History

Merge tag 'amd-drm-next-5.13-2021-03-23' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-5.13-2021-03-23: amdgpu: - Debugfs cleanup - Various cleanups and spelling fixes - Flexible array cleanups - Initial AMD Freesync HDMI - Display fixes - 10bpc dithering improvements - Display ASSR support - Clean up and unify powerplay and swsmu interfaces - Vangogh fixes - Add SMU gfx busy queues for RV/PCO - PCIE DPM fixes - S0ix fixes - GPU metrics data fixes - DCN secure display support - Backlight type override - Add initial support for Aldebaran - RAS fixes - Prime fixes for A+A systems - Reset fixes - Initial resource cursor support - Drop legacy IO BAR requirements - Various power fixes amdkfd: - MMU notifier fixes - APU fixes radeon: - Debugfs cleanups - Flexible array cleanups UAPI: - amdgpu: Add a new INFO ioctl interface to query video capabilities rather than hardcoding them in userspace. This allows us to provide fine grained asic capabilities (e.g., if a particular part is bandwidth limited, we can limit the capabilities). Proposed userspace: https://gitlab.freedesktop.org/leoliu/drm/-/commits/info_video_caps https://gitlab.freedesktop.org/leoliu/mesa/-/commits/info_video_caps - amdkfd: bump the driver version. There was a problem with reporting some RAS features on older versions of the driver. Proposed userspace: https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/commit/7cdd63475c36bb9f49bb960f90f9a8cdb7e80a21 Danvet: A bunch of conflicts all over, but it seems to compile ... I did put the call to dc_allow_idle_optimizations() on a single line since it looked a bit too jarring to be left alone.

