Final block of feature work for 4.11: - gen8 pd cleanup from Matthew Auld - more cleanups for view/vma (Chris) - dmc support on glk (Anusha Srivatsa) - use core crc api (Tomue) - track wedged requests using fence.error (Chris) - lots of psr fixes (Nagaraju, Vathsala) - dp mst support, acked for merging through drm-intel by Takashi (Libin) - huc loading support, including uapi for libva to use it (Anusha Srivatsa) * tag 'drm-intel-next-2017-01-23' of git://anongit.freedesktop.org/git/drm-intel: (111 commits) drm/i915: Update DRIVER_DATE to 20170123 drm/i915: reinstate call to trace_i915_vma_bind drm/i915: Assert that created vma has a whole number of pages drm/i915: Assert the drm_mm_node is allocated when on the VM lists drm/i915: Treat an error from i915_vma_instance() as unlikely drm/i915: Reject vma creation larger than address space drm/i915: Use common LRU inactive vma bumping for unpin_from_display drm/i915: Do an unlocked wait before set-cache-level ioctl drm/i915/huc: Assert that HuC vma is placed in GuC accessible range drm/i915/huc: Avoid attempting to authenticate non-existent fw drm/i915: Set adjustment to zero on Up/Down interrupts if freq is already max/min drm/i915: Remove the double handling of 'flags from intel_mode_from_pipe_config() drm/i915: Remove crtc->config usage from intel_modeset_readout_hw_state() drm/i915: Release temporary load-detect state upon switching drm/i915: Remove i915_gem_object_to_ggtt() drm/i915: Remove i915_vma_create from VMA API drm/i915: Add a check that the VMA instance we lookup matches the request drm/i915: Rename some warts in the VMA API drm/i915: Track pinned vma in intel_plane_state drm/i915/get_params: Add HuC status to getparams ...
Unified · Split
| @@ -19,6 +19,23 @@ PCM · 18 unmodified lines | ||
| 19 | 19 | === |
| 20 | 20 | To be added |
| 21 | 21 | |
| 22 | +Pin Initialization | |
| 23 | +================== | |
| 24 | +Each pin may have several device entries (virtual pins). On Intel platform, | |
| 25 | +the device entries number is dynamically changed. If DP MST hub is connected, | |
| 26 | +it is in DP MST mode, and the device entries number is 3. Otherwise, the | |
| 27 | +device entries number is 1. | |
| 28 | + | |
| 29 | +To simplify the implementation, all the device entries will be initialized | |
| 30 | +when bootup no matter whether it is in DP MST mode or not. | |
| 31 | + | |
| 32 | +Connection list | |
| 33 | +=============== | |
| 34 | +DP MST reuses connection list code. The code can be reused because | |
| 35 | +device entries on the same pin have the same connection list. | |
| 36 | + | |
| 37 | +This means DP MST gets the device entry connection list without the | |
| 38 | +device entry setting. | |
| 22 | 39 | |
| 23 | 40 | Jack |
| 24 | 41 | ==== |
| @@ -56,7 +56,9 @@ i915-y += i915_cmd_parser.o \ · 55 unmodified lines | ||
| 56 | 56 | |
| 57 | 57 | # general-purpose microcontroller (GuC) support |
| 58 | 58 | i915-y += intel_uc.o \ |
| 59 | + intel_guc_log.o \ | |
| 59 | 60 | intel_guc_loader.o \ |
| 61 | + intel_huc.o \ | |
| 60 | 62 | i915_guc_submission.o |
| 61 | 63 | |
| 62 | 64 | # autogenerated null render state |
| @@ -41,47 +41,34 @@ static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm) · 40 unmodified lines | ||
| 41 | 41 | { |
| 42 | 42 | struct intel_gvt *gvt = vgpu->gvt; |
| 43 | 43 | struct drm_i915_private *dev_priv = gvt->dev_priv; |
| 44 | − u32 alloc_flag, search_flag; | |
| 44 | + unsigned int flags; | |
| 45 | 45 | u64 start, end, size; |
| 46 | 46 | struct drm_mm_node *node; |
| 47 | − int retried = 0; | |
| 48 | 47 | int ret; |
| 49 | 48 | |
| 50 | 49 | if (high_gm) { |
| 51 | − search_flag = DRM_MM_SEARCH_BELOW; | |
| 52 | − alloc_flag = DRM_MM_CREATE_TOP; | |
| 53 | 50 | node = &vgpu->gm.high_gm_node; |
| 54 | 51 | size = vgpu_hidden_sz(vgpu); |
| 55 | 52 | start = gvt_hidden_gmadr_base(gvt); |
| 56 | 53 | end = gvt_hidden_gmadr_end(gvt); |
| 54 | + flags = PIN_HIGH; | |
| 57 | 55 | } else { |
| 58 | − search_flag = DRM_MM_SEARCH_DEFAULT; | |
| 59 | − alloc_flag = DRM_MM_CREATE_DEFAULT; | |
| 60 | 56 | node = &vgpu->gm.low_gm_node; |
| 61 | 57 | size = vgpu_aperture_sz(vgpu); |
| 62 | 58 | start = gvt_aperture_gmadr_base(gvt); |
| 63 | 59 | end = gvt_aperture_gmadr_end(gvt); |
| 60 | + flags = PIN_MAPPABLE; | |
| 64 | 61 | } |
| 65 | 62 | |
| 66 | 63 | mutex_lock(&dev_priv->drm.struct_mutex); |
| 67 | −search_again: | |
| 68 | − ret = drm_mm_insert_node_in_range_generic(&dev_priv->ggtt.base.mm, | |
| 69 | − node, size, 4096, | |
| 70 | − I915_COLOR_UNEVICTABLE, | |
| 71 | − start, end, search_flag, | |
| 72 | − alloc_flag); | |
| 73 | − if (ret) { | |
| 74 | − ret = i915_gem_evict_something(&dev_priv->ggtt.base, | |
| 75 | − size, 4096, | |
| 76 | − I915_COLOR_UNEVICTABLE, | |
| 77 | − start, end, 0); | |
| 78 | − if (ret == 0 && ++retried < 3) | |
| 79 | − goto search_again; | |
| 80 | − | |
| 81 | − gvt_err("fail to alloc %s gm space from host, retried %d\n", | |
| 82 | − high_gm ? "high" : "low", retried); | |
| 83 | − } | |
| 64 | + ret = i915_gem_gtt_insert(&dev_priv->ggtt.base, node, | |
| 65 | + size, 4096, I915_COLOR_UNEVICTABLE, | |
| 66 | + start, end, flags); | |
| 84 | 67 | mutex_unlock(&dev_priv->drm.struct_mutex); |
| 68 | + if (ret) | |
| 69 | + gvt_err("fail to alloc %s gm space from host\n", | |
| 70 | + high_gm ? "high" : "low"); | |
| 71 | + | |
| 85 | 72 | return ret; |
| 86 | 73 | } |
| 87 | 74 | |
| @@ -159,8 +159,35 @@ describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj) · 158 unmodified lines | ||
| 159 | 159 | seq_printf(m, " (%sgtt offset: %08llx, size: %08llx", |
| 160 | 160 | i915_vma_is_ggtt(vma) ? "g" : "pp", |
| 161 | 161 | vma->node.start, vma->node.size); |
| 162 | − if (i915_vma_is_ggtt(vma)) | |
| 163 | − seq_printf(m, ", type: %u", vma->ggtt_view.type); | |
| 162 | + if (i915_vma_is_ggtt(vma)) { | |
| 163 | + switch (vma->ggtt_view.type) { | |
| 164 | + case I915_GGTT_VIEW_NORMAL: | |
| 165 | + seq_puts(m, ", normal"); | |
| 166 | + break; | |
| 167 | + | |
| 168 | + case I915_GGTT_VIEW_PARTIAL: | |
| 169 | + seq_printf(m, ", partial [%08llx+%x]", | |
| 170 | + vma->ggtt_view.partial.offset << PAGE_SHIFT, | |
| 171 | + vma->ggtt_view.partial.size << PAGE_SHIFT); | |
| 172 | + break; | |
| 173 | + | |
| 174 | + case I915_GGTT_VIEW_ROTATED: | |
| 175 | + seq_printf(m, ", rotated [(%ux%u, stride=%u, offset=%u), (%ux%u, stride=%u, offset=%u)]", | |
| 176 | + vma->ggtt_view.rotated.plane[0].width, | |
| 177 | + vma->ggtt_view.rotated.plane[0].height, | |
| 178 | + vma->ggtt_view.rotated.plane[0].stride, | |
| 179 | + vma->ggtt_view.rotated.plane[0].offset, | |
| 180 | + vma->ggtt_view.rotated.plane[1].width, | |
| 181 | + vma->ggtt_view.rotated.plane[1].height, | |
| 182 | + vma->ggtt_view.rotated.plane[1].stride, | |
| 183 | + vma->ggtt_view.rotated.plane[1].offset); | |
| 184 | + break; | |
| 185 | + | |
| 186 | + default: | |
| 187 | + MISSING_CASE(vma->ggtt_view.type); | |
| 188 | + break; | |
| 189 | + } | |
| 190 | + } | |
| 164 | 191 | if (vma->fence) |
| 165 | 192 | seq_printf(m, " , fence: %d%s", |
| 166 | 193 | vma->fence->id, |
| @@ -2325,10 +2352,40 @@ static int i915_llc(struct seq_file *m, void *data) · 2158 unmodified lines | ||
| 2325 | 2352 | return 0; |
| 2326 | 2353 | } |
| 2327 | 2354 | |
| 2355 | +static int i915_huc_load_status_info(struct seq_file *m, void *data) | |
| 2356 | +{ | |
| 2357 | + struct drm_i915_private *dev_priv = node_to_i915(m->private); | |
| 2358 | + struct intel_uc_fw *huc_fw = &dev_priv->huc.fw; | |
| 2359 | + | |
| 2360 | + if (!HAS_HUC_UCODE(dev_priv)) | |
| 2361 | + return 0; | |
| 2362 | + | |
| 2363 | + seq_puts(m, "HuC firmware status:\n"); | |
| 2364 | + seq_printf(m, "\tpath: %s\n", huc_fw->path); | |
| 2365 | + seq_printf(m, "\tfetch: %s\n", | |
| 2366 | + intel_uc_fw_status_repr(huc_fw->fetch_status)); | |
| 2367 | + seq_printf(m, "\tload: %s\n", | |
| 2368 | + intel_uc_fw_status_repr(huc_fw->load_status)); | |
| 2369 | + seq_printf(m, "\tversion wanted: %d.%d\n", | |
| 2370 | + huc_fw->major_ver_wanted, huc_fw->minor_ver_wanted); | |
| 2371 | + seq_printf(m, "\tversion found: %d.%d\n", | |
| 2372 | + huc_fw->major_ver_found, huc_fw->minor_ver_found); | |
| 2373 | + seq_printf(m, "\theader: offset is %d; size = %d\n", | |
| 2374 | + huc_fw->header_offset, huc_fw->header_size); | |
| 2375 | + seq_printf(m, "\tuCode: offset is %d; size = %d\n", | |
| 2376 | + huc_fw->ucode_offset, huc_fw->ucode_size); | |
| 2377 | + seq_printf(m, "\tRSA: offset is %d; size = %d\n", | |
| 2378 | + huc_fw->rsa_offset, huc_fw->rsa_size); | |
| 2379 | + | |
| 2380 | + seq_printf(m, "\nHuC status 0x%08x:\n", I915_READ(HUC_STATUS2)); | |
| 2381 | + | |
| 2382 | + return 0; | |
| 2383 | +} | |
| 2384 | + | |
| 2328 | 2385 | static int i915_guc_load_status_info(struct seq_file *m, void *data) |
| 2329 | 2386 | { |
| 2330 | 2387 | struct drm_i915_private *dev_priv = node_to_i915(m->private); |
| 2331 | − struct intel_guc_fw *guc_fw = &dev_priv->guc.guc_fw; | |
| 2388 | + struct intel_uc_fw *guc_fw = &dev_priv->guc.fw; | |
| 2332 | 2389 | u32 tmp, i; |
| 2333 | 2390 | |
| 2334 | 2391 | if (!HAS_GUC_UCODE(dev_priv)) |
| @@ -2336,15 +2393,15 @@ static int i915_guc_load_status_info(struct seq_file *m, void *data) · 1 unmodified line | ||
| 2336 | 2393 | |
| 2337 | 2394 | seq_printf(m, "GuC firmware status:\n"); |
| 2338 | 2395 | seq_printf(m, "\tpath: %s\n", |
| 2339 | − guc_fw->guc_fw_path); | |
| 2396 | + guc_fw->path); | |
| 2340 | 2397 | seq_printf(m, "\tfetch: %s\n", |
| 2341 | − intel_guc_fw_status_repr(guc_fw->guc_fw_fetch_status)); | |
| 2398 | + intel_uc_fw_status_repr(guc_fw->fetch_status)); | |
| 2342 | 2399 | seq_printf(m, "\tload: %s\n", |
| 2343 | − intel_guc_fw_status_repr(guc_fw->guc_fw_load_status)); | |
| 2400 | + intel_uc_fw_status_repr(guc_fw->load_status)); | |
| 2344 | 2401 | seq_printf(m, "\tversion wanted: %d.%d\n", |
| 2345 | − guc_fw->guc_fw_major_wanted, guc_fw->guc_fw_minor_wanted); | |
| 2402 | + guc_fw->major_ver_wanted, guc_fw->minor_ver_wanted); | |
| 2346 | 2403 | seq_printf(m, "\tversion found: %d.%d\n", |
| 2347 | − guc_fw->guc_fw_major_found, guc_fw->guc_fw_minor_found); | |
| 2404 | + guc_fw->major_ver_found, guc_fw->minor_ver_found); | |
| 2348 | 2405 | seq_printf(m, "\theader: offset is %d; size = %d\n", |
| 2349 | 2406 | guc_fw->header_offset, guc_fw->header_size); |
| 2350 | 2407 | seq_printf(m, "\tuCode: offset is %d; size = %d\n", |
| @@ -2532,6 +2589,29 @@ DEFINE_SIMPLE_ATTRIBUTE(i915_guc_log_control_fops, · 181 unmodified lines | ||
| 2532 | 2589 | i915_guc_log_control_get, i915_guc_log_control_set, |
| 2533 | 2590 | "%lld\n"); |
| 2534 | 2591 | |
| 2592 | +static const char *psr2_live_status(u32 val) | |
| 2593 | +{ | |
| 2594 | + static const char * const live_status[] = { | |
| 2595 | + "IDLE", | |
| 2596 | + "CAPTURE", | |
| 2597 | + "CAPTURE_FS", | |
| 2598 | + "SLEEP", | |
| 2599 | + "BUFON_FW", | |
| 2600 | + "ML_UP", | |
| 2601 | + "SU_STANDBY", | |
| 2602 | + "FAST_SLEEP", | |
| 2603 | + "DEEP_SLEEP", | |
| 2604 | + "BUF_ON", | |
| 2605 | + "TG_ON" | |
| 2606 | + }; | |
| 2607 | + | |
| 2608 | + val = (val & EDP_PSR2_STATUS_STATE_MASK) >> EDP_PSR2_STATUS_STATE_SHIFT; | |
| 2609 | + if (val < ARRAY_SIZE(live_status)) | |
| 2610 | + return live_status[val]; | |
| 2611 | + | |
| 2612 | + return "unknown"; | |
| 2613 | +} | |
| 2614 | + | |
| 2535 | 2615 | static int i915_edp_psr_status(struct seq_file *m, void *data) |
| 2536 | 2616 | { |
| 2537 | 2617 | struct drm_i915_private *dev_priv = node_to_i915(m->private); |
| @@ -2606,6 +2686,12 @@ static int i915_edp_psr_status(struct seq_file *m, void *data) · 68 unmodified lines | ||
| 2606 | 2686 | |
| 2607 | 2687 | seq_printf(m, "Performance_Counter: %u\n", psrperf); |
| 2608 | 2688 | } |
| 2689 | + if (dev_priv->psr.psr2_support) { | |
| 2690 | + u32 psr2 = I915_READ(EDP_PSR2_STATUS_CTL); | |
| 2691 | + | |
| 2692 | + seq_printf(m, "EDP_PSR2_STATUS_CTL: %x [%s]\n", | |
| 2693 | + psr2, psr2_live_status(psr2)); | |
| 2694 | + } | |
| 2609 | 2695 | mutex_unlock(&dev_priv->psr.lock); |
| 2610 | 2696 | |
| 2611 | 2697 | intel_runtime_pm_put(dev_priv); |
| @@ -4553,6 +4639,7 @@ static const struct drm_info_list i915_debugfs_list[] = { · 1941 unmodified lines | ||
| 4553 | 4639 | {"i915_guc_info", i915_guc_info, 0}, |
| 4554 | 4640 | {"i915_guc_load_status", i915_guc_load_status_info, 0}, |
| 4555 | 4641 | {"i915_guc_log_dump", i915_guc_log_dump, 0}, |
| 4642 | + {"i915_huc_load_status", i915_huc_load_status_info, 0}, | |
| 4556 | 4643 | {"i915_frequency_info", i915_frequency_info, 0}, |
| 4557 | 4644 | {"i915_hangcheck_info", i915_hangcheck_info, 0}, |
| 4558 | 4645 | {"i915_drpc_info", i915_drpc_info, 0}, |
| @@ -49,6 +49,7 @@ · 48 unmodified lines | ||
| 49 | 49 | #include "i915_trace.h" |
| 50 | 50 | #include "i915_vgpu.h" |
| 51 | 51 | #include "intel_drv.h" |
| 52 | +#include "intel_uc.h" | |
| 52 | 53 | |
| 53 | 54 | static struct drm_driver driver; |
| 54 | 55 | |
| @@ -315,6 +316,12 @@ static int i915_getparam(struct drm_device *dev, void *data, · 260 unmodified lines | ||
| 315 | 316 | case I915_PARAM_MIN_EU_IN_POOL: |
| 316 | 317 | value = INTEL_INFO(dev_priv)->sseu.min_eu_in_pool; |
| 317 | 318 | break; |
| 319 | + case I915_PARAM_HUC_STATUS: | |
| 320 | + /* The register is already force-woken. We dont need | |
| 321 | + * any rpm here | |
| 322 | + */ | |
| 323 | + value = I915_READ(HUC_STATUS2) & HUC_FW_VERIFIED; | |
| 324 | + break; | |
| 318 | 325 | case I915_PARAM_MMAP_GTT_VERSION: |
| 319 | 326 | /* Though we've started our numbering from 1, and so class all |
| 320 | 327 | * earlier versions as 0, in effect their value is undefined as |
| @@ -599,6 +606,7 @@ static int i915_load_modeset_init(struct drm_device *dev) · 278 unmodified lines | ||
| 599 | 606 | if (ret) |
| 600 | 607 | goto cleanup_irq; |
| 601 | 608 | |
| 609 | + intel_huc_init(dev_priv); | |
| 602 | 610 | intel_guc_init(dev_priv); |
| 603 | 611 | |
| 604 | 612 | ret = i915_gem_init(dev_priv); |
| @@ -627,6 +635,7 @@ cleanup_gem: · 22 unmodified lines | ||
| 627 | 635 | i915_gem_fini(dev_priv); |
| 628 | 636 | cleanup_irq: |
| 629 | 637 | intel_guc_fini(dev_priv); |
| 638 | + intel_huc_fini(dev_priv); | |
| 630 | 639 | drm_irq_uninstall(dev); |
| 631 | 640 | intel_teardown_gmbus(dev_priv); |
| 632 | 641 | cleanup_csr: |
| @@ -1114,7 +1123,7 @@ static void i915_driver_register(struct drm_i915_private *dev_priv) · 481 unmodified lines | ||
| 1114 | 1123 | /* Reveal our presence to userspace */ |
| 1115 | 1124 | if (drm_dev_register(dev, 0) == 0) { |
| 1116 | 1125 | i915_debugfs_register(dev_priv); |
| 1117 | − i915_guc_register(dev_priv); | |
| 1126 | + i915_guc_log_register(dev_priv); | |
| 1118 | 1127 | i915_setup_sysfs(dev_priv); |
| 1119 | 1128 | |
| 1120 | 1129 | /* Depends on sysfs having been initialized */ |
| @@ -1158,7 +1167,7 @@ static void i915_driver_unregister(struct drm_i915_private *dev_priv) · 37 unmodified lines | ||
| 1158 | 1167 | i915_perf_unregister(dev_priv); |
| 1159 | 1168 | |
| 1160 | 1169 | i915_teardown_sysfs(dev_priv); |
| 1161 | − i915_guc_unregister(dev_priv); | |
| 1170 | + i915_guc_log_unregister(dev_priv); | |
| 1162 | 1171 | i915_debugfs_unregister(dev_priv); |
| 1163 | 1172 | drm_dev_unregister(&dev_priv->drm); |
| 1164 | 1173 | |
| @@ -1314,6 +1323,7 @@ void i915_driver_unload(struct drm_device *dev) · 149 unmodified lines | ||
| 1314 | 1323 | drain_workqueue(dev_priv->wq); |
| 1315 | 1324 | |
| 1316 | 1325 | intel_guc_fini(dev_priv); |
| 1326 | + intel_huc_fini(dev_priv); | |
| 1317 | 1327 | i915_gem_fini(dev_priv); |
| 1318 | 1328 | intel_fbc_cleanup_cfb(dev_priv); |
| 1319 | 1329 | |
| @@ -1471,7 +1481,7 @@ static int i915_drm_suspend_late(struct drm_device *dev, bool hibernation) · 151 unmodified lines | ||
| 1471 | 1481 | |
| 1472 | 1482 | intel_display_set_init_power(dev_priv, false); |
| 1473 | 1483 | |
| 1474 | − fw_csr = !IS_BROXTON(dev_priv) && | |
| 1484 | + fw_csr = !IS_GEN9_LP(dev_priv) && | |
| 1475 | 1485 | suspend_to_idle(dev_priv) && dev_priv->csr.dmc_payload; |
| 1476 | 1486 | /* |
| 1477 | 1487 | * In case of firmware assisted context save/restore don't manually |
| @@ -1484,7 +1494,7 @@ static int i915_drm_suspend_late(struct drm_device *dev, bool hibernation) · 6 unmodified lines | ||
| 1484 | 1494 | intel_power_domains_suspend(dev_priv); |
| 1485 | 1495 | |
| 1486 | 1496 | ret = 0; |
| 1487 | − if (IS_BROXTON(dev_priv)) | |
| 1497 | + if (IS_GEN9_LP(dev_priv)) | |
| 1488 | 1498 | bxt_enable_dc9(dev_priv); |
| 1489 | 1499 | else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
| 1490 | 1500 | hsw_enable_pc8(dev_priv); |
| @@ -1692,7 +1702,7 @@ static int i915_drm_resume_early(struct drm_device *dev) · 201 unmodified lines | ||
| 1692 | 1702 | |
| 1693 | 1703 | intel_uncore_early_sanitize(dev_priv, true); |
| 1694 | 1704 | |
| 1695 | − if (IS_BROXTON(dev_priv)) { | |
| 1705 | + if (IS_GEN9_LP(dev_priv)) { | |
| 1696 | 1706 | if (!dev_priv->suspended_to_idle) |
| 1697 | 1707 | gen9_sanitize_dc_state(dev_priv); |
| 1698 | 1708 | bxt_disable_dc9(dev_priv); |
| @@ -1702,7 +1712,7 @@ static int i915_drm_resume_early(struct drm_device *dev) · 3 unmodified lines | ||
| 1702 | 1712 | |
| 1703 | 1713 | intel_uncore_sanitize(dev_priv); |
| 1704 | 1714 | |
| 1705 | − if (IS_BROXTON(dev_priv) || | |
| 1715 | + if (IS_GEN9_LP(dev_priv) || | |
| 1706 | 1716 | !(dev_priv->suspended_to_idle && dev_priv->csr.dmc_payload)) |
| 1707 | 1717 | intel_power_domains_init_hw(dev_priv, true); |
| 1708 | 1718 | |
| @@ -1728,25 +1738,9 @@ static int i915_resume_switcheroo(struct drm_device *dev) · 19 unmodified lines | ||
| 1728 | 1738 | return i915_drm_resume(dev); |
| 1729 | 1739 | } |
| 1730 | 1740 | |
| 1731 | −static void disable_engines_irq(struct drm_i915_private *dev_priv) | |
| 1732 | −{ | |
| 1733 | − struct intel_engine_cs *engine; | |
| 1734 | − enum intel_engine_id id; | |
| 1735 | − | |
| 1736 | − /* Ensure irq handler finishes, and not run again. */ | |
| 1737 | − disable_irq(dev_priv->drm.irq); | |
| 1738 | − for_each_engine(engine, dev_priv, id) | |
| 1739 | − tasklet_kill(&engine->irq_tasklet); | |
| 1740 | −} | |
| 1741 | − | |
| 1742 | −static void enable_engines_irq(struct drm_i915_private *dev_priv) | |
| 1743 | −{ | |
| 1744 | − enable_irq(dev_priv->drm.irq); | |
| 1745 | −} | |
| 1746 | − | |
| 1747 | 1741 | /** |
| 1748 | 1742 | * i915_reset - reset chip after a hang |
| 1749 | − * @dev: drm device to reset | |
| 1743 | + * @dev_priv: device private to reset | |
| 1750 | 1744 | * |
| 1751 | 1745 | * Reset the chip. Useful if a hang is detected. Marks the device as wedged |
| 1752 | 1746 | * on failure. |
| @@ -1776,12 +1770,15 @@ void i915_reset(struct drm_i915_private *dev_priv) · 23 unmodified lines | ||
| 1776 | 1770 | error->reset_count++; |
| 1777 | 1771 | |
| 1778 | 1772 | pr_notice("drm/i915: Resetting chip after gpu hang\n"); |
| 1779 | − i915_gem_reset_prepare(dev_priv); | |
| 1773 | + disable_irq(dev_priv->drm.irq); | |
| 1774 | + ret = i915_gem_reset_prepare(dev_priv); | |
| 1775 | + if (ret) { | |
| 1776 | + DRM_ERROR("GPU recovery failed\n"); | |
| 1777 | + intel_gpu_reset(dev_priv, ALL_ENGINES); | |
| 1778 | + goto error; | |
| 1779 | + } | |
| 1780 | 1780 | |
| 1781 | − disable_engines_irq(dev_priv); | |
| 1782 | 1781 | ret = intel_gpu_reset(dev_priv, ALL_ENGINES); |
| 1783 | − enable_engines_irq(dev_priv); | |
| 1784 | − | |
| 1785 | 1782 | if (ret) { |
| 1786 | 1783 | if (ret != -ENODEV) |
| 1787 | 1784 | DRM_ERROR("Failed to reset chip: %i\n", ret); |
| @@ -1816,6 +1813,7 @@ void i915_reset(struct drm_i915_private *dev_priv) · 28 unmodified lines | ||
| 1816 | 1813 | i915_queue_hangcheck(dev_priv); |
| 1817 | 1814 | |
| 1818 | 1815 | wakeup: |
| 1816 | + enable_irq(dev_priv->drm.irq); | |
| 1819 | 1817 | wake_up_bit(&error->flags, I915_RESET_IN_PROGRESS); |
| 1820 | 1818 | return; |
| 1821 | 1819 | |
| @@ -2326,7 +2324,7 @@ static int intel_runtime_suspend(struct device *kdev) · 504 unmodified lines | ||
| 2326 | 2324 | intel_runtime_pm_disable_interrupts(dev_priv); |
| 2327 | 2325 | |
| 2328 | 2326 | ret = 0; |
| 2329 | − if (IS_BROXTON(dev_priv)) { | |
| 2327 | + if (IS_GEN9_LP(dev_priv)) { | |
| 2330 | 2328 | bxt_display_core_uninit(dev_priv); |
| 2331 | 2329 | bxt_enable_dc9(dev_priv); |
| 2332 | 2330 | } else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) { |
| @@ -2411,7 +2409,7 @@ static int intel_runtime_resume(struct device *kdev) · 78 unmodified lines | ||
| 2411 | 2409 | if (IS_GEN6(dev_priv)) |
| 2412 | 2410 | intel_init_pch_refclk(dev_priv); |
| 2413 | 2411 | |
| 2414 | − if (IS_BROXTON(dev_priv)) { | |
| 2412 | + if (IS_GEN9_LP(dev_priv)) { | |
| 2415 | 2413 | bxt_disable_dc9(dev_priv); |
| 2416 | 2414 | bxt_display_core_init(dev_priv, true); |
| 2417 | 2415 | if (dev_priv->csr.dmc_payload && |
| @@ -2549,8 +2547,8 @@ static const struct drm_ioctl_desc i915_ioctls[] = { · 131 unmodified lines | ||
| 2549 | 2547 | DRM_IOCTL_DEF_DRV(I915_GEM_MMAP_GTT, i915_gem_mmap_gtt_ioctl, DRM_RENDER_ALLOW), |
| 2550 | 2548 | DRM_IOCTL_DEF_DRV(I915_GEM_SET_DOMAIN, i915_gem_set_domain_ioctl, DRM_RENDER_ALLOW), |
| 2551 | 2549 | DRM_IOCTL_DEF_DRV(I915_GEM_SW_FINISH, i915_gem_sw_finish_ioctl, DRM_RENDER_ALLOW), |
| 2552 | − DRM_IOCTL_DEF_DRV(I915_GEM_SET_TILING, i915_gem_set_tiling, DRM_RENDER_ALLOW), | |
| 2553 | − DRM_IOCTL_DEF_DRV(I915_GEM_GET_TILING, i915_gem_get_tiling, DRM_RENDER_ALLOW), | |
| 2550 | + DRM_IOCTL_DEF_DRV(I915_GEM_SET_TILING, i915_gem_set_tiling_ioctl, DRM_RENDER_ALLOW), | |
| 2551 | + DRM_IOCTL_DEF_DRV(I915_GEM_GET_TILING, i915_gem_get_tiling_ioctl, DRM_RENDER_ALLOW), | |
| 2554 | 2552 | DRM_IOCTL_DEF_DRV(I915_GEM_GET_APERTURE, i915_gem_get_aperture_ioctl, DRM_RENDER_ALLOW), |
| 2555 | 2553 | DRM_IOCTL_DEF_DRV(I915_GET_PIPE_FROM_CRTC_ID, intel_get_pipe_from_crtc_id, 0), |
| 2556 | 2554 | DRM_IOCTL_DEF_DRV(I915_GEM_MADVISE, i915_gem_madvise_ioctl, DRM_RENDER_ALLOW), |
| @@ -78,8 +78,8 @@ · 77 unmodified lines | ||
| 78 | 78 | |
| 79 | 79 | #define DRIVER_NAME "i915" |
| 80 | 80 | #define DRIVER_DESC "Intel Graphics" |
| 81 | −#define DRIVER_DATE "20170109" | |
| 82 | −#define DRIVER_TIMESTAMP 1483953121 | |
| 81 | +#define DRIVER_DATE "20170123" | |
| 82 | +#define DRIVER_TIMESTAMP 1485156432 | |
| 83 | 83 | |
| 84 | 84 | #undef WARN_ON |
| 85 | 85 | /* Many gcc seem to no see through this and fall over :( */ |
| @@ -1069,6 +1069,8 @@ struct intel_fbc { · 983 unmodified lines | ||
| 1069 | 1069 | struct work_struct underrun_work; |
| 1070 | 1070 | |
| 1071 | 1071 | struct intel_fbc_state_cache { |
| 1072 | + struct i915_vma *vma; | |
| 1073 | + | |
| 1072 | 1074 | struct { |
| 1073 | 1075 | unsigned int mode_flags; |
| 1074 | 1076 | uint32_t hsw_bdw_pixel_rate; |
| @@ -1082,15 +1084,14 @@ struct intel_fbc { · 7 unmodified lines | ||
| 1082 | 1084 | } plane; |
| 1083 | 1085 | |
| 1084 | 1086 | struct { |
| 1085 | − u64 ilk_ggtt_offset; | |
| 1086 | 1087 | const struct drm_format_info *format; |
| 1087 | 1088 | unsigned int stride; |
| 1088 | − int fence_reg; | |
| 1089 | − unsigned int tiling_mode; | |
| 1090 | 1089 | } fb; |
| 1091 | 1090 | } state_cache; |
| 1092 | 1091 | |
| 1093 | 1092 | struct intel_fbc_reg_params { |
| 1093 | + struct i915_vma *vma; | |
| 1094 | + | |
| 1094 | 1095 | struct { |
| 1095 | 1096 | enum pipe pipe; |
| 1096 | 1097 | enum plane plane; |
| @@ -1098,10 +1099,8 @@ struct intel_fbc { · 1 unmodified line | ||
| 1098 | 1099 | } crtc; |
| 1099 | 1100 | |
| 1100 | 1101 | struct { |
| 1101 | − u64 ggtt_offset; | |
| 1102 | 1102 | const struct drm_format_info *format; |
| 1103 | 1103 | unsigned int stride; |
| 1104 | − int fence_reg; | |
| 1105 | 1104 | } fb; |
| 1106 | 1105 | |
| 1107 | 1106 | int cfb_size; |
| @@ -1154,6 +1153,9 @@ struct i915_psr { · 46 unmodified lines | ||
| 1154 | 1153 | bool psr2_support; |
| 1155 | 1154 | bool aux_frame_sync; |
| 1156 | 1155 | bool link_standby; |
| 1156 | + bool y_cord_support; | |
| 1157 | + bool colorimetry_support; | |
| 1158 | + bool alpm; | |
| 1157 | 1159 | }; |
| 1158 | 1160 | |
| 1159 | 1161 | enum intel_pch { |
| @@ -1809,6 +1811,7 @@ struct intel_pipe_crc { · 649 unmodified lines | ||
| 1809 | 1811 | enum intel_pipe_crc_source source; |
| 1810 | 1812 | int head, tail; |
| 1811 | 1813 | wait_queue_head_t wq; |
| 1814 | + int skipped; | |
| 1812 | 1815 | }; |
| 1813 | 1816 | |
| 1814 | 1817 | struct i915_frontbuffer_tracking { |
| @@ -2069,6 +2072,7 @@ struct drm_i915_private { · 254 unmodified lines | ||
| 2069 | 2072 | |
| 2070 | 2073 | struct intel_gvt *gvt; |
| 2071 | 2074 | |
| 2075 | + struct intel_huc huc; | |
| 2072 | 2076 | struct intel_guc guc; |
| 2073 | 2077 | |
| 2074 | 2078 | struct intel_csr csr; |
| @@ -2843,6 +2847,7 @@ intel_info(const struct drm_i915_private *dev_priv) · 768 unmodified lines | ||
| 2843 | 2847 | #define HAS_GUC(dev_priv) ((dev_priv)->info.has_guc) |
| 2844 | 2848 | #define HAS_GUC_UCODE(dev_priv) (HAS_GUC(dev_priv)) |
| 2845 | 2849 | #define HAS_GUC_SCHED(dev_priv) (HAS_GUC(dev_priv)) |
| 2850 | +#define HAS_HUC_UCODE(dev_priv) (HAS_GUC(dev_priv)) | |
| 2846 | 2851 | |
| 2847 | 2852 | #define HAS_RESOURCE_STREAMER(dev_priv) ((dev_priv)->info.has_resource_streamer) |
| 2848 | 2853 | |
| @@ -3101,10 +3106,10 @@ int i915_gem_throttle_ioctl(struct drm_device *dev, void *data, · 252 unmodified lines | ||
| 3101 | 3106 | struct drm_file *file_priv); |
| 3102 | 3107 | int i915_gem_madvise_ioctl(struct drm_device *dev, void *data, |
| 3103 | 3108 | struct drm_file *file_priv); |
| 3104 | −int i915_gem_set_tiling(struct drm_device *dev, void *data, | |
| 3105 | − struct drm_file *file_priv); | |
| 3106 | −int i915_gem_get_tiling(struct drm_device *dev, void *data, | |
| 3107 | − struct drm_file *file_priv); | |
| 3109 | +int i915_gem_set_tiling_ioctl(struct drm_device *dev, void *data, | |
| 3110 | + struct drm_file *file_priv); | |
| 3111 | +int i915_gem_get_tiling_ioctl(struct drm_device *dev, void *data, | |
| 3112 | + struct drm_file *file_priv); | |
| 3108 | 3113 | void i915_gem_init_userptr(struct drm_i915_private *dev_priv); |
| 3109 | 3114 | int i915_gem_userptr_ioctl(struct drm_device *dev, void *data, |
| 3110 | 3115 | struct drm_file *file); |
| @@ -3323,7 +3328,7 @@ static inline u32 i915_reset_count(struct i915_gpu_error *error) · 212 unmodified lines | ||
| 3323 | 3328 | return READ_ONCE(error->reset_count); |
| 3324 | 3329 | } |
| 3325 | 3330 | |
| 3326 | −void i915_gem_reset_prepare(struct drm_i915_private *dev_priv); | |
| 3331 | +int i915_gem_reset_prepare(struct drm_i915_private *dev_priv); | |
| 3327 | 3332 | void i915_gem_reset_finish(struct drm_i915_private *dev_priv); |
| 3328 | 3333 | void i915_gem_set_wedged(struct drm_i915_private *dev_priv); |
| 3329 | 3334 | void i915_gem_clflush_object(struct drm_i915_gem_object *obj, bool force); |
| @@ -3360,11 +3365,6 @@ int i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, · 30 unmodified lines | ||
| 3360 | 3365 | int i915_gem_open(struct drm_device *dev, struct drm_file *file); |
| 3361 | 3366 | void i915_gem_release(struct drm_device *dev, struct drm_file *file); |
| 3362 | 3367 | |
| 3363 | −u64 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv, u64 size, | |
| 3364 | − int tiling_mode); | |
| 3365 | −u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size, | |
| 3366 | − int tiling_mode, bool fenced); | |
| 3367 | − | |
| 3368 | 3368 | int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, |
| 3369 | 3369 | enum i915_cache_level cache_level); |
| 3370 | 3370 | |
| @@ -3374,36 +3374,12 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev, · 3 unmodified lines | ||
| 3374 | 3374 | struct dma_buf *i915_gem_prime_export(struct drm_device *dev, |
| 3375 | 3375 | struct drm_gem_object *gem_obj, int flags); |
| 3376 | 3376 | |
| 3377 | −struct i915_vma * | |
| 3378 | −i915_gem_obj_to_vma(struct drm_i915_gem_object *obj, | |
| 3379 | − struct i915_address_space *vm, | |
| 3380 | − const struct i915_ggtt_view *view); | |
| 3381 | − | |
| 3382 | −struct i915_vma * | |
| 3383 | −i915_gem_obj_lookup_or_create_vma(struct drm_i915_gem_object *obj, | |
| 3384 | − struct i915_address_space *vm, | |
| 3385 | − const struct i915_ggtt_view *view); | |
| 3386 | − | |
| 3387 | 3377 | static inline struct i915_hw_ppgtt * |
| 3388 | 3378 | i915_vm_to_ppgtt(struct i915_address_space *vm) |
| 3389 | 3379 | { |
| 3390 | 3380 | return container_of(vm, struct i915_hw_ppgtt, base); |
| 3391 | 3381 | } |
| 3392 | 3382 | |
| 3393 | −static inline struct i915_vma * | |
| 3394 | −i915_gem_object_to_ggtt(struct drm_i915_gem_object *obj, | |
| 3395 | − const struct i915_ggtt_view *view) | |
| 3396 | −{ | |
| 3397 | − return i915_gem_obj_to_vma(obj, &to_i915(obj->base.dev)->ggtt.base, view); | |
| 3398 | −} | |
| 3399 | − | |
| 3400 | −static inline unsigned long | |
| 3401 | −i915_gem_object_ggtt_offset(struct drm_i915_gem_object *o, | |
| 3402 | − const struct i915_ggtt_view *view) | |
| 3403 | −{ | |
| 3404 | − return i915_ggtt_offset(i915_gem_object_to_ggtt(o, view)); | |
| 3405 | −} | |
| 3406 | − | |
| 3407 | 3383 | /* i915_gem_fence_reg.c */ |
| 3408 | 3384 | int __must_check i915_vma_get_fence(struct i915_vma *vma); |
| 3409 | 3385 | int __must_check i915_vma_put_fence(struct i915_vma *vma); |
| @@ -3471,8 +3447,9 @@ int __must_check i915_gem_evict_something(struct i915_address_space *vm, · 61 unmodified lines | ||
| 3471 | 3447 | unsigned cache_level, |
| 3472 | 3448 | u64 start, u64 end, |
| 3473 | 3449 | unsigned flags); |
| 3474 | −int __must_check i915_gem_evict_for_vma(struct i915_vma *vma, | |
| 3475 | − unsigned int flags); | |
| 3450 | +int __must_check i915_gem_evict_for_node(struct i915_address_space *vm, | |
| 3451 | + struct drm_mm_node *node, | |
| 3452 | + unsigned int flags); | |
| 3476 | 3453 | int i915_gem_evict_vm(struct i915_address_space *vm, bool do_idle); |
| 3477 | 3454 | |
| 3478 | 3455 | /* belongs in i915_gem_gtt.h */ |
| @@ -3506,7 +3483,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_i915_private *dev_priv · 27 unmodified lines | ||