Daniel Vetter committed 5 years ago

Commit 2cbcb782 parents 06debd6 8c44390323 files · +142321 −3917Browse files

Unified · Split

241 files
  1. drivers/gpu/drm/amd/amdgpu/Kconfig +0 −9
  2. drivers/gpu/drm/amd/amdgpu/Makefile +10 −4
  3. drivers/gpu/drm/amd/amdgpu/aldebaran_reg_init.c +54 −0
  4. drivers/gpu/drm/amd/amdgpu/amdgpu.h +43 −13
  5. drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +3 −9
  6. drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +1 −0
  7. drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c +47 −0
  8. drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c +8 −8
  9. drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.h +30 −0
  10. drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +35 −2
  11. drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c +0 −43
  12. drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c +67 −6
  13. drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c +10 −6
  14. drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +7 −7
  15. drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c +59 −125
  16. drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h +3 −11
  17. drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +162 −107
  18. drivers/gpu/drm/amd/amdgpu/amdgpu_display.c +58 −14
  19. drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +12 −17
  20. drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +125 −3
  21. drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c +2 −2
  22. drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +22 −29
  23. drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +1 −25
  24. drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h +0 −3
  25. drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +11 −11
  26. drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +8 −1
  27. drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h +3 −0
  28. drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c +147 −3
  29. drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h +27 −4
  30. drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +10 −13
  31. drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c +49 −5
  32. drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h +5 −1
  33. drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c +36 −1
  34. drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +95 −38
  35. drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +8 −0
  36. drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +14 −8
  37. drivers/gpu/drm/amd/amdgpu/amdgpu_object.h +1 −2
  38. drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c +77 −18
  39. drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h +9 −6
  40. drivers/gpu/drm/amd/amdgpu/amdgpu_rap.c +4 −3
  41. drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +119 −72
  42. drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +6 −7
  43. drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c +12 −31
  44. drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h +1 −3
  45. drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h +105 −0
  46. drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h +5 −0
  47. drivers/gpu/drm/amd/amdgpu/amdgpu_securedisplay.c +3 −0
  48. drivers/gpu/drm/amd/amdgpu/amdgpu_smuio.h +2 −0
  49. drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +171 −287
  50. drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 −7
  51. drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c +1 −0
  52. drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.h +4 −0
  53. drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +8 −0
  54. drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +11 −2
  55. drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +1 −0
  56. drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3 −3
  57. drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +1 −3
  58. drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c +13 −4
  59. drivers/gpu/drm/amd/amdgpu/athub_v2_1.c +2 −4
  60. drivers/gpu/drm/amd/amdgpu/atom.c +2 −2
  61. drivers/gpu/drm/amd/amdgpu/atom.h +0 −2
  62. drivers/gpu/drm/amd/amdgpu/cik.c +77 −0
  63. drivers/gpu/drm/amd/amdgpu/df_v3_6.c +2 −0
  64. drivers/gpu/drm/amd/amdgpu/dimgrey_cavefish_reg_init.c +0 −0
  65. drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +5 −4
  66. drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +123 −38
  67. drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.c +1286 −0
  68. drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.h +40 −0
  69. drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +62 −13
  70. drivers/gpu/drm/amd/amdgpu/gfxhub_v1_1.c +45 −9
  71. drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c +4 −3
  72. drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +1 −0
  73. drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +1 −0
  74. drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +1 −0
  75. drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +137 −15
  76. drivers/gpu/drm/amd/amdgpu/hdp_v4_0.c +3 −0
  77. drivers/gpu/drm/amd/amdgpu/jpeg_v2_5.c +63 −2
  78. drivers/gpu/drm/amd/amdgpu/jpeg_v2_5.h +1 −0
  79. drivers/gpu/drm/amd/amdgpu/mmhub_v1_7.c +1329 −0
  80. drivers/gpu/drm/amd/amdgpu/mmhub_v1_7.h +28 −0
  81. drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c +40 −8
  82. drivers/gpu/drm/amd/amdgpu/nv.c +182 −39
  83. drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h +1 −0
  84. drivers/gpu/drm/amd/amdgpu/psp_v13_0.c +378 −0
  85. drivers/gpu/drm/amd/amdgpu/psp_v13_0.h +30 −0
  86. drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +162 −19
  87. drivers/gpu/drm/amd/amdgpu/sdma_v4_4.c +232 −0
  88. drivers/gpu/drm/amd/amdgpu/sdma_v4_4.h +28 −0
  89. drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +2 −2
  90. drivers/gpu/drm/amd/amdgpu/si.c +111 −0
  91. drivers/gpu/drm/amd/amdgpu/smuio_v13_0.c +121 −0
  92. drivers/gpu/drm/amd/amdgpu/smuio_v13_0.h +30 −0
  93. drivers/gpu/drm/amd/amdgpu/soc15.c +310 −50
  94. drivers/gpu/drm/amd/amdgpu/soc15.h +12 −0
  95. drivers/gpu/drm/amd/amdgpu/soc15_common.h +48 −0
  96. drivers/gpu/drm/amd/amdgpu/ta_secureDisplay_if.h +1 −0
  97. drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +97 −2
  98. drivers/gpu/drm/amd/amdgpu/vcn_v2_5.h +1 −0
  99. drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +23 −1
  100. drivers/gpu/drm/amd/amdgpu/vega10_ih.c +11 −21
  101. drivers/gpu/drm/amd/amdgpu/vega20_ih.c +28 −21
  102. drivers/gpu/drm/amd/amdgpu/vi.c +190 −0
  103. drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c +3 −2
  104. drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h +492 −0
  105. drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm +1 −0
  106. drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm +123 −81
  107. drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +1 −5
  108. drivers/gpu/drm/amd/amdkfd/kfd_crat.c +16 −1
  109. drivers/gpu/drm/amd/amdkfd/kfd_device.c +25 −0
  110. drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +33 −24
  111. drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h +0 −5
  112. drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c +1 −0
  113. drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +3 −2
  114. drivers/gpu/drm/amd/amdkfd/kfd_iommu.c +6 −0
  115. drivers/gpu/drm/amd/amdkfd/kfd_iommu.h +7 −2
  116. drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h +1 −0
  117. drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c +8 −0
  118. drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c +8 −0
  119. drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +8 −0
  120. drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c +8 −0
  121. drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c +1 −0
  122. drivers/gpu/drm/amd/amdkfd/kfd_priv.h +9 −0
  123. drivers/gpu/drm/amd/amdkfd/kfd_process.c +56 −10
  124. drivers/gpu/drm/amd/amdkfd/kfd_topology.c +1 −0
  125. drivers/gpu/drm/amd/amdkfd/kfd_topology.h +3 −2
  126. drivers/gpu/drm/amd/display/Kconfig +14 −0
  127. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +622 −102
  128. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h +50 −16
  129. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crc.c +288 −5
  130. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crc.h +37 −0
  131. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c +438 −36
  132. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.h +4 −1
  133. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c +42 −2
  134. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_irq.c +77 −36
  135. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_irq_params.h +9 −0
  136. drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_mst_types.c +10 −7
  137. drivers/gpu/drm/amd/display/dc/Makefile +1 −1
  138. drivers/gpu/drm/amd/display/dc/basics/dc_common.c +6 −14
  139. drivers/gpu/drm/amd/display/dc/basics/dc_common.h +2 −2
  140. drivers/gpu/drm/amd/display/dc/bios/bios_parser2.c +3 −2
  141. drivers/gpu/drm/amd/display/dc/bios/bios_parser_helper.c +3 −2
  142. drivers/gpu/drm/amd/display/dc/bios/bios_parser_helper.h +1 −1
  143. drivers/gpu/drm/amd/display/dc/bios/command_table.c +21 −0
  144. drivers/gpu/drm/amd/display/dc/bios/command_table2.c +84 −9
  145. drivers/gpu/drm/amd/display/dc/bios/command_table2.h +2 −1
  146. drivers/gpu/drm/amd/display/dc/clk_mgr/clk_mgr.c +13 −4
  147. drivers/gpu/drm/amd/display/dc/clk_mgr/dcn20/dcn20_clk_mgr.c +1 −1
  148. drivers/gpu/drm/amd/display/dc/clk_mgr/dcn30/dcn30_clk_mgr.c +3 −1
  149. drivers/gpu/drm/amd/display/dc/core/dc.c +260 −46
  150. drivers/gpu/drm/amd/display/dc/core/dc_link.c +28 −8
  151. drivers/gpu/drm/amd/display/dc/core/dc_link_ddc.c +3 −2
  152. drivers/gpu/drm/amd/display/dc/core/dc_link_dp.c +17 −1
  153. drivers/gpu/drm/amd/display/dc/core/dc_resource.c +43 −0
  154. drivers/gpu/drm/amd/display/dc/core/dc_stat.c +64 −0
  155. drivers/gpu/drm/amd/display/dc/dc.h +25 −2
  156. drivers/gpu/drm/amd/display/dc/dc_bios_types.h +2 −1
  157. drivers/gpu/drm/amd/display/dc/dc_ddc_types.h +0 −10
  158. drivers/gpu/drm/amd/display/dc/dc_dmub_srv.c +34 −0
  159. drivers/gpu/drm/amd/display/dc/dc_dmub_srv.h +8 −0
  160. drivers/gpu/drm/amd/display/dc/dc_dsc.h +8 −3
  161. drivers/gpu/drm/amd/display/dc/dc_hw_types.h +2 −0
  162. drivers/gpu/drm/amd/display/dc/dc_link.h +14 −5
  163. drivers/gpu/drm/amd/display/dc/dc_stat.h +42 −0
  164. drivers/gpu/drm/amd/display/dc/dc_stream.h +10 −11
  165. drivers/gpu/drm/amd/display/dc/dce/Makefile +1 −1
  166. drivers/gpu/drm/amd/display/dc/dce/dce_aux.c +22 −18
  167. drivers/gpu/drm/amd/display/dc/dce/dce_aux.h +2 −2
  168. drivers/gpu/drm/amd/display/dc/dce/dce_clock_source.c +46 −6
  169. drivers/gpu/drm/amd/display/dc/dce/dce_dmcu.c +84 −0
  170. drivers/gpu/drm/amd/display/dc/dce/dce_opp.c +1 −3
  171. drivers/gpu/drm/amd/display/dc/dce/dmub_abm.c +5 −5
  172. drivers/gpu/drm/amd/display/dc/dce/dmub_outbox.c +60 −0
  173. drivers/gpu/drm/amd/display/dc/dce/dmub_outbox.h +33 −0
  174. drivers/gpu/drm/amd/display/dc/dce/dmub_psr.c +12 −7
  175. drivers/gpu/drm/amd/display/dc/dce110/dce110_hw_sequencer.c +59 −32
  176. drivers/gpu/drm/amd/display/dc/dce110/dce110_resource.c +2 −0
  177. drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c +260 −40
  178. drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.h +5 −3
  179. drivers/gpu/drm/amd/display/dc/dcn10/dcn10_init.c +0 −1
  180. drivers/gpu/drm/amd/display/dc/dcn10/dcn10_optc.h +2 −0
  181. drivers/gpu/drm/amd/display/dc/dcn10/dcn10_resource.c +5 −1
  182. drivers/gpu/drm/amd/display/dc/dcn10/dcn10_stream_encoder.c +0 −1
  183. drivers/gpu/drm/amd/display/dc/dcn20/dcn20_hwseq.c +14 −35
  184. drivers/gpu/drm/amd/display/dc/dcn20/dcn20_init.c +1 −1
  185. drivers/gpu/drm/amd/display/dc/dcn20/dcn20_optc.c +124 −0
  186. drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c +12 −4
  187. drivers/gpu/drm/amd/display/dc/dcn21/dcn21_init.c +0 −1
  188. drivers/gpu/drm/amd/display/dc/dcn21/dcn21_resource.c +15 −6
  189. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_cm_common.c +18 −8
  190. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_dio_stream_encoder.c +0 −1
  191. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_dpp.c +2 −2
  192. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_dpp_cm.c +2 −2
  193. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_dwb_cm.c +1 −1
  194. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_hwseq.c +34 −66
  195. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_hwseq.h +0 −4
  196. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_init.c +0 −1
  197. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_mpc.c +14 −12
  198. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_resource.c +18 −7
  199. drivers/gpu/drm/amd/display/dc/dcn30/dcn30_resource.h +1 −0
  200. drivers/gpu/drm/amd/display/dc/dcn301/dcn301_init.c +0 −1
  201. drivers/gpu/drm/amd/display/dc/dcn301/dcn301_resource.c +6 −1
  202. drivers/gpu/drm/amd/display/dc/dcn302/dcn302_resource.c +6 −1
  203. drivers/gpu/drm/amd/display/dc/dm_helpers.h +2 −0
  204. drivers/gpu/drm/amd/display/dc/dml/dcn20/display_mode_vba_20.c +1 −0
  205. drivers/gpu/drm/amd/display/dc/dml/dcn20/display_mode_vba_20v2.c +1 −0
  206. drivers/gpu/drm/amd/display/dc/dml/dcn30/display_mode_vba_30.c +1 −1
  207. drivers/gpu/drm/amd/display/dc/dml/display_mode_vba.c +0 −3
  208. drivers/gpu/drm/amd/display/dc/dsc/dc_dsc.c +34 −71
  209. drivers/gpu/drm/amd/display/dc/gpio/dce110/hw_factory_dce110.c +0 −4
  210. drivers/gpu/drm/amd/display/dc/hdcp/hdcp_msg.c +5 −4
  211. drivers/gpu/drm/amd/display/dc/inc/clock_source.h +5 −0
  212. drivers/gpu/drm/amd/display/dc/inc/core_types.h +15 −0
  213. drivers/gpu/drm/amd/display/dc/inc/dc_link_ddc.h +2 −1
  214. drivers/gpu/drm/amd/display/dc/inc/hw/aux_engine.h +3 −1
  215. drivers/gpu/drm/amd/display/dc/inc/hw/dmcu.h +21 −0
  216. drivers/gpu/drm/amd/display/dc/inc/hw/hw_shared.h +1 −0
  217. drivers/gpu/drm/amd/display/dc/inc/hw/timing_generator.h +12 −0
  218. drivers/gpu/drm/amd/display/dc/inc/hw_sequencer.h +3 −4
  219. drivers/gpu/drm/amd/display/dc/inc/resource.h +5 −0
  220. drivers/gpu/drm/amd/display/dc/irq/dcn20/irq_service_dcn20.c +31 −0
  221. drivers/gpu/drm/amd/display/dc/irq/dcn21/irq_service_dcn21.c +34 −0
  222. drivers/gpu/drm/amd/display/dc/irq/dcn30/irq_service_dcn30.c +38 −0
  223. drivers/gpu/drm/amd/display/dc/irq/dcn302/irq_service_dcn302.c +30 −0
  224. drivers/gpu/drm/amd/display/dc/irq_types.h +2 −1
  225. drivers/gpu/drm/amd/display/dmub/dmub_srv.h +55 −0
  226. drivers/gpu/drm/amd/display/dmub/dmub_srv_stat.h +41 −0
  227. drivers/gpu/drm/amd/display/dmub/inc/dmub_cmd.h +1326 −169
  228. drivers/gpu/drm/amd/display/dmub/inc/dmub_trace_buffer.h +1 −2
  229. drivers/gpu/drm/amd/display/dmub/src/Makefile +1 −1
  230. drivers/gpu/drm/amd/display/dmub/src/dmub_dcn20.c +57 −0
  231. drivers/gpu/drm/amd/display/dmub/src/dmub_dcn20.h +33 −2
  232. drivers/gpu/drm/amd/display/dmub/src/dmub_dcn30.c +7 −0
  233. drivers/gpu/drm/amd/display/dmub/src/dmub_srv.c +79 −5
  234. drivers/gpu/drm/amd/display/dmub/src/dmub_srv_stat.c +105 −0
  235. drivers/gpu/drm/amd/display/modules/freesync/freesync.c +26 −2
  236. drivers/gpu/drm/amd/display/modules/inc/mod_freesync.h +2 −1
  237. drivers/gpu/drm/amd/include/aldebaran_ip_offset.h +1738 −0
  238. drivers/gpu/drm/amd/include/asic_reg/dcn/dcn_3_0_0_offset.h +1 −1
  239. drivers/gpu/drm/amd/include/asic_reg/gc/gc_9_4_2_offset.h +7683 −0
  240. drivers/gpu/drm/amd/include/asic_reg/gc/gc_9_4_2_sh_mask.h +32949 −0
  241. drivers/gpu/drm/amd/include/asic_reg/mmhub/mmhub_1_7_offset.h +5125 −0
drivers/gpu/drm/amd/amdgpu/Kconfig+0−9
@@ -34,15 +34,6 @@ config DRM_AMDGPU_USERPTR · 33 unmodified lines
3434 This option selects CONFIG_HMM and CONFIG_HMM_MIRROR if it
3535 isn't already selected to enabled full userptr support.
3636
37−config DRM_AMDGPU_GART_DEBUGFS
38− bool "Allow GART access through debugfs"
39− depends on DRM_AMDGPU
40− depends on DEBUG_FS
41− default n
42− help
43− Selecting this option creates a debugfs file to inspect the mapped
44− pages. Uses more memory for housekeeping, enable only for debugging.
45−
4637 source "drivers/gpu/drm/amd/acp/Kconfig"
4738 source "drivers/gpu/drm/amd/display/Kconfig"
4839 source "drivers/gpu/drm/amd/amdkfd/Kconfig"
drivers/gpu/drm/amd/amdgpu/Makefile+10−4
@@ -71,7 +71,7 @@ amdgpu-y += \ · 70 unmodified lines
7171 vi.o mxgpu_vi.o nbio_v6_1.o soc15.o emu_soc.o mxgpu_ai.o nbio_v7_0.o vega10_reg_init.o \
7272 vega20_reg_init.o nbio_v7_4.o nbio_v2_3.o nv.o navi10_reg_init.o navi14_reg_init.o \
7373 arct_reg_init.o navi12_reg_init.o mxgpu_nv.o sienna_cichlid_reg_init.o vangogh_reg_init.o \
74− nbio_v7_2.o dimgrey_cavefish_reg_init.o hdp_v4_0.o hdp_v5_0.o
74+ nbio_v7_2.o dimgrey_cavefish_reg_init.o hdp_v4_0.o hdp_v5_0.o aldebaran_reg_init.o
7575
7676 # add DF block
7777 amdgpu-y += \
@@ -83,7 +83,8 @@ amdgpu-y += \ · 5 unmodified lines
8383 gmc_v7_0.o \
8484 gmc_v8_0.o \
8585 gfxhub_v1_0.o mmhub_v1_0.o gmc_v9_0.o gfxhub_v1_1.o mmhub_v9_4.o \
86− gfxhub_v2_0.o mmhub_v2_0.o gmc_v10_0.o gfxhub_v2_1.o mmhub_v2_3.o
86+ gfxhub_v2_0.o mmhub_v2_0.o gmc_v10_0.o gfxhub_v2_1.o mmhub_v2_3.o \
87+ mmhub_v1_7.o
8788
8889 # add UMC block
8990 amdgpu-y += \
@@ -106,7 +107,8 @@ amdgpu-y += \ · 16 unmodified lines
106107 psp_v3_1.o \
107108 psp_v10_0.o \
108109 psp_v11_0.o \
109− psp_v12_0.o
110+ psp_v12_0.o \
111+ psp_v13_0.o
110112
111113 # add DCE block
112114 amdgpu-y += \
@@ -121,6 +123,7 @@ amdgpu-y += \ · 8 unmodified lines
121123 gfx_v8_0.o \
122124 gfx_v9_0.o \
123125 gfx_v9_4.o \
126+ gfx_v9_4_2.o \
124127 gfx_v10_0.o
125128
126129 # add async DMA block
@@ -129,6 +132,7 @@ amdgpu-y += \ · 2 unmodified lines
129132 sdma_v2_4.o \
130133 sdma_v3_0.o \
131134 sdma_v4_0.o \
135+ sdma_v4_4.o \
132136 sdma_v5_0.o \
133137 sdma_v5_2.o
134138
@@ -172,7 +176,8 @@ amdgpu-y += \ · 37 unmodified lines
172176 amdgpu-y += \
173177 smuio_v9_0.o \
174178 smuio_v11_0.o \
175− smuio_v11_0_6.o
179+ smuio_v11_0_6.o \
180+ smuio_v13_0.o
176181
177182 # add amdkfd interfaces
178183 amdgpu-y += amdgpu_amdkfd.o
@@ -187,6 +192,7 @@ amdgpu-y += \ · 8 unmodified lines
187192 amdgpu_amdkfd_gfx_v8.o \
188193 amdgpu_amdkfd_gfx_v9.o \
189194 amdgpu_amdkfd_arcturus.o \
195+ amdgpu_amdkfd_aldebaran.o \
190196 amdgpu_amdkfd_gfx_v10.o \
191197 amdgpu_amdkfd_gfx_v10_3.o
192198
drivers/gpu/drm/amd/amdgpu/aldebaran_reg_init.c+54−0
@@ -0,0 +1,54 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#include "amdgpu.h"
24+#include "soc15.h"
25+
26+#include "soc15_common.h"
27+#include "aldebaran_ip_offset.h"
28+
29+int aldebaran_reg_base_init(struct amdgpu_device *adev)
30+{
31+ /* HW has more IP blocks, only initialized the block needed by our driver */
32+ uint32_t i;
33+ for (i = 0 ; i < MAX_INSTANCE ; ++i) {
34+ adev->reg_offset[GC_HWIP][i] = (uint32_t *)(&(GC_BASE.instance[i]));
35+ adev->reg_offset[HDP_HWIP][i] = (uint32_t *)(&(HDP_BASE.instance[i]));
36+ adev->reg_offset[MMHUB_HWIP][i] = (uint32_t *)(&(MMHUB_BASE.instance[i]));
37+ adev->reg_offset[ATHUB_HWIP][i] = (uint32_t *)(&(ATHUB_BASE.instance[i]));
38+ adev->reg_offset[NBIO_HWIP][i] = (uint32_t *)(&(NBIO_BASE.instance[i]));
39+ adev->reg_offset[MP0_HWIP][i] = (uint32_t *)(&(MP0_BASE.instance[i]));
40+ adev->reg_offset[MP1_HWIP][i] = (uint32_t *)(&(MP1_BASE.instance[i]));
41+ adev->reg_offset[DF_HWIP][i] = (uint32_t *)(&(DF_BASE.instance[i]));
42+ adev->reg_offset[OSSSYS_HWIP][i] = (uint32_t *)(&(OSSSYS_BASE.instance[i]));
43+ adev->reg_offset[SDMA0_HWIP][i] = (uint32_t *)(&(SDMA0_BASE.instance[i]));
44+ adev->reg_offset[SDMA1_HWIP][i] = (uint32_t *)(&(SDMA1_BASE.instance[i]));
45+ adev->reg_offset[SDMA2_HWIP][i] = (uint32_t *)(&(SDMA2_BASE.instance[i]));
46+ adev->reg_offset[SDMA3_HWIP][i] = (uint32_t *)(&(SDMA3_BASE.instance[i]));
47+ adev->reg_offset[SDMA4_HWIP][i] = (uint32_t *)(&(SDMA4_BASE.instance[i]));
48+ adev->reg_offset[SMUIO_HWIP][i] = (uint32_t *)(&(SMUIO_BASE.instance[i]));
49+ adev->reg_offset[THM_HWIP][i] = (uint32_t *)(&(THM_BASE.instance[i]));
50+ adev->reg_offset[UMC_HWIP][i] = (uint32_t *)(&(UMC_BASE.instance[i]));
51+ adev->reg_offset[VCN_HWIP][i] = (uint32_t *)(&(VCN_BASE.instance[i]));
52+ }
53+ return 0;
54+}
drivers/gpu/drm/amd/amdgpu/amdgpu.h+43−13
@@ -124,6 +124,16 @@ struct amdgpu_mgpu_info · 123 unmodified lines
124124 uint32_t num_gpu;
125125 uint32_t num_dgpu;
126126 uint32_t num_apu;
127+
128+ /* delayed reset_func for XGMI configuration if necessary */
129+ struct delayed_work delayed_reset_work;
130+ bool pending_reset;
131+};
132+
133+struct amdgpu_watchdog_timer
134+{
135+ bool timeout_fatal_disable;
136+ uint32_t period; /* maxCycles = (1 << period), the number of cycles before a timeout */
127137 };
128138
129139 #define AMDGPU_MAX_TIMEOUT_PARAM_LENGTH 256
@@ -177,7 +187,9 @@ extern int amdgpu_compute_multipipe; · 47 unmodified lines
177187 extern int amdgpu_gpu_recovery;
178188 extern int amdgpu_emu_mode;
179189 extern uint amdgpu_smu_memory_pool_size;
190+extern int amdgpu_smu_pptable_id;
180191 extern uint amdgpu_dc_feature_mask;
192+extern uint amdgpu_freesync_vid_mode;
181193 extern uint amdgpu_dc_debug_mask;
182194 extern uint amdgpu_dm_abm_level;
183195 extern int amdgpu_backlight;
@@ -185,6 +197,7 @@ extern struct amdgpu_mgpu_info mgpu_info; · 1 unmodified line
185197 extern int amdgpu_ras_enable;
186198 extern uint amdgpu_ras_mask;
187199 extern int amdgpu_bad_page_threshold;
200+extern struct amdgpu_watchdog_timer amdgpu_watchdog_timer;
188201 extern int amdgpu_async_gfx_ring;
189202 extern int amdgpu_mcbp;
190203 extern int amdgpu_discovery;
@@ -584,6 +597,19 @@ enum amd_reset_method { · 393 unmodified lines
584597 AMD_RESET_METHOD_PCI,
585598 };
586599
600+struct amdgpu_video_codec_info {
601+ u32 codec_type;
602+ u32 max_width;
603+ u32 max_height;
604+ u32 max_pixels_per_frame;
605+ u32 max_level;
606+};
607+
608+struct amdgpu_video_codecs {
609+ const u32 codec_count;
610+ const struct amdgpu_video_codec_info *codec_array;
611+};
612+
587613 /*
588614 * ASIC specific functions.
589615 */
@@ -628,6 +654,9 @@ struct amdgpu_asic_funcs { · 38 unmodified lines
628654 void (*pre_asic_init)(struct amdgpu_device *adev);
629655 /* enter/exit umd stable pstate */
630656 int (*update_umd_stable_pstate)(struct amdgpu_device *adev, bool enter);
657+ /* query video codecs */
658+ int (*query_video_codecs)(struct amdgpu_device *adev, bool encode,
659+ const struct amdgpu_video_codecs **codecs);
631660 };
632661
633662 /*
@@ -792,12 +821,7 @@ struct amdgpu_device { · 158 unmodified lines
792821 bool accel_working;
793822 struct notifier_block acpi_nb;
794823 struct amdgpu_i2c_chan *i2c_bus[AMDGPU_MAX_I2C_BUS];
795− struct amdgpu_debugfs debugfs[AMDGPU_DEBUGFS_MAX_COMPONENTS];
796− unsigned debugfs_count;
797−#if defined(CONFIG_DEBUG_FS)
798− struct dentry *debugfs_preempt;
799− struct dentry *debugfs_regs[AMDGPU_DEBUGFS_MAX_COMPONENTS];
800−#endif
824+ struct debugfs_blob_wrapper debugfs_vbios_blob;
801825 struct amdgpu_atif *atif;
802826 struct amdgpu_atcs atcs;
803827 struct mutex srbm_mutex;
@@ -853,8 +877,6 @@ struct amdgpu_device { · 49 unmodified lines
853877 spinlock_t audio_endpt_idx_lock;
854878 amdgpu_block_rreg_t audio_endpt_rreg;
855879 amdgpu_block_wreg_t audio_endpt_wreg;
856− void __iomem *rio_mem;
857− resource_size_t rio_mem_size;
858880 struct amdgpu_doorbell doorbell;
859881
860882 /* clock/pll info */
@@ -897,6 +919,7 @@ struct amdgpu_device { · 36 unmodified lines
897919 struct amdgpu_irq_src vupdate_irq;
898920 struct amdgpu_irq_src pageflip_irq;
899921 struct amdgpu_irq_src hpd_irq;
922+ struct amdgpu_irq_src dmub_trace_irq;
900923
901924 /* rings */
902925 u64 fence_context;
@@ -1024,6 +1047,7 @@ struct amdgpu_device { · 121 unmodified lines
10241047
10251048 int asic_reset_res;
10261049 struct work_struct xgmi_reset_work;
1050+ struct list_head reset_list;
10271051
10281052 long gfx_timeout;
10291053 long sdma_timeout;
@@ -1088,9 +1112,6 @@ void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev, · 58 unmodified lines
10881112 void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value);
10891113 uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset);
10901114
1091−u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg);
1092−void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v);
1093−
10941115 u32 amdgpu_device_indirect_rreg(struct amdgpu_device *adev,
10951116 u32 pcie_index, u32 pcie_data,
10961117 u32 reg_addr);
@@ -1107,6 +1128,15 @@ void amdgpu_device_indirect_wreg64(struct amdgpu_device *adev, · 10 unmodified lines
11071128 bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type);
11081129 bool amdgpu_device_has_dc_support(struct amdgpu_device *adev);
11091130
1131+int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
1132+ struct amdgpu_job *job,
1133+ bool *need_full_reset_arg);
1134+
1135+int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive,
1136+ struct list_head *device_list_handle,
1137+ bool *need_full_reset_arg,
1138+ bool skip_hw_reset);
1139+
11101140 int emu_soc_asic_init(struct amdgpu_device *adev);
11111141
11121142 /*
@@ -1172,8 +1202,6 @@ int emu_soc_asic_init(struct amdgpu_device *adev); · 59 unmodified lines
11721202 } while (0)
11731203
11741204 #define DREG32_SYS(sqf, adev, reg) seq_printf((sqf), #reg " : 0x%08X\n", amdgpu_device_rreg((adev), (reg), false))
1175−#define RREG32_IO(reg) amdgpu_io_rreg(adev, (reg))
1176−#define WREG32_IO(reg, v) amdgpu_io_wreg(adev, (reg), (v))
11771205
11781206 #define REG_FIELD_SHIFT(reg, field) reg##__##field##__SHIFT
11791207 #define REG_FIELD_MASK(reg, field) reg##__##field##_MASK
@@ -1227,6 +1255,7 @@ int emu_soc_asic_init(struct amdgpu_device *adev); · 47 unmodified lines
12271255 #define amdgpu_asic_pre_asic_init(adev) (adev)->asic_funcs->pre_asic_init((adev))
12281256 #define amdgpu_asic_update_umd_stable_pstate(adev, enter) \
12291257 ((adev)->asic_funcs->update_umd_stable_pstate ? (adev)->asic_funcs->update_umd_stable_pstate((adev), (enter)) : 0)
1258+#define amdgpu_asic_query_video_codecs(adev, e, c) (adev)->asic_funcs->query_video_codecs((adev), (e), (c))
12301259
12311260 #define amdgpu_inc_vram_lost(adev) atomic_inc(&((adev)->vram_lost_counter));
12321261
@@ -1247,6 +1276,7 @@ void amdgpu_device_program_register_sequence(struct amdgpu_device *adev, · 14 unmodified lines
12471276 const u32 array_size);
12481277
12491278 bool amdgpu_device_supports_atpx(struct drm_device *dev);
1279+int amdgpu_device_mode1_reset(struct amdgpu_device *adev);
12501280 bool amdgpu_device_supports_boco(struct drm_device *dev);
12511281 bool amdgpu_device_supports_baco(struct drm_device *dev);
12521282 bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev,
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c+3−9
@@ -44,7 +44,7 @@ int amdgpu_amdkfd_init(void) · 43 unmodified lines
4444 int ret;
4545
4646 si_meminfo(&si);
47− amdgpu_amdkfd_total_mem_size = si.totalram - si.totalhigh;
47+ amdgpu_amdkfd_total_mem_size = si.freeram - si.freehigh;
4848 amdgpu_amdkfd_total_mem_size *= si.mem_unit;
4949
5050 ret = kgd2kfd_init();
@@ -165,7 +165,8 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) · 114 unmodified lines
165165 adev->doorbell_index.last_non_cp;
166166 }
167167
168− kgd2kfd_device_init(adev->kfd.dev, adev_to_drm(adev), &gpu_resources);
168+ adev->kfd.init_complete = kgd2kfd_device_init(adev->kfd.dev,
169+ adev_to_drm(adev), &gpu_resources);
169170 }
170171 }
171172
@@ -638,13 +639,6 @@ void amdgpu_amdkfd_set_compute_idle(struct kgd_dev *kgd, bool idle) · 466 unmodified lines
638639 {
639640 struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
640641
641− /* Temp workaround to fix the soft hang observed in certain compute
642− * applications if GFXOFF is enabled.
643− */
644− if (adev->asic_type == CHIP_SIENNA_CICHLID) {
645− pr_debug("GFXOFF is %s\n", idle ? "enabled" : "disabled");
646− amdgpu_gfx_off_ctrl(adev, idle);
647− }
648642 amdgpu_dpm_switch_power_profile(adev,
649643 PP_SMC_POWER_PROFILE_COMPUTE,
650644 !idle);
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h+1−0
@@ -80,6 +80,7 @@ struct amdgpu_amdkfd_fence { · 79 unmodified lines
8080 struct amdgpu_kfd_dev {
8181 struct kfd_dev *dev;
8282 uint64_t vram_used;
83+ bool init_complete;
8384 };
8485
8586 enum kgd_engine_type {
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c+47−0
@@ -0,0 +1,47 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ */
22+#include "amdgpu.h"
23+#include "amdgpu_amdkfd.h"
24+#include "amdgpu_amdkfd_arcturus.h"
25+#include "amdgpu_amdkfd_gfx_v9.h"
26+
27+const struct kfd2kgd_calls aldebaran_kfd2kgd = {
28+ .program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
29+ .set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
30+ .init_interrupts = kgd_gfx_v9_init_interrupts,
31+ .hqd_load = kgd_gfx_v9_hqd_load,
32+ .hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
33+ .hqd_sdma_load = kgd_arcturus_hqd_sdma_load,
34+ .hqd_dump = kgd_gfx_v9_hqd_dump,
35+ .hqd_sdma_dump = kgd_arcturus_hqd_sdma_dump,
36+ .hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
37+ .hqd_sdma_is_occupied = kgd_arcturus_hqd_sdma_is_occupied,
38+ .hqd_destroy = kgd_gfx_v9_hqd_destroy,
39+ .hqd_sdma_destroy = kgd_arcturus_hqd_sdma_destroy,
40+ .address_watch_disable = kgd_gfx_v9_address_watch_disable,
41+ .address_watch_execute = kgd_gfx_v9_address_watch_execute,
42+ .wave_control_execute = kgd_gfx_v9_wave_control_execute,
43+ .address_watch_get_offset = kgd_gfx_v9_address_watch_get_offset,
44+ .get_atc_vmid_pasid_mapping_info =
45+ kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
46+ .set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
47+};
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c+8−8
@@ -122,7 +122,7 @@ static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev, · 121 unmodified lines
122122 return sdma_rlc_reg_offset;
123123 }
124124
125−static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
125+int kgd_arcturus_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
126126 uint32_t __user *wptr, struct mm_struct *mm)
127127 {
128128 struct amdgpu_device *adev = get_amdgpu_device(kgd);
@@ -192,7 +192,7 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, · 63 unmodified lines
192192 return 0;
193193 }
194194
195−static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
195+int kgd_arcturus_hqd_sdma_dump(struct kgd_dev *kgd,
196196 uint32_t engine_id, uint32_t queue_id,
197197 uint32_t (**dump)[2], uint32_t *n_regs)
198198 {
@@ -224,7 +224,7 @@ static int kgd_hqd_sdma_dump(struct kgd_dev *kgd, · 25 unmodified lines
224224 return 0;
225225 }
226226
227−static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
227+bool kgd_arcturus_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
228228 {
229229 struct amdgpu_device *adev = get_amdgpu_device(kgd);
230230 struct v9_sdma_mqd *m;
@@ -243,7 +243,7 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) · 12 unmodified lines
243243 return false;
244244 }
245245
246−static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
246+int kgd_arcturus_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
247247 unsigned int utimeout)
248248 {
249249 struct amdgpu_device *adev = get_amdgpu_device(kgd);
@@ -289,13 +289,13 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = { · 39 unmodified lines
289289 .init_interrupts = kgd_gfx_v9_init_interrupts,
290290 .hqd_load = kgd_gfx_v9_hqd_load,
291291 .hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
292− .hqd_sdma_load = kgd_hqd_sdma_load,
292+ .hqd_sdma_load = kgd_arcturus_hqd_sdma_load,
293293 .hqd_dump = kgd_gfx_v9_hqd_dump,
294− .hqd_sdma_dump = kgd_hqd_sdma_dump,
294+ .hqd_sdma_dump = kgd_arcturus_hqd_sdma_dump,
295295 .hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
296− .hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
296+ .hqd_sdma_is_occupied = kgd_arcturus_hqd_sdma_is_occupied,
297297 .hqd_destroy = kgd_gfx_v9_hqd_destroy,
298− .hqd_sdma_destroy = kgd_hqd_sdma_destroy,
298+ .hqd_sdma_destroy = kgd_arcturus_hqd_sdma_destroy,
299299 .address_watch_disable = kgd_gfx_v9_address_watch_disable,
300300 .address_watch_execute = kgd_gfx_v9_address_watch_execute,
301301 .wave_control_execute = kgd_gfx_v9_wave_control_execute,
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.h+30−0
@@ -0,0 +1,30 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ */
22+
23+int kgd_arcturus_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
24+ uint32_t __user *wptr, struct mm_struct *mm);
25+int kgd_arcturus_hqd_sdma_dump(struct kgd_dev *kgd,
26+ uint32_t engine_id, uint32_t queue_id,
27+ uint32_t (**dump)[2], uint32_t *n_regs);
28+bool kgd_arcturus_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd);
29+int kgd_arcturus_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
30+ unsigned int utimeout);
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c+35−2
@@ -31,6 +31,7 @@ · 30 unmodified lines
3131 #include "amdgpu_amdkfd.h"
3232 #include "amdgpu_dma_buf.h"
3333 #include <uapi/linux/kfd_ioctl.h>
34+#include "amdgpu_xgmi.h"
3435
3536 /* BO flag to indicate a KFD userptr BO */
3637 #define AMDGPU_AMDKFD_USERPTR_BO (1ULL << 63)
@@ -96,7 +97,7 @@ void amdgpu_amdkfd_gpuvm_init_mem_limits(void) · 59 unmodified lines
9697 uint64_t mem;
9798
9899 si_meminfo(&si);
99− mem = si.totalram - si.totalhigh;
100+ mem = si.freeram - si.freehigh;
100101 mem *= si.mem_unit;
101102
102103 spin_lock_init(&kfd_mem_limit.mem_limit_lock);
@@ -412,7 +413,10 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem) · 309 unmodified lines
412413 {
413414 struct amdgpu_device *bo_adev = amdgpu_ttm_adev(mem->bo->tbo.bdev);
414415 bool coherent = mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_COHERENT;
416+ bool uncached = mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED;
415417 uint32_t mapping_flags;
418+ uint64_t pte_flags;
419+ bool snoop = false;
416420
417421 mapping_flags = AMDGPU_VM_PAGE_READABLE;
418422 if (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE)
@@ -433,12 +437,41 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem) · 14 unmodified lines
433437 AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC;
434438 }
435439 break;
440+ case CHIP_ALDEBARAN:
441+ if (coherent && uncached) {
442+ if (adev->gmc.xgmi.connected_to_cpu ||
443+ !(mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM))
444+ snoop = true;
445+ mapping_flags |= AMDGPU_VM_MTYPE_UC;
446+ } else if (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
447+ if (bo_adev == adev) {
448+ mapping_flags |= AMDGPU_VM_MTYPE_RW;
449+ if (adev->gmc.xgmi.connected_to_cpu)
450+ snoop = true;
451+ } else {
452+ mapping_flags |= AMDGPU_VM_MTYPE_NC;
453+ if (amdgpu_xgmi_same_hive(adev, bo_adev))
454+ snoop = true;
455+ }
456+ } else {
457+ snoop = true;
458+ if (adev->gmc.xgmi.connected_to_cpu)
459+ /* system memory uses NC on A+A */
460+ mapping_flags |= AMDGPU_VM_MTYPE_NC;
461+ else
462+ mapping_flags |= coherent ?
463+ AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC;
464+ }
465+ break;
436466 default:
437467 mapping_flags |= coherent ?
438468 AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC;
439469 }
440470
441− return amdgpu_gem_va_map_flags(adev, mapping_flags);
471+ pte_flags = amdgpu_gem_va_map_flags(adev, mapping_flags);
472+ pte_flags |= snoop ? AMDGPU_PTE_SNOOPED : 0;
473+
474+ return pte_flags;
442475 }
443476
444477 /* add_bo_to_vm - Add a BO to a VM
drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c+0−43
@@ -1905,40 +1905,6 @@ static uint32_t cail_reg_read(struct card_info *info, uint32_t reg) · 1904 unmodified lines
19051905 return r;
19061906 }
19071907
1908−/**
1909− * cail_ioreg_write - write IO register
1910− *
1911− * @info: atom card_info pointer
1912− * @reg: IO register offset
1913− * @val: value to write to the pll register
1914− *
1915− * Provides a IO register accessor for the atom interpreter (r4xx+).
1916− */
1917−static void cail_ioreg_write(struct card_info *info, uint32_t reg, uint32_t val)
1918−{
1919− struct amdgpu_device *adev = drm_to_adev(info->dev);
1920−
1921− WREG32_IO(reg, val);
1922−}
1923−
1924−/**
1925− * cail_ioreg_read - read IO register
1926− *
1927− * @info: atom card_info pointer
1928− * @reg: IO register offset
1929− *
1930− * Provides an IO register accessor for the atom interpreter (r4xx+).
1931− * Returns the value of the IO register.
1932− */
1933−static uint32_t cail_ioreg_read(struct card_info *info, uint32_t reg)
1934−{
1935− struct amdgpu_device *adev = drm_to_adev(info->dev);
1936− uint32_t r;
1937−
1938− r = RREG32_IO(reg);
1939− return r;
1940−}
1941−
19421908 static ssize_t amdgpu_atombios_get_vbios_version(struct device *dev,
19431909 struct device_attribute *attr,
19441910 char *buf)
@@ -1998,15 +1964,6 @@ int amdgpu_atombios_init(struct amdgpu_device *adev) · 53 unmodified lines
19981964 atom_card_info->dev = adev_to_drm(adev);
19991965 atom_card_info->reg_read = cail_reg_read;
20001966 atom_card_info->reg_write = cail_reg_write;
2001− /* needed for iio ops */
2002− if (adev->rio_mem) {
2003− atom_card_info->ioreg_read = cail_ioreg_read;
2004− atom_card_info->ioreg_write = cail_ioreg_write;
2005− } else {
2006− DRM_DEBUG("PCI I/O BAR is not found. Using MMIO to access ATOM BIOS\n");
2007− atom_card_info->ioreg_read = cail_reg_read;
2008− atom_card_info->ioreg_write = cail_reg_write;
2009− }
20101967 atom_card_info->mc_read = cail_mc_read;
20111968 atom_card_info->mc_write = cail_mc_write;
20121969 atom_card_info->pll_read = cail_pll_read;
drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c+67−6
@@ -117,12 +117,15 @@ union igp_info { · 116 unmodified lines
117117
118118 union umc_info {
119119 struct atom_umc_info_v3_1 v31;
120+ struct atom_umc_info_v3_2 v32;
121+ struct atom_umc_info_v3_3 v33;
120122 };
121123
122124 union vram_info {
123125 struct atom_vram_info_header_v2_3 v23;
124126 struct atom_vram_info_header_v2_4 v24;
125127 struct atom_vram_info_header_v2_5 v25;
128+ struct atom_vram_info_header_v2_6 v26;
126129 };
127130
128131 union vram_module {
@@ -164,6 +167,7 @@ static int convert_atom_mem_type_to_vram_type(struct amdgpu_device *adev, · 35 unmodified lines
164167 vram_type = AMDGPU_VRAM_TYPE_GDDR5;
165168 break;
166169 case ATOM_DGPU_VRAM_TYPE_HBM2:
170+ case ATOM_DGPU_VRAM_TYPE_HBM2E:
167171 vram_type = AMDGPU_VRAM_TYPE_HBM;
168172 break;
169173 case ATOM_DGPU_VRAM_TYPE_GDDR6:
@@ -315,6 +319,26 @@ amdgpu_atomfirmware_get_vram_info(struct amdgpu_device *adev, · 145 unmodified lines
315319 if (vram_vendor)
316320 *vram_vendor = mem_vendor;
317321 break;
322+ case 6:
323+ if (module_id > vram_info->v26.vram_module_num)
324+ module_id = 0;
325+ vram_module = (union vram_module *)vram_info->v26.vram_module;
326+ while (i < module_id) {
327+ vram_module = (union vram_module *)
328+ ((u8 *)vram_module + vram_module->v9.vram_module_size);
329+ i++;
330+ }
331+ mem_type = vram_module->v9.memory_type;
332+ if (vram_type)
333+ *vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
334+ mem_channel_number = vram_module->v9.channel_num;
335+ mem_channel_width = vram_module->v9.channel_width;
336+ if (vram_width)
337+ *vram_width = mem_channel_number * (1 << mem_channel_width);
338+ mem_vendor = (vram_module->v9.vender_rev_id) & 0xF;
339+ if (vram_vendor)
340+ *vram_vendor = mem_vendor;
341+ break;
318342 default:
319343 return -EINVAL;
320344 }
@@ -337,19 +361,39 @@ bool amdgpu_atomfirmware_mem_ecc_supported(struct amdgpu_device *adev) · 16 unmodified lines
337361 union umc_info *umc_info;
338362 u8 frev, crev;
339363 bool ecc_default_enabled = false;
364+ u8 umc_config;
365+ u32 umc_config1;
340366
341367 index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
342368 umc_info);
343369
344370 if (amdgpu_atom_parse_data_header(mode_info->atom_context,
345371 index, &size, &frev, &crev, &data_offset)) {
346− /* support umc_info 3.1+ */
347− if ((frev == 3 && crev >= 1) || (frev > 3)) {
372+ if (frev == 3) {
348373 umc_info = (union umc_info *)
349374 (mode_info->atom_context->bios + data_offset);
350− ecc_default_enabled =
351− (le32_to_cpu(umc_info->v31.umc_config) &
352− UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ? true : false;
375+ switch (crev) {
376+ case 1:
377+ umc_config = le32_to_cpu(umc_info->v31.umc_config);
378+ ecc_default_enabled =
379+ (umc_config & UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ? true : false;
380+ break;
381+ case 2:
382+ umc_config = le32_to_cpu(umc_info->v32.umc_config);
383+ ecc_default_enabled =
384+ (umc_config & UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ? true : false;
385+ break;
386+ case 3:
387+ umc_config = le32_to_cpu(umc_info->v33.umc_config);
388+ umc_config1 = le32_to_cpu(umc_info->v33.umc_config1);
389+ ecc_default_enabled =
390+ ((umc_config & UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ||
391+ (umc_config1 & UMC_CONFIG1__ENABLE_ECC_CAPABLE)) ? true : false;
392+ break;
393+ default:
394+ /* unsupported crev */
395+ return false;
396+ }
353397 }
354398 }
355399
@@ -479,7 +523,8 @@ int amdgpu_atomfirmware_get_clock_info(struct amdgpu_device *adev) · 123 unmodified lines
479523 }
480524
481525 union gfx_info {
482− struct atom_gfx_info_v2_4 v24;
526+ struct atom_gfx_info_v2_4 v24;
527+ struct atom_gfx_info_v2_7 v27;
483528 };
484529
485530 int amdgpu_atomfirmware_get_gfx_info(struct amdgpu_device *adev)
@@ -514,6 +559,22 @@ int amdgpu_atomfirmware_get_gfx_info(struct amdgpu_device *adev) · 28 unmodified lines
514559 adev->gfx.cu_info.max_scratch_slots_per_cu = gfx_info->v24.gc_max_scratch_slots_per_cu;
515560 adev->gfx.cu_info.lds_size = le16_to_cpu(gfx_info->v24.gc_lds_size);
516561 return 0;
562+ case 7:
563+ adev->gfx.config.max_shader_engines = gfx_info->v27.max_shader_engines;
564+ adev->gfx.config.max_cu_per_sh = gfx_info->v27.max_cu_per_sh;
565+ adev->gfx.config.max_sh_per_se = gfx_info->v27.max_sh_per_se;
566+ adev->gfx.config.max_backends_per_se = gfx_info->v27.max_backends_per_se;
567+ adev->gfx.config.max_texture_channel_caches = gfx_info->v27.max_texture_channel_caches;
568+ adev->gfx.config.max_gprs = le16_to_cpu(gfx_info->v27.gc_num_gprs);
569+ adev->gfx.config.max_gs_threads = gfx_info->v27.gc_num_max_gs_thds;
570+ adev->gfx.config.gs_vgt_table_depth = gfx_info->v27.gc_gs_table_depth;
571+ adev->gfx.config.gs_prim_buffer_depth = le16_to_cpu(gfx_info->v27.gc_gsprim_buff_depth);
572+ adev->gfx.config.double_offchip_lds_buf = gfx_info->v27.gc_double_offchip_lds_buffer;
573+ adev->gfx.cu_info.wave_front_size = le16_to_cpu(gfx_info->v27.gc_wave_size);
574+ adev->gfx.cu_info.max_waves_per_simd = le16_to_cpu(gfx_info->v27.gc_max_waves_per_simd);
575+ adev->gfx.cu_info.max_scratch_slots_per_cu = gfx_info->v27.gc_max_scratch_slots_per_cu;
576+ adev->gfx.cu_info.lds_size = le16_to_cpu(gfx_info->v27.gc_lds_size);
577+ return 0;
517578 default:
518579 return -EINVAL;
519580 }
drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c+10−6
@@ -97,6 +97,10 @@ static bool igp_read_bios_from_vram(struct amdgpu_device *adev) · 96 unmodified lines
9797 if (amdgpu_device_need_post(adev))
9898 return false;
9999
100+ /* FB BAR not enabled */
101+ if (pci_resource_len(adev->pdev, 0) == 0)
102+ return false;
103+
100104 adev->bios = NULL;
101105 vram_base = pci_resource_start(adev->pdev, 0);
102106 bios = ioremap_wc(vram_base, size);
@@ -316,7 +320,7 @@ static bool amdgpu_atrm_get_bios(struct amdgpu_device *adev) · 213 unmodified lines
316320
317321 adev->bios = kmalloc(size, GFP_KERNEL);
318322 if (!adev->bios) {
319− DRM_ERROR("Unable to allocate bios\n");
323+ dev_err(adev->dev, "Unable to allocate bios\n");
320324 return false;
321325 }
322326
@@ -364,7 +368,7 @@ static bool amdgpu_acpi_vfct_bios(struct amdgpu_device *adev) · 41 unmodified lines
364368 return false;
365369 tbl_size = hdr->length;
366370 if (tbl_size < sizeof(UEFI_ACPI_VFCT)) {
367− DRM_ERROR("ACPI VFCT table present but broken (too short #1)\n");
371+ dev_info(adev->dev, "ACPI VFCT table present but broken (too short #1),skipping\n");
368372 return false;
369373 }
370374
@@ -377,13 +381,13 @@ static bool amdgpu_acpi_vfct_bios(struct amdgpu_device *adev) · 6 unmodified lines
377381
378382 offset += sizeof(VFCT_IMAGE_HEADER);
379383 if (offset > tbl_size) {
380− DRM_ERROR("ACPI VFCT image header truncated\n");
384+ dev_info(adev->dev, "ACPI VFCT image header truncated,skipping\n");
381385 return false;
382386 }
383387
384388 offset += vhdr->ImageLength;
385389 if (offset > tbl_size) {
386− DRM_ERROR("ACPI VFCT image truncated\n");
390+ dev_info(adev->dev, "ACPI VFCT image truncated,skipping\n");
387391 return false;
388392 }
389393
@@ -406,7 +410,7 @@ static bool amdgpu_acpi_vfct_bios(struct amdgpu_device *adev) · 16 unmodified lines
406410 }
407411 }
408412
409− DRM_ERROR("ACPI VFCT table present but broken (too short #2)\n");
413+ dev_info(adev->dev, "ACPI VFCT table present but broken (too short #2),skipping\n");
410414 return false;
411415 }
412416 #else
@@ -453,7 +457,7 @@ bool amdgpu_get_bios(struct amdgpu_device *adev) · 40 unmodified lines
453457 goto success;
454458 }
455459
456− DRM_ERROR("Unable to locate a BIOS ROM\n");
460+ dev_err(adev->dev, "Unable to locate a BIOS ROM\n");
457461 return false;
458462
459463 success:
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c+7−7
@@ -117,7 +117,7 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, union drm_amdgpu_cs · 116 unmodified lines
117117 if (cs->in.num_chunks == 0)
118118 return 0;
119119
120− chunk_array = kmalloc_array(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
120+ chunk_array = kvmalloc_array(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
121121 if (!chunk_array)
122122 return -ENOMEM;
123123
@@ -144,7 +144,7 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, union drm_amdgpu_cs · 20 unmodified lines
144144 }
145145
146146 p->nchunks = cs->in.num_chunks;
147− p->chunks = kmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
147+ p->chunks = kvmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
148148 GFP_KERNEL);
149149 if (!p->chunks) {
150150 ret = -ENOMEM;
@@ -238,7 +238,7 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, union drm_amdgpu_cs · 87 unmodified lines
238238
239239 if (p->uf_entry.tv.bo)
240240 p->job->uf_addr = uf_offset;
241− kfree(chunk_array);
241+ kvfree(chunk_array);
242242
243243 /* Use this opportunity to fill in task info for the vm */
244244 amdgpu_vm_set_task_info(vm);
@@ -250,11 +250,11 @@ free_all_kdata: · 5 unmodified lines
250250 free_partial_kdata:
251251 for (; i >= 0; i--)
252252 kvfree(p->chunks[i].kdata);
253− kfree(p->chunks);
253+ kvfree(p->chunks);
254254 p->chunks = NULL;
255255 p->nchunks = 0;
256256 free_chunk:
257− kfree(chunk_array);
257+ kvfree(chunk_array);
258258
259259 return ret;
260260 }
@@ -559,7 +559,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, · 298 unmodified lines
559559 sizeof(struct page *),
560560 GFP_KERNEL | __GFP_ZERO);
561561 if (!e->user_pages) {
562− DRM_ERROR("calloc failure\n");
562+ DRM_ERROR("kvmalloc_array failure\n");
563563 return -ENOMEM;
564564 }
565565
@@ -706,7 +706,7 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, · 140 unmodified lines
706706
707707 for (i = 0; i < parser->nchunks; i++)
708708 kvfree(parser->chunks[i].kdata);
709− kfree(parser->chunks);
709+ kvfree(parser->chunks);
710710 if (parser->job)
711711 amdgpu_job_free(parser->job);
712712 if (parser->uf_entry.tv.bo) {
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c+59−125
@@ -28,7 +28,6 @@ · 27 unmodified lines
2828 #include <linux/uaccess.h>
2929 #include <linux/pm_runtime.h>
3030 #include <linux/poll.h>
31−#include <drm/drm_debugfs.h>
3231
3332 #include "amdgpu.h"
3433 #include "amdgpu_pm.h"
@@ -38,45 +37,6 @@ · 3 unmodified lines
3837 #include "amdgpu_securedisplay.h"
3938 #include "amdgpu_fw_attestation.h"
4039
41−/**
42− * amdgpu_debugfs_add_files - Add simple debugfs entries
43− *
44− * @adev: Device to attach debugfs entries to
45− * @files: Array of function callbacks that respond to reads
46− * @nfiles: Number of callbacks to register
47− *
48− */
49−int amdgpu_debugfs_add_files(struct amdgpu_device *adev,
50− const struct drm_info_list *files,
51− unsigned nfiles)
52−{
53− unsigned i;
54−
55− for (i = 0; i < adev->debugfs_count; i++) {
56− if (adev->debugfs[i].files == files) {
57− /* Already registered */
58− return 0;
59− }
60− }
61−
62− i = adev->debugfs_count + 1;
63− if (i > AMDGPU_DEBUGFS_MAX_COMPONENTS) {
64− DRM_ERROR("Reached maximum number of debugfs components.\n");
65− DRM_ERROR("Report so we increase "
66− "AMDGPU_DEBUGFS_MAX_COMPONENTS.\n");
67− return -EINVAL;
68− }
69− adev->debugfs[adev->debugfs_count].files = files;
70− adev->debugfs[adev->debugfs_count].num_files = nfiles;
71− adev->debugfs_count = i;
72−#if defined(CONFIG_DEBUG_FS)
73− drm_debugfs_create_files(files, nfiles,
74− adev_to_drm(adev)->primary->debugfs_root,
75− adev_to_drm(adev)->primary);
76−#endif
77− return 0;
78−}
79−
8040 int amdgpu_debugfs_wait_dump(struct amdgpu_device *adev)
8141 {
8242 #if defined(CONFIG_DEBUG_FS)
@@ -1228,22 +1188,20 @@ int amdgpu_debugfs_regs_init(struct amdgpu_device *adev) · 1145 unmodified lines
12281188 adev, debugfs_regs[i]);
12291189 if (!i && !IS_ERR_OR_NULL(ent))
12301190 i_size_write(ent->d_inode, adev->rmmio_size);
1231− adev->debugfs_regs[i] = ent;
12321191 }
12331192
12341193 return 0;
12351194 }
12361195
1237−static int amdgpu_debugfs_test_ib(struct seq_file *m, void *data)
1196+static int amdgpu_debugfs_test_ib_show(struct seq_file *m, void *unused)
12381197 {
1239− struct drm_info_node *node = (struct drm_info_node *) m->private;
1240− struct drm_device *dev = node->minor->dev;
1241− struct amdgpu_device *adev = drm_to_adev(dev);
1198+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
1199+ struct drm_device *dev = adev_to_drm(adev);
12421200 int r = 0, i;
12431201
12441202 r = pm_runtime_get_sync(dev->dev);
12451203 if (r < 0) {
1246− pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1204+ pm_runtime_put_autosuspend(dev->dev);
12471205 return r;
12481206 }
12491207
@@ -1285,30 +1243,19 @@ static int amdgpu_debugfs_test_ib(struct seq_file *m, void *data) · 35 unmodified lines
12851243 return 0;
12861244 }
12871245
1288−static int amdgpu_debugfs_get_vbios_dump(struct seq_file *m, void *data)
1246+static int amdgpu_debugfs_evict_vram(void *data, u64 *val)
12891247 {
1290− struct drm_info_node *node = (struct drm_info_node *) m->private;
1291− struct drm_device *dev = node->minor->dev;
1292− struct amdgpu_device *adev = drm_to_adev(dev);
1293−
1294− seq_write(m, adev->bios, adev->bios_size);
1295− return 0;
1296−}
1297−
1298−static int amdgpu_debugfs_evict_vram(struct seq_file *m, void *data)
1299−{
1300− struct drm_info_node *node = (struct drm_info_node *)m->private;
1301− struct drm_device *dev = node->minor->dev;
1302− struct amdgpu_device *adev = drm_to_adev(dev);
1248+ struct amdgpu_device *adev = (struct amdgpu_device *)data;
1249+ struct drm_device *dev = adev_to_drm(adev);
13031250 int r;
13041251
13051252 r = pm_runtime_get_sync(dev->dev);
13061253 if (r < 0) {
1307− pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1254+ pm_runtime_put_autosuspend(dev->dev);
13081255 return r;
13091256 }
13101257
1311− seq_printf(m, "(%d)\n", amdgpu_bo_evict_vram(adev));
1258+ *val = amdgpu_bo_evict_vram(adev);
13121259
13131260 pm_runtime_mark_last_busy(dev->dev);
13141261 pm_runtime_put_autosuspend(dev->dev);
@@ -1316,11 +1263,11 @@ static int amdgpu_debugfs_evict_vram(struct seq_file *m, void *data) · 1 unmodified line
13161263 return 0;
13171264 }
13181265
1319−static int amdgpu_debugfs_evict_gtt(struct seq_file *m, void *data)
1266+
1267+static int amdgpu_debugfs_evict_gtt(void *data, u64 *val)
13201268 {
1321− struct drm_info_node *node = (struct drm_info_node *)m->private;
1322− struct drm_device *dev = node->minor->dev;
1323− struct amdgpu_device *adev = drm_to_adev(dev);
1269+ struct amdgpu_device *adev = (struct amdgpu_device *)data;
1270+ struct drm_device *dev = adev_to_drm(adev);
13241271 struct ttm_resource_manager *man;
13251272 int r;
13261273
@@ -1331,8 +1278,7 @@ static int amdgpu_debugfs_evict_gtt(struct seq_file *m, void *data) · 4 unmodified lines
13311278 }
13321279
13331280 man = ttm_manager_type(&adev->mman.bdev, TTM_PL_TT);
1334− r = ttm_resource_manager_evict_all(&adev->mman.bdev, man);
1335− seq_printf(m, "(%d)\n", r);
1281+ *val = ttm_resource_manager_evict_all(&adev->mman.bdev, man);
13361282
13371283 pm_runtime_mark_last_busy(dev->dev);
13381284 pm_runtime_put_autosuspend(dev->dev);
@@ -1340,10 +1286,11 @@ static int amdgpu_debugfs_evict_gtt(struct seq_file *m, void *data) · 1 unmodified line
13401286 return 0;
13411287 }
13421288
1343−static int amdgpu_debugfs_vm_info(struct seq_file *m, void *data)
1289+
1290+static int amdgpu_debugfs_vm_info_show(struct seq_file *m, void *unused)
13441291 {
1345− struct drm_info_node *node = (struct drm_info_node *)m->private;
1346− struct drm_device *dev = node->minor->dev;
1292+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
1293+ struct drm_device *dev = adev_to_drm(adev);
13471294 struct drm_file *file;
13481295 int r;
13491296
@@ -1369,13 +1316,12 @@ static int amdgpu_debugfs_vm_info(struct seq_file *m, void *data) · 19 unmodified lines
13691316 return r;
13701317 }
13711318
1372−static const struct drm_info_list amdgpu_debugfs_list[] = {
1373− {"amdgpu_vbios", amdgpu_debugfs_get_vbios_dump},
1374− {"amdgpu_test_ib", &amdgpu_debugfs_test_ib},
1375− {"amdgpu_evict_vram", &amdgpu_debugfs_evict_vram},
1376− {"amdgpu_evict_gtt", &amdgpu_debugfs_evict_gtt},
1377− {"amdgpu_vm_info", &amdgpu_debugfs_vm_info},
1378−};
1319+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_test_ib);
1320+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_vm_info);
1321+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_evict_vram_fops, amdgpu_debugfs_evict_vram,
1322+ NULL, "%lld\n");
1323+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_evict_gtt_fops, amdgpu_debugfs_evict_gtt,
1324+ NULL, "%lld\n");
13791325
13801326 static void amdgpu_ib_preempt_fences_swap(struct amdgpu_ring *ring,
13811327 struct dma_fence **fences)
@@ -1586,71 +1532,50 @@ static int amdgpu_debugfs_sclk_set(void *data, u64 val) · 204 unmodified lines
15861532 return 0;
15871533 }
15881534
1589−DEFINE_SIMPLE_ATTRIBUTE(fops_ib_preempt, NULL,
1535+DEFINE_DEBUGFS_ATTRIBUTE(fops_ib_preempt, NULL,
15901536 amdgpu_debugfs_ib_preempt, "%llu\n");
15911537
1592−DEFINE_SIMPLE_ATTRIBUTE(fops_sclk_set, NULL,
1538+DEFINE_DEBUGFS_ATTRIBUTE(fops_sclk_set, NULL,
15931539 amdgpu_debugfs_sclk_set, "%llu\n");
15941540
15951541 int amdgpu_debugfs_init(struct amdgpu_device *adev)
15961542 {
1543+ struct dentry *root = adev_to_drm(adev)->primary->debugfs_root;
1544+ struct dentry *ent;
15971545 int r, i;
15981546
1599− adev->debugfs_preempt =
1600− debugfs_create_file("amdgpu_preempt_ib", 0600,
1601− adev_to_drm(adev)->primary->debugfs_root, adev,
1602− &fops_ib_preempt);
1603− if (!(adev->debugfs_preempt)) {
1547+
1548+
1549+ ent = debugfs_create_file("amdgpu_preempt_ib", 0600, root, adev,
1550+ &fops_ib_preempt);
1551+ if (!ent) {
16041552 DRM_ERROR("unable to create amdgpu_preempt_ib debugsfs file\n");
16051553 return -EIO;
16061554 }
16071555
1608− adev->smu.debugfs_sclk =
1609− debugfs_create_file("amdgpu_force_sclk", 0200,
1610− adev_to_drm(adev)->primary->debugfs_root, adev,
1611− &fops_sclk_set);
1612− if (!(adev->smu.debugfs_sclk)) {
1556+ ent = debugfs_create_file("amdgpu_force_sclk", 0200, root, adev,
1557+ &fops_sclk_set);
1558+ if (!ent) {
16131559 DRM_ERROR("unable to create amdgpu_set_sclk debugsfs file\n");
16141560 return -EIO;
16151561 }
16161562
16171563 /* Register debugfs entries for amdgpu_ttm */
1618− r = amdgpu_ttm_debugfs_init(adev);
1619− if (r) {
1620− DRM_ERROR("Failed to init debugfs\n");
1621− return r;
1622− }
1623−
1624− r = amdgpu_debugfs_pm_init(adev);
1625− if (r) {
1626− DRM_ERROR("Failed to register debugfs file for dpm!\n");
1627− return r;
1628− }
1629−
1630− if (amdgpu_debugfs_sa_init(adev)) {
1631− dev_err(adev->dev, "failed to register debugfs file for SA\n");
1632− }
1633−
1634− if (amdgpu_debugfs_fence_init(adev))
1635− dev_err(adev->dev, "fence debugfs file creation failed\n");
1636−
1637− r = amdgpu_debugfs_gem_init(adev);
1638− if (r)
1639− DRM_ERROR("registering gem debugfs failed (%d).\n", r);
1564+ amdgpu_ttm_debugfs_init(adev);
1565+ amdgpu_debugfs_pm_init(adev);
1566+ amdgpu_debugfs_sa_init(adev);
1567+ amdgpu_debugfs_fence_init(adev);
1568+ amdgpu_debugfs_gem_init(adev);
16401569
16411570 r = amdgpu_debugfs_regs_init(adev);
16421571 if (r)
16431572 DRM_ERROR("registering register debugfs failed (%d).\n", r);
16441573
1645− r = amdgpu_debugfs_firmware_init(adev);
1646− if (r)
1647− DRM_ERROR("registering firmware debugfs failed (%d).\n", r);
1574+ amdgpu_debugfs_firmware_init(adev);
16481575
16491576 #if defined(CONFIG_DRM_AMD_DC)
1650− if (amdgpu_device_has_dc_support(adev)) {
1651− if (dtn_debugfs_init(adev))
1652− DRM_ERROR("amdgpu: failed initialize dtn debugfs support.\n");
1653− }
1577+ if (amdgpu_device_has_dc_support(adev))
1578+ dtn_debugfs_init(adev);
16541579 #endif
16551580
16561581 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
@@ -1665,17 +1590,26 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev) · 8 unmodified lines
16651590 }
16661591
16671592 amdgpu_ras_debugfs_create_all(adev);
1668−
16691593 amdgpu_debugfs_autodump_init(adev);
1670−
16711594 amdgpu_rap_debugfs_init(adev);
1672−
16731595 amdgpu_securedisplay_debugfs_init(adev);
1674−
16751596 amdgpu_fw_attestation_debugfs_init(adev);
16761597
1677− return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_list,
1678− ARRAY_SIZE(amdgpu_debugfs_list));
1598+ debugfs_create_file("amdgpu_evict_vram", 0444, root, adev,
1599+ &amdgpu_evict_vram_fops);
1600+ debugfs_create_file("amdgpu_evict_gtt", 0444, root, adev,
1601+ &amdgpu_evict_gtt_fops);
1602+ debugfs_create_file("amdgpu_test_ib", 0444, root, adev,
1603+ &amdgpu_debugfs_test_ib_fops);
1604+ debugfs_create_file("amdgpu_vm_info", 0444, root, adev,
1605+ &amdgpu_debugfs_vm_info_fops);
1606+
1607+ adev->debugfs_vbios_blob.data = adev->bios;
1608+ adev->debugfs_vbios_blob.size = adev->bios_size;
1609+ debugfs_create_blob("amdgpu_vbios", 0444, root,
1610+ &adev->debugfs_vbios_blob);
1611+
1612+ return 0;
16791613 }
16801614
16811615 #else
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h+3−11
@@ -26,11 +26,6 @@ · 25 unmodified lines
2626 /*
2727 * Debugfs
2828 */
29−struct amdgpu_debugfs {
30− const struct drm_info_list *files;
31− unsigned num_files;
32−};
33−
3429 struct amdgpu_autodump {
3530 struct completion dumping;
3631 struct wait_queue_head gpu_hang;
@@ -39,10 +34,7 @@ struct amdgpu_autodump { · 2 unmodified lines
3934 int amdgpu_debugfs_regs_init(struct amdgpu_device *adev);
4035 int amdgpu_debugfs_init(struct amdgpu_device *adev);
4136 void amdgpu_debugfs_fini(struct amdgpu_device *adev);
42−int amdgpu_debugfs_add_files(struct amdgpu_device *adev,
43− const struct drm_info_list *files,
44− unsigned nfiles);
45−int amdgpu_debugfs_fence_init(struct amdgpu_device *adev);
46−int amdgpu_debugfs_firmware_init(struct amdgpu_device *adev);
47−int amdgpu_debugfs_gem_init(struct amdgpu_device *adev);
37+void amdgpu_debugfs_fence_init(struct amdgpu_device *adev);
38+void amdgpu_debugfs_firmware_init(struct amdgpu_device *adev);
39+void amdgpu_debugfs_gem_init(struct amdgpu_device *adev);
4840 int amdgpu_debugfs_wait_dump(struct amdgpu_device *adev);
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c+162−107
@@ -110,6 +110,7 @@ const char *amdgpu_asic_name[] = { · 109 unmodified lines
110110 "RAVEN",
111111 "ARCTURUS",
112112 "RENOIR",
113+ "ALDEBARAN",
113114 "NAVI10",
114115 "NAVI14",
115116 "NAVI12",
@@ -464,49 +465,6 @@ void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev, · 348 unmodified lines
464465 }
465466 }
466467
467−/**
468− * amdgpu_io_rreg - read an IO register
469− *
470− * @adev: amdgpu_device pointer
471− * @reg: dword aligned register offset
472− *
473− * Returns the 32 bit value from the offset specified.
474− */
475−u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg)
476−{
477− if (adev->in_pci_err_recovery)
478− return 0;
479−
480− if ((reg * 4) < adev->rio_mem_size)
481− return ioread32(adev->rio_mem + (reg * 4));
482− else {
483− iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4));
484− return ioread32(adev->rio_mem + (mmMM_DATA * 4));
485− }
486−}
487−
488−/**
489− * amdgpu_io_wreg - write to an IO register
490− *
491− * @adev: amdgpu_device pointer
492− * @reg: dword aligned register offset
493− * @v: 32 bit value to write to the register
494− *
495− * Writes the value specified to the offset specified.
496− */
497−void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
498−{
499− if (adev->in_pci_err_recovery)
500− return;
501−
502− if ((reg * 4) < adev->rio_mem_size)
503− iowrite32(v, adev->rio_mem + (reg * 4));
504− else {
505− iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4));
506− iowrite32(v, adev->rio_mem + (mmMM_DATA * 4));
507− }
508−}
509−
510468 /**
511469 * amdgpu_mm_rdoorbell - read a doorbell dword
512470 *
@@ -1223,6 +1181,10 @@ bool amdgpu_device_need_post(struct amdgpu_device *adev) · 710 unmodified lines
12231181 }
12241182 }
12251183
1184+ /* Don't post if we need to reset whole hive on init */
1185+ if (adev->gmc.xgmi.pending_reset)
1186+ return false;
1187+
12261188 if (adev->has_hw_reset) {
12271189 adev->has_hw_reset = false;
12281190 return true;
@@ -1810,6 +1772,7 @@ static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev) · 581 unmodified lines
18101772 case CHIP_CARRIZO:
18111773 case CHIP_STONEY:
18121774 case CHIP_VEGA20:
1775+ case CHIP_ALDEBARAN:
18131776 case CHIP_SIENNA_CICHLID:
18141777 case CHIP_NAVY_FLOUNDER:
18151778 case CHIP_DIMGREY_CAVEFISH:
@@ -2010,6 +1973,7 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) · 194 unmodified lines
20101973 case CHIP_RAVEN:
20111974 case CHIP_ARCTURUS:
20121975 case CHIP_RENOIR:
1976+ case CHIP_ALDEBARAN:
20131977 if (adev->flags & AMD_IS_APU)
20141978 adev->family = AMDGPU_FAMILY_RV;
20151979 else
@@ -2045,6 +2009,8 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) · 29 unmodified lines
20452009 adev->pm.pp_feature = amdgpu_pp_feature_mask;
20462010 if (amdgpu_sriov_vf(adev) || sched_policy == KFD_SCHED_POLICY_NO_HWS)
20472011 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
2012+ if (amdgpu_sriov_vf(adev) && adev->asic_type == CHIP_SIENNA_CICHLID)
2013+ adev->pm.pp_feature &= ~PP_OVERDRIVE_MASK;
20482014
20492015 for (i = 0; i < adev->num_ip_blocks; i++) {
20502016 if ((amdgpu_ip_block_mask & (1 << i)) == 0) {
@@ -2149,6 +2115,9 @@ static int amdgpu_device_fw_loading(struct amdgpu_device *adev) · 98 unmodified lines
21492115 if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_PSP)
21502116 continue;
21512117
2118+ if (!adev->ip_blocks[i].status.sw)
2119+ continue;
2120+
21522121 /* no need to do the fw loading again if already done*/
21532122 if (adev->ip_blocks[i].status.hw == true)
21542123 break;
@@ -2289,7 +2258,10 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev) · 134 unmodified lines
22892258
22902259 if (adev->gmc.xgmi.num_physical_nodes > 1)
22912260 amdgpu_xgmi_add_device(adev);
2292− amdgpu_amdkfd_device_init(adev);
2261+
2262+ /* Don't init kfd if whole hive need to be reset during init */
2263+ if (!adev->gmc.xgmi.pending_reset)
2264+ amdgpu_amdkfd_device_init(adev);
22932265
22942266 amdgpu_fru_get_product_info(adev);
22952267
@@ -2498,6 +2470,11 @@ static int amdgpu_device_ip_late_init(struct amdgpu_device *adev) · 202 unmodified lines
24982470 if (r)
24992471 DRM_ERROR("enable mgpu fan boost failed (%d).\n", r);
25002472
2473+ /* For XGMI + passthrough configuration on arcturus, enable light SBR */
2474+ if (adev->asic_type == CHIP_ARCTURUS &&
2475+ amdgpu_passthrough(adev) &&
2476+ adev->gmc.xgmi.num_physical_nodes > 1)
2477+ smu_set_light_sbr(&adev->smu, true);
25012478
25022479 if (adev->gmc.xgmi.num_physical_nodes > 1) {
25032480 mutex_lock(&mgpu_info.mutex);
@@ -2734,6 +2711,16 @@ static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev) · 230 unmodified lines
27342711 adev->ip_blocks[i].status.hw = false;
27352712 continue;
27362713 }
2714+
2715+ /* skip unnecessary suspend if we do not initialize them yet */
2716+ if (adev->gmc.xgmi.pending_reset &&
2717+ !(adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
2718+ adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC ||
2719+ adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
2720+ adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH)) {
2721+ adev->ip_blocks[i].status.hw = false;
2722+ continue;
2723+ }
27372724 /* XXX handle errors */
27382725 r = adev->ip_blocks[i].version->funcs->suspend(adev);
27392726 /* XXX handle errors */
@@ -2753,7 +2740,6 @@ static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev) · 13 unmodified lines
27532740 }
27542741 }
27552742 }
2756− adev->ip_blocks[i].status.hw = false;
27572743 }
27582744
27592745 return 0;
@@ -2774,8 +2760,10 @@ int amdgpu_device_ip_suspend(struct amdgpu_device *adev) · 14 unmodified lines
27742760 {
27752761 int r;
27762762
2777− if (amdgpu_sriov_vf(adev))
2763+ if (amdgpu_sriov_vf(adev)) {
2764+ amdgpu_virt_fini_data_exchange(adev);
27782765 amdgpu_virt_request_full_gpu(adev, false);
2766+ }
27792767
27802768 r = amdgpu_device_ip_suspend_phase1(adev);
27812769 if (r)
@@ -3257,7 +3245,6 @@ int amdgpu_device_init(struct amdgpu_device *adev, · 475 unmodified lines
32573245
32583246 /* mutex initialization are all done here so we
32593247 * can recall function without having locking issues */
3260− atomic_set(&adev->irq.ih.lock, 0);
32613248 mutex_init(&adev->firmware.mutex);
32623249 mutex_init(&adev->pm.mutex);
32633250 mutex_init(&adev->gfx.gpu_clock_mutex);
@@ -3290,6 +3277,8 @@ int amdgpu_device_init(struct amdgpu_device *adev, · 26 unmodified lines
32903277 INIT_LIST_HEAD(&adev->shadow_list);
32913278 mutex_init(&adev->shadow_list_lock);
32923279
3280+ INIT_LIST_HEAD(&adev->reset_list);
3281+
32933282 INIT_DELAYED_WORK(&adev->delayed_init_work,
32943283 amdgpu_device_delayed_init_work_handler);
32953284 INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
@@ -3328,17 +3317,6 @@ int amdgpu_device_init(struct amdgpu_device *adev, · 32 unmodified lines
33283317 DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base);
33293318 DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size);
33303319
3331− /* io port mapping */
3332− for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3333− if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) {
3334− adev->rio_mem_size = pci_resource_len(adev->pdev, i);
3335− adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size);
3336− break;
3337− }
3338− }
3339− if (adev->rio_mem == NULL)
3340− DRM_INFO("PCI I/O BAR is not found.\n");
3341−
33423320 /* enable PCIE atomic ops */
33433321 r = pci_enable_atomic_ops_to_root(adev->pdev,
33443322 PCI_EXP_DEVCAP2_ATOMIC_COMP32 |
@@ -3405,10 +3383,28 @@ int amdgpu_device_init(struct amdgpu_device *adev, · 60 unmodified lines
34053383 * E.g., driver was not cleanly unloaded previously, etc.
34063384 */
34073385 if (!amdgpu_sriov_vf(adev) && amdgpu_asic_need_reset_on_init(adev)) {
3408− r = amdgpu_asic_reset(adev);
3409− if (r) {
3410− dev_err(adev->dev, "asic reset on init failed\n");
3411− goto failed;
3386+ if (adev->gmc.xgmi.num_physical_nodes) {
3387+ dev_info(adev->dev, "Pending hive reset.\n");
3388+ adev->gmc.xgmi.pending_reset = true;
3389+ /* Only need to init necessary block for SMU to handle the reset */
3390+ for (i = 0; i < adev->num_ip_blocks; i++) {
3391+ if (!adev->ip_blocks[i].status.valid)
3392+ continue;
3393+ if (!(adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
3394+ adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
3395+ adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH ||
3396+ adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC)) {
3397+ DRM_DEBUG("IP %s disabled for hw_init.\n",
3398+ adev->ip_blocks[i].version->funcs->name);
3399+ adev->ip_blocks[i].status.hw = true;
3400+ }
3401+ }
3402+ } else {
3403+ r = amdgpu_asic_reset(adev);
3404+ if (r) {
3405+ dev_err(adev->dev, "asic reset on init failed\n");
3406+ goto failed;
3407+ }
34123408 }
34133409 }
34143410
@@ -3474,11 +3470,11 @@ fence_driver_init: · 59 unmodified lines
34743470 adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME;
34753471 adev->virt.ops = NULL;
34763472 r = -EAGAIN;
3477− goto failed;
3473+ goto release_ras_con;
34783474 }
34793475 dev_err(adev->dev, "amdgpu_device_ip_init failed\n");
34803476 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
3481− goto failed;
3477+ goto release_ras_con;
34823478 }
34833479
34843480 dev_info(adev->dev,
@@ -3539,19 +3535,19 @@ fence_driver_init: · 54 unmodified lines
35393535 /* enable clockgating, etc. after ib tests, etc. since some blocks require
35403536 * explicit gating rather than handling it automatically.
35413537 */
3542− r = amdgpu_device_ip_late_init(adev);
3543− if (r) {
3544− dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n");
3545− amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r);
3546− goto failed;
3538+ if (!adev->gmc.xgmi.pending_reset) {
3539+ r = amdgpu_device_ip_late_init(adev);
3540+ if (r) {
3541+ dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n");
3542+ amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r);
3543+ goto release_ras_con;
3544+ }
3545+ /* must succeed. */
3546+ amdgpu_ras_resume(adev);
3547+ queue_delayed_work(system_wq, &adev->delayed_init_work,
3548+ msecs_to_jiffies(AMDGPU_RESUME_MS));
35473549 }
35483550
3549− /* must succeed. */
3550− amdgpu_ras_resume(adev);
3551−
3552− queue_delayed_work(system_wq, &adev->delayed_init_work,
3553− msecs_to_jiffies(AMDGPU_RESUME_MS));
3554−
35553551 if (amdgpu_sriov_vf(adev))
35563552 flush_delayed_work(&adev->delayed_init_work);
35573553
@@ -3568,8 +3564,15 @@ fence_driver_init: · 10 unmodified lines
35683564 if (amdgpu_device_cache_pci_state(adev->pdev))
35693565 pci_restore_state(pdev);
35703566
3567+ if (adev->gmc.xgmi.pending_reset)
3568+ queue_delayed_work(system_wq, &mgpu_info.delayed_reset_work,
3569+ msecs_to_jiffies(AMDGPU_RESUME_MS));
3570+
35713571 return 0;
35723572
3573+release_ras_con:
3574+ amdgpu_release_ras_context(adev);
3575+
35733576 failed:
35743577 amdgpu_vf_error_trans_all(adev);
35753578 if (atpx)
@@ -3594,6 +3597,7 @@ void amdgpu_device_fini(struct amdgpu_device *adev) · 18 unmodified lines
35943597 {
35953598 dev_info(adev->dev, "amdgpu: finishing device.\n");
35963599 flush_delayed_work(&adev->delayed_init_work);
3600+ ttm_bo_lock_delayed_workqueue(&adev->mman.bdev);
35973601 adev->shutdown = true;
35983602
35993603 kfree(adev->pci_state);
@@ -3640,9 +3644,6 @@ void amdgpu_device_fini(struct amdgpu_device *adev) · 40 unmodified lines
36403644 vga_switcheroo_fini_domain_pm_ops(adev->dev);
36413645 if ((adev->pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA)
36423646 vga_client_register(adev->pdev, NULL, NULL, NULL);
3643− if (adev->rio_mem)
3644− pci_iounmap(adev->pdev, adev->rio_mem);
3645− adev->rio_mem = NULL;
36463647 iounmap(adev->rmmio);
36473648 adev->rmmio = NULL;
36483649 amdgpu_device_doorbell_fini(adev);
@@ -4237,15 +4238,56 @@ disabled: · 588 unmodified lines
42374238 return false;
42384239 }
42394240
4241+int amdgpu_device_mode1_reset(struct amdgpu_device *adev)
4242+{
4243+ u32 i;
4244+ int ret = 0;
4245+
4246+ amdgpu_atombios_scratch_regs_engine_hung(adev, true);
4247+
4248+ dev_info(adev->dev, "GPU mode1 reset\n");
4249+
4250+ /* disable BM */
4251+ pci_clear_master(adev->pdev);
4252+
4253+ amdgpu_device_cache_pci_state(adev->pdev);
4254+
4255+ if (amdgpu_dpm_is_mode1_reset_supported(adev)) {
4256+ dev_info(adev->dev, "GPU smu mode1 reset\n");
4257+ ret = amdgpu_dpm_mode1_reset(adev);
4258+ } else {
4259+ dev_info(adev->dev, "GPU psp mode1 reset\n");
4260+ ret = psp_gpu_reset(adev);
4261+ }
4262+
4263+ if (ret)
4264+ dev_err(adev->dev, "GPU mode1 reset failed\n");
4265+
4266+ amdgpu_device_load_pci_state(adev->pdev);
4267+
4268+ /* wait for asic to come out of reset */
4269+ for (i = 0; i < adev->usec_timeout; i++) {
4270+ u32 memsize = adev->nbio.funcs->get_memsize(adev);
42404271
4241−static int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
4242− struct amdgpu_job *job,
4243− bool *need_full_reset_arg)
4272+ if (memsize != 0xffffffff)
4273+ break;
4274+ udelay(1);
4275+ }
4276+
4277+ amdgpu_atombios_scratch_regs_engine_hung(adev, false);
4278+ return ret;
4279+}
4280+
4281+int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
4282+ struct amdgpu_job *job,
4283+ bool *need_full_reset_arg)
42444284 {
42454285 int i, r = 0;
42464286 bool need_full_reset = *need_full_reset_arg;
42474287
4248− amdgpu_debugfs_wait_dump(adev);
4288+ /* no need to dump if device is not in good state during probe period */
4289+ if (!adev->gmc.xgmi.pending_reset)
4290+ amdgpu_debugfs_wait_dump(adev);
42494291
42504292 if (amdgpu_sriov_vf(adev)) {
42514293 /* stop the data exchange thread */
@@ -4291,23 +4333,24 @@ static int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, · 39 unmodified lines
42914333 return r;
42924334 }
42934335
4294−static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive,
4295− struct list_head *device_list_handle,
4296− bool *need_full_reset_arg,
4297− bool skip_hw_reset)
4336+int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive,
4337+ struct list_head *device_list_handle,
4338+ bool *need_full_reset_arg,
4339+ bool skip_hw_reset)
42984340 {
42994341 struct amdgpu_device *tmp_adev = NULL;
43004342 bool need_full_reset = *need_full_reset_arg, vram_lost = false;
43014343 int r = 0;
43024344
43034345 /*
4304− * ASIC reset has to be done on all HGMI hive nodes ASAP
4346+ * ASIC reset has to be done on all XGMI hive nodes ASAP
43054347 * to allow proper links negotiation in FW (within 1 sec)
43064348 */
43074349 if (!skip_hw_reset && need_full_reset) {
4308− list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
4350+ list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
43094351 /* For XGMI run all resets in parallel to speed up the process */
43104352 if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
4353+ tmp_adev->gmc.xgmi.pending_reset = false;
43114354 if (!queue_work(system_unbound_wq, &tmp_adev->xgmi_reset_work))
43124355 r = -EALREADY;
43134356 } else
@@ -4322,8 +4365,7 @@ static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, · 8 unmodified lines
43224365
43234366 /* For XGMI wait for all resets to complete before proceed */
43244367 if (!r) {
4325− list_for_each_entry(tmp_adev, device_list_handle,
4326− gmc.xgmi.head) {
4368+ list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
43274369 if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
43284370 flush_work(&tmp_adev->xgmi_reset_work);
43294371 r = tmp_adev->asic_reset_res;
@@ -4335,7 +4377,7 @@ static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, · 5 unmodified lines
43354377 }
43364378
43374379 if (!r && amdgpu_ras_intr_triggered()) {
4338− list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
4380+ list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
43394381 if (tmp_adev->mmhub.funcs &&
43404382 tmp_adev->mmhub.funcs->reset_ras_error_count)
43414383 tmp_adev->mmhub.funcs->reset_ras_error_count(tmp_adev);
@@ -4344,13 +4386,13 @@ static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, · 2 unmodified lines
43444386 amdgpu_ras_intr_cleared();
43454387 }
43464388
4347− list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
4389+ list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
43484390 if (need_full_reset) {
43494391 /* post card */
4350− if (amdgpu_device_asic_init(tmp_adev))
4392+ r = amdgpu_device_asic_init(tmp_adev);
4393+ if (r) {
43514394 dev_warn(tmp_adev->dev, "asic atom init failed!");
4352−
4353− if (!r) {
4395+ } else {
43544396 dev_info(tmp_adev->dev, "GPU reset succeeded, trying to resume\n");
43554397 r = amdgpu_device_ip_resume_phase1(tmp_adev);
43564398 if (r)
@@ -4383,6 +4425,9 @@ static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, · 26 unmodified lines
43834425 */
43844426 amdgpu_register_gpu_instance(tmp_adev);
43854427
4428+ if (!hive && tmp_adev->gmc.xgmi.num_physical_nodes > 1)
4429+ amdgpu_xgmi_add_device(tmp_adev);
4430+
43864431 r = amdgpu_device_ip_late_init(tmp_adev);
43874432 if (r)
43884433 goto out;
@@ -4399,7 +4444,7 @@ static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, · 10 unmodified lines
43994444 * bad_page_threshold value to fix this once
44004445 * probing driver again.
44014446 */
4402− if (!amdgpu_ras_check_err_threshold(tmp_adev)) {
4447+ if (!amdgpu_ras_eeprom_check_err_threshold(tmp_adev)) {
44034448 /* must succeed. */
44044449 amdgpu_ras_resume(tmp_adev);
44054450 } else {
@@ -4449,7 +4494,6 @@ static bool amdgpu_device_lock_adev(struct amdgpu_device *adev, · 43 unmodified lines
44494494 down_write(&adev->reset_sem);
44504495 }
44514496
4452− atomic_inc(&adev->gpu_reset_counter);
44534497 switch (amdgpu_asic_reset_method(adev)) {
44544498 case AMD_RESET_METHOD_MODE1:
44554499 adev->mp1_state = PP_MP1_STATE_SHUTDOWN;
@@ -4656,16 +4700,18 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, · 200 unmodified lines
46564700 */
46574701 INIT_LIST_HEAD(&device_list);
46584702 if (adev->gmc.xgmi.num_physical_nodes > 1) {
4659− if (!list_is_first(&adev->gmc.xgmi.head, &hive->device_list))
4660− list_rotate_to_front(&adev->gmc.xgmi.head, &hive->device_list);
4661− device_list_handle = &hive->device_list;
4703+ list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head)
4704+ list_add_tail(&tmp_adev->reset_list, &device_list);
4705+ if (!list_is_first(&adev->reset_list, &device_list))
4706+ list_rotate_to_front(&adev->reset_list, &device_list);
4707+ device_list_handle = &device_list;
46624708 } else {
4663− list_add_tail(&adev->gmc.xgmi.head, &device_list);
4709+ list_add_tail(&adev->reset_list, &device_list);
46644710 device_list_handle = &device_list;
46654711 }
46664712
46674713 /* block all schedulers and reset given job's ring */
4668− list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
4714+ list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
46694715 /*
46704716 * Try to put the audio codec into suspend state
46714717 * before gpu reset started.
@@ -4710,6 +4756,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, · 38 unmodified lines
47104756 if (need_emergency_restart)
47114757 amdgpu_job_stop_all_jobs_on_sched(&ring->sched);
47124758 }
4759+ atomic_inc(&tmp_adev->gpu_reset_counter);
47134760 }
47144761
47154762 if (need_emergency_restart)
@@ -4729,7 +4776,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, · 13 unmodified lines
47294776 }
47304777
47314778 retry: /* Rest of adevs pre asic reset from XGMI hive. */
4732− list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
4779+ list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
47334780 r = amdgpu_device_pre_asic_reset(tmp_adev,
47344781 (tmp_adev == adev) ? job : NULL,
47354782 &need_full_reset);
@@ -4756,7 +4803,7 @@ retry: /* Rest of adevs pre asic reset from XGMI hive. */ · 20 unmodified lines
47564803 skip_hw_reset:
47574804
47584805 /* Post ASIC reset for all devs .*/
4759− list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
4806+ list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
47604807
47614808 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
47624809 struct amdgpu_ring *ring = tmp_adev->rings[i];
@@ -4787,10 +4834,17 @@ skip_hw_reset: · 24 unmodified lines
47874834 }
47884835
47894836 skip_sched_resume:
4790− list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
4791− /*unlock kfd: SRIOV would do it separately */
4837+ list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
4838+ /* unlock kfd: SRIOV would do it separately */
47924839 if (!need_emergency_restart && !amdgpu_sriov_vf(tmp_adev))
47934840 amdgpu_amdkfd_post_reset(tmp_adev);
4841+
4842+ /* kfd_post_reset will do nothing if kfd device is not initialized,
4843+ * need to bring up kfd here if it's not be initialized before
4844+ */
4845+ if (!adev->kfd.init_complete)
4846+ amdgpu_amdkfd_device_init(adev);
4847+
47944848 if (audio_suspended)
47954849 amdgpu_device_resume_display_audio(tmp_adev);
47964850 amdgpu_device_unlock_adev(tmp_adev);
@@ -5052,6 +5106,7 @@ pci_ers_result_t amdgpu_pci_error_detected(struct pci_dev *pdev, pci_channel_sta · 255 unmodified lines
50525106
50535107 drm_sched_stop(&ring->sched, NULL);
50545108 }
5109+ atomic_inc(&adev->gpu_reset_counter);
50555110 return PCI_ERS_RESULT_NEED_RESET;
50565111 case pci_channel_io_perm_failure:
50575112 /* Permanent error, prepare for device removal */
@@ -5100,7 +5155,7 @@ pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev) · 42 unmodified lines
51005155 DRM_INFO("PCI error: slot reset callback!!\n");
51015156
51025157 INIT_LIST_HEAD(&device_list);
5103− list_add_tail(&adev->gmc.xgmi.head, &device_list);
5158+ list_add_tail(&adev->reset_list, &device_list);
51045159
51055160 /* wait for asic to come out of reset */
51065161 msleep(500);
drivers/gpu/drm/amd/amdgpu/amdgpu_display.c+58−14
@@ -870,17 +870,62 @@ static int amdgpu_display_get_fb_info(const struct amdgpu_framebuffer *amdgpu_fb · 869 unmodified lines
870870 return r;
871871 }
872872
873+int amdgpu_display_gem_fb_init(struct drm_device *dev,
874+ struct amdgpu_framebuffer *rfb,
875+ const struct drm_mode_fb_cmd2 *mode_cmd,
876+ struct drm_gem_object *obj)
877+{
878+ int ret;
879+
880+ rfb->base.obj[0] = obj;
881+ drm_helper_mode_fill_fb_struct(dev, &rfb->base, mode_cmd);
882+ ret = drm_framebuffer_init(dev, &rfb->base, &amdgpu_fb_funcs);
883+ if (ret)
884+ goto err;
885+
886+ ret = amdgpu_display_framebuffer_init(dev, rfb, mode_cmd, obj);
887+ if (ret)
888+ goto err;
889+
890+ return 0;
891+err:
892+ drm_err(dev, "Failed to init gem fb: %d\n", ret);
893+ rfb->base.obj[0] = NULL;
894+ return ret;
895+}
896+
897+int amdgpu_display_gem_fb_verify_and_init(
898+ struct drm_device *dev, struct amdgpu_framebuffer *rfb,
899+ struct drm_file *file_priv, const struct drm_mode_fb_cmd2 *mode_cmd,
900+ struct drm_gem_object *obj)
901+{
902+ int ret;
903+
904+ rfb->base.obj[0] = obj;
905+
906+ /* Verify that bo size can fit the fb size. */
907+ ret = drm_gem_fb_init_with_funcs(dev, &rfb->base, file_priv, mode_cmd,
908+ &amdgpu_fb_funcs);
909+ if (ret)
910+ goto err;
911+
912+ ret = amdgpu_display_framebuffer_init(dev, rfb, mode_cmd, obj);
913+ if (ret)
914+ goto err;
915+
916+ return 0;
917+err:
918+ drm_err(dev, "Failed to verify and init gem fb: %d\n", ret);
919+ rfb->base.obj[0] = NULL;
920+ return ret;
921+}
922+
873923 int amdgpu_display_framebuffer_init(struct drm_device *dev,
874924 struct amdgpu_framebuffer *rfb,
875925 const struct drm_mode_fb_cmd2 *mode_cmd,
876926 struct drm_gem_object *obj)
877927 {
878928 int ret, i;
879− rfb->base.obj[0] = obj;
880− drm_helper_mode_fill_fb_struct(dev, &rfb->base, mode_cmd);
881− ret = drm_framebuffer_init(dev, &rfb->base, &amdgpu_fb_funcs);
882− if (ret)
883− goto fail;
884929
885930 /*
886931 * This needs to happen before modifier conversion as that might change
@@ -891,13 +936,13 @@ int amdgpu_display_framebuffer_init(struct drm_device *dev, · 4 unmodified lines
891936 drm_dbg_kms(dev, "Plane 0 and %d have different BOs: %u vs. %u\n",
892937 i, mode_cmd->handles[0], mode_cmd->handles[i]);
893938 ret = -EINVAL;
894− goto fail;
939+ return ret;
895940 }
896941 }
897942
898943 ret = amdgpu_display_get_fb_info(rfb, &rfb->tiling_flags, &rfb->tmz_surface);
899944 if (ret)
900− goto fail;
945+ return ret;
901946
902947 if (dev->mode_config.allow_fb_modifiers &&
903948 !(rfb->base.flags & DRM_MODE_FB_MODIFIERS)) {
@@ -905,20 +950,17 @@ int amdgpu_display_framebuffer_init(struct drm_device *dev, · 1 unmodified line
905950 if (ret) {
906951 drm_dbg_kms(dev, "Failed to convert tiling flags 0x%llX to a modifier",
907952 rfb->tiling_flags);
908− goto fail;
953+ return ret;
909954 }
910955 }
911956
912957 for (i = 1; i < rfb->base.format->num_planes; ++i) {
958+ drm_gem_object_get(rfb->base.obj[0]);
959+ drm_gem_object_put(rfb->base.obj[i]);
913960 rfb->base.obj[i] = rfb->base.obj[0];
914− drm_gem_object_get(rfb->base.obj[i]);
915961 }
916962
917963 return 0;
918−
919−fail:
920− rfb->base.obj[0] = NULL;
921− return ret;
922964 }
923965
924966 struct drm_framebuffer *
@@ -953,13 +995,15 @@ amdgpu_display_user_framebuffer_create(struct drm_device *dev, · 28 unmodified lines
953995 return ERR_PTR(-ENOMEM);
954996 }
955997
956− ret = amdgpu_display_framebuffer_init(dev, amdgpu_fb, mode_cmd, obj);
998+ ret = amdgpu_display_gem_fb_verify_and_init(dev, amdgpu_fb, file_priv,
999+ mode_cmd, obj);
9571000 if (ret) {
9581001 kfree(amdgpu_fb);
9591002 drm_gem_object_put(obj);
9601003 return ERR_PTR(ret);
9611004 }
9621005
1006+ drm_gem_object_put(obj);
9631007 return &amdgpu_fb->base;
9641008 }
9651009
drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c+12−17
@@ -321,17 +321,12 @@ static void amdgpu_dma_buf_unmap(struct dma_buf_attachment *attach, · 320 unmodified lines
321321 struct sg_table *sgt,
322322 enum dma_data_direction dir)
323323 {
324− struct dma_buf *dma_buf = attach->dmabuf;
325− struct drm_gem_object *obj = dma_buf->priv;
326− struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
327− struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
328−
329324 if (sgt->sgl->page_link) {
330325 dma_unmap_sgtable(attach->dev, sgt, dir, 0);
331326 sg_free_table(sgt);
332327 kfree(sgt);
333328 } else {
334− amdgpu_vram_mgr_free_sgt(adev, attach->dev, dir, sgt);
329+ amdgpu_vram_mgr_free_sgt(attach->dev, dir, sgt);
335330 }
336331 }
337332
@@ -434,22 +429,22 @@ amdgpu_dma_buf_create_obj(struct drm_device *dev, struct dma_buf *dma_buf) · 96 unmodified lines
434429 {
435430 struct dma_resv *resv = dma_buf->resv;
436431 struct amdgpu_device *adev = drm_to_adev(dev);
437− struct amdgpu_bo *bo;
438− struct amdgpu_bo_param bp;
439432 struct drm_gem_object *gobj;
433+ struct amdgpu_bo *bo;
434+ uint64_t flags = 0;
440435 int ret;
441436
442− memset(&bp, 0, sizeof(bp));
443− bp.size = dma_buf->size;
444− bp.byte_align = PAGE_SIZE;
445− bp.domain = AMDGPU_GEM_DOMAIN_CPU;
446− bp.flags = 0;
447− bp.type = ttm_bo_type_sg;
448− bp.resv = resv;
449437 dma_resv_lock(resv, NULL);
438+
439+ if (dma_buf->ops == &amdgpu_dmabuf_ops) {
440+ struct amdgpu_bo *other = gem_to_amdgpu_bo(dma_buf->priv);
441+
442+ flags |= other->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC;
443+ }
444+
450445 ret = amdgpu_gem_object_create(adev, dma_buf->size, PAGE_SIZE,
451− AMDGPU_GEM_DOMAIN_CPU,
452− 0, ttm_bo_type_sg, resv, &gobj);
446+ AMDGPU_GEM_DOMAIN_CPU, flags,
447+ ttm_bo_type_sg, resv, &gobj);
453448 if (ret)
454449 goto error;
455450
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c+125−3
@@ -45,6 +45,7 @@ · 44 unmodified lines
4545 #include "amdgpu_amdkfd.h"
4646
4747 #include "amdgpu_ras.h"
48+#include "amdgpu_xgmi.h"
4849
4950 /*
5051 * KMS wrapper.
@@ -90,9 +91,10 @@ · 39 unmodified lines
9091 * - 3.38.0 - Add AMDGPU_IB_FLAG_EMIT_MEM_SYNC
9192 * - 3.39.0 - DMABUF implicit sync does a full pipeline sync
9293 * - 3.40.0 - Add AMDGPU_IDS_FLAGS_TMZ
94+ * - 3.41.0 - Add video codec query
9395 */
9496 #define KMS_DRIVER_MAJOR 3
95−#define KMS_DRIVER_MINOR 40
97+#define KMS_DRIVER_MINOR 41
9698 #define KMS_DRIVER_PATCHLEVEL 0
9799
98100 int amdgpu_vram_limit;
@@ -145,6 +147,7 @@ int amdgpu_compute_multipipe = -1; · 46 unmodified lines
145147 int amdgpu_gpu_recovery = -1; /* auto */
146148 int amdgpu_emu_mode;
147149 uint amdgpu_smu_memory_pool_size;
150+int amdgpu_smu_pptable_id = -1;
148151 /*
149152 * FBC (bit 0) disabled by default
150153 * MULTI_MON_PP_MCLK_SWITCH (bit 1) enabled by default
@@ -162,16 +165,26 @@ int amdgpu_discovery = -1; · 11 unmodified lines
162165 int amdgpu_mes;
163166 int amdgpu_noretry = -1;
164167 int amdgpu_force_asic_type = -1;
165−int amdgpu_tmz;
168+int amdgpu_tmz = -1; /* auto */
169+uint amdgpu_freesync_vid_mode;
166170 int amdgpu_reset_method = -1; /* auto */
167171 int amdgpu_num_kcq = -1;
168172
173+static void amdgpu_drv_delayed_reset_work_handler(struct work_struct *work);
174+
169175 struct amdgpu_mgpu_info mgpu_info = {
170176 .mutex = __MUTEX_INITIALIZER(mgpu_info.mutex),
177+ .delayed_reset_work = __DELAYED_WORK_INITIALIZER(
178+ mgpu_info.delayed_reset_work,
179+ amdgpu_drv_delayed_reset_work_handler, 0),
171180 };
172181 int amdgpu_ras_enable = -1;
173182 uint amdgpu_ras_mask = 0xffffffff;
174183 int amdgpu_bad_page_threshold = -1;
184+struct amdgpu_watchdog_timer amdgpu_watchdog_timer = {
185+ .timeout_fatal_disable = false,
186+ .period = 0x23, /* default to max. timeout = 1 << 0x23 cycles */
187+};
175188
176189 /**
177190 * DOC: vramlimit (int)
@@ -527,6 +540,20 @@ module_param_named(ras_enable, amdgpu_ras_enable, int, 0444); · 349 unmodified lines
527540 MODULE_PARM_DESC(ras_mask, "Mask of RAS features to enable (default 0xffffffff), only valid when ras_enable == 1");
528541 module_param_named(ras_mask, amdgpu_ras_mask, uint, 0444);
529542
543+/**
544+ * DOC: timeout_fatal_disable (bool)
545+ * Disable Watchdog timeout fatal error event
546+ */
547+MODULE_PARM_DESC(timeout_fatal_disable, "disable watchdog timeout fatal error (false = default)");
548+module_param_named(timeout_fatal_disable, amdgpu_watchdog_timer.timeout_fatal_disable, bool, 0644);
549+
550+/**
551+ * DOC: timeout_period (uint)
552+ * Modify the watchdog timeout max_cycles as (1 << period)
553+ */
554+MODULE_PARM_DESC(timeout_period, "watchdog timeout period (1 to 0x23(default), timeout maxCycles = (1 << period)");
555+module_param_named(timeout_period, amdgpu_watchdog_timer.period, uint, 0644);
556+
530557 /**
531558 * DOC: si_support (int)
532559 * Set SI support driver. This parameter works after set config CONFIG_DRM_AMDGPU_SI. For SI asic, when radeon driver is enabled,
@@ -748,6 +775,13 @@ bool no_system_mem_limit; · 215 unmodified lines
748775 module_param(no_system_mem_limit, bool, 0644);
749776 MODULE_PARM_DESC(no_system_mem_limit, "disable system memory limit (false = default)");
750777
778+/**
779+ * DOC: no_queue_eviction_on_vm_fault (int)
780+ * If set, process queues will not be evicted on gpuvm fault. This is to keep the wavefront context for debugging (0 = queue eviction, 1 = no queue eviction). The default is 0 (queue eviction).
781+ */
782+int amdgpu_no_queue_eviction_on_vm_fault = 0;
783+MODULE_PARM_DESC(no_queue_eviction_on_vm_fault, "No queue eviction on VM fault (0 = queue eviction, 1 = no queue eviction)");
784+module_param_named(no_queue_eviction_on_vm_fault, amdgpu_no_queue_eviction_on_vm_fault, int, 0444);
751785 #endif
752786
753787 /**
@@ -792,9 +826,20 @@ module_param_named(backlight, amdgpu_backlight, bint, 0444); · 38 unmodified lines
792826 *
793827 * The default value: 0 (off). TODO: change to auto till it is completed.
794828 */
795−MODULE_PARM_DESC(tmz, "Enable TMZ feature (-1 = auto, 0 = off (default), 1 = on)");
829+MODULE_PARM_DESC(tmz, "Enable TMZ feature (-1 = auto (default), 0 = off, 1 = on)");
796830 module_param_named(tmz, amdgpu_tmz, int, 0444);
797831
832+/**
833+ * DOC: freesync_video (uint)
834+ * Enabled the optimization to adjust front porch timing to achieve seamless mode change experience
835+ * when setting a freesync supported mode for which full modeset is not needed.
836+ * The default value: 0 (off).
837+ */
838+MODULE_PARM_DESC(
839+ freesync_video,
840+ "Enable freesync modesetting optimization feature (0 = off (default), 1 = on)");
841+module_param_named(freesync_video, amdgpu_freesync_vid_mode, uint, 0444);
842+
798843 /**
799844 * DOC: reset_method (int)
800845 * GPU reset method (-1 = auto (default), 0 = legacy, 1 = mode0, 2 = mode1, 3 = mode2, 4 = baco, 5 = pci)
@@ -815,6 +860,15 @@ module_param_named(bad_page_threshold, amdgpu_bad_page_threshold, int, 0444); · 14 unmodified lines
815860 MODULE_PARM_DESC(num_kcq, "number of kernel compute queue user want to setup (8 if set to greater than 8 or less than 0, only affect gfx 8+)");
816861 module_param_named(num_kcq, amdgpu_num_kcq, int, 0444);
817862
863+/**
864+ * DOC: smu_pptable_id (int)
865+ * Used to override pptable id. id = 0 use VBIOS pptable.
866+ * id > 0 use the soft pptable with specicfied id.
867+ */
868+MODULE_PARM_DESC(smu_pptable_id,
869+ "specify pptable id to be used (-1 = auto(default) value, 0 = use pptable from vbios, > 0 = soft pptable id)");
870+module_param_named(smu_pptable_id, amdgpu_smu_pptable_id, int, 0444);
871+
818872 static const struct pci_device_id pciidlist[] = {
819873 #ifdef CONFIG_DRM_AMDGPU_SI
820874 {0x1002, 0x6780, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TAHITI},
@@ -1124,6 +1178,11 @@ static const struct pci_device_id pciidlist[] = { · 303 unmodified lines
11241178 {0x1002, 0x73E2, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
11251179 {0x1002, 0x73FF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
11261180
1181+ /* Aldebaran */
1182+ {0x1002, 0x7408, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ALDEBARAN|AMD_EXP_HW_SUPPORT},
1183+ {0x1002, 0x740C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ALDEBARAN|AMD_EXP_HW_SUPPORT},
1184+ {0x1002, 0x740F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ALDEBARAN|AMD_EXP_HW_SUPPORT},
1185+
11271186 {0, 0, 0}
11281187 };
11291188
@@ -1280,6 +1339,69 @@ amdgpu_pci_shutdown(struct pci_dev *pdev) · 150 unmodified lines
12801339 adev->mp1_state = PP_MP1_STATE_NONE;
12811340 }
12821341
1342+/**
1343+ * amdgpu_drv_delayed_reset_work_handler - work handler for reset
1344+ *
1345+ * @work: work_struct.
1346+ */
1347+static void amdgpu_drv_delayed_reset_work_handler(struct work_struct *work)
1348+{
1349+ struct list_head device_list;
1350+ struct amdgpu_device *adev;
1351+ int i, r;
1352+ bool need_full_reset = true;
1353+
1354+ mutex_lock(&mgpu_info.mutex);
1355+ if (mgpu_info.pending_reset == true) {
1356+ mutex_unlock(&mgpu_info.mutex);
1357+ return;
1358+ }
1359+ mgpu_info.pending_reset = true;
1360+ mutex_unlock(&mgpu_info.mutex);
1361+
1362+ for (i = 0; i < mgpu_info.num_dgpu; i++) {
1363+ adev = mgpu_info.gpu_ins[i].adev;
1364+ r = amdgpu_device_pre_asic_reset(adev, NULL, &need_full_reset);
1365+ if (r) {
1366+ dev_err(adev->dev, "GPU pre asic reset failed with err, %d for drm dev, %s ",
1367+ r, adev_to_drm(adev)->unique);
1368+ }
1369+ if (!queue_work(system_unbound_wq, &adev->xgmi_reset_work))
1370+ r = -EALREADY;
1371+ }
1372+ for (i = 0; i < mgpu_info.num_dgpu; i++) {
1373+ adev = mgpu_info.gpu_ins[i].adev;
1374+ flush_work(&adev->xgmi_reset_work);
1375+ adev->gmc.xgmi.pending_reset = false;
1376+ }
1377+
1378+ /* reset function will rebuild the xgmi hive info , clear it now */
1379+ for (i = 0; i < mgpu_info.num_dgpu; i++)
1380+ amdgpu_xgmi_remove_device(mgpu_info.gpu_ins[i].adev);
1381+
1382+ INIT_LIST_HEAD(&device_list);
1383+
1384+ for (i = 0; i < mgpu_info.num_dgpu; i++)
1385+ list_add_tail(&mgpu_info.gpu_ins[i].adev->reset_list, &device_list);
1386+
1387+ /* unregister the GPU first, reset function will add them back */
1388+ list_for_each_entry(adev, &device_list, reset_list)
1389+ amdgpu_unregister_gpu_instance(adev);
1390+
1391+ r = amdgpu_do_asic_reset(NULL, &device_list, &need_full_reset, true);
1392+ if (r) {
1393+ DRM_ERROR("reinit gpus failure");
1394+ return;
1395+ }
1396+ for (i = 0; i < mgpu_info.num_dgpu; i++) {
1397+ adev = mgpu_info.gpu_ins[i].adev;
1398+ if (!adev->kfd.init_complete)
1399+ amdgpu_amdkfd_device_init(adev);
1400+ amdgpu_ttm_set_buffer_funcs_status(adev, true);
1401+ }
1402+ return;
1403+}
1404+
12831405 static int amdgpu_pmops_suspend(struct device *dev)
12841406 {
12851407 struct drm_device *drm_dev = dev_get_drvdata(dev);
drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c+2−2
@@ -232,8 +232,8 @@ static int amdgpufb_create(struct drm_fb_helper *helper, · 231 unmodified lines
232232 goto out;
233233 }
234234
235− ret = amdgpu_display_framebuffer_init(adev_to_drm(adev), &rfbdev->rfb,
236− &mode_cmd, gobj);
235+ ret = amdgpu_display_gem_fb_init(adev_to_drm(adev), &rfbdev->rfb,
236+ &mode_cmd, gobj);
237237 if (ret) {
238238 DRM_ERROR("failed to initialize framebuffer %d\n", ret);
239239 goto out;
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c+22−29
@@ -36,8 +36,6 @@ · 35 unmodified lines
3636 #include <linux/firmware.h>
3737 #include <linux/pm_runtime.h>
3838
39−#include <drm/drm_debugfs.h>
40−
4139 #include "amdgpu.h"
4240 #include "amdgpu_trace.h"
4341
@@ -533,6 +531,8 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) · 489 unmodified lines
533531
534532 if (!ring || !ring->fence_drv.initialized)
535533 continue;
534+ if (!ring->no_scheduler)
535+ drm_sched_fini(&ring->sched);
536536 r = amdgpu_fence_wait_empty(ring);
537537 if (r) {
538538 /* no need to trigger GPU reset as we are unloading */
@@ -541,8 +541,7 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) · 2 unmodified lines
541541 if (ring->fence_drv.irq_src)
542542 amdgpu_irq_put(adev, ring->fence_drv.irq_src,
543543 ring->fence_drv.irq_type);
544− if (!ring->no_scheduler)
545− drm_sched_fini(&ring->sched);
544+
546545 del_timer_sync(&ring->fence_drv.fallback_timer);
547546 for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
548547 dma_fence_put(ring->fence_drv.fences[j]);
@@ -697,11 +696,9 @@ static const struct dma_fence_ops amdgpu_fence_ops = { · 148 unmodified lines
697696 * Fence debugfs
698697 */
699698 #if defined(CONFIG_DEBUG_FS)
700−static int amdgpu_debugfs_fence_info(struct seq_file *m, void *data)
699+static int amdgpu_debugfs_fence_info_show(struct seq_file *m, void *unused)
701700 {
702− struct drm_info_node *node = (struct drm_info_node *)m->private;
703− struct drm_device *dev = node->minor->dev;
704− struct amdgpu_device *adev = drm_to_adev(dev);
701+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
705702 int i;
706703
707704 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
@@ -746,11 +743,10 @@ static int amdgpu_debugfs_fence_info(struct seq_file *m, void *data) · 38 unmodified lines
746743 *
747744 * Manually trigger a gpu reset at the next fence wait.
748745 */
749−static int amdgpu_debugfs_gpu_recover(struct seq_file *m, void *data)
746+static int gpu_recover_get(void *data, u64 *val)
750747 {
751− struct drm_info_node *node = (struct drm_info_node *) m->private;
752− struct drm_device *dev = node->minor->dev;
753− struct amdgpu_device *adev = drm_to_adev(dev);
748+ struct amdgpu_device *adev = (struct amdgpu_device *)data;
749+ struct drm_device *dev = adev_to_drm(adev);
754750 int r;
755751
756752 r = pm_runtime_get_sync(dev->dev);
@@ -759,8 +755,7 @@ static int amdgpu_debugfs_gpu_recover(struct seq_file *m, void *data) · 2 unmodified lines
759755 return 0;
760756 }
761757
762− seq_printf(m, "gpu recover\n");
763− amdgpu_device_gpu_recover(adev, NULL);
758+ *val = amdgpu_device_gpu_recover(adev, NULL);
764759
765760 pm_runtime_mark_last_busy(dev->dev);
766761 pm_runtime_put_autosuspend(dev->dev);
@@ -768,26 +763,24 @@ static int amdgpu_debugfs_gpu_recover(struct seq_file *m, void *data) · 1 unmodified line
768763 return 0;
769764 }
770765
771−static const struct drm_info_list amdgpu_debugfs_fence_list[] = {
772− {"amdgpu_fence_info", &amdgpu_debugfs_fence_info, 0, NULL},
773− {"amdgpu_gpu_recover", &amdgpu_debugfs_gpu_recover, 0, NULL}
774−};
766+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_fence_info);
767+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_debugfs_gpu_recover_fops, gpu_recover_get, NULL,
768+ "%lld\n");
775769
776−static const struct drm_info_list amdgpu_debugfs_fence_list_sriov[] = {
777− {"amdgpu_fence_info", &amdgpu_debugfs_fence_info, 0, NULL},
778−};
779770 #endif
780771
781−int amdgpu_debugfs_fence_init(struct amdgpu_device *adev)
772+void amdgpu_debugfs_fence_init(struct amdgpu_device *adev)
782773 {
783774 #if defined(CONFIG_DEBUG_FS)
784− if (amdgpu_sriov_vf(adev))
785− return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_fence_list_sriov,
786− ARRAY_SIZE(amdgpu_debugfs_fence_list_sriov));
787− return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_fence_list,
788− ARRAY_SIZE(amdgpu_debugfs_fence_list));
789−#else
790− return 0;
775+ struct drm_minor *minor = adev_to_drm(adev)->primary;
776+ struct dentry *root = minor->debugfs_root;
777+
778+ debugfs_create_file("amdgpu_fence_info", 0444, root, adev,
779+ &amdgpu_debugfs_fence_info_fops);
780+
781+ if (!amdgpu_sriov_vf(adev))
782+ debugfs_create_file("amdgpu_gpu_recover", 0444, root, adev,
783+ &amdgpu_debugfs_gpu_recover_fops);
791784 #endif
792785 }
793786
drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c+1−25
@@ -202,6 +202,7 @@ void amdgpu_gart_table_vram_free(struct amdgpu_device *adev) · 201 unmodified lines
202202 return;
203203 }
204204 amdgpu_bo_unref(&adev->gart.bo);
205+ adev->gart.ptr = NULL;
205206 }
206207
207208 /*
@@ -236,9 +237,6 @@ int amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, · 28 unmodified lines
236237 t = offset / AMDGPU_GPU_PAGE_SIZE;
237238 p = t / AMDGPU_GPU_PAGES_IN_CPU_PAGE;
238239 for (i = 0; i < pages; i++, p++) {
239−#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
240− adev->gart.pages[p] = NULL;
241−#endif
242240 page_base = adev->dummy_page_addr;
243241 if (!adev->gart.ptr)
244242 continue;
@@ -312,9 +310,6 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset, · 67 unmodified lines
312310 int pages, struct page **pagelist, dma_addr_t *dma_addr,
313311 uint64_t flags)
314312 {
315−#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
316− unsigned t,p;
317−#endif
318313 int r, i;
319314
320315 if (!adev->gart.ready) {
@@ -322,13 +317,6 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset, · 1 unmodified line
322317 return -EINVAL;
323318 }
324319
325−#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
326− t = offset / AMDGPU_GPU_PAGE_SIZE;
327− p = t / AMDGPU_GPU_PAGES_IN_CPU_PAGE;
328− for (i = 0; i < pages; i++, p++)
329− adev->gart.pages[p] = pagelist ? pagelist[i] : NULL;
330−#endif
331−
332320 if (!adev->gart.ptr)
333321 return 0;
334322
@@ -373,14 +361,6 @@ int amdgpu_gart_init(struct amdgpu_device *adev) · 38 unmodified lines
373361 DRM_INFO("GART: num cpu pages %u, num gpu pages %u\n",
374362 adev->gart.num_cpu_pages, adev->gart.num_gpu_pages);
375363
376−#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
377− /* Allocate pages table */
378− adev->gart.pages = vzalloc(array_size(sizeof(void *),
379− adev->gart.num_cpu_pages));
380− if (adev->gart.pages == NULL)
381− return -ENOMEM;
382−#endif
383−
384364 return 0;
385365 }
386366
@@ -393,9 +373,5 @@ int amdgpu_gart_init(struct amdgpu_device *adev) · 6 unmodified lines
393373 */
394374 void amdgpu_gart_fini(struct amdgpu_device *adev)
395375 {
396−#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
397− vfree(adev->gart.pages);
398− adev->gart.pages = NULL;
399−#endif
400376 amdgpu_gart_dummy_page_fini(adev);
401377 }
drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h+0−3
@@ -46,9 +46,6 @@ struct amdgpu_gart { · 45 unmodified lines
4646 unsigned num_gpu_pages;
4747 unsigned num_cpu_pages;
4848 unsigned table_size;
49−#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
50− struct page **pages;
51−#endif
5249 bool ready;
5350
5451 /* Asic default pte flags */
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c+11−11
@@ -32,7 +32,6 @@ · 31 unmodified lines
3232 #include <linux/dma-buf.h>
3333
3434 #include <drm/amdgpu_drm.h>
35−#include <drm/drm_debugfs.h>
3635 #include <drm/drm_gem_ttm_helper.h>
3736
3837 #include "amdgpu.h"
@@ -855,10 +854,10 @@ int amdgpu_mode_dumb_create(struct drm_file *file_priv, · 816 unmodified lines
855854 }
856855
857856 #if defined(CONFIG_DEBUG_FS)
858−static int amdgpu_debugfs_gem_info(struct seq_file *m, void *data)
857+static int amdgpu_debugfs_gem_info_show(struct seq_file *m, void *unused)
859858 {
860− struct drm_info_node *node = (struct drm_info_node *)m->private;
861− struct drm_device *dev = node->minor->dev;
859+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
860+ struct drm_device *dev = adev_to_drm(adev);
862861 struct drm_file *file;
863862 int r;
864863
@@ -896,16 +895,17 @@ static int amdgpu_debugfs_gem_info(struct seq_file *m, void *data) · 31 unmodified lines
896895 return 0;
897896 }
898897
899−static const struct drm_info_list amdgpu_debugfs_gem_list[] = {
900− {"amdgpu_gem_info", &amdgpu_debugfs_gem_info, 0, NULL},
901−};
898+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_gem_info);
899+
902900 #endif
903901
904−int amdgpu_debugfs_gem_init(struct amdgpu_device *adev)
902+void amdgpu_debugfs_gem_init(struct amdgpu_device *adev)
905903 {
906904 #if defined(CONFIG_DEBUG_FS)
907− return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_gem_list,
908− ARRAY_SIZE(amdgpu_debugfs_gem_list));
905+ struct drm_minor *minor = adev_to_drm(adev)->primary;
906+ struct dentry *root = minor->debugfs_root;
907+
908+ debugfs_create_file("amdgpu_gem_info", 0444, root, adev,
909+ &amdgpu_debugfs_gem_info_fops);
909910 #endif
910− return 0;
911911 }
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c+8−1
@@ -601,6 +601,7 @@ int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev) · 600 unmodified lines
601601 struct ras_ih_if ih_info = {
602602 .cb = amdgpu_gfx_process_ras_data_cb,
603603 };
604+ struct ras_query_if info = { 0 };
604605
605606 if (!adev->gfx.ras_if) {
606607 adev->gfx.ras_if = kmalloc(sizeof(struct ras_common_if), GFP_KERNEL);
@@ -612,13 +613,19 @@ int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev) · 5 unmodified lines
612613 strcpy(adev->gfx.ras_if->name, "gfx");
613614 }
614615 fs_info.head = ih_info.head = *adev->gfx.ras_if;
615−
616616 r = amdgpu_ras_late_init(adev, adev->gfx.ras_if,
617617 &fs_info, &ih_info);
618618 if (r)
619619 goto free;
620620
621621 if (amdgpu_ras_is_supported(adev, adev->gfx.ras_if->block)) {
622+ if (adev->gmc.xgmi.connected_to_cpu) {
623+ info.head = *adev->gfx.ras_if;
624+ amdgpu_ras_query_error_status(adev, &info);
625+ } else {
626+ amdgpu_ras_reset_error_status(adev, AMDGPU_RAS_BLOCK__GFX);
627+ }
628+
622629 r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
623630 if (r)
624631 goto late_fini;
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h+3−0
@@ -30,6 +30,7 @@ · 29 unmodified lines
3030 #include "clearstate_defs.h"
3131 #include "amdgpu_ring.h"
3232 #include "amdgpu_rlc.h"
33+#include "soc15.h"
3334
3435 /* GFX current status */
3536 #define AMDGPU_GFX_NORMAL_MODE 0x00000000L
@@ -224,7 +225,9 @@ struct amdgpu_gfx_funcs { · 188 unmodified lines
224225 void (*reset_ras_error_count) (struct amdgpu_device *adev);
225226 void (*init_spm_golden)(struct amdgpu_device *adev);
226227 void (*query_ras_error_status) (struct amdgpu_device *adev);
228+ void (*reset_ras_error_status) (struct amdgpu_device *adev);
227229 void (*update_perfmon_mgcg)(struct amdgpu_device *adev, bool enable);
230+ void (*enable_watchdog_timer)(struct amdgpu_device *adev);
228231 };
229232
230233 struct sq_work {
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c+147−3
@@ -31,6 +31,57 @@ · 30 unmodified lines
3131 #include "amdgpu_ras.h"
3232 #include "amdgpu_xgmi.h"
3333
34+/**
35+ * amdgpu_gmc_pdb0_alloc - allocate vram for pdb0
36+ *
37+ * @adev: amdgpu_device pointer
38+ *
39+ * Allocate video memory for pdb0 and map it for CPU access
40+ * Returns 0 for success, error for failure.
41+ */
42+int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev)
43+{
44+ int r;
45+ struct amdgpu_bo_param bp;
46+ u64 vram_size = adev->gmc.xgmi.node_segment_size * adev->gmc.xgmi.num_physical_nodes;
47+ uint32_t pde0_page_shift = adev->gmc.vmid0_page_table_block_size + 21;
48+ uint32_t npdes = (vram_size + (1ULL << pde0_page_shift) -1) >> pde0_page_shift;
49+
50+ memset(&bp, 0, sizeof(bp));
51+ bp.size = PAGE_ALIGN((npdes + 1) * 8);
52+ bp.byte_align = PAGE_SIZE;
53+ bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
54+ bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
55+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
56+ bp.type = ttm_bo_type_kernel;
57+ bp.resv = NULL;
58+ r = amdgpu_bo_create(adev, &bp, &adev->gmc.pdb0_bo);
59+ if (r)
60+ return r;
61+
62+ r = amdgpu_bo_reserve(adev->gmc.pdb0_bo, false);
63+ if (unlikely(r != 0))
64+ goto bo_reserve_failure;
65+
66+ r = amdgpu_bo_pin(adev->gmc.pdb0_bo, AMDGPU_GEM_DOMAIN_VRAM);
67+ if (r)
68+ goto bo_pin_failure;
69+ r = amdgpu_bo_kmap(adev->gmc.pdb0_bo, &adev->gmc.ptr_pdb0);
70+ if (r)
71+ goto bo_kmap_failure;
72+
73+ amdgpu_bo_unreserve(adev->gmc.pdb0_bo);
74+ return 0;
75+
76+bo_kmap_failure:
77+ amdgpu_bo_unpin(adev->gmc.pdb0_bo);
78+bo_pin_failure:
79+ amdgpu_bo_unreserve(adev->gmc.pdb0_bo);
80+bo_reserve_failure:
81+ amdgpu_bo_unref(&adev->gmc.pdb0_bo);
82+ return r;
83+}
84+
3485 /**
3586 * amdgpu_gmc_get_pde_for_bo - get the PDE for a BO
3687 *
@@ -158,6 +209,39 @@ void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc, · 121 unmodified lines
158209 mc->vram_end, mc->real_vram_size >> 20);
159210 }
160211
212+/** amdgpu_gmc_sysvm_location - place vram and gart in sysvm aperture
213+ *
214+ * @adev: amdgpu device structure holding all necessary information
215+ * @mc: memory controller structure holding memory information
216+ *
217+ * This function is only used if use GART for FB translation. In such
218+ * case, we use sysvm aperture (vmid0 page tables) for both vram
219+ * and gart (aka system memory) access.
220+ *
221+ * GPUVM (and our organization of vmid0 page tables) require sysvm
222+ * aperture to be placed at a location aligned with 8 times of native
223+ * page size. For example, if vm_context0_cntl.page_table_block_size
224+ * is 12, then native page size is 8G (2M*2^12), sysvm should start
225+ * with a 64G aligned address. For simplicity, we just put sysvm at
226+ * address 0. So vram start at address 0 and gart is right after vram.
227+ */
228+void amdgpu_gmc_sysvm_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc)
229+{
230+ u64 hive_vram_start = 0;
231+ u64 hive_vram_end = mc->xgmi.node_segment_size * mc->xgmi.num_physical_nodes - 1;
232+ mc->vram_start = mc->xgmi.node_segment_size * mc->xgmi.physical_node_id;
233+ mc->vram_end = mc->vram_start + mc->xgmi.node_segment_size - 1;
234+ mc->gart_start = hive_vram_end + 1;
235+ mc->gart_end = mc->gart_start + mc->gart_size - 1;
236+ mc->fb_start = hive_vram_start;
237+ mc->fb_end = hive_vram_end;
238+ dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
239+ mc->mc_vram_size >> 20, mc->vram_start,
240+ mc->vram_end, mc->real_vram_size >> 20);
241+ dev_info(adev->dev, "GART: %lluM 0x%016llX - 0x%016llX\n",
242+ mc->gart_size >> 20, mc->gart_start, mc->gart_end);
243+}
244+
161245 /**
162246 * amdgpu_gmc_gart_location - try to find GART location
163247 *
@@ -165,7 +249,6 @@ void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc, · 1 unmodified line
165249 * @mc: memory controller structure holding memory information
166250 *
167251 * Function will place try to place GART before or after VRAM.
168− *
169252 * If GART size is bigger than space left then we ajust GART size.
170253 * Thus function will never fails.
171254 */
@@ -176,8 +259,6 @@ void amdgpu_gmc_gart_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc) · 4 unmodified lines
176259 /*To avoid the hole, limit the max mc address to AMDGPU_GMC_HOLE_START*/
177260 u64 max_mc_address = min(adev->gmc.mc_mask, AMDGPU_GMC_HOLE_START - 1);
178261
179− mc->gart_size += adev->pm.smu_prv_buffer_size;
180−
181262 /* VCE doesn't like it when BOs cross a 4GB segment, so align
182263 * the GART base on a 4GB boundary as well.
183264 */
@@ -384,6 +465,16 @@ void amdgpu_gmc_tmz_set(struct amdgpu_device *adev) · 200 unmodified lines
384465 {
385466 switch (adev->asic_type) {
386467 case CHIP_RAVEN:
468+ if (amdgpu_tmz == 0) {
469+ adev->gmc.tmz_enabled = false;
470+ dev_info(adev->dev,
471+ "Trusted Memory Zone (TMZ) feature disabled (cmd line)\n");
472+ } else {
473+ adev->gmc.tmz_enabled = true;
474+ dev_info(adev->dev,
475+ "Trusted Memory Zone (TMZ) feature enabled\n");
476+ }
477+ break;
387478 case CHIP_RENOIR:
388479 case CHIP_NAVI10:
389480 case CHIP_NAVI14:
@@ -423,6 +514,7 @@ void amdgpu_gmc_noretry_set(struct amdgpu_device *adev) · 33 unmodified lines
423514 switch (adev->asic_type) {
424515 case CHIP_VEGA10:
425516 case CHIP_VEGA20:
517+ case CHIP_ALDEBARAN:
426518 /*
427519 * noretry = 0 will cause kfd page fault tests fail
428520 * for some ASICs, so set default to 1 for these ASICs.
@@ -518,3 +610,55 @@ void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev) · 89 unmodified lines
518610 adev->mman.stolen_extended_size = 0;
519611 }
520612 }
613+
614+/**
615+ * amdgpu_gmc_init_pdb0 - initialize PDB0
616+ *
617+ * @adev: amdgpu_device pointer
618+ *
619+ * This function is only used when GART page table is used
620+ * for FB address translatioin. In such a case, we construct
621+ * a 2-level system VM page table: PDB0->PTB, to cover both
622+ * VRAM of the hive and system memory.
623+ *
624+ * PDB0 is static, initialized once on driver initialization.
625+ * The first n entries of PDB0 are used as PTE by setting
626+ * P bit to 1, pointing to VRAM. The n+1'th entry points
627+ * to a big PTB covering system memory.
628+ *
629+ */
630+void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev)
631+{
632+ int i;
633+ uint64_t flags = adev->gart.gart_pte_flags; //TODO it is UC. explore NC/RW?
634+ /* Each PDE0 (used as PTE) covers (2^vmid0_page_table_block_size)*2M
635+ */
636+ u64 vram_size = adev->gmc.xgmi.node_segment_size * adev->gmc.xgmi.num_physical_nodes;
637+ u64 pde0_page_size = (1ULL<<adev->gmc.vmid0_page_table_block_size)<<21;
638+ u64 vram_addr = adev->vm_manager.vram_base_offset -
639+ adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
640+ u64 vram_end = vram_addr + vram_size;
641+ u64 gart_ptb_gpu_pa = amdgpu_bo_gpu_offset(adev->gart.bo) +
642+ adev->vm_manager.vram_base_offset - adev->gmc.vram_start;
643+
644+ flags |= AMDGPU_PTE_VALID | AMDGPU_PTE_READABLE;
645+ flags |= AMDGPU_PTE_WRITEABLE;
646+ flags |= AMDGPU_PTE_SNOOPED;
647+ flags |= AMDGPU_PTE_FRAG((adev->gmc.vmid0_page_table_block_size + 9*1));
648+ flags |= AMDGPU_PDE_PTE;
649+
650+ /* The first n PDE0 entries are used as PTE,
651+ * pointing to vram
652+ */
653+ for (i = 0; vram_addr < vram_end; i++, vram_addr += pde0_page_size)
654+ amdgpu_gmc_set_pte_pde(adev, adev->gmc.ptr_pdb0, i, vram_addr, flags);
655+
656+ /* The n+1'th PDE0 entry points to a huge
657+ * PTB who has more than 512 entries each
658+ * pointing to a 4K system page
659+ */
660+ flags = AMDGPU_PTE_VALID;
661+ flags |= AMDGPU_PDE_BFS(0) | AMDGPU_PTE_SNOOPED;
662+ /* Requires gart_ptb_gpu_pa to be 4K aligned */
663+ amdgpu_gmc_set_pte_pde(adev, adev->gmc.ptr_pdb0, i, gart_ptb_gpu_pa, flags);
664+}
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h+27−4
@@ -149,6 +149,8 @@ struct amdgpu_xgmi { · 148 unmodified lines
149149 struct list_head head;
150150 bool supported;
151151 struct ras_common_if *ras_if;
152+ bool connected_to_cpu;
153+ bool pending_reset;
152154 };
153155
154156 struct amdgpu_gmc {
@@ -189,10 +191,13 @@ struct amdgpu_gmc { · 34 unmodified lines
189191 u64 gart_end;
190192 /* Frame buffer aperture of this GPU device. Different from
191193 * fb_start (see below), this only covers the local GPU device.
192− * Driver get fb_start from MC_VM_FB_LOCATION_BASE (set by vbios)
193− * and calculate vram_start of this local device by adding an
194− * offset inside the XGMI hive.
195− * Under VMID0, logical address == MC address
194+ * If driver uses FB aperture to access FB, driver get fb_start from
195+ * MC_VM_FB_LOCATION_BASE (set by vbios) and calculate vram_start
196+ * of this local device by adding an offset inside the XGMI hive.
197+ * If driver uses GART table for VMID0 FB access, driver finds a hole in
198+ * VMID0's virtual address space to place the SYSVM aperture inside
199+ * which the first part is vram and the second part is gart (covering
200+ * system ram).
196201 */
197202 u64 vram_start;
198203 u64 vram_end;
@@ -204,6 +209,15 @@ struct amdgpu_gmc { · 5 unmodified lines
204209 */
205210 u64 fb_start;
206211 u64 fb_end;
212+ /* In the case of use GART table for vmid0 FB access, [fb_start, fb_end]
213+ * will be squeezed to GART aperture. But we have a PSP FW issue to fix
214+ * for now. To temporarily workaround the PSP FW issue, added below two
215+ * variables to remember the original fb_start/end to re-enable FB
216+ * aperture to workaround the PSP FW issue. Will delete it after we
217+ * get a proper PSP FW fix.
218+ */
219+ u64 fb_start_original;
220+ u64 fb_end_original;
207221 unsigned vram_width;
208222 u64 real_vram_size;
209223 int vram_mtrr;
@@ -240,6 +254,12 @@ struct amdgpu_gmc { · 30 unmodified lines
240254 struct amdgpu_xgmi xgmi;
241255 struct amdgpu_irq_src ecc_irq;
242256 int noretry;
257+
258+ uint32_t vmid0_page_table_block_size;
259+ uint32_t vmid0_page_table_depth;
260+ struct amdgpu_bo *pdb0_bo;
261+ /* CPU kmapped address of pdb0*/
262+ void *ptr_pdb0;
243263 };
244264
245265 #define amdgpu_gmc_flush_gpu_tlb(adev, vmid, vmhub, type) ((adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (vmhub), (type)))
@@ -281,6 +301,7 @@ static inline uint64_t amdgpu_gmc_sign_extend(uint64_t addr) · 35 unmodified lines
281301 return addr;
282302 }
283303
304+int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev);
284305 void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level,
285306 uint64_t *addr, uint64_t *flags);
286307 int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
@@ -288,6 +309,7 @@ int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr, · 1 unmodified line
288309 uint64_t flags);
289310 uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo);
290311 uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo);
312+void amdgpu_gmc_sysvm_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc);
291313 void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
292314 u64 base);
293315 void amdgpu_gmc_gart_location(struct amdgpu_device *adev,
@@ -309,4 +331,5 @@ amdgpu_gmc_set_vm_fault_masks(struct amdgpu_device *adev, int hub_type, · 15 unmodified lines
309331
310332 void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev);
311333
334+void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev);
312335 #endif
drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c+10−13
@@ -30,7 +30,6 @@ · 29 unmodified lines
3030 #include <linux/slab.h>
3131
3232 #include <drm/amdgpu_drm.h>
33−#include <drm/drm_debugfs.h>
3433
3534 #include "amdgpu.h"
3635 #include "atom.h"
@@ -453,11 +452,9 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev) · 416 unmodified lines
453452 */
454453 #if defined(CONFIG_DEBUG_FS)
455454
456−static int amdgpu_debugfs_sa_info(struct seq_file *m, void *data)
455+static int amdgpu_debugfs_sa_info_show(struct seq_file *m, void *unused)
457456 {
458− struct drm_info_node *node = (struct drm_info_node *) m->private;
459− struct drm_device *dev = node->minor->dev;
460− struct amdgpu_device *adev = drm_to_adev(dev);
457+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
461458
462459 seq_printf(m, "--------------------- DELAYED --------------------- \n");
463460 amdgpu_sa_bo_dump_debug_info(&adev->ib_pools[AMDGPU_IB_POOL_DELAYED],
@@ -471,18 +468,18 @@ static int amdgpu_debugfs_sa_info(struct seq_file *m, void *data) · 7 unmodified lines
471468 return 0;
472469 }
473470
474−static const struct drm_info_list amdgpu_debugfs_sa_list[] = {
475− {"amdgpu_sa_info", &amdgpu_debugfs_sa_info, 0, NULL},
476−};
471+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_sa_info);
477472
478473 #endif
479474
480−int amdgpu_debugfs_sa_init(struct amdgpu_device *adev)
475+void amdgpu_debugfs_sa_init(struct amdgpu_device *adev)
481476 {
482477 #if defined(CONFIG_DEBUG_FS)
483− return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_sa_list,
484− ARRAY_SIZE(amdgpu_debugfs_sa_list));
485−#else
486− return 0;
478+ struct drm_minor *minor = adev_to_drm(adev)->primary;
479+ struct dentry *root = minor->debugfs_root;
480+
481+ debugfs_create_file("amdgpu_sa_info", 0444, root, adev,
482+ &amdgpu_debugfs_sa_info_fops);
483+
487484 #endif
488485 }
drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c+49−5
@@ -99,6 +99,8 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, · 98 unmodified lines
9999 ih->rptr_addr = adev->wb.gpu_addr + rptr_offs * 4;
100100 ih->rptr_cpu = &adev->wb.wb[rptr_offs];
101101 }
102+
103+ init_waitqueue_head(&ih->wait_process);
102104 return 0;
103105 }
104106
@@ -160,6 +162,52 @@ void amdgpu_ih_ring_write(struct amdgpu_ih_ring *ih, const uint32_t *iv, · 55 unmodified lines
160162 }
161163 }
162164
165+/* Waiter helper that checks current rptr matches or passes checkpoint wptr */
166+static bool amdgpu_ih_has_checkpoint_processed(struct amdgpu_device *adev,
167+ struct amdgpu_ih_ring *ih,
168+ uint32_t checkpoint_wptr,
169+ uint32_t *prev_rptr)
170+{
171+ uint32_t cur_rptr = ih->rptr | (*prev_rptr & ~ih->ptr_mask);
172+
173+ /* rptr has wrapped. */
174+ if (cur_rptr < *prev_rptr)
175+ cur_rptr += ih->ptr_mask + 1;
176+ *prev_rptr = cur_rptr;
177+
178+ return cur_rptr >= checkpoint_wptr;
179+}
180+
181+/**
182+ * amdgpu_ih_wait_on_checkpoint_process - wait to process IVs up to checkpoint
183+ *
184+ * @adev: amdgpu_device pointer
185+ * @ih: ih ring to process
186+ *
187+ * Used to ensure ring has processed IVs up to the checkpoint write pointer.
188+ */
189+int amdgpu_ih_wait_on_checkpoint_process(struct amdgpu_device *adev,
190+ struct amdgpu_ih_ring *ih)
191+{
192+ uint32_t checkpoint_wptr, rptr;
193+
194+ if (!ih->enabled || adev->shutdown)
195+ return -ENODEV;
196+
197+ checkpoint_wptr = amdgpu_ih_get_wptr(adev, ih);
198+ /* Order wptr with rptr. */
199+ rmb();
200+ rptr = READ_ONCE(ih->rptr);
201+
202+ /* wptr has wrapped. */
203+ if (rptr > checkpoint_wptr)
204+ checkpoint_wptr += ih->ptr_mask + 1;
205+
206+ return wait_event_interruptible(ih->wait_process,
207+ amdgpu_ih_has_checkpoint_processed(adev, ih,
208+ checkpoint_wptr, &rptr));
209+}
210+
163211 /**
164212 * amdgpu_ih_process - interrupt handler
165213 *
@@ -180,10 +228,6 @@ int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) · 14 unmodified lines
180228 wptr = amdgpu_ih_get_wptr(adev, ih);
181229
182230 restart_ih:
183− /* is somebody else already processing irqs? */
184− if (atomic_xchg(&ih->lock, 1))
185− return IRQ_NONE;
186−
187231 DRM_DEBUG("%s: rptr %d, wptr %d\n", __func__, ih->rptr, wptr);
188232
189233 /* Order reading of wptr vs. reading of IH ring data */
@@ -195,7 +239,7 @@ restart_ih: · 5 unmodified lines
195239 }
196240
197241 amdgpu_ih_set_rptr(adev, ih);
198− atomic_set(&ih->lock, 0);
242+ wake_up_all(&ih->wait_process);
199243
200244 /* make sure wptr hasn't changed while processing */
201245 wptr = amdgpu_ih_get_wptr(adev, ih);
drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h+5−1
@@ -64,8 +64,10 @@ struct amdgpu_ih_ring { · 63 unmodified lines
6464
6565 bool enabled;
6666 unsigned rptr;
67− atomic_t lock;
6867 struct amdgpu_ih_regs ih_regs;
68+
69+ /* For waiting on IH processing at checkpoint. */
70+ wait_queue_head_t wait_process;
6971 };
7072
7173 /* provided by the ih block */
@@ -87,6 +89,8 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, · 15 unmodified lines
8789 void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih);
8890 void amdgpu_ih_ring_write(struct amdgpu_ih_ring *ih, const uint32_t *iv,
8991 unsigned int num_dw);
92+int amdgpu_ih_wait_on_checkpoint_process(struct amdgpu_device *adev,
93+ struct amdgpu_ih_ring *ih);
9094 int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih);
9195 void amdgpu_ih_decode_iv_helper(struct amdgpu_device *adev,
9296 struct amdgpu_ih_ring *ih,
drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c+36−1
@@ -65,6 +65,41 @@ · 64 unmodified lines
6565
6666 #define AMDGPU_WAIT_IDLE_TIMEOUT 200
6767
68+const char *soc15_ih_clientid_name[] = {
69+ "IH",
70+ "SDMA2 or ACP",
71+ "ATHUB",
72+ "BIF",
73+ "SDMA3 or DCE",
74+ "SDMA4 or ISP",
75+ "VMC1 or PCIE0",
76+ "RLC",
77+ "SDMA0",
78+ "SDMA1",
79+ "SE0SH",
80+ "SE1SH",
81+ "SE2SH",
82+ "SE3SH",
83+ "VCN1 or UVD1",
84+ "THM",
85+ "VCN or UVD",
86+ "SDMA5 or VCE0",
87+ "VMC",
88+ "SDMA6 or XDMA",
89+ "GRBM_CP",
90+ "ATS",
91+ "ROM_SMUIO",
92+ "DF",
93+ "SDMA7 or VCE1",
94+ "PWR",
95+ "reserved",
96+ "UTCL2",
97+ "EA",
98+ "UTCL2LOG",
99+ "MP0",
100+ "MP1"
101+};
102+
68103 /**
69104 * amdgpu_hotplug_work_func - work handler for display hotplug event
70105 *
@@ -535,7 +570,7 @@ void amdgpu_irq_gpu_reset_resume_helper(struct amdgpu_device *adev) · 464 unmodified lines
535570 for (j = 0; j < AMDGPU_MAX_IRQ_SRC_ID; ++j) {
536571 struct amdgpu_irq_src *src = adev->irq.client[i].sources[j];
537572
538− if (!src)
573+ if (!src || !src->funcs || !src->funcs->set)
539574 continue;
540575 for (k = 0; k < src->num_types; k++)
541576 amdgpu_irq_update(adev, src, k);
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c+95−38
@@ -27,7 +27,6 @@ · 26 unmodified lines
2727 */
2828
2929 #include "amdgpu.h"
30−#include <drm/drm_debugfs.h>
3130 #include <drm/amdgpu_drm.h>
3231 #include "amdgpu_uvd.h"
3332 #include "amdgpu_vce.h"
@@ -287,22 +286,30 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info, · 253 unmodified lines
287286 break;
288287 case AMDGPU_INFO_FW_TA:
289288 switch (query_fw->index) {
290− case 0:
289+ case TA_FW_TYPE_PSP_XGMI:
291290 fw_info->ver = adev->psp.ta_fw_version;
292291 fw_info->feature = adev->psp.ta_xgmi_ucode_version;
293292 break;
294− case 1:
293+ case TA_FW_TYPE_PSP_RAS:
295294 fw_info->ver = adev->psp.ta_fw_version;
296295 fw_info->feature = adev->psp.ta_ras_ucode_version;
297296 break;
298− case 2:
297+ case TA_FW_TYPE_PSP_HDCP:
299298 fw_info->ver = adev->psp.ta_fw_version;
300299 fw_info->feature = adev->psp.ta_hdcp_ucode_version;
301300 break;
302− case 3:
301+ case TA_FW_TYPE_PSP_DTM:
303302 fw_info->ver = adev->psp.ta_fw_version;
304303 fw_info->feature = adev->psp.ta_dtm_ucode_version;
305304 break;
305+ case TA_FW_TYPE_PSP_RAP:
306+ fw_info->ver = adev->psp.ta_fw_version;
307+ fw_info->feature = adev->psp.ta_rap_ucode_version;
308+ break;
309+ case TA_FW_TYPE_PSP_SECUREDISPLAY:
310+ fw_info->ver = adev->psp.ta_fw_version;
311+ fw_info->feature = adev->psp.ta_securedisplay_ucode_version;
312+ break;
306313 default:
307314 return -EINVAL;
308315 }
@@ -981,6 +988,63 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) · 672 unmodified lines
981988 min_t(u64, size, sizeof(ras_mask))) ?
982989 -EFAULT : 0;
983990 }
991+ case AMDGPU_INFO_VIDEO_CAPS: {
992+ const struct amdgpu_video_codecs *codecs;
993+ struct drm_amdgpu_info_video_caps *caps;
994+ int r;
995+
996+ switch (info->video_cap.type) {
997+ case AMDGPU_INFO_VIDEO_CAPS_DECODE:
998+ r = amdgpu_asic_query_video_codecs(adev, false, &codecs);
999+ if (r)
1000+ return -EINVAL;
1001+ break;
1002+ case AMDGPU_INFO_VIDEO_CAPS_ENCODE:
1003+ r = amdgpu_asic_query_video_codecs(adev, true, &codecs);
1004+ if (r)
1005+ return -EINVAL;
1006+ break;
1007+ default:
1008+ DRM_DEBUG_KMS("Invalid request %d\n",
1009+ info->video_cap.type);
1010+ return -EINVAL;
1011+ }
1012+
1013+ caps = kzalloc(sizeof(*caps), GFP_KERNEL);
1014+ if (!caps)
1015+ return -ENOMEM;
1016+
1017+ for (i = 0; i < codecs->codec_count; i++) {
1018+ int idx = codecs->codec_array[i].codec_type;
1019+
1020+ switch (idx) {
1021+ case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2:
1022+ case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4:
1023+ case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1:
1024+ case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC:
1025+ case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC:
1026+ case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG:
1027+ case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9:
1028+ case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_AV1:
1029+ caps->codec_info[idx].valid = 1;
1030+ caps->codec_info[idx].max_width =
1031+ codecs->codec_array[i].max_width;
1032+ caps->codec_info[idx].max_height =
1033+ codecs->codec_array[i].max_height;
1034+ caps->codec_info[idx].max_pixels_per_frame =
1035+ codecs->codec_array[i].max_pixels_per_frame;
1036+ caps->codec_info[idx].max_level =
1037+ codecs->codec_array[i].max_level;
1038+ break;
1039+ default:
1040+ break;
1041+ }
1042+ }
1043+ r = copy_to_user(out, caps,
1044+ min((size_t)size, sizeof(*caps))) ? -EFAULT : 0;
1045+ kfree(caps);
1046+ return r;
1047+ }
9841048 default:
9851049 DRM_DEBUG_KMS("Invalid request %d\n", info->query);
9861050 return -EINVAL;
@@ -1262,16 +1326,25 @@ void amdgpu_disable_vblank_kms(struct drm_crtc *crtc) · 275 unmodified lines
12621326 */
12631327 #if defined(CONFIG_DEBUG_FS)
12641328
1265−static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
1329+static int amdgpu_debugfs_firmware_info_show(struct seq_file *m, void *unused)
12661330 {
1267− struct drm_info_node *node = (struct drm_info_node *) m->private;
1268− struct drm_device *dev = node->minor->dev;
1269− struct amdgpu_device *adev = drm_to_adev(dev);
1331+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
12701332 struct drm_amdgpu_info_firmware fw_info;
12711333 struct drm_amdgpu_query_fw query_fw;
12721334 struct atom_context *ctx = adev->mode_info.atom_context;
12731335 int ret, i;
12741336
1337+ static const char *ta_fw_name[TA_FW_TYPE_MAX_INDEX] = {
1338+#define TA_FW_NAME(type) [TA_FW_TYPE_PSP_##type] = #type
1339+ TA_FW_NAME(XGMI),
1340+ TA_FW_NAME(RAS),
1341+ TA_FW_NAME(HDCP),
1342+ TA_FW_NAME(DTM),
1343+ TA_FW_NAME(RAP),
1344+ TA_FW_NAME(SECUREDISPLAY),
1345+#undef TA_FW_NAME
1346+ };
1347+
12751348 /* VCE */
12761349 query_fw.fw_type = AMDGPU_INFO_FW_VCE;
12771350 ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
@@ -1389,31 +1462,14 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data) · 111 unmodified lines
13891462 fw_info.feature, fw_info.ver);
13901463
13911464 query_fw.fw_type = AMDGPU_INFO_FW_TA;
1392− for (i = 0; i < 4; i++) {
1465+ for (i = TA_FW_TYPE_PSP_XGMI; i < TA_FW_TYPE_MAX_INDEX; i++) {
13931466 query_fw.index = i;
13941467 ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
13951468 if (ret)
13961469 continue;
1397− switch (query_fw.index) {
1398− case 0:
1399− seq_printf(m, "TA %s feature version: 0x%08x, firmware version: 0x%08x\n",
1400− "RAS", fw_info.feature, fw_info.ver);
1401− break;
1402− case 1:
1403− seq_printf(m, "TA %s feature version: 0x%08x, firmware version: 0x%08x\n",
1404− "XGMI", fw_info.feature, fw_info.ver);
1405− break;
1406− case 2:
1407− seq_printf(m, "TA %s feature version: 0x%08x, firmware version: 0x%08x\n",
1408− "HDCP", fw_info.feature, fw_info.ver);
1409− break;
1410− case 3:
1411− seq_printf(m, "TA %s feature version: 0x%08x, firmware version: 0x%08x\n",
1412− "DTM", fw_info.feature, fw_info.ver);
1413− break;
1414− default:
1415− return -EINVAL;
1416− }
1470+
1471+ seq_printf(m, "TA %s feature version: 0x%08x, firmware version: 0x%08x\n",
1472+ ta_fw_name[i], fw_info.feature, fw_info.ver);
14171473 }
14181474
14191475 /* SMC */
@@ -1472,17 +1528,18 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data) · 52 unmodified lines
14721528 return 0;
14731529 }
14741530
1475−static const struct drm_info_list amdgpu_firmware_info_list[] = {
1476− {"amdgpu_firmware_info", amdgpu_debugfs_firmware_info, 0, NULL},
1477−};
1531+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_firmware_info);
1532+
14781533 #endif
14791534
1480−int amdgpu_debugfs_firmware_init(struct amdgpu_device *adev)
1535+void amdgpu_debugfs_firmware_init(struct amdgpu_device *adev)
14811536 {
14821537 #if defined(CONFIG_DEBUG_FS)
1483− return amdgpu_debugfs_add_files(adev, amdgpu_firmware_info_list,
1484− ARRAY_SIZE(amdgpu_firmware_info_list));
1485−#else
1486− return 0;
1538+ struct drm_minor *minor = adev_to_drm(adev)->primary;
1539+ struct dentry *root = minor->debugfs_root;
1540+
1541+ debugfs_create_file("amdgpu_firmware_info", 0444, root,
1542+ adev, &amdgpu_debugfs_firmware_info_fops);
1543+
14871544 #endif
14881545 }
drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h+8−0
@@ -602,6 +602,14 @@ int amdgpu_display_get_crtc_scanoutpos(struct drm_device *dev, · 601 unmodified lines
602602 int *hpos, ktime_t *stime, ktime_t *etime,
603603 const struct drm_display_mode *mode);
604604
605+int amdgpu_display_gem_fb_init(struct drm_device *dev,
606+ struct amdgpu_framebuffer *rfb,
607+ const struct drm_mode_fb_cmd2 *mode_cmd,
608+ struct drm_gem_object *obj);
609+int amdgpu_display_gem_fb_verify_and_init(
610+ struct drm_device *dev, struct amdgpu_framebuffer *rfb,
611+ struct drm_file *file_priv, const struct drm_mode_fb_cmd2 *mode_cmd,
612+ struct drm_gem_object *obj);
605613 int amdgpu_display_framebuffer_init(struct drm_device *dev,
606614 struct amdgpu_framebuffer *rfb,
607615 const struct drm_mode_fb_cmd2 *mode_cmd,
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c+14−8
@@ -1061,13 +1061,17 @@ static const char *amdgpu_vram_names[] = { · 1060 unmodified lines
10611061 */
10621062 int amdgpu_bo_init(struct amdgpu_device *adev)
10631063 {
1064− /* reserve PAT memory space to WC for VRAM */
1065− arch_io_reserve_memtype_wc(adev->gmc.aper_base,
1066− adev->gmc.aper_size);
1064+ /* On A+A platform, VRAM can be mapped as WB */
1065+ if (!adev->gmc.xgmi.connected_to_cpu) {
1066+ /* reserve PAT memory space to WC for VRAM */
1067+ arch_io_reserve_memtype_wc(adev->gmc.aper_base,
1068+ adev->gmc.aper_size);
1069+
1070+ /* Add an MTRR for the VRAM */
1071+ adev->gmc.vram_mtrr = arch_phys_wc_add(adev->gmc.aper_base,
1072+ adev->gmc.aper_size);
1073+ }
10671074
1068− /* Add an MTRR for the VRAM */
1069− adev->gmc.vram_mtrr = arch_phys_wc_add(adev->gmc.aper_base,
1070− adev->gmc.aper_size);
10711075 DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
10721076 adev->gmc.mc_vram_size >> 20,
10731077 (unsigned long long)adev->gmc.aper_size >> 20);
@@ -1085,8 +1089,10 @@ int amdgpu_bo_init(struct amdgpu_device *adev) · 11 unmodified lines
10851089 void amdgpu_bo_fini(struct amdgpu_device *adev)
10861090 {
10871091 amdgpu_ttm_fini(adev);
1088− arch_phys_wc_del(adev->gmc.vram_mtrr);
1089− arch_io_free_memtype_wc(adev->gmc.aper_base, adev->gmc.aper_size);
1092+ if (!adev->gmc.xgmi.connected_to_cpu) {
1093+ arch_phys_wc_del(adev->gmc.vram_mtrr);
1094+ arch_io_free_memtype_wc(adev->gmc.aper_base, adev->gmc.aper_size);
1095+ }
10901096 }
10911097
10921098 /**
drivers/gpu/drm/amd/amdgpu/amdgpu_object.h+1−2
@@ -100,7 +100,6 @@ struct amdgpu_bo { · 99 unmodified lines
100100 struct amdgpu_bo *parent;
101101 struct amdgpu_bo *shadow;
102102
103− struct amdgpu_mn *mn;
104103
105104
106105 #ifdef CONFIG_MMU_NOTIFIER
@@ -329,7 +328,7 @@ void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager, · 222 unmodified lines
329328 struct seq_file *m);
330329 u64 amdgpu_bo_print_info(int id, struct amdgpu_bo *bo, struct seq_file *m);
331330 #endif
332−int amdgpu_debugfs_sa_init(struct amdgpu_device *adev);
331+void amdgpu_debugfs_sa_init(struct amdgpu_device *adev);
333332
334333 bool amdgpu_bo_support_uswc(u64 bo_flags);
335334
drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c+77−18
@@ -34,6 +34,7 @@ · 33 unmodified lines
3434 #include "psp_v10_0.h"
3535 #include "psp_v11_0.h"
3636 #include "psp_v12_0.h"
37+#include "psp_v13_0.h"
3738
3839 #include "amdgpu_ras.h"
3940 #include "amdgpu_securedisplay.h"
@@ -56,7 +57,7 @@ static int psp_load_smu_fw(struct psp_context *psp); · 16 unmodified lines
5657 * - Load XGMI/RAS/HDCP/DTM TA if any
5758 *
5859 * This new sequence is required for
59− * - Arcturus
60+ * - Arcturus and onwards
6061 * - Navi12 and onwards
6162 */
6263 static void psp_check_pmfw_centralized_cstate_management(struct psp_context *psp)
@@ -71,7 +72,7 @@ static void psp_check_pmfw_centralized_cstate_management(struct psp_context *psp · 8 unmodified lines
7172 if (adev->flags & AMD_IS_APU)
7273 return;
7374
74− if ((adev->asic_type == CHIP_ARCTURUS) ||
75+ if ((adev->asic_type >= CHIP_ARCTURUS) ||
7576 (adev->asic_type >= CHIP_NAVI12))
7677 psp->pmfw_centralized_cstate_management = true;
7778 }
@@ -109,6 +110,9 @@ static int psp_early_init(void *handle) · 31 unmodified lines
109110 case CHIP_RENOIR:
110111 psp_v12_0_set_psp_funcs(psp);
111112 break;
113+ case CHIP_ALDEBARAN:
114+ psp_v13_0_set_psp_funcs(psp);
115+ break;
112116 default:
113117 return -EINVAL;
114118 }
@@ -383,7 +387,7 @@ static int psp_tmr_init(struct psp_context *psp) · 268 unmodified lines
383387 * Note: this memory need be reserved till the driver
384388 * uninitializes.
385389 */
386− tmr_size = PSP_TMR_SIZE;
390+ tmr_size = PSP_TMR_SIZE(psp->adev);
387391
388392 /* For ASICs support RLC autoload, psp will parse the toc
389393 * and calculate the total size of TMR needed */
@@ -399,10 +403,20 @@ static int psp_tmr_init(struct psp_context *psp) · 9 unmodified lines
399403 }
400404
401405 pptr = amdgpu_sriov_vf(psp->adev) ? &tmr_buf : NULL;
402− ret = amdgpu_bo_create_kernel(psp->adev, tmr_size, PSP_TMR_SIZE,
406+ ret = amdgpu_bo_create_kernel(psp->adev, tmr_size, PSP_TMR_SIZE(psp->adev),
403407 AMDGPU_GEM_DOMAIN_VRAM,
404408 &psp->tmr_bo, &psp->tmr_mc_addr, pptr);
405409
410+ /* workaround the tmr_mc_addr:
411+ * PSP requires an address in FB aperture. Right now driver produce
412+ * tmr_mc_addr in the GART aperture. Convert it back to FB aperture
413+ * for PSP. Will revert it after we get a fix from PSP FW.
414+ */
415+ if (psp->adev->asic_type == CHIP_ALDEBARAN) {
416+ psp->tmr_mc_addr -= psp->adev->gmc.fb_start;
417+ psp->tmr_mc_addr += psp->adev->gmc.fb_start_original;
418+ }
419+
406420 return ret;
407421 }
408422
@@ -542,6 +556,28 @@ int psp_get_fw_attestation_records_addr(struct psp_context *psp, · 133 unmodified lines
542556 return ret;
543557 }
544558
559+static int psp_rl_load(struct amdgpu_device *adev)
560+{
561+ struct psp_context *psp = &adev->psp;
562+ struct psp_gfx_cmd_resp *cmd = psp->cmd;
563+
564+ if (psp->rl_bin_size == 0)
565+ return 0;
566+
567+ memset(psp->fw_pri_buf, 0, PSP_1_MEG);
568+ memcpy(psp->fw_pri_buf, psp->rl_start_addr, psp->rl_bin_size);
569+
570+ memset(cmd, 0, sizeof(struct psp_gfx_cmd_resp));
571+
572+ cmd->cmd_id = GFX_CMD_ID_LOAD_IP_FW;
573+ cmd->cmd.cmd_load_ip_fw.fw_phy_addr_lo = lower_32_bits(psp->fw_pri_mc_addr);
574+ cmd->cmd.cmd_load_ip_fw.fw_phy_addr_hi = upper_32_bits(psp->fw_pri_mc_addr);
575+ cmd->cmd.cmd_load_ip_fw.fw_size = psp->rl_bin_size;
576+ cmd->cmd.cmd_load_ip_fw.fw_type = GFX_FW_TYPE_REG_LIST;
577+
578+ return psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
579+}
580+
545581 static void psp_prep_asd_load_cmd_buf(struct psp_gfx_cmd_resp *cmd,
546582 uint64_t asd_mc, uint32_t size)
547583 {
@@ -755,8 +791,9 @@ static int psp_xgmi_unload(struct psp_context *psp) · 207 unmodified lines
755791 struct psp_gfx_cmd_resp *cmd;
756792 struct amdgpu_device *adev = psp->adev;
757793
758− /* XGMI TA unload currently is not supported on Arcturus */
759− if (adev->asic_type == CHIP_ARCTURUS)
794+ /* XGMI TA unload currently is not supported on Arcturus/Aldebaran A+A */
795+ if (adev->asic_type == CHIP_ARCTURUS ||
796+ (adev->asic_type == CHIP_ALDEBARAN && adev->gmc.xgmi.connected_to_cpu))
760797 return 0;
761798
762799 /*
@@ -1561,6 +1598,7 @@ static int psp_rap_unload(struct psp_context *psp) · 798 unmodified lines
15611598 static int psp_rap_initialize(struct psp_context *psp)
15621599 {
15631600 int ret;
1601+ enum ta_rap_status status = TA_RAP_STATUS__SUCCESS;
15641602
15651603 /*
15661604 * TODO: bypass the initialize in sriov for now
@@ -1584,8 +1622,8 @@ static int psp_rap_initialize(struct psp_context *psp) · 17 unmodified lines
15841622 if (ret)
15851623 return ret;
15861624
1587− ret = psp_rap_invoke(psp, TA_CMD_RAP__INITIALIZE);
1588− if (ret != TA_RAP_STATUS__SUCCESS) {
1625+ ret = psp_rap_invoke(psp, TA_CMD_RAP__INITIALIZE, &status);
1626+ if (ret || status != TA_RAP_STATUS__SUCCESS) {
15891627 psp_rap_unload(psp);
15901628
15911629 amdgpu_bo_free_kernel(&psp->rap_context.rap_shared_bo,
@@ -1594,8 +1632,10 @@ static int psp_rap_initialize(struct psp_context *psp) · 2 unmodified lines
15941632
15951633 psp->rap_context.rap_initialized = false;
15961634
1597− dev_warn(psp->adev->dev, "RAP TA initialize fail.\n");
1598− return -EINVAL;
1635+ dev_warn(psp->adev->dev, "RAP TA initialize fail (%d) status %d.\n",
1636+ ret, status);
1637+
1638+ return ret;
15991639 }
16001640
16011641 return 0;
@@ -1620,13 +1660,13 @@ static int psp_rap_terminate(struct psp_context *psp) · 18 unmodified lines
16201660 return ret;
16211661 }
16221662
1623−int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id)
1663+int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id, enum ta_rap_status *status)
16241664 {
16251665 struct ta_rap_shared_memory *rap_cmd;
1626− int ret;
1666+ int ret = 0;
16271667
16281668 if (!psp->rap_context.rap_initialized)
1629− return -EINVAL;
1669+ return 0;
16301670
16311671 if (ta_cmd_id != TA_CMD_RAP__INITIALIZE &&
16321672 ta_cmd_id != TA_CMD_RAP__VALIDATE_L0)
@@ -1642,14 +1682,16 @@ int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id) · 9 unmodified lines
16421682 rap_cmd->validation_method_id = METHOD_A;
16431683
16441684 ret = psp_ta_invoke(psp, rap_cmd->cmd_id, psp->rap_context.session_id);
1645− if (ret) {
1646− mutex_unlock(&psp->rap_context.mutex);
1647− return ret;
1648− }
1685+ if (ret)
1686+ goto out_unlock;
16491687
1688+ if (status)
1689+ *status = rap_cmd->rap_status;
1690+
1691+out_unlock:
16501692 mutex_unlock(&psp->rap_context.mutex);
16511693
1652− return rap_cmd->rap_status;
1694+ return ret;
16531695 }
16541696 // RAP end
16551697
@@ -2276,6 +2318,12 @@ skip_memalloc: · 620 unmodified lines
22762318 return ret;
22772319 }
22782320
2321+ ret = psp_rl_load(adev);
2322+ if (ret) {
2323+ DRM_ERROR("PSP load RL failed!\n");
2324+ return ret;
2325+ }
2326+
22792327 if (psp->adev->psp.ta_fw) {
22802328 ret = psp_ras_initialize(psp);
22812329 if (ret)
@@ -2751,6 +2799,9 @@ int psp_init_sos_microcode(struct psp_context *psp, · 469 unmodified lines
27512799 adev->psp.spl_bin_size = le32_to_cpu(sos_hdr_v1_3->spl_size_bytes);
27522800 adev->psp.spl_start_addr = (uint8_t *)adev->psp.sys_start_addr +
27532801 le32_to_cpu(sos_hdr_v1_3->spl_offset_bytes);
2802+ adev->psp.rl_bin_size = le32_to_cpu(sos_hdr_v1_3->rl_size_bytes);
2803+ adev->psp.rl_start_addr = (uint8_t *)adev->psp.sys_start_addr +
2804+ le32_to_cpu(sos_hdr_v1_3->rl_offset_bytes);
27542805 }
27552806 break;
27562807 default:
@@ -3052,3 +3103,11 @@ const struct amdgpu_ip_block_version psp_v12_0_ip_block = · 295 unmodified lines
30523103 .rev = 0,
30533104 .funcs = &psp_ip_funcs,
30543105 };
3106+
3107+const struct amdgpu_ip_block_version psp_v13_0_ip_block = {
3108+ .type = AMD_IP_BLOCK_TYPE_PSP,
3109+ .major = 13,
3110+ .minor = 0,
3111+ .rev = 0,
3112+ .funcs = &psp_ip_funcs,
3113+};
drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h+9−6
@@ -37,7 +37,7 @@ · 36 unmodified lines
3737 #define PSP_XGMI_SHARED_MEM_SIZE 0x4000
3838 #define PSP_RAS_SHARED_MEM_SIZE 0x4000
3939 #define PSP_1_MEG 0x100000
40−#define PSP_TMR_SIZE 0x400000
40+#define PSP_TMR_SIZE(adev) ((adev)->asic_type == CHIP_ALDEBARAN ? 0x800000 : 0x400000)
4141 #define PSP_HDCP_SHARED_MEM_SIZE 0x4000
4242 #define PSP_DTM_SHARED_MEM_SIZE 0x4000
4343 #define PSP_RAP_SHARED_MEM_SIZE 0x4000
@@ -248,11 +248,13 @@ struct psp_context · 204 unmodified lines
248248 uint32_t toc_bin_size;
249249 uint32_t kdb_bin_size;
250250 uint32_t spl_bin_size;
251+ uint32_t rl_bin_size;
251252 uint8_t *sys_start_addr;
252253 uint8_t *sos_start_addr;
253254 uint8_t *toc_start_addr;
254255 uint8_t *kdb_start_addr;
255256 uint8_t *spl_start_addr;
257+ uint8_t *rl_start_addr;
256258
257259 /* tmr buffer */
258260 struct amdgpu_bo *tmr_bo;
@@ -365,11 +367,13 @@ struct amdgpu_psp_funcs { · 106 unmodified lines
365367 extern const struct amd_ip_funcs psp_ip_funcs;
366368
367369 extern const struct amdgpu_ip_block_version psp_v3_1_ip_block;
368−extern int psp_wait_for(struct psp_context *psp, uint32_t reg_index,
369− uint32_t field_val, uint32_t mask, bool check_changed);
370−
371370 extern const struct amdgpu_ip_block_version psp_v10_0_ip_block;
371+extern const struct amdgpu_ip_block_version psp_v11_0_ip_block;
372372 extern const struct amdgpu_ip_block_version psp_v12_0_ip_block;
373+extern const struct amdgpu_ip_block_version psp_v13_0_ip_block;
374+
375+extern int psp_wait_for(struct psp_context *psp, uint32_t reg_index,
376+ uint32_t field_val, uint32_t mask, bool check_changed);
373377
374378 int psp_gpu_reset(struct amdgpu_device *adev);
375379 int psp_update_vcn_sram(struct amdgpu_device *adev, int inst_idx,
@@ -395,12 +399,11 @@ int psp_ras_trigger_error(struct psp_context *psp, · 19 unmodified lines
395399
396400 int psp_hdcp_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
397401 int psp_dtm_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
398−int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
402+int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id, enum ta_rap_status *status);
399403 int psp_securedisplay_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
400404
401405 int psp_rlc_autoload_start(struct psp_context *psp);
402406
403−extern const struct amdgpu_ip_block_version psp_v11_0_ip_block;
404407 int psp_reg_program(struct psp_context *psp, enum psp_reg_prog_id reg,
405408 uint32_t value);
406409 int psp_ring_cmd_submit(struct psp_context *psp,
drivers/gpu/drm/amd/amdgpu/amdgpu_rap.c+4−3
@@ -48,6 +48,7 @@ static ssize_t amdgpu_rap_debugfs_write(struct file *f, const char __user *buf, · 47 unmodified lines
4848 struct ta_rap_cmd_output_data *rap_cmd_output;
4949 struct drm_device *dev = adev_to_drm(adev);
5050 uint32_t op;
51+ enum ta_rap_status status;
5152 int ret;
5253
5354 if (*pos || size != 2)
@@ -70,9 +71,8 @@ static ssize_t amdgpu_rap_debugfs_write(struct file *f, const char __user *buf, · 16 unmodified lines
7071
7172 switch (op) {
7273 case 2:
73− ret = psp_rap_invoke(&adev->psp, op);
74−
75− if (ret == TA_RAP_STATUS__SUCCESS) {
74+ ret = psp_rap_invoke(&adev->psp, op, &status);
75+ if (!ret && status == TA_RAP_STATUS__SUCCESS) {
7676 dev_info(adev->dev, "RAP L0 validate test success.\n");
7777 } else {
7878 rap_shared_mem = (struct ta_rap_shared_memory *)
@@ -97,6 +97,7 @@ static ssize_t amdgpu_rap_debugfs_write(struct file *f, const char __user *buf, · 18 unmodified lines
9797 default:
9898 dev_info(adev->dev, "Unsupported op id: %d, ", op);
9999 dev_info(adev->dev, "Only support op 2(L0 validate test).\n");
100+ break;
100101 }
101102
102103 amdgpu_gfx_off_ctrl(adev, true);
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c+119−72
@@ -109,7 +109,7 @@ static ssize_t amdgpu_ras_debugfs_read(struct file *f, char __user *buf, · 108 unmodified lines
109109 ssize_t s;
110110 char val[128];
111111
112− if (amdgpu_ras_error_query(obj->adev, &info))
112+ if (amdgpu_ras_query_error_status(obj->adev, &info))
113113 return -EINVAL;
114114
115115 s = snprintf(val, sizeof(val), "%s: %lu\n%s: %lu\n",
@@ -434,7 +434,7 @@ static ssize_t amdgpu_ras_sysfs_read(struct device *dev, · 318 unmodified lines
434434 return snprintf(buf, PAGE_SIZE,
435435 "Query currently inaccessible\n");
436436
437− if (amdgpu_ras_error_query(obj->adev, &info))
437+ if (amdgpu_ras_query_error_status(obj->adev, &info))
438438 return -EINVAL;
439439
440440 return snprintf(buf, PAGE_SIZE, "%s: %lu\n%s: %lu\n",
@@ -463,7 +463,7 @@ static struct ras_manager *amdgpu_ras_create_obj(struct amdgpu_device *adev, · 22 unmodified lines
463463 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
464464 struct ras_manager *obj;
465465
466− if (!con)
466+ if (!adev->ras_features || !con)
467467 return NULL;
468468
469469 if (head->block >= AMDGPU_RAS_BLOCK_COUNT)
@@ -490,7 +490,7 @@ struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev, · 20 unmodified lines
490490 struct ras_manager *obj;
491491 int i;
492492
493− if (!con)
493+ if (!adev->ras_features || !con)
494494 return NULL;
495495
496496 if (head) {
@@ -590,7 +590,11 @@ static int __amdgpu_ras_feature_enable(struct amdgpu_device *adev, · 93 unmodified lines
590590 con->features |= BIT(head->block);
591591 } else {
592592 if (obj && amdgpu_ras_is_feature_enabled(adev, head)) {
593− con->features &= ~BIT(head->block);
593+ /* skip clean gfx ras context feature for VEGA20 Gaming.
594+ * will clean later
595+ */
596+ if (!(!adev->ras_features && con->features & BIT(AMDGPU_RAS_BLOCK__GFX)))
597+ con->features &= ~BIT(head->block);
594598 put_obj(obj);
595599 }
596600 }
@@ -693,6 +697,10 @@ int amdgpu_ras_feature_enable_on_boot(struct amdgpu_device *adev, · 96 unmodified lines
693697 if (ret)
694698 return ret;
695699
700+ /* gfx block ras dsiable cmd must send to ras-ta */
701+ if (head->block == AMDGPU_RAS_BLOCK__GFX)
702+ con->features |= BIT(head->block);
703+
696704 ret = amdgpu_ras_feature_enable(adev, head, 0);
697705 }
698706 } else
@@ -757,8 +765,8 @@ static int amdgpu_ras_enable_all_features(struct amdgpu_device *adev, · 58 unmodified lines
757765 /* feature ctl end */
758766
759767 /* query/inject/cure begin */
760−int amdgpu_ras_error_query(struct amdgpu_device *adev,
761− struct ras_query_if *info)
768+int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
769+ struct ras_query_if *info)
762770 {
763771 struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head);
764772 struct ras_err_data err_data = {0, 0, 0, NULL};
@@ -787,10 +795,16 @@ int amdgpu_ras_error_query(struct amdgpu_device *adev, · 22 unmodified lines
787795 case AMDGPU_RAS_BLOCK__GFX:
788796 if (adev->gfx.funcs->query_ras_error_count)
789797 adev->gfx.funcs->query_ras_error_count(adev, &err_data);
798+
799+ if (adev->gfx.funcs->query_ras_error_status)
800+ adev->gfx.funcs->query_ras_error_status(adev);
790801 break;
791802 case AMDGPU_RAS_BLOCK__MMHUB:
792803 if (adev->mmhub.funcs->query_ras_error_count)
793804 adev->mmhub.funcs->query_ras_error_count(adev, &err_data);
805+
806+ if (adev->mmhub.funcs->query_ras_error_status)
807+ adev->mmhub.funcs->query_ras_error_status(adev);
794808 break;
795809 case AMDGPU_RAS_BLOCK__PCIE_BIF:
796810 if (adev->nbio.funcs->query_ras_error_count)
@@ -826,6 +840,35 @@ int amdgpu_ras_error_query(struct amdgpu_device *adev, · 29 unmodified lines
826840 return 0;
827841 }
828842
843+int amdgpu_ras_reset_error_status(struct amdgpu_device *adev,
844+ enum amdgpu_ras_block block)
845+{
846+ if (!amdgpu_ras_is_supported(adev, block))
847+ return -EINVAL;
848+
849+ switch (block) {
850+ case AMDGPU_RAS_BLOCK__GFX:
851+ if (adev->gfx.funcs->reset_ras_error_count)
852+ adev->gfx.funcs->reset_ras_error_count(adev);
853+
854+ if (adev->gfx.funcs->reset_ras_error_status)
855+ adev->gfx.funcs->reset_ras_error_status(adev);
856+ break;
857+ case AMDGPU_RAS_BLOCK__MMHUB:
858+ if (adev->mmhub.funcs->reset_ras_error_count)
859+ adev->mmhub.funcs->reset_ras_error_count(adev);
860+ break;
861+ case AMDGPU_RAS_BLOCK__SDMA:
862+ if (adev->sdma.funcs->reset_ras_error_count)
863+ adev->sdma.funcs->reset_ras_error_count(adev);
864+ break;
865+ default:
866+ break;
867+ }
868+
869+ return 0;
870+}
871+
829872 /* Trigger XGMI/WAFL error */
830873 static int amdgpu_ras_error_inject_xgmi(struct amdgpu_device *adev,
831874 struct ta_ras_trigger_error_input *block_info)
@@ -913,7 +956,7 @@ unsigned long amdgpu_ras_query_error_count(struct amdgpu_device *adev, · 81 unmodified lines
913956 struct ras_manager *obj;
914957 struct ras_err_data data = {0, 0};
915958
916− if (!con)
959+ if (!adev->ras_features || !con)
917960 return 0;
918961
919962 list_for_each_entry(obj, &con->head, node) {
@@ -921,7 +964,7 @@ unsigned long amdgpu_ras_query_error_count(struct amdgpu_device *adev, · 1 unmodified line
921964 .head = obj->head,
922965 };
923966
924− if (amdgpu_ras_error_query(adev, &info))
967+ if (amdgpu_ras_query_error_status(adev, &info))
925968 return 0;
926969
927970 data.ce_count += info.ce_count;
@@ -1137,16 +1180,17 @@ static int amdgpu_ras_sysfs_remove_all(struct amdgpu_device *adev) · 209 unmodified lines
11371180 *
11381181 */
11391182 /* debugfs begin */
1140−static void amdgpu_ras_debugfs_create_ctrl_node(struct amdgpu_device *adev)
1183+static struct dentry *amdgpu_ras_debugfs_create_ctrl_node(struct amdgpu_device *adev)
11411184 {
11421185 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
1186+ struct dentry *dir;
11431187 struct drm_minor *minor = adev_to_drm(adev)->primary;
11441188
1145− con->dir = debugfs_create_dir(RAS_FS_NAME, minor->debugfs_root);
1146− debugfs_create_file("ras_ctrl", S_IWUGO | S_IRUGO, con->dir,
1147− adev, &amdgpu_ras_debugfs_ctrl_ops);
1148− debugfs_create_file("ras_eeprom_reset", S_IWUGO | S_IRUGO, con->dir,
1149− adev, &amdgpu_ras_debugfs_eeprom_ops);
1189+ dir = debugfs_create_dir(RAS_FS_NAME, minor->debugfs_root);
1190+ debugfs_create_file("ras_ctrl", S_IWUGO | S_IRUGO, dir, adev,
1191+ &amdgpu_ras_debugfs_ctrl_ops);
1192+ debugfs_create_file("ras_eeprom_reset", S_IWUGO | S_IRUGO, dir, adev,
1193+ &amdgpu_ras_debugfs_eeprom_ops);
11501194
11511195 /*
11521196 * After one uncorrectable error happens, usually GPU recovery will
@@ -1156,24 +1200,24 @@ static void amdgpu_ras_debugfs_create_ctrl_node(struct amdgpu_device *adev) · 3 unmodified lines
11561200 * ERREVENT_ATHUB_INTERRUPT generated. Normal GPU recovery routine
11571201 * will never be called.
11581202 */
1159− debugfs_create_bool("auto_reboot", S_IWUGO | S_IRUGO, con->dir,
1160− &con->reboot);
1203+ debugfs_create_bool("auto_reboot", S_IWUGO | S_IRUGO, dir, &con->reboot);
11611204
11621205 /*
11631206 * User could set this not to clean up hardware's error count register
11641207 * of RAS IPs during ras recovery.
11651208 */
1166− debugfs_create_bool("disable_ras_err_cnt_harvest", 0644,
1167− con->dir, &con->disable_ras_err_cnt_harvest);
1209+ debugfs_create_bool("disable_ras_err_cnt_harvest", 0644, dir,
1210+ &con->disable_ras_err_cnt_harvest);
1211+ return dir;
11681212 }
11691213
11701214 static void amdgpu_ras_debugfs_create(struct amdgpu_device *adev,
1171− struct ras_fs_if *head)
1215+ struct ras_fs_if *head,
1216+ struct dentry *dir)
11721217 {
1173− struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
11741218 struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head->head);
11751219
1176− if (!obj || obj->ent)
1220+ if (!obj || !dir)
11771221 return;
11781222
11791223 get_obj(obj);
@@ -1182,14 +1226,14 @@ static void amdgpu_ras_debugfs_create(struct amdgpu_device *adev, · 2 unmodified lines
11821226 head->debugfs_name,
11831227 sizeof(obj->fs_data.debugfs_name));
11841228
1185− obj->ent = debugfs_create_file(obj->fs_data.debugfs_name,
1186− S_IWUGO | S_IRUGO, con->dir, obj,
1187− &amdgpu_ras_debugfs_ops);
1229+ debugfs_create_file(obj->fs_data.debugfs_name, S_IWUGO | S_IRUGO, dir,
1230+ obj, &amdgpu_ras_debugfs_ops);
11881231 }
11891232
11901233 void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev)
11911234 {
11921235 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
1236+ struct dentry *dir;
11931237 struct ras_manager *obj;
11941238 struct ras_fs_if fs_info;
11951239
@@ -1200,7 +1244,7 @@ void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev) · 4 unmodified lines
12001244 if (!IS_ENABLED(CONFIG_DEBUG_FS) || !con)
12011245 return;
12021246
1203− amdgpu_ras_debugfs_create_ctrl_node(adev);
1247+ dir = amdgpu_ras_debugfs_create_ctrl_node(adev);
12041248
12051249 list_for_each_entry(obj, &con->head, node) {
12061250 if (amdgpu_ras_is_supported(adev, obj->head.block) &&
@@ -1208,34 +1252,11 @@ void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev) · 1 unmodified line
12081252 sprintf(fs_info.debugfs_name, "%s_err_inject",
12091253 ras_block_str(obj->head.block));
12101254 fs_info.head = obj->head;
1211− amdgpu_ras_debugfs_create(adev, &fs_info);
1255+ amdgpu_ras_debugfs_create(adev, &fs_info, dir);
12121256 }
12131257 }
12141258 }
12151259
1216−static void amdgpu_ras_debugfs_remove(struct amdgpu_device *adev,
1217− struct ras_common_if *head)
1218−{
1219− struct ras_manager *obj = amdgpu_ras_find_obj(adev, head);
1220−
1221− if (!obj || !obj->ent)
1222− return;
1223−
1224− obj->ent = NULL;
1225− put_obj(obj);
1226−}
1227−
1228−static void amdgpu_ras_debugfs_remove_all(struct amdgpu_device *adev)
1229−{
1230− struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
1231− struct ras_manager *obj, *tmp;
1232−
1233− list_for_each_entry_safe(obj, tmp, &con->head, node) {
1234− amdgpu_ras_debugfs_remove(adev, &obj->head);
1235− }
1236−
1237− con->dir = NULL;
1238−}
12391260 /* debugfs end */
12401261
12411262 /* ras fs */
@@ -1282,8 +1303,17 @@ static int amdgpu_ras_fs_init(struct amdgpu_device *adev) · 40 unmodified lines
12821303
12831304 static int amdgpu_ras_fs_fini(struct amdgpu_device *adev)
12841305 {
1285− if (IS_ENABLED(CONFIG_DEBUG_FS))
1286− amdgpu_ras_debugfs_remove_all(adev);
1306+ struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
1307+ struct ras_manager *con_obj, *ip_obj, *tmp;
1308+
1309+ if (IS_ENABLED(CONFIG_DEBUG_FS)) {
1310+ list_for_each_entry_safe(con_obj, tmp, &con->head, node) {
1311+ ip_obj = amdgpu_ras_find_obj(adev, &con_obj->head);
1312+ if (ip_obj)
1313+ put_obj(ip_obj);
1314+ }
1315+ }
1316+
12871317 amdgpu_ras_sysfs_remove_all(adev);
12881318 return 0;
12891319 }
@@ -1447,7 +1477,7 @@ static void amdgpu_ras_log_on_err_counter(struct amdgpu_device *adev) · 157 unmodified lines
14471477 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
14481478 struct ras_manager *obj;
14491479
1450− if (!con)
1480+ if (!adev->ras_features || !con)
14511481 return;
14521482
14531483 list_for_each_entry(obj, &con->head, node) {
@@ -1464,7 +1494,7 @@ static void amdgpu_ras_log_on_err_counter(struct amdgpu_device *adev) · 10 unmodified lines
14641494 if (info.head.block == AMDGPU_RAS_BLOCK__PCIE_BIF)
14651495 continue;
14661496
1467− amdgpu_ras_error_query(adev, &info);
1497+ amdgpu_ras_query_error_status(adev, &info);
14681498 }
14691499 }
14701500
@@ -1495,7 +1525,7 @@ static void amdgpu_ras_query_err_status(struct amdgpu_device *adev) · 24 unmodified lines
14951525 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
14961526 struct ras_manager *obj;
14971527
1498− if (!con)
1528+ if (!adev->ras_features || !con)
14991529 return;
15001530
15011531 list_for_each_entry(obj, &con->head, node) {
@@ -1809,7 +1839,7 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) · 307 unmodified lines
18091839 bool exc_err_limit = false;
18101840 int ret;
18111841
1812− if (con)
1842+ if (adev->ras_features && con)
18131843 data = &con->eh_data;
18141844 else
18151845 return 0;
@@ -1828,6 +1858,12 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) · 12 unmodified lines
18281858 max_eeprom_records_len = amdgpu_ras_eeprom_get_record_max_length();
18291859 amdgpu_ras_validate_threshold(adev, max_eeprom_records_len);
18301860
1861+ /* Todo: During test the SMU might fail to read the eeprom through I2C
1862+ * when the GPU is pending on XGMI reset during probe time
1863+ * (Mostly after second bus reset), skip it now
1864+ */
1865+ if (adev->gmc.xgmi.pending_reset)
1866+ return 0;
18311867 ret = amdgpu_ras_eeprom_init(&con->eeprom_control, &exc_err_limit);
18321868 /*
18331869 * This calling fails when exc_err_limit is true or
@@ -1928,11 +1964,11 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev, · 94 unmodified lines
19281964 return;
19291965
19301966 if (amdgpu_atomfirmware_mem_ecc_supported(adev)) {
1931− dev_info(adev->dev, "HBM ECC is active.\n");
1967+ dev_info(adev->dev, "MEM ECC is active.\n");
19321968 *hw_supported |= (1 << AMDGPU_RAS_BLOCK__UMC |
19331969 1 << AMDGPU_RAS_BLOCK__DF);
19341970 } else
1935− dev_info(adev->dev, "HBM ECC is not presented.\n");
1971+ dev_info(adev->dev, "MEM ECC is not presented.\n");
19361972
19371973 if (amdgpu_atomfirmware_sram_ecc_supported(adev)) {
19381974 dev_info(adev->dev, "SRAM ECC is active.\n");
@@ -1970,6 +2006,15 @@ int amdgpu_ras_init(struct amdgpu_device *adev) · 31 unmodified lines
19702006 amdgpu_ras_check_supported(adev, &con->hw_supported,
19712007 &con->supported);
19722008 if (!con->hw_supported || (adev->asic_type == CHIP_VEGA10)) {
2009+ /* set gfx block ras context feature for VEGA20 Gaming
2010+ * send ras disable cmd to ras ta during ras late init.
2011+ */
2012+ if (!adev->ras_features && adev->asic_type == CHIP_VEGA20) {
2013+ con->features |= BIT(AMDGPU_RAS_BLOCK__GFX);
2014+
2015+ return 0;
2016+ }
2017+
19732018 r = 0;
19742019 goto release_con;
19752020 }
@@ -2083,8 +2128,12 @@ void amdgpu_ras_resume(struct amdgpu_device *adev) · 107 unmodified lines
20832128 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
20842129 struct ras_manager *obj, *tmp;
20852130
2086− if (!con)
2131+ if (!adev->ras_features || !con) {
2132+ /* clean ras context for VEGA20 Gaming after send ras disable cmd */
2133+ amdgpu_release_ras_context(adev);
2134+
20872135 return;
2136+ }
20882137
20892138 if (con->flags & AMDGPU_RAS_FLAG_INIT_BY_VBIOS) {
20902139 /* Set up all other IPs which are not implemented. There is a
@@ -2125,7 +2174,7 @@ void amdgpu_ras_suspend(struct amdgpu_device *adev) · 34 unmodified lines
21252174 {
21262175 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
21272176
2128− if (!con)
2177+ if (!adev->ras_features || !con)
21292178 return;
21302179
21312180 amdgpu_ras_disable_all_features(adev, 0);
@@ -2139,7 +2188,7 @@ int amdgpu_ras_pre_fini(struct amdgpu_device *adev) · 7 unmodified lines
21392188 {
21402189 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
21412190
2142− if (!con)
2191+ if (!adev->ras_features || !con)
21432192 return 0;
21442193
21452194 /* Need disable ras on all IPs here before ip [hw/sw]fini */
@@ -2152,7 +2201,7 @@ int amdgpu_ras_fini(struct amdgpu_device *adev) · 6 unmodified lines
21522201 {
21532202 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
21542203
2155− if (!con)
2204+ if (!adev->ras_features || !con)
21562205 return 0;
21572206
21582207 amdgpu_ras_fs_fini(adev);
@@ -2196,18 +2245,16 @@ bool amdgpu_ras_need_emergency_restart(struct amdgpu_device *adev) · 37 unmodified lines
21962245 return false;
21972246 }
21982247
2199−bool amdgpu_ras_check_err_threshold(struct amdgpu_device *adev)
2248+void amdgpu_release_ras_context(struct amdgpu_device *adev)
22002249 {
22012250 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
2202− bool exc_err_limit = false;
22032251
2204− if (con && (amdgpu_bad_page_threshold != 0))
2205− amdgpu_ras_eeprom_check_err_threshold(&con->eeprom_control,
2206− &exc_err_limit);
2252+ if (!con)
2253+ return;
22072254
2208− /*
2209− * We are only interested in variable exc_err_limit,
2210− * as it says if GPU is in bad state or not.
2211− */
2212− return exc_err_limit;
2255+ if (!adev->ras_features && con->features & BIT(AMDGPU_RAS_BLOCK__GFX)) {
2256+ con->features &= ~BIT(AMDGPU_RAS_BLOCK__GFX);
2257+ amdgpu_ras_set_context(adev, NULL);
2258+ kfree(con);
2259+ }
22132260 }
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h+6−7
@@ -318,8 +318,6 @@ struct amdgpu_ras { · 317 unmodified lines
318318 uint32_t supported;
319319 uint32_t features;
320320 struct list_head head;
321− /* debugfs */
322− struct dentry *dir;
323321 /* sysfs */
324322 struct device_attribute features_attr;
325323 struct bin_attribute badpages_attr;
@@ -395,8 +393,6 @@ struct ras_manager { · 69 unmodified lines
395393 struct list_head node;
396394 /* the device */
397395 struct amdgpu_device *adev;
398− /* debugfs */
399− struct dentry *ent;
400396 /* sysfs */
401397 struct device_attribute sysfs_attr;
402398 int attr_inuse;
@@ -495,8 +491,6 @@ void amdgpu_ras_suspend(struct amdgpu_device *adev); · 92 unmodified lines
495491 unsigned long amdgpu_ras_query_error_count(struct amdgpu_device *adev,
496492 bool is_ce);
497493
498−bool amdgpu_ras_check_err_threshold(struct amdgpu_device *adev);
499−
500494 /* error handling functions */
501495 int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
502496 struct eeprom_table_record *bps, int pages);
@@ -594,9 +588,12 @@ int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev, · 91 unmodified lines
594588
595589 void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev);
596590
597−int amdgpu_ras_error_query(struct amdgpu_device *adev,
591+int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
598592 struct ras_query_if *info);
599593
594+int amdgpu_ras_reset_error_status(struct amdgpu_device *adev,
595+ enum amdgpu_ras_block block);
596+
600597 int amdgpu_ras_error_inject(struct amdgpu_device *adev,
601598 struct ras_inject_if *info);
602599
@@ -629,4 +626,6 @@ void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev); · 26 unmodified lines
629626 void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool ready);
630627
631628 bool amdgpu_ras_need_emergency_restart(struct amdgpu_device *adev);
629+
630+void amdgpu_release_ras_context(struct amdgpu_device *adev);
632631 #endif
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c+12−31
@@ -434,47 +434,28 @@ static uint32_t __correct_eeprom_dest_address(uint32_t curr_address) · 433 unmodified lines
434434 return curr_address;
435435 }
436436
437−int amdgpu_ras_eeprom_check_err_threshold(
438− struct amdgpu_ras_eeprom_control *control,
439− bool *exceed_err_limit)
437+bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
440438 {
441− struct amdgpu_device *adev = to_amdgpu_device(control);
442− unsigned char buff[EEPROM_ADDRESS_SIZE +
443− EEPROM_TABLE_HEADER_SIZE] = { 0 };
444− struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
445− struct i2c_msg msg = {
446− .addr = control->i2c_address,
447− .flags = I2C_M_RD,
448− .len = EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
449− .buf = buff,
450− };
451− int ret;
452−
453− *exceed_err_limit = false;
439+ struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
454440
455441 if (!__is_ras_eeprom_supported(adev))
456− return 0;
457−
458− /* read EEPROM table header */
459− mutex_lock(&control->tbl_mutex);
460− ret = i2c_transfer(&adev->pm.smu_i2c, &msg, 1);
461− if (ret < 1) {
462− dev_err(adev->dev, "Failed to read EEPROM table header.\n");
463− goto err;
464− }
442+ return false;
465443
466− __decode_table_header_from_buff(hdr, &buff[2]);
444+ /* skip check eeprom table for VEGA20 Gaming */
445+ if (!con)
446+ return false;
447+ else
448+ if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
449+ return false;
467450
468− if (hdr->header == EEPROM_TABLE_HDR_BAD) {
451+ if (con->eeprom_control.tbl_hdr.header == EEPROM_TABLE_HDR_BAD) {
469452 dev_warn(adev->dev, "This GPU is in BAD status.");
470453 dev_warn(adev->dev, "Please retire it or setting one bigger "
471454 "threshold value when reloading driver.\n");
472− *exceed_err_limit = true;
455+ return true;
473456 }
474457
475−err:
476− mutex_unlock(&control->tbl_mutex);
477− return 0;
458+ return false;
478459 }
479460
480461 int amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control *control,
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h+1−3
@@ -80,9 +80,7 @@ int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control, · 79 unmodified lines
8080 bool *exceed_err_limit);
8181 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control);
8282
83−int amdgpu_ras_eeprom_check_err_threshold(
84− struct amdgpu_ras_eeprom_control *control,
85− bool *exceed_err_limit);
83+bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev);
8684
8785 int amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control *control,
8886 struct eeprom_table_record *records,
drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h+105−0
@@ -0,0 +1,105 @@
1+// SPDX-License-Identifier: GPL-2.0 OR MIT
2+/*
3+ * Copyright 2020 Advanced Micro Devices, Inc.
4+ *
5+ * Permission is hereby granted, free of charge, to any person obtaining a
6+ * copy of this software and associated documentation files (the "Software"),
7+ * to deal in the Software without restriction, including without limitation
8+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9+ * and/or sell copies of the Software, and to permit persons to whom the
10+ * Software is furnished to do so, subject to the following conditions:
11+ *
12+ * The above copyright notice and this permission notice shall be included in
13+ * all copies or substantial portions of the Software.
14+ *
15+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21+ * OTHER DEALINGS IN THE SOFTWARE.
22+ *
23+ * Authors: Christian König
24+ */
25+
26+#ifndef __AMDGPU_RES_CURSOR_H__
27+#define __AMDGPU_RES_CURSOR_H__
28+
29+#include <drm/drm_mm.h>
30+#include <drm/ttm/ttm_resource.h>
31+
32+/* state back for walking over vram_mgr and gtt_mgr allocations */
33+struct amdgpu_res_cursor {
34+ uint64_t start;
35+ uint64_t size;
36+ uint64_t remaining;
37+ struct drm_mm_node *node;
38+};
39+
40+/**
41+ * amdgpu_res_first - initialize a amdgpu_res_cursor
42+ *
43+ * @res: TTM resource object to walk
44+ * @start: Start of the range
45+ * @size: Size of the range
46+ * @cur: cursor object to initialize
47+ *
48+ * Start walking over the range of allocations between @start and @size.
49+ */
50+static inline void amdgpu_res_first(struct ttm_resource *res,
51+ uint64_t start, uint64_t size,
52+ struct amdgpu_res_cursor *cur)
53+{
54+ struct drm_mm_node *node;
55+
56+ if (!res || !res->mm_node) {
57+ cur->start = start;
58+ cur->size = size;
59+ cur->remaining = size;
60+ cur->node = NULL;
61+ return;
62+ }
63+
64+ BUG_ON(start + size > res->num_pages << PAGE_SHIFT);
65+
66+ node = res->mm_node;
67+ while (start > node->size << PAGE_SHIFT)
68+ start -= node++->size << PAGE_SHIFT;
69+
70+ cur->start = (node->start << PAGE_SHIFT) + start;
71+ cur->size = min((node->size << PAGE_SHIFT) - start, size);
72+ cur->remaining = size;
73+ cur->node = node;
74+}
75+
76+/**
77+ * amdgpu_res_next - advance the cursor
78+ *
79+ * @cur: the cursor to advance
80+ * @size: number of bytes to move forward
81+ *
82+ * Move the cursor @size bytes forwrad, walking to the next node if necessary.
83+ */
84+static inline void amdgpu_res_next(struct amdgpu_res_cursor *cur, uint64_t size)
85+{
86+ struct drm_mm_node *node = cur->node;
87+
88+ BUG_ON(size > cur->remaining);
89+
90+ cur->remaining -= size;
91+ if (!cur->remaining)
92+ return;
93+
94+ cur->size -= size;
95+ if (cur->size) {
96+ cur->start += size;
97+ return;
98+ }
99+
100+ cur->node = ++node;
101+ cur->start = node->start << PAGE_SHIFT;
102+ cur->size = min(node->size << PAGE_SHIFT, cur->remaining);
103+}
104+
105+#endif
drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h+5−0
@@ -64,6 +64,11 @@ struct amdgpu_sdma { · 63 unmodified lines
6464 struct amdgpu_irq_src trap_irq;
6565 struct amdgpu_irq_src illegal_inst_irq;
6666 struct amdgpu_irq_src ecc_irq;
67+ struct amdgpu_irq_src vm_hole_irq;
68+ struct amdgpu_irq_src doorbell_invalid_irq;
69+ struct amdgpu_irq_src pool_timeout_irq;
70+ struct amdgpu_irq_src srbm_write_irq;
71+
6772 int num_instances;
6873 uint32_t srbm_soft_reset;
6974 bool has_page_queue;
drivers/gpu/drm/amd/amdgpu/amdgpu_securedisplay.c+3−0
@@ -69,6 +69,9 @@ void psp_securedisplay_parse_resp_status(struct psp_context *psp, · 68 unmodified lines
6969 case TA_SECUREDISPLAY_STATUS__READ_CRC_ERROR:
7070 dev_err(psp->adev->dev, "Secure display: Failed to Read CRC");
7171 break;
72+ case TA_SECUREDISPLAY_STATUS__I2C_INIT_ERROR:
73+ dev_err(psp->adev->dev, "Secure display: Failed to initialize I2C.");
74+ break;
7275 default:
7376 dev_err(psp->adev->dev, "Secure display: Failed to parse status: %d\n", status);
7477 }
drivers/gpu/drm/amd/amdgpu/amdgpu_smuio.h+2−0
@@ -28,6 +28,8 @@ struct amdgpu_smuio_funcs { · 27 unmodified lines
2828 u32 (*get_rom_data_offset)(struct amdgpu_device *adev);
2929 void (*update_rom_clock_gating)(struct amdgpu_device *adev, bool enable);
3030 void (*get_clock_gating_state)(struct amdgpu_device *adev, u32 *flags);
31+ u32 (*get_die_id)(struct amdgpu_device *adev);
32+ bool (*is_host_gpu_xgmi_supported)(struct amdgpu_device *adev);
3133 };
3234
3335 struct amdgpu_smuio {
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c+171−287
@@ -47,7 +47,6 @@ · 46 unmodified lines
4747 #include <drm/ttm/ttm_bo_driver.h>
4848 #include <drm/ttm/ttm_placement.h>
4949
50−#include <drm/drm_debugfs.h>
5150 #include <drm/amdgpu_drm.h>
5251
5352 #include "amdgpu.h"
@@ -57,6 +56,7 @@ · 3 unmodified lines
5756 #include "amdgpu_sdma.h"
5857 #include "amdgpu_ras.h"
5958 #include "amdgpu_atomfirmware.h"
59+#include "amdgpu_res_cursor.h"
6060 #include "bif/bif_4_1_d.h"
6161
6262 #define AMDGPU_TTM_VRAM_MAX_DW_READ (size_t)128
@@ -178,55 +178,12 @@ static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp) · 115 unmodified lines
178178 filp->private_data);
179179 }
180180
181−/**
182− * amdgpu_mm_node_addr - Compute the GPU relative offset of a GTT buffer.
183− *
184− * @bo: The bo to assign the memory to.
185− * @mm_node: Memory manager node for drm allocator.
186− * @mem: The region where the bo resides.
187− *
188− */
189−static uint64_t amdgpu_mm_node_addr(struct ttm_buffer_object *bo,
190− struct drm_mm_node *mm_node,
191− struct ttm_resource *mem)
192−{
193− uint64_t addr = 0;
194−
195− if (mm_node->start != AMDGPU_BO_INVALID_OFFSET) {
196− addr = mm_node->start << PAGE_SHIFT;
197− addr += amdgpu_ttm_domain_start(amdgpu_ttm_adev(bo->bdev),
198− mem->mem_type);
199− }
200− return addr;
201−}
202−
203−/**
204− * amdgpu_find_mm_node - Helper function finds the drm_mm_node corresponding to
205− * @offset. It also modifies the offset to be within the drm_mm_node returned
206− *
207− * @mem: The region where the bo resides.
208− * @offset: The offset that drm_mm_node is used for finding.
209− *
210− */
211−static struct drm_mm_node *amdgpu_find_mm_node(struct ttm_resource *mem,
212− uint64_t *offset)
213−{
214− struct drm_mm_node *mm_node = mem->mm_node;
215−
216− while (*offset >= (mm_node->size << PAGE_SHIFT)) {
217− *offset -= (mm_node->size << PAGE_SHIFT);
218− ++mm_node;
219− }
220− return mm_node;
221−}
222−
223181 /**
224182 * amdgpu_ttm_map_buffer - Map memory into the GART windows
225183 * @bo: buffer object to map
226184 * @mem: memory object to map
227− * @mm_node: drm_mm node object to map
185+ * @mm_cur: range to map
228186 * @num_pages: number of pages to map
229− * @offset: offset into @mm_node where to start
230187 * @window: which GART window to use
231188 * @ring: DMA ring to use for the copy
232189 * @tmz: if we should setup a TMZ enabled mapping
@@ -237,10 +194,10 @@ static struct drm_mm_node *amdgpu_find_mm_node(struct ttm_resource *mem, · 4 unmodified lines
237194 */
238195 static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo,
239196 struct ttm_resource *mem,
240− struct drm_mm_node *mm_node,
241− unsigned num_pages, uint64_t offset,
242− unsigned window, struct amdgpu_ring *ring,
243− bool tmz, uint64_t *addr)
197+ struct amdgpu_res_cursor *mm_cur,
198+ unsigned num_pages, unsigned window,
199+ struct amdgpu_ring *ring, bool tmz,
200+ uint64_t *addr)
244201 {
245202 struct amdgpu_device *adev = ring->adev;
246203 struct amdgpu_job *job;
@@ -257,14 +214,15 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, · 10 unmodified lines
257214
258215 /* Map only what can't be accessed directly */
259216 if (!tmz && mem->start != AMDGPU_BO_INVALID_OFFSET) {
260− *addr = amdgpu_mm_node_addr(bo, mm_node, mem) + offset;
217+ *addr = amdgpu_ttm_domain_start(adev, mem->mem_type) +
218+ mm_cur->start;
261219 return 0;
262220 }
263221
264222 *addr = adev->gmc.gart_start;
265223 *addr += (u64)window * AMDGPU_GTT_MAX_TRANSFER_SIZE *
266224 AMDGPU_GPU_PAGE_SIZE;
267− *addr += offset & ~PAGE_MASK;
225+ *addr += mm_cur->start & ~PAGE_MASK;
268226
269227 num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8);
270228 num_bytes = num_pages * 8;
@@ -292,17 +250,17 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, · 21 unmodified lines
292250 cpu_addr = &job->ibs[0].ptr[num_dw];
293251
294252 if (mem->mem_type == TTM_PL_TT) {
295− dma_addr_t *dma_address;
253+ dma_addr_t *dma_addr;
296254
297− dma_address = &bo->ttm->dma_address[offset >> PAGE_SHIFT];
298− r = amdgpu_gart_map(adev, 0, num_pages, dma_address, flags,
255+ dma_addr = &bo->ttm->dma_address[mm_cur->start >> PAGE_SHIFT];
256+ r = amdgpu_gart_map(adev, 0, num_pages, dma_addr, flags,
299257 cpu_addr);
300258 if (r)
301259 goto error_free;
302260 } else {
303261 dma_addr_t dma_address;
304262
305− dma_address = (mm_node->start << PAGE_SHIFT) + offset;
263+ dma_address = mm_cur->start;
306264 dma_address += adev->vm_manager.vram_base_offset;
307265
308266 for (i = 0; i < num_pages; ++i) {
@@ -354,9 +312,8 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, · 45 unmodified lines
354312 const uint32_t GTT_MAX_BYTES = (AMDGPU_GTT_MAX_TRANSFER_SIZE *
355313 AMDGPU_GPU_PAGE_SIZE);
356314
357− uint64_t src_node_size, dst_node_size, src_offset, dst_offset;
358315 struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
359− struct drm_mm_node *src_mm, *dst_mm;
316+ struct amdgpu_res_cursor src_mm, dst_mm;
360317 struct dma_fence *fence = NULL;
361318 int r = 0;
362319
@@ -365,29 +322,13 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, · 2 unmodified lines
365322 return -EINVAL;
366323 }
367324
368− src_offset = src->offset;
369− if (src->mem->mm_node) {
370− src_mm = amdgpu_find_mm_node(src->mem, &src_offset);
371− src_node_size = (src_mm->size << PAGE_SHIFT) - src_offset;
372− } else {
373− src_mm = NULL;
374− src_node_size = ULLONG_MAX;
375− }
376−
377− dst_offset = dst->offset;
378− if (dst->mem->mm_node) {
379− dst_mm = amdgpu_find_mm_node(dst->mem, &dst_offset);
380− dst_node_size = (dst_mm->size << PAGE_SHIFT) - dst_offset;
381− } else {
382− dst_mm = NULL;
383− dst_node_size = ULLONG_MAX;
384− }
325+ amdgpu_res_first(src->mem, src->offset, size, &src_mm);
326+ amdgpu_res_first(dst->mem, dst->offset, size, &dst_mm);
385327
386328 mutex_lock(&adev->mman.gtt_window_lock);
387−
388− while (size) {
389− uint32_t src_page_offset = src_offset & ~PAGE_MASK;
390− uint32_t dst_page_offset = dst_offset & ~PAGE_MASK;
329+ while (src_mm.remaining) {
330+ uint32_t src_page_offset = src_mm.start & ~PAGE_MASK;
331+ uint32_t dst_page_offset = dst_mm.start & ~PAGE_MASK;
391332 struct dma_fence *next;
392333 uint32_t cur_size;
393334 uint64_t from, to;
@@ -396,19 +337,19 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, · 2 unmodified lines
396337 * begins at an offset, then adjust the size accordingly
397338 */
398339 cur_size = max(src_page_offset, dst_page_offset);
399− cur_size = min(min3(src_node_size, dst_node_size, size),
340+ cur_size = min(min3(src_mm.size, dst_mm.size, size),
400341 (uint64_t)(GTT_MAX_BYTES - cur_size));
401342
402343 /* Map src to window 0 and dst to window 1. */
403− r = amdgpu_ttm_map_buffer(src->bo, src->mem, src_mm,
344+ r = amdgpu_ttm_map_buffer(src->bo, src->mem, &src_mm,
404345 PFN_UP(cur_size + src_page_offset),
405− src_offset, 0, ring, tmz, &from);
346+ 0, ring, tmz, &from);
406347 if (r)
407348 goto error;
408349
409− r = amdgpu_ttm_map_buffer(dst->bo, dst->mem, dst_mm,
350+ r = amdgpu_ttm_map_buffer(dst->bo, dst->mem, &dst_mm,
410351 PFN_UP(cur_size + dst_page_offset),
411− dst_offset, 1, ring, tmz, &to);
352+ 1, ring, tmz, &to);
412353 if (r)
413354 goto error;
414355
@@ -420,27 +361,8 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, · 5 unmodified lines
420361 dma_fence_put(fence);
421362 fence = next;
422363
423− size -= cur_size;
424− if (!size)
425− break;
426−
427− src_node_size -= cur_size;
428− if (!src_node_size) {
429− ++src_mm;
430− src_node_size = src_mm->size << PAGE_SHIFT;
431− src_offset = 0;
432− } else {
433− src_offset += cur_size;
434− }
435−
436− dst_node_size -= cur_size;
437− if (!dst_node_size) {
438− ++dst_mm;
439− dst_node_size = dst_mm->size << PAGE_SHIFT;
440− dst_offset = 0;
441− } else {
442− dst_offset += cur_size;
443− }
364+ amdgpu_res_next(&src_mm, cur_size);
365+ amdgpu_res_next(&dst_mm, cur_size);
444366 }
445367 error:
446368 mutex_unlock(&adev->mman.gtt_window_lock);
@@ -519,7 +441,8 @@ error: · 72 unmodified lines
519441 static bool amdgpu_mem_visible(struct amdgpu_device *adev,
520442 struct ttm_resource *mem)
521443 {
522− struct drm_mm_node *nodes = mem->mm_node;
444+ uint64_t mem_size = (u64)mem->num_pages << PAGE_SHIFT;
445+ struct amdgpu_res_cursor cursor;
523446
524447 if (mem->mem_type == TTM_PL_SYSTEM ||
525448 mem->mem_type == TTM_PL_TT)
@@ -527,12 +450,13 @@ static bool amdgpu_mem_visible(struct amdgpu_device *adev, · 1 unmodified line
527450 if (mem->mem_type != TTM_PL_VRAM)
528451 return false;
529452
453+ amdgpu_res_first(mem, 0, mem_size, &cursor);
454+
530455 /* ttm_resource_ioremap only supports contiguous memory */
531− if (nodes->size != mem->num_pages)
456+ if (cursor.size != mem_size)
532457 return false;
533458
534− return ((nodes->start + nodes->size) << PAGE_SHIFT)
535− <= adev->gmc.visible_vram_size;
459+ return cursor.start + cursor.size <= adev->gmc.visible_vram_size;
536460 }
537461
538462 /*
@@ -674,7 +598,10 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_device *bdev, struct ttm_resourc · 135 unmodified lines
674598
675599 mem->bus.offset += adev->gmc.aper_base;
676600 mem->bus.is_iomem = true;
677− mem->bus.caching = ttm_write_combined;
601+ if (adev->gmc.xgmi.connected_to_cpu)
602+ mem->bus.caching = ttm_cached;
603+ else
604+ mem->bus.caching = ttm_write_combined;
678605 break;
679606 default:
680607 return -EINVAL;
@@ -686,12 +613,10 @@ static unsigned long amdgpu_ttm_io_mem_pfn(struct ttm_buffer_object *bo, · 5 unmodified lines
686613 unsigned long page_offset)
687614 {
688615 struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
689− uint64_t offset = (page_offset << PAGE_SHIFT);
690− struct drm_mm_node *mm;
616+ struct amdgpu_res_cursor cursor;
691617
692− mm = amdgpu_find_mm_node(&bo->mem, &offset);
693− offset += adev->gmc.aper_base;
694− return mm->start + (offset >> PAGE_SHIFT);
618+ amdgpu_res_first(&bo->mem, (u64)page_offset << PAGE_SHIFT, 0, &cursor);
619+ return (adev->gmc.aper_base + cursor.start) >> PAGE_SHIFT;
695620 }
696621
697622 /**
@@ -1430,6 +1355,10 @@ uint64_t amdgpu_ttm_tt_pde_flags(struct ttm_tt *ttm, struct ttm_resource *mem) · 732 unmodified lines
14301355 flags |= AMDGPU_PTE_SNOOPED;
14311356 }
14321357
1358+ if (mem && mem->mem_type == TTM_PL_VRAM &&
1359+ mem->bus.caching == ttm_cached)
1360+ flags |= AMDGPU_PTE_SNOOPED;
1361+
14331362 return flags;
14341363 }
14351364
@@ -1469,7 +1398,7 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo, · 33 unmodified lines
14691398 const struct ttm_place *place)
14701399 {
14711400 unsigned long num_pages = bo->mem.num_pages;
1472− struct drm_mm_node *node = bo->mem.mm_node;
1401+ struct amdgpu_res_cursor cursor;
14731402 struct dma_resv_list *flist;
14741403 struct dma_fence *f;
14751404 int i;
@@ -1501,13 +1430,15 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo, · 25 unmodified lines
15011430
15021431 case TTM_PL_VRAM:
15031432 /* Check each drm MM node individually */
1504− while (num_pages) {
1505− if (place->fpfn < (node->start + node->size) &&
1506− !(place->lpfn && place->lpfn <= node->start))
1433+ amdgpu_res_first(&bo->mem, 0, (u64)num_pages << PAGE_SHIFT,
1434+ &cursor);
1435+ while (cursor.remaining) {
1436+ if (place->fpfn < PFN_DOWN(cursor.start + cursor.size)
1437+ && !(place->lpfn &&
1438+ place->lpfn <= PFN_DOWN(cursor.start)))
15071439 return true;
15081440
1509− num_pages -= node->size;
1510− ++node;
1441+ amdgpu_res_next(&cursor, cursor.size);
15111442 }
15121443 return false;
15131444
@@ -1531,41 +1462,36 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo, · 17 unmodified lines
15311462 * access for debugging purposes.
15321463 */
15331464 static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo,
1534− unsigned long offset,
1535− void *buf, int len, int write)
1465+ unsigned long offset, void *buf, int len,
1466+ int write)
15361467 {
15371468 struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
15381469 struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
1539− struct drm_mm_node *nodes;
1470+ struct amdgpu_res_cursor cursor;
1471+ unsigned long flags;
15401472 uint32_t value = 0;
15411473 int ret = 0;
1542− uint64_t pos;
1543− unsigned long flags;
15441474
15451475 if (bo->mem.mem_type != TTM_PL_VRAM)
15461476 return -EIO;
15471477
1548− pos = offset;
1549− nodes = amdgpu_find_mm_node(&abo->tbo.mem, &pos);
1550− pos += (nodes->start << PAGE_SHIFT);
1551−
1552− while (len && pos < adev->gmc.mc_vram_size) {
1553− uint64_t aligned_pos = pos & ~(uint64_t)3;
1554− uint64_t bytes = 4 - (pos & 3);
1555− uint32_t shift = (pos & 3) * 8;
1478+ amdgpu_res_first(&bo->mem, offset, len, &cursor);
1479+ while (cursor.remaining) {
1480+ uint64_t aligned_pos = cursor.start & ~(uint64_t)3;
1481+ uint64_t bytes = 4 - (cursor.start & 3);
1482+ uint32_t shift = (cursor.start & 3) * 8;
15561483 uint32_t mask = 0xffffffff << shift;
15571484
1558− if (len < bytes) {
1559− mask &= 0xffffffff >> (bytes - len) * 8;
1560− bytes = len;
1485+ if (cursor.size < bytes) {
1486+ mask &= 0xffffffff >> (bytes - cursor.size) * 8;
1487+ bytes = cursor.size;
15611488 }
15621489
15631490 if (mask != 0xffffffff) {
15641491 spin_lock_irqsave(&adev->mmio_idx_lock, flags);
15651492 WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)aligned_pos) | 0x80000000);
15661493 WREG32_NO_KIQ(mmMM_INDEX_HI, aligned_pos >> 31);
1567− if (!write || mask != 0xffffffff)
1568− value = RREG32_NO_KIQ(mmMM_DATA);
1494+ value = RREG32_NO_KIQ(mmMM_DATA);
15691495 if (write) {
15701496 value &= ~mask;
15711497 value |= (*(uint32_t *)buf << shift) & mask;
@@ -1577,21 +1503,15 @@ static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo, · 5 unmodified lines
15771503 memcpy(buf, &value, bytes);
15781504 }
15791505 } else {
1580− bytes = (nodes->start + nodes->size) << PAGE_SHIFT;
1581− bytes = min(bytes - pos, (uint64_t)len & ~0x3ull);
1582−
1583− amdgpu_device_vram_access(adev, pos, (uint32_t *)buf,
1584− bytes, write);
1506+ bytes = cursor.size & 0x3ull;
1507+ amdgpu_device_vram_access(adev, cursor.start,
1508+ (uint32_t *)buf, bytes,
1509+ write);
15851510 }
15861511
15871512 ret += bytes;
15881513 buf = (uint8_t *)buf + bytes;
1589− pos += bytes;
1590− len -= bytes;
1591− if (pos >= (nodes->start + nodes->size) << PAGE_SHIFT) {
1592− ++nodes;
1593− pos = (nodes->start << PAGE_SHIFT);
1594− }
1514+ amdgpu_res_next(&cursor, bytes);
15951515 }
15961516
15971517 return ret;
@@ -1696,7 +1616,7 @@ static void amdgpu_ttm_training_data_block_init(struct amdgpu_device *adev) · 98 unmodified lines
16961616 (adev->gmc.mc_vram_size - GDDR6_MEM_TRAINING_OFFSET);
16971617 ctx->train_data_size =
16981618 GDDR6_MEM_TRAINING_DATA_SIZE_IN_BYTES;
1699−
1619+
17001620 DRM_DEBUG("train_data_size:%llx,p2c_train_data_offset:%llx,c2p_train_data_offset:%llx.\n",
17011621 ctx->train_data_size,
17021622 ctx->p2c_train_data_offset,
@@ -1812,8 +1732,15 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) · 109 unmodified lines
18121732 /* Change the size here instead of the init above so only lpfn is affected */
18131733 amdgpu_ttm_set_buffer_funcs_status(adev, false);
18141734 #ifdef CONFIG_64BIT
1815− adev->mman.aper_base_kaddr = ioremap_wc(adev->gmc.aper_base,
1816− adev->gmc.visible_vram_size);
1735+#ifdef CONFIG_X86
1736+ if (adev->gmc.xgmi.connected_to_cpu)
1737+ adev->mman.aper_base_kaddr = ioremap_cache(adev->gmc.aper_base,
1738+ adev->gmc.visible_vram_size);
1739+
1740+ else
1741+#endif
1742+ adev->mman.aper_base_kaddr = ioremap_wc(adev->gmc.aper_base,
1743+ adev->gmc.visible_vram_size);
18171744 #endif
18181745
18191746 /*
@@ -2053,7 +1980,8 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, · 233 unmodified lines
20531980 return r;
20541981
20551982 if (vm_needs_flush) {
2056− job->vm_pd_addr = amdgpu_gmc_pd_addr(adev->gart.bo);
1983+ job->vm_pd_addr = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo ?
1984+ adev->gmc.pdb0_bo : adev->gart.bo);
20571985 job->vm_needs_flush = true;
20581986 }
20591987 if (resv) {
@@ -2104,9 +2032,9 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo, · 44 unmodified lines
21042032 uint32_t max_bytes = adev->mman.buffer_funcs->fill_max_bytes;
21052033 struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
21062034
2107− struct drm_mm_node *mm_node;
2108− unsigned long num_pages;
2035+ struct amdgpu_res_cursor cursor;
21092036 unsigned int num_loops, num_dw;
2037+ uint64_t num_bytes;
21102038
21112039 struct amdgpu_job *job;
21122040 int r;
@@ -2122,15 +2050,13 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo, · 9 unmodified lines
21222050 return r;
21232051 }
21242052
2125− num_pages = bo->tbo.mem.num_pages;
2126− mm_node = bo->tbo.mem.mm_node;
2053+ num_bytes = bo->tbo.mem.num_pages << PAGE_SHIFT;
21272054 num_loops = 0;
2128− while (num_pages) {
2129− uint64_t byte_count = mm_node->size << PAGE_SHIFT;
21302055
2131− num_loops += DIV_ROUND_UP_ULL(byte_count, max_bytes);
2132− num_pages -= mm_node->size;
2133− ++mm_node;
2056+ amdgpu_res_first(&bo->tbo.mem, 0, num_bytes, &cursor);
2057+ while (cursor.remaining) {
2058+ num_loops += DIV_ROUND_UP_ULL(cursor.size, max_bytes);
2059+ amdgpu_res_next(&cursor, cursor.size);
21342060 }
21352061 num_dw = num_loops * adev->mman.buffer_funcs->fill_num_dw;
21362062
@@ -2152,27 +2078,16 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo, · 15 unmodified lines
21522078 }
21532079 }
21542080
2155− num_pages = bo->tbo.mem.num_pages;
2156− mm_node = bo->tbo.mem.mm_node;
2157−
2158− while (num_pages) {
2159− uint64_t byte_count = mm_node->size << PAGE_SHIFT;
2160− uint64_t dst_addr;
2161−
2162− dst_addr = amdgpu_mm_node_addr(&bo->tbo, mm_node, &bo->tbo.mem);
2163− while (byte_count) {
2164− uint32_t cur_size_in_bytes = min_t(uint64_t, byte_count,
2165− max_bytes);
2081+ amdgpu_res_first(&bo->tbo.mem, 0, num_bytes, &cursor);
2082+ while (cursor.remaining) {
2083+ uint32_t cur_size = min_t(uint64_t, cursor.size, max_bytes);
2084+ uint64_t dst_addr = cursor.start;
21662085
2167− amdgpu_emit_fill_buffer(adev, &job->ibs[0], src_data,
2168− dst_addr, cur_size_in_bytes);
2086+ dst_addr += amdgpu_ttm_domain_start(adev, bo->tbo.mem.mem_type);
2087+ amdgpu_emit_fill_buffer(adev, &job->ibs[0], src_data, dst_addr,
2088+ cur_size);
21692089
2170− dst_addr += cur_size_in_bytes;
2171− byte_count -= cur_size_in_bytes;
2172− }
2173−
2174− num_pages -= mm_node->size;
2175− ++mm_node;
2090+ amdgpu_res_next(&cursor, cur_size);
21762091 }
21772092
21782093 amdgpu_ring_pad_ib(ring, &job->ibs[0]);
@@ -2191,36 +2106,74 @@ error_free: · 12 unmodified lines
21912106
21922107 #if defined(CONFIG_DEBUG_FS)
21932108
2194−static int amdgpu_mm_dump_table(struct seq_file *m, void *data)
2109+static int amdgpu_mm_vram_table_show(struct seq_file *m, void *unused)
21952110 {
2196− struct drm_info_node *node = (struct drm_info_node *)m->private;
2197− unsigned ttm_pl = (uintptr_t)node->info_ent->data;
2198− struct drm_device *dev = node->minor->dev;
2199− struct amdgpu_device *adev = drm_to_adev(dev);
2200− struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, ttm_pl);
2111+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
2112+ struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev,
2113+ TTM_PL_VRAM);
22012114 struct drm_printer p = drm_seq_file_printer(m);
22022115
22032116 man->func->debug(man, &p);
22042117 return 0;
22052118 }
22062119
2207−static int amdgpu_ttm_pool_debugfs(struct seq_file *m, void *data)
2120+static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused)
22082121 {
2209− struct drm_info_node *node = (struct drm_info_node *)m->private;
2210− struct drm_device *dev = node->minor->dev;
2211− struct amdgpu_device *adev = drm_to_adev(dev);
2122+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
22122123
22132124 return ttm_pool_debugfs(&adev->mman.bdev.pool, m);
22142125 }
22152126
2216−static const struct drm_info_list amdgpu_ttm_debugfs_list[] = {
2217− {"amdgpu_vram_mm", amdgpu_mm_dump_table, 0, (void *)TTM_PL_VRAM},
2218− {"amdgpu_gtt_mm", amdgpu_mm_dump_table, 0, (void *)TTM_PL_TT},
2219− {"amdgpu_gds_mm", amdgpu_mm_dump_table, 0, (void *)AMDGPU_PL_GDS},
2220− {"amdgpu_gws_mm", amdgpu_mm_dump_table, 0, (void *)AMDGPU_PL_GWS},
2221− {"amdgpu_oa_mm", amdgpu_mm_dump_table, 0, (void *)AMDGPU_PL_OA},
2222− {"ttm_page_pool", amdgpu_ttm_pool_debugfs, 0, NULL},
2223−};
2127+static int amdgpu_mm_tt_table_show(struct seq_file *m, void *unused)
2128+{
2129+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
2130+ struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev,
2131+ TTM_PL_TT);
2132+ struct drm_printer p = drm_seq_file_printer(m);
2133+
2134+ man->func->debug(man, &p);
2135+ return 0;
2136+}
2137+
2138+static int amdgpu_mm_gds_table_show(struct seq_file *m, void *unused)
2139+{
2140+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
2141+ struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev,
2142+ AMDGPU_PL_GDS);
2143+ struct drm_printer p = drm_seq_file_printer(m);
2144+
2145+ man->func->debug(man, &p);
2146+ return 0;
2147+}
2148+
2149+static int amdgpu_mm_gws_table_show(struct seq_file *m, void *unused)
2150+{
2151+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
2152+ struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev,
2153+ AMDGPU_PL_GWS);
2154+ struct drm_printer p = drm_seq_file_printer(m);
2155+
2156+ man->func->debug(man, &p);
2157+ return 0;
2158+}
2159+
2160+static int amdgpu_mm_oa_table_show(struct seq_file *m, void *unused)
2161+{
2162+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
2163+ struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev,
2164+ AMDGPU_PL_OA);
2165+ struct drm_printer p = drm_seq_file_printer(m);
2166+
2167+ man->func->debug(man, &p);
2168+ return 0;
2169+}
2170+
2171+DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_vram_table);
2172+DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_tt_table);
2173+DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_gds_table);
2174+DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_gws_table);
2175+DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_oa_table);
2176+DEFINE_SHOW_ATTRIBUTE(amdgpu_ttm_page_pool);
22242177
22252178 /*
22262179 * amdgpu_ttm_vram_read - Linear read access to VRAM
@@ -2308,58 +2261,6 @@ static const struct file_operations amdgpu_ttm_vram_fops = { · 81 unmodified lines
23082261 .llseek = default_llseek,
23092262 };
23102263
2311−#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
2312−
2313−/*
2314− * amdgpu_ttm_gtt_read - Linear read access to GTT memory
2315− */
2316−static ssize_t amdgpu_ttm_gtt_read(struct file *f, char __user *buf,
2317− size_t size, loff_t *pos)
2318−{
2319− struct amdgpu_device *adev = file_inode(f)->i_private;
2320− ssize_t result = 0;
2321− int r;
2322−
2323− while (size) {
2324− loff_t p = *pos / PAGE_SIZE;
2325− unsigned off = *pos & ~PAGE_MASK;
2326− size_t cur_size = min_t(size_t, size, PAGE_SIZE - off);
2327− struct page *page;
2328− void *ptr;
2329−
2330− if (p >= adev->gart.num_cpu_pages)
2331− return result;
2332−
2333− page = adev->gart.pages[p];
2334− if (page) {
2335− ptr = kmap(page);
2336− ptr += off;
2337−
2338− r = copy_to_user(buf, ptr, cur_size);
2339− kunmap(adev->gart.pages[p]);
2340− } else
2341− r = clear_user(buf, cur_size);
2342−
2343− if (r)
2344− return -EFAULT;
2345−
2346− result += cur_size;
2347− buf += cur_size;
2348− *pos += cur_size;
2349− size -= cur_size;
2350− }
2351−
2352− return result;
2353−}
2354−
2355−static const struct file_operations amdgpu_ttm_gtt_fops = {
2356− .owner = THIS_MODULE,
2357− .read = amdgpu_ttm_gtt_read,
2358− .llseek = default_llseek
2359−};
2360−
2361−#endif
2362−
23632264 /*
23642265 * amdgpu_iomem_read - Virtual read access to GPU mapped memory
23652266 *
@@ -2474,46 +2375,29 @@ static const struct file_operations amdgpu_ttm_iomem_fops = { · 108 unmodified lines
24742375 .llseek = default_llseek
24752376 };
24762377
2477−static const struct {
2478− char *name;
2479− const struct file_operations *fops;
2480− int domain;
2481−} ttm_debugfs_entries[] = {
2482− { "amdgpu_vram", &amdgpu_ttm_vram_fops, TTM_PL_VRAM },
2483−#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
2484− { "amdgpu_gtt", &amdgpu_ttm_gtt_fops, TTM_PL_TT },
2485−#endif
2486− { "amdgpu_iomem", &amdgpu_ttm_iomem_fops, TTM_PL_SYSTEM },
2487−};
2488−
24892378 #endif
24902379
2491−int amdgpu_ttm_debugfs_init(struct amdgpu_device *adev)
2380+void amdgpu_ttm_debugfs_init(struct amdgpu_device *adev)
24922381 {
24932382 #if defined(CONFIG_DEBUG_FS)
2494− unsigned count;
2495−
24962383 struct drm_minor *minor = adev_to_drm(adev)->primary;
2497− struct dentry *ent, *root = minor->debugfs_root;
2498−
2499− for (count = 0; count < ARRAY_SIZE(ttm_debugfs_entries); count++) {
2500− ent = debugfs_create_file(
2501− ttm_debugfs_entries[count].name,
2502− S_IFREG | S_IRUGO, root,
2503− adev,
2504− ttm_debugfs_entries[count].fops);
2505− if (IS_ERR(ent))
2506− return PTR_ERR(ent);
2507− if (ttm_debugfs_entries[count].domain == TTM_PL_VRAM)
2508− i_size_write(ent->d_inode, adev->gmc.mc_vram_size);
2509− else if (ttm_debugfs_entries[count].domain == TTM_PL_TT)
2510− i_size_write(ent->d_inode, adev->gmc.gart_size);
2511− adev->mman.debugfs_entries[count] = ent;
2512− }
2513−
2514− count = ARRAY_SIZE(amdgpu_ttm_debugfs_list);
2515− return amdgpu_debugfs_add_files(adev, amdgpu_ttm_debugfs_list, count);
2516−#else
2517− return 0;
2384+ struct dentry *root = minor->debugfs_root;
2385+
2386+ debugfs_create_file_size("amdgpu_vram", 0444, root, adev,
2387+ &amdgpu_ttm_vram_fops, adev->gmc.mc_vram_size);
2388+ debugfs_create_file("amdgpu_iomem", 0444, root, adev,
2389+ &amdgpu_ttm_iomem_fops);
2390+ debugfs_create_file("amdgpu_vram_mm", 0444, root, adev,
2391+ &amdgpu_mm_vram_table_fops);
2392+ debugfs_create_file("amdgpu_gtt_mm", 0444, root, adev,
2393+ &amdgpu_mm_tt_table_fops);
2394+ debugfs_create_file("amdgpu_gds_mm", 0444, root, adev,
2395+ &amdgpu_mm_gds_table_fops);
2396+ debugfs_create_file("amdgpu_gws_mm", 0444, root, adev,
2397+ &amdgpu_mm_gws_table_fops);
2398+ debugfs_create_file("amdgpu_oa_mm", 0444, root, adev,
2399+ &amdgpu_mm_oa_table_fops);
2400+ debugfs_create_file("ttm_page_pool", 0444, root, adev,
2401+ &amdgpu_ttm_page_pool_fops);
25182402 #endif
25192403 }
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h+2−7
@@ -64,10 +64,6 @@ struct amdgpu_mman { · 63 unmodified lines
6464 bool initialized;
6565 void __iomem *aper_base_kaddr;
6666
67−#if defined(CONFIG_DEBUG_FS)
68− struct dentry *debugfs_entries[8];
69−#endif
70−
7167 /* buffer handling */
7268 const struct amdgpu_buffer_funcs *buffer_funcs;
7369 struct amdgpu_ring *buffer_funcs_ring;
@@ -119,8 +115,7 @@ int amdgpu_vram_mgr_alloc_sgt(struct amdgpu_device *adev, · 45 unmodified lines
119115 struct device *dev,
120116 enum dma_data_direction dir,
121117 struct sg_table **sgt);
122−void amdgpu_vram_mgr_free_sgt(struct amdgpu_device *adev,
123− struct device *dev,
118+void amdgpu_vram_mgr_free_sgt(struct device *dev,
124119 enum dma_data_direction dir,
125120 struct sg_table *sgt);
126121 uint64_t amdgpu_vram_mgr_usage(struct ttm_resource_manager *man);
@@ -186,6 +181,6 @@ uint64_t amdgpu_ttm_tt_pde_flags(struct ttm_tt *ttm, struct ttm_resource *mem); · 59 unmodified lines
186181 uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
187182 struct ttm_resource *mem);
188183
189−int amdgpu_ttm_debugfs_init(struct amdgpu_device *adev);
184+void amdgpu_ttm_debugfs_init(struct amdgpu_device *adev);
190185
191186 #endif
drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c+1−0
@@ -402,6 +402,7 @@ amdgpu_ucode_get_load_type(struct amdgpu_device *adev, int load_type) · 401 unmodified lines
402402 case CHIP_NAVY_FLOUNDER:
403403 case CHIP_VANGOGH:
404404 case CHIP_DIMGREY_CAVEFISH:
405+ case CHIP_ALDEBARAN:
405406 if (!load_type)
406407 return AMDGPU_FW_LOAD_DIRECT;
407408 else
drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.h+4−0
@@ -105,6 +105,9 @@ struct psp_firmware_header_v1_3 { · 104 unmodified lines
105105 uint32_t spl_header_version;
106106 uint32_t spl_offset_bytes;
107107 uint32_t spl_size_bytes;
108+ uint32_t rl_header_version;
109+ uint32_t rl_offset_bytes;
110+ uint32_t rl_size_bytes;
108111 };
109112
110113 /* version_major=1, version_minor=0 */
@@ -136,6 +139,7 @@ enum ta_fw_type { · 25 unmodified lines
136139 TA_FW_TYPE_PSP_DTM,
137140 TA_FW_TYPE_PSP_RAP,
138141 TA_FW_TYPE_PSP_SECUREDISPLAY,
142+ TA_FW_TYPE_MAX_INDEX,
139143 };
140144
141145 struct ta_fw_bin_desc {
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c+8−0
@@ -47,6 +47,7 @@ · 46 unmodified lines
4747 #define FIRMWARE_NAVY_FLOUNDER "amdgpu/navy_flounder_vcn.bin"
4848 #define FIRMWARE_VANGOGH "amdgpu/vangogh_vcn.bin"
4949 #define FIRMWARE_DIMGREY_CAVEFISH "amdgpu/dimgrey_cavefish_vcn.bin"
50+#define FIRMWARE_ALDEBARAN "amdgpu/aldebaran_vcn.bin"
5051
5152 MODULE_FIRMWARE(FIRMWARE_RAVEN);
5253 MODULE_FIRMWARE(FIRMWARE_PICASSO);
@@ -54,6 +55,7 @@ MODULE_FIRMWARE(FIRMWARE_RAVEN2); · 1 unmodified line
5455 MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
5556 MODULE_FIRMWARE(FIRMWARE_RENOIR);
5657 MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
58+MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
5759 MODULE_FIRMWARE(FIRMWARE_NAVI10);
5860 MODULE_FIRMWARE(FIRMWARE_NAVI14);
5961 MODULE_FIRMWARE(FIRMWARE_NAVI12);
@@ -100,6 +102,12 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev) · 40 unmodified lines
100102 else
101103 fw_name = FIRMWARE_GREEN_SARDINE;
102104
105+ if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
106+ (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
107+ adev->vcn.indirect_sram = true;
108+ break;
109+ case CHIP_ALDEBARAN:
110+ fw_name = FIRMWARE_ALDEBARAN;
103111 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
104112 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
105113 adev->vcn.indirect_sram = true;
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h+11−2
@@ -155,6 +155,7 @@ · 154 unmodified lines
155155 } \
156156 } while (0)
157157
158+#define AMDGPU_VCN_FW_SHARED_FLAG_0_RB (1 << 6)
158159 #define AMDGPU_VCN_MULTI_QUEUE_FLAG (1 << 8)
159160 #define AMDGPU_VCN_SW_RING_FLAG (1 << 9)
160161
@@ -243,6 +244,12 @@ struct amdgpu_vcn { · 82 unmodified lines
243244 int inst_idx, struct dpg_pause_state *new_state);
244245 };
245246
247+struct amdgpu_fw_shared_rb_ptrs_struct {
248+ /* to WA DPG R/W ptr issues.*/
249+ uint32_t rptr;
250+ uint32_t wptr;
251+};
252+
246253 struct amdgpu_fw_shared_multi_queue {
247254 uint8_t decode_queue_mode;
248255 uint8_t encode_generalpurpose_queue_mode;
@@ -258,10 +265,12 @@ struct amdgpu_fw_shared_sw_ring { · 9 unmodified lines
258265
259266 struct amdgpu_fw_shared {
260267 uint32_t present_flag_0;
261− uint8_t pad[53];
268+ uint8_t pad[44];
269+ struct amdgpu_fw_shared_rb_ptrs_struct rb;
270+ uint8_t pad1[1];
262271 struct amdgpu_fw_shared_multi_queue multi_queue;
263272 struct amdgpu_fw_shared_sw_ring sw_ring;
264−} __attribute__((__packed__));
273+};
265274
266275 struct amdgpu_vcn_decode_buffer {
267276 uint32_t valid_buf_flag;
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c+1−0
@@ -640,6 +640,7 @@ void amdgpu_detect_virtualization(struct amdgpu_device *adev) · 639 unmodified lines
640640 case CHIP_NAVI12:
641641 case CHIP_SIENNA_CICHLID:
642642 case CHIP_ARCTURUS:
643+ case CHIP_ALDEBARAN:
643644 reg = RREG32(mmRCC_IOV_FUNC_IDENTIFIER);
644645 break;
645646 default: /* other chip doesn't support SRIOV */
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c+3−3
@@ -92,13 +92,13 @@ struct amdgpu_prt_cb { · 91 unmodified lines
9292 static inline void amdgpu_vm_eviction_lock(struct amdgpu_vm *vm)
9393 {
9494 mutex_lock(&vm->eviction_lock);
95− vm->saved_flags = memalloc_nofs_save();
95+ vm->saved_flags = memalloc_noreclaim_save();
9696 }
9797
9898 static inline int amdgpu_vm_eviction_trylock(struct amdgpu_vm *vm)
9999 {
100100 if (mutex_trylock(&vm->eviction_lock)) {
101− vm->saved_flags = memalloc_nofs_save();
101+ vm->saved_flags = memalloc_noreclaim_save();
102102 return 1;
103103 }
104104 return 0;
@@ -106,7 +106,7 @@ static inline int amdgpu_vm_eviction_trylock(struct amdgpu_vm *vm) · 1 unmodified line
106106
107107 static inline void amdgpu_vm_eviction_unlock(struct amdgpu_vm *vm)
108108 {
109− memalloc_nofs_restore(vm->saved_flags);
109+ memalloc_noreclaim_restore(vm->saved_flags);
110110 mutex_unlock(&vm->eviction_lock);
111111 }
112112
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c+1−3
@@ -639,15 +639,13 @@ error_free: · 638 unmodified lines
639639 /**
640640 * amdgpu_vram_mgr_free_sgt - allocate and fill a sg table
641641 *
642− * @adev: amdgpu device pointer
643642 * @dev: device pointer
644643 * @dir: data direction of resource to unmap
645644 * @sgt: sg table to free
646645 *
647646 * Free a previously allocate sg table.
648647 */
649−void amdgpu_vram_mgr_free_sgt(struct amdgpu_device *adev,
650− struct device *dev,
648+void amdgpu_vram_mgr_free_sgt(struct device *dev,
651649 enum dma_data_direction dir,
652650 struct sg_table *sgt)
653651 {
drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c+13−4
@@ -468,15 +468,22 @@ int amdgpu_xgmi_update_topology(struct amdgpu_hive_info *hive, struct amdgpu_dev · 467 unmodified lines
468468 }
469469
470470
471+/*
472+ * NOTE psp_xgmi_node_info.num_hops layout is as follows:
473+ * num_hops[7:6] = link type (0 = xGMI2, 1 = xGMI3, 2/3 = reserved)
474+ * num_hops[5:3] = reserved
475+ * num_hops[2:0] = number of hops
476+ */
471477 int amdgpu_xgmi_get_hops_count(struct amdgpu_device *adev,
472478 struct amdgpu_device *peer_adev)
473479 {
474480 struct psp_xgmi_topology_info *top = &adev->psp.xgmi_context.top_info;
481+ uint8_t num_hops_mask = 0x7;
475482 int i;
476483
477484 for (i = 0 ; i < top->num_nodes; ++i)
478485 if (top->nodes[i].node_id == peer_adev->gmc.xgmi.node_id)
479− return top->nodes[i].num_hops;
486+ return top->nodes[i].num_hops & num_hops_mask;
480487 return -EINVAL;
481488 }
482489
@@ -492,7 +499,8 @@ int amdgpu_xgmi_add_device(struct amdgpu_device *adev) · 9 unmodified lines
492499 if (!adev->gmc.xgmi.supported)
493500 return 0;
494501
495− if (amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_PSP)) {
502+ if (!adev->gmc.xgmi.pending_reset &&
503+ amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_PSP)) {
496504 ret = psp_xgmi_initialize(&adev->psp);
497505 if (ret) {
498506 dev_err(adev->dev,
@@ -538,7 +546,8 @@ int amdgpu_xgmi_add_device(struct amdgpu_device *adev) · 39 unmodified lines
538546
539547 task_barrier_add_task(&hive->tb);
540548
541− if (amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_PSP)) {
549+ if (!adev->gmc.xgmi.pending_reset &&
550+ amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_PSP)) {
542551 list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
543552 /* update node list for other device in the hive */
544553 if (tmp_adev != adev) {
@@ -567,7 +576,7 @@ int amdgpu_xgmi_add_device(struct amdgpu_device *adev) · 22 unmodified lines
567576 }
568577 }
569578
570− if (!ret)
579+ if (!ret && !adev->gmc.xgmi.pending_reset)
571580 ret = amdgpu_xgmi_sysfs_add_dev_info(adev, hive);
572581
573582 exit_unlock:
drivers/gpu/drm/amd/amdgpu/athub_v2_1.c+2−4
@@ -74,10 +74,8 @@ int athub_v2_1_set_clockgating(struct amdgpu_device *adev, · 73 unmodified lines
7474 case CHIP_SIENNA_CICHLID:
7575 case CHIP_NAVY_FLOUNDER:
7676 case CHIP_DIMGREY_CAVEFISH:
77− athub_v2_1_update_medium_grain_clock_gating(adev,
78− state == AMD_CG_STATE_GATE ? true : false);
79− athub_v2_1_update_medium_grain_light_sleep(adev,
80− state == AMD_CG_STATE_GATE ? true : false);
77+ athub_v2_1_update_medium_grain_clock_gating(adev, state == AMD_CG_STATE_GATE);
78+ athub_v2_1_update_medium_grain_light_sleep(adev, state == AMD_CG_STATE_GATE);
8179 break;
8280 default:
8381 break;
drivers/gpu/drm/amd/amdgpu/atom.c+2−2
@@ -114,11 +114,11 @@ static uint32_t atom_iio_execute(struct atom_context *ctx, int base, · 113 unmodified lines
114114 base++;
115115 break;
116116 case ATOM_IIO_READ:
117− temp = ctx->card->ioreg_read(ctx->card, CU16(base + 1));
117+ temp = ctx->card->reg_read(ctx->card, CU16(base + 1));
118118 base += 3;
119119 break;
120120 case ATOM_IIO_WRITE:
121− ctx->card->ioreg_write(ctx->card, CU16(base + 1), temp);
121+ ctx->card->reg_write(ctx->card, CU16(base + 1), temp);
122122 base += 3;
123123 break;
124124 case ATOM_IIO_CLEAR:
drivers/gpu/drm/amd/amdgpu/atom.h+0−2
@@ -116,8 +116,6 @@ struct card_info { · 115 unmodified lines
116116 struct drm_device *dev;
117117 void (* reg_write)(struct card_info *, uint32_t, uint32_t); /* filled by driver */
118118 uint32_t (* reg_read)(struct card_info *, uint32_t); /* filled by driver */
119− void (* ioreg_write)(struct card_info *, uint32_t, uint32_t); /* filled by driver */
120− uint32_t (* ioreg_read)(struct card_info *, uint32_t); /* filled by driver */
121119 void (* mc_write)(struct card_info *, uint32_t, uint32_t); /* filled by driver */
122120 uint32_t (* mc_read)(struct card_info *, uint32_t); /* filled by driver */
123121 void (* pll_write)(struct card_info *, uint32_t, uint32_t); /* filled by driver */
drivers/gpu/drm/amd/amdgpu/cik.c+77−0
@@ -26,6 +26,8 @@ · 25 unmodified lines
2626 #include <linux/module.h>
2727 #include <linux/pci.h>
2828
29+#include <drm/amdgpu_drm.h>
30+
2931 #include "amdgpu.h"
3032 #include "amdgpu_atombios.h"
3133 #include "amdgpu_ih.h"
@@ -70,6 +72,80 @@ · 38 unmodified lines
7072 #include "amdgpu_amdkfd.h"
7173 #include "dce_virtual.h"
7274
75+static const struct amdgpu_video_codec_info cik_video_codecs_encode_array[] =
76+{
77+ {
78+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
79+ .max_width = 2048,
80+ .max_height = 1152,
81+ .max_pixels_per_frame = 2048 * 1152,
82+ .max_level = 0,
83+ },
84+};
85+
86+static const struct amdgpu_video_codecs cik_video_codecs_encode =
87+{
88+ .codec_count = ARRAY_SIZE(cik_video_codecs_encode_array),
89+ .codec_array = cik_video_codecs_encode_array,
90+};
91+
92+static const struct amdgpu_video_codec_info cik_video_codecs_decode_array[] =
93+{
94+ {
95+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
96+ .max_width = 2048,
97+ .max_height = 1152,
98+ .max_pixels_per_frame = 2048 * 1152,
99+ .max_level = 3,
100+ },
101+ {
102+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
103+ .max_width = 2048,
104+ .max_height = 1152,
105+ .max_pixels_per_frame = 2048 * 1152,
106+ .max_level = 5,
107+ },
108+ {
109+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
110+ .max_width = 2048,
111+ .max_height = 1152,
112+ .max_pixels_per_frame = 2048 * 1152,
113+ .max_level = 41,
114+ },
115+ {
116+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
117+ .max_width = 2048,
118+ .max_height = 1152,
119+ .max_pixels_per_frame = 2048 * 1152,
120+ .max_level = 4,
121+ },
122+};
123+
124+static const struct amdgpu_video_codecs cik_video_codecs_decode =
125+{
126+ .codec_count = ARRAY_SIZE(cik_video_codecs_decode_array),
127+ .codec_array = cik_video_codecs_decode_array,
128+};
129+
130+static int cik_query_video_codecs(struct amdgpu_device *adev, bool encode,
131+ const struct amdgpu_video_codecs **codecs)
132+{
133+ switch (adev->asic_type) {
134+ case CHIP_BONAIRE:
135+ case CHIP_HAWAII:
136+ case CHIP_KAVERI:
137+ case CHIP_KABINI:
138+ case CHIP_MULLINS:
139+ if (encode)
140+ *codecs = &cik_video_codecs_encode;
141+ else
142+ *codecs = &cik_video_codecs_decode;
143+ return 0;
144+ default:
145+ return -EINVAL;
146+ }
147+}
148+
73149 /*
74150 * Indirect registers accessor
75151 */
@@ -1933,6 +2009,7 @@ static const struct amdgpu_asic_funcs cik_asic_funcs = · 1857 unmodified lines
19332009 .get_pcie_replay_count = &cik_get_pcie_replay_count,
19342010 .supports_baco = &cik_asic_supports_baco,
19352011 .pre_asic_init = &cik_pre_asic_init,
2012+ .query_video_codecs = &cik_query_video_codecs,
19362013 };
19372014
19382015 static int cik_common_early_init(void *handle)
drivers/gpu/drm/amd/amdgpu/df_v3_6.c+2−0
@@ -568,6 +568,8 @@ static int df_v3_6_pmc_stop(struct amdgpu_device *adev, uint64_t config, · 567 unmodified lines
568568 if (ret)
569569 return ret;
570570
571+ df_v3_6_perfmon_wreg(adev, lo_base_addr, lo_val,
572+ hi_base_addr, hi_val);
571573
572574 if (is_remove) {
573575 df_v3_6_reset_perfmon_cntr(adev, config, counter_idx);
drivers/gpu/drm/amd/amdgpu/dimgrey_cavefish_reg_init.c+0−0

File mode changed from 100755 to 100644.

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c+5−4
@@ -6718,7 +6718,8 @@ static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev, · 6717 unmodified lines
67186718 return 0;
67196719 }
67206720
6721−static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data)
6721+static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data,
6722+ bool from_wq)
67226723 {
67236724 u32 enc, se_id, sh_id, cu_id;
67246725 char type[20];
@@ -6756,7 +6757,7 @@ static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data) · 31 unmodified lines
67566757 * or from BH in which case we can access SQ_EDC_INFO
67576758 * instance
67586759 */
6759− if (in_task()) {
6760+ if (from_wq) {
67606761 mutex_lock(&adev->grbm_idx_mutex);
67616762 gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id);
67626763
@@ -6794,7 +6795,7 @@ static void gfx_v8_0_sq_irq_work_func(struct work_struct *work) · 31 unmodified lines
67946795 struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work);
67956796 struct sq_work *sq_work = container_of(work, struct sq_work, work);
67966797
6797− gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data);
6798+ gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data, true);
67986799 }
67996800
68006801 static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
@@ -6809,7 +6810,7 @@ static int gfx_v8_0_sq_irq(struct amdgpu_device *adev, · 8 unmodified lines
68096810 * just print whatever info is possible directly from the ISR.
68106811 */
68116812 if (work_pending(&adev->gfx.sq_work.work)) {
6812− gfx_v8_0_parse_sq_irq(adev, ih_data);
6813+ gfx_v8_0_parse_sq_irq(adev, ih_data, false);
68136814 } else {
68146815 adev->gfx.sq_work.ih_data = ih_data;
68156816 schedule_work(&adev->gfx.sq_work.work);
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c+123−38
@@ -49,6 +49,7 @@ · 48 unmodified lines
4949
5050 #include "gfx_v9_4.h"
5151 #include "gfx_v9_0.h"
52+#include "gfx_v9_4_2.h"
5253
5354 #include "asic_reg/pwr/pwr_10_0_offset.h"
5455 #include "asic_reg/pwr/pwr_10_0_sh_mask.h"
@@ -107,14 +108,12 @@ MODULE_FIRMWARE("amdgpu/raven2_rlc.bin"); · 52 unmodified lines
107108 MODULE_FIRMWARE("amdgpu/raven_kicker_rlc.bin");
108109
109110 MODULE_FIRMWARE("amdgpu/arcturus_mec.bin");
110−MODULE_FIRMWARE("amdgpu/arcturus_mec2.bin");
111111 MODULE_FIRMWARE("amdgpu/arcturus_rlc.bin");
112112
113113 MODULE_FIRMWARE("amdgpu/renoir_ce.bin");
114114 MODULE_FIRMWARE("amdgpu/renoir_pfp.bin");
115115 MODULE_FIRMWARE("amdgpu/renoir_me.bin");
116116 MODULE_FIRMWARE("amdgpu/renoir_mec.bin");
117−MODULE_FIRMWARE("amdgpu/renoir_mec2.bin");
118117 MODULE_FIRMWARE("amdgpu/renoir_rlc.bin");
119118
120119 MODULE_FIRMWARE("amdgpu/green_sardine_ce.bin");
@@ -124,6 +123,10 @@ MODULE_FIRMWARE("amdgpu/green_sardine_mec.bin"); · 3 unmodified lines
124123 MODULE_FIRMWARE("amdgpu/green_sardine_mec2.bin");
125124 MODULE_FIRMWARE("amdgpu/green_sardine_rlc.bin");
126125
126+MODULE_FIRMWARE("amdgpu/aldebaran_mec.bin");
127+MODULE_FIRMWARE("amdgpu/aldebaran_mec2.bin");
128+MODULE_FIRMWARE("amdgpu/aldebaran_rlc.bin");
129+
127130 #define mmTCP_CHAN_STEER_0_ARCT 0x0b03
128131 #define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX 0
129132 #define mmTCP_CHAN_STEER_1_ARCT 0x0b04
@@ -981,11 +984,16 @@ static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev) · 851 unmodified lines
981984 golden_settings_gc_9_1_rn,
982985 ARRAY_SIZE(golden_settings_gc_9_1_rn));
983986 return; /* for renoir, don't need common goldensetting */
987+ case CHIP_ALDEBARAN:
988+ gfx_v9_4_2_init_golden_registers(adev,
989+ adev->smuio.funcs->get_die_id(adev));
990+ break;
984991 default:
985992 break;
986993 }
987994
988− if (adev->asic_type != CHIP_ARCTURUS)
995+ if ((adev->asic_type != CHIP_ARCTURUS) &&
996+ (adev->asic_type != CHIP_ALDEBARAN))
989997 soc15_program_register_sequence(adev, golden_settings_gc_9_x_common,
990998 (const u32)ARRAY_SIZE(golden_settings_gc_9_x_common));
991999 }
@@ -1517,6 +1525,16 @@ out: · 525 unmodified lines
15171525 return err;
15181526 }
15191527
1528+static bool gfx_v9_0_load_mec2_fw_bin_support(struct amdgpu_device *adev)
1529+{
1530+ if (adev->asic_type == CHIP_ALDEBARAN ||
1531+ adev->asic_type == CHIP_ARCTURUS ||
1532+ adev->asic_type == CHIP_RENOIR)
1533+ return false;
1534+
1535+ return true;
1536+}
1537+
15201538 static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev,
15211539 const char *chip_name)
15221540 {
@@ -1538,21 +1556,23 @@ static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev, · 15 unmodified lines
15381556 adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
15391557
15401558
1541− snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1542− err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1543− if (!err) {
1544− err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1545− if (err)
1546− goto out;
1547− cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1548− adev->gfx.mec2_fw->data;
1549− adev->gfx.mec2_fw_version =
1550− le32_to_cpu(cp_hdr->header.ucode_version);
1551− adev->gfx.mec2_feature_version =
1552− le32_to_cpu(cp_hdr->ucode_feature_version);
1553− } else {
1554− err = 0;
1555− adev->gfx.mec2_fw = NULL;
1559+ if (gfx_v9_0_load_mec2_fw_bin_support(adev)) {
1560+ snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1561+ err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1562+ if (!err) {
1563+ err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1564+ if (err)
1565+ goto out;
1566+ cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1567+ adev->gfx.mec2_fw->data;
1568+ adev->gfx.mec2_fw_version =
1569+ le32_to_cpu(cp_hdr->header.ucode_version);
1570+ adev->gfx.mec2_feature_version =
1571+ le32_to_cpu(cp_hdr->ucode_feature_version);
1572+ } else {
1573+ err = 0;
1574+ adev->gfx.mec2_fw = NULL;
1575+ }
15561576 }
15571577
15581578 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
@@ -1581,8 +1601,7 @@ static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev, · 22 unmodified lines
15811601
15821602 /* TODO: Determine if MEC2 JT FW loading can be removed
15831603 for all GFX V9 asic and above */
1584− if (adev->asic_type != CHIP_ARCTURUS &&
1585− adev->asic_type != CHIP_RENOIR) {
1604+ if (gfx_v9_0_load_mec2_fw_bin_support(adev)) {
15861605 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2_JT];
15871606 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2_JT;
15881607 info->fw = adev->gfx.mec2_fw;
@@ -1642,6 +1661,9 @@ static int gfx_v9_0_init_microcode(struct amdgpu_device *adev) · 53 unmodified lines
16421661 else
16431662 chip_name = "green_sardine";
16441663 break;
1664+ case CHIP_ALDEBARAN:
1665+ chip_name = "aldebaran";
1666+ break;
16451667 default:
16461668 BUG();
16471669 }
@@ -1882,7 +1904,10 @@ static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable) · 234 unmodified lines
18821904
18831905 static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev)
18841906 {
1885− return 5;
1907+ if (gfx_v9_0_load_mec2_fw_bin_support(adev))
1908+ return 5;
1909+ else
1910+ return 4;
18861911 }
18871912
18881913 static int gfx_v9_0_rlc_init(struct amdgpu_device *adev)
@@ -2088,6 +2113,21 @@ static const struct amdgpu_gfx_funcs gfx_v9_4_gfx_funcs = { · 199 unmodified lines
20882113 .query_ras_error_status = &gfx_v9_4_query_ras_error_status,
20892114 };
20902115
2116+static const struct amdgpu_gfx_funcs gfx_v9_4_2_gfx_funcs = {
2117+ .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
2118+ .select_se_sh = &gfx_v9_0_select_se_sh,
2119+ .read_wave_data = &gfx_v9_0_read_wave_data,
2120+ .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
2121+ .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
2122+ .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
2123+ .ras_error_inject = &gfx_v9_4_2_ras_error_inject,
2124+ .query_ras_error_count = &gfx_v9_4_2_query_ras_error_count,
2125+ .reset_ras_error_count = &gfx_v9_4_2_reset_ras_error_count,
2126+ .query_ras_error_status = &gfx_v9_4_2_query_ras_error_status,
2127+ .reset_ras_error_status = &gfx_v9_4_2_reset_ras_error_status,
2128+ .enable_watchdog_timer = &gfx_v9_4_2_enable_watchdog_timer,
2129+};
2130+
20912131 static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
20922132 {
20932133 u32 gb_addr_config;
@@ -2159,6 +2199,21 @@ static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev) · 65 unmodified lines
21592199 gb_addr_config &= ~0xf3e777ff;
21602200 gb_addr_config |= 0x22010042;
21612201 break;
2202+ case CHIP_ALDEBARAN:
2203+ adev->gfx.funcs = &gfx_v9_4_2_gfx_funcs;
2204+ adev->gfx.config.max_hw_contexts = 8;
2205+ adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2206+ adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2207+ adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2208+ adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2209+ gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2210+ gb_addr_config &= ~0xf3e777ff;
2211+ gb_addr_config |= 0x22014042;
2212+ /* check vbios table if gpu info is not available */
2213+ err = amdgpu_atomfirmware_get_gfx_info(adev);
2214+ if (err)
2215+ return err;
2216+ break;
21622217 default:
21632218 BUG();
21642219 break;
@@ -2249,6 +2304,7 @@ static int gfx_v9_0_sw_init(void *handle) · 84 unmodified lines
22492304 case CHIP_RAVEN:
22502305 case CHIP_ARCTURUS:
22512306 case CHIP_RENOIR:
2307+ case CHIP_ALDEBARAN:
22522308 adev->gfx.mec.num_mec = 2;
22532309 break;
22542310 default:
@@ -2634,17 +2690,15 @@ static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev, · 379 unmodified lines
26342690 {
26352691 u32 tmp;
26362692
2637− /* don't toggle interrupts that are only applicable
2638− * to me0 pipe0 on AISCs that have me0 removed */
2639− if (!adev->gfx.num_gfx_rings)
2640− return;
2693+ /* These interrupts should be enabled to drive DS clock */
26412694
26422695 tmp= RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0);
26432696
26442697 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
26452698 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
26462699 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
2647− tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
2700+ if(adev->gfx.num_gfx_rings)
2701+ tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
26482702
26492703 WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
26502704 }
@@ -3696,11 +3750,18 @@ static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring) · 1045 unmodified lines
36963750 struct amdgpu_device *adev = ring->adev;
36973751 struct v9_mqd *mqd = ring->mqd_ptr;
36983752 int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
3753+ struct v9_mqd *tmp_mqd;
36993754
37003755 gfx_v9_0_kiq_setting(ring);
37013756
3702− if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
3703− /* reset MQD to a clean status */
3757+ /* GPU could be in bad state during probe, driver trigger the reset
3758+ * after load the SMU, in this case , the mqd is not be initialized.
3759+ * driver need to re-init the mqd.
3760+ * check mqd->cp_hqd_pq_control since this value should not be 0
3761+ */
3762+ tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3763+ if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control){
3764+ /* for GPU_RESET case , reset MQD to a clean status */
37043765 if (adev->gfx.mec.mqd_backup[mqd_idx])
37053766 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
37063767
@@ -3736,8 +3797,15 @@ static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring) · 29 unmodified lines
37363797 struct amdgpu_device *adev = ring->adev;
37373798 struct v9_mqd *mqd = ring->mqd_ptr;
37383799 int mqd_idx = ring - &adev->gfx.compute_ring[0];
3800+ struct v9_mqd *tmp_mqd;
37393801
3740− if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
3802+ /* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control
3803+ * is not be initialized before
3804+ */
3805+ tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3806+
3807+ if (!tmp_mqd->cp_hqd_pq_control ||
3808+ (!amdgpu_in_reset(adev) && !adev->in_suspend)) {
37413809 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
37423810 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
37433811 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
@@ -3913,6 +3981,9 @@ static int gfx_v9_0_hw_init(void *handle) · 169 unmodified lines
39133981 if (r)
39143982 return r;
39153983
3984+ if (adev->asic_type == CHIP_ALDEBARAN)
3985+ gfx_v9_4_2_set_power_brake_sequence(adev);
3986+
39163987 return r;
39173988 }
39183989
@@ -4101,7 +4172,7 @@ static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev) · 182 unmodified lines
41014172 *
41024173 * also don't wait anymore for IRQ context
41034174 * */
4104− if (r < 1 && (amdgpu_in_reset(adev) || in_interrupt()))
4175+ if (r < 1 && (amdgpu_in_reset(adev)))
41054176 goto failed_kiq_read;
41064177
41074178 might_sleep();
@@ -4486,7 +4557,8 @@ static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev) · 378 unmodified lines
44864557 if (!ring->sched.ready)
44874558 return 0;
44884559
4489− if (adev->asic_type == CHIP_ARCTURUS) {
4560+ if (adev->asic_type == CHIP_ARCTURUS ||
4561+ adev->asic_type == CHIP_ALDEBARAN) {
44904562 vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus;
44914563 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus);
44924564 vgpr_init_regs_ptr = vgpr_init_regs_arcturus;
@@ -4636,7 +4708,8 @@ static int gfx_v9_0_early_init(void *handle) · 143 unmodified lines
46364708 {
46374709 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
46384710
4639− if (adev->asic_type == CHIP_ARCTURUS)
4711+ if (adev->asic_type == CHIP_ARCTURUS ||
4712+ adev->asic_type == CHIP_ALDEBARAN)
46404713 adev->gfx.num_gfx_rings = 0;
46414714 else
46424715 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
@@ -4662,7 +4735,8 @@ static int gfx_v9_0_ecc_late_init(void *handle) · 19 unmodified lines
46624735 * to GDS in suspend/resume sequence on several cards. So just
46634736 * limit this operation in cold boot sequence.
46644737 */
4665− if (!adev->in_suspend) {
4738+ if ((!adev->in_suspend) &&
4739+ (adev->gds.gds_size)) {
46664740 r = gfx_v9_0_do_edc_gds_workarounds(adev);
46674741 if (r)
46684742 return r;
@@ -4673,14 +4747,13 @@ static int gfx_v9_0_ecc_late_init(void *handle) · 4 unmodified lines
46734747 if (r)
46744748 return r;
46754749
4676− if (adev->gfx.funcs &&
4677− adev->gfx.funcs->reset_ras_error_count)
4678− adev->gfx.funcs->reset_ras_error_count(adev);
4679−
46804750 r = amdgpu_gfx_ras_late_init(adev);
46814751 if (r)
46824752 return r;
46834753
4754+ if (adev->gfx.funcs->enable_watchdog_timer)
4755+ adev->gfx.funcs->enable_watchdog_timer(adev);
4756+
46844757 return 0;
46854758 }
46864759
@@ -4858,7 +4931,7 @@ static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev, · 171 unmodified lines
48584931 {
48594932 uint32_t data, def;
48604933
4861− if (adev->asic_type == CHIP_ARCTURUS)
4934+ if (!adev->gfx.num_gfx_rings)
48624935 return;
48634936
48644937 amdgpu_gfx_rlc_enter_safe_mode(adev);
@@ -5105,6 +5178,7 @@ static int gfx_v9_0_set_clockgating_state(void *handle, · 240 unmodified lines
51055178 case CHIP_RAVEN:
51065179 case CHIP_ARCTURUS:
51075180 case CHIP_RENOIR:
5181+ case CHIP_ALDEBARAN:
51085182 gfx_v9_0_update_gfx_clock_gating(adev,
51095183 state == AMD_CG_STATE_GATE);
51105184 break;
@@ -6924,6 +6998,7 @@ static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev) · 1813 unmodified lines
69246998 case CHIP_RAVEN:
69256999 case CHIP_ARCTURUS:
69267000 case CHIP_RENOIR:
7001+ case CHIP_ALDEBARAN:
69277002 adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
69287003 break;
69297004 default:
@@ -6944,6 +7019,12 @@ static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev) · 14 unmodified lines
69447019 case CHIP_ARCTURUS:
69457020 adev->gds.gds_size = 0x1000;
69467021 break;
7022+ case CHIP_ALDEBARAN:
7023+ /* aldebaran removed all the GDS internal memory,
7024+ * only support GWS opcode in kernel, like barrier
7025+ * semaphore.etc */
7026+ adev->gds.gds_size = 0;
7027+ break;
69477028 default:
69487029 adev->gds.gds_size = 0x10000;
69497030 break;
@@ -6966,6 +7047,10 @@ static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev) · 16 unmodified lines
69667047 case CHIP_ARCTURUS:
69677048 adev->gds.gds_compute_max_wave_id = 0xfff;
69687049 break;
7050+ case CHIP_ALDEBARAN:
7051+ /* deprecated for Aldebaran, no usage at all */
7052+ adev->gds.gds_compute_max_wave_id = 0;
7053+ break;
69697054 default:
69707055 /* this really depends on the chip */
69717056 adev->gds.gds_compute_max_wave_id = 0x7ff;
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.c+1286−0
@@ -0,0 +1,1286 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#include "amdgpu.h"
24+#include "soc15.h"
25+
26+#include "gc/gc_9_4_2_offset.h"
27+#include "gc/gc_9_4_2_sh_mask.h"
28+#include "gfx_v9_0.h"
29+
30+#include "gfx_v9_4_2.h"
31+#include "amdgpu_ras.h"
32+#include "amdgpu_gfx.h"
33+
34+enum gfx_v9_4_2_utc_type {
35+ VML2_MEM,
36+ VML2_WALKER_MEM,
37+ UTCL2_MEM,
38+ ATC_L2_CACHE_2M,
39+ ATC_L2_CACHE_32K,
40+ ATC_L2_CACHE_4K
41+};
42+
43+struct gfx_v9_4_2_utc_block {
44+ enum gfx_v9_4_2_utc_type type;
45+ uint32_t num_banks;
46+ uint32_t num_ways;
47+ uint32_t num_mem_blocks;
48+ struct soc15_reg idx_reg;
49+ struct soc15_reg data_reg;
50+ uint32_t sec_count_mask;
51+ uint32_t sec_count_shift;
52+ uint32_t ded_count_mask;
53+ uint32_t ded_count_shift;
54+ uint32_t clear;
55+};
56+
57+static const struct soc15_reg_golden golden_settings_gc_9_4_2_alde_die_0[] = {
58+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_0, 0x3fffffff, 0x141dc920),
59+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_1, 0x3fffffff, 0x3b458b93),
60+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_2, 0x3fffffff, 0x1a4f5583),
61+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_3, 0x3fffffff, 0x317717f6),
62+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_4, 0x3fffffff, 0x107cc1e6),
63+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_5, 0x3ff, 0x351),
64+};
65+
66+static const struct soc15_reg_golden golden_settings_gc_9_4_2_alde_die_1[] = {
67+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_0, 0x3fffffff, 0x2591aa38),
68+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_1, 0x3fffffff, 0xac9e88b),
69+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_2, 0x3fffffff, 0x2bc3369b),
70+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_3, 0x3fffffff, 0xfb74ee),
71+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_4, 0x3fffffff, 0x21f0a2fe),
72+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CHAN_STEER_5, 0x3ff, 0x49),
73+};
74+
75+static const struct soc15_reg_golden golden_settings_gc_9_4_2_alde[] = {
76+ SOC15_REG_GOLDEN_VALUE(GC, 0, regGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
77+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTA_CNTL_AUX, 0xfffffeef, 0x10b0000),
78+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_UTCL1_CNTL1, 0xffffffff, 0x30800400),
79+ SOC15_REG_GOLDEN_VALUE(GC, 0, regTCI_CNTL_3, 0xff, 0x20),
80+};
81+
82+static void gfx_v9_4_2_query_sq_timeout_status(struct amdgpu_device *adev);
83+static void gfx_v9_4_2_reset_sq_timeout_status(struct amdgpu_device *adev);
84+
85+void gfx_v9_4_2_init_golden_registers(struct amdgpu_device *adev,
86+ uint32_t die_id)
87+{
88+ soc15_program_register_sequence(adev,
89+ golden_settings_gc_9_4_2_alde,
90+ ARRAY_SIZE(golden_settings_gc_9_4_2_alde));
91+
92+ /* apply golden settings per die */
93+ switch (die_id) {
94+ case 0:
95+ soc15_program_register_sequence(adev,
96+ golden_settings_gc_9_4_2_alde_die_0,
97+ ARRAY_SIZE(golden_settings_gc_9_4_2_alde_die_0));
98+ break;
99+ case 1:
100+ soc15_program_register_sequence(adev,
101+ golden_settings_gc_9_4_2_alde_die_1,
102+ ARRAY_SIZE(golden_settings_gc_9_4_2_alde_die_1));
103+ break;
104+ default:
105+ dev_warn(adev->dev,
106+ "invalid die id %d, ignore channel fabricid remap settings\n",
107+ die_id);
108+ break;
109+ }
110+
111+ return;
112+}
113+
114+void gfx_v9_4_2_debug_trap_config_init(struct amdgpu_device *adev,
115+ uint32_t first_vmid,
116+ uint32_t last_vmid)
117+{
118+ uint32_t data;
119+ int i;
120+
121+ mutex_lock(&adev->srbm_mutex);
122+
123+ for (i = first_vmid; i < last_vmid; i++) {
124+ data = 0;
125+ soc15_grbm_select(adev, 0, 0, 0, i);
126+ data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
127+ data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
128+ data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE,
129+ 0);
130+ WREG32(SOC15_REG_OFFSET(GC, 0, regSPI_GDBG_PER_VMID_CNTL), data);
131+ }
132+
133+ soc15_grbm_select(adev, 0, 0, 0, 0);
134+ mutex_unlock(&adev->srbm_mutex);
135+}
136+
137+void gfx_v9_4_2_set_power_brake_sequence(struct amdgpu_device *adev)
138+{
139+ u32 tmp;
140+
141+ gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
142+
143+ tmp = 0;
144+ tmp = REG_SET_FIELD(tmp, GC_THROTTLE_CTRL, PATTERN_MODE, 1);
145+ WREG32_SOC15(GC, 0, regGC_THROTTLE_CTRL, tmp);
146+
147+ tmp = 0;
148+ tmp = REG_SET_FIELD(tmp, GC_THROTTLE_CTRL1, PWRBRK_STALL_EN, 1);
149+ WREG32_SOC15(GC, 0, regGC_THROTTLE_CTRL1, tmp);
150+
151+ WREG32_SOC15(GC, 0, regDIDT_IND_INDEX, ixDIDT_SQ_THROTTLE_CTRL);
152+ tmp = 0;
153+ tmp = REG_SET_FIELD(tmp, DIDT_SQ_THROTTLE_CTRL, PWRBRK_STALL_EN, 1);
154+ WREG32_SOC15(GC, 0, regDIDT_IND_DATA, tmp);
155+
156+ WREG32_SOC15(GC, 0, regGC_CAC_IND_INDEX, ixPWRBRK_STALL_PATTERN_CTRL);
157+ tmp = 0;
158+ tmp = REG_SET_FIELD(tmp, PWRBRK_STALL_PATTERN_CTRL, PWRBRK_END_STEP, 0x12);
159+ WREG32_SOC15(GC, 0, regGC_CAC_IND_DATA, tmp);
160+}
161+
162+static const struct soc15_reg_entry gfx_v9_4_2_edc_counter_regs[] = {
163+ /* CPF */
164+ { SOC15_REG_ENTRY(GC, 0, regCPF_EDC_ROQ_CNT), 0, 1, 1 },
165+ { SOC15_REG_ENTRY(GC, 0, regCPF_EDC_TAG_CNT), 0, 1, 1 },
166+ /* CPC */
167+ { SOC15_REG_ENTRY(GC, 0, regCPC_EDC_SCRATCH_CNT), 0, 1, 1 },
168+ { SOC15_REG_ENTRY(GC, 0, regCPC_EDC_UCODE_CNT), 0, 1, 1 },
169+ { SOC15_REG_ENTRY(GC, 0, regDC_EDC_STATE_CNT), 0, 1, 1 },
170+ { SOC15_REG_ENTRY(GC, 0, regDC_EDC_CSINVOC_CNT), 0, 1, 1 },
171+ { SOC15_REG_ENTRY(GC, 0, regDC_EDC_RESTORE_CNT), 0, 1, 1 },
172+ /* GDS */
173+ { SOC15_REG_ENTRY(GC, 0, regGDS_EDC_CNT), 0, 1, 1 },
174+ { SOC15_REG_ENTRY(GC, 0, regGDS_EDC_GRBM_CNT), 0, 1, 1 },
175+ { SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 1, 1 },
176+ { SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PHY_CNT), 0, 1, 1 },
177+ { SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PIPE_CNT), 0, 1, 1 },
178+ /* RLC */
179+ { SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT), 0, 1, 1 },
180+ { SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2), 0, 1, 1 },
181+ /* SPI */
182+ { SOC15_REG_ENTRY(GC, 0, regSPI_EDC_CNT), 0, 8, 1 },
183+ /* SQC */
184+ { SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT), 0, 8, 7 },
185+ { SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT2), 0, 8, 7 },
186+ { SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT3), 0, 8, 7 },
187+ { SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3), 0, 8, 7 },
188+ /* SQ */
189+ { SOC15_REG_ENTRY(GC, 0, regSQ_EDC_CNT), 0, 8, 14 },
190+ /* TCP */
191+ { SOC15_REG_ENTRY(GC, 0, regTCP_EDC_CNT_NEW), 0, 8, 14 },
192+ /* TCI */
193+ { SOC15_REG_ENTRY(GC, 0, regTCI_EDC_CNT), 0, 1, 69 },
194+ /* TCC */
195+ { SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT), 0, 1, 16 },
196+ { SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2), 0, 1, 16 },
197+ /* TCA */
198+ { SOC15_REG_ENTRY(GC, 0, regTCA_EDC_CNT), 0, 1, 2 },
199+ /* TCX */
200+ { SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT), 0, 1, 2 },
201+ { SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT2), 0, 1, 2 },
202+ /* TD */
203+ { SOC15_REG_ENTRY(GC, 0, regTD_EDC_CNT), 0, 8, 14 },
204+ /* TA */
205+ { SOC15_REG_ENTRY(GC, 0, regTA_EDC_CNT), 0, 8, 14 },
206+ /* GCEA */
207+ { SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT), 0, 1, 16 },
208+ { SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2), 0, 1, 16 },
209+ { SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3), 0, 1, 16 },
210+};
211+
212+static void gfx_v9_4_2_select_se_sh(struct amdgpu_device *adev, u32 se_num,
213+ u32 sh_num, u32 instance)
214+{
215+ u32 data;
216+
217+ if (instance == 0xffffffff)
218+ data = REG_SET_FIELD(0, GRBM_GFX_INDEX,
219+ INSTANCE_BROADCAST_WRITES, 1);
220+ else
221+ data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX,
222+ instance);
223+
224+ if (se_num == 0xffffffff)
225+ data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES,
226+ 1);
227+ else
228+ data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
229+
230+ if (sh_num == 0xffffffff)
231+ data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES,
232+ 1);
233+ else
234+ data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
235+
236+ WREG32_SOC15_RLC_SHADOW_EX(reg, GC, 0, regGRBM_GFX_INDEX, data);
237+}
238+
239+static const struct soc15_ras_field_entry gfx_v9_4_2_ras_fields[] = {
240+ /* CPF */
241+ { "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, regCPF_EDC_ROQ_CNT),
242+ SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, SEC_COUNT_ME2),
243+ SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, DED_COUNT_ME2) },
244+ { "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, regCPF_EDC_ROQ_CNT),
245+ SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, SEC_COUNT_ME1),
246+ SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, DED_COUNT_ME1) },
247+ { "CPF_TCIU_TAG", SOC15_REG_ENTRY(GC, 0, regCPF_EDC_TAG_CNT),
248+ SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT),
249+ SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT) },
250+
251+ /* CPC */
252+ { "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, regCPC_EDC_SCRATCH_CNT),
253+ SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT),
254+ SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT) },
255+ { "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, regCPC_EDC_UCODE_CNT),
256+ SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT),
257+ SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT) },
258+ { "CPC_DC_STATE_RAM_ME1", SOC15_REG_ENTRY(GC, 0, regDC_EDC_STATE_CNT),
259+ SOC15_REG_FIELD(DC_EDC_STATE_CNT, SEC_COUNT_ME1),
260+ SOC15_REG_FIELD(DC_EDC_STATE_CNT, DED_COUNT_ME1) },
261+ { "CPC_DC_CSINVOC_RAM_ME1",
262+ SOC15_REG_ENTRY(GC, 0, regDC_EDC_CSINVOC_CNT),
263+ SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, SEC_COUNT_ME1),
264+ SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, DED_COUNT_ME1) },
265+ { "CPC_DC_RESTORE_RAM_ME1",
266+ SOC15_REG_ENTRY(GC, 0, regDC_EDC_RESTORE_CNT),
267+ SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, SEC_COUNT_ME1),
268+ SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, DED_COUNT_ME1) },
269+ { "CPC_DC_CSINVOC_RAM1_ME1",
270+ SOC15_REG_ENTRY(GC, 0, regDC_EDC_CSINVOC_CNT),
271+ SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, SEC_COUNT1_ME1),
272+ SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, DED_COUNT1_ME1) },
273+ { "CPC_DC_RESTORE_RAM1_ME1",
274+ SOC15_REG_ENTRY(GC, 0, regDC_EDC_RESTORE_CNT),
275+ SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, SEC_COUNT1_ME1),
276+ SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, DED_COUNT1_ME1) },
277+
278+ /* GDS */
279+ { "GDS_GRBM", SOC15_REG_ENTRY(GC, 0, regGDS_EDC_GRBM_CNT),
280+ SOC15_REG_FIELD(GDS_EDC_GRBM_CNT, SEC),
281+ SOC15_REG_FIELD(GDS_EDC_GRBM_CNT, DED) },
282+ { "GDS_MEM", SOC15_REG_ENTRY(GC, 0, regGDS_EDC_CNT),
283+ SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC),
284+ SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED) },
285+ { "GDS_PHY_CMD_RAM_MEM", SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PHY_CNT),
286+ SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC),
287+ SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED) },
288+ { "GDS_PHY_DATA_RAM_MEM", SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PHY_CNT),
289+ SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SEC),
290+ SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_DED) },
291+ { "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PHY_CNT),
292+ SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC),
293+ SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED) },
294+ { "GDS_ME1_PIPE0_PIPE_MEM",
295+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PIPE_CNT),
296+ SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC),
297+ SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED) },
298+ { "GDS_ME1_PIPE1_PIPE_MEM",
299+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PIPE_CNT),
300+ SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC),
301+ SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED) },
302+ { "GDS_ME1_PIPE2_PIPE_MEM",
303+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PIPE_CNT),
304+ SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC),
305+ SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED) },
306+ { "GDS_ME1_PIPE3_PIPE_MEM",
307+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_PIPE_CNT),
308+ SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC),
309+ SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED) },
310+ { "GDS_ME0_GFXHP3D_PIX_DED",
311+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
312+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME0_GFXHP3D_PIX_DED) },
313+ { "GDS_ME0_GFXHP3D_VTX_DED",
314+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
315+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME0_GFXHP3D_VTX_DED) },
316+ { "GDS_ME0_CS_DED",
317+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
318+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME0_CS_DED) },
319+ { "GDS_ME0_GFXHP3D_GS_DED",
320+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
321+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME0_GFXHP3D_GS_DED) },
322+ { "GDS_ME1_PIPE0_DED",
323+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
324+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME1_PIPE0_DED) },
325+ { "GDS_ME1_PIPE1_DED",
326+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
327+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME1_PIPE1_DED) },
328+ { "GDS_ME1_PIPE2_DED",
329+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
330+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME1_PIPE2_DED) },
331+ { "GDS_ME1_PIPE3_DED",
332+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
333+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME1_PIPE3_DED) },
334+ { "GDS_ME2_PIPE0_DED",
335+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
336+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME2_PIPE0_DED) },
337+ { "GDS_ME2_PIPE1_DED",
338+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
339+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME2_PIPE1_DED) },
340+ { "GDS_ME2_PIPE2_DED",
341+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
342+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME2_PIPE2_DED) },
343+ { "GDS_ME2_PIPE3_DED",
344+ SOC15_REG_ENTRY(GC, 0, regGDS_EDC_OA_DED), 0, 0,
345+ SOC15_REG_FIELD(GDS_EDC_OA_DED, ME2_PIPE3_DED) },
346+
347+ /* RLC */
348+ { "RLCG_INSTR_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT),
349+ SOC15_REG_FIELD(RLC_EDC_CNT, RLCG_INSTR_RAM_SEC_COUNT),
350+ SOC15_REG_FIELD(RLC_EDC_CNT, RLCG_INSTR_RAM_DED_COUNT) },
351+ { "RLCG_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT),
352+ SOC15_REG_FIELD(RLC_EDC_CNT, RLCG_SCRATCH_RAM_SEC_COUNT),
353+ SOC15_REG_FIELD(RLC_EDC_CNT, RLCG_SCRATCH_RAM_DED_COUNT) },
354+ { "RLCV_INSTR_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT),
355+ SOC15_REG_FIELD(RLC_EDC_CNT, RLCV_INSTR_RAM_SEC_COUNT),
356+ SOC15_REG_FIELD(RLC_EDC_CNT, RLCV_INSTR_RAM_DED_COUNT) },
357+ { "RLCV_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT),
358+ SOC15_REG_FIELD(RLC_EDC_CNT, RLCV_SCRATCH_RAM_SEC_COUNT),
359+ SOC15_REG_FIELD(RLC_EDC_CNT, RLCV_SCRATCH_RAM_DED_COUNT) },
360+ { "RLC_TCTAG_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT),
361+ SOC15_REG_FIELD(RLC_EDC_CNT, RLC_TCTAG_RAM_SEC_COUNT),
362+ SOC15_REG_FIELD(RLC_EDC_CNT, RLC_TCTAG_RAM_DED_COUNT) },
363+ { "RLC_SPM_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT),
364+ SOC15_REG_FIELD(RLC_EDC_CNT, RLC_SPM_SCRATCH_RAM_SEC_COUNT),
365+ SOC15_REG_FIELD(RLC_EDC_CNT, RLC_SPM_SCRATCH_RAM_DED_COUNT) },
366+ { "RLC_SRM_DATA_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT),
367+ SOC15_REG_FIELD(RLC_EDC_CNT, RLC_SRM_DATA_RAM_SEC_COUNT),
368+ SOC15_REG_FIELD(RLC_EDC_CNT, RLC_SRM_DATA_RAM_DED_COUNT) },
369+ { "RLC_SRM_ADDR_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT),
370+ SOC15_REG_FIELD(RLC_EDC_CNT, RLC_SRM_ADDR_RAM_SEC_COUNT),
371+ SOC15_REG_FIELD(RLC_EDC_CNT, RLC_SRM_ADDR_RAM_DED_COUNT) },
372+ { "RLC_SPM_SE0_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2),
373+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE0_SCRATCH_RAM_SEC_COUNT),
374+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE0_SCRATCH_RAM_DED_COUNT) },
375+ { "RLC_SPM_SE1_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2),
376+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE1_SCRATCH_RAM_SEC_COUNT),
377+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE1_SCRATCH_RAM_DED_COUNT) },
378+ { "RLC_SPM_SE2_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2),
379+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE2_SCRATCH_RAM_SEC_COUNT),
380+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE2_SCRATCH_RAM_DED_COUNT) },
381+ { "RLC_SPM_SE3_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2),
382+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE3_SCRATCH_RAM_SEC_COUNT),
383+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE3_SCRATCH_RAM_DED_COUNT) },
384+ { "RLC_SPM_SE4_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2),
385+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE4_SCRATCH_RAM_SEC_COUNT),
386+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE4_SCRATCH_RAM_DED_COUNT) },
387+ { "RLC_SPM_SE5_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2),
388+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE5_SCRATCH_RAM_SEC_COUNT),
389+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE5_SCRATCH_RAM_DED_COUNT) },
390+ { "RLC_SPM_SE6_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2),
391+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE6_SCRATCH_RAM_SEC_COUNT),
392+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE6_SCRATCH_RAM_DED_COUNT) },
393+ { "RLC_SPM_SE7_SCRATCH_RAM", SOC15_REG_ENTRY(GC, 0, regRLC_EDC_CNT2),
394+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE7_SCRATCH_RAM_SEC_COUNT),
395+ SOC15_REG_FIELD(RLC_EDC_CNT2, RLC_SPM_SE7_SCRATCH_RAM_DED_COUNT) },
396+
397+ /* SPI */
398+ { "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, regSPI_EDC_CNT),
399+ SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SEC_COUNT),
400+ SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_DED_COUNT) },
401+ { "SPI_GDS_EXPREQ", SOC15_REG_ENTRY(GC, 0, regSPI_EDC_CNT),
402+ SOC15_REG_FIELD(SPI_EDC_CNT, SPI_GDS_EXPREQ_SEC_COUNT),
403+ SOC15_REG_FIELD(SPI_EDC_CNT, SPI_GDS_EXPREQ_DED_COUNT) },
404+ { "SPI_WB_GRANT_30", SOC15_REG_ENTRY(GC, 0, regSPI_EDC_CNT),
405+ SOC15_REG_FIELD(SPI_EDC_CNT, SPI_WB_GRANT_30_SEC_COUNT),
406+ SOC15_REG_FIELD(SPI_EDC_CNT, SPI_WB_GRANT_30_DED_COUNT) },
407+ { "SPI_LIFE_CNT", SOC15_REG_ENTRY(GC, 0, regSPI_EDC_CNT),
408+ SOC15_REG_FIELD(SPI_EDC_CNT, SPI_LIFE_CNT_SEC_COUNT),
409+ SOC15_REG_FIELD(SPI_EDC_CNT, SPI_LIFE_CNT_DED_COUNT) },
410+
411+ /* SQC - regSQC_EDC_CNT */
412+ { "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT),
413+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT),
414+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT) },
415+ { "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT),
416+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT),
417+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT) },
418+ { "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT),
419+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT),
420+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT) },
421+ { "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT),
422+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT),
423+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT) },
424+ { "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT),
425+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT),
426+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT) },
427+ { "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT),
428+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT),
429+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT) },
430+ { "SQC_DATA_CU3_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT),
431+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU3_WRITE_DATA_BUF_SEC_COUNT),
432+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU3_WRITE_DATA_BUF_DED_COUNT) },
433+ { "SQC_DATA_CU3_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT),
434+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU3_UTCL1_LFIFO_SEC_COUNT),
435+ SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU3_UTCL1_LFIFO_DED_COUNT) },
436+
437+ /* SQC - regSQC_EDC_CNT2 */
438+ { "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT2),
439+ SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT),
440+ SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT) },
441+ { "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT2),
442+ SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT),
443+ SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT) },
444+ { "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT2),
445+ SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT),
446+ SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT) },
447+ { "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT2),
448+ SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT),
449+ SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT) },
450+ { "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT2),
451+ SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT),
452+ SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT) },
453+ { "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT2),
454+ SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SEC_COUNT),
455+ SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_DED_COUNT) },
456+
457+ /* SQC - regSQC_EDC_CNT3 */
458+ { "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT3),
459+ SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT),
460+ SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT) },
461+ { "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT3),
462+ SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT),
463+ SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT) },
464+ { "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT3),
465+ SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT),
466+ SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT) },
467+ { "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT3),
468+ SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT),
469+ SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT) },
470+ { "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_CNT3),
471+ SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SEC_COUNT),
472+ SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_DED_COUNT) },
473+
474+ /* SQC - regSQC_EDC_PARITY_CNT3 */
475+ { "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3),
476+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, INST_BANKA_UTCL1_MISS_FIFO_SEC_COUNT),
477+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, INST_BANKA_UTCL1_MISS_FIFO_DED_COUNT) },
478+ { "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3),
479+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, INST_BANKA_MISS_FIFO_SEC_COUNT),
480+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, INST_BANKA_MISS_FIFO_DED_COUNT) },
481+ { "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3),
482+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, DATA_BANKA_HIT_FIFO_SEC_COUNT),
483+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, DATA_BANKA_HIT_FIFO_DED_COUNT) },
484+ { "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3),
485+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, DATA_BANKA_MISS_FIFO_SEC_COUNT),
486+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, DATA_BANKA_MISS_FIFO_DED_COUNT) },
487+ { "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3),
488+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SEC_COUNT),
489+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, INST_BANKB_UTCL1_MISS_FIFO_DED_COUNT) },
490+ { "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3),
491+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, INST_BANKB_MISS_FIFO_SEC_COUNT),
492+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, INST_BANKB_MISS_FIFO_DED_COUNT) },
493+ { "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3),
494+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, DATA_BANKB_HIT_FIFO_SEC_COUNT),
495+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, DATA_BANKB_HIT_FIFO_DED_COUNT) },
496+ { "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, regSQC_EDC_PARITY_CNT3),
497+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, DATA_BANKB_MISS_FIFO_SEC_COUNT),
498+ SOC15_REG_FIELD(SQC_EDC_PARITY_CNT3, DATA_BANKB_MISS_FIFO_DED_COUNT) },
499+
500+ /* SQ */
501+ { "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, regSQ_EDC_CNT),
502+ SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT),
503+ SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT) },
504+ { "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, regSQ_EDC_CNT),
505+ SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT),
506+ SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT) },
507+ { "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, regSQ_EDC_CNT),
508+ SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT),
509+ SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT) },
510+ { "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, regSQ_EDC_CNT),
511+ SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT),
512+ SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT) },
513+ { "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, regSQ_EDC_CNT),
514+ SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT),
515+ SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT) },
516+ { "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, regSQ_EDC_CNT),
517+ SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT),
518+ SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT) },
519+ { "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, regSQ_EDC_CNT),
520+ SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT),
521+ SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT) },
522+
523+ /* TCP */
524+ { "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, regTCP_EDC_CNT_NEW),
525+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT),
526+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT) },
527+ { "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, regTCP_EDC_CNT_NEW),
528+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT),
529+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT) },
530+ { "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, regTCP_EDC_CNT_NEW),
531+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SEC_COUNT),
532+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_DED_COUNT) },
533+ { "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, regTCP_EDC_CNT_NEW),
534+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT),
535+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_DED_COUNT) },
536+ { "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, regTCP_EDC_CNT_NEW),
537+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SEC_COUNT),
538+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_DED_COUNT) },
539+ { "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, regTCP_EDC_CNT_NEW),
540+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT),
541+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT) },
542+ { "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, regTCP_EDC_CNT_NEW),
543+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT),
544+ SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT) },
545+
546+ /* TCI */
547+ { "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, regTCI_EDC_CNT),
548+ SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SEC_COUNT),
549+ SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_DED_COUNT) },
550+
551+ /* TCC */
552+ { "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT),
553+ SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT),
554+ SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT) },
555+ { "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT),
556+ SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT),
557+ SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT) },
558+ { "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT),
559+ SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT),
560+ SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT) },
561+ { "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT),
562+ SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT),
563+ SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT) },
564+ { "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT),
565+ SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT),
566+ SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT) },
567+ { "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT),
568+ SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SEC_COUNT),
569+ SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_DED_COUNT) },
570+ { "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT),
571+ SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_NEXT_RAM_SEC_COUNT),
572+ SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_NEXT_RAM_DED_COUNT) },
573+ { "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2),
574+ SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SEC_COUNT),
575+ SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_DED_COUNT) },
576+ { "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2),
577+ SOC15_REG_FIELD(TCC_EDC_CNT2, UC_ATOMIC_FIFO_SEC_COUNT),
578+ SOC15_REG_FIELD(TCC_EDC_CNT2, UC_ATOMIC_FIFO_DED_COUNT) },
579+ { "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2),
580+ SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SEC_COUNT),
581+ SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_DED_COUNT) },
582+ { "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2),
583+ SOC15_REG_FIELD(TCC_EDC_CNT2, RETURN_CONTROL_SEC_COUNT),
584+ SOC15_REG_FIELD(TCC_EDC_CNT2, RETURN_CONTROL_DED_COUNT) },
585+ { "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2),
586+ SOC15_REG_FIELD(TCC_EDC_CNT2, IN_USE_TRANSFER_SEC_COUNT),
587+ SOC15_REG_FIELD(TCC_EDC_CNT2, IN_USE_TRANSFER_DED_COUNT) },
588+ { "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2),
589+ SOC15_REG_FIELD(TCC_EDC_CNT2, IN_USE_DEC_SEC_COUNT),
590+ SOC15_REG_FIELD(TCC_EDC_CNT2, IN_USE_DEC_DED_COUNT) },
591+ { "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2),
592+ SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SEC_COUNT),
593+ SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_DED_COUNT) },
594+ { "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, regTCC_EDC_CNT2),
595+ SOC15_REG_FIELD(TCC_EDC_CNT2, RETURN_DATA_SEC_COUNT),
596+ SOC15_REG_FIELD(TCC_EDC_CNT2, RETURN_DATA_DED_COUNT) },
597+
598+ /* TCA */
599+ { "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, regTCA_EDC_CNT),
600+ SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SEC_COUNT),
601+ SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_DED_COUNT) },
602+ { "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, regTCA_EDC_CNT),
603+ SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SEC_COUNT),
604+ SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_DED_COUNT) },
605+
606+ /* TCX */
607+ { "TCX_GROUP0", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
608+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP0_SEC_COUNT),
609+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP0_DED_COUNT) },
610+ { "TCX_GROUP1", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
611+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP1_SEC_COUNT),
612+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP1_DED_COUNT) },
613+ { "TCX_GROUP2", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
614+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP2_SEC_COUNT),
615+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP2_DED_COUNT) },
616+ { "TCX_GROUP3", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
617+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP3_SEC_COUNT),
618+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP3_DED_COUNT) },
619+ { "TCX_GROUP4", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
620+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP4_SEC_COUNT),
621+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP4_DED_COUNT) },
622+ { "TCX_GROUP5", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
623+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP5_SED_COUNT), 0, 0 },
624+ { "TCX_GROUP6", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
625+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP6_SED_COUNT), 0, 0 },
626+ { "TCX_GROUP7", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
627+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP7_SED_COUNT), 0, 0 },
628+ { "TCX_GROUP8", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
629+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP8_SED_COUNT), 0, 0 },
630+ { "TCX_GROUP9", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
631+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP9_SED_COUNT), 0, 0 },
632+ { "TCX_GROUP10", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT),
633+ SOC15_REG_FIELD(TCX_EDC_CNT, GROUP10_SED_COUNT), 0, 0 },
634+ { "TCX_GROUP11", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT2),
635+ SOC15_REG_FIELD(TCX_EDC_CNT2, GROUP11_SED_COUNT), 0, 0 },
636+ { "TCX_GROUP12", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT2),
637+ SOC15_REG_FIELD(TCX_EDC_CNT2, GROUP12_SED_COUNT), 0, 0 },
638+ { "TCX_GROUP13", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT2),
639+ SOC15_REG_FIELD(TCX_EDC_CNT2, GROUP13_SED_COUNT), 0, 0 },
640+ { "TCX_GROUP14", SOC15_REG_ENTRY(GC, 0, regTCX_EDC_CNT2),
641+ SOC15_REG_FIELD(TCX_EDC_CNT2, GROUP14_SED_COUNT), 0, 0 },
642+
643+ /* TD */
644+ { "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, regTD_EDC_CNT),
645+ SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT),
646+ SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT) },
647+ { "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, regTD_EDC_CNT),
648+ SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT),
649+ SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT) },
650+ { "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, regTD_EDC_CNT),
651+ SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SEC_COUNT),
652+ SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_DED_COUNT) },
653+
654+ /* TA */
655+ { "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, regTA_EDC_CNT),
656+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT),
657+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT) },
658+ { "TA_FS_AFIFO_LO", SOC15_REG_ENTRY(GC, 0, regTA_EDC_CNT),
659+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_LO_SEC_COUNT),
660+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_LO_DED_COUNT) },
661+ { "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, regTA_EDC_CNT),
662+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SEC_COUNT),
663+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_DED_COUNT) },
664+ { "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, regTA_EDC_CNT),
665+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SEC_COUNT),
666+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_DED_COUNT) },
667+ { "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, regTA_EDC_CNT),
668+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SEC_COUNT),
669+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_DED_COUNT) },
670+ { "TA_FS_AFIFO_HI", SOC15_REG_ENTRY(GC, 0, regTA_EDC_CNT),
671+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_HI_SEC_COUNT),
672+ SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_HI_DED_COUNT) },
673+
674+ /* EA - regGCEA_EDC_CNT */
675+ { "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
676+ SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
677+ SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT) },
678+ { "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
679+ SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
680+ SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT) },
681+ { "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
682+ SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
683+ SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT) },
684+ { "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
685+ SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
686+ SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT) },
687+ { "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
688+ SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
689+ SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT) },
690+ { "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
691+ SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SEC_COUNT),
692+ SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_DED_COUNT) },
693+ { "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
694+ SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT), 0, 0 },
695+ { "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
696+ SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT), 0, 0 },
697+ { "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
698+ SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT), 0, 0 },
699+ { "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT),
700+ SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT), 0, 0 },
701+
702+ /* EA - regGCEA_EDC_CNT2 */
703+ { "EA_GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
704+ SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
705+ SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT) },
706+ { "EA_GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
707+ SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
708+ SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT) },
709+ { "EA_GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
710+ SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
711+ SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT) },
712+ { "EA_GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
713+ SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT), 0, 0 },
714+ { "EA_GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
715+ SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT), 0, 0 },
716+ { "EA_MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
717+ SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT),
718+ SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_DED_COUNT) },
719+ { "EA_MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
720+ SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT),
721+ SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_DED_COUNT) },
722+ { "EA_MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
723+ SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT),
724+ SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_DED_COUNT) },
725+ { "EA_MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT2),
726+ SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT),
727+ SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_DED_COUNT) },
728+
729+ /* EA - regGCEA_EDC_CNT3 */
730+ { "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3), 0, 0,
731+ SOC15_REG_FIELD(GCEA_EDC_CNT3, DRAMRD_PAGEMEM_DED_COUNT) },
732+ { "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3), 0, 0,
733+ SOC15_REG_FIELD(GCEA_EDC_CNT3, DRAMWR_PAGEMEM_DED_COUNT) },
734+ { "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3), 0, 0,
735+ SOC15_REG_FIELD(GCEA_EDC_CNT3, IORD_CMDMEM_DED_COUNT) },
736+ { "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3), 0, 0,
737+ SOC15_REG_FIELD(GCEA_EDC_CNT3, IOWR_CMDMEM_DED_COUNT) },
738+ { "EA_GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3), 0, 0,
739+ SOC15_REG_FIELD(GCEA_EDC_CNT3, GMIRD_PAGEMEM_DED_COUNT) },
740+ { "EA_GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3), 0, 0,
741+ SOC15_REG_FIELD(GCEA_EDC_CNT3, GMIWR_PAGEMEM_DED_COUNT) },
742+ { "EA_MAM_A0MEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3),
743+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_A0MEM_SEC_COUNT),
744+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_A0MEM_DED_COUNT) },
745+ { "EA_MAM_A1MEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3),
746+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_A1MEM_SEC_COUNT),
747+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_A1MEM_DED_COUNT) },
748+ { "EA_MAM_A2MEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3),
749+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_A2MEM_SEC_COUNT),
750+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_A2MEM_DED_COUNT) },
751+ { "EA_MAM_A3MEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3),
752+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_A3MEM_SEC_COUNT),
753+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_A3MEM_DED_COUNT) },
754+ { "EA_MAM_AFMEM", SOC15_REG_ENTRY(GC, 0, regGCEA_EDC_CNT3),
755+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_AFMEM_SEC_COUNT),
756+ SOC15_REG_FIELD(GCEA_EDC_CNT3, MAM_AFMEM_DED_COUNT) },
757+};
758+
759+static const char * const vml2_walker_mems[] = {
760+ "UTC_VML2_CACHE_PDE0_MEM0",
761+ "UTC_VML2_CACHE_PDE0_MEM1",
762+ "UTC_VML2_CACHE_PDE1_MEM0",
763+ "UTC_VML2_CACHE_PDE1_MEM1",
764+ "UTC_VML2_CACHE_PDE2_MEM0",
765+ "UTC_VML2_CACHE_PDE2_MEM1",
766+ "UTC_VML2_RDIF_ARADDRS",
767+ "UTC_VML2_RDIF_LOG_FIFO",
768+ "UTC_VML2_QUEUE_REQ",
769+ "UTC_VML2_QUEUE_RET",
770+};
771+
772+static struct gfx_v9_4_2_utc_block gfx_v9_4_2_utc_blocks[] = {
773+ { VML2_MEM, 8, 2, 2,
774+ { SOC15_REG_ENTRY(GC, 0, regVML2_MEM_ECC_INDEX) },
775+ { SOC15_REG_ENTRY(GC, 0, regVML2_MEM_ECC_CNTL) },
776+ SOC15_REG_FIELD(VML2_MEM_ECC_CNTL, SEC_COUNT),
777+ SOC15_REG_FIELD(VML2_MEM_ECC_CNTL, DED_COUNT),
778+ REG_SET_FIELD(0, VML2_MEM_ECC_CNTL, WRITE_COUNTERS, 1) },
779+ { VML2_WALKER_MEM, ARRAY_SIZE(vml2_walker_mems), 1, 1,
780+ { SOC15_REG_ENTRY(GC, 0, regVML2_WALKER_MEM_ECC_INDEX) },
781+ { SOC15_REG_ENTRY(GC, 0, regVML2_WALKER_MEM_ECC_CNTL) },
782+ SOC15_REG_FIELD(VML2_WALKER_MEM_ECC_CNTL, SEC_COUNT),
783+ SOC15_REG_FIELD(VML2_WALKER_MEM_ECC_CNTL, DED_COUNT),
784+ REG_SET_FIELD(0, VML2_WALKER_MEM_ECC_CNTL, WRITE_COUNTERS, 1) },
785+ { UTCL2_MEM, 18, 1, 2,
786+ { SOC15_REG_ENTRY(GC, 0, regUTCL2_MEM_ECC_INDEX) },
787+ { SOC15_REG_ENTRY(GC, 0, regUTCL2_MEM_ECC_CNTL) },
788+ SOC15_REG_FIELD(UTCL2_MEM_ECC_CNTL, SEC_COUNT),
789+ SOC15_REG_FIELD(UTCL2_MEM_ECC_CNTL, DED_COUNT),
790+ REG_SET_FIELD(0, UTCL2_MEM_ECC_CNTL, WRITE_COUNTERS, 1) },
791+ { ATC_L2_CACHE_2M, 8, 2, 1,
792+ { SOC15_REG_ENTRY(GC, 0, regATC_L2_CACHE_2M_DSM_INDEX) },
793+ { SOC15_REG_ENTRY(GC, 0, regATC_L2_CACHE_2M_DSM_CNTL) },
794+ SOC15_REG_FIELD(ATC_L2_CACHE_2M_DSM_CNTL, SEC_COUNT),
795+ SOC15_REG_FIELD(ATC_L2_CACHE_2M_DSM_CNTL, DED_COUNT),
796+ REG_SET_FIELD(0, ATC_L2_CACHE_2M_DSM_CNTL, WRITE_COUNTERS, 1) },
797+ { ATC_L2_CACHE_32K, 8, 2, 2,
798+ { SOC15_REG_ENTRY(GC, 0, regATC_L2_CACHE_32K_DSM_INDEX) },
799+ { SOC15_REG_ENTRY(GC, 0, regATC_L2_CACHE_32K_DSM_CNTL) },
800+ SOC15_REG_FIELD(ATC_L2_CACHE_32K_DSM_CNTL, SEC_COUNT),
801+ SOC15_REG_FIELD(ATC_L2_CACHE_32K_DSM_CNTL, DED_COUNT),
802+ REG_SET_FIELD(0, ATC_L2_CACHE_32K_DSM_CNTL, WRITE_COUNTERS, 1) },
803+ { ATC_L2_CACHE_4K, 8, 2, 8,
804+ { SOC15_REG_ENTRY(GC, 0, regATC_L2_CACHE_4K_DSM_INDEX) },
805+ { SOC15_REG_ENTRY(GC, 0, regATC_L2_CACHE_4K_DSM_CNTL) },
806+ SOC15_REG_FIELD(ATC_L2_CACHE_4K_DSM_CNTL, SEC_COUNT),
807+ SOC15_REG_FIELD(ATC_L2_CACHE_4K_DSM_CNTL, DED_COUNT),
808+ REG_SET_FIELD(0, ATC_L2_CACHE_4K_DSM_CNTL, WRITE_COUNTERS, 1) },
809+};
810+
811+static const struct soc15_reg_entry gfx_v9_4_2_rdrsp_status_regs =
812+ { SOC15_REG_ENTRY(GC, 0, regGCEA_ERR_STATUS), 0, 1, 16 };
813+
814+static int gfx_v9_4_2_get_reg_error_count(struct amdgpu_device *adev,
815+ const struct soc15_reg_entry *reg,
816+ uint32_t se_id, uint32_t inst_id,
817+ uint32_t value, uint32_t *sec_count,
818+ uint32_t *ded_count)
819+{
820+ uint32_t i;
821+ uint32_t sec_cnt, ded_cnt;
822+
823+ for (i = 0; i < ARRAY_SIZE(gfx_v9_4_2_ras_fields); i++) {
824+ if (gfx_v9_4_2_ras_fields[i].reg_offset != reg->reg_offset ||
825+ gfx_v9_4_2_ras_fields[i].seg != reg->seg ||
826+ gfx_v9_4_2_ras_fields[i].inst != reg->inst)
827+ continue;
828+
829+ sec_cnt = SOC15_RAS_REG_FIELD_VAL(
830+ value, gfx_v9_4_2_ras_fields[i], sec);
831+ if (sec_cnt) {
832+ dev_info(adev->dev,
833+ "GFX SubBlock %s, Instance[%d][%d], SEC %d\n",
834+ gfx_v9_4_2_ras_fields[i].name, se_id, inst_id,
835+ sec_cnt);
836+ *sec_count += sec_cnt;
837+ }
838+
839+ ded_cnt = SOC15_RAS_REG_FIELD_VAL(
840+ value, gfx_v9_4_2_ras_fields[i], ded);
841+ if (ded_cnt) {
842+ dev_info(adev->dev,
843+ "GFX SubBlock %s, Instance[%d][%d], DED %d\n",
844+ gfx_v9_4_2_ras_fields[i].name, se_id, inst_id,
845+ ded_cnt);
846+ *ded_count += ded_cnt;
847+ }
848+ }
849+
850+ return 0;
851+}
852+
853+static int gfx_v9_4_2_query_sram_edc_count(struct amdgpu_device *adev,
854+ uint32_t *sec_count, uint32_t *ded_count)
855+{
856+ uint32_t i, j, k, data;
857+ uint32_t sec_cnt = 0, ded_cnt = 0;
858+
859+ if (sec_count && ded_count) {
860+ *sec_count = 0;
861+ *ded_count = 0;
862+ }
863+
864+ mutex_lock(&adev->grbm_idx_mutex);
865+
866+ for (i = 0; i < ARRAY_SIZE(gfx_v9_4_2_edc_counter_regs); i++) {
867+ for (j = 0; j < gfx_v9_4_2_edc_counter_regs[i].se_num; j++) {
868+ for (k = 0; k < gfx_v9_4_2_edc_counter_regs[i].instance;
869+ k++) {
870+ gfx_v9_4_2_select_se_sh(adev, j, 0, k);
871+
872+ /* if sec/ded_count is null, just clear counter */
873+ if (!sec_count || !ded_count) {
874+ WREG32(SOC15_REG_ENTRY_OFFSET(
875+ gfx_v9_4_2_edc_counter_regs[i]), 0);
876+ continue;
877+ }
878+
879+ data = RREG32(SOC15_REG_ENTRY_OFFSET(
880+ gfx_v9_4_2_edc_counter_regs[i]));
881+
882+ if (!data)
883+ continue;
884+
885+ gfx_v9_4_2_get_reg_error_count(adev,
886+ &gfx_v9_4_2_edc_counter_regs[i],
887+ j, k, data, &sec_cnt, &ded_cnt);
888+
889+ /* clear counter after read */
890+ WREG32(SOC15_REG_ENTRY_OFFSET(
891+ gfx_v9_4_2_edc_counter_regs[i]), 0);
892+ }
893+ }
894+ }
895+
896+ if (sec_count && ded_count) {
897+ *sec_count += sec_cnt;
898+ *ded_count += ded_cnt;
899+ }
900+
901+ gfx_v9_4_2_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
902+ mutex_unlock(&adev->grbm_idx_mutex);
903+
904+ return 0;
905+}
906+
907+static void gfx_v9_4_2_log_utc_edc_count(struct amdgpu_device *adev,
908+ struct gfx_v9_4_2_utc_block *blk,
909+ uint32_t instance, uint32_t sec_cnt,
910+ uint32_t ded_cnt)
911+{
912+ uint32_t bank, way, mem;
913+ static const char *vml2_way_str[] = { "BIGK", "4K" };
914+ static const char *utcl2_rounter_str[] = { "VMC", "APT" };
915+
916+ mem = instance % blk->num_mem_blocks;
917+ way = (instance / blk->num_mem_blocks) % blk->num_ways;
918+ bank = instance / (blk->num_mem_blocks * blk->num_ways);
919+
920+ switch (blk->type) {
921+ case VML2_MEM:
922+ dev_info(
923+ adev->dev,
924+ "GFX SubBlock UTC_VML2_BANK_CACHE_%d_%s_MEM%d, SED %d, DED %d\n",
925+ bank, vml2_way_str[way], mem, sec_cnt, ded_cnt);
926+ break;
927+ case VML2_WALKER_MEM:
928+ dev_info(adev->dev, "GFX SubBlock %s, SED %d, DED %d\n",
929+ vml2_walker_mems[bank], sec_cnt, ded_cnt);
930+ break;
931+ case UTCL2_MEM:
932+ dev_info(
933+ adev->dev,
934+ "GFX SubBlock UTCL2_ROUTER_IFIF%d_GROUP0_%s, SED %d, DED %d\n",
935+ bank, utcl2_rounter_str[mem], sec_cnt, ded_cnt);
936+ break;
937+ case ATC_L2_CACHE_2M:
938+ dev_info(
939+ adev->dev,
940+ "GFX SubBlock UTC_ATCL2_CACHE_2M_BANK%d_WAY%d_MEM, SED %d, DED %d\n",
941+ bank, way, sec_cnt, ded_cnt);
942+ break;
943+ case ATC_L2_CACHE_32K:
944+ dev_info(
945+ adev->dev,
946+ "GFX SubBlock UTC_ATCL2_CACHE_32K_BANK%d_WAY%d_MEM%d, SED %d, DED %d\n",
947+ bank, way, mem, sec_cnt, ded_cnt);
948+ break;
949+ case ATC_L2_CACHE_4K:
950+ dev_info(
951+ adev->dev,
952+ "GFX SubBlock UTC_ATCL2_CACHE_4K_BANK%d_WAY%d_MEM%d, SED %d, DED %d\n",
953+ bank, way, mem, sec_cnt, ded_cnt);
954+ break;
955+ }
956+}
957+
958+static int gfx_v9_4_2_query_utc_edc_count(struct amdgpu_device *adev,
959+ uint32_t *sec_count,
960+ uint32_t *ded_count)
961+{
962+ uint32_t i, j, data;
963+ uint32_t sec_cnt, ded_cnt;
964+ uint32_t num_instances;
965+ struct gfx_v9_4_2_utc_block *blk;
966+
967+ if (sec_count && ded_count) {
968+ *sec_count = 0;
969+ *ded_count = 0;
970+ }
971+
972+ for (i = 0; i < ARRAY_SIZE(gfx_v9_4_2_utc_blocks); i++) {
973+ blk = &gfx_v9_4_2_utc_blocks[i];
974+ num_instances =
975+ blk->num_banks * blk->num_ways * blk->num_mem_blocks;
976+ for (j = 0; j < num_instances; j++) {
977+ WREG32(SOC15_REG_ENTRY_OFFSET(blk->idx_reg), j);
978+
979+ /* if sec/ded_count is NULL, just clear counter */
980+ if (!sec_count || !ded_count) {
981+ WREG32(SOC15_REG_ENTRY_OFFSET(blk->data_reg),
982+ blk->clear);
983+ continue;
984+ }
985+
986+ data = RREG32(SOC15_REG_ENTRY_OFFSET(blk->data_reg));
987+ if (!data)
988+ continue;
989+
990+ sec_cnt = SOC15_RAS_REG_FIELD_VAL(data, *blk, sec);
991+ *sec_count += sec_cnt;
992+ ded_cnt = SOC15_RAS_REG_FIELD_VAL(data, *blk, ded);
993+ *ded_count += ded_cnt;
994+
995+ /* clear counter after read */
996+ WREG32(SOC15_REG_ENTRY_OFFSET(blk->data_reg),
997+ blk->clear);
998+
999+ /* print the edc count */
1000+ gfx_v9_4_2_log_utc_edc_count(adev, blk, j, sec_cnt,
1001+ ded_cnt);
1002+ }
1003+ }
1004+
1005+ return 0;
1006+}
1007+
1008+int gfx_v9_4_2_query_ras_error_count(struct amdgpu_device *adev,
1009+ void *ras_error_status)
1010+{
1011+ struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
1012+ uint32_t sec_count = 0, ded_count = 0;
1013+
1014+ if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
1015+ return -EINVAL;
1016+
1017+ err_data->ue_count = 0;
1018+ err_data->ce_count = 0;
1019+
1020+ gfx_v9_4_2_query_sram_edc_count(adev, &sec_count, &ded_count);
1021+ err_data->ce_count += sec_count;
1022+ err_data->ue_count += ded_count;
1023+
1024+ gfx_v9_4_2_query_utc_edc_count(adev, &sec_count, &ded_count);
1025+ err_data->ce_count += sec_count;
1026+ err_data->ue_count += ded_count;
1027+
1028+ return 0;
1029+}
1030+
1031+static void gfx_v9_4_2_reset_utc_err_status(struct amdgpu_device *adev)
1032+{
1033+ WREG32_SOC15(GC, 0, regUTCL2_MEM_ECC_STATUS, 0x3);
1034+ WREG32_SOC15(GC, 0, regVML2_MEM_ECC_STATUS, 0x3);
1035+ WREG32_SOC15(GC, 0, regVML2_WALKER_MEM_ECC_STATUS, 0x3);
1036+}
1037+
1038+static void gfx_v9_4_2_reset_ea_err_status(struct amdgpu_device *adev)
1039+{
1040+ uint32_t i, j;
1041+
1042+ mutex_lock(&adev->grbm_idx_mutex);
1043+ for (i = 0; i < gfx_v9_4_2_rdrsp_status_regs.se_num; i++) {
1044+ for (j = 0; j < gfx_v9_4_2_rdrsp_status_regs.instance;
1045+ j++) {
1046+ gfx_v9_4_2_select_se_sh(adev, i, 0, j);
1047+ WREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_4_2_rdrsp_status_regs), 0x10);
1048+ }
1049+ }
1050+ gfx_v9_4_2_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1051+ mutex_unlock(&adev->grbm_idx_mutex);
1052+}
1053+
1054+void gfx_v9_4_2_reset_ras_error_count(struct amdgpu_device *adev)
1055+{
1056+ if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
1057+ return;
1058+
1059+ gfx_v9_4_2_query_sram_edc_count(adev, NULL, NULL);
1060+ gfx_v9_4_2_query_utc_edc_count(adev, NULL, NULL);
1061+}
1062+
1063+int gfx_v9_4_2_ras_error_inject(struct amdgpu_device *adev, void *inject_if)
1064+{
1065+ struct ras_inject_if *info = (struct ras_inject_if *)inject_if;
1066+ int ret;
1067+ struct ta_ras_trigger_error_input block_info = { 0 };
1068+
1069+ if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
1070+ return -EINVAL;
1071+
1072+ block_info.block_id = amdgpu_ras_block_to_ta(info->head.block);
1073+ block_info.sub_block_index = info->head.sub_block_index;
1074+ block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type);
1075+ block_info.address = info->address;
1076+ block_info.value = info->value;
1077+
1078+ mutex_lock(&adev->grbm_idx_mutex);
1079+ ret = psp_ras_trigger_error(&adev->psp, &block_info);
1080+ mutex_unlock(&adev->grbm_idx_mutex);
1081+
1082+ return ret;
1083+}
1084+
1085+static void gfx_v9_4_2_query_ea_err_status(struct amdgpu_device *adev)
1086+{
1087+ uint32_t i, j;
1088+ uint32_t reg_value;
1089+
1090+ mutex_lock(&adev->grbm_idx_mutex);
1091+
1092+ for (i = 0; i < gfx_v9_4_2_rdrsp_status_regs.se_num; i++) {
1093+ for (j = 0; j < gfx_v9_4_2_rdrsp_status_regs.instance;
1094+ j++) {
1095+ gfx_v9_4_2_select_se_sh(adev, i, 0, j);
1096+ reg_value = RREG32(SOC15_REG_ENTRY_OFFSET(
1097+ gfx_v9_4_2_rdrsp_status_regs));
1098+ if (reg_value)
1099+ dev_warn(adev->dev, "GCEA err detected at instance: %d, status: 0x%x!\n",
1100+ j, reg_value);
1101+ /* clear after read */
1102+ WREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_4_2_rdrsp_status_regs), 0x10);
1103+ }
1104+ }
1105+
1106+ gfx_v9_4_2_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1107+ mutex_unlock(&adev->grbm_idx_mutex);
1108+}
1109+
1110+static void gfx_v9_4_2_query_utc_err_status(struct amdgpu_device *adev)
1111+{
1112+ uint32_t data;
1113+
1114+ data = RREG32_SOC15(GC, 0, regUTCL2_MEM_ECC_STATUS);
1115+ if (!data) {
1116+ dev_warn(adev->dev, "GFX UTCL2 Mem Ecc Status: 0x%x!\n", data);
1117+ WREG32_SOC15(GC, 0, regUTCL2_MEM_ECC_STATUS, 0x3);
1118+ }
1119+
1120+ data = RREG32_SOC15(GC, 0, regVML2_MEM_ECC_STATUS);
1121+ if (!data) {
1122+ dev_warn(adev->dev, "GFX VML2 Mem Ecc Status: 0x%x!\n", data);
1123+ WREG32_SOC15(GC, 0, regVML2_MEM_ECC_STATUS, 0x3);
1124+ }
1125+
1126+ data = RREG32_SOC15(GC, 0, regVML2_WALKER_MEM_ECC_STATUS);
1127+ if (!data) {
1128+ dev_warn(adev->dev, "GFX VML2 Walker Mem Ecc Status: 0x%x!\n", data);
1129+ WREG32_SOC15(GC, 0, regVML2_WALKER_MEM_ECC_STATUS, 0x3);
1130+ }
1131+}
1132+
1133+void gfx_v9_4_2_query_ras_error_status(struct amdgpu_device *adev)
1134+{
1135+ if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
1136+ return;
1137+
1138+ gfx_v9_4_2_query_ea_err_status(adev);
1139+ gfx_v9_4_2_query_utc_err_status(adev);
1140+ gfx_v9_4_2_query_sq_timeout_status(adev);
1141+}
1142+
1143+void gfx_v9_4_2_reset_ras_error_status(struct amdgpu_device *adev)
1144+{
1145+ if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
1146+ return;
1147+
1148+ gfx_v9_4_2_reset_utc_err_status(adev);
1149+ gfx_v9_4_2_reset_ea_err_status(adev);
1150+ gfx_v9_4_2_reset_sq_timeout_status(adev);
1151+}
1152+
1153+void gfx_v9_4_2_enable_watchdog_timer(struct amdgpu_device *adev)
1154+{
1155+ uint32_t i;
1156+ uint32_t data;
1157+
1158+ data = REG_SET_FIELD(0, SQ_TIMEOUT_CONFIG, TIMEOUT_FATAL_DISABLE,
1159+ amdgpu_watchdog_timer.timeout_fatal_disable ? 1 :
1160+ 0);
1161+
1162+ if (amdgpu_watchdog_timer.timeout_fatal_disable &&
1163+ (amdgpu_watchdog_timer.period < 1 ||
1164+ amdgpu_watchdog_timer.period > 0x23)) {
1165+ dev_warn(adev->dev, "Watchdog period range is 1 to 0x23\n");
1166+ amdgpu_watchdog_timer.period = 0x23;
1167+ }
1168+ data = REG_SET_FIELD(data, SQ_TIMEOUT_CONFIG, PERIOD_SEL,
1169+ amdgpu_watchdog_timer.period);
1170+
1171+ mutex_lock(&adev->grbm_idx_mutex);
1172+ for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1173+ gfx_v9_4_2_select_se_sh(adev, i, 0xffffffff, 0xffffffff);
1174+ WREG32_SOC15(GC, 0, regSQ_TIMEOUT_CONFIG, data);
1175+ }
1176+ gfx_v9_4_2_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1177+ mutex_unlock(&adev->grbm_idx_mutex);
1178+}
1179+
1180+static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
1181+{
1182+ WREG32_SOC15_RLC_EX(reg, GC, 0, regSQ_IND_INDEX,
1183+ (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1184+ (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
1185+ (address << SQ_IND_INDEX__INDEX__SHIFT) |
1186+ (SQ_IND_INDEX__FORCE_READ_MASK));
1187+ return RREG32_SOC15(GC, 0, regSQ_IND_DATA);
1188+}
1189+
1190+static void gfx_v9_4_2_log_cu_timeout_status(struct amdgpu_device *adev,
1191+ uint32_t status)
1192+{
1193+ struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
1194+ uint32_t i, simd, wave;
1195+ uint32_t wave_status;
1196+ uint32_t wave_pc_lo, wave_pc_hi;
1197+ uint32_t wave_exec_lo, wave_exec_hi;
1198+ uint32_t wave_inst_dw0, wave_inst_dw1;
1199+ uint32_t wave_ib_sts;
1200+
1201+ for (i = 0; i < 32; i++) {
1202+ if (!((i << 1) & status))
1203+ continue;
1204+
1205+ simd = i / cu_info->max_waves_per_simd;
1206+ wave = i % cu_info->max_waves_per_simd;
1207+
1208+ wave_status = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
1209+ wave_pc_lo = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
1210+ wave_pc_hi = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
1211+ wave_exec_lo =
1212+ wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
1213+ wave_exec_hi =
1214+ wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
1215+ wave_inst_dw0 =
1216+ wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
1217+ wave_inst_dw1 =
1218+ wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
1219+ wave_ib_sts = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
1220+
1221+ dev_info(
1222+ adev->dev,
1223+ "\t SIMD %d, Wave %d: status 0x%x, pc 0x%llx, exec 0x%llx, inst 0x%llx, ib_sts 0x%x\n",
1224+ simd, wave, wave_status,
1225+ ((uint64_t)wave_pc_hi << 32 | wave_pc_lo),
1226+ ((uint64_t)wave_exec_hi << 32 | wave_exec_lo),
1227+ ((uint64_t)wave_inst_dw1 << 32 | wave_inst_dw0),
1228+ wave_ib_sts);
1229+ }
1230+}
1231+
1232+static void gfx_v9_4_2_query_sq_timeout_status(struct amdgpu_device *adev)
1233+{
1234+ uint32_t se_idx, sh_idx, cu_idx;
1235+ uint32_t status;
1236+
1237+ mutex_lock(&adev->grbm_idx_mutex);
1238+ for (se_idx = 0; se_idx < adev->gfx.config.max_shader_engines;
1239+ se_idx++) {
1240+ for (sh_idx = 0; sh_idx < adev->gfx.config.max_sh_per_se;
1241+ sh_idx++) {
1242+ for (cu_idx = 0;
1243+ cu_idx < adev->gfx.config.max_cu_per_sh;
1244+ cu_idx++) {
1245+ gfx_v9_4_2_select_se_sh(adev, se_idx, sh_idx,
1246+ cu_idx);
1247+ status = RREG32_SOC15(GC, 0,
1248+ regSQ_TIMEOUT_STATUS);
1249+ if (status != 0) {
1250+ dev_info(
1251+ adev->dev,
1252+ "GFX Watchdog Timeout: SE %d, SH %d, CU %d\n",
1253+ se_idx, sh_idx, cu_idx);
1254+ gfx_v9_4_2_log_cu_timeout_status(
1255+ adev, status);
1256+ }
1257+ /* clear old status */
1258+ WREG32_SOC15(GC, 0, regSQ_TIMEOUT_STATUS, 0);
1259+ }
1260+ }
1261+ }
1262+ gfx_v9_4_2_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1263+ mutex_unlock(&adev->grbm_idx_mutex);
1264+}
1265+
1266+static void gfx_v9_4_2_reset_sq_timeout_status(struct amdgpu_device *adev)
1267+{
1268+ uint32_t se_idx, sh_idx, cu_idx;
1269+
1270+ mutex_lock(&adev->grbm_idx_mutex);
1271+ for (se_idx = 0; se_idx < adev->gfx.config.max_shader_engines;
1272+ se_idx++) {
1273+ for (sh_idx = 0; sh_idx < adev->gfx.config.max_sh_per_se;
1274+ sh_idx++) {
1275+ for (cu_idx = 0;
1276+ cu_idx < adev->gfx.config.max_cu_per_sh;
1277+ cu_idx++) {
1278+ gfx_v9_4_2_select_se_sh(adev, se_idx, sh_idx,
1279+ cu_idx);
1280+ WREG32_SOC15(GC, 0, regSQ_TIMEOUT_STATUS, 0);
1281+ }
1282+ }
1283+ }
1284+ gfx_v9_4_2_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1285+ mutex_unlock(&adev->grbm_idx_mutex);
1286+}
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.h+40−0
@@ -0,0 +1,40 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+
24+#ifndef __GFX_V9_4_2_H__
25+#define __GFX_V9_4_2_H__
26+
27+void gfx_v9_4_2_debug_trap_config_init(struct amdgpu_device *adev,
28+ uint32_t first_vmid, uint32_t last_vmid);
29+void gfx_v9_4_2_init_golden_registers(struct amdgpu_device *adev,
30+ uint32_t die_id);
31+void gfx_v9_4_2_set_power_brake_sequence(struct amdgpu_device *adev);
32+
33+void gfx_v9_4_2_reset_ras_error_count(struct amdgpu_device *adev);
34+int gfx_v9_4_2_ras_error_inject(struct amdgpu_device *adev, void *inject_if);
35+void gfx_v9_4_2_query_ras_error_status(struct amdgpu_device *adev);
36+int gfx_v9_4_2_query_ras_error_count(struct amdgpu_device *adev,
37+ void *ras_error_status);
38+void gfx_v9_4_2_reset_ras_error_status(struct amdgpu_device *adev);
39+void gfx_v9_4_2_enable_watchdog_timer(struct amdgpu_device *adev);
40+#endif /* __GFX_V9_4_2_H__ */
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c+62−13
@@ -53,19 +53,39 @@ static void gfxhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, · 52 unmodified lines
5353
5454 static void gfxhub_v1_0_init_gart_aperture_regs(struct amdgpu_device *adev)
5555 {
56− uint64_t pt_base = amdgpu_gmc_pd_addr(adev->gart.bo);
56+ uint64_t pt_base;
5757
58− gfxhub_v1_0_setup_vm_pt_regs(adev, 0, pt_base);
58+ if (adev->gmc.pdb0_bo)
59+ pt_base = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo);
60+ else
61+ pt_base = amdgpu_gmc_pd_addr(adev->gart.bo);
5962
60− WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
61− (u32)(adev->gmc.gart_start >> 12));
62− WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
63− (u32)(adev->gmc.gart_start >> 44));
63+ gfxhub_v1_0_setup_vm_pt_regs(adev, 0, pt_base);
6464
65− WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
66− (u32)(adev->gmc.gart_end >> 12));
67− WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
68− (u32)(adev->gmc.gart_end >> 44));
65+ /* If use GART for FB translation, vmid0 page table covers both
66+ * vram and system memory (gart)
67+ */
68+ if (adev->gmc.pdb0_bo) {
69+ WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
70+ (u32)(adev->gmc.fb_start >> 12));
71+ WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
72+ (u32)(adev->gmc.fb_start >> 44));
73+
74+ WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
75+ (u32)(adev->gmc.gart_end >> 12));
76+ WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
77+ (u32)(adev->gmc.gart_end >> 44));
78+ } else {
79+ WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
80+ (u32)(adev->gmc.gart_start >> 12));
81+ WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
82+ (u32)(adev->gmc.gart_start >> 44));
83+
84+ WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
85+ (u32)(adev->gmc.gart_end >> 12));
86+ WREG32_SOC15(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
87+ (u32)(adev->gmc.gart_end >> 44));
88+ }
6989 }
7090
7191 static void gfxhub_v1_0_init_system_aperture_regs(struct amdgpu_device *adev)
@@ -116,6 +136,27 @@ static void gfxhub_v1_0_init_system_aperture_regs(struct amdgpu_device *adev) · 44 unmodified lines
116136 WREG32_FIELD15(GC, 0, VM_L2_PROTECTION_FAULT_CNTL2,
117137 ACTIVE_PAGE_MIGRATION_PTE_READ_RETRY, 1);
118138 }
139+
140+ /* In the case squeezing vram into GART aperture, we don't use
141+ * FB aperture and AGP aperture. Disable them.
142+ */
143+ if (adev->gmc.pdb0_bo) {
144+ if (adev->asic_type == CHIP_ALDEBARAN) {
145+ WREG32_SOC15(GC, 0, mmMC_VM_FB_LOCATION_TOP, adev->gmc.fb_end_original >> 24);
146+ WREG32_SOC15(GC, 0, mmMC_VM_FB_LOCATION_BASE, adev->gmc.fb_start_original >> 24);
147+ WREG32_SOC15(GC, 0, mmMC_VM_AGP_TOP, 0);
148+ WREG32_SOC15(GC, 0, mmMC_VM_AGP_BOT, 0xFFFFFF);
149+ WREG32_SOC15(GC, 0, mmMC_VM_SYSTEM_APERTURE_LOW_ADDR, adev->gmc.fb_start_original >> 18);
150+ WREG32_SOC15(GC, 0, mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR, adev->gmc.fb_end_original >> 18);
151+ } else {
152+ WREG32_SOC15(GC, 0, mmMC_VM_FB_LOCATION_TOP, 0);
153+ WREG32_SOC15(GC, 0, mmMC_VM_FB_LOCATION_BASE, 0x00FFFFFF);
154+ WREG32_SOC15(GC, 0, mmMC_VM_AGP_TOP, 0);
155+ WREG32_SOC15(GC, 0, mmMC_VM_AGP_BOT, 0xFFFFFF);
156+ WREG32_SOC15(GC, 0, mmMC_VM_SYSTEM_APERTURE_LOW_ADDR, 0x3FFFFFFF);
157+ WREG32_SOC15(GC, 0, mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR, 0);
158+ }
159+ }
119160 }
120161
121162 static void gfxhub_v1_0_init_tlb_regs(struct amdgpu_device *adev)
@@ -173,8 +214,13 @@ static void gfxhub_v1_0_init_cache_regs(struct amdgpu_device *adev) · 51 unmodified lines
173214 WREG32_SOC15_RLC(GC, 0, mmVM_L2_CNTL3, tmp);
174215
175216 tmp = mmVM_L2_CNTL4_DEFAULT;
176− tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
177− tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
217+ if (adev->gmc.xgmi.connected_to_cpu) {
218+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 1);
219+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 1);
220+ } else {
221+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
222+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
223+ }
178224 WREG32_SOC15_RLC(GC, 0, mmVM_L2_CNTL4, tmp);
179225 }
180226
@@ -184,7 +230,10 @@ static void gfxhub_v1_0_enable_system_domain(struct amdgpu_device *adev) · 3 unmodified lines
184230
185231 tmp = RREG32_SOC15(GC, 0, mmVM_CONTEXT0_CNTL);
186232 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
187− tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
233+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH,
234+ adev->gmc.vmid0_page_table_depth);
235+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_BLOCK_SIZE,
236+ adev->gmc.vmid0_page_table_block_size);
188237 tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL,
189238 RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
190239 WREG32_SOC15(GC, 0, mmVM_CONTEXT0_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_1.c+45−9
@@ -28,13 +28,42 @@ · 27 unmodified lines
2828
2929 #include "soc15_common.h"
3030
31+#define mmMC_VM_XGMI_LFB_CNTL_ALDE 0x0978
32+#define mmMC_VM_XGMI_LFB_CNTL_ALDE_BASE_IDX 0
33+#define mmMC_VM_XGMI_LFB_SIZE_ALDE 0x0979
34+#define mmMC_VM_XGMI_LFB_SIZE_ALDE_BASE_IDX 0
35+//MC_VM_XGMI_LFB_CNTL
36+#define MC_VM_XGMI_LFB_CNTL_ALDE__PF_LFB_REGION__SHIFT 0x0
37+#define MC_VM_XGMI_LFB_CNTL_ALDE__PF_MAX_REGION__SHIFT 0x4
38+#define MC_VM_XGMI_LFB_CNTL_ALDE__PF_LFB_REGION_MASK 0x0000000FL
39+#define MC_VM_XGMI_LFB_CNTL_ALDE__PF_MAX_REGION_MASK 0x000000F0L
40+//MC_VM_XGMI_LFB_SIZE
41+#define MC_VM_XGMI_LFB_SIZE_ALDE__PF_LFB_SIZE__SHIFT 0x0
42+#define MC_VM_XGMI_LFB_SIZE_ALDE__PF_LFB_SIZE_MASK 0x0001FFFFL
43+
3144 int gfxhub_v1_1_get_xgmi_info(struct amdgpu_device *adev)
3245 {
33− u32 xgmi_lfb_cntl = RREG32_SOC15(GC, 0, mmMC_VM_XGMI_LFB_CNTL);
34− u32 max_region =
46+ u32 max_num_physical_nodes;
47+ u32 max_physical_node_id;
48+ u32 xgmi_lfb_cntl;
49+ u32 max_region;
50+ u64 seg_size;
51+
52+ if (adev->asic_type == CHIP_ALDEBARAN) {
53+ xgmi_lfb_cntl = RREG32_SOC15(GC, 0, mmMC_VM_XGMI_LFB_CNTL_ALDE);
54+ seg_size = REG_GET_FIELD(
55+ RREG32_SOC15(GC, 0, mmMC_VM_XGMI_LFB_SIZE_ALDE),
56+ MC_VM_XGMI_LFB_SIZE, PF_LFB_SIZE) << 24;
57+ } else {
58+ xgmi_lfb_cntl = RREG32_SOC15(GC, 0, mmMC_VM_XGMI_LFB_CNTL);
59+ seg_size = REG_GET_FIELD(
60+ RREG32_SOC15(GC, 0, mmMC_VM_XGMI_LFB_SIZE),
61+ MC_VM_XGMI_LFB_SIZE, PF_LFB_SIZE) << 24;
62+ }
63+
64+ max_region =
3565 REG_GET_FIELD(xgmi_lfb_cntl, MC_VM_XGMI_LFB_CNTL, PF_MAX_REGION);
36− u32 max_num_physical_nodes = 0;
37− u32 max_physical_node_id = 0;
66+
3867
3968 switch (adev->asic_type) {
4069 case CHIP_VEGA20:
@@ -45,23 +74,30 @@ int gfxhub_v1_1_get_xgmi_info(struct amdgpu_device *adev) · 4 unmodified lines
4574 max_num_physical_nodes = 8;
4675 max_physical_node_id = 7;
4776 break;
77+ case CHIP_ALDEBARAN:
78+ /* just using duplicates for Aldebaran support, revisit later */
79+ max_num_physical_nodes = 8;
80+ max_physical_node_id = 7;
81+ break;
4882 default:
4983 return -EINVAL;
5084 }
5185
5286 /* PF_MAX_REGION=0 means xgmi is disabled */
53− if (max_region) {
87+ if (max_region || adev->gmc.xgmi.connected_to_cpu) {
5488 adev->gmc.xgmi.num_physical_nodes = max_region + 1;
89+
5590 if (adev->gmc.xgmi.num_physical_nodes > max_num_physical_nodes)
5691 return -EINVAL;
5792
5893 adev->gmc.xgmi.physical_node_id =
59− REG_GET_FIELD(xgmi_lfb_cntl, MC_VM_XGMI_LFB_CNTL, PF_LFB_REGION);
94+ REG_GET_FIELD(xgmi_lfb_cntl, MC_VM_XGMI_LFB_CNTL,
95+ PF_LFB_REGION);
96+
6097 if (adev->gmc.xgmi.physical_node_id > max_physical_node_id)
6198 return -EINVAL;
62− adev->gmc.xgmi.node_segment_size = REG_GET_FIELD(
63− RREG32_SOC15(GC, 0, mmMC_VM_XGMI_LFB_SIZE),
64− MC_VM_XGMI_LFB_SIZE, PF_LFB_SIZE) << 24;
99+
100+ adev->gmc.xgmi.node_segment_size = seg_size;
65101 }
66102
67103 return 0;
drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c+4−3
@@ -113,7 +113,7 @@ static int gmc_v10_0_process_interrupt(struct amdgpu_device *adev, · 112 unmodified lines
113113 /* Delegate it to a different ring if the hardware hasn't
114114 * already done it.
115115 */
116− if (in_interrupt()) {
116+ if (entry->ih == &adev->irq.ih) {
117117 amdgpu_irq_delegate(adev, entry, 8);
118118 return 1;
119119 }
@@ -152,8 +152,9 @@ static int gmc_v10_0_process_interrupt(struct amdgpu_device *adev, · 32 unmodified lines
152152 entry->src_id, entry->ring_id, entry->vmid,
153153 entry->pasid, task_info.process_name, task_info.tgid,
154154 task_info.task_name, task_info.pid);
155− dev_err(adev->dev, " in page starting at address 0x%012llx from client %d\n",
156− addr, entry->client_id);
155+ dev_err(adev->dev, " in page starting at address 0x%016llx from client 0x%x (%s)\n",
156+ addr, entry->client_id,
157+ soc15_ih_clientid_name[entry->client_id]);
157158
158159 if (!amdgpu_sriov_vf(adev))
159160 hub->vmhub_funcs->print_l2_protection_fault_status(adev,
drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c+1−0
@@ -346,6 +346,7 @@ static int gmc_v6_0_mc_init(struct amdgpu_device *adev) · 345 unmodified lines
346346 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
347347 }
348348
349+ adev->gmc.gart_size += adev->pm.smu_prv_buffer_size;
349350 gmc_v6_0_vram_gtt_location(adev, &adev->gmc);
350351
351352 return 0;
drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c+1−0
@@ -414,6 +414,7 @@ static int gmc_v7_0_mc_init(struct amdgpu_device *adev) · 413 unmodified lines
414414 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
415415 }
416416
417+ adev->gmc.gart_size += adev->pm.smu_prv_buffer_size;
417418 gmc_v7_0_vram_gtt_location(adev, &adev->gmc);
418419
419420 return 0;
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c+1−0
@@ -599,6 +599,7 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev) · 598 unmodified lines
599599 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
600600 }
601601
602+ adev->gmc.gart_size += adev->pm.smu_prv_buffer_size;
602603 gmc_v8_0_vram_gtt_location(adev, &adev->gmc);
603604
604605 return 0;
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c+137−15
@@ -50,6 +50,7 @@ · 49 unmodified lines
5050 #include "athub_v1_0.h"
5151 #include "gfxhub_v1_1.h"
5252 #include "mmhub_v9_4.h"
53+#include "mmhub_v1_7.h"
5354 #include "umc_v6_1.h"
5455 #include "umc_v6_0.h"
5556
@@ -279,6 +280,47 @@ static const char *mmhub_client_ids_arcturus[][2] = { · 223 unmodified lines
279280 [384][1] = "OSS",
280281 };
281282
283+static const char *mmhub_client_ids_aldebaran[][2] = {
284+ [2][0] = "MP1",
285+ [3][0] = "MP0",
286+ [32+1][0] = "DBGU_IO0",
287+ [32+2][0] = "DBGU_IO2",
288+ [32+4][0] = "MPIO",
289+ [96+11][0] = "JPEG0",
290+ [96+12][0] = "VCN0",
291+ [96+13][0] = "VCNU0",
292+ [128+11][0] = "JPEG1",
293+ [128+12][0] = "VCN1",
294+ [128+13][0] = "VCNU1",
295+ [160+1][0] = "XDP",
296+ [160+14][0] = "HDP",
297+ [256+0][0] = "SDMA0",
298+ [256+1][0] = "SDMA1",
299+ [256+2][0] = "SDMA2",
300+ [256+3][0] = "SDMA3",
301+ [256+4][0] = "SDMA4",
302+ [384+0][0] = "OSS",
303+ [2][1] = "MP1",
304+ [3][1] = "MP0",
305+ [32+1][1] = "DBGU_IO0",
306+ [32+2][1] = "DBGU_IO2",
307+ [32+4][1] = "MPIO",
308+ [96+11][1] = "JPEG0",
309+ [96+12][1] = "VCN0",
310+ [96+13][1] = "VCNU0",
311+ [128+11][1] = "JPEG1",
312+ [128+12][1] = "VCN1",
313+ [128+13][1] = "VCNU1",
314+ [160+1][1] = "XDP",
315+ [160+14][1] = "HDP",
316+ [256+0][1] = "SDMA0",
317+ [256+1][1] = "SDMA1",
318+ [256+2][1] = "SDMA2",
319+ [256+3][1] = "SDMA3",
320+ [256+4][1] = "SDMA4",
321+ [384+0][1] = "OSS",
322+};
323+
282324 static const struct soc15_reg_golden golden_settings_mmhub_1_0_0[] =
283325 {
284326 SOC15_REG_GOLDEN_VALUE(MMHUB, 0, mmDAGB1_WRCLI2, 0x00000007, 0xfe5fe0fa),
@@ -484,7 +526,7 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev, · 199 unmodified lines
484526 /* Delegate it to a different ring if the hardware hasn't
485527 * already done it.
486528 */
487− if (in_interrupt()) {
529+ if (entry->ih == &adev->irq.ih) {
488530 amdgpu_irq_delegate(adev, entry, 8);
489531 return 1;
490532 }
@@ -520,8 +562,9 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev, · 29 unmodified lines
520562 entry->src_id, entry->ring_id, entry->vmid,
521563 entry->pasid, task_info.process_name, task_info.tgid,
522564 task_info.task_name, task_info.pid);
523− dev_err(adev->dev, " in page starting at address 0x%012llx from client %d\n",
524− addr, entry->client_id);
565+ dev_err(adev->dev, " in page starting at address 0x%016llx from IH client 0x%x (%s)\n",
566+ addr, entry->client_id,
567+ soc15_ih_clientid_name[entry->client_id]);
525568
526569 if (amdgpu_sriov_vf(adev))
527570 return 0;
@@ -568,6 +611,9 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev, · 40 unmodified lines
568611 case CHIP_RENOIR:
569612 mmhub_cid = mmhub_client_ids_renoir[cid][rw];
570613 break;
614+ case CHIP_ALDEBARAN:
615+ mmhub_cid = mmhub_client_ids_aldebaran[cid][rw];
616+ break;
571617 default:
572618 mmhub_cid = NULL;
573619 break;
@@ -642,6 +688,9 @@ static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid, · 68 unmodified lines
642688 static bool gmc_v9_0_use_invalidate_semaphore(struct amdgpu_device *adev,
643689 uint32_t vmhub)
644690 {
691+ if (adev->asic_type == CHIP_ALDEBARAN)
692+ return false;
693+
645694 return ((vmhub == AMDGPU_MMHUB_0 ||
646695 vmhub == AMDGPU_MMHUB_1) &&
647696 (!amdgpu_sriov_vf(adev)) &&
@@ -1033,10 +1082,14 @@ static void gmc_v9_0_get_vm_pte(struct amdgpu_device *adev, · 385 unmodified lines
10331082 *flags &= ~AMDGPU_PTE_VALID;
10341083 }
10351084
1036− if (adev->asic_type == CHIP_ARCTURUS &&
1085+ if ((adev->asic_type == CHIP_ARCTURUS ||
1086+ adev->asic_type == CHIP_ALDEBARAN) &&
10371087 !(*flags & AMDGPU_PTE_SYSTEM) &&
10381088 mapping->bo_va->is_xgmi)
10391089 *flags |= AMDGPU_PTE_SNOOPED;
1090+
1091+ if (adev->asic_type == CHIP_ALDEBARAN)
1092+ *flags |= mapping->flags & AMDGPU_PTE_SNOOPED;
10401093 }
10411094
10421095 static unsigned gmc_v9_0_get_vbios_fb_size(struct amdgpu_device *adev)
@@ -1123,6 +1176,9 @@ static void gmc_v9_0_set_mmhub_funcs(struct amdgpu_device *adev) · 80 unmodified lines
11231176 case CHIP_ARCTURUS:
11241177 adev->mmhub.funcs = &mmhub_v9_4_funcs;
11251178 break;
1179+ case CHIP_ALDEBARAN:
1180+ adev->mmhub.funcs = &mmhub_v1_7_funcs;
1181+ break;
11261182 default:
11271183 adev->mmhub.funcs = &mmhub_v1_0_funcs;
11281184 break;
@@ -1144,6 +1200,16 @@ static int gmc_v9_0_early_init(void *handle) · 15 unmodified lines
11441200 gmc_v9_0_set_mmhub_funcs(adev);
11451201 gmc_v9_0_set_gfxhub_funcs(adev);
11461202
1203+ if (adev->asic_type == CHIP_VEGA20 ||
1204+ adev->asic_type == CHIP_ARCTURUS)
1205+ adev->gmc.xgmi.supported = true;
1206+
1207+ if (adev->asic_type == CHIP_ALDEBARAN) {
1208+ adev->gmc.xgmi.supported = true;
1209+ adev->gmc.xgmi.connected_to_cpu =
1210+ adev->smuio.funcs->is_host_gpu_xgmi_supported(adev);
1211+ }
1212+
11471213 adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
11481214 adev->gmc.shared_aperture_end =
11491215 adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
@@ -1194,9 +1260,13 @@ static void gmc_v9_0_vram_gtt_location(struct amdgpu_device *adev, · 44 unmodified lines
11941260
11951261 /* add the xgmi offset of the physical node */
11961262 base += adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
1197− amdgpu_gmc_vram_location(adev, mc, base);
1198− amdgpu_gmc_gart_location(adev, mc);
1199− amdgpu_gmc_agp_location(adev, mc);
1263+ if (adev->gmc.xgmi.connected_to_cpu) {
1264+ amdgpu_gmc_sysvm_location(adev, mc);
1265+ } else {
1266+ amdgpu_gmc_vram_location(adev, mc, base);
1267+ amdgpu_gmc_gart_location(adev, mc);
1268+ amdgpu_gmc_agp_location(adev, mc);
1269+ }
12001270 /* base offset of vram pages */
12011271 adev->vm_manager.vram_base_offset = adev->gfxhub.funcs->get_mc_fb_offset(adev);
12021272
@@ -1223,7 +1293,8 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev) · 20 unmodified lines
12231293 adev->nbio.funcs->get_memsize(adev) * 1024ULL * 1024ULL;
12241294 adev->gmc.real_vram_size = adev->gmc.mc_vram_size;
12251295
1226− if (!(adev->flags & AMD_IS_APU)) {
1296+ if (!(adev->flags & AMD_IS_APU) &&
1297+ !adev->gmc.xgmi.connected_to_cpu) {
12271298 r = amdgpu_device_resize_fb_bar(adev);
12281299 if (r)
12291300 return r;
@@ -1232,10 +1303,28 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev) · 2 unmodified lines
12321303 adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
12331304
12341305 #ifdef CONFIG_X86_64
1235− if (adev->flags & AMD_IS_APU) {
1236− adev->gmc.aper_base = adev->gfxhub.funcs->get_mc_fb_offset(adev);
1306+ /*
1307+ * AMD Accelerated Processing Platform (APP) supporting GPU-HOST xgmi
1308+ * interface can use VRAM through here as it appears system reserved
1309+ * memory in host address space.
1310+ *
1311+ * For APUs, VRAM is just the stolen system memory and can be accessed
1312+ * directly.
1313+ *
1314+ * Otherwise, use the legacy Host Data Path (HDP) through PCIe BAR.
1315+ */
1316+
1317+ /* check whether both host-gpu and gpu-gpu xgmi links exist */
1318+ if ((adev->flags & AMD_IS_APU) ||
1319+ (adev->gmc.xgmi.supported &&
1320+ adev->gmc.xgmi.connected_to_cpu)) {
1321+ adev->gmc.aper_base =
1322+ adev->gfxhub.funcs->get_mc_fb_offset(adev) +
1323+ adev->gmc.xgmi.physical_node_id *
1324+ adev->gmc.xgmi.node_segment_size;
12371325 adev->gmc.aper_size = adev->gmc.real_vram_size;
12381326 }
1327+
12391328 #endif
12401329 /* In case the PCI BAR is larger than the actual amount of vram */
12411330 adev->gmc.visible_vram_size = adev->gmc.aper_size;
@@ -1249,6 +1338,7 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev) · 7 unmodified lines
12491338 case CHIP_VEGA12: /* all engines support GPUVM */
12501339 case CHIP_VEGA20:
12511340 case CHIP_ARCTURUS:
1341+ case CHIP_ALDEBARAN:
12521342 default:
12531343 adev->gmc.gart_size = 512ULL << 20;
12541344 break;
@@ -1261,6 +1351,8 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev) · 6 unmodified lines
12611351 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
12621352 }
12631353
1354+ adev->gmc.gart_size += adev->pm.smu_prv_buffer_size;
1355+
12641356 gmc_v9_0_vram_gtt_location(adev, &adev->gmc);
12651357
12661358 return 0;
@@ -1274,6 +1366,15 @@ static int gmc_v9_0_gart_init(struct amdgpu_device *adev) · 7 unmodified lines
12741366 WARN(1, "VEGA10 PCIE GART already initialized\n");
12751367 return 0;
12761368 }
1369+
1370+ if (adev->gmc.xgmi.connected_to_cpu) {
1371+ adev->gmc.vmid0_page_table_depth = 1;
1372+ adev->gmc.vmid0_page_table_block_size = 12;
1373+ } else {
1374+ adev->gmc.vmid0_page_table_depth = 0;
1375+ adev->gmc.vmid0_page_table_block_size = 0;
1376+ }
1377+
12771378 /* Initialize common gart structure */
12781379 r = amdgpu_gart_init(adev);
12791380 if (r)
@@ -1281,7 +1382,16 @@ static int gmc_v9_0_gart_init(struct amdgpu_device *adev) · 1 unmodified line
12811382 adev->gart.table_size = adev->gart.num_gpu_pages * 8;
12821383 adev->gart.gart_pte_flags = AMDGPU_PTE_MTYPE_VG10(MTYPE_UC) |
12831384 AMDGPU_PTE_EXECUTABLE;
1284− return amdgpu_gart_table_vram_alloc(adev);
1385+
1386+ r = amdgpu_gart_table_vram_alloc(adev);
1387+ if (r)
1388+ return r;
1389+
1390+ if (adev->gmc.xgmi.connected_to_cpu) {
1391+ r = amdgpu_gmc_pdb0_alloc(adev);
1392+ }
1393+
1394+ return r;
12851395 }
12861396
12871397 /**
@@ -1352,6 +1462,7 @@ static int gmc_v9_0_sw_init(void *handle) · 64 unmodified lines
13521462 case CHIP_VEGA12:
13531463 case CHIP_VEGA20:
13541464 case CHIP_RENOIR:
1465+ case CHIP_ALDEBARAN:
13551466 adev->num_vmhubs = 2;
13561467
13571468
@@ -1448,7 +1559,8 @@ static int gmc_v9_0_sw_init(void *handle) · 90 unmodified lines
14481559 * for video processing.
14491560 */
14501561 adev->vm_manager.first_kfd_vmid =
1451− adev->asic_type == CHIP_ARCTURUS ? 3 : 8;
1562+ (adev->asic_type == CHIP_ARCTURUS ||
1563+ adev->asic_type == CHIP_ALDEBARAN) ? 3 : 8;
14521564
14531565 amdgpu_vm_manager_init(adev);
14541566
@@ -1465,6 +1577,7 @@ static int gmc_v9_0_sw_fini(void *handle) · 10 unmodified lines
14651577 amdgpu_gem_force_release(adev);
14661578 amdgpu_vm_manager_fini(adev);
14671579 amdgpu_gart_table_vram_free(adev);
1580+ amdgpu_bo_unref(&adev->gmc.pdb0_bo);
14681581 amdgpu_bo_fini(adev);
14691582 amdgpu_gart_fini(adev);
14701583
@@ -1525,10 +1638,14 @@ static int gmc_v9_0_gart_enable(struct amdgpu_device *adev) · 54 unmodified lines
15251638 {
15261639 int r;
15271640
1641+ if (adev->gmc.xgmi.connected_to_cpu)
1642+ amdgpu_gmc_init_pdb0(adev);
1643+
15281644 if (adev->gart.bo == NULL) {
15291645 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
15301646 return -EINVAL;
15311647 }
1648+
15321649 r = amdgpu_gart_table_vram_pin(adev);
15331650 if (r)
15341651 return r;
@@ -1541,9 +1658,14 @@ static int gmc_v9_0_gart_enable(struct amdgpu_device *adev) · 6 unmodified lines
15411658 if (r)
15421659 return r;
15431660
1544− DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
1545− (unsigned)(adev->gmc.gart_size >> 20),
1546− (unsigned long long)amdgpu_bo_gpu_offset(adev->gart.bo));
1661+ DRM_INFO("PCIE GART of %uM enabled.\n",
1662+ (unsigned)(adev->gmc.gart_size >> 20));
1663+ if (adev->gmc.pdb0_bo)
1664+ DRM_INFO("PDB0 located at 0x%016llX\n",
1665+ (unsigned long long)amdgpu_bo_gpu_offset(adev->gmc.pdb0_bo));
1666+ DRM_INFO("PTB located at 0x%016llX\n",
1667+ (unsigned long long)amdgpu_bo_gpu_offset(adev->gart.bo));
1668+
15471669 adev->gart.ready = true;
15481670 return 0;
15491671 }
drivers/gpu/drm/amd/amdgpu/hdp_v4_0.c+3−0
@@ -49,6 +49,9 @@ static void hdp_v4_0_flush_hdp(struct amdgpu_device *adev, · 48 unmodified lines
4949 static void hdp_v4_0_invalidate_hdp(struct amdgpu_device *adev,
5050 struct amdgpu_ring *ring)
5151 {
52+ if (adev->asic_type == CHIP_ALDEBARAN)
53+ return;
54+
5255 if (!ring || !ring->funcs->emit_wreg)
5356 WREG32_SOC15_NO_KIQ(HDP, 0, mmHDP_READ_CACHE_INVALIDATE, 1);
5457 else
drivers/gpu/drm/amd/amdgpu/jpeg_v2_5.c+63−2
@@ -565,6 +565,26 @@ static const struct amd_ip_funcs jpeg_v2_5_ip_funcs = { · 564 unmodified lines
565565 .set_powergating_state = jpeg_v2_5_set_powergating_state,
566566 };
567567
568+static const struct amd_ip_funcs jpeg_v2_6_ip_funcs = {
569+ .name = "jpeg_v2_6",
570+ .early_init = jpeg_v2_5_early_init,
571+ .late_init = NULL,
572+ .sw_init = jpeg_v2_5_sw_init,
573+ .sw_fini = jpeg_v2_5_sw_fini,
574+ .hw_init = jpeg_v2_5_hw_init,
575+ .hw_fini = jpeg_v2_5_hw_fini,
576+ .suspend = jpeg_v2_5_suspend,
577+ .resume = jpeg_v2_5_resume,
578+ .is_idle = jpeg_v2_5_is_idle,
579+ .wait_for_idle = jpeg_v2_5_wait_for_idle,
580+ .check_soft_reset = NULL,
581+ .pre_soft_reset = NULL,
582+ .soft_reset = NULL,
583+ .post_soft_reset = NULL,
584+ .set_clockgating_state = jpeg_v2_5_set_clockgating_state,
585+ .set_powergating_state = jpeg_v2_5_set_powergating_state,
586+};
587+
568588 static const struct amdgpu_ring_funcs jpeg_v2_5_dec_ring_vm_funcs = {
569589 .type = AMDGPU_RING_TYPE_VCN_JPEG,
570590 .align_mask = 0xf,
@@ -595,6 +615,36 @@ static const struct amdgpu_ring_funcs jpeg_v2_5_dec_ring_vm_funcs = { · 24 unmodified lines
595615 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
596616 };
597617
618+static const struct amdgpu_ring_funcs jpeg_v2_6_dec_ring_vm_funcs = {
619+ .type = AMDGPU_RING_TYPE_VCN_JPEG,
620+ .align_mask = 0xf,
621+ .vmhub = AMDGPU_MMHUB_0,
622+ .get_rptr = jpeg_v2_5_dec_ring_get_rptr,
623+ .get_wptr = jpeg_v2_5_dec_ring_get_wptr,
624+ .set_wptr = jpeg_v2_5_dec_ring_set_wptr,
625+ .emit_frame_size =
626+ SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
627+ SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
628+ 8 + /* jpeg_v2_5_dec_ring_emit_vm_flush */
629+ 18 + 18 + /* jpeg_v2_5_dec_ring_emit_fence x2 vm fence */
630+ 8 + 16,
631+ .emit_ib_size = 22, /* jpeg_v2_5_dec_ring_emit_ib */
632+ .emit_ib = jpeg_v2_0_dec_ring_emit_ib,
633+ .emit_fence = jpeg_v2_0_dec_ring_emit_fence,
634+ .emit_vm_flush = jpeg_v2_0_dec_ring_emit_vm_flush,
635+ .test_ring = amdgpu_jpeg_dec_ring_test_ring,
636+ .test_ib = amdgpu_jpeg_dec_ring_test_ib,
637+ .insert_nop = jpeg_v2_0_dec_ring_nop,
638+ .insert_start = jpeg_v2_0_dec_ring_insert_start,
639+ .insert_end = jpeg_v2_0_dec_ring_insert_end,
640+ .pad_ib = amdgpu_ring_generic_pad_ib,
641+ .begin_use = amdgpu_jpeg_ring_begin_use,
642+ .end_use = amdgpu_jpeg_ring_end_use,
643+ .emit_wreg = jpeg_v2_0_dec_ring_emit_wreg,
644+ .emit_reg_wait = jpeg_v2_0_dec_ring_emit_reg_wait,
645+ .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
646+};
647+
598648 static void jpeg_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
599649 {
600650 int i;
@@ -602,8 +652,10 @@ static void jpeg_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev) · 1 unmodified line
602652 for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
603653 if (adev->jpeg.harvest_config & (1 << i))
604654 continue;
605−
606− adev->jpeg.inst[i].ring_dec.funcs = &jpeg_v2_5_dec_ring_vm_funcs;
655+ if (adev->asic_type == CHIP_ARCTURUS)
656+ adev->jpeg.inst[i].ring_dec.funcs = &jpeg_v2_5_dec_ring_vm_funcs;
657+ else /* CHIP_ALDEBARAN */
658+ adev->jpeg.inst[i].ring_dec.funcs = &jpeg_v2_6_dec_ring_vm_funcs;
607659 adev->jpeg.inst[i].ring_dec.me = i;
608660 DRM_INFO("JPEG(%d) JPEG decode is enabled in VM mode\n", i);
609661 }
@@ -635,3 +687,12 @@ const struct amdgpu_ip_block_version jpeg_v2_5_ip_block = · 25 unmodified lines
635687 .rev = 0,
636688 .funcs = &jpeg_v2_5_ip_funcs,
637689 };
690+
691+const struct amdgpu_ip_block_version jpeg_v2_6_ip_block =
692+{
693+ .type = AMD_IP_BLOCK_TYPE_JPEG,
694+ .major = 2,
695+ .minor = 6,
696+ .rev = 0,
697+ .funcs = &jpeg_v2_6_ip_funcs,
698+};
drivers/gpu/drm/amd/amdgpu/jpeg_v2_5.h+1−0
@@ -25,5 +25,6 @@ · 24 unmodified lines
2525 #define __JPEG_V2_5_H__
2626
2727 extern const struct amdgpu_ip_block_version jpeg_v2_5_ip_block;
28+extern const struct amdgpu_ip_block_version jpeg_v2_6_ip_block;
2829
2930 #endif /* __JPEG_V2_5_H__ */
drivers/gpu/drm/amd/amdgpu/mmhub_v1_7.c+1329−0
@@ -0,0 +1,1329 @@
1+/*
2+ * Copyright 2019 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#include "amdgpu.h"
24+#include "amdgpu_ras.h"
25+#include "mmhub_v1_7.h"
26+
27+#include "mmhub/mmhub_1_7_offset.h"
28+#include "mmhub/mmhub_1_7_sh_mask.h"
29+#include "vega10_enum.h"
30+
31+#include "soc15_common.h"
32+#include "soc15.h"
33+
34+#define regVM_L2_CNTL3_DEFAULT 0x80100007
35+#define regVM_L2_CNTL4_DEFAULT 0x000000c1
36+
37+static u64 mmhub_v1_7_get_fb_location(struct amdgpu_device *adev)
38+{
39+ u64 base = RREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_BASE);
40+ u64 top = RREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_TOP);
41+
42+ base &= MC_VM_FB_LOCATION_BASE__FB_BASE_MASK;
43+ base <<= 24;
44+
45+ top &= MC_VM_FB_LOCATION_TOP__FB_TOP_MASK;
46+ top <<= 24;
47+
48+ adev->gmc.fb_start = base;
49+ adev->gmc.fb_end = top;
50+ adev->gmc.fb_start_original = base;
51+ adev->gmc.fb_end_original = top;
52+
53+ return base;
54+}
55+
56+static void mmhub_v1_7_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid,
57+ uint64_t page_table_base)
58+{
59+ struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB_0];
60+
61+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32,
62+ hub->ctx_addr_distance * vmid, lower_32_bits(page_table_base));
63+
64+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32,
65+ hub->ctx_addr_distance * vmid, upper_32_bits(page_table_base));
66+}
67+
68+static void mmhub_v1_7_init_gart_aperture_regs(struct amdgpu_device *adev)
69+{
70+ uint64_t pt_base;
71+
72+ if (adev->gmc.pdb0_bo)
73+ pt_base = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo);
74+ else
75+ pt_base = amdgpu_gmc_pd_addr(adev->gart.bo);
76+
77+ mmhub_v1_7_setup_vm_pt_regs(adev, 0, pt_base);
78+
79+ /* If use GART for FB translation, vmid0 page table covers both
80+ * vram and system memory (gart)
81+ */
82+ if (adev->gmc.pdb0_bo) {
83+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
84+ (u32)(adev->gmc.fb_start >> 12));
85+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
86+ (u32)(adev->gmc.fb_start >> 44));
87+
88+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
89+ (u32)(adev->gmc.gart_end >> 12));
90+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
91+ (u32)(adev->gmc.gart_end >> 44));
92+
93+ } else {
94+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
95+ (u32)(adev->gmc.gart_start >> 12));
96+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
97+ (u32)(adev->gmc.gart_start >> 44));
98+
99+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
100+ (u32)(adev->gmc.gart_end >> 12));
101+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
102+ (u32)(adev->gmc.gart_end >> 44));
103+ }
104+}
105+
106+static void mmhub_v1_7_init_system_aperture_regs(struct amdgpu_device *adev)
107+{
108+ uint64_t value;
109+ uint32_t tmp;
110+
111+ /* Program the AGP BAR */
112+ WREG32_SOC15(MMHUB, 0, regMC_VM_AGP_BASE, 0);
113+ WREG32_SOC15(MMHUB, 0, regMC_VM_AGP_BOT, adev->gmc.agp_start >> 24);
114+ WREG32_SOC15(MMHUB, 0, regMC_VM_AGP_TOP, adev->gmc.agp_end >> 24);
115+
116+ /* Program the system aperture low logical page number. */
117+ WREG32_SOC15(MMHUB, 0, regMC_VM_SYSTEM_APERTURE_LOW_ADDR,
118+ min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
119+
120+ WREG32_SOC15(MMHUB, 0, regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
121+ max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
122+
123+ /* In the case squeezing vram into GART aperture, we don't use
124+ * FB aperture and AGP aperture. Disable them.
125+ */
126+ if (adev->gmc.pdb0_bo) {
127+ WREG32_SOC15(MMHUB, 0, regMC_VM_AGP_BOT, 0xFFFFFF);
128+ WREG32_SOC15(MMHUB, 0, regMC_VM_AGP_TOP, 0);
129+ WREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_TOP, adev->gmc.fb_end_original >> 24);
130+ WREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_BASE, adev->gmc.fb_start_original >> 24);
131+ WREG32_SOC15(MMHUB, 0, regMC_VM_SYSTEM_APERTURE_LOW_ADDR, adev->gmc.fb_start_original >> 18);
132+ WREG32_SOC15(MMHUB, 0, regMC_VM_SYSTEM_APERTURE_HIGH_ADDR, adev->gmc.fb_end_original >> 18);
133+ }
134+ if (amdgpu_sriov_vf(adev))
135+ return;
136+
137+ /* Set default page address. */
138+ value = adev->vram_scratch.gpu_addr - adev->gmc.vram_start +
139+ adev->vm_manager.vram_base_offset;
140+ WREG32_SOC15(MMHUB, 0, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_LSB,
141+ (u32)(value >> 12));
142+ WREG32_SOC15(MMHUB, 0, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_MSB,
143+ (u32)(value >> 44));
144+
145+ /* Program "protection fault". */
146+ WREG32_SOC15(MMHUB, 0, regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_LO32,
147+ (u32)(adev->dummy_page_addr >> 12));
148+ WREG32_SOC15(MMHUB, 0, regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_HI32,
149+ (u32)((u64)adev->dummy_page_addr >> 44));
150+
151+ tmp = RREG32_SOC15(MMHUB, 0, regVM_L2_PROTECTION_FAULT_CNTL2);
152+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL2,
153+ ACTIVE_PAGE_MIGRATION_PTE_READ_RETRY, 1);
154+ WREG32_SOC15(MMHUB, 0, regVM_L2_PROTECTION_FAULT_CNTL2, tmp);
155+}
156+
157+static void mmhub_v1_7_init_tlb_regs(struct amdgpu_device *adev)
158+{
159+ uint32_t tmp;
160+
161+ /* Setup TLB control */
162+ tmp = RREG32_SOC15(MMHUB, 0, regMC_VM_MX_L1_TLB_CNTL);
163+
164+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
165+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
166+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
167+ ENABLE_ADVANCED_DRIVER_MODEL, 1);
168+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
169+ SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
170+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ECO_BITS, 0);
171+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
172+ MTYPE, MTYPE_UC);/* XXX for emulation. */
173+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ATC_EN, 1);
174+
175+ WREG32_SOC15(MMHUB, 0, regMC_VM_MX_L1_TLB_CNTL, tmp);
176+}
177+
178+static void mmhub_v1_7_init_cache_regs(struct amdgpu_device *adev)
179+{
180+ uint32_t tmp;
181+
182+ if (amdgpu_sriov_vf(adev))
183+ return;
184+
185+ /* Setup L2 cache */
186+ tmp = RREG32_SOC15(MMHUB, 0, regVM_L2_CNTL);
187+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
188+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
189+ /* XXX for emulation, Refer to closed source code.*/
190+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
191+ 0);
192+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
193+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
194+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
195+ WREG32_SOC15(MMHUB, 0, regVM_L2_CNTL, tmp);
196+
197+ tmp = RREG32_SOC15(MMHUB, 0, regVM_L2_CNTL2);
198+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
199+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
200+ WREG32_SOC15(MMHUB, 0, regVM_L2_CNTL2, tmp);
201+
202+ tmp = regVM_L2_CNTL3_DEFAULT;
203+ if (adev->gmc.translate_further) {
204+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 12);
205+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
206+ L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
207+ } else {
208+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 9);
209+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
210+ L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
211+ }
212+ WREG32_SOC15(MMHUB, 0, regVM_L2_CNTL3, tmp);
213+
214+ tmp = regVM_L2_CNTL4_DEFAULT;
215+ if (adev->gmc.xgmi.connected_to_cpu) {
216+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
217+ VMC_TAP_PDE_REQUEST_PHYSICAL, 1);
218+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
219+ VMC_TAP_PTE_REQUEST_PHYSICAL, 1);
220+ } else {
221+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
222+ VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
223+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4,
224+ VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
225+ }
226+ WREG32_SOC15(MMHUB, 0, regVM_L2_CNTL4, tmp);
227+}
228+
229+static void mmhub_v1_7_enable_system_domain(struct amdgpu_device *adev)
230+{
231+ uint32_t tmp;
232+
233+ tmp = RREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_CNTL);
234+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
235+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH,
236+ adev->gmc.vmid0_page_table_depth);
237+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_BLOCK_SIZE,
238+ adev->gmc.vmid0_page_table_block_size);
239+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL,
240+ RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
241+ WREG32_SOC15(MMHUB, 0, regVM_CONTEXT0_CNTL, tmp);
242+}
243+
244+static void mmhub_v1_7_disable_identity_aperture(struct amdgpu_device *adev)
245+{
246+ if (amdgpu_sriov_vf(adev))
247+ return;
248+
249+ WREG32_SOC15(MMHUB, 0, regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_LO32,
250+ 0XFFFFFFFF);
251+ WREG32_SOC15(MMHUB, 0, regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_HI32,
252+ 0x0000000F);
253+
254+ WREG32_SOC15(MMHUB, 0,
255+ regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_LO32, 0);
256+ WREG32_SOC15(MMHUB, 0,
257+ regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_HI32, 0);
258+
259+ WREG32_SOC15(MMHUB, 0, regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_LO32,
260+ 0);
261+ WREG32_SOC15(MMHUB, 0, regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_HI32,
262+ 0);
263+}
264+
265+static void mmhub_v1_7_setup_vmid_config(struct amdgpu_device *adev)
266+{
267+ struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB_0];
268+ unsigned num_level, block_size;
269+ uint32_t tmp;
270+ int i;
271+
272+ num_level = adev->vm_manager.num_level;
273+ block_size = adev->vm_manager.block_size;
274+ if (adev->gmc.translate_further)
275+ num_level -= 1;
276+ else
277+ block_size -= 9;
278+
279+ for (i = 0; i <= 14; i++) {
280+ tmp = RREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT1_CNTL, i);
281+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
282+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH,
283+ num_level);
284+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
285+ RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
286+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
287+ DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT,
288+ 1);
289+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
290+ PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
291+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
292+ VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
293+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
294+ READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
295+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
296+ WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
297+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
298+ EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
299+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