| 3506 | 3483 | /* i915_gem_internal.c */ |
| 3507 | 3484 | struct drm_i915_gem_object * |
| 3508 | 3485 | i915_gem_object_create_internal(struct drm_i915_private *dev_priv, |
| 3509 | − unsigned int size); | |
| 3486 | + phys_addr_t size); | |
| 3510 | 3487 | |
| 3511 | 3488 | /* i915_gem_shrinker.c */ |
| 3512 | 3489 | unsigned long i915_gem_shrink(struct drm_i915_private *dev_priv, |
| @@ -3531,6 +3508,11 @@ static inline bool i915_gem_object_needs_bit17_swizzle(struct drm_i915_gem_objec · 18 unmodified lines | ||
| 3531 | 3508 | i915_gem_object_is_tiled(obj); |
| 3532 | 3509 | } |
| 3533 | 3510 | |
| 3511 | +u32 i915_gem_fence_size(struct drm_i915_private *dev_priv, u32 size, | |
| 3512 | + unsigned int tiling, unsigned int stride); | |
| 3513 | +u32 i915_gem_fence_alignment(struct drm_i915_private *dev_priv, u32 size, | |
| 3514 | + unsigned int tiling, unsigned int stride); | |
| 3515 | + | |
| 3534 | 3516 | /* i915_debugfs.c */ |
| 3535 | 3517 | #ifdef CONFIG_DEBUG_FS |
| 3536 | 3518 | int i915_debugfs_register(struct drm_i915_private *dev_priv); |
| @@ -1696,12 +1696,7 @@ i915_gem_mmap_ioctl(struct drm_device *dev, void *data, · 1695 unmodified lines | ||
| 1696 | 1696 | |
| 1697 | 1697 | static unsigned int tile_row_pages(struct drm_i915_gem_object *obj) |
| 1698 | 1698 | { |
| 1699 | − u64 size; | |
| 1700 | − | |
| 1701 | − size = i915_gem_object_get_stride(obj); | |
| 1702 | − size *= i915_gem_object_get_tiling(obj) == I915_TILING_Y ? 32 : 8; | |
| 1703 | − | |
| 1704 | − return size >> PAGE_SHIFT; | |
| 1699 | + return i915_gem_object_get_tile_row_size(obj) >> PAGE_SHIFT; | |
| 1705 | 1700 | } |
| 1706 | 1701 | |
| 1707 | 1702 | /** |
| @@ -1754,6 +1749,29 @@ int i915_gem_mmap_gtt_version(void) · 46 unmodified lines | ||
| 1754 | 1749 | return 1; |
| 1755 | 1750 | } |
| 1756 | 1751 | |
| 1752 | +static inline struct i915_ggtt_view | |
| 1753 | +compute_partial_view(struct drm_i915_gem_object *obj, | |
| 1754 | + pgoff_t page_offset, | |
| 1755 | + unsigned int chunk) | |
| 1756 | +{ | |
| 1757 | + struct i915_ggtt_view view; | |
| 1758 | + | |
| 1759 | + if (i915_gem_object_is_tiled(obj)) | |
| 1760 | + chunk = roundup(chunk, tile_row_pages(obj)); | |
| 1761 | + | |
| 1762 | + view.type = I915_GGTT_VIEW_PARTIAL; | |
| 1763 | + view.partial.offset = rounddown(page_offset, chunk); | |
| 1764 | + view.partial.size = | |
| 1765 | + min_t(unsigned int, chunk, | |
| 1766 | + (obj->base.size >> PAGE_SHIFT) - view.partial.offset); | |
| 1767 | + | |
| 1768 | + /* If the partial covers the entire object, just create a normal VMA. */ | |
| 1769 | + if (chunk >= obj->base.size >> PAGE_SHIFT) | |
| 1770 | + view.type = I915_GGTT_VIEW_NORMAL; | |
| 1771 | + | |
| 1772 | + return view; | |
| 1773 | +} | |
| 1774 | + | |
| 1757 | 1775 | /** |
| 1758 | 1776 | * i915_gem_fault - fault a page into the GTT |
| 1759 | 1777 | * @area: CPU VMA in question |
| @@ -1830,26 +1848,9 @@ int i915_gem_fault(struct vm_area_struct *area, struct vm_fault *vmf) · 70 unmodified lines | ||
| 1830 | 1848 | /* Now pin it into the GTT as needed */ |
| 1831 | 1849 | vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, flags); |
| 1832 | 1850 | if (IS_ERR(vma)) { |
| 1833 | − struct i915_ggtt_view view; | |
| 1834 | − unsigned int chunk_size; | |
| 1835 | − | |
| 1836 | 1851 | /* Use a partial view if it is bigger than available space */ |
| 1837 | − chunk_size = MIN_CHUNK_PAGES; | |
| 1838 | − if (i915_gem_object_is_tiled(obj)) | |
| 1839 | − chunk_size = roundup(chunk_size, tile_row_pages(obj)); | |
| 1840 | − | |
| 1841 | − memset(&view, 0, sizeof(view)); | |
| 1842 | − view.type = I915_GGTT_VIEW_PARTIAL; | |
| 1843 | − view.params.partial.offset = rounddown(page_offset, chunk_size); | |
| 1844 | − view.params.partial.size = | |
| 1845 | − min_t(unsigned int, chunk_size, | |
| 1846 | − vma_pages(area) - view.params.partial.offset); | |
| 1847 | − | |
| 1848 | − /* If the partial covers the entire object, just create a | |
| 1849 | − * normal VMA. | |
| 1850 | − */ | |
| 1851 | − if (chunk_size >= obj->base.size >> PAGE_SHIFT) | |
| 1852 | − view.type = I915_GGTT_VIEW_NORMAL; | |
| 1852 | + struct i915_ggtt_view view = | |
| 1853 | + compute_partial_view(obj, page_offset, MIN_CHUNK_PAGES); | |
| 1853 | 1854 | |
| 1854 | 1855 | /* Userspace is now writing through an untracked VMA, abandon |
| 1855 | 1856 | * all hope that the hardware is able to track future writes. |
| @@ -1878,7 +1879,7 @@ int i915_gem_fault(struct vm_area_struct *area, struct vm_fault *vmf) · 22 unmodified lines | ||
| 1878 | 1879 | |
| 1879 | 1880 | /* Finally, remap it using the new GTT offset */ |
| 1880 | 1881 | ret = remap_io_mapping(area, |
| 1881 | − area->vm_start + (vma->ggtt_view.params.partial.offset << PAGE_SHIFT), | |
| 1882 | + area->vm_start + (vma->ggtt_view.partial.offset << PAGE_SHIFT), | |
| 1882 | 1883 | (ggtt->mappable_base + vma->node.start) >> PAGE_SHIFT, |
| 1883 | 1884 | min_t(u64, vma->size, area->vm_end - area->vm_start), |
| 1884 | 1885 | &ggtt->mappable); |
| @@ -2021,69 +2022,6 @@ void i915_gem_runtime_suspend(struct drm_i915_private *dev_priv) · 136 unmodified lines | ||
| 2021 | 2022 | } |
| 2022 | 2023 | } |
| 2023 | 2024 | |
| 2024 | −/** | |
| 2025 | − * i915_gem_get_ggtt_size - return required global GTT size for an object | |
| 2026 | − * @dev_priv: i915 device | |
| 2027 | − * @size: object size | |
| 2028 | − * @tiling_mode: tiling mode | |
| 2029 | − * | |
| 2030 | − * Return the required global GTT size for an object, taking into account | |
| 2031 | − * potential fence register mapping. | |
| 2032 | − */ | |
| 2033 | −u64 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv, | |
| 2034 | − u64 size, int tiling_mode) | |
| 2035 | −{ | |
| 2036 | − u64 ggtt_size; | |
| 2037 | − | |
| 2038 | − GEM_BUG_ON(size == 0); | |
| 2039 | − | |
| 2040 | − if (INTEL_GEN(dev_priv) >= 4 || | |
| 2041 | − tiling_mode == I915_TILING_NONE) | |
| 2042 | − return size; | |
| 2043 | − | |
| 2044 | − /* Previous chips need a power-of-two fence region when tiling */ | |
| 2045 | − if (IS_GEN3(dev_priv)) | |
| 2046 | − ggtt_size = 1024*1024; | |
| 2047 | − else | |
| 2048 | − ggtt_size = 512*1024; | |
| 2049 | − | |
| 2050 | − while (ggtt_size < size) | |
| 2051 | − ggtt_size <<= 1; | |
| 2052 | − | |
| 2053 | − return ggtt_size; | |
| 2054 | −} | |
| 2055 | − | |
| 2056 | −/** | |
| 2057 | − * i915_gem_get_ggtt_alignment - return required global GTT alignment | |
| 2058 | − * @dev_priv: i915 device | |
| 2059 | − * @size: object size | |
| 2060 | − * @tiling_mode: tiling mode | |
| 2061 | − * @fenced: is fenced alignment required or not | |
| 2062 | − * | |
| 2063 | − * Return the required global GTT alignment for an object, taking into account | |
| 2064 | − * potential fence register mapping. | |
| 2065 | − */ | |
| 2066 | −u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size, | |
| 2067 | − int tiling_mode, bool fenced) | |
| 2068 | −{ | |
| 2069 | − GEM_BUG_ON(size == 0); | |
| 2070 | − | |
| 2071 | − /* | |
| 2072 | − * Minimum alignment is 4k (GTT page size), but might be greater | |
| 2073 | − * if a fence register is needed for the object. | |
| 2074 | − */ | |
| 2075 | − if (INTEL_GEN(dev_priv) >= 4 || | |
| 2076 | − (!fenced && (IS_G33(dev_priv) || IS_PINEVIEW(dev_priv))) || | |
| 2077 | − tiling_mode == I915_TILING_NONE) | |
| 2078 | − return 4096; | |
| 2079 | − | |
| 2080 | − /* | |
| 2081 | − * Previous chips need to be aligned to the size of the smallest | |
| 2082 | − * fence register that can contain the object. | |
| 2083 | − */ | |
| 2084 | − return i915_gem_get_ggtt_size(dev_priv, size, tiling_mode); | |
| 2085 | −} | |
| 2086 | − | |
| 2087 | 2025 | static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj) |
| 2088 | 2026 | { |
| 2089 | 2027 | struct drm_i915_private *dev_priv = to_i915(obj->base.dev); |
| @@ -2666,13 +2604,52 @@ i915_gem_find_active_request(struct intel_engine_cs *engine) · 576 unmodified lines | ||
| 2666 | 2604 | if (__i915_gem_request_completed(request)) |
| 2667 | 2605 | continue; |
| 2668 | 2606 | |
| 2607 | + GEM_BUG_ON(request->engine != engine); | |
| 2669 | 2608 | return request; |
| 2670 | 2609 | } |
| 2671 | 2610 | |
| 2672 | 2611 | return NULL; |
| 2673 | 2612 | } |
| 2674 | 2613 | |
| 2675 | −static void reset_request(struct drm_i915_gem_request *request) | |
| 2614 | +static bool engine_stalled(struct intel_engine_cs *engine) | |
| 2615 | +{ | |
| 2616 | + if (!engine->hangcheck.stalled) | |
| 2617 | + return false; | |
| 2618 | + | |
| 2619 | + /* Check for possible seqno movement after hang declaration */ | |
| 2620 | + if (engine->hangcheck.seqno != intel_engine_get_seqno(engine)) { | |
| 2621 | + DRM_DEBUG_DRIVER("%s pardoned\n", engine->name); | |
| 2622 | + return false; | |
| 2623 | + } | |
| 2624 | + | |
| 2625 | + return true; | |
| 2626 | +} | |
| 2627 | + | |
| 2628 | +int i915_gem_reset_prepare(struct drm_i915_private *dev_priv) | |
| 2629 | +{ | |
| 2630 | + struct intel_engine_cs *engine; | |
| 2631 | + enum intel_engine_id id; | |
| 2632 | + int err = 0; | |
| 2633 | + | |
| 2634 | + /* Ensure irq handler finishes, and not run again. */ | |
| 2635 | + for_each_engine(engine, dev_priv, id) { | |
| 2636 | + struct drm_i915_gem_request *request; | |
| 2637 | + | |
| 2638 | + tasklet_kill(&engine->irq_tasklet); | |
| 2639 | + | |
| 2640 | + if (engine_stalled(engine)) { | |
| 2641 | + request = i915_gem_find_active_request(engine); | |
| 2642 | + if (request && request->fence.error == -EIO) | |
| 2643 | + err = -EIO; /* Previous reset failed! */ | |
| 2644 | + } | |
| 2645 | + } | |
| 2646 | + | |
| 2647 | + i915_gem_revoke_fences(dev_priv); | |
| 2648 | + | |
| 2649 | + return err; | |
| 2650 | +} | |
| 2651 | + | |
| 2652 | +static void skip_request(struct drm_i915_gem_request *request) | |
| 2676 | 2653 | { |
| 2677 | 2654 | void *vaddr = request->ring->vaddr; |
| 2678 | 2655 | u32 head; |
| @@ -2687,20 +2664,74 @@ static void reset_request(struct drm_i915_gem_request *request) · 8 unmodified lines | ||
| 2687 | 2664 | head = 0; |
| 2688 | 2665 | } |
| 2689 | 2666 | memset(vaddr + head, 0, request->postfix - head); |
| 2667 | + | |
| 2668 | + dma_fence_set_error(&request->fence, -EIO); | |
| 2690 | 2669 | } |
| 2691 | 2670 | |
| 2692 | −void i915_gem_reset_prepare(struct drm_i915_private *dev_priv) | |
| 2671 | +static void engine_skip_context(struct drm_i915_gem_request *request) | |
| 2693 | 2672 | { |
| 2694 | − i915_gem_revoke_fences(dev_priv); | |
| 2673 | + struct intel_engine_cs *engine = request->engine; | |
| 2674 | + struct i915_gem_context *hung_ctx = request->ctx; | |
| 2675 | + struct intel_timeline *timeline; | |
| 2676 | + unsigned long flags; | |
| 2677 | + | |
| 2678 | + timeline = i915_gem_context_lookup_timeline(hung_ctx, engine); | |
| 2679 | + | |
| 2680 | + spin_lock_irqsave(&engine->timeline->lock, flags); | |
| 2681 | + spin_lock(&timeline->lock); | |
| 2682 | + | |
| 2683 | + list_for_each_entry_continue(request, &engine->timeline->requests, link) | |
| 2684 | + if (request->ctx == hung_ctx) | |
| 2685 | + skip_request(request); | |
| 2686 | + | |
| 2687 | + list_for_each_entry(request, &timeline->requests, link) | |
| 2688 | + skip_request(request); | |
| 2689 | + | |
| 2690 | + spin_unlock(&timeline->lock); | |
| 2691 | + spin_unlock_irqrestore(&engine->timeline->lock, flags); | |
| 2692 | +} | |
| 2693 | + | |
| 2694 | +/* Returns true if the request was guilty of hang */ | |
| 2695 | +static bool i915_gem_reset_request(struct drm_i915_gem_request *request) | |
| 2696 | +{ | |
| 2697 | + /* Read once and return the resolution */ | |
| 2698 | + const bool guilty = engine_stalled(request->engine); | |
| 2699 | + | |
| 2700 | + /* The guilty request will get skipped on a hung engine. | |
| 2701 | + * | |
| 2702 | + * Users of client default contexts do not rely on logical | |
| 2703 | + * state preserved between batches so it is safe to execute | |
| 2704 | + * queued requests following the hang. Non default contexts | |
| 2705 | + * rely on preserved state, so skipping a batch loses the | |
| 2706 | + * evolution of the state and it needs to be considered corrupted. | |
| 2707 | + * Executing more queued batches on top of corrupted state is | |
| 2708 | + * risky. But we take the risk by trying to advance through | |
| 2709 | + * the queued requests in order to make the client behaviour | |
| 2710 | + * more predictable around resets, by not throwing away random | |
| 2711 | + * amount of batches it has prepared for execution. Sophisticated | |
| 2712 | + * clients can use gem_reset_stats_ioctl and dma fence status | |
| 2713 | + * (exported via sync_file info ioctl on explicit fences) to observe | |
| 2714 | + * when it loses the context state and should rebuild accordingly. | |
| 2715 | + * | |
| 2716 | + * The context ban, and ultimately the client ban, mechanism are safety | |
| 2717 | + * valves if client submission ends up resulting in nothing more than | |
| 2718 | + * subsequent hangs. | |
| 2719 | + */ | |
| 2720 | + | |
| 2721 | + if (guilty) { | |
| 2722 | + i915_gem_context_mark_guilty(request->ctx); | |
| 2723 | + skip_request(request); | |
| 2724 | + } else { | |
| 2725 | + i915_gem_context_mark_innocent(request->ctx); | |
| 2726 | + dma_fence_set_error(&request->fence, -EAGAIN); | |
| 2727 | + } | |
| 2728 | + | |
| 2729 | + return guilty; | |
| 2695 | 2730 | } |
| 2696 | 2731 | |
| 2697 | 2732 | static void i915_gem_reset_engine(struct intel_engine_cs *engine) |
| 2698 | 2733 | { |
| 2699 | 2734 | struct drm_i915_gem_request *request; |
| 2700 | − struct i915_gem_context *hung_ctx; | |
| 2701 | − struct intel_timeline *timeline; | |
| 2702 | − unsigned long flags; | |
| 2703 | − bool ring_hung; | |
| 2704 | 2735 | |
| 2705 | 2736 | if (engine->irq_seqno_barrier) |
| 2706 | 2737 | engine->irq_seqno_barrier(engine); |
| @@ -2709,22 +2740,7 @@ static void i915_gem_reset_engine(struct intel_engine_cs *engine) · 2 unmodified lines | ||
| 2709 | 2740 | if (!request) |
| 2710 | 2741 | return; |
| 2711 | 2742 | |
| 2712 | − hung_ctx = request->ctx; | |
| 2713 | − | |
| 2714 | − ring_hung = engine->hangcheck.stalled; | |
| 2715 | − if (engine->hangcheck.seqno != intel_engine_get_seqno(engine)) { | |
| 2716 | − DRM_DEBUG_DRIVER("%s pardoned, was guilty? %s\n", | |
| 2717 | − engine->name, | |
| 2718 | − yesno(ring_hung)); | |
| 2719 | − ring_hung = false; | |
| 2720 | − } | |
| 2721 | − | |
| 2722 | − if (ring_hung) | |
| 2723 | − i915_gem_context_mark_guilty(hung_ctx); | |
| 2724 | − else | |
| 2725 | − i915_gem_context_mark_innocent(hung_ctx); | |
| 2726 | − | |
| 2727 | − if (!ring_hung) | |
| 2743 | + if (!i915_gem_reset_request(request)) | |
| 2728 | 2744 | return; |
| 2729 | 2745 | |
| 2730 | 2746 | DRM_DEBUG_DRIVER("resetting %s to restart from tail of request 0x%x\n", |
| @@ -2734,34 +2750,8 @@ static void i915_gem_reset_engine(struct intel_engine_cs *engine) · 3 unmodified lines | ||
| 2734 | 2750 | engine->reset_hw(engine, request); |
| 2735 | 2751 | |
| 2736 | 2752 | /* If this context is now banned, skip all of its pending requests. */ |
| 2737 | − if (!i915_gem_context_is_banned(hung_ctx)) | |
| 2738 | − return; | |
| 2739 | − | |
| 2740 | − /* Users of the default context do not rely on logical state | |
| 2741 | − * preserved between batches. They have to emit full state on | |
| 2742 | − * every batch and so it is safe to execute queued requests following | |
| 2743 | − * the hang. | |
| 2744 | − * | |
| 2745 | − * Other contexts preserve state, now corrupt. We want to skip all | |
| 2746 | − * queued requests that reference the corrupt context. | |
| 2747 | − */ | |
| 2748 | − if (i915_gem_context_is_default(hung_ctx)) | |
| 2749 | − return; | |
| 2750 | − | |
| 2751 | − timeline = i915_gem_context_lookup_timeline(hung_ctx, engine); | |
| 2752 | − | |
| 2753 | − spin_lock_irqsave(&engine->timeline->lock, flags); | |
| 2754 | − spin_lock(&timeline->lock); | |
| 2755 | − | |
| 2756 | − list_for_each_entry_continue(request, &engine->timeline->requests, link) | |
| 2757 | − if (request->ctx == hung_ctx) | |
| 2758 | − reset_request(request); | |
| 2759 | − | |
| 2760 | − list_for_each_entry(request, &timeline->requests, link) | |
| 2761 | − reset_request(request); | |
| 2762 | − | |
| 2763 | − spin_unlock(&timeline->lock); | |
| 2764 | − spin_unlock_irqrestore(&engine->timeline->lock, flags); | |
| 2753 | + if (i915_gem_context_is_banned(request->ctx)) | |
| 2754 | + engine_skip_context(request); | |
| 2765 | 2755 | } |
| 2766 | 2756 | |
| 2767 | 2757 | void i915_gem_reset_finish(struct drm_i915_private *dev_priv) |
| @@ -2788,12 +2778,16 @@ void i915_gem_reset_finish(struct drm_i915_private *dev_priv) · 20 unmodified lines | ||
| 2788 | 2778 | |
| 2789 | 2779 | static void nop_submit_request(struct drm_i915_gem_request *request) |
| 2790 | 2780 | { |
| 2781 | + dma_fence_set_error(&request->fence, -EIO); | |
| 2791 | 2782 | i915_gem_request_submit(request); |
| 2792 | 2783 | intel_engine_init_global_seqno(request->engine, request->global_seqno); |
| 2793 | 2784 | } |
| 2794 | 2785 | |
| 2795 | −static void i915_gem_cleanup_engine(struct intel_engine_cs *engine) | |
| 2786 | +static void engine_set_wedged(struct intel_engine_cs *engine) | |
| 2796 | 2787 | { |
| 2788 | + struct drm_i915_gem_request *request; | |
| 2789 | + unsigned long flags; | |
| 2790 | + | |
| 2797 | 2791 | /* We need to be sure that no thread is running the old callback as |
| 2798 | 2792 | * we install the nop handler (otherwise we would submit a request |
| 2799 | 2793 | * to hardware that will never complete). In order to prevent this |
| @@ -2802,6 +2796,12 @@ static void i915_gem_cleanup_engine(struct intel_engine_cs *engine) · 2 unmodified lines | ||
| 2802 | 2796 | */ |
| 2803 | 2797 | engine->submit_request = nop_submit_request; |
| 2804 | 2798 | |
| 2799 | + /* Mark all executing requests as skipped */ | |
| 2800 | + spin_lock_irqsave(&engine->timeline->lock, flags); | |
| 2801 | + list_for_each_entry(request, &engine->timeline->requests, link) | |
| 2802 | + dma_fence_set_error(&request->fence, -EIO); | |
| 2803 | + spin_unlock_irqrestore(&engine->timeline->lock, flags); | |
| 2804 | + | |
| 2805 | 2805 | /* Mark all pending requests as complete so that any concurrent |
| 2806 | 2806 | * (lockless) lookup doesn't try and wait upon the request as we |
| 2807 | 2807 | * reset it. |
| @@ -2837,7 +2837,7 @@ static int __i915_gem_set_wedged_BKL(void *data) · 29 unmodified lines | ||
| 2837 | 2837 | enum intel_engine_id id; |
| 2838 | 2838 | |
| 2839 | 2839 | for_each_engine(engine, i915, id) |
| 2840 | − i915_gem_cleanup_engine(engine); | |
| 2840 | + engine_set_wedged(engine); | |
| 2841 | 2841 | |
| 2842 | 2842 | return 0; |
| 2843 | 2843 | } |
| @@ -3397,7 +3397,7 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data, · 553 unmodified lines | ||
| 3397 | 3397 | struct drm_i915_gem_caching *args = data; |
| 3398 | 3398 | struct drm_i915_gem_object *obj; |
| 3399 | 3399 | enum i915_cache_level level; |
| 3400 | − int ret; | |
| 3400 | + int ret = 0; | |
| 3401 | 3401 | |
| 3402 | 3402 | switch (args->caching) { |
| 3403 | 3403 | case I915_CACHING_NONE: |
| @@ -3422,20 +3422,29 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data, · 18 unmodified lines | ||
| 3422 | 3422 | return -EINVAL; |
| 3423 | 3423 | } |
| 3424 | 3424 | |
| 3425 | − ret = i915_mutex_lock_interruptible(dev); | |
| 3425 | + obj = i915_gem_object_lookup(file, args->handle); | |
| 3426 | + if (!obj) | |
| 3427 | + return -ENOENT; | |
| 3428 | + | |
| 3429 | + if (obj->cache_level == level) | |
| 3430 | + goto out; | |
| 3431 | + | |
| 3432 | + ret = i915_gem_object_wait(obj, | |
| 3433 | + I915_WAIT_INTERRUPTIBLE, | |
| 3434 | + MAX_SCHEDULE_TIMEOUT, | |
| 3435 | + to_rps_client(file)); | |
| 3426 | 3436 | if (ret) |
| 3427 | − return ret; | |
| 3437 | + goto out; | |
| 3428 | 3438 | |
| 3429 | − obj = i915_gem_object_lookup(file, args->handle); | |
| 3430 | − if (!obj) { | |
| 3431 | − ret = -ENOENT; | |
| 3432 | − goto unlock; | |
| 3433 | − } | |
| 3439 | + ret = i915_mutex_lock_interruptible(dev); | |
| 3440 | + if (ret) | |
| 3441 | + goto out; | |
| 3434 | 3442 | |
| 3435 | 3443 | ret = i915_gem_object_set_cache_level(obj, level); |
| 3436 | − i915_gem_object_put(obj); | |
| 3437 | −unlock: | |
| 3438 | 3444 | mutex_unlock(&dev->struct_mutex); |
| 3445 | + | |
| 3446 | +out: | |
| 3447 | + i915_gem_object_put(obj); | |
| 3439 | 3448 | return ret; |
| 3440 | 3449 | } |
| 3441 | 3450 | |
| @@ -3485,7 +3494,7 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, · 43 unmodified lines | ||
| 3485 | 3494 | * try to preserve the existing ABI). |
| 3486 | 3495 | */ |
| 3487 | 3496 | vma = ERR_PTR(-ENOSPC); |
| 3488 | − if (view->type == I915_GGTT_VIEW_NORMAL) | |
| 3497 | + if (!view || view->type == I915_GGTT_VIEW_NORMAL) | |
| 3489 | 3498 | vma = i915_gem_object_ggtt_pin(obj, view, 0, alignment, |
| 3490 | 3499 | PIN_MAPPABLE | PIN_NONBLOCK); |
| 3491 | 3500 | if (IS_ERR(vma)) { |
| @@ -3544,11 +3553,10 @@ i915_gem_object_unpin_from_display_plane(struct i915_vma *vma) · 52 unmodified lines | ||
| 3544 | 3553 | return; |
| 3545 | 3554 | |
| 3546 | 3555 | if (--vma->obj->pin_display == 0) |
| 3547 | − vma->display_alignment = 0; | |
| 3556 | + vma->display_alignment = I915_GTT_MIN_ALIGNMENT; | |
| 3548 | 3557 | |
| 3549 | 3558 | /* Bump the LRU to try and avoid premature eviction whilst flipping */ |
| 3550 | − if (!i915_vma_is_active(vma)) | |
| 3551 | − list_move_tail(&vma->vm_link, &vma->vm->inactive_list); | |
| 3559 | + i915_gem_object_bump_inactive_ggtt(vma->obj); | |
| 3552 | 3560 | |
| 3553 | 3561 | i915_vma_unpin(vma); |
| 3554 | 3562 | } |
| @@ -3679,8 +3687,8 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, · 124 unmodified lines | ||
| 3679 | 3687 | |
| 3680 | 3688 | lockdep_assert_held(&obj->base.dev->struct_mutex); |
| 3681 | 3689 | |
| 3682 | − vma = i915_gem_obj_lookup_or_create_vma(obj, vm, view); | |
| 3683 | − if (IS_ERR(vma)) | |
| 3690 | + vma = i915_vma_instance(obj, vm, view); | |
| 3691 | + if (unlikely(IS_ERR(vma))) | |
| 3684 | 3692 | return vma; |
| 3685 | 3693 | |
| 3686 | 3694 | if (i915_vma_misplaced(vma, size, alignment, flags)) { |
| @@ -3689,10 +3697,6 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, · 2 unmodified lines | ||
| 3689 | 3697 | return ERR_PTR(-ENOSPC); |
| 3690 | 3698 | |
| 3691 | 3699 | if (flags & PIN_MAPPABLE) { |
| 3692 | − u32 fence_size; | |
| 3693 | − | |
| 3694 | − fence_size = i915_gem_get_ggtt_size(dev_priv, vma->size, | |
| 3695 | − i915_gem_object_get_tiling(obj)); | |
| 3696 | 3700 | /* If the required space is larger than the available |
| 3697 | 3701 | * aperture, we will not able to find a slot for the |
| 3698 | 3702 | * object and unbinding the object now will be in |
| @@ -3700,7 +3704,7 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, · 1 unmodified line | ||
| 3700 | 3704 | * the object in and out of the Global GTT and |
| 3701 | 3705 | * waste a lot of cycles under the mutex. |
| 3702 | 3706 | */ |
| 3703 | − if (fence_size > dev_priv->ggtt.mappable_end) | |
| 3707 | + if (vma->fence_size > dev_priv->ggtt.mappable_end) | |
| 3704 | 3708 | return ERR_PTR(-E2BIG); |
| 3705 | 3709 | |
| 3706 | 3710 | /* If NONBLOCK is set the caller is optimistically |
| @@ -3719,7 +3723,7 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, · 12 unmodified lines | ||
| 3719 | 3723 | * we could try to minimise harm to others. |
| 3720 | 3724 | */ |
| 3721 | 3725 | if (flags & PIN_NONBLOCK && |
| 3722 | − fence_size > dev_priv->ggtt.mappable_end / 2) | |
| 3726 | + vma->fence_size > dev_priv->ggtt.mappable_end / 2) | |
| 3723 | 3727 | return ERR_PTR(-ENOSPC); |
| 3724 | 3728 | } |
| 3725 | 3729 | |
| @@ -4193,7 +4197,8 @@ static void assert_kernel_context_is_current(struct drm_i915_private *dev_priv) · 467 unmodified lines | ||
| 4193 | 4197 | enum intel_engine_id id; |
| 4194 | 4198 | |
| 4195 | 4199 | for_each_engine(engine, dev_priv, id) |
| 4196 | − GEM_BUG_ON(!i915_gem_context_is_kernel(engine->last_retired_context)); | |
| 4200 | + GEM_BUG_ON(engine->last_retired_context && | |
| 4201 | + !i915_gem_context_is_kernel(engine->last_retired_context)); | |
| 4197 | 4202 | } |
| 4198 | 4203 | |
| 4199 | 4204 | int i915_gem_suspend(struct drm_i915_private *dev_priv) |
| @@ -97,7 +97,7 @@ · 96 unmodified lines | ||
| 97 | 97 | * part. It should be safe to decrease this, but it's more future proof as is. |
| 98 | 98 | */ |
| 99 | 99 | #define GEN6_CONTEXT_ALIGN (64<<10) |
| 100 | −#define GEN7_CONTEXT_ALIGN 4096 | |
| 100 | +#define GEN7_CONTEXT_ALIGN I915_GTT_MIN_ALIGNMENT | |
| 101 | 101 | |
| 102 | 102 | static size_t get_context_alignment(struct drm_i915_private *dev_priv) |
| 103 | 103 | { |
| @@ -205,27 +205,6 @@ alloc_context_obj(struct drm_i915_private *dev_priv, u64 size) · 101 unmodified lines | ||
| 205 | 205 | return obj; |
| 206 | 206 | } |
| 207 | 207 | |
| 208 | −static void i915_ppgtt_close(struct i915_address_space *vm) | |
| 209 | −{ | |
| 210 | − struct list_head *phases[] = { | |
| 211 | − &vm->active_list, | |
| 212 | − &vm->inactive_list, | |
| 213 | − &vm->unbound_list, | |
| 214 | − NULL, | |
| 215 | − }, **phase; | |
| 216 | − | |
| 217 | − GEM_BUG_ON(vm->closed); | |
| 218 | − vm->closed = true; | |
| 219 | − | |
| 220 | − for (phase = phases; *phase; phase++) { | |
| 221 | − struct i915_vma *vma, *vn; | |
| 222 | − | |
| 223 | − list_for_each_entry_safe(vma, vn, *phase, vm_link) | |
| 224 | − if (!i915_vma_is_closed(vma)) | |
| 225 | − i915_vma_close(vma); | |
| 226 | − } | |
| 227 | −} | |
| 228 | − | |
| 229 | 208 | static void context_close(struct i915_gem_context *ctx) |
| 230 | 209 | { |
| 231 | 210 | i915_gem_context_set_closed(ctx); |
| @@ -290,7 +269,7 @@ __create_hw_context(struct drm_i915_private *dev_priv, · 58 unmodified lines | ||
| 290 | 269 | goto err_out; |
| 291 | 270 | } |
| 292 | 271 | |
| 293 | − vma = i915_vma_create(obj, &dev_priv->ggtt.base, NULL); | |
| 272 | + vma = i915_vma_instance(obj, &dev_priv->ggtt.base, NULL); | |
| 294 | 273 | if (IS_ERR(vma)) { |
| 295 | 274 | i915_gem_object_put(obj); |
| 296 | 275 | ret = PTR_ERR(vma); |
| @@ -341,7 +320,7 @@ __create_hw_context(struct drm_i915_private *dev_priv, · 44 unmodified lines | ||
| 341 | 320 | if (HAS_GUC(dev_priv) && i915.enable_guc_loading) |
| 342 | 321 | ctx->ggtt_offset_bias = GUC_WOPCM_TOP; |
| 343 | 322 | else |
| 344 | − ctx->ggtt_offset_bias = 4096; | |
| 323 | + ctx->ggtt_offset_bias = I915_GTT_PAGE_SIZE; | |
| 345 | 324 | |
| 346 | 325 | return ctx; |
| 347 | 326 | |
| @@ -456,7 +435,8 @@ int i915_gem_context_init(struct drm_i915_private *dev_priv) · 108 unmodified lines | ||
| 456 | 435 | dev_priv->hw_context_size = 0; |
| 457 | 436 | } else if (HAS_HW_CONTEXTS(dev_priv)) { |
| 458 | 437 | dev_priv->hw_context_size = |
| 459 | − round_up(get_context_size(dev_priv), 4096); | |
| 438 | + round_up(get_context_size(dev_priv), | |
| 439 | + I915_GTT_PAGE_SIZE); | |
| 460 | 440 | if (dev_priv->hw_context_size > (1<<20)) { |
| 461 | 441 | DRM_DEBUG_DRIVER("Disabling HW Contexts; invalid size %d\n", |
| 462 | 442 | dev_priv->hw_context_size); |
| @@ -897,6 +877,26 @@ int i915_switch_context(struct drm_i915_gem_request *req) · 434 unmodified lines | ||
| 897 | 877 | return do_rcs_switch(req); |
| 898 | 878 | } |
| 899 | 879 | |
| 880 | +static bool engine_has_kernel_context(struct intel_engine_cs *engine) | |
| 881 | +{ | |
| 882 | + struct i915_gem_timeline *timeline; | |
| 883 | + | |
| 884 | + list_for_each_entry(timeline, &engine->i915->gt.timelines, link) { | |
| 885 | + struct intel_timeline *tl; | |
| 886 | + | |
| 887 | + if (timeline == &engine->i915->gt.global_timeline) | |
| 888 | + continue; | |
| 889 | + | |
| 890 | + tl = &timeline->engine[engine->id]; | |
| 891 | + if (i915_gem_active_peek(&tl->last_request, | |
| 892 | + &engine->i915->drm.struct_mutex)) | |
| 893 | + return false; | |
| 894 | + } | |
| 895 | + | |
| 896 | + return (!engine->last_retired_context || | |
| 897 | + i915_gem_context_is_kernel(engine->last_retired_context)); | |
| 898 | +} | |
| 899 | + | |
| 900 | 900 | int i915_gem_switch_to_kernel_context(struct drm_i915_private *dev_priv) |
| 901 | 901 | { |
| 902 | 902 | struct intel_engine_cs *engine; |
| @@ -905,10 +905,15 @@ int i915_gem_switch_to_kernel_context(struct drm_i915_private *dev_priv) · 2 unmodified lines | ||
| 905 | 905 | |
| 906 | 906 | lockdep_assert_held(&dev_priv->drm.struct_mutex); |
| 907 | 907 | |
| 908 | + i915_gem_retire_requests(dev_priv); | |
| 909 | + | |
| 908 | 910 | for_each_engine(engine, dev_priv, id) { |
| 909 | 911 | struct drm_i915_gem_request *req; |
| 910 | 912 | int ret; |
| 911 | 913 | |
| 914 | + if (engine_has_kernel_context(engine)) | |
| 915 | + continue; | |
| 916 | + | |
| 912 | 917 | req = i915_gem_request_alloc(engine, dev_priv->kernel_context); |
| 913 | 918 | if (IS_ERR(req)) |
| 914 | 919 | return PTR_ERR(req); |
| @@ -231,7 +231,8 @@ found: · 230 unmodified lines | ||
| 231 | 231 | |
| 232 | 232 | /** |
| 233 | 233 | * i915_gem_evict_for_vma - Evict vmas to make room for binding a new one |
| 234 | − * @target: address space and range to evict for | |
| 234 | + * @vm: address space to evict from | |
| 235 | + * @target: range (and color) to evict for | |
| 235 | 236 | * @flags: additional flags to control the eviction algorithm |
| 236 | 237 | * |
| 237 | 238 | * This function will try to evict vmas that overlap the target node. |
| @@ -239,18 +240,20 @@ found: · 1 unmodified line | ||
| 239 | 240 | * To clarify: This is for freeing up virtual address space, not for freeing |
| 240 | 241 | * memory in e.g. the shrinker. |
| 241 | 242 | */ |
| 242 | −int i915_gem_evict_for_vma(struct i915_vma *target, unsigned int flags) | |
| 243 | +int i915_gem_evict_for_node(struct i915_address_space *vm, | |
| 244 | + struct drm_mm_node *target, | |
| 245 | + unsigned int flags) | |