300+ PAGE_TABLE_BLOCK_SIZE,
301+ block_size);
302+ /* Send no-retry XNACK on fault to suppress VM fault storm. */
303+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
304+ RETRY_PERMISSION_OR_INVALID_PAGE_FAULT,
305+ !adev->gmc.noretry);
306+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT1_CNTL,
307+ i * hub->ctx_distance, tmp);
308+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32,
309+ i * hub->ctx_addr_distance, 0);
310+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32,
311+ i * hub->ctx_addr_distance, 0);
312+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT1_PAGE_TABLE_END_ADDR_LO32,
313+ i * hub->ctx_addr_distance,
314+ lower_32_bits(adev->vm_manager.max_pfn - 1));
315+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT1_PAGE_TABLE_END_ADDR_HI32,
316+ i * hub->ctx_addr_distance,
317+ upper_32_bits(adev->vm_manager.max_pfn - 1));
318+ }
319+}
320+
321+static void mmhub_v1_7_program_invalidation(struct amdgpu_device *adev)
322+{
323+ struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB_0];
324+ unsigned i;
325+
326+ for (i = 0; i < 18; ++i) {
327+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32,
328+ i * hub->eng_addr_distance, 0xffffffff);
329+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_INVALIDATE_ENG0_ADDR_RANGE_HI32,
330+ i * hub->eng_addr_distance, 0x1f);
331+ }
332+}
333+
334+static int mmhub_v1_7_gart_enable(struct amdgpu_device *adev)
335+{
336+ if (amdgpu_sriov_vf(adev)) {
337+ /*
338+ * MC_VM_FB_LOCATION_BASE/TOP is NULL for VF, becuase they are
339+ * VF copy registers so vbios post doesn't program them, for
340+ * SRIOV driver need to program them
341+ */
342+ WREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_BASE,
343+ adev->gmc.vram_start >> 24);
344+ WREG32_SOC15(MMHUB, 0, regMC_VM_FB_LOCATION_TOP,
345+ adev->gmc.vram_end >> 24);
346+ }
347+
348+ /* GART Enable. */
349+ mmhub_v1_7_init_gart_aperture_regs(adev);
350+ mmhub_v1_7_init_system_aperture_regs(adev);
351+ mmhub_v1_7_init_tlb_regs(adev);
352+ mmhub_v1_7_init_cache_regs(adev);
353+
354+ mmhub_v1_7_enable_system_domain(adev);
355+ mmhub_v1_7_disable_identity_aperture(adev);
356+ mmhub_v1_7_setup_vmid_config(adev);
357+ mmhub_v1_7_program_invalidation(adev);
358+
359+ return 0;
360+}
361+
362+static void mmhub_v1_7_gart_disable(struct amdgpu_device *adev)
363+{
364+ struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB_0];
365+ u32 tmp;
366+ u32 i;
367+
368+ /* Disable all tables */
369+ for (i = 0; i < 16; i++)
370+ WREG32_SOC15_OFFSET(MMHUB, 0, regVM_CONTEXT0_CNTL,
371+ i * hub->ctx_distance, 0);
372+
373+ /* Setup TLB control */
374+ tmp = RREG32_SOC15(MMHUB, 0, regMC_VM_MX_L1_TLB_CNTL);
375+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
376+ tmp = REG_SET_FIELD(tmp,
377+ MC_VM_MX_L1_TLB_CNTL,
378+ ENABLE_ADVANCED_DRIVER_MODEL,
379+ 0);
380+ WREG32_SOC15(MMHUB, 0, regMC_VM_MX_L1_TLB_CNTL, tmp);
381+
382+ if (!amdgpu_sriov_vf(adev)) {
383+ /* Setup L2 cache */
384+ tmp = RREG32_SOC15(MMHUB, 0, regVM_L2_CNTL);
385+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
386+ WREG32_SOC15(MMHUB, 0, regVM_L2_CNTL, tmp);
387+ WREG32_SOC15(MMHUB, 0, regVM_L2_CNTL3, 0);
388+ }
389+}
390+
391+/**
392+ * mmhub_v1_7_set_fault_enable_default - update GART/VM fault handling
393+ *
394+ * @adev: amdgpu_device pointer
395+ * @value: true redirects VM faults to the default page
396+ */
397+static void mmhub_v1_7_set_fault_enable_default(struct amdgpu_device *adev, bool value)
398+{
399+ u32 tmp;
400+
401+ if (amdgpu_sriov_vf(adev))
402+ return;
403+
404+ tmp = RREG32_SOC15(MMHUB, 0, regVM_L2_PROTECTION_FAULT_CNTL);
405+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
406+ RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
407+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
408+ PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
409+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
410+ PDE1_PROTECTION_FAULT_ENABLE_DEFAULT, value);
411+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
412+ PDE2_PROTECTION_FAULT_ENABLE_DEFAULT, value);
413+ tmp = REG_SET_FIELD(tmp,
414+ VM_L2_PROTECTION_FAULT_CNTL,
415+ TRANSLATE_FURTHER_PROTECTION_FAULT_ENABLE_DEFAULT,
416+ value);
417+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
418+ NACK_PROTECTION_FAULT_ENABLE_DEFAULT, value);
419+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
420+ DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
421+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
422+ VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
423+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
424+ READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
425+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
426+ WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
427+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
428+ EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
429+ if (!value) {
430+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
431+ CRASH_ON_NO_RETRY_FAULT, 1);
432+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
433+ CRASH_ON_RETRY_FAULT, 1);
434+ }
435+
436+ WREG32_SOC15(MMHUB, 0, regVM_L2_PROTECTION_FAULT_CNTL, tmp);
437+}
438+
439+static void mmhub_v1_7_init(struct amdgpu_device *adev)
440+{
441+ struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB_0];
442+
443+ hub->ctx0_ptb_addr_lo32 =
444+ SOC15_REG_OFFSET(MMHUB, 0,
445+ regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32);
446+ hub->ctx0_ptb_addr_hi32 =
447+ SOC15_REG_OFFSET(MMHUB, 0,
448+ regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32);
449+ hub->vm_inv_eng0_req =
450+ SOC15_REG_OFFSET(MMHUB, 0, regVM_INVALIDATE_ENG0_REQ);
451+ hub->vm_inv_eng0_ack =
452+ SOC15_REG_OFFSET(MMHUB, 0, regVM_INVALIDATE_ENG0_ACK);
453+ hub->vm_context0_cntl =
454+ SOC15_REG_OFFSET(MMHUB, 0, regVM_CONTEXT0_CNTL);
455+ hub->vm_l2_pro_fault_status =
456+ SOC15_REG_OFFSET(MMHUB, 0, regVM_L2_PROTECTION_FAULT_STATUS);
457+ hub->vm_l2_pro_fault_cntl =
458+ SOC15_REG_OFFSET(MMHUB, 0, regVM_L2_PROTECTION_FAULT_CNTL);
459+
460+ hub->ctx_distance = regVM_CONTEXT1_CNTL - regVM_CONTEXT0_CNTL;
461+ hub->ctx_addr_distance = regVM_CONTEXT1_PAGE_TABLE_BASE_ADDR_LO32 -
462+ regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32;
463+ hub->eng_distance = regVM_INVALIDATE_ENG1_REQ - regVM_INVALIDATE_ENG0_REQ;
464+ hub->eng_addr_distance = regVM_INVALIDATE_ENG1_ADDR_RANGE_LO32 -
465+ regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32;
466+
467+}
468+
469+static void mmhub_v1_7_update_medium_grain_clock_gating(struct amdgpu_device *adev,
470+ bool enable)
471+{
472+ uint32_t def, data, def1, data1, def2 = 0, data2 = 0;
473+
474+ def = data = RREG32_SOC15(MMHUB, 0, regATC_L2_MISC_CG);
475+
476+ def1 = data1 = RREG32_SOC15(MMHUB, 0, regDAGB0_CNTL_MISC2);
477+ def2 = data2 = RREG32_SOC15(MMHUB, 0, regDAGB1_CNTL_MISC2);
478+
479+ if (enable) {
480+ data |= ATC_L2_MISC_CG__ENABLE_MASK;
481+
482+ data1 &= ~(DAGB0_CNTL_MISC2__DISABLE_WRREQ_CG_MASK |
483+ DAGB0_CNTL_MISC2__DISABLE_WRRET_CG_MASK |
484+ DAGB0_CNTL_MISC2__DISABLE_RDREQ_CG_MASK |
485+ DAGB0_CNTL_MISC2__DISABLE_RDRET_CG_MASK |
486+ DAGB0_CNTL_MISC2__DISABLE_TLBWR_CG_MASK |
487+ DAGB0_CNTL_MISC2__DISABLE_TLBRD_CG_MASK);
488+
489+ data2 &= ~(DAGB1_CNTL_MISC2__DISABLE_WRREQ_CG_MASK |
490+ DAGB1_CNTL_MISC2__DISABLE_WRRET_CG_MASK |
491+ DAGB1_CNTL_MISC2__DISABLE_RDREQ_CG_MASK |
492+ DAGB1_CNTL_MISC2__DISABLE_RDRET_CG_MASK |
493+ DAGB1_CNTL_MISC2__DISABLE_TLBWR_CG_MASK |
494+ DAGB1_CNTL_MISC2__DISABLE_TLBRD_CG_MASK);
495+ } else {
496+ data &= ~ATC_L2_MISC_CG__ENABLE_MASK;
497+
498+ data1 |= (DAGB0_CNTL_MISC2__DISABLE_WRREQ_CG_MASK |
499+ DAGB0_CNTL_MISC2__DISABLE_WRRET_CG_MASK |
500+ DAGB0_CNTL_MISC2__DISABLE_RDREQ_CG_MASK |
501+ DAGB0_CNTL_MISC2__DISABLE_RDRET_CG_MASK |
502+ DAGB0_CNTL_MISC2__DISABLE_TLBWR_CG_MASK |
503+ DAGB0_CNTL_MISC2__DISABLE_TLBRD_CG_MASK);
504+
505+ data2 |= (DAGB1_CNTL_MISC2__DISABLE_WRREQ_CG_MASK |
506+ DAGB1_CNTL_MISC2__DISABLE_WRRET_CG_MASK |
507+ DAGB1_CNTL_MISC2__DISABLE_RDREQ_CG_MASK |
508+ DAGB1_CNTL_MISC2__DISABLE_RDRET_CG_MASK |
509+ DAGB1_CNTL_MISC2__DISABLE_TLBWR_CG_MASK |
510+ DAGB1_CNTL_MISC2__DISABLE_TLBRD_CG_MASK);
511+ }
512+
513+ if (def != data)
514+ WREG32_SOC15(MMHUB, 0, regATC_L2_MISC_CG, data);
515+
516+ if (def1 != data1)
517+ WREG32_SOC15(MMHUB, 0, regDAGB0_CNTL_MISC2, data1);
518+
519+ if (def2 != data2)
520+ WREG32_SOC15(MMHUB, 0, regDAGB1_CNTL_MISC2, data2);
521+}
522+
523+static void mmhub_v1_7_update_medium_grain_light_sleep(struct amdgpu_device *adev,
524+ bool enable)
525+{
526+ uint32_t def, data;
527+
528+ def = data = RREG32_SOC15(MMHUB, 0, regATC_L2_MISC_CG);
529+
530+ if (enable)
531+ data |= ATC_L2_MISC_CG__MEM_LS_ENABLE_MASK;
532+ else
533+ data &= ~ATC_L2_MISC_CG__MEM_LS_ENABLE_MASK;
534+
535+ if (def != data)
536+ WREG32_SOC15(MMHUB, 0, regATC_L2_MISC_CG, data);
537+}
538+
539+static int mmhub_v1_7_set_clockgating(struct amdgpu_device *adev,
540+ enum amd_clockgating_state state)
541+{
542+ if (amdgpu_sriov_vf(adev))
543+ return 0;
544+
545+ /* Change state only if MCCG support is enabled through driver */
546+ if (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)
547+ mmhub_v1_7_update_medium_grain_clock_gating(adev,
548+ state == AMD_CG_STATE_GATE);
549+
550+ /* Change state only if LS support is enabled through driver */
551+ if (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)
552+ mmhub_v1_7_update_medium_grain_light_sleep(adev,
553+ state == AMD_CG_STATE_GATE);
554+
555+ return 0;
556+}
557+
558+static void mmhub_v1_7_get_clockgating(struct amdgpu_device *adev, u32 *flags)
559+{
560+ int data, data1;
561+
562+ if (amdgpu_sriov_vf(adev))
563+ *flags = 0;
564+
565+ data = RREG32_SOC15(MMHUB, 0, regATC_L2_MISC_CG);
566+
567+ data1 = RREG32_SOC15(MMHUB, 0, regDAGB0_CNTL_MISC2);
568+
569+ /* AMD_CG_SUPPORT_MC_MGCG */
570+ if ((data & ATC_L2_MISC_CG__ENABLE_MASK) &&
571+ !(data1 & (DAGB0_CNTL_MISC2__DISABLE_WRREQ_CG_MASK |
572+ DAGB0_CNTL_MISC2__DISABLE_WRRET_CG_MASK |
573+ DAGB0_CNTL_MISC2__DISABLE_RDREQ_CG_MASK |
574+ DAGB0_CNTL_MISC2__DISABLE_RDRET_CG_MASK |
575+ DAGB0_CNTL_MISC2__DISABLE_TLBWR_CG_MASK |
576+ DAGB0_CNTL_MISC2__DISABLE_TLBRD_CG_MASK)))
577+ *flags |= AMD_CG_SUPPORT_MC_MGCG;
578+
579+ /* AMD_CG_SUPPORT_MC_LS */
580+ if (data & ATC_L2_MISC_CG__MEM_LS_ENABLE_MASK)
581+ *flags |= AMD_CG_SUPPORT_MC_LS;
582+}
583+
584+static const struct soc15_ras_field_entry mmhub_v1_7_ras_fields[] = {
585+ /* MMHUB Range 0 */
586+ { "MMEA0_DRAMRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
587+ SOC15_REG_FIELD(MMEA0_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
588+ SOC15_REG_FIELD(MMEA0_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT),
589+ },
590+ { "MMEA0_DRAMWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
591+ SOC15_REG_FIELD(MMEA0_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
592+ SOC15_REG_FIELD(MMEA0_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT),
593+ },
594+ { "MMEA0_DRAMWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
595+ SOC15_REG_FIELD(MMEA0_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
596+ SOC15_REG_FIELD(MMEA0_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT),
597+ },
598+ { "MMEA0_RRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
599+ SOC15_REG_FIELD(MMEA0_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
600+ SOC15_REG_FIELD(MMEA0_EDC_CNT, RRET_TAGMEM_DED_COUNT),
601+ },
602+ { "MMEA0_WRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
603+ SOC15_REG_FIELD(MMEA0_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
604+ SOC15_REG_FIELD(MMEA0_EDC_CNT, WRET_TAGMEM_DED_COUNT),
605+ },
606+ { "MMEA0_IOWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
607+ SOC15_REG_FIELD(MMEA0_EDC_CNT, IOWR_DATAMEM_SEC_COUNT),
608+ SOC15_REG_FIELD(MMEA0_EDC_CNT, IOWR_DATAMEM_DED_COUNT),
609+ },
610+ { "MMEA0_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
611+ SOC15_REG_FIELD(MMEA0_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
612+ 0, 0,
613+ },
614+ { "MMEA0_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
615+ SOC15_REG_FIELD(MMEA0_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
616+ 0, 0,
617+ },
618+ { "MMEA0_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
619+ SOC15_REG_FIELD(MMEA0_EDC_CNT, IORD_CMDMEM_SED_COUNT),
620+ 0, 0,
621+ },
622+ { "MMEA0_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT),
623+ SOC15_REG_FIELD(MMEA0_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
624+ 0, 0,
625+ },
626+ { "MMEA0_GMIRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
627+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
628+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT),
629+ },
630+ { "MMEA0_GMIWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
631+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
632+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT),
633+ },
634+ { "MMEA0_GMIWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
635+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
636+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT),
637+ },
638+ { "MMEA0_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
639+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
640+ 0, 0,
641+ },
642+ { "MMEA0_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
643+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
644+ 0, 0,
645+ },
646+ { "MMEA0_MAM_D0MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
647+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D0MEM_SED_COUNT),
648+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D0MEM_DED_COUNT),
649+ },
650+ { "MMEA0_MAM_D1MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
651+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D1MEM_SED_COUNT),
652+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D1MEM_DED_COUNT),
653+ },
654+ { "MMEA0_MAM_D2MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
655+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D2MEM_SED_COUNT),
656+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D2MEM_DED_COUNT),
657+ },
658+ { "MMEA0_MAM_D3MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2),
659+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D3MEM_SED_COUNT),
660+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D3MEM_DED_COUNT),
661+ },
662+ { "MMEA0_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT3),
663+ 0, 0,
664+ SOC15_REG_FIELD(MMEA0_EDC_CNT3, DRAMRD_PAGEMEM_DED_COUNT),
665+ },
666+ { "MMEA0_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT3),
667+ 0, 0,
668+ SOC15_REG_FIELD(MMEA0_EDC_CNT3, DRAMWR_PAGEMEM_DED_COUNT),
669+ },
670+ { "MMEA0_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT3),
671+ 0, 0,
672+ SOC15_REG_FIELD(MMEA0_EDC_CNT3, IORD_CMDMEM_DED_COUNT),
673+ },
674+ { "MMEA0_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT3),
675+ 0, 0,
676+ SOC15_REG_FIELD(MMEA0_EDC_CNT3, IOWR_CMDMEM_DED_COUNT),
677+ },
678+ { "MMEA0_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT3),
679+ 0, 0,
680+ SOC15_REG_FIELD(MMEA0_EDC_CNT3, GMIRD_PAGEMEM_DED_COUNT),
681+ },
682+ { "MMEA0_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT3),
683+ 0, 0,
684+ SOC15_REG_FIELD(MMEA0_EDC_CNT3, GMIWR_PAGEMEM_DED_COUNT),
685+ },
686+
687+ /* MMHUB Range 1 */
688+ { "MMEA1_DRAMRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
689+ SOC15_REG_FIELD(MMEA1_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
690+ SOC15_REG_FIELD(MMEA1_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT),
691+ },
692+ { "MMEA1_DRAMWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
693+ SOC15_REG_FIELD(MMEA1_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
694+ SOC15_REG_FIELD(MMEA1_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT),
695+ },
696+ { "MMEA1_DRAMWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
697+ SOC15_REG_FIELD(MMEA1_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
698+ SOC15_REG_FIELD(MMEA1_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT),
699+ },
700+ { "MMEA1_RRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
701+ SOC15_REG_FIELD(MMEA1_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
702+ SOC15_REG_FIELD(MMEA1_EDC_CNT, RRET_TAGMEM_DED_COUNT),
703+ },
704+ { "MMEA1_WRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
705+ SOC15_REG_FIELD(MMEA1_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
706+ SOC15_REG_FIELD(MMEA1_EDC_CNT, WRET_TAGMEM_DED_COUNT),
707+ },
708+ { "MMEA1_IOWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
709+ SOC15_REG_FIELD(MMEA1_EDC_CNT, IOWR_DATAMEM_SEC_COUNT),
710+ SOC15_REG_FIELD(MMEA1_EDC_CNT, IOWR_DATAMEM_DED_COUNT),
711+ },
712+ { "MMEA1_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
713+ SOC15_REG_FIELD(MMEA1_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
714+ 0, 0,
715+ },
716+ { "MMEA1_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
717+ SOC15_REG_FIELD(MMEA1_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
718+ 0, 0,
719+ },
720+ { "MMEA1_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
721+ SOC15_REG_FIELD(MMEA1_EDC_CNT, IORD_CMDMEM_SED_COUNT),
722+ 0, 0,
723+ },
724+ { "MMEA1_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT),
725+ SOC15_REG_FIELD(MMEA1_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
726+ 0, 0,
727+ },
728+ { "MMEA1_GMIRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
729+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
730+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT),
731+ },
732+ { "MMEA1_GMIWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
733+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
734+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT),
735+ },
736+ { "MMEA1_GMIWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
737+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
738+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT),
739+ },
740+ { "MMEA1_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
741+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
742+ 0, 0,
743+ },
744+ { "MMEA1_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
745+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
746+ 0, 0,
747+ },
748+ { "MMEA1_MAM_D0MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
749+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, MAM_D0MEM_SED_COUNT),
750+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, MAM_D0MEM_DED_COUNT),
751+ },
752+ { "MMEA1_MAM_D1MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
753+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, MAM_D1MEM_SED_COUNT),
754+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, MAM_D1MEM_DED_COUNT),
755+ },
756+ { "MMEA1_MAM_D2MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
757+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, MAM_D2MEM_SED_COUNT),
758+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, MAM_D2MEM_DED_COUNT),
759+ },
760+ { "MMEA1_MAM_D3MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2),
761+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, MAM_D3MEM_SED_COUNT),
762+ SOC15_REG_FIELD(MMEA1_EDC_CNT2, MAM_D3MEM_DED_COUNT),
763+ },
764+ { "MMEA1_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT3),
765+ 0, 0,
766+ SOC15_REG_FIELD(MMEA1_EDC_CNT3, DRAMRD_PAGEMEM_DED_COUNT),
767+ },
768+ { "MMEA1_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT3),
769+ 0, 0,
770+ SOC15_REG_FIELD(MMEA1_EDC_CNT3, DRAMWR_PAGEMEM_DED_COUNT),
771+ },
772+ { "MMEA1_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT3),
773+ 0, 0,
774+ SOC15_REG_FIELD(MMEA1_EDC_CNT3, IORD_CMDMEM_DED_COUNT),
775+ },
776+ { "MMEA1_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT3),
777+ 0, 0,
778+ SOC15_REG_FIELD(MMEA1_EDC_CNT3, IOWR_CMDMEM_DED_COUNT),
779+ },
780+ { "MMEA1_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT3),
781+ 0, 0,
782+ SOC15_REG_FIELD(MMEA1_EDC_CNT3, GMIRD_PAGEMEM_DED_COUNT),
783+ },
784+ { "MMEA1_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT3),
785+ 0, 0,
786+ SOC15_REG_FIELD(MMEA1_EDC_CNT3, GMIWR_PAGEMEM_DED_COUNT),
787+ },
788+
789+ /* MMHAB Range 2*/
790+ { "MMEA2_DRAMRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
791+ SOC15_REG_FIELD(MMEA2_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
792+ SOC15_REG_FIELD(MMEA2_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT),
793+ },
794+ { "MMEA2_DRAMWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
795+ SOC15_REG_FIELD(MMEA2_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
796+ SOC15_REG_FIELD(MMEA2_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT),
797+ },
798+ { "MMEA2_DRAMWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
799+ SOC15_REG_FIELD(MMEA2_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
800+ SOC15_REG_FIELD(MMEA2_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT),
801+ },
802+ { "MMEA2_RRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
803+ SOC15_REG_FIELD(MMEA2_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
804+ SOC15_REG_FIELD(MMEA2_EDC_CNT, RRET_TAGMEM_DED_COUNT),
805+ },
806+ { "MMEA2_WRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
807+ SOC15_REG_FIELD(MMEA2_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
808+ SOC15_REG_FIELD(MMEA2_EDC_CNT, WRET_TAGMEM_DED_COUNT),
809+ },
810+ { "MMEA2_IOWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
811+ SOC15_REG_FIELD(MMEA2_EDC_CNT, IOWR_DATAMEM_SEC_COUNT),
812+ SOC15_REG_FIELD(MMEA2_EDC_CNT, IOWR_DATAMEM_DED_COUNT),
813+ },
814+ { "MMEA2_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
815+ SOC15_REG_FIELD(MMEA2_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
816+ 0, 0,
817+ },
818+ { "MMEA2_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
819+ SOC15_REG_FIELD(MMEA2_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
820+ 0, 0,
821+ },
822+ { "MMEA2_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
823+ SOC15_REG_FIELD(MMEA2_EDC_CNT, IORD_CMDMEM_SED_COUNT),
824+ 0, 0,
825+ },
826+ { "MMEA2_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT),
827+ SOC15_REG_FIELD(MMEA2_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
828+ 0, 0,
829+ },
830+ { "MMEA2_GMIRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
831+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
832+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT),
833+ },
834+ { "MMEA2_GMIWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
835+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
836+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT),
837+ },
838+ { "MMEA2_GMIWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
839+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
840+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT),
841+ },
842+ { "MMEA2_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
843+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
844+ 0, 0,
845+ },
846+ { "MMEA2_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
847+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
848+ 0, 0,
849+ },
850+ { "MMEA2_MAM_D0MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
851+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D0MEM_SED_COUNT),
852+ SOC15_REG_FIELD(MMEA0_EDC_CNT2, MAM_D0MEM_DED_COUNT),
853+ },
854+ { "MMEA2_MAM_D1MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
855+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, MAM_D1MEM_SED_COUNT),
856+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, MAM_D1MEM_DED_COUNT),
857+ },
858+ { "MMEA2_MAM_D2MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
859+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, MAM_D2MEM_SED_COUNT),
860+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, MAM_D2MEM_DED_COUNT),
861+ },
862+ { "MMEA2_MAM_D3MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2),
863+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, MAM_D3MEM_SED_COUNT),
864+ SOC15_REG_FIELD(MMEA2_EDC_CNT2, MAM_D3MEM_DED_COUNT),
865+ },
866+ { "MMEA2_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT3),
867+ 0, 0,
868+ SOC15_REG_FIELD(MMEA2_EDC_CNT3, DRAMRD_PAGEMEM_DED_COUNT),
869+ },
870+ { "MMEA2_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT3),
871+ 0, 0,
872+ SOC15_REG_FIELD(MMEA2_EDC_CNT3, DRAMWR_PAGEMEM_DED_COUNT),
873+ },
874+ { "MMEA2_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT3),
875+ 0, 0,
876+ SOC15_REG_FIELD(MMEA2_EDC_CNT3, IORD_CMDMEM_DED_COUNT),
877+ },
878+ { "MMEA2_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT3),
879+ 0, 0,
880+ SOC15_REG_FIELD(MMEA2_EDC_CNT3, IOWR_CMDMEM_DED_COUNT),
881+ },
882+ { "MMEA2_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT3),
883+ 0, 0,
884+ SOC15_REG_FIELD(MMEA2_EDC_CNT3, GMIRD_PAGEMEM_DED_COUNT),
885+ },
886+ { "MMEA2_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT3),
887+ 0, 0,
888+ SOC15_REG_FIELD(MMEA2_EDC_CNT3, GMIWR_PAGEMEM_DED_COUNT),
889+ },
890+
891+ /* MMHUB Rang 3 */
892+ { "MMEA3_DRAMRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
893+ SOC15_REG_FIELD(MMEA3_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
894+ SOC15_REG_FIELD(MMEA3_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT),
895+ },
896+ { "MMEA3_DRAMWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
897+ SOC15_REG_FIELD(MMEA3_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
898+ SOC15_REG_FIELD(MMEA3_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT),
899+ },
900+ { "MMEA3_DRAMWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
901+ SOC15_REG_FIELD(MMEA3_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
902+ SOC15_REG_FIELD(MMEA3_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT),
903+ },
904+ { "MMEA3_RRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
905+ SOC15_REG_FIELD(MMEA3_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
906+ SOC15_REG_FIELD(MMEA3_EDC_CNT, RRET_TAGMEM_DED_COUNT),
907+ },
908+ { "MMEA3_WRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
909+ SOC15_REG_FIELD(MMEA3_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
910+ SOC15_REG_FIELD(MMEA3_EDC_CNT, WRET_TAGMEM_DED_COUNT),
911+ },
912+ { "MMEA3_IOWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
913+ SOC15_REG_FIELD(MMEA3_EDC_CNT, IOWR_DATAMEM_SEC_COUNT),
914+ SOC15_REG_FIELD(MMEA3_EDC_CNT, IOWR_DATAMEM_DED_COUNT),
915+ },
916+ { "MMEA3_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
917+ SOC15_REG_FIELD(MMEA3_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
918+ 0, 0,
919+ },
920+ { "MMEA3_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
921+ SOC15_REG_FIELD(MMEA3_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
922+ 0, 0,
923+ },
924+ { "MMEA3_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
925+ SOC15_REG_FIELD(MMEA3_EDC_CNT, IORD_CMDMEM_SED_COUNT),
926+ 0, 0,
927+ },
928+ { "MMEA3_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT),
929+ SOC15_REG_FIELD(MMEA3_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
930+ 0, 0,
931+ },
932+ { "MMEA3_GMIRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
933+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
934+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT),
935+ },
936+ { "MMEA3_GMIWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
937+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
938+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT),
939+ },
940+ { "MMEA3_GMIWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
941+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
942+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT),
943+ },
944+ { "MMEA3_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
945+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
946+ 0, 0,
947+ },
948+ { "MMEA3_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
949+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
950+ 0, 0,
951+ },
952+ { "MMEA3_MAM_D0MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
953+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, MAM_D0MEM_SED_COUNT),
954+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, MAM_D0MEM_DED_COUNT),
955+ },
956+ { "MMEA3_MAM_D1MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
957+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, MAM_D1MEM_SED_COUNT),
958+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, MAM_D1MEM_DED_COUNT),
959+ },
960+ { "MMEA3_MAM_D2MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
961+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, MAM_D2MEM_SED_COUNT),
962+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, MAM_D2MEM_DED_COUNT),
963+ },
964+ { "MMEA3_MAM_D3MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2),
965+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, MAM_D3MEM_SED_COUNT),
966+ SOC15_REG_FIELD(MMEA3_EDC_CNT2, MAM_D3MEM_DED_COUNT),
967+ },
968+ { "MMEA3_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT3),
969+ 0, 0,
970+ SOC15_REG_FIELD(MMEA3_EDC_CNT3, DRAMRD_PAGEMEM_DED_COUNT),
971+ },
972+ { "MMEA3_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT3),
973+ 0, 0,
974+ SOC15_REG_FIELD(MMEA3_EDC_CNT3, DRAMWR_PAGEMEM_DED_COUNT),
975+ },
976+ { "MMEA3_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT3),
977+ 0, 0,
978+ SOC15_REG_FIELD(MMEA3_EDC_CNT3, IORD_CMDMEM_DED_COUNT),
979+ },
980+ { "MMEA3_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT3),
981+ 0, 0,
982+ SOC15_REG_FIELD(MMEA3_EDC_CNT3, IOWR_CMDMEM_DED_COUNT),
983+ },
984+ { "MMEA3_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT3),
985+ 0, 0,
986+ SOC15_REG_FIELD(MMEA3_EDC_CNT3, GMIRD_PAGEMEM_DED_COUNT),
987+ },
988+ { "MMEA3_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT3),
989+ 0, 0,
990+ SOC15_REG_FIELD(MMEA3_EDC_CNT3, GMIWR_PAGEMEM_DED_COUNT),
991+ },
992+
993+ /* MMHUB Range 4 */
994+ { "MMEA4_DRAMRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
995+ SOC15_REG_FIELD(MMEA4_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
996+ SOC15_REG_FIELD(MMEA4_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT),
997+ },
998+ { "MMEA4_DRAMWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
999+ SOC15_REG_FIELD(MMEA4_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
1000+ SOC15_REG_FIELD(MMEA4_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT),
1001+ },
1002+ { "MMEA4_DRAMWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
1003+ SOC15_REG_FIELD(MMEA4_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
1004+ SOC15_REG_FIELD(MMEA4_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT),
1005+ },
1006+ { "MMEA4_RRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
1007+ SOC15_REG_FIELD(MMEA4_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
1008+ SOC15_REG_FIELD(MMEA4_EDC_CNT, RRET_TAGMEM_DED_COUNT),
1009+ },
1010+ { "MMEA4_WRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
1011+ SOC15_REG_FIELD(MMEA4_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
1012+ SOC15_REG_FIELD(MMEA4_EDC_CNT, WRET_TAGMEM_DED_COUNT),
1013+ },
1014+ { "MMEA4_IOWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
1015+ SOC15_REG_FIELD(MMEA4_EDC_CNT, IOWR_DATAMEM_SEC_COUNT),
1016+ SOC15_REG_FIELD(MMEA4_EDC_CNT, IOWR_DATAMEM_DED_COUNT),
1017+ },
1018+ { "MMEA4_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
1019+ SOC15_REG_FIELD(MMEA4_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
1020+ 0, 0,
1021+ },
1022+ { "MMEA4_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
1023+ SOC15_REG_FIELD(MMEA4_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
1024+ 0, 0,
1025+ },
1026+ { "MMEA4_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
1027+ SOC15_REG_FIELD(MMEA4_EDC_CNT, IORD_CMDMEM_SED_COUNT),
1028+ 0, 0,
1029+ },
1030+ { "MMEA4_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT),
1031+ SOC15_REG_FIELD(MMEA4_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
1032+ 0, 0,
1033+ },
1034+ { "MMEA4_GMIRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1035+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
1036+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT),
1037+ },
1038+ { "MMEA4_GMIWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1039+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
1040+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT),
1041+ },
1042+ { "MMEA4_GMIWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1043+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
1044+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT),
1045+ },
1046+ { "MMEA4_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1047+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
1048+ 0, 0,
1049+ },
1050+ { "MMEA4_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1051+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
1052+ 0, 0,
1053+ },
1054+ { "MMEA4_MAM_D0MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1055+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, MAM_D0MEM_SED_COUNT),
1056+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, MAM_D0MEM_DED_COUNT),
1057+ },
1058+ { "MMEA4_MAM_D1MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1059+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, MAM_D1MEM_SED_COUNT),
1060+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, MAM_D1MEM_DED_COUNT),
1061+ },
1062+ { "MMEA4_MAM_D2MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1063+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, MAM_D2MEM_SED_COUNT),
1064+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, MAM_D2MEM_DED_COUNT),
1065+ },
1066+ { "MMEA4_MAM_D3MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2),
1067+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, MAM_D3MEM_SED_COUNT),
1068+ SOC15_REG_FIELD(MMEA4_EDC_CNT2, MAM_D3MEM_DED_COUNT),
1069+ },
1070+ { "MMEA4_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT3),
1071+ 0, 0,
1072+ SOC15_REG_FIELD(MMEA4_EDC_CNT3, DRAMRD_PAGEMEM_DED_COUNT),
1073+ },
1074+ { "MMEA4_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT3),
1075+ 0, 0,
1076+ SOC15_REG_FIELD(MMEA4_EDC_CNT3, DRAMWR_PAGEMEM_DED_COUNT),
1077+ },
1078+ { "MMEA4_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT3),
1079+ 0, 0,
1080+ SOC15_REG_FIELD(MMEA4_EDC_CNT3, IORD_CMDMEM_DED_COUNT),
1081+ },
1082+ { "MMEA4_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT3),
1083+ 0, 0,
1084+ SOC15_REG_FIELD(MMEA4_EDC_CNT3, IOWR_CMDMEM_DED_COUNT),
1085+ },
1086+ { "MMEA4_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT3),
1087+ 0, 0,
1088+ SOC15_REG_FIELD(MMEA4_EDC_CNT3, GMIRD_PAGEMEM_DED_COUNT),
1089+ },
1090+ { "MMEA4_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT3),
1091+ 0, 0,
1092+ SOC15_REG_FIELD(MMEA4_EDC_CNT3, GMIWR_PAGEMEM_DED_COUNT),
1093+ },
1094+
1095+ /* MMHUAB Range 5 */
1096+ { "MMEA5_DRAMRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1097+ SOC15_REG_FIELD(MMEA5_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
1098+ SOC15_REG_FIELD(MMEA5_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT),
1099+ },
1100+ { "MMEA5_DRAMWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1101+ SOC15_REG_FIELD(MMEA5_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
1102+ SOC15_REG_FIELD(MMEA5_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT),
1103+ },
1104+ { "MMEA5_DRAMWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1105+ SOC15_REG_FIELD(MMEA5_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
1106+ SOC15_REG_FIELD(MMEA5_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT),
1107+ },
1108+ { "MMEA5_RRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1109+ SOC15_REG_FIELD(MMEA5_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
1110+ SOC15_REG_FIELD(MMEA5_EDC_CNT, RRET_TAGMEM_DED_COUNT),
1111+ },
1112+ { "MMEA5_WRET_TAGMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1113+ SOC15_REG_FIELD(MMEA5_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
1114+ SOC15_REG_FIELD(MMEA5_EDC_CNT, WRET_TAGMEM_DED_COUNT),
1115+ },
1116+ { "MMEA5_IOWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1117+ SOC15_REG_FIELD(MMEA5_EDC_CNT, IOWR_DATAMEM_SEC_COUNT),
1118+ SOC15_REG_FIELD(MMEA5_EDC_CNT, IOWR_DATAMEM_DED_COUNT),
1119+ },
1120+ { "MMEA5_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1121+ SOC15_REG_FIELD(MMEA5_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
1122+ 0, 0,
1123+ },
1124+ { "MMEA5_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1125+ SOC15_REG_FIELD(MMEA5_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
1126+ 0, 0,
1127+ },
1128+ { "MMEA5_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1129+ SOC15_REG_FIELD(MMEA5_EDC_CNT, IORD_CMDMEM_SED_COUNT),
1130+ 0, 0,
1131+ },
1132+ { "MMEA5_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT),
1133+ SOC15_REG_FIELD(MMEA5_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
1134+ 0, 0,
1135+ },
1136+ { "MMEA5_GMIRD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1137+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
1138+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT),
1139+ },
1140+ { "MMEA5_GMIWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1141+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
1142+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT),
1143+ },
1144+ { "MMEA5_GMIWR_DATAMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1145+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
1146+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT),
1147+ },
1148+ { "MMEA5_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1149+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
1150+ 0, 0,
1151+ },
1152+ { "MMEA5_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1153+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
1154+ 0, 0,
1155+ },
1156+ { "MMEA5_MAM_D0MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1157+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, MAM_D0MEM_SED_COUNT),
1158+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, MAM_D0MEM_DED_COUNT),
1159+ },
1160+ { "MMEA5_MAM_D1MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1161+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, MAM_D1MEM_SED_COUNT),
1162+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, MAM_D1MEM_DED_COUNT),
1163+ },
1164+ { "MMEA5_MAM_D2MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1165+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, MAM_D2MEM_SED_COUNT),
1166+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, MAM_D2MEM_DED_COUNT),
1167+ },
1168+ { "MMEA5_MAM_D3MEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2),
1169+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, MAM_D3MEM_SED_COUNT),
1170+ SOC15_REG_FIELD(MMEA5_EDC_CNT2, MAM_D3MEM_DED_COUNT),
1171+ },
1172+ { "MMEA5_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT3),
1173+ 0, 0,
1174+ SOC15_REG_FIELD(MMEA5_EDC_CNT3, DRAMRD_PAGEMEM_DED_COUNT),
1175+ },
1176+ { "MMEA5_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT3),
1177+ 0, 0,
1178+ SOC15_REG_FIELD(MMEA5_EDC_CNT3, DRAMWR_PAGEMEM_DED_COUNT),
1179+ },
1180+ { "MMEA5_IORD_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT3),
1181+ 0, 0,
1182+ SOC15_REG_FIELD(MMEA5_EDC_CNT3, IORD_CMDMEM_DED_COUNT),
1183+ },
1184+ { "MMEA5_IOWR_CMDMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT3),
1185+ 0, 0,
1186+ SOC15_REG_FIELD(MMEA5_EDC_CNT3, IOWR_CMDMEM_DED_COUNT),
1187+ },
1188+ { "MMEA5_GMIRD_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT3),
1189+ 0, 0,
1190+ SOC15_REG_FIELD(MMEA5_EDC_CNT3, GMIRD_PAGEMEM_DED_COUNT),
1191+ },
1192+ { "MMEA5_GMIWR_PAGEMEM", SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT3),
1193+ 0, 0,
1194+ SOC15_REG_FIELD(MMEA5_EDC_CNT3, GMIWR_PAGEMEM_DED_COUNT),
1195+ },
1196+};
1197+
1198+static const struct soc15_reg_entry mmhub_v1_7_edc_cnt_regs[] = {
1199+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT), 0, 0, 0 },
1200+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT2), 0, 0, 0 },
1201+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_EDC_CNT3), 0, 0, 0 },
1202+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT), 0, 0, 0 },
1203+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT2), 0, 0, 0 },
1204+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_EDC_CNT3), 0, 0, 0 },
1205+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT), 0, 0, 0 },
1206+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT2), 0, 0, 0 },
1207+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_EDC_CNT3), 0, 0, 0 },
1208+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT), 0, 0, 0 },
1209+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT2), 0, 0, 0 },
1210+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_EDC_CNT3), 0, 0, 0 },
1211+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT), 0, 0, 0 },
1212+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT2), 0, 0, 0 },
1213+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_EDC_CNT3), 0, 0, 0 },
1214+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT), 0, 0, 0 },
1215+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT2), 0, 0, 0 },
1216+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_EDC_CNT3), 0, 0, 0 },
1217+};
1218+
1219+static int mmhub_v1_7_get_ras_error_count(struct amdgpu_device *adev,
1220+ const struct soc15_reg_entry *reg,
1221+ uint32_t value,
1222+ uint32_t *sec_count,
1223+ uint32_t *ded_count)
1224+{
1225+ uint32_t i;
1226+ uint32_t sec_cnt, ded_cnt;
1227+
1228+ for (i = 0; i < ARRAY_SIZE(mmhub_v1_7_ras_fields); i++) {
1229+ if(mmhub_v1_7_ras_fields[i].reg_offset != reg->reg_offset)
1230+ continue;
1231+
1232+ sec_cnt = (value &
1233+ mmhub_v1_7_ras_fields[i].sec_count_mask) >>
1234+ mmhub_v1_7_ras_fields[i].sec_count_shift;
1235+ if (sec_cnt) {
1236+ dev_info(adev->dev, "MMHUB SubBlock %s, SEC %d\n",
1237+ mmhub_v1_7_ras_fields[i].name,
1238+ sec_cnt);
1239+ *sec_count += sec_cnt;
1240+ }
1241+
1242+ ded_cnt = (value &
1243+ mmhub_v1_7_ras_fields[i].ded_count_mask) >>
1244+ mmhub_v1_7_ras_fields[i].ded_count_shift;
1245+ if (ded_cnt) {
1246+ dev_info(adev->dev, "MMHUB SubBlock %s, DED %d\n",
1247+ mmhub_v1_7_ras_fields[i].name,
1248+ ded_cnt);
1249+ *ded_count += ded_cnt;
1250+ }
1251+ }
1252+
1253+ return 0;
1254+}
1255+
1256+static void mmhub_v1_7_query_ras_error_count(struct amdgpu_device *adev,
1257+ void *ras_error_status)
1258+{
1259+ struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
1260+ uint32_t sec_count = 0, ded_count = 0;
1261+ uint32_t i;
1262+ uint32_t reg_value;
1263+
1264+ err_data->ue_count = 0;
1265+ err_data->ce_count = 0;
1266+
1267+ for (i = 0; i < ARRAY_SIZE(mmhub_v1_7_edc_cnt_regs); i++) {
1268+ reg_value =
1269+ RREG32(SOC15_REG_ENTRY_OFFSET(mmhub_v1_7_edc_cnt_regs[i]));
1270+ if (reg_value)
1271+ mmhub_v1_7_get_ras_error_count(adev, &mmhub_v1_7_edc_cnt_regs[i],
1272+ reg_value, &sec_count, &ded_count);
1273+ }
1274+
1275+ err_data->ce_count += sec_count;
1276+ err_data->ue_count += ded_count;
1277+}
1278+
1279+static void mmhub_v1_7_reset_ras_error_count(struct amdgpu_device *adev)
1280+{
1281+ uint32_t i;
1282+
1283+ /* write 0 to reset the edc counters */
1284+ if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__MMHUB)) {
1285+ for (i = 0; i < ARRAY_SIZE(mmhub_v1_7_edc_cnt_regs); i++)
1286+ WREG32(SOC15_REG_ENTRY_OFFSET(mmhub_v1_7_edc_cnt_regs[i]), 0);
1287+ }
1288+}
1289+
1290+static const struct soc15_reg_entry mmhub_v1_7_err_status_regs[] = {
1291+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA0_ERR_STATUS), 0, 0, 0 },
1292+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA1_ERR_STATUS), 0, 0, 0 },
1293+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA2_ERR_STATUS), 0, 0, 0 },
1294+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA3_ERR_STATUS), 0, 0, 0 },
1295+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA4_ERR_STATUS), 0, 0, 0 },
1296+ { SOC15_REG_ENTRY(MMHUB, 0, regMMEA5_ERR_STATUS), 0, 0, 0 },
1297+};
1298+
1299+static void mmhub_v1_7_query_ras_error_status(struct amdgpu_device *adev)
1300+{
1301+ int i;
1302+ uint32_t reg_value;
1303+
1304+ if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__MMHUB))
1305+ return;
1306+
1307+ for (i = 0; i < ARRAY_SIZE(mmhub_v1_7_err_status_regs); i++) {
1308+ reg_value =
1309+ RREG32(SOC15_REG_ENTRY_OFFSET(mmhub_v1_7_err_status_regs[i]));
1310+ if (reg_value)
1311+ dev_warn(adev->dev, "MMHUB EA err detected at instance: %d, status: 0x%x!\n",
1312+ i, reg_value);
1313+ }
1314+}
1315+
1316+const struct amdgpu_mmhub_funcs mmhub_v1_7_funcs = {
1317+ .ras_late_init = amdgpu_mmhub_ras_late_init,
1318+ .query_ras_error_count = mmhub_v1_7_query_ras_error_count,
1319+ .reset_ras_error_count = mmhub_v1_7_reset_ras_error_count,
1320+ .get_fb_location = mmhub_v1_7_get_fb_location,
1321+ .init = mmhub_v1_7_init,
1322+ .gart_enable = mmhub_v1_7_gart_enable,
1323+ .set_fault_enable_default = mmhub_v1_7_set_fault_enable_default,
1324+ .gart_disable = mmhub_v1_7_gart_disable,
1325+ .set_clockgating = mmhub_v1_7_set_clockgating,
1326+ .get_clockgating = mmhub_v1_7_get_clockgating,
1327+ .setup_vm_pt_regs = mmhub_v1_7_setup_vm_pt_regs,
1328+ .query_ras_error_status = mmhub_v1_7_query_ras_error_status,
1329+};
drivers/gpu/drm/amd/amdgpu/mmhub_v1_7.h+28−0
@@ -0,0 +1,28 @@
1+/*
2+ * Copyright 2016 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#ifndef __MMHUB_V1_7_H__
24+#define __MMHUB_V1_7_H__
25+
26+extern const struct amdgpu_mmhub_funcs mmhub_v1_7_funcs;
27+
28+#endif
drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c+40−8
@@ -52,6 +52,20 @@ · 51 unmodified lines
5252 #define BIF_MMSCH1_DOORBELL_RANGE__OFFSET_MASK 0x00000FFCL
5353 #define BIF_MMSCH1_DOORBELL_RANGE__SIZE_MASK 0x001F0000L
5454
55+#define BIF_MMSCH1_DOORBELL_RANGE__OFFSET_MASK 0x00000FFCL
56+#define BIF_MMSCH1_DOORBELL_RANGE__SIZE_MASK 0x001F0000L
57+
58+#define mmBIF_MMSCH1_DOORBELL_RANGE_ALDE 0x01d8
59+#define mmBIF_MMSCH1_DOORBELL_RANGE_ALDE_BASE_IDX 2
60+//BIF_MMSCH1_DOORBELL_ALDE_RANGE
61+#define BIF_MMSCH1_DOORBELL_RANGE_ALDE__OFFSET__SHIFT 0x2
62+#define BIF_MMSCH1_DOORBELL_RANGE_ALDE__SIZE__SHIFT 0x10
63+#define BIF_MMSCH1_DOORBELL_RANGE_ALDE__OFFSET_MASK 0x00000FFCL
64+#define BIF_MMSCH1_DOORBELL_RANGE_ALDE__SIZE_MASK 0x001F0000L
65+
66+#define mmRCC_DEV0_EPF0_STRAP0_ALDE 0x0015
67+#define mmRCC_DEV0_EPF0_STRAP0_ALDE_BASE_IDX 2
68+
5569 static void nbio_v7_4_query_ras_error_count(struct amdgpu_device *adev,
5670 void *ras_error_status);
5771
@@ -65,7 +79,12 @@ static void nbio_v7_4_remap_hdp_registers(struct amdgpu_device *adev) · 7 unmodified lines
6579
6680 static u32 nbio_v7_4_get_rev_id(struct amdgpu_device *adev)
6781 {
68− u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0);
82+ u32 tmp;
83+
84+ if (adev->asic_type == CHIP_ALDEBARAN)
85+ tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0_ALDE);
86+ else
87+ tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0);
6988
7089 tmp &= RCC_DEV0_EPF0_STRAP0__STRAP_ATI_REV_ID_DEV0_F0_MASK;
7190 tmp >>= RCC_DEV0_EPF0_STRAP0__STRAP_ATI_REV_ID_DEV0_F0__SHIFT;
@@ -92,10 +111,10 @@ static void nbio_v7_4_sdma_doorbell_range(struct amdgpu_device *adev, int instan · 20 unmodified lines
92111 {
93112 u32 reg, doorbell_range;
94113
95− if (instance < 2)
114+ if (instance < 2) {
96115 reg = instance +
97116 SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA0_DOORBELL_RANGE);
98− else
117+ } else {
99118 /*
100119 * These registers address of SDMA2~7 is not consecutive
101120 * from SDMA0~1. Need plus 4 dwords offset.
@@ -103,9 +122,19 @@ static void nbio_v7_4_sdma_doorbell_range(struct amdgpu_device *adev, int instan · 1 unmodified line
103122 * BIF_SDMA0_DOORBELL_RANGE: 0x3bc0
104123 * BIF_SDMA1_DOORBELL_RANGE: 0x3bc4
105124 * BIF_SDMA2_DOORBELL_RANGE: 0x3bd8
125++ * BIF_SDMA4_DOORBELL_RANGE:
126++ * ARCTURUS: 0x3be0
127++ * ALDEBARAN: 0x3be4
106128 */
107− reg = instance + 0x4 +
108− SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA0_DOORBELL_RANGE);
129+ if (adev->asic_type == CHIP_ALDEBARAN && instance == 4)
130+ reg = instance + 0x4 + 0x1 +
131+ SOC15_REG_OFFSET(NBIO, 0,
132+ mmBIF_SDMA0_DOORBELL_RANGE);
133+ else
134+ reg = instance + 0x4 +
135+ SOC15_REG_OFFSET(NBIO, 0,
136+ mmBIF_SDMA0_DOORBELL_RANGE);
137+ }
109138
110139 doorbell_range = RREG32(reg);
111140
@@ -124,9 +153,12 @@ static void nbio_v7_4_vcn_doorbell_range(struct amdgpu_device *adev, bool use_do · 12 unmodified lines
124153 u32 reg;
125154 u32 doorbell_range;
126155
127− if (instance)
128− reg = SOC15_REG_OFFSET(NBIO, 0, mmBIF_MMSCH1_DOORBELL_RANGE);
129− else
156+ if (instance) {
157+ if (adev->asic_type == CHIP_ALDEBARAN)
158+ reg = SOC15_REG_OFFSET(NBIO, 0, mmBIF_MMSCH1_DOORBELL_RANGE_ALDE);
159+ else
160+ reg = SOC15_REG_OFFSET(NBIO, 0, mmBIF_MMSCH1_DOORBELL_RANGE);
161+ } else
130162 reg = SOC15_REG_OFFSET(NBIO, 0, mmBIF_MMSCH0_DOORBELL_RANGE);
131163
132164 doorbell_range = RREG32(reg);
drivers/gpu/drm/amd/amdgpu/nv.c+182−39
@@ -25,6 +25,8 @@ · 24 unmodified lines
2525 #include <linux/module.h>
2626 #include <linux/pci.h>
2727
28+#include <drm/amdgpu_drm.h>
29+
2830 #include "amdgpu.h"
2931 #include "amdgpu_atombios.h"
3032 #include "amdgpu_ih.h"
@@ -65,6 +67,184 @@ · 34 unmodified lines
6567
6668 static const struct amd_ip_funcs nv_common_ip_funcs;
6769
70+/* Navi */
71+static const struct amdgpu_video_codec_info nv_video_codecs_encode_array[] =
72+{
73+ {
74+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
75+ .max_width = 4096,
76+ .max_height = 2304,
77+ .max_pixels_per_frame = 4096 * 2304,
78+ .max_level = 0,
79+ },
80+ {
81+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
82+ .max_width = 4096,
83+ .max_height = 2304,
84+ .max_pixels_per_frame = 4096 * 2304,
85+ .max_level = 0,
86+ },
87+};
88+
89+static const struct amdgpu_video_codecs nv_video_codecs_encode =
90+{
91+ .codec_count = ARRAY_SIZE(nv_video_codecs_encode_array),
92+ .codec_array = nv_video_codecs_encode_array,
93+};
94+
95+/* Navi1x */
96+static const struct amdgpu_video_codec_info nv_video_codecs_decode_array[] =
97+{
98+ {
99+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
100+ .max_width = 4096,
101+ .max_height = 4096,
102+ .max_pixels_per_frame = 4096 * 4096,
103+ .max_level = 3,
104+ },
105+ {
106+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
107+ .max_width = 4096,
108+ .max_height = 4096,
109+ .max_pixels_per_frame = 4096 * 4096,
110+ .max_level = 5,
111+ },
112+ {
113+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
114+ .max_width = 4096,
115+ .max_height = 4096,
116+ .max_pixels_per_frame = 4096 * 4096,
117+ .max_level = 52,
118+ },
119+ {
120+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
121+ .max_width = 4096,
122+ .max_height = 4096,
123+ .max_pixels_per_frame = 4096 * 4096,
124+ .max_level = 4,
125+ },
126+ {
127+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
128+ .max_width = 8192,
129+ .max_height = 4352,
130+ .max_pixels_per_frame = 8192 * 4352,
131+ .max_level = 186,
132+ },
133+ {
134+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG,
135+ .max_width = 4096,
136+ .max_height = 4096,
137+ .max_pixels_per_frame = 4096 * 4096,
138+ .max_level = 0,
139+ },
140+ {
141+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9,
142+ .max_width = 8192,
143+ .max_height = 4352,
144+ .max_pixels_per_frame = 8192 * 4352,
145+ .max_level = 0,
146+ },
147+};
148+
149+static const struct amdgpu_video_codecs nv_video_codecs_decode =
150+{
151+ .codec_count = ARRAY_SIZE(nv_video_codecs_decode_array),
152+ .codec_array = nv_video_codecs_decode_array,
153+};
154+
155+/* Sienna Cichlid */
156+static const struct amdgpu_video_codec_info sc_video_codecs_decode_array[] =
157+{
158+ {
159+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
160+ .max_width = 4096,
161+ .max_height = 4096,
162+ .max_pixels_per_frame = 4096 * 4096,
163+ .max_level = 3,
164+ },
165+ {
166+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
167+ .max_width = 4096,
168+ .max_height = 4096,
169+ .max_pixels_per_frame = 4096 * 4096,
170+ .max_level = 5,
171+ },
172+ {
173+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
174+ .max_width = 4096,
175+ .max_height = 4096,
176+ .max_pixels_per_frame = 4096 * 4096,
177+ .max_level = 52,
178+ },
179+ {
180+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
181+ .max_width = 4096,
182+ .max_height = 4096,
183+ .max_pixels_per_frame = 4096 * 4096,
184+ .max_level = 4,
185+ },
186+ {
187+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
188+ .max_width = 8192,
189+ .max_height = 4352,
190+ .max_pixels_per_frame = 8192 * 4352,
191+ .max_level = 186,
192+ },
193+ {
194+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG,
195+ .max_width = 4096,
196+ .max_height = 4096,
197+ .max_pixels_per_frame = 4096 * 4096,
198+ .max_level = 0,
199+ },
200+ {
201+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9,
202+ .max_width = 8192,
203+ .max_height = 4352,
204+ .max_pixels_per_frame = 8192 * 4352,
205+ .max_level = 0,
206+ },
207+ {
208+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_AV1,
209+ .max_width = 8192,
210+ .max_height = 4352,
211+ .max_pixels_per_frame = 8192 * 4352,
212+ .max_level = 0,
213+ },
214+};
215+
216+static const struct amdgpu_video_codecs sc_video_codecs_decode =
217+{
218+ .codec_count = ARRAY_SIZE(sc_video_codecs_decode_array),
219+ .codec_array = sc_video_codecs_decode_array,
220+};
221+
222+static int nv_query_video_codecs(struct amdgpu_device *adev, bool encode,
223+ const struct amdgpu_video_codecs **codecs)
224+{
225+ switch (adev->asic_type) {
226+ case CHIP_SIENNA_CICHLID:
227+ case CHIP_NAVY_FLOUNDER:
228+ case CHIP_DIMGREY_CAVEFISH:
229+ case CHIP_VANGOGH:
230+ if (encode)
231+ *codecs = &nv_video_codecs_encode;
232+ else
233+ *codecs = &sc_video_codecs_decode;
234+ return 0;
235+ case CHIP_NAVI10:
236+ case CHIP_NAVI14:
237+ case CHIP_NAVI12:
238+ if (encode)
239+ *codecs = &nv_video_codecs_encode;
240+ else
241+ *codecs = &nv_video_codecs_decode;
242+ return 0;
243+ default:
244+ return -EINVAL;
245+ }
246+}
247+
68248 /*
69249 * Indirect registers accessor
70250 */
@@ -304,44 +484,6 @@ static int nv_read_register(struct amdgpu_device *adev, u32 se_num, · 233 unmodified lines
304484 return -EINVAL;
305485 }
306486
307−static int nv_asic_mode1_reset(struct amdgpu_device *adev)
308−{
309− u32 i;
310− int ret = 0;
311−
312− amdgpu_atombios_scratch_regs_engine_hung(adev, true);
313−
314− /* disable BM */
315− pci_clear_master(adev->pdev);
316−
317− amdgpu_device_cache_pci_state(adev->pdev);
318−
319− if (amdgpu_dpm_is_mode1_reset_supported(adev)) {
320− dev_info(adev->dev, "GPU smu mode1 reset\n");
321− ret = amdgpu_dpm_mode1_reset(adev);
322− } else {
323− dev_info(adev->dev, "GPU psp mode1 reset\n");
324− ret = psp_gpu_reset(adev);
325− }
326−
327− if (ret)
328− dev_err(adev->dev, "GPU mode1 reset failed\n");
329− amdgpu_device_load_pci_state(adev->pdev);
330−
331− /* wait for asic to come out of reset */
332− for (i = 0; i < adev->usec_timeout; i++) {
333− u32 memsize = adev->nbio.funcs->get_memsize(adev);
334−
335− if (memsize != 0xffffffff)
336− break;
337− udelay(1);
338− }
339−
340− amdgpu_atombios_scratch_regs_engine_hung(adev, false);
341−
342− return ret;
343−}
344−
345487 static int nv_asic_mode2_reset(struct amdgpu_device *adev)
346488 {
347489 u32 i;
@@ -444,7 +586,7 @@ static int nv_asic_reset(struct amdgpu_device *adev) · 96 unmodified lines
444586 break;
445587 default:
446588 dev_info(adev->dev, "MODE1 reset\n");
447− ret = nv_asic_mode1_reset(adev);
589+ ret = amdgpu_device_mode1_reset(adev);
448590 break;
449591 }
450592
@@ -847,6 +989,7 @@ static const struct amdgpu_asic_funcs nv_asic_funcs = · 396 unmodified lines
847989 .supports_baco = &nv_asic_supports_baco,
848990 .pre_asic_init = &nv_pre_asic_init,
849991 .update_umd_stable_pstate = &nv_update_umd_stable_pstate,
992+ .query_video_codecs = &nv_query_video_codecs,
850993 };
851994
852995 static int nv_common_early_init(void *handle)
drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h+1−0
@@ -235,6 +235,7 @@ enum psp_gfx_fw_type { · 234 unmodified lines
235235 GFX_FW_TYPE_SDMA6 = 56, /* SDMA6 MI */
236236 GFX_FW_TYPE_SDMA7 = 57, /* SDMA7 MI */
237237 GFX_FW_TYPE_VCN1 = 58, /* VCN1 MI */
238+ GFX_FW_TYPE_REG_LIST = 67, /* REG_LIST MI */
238239 GFX_FW_TYPE_MAX
239240 };
240241
drivers/gpu/drm/amd/amdgpu/psp_v13_0.c+378−0
@@ -0,0 +1,378 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#include "amdgpu.h"
24+#include "amdgpu_psp.h"
25+#include "amdgpu_ucode.h"
26+#include "soc15_common.h"
27+#include "psp_v13_0.h"
28+
29+#include "mp/mp_13_0_2_offset.h"
30+#include "mp/mp_13_0_2_sh_mask.h"
31+
32+MODULE_FIRMWARE("amdgpu/aldebaran_sos.bin");
33+MODULE_FIRMWARE("amdgpu/aldebaran_ta.bin");
34+
35+static int psp_v13_0_init_microcode(struct psp_context *psp)
36+{
37+ struct amdgpu_device *adev = psp->adev;
38+ const char *chip_name;
39+ int err = 0;
40+
41+ switch (adev->asic_type) {
42+ case CHIP_ALDEBARAN:
43+ chip_name = "aldebaran";
44+ break;
45+ default:
46+ BUG();
47+ }
48+
49+ err = psp_init_sos_microcode(psp, chip_name);
50+ if (err)
51+ return err;
52+
53+ err = psp_init_ta_microcode(&adev->psp, chip_name);
54+
55+ return err;
56+}
57+
58+static bool psp_v13_0_is_sos_alive(struct psp_context *psp)
59+{
60+ struct amdgpu_device *adev = psp->adev;
61+ uint32_t sol_reg;
62+
63+ sol_reg = RREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_81);
64+
65+ return sol_reg != 0x0;
66+}
67+
68+static int psp_v13_0_wait_for_bootloader(struct psp_context *psp)
69+{
70+ struct amdgpu_device *adev = psp->adev;
71+
72+ int ret;
73+ int retry_loop;
74+
75+ for (retry_loop = 0; retry_loop < 10; retry_loop++) {
76+ /* Wait for bootloader to signify that is
77+ ready having bit 31 of C2PMSG_35 set to 1 */
78+ ret = psp_wait_for(psp,
79+ SOC15_REG_OFFSET(MP0, 0, regMP0_SMN_C2PMSG_35),
80+ 0x80000000,
81+ 0x80000000,
82+ false);
83+
84+ if (ret == 0)
85+ return 0;
86+ }
87+
88+ return ret;
89+}
90+
91+static int psp_v13_0_bootloader_load_kdb(struct psp_context *psp)
92+{
93+ int ret;
94+ uint32_t psp_gfxdrv_command_reg = 0;
95+ struct amdgpu_device *adev = psp->adev;
96+
97+ /* Check tOS sign of life register to confirm sys driver and sOS
98+ * are already been loaded.
99+ */
100+ if (psp_v13_0_is_sos_alive(psp))
101+ return 0;
102+
103+ ret = psp_v13_0_wait_for_bootloader(psp);
104+ if (ret)
105+ return ret;
106+
107+ memset(psp->fw_pri_buf, 0, PSP_1_MEG);
108+
109+ /* Copy PSP KDB binary to memory */
110+ memcpy(psp->fw_pri_buf, psp->kdb_start_addr, psp->kdb_bin_size);
111+
112+ /* Provide the PSP KDB to bootloader */
113+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_36,
114+ (uint32_t)(psp->fw_pri_mc_addr >> 20));
115+ psp_gfxdrv_command_reg = PSP_BL__LOAD_KEY_DATABASE;
116+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_35,
117+ psp_gfxdrv_command_reg);
118+
119+ ret = psp_v13_0_wait_for_bootloader(psp);
120+
121+ return ret;
122+}
123+
124+static int psp_v13_0_bootloader_load_sysdrv(struct psp_context *psp)
125+{
126+ int ret;
127+ uint32_t psp_gfxdrv_command_reg = 0;
128+ struct amdgpu_device *adev = psp->adev;
129+
130+ /* Check sOS sign of life register to confirm sys driver and sOS
131+ * are already been loaded.
132+ */
133+ if (psp_v13_0_is_sos_alive(psp))
134+ return 0;
135+
136+ ret = psp_v13_0_wait_for_bootloader(psp);
137+ if (ret)
138+ return ret;
139+
140+ memset(psp->fw_pri_buf, 0, PSP_1_MEG);
141+
142+ /* Copy PSP System Driver binary to memory */
143+ memcpy(psp->fw_pri_buf, psp->sys_start_addr, psp->sys_bin_size);
144+
145+ /* Provide the sys driver to bootloader */
146+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_36,
147+ (uint32_t)(psp->fw_pri_mc_addr >> 20));
148+ psp_gfxdrv_command_reg = PSP_BL__LOAD_SYSDRV;
149+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_35,
150+ psp_gfxdrv_command_reg);
151+
152+ /* there might be handshake issue with hardware which needs delay */
153+ mdelay(20);
154+
155+ ret = psp_v13_0_wait_for_bootloader(psp);
156+
157+ return ret;
158+}
159+
160+static int psp_v13_0_bootloader_load_sos(struct psp_context *psp)
161+{
162+ int ret;
163+ unsigned int psp_gfxdrv_command_reg = 0;
164+ struct amdgpu_device *adev = psp->adev;
165+
166+ /* Check sOS sign of life register to confirm sys driver and sOS
167+ * are already been loaded.
168+ */
169+ if (psp_v13_0_is_sos_alive(psp))
170+ return 0;
171+
172+ ret = psp_v13_0_wait_for_bootloader(psp);
173+ if (ret)
174+ return ret;
175+
176+ memset(psp->fw_pri_buf, 0, PSP_1_MEG);
177+
178+ /* Copy Secure OS binary to PSP memory */
179+ memcpy(psp->fw_pri_buf, psp->sos_start_addr, psp->sos_bin_size);
180+
181+ /* Provide the PSP secure OS to bootloader */
182+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_36,
183+ (uint32_t)(psp->fw_pri_mc_addr >> 20));
184+ psp_gfxdrv_command_reg = PSP_BL__LOAD_SOSDRV;
185+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_35,
186+ psp_gfxdrv_command_reg);
187+
188+ /* there might be handshake issue with hardware which needs delay */
189+ mdelay(20);
190+ ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, regMP0_SMN_C2PMSG_81),
191+ RREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_81),
192+ 0, true);
193+
194+ return ret;
195+}
196+
197+static int psp_v13_0_ring_init(struct psp_context *psp,
198+ enum psp_ring_type ring_type)
199+{
200+ int ret = 0;
201+ struct psp_ring *ring;
202+ struct amdgpu_device *adev = psp->adev;
203+
204+ ring = &psp->km_ring;
205+
206+ ring->ring_type = ring_type;
207+
208+ /* allocate 4k Page of Local Frame Buffer memory for ring */
209+ ring->ring_size = 0x1000;
210+ ret = amdgpu_bo_create_kernel(adev, ring->ring_size, PAGE_SIZE,
211+ AMDGPU_GEM_DOMAIN_VRAM,
212+ &adev->firmware.rbuf,
213+ &ring->ring_mem_mc_addr,
214+ (void **)&ring->ring_mem);
215+ if (ret) {
216+ ring->ring_size = 0;
217+ return ret;
218+ }
219+
220+ return 0;
221+}
222+
223+static int psp_v13_0_ring_stop(struct psp_context *psp,
224+ enum psp_ring_type ring_type)
225+{
226+ int ret = 0;
227+ struct amdgpu_device *adev = psp->adev;
228+
229+ if (amdgpu_sriov_vf(adev)) {
230+ /* Write the ring destroy command*/
231+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_101,
232+ GFX_CTRL_CMD_ID_DESTROY_GPCOM_RING);
233+ /* there might be handshake issue with hardware which needs delay */
234+ mdelay(20);
235+ /* Wait for response flag (bit 31) */
236+ ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, regMP0_SMN_C2PMSG_101),
237+ 0x80000000, 0x80000000, false);
238+ } else {
239+ /* Write the ring destroy command*/
240+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_64,
241+ GFX_CTRL_CMD_ID_DESTROY_RINGS);
242+ /* there might be handshake issue with hardware which needs delay */
243+ mdelay(20);
244+ /* Wait for response flag (bit 31) */
245+ ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, regMP0_SMN_C2PMSG_64),
246+ 0x80000000, 0x80000000, false);
247+ }
248+
249+ return ret;
250+}
251+
252+static int psp_v13_0_ring_create(struct psp_context *psp,
253+ enum psp_ring_type ring_type)
254+{
255+ int ret = 0;
256+ unsigned int psp_ring_reg = 0;
257+ struct psp_ring *ring = &psp->km_ring;
258+ struct amdgpu_device *adev = psp->adev;
259+
260+ if (amdgpu_sriov_vf(adev)) {
261+ ret = psp_v13_0_ring_stop(psp, ring_type);
262+ if (ret) {
263+ DRM_ERROR("psp_v13_0_ring_stop_sriov failed!\n");
264+ return ret;
265+ }
266+
267+ /* Write low address of the ring to C2PMSG_102 */
268+ psp_ring_reg = lower_32_bits(ring->ring_mem_mc_addr);
269+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_102, psp_ring_reg);
270+ /* Write high address of the ring to C2PMSG_103 */
271+ psp_ring_reg = upper_32_bits(ring->ring_mem_mc_addr);
272+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_103, psp_ring_reg);
273+
274+ /* Write the ring initialization command to C2PMSG_101 */
275+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_101,
276+ GFX_CTRL_CMD_ID_INIT_GPCOM_RING);
277+
278+ /* there might be handshake issue with hardware which needs delay */
279+ mdelay(20);
280+
281+ /* Wait for response flag (bit 31) in C2PMSG_101 */
282+ ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, regMP0_SMN_C2PMSG_101),
283+ 0x80000000, 0x8000FFFF, false);
284+
285+ } else {
286+ /* Wait for sOS ready for ring creation */
287+ ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, regMP0_SMN_C2PMSG_64),
288+ 0x80000000, 0x80000000, false);
289+ if (ret) {
290+ DRM_ERROR("Failed to wait for trust OS ready for ring creation\n");
291+ return ret;
292+ }
293+
294+ /* Write low address of the ring to C2PMSG_69 */
295+ psp_ring_reg = lower_32_bits(ring->ring_mem_mc_addr);
296+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_69, psp_ring_reg);
297+ /* Write high address of the ring to C2PMSG_70 */
298+ psp_ring_reg = upper_32_bits(ring->ring_mem_mc_addr);
299+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_70, psp_ring_reg);
300+ /* Write size of ring to C2PMSG_71 */
301+ psp_ring_reg = ring->ring_size;
302+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_71, psp_ring_reg);
303+ /* Write the ring initialization command to C2PMSG_64 */
304+ psp_ring_reg = ring_type;
305+ psp_ring_reg = psp_ring_reg << 16;
306+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_64, psp_ring_reg);
307+
308+ /* there might be handshake issue with hardware which needs delay */
309+ mdelay(20);
310+
311+ /* Wait for response flag (bit 31) in C2PMSG_64 */
312+ ret = psp_wait_for(psp, SOC15_REG_OFFSET(MP0, 0, regMP0_SMN_C2PMSG_64),
313+ 0x80000000, 0x8000FFFF, false);
314+ }
315+
316+ return ret;
317+}
318+
319+static int psp_v13_0_ring_destroy(struct psp_context *psp,
320+ enum psp_ring_type ring_type)
321+{
322+ int ret = 0;
323+ struct psp_ring *ring = &psp->km_ring;
324+ struct amdgpu_device *adev = psp->adev;
325+
326+ ret = psp_v13_0_ring_stop(psp, ring_type);
327+ if (ret)
328+ DRM_ERROR("Fail to stop psp ring\n");
329+
330+ amdgpu_bo_free_kernel(&adev->firmware.rbuf,
331+ &ring->ring_mem_mc_addr,
332+ (void **)&ring->ring_mem);
333+
334+ return ret;
335+}
336+
337+static uint32_t psp_v13_0_ring_get_wptr(struct psp_context *psp)
338+{
339+ uint32_t data;
340+ struct amdgpu_device *adev = psp->adev;
341+
342+ if (amdgpu_sriov_vf(adev))
343+ data = RREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_102);
344+ else
345+ data = RREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_67);
346+
347+ return data;
348+}
349+
350+static void psp_v13_0_ring_set_wptr(struct psp_context *psp, uint32_t value)
351+{
352+ struct amdgpu_device *adev = psp->adev;
353+
354+ if (amdgpu_sriov_vf(adev)) {
355+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_102, value);
356+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_101,
357+ GFX_CTRL_CMD_ID_CONSUME_CMD);
358+ } else
359+ WREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_67, value);
360+}
361+
362+static const struct psp_funcs psp_v13_0_funcs = {
363+ .init_microcode = psp_v13_0_init_microcode,
364+ .bootloader_load_kdb = psp_v13_0_bootloader_load_kdb,
365+ .bootloader_load_sysdrv = psp_v13_0_bootloader_load_sysdrv,
366+ .bootloader_load_sos = psp_v13_0_bootloader_load_sos,
367+ .ring_init = psp_v13_0_ring_init,
368+ .ring_create = psp_v13_0_ring_create,
369+ .ring_stop = psp_v13_0_ring_stop,
370+ .ring_destroy = psp_v13_0_ring_destroy,
371+ .ring_get_wptr = psp_v13_0_ring_get_wptr,
372+ .ring_set_wptr = psp_v13_0_ring_set_wptr,
373+};
374+
375+void psp_v13_0_set_psp_funcs(struct psp_context *psp)
376+{
377+ psp->funcs = &psp_v13_0_funcs;
378+}
drivers/gpu/drm/amd/amdgpu/psp_v13_0.h+30−0
@@ -0,0 +1,30 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#ifndef __PSP_V13_0_H__
24+#define __PSP_V13_0_H__
25+
26+#include "amdgpu_psp.h"
27+
28+void psp_v13_0_set_psp_funcs(struct psp_context *psp);
29+
30+#endif
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c+162−19
@@ -56,6 +56,7 @@ · 55 unmodified lines
5656 #include "ivsrcid/sdma1/irqsrcs_sdma1_4_0.h"
5757
5858 #include "amdgpu_ras.h"
59+#include "sdma_v4_4.h"
5960
6061 MODULE_FIRMWARE("amdgpu/vega10_sdma.bin");
6162 MODULE_FIRMWARE("amdgpu/vega10_sdma1.bin");
@@ -69,6 +70,7 @@ MODULE_FIRMWARE("amdgpu/raven2_sdma.bin"); · 7 unmodified lines
6970 MODULE_FIRMWARE("amdgpu/arcturus_sdma.bin");
7071 MODULE_FIRMWARE("amdgpu/renoir_sdma.bin");
7172 MODULE_FIRMWARE("amdgpu/green_sardine_sdma.bin");
73+MODULE_FIRMWARE("amdgpu/aldebaran_sdma.bin");
7274
7375 #define SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK 0x000000F8L
7476 #define SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK 0xFC000000L
@@ -259,6 +261,24 @@ static const struct soc15_reg_golden golden_settings_sdma_arct[] = · 184 unmodified lines
259261 SOC15_REG_GOLDEN_VALUE(SDMA7, 0, mmSDMA7_UTCL1_TIMEOUT, 0xffffffff, 0x00010001)
260262 };
261263
264+static const struct soc15_reg_golden golden_settings_sdma_aldebaran[] = {
265+ SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
266+ SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
267+ SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
268+ SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
269+ SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
270+ SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
271+ SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
272+ SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
273+ SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA2_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
274+ SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
275+ SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
276+ SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
277+ SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
278+ SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
279+ SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
280+};
281+
262282 static const struct soc15_reg_golden golden_settings_sdma_4_3[] = {
263283 SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
264284 SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
@@ -482,6 +502,11 @@ static void sdma_v4_0_init_golden_registers(struct amdgpu_device *adev) · 217 unmodified lines
482502 golden_settings_sdma_arct,
483503 ARRAY_SIZE(golden_settings_sdma_arct));
484504 break;
505+ case CHIP_ALDEBARAN:
506+ soc15_program_register_sequence(adev,
507+ golden_settings_sdma_aldebaran,
508+ ARRAY_SIZE(golden_settings_sdma_aldebaran));
509+ break;
485510 case CHIP_RAVEN:
486511 soc15_program_register_sequence(adev,
487512 golden_settings_sdma_4_1,
@@ -564,7 +589,8 @@ static void sdma_v4_0_destroy_inst_ctx(struct amdgpu_device *adev) · 76 unmodified lines
564589
565590 /* arcturus shares the same FW memory across
566591 all SDMA isntances */
567− if (adev->asic_type == CHIP_ARCTURUS)
592+ if (adev->asic_type == CHIP_ARCTURUS ||
593+ adev->asic_type == CHIP_ALDEBARAN)
568594 break;
569595 }
570596
@@ -621,6 +647,9 @@ static int sdma_v4_0_init_microcode(struct amdgpu_device *adev) · 50 unmodified lines
621647 else
622648 chip_name = "green_sardine";
623649 break;
650+ case CHIP_ALDEBARAN:
651+ chip_name = "aldebaran";
652+ break;
624653 default:
625654 BUG();
626655 }
@@ -636,8 +665,9 @@ static int sdma_v4_0_init_microcode(struct amdgpu_device *adev) · 9 unmodified lines
636665 goto out;
637666
638667 for (i = 1; i < adev->sdma.num_instances; i++) {
639− if (adev->asic_type == CHIP_ARCTURUS) {
640− /* Acturus will leverage the same FW memory
668+ if (adev->asic_type == CHIP_ARCTURUS ||
669+ adev->asic_type == CHIP_ALDEBARAN) {
670+ /* Acturus & Aldebaran will leverage the same FW memory
641671 for every SDMA instance */
642672 memcpy((void *)&adev->sdma.instance[i],
643673 (void *)&adev->sdma.instance[0],
@@ -1825,6 +1855,8 @@ static int sdma_v4_0_early_init(void *handle) · 1181 unmodified lines
18251855 adev->sdma.num_instances = 1;
18261856 else if (adev->asic_type == CHIP_ARCTURUS)
18271857 adev->sdma.num_instances = 8;
1858+ else if (adev->asic_type == CHIP_ALDEBARAN)
1859+ adev->sdma.num_instances = 5;
18281860 else
18291861 adev->sdma.num_instances = 2;
18301862
@@ -1895,6 +1927,33 @@ static int sdma_v4_0_sw_init(void *handle) · 64 unmodified lines
18951927 return r;
18961928 }
18971929
1930+ /* SDMA VM_HOLE/DOORBELL_INV/POLL_TIMEOUT/SRBM_WRITE_PROTECTION event*/
1931+ for (i = 0; i < adev->sdma.num_instances; i++) {
1932+ r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1933+ SDMA0_4_0__SRCID__SDMA_VM_HOLE,
1934+ &adev->sdma.vm_hole_irq);
1935+ if (r)
1936+ return r;
1937+
1938+ r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1939+ SDMA0_4_0__SRCID__SDMA_DOORBELL_INVALID,
1940+ &adev->sdma.doorbell_invalid_irq);
1941+ if (r)
1942+ return r;
1943+
1944+ r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1945+ SDMA0_4_0__SRCID__SDMA_POLL_TIMEOUT,
1946+ &adev->sdma.pool_timeout_irq);
1947+ if (r)
1948+ return r;
1949+
1950+ r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1951+ SDMA0_4_0__SRCID__SDMA_SRBMWRITE,
1952+ &adev->sdma.srbm_write_irq);
1953+ if (r)
1954+ return r;
1955+ }
1956+
18981957 for (i = 0; i < adev->sdma.num_instances; i++) {
18991958 ring = &adev->sdma.instance[i].ring;
19001959 ring->ring_obj = NULL;
@@ -2149,6 +2208,72 @@ static int sdma_v4_0_set_ecc_irq_state(struct amdgpu_device *adev, · 248 unmodified lines
21492208 return 0;
21502209 }
21512210
2211+static int sdma_v4_0_print_iv_entry(struct amdgpu_device *adev,
2212+ struct amdgpu_iv_entry *entry)
2213+{
2214+ int instance;
2215+ struct amdgpu_task_info task_info;
2216+ u64 addr;
2217+
2218+ instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2219+ if (instance < 0 || instance >= adev->sdma.num_instances) {
2220+ dev_err(adev->dev, "sdma instance invalid %d\n", instance);
2221+ return -EINVAL;
2222+ }
2223+
2224+ addr = (u64)entry->src_data[0] << 12;
2225+ addr |= ((u64)entry->src_data[1] & 0xf) << 44;
2226+
2227+ memset(&task_info, 0, sizeof(struct amdgpu_task_info));
2228+ amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
2229+
2230+ dev_info(adev->dev,
2231+ "[sdma%d] address:0x%016llx src_id:%u ring:%u vmid:%u "
2232+ "pasid:%u, for process %s pid %d thread %s pid %d\n",
2233+ instance, addr, entry->src_id, entry->ring_id, entry->vmid,
2234+ entry->pasid, task_info.process_name, task_info.tgid,
2235+ task_info.task_name, task_info.pid);
2236+ return 0;
2237+}
2238+
2239+static int sdma_v4_0_process_vm_hole_irq(struct amdgpu_device *adev,
2240+ struct amdgpu_irq_src *source,
2241+ struct amdgpu_iv_entry *entry)
2242+{
2243+ dev_err(adev->dev, "MC or SEM address in VM hole\n");
2244+ sdma_v4_0_print_iv_entry(adev, entry);
2245+ return 0;
2246+}
2247+
2248+static int sdma_v4_0_process_doorbell_invalid_irq(struct amdgpu_device *adev,
2249+ struct amdgpu_irq_src *source,
2250+ struct amdgpu_iv_entry *entry)
2251+{
2252+ dev_err(adev->dev, "SDMA received a doorbell from BIF with byte_enable !=0xff\n");
2253+ sdma_v4_0_print_iv_entry(adev, entry);
2254+ return 0;
2255+}
2256+
2257+static int sdma_v4_0_process_pool_timeout_irq(struct amdgpu_device *adev,
2258+ struct amdgpu_irq_src *source,
2259+ struct amdgpu_iv_entry *entry)
2260+{
2261+ dev_err(adev->dev,
2262+ "Polling register/memory timeout executing POLL_REG/MEM with finite timer\n");
2263+ sdma_v4_0_print_iv_entry(adev, entry);
2264+ return 0;
2265+}
2266+
2267+static int sdma_v4_0_process_srbm_write_irq(struct amdgpu_device *adev,
2268+ struct amdgpu_irq_src *source,
2269+ struct amdgpu_iv_entry *entry)
2270+{
2271+ dev_err(adev->dev,
2272+ "SDMA gets an Register Write SRBM_WRITE command in non-privilege command buffer\n");
2273+ sdma_v4_0_print_iv_entry(adev, entry);
2274+ return 0;
2275+}
2276+
21522277 static void sdma_v4_0_update_medium_grain_clock_gating(
21532278 struct amdgpu_device *adev,
21542279 bool enable)
@@ -2222,21 +2347,10 @@ static int sdma_v4_0_set_clockgating_state(void *handle, · 67 unmodified lines
22222347 if (amdgpu_sriov_vf(adev))
22232348 return 0;
22242349
2225− switch (adev->asic_type) {
2226− case CHIP_VEGA10:
2227− case CHIP_VEGA12:
2228− case CHIP_VEGA20:
2229− case CHIP_RAVEN:
2230− case CHIP_ARCTURUS:
2231− case CHIP_RENOIR:
2232− sdma_v4_0_update_medium_grain_clock_gating(adev,
2233− state == AMD_CG_STATE_GATE);
2234− sdma_v4_0_update_medium_grain_light_sleep(adev,
2235− state == AMD_CG_STATE_GATE);
2236− break;
2237− default:
2238− break;
2239− }
2350+ sdma_v4_0_update_medium_grain_clock_gating(adev,
2351+ state == AMD_CG_STATE_GATE);
2352+ sdma_v4_0_update_medium_grain_light_sleep(adev,
2353+ state == AMD_CG_STATE_GATE);
22402354 return 0;
22412355 }
22422356
@@ -2249,7 +2363,7 @@ static int sdma_v4_0_set_powergating_state(void *handle, · 6 unmodified lines
22492363 case CHIP_RAVEN:
22502364 case CHIP_RENOIR:
22512365 sdma_v4_1_update_power_gating(adev,
2252− state == AMD_PG_STATE_GATE ? true : false);
2366+ state == AMD_PG_STATE_GATE);
22532367 break;
22542368 default:
22552369 break;
@@ -2465,7 +2579,21 @@ static const struct amdgpu_irq_src_funcs sdma_v4_0_ecc_irq_funcs = { · 209 unmodified lines
24652579 .process = amdgpu_sdma_process_ecc_irq,
24662580 };
24672581
2582+static const struct amdgpu_irq_src_funcs sdma_v4_0_vm_hole_irq_funcs = {
2583+ .process = sdma_v4_0_process_vm_hole_irq,
2584+};
2585+
2586+static const struct amdgpu_irq_src_funcs sdma_v4_0_doorbell_invalid_irq_funcs = {
2587+ .process = sdma_v4_0_process_doorbell_invalid_irq,
2588+};
2589+
2590+static const struct amdgpu_irq_src_funcs sdma_v4_0_pool_timeout_irq_funcs = {
2591+ .process = sdma_v4_0_process_pool_timeout_irq,
2592+};
24682593
2594+static const struct amdgpu_irq_src_funcs sdma_v4_0_srbm_write_irq_funcs = {
2595+ .process = sdma_v4_0_process_srbm_write_irq,
2596+};
24692597
24702598 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev)
24712599 {
@@ -2474,9 +2602,17 @@ static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev) · 2 unmodified lines
24742602 adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE1;
24752603 adev->sdma.ecc_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE1;
24762604 break;
2605+ case 5:
2606+ adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE5;
2607+ adev->sdma.ecc_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE5;
2608+ break;
24772609 case 8:
24782610 adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
24792611 adev->sdma.ecc_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
2612+ adev->sdma.vm_hole_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE5;
2613+ adev->sdma.doorbell_invalid_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
2614+ adev->sdma.pool_timeout_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
2615+ adev->sdma.srbm_write_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
24802616 break;
24812617 case 2:
24822618 default:
@@ -2487,6 +2623,10 @@ static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev) · 4 unmodified lines
24872623 adev->sdma.trap_irq.funcs = &sdma_v4_0_trap_irq_funcs;
24882624 adev->sdma.illegal_inst_irq.funcs = &sdma_v4_0_illegal_inst_irq_funcs;
24892625 adev->sdma.ecc_irq.funcs = &sdma_v4_0_ecc_irq_funcs;
2626+ adev->sdma.vm_hole_irq.funcs = &sdma_v4_0_vm_hole_irq_funcs;
2627+ adev->sdma.doorbell_invalid_irq.funcs = &sdma_v4_0_doorbell_invalid_irq_funcs;
2628+ adev->sdma.pool_timeout_irq.funcs = &sdma_v4_0_pool_timeout_irq_funcs;
2629+ adev->sdma.srbm_write_irq.funcs = &sdma_v4_0_srbm_write_irq_funcs;
24902630 }
24912631
24922632 /**
@@ -2655,6 +2795,9 @@ static void sdma_v4_0_set_ras_funcs(struct amdgpu_device *adev) · 162 unmodified lines
26552795 case CHIP_ARCTURUS:
26562796 adev->sdma.funcs = &sdma_v4_0_ras_funcs;
26572797 break;
2798+ case CHIP_ALDEBARAN:
2799+ adev->sdma.funcs = &sdma_v4_4_ras_funcs;
2800+ break;
26582801 default:
26592802 break;
26602803 }
drivers/gpu/drm/amd/amdgpu/sdma_v4_4.c+232−0
@@ -0,0 +1,232 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#include "amdgpu.h"
24+#include "sdma/sdma_4_4_0_offset.h"
25+#include "sdma/sdma_4_4_0_sh_mask.h"
26+#include "soc15.h"
27+#include "amdgpu_ras.h"
28+
29+#define SDMA1_REG_OFFSET 0x600
30+#define SDMA2_REG_OFFSET 0x1cda0
31+#define SDMA3_REG_OFFSET 0x1d1a0
32+#define SDMA4_REG_OFFSET 0x1d5a0
33+
34+/* helper function that allow only use sdma0 register offset
35+ * to calculate register offset for all the sdma instances */
36+static uint32_t sdma_v4_4_get_reg_offset(struct amdgpu_device *adev,
37+ uint32_t instance,
38+ uint32_t offset)
39+{
40+ uint32_t sdma_base = adev->reg_offset[SDMA0_HWIP][0][0];
41+
42+ switch (instance) {
43+ case 0:
44+ return (sdma_base + offset);
45+ case 1:
46+ return (sdma_base + SDMA1_REG_OFFSET + offset);
47+ case 2:
48+ return (sdma_base + SDMA2_REG_OFFSET + offset);
49+ case 3:
50+ return (sdma_base + SDMA3_REG_OFFSET + offset);
51+ case 4:
52+ return (sdma_base + SDMA4_REG_OFFSET + offset);
53+ default:
54+ break;
55+ }
56+ return 0;
57+}
58+
59+static const struct soc15_ras_field_entry sdma_v4_4_ras_fields[] = {
60+ { "SDMA_MBANK_DATA_BUF0_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
61+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF0_SED),
62+ 0, 0,
63+ },
64+ { "SDMA_MBANK_DATA_BUF1_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
65+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF1_SED),
66+ 0, 0,
67+ },
68+ { "SDMA_MBANK_DATA_BUF2_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
69+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF2_SED),
70+ 0, 0,
71+ },
72+ { "SDMA_MBANK_DATA_BUF3_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
73+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF3_SED),
74+ 0, 0,
75+ },
76+ { "SDMA_MBANK_DATA_BUF4_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
77+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF4_SED),
78+ 0, 0,
79+ },
80+ { "SDMA_MBANK_DATA_BUF5_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
81+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF5_SED),
82+ 0, 0,
83+ },
84+ { "SDMA_MBANK_DATA_BUF6_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
85+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF6_SED),
86+ 0, 0,
87+ },
88+ { "SDMA_MBANK_DATA_BUF7_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
89+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF7_SED),
90+ 0, 0,
91+ },
92+ { "SDMA_MBANK_DATA_BUF8_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
93+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF8_SED),
94+ 0, 0,
95+ },
96+ { "SDMA_MBANK_DATA_BUF9_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
97+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF9_SED),
98+ 0, 0,
99+ },
100+ { "SDMA_MBANK_DATA_BUF10_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
101+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF10_SED),
102+ 0, 0,
103+ },
104+ { "SDMA_MBANK_DATA_BUF11_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
105+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF11_SED),
106+ 0, 0,
107+ },
108+ { "SDMA_MBANK_DATA_BUF12_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
109+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF12_SED),
110+ 0, 0,
111+ },
112+ { "SDMA_MBANK_DATA_BUF13_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
113+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF13_SED),
114+ 0, 0,
115+ },
116+ { "SDMA_MBANK_DATA_BUF14_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
117+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF14_SED),
118+ 0, 0,
119+ },
120+ { "SDMA_MBANK_DATA_BUF15_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER),
121+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF15_SED),
122+ 0, 0,
123+ },
124+ { "SDMA_UCODE_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
125+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_UCODE_BUF_SED),
126+ 0, 0,
127+ },
128+ { "SDMA_RB_CMD_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
129+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_RB_CMD_BUF_SED),
130+ 0, 0,
131+ },
132+ { "SDMA_IB_CMD_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
133+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_IB_CMD_BUF_SED),
134+ 0, 0,
135+ },
136+ { "SDMA_UTCL1_RD_FIFO_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
137+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_UTCL1_RD_FIFO_SED),
138+ 0, 0,
139+ },
140+ { "SDMA_UTCL1_RDBST_FIFO_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
141+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_UTCL1_RDBST_FIFO_SED),
142+ 0, 0,
143+ },
144+ { "SDMA_DATA_LUT_FIFO_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
145+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_DATA_LUT_FIFO_SED),
146+ 0, 0,
147+ },
148+ { "SDMA_SPLIT_DATA_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
149+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_SPLIT_DATA_BUF_SED),
150+ 0, 0,
151+ },
152+ { "SDMA_MC_WR_ADDR_FIFO_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
153+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_MC_WR_ADDR_FIFO_SED),
154+ 0, 0,
155+ },
156+ { "SDMA_MC_RDRET_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, regSDMA0_EDC_COUNTER2),
157+ SOC15_REG_FIELD(SDMA0_EDC_COUNTER2, SDMA_MC_WR_ADDR_FIFO_SED),
158+ 0, 0,
159+ },
160+};
161+
162+static void sdma_v4_4_get_ras_error_count(struct amdgpu_device *adev,
163+ uint32_t value,
164+ uint32_t instance,
165+ uint32_t *sec_count)
166+{
167+ uint32_t i;
168+ uint32_t sec_cnt;
169+
170+ /* double bits error (multiple bits) error detection is not supported */
171+ for (i = 0; i < ARRAY_SIZE(sdma_v4_4_ras_fields); i++) {
172+ /* the SDMA_EDC_COUNTER register in each sdma instance
173+ * shares the same sed shift_mask
174+ * */
175+ sec_cnt = (value &
176+ sdma_v4_4_ras_fields[i].sec_count_mask) >>
177+ sdma_v4_4_ras_fields[i].sec_count_shift;
178+ if (sec_cnt) {
179+ dev_info(adev->dev, "Detected %s in SDMA%d, SED %d\n",
180+ sdma_v4_4_ras_fields[i].name,
181+ instance, sec_cnt);
182+ *sec_count += sec_cnt;
183+ }
184+ }
185+}
186+
187+static int sdma_v4_4_query_ras_error_count(struct amdgpu_device *adev,
188+ uint32_t instance,
189+ void *ras_error_status)
190+{
191+ struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
192+ uint32_t sec_count = 0;
193+ uint32_t reg_value = 0;
194+ uint32_t reg_offset = 0;
195+
196+ reg_offset = sdma_v4_4_get_reg_offset(adev, instance, regSDMA0_EDC_COUNTER);
197+ reg_value = RREG32(reg_offset);
198+ /* double bit error is not supported */
199+ if (reg_value)
200+ sdma_v4_4_get_ras_error_count(adev, reg_value, instance, &sec_count);
201+ /* err_data->ce_count should be initialized to 0
202+ * before calling into this function */
203+ err_data->ce_count += sec_count;
204+ /* double bit error is not supported
205+ * set ue count to 0 */
206+ err_data->ue_count = 0;
207+
208+ return 0;
209+};
210+
211+static void sdma_v4_4_reset_ras_error_count(struct amdgpu_device *adev)
212+{
213+ int i;
214+ uint32_t reg_offset;
215+
216+ /* write 0 to EDC_COUNTER reg to clear sdma edc counters */
217+ if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA)) {
218+ for (i = 0; i < adev->sdma.num_instances; i++) {
219+ reg_offset = sdma_v4_4_get_reg_offset(adev, i, regSDMA0_EDC_COUNTER);
220+ WREG32(reg_offset, 0);
221+ reg_offset = sdma_v4_4_get_reg_offset(adev, i, regSDMA0_EDC_COUNTER2);
222+ WREG32(reg_offset, 0);
223+ }
224+ }
225+}
226+
227+const struct amdgpu_sdma_ras_funcs sdma_v4_4_ras_funcs = {
228+ .ras_late_init = amdgpu_sdma_ras_late_init,
229+ .ras_fini = amdgpu_sdma_ras_fini,
230+ .query_ras_error_count = sdma_v4_4_query_ras_error_count,
231+ .reset_ras_error_count = sdma_v4_4_reset_ras_error_count,
232+};
drivers/gpu/drm/amd/amdgpu/sdma_v4_4.h+28−0
@@ -0,0 +1,28 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#ifndef __SDMA_V4_4_H__
24+#define __SDMA_V4_4_H__
25+
26+extern const struct amdgpu_sdma_ras_funcs sdma_v4_4_ras_funcs;
27+
28+#endif
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c+2−2
@@ -1595,9 +1595,9 @@ static int sdma_v5_2_set_clockgating_state(void *handle, · 1594 unmodified lines
15951595 case CHIP_VANGOGH:
15961596 case CHIP_DIMGREY_CAVEFISH:
15971597 sdma_v5_2_update_medium_grain_clock_gating(adev,
1598− state == AMD_CG_STATE_GATE ? true : false);
1598+ state == AMD_CG_STATE_GATE);
15991599 sdma_v5_2_update_medium_grain_light_sleep(adev,
1600− state == AMD_CG_STATE_GATE ? true : false);
1600+ state == AMD_CG_STATE_GATE);
16011601 break;
16021602 default:
16031603 break;
drivers/gpu/drm/amd/amdgpu/si.c+111−0
@@ -26,6 +26,8 @@ · 25 unmodified lines
2626 #include <linux/module.h>
2727 #include <linux/pci.h>
2828
29+#include <drm/amdgpu_drm.h>
30+
2931 #include "amdgpu.h"
3032 #include "amdgpu_atombios.h"
3133 #include "amdgpu_ih.h"
@@ -905,6 +907,114 @@ static const u32 hainan_mgcg_cgcg_init[] = · 873 unmodified lines
905907 0x3630, 0xfffffff0, 0x00000100,
906908 };
907909
910+/* XXX: update when we support VCE */
911+#if 0
912+/* tahiti, pitcarin, verde */
913+static const struct amdgpu_video_codec_info tahiti_video_codecs_encode_array[] =
914+{
915+ {
916+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
917+ .max_width = 2048,
918+ .max_height = 1152,
919+ .max_pixels_per_frame = 2048 * 1152,
920+ .max_level = 0,
921+ },
922+};
923+
924+static const struct amdgpu_video_codecs tahiti_video_codecs_encode =
925+{
926+ .codec_count = ARRAY_SIZE(tahiti_video_codecs_encode_array),
927+ .codec_array = tahiti_video_codecs_encode_array,
928+};
929+#else
930+static const struct amdgpu_video_codecs tahiti_video_codecs_encode =
931+{
932+ .codec_count = 0,
933+ .codec_array = NULL,
934+};
935+#endif
936+/* oland and hainan don't support encode */
937+static const struct amdgpu_video_codecs hainan_video_codecs_encode =
938+{
939+ .codec_count = 0,
940+ .codec_array = NULL,
941+};
942+
943+/* tahiti, pitcarin, verde, oland */
944+static const struct amdgpu_video_codec_info tahiti_video_codecs_decode_array[] =
945+{
946+ {
947+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
948+ .max_width = 2048,
949+ .max_height = 1152,
950+ .max_pixels_per_frame = 2048 * 1152,
951+ .max_level = 3,
952+ },
953+ {
954+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
955+ .max_width = 2048,
956+ .max_height = 1152,
957+ .max_pixels_per_frame = 2048 * 1152,
958+ .max_level = 5,
959+ },
960+ {
961+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
962+ .max_width = 2048,
963+ .max_height = 1152,
964+ .max_pixels_per_frame = 2048 * 1152,
965+ .max_level = 41,
966+ },
967+ {
968+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
969+ .max_width = 2048,
970+ .max_height = 1152,
971+ .max_pixels_per_frame = 2048 * 1152,
972+ .max_level = 4,
973+ },
974+};
975+
976+static const struct amdgpu_video_codecs tahiti_video_codecs_decode =
977+{
978+ .codec_count = ARRAY_SIZE(tahiti_video_codecs_decode_array),
979+ .codec_array = tahiti_video_codecs_decode_array,
980+};
981+
982+/* hainan doesn't support decode */
983+static const struct amdgpu_video_codecs hainan_video_codecs_decode =
984+{
985+ .codec_count = 0,
986+ .codec_array = NULL,
987+};
988+
989+static int si_query_video_codecs(struct amdgpu_device *adev, bool encode,
990+ const struct amdgpu_video_codecs **codecs)
991+{
992+ switch (adev->asic_type) {
993+ case CHIP_VERDE:
994+ case CHIP_TAHITI:
995+ case CHIP_PITCAIRN:
996+ if (encode)
997+ *codecs = &tahiti_video_codecs_encode;
998+ else
999+ *codecs = &tahiti_video_codecs_decode;
1000+ return 0;
1001+ case CHIP_OLAND:
1002+ if (encode)
1003+ *codecs = &hainan_video_codecs_encode;
1004+ else
1005+ *codecs = &tahiti_video_codecs_decode;
1006+ return 0;
1007+ case CHIP_HAINAN:
1008+ if (encode)
1009+ *codecs = &hainan_video_codecs_encode;
1010+ else
1011+ *codecs = &hainan_video_codecs_decode;
1012+ return 0;
1013+ default:
1014+ return -EINVAL;
1015+ }
1016+}
1017+
9081018 static u32 si_pcie_rreg(struct amdgpu_device *adev, u32 reg)
9091019 {
9101020 unsigned long flags;
@@ -1903,6 +2013,7 @@ static const struct amdgpu_asic_funcs si_asic_funcs = · 992 unmodified lines
19032013 .get_pcie_replay_count = &si_get_pcie_replay_count,
19042014 .supports_baco = &si_asic_supports_baco,
19052015 .pre_asic_init = &si_pre_asic_init,
2016+ .query_video_codecs = &si_query_video_codecs,
19062017 };
19072018
19082019 static uint32_t si_get_rev_id(struct amdgpu_device *adev)
drivers/gpu/drm/amd/amdgpu/smuio_v13_0.c+121−0
@@ -0,0 +1,121 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#include "amdgpu.h"
24+#include "smuio_v13_0.h"
25+#include "smuio/smuio_13_0_2_offset.h"
26+#include "smuio/smuio_13_0_2_sh_mask.h"
27+
28+#define SMUIO_MCM_CONFIG__HOST_GPU_XGMI_MASK 0x00000001L
29+
30+static u32 smuio_v13_0_get_rom_index_offset(struct amdgpu_device *adev)
31+{
32+ return SOC15_REG_OFFSET(SMUIO, 0, regROM_INDEX);
33+}
34+
35+static u32 smuio_v13_0_get_rom_data_offset(struct amdgpu_device *adev)
36+{
37+ return SOC15_REG_OFFSET(SMUIO, 0, regROM_DATA);
38+}
39+
40+static void smuio_v13_0_update_rom_clock_gating(struct amdgpu_device *adev, bool enable)
41+{
42+ u32 def, data;
43+
44+ /* enable/disable ROM CG is not supported on APU */
45+ if (adev->flags & AMD_IS_APU)
46+ return;
47+
48+ def = data = RREG32_SOC15(SMUIO, 0, regCGTT_ROM_CLK_CTRL0);
49+
50+ if (enable && (adev->cg_flags & AMD_CG_SUPPORT_ROM_MGCG))
51+ data &= ~(CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE0_MASK |
52+ CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE1_MASK);
53+ else
54+ data |= CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE0_MASK |
55+ CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE1_MASK;
56+
57+ if (def != data)
58+ WREG32_SOC15(SMUIO, 0, regCGTT_ROM_CLK_CTRL0, data);
59+}
60+
61+static void smuio_v13_0_get_clock_gating_state(struct amdgpu_device *adev, u32 *flags)
62+{
63+ u32 data;
64+
65+ /* CGTT_ROM_CLK_CTRL0 is not available for APU */
66+ if (adev->flags & AMD_IS_APU)
67+ return;
68+
69+ data = RREG32_SOC15(SMUIO, 0, regCGTT_ROM_CLK_CTRL0);
70+ if (!(data & CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE0_MASK))
71+ *flags |= AMD_CG_SUPPORT_ROM_MGCG;
72+}
73+
74+/**
75+ * smuio_v13_0_get_die_id - query die id from FCH.
76+ *
77+ * @adev: amdgpu device pointer
78+ *
79+ * Returns die id
80+ */
81+static u32 smuio_v13_0_get_die_id(struct amdgpu_device *adev)
82+{
83+ u32 data, die_id;
84+
85+ data = RREG32_SOC15(SMUIO, 0, regSMUIO_MCM_CONFIG);
86+ die_id = REG_GET_FIELD(data, SMUIO_MCM_CONFIG, DIE_ID);
87+
88+ return die_id;
89+}
90+
91+/**
92+ * smuio_v13_0_supports_host_gpu_xgmi - detect xgmi interface between cpu and gpu/s.
93+ *
94+ * @adev: amdgpu device pointer
95+ *
96+ * Returns true on success or false otherwise.
97+ */
98+static bool smuio_v13_0_is_host_gpu_xgmi_supported(struct amdgpu_device *adev)
99+{
100+ u32 data;
101+
102+ data = RREG32_SOC15(SMUIO, 0, regSMUIO_MCM_CONFIG);
103+ data = REG_GET_FIELD(data, SMUIO_MCM_CONFIG, TOPOLOGY_ID);
104+ /* data[4:0]
105+ * bit 0 == 0 host-gpu interface is PCIE
106+ * bit 0 == 1 host-gpu interface is Alternate Protocal
107+ * for AMD, this is XGMI
108+ */
109+ data &= SMUIO_MCM_CONFIG__HOST_GPU_XGMI_MASK;
110+
111+ return data ? true : false;
112+}
113+
114+const struct amdgpu_smuio_funcs smuio_v13_0_funcs = {
115+ .get_rom_index_offset = smuio_v13_0_get_rom_index_offset,
116+ .get_rom_data_offset = smuio_v13_0_get_rom_data_offset,
117+ .get_die_id = smuio_v13_0_get_die_id,
118+ .is_host_gpu_xgmi_supported = smuio_v13_0_is_host_gpu_xgmi_supported,
119+ .update_rom_clock_gating = smuio_v13_0_update_rom_clock_gating,
120+ .get_clock_gating_state = smuio_v13_0_get_clock_gating_state,
121+};
drivers/gpu/drm/amd/amdgpu/smuio_v13_0.h+30−0
@@ -0,0 +1,30 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ */
23+#ifndef __SMUIO_V13_0_H__
24+#define __SMUIO_V13_0_H__
25+
26+#include "soc15_common.h"
27+
28+extern const struct amdgpu_smuio_funcs smuio_v13_0_funcs;
29+
30+#endif /* __SMUIO_V13_0_H__ */
drivers/gpu/drm/amd/amdgpu/soc15.c+310−50
@@ -25,6 +25,8 @@ · 24 unmodified lines
2525 #include <linux/module.h>
2626 #include <linux/pci.h>
2727
28+#include <drm/amdgpu_drm.h>
29+
2830 #include "amdgpu.h"
2931 #include "amdgpu_atombios.h"
3032 #include "amdgpu_ih.h"
@@ -71,6 +73,7 @@ · 40 unmodified lines
7173 #include "jpeg_v2_5.h"
7274 #include "smuio_v9_0.h"
7375 #include "smuio_v11_0.h"
76+#include "smuio_v13_0.h"
7477 #include "dce_virtual.h"
7578 #include "mxgpu_ai.h"
7679 #include "amdgpu_smu.h"
@@ -83,6 +86,234 @@ · 6 unmodified lines
8386 #define mmMP0_MISC_LIGHT_SLEEP_CTRL 0x01ba
8487 #define mmMP0_MISC_LIGHT_SLEEP_CTRL_BASE_IDX 0
8588
89+/* Vega, Raven, Arcturus */
90+static const struct amdgpu_video_codec_info vega_video_codecs_encode_array[] =
91+{
92+ {
93+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
94+ .max_width = 4096,
95+ .max_height = 2304,
96+ .max_pixels_per_frame = 4096 * 2304,
97+ .max_level = 0,
98+ },
99+ {
100+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
101+ .max_width = 4096,
102+ .max_height = 2304,
103+ .max_pixels_per_frame = 4096 * 2304,
104+ .max_level = 0,
105+ },
106+};
107+
108+static const struct amdgpu_video_codecs vega_video_codecs_encode =
109+{
110+ .codec_count = ARRAY_SIZE(vega_video_codecs_encode_array),
111+ .codec_array = vega_video_codecs_encode_array,
112+};
113+
114+/* Vega */
115+static const struct amdgpu_video_codec_info vega_video_codecs_decode_array[] =
116+{
117+ {
118+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
119+ .max_width = 4096,
120+ .max_height = 4096,
121+ .max_pixels_per_frame = 4096 * 4096,
122+ .max_level = 3,
123+ },
124+ {
125+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
126+ .max_width = 4096,
127+ .max_height = 4096,
128+ .max_pixels_per_frame = 4096 * 4096,
129+ .max_level = 5,
130+ },
131+ {
132+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
133+ .max_width = 4096,
134+ .max_height = 4096,
135+ .max_pixels_per_frame = 4096 * 4096,
136+ .max_level = 52,
137+ },
138+ {
139+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
140+ .max_width = 4096,
141+ .max_height = 4096,
142+ .max_pixels_per_frame = 4096 * 4096,
143+ .max_level = 4,
144+ },
145+ {
146+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
147+ .max_width = 4096,
148+ .max_height = 4096,
149+ .max_pixels_per_frame = 4096 * 4096,
150+ .max_level = 186,
151+ },
152+ {
153+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG,
154+ .max_width = 4096,
155+ .max_height = 4096,
156+ .max_pixels_per_frame = 4096 * 4096,
157+ .max_level = 0,
158+ },
159+};
160+
161+static const struct amdgpu_video_codecs vega_video_codecs_decode =
162+{
163+ .codec_count = ARRAY_SIZE(vega_video_codecs_decode_array),
164+ .codec_array = vega_video_codecs_decode_array,
165+};
166+
167+/* Raven */
168+static const struct amdgpu_video_codec_info rv_video_codecs_decode_array[] =
169+{
170+ {
171+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
172+ .max_width = 4096,
173+ .max_height = 4096,
174+ .max_pixels_per_frame = 4096 * 4096,
175+ .max_level = 3,
176+ },
177+ {
178+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
179+ .max_width = 4096,
180+ .max_height = 4096,
181+ .max_pixels_per_frame = 4096 * 4096,
182+ .max_level = 5,
183+ },
184+ {
185+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
186+ .max_width = 4096,
187+ .max_height = 4096,
188+ .max_pixels_per_frame = 4096 * 4096,
189+ .max_level = 52,
190+ },
191+ {
192+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
193+ .max_width = 4096,
194+ .max_height = 4096,
195+ .max_pixels_per_frame = 4096 * 4096,
196+ .max_level = 4,
197+ },
198+ {
199+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
200+ .max_width = 4096,
201+ .max_height = 4096,
202+ .max_pixels_per_frame = 4096 * 4096,
203+ .max_level = 186,
204+ },
205+ {
206+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG,
207+ .max_width = 4096,
208+ .max_height = 4096,
209+ .max_pixels_per_frame = 4096 * 4096,
210+ .max_level = 0,
211+ },
212+ {
213+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9,
214+ .max_width = 4096,
215+ .max_height = 4096,
216+ .max_pixels_per_frame = 4096 * 4096,
217+ .max_level = 0,
218+ },
219+};
220+
221+static const struct amdgpu_video_codecs rv_video_codecs_decode =
222+{
223+ .codec_count = ARRAY_SIZE(rv_video_codecs_decode_array),
224+ .codec_array = rv_video_codecs_decode_array,
225+};
226+
227+/* Renoir, Arcturus */
228+static const struct amdgpu_video_codec_info rn_video_codecs_decode_array[] =
229+{
230+ {
231+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
232+ .max_width = 4096,
233+ .max_height = 4096,
234+ .max_pixels_per_frame = 4096 * 4096,
235+ .max_level = 3,
236+ },
237+ {
238+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
239+ .max_width = 4096,
240+ .max_height = 4096,
241+ .max_pixels_per_frame = 4096 * 4096,
242+ .max_level = 5,
243+ },
244+ {
245+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
246+ .max_width = 4096,
247+ .max_height = 4096,
248+ .max_pixels_per_frame = 4096 * 4096,
249+ .max_level = 52,
250+ },
251+ {
252+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
253+ .max_width = 4096,
254+ .max_height = 4096,
255+ .max_pixels_per_frame = 4096 * 4096,
256+ .max_level = 4,
257+ },
258+ {
259+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
260+ .max_width = 8192,
261+ .max_height = 4352,
262+ .max_pixels_per_frame = 4096 * 4096,
263+ .max_level = 186,
264+ },
265+ {
266+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG,
267+ .max_width = 4096,
268+ .max_height = 4096,
269+ .max_pixels_per_frame = 4096 * 4096,
270+ .max_level = 0,
271+ },
272+ {
273+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9,
274+ .max_width = 8192,
275+ .max_height = 4352,
276+ .max_pixels_per_frame = 4096 * 4096,
277+ .max_level = 0,
278+ },
279+};
280+
281+static const struct amdgpu_video_codecs rn_video_codecs_decode =
282+{
283+ .codec_count = ARRAY_SIZE(rn_video_codecs_decode_array),
284+ .codec_array = rn_video_codecs_decode_array,
285+};
286+
287+static int soc15_query_video_codecs(struct amdgpu_device *adev, bool encode,
288+ const struct amdgpu_video_codecs **codecs)
289+{
290+ switch (adev->asic_type) {
291+ case CHIP_VEGA20:
292+ case CHIP_VEGA10:
293+ case CHIP_VEGA12:
294+ if (encode)
295+ *codecs = &vega_video_codecs_encode;
296+ else
297+ *codecs = &vega_video_codecs_decode;
298+ return 0;
299+ case CHIP_RAVEN:
300+ if (encode)
301+ *codecs = &vega_video_codecs_encode;
302+ else
303+ *codecs = &rv_video_codecs_decode;
304+ return 0;
305+ case CHIP_ARCTURUS:
306+ case CHIP_RENOIR:
307+ if (encode)
308+ *codecs = &vega_video_codecs_encode;
309+ else
310+ *codecs = &rn_video_codecs_decode;
311+ return 0;
312+ default:
313+ return -EINVAL;
314+ }
315+}
316+
86317 /*
87318 * Indirect registers accessor
88319 */
@@ -419,40 +650,6 @@ void soc15_program_register_sequence(struct amdgpu_device *adev, · 330 unmodified lines
419650
420651 }
421652
422−static int soc15_asic_mode1_reset(struct amdgpu_device *adev)
423−{
424− u32 i;
425− int ret = 0;
426−
427− amdgpu_atombios_scratch_regs_engine_hung(adev, true);
428−
429− dev_info(adev->dev, "GPU mode1 reset\n");
430−
431− /* disable BM */
432− pci_clear_master(adev->pdev);
433−
434− amdgpu_device_cache_pci_state(adev->pdev);
435−
436− ret = psp_gpu_reset(adev);
437− if (ret)
438− dev_err(adev->dev, "GPU mode1 reset failed\n");
439−
440− amdgpu_device_load_pci_state(adev->pdev);
441−
442− /* wait for asic to come out of reset */
443− for (i = 0; i < adev->usec_timeout; i++) {
444− u32 memsize = adev->nbio.funcs->get_memsize(adev);
445−
446− if (memsize != 0xffffffff)
447− break;
448− udelay(1);
449− }
450−
451− amdgpu_atombios_scratch_regs_engine_hung(adev, false);
452−
453− return ret;
454−}
455−
456653 static int soc15_asic_baco_reset(struct amdgpu_device *adev)
457654 {
458655 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
@@ -477,13 +674,21 @@ static enum amd_reset_method · 18 unmodified lines
477674 soc15_asic_reset_method(struct amdgpu_device *adev)
478675 {
479676 bool baco_reset = false;
677+ bool connected_to_cpu = false;
480678 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
481679
680+ if (adev->gmc.xgmi.supported && adev->gmc.xgmi.connected_to_cpu)
681+ connected_to_cpu = true;
682+
482683 if (amdgpu_reset_method == AMD_RESET_METHOD_MODE1 ||
483684 amdgpu_reset_method == AMD_RESET_METHOD_MODE2 ||
484685 amdgpu_reset_method == AMD_RESET_METHOD_BACO ||
485− amdgpu_reset_method == AMD_RESET_METHOD_PCI)
486− return amdgpu_reset_method;
686+ amdgpu_reset_method == AMD_RESET_METHOD_PCI) {
687+ /* If connected to cpu, driver only support mode2 */
688+ if (connected_to_cpu)
689+ return AMD_RESET_METHOD_MODE2;
690+ return amdgpu_reset_method;
691+ }
487692
488693 if (amdgpu_reset_method != -1)
489694 dev_warn(adev->dev, "Specified reset method:%d isn't supported, using AUTO instead.\n",
@@ -509,6 +714,14 @@ soc15_asic_reset_method(struct amdgpu_device *adev) · 19 unmodified lines
509714 if ((ras && ras->supported) && adev->pm.fw_version <= 0x283400)
510715 baco_reset = false;
511716 break;
717+ case CHIP_ALDEBARAN:
718+ /*
719+ * 1.connected to cpu: driver issue mode2 reset
720+ * 2.discret gpu: driver issue mode1 reset
721+ */
722+ if (connected_to_cpu)
723+ return AMD_RESET_METHOD_MODE2;
724+ break;
512725 default:
513726 break;
514727 }
@@ -538,7 +751,7 @@ static int soc15_asic_reset(struct amdgpu_device *adev) · 23 unmodified lines
538751 return amdgpu_dpm_mode2_reset(adev);
539752 default:
540753 dev_info(adev->dev, "MODE1 reset\n");
541− return soc15_asic_mode1_reset(adev);
754+ return amdgpu_device_mode1_reset(adev);
542755 }
543756 }
544757
@@ -661,6 +874,9 @@ static void soc15_reg_base_init(struct amdgpu_device *adev) · 116 unmodified lines
661874 case CHIP_ARCTURUS:
662875 arct_reg_base_init(adev);
663876 break;
877+ case CHIP_ALDEBARAN:
878+ aldebaran_reg_base_init(adev);
879+ break;
664880 default:
665881 DRM_ERROR("Unsupported asic type: %d!\n", adev->asic_type);
666882 break;
@@ -683,14 +899,12 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev) · 16 unmodified lines
683899 if (!amdgpu_sriov_vf(adev))
684900 soc15_reg_base_init(adev);
685901
686− if (adev->asic_type == CHIP_VEGA20 || adev->asic_type == CHIP_ARCTURUS)
687− adev->gmc.xgmi.supported = true;
688−
689902 if (adev->flags & AMD_IS_APU) {
690903 adev->nbio.funcs = &nbio_v7_0_funcs;
691904 adev->nbio.hdp_flush_reg = &nbio_v7_0_hdp_flush_reg;
692905 } else if (adev->asic_type == CHIP_VEGA20 ||
693− adev->asic_type == CHIP_ARCTURUS) {
906+ adev->asic_type == CHIP_ARCTURUS ||
907+ adev->asic_type == CHIP_ALDEBARAN) {
694908 adev->nbio.funcs = &nbio_v7_4_funcs;
695909 adev->nbio.hdp_flush_reg = &nbio_v7_4_hdp_flush_reg;
696910 } else {
@@ -699,7 +913,9 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev) · 2 unmodified lines
699913 }
700914 adev->hdp.funcs = &hdp_v4_0_funcs;
701915
702− if (adev->asic_type == CHIP_VEGA20 || adev->asic_type == CHIP_ARCTURUS)
916+ if (adev->asic_type == CHIP_VEGA20 ||
917+ adev->asic_type == CHIP_ARCTURUS ||
918+ adev->asic_type == CHIP_ALDEBARAN)
703919 adev->df.funcs = &df_v3_6_funcs;
704920 else
705921 adev->df.funcs = &df_v1_7_funcs;
@@ -707,6 +923,8 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev) · 1 unmodified line
707923 if (adev->asic_type == CHIP_VEGA20 ||
708924 adev->asic_type == CHIP_ARCTURUS)
709925 adev->smuio.funcs = &smuio_v11_0_funcs;
926+ else if (adev->asic_type == CHIP_ALDEBARAN)
927+ adev->smuio.funcs = &smuio_v13_0_funcs;
710928 else
711929 adev->smuio.funcs = &smuio_v9_0_funcs;
712930
@@ -826,6 +1044,27 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev) · 113 unmodified lines
8261044 amdgpu_device_ip_block_add(adev, &vcn_v2_0_ip_block);
8271045 amdgpu_device_ip_block_add(adev, &jpeg_v2_0_ip_block);
8281046 break;
1047+ case CHIP_ALDEBARAN:
1048+ amdgpu_device_ip_block_add(adev, &vega10_common_ip_block);
1049+ amdgpu_device_ip_block_add(adev, &gmc_v9_0_ip_block);
1050+
1051+ if (amdgpu_sriov_vf(adev)) {
1052+ if (likely(adev->firmware.load_type == AMDGPU_FW_LOAD_PSP))
1053+ amdgpu_device_ip_block_add(adev, &psp_v13_0_ip_block);
1054+ amdgpu_device_ip_block_add(adev, &vega20_ih_ip_block);
1055+ } else {
1056+ amdgpu_device_ip_block_add(adev, &vega20_ih_ip_block);
1057+ if (likely(adev->firmware.load_type == AMDGPU_FW_LOAD_PSP))
1058+ amdgpu_device_ip_block_add(adev, &psp_v13_0_ip_block);
1059+ }
1060+
1061+ amdgpu_device_ip_block_add(adev, &gfx_v9_0_ip_block);
1062+ amdgpu_device_ip_block_add(adev, &sdma_v4_0_ip_block);
1063+
1064+ amdgpu_device_ip_block_add(adev, &smu_v13_0_ip_block);
1065+ amdgpu_device_ip_block_add(adev, &vcn_v2_6_ip_block);
1066+ amdgpu_device_ip_block_add(adev, &jpeg_v2_6_ip_block);
1067+ break;
8291068 default:
8301069 return -EINVAL;
8311070 }
@@ -994,6 +1233,7 @@ static const struct amdgpu_asic_funcs soc15_asic_funcs = · 162 unmodified lines
9941233 .get_pcie_replay_count = &soc15_get_pcie_replay_count,
9951234 .supports_baco = &soc15_supports_baco,
9961235 .pre_asic_init = &soc15_pre_asic_init,
1236+ .query_video_codecs = &soc15_query_video_codecs,
9971237 };
9981238
9991239 static const struct amdgpu_asic_funcs vega20_asic_funcs =
@@ -1015,6 +1255,7 @@ static const struct amdgpu_asic_funcs vega20_asic_funcs = · 15 unmodified lines
10151255 .get_pcie_replay_count = &soc15_get_pcie_replay_count,
10161256 .supports_baco = &soc15_supports_baco,
10171257 .pre_asic_init = &soc15_pre_asic_init,
1258+ .query_video_codecs = &soc15_query_video_codecs,
10181259 };
10191260
10201261 static int soc15_common_early_init(void *handle)
@@ -1244,6 +1485,21 @@ static int soc15_common_early_init(void *handle) · 223 unmodified lines
12441485 AMD_PG_SUPPORT_JPEG |
12451486 AMD_PG_SUPPORT_VCN_DPG;
12461487 break;
1488+ case CHIP_ALDEBARAN:
1489+ adev->asic_funcs = &vega20_asic_funcs;
1490+ adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
1491+ AMD_CG_SUPPORT_GFX_MGLS |
1492+ AMD_CG_SUPPORT_GFX_CGCG |
1493+ AMD_CG_SUPPORT_GFX_CGLS |
1494+ AMD_CG_SUPPORT_GFX_CP_LS |
1495+ AMD_CG_SUPPORT_HDP_LS |
1496+ AMD_CG_SUPPORT_SDMA_MGCG |
1497+ AMD_CG_SUPPORT_SDMA_LS |
1498+ AMD_CG_SUPPORT_IH_CG;
1499+ /*AMD_CG_SUPPORT_VCN_MGCG |AMD_CG_SUPPORT_JPEG_MGCG;*/
1500+ adev->pg_flags = AMD_PG_SUPPORT_VCN_DPG;
1501+ adev->external_rev_id = adev->rev_id + 0x3c;
1502+ break;
12471503 default:
12481504 /* FIXME: not supported yet */
12491505 return -EINVAL;
@@ -1477,6 +1733,7 @@ static int soc15_common_set_clockgating_state(void *handle, · 227 unmodified lines
14771733 state == AMD_CG_STATE_GATE);
14781734 break;
14791735 case CHIP_ARCTURUS:
1736+ case CHIP_ALDEBARAN:
14801737 adev->hdp.funcs->update_clock_gating(adev,
14811738 state == AMD_CG_STATE_GATE);
14821739 break;
@@ -1498,15 +1755,18 @@ static void soc15_common_get_clockgating_state(void *handle, u32 *flags) · 15 unmodified lines
14981755
14991756 adev->hdp.funcs->get_clock_gating_state(adev, flags);
15001757
1501− /* AMD_CG_SUPPORT_DRM_MGCG */
1502− data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0));
1503− if (!(data & 0x01000000))
1504− *flags |= AMD_CG_SUPPORT_DRM_MGCG;
1758+ if (adev->asic_type != CHIP_ALDEBARAN) {
15051759
1506− /* AMD_CG_SUPPORT_DRM_LS */
1507− data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL));
1508− if (data & 0x1)
1509− *flags |= AMD_CG_SUPPORT_DRM_LS;
1760+ /* AMD_CG_SUPPORT_DRM_MGCG */
1761+ data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0));
1762+ if (!(data & 0x01000000))
1763+ *flags |= AMD_CG_SUPPORT_DRM_MGCG;
1764+
1765+ /* AMD_CG_SUPPORT_DRM_LS */
1766+ data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL));
1767+ if (data & 0x1)
1768+ *flags |= AMD_CG_SUPPORT_DRM_LS;
1769+ }
15101770
15111771 /* AMD_CG_SUPPORT_ROM_MGCG */
15121772 adev->smuio.funcs->get_clock_gating_state(adev, flags);
drivers/gpu/drm/amd/amdgpu/soc15.h+12−0
@@ -49,6 +49,13 @@ struct soc15_reg_rlcg { · 48 unmodified lines
4949 u32 reg;
5050 };
5151
52+struct soc15_reg {
53+ uint32_t hwip;
54+ uint32_t inst;
55+ uint32_t seg;
56+ uint32_t reg_offset;
57+};
58+
5259 struct soc15_reg_entry {
5360 uint32_t hwip;
5461 uint32_t inst;
@@ -88,6 +95,10 @@ struct soc15_ras_field_entry { · 33 unmodified lines
8895
8996 #define SOC15_REG_FIELD(reg, field) reg##__##field##_MASK, reg##__##field##__SHIFT
9097
98+#define SOC15_REG_FIELD_VAL(val, mask, shift) (((val) & mask) >> shift)
99+
100+#define SOC15_RAS_REG_FIELD_VAL(val, entry, field) SOC15_REG_FIELD_VAL((val), (entry).field##_count_mask, (entry).field##_count_shift)
101+
91102 void soc15_grbm_select(struct amdgpu_device *adev,
92103 u32 me, u32 pipe, u32 queue, u32 vmid);
93104 void soc15_set_virt_ops(struct amdgpu_device *adev);
@@ -100,6 +111,7 @@ void soc15_program_register_sequence(struct amdgpu_device *adev, · 6 unmodified lines
100111 int vega10_reg_base_init(struct amdgpu_device *adev);
101112 int vega20_reg_base_init(struct amdgpu_device *adev);
102113 int arct_reg_base_init(struct amdgpu_device *adev);
114+int aldebaran_reg_base_init(struct amdgpu_device *adev);
103115
104116 void vega10_doorbell_index_init(struct amdgpu_device *adev);
105117 void vega20_doorbell_index_init(struct amdgpu_device *adev);
drivers/gpu/drm/amd/amdgpu/soc15_common.h+48−0
@@ -100,6 +100,30 @@ · 99 unmodified lines
100100 } \
101101 } while (0)
102102
103+#define WREG32_RLC_EX(prefix, reg, value) \
104+ do { \
105+ if (amdgpu_sriov_fullaccess(adev)) { \
106+ uint32_t i = 0; \
107+ uint32_t retries = 50000; \
108+ uint32_t r0 = adev->reg_offset[GC_HWIP][0][prefix##SCRATCH_REG0_BASE_IDX] + prefix##SCRATCH_REG0; \
109+ uint32_t r1 = adev->reg_offset[GC_HWIP][0][prefix##SCRATCH_REG1_BASE_IDX] + prefix##SCRATCH_REG1; \
110+ uint32_t spare_int = adev->reg_offset[GC_HWIP][0][prefix##RLC_SPARE_INT_BASE_IDX] + prefix##RLC_SPARE_INT; \
111+ WREG32(r0, value); \
112+ WREG32(r1, (reg | 0x80000000)); \
113+ WREG32(spare_int, 0x1); \
114+ for (i = 0; i < retries; i++) { \
115+ u32 tmp = RREG32(r1); \
116+ if (!(tmp & 0x80000000)) \
117+ break; \
118+ udelay(10); \
119+ } \
120+ if (i >= retries) \
121+ pr_err("timeout: rlcg program reg:0x%05x failed !\n", reg); \
122+ } else { \
123+ WREG32(reg, value); \
124+ } \
125+ } while (0)
126+
103127 #define WREG32_SOC15_RLC_SHADOW(ip, inst, reg, value) \
104128 do { \
105129 uint32_t target_reg = adev->reg_offset[ip##_HWIP][inst][reg##_BASE_IDX] + reg;\
@@ -118,12 +142,36 @@ · 12 unmodified lines
118142 } \
119143 } while (0)
120144
145+#define WREG32_SOC15_RLC_SHADOW_EX(prefix, ip, inst, reg, value) \
146+ do { \
147+ uint32_t target_reg = adev->reg_offset[ip##_HWIP][inst][reg##_BASE_IDX] + reg;\
148+ if (amdgpu_sriov_fullaccess(adev)) { \
149+ uint32_t r2 = adev->reg_offset[GC_HWIP][0][prefix##SCRATCH_REG1_BASE_IDX] + prefix##SCRATCH_REG2; \
150+ uint32_t r3 = adev->reg_offset[GC_HWIP][0][prefix##SCRATCH_REG1_BASE_IDX] + prefix##SCRATCH_REG3; \
151+ uint32_t grbm_cntl = adev->reg_offset[GC_HWIP][0][prefix##GRBM_GFX_CNTL_BASE_IDX] + prefix##GRBM_GFX_CNTL; \
152+ uint32_t grbm_idx = adev->reg_offset[GC_HWIP][0][prefix##GRBM_GFX_INDEX_BASE_IDX] + prefix##GRBM_GFX_INDEX; \
153+ if (target_reg == grbm_cntl) \
154+ WREG32(r2, value); \
155+ else if (target_reg == grbm_idx) \
156+ WREG32(r3, value); \
157+ WREG32(target_reg, value); \
158+ } else { \
159+ WREG32(target_reg, value); \
160+ } \
161+ } while (0)
162+
121163 #define WREG32_SOC15_RLC(ip, inst, reg, value) \
122164 do { \
123165 uint32_t target_reg = adev->reg_offset[GC_HWIP][0][reg##_BASE_IDX] + reg;\
124166 WREG32_RLC(target_reg, value); \
125167 } while (0)
126168
169+#define WREG32_SOC15_RLC_EX(prefix, ip, inst, reg, value) \
170+ do { \
171+ uint32_t target_reg = adev->reg_offset[GC_HWIP][0][reg##_BASE_IDX] + reg;\
172+ WREG32_RLC_EX(prefix, target_reg, value); \
173+ } while (0)
174+
127175 #define WREG32_FIELD15_RLC(ip, idx, reg, field, val) \
128176 WREG32_RLC((adev->reg_offset[ip##_HWIP][idx][mm##reg##_BASE_IDX] + mm##reg), \
129177 (RREG32(adev->reg_offset[ip##_HWIP][idx][mm##reg##_BASE_IDX] + mm##reg) \
drivers/gpu/drm/amd/amdgpu/ta_secureDisplay_if.h+1−0
@@ -50,6 +50,7 @@ enum ta_securedisplay_status { · 49 unmodified lines
5050 TA_SECUREDISPLAY_STATUS__I2C_WRITE_ERROR = 0x04, /* Fail to Write to I2C */
5151 TA_SECUREDISPLAY_STATUS__READ_DIO_SCRATCH_ERROR = 0x05, /*Fail Read DIO Scratch Register*/
5252 TA_SECUREDISPLAY_STATUS__READ_CRC_ERROR = 0x06, /* Fail to Read CRC*/
53+ TA_SECUREDISPLAY_STATUS__I2C_INIT_ERROR = 0x07, /* Failed to initialize I2C */
5354
5455 TA_SECUREDISPLAY_STATUS__MAX = 0x7FFFFFFF,/* Maximum Value for status*/
5556 };
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c+97−2
@@ -1545,6 +1545,36 @@ static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = { · 1544 unmodified lines
15451545 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
15461546 };
15471547
1548+static const struct amdgpu_ring_funcs vcn_v2_6_dec_ring_vm_funcs = {
1549+ .type = AMDGPU_RING_TYPE_VCN_DEC,
1550+ .align_mask = 0xf,
1551+ .vmhub = AMDGPU_MMHUB_0,
1552+ .get_rptr = vcn_v2_5_dec_ring_get_rptr,
1553+ .get_wptr = vcn_v2_5_dec_ring_get_wptr,
1554+ .set_wptr = vcn_v2_5_dec_ring_set_wptr,
1555+ .emit_frame_size =
1556+ SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1557+ SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1558+ 8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1559+ 14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1560+ 6,
1561+ .emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1562+ .emit_ib = vcn_v2_0_dec_ring_emit_ib,
1563+ .emit_fence = vcn_v2_0_dec_ring_emit_fence,
1564+ .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1565+ .test_ring = vcn_v2_0_dec_ring_test_ring,
1566+ .test_ib = amdgpu_vcn_dec_ring_test_ib,
1567+ .insert_nop = vcn_v2_0_dec_ring_insert_nop,
1568+ .insert_start = vcn_v2_0_dec_ring_insert_start,
1569+ .insert_end = vcn_v2_0_dec_ring_insert_end,
1570+ .pad_ib = amdgpu_ring_generic_pad_ib,
1571+ .begin_use = amdgpu_vcn_ring_begin_use,
1572+ .end_use = amdgpu_vcn_ring_end_use,
1573+ .emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1574+ .emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1575+ .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1576+};
1577+
15481578 /**
15491579 * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
15501580 *
@@ -1644,6 +1674,36 @@ static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = { · 93 unmodified lines
16441674 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
16451675 };
16461676
1677+static const struct amdgpu_ring_funcs vcn_v2_6_enc_ring_vm_funcs = {
1678+ .type = AMDGPU_RING_TYPE_VCN_ENC,
1679+ .align_mask = 0x3f,
1680+ .nop = VCN_ENC_CMD_NO_OP,
1681+ .vmhub = AMDGPU_MMHUB_0,
1682+ .get_rptr = vcn_v2_5_enc_ring_get_rptr,
1683+ .get_wptr = vcn_v2_5_enc_ring_get_wptr,
1684+ .set_wptr = vcn_v2_5_enc_ring_set_wptr,
1685+ .emit_frame_size =
1686+ SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1687+ SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1688+ 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1689+ 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1690+ 1, /* vcn_v2_0_enc_ring_insert_end */
1691+ .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1692+ .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1693+ .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1694+ .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1695+ .test_ring = amdgpu_vcn_enc_ring_test_ring,
1696+ .test_ib = amdgpu_vcn_enc_ring_test_ib,
1697+ .insert_nop = amdgpu_ring_insert_nop,
1698+ .insert_end = vcn_v2_0_enc_ring_insert_end,
1699+ .pad_ib = amdgpu_ring_generic_pad_ib,
1700+ .begin_use = amdgpu_vcn_ring_begin_use,
1701+ .end_use = amdgpu_vcn_ring_end_use,
1702+ .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1703+ .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1704+ .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1705+};
1706+
16471707 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
16481708 {
16491709 int i;
@@ -1651,7 +1711,10 @@ static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev) · 1 unmodified line
16511711 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
16521712 if (adev->vcn.harvest_config & (1 << i))
16531713 continue;
1654− adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1714+ if (adev->asic_type == CHIP_ARCTURUS)
1715+ adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1716+ else /* CHIP_ALDEBARAN */
1717+ adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_6_dec_ring_vm_funcs;
16551718 adev->vcn.inst[i].ring_dec.me = i;
16561719 DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
16571720 }
@@ -1665,7 +1728,10 @@ static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev) · 7 unmodified lines
16651728 if (adev->vcn.harvest_config & (1 << j))
16661729 continue;
16671730 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1668− adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1731+ if (adev->asic_type == CHIP_ARCTURUS)
1732+ adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1733+ else /* CHIP_ALDEBARAN */
1734+ adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_6_enc_ring_vm_funcs;
16691735 adev->vcn.inst[j].ring_enc[i].me = j;
16701736 }
16711737 DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
@@ -1830,6 +1896,26 @@ static const struct amd_ip_funcs vcn_v2_5_ip_funcs = { · 158 unmodified lines
18301896 .set_powergating_state = vcn_v2_5_set_powergating_state,
18311897 };
18321898
1899+static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
1900+ .name = "vcn_v2_6",
1901+ .early_init = vcn_v2_5_early_init,
1902+ .late_init = NULL,
1903+ .sw_init = vcn_v2_5_sw_init,
1904+ .sw_fini = vcn_v2_5_sw_fini,
1905+ .hw_init = vcn_v2_5_hw_init,
1906+ .hw_fini = vcn_v2_5_hw_fini,
1907+ .suspend = vcn_v2_5_suspend,
1908+ .resume = vcn_v2_5_resume,
1909+ .is_idle = vcn_v2_5_is_idle,
1910+ .wait_for_idle = vcn_v2_5_wait_for_idle,
1911+ .check_soft_reset = NULL,
1912+ .pre_soft_reset = NULL,
1913+ .soft_reset = NULL,
1914+ .post_soft_reset = NULL,
1915+ .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1916+ .set_powergating_state = vcn_v2_5_set_powergating_state,
1917+};
1918+
18331919 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
18341920 {
18351921 .type = AMD_IP_BLOCK_TYPE_VCN,
@@ -1838,3 +1924,12 @@ const struct amdgpu_ip_block_version vcn_v2_5_ip_block = · 2 unmodified lines
18381924 .rev = 0,
18391925 .funcs = &vcn_v2_5_ip_funcs,
18401926 };
1927+
1928+const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
1929+{
1930+ .type = AMD_IP_BLOCK_TYPE_VCN,
1931+ .major = 2,
1932+ .minor = 6,
1933+ .rev = 0,
1934+ .funcs = &vcn_v2_6_ip_funcs,
1935+};
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.h+1−0
@@ -25,5 +25,6 @@ · 24 unmodified lines
2525 #define __VCN_V2_5_H__
2626
2727 extern const struct amdgpu_ip_block_version vcn_v2_5_ip_block;
28+extern const struct amdgpu_ip_block_version vcn_v2_6_ip_block;
2829
2930 #endif /* __VCN_V2_5_H__ */
drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c+23−1
@@ -238,7 +238,8 @@ static int vcn_v3_0_sw_init(void *handle) · 237 unmodified lines
238238
239239 fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
240240 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
241− cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
241+ cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG) |
242+ cpu_to_le32(AMDGPU_VCN_FW_SHARED_FLAG_0_RB);
242243 fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
243244 }
244245
@@ -1074,7 +1075,13 @@ static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, boo · 829 unmodified lines
10741075 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
10751076 lower_32_bits(ring->wptr));
10761077
1078+ /* Reset FW shared memory RBC WPTR/RPTR */
1079+ fw_shared->rb.rptr = 0;
1080+ fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1081+
1082+ /*resetting done, fw can check RB ring */
10771083 fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1084+
10781085 /* Unstall DPG */
10791086 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
10801087 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
@@ -1239,9 +1246,11 @@ static int vcn_v3_0_start(struct amdgpu_device *adev) · 158 unmodified lines
12391246 /* Initialize the ring buffer's read and write pointers */
12401247 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
12411248
1249+ WREG32_SOC15(VCN, i, mmUVD_SCRATCH2, 0);
12421250 ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
12431251 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
12441252 lower_32_bits(ring->wptr));
1253+ fw_shared->rb.wptr = lower_32_bits(ring->wptr);
12451254 fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
12461255
12471256 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
@@ -1662,6 +1671,10 @@ static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev, · 414 unmodified lines
16621671 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
16631672 fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
16641673
1674+ /* restore wptr/rptr with pointers saved in FW shared memory*/
1675+ WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, fw_shared->rb.rptr);
1676+ WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, fw_shared->rb.wptr);
1677+
16651678 /* Unstall DPG */
16661679 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
16671680 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
@@ -1721,6 +1734,15 @@ static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring) · 53 unmodified lines
17211734 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
17221735 {
17231736 struct amdgpu_device *adev = ring->adev;
1737+ volatile struct amdgpu_fw_shared *fw_shared;
1738+
1739+ if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1740+ /*whenever update RBC_RB_WPTR, we save the wptr in shared rb.wptr and scratch2 */
1741+ fw_shared = adev->vcn.inst[ring->me].fw_shared_cpu_addr;
1742+ fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1743+ WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1744+ lower_32_bits(ring->wptr));
1745+ }
17241746
17251747 if (ring->use_doorbell) {
17261748 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
drivers/gpu/drm/amd/amdgpu/vega10_ih.c+11−21
@@ -220,10 +220,8 @@ static int vega10_ih_enable_ring(struct amdgpu_device *adev, · 219 unmodified lines
220220 tmp = vega10_ih_rb_cntl(ih, tmp);
221221 if (ih == &adev->irq.ih)
222222 tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RPTR_REARM, !!adev->irq.msi_enabled);
223− if (ih == &adev->irq.ih1) {
224− tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 0);
223+ if (ih == &adev->irq.ih1)
225224 tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_FULL_DRAIN_ENABLE, 1);
226− }
227225 if (amdgpu_sriov_vf(adev)) {
228226 if (psp_reg_program(&adev->psp, ih_regs->psp_reg_id, tmp)) {
229227 dev_err(adev->dev, "PSP program IH_RB_CNTL failed!\n");
@@ -265,7 +263,6 @@ static int vega10_ih_irq_init(struct amdgpu_device *adev) · 35 unmodified lines
265263 u32 ih_chicken;
266264 int ret;
267265 int i;
268− u32 tmp;
269266
270267 /* disable irqs */
271268 ret = vega10_ih_toggle_interrupts(adev, false);
@@ -291,15 +288,6 @@ static int vega10_ih_irq_init(struct amdgpu_device *adev) · 19 unmodified lines
291288 }
292289 }
293290
294− tmp = RREG32_SOC15(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL);
295− tmp = REG_SET_FIELD(tmp, IH_STORM_CLIENT_LIST_CNTL,
296− CLIENT18_IS_STORM_CLIENT, 1);
297− WREG32_SOC15(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL, tmp);
298−
299− tmp = RREG32_SOC15(OSSSYS, 0, mmIH_INT_FLOOD_CNTL);
300− tmp = REG_SET_FIELD(tmp, IH_INT_FLOOD_CNTL, FLOOD_CNTL_ENABLE, 1);
301− WREG32_SOC15(OSSSYS, 0, mmIH_INT_FLOOD_CNTL, tmp);
302−
303291 pci_set_master(adev->pdev);
304292
305293 /* enable interrupts */
@@ -345,11 +333,17 @@ static u32 vega10_ih_get_wptr(struct amdgpu_device *adev, · 39 unmodified lines
345333 u32 wptr, tmp;
346334 struct amdgpu_ih_regs *ih_regs;
347335
348− wptr = le32_to_cpu(*ih->wptr_cpu);
349− ih_regs = &ih->ih_regs;
336+ if (ih == &adev->irq.ih) {
337+ /* Only ring0 supports writeback. On other rings fall back
338+ * to register-based code with overflow checking below.
339+ */
340+ wptr = le32_to_cpu(*ih->wptr_cpu);
350341
351− if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW))
352− goto out;
342+ if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW))
343+ goto out;
344+ }
345+
346+ ih_regs = &ih->ih_regs;
353347
354348 /* Double check that the overflow wasn't already cleared. */
355349 wptr = RREG32_NO_KIQ(ih_regs->ih_rb_wptr);
@@ -440,15 +434,11 @@ static int vega10_ih_self_irq(struct amdgpu_device *adev, · 84 unmodified lines
440434 struct amdgpu_irq_src *source,
441435 struct amdgpu_iv_entry *entry)
442436 {
443− uint32_t wptr = cpu_to_le32(entry->src_data[0]);
444−
445437 switch (entry->ring_id) {
446438 case 1:
447− *adev->irq.ih1.wptr_cpu = wptr;
448439 schedule_work(&adev->irq.ih1_work);
449440 break;
450441 case 2:
451− *adev->irq.ih2.wptr_cpu = wptr;
452442 schedule_work(&adev->irq.ih2_work);
453443 break;
454444 default: break;
drivers/gpu/drm/amd/amdgpu/vega20_ih.c+28−21
@@ -35,6 +35,9 @@ · 34 unmodified lines
3535
3636 #define MAX_REARM_RETRY 10
3737
38+#define mmIH_CHICKEN_ALDEBARAN 0x18d
39+#define mmIH_CHICKEN_ALDEBARAN_BASE_IDX 0
40+
3841 static void vega20_ih_set_interrupt_funcs(struct amdgpu_device *adev);
3942
4043 /**
@@ -104,6 +107,8 @@ static int vega20_ih_toggle_ring_interrupts(struct amdgpu_device *adev, · 63 unmodified lines
104107
105108 tmp = RREG32(ih_regs->ih_rb_cntl);
106109 tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_ENABLE, (enable ? 1 : 0));
110+ tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_GPU_TS_ENABLE, 1);
111+
107112 /* enable_intr field is only valid in ring0 */
108113 if (ih == &adev->irq.ih)
109114 tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, ENABLE_INTR, (enable ? 1 : 0));
@@ -220,10 +225,8 @@ static int vega20_ih_enable_ring(struct amdgpu_device *adev, · 110 unmodified lines
220225 tmp = vega20_ih_rb_cntl(ih, tmp);
221226 if (ih == &adev->irq.ih)
222227 tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RPTR_REARM, !!adev->irq.msi_enabled);
223− if (ih == &adev->irq.ih1) {
224− tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 0);
228+ if (ih == &adev->irq.ih1)
225229 tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_FULL_DRAIN_ENABLE, 1);
226− }
227230 if (amdgpu_sriov_vf(adev)) {
228231 if (psp_reg_program(&adev->psp, ih_regs->psp_reg_id, tmp)) {
229232 dev_err(adev->dev, "PSP program IH_RB_CNTL failed!\n");
@@ -297,7 +300,6 @@ static int vega20_ih_irq_init(struct amdgpu_device *adev) · 67 unmodified lines
297300 u32 ih_chicken;
298301 int ret;
299302 int i;
300− u32 tmp;
301303
302304 /* disable irqs */
303305 ret = vega20_ih_toggle_interrupts(adev, false);
@@ -316,6 +318,18 @@ static int vega20_ih_irq_init(struct amdgpu_device *adev) · 12 unmodified lines
316318 WREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN, ih_chicken);
317319 }
318320
321+ /* psp firmware won't program IH_CHICKEN for aldebaran
322+ * driver needs to program it properly according to
323+ * MC_SPACE type in IH_RB_CNTL */
324+ if (adev->asic_type == CHIP_ALDEBARAN) {
325+ ih_chicken = RREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN_ALDEBARAN);
326+ if (adev->irq.ih.use_bus_addr) {
327+ ih_chicken = REG_SET_FIELD(ih_chicken, IH_CHICKEN,
328+ MC_SPACE_GPA_ENABLE, 1);
329+ }
330+ WREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN_ALDEBARAN, ih_chicken);
331+ }
332+
319333 for (i = 0; i < ARRAY_SIZE(ih); i++) {
320334 if (ih[i]->ring_size) {
321335 if (i == 1)
@@ -326,15 +340,6 @@ static int vega20_ih_irq_init(struct amdgpu_device *adev) · 4 unmodified lines
326340 }
327341 }
328342
329− tmp = RREG32_SOC15(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL);
330− tmp = REG_SET_FIELD(tmp, IH_STORM_CLIENT_LIST_CNTL,
331− CLIENT18_IS_STORM_CLIENT, 1);
332− WREG32_SOC15(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL, tmp);
333−
334− tmp = RREG32_SOC15(OSSSYS, 0, mmIH_INT_FLOOD_CNTL);
335− tmp = REG_SET_FIELD(tmp, IH_INT_FLOOD_CNTL, FLOOD_CNTL_ENABLE, 1);
336− WREG32_SOC15(OSSSYS, 0, mmIH_INT_FLOOD_CNTL, tmp);
337−
338343 pci_set_master(adev->pdev);
339344
340345 /* enable interrupts */
@@ -380,11 +385,17 @@ static u32 vega20_ih_get_wptr(struct amdgpu_device *adev, · 39 unmodified lines
380385 u32 wptr, tmp;
381386 struct amdgpu_ih_regs *ih_regs;
382387
383− wptr = le32_to_cpu(*ih->wptr_cpu);
384− ih_regs = &ih->ih_regs;
388+ if (ih == &adev->irq.ih) {
389+ /* Only ring0 supports writeback. On other rings fall back
390+ * to register-based code with overflow checking below.
391+ */
392+ wptr = le32_to_cpu(*ih->wptr_cpu);
385393
386− if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW))
387− goto out;
394+ if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW))
395+ goto out;
396+ }
397+
398+ ih_regs = &ih->ih_regs;
388399
389400 /* Double check that the overflow wasn't already cleared. */
390401 wptr = RREG32_NO_KIQ(ih_regs->ih_rb_wptr);
@@ -476,15 +487,11 @@ static int vega20_ih_self_irq(struct amdgpu_device *adev, · 85 unmodified lines
476487 struct amdgpu_irq_src *source,
477488 struct amdgpu_iv_entry *entry)
478489 {
479− uint32_t wptr = cpu_to_le32(entry->src_data[0]);
480−
481490 switch (entry->ring_id) {
482491 case 1:
483− *adev->irq.ih1.wptr_cpu = wptr;
484492 schedule_work(&adev->irq.ih1_work);
485493 break;
486494 case 2:
487− *adev->irq.ih2.wptr_cpu = wptr;
488495 schedule_work(&adev->irq.ih2_work);
489496 break;
490497 default: break;
drivers/gpu/drm/amd/amdgpu/vi.c+190−0
@@ -24,6 +24,8 @@ · 23 unmodified lines
2424 #include <linux/pci.h>
2525 #include <linux/slab.h>
2626
27+#include <drm/amdgpu_drm.h>
28+
2729 #include "amdgpu.h"
2830 #include "amdgpu_atombios.h"
2931 #include "amdgpu_ih.h"
@@ -79,6 +81,193 @@ · 49 unmodified lines
7981 #include "mxgpu_vi.h"
8082 #include "amdgpu_dm.h"
8183
84+/* Topaz */
85+static const struct amdgpu_video_codecs topaz_video_codecs_encode =
86+{
87+ .codec_count = 0,
88+ .codec_array = NULL,
89+};
90+
91+/* Tonga, CZ, ST, Fiji */
92+static const struct amdgpu_video_codec_info tonga_video_codecs_encode_array[] =
93+{
94+ {
95+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
96+ .max_width = 4096,
97+ .max_height = 2304,
98+ .max_pixels_per_frame = 4096 * 2304,
99+ .max_level = 0,
100+ },
101+};
102+
103+static const struct amdgpu_video_codecs tonga_video_codecs_encode =
104+{
105+ .codec_count = ARRAY_SIZE(tonga_video_codecs_encode_array),
106+ .codec_array = tonga_video_codecs_encode_array,
107+};
108+
109+/* Polaris */
110+static const struct amdgpu_video_codec_info polaris_video_codecs_encode_array[] =
111+{
112+ {
113+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
114+ .max_width = 4096,
115+ .max_height = 2304,
116+ .max_pixels_per_frame = 4096 * 2304,
117+ .max_level = 0,
118+ },
119+ {
120+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
121+ .max_width = 4096,
122+ .max_height = 2304,
123+ .max_pixels_per_frame = 4096 * 2304,
124+ .max_level = 0,
125+ },
126+};
127+
128+static const struct amdgpu_video_codecs polaris_video_codecs_encode =
129+{
130+ .codec_count = ARRAY_SIZE(polaris_video_codecs_encode_array),
131+ .codec_array = polaris_video_codecs_encode_array,
132+};
133+
134+/* Topaz */
135+static const struct amdgpu_video_codecs topaz_video_codecs_decode =
136+{
137+ .codec_count = 0,
138+ .codec_array = NULL,
139+};
140+
141+/* Tonga */
142+static const struct amdgpu_video_codec_info tonga_video_codecs_decode_array[] =
143+{
144+ {
145+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
146+ .max_width = 4096,
147+ .max_height = 4096,
148+ .max_pixels_per_frame = 4096 * 4096,
149+ .max_level = 3,
150+ },
151+ {
152+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
153+ .max_width = 4096,
154+ .max_height = 4096,
155+ .max_pixels_per_frame = 4096 * 4096,
156+ .max_level = 5,
157+ },
158+ {
159+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
160+ .max_width = 4096,
161+ .max_height = 4096,
162+ .max_pixels_per_frame = 4096 * 4096,
163+ .max_level = 52,
164+ },
165+ {
166+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
167+ .max_width = 4096,
168+ .max_height = 4096,
169+ .max_pixels_per_frame = 4096 * 4096,
170+ .max_level = 4,
171+ },
172+};
173+
174+static const struct amdgpu_video_codecs tonga_video_codecs_decode =
175+{
176+ .codec_count = ARRAY_SIZE(tonga_video_codecs_decode_array),
177+ .codec_array = tonga_video_codecs_decode_array,
178+};
179+
180+/* CZ, ST, Fiji, Polaris */
181+static const struct amdgpu_video_codec_info cz_video_codecs_decode_array[] =
182+{
183+ {
184+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2,
185+ .max_width = 4096,
186+ .max_height = 4096,
187+ .max_pixels_per_frame = 4096 * 4096,
188+ .max_level = 3,
189+ },
190+ {
191+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4,
192+ .max_width = 4096,
193+ .max_height = 4096,
194+ .max_pixels_per_frame = 4096 * 4096,
195+ .max_level = 5,
196+ },
197+ {
198+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC,
199+ .max_width = 4096,
200+ .max_height = 4096,
201+ .max_pixels_per_frame = 4096 * 4096,
202+ .max_level = 52,
203+ },
204+ {
205+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1,
206+ .max_width = 4096,
207+ .max_height = 4096,
208+ .max_pixels_per_frame = 4096 * 4096,
209+ .max_level = 4,
210+ },
211+ {
212+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC,
213+ .max_width = 4096,
214+ .max_height = 4096,
215+ .max_pixels_per_frame = 4096 * 4096,
216+ .max_level = 186,
217+ },
218+ {
219+ .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG,
220+ .max_width = 4096,
221+ .max_height = 4096,
222+ .max_pixels_per_frame = 4096 * 4096,
223+ .max_level = 0,
224+ },
225+};
226+
227+static const struct amdgpu_video_codecs cz_video_codecs_decode =
228+{
229+ .codec_count = ARRAY_SIZE(cz_video_codecs_decode_array),
230+ .codec_array = cz_video_codecs_decode_array,
231+};
232+
233+static int vi_query_video_codecs(struct amdgpu_device *adev, bool encode,
234+ const struct amdgpu_video_codecs **codecs)
235+{
236+ switch (adev->asic_type) {
237+ case CHIP_TOPAZ:
238+ if (encode)
239+ *codecs = &topaz_video_codecs_encode;
240+ else
241+ *codecs = &topaz_video_codecs_decode;
242+ return 0;
243+ case CHIP_TONGA:
244+ if (encode)
245+ *codecs = &tonga_video_codecs_encode;
246+ else
247+ *codecs = &tonga_video_codecs_decode;
248+ return 0;
249+ case CHIP_POLARIS10:
250+ case CHIP_POLARIS11:
251+ case CHIP_POLARIS12:
252+ case CHIP_VEGAM:
253+ if (encode)
254+ *codecs = &polaris_video_codecs_encode;
255+ else
256+ *codecs = &cz_video_codecs_decode;
257+ return 0;
258+ case CHIP_FIJI:
259+ case CHIP_CARRIZO:
260+ case CHIP_STONEY:
261+ if (encode)
262+ *codecs = &tonga_video_codecs_encode;
263+ else
264+ *codecs = &cz_video_codecs_decode;
265+ return 0;
266+ default:
267+ return -EINVAL;
268+ }
269+}
270+
82271 /*
83272 * Indirect registers accessor
84273 */
@@ -1085,6 +1274,7 @@ static const struct amdgpu_asic_funcs vi_asic_funcs = · 1000 unmodified lines
10851274 .get_pcie_replay_count = &vi_get_pcie_replay_count,
10861275 .supports_baco = &vi_asic_supports_baco,
10871276 .pre_asic_init = &vi_pre_asic_init,
1277+ .query_video_codecs = &vi_query_video_codecs,
10881278 };
10891279
10901280 #define CZ_REV_BRISTOL(rev) \
drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c+3−2
@@ -80,8 +80,9 @@ static bool cik_event_interrupt_isr(struct kfd_dev *dev, · 79 unmodified lines
8080 ihre->source_id == CIK_INTSRC_SDMA_TRAP ||
8181 ihre->source_id == CIK_INTSRC_SQ_INTERRUPT_MSG ||
8282 ihre->source_id == CIK_INTSRC_CP_BAD_OPCODE ||
83− ihre->source_id == CIK_INTSRC_GFX_PAGE_INV_FAULT ||
84− ihre->source_id == CIK_INTSRC_GFX_MEM_PROT_FAULT;
83+ ((ihre->source_id == CIK_INTSRC_GFX_PAGE_INV_FAULT ||
84+ ihre->source_id == CIK_INTSRC_GFX_MEM_PROT_FAULT) &&
85+ !amdgpu_no_queue_eviction_on_vm_fault);
8586 }
8687
8788 static void cik_event_interrupt_wq(struct kfd_dev *dev,
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h+492−0
@@ -1575,6 +1575,498 @@ static const uint32_t cwsr_trap_arcturus_hex[] = { · 1574 unmodified lines
15751575 0xbf810000, 0x00000000,
15761576 };
15771577
1578+static const uint32_t cwsr_trap_aldebaran_hex[] = {
1579+ 0xbf820001, 0xbf8202ce,
1580+ 0xb8f8f802, 0x89788678,
1581+ 0xb8eef801, 0x866eff6e,
1582+ 0x00000800, 0xbf840003,
1583+ 0x866eff78, 0x00002000,
1584+ 0xbf840016, 0xb8fbf803,
1585+ 0x866eff7b, 0x00000400,
1586+ 0xbf85003b, 0x866eff7b,
1587+ 0x00000800, 0xbf850003,
1588+ 0x866eff7b, 0x00000100,
1589+ 0xbf84000c, 0x866eff78,
1590+ 0x00002000, 0xbf840005,
1591+ 0xbf8e0010, 0xb8eef803,
1592+ 0x866eff6e, 0x00000400,
1593+ 0xbf84fffb, 0x8778ff78,
1594+ 0x00002000, 0x80ec886c,
1595+ 0x82ed806d, 0xb8eef807,
1596+ 0x866fff6e, 0x001f8000,
1597+ 0x8e6f8b6f, 0x8977ff77,
1598+ 0xfc000000, 0x87776f77,
1599+ 0x896eff6e, 0x001f8000,
1600+ 0xb96ef807, 0xb8faf812,
1601+ 0xb8fbf813, 0x8efa887a,
1602+ 0xc0071bbd, 0x00000000,
1603+ 0xbf8cc07f, 0xc0071ebd,
1604+ 0x00000008, 0xbf8cc07f,
1605+ 0x86ee6e6e, 0xbf840001,
1606+ 0xbe801d6e, 0xb8fbf803,
1607+ 0x867bff7b, 0x000001ff,
1608+ 0xbf850002, 0x806c846c,
1609+ 0x826d806d, 0x866dff6d,
1610+ 0x0000ffff, 0x8f6e8b77,
1611+ 0x866eff6e, 0x001f8000,
1612+ 0xb96ef807, 0x86fe7e7e,
1613+ 0x86ea6a6a, 0x8f6e8378,
1614+ 0xb96ee0c2, 0xbf800002,
1615+ 0xb9780002, 0xbe801f6c,
1616+ 0x866dff6d, 0x0000ffff,
1617+ 0xbefa0080, 0xb97a0283,
1618+ 0xb8fa2407, 0x8e7a9b7a,
1619+ 0x876d7a6d, 0xb8fa03c7,
1620+ 0x8e7a9a7a, 0x876d7a6d,
1621+ 0xb8faf807, 0x867aff7a,
1622+ 0x00007fff, 0xb97af807,
1623+ 0xbeee007e, 0xbeef007f,
1624+ 0xbefe0180, 0xbf900004,
1625+ 0x877a8478, 0xb97af802,
1626+ 0xbf8e0002, 0xbf88fffe,
1627+ 0xb8fa2985, 0x807a817a,
1628+ 0x8e7a8a7a, 0x8e7a817a,
1629+ 0xb8fb1605, 0x807b817b,
1630+ 0x8e7b867b, 0x807a7b7a,
1631+ 0x807a7e7a, 0x827b807f,
1632+ 0x867bff7b, 0x0000ffff,
1633+ 0xc04b1c3d, 0x00000050,
1634+ 0xbf8cc07f, 0xc04b1d3d,
1635+ 0x00000060, 0xbf8cc07f,
1636+ 0xc0431e7d, 0x00000074,
1637+ 0xbf8cc07f, 0xbef4007e,
1638+ 0x8675ff7f, 0x0000ffff,
1639+ 0x8775ff75, 0x00040000,
1640+ 0xbef60080, 0xbef700ff,
1641+ 0x00807fac, 0x867aff7f,
1642+ 0x08000000, 0x8f7a837a,
1643+ 0x87777a77, 0x867aff7f,
1644+ 0x70000000, 0x8f7a817a,
1645+ 0x87777a77, 0xbef1007c,
1646+ 0xbef00080, 0xb8f02985,
1647+ 0x80708170, 0x8e708a70,
1648+ 0x8e708170, 0xb8fa1605,
1649+ 0x807a817a, 0x8e7a867a,
1650+ 0x80707a70, 0xbef60084,
1651+ 0xbef600ff, 0x01000000,
1652+ 0xbefe007c, 0xbefc0070,
1653+ 0xc0611c7a, 0x0000007c,
1654+ 0xbf8cc07f, 0x80708470,
1655+ 0xbefc007e, 0xbefe007c,
1656+ 0xbefc0070, 0xc0611b3a,
1657+ 0x0000007c, 0xbf8cc07f,
1658+ 0x80708470, 0xbefc007e,
1659+ 0xbefe007c, 0xbefc0070,
1660+ 0xc0611b7a, 0x0000007c,
1661+ 0xbf8cc07f, 0x80708470,
1662+ 0xbefc007e, 0xbefe007c,
1663+ 0xbefc0070, 0xc0611bba,
1664+ 0x0000007c, 0xbf8cc07f,
1665+ 0x80708470, 0xbefc007e,
1666+ 0xbefe007c, 0xbefc0070,
1667+ 0xc0611bfa, 0x0000007c,
1668+ 0xbf8cc07f, 0x80708470,
1669+ 0xbefc007e, 0xbefe007c,
1670+ 0xbefc0070, 0xc0611e3a,
1671+ 0x0000007c, 0xbf8cc07f,
1672+ 0x80708470, 0xbefc007e,
1673+ 0xb8fbf803, 0xbefe007c,
1674+ 0xbefc0070, 0xc0611efa,
1675+ 0x0000007c, 0xbf8cc07f,
1676+ 0x80708470, 0xbefc007e,
1677+ 0xbefe007c, 0xbefc0070,
1678+ 0xc0611a3a, 0x0000007c,
1679+ 0xbf8cc07f, 0x80708470,
1680+ 0xbefc007e, 0xbefe007c,
1681+ 0xbefc0070, 0xc0611a7a,
1682+ 0x0000007c, 0xbf8cc07f,
1683+ 0x80708470, 0xbefc007e,
1684+ 0xb8f1f801, 0xbefe007c,
1685+ 0xbefc0070, 0xc0611c7a,
1686+ 0x0000007c, 0xbf8cc07f,
1687+ 0x80708470, 0xbefc007e,
1688+ 0x867aff7f, 0x04000000,
1689+ 0xbeef0080, 0x876f6f7a,
1690+ 0xb8f02985, 0x80708170,
1691+ 0x8e708a70, 0x8e708170,
1692+ 0xb8fb1605, 0x807b817b,
1693+ 0x8e7b847b, 0x8e76827b,
1694+ 0xbef600ff, 0x01000000,
1695+ 0xbef20174, 0x80747074,
1696+ 0x82758075, 0xbefc0080,
1697+ 0xbf800000, 0xbe802b00,
1698+ 0xbe822b02, 0xbe842b04,
1699+ 0xbe862b06, 0xbe882b08,
1700+ 0xbe8a2b0a, 0xbe8c2b0c,
1701+ 0xbe8e2b0e, 0xc06b003a,
1702+ 0x00000000, 0xbf8cc07f,
1703+ 0xc06b013a, 0x00000010,
1704+ 0xbf8cc07f, 0xc06b023a,
1705+ 0x00000020, 0xbf8cc07f,
1706+ 0xc06b033a, 0x00000030,
1707+ 0xbf8cc07f, 0x8074c074,
1708+ 0x82758075, 0x807c907c,
1709+ 0xbf0a7b7c, 0xbf85ffe7,
1710+ 0xbef40172, 0xbef00080,
1711+ 0xbefe00c1, 0xbeff00c1,
1712+ 0xbee80080, 0xbee90080,
1713+ 0xbef600ff, 0x01000000,
1714+ 0x867aff78, 0x00400000,
1715+ 0xbf850003, 0xb8faf803,
1716+ 0x897a7aff, 0x10000000,
1717+ 0xbf85004d, 0xbe840080,
1718+ 0xd2890000, 0x00000900,
1719+ 0x80048104, 0xd2890001,
1720+ 0x00000900, 0x80048104,
1721+ 0xd2890002, 0x00000900,
1722+ 0x80048104, 0xd2890003,
1723+ 0x00000900, 0x80048104,
1724+ 0xc069003a, 0x00000070,
1725+ 0xbf8cc07f, 0x80709070,
1726+ 0xbf06c004, 0xbf84ffee,
1727+ 0xbe840080, 0xd2890000,
1728+ 0x00000901, 0x80048104,
1729+ 0xd2890001, 0x00000901,
1730+ 0x80048104, 0xd2890002,
1731+ 0x00000901, 0x80048104,
1732+ 0xd2890003, 0x00000901,
1733+ 0x80048104, 0xc069003a,
1734+ 0x00000070, 0xbf8cc07f,
1735+ 0x80709070, 0xbf06c004,
1736+ 0xbf84ffee, 0xbe840080,
1737+ 0xd2890000, 0x00000902,
1738+ 0x80048104, 0xd2890001,
1739+ 0x00000902, 0x80048104,
1740+ 0xd2890002, 0x00000902,
1741+ 0x80048104, 0xd2890003,
1742+ 0x00000902, 0x80048104,
1743+ 0xc069003a, 0x00000070,
1744+ 0xbf8cc07f, 0x80709070,
1745+ 0xbf06c004, 0xbf84ffee,
1746+ 0xbe840080, 0xd2890000,
1747+ 0x00000903, 0x80048104,
1748+ 0xd2890001, 0x00000903,
1749+ 0x80048104, 0xd2890002,
1750+ 0x00000903, 0x80048104,
1751+ 0xd2890003, 0x00000903,
1752+ 0x80048104, 0xc069003a,
1753+ 0x00000070, 0xbf8cc07f,
1754+ 0x80709070, 0xbf06c004,
1755+ 0xbf84ffee, 0xbf820008,
1756+ 0xe0724000, 0x701d0000,
1757+ 0xe0724100, 0x701d0100,
1758+ 0xe0724200, 0x701d0200,
1759+ 0xe0724300, 0x701d0300,
1760+ 0xbefe00c1, 0xbeff00c1,
1761+ 0xb8fb4306, 0x867bc17b,
1762+ 0xbf840064, 0xbf8a0000,
1763+ 0x867aff6f, 0x04000000,
1764+ 0xbf840060, 0x8e7b867b,
1765+ 0x8e7b827b, 0xbef6007b,
1766+ 0xb8f02985, 0x80708170,
1767+ 0x8e708a70, 0x8e708170,
1768+ 0xb8fa1605, 0x807a817a,
1769+ 0x8e7a867a, 0x80707a70,
1770+ 0x8070ff70, 0x00000080,
1771+ 0xbef600ff, 0x01000000,
1772+ 0xbefc0080, 0xd28c0002,
1773+ 0x000100c1, 0xd28d0003,
1774+ 0x000204c1, 0x867aff78,
1775+ 0x00400000, 0xbf850003,
1776+ 0xb8faf803, 0x897a7aff,
1777+ 0x10000000, 0xbf850030,
1778+ 0x24040682, 0xd86e4000,
1779+ 0x00000002, 0xbf8cc07f,
1780+ 0xbe840080, 0xd2890000,
1781+ 0x00000900, 0x80048104,
1782+ 0xd2890001, 0x00000900,
1783+ 0x80048104, 0xd2890002,
1784+ 0x00000900, 0x80048104,
1785+ 0xd2890003, 0x00000900,
1786+ 0x80048104, 0xc069003a,
1787+ 0x00000070, 0xbf8cc07f,
1788+ 0x80709070, 0xbf06c004,
1789+ 0xbf84ffee, 0xbe840080,
1790+ 0xd2890000, 0x00000901,
1791+ 0x80048104, 0xd2890001,
1792+ 0x00000901, 0x80048104,
1793+ 0xd2890002, 0x00000901,
1794+ 0x80048104, 0xd2890003,
1795+ 0x00000901, 0x80048104,
1796+ 0xc069003a, 0x00000070,
1797+ 0xbf8cc07f, 0x80709070,
1798+ 0xbf06c004, 0xbf84ffee,
1799+ 0x680404ff, 0x00000200,
1800+ 0xd0c9006a, 0x0000f702,
1801+ 0xbf87ffd2, 0xbf820015,
1802+ 0xd1060002, 0x00011103,
1803+ 0x7e0602ff, 0x00000200,
1804+ 0xbefc00ff, 0x00010000,
1805+ 0xbe800077, 0x8677ff77,
1806+ 0xff7fffff, 0x8777ff77,
1807+ 0x00058000, 0xd8ec0000,
1808+ 0x00000002, 0xbf8cc07f,
1809+ 0xe0765000, 0x701d0002,
1810+ 0x68040702, 0xd0c9006a,
1811+ 0x0000f702, 0xbf87fff7,
1812+ 0xbef70000, 0xbef000ff,
1813+ 0x00000400, 0xbefe00c1,
1814+ 0xbeff00c1, 0xb8fb2b05,
1815+ 0x807b817b, 0x8e7b827b,
1816+ 0xbef600ff, 0x01000000,
1817+ 0xbefc0084, 0xbf0a7b7c,
1818+ 0xbf84006d, 0xbf11017c,
1819+ 0x807bff7b, 0x00001000,
1820+ 0x867aff78, 0x00400000,
1821+ 0xbf850003, 0xb8faf803,
1822+ 0x897a7aff, 0x10000000,
1823+ 0xbf850051, 0xbe840080,
1824+ 0xd2890000, 0x00000900,
1825+ 0x80048104, 0xd2890001,
1826+ 0x00000900, 0x80048104,
1827+ 0xd2890002, 0x00000900,
1828+ 0x80048104, 0xd2890003,
1829+ 0x00000900, 0x80048104,
1830+ 0xc069003a, 0x00000070,
1831+ 0xbf8cc07f, 0x80709070,
1832+ 0xbf06c004, 0xbf84ffee,
1833+ 0xbe840080, 0xd2890000,
1834+ 0x00000901, 0x80048104,
1835+ 0xd2890001, 0x00000901,
1836+ 0x80048104, 0xd2890002,
1837+ 0x00000901, 0x80048104,
1838+ 0xd2890003, 0x00000901,
1839+ 0x80048104, 0xc069003a,
1840+ 0x00000070, 0xbf8cc07f,
1841+ 0x80709070, 0xbf06c004,
1842+ 0xbf84ffee, 0xbe840080,
1843+ 0xd2890000, 0x00000902,
1844+ 0x80048104, 0xd2890001,
1845+ 0x00000902, 0x80048104,
1846+ 0xd2890002, 0x00000902,
1847+ 0x80048104, 0xd2890003,
1848+ 0x00000902, 0x80048104,
1849+ 0xc069003a, 0x00000070,
1850+ 0xbf8cc07f, 0x80709070,
1851+ 0xbf06c004, 0xbf84ffee,
1852+ 0xbe840080, 0xd2890000,
1853+ 0x00000903, 0x80048104,
1854+ 0xd2890001, 0x00000903,
1855+ 0x80048104, 0xd2890002,
1856+ 0x00000903, 0x80048104,
1857+ 0xd2890003, 0x00000903,
1858+ 0x80048104, 0xc069003a,
1859+ 0x00000070, 0xbf8cc07f,
1860+ 0x80709070, 0xbf06c004,
1861+ 0xbf84ffee, 0x807c847c,
1862+ 0xbf0a7b7c, 0xbf85ffb1,
1863+ 0xbf9c0000, 0xbf820012,
1864+ 0x7e000300, 0x7e020301,
1865+ 0x7e040302, 0x7e060303,
1866+ 0xe0724000, 0x701d0000,
1867+ 0xe0724100, 0x701d0100,
1868+ 0xe0724200, 0x701d0200,
1869+ 0xe0724300, 0x701d0300,
1870+ 0x807c847c, 0x8070ff70,
1871+ 0x00000400, 0xbf0a7b7c,
1872+ 0xbf85ffef, 0xbf9c0000,
1873+ 0xb8fb2985, 0x807b817b,
1874+ 0x8e7b837b, 0xb8fa2b05,
1875+ 0x807a817a, 0x8e7a827a,
1876+ 0x80fb7a7b, 0x867b7b7b,
1877+ 0xbf84007a, 0x807bff7b,
1878+ 0x00001000, 0xbefc0080,
1879+ 0xbf11017c, 0x867aff78,
1880+ 0x00400000, 0xbf850003,
1881+ 0xb8faf803, 0x897a7aff,
1882+ 0x10000000, 0xbf850059,
1883+ 0xd3d84000, 0x18000100,
1884+ 0xd3d84001, 0x18000101,
1885+ 0xd3d84002, 0x18000102,
1886+ 0xd3d84003, 0x18000103,
1887+ 0xbe840080, 0xd2890000,
1888+ 0x00000900, 0x80048104,
1889+ 0xd2890001, 0x00000900,
1890+ 0x80048104, 0xd2890002,
1891+ 0x00000900, 0x80048104,
1892+ 0xd2890003, 0x00000900,
1893+ 0x80048104, 0xc069003a,
1894+ 0x00000070, 0xbf8cc07f,
1895+ 0x80709070, 0xbf06c004,
1896+ 0xbf84ffee, 0xbe840080,
1897+ 0xd2890000, 0x00000901,
1898+ 0x80048104, 0xd2890001,
1899+ 0x00000901, 0x80048104,
1900+ 0xd2890002, 0x00000901,
1901+ 0x80048104, 0xd2890003,
1902+ 0x00000901, 0x80048104,
1903+ 0xc069003a, 0x00000070,
1904+ 0xbf8cc07f, 0x80709070,
1905+ 0xbf06c004, 0xbf84ffee,
1906+ 0xbe840080, 0xd2890000,
1907+ 0x00000902, 0x80048104,
1908+ 0xd2890001, 0x00000902,
1909+ 0x80048104, 0xd2890002,
1910+ 0x00000902, 0x80048104,
1911+ 0xd2890003, 0x00000902,
1912+ 0x80048104, 0xc069003a,
1913+ 0x00000070, 0xbf8cc07f,
1914+ 0x80709070, 0xbf06c004,
1915+ 0xbf84ffee, 0xbe840080,
1916+ 0xd2890000, 0x00000903,
1917+ 0x80048104, 0xd2890001,
1918+ 0x00000903, 0x80048104,
1919+ 0xd2890002, 0x00000903,
1920+ 0x80048104, 0xd2890003,
1921+ 0x00000903, 0x80048104,
1922+ 0xc069003a, 0x00000070,
1923+ 0xbf8cc07f, 0x80709070,
1924+ 0xbf06c004, 0xbf84ffee,
1925+ 0x807c847c, 0xbf0a7b7c,
1926+ 0xbf85ffa9, 0xbf9c0000,
1927+ 0xbf820016, 0xd3d84000,
1928+ 0x18000100, 0xd3d84001,
1929+ 0x18000101, 0xd3d84002,
1930+ 0x18000102, 0xd3d84003,
1931+ 0x18000103, 0xe0724000,
1932+ 0x701d0000, 0xe0724100,
1933+ 0x701d0100, 0xe0724200,
1934+ 0x701d0200, 0xe0724300,
1935+ 0x701d0300, 0x807c847c,
1936+ 0x8070ff70, 0x00000400,
1937+ 0xbf0a7b7c, 0xbf85ffeb,
1938+ 0xbf9c0000, 0xbf820101,
1939+ 0xbef4007e, 0x8675ff7f,
1940+ 0x0000ffff, 0x8775ff75,
1941+ 0x00040000, 0xbef60080,
1942+ 0xbef700ff, 0x00807fac,
1943+ 0x866eff7f, 0x08000000,
1944+ 0x8f6e836e, 0x87776e77,
1945+ 0x866eff7f, 0x70000000,
1946+ 0x8f6e816e, 0x87776e77,
1947+ 0x866eff7f, 0x04000000,
1948+ 0xbf84001f, 0xbefe00c1,
1949+ 0xbeff00c1, 0xb8ef4306,
1950+ 0x866fc16f, 0xbf84001a,
1951+ 0x8e6f866f, 0x8e6f826f,
1952+ 0xbef6006f, 0xb8f82985,
1953+ 0x80788178, 0x8e788a78,
1954+ 0x8e788178, 0xb8ee1605,
1955+ 0x806e816e, 0x8e6e866e,
1956+ 0x80786e78, 0x8078ff78,
1957+ 0x00000080, 0xbef600ff,
1958+ 0x01000000, 0xbefc0080,
1959+ 0xe0510000, 0x781d0000,
1960+ 0xe0510100, 0x781d0000,
1961+ 0x807cff7c, 0x00000200,
1962+ 0x8078ff78, 0x00000200,
1963+ 0xbf0a6f7c, 0xbf85fff6,
1964+ 0xbefe00c1, 0xbeff00c1,
1965+ 0xbef600ff, 0x01000000,
1966+ 0xb8ef2b05, 0x806f816f,
1967+ 0x8e6f826f, 0x806fff6f,
1968+ 0x00008000, 0xbef80080,
1969+ 0xbeee0078, 0x8078ff78,
1970+ 0x00000400, 0xbefc0084,
1971+ 0xbf11087c, 0xe0524000,
1972+ 0x781d0000, 0xe0524100,
1973+ 0x781d0100, 0xe0524200,
1974+ 0x781d0200, 0xe0524300,
1975+ 0x781d0300, 0xbf8c0f70,
1976+ 0x7e000300, 0x7e020301,
1977+ 0x7e040302, 0x7e060303,
1978+ 0x807c847c, 0x8078ff78,
1979+ 0x00000400, 0xbf0a6f7c,
1980+ 0xbf85ffee, 0xb8ef2985,
1981+ 0x806f816f, 0x8e6f836f,
1982+ 0xb8f92b05, 0x80798179,
1983+ 0x8e798279, 0x80ef796f,
1984+ 0x866f6f6f, 0xbf84001a,
1985+ 0x806fff6f, 0x00008000,
1986+ 0xbefc0080, 0xbf11087c,
1987+ 0xe0524000, 0x781d0000,
1988+ 0xe0524100, 0x781d0100,
1989+ 0xe0524200, 0x781d0200,
1990+ 0xe0524300, 0x781d0300,
1991+ 0xbf8c0f70, 0xd3d94000,
1992+ 0x18000100, 0xd3d94001,
1993+ 0x18000101, 0xd3d94002,
1994+ 0x18000102, 0xd3d94003,
1995+ 0x18000103, 0x807c847c,
1996+ 0x8078ff78, 0x00000400,
1997+ 0xbf0a6f7c, 0xbf85ffea,
1998+ 0xbf9c0000, 0xe0524000,
1999+ 0x6e1d0000, 0xe0524100,
2000+ 0x6e1d0100, 0xe0524200,
2001+ 0x6e1d0200, 0xe0524300,
2002+ 0x6e1d0300, 0xbf8c0f70,
2003+ 0xb8f82985, 0x80788178,
2004+ 0x8e788a78, 0x8e788178,
2005+ 0xb8ee1605, 0x806e816e,
2006+ 0x8e6e866e, 0x80786e78,
2007+ 0x80f8c078, 0xb8ef1605,
2008+ 0x806f816f, 0x8e6f846f,
2009+ 0x8e76826f, 0xbef600ff,
2010+ 0x01000000, 0xbefc006f,
2011+ 0xc031003a, 0x00000078,
2012+ 0x80f8c078, 0xbf8cc07f,
2013+ 0x80fc907c, 0xbf800000,
2014+ 0xbe802d00, 0xbe822d02,
2015+ 0xbe842d04, 0xbe862d06,
2016+ 0xbe882d08, 0xbe8a2d0a,
2017+ 0xbe8c2d0c, 0xbe8e2d0e,
2018+ 0xbf06807c, 0xbf84fff0,
2019+ 0xb8f82985, 0x80788178,
2020+ 0x8e788a78, 0x8e788178,
2021+ 0xb8ee1605, 0x806e816e,
2022+ 0x8e6e866e, 0x80786e78,
2023+ 0xbef60084, 0xbef600ff,
2024+ 0x01000000, 0xc0211bfa,
2025+ 0x00000078, 0x80788478,
2026+ 0xc0211b3a, 0x00000078,
2027+ 0x80788478, 0xc0211b7a,
2028+ 0x00000078, 0x80788478,
2029+ 0xc0211c3a, 0x00000078,
2030+ 0x80788478, 0xc0211c7a,
2031+ 0x00000078, 0x80788478,
2032+ 0xc0211eba, 0x00000078,
2033+ 0x80788478, 0xc0211efa,
2034+ 0x00000078, 0x80788478,
2035+ 0xc0211a3a, 0x00000078,
2036+ 0x80788478, 0xc0211a7a,
2037+ 0x00000078, 0x80788478,
2038+ 0xc0211cfa, 0x00000078,
2039+ 0x80788478, 0xbf8cc07f,
2040+ 0xbefc006f, 0xbefe0070,
2041+ 0xbeff0071, 0x866f7bff,
2042+ 0x000003ff, 0xb96f4803,
2043+ 0x866f7bff, 0xfffff800,
2044+ 0x8f6f8b6f, 0xb96fa2c3,
2045+ 0xb973f801, 0xb8ee2985,
2046+ 0x806e816e, 0x8e6e8a6e,
2047+ 0x8e6e816e, 0xb8ef1605,
2048+ 0x806f816f, 0x8e6f866f,
2049+ 0x806e6f6e, 0x806e746e,
2050+ 0x826f8075, 0x866fff6f,
2051+ 0x0000ffff, 0xc00b1c37,
2052+ 0x00000050, 0xc00b1d37,
2053+ 0x00000060, 0xc0031e77,
2054+ 0x00000074, 0xbf8cc07f,
2055+ 0x866fff6d, 0xf8000000,
2056+ 0x8f6f9b6f, 0x8e6f906f,
2057+ 0xbeee0080, 0x876e6f6e,
2058+ 0x866fff6d, 0x04000000,
2059+ 0x8f6f9a6f, 0x8e6f8f6f,
2060+ 0x876e6f6e, 0x866fff7a,
2061+ 0x00800000, 0x8f6f976f,
2062+ 0xb96ef807, 0x866dff6d,
2063+ 0x0000ffff, 0x86fe7e7e,
2064+ 0x86ea6a6a, 0x8f6e837a,
2065+ 0xb96ee0c2, 0xbf800002,
2066+ 0xb97a0002, 0xbf8a0000,
2067+ 0x95806f6c, 0xbf810000,
2068+};
2069+
15782070 static const uint32_t cwsr_trap_gfx10_hex[] = {
15792071 0xbf820001, 0xbf8201cf,
15802072 0xb0804004, 0xb978f802,
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm+1−0
@@ -563,6 +563,7 @@ L_RESTORE: · 562 unmodified lines
563563 buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256
564564 buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256*2
565565 buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256*3
566+ s_waitcnt vmcnt(0)
566567
567568 /* restore SGPRs */
568569 //////////////////////////////
drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm+123−81
@@ -21,9 +21,24 @@ · 20 unmodified lines
2121 */
2222
2323 /* To compile this assembly code:
24− * PROJECT=greenland ./sp3 cwsr_trap_handler_gfx9.asm -hex tmp.hex
24+ *
25+ * gfx9:
26+ * cpp -DASIC_FAMILY=CHIP_VEGAM cwsr_trap_handler_gfx9.asm -P -o gfx9.sp3
27+ * sp3 gfx9.sp3 -hex gfx9.hex
28+ *
29+ * arcturus:
30+ * cpp -DASIC_FAMILY=CHIP_ARCTURUS cwsr_trap_handler_gfx9.asm -P -o arcturus.sp3
31+ * sp3 arcturus.sp3 -hex arcturus.hex
32+ *
33+ * aldebaran:
34+ * cpp -DASIC_FAMILY=CHIP_ALDEBARAN cwsr_trap_handler_gfx9.asm -P -o aldebaran.sp3
35+ * sp3 aldebaran.sp3 -hex aldebaran.hex
2536 */
2637
38+#define CHIP_VEGAM 18
39+#define CHIP_ARCTURUS 23
40+#define CHIP_ALDEBARAN 25
41+
2742 var ACK_SQC_STORE = 1 //workaround for suspected SQC store bug causing incorrect stores under concurrency
2843 var SAVE_AFTER_XNACK_ERROR = 1 //workaround for TCP store failure after XNACK error when ALLOW_REPLAY=0, for debugger
2944 var SINGLE_STEP_MISSED_WORKAROUND = 1 //workaround for lost MODE.DEBUG_EN exception when SAVECTX raised
@@ -44,10 +59,17 @@ var SQ_WAVE_STATUS_ALLOW_REPLAY_MASK = 0x400000 · 14 unmodified lines
4459
4560 var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT = 12
4661 var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE = 9
47−var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT = 8
4862 var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE = 6
49−var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT = 24
5063 var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE = 3 //FIXME sq.blk still has 4 bits at this time while SQ programming guide has 3 bits
64+var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT = 24
65+
66+#if ASIC_FAMILY >= CHIP_ALDEBARAN
67+var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT = 6
68+var SQ_WAVE_GPR_ALLOC_ACCV_OFFSET_SHIFT = 12
69+var SQ_WAVE_GPR_ALLOC_ACCV_OFFSET_SIZE = 6
70+#else
71+var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT = 8
72+#endif
5173
5274 var SQ_WAVE_TRAPSTS_SAVECTX_MASK = 0x400
5375 var SQ_WAVE_TRAPSTS_EXCE_MASK = 0x1FF // Exception mask
@@ -134,7 +156,7 @@ var s_restore_spi_init_lo = exec_lo · 80 unmodified lines
134156 var s_restore_spi_init_hi = exec_hi
135157
136158 var s_restore_mem_offset = ttmp12
137−var s_restore_accvgpr_offset = ttmp13
159+var s_restore_tmp2 = ttmp13
138160 var s_restore_alloc_size = ttmp3
139161 var s_restore_tmp = ttmp2
140162 var s_restore_mem_offset_save = s_restore_tmp //no conflict
@@ -466,12 +488,7 @@ if SAVE_AFTER_XNACK_ERROR · 325 unmodified lines
466488 L_SAVE_FIRST_VGPRS_WITH_TCP:
467489 end
468490
469− buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
470− buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256
471− buffer_store_dword v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*2
472− buffer_store_dword v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*3
473−
474−
491+ write_4vgprs_to_mem(s_save_buf_rsrc0, s_save_mem_offset)
475492
476493 /* save LDS */
477494 //////////////////////////////
@@ -565,11 +582,8 @@ L_SAVE_LDS_DONE: · 87 unmodified lines
565582 s_mov_b32 exec_lo, 0xFFFFFFFF //need every thread from now on
566583 s_mov_b32 exec_hi, 0xFFFFFFFF
567584
568− s_getreg_b32 s_save_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE) //vpgr_size
569− s_add_u32 s_save_alloc_size, s_save_alloc_size, 1
570− s_lshl_b32 s_save_alloc_size, s_save_alloc_size, 2 //Number of VGPRs = (vgpr_size + 1) * 4 (non-zero value) //FIXME for GFX, zero is possible
571− s_lshl_b32 s_save_buf_rsrc2, s_save_alloc_size, 8 //NUM_RECORDS in bytes (64 threads*4)
572− s_mov_b32 s_save_buf_rsrc2, 0x1000000 //NUM_RECORDS in bytes
585+ get_num_arch_vgprs(s_save_alloc_size)
586+ s_mov_b32 s_save_buf_rsrc2, 0x1000000 //NUM_RECORDS in bytes
573587
574588
575589 // VGPR store using dw burst
@@ -602,10 +616,7 @@ end · 26 unmodified lines
602616 v_mov_b32 v2, v2 //v0 = v[0+m0]
603617 v_mov_b32 v3, v3 //v0 = v[0+m0]
604618
605− buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
606− buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256
607− buffer_store_dword v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*2
608− buffer_store_dword v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*3
619+ write_4vgprs_to_mem(s_save_buf_rsrc0, s_save_mem_offset)
609620
610621 s_add_u32 m0, m0, 4 //next vgpr index
611622 s_add_u32 s_save_mem_offset, s_save_mem_offset, 256*4 //every buffer_store_dword does 256 bytes
@@ -615,8 +626,17 @@ end · 3 unmodified lines
615626
616627 L_SAVE_VGPR_END:
617628
618−if ASIC_TARGET_ARCTURUS
629+#if ASIC_FAMILY >= CHIP_ARCTURUS
619630 // Save ACC VGPRs
631+
632+#if ASIC_FAMILY >= CHIP_ALDEBARAN
633+ // ACC VGPR count may differ from ARCH VGPR count.
634+ get_num_acc_vgprs(s_save_alloc_size, s_save_tmp)
635+ s_and_b32 s_save_alloc_size, s_save_alloc_size, s_save_alloc_size
636+ s_cbranch_scc0 L_SAVE_ACCVGPR_END
637+ s_add_u32 s_save_alloc_size, s_save_alloc_size, 0x1000 //add 0x1000 since we compare m0 against it later
638+#endif
639+
620640 s_mov_b32 m0, 0x0 //VGPR initial index value =0
621641 s_set_gpr_idx_on m0, 0x1 //M0[7:0] = M0[7:0] and M0[15:12] = 0x1
622642
@@ -644,10 +664,7 @@ L_SAVE_ACCVGPR_LOOP: · 21 unmodified lines
644664 v_accvgpr_read v[vgpr], acc[vgpr] // v[N] = acc[N+m0]
645665 end
646666
647− buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
648− buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256
649− buffer_store_dword v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*2
650− buffer_store_dword v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*3
667+ write_4vgprs_to_mem(s_save_buf_rsrc0, s_save_mem_offset)
651668
652669 s_add_u32 m0, m0, 4
653670 s_add_u32 s_save_mem_offset, s_save_mem_offset, 256*4
@@ -656,7 +673,7 @@ L_SAVE_ACCVGPR_LOOP: · 2 unmodified lines
656673 s_set_gpr_idx_off
657674
658675 L_SAVE_ACCVGPR_END:
659−end
676+#endif
660677
661678 s_branch L_END_PGM
662679
@@ -724,53 +741,23 @@ L_RESTORE: · 61 unmodified lines
724741 /* restore VGPRs */
725742 //////////////////////////////
726743 L_RESTORE_VGPR:
727− // VGPR SR memory offset : 0
728− s_mov_b32 s_restore_mem_offset, 0x0
729744 s_mov_b32 exec_lo, 0xFFFFFFFF //need every thread from now on //be consistent with SAVE although can be moved ahead
730745 s_mov_b32 exec_hi, 0xFFFFFFFF
746+ s_mov_b32 s_restore_buf_rsrc2, 0x1000000 //NUM_RECORDS in bytes
731747
732− s_getreg_b32 s_restore_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE) //vpgr_size
733− s_add_u32 s_restore_alloc_size, s_restore_alloc_size, 1
734− s_lshl_b32 s_restore_alloc_size, s_restore_alloc_size, 2 //Number of VGPRs = (vgpr_size + 1) * 4 (non-zero value)
735− s_lshl_b32 s_restore_buf_rsrc2, s_restore_alloc_size, 8 //NUM_RECORDS in bytes (64 threads*4)
736−
737−if ASIC_TARGET_ARCTURUS
738− s_mov_b32 s_restore_accvgpr_offset, s_restore_buf_rsrc2 //ACC VGPRs at end of VGPRs
739−end
740−
741− s_mov_b32 s_restore_buf_rsrc2, 0x1000000 //NUM_RECORDS in bytes
748+ // Save ARCH VGPRs 4-N, then all ACC VGPRs, then ARCH VGPRs 0-3.
749+ get_num_arch_vgprs(s_restore_alloc_size)
750+ s_add_u32 s_restore_alloc_size, s_restore_alloc_size, 0x8000 //add 0x8000 since we compare m0 against it later
742751
743− // VGPR load using dw burst
752+ // ARCH VGPRs at offset: 0
753+ s_mov_b32 s_restore_mem_offset, 0x0
744754 s_mov_b32 s_restore_mem_offset_save, s_restore_mem_offset // restore start with v1, v0 will be the last
745755 s_add_u32 s_restore_mem_offset, s_restore_mem_offset, 256*4
746−if ASIC_TARGET_ARCTURUS
747− s_mov_b32 s_restore_accvgpr_offset_save, s_restore_accvgpr_offset
748− s_add_u32 s_restore_accvgpr_offset, s_restore_accvgpr_offset, 256*4
749−end
750756 s_mov_b32 m0, 4 //VGPR initial index value = 1
751− s_set_gpr_idx_on m0, 0x8 //M0[7:0] = M0[7:0] and M0[15:12] = 0x8
752− s_add_u32 s_restore_alloc_size, s_restore_alloc_size, 0x8000 //add 0x8000 since we compare m0 against it later
757+ s_set_gpr_idx_on m0, 0x8 //M0[7:0] = M0[7:0] and M0[15:12] = 0x8
753758
754759 L_RESTORE_VGPR_LOOP:
755−
756−if ASIC_TARGET_ARCTURUS
757− buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset slc:1 glc:1
758− buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset slc:1 glc:1 offset:256
759− buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset slc:1 glc:1 offset:256*2
760− buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset slc:1 glc:1 offset:256*3
761− s_add_u32 s_restore_accvgpr_offset, s_restore_accvgpr_offset, 256*4
762− s_waitcnt vmcnt(0)
763−
764− for var vgpr = 0; vgpr < 4; ++ vgpr
765− v_accvgpr_write acc[vgpr], v[vgpr]
766− end
767−end
768−
769− buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
770− buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256
771− buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*2
772− buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*3
773− s_waitcnt vmcnt(0) //ensure data ready
760+ read_4vgprs_from_mem(s_restore_buf_rsrc0, s_restore_mem_offset)
774761 v_mov_b32 v0, v0 //v[0+m0] = v0
775762 v_mov_b32 v1, v1
776763 v_mov_b32 v2, v2
@@ -779,24 +766,38 @@ end · 2 unmodified lines
779766 s_add_u32 s_restore_mem_offset, s_restore_mem_offset, 256*4 //every buffer_load_dword does 256 bytes
780767 s_cmp_lt_u32 m0, s_restore_alloc_size //scc = (m0 < s_restore_alloc_size) ? 1 : 0
781768 s_cbranch_scc1 L_RESTORE_VGPR_LOOP //VGPR restore (except v0) is complete?
782− s_set_gpr_idx_off
783− /* VGPR restore on v0 */
784−if ASIC_TARGET_ARCTURUS
785− buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset_save slc:1 glc:1
786− buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset_save slc:1 glc:1 offset:256
787− buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset_save slc:1 glc:1 offset:256*2
788− buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset_save slc:1 glc:1 offset:256*3
789− s_waitcnt vmcnt(0)
790769
791− for var vgpr = 0; vgpr < 4; ++ vgpr
792− v_accvgpr_write acc[vgpr], v[vgpr]
793− end
794−end
770+#if ASIC_FAMILY >= CHIP_ALDEBARAN
771+ // ACC VGPR count may differ from ARCH VGPR count.
772+ get_num_acc_vgprs(s_restore_alloc_size, s_restore_tmp2)
773+ s_and_b32 s_restore_alloc_size, s_restore_alloc_size, s_restore_alloc_size
774+ s_cbranch_scc0 L_RESTORE_ACCVGPR_END
775+ s_add_u32 s_restore_alloc_size, s_restore_alloc_size, 0x8000 //add 0x8000 since we compare m0 against it later
776+#endif
795777
796− buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1
797− buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256
798− buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256*2
799− buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256*3
778+#if ASIC_FAMILY >= CHIP_ARCTURUS
779+ // ACC VGPRs at offset: size(ARCH VGPRs)
780+ s_mov_b32 m0, 0
781+ s_set_gpr_idx_on m0, 0x8 //M0[7:0] = M0[7:0] and M0[15:12] = 0x8
782+
783+ L_RESTORE_ACCVGPR_LOOP:
784+ read_4vgprs_from_mem(s_restore_buf_rsrc0, s_restore_mem_offset)
785+
786+ for var vgpr = 0; vgpr < 4; ++ vgpr
787+ v_accvgpr_write acc[vgpr], v[vgpr]
788+ end
789+
790+ s_add_u32 m0, m0, 4 //next vgpr index
791+ s_add_u32 s_restore_mem_offset, s_restore_mem_offset, 256*4 //every buffer_load_dword does 256 bytes
792+ s_cmp_lt_u32 m0, s_restore_alloc_size //scc = (m0 < s_restore_alloc_size) ? 1 : 0
793+ s_cbranch_scc1 L_RESTORE_ACCVGPR_LOOP //VGPR restore (except v0) is complete?
794+ L_RESTORE_ACCVGPR_END:
795+#endif
796+
797+ s_set_gpr_idx_off
798+
799+ // Restore VGPRs 0-3 last, no longer needed.
800+ read_4vgprs_from_mem(s_restore_buf_rsrc0, s_restore_mem_offset_save)
800801
801802 /* restore SGPRs */
802803 //////////////////////////////
@@ -974,6 +975,21 @@ function check_if_tcp_store_ok · 171 unmodified lines
974975 L_TCP_STORE_CHECK_DONE:
975976 end
976977
978+function write_4vgprs_to_mem(s_rsrc, s_mem_offset)
979+ buffer_store_dword v0, v0, s_rsrc, s_mem_offset slc:1 glc:1
980+ buffer_store_dword v1, v0, s_rsrc, s_mem_offset slc:1 glc:1 offset:256
981+ buffer_store_dword v2, v0, s_rsrc, s_mem_offset slc:1 glc:1 offset:256*2
982+ buffer_store_dword v3, v0, s_rsrc, s_mem_offset slc:1 glc:1 offset:256*3
983+end
984+
985+function read_4vgprs_from_mem(s_rsrc, s_mem_offset)
986+ buffer_load_dword v0, v0, s_rsrc, s_mem_offset slc:1 glc:1
987+ buffer_load_dword v1, v0, s_rsrc, s_mem_offset slc:1 glc:1 offset:256
988+ buffer_load_dword v2, v0, s_rsrc, s_mem_offset slc:1 glc:1 offset:256*2
989+ buffer_load_dword v3, v0, s_rsrc, s_mem_offset slc:1 glc:1 offset:256*3
990+ s_waitcnt vmcnt(0)
991+end
992+
977993 function write_vgpr_to_mem_with_sqc(v, s_rsrc, s_mem_offset)
978994 s_mov_b32 s4, 0
979995
@@ -1008,9 +1024,9 @@ function get_vgpr_size_bytes(s_vgpr_size_byte) · 28 unmodified lines
10081024 s_add_u32 s_vgpr_size_byte, s_vgpr_size_byte, 1
10091025 s_lshl_b32 s_vgpr_size_byte, s_vgpr_size_byte, (2+8) //Number of VGPRs = (vgpr_size + 1) * 4 * 64 * 4 (non-zero value) //FIXME for GFX, zero is possible
10101026
1011−if ASIC_TARGET_ARCTURUS
1027+#if ASIC_FAMILY >= CHIP_ARCTURUS
10121028 s_lshl_b32 s_vgpr_size_byte, s_vgpr_size_byte, 1 // Double size for ACC VGPRs
1013−end
1029+#endif
10141030 end
10151031
10161032 function get_sgpr_size_bytes(s_sgpr_size_byte)
@@ -1023,6 +1039,32 @@ function get_hwreg_size_bytes · 6 unmodified lines
10231039 return 128 //HWREG size 128 bytes
10241040 end
10251041
1042+function get_num_arch_vgprs(s_num_arch_vgprs)
1043+#if ASIC_FAMILY >= CHIP_ALDEBARAN
1044+ // VGPR count includes ACC VGPRs, use ACC VGPR offset for ARCH VGPR count.
1045+ s_getreg_b32 s_num_arch_vgprs, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_ACCV_OFFSET_SHIFT,SQ_WAVE_GPR_ALLOC_ACCV_OFFSET_SIZE)
1046+#else
1047+ s_getreg_b32 s_num_arch_vgprs, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
1048+#endif
1049+
1050+ // Number of VGPRs = (vgpr_size + 1) * 4
1051+ s_add_u32 s_num_arch_vgprs, s_num_arch_vgprs, 1
1052+ s_lshl_b32 s_num_arch_vgprs, s_num_arch_vgprs, 2
1053+end
1054+
1055+#if ASIC_FAMILY >= CHIP_ALDEBARAN
1056+function get_num_acc_vgprs(s_num_acc_vgprs, s_tmp)
1057+ // VGPR count = (GPR_ALLOC.VGPR_SIZE + 1) * 8
1058+ s_getreg_b32 s_num_acc_vgprs, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
1059+ s_add_u32 s_num_acc_vgprs, s_num_acc_vgprs, 1
1060+ s_lshl_b32 s_num_acc_vgprs, s_num_acc_vgprs, 3
1061+
1062+ // ACC VGPR count = VGPR count - ARCH VGPR count.
1063+ get_num_arch_vgprs(s_tmp)
1064+ s_sub_u32 s_num_acc_vgprs, s_num_acc_vgprs, s_tmp
1065+end
1066+#endif
1067+
10261068 function ack_sqc_store_workaround
10271069 if ACK_SQC_STORE
10281070 s_waitcnt lgkmcnt(0)
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c+1−5
@@ -556,11 +556,7 @@ static int kfd_ioctl_set_trap_handler(struct file *filep, · 555 unmodified lines
556556 goto out;
557557 }
558558
559− if (dev->dqm->ops.set_trap_handler(dev->dqm,
560− &pdd->qpd,
561− args->tba_addr,
562− args->tma_addr))
563− err = -EINVAL;
559+ kfd_process_set_trap_handler(&pdd->qpd, args->tba_addr, args->tma_addr);
564560
565561 out:
566562 mutex_unlock(&p->mutex);
drivers/gpu/drm/amd/amdkfd/kfd_crat.c+16−1
@@ -26,6 +26,7 @@ · 25 unmodified lines
2626 #include "kfd_priv.h"
2727 #include "kfd_topology.h"
2828 #include "kfd_iommu.h"
29+#include "amdgpu.h"
2930 #include "amdgpu_amdkfd.h"
3031
3132 /* GPU Processor ID base for dGPUs for which VCRAT needs to be created.
@@ -665,6 +666,7 @@ static int kfd_fill_gpu_cache_info(struct kfd_dev *kdev, · 633 unmodified lines
665666 case CHIP_VEGA12:
666667 case CHIP_VEGA20:
667668 case CHIP_ARCTURUS:
669+ case CHIP_ALDEBARAN:
668670 pcache_info = vega10_cache_info;
669671 num_of_cache_types = ARRAY_SIZE(vega10_cache_info);
670672 break;
@@ -1112,6 +1114,8 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, · 441 unmodified lines
11121114 struct crat_subtype_iolink *sub_type_hdr,
11131115 uint32_t proximity_domain)
11141116 {
1117+ struct amdgpu_device *adev = (struct amdgpu_device *)kdev->kgd;
1118+
11151119 *avail_size -= sizeof(struct crat_subtype_iolink);
11161120 if (*avail_size < 0)
11171121 return -ENOMEM;
@@ -1128,7 +1132,18 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, · 10 unmodified lines
11281132 /* Fill in IOLINK subtype.
11291133 * TODO: Fill-in other fields of iolink subtype
11301134 */
1131− sub_type_hdr->io_interface_type = CRAT_IOLINK_TYPE_PCIEXPRESS;
1135+ if (adev->gmc.xgmi.connected_to_cpu) {
1136+ /*
1137+ * with host gpu xgmi link, host can access gpu memory whether
1138+ * or not pcie bar type is large, so always create bidirectional
1139+ * io link.
1140+ */
1141+ sub_type_hdr->flags |= CRAT_IOLINK_FLAGS_BI_DIRECTIONAL;
1142+ sub_type_hdr->io_interface_type = CRAT_IOLINK_TYPE_XGMI;
1143+ } else {
1144+ sub_type_hdr->io_interface_type = CRAT_IOLINK_TYPE_PCIEXPRESS;
1145+ }
1146+
11321147 sub_type_hdr->proximity_domain_from = proximity_domain;
11331148 #ifdef CONFIG_NUMA
11341149 if (kdev->pdev->dev.numa_node == NUMA_NO_NODE)
drivers/gpu/drm/amd/amdkfd/kfd_device.c+25−0
@@ -46,6 +46,7 @@ extern const struct kfd2kgd_calls gfx_v7_kfd2kgd; · 45 unmodified lines
4646 extern const struct kfd2kgd_calls gfx_v8_kfd2kgd;
4747 extern const struct kfd2kgd_calls gfx_v9_kfd2kgd;
4848 extern const struct kfd2kgd_calls arcturus_kfd2kgd;
49+extern const struct kfd2kgd_calls aldebaran_kfd2kgd;
4950 extern const struct kfd2kgd_calls gfx_v10_kfd2kgd;
5051 extern const struct kfd2kgd_calls gfx_v10_3_kfd2kgd;
5152
@@ -71,6 +72,7 @@ static const struct kfd2kgd_calls *kfd2kgd_funcs[] = { · 19 unmodified lines
7172 [CHIP_VEGA20] = &gfx_v9_kfd2kgd,
7273 [CHIP_RENOIR] = &gfx_v9_kfd2kgd,
7374 [CHIP_ARCTURUS] = &arcturus_kfd2kgd,
75+ [CHIP_ALDEBARAN] = &aldebaran_kfd2kgd,
7476 [CHIP_NAVI10] = &gfx_v10_kfd2kgd,
7577 [CHIP_NAVI12] = &gfx_v10_kfd2kgd,
7678 [CHIP_NAVI14] = &gfx_v10_kfd2kgd,
@@ -392,6 +394,24 @@ static const struct kfd_device_info arcturus_device_info = { · 315 unmodified lines
392394 .num_sdma_queues_per_engine = 8,
393395 };
394396
397+static const struct kfd_device_info aldebaran_device_info = {
398+ .asic_family = CHIP_ALDEBARAN,
399+ .asic_name = "aldebaran",
400+ .max_pasid_bits = 16,
401+ .max_no_of_hqd = 24,
402+ .doorbell_size = 8,
403+ .ih_ring_entry_size = 8 * sizeof(uint32_t),
404+ .event_interrupt_class = &event_interrupt_class_v9,
405+ .num_of_watch_points = 4,
406+ .mqd_size_aligned = MQD_SIZE_ALIGNED,
407+ .supports_cwsr = true,
408+ .needs_iommu_device = false,
409+ .needs_pci_atomics = false,
410+ .num_sdma_engines = 2,
411+ .num_xgmi_sdma_engines = 3,
412+ .num_sdma_queues_per_engine = 8,
413+};
414+
395415 static const struct kfd_device_info renoir_device_info = {
396416 .asic_family = CHIP_RENOIR,
397417 .asic_name = "renoir",
@@ -556,6 +576,7 @@ static const struct kfd_device_info *kfd_supported_devices[][2] = { · 158 unmodified lines
556576 [CHIP_VEGA20] = {&vega20_device_info, NULL},
557577 [CHIP_RENOIR] = {&renoir_device_info, NULL},
558578 [CHIP_ARCTURUS] = {&arcturus_device_info, &arcturus_device_info},
579+ [CHIP_ALDEBARAN] = {&aldebaran_device_info, NULL},
559580 [CHIP_NAVI10] = {&navi10_device_info, NULL},
560581 [CHIP_NAVI12] = {&navi12_device_info, &navi12_device_info},
561582 [CHIP_NAVI14] = {&navi14_device_info, NULL},
@@ -640,6 +661,10 @@ static void kfd_cwsr_init(struct kfd_dev *kfd) · 78 unmodified lines
640661 BUILD_BUG_ON(sizeof(cwsr_trap_arcturus_hex) > PAGE_SIZE);
641662 kfd->cwsr_isa = cwsr_trap_arcturus_hex;
642663 kfd->cwsr_isa_size = sizeof(cwsr_trap_arcturus_hex);
664+ } else if (kfd->device_info->asic_family == CHIP_ALDEBARAN) {
665+ BUILD_BUG_ON(sizeof(cwsr_trap_aldebaran_hex) > PAGE_SIZE);
666+ kfd->cwsr_isa = cwsr_trap_aldebaran_hex;
667+ kfd->cwsr_isa_size = sizeof(cwsr_trap_aldebaran_hex);
643668 } else if (kfd->device_info->asic_family < CHIP_NAVI10) {
644669 BUILD_BUG_ON(sizeof(cwsr_trap_gfx9_hex) > PAGE_SIZE);
645670 kfd->cwsr_isa = cwsr_trap_gfx9_hex;
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c+33−24
@@ -1128,6 +1128,9 @@ static int set_sched_resources(struct device_queue_manager *dqm) · 1127 unmodified lines
11281128
11291129 static int initialize_cpsch(struct device_queue_manager *dqm)
11301130 {
1131+ uint64_t num_sdma_queues;
1132+ uint64_t num_xgmi_sdma_queues;
1133+
11311134 pr_debug("num of pipes: %d\n", get_pipes_per_mec(dqm));
11321135
11331136 mutex_init(&dqm->lock_hidden);
@@ -1136,8 +1139,18 @@ static int initialize_cpsch(struct device_queue_manager *dqm) · 2 unmodified lines
11361139 dqm->active_cp_queue_count = 0;
11371140 dqm->gws_queue_count = 0;
11381141 dqm->active_runlist = false;
1139− dqm->sdma_bitmap = ~0ULL >> (64 - get_num_sdma_queues(dqm));
1140− dqm->xgmi_sdma_bitmap = ~0ULL >> (64 - get_num_xgmi_sdma_queues(dqm));
1142+
1143+ num_sdma_queues = get_num_sdma_queues(dqm);
1144+ if (num_sdma_queues >= BITS_PER_TYPE(dqm->sdma_bitmap))
1145+ dqm->sdma_bitmap = ULLONG_MAX;
1146+ else
1147+ dqm->sdma_bitmap = (BIT_ULL(num_sdma_queues) - 1);
1148+
1149+ num_xgmi_sdma_queues = get_num_xgmi_sdma_queues(dqm);
1150+ if (num_xgmi_sdma_queues >= BITS_PER_TYPE(dqm->xgmi_sdma_bitmap))
1151+ dqm->xgmi_sdma_bitmap = ULLONG_MAX;
1152+ else
1153+ dqm->xgmi_sdma_bitmap = (BIT_ULL(num_xgmi_sdma_queues) - 1);
11411154
11421155 INIT_WORK(&dqm->hw_exception_work, kfd_process_hw_exception);
11431156
@@ -1393,6 +1406,7 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm, · 249 unmodified lines
13931406 uint32_t filter_param)
13941407 {
13951408 int retval = 0;
1409+ struct mqd_manager *mqd_mgr;
13961410
13971411 if (!dqm->sched_running)
13981412 return 0;
@@ -1424,6 +1438,22 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm, · 25 unmodified lines
14241438 return retval;
14251439 }
14261440
1441+ /* In the current MEC firmware implementation, if compute queue
1442+ * doesn't response to the preemption request in time, HIQ will
1443+ * abandon the unmap request without returning any timeout error
1444+ * to driver. Instead, MEC firmware will log the doorbell of the
1445+ * unresponding compute queue to HIQ.MQD.queue_doorbell_id fields.
1446+ * To make sure the queue unmap was successful, driver need to
1447+ * check those fields
1448+ */
1449+ mqd_mgr = dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ];
1450+ if (mqd_mgr->read_doorbell_id(dqm->packets.priv_queue->queue->mqd)) {
1451+ pr_err("HIQ MQD's queue_doorbell_id0 is not 0, Queue preemption time out\n");
1452+ while (halt_if_hws_hang)
1453+ schedule();
1454+ return -ETIME;
1455+ }
1456+
14271457 pm_release_ib(&dqm->packets);
14281458 dqm->active_runlist = false;
14291459
@@ -1596,26 +1626,6 @@ out: · 166 unmodified lines
15961626 return retval;
15971627 }
15981628
1599−static int set_trap_handler(struct device_queue_manager *dqm,
1600− struct qcm_process_device *qpd,
1601− uint64_t tba_addr,
1602− uint64_t tma_addr)
1603−{
1604− uint64_t *tma;
1605−
1606− if (dqm->dev->cwsr_enabled) {
1607− /* Jump from CWSR trap handler to user trap */
1608− tma = (uint64_t *)(qpd->cwsr_kaddr + KFD_CWSR_TMA_OFFSET);
1609− tma[0] = tba_addr;
1610− tma[1] = tma_addr;
1611− } else {
1612− qpd->tba_addr = tba_addr;
1613− qpd->tma_addr = tma_addr;
1614− }
1615−
1616− return 0;
1617−}
1618−
16191629 static int process_termination_nocpsch(struct device_queue_manager *dqm,
16201630 struct qcm_process_device *qpd)
16211631 {
@@ -1859,7 +1869,6 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) · 237 unmodified lines
18591869 dqm->ops.create_kernel_queue = create_kernel_queue_cpsch;
18601870 dqm->ops.destroy_kernel_queue = destroy_kernel_queue_cpsch;
18611871 dqm->ops.set_cache_memory_policy = set_cache_memory_policy;
1862− dqm->ops.set_trap_handler = set_trap_handler;
18631872 dqm->ops.process_termination = process_termination_cpsch;
18641873 dqm->ops.evict_process_queues = evict_process_queues_cpsch;
18651874 dqm->ops.restore_process_queues = restore_process_queues_cpsch;
@@ -1878,7 +1887,6 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) · 12 unmodified lines
18781887 dqm->ops.initialize = initialize_nocpsch;
18791888 dqm->ops.uninitialize = uninitialize;
18801889 dqm->ops.set_cache_memory_policy = set_cache_memory_policy;
1881− dqm->ops.set_trap_handler = set_trap_handler;
18821890 dqm->ops.process_termination = process_termination_nocpsch;
18831891 dqm->ops.evict_process_queues = evict_process_queues_nocpsch;
18841892 dqm->ops.restore_process_queues =
@@ -1918,6 +1926,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) · 33 unmodified lines
19181926 case CHIP_RAVEN:
19191927 case CHIP_RENOIR:
19201928 case CHIP_ARCTURUS:
1929+ case CHIP_ALDEBARAN:
19211930 device_queue_manager_init_v9(&dqm->asic_ops);
19221931 break;
19231932 case CHIP_NAVI10:
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h+0−5
@@ -121,11 +121,6 @@ struct device_queue_manager_ops { · 120 unmodified lines
121121 void __user *alternate_aperture_base,
122122 uint64_t alternate_aperture_size);
123123
124− int (*set_trap_handler)(struct device_queue_manager *dqm,
125− struct qcm_process_device *qpd,
126− uint64_t tba_addr,
127− uint64_t tma_addr);
128−
129124 int (*process_termination)(struct device_queue_manager *dqm,
130125 struct qcm_process_device *qpd);
131126
drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c+1−0
@@ -412,6 +412,7 @@ int kfd_init_apertures(struct kfd_process *process) · 411 unmodified lines
412412 case CHIP_RAVEN:
413413 case CHIP_RENOIR:
414414 case CHIP_ARCTURUS:
415+ case CHIP_ALDEBARAN:
415416 case CHIP_NAVI10:
416417 case CHIP_NAVI12:
417418 case CHIP_NAVI14:
drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c+3−2
@@ -98,9 +98,10 @@ static bool event_interrupt_isr_v9(struct kfd_dev *dev, · 97 unmodified lines
9898 source_id == SOC15_INTSRC_SDMA_TRAP ||
9999 source_id == SOC15_INTSRC_SQ_INTERRUPT_MSG ||
100100 source_id == SOC15_INTSRC_CP_BAD_OPCODE ||
101− client_id == SOC15_IH_CLIENTID_VMC ||
101+ ((client_id == SOC15_IH_CLIENTID_VMC ||
102102 client_id == SOC15_IH_CLIENTID_VMC1 ||
103− client_id == SOC15_IH_CLIENTID_UTCL2;
103+ client_id == SOC15_IH_CLIENTID_UTCL2) &&
104+ !amdgpu_no_queue_eviction_on_vm_fault);
104105 }
105106
106107 static void event_interrupt_wq_v9(struct kfd_dev *dev,
drivers/gpu/drm/amd/amdkfd/kfd_iommu.c+6−0
@@ -20,6 +20,10 @@ · 19 unmodified lines
2020 * OTHER DEALINGS IN THE SOFTWARE.
2121 */
2222
23+#include <linux/kconfig.h>
24+
25+#if IS_REACHABLE(CONFIG_AMD_IOMMU_V2)
26+
2327 #include <linux/printk.h>
2428 #include <linux/device.h>
2529 #include <linux/slab.h>
@@ -355,3 +359,5 @@ int kfd_iommu_add_perf_counters(struct kfd_topology_device *kdev) · 329 unmodified lines
355359
356360 return 0;
357361 }
362+
363+#endif
drivers/gpu/drm/amd/amdkfd/kfd_iommu.h+7−2
@@ -23,7 +23,9 @@ · 22 unmodified lines
2323 #ifndef __KFD_IOMMU_H__
2424 #define __KFD_IOMMU_H__
2525
26−#if defined(CONFIG_AMD_IOMMU_V2_MODULE) || defined(CONFIG_AMD_IOMMU_V2)
26+#include <linux/kconfig.h>
27+
28+#if IS_REACHABLE(CONFIG_AMD_IOMMU_V2)
2729
2830 #define KFD_SUPPORT_IOMMU_V2
2931
@@ -46,6 +48,9 @@ static inline int kfd_iommu_check_device(struct kfd_dev *kfd) · 16 unmodified lines
4648 }
4749 static inline int kfd_iommu_device_init(struct kfd_dev *kfd)
4850 {
51+#if IS_MODULE(CONFIG_AMD_IOMMU_V2)
52+ WARN_ONCE(1, "iommu_v2 module is not usable by built-in KFD");
53+#endif
4954 return 0;
5055 }
5156
@@ -73,6 +78,6 @@ static inline int kfd_iommu_add_perf_counters(struct kfd_topology_device *kdev) · 21 unmodified lines
7378 return 0;
7479 }
7580
76−#endif /* defined(CONFIG_AMD_IOMMU_V2) */
81+#endif /* IS_REACHABLE(CONFIG_AMD_IOMMU_V2) */
7782
7883 #endif /* __KFD_IOMMU_H__ */
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h+1−0
@@ -101,6 +101,7 @@ struct mqd_manager { · 100 unmodified lines
101101 #if defined(CONFIG_DEBUG_FS)
102102 int (*debugfs_show_mqd)(struct seq_file *m, void *data);
103103 #endif
104+ uint32_t (*read_doorbell_id)(void *mqd);
104105
105106 struct mutex mqd_mutex;
106107 struct kfd_dev *dev;
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c+8−0
@@ -226,6 +226,13 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, · 225 unmodified lines
226226 __update_mqd(mm, mqd, q, 1);
227227 }
228228
229+static uint32_t read_doorbell_id(void *mqd)
230+{
231+ struct cik_mqd *m = (struct cik_mqd *)mqd;
232+
233+ return m->queue_doorbell_id0;
234+}
235+
229236 static void update_mqd_hawaii(struct mqd_manager *mm, void *mqd,
230237 struct queue_properties *q)
231238 {
@@ -398,6 +405,7 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type, · 166 unmodified lines
398405 #if defined(CONFIG_DEBUG_FS)
399406 mqd->debugfs_show_mqd = debugfs_show_mqd;
400407 #endif
408+ mqd->read_doorbell_id = read_doorbell_id;
401409 break;
402410 case KFD_MQD_TYPE_DIQ:
403411 mqd->allocate_mqd = allocate_mqd;
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c+8−0
@@ -224,6 +224,13 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, · 223 unmodified lines
224224 q->is_active = QUEUE_IS_ACTIVE(*q);
225225 }
226226
227+static uint32_t read_doorbell_id(void *mqd)
228+{
229+ struct v10_compute_mqd *m = (struct v10_compute_mqd *)mqd;
230+
231+ return m->queue_doorbell_id0;
232+}
233+
227234 static int destroy_mqd(struct mqd_manager *mm, void *mqd,
228235 enum kfd_preempt_type type,
229236 unsigned int timeout, uint32_t pipe_id,
@@ -425,6 +432,7 @@ struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type, · 195 unmodified lines
425432 #if defined(CONFIG_DEBUG_FS)
426433 mqd->debugfs_show_mqd = debugfs_show_mqd;
427434 #endif
435+ mqd->read_doorbell_id = read_doorbell_id;
428436 pr_debug("%s@%i\n", __func__, __LINE__);
429437 break;
430438 case KFD_MQD_TYPE_DIQ:
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c+8−0
@@ -276,6 +276,13 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, · 275 unmodified lines
276276 }
277277
278278
279+static uint32_t read_doorbell_id(void *mqd)
280+{
281+ struct v9_mqd *m = (struct v9_mqd *)mqd;
282+
283+ return m->queue_doorbell_id0;
284+}
285+
279286 static int destroy_mqd(struct mqd_manager *mm, void *mqd,
280287 enum kfd_preempt_type type,
281288 unsigned int timeout, uint32_t pipe_id,
@@ -477,6 +484,7 @@ struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type, · 195 unmodified lines
477484 #if defined(CONFIG_DEBUG_FS)
478485 mqd->debugfs_show_mqd = debugfs_show_mqd;
479486 #endif
487+ mqd->read_doorbell_id = read_doorbell_id;
480488 break;
481489 case KFD_MQD_TYPE_DIQ:
482490 mqd->allocate_mqd = allocate_mqd;
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c+8−0
@@ -243,6 +243,13 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, · 242 unmodified lines
243243 __update_mqd(mm, mqd, q, MTYPE_CC, 1);
244244 }
245245
246+static uint32_t read_doorbell_id(void *mqd)
247+{
248+ struct vi_mqd *m = (struct vi_mqd *)mqd;
249+
250+ return m->queue_doorbell_id0;
251+}
252+
246253 static void update_mqd_tonga(struct mqd_manager *mm, void *mqd,
247254 struct queue_properties *q)
248255 {
@@ -446,6 +453,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type, · 197 unmodified lines
446453 #if defined(CONFIG_DEBUG_FS)
447454 mqd->debugfs_show_mqd = debugfs_show_mqd;
448455 #endif
456+ mqd->read_doorbell_id = read_doorbell_id;
449457 break;
450458 case KFD_MQD_TYPE_DIQ:
451459 mqd->allocate_mqd = allocate_mqd;
drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c+1−0
@@ -242,6 +242,7 @@ int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm) · 241 unmodified lines
242242 case CHIP_RAVEN:
243243 case CHIP_RENOIR:
244244 case CHIP_ARCTURUS:
245+ case CHIP_ALDEBARAN:
245246 case CHIP_NAVI10:
246247 case CHIP_NAVI12:
247248 case CHIP_NAVI14:
drivers/gpu/drm/amd/amdkfd/kfd_priv.h+9−0
@@ -169,6 +169,11 @@ extern bool hws_gws_support; · 168 unmodified lines
169169 /* Queue preemption timeout in ms */
170170 extern int queue_preemption_timeout_ms;
171171
172+/*
173+ * Don't evict process queues on vm fault
174+ */
175+extern int amdgpu_no_queue_eviction_on_vm_fault;
176+
172177 /* Enable eviction debug messages */
173178 extern bool debug_evictions;
174179
@@ -944,6 +949,10 @@ bool interrupt_is_wanted(struct kfd_dev *dev, · 769 unmodified lines
944949 /* amdkfd Apertures */
945950 int kfd_init_apertures(struct kfd_process *process);
946951
952+void kfd_process_set_trap_handler(struct qcm_process_device *qpd,
953+ uint64_t tba_addr,
954+ uint64_t tma_addr);
955+
947956 /* Queue Context Management */
948957 int init_queue(struct queue **q, const struct queue_properties *properties);
949958 void uninit_queue(struct queue *q);
drivers/gpu/drm/amd/amdkfd/kfd_process.c+56−10
@@ -775,10 +775,8 @@ struct kfd_process *kfd_create_process(struct file *filep) · 774 unmodified lines
775775 goto out;
776776
777777 ret = kfd_process_init_cwsr_apu(process, filep);
778− if (ret) {
779− process = ERR_PTR(ret);
780− goto out;
781− }
778+ if (ret)
779+ goto out_destroy;
782780
783781 if (!procfs.kobj)
784782 goto out;
@@ -826,6 +824,14 @@ out: · 41 unmodified lines
826824 mutex_unlock(&kfd_processes_mutex);
827825
828826 return process;
827+
828+out_destroy:
829+ hash_del_rcu(&process->kfd_processes);
830+ mutex_unlock(&kfd_processes_mutex);
831+ synchronize_srcu(&kfd_processes_srcu);
832+ /* kfd_process_free_notifier will trigger the cleanup */
833+ mmu_notifier_put(&process->mmu_notifier);
834+ return ERR_PTR(ret);
829835 }
830836
831837 struct kfd_process *kfd_get_process(const struct task_struct *thread)
@@ -1011,6 +1017,16 @@ static void kfd_process_ref_release(struct kref *ref) · 179 unmodified lines
10111017 queue_work(kfd_process_wq, &p->release_work);
10121018 }
10131019
1020+static struct mmu_notifier *kfd_process_alloc_notifier(struct mm_struct *mm)
1021+{
1022+ int idx = srcu_read_lock(&kfd_processes_srcu);
1023+ struct kfd_process *p = find_process_by_mm(mm);
1024+
1025+ srcu_read_unlock(&kfd_processes_srcu, idx);
1026+
1027+ return p ? &p->mmu_notifier : ERR_PTR(-ESRCH);
1028+}
1029+
10141030 static void kfd_process_free_notifier(struct mmu_notifier *mn)
10151031 {
10161032 kfd_unref_process(container_of(mn, struct kfd_process, mmu_notifier));
@@ -1075,6 +1091,7 @@ static void kfd_process_notifier_release(struct mmu_notifier *mn, · 58 unmodified lines
10751091
10761092 static const struct mmu_notifier_ops kfd_process_mmu_notifier_ops = {
10771093 .release = kfd_process_notifier_release,
1094+ .alloc_notifier = kfd_process_alloc_notifier,
10781095 .free_notifier = kfd_process_free_notifier,
10791096 };
10801097
@@ -1145,6 +1162,25 @@ static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd) · 64 unmodified lines
11451162 return 0;
11461163 }
11471164
1165+void kfd_process_set_trap_handler(struct qcm_process_device *qpd,
1166+ uint64_t tba_addr,
1167+ uint64_t tma_addr)
1168+{
1169+ if (qpd->cwsr_kaddr) {
1170+ /* KFD trap handler is bound, record as second-level TBA/TMA
1171+ * in first-level TMA. First-level trap will jump to second.
1172+ */
1173+ uint64_t *tma =
1174+ (uint64_t *)(qpd->cwsr_kaddr + KFD_CWSR_TMA_OFFSET);
1175+ tma[0] = tba_addr;
1176+ tma[1] = tma_addr;
1177+ } else {
1178+ /* No trap handler bound, bind as first-level TBA/TMA. */
1179+ qpd->tba_addr = tba_addr;
1180+ qpd->tma_addr = tma_addr;
1181+ }
1182+}
1183+
11481184 /*
11491185 * On return the kfd_process is fully operational and will be freed when the
11501186 * mm is released
@@ -1152,6 +1188,7 @@ static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd) · 1 unmodified line
11521188 static struct kfd_process *create_process(const struct task_struct *thread)
11531189 {
11541190 struct kfd_process *process;
1191+ struct mmu_notifier *mn;
11551192 int err = -ENOMEM;
11561193
11571194 process = kzalloc(sizeof(*process), GFP_KERNEL);
@@ -1182,19 +1219,28 @@ static struct kfd_process *create_process(const struct task_struct *thread) · 24 unmodified lines
11821219 if (err != 0)
11831220 goto err_init_apertures;
11841221
1185− /* Must be last, have to use release destruction after this */
1186− process->mmu_notifier.ops = &kfd_process_mmu_notifier_ops;
1187− err = mmu_notifier_register(&process->mmu_notifier, process->mm);
1188− if (err)
1222+ /* alloc_notifier needs to find the process in the hash table */
1223+ hash_add_rcu(kfd_processes_table, &process->kfd_processes,
1224+ (uintptr_t)process->mm);
1225+
1226+ /* MMU notifier registration must be the last call that can fail
1227+ * because after this point we cannot unwind the process creation.
1228+ * After this point, mmu_notifier_put will trigger the cleanup by
1229+ * dropping the last process reference in the free_notifier.
1230+ */
1231+ mn = mmu_notifier_get(&kfd_process_mmu_notifier_ops, process->mm);
1232+ if (IS_ERR(mn)) {
1233+ err = PTR_ERR(mn);
11891234 goto err_register_notifier;
1235+ }
1236+ BUG_ON(mn != &process->mmu_notifier);
11901237
11911238 get_task_struct(process->lead_thread);
1192− hash_add_rcu(kfd_processes_table, &process->kfd_processes,
1193− (uintptr_t)process->mm);
11941239
11951240 return process;
11961241
11971242 err_register_notifier:
1243+ hash_del_rcu(&process->kfd_processes);
11981244 kfd_process_free_outstanding_kfd_bos(process);
11991245 kfd_process_destroy_pdds(process);
12001246 err_init_apertures:
drivers/gpu/drm/amd/amdkfd/kfd_topology.c+1−0
@@ -1370,6 +1370,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu) · 1369 unmodified lines
13701370 case CHIP_RAVEN:
13711371 case CHIP_RENOIR:
13721372 case CHIP_ARCTURUS:
1373+ case CHIP_ALDEBARAN:
13731374 case CHIP_NAVI10:
13741375 case CHIP_NAVI12:
13751376 case CHIP_NAVI14:
drivers/gpu/drm/amd/amdkfd/kfd_topology.h+3−2
@@ -47,13 +47,14 @@ · 46 unmodified lines
4747 #define HSA_CAP_DOORBELL_TYPE_2_0 0x2
4848 #define HSA_CAP_AQL_QUEUE_DOUBLE_MAP 0x00004000
4949
50−#define HSA_CAP_SRAM_EDCSUPPORTED 0x00080000
50+#define HSA_CAP_RESERVED_WAS_SRAM_EDCSUPPORTED 0x00080000 /* Old buggy user mode depends on this being 0 */
5151 #define HSA_CAP_MEM_EDCSUPPORTED 0x00100000
5252 #define HSA_CAP_RASEVENTNOTIFY 0x00200000
5353 #define HSA_CAP_ASIC_REVISION_MASK 0x03c00000
5454 #define HSA_CAP_ASIC_REVISION_SHIFT 22
55+#define HSA_CAP_SRAM_EDCSUPPORTED 0x04000000
5556
56−#define HSA_CAP_RESERVED 0xfc078000
57+#define HSA_CAP_RESERVED 0xf80f8000
5758
5859 struct kfd_node_properties {
5960 uint64_t hive_id;
drivers/gpu/drm/amd/display/Kconfig+14−0
@@ -38,4 +38,18 @@ config DEBUG_KERNEL_DC · 37 unmodified lines
3838 help
3939 Choose this option if you want to hit kdgb_break in assert.
4040
41+config DRM_AMD_SECURE_DISPLAY
42+ bool "Enable secure display support"
43+ default n
44+ depends on DEBUG_FS
45+ depends on DRM_AMD_DC_DCN
46+ help
47+ Choose this option if you want to
48+ support secure display
49+
50+ This option enables the calculation
51+ of crc of specific region via debugfs.
52+ Cooperate with specific DMCU FW.
53+
54+
4155 endmenu
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c+622−102
@@ -34,6 +34,7 @@ · 33 unmodified lines
3434 #include "dc/inc/hw/dmcu.h"
3535 #include "dc/inc/hw/abm.h"
3636 #include "dc/dc_dmub_srv.h"
37+#include "dc/dc_edid_parser.h"
3738 #include "amdgpu_dm_trace.h"
3839
3940 #include "vid.h"
@@ -75,7 +76,6 @@ · 35 unmodified lines
7576 #include <drm/drm_edid.h>
7677 #include <drm/drm_vblank.h>
7778 #include <drm/drm_audio_component.h>
78−#include <drm/drm_hdcp.h>
7979
8080 #if defined(CONFIG_DRM_AMD_DC_DCN)
8181 #include "ivsrcid/dcn/irqsrcs_dcn_1_0.h"
@@ -212,6 +212,9 @@ static bool amdgpu_dm_psr_disable_all(struct amdgpu_display_manager *dm); · 130 unmodified lines
212212 static const struct drm_format_info *
213213 amd_get_format_info(const struct drm_mode_fb_cmd2 *cmd);
214214
215+static bool
216+is_timing_unchanged_for_freesync(struct drm_crtc_state *old_crtc_state,
217+ struct drm_crtc_state *new_crtc_state);
215218 /*
216219 * dm_vblank_get_counter
217220 *
@@ -335,6 +338,17 @@ static inline bool amdgpu_dm_vrr_active(struct dm_crtc_state *dm_state) · 117 unmodified lines
335338 dm_state->freesync_config.state == VRR_STATE_ACTIVE_FIXED;
336339 }
337340
341+static inline bool is_dc_timing_adjust_needed(struct dm_crtc_state *old_state,
342+ struct dm_crtc_state *new_state)
343+{
344+ if (new_state->freesync_config.state == VRR_STATE_ACTIVE_FIXED)
345+ return true;
346+ else if (amdgpu_dm_vrr_active(old_state) != amdgpu_dm_vrr_active(new_state))
347+ return true;
348+ else
349+ return false;
350+}
351+
338352 /**
339353 * dm_pflip_high_irq() - Handle pageflip interrupt
340354 * @interrupt_params: ignored
@@ -566,6 +580,31 @@ static void dm_crtc_high_irq(void *interrupt_params) · 225 unmodified lines
566580 spin_unlock_irqrestore(&adev_to_drm(adev)->event_lock, flags);
567581 }
568582
583+#if defined(CONFIG_DRM_AMD_DC_DCN)
584+/**
585+ * dm_dcn_vertical_interrupt0_high_irq() - Handles OTG Vertical interrupt0 for
586+ * DCN generation ASICs
587+ * @interrupt params - interrupt parameters
588+ *
589+ * Used to set crc window/read out crc value at vertical line 0 position
590+ */
591+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
592+static void dm_dcn_vertical_interrupt0_high_irq(void *interrupt_params)
593+{
594+ struct common_irq_params *irq_params = interrupt_params;
595+ struct amdgpu_device *adev = irq_params->adev;
596+ struct amdgpu_crtc *acrtc;
597+
598+ acrtc = get_crtc_by_otg_inst(adev, irq_params->irq_src - IRQ_TYPE_VLINE0);
599+
600+ if (!acrtc)
601+ return;
602+
603+ amdgpu_dm_crtc_handle_crc_window_irq(&acrtc->base);
604+}
605+#endif
606+#endif
607+
569608 static int dm_set_clockgating_state(void *handle,
570609 enum amd_clockgating_state state)
571610 {
@@ -951,9 +990,7 @@ static void event_mall_stutter(struct work_struct *work) · 379 unmodified lines
951990 else
952991 dm->active_vblank_irq_count--;
953992
954−
955− dc_allow_idle_optimizations(
956− dm->dc, dm->active_vblank_irq_count == 0 ? true : false);
993+ dc_allow_idle_optimizations(dm->dc, dm->active_vblank_irq_count == 0);
957994
958995 DRM_DEBUG_DRIVER("Allow idle optimizations (MALL): %d\n", dm->active_vblank_irq_count == 0);
959996
@@ -1060,6 +1097,7 @@ static int amdgpu_dm_init(struct amdgpu_device *adev) · 100 unmodified lines
10601097
10611098 init_data.flags.power_down_display_on_boot = true;
10621099
1100+ INIT_LIST_HEAD(&adev->dm.da_list);
10631101 /* Display Core create. */
10641102 adev->dm.dc = dc_create(&init_data);
10651103
@@ -1138,6 +1176,9 @@ static int amdgpu_dm_init(struct amdgpu_device *adev) · 72 unmodified lines
11381176
11391177 dc_init_callbacks(adev->dm.dc, &init_params);
11401178 }
1179+#endif
1180+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
1181+ adev->dm.crc_rd_wrk = amdgpu_dm_crtc_secure_display_create_work();
11411182 #endif
11421183 if (amdgpu_dm_initialize_drm_device(adev)) {
11431184 DRM_ERROR(
@@ -1182,6 +1223,13 @@ static void amdgpu_dm_fini(struct amdgpu_device *adev) · 38 unmodified lines
11821223
11831224 amdgpu_dm_destroy_drm_device(&adev->dm);
11841225
1226+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
1227+ if (adev->dm.crc_rd_wrk) {
1228+ flush_work(&adev->dm.crc_rd_wrk->notify_ta_work);
1229+ kfree(adev->dm.crc_rd_wrk);
1230+ adev->dm.crc_rd_wrk = NULL;
1231+ }
1232+#endif
11851233 #ifdef CONFIG_DRM_AMD_DC_HDCP
11861234 if (adev->dm.hdcp_workqueue) {
11871235 hdcp_destroy(&adev->dev->kobj, adev->dm.hdcp_workqueue);
@@ -1191,6 +1239,15 @@ static void amdgpu_dm_fini(struct amdgpu_device *adev) · 3 unmodified lines
11911239 if (adev->dm.dc)
11921240 dc_deinit_callbacks(adev->dm.dc);
11931241 #endif
1242+
1243+#if defined(CONFIG_DRM_AMD_DC_DCN)
1244+ if (adev->dm.vblank_workqueue) {
1245+ adev->dm.vblank_workqueue->dm = NULL;
1246+ kfree(adev->dm.vblank_workqueue);
1247+ adev->dm.vblank_workqueue = NULL;
1248+ }
1249+#endif
1250+
11941251 if (adev->dm.dc->ctx->dmub_srv) {
11951252 dc_dmub_srv_destroy(&adev->dm.dc->ctx->dmub_srv);
11961253 adev->dm.dc->ctx->dmub_srv = NULL;
@@ -1847,6 +1904,9 @@ static int dm_suspend(void *handle) · 650 unmodified lines
18471904 return ret;
18481905 }
18491906
1907+#ifdef CONFIG_DRM_AMD_SECURE_DISPLAY
1908+ amdgpu_dm_crtc_secure_display_suspend(adev);
1909+#endif
18501910 WARN_ON(adev->dm.cached_state);
18511911 adev->dm.cached_state = drm_atomic_helper_suspend(adev_to_drm(adev));
18521912
@@ -2171,6 +2231,10 @@ static int dm_resume(void *handle) · 318 unmodified lines
21712231
21722232 dm->cached_state = NULL;
21732233
2234+#ifdef CONFIG_DRM_AMD_SECURE_DISPLAY
2235+ amdgpu_dm_crtc_secure_display_resume(adev);
2236+#endif
2237+
21742238 amdgpu_dm_irq_resume_late(adev);
21752239
21762240 amdgpu_dm_smu_write_watermarks_table(adev);
@@ -2907,6 +2971,16 @@ static int dcn10_register_irq_handlers(struct amdgpu_device *adev) · 730 unmodified lines
29072971 struct dc_interrupt_params int_params = {0};
29082972 int r;
29092973 int i;
2974+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
2975+ static const unsigned int vrtl_int_srcid[] = {
2976+ DCN_1_0__SRCID__OTG1_VERTICAL_INTERRUPT0_CONTROL,
2977+ DCN_1_0__SRCID__OTG2_VERTICAL_INTERRUPT0_CONTROL,
2978+ DCN_1_0__SRCID__OTG3_VERTICAL_INTERRUPT0_CONTROL,
2979+ DCN_1_0__SRCID__OTG4_VERTICAL_INTERRUPT0_CONTROL,
2980+ DCN_1_0__SRCID__OTG5_VERTICAL_INTERRUPT0_CONTROL,
2981+ DCN_1_0__SRCID__OTG6_VERTICAL_INTERRUPT0_CONTROL
2982+ };
2983+#endif
29102984
29112985 int_params.requested_polarity = INTERRUPT_POLARITY_DEFAULT;
29122986 int_params.current_polarity = INTERRUPT_POLARITY_DEFAULT;
@@ -2947,6 +3021,37 @@ static int dcn10_register_irq_handlers(struct amdgpu_device *adev) · 34 unmodified lines
29473021 adev, &int_params, dm_crtc_high_irq, c_irq_params);
29483022 }
29493023
3024+ /* Use otg vertical line interrupt */
3025+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
3026+ for (i = 0; i <= adev->mode_info.num_crtc - 1; i++) {
3027+ r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_DCE,
3028+ vrtl_int_srcid[i], &adev->vline0_irq);
3029+
3030+ if (r) {
3031+ DRM_ERROR("Failed to add vline0 irq id!\n");
3032+ return r;
3033+ }
3034+
3035+ int_params.int_context = INTERRUPT_HIGH_IRQ_CONTEXT;
3036+ int_params.irq_source =
3037+ dc_interrupt_to_irq_source(dc, vrtl_int_srcid[i], 0);
3038+
3039+ if (int_params.irq_source == DC_IRQ_SOURCE_INVALID) {
3040+ DRM_ERROR("Failed to register vline0 irq %d!\n", vrtl_int_srcid[i]);
3041+ break;
3042+ }
3043+
3044+ c_irq_params = &adev->dm.vline0_params[int_params.irq_source
3045+ - DC_IRQ_SOURCE_DC1_VLINE0];
3046+
3047+ c_irq_params->adev = adev;
3048+ c_irq_params->irq_src = int_params.irq_source;
3049+
3050+ amdgpu_dm_irq_register_interrupt(adev, &int_params,
3051+ dm_dcn_vertical_interrupt0_high_irq, c_irq_params);
3052+ }
3053+#endif
3054+
29503055 /* Use VUPDATE_NO_LOCK interrupt on DCN, which seems to correspond to
29513056 * the regular VUPDATE interrupt on DCE. We want DC_IRQ_SOURCE_VUPDATEx
29523057 * to trigger at end of each vblank, regardless of state of the lock,
@@ -5001,19 +5106,16 @@ static void fill_stream_properties_from_drm_display_mode( · 2048 unmodified lines
50015106 timing_out->hdmi_vic = hv_frame.vic;
50025107 }
50035108
5004− timing_out->h_addressable = mode_in->crtc_hdisplay;
5005− timing_out->h_total = mode_in->crtc_htotal;
5006− timing_out->h_sync_width =
5007− mode_in->crtc_hsync_end - mode_in->crtc_hsync_start;
5008− timing_out->h_front_porch =
5009− mode_in->crtc_hsync_start - mode_in->crtc_hdisplay;
5010− timing_out->v_total = mode_in->crtc_vtotal;
5011− timing_out->v_addressable = mode_in->crtc_vdisplay;
5012− timing_out->v_front_porch =
5013− mode_in->crtc_vsync_start - mode_in->crtc_vdisplay;
5014− timing_out->v_sync_width =
5015− mode_in->crtc_vsync_end - mode_in->crtc_vsync_start;
5016− timing_out->pix_clk_100hz = mode_in->crtc_clock * 10;
5109+ timing_out->h_addressable = mode_in->hdisplay;
5110+ timing_out->h_total = mode_in->htotal;
5111+ timing_out->h_sync_width = mode_in->hsync_end - mode_in->hsync_start;
5112+ timing_out->h_front_porch = mode_in->hsync_start - mode_in->hdisplay;
5113+ timing_out->v_total = mode_in->vtotal;
5114+ timing_out->v_addressable = mode_in->vdisplay;
5115+ timing_out->v_front_porch = mode_in->vsync_start - mode_in->vdisplay;
5116+ timing_out->v_sync_width = mode_in->vsync_end - mode_in->vsync_start;
5117+ timing_out->pix_clk_100hz = mode_in->clock * 10;
5118+
50175119 timing_out->aspect_ratio = get_aspect_ratio(mode_in);
50185120
50195121 stream->output_color_space = get_output_color_space(timing_out);
@@ -5180,6 +5282,86 @@ static void dm_enable_per_frame_crtc_master_sync(struct dc_state *context) · 160 unmodified lines
51805282 set_master_stream(context->streams, context->stream_count);
51815283 }
51825284
5285+static struct drm_display_mode *
5286+get_highest_refresh_rate_mode(struct amdgpu_dm_connector *aconnector,
5287+ bool use_probed_modes)
5288+{
5289+ struct drm_display_mode *m, *m_pref = NULL;
5290+ u16 current_refresh, highest_refresh;
5291+ struct list_head *list_head = use_probed_modes ?
5292+ &aconnector->base.probed_modes :
5293+ &aconnector->base.modes;
5294+
5295+ if (aconnector->freesync_vid_base.clock != 0)
5296+ return &aconnector->freesync_vid_base;
5297+
5298+ /* Find the preferred mode */
5299+ list_for_each_entry (m, list_head, head) {
5300+ if (m->type & DRM_MODE_TYPE_PREFERRED) {
5301+ m_pref = m;
5302+ break;
5303+ }
5304+ }
5305+
5306+ if (!m_pref) {
5307+ /* Probably an EDID with no preferred mode. Fallback to first entry */
5308+ m_pref = list_first_entry_or_null(
5309+ &aconnector->base.modes, struct drm_display_mode, head);
5310+ if (!m_pref) {
5311+ DRM_DEBUG_DRIVER("No preferred mode found in EDID\n");
5312+ return NULL;
5313+ }
5314+ }
5315+
5316+ highest_refresh = drm_mode_vrefresh(m_pref);
5317+
5318+ /*
5319+ * Find the mode with highest refresh rate with same resolution.
5320+ * For some monitors, preferred mode is not the mode with highest
5321+ * supported refresh rate.
5322+ */
5323+ list_for_each_entry (m, list_head, head) {
5324+ current_refresh = drm_mode_vrefresh(m);
5325+
5326+ if (m->hdisplay == m_pref->hdisplay &&
5327+ m->vdisplay == m_pref->vdisplay &&
5328+ highest_refresh < current_refresh) {
5329+ highest_refresh = current_refresh;
5330+ m_pref = m;
5331+ }
5332+ }
5333+
5334+ aconnector->freesync_vid_base = *m_pref;
5335+ return m_pref;
5336+}
5337+
5338+static bool is_freesync_video_mode(struct drm_display_mode *mode,
5339+ struct amdgpu_dm_connector *aconnector)
5340+{
5341+ struct drm_display_mode *high_mode;
5342+ int timing_diff;
5343+
5344+ high_mode = get_highest_refresh_rate_mode(aconnector, false);
5345+ if (!high_mode || !mode)
5346+ return false;
5347+
5348+ timing_diff = high_mode->vtotal - mode->vtotal;
5349+
5350+ if (high_mode->clock == 0 || high_mode->clock != mode->clock ||
5351+ high_mode->hdisplay != mode->hdisplay ||
5352+ high_mode->vdisplay != mode->vdisplay ||
5353+ high_mode->hsync_start != mode->hsync_start ||
5354+ high_mode->hsync_end != mode->hsync_end ||
5355+ high_mode->htotal != mode->htotal ||
5356+ high_mode->hskew != mode->hskew ||
5357+ high_mode->vscan != mode->vscan ||
5358+ high_mode->vsync_start - mode->vsync_start != timing_diff ||
5359+ high_mode->vsync_end - mode->vsync_end != timing_diff)
5360+ return false;
5361+ else
5362+ return true;
5363+}
5364+
51835365 static struct dc_stream_state *
51845366 create_stream_for_sink(struct amdgpu_dm_connector *aconnector,
51855367 const struct drm_display_mode *drm_mode,
@@ -5193,8 +5375,10 @@ create_stream_for_sink(struct amdgpu_dm_connector *aconnector, · 7 unmodified lines
51935375 dm_state ? &dm_state->base : NULL;
51945376 struct dc_stream_state *stream = NULL;
51955377 struct drm_display_mode mode = *drm_mode;
5378+ struct drm_display_mode saved_mode;
5379+ struct drm_display_mode *freesync_mode = NULL;
51965380 bool native_mode_found = false;
5197− bool scale = dm_state ? (dm_state->scaling != RMX_OFF) : false;
5381+ bool recalculate_timing = dm_state ? (dm_state->scaling != RMX_OFF) : false;
51985382 int mode_refresh;
51995383 int preferred_refresh = 0;
52005384 #if defined(CONFIG_DRM_AMD_DC_DCN)
@@ -5202,6 +5386,9 @@ create_stream_for_sink(struct amdgpu_dm_connector *aconnector, · 1 unmodified line
52025386 uint32_t link_bandwidth_kbps;
52035387 #endif
52045388 struct dc_sink *sink = NULL;
5389+
5390+ memset(&saved_mode, 0, sizeof(saved_mode));
5391+
52055392 if (aconnector == NULL) {
52065393 DRM_ERROR("aconnector is NULL!\n");
52075394 return stream;
@@ -5254,25 +5441,38 @@ create_stream_for_sink(struct amdgpu_dm_connector *aconnector, · 46 unmodified lines
52545441 */
52555442 DRM_DEBUG_DRIVER("No preferred mode found\n");
52565443 } else {
5257− decide_crtc_timing_for_drm_display_mode(
5444+ recalculate_timing |= amdgpu_freesync_vid_mode &&
5445+ is_freesync_video_mode(&mode, aconnector);
5446+ if (recalculate_timing) {
5447+ freesync_mode = get_highest_refresh_rate_mode(aconnector, false);
5448+ saved_mode = mode;
5449+ mode = *freesync_mode;
5450+ } else {
5451+ decide_crtc_timing_for_drm_display_mode(
52585452 &mode, preferred_mode,
52595453 dm_state ? (dm_state->scaling != RMX_OFF) : false);
5454+ }
5455+
52605456 preferred_refresh = drm_mode_vrefresh(preferred_mode);
52615457 }
52625458
5263− if (!dm_state)
5459+ if (recalculate_timing)
5460+ drm_mode_set_crtcinfo(&saved_mode, 0);
5461+ else
52645462 drm_mode_set_crtcinfo(&mode, 0);
52655463
5266− /*
5464+ /*
52675465 * If scaling is enabled and refresh rate didn't change
52685466 * we copy the vic and polarities of the old timings
52695467 */
5270− if (!scale || mode_refresh != preferred_refresh)
5271− fill_stream_properties_from_drm_display_mode(stream,
5272− &mode, &aconnector->base, con_state, NULL, requested_bpc);
5468+ if (!recalculate_timing || mode_refresh != preferred_refresh)
5469+ fill_stream_properties_from_drm_display_mode(
5470+ stream, &mode, &aconnector->base, con_state, NULL,
5471+ requested_bpc);
52735472 else
5274− fill_stream_properties_from_drm_display_mode(stream,
5275− &mode, &aconnector->base, con_state, old_stream, requested_bpc);
5473+ fill_stream_properties_from_drm_display_mode(
5474+ stream, &mode, &aconnector->base, con_state, old_stream,
5475+ requested_bpc);
52765476
52775477 stream->timing.flags.DSC = 0;
52785478
@@ -5409,15 +5609,22 @@ dm_crtc_duplicate_state(struct drm_crtc *crtc) · 130 unmodified lines
54095609 state->abm_level = cur->abm_level;
54105610 state->vrr_supported = cur->vrr_supported;
54115611 state->freesync_config = cur->freesync_config;
5412− state->crc_src = cur->crc_src;
54135612 state->cm_has_degamma = cur->cm_has_degamma;
54145613 state->cm_is_degamma_srgb = cur->cm_is_degamma_srgb;
5415−
54165614 /* TODO Duplicate dc_stream after objects are stream object is flattened */
54175615
54185616 return &state->base;
54195617 }
54205618
5619+#ifdef CONFIG_DRM_AMD_SECURE_DISPLAY
5620+static int amdgpu_dm_crtc_late_register(struct drm_crtc *crtc)
5621+{
5622+ crtc_debugfs_init(crtc);
5623+
5624+ return 0;
5625+}
5626+#endif
5627+
54215628 static inline int dm_set_vupdate_irq(struct drm_crtc *crtc, bool enable)
54225629 {
54235630 enum dc_irq_source irq_source;
@@ -5503,6 +5710,9 @@ static const struct drm_crtc_funcs amdgpu_dm_crtc_funcs = { · 79 unmodified lines
55035710 .enable_vblank = dm_enable_vblank,
55045711 .disable_vblank = dm_disable_vblank,
55055712 .get_vblank_timestamp = drm_crtc_vblank_helper_get_vblank_timestamp,
5713+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
5714+ .late_register = amdgpu_dm_crtc_late_register,
5715+#endif
55065716 };
55075717
55085718 static enum drm_connector_status
@@ -6488,13 +6698,17 @@ static int dm_plane_helper_check_state(struct drm_plane_state *state, · 979 unmodified lines
64886698 else if (state->crtc_y + state->crtc_h > new_crtc_state->mode.crtc_vdisplay)
64896699 viewport_height = new_crtc_state->mode.crtc_vdisplay - state->crtc_y;
64906700
6491− /* If completely outside of screen, viewport_width and/or viewport_height will be negative,
6492− * which is still OK to satisfy the condition below, thereby also covering these cases
6493− * (when plane is completely outside of screen).
6494− * x2 for width is because of pipe-split.
6495− */
6496− if (viewport_width < MIN_VIEWPORT_SIZE*2 || viewport_height < MIN_VIEWPORT_SIZE)
6701+ if (viewport_width < 0 || viewport_height < 0) {
6702+ DRM_DEBUG_ATOMIC("Plane completely outside of screen\n");
6703+ return -EINVAL;
6704+ } else if (viewport_width < MIN_VIEWPORT_SIZE*2) { /* x2 for width is because of pipe-split. */
6705+ DRM_DEBUG_ATOMIC("Viewport width %d smaller than %d\n", viewport_width, MIN_VIEWPORT_SIZE*2);
64976706 return -EINVAL;
6707+ } else if (viewport_height < MIN_VIEWPORT_SIZE) {
6708+ DRM_DEBUG_ATOMIC("Viewport height %d smaller than %d\n", viewport_height, MIN_VIEWPORT_SIZE);
6709+ return -EINVAL;
6710+ }
6711+
64986712 }
64996713
65006714 /* Get min/max allowed scaling factors from plane caps. */
@@ -6975,11 +7189,118 @@ static void amdgpu_dm_connector_ddc_get_modes(struct drm_connector *connector, · 474 unmodified lines
69757189 */
69767190 drm_mode_sort(&connector->probed_modes);
69777191 amdgpu_dm_get_native_mode(connector);
7192+
7193+ /* Freesync capabilities are reset by calling
7194+ * drm_add_edid_modes() and need to be
7195+ * restored here.
7196+ */
7197+ amdgpu_dm_update_freesync_caps(connector, edid);
69787198 } else {
69797199 amdgpu_dm_connector->num_modes = 0;
69807200 }
69817201 }
69827202
7203+static bool is_duplicate_mode(struct amdgpu_dm_connector *aconnector,
7204+ struct drm_display_mode *mode)
7205+{
7206+ struct drm_display_mode *m;
7207+
7208+ list_for_each_entry (m, &aconnector->base.probed_modes, head) {
7209+ if (drm_mode_equal(m, mode))
7210+ return true;
7211+ }
7212+
7213+ return false;
7214+}
7215+
7216+static uint add_fs_modes(struct amdgpu_dm_connector *aconnector)
7217+{
7218+ const struct drm_display_mode *m;
7219+ struct drm_display_mode *new_mode;
7220+ uint i;
7221+ uint32_t new_modes_count = 0;
7222+
7223+ /* Standard FPS values
7224+ *
7225+ * 23.976 - TV/NTSC
7226+ * 24 - Cinema
7227+ * 25 - TV/PAL
7228+ * 29.97 - TV/NTSC
7229+ * 30 - TV/NTSC
7230+ * 48 - Cinema HFR
7231+ * 50 - TV/PAL
7232+ * 60 - Commonly used
7233+ * 48,72,96 - Multiples of 24
7234+ */
7235+ const uint32_t common_rates[] = { 23976, 24000, 25000, 29970, 30000,
7236+ 48000, 50000, 60000, 72000, 96000 };
7237+
7238+ /*
7239+ * Find mode with highest refresh rate with the same resolution
7240+ * as the preferred mode. Some monitors report a preferred mode
7241+ * with lower resolution than the highest refresh rate supported.
7242+ */
7243+
7244+ m = get_highest_refresh_rate_mode(aconnector, true);
7245+ if (!m)
7246+ return 0;
7247+
7248+ for (i = 0; i < ARRAY_SIZE(common_rates); i++) {
7249+ uint64_t target_vtotal, target_vtotal_diff;
7250+ uint64_t num, den;
7251+
7252+ if (drm_mode_vrefresh(m) * 1000 < common_rates[i])
7253+ continue;
7254+
7255+ if (common_rates[i] < aconnector->min_vfreq * 1000 ||
7256+ common_rates[i] > aconnector->max_vfreq * 1000)
7257+ continue;
7258+
7259+ num = (unsigned long long)m->clock * 1000 * 1000;
7260+ den = common_rates[i] * (unsigned long long)m->htotal;
7261+ target_vtotal = div_u64(num, den);
7262+ target_vtotal_diff = target_vtotal - m->vtotal;
7263+
7264+ /* Check for illegal modes */
7265+ if (m->vsync_start + target_vtotal_diff < m->vdisplay ||
7266+ m->vsync_end + target_vtotal_diff < m->vsync_start ||
7267+ m->vtotal + target_vtotal_diff < m->vsync_end)
7268+ continue;
7269+
7270+ new_mode = drm_mode_duplicate(aconnector->base.dev, m);
7271+ if (!new_mode)
7272+ goto out;
7273+
7274+ new_mode->vtotal += (u16)target_vtotal_diff;
7275+ new_mode->vsync_start += (u16)target_vtotal_diff;
7276+ new_mode->vsync_end += (u16)target_vtotal_diff;
7277+ new_mode->type &= ~DRM_MODE_TYPE_PREFERRED;
7278+ new_mode->type |= DRM_MODE_TYPE_DRIVER;
7279+
7280+ if (!is_duplicate_mode(aconnector, new_mode)) {
7281+ drm_mode_probed_add(&aconnector->base, new_mode);
7282+ new_modes_count += 1;
7283+ } else
7284+ drm_mode_destroy(aconnector->base.dev, new_mode);
7285+ }
7286+ out:
7287+ return new_modes_count;
7288+}
7289+
7290+static void amdgpu_dm_connector_add_freesync_modes(struct drm_connector *connector,
7291+ struct edid *edid)
7292+{
7293+ struct amdgpu_dm_connector *amdgpu_dm_connector =
7294+ to_amdgpu_dm_connector(connector);
7295+
7296+ if (!(amdgpu_freesync_vid_mode && edid))
7297+ return;
7298+
7299+ if (amdgpu_dm_connector->max_vfreq - amdgpu_dm_connector->min_vfreq > 10)
7300+ amdgpu_dm_connector->num_modes +=
7301+ add_fs_modes(amdgpu_dm_connector);
7302+}
7303+
69837304 static int amdgpu_dm_connector_get_modes(struct drm_connector *connector)
69847305 {
69857306 struct amdgpu_dm_connector *amdgpu_dm_connector =
@@ -6995,6 +7316,7 @@ static int amdgpu_dm_connector_get_modes(struct drm_connector *connector) · 9 unmodified lines
69957316 } else {
69967317 amdgpu_dm_connector_ddc_get_modes(connector, edid);
69977318 amdgpu_dm_connector_add_common_modes(encoder, connector);
7319+ amdgpu_dm_connector_add_freesync_modes(connector, edid);
69987320 }
69997321 amdgpu_dm_fbc_init(connector);
70007322
@@ -7299,8 +7621,19 @@ static void manage_dm_interrupts(struct amdgpu_device *adev, · 298 unmodified lines
72997621 adev,
73007622 &adev->pageflip_irq,
73017623 irq_type);
7624+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
7625+ amdgpu_irq_get(
7626+ adev,
7627+ &adev->vline0_irq,
7628+ irq_type);
7629+#endif
73027630 } else {
7303−
7631+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
7632+ amdgpu_irq_put(
7633+ adev,
7634+ &adev->vline0_irq,
7635+ irq_type);
7636+#endif
73047637 amdgpu_irq_put(
73057638 adev,
73067639 &adev->pageflip_irq,
@@ -7424,10 +7757,6 @@ static int get_cursor_position(struct drm_plane *plane, struct drm_crtc *crtc, · 117 unmodified lines
74247757 int x, y;
74257758 int xorigin = 0, yorigin = 0;
74267759
7427− position->enable = false;
7428− position->x = 0;
7429− position->y = 0;
7430−
74317760 if (!crtc || !plane->state->fb)
74327761 return 0;
74337762
@@ -7474,7 +7803,7 @@ static void handle_cursor_update(struct drm_plane *plane, · 40 unmodified lines
74747803 struct dm_crtc_state *crtc_state = crtc ? to_dm_crtc_state(crtc->state) : NULL;
74757804 struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
74767805 uint64_t address = afb ? afb->address : 0;
7477− struct dc_cursor_position position;
7806+ struct dc_cursor_position position = {0};
74787807 struct dc_cursor_attributes attributes;
74797808 int ret;
74807809
@@ -7559,6 +7888,7 @@ static void update_freesync_state_on_stream( · 78 unmodified lines
75597888 struct amdgpu_device *adev = dm->adev;
75607889 struct amdgpu_crtc *acrtc = to_amdgpu_crtc(new_crtc_state->base.crtc);
75617890 unsigned long flags;
7891+ bool pack_sdp_v1_3 = false;
75627892
75637893 if (!new_stream)
75647894 return;
@@ -7600,7 +7930,8 @@ static void update_freesync_state_on_stream( · 35 unmodified lines
76007930 &vrr_params,
76017931 PACKET_TYPE_VRR,
76027932 TRANSFER_FUNC_UNKNOWN,
7603− &vrr_infopacket);
7933+ &vrr_infopacket,
7934+ pack_sdp_v1_3);
76047935
76057936 new_crtc_state->freesync_timing_changed |=
76067937 (memcmp(&acrtc->dm_irq_params.vrr_params.adjust,
@@ -7654,9 +7985,22 @@ static void update_stream_irq_parameters( · 47 unmodified lines
76547985 if (new_crtc_state->vrr_supported &&
76557986 config.min_refresh_in_uhz &&
76567987 config.max_refresh_in_uhz) {
7657− config.state = new_crtc_state->base.vrr_enabled ?
7658− VRR_STATE_ACTIVE_VARIABLE :
7659− VRR_STATE_INACTIVE;
7988+ /*
7989+ * if freesync compatible mode was set, config.state will be set
7990+ * in atomic check
7991+ */
7992+ if (config.state == VRR_STATE_ACTIVE_FIXED && config.fixed_refresh_in_uhz &&
7993+ (!drm_atomic_crtc_needs_modeset(&new_crtc_state->base) ||
7994+ new_crtc_state->freesync_config.state == VRR_STATE_ACTIVE_FIXED)) {
7995+ vrr_params.max_refresh_in_uhz = config.max_refresh_in_uhz;
7996+ vrr_params.min_refresh_in_uhz = config.min_refresh_in_uhz;
7997+ vrr_params.fixed_refresh_in_uhz = config.fixed_refresh_in_uhz;
7998+ vrr_params.state = VRR_STATE_ACTIVE_FIXED;
7999+ } else {
8000+ config.state = new_crtc_state->base.vrr_enabled ?
8001+ VRR_STATE_ACTIVE_VARIABLE :
8002+ VRR_STATE_INACTIVE;
8003+ }
76608004 } else {
76618005 config.state = VRR_STATE_UNSUPPORTED;
76628006 }
@@ -7977,8 +8321,7 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state, · 314 unmodified lines
79778321 * re-adjust the min/max bounds now that DC doesn't handle this
79788322 * as part of commit.
79798323 */
7980− if (amdgpu_dm_vrr_active(dm_old_crtc_state) !=
7981− amdgpu_dm_vrr_active(acrtc_state)) {
8324+ if (is_dc_timing_adjust_needed(dm_old_crtc_state, acrtc_state)) {
79828325 spin_lock_irqsave(&pcrtc->dev->event_lock, flags);
79838326 dc_stream_adjust_vmin_vmax(
79848327 dm->dc, acrtc_state->stream,
@@ -8263,6 +8606,7 @@ static void amdgpu_dm_atomic_commit_tail(struct drm_atomic_state *state) · 278 unmodified lines
82638606 /* i.e. reset mode */
82648607 if (dm_old_crtc_state->stream)
82658608 remove_stream(adev, acrtc, dm_old_crtc_state->stream);
8609+
82668610 mode_set_reset_required = true;
82678611 }
82688612 } /* for_each_crtc_in_state() */
@@ -8321,8 +8665,7 @@ static void amdgpu_dm_atomic_commit_tail(struct drm_atomic_state *state) · 52 unmodified lines
83218665 hdcp_update_display(
83228666 adev->dm.hdcp_workqueue, aconnector->dc_link->link_index, aconnector,
83238667 new_con_state->hdcp_content_type,
8324− new_con_state->content_protection == DRM_MODE_CONTENT_PROTECTION_DESIRED ? true
8325− : false);
8668+ new_con_state->content_protection == DRM_MODE_CONTENT_PROTECTION_DESIRED);
83268669 }
83278670 #endif
83288671
@@ -8432,7 +8775,10 @@ static void amdgpu_dm_atomic_commit_tail(struct drm_atomic_state *state) · 103 unmodified lines
84328775 */
84338776 for_each_oldnew_crtc_in_state(state, crtc, old_crtc_state, new_crtc_state, i) {
84348777 struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
8435−
8778+#ifdef CONFIG_DEBUG_FS
8779+ bool configure_crc = false;
8780+ enum amdgpu_dm_pipe_crc_source cur_crc_src;
8781+#endif
84368782 dm_new_crtc_state = to_dm_crtc_state(new_crtc_state);
84378783
84388784 if (new_crtc_state->active &&
@@ -8448,12 +8794,21 @@ static void amdgpu_dm_atomic_commit_tail(struct drm_atomic_state *state) · 9 unmodified lines
84488794 * settings for the stream.
84498795 */
84508796 dm_new_crtc_state = to_dm_crtc_state(new_crtc_state);
8797+ spin_lock_irqsave(&adev_to_drm(adev)->event_lock, flags);
8798+ cur_crc_src = acrtc->dm_irq_params.crc_src;
8799+ spin_unlock_irqrestore(&adev_to_drm(adev)->event_lock, flags);
8800+
8801+ if (amdgpu_dm_is_valid_crc_source(cur_crc_src)) {
8802+ configure_crc = true;
8803+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
8804+ if (amdgpu_dm_crc_window_is_activated(crtc))
8805+ configure_crc = false;
8806+#endif
8807+ }
84518808
8452− if (amdgpu_dm_is_valid_crc_source(dm_new_crtc_state->crc_src)) {
8809+ if (configure_crc)
84538810 amdgpu_dm_crtc_configure_crc_source(
8454− crtc, dm_new_crtc_state,
8455− dm_new_crtc_state->crc_src);
8456− }
8811+ crtc, dm_new_crtc_state, cur_crc_src);
84578812 #endif
84588813 }
84598814 }
@@ -8662,6 +9017,7 @@ static void get_freesync_config_for_crtc( · 202 unmodified lines
86629017 to_amdgpu_dm_connector(new_con_state->base.connector);
86639018 struct drm_display_mode *mode = &new_crtc_state->base.mode;
86649019 int vrefresh = drm_mode_vrefresh(mode);
9020+ bool fs_vid_mode = false;
86659021
86669022 new_crtc_state->vrr_supported = new_con_state->freesync_capable &&
86679023 vrefresh >= aconnector->min_vfreq &&
@@ -8669,17 +9025,24 @@ static void get_freesync_config_for_crtc( · 1 unmodified line
86699025
86709026 if (new_crtc_state->vrr_supported) {
86719027 new_crtc_state->stream->ignore_msa_timing_param = true;
8672− config.state = new_crtc_state->base.vrr_enabled ?
8673− VRR_STATE_ACTIVE_VARIABLE :
8674− VRR_STATE_INACTIVE;
8675− config.min_refresh_in_uhz =
8676− aconnector->min_vfreq * 1000000;
8677− config.max_refresh_in_uhz =
8678− aconnector->max_vfreq * 1000000;
9028+ fs_vid_mode = new_crtc_state->freesync_config.state == VRR_STATE_ACTIVE_FIXED;
9029+
9030+ config.min_refresh_in_uhz = aconnector->min_vfreq * 1000000;
9031+ config.max_refresh_in_uhz = aconnector->max_vfreq * 1000000;
86799032 config.vsif_supported = true;
86809033 config.btr = true;
8681− }
86829034
9035+ if (fs_vid_mode) {
9036+ config.state = VRR_STATE_ACTIVE_FIXED;
9037+ config.fixed_refresh_in_uhz = new_crtc_state->freesync_config.fixed_refresh_in_uhz;
9038+ goto out;
9039+ } else if (new_crtc_state->base.vrr_enabled) {
9040+ config.state = VRR_STATE_ACTIVE_VARIABLE;
9041+ } else {
9042+ config.state = VRR_STATE_INACTIVE;
9043+ }
9044+ }
9045+out:
86839046 new_crtc_state->freesync_config = config;
86849047 }
86859048
@@ -8692,6 +9055,50 @@ static void reset_freesync_config_for_crtc( · 6 unmodified lines
86929055 sizeof(new_crtc_state->vrr_infopacket));
86939056 }
86949057
9058+static bool
9059+is_timing_unchanged_for_freesync(struct drm_crtc_state *old_crtc_state,
9060+ struct drm_crtc_state *new_crtc_state)
9061+{
9062+ struct drm_display_mode old_mode, new_mode;
9063+
9064+ if (!old_crtc_state || !new_crtc_state)
9065+ return false;
9066+
9067+ old_mode = old_crtc_state->mode;
9068+ new_mode = new_crtc_state->mode;
9069+
9070+ if (old_mode.clock == new_mode.clock &&
9071+ old_mode.hdisplay == new_mode.hdisplay &&
9072+ old_mode.vdisplay == new_mode.vdisplay &&
9073+ old_mode.htotal == new_mode.htotal &&
9074+ old_mode.vtotal != new_mode.vtotal &&
9075+ old_mode.hsync_start == new_mode.hsync_start &&
9076+ old_mode.vsync_start != new_mode.vsync_start &&
9077+ old_mode.hsync_end == new_mode.hsync_end &&
9078+ old_mode.vsync_end != new_mode.vsync_end &&
9079+ old_mode.hskew == new_mode.hskew &&
9080+ old_mode.vscan == new_mode.vscan &&
9081+ (old_mode.vsync_end - old_mode.vsync_start) ==
9082+ (new_mode.vsync_end - new_mode.vsync_start))
9083+ return true;
9084+
9085+ return false;
9086+}
9087+
9088+static void set_freesync_fixed_config(struct dm_crtc_state *dm_new_crtc_state) {
9089+ uint64_t num, den, res;
9090+ struct drm_crtc_state *new_crtc_state = &dm_new_crtc_state->base;
9091+
9092+ dm_new_crtc_state->freesync_config.state = VRR_STATE_ACTIVE_FIXED;
9093+
9094+ num = (unsigned long long)new_crtc_state->mode.clock * 1000 * 1000000;
9095+ den = (unsigned long long)new_crtc_state->mode.htotal *
9096+ (unsigned long long)new_crtc_state->mode.vtotal;
9097+
9098+ res = div_u64(num, den);
9099+ dm_new_crtc_state->freesync_config.fixed_refresh_in_uhz = res;
9100+}
9101+
86959102 static int dm_update_crtc_state(struct amdgpu_display_manager *dm,
86969103 struct drm_atomic_state *state,
86979104 struct drm_crtc *crtc,
@@ -8782,6 +9189,11 @@ static int dm_update_crtc_state(struct amdgpu_display_manager *dm, · 84 unmodified lines
87829189 * TODO: Refactor this function to allow this check to work
87839190 * in all conditions.
87849191 */
9192+ if (amdgpu_freesync_vid_mode &&
9193+ dm_new_crtc_state->stream &&
9194+ is_timing_unchanged_for_freesync(new_crtc_state, old_crtc_state))
9195+ goto skip_modeset;
9196+
87859197 if (dm_new_crtc_state->stream &&
87869198 dc_is_stream_unchanged(new_stream, dm_old_crtc_state->stream) &&
87879199 dc_is_stream_scaling_unchanged(new_stream, dm_old_crtc_state->stream)) {
@@ -8813,6 +9225,24 @@ static int dm_update_crtc_state(struct amdgpu_display_manager *dm, · 25 unmodified lines
88139225 if (!dm_old_crtc_state->stream)
88149226 goto skip_modeset;
88159227
9228+ if (amdgpu_freesync_vid_mode && dm_new_crtc_state->stream &&
9229+ is_timing_unchanged_for_freesync(new_crtc_state,
9230+ old_crtc_state)) {
9231+ new_crtc_state->mode_changed = false;
9232+ DRM_DEBUG_DRIVER(
9233+ "Mode change not required for front porch change, "
9234+ "setting mode_changed to %d",
9235+ new_crtc_state->mode_changed);
9236+
9237+ set_freesync_fixed_config(dm_new_crtc_state);
9238+
9239+ goto skip_modeset;
9240+ } else if (amdgpu_freesync_vid_mode && aconnector &&
9241+ is_freesync_video_mode(&new_crtc_state->mode,
9242+ aconnector)) {
9243+ set_freesync_fixed_config(dm_new_crtc_state);
9244+ }
9245+
88169246 ret = dm_atomic_get_state(state, &dm_state);
88179247 if (ret)
88189248 goto fail;
@@ -9390,7 +9820,7 @@ static int amdgpu_dm_atomic_check(struct drm_device *dev, · 571 unmodified lines
93909820 }
93919821
93929822 #if defined(CONFIG_DRM_AMD_DC_DCN)
9393− if (adev->asic_type >= CHIP_NAVI10) {
9823+ if (dc_resource_is_dsc_encoding_supported(dc)) {