| 243 | 246 | { |
| 244 | 247 | LIST_HEAD(eviction_list); |
| 245 | 248 | struct drm_mm_node *node; |
| 246 | − u64 start = target->node.start; | |
| 247 | − u64 end = start + target->node.size; | |
| 249 | + u64 start = target->start; | |
| 250 | + u64 end = start + target->size; | |
| 248 | 251 | struct i915_vma *vma, *next; |
| 249 | 252 | bool check_color; |
| 250 | 253 | int ret = 0; |
| 251 | 254 | |
| 252 | − lockdep_assert_held(&target->vm->i915->drm.struct_mutex); | |
| 253 | − trace_i915_gem_evict_vma(target, flags); | |
| 255 | + lockdep_assert_held(&vm->i915->drm.struct_mutex); | |
| 256 | + trace_i915_gem_evict_node(vm, target, flags); | |
| 254 | 257 | |
| 255 | 258 | /* Retire before we search the active list. Although we have |
| 256 | 259 | * reasonable accuracy in our retirement lists, we may have |
| @@ -258,18 +261,18 @@ int i915_gem_evict_for_vma(struct i915_vma *target, unsigned int flags) · 1 unmodified line | ||
| 258 | 261 | * retiring. |
| 259 | 262 | */ |
| 260 | 263 | if (!(flags & PIN_NONBLOCK)) |
| 261 | − i915_gem_retire_requests(target->vm->i915); | |
| 264 | + i915_gem_retire_requests(vm->i915); | |
| 262 | 265 | |
| 263 | − check_color = target->vm->mm.color_adjust; | |
| 266 | + check_color = vm->mm.color_adjust; | |
| 264 | 267 | if (check_color) { |
| 265 | 268 | /* Expand search to cover neighbouring guard pages (or lack!) */ |
| 266 | − if (start > target->vm->start) | |
| 267 | − start -= 4096; | |
| 268 | − if (end < target->vm->start + target->vm->total) | |
| 269 | − end += 4096; | |
| 269 | + if (start > vm->start) | |
| 270 | + start -= I915_GTT_PAGE_SIZE; | |
| 271 | + if (end < vm->start + vm->total) | |
| 272 | + end += I915_GTT_PAGE_SIZE; | |
| 270 | 273 | } |
| 271 | 274 | |
| 272 | − drm_mm_for_each_node_in_range(node, &target->vm->mm, start, end) { | |
| 275 | + drm_mm_for_each_node_in_range(node, &vm->mm, start, end) { | |
| 273 | 276 | /* If we find any non-objects (!vma), we cannot evict them */ |
| 274 | 277 | if (node->color == I915_COLOR_UNEVICTABLE) { |
| 275 | 278 | ret = -ENOSPC; |
| @@ -285,12 +288,12 @@ int i915_gem_evict_for_vma(struct i915_vma *target, unsigned int flags) · 9 unmodified lines | ||
| 285 | 288 | * those as well to make room for our guard pages. |
| 286 | 289 | */ |
| 287 | 290 | if (check_color) { |
| 288 | − if (vma->node.start + vma->node.size == target->node.start) { | |
| 289 | − if (vma->node.color == target->node.color) | |
| 291 | + if (vma->node.start + vma->node.size == node->start) { | |
| 292 | + if (vma->node.color == node->color) | |
| 290 | 293 | continue; |
| 291 | 294 | } |
| 292 | − if (vma->node.start == target->node.start + target->node.size) { | |
| 293 | − if (vma->node.color == target->node.color) | |
| 295 | + if (vma->node.start == node->start + node->size) { | |
| 296 | + if (vma->node.color == node->color) | |
| 294 | 297 | continue; |
| 295 | 298 | } |
| 296 | 299 | } |
| @@ -302,7 +305,7 @@ int i915_gem_evict_for_vma(struct i915_vma *target, unsigned int flags) · 5 unmodified lines | ||
| 302 | 305 | } |
| 303 | 306 | |
| 304 | 307 | /* Overlap of objects in the same batch? */ |
| 305 | − if (i915_vma_is_pinned(vma)) { | |
| 308 | + if (i915_vma_is_pinned(vma) || !list_empty(&vma->exec_list)) { | |
| 306 | 309 | ret = -ENOSPC; |
| 307 | 310 | if (vma->exec_entry && |
| 308 | 311 | vma->exec_entry->flags & EXEC_OBJECT_PINNED) |
| @@ -184,7 +184,7 @@ eb_lookup_vmas(struct eb_vmas *eb, · 183 unmodified lines | ||
| 184 | 184 | * from the (obj, vm) we don't run the risk of creating |
| 185 | 185 | * duplicated vmas for the same vm. |
| 186 | 186 | */ |
| 187 | − vma = i915_gem_obj_lookup_or_create_vma(obj, vm, NULL); | |
| 187 | + vma = i915_vma_instance(obj, vm, NULL); | |
| 188 | 188 | if (unlikely(IS_ERR(vma))) { |
| 189 | 189 | DRM_DEBUG("Failed to lookup VMA\n"); |
| 190 | 190 | ret = PTR_ERR(vma); |
| @@ -438,7 +438,7 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj, · 247 unmodified lines | ||
| 438 | 438 | memset(&cache->node, 0, sizeof(cache->node)); |
| 439 | 439 | ret = drm_mm_insert_node_in_range_generic |
| 440 | 440 | (&ggtt->base.mm, &cache->node, |
| 441 | − 4096, 0, I915_COLOR_UNEVICTABLE, | |
| 441 | + PAGE_SIZE, 0, I915_COLOR_UNEVICTABLE, | |
| 442 | 442 | 0, ggtt->mappable_end, |
| 443 | 443 | DRM_MM_SEARCH_DEFAULT, |
| 444 | 444 | DRM_MM_CREATE_DEFAULT); |
| @@ -851,8 +851,7 @@ eb_vma_misplaced(struct i915_vma *vma) · 406 unmodified lines | ||
| 851 | 851 | WARN_ON(entry->flags & __EXEC_OBJECT_NEEDS_MAP && |
| 852 | 852 | !i915_vma_is_ggtt(vma)); |
| 853 | 853 | |
| 854 | − if (entry->alignment && | |
| 855 | − vma->node.start & (entry->alignment - 1)) | |
| 854 | + if (entry->alignment && !IS_ALIGNED(vma->node.start, entry->alignment)) | |
| 856 | 855 | return true; |
| 857 | 856 | |
| 858 | 857 | if (vma->node.size < entry->pad_to_size) |
| @@ -77,16 +77,17 @@ static void i965_write_fence_reg(struct drm_i915_fence_reg *fence, · 76 unmodified lines | ||
| 77 | 77 | |
| 78 | 78 | val = 0; |
| 79 | 79 | if (vma) { |
| 80 | − unsigned int tiling = i915_gem_object_get_tiling(vma->obj); | |
| 81 | − bool is_y_tiled = tiling == I915_TILING_Y; | |
| 82 | 80 | unsigned int stride = i915_gem_object_get_stride(vma->obj); |
| 83 | − u32 row_size = stride * (is_y_tiled ? 32 : 8); | |
| 84 | − u32 size = rounddown((u32)vma->node.size, row_size); | |
| 85 | 81 | |
| 86 | − val = ((vma->node.start + size - 4096) & 0xfffff000) << 32; | |
| 87 | − val |= vma->node.start & 0xfffff000; | |
| 82 | + GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma)); | |
| 83 | + GEM_BUG_ON(!IS_ALIGNED(vma->node.start, I965_FENCE_PAGE)); | |
| 84 | + GEM_BUG_ON(!IS_ALIGNED(vma->fence_size, I965_FENCE_PAGE)); | |
| 85 | + GEM_BUG_ON(!IS_ALIGNED(stride, 128)); | |
| 86 | + | |
| 87 | + val = (vma->node.start + vma->fence_size - I965_FENCE_PAGE) << 32; | |
| 88 | + val |= vma->node.start; | |
| 88 | 89 | val |= (u64)((stride / 128) - 1) << fence_pitch_shift; |
| 89 | − if (is_y_tiled) | |
| 90 | + if (i915_gem_object_get_tiling(vma->obj) == I915_TILING_Y) | |
| 90 | 91 | val |= BIT(I965_FENCE_TILING_Y_SHIFT); |
| 91 | 92 | val |= I965_FENCE_REG_VALID; |
| 92 | 93 | } |
| @@ -122,31 +123,24 @@ static void i915_write_fence_reg(struct drm_i915_fence_reg *fence, · 29 unmodified lines | ||
| 122 | 123 | unsigned int tiling = i915_gem_object_get_tiling(vma->obj); |
| 123 | 124 | bool is_y_tiled = tiling == I915_TILING_Y; |
| 124 | 125 | unsigned int stride = i915_gem_object_get_stride(vma->obj); |
| 125 | − int pitch_val; | |
| 126 | − int tile_width; | |
| 127 | 126 | |
| 128 | − WARN((vma->node.start & ~I915_FENCE_START_MASK) || | |
| 129 | − !is_power_of_2(vma->node.size) || | |
| 130 | − (vma->node.start & (vma->node.size - 1)), | |
| 131 | − "object 0x%08llx [fenceable? %d] not 1M or pot-size (0x%08llx) aligned\n", | |
| 132 | − vma->node.start, | |
| 133 | − i915_vma_is_map_and_fenceable(vma), | |
| 134 | − vma->node.size); | |
| 127 | + GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma)); | |
| 128 | + GEM_BUG_ON(vma->node.start & ~I915_FENCE_START_MASK); | |
| 129 | + GEM_BUG_ON(!is_power_of_2(vma->fence_size)); | |
| 130 | + GEM_BUG_ON(!IS_ALIGNED(vma->node.start, vma->fence_size)); | |
| 135 | 131 | |
| 136 | 132 | if (is_y_tiled && HAS_128_BYTE_Y_TILING(fence->i915)) |
| 137 | − tile_width = 128; | |
| 133 | + stride /= 128; | |
| 138 | 134 | else |
| 139 | − tile_width = 512; | |
| 140 | − | |
| 141 | − /* Note: pitch better be a power of two tile widths */ | |
| 142 | − pitch_val = stride / tile_width; | |
| 143 | − pitch_val = ffs(pitch_val) - 1; | |
| 135 | + stride /= 512; | |
| 136 | + GEM_BUG_ON(!is_power_of_2(stride)); | |
| 144 | 137 | |
| 145 | 138 | val = vma->node.start; |
| 146 | 139 | if (is_y_tiled) |
| 147 | 140 | val |= BIT(I830_FENCE_TILING_Y_SHIFT); |
| 148 | − val |= I915_FENCE_SIZE_BITS(vma->node.size); | |
| 149 | − val |= pitch_val << I830_FENCE_PITCH_SHIFT; | |
| 141 | + val |= I915_FENCE_SIZE_BITS(vma->fence_size); | |
| 142 | + val |= ilog2(stride) << I830_FENCE_PITCH_SHIFT; | |
| 143 | + | |
| 150 | 144 | val |= I830_FENCE_REG_VALID; |
| 151 | 145 | } |
| 152 | 146 | |
| @@ -166,25 +160,19 @@ static void i830_write_fence_reg(struct drm_i915_fence_reg *fence, · 13 unmodified lines | ||
| 166 | 160 | |
| 167 | 161 | val = 0; |
| 168 | 162 | if (vma) { |
| 169 | − unsigned int tiling = i915_gem_object_get_tiling(vma->obj); | |
| 170 | − bool is_y_tiled = tiling == I915_TILING_Y; | |
| 171 | 163 | unsigned int stride = i915_gem_object_get_stride(vma->obj); |
| 172 | − u32 pitch_val; | |
| 173 | 164 | |
| 174 | − WARN((vma->node.start & ~I830_FENCE_START_MASK) || | |
| 175 | − !is_power_of_2(vma->node.size) || | |
| 176 | − (vma->node.start & (vma->node.size - 1)), | |
| 177 | − "object 0x%08llx not 512K or pot-size 0x%08llx aligned\n", | |
| 178 | − vma->node.start, vma->node.size); | |
| 179 | − | |
| 180 | − pitch_val = stride / 128; | |
| 181 | − pitch_val = ffs(pitch_val) - 1; | |
| 165 | + GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma)); | |
| 166 | + GEM_BUG_ON(vma->node.start & ~I830_FENCE_START_MASK); | |
| 167 | + GEM_BUG_ON(!is_power_of_2(vma->fence_size)); | |
| 168 | + GEM_BUG_ON(!is_power_of_2(stride / 128)); | |
| 169 | + GEM_BUG_ON(!IS_ALIGNED(vma->node.start, vma->fence_size)); | |
| 182 | 170 | |
| 183 | 171 | val = vma->node.start; |
| 184 | − if (is_y_tiled) | |
| 172 | + if (i915_gem_object_get_tiling(vma->obj) == I915_TILING_Y) | |
| 185 | 173 | val |= BIT(I830_FENCE_TILING_Y_SHIFT); |
| 186 | − val |= I830_FENCE_SIZE_BITS(vma->node.size); | |
| 187 | − val |= pitch_val << I830_FENCE_PITCH_SHIFT; | |
| 174 | + val |= I830_FENCE_SIZE_BITS(vma->fence_size); | |
| 175 | + val |= ilog2(stride / 128) << I830_FENCE_PITCH_SHIFT; | |
| 188 | 176 | val |= I830_FENCE_REG_VALID; |
| 189 | 177 | } |
| 190 | 178 | |
| @@ -30,6 +30,8 @@ · 29 unmodified lines | ||
| 30 | 30 | struct drm_i915_private; |
| 31 | 31 | struct i915_vma; |
| 32 | 32 | |
| 33 | +#define I965_FENCE_PAGE 4096UL | |
| 34 | + | |
| 33 | 35 | struct drm_i915_fence_reg { |
| 34 | 36 | struct list_head link; |
| 35 | 37 | struct drm_i915_private *i915; |
| @@ -23,10 +23,14 @@ · 22 unmodified lines | ||
| 23 | 23 | * |
| 24 | 24 | */ |
| 25 | 25 | |
| 26 | +#include <linux/log2.h> | |
| 27 | +#include <linux/random.h> | |
| 26 | 28 | #include <linux/seq_file.h> |
| 27 | 29 | #include <linux/stop_machine.h> |
| 30 | + | |
| 28 | 31 | #include <drm/drmP.h> |
| 29 | 32 | #include <drm/i915_drm.h> |
| 33 | + | |
| 30 | 34 | #include "i915_drv.h" |
| 31 | 35 | #include "i915_vgpu.h" |
| 32 | 36 | #include "i915_trace.h" |
| @@ -99,12 +103,29 @@ · 66 unmodified lines | ||
| 99 | 103 | static int |
| 100 | 104 | i915_get_ggtt_vma_pages(struct i915_vma *vma); |
| 101 | 105 | |
| 102 | −const struct i915_ggtt_view i915_ggtt_view_normal = { | |
| 103 | − .type = I915_GGTT_VIEW_NORMAL, | |
| 104 | −}; | |
| 105 | −const struct i915_ggtt_view i915_ggtt_view_rotated = { | |
| 106 | − .type = I915_GGTT_VIEW_ROTATED, | |
| 107 | −}; | |
| 106 | +static void gen6_ggtt_invalidate(struct drm_i915_private *dev_priv) | |
| 107 | +{ | |
| 108 | + /* Note that as an uncached mmio write, this should flush the | |
| 109 | + * WCB of the writes into the GGTT before it triggers the invalidate. | |
| 110 | + */ | |
| 111 | + I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN); | |
| 112 | +} | |
| 113 | + | |
| 114 | +static void guc_ggtt_invalidate(struct drm_i915_private *dev_priv) | |
| 115 | +{ | |
| 116 | + gen6_ggtt_invalidate(dev_priv); | |
| 117 | + I915_WRITE(GEN8_GTCR, GEN8_GTCR_INVALIDATE); | |
| 118 | +} | |
| 119 | + | |
| 120 | +static void gmch_ggtt_invalidate(struct drm_i915_private *dev_priv) | |
| 121 | +{ | |
| 122 | + intel_gtt_chipset_flush(); | |
| 123 | +} | |
| 124 | + | |
| 125 | +static inline void i915_ggtt_invalidate(struct drm_i915_private *i915) | |
| 126 | +{ | |
| 127 | + i915->ggtt.invalidate(i915); | |
| 128 | +} | |
| 108 | 129 | |
| 109 | 130 | int intel_sanitize_enable_ppgtt(struct drm_i915_private *dev_priv, |
| 110 | 131 | int enable_ppgtt) |
| @@ -329,7 +350,7 @@ static int __setup_page_dma(struct drm_i915_private *dev_priv, · 218 unmodified lines | ||
| 329 | 350 | return -ENOMEM; |
| 330 | 351 | |
| 331 | 352 | p->daddr = dma_map_page(kdev, |
| 332 | − p->page, 0, 4096, PCI_DMA_BIDIRECTIONAL); | |
| 353 | + p->page, 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); | |
| 333 | 354 | |
| 334 | 355 | if (dma_mapping_error(kdev, p->daddr)) { |
| 335 | 356 | __free_page(p->page); |
| @@ -353,7 +374,7 @@ static void cleanup_page_dma(struct drm_i915_private *dev_priv, · 17 unmodified lines | ||
| 353 | 374 | if (WARN_ON(!p->page)) |
| 354 | 375 | return; |
| 355 | 376 | |
| 356 | − dma_unmap_page(&pdev->dev, p->daddr, 4096, PCI_DMA_BIDIRECTIONAL); | |
| 377 | + dma_unmap_page(&pdev->dev, p->daddr, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); | |
| 357 | 378 | __free_page(p->page); |
| 358 | 379 | memset(p, 0, sizeof(*p)); |
| 359 | 380 | } |
| @@ -626,10 +647,10 @@ static void gen8_initialize_pml4(struct i915_address_space *vm, · 266 unmodified lines | ||
| 626 | 647 | } |
| 627 | 648 | |
| 628 | 649 | static void |
| 629 | −gen8_setup_page_directory(struct i915_hw_ppgtt *ppgtt, | |
| 630 | − struct i915_page_directory_pointer *pdp, | |
| 631 | − struct i915_page_directory *pd, | |
| 632 | − int index) | |
| 650 | +gen8_setup_pdpe(struct i915_hw_ppgtt *ppgtt, | |
| 651 | + struct i915_page_directory_pointer *pdp, | |
| 652 | + struct i915_page_directory *pd, | |
| 653 | + int index) | |
| 633 | 654 | { |
| 634 | 655 | gen8_ppgtt_pdpe_t *page_directorypo; |
| 635 | 656 | |
| @@ -642,10 +663,10 @@ gen8_setup_page_directory(struct i915_hw_ppgtt *ppgtt, · 6 unmodified lines | ||
| 642 | 663 | } |
| 643 | 664 | |
| 644 | 665 | static void |
| 645 | −gen8_setup_page_directory_pointer(struct i915_hw_ppgtt *ppgtt, | |
| 646 | − struct i915_pml4 *pml4, | |
| 647 | − struct i915_page_directory_pointer *pdp, | |
| 648 | − int index) | |
| 666 | +gen8_setup_pml4e(struct i915_hw_ppgtt *ppgtt, | |
| 667 | + struct i915_pml4 *pml4, | |
| 668 | + struct i915_page_directory_pointer *pdp, | |
| 669 | + int index) | |
| 649 | 670 | { |
| 650 | 671 | gen8_ppgtt_pml4e_t *pagemap = kmap_px(pml4); |
| 651 | 672 | |
| @@ -793,9 +814,6 @@ static bool gen8_ppgtt_clear_pdp(struct i915_address_space *vm, · 141 unmodified lines | ||
| 793 | 814 | struct i915_hw_ppgtt *ppgtt = i915_vm_to_ppgtt(vm); |
| 794 | 815 | struct i915_page_directory *pd; |
| 795 | 816 | uint64_t pdpe; |
| 796 | − gen8_ppgtt_pdpe_t *pdpe_vaddr; | |
| 797 | − gen8_ppgtt_pdpe_t scratch_pdpe = | |
| 798 | − gen8_pdpe_encode(px_dma(vm->scratch_pd), I915_CACHE_LLC); | |
| 799 | 817 | |
| 800 | 818 | gen8_for_each_pdpe(pd, pdp, start, length, pdpe) { |
| 801 | 819 | if (WARN_ON(!pdp->page_directory[pdpe])) |
| @@ -803,11 +821,7 @@ static bool gen8_ppgtt_clear_pdp(struct i915_address_space *vm, · 1 unmodified line | ||
| 803 | 821 | |
| 804 | 822 | if (gen8_ppgtt_clear_pd(vm, pd, start, length)) { |
| 805 | 823 | __clear_bit(pdpe, pdp->used_pdpes); |
| 806 | − if (USES_FULL_48BIT_PPGTT(dev_priv)) { | |
| 807 | − pdpe_vaddr = kmap_px(pdp); | |
| 808 | − pdpe_vaddr[pdpe] = scratch_pdpe; | |
| 809 | − kunmap_px(ppgtt, pdpe_vaddr); | |
| 810 | − } | |
| 824 | + gen8_setup_pdpe(ppgtt, pdp, vm->scratch_pd, pdpe); | |
| 811 | 825 | free_pd(vm->i915, pd); |
| 812 | 826 | } |
| 813 | 827 | } |
| @@ -832,9 +846,6 @@ static void gen8_ppgtt_clear_pml4(struct i915_address_space *vm, · 18 unmodified lines | ||
| 832 | 846 | struct i915_hw_ppgtt *ppgtt = i915_vm_to_ppgtt(vm); |
| 833 | 847 | struct i915_page_directory_pointer *pdp; |
| 834 | 848 | uint64_t pml4e; |
| 835 | − gen8_ppgtt_pml4e_t *pml4e_vaddr; | |
| 836 | − gen8_ppgtt_pml4e_t scratch_pml4e = | |
| 837 | − gen8_pml4e_encode(px_dma(vm->scratch_pdp), I915_CACHE_LLC); | |
| 838 | 849 | |
| 839 | 850 | GEM_BUG_ON(!USES_FULL_48BIT_PPGTT(vm->i915)); |
| 840 | 851 | |
| @@ -844,9 +855,7 @@ static void gen8_ppgtt_clear_pml4(struct i915_address_space *vm, · 3 unmodified lines | ||
| 844 | 855 | |
| 845 | 856 | if (gen8_ppgtt_clear_pdp(vm, pdp, start, length)) { |
| 846 | 857 | __clear_bit(pml4e, pml4->used_pml4es); |
| 847 | − pml4e_vaddr = kmap_px(pml4); | |
| 848 | − pml4e_vaddr[pml4e] = scratch_pml4e; | |
| 849 | − kunmap_px(ppgtt, pml4e_vaddr); | |
| 858 | + gen8_setup_pml4e(ppgtt, pml4, vm->scratch_pdp, pml4e); | |
| 850 | 859 | free_pdp(vm->i915, pdp); |
| 851 | 860 | } |
| 852 | 861 | } |
| @@ -1366,7 +1375,7 @@ static int gen8_alloc_va_range_3lvl(struct i915_address_space *vm, · 513 unmodified lines | ||
| 1366 | 1375 | |
| 1367 | 1376 | kunmap_px(ppgtt, page_directory); |
| 1368 | 1377 | __set_bit(pdpe, pdp->used_pdpes); |
| 1369 | − gen8_setup_page_directory(ppgtt, pdp, pd, pdpe); | |
| 1378 | + gen8_setup_pdpe(ppgtt, pdp, pd, pdpe); | |
| 1370 | 1379 | } |
| 1371 | 1380 | |
| 1372 | 1381 | free_gen8_temp_bitmaps(new_page_dirs, new_page_tables); |
| @@ -1425,7 +1434,7 @@ static int gen8_alloc_va_range_4lvl(struct i915_address_space *vm, · 52 unmodified lines | ||
| 1425 | 1434 | if (ret) |
| 1426 | 1435 | goto err_out; |
| 1427 | 1436 | |
| 1428 | − gen8_setup_page_directory_pointer(ppgtt, pml4, pdp, pml4e); | |
| 1437 | + gen8_setup_pml4e(ppgtt, pml4, pdp, pml4e); | |
| 1429 | 1438 | } |
| 1430 | 1439 | |
| 1431 | 1440 | bitmap_or(pml4->used_pml4es, new_pdps, pml4->used_pml4es, |
| @@ -2044,7 +2053,6 @@ static int gen6_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt) · 612 unmodified lines | ||
| 2044 | 2053 | struct i915_address_space *vm = &ppgtt->base; |
| 2045 | 2054 | struct drm_i915_private *dev_priv = ppgtt->base.i915; |
| 2046 | 2055 | struct i915_ggtt *ggtt = &dev_priv->ggtt; |
| 2047 | − bool retried = false; | |
| 2048 | 2056 | int ret; |
| 2049 | 2057 | |
| 2050 | 2058 | /* PPGTT PDEs reside in the GGTT and consists of 512 entries. The |
| @@ -2057,29 +2065,14 @@ static int gen6_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt) · 6 unmodified lines | ||
| 2057 | 2065 | if (ret) |
| 2058 | 2066 | return ret; |
| 2059 | 2067 | |
| 2060 | −alloc: | |
| 2061 | − ret = drm_mm_insert_node_in_range_generic(&ggtt->base.mm, &ppgtt->node, | |
| 2062 | − GEN6_PD_SIZE, GEN6_PD_ALIGN, | |
| 2063 | − I915_COLOR_UNEVICTABLE, | |
| 2064 | − 0, ggtt->base.total, | |
| 2065 | − DRM_MM_TOPDOWN); | |
| 2066 | − if (ret == -ENOSPC && !retried) { | |
| 2067 | − ret = i915_gem_evict_something(&ggtt->base, | |
| 2068 | − GEN6_PD_SIZE, GEN6_PD_ALIGN, | |
| 2069 | − I915_COLOR_UNEVICTABLE, | |
| 2070 | − 0, ggtt->base.total, | |
| 2071 | − 0); | |
| 2072 | − if (ret) | |
| 2073 | − goto err_out; | |
| 2074 | − | |
| 2075 | − retried = true; | |
| 2076 | − goto alloc; | |
| 2077 | − } | |
| 2078 | − | |
| 2068 | + ret = i915_gem_gtt_insert(&ggtt->base, &ppgtt->node, | |
| 2069 | + GEN6_PD_SIZE, GEN6_PD_ALIGN, | |
| 2070 | + I915_COLOR_UNEVICTABLE, | |
| 2071 | + 0, ggtt->base.total, | |
| 2072 | + PIN_HIGH); | |
| 2079 | 2073 | if (ret) |
| 2080 | 2074 | goto err_out; |
| 2081 | 2075 | |
| 2082 | − | |
| 2083 | 2076 | if (ppgtt->node.start < ggtt->mappable_end) |
| 2084 | 2077 | DRM_DEBUG("Forced to use aperture for PDEs\n"); |
| 2085 | 2078 | |
| @@ -2267,6 +2260,27 @@ i915_ppgtt_create(struct drm_i915_private *dev_priv, · 181 unmodified lines | ||
| 2267 | 2260 | return ppgtt; |
| 2268 | 2261 | } |
| 2269 | 2262 | |
| 2263 | +void i915_ppgtt_close(struct i915_address_space *vm) | |
| 2264 | +{ | |
| 2265 | + struct list_head *phases[] = { | |
| 2266 | + &vm->active_list, | |
| 2267 | + &vm->inactive_list, | |
| 2268 | + &vm->unbound_list, | |
| 2269 | + NULL, | |
| 2270 | + }, **phase; | |
| 2271 | + | |
| 2272 | + GEM_BUG_ON(vm->closed); | |
| 2273 | + vm->closed = true; | |
| 2274 | + | |
| 2275 | + for (phase = phases; *phase; phase++) { | |
| 2276 | + struct i915_vma *vma, *vn; | |
| 2277 | + | |
| 2278 | + list_for_each_entry_safe(vma, vn, *phase, vm_link) | |
| 2279 | + if (!i915_vma_is_closed(vma)) | |
| 2280 | + i915_vma_close(vma); | |
| 2281 | + } | |
| 2282 | +} | |
| 2283 | + | |
| 2270 | 2284 | void i915_ppgtt_release(struct kref *kref) |
| 2271 | 2285 | { |
| 2272 | 2286 | struct i915_hw_ppgtt *ppgtt = |
| @@ -2331,16 +2345,6 @@ void i915_check_and_clear_faults(struct drm_i915_private *dev_priv) · 58 unmodified lines | ||
| 2331 | 2345 | POSTING_READ(RING_FAULT_REG(dev_priv->engine[RCS])); |
| 2332 | 2346 | } |
| 2333 | 2347 | |
| 2334 | −static void i915_ggtt_flush(struct drm_i915_private *dev_priv) | |
| 2335 | −{ | |
| 2336 | − if (INTEL_INFO(dev_priv)->gen < 6) { | |
| 2337 | − intel_gtt_chipset_flush(); | |
| 2338 | − } else { | |
| 2339 | − I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN); | |
| 2340 | − POSTING_READ(GFX_FLSH_CNTL_GEN6); | |
| 2341 | − } | |
| 2342 | −} | |
| 2343 | − | |
| 2344 | 2348 | void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv) |
| 2345 | 2349 | { |
| 2346 | 2350 | struct i915_ggtt *ggtt = &dev_priv->ggtt; |
| @@ -2355,7 +2359,7 @@ void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv) · 8 unmodified lines | ||
| 2355 | 2359 | |
| 2356 | 2360 | ggtt->base.clear_range(&ggtt->base, ggtt->base.start, ggtt->base.total); |
| 2357 | 2361 | |
| 2358 | − i915_ggtt_flush(dev_priv); | |
| 2362 | + i915_ggtt_invalidate(dev_priv); | |
| 2359 | 2363 | } |
| 2360 | 2364 | |
| 2361 | 2365 | int i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj, |
| @@ -2394,15 +2398,13 @@ static void gen8_ggtt_insert_page(struct i915_address_space *vm, · 32 unmodified lines | ||
| 2394 | 2398 | enum i915_cache_level level, |
| 2395 | 2399 | u32 unused) |
| 2396 | 2400 | { |
| 2397 | − struct drm_i915_private *dev_priv = vm->i915; | |
| 2401 | + struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); | |
| 2398 | 2402 | gen8_pte_t __iomem *pte = |
| 2399 | − (gen8_pte_t __iomem *)dev_priv->ggtt.gsm + | |
| 2400 | − (offset >> PAGE_SHIFT); | |
| 2403 | + (gen8_pte_t __iomem *)ggtt->gsm + (offset >> PAGE_SHIFT); | |
| 2401 | 2404 | |
| 2402 | 2405 | gen8_set_pte(pte, gen8_pte_encode(addr, level)); |
| 2403 | 2406 | |
| 2404 | − I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN); | |
| 2405 | − POSTING_READ(GFX_FLSH_CNTL_GEN6); | |
| 2407 | + ggtt->invalidate(vm->i915); | |
| 2406 | 2408 | } |
| 2407 | 2409 | |
| 2408 | 2410 | static void gen8_ggtt_insert_entries(struct i915_address_space *vm, |
| @@ -2410,7 +2412,6 @@ static void gen8_ggtt_insert_entries(struct i915_address_space *vm, · 1 unmodified line | ||
| 2410 | 2412 | uint64_t start, |
| 2411 | 2413 | enum i915_cache_level level, u32 unused) |
| 2412 | 2414 | { |
| 2413 | − struct drm_i915_private *dev_priv = vm->i915; | |
| 2414 | 2415 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
| 2415 | 2416 | struct sgt_iter sgt_iter; |
| 2416 | 2417 | gen8_pte_t __iomem *gtt_entries; |
| @@ -2439,8 +2440,7 @@ static void gen8_ggtt_insert_entries(struct i915_address_space *vm, · 22 unmodified lines | ||
| 2439 | 2440 | * want to flush the TLBs only after we're certain all the PTE updates |
| 2440 | 2441 | * have finished. |
| 2441 | 2442 | */ |
| 2442 | − I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN); | |
| 2443 | − POSTING_READ(GFX_FLSH_CNTL_GEN6); | |
| 2443 | + ggtt->invalidate(vm->i915); | |
| 2444 | 2444 | } |
| 2445 | 2445 | |
| 2446 | 2446 | struct insert_entries { |
| @@ -2475,15 +2475,13 @@ static void gen6_ggtt_insert_page(struct i915_address_space *vm, · 28 unmodified lines | ||
| 2475 | 2475 | enum i915_cache_level level, |
| 2476 | 2476 | u32 flags) |
| 2477 | 2477 | { |
| 2478 | − struct drm_i915_private *dev_priv = vm->i915; | |
| 2478 | + struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); | |
| 2479 | 2479 | gen6_pte_t __iomem *pte = |
| 2480 | − (gen6_pte_t __iomem *)dev_priv->ggtt.gsm + | |
| 2481 | − (offset >> PAGE_SHIFT); | |
| 2480 | + (gen6_pte_t __iomem *)ggtt->gsm + (offset >> PAGE_SHIFT); | |
| 2482 | 2481 | |
| 2483 | 2482 | iowrite32(vm->pte_encode(addr, level, flags), pte); |
| 2484 | 2483 | |
| 2485 | − I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN); | |
| 2486 | − POSTING_READ(GFX_FLSH_CNTL_GEN6); | |
| 2484 | + ggtt->invalidate(vm->i915); | |
| 2487 | 2485 | } |
| 2488 | 2486 | |
| 2489 | 2487 | /* |
| @@ -2497,7 +2495,6 @@ static void gen6_ggtt_insert_entries(struct i915_address_space *vm, · 7 unmodified lines | ||
| 2497 | 2495 | uint64_t start, |
| 2498 | 2496 | enum i915_cache_level level, u32 flags) |
| 2499 | 2497 | { |
| 2500 | − struct drm_i915_private *dev_priv = vm->i915; | |
| 2501 | 2498 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
| 2502 | 2499 | struct sgt_iter sgt_iter; |
| 2503 | 2500 | gen6_pte_t __iomem *gtt_entries; |
| @@ -2525,8 +2522,7 @@ static void gen6_ggtt_insert_entries(struct i915_address_space *vm, · 21 unmodified lines | ||
| 2525 | 2522 | * want to flush the TLBs only after we're certain all the PTE updates |
| 2526 | 2523 | * have finished. |
| 2527 | 2524 | */ |
| 2528 | − I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN); | |
| 2529 | − POSTING_READ(GFX_FLSH_CNTL_GEN6); | |
| 2525 | + ggtt->invalidate(vm->i915); | |
| 2530 | 2526 | } |
| 2531 | 2527 | |
| 2532 | 2528 | static void nop_clear_range(struct i915_address_space *vm, |
| @@ -2723,11 +2719,11 @@ static void i915_gtt_color_adjust(const struct drm_mm_node *node, · 190 unmodified lines | ||
| 2723 | 2719 | u64 *end) |
| 2724 | 2720 | { |
| 2725 | 2721 | if (node->color != color) |
| 2726 | − *start += 4096; | |
| 2722 | + *start += I915_GTT_PAGE_SIZE; | |
| 2727 | 2723 | |
| 2728 | 2724 | node = list_next_entry(node, node_list); |
| 2729 | 2725 | if (node->allocated && node->color != color) |
| 2730 | − *end -= 4096; | |
| 2726 | + *end -= I915_GTT_PAGE_SIZE; | |
| 2731 | 2727 | } |
| 2732 | 2728 | |
| 2733 | 2729 | int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) |
| @@ -2754,7 +2750,7 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) · 20 unmodified lines | ||
| 2754 | 2750 | /* Reserve a mappable slot for our lockless error capture */ |
| 2755 | 2751 | ret = drm_mm_insert_node_in_range_generic(&ggtt->base.mm, |
| 2756 | 2752 | &ggtt->error_capture, |
| 2757 | − 4096, 0, | |
| 2753 | + PAGE_SIZE, 0, | |
| 2758 | 2754 | I915_COLOR_UNEVICTABLE, |
| 2759 | 2755 | 0, ggtt->mappable_end, |
| 2760 | 2756 | 0, 0); |
| @@ -3086,6 +3082,8 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt) · 325 unmodified lines | ||
| 3086 | 3082 | if (IS_CHERRYVIEW(dev_priv)) |
| 3087 | 3083 | ggtt->base.insert_entries = gen8_ggtt_insert_entries__BKL; |
| 3088 | 3084 | |
| 3085 | + ggtt->invalidate = gen6_ggtt_invalidate; | |
| 3086 | + | |
| 3089 | 3087 | return ggtt_probe_common(ggtt, size); |
| 3090 | 3088 | } |
| 3091 | 3089 | |
| @@ -3123,6 +3121,8 @@ static int gen6_gmch_probe(struct i915_ggtt *ggtt) · 31 unmodified lines | ||
| 3123 | 3121 | ggtt->base.unbind_vma = ggtt_unbind_vma; |
| 3124 | 3122 | ggtt->base.cleanup = gen6_gmch_remove; |
| 3125 | 3123 | |
| 3124 | + ggtt->invalidate = gen6_ggtt_invalidate; | |
| 3125 | + | |
| 3126 | 3126 | if (HAS_EDRAM(dev_priv)) |
| 3127 | 3127 | ggtt->base.pte_encode = iris_pte_encode; |
| 3128 | 3128 | else if (IS_HASWELL(dev_priv)) |
| @@ -3166,6 +3166,8 @@ static int i915_gmch_probe(struct i915_ggtt *ggtt) · 37 unmodified lines | ||
| 3166 | 3166 | ggtt->base.unbind_vma = ggtt_unbind_vma; |
| 3167 | 3167 | ggtt->base.cleanup = i915_gmch_remove; |
| 3168 | 3168 | |
| 3169 | + ggtt->invalidate = gmch_ggtt_invalidate; | |
| 3170 | + | |
| 3169 | 3171 | if (unlikely(ggtt->do_idle_maps)) |
| 3170 | 3172 | DRM_INFO("applying Ironlake quirks for intel_iommu\n"); |
| 3171 | 3173 | |
| @@ -3284,6 +3286,16 @@ int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv) · 112 unmodified lines | ||
| 3284 | 3286 | return 0; |
| 3285 | 3287 | } |
| 3286 | 3288 | |
| 3289 | +void i915_ggtt_enable_guc(struct drm_i915_private *i915) | |
| 3290 | +{ | |
| 3291 | + i915->ggtt.invalidate = guc_ggtt_invalidate; | |
| 3292 | +} | |
| 3293 | + | |
| 3294 | +void i915_ggtt_disable_guc(struct drm_i915_private *i915) | |
| 3295 | +{ | |
| 3296 | + i915->ggtt.invalidate = gen6_ggtt_invalidate; | |
| 3297 | +} | |
| 3298 | + | |
| 3287 | 3299 | void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv) |
| 3288 | 3300 | { |
| 3289 | 3301 | struct i915_ggtt *ggtt = &dev_priv->ggtt; |
| @@ -3347,52 +3359,7 @@ void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv) · 57 unmodified lines | ||
| 3347 | 3359 | } |
| 3348 | 3360 | } |
| 3349 | 3361 | |
| 3350 | − i915_ggtt_flush(dev_priv); | |
| 3351 | −} | |
| 3352 | − | |
| 3353 | −struct i915_vma * | |
| 3354 | −i915_gem_obj_to_vma(struct drm_i915_gem_object *obj, | |
| 3355 | − struct i915_address_space *vm, | |
| 3356 | − const struct i915_ggtt_view *view) | |
| 3357 | −{ | |
| 3358 | − struct rb_node *rb; | |
| 3359 | − | |
| 3360 | − rb = obj->vma_tree.rb_node; | |
| 3361 | − while (rb) { | |
| 3362 | − struct i915_vma *vma = rb_entry(rb, struct i915_vma, obj_node); | |
| 3363 | − long cmp; | |
| 3364 | − | |
| 3365 | − cmp = i915_vma_compare(vma, vm, view); | |
| 3366 | − if (cmp == 0) | |
| 3367 | − return vma; | |
| 3368 | − | |
| 3369 | − if (cmp < 0) | |
| 3370 | − rb = rb->rb_right; | |
| 3371 | − else | |
| 3372 | − rb = rb->rb_left; | |
| 3373 | − } | |
| 3374 | − | |
| 3375 | − return NULL; | |
| 3376 | −} | |
| 3377 | − | |
| 3378 | −struct i915_vma * | |
| 3379 | −i915_gem_obj_lookup_or_create_vma(struct drm_i915_gem_object *obj, | |
| 3380 | − struct i915_address_space *vm, | |
| 3381 | − const struct i915_ggtt_view *view) | |
| 3382 | −{ | |
| 3383 | − struct i915_vma *vma; | |
| 3384 | − | |
| 3385 | − lockdep_assert_held(&obj->base.dev->struct_mutex); | |
| 3386 | − GEM_BUG_ON(view && !i915_is_ggtt(vm)); | |
| 3387 | − | |
| 3388 | − vma = i915_gem_obj_to_vma(obj, vm, view); | |
| 3389 | − if (!vma) { | |
| 3390 | − vma = i915_vma_create(obj, vm, view); | |
| 3391 | − GEM_BUG_ON(vma != i915_gem_obj_to_vma(obj, vm, view)); | |