93949824 for_each_oldnew_crtc_in_state(state, crtc, old_crtc_state, new_crtc_state, i) {
93959825 if (drm_atomic_crtc_needs_modeset(new_crtc_state)) {
93969826 ret = add_affected_mst_dsc_crtcs(state, crtc);
@@ -9696,11 +10126,85 @@ static bool is_dp_capable_without_timing_msa(struct dc *dc, · 299 unmodified lines
969610126
969710127 return capable;
969810128 }
10129+
10130+static bool parse_edid_cea(struct amdgpu_dm_connector *aconnector,
10131+ uint8_t *edid_ext, int len,
10132+ struct amdgpu_hdmi_vsdb_info *vsdb_info)
10133+{
10134+ int i;
10135+ struct amdgpu_device *adev = drm_to_adev(aconnector->base.dev);
10136+ struct dc *dc = adev->dm.dc;
10137+
10138+ /* send extension block to DMCU for parsing */
10139+ for (i = 0; i < len; i += 8) {
10140+ bool res;
10141+ int offset;
10142+
10143+ /* send 8 bytes a time */
10144+ if (!dc_edid_parser_send_cea(dc, i, len, &edid_ext[i], 8))
10145+ return false;
10146+
10147+ if (i+8 == len) {
10148+ /* EDID block sent completed, expect result */
10149+ int version, min_rate, max_rate;
10150+
10151+ res = dc_edid_parser_recv_amd_vsdb(dc, &version, &min_rate, &max_rate);
10152+ if (res) {
10153+ /* amd vsdb found */
10154+ vsdb_info->freesync_supported = 1;
10155+ vsdb_info->amd_vsdb_version = version;
10156+ vsdb_info->min_refresh_rate_hz = min_rate;
10157+ vsdb_info->max_refresh_rate_hz = max_rate;
10158+ return true;
10159+ }
10160+ /* not amd vsdb */
10161+ return false;
10162+ }
10163+
10164+ /* check for ack*/
10165+ res = dc_edid_parser_recv_cea_ack(dc, &offset);
10166+ if (!res)
10167+ return false;
10168+ }
10169+
10170+ return false;
10171+}
10172+
10173+static int parse_hdmi_amd_vsdb(struct amdgpu_dm_connector *aconnector,
10174+ struct edid *edid, struct amdgpu_hdmi_vsdb_info *vsdb_info)
10175+{
10176+ uint8_t *edid_ext = NULL;
10177+ int i;
10178+ bool valid_vsdb_found = false;
10179+
10180+ /*----- drm_find_cea_extension() -----*/
10181+ /* No EDID or EDID extensions */
10182+ if (edid == NULL || edid->extensions == 0)
10183+ return -ENODEV;
10184+
10185+ /* Find CEA extension */
10186+ for (i = 0; i < edid->extensions; i++) {
10187+ edid_ext = (uint8_t *)edid + EDID_LENGTH * (i + 1);
10188+ if (edid_ext[0] == CEA_EXT)
10189+ break;
10190+ }
10191+
10192+ if (i == edid->extensions)
10193+ return -ENODEV;
10194+
10195+ /*----- cea_db_offsets() -----*/
10196+ if (edid_ext[0] != CEA_EXT)
10197+ return -ENODEV;
10198+
10199+ valid_vsdb_found = parse_edid_cea(aconnector, edid_ext, EDID_LENGTH, vsdb_info);
10200+
10201+ return valid_vsdb_found ? i : -ENODEV;
10202+}
10203+
969910204 void amdgpu_dm_update_freesync_caps(struct drm_connector *connector,
970010205 struct edid *edid)
970110206 {
9702− int i;
9703− bool edid_check_required;
10207+ int i = 0;
970410208 struct detailed_timing *timing;
970510209 struct detailed_non_pixel *data;
970610210 struct detailed_data_monitor_range *range;
@@ -9711,6 +10215,7 @@ void amdgpu_dm_update_freesync_caps(struct drm_connector *connector, · 4 unmodified lines
971110215 struct drm_device *dev = connector->dev;
971210216 struct amdgpu_device *adev = drm_to_adev(dev);
971310217 bool freesync_capable = false;
10218+ struct amdgpu_hdmi_vsdb_info vsdb_info = {0};
971410219
971510220 if (!connector->state) {
971610221 DRM_ERROR("%s - Connector has no state", __func__);
@@ -9729,60 +10234,75 @@ void amdgpu_dm_update_freesync_caps(struct drm_connector *connector, · 12 unmodified lines
972910234
973010235 dm_con_state = to_dm_connector_state(connector->state);
973110236
9732− edid_check_required = false;
973310237 if (!amdgpu_dm_connector->dc_sink) {
973410238 DRM_ERROR("dc_sink NULL, could not add free_sync module.\n");
973510239 goto update;
973610240 }
973710241 if (!adev->dm.freesync_module)
973810242 goto update;
9739− /*
9740− * if edid non zero restrict freesync only for dp and edp
9741− */
9742− if (edid) {
9743− if (amdgpu_dm_connector->dc_sink->sink_signal == SIGNAL_TYPE_DISPLAY_PORT
9744− || amdgpu_dm_connector->dc_sink->sink_signal == SIGNAL_TYPE_EDP) {
10243+
10244+
10245+ if (amdgpu_dm_connector->dc_sink->sink_signal == SIGNAL_TYPE_DISPLAY_PORT
10246+ || amdgpu_dm_connector->dc_sink->sink_signal == SIGNAL_TYPE_EDP) {
10247+ bool edid_check_required = false;
10248+
10249+ if (edid) {
974510250 edid_check_required = is_dp_capable_without_timing_msa(
974610251 adev->dm.dc,
974710252 amdgpu_dm_connector);
974810253 }
9749− }
9750− if (edid_check_required == true && (edid->version > 1 ||
9751− (edid->version == 1 && edid->revision > 1))) {
9752− for (i = 0; i < 4; i++) {
975310254
9754− timing = &edid->detailed_timings[i];
9755− data = &timing->data.other_data;
9756− range = &data->data.range;
9757− /*
9758− * Check if monitor has continuous frequency mode
9759− */
9760− if (data->type != EDID_DETAIL_MONITOR_RANGE)
9761− continue;
9762− /*
9763− * Check for flag range limits only. If flag == 1 then
9764− * no additional timing information provided.
9765− * Default GTF, GTF Secondary curve and CVT are not
9766− * supported
9767− */
9768− if (range->flags != 1)
9769− continue;
10255+ if (edid_check_required == true && (edid->version > 1 ||
10256+ (edid->version == 1 && edid->revision > 1))) {
10257+ for (i = 0; i < 4; i++) {
977010258
9771− amdgpu_dm_connector->min_vfreq = range->min_vfreq;
9772− amdgpu_dm_connector->max_vfreq = range->max_vfreq;
9773− amdgpu_dm_connector->pixel_clock_mhz =
9774− range->pixel_clock_mhz * 10;
10259+ timing = &edid->detailed_timings[i];
10260+ data = &timing->data.other_data;
10261+ range = &data->data.range;
10262+ /*
10263+ * Check if monitor has continuous frequency mode
10264+ */
10265+ if (data->type != EDID_DETAIL_MONITOR_RANGE)
10266+ continue;
10267+ /*
10268+ * Check for flag range limits only. If flag == 1 then
10269+ * no additional timing information provided.
10270+ * Default GTF, GTF Secondary curve and CVT are not
10271+ * supported
10272+ */
10273+ if (range->flags != 1)
10274+ continue;
977510275
9776− connector->display_info.monitor_range.min_vfreq = range->min_vfreq;
9777− connector->display_info.monitor_range.max_vfreq = range->max_vfreq;
10276+ amdgpu_dm_connector->min_vfreq = range->min_vfreq;
10277+ amdgpu_dm_connector->max_vfreq = range->max_vfreq;
10278+ amdgpu_dm_connector->pixel_clock_mhz =
10279+ range->pixel_clock_mhz * 10;
977810280
9779− break;
9780− }
10281+ connector->display_info.monitor_range.min_vfreq = range->min_vfreq;
10282+ connector->display_info.monitor_range.max_vfreq = range->max_vfreq;
978110283
9782− if (amdgpu_dm_connector->max_vfreq -
9783− amdgpu_dm_connector->min_vfreq > 10) {
10284+ break;
10285+ }
978410286
9785− freesync_capable = true;
10287+ if (amdgpu_dm_connector->max_vfreq -
10288+ amdgpu_dm_connector->min_vfreq > 10) {
10289+
10290+ freesync_capable = true;
10291+ }
10292+ }
10293+ } else if (edid && amdgpu_dm_connector->dc_sink->sink_signal == SIGNAL_TYPE_HDMI_TYPE_A) {
10294+ i = parse_hdmi_amd_vsdb(amdgpu_dm_connector, edid, &vsdb_info);
10295+ if (i >= 0 && vsdb_info.freesync_supported) {
10296+ timing = &edid->detailed_timings[i];
10297+ data = &timing->data.other_data;
10298+
10299+ amdgpu_dm_connector->min_vfreq = vsdb_info.min_refresh_rate_hz;
10300+ amdgpu_dm_connector->max_vfreq = vsdb_info.max_refresh_rate_hz;
10301+ if (amdgpu_dm_connector->max_vfreq - amdgpu_dm_connector->min_vfreq > 10)
10302+ freesync_capable = true;
10303+
10304+ connector->display_info.monitor_range.min_vfreq = vsdb_info.min_refresh_rate_hz;
10305+ connector->display_info.monitor_range.max_vfreq = vsdb_info.max_refresh_rate_hz;
978610306 }
978710307 }
978810308
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h+50−16
@@ -68,18 +68,6 @@ struct common_irq_params { · 67 unmodified lines
6868 enum dc_irq_source irq_src;
6969 };
7070
71−/**
72− * struct irq_list_head - Linked-list for low context IRQ handlers.
73− *
74− * @head: The list_head within &struct handler_data
75− * @work: A work_struct containing the deferred handler work
76− */
77−struct irq_list_head {
78− struct list_head head;
79− /* In case this interrupt needs post-processing, 'work' will be queued*/
80− struct work_struct work;
81−};
82−
8371 /**
8472 * struct dm_compressor_info - Buffer info used by frame buffer compression
8573 * @cpu_addr: MMIO cpu addr
@@ -144,6 +132,16 @@ struct amdgpu_dm_backlight_caps { · 58 unmodified lines
144132 bool aux_support;
145133 };
146134
135+/**
136+ * struct dal_allocation - Tracks mapped FB memory for SMU communication
137+ */
138+struct dal_allocation {
139+ struct list_head list;
140+ struct amdgpu_bo *bo;
141+ void *cpu_ptr;
142+ u64 gpu_addr;
143+};
144+
147145 /**
148146 * struct amdgpu_display_manager - Central amdgpu display manager device
149147 *
@@ -257,12 +255,12 @@ struct amdgpu_display_manager { · 107 unmodified lines
257255 */
258256 struct mutex audio_lock;
259257
258+#if defined(CONFIG_DRM_AMD_DC_DCN)
260259 /**
261− * @vblank_work_lock:
260+ * @vblank_lock:
262261 *
263262 * Guards access to deferred vblank work state.
264263 */
265−#if defined(CONFIG_DRM_AMD_DC_DCN)
266264 spinlock_t vblank_lock;
267265 #endif
268266
@@ -293,7 +291,7 @@ struct amdgpu_display_manager { · 24 unmodified lines
293291 * Note that handlers are called in the same order as they were
294292 * registered (FIFO).
295293 */
296− struct irq_list_head irq_handler_list_low_tab[DAL_IRQ_SOURCES_NUMBER];
294+ struct list_head irq_handler_list_low_tab[DAL_IRQ_SOURCES_NUMBER];
297295
298296 /**
299297 * @irq_handler_list_high_tab:
@@ -323,6 +321,15 @@ struct amdgpu_display_manager { · 23 unmodified lines
323321 struct common_irq_params
324322 vblank_params[DC_IRQ_SOURCE_VBLANK6 - DC_IRQ_SOURCE_VBLANK1 + 1];
325323
324+ /**
325+ * @vline0_params:
326+ *
327+ * OTG vertical interrupt0 IRQ parameters, passed to registered
328+ * handlers when triggered.
329+ */
330+ struct common_irq_params
331+ vline0_params[DC_IRQ_SOURCE_DC6_VLINE0 - DC_IRQ_SOURCE_DC1_VLINE0 + 1];
332+
326333 /**
327334 * @vupdate_params:
328335 *
@@ -345,6 +352,11 @@ struct amdgpu_display_manager { · 16 unmodified lines
345352 #endif
346353
347354 #if defined(CONFIG_DRM_AMD_DC_DCN)
355+ /**
356+ * @vblank_workqueue:
357+ *
358+ * amdgpu workqueue during vblank
359+ */
348360 struct vblank_workqueue *vblank_workqueue;
349361 #endif
350362
@@ -363,12 +375,18 @@ struct amdgpu_display_manager { · 12 unmodified lines
363375 */
364376 const struct gpu_info_soc_bounding_box_v1_0 *soc_bounding_box;
365377
378+#if defined(CONFIG_DRM_AMD_DC_DCN)
366379 /**
367380 * @active_vblank_irq_count:
368381 *
369382 * number of currently active vblank irqs
370383 */
371384 uint32_t active_vblank_irq_count;
385+#endif
386+
387+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
388+ struct crc_rd_work *crc_rd_wrk;
389+#endif
372390
373391 /**
374392 * @mst_encoders:
@@ -377,6 +395,13 @@ struct amdgpu_display_manager { · 2 unmodified lines
377395 */
378396 struct amdgpu_encoder mst_encoders[AMDGPU_DM_MAX_CRTC];
379397 bool force_timing_sync;
398+ bool dmcub_trace_event_en;
399+ /**
400+ * @da_list:
401+ *
402+ * DAL fb memory allocation list, for communication with SMU.
403+ */
404+ struct list_head da_list;
380405 };
381406
382407 enum dsc_clock_force_state {
@@ -440,6 +465,8 @@ struct amdgpu_dm_connector { · 57 unmodified lines
440465 #endif
441466 bool force_yuv420_output;
442467 struct dsc_preferred_settings dsc_settings;
468+ /* Cached display modes */
469+ struct drm_display_mode freesync_vid_base;
443470 };
444471
445472 #define to_amdgpu_dm_connector(x) container_of(x, struct amdgpu_dm_connector, base)
@@ -462,7 +489,6 @@ struct dm_crtc_state { · 16 unmodified lines
462489 int active_planes;
463490
464491 int crc_skip_count;
465− enum amdgpu_dm_pipe_crc_source crc_src;
466492
467493 bool freesync_timing_changed;
468494 bool freesync_vrr_info_changed;
@@ -501,6 +527,14 @@ struct dm_connector_state { · 32 unmodified lines
501527 uint64_t pbn;
502528 };
503529
530+struct amdgpu_hdmi_vsdb_info {
531+ unsigned int amd_vsdb_version; /* VSDB version, should be used to determine which VSIF to send */
532+ bool freesync_supported; /* FreeSync Supported */
533+ unsigned int min_refresh_rate_hz; /* FreeSync Minimum Refresh Rate in Hz */
534+ unsigned int max_refresh_rate_hz; /* FreeSync Maximum Refresh Rate in Hz */
535+};
536+
537+
504538 #define to_dm_connector_state(x)\
505539 container_of((x), struct dm_connector_state, base)
506540
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crc.c+288−5
@@ -29,6 +29,7 @@ · 28 unmodified lines
2929 #include "amdgpu.h"
3030 #include "amdgpu_dm.h"
3131 #include "dc.h"
32+#include "amdgpu_securedisplay.h"
3233
3334 static const char *const pipe_crc_sources[] = {
3435 "none",
@@ -81,6 +82,73 @@ const char *const *amdgpu_dm_crtc_get_crc_sources(struct drm_crtc *crtc, · 46 unmodified lines
8182 return pipe_crc_sources;
8283 }
8384
85+#ifdef CONFIG_DRM_AMD_SECURE_DISPLAY
86+static void amdgpu_dm_set_crc_window_default(struct drm_crtc *crtc)
87+{
88+ struct drm_device *drm_dev = crtc->dev;
89+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
90+
91+ spin_lock_irq(&drm_dev->event_lock);
92+ acrtc->dm_irq_params.crc_window.x_start = 0;
93+ acrtc->dm_irq_params.crc_window.y_start = 0;
94+ acrtc->dm_irq_params.crc_window.x_end = 0;
95+ acrtc->dm_irq_params.crc_window.y_end = 0;
96+ acrtc->dm_irq_params.crc_window.activated = false;
97+ acrtc->dm_irq_params.crc_window.update_win = false;
98+ acrtc->dm_irq_params.crc_window.skip_frame_cnt = 0;
99+ spin_unlock_irq(&drm_dev->event_lock);
100+}
101+
102+static void amdgpu_dm_crtc_notify_ta_to_read(struct work_struct *work)
103+{
104+ struct crc_rd_work *crc_rd_wrk;
105+ struct amdgpu_device *adev;
106+ struct psp_context *psp;
107+ struct securedisplay_cmd *securedisplay_cmd;
108+ struct drm_crtc *crtc;
109+ uint8_t phy_id;
110+ int ret;
111+
112+ crc_rd_wrk = container_of(work, struct crc_rd_work, notify_ta_work);
113+ spin_lock_irq(&crc_rd_wrk->crc_rd_work_lock);
114+ crtc = crc_rd_wrk->crtc;
115+
116+ if (!crtc) {
117+ spin_unlock_irq(&crc_rd_wrk->crc_rd_work_lock);
118+ return;
119+ }
120+
121+ adev = drm_to_adev(crtc->dev);
122+ psp = &adev->psp;
123+ phy_id = crc_rd_wrk->phy_inst;
124+ spin_unlock_irq(&crc_rd_wrk->crc_rd_work_lock);
125+
126+ psp_prep_securedisplay_cmd_buf(psp, &securedisplay_cmd,
127+ TA_SECUREDISPLAY_COMMAND__SEND_ROI_CRC);
128+ securedisplay_cmd->securedisplay_in_message.send_roi_crc.phy_id =
129+ phy_id;
130+ ret = psp_securedisplay_invoke(psp, TA_SECUREDISPLAY_COMMAND__SEND_ROI_CRC);
131+ if (!ret) {
132+ if (securedisplay_cmd->status != TA_SECUREDISPLAY_STATUS__SUCCESS) {
133+ psp_securedisplay_parse_resp_status(psp, securedisplay_cmd->status);
134+ }
135+ }
136+}
137+
138+bool amdgpu_dm_crc_window_is_activated(struct drm_crtc *crtc)
139+{
140+ struct drm_device *drm_dev = crtc->dev;
141+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
142+ bool ret = false;
143+
144+ spin_lock_irq(&drm_dev->event_lock);
145+ ret = acrtc->dm_irq_params.crc_window.activated;
146+ spin_unlock_irq(&drm_dev->event_lock);
147+
148+ return ret;
149+}
150+#endif
151+
84152 int
85153 amdgpu_dm_crtc_verify_crc_source(struct drm_crtc *crtc, const char *src_name,
86154 size_t *values_cnt)
@@ -114,6 +182,20 @@ int amdgpu_dm_crtc_configure_crc_source(struct drm_crtc *crtc, · 27 unmodified lines
114182
115183 /* Enable CRTC CRC generation if necessary. */
116184 if (dm_is_crc_source_crtc(source) || source == AMDGPU_DM_PIPE_CRC_SOURCE_NONE) {
185+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
186+ if (!enable) {
187+ if (adev->dm.crc_rd_wrk) {
188+ flush_work(&adev->dm.crc_rd_wrk->notify_ta_work);
189+ spin_lock_irq(&adev->dm.crc_rd_wrk->crc_rd_work_lock);
190+ if (adev->dm.crc_rd_wrk->crtc == crtc) {
191+ dc_stream_stop_dmcu_crc_win_update(stream_state->ctx->dc,
192+ dm_crtc_state->stream);
193+ adev->dm.crc_rd_wrk->crtc = NULL;
194+ }
195+ spin_unlock_irq(&adev->dm.crc_rd_wrk->crc_rd_work_lock);
196+ }
197+ }
198+#endif
117199 if (!dc_stream_configure_crc(stream_state->ctx->dc,
118200 stream_state, NULL, enable, enable)) {
119201 ret = -EINVAL;
@@ -142,8 +224,11 @@ unlock: · 22 unmodified lines
142224 int amdgpu_dm_crtc_set_crc_source(struct drm_crtc *crtc, const char *src_name)
143225 {
144226 enum amdgpu_dm_pipe_crc_source source = dm_parse_crc_source(src_name);
227+ enum amdgpu_dm_pipe_crc_source cur_crc_src;
145228 struct drm_crtc_commit *commit;
146229 struct dm_crtc_state *crtc_state;
230+ struct drm_device *drm_dev = crtc->dev;
231+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
147232 struct drm_dp_aux *aux = NULL;
148233 bool enable = false;
149234 bool enabled = false;
@@ -182,6 +267,9 @@ int amdgpu_dm_crtc_set_crc_source(struct drm_crtc *crtc, const char *src_name) · 32 unmodified lines
182267
183268 enable = amdgpu_dm_is_valid_crc_source(source);
184269 crtc_state = to_dm_crtc_state(crtc->state);
270+ spin_lock_irq(&drm_dev->event_lock);
271+ cur_crc_src = acrtc->dm_irq_params.crc_src;
272+ spin_unlock_irq(&drm_dev->event_lock);
185273
186274 /*
187275 * USER REQ SRC | CURRENT SRC | BEHAVIOR
@@ -198,7 +286,7 @@ int amdgpu_dm_crtc_set_crc_source(struct drm_crtc *crtc, const char *src_name) · 10 unmodified lines
198286 */
199287 if (dm_is_crc_source_dprx(source) ||
200288 (source == AMDGPU_DM_PIPE_CRC_SOURCE_NONE &&
201− dm_is_crc_source_dprx(crtc_state->crc_src))) {
289+ dm_is_crc_source_dprx(cur_crc_src))) {
202290 struct amdgpu_dm_connector *aconn = NULL;
203291 struct drm_connector *connector;
204292 struct drm_connector_list_iter conn_iter;
@@ -228,6 +316,10 @@ int amdgpu_dm_crtc_set_crc_source(struct drm_crtc *crtc, const char *src_name) · 23 unmodified lines
228316 }
229317 }
230318
319+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
320+ amdgpu_dm_set_crc_window_default(crtc);
321+#endif
322+
231323 if (amdgpu_dm_crtc_configure_crc_source(crtc, crtc_state, source)) {
232324 ret = -EINVAL;
233325 goto cleanup;
@@ -237,7 +329,7 @@ int amdgpu_dm_crtc_set_crc_source(struct drm_crtc *crtc, const char *src_name) · 3 unmodified lines
237329 * Reading the CRC requires the vblank interrupt handler to be
238330 * enabled. Keep a reference until CRC capture stops.
239331 */
240− enabled = amdgpu_dm_is_valid_crc_source(crtc_state->crc_src);
332+ enabled = amdgpu_dm_is_valid_crc_source(cur_crc_src);
241333 if (!enabled && enable) {
242334 ret = drm_crtc_vblank_get(crtc);
243335 if (ret)
@@ -261,7 +353,9 @@ int amdgpu_dm_crtc_set_crc_source(struct drm_crtc *crtc, const char *src_name) · 17 unmodified lines
261353 }
262354 }
263355
264− crtc_state->crc_src = source;
356+ spin_lock_irq(&drm_dev->event_lock);
357+ acrtc->dm_irq_params.crc_src = source;
358+ spin_unlock_irq(&drm_dev->event_lock);
265359
266360 /* Reset crc_skipped on dm state */
267361 crtc_state->crc_skip_count = 0;
@@ -286,16 +380,26 @@ void amdgpu_dm_crtc_handle_crc_irq(struct drm_crtc *crtc) · 18 unmodified lines
286380 {
287381 struct dm_crtc_state *crtc_state;
288382 struct dc_stream_state *stream_state;
383+ struct drm_device *drm_dev = NULL;
384+ enum amdgpu_dm_pipe_crc_source cur_crc_src;
385+ struct amdgpu_crtc *acrtc = NULL;
289386 uint32_t crcs[3];
387+ unsigned long flags;
290388
291389 if (crtc == NULL)
292390 return;
293391
294392 crtc_state = to_dm_crtc_state(crtc->state);
295393 stream_state = crtc_state->stream;
394+ acrtc = to_amdgpu_crtc(crtc);
395+ drm_dev = crtc->dev;
396+
397+ spin_lock_irqsave(&drm_dev->event_lock, flags);
398+ cur_crc_src = acrtc->dm_irq_params.crc_src;
399+ spin_unlock_irqrestore(&drm_dev->event_lock, flags);
296400
297401 /* Early return if CRC capture is not enabled. */
298− if (!amdgpu_dm_is_valid_crc_source(crtc_state->crc_src))
402+ if (!amdgpu_dm_is_valid_crc_source(cur_crc_src))
299403 return;
300404
301405 /*
@@ -309,7 +413,7 @@ void amdgpu_dm_crtc_handle_crc_irq(struct drm_crtc *crtc) · 7 unmodified lines
309413 return;
310414 }
311415
312− if (dm_is_crc_source_crtc(crtc_state->crc_src)) {
416+ if (dm_is_crc_source_crtc(cur_crc_src)) {
313417 if (!dc_stream_get_crc(stream_state->ctx->dc, stream_state,
314418 &crcs[0], &crcs[1], &crcs[2]))
315419 return;
@@ -318,3 +422,182 @@ void amdgpu_dm_crtc_handle_crc_irq(struct drm_crtc *crtc) · 2 unmodified lines
318422 drm_crtc_accurate_vblank_count(crtc), crcs);
319423 }
320424 }
425+
426+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
427+void amdgpu_dm_crtc_handle_crc_window_irq(struct drm_crtc *crtc)
428+{
429+ struct dc_stream_state *stream_state;
430+ struct drm_device *drm_dev = NULL;
431+ enum amdgpu_dm_pipe_crc_source cur_crc_src;
432+ struct amdgpu_crtc *acrtc = NULL;
433+ struct amdgpu_device *adev = NULL;
434+ struct crc_rd_work *crc_rd_wrk = NULL;
435+ struct crc_params *crc_window = NULL, tmp_window;
436+ unsigned long flags1, flags2;
437+ struct crtc_position position;
438+ uint32_t v_blank;
439+ uint32_t v_back_porch;
440+ uint32_t crc_window_latch_up_line;
441+ struct dc_crtc_timing *timing_out;
442+
443+ if (crtc == NULL)
444+ return;
445+
446+ acrtc = to_amdgpu_crtc(crtc);
447+ adev = drm_to_adev(crtc->dev);
448+ drm_dev = crtc->dev;
449+
450+ spin_lock_irqsave(&drm_dev->event_lock, flags1);
451+ stream_state = acrtc->dm_irq_params.stream;
452+ cur_crc_src = acrtc->dm_irq_params.crc_src;
453+ timing_out = &stream_state->timing;
454+
455+ /* Early return if CRC capture is not enabled. */
456+ if (!amdgpu_dm_is_valid_crc_source(cur_crc_src))
457+ goto cleanup;
458+
459+ if (dm_is_crc_source_crtc(cur_crc_src)) {
460+ if (acrtc->dm_irq_params.crc_window.activated) {
461+ if (acrtc->dm_irq_params.crc_window.update_win) {
462+ if (acrtc->dm_irq_params.crc_window.skip_frame_cnt) {
463+ acrtc->dm_irq_params.crc_window.skip_frame_cnt -= 1;
464+ goto cleanup;
465+ }
466+ crc_window = &tmp_window;
467+
468+ tmp_window.windowa_x_start =
469+ acrtc->dm_irq_params.crc_window.x_start;
470+ tmp_window.windowa_y_start =
471+ acrtc->dm_irq_params.crc_window.y_start;
472+ tmp_window.windowa_x_end =
473+ acrtc->dm_irq_params.crc_window.x_end;
474+ tmp_window.windowa_y_end =
475+ acrtc->dm_irq_params.crc_window.y_end;
476+ tmp_window.windowb_x_start =
477+ acrtc->dm_irq_params.crc_window.x_start;
478+ tmp_window.windowb_y_start =
479+ acrtc->dm_irq_params.crc_window.y_start;
480+ tmp_window.windowb_x_end =
481+ acrtc->dm_irq_params.crc_window.x_end;
482+ tmp_window.windowb_y_end =
483+ acrtc->dm_irq_params.crc_window.y_end;
484+
485+ dc_stream_forward_dmcu_crc_window(stream_state->ctx->dc,
486+ stream_state, crc_window);
487+
488+ acrtc->dm_irq_params.crc_window.update_win = false;
489+
490+ dc_stream_get_crtc_position(stream_state->ctx->dc, &stream_state, 1,
491+ &position.vertical_count,
492+ &position.nominal_vcount);
493+
494+ v_blank = timing_out->v_total - timing_out->v_border_top -
495+ timing_out->v_addressable - timing_out->v_border_bottom;
496+
497+ v_back_porch = v_blank - timing_out->v_front_porch -
498+ timing_out->v_sync_width;
499+
500+ crc_window_latch_up_line = v_back_porch + timing_out->v_sync_width;
501+
502+ /* take 3 lines margin*/
503+ if ((position.vertical_count + 3) >= crc_window_latch_up_line)
504+ acrtc->dm_irq_params.crc_window.skip_frame_cnt = 1;
505+ else
506+ acrtc->dm_irq_params.crc_window.skip_frame_cnt = 0;
507+ } else {
508+ if (acrtc->dm_irq_params.crc_window.skip_frame_cnt == 0) {
509+ if (adev->dm.crc_rd_wrk) {
510+ crc_rd_wrk = adev->dm.crc_rd_wrk;
511+ spin_lock_irqsave(&crc_rd_wrk->crc_rd_work_lock, flags2);
512+ crc_rd_wrk->phy_inst =
513+ stream_state->link->link_enc_hw_inst;
514+ spin_unlock_irqrestore(&crc_rd_wrk->crc_rd_work_lock, flags2);
515+ schedule_work(&crc_rd_wrk->notify_ta_work);
516+ }
517+ } else {
518+ acrtc->dm_irq_params.crc_window.skip_frame_cnt -= 1;
519+ }
520+ }
521+ }
522+ }
523+
524+cleanup:
525+ spin_unlock_irqrestore(&drm_dev->event_lock, flags1);
526+}
527+
528+void amdgpu_dm_crtc_secure_display_resume(struct amdgpu_device *adev)
529+{
530+ struct drm_crtc *crtc;
531+ enum amdgpu_dm_pipe_crc_source cur_crc_src;
532+ struct crc_rd_work *crc_rd_wrk = adev->dm.crc_rd_wrk;
533+ struct crc_window_parm cur_crc_window;
534+ struct amdgpu_crtc *acrtc = NULL;
535+
536+ drm_for_each_crtc(crtc, &adev->ddev) {
537+ acrtc = to_amdgpu_crtc(crtc);
538+
539+ spin_lock_irq(&adev_to_drm(adev)->event_lock);
540+ cur_crc_src = acrtc->dm_irq_params.crc_src;
541+ cur_crc_window = acrtc->dm_irq_params.crc_window;
542+ spin_unlock_irq(&adev_to_drm(adev)->event_lock);
543+
544+ if (amdgpu_dm_is_valid_crc_source(cur_crc_src)) {
545+ amdgpu_dm_crtc_set_crc_source(crtc,
546+ pipe_crc_sources[cur_crc_src]);
547+ spin_lock_irq(&adev_to_drm(adev)->event_lock);
548+ acrtc->dm_irq_params.crc_window = cur_crc_window;
549+ if (acrtc->dm_irq_params.crc_window.activated) {
550+ acrtc->dm_irq_params.crc_window.update_win = true;
551+ acrtc->dm_irq_params.crc_window.skip_frame_cnt = 1;
552+ spin_lock_irq(&crc_rd_wrk->crc_rd_work_lock);
553+ crc_rd_wrk->crtc = crtc;
554+ spin_unlock_irq(&crc_rd_wrk->crc_rd_work_lock);
555+ }
556+ spin_unlock_irq(&adev_to_drm(adev)->event_lock);
557+ }
558+ }
559+}
560+
561+void amdgpu_dm_crtc_secure_display_suspend(struct amdgpu_device *adev)
562+{
563+ struct drm_crtc *crtc;
564+ struct crc_window_parm cur_crc_window;
565+ enum amdgpu_dm_pipe_crc_source cur_crc_src;
566+ struct amdgpu_crtc *acrtc = NULL;
567+
568+ drm_for_each_crtc(crtc, &adev->ddev) {
569+ acrtc = to_amdgpu_crtc(crtc);
570+
571+ spin_lock_irq(&adev_to_drm(adev)->event_lock);
572+ cur_crc_src = acrtc->dm_irq_params.crc_src;
573+ cur_crc_window = acrtc->dm_irq_params.crc_window;
574+ cur_crc_window.update_win = false;
575+ spin_unlock_irq(&adev_to_drm(adev)->event_lock);
576+
577+ if (amdgpu_dm_is_valid_crc_source(cur_crc_src)) {
578+ amdgpu_dm_crtc_set_crc_source(crtc, NULL);
579+ spin_lock_irq(&adev_to_drm(adev)->event_lock);
580+ /* For resume to set back crc source*/
581+ acrtc->dm_irq_params.crc_src = cur_crc_src;
582+ acrtc->dm_irq_params.crc_window = cur_crc_window;
583+ spin_unlock_irq(&adev_to_drm(adev)->event_lock);
584+ }
585+ }
586+
587+}
588+
589+struct crc_rd_work *amdgpu_dm_crtc_secure_display_create_work(void)
590+{
591+ struct crc_rd_work *crc_rd_wrk = NULL;
592+
593+ crc_rd_wrk = kzalloc(sizeof(*crc_rd_wrk), GFP_KERNEL);
594+
595+ if (!crc_rd_wrk)
596+ return NULL;
597+
598+ spin_lock_init(&crc_rd_wrk->crc_rd_work_lock);
599+ INIT_WORK(&crc_rd_wrk->notify_ta_work, amdgpu_dm_crtc_notify_ta_to_read);
600+
601+ return crc_rd_wrk;
602+}
603+#endif
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crc.h+37−0
@@ -39,6 +39,29 @@ enum amdgpu_dm_pipe_crc_source { · 38 unmodified lines
3939 AMDGPU_DM_PIPE_CRC_SOURCE_INVALID = -1,
4040 };
4141
42+#ifdef CONFIG_DRM_AMD_SECURE_DISPLAY
43+struct crc_window_parm {
44+ uint16_t x_start;
45+ uint16_t y_start;
46+ uint16_t x_end;
47+ uint16_t y_end;
48+ /* CRC windwo is activated or not*/
49+ bool activated;
50+ /* Update crc window during vertical blank or not */
51+ bool update_win;
52+ /* skip reading/writing for few frames */
53+ int skip_frame_cnt;
54+};
55+
56+struct crc_rd_work {
57+ struct work_struct notify_ta_work;
58+ /* To protect crc_rd_work carried fields*/
59+ spinlock_t crc_rd_work_lock;
60+ struct drm_crtc *crtc;
61+ uint8_t phy_inst;
62+};
63+#endif
64+
4265 static inline bool amdgpu_dm_is_valid_crc_source(enum amdgpu_dm_pipe_crc_source source)
4366 {
4467 return (source > AMDGPU_DM_PIPE_CRC_SOURCE_NONE) &&
@@ -64,4 +87,18 @@ void amdgpu_dm_crtc_handle_crc_irq(struct drm_crtc *crtc); · 19 unmodified lines
6487 #define amdgpu_dm_crtc_handle_crc_irq(x)
6588 #endif
6689
90+#ifdef CONFIG_DRM_AMD_SECURE_DISPLAY
91+bool amdgpu_dm_crc_window_is_activated(struct drm_crtc *crtc);
92+void amdgpu_dm_crtc_handle_crc_window_irq(struct drm_crtc *crtc);
93+struct crc_rd_work *amdgpu_dm_crtc_secure_display_create_work(void);
94+void amdgpu_dm_crtc_secure_display_resume(struct amdgpu_device *adev);
95+void amdgpu_dm_crtc_secure_display_suspend(struct amdgpu_device *adev);
96+#else
97+#define amdgpu_dm_crc_window_is_activated(x)
98+#define amdgpu_dm_crtc_handle_crc_window_irq(x)
99+#define amdgpu_dm_crtc_secure_display_create_work()
100+#define amdgpu_dm_crtc_secure_display_resume(x)
101+#define amdgpu_dm_crtc_secure_display_suspend(x)
102+#endif
103+
67104 #endif /* AMD_DAL_DEV_AMDGPU_DM_AMDGPU_DM_CRC_H_ */
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c+438−36
@@ -25,8 +25,6 @@ · 24 unmodified lines
2525
2626 #include <linux/uaccess.h>
2727
28−#include <drm/drm_debugfs.h>
29−
3028 #include "dc.h"
3129 #include "amdgpu.h"
3230 #include "amdgpu_dm.h"
@@ -36,6 +34,7 @@ · 3 unmodified lines
3634 #include "resource.h"
3735 #include "dsc.h"
3836 #include "dc_link_dp.h"
37+#include "dc/dc_dmub_srv.h"
3938
4039 struct dmub_debugfs_trace_header {
4140 uint32_t entry_count;
@@ -2154,6 +2153,149 @@ static ssize_t dp_dsc_slice_bpg_offset_read(struct file *f, char __user *buf, · 2112 unmodified lines
21542153 return result;
21552154 }
21562155
2156+
2157+/*
2158+ * function description: Read max_requested_bpc property from the connector
2159+ *
2160+ * Access it with the following command:
2161+ *
2162+ * cat /sys/kernel/debug/dri/0/DP-X/max_bpc
2163+ *
2164+ */
2165+static ssize_t dp_max_bpc_read(struct file *f, char __user *buf,
2166+ size_t size, loff_t *pos)
2167+{
2168+ struct amdgpu_dm_connector *aconnector = file_inode(f)->i_private;
2169+ struct drm_connector *connector = &aconnector->base;
2170+ struct drm_device *dev = connector->dev;
2171+ struct dm_connector_state *state;
2172+ ssize_t result = 0;
2173+ char *rd_buf = NULL;
2174+ char *rd_buf_ptr = NULL;
2175+ const uint32_t rd_buf_size = 10;
2176+ int r;
2177+
2178+ rd_buf = kcalloc(rd_buf_size, sizeof(char), GFP_KERNEL);
2179+
2180+ if (!rd_buf)
2181+ return -ENOMEM;
2182+
2183+ mutex_lock(&dev->mode_config.mutex);
2184+ drm_modeset_lock(&dev->mode_config.connection_mutex, NULL);
2185+
2186+ if (connector->state == NULL)
2187+ goto unlock;
2188+
2189+ state = to_dm_connector_state(connector->state);
2190+
2191+ rd_buf_ptr = rd_buf;
2192+ snprintf(rd_buf_ptr, rd_buf_size,
2193+ "%u\n",
2194+ state->base.max_requested_bpc);
2195+
2196+ while (size) {
2197+ if (*pos >= rd_buf_size)
2198+ break;
2199+
2200+ r = put_user(*(rd_buf + result), buf);
2201+ if (r) {
2202+ result = r; /* r = -EFAULT */
2203+ goto unlock;
2204+ }
2205+ buf += 1;
2206+ size -= 1;
2207+ *pos += 1;
2208+ result += 1;
2209+ }
2210+unlock:
2211+ drm_modeset_unlock(&dev->mode_config.connection_mutex);
2212+ mutex_unlock(&dev->mode_config.mutex);
2213+ kfree(rd_buf);
2214+ return result;
2215+}
2216+
2217+
2218+/*
2219+ * function description: Set max_requested_bpc property on the connector
2220+ *
2221+ * This function will not force the input BPC on connector, it will only
2222+ * change the max value. This is equivalent to setting max_bpc through
2223+ * xrandr.
2224+ *
2225+ * The BPC value written must be >= 6 and <= 16. Values outside of this
2226+ * range will result in errors.
2227+ *
2228+ * BPC values:
2229+ * 0x6 - 6 BPC
2230+ * 0x8 - 8 BPC
2231+ * 0xa - 10 BPC
2232+ * 0xc - 12 BPC
2233+ * 0x10 - 16 BPC
2234+ *
2235+ * Write the max_bpc in the following way:
2236+ *
2237+ * echo 0x6 > /sys/kernel/debug/dri/0/DP-X/max_bpc
2238+ *
2239+ */
2240+static ssize_t dp_max_bpc_write(struct file *f, const char __user *buf,
2241+ size_t size, loff_t *pos)
2242+{
2243+ struct amdgpu_dm_connector *aconnector = file_inode(f)->i_private;
2244+ struct drm_connector *connector = &aconnector->base;
2245+ struct dm_connector_state *state;
2246+ struct drm_device *dev = connector->dev;
2247+ char *wr_buf = NULL;
2248+ uint32_t wr_buf_size = 42;
2249+ int max_param_num = 1;
2250+ long param[1] = {0};
2251+ uint8_t param_nums = 0;
2252+
2253+ if (size == 0)
2254+ return -EINVAL;
2255+
2256+ wr_buf = kcalloc(wr_buf_size, sizeof(char), GFP_KERNEL);
2257+
2258+ if (!wr_buf) {
2259+ DRM_DEBUG_DRIVER("no memory to allocate write buffer\n");
2260+ return -ENOSPC;
2261+ }
2262+
2263+ if (parse_write_buffer_into_params(wr_buf, size,
2264+ (long *)param, buf,
2265+ max_param_num,
2266+ &param_nums)) {
2267+ kfree(wr_buf);
2268+ return -EINVAL;
2269+ }
2270+
2271+ if (param_nums <= 0) {
2272+ DRM_DEBUG_DRIVER("user data not be read\n");
2273+ kfree(wr_buf);
2274+ return -EINVAL;
2275+ }
2276+
2277+ if (param[0] < 6 || param[0] > 16) {
2278+ DRM_DEBUG_DRIVER("bad max_bpc value\n");
2279+ kfree(wr_buf);
2280+ return -EINVAL;
2281+ }
2282+
2283+ mutex_lock(&dev->mode_config.mutex);
2284+ drm_modeset_lock(&dev->mode_config.connection_mutex, NULL);
2285+
2286+ if (connector->state == NULL)
2287+ goto unlock;
2288+
2289+ state = to_dm_connector_state(connector->state);
2290+ state->base.max_requested_bpc = param[0];
2291+unlock:
2292+ drm_modeset_unlock(&dev->mode_config.connection_mutex);
2293+ mutex_unlock(&dev->mode_config.mutex);
2294+
2295+ kfree(wr_buf);
2296+ return size;
2297+}
2298+
21572299 DEFINE_SHOW_ATTRIBUTE(dp_dsc_fec_support);
21582300 DEFINE_SHOW_ATTRIBUTE(dmub_fw_state);
21592301 DEFINE_SHOW_ATTRIBUTE(dmub_tracebuffer);
@@ -2265,6 +2407,13 @@ static const struct file_operations dp_dpcd_data_debugfs_fops = { · 105 unmodified lines
22652407 .llseek = default_llseek
22662408 };
22672409
2410+static const struct file_operations dp_max_bpc_debugfs_fops = {
2411+ .owner = THIS_MODULE,
2412+ .read = dp_max_bpc_read,
2413+ .write = dp_max_bpc_write,
2414+ .llseek = default_llseek
2415+};
2416+
22682417 static const struct {
22692418 char *name;
22702419 const struct file_operations *fops;
@@ -2287,7 +2436,8 @@ static const struct { · 16 unmodified lines
22872436 {"dsc_pic_height", &dp_dsc_pic_height_debugfs_fops},
22882437 {"dsc_chunk_size", &dp_dsc_chunk_size_debugfs_fops},
22892438 {"dsc_slice_bpg", &dp_dsc_slice_bpg_offset_debugfs_fops},
2290− {"dp_dsc_fec_support", &dp_dsc_fec_support_fops}
2439+ {"dp_dsc_fec_support", &dp_dsc_fec_support_fops},
2440+ {"max_bpc", &dp_max_bpc_debugfs_fops}
22912441 };
22922442
22932443 #ifdef CONFIG_DRM_AMD_DC_HDCP
@@ -2341,9 +2491,51 @@ static int psr_get(void *data, u64 *val) · 47 unmodified lines
23412491 return 0;
23422492 }
23432493
2494+/*
2495+ * Set dmcub trace event IRQ enable or disable.
2496+ * Usage to enable dmcub trace event IRQ: echo 1 > /sys/kernel/debug/dri/0/amdgpu_dm_dmcub_trace_event_en
2497+ * Usage to disable dmcub trace event IRQ: echo 0 > /sys/kernel/debug/dri/0/amdgpu_dm_dmcub_trace_event_en
2498+ */
2499+static int dmcub_trace_event_state_set(void *data, u64 val)
2500+{
2501+ struct amdgpu_device *adev = data;
2502+
2503+ if (val == 1 || val == 0) {
2504+ dc_dmub_trace_event_control(adev->dm.dc, val);
2505+ adev->dm.dmcub_trace_event_en = (bool)val;
2506+ } else
2507+ return 0;
2508+
2509+ return 0;
2510+}
2511+
2512+/*
2513+ * The interface doesn't need get function, so it will return the
2514+ * value of zero
2515+ * Usage: cat /sys/kernel/debug/dri/0/amdgpu_dm_dmcub_trace_event_en
2516+ */
2517+static int dmcub_trace_event_state_get(void *data, u64 *val)
2518+{
2519+ struct amdgpu_device *adev = data;
2520+
2521+ *val = adev->dm.dmcub_trace_event_en;
2522+ return 0;
2523+}
2524+
2525+DEFINE_DEBUGFS_ATTRIBUTE(dmcub_trace_event_state_fops, dmcub_trace_event_state_get,
2526+ dmcub_trace_event_state_set, "%llu\n");
23442527
23452528 DEFINE_DEBUGFS_ATTRIBUTE(psr_fops, psr_get, NULL, "%llu\n");
23462529
2530+static const struct {
2531+ char *name;
2532+ const struct file_operations *fops;
2533+} connector_debugfs_entries[] = {
2534+ {"force_yuv420_output", &force_yuv420_output_fops},
2535+ {"output_bpc", &output_bpc_fops},
2536+ {"trigger_hotplug", &trigger_hotplug_debugfs_fops}
2537+};
2538+
23472539 void connector_debugfs_init(struct amdgpu_dm_connector *connector)
23482540 {
23492541 int i;
@@ -2360,14 +2552,11 @@ void connector_debugfs_init(struct amdgpu_dm_connector *connector) · 10 unmodified lines
23602552 if (connector->base.connector_type == DRM_MODE_CONNECTOR_eDP)
23612553 debugfs_create_file_unsafe("psr_state", 0444, dir, connector, &psr_fops);
23622554
2363− debugfs_create_file_unsafe("force_yuv420_output", 0644, dir, connector,
2364− &force_yuv420_output_fops);
2365−
2366− debugfs_create_file("output_bpc", 0644, dir, connector,
2367− &output_bpc_fops);
2368−
2369− debugfs_create_file("trigger_hotplug", 0644, dir, connector,
2370− &trigger_hotplug_debugfs_fops);
2555+ for (i = 0; i < ARRAY_SIZE(connector_debugfs_entries); i++) {
2556+ debugfs_create_file(connector_debugfs_entries[i].name,
2557+ 0644, dir, connector,
2558+ connector_debugfs_entries[i].fops);
2559+ }
23712560
23722561 connector->debugfs_dpcd_address = 0;
23732562 connector->debugfs_dpcd_size = 0;
@@ -2383,6 +2572,225 @@ void connector_debugfs_init(struct amdgpu_dm_connector *connector) · 9 unmodified lines
23832572 #endif
23842573 }
23852574
2575+#ifdef CONFIG_DRM_AMD_SECURE_DISPLAY
2576+/*
2577+ * Set crc window coordinate x start
2578+ */
2579+static int crc_win_x_start_set(void *data, u64 val)
2580+{
2581+ struct drm_crtc *crtc = data;
2582+ struct drm_device *drm_dev = crtc->dev;
2583+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
2584+
2585+ spin_lock_irq(&drm_dev->event_lock);
2586+ acrtc->dm_irq_params.crc_window.x_start = (uint16_t) val;
2587+ acrtc->dm_irq_params.crc_window.update_win = false;
2588+ spin_unlock_irq(&drm_dev->event_lock);
2589+
2590+ return 0;
2591+}
2592+
2593+/*
2594+ * Get crc window coordinate x start
2595+ */
2596+static int crc_win_x_start_get(void *data, u64 *val)
2597+{
2598+ struct drm_crtc *crtc = data;
2599+ struct drm_device *drm_dev = crtc->dev;
2600+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
2601+
2602+ spin_lock_irq(&drm_dev->event_lock);
2603+ *val = acrtc->dm_irq_params.crc_window.x_start;
2604+ spin_unlock_irq(&drm_dev->event_lock);
2605+
2606+ return 0;
2607+}
2608+
2609+DEFINE_DEBUGFS_ATTRIBUTE(crc_win_x_start_fops, crc_win_x_start_get,
2610+ crc_win_x_start_set, "%llu\n");
2611+
2612+
2613+/*
2614+ * Set crc window coordinate y start
2615+ */
2616+static int crc_win_y_start_set(void *data, u64 val)
2617+{
2618+ struct drm_crtc *crtc = data;
2619+ struct drm_device *drm_dev = crtc->dev;
2620+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
2621+
2622+ spin_lock_irq(&drm_dev->event_lock);
2623+ acrtc->dm_irq_params.crc_window.y_start = (uint16_t) val;
2624+ acrtc->dm_irq_params.crc_window.update_win = false;
2625+ spin_unlock_irq(&drm_dev->event_lock);
2626+
2627+ return 0;
2628+}
2629+
2630+/*
2631+ * Get crc window coordinate y start
2632+ */
2633+static int crc_win_y_start_get(void *data, u64 *val)
2634+{
2635+ struct drm_crtc *crtc = data;
2636+ struct drm_device *drm_dev = crtc->dev;
2637+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
2638+
2639+ spin_lock_irq(&drm_dev->event_lock);
2640+ *val = acrtc->dm_irq_params.crc_window.y_start;
2641+ spin_unlock_irq(&drm_dev->event_lock);
2642+
2643+ return 0;
2644+}
2645+
2646+DEFINE_DEBUGFS_ATTRIBUTE(crc_win_y_start_fops, crc_win_y_start_get,
2647+ crc_win_y_start_set, "%llu\n");
2648+
2649+/*
2650+ * Set crc window coordinate x end
2651+ */
2652+static int crc_win_x_end_set(void *data, u64 val)
2653+{
2654+ struct drm_crtc *crtc = data;
2655+ struct drm_device *drm_dev = crtc->dev;
2656+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
2657+
2658+ spin_lock_irq(&drm_dev->event_lock);
2659+ acrtc->dm_irq_params.crc_window.x_end = (uint16_t) val;
2660+ acrtc->dm_irq_params.crc_window.update_win = false;
2661+ spin_unlock_irq(&drm_dev->event_lock);
2662+
2663+ return 0;
2664+}
2665+
2666+/*
2667+ * Get crc window coordinate x end
2668+ */
2669+static int crc_win_x_end_get(void *data, u64 *val)
2670+{
2671+ struct drm_crtc *crtc = data;
2672+ struct drm_device *drm_dev = crtc->dev;
2673+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
2674+
2675+ spin_lock_irq(&drm_dev->event_lock);
2676+ *val = acrtc->dm_irq_params.crc_window.x_end;
2677+ spin_unlock_irq(&drm_dev->event_lock);
2678+
2679+ return 0;
2680+}
2681+
2682+DEFINE_DEBUGFS_ATTRIBUTE(crc_win_x_end_fops, crc_win_x_end_get,
2683+ crc_win_x_end_set, "%llu\n");
2684+
2685+/*
2686+ * Set crc window coordinate y end
2687+ */
2688+static int crc_win_y_end_set(void *data, u64 val)
2689+{
2690+ struct drm_crtc *crtc = data;
2691+ struct drm_device *drm_dev = crtc->dev;
2692+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
2693+
2694+ spin_lock_irq(&drm_dev->event_lock);
2695+ acrtc->dm_irq_params.crc_window.y_end = (uint16_t) val;
2696+ acrtc->dm_irq_params.crc_window.update_win = false;
2697+ spin_unlock_irq(&drm_dev->event_lock);
2698+
2699+ return 0;
2700+}
2701+
2702+/*
2703+ * Get crc window coordinate y end
2704+ */
2705+static int crc_win_y_end_get(void *data, u64 *val)
2706+{
2707+ struct drm_crtc *crtc = data;
2708+ struct drm_device *drm_dev = crtc->dev;
2709+ struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
2710+
2711+ spin_lock_irq(&drm_dev->event_lock);
2712+ *val = acrtc->dm_irq_params.crc_window.y_end;
2713+ spin_unlock_irq(&drm_dev->event_lock);
2714+
2715+ return 0;
2716+}
2717+
2718+DEFINE_DEBUGFS_ATTRIBUTE(crc_win_y_end_fops, crc_win_y_end_get,
2719+ crc_win_y_end_set, "%llu\n");
2720+/*
2721+ * Trigger to commit crc window
2722+ */
2723+static int crc_win_update_set(void *data, u64 val)
2724+{
2725+ struct drm_crtc *new_crtc = data;
2726+ struct drm_crtc *old_crtc = NULL;
2727+ struct amdgpu_crtc *new_acrtc, *old_acrtc;
2728+ struct amdgpu_device *adev = drm_to_adev(new_crtc->dev);
2729+ struct crc_rd_work *crc_rd_wrk = adev->dm.crc_rd_wrk;
2730+
2731+ if (val) {
2732+ spin_lock_irq(&adev_to_drm(adev)->event_lock);
2733+ spin_lock_irq(&crc_rd_wrk->crc_rd_work_lock);
2734+ if (crc_rd_wrk && crc_rd_wrk->crtc) {
2735+ old_crtc = crc_rd_wrk->crtc;
2736+ old_acrtc = to_amdgpu_crtc(old_crtc);
2737+ }
2738+ new_acrtc = to_amdgpu_crtc(new_crtc);
2739+
2740+ if (old_crtc && old_crtc != new_crtc) {
2741+ old_acrtc->dm_irq_params.crc_window.activated = false;
2742+ old_acrtc->dm_irq_params.crc_window.update_win = false;
2743+ old_acrtc->dm_irq_params.crc_window.skip_frame_cnt = 0;
2744+
2745+ new_acrtc->dm_irq_params.crc_window.activated = true;
2746+ new_acrtc->dm_irq_params.crc_window.update_win = true;
2747+ new_acrtc->dm_irq_params.crc_window.skip_frame_cnt = 0;
2748+ crc_rd_wrk->crtc = new_crtc;
2749+ } else {
2750+ new_acrtc->dm_irq_params.crc_window.activated = true;
2751+ new_acrtc->dm_irq_params.crc_window.update_win = true;
2752+ new_acrtc->dm_irq_params.crc_window.skip_frame_cnt = 0;
2753+ crc_rd_wrk->crtc = new_crtc;
2754+ }
2755+ spin_unlock_irq(&crc_rd_wrk->crc_rd_work_lock);
2756+ spin_unlock_irq(&adev_to_drm(adev)->event_lock);
2757+ }
2758+
2759+ return 0;
2760+}
2761+
2762+/*
2763+ * Get crc window update flag
2764+ */
2765+static int crc_win_update_get(void *data, u64 *val)
2766+{
2767+ *val = 0;
2768+ return 0;
2769+}
2770+
2771+DEFINE_DEBUGFS_ATTRIBUTE(crc_win_update_fops, crc_win_update_get,
2772+ crc_win_update_set, "%llu\n");
2773+
2774+void crtc_debugfs_init(struct drm_crtc *crtc)
2775+{
2776+ struct dentry *dir = debugfs_lookup("crc", crtc->debugfs_entry);
2777+
2778+ if (!dir)
2779+ return;
2780+
2781+ debugfs_create_file_unsafe("crc_win_x_start", 0644, dir, crtc,
2782+ &crc_win_x_start_fops);
2783+ debugfs_create_file_unsafe("crc_win_y_start", 0644, dir, crtc,
2784+ &crc_win_y_start_fops);
2785+ debugfs_create_file_unsafe("crc_win_x_end", 0644, dir, crtc,
2786+ &crc_win_x_end_fops);
2787+ debugfs_create_file_unsafe("crc_win_y_end", 0644, dir, crtc,
2788+ &crc_win_y_end_fops);
2789+ debugfs_create_file_unsafe("crc_win_update", 0644, dir, crtc,
2790+ &crc_win_update_fops);
2791+
2792+}
2793+#endif
23862794 /*
23872795 * Writes DTN log state to the user supplied buffer.
23882796 * Example usage: cat /sys/kernel/debug/dri/0/amdgpu_dm_dtn_log
@@ -2450,11 +2858,9 @@ static ssize_t dtn_log_write( · 61 unmodified lines
24502858 * As written to display, taking ABM and backlight lut into account.
24512859 * Ranges from 0x0 to 0x10000 (= 100% PWM)
24522860 */
2453−static int current_backlight_read(struct seq_file *m, void *data)
2861+static int current_backlight_show(struct seq_file *m, void *unused)
24542862 {
2455− struct drm_info_node *node = (struct drm_info_node *)m->private;
2456− struct drm_device *dev = node->minor->dev;
2457− struct amdgpu_device *adev = drm_to_adev(dev);
2863+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
24582864 struct amdgpu_display_manager *dm = &adev->dm;
24592865
24602866 unsigned int backlight = dc_link_get_backlight_level(dm->backlight_link);
@@ -2468,11 +2874,9 @@ static int current_backlight_read(struct seq_file *m, void *data) · 7 unmodified lines
24682874 * As written to display, taking ABM and backlight lut into account.
24692875 * Ranges from 0x0 to 0x10000 (= 100% PWM)
24702876 */
2471−static int target_backlight_read(struct seq_file *m, void *data)
2877+static int target_backlight_show(struct seq_file *m, void *unused)
24722878 {
2473− struct drm_info_node *node = (struct drm_info_node *)m->private;
2474− struct drm_device *dev = node->minor->dev;
2475− struct amdgpu_device *adev = drm_to_adev(dev);
2879+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
24762880 struct amdgpu_display_manager *dm = &adev->dm;
24772881
24782882 unsigned int backlight = dc_link_get_target_backlight_pwm(dm->backlight_link);
@@ -2481,10 +2885,10 @@ static int target_backlight_read(struct seq_file *m, void *data) · 2 unmodified lines
24812885 return 0;
24822886 }
24832887
2484−static int mst_topo(struct seq_file *m, void *unused)
2888+static int mst_topo_show(struct seq_file *m, void *unused)
24852889 {
2486− struct drm_info_node *node = (struct drm_info_node *)m->private;
2487− struct drm_device *dev = node->minor->dev;
2890+ struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
2891+ struct drm_device *dev = adev_to_drm(adev);
24882892 struct drm_connector *connector;
24892893 struct drm_connector_list_iter conn_iter;
24902894 struct amdgpu_dm_connector *aconnector;
@@ -2504,12 +2908,6 @@ static int mst_topo(struct seq_file *m, void *unused) · 13 unmodified lines
25042908 return 0;
25052909 }
25062910
2507−static const struct drm_info_list amdgpu_dm_debugfs_list[] = {
2508− {"amdgpu_current_backlight_pwm", &current_backlight_read},
2509− {"amdgpu_target_backlight_pwm", &target_backlight_read},
2510− {"amdgpu_mst_topology", &mst_topo},
2511−};
2512−
25132911 /*
25142912 * Sets the force_timing_sync debug optino from the given string.
25152913 * All connected displays will be force synchronized immediately.
@@ -2568,10 +2966,13 @@ static int visual_confirm_get(void *data, u64 *val) · 52 unmodified lines
25682966 return 0;
25692967 }
25702968
2969+DEFINE_SHOW_ATTRIBUTE(current_backlight);
2970+DEFINE_SHOW_ATTRIBUTE(target_backlight);
2971+DEFINE_SHOW_ATTRIBUTE(mst_topo);
25712972 DEFINE_DEBUGFS_ATTRIBUTE(visual_confirm_fops, visual_confirm_get,
25722973 visual_confirm_set, "%llu\n");
25732974
2574−int dtn_debugfs_init(struct amdgpu_device *adev)
2975+void dtn_debugfs_init(struct amdgpu_device *adev)
25752976 {
25762977 static const struct file_operations dtn_log_fops = {
25772978 .owner = THIS_MODULE,
@@ -2582,13 +2983,13 @@ int dtn_debugfs_init(struct amdgpu_device *adev) · 4 unmodified lines
25822983
25832984 struct drm_minor *minor = adev_to_drm(adev)->primary;
25842985 struct dentry *root = minor->debugfs_root;
2585− int ret;
2586−
2587− ret = amdgpu_debugfs_add_files(adev, amdgpu_dm_debugfs_list,
2588− ARRAY_SIZE(amdgpu_dm_debugfs_list));
2589− if (ret)
2590− return ret;
25912986
2987+ debugfs_create_file("amdgpu_current_backlight_pwm", 0444,
2988+ root, adev, &current_backlight_fops);
2989+ debugfs_create_file("amdgpu_target_backlight_pwm", 0444,
2990+ root, adev, &target_backlight_fops);
2991+ debugfs_create_file("amdgpu_mst_topology", 0444, root,
2992+ adev, &mst_topo_fops);
25922993 debugfs_create_file("amdgpu_dm_dtn_log", 0644, root, adev,
25932994 &dtn_log_fops);
25942995
@@ -2604,5 +3005,6 @@ int dtn_debugfs_init(struct amdgpu_device *adev) · 9 unmodified lines
26043005 debugfs_create_file_unsafe("amdgpu_dm_force_timing_sync", 0644, root,
26053006 adev, &force_timing_sync_ops);
26063007
2607− return 0;
3008+ debugfs_create_file_unsafe("amdgpu_dm_dmcub_trace_event_en", 0644, root,
3009+ adev, &dmcub_trace_event_state_fops);
26083010 }
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.h+4−1
@@ -30,6 +30,9 @@ · 29 unmodified lines
3030 #include "amdgpu_dm.h"
3131
3232 void connector_debugfs_init(struct amdgpu_dm_connector *connector);
33−int dtn_debugfs_init(struct amdgpu_device *adev);
33+void dtn_debugfs_init(struct amdgpu_device *adev);
34+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
35+void crtc_debugfs_init(struct drm_crtc *crtc);
36+#endif
3437
3538 #endif
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c+42−2
@@ -652,14 +652,54 @@ void *dm_helpers_allocate_gpu_mem( · 651 unmodified lines
652652 size_t size,
653653 long long *addr)
654654 {
655− // TODO
656− return NULL;
655+ struct amdgpu_device *adev = ctx->driver_context;
656+ struct dal_allocation *da;
657+ u32 domain = (type == DC_MEM_ALLOC_TYPE_GART) ?
658+ AMDGPU_GEM_DOMAIN_GTT : AMDGPU_GEM_DOMAIN_VRAM;
659+ int ret;
660+
661+ da = kzalloc(sizeof(struct dal_allocation), GFP_KERNEL);
662+ if (!da)
663+ return NULL;
664+
665+ ret = amdgpu_bo_create_kernel(adev, size, PAGE_SIZE,
666+ domain, &da->bo,
667+ &da->gpu_addr, &da->cpu_ptr);
668+
669+ *addr = da->gpu_addr;
670+
671+ if (ret) {
672+ kfree(da);
673+ return NULL;
674+ }
675+
676+ /* add da to list in dm */
677+ list_add(&da->list, &adev->dm.da_list);
678+
679+ return da->cpu_ptr;
657680 }
658681
659682 void dm_helpers_free_gpu_mem(
660683 struct dc_context *ctx,
661684 enum dc_gpu_mem_alloc_type type,
662685 void *pvMem)
686+{
687+ struct amdgpu_device *adev = ctx->driver_context;
688+ struct dal_allocation *da;
689+
690+ /* walk the da list in DM */
691+ list_for_each_entry(da, &adev->dm.da_list, list) {
692+ if (pvMem == da->cpu_ptr) {
693+ amdgpu_bo_free_kernel(&da->bo, &da->gpu_addr, &da->cpu_ptr);
694+ list_del(&da->list);
695+ kfree(da);
696+ break;
697+ }
698+ }
699+}
700+
701+bool dm_helpers_dmub_outbox0_interrupt_control(struct dc_context *ctx, bool enable)
663702 {
664703 // TODO
704+ return true;
665705 }
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_irq.c+77−36
@@ -82,6 +82,7 @@ struct amdgpu_dm_irq_handler_data { · 81 unmodified lines
8282 struct amdgpu_display_manager *dm;
8383 /* DAL irq source which registered for this interrupt. */
8484 enum dc_irq_source irq_source;
85+ struct work_struct work;
8586 };
8687
8788 #define DM_IRQ_TABLE_LOCK(adev, flags) \
@@ -111,20 +112,10 @@ static void init_handler_common_data(struct amdgpu_dm_irq_handler_data *hcd, · 23 unmodified lines
111112 */
112113 static void dm_irq_work_func(struct work_struct *work)
113114 {
114− struct irq_list_head *irq_list_head =
115− container_of(work, struct irq_list_head, work);
116− struct list_head *handler_list = &irq_list_head->head;
117− struct amdgpu_dm_irq_handler_data *handler_data;
118−
119− list_for_each_entry(handler_data, handler_list, list) {
120− DRM_DEBUG_KMS("DM_IRQ: work_func: for dal_src=%d\n",
121− handler_data->irq_source);
115+ struct amdgpu_dm_irq_handler_data *handler_data =
116+ container_of(work, struct amdgpu_dm_irq_handler_data, work);
122117
123− DRM_DEBUG_KMS("DM_IRQ: schedule_work: for dal_src=%d\n",
124− handler_data->irq_source);
125−
126− handler_data->handler(handler_data->handler_arg);
127− }
118+ handler_data->handler(handler_data->handler_arg);
128119
129120 /* Call a DAL subcomponent which registered for interrupt notification
130121 * at INTERRUPT_LOW_IRQ_CONTEXT.
@@ -156,7 +147,7 @@ static struct list_head *remove_irq_handler(struct amdgpu_device *adev, · 25 unmodified lines
156147 break;
157148 case INTERRUPT_LOW_IRQ_CONTEXT:
158149 default:
159− hnd_list = &adev->dm.irq_handler_list_low_tab[irq_source].head;
150+ hnd_list = &adev->dm.irq_handler_list_low_tab[irq_source];
160151 break;
161152 }
162153
@@ -290,7 +281,8 @@ void *amdgpu_dm_irq_register_interrupt(struct amdgpu_device *adev, · 127 unmodified lines
290281 break;
291282 case INTERRUPT_LOW_IRQ_CONTEXT:
292283 default:
293− hnd_list = &adev->dm.irq_handler_list_low_tab[irq_source].head;
284+ hnd_list = &adev->dm.irq_handler_list_low_tab[irq_source];
285+ INIT_WORK(&handler_data->work, dm_irq_work_func);
294286 break;
295287 }
296288
@@ -372,7 +364,7 @@ void amdgpu_dm_irq_unregister_interrupt(struct amdgpu_device *adev, · 75 unmodified lines
372364 int amdgpu_dm_irq_init(struct amdgpu_device *adev)
373365 {
374366 int src;
375− struct irq_list_head *lh;
367+ struct list_head *lh;
376368
377369 DRM_DEBUG_KMS("DM_IRQ\n");
378370
@@ -381,9 +373,7 @@ int amdgpu_dm_irq_init(struct amdgpu_device *adev) · 2 unmodified lines
381373 for (src = 0; src < DAL_IRQ_SOURCES_NUMBER; src++) {
382374 /* low context handler list init */
383375 lh = &adev->dm.irq_handler_list_low_tab[src];
384− INIT_LIST_HEAD(&lh->head);
385− INIT_WORK(&lh->work, dm_irq_work_func);
386−
376+ INIT_LIST_HEAD(lh);
387377 /* high context handler init */
388378 INIT_LIST_HEAD(&adev->dm.irq_handler_list_high_tab[src]);
389379 }
@@ -400,8 +390,11 @@ int amdgpu_dm_irq_init(struct amdgpu_device *adev) · 10 unmodified lines
400390 void amdgpu_dm_irq_fini(struct amdgpu_device *adev)
401391 {
402392 int src;
403− struct irq_list_head *lh;
393+ struct list_head *lh;
394+ struct list_head *entry, *tmp;
395+ struct amdgpu_dm_irq_handler_data *handler;
404396 unsigned long irq_table_flags;
397+
405398 DRM_DEBUG_KMS("DM_IRQ: releasing resources.\n");
406399 for (src = 0; src < DAL_IRQ_SOURCES_NUMBER; src++) {
407400 DM_IRQ_TABLE_LOCK(adev, irq_table_flags);
@@ -410,7 +403,16 @@ void amdgpu_dm_irq_fini(struct amdgpu_device *adev) · 2 unmodified lines
410403 * (because no code can schedule a new one). */
411404 lh = &adev->dm.irq_handler_list_low_tab[src];
412405 DM_IRQ_TABLE_UNLOCK(adev, irq_table_flags);
413− flush_work(&lh->work);
406+
407+ if (!list_empty(lh)) {
408+ list_for_each_safe(entry, tmp, lh) {
409+ handler = list_entry(
410+ entry,
411+ struct amdgpu_dm_irq_handler_data,
412+ list);
413+ flush_work(&handler->work);
414+ }
415+ }
414416 }
415417 }
416418
@@ -420,6 +422,8 @@ int amdgpu_dm_irq_suspend(struct amdgpu_device *adev) · 3 unmodified lines
420422 struct list_head *hnd_list_h;
421423 struct list_head *hnd_list_l;
422424 unsigned long irq_table_flags;
425+ struct list_head *entry, *tmp;
426+ struct amdgpu_dm_irq_handler_data *handler;
423427
424428 DM_IRQ_TABLE_LOCK(adev, irq_table_flags);
425429
@@ -430,14 +434,22 @@ int amdgpu_dm_irq_suspend(struct amdgpu_device *adev) · 4 unmodified lines
430434 * will be disabled from manage_dm_interrupts on disable CRTC.
431435 */
432436 for (src = DC_IRQ_SOURCE_HPD1; src <= DC_IRQ_SOURCE_HPD6RX; src++) {
433− hnd_list_l = &adev->dm.irq_handler_list_low_tab[src].head;
437+ hnd_list_l = &adev->dm.irq_handler_list_low_tab[src];
434438 hnd_list_h = &adev->dm.irq_handler_list_high_tab[src];
435439 if (!list_empty(hnd_list_l) || !list_empty(hnd_list_h))
436440 dc_interrupt_set(adev->dm.dc, src, false);
437441
438442 DM_IRQ_TABLE_UNLOCK(adev, irq_table_flags);
439− flush_work(&adev->dm.irq_handler_list_low_tab[src].work);
440443
444+ if (!list_empty(hnd_list_l)) {
445+ list_for_each_safe (entry, tmp, hnd_list_l) {
446+ handler = list_entry(
447+ entry,
448+ struct amdgpu_dm_irq_handler_data,
449+ list);
450+ flush_work(&handler->work);
451+ }
452+ }
441453 DM_IRQ_TABLE_LOCK(adev, irq_table_flags);
442454 }
443455
@@ -457,7 +469,7 @@ int amdgpu_dm_irq_resume_early(struct amdgpu_device *adev) · 13 unmodified lines
457469
458470 /* re-enable short pulse interrupts HW interrupt */
459471 for (src = DC_IRQ_SOURCE_HPD1RX; src <= DC_IRQ_SOURCE_HPD6RX; src++) {
460− hnd_list_l = &adev->dm.irq_handler_list_low_tab[src].head;
472+ hnd_list_l = &adev->dm.irq_handler_list_low_tab[src];
461473 hnd_list_h = &adev->dm.irq_handler_list_high_tab[src];
462474 if (!list_empty(hnd_list_l) || !list_empty(hnd_list_h))
463475 dc_interrupt_set(adev->dm.dc, src, true);
@@ -483,7 +495,7 @@ int amdgpu_dm_irq_resume_late(struct amdgpu_device *adev) · 19 unmodified lines
483495 * will be enabled from manage_dm_interrupts on enable CRTC.
484496 */
485497 for (src = DC_IRQ_SOURCE_HPD1; src <= DC_IRQ_SOURCE_HPD6; src++) {
486− hnd_list_l = &adev->dm.irq_handler_list_low_tab[src].head;
498+ hnd_list_l = &adev->dm.irq_handler_list_low_tab[src];
487499 hnd_list_h = &adev->dm.irq_handler_list_high_tab[src];
488500 if (!list_empty(hnd_list_l) || !list_empty(hnd_list_h))
489501 dc_interrupt_set(adev->dm.dc, src, true);
@@ -500,22 +512,51 @@ int amdgpu_dm_irq_resume_late(struct amdgpu_device *adev) · 10 unmodified lines
500512 static void amdgpu_dm_irq_schedule_work(struct amdgpu_device *adev,
501513 enum dc_irq_source irq_source)
502514 {
503− unsigned long irq_table_flags;
504− struct work_struct *work = NULL;
515+ struct list_head *handler_list = &adev->dm.irq_handler_list_low_tab[irq_source];
516+ struct amdgpu_dm_irq_handler_data *handler_data;
517+ bool work_queued = false;
505518
506− DM_IRQ_TABLE_LOCK(adev, irq_table_flags);
519+ if (list_empty(handler_list))
520+ return;
507521
508− if (!list_empty(&adev->dm.irq_handler_list_low_tab[irq_source].head))
509− work = &adev->dm.irq_handler_list_low_tab[irq_source].work;
522+ list_for_each_entry (handler_data, handler_list, list) {
523+ if (queue_work(system_highpri_wq, &handler_data->work)) {
524+ work_queued = true;
525+ break;
526+ }
527+ }
510528
511− DM_IRQ_TABLE_UNLOCK(adev, irq_table_flags);
529+ if (!work_queued) {
530+ struct amdgpu_dm_irq_handler_data *handler_data_add;
531+ /*get the amdgpu_dm_irq_handler_data of first item pointed by handler_list*/
532+ handler_data = container_of(handler_list->next, struct amdgpu_dm_irq_handler_data, list);
512533
513− if (work) {
514− if (!schedule_work(work))
515− DRM_INFO("amdgpu_dm_irq_schedule_work FAILED src %d\n",
516− irq_source);
517− }
534+ /*allocate a new amdgpu_dm_irq_handler_data*/
535+ handler_data_add = kzalloc(sizeof(*handler_data), GFP_KERNEL);
536+ if (!handler_data_add) {
537+ DRM_ERROR("DM_IRQ: failed to allocate irq handler!\n");
538+ return;
539+ }
540+
541+ /*copy new amdgpu_dm_irq_handler_data members from handler_data*/
542+ handler_data_add->handler = handler_data->handler;
543+ handler_data_add->handler_arg = handler_data->handler_arg;
544+ handler_data_add->dm = handler_data->dm;
545+ handler_data_add->irq_source = irq_source;
518546
547+ list_add_tail(&handler_data_add->list, handler_list);
548+
549+ INIT_WORK(&handler_data_add->work, dm_irq_work_func);
550+
551+ if (queue_work(system_highpri_wq, &handler_data_add->work))
552+ DRM_DEBUG("Queued work for handling interrupt from "
553+ "display for IRQ source %d\n",
554+ irq_source);
555+ else
556+ DRM_ERROR("Failed to queue work for handling interrupt "
557+ "from display for IRQ source %d\n",
558+ irq_source);
559+ }
519560 }
520561
521562 /*
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_irq_params.h+9−0
@@ -26,12 +26,21 @@ · 25 unmodified lines
2626 #ifndef __AMDGPU_DM_IRQ_PARAMS_H__
2727 #define __AMDGPU_DM_IRQ_PARAMS_H__
2828
29+#include "amdgpu_dm_crc.h"
30+
2931 struct dm_irq_params {
3032 u32 last_flip_vblank;
3133 struct mod_vrr_params vrr_params;
3234 struct dc_stream_state *stream;
3335 int active_planes;
3436 struct mod_freesync_config freesync_config;
37+
38+#ifdef CONFIG_DEBUG_FS
39+ enum amdgpu_dm_pipe_crc_source crc_src;
40+#ifdef CONFIG_DRM_AMD_SECURE_DISPLAY
41+ struct crc_window_parm crc_window;
42+#endif
43+#endif
3544 };
3645
3746 #endif /* __AMDGPU_DM_IRQ_PARAMS_H__ */
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_mst_types.c+10−7
@@ -38,6 +38,7 @@ · 37 unmodified lines
3838 #include "dc_link_ddc.h"
3939
4040 #include "i2caux_interface.h"
41+#include "dmub_cmd.h"
4142 #if defined(CONFIG_DEBUG_FS)
4243 #include "amdgpu_dm_debugfs.h"
4344 #endif
@@ -51,7 +52,7 @@ static ssize_t dm_dp_aux_transfer(struct drm_dp_aux *aux, · 7 unmodified lines
5152 {
5253 ssize_t result = 0;
5354 struct aux_payload payload;
54− enum aux_channel_operation_result operation_result;
55+ enum aux_return_code_type operation_result;
5556
5657 if (WARN_ON(msg->size > 16))
5758 return -E2BIG;
@@ -73,17 +74,19 @@ static ssize_t dm_dp_aux_transfer(struct drm_dp_aux *aux, · 15 unmodified lines
7374
7475 if (result < 0)
7576 switch (operation_result) {
76− case AUX_CHANNEL_OPERATION_SUCCEEDED:
77+ case AUX_RET_SUCCESS:
7778 break;
78− case AUX_CHANNEL_OPERATION_FAILED_HPD_DISCON:
79− case AUX_CHANNEL_OPERATION_FAILED_REASON_UNKNOWN:
79+ case AUX_RET_ERROR_HPD_DISCON:
80+ case AUX_RET_ERROR_UNKNOWN:
81+ case AUX_RET_ERROR_INVALID_OPERATION:
82+ case AUX_RET_ERROR_PROTOCOL_ERROR:
8083 result = -EIO;
8184 break;
82− case AUX_CHANNEL_OPERATION_FAILED_INVALID_REPLY:
83− case AUX_CHANNEL_OPERATION_FAILED_ENGINE_ACQUIRE:
85+ case AUX_RET_ERROR_INVALID_REPLY:
86+ case AUX_RET_ERROR_ENGINE_ACQUIRE:
8487 result = -EBUSY;
8588 break;
86− case AUX_CHANNEL_OPERATION_FAILED_TIMEOUT:
89+ case AUX_RET_ERROR_TIMEOUT:
8790 result = -ETIMEDOUT;
8891 break;
8992 }
drivers/gpu/drm/amd/display/dc/Makefile+1−1
@@ -54,7 +54,7 @@ AMD_DC = $(addsuffix /Makefile, $(addprefix $(FULL_AMD_DISPLAY_PATH)/dc/,$(DC_LI · 53 unmodified lines
5454
5555 include $(AMD_DC)
5656
57−DISPLAY_CORE = dc.o dc_link.o dc_resource.o dc_hw_sequencer.o dc_sink.o \
57+DISPLAY_CORE = dc.o dc_stat.o dc_link.o dc_resource.o dc_hw_sequencer.o dc_sink.o \
5858 dc_surface.o dc_link_hwss.o dc_link_dp.o dc_link_ddc.o dc_debug.o dc_stream.o
5959
6060 ifdef CONFIG_DRM_AMD_DC_DCN
drivers/gpu/drm/amd/display/dc/basics/dc_common.c+6−14
@@ -49,24 +49,20 @@ bool is_rgb_cspace(enum dc_color_space output_color_space) · 48 unmodified lines
4949 }
5050 }
5151
52−bool is_child_pipe_tree_visible(struct pipe_ctx *pipe_ctx)
52+bool is_lower_pipe_tree_visible(struct pipe_ctx *pipe_ctx)
5353 {
5454 if (pipe_ctx->plane_state && pipe_ctx->plane_state->visible)
5555 return true;
56− if (pipe_ctx->bottom_pipe && is_child_pipe_tree_visible(pipe_ctx->bottom_pipe))
57− return true;
58− if (pipe_ctx->next_odm_pipe && is_child_pipe_tree_visible(pipe_ctx->next_odm_pipe))
56+ if (pipe_ctx->bottom_pipe && is_lower_pipe_tree_visible(pipe_ctx->bottom_pipe))
5957 return true;
6058 return false;
6159 }
6260
63−bool is_parent_pipe_tree_visible(struct pipe_ctx *pipe_ctx)
61+bool is_upper_pipe_tree_visible(struct pipe_ctx *pipe_ctx)
6462 {
6563 if (pipe_ctx->plane_state && pipe_ctx->plane_state->visible)
6664 return true;
67− if (pipe_ctx->top_pipe && is_parent_pipe_tree_visible(pipe_ctx->top_pipe))
68− return true;
69− if (pipe_ctx->prev_odm_pipe && is_parent_pipe_tree_visible(pipe_ctx->prev_odm_pipe))
65+ if (pipe_ctx->top_pipe && is_upper_pipe_tree_visible(pipe_ctx->top_pipe))
7066 return true;
7167 return false;
7268 }
@@ -75,13 +71,9 @@ bool is_pipe_tree_visible(struct pipe_ctx *pipe_ctx) · 2 unmodified lines
7571 {
7672 if (pipe_ctx->plane_state && pipe_ctx->plane_state->visible)
7773 return true;
78− if (pipe_ctx->top_pipe && is_parent_pipe_tree_visible(pipe_ctx->top_pipe))
79− return true;
80− if (pipe_ctx->bottom_pipe && is_child_pipe_tree_visible(pipe_ctx->bottom_pipe))
81− return true;
82− if (pipe_ctx->prev_odm_pipe && is_parent_pipe_tree_visible(pipe_ctx->prev_odm_pipe))
74+ if (pipe_ctx->top_pipe && is_upper_pipe_tree_visible(pipe_ctx->top_pipe))
8375 return true;
84− if (pipe_ctx->next_odm_pipe && is_child_pipe_tree_visible(pipe_ctx->next_odm_pipe))
76+ if (pipe_ctx->bottom_pipe && is_lower_pipe_tree_visible(pipe_ctx->bottom_pipe))
8577 return true;
8678 return false;
8779 }
drivers/gpu/drm/amd/display/dc/basics/dc_common.h+2−2
@@ -30,9 +30,9 @@ · 29 unmodified lines
3030
3131 bool is_rgb_cspace(enum dc_color_space output_color_space);
3232
33−bool is_child_pipe_tree_visible(struct pipe_ctx *pipe_ctx);
33+bool is_lower_pipe_tree_visible(struct pipe_ctx *pipe_ctx);
3434
35−bool is_parent_pipe_tree_visible(struct pipe_ctx *pipe_ctx);
35+bool is_upper_pipe_tree_visible(struct pipe_ctx *pipe_ctx);
3636
3737 bool is_pipe_tree_visible(struct pipe_ctx *pipe_ctx);
3838
drivers/gpu/drm/amd/display/dc/bios/bios_parser2.c+3−2
@@ -1180,14 +1180,15 @@ static enum bp_result bios_parser_enable_disp_power_gating( · 1179 unmodified lines
11801180
11811181 static enum bp_result bios_parser_enable_lvtma_control(
11821182 struct dc_bios *dcb,
1183− uint8_t uc_pwr_on)
1183+ uint8_t uc_pwr_on,
1184+ uint8_t panel_instance)
11841185 {
11851186 struct bios_parser *bp = BP_FROM_DCB(dcb);
11861187
11871188 if (!bp->cmd_tbl.enable_lvtma_control)
11881189 return BP_RESULT_FAILURE;
11891190
1190− return bp->cmd_tbl.enable_lvtma_control(bp, uc_pwr_on);
1191+ return bp->cmd_tbl.enable_lvtma_control(bp, uc_pwr_on, panel_instance);
11911192 }
11921193
11931194 static bool bios_parser_is_accelerated_mode(
drivers/gpu/drm/amd/display/dc/bios/bios_parser_helper.c+3−2
@@ -64,9 +64,10 @@ bool bios_is_accelerated_mode( · 63 unmodified lines
6464
6565
6666 void bios_set_scratch_acc_mode_change(
67− struct dc_bios *bios)
67+ struct dc_bios *bios,
68+ uint32_t state)
6869 {
69− REG_UPDATE(BIOS_SCRATCH_6, S6_ACC_MODE, 1);
70+ REG_UPDATE(BIOS_SCRATCH_6, S6_ACC_MODE, state);
7071 }
7172
7273
drivers/gpu/drm/amd/display/dc/bios/bios_parser_helper.h+1−1
@@ -32,7 +32,7 @@ uint8_t *bios_get_image(struct dc_bios *bp, uint32_t offset, · 31 unmodified lines
3232 uint32_t size);
3333
3434 bool bios_is_accelerated_mode(struct dc_bios *bios);
35−void bios_set_scratch_acc_mode_change(struct dc_bios *bios);
35+void bios_set_scratch_acc_mode_change(struct dc_bios *bios, uint32_t state);
3636 void bios_set_scratch_critical_state(struct dc_bios *bios, bool state);
3737 uint32_t bios_get_vga_enabled_displays(struct dc_bios *bios);
3838
drivers/gpu/drm/amd/display/dc/bios/command_table.c+21−0
@@ -1531,6 +1531,27 @@ static enum bp_result adjust_display_pll_v2( · 1530 unmodified lines
15311531 params.ucEncodeMode =
15321532 (uint8_t)bp->cmd_helper->encoder_mode_bp_to_atom(
15331533 bp_params->signal_type, false);
1534+
1535+ if (EXEC_BIOS_CMD_TABLE(AdjustDisplayPll, params)) {
1536+ /* Convert output pixel clock back 10KHz-->KHz: multiply
1537+ * original pixel clock in KHz by ratio
1538+ * [output pxlClk/input pxlClk] */
1539+ uint64_t pixel_clk_10_khz_out =
1540+ (uint64_t)le16_to_cpu(params.usPixelClock);
1541+ uint64_t pixel_clk = (uint64_t)bp_params->pixel_clock;
1542+
1543+ if (pixel_clock_10KHz_in != 0) {
1544+ bp_params->adjusted_pixel_clock =
1545+ div_u64(pixel_clk * pixel_clk_10_khz_out,
1546+ pixel_clock_10KHz_in);
1547+ } else {
1548+ bp_params->adjusted_pixel_clock = 0;
1549+ BREAK_TO_DEBUGGER();
1550+ }
1551+
1552+ result = BP_RESULT_OK;
1553+ }
1554+
15341555 return result;
15351556 }
15361557
drivers/gpu/drm/amd/display/dc/bios/command_table2.c+84−9
@@ -218,6 +218,10 @@ static enum bp_result transmitter_control_v1_6( · 217 unmodified lines
218218 struct bios_parser *bp,
219219 struct bp_transmitter_control *cntl);
220220
221+static enum bp_result transmitter_control_v1_7(
222+ struct bios_parser *bp,
223+ struct bp_transmitter_control *cntl);
224+
221225 static enum bp_result transmitter_control_fallback(
222226 struct bios_parser *bp,
223227 struct bp_transmitter_control *cntl);
@@ -233,6 +237,9 @@ static void init_transmitter_control(struct bios_parser *bp) · 9 unmodified lines
233237 case 6:
234238 bp->cmd_tbl.transmitter_control = transmitter_control_v1_6;
235239 break;
240+ case 7:
241+ bp->cmd_tbl.transmitter_control = transmitter_control_v1_7;
242+ break;
236243 default:
237244 dm_output_to_console("Don't have transmitter_control for v%d\n", crev);
238245 bp->cmd_tbl.transmitter_control = transmitter_control_fallback;
@@ -304,13 +311,76 @@ static enum bp_result transmitter_control_v1_6( · 65 unmodified lines
304311 return result;
305312 }
306313
314+static void transmitter_control_dmcub_v1_7(
315+ struct dc_dmub_srv *dmcub,
316+ struct dmub_dig_transmitter_control_data_v1_7 *dig)
317+{
318+ union dmub_rb_cmd cmd;
319+
320+ memset(&cmd, 0, sizeof(cmd));
321+
322+ cmd.dig1_transmitter_control.header.type = DMUB_CMD__VBIOS;
323+ cmd.dig1_transmitter_control.header.sub_type =
324+ DMUB_CMD__VBIOS_DIG1_TRANSMITTER_CONTROL;
325+ cmd.dig1_transmitter_control.header.payload_bytes =
326+ sizeof(cmd.dig1_transmitter_control) -
327+ sizeof(cmd.dig1_transmitter_control.header);
328+ cmd.dig1_transmitter_control.transmitter_control.dig_v1_7 = *dig;
329+
330+ dc_dmub_srv_cmd_queue(dmcub, &cmd);
331+ dc_dmub_srv_cmd_execute(dmcub);
332+ dc_dmub_srv_wait_idle(dmcub);
333+}
334+
335+static enum bp_result transmitter_control_v1_7(
336+ struct bios_parser *bp,
337+ struct bp_transmitter_control *cntl)
338+{
339+ enum bp_result result = BP_RESULT_FAILURE;
340+ const struct command_table_helper *cmd = bp->cmd_helper;
341+ struct dmub_dig_transmitter_control_data_v1_7 dig_v1_7 = {0};
342+
343+ dig_v1_7.phyid = cmd->phy_id_to_atom(cntl->transmitter);
344+ dig_v1_7.action = (uint8_t)cntl->action;
345+
346+ if (cntl->action == TRANSMITTER_CONTROL_SET_VOLTAGE_AND_PREEMPASIS)
347+ dig_v1_7.mode_laneset.dplaneset = (uint8_t)cntl->lane_settings;
348+ else
349+ dig_v1_7.mode_laneset.digmode =
350+ cmd->signal_type_to_atom_dig_mode(cntl->signal);
351+
352+ dig_v1_7.lanenum = (uint8_t)cntl->lanes_number;
353+ dig_v1_7.hpdsel = cmd->hpd_sel_to_atom(cntl->hpd_sel);
354+ dig_v1_7.digfe_sel = cmd->dig_encoder_sel_to_atom(cntl->engine_id);
355+ dig_v1_7.connobj_id = (uint8_t)cntl->connector_obj_id.id;
356+ dig_v1_7.symclk_units.symclk_10khz = cntl->pixel_clock/10;
357+
358+ if (cntl->action == TRANSMITTER_CONTROL_ENABLE ||
359+ cntl->action == TRANSMITTER_CONTROL_ACTIAVATE ||
360+ cntl->action == TRANSMITTER_CONTROL_DEACTIVATE) {
361+ DC_LOG_BIOS("%s:dig_v1_7.symclk_units.symclk_10khz = %d\n",
362+ __func__, dig_v1_7.symclk_units.symclk_10khz);
363+ }
364+
365+ if (bp->base.ctx->dc->ctx->dmub_srv &&
366+ bp->base.ctx->dc->debug.dmub_command_table) {
367+ transmitter_control_dmcub_v1_7(bp->base.ctx->dmub_srv, &dig_v1_7);
368+ return BP_RESULT_OK;
369+ }
370+
371+/*color_depth not used any more, driver has deep color factor in the Phyclk*/
372+ if (EXEC_BIOS_CMD_TABLE(dig1transmittercontrol, dig_v1_7))
373+ result = BP_RESULT_OK;
374+ return result;
375+}
376+
307377 static enum bp_result transmitter_control_fallback(
308378 struct bios_parser *bp,
309379 struct bp_transmitter_control *cntl)
310380 {
311381 if (bp->base.ctx->dc->ctx->dmub_srv &&
312382 bp->base.ctx->dc->debug.dmub_command_table) {
313− return transmitter_control_v1_6(bp, cntl);
383+ return transmitter_control_v1_7(bp, cntl);
314384 }
315385
316386 return BP_RESULT_FAILURE;
@@ -911,7 +981,8 @@ static unsigned int get_smu_clock_info_v3_1(struct bios_parser *bp, uint8_t id) · 594 unmodified lines
911981
912982 static enum bp_result enable_lvtma_control(
913983 struct bios_parser *bp,
914− uint8_t uc_pwr_on);
984+ uint8_t uc_pwr_on,
985+ uint8_t panel_instance);
915986
916987 static void init_enable_lvtma_control(struct bios_parser *bp)
917988 {
@@ -922,19 +993,21 @@ static void init_enable_lvtma_control(struct bios_parser *bp) · 4 unmodified lines
922993
923994 static void enable_lvtma_control_dmcub(
924995 struct dc_dmub_srv *dmcub,
925− uint8_t uc_pwr_on)
996+ uint8_t uc_pwr_on,
997+ uint8_t panel_instance)
926998 {
927999
9281000 union dmub_rb_cmd cmd;
9291001
9301002 memset(&cmd, 0, sizeof(cmd));
9311003
932− cmd.cmd_common.header.type = DMUB_CMD__VBIOS;
933− cmd.cmd_common.header.sub_type =
1004+ cmd.lvtma_control.header.type = DMUB_CMD__VBIOS;
1005+ cmd.lvtma_control.header.sub_type =
9341006 DMUB_CMD__VBIOS_LVTMA_CONTROL;
935− cmd.cmd_common.cmd_buffer[0] =
1007+ cmd.lvtma_control.data.uc_pwr_action =
9361008 uc_pwr_on;
937−
1009+ cmd.lvtma_control.data.panel_inst =
1010+ panel_instance;
9381011 dc_dmub_srv_cmd_queue(dmcub, &cmd);
9391012 dc_dmub_srv_cmd_execute(dmcub);
9401013 dc_dmub_srv_wait_idle(dmcub);
@@ -943,14 +1016,16 @@ static void enable_lvtma_control_dmcub( · 2 unmodified lines
9431016
9441017 static enum bp_result enable_lvtma_control(
9451018 struct bios_parser *bp,
946− uint8_t uc_pwr_on)
1019+ uint8_t uc_pwr_on,
1020+ uint8_t panel_instance)
9471021 {
9481022 enum bp_result result = BP_RESULT_FAILURE;
9491023
9501024 if (bp->base.ctx->dc->ctx->dmub_srv &&
9511025 bp->base.ctx->dc->debug.dmub_command_table) {
9521026 enable_lvtma_control_dmcub(bp->base.ctx->dmub_srv,
953− uc_pwr_on);
1027+ uc_pwr_on,
1028+ panel_instance);
9541029 return BP_RESULT_OK;
9551030 }
9561031 return result;
drivers/gpu/drm/amd/display/dc/bios/command_table2.h+2−1
@@ -95,7 +95,8 @@ struct cmd_tbl { · 94 unmodified lines
9595 unsigned int (*get_smu_clock_info)(
9696 struct bios_parser *bp, uint8_t id);
9797 enum bp_result (*enable_lvtma_control)(struct bios_parser *bp,
98− uint8_t uc_pwr_on);
98+ uint8_t uc_pwr_on,
99+ uint8_t panel_instance);
99100 };
100101
101102 void dal_firmware_parser_init_cmd_tbl(struct bios_parser *bp);
drivers/gpu/drm/amd/display/dc/clk_mgr/clk_mgr.c+13−4
@@ -87,12 +87,16 @@ int clk_mgr_helper_get_active_plane_cnt( · 86 unmodified lines
8787
8888 void clk_mgr_exit_optimized_pwr_state(const struct dc *dc, struct clk_mgr *clk_mgr)
8989 {
90− struct dc_link *edp_link = get_edp_link(dc);
90+ struct dc_link *edp_links[MAX_NUM_EDP];
91+ struct dc_link *edp_link = NULL;
92+ int edp_num;
9193
94+ get_edp_links(dc, edp_links, &edp_num);
9295 if (dc->hwss.exit_optimized_pwr_state)
9396 dc->hwss.exit_optimized_pwr_state(dc, dc->current_state);
9497
95− if (edp_link) {
98+ if (edp_num) {
99+ edp_link = edp_links[0];
96100 clk_mgr->psr_allow_active_cache = edp_link->psr_settings.psr_allow_active;
97101 dc_link_set_psr_allow_active(edp_link, false, false, false);
98102 }
@@ -101,11 +105,16 @@ void clk_mgr_exit_optimized_pwr_state(const struct dc *dc, struct clk_mgr *clk_m · 2 unmodified lines
101105
102106 void clk_mgr_optimize_pwr_state(const struct dc *dc, struct clk_mgr *clk_mgr)
103107 {
104− struct dc_link *edp_link = get_edp_link(dc);
108+ struct dc_link *edp_links[MAX_NUM_EDP];
109+ struct dc_link *edp_link = NULL;
110+ int edp_num;
105111
106− if (edp_link)
112+ get_edp_links(dc, edp_links, &edp_num);
113+ if (edp_num) {
114+ edp_link = edp_links[0];
107115 dc_link_set_psr_allow_active(edp_link,
108116 clk_mgr->psr_allow_active_cache, false, false);
117+ }
109118
110119 if (dc->hwss.optimize_pwr_state)
111120 dc->hwss.optimize_pwr_state(dc, dc->current_state);
drivers/gpu/drm/amd/display/dc/clk_mgr/dcn20/dcn20_clk_mgr.c+1−1
@@ -361,7 +361,7 @@ void dcn2_read_clocks_from_hw_dentist(struct clk_mgr *clk_mgr_base) · 360 unmodified lines
361361 REG_GET(DENTIST_DISPCLK_CNTL, DENTIST_DPPCLK_WDIVIDER, &dppclk_wdivider);
362362
363363 disp_divider = dentist_get_divider_from_did(dispclk_wdivider);
364− dpp_divider = dentist_get_divider_from_did(dispclk_wdivider);
364+ dpp_divider = dentist_get_divider_from_did(dppclk_wdivider);
365365
366366 if (disp_divider && dpp_divider) {
367367 /* Calculate the current DFS clock, in kHz.*/
drivers/gpu/drm/amd/display/dc/clk_mgr/dcn30/dcn30_clk_mgr.c+3−1
@@ -252,6 +252,7 @@ static void dcn3_update_clocks(struct clk_mgr *clk_mgr_base, · 251 unmodified lines
252252 bool force_reset = false;
253253 bool update_uclk = false;
254254 bool p_state_change_support;
255+ int total_plane_count;
255256
256257 if (dc->work_arounds.skip_clock_update || !clk_mgr->smu_present)
257258 return;
@@ -292,7 +293,8 @@ static void dcn3_update_clocks(struct clk_mgr *clk_mgr_base, · 34 unmodified lines
292293 clk_mgr_base->clks.socclk_khz = new_clocks->socclk_khz;
293294
294295 clk_mgr_base->clks.prev_p_state_change_support = clk_mgr_base->clks.p_state_change_support;
295− p_state_change_support = new_clocks->p_state_change_support || (display_count == 0);
296+ total_plane_count = clk_mgr_helper_get_active_plane_cnt(dc, context);
297+ p_state_change_support = new_clocks->p_state_change_support || (total_plane_count == 0);
296298 if (should_update_pstate_support(safe_to_lower, p_state_change_support, clk_mgr_base->clks.p_state_change_support)) {
297299 clk_mgr_base->clks.p_state_change_support = p_state_change_support;
298300
drivers/gpu/drm/amd/display/dc/core/dc.c+260−46
@@ -41,6 +41,7 @@ · 40 unmodified lines
4141 #include "dc_bios_types.h"
4242
4343 #include "bios_parser_interface.h"
44+#include "bios/bios_parser_helper.h"
4445 #include "include/irq_service_interface.h"
4546 #include "transform.h"
4647 #include "dmcu.h"
@@ -68,6 +69,7 @@ · 21 unmodified lines
6869
6970 #include "dmub/dmub_srv.h"
7071
72+#include "i2caux_interface.h"
7173 #include "dce/dmub_hw_lock_mgr.h"
7274
7375 #include "dc_trace.h"
@@ -163,6 +165,18 @@ static uint32_t get_num_of_internal_disp(struct dc_link **links, uint32_t num_li · 89 unmodified lines
163165 return count;
164166 }
165167
168+static int get_seamless_boot_stream_count(struct dc_state *ctx)
169+{
170+ uint8_t i;
171+ uint8_t seamless_boot_stream_count = 0;
172+
173+ for (i = 0; i < ctx->stream_count; i++)
174+ if (ctx->streams[i]->apply_seamless_boot_optimization)
175+ seamless_boot_stream_count++;
176+
177+ return seamless_boot_stream_count;
178+}
179+
166180 static bool create_links(
167181 struct dc *dc,
168182 uint32_t num_virtual_links)
@@ -334,6 +348,88 @@ bool dc_stream_get_crtc_position(struct dc *dc, · 165 unmodified lines
334348 return ret;
335349 }
336350
351+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
352+bool dc_stream_forward_dmcu_crc_window(struct dc *dc, struct dc_stream_state *stream,
353+ struct crc_params *crc_window)
354+{
355+ int i;
356+ struct dmcu *dmcu = dc->res_pool->dmcu;
357+ struct pipe_ctx *pipe;
358+ struct crc_region tmp_win, *crc_win;
359+ struct otg_phy_mux mapping_tmp, *mux_mapping;
360+
361+ /*crc window can't be null*/
362+ if (!crc_window)
363+ return false;
364+
365+ if ((dmcu != NULL && dmcu->funcs->is_dmcu_initialized(dmcu))) {
366+ crc_win = &tmp_win;
367+ mux_mapping = &mapping_tmp;
368+ /*set crc window*/
369+ tmp_win.x_start = crc_window->windowa_x_start;
370+ tmp_win.y_start = crc_window->windowa_y_start;
371+ tmp_win.x_end = crc_window->windowa_x_end;
372+ tmp_win.y_end = crc_window->windowa_y_end;
373+
374+ for (i = 0; i < MAX_PIPES; i++) {
375+ pipe = &dc->current_state->res_ctx.pipe_ctx[i];
376+ if (pipe->stream == stream && !pipe->top_pipe && !pipe->prev_odm_pipe)
377+ break;
378+ }
379+
380+ /* Stream not found */
381+ if (i == MAX_PIPES)
382+ return false;
383+
384+
385+ /*set mux routing info*/
386+ mapping_tmp.phy_output_num = stream->link->link_enc_hw_inst;
387+ mapping_tmp.otg_output_num = pipe->stream_res.tg->inst;
388+
389+ dmcu->funcs->forward_crc_window(dmcu, crc_win, mux_mapping);
390+ } else {
391+ DC_LOG_DC("dmcu is not initialized");
392+ return false;
393+ }
394+
395+ return true;
396+}
397+
398+bool dc_stream_stop_dmcu_crc_win_update(struct dc *dc, struct dc_stream_state *stream)
399+{
400+ int i;
401+ struct dmcu *dmcu = dc->res_pool->dmcu;
402+ struct pipe_ctx *pipe;
403+ struct otg_phy_mux mapping_tmp, *mux_mapping;
404+
405+ if ((dmcu != NULL && dmcu->funcs->is_dmcu_initialized(dmcu))) {
406+ mux_mapping = &mapping_tmp;
407+
408+ for (i = 0; i < MAX_PIPES; i++) {
409+ pipe = &dc->current_state->res_ctx.pipe_ctx[i];
410+ if (pipe->stream == stream && !pipe->top_pipe && !pipe->prev_odm_pipe)
411+ break;
412+ }
413+
414+ /* Stream not found */
415+ if (i == MAX_PIPES)
416+ return false;
417+
418+
419+ /*set mux routing info*/
420+ mapping_tmp.phy_output_num = stream->link->link_enc_hw_inst;
421+ mapping_tmp.otg_output_num = pipe->stream_res.tg->inst;
422+
423+ dmcu->funcs->stop_crc_win_update(dmcu, mux_mapping);
424+ } else {
425+ DC_LOG_DC("dmcu is not initialized");
426+ return false;
427+ }
428+
429+ return true;
430+}
431+#endif
432+
337433 /**
338434 * dc_stream_configure_crc() - Configure CRC capture for the given stream.
339435 * @dc: DC Object
@@ -970,7 +1066,6 @@ struct dc *dc_create(const struct dc_init_data *init_params) · 630 unmodified lines
9701066 full_pipe_count,
9711067 dc->res_pool->stream_enc_count);
9721068
973− dc->optimize_seamless_boot_streams = 0;
9741069 dc->caps.max_links = dc->link_count;
9751070 dc->caps.max_audios = dc->res_pool->audio_count;
9761071 dc->caps.linear_pitch_alignment = 64;
@@ -1000,22 +1095,25 @@ destruct_dc: · 23 unmodified lines
10001095
10011096 static void detect_edp_presence(struct dc *dc)
10021097 {
1003− struct dc_link *edp_link = get_edp_link(dc);
1004− bool edp_sink_present = true;
1098+ struct dc_link *edp_links[MAX_NUM_EDP];
1099+ struct dc_link *edp_link = NULL;
1100+ enum dc_connection_type type;
1101+ int i;
1102+ int edp_num;
10051103
1006− if (!edp_link)
1104+ get_edp_links(dc, edp_links, &edp_num);
1105+ if (!edp_num)
10071106 return;
10081107
1009− if (dc->config.edp_not_connected) {
1010− edp_sink_present = false;
1011− } else {
1012− enum dc_connection_type type;
1013− dc_link_detect_sink(edp_link, &type);
1014− if (type == dc_connection_none)
1015− edp_sink_present = false;
1108+ for (i = 0; i < edp_num; i++) {
1109+ edp_link = edp_links[i];
1110+ if (dc->config.edp_not_connected) {
1111+ edp_link->edp_sink_present = false;
1112+ } else {
1113+ dc_link_detect_sink(edp_link, &type);
1114+ edp_link->edp_sink_present = (type != dc_connection_none);
1115+ }
10161116 }
1017−
1018− edp_link->edp_sink_present = edp_sink_present;
10191117 }
10201118
10211119 void dc_hardware_init(struct dc *dc)
@@ -1091,6 +1189,7 @@ static void program_timing_sync( · 69 unmodified lines
10911189
10921190 for (i = 0; i < pipe_count; i++) {
10931191 int group_size = 1;
1192+ enum timing_synchronization_type sync_type = NOT_SYNCHRONIZABLE;
10941193 struct pipe_ctx *pipe_set[MAX_PIPES];
10951194
10961195 if (!unsynced_pipes[i])
@@ -1105,10 +1204,22 @@ static void program_timing_sync( · 8 unmodified lines
11051204 for (j = i + 1; j < pipe_count; j++) {
11061205 if (!unsynced_pipes[j])
11071206 continue;
1108−
1109− if (resource_are_streams_timing_synchronizable(
1207+ if (sync_type != TIMING_SYNCHRONIZABLE &&
1208+ dc->hwss.enable_vblanks_synchronization &&
1209+ unsynced_pipes[j]->stream_res.tg->funcs->align_vblanks &&
1210+ resource_are_vblanks_synchronizable(
1211+ unsynced_pipes[j]->stream,
1212+ pipe_set[0]->stream)) {
1213+ sync_type = VBLANK_SYNCHRONIZABLE;
1214+ pipe_set[group_size] = unsynced_pipes[j];
1215+ unsynced_pipes[j] = NULL;
1216+ group_size++;
1217+ } else
1218+ if (sync_type != VBLANK_SYNCHRONIZABLE &&
1219+ resource_are_streams_timing_synchronizable(
11101220 unsynced_pipes[j]->stream,
11111221 pipe_set[0]->stream)) {
1222+ sync_type = TIMING_SYNCHRONIZABLE;
11121223 pipe_set[group_size] = unsynced_pipes[j];
11131224 unsynced_pipes[j] = NULL;
11141225 group_size++;
@@ -1134,7 +1245,6 @@ static void program_timing_sync( · 19 unmodified lines
11341245 }
11351246 }
11361247
1137−
11381248 for (k = 0; k < group_size; k++) {
11391249 struct dc_stream_status *status = dc_stream_get_status_from_state(ctx, pipe_set[k]->stream);
11401250
@@ -1164,8 +1274,14 @@ static void program_timing_sync( · 23 unmodified lines
11641274 }
11651275
11661276 if (group_size > 1) {
1167− dc->hwss.enable_timing_synchronization(
1168− dc, group_index, group_size, pipe_set);
1277+ if (sync_type == TIMING_SYNCHRONIZABLE) {
1278+ dc->hwss.enable_timing_synchronization(
1279+ dc, group_index, group_size, pipe_set);
1280+ } else
1281+ if (sync_type == VBLANK_SYNCHRONIZABLE) {
1282+ dc->hwss.enable_vblanks_synchronization(
1283+ dc, group_index, group_size, pipe_set);
1284+ }
11691285 group_index++;
11701286 }
11711287 num_group++;
@@ -1202,8 +1318,9 @@ bool dc_validate_seamless_boot_timing(const struct dc *dc, · 30 unmodified lines
12021318 unsigned int i, enc_inst, tg_inst = 0;
12031319
12041320 // Seamless port only support single DP and EDP so far
1205− if (sink->sink_signal != SIGNAL_TYPE_DISPLAY_PORT &&
1206− sink->sink_signal != SIGNAL_TYPE_EDP)
1321+ if ((sink->sink_signal != SIGNAL_TYPE_DISPLAY_PORT &&
1322+ sink->sink_signal != SIGNAL_TYPE_EDP) ||
1323+ sink->sink_signal == SIGNAL_TYPE_DISPLAY_PORT_MST)
12071324 return false;
12081325
12091326 /* Check for enabled DIG to identify enabled display */
@@ -1377,11 +1494,7 @@ static enum dc_status dc_commit_state_no_check(struct dc *dc, struct dc_state *c · 167 unmodified lines
13771494 dc->hwss.enable_accelerated_mode(dc, context);
13781495 }
13791496
1380− for (i = 0; i < context->stream_count; i++)
1381− if (context->streams[i]->apply_seamless_boot_optimization)
1382− dc->optimize_seamless_boot_streams++;
1383−
1384− if (context->stream_count > dc->optimize_seamless_boot_streams ||
1497+ if (context->stream_count > get_seamless_boot_stream_count(context) ||
13851498 context->stream_count == 0)
13861499 dc->hwss.prepare_bandwidth(dc, context);
13871500
@@ -1464,7 +1577,7 @@ static enum dc_status dc_commit_state_no_check(struct dc *dc, struct dc_state *c · 76 unmodified lines
14641577
14651578 dc_enable_stereo(dc, context, dc_streams, context->stream_count);
14661579
1467− if (context->stream_count > dc->optimize_seamless_boot_streams ||
1580+ if (context->stream_count > get_seamless_boot_stream_count(context) ||
14681581 context->stream_count == 0) {
14691582 /* Must wait for no flips to be pending before doing optimize bw */
14701583 wait_for_no_pipes_pending(dc, context);
@@ -1578,7 +1691,7 @@ void dc_post_update_surfaces_to_stream(struct dc *dc) · 107 unmodified lines
15781691 int i;
15791692 struct dc_state *context = dc->current_state;
15801693
1581− if ((!dc->optimized_required) || dc->optimize_seamless_boot_streams > 0)
1694+ if ((!dc->optimized_required) || get_seamless_boot_stream_count(context) > 0)
15821695 return;
15831696
15841697 post_surface_trace(dc);
@@ -2277,8 +2390,6 @@ static void copy_stream_update_to_stream(struct dc *dc, · 692 unmodified lines
22772390 if (update->dither_option)
22782391 stream->dither_option = *update->dither_option;
22792392
2280− if (update->pending_test_pattern)
2281− stream->test_pattern = *update->pending_test_pattern;
22822393 /* update current stream with writeback info */
22832394 if (update->wb_update) {
22842395 int i;
@@ -2324,7 +2435,6 @@ static void commit_planes_do_stream_update(struct dc *dc, · 39 unmodified lines
23242435 struct dc_state *context)
23252436 {
23262437 int j;
2327− bool should_program_abm;
23282438
23292439 // Stream updates
23302440 for (j = 0; j < dc->res_pool->pipe_count; j++) {
@@ -2375,15 +2485,6 @@ static void commit_planes_do_stream_update(struct dc *dc, · 44 unmodified lines
23752485 }
23762486 }
23772487
2378− if (stream_update->pending_test_pattern) {
2379− dc_link_dp_set_test_pattern(stream->link,
2380− stream->test_pattern.type,
2381− stream->test_pattern.color_space,
2382− stream->test_pattern.p_link_settings,
2383− stream->test_pattern.p_custom_pattern,
2384− stream->test_pattern.cust_pattern_size);
2385− }
2386−
23872488 /* Full fe update*/
23882489 if (update_type == UPDATE_TYPE_FAST)
23892490 continue;
@@ -2398,9 +2499,10 @@ static void commit_planes_do_stream_update(struct dc *dc, · 8 unmodified lines
23982499 if (pipe_ctx->stream_res.audio && !dc->debug.az_endpoint_mute_only)
23992500 pipe_ctx->stream_res.audio->funcs->az_disable(pipe_ctx->stream_res.audio);
24002501
2401− dc->hwss.optimize_bandwidth(dc, dc->current_state);
2502+ dc->optimized_required = true;
2503+
24022504 } else {
2403− if (dc->optimize_seamless_boot_streams == 0)
2505+ if (get_seamless_boot_stream_count(context) == 0)
24042506 dc->hwss.prepare_bandwidth(dc, dc->current_state);
24052507
24062508 core_link_enable_stream(dc->current_state, pipe_ctx);
@@ -2408,7 +2510,7 @@ static void commit_planes_do_stream_update(struct dc *dc, · 1 unmodified line
24082510 }
24092511
24102512 if (stream_update->abm_level && pipe_ctx->stream_res.abm) {
2411− should_program_abm = true;
2513+ bool should_program_abm = true;
24122514
24132515 // if otg funcs defined check if blanked before programming
24142516 if (pipe_ctx->stream_res.tg->funcs->is_blanked)
@@ -2439,7 +2541,7 @@ static void commit_planes_for_stream(struct dc *dc, · 24 unmodified lines
24392541 int i, j;
24402542 struct pipe_ctx *top_pipe_to_program = NULL;
24412543
2442− if (dc->optimize_seamless_boot_streams > 0 && surface_count > 0) {
2544+ if (get_seamless_boot_stream_count(context) > 0 && surface_count > 0) {
24432545 /* Optimize seamless boot flag keeps clocks and watermarks high until
24442546 * first flip. After first flip, optimization is required to lower
24452547 * bandwidth. Important to note that it is expected UEFI will
@@ -2448,9 +2550,8 @@ static void commit_planes_for_stream(struct dc *dc, · 2 unmodified lines
24482550 */
24492551 if (stream->apply_seamless_boot_optimization) {
24502552 stream->apply_seamless_boot_optimization = false;
2451− dc->optimize_seamless_boot_streams--;
24522553
2453− if (dc->optimize_seamless_boot_streams == 0)
2554+ if (get_seamless_boot_stream_count(context) == 0)
24542555 dc->optimized_required = true;
24552556 }
24562557 }
@@ -2460,7 +2561,7 @@ static void commit_planes_for_stream(struct dc *dc, · 3 unmodified lines
24602561 dc_allow_idle_optimizations(dc, false);
24612562
24622563 #endif
2463− if (dc->optimize_seamless_boot_streams == 0)
2564+ if (get_seamless_boot_stream_count(context) == 0)
24642565 dc->hwss.prepare_bandwidth(dc, context);
24652566
24662567 context_clock_trace(dc, context);
@@ -2868,6 +2969,9 @@ void dc_set_power_state( · 401 unmodified lines
28682969 struct kref refcount;
28692970 struct display_mode_lib *dml;
28702971
2972+ if (!dc->current_state)
2973+ return;
2974+
28712975 switch (power_state) {
28722976 case DC_ACPI_CM_POWER_STATE_D0:
28732977 dc_resource_state_construct(dc, dc->current_state);
@@ -3176,3 +3280,113 @@ void dc_hardware_release(struct dc *dc) · 302 unmodified lines
31763280 dc->hwss.hardware_release(dc);
31773281 }
31783282 #endif
3283+
3284+/**
3285+ *****************************************************************************
3286+ * Function: dc_enable_dmub_notifications
3287+ *
3288+ * @brief
3289+ * Returns whether dmub notification can be enabled
3290+ *
3291+ * @param
3292+ * [in] dc: dc structure
3293+ *
3294+ * @return
3295+ * True to enable dmub notifications, False otherwise
3296+ *****************************************************************************
3297+ */
3298+bool dc_enable_dmub_notifications(struct dc *dc)
3299+{
3300+ /* dmub aux needs dmub notifications to be enabled */
3301+ return dc->debug.enable_dmub_aux_for_legacy_ddc;
3302+}
3303+
3304+/**
3305+ *****************************************************************************
3306+ * Function: dc_process_dmub_aux_transfer_async
3307+ *
3308+ * @brief
3309+ * Submits aux command to dmub via inbox message
3310+ * Sets port index appropriately for legacy DDC
3311+ *
3312+ * @param
3313+ * [in] dc: dc structure
3314+ * [in] link_index: link index
3315+ * [in] payload: aux payload
3316+ *
3317+ * @return
3318+ * True if successful, False if failure
3319+ *****************************************************************************
3320+ */
3321+bool dc_process_dmub_aux_transfer_async(struct dc *dc,
3322+ uint32_t link_index,
3323+ struct aux_payload *payload)
3324+{
3325+ uint8_t action;
3326+ union dmub_rb_cmd cmd = {0};
3327+ struct dc_dmub_srv *dmub_srv = dc->ctx->dmub_srv;
3328+
3329+ ASSERT(payload->length <= 16);
3330+
3331+ cmd.dp_aux_access.header.type = DMUB_CMD__DP_AUX_ACCESS;
3332+ cmd.dp_aux_access.header.payload_bytes = 0;
3333+ cmd.dp_aux_access.aux_control.type = AUX_CHANNEL_LEGACY_DDC;
3334+ cmd.dp_aux_access.aux_control.instance = dc->links[link_index]->ddc_hw_inst;
3335+ cmd.dp_aux_access.aux_control.sw_crc_enabled = 0;
3336+ cmd.dp_aux_access.aux_control.timeout = 0;
3337+ cmd.dp_aux_access.aux_control.dpaux.address = payload->address;
3338+ cmd.dp_aux_access.aux_control.dpaux.is_i2c_over_aux = payload->i2c_over_aux;
3339+ cmd.dp_aux_access.aux_control.dpaux.length = payload->length;
3340+
3341+ /* set aux action */
3342+ if (payload->i2c_over_aux) {
3343+ if (payload->write) {
3344+ if (payload->mot)
3345+ action = DP_AUX_REQ_ACTION_I2C_WRITE_MOT;
3346+ else
3347+ action = DP_AUX_REQ_ACTION_I2C_WRITE;
3348+ } else {
3349+ if (payload->mot)
3350+ action = DP_AUX_REQ_ACTION_I2C_READ_MOT;
3351+ else
3352+ action = DP_AUX_REQ_ACTION_I2C_READ;
3353+ }
3354+ } else {
3355+ if (payload->write)
3356+ action = DP_AUX_REQ_ACTION_DPCD_WRITE;
3357+ else
3358+ action = DP_AUX_REQ_ACTION_DPCD_READ;
3359+ }
3360+
3361+ cmd.dp_aux_access.aux_control.dpaux.action = action;
3362+
3363+ if (payload->length && payload->write) {
3364+ memcpy(cmd.dp_aux_access.aux_control.dpaux.data,
3365+ payload->data,
3366+ payload->length
3367+ );
3368+ }
3369+
3370+ dc_dmub_srv_cmd_queue(dmub_srv, &cmd);
3371+ dc_dmub_srv_cmd_execute(dmub_srv);
3372+ dc_dmub_srv_wait_idle(dmub_srv);
3373+
3374+ return true;
3375+}
3376+
3377+/**
3378+ *****************************************************************************
3379+ * Function: dc_disable_accelerated_mode
3380+ *
3381+ * @brief
3382+ * disable accelerated mode
3383+ *
3384+ * @param
3385+ * [in] dc: dc structure
3386+ *
3387+ *****************************************************************************
3388+ */
3389+void dc_disable_accelerated_mode(struct dc *dc)
3390+{
3391+ bios_set_scratch_acc_mode_change(dc->ctx->dc_bios, 0);
3392+}
drivers/gpu/drm/amd/display/dc/core/dc_link.c+28−8
@@ -91,8 +91,14 @@ static void dc_link_destruct(struct dc_link *link) · 90 unmodified lines
9191 if (link->panel_cntl)
9292 link->panel_cntl->funcs->destroy(&link->panel_cntl);
9393
94− if (link->link_enc)
94+ if (link->link_enc) {
95+ /* Update link encoder tracking variables. These are used for the dynamic
96+ * assignment of link encoders to streams.
97+ */
98+ link->dc->res_pool->link_encoders[link->link_enc->preferred_engine] = NULL;
99+ link->dc->res_pool->dig_link_enc_count--;
95100 link->link_enc->funcs->destroy(&link->link_enc);
101+ }
96102
97103 if (link->local_sink)
98104 dc_sink_release(link->local_sink);
@@ -1500,7 +1506,7 @@ static bool dc_link_construct(struct dc_link *link, · 1401 unmodified lines
15001506 (link->link_id.id == CONNECTOR_ID_EDP ||
15011507 link->link_id.id == CONNECTOR_ID_LVDS)) {
15021508 panel_cntl_init_data.ctx = dc_ctx;
1503− panel_cntl_init_data.inst = 0;
1509+ panel_cntl_init_data.inst = link->link_index;
15041510 link->panel_cntl =
15051511 link->dc->res_pool->funcs->panel_cntl_create(
15061512 &panel_cntl_init_data);
@@ -1532,6 +1538,12 @@ static bool dc_link_construct(struct dc_link *link, · 25 unmodified lines
15321538
15331539 DC_LOG_DC("BIOS object table - DP_IS_USB_C: %d", link->link_enc->features.flags.bits.DP_IS_USB_C);
15341540
1541+ /* Update link encoder tracking variables. These are used for the dynamic
1542+ * assignment of link encoders to streams.
1543+ */
1544+ link->dc->res_pool->link_encoders[link->link_enc->preferred_engine] = link->link_enc;
1545+ link->dc->res_pool->dig_link_enc_count++;
1546+
15351547 link->link_enc_hw_inst = link->link_enc->transmitter;
15361548
15371549 for (i = 0; i < 4; i++) {
@@ -1603,6 +1615,7 @@ static bool dc_link_construct(struct dc_link *link, · 65 unmodified lines
16031615 link->psr_settings.psr_version = DC_PSR_VERSION_UNSUPPORTED;
16041616
16051617 DC_LOG_DC("BIOS object table - %s finished successfully.\n", __func__);
1618+ kfree(info);
16061619 return true;
16071620 device_tag_fail:
16081621 link->link_enc->funcs->destroy(&link->link_enc);
@@ -3257,6 +3270,16 @@ void core_link_enable_stream( · 1648 unmodified lines
32573270 /* Do not touch link on seamless boot optimization. */
32583271 if (pipe_ctx->stream->apply_seamless_boot_optimization) {
32593272 pipe_ctx->stream->dpms_off = false;
3273+
3274+ /* Still enable stream features & audio on seamless boot for DP external displays */
3275+ if (pipe_ctx->stream->signal == SIGNAL_TYPE_DISPLAY_PORT) {
3276+ enable_stream_features(pipe_ctx);
3277+ if (pipe_ctx->stream_res.audio != NULL) {
3278+ pipe_ctx->stream_res.stream_enc->funcs->dp_audio_enable(pipe_ctx->stream_res.stream_enc);
3279+ dc->hwss.enable_audio_stream(pipe_ctx);
3280+ }
3281+ }
3282+
32603283 #if defined(CONFIG_DRM_AMD_DC_HDCP)
32613284 update_psp_stream_config(pipe_ctx, false);
32623285 #endif
@@ -3480,15 +3503,12 @@ uint32_t dc_bandwidth_in_kbps_from_timing( · 217 unmodified lines
34803503 {
34813504 uint32_t bits_per_channel = 0;
34823505 uint32_t kbps;
3483− struct fixed31_32 link_bw_kbps;
34843506
3507+#if defined(CONFIG_DRM_AMD_DC_DCN)
34853508 if (timing->flags.DSC) {
3486− link_bw_kbps = dc_fixpt_from_int(timing->pix_clk_100hz);
3487− link_bw_kbps = dc_fixpt_div_int(link_bw_kbps, 160);
3488− link_bw_kbps = dc_fixpt_mul_int(link_bw_kbps, timing->dsc_cfg.bits_per_pixel);
3489− kbps = dc_fixpt_ceil(link_bw_kbps);
3490− return kbps;
3509+ return dc_dsc_stream_bandwidth_in_kbps(timing->pix_clk_100hz, timing->dsc_cfg.bits_per_pixel);
34913510 }
3511+#endif
34923512
34933513 switch (timing->display_color_depth) {
34943514 case COLOR_DEPTH_666:
drivers/gpu/drm/amd/display/dc/core/dc_link_ddc.c+3−2
@@ -36,6 +36,7 @@ · 35 unmodified lines
3636 #include "core_types.h"
3737 #include "dc_link_ddc.h"
3838 #include "dce/dce_aux.h"
39+#include "dmub/inc/dmub_cmd.h"
3940
4041 #define DC_LOGGER_INIT(logger)
4142
@@ -558,7 +559,7 @@ bool dal_ddc_service_query_ddc_data( · 516 unmodified lines
558559 /* should not set mot (middle of transaction) to 0
559560 * if there are pending read payloads
560561 */
561− payload.mot = read_size == 0 ? false : true;
562+ payload.mot = !(read_size == 0);
562563 payload.length = write_size;
563564 payload.data = write_buf;
564565
@@ -655,7 +656,7 @@ bool dal_ddc_submit_aux_command(struct ddc_service *ddc, · 90 unmodified lines
655656 */
656657 int dc_link_aux_transfer_raw(struct ddc_service *ddc,
657658 struct aux_payload *payload,
658− enum aux_channel_operation_result *operation_result)
659+ enum aux_return_code_type *operation_result)
659660 {
660661 return dce_aux_transfer_raw(ddc, payload, operation_result);
661662 }
drivers/gpu/drm/amd/display/dc/core/dc_link_dp.c+17−1
@@ -1669,6 +1669,22 @@ bool perform_link_training_with_retries( · 1668 unmodified lines
16691669 msleep(delay_dp_power_up_in_ms);
16701670 }
16711671
1672+#ifdef CONFIG_DRM_AMD_DC_HDCP
1673+ if (panel_mode == DP_PANEL_MODE_EDP) {
1674+ struct cp_psp *cp_psp = &stream->ctx->cp_psp;
1675+
1676+ if (cp_psp && cp_psp->funcs.enable_assr) {
1677+ if (!cp_psp->funcs.enable_assr(cp_psp->handle, link)) {
1678+ /* since eDP implies ASSR on, change panel
1679+ * mode to disable ASSR
1680+ */
1681+ panel_mode = DP_PANEL_MODE_DEFAULT;
1682+ }
1683+ } else
1684+ panel_mode = DP_PANEL_MODE_DEFAULT;
1685+ }
1686+#endif
1687+
16721688 dp_set_panel_mode(link, panel_mode);
16731689
16741690 if (link->aux_access_disabled) {
@@ -4265,7 +4281,7 @@ void dp_set_panel_mode(struct dc_link *link, enum dp_panel_mode panel_mode) · 2590 unmodified lines
42654281
42664282 if (edp_config_set.bits.PANEL_MODE_EDP
42674283 != panel_mode_edp) {
4268− enum dc_status result = DC_ERROR_UNEXPECTED;
4284+ enum dc_status result;
42694285
42704286 edp_config_set.bits.PANEL_MODE_EDP =
42714287 panel_mode_edp;
drivers/gpu/drm/amd/display/dc/core/dc_resource.c+43−0
@@ -417,6 +417,49 @@ int resource_get_clock_source_reference( · 416 unmodified lines
417417 return -1;
418418 }
419419
420+bool resource_are_vblanks_synchronizable(
421+ struct dc_stream_state *stream1,
422+ struct dc_stream_state *stream2)
423+{
424+ uint32_t base60_refresh_rates[] = {10, 20, 5};
425+ uint8_t i;
426+ uint8_t rr_count = sizeof(base60_refresh_rates)/sizeof(base60_refresh_rates[0]);
427+ uint64_t frame_time_diff;
428+
429+ if (stream1->ctx->dc->config.vblank_alignment_dto_params &&
430+ stream1->ctx->dc->config.vblank_alignment_max_frame_time_diff > 0 &&
431+ dc_is_dp_signal(stream1->signal) &&
432+ dc_is_dp_signal(stream2->signal) &&
433+ false == stream1->has_non_synchronizable_pclk &&
434+ false == stream2->has_non_synchronizable_pclk &&
435+ stream1->timing.flags.VBLANK_SYNCHRONIZABLE &&
436+ stream2->timing.flags.VBLANK_SYNCHRONIZABLE) {
437+ /* disable refresh rates higher than 60Hz for now */
438+ if (stream1->timing.pix_clk_100hz*100/stream1->timing.h_total/
439+ stream1->timing.v_total > 60)
440+ return false;
441+ if (stream2->timing.pix_clk_100hz*100/stream2->timing.h_total/
442+ stream2->timing.v_total > 60)
443+ return false;
444+ frame_time_diff = (uint64_t)10000 *
445+ stream1->timing.h_total *
446+ stream1->timing.v_total *
447+ stream2->timing.pix_clk_100hz;
448+ frame_time_diff = div_u64(frame_time_diff, stream1->timing.pix_clk_100hz);
449+ frame_time_diff = div_u64(frame_time_diff, stream2->timing.h_total);
450+ frame_time_diff = div_u64(frame_time_diff, stream2->timing.v_total);
451+ for (i = 0; i < rr_count; i++) {
452+ int64_t diff = (int64_t)div_u64(frame_time_diff * base60_refresh_rates[i], 10) - 10000;
453+
454+ if (diff < 0)
455+ diff = -diff;
456+ if (diff < stream1->ctx->dc->config.vblank_alignment_max_frame_time_diff)
457+ return true;
458+ }
459+ }
460+ return false;
461+}
462+
420463 bool resource_are_streams_timing_synchronizable(
421464 struct dc_stream_state *stream1,
422465 struct dc_stream_state *stream2)
drivers/gpu/drm/amd/display/dc/core/dc_stat.c+64−0
@@ -0,0 +1,64 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ * Authors: AMD
23+ */
24+
25+#include "dc/dc_stat.h"
26+#include "dmub/dmub_srv_stat.h"
27+#include "dc_dmub_srv.h"
28+
29+/**
30+ * DOC: DC STAT Interface
31+ *
32+ * These interfaces are called without acquiring DAL and DC locks.
33+ * Hence, there is limitations on whese interfaces can access. Only
34+ * variables exclusively defined for these interfaces can be modified.
35+ */
36+
37+/**
38+ *****************************************************************************
39+ * Function: dc_stat_get_dmub_notification
40+ *
41+ * @brief
42+ * Calls dmub layer to retrieve dmub notification
43+ *
44+ * @param
45+ * [in] dc: dc structure
46+ * [in] notify: dmub notification structure
47+ *
48+ * @return
49+ * None
50+ *****************************************************************************
51+ */
52+void dc_stat_get_dmub_notification(const struct dc *dc, struct dmub_notification *notify)
53+{
54+ /**
55+ * This function is called without dal and dc locks, so
56+ * we shall not modify any dc, dc_dmub_srv or dmub variables
57+ * except variables exclusively accessed by this function
58+ */
59+ struct dmub_srv *dmub = dc->ctx->dmub_srv->dmub;
60+ enum dmub_status status;
61+
62+ status = dmub_srv_stat_get_notification(dmub, notify);
63+ ASSERT(status == DMUB_STATUS_OK);
64+}
drivers/gpu/drm/amd/display/dc/dc.h+25−2
@@ -42,13 +42,17 @@ · 41 unmodified lines
4242 #include "inc/hw/dmcu.h"
4343 #include "dml/display_mode_lib.h"
4444
45−#define DC_VER "3.2.122"
45+/* forward declaration */
46+struct aux_payload;
47+
48+#define DC_VER "3.2.127"
4649
4750 #define MAX_SURFACES 3
4851 #define MAX_PLANES 6
4952 #define MAX_STREAMS 6
5053 #define MAX_SINKS_PER_LINK 4
5154 #define MIN_VIEWPORT_SIZE 12
55+#define MAX_NUM_EDP 2
5256
5357 /*******************************************************************************
5458 * Display Core Interfaces
@@ -151,6 +155,8 @@ struct dc_caps { · 96 unmodified lines
151155 uint32_t max_links;
152156 uint32_t max_audios;
153157 uint32_t max_slave_planes;
158+ uint32_t max_slave_yuv_planes;
159+ uint32_t max_slave_rgb_planes;
154160 uint32_t max_planes;
155161 uint32_t max_downscale_ratio;
156162 uint32_t i2c_speed_in_khz;
@@ -301,6 +307,8 @@ struct dc_config { · 144 unmodified lines
301307 #if defined(CONFIG_DRM_AMD_DC_DCN)
302308 bool clamp_min_dcfclk;
303309 #endif
310+ uint64_t vblank_alignment_dto_params;
311+ uint8_t vblank_alignment_max_frame_time_diff;
304312 };
305313
306314 enum visual_confirm {
@@ -528,6 +536,10 @@ struct dc_debug_options { · 221 unmodified lines
528536 bool disable_dsc;
529537 bool enable_dram_clock_change_one_display_vactive;
530538 union mem_low_power_enable_options enable_mem_low_power;
539+ bool force_vblank_alignment;
540+
541+ /* Enable dmub aux for legacy ddc */
542+ bool enable_dmub_aux_for_legacy_ddc;
531543 };
532544
533545 struct dc_debug_data {
@@ -628,7 +640,6 @@ struct dc { · 94 unmodified lines
628640 #endif
629641
630642 /* Require to maintain clocks and bandwidth for UEFI enabled HW */
631− int optimize_seamless_boot_streams;
632643
633644 /* FBC compressor */
634645 struct compressor *fbc_compressor;
@@ -1292,8 +1303,20 @@ void dc_hardware_release(struct dc *dc); · 657 unmodified lines
12921303
12931304 bool dc_set_psr_allow_active(struct dc *dc, bool enable);
12941305
1306+bool dc_enable_dmub_notifications(struct dc *dc);
1307+
1308+bool dc_process_dmub_aux_transfer_async(struct dc *dc,
1309+ uint32_t link_index,
1310+ struct aux_payload *payload);
1311+
12951312 /*******************************************************************************
12961313 * DSC Interfaces
12971314 ******************************************************************************/
12981315 #include "dc_dsc.h"
1316+
1317+/*******************************************************************************
1318+ * Disable acc mode Interfaces
1319+ ******************************************************************************/
1320+void dc_disable_accelerated_mode(struct dc *dc);
1321+
12991322 #endif /* DC_INTERFACE_H_ */
drivers/gpu/drm/amd/display/dc/dc_bios_types.h+2−1
@@ -139,7 +139,8 @@ struct dc_vbios_funcs { · 138 unmodified lines
139139
140140 enum bp_result (*enable_lvtma_control)(
141141 struct dc_bios *bios,
142− uint8_t uc_pwr_on);
142+ uint8_t uc_pwr_on,
143+ uint8_t panel_instance);
143144
144145 enum bp_result (*get_soc_bb_info)(
145146 struct dc_bios *dcb,
drivers/gpu/drm/amd/display/dc/dc_ddc_types.h+0−10
@@ -44,16 +44,6 @@ enum i2caux_transaction_action { · 43 unmodified lines
4444 I2CAUX_TRANSACTION_ACTION_DP_READ = 0x90
4545 };
4646
47−enum aux_channel_operation_result {
48− AUX_CHANNEL_OPERATION_SUCCEEDED,
49− AUX_CHANNEL_OPERATION_FAILED_REASON_UNKNOWN,
50− AUX_CHANNEL_OPERATION_FAILED_INVALID_REPLY,
51− AUX_CHANNEL_OPERATION_FAILED_TIMEOUT,
52− AUX_CHANNEL_OPERATION_FAILED_HPD_DISCON,
53− AUX_CHANNEL_OPERATION_FAILED_ENGINE_ACQUIRE
54−};
55−
56−
5747 struct aux_request_transaction_data {
5848 enum aux_transaction_type type;
5949 enum i2caux_transaction_action action;
drivers/gpu/drm/amd/display/dc/dc_dmub_srv.c+34−0
@@ -26,6 +26,10 @@ · 25 unmodified lines
2626 #include "dc.h"
2727 #include "dc_dmub_srv.h"
2828 #include "../dmub/dmub_srv.h"
29+#include "dm_helpers.h"
30+
31+#define CTX dc_dmub_srv->ctx
32+#define DC_LOGGER CTX->logger
2933
3034 static void dc_dmub_srv_construct(struct dc_dmub_srv *dc_srv, struct dc *dc,
3135 struct dmub_srv *dmub)
@@ -106,6 +110,25 @@ void dc_dmub_srv_wait_idle(struct dc_dmub_srv *dc_dmub_srv) · 74 unmodified lines
106110 DC_ERROR("Error waiting for DMUB idle: status=%d\n", status);
107111 }
108112
113+bool dc_dmub_srv_cmd_with_reply_data(struct dc_dmub_srv *dc_dmub_srv, union dmub_rb_cmd *cmd)
114+{
115+ struct dmub_srv *dmub;
116+ enum dmub_status status;
117+
118+ if (!dc_dmub_srv || !dc_dmub_srv->dmub)
119+ return false;
120+
121+ dmub = dc_dmub_srv->dmub;
122+
123+ status = dmub_srv_cmd_with_reply_data(dmub, cmd);
124+ if (status != DMUB_STATUS_OK) {
125+ DC_LOG_DEBUG("No reply for DMUB command: status=%d\n", status);
126+ return false;
127+ }
128+
129+ return true;
130+}
131+
109132 void dc_dmub_srv_wait_phy_init(struct dc_dmub_srv *dc_dmub_srv)
110133 {
111134 struct dmub_srv *dmub = dc_dmub_srv->dmub;
@@ -148,3 +171,14 @@ bool dc_dmub_srv_notify_stream_mask(struct dc_dmub_srv *dc_dmub_srv, · 36 unmodified lines
148171 dmub, DMUB_GPINT__IDLE_OPT_NOTIFY_STREAM_MASK,
149172 stream_mask, timeout) == DMUB_STATUS_OK;
150173 }
174+
175+bool dc_dmub_srv_get_dmub_outbox0_msg(const struct dc *dc, struct dmcub_trace_buf_entry *entry)
176+{
177+ struct dmub_srv *dmub = dc->ctx->dmub_srv->dmub;
178+ return dmub_srv_get_outbox0_msg(dmub, entry);
179+}
180+
181+void dc_dmub_trace_event_control(struct dc *dc, bool enable)
182+{
183+ dm_helpers_dmub_outbox0_interrupt_control(dc->ctx, enable);
184+}
drivers/gpu/drm/amd/display/dc/dc_dmub_srv.h+8−0
@@ -30,6 +30,7 @@ · 29 unmodified lines
3030 #include "dmub/dmub_srv.h"
3131
3232 struct dmub_srv;
33+struct dc;
3334
3435 struct dc_reg_helper_state {
3536 bool gather_in_progress;
@@ -56,6 +57,13 @@ void dc_dmub_srv_wait_idle(struct dc_dmub_srv *dc_dmub_srv); · 20 unmodified lines
5657
5758 void dc_dmub_srv_wait_phy_init(struct dc_dmub_srv *dc_dmub_srv);
5859
60+bool dc_dmub_srv_cmd_with_reply_data(struct dc_dmub_srv *dc_dmub_srv, union dmub_rb_cmd *cmd);
61+
5962 bool dc_dmub_srv_notify_stream_mask(struct dc_dmub_srv *dc_dmub_srv,
6063 unsigned int stream_mask);
64+
65+bool dc_dmub_srv_get_dmub_outbox0_msg(const struct dc *dc, struct dmcub_trace_buf_entry *entry);
66+
67+void dc_dmub_trace_event_control(struct dc *dc, bool enable);
68+
6169 #endif /* _DMUB_DC_SRV_H_ */
drivers/gpu/drm/amd/display/dc/dc_dsc.h+8−3
@@ -51,6 +51,7 @@ struct dc_dsc_policy { · 50 unmodified lines
5151 int min_slice_height; // Must not be less than 8
5252 uint32_t max_target_bpp;
5353 uint32_t min_target_bpp;
54+ uint32_t preferred_bpp_x16;
5455 bool enable_dsc_when_not_needed;
5556 };
5657
@@ -62,8 +63,8 @@ bool dc_dsc_parse_dsc_dpcd(const struct dc *dc, · 5 unmodified lines
6263 bool dc_dsc_compute_bandwidth_range(
6364 const struct display_stream_compressor *dsc,
6465 uint32_t dsc_min_slice_height_override,
65− uint32_t min_bpp,
66− uint32_t max_bpp,
66+ uint32_t min_bpp_x16,
67+ uint32_t max_bpp_x16,
6768 const struct dsc_dec_dpcd_caps *dsc_sink_caps,
6869 const struct dc_crtc_timing *timing,
6970 struct dc_dsc_bw_range *range);
@@ -77,12 +78,16 @@ bool dc_dsc_compute_config( · 7 unmodified lines
7778 const struct dc_crtc_timing *timing,
7879 struct dc_dsc_config *dsc_cfg);
7980
81+uint32_t dc_dsc_stream_bandwidth_in_kbps(uint32_t pix_clk_100hz, uint32_t bpp_x16);
82+
8083 void dc_dsc_get_policy_for_timing(const struct dc_crtc_timing *timing,
81− uint32_t max_target_bpp_limit_override,
84+ uint32_t max_target_bpp_limit_override_x16,
8285 struct dc_dsc_policy *policy);
8386
8487 void dc_dsc_policy_set_max_target_bpp_limit(uint32_t limit);
8588
8689 void dc_dsc_policy_set_enable_dsc_when_not_needed(bool enable);
8790
91+uint32_t dc_dsc_stream_bandwidth_in_kbps(uint32_t pix_clk_100hz, uint32_t bpp_x16);
92+
8893 #endif
drivers/gpu/drm/amd/display/dc/dc_hw_types.h+2−0
@@ -705,6 +705,7 @@ struct dc_crtc_timing_flags { · 704 unmodified lines
705705 #ifndef TRIM_FSFT
706706 uint32_t FAST_TRANSPORT: 1;
707707 #endif
708+ uint32_t VBLANK_SYNCHRONIZABLE: 1;
708709 };
709710
710711 enum dc_timing_3d_format {
@@ -769,6 +770,7 @@ struct dc_crtc_timing { · 58 unmodified lines
769770 #endif
770771
771772 struct dc_crtc_timing_flags flags;
773+ uint32_t dsc_fixed_bits_per_pixel_x16; /* DSC target bitrate in 1/16 of bpp (e.g. 128 -> 8bpp) */
772774 struct dc_dsc_config dsc_cfg;
773775 };
774776
drivers/gpu/drm/amd/display/dc/dc_link.h+14−5
@@ -103,6 +103,10 @@ struct dc_link { · 102 unmodified lines
103103 bool lttpr_non_transparent_mode;
104104 bool is_internal_display;
105105
106+ /* TODO: Rename. Flag an endpoint as having a programmable mapping to a
107+ * DIG encoder. */
108+ bool is_dig_mapping_flexible;
109+
106110 bool edp_sink_present;
107111
108112 /* caps is the same as reported_link_cap. link_traing use
@@ -183,16 +187,21 @@ static inline struct dc_link *dc_get_link_at_index(struct dc *dc, uint32_t link_ · 74 unmodified lines
183187 return dc->links[link_index];
184188 }
185189
186−static inline struct dc_link *get_edp_link(const struct dc *dc)
190+static inline void get_edp_links(const struct dc *dc,
191+ struct dc_link **edp_links,
192+ int *edp_num)
187193 {
188194 int i;
189195
190− // report any eDP links, even unconnected DDI's
196+ *edp_num = 0;
191197 for (i = 0; i < dc->link_count; i++) {
192− if (dc->links[i]->connector_signal == SIGNAL_TYPE_EDP)
193− return dc->links[i];
198+ // report any eDP links, even unconnected DDI's
199+ if (dc->links[i]->connector_signal == SIGNAL_TYPE_EDP) {
200+ edp_links[*edp_num] = dc->links[i];
201+ if (++(*edp_num) == MAX_NUM_EDP)
202+ return;
203+ }
194204 }
195− return NULL;
196205 }
197206
198207 /* Set backlight level of an embedded panel (eDP, LVDS).
drivers/gpu/drm/amd/display/dc/dc_stat.h+42−0
@@ -0,0 +1,42 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ * Authors: AMD
23+ *
24+ */
25+
26+#ifndef _DC_STAT_H_
27+#define _DC_STAT_H_
28+
29+/**
30+ * DOC: DC STAT Interface
31+ *
32+ * These interfaces are called without acquiring DAL and DC locks.
33+ * Hence, there is limitations on whese interfaces can access. Only
34+ * variables exclusively defined for these interfaces can be modified.
35+ */
36+
37+#include "dc.h"
38+#include "dmub/dmub_srv.h"
39+
40+void dc_stat_get_dmub_notification(const struct dc *dc, struct dmub_notification *notify);
41+
42+#endif /* _DC_STAT_H_ */
drivers/gpu/drm/amd/display/dc/dc_stream.h+10−11
@@ -130,14 +130,6 @@ union stream_update_flags { · 129 unmodified lines
130130 uint32_t raw;
131131 };
132132
133−struct test_pattern {
134− enum dp_test_pattern type;
135− enum dp_test_pattern_color_space color_space;
136− struct link_training_settings const *p_link_settings;
137− unsigned char const *p_custom_pattern;
138− unsigned int cust_pattern_size;
139−};
140−
141133 struct dc_stream_state {
142134 // sink is deprecated, new code should not reference
143135 // this pointer
@@ -235,9 +227,10 @@ struct dc_stream_state { · 91 unmodified lines
235227
236228 uint32_t stream_id;
237229 bool is_dsc_enabled;
238−
239− struct test_pattern test_pattern;
240230 union stream_update_flags update_flags;
231+
232+ bool has_non_synchronizable_pclk;
233+ bool vblank_synchronized;
241234 };
242235
243236 #define ABM_LEVEL_IMMEDIATE_DISABLE 255
@@ -271,7 +264,6 @@ struct dc_stream_update { · 27 unmodified lines
271264 struct dc_dsc_config *dsc_config;
272265 struct dc_transfer_func *func_shaper;
273266 struct dc_3dlut *lut3d_func;
274− struct test_pattern *pending_test_pattern;
275267 };
276268
277269 bool dc_is_stream_unchanged(
@@ -461,6 +453,13 @@ bool dc_stream_get_crtc_position(struct dc *dc, · 183 unmodified lines
461453 unsigned int *v_pos,
462454 unsigned int *nom_v_pos);
463455
456+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
457+bool dc_stream_forward_dmcu_crc_window(struct dc *dc, struct dc_stream_state *stream,
458+ struct crc_params *crc_window);
459+bool dc_stream_stop_dmcu_crc_win_update(struct dc *dc,
460+ struct dc_stream_state *stream);
461+#endif
462+
464463 bool dc_stream_configure_crc(struct dc *dc,
465464 struct dc_stream_state *stream,
466465 struct crc_params *crc_window,
drivers/gpu/drm/amd/display/dc/dce/Makefile+1−1
@@ -30,7 +30,7 @@ DCE = dce_audio.o dce_stream_encoder.o dce_link_encoder.o dce_hwseq.o \ · 29 unmodified lines
3030 dce_mem_input.o dce_clock_source.o dce_scl_filters.o dce_transform.o \
3131 dce_opp.o dce_dmcu.o dce_abm.o dce_ipp.o dce_aux.o \
3232 dce_i2c.o dce_i2c_hw.o dce_i2c_sw.o dmub_psr.o dmub_abm.o dce_panel_cntl.o \
33−dmub_hw_lock_mgr.o
33+dmub_hw_lock_mgr.o dmub_outbox.o
3434
3535 AMD_DAL_DCE = $(addprefix $(AMDDALPATH)/dc/dce/,$(DCE))
3636
drivers/gpu/drm/amd/display/dc/dce/dce_aux.c+22−18
@@ -31,6 +31,8 @@ · 30 unmodified lines
3131 #include "dce_aux.h"
3232 #include "dce/dce_11_0_sh_mask.h"
3333 #include "dm_event_log.h"
34+#include "dm_helpers.h"
35+#include "dmub/inc/dmub_cmd.h"
3436
3537 #define CTX \
3638 aux110->base.ctx
@@ -324,7 +326,7 @@ static int read_channel_reply(struct dce_aux *engine, uint32_t size, · 287 unmodified lines
324326 return 0;
325327 }
326328
327−static enum aux_channel_operation_result get_channel_status(
329+static enum aux_return_code_type get_channel_status(
328330 struct dce_aux *engine,
329331 uint8_t *returned_bytes)
330332 {
@@ -335,7 +337,7 @@ static enum aux_channel_operation_result get_channel_status( · 4 unmodified lines
335337 if (returned_bytes == NULL) {
336338 /*caller pass NULL pointer*/
337339 ASSERT_CRITICAL(false);
338− return AUX_CHANNEL_OPERATION_FAILED_REASON_UNKNOWN;
340+ return AUX_RET_ERROR_UNKNOWN;
339341 }
340342 *returned_bytes = 0;
341343
@@ -346,7 +348,7 @@ static enum aux_channel_operation_result get_channel_status( · 4 unmodified lines
346348 value = REG_READ(AUX_SW_STATUS);
347349 /* in case HPD is LOW, exit AUX transaction */
348350 if ((value & AUX_SW_STATUS__AUX_SW_HPD_DISCON_MASK))
349− return AUX_CHANNEL_OPERATION_FAILED_HPD_DISCON;
351+ return AUX_RET_ERROR_HPD_DISCON;
350352
351353 /* Note that the following bits are set in 'status.bits'
352354 * during CTS 4.2.1.2 (FW 3.3.1):
@@ -359,14 +361,14 @@ static enum aux_channel_operation_result get_channel_status( · 6 unmodified lines
359361 if (value & AUX_SW_STATUS__AUX_SW_DONE_MASK) {
360362 if ((value & AUX_SW_STATUS__AUX_SW_RX_TIMEOUT_STATE_MASK) ||
361363 (value & AUX_SW_STATUS__AUX_SW_RX_TIMEOUT_MASK))
362− return AUX_CHANNEL_OPERATION_FAILED_TIMEOUT;
364+ return AUX_RET_ERROR_TIMEOUT;
363365
364366 else if ((value & AUX_SW_STATUS__AUX_SW_RX_INVALID_STOP_MASK) ||
365367 (value & AUX_SW_STATUS__AUX_SW_RX_RECV_NO_DET_MASK) ||
366368 (value &
367369 AUX_SW_STATUS__AUX_SW_RX_RECV_INVALID_H_MASK) ||
368370 (value & AUX_SW_STATUS__AUX_SW_RX_RECV_INVALID_L_MASK))
369− return AUX_CHANNEL_OPERATION_FAILED_INVALID_REPLY;
371+ return AUX_RET_ERROR_INVALID_REPLY;
370372
371373 *returned_bytes = get_reg_field_value(value,
372374 AUX_SW_STATUS,
@@ -374,17 +376,17 @@ static enum aux_channel_operation_result get_channel_status( · 1 unmodified line
374376
375377 if (*returned_bytes == 0)
376378 return
377− AUX_CHANNEL_OPERATION_FAILED_INVALID_REPLY;
379+ AUX_RET_ERROR_INVALID_REPLY;
378380 else {
379381 *returned_bytes -= 1;
380− return AUX_CHANNEL_OPERATION_SUCCEEDED;
382+ return AUX_RET_SUCCESS;
381383 }
382384 } else {
383385 /*time_elapsed >= aux_engine->timeout_period
384386 * AUX_SW_STATUS__AUX_SW_HPD_DISCON = at this point
385387 */
386388 ASSERT_CRITICAL(false);
387− return AUX_CHANNEL_OPERATION_FAILED_TIMEOUT;
389+ return AUX_RET_ERROR_TIMEOUT;
388390 }
389391 }
390392
@@ -541,7 +543,7 @@ static enum i2caux_transaction_action i2caux_action_from_payload(struct aux_payl · 150 unmodified lines
541543
542544 int dce_aux_transfer_raw(struct ddc_service *ddc,
543545 struct aux_payload *payload,
544− enum aux_channel_operation_result *operation_result)
546+ enum aux_return_code_type *operation_result)
545547 {
546548 struct ddc *ddc_pin = ddc->ddc_pin;
547549 struct dce_aux *aux_engine;
@@ -556,7 +558,7 @@ int dce_aux_transfer_raw(struct ddc_service *ddc, · 8 unmodified lines
556558
557559 aux_engine = ddc->ctx->dc->res_pool->engines[ddc_pin->pin_data->en];
558560 if (!acquire(aux_engine, ddc_pin)) {
559− *operation_result = AUX_CHANNEL_OPERATION_FAILED_ENGINE_ACQUIRE;
561+ *operation_result = AUX_RET_ERROR_ENGINE_ACQUIRE;
560562 return -1;
561563 }
562564
@@ -575,8 +577,9 @@ int dce_aux_transfer_raw(struct ddc_service *ddc, · 12 unmodified lines
575577 submit_channel_request(aux_engine, &aux_req);
576578 *operation_result = get_channel_status(aux_engine, &returned_bytes);
577579
578− if (*operation_result == AUX_CHANNEL_OPERATION_SUCCEEDED) {
580+ if (*operation_result == AUX_RET_SUCCESS) {
579581 int __maybe_unused bytes_replied = 0;
582+
580583 bytes_replied = read_channel_reply(aux_engine, payload->length,
581584 payload->data, payload->reply,
582585 &status);
@@ -604,7 +607,7 @@ bool dce_aux_transfer_with_retries(struct ddc_service *ddc, · 21 unmodified lines
604607 int i, ret = 0;
605608 uint8_t reply;
606609 bool payload_reply = true;
607− enum aux_channel_operation_result operation_result;
610+ enum aux_return_code_type operation_result;
608611 bool retry_on_defer = false;
609612
610613 int aux_ack_retries = 0,
@@ -620,8 +623,9 @@ bool dce_aux_transfer_with_retries(struct ddc_service *ddc, · 9 unmodified lines
620623
621624 for (i = 0; i < AUX_MAX_RETRIES; i++) {
622625 ret = dce_aux_transfer_raw(ddc, payload, &operation_result);
626+
623627 switch (operation_result) {
624− case AUX_CHANNEL_OPERATION_SUCCEEDED:
628+ case AUX_RET_SUCCESS:
625629 aux_timeout_retries = 0;
626630 aux_invalid_reply_retries = 0;
627631
@@ -667,14 +671,14 @@ bool dce_aux_transfer_with_retries(struct ddc_service *ddc, · 39 unmodified lines
667671 }
668672 break;
669673
670− case AUX_CHANNEL_OPERATION_FAILED_INVALID_REPLY:
674+ case AUX_RET_ERROR_INVALID_REPLY:
671675 if (++aux_invalid_reply_retries >= AUX_MAX_INVALID_REPLY_RETRIES)
672676 goto fail;
673677 else
674678 udelay(400);
675679 break;
676680
677− case AUX_CHANNEL_OPERATION_FAILED_TIMEOUT:
681+ case AUX_RET_ERROR_TIMEOUT:
678682 // Check whether a DEFER had occurred before the timeout.
679683 // If so, treat timeout as a DEFER.
680684 if (retry_on_defer) {
@@ -696,9 +700,9 @@ bool dce_aux_transfer_with_retries(struct ddc_service *ddc, · 15 unmodified lines
696700 }
697701 break;
698702
699− case AUX_CHANNEL_OPERATION_FAILED_HPD_DISCON:
700− case AUX_CHANNEL_OPERATION_FAILED_ENGINE_ACQUIRE:
701− case AUX_CHANNEL_OPERATION_FAILED_REASON_UNKNOWN:
703+ case AUX_RET_ERROR_HPD_DISCON:
704+ case AUX_RET_ERROR_ENGINE_ACQUIRE:
705+ case AUX_RET_ERROR_UNKNOWN:
702706 default:
703707 goto fail;
704708 }
drivers/gpu/drm/amd/display/dc/dce/dce_aux.h+2−2
@@ -29,6 +29,7 @@ · 28 unmodified lines
2929 #include "i2caux_interface.h"
3030 #include "inc/hw/aux_engine.h"
3131
32+enum aux_return_code_type;
3233
3334 #define AUX_COMMON_REG_LIST0(id)\
3435 SRI(AUX_CONTROL, DP_AUX, id), \
@@ -99,7 +100,6 @@ struct dce110_aux_registers { · 64 unmodified lines
99100 AUX_SF(AUX_SW_CONTROL, AUX_SW_GO, mask_sh),\
100101 AUX_SF(AUX_SW_DATA, AUX_SW_AUTOINCREMENT_DISABLE, mask_sh),\
101102 AUX_SF(AUX_SW_DATA, AUX_SW_DATA_RW, mask_sh),\
102− AUX_SF(AUX_SW_DATA, AUX_SW_AUTOINCREMENT_DISABLE, mask_sh),\
103103 AUX_SF(AUX_SW_DATA, AUX_SW_INDEX, mask_sh),\
104104 AUX_SF(AUX_SW_DATA, AUX_SW_DATA, mask_sh),\
105105 AUX_SF(AUX_SW_STATUS, AUX_SW_REPLY_BYTE_COUNT, mask_sh),\
@@ -302,7 +302,7 @@ bool dce110_aux_engine_acquire( · 196 unmodified lines
302302
303303 int dce_aux_transfer_raw(struct ddc_service *ddc,
304304 struct aux_payload *cmd,
305− enum aux_channel_operation_result *operation_result);
305+ enum aux_return_code_type *operation_result);
306306
307307 bool dce_aux_transfer_with_retries(struct ddc_service *ddc,
308308 struct aux_payload *cmd);
drivers/gpu/drm/amd/display/dc/dce/dce_clock_source.c+46−6
@@ -1002,15 +1002,27 @@ static bool get_pixel_clk_frequency_100hz( · 1001 unmodified lines
10021002 {
10031003 struct dce110_clk_src *clk_src = TO_DCE110_CLK_SRC(clock_source);
10041004 unsigned int clock_hz = 0;
1005+ unsigned int modulo_hz = 0;
10051006
10061007 if (clock_source->id == CLOCK_SOURCE_ID_DP_DTO) {
10071008 clock_hz = REG_READ(PHASE[inst]);
10081009
1009− /* NOTE: There is agreement with VBIOS here that MODULO is
1010− * programmed equal to DPREFCLK, in which case PHASE will be
1011− * equivalent to pixel clock.
1012− */
1013− *pixel_clk_khz = clock_hz / 100;
1010+ if (clock_source->ctx->dc->hwss.enable_vblanks_synchronization &&
1011+ clock_source->ctx->dc->config.vblank_alignment_max_frame_time_diff > 0) {
1012+ /* NOTE: In case VBLANK syncronization is enabled, MODULO may
1013+ * not be programmed equal to DPREFCLK
1014+ */
1015+ modulo_hz = REG_READ(MODULO[inst]);
1016+ *pixel_clk_khz = div_u64((uint64_t)clock_hz*
1017+ clock_source->ctx->dc->clk_mgr->dprefclk_khz*10,
1018+ modulo_hz);
1019+ } else {
1020+ /* NOTE: There is agreement with VBIOS here that MODULO is
1021+ * programmed equal to DPREFCLK, in which case PHASE will be
1022+ * equivalent to pixel clock.
1023+ */
1024+ *pixel_clk_khz = clock_hz / 100;
1025+ }
10141026 return true;
10151027 }
10161028
@@ -1074,8 +1086,35 @@ static bool dcn20_program_pix_clk( · 57 unmodified lines
10741086 struct pixel_clk_params *pix_clk_params,
10751087 struct pll_settings *pll_settings)
10761088 {
1089+ struct dce110_clk_src *clk_src = TO_DCE110_CLK_SRC(clock_source);
1090+ unsigned int inst = pix_clk_params->controller_id - CONTROLLER_ID_D0;
1091+
10771092 dce112_program_pix_clk(clock_source, pix_clk_params, pll_settings);
10781093
1094+ if (clock_source->ctx->dc->hwss.enable_vblanks_synchronization &&
1095+ clock_source->ctx->dc->config.vblank_alignment_max_frame_time_diff > 0) {
1096+ /* NOTE: In case VBLANK syncronization is enabled,
1097+ * we need to set modulo to default DPREFCLK first
1098+ * dce112_program_pix_clk does not set default DPREFCLK
1099+ */
1100+ REG_WRITE(MODULO[inst],
1101+ clock_source->ctx->dc->clk_mgr->dprefclk_khz*1000);
1102+ }
1103+ return true;
1104+}
1105+
1106+static bool dcn20_override_dp_pix_clk(
1107+ struct clock_source *clock_source,
1108+ unsigned int inst,
1109+ unsigned int pixel_clk,
1110+ unsigned int ref_clk)
1111+{
1112+ struct dce110_clk_src *clk_src = TO_DCE110_CLK_SRC(clock_source);
1113+
1114+ REG_UPDATE(PIXEL_RATE_CNTL[inst], DP_DTO0_ENABLE, 0);
1115+ REG_WRITE(PHASE[inst], pixel_clk);
1116+ REG_WRITE(MODULO[inst], ref_clk);
1117+ REG_UPDATE(PIXEL_RATE_CNTL[inst], DP_DTO0_ENABLE, 1);
10791118 return true;
10801119 }
10811120
@@ -1083,7 +1122,8 @@ static const struct clock_source_funcs dcn20_clk_src_funcs = { · 1 unmodified line
10831122 .cs_power_down = dce110_clock_source_power_down,
10841123 .program_pix_clk = dcn20_program_pix_clk,
10851124 .get_pix_clk_dividers = dce112_get_pix_clk_dividers,
1086− .get_pixel_clk_frequency_100hz = get_pixel_clk_frequency_100hz
1125+ .get_pixel_clk_frequency_100hz = get_pixel_clk_frequency_100hz,
1126+ .override_dp_pix_clk = dcn20_override_dp_pix_clk
10871127 };
10881128
10891129 #if defined(CONFIG_DRM_AMD_DC_DCN)
drivers/gpu/drm/amd/display/dc/dce/dce_dmcu.c+84−0
@@ -57,6 +57,8 @@ · 56 unmodified lines
5757 #define MCP_SYNC_PHY_LOCK 0x90
5858 #define MCP_SYNC_PHY_UNLOCK 0x91
5959 #define MCP_BL_SET_PWM_FRAC 0x6A /* Enable or disable Fractional PWM */
60+#define CRC_WIN_NOTIFY 0x92
61+#define CRC_STOP_UPDATE 0x93
6062 #define MCP_SEND_EDID_CEA 0xA0
6163 #define EDID_CEA_CMD_ACK 1
6264 #define EDID_CEA_CMD_NACK 2
@@ -930,6 +932,84 @@ static bool dcn10_recv_edid_cea_ack(struct dmcu *dmcu, int *offset) · 867 unmodified lines
930932
931933 #endif //(CONFIG_DRM_AMD_DC_DCN)
932934
935+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
936+static void dcn10_forward_crc_window(struct dmcu *dmcu,
937+ struct crc_region *crc_win,
938+ struct otg_phy_mux *mux_mapping)
939+{
940+ struct dce_dmcu *dmcu_dce = TO_DCE_DMCU(dmcu);
941+ unsigned int dmcu_max_retry_on_wait_reg_ready = 801;
942+ unsigned int dmcu_wait_reg_ready_interval = 100;
943+ unsigned int crc_start = 0, crc_end = 0, otg_phy_mux = 0;
944+
945+ /* If microcontroller is not running, do nothing */
946+ if (dmcu->dmcu_state != DMCU_RUNNING)
947+ return;
948+
949+ if (!crc_win)
950+ return;
951+
952+ /* waitDMCUReadyForCmd */
953+ REG_WAIT(MASTER_COMM_CNTL_REG, MASTER_COMM_INTERRUPT, 0,
954+ dmcu_wait_reg_ready_interval,
955+ dmcu_max_retry_on_wait_reg_ready);
956+
957+ /* build up nitification data */
958+ crc_start = (((unsigned int) crc_win->x_start) << 16) | crc_win->y_start;
959+ crc_end = (((unsigned int) crc_win->x_end) << 16) | crc_win->y_end;
960+ otg_phy_mux =
961+ (((unsigned int) mux_mapping->otg_output_num) << 16) | mux_mapping->phy_output_num;
962+
963+ dm_write_reg(dmcu->ctx, REG(MASTER_COMM_DATA_REG1),
964+ crc_start);
965+
966+ dm_write_reg(dmcu->ctx, REG(MASTER_COMM_DATA_REG2),
967+ crc_end);
968+
969+ dm_write_reg(dmcu->ctx, REG(MASTER_COMM_DATA_REG3),
970+ otg_phy_mux);
971+
972+ /* setDMCUParam_Cmd */
973+ REG_UPDATE(MASTER_COMM_CMD_REG, MASTER_COMM_CMD_REG_BYTE0,
974+ CRC_WIN_NOTIFY);
975+
976+ /* notifyDMCUMsg */
977+ REG_UPDATE(MASTER_COMM_CNTL_REG, MASTER_COMM_INTERRUPT, 1);
978+}
979+
980+static void dcn10_stop_crc_win_update(struct dmcu *dmcu,
981+ struct otg_phy_mux *mux_mapping)
982+{
983+ struct dce_dmcu *dmcu_dce = TO_DCE_DMCU(dmcu);
984+ unsigned int dmcu_max_retry_on_wait_reg_ready = 801;
985+ unsigned int dmcu_wait_reg_ready_interval = 100;
986+ unsigned int otg_phy_mux = 0;
987+
988+ /* If microcontroller is not running, do nothing */
989+ if (dmcu->dmcu_state != DMCU_RUNNING)
990+ return;
991+
992+ /* waitDMCUReadyForCmd */
993+ REG_WAIT(MASTER_COMM_CNTL_REG, MASTER_COMM_INTERRUPT, 0,
994+ dmcu_wait_reg_ready_interval,
995+ dmcu_max_retry_on_wait_reg_ready);
996+
997+ /* build up nitification data */
998+ otg_phy_mux =
999+ (((unsigned int) mux_mapping->otg_output_num) << 16) | mux_mapping->phy_output_num;
1000+
1001+ dm_write_reg(dmcu->ctx, REG(MASTER_COMM_DATA_REG1),
1002+ otg_phy_mux);
1003+
1004+ /* setDMCUParam_Cmd */
1005+ REG_UPDATE(MASTER_COMM_CMD_REG, MASTER_COMM_CMD_REG_BYTE0,
1006+ CRC_STOP_UPDATE);
1007+
1008+ /* notifyDMCUMsg */
1009+ REG_UPDATE(MASTER_COMM_CNTL_REG, MASTER_COMM_INTERRUPT, 1);
1010+}
1011+#endif
1012+
9331013 static const struct dmcu_funcs dce_funcs = {
9341014 .dmcu_init = dce_dmcu_init,
9351015 .load_iram = dce_dmcu_load_iram,
@@ -953,6 +1033,10 @@ static const struct dmcu_funcs dcn10_funcs = { · 17 unmodified lines
9531033 .send_edid_cea = dcn10_send_edid_cea,
9541034 .recv_amd_vsdb = dcn10_recv_amd_vsdb,
9551035 .recv_edid_cea_ack = dcn10_recv_edid_cea_ack,
1036+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
1037+ .forward_crc_window = dcn10_forward_crc_window,
1038+ .stop_crc_win_update = dcn10_stop_crc_win_update,
1039+#endif
9561040 .is_dmcu_initialized = dcn10_is_dmcu_initialized
9571041 };
9581042
drivers/gpu/drm/amd/display/dc/dce/dce_opp.c+1−3
@@ -216,9 +216,7 @@ static void set_spatial_dither( · 215 unmodified lines
216216 REG_UPDATE(FMT_BIT_DEPTH_CONTROL,
217217 FMT_TEMPORAL_DITHER_EN, 0);
218218
219− /* no 10bpc on DCE11*/
220− if (params->flags.SPATIAL_DITHER_ENABLED == 0 ||
221− params->flags.SPATIAL_DITHER_DEPTH == 2)
219+ if (params->flags.SPATIAL_DITHER_ENABLED == 0)
222220 return;
223221
224222 /* only use FRAME_COUNTER_MAX if frameRandom == 1*/
drivers/gpu/drm/amd/display/dc/dce/dmub_abm.c+5−5
@@ -72,11 +72,11 @@ static void dmub_abm_init(struct abm *abm, uint32_t backlight) · 71 unmodified lines
7272 {
7373 struct dce_abm *dce_abm = TO_DMUB_ABM(abm);
7474
75− REG_WRITE(DC_ABM1_HG_SAMPLE_RATE, 0x103);
76− REG_WRITE(DC_ABM1_HG_SAMPLE_RATE, 0x101);
77− REG_WRITE(DC_ABM1_LS_SAMPLE_RATE, 0x103);
78− REG_WRITE(DC_ABM1_LS_SAMPLE_RATE, 0x101);
79− REG_WRITE(BL1_PWM_BL_UPDATE_SAMPLE_RATE, 0x101);
75+ REG_WRITE(DC_ABM1_HG_SAMPLE_RATE, 0x3);
76+ REG_WRITE(DC_ABM1_HG_SAMPLE_RATE, 0x1);
77+ REG_WRITE(DC_ABM1_LS_SAMPLE_RATE, 0x3);
78+ REG_WRITE(DC_ABM1_LS_SAMPLE_RATE, 0x1);
79+ REG_WRITE(BL1_PWM_BL_UPDATE_SAMPLE_RATE, 0x1);
8080
8181 REG_SET_3(DC_ABM1_HG_MISC_CTRL, 0,
8282 ABM1_HG_NUM_OF_BINS_SEL, 0,
drivers/gpu/drm/amd/display/dc/dce/dmub_outbox.c+60−0
@@ -0,0 +1,60 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ * Authors: AMD
23+ */
24+
25+#include "dmub_outbox.h"
26+#include "dc_dmub_srv.h"
27+#include "dmub/inc/dmub_cmd.h"
28+
29+/**
30+ *****************************************************************************
31+ * Function: dmub_enable_outbox_notification
32+ *
33+ * @brief
34+ * Sends inbox cmd to dmub to enable outbox1 messages with interrupt.
35+ * Dmub sends outbox1 message and triggers outbox1 interrupt.
36+ *
37+ * @param
38+ * [in] dc: dc structure
39+ *
40+ * @return
41+ * None
42+ *****************************************************************************
43+ */
44+void dmub_enable_outbox_notification(struct dc *dc)
45+{
46+ union dmub_rb_cmd cmd;
47+ struct dc_context *dc_ctx = dc->ctx;
48+
49+ memset(&cmd, 0x0, sizeof(cmd));
50+ cmd.outbox1_enable.header.type = DMUB_CMD__OUTBOX1_ENABLE;
51+ cmd.outbox1_enable.header.sub_type = 0;
52+ cmd.outbox1_enable.header.payload_bytes =
53+ sizeof(cmd.outbox1_enable) -
54+ sizeof(cmd.outbox1_enable.header);
55+ cmd.outbox1_enable.enable = true;
56+
57+ dc_dmub_srv_cmd_queue(dc_ctx->dmub_srv, &cmd);
58+ dc_dmub_srv_cmd_execute(dc_ctx->dmub_srv);
59+ dc_dmub_srv_wait_idle(dc_ctx->dmub_srv);
60+}
drivers/gpu/drm/amd/display/dc/dce/dmub_outbox.h+33−0
@@ -0,0 +1,33 @@
1+/*
2+ * Copyright 2020 Advanced Micro Devices, Inc.
3+ *
4+ * Permission is hereby granted, free of charge, to any person obtaining a
5+ * copy of this software and associated documentation files (the "Software"),
6+ * to deal in the Software without restriction, including without limitation
7+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8+ * and/or sell copies of the Software, and to permit persons to whom the
9+ * Software is furnished to do so, subject to the following conditions:
10+ *
11+ * The above copyright notice and this permission notice shall be included in
12+ * all copies or substantial portions of the Software.
13+ *
14+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20+ * OTHER DEALINGS IN THE SOFTWARE.
21+ *
22+ * Authors: AMD
23+ *
24+ */
25+
26+#ifndef _DMUB_OUTBOX_H_
27+#define _DMUB_OUTBOX_H_
28+
29+#include "dc.h"
30+
31+void dmub_enable_outbox_notification(struct dc *dc);
32+
33+#endif /* _DMUB_OUTBOX_H_ */
drivers/gpu/drm/amd/display/dc/dce/dmub_psr.c+12−7
@@ -81,13 +81,18 @@ static void dmub_psr_get_state(struct dmub_psr *dmub, enum dc_psr_state *state) · 80 unmodified lines
8181 {
8282 struct dmub_srv *srv = dmub->ctx->dmub_srv->dmub;
8383 uint32_t raw_state;
84+ enum dmub_status status;
8485
8586 // Send gpint command and wait for ack
86− dmub_srv_send_gpint_command(srv, DMUB_GPINT__GET_PSR_STATE, 0, 30);
87−
88− dmub_srv_get_gpint_response(srv, &raw_state);
89−
90− *state = convert_psr_state(raw_state);
87+ status = dmub_srv_send_gpint_command(srv, DMUB_GPINT__GET_PSR_STATE, 0, 30);
88+
89+ if (status == DMUB_STATUS_OK) {
90+ // GPINT was executed, get response
91+ dmub_srv_get_gpint_response(srv, &raw_state);
92+ *state = convert_psr_state(raw_state);
93+ } else
94+ // Return invalid state when GPINT times out
95+ *state = 0xFF;
9196 }
9297
9398 /*
@@ -216,6 +221,7 @@ static bool dmub_psr_copy_settings(struct dmub_psr *dmub, · 122 unmodified lines
216221 res_ctx->pipe_ctx[i].stream->link == link &&
217222 res_ctx->pipe_ctx[i].stream->link->connector_signal == SIGNAL_TYPE_EDP) {
218223 pipe_ctx = &res_ctx->pipe_ctx[i];
224+ //TODO: refactor for multi edp support
219225 break;
220226 }
221227 }
@@ -269,8 +275,7 @@ static bool dmub_psr_copy_settings(struct dmub_psr *dmub, · 47 unmodified lines
269275 copy_settings_data->frame_cap_ind = psr_context->psrFrameCaptureIndicationReq;
270276 copy_settings_data->init_sdp_deadline = psr_context->sdpTransmitLineNumDeadline;
271277 copy_settings_data->debug.u32All = 0;
272− copy_settings_data->debug.bitfields.visual_confirm = dc->dc->debug.visual_confirm == VISUAL_CONFIRM_PSR ?
273− true : false;
278+ copy_settings_data->debug.bitfields.visual_confirm = dc->dc->debug.visual_confirm == VISUAL_CONFIRM_PSR;
274279 copy_settings_data->debug.bitfields.use_hw_lock_mgr = 1;
275280
276281 dc_dmub_srv_cmd_queue(dc->dmub_srv, &cmd);
drivers/gpu/drm/amd/display/dc/dce110/dce110_hw_sequencer.c+59−32
@@ -797,6 +797,7 @@ void dce110_edp_power_control( · 796 unmodified lines
797797 struct dc_context *ctx = link->ctx;
798798 struct bp_transmitter_control cntl = { 0 };
799799 enum bp_result bp_result;
800+ uint8_t panel_instance;
800801
801802
802803 if (dal_graphics_object_id_get_connector_id(link->link_enc->connector)
@@ -807,7 +808,6 @@ void dce110_edp_power_control( · 4 unmodified lines
807808
808809 if (!link->panel_cntl)
809810 return;
810−
811811 if (power_up !=
812812 link->panel_cntl->funcs->is_panel_powered_on(link->panel_cntl)) {
813813
@@ -880,15 +880,18 @@ void dce110_edp_power_control( · 66 unmodified lines
880880 cntl.coherent = false;
881881 cntl.lanes_number = LANE_COUNT_FOUR;
882882 cntl.hpd_sel = link->link_enc->hpd_source;
883+ panel_instance = link->panel_cntl->inst;
883884
884885 if (ctx->dc->ctx->dmub_srv &&
885886 ctx->dc->debug.dmub_command_table) {
886887 if (cntl.action == TRANSMITTER_CONTROL_POWER_ON)
887888 bp_result = ctx->dc_bios->funcs->enable_lvtma_control(ctx->dc_bios,
888− LVTMA_CONTROL_POWER_ON);
889+ LVTMA_CONTROL_POWER_ON,
890+ panel_instance);
889891 else
890892 bp_result = ctx->dc_bios->funcs->enable_lvtma_control(ctx->dc_bios,
891− LVTMA_CONTROL_POWER_OFF);
893+ LVTMA_CONTROL_POWER_OFF,
894+ panel_instance);
892895 }
893896
894897 bp_result = link_transmitter_control(ctx->dc_bios, &cntl);
@@ -963,6 +966,7 @@ void dce110_edp_backlight_control( · 68 unmodified lines
963966 {
964967 struct dc_context *ctx = link->ctx;
965968 struct bp_transmitter_control cntl = { 0 };
969+ uint8_t panel_instance;
966970
967971 if (dal_graphics_object_id_get_connector_id(link->link_enc->connector)
968972 != CONNECTOR_ID_EDP) {
@@ -1011,6 +1015,7 @@ void dce110_edp_backlight_control( · 42 unmodified lines
10111015 */
10121016 /* dc_service_sleep_in_milliseconds(50); */
10131017 /*edp 1.2*/
1018+ panel_instance = link->panel_cntl->inst;
10141019 if (cntl.action == TRANSMITTER_CONTROL_BACKLIGHT_ON)
10151020 edp_receiver_ready_T7(link);
10161021
@@ -1018,10 +1023,12 @@ void dce110_edp_backlight_control( · 1 unmodified line
10181023 ctx->dc->debug.dmub_command_table) {
10191024 if (cntl.action == TRANSMITTER_CONTROL_BACKLIGHT_ON)
10201025 ctx->dc_bios->funcs->enable_lvtma_control(ctx->dc_bios,
1021− LVTMA_CONTROL_LCD_BLON);
1026+ LVTMA_CONTROL_LCD_BLON,
1027+ panel_instance);
10221028 else
10231029 ctx->dc_bios->funcs->enable_lvtma_control(ctx->dc_bios,
1024− LVTMA_CONTROL_LCD_BLOFF);
1030+ LVTMA_CONTROL_LCD_BLOFF,
1031+ panel_instance);
10251032 }
10261033
10271034 link_transmitter_control(ctx->dc_bios, &cntl);
@@ -1629,34 +1636,39 @@ static void disable_vga_and_power_gate_all_controllers( · 601 unmodified lines
16291636 }
16301637
16311638
1632−static struct dc_stream_state *get_edp_stream(struct dc_state *context)
1639+static void get_edp_streams(struct dc_state *context,
1640+ struct dc_stream_state **edp_streams,
1641+ int *edp_stream_num)
16331642 {
16341643 int i;
16351644
1645+ *edp_stream_num = 0;
16361646 for (i = 0; i < context->stream_count; i++) {
1637− if (context->streams[i]->signal == SIGNAL_TYPE_EDP)
1638− return context->streams[i];
1647+ if (context->streams[i]->signal == SIGNAL_TYPE_EDP) {
1648+ edp_streams[*edp_stream_num] = context->streams[i];
1649+ if (++(*edp_stream_num) == MAX_NUM_EDP)
1650+ return;
1651+ }
16391652 }
1640− return NULL;
16411653 }
16421654
1643−static struct dc_link *get_edp_link_with_sink(
1655+static void get_edp_links_with_sink(
16441656 struct dc *dc,
1645− struct dc_state *context)
1657+ struct dc_link **edp_links_with_sink,
1658+ int *edp_with_sink_num)
16461659 {
16471660 int i;
1648− struct dc_link *link = NULL;
16491661
16501662 /* check if there is an eDP panel not in use */
1663+ *edp_with_sink_num = 0;
16511664 for (i = 0; i < dc->link_count; i++) {
16521665 if (dc->links[i]->local_sink &&
16531666 dc->links[i]->local_sink->sink_signal == SIGNAL_TYPE_EDP) {
1654− link = dc->links[i];
1655− break;
1667+ edp_links_with_sink[*edp_with_sink_num] = dc->links[i];
1668+ if (++(*edp_with_sink_num) == MAX_NUM_EDP)
1669+ return;
16561670 }
16571671 }
1658−
1659− return link;
16601672 }
16611673
16621674 /*
@@ -1668,36 +1680,48 @@ static struct dc_link *get_edp_link_with_sink( · 5 unmodified lines
16681680 */
16691681 void dce110_enable_accelerated_mode(struct dc *dc, struct dc_state *context)
16701682 {
1671− int i;
1672− struct dc_link *edp_link_with_sink = get_edp_link_with_sink(dc, context);
1673− struct dc_link *edp_link = get_edp_link(dc);
1683+ struct dc_link *edp_links_with_sink[MAX_NUM_EDP];
1684+ struct dc_link *edp_links[MAX_NUM_EDP];
1685+ struct dc_stream_state *edp_streams[MAX_NUM_EDP];
1686+ struct dc_link *edp_link_with_sink = NULL;
1687+ struct dc_link *edp_link = NULL;
16741688 struct dc_stream_state *edp_stream = NULL;
1689+ struct dce_hwseq *hws = dc->hwseq;
1690+ int edp_with_sink_num;
1691+ int edp_num;
1692+ int edp_stream_num;
1693+ int i;
16751694 bool can_apply_edp_fast_boot = false;
16761695 bool can_apply_seamless_boot = false;
16771696 bool keep_edp_vdd_on = false;
1678− struct dce_hwseq *hws = dc->hwseq;
1697+
1698+ get_edp_links_with_sink(dc, edp_links_with_sink, &edp_with_sink_num);
1699+ get_edp_links(dc, edp_links, &edp_num);
16791700
16801701 if (hws->funcs.init_pipes)
16811702 hws->funcs.init_pipes(dc, context);
16821703
1683− edp_stream = get_edp_stream(context);
1704+ get_edp_streams(context, edp_streams, &edp_stream_num);
16841705
16851706 // Check fastboot support, disable on DCE8 because of blank screens
1686− if (edp_link && dc->ctx->dce_version != DCE_VERSION_8_0 &&
1707+ if (edp_num && dc->ctx->dce_version != DCE_VERSION_8_0 &&
16871708 dc->ctx->dce_version != DCE_VERSION_8_1 &&
16881709 dc->ctx->dce_version != DCE_VERSION_8_3) {
1689−
1690− // enable fastboot if backend is enabled on eDP
1691− if (edp_link->link_enc->funcs->is_dig_enabled(edp_link->link_enc)) {
1692− /* Set optimization flag on eDP stream*/
1693− if (edp_stream && edp_link->link_status.link_active) {
1694− edp_stream->apply_edp_fast_boot_optimization = true;
1695− can_apply_edp_fast_boot = true;
1710+ for (i = 0; i < edp_num; i++) {
1711+ edp_link = edp_links[i];
1712+ // enable fastboot if backend is enabled on eDP
1713+ if (edp_link->link_enc->funcs->is_dig_enabled(edp_link->link_enc)) {
1714+ /* Set optimization flag on eDP stream*/
1715+ if (edp_stream_num && edp_link->link_status.link_active) {
1716+ edp_stream = edp_streams[0];
1717+ edp_stream->apply_edp_fast_boot_optimization = true;
1718+ can_apply_edp_fast_boot = true;
1719+ break;
1720+ }
16961721 }
16971722 }
1698−
16991723 // We are trying to enable eDP, don't power down VDD
1700− if (edp_stream)
1724+ if (edp_stream_num)
17011725 keep_edp_vdd_on = true;
17021726 }
17031727
@@ -1712,6 +1736,9 @@ void dce110_enable_accelerated_mode(struct dc *dc, struct dc_state *context) · 8 unmodified lines
17121736 /* eDP should not have stream in resume from S4 and so even with VBios post
17131737 * it should get turned off
17141738 */
1739+ if (edp_with_sink_num)
1740+ edp_link_with_sink = edp_links_with_sink[0];
1741+
17151742 if (!can_apply_edp_fast_boot && !can_apply_seamless_boot) {
17161743 if (edp_link_with_sink && !keep_edp_vdd_on) {
17171744 /*turn off backlight before DP_blank and encoder powered down*/
@@ -1723,7 +1750,7 @@ void dce110_enable_accelerated_mode(struct dc *dc, struct dc_state *context) · 5 unmodified lines
17231750 if (edp_link_with_sink && !keep_edp_vdd_on)
17241751 dc->hwss.edp_power_control(edp_link_with_sink, false);
17251752 }
1726− bios_set_scratch_acc_mode_change(dc->ctx->dc_bios);
1753+ bios_set_scratch_acc_mode_change(dc->ctx->dc_bios, 1);
17271754 }
17281755
17291756 static uint32_t compute_pstate_blackout_duration(
drivers/gpu/drm/amd/display/dc/dce110/dce110_resource.c+2−0
@@ -1272,6 +1272,8 @@ static bool underlay_create(struct dc_context *ctx, struct resource_pool *pool) · 1271 unmodified lines
12721272
12731273 /* update the public caps to indicate an underlay is available */
12741274 ctx->dc->caps.max_slave_planes = 1;
1275+ ctx->dc->caps.max_slave_yuv_planes = 1;
1276+ ctx->dc->caps.max_slave_rgb_planes = 0;
12751277
12761278 return true;
12771279 }
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c+260−40
@@ -53,6 +53,7 @@ · 52 unmodified lines
5353 #include "dsc.h"
5454 #include "dce/dmub_hw_lock_mgr.h"
5555 #include "dc_trace.h"
56+#include "dce/dmub_outbox.h"
5657
5758 #define DC_LOGGER_INIT(logger)
5859
@@ -1355,6 +1356,10 @@ void dcn10_init_hw(struct dc *dc) · 1296 unmodified lines
13551356 hws->funcs.dsc_pg_control(hws, res_pool->dscs[i]->inst, false);
13561357 }
13571358
1359+ /* Enable outbox notification feature of dmub */
1360+ if (dc->debug.enable_dmub_aux_for_legacy_ddc)
1361+ dmub_enable_outbox_notification(dc);
1362+
13581363 /* we want to turn off all dp displays before doing detection */
13591364 if (dc->config.power_down_display_on_boot) {
13601365 uint8_t dpcd_power_state = '\0';
@@ -1457,19 +1462,26 @@ void dcn10_init_hw(struct dc *dc) · 96 unmodified lines
14571462 */
14581463 void dcn10_power_down_on_boot(struct dc *dc)
14591464 {
1460− int i = 0;
1465+ struct dc_link *edp_links[MAX_NUM_EDP];
14611466 struct dc_link *edp_link;
1467+ int edp_num;
1468+ int i = 0;
14621469
1463− edp_link = get_edp_link(dc);
1464− if (edp_link &&
1465− edp_link->link_enc->funcs->is_dig_enabled &&
1466− edp_link->link_enc->funcs->is_dig_enabled(edp_link->link_enc) &&
1467− dc->hwseq->funcs.edp_backlight_control &&
1468− dc->hwss.power_down &&
1469− dc->hwss.edp_power_control) {
1470− dc->hwseq->funcs.edp_backlight_control(edp_link, false);
1471− dc->hwss.power_down(dc);
1472− dc->hwss.edp_power_control(edp_link, false);
1470+ get_edp_links(dc, edp_links, &edp_num);
1471+
1472+ if (edp_num) {
1473+ for (i = 0; i < edp_num; i++) {
1474+ edp_link = edp_links[i];
1475+ if (edp_link->link_enc->funcs->is_dig_enabled &&
1476+ edp_link->link_enc->funcs->is_dig_enabled(edp_link->link_enc) &&
1477+ dc->hwseq->funcs.edp_backlight_control &&
1478+ dc->hwss.power_down &&
1479+ dc->hwss.edp_power_control) {
1480+ dc->hwseq->funcs.edp_backlight_control(edp_link, false);
1481+ dc->hwss.power_down(dc);
1482+ dc->hwss.edp_power_control(edp_link, false);
1483+ }
1484+ }
14731485 } else {
14741486 for (i = 0; i < dc->link_count; i++) {
14751487 struct dc_link *link = dc->links[i];
@@ -1851,6 +1863,230 @@ static bool wait_for_reset_trigger_to_occur( · 375 unmodified lines
18511863 return rc;
18521864 }
18531865
1866+uint64_t reduceSizeAndFraction(
1867+ uint64_t *numerator,
1868+ uint64_t *denominator,
1869+ bool checkUint32Bounary)
1870+{
1871+ int i;
1872+ bool ret = checkUint32Bounary == false;
1873+ uint64_t max_int32 = 0xffffffff;
1874+ uint64_t num, denom;
1875+ static const uint16_t prime_numbers[] = {
1876+ 2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43,
1877+ 47, 53, 59, 61, 67, 71, 73, 79, 83, 89, 97, 101, 103,
1878+ 107, 109, 113, 127, 131, 137, 139, 149, 151, 157, 163,
1879+ 167, 173, 179, 181, 191, 193, 197, 199, 211, 223, 227,
1880+ 229, 233, 239, 241, 251, 257, 263, 269, 271, 277, 281,
1881+ 283, 293, 307, 311, 313, 317, 331, 337, 347, 349, 353,
1882+ 359, 367, 373, 379, 383, 389, 397, 401, 409, 419, 421,
1883+ 431, 433, 439, 443, 449, 457, 461, 463, 467, 479, 487,
1884+ 491, 499, 503, 509, 521, 523, 541, 547, 557, 563, 569,
1885+ 571, 577, 587, 593, 599, 601, 607, 613, 617, 619, 631,
1886+ 641, 643, 647, 653, 659, 661, 673, 677, 683, 691, 701,
1887+ 709, 719, 727, 733, 739, 743, 751, 757, 761, 769, 773,
1888+ 787, 797, 809, 811, 821, 823, 827, 829, 839, 853, 857,
1889+ 859, 863, 877, 881, 883, 887, 907, 911, 919, 929, 937,
1890+ 941, 947, 953, 967, 971, 977, 983, 991, 997};
1891+ int count = ARRAY_SIZE(prime_numbers);
1892+
1893+ num = *numerator;
1894+ denom = *denominator;
1895+ for (i = 0; i < count; i++) {
1896+ uint32_t num_reminder, denom_reminder;
1897+ uint64_t num_result, denom_result;
1898+ if (checkUint32Bounary &&
1899+ num <= max_int32 && denom <= max_int32) {
1900+ ret = true;
1901+ break;
1902+ }
1903+ do {
1904+ num_result = div_u64_rem(num, prime_numbers[i], &num_reminder);
1905+ denom_result = div_u64_rem(denom, prime_numbers[i], &denom_reminder);
1906+ if (num_reminder == 0 && denom_reminder == 0) {
1907+ num = num_result;
1908+ denom = denom_result;
1909+ }
1910+ } while (num_reminder == 0 && denom_reminder == 0);
1911+ }
1912+ *numerator = num;
1913+ *denominator = denom;
1914+ return ret;
1915+}
1916+
1917+bool is_low_refresh_rate(struct pipe_ctx *pipe)
1918+{
1919+ uint32_t master_pipe_refresh_rate =
1920+ pipe->stream->timing.pix_clk_100hz * 100 /
1921+ pipe->stream->timing.h_total /
1922+ pipe->stream->timing.v_total;
1923+ return master_pipe_refresh_rate <= 30;
1924+}
1925+
1926+uint8_t get_clock_divider(struct pipe_ctx *pipe, bool account_low_refresh_rate)
1927+{
1928+ uint32_t clock_divider = 1;
1929+ uint32_t numpipes = 1;
1930+
1931+ if (account_low_refresh_rate && is_low_refresh_rate(pipe))
1932+ clock_divider *= 2;
1933+
1934+ if (pipe->stream_res.pix_clk_params.pixel_encoding == PIXEL_ENCODING_YCBCR420)
1935+ clock_divider *= 2;
1936+
1937+ while (pipe->next_odm_pipe) {
1938+ pipe = pipe->next_odm_pipe;
1939+ numpipes++;
1940+ }
1941+ clock_divider *= numpipes;
1942+
1943+ return clock_divider;
1944+}
1945+
1946+int dcn10_align_pixel_clocks(
1947+ struct dc *dc,
1948+ int group_size,
1949+ struct pipe_ctx *grouped_pipes[])
1950+{
1951+ struct dc_context *dc_ctx = dc->ctx;
1952+ int i, master = -1, embedded = -1;
1953+ struct dc_crtc_timing hw_crtc_timing[MAX_PIPES] = {0};
1954+ uint64_t phase[MAX_PIPES];
1955+ uint64_t modulo[MAX_PIPES];
1956+ unsigned int pclk;
1957+
1958+ uint32_t embedded_pix_clk_100hz;
1959+ uint16_t embedded_h_total;
1960+ uint16_t embedded_v_total;
1961+ bool clamshell_closed = false;
1962+ uint32_t dp_ref_clk_100hz =
1963+ dc->res_pool->dp_clock_source->ctx->dc->clk_mgr->dprefclk_khz*10;
1964+
1965+ if (dc->config.vblank_alignment_dto_params &&
1966+ dc->res_pool->dp_clock_source->funcs->override_dp_pix_clk) {
1967+ clamshell_closed =
1968+ (dc->config.vblank_alignment_dto_params >> 63);
1969+ embedded_h_total =
1970+ (dc->config.vblank_alignment_dto_params >> 32) & 0x7FFF;
1971+ embedded_v_total =
1972+ (dc->config.vblank_alignment_dto_params >> 48) & 0x7FFF;
1973+ embedded_pix_clk_100hz =
1974+ dc->config.vblank_alignment_dto_params & 0xFFFFFFFF;
1975+
1976+ for (i = 0; i < group_size; i++) {
1977+ grouped_pipes[i]->stream_res.tg->funcs->get_hw_timing(
1978+ grouped_pipes[i]->stream_res.tg,
1979+ &hw_crtc_timing[i]);
1980+ dc->res_pool->dp_clock_source->funcs->get_pixel_clk_frequency_100hz(
1981+ dc->res_pool->dp_clock_source,
1982+ grouped_pipes[i]->stream_res.tg->inst,
1983+ &pclk);
1984+ hw_crtc_timing[i].pix_clk_100hz = pclk;
1985+ if (dc_is_embedded_signal(
1986+ grouped_pipes[i]->stream->signal)) {
1987+ embedded = i;
1988+ master = i;
1989+ phase[i] = embedded_pix_clk_100hz*100;
1990+ modulo[i] = dp_ref_clk_100hz*100;
1991+ } else {
1992+
1993+ phase[i] = (uint64_t)embedded_pix_clk_100hz*
1994+ hw_crtc_timing[i].h_total*
1995+ hw_crtc_timing[i].v_total;
1996+ phase[i] = div_u64(phase[i], get_clock_divider(grouped_pipes[i], true));
1997+ modulo[i] = (uint64_t)dp_ref_clk_100hz*
1998+ embedded_h_total*
1999+ embedded_v_total;
2000+
2001+ if (reduceSizeAndFraction(&phase[i],
2002+ &modulo[i], true) == false) {
2003+ /*
2004+ * this will help to stop reporting
2005+ * this timing synchronizable
2006+ */
2007+ DC_SYNC_INFO("Failed to reduce DTO parameters\n");
2008+ grouped_pipes[i]->stream->has_non_synchronizable_pclk = true;
2009+ }
2010+ }
2011+ }
2012+
2013+ for (i = 0; i < group_size; i++) {
2014+ if (i != embedded && !grouped_pipes[i]->stream->has_non_synchronizable_pclk) {
2015+ dc->res_pool->dp_clock_source->funcs->override_dp_pix_clk(
2016+ dc->res_pool->dp_clock_source,
2017+ grouped_pipes[i]->stream_res.tg->inst,
2018+ phase[i], modulo[i]);
2019+ dc->res_pool->dp_clock_source->funcs->get_pixel_clk_frequency_100hz(
2020+ dc->res_pool->dp_clock_source,
2021+ grouped_pipes[i]->stream_res.tg->inst, &pclk);
2022+ grouped_pipes[i]->stream->timing.pix_clk_100hz =
2023+ pclk*get_clock_divider(grouped_pipes[i], false);
2024+ if (master == -1)
2025+ master = i;
2026+ }
2027+ }
2028+
2029+ }
2030+ return master;
2031+}
2032+
2033+void dcn10_enable_vblanks_synchronization(
2034+ struct dc *dc,
2035+ int group_index,
2036+ int group_size,
2037+ struct pipe_ctx *grouped_pipes[])
2038+{
2039+ struct dc_context *dc_ctx = dc->ctx;
2040+ struct output_pixel_processor *opp;
2041+ struct timing_generator *tg;
2042+ int i, width, height, master;
2043+
2044+ for (i = 1; i < group_size; i++) {
2045+ opp = grouped_pipes[i]->stream_res.opp;
2046+ tg = grouped_pipes[i]->stream_res.tg;
2047+ tg->funcs->get_otg_active_size(tg, &width, &height);
2048+ if (opp->funcs->opp_program_dpg_dimensions)
2049+ opp->funcs->opp_program_dpg_dimensions(opp, width, 2*(height) + 1);
2050+ }
2051+
2052+ for (i = 0; i < group_size; i++) {
2053+ if (grouped_pipes[i]->stream == NULL)
2054+ continue;
2055+ grouped_pipes[i]->stream->vblank_synchronized = false;
2056+ grouped_pipes[i]->stream->has_non_synchronizable_pclk = false;
2057+ }
2058+
2059+ DC_SYNC_INFO("Aligning DP DTOs\n");
2060+
2061+ master = dcn10_align_pixel_clocks(dc, group_size, grouped_pipes);
2062+
2063+ DC_SYNC_INFO("Synchronizing VBlanks\n");
2064+
2065+ if (master >= 0) {
2066+ for (i = 0; i < group_size; i++) {
2067+ if (i != master && !grouped_pipes[i]->stream->has_non_synchronizable_pclk)
2068+ grouped_pipes[i]->stream_res.tg->funcs->align_vblanks(
2069+ grouped_pipes[master]->stream_res.tg,
2070+ grouped_pipes[i]->stream_res.tg,
2071+ grouped_pipes[master]->stream->timing.pix_clk_100hz,
2072+ grouped_pipes[i]->stream->timing.pix_clk_100hz,
2073+ get_clock_divider(grouped_pipes[master], false),
2074+ get_clock_divider(grouped_pipes[i], false));
2075+ grouped_pipes[i]->stream->vblank_synchronized = true;
2076+ }
2077+ grouped_pipes[master]->stream->vblank_synchronized = true;
2078+ DC_SYNC_INFO("Sync complete\n");
2079+ }
2080+
2081+ for (i = 1; i < group_size; i++) {
2082+ opp = grouped_pipes[i]->stream_res.opp;
2083+ tg = grouped_pipes[i]->stream_res.tg;
2084+ tg->funcs->get_otg_active_size(tg, &width, &height);
2085+ if (opp->funcs->opp_program_dpg_dimensions)
2086+ opp->funcs->opp_program_dpg_dimensions(opp, width, height);
2087+ }
2088+}
2089+
18542090 void dcn10_enable_timing_synchronization(
18552091 struct dc *dc,
18562092 int group_index,
@@ -1872,6 +2108,12 @@ void dcn10_enable_timing_synchronization( · 15 unmodified lines
18722108 opp->funcs->opp_program_dpg_dimensions(opp, width, 2*(height) + 1);
18732109 }
18742110
2111+ for (i = 0; i < group_size; i++) {
2112+ if (grouped_pipes[i]->stream == NULL)
2113+ continue;
2114+ grouped_pipes[i]->stream->vblank_synchronized = false;
2115+ }
2116+
18752117 for (i = 1; i < group_size; i++)
18762118 grouped_pipes[i]->stream_res.tg->funcs->enable_reset_trigger(
18772119 grouped_pipes[i]->stream_res.tg,
@@ -2642,7 +2884,7 @@ static void dcn10_update_dchubp_dpp( · 764 unmodified lines
26422884 hws->funcs.update_plane_addr(dc, pipe_ctx);
26432885
26442886 if (is_pipe_tree_visible(pipe_ctx))
2645− dc->hwss.set_hubp_blank(dc, pipe_ctx, false);
2887+ hubp->funcs->set_blank(hubp, false);
26462888 }
26472889
26482890 void dcn10_blank_pixel_data(
@@ -3153,16 +3395,13 @@ void dcn10_setup_stereo(struct pipe_ctx *pipe_ctx, struct dc *dc) · 504 unmodified lines
31533395 return;
31543396 }
31553397
3156−static struct pipe_ctx *get_pipe_ctx_by_hubp_inst(struct dc_state *context, int mpcc_inst)
3398+static struct hubp *get_hubp_by_inst(struct resource_pool *res_pool, int mpcc_inst)
31573399 {
31583400 int i;
31593401
3160− for (i = 0; i < MAX_PIPES; i++) {
3161− if (context->res_ctx.pipe_ctx[i].plane_res.hubp
3162− && context->res_ctx.pipe_ctx[i].plane_res.hubp->inst == mpcc_inst) {
3163− return &context->res_ctx.pipe_ctx[i];
3164− }
3165−
3402+ for (i = 0; i < res_pool->pipe_count; i++) {
3403+ if (res_pool->hubps[i]->inst == mpcc_inst)
3404+ return res_pool->hubps[i];
31663405 }
31673406 ASSERT(false);
31683407 return NULL;
@@ -3185,23 +3424,11 @@ void dcn10_wait_for_mpcc_disconnect( · 16 unmodified lines
31853424
31863425 for (mpcc_inst = 0; mpcc_inst < MAX_PIPES; mpcc_inst++) {
31873426 if (pipe_ctx->stream_res.opp->mpcc_disconnect_pending[mpcc_inst]) {
3188− struct pipe_ctx *restore_bottom_pipe;
3189− struct pipe_ctx *restore_top_pipe;
3190− struct pipe_ctx *inst_pipe_ctx = get_pipe_ctx_by_hubp_inst(dc->current_state, mpcc_inst);
3427+ struct hubp *hubp = get_hubp_by_inst(res_pool, mpcc_inst);
31913428
3192− ASSERT(inst_pipe_ctx);
31933429 res_pool->mpc->funcs->wait_for_idle(res_pool->mpc, mpcc_inst);
31943430 pipe_ctx->stream_res.opp->mpcc_disconnect_pending[mpcc_inst] = false;
3195− /*
3196− * Set top and bottom pipes NULL, as we don't want
3197− * to blank those pipes when disconnecting from MPCC
3198− */
3199− restore_bottom_pipe = inst_pipe_ctx->bottom_pipe;
3200− restore_top_pipe = inst_pipe_ctx->top_pipe;
3201− inst_pipe_ctx->top_pipe = inst_pipe_ctx->bottom_pipe = NULL;
3202− dc->hwss.set_hubp_blank(dc, inst_pipe_ctx, true);
3203− inst_pipe_ctx->top_pipe = restore_top_pipe;
3204− inst_pipe_ctx->bottom_pipe = restore_bottom_pipe;
3431+ hubp->funcs->set_blank(hubp, true);
32053432 }
32063433 }
32073434
@@ -3754,10 +3981,3 @@ void dcn10_get_clock(struct dc *dc, · 546 unmodified lines
37543981 dc->clk_mgr->funcs->get_clock(dc->clk_mgr, context, clock_type, clock_cfg);
37553982
37563983 }
3757−
3758−void dcn10_set_hubp_blank(const struct dc *dc,
3759− struct pipe_ctx *pipe_ctx,
3760− bool blank_enable)
3761−{
3762− pipe_ctx->plane_res.hubp->funcs->set_blank(pipe_ctx->plane_res.hubp, blank_enable);
3763−}
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.h+5−3
@@ -123,6 +123,11 @@ void dcn10_enable_timing_synchronization( · 122 unmodified lines
123123 int group_index,
124124 int group_size,
125125 struct pipe_ctx *grouped_pipes[]);
126+void dcn10_enable_vblanks_synchronization(
127+ struct dc *dc,
128+ int group_index,
129+ int group_size,
130+ struct pipe_ctx *grouped_pipes[]);
126131 void dcn10_enable_per_frame_crtc_position_reset(
127132 struct dc *dc,
128133 int group_size,
@@ -204,8 +209,5 @@ void dcn10_wait_for_pending_cleared(struct dc *dc, · 75 unmodified lines
204209 struct dc_state *context);
205210 void dcn10_set_hdr_multiplier(struct pipe_ctx *pipe_ctx);
206211 void dcn10_verify_allow_pstate_change_high(struct dc *dc);
207−void dcn10_set_hubp_blank(const struct dc *dc,
208− struct pipe_ctx *pipe_ctx,
209− bool blank_enable);
210212
211213 #endif /* __DC_HWSS_DCN10_H__ */
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_init.c+0−1
@@ -79,7 +79,6 @@ static const struct hw_sequencer_funcs dcn10_funcs = { · 78 unmodified lines
7979 .set_backlight_level = dce110_set_backlight_level,
8080 .set_abm_immediate_disable = dce110_set_abm_immediate_disable,
8181 .set_pipe = dce110_set_pipe,
82− .set_hubp_blank = dcn10_set_hubp_blank,
8382 };
8483
8584 static const struct hwseq_private_funcs dcn10_private_funcs = {
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_optc.h+2−0
@@ -212,6 +212,7 @@ struct dcn_optc_registers { · 211 unmodified lines
212212 SF(OTG0_OTG_CONTROL, OTG_START_POINT_CNTL, mask_sh),\
213213 SF(OTG0_OTG_CONTROL, OTG_DISABLE_POINT_CNTL, mask_sh),\
214214 SF(OTG0_OTG_CONTROL, OTG_FIELD_NUMBER_CNTL, mask_sh),\
215+ SF(OTG0_OTG_CONTROL, OTG_CURRENT_MASTER_EN_STATE, mask_sh),\
215216 SF(OTG0_OTG_STEREO_CONTROL, OTG_STEREO_EN, mask_sh),\
216217 SF(OTG0_OTG_STEREO_CONTROL, OTG_STEREO_SYNC_OUTPUT_LINE_NUM, mask_sh),\
217218 SF(OTG0_OTG_STEREO_CONTROL, OTG_STEREO_SYNC_OUTPUT_POLARITY, mask_sh),\
@@ -352,6 +353,7 @@ struct dcn_optc_registers { · 134 unmodified lines
352353 type OTG_START_POINT_CNTL;\
353354 type OTG_DISABLE_POINT_CNTL;\
354355 type OTG_FIELD_NUMBER_CNTL;\
356+ type OTG_CURRENT_MASTER_EN_STATE;\
355357 type OTG_STEREO_EN;\
356358 type OTG_STEREO_SYNC_OUTPUT_LINE_NUM;\
357359 type OTG_STEREO_SYNC_OUTPUT_POLARITY;\
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_resource.c+5−1
@@ -82,7 +82,7 @@ const struct _vcs_dpi_ip_params_st dcn1_0_ip = { · 81 unmodified lines
8282 .meta_chunk_size_kbytes = 2,
8383 .writeback_chunk_size_kbytes = 2,
8484 .line_buffer_size_bits = 589824,
85− .max_line_buffer_lines = 12,
85+ .max_line_buffer_lines = 32,
8686 .IsLineBufferBppFixed = 0,
8787 .LineBufferFixedBpp = -1,
8888 .writeback_luma_buffer_size_kbytes = 12,
@@ -619,6 +619,7 @@ static const struct dc_debug_options debug_defaults_drv = { · 530 unmodified lines
619619 .recovery_enabled = false, /*enable this by default after testing.*/
620620 .max_downscale_src_width = 3840,
621621 .underflow_assert_delay_us = 0xFFFFFFFF,
622+ .use_max_lb = true
622623 };
623624
624625 static const struct dc_debug_options debug_defaults_diags = {
@@ -630,6 +631,7 @@ static const struct dc_debug_options debug_defaults_diags = { · 5 unmodified lines
630631 .disable_pplib_clock_request = true,
631632 .disable_pplib_wm_range = true,
632633 .underflow_assert_delay_us = 0xFFFFFFFF,
634+ .use_max_lb = true
633635 };
634636
635637 static void dcn10_dpp_destroy(struct dpp **dpp)
@@ -1420,6 +1422,8 @@ static bool dcn10_resource_construct( · 784 unmodified lines
14201422 dc->caps.max_cursor_size = 256;
14211423 dc->caps.min_horizontal_blanking_period = 80;
14221424 dc->caps.max_slave_planes = 1;
1425+ dc->caps.max_slave_yuv_planes = 1;
1426+ dc->caps.max_slave_rgb_planes = 0;
14231427 dc->caps.is_apu = true;
14241428 dc->caps.post_blend_color_processing = false;
14251429 dc->caps.extended_aux_timeout_support = false;
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_stream_encoder.c+0−1
@@ -738,7 +738,6 @@ void enc1_stream_encoder_update_dp_info_packets( · 737 unmodified lines
738738 REG_UPDATE(DP_SEC_CNTL, DP_SEC_GSP2_ENABLE, info_frame->spd.valid);
739739 REG_UPDATE(DP_SEC_CNTL, DP_SEC_GSP3_ENABLE, info_frame->hdrsmd.valid);
740740
741−
742741 /* This bit is the master enable bit.
743742 * When enabling secondary stream engine,
744743 * this master bit must also be set.
drivers/gpu/drm/amd/display/dc/dcn20/dcn20_hwseq.c+14−35
@@ -1507,38 +1507,8 @@ static void dcn20_update_dchubp_dpp( · 1506 unmodified lines
15071507 if (pipe_ctx->update_flags.bits.enable || pipe_ctx->update_flags.bits.opp_changed
15081508 || pipe_ctx->stream->update_flags.bits.gamut_remap
15091509 || pipe_ctx->stream->update_flags.bits.out_csc) {
1510− struct mpc *mpc = pipe_ctx->stream_res.opp->ctx->dc->res_pool->mpc;
1511−
1512− if (mpc->funcs->set_gamut_remap) {
1513− int i;
1514− int mpcc_id = hubp->inst;
1515− struct mpc_grph_gamut_adjustment adjust;
1516− bool enable_remap_dpp = false;
1517−
1518− memset(&adjust, 0, sizeof(adjust));
1519− adjust.gamut_adjust_type = GRAPHICS_GAMUT_ADJUST_TYPE_BYPASS;
1520−
1521− /* save the enablement of gamut remap for dpp */
1522− enable_remap_dpp = pipe_ctx->stream->gamut_remap_matrix.enable_remap;
1523−
1524− /* force bypass gamut remap for dpp/cm */
1525− pipe_ctx->stream->gamut_remap_matrix.enable_remap = false;
1526− dc->hwss.program_gamut_remap(pipe_ctx);
1527−
1528− /* restore gamut remap flag and use this remap into mpc */
1529− pipe_ctx->stream->gamut_remap_matrix.enable_remap = enable_remap_dpp;
1530−
1531− /* build remap matrix for top plane if enabled */
1532− if (enable_remap_dpp && pipe_ctx->top_pipe == NULL) {
1533− adjust.gamut_adjust_type = GRAPHICS_GAMUT_ADJUST_TYPE_SW;
1534− for (i = 0; i < CSC_TEMPERATURE_MATRIX_SIZE; i++)
1535− adjust.temperature_matrix[i] =
1536− pipe_ctx->stream->gamut_remap_matrix.matrix[i];
1537− }
1538− mpc->funcs->set_gamut_remap(mpc, mpcc_id, &adjust);
1539− } else
1540− /* dpp/cm gamut remap*/
1541− dc->hwss.program_gamut_remap(pipe_ctx);
1510+ /* dpp/cm gamut remap*/
1511+ dc->hwss.program_gamut_remap(pipe_ctx);
15421512
15431513 /*call the dcn2 method which uses mpc csc*/
15441514 dc->hwss.program_output_csc(dc,
@@ -1581,8 +1551,8 @@ static void dcn20_update_dchubp_dpp( · 36 unmodified lines
15811551
15821552
15831553
1584− if (is_pipe_tree_visible(pipe_ctx))
1585− dc->hwss.set_hubp_blank(dc, pipe_ctx, false);
1554+ if (pipe_ctx->update_flags.bits.enable)
1555+ hubp->funcs->set_blank(hubp, false);
15861556 }
15871557
15881558
@@ -1778,10 +1748,19 @@ void dcn20_post_unlock_program_front_end( · 189 unmodified lines
17781748
17791749 for (i = 0; i < dc->res_pool->pipe_count; i++) {
17801750 struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
1751+ struct pipe_ctx *mpcc_pipe;
17811752
17821753 if (pipe->vtp_locked) {
1783− dc->hwss.set_hubp_blank(dc, pipe, true);
1754+ dc->hwseq->funcs.wait_for_blank_complete(pipe->stream_res.opp);
1755+ pipe->plane_res.hubp->funcs->set_blank(pipe->plane_res.hubp, true);
17841756 pipe->vtp_locked = false;
1757+
1758+ for (mpcc_pipe = pipe->bottom_pipe; mpcc_pipe; mpcc_pipe = mpcc_pipe->bottom_pipe)
1759+ mpcc_pipe->plane_res.hubp->funcs->set_blank(mpcc_pipe->plane_res.hubp, true);
1760+
1761+ for (i = 0; i < dc->res_pool->pipe_count; i++)
1762+ if (context->res_ctx.pipe_ctx[i].update_flags.bits.disable)
1763+ dc->hwss.disable_plane(dc, &dc->current_state->res_ctx.pipe_ctx[i]);
17851764 }
17861765 }
17871766 /* WA to apply WM setting*/
drivers/gpu/drm/amd/display/dc/dcn20/dcn20_init.c+1−1
@@ -42,6 +42,7 @@ static const struct hw_sequencer_funcs dcn20_funcs = { · 41 unmodified lines
4242 .program_output_csc = dcn20_program_output_csc,
4343 .enable_accelerated_mode = dce110_enable_accelerated_mode,
4444 .enable_timing_synchronization = dcn10_enable_timing_synchronization,
45+ .enable_vblanks_synchronization = dcn10_enable_vblanks_synchronization,
4546 .enable_per_frame_crtc_position_reset = dcn10_enable_per_frame_crtc_position_reset,
4647 .update_info_frame = dce110_update_info_frame,
4748 .send_immediate_sdp_message = dcn10_send_immediate_sdp_message,
@@ -94,7 +95,6 @@ static const struct hw_sequencer_funcs dcn20_funcs = { · 46 unmodified lines
9495 .optimize_timing_for_fsft = dcn20_optimize_timing_for_fsft,
9596 #endif
9697 .set_disp_pattern_generator = dcn20_set_disp_pattern_generator,
97− .set_hubp_blank = dcn10_set_hubp_blank,
9898 };
9999
100100 static const struct hwseq_private_funcs dcn20_private_funcs = {
drivers/gpu/drm/amd/display/dc/dcn20/dcn20_optc.c+124−0
@@ -309,6 +309,129 @@ void optc2_set_dwb_source(struct timing_generator *optc, · 308 unmodified lines
309309 OPTC_DWB1_SOURCE_SELECT, optc->inst);
310310 }
311311
312+void optc2_align_vblanks(
313+ struct timing_generator *optc_master,
314+ struct timing_generator *optc_slave,
315+ uint32_t master_pixel_clock_100Hz,
316+ uint32_t slave_pixel_clock_100Hz,
317+ uint8_t master_clock_divider,
318+ uint8_t slave_clock_divider)
319+{
320+ /* accessing slave OTG registers */
321+ struct optc *optc1 = DCN10TG_FROM_TG(optc_slave);
322+
323+ uint32_t master_v_active = 0;
324+ uint32_t master_h_total = 0;
325+ uint32_t slave_h_total = 0;
326+ uint64_t L, XY;
327+ uint32_t X, Y, p = 10000;
328+ uint32_t master_update_lock;
329+
330+ /* disable slave OTG */
331+ REG_UPDATE(OTG_CONTROL, OTG_MASTER_EN, 0);
332+ /* wait until disabled */
333+ REG_WAIT(OTG_CONTROL,
334+ OTG_CURRENT_MASTER_EN_STATE,
335+ 0, 10, 5000);
336+
337+ REG_GET(OTG_H_TOTAL, OTG_H_TOTAL, &slave_h_total);
338+
339+ /* assign slave OTG to be controlled by master update lock */
340+ REG_SET(OTG_GLOBAL_CONTROL0, 0,
341+ OTG_MASTER_UPDATE_LOCK_SEL, optc_master->inst);
342+
343+ /* accessing master OTG registers */
344+ optc1 = DCN10TG_FROM_TG(optc_master);
345+
346+ /* saving update lock state, not sure if it's needed */
347+ REG_GET(OTG_MASTER_UPDATE_LOCK,
348+ OTG_MASTER_UPDATE_LOCK, &master_update_lock);
349+ /* unlocking master OTG */
350+ REG_SET(OTG_MASTER_UPDATE_LOCK, 0,
351+ OTG_MASTER_UPDATE_LOCK, 0);
352+
353+ REG_GET(OTG_V_BLANK_START_END,
354+ OTG_V_BLANK_START, &master_v_active);
355+ REG_GET(OTG_H_TOTAL, OTG_H_TOTAL, &master_h_total);
356+
357+ /* calculate when to enable slave OTG */
358+ L = (uint64_t)p * slave_h_total * master_pixel_clock_100Hz;
359+ L = div_u64(L, master_h_total);
360+ L = div_u64(L, slave_pixel_clock_100Hz);
361+ XY = div_u64(L, p);
362+ Y = master_v_active - XY - 1;
363+ X = div_u64(((XY + 1) * p - L) * master_h_total, p * master_clock_divider);
364+
365+ /*
366+ * set master OTG to unlock when V/H
367+ * counters reach calculated values
368+ */
369+ REG_UPDATE(OTG_GLOBAL_CONTROL1,
370+ MASTER_UPDATE_LOCK_DB_EN, 1);
371+ REG_UPDATE_2(OTG_GLOBAL_CONTROL1,
372+ MASTER_UPDATE_LOCK_DB_X,
373+ X,
374+ MASTER_UPDATE_LOCK_DB_Y,
375+ Y);
376+
377+ /* lock master OTG */
378+ REG_SET(OTG_MASTER_UPDATE_LOCK, 0,
379+ OTG_MASTER_UPDATE_LOCK, 1);
380+ REG_WAIT(OTG_MASTER_UPDATE_LOCK,
381+ UPDATE_LOCK_STATUS, 1, 1, 10);
382+
383+ /* accessing slave OTG registers */
384+ optc1 = DCN10TG_FROM_TG(optc_slave);
385+
386+ /*
387+ * enable slave OTG, the OTG is locked with
388+ * master's update lock, so it will not run
389+ */
390+ REG_UPDATE(OTG_CONTROL,
391+ OTG_MASTER_EN, 1);
392+
393+ /* accessing master OTG registers */
394+ optc1 = DCN10TG_FROM_TG(optc_master);
395+
396+ /*
397+ * unlock master OTG. When master H/V counters reach
398+ * DB_XY point, slave OTG will start
399+ */
400+ REG_SET(OTG_MASTER_UPDATE_LOCK, 0,
401+ OTG_MASTER_UPDATE_LOCK, 0);
402+
403+ /* accessing slave OTG registers */
404+ optc1 = DCN10TG_FROM_TG(optc_slave);
405+
406+ /* wait for slave OTG to start running*/
407+ REG_WAIT(OTG_CONTROL,
408+ OTG_CURRENT_MASTER_EN_STATE,
409+ 1, 10, 5000);
410+
411+ /* accessing master OTG registers */
412+ optc1 = DCN10TG_FROM_TG(optc_master);
413+
414+ /* disable the XY point*/
415+ REG_UPDATE(OTG_GLOBAL_CONTROL1,
416+ MASTER_UPDATE_LOCK_DB_EN, 0);
417+ REG_UPDATE_2(OTG_GLOBAL_CONTROL1,
418+ MASTER_UPDATE_LOCK_DB_X,
419+ 0,
420+ MASTER_UPDATE_LOCK_DB_Y,
421+ 0);
422+
423+ /*restore master update lock*/
424+ REG_SET(OTG_MASTER_UPDATE_LOCK, 0,
425+ OTG_MASTER_UPDATE_LOCK, master_update_lock);
426+
427+ /* accessing slave OTG registers */
428+ optc1 = DCN10TG_FROM_TG(optc_slave);
429+ /* restore slave to be controlled by it's own */
430+ REG_SET(OTG_GLOBAL_CONTROL0, 0,
431+ OTG_MASTER_UPDATE_LOCK_SEL, optc_slave->inst);
432+
433+}
434+
312435 void optc2_triplebuffer_lock(struct timing_generator *optc)
313436 {
314437 struct optc *optc1 = DCN10TG_FROM_TG(optc);
@@ -468,6 +591,7 @@ static struct timing_generator_funcs dcn20_tg_funcs = { · 153 unmodified lines
468591 .program_manual_trigger = optc2_program_manual_trigger,
469592 .setup_manual_trigger = optc2_setup_manual_trigger,
470593 .get_hw_timing = optc1_get_hw_timing,
594+ .align_vblanks = optc2_align_vblanks,
471595 };
472596
473597 void dcn20_timing_generator_init(struct optc *optc1)
drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c+12−4
@@ -112,7 +112,7 @@ struct _vcs_dpi_ip_params_st dcn2_0_ip = { · 111 unmodified lines
112112 .is_line_buffer_bpp_fixed = 0,
113113 .line_buffer_fixed_bpp = 0,
114114 .dcc_supported = true,
115− .max_line_buffer_lines = 12,
115+ .max_line_buffer_lines = 32,
116116 .writeback_luma_buffer_size_kbytes = 12,
117117 .writeback_chroma_buffer_size_kbytes = 8,
118118 .writeback_chroma_line_buffer_width_pixels = 4,
@@ -180,7 +180,7 @@ static struct _vcs_dpi_ip_params_st dcn2_0_nv14_ip = { · 61 unmodified lines
180180 .is_line_buffer_bpp_fixed = 0,
181181 .line_buffer_fixed_bpp = 0,
182182 .dcc_supported = true,
183− .max_line_buffer_lines = 12,
183+ .max_line_buffer_lines = 32,
184184 .writeback_luma_buffer_size_kbytes = 12,
185185 .writeback_chroma_buffer_size_kbytes = 8,
186186 .writeback_chroma_line_buffer_width_pixels = 4,
@@ -1075,6 +1075,7 @@ static const struct dc_debug_options debug_defaults_drv = { · 888 unmodified lines
10751075 .scl_reset_length10 = true,
10761076 .sanity_checks = false,
10771077 .underflow_assert_delay_us = 0xFFFFFFFF,
1078+ .use_max_lb = true
10781079 };
10791080
10801081 static const struct dc_debug_options debug_defaults_diags = {
@@ -1091,6 +1092,7 @@ static const struct dc_debug_options debug_defaults_diags = { · 10 unmodified lines
10911092 .scl_reset_length10 = true,
10921093 .underflow_assert_delay_us = 0xFFFFFFFF,
10931094 .enable_tri_buf = true,
1095+ .use_max_lb = true
10941096 };
10951097
10961098 void dcn20_dpp_destroy(struct dpp **dpp)
@@ -2033,9 +2035,13 @@ int dcn20_populate_dml_pipes_from_context( · 936 unmodified lines
20332035 if (res_ctx->pipe_ctx[pipe_cnt].stream == res_ctx->pipe_ctx[i].stream)
20342036 continue;
20352037
2036− if (dc->debug.disable_timing_sync || !resource_are_streams_timing_synchronizable(
2038+ if (dc->debug.disable_timing_sync ||
2039+ (!resource_are_streams_timing_synchronizable(
20372040 res_ctx->pipe_ctx[pipe_cnt].stream,
2038− res_ctx->pipe_ctx[i].stream)) {
2041+ res_ctx->pipe_ctx[i].stream) &&
2042+ !resource_are_vblanks_synchronizable(
2043+ res_ctx->pipe_ctx[pipe_cnt].stream,
2044+ res_ctx->pipe_ctx[i].stream))) {
20392045 synchronized_vblank = false;
20402046 break;
20412047 }
@@ -3697,6 +3703,8 @@ static bool dcn20_resource_construct( · 1655 unmodified lines
36973703 dc->caps.dmdata_alloc_size = 2048;
36983704
36993705 dc->caps.max_slave_planes = 1;
3706+ dc->caps.max_slave_yuv_planes = 1;
3707+ dc->caps.max_slave_rgb_planes = 1;
37003708 dc->caps.post_blend_color_processing = true;
37013709 dc->caps.force_dp_tps4_for_cp2520 = true;
37023710 dc->caps.extended_aux_timeout_support = true;
drivers/gpu/drm/amd/display/dc/dcn21/dcn21_init.c+0−1
@@ -99,7 +99,6 @@ static const struct hw_sequencer_funcs dcn21_funcs = { · 98 unmodified lines
9999 #endif
100100 .is_abm_supported = dcn21_is_abm_supported,
101101 .set_disp_pattern_generator = dcn20_set_disp_pattern_generator,
102− .set_hubp_blank = dcn10_set_hubp_blank,
103102 };
104103
105104 static const struct hwseq_private_funcs dcn21_private_funcs = {
drivers/gpu/drm/amd/display/dc/dcn21/dcn21_resource.c+15−6
@@ -61,7 +61,6 @@ · 60 unmodified lines
6161 #include "dcn21/dcn21_dccg.h"
6262 #include "dcn21_hubbub.h"
6363 #include "dcn10/dcn10_resource.h"
64−#include "dce110/dce110_resource.h"
6564 #include "dce/dce_panel_cntl.h"
6665
6766 #include "dcn20/dcn20_dwb.h"
@@ -116,7 +115,7 @@ struct _vcs_dpi_ip_params_st dcn2_1_ip = { · 48 unmodified lines
116115 .is_line_buffer_bpp_fixed = 0,
117116 .line_buffer_fixed_bpp = 0,
118117 .dcc_supported = true,
119− .max_line_buffer_lines = 12,
118+ .max_line_buffer_lines = 32,
120119 .writeback_luma_buffer_size_kbytes = 12,
121120 .writeback_chroma_buffer_size_kbytes = 8,
122121 .writeback_chroma_line_buffer_width_pixels = 4,
@@ -883,7 +882,9 @@ static const struct dc_debug_options debug_defaults_drv = { · 760 unmodified lines
883882 .scl_reset_length10 = true,
884883 .sanity_checks = true,
885884 .disable_48mhz_pwrdwn = false,
886− .usbc_combo_phy_reset_wa = true
885+ .usbc_combo_phy_reset_wa = true,
886+ .dmub_command_table = true,
887+ .use_max_lb = true
887888 };
888889
889890 static const struct dc_debug_options debug_defaults_diags = {
@@ -899,7 +900,8 @@ static const struct dc_debug_options debug_defaults_diags = { · 9 unmodified lines
899900 .disable_stutter = true,
900901 .disable_48mhz_pwrdwn = true,
901902 .disable_psr = true,
902− .enable_tri_buf = true
903+ .enable_tri_buf = true,
904+ .use_max_lb = true
903905 };
904906
905907 enum dcn20_clk_src_array_id {
@@ -1595,6 +1597,11 @@ static void update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_param · 689 unmodified lines
15951597 dcn2_1_soc.num_chans = bw_params->num_channels;
15961598
15971599 ASSERT(clk_table->num_entries);
1600+ /* Copy dcn2_1_soc.clock_limits to clock_limits to avoid copying over null states later */
1601+ for (i = 0; i < dcn2_1_soc.num_states + 1; i++) {
1602+ clock_limits[i] = dcn2_1_soc.clock_limits[i];
1603+ }
1604+
15981605 for (i = 0; i < clk_table->num_entries; i++) {
15991606 /* loop backwards*/
16001607 for (closest_clk_lvl = 0, j = dcn2_1_soc.num_states - 1; j >= 0; j--) {
@@ -1628,11 +1635,11 @@ static void update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_param · 27 unmodified lines
16281635 dcn2_1_soc.clock_limits[i] = clock_limits[i];
16291636 if (clk_table->num_entries) {
16301637 dcn2_1_soc.num_states = clk_table->num_entries + 1;
1638+ /* fill in min DF PState */
1639+ dcn2_1_soc.clock_limits[1] = construct_low_pstate_lvl(clk_table, closest_clk_lvl);
16311640 /* duplicate last level */
16321641 dcn2_1_soc.clock_limits[dcn2_1_soc.num_states] = dcn2_1_soc.clock_limits[dcn2_1_soc.num_states - 1];
16331642 dcn2_1_soc.clock_limits[dcn2_1_soc.num_states].state = dcn2_1_soc.num_states;
1634− /* fill in min DF PState */
1635− dcn2_1_soc.clock_limits[1] = construct_low_pstate_lvl(clk_table, closest_clk_lvl);
16361643 }
16371644
16381645 dml_init_instance(&dc->dml, &dcn2_1_soc, &dcn2_1_ip, DML_PROJECT_DCN21);
@@ -1975,6 +1982,8 @@ static bool dcn21_resource_construct( · 336 unmodified lines
19751982 dc->caps.dmdata_alloc_size = 2048;
19761983
19771984 dc->caps.max_slave_planes = 1;
1985+ dc->caps.max_slave_yuv_planes = 1;
1986+ dc->caps.max_slave_rgb_planes = 1;
19781987 dc->caps.post_blend_color_processing = true;
19791988 dc->caps.force_dp_tps4_for_cp2520 = true;
19801989 dc->caps.extended_aux_timeout_support = true;
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_cm_common.c+18−8
@@ -113,6 +113,7 @@ bool cm3_helper_translate_curve_to_hw_format( · 112 unmodified lines
113113 struct pwl_result_data *rgb_resulted;
114114 struct pwl_result_data *rgb;
115115 struct pwl_result_data *rgb_plus_1;
116+ struct pwl_result_data *rgb_minus_1;
116117 struct fixed31_32 end_value;
117118
118119 int32_t region_start, region_end;
@@ -140,7 +141,7 @@ bool cm3_helper_translate_curve_to_hw_format( · 21 unmodified lines
140141 region_start = -MAX_LOW_POINT;
141142 region_end = NUMBER_REGIONS - MAX_LOW_POINT;
142143 } else {
143− /* 10 segments
144+ /* 11 segments
144145 * segment is from 2^-10 to 2^0
145146 * There are less than 256 points, for optimization
146147 */
@@ -154,9 +155,10 @@ bool cm3_helper_translate_curve_to_hw_format( · 7 unmodified lines
154155 seg_distr[7] = 4;
155156 seg_distr[8] = 4;
156157 seg_distr[9] = 4;
158+ seg_distr[10] = 1;
157159
158160 region_start = -10;
159− region_end = 0;
161+ region_end = 1;
160162 }
161163
162164 for (i = region_end - region_start; i < MAX_REGIONS_NUMBER ; i++)
@@ -189,6 +191,10 @@ bool cm3_helper_translate_curve_to_hw_format( · 26 unmodified lines
189191 rgb_resulted[hw_points - 1].green = output_tf->tf_pts.green[start_index];
190192 rgb_resulted[hw_points - 1].blue = output_tf->tf_pts.blue[start_index];
191193
194+ rgb_resulted[hw_points].red = rgb_resulted[hw_points - 1].red;
195+ rgb_resulted[hw_points].green = rgb_resulted[hw_points - 1].green;
196+ rgb_resulted[hw_points].blue = rgb_resulted[hw_points - 1].blue;
197+
192198 // All 3 color channels have same x
193199 corner_points[0].red.x = dc_fixpt_pow(dc_fixpt_from_int(2),
194200 dc_fixpt_from_int(region_start));
@@ -259,15 +265,18 @@ bool cm3_helper_translate_curve_to_hw_format( · 64 unmodified lines
259265
260266 rgb = rgb_resulted;
261267 rgb_plus_1 = rgb_resulted + 1;
268+ rgb_minus_1 = rgb;
262269
263270 i = 1;
264271 while (i != hw_points + 1) {
265− if (dc_fixpt_lt(rgb_plus_1->red, rgb->red))
266− rgb_plus_1->red = rgb->red;
267− if (dc_fixpt_lt(rgb_plus_1->green, rgb->green))
268− rgb_plus_1->green = rgb->green;
269− if (dc_fixpt_lt(rgb_plus_1->blue, rgb->blue))
270− rgb_plus_1->blue = rgb->blue;
272+ if (i >= hw_points - 1) {
273+ if (dc_fixpt_lt(rgb_plus_1->red, rgb->red))
274+ rgb_plus_1->red = dc_fixpt_add(rgb->red, rgb_minus_1->delta_red);
275+ if (dc_fixpt_lt(rgb_plus_1->green, rgb->green))
276+ rgb_plus_1->green = dc_fixpt_add(rgb->green, rgb_minus_1->delta_green);
277+ if (dc_fixpt_lt(rgb_plus_1->blue, rgb->blue))
278+ rgb_plus_1->blue = dc_fixpt_add(rgb->blue, rgb_minus_1->delta_blue);
279+ }
271280
272281 rgb->delta_red = dc_fixpt_sub(rgb_plus_1->red, rgb->red);
273282 rgb->delta_green = dc_fixpt_sub(rgb_plus_1->green, rgb->green);
@@ -283,6 +292,7 @@ bool cm3_helper_translate_curve_to_hw_format( · 9 unmodified lines
283292 }
284293
285294 ++rgb_plus_1;
295+ rgb_minus_1 = rgb;
286296 ++rgb;
287297 ++i;
288298 }
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_dio_stream_encoder.c+0−1
@@ -454,7 +454,6 @@ static void enc3_stream_encoder_update_dp_info_packets( · 453 unmodified lines
454454 REG_UPDATE(DP_SEC_CNTL, DP_SEC_GSP2_ENABLE, info_frame->spd.valid);
455455 REG_UPDATE(DP_SEC_CNTL, DP_SEC_GSP3_ENABLE, info_frame->hdrsmd.valid);
456456
457−
458457 /* This bit is the master enable bit.
459458 * When enabling secondary stream engine,
460459 * this master bit must also be set.
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_dpp.c+2−2
@@ -718,7 +718,7 @@ bool dpp3_program_blnd_lut( · 717 unmodified lines
718718 next_mode = LUT_RAM_B;
719719
720720 dpp3_power_on_blnd_lut(dpp_base, true);
721− dpp3_configure_blnd_lut(dpp_base, next_mode == LUT_RAM_A ? true:false);
721+ dpp3_configure_blnd_lut(dpp_base, next_mode == LUT_RAM_A);
722722
723723 if (next_mode == LUT_RAM_A)
724724 dpp3_program_blnd_luta_settings(dpp_base, params);
@@ -1136,7 +1136,7 @@ bool dpp3_program_shaper( · 411 unmodified lines
11361136 else
11371137 next_mode = LUT_RAM_A;
11381138
1139− dpp3_configure_shaper_lut(dpp_base, next_mode == LUT_RAM_A ? true:false);
1139+ dpp3_configure_shaper_lut(dpp_base, next_mode == LUT_RAM_A);
11401140
11411141 if (next_mode == LUT_RAM_A)
11421142 dpp3_program_shaper_luta_settings(dpp_base, params);
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_dpp_cm.c+2−2
@@ -240,7 +240,7 @@ bool dpp3_program_gamcor_lut( · 239 unmodified lines
240240 next_mode = LUT_RAM_A;
241241
242242 dpp3_power_on_gamcor_lut(dpp_base, true);
243− dpp3_configure_gamcor_lut(dpp_base, next_mode == LUT_RAM_A ? true:false);
243+ dpp3_configure_gamcor_lut(dpp_base, next_mode == LUT_RAM_A);
244244
245245 if (next_mode == LUT_RAM_B) {
246246 gam_regs.start_cntl_b = REG(CM_GAMCOR_RAMB_START_CNTL_B);
@@ -295,7 +295,7 @@ bool dpp3_program_gamcor_lut( · 48 unmodified lines
295295 cm_helper_program_gamcor_xfer_func(dpp_base->ctx, params, &gam_regs);
296296
297297 dpp3_program_gammcor_lut(dpp_base, params->rgb_resulted, params->hw_points_num,
298− next_mode == LUT_RAM_A ? true:false);
298+ next_mode == LUT_RAM_A);
299299
300300 //select Gamma LUT to use for next frame
301301 REG_UPDATE(CM_GAMCOR_CONTROL, CM_GAMCOR_SELECT, next_mode == LUT_RAM_A ? 0:1);
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_dwb_cm.c+1−1
@@ -217,7 +217,7 @@ static bool dwb3_program_ogam_lut( · 216 unmodified lines
217217 else
218218 next_mode = LUT_RAM_A;
219219
220− dwb3_configure_ogam_lut(dwbc30, next_mode == LUT_RAM_A ? true : false);
220+ dwb3_configure_ogam_lut(dwbc30, next_mode == LUT_RAM_A);
221221
222222 if (next_mode == LUT_RAM_A)
223223 dwb3_program_ogam_luta_settings(dwbc30, params);
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_hwseq.c+34−66
@@ -421,11 +421,12 @@ void dcn30_program_all_writeback_pipes_in_tree( · 420 unmodified lines
421421
422422 void dcn30_init_hw(struct dc *dc)
423423 {
424− int i, j;
425424 struct abm **abms = dc->res_pool->multiple_abms;
426425 struct dce_hwseq *hws = dc->hwseq;
427426 struct dc_bios *dcb = dc->ctx->dc_bios;
428427 struct resource_pool *res_pool = dc->res_pool;
428+ int i, j;
429+ int edp_num;
429430 uint32_t backlight = MAX_BACKLIGHT_LEVEL;
430431
431432 if (dc->clk_mgr && dc->clk_mgr->funcs->init_clocks)
@@ -574,17 +575,23 @@ void dcn30_init_hw(struct dc *dc) · 142 unmodified lines
574575 * if DIG is turned on and seamless boot not enabled
575576 */
576577 if (dc->config.power_down_display_on_boot) {
577− struct dc_link *edp_link = get_edp_link(dc);
578−
579− if (edp_link &&
580− edp_link->link_enc->funcs->is_dig_enabled &&
581− edp_link->link_enc->funcs->is_dig_enabled(edp_link->link_enc) &&
582− dc->hwss.edp_backlight_control &&
583− dc->hwss.power_down &&
584− dc->hwss.edp_power_control) {
585− dc->hwss.edp_backlight_control(edp_link, false);
586− dc->hwss.power_down(dc);
587− dc->hwss.edp_power_control(edp_link, false);
578+ struct dc_link *edp_links[MAX_NUM_EDP];
579+ struct dc_link *edp_link;
580+
581+ get_edp_links(dc, edp_links, &edp_num);
582+ if (edp_num) {
583+ for (i = 0; i < edp_num; i++) {
584+ edp_link = edp_links[i];
585+ if (edp_link->link_enc->funcs->is_dig_enabled &&
586+ edp_link->link_enc->funcs->is_dig_enabled(edp_link->link_enc) &&
587+ dc->hwss.edp_backlight_control &&
588+ dc->hwss.power_down &&
589+ dc->hwss.edp_power_control) {
590+ dc->hwss.edp_backlight_control(edp_link, false);
591+ dc->hwss.power_down(dc);
592+ dc->hwss.edp_power_control(edp_link, false);
593+ }
594+ }
588595 } else {
589596 for (i = 0; i < dc->link_count; i++) {
590597 struct dc_link *link = dc->links[i];
@@ -651,7 +658,7 @@ void dcn30_set_avmute(struct pipe_ctx *pipe_ctx, bool enable) · 60 unmodified lines
651658 if (pipe_ctx == NULL)
652659 return;
653660
654− if (dc_is_hdmi_tmds_signal(pipe_ctx->stream->signal) && pipe_ctx->stream_res.stream_enc != NULL)
661+ if (dc_is_hdmi_signal(pipe_ctx->stream->signal) && pipe_ctx->stream_res.stream_enc != NULL)
655662 pipe_ctx->stream_res.stream_enc->funcs->set_avmute(
656663 pipe_ctx->stream_res.stream_enc,
657664 enable);
@@ -848,7 +855,7 @@ bool dcn30_apply_idle_power_optimizations(struct dc *dc, bool enable) · 190 unmodified lines
848855
849856 cmd.mall.cursor_copy_src.quad_part = cursor_attr.address.quad_part;
850857 cmd.mall.cursor_copy_dst.quad_part =
851− plane->address.grph.cursor_cache_addr.quad_part;
858+ (plane->address.grph.cursor_cache_addr.quad_part + 2047) & ~2047;
852859 cmd.mall.cursor_width = cursor_attr.width;
853860 cmd.mall.cursor_height = cursor_attr.height;
854861 cmd.mall.cursor_pitch = cursor_attr.pitch;
@@ -858,8 +865,7 @@ bool dcn30_apply_idle_power_optimizations(struct dc *dc, bool enable) · 3 unmodified lines
858865 dc_dmub_srv_wait_idle(dc->ctx->dmub_srv);
859866
860867 /* Use copied cursor, and it's okay to not switch back */
861− cursor_attr.address.quad_part =
862− plane->address.grph.cursor_cache_addr.quad_part;
868+ cursor_attr.address.quad_part = cmd.mall.cursor_copy_dst.quad_part;
863869 dc_stream_set_cursor_attributes(stream, &cursor_attr);
864870 }
865871
@@ -940,53 +946,6 @@ void dcn30_hardware_release(struct dc *dc) · 74 unmodified lines
940946 dc->res_pool->hubbub, true, true);
941947 }
942948
943−void dcn30_set_hubp_blank(const struct dc *dc,
944− struct pipe_ctx *pipe_ctx,
945− bool blank_enable)
946−{
947− struct pipe_ctx *mpcc_pipe;
948− struct pipe_ctx *odm_pipe;
949−
950− if (blank_enable) {
951− struct plane_resource *plane_res = &pipe_ctx->plane_res;
952− struct stream_resource *stream_res = &pipe_ctx->stream_res;
953−
954− /* Wait for enter vblank */
955− stream_res->tg->funcs->wait_for_state(stream_res->tg, CRTC_STATE_VBLANK);
956−
957− /* Blank HUBP to allow p-state during blank on all timings */
958− pipe_ctx->plane_res.hubp->funcs->set_blank(pipe_ctx->plane_res.hubp, true);
959− /* Confirm hubp in blank */
960− ASSERT(plane_res->hubp->funcs->hubp_in_blank(plane_res->hubp));
961− /* Toggle HUBP_DISABLE */
962− plane_res->hubp->funcs->hubp_soft_reset(plane_res->hubp, true);
963− plane_res->hubp->funcs->hubp_soft_reset(plane_res->hubp, false);
964− for (mpcc_pipe = pipe_ctx->bottom_pipe; mpcc_pipe; mpcc_pipe = mpcc_pipe->bottom_pipe) {
965− mpcc_pipe->plane_res.hubp->funcs->set_blank(mpcc_pipe->plane_res.hubp, true);
966− /* Confirm hubp in blank */
967− ASSERT(mpcc_pipe->plane_res.hubp->funcs->hubp_in_blank(mpcc_pipe->plane_res.hubp));
968− /* Toggle HUBP_DISABLE */
969− mpcc_pipe->plane_res.hubp->funcs->hubp_soft_reset(mpcc_pipe->plane_res.hubp, true);
970− mpcc_pipe->plane_res.hubp->funcs->hubp_soft_reset(mpcc_pipe->plane_res.hubp, false);
971−
972− }
973− for (odm_pipe = pipe_ctx->next_odm_pipe; odm_pipe; odm_pipe = odm_pipe->next_odm_pipe) {
974− odm_pipe->plane_res.hubp->funcs->set_blank(odm_pipe->plane_res.hubp, true);
975− /* Confirm hubp in blank */
976− ASSERT(odm_pipe->plane_res.hubp->funcs->hubp_in_blank(odm_pipe->plane_res.hubp));
977− /* Toggle HUBP_DISABLE */
978− odm_pipe->plane_res.hubp->funcs->hubp_soft_reset(odm_pipe->plane_res.hubp, true);
979− odm_pipe->plane_res.hubp->funcs->hubp_soft_reset(odm_pipe->plane_res.hubp, false);
980− }
981− } else {
982− pipe_ctx->plane_res.hubp->funcs->set_blank(pipe_ctx->plane_res.hubp, false);
983− for (mpcc_pipe = pipe_ctx->bottom_pipe; mpcc_pipe; mpcc_pipe = mpcc_pipe->bottom_pipe)
984− mpcc_pipe->plane_res.hubp->funcs->set_blank(mpcc_pipe->plane_res.hubp, false);
985− for (odm_pipe = pipe_ctx->next_odm_pipe; odm_pipe; odm_pipe = odm_pipe->next_odm_pipe)
986− odm_pipe->plane_res.hubp->funcs->set_blank(odm_pipe->plane_res.hubp, false);
987− }
988−}
989−
990949 void dcn30_set_disp_pattern_generator(const struct dc *dc,
991950 struct pipe_ctx *pipe_ctx,
992951 enum controller_dp_test_pattern test_pattern,
@@ -996,6 +955,7 @@ void dcn30_set_disp_pattern_generator(const struct dc *dc, · 3 unmodified lines
996955 int width, int height, int offset)
997956 {
998957 struct stream_resource *stream_res = &pipe_ctx->stream_res;
958+ struct pipe_ctx *mpcc_pipe;
999959
1000960 if (test_pattern != CONTROLLER_DP_TEST_PATTERN_VIDEOMODE) {
1001961 pipe_ctx->vtp_locked = false;
@@ -1007,12 +967,20 @@ void dcn30_set_disp_pattern_generator(const struct dc *dc, · 5 unmodified lines
1007967 if (stream_res->tg->funcs->is_tg_enabled(stream_res->tg)) {
1008968 if (stream_res->tg->funcs->is_locked(stream_res->tg))
1009969 pipe_ctx->vtp_locked = true;
1010− else
1011− dc->hwss.set_hubp_blank(dc, pipe_ctx, true);
970+ else {
971+ /* Blank HUBP to allow p-state during blank on all timings */
972+ pipe_ctx->plane_res.hubp->funcs->set_blank(pipe_ctx->plane_res.hubp, true);
973+
974+ for (mpcc_pipe = pipe_ctx->bottom_pipe; mpcc_pipe; mpcc_pipe = mpcc_pipe->bottom_pipe)
975+ mpcc_pipe->plane_res.hubp->funcs->set_blank(mpcc_pipe->plane_res.hubp, true);
976+ }
1012977 }
1013978 } else {
1014− dc->hwss.set_hubp_blank(dc, pipe_ctx, false);
1015979 /* turning off DPG */
980+ pipe_ctx->plane_res.hubp->funcs->set_blank(pipe_ctx->plane_res.hubp, false);
981+ for (mpcc_pipe = pipe_ctx->bottom_pipe; mpcc_pipe; mpcc_pipe = mpcc_pipe->bottom_pipe)
982+ mpcc_pipe->plane_res.hubp->funcs->set_blank(mpcc_pipe->plane_res.hubp, false);
983+
1016984 stream_res->opp->funcs->opp_set_disp_pattern_generator(stream_res->opp, test_pattern, color_space,
1017985 color_depth, solid_color, width, height, offset);
1018986 }
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_hwseq.h+0−4
@@ -80,8 +80,4 @@ void dcn30_set_disp_pattern_generator(const struct dc *dc, · 79 unmodified lines
8080 const struct tg_color *solid_color,
8181 int width, int height, int offset);
8282
83−void dcn30_set_hubp_blank(const struct dc *dc,
84− struct pipe_ctx *pipe_ctx,
85− bool blank_enable);
86−
8783 #endif /* __DC_HWSS_DCN30_H__ */
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_init.c+0−1
@@ -98,7 +98,6 @@ static const struct hw_sequencer_funcs dcn30_funcs = { · 97 unmodified lines
9898 .hardware_release = dcn30_hardware_release,
9999 .set_pipe = dcn21_set_pipe,
100100 .set_disp_pattern_generator = dcn30_set_disp_pattern_generator,
101− .set_hubp_blank = dcn30_set_hubp_blank,
102101 };
103102
104103 static const struct hwseq_private_funcs dcn30_private_funcs = {
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_mpc.c+14−12
@@ -143,16 +143,18 @@ static void mpc3_power_on_ogam_lut( · 142 unmodified lines
143143 {
144144 struct dcn30_mpc *mpc30 = TO_DCN30_MPC(mpc);
145145
146− if (mpc->ctx->dc->debug.enable_mem_low_power.bits.mpc) {
147− // Force power on
148− REG_UPDATE(MPCC_MEM_PWR_CTRL[mpcc_id], MPCC_OGAM_MEM_PWR_DIS, power_on == true ? 1:0);
149− // Wait for confirmation when powering on
150− if (power_on)
151− REG_WAIT(MPCC_MEM_PWR_CTRL[mpcc_id], MPCC_OGAM_MEM_PWR_STATE, 0, 10, 10);
152− } else {
153− REG_SET(MPCC_MEM_PWR_CTRL[mpcc_id], 0,
154− MPCC_OGAM_MEM_PWR_FORCE, power_on == true ? 0 : 1);
155− }
146+ /*
147+ * Powering on: force memory active so the LUT can be updated.
148+ * Powering off: allow entering memory low power mode
149+ *
150+ * Memory low power mode is controlled during MPC OGAM LUT init.
151+ */
152+ REG_UPDATE(MPCC_MEM_PWR_CTRL[mpcc_id],
153+ MPCC_OGAM_MEM_PWR_DIS, power_on != 0);
154+
155+ /* Wait for memory to be powered on - we won't be able to write to it otherwise. */
156+ if (power_on)
157+ REG_WAIT(MPCC_MEM_PWR_CTRL[mpcc_id], MPCC_OGAM_MEM_PWR_STATE, 0, 10, 10);
156158 }
157159
158160 static void mpc3_configure_ogam_lut(
@@ -355,7 +357,7 @@ void mpc3_set_output_gamma( · 196 unmodified lines
355357 next_mode = LUT_RAM_A;
356358
357359 mpc3_power_on_ogam_lut(mpc, mpcc_id, true);
358− mpc3_configure_ogam_lut(mpc, mpcc_id, next_mode == LUT_RAM_A ? true:false);
360+ mpc3_configure_ogam_lut(mpc, mpcc_id, next_mode == LUT_RAM_A);
359361
360362 if (next_mode == LUT_RAM_A)
361363 mpc3_program_luta(mpc, mpcc_id, params);
@@ -1427,7 +1429,7 @@ const struct mpc_funcs dcn30_mpc_funcs = { · 1065 unmodified lines
14271429 .acquire_rmu = mpcc3_acquire_rmu,
14281430 .program_3dlut = mpc3_program_3dlut,
14291431 .release_rmu = mpcc3_release_rmu,
1430− .power_on_mpc_mem_pwr = mpc20_power_on_ogam_lut,
1432+ .power_on_mpc_mem_pwr = mpc3_power_on_ogam_lut,
14311433 .get_mpc_out_mux = mpc1_get_mpc_out_mux,
14321434
14331435 };
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_resource.c+18−7
@@ -120,7 +120,7 @@ struct _vcs_dpi_ip_params_st dcn3_0_ip = { · 119 unmodified lines
120120 .dcc_supported = true,
121121 .writeback_interface_buffer_size_kbytes = 90,
122122 .writeback_line_buffer_buffer_size = 0,
123− .max_line_buffer_lines = 12,
123+ .max_line_buffer_lines = 32,
124124 .writeback_luma_buffer_size_kbytes = 12, // writeback_line_buffer_buffer_size = 656640
125125 .writeback_chroma_buffer_size_kbytes = 8,
126126 .writeback_chroma_line_buffer_width_pixels = 4,
@@ -181,7 +181,7 @@ struct _vcs_dpi_soc_bounding_box_st dcn3_0_soc = { · 54 unmodified lines
181181 },
182182 .min_dcfclk = 500.0, /* TODO: set this to actual min DCFCLK */
183183 .num_states = 1,
184− .sr_exit_time_us = 12,
184+ .sr_exit_time_us = 15.5,
185185 .sr_enter_plus_exit_time_us = 20,
186186 .urgent_latency_us = 4.0,
187187 .urgent_latency_pixel_data_only_us = 4.0,
@@ -852,6 +852,7 @@ static const struct dc_debug_options debug_defaults_drv = { · 664 unmodified lines
852852 .dwb_fi_phase = -1, // -1 = disable,
853853 .dmub_command_table = true,
854854 .disable_psr = false,
855+ .use_max_lb = true
855856 };
856857
857858 static const struct dc_debug_options debug_defaults_diags = {
@@ -870,6 +871,7 @@ static const struct dc_debug_options debug_defaults_diags = { · 12 unmodified lines
870871 .dmub_command_table = true,
871872 .disable_psr = true,
872873 .enable_tri_buf = true,
874+ .use_max_lb = true
873875 };
874876
875877 void dcn30_dpp_destroy(struct dpp **dpp)
@@ -1874,6 +1876,7 @@ static noinline bool dcn30_internal_validate_bw( · 998 unmodified lines
18741876 if (!pipes)
18751877 return false;
18761878
1879+ dc->res_pool->funcs->update_soc_for_wm_a(dc, context);
18771880 pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate);
18781881
18791882 DC_FP_START();
@@ -2223,11 +2226,7 @@ static noinline void dcn30_calculate_wm_and_dlg_fp( · 343 unmodified lines
22232226 *
22242227 * Set A calculated last so that following calculations are based on Set A
22252228 */
2226− if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].valid) {
2227− context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
2228− context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_enter_plus_exit_time_us;
2229− context->bw_ctx.dml.soc.sr_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_exit_time_us;
2230− }
2229+ dc->res_pool->funcs->update_soc_for_wm_a(dc, context);
22312230 context->bw_ctx.bw.dcn.watermarks.a.urgent_ns = get_wm_urgent(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
22322231 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_enter_plus_exit_ns = get_wm_stutter_enter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
22332232 context->bw_ctx.bw.dcn.watermarks.a.cstate_pstate.cstate_exit_ns = get_wm_stutter_exit(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000;
@@ -2270,6 +2269,15 @@ static noinline void dcn30_calculate_wm_and_dlg_fp( · 36 unmodified lines
22702269 dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
22712270 }
22722271
2272+void dcn30_update_soc_for_wm_a(struct dc *dc, struct dc_state *context)
2273+{
2274+ if (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].valid) {
2275+ context->bw_ctx.dml.soc.dram_clock_change_latency_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
2276+ context->bw_ctx.dml.soc.sr_enter_plus_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_enter_plus_exit_time_us;
2277+ context->bw_ctx.dml.soc.sr_exit_time_us = dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.sr_exit_time_us;
2278+ }
2279+}
2280+
22732281 void dcn30_calculate_wm_and_dlg(
22742282 struct dc *dc, struct dc_state *context,
22752283 display_e2e_pipe_params_st *pipes,
@@ -2494,6 +2502,7 @@ static const struct resource_funcs dcn30_res_pool_funcs = { · 218 unmodified lines
24942502 .panel_cntl_create = dcn30_panel_cntl_create,
24952503 .validate_bandwidth = dcn30_validate_bandwidth,
24962504 .calculate_wm_and_dlg = dcn30_calculate_wm_and_dlg,
2505+ .update_soc_for_wm_a = dcn30_update_soc_for_wm_a,
24972506 .populate_dml_pipes = dcn30_populate_dml_pipes_from_context,
24982507 .acquire_idle_pipe_for_layer = dcn20_acquire_idle_pipe_for_layer,
24992508 .add_stream_to_ctx = dcn30_add_stream_to_ctx,
@@ -2566,6 +2575,8 @@ static bool dcn30_resource_construct( · 66 unmodified lines
25662575 dc->caps.cursor_cache_size = dc->caps.max_cursor_size * dc->caps.max_cursor_size * 8;
25672576
25682577 dc->caps.max_slave_planes = 1;
2578+ dc->caps.max_slave_yuv_planes = 1;
2579+ dc->caps.max_slave_rgb_planes = 1;
25692580 dc->caps.post_blend_color_processing = true;
25702581 dc->caps.force_dp_tps4_for_cp2520 = true;
25712582 dc->caps.extended_aux_timeout_support = true;
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_resource.h+1−0
@@ -60,6 +60,7 @@ void dcn30_calculate_wm_and_dlg( · 59 unmodified lines
6060 display_e2e_pipe_params_st *pipes,
6161 int pipe_cnt,
6262 int vlevel);
63+void dcn30_update_soc_for_wm_a(struct dc *dc, struct dc_state *context);
6364 void dcn30_populate_dml_writeback_from_context(
6465 struct dc *dc, struct resource_context *res_ctx, display_e2e_pipe_params_st *pipes);
6566
drivers/gpu/drm/amd/display/dc/dcn301/dcn301_init.c+0−1
@@ -98,7 +98,6 @@ static const struct hw_sequencer_funcs dcn301_funcs = { · 97 unmodified lines
9898 .set_abm_immediate_disable = dcn21_set_abm_immediate_disable,
9999 .set_pipe = dcn21_set_pipe,
100100 .set_disp_pattern_generator = dcn30_set_disp_pattern_generator,
101− .set_hubp_blank = dcn30_set_hubp_blank,
102101 };
103102
104103 static const struct hwseq_private_funcs dcn301_private_funcs = {
drivers/gpu/drm/amd/display/dc/dcn301/dcn301_resource.c+6−1
@@ -116,7 +116,7 @@ struct _vcs_dpi_ip_params_st dcn3_01_ip = { · 115 unmodified lines
116116 .dcc_supported = true,
117117 .writeback_interface_buffer_size_kbytes = 90,
118118 .writeback_line_buffer_buffer_size = 656640,
119− .max_line_buffer_lines = 12,
119+ .max_line_buffer_lines = 32,
120120 .writeback_luma_buffer_size_kbytes = 12, // writeback_line_buffer_buffer_size = 656640
121121 .writeback_chroma_buffer_size_kbytes = 8,
122122 .writeback_chroma_line_buffer_width_pixels = 4,
@@ -873,6 +873,7 @@ static const struct dc_debug_options debug_defaults_drv = { · 750 unmodified lines
873873 .underflow_assert_delay_us = 0xFFFFFFFF,
874874 .dwb_fi_phase = -1, // -1 = disable
875875 .dmub_command_table = true,
876+ .use_max_lb = false,
876877 };
877878
878879 static const struct dc_debug_options debug_defaults_diags = {
@@ -889,6 +890,7 @@ static const struct dc_debug_options debug_defaults_diags = { · 10 unmodified lines
889890 .scl_reset_length10 = true,
890891 .dwb_fi_phase = -1, // -1 = disable
891892 .dmub_command_table = true,
893+ .use_max_lb = false,
892894 };
893895
894896 void dcn301_dpp_destroy(struct dpp **dpp)
@@ -1719,6 +1721,7 @@ static struct resource_funcs dcn301_res_pool_funcs = { · 824 unmodified lines
17191721 .panel_cntl_create = dcn301_panel_cntl_create,
17201722 .validate_bandwidth = dcn30_validate_bandwidth,
17211723 .calculate_wm_and_dlg = dcn301_calculate_wm_and_dlg,
1724+ .update_soc_for_wm_a = dcn30_update_soc_for_wm_a,
17221725 .populate_dml_pipes = dcn30_populate_dml_pipes_from_context,
17231726 .acquire_idle_pipe_for_layer = dcn20_acquire_idle_pipe_for_layer,
17241727 .add_stream_to_ctx = dcn30_add_stream_to_ctx,
@@ -1764,6 +1767,8 @@ static bool dcn301_resource_construct( · 39 unmodified lines
17641767 dc->caps.min_horizontal_blanking_period = 80;
17651768 dc->caps.dmdata_alloc_size = 2048;
17661769 dc->caps.max_slave_planes = 1;
1770+ dc->caps.max_slave_yuv_planes = 1;
1771+ dc->caps.max_slave_rgb_planes = 1;
17671772 dc->caps.is_apu = true;
17681773 dc->caps.post_blend_color_processing = true;
17691774 dc->caps.force_dp_tps4_for_cp2520 = true;
drivers/gpu/drm/amd/display/dc/dcn302/dcn302_resource.c+6−1
@@ -101,7 +101,7 @@ struct _vcs_dpi_ip_params_st dcn3_02_ip = { · 100 unmodified lines
101101 .dcc_supported = true,
102102 .writeback_interface_buffer_size_kbytes = 90,
103103 .writeback_line_buffer_buffer_size = 0,
104− .max_line_buffer_lines = 12,
104+ .max_line_buffer_lines = 32,
105105 .writeback_luma_buffer_size_kbytes = 12, // writeback_line_buffer_buffer_size = 656640
106106 .writeback_chroma_buffer_size_kbytes = 8,
107107 .writeback_chroma_line_buffer_width_pixels = 4,
@@ -223,6 +223,7 @@ static const struct dc_debug_options debug_defaults_drv = { · 115 unmodified lines
223223 .underflow_assert_delay_us = 0xFFFFFFFF,
224224 .dwb_fi_phase = -1, // -1 = disable,
225225 .dmub_command_table = true,
226+ .use_max_lb = true
226227 };
227228
228229 static const struct dc_debug_options debug_defaults_diags = {
@@ -241,6 +242,7 @@ static const struct dc_debug_options debug_defaults_diags = { · 12 unmodified lines
241242 .dmub_command_table = true,
242243 .enable_tri_buf = true,
243244 .disable_psr = true,
245+ .use_max_lb = true
244246 };
245247
246248 enum dcn302_clk_src_array_id {
@@ -1395,6 +1397,7 @@ static struct resource_funcs dcn302_res_pool_funcs = { · 1148 unmodified lines
13951397 .panel_cntl_create = dcn302_panel_cntl_create,
13961398 .validate_bandwidth = dcn30_validate_bandwidth,
13971399 .calculate_wm_and_dlg = dcn30_calculate_wm_and_dlg,
1400+ .update_soc_for_wm_a = dcn30_update_soc_for_wm_a,
13981401 .populate_dml_pipes = dcn30_populate_dml_pipes_from_context,
13991402 .acquire_idle_pipe_for_layer = dcn20_acquire_idle_pipe_for_layer,
14001403 .add_stream_to_ctx = dcn30_add_stream_to_ctx,
@@ -1481,6 +1484,8 @@ static bool dcn302_resource_construct( · 80 unmodified lines
14811484 dc->caps.mall_size_total = dc->caps.mall_size_per_mem_channel * dc->ctx->dc_bios->vram_info.num_chans * 1048576;
14821485 dc->caps.cursor_cache_size = dc->caps.max_cursor_size * dc->caps.max_cursor_size * 8;
14831486 dc->caps.max_slave_planes = 1;
1487+ dc->caps.max_slave_yuv_planes = 1;
1488+ dc->caps.max_slave_rgb_planes = 1;
14841489 dc->caps.post_blend_color_processing = true;
14851490 dc->caps.force_dp_tps4_for_cp2520 = true;
14861491 dc->caps.extended_aux_timeout_support = true;
drivers/gpu/drm/amd/display/dc/dm_helpers.h+2−0
@@ -156,4 +156,6 @@ void dm_set_dcn_clocks( · 155 unmodified lines
156156 struct dc_context *ctx,
157157 struct dc_clocks *clks);
158158
159+bool dm_helpers_dmub_outbox0_interrupt_control(struct dc_context *ctx, bool enable);
160+
159161 #endif /* __DM_HELPERS__ */
drivers/gpu/drm/amd/display/dc/dml/dcn20/display_mode_vba_20.c+1−0
@@ -3437,6 +3437,7 @@ void dml20_ModeSupportAndSystemConfigurationFull(struct display_mode_lib *mode_l · 3436 unmodified lines
34373437 mode_lib->vba.DCCEnabledInAnyPlane = true;
34383438 }
34393439 }
3440+ mode_lib->vba.UrgentLatency = mode_lib->vba.UrgentLatencyPixelDataOnly;
34403441 for (i = 0; i <= mode_lib->vba.soc.num_states; i++) {
34413442 locals->FabricAndDRAMBandwidthPerState[i] = dml_min(
34423443 mode_lib->vba.DRAMSpeedPerState[i] * mode_lib->vba.NumberOfChannels
drivers/gpu/drm/amd/display/dc/dml/dcn20/display_mode_vba_20v2.c+1−0
@@ -3544,6 +3544,7 @@ void dml20v2_ModeSupportAndSystemConfigurationFull(struct display_mode_lib *mode · 3543 unmodified lines
35443544 mode_lib->vba.DCCEnabledInAnyPlane = true;
35453545 }
35463546 }
3547+ mode_lib->vba.UrgentLatency = mode_lib->vba.UrgentLatencyPixelDataOnly;
35473548 for (i = 0; i <= mode_lib->vba.soc.num_states; i++) {
35483549 locals->FabricAndDRAMBandwidthPerState[i] = dml_min(
35493550 mode_lib->vba.DRAMSpeedPerState[i] * mode_lib->vba.NumberOfChannels
drivers/gpu/drm/amd/display/dc/dml/dcn30/display_mode_vba_30.c+1−1
@@ -4050,7 +4050,7 @@ void dml30_ModeSupportAndSystemConfigurationFull(struct display_mode_lib *mode_l · 4049 unmodified lines
40504050 v->RequiredDPPCLK[i][j][NumberOfNonSplitPlaneOfMaximumBandwidth] = v->MinDPPCLKUsingSingleDPP[NumberOfNonSplitPlaneOfMaximumBandwidth]
40514051 * (1 + v->DISPCLKDPPCLKDSCCLKDownSpreading / 100) / 2;
40524052 v->TotalNumberOfActiveDPP[i][j] = v->TotalNumberOfActiveDPP[i][j] + 1;
4053− v->TotalNumberOfSingleDPPPlanes[i][j] = v->TotalNumberOfSingleDPPPlanes[i][j] + 1;
4053+ v->TotalNumberOfSingleDPPPlanes[i][j] = v->TotalNumberOfSingleDPPPlanes[i][j] - 1;
40544054 }
40554055 }
40564056 if (v->TotalNumberOfActiveDPP[i][j] > v->MaxNumDPP) {
drivers/gpu/drm/amd/display/dc/dml/display_mode_vba.c+0−3
@@ -599,7 +599,6 @@ static void fetch_pipe_params(struct display_mode_lib *mode_lib) · 598 unmodified lines
599599 for (k = j + 1; k < mode_lib->vba.cache_num_pipes; ++k) {
600600 display_pipe_source_params_st *src_k = &pipes[k].pipe.src;
601601 display_pipe_dest_params_st *dst_k = &pipes[k].pipe.dest;
602− display_output_params_st *dout_k = &pipes[j].dout;
603602
604603 if (src_k->is_hsplit && !visited[k]
605604 && src->hsplit_grp == src_k->hsplit_grp) {
@@ -620,8 +619,6 @@ static void fetch_pipe_params(struct display_mode_lib *mode_lib) · 14 unmodified lines
620619 mode_lib->vba.ViewportHeightChroma[mode_lib->vba.NumberOfActivePlanes] +=
621620 src_k->viewport_height_c;
622621 }
623− mode_lib->vba.NumberOfDSCSlices[mode_lib->vba.NumberOfActivePlanes] +=
624− dout_k->dsc_slices;
625622
626623 visited[k] = true;
627624 }
drivers/gpu/drm/amd/display/dc/dsc/dc_dsc.c+34−71
@@ -37,59 +37,6 @@ static uint32_t dsc_policy_max_target_bpp_limit = 16; · 36 unmodified lines
3737 /* default DSC policy enables DSC only when needed */
3838 static bool dsc_policy_enable_dsc_when_not_needed;
3939
40−static uint32_t dc_dsc_bandwidth_in_kbps_from_timing(
41− const struct dc_crtc_timing *timing)
42−{
43− uint32_t bits_per_channel = 0;
44− uint32_t kbps;
45−
46− if (timing->flags.DSC) {
47− kbps = (timing->pix_clk_100hz * timing->dsc_cfg.bits_per_pixel);
48− kbps = kbps / 160 + ((kbps % 160) ? 1 : 0);
49− return kbps;
50− }
51−
52− switch (timing->display_color_depth) {
53− case COLOR_DEPTH_666:
54− bits_per_channel = 6;
55− break;
56− case COLOR_DEPTH_888:
57− bits_per_channel = 8;
58− break;
59− case COLOR_DEPTH_101010:
60− bits_per_channel = 10;
61− break;
62− case COLOR_DEPTH_121212:
63− bits_per_channel = 12;
64− break;
65− case COLOR_DEPTH_141414:
66− bits_per_channel = 14;
67− break;
68− case COLOR_DEPTH_161616:
69− bits_per_channel = 16;
70− break;
71− default:
72− break;
73− }
74−
75− ASSERT(bits_per_channel != 0);
76−
77− kbps = timing->pix_clk_100hz / 10;
78− kbps *= bits_per_channel;
79−
80− if (timing->flags.Y_ONLY != 1) {
81− /*Only YOnly make reduce bandwidth by 1/3 compares to RGB*/
82− kbps *= 3;
83− if (timing->pixel_encoding == PIXEL_ENCODING_YCBCR420)
84− kbps /= 2;
85− else if (timing->pixel_encoding == PIXEL_ENCODING_YCBCR422)
86− kbps = kbps * 2 / 3;
87− }
88−
89− return kbps;
90−
91−}
92−
9340 static bool dsc_buff_block_size_from_dpcd(int dpcd_buff_block_size, int *buff_block_size)
9441 {
9542
@@ -315,18 +262,18 @@ static inline uint32_t dsc_div_by_10_round_up(uint32_t value) · 219 unmodified lines
315262 * and uncompressed bandwidth.
316263 */
317264 static void get_dsc_bandwidth_range(
318− const uint32_t min_bpp,
319− const uint32_t max_bpp,
265+ const uint32_t min_bpp_x16,
266+ const uint32_t max_bpp_x16,
320267 const struct dsc_enc_caps *dsc_caps,
321268 const struct dc_crtc_timing *timing,
322269 struct dc_dsc_bw_range *range)
323270 {
324271 /* native stream bandwidth */
325− range->stream_kbps = dc_dsc_bandwidth_in_kbps_from_timing(timing);
272+ range->stream_kbps = dc_bandwidth_in_kbps_from_timing(timing);
326273
327274 /* max dsc target bpp */
328− range->max_kbps = dsc_div_by_10_round_up(max_bpp * timing->pix_clk_100hz);
329− range->max_target_bpp_x16 = max_bpp * 16;
275+ range->max_kbps = dc_dsc_stream_bandwidth_in_kbps(timing->pix_clk_100hz, max_bpp_x16);
276+ range->max_target_bpp_x16 = max_bpp_x16;
330277 if (range->max_kbps > range->stream_kbps) {
331278 /* max dsc target bpp is capped to native bandwidth */
332279 range->max_kbps = range->stream_kbps;
@@ -334,8 +281,8 @@ static void get_dsc_bandwidth_range( · 1 unmodified line
334281 }
335282
336283 /* min dsc target bpp */
337− range->min_kbps = dsc_div_by_10_round_up(min_bpp * timing->pix_clk_100hz);
338− range->min_target_bpp_x16 = min_bpp * 16;
284+ range->min_kbps = dc_dsc_stream_bandwidth_in_kbps(timing->pix_clk_100hz, min_bpp_x16);
285+ range->min_target_bpp_x16 = min_bpp_x16;
339286 if (range->min_kbps > range->max_kbps) {
340287 /* min dsc target bpp is capped to max dsc bandwidth*/
341288 range->min_kbps = range->max_kbps;
@@ -363,12 +310,17 @@ static bool decide_dsc_target_bpp_x16( · 21 unmodified lines
363310
364311 memset(&range, 0, sizeof(range));
365312
366− get_dsc_bandwidth_range(policy->min_target_bpp, policy->max_target_bpp,
313+ get_dsc_bandwidth_range(policy->min_target_bpp * 16, policy->max_target_bpp * 16,
367314 dsc_common_caps, timing, &range);
368315 if (!policy->enable_dsc_when_not_needed && target_bandwidth_kbps >= range.stream_kbps) {
369316 /* enough bandwidth without dsc */
370317 *target_bpp_x16 = 0;
371318 should_use_dsc = false;
319+ } else if (policy->preferred_bpp_x16 > 0 &&
320+ policy->preferred_bpp_x16 <= range.max_target_bpp_x16 &&
321+ policy->preferred_bpp_x16 >= range.min_target_bpp_x16) {
322+ *target_bpp_x16 = policy->preferred_bpp_x16;
323+ should_use_dsc = true;
372324 } else if (target_bandwidth_kbps >= range.max_kbps) {
373325 /* use max target bpp allowed */
374326 *target_bpp_x16 = range.max_target_bpp_x16;
@@ -545,7 +497,7 @@ static bool setup_dsc_config( · 170 unmodified lines
545497 int target_bandwidth_kbps,
546498 const struct dc_crtc_timing *timing,
547499 int min_slice_height_override,
548− int max_dsc_target_bpp_limit_override,
500+ int max_dsc_target_bpp_limit_override_x16,
549501 struct dc_dsc_config *dsc_cfg)
550502 {
551503 struct dsc_enc_caps dsc_common_caps;
@@ -564,7 +516,7 @@ static bool setup_dsc_config( · 12 unmodified lines
564516
565517 memset(dsc_cfg, 0, sizeof(struct dc_dsc_config));
566518
567− dc_dsc_get_policy_for_timing(timing, max_dsc_target_bpp_limit_override, &policy);
519+ dc_dsc_get_policy_for_timing(timing, max_dsc_target_bpp_limit_override_x16, &policy);
568520 pic_width = timing->h_addressable + timing->h_border_left + timing->h_border_right;
569521 pic_height = timing->v_addressable + timing->v_border_top + timing->v_border_bottom;
570522
@@ -865,8 +817,8 @@ bool dc_dsc_parse_dsc_dpcd(const struct dc *dc, const uint8_t *dpcd_dsc_basic_da · 294 unmodified lines
865817 bool dc_dsc_compute_bandwidth_range(
866818 const struct display_stream_compressor *dsc,
867819 uint32_t dsc_min_slice_height_override,
868− uint32_t min_bpp,
869− uint32_t max_bpp,
820+ uint32_t min_bpp_x16,
821+ uint32_t max_bpp_x16,
870822 const struct dsc_dec_dpcd_caps *dsc_sink_caps,
871823 const struct dc_crtc_timing *timing,
872824 struct dc_dsc_bw_range *range)
@@ -883,10 +835,10 @@ bool dc_dsc_compute_bandwidth_range( · 10 unmodified lines
883835
884836 if (is_dsc_possible)
885837 is_dsc_possible = setup_dsc_config(dsc_sink_caps, &dsc_enc_caps, 0, timing,
886− dsc_min_slice_height_override, max_bpp, &config);
838+ dsc_min_slice_height_override, max_bpp_x16, &config);
887839
888840 if (is_dsc_possible)
889− get_dsc_bandwidth_range(min_bpp, max_bpp, &dsc_common_caps, timing, range);
841+ get_dsc_bandwidth_range(min_bpp_x16, max_bpp_x16, &dsc_common_caps, timing, range);
890842
891843 return is_dsc_possible;
892844 }
@@ -908,11 +860,20 @@ bool dc_dsc_compute_config( · 15 unmodified lines
908860 &dsc_enc_caps,
909861 target_bandwidth_kbps,
910862 timing, dsc_min_slice_height_override,
911− max_target_bpp_limit_override, dsc_cfg);
863+ max_target_bpp_limit_override * 16, dsc_cfg);
912864 return is_dsc_possible;
913865 }
914866
915−void dc_dsc_get_policy_for_timing(const struct dc_crtc_timing *timing, uint32_t max_target_bpp_limit_override, struct dc_dsc_policy *policy)
867+uint32_t dc_dsc_stream_bandwidth_in_kbps(uint32_t pix_clk_100hz, uint32_t bpp_x16)
868+{
869+ struct fixed31_32 link_bw_kbps;
870+ link_bw_kbps = dc_fixpt_from_int(pix_clk_100hz);
871+ link_bw_kbps = dc_fixpt_div_int(link_bw_kbps, 160);
872+ link_bw_kbps = dc_fixpt_mul_int(link_bw_kbps, bpp_x16);
873+ return dc_fixpt_ceil(link_bw_kbps);
874+}
875+
876+void dc_dsc_get_policy_for_timing(const struct dc_crtc_timing *timing, uint32_t max_target_bpp_limit_override_x16, struct dc_dsc_policy *policy)
916877 {
917878 uint32_t bpc = 0;
918879
@@ -967,13 +928,15 @@ void dc_dsc_get_policy_for_timing(const struct dc_crtc_timing *timing, uint32_t · 48 unmodified lines
967928 return;
968929 }
969930
931+ policy->preferred_bpp_x16 = timing->dsc_fixed_bits_per_pixel_x16;
932+
970933 /* internal upper limit, default 16 bpp */
971934 if (policy->max_target_bpp > dsc_policy_max_target_bpp_limit)
972935 policy->max_target_bpp = dsc_policy_max_target_bpp_limit;
973936
974937 /* apply override */
975− if (max_target_bpp_limit_override && policy->max_target_bpp > max_target_bpp_limit_override)
976− policy->max_target_bpp = max_target_bpp_limit_override;
938+ if (max_target_bpp_limit_override_x16 && policy->max_target_bpp > max_target_bpp_limit_override_x16 / 16)
939+ policy->max_target_bpp = max_target_bpp_limit_override_x16 / 16;
977940
978941 /* enable DSC when not needed, default false */
979942 if (dsc_policy_enable_dsc_when_not_needed)
drivers/gpu/drm/amd/display/dc/gpio/dce110/hw_factory_dce110.c+0−4
@@ -48,10 +48,6 @@ · 47 unmodified lines
4848 #define REGI(reg_name, block, id)\
4949 mm ## block ## id ## _ ## reg_name
5050
51−#include "../hw_gpio.h"
52−#include "../hw_ddc.h"
53−#include "../hw_hpd.h"
54−
5551 #include "reg_helper.h"
5652 #include "../hpd_regs.h"
5753
drivers/gpu/drm/amd/display/dc/hdcp/hdcp_msg.c+5−4
@@ -39,7 +39,7 @@ · 38 unmodified lines
3939 #define HDCP14_KSV_SIZE 5
4040 #define HDCP14_MAX_KSV_FIFO_SIZE 127*HDCP14_KSV_SIZE
4141
42−static const bool hdcp_cmd_is_read[] = {
42+static const bool hdcp_cmd_is_read[HDCP_MESSAGE_ID_MAX] = {
4343 [HDCP_MESSAGE_ID_READ_BKSV] = true,
4444 [HDCP_MESSAGE_ID_READ_RI_R0] = true,
4545 [HDCP_MESSAGE_ID_READ_PJ] = true,
@@ -75,7 +75,7 @@ static const bool hdcp_cmd_is_read[] = { · 29 unmodified lines
7575 [HDCP_MESSAGE_ID_WRITE_CONTENT_STREAM_TYPE] = false
7676 };
7777
78−static const uint8_t hdcp_i2c_offsets[] = {
78+static const uint8_t hdcp_i2c_offsets[HDCP_MESSAGE_ID_MAX] = {
7979 [HDCP_MESSAGE_ID_READ_BKSV] = 0x0,
8080 [HDCP_MESSAGE_ID_READ_RI_R0] = 0x8,
8181 [HDCP_MESSAGE_ID_READ_PJ] = 0xA,
@@ -106,7 +106,8 @@ static const uint8_t hdcp_i2c_offsets[] = { · 24 unmodified lines
106106 [HDCP_MESSAGE_ID_WRITE_REPEATER_AUTH_SEND_ACK] = 0x60,
107107 [HDCP_MESSAGE_ID_WRITE_REPEATER_AUTH_STREAM_MANAGE] = 0x60,
108108 [HDCP_MESSAGE_ID_READ_REPEATER_AUTH_STREAM_READY] = 0x80,
109− [HDCP_MESSAGE_ID_READ_RXSTATUS] = 0x70
109+ [HDCP_MESSAGE_ID_READ_RXSTATUS] = 0x70,
110+ [HDCP_MESSAGE_ID_WRITE_CONTENT_STREAM_TYPE] = 0x0,
110111 };
111112
112113 struct protection_properties {
@@ -184,7 +185,7 @@ static const struct protection_properties hdmi_14_protection = { · 71 unmodified lines
184185 .process_transaction = hdmi_14_process_transaction
185186 };
186187
187−static const uint32_t hdcp_dpcd_addrs[] = {
188+static const uint32_t hdcp_dpcd_addrs[HDCP_MESSAGE_ID_MAX] = {
188189 [HDCP_MESSAGE_ID_READ_BKSV] = 0x68000,
189190 [HDCP_MESSAGE_ID_READ_RI_R0] = 0x68005,
190191 [HDCP_MESSAGE_ID_READ_PJ] = 0xFFFFFFFF,
drivers/gpu/drm/amd/display/dc/inc/clock_source.h+5−0
@@ -170,6 +170,11 @@ struct clock_source_funcs { · 169 unmodified lines
170170 const struct clock_source *clock_source,
171171 unsigned int inst,
172172 unsigned int *pixel_clk_khz);
173+ bool (*override_dp_pix_clk)(
174+ struct clock_source *clock_source,
175+ unsigned int inst,
176+ unsigned int pixel_clk,
177+ unsigned int ref_clk);
173178 };
174179
175180 struct clock_source {
drivers/gpu/drm/amd/display/dc/inc/core_types.h+15−0
@@ -97,6 +97,10 @@ struct resource_funcs { · 96 unmodified lines
9797 const struct panel_cntl_init_data *panel_cntl_init_data);
9898 struct link_encoder *(*link_enc_create)(
9999 const struct encoder_init_data *init);
100+ /* Create a minimal link encoder object with no dc_link object
101+ * associated with it. */
102+ struct link_encoder *(*link_enc_create_minimal)(struct dc_context *ctx, enum engine_id eng_id);
103+
100104 bool (*validate_bandwidth)(
101105 struct dc *dc,
102106 struct dc_state *context,
@@ -106,6 +110,8 @@ struct resource_funcs { · 3 unmodified lines
106110 display_e2e_pipe_params_st *pipes,
107111 int pipe_cnt,
108112 int vlevel);
113+ void (*update_soc_for_wm_a)(
114+ struct dc *dc, struct dc_state *context);
109115 int (*populate_dml_pipes)(
110116 struct dc *dc,
111117 struct dc_state *context,
@@ -210,6 +216,15 @@ struct resource_pool { · 98 unmodified lines
210216 unsigned int underlay_pipe_index;
211217 unsigned int stream_enc_count;
212218
219+ /* An array for accessing the link encoder objects that have been created.
220+ * Index in array corresponds to engine ID - viz. 0: ENGINE_ID_DIGA
221+ */
222+ struct link_encoder *link_encoders[MAX_DIG_LINK_ENCODERS];
223+ /* Number of DIG link encoder objects created - i.e. number of valid
224+ * entries in link_encoders array.
225+ */
226+ unsigned int dig_link_enc_count;
227+
213228 #if defined(CONFIG_DRM_AMD_DC_DCN)
214229 struct dc_3dlut *mpc_lut[MAX_PIPES];
215230 struct dc_transfer_func *mpc_shaper[MAX_PIPES];
drivers/gpu/drm/amd/display/dc/inc/dc_link_ddc.h+2−1
@@ -56,6 +56,7 @@ struct dp_receiver_id_info; · 55 unmodified lines
5656
5757 struct i2c_payloads;
5858 struct aux_payloads;
59+enum aux_return_code_type;
5960
6061 void dal_ddc_i2c_payloads_add(
6162 struct i2c_payloads *payloads,
@@ -100,7 +101,7 @@ bool dal_ddc_submit_aux_command(struct ddc_service *ddc, · 38 unmodified lines
100101
101102 int dc_link_aux_transfer_raw(struct ddc_service *ddc,
102103 struct aux_payload *payload,
103− enum aux_channel_operation_result *operation_result);
104+ enum aux_return_code_type *operation_result);
104105
105106 bool dc_link_aux_transfer_with_retries(struct ddc_service *ddc,
106107 struct aux_payload *payload);
drivers/gpu/drm/amd/display/dc/inc/hw/aux_engine.h+3−1
@@ -29,6 +29,8 @@ · 28 unmodified lines
2929 #include "dc_ddc_types.h"
3030 #include "include/i2caux_interface.h"
3131
32+enum aux_return_code_type;
33+
3234 enum i2caux_transaction_operation {
3335 I2CAUX_TRANSACTION_READ,
3436 I2CAUX_TRANSACTION_WRITE
@@ -162,7 +164,7 @@ struct aux_engine_funcs { · 127 unmodified lines
162164 uint8_t *buffer,
163165 uint8_t *reply_result,
164166 uint32_t *sw_status);
165− enum aux_channel_operation_result (*get_channel_status)(
167+ enum aux_return_code_type (*get_channel_status)(
166168 struct aux_engine *engine,
167169 uint8_t *returned_bytes);
168170 bool (*is_engine_available)(struct aux_engine *engine);
drivers/gpu/drm/amd/display/dc/inc/hw/dmcu.h+21−0
@@ -56,6 +56,20 @@ struct dmcu { · 55 unmodified lines
5656 bool auto_load_dmcu;
5757 };
5858
59+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
60+struct crc_region {
61+ uint16_t x_start;
62+ uint16_t y_start;
63+ uint16_t x_end;
64+ uint16_t y_end;
65+};
66+
67+struct otg_phy_mux {
68+ uint8_t phy_output_num;
69+ uint8_t otg_output_num;
70+};
71+#endif
72+
5973 struct dmcu_funcs {
6074 bool (*dmcu_init)(struct dmcu *dmcu);
6175 bool (*load_iram)(struct dmcu *dmcu,
@@ -84,6 +98,13 @@ struct dmcu_funcs { · 22 unmodified lines
8498 int *min_frame_rate,
8599 int *max_frame_rate);
86100 bool (*recv_edid_cea_ack)(struct dmcu *dmcu, int *offset);
101+#if defined(CONFIG_DRM_AMD_SECURE_DISPLAY)
102+ void (*forward_crc_window)(struct dmcu *dmcu,
103+ struct crc_region *crc_win,
104+ struct otg_phy_mux *mux_mapping);
105+ void (*stop_crc_win_update)(struct dmcu *dmcu,
106+ struct otg_phy_mux *mux_mapping);
107+#endif
87108 };
88109
89110 #endif
drivers/gpu/drm/amd/display/dc/inc/hw/hw_shared.h+1−0
@@ -36,6 +36,7 @@ · 35 unmodified lines
3636
3737 #define MAX_AUDIOS 7
3838 #define MAX_PIPES 6
39+#define MAX_DIG_LINK_ENCODERS 7
3940 #define MAX_DWB_PIPES 1
4041
4142 struct gamma_curve {
drivers/gpu/drm/amd/display/dc/inc/hw/timing_generator.h+12−0
@@ -109,6 +109,12 @@ enum h_timing_div_mode { · 108 unmodified lines
109109 H_TIMING_DIV_BY4,
110110 };
111111
112+enum timing_synchronization_type {
113+ NOT_SYNCHRONIZABLE,
114+ TIMING_SYNCHRONIZABLE,
115+ VBLANK_SYNCHRONIZABLE
116+};
117+
112118 struct crc_params {
113119 /* Regions used to calculate CRC*/
114120 uint16_t windowa_x_start;
@@ -292,6 +298,12 @@ struct timing_generator_funcs { · 177 unmodified lines
292298 uint32_t window_start, uint32_t window_end);
293299 void (*set_vtotal_change_limit)(struct timing_generator *optc,
294300 uint32_t limit);
301+ void (*align_vblanks)(struct timing_generator *master_optc,
302+ struct timing_generator *slave_optc,
303+ uint32_t master_pixel_clock_100Hz,
304+ uint32_t slave_pixel_clock_100Hz,
305+ uint8_t master_clock_divider,
306+ uint8_t slave_clock_divider);
295307 };
296308
297309 #endif
drivers/gpu/drm/amd/display/dc/inc/hw_sequencer.h+3−4
@@ -111,6 +111,9 @@ struct hw_sequencer_funcs { · 110 unmodified lines
111111 void (*enable_timing_synchronization)(struct dc *dc,
112112 int group_index, int group_size,
113113 struct pipe_ctx *grouped_pipes[]);
114+ void (*enable_vblanks_synchronization)(struct dc *dc,
115+ int group_index, int group_size,
116+ struct pipe_ctx *grouped_pipes[]);
114117 void (*setup_periodic_interrupt)(struct dc *dc,
115118 struct pipe_ctx *pipe_ctx,
116119 enum vline_select vline);
@@ -231,10 +234,6 @@ struct hw_sequencer_funcs { · 114 unmodified lines
231234 enum dc_color_depth color_depth,
232235 const struct tg_color *solid_color,
233236 int width, int height, int offset);
234−
235− void (*set_hubp_blank)(const struct dc *dc,
236− struct pipe_ctx *pipe_ctx,
237− bool blank_enable);
238237 };
239238
240239 void color_space_to_black_color(
drivers/gpu/drm/amd/display/dc/inc/resource.h+5−0
@@ -48,6 +48,7 @@ struct resource_caps { · 47 unmodified lines
4848 int num_ddc;
4949 int num_vmid;
5050 int num_dsc;
51+ unsigned int num_dig_link_enc; // Total number of DIGs (digital encoders) in DIO (Display Input/Output).
5152 int num_mpc_3dlut;
5253 };
5354
@@ -115,6 +116,10 @@ bool resource_are_streams_timing_synchronizable( · 61 unmodified lines
115116 struct dc_stream_state *stream1,
116117 struct dc_stream_state *stream2);
117118
119+bool resource_are_vblanks_synchronizable(
120+ struct dc_stream_state *stream1,
121+ struct dc_stream_state *stream2);
122+
118123 struct clock_source *resource_find_used_clk_src_for_sharing(
119124 struct resource_context *res_ctx,
120125 struct pipe_ctx *pipe_ctx);
drivers/gpu/drm/amd/display/dc/irq/dcn20/irq_service_dcn20.c+31−0
@@ -58,6 +58,18 @@ enum dc_irq_source to_dal_irq_source_dcn20( · 57 unmodified lines
5858 return DC_IRQ_SOURCE_VBLANK5;
5959 case DCN_1_0__SRCID__DC_D6_OTG_VSTARTUP:
6060 return DC_IRQ_SOURCE_VBLANK6;
61+ case DCN_1_0__SRCID__OTG1_VERTICAL_INTERRUPT0_CONTROL:
62+ return DC_IRQ_SOURCE_DC1_VLINE0;
63+ case DCN_1_0__SRCID__OTG2_VERTICAL_INTERRUPT0_CONTROL:
64+ return DC_IRQ_SOURCE_DC2_VLINE0;
65+ case DCN_1_0__SRCID__OTG3_VERTICAL_INTERRUPT0_CONTROL:
66+ return DC_IRQ_SOURCE_DC3_VLINE0;
67+ case DCN_1_0__SRCID__OTG4_VERTICAL_INTERRUPT0_CONTROL:
68+ return DC_IRQ_SOURCE_DC4_VLINE0;
69+ case DCN_1_0__SRCID__OTG5_VERTICAL_INTERRUPT0_CONTROL:
70+ return DC_IRQ_SOURCE_DC5_VLINE0;
71+ case DCN_1_0__SRCID__OTG6_VERTICAL_INTERRUPT0_CONTROL:
72+ return DC_IRQ_SOURCE_DC6_VLINE0;
6173 case DCN_1_0__SRCID__HUBP0_FLIP_INTERRUPT:
6274 return DC_IRQ_SOURCE_PFLIP1;
6375 case DCN_1_0__SRCID__HUBP1_FLIP_INTERRUPT:
@@ -172,6 +184,11 @@ static const struct irq_source_info_funcs vupdate_no_lock_irq_info_funcs = { · 108 unmodified lines
172184 .ack = NULL
173185 };
174186
187+static const struct irq_source_info_funcs vline0_irq_info_funcs = {
188+ .set = NULL,
189+ .ack = NULL
190+};
191+
175192 #undef BASE_INNER
176193 #define BASE_INNER(seg) DCN_BASE__INST0_SEG ## seg
177194
@@ -245,6 +262,14 @@ static const struct irq_source_info_funcs vupdate_no_lock_irq_info_funcs = { · 67 unmodified lines
245262 .funcs = &vblank_irq_info_funcs\
246263 }
247264
265+#define vline0_int_entry(reg_num)\
266+ [DC_IRQ_SOURCE_DC1_VLINE0 + reg_num] = {\
267+ IRQ_REG_ENTRY(OTG, reg_num,\
268+ OTG_VERTICAL_INTERRUPT0_CONTROL, OTG_VERTICAL_INTERRUPT0_INT_ENABLE,\
269+ OTG_VERTICAL_INTERRUPT0_CONTROL, OTG_VERTICAL_INTERRUPT0_CLEAR),\
270+ .funcs = &vline0_irq_info_funcs\
271+ }
272+
248273 #define dummy_irq_entry() \
249274 {\
250275 .funcs = &dummy_irq_info_funcs\
@@ -353,6 +378,12 @@ irq_source_info_dcn20[DAL_IRQ_SOURCES_NUMBER] = { · 102 unmodified lines
353378 vblank_int_entry(3),
354379 vblank_int_entry(4),
355380 vblank_int_entry(5),
381+ vline0_int_entry(0),
382+ vline0_int_entry(1),
383+ vline0_int_entry(2),
384+ vline0_int_entry(3),
385+ vline0_int_entry(4),
386+ vline0_int_entry(5),
356387 };
357388
358389 static const struct irq_service_funcs irq_service_funcs_dcn20 = {
drivers/gpu/drm/amd/display/dc/irq/dcn21/irq_service_dcn21.c+34−0
@@ -58,6 +58,20 @@ enum dc_irq_source to_dal_irq_source_dcn21( · 57 unmodified lines
5858 return DC_IRQ_SOURCE_VBLANK5;
5959 case DCN_1_0__SRCID__DC_D6_OTG_VSTARTUP:
6060 return DC_IRQ_SOURCE_VBLANK6;
61+ case DCN_1_0__SRCID__DMCUB_OUTBOX_HIGH_PRIORITY_READY_INT:
62+ return DC_IRQ_SOURCE_DMCUB_OUTBOX0;
63+ case DCN_1_0__SRCID__OTG1_VERTICAL_INTERRUPT0_CONTROL:
64+ return DC_IRQ_SOURCE_DC1_VLINE0;
65+ case DCN_1_0__SRCID__OTG2_VERTICAL_INTERRUPT0_CONTROL:
66+ return DC_IRQ_SOURCE_DC2_VLINE0;
67+ case DCN_1_0__SRCID__OTG3_VERTICAL_INTERRUPT0_CONTROL:
68+ return DC_IRQ_SOURCE_DC3_VLINE0;
69+ case DCN_1_0__SRCID__OTG4_VERTICAL_INTERRUPT0_CONTROL:
70+ return DC_IRQ_SOURCE_DC4_VLINE0;
71+ case DCN_1_0__SRCID__OTG5_VERTICAL_INTERRUPT0_CONTROL:
72+ return DC_IRQ_SOURCE_DC5_VLINE0;
73+ case DCN_1_0__SRCID__OTG6_VERTICAL_INTERRUPT0_CONTROL:
74+ return DC_IRQ_SOURCE_DC6_VLINE0;
6175 case DCN_1_0__SRCID__HUBP0_FLIP_INTERRUPT:
6276 return DC_IRQ_SOURCE_PFLIP1;
6377 case DCN_1_0__SRCID__HUBP1_FLIP_INTERRUPT:
@@ -173,6 +187,12 @@ static const struct irq_source_info_funcs vupdate_no_lock_irq_info_funcs = { · 109 unmodified lines
173187 .ack = NULL
174188 };
175189
190+
191+static const struct irq_source_info_funcs vline0_irq_info_funcs = {
192+ .set = NULL,
193+ .ack = NULL
194+};
195+
176196 #undef BASE_INNER
177197 #define BASE_INNER(seg) DMU_BASE__INST0_SEG ## seg
178198
@@ -254,6 +274,14 @@ static const struct irq_source_info_funcs vupdate_no_lock_irq_info_funcs = { · 75 unmodified lines
254274 .funcs = &vblank_irq_info_funcs\
255275 }
256276
277+#define vline0_int_entry(reg_num)\
278+ [DC_IRQ_SOURCE_DC1_VLINE0 + reg_num] = {\
279+ IRQ_REG_ENTRY(OTG, reg_num,\
280+ OTG_VERTICAL_INTERRUPT0_CONTROL, OTG_VERTICAL_INTERRUPT0_INT_ENABLE,\
281+ OTG_VERTICAL_INTERRUPT0_CONTROL, OTG_VERTICAL_INTERRUPT0_CLEAR),\
282+ .funcs = &vline0_irq_info_funcs\
283+ }
284+
257285 #define dummy_irq_entry() \
258286 {\
259287 .funcs = &dummy_irq_info_funcs\
@@ -366,6 +394,12 @@ irq_source_info_dcn21[DAL_IRQ_SOURCES_NUMBER] = { · 106 unmodified lines
366394 vblank_int_entry(3),
367395 vblank_int_entry(4),