| 3392 | − } | |
| 3393 | − | |
| 3394 | − GEM_BUG_ON(i915_vma_is_closed(vma)); | |
| 3395 | − return vma; | |
| 3362 | + i915_ggtt_invalidate(dev_priv); | |
| 3396 | 3363 | } |
| 3397 | 3364 | |
| 3398 | 3365 | static struct scatterlist * |
| @@ -3492,7 +3459,7 @@ intel_partial_pages(const struct i915_ggtt_view *view, · 93 unmodified lines | ||
| 3492 | 3459 | { |
| 3493 | 3460 | struct sg_table *st; |
| 3494 | 3461 | struct scatterlist *sg, *iter; |
| 3495 | − unsigned int count = view->params.partial.size; | |
| 3462 | + unsigned int count = view->partial.size; | |
| 3496 | 3463 | unsigned int offset; |
| 3497 | 3464 | int ret = -ENOMEM; |
| 3498 | 3465 | |
| @@ -3504,9 +3471,7 @@ intel_partial_pages(const struct i915_ggtt_view *view, · 5 unmodified lines | ||
| 3504 | 3471 | if (ret) |
| 3505 | 3472 | goto err_sg_alloc; |
| 3506 | 3473 | |
| 3507 | − iter = i915_gem_object_get_sg(obj, | |
| 3508 | − view->params.partial.offset, | |
| 3509 | − &offset); | |
| 3474 | + iter = i915_gem_object_get_sg(obj, view->partial.offset, &offset); | |
| 3510 | 3475 | GEM_BUG_ON(!iter); |
| 3511 | 3476 | |
| 3512 | 3477 | sg = st->sgl; |
| @@ -3558,7 +3523,8 @@ i915_get_ggtt_vma_pages(struct i915_vma *vma) · 45 unmodified lines | ||
| 3558 | 3523 | vma->pages = vma->obj->mm.pages; |
| 3559 | 3524 | else if (vma->ggtt_view.type == I915_GGTT_VIEW_ROTATED) |
| 3560 | 3525 | vma->pages = |
| 3561 | − intel_rotate_fb_obj_pages(&vma->ggtt_view.params.rotated, vma->obj); | |
| 3526 | + intel_rotate_fb_obj_pages(&vma->ggtt_view.rotated, | |
| 3527 | + vma->obj); | |
| 3562 | 3528 | else if (vma->ggtt_view.type == I915_GGTT_VIEW_PARTIAL) |
| 3563 | 3529 | vma->pages = intel_partial_pages(&vma->ggtt_view, vma->obj); |
| 3564 | 3530 | else |
| @@ -3579,3 +3545,207 @@ i915_get_ggtt_vma_pages(struct i915_vma *vma) · 14 unmodified lines | ||
| 3579 | 3545 | return ret; |
| 3580 | 3546 | } |
| 3581 | 3547 | |
| 3548 | +/** | |
| 3549 | + * i915_gem_gtt_reserve - reserve a node in an address_space (GTT) | |
| 3550 | + * @vm: the &struct i915_address_space | |
| 3551 | + * @node: the &struct drm_mm_node (typically i915_vma.mode) | |
| 3552 | + * @size: how much space to allocate inside the GTT, | |
| 3553 | + * must be #I915_GTT_PAGE_SIZE aligned | |
| 3554 | + * @offset: where to insert inside the GTT, | |
| 3555 | + * must be #I915_GTT_MIN_ALIGNMENT aligned, and the node | |
| 3556 | + * (@offset + @size) must fit within the address space | |
| 3557 | + * @color: color to apply to node, if this node is not from a VMA, | |
| 3558 | + * color must be #I915_COLOR_UNEVICTABLE | |
| 3559 | + * @flags: control search and eviction behaviour | |
| 3560 | + * | |
| 3561 | + * i915_gem_gtt_reserve() tries to insert the @node at the exact @offset inside | |
| 3562 | + * the address space (using @size and @color). If the @node does not fit, it | |
| 3563 | + * tries to evict any overlapping nodes from the GTT, including any | |
| 3564 | + * neighbouring nodes if the colors do not match (to ensure guard pages between | |
| 3565 | + * differing domains). See i915_gem_evict_for_node() for the gory details | |
| 3566 | + * on the eviction algorithm. #PIN_NONBLOCK may used to prevent waiting on | |
| 3567 | + * evicting active overlapping objects, and any overlapping node that is pinned | |
| 3568 | + * or marked as unevictable will also result in failure. | |
| 3569 | + * | |
| 3570 | + * Returns: 0 on success, -ENOSPC if no suitable hole is found, -EINTR if | |
| 3571 | + * asked to wait for eviction and interrupted. | |
| 3572 | + */ | |
| 3573 | +int i915_gem_gtt_reserve(struct i915_address_space *vm, | |
| 3574 | + struct drm_mm_node *node, | |
| 3575 | + u64 size, u64 offset, unsigned long color, | |
| 3576 | + unsigned int flags) | |
| 3577 | +{ | |
| 3578 | + int err; | |
| 3579 | + | |
| 3580 | + GEM_BUG_ON(!size); | |
| 3581 | + GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)); | |
| 3582 | + GEM_BUG_ON(!IS_ALIGNED(offset, I915_GTT_MIN_ALIGNMENT)); | |
| 3583 | + GEM_BUG_ON(range_overflows(offset, size, vm->total)); | |
| 3584 | + GEM_BUG_ON(vm == &vm->i915->mm.aliasing_ppgtt->base); | |
| 3585 | + GEM_BUG_ON(drm_mm_node_allocated(node)); | |
| 3586 | + | |
| 3587 | + node->size = size; | |
| 3588 | + node->start = offset; | |
| 3589 | + node->color = color; | |
| 3590 | + | |
| 3591 | + err = drm_mm_reserve_node(&vm->mm, node); | |
| 3592 | + if (err != -ENOSPC) | |
| 3593 | + return err; | |
| 3594 | + | |
| 3595 | + err = i915_gem_evict_for_node(vm, node, flags); | |
| 3596 | + if (err == 0) | |
| 3597 | + err = drm_mm_reserve_node(&vm->mm, node); | |
| 3598 | + | |
| 3599 | + return err; | |
| 3600 | +} | |
| 3601 | + | |
| 3602 | +static u64 random_offset(u64 start, u64 end, u64 len, u64 align) | |
| 3603 | +{ | |
| 3604 | + u64 range, addr; | |
| 3605 | + | |
| 3606 | + GEM_BUG_ON(range_overflows(start, len, end)); | |
| 3607 | + GEM_BUG_ON(round_up(start, align) > round_down(end - len, align)); | |
| 3608 | + | |
| 3609 | + range = round_down(end - len, align) - round_up(start, align); | |
| 3610 | + if (range) { | |
| 3611 | + if (sizeof(unsigned long) == sizeof(u64)) { | |
| 3612 | + addr = get_random_long(); | |
| 3613 | + } else { | |
| 3614 | + addr = get_random_int(); | |
| 3615 | + if (range > U32_MAX) { | |
| 3616 | + addr <<= 32; | |
| 3617 | + addr |= get_random_int(); | |
| 3618 | + } | |
| 3619 | + } | |
| 3620 | + div64_u64_rem(addr, range, &addr); | |
| 3621 | + start += addr; | |
| 3622 | + } | |
| 3623 | + | |
| 3624 | + return round_up(start, align); | |
| 3625 | +} | |
| 3626 | + | |
| 3627 | +/** | |
| 3628 | + * i915_gem_gtt_insert - insert a node into an address_space (GTT) | |
| 3629 | + * @vm: the &struct i915_address_space | |
| 3630 | + * @node: the &struct drm_mm_node (typically i915_vma.node) | |
| 3631 | + * @size: how much space to allocate inside the GTT, | |
| 3632 | + * must be #I915_GTT_PAGE_SIZE aligned | |
| 3633 | + * @alignment: required alignment of starting offset, may be 0 but | |
| 3634 | + * if specified, this must be a power-of-two and at least | |
| 3635 | + * #I915_GTT_MIN_ALIGNMENT | |
| 3636 | + * @color: color to apply to node | |
| 3637 | + * @start: start of any range restriction inside GTT (0 for all), | |
| 3638 | + * must be #I915_GTT_PAGE_SIZE aligned | |
| 3639 | + * @end: end of any range restriction inside GTT (U64_MAX for all), | |
| 3640 | + * must be #I915_GTT_PAGE_SIZE aligned if not U64_MAX | |
| 3641 | + * @flags: control search and eviction behaviour | |
| 3642 | + * | |
| 3643 | + * i915_gem_gtt_insert() first searches for an available hole into which | |
| 3644 | + * is can insert the node. The hole address is aligned to @alignment and | |
| 3645 | + * its @size must then fit entirely within the [@start, @end] bounds. The | |
| 3646 | + * nodes on either side of the hole must match @color, or else a guard page | |
| 3647 | + * will be inserted between the two nodes (or the node evicted). If no | |
| 3648 | + * suitable hole is found, first a victim is randomly selected and tested | |
| 3649 | + * for eviction, otherwise then the LRU list of objects within the GTT | |
| 3650 | + * is scanned to find the first set of replacement nodes to create the hole. | |
| 3651 | + * Those old overlapping nodes are evicted from the GTT (and so must be | |
| 3652 | + * rebound before any future use). Any node that is currently pinned cannot | |
| 3653 | + * be evicted (see i915_vma_pin()). Similar if the node's VMA is currently | |
| 3654 | + * active and #PIN_NONBLOCK is specified, that node is also skipped when | |
| 3655 | + * searching for an eviction candidate. See i915_gem_evict_something() for | |
| 3656 | + * the gory details on the eviction algorithm. | |
| 3657 | + * | |
| 3658 | + * Returns: 0 on success, -ENOSPC if no suitable hole is found, -EINTR if | |
| 3659 | + * asked to wait for eviction and interrupted. | |
| 3660 | + */ | |
| 3661 | +int i915_gem_gtt_insert(struct i915_address_space *vm, | |
| 3662 | + struct drm_mm_node *node, | |
| 3663 | + u64 size, u64 alignment, unsigned long color, | |
| 3664 | + u64 start, u64 end, unsigned int flags) | |
| 3665 | +{ | |
| 3666 | + u32 search_flag, alloc_flag; | |
| 3667 | + u64 offset; | |
| 3668 | + int err; | |
| 3669 | + | |
| 3670 | + lockdep_assert_held(&vm->i915->drm.struct_mutex); | |
| 3671 | + GEM_BUG_ON(!size); | |
| 3672 | + GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)); | |
| 3673 | + GEM_BUG_ON(alignment && !is_power_of_2(alignment)); | |
| 3674 | + GEM_BUG_ON(alignment && !IS_ALIGNED(alignment, I915_GTT_MIN_ALIGNMENT)); | |
| 3675 | + GEM_BUG_ON(start >= end); | |
| 3676 | + GEM_BUG_ON(start > 0 && !IS_ALIGNED(start, I915_GTT_PAGE_SIZE)); | |
| 3677 | + GEM_BUG_ON(end < U64_MAX && !IS_ALIGNED(end, I915_GTT_PAGE_SIZE)); | |
| 3678 | + GEM_BUG_ON(vm == &vm->i915->mm.aliasing_ppgtt->base); | |
| 3679 | + GEM_BUG_ON(drm_mm_node_allocated(node)); | |
| 3680 | + | |
| 3681 | + if (unlikely(range_overflows(start, size, end))) | |
| 3682 | + return -ENOSPC; | |
| 3683 | + | |
| 3684 | + if (unlikely(round_up(start, alignment) > round_down(end - size, alignment))) | |
| 3685 | + return -ENOSPC; | |
| 3686 | + | |
| 3687 | + if (flags & PIN_HIGH) { | |
| 3688 | + search_flag = DRM_MM_SEARCH_BELOW; | |
| 3689 | + alloc_flag = DRM_MM_CREATE_TOP; | |
| 3690 | + } else { | |
| 3691 | + search_flag = DRM_MM_SEARCH_DEFAULT; | |
| 3692 | + alloc_flag = DRM_MM_CREATE_DEFAULT; | |
| 3693 | + } | |
| 3694 | + | |
| 3695 | + /* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks, | |
| 3696 | + * so we know that we always have a minimum alignment of 4096. | |
| 3697 | + * The drm_mm range manager is optimised to return results | |
| 3698 | + * with zero alignment, so where possible use the optimal | |
| 3699 | + * path. | |
| 3700 | + */ | |
| 3701 | + BUILD_BUG_ON(I915_GTT_MIN_ALIGNMENT > I915_GTT_PAGE_SIZE); | |
| 3702 | + if (alignment <= I915_GTT_MIN_ALIGNMENT) | |
| 3703 | + alignment = 0; | |
| 3704 | + | |
| 3705 | + err = drm_mm_insert_node_in_range_generic(&vm->mm, node, | |
| 3706 | + size, alignment, color, | |
| 3707 | + start, end, | |
| 3708 | + search_flag, alloc_flag); | |
| 3709 | + if (err != -ENOSPC) | |
| 3710 | + return err; | |
| 3711 | + | |
| 3712 | + /* No free space, pick a slot at random. | |
| 3713 | + * | |
| 3714 | + * There is a pathological case here using a GTT shared between | |
| 3715 | + * mmap and GPU (i.e. ggtt/aliasing_ppgtt but not full-ppgtt): | |
| 3716 | + * | |
| 3717 | + * |<-- 256 MiB aperture -->||<-- 1792 MiB unmappable -->| | |
| 3718 | + * (64k objects) (448k objects) | |
| 3719 | + * | |
| 3720 | + * Now imagine that the eviction LRU is ordered top-down (just because | |
| 3721 | + * pathology meets real life), and that we need to evict an object to | |
| 3722 | + * make room inside the aperture. The eviction scan then has to walk | |
| 3723 | + * the 448k list before it finds one within range. And now imagine that | |
| 3724 | + * it has to search for a new hole between every byte inside the memcpy, | |
| 3725 | + * for several simultaneous clients. | |
| 3726 | + * | |
| 3727 | + * On a full-ppgtt system, if we have run out of available space, there | |
| 3728 | + * will be lots and lots of objects in the eviction list! Again, | |
| 3729 | + * searching that LRU list may be slow if we are also applying any | |
| 3730 | + * range restrictions (e.g. restriction to low 4GiB) and so, for | |
| 3731 | + * simplicity and similarilty between different GTT, try the single | |
| 3732 | + * random replacement first. | |
| 3733 | + */ | |
| 3734 | + offset = random_offset(start, end, | |
| 3735 | + size, alignment ?: I915_GTT_MIN_ALIGNMENT); | |
| 3736 | + err = i915_gem_gtt_reserve(vm, node, size, offset, color, flags); | |
| 3737 | + if (err != -ENOSPC) | |
| 3738 | + return err; | |
| 3739 | + | |
| 3740 | + /* Randomly selected placement is pinned, do a search */ | |
| 3741 | + err = i915_gem_evict_something(vm, size, alignment, color, | |
| 3742 | + start, end, flags); | |
| 3743 | + if (err) | |
| 3744 | + return err; | |
| 3745 | + | |
| 3746 | + search_flag = DRM_MM_SEARCH_DEFAULT; | |
| 3747 | + return drm_mm_insert_node_in_range_generic(&vm->mm, node, | |
| 3748 | + size, alignment, color, | |
| 3749 | + start, end, | |
| 3750 | + search_flag, alloc_flag); | |
| 3751 | +} | |
| @@ -40,6 +40,9 @@ · 39 unmodified lines | ||
| 40 | 40 | #include "i915_gem_timeline.h" |
| 41 | 41 | #include "i915_gem_request.h" |
| 42 | 42 | |
| 43 | +#define I915_GTT_PAGE_SIZE 4096UL | |
| 44 | +#define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE | |
| 45 | + | |
| 43 | 46 | #define I915_FENCE_REG_NONE -1 |
| 44 | 47 | #define I915_MAX_NUM_FENCES 32 |
| 45 | 48 | /* 32 fences + sign bit for FENCE_REG_NONE */ |
| @@ -142,34 +145,57 @@ typedef uint64_t gen8_ppgtt_pml4e_t; · 96 unmodified lines | ||
| 142 | 145 | |
| 143 | 146 | struct sg_table; |
| 144 | 147 | |
| 145 | −enum i915_ggtt_view_type { | |
| 146 | − I915_GGTT_VIEW_NORMAL = 0, | |
| 147 | − I915_GGTT_VIEW_ROTATED, | |
| 148 | − I915_GGTT_VIEW_PARTIAL, | |
| 149 | −}; | |
| 150 | − | |
| 151 | 148 | struct intel_rotation_info { |
| 152 | − struct { | |
| 149 | + struct intel_rotation_plane_info { | |
| 153 | 150 | /* tiles */ |
| 154 | 151 | unsigned int width, height, stride, offset; |
| 155 | 152 | } plane[2]; |
| 153 | +} __packed; | |
| 154 | + | |
| 155 | +static inline void assert_intel_rotation_info_is_packed(void) | |
| 156 | +{ | |
| 157 | + BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 8*sizeof(unsigned int)); | |
| 158 | +} | |
| 159 | + | |
| 160 | +struct intel_partial_info { | |
| 161 | + u64 offset; | |
| 162 | + unsigned int size; | |
| 163 | +} __packed; | |
| 164 | + | |
| 165 | +static inline void assert_intel_partial_info_is_packed(void) | |
| 166 | +{ | |
| 167 | + BUILD_BUG_ON(sizeof(struct intel_partial_info) != sizeof(u64) + sizeof(unsigned int)); | |
| 168 | +} | |
| 169 | + | |
| 170 | +enum i915_ggtt_view_type { | |
| 171 | + I915_GGTT_VIEW_NORMAL = 0, | |
| 172 | + I915_GGTT_VIEW_ROTATED = sizeof(struct intel_rotation_info), | |
| 173 | + I915_GGTT_VIEW_PARTIAL = sizeof(struct intel_partial_info), | |
| 156 | 174 | }; |
| 157 | 175 | |
| 176 | +static inline void assert_i915_ggtt_view_type_is_unique(void) | |
| 177 | +{ | |
| 178 | + /* As we encode the size of each branch inside the union into its type, | |
| 179 | + * we have to be careful that each branch has a unique size. | |
| 180 | + */ | |
| 181 | + switch ((enum i915_ggtt_view_type)0) { | |
| 182 | + case I915_GGTT_VIEW_NORMAL: | |
| 183 | + case I915_GGTT_VIEW_PARTIAL: | |
| 184 | + case I915_GGTT_VIEW_ROTATED: | |
| 185 | + /* gcc complains if these are identical cases */ | |
| 186 | + break; | |
| 187 | + } | |
| 188 | +} | |
| 189 | + | |
| 158 | 190 | struct i915_ggtt_view { |
| 159 | 191 | enum i915_ggtt_view_type type; |
| 160 | − | |
| 161 | 192 | union { |
| 162 | − struct { | |
| 163 | − u64 offset; | |
| 164 | − unsigned int size; | |
| 165 | − } partial; | |
| 193 | + /* Members need to contain no holes/padding */ | |
| 194 | + struct intel_partial_info partial; | |
| 166 | 195 | struct intel_rotation_info rotated; |
| 167 | − } params; | |
| 196 | + }; | |
| 168 | 197 | }; |
| 169 | 198 | |
| 170 | −extern const struct i915_ggtt_view i915_ggtt_view_normal; | |
| 171 | −extern const struct i915_ggtt_view i915_ggtt_view_rotated; | |
| 172 | − | |
| 173 | 199 | enum i915_cache_level; |
| 174 | 200 | |
| 175 | 201 | struct i915_vma; |
| @@ -333,6 +359,7 @@ struct i915_ggtt { · 157 unmodified lines | ||
| 333 | 359 | |
| 334 | 360 | /** "Graphics Stolen Memory" holds the global PTEs */ |
| 335 | 361 | void __iomem *gsm; |
| 362 | + void (*invalidate)(struct drm_i915_private *dev_priv); | |
| 336 | 363 | |
| 337 | 364 | bool do_idle_maps; |
| 338 | 365 | |
| @@ -501,6 +528,8 @@ i915_vm_to_ggtt(struct i915_address_space *vm) · 162 unmodified lines | ||
| 501 | 528 | int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv); |
| 502 | 529 | int i915_ggtt_init_hw(struct drm_i915_private *dev_priv); |
| 503 | 530 | int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv); |
| 531 | +void i915_ggtt_enable_guc(struct drm_i915_private *i915); | |
| 532 | +void i915_ggtt_disable_guc(struct drm_i915_private *i915); | |
| 504 | 533 | int i915_gem_init_ggtt(struct drm_i915_private *dev_priv); |
| 505 | 534 | void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv); |
| 506 | 535 | |
| @@ -509,6 +538,7 @@ void i915_ppgtt_release(struct kref *kref); · 2 unmodified lines | ||
| 509 | 538 | struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_i915_private *dev_priv, |
| 510 | 539 | struct drm_i915_file_private *fpriv, |
| 511 | 540 | const char *name); |
| 541 | +void i915_ppgtt_close(struct i915_address_space *vm); | |
| 512 | 542 | static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt) |
| 513 | 543 | { |
| 514 | 544 | if (ppgtt) |
| @@ -529,6 +559,16 @@ int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj, · 14 unmodified lines | ||
| 529 | 559 | void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj, |
| 530 | 560 | struct sg_table *pages); |
| 531 | 561 | |
| 562 | +int i915_gem_gtt_reserve(struct i915_address_space *vm, | |
| 563 | + struct drm_mm_node *node, | |
| 564 | + u64 size, u64 offset, unsigned long color, | |
| 565 | + unsigned int flags); | |
| 566 | + | |
| 567 | +int i915_gem_gtt_insert(struct i915_address_space *vm, | |
| 568 | + struct drm_mm_node *node, | |
| 569 | + u64 size, u64 alignment, unsigned long color, | |
| 570 | + u64 start, u64 end, unsigned int flags); | |
| 571 | + | |
| 532 | 572 | /* Flags used by pin/bind&friends. */ |
| 533 | 573 | #define PIN_NONBLOCK BIT(0) |
| 534 | 574 | #define PIN_MAPPABLE BIT(1) |
| @@ -543,6 +583,6 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj, · 8 unmodified lines | ||
| 543 | 583 | #define PIN_HIGH BIT(9) |
| 544 | 584 | #define PIN_OFFSET_BIAS BIT(10) |
| 545 | 585 | #define PIN_OFFSET_FIXED BIT(11) |
| 546 | −#define PIN_OFFSET_MASK (~4095) | |
| 586 | +#define PIN_OFFSET_MASK (-I915_GTT_PAGE_SIZE) | |
| 547 | 587 | |
| 548 | 588 | #endif |
| @@ -151,10 +151,16 @@ static const struct drm_i915_gem_object_ops i915_gem_object_internal_ops = { · 150 unmodified lines | ||
| 151 | 151 | */ |
| 152 | 152 | struct drm_i915_gem_object * |
| 153 | 153 | i915_gem_object_create_internal(struct drm_i915_private *i915, |
| 154 | − unsigned int size) | |
| 154 | + phys_addr_t size) | |
| 155 | 155 | { |
| 156 | 156 | struct drm_i915_gem_object *obj; |
| 157 | 157 | |
| 158 | + GEM_BUG_ON(!size); | |
| 159 | + GEM_BUG_ON(!IS_ALIGNED(size, PAGE_SIZE)); | |
| 160 | + | |
| 161 | + if (overflows_type(size, obj->base.size)) | |
| 162 | + return ERR_PTR(-E2BIG); | |
| 163 | + | |
| 158 | 164 | obj = i915_gem_object_alloc(i915); |
| 159 | 165 | if (!obj) |
| 160 | 166 | return ERR_PTR(-ENOMEM); |
| @@ -317,6 +317,29 @@ i915_gem_object_get_stride(struct drm_i915_gem_object *obj) · 316 unmodified lines | ||
| 317 | 317 | return obj->tiling_and_stride & STRIDE_MASK; |
| 318 | 318 | } |
| 319 | 319 | |
| 320 | +static inline unsigned int | |
| 321 | +i915_gem_tile_height(unsigned int tiling) | |
| 322 | +{ | |
| 323 | + GEM_BUG_ON(!tiling); | |
| 324 | + return tiling == I915_TILING_Y ? 32 : 8; | |
| 325 | +} | |
| 326 | + | |
| 327 | +static inline unsigned int | |
| 328 | +i915_gem_object_get_tile_height(struct drm_i915_gem_object *obj) | |
| 329 | +{ | |
| 330 | + return i915_gem_tile_height(i915_gem_object_get_tiling(obj)); | |
| 331 | +} | |
| 332 | + | |
| 333 | +static inline unsigned int | |
| 334 | +i915_gem_object_get_tile_row_size(struct drm_i915_gem_object *obj) | |
| 335 | +{ | |
| 336 | + return (i915_gem_object_get_stride(obj) * | |
| 337 | + i915_gem_object_get_tile_height(obj)); | |
| 338 | +} | |
| 339 | + | |
| 340 | +int i915_gem_object_set_tiling(struct drm_i915_gem_object *obj, | |
| 341 | + unsigned int tiling, unsigned int stride); | |
| 342 | + | |
| 320 | 343 | static inline struct intel_engine_cs * |
| 321 | 344 | i915_gem_object_last_write_engine(struct drm_i915_gem_object *obj) |
| 322 | 345 | { |
| @@ -187,20 +187,20 @@ int i915_gem_render_state_init(struct intel_engine_cs *engine) · 186 unmodified lines | ||
| 187 | 187 | if (!rodata) |
| 188 | 188 | return 0; |
| 189 | 189 | |
| 190 | − if (rodata->batch_items * 4 > 4096) | |
| 190 | + if (rodata->batch_items * 4 > PAGE_SIZE) | |
| 191 | 191 | return -EINVAL; |
| 192 | 192 | |
| 193 | 193 | so = kmalloc(sizeof(*so), GFP_KERNEL); |
| 194 | 194 | if (!so) |
| 195 | 195 | return -ENOMEM; |
| 196 | 196 | |
| 197 | − obj = i915_gem_object_create_internal(engine->i915, 4096); | |
| 197 | + obj = i915_gem_object_create_internal(engine->i915, PAGE_SIZE); | |
| 198 | 198 | if (IS_ERR(obj)) { |
| 199 | 199 | ret = PTR_ERR(obj); |
| 200 | 200 | goto err_free; |
| 201 | 201 | } |
| 202 | 202 | |
| 203 | − so->vma = i915_vma_create(obj, &engine->i915->ggtt.base, NULL); | |
| 203 | + so->vma = i915_vma_instance(obj, &engine->i915->ggtt.base, NULL); | |
| 204 | 204 | if (IS_ERR(so->vma)) { |
| 205 | 205 | ret = PTR_ERR(so->vma); |
| 206 | 206 | goto err_obj; |
| @@ -307,26 +307,6 @@ void i915_gem_request_retire_upto(struct drm_i915_gem_request *req) · 306 unmodified lines | ||
| 307 | 307 | } while (tmp != req); |
| 308 | 308 | } |
| 309 | 309 | |
| 310 | −static int i915_gem_check_wedge(struct drm_i915_private *dev_priv) | |
| 311 | −{ | |
| 312 | − struct i915_gpu_error *error = &dev_priv->gpu_error; | |
| 313 | − | |
| 314 | − if (i915_terminally_wedged(error)) | |
| 315 | − return -EIO; | |
| 316 | − | |
| 317 | − if (i915_reset_in_progress(error)) { | |
| 318 | − /* Non-interruptible callers can't handle -EAGAIN, hence return | |
| 319 | − * -EIO unconditionally for these. | |
| 320 | − */ | |
| 321 | − if (!dev_priv->mm.interruptible) | |
| 322 | − return -EIO; | |
| 323 | − | |
| 324 | − return -EAGAIN; | |
| 325 | − } | |
| 326 | − | |
| 327 | − return 0; | |
| 328 | −} | |
| 329 | − | |
| 330 | 310 | static int i915_gem_init_global_seqno(struct drm_i915_private *i915, u32 seqno) |
| 331 | 311 | { |
| 332 | 312 | struct i915_gem_timeline *timeline = &i915->gt.global_timeline; |
| @@ -521,12 +501,10 @@ i915_gem_request_alloc(struct intel_engine_cs *engine, · 188 unmodified lines | ||
| 521 | 501 | lockdep_assert_held(&dev_priv->drm.struct_mutex); |
| 522 | 502 | |
| 523 | 503 | /* ABI: Before userspace accesses the GPU (e.g. execbuffer), report |
| 524 | − * EIO if the GPU is already wedged, or EAGAIN to drop the struct_mutex | |
| 525 | − * and restart. | |
| 504 | + * EIO if the GPU is already wedged. | |
| 526 | 505 | */ |
| 527 | − ret = i915_gem_check_wedge(dev_priv); | |
| 528 | − if (ret) | |
| 529 | − return ERR_PTR(ret); | |
| 506 | + if (i915_terminally_wedged(&dev_priv->gpu_error)) | |
| 507 | + return ERR_PTR(-EIO); | |
| 530 | 508 | |
| 531 | 509 | /* Pinning the contexts may generate requests in order to acquire |
| 532 | 510 | * GGTT space, so do this first before we reserve a seqno for |
| @@ -851,6 +829,13 @@ void __i915_add_request(struct drm_i915_gem_request *request, bool flush_caches) · 318 unmodified lines | ||
| 851 | 829 | lockdep_assert_held(&request->i915->drm.struct_mutex); |
| 852 | 830 | trace_i915_gem_request_add(request); |
| 853 | 831 | |
| 832 | + /* Make sure that no request gazumped us - if it was allocated after | |
| 833 | + * our i915_gem_request_alloc() and called __i915_add_request() before | |
| 834 | + * us, the timeline will hold its seqno which is later than ours. | |
| 835 | + */ | |
| 836 | + GEM_BUG_ON(i915_seqno_passed(timeline->last_submitted_seqno, | |
| 837 | + request->fence.seqno)); | |
| 838 | + | |
| 854 | 839 | /* |
| 855 | 840 | * To ensure that this call will not fail, space for its emissions |
| 856 | 841 | * should already have been reserved in the ring buffer. Let the ring |
| @@ -647,8 +647,9 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_i915_private *dev_priv · 646 unmodified lines | ||
| 647 | 647 | stolen_offset, gtt_offset, size); |
| 648 | 648 | |
| 649 | 649 | /* KISS and expect everything to be page-aligned */ |
| 650 | − if (WARN_ON(size == 0) || WARN_ON(size & 4095) || | |
| 651 | − WARN_ON(stolen_offset & 4095)) | |
| 650 | + if (WARN_ON(size == 0) || | |
| 651 | + WARN_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)) || | |
| 652 | + WARN_ON(!IS_ALIGNED(stolen_offset, I915_GTT_MIN_ALIGNMENT))) | |
| 652 | 653 | return NULL; |
| 653 | 654 | |
| 654 | 655 | stolen = kzalloc(sizeof(*stolen), GFP_KERNEL); |
| @@ -682,7 +683,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_i915_private *dev_priv · 27 unmodified lines | ||
| 682 | 683 | if (ret) |
| 683 | 684 | goto err; |
| 684 | 685 | |
| 685 | − vma = i915_gem_obj_lookup_or_create_vma(obj, &ggtt->base, NULL); | |
| 686 | + vma = i915_vma_instance(obj, &ggtt->base, NULL); | |
| 686 | 687 | if (IS_ERR(vma)) { |
| 687 | 688 | ret = PTR_ERR(vma); |
| 688 | 689 | goto err_pages; |
| @@ -693,15 +694,16 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_i915_private *dev_priv · 4 unmodified lines | ||
| 693 | 694 | * setting up the GTT space. The actual reservation will occur |
| 694 | 695 | * later. |
| 695 | 696 | */ |
| 696 | − vma->node.start = gtt_offset; | |
| 697 | − vma->node.size = size; | |
| 698 | − | |
| 699 | − ret = drm_mm_reserve_node(&ggtt->base.mm, &vma->node); | |
| 697 | + ret = i915_gem_gtt_reserve(&ggtt->base, &vma->node, | |
| 698 | + size, gtt_offset, obj->cache_level, | |
| 699 | + 0); | |
| 700 | 700 | if (ret) { |
| 701 | 701 | DRM_DEBUG_KMS("failed to allocate stolen GTT space\n"); |
| 702 | 702 | goto err_pages; |
| 703 | 703 | } |
| 704 | 704 | |
| 705 | + GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); | |
| 706 | + | |
| 705 | 707 | vma->pages = obj->mm.pages; |
| 706 | 708 | vma->flags |= I915_VMA_GLOBAL_BIND; |
| 707 | 709 | __i915_vma_set_map_and_fenceable(vma); |
| @@ -34,8 +34,8 @@ · 33 unmodified lines | ||
| 34 | 34 | /** |
| 35 | 35 | * DOC: buffer object tiling |
| 36 | 36 | * |
| 37 | − * i915_gem_set_tiling() and i915_gem_get_tiling() is the userspace interface to | |
| 38 | − * declare fence register requirements. | |
| 37 | + * i915_gem_set_tiling_ioctl() and i915_gem_get_tiling_ioctl() is the userspace | |
| 38 | + * interface to declare fence register requirements. | |
| 39 | 39 | * |
| 40 | 40 | * In principle GEM doesn't care at all about the internal data layout of an |
| 41 | 41 | * object, and hence it also doesn't care about tiling or swizzling. There's two |
| @@ -58,86 +58,147 @@ · 16 unmodified lines | ||
| 58 | 58 | * invovlement. |
| 59 | 59 | */ |
| 60 | 60 | |
| 61 | +/** | |
| 62 | + * i915_gem_fence_size - required global GTT size for a fence | |
| 63 | + * @i915: i915 device | |
| 64 | + * @size: object size | |
| 65 | + * @tiling: tiling mode | |
| 66 | + * @stride: tiling stride | |
| 67 | + * | |
| 68 | + * Return the required global GTT size for a fence (view of a tiled object), | |
| 69 | + * taking into account potential fence register mapping. | |
| 70 | + */ | |
| 71 | +u32 i915_gem_fence_size(struct drm_i915_private *i915, | |
| 72 | + u32 size, unsigned int tiling, unsigned int stride) | |
| 73 | +{ | |
| 74 | + u32 ggtt_size; | |
| 75 | + | |
| 76 | + GEM_BUG_ON(!size); | |
| 77 | + | |
| 78 | + if (tiling == I915_TILING_NONE) | |
| 79 | + return size; | |
| 80 | + | |
| 81 | + GEM_BUG_ON(!stride); | |
| 82 | + | |
| 83 | + if (INTEL_GEN(i915) >= 4) { | |
| 84 | + stride *= i915_gem_tile_height(tiling); | |
| 85 | + GEM_BUG_ON(!IS_ALIGNED(stride, I965_FENCE_PAGE)); | |
| 86 | + return roundup(size, stride); | |
| 87 | + } | |
| 88 | + | |
| 89 | + /* Previous chips need a power-of-two fence region when tiling */ | |
| 90 | + if (IS_GEN3(i915)) | |
| 91 | + ggtt_size = 1024*1024; | |
| 92 | + else | |
| 93 | + ggtt_size = 512*1024; | |
| 94 | + | |
| 95 | + while (ggtt_size < size) | |
| 96 | + ggtt_size <<= 1; | |
| 97 | + | |
| 98 | + return ggtt_size; | |
| 99 | +} | |
| 100 | + | |
| 101 | +/** | |
| 102 | + * i915_gem_fence_alignment - required global GTT alignment for a fence | |
| 103 | + * @i915: i915 device | |
| 104 | + * @size: object size | |
| 105 | + * @tiling: tiling mode | |
| 106 | + * @stride: tiling stride | |
| 107 | + * | |
| 108 | + * Return the required global GTT alignment for a fence (a view of a tiled | |
| 109 | + * object), taking into account potential fence register mapping. | |
| 110 | + */ | |
| 111 | +u32 i915_gem_fence_alignment(struct drm_i915_private *i915, u32 size, | |
| 112 | + unsigned int tiling, unsigned int stride) | |
| 113 | +{ | |
| 114 | + GEM_BUG_ON(!size); | |
| 115 | + | |
| 116 | + /* | |
| 117 | + * Minimum alignment is 4k (GTT page size), but might be greater | |
| 118 | + * if a fence register is needed for the object. | |
| 119 | + */ | |
| 120 | + if (tiling == I915_TILING_NONE) | |
| 121 | + return I915_GTT_MIN_ALIGNMENT; | |
| 122 | + | |
| 123 | + if (INTEL_GEN(i915) >= 4) | |
| 124 | + return I965_FENCE_PAGE; | |
| 125 | + | |
| 126 | + /* | |
| 127 | + * Previous chips need to be aligned to the size of the smallest | |
| 128 | + * fence register that can contain the object. | |
| 129 | + */ | |
| 130 | + return i915_gem_fence_size(i915, size, tiling, stride); | |
| 131 | +} | |
| 132 | + | |
| 61 | 133 | /* Check pitch constriants for all chips & tiling formats */ |
| 62 | 134 | static bool |
| 63 | −i915_tiling_ok(struct drm_i915_private *dev_priv, | |
| 64 | − int stride, int size, int tiling_mode) | |
| 135 | +i915_tiling_ok(struct drm_i915_gem_object *obj, | |
| 136 | + unsigned int tiling, unsigned int stride) | |
| 65 | 137 | { |
| 66 | − int tile_width; | |
| 138 | + struct drm_i915_private *i915 = to_i915(obj->base.dev); | |
| 139 | + unsigned int tile_width; | |
| 67 | 140 | |
| 68 | 141 | /* Linear is always fine */ |
| 69 | − if (tiling_mode == I915_TILING_NONE) | |
| 142 | + if (tiling == I915_TILING_NONE) | |
| 70 | 143 | return true; |
| 71 | 144 | |
| 72 | − if (tiling_mode > I915_TILING_LAST) | |
| 145 | + if (tiling > I915_TILING_LAST) | |
| 73 | 146 | return false; |
| 74 | 147 | |
| 75 | − if (IS_GEN2(dev_priv) || | |
| 76 | − (tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev_priv))) | |
| 77 | − tile_width = 128; | |
| 78 | − else | |
| 79 | − tile_width = 512; | |
| 80 | − | |
| 81 | 148 | /* check maximum stride & object size */ |
| 82 | 149 | /* i965+ stores the end address of the gtt mapping in the fence |
| 83 | 150 | * reg, so dont bother to check the size */ |
| 84 | − if (INTEL_GEN(dev_priv) >= 7) { | |
| 151 | + if (INTEL_GEN(i915) >= 7) { | |
| 85 | 152 | if (stride / 128 > GEN7_FENCE_MAX_PITCH_VAL) |
| 86 | 153 | return false; |
| 87 | − } else if (INTEL_GEN(dev_priv) >= 4) { | |
| 154 | + } else if (INTEL_GEN(i915) >= 4) { | |
| 88 | 155 | if (stride / 128 > I965_FENCE_MAX_PITCH_VAL) |
| 89 | 156 | return false; |
| 90 | 157 | } else { |
| 91 | 158 | if (stride > 8192) |
| 92 | 159 | return false; |
| 93 | 160 | |
| 94 | − if (IS_GEN3(dev_priv)) { | |
| 95 | − if (size > I830_FENCE_MAX_SIZE_VAL << 20) | |
| 161 | + if (IS_GEN3(i915)) { | |
| 162 | + if (obj->base.size > I830_FENCE_MAX_SIZE_VAL << 20) | |
| 96 | 163 | return false; |
| 97 | 164 | } else { |
| 98 | − if (size > I830_FENCE_MAX_SIZE_VAL << 19) | |
| 165 | + if (obj->base.size > I830_FENCE_MAX_SIZE_VAL << 19) | |
| 99 | 166 | return false; |
| 100 | 167 | } |
| 101 | 168 | } |
| 102 | 169 | |
| 103 | − if (stride < tile_width) | |
| 170 | + if (IS_GEN2(i915) || | |
| 171 | + (tiling == I915_TILING_Y && HAS_128_BYTE_Y_TILING(i915))) | |
| 172 | + tile_width = 128; | |
| 173 | + else | |
| 174 | + tile_width = 512; | |
| 175 | + | |
| 176 | + if (!IS_ALIGNED(stride, tile_width)) | |
| 104 | 177 | return false; |
| 105 | 178 | |
| 106 | 179 | /* 965+ just needs multiples of tile width */ |
| 107 | − if (INTEL_GEN(dev_priv) >= 4) { | |
| 108 | − if (stride & (tile_width - 1)) | |
| 109 | − return false; | |
| 180 | + if (INTEL_GEN(i915) >= 4) | |
| 110 | 181 | return true; |
| 111 | − } | |
| 112 | 182 | |
| 113 | 183 | /* Pre-965 needs power of two tile widths */ |
| 114 | − if (stride & (stride - 1)) | |
| 115 | − return false; | |
| 116 | − | |
| 117 | − return true; | |
| 184 | + return is_power_of_2(stride); | |
| 118 | 185 | } |
| 119 | 186 | |
| 120 | −static bool i915_vma_fence_prepare(struct i915_vma *vma, int tiling_mode) | |
| 187 | +static bool i915_vma_fence_prepare(struct i915_vma *vma, | |
| 188 | + int tiling_mode, unsigned int stride) | |
| 121 | 189 | { |
| 122 | − struct drm_i915_private *dev_priv = vma->vm->i915; | |
| 123 | − u32 size; | |
| 190 | + struct drm_i915_private *i915 = vma->vm->i915; | |
| 191 | + u32 size, alignment; | |
| 124 | 192 | |
| 125 | 193 | if (!i915_vma_is_map_and_fenceable(vma)) |
| 126 | 194 | return true; |
| 127 | 195 | |
| 128 | − if (INTEL_GEN(dev_priv) == 3) { | |
| 129 | − if (vma->node.start & ~I915_FENCE_START_MASK) | |
| 130 | − return false; | |
| 131 | − } else { | |
| 132 | − if (vma->node.start & ~I830_FENCE_START_MASK) | |
| 133 | − return false; | |
| 134 | − } | |
| 135 | − | |
| 136 | − size = i915_gem_get_ggtt_size(dev_priv, vma->size, tiling_mode); | |
| 196 | + size = i915_gem_fence_size(i915, vma->size, tiling_mode, stride); | |
| 137 | 197 | if (vma->node.size < size) |
| 138 | 198 | return false; |
| 139 | 199 | |
| 140 | − if (vma->node.start & (size - 1)) | |
| 200 | + alignment = i915_gem_fence_alignment(i915, vma->size, tiling_mode, stride); | |
| 201 | + if (!IS_ALIGNED(vma->node.start, alignment)) | |
| 141 | 202 | return false; |
| 142 | 203 | |
| 143 | 204 | return true; |
| @@ -145,20 +206,20 @@ static bool i915_vma_fence_prepare(struct i915_vma *vma, int tiling_mode) · 1 unmodified line | ||
| 145 | 206 | |
| 146 | 207 | /* Make the current GTT allocation valid for the change in tiling. */ |
| 147 | 208 | static int |
| 148 | −i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj, int tiling_mode) | |
| 209 | +i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj, | |
| 210 | + int tiling_mode, unsigned int stride) | |
| 149 | 211 | { |
| 150 | − struct drm_i915_private *dev_priv = to_i915(obj->base.dev); | |
| 151 | 212 | struct i915_vma *vma; |
| 152 | 213 | int ret; |
| 153 | 214 | |
| 154 | 215 | if (tiling_mode == I915_TILING_NONE) |
| 155 | 216 | return 0; |
| 156 | 217 | |
| 157 | − if (INTEL_GEN(dev_priv) >= 4) | |
| 158 | − return 0; | |
| 159 | − | |
| 160 | 218 | list_for_each_entry(vma, &obj->vma_list, obj_link) { |
| 161 | − if (i915_vma_fence_prepare(vma, tiling_mode)) | |
| 219 | + if (!i915_vma_is_ggtt(vma)) | |
| 220 | + break; | |
| 221 | + | |
| 222 | + if (i915_vma_fence_prepare(vma, tiling_mode, stride)) | |
| 162 | 223 | continue; |
| 163 | 224 | |
| 164 | 225 | ret = i915_vma_unbind(vma); |
| @@ -169,8 +230,100 @@ i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj, int tiling_mode) · 4 unmodified lines | ||
| 169 | 230 | return 0; |
| 170 | 231 | } |
| 171 | 232 | |
| 233 | +int | |
| 234 | +i915_gem_object_set_tiling(struct drm_i915_gem_object *obj, | |
| 235 | + unsigned int tiling, unsigned int stride) | |
| 236 | +{ | |
| 237 | + struct drm_i915_private *i915 = to_i915(obj->base.dev); | |
| 238 | + struct i915_vma *vma; | |
| 239 | + int err; | |
| 240 | + | |
| 241 | + /* Make sure we don't cross-contaminate obj->tiling_and_stride */ | |
| 242 | + BUILD_BUG_ON(I915_TILING_LAST & STRIDE_MASK); | |
| 243 | + | |
| 244 | + GEM_BUG_ON(!i915_tiling_ok(obj, tiling, stride)); | |
| 245 | + GEM_BUG_ON(!stride ^ (tiling == I915_TILING_NONE)); | |
| 246 | + lockdep_assert_held(&i915->drm.struct_mutex); | |
| 247 | + | |
| 248 | + if ((tiling | stride) == obj->tiling_and_stride) | |
| 249 | + return 0; | |
| 250 | + | |
| 251 | + if (obj->framebuffer_references) | |
| 252 | + return -EBUSY; | |
| 253 | + | |
| 254 | + /* We need to rebind the object if its current allocation | |
| 255 | + * no longer meets the alignment restrictions for its new | |
| 256 | + * tiling mode. Otherwise we can just leave it alone, but | |
| 257 | + * need to ensure that any fence register is updated before | |
| 258 | + * the next fenced (either through the GTT or by the BLT unit | |
| 259 | + * on older GPUs) access. | |
| 260 | + * | |
| 261 | + * After updating the tiling parameters, we then flag whether | |
| 262 | + * we need to update an associated fence register. Note this | |
| 263 | + * has to also include the unfenced register the GPU uses | |
| 264 | + * whilst executing a fenced command for an untiled object. | |
| 265 | + */ | |
| 266 | + | |
| 267 | + err = i915_gem_object_fence_prepare(obj, tiling, stride); | |
| 268 | + if (err) | |
| 269 | + return err; | |
| 270 | + | |
| 271 | + /* If the memory has unknown (i.e. varying) swizzling, we pin the | |
| 272 | + * pages to prevent them being swapped out and causing corruption | |
| 273 | + * due to the change in swizzling. | |
| 274 | + */ | |
| 275 | + mutex_lock(&obj->mm.lock); | |
| 276 | + if (obj->mm.pages && | |
| 277 | + obj->mm.madv == I915_MADV_WILLNEED && | |
| 278 | + i915->quirks & QUIRK_PIN_SWIZZLED_PAGES) { | |
| 279 | + if (tiling == I915_TILING_NONE) { | |
| 280 | + GEM_BUG_ON(!obj->mm.quirked); | |
| 281 | + __i915_gem_object_unpin_pages(obj); | |
| 282 | + obj->mm.quirked = false; | |
| 283 | + } | |
| 284 | + if (!i915_gem_object_is_tiled(obj)) { | |
| 285 | + GEM_BUG_ON(!obj->mm.quirked); | |
| 286 | + __i915_gem_object_pin_pages(obj); | |
| 287 | + obj->mm.quirked = true; | |
| 288 | + } | |
| 289 | + } | |
| 290 | + mutex_unlock(&obj->mm.lock); | |
| 291 | + | |
| 292 | + list_for_each_entry(vma, &obj->vma_list, obj_link) { | |
| 293 | + if (!i915_vma_is_ggtt(vma)) | |
| 294 | + break; | |
| 295 | + | |
| 296 | + vma->fence_size = | |
| 297 | + i915_gem_fence_size(i915, vma->size, tiling, stride); | |
| 298 | + vma->fence_alignment = | |
| 299 | + i915_gem_fence_alignment(i915, | |
| 300 | + vma->size, tiling, stride); | |
| 301 | + | |
| 302 | + if (vma->fence) | |
| 303 | + vma->fence->dirty = true; | |
| 304 | + } | |
| 305 | + | |
| 306 | + obj->tiling_and_stride = tiling | stride; | |
| 307 | + | |
| 308 | + /* Force the fence to be reacquired for GTT access */ | |
| 309 | + i915_gem_release_mmap(obj); | |
| 310 | + | |
| 311 | + /* Try to preallocate memory required to save swizzling on put-pages */ | |
| 312 | + if (i915_gem_object_needs_bit17_swizzle(obj)) { | |
| 313 | + if (!obj->bit_17) { | |
| 314 | + obj->bit_17 = kcalloc(BITS_TO_LONGS(obj->base.size >> PAGE_SHIFT), | |
| 315 | + sizeof(long), GFP_KERNEL); | |
| 316 | + } | |
| 317 | + } else { | |
| 318 | + kfree(obj->bit_17); | |
| 319 | + obj->bit_17 = NULL; | |
| 320 | + } | |
| 321 | + | |
| 322 | + return 0; | |
| 323 | +} | |
| 324 | + | |
| 172 | 325 | /** |
| 173 | − * i915_gem_set_tiling - IOCTL handler to set tiling mode | |
| 326 | + * i915_gem_set_tiling_ioctl - IOCTL handler to set tiling mode | |
| 174 | 327 | * @dev: DRM device |
| 175 | 328 | * @data: data pointer for the ioctl |
| 176 | 329 | * @file: DRM file for the ioctl call |
| @@ -184,30 +337,19 @@ i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj, int tiling_mode) · 7 unmodified lines | ||
| 184 | 337 | * Zero on success, negative errno on failure. |
| 185 | 338 | */ |
| 186 | 339 | int |
| 187 | −i915_gem_set_tiling(struct drm_device *dev, void *data, | |
| 188 | − struct drm_file *file) | |
| 340 | +i915_gem_set_tiling_ioctl(struct drm_device *dev, void *data, | |
| 341 | + struct drm_file *file) | |
| 189 | 342 | { |
| 190 | 343 | struct drm_i915_gem_set_tiling *args = data; |
| 191 | − struct drm_i915_private *dev_priv = to_i915(dev); | |
| 192 | 344 | struct drm_i915_gem_object *obj; |
| 193 | − int err = 0; | |
| 194 | − | |
| 195 | − /* Make sure we don't cross-contaminate obj->tiling_and_stride */ | |
| 196 | − BUILD_BUG_ON(I915_TILING_LAST & STRIDE_MASK); | |
| 345 | + int err; | |
| 197 | 346 | |
| 198 | 347 | obj = i915_gem_object_lookup(file, args->handle); |
| 199 | 348 | if (!obj) |
| 200 | 349 | return -ENOENT; |
| 201 | 350 | |
| 202 | − if (!i915_tiling_ok(dev_priv, | |
| 203 | − args->stride, obj->base.size, args->tiling_mode)) { | |
| 204 | − i915_gem_object_put(obj); | |
| 205 | − return -EINVAL; | |
| 206 | − } | |
| 207 | − | |
| 208 | − mutex_lock(&dev->struct_mutex); | |
| 209 | − if (obj->pin_display || obj->framebuffer_references) { | |
| 210 | − err = -EBUSY; | |
| 351 | + if (!i915_tiling_ok(obj, args->tiling_mode, args->stride)) { | |
| 352 | + err = -EINVAL; | |
| 211 | 353 | goto err; |
| 212 | 354 | } |
| 213 | 355 | |
| @@ -216,9 +358,9 @@ i915_gem_set_tiling(struct drm_device *dev, void *data, · 2 unmodified lines | ||
| 216 | 358 | args->stride = 0; |
| 217 | 359 | } else { |
| 218 | 360 | if (args->tiling_mode == I915_TILING_X) |
| 219 | − args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x; | |
| 361 | + args->swizzle_mode = to_i915(dev)->mm.bit_6_swizzle_x; | |
| 220 | 362 | else |
| 221 | − args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y; | |
| 363 | + args->swizzle_mode = to_i915(dev)->mm.bit_6_swizzle_y; | |
| 222 | 364 | |
| 223 | 365 | /* Hide bit 17 swizzling from the user. This prevents old Mesa |
| 224 | 366 | * from aborting the application on sw fallbacks to bit 17, |
| @@ -240,79 +382,24 @@ i915_gem_set_tiling(struct drm_device *dev, void *data, · 15 unmodified lines | ||
| 240 | 382 | } |
| 241 | 383 | } |
| 242 | 384 | |
| 243 | − if (args->tiling_mode != i915_gem_object_get_tiling(obj) || | |
| 244 | − args->stride != i915_gem_object_get_stride(obj)) { | |
| 245 | − /* We need to rebind the object if its current allocation | |
| 246 | − * no longer meets the alignment restrictions for its new | |
| 247 | − * tiling mode. Otherwise we can just leave it alone, but | |
| 248 | − * need to ensure that any fence register is updated before | |
| 249 | − * the next fenced (either through the GTT or by the BLT unit | |
| 250 | − * on older GPUs) access. | |
| 251 | − * | |
| 252 | − * After updating the tiling parameters, we then flag whether | |
| 253 | − * we need to update an associated fence register. Note this | |
| 254 | − * has to also include the unfenced register the GPU uses | |
| 255 | − * whilst executing a fenced command for an untiled object. | |
| 256 | − */ | |
| 385 | + err = mutex_lock_interruptible(&dev->struct_mutex); | |
| 386 | + if (err) | |
| 387 | + goto err; | |
| 257 | 388 | |
| 258 | − err = i915_gem_object_fence_prepare(obj, args->tiling_mode); | |
| 259 | − if (!err) { | |
| 260 | − struct i915_vma *vma; | |
| 261 | − | |
| 262 | − mutex_lock(&obj->mm.lock); | |
| 263 | − if (obj->mm.pages && | |
| 264 | − obj->mm.madv == I915_MADV_WILLNEED && | |
| 265 | − dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES) { | |
| 266 | − if (args->tiling_mode == I915_TILING_NONE) { | |
| 267 | − GEM_BUG_ON(!obj->mm.quirked); | |
| 268 | − __i915_gem_object_unpin_pages(obj); | |
| 269 | − obj->mm.quirked = false; | |
| 270 | − } | |
| 271 | − if (!i915_gem_object_is_tiled(obj)) { | |
| 272 | − GEM_BUG_ON(!obj->mm.quirked); | |
| 273 | − __i915_gem_object_pin_pages(obj); | |
| 274 | − obj->mm.quirked = true; | |
| 275 | − } | |
| 276 | − } | |
| 277 | − mutex_unlock(&obj->mm.lock); | |
| 278 | − | |
| 279 | − list_for_each_entry(vma, &obj->vma_list, obj_link) { | |
| 280 | − if (!vma->fence) | |
| 281 | − continue; | |
| 282 | − | |
| 283 | − vma->fence->dirty = true; | |
| 284 | − } | |
| 285 | − obj->tiling_and_stride = | |
| 286 | − args->stride | args->tiling_mode; | |
| 287 | − | |
| 288 | − /* Force the fence to be reacquired for GTT access */ | |
| 289 | − i915_gem_release_mmap(obj); | |
| 290 | − } | |
| 291 | − } | |
| 292 | − /* we have to maintain this existing ABI... */ | |
| 389 | + err = i915_gem_object_set_tiling(obj, args->tiling_mode, args->stride); | |
| 390 | + mutex_unlock(&dev->struct_mutex); | |
| 391 | + | |
| 392 | + /* We have to maintain this existing ABI... */ | |
| 293 | 393 | args->stride = i915_gem_object_get_stride(obj); |
| 294 | 394 | args->tiling_mode = i915_gem_object_get_tiling(obj); |
| 295 | 395 | |
| 296 | − /* Try to preallocate memory required to save swizzling on put-pages */ | |
| 297 | − if (i915_gem_object_needs_bit17_swizzle(obj)) { | |
| 298 | − if (obj->bit_17 == NULL) { | |
| 299 | − obj->bit_17 = kcalloc(BITS_TO_LONGS(obj->base.size >> PAGE_SHIFT), | |
| 300 | − sizeof(long), GFP_KERNEL); | |
| 301 | − } | |
| 302 | − } else { | |
| 303 | − kfree(obj->bit_17); | |
| 304 | − obj->bit_17 = NULL; | |
| 305 | − } | |
| 306 | − | |
| 307 | 396 | err: |
| 308 | 397 | i915_gem_object_put(obj); |
| 309 | − mutex_unlock(&dev->struct_mutex); | |
| 310 | − | |
| 311 | 398 | return err; |
| 312 | 399 | } |
| 313 | 400 | |
| 314 | 401 | /** |
| 315 | − * i915_gem_get_tiling - IOCTL handler to get tiling mode | |
| 402 | + * i915_gem_get_tiling_ioctl - IOCTL handler to get tiling mode | |
| 316 | 403 | * @dev: DRM device |
| 317 | 404 | * @data: data pointer for the ioctl |
| 318 | 405 | * @file: DRM file for the ioctl call |
| @@ -325,8 +412,8 @@ err: · 6 unmodified lines | ||
| 325 | 412 | * Zero on success, negative errno on failure. |
| 326 | 413 | */ |
| 327 | 414 | int |
| 328 | −i915_gem_get_tiling(struct drm_device *dev, void *data, | |
| 329 | − struct drm_file *file) | |
| 415 | +i915_gem_get_tiling_ioctl(struct drm_device *dev, void *data, | |
| 416 | + struct drm_file *file) | |
| 330 | 417 | { |
| 331 | 418 | struct drm_i915_gem_get_tiling *args = data; |
| 332 | 419 | struct drm_i915_private *dev_priv = to_i915(dev); |
| @@ -121,6 +121,7 @@ static void __i915_error_advance(struct drm_i915_error_state_buf *e, · 120 unmodified lines | ||
| 121 | 121 | e->pos += len; |
| 122 | 122 | } |
| 123 | 123 | |
| 124 | +__printf(2, 0) | |
| 124 | 125 | static void i915_error_vprintf(struct drm_i915_error_state_buf *e, |
| 125 | 126 | const char *f, va_list args) |
| 126 | 127 | { |
| @@ -61,12 +61,18 @@ · 60 unmodified lines | ||
| 61 | 61 | #define DMA_ADDRESS_SPACE_GTT (8 << 16) |
| 62 | 62 | #define DMA_COPY_SIZE _MMIO(0xc310) |
| 63 | 63 | #define DMA_CTRL _MMIO(0xc314) |
| 64 | +#define HUC_UKERNEL (1<<9) | |
| 64 | 65 | #define UOS_MOVE (1<<4) |
| 65 | 66 | #define START_DMA (1<<0) |
| 66 | 67 | #define DMA_GUC_WOPCM_OFFSET _MMIO(0xc340) |
| 68 | +#define HUC_LOADING_AGENT_VCR (0<<1) | |
| 69 | +#define HUC_LOADING_AGENT_GUC (1<<1) | |
| 67 | 70 | #define GUC_WOPCM_OFFSET_VALUE 0x80000 /* 512KB */ |
| 68 | 71 | #define GUC_MAX_IDLE_COUNT _MMIO(0xC3E4) |
| 69 | 72 | |
| 73 | +#define HUC_STATUS2 _MMIO(0xD3B0) | |
| 74 | +#define HUC_FW_VERIFIED (1<<7) | |
| 75 | + | |
| 70 | 76 | /* Defines WOPCM space available to GuC firmware */ |
| 71 | 77 | #define GUC_WOPCM_SIZE _MMIO(0xc050) |
| 72 | 78 | /* GuC addresses below GUC_WOPCM_TOP don't map through the GTT */ |
| @@ -22,8 +22,6 @@ · 21 unmodified lines | ||
| 22 | 22 | * |
| 23 | 23 | */ |
| 24 | 24 | #include <linux/circ_buf.h> |
| 25 | −#include <linux/debugfs.h> | |
| 26 | −#include <linux/relay.h> | |
| 27 | 25 | #include "i915_drv.h" |
| 28 | 26 | #include "intel_uc.h" |
| 29 | 27 | |
| @@ -545,7 +543,7 @@ static void i915_guc_submit(struct drm_i915_gem_request *rq) · 515 unmodified lines | ||
| 545 | 543 | */ |
| 546 | 544 | |
| 547 | 545 | /** |
| 548 | − * guc_allocate_vma() - Allocate a GGTT VMA for GuC usage | |
| 546 | + * intel_guc_allocate_vma() - Allocate a GGTT VMA for GuC usage | |
| 549 | 547 | * @guc: the guc |
| 550 | 548 | * @size: size of area to allocate (both virtual space and memory) |
| 551 | 549 | * |
| @@ -557,7 +555,7 @@ static void i915_guc_submit(struct drm_i915_gem_request *rq) · 5 unmodified lines | ||
| 557 | 555 | * |
| 558 | 556 | * Return: A i915_vma if successful, otherwise an ERR_PTR. |
| 559 | 557 | */ |
| 560 | −static struct i915_vma *guc_allocate_vma(struct intel_guc *guc, u32 size) | |
| 558 | +struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size) | |
| 561 | 559 | { |
| 562 | 560 | struct drm_i915_private *dev_priv = guc_to_i915(guc); |
| 563 | 561 | struct drm_i915_gem_object *obj; |
| @@ -568,7 +566,7 @@ static struct i915_vma *guc_allocate_vma(struct intel_guc *guc, u32 size) · 4 unmodified lines | ||
| 568 | 566 | if (IS_ERR(obj)) |
| 569 | 567 | return ERR_CAST(obj); |
| 570 | 568 | |
| 571 | − vma = i915_vma_create(obj, &dev_priv->ggtt.base, NULL); | |
| 569 | + vma = i915_vma_instance(obj, &dev_priv->ggtt.base, NULL); | |
| 572 | 570 | if (IS_ERR(vma)) |
| 573 | 571 | goto err; |
| 574 | 572 | |
| @@ -579,9 +577,6 @@ static struct i915_vma *guc_allocate_vma(struct intel_guc *guc, u32 size) · 4 unmodified lines | ||
| 579 | 577 | goto err; |
| 580 | 578 | } |
| 581 | 579 | |
| 582 | − /* Invalidate GuC TLB to let GuC take the latest updates to GTT. */ | |
| 583 | − I915_WRITE(GEN8_GTCR, GEN8_GTCR_INVALIDATE); | |
| 584 | − | |
| 585 | 580 | return vma; |
| 586 | 581 | |
| 587 | 582 | err: |
| @@ -721,7 +716,7 @@ guc_client_alloc(struct drm_i915_private *dev_priv, · 133 unmodified lines | ||
| 721 | 716 | } |
| 722 | 717 | |
| 723 | 718 | /* The first page is doorbell/proc_desc. Two followed pages are wq. */ |
| 724 | − vma = guc_allocate_vma(guc, GUC_DB_SIZE + GUC_WQ_SIZE); | |
| 719 | + vma = intel_guc_allocate_vma(guc, GUC_DB_SIZE + GUC_WQ_SIZE); | |
| 725 | 720 | if (IS_ERR(vma)) |
| 726 | 721 | goto err; |
| 727 | 722 | |
| @@ -777,488 +772,7 @@ err: · 49 unmodified lines | ||
| 777 | 772 | return NULL; |
| 778 | 773 | } |
| 779 | 774 | |
| 780 | −/* | |
| 781 | − * Sub buffer switch callback. Called whenever relay has to switch to a new | |
| 782 | − * sub buffer, relay stays on the same sub buffer if 0 is returned. | |
| 783 | − */ | |
| 784 | −static int subbuf_start_callback(struct rchan_buf *buf, | |
| 785 | − void *subbuf, | |
| 786 | − void *prev_subbuf, | |
| 787 | − size_t prev_padding) | |
| 788 | −{ | |
| 789 | − /* Use no-overwrite mode by default, where relay will stop accepting | |
| 790 | − * new data if there are no empty sub buffers left. | |
| 791 | − * There is no strict synchronization enforced by relay between Consumer | |
| 792 | − * and Producer. In overwrite mode, there is a possibility of getting | |
| 793 | − * inconsistent/garbled data, the producer could be writing on to the | |
| 794 | − * same sub buffer from which Consumer is reading. This can't be avoided | |
| 795 | − * unless Consumer is fast enough and can always run in tandem with | |
| 796 | − * Producer. | |
| 797 | − */ | |
| 798 | − if (relay_buf_full(buf)) | |
| 799 | − return 0; | |
| 800 | − | |
| 801 | − return 1; | |
| 802 | −} | |
| 803 | − | |
| 804 | −/* | |
| 805 | − * file_create() callback. Creates relay file in debugfs. | |
| 806 | − */ | |
| 807 | −static struct dentry *create_buf_file_callback(const char *filename, | |
| 808 | − struct dentry *parent, | |
| 809 | − umode_t mode, | |
| 810 | − struct rchan_buf *buf, | |
| 811 | − int *is_global) | |
| 812 | −{ | |
| 813 | − struct dentry *buf_file; | |
| 814 | − | |
| 815 | − /* This to enable the use of a single buffer for the relay channel and | |
| 816 | − * correspondingly have a single file exposed to User, through which | |
| 817 | − * it can collect the logs in order without any post-processing. | |
| 818 | − * Need to set 'is_global' even if parent is NULL for early logging. | |
| 819 | − */ | |
| 820 | − *is_global = 1; | |
| 821 | − | |
| 822 | − if (!parent) | |
| 823 | − return NULL; | |
| 824 | − | |
| 825 | − /* Not using the channel filename passed as an argument, since for each | |
| 826 | − * channel relay appends the corresponding CPU number to the filename | |
| 827 | − * passed in relay_open(). This should be fine as relay just needs a | |
| 828 | − * dentry of the file associated with the channel buffer and that file's | |
| 829 | − * name need not be same as the filename passed as an argument. | |
| 830 | − */ | |
| 831 | − buf_file = debugfs_create_file("guc_log", mode, | |
| 832 | − parent, buf, &relay_file_operations); | |
| 833 | − return buf_file; | |
| 834 | −} | |
| 835 | − | |
| 836 | −/* | |
| 837 | − * file_remove() default callback. Removes relay file in debugfs. | |
| 838 | − */ | |
| 839 | −static int remove_buf_file_callback(struct dentry *dentry) | |
| 840 | −{ | |
| 841 | − debugfs_remove(dentry); | |
| 842 | − return 0; | |
| 843 | −} | |
| 844 | − | |
| 845 | −/* relay channel callbacks */ | |
| 846 | −static struct rchan_callbacks relay_callbacks = { | |
| 847 | − .subbuf_start = subbuf_start_callback, | |
| 848 | − .create_buf_file = create_buf_file_callback, | |
| 849 | − .remove_buf_file = remove_buf_file_callback, | |
| 850 | −}; | |
| 851 | − | |
| 852 | −static void guc_log_remove_relay_file(struct intel_guc *guc) | |
| 853 | −{ | |
| 854 | − relay_close(guc->log.relay_chan); | |
| 855 | −} | |
| 856 | − | |
| 857 | −static int guc_log_create_relay_channel(struct intel_guc *guc) | |
| 858 | −{ | |
| 859 | − struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 860 | − struct rchan *guc_log_relay_chan; | |
| 861 | − size_t n_subbufs, subbuf_size; | |
| 862 | − | |
| 863 | − /* Keep the size of sub buffers same as shared log buffer */ | |
| 864 | − subbuf_size = guc->log.vma->obj->base.size; | |
| 865 | − | |
| 866 | − /* Store up to 8 snapshots, which is large enough to buffer sufficient | |
| 867 | − * boot time logs and provides enough leeway to User, in terms of | |
| 868 | − * latency, for consuming the logs from relay. Also doesn't take | |
| 869 | − * up too much memory. | |
| 870 | − */ | |
| 871 | − n_subbufs = 8; | |
| 872 | − | |
| 873 | − guc_log_relay_chan = relay_open(NULL, NULL, subbuf_size, | |
| 874 | − n_subbufs, &relay_callbacks, dev_priv); | |
| 875 | − if (!guc_log_relay_chan) { | |
| 876 | − DRM_ERROR("Couldn't create relay chan for GuC logging\n"); | |
| 877 | − return -ENOMEM; | |
| 878 | − } | |
| 879 | − | |
| 880 | − GEM_BUG_ON(guc_log_relay_chan->subbuf_size < subbuf_size); | |
| 881 | − guc->log.relay_chan = guc_log_relay_chan; | |
| 882 | − return 0; | |
| 883 | −} | |
| 884 | − | |
| 885 | −static int guc_log_create_relay_file(struct intel_guc *guc) | |
| 886 | −{ | |
| 887 | − struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 888 | − struct dentry *log_dir; | |
| 889 | − int ret; | |
| 890 | − | |
| 891 | − /* For now create the log file in /sys/kernel/debug/dri/0 dir */ | |
| 892 | − log_dir = dev_priv->drm.primary->debugfs_root; | |
| 893 | − | |
| 894 | − /* If /sys/kernel/debug/dri/0 location do not exist, then debugfs is | |
| 895 | − * not mounted and so can't create the relay file. | |
| 896 | − * The relay API seems to fit well with debugfs only, for availing relay | |
| 897 | − * there are 3 requirements which can be met for debugfs file only in a | |
| 898 | − * straightforward/clean manner :- | |
| 899 | − * i) Need the associated dentry pointer of the file, while opening the | |
| 900 | − * relay channel. | |
| 901 | − * ii) Should be able to use 'relay_file_operations' fops for the file. | |
| 902 | − * iii) Set the 'i_private' field of file's inode to the pointer of | |
| 903 | − * relay channel buffer. | |
| 904 | − */ | |
| 905 | − if (!log_dir) { | |
| 906 | − DRM_ERROR("Debugfs dir not available yet for GuC log file\n"); | |
| 907 | − return -ENODEV; | |
| 908 | − } | |
| 909 | − | |
| 910 | − ret = relay_late_setup_files(guc->log.relay_chan, "guc_log", log_dir); | |
| 911 | − if (ret) { | |
| 912 | − DRM_ERROR("Couldn't associate relay chan with file %d\n", ret); | |
| 913 | − return ret; | |
| 914 | − } | |
| 915 | − | |
| 916 | − return 0; | |
| 917 | −} | |
| 918 | − | |
| 919 | −static void guc_move_to_next_buf(struct intel_guc *guc) | |
| 920 | −{ | |
| 921 | − /* Make sure the updates made in the sub buffer are visible when | |
| 922 | − * Consumer sees the following update to offset inside the sub buffer. | |
| 923 | − */ | |
| 924 | − smp_wmb(); | |
| 925 | − | |
| 926 | − /* All data has been written, so now move the offset of sub buffer. */ | |
| 927 | − relay_reserve(guc->log.relay_chan, guc->log.vma->obj->base.size); | |
| 928 | − | |
| 929 | − /* Switch to the next sub buffer */ | |
| 930 | − relay_flush(guc->log.relay_chan); | |
| 931 | −} | |
| 932 | − | |
| 933 | −static void *guc_get_write_buffer(struct intel_guc *guc) | |
| 934 | −{ | |
| 935 | − if (!guc->log.relay_chan) | |
| 936 | − return NULL; | |
| 937 | − | |
| 938 | − /* Just get the base address of a new sub buffer and copy data into it | |
| 939 | − * ourselves. NULL will be returned in no-overwrite mode, if all sub | |
| 940 | − * buffers are full. Could have used the relay_write() to indirectly | |
| 941 | − * copy the data, but that would have been bit convoluted, as we need to | |
| 942 | − * write to only certain locations inside a sub buffer which cannot be | |
| 943 | − * done without using relay_reserve() along with relay_write(). So its | |
| 944 | − * better to use relay_reserve() alone. | |
| 945 | − */ | |
| 946 | − return relay_reserve(guc->log.relay_chan, 0); | |
| 947 | −} | |
| 948 | − | |
| 949 | −static bool | |
| 950 | −guc_check_log_buf_overflow(struct intel_guc *guc, | |
| 951 | − enum guc_log_buffer_type type, unsigned int full_cnt) | |
| 952 | −{ | |
| 953 | − unsigned int prev_full_cnt = guc->log.prev_overflow_count[type]; | |
| 954 | − bool overflow = false; | |
| 955 | − | |
| 956 | − if (full_cnt != prev_full_cnt) { | |
| 957 | − overflow = true; | |
| 958 | − | |
| 959 | − guc->log.prev_overflow_count[type] = full_cnt; | |
| 960 | − guc->log.total_overflow_count[type] += full_cnt - prev_full_cnt; | |
| 961 | − | |
| 962 | − if (full_cnt < prev_full_cnt) { | |
| 963 | − /* buffer_full_cnt is a 4 bit counter */ | |
| 964 | − guc->log.total_overflow_count[type] += 16; | |
| 965 | − } | |
| 966 | − DRM_ERROR_RATELIMITED("GuC log buffer overflow\n"); | |
| 967 | − } | |
| 968 | − | |
| 969 | − return overflow; | |
| 970 | −} | |
| 971 | − | |
| 972 | −static unsigned int guc_get_log_buffer_size(enum guc_log_buffer_type type) | |
| 973 | −{ | |
| 974 | − switch (type) { | |
| 975 | − case GUC_ISR_LOG_BUFFER: | |
| 976 | − return (GUC_LOG_ISR_PAGES + 1) * PAGE_SIZE; | |
| 977 | − case GUC_DPC_LOG_BUFFER: | |
| 978 | − return (GUC_LOG_DPC_PAGES + 1) * PAGE_SIZE; | |
| 979 | − case GUC_CRASH_DUMP_LOG_BUFFER: | |
| 980 | − return (GUC_LOG_CRASH_PAGES + 1) * PAGE_SIZE; | |
| 981 | − default: | |
| 982 | − MISSING_CASE(type); | |
| 983 | − } | |
| 984 | − | |
| 985 | − return 0; | |
| 986 | −} | |
| 987 | − | |
| 988 | −static void guc_read_update_log_buffer(struct intel_guc *guc) | |
| 989 | −{ | |
| 990 | − unsigned int buffer_size, read_offset, write_offset, bytes_to_copy, full_cnt; | |
| 991 | − struct guc_log_buffer_state *log_buf_state, *log_buf_snapshot_state; | |
| 992 | − struct guc_log_buffer_state log_buf_state_local; | |
| 993 | − enum guc_log_buffer_type type; | |
| 994 | − void *src_data, *dst_data; | |
| 995 | − bool new_overflow; | |
| 996 | − | |
| 997 | − if (WARN_ON(!guc->log.buf_addr)) | |
| 998 | − return; | |
| 999 | − | |
| 1000 | − /* Get the pointer to shared GuC log buffer */ | |
| 1001 | − log_buf_state = src_data = guc->log.buf_addr; | |
| 1002 | − | |
| 1003 | − /* Get the pointer to local buffer to store the logs */ | |
| 1004 | − log_buf_snapshot_state = dst_data = guc_get_write_buffer(guc); | |
| 1005 | − | |
| 1006 | − /* Actual logs are present from the 2nd page */ | |
| 1007 | − src_data += PAGE_SIZE; | |
| 1008 | − dst_data += PAGE_SIZE; | |
| 1009 | − | |
| 1010 | − for (type = GUC_ISR_LOG_BUFFER; type < GUC_MAX_LOG_BUFFER; type++) { | |
| 1011 | − /* Make a copy of the state structure, inside GuC log buffer | |
| 1012 | − * (which is uncached mapped), on the stack to avoid reading | |
| 1013 | − * from it multiple times. | |
| 1014 | − */ | |
| 1015 | − memcpy(&log_buf_state_local, log_buf_state, | |
| 1016 | − sizeof(struct guc_log_buffer_state)); | |
| 1017 | − buffer_size = guc_get_log_buffer_size(type); | |
| 1018 | − read_offset = log_buf_state_local.read_ptr; | |
| 1019 | − write_offset = log_buf_state_local.sampled_write_ptr; | |
| 1020 | − full_cnt = log_buf_state_local.buffer_full_cnt; | |
| 1021 | − | |
| 1022 | − /* Bookkeeping stuff */ | |
| 1023 | − guc->log.flush_count[type] += log_buf_state_local.flush_to_file; | |
| 1024 | − new_overflow = guc_check_log_buf_overflow(guc, type, full_cnt); | |
| 1025 | − | |
| 1026 | − /* Update the state of shared log buffer */ | |
| 1027 | − log_buf_state->read_ptr = write_offset; | |
| 1028 | − log_buf_state->flush_to_file = 0; | |
| 1029 | − log_buf_state++; | |
| 1030 | − | |
| 1031 | − if (unlikely(!log_buf_snapshot_state)) | |
| 1032 | − continue; | |
| 1033 | − | |
| 1034 | − /* First copy the state structure in snapshot buffer */ | |
| 1035 | − memcpy(log_buf_snapshot_state, &log_buf_state_local, | |
| 1036 | − sizeof(struct guc_log_buffer_state)); | |
| 1037 | − | |
| 1038 | − /* The write pointer could have been updated by GuC firmware, | |
| 1039 | − * after sending the flush interrupt to Host, for consistency | |
| 1040 | − * set write pointer value to same value of sampled_write_ptr | |
| 1041 | − * in the snapshot buffer. | |
| 1042 | − */ | |
| 1043 | − log_buf_snapshot_state->write_ptr = write_offset; | |
| 1044 | − log_buf_snapshot_state++; | |
| 1045 | − | |
| 1046 | − /* Now copy the actual logs. */ | |
| 1047 | − if (unlikely(new_overflow)) { | |
| 1048 | − /* copy the whole buffer in case of overflow */ | |
| 1049 | − read_offset = 0; | |
| 1050 | − write_offset = buffer_size; | |
| 1051 | − } else if (unlikely((read_offset > buffer_size) || | |
| 1052 | − (write_offset > buffer_size))) { | |
| 1053 | − DRM_ERROR("invalid log buffer state\n"); | |
| 1054 | − /* copy whole buffer as offsets are unreliable */ | |
| 1055 | − read_offset = 0; | |
| 1056 | − write_offset = buffer_size; | |
| 1057 | − } | |
| 1058 | − | |
| 1059 | − /* Just copy the newly written data */ | |
| 1060 | − if (read_offset > write_offset) { | |
| 1061 | − i915_memcpy_from_wc(dst_data, src_data, write_offset); | |
| 1062 | − bytes_to_copy = buffer_size - read_offset; | |
| 1063 | − } else { | |
| 1064 | − bytes_to_copy = write_offset - read_offset; | |
| 1065 | − } | |
| 1066 | − i915_memcpy_from_wc(dst_data + read_offset, | |
| 1067 | − src_data + read_offset, bytes_to_copy); | |
| 1068 | − | |
| 1069 | − src_data += buffer_size; | |
| 1070 | − dst_data += buffer_size; | |
| 1071 | − } | |
| 1072 | − | |
| 1073 | − if (log_buf_snapshot_state) | |
| 1074 | − guc_move_to_next_buf(guc); | |
| 1075 | − else { | |
| 1076 | − /* Used rate limited to avoid deluge of messages, logs might be | |
| 1077 | − * getting consumed by User at a slow rate. | |
| 1078 | − */ | |
| 1079 | − DRM_ERROR_RATELIMITED("no sub-buffer to capture logs\n"); | |
| 1080 | − guc->log.capture_miss_count++; | |
| 1081 | − } | |
| 1082 | −} | |
| 1083 | − | |
| 1084 | −static void guc_capture_logs_work(struct work_struct *work) | |
| 1085 | −{ | |
| 1086 | − struct drm_i915_private *dev_priv = | |
| 1087 | − container_of(work, struct drm_i915_private, guc.log.flush_work); | |
| 1088 | − | |
| 1089 | − i915_guc_capture_logs(dev_priv); | |
| 1090 | −} | |
| 1091 | − | |
| 1092 | −static void guc_log_cleanup(struct intel_guc *guc) | |
| 1093 | −{ | |
| 1094 | − struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 1095 | − | |
| 1096 | − lockdep_assert_held(&dev_priv->drm.struct_mutex); | |
| 1097 | − | |
| 1098 | − /* First disable the flush interrupt */ | |
| 1099 | − gen9_disable_guc_interrupts(dev_priv); | |
| 1100 | 775 | |
| 1101 | − if (guc->log.flush_wq) | |
| 1102 | − destroy_workqueue(guc->log.flush_wq); | |
| 1103 | − | |
| 1104 | − guc->log.flush_wq = NULL; | |
| 1105 | − | |
| 1106 | − if (guc->log.relay_chan) | |
| 1107 | − guc_log_remove_relay_file(guc); | |
| 1108 | − | |
| 1109 | − guc->log.relay_chan = NULL; | |
| 1110 | − | |
| 1111 | − if (guc->log.buf_addr) | |
| 1112 | − i915_gem_object_unpin_map(guc->log.vma->obj); | |
| 1113 | − | |
| 1114 | − guc->log.buf_addr = NULL; | |
| 1115 | −} | |
| 1116 | − | |
| 1117 | −static int guc_log_create_extras(struct intel_guc *guc) | |
| 1118 | −{ | |
| 1119 | − struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 1120 | − void *vaddr; | |
| 1121 | − int ret; | |
| 1122 | − | |
| 1123 | − lockdep_assert_held(&dev_priv->drm.struct_mutex); | |
| 1124 | − | |
| 1125 | − /* Nothing to do */ | |
| 1126 | − if (i915.guc_log_level < 0) | |
| 1127 | − return 0; | |
| 1128 | − | |
| 1129 | − if (!guc->log.buf_addr) { | |
| 1130 | − /* Create a WC (Uncached for read) vmalloc mapping of log | |
| 1131 | − * buffer pages, so that we can directly get the data | |
| 1132 | − * (up-to-date) from memory. | |
| 1133 | − */ | |
| 1134 | − vaddr = i915_gem_object_pin_map(guc->log.vma->obj, I915_MAP_WC); | |
| 1135 | − if (IS_ERR(vaddr)) { | |
| 1136 | − ret = PTR_ERR(vaddr); | |
| 1137 | − DRM_ERROR("Couldn't map log buffer pages %d\n", ret); | |
| 1138 | − return ret; | |
| 1139 | − } | |
| 1140 | − | |
| 1141 | − guc->log.buf_addr = vaddr; | |
| 1142 | − } | |
| 1143 | − | |
| 1144 | − if (!guc->log.relay_chan) { | |
| 1145 | − /* Create a relay channel, so that we have buffers for storing | |
| 1146 | − * the GuC firmware logs, the channel will be linked with a file | |
| 1147 | − * later on when debugfs is registered. | |
| 1148 | − */ | |
| 1149 | − ret = guc_log_create_relay_channel(guc); | |
| 1150 | − if (ret) | |
| 1151 | − return ret; | |
| 1152 | − } | |
| 1153 | − | |
| 1154 | − if (!guc->log.flush_wq) { | |
| 1155 | − INIT_WORK(&guc->log.flush_work, guc_capture_logs_work); | |
| 1156 | − | |
| 1157 | − /* | |
| 1158 | − * GuC log buffer flush work item has to do register access to | |
| 1159 | − * send the ack to GuC and this work item, if not synced before | |
| 1160 | − * suspend, can potentially get executed after the GFX device is | |
| 1161 | − * suspended. | |
| 1162 | − * By marking the WQ as freezable, we don't have to bother about | |
| 1163 | − * flushing of this work item from the suspend hooks, the pending | |
| 1164 | − * work item if any will be either executed before the suspend | |
| 1165 | − * or scheduled later on resume. This way the handling of work | |
| 1166 | − * item can be kept same between system suspend & rpm suspend. | |
| 1167 | − */ | |
| 1168 | − guc->log.flush_wq = alloc_ordered_workqueue("i915-guc_log", | |
| 1169 | − WQ_HIGHPRI | WQ_FREEZABLE); | |
| 1170 | − if (guc->log.flush_wq == NULL) { | |
| 1171 | − DRM_ERROR("Couldn't allocate the wq for GuC logging\n"); | |
| 1172 | − return -ENOMEM; | |
| 1173 | − } | |
| 1174 | − } | |
| 1175 | − | |
| 1176 | − return 0; | |
| 1177 | −} | |
| 1178 | − | |
| 1179 | −static void guc_log_create(struct intel_guc *guc) | |
| 1180 | −{ | |
| 1181 | − struct i915_vma *vma; | |
| 1182 | − unsigned long offset; | |
| 1183 | − uint32_t size, flags; | |
| 1184 | − | |
| 1185 | − if (i915.guc_log_level > GUC_LOG_VERBOSITY_MAX) | |
| 1186 | − i915.guc_log_level = GUC_LOG_VERBOSITY_MAX; | |
| 1187 | − | |
| 1188 | − /* The first page is to save log buffer state. Allocate one | |
| 1189 | − * extra page for others in case for overlap */ | |
| 1190 | − size = (1 + GUC_LOG_DPC_PAGES + 1 + | |
| 1191 | − GUC_LOG_ISR_PAGES + 1 + | |
| 1192 | − GUC_LOG_CRASH_PAGES + 1) << PAGE_SHIFT; | |
| 1193 | − | |
| 1194 | − vma = guc->log.vma; | |
| 1195 | − if (!vma) { | |
| 1196 | − /* We require SSE 4.1 for fast reads from the GuC log buffer and | |
| 1197 | − * it should be present on the chipsets supporting GuC based | |
| 1198 | − * submisssions. | |
| 1199 | − */ | |
| 1200 | − if (WARN_ON(!i915_has_memcpy_from_wc())) { | |
| 1201 | − /* logging will not be enabled */ | |
| 1202 | − i915.guc_log_level = -1; | |
| 1203 | − return; | |
| 1204 | − } | |
| 1205 | − | |
| 1206 | − vma = guc_allocate_vma(guc, size); | |
| 1207 | − if (IS_ERR(vma)) { | |
| 1208 | − /* logging will be off */ | |
| 1209 | − i915.guc_log_level = -1; | |
| 1210 | − return; | |
| 1211 | − } | |
| 1212 | − | |
| 1213 | − guc->log.vma = vma; | |
| 1214 | − | |
| 1215 | − if (guc_log_create_extras(guc)) { | |
| 1216 | − guc_log_cleanup(guc); | |
| 1217 | − i915_vma_unpin_and_release(&guc->log.vma); | |
| 1218 | − i915.guc_log_level = -1; | |
| 1219 | − return; | |
| 1220 | − } | |
| 1221 | − } | |
| 1222 | − | |
| 1223 | − /* each allocated unit is a page */ | |
| 1224 | − flags = GUC_LOG_VALID | GUC_LOG_NOTIFY_ON_HALF_FULL | | |
| 1225 | − (GUC_LOG_DPC_PAGES << GUC_LOG_DPC_SHIFT) | | |
| 1226 | − (GUC_LOG_ISR_PAGES << GUC_LOG_ISR_SHIFT) | | |
| 1227 | − (GUC_LOG_CRASH_PAGES << GUC_LOG_CRASH_SHIFT); | |
| 1228 | − | |
| 1229 | − offset = guc_ggtt_offset(vma) >> PAGE_SHIFT; /* in pages */ | |
| 1230 | − guc->log.flags = (offset << GUC_LOG_BUF_ADDR_SHIFT) | flags; | |
| 1231 | −} | |
| 1232 | − | |
| 1233 | −static int guc_log_late_setup(struct intel_guc *guc) | |
| 1234 | −{ | |
| 1235 | − struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 1236 | − int ret; | |
| 1237 | − | |
| 1238 | − lockdep_assert_held(&dev_priv->drm.struct_mutex); | |
| 1239 | − | |
| 1240 | − if (i915.guc_log_level < 0) | |
| 1241 | − return -EINVAL; | |
| 1242 | − | |
| 1243 | − /* If log_level was set as -1 at boot time, then setup needed to | |
| 1244 | − * handle log buffer flush interrupts would not have been done yet, | |
| 1245 | − * so do that now. | |
| 1246 | − */ | |
| 1247 | − ret = guc_log_create_extras(guc); | |
| 1248 | − if (ret) | |
| 1249 | − goto err; | |
| 1250 | − | |
| 1251 | − ret = guc_log_create_relay_file(guc); | |
| 1252 | − if (ret) | |
| 1253 | − goto err; | |
| 1254 | − | |
| 1255 | − return 0; | |
| 1256 | −err: | |
| 1257 | − guc_log_cleanup(guc); | |
| 1258 | − /* logging will remain off */ | |
| 1259 | − i915.guc_log_level = -1; | |
| 1260 | − return ret; | |
| 1261 | −} | |
| 1262 | 776 | |
| 1263 | 777 | static void guc_policies_init(struct guc_policies *policies) |
| 1264 | 778 | { |
| @@ -1301,7 +815,7 @@ static void guc_addon_create(struct intel_guc *guc) · 36 unmodified lines | ||
| 1301 | 815 | |
| 1302 | 816 | vma = guc->ads_vma; |
| 1303 | 817 | if (!vma) { |
| 1304 | − vma = guc_allocate_vma(guc, PAGE_ALIGN(size)); | |
| 818 | + vma = intel_guc_allocate_vma(guc, PAGE_ALIGN(size)); | |
| 1305 | 819 | if (IS_ERR(vma)) |
| 1306 | 820 | return; |
| 1307 | 821 | |
| @@ -1376,13 +890,13 @@ int i915_guc_submission_init(struct drm_i915_private *dev_priv) · 68 unmodified lines | ||
| 1376 | 890 | if (guc->ctx_pool_vma) |
| 1377 | 891 | return 0; /* already allocated */ |
| 1378 | 892 | |
| 1379 | − vma = guc_allocate_vma(guc, gemsize); | |
| 893 | + vma = intel_guc_allocate_vma(guc, gemsize); | |
| 1380 | 894 | if (IS_ERR(vma)) |
| 1381 | 895 | return PTR_ERR(vma); |
| 1382 | 896 | |
| 1383 | 897 | guc->ctx_pool_vma = vma; |
| 1384 | 898 | ida_init(&guc->ctx_ids); |
| 1385 | − guc_log_create(guc); | |
| 899 | + intel_guc_log_create(guc); | |
| 1386 | 900 | guc_addon_create(guc); |
| 1387 | 901 | |
| 1388 | 902 | guc->execbuf_client = guc_client_alloc(dev_priv, |
| @@ -1484,7 +998,7 @@ int intel_guc_suspend(struct drm_i915_private *dev_priv) · 95 unmodified lines | ||
| 1484 | 998 | struct i915_gem_context *ctx; |
| 1485 | 999 | u32 data[3]; |
| 1486 | 1000 | |
| 1487 | − if (guc->guc_fw.guc_fw_load_status != GUC_FIRMWARE_SUCCESS) | |
| 1001 | + if (guc->fw.load_status != INTEL_UC_FIRMWARE_SUCCESS) | |
| 1488 | 1002 | return 0; |
| 1489 | 1003 | |
| 1490 | 1004 | gen9_disable_guc_interrupts(dev_priv); |
| @@ -1511,7 +1025,7 @@ int intel_guc_resume(struct drm_i915_private *dev_priv) · 20 unmodified lines | ||
| 1511 | 1025 | struct i915_gem_context *ctx; |
| 1512 | 1026 | u32 data[3]; |
| 1513 | 1027 | |
| 1514 | − if (guc->guc_fw.guc_fw_load_status != GUC_FIRMWARE_SUCCESS) | |
| 1028 | + if (guc->fw.load_status != INTEL_UC_FIRMWARE_SUCCESS) | |
| 1515 | 1029 | return 0; |
| 1516 | 1030 | |
| 1517 | 1031 | if (i915.guc_log_level >= 0) |
| @@ -1527,103 +1041,4 @@ int intel_guc_resume(struct drm_i915_private *dev_priv) · 9 unmodified lines | ||
| 1527 | 1041 | return intel_guc_send(guc, data, ARRAY_SIZE(data)); |
| 1528 | 1042 | } |
| 1529 | 1043 | |
| 1530 | −void i915_guc_capture_logs(struct drm_i915_private *dev_priv) | |
| 1531 | −{ | |
| 1532 | − guc_read_update_log_buffer(&dev_priv->guc); | |
| 1533 | − | |
| 1534 | − /* Generally device is expected to be active only at this | |
| 1535 | − * time, so get/put should be really quick. | |
| 1536 | − */ | |
| 1537 | − intel_runtime_pm_get(dev_priv); | |
| 1538 | − intel_guc_log_flush_complete(&dev_priv->guc); | |
| 1539 | − intel_runtime_pm_put(dev_priv); | |
| 1540 | −} | |
| 1541 | − | |
| 1542 | −void i915_guc_flush_logs(struct drm_i915_private *dev_priv) | |
| 1543 | −{ | |
| 1544 | − if (!i915.enable_guc_submission || (i915.guc_log_level < 0)) | |
| 1545 | − return; | |
| 1546 | − | |
| 1547 | − /* First disable the interrupts, will be renabled afterwards */ | |
| 1548 | − gen9_disable_guc_interrupts(dev_priv); | |
| 1549 | − | |
| 1550 | − /* Before initiating the forceful flush, wait for any pending/ongoing | |
| 1551 | − * flush to complete otherwise forceful flush may not actually happen. | |
| 1552 | − */ | |
| 1553 | − flush_work(&dev_priv->guc.log.flush_work); | |
| 1554 | − | |
| 1555 | − /* Ask GuC to update the log buffer state */ | |
| 1556 | − intel_guc_log_flush(&dev_priv->guc); | |
| 1557 | − | |
| 1558 | − /* GuC would have updated log buffer by now, so capture it */ | |
| 1559 | − i915_guc_capture_logs(dev_priv); | |
| 1560 | −} | |
| 1561 | − | |
| 1562 | −void i915_guc_unregister(struct drm_i915_private *dev_priv) | |
| 1563 | −{ | |
| 1564 | − if (!i915.enable_guc_submission) | |
| 1565 | − return; | |
| 1566 | 1044 | |
| 1567 | − mutex_lock(&dev_priv->drm.struct_mutex); | |
| 1568 | − guc_log_cleanup(&dev_priv->guc); | |
| 1569 | − mutex_unlock(&dev_priv->drm.struct_mutex); | |
| 1570 | −} | |
| 1571 | − | |
| 1572 | −void i915_guc_register(struct drm_i915_private *dev_priv) | |
| 1573 | −{ | |
| 1574 | − if (!i915.enable_guc_submission) | |
| 1575 | − return; | |
| 1576 | − | |
| 1577 | − mutex_lock(&dev_priv->drm.struct_mutex); | |
| 1578 | − guc_log_late_setup(&dev_priv->guc); | |
| 1579 | − mutex_unlock(&dev_priv->drm.struct_mutex); | |
| 1580 | −} | |
| 1581 | − | |
| 1582 | −int i915_guc_log_control(struct drm_i915_private *dev_priv, u64 control_val) | |
| 1583 | −{ | |
| 1584 | − union guc_log_control log_param; | |
| 1585 | − int ret; | |
| 1586 | − | |
| 1587 | − log_param.value = control_val; | |
| 1588 | − | |
| 1589 | − if (log_param.verbosity < GUC_LOG_VERBOSITY_MIN || | |
| 1590 | − log_param.verbosity > GUC_LOG_VERBOSITY_MAX) | |
| 1591 | − return -EINVAL; | |
| 1592 | − | |
| 1593 | − /* This combination doesn't make sense & won't have any effect */ | |
| 1594 | − if (!log_param.logging_enabled && (i915.guc_log_level < 0)) | |
| 1595 | − return 0; | |
| 1596 | − | |
| 1597 | − ret = intel_guc_log_control(&dev_priv->guc, log_param.value); | |
| 1598 | − if (ret < 0) { | |
| 1599 | − DRM_DEBUG_DRIVER("guc_logging_control action failed %d\n", ret); | |
| 1600 | − return ret; | |
| 1601 | − } | |
| 1602 | − | |
| 1603 | − i915.guc_log_level = log_param.verbosity; | |
| 1604 | − | |
| 1605 | − /* If log_level was set as -1 at boot time, then the relay channel file | |
| 1606 | − * wouldn't have been created by now and interrupts also would not have | |
| 1607 | − * been enabled. | |
| 1608 | − */ | |
| 1609 | − if (!dev_priv->guc.log.relay_chan) { | |
| 1610 | − ret = guc_log_late_setup(&dev_priv->guc); | |
| 1611 | − if (!ret) | |
| 1612 | − gen9_enable_guc_interrupts(dev_priv); | |
| 1613 | − } else if (!log_param.logging_enabled) { | |
| 1614 | − /* Once logging is disabled, GuC won't generate logs & send an | |
| 1615 | − * interrupt. But there could be some data in the log buffer | |
| 1616 | − * which is yet to be captured. So request GuC to update the log | |
| 1617 | − * buffer state and then collect the left over logs. | |
| 1618 | − */ | |
| 1619 | − i915_guc_flush_logs(dev_priv); | |
| 1620 | − | |
| 1621 | − /* As logging is disabled, update log level to reflect that */ | |
| 1622 | − i915.guc_log_level = -1; | |
| 1623 | − } else { | |
| 1624 | − /* In case interrupts were disabled, enable them now */ | |
| 1625 | − gen9_enable_guc_interrupts(dev_priv); | |
| 1626 | − } | |
| 1627 | − | |
| 1628 | − return ret; | |
| 1629 | −} | |
| @@ -1170,6 +1170,9 @@ static void gen6_pm_rps_work(struct work_struct *work) · 1169 unmodified lines | ||
| 1170 | 1170 | adj *= 2; |
| 1171 | 1171 | else /* CHV needs even encode values */ |
| 1172 | 1172 | adj = IS_CHERRYVIEW(dev_priv) ? 2 : 1; |
| 1173 | + | |
| 1174 | + if (new_delay >= dev_priv->rps.max_freq_softlimit) | |
| 1175 | + adj = 0; | |
| 1173 | 1176 | /* |
| 1174 | 1177 | * For better performance, jump directly |
| 1175 | 1178 | * to RPe if we're below it. |
| @@ -1191,6 +1194,9 @@ static void gen6_pm_rps_work(struct work_struct *work) · 15 unmodified lines | ||
| 1191 | 1194 | adj *= 2; |
| 1192 | 1195 | else /* CHV needs even encode values */ |
| 1193 | 1196 | adj = IS_CHERRYVIEW(dev_priv) ? -2 : -1; |
| 1197 | + | |
| 1198 | + if (new_delay <= dev_priv->rps.min_freq_softlimit) | |
| 1199 | + adj = 0; | |
| 1194 | 1200 | } else { /* unknown event */ |
| 1195 | 1201 | adj = 0; |
| 1196 | 1202 | } |
| @@ -1553,41 +1559,68 @@ static void display_pipe_crc_irq_handler(struct drm_i915_private *dev_priv, · 356 unmodified lines | ||
| 1553 | 1559 | { |
| 1554 | 1560 | struct intel_pipe_crc *pipe_crc = &dev_priv->pipe_crc[pipe]; |
| 1555 | 1561 | struct intel_pipe_crc_entry *entry; |
| 1562 | + struct intel_crtc *crtc = intel_get_crtc_for_pipe(dev_priv, pipe); | |
| 1563 | + struct drm_driver *driver = dev_priv->drm.driver; | |
| 1564 | + uint32_t crcs[5]; | |
| 1556 | 1565 | int head, tail; |
| 1557 | 1566 | |
| 1558 | 1567 | spin_lock(&pipe_crc->lock); |
| 1568 | + if (pipe_crc->source) { | |
| 1569 | + if (!pipe_crc->entries) { | |
| 1570 | + spin_unlock(&pipe_crc->lock); | |
| 1571 | + DRM_DEBUG_KMS("spurious interrupt\n"); | |
| 1572 | + return; | |
| 1573 | + } | |
| 1559 | 1574 | |
| 1560 | − if (!pipe_crc->entries) { | |
| 1561 | − spin_unlock(&pipe_crc->lock); | |
| 1562 | − DRM_DEBUG_KMS("spurious interrupt\n"); | |
| 1563 | − return; | |
| 1564 | − } | |
| 1565 | − | |
| 1566 | − head = pipe_crc->head; | |
| 1567 | − tail = pipe_crc->tail; | |
| 1575 | + head = pipe_crc->head; | |
| 1576 | + tail = pipe_crc->tail; | |
| 1568 | 1577 | |
| 1569 | − if (CIRC_SPACE(head, tail, INTEL_PIPE_CRC_ENTRIES_NR) < 1) { | |
| 1570 | − spin_unlock(&pipe_crc->lock); | |
| 1571 | − DRM_ERROR("CRC buffer overflowing\n"); | |
| 1572 | − return; | |
| 1573 | − } | |
| 1578 | + if (CIRC_SPACE(head, tail, INTEL_PIPE_CRC_ENTRIES_NR) < 1) { | |
| 1579 | + spin_unlock(&pipe_crc->lock); | |
| 1580 | + DRM_ERROR("CRC buffer overflowing\n"); | |
| 1581 | + return; | |
| 1582 | + } | |
| 1574 | 1583 | |
| 1575 | − entry = &pipe_crc->entries[head]; | |
| 1584 | + entry = &pipe_crc->entries[head]; | |
| 1576 | 1585 | |
| 1577 | − entry->frame = dev_priv->drm.driver->get_vblank_counter(&dev_priv->drm, | |
| 1578 | − pipe); | |
| 1579 | − entry->crc[0] = crc0; | |
| 1580 | − entry->crc[1] = crc1; | |
| 1581 | − entry->crc[2] = crc2; | |
| 1582 | − entry->crc[3] = crc3; | |
| 1583 | − entry->crc[4] = crc4; | |
| 1586 | + entry->frame = driver->get_vblank_counter(&dev_priv->drm, pipe); | |
| 1587 | + entry->crc[0] = crc0; | |
| 1588 | + entry->crc[1] = crc1; | |
| 1589 | + entry->crc[2] = crc2; | |
| 1590 | + entry->crc[3] = crc3; | |
| 1591 | + entry->crc[4] = crc4; | |
| 1584 | 1592 | |
| 1585 | − head = (head + 1) & (INTEL_PIPE_CRC_ENTRIES_NR - 1); | |
| 1586 | − pipe_crc->head = head; | |
| 1593 | + head = (head + 1) & (INTEL_PIPE_CRC_ENTRIES_NR - 1); | |
| 1594 | + pipe_crc->head = head; | |
| 1587 | 1595 | |
| 1588 | − spin_unlock(&pipe_crc->lock); | |
| 1596 | + spin_unlock(&pipe_crc->lock); | |
| 1589 | 1597 | |
| 1590 | − wake_up_interruptible(&pipe_crc->wq); | |
| 1598 | + wake_up_interruptible(&pipe_crc->wq); | |
| 1599 | + } else { | |
| 1600 | + /* | |
| 1601 | + * For some not yet identified reason, the first CRC is | |
| 1602 | + * bonkers. So let's just wait for the next vblank and read | |
| 1603 | + * out the buggy result. | |
| 1604 | + * | |
| 1605 | + * On CHV sometimes the second CRC is bonkers as well, so | |
| 1606 | + * don't trust that one either. | |
| 1607 | + */ | |
| 1608 | + if (pipe_crc->skipped == 0 || | |
| 1609 | + (IS_CHERRYVIEW(dev_priv) && pipe_crc->skipped == 1)) { | |
| 1610 | + pipe_crc->skipped++; | |
| 1611 | + spin_unlock(&pipe_crc->lock); | |
| 1612 | + return; | |
| 1613 | + } | |
| 1614 | + spin_unlock(&pipe_crc->lock); | |
| 1615 | + crcs[0] = crc0; | |
| 1616 | + crcs[1] = crc1; | |
| 1617 | + crcs[2] = crc2; | |
| 1618 | + crcs[3] = crc3; | |
| 1619 | + crcs[4] = crc4; | |
| 1620 | + drm_crtc_add_crc_entry(&crtc->base, true, | |
| 1621 | + drm_accurate_vblank_count(&crtc->base), | |
| 1622 | + crcs); | |
| 1623 | + } | |
| 1591 | 1624 | } |
| 1592 | 1625 | #else |
| 1593 | 1626 | static inline void |
| @@ -2703,12 +2736,13 @@ static void i915_clear_error_registers(struct drm_i915_private *dev_priv) · 1109 unmodified lines | ||
| 2703 | 2736 | * i915_handle_error - handle a gpu error |
| 2704 | 2737 | * @dev_priv: i915 device private |
| 2705 | 2738 | * @engine_mask: mask representing engines that are hung |
| 2739 | + * @fmt: Error message format string | |
| 2740 | + * | |
| 2706 | 2741 | * Do some basic checking of register state at error time and |
| 2707 | 2742 | * dump it to the syslog. Also call i915_capture_error_state() to make |
| 2708 | 2743 | * sure we get a record and make it available in debugfs. Fire a uevent |
| 2709 | 2744 | * so userspace knows something bad happened (should trigger collection |
| 2710 | 2745 | * of a ring dump etc.). |
| 2711 | − * @fmt: Error message format string | |
| 2712 | 2746 | */ |
| 2713 | 2747 | void i915_handle_error(struct drm_i915_private *dev_priv, |
| 2714 | 2748 | u32 engine_mask, |
| @@ -3597,9 +3597,12 @@ enum { · 3596 unmodified lines | ||
| 3597 | 3597 | #define EDP_PSR_PERF_CNT_MASK 0xffffff |
| 3598 | 3598 | |
| 3599 | 3599 | #define EDP_PSR_DEBUG_CTL _MMIO(dev_priv->psr_mmio_base + 0x60) |
| 3600 | −#define EDP_PSR_DEBUG_MASK_LPSP (1<<27) | |
| 3601 | −#define EDP_PSR_DEBUG_MASK_MEMUP (1<<26) | |
| 3602 | −#define EDP_PSR_DEBUG_MASK_HPD (1<<25) | |
| 3600 | +#define EDP_PSR_DEBUG_MASK_MAX_SLEEP (1<<28) | |
| 3601 | +#define EDP_PSR_DEBUG_MASK_LPSP (1<<27) | |
| 3602 | +#define EDP_PSR_DEBUG_MASK_MEMUP (1<<26) | |
| 3603 | +#define EDP_PSR_DEBUG_MASK_HPD (1<<25) | |
| 3604 | +#define EDP_PSR_DEBUG_MASK_DISP_REG_WRITE (1<<16) | |
| 3605 | +#define EDP_PSR_DEBUG_EXIT_ON_PIXEL_UNDERRUN (1<<15) | |
| 3603 | 3606 | |
| 3604 | 3607 | #define EDP_PSR2_CTL _MMIO(0x6f900) |
| 3605 | 3608 | #define EDP_PSR2_ENABLE (1<<31) |
| @@ -3614,6 +3617,11 @@ enum { · 8 unmodified lines | ||
| 3614 | 3617 | #define EDP_PSR2_FRAME_BEFORE_SU_SHIFT 4 |
| 3615 | 3618 | #define EDP_PSR2_FRAME_BEFORE_SU_MASK (0xf<<4) |
| 3616 | 3619 | #define EDP_PSR2_IDLE_MASK 0xf |
| 3620 | +#define EDP_FRAMES_BEFORE_SU_ENTRY (1<<4) | |
| 3621 | + | |
| 3622 | +#define EDP_PSR2_STATUS_CTL _MMIO(0x6f940) | |
| 3623 | +#define EDP_PSR2_STATUS_STATE_MASK (0xf<<28) | |
| 3624 | +#define EDP_PSR2_STATUS_STATE_SHIFT 28 | |
| 3617 | 3625 | |
| 3618 | 3626 | /* VGA port control */ |
| 3619 | 3627 | #define ADPA _MMIO(0x61100) |
| @@ -6446,6 +6454,12 @@ enum { · 2826 unmodified lines | ||
| 6446 | 6454 | #define BDW_DPRS_MASK_VBLANK_SRD (1 << 0) |
| 6447 | 6455 | #define CHICKEN_PIPESL_1(pipe) _MMIO_PIPE(pipe, _CHICKEN_PIPESL_1_A, _CHICKEN_PIPESL_1_B) |
| 6448 | 6456 | |
| 6457 | +#define CHICKEN_TRANS_A 0x420c0 | |
| 6458 | +#define CHICKEN_TRANS_B 0x420c4 | |
| 6459 | +#define CHICKEN_TRANS(trans) _MMIO_TRANS(trans, CHICKEN_TRANS_A, CHICKEN_TRANS_B) | |
| 6460 | +#define PSR2_VSC_ENABLE_PROG_HEADER (1<<12) | |
| 6461 | +#define PSR2_ADD_VERTICAL_LINE_COUNT (1<<15) | |
| 6462 | + | |
| 6449 | 6463 | #define DISP_ARB_CTL _MMIO(0x45000) |
| 6450 | 6464 | #define DISP_FBC_MEMORY_WAKE (1<<31) |
| 6451 | 6465 | #define DISP_TILE_SURFACE_SWIZZLING (1<<13) |
| @@ -63,6 +63,7 @@ static inline void debug_fence_destroy(struct i915_sw_fence *fence) · 62 unmodified lines | ||
| 63 | 63 | static inline void debug_fence_free(struct i915_sw_fence *fence) |
| 64 | 64 | { |
| 65 | 65 | debug_object_free(fence, &i915_sw_fence_debug_descr); |
| 66 | + smp_wmb(); /* flush the change in state before reallocation */ | |
| 66 | 67 | } |
| 67 | 68 | |
| 68 | 69 | static inline void debug_fence_assert(struct i915_sw_fence *fence) |
| @@ -58,7 +58,7 @@ static u32 calc_residency(struct drm_i915_private *dev_priv, · 57 unmodified lines | ||
| 58 | 58 | |
| 59 | 59 | if (I915_READ(VLV_COUNTER_CONTROL) & VLV_COUNT_RANGE_HIGH) |
| 60 | 60 | units <<= 8; |
| 61 | − } else if (IS_BROXTON(dev_priv)) { | |
| 61 | + } else if (IS_GEN9_LP(dev_priv)) { | |
| 62 | 62 | units = 1; |
| 63 | 63 | div = 1200; /* 833.33ns */ |
| 64 | 64 | } |
| @@ -450,9 +450,9 @@ TRACE_EVENT(i915_gem_evict_vm, · 449 unmodified lines | ||
| 450 | 450 | TP_printk("dev=%d, vm=%p", __entry->dev, __entry->vm) |
| 451 | 451 | ); |
| 452 | 452 | |
| 453 | −TRACE_EVENT(i915_gem_evict_vma, | |
| 454 | − TP_PROTO(struct i915_vma *vma, unsigned int flags), | |
| 455 | − TP_ARGS(vma, flags), | |
| 453 | +TRACE_EVENT(i915_gem_evict_node, | |
| 454 | + TP_PROTO(struct i915_address_space *vm, struct drm_mm_node *node, unsigned int flags), | |
| 455 | + TP_ARGS(vm, node, flags), | |
| 456 | 456 | |
| 457 | 457 | TP_STRUCT__entry( |
| 458 | 458 | __field(u32, dev) |
| @@ -464,11 +464,11 @@ TRACE_EVENT(i915_gem_evict_vma, · 5 unmodified lines | ||
| 464 | 464 | ), |
| 465 | 465 | |
| 466 | 466 | TP_fast_assign( |
| 467 | − __entry->dev = vma->vm->i915->drm.primary->index; | |
| 468 | − __entry->vm = vma->vm; | |
| 469 | − __entry->start = vma->node.start; | |
| 470 | − __entry->size = vma->node.size; | |
| 471 | − __entry->color = vma->node.color; | |
| 467 | + __entry->dev = vm->i915->drm.primary->index; | |
| 468 | + __entry->vm = vm; | |
| 469 | + __entry->start = node->start; | |
| 470 | + __entry->size = node->size; | |
| 471 | + __entry->color = node->color; | |
| 472 | 472 | __entry->flags = flags; |
| 473 | 473 | ), |
| 474 | 474 | |
| @@ -116,22 +116,20 @@ void intel_vgt_deballoon(struct drm_i915_private *dev_priv) · 115 unmodified lines | ||
| 116 | 116 | memset(&bl_info, 0, sizeof(bl_info)); |
| 117 | 117 | } |
| 118 | 118 | |
| 119 | −static int vgt_balloon_space(struct drm_mm *mm, | |
| 119 | +static int vgt_balloon_space(struct i915_ggtt *ggtt, | |
| 120 | 120 | struct drm_mm_node *node, |
| 121 | 121 | unsigned long start, unsigned long end) |
| 122 | 122 | { |
| 123 | 123 | unsigned long size = end - start; |
| 124 | 124 | |
| 125 | − if (start == end) | |
| 125 | + if (start >= end) | |
| 126 | 126 | return -EINVAL; |
| 127 | 127 | |
| 128 | 128 | DRM_INFO("balloon space: range [ 0x%lx - 0x%lx ] %lu KiB.\n", |
| 129 | 129 | start, end, size / 1024); |
| 130 | − | |
| 131 | − node->start = start; | |
| 132 | − node->size = size; | |
| 133 | − | |
| 134 | − return drm_mm_reserve_node(mm, node); | |
| 130 | + return i915_gem_gtt_reserve(&ggtt->base, node, | |
| 131 | + size, start, I915_COLOR_UNEVICTABLE, | |
| 132 | + 0); | |
| 135 | 133 | } |
| 136 | 134 | |
| 137 | 135 | /** |
| @@ -214,10 +212,8 @@ int intel_vgt_balloon(struct drm_i915_private *dev_priv) · 76 unmodified lines | ||
| 214 | 212 | |
| 215 | 213 | /* Unmappable graphic memory ballooning */ |
| 216 | 214 | if (unmappable_base > ggtt->mappable_end) { |
| 217 | − ret = vgt_balloon_space(&ggtt->base.mm, | |
| 218 | − &bl_info.space[2], | |
| 219 | − ggtt->mappable_end, | |
| 220 | − unmappable_base); | |
| 215 | + ret = vgt_balloon_space(ggtt, &bl_info.space[2], | |
| 216 | + ggtt->mappable_end, unmappable_base); | |
| 221 | 217 | |
| 222 | 218 | if (ret) |
| 223 | 219 | goto err; |
| @@ -228,18 +224,15 @@ int intel_vgt_balloon(struct drm_i915_private *dev_priv) · 4 unmodified lines | ||
| 228 | 224 | * because it is reserved to the guard page. |
| 229 | 225 | */ |
| 230 | 226 | if (unmappable_end < ggtt_end - PAGE_SIZE) { |
| 231 | − ret = vgt_balloon_space(&ggtt->base.mm, | |
| 232 | − &bl_info.space[3], | |
| 233 | − unmappable_end, | |
| 234 | − ggtt_end - PAGE_SIZE); | |
| 227 | + ret = vgt_balloon_space(ggtt, &bl_info.space[3], | |
| 228 | + unmappable_end, ggtt_end - PAGE_SIZE); | |
| 235 | 229 | if (ret) |
| 236 | 230 | goto err; |
| 237 | 231 | } |
| 238 | 232 | |
| 239 | 233 | /* Mappable graphic memory ballooning */ |
| 240 | 234 | if (mappable_base > ggtt->base.start) { |
| 241 | − ret = vgt_balloon_space(&ggtt->base.mm, | |
| 242 | − &bl_info.space[0], | |
| 235 | + ret = vgt_balloon_space(ggtt, &bl_info.space[0], | |
| 243 | 236 | ggtt->base.start, mappable_base); |
| 244 | 237 | |
| 245 | 238 | if (ret) |
| @@ -247,10 +240,8 @@ int intel_vgt_balloon(struct drm_i915_private *dev_priv) · 1 unmodified line | ||
| 247 | 240 | } |
| 248 | 241 | |
| 249 | 242 | if (mappable_end < ggtt->mappable_end) { |
| 250 | − ret = vgt_balloon_space(&ggtt->base.mm, | |
| 251 | − &bl_info.space[1], | |
| 252 | − mappable_end, | |
| 253 | − ggtt->mappable_end); | |
| 243 | + ret = vgt_balloon_space(ggtt, &bl_info.space[1], | |
| 244 | + mappable_end, ggtt->mappable_end); | |
| 254 | 245 | |
| 255 | 246 | if (ret) |
| 256 | 247 | goto err; |
| @@ -45,6 +45,7 @@ i915_vma_retire(struct i915_gem_active *active, · 44 unmodified lines | ||
| 45 | 45 | if (i915_vma_is_active(vma)) |
| 46 | 46 | return; |
| 47 | 47 | |
| 48 | + GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); | |
| 48 | 49 | list_move_tail(&vma->vm_link, &vma->vm->inactive_list); |
| 49 | 50 | if (unlikely(i915_vma_is_closed(vma) && !i915_vma_is_pinned(vma))) |
| 50 | 51 | WARN_ON(i915_vma_unbind(vma)); |
| @@ -69,17 +70,15 @@ i915_vma_retire(struct i915_gem_active *active, · 18 unmodified lines | ||
| 69 | 70 | } |
| 70 | 71 | |
| 71 | 72 | static struct i915_vma * |
| 72 | −__i915_vma_create(struct drm_i915_gem_object *obj, | |
| 73 | − struct i915_address_space *vm, | |
| 74 | − const struct i915_ggtt_view *view) | |
| 73 | +vma_create(struct drm_i915_gem_object *obj, | |
| 74 | + struct i915_address_space *vm, | |
| 75 | + const struct i915_ggtt_view *view) | |
| 75 | 76 | { |
| 76 | 77 | struct i915_vma *vma; |
| 77 | 78 | struct rb_node *rb, **p; |
| 78 | 79 | int i; |
| 79 | 80 | |
| 80 | − GEM_BUG_ON(vm->closed); | |
| 81 | − | |
| 82 | − vma = kmem_cache_zalloc(to_i915(obj->base.dev)->vmas, GFP_KERNEL); | |
| 81 | + vma = kmem_cache_zalloc(vm->i915->vmas, GFP_KERNEL); | |
| 83 | 82 | if (vma == NULL) |
| 84 | 83 | return ERR_PTR(-ENOMEM); |
| 85 | 84 | |
| @@ -87,29 +86,50 @@ __i915_vma_create(struct drm_i915_gem_object *obj, · 1 unmodified line | ||
| 87 | 86 | for (i = 0; i < ARRAY_SIZE(vma->last_read); i++) |
| 88 | 87 | init_request_active(&vma->last_read[i], i915_vma_retire); |
| 89 | 88 | init_request_active(&vma->last_fence, NULL); |
| 90 | − list_add(&vma->vm_link, &vm->unbound_list); | |
| 91 | 89 | vma->vm = vm; |
| 92 | 90 | vma->obj = obj; |
| 93 | 91 | vma->size = obj->base.size; |
| 92 | + vma->display_alignment = I915_GTT_MIN_ALIGNMENT; | |
| 94 | 93 | |
| 95 | 94 | if (view) { |
| 96 | 95 | vma->ggtt_view = *view; |
| 97 | 96 | if (view->type == I915_GGTT_VIEW_PARTIAL) { |
| 98 | 97 | GEM_BUG_ON(range_overflows_t(u64, |
| 99 | − view->params.partial.offset, | |
| 100 | − view->params.partial.size, | |
| 98 | + view->partial.offset, | |
| 99 | + view->partial.size, | |
| 101 | 100 | obj->base.size >> PAGE_SHIFT)); |
| 102 | − vma->size = view->params.partial.size; | |
| 101 | + vma->size = view->partial.size; | |
| 103 | 102 | vma->size <<= PAGE_SHIFT; |
| 104 | 103 | GEM_BUG_ON(vma->size >= obj->base.size); |
| 105 | 104 | } else if (view->type == I915_GGTT_VIEW_ROTATED) { |
| 106 | − vma->size = | |
| 107 | − intel_rotation_info_size(&view->params.rotated); | |
| 105 | + vma->size = intel_rotation_info_size(&view->rotated); | |
| 108 | 106 | vma->size <<= PAGE_SHIFT; |
| 109 | 107 | } |
| 110 | 108 | } |
| 111 | 109 | |
| 110 | + if (unlikely(vma->size > vm->total)) | |
| 111 | + goto err_vma; | |
| 112 | + | |
| 113 | + GEM_BUG_ON(!IS_ALIGNED(vma->size, I915_GTT_PAGE_SIZE)); | |
| 114 | + | |
| 112 | 115 | if (i915_is_ggtt(vm)) { |
| 116 | + if (unlikely(overflows_type(vma->size, u32))) | |
| 117 | + goto err_vma; | |
| 118 | + | |
| 119 | + vma->fence_size = i915_gem_fence_size(vm->i915, vma->size, | |
| 120 | + i915_gem_object_get_tiling(obj), | |
| 121 | + i915_gem_object_get_stride(obj)); | |
| 122 | + if (unlikely(vma->fence_size < vma->size || /* overflow */ | |
| 123 | + vma->fence_size > vm->total)) | |
| 124 | + goto err_vma; | |
| 125 | + | |
| 126 | + GEM_BUG_ON(!IS_ALIGNED(vma->fence_size, I915_GTT_MIN_ALIGNMENT)); | |
| 127 | + | |
| 128 | + vma->fence_alignment = i915_gem_fence_alignment(vm->i915, vma->size, | |
| 129 | + i915_gem_object_get_tiling(obj), | |
| 130 | + i915_gem_object_get_stride(obj)); | |
| 131 | + GEM_BUG_ON(!is_power_of_2(vma->fence_alignment)); | |
| 132 | + | |
| 113 | 133 | vma->flags |= I915_VMA_GGTT; |
| 114 | 134 | list_add(&vma->obj_link, &obj->vma_list); |
| 115 | 135 | } else { |
| @@ -131,20 +151,74 @@ __i915_vma_create(struct drm_i915_gem_object *obj, · 15 unmodified lines | ||
| 131 | 151 | } |
| 132 | 152 | rb_link_node(&vma->obj_node, rb, p); |
| 133 | 153 | rb_insert_color(&vma->obj_node, &obj->vma_tree); |
| 154 | + list_add(&vma->vm_link, &vm->unbound_list); | |
| 134 | 155 | |
| 135 | 156 | return vma; |
| 157 | + | |
| 158 | +err_vma: | |
| 159 | + kmem_cache_free(vm->i915->vmas, vma); | |
| 160 | + return ERR_PTR(-E2BIG); | |
| 136 | 161 | } |
| 137 | 162 | |
| 163 | +static struct i915_vma * | |
| 164 | +vma_lookup(struct drm_i915_gem_object *obj, | |
| 165 | + struct i915_address_space *vm, | |
| 166 | + const struct i915_ggtt_view *view) | |
| 167 | +{ | |
| 168 | + struct rb_node *rb; | |
| 169 | + | |
| 170 | + rb = obj->vma_tree.rb_node; | |
| 171 | + while (rb) { | |
| 172 | + struct i915_vma *vma = rb_entry(rb, struct i915_vma, obj_node); | |
| 173 | + long cmp; | |
| 174 | + | |
| 175 | + cmp = i915_vma_compare(vma, vm, view); | |
| 176 | + if (cmp == 0) | |
| 177 | + return vma; | |
| 178 | + | |
| 179 | + if (cmp < 0) | |
| 180 | + rb = rb->rb_right; | |
| 181 | + else | |
| 182 | + rb = rb->rb_left; | |
| 183 | + } | |
| 184 | + | |
| 185 | + return NULL; | |
| 186 | +} | |
| 187 | + | |
| 188 | +/** | |
| 189 | + * i915_vma_instance - return the singleton instance of the VMA | |
| 190 | + * @obj: parent &struct drm_i915_gem_object to be mapped | |
| 191 | + * @vm: address space in which the mapping is located | |
| 192 | + * @view: additional mapping requirements | |
| 193 | + * | |
| 194 | + * i915_vma_instance() looks up an existing VMA of the @obj in the @vm with | |
| 195 | + * the same @view characteristics. If a match is not found, one is created. | |
| 196 | + * Once created, the VMA is kept until either the object is freed, or the | |
| 197 | + * address space is closed. | |
| 198 | + * | |
| 199 | + * Must be called with struct_mutex held. | |
| 200 | + * | |
| 201 | + * Returns the vma, or an error pointer. | |
| 202 | + */ | |
| 138 | 203 | struct i915_vma * |
| 139 | −i915_vma_create(struct drm_i915_gem_object *obj, | |
| 140 | − struct i915_address_space *vm, | |
| 141 | − const struct i915_ggtt_view *view) | |
| 204 | +i915_vma_instance(struct drm_i915_gem_object *obj, | |
| 205 | + struct i915_address_space *vm, | |
| 206 | + const struct i915_ggtt_view *view) | |
| 142 | 207 | { |
| 208 | + struct i915_vma *vma; | |
| 209 | + | |
| 143 | 210 | lockdep_assert_held(&obj->base.dev->struct_mutex); |
| 144 | 211 | GEM_BUG_ON(view && !i915_is_ggtt(vm)); |
| 145 | − GEM_BUG_ON(i915_gem_obj_to_vma(obj, vm, view)); | |
| 212 | + GEM_BUG_ON(vm->closed); | |
| 213 | + | |
| 214 | + vma = vma_lookup(obj, vm, view); | |
| 215 | + if (!vma) | |
| 216 | + vma = vma_create(obj, vm, view); | |
| 146 | 217 | |
| 147 | − return __i915_vma_create(obj, vm, view); | |
| 218 | + GEM_BUG_ON(!IS_ERR(vma) && i915_vma_is_closed(vma)); | |
| 219 | + GEM_BUG_ON(!IS_ERR(vma) && i915_vma_compare(vma, vm, view)); | |
| 220 | + GEM_BUG_ON(!IS_ERR(vma) && vma_lookup(obj, vm, view) != vma); | |
| 221 | + return vma; | |
| 148 | 222 | } |
| 149 | 223 | |
| 150 | 224 | /** |
| @@ -195,6 +269,7 @@ int i915_vma_bind(struct i915_vma *vma, enum i915_cache_level cache_level, · 44 unmodified lines | ||
| 195 | 269 | return ret; |
| 196 | 270 | } |
| 197 | 271 | |
| 272 | + trace_i915_vma_bind(vma, bind_flags); | |
| 198 | 273 | ret = vma->vm->bind_vma(vma, cache_level, bind_flags); |
| 199 | 274 | if (ret) |
| 200 | 275 | return ret; |
| @@ -258,7 +333,8 @@ i915_vma_misplaced(struct i915_vma *vma, u64 size, u64 alignment, u64 flags) · 57 unmodified lines | ||
| 258 | 333 | if (vma->node.size < size) |
| 259 | 334 | return true; |
| 260 | 335 | |
| 261 | − if (alignment && vma->node.start & (alignment - 1)) | |
| 336 | + GEM_BUG_ON(alignment && !is_power_of_2(alignment)); | |
| 337 | + if (alignment && !IS_ALIGNED(vma->node.start, alignment)) | |
| 262 | 338 | return true; |
| 263 | 339 | |
| 264 | 340 | if (flags & PIN_MAPPABLE && !i915_vma_is_map_and_fenceable(vma)) |
| @@ -277,31 +353,24 @@ i915_vma_misplaced(struct i915_vma *vma, u64 size, u64 alignment, u64 flags) · 12 unmodified lines | ||
| 277 | 353 | |
| 278 | 354 | void __i915_vma_set_map_and_fenceable(struct i915_vma *vma) |
| 279 | 355 | { |
| 280 | − struct drm_i915_gem_object *obj = vma->obj; | |
| 281 | − struct drm_i915_private *dev_priv = to_i915(obj->base.dev); | |
| 282 | 356 | bool mappable, fenceable; |
| 283 | − u32 fence_size, fence_alignment; | |
| 284 | − | |
| 285 | − fence_size = i915_gem_get_ggtt_size(dev_priv, | |
| 286 | − vma->size, | |
| 287 | − i915_gem_object_get_tiling(obj)); | |
| 288 | − fence_alignment = i915_gem_get_ggtt_alignment(dev_priv, | |
| 289 | − vma->size, | |
| 290 | − i915_gem_object_get_tiling(obj), | |
| 291 | − true); | |
| 292 | − | |
| 293 | − fenceable = (vma->node.size == fence_size && | |
| 294 | − (vma->node.start & (fence_alignment - 1)) == 0); | |
| 295 | 357 | |
| 296 | − mappable = (vma->node.start + fence_size <= | |
| 297 | − dev_priv->ggtt.mappable_end); | |
| 358 | + GEM_BUG_ON(!i915_vma_is_ggtt(vma)); | |
| 359 | + GEM_BUG_ON(!vma->fence_size); | |
| 298 | 360 | |
| 299 | 361 | /* |
| 300 | 362 | * Explicitly disable for rotated VMA since the display does not |
| 301 | 363 | * need the fence and the VMA is not accessible to other users. |
| 302 | 364 | */ |
| 303 | − if (mappable && fenceable && | |
| 304 | − vma->ggtt_view.type != I915_GGTT_VIEW_ROTATED) | |
| 365 | + if (vma->ggtt_view.type == I915_GGTT_VIEW_ROTATED) | |
| 366 | + return; | |
| 367 | + | |
| 368 | + fenceable = (vma->node.size >= vma->fence_size && | |
| 369 | + IS_ALIGNED(vma->node.start, vma->fence_alignment)); | |
| 370 | + | |
| 371 | + mappable = vma->node.start + vma->fence_size <= i915_vm_to_ggtt(vma->vm)->mappable_end; | |
| 372 | + | |
| 373 | + if (mappable && fenceable) | |
| 305 | 374 | vma->flags |= I915_VMA_CAN_FENCE; |
| 306 | 375 | else |
| 307 | 376 | vma->flags &= ~I915_VMA_CAN_FENCE; |
| @@ -368,22 +437,26 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags) · 60 unmodified lines | ||
| 368 | 437 | GEM_BUG_ON(drm_mm_node_allocated(&vma->node)); |
| 369 | 438 | |
| 370 | 439 | size = max(size, vma->size); |
| 371 | − if (flags & PIN_MAPPABLE) | |
| 372 | − size = i915_gem_get_ggtt_size(dev_priv, size, | |
| 373 | − i915_gem_object_get_tiling(obj)); | |
| 440 | + alignment = max(alignment, vma->display_alignment); | |
| 441 | + if (flags & PIN_MAPPABLE) { | |
| 442 | + size = max_t(typeof(size), size, vma->fence_size); | |
| 443 | + alignment = max_t(typeof(alignment), | |
| 444 | + alignment, vma->fence_alignment); | |
| 445 | + } | |
| 374 | 446 | |
| 375 | − alignment = max(max(alignment, vma->display_alignment), | |
| 376 | − i915_gem_get_ggtt_alignment(dev_priv, size, | |
| 377 | − i915_gem_object_get_tiling(obj), | |
| 378 | − flags & PIN_MAPPABLE)); | |
| 447 | + GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)); | |
| 448 | + GEM_BUG_ON(!IS_ALIGNED(alignment, I915_GTT_MIN_ALIGNMENT)); | |
| 449 | + GEM_BUG_ON(!is_power_of_2(alignment)); | |
| 379 | 450 | |
| 380 | 451 | start = flags & PIN_OFFSET_BIAS ? flags & PIN_OFFSET_MASK : 0; |
| 452 | + GEM_BUG_ON(!IS_ALIGNED(start, I915_GTT_PAGE_SIZE)); | |
| 381 | 453 | |
| 382 | 454 | end = vma->vm->total; |
| 383 | 455 | if (flags & PIN_MAPPABLE) |
| 384 | 456 | end = min_t(u64, end, dev_priv->ggtt.mappable_end); |
| 385 | 457 | if (flags & PIN_ZONE_4G) |
| 386 | − end = min_t(u64, end, (1ULL << 32) - PAGE_SIZE); | |
| 458 | + end = min_t(u64, end, (1ULL << 32) - I915_GTT_PAGE_SIZE); | |
| 459 | + GEM_BUG_ON(!IS_ALIGNED(end, I915_GTT_PAGE_SIZE)); | |
| 387 | 460 | |
| 388 | 461 | /* If binding the object/GGTT view requires more space than the entire |
| 389 | 462 | * aperture has, reject it early before evicting everything in a vain |
| @@ -403,65 +476,28 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags) · 13 unmodified lines | ||
| 403 | 476 | |
| 404 | 477 | if (flags & PIN_OFFSET_FIXED) { |
| 405 | 478 | u64 offset = flags & PIN_OFFSET_MASK; |
| 406 | − if (offset & (alignment - 1) || | |
| 479 | + if (!IS_ALIGNED(offset, alignment) || | |
| 407 | 480 | range_overflows(offset, size, end)) { |
| 408 | 481 | ret = -EINVAL; |
| 409 | 482 | goto err_unpin; |
| 410 | 483 | } |
| 411 | 484 | |
| 412 | − vma->node.start = offset; | |
| 413 | − vma->node.size = size; | |
| 414 | − vma->node.color = obj->cache_level; | |
| 415 | − ret = drm_mm_reserve_node(&vma->vm->mm, &vma->node); | |
| 416 | − if (ret) { | |
| 417 | − ret = i915_gem_evict_for_vma(vma, flags); | |
| 418 | − if (ret == 0) | |
| 419 | − ret = drm_mm_reserve_node(&vma->vm->mm, &vma->node); | |
| 420 | − if (ret) | |
| 421 | − goto err_unpin; | |
| 422 | − } | |
| 485 | + ret = i915_gem_gtt_reserve(vma->vm, &vma->node, | |
| 486 | + size, offset, obj->cache_level, | |
| 487 | + flags); | |
| 488 | + if (ret) | |
| 489 | + goto err_unpin; | |
| 423 | 490 | } else { |
| 424 | − u32 search_flag, alloc_flag; | |
| 425 | − | |
| 426 | − if (flags & PIN_HIGH) { | |
| 427 | − search_flag = DRM_MM_SEARCH_BELOW; | |
| 428 | − alloc_flag = DRM_MM_CREATE_TOP; | |
| 429 | − } else { | |
| 430 | − search_flag = DRM_MM_SEARCH_DEFAULT; | |
| 431 | − alloc_flag = DRM_MM_CREATE_DEFAULT; | |
| 432 | − } | |
| 433 | − | |
| 434 | − /* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks, | |
| 435 | − * so we know that we always have a minimum alignment of 4096. | |
| 436 | − * The drm_mm range manager is optimised to return results | |
| 437 | − * with zero alignment, so where possible use the optimal | |
| 438 | − * path. | |
| 439 | − */ | |
| 440 | − if (alignment <= 4096) | |
| 441 | − alignment = 0; | |
| 442 | − | |
| 443 | −search_free: | |
| 444 | − ret = drm_mm_insert_node_in_range_generic(&vma->vm->mm, | |
| 445 | − &vma->node, | |
| 446 | − size, alignment, | |
| 447 | − obj->cache_level, | |
| 448 | − start, end, | |
| 449 | − search_flag, | |
| 450 | − alloc_flag); | |
| 451 | − if (ret) { | |
| 452 | − ret = i915_gem_evict_something(vma->vm, size, alignment, | |
| 453 | − obj->cache_level, | |
| 454 | − start, end, | |
| 455 | − flags); | |
| 456 | − if (ret == 0) | |
| 457 | − goto search_free; | |
| 458 | − | |
| 491 | + ret = i915_gem_gtt_insert(vma->vm, &vma->node, | |
| 492 | + size, alignment, obj->cache_level, | |
| 493 | + start, end, flags); | |
| 494 | + if (ret) | |
| 459 | 495 | goto err_unpin; |
| 460 | − } | |
| 461 | 496 | |
| 462 | 497 | GEM_BUG_ON(vma->node.start < start); |
| 463 | 498 | GEM_BUG_ON(vma->node.start + vma->node.size > end); |
| 464 | 499 | } |
| 500 | + GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); | |
| 465 | 501 | GEM_BUG_ON(!i915_gem_valid_gtt_space(vma, obj->cache_level)); |
| 466 | 502 | |
| 467 | 503 | list_move_tail(&obj->global_link, &dev_priv->mm.bound_list); |
| @@ -504,6 +540,7 @@ int __i915_vma_do_pin(struct i915_vma *vma, · 36 unmodified lines | ||
| 504 | 540 | if ((bound ^ vma->flags) & I915_VMA_GLOBAL_BIND) |
| 505 | 541 | __i915_vma_set_map_and_fenceable(vma); |
| 506 | 542 | |
| 543 | + GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); | |
| 507 | 544 | GEM_BUG_ON(i915_vma_misplaced(vma, size, alignment, flags)); |
| 508 | 545 | return 0; |
| 509 | 546 | |
| @@ -55,6 +55,9 @@ struct i915_vma { · 54 unmodified lines | ||
| 55 | 55 | u64 size; |
| 56 | 56 | u64 display_alignment; |
| 57 | 57 | |
| 58 | + u32 fence_size; | |
| 59 | + u32 fence_alignment; | |
| 60 | + | |
| 58 | 61 | unsigned int flags; |
| 59 | 62 | /** |
| 60 | 63 | * How many users have pinned this object in GTT space. The following |
| @@ -109,9 +112,9 @@ struct i915_vma { · 48 unmodified lines | ||
| 109 | 112 | }; |
| 110 | 113 | |
| 111 | 114 | struct i915_vma * |
| 112 | −i915_vma_create(struct drm_i915_gem_object *obj, | |
| 113 | − struct i915_address_space *vm, | |
| 114 | − const struct i915_ggtt_view *view); | |
| 115 | +i915_vma_instance(struct drm_i915_gem_object *obj, | |
| 116 | + struct i915_address_space *vm, | |
| 117 | + const struct i915_ggtt_view *view); | |
| 115 | 118 | |
| 116 | 119 | void i915_vma_unpin_and_release(struct i915_vma **p_vma); |
| 117 | 120 | |
| @@ -196,15 +199,30 @@ i915_vma_compare(struct i915_vma *vma, · 78 unmodified lines | ||
| 196 | 199 | if (cmp) |
| 197 | 200 | return cmp; |
| 198 | 201 | |
| 202 | + BUILD_BUG_ON(I915_GGTT_VIEW_NORMAL != 0); | |
| 203 | + cmp = vma->ggtt_view.type; | |
| 199 | 204 | if (!view) |
| 200 | − return vma->ggtt_view.type; | |
| 205 | + return cmp; | |
| 201 | 206 | |
| 202 | − if (vma->ggtt_view.type != view->type) | |
| 203 | − return vma->ggtt_view.type - view->type; | |
| 207 | + cmp -= view->type; | |
| 208 | + if (cmp) | |
| 209 | + return cmp; | |
| 204 | 210 | |
| 205 | − return memcmp(&vma->ggtt_view.params, | |
| 206 | − &view->params, | |
| 207 | − sizeof(view->params)); | |
| 211 | + /* ggtt_view.type also encodes its size so that we both distinguish | |
| 212 | + * different views using it as a "type" and also use a compact (no | |
| 213 | + * accessing of uninitialised padding bytes) memcmp without storing | |
| 214 | + * an extra parameter or adding more code. | |
| 215 | + * | |
| 216 | + * To ensure that the memcmp is valid for all branches of the union, | |
| 217 | + * even though the code looks like it is just comparing one branch, | |
| 218 | + * we assert above that all branches have the same address, and that | |
| 219 | + * each branch has a unique type/size. | |
| 220 | + */ | |
| 221 | + BUILD_BUG_ON(I915_GGTT_VIEW_NORMAL >= I915_GGTT_VIEW_PARTIAL); | |
| 222 | + BUILD_BUG_ON(I915_GGTT_VIEW_PARTIAL >= I915_GGTT_VIEW_ROTATED); | |
| 223 | + BUILD_BUG_ON(offsetof(typeof(*view), rotated) != | |
| 224 | + offsetof(typeof(*view), partial)); | |
| 225 | + return memcmp(&vma->ggtt_view.partial, &view->partial, view->type); | |
| 208 | 226 | } |
| 209 | 227 | |
| 210 | 228 | int i915_vma_bind(struct i915_vma *vma, enum i915_cache_level cache_level, |
| @@ -229,8 +247,11 @@ i915_vma_pin(struct i915_vma *vma, u64 size, u64 alignment, u64 flags) · 18 unmodified lines | ||
| 229 | 247 | /* Pin early to prevent the shrinker/eviction logic from destroying |
| 230 | 248 | * our vma as we insert and bind. |
| 231 | 249 | */ |
| 232 | − if (likely(((++vma->flags ^ flags) & I915_VMA_BIND_MASK) == 0)) | |
| 250 | + if (likely(((++vma->flags ^ flags) & I915_VMA_BIND_MASK) == 0)) { | |
| 251 | + GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); | |
| 252 | + GEM_BUG_ON(i915_vma_misplaced(vma, size, alignment, flags)); | |
| 233 | 253 | return 0; |
| 254 | + } | |
| 234 | 255 | |
| 235 | 256 | return __i915_vma_do_pin(vma, size, alignment, flags); |
| 236 | 257 | } |
| @@ -85,6 +85,8 @@ intel_plane_duplicate_state(struct drm_plane *plane) · 84 unmodified lines | ||
| 85 | 85 | |
| 86 | 86 | __drm_atomic_helper_plane_duplicate_state(plane, state); |
| 87 | 87 | |
| 88 | + intel_state->vma = NULL; | |
| 89 | + | |
| 88 | 90 | return state; |
| 89 | 91 | } |
| 90 | 92 | |
| @@ -100,6 +102,24 @@ void · 9 unmodified lines | ||
| 100 | 102 | intel_plane_destroy_state(struct drm_plane *plane, |
| 101 | 103 | struct drm_plane_state *state) |
| 102 | 104 | { |
| 105 | + struct i915_vma *vma; | |
| 106 | + | |
| 107 | + vma = fetch_and_zero(&to_intel_plane_state(state)->vma); | |
| 108 | + | |
| 109 | + /* | |
| 110 | + * FIXME: Normally intel_cleanup_plane_fb handles destruction of vma. | |
| 111 | + * We currently don't clear all planes during driver unload, so we have | |
| 112 | + * to be able to unpin vma here for now. | |
| 113 | + * | |
| 114 | + * Normally this can only happen during unload when kmscon is disabled | |
| 115 | + * and userspace doesn't attempt to set a framebuffer at all. | |
| 116 | + */ | |
| 117 | + if (vma) { | |
| 118 | + mutex_lock(&plane->dev->struct_mutex); | |
| 119 | + intel_unpin_fb_vma(vma); | |
| 120 | + mutex_unlock(&plane->dev->struct_mutex); | |
| 121 | + } | |
| 122 | + | |
| 103 | 123 | drm_atomic_helper_plane_destroy_state(plane, state); |
| 104 | 124 | } |
| 105 | 125 | |
| @@ -34,6 +34,10 @@ · 33 unmodified lines | ||
| 34 | 34 | * low-power state and comes back to normal. |
| 35 | 35 | */ |
| 36 | 36 | |
| 37 | +#define I915_CSR_GLK "i915/glk_dmc_ver1_01.bin" | |
| 38 | +MODULE_FIRMWARE(I915_CSR_GLK); | |
| 39 | +#define GLK_CSR_VERSION_REQUIRED CSR_VERSION(1, 1) | |
| 40 | + | |
| 37 | 41 | #define I915_CSR_KBL "i915/kbl_dmc_ver1_01.bin" |
| 38 | 42 | MODULE_FIRMWARE(I915_CSR_KBL); |
| 39 | 43 | #define KBL_CSR_VERSION_REQUIRED CSR_VERSION(1, 1) |
| @@ -286,7 +290,9 @@ static uint32_t *parse_csr_fw(struct drm_i915_private *dev_priv, · 246 unmodified lines | ||
| 286 | 290 | |
| 287 | 291 | csr->version = css_header->version; |
| 288 | 292 | |
| 289 | − if (IS_KABYLAKE(dev_priv)) { | |
| 293 | + if (IS_GEMINILAKE(dev_priv)) { | |
| 294 | + required_version = GLK_CSR_VERSION_REQUIRED; | |
| 295 | + } else if (IS_KABYLAKE(dev_priv)) { | |
| 290 | 296 | required_version = KBL_CSR_VERSION_REQUIRED; |
| 291 | 297 | } else if (IS_SKYLAKE(dev_priv)) { |
| 292 | 298 | required_version = SKL_CSR_VERSION_REQUIRED; |
| @@ -435,7 +441,9 @@ void intel_csr_ucode_init(struct drm_i915_private *dev_priv) · 142 unmodified lines | ||
| 435 | 441 | if (!HAS_CSR(dev_priv)) |
| 436 | 442 | return; |
| 437 | 443 | |
| 438 | − if (IS_KABYLAKE(dev_priv)) | |
| 444 | + if (IS_GEMINILAKE(dev_priv)) | |
| 445 | + csr->fw_path = I915_CSR_GLK; | |
| 446 | + else if (IS_KABYLAKE(dev_priv)) | |
| 439 | 447 | csr->fw_path = I915_CSR_KBL; |
| 440 | 448 | else if (IS_SKYLAKE(dev_priv)) |
| 441 | 449 | csr->fw_path = I915_CSR_SKL; |
| @@ -192,7 +192,7 @@ static void gen9_sseu_info_init(struct drm_i915_private *dev_priv) · 191 unmodified lines | ||
| 192 | 192 | (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) && |
| 193 | 193 | hweight8(sseu->slice_mask) > 1; |
| 194 | 194 | sseu->has_subslice_pg = |
| 195 | − IS_BROXTON(dev_priv) && sseu_subslice_total(sseu) > 1; | |
| 195 | + IS_GEN9_LP(dev_priv) && sseu_subslice_total(sseu) > 1; | |
| 196 | 196 | sseu->has_eu_pg = sseu->eu_per_subslice > 2; |
| 197 | 197 | |
| 198 | 198 | if (IS_BROXTON(dev_priv)) { |
| @@ -2137,11 +2137,10 @@ intel_fill_fb_ggtt_view(struct i915_ggtt_view *view, · 2136 unmodified lines | ||
| 2137 | 2137 | const struct drm_framebuffer *fb, |
| 2138 | 2138 | unsigned int rotation) |
| 2139 | 2139 | { |
| 2140 | + view->type = I915_GGTT_VIEW_NORMAL; | |
| 2140 | 2141 | if (drm_rotation_90_or_270(rotation)) { |
| 2141 | − *view = i915_ggtt_view_rotated; | |
| 2142 | − view->params.rotated = to_intel_framebuffer(fb)->rot_info; | |
| 2143 | − } else { | |
| 2144 | − *view = i915_ggtt_view_normal; | |
| 2142 | + view->type = I915_GGTT_VIEW_ROTATED; | |
| 2143 | + view->rotated = to_intel_framebuffer(fb)->rot_info; | |
| 2145 | 2144 | } |
| 2146 | 2145 | } |
| 2147 | 2146 | |
| @@ -2235,24 +2234,19 @@ intel_pin_and_fence_fb_obj(struct drm_framebuffer *fb, unsigned int rotation) · 87 unmodified lines | ||
| 2235 | 2234 | i915_vma_pin_fence(vma); |
| 2236 | 2235 | } |
| 2237 | 2236 | |
| 2237 | + i915_vma_get(vma); | |
| 2238 | 2238 | err: |
| 2239 | 2239 | intel_runtime_pm_put(dev_priv); |
| 2240 | 2240 | return vma; |
| 2241 | 2241 | } |
| 2242 | 2242 | |
| 2243 | −void intel_unpin_fb_obj(struct drm_framebuffer *fb, unsigned int rotation) | |
| 2243 | +void intel_unpin_fb_vma(struct i915_vma *vma) | |
| 2244 | 2244 | { |
| 2245 | − struct drm_i915_gem_object *obj = intel_fb_obj(fb); | |
| 2246 | − struct i915_ggtt_view view; | |
| 2247 | − struct i915_vma *vma; | |
| 2248 | − | |
| 2249 | − WARN_ON(!mutex_is_locked(&obj->base.dev->struct_mutex)); | |
| 2250 | − | |
| 2251 | − intel_fill_fb_ggtt_view(&view, fb, rotation); | |
| 2252 | − vma = i915_gem_object_to_ggtt(obj, &view); | |
| 2245 | + lockdep_assert_held(&vma->vm->i915->drm.struct_mutex); | |
| 2253 | 2246 | |
| 2254 | 2247 | i915_vma_unpin_fence(vma); |
| 2255 | 2248 | i915_gem_object_unpin_from_display_plane(vma); |
| 2249 | + i915_vma_put(vma); | |
| 2256 | 2250 | } |
| 2257 | 2251 | |
| 2258 | 2252 | static int intel_fb_pitch(const struct drm_framebuffer *fb, int plane, |
| @@ -2745,7 +2739,6 @@ intel_find_initial_plane_obj(struct intel_crtc *intel_crtc, · 486 unmodified lines | ||
| 2745 | 2739 | struct drm_device *dev = intel_crtc->base.dev; |
| 2746 | 2740 | struct drm_i915_private *dev_priv = to_i915(dev); |
| 2747 | 2741 | struct drm_crtc *c; |
| 2748 | − struct intel_crtc *i; | |
| 2749 | 2742 | struct drm_i915_gem_object *obj; |
| 2750 | 2743 | struct drm_plane *primary = intel_crtc->base.primary; |
| 2751 | 2744 | struct drm_plane_state *plane_state = primary->state; |
| @@ -2770,20 +2763,20 @@ intel_find_initial_plane_obj(struct intel_crtc *intel_crtc, · 18 unmodified lines | ||
| 2770 | 2763 | * an fb with another CRTC instead |
| 2771 | 2764 | */ |
| 2772 | 2765 | for_each_crtc(dev, c) { |
| 2773 | − i = to_intel_crtc(c); | |
| 2766 | + struct intel_plane_state *state; | |
| 2774 | 2767 | |
| 2775 | 2768 | if (c == &intel_crtc->base) |
| 2776 | 2769 | continue; |
| 2777 | 2770 | |
| 2778 | − if (!i->active) | |
| 2771 | + if (!to_intel_crtc(c)->active) | |
| 2779 | 2772 | continue; |
| 2780 | 2773 | |
| 2781 | − fb = c->primary->fb; | |
| 2782 | − if (!fb) | |
| 2774 | + state = to_intel_plane_state(c->primary->state); | |
| 2775 | + if (!state->vma) | |
| 2783 | 2776 | continue; |
| 2784 | 2777 | |
| 2785 | − obj = intel_fb_obj(fb); | |
| 2786 | − if (i915_gem_object_ggtt_offset(obj, NULL) == plane_config->base) { | |
| 2778 | + if (intel_plane_ggtt_offset(state) == plane_config->base) { | |
| 2779 | + fb = c->primary->fb; | |
| 2787 | 2780 | drm_framebuffer_reference(fb); |
| 2788 | 2781 | goto valid_fb; |
| 2789 | 2782 | } |
| @@ -2796,7 +2789,7 @@ intel_find_initial_plane_obj(struct intel_crtc *intel_crtc, · 6 unmodified lines | ||
| 2796 | 2789 | * simplest solution is to just disable the primary plane now and |
| 2797 | 2790 | * pretend the BIOS never had it enabled. |
| 2798 | 2791 | */ |
| 2799 | − to_intel_plane_state(plane_state)->base.visible = false; | |
| 2792 | + plane_state->visible = false; | |
| 2800 | 2793 | crtc_state->plane_mask &= ~(1 << drm_plane_index(primary)); |
| 2801 | 2794 | intel_pre_disable_primary_noatomic(&intel_crtc->base); |
| 2802 | 2795 | intel_plane->disable_plane(primary, &intel_crtc->base); |
| @@ -2804,6 +2797,19 @@ intel_find_initial_plane_obj(struct intel_crtc *intel_crtc, · 1 unmodified line | ||
| 2804 | 2797 | return; |
| 2805 | 2798 | |
| 2806 | 2799 | valid_fb: |
| 2800 | + mutex_lock(&dev->struct_mutex); | |
| 2801 | + intel_state->vma = | |
| 2802 | + intel_pin_and_fence_fb_obj(fb, primary->state->rotation); | |
| 2803 | + mutex_unlock(&dev->struct_mutex); | |
| 2804 | + if (IS_ERR(intel_state->vma)) { | |
| 2805 | + DRM_ERROR("failed to pin boot fb on pipe %d: %li\n", | |
| 2806 | + intel_crtc->pipe, PTR_ERR(intel_state->vma)); | |
| 2807 | + | |
| 2808 | + intel_state->vma = NULL; | |
| 2809 | + drm_framebuffer_unreference(fb); | |
| 2810 | + return; | |
| 2811 | + } | |
| 2812 | + | |
| 2807 | 2813 | plane_state->src_x = 0; |
| 2808 | 2814 | plane_state->src_y = 0; |
| 2809 | 2815 | plane_state->src_w = fb->width << 16; |
| @@ -3099,13 +3105,13 @@ static void i9xx_update_primary_plane(struct drm_plane *primary, · 289 unmodified lines | ||
| 3099 | 3105 | I915_WRITE(DSPSTRIDE(plane), fb->pitches[0]); |
| 3100 | 3106 | if (INTEL_GEN(dev_priv) >= 4) { |
| 3101 | 3107 | I915_WRITE(DSPSURF(plane), |
| 3102 | − intel_fb_gtt_offset(fb, rotation) + | |
| 3108 | + intel_plane_ggtt_offset(plane_state) + | |
| 3103 | 3109 | intel_crtc->dspaddr_offset); |
| 3104 | 3110 | I915_WRITE(DSPTILEOFF(plane), (y << 16) | x); |
| 3105 | 3111 | I915_WRITE(DSPLINOFF(plane), linear_offset); |
| 3106 | 3112 | } else { |
| 3107 | 3113 | I915_WRITE(DSPADDR(plane), |
| 3108 | − intel_fb_gtt_offset(fb, rotation) + | |
| 3114 | + intel_plane_ggtt_offset(plane_state) + | |
| 3109 | 3115 | intel_crtc->dspaddr_offset); |
| 3110 | 3116 | } |
| 3111 | 3117 | POSTING_READ(reg); |
| @@ -3202,7 +3208,7 @@ static void ironlake_update_primary_plane(struct drm_plane *primary, · 90 unmodified lines | ||
| 3202 | 3208 | |
| 3203 | 3209 | I915_WRITE(DSPSTRIDE(plane), fb->pitches[0]); |
| 3204 | 3210 | I915_WRITE(DSPSURF(plane), |
| 3205 | − intel_fb_gtt_offset(fb, rotation) + | |
| 3211 | + intel_plane_ggtt_offset(plane_state) + | |
| 3206 | 3212 | intel_crtc->dspaddr_offset); |
| 3207 | 3213 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) { |
| 3208 | 3214 | I915_WRITE(DSPOFFSET(plane), (y << 16) | x); |
| @@ -3225,23 +3231,6 @@ u32 intel_fb_stride_alignment(const struct drm_i915_private *dev_priv, · 16 unmodified lines | ||
| 3225 | 3231 | } |
| 3226 | 3232 | } |
| 3227 | 3233 | |
| 3228 | −u32 intel_fb_gtt_offset(struct drm_framebuffer *fb, | |
| 3229 | − unsigned int rotation) | |
| 3230 | −{ | |
| 3231 | − struct drm_i915_gem_object *obj = intel_fb_obj(fb); | |
| 3232 | − struct i915_ggtt_view view; | |
| 3233 | − struct i915_vma *vma; | |
| 3234 | − | |
| 3235 | − intel_fill_fb_ggtt_view(&view, fb, rotation); | |
| 3236 | − | |
| 3237 | − vma = i915_gem_object_to_ggtt(obj, &view); | |
| 3238 | − if (WARN(!vma, "ggtt vma for display object not found! (view=%u)\n", | |
| 3239 | − view.type)) | |
| 3240 | − return -1; | |
| 3241 | − | |
| 3242 | − return i915_ggtt_offset(vma); | |
| 3243 | −} | |
| 3244 | − | |
| 3245 | 3234 | static void skl_detach_scaler(struct intel_crtc *intel_crtc, int id) |
| 3246 | 3235 | { |
| 3247 | 3236 | struct drm_device *dev = intel_crtc->base.dev; |
| @@ -3437,7 +3426,7 @@ static void skylake_update_primary_plane(struct drm_plane *plane, · 189 unmodified lines | ||
| 3437 | 3426 | } |
| 3438 | 3427 | |
| 3439 | 3428 | I915_WRITE(PLANE_SURF(pipe, plane_id), |
| 3440 | − intel_fb_gtt_offset(fb, rotation) + surf_addr); | |
| 3429 | + intel_plane_ggtt_offset(plane_state) + surf_addr); | |
| 3441 | 3430 | |
| 3442 | 3431 | POSTING_READ(PLANE_SURF(pipe, plane_id)); |
| 3443 | 3432 | } |
| @@ -3561,23 +3550,19 @@ void intel_prepare_reset(struct drm_i915_private *dev_priv) · 117 unmodified lines | ||
| 3561 | 3550 | state = drm_atomic_helper_duplicate_state(dev, ctx); |
| 3562 | 3551 | if (IS_ERR(state)) { |
| 3563 | 3552 | ret = PTR_ERR(state); |
| 3564 | − state = NULL; | |
| 3565 | 3553 | DRM_ERROR("Duplicating state failed with %i\n", ret); |
| 3566 | − goto err; | |
| 3554 | + return; | |
| 3567 | 3555 | } |
| 3568 | 3556 | |
| 3569 | 3557 | ret = drm_atomic_helper_disable_all(dev, ctx); |
| 3570 | 3558 | if (ret) { |
| 3571 | 3559 | DRM_ERROR("Suspending crtc's failed with %i\n", ret); |
| 3572 | − goto err; | |
| 3560 | + drm_atomic_state_put(state); | |
| 3561 | + return; | |
| 3573 | 3562 | } |
| 3574 | 3563 | |
| 3575 | 3564 | dev_priv->modeset_restore_state = state; |
| 3576 | 3565 | state->acquire_ctx = ctx; |
| 3577 | − return; | |
| 3578 | − | |
| 3579 | −err: | |
| 3580 | − drm_atomic_state_put(state); | |
| 3581 | 3566 | } |
| 3582 | 3567 | |
| 3583 | 3568 | void intel_finish_reset(struct drm_i915_private *dev_priv) |
| @@ -6886,13 +6871,13 @@ static void intel_crtc_disable_noatomic(struct drm_crtc *crtc) · 3302 unmodified lines | ||
| 6886 | 6871 | if (!intel_crtc->active) |
| 6887 | 6872 | return; |
| 6888 | 6873 | |
| 6889 | − if (to_intel_plane_state(crtc->primary->state)->base.visible) { | |
| 6874 | + if (crtc->primary->state->visible) { | |
| 6890 | 6875 | WARN_ON(intel_crtc->flip_work); |
| 6891 | 6876 | |
| 6892 | 6877 | intel_pre_disable_primary_noatomic(crtc); |
| 6893 | 6878 | |
| 6894 | 6879 | intel_crtc_disable_planes(crtc, 1 << drm_plane_index(crtc->primary)); |
| 6895 | − to_intel_plane_state(crtc->primary->state)->base.visible = false; | |
| 6880 | + crtc->primary->state->visible = false; | |
| 6896 | 6881 | } |
| 6897 | 6882 | |
| 6898 | 6883 | state = drm_atomic_state_alloc(crtc->dev); |
| @@ -8406,7 +8391,6 @@ void intel_mode_from_pipe_config(struct drm_display_mode *mode, · 1507 unmodified lines | ||
| 8406 | 8391 | mode->type = DRM_MODE_TYPE_DRIVER; |
| 8407 | 8392 | |
| 8408 | 8393 | mode->clock = pipe_config->base.adjusted_mode.crtc_clock; |
| 8409 | − mode->flags |= pipe_config->base.adjusted_mode.flags; | |
| 8410 | 8394 | |
| 8411 | 8395 | mode->hsync = drm_mode_hsync(mode); |
| 8412 | 8396 | mode->vrefresh = drm_mode_vrefresh(mode); |
| @@ -11291,6 +11275,7 @@ found: · 2878 unmodified lines | ||
| 11291 | 11275 | } |
| 11292 | 11276 | |
| 11293 | 11277 | old->restore_state = restore_state; |
| 11278 | + drm_atomic_state_put(state); | |
| 11294 | 11279 | |
| 11295 | 11280 | /* let the connector get through one full cycle before testing */ |
| 11296 | 11281 | intel_wait_for_vblank(dev_priv, intel_crtc->pipe); |
| @@ -11570,7 +11555,7 @@ static void intel_unpin_work_fn(struct work_struct *__work) · 273 unmodified lines | ||
| 11570 | 11555 | flush_work(&work->mmio_work); |
| 11571 | 11556 | |
| 11572 | 11557 | mutex_lock(&dev->struct_mutex); |
| 11573 | − intel_unpin_fb_obj(work->old_fb, primary->state->rotation); | |
| 11558 | + intel_unpin_fb_vma(work->old_vma); | |
| 11574 | 11559 | i915_gem_object_put(work->pending_flip_obj); |
| 11575 | 11560 | mutex_unlock(&dev->struct_mutex); |
| 11576 | 11561 | |
| @@ -12280,8 +12265,10 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc, · 703 unmodified lines | ||
| 12280 | 12265 | goto cleanup_pending; |
| 12281 | 12266 | } |
| 12282 | 12267 | |
| 12283 | − work->gtt_offset = intel_fb_gtt_offset(fb, primary->state->rotation); | |
| 12284 | − work->gtt_offset += intel_crtc->dspaddr_offset; | |
| 12268 | + work->old_vma = to_intel_plane_state(primary->state)->vma; | |
| 12269 | + to_intel_plane_state(primary->state)->vma = vma; | |
| 12270 | + | |
| 12271 | + work->gtt_offset = i915_ggtt_offset(vma) + intel_crtc->dspaddr_offset; | |
| 12285 | 12272 | work->rotation = crtc->primary->state->rotation; |
| 12286 | 12273 | |
| 12287 | 12274 | /* |
| @@ -12336,7 +12323,8 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc, · 48 unmodified lines | ||
| 12336 | 12323 | cleanup_request: |
| 12337 | 12324 | i915_add_request_no_flush(request); |
| 12338 | 12325 | cleanup_unpin: |
| 12339 | − intel_unpin_fb_obj(fb, crtc->primary->state->rotation); | |
| 12326 | + to_intel_plane_state(primary->state)->vma = work->old_vma; | |
| 12327 | + intel_unpin_fb_vma(vma); | |
| 12340 | 12328 | cleanup_pending: |
| 12341 | 12329 | atomic_dec(&intel_crtc->unpin_work_count); |
| 12342 | 12330 | unlock: |
| @@ -12466,7 +12454,7 @@ int intel_plane_atomic_calc_changes(struct drm_crtc_state *crtc_state, · 123 unmodified lines | ||
| 12466 | 12454 | } |
| 12467 | 12455 | |
| 12468 | 12456 | was_visible = old_plane_state->base.visible; |
| 12469 | − visible = to_intel_plane_state(plane_state)->base.visible; | |
| 12457 | + visible = plane_state->visible; | |
| 12470 | 12458 | |
| 12471 | 12459 | if (!was_crtc_enabled && WARN_ON(was_visible)) |
| 12472 | 12460 | was_visible = false; |
| @@ -12482,7 +12470,7 @@ int intel_plane_atomic_calc_changes(struct drm_crtc_state *crtc_state, · 9 unmodified lines | ||
| 12482 | 12470 | * only combine the results from all planes in the current place? |
| 12483 | 12471 | */ |
| 12484 | 12472 | if (!is_crtc_enabled) |
| 12485 | − to_intel_plane_state(plane_state)->base.visible = visible = false; | |
| 12473 | + plane_state->visible = visible = false; | |
| 12486 | 12474 | |
| 12487 | 12475 | if (!was_visible && !visible) |
| 12488 | 12476 | return 0; |
| @@ -14740,6 +14728,7 @@ static const struct drm_crtc_funcs intel_crtc_funcs = { · 2251 unmodified lines | ||
| 14740 | 14728 | .page_flip = intel_crtc_page_flip, |
| 14741 | 14729 | .atomic_duplicate_state = intel_crtc_duplicate_state, |
| 14742 | 14730 | .atomic_destroy_state = intel_crtc_destroy_state, |
| 14731 | + .set_crc_source = intel_crtc_set_crc_source, | |
| 14743 | 14732 | }; |
| 14744 | 14733 | |
| 14745 | 14734 | /** |
| @@ -14836,6 +14825,8 @@ intel_prepare_plane_fb(struct drm_plane *plane, · 90 unmodified lines | ||
| 14836 | 14825 | DRM_DEBUG_KMS("failed to pin object\n"); |
| 14837 | 14826 | return PTR_ERR(vma); |
| 14838 | 14827 | } |
| 14828 | + | |
| 14829 | + to_intel_plane_state(new_state)->vma = vma; | |
| 14839 | 14830 | } |
| 14840 | 14831 | |
| 14841 | 14832 | return 0; |
| @@ -14854,19 +14845,12 @@ void · 12 unmodified lines | ||
| 14854 | 14845 | intel_cleanup_plane_fb(struct drm_plane *plane, |
| 14855 | 14846 | struct drm_plane_state *old_state) |
| 14856 | 14847 | { |
| 14857 | − struct drm_i915_private *dev_priv = to_i915(plane->dev); | |
| 14858 | − struct intel_plane_state *old_intel_state; | |
| 14859 | − struct drm_i915_gem_object *old_obj = intel_fb_obj(old_state->fb); | |
| 14860 | − struct drm_i915_gem_object *obj = intel_fb_obj(plane->state->fb); | |
| 14861 | − | |
| 14862 | − old_intel_state = to_intel_plane_state(old_state); | |
| 14863 | − | |
| 14864 | − if (!obj && !old_obj) | |
| 14865 | − return; | |
| 14848 | + struct i915_vma *vma; | |
| 14866 | 14849 | |
| 14867 | − if (old_obj && (plane->type != DRM_PLANE_TYPE_CURSOR || | |
| 14868 | − !INTEL_INFO(dev_priv)->cursor_needs_physical)) | |
| 14869 | − intel_unpin_fb_obj(old_state->fb, old_state->rotation); | |
| 14850 | + /* Should only be called after a successful intel_prepare_plane_fb()! */ | |
| 14851 | + vma = fetch_and_zero(&to_intel_plane_state(old_state)->vma); | |
| 14852 | + if (vma) | |
| 14853 | + intel_unpin_fb_vma(vma); | |
| 14870 | 14854 | } |
| 14871 | 14855 | |
| 14872 | 14856 | int |
| @@ -15018,6 +15002,7 @@ intel_legacy_cursor_update(struct drm_plane *plane, · 145 unmodified lines | ||
| 15018 | 15002 | struct intel_plane *intel_plane = to_intel_plane(plane); |
| 15019 | 15003 | struct drm_framebuffer *old_fb; |
| 15020 | 15004 | struct drm_crtc_state *crtc_state = crtc->state; |
| 15005 | + struct i915_vma *old_vma; | |
| 15021 | 15006 | |
| 15022 | 15007 | /* |
| 15023 | 15008 | * When crtc is inactive or there is a modeset pending, |
| @@ -15089,9 +15074,12 @@ intel_legacy_cursor_update(struct drm_plane *plane, · 65 unmodified lines | ||
| 15089 | 15074 | ret = PTR_ERR(vma); |
| 15090 | 15075 | goto out_unlock; |
| 15091 | 15076 | } |
| 15077 | + | |
| 15078 | + to_intel_plane_state(new_plane_state)->vma = vma; | |
| 15092 | 15079 | } |
| 15093 | 15080 | |
| 15094 | 15081 | old_fb = old_plane_state->fb; |
| 15082 | + old_vma = to_intel_plane_state(old_plane_state)->vma; | |
| 15095 | 15083 | |
| 15096 | 15084 | i915_gem_track_fb(intel_fb_obj(old_fb), intel_fb_obj(fb), |
| 15097 | 15085 | intel_plane->frontbuffer_bit); |
| @@ -15101,6 +15089,7 @@ intel_legacy_cursor_update(struct drm_plane *plane, · 3 unmodified lines | ||
| 15101 | 15089 | *to_intel_plane_state(old_plane_state) = *to_intel_plane_state(new_plane_state); |
| 15102 | 15090 | new_plane_state->fence = NULL; |
| 15103 | 15091 | new_plane_state->fb = old_fb; |
| 15092 | + to_intel_plane_state(new_plane_state)->vma = old_vma; | |
| 15104 | 15093 | |
| 15105 | 15094 | intel_plane->update_plane(plane, |
| 15106 | 15095 | to_intel_crtc_state(crtc->state), |
| @@ -15339,7 +15328,7 @@ intel_update_cursor_plane(struct drm_plane *plane, · 232 unmodified lines | ||
| 15339 | 15328 | if (!obj) |
| 15340 | 15329 | addr = 0; |
| 15341 | 15330 | else if (!INTEL_INFO(dev_priv)->cursor_needs_physical) |
| 15342 | − addr = i915_gem_object_ggtt_offset(obj, NULL); | |
| 15331 | + addr = intel_plane_ggtt_offset(state); | |
| 15343 | 15332 | else |
| 15344 | 15333 | addr = obj->phys_handle->busaddr; |
| 15345 | 15334 | |
| @@ -16843,7 +16832,7 @@ static void intel_sanitize_crtc(struct intel_crtc *crtc) · 1497 unmodified lines | ||
| 16843 | 16832 | * Temporarily change the plane mapping and disable everything |
| 16844 | 16833 | * ... */ |
| 16845 | 16834 | plane = crtc->plane; |
| 16846 | − to_intel_plane_state(crtc->base.primary->state)->base.visible = true; | |
| 16835 | + crtc->base.primary->state->visible = true; | |
| 16847 | 16836 | crtc->plane = !plane; |
| 16848 | 16837 | intel_crtc_disable_noatomic(&crtc->base); |
| 16849 | 16838 | crtc->plane = plane; |
| @@ -16995,7 +16984,8 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev) · 145 unmodified lines | ||
| 16995 | 16984 | dev_priv->active_crtcs = 0; |
| 16996 | 16985 | |
| 16997 | 16986 | for_each_intel_crtc(dev, crtc) { |
| 16998 | − struct intel_crtc_state *crtc_state = crtc->config; | |
| 16987 | + struct intel_crtc_state *crtc_state = | |
| 16988 | + to_intel_crtc_state(crtc->base.state); | |
| 16999 | 16989 | |
| 17000 | 16990 | __drm_atomic_helper_crtc_destroy_state(&crtc_state->base); |
| 17001 | 16991 | memset(crtc_state, 0, sizeof(*crtc_state)); |
| @@ -17014,7 +17004,7 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev) · 12 unmodified lines | ||
| 17014 | 17004 | |
| 17015 | 17005 | DRM_DEBUG_KMS("[CRTC:%d:%s] hw state readout: %s\n", |
| 17016 | 17006 | crtc->base.base.id, crtc->base.name, |
| 17017 | − enableddisabled(crtc->active)); | |
| 17007 | + enableddisabled(crtc_state->base.active)); | |
| 17018 | 17008 | } |
| 17019 | 17009 | |
| 17020 | 17010 | for (i = 0; i < dev_priv->num_shared_dpll; i++) { |
| @@ -17024,7 +17014,11 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev) · 3 unmodified lines | ||
| 17024 | 17014 | &pll->state.hw_state); |
| 17025 | 17015 | pll->state.crtc_mask = 0; |
| 17026 | 17016 | for_each_intel_crtc(dev, crtc) { |
| 17027 | − if (crtc->active && crtc->config->shared_dpll == pll) | |
| 17017 | + struct intel_crtc_state *crtc_state = | |
| 17018 | + to_intel_crtc_state(crtc->base.state); | |
| 17019 | + | |
| 17020 | + if (crtc_state->base.active && | |
| 17021 | + crtc_state->shared_dpll == pll) | |
| 17028 | 17022 | pll->state.crtc_mask |= 1 << crtc->pipe; |
| 17029 | 17023 | } |
| 17030 | 17024 | pll->active_mask = pll->state.crtc_mask; |
| @@ -17037,11 +17031,14 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev) · 6 unmodified lines | ||
| 17037 | 17031 | pipe = 0; |
| 17038 | 17032 | |
| 17039 | 17033 | if (encoder->get_hw_state(encoder, &pipe)) { |
| 17034 | + struct intel_crtc_state *crtc_state; | |
| 17035 | + | |
| 17040 | 17036 | crtc = intel_get_crtc_for_pipe(dev_priv, pipe); |
| 17037 | + crtc_state = to_intel_crtc_state(crtc->base.state); | |
| 17041 | 17038 | |
| 17042 | 17039 | encoder->base.crtc = &crtc->base; |
| 17043 | − crtc->config->output_types |= 1 << encoder->type; | |
| 17044 | − encoder->get_config(encoder, crtc->config); | |
| 17040 | + crtc_state->output_types |= 1 << encoder->type; | |
| 17041 | + encoder->get_config(encoder, crtc_state); | |
| 17045 | 17042 | } else { |
| 17046 | 17043 | encoder->base.crtc = NULL; |
| 17047 | 17044 | } |
| @@ -17082,14 +17079,16 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev) · 34 unmodified lines | ||
| 17082 | 17079 | } |
| 17083 | 17080 | |
| 17084 | 17081 | for_each_intel_crtc(dev, crtc) { |
| 17082 | + struct intel_crtc_state *crtc_state = | |
| 17083 | + to_intel_crtc_state(crtc->base.state); | |
| 17085 | 17084 | int pixclk = 0; |
| 17086 | 17085 | |
| 17087 | − crtc->base.hwmode = crtc->config->base.adjusted_mode; | |
| 17086 | + crtc->base.hwmode = crtc_state->base.adjusted_mode; | |
| 17088 | 17087 | |
| 17089 | 17088 | memset(&crtc->base.mode, 0, sizeof(crtc->base.mode)); |
| 17090 | − if (crtc->base.state->active) { | |
| 17091 | − intel_mode_from_pipe_config(&crtc->base.mode, crtc->config); | |
| 17092 | − intel_mode_from_pipe_config(&crtc->base.state->adjusted_mode, crtc->config); | |
| 17089 | + if (crtc_state->base.active) { | |
| 17090 | + intel_mode_from_pipe_config(&crtc->base.mode, crtc_state); | |
| 17091 | + intel_mode_from_pipe_config(&crtc_state->base.adjusted_mode, crtc_state); | |
| 17093 | 17092 | WARN_ON(drm_atomic_set_mode_for_crtc(crtc->base.state, &crtc->base.mode)); |
| 17094 | 17093 | |
| 17095 | 17094 | /* |
| @@ -17101,17 +17100,17 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev) · 5 unmodified lines | ||
| 17101 | 17100 | * set a flag to indicate that a full recalculation is |
| 17102 | 17101 | * needed on the next commit. |
| 17103 | 17102 | */ |
| 17104 | − crtc->base.state->mode.private_flags = I915_MODE_FLAG_INHERITED; | |
| 17103 | + crtc_state->base.mode.private_flags = I915_MODE_FLAG_INHERITED; | |
| 17105 | 17104 | |
| 17106 | 17105 | if (INTEL_GEN(dev_priv) >= 9 || IS_BROADWELL(dev_priv)) |
| 17107 | − pixclk = ilk_pipe_pixel_rate(crtc->config); | |
| 17106 | + pixclk = ilk_pipe_pixel_rate(crtc_state); | |
| 17108 | 17107 | else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) |
| 17109 | − pixclk = crtc->config->base.adjusted_mode.crtc_clock; | |
| 17108 | + pixclk = crtc_state->base.adjusted_mode.crtc_clock; | |
| 17110 | 17109 | else |
| 17111 | 17110 | WARN_ON(dev_priv->display.modeset_calc_cdclk); |
| 17112 | 17111 | |
| 17113 | 17112 | /* pixel rate mustn't exceed 95% of cdclk with IPS on BDW */ |
| 17114 | − if (IS_BROADWELL(dev_priv) && crtc->config->ips_enabled) | |
| 17113 | + if (IS_BROADWELL(dev_priv) && crtc_state->ips_enabled) | |
| 17115 | 17114 | pixclk = DIV_ROUND_UP(pixclk * 100, 95); |
| 17116 | 17115 | |
| 17117 | 17116 | drm_calc_timestamping_constants(&crtc->base, &crtc->base.hwmode); |
| @@ -17120,7 +17119,7 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev) · 2 unmodified lines | ||
| 17120 | 17119 | |
| 17121 | 17120 | dev_priv->min_pixclk[crtc->pipe] = pixclk; |
| 17122 | 17121 | |
| 17123 | − intel_pipe_config_sanity_check(dev_priv, crtc->config); | |
| 17122 | + intel_pipe_config_sanity_check(dev_priv, crtc_state); | |
| 17124 | 17123 | } |
| 17125 | 17124 | } |
| 17126 | 17125 | |
| @@ -17221,47 +17220,19 @@ void intel_display_resume(struct drm_device *dev) · 94 unmodified lines | ||
| 17221 | 17220 | |
| 17222 | 17221 | if (ret) |
| 17223 | 17222 | DRM_ERROR("Restoring old state failed with %i\n", ret); |
| 17224 | − drm_atomic_state_put(state); | |
| 17223 | + if (state) | |
| 17224 | + drm_atomic_state_put(state); | |
| 17225 | 17225 | } |
| 17226 | 17226 | |
| 17227 | 17227 | void intel_modeset_gem_init(struct drm_device *dev) |
| 17228 | 17228 | { |
| 17229 | 17229 | struct drm_i915_private *dev_priv = to_i915(dev); |
| 17230 | − struct drm_crtc *c; | |
| 17231 | − struct drm_i915_gem_object *obj; | |
| 17232 | 17230 | |
| 17233 | 17231 | intel_init_gt_powersave(dev_priv); |
| 17234 | 17232 | |
| 17235 | 17233 | intel_modeset_init_hw(dev); |
| 17236 | 17234 | |
| 17237 | 17235 | intel_setup_overlay(dev_priv); |
| 17238 | − | |
| 17239 | − /* | |
| 17240 | − * Make sure any fbs we allocated at startup are properly | |
| 17241 | − * pinned & fenced. When we do the allocation it's too early | |
| 17242 | − * for this. | |
| 17243 | − */ | |
| 17244 | − for_each_crtc(dev, c) { | |
| 17245 | − struct i915_vma *vma; | |
| 17246 | − | |
| 17247 | − obj = intel_fb_obj(c->primary->fb); | |
| 17248 | − if (obj == NULL) | |
| 17249 | − continue; | |
| 17250 | − | |
| 17251 | − mutex_lock(&dev->struct_mutex); | |
| 17252 | − vma = intel_pin_and_fence_fb_obj(c->primary->fb, | |
| 17253 | − c->primary->state->rotation); | |
| 17254 | − mutex_unlock(&dev->struct_mutex); | |
| 17255 | − if (IS_ERR(vma)) { | |
| 17256 | − DRM_ERROR("failed to pin boot fb on pipe %d\n", | |
| 17257 | − to_intel_crtc(c)->pipe); | |
| 17258 | − drm_framebuffer_unreference(c->primary->fb); | |
| 17259 | − c->primary->fb = NULL; | |
| 17260 | − c->primary->crtc = c->primary->state->crtc = NULL; | |
| 17261 | − update_state_fb(c->primary); | |
| 17262 | − c->state->plane_mask &= ~(1 << drm_plane_index(c->primary)); | |
| 17263 | − } | |
| 17264 | − } | |
| 17265 | 17236 | } |
| 17266 | 17237 | |
| 17267 | 17238 | int intel_connector_register(struct drm_connector *connector) |
| @@ -3042,6 +3042,32 @@ intel_dp_get_link_status(struct intel_dp *intel_dp, uint8_t link_status[DP_LINK_ · 3041 unmodified lines | ||
| 3042 | 3042 | DP_LINK_STATUS_SIZE) == DP_LINK_STATUS_SIZE; |
| 3043 | 3043 | } |
| 3044 | 3044 | |
| 3045 | +static bool intel_dp_get_y_cord_status(struct intel_dp *intel_dp) | |
| 3046 | +{ | |
| 3047 | + uint8_t psr_caps = 0; | |
| 3048 | + | |
| 3049 | + drm_dp_dpcd_readb(&intel_dp->aux, DP_PSR_CAPS, &psr_caps); | |
| 3050 | + return psr_caps & DP_PSR2_SU_Y_COORDINATE_REQUIRED; | |
| 3051 | +} | |
| 3052 | + | |
| 3053 | +static bool intel_dp_get_colorimetry_status(struct intel_dp *intel_dp) | |
| 3054 | +{ | |
| 3055 | + uint8_t dprx = 0; | |
| 3056 | + | |
| 3057 | + drm_dp_dpcd_readb(&intel_dp->aux, | |
| 3058 | + DP_DPRX_FEATURE_ENUMERATION_LIST, | |
| 3059 | + &dprx); | |
| 3060 | + return dprx & DP_VSC_SDP_EXT_FOR_COLORIMETRY_SUPPORTED; | |
| 3061 | +} | |
| 3062 | + | |
| 3063 | +static bool intel_dp_get_alpm_status(struct intel_dp *intel_dp) | |
| 3064 | +{ | |
| 3065 | + uint8_t alpm_caps = 0; | |
| 3066 | + | |
| 3067 | + drm_dp_dpcd_readb(&intel_dp->aux, DP_RECEIVER_ALPM_CAP, &alpm_caps); | |
| 3068 | + return alpm_caps & DP_ALPM_CAP; | |
| 3069 | +} | |
| 3070 | + | |
| 3045 | 3071 | /* These are source-specific values. */ |
| 3046 | 3072 | uint8_t |
| 3047 | 3073 | intel_dp_voltage_max(struct intel_dp *intel_dp) |
| @@ -3412,7 +3438,7 @@ intel_dp_set_signal_levels(struct intel_dp *intel_dp) · 364 unmodified lines | ||
| 3412 | 3438 | if (HAS_DDI(dev_priv)) { |
| 3413 | 3439 | signal_levels = ddi_signal_levels(intel_dp); |
| 3414 | 3440 | |
| 3415 | − if (IS_BROXTON(dev_priv)) | |
| 3441 | + if (IS_GEN9_LP(dev_priv)) | |
| 3416 | 3442 | signal_levels = 0; |
| 3417 | 3443 | else |
| 3418 | 3444 | mask = DDI_BUF_EMP_MASK; |
| @@ -3620,6 +3646,16 @@ intel_edp_init_dpcd(struct intel_dp *intel_dp) · 201 unmodified lines | ||
| 3620 | 3646 | dev_priv->psr.psr2_support = dev_priv->psr.aux_frame_sync; |
| 3621 | 3647 | DRM_DEBUG_KMS("PSR2 %s on sink", |
| 3622 | 3648 | dev_priv->psr.psr2_support ? "supported" : "not supported"); |
| 3649 | + | |
| 3650 | + if (dev_priv->psr.psr2_support) { | |
| 3651 | + dev_priv->psr.y_cord_support = | |
| 3652 | + intel_dp_get_y_cord_status(intel_dp); | |
| 3653 | + dev_priv->psr.colorimetry_support = | |
| 3654 | + intel_dp_get_colorimetry_status(intel_dp); | |
| 3655 | + dev_priv->psr.alpm = | |
| 3656 | + intel_dp_get_alpm_status(intel_dp); | |
| 3657 | + } | |
| 3658 | + | |
| 3623 | 3659 | } |
| 3624 | 3660 | |
| 3625 | 3661 | /* Read the eDP Display control capabilities registers */ |
| @@ -376,6 +376,7 @@ struct intel_atomic_state { · 375 unmodified lines | ||
| 376 | 376 | struct intel_plane_state { |
| 377 | 377 | struct drm_plane_state base; |
| 378 | 378 | struct drm_rect clip; |
| 379 | + struct i915_vma *vma; | |
| 379 | 380 | |
| 380 | 381 | struct { |
| 381 | 382 | u32 offset; |
| @@ -1067,6 +1068,7 @@ struct intel_flip_work { · 685 unmodified lines | ||
| 1067 | 1068 | struct work_struct mmio_work; |
| 1068 | 1069 | |
| 1069 | 1070 | struct drm_crtc *crtc; |
| 1071 | + struct i915_vma *old_vma; | |
| 1070 | 1072 | struct drm_framebuffer *old_fb; |
| 1071 | 1073 | struct drm_i915_gem_object *pending_flip_obj; |
| 1072 | 1074 | struct drm_pending_vblank_event *event; |
| @@ -1302,7 +1304,7 @@ void intel_release_load_detect_pipe(struct drm_connector *connector, · 229 unmodified lines | ||
| 1302 | 1304 | struct drm_modeset_acquire_ctx *ctx); |
| 1303 | 1305 | struct i915_vma * |
| 1304 | 1306 | intel_pin_and_fence_fb_obj(struct drm_framebuffer *fb, unsigned int rotation); |
| 1305 | −void intel_unpin_fb_obj(struct drm_framebuffer *fb, unsigned int rotation); | |
| 1307 | +void intel_unpin_fb_vma(struct i915_vma *vma); | |
| 1306 | 1308 | struct drm_framebuffer * |
| 1307 | 1309 | __intel_framebuffer_create(struct drm_device *dev, |
| 1308 | 1310 | struct drm_mode_fb_cmd2 *mode_cmd, |
| @@ -1391,7 +1393,10 @@ void intel_mode_from_pipe_config(struct drm_display_mode *mode, · 82 unmodified lines | ||
| 1391 | 1393 | int skl_update_scaler_crtc(struct intel_crtc_state *crtc_state); |
| 1392 | 1394 | int skl_max_scale(struct intel_crtc *crtc, struct intel_crtc_state *crtc_state); |
| 1393 | 1395 | |
| 1394 | −u32 intel_fb_gtt_offset(struct drm_framebuffer *fb, unsigned int rotation); | |
| 1396 | +static inline u32 intel_plane_ggtt_offset(const struct intel_plane_state *state) | |
| 1397 | +{ | |
| 1398 | + return i915_ggtt_offset(state->vma); | |
| 1399 | +} | |
| 1395 | 1400 | |
| 1396 | 1401 | u32 skl_plane_ctl_format(uint32_t pixel_format); |
| 1397 | 1402 | u32 skl_plane_ctl_tiling(uint64_t fb_modifier); |
| @@ -1880,5 +1885,11 @@ void lspcon_wait_pcon_mode(struct intel_lspcon *lspcon); · 482 unmodified lines | ||
| 1880 | 1885 | /* intel_pipe_crc.c */ |
| 1881 | 1886 | int intel_pipe_crc_create(struct drm_minor *minor); |
| 1882 | 1887 | void intel_pipe_crc_cleanup(struct drm_minor *minor); |
| 1888 | +#ifdef CONFIG_DEBUG_FS | |
| 1889 | +int intel_crtc_set_crc_source(struct drm_crtc *crtc, const char *source_name, | |
| 1890 | + size_t *values_cnt); | |
| 1891 | +#else | |
| 1892 | +#define intel_crtc_set_crc_source NULL | |
| 1893 | +#endif | |
| 1883 | 1894 | extern const struct file_operations i915_display_crc_ctl_fops; |
| 1884 | 1895 | #endif /* __INTEL_DRV_H__ */ |
| @@ -264,7 +264,7 @@ int intel_engine_create_scratch(struct intel_engine_cs *engine, int size) · 263 unmodified lines | ||
| 264 | 264 | return PTR_ERR(obj); |
| 265 | 265 | } |
| 266 | 266 | |
| 267 | − vma = i915_vma_create(obj, &engine->i915->ggtt.base, NULL); | |
| 267 | + vma = i915_vma_instance(obj, &engine->i915->ggtt.base, NULL); | |
| 268 | 268 | if (IS_ERR(vma)) { |
| 269 | 269 | ret = PTR_ERR(vma); |
| 270 | 270 | goto err_unref; |
| @@ -173,7 +173,7 @@ static void i8xx_fbc_activate(struct drm_i915_private *dev_priv) · 172 unmodified lines | ||
| 173 | 173 | if (IS_I945GM(dev_priv)) |
| 174 | 174 | fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */ |
| 175 | 175 | fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT; |
| 176 | − fbc_ctl |= params->fb.fence_reg; | |
| 176 | + fbc_ctl |= params->vma->fence->id; | |
| 177 | 177 | I915_WRITE(FBC_CONTROL, fbc_ctl); |
| 178 | 178 | } |
| 179 | 179 | |
| @@ -193,8 +193,8 @@ static void g4x_fbc_activate(struct drm_i915_private *dev_priv) · 13 unmodified lines | ||
| 193 | 193 | else |
| 194 | 194 | dpfc_ctl |= DPFC_CTL_LIMIT_1X; |
| 195 | 195 | |
| 196 | − if (params->fb.fence_reg != I915_FENCE_REG_NONE) { | |
| 197 | − dpfc_ctl |= DPFC_CTL_FENCE_EN | params->fb.fence_reg; | |
| 196 | + if (params->vma->fence) { | |
| 197 | + dpfc_ctl |= DPFC_CTL_FENCE_EN | params->vma->fence->id; | |
| 198 | 198 | I915_WRITE(DPFC_FENCE_YOFF, params->crtc.fence_y_offset); |
| 199 | 199 | } else { |
| 200 | 200 | I915_WRITE(DPFC_FENCE_YOFF, 0); |
| @@ -251,13 +251,14 @@ static void ilk_fbc_activate(struct drm_i915_private *dev_priv) · 50 unmodified lines | ||
| 251 | 251 | break; |
| 252 | 252 | } |
| 253 | 253 | |
| 254 | − if (params->fb.fence_reg != I915_FENCE_REG_NONE) { | |
| 254 | + if (params->vma->fence) { | |
| 255 | 255 | dpfc_ctl |= DPFC_CTL_FENCE_EN; |
| 256 | 256 | if (IS_GEN5(dev_priv)) |
| 257 | − dpfc_ctl |= params->fb.fence_reg; | |
| 257 | + dpfc_ctl |= params->vma->fence->id; | |
| 258 | 258 | if (IS_GEN6(dev_priv)) { |
| 259 | 259 | I915_WRITE(SNB_DPFC_CTL_SA, |
| 260 | − SNB_CPU_FENCE_ENABLE | params->fb.fence_reg); | |
| 260 | + SNB_CPU_FENCE_ENABLE | | |
| 261 | + params->vma->fence->id); | |
| 261 | 262 | I915_WRITE(DPFC_CPU_FENCE_OFFSET, |
| 262 | 263 | params->crtc.fence_y_offset); |
| 263 | 264 | } |
| @@ -269,7 +270,8 @@ static void ilk_fbc_activate(struct drm_i915_private *dev_priv) · 5 unmodified lines | ||
| 269 | 270 | } |
| 270 | 271 | |
| 271 | 272 | I915_WRITE(ILK_DPFC_FENCE_YOFF, params->crtc.fence_y_offset); |
| 272 | − I915_WRITE(ILK_FBC_RT_BASE, params->fb.ggtt_offset | ILK_FBC_RT_VALID); | |
| 273 | + I915_WRITE(ILK_FBC_RT_BASE, | |
| 274 | + i915_ggtt_offset(params->vma) | ILK_FBC_RT_VALID); | |
| 273 | 275 | /* enable it... */ |
| 274 | 276 | I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN); |
| 275 | 277 | |
| @@ -319,10 +321,11 @@ static void gen7_fbc_activate(struct drm_i915_private *dev_priv) · 43 unmodified lines | ||
| 319 | 321 | break; |
| 320 | 322 | } |
| 321 | 323 | |
| 322 | − if (params->fb.fence_reg != I915_FENCE_REG_NONE) { | |
| 324 | + if (params->vma->fence) { | |
| 323 | 325 | dpfc_ctl |= IVB_DPFC_CTL_FENCE_EN; |
| 324 | 326 | I915_WRITE(SNB_DPFC_CTL_SA, |
| 325 | − SNB_CPU_FENCE_ENABLE | params->fb.fence_reg); | |
| 327 | + SNB_CPU_FENCE_ENABLE | | |
| 328 | + params->vma->fence->id); | |
| 326 | 329 | I915_WRITE(DPFC_CPU_FENCE_OFFSET, params->crtc.fence_y_offset); |
| 327 | 330 | } else { |
| 328 | 331 | I915_WRITE(SNB_DPFC_CTL_SA,0); |
| @@ -727,14 +730,6 @@ static bool intel_fbc_hw_tracking_covers_screen(struct intel_crtc *crtc) · 398 unmodified lines | ||
| 727 | 730 | return effective_w <= max_w && effective_h <= max_h; |
| 728 | 731 | } |
| 729 | 732 | |
| 730 | −/* XXX replace me when we have VMA tracking for intel_plane_state */ | |
| 731 | −static int get_fence_id(struct drm_framebuffer *fb) | |
| 732 | −{ | |
| 733 | − struct i915_vma *vma = i915_gem_object_to_ggtt(intel_fb_obj(fb), NULL); | |
| 734 | − | |
| 735 | − return vma && vma->fence ? vma->fence->id : I915_FENCE_REG_NONE; | |
| 736 | −} | |
| 737 | − | |
| 738 | 733 | static void intel_fbc_update_state_cache(struct intel_crtc *crtc, |
| 739 | 734 | struct intel_crtc_state *crtc_state, |
| 740 | 735 | struct intel_plane_state *plane_state) |
| @@ -743,7 +738,8 @@ static void intel_fbc_update_state_cache(struct intel_crtc *crtc, · 2 unmodified lines | ||
| 743 | 738 | struct intel_fbc *fbc = &dev_priv->fbc; |
| 744 | 739 | struct intel_fbc_state_cache *cache = &fbc->state_cache; |
| 745 | 740 | struct drm_framebuffer *fb = plane_state->base.fb; |
| 746 | − struct drm_i915_gem_object *obj; | |
| 741 | + | |
| 742 | + cache->vma = NULL; | |
| 747 | 743 | |
| 748 | 744 | cache->crtc.mode_flags = crtc_state->base.adjusted_mode.flags; |
| 749 | 745 | if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) |
| @@ -758,16 +754,10 @@ static void intel_fbc_update_state_cache(struct intel_crtc *crtc, · 8 unmodified lines | ||
| 758 | 754 | if (!cache->plane.visible) |
| 759 | 755 | return; |
| 760 | 756 | |
| 761 | − obj = intel_fb_obj(fb); | |
| 762 | − | |
| 763 | − /* FIXME: We lack the proper locking here, so only run this on the | |
| 764 | − * platforms that need. */ | |
| 765 | − if (IS_GEN(dev_priv, 5, 6)) | |
| 766 | − cache->fb.ilk_ggtt_offset = i915_gem_object_ggtt_offset(obj, NULL); | |
| 767 | 757 | cache->fb.format = fb->format; |
| 768 | 758 | cache->fb.stride = fb->pitches[0]; |
| 769 | − cache->fb.fence_reg = get_fence_id(fb); | |
| 770 | − cache->fb.tiling_mode = i915_gem_object_get_tiling(obj); | |
| 759 | + | |
| 760 | + cache->vma = plane_state->vma; | |
| 771 | 761 | } |
| 772 | 762 | |
| 773 | 763 | static bool intel_fbc_can_activate(struct intel_crtc *crtc) |
| @@ -784,7 +774,7 @@ static bool intel_fbc_can_activate(struct intel_crtc *crtc) · 10 unmodified lines | ||
| 784 | 774 | return false; |
| 785 | 775 | } |
| 786 | 776 | |
| 787 | − if (!cache->plane.visible) { | |
| 777 | + if (!cache->vma) { | |
| 788 | 778 | fbc->no_fbc_reason = "primary plane not visible"; |
| 789 | 779 | return false; |
| 790 | 780 | } |
| @@ -807,8 +797,7 @@ static bool intel_fbc_can_activate(struct intel_crtc *crtc) · 16 unmodified lines | ||
| 807 | 797 | * so have no fence associated with it) due to aperture constaints |
| 808 | 798 | * at the time of pinning. |
| 809 | 799 | */ |
| 810 | − if (cache->fb.tiling_mode != I915_TILING_X || | |
| 811 | − cache->fb.fence_reg == I915_FENCE_REG_NONE) { | |
| 800 | + if (!cache->vma->fence) { | |
| 812 | 801 | fbc->no_fbc_reason = "framebuffer not tiled or fenced"; |
| 813 | 802 | return false; |
| 814 | 803 | } |
| @@ -888,17 +877,16 @@ static void intel_fbc_get_reg_params(struct intel_crtc *crtc, · 73 unmodified lines | ||
| 888 | 877 | * zero. */ |
| 889 | 878 | memset(params, 0, sizeof(*params)); |
| 890 | 879 | |
| 880 | + params->vma = cache->vma; | |
| 881 | + | |
| 891 | 882 | params->crtc.pipe = crtc->pipe; |
| 892 | 883 | params->crtc.plane = crtc->plane; |
| 893 | 884 | params->crtc.fence_y_offset = get_crtc_fence_y_offset(crtc); |
| 894 | 885 | |
| 895 | 886 | params->fb.format = cache->fb.format; |
| 896 | 887 | params->fb.stride = cache->fb.stride; |
| 897 | − params->fb.fence_reg = cache->fb.fence_reg; | |
| 898 | 888 | |
| 899 | 889 | params->cfb_size = intel_fbc_calculate_cfb_size(dev_priv, cache); |
| 900 | − | |
| 901 | − params->fb.ggtt_offset = cache->fb.ilk_ggtt_offset; | |
| 