10 more lines aren’t shown here.

drivers/gpu/drm/amd/display/dc/irq/dcn30/irq_service_dcn30.c+38−0

This diff is large: its 66 lines aren’t shown here.

drivers/gpu/drm/amd/display/dc/irq/dcn302/irq_service_dcn302.c+30−0

This diff is large: its 58 lines aren’t shown here.

drivers/gpu/drm/amd/display/dc/irq_types.h+2−1

This diff is large: its 10 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/dmub_srv.h+55−0

This diff is large: its 104 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/dmub_srv_stat.h+41−0

This diff is large: its 42 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/inc/dmub_cmd.h+1326−169

This diff is large: its 2175 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/inc/dmub_trace_buffer.h+1−2

This diff is large: its 15 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/src/Makefile+1−1

This diff is large: its 9 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/src/dmub_dcn20.c+57−0

This diff is large: its 78 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/src/dmub_dcn20.h+33−2

This diff is large: its 78 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/src/dmub_dcn30.c+7−0

This diff is large: its 14 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/src/dmub_srv.c+79−5

This diff is large: its 155 lines aren’t shown here.

drivers/gpu/drm/amd/display/dmub/src/dmub_srv_stat.c+105−0

This diff is large: its 106 lines aren’t shown here.

drivers/gpu/drm/amd/display/modules/freesync/freesync.c+26−2

This diff is large: its 51 lines aren’t shown here.

drivers/gpu/drm/amd/display/modules/inc/mod_freesync.h+2−1

This diff is large: its 10 lines aren’t shown here.

drivers/gpu/drm/amd/include/aldebaran_ip_offset.h+1738−0

This diff is large: its 1739 lines aren’t shown here.

drivers/gpu/drm/amd/include/asic_reg/dcn/dcn_3_0_0_offset.h+1−1

This diff is large: its 9 lines aren’t shown here.

drivers/gpu/drm/amd/include/asic_reg/gc/gc_9_4_2_offset.h+7683−0

This diff is large: its 7684 lines aren’t shown here.

drivers/gpu/drm/amd/include/asic_reg/gc/gc_9_4_2_sh_mask.h+32949−0

This diff is large: its 32950 lines aren’t shown here.

drivers/gpu/drm/amd/include/asic_reg/mmhub/mmhub_1_7_offset.h+5125−0

This diff is large: its 5126 lines aren’t shown here.

This diff is larger than a page shows: the files after these aren’t listed. Clone the repository to see all of it.

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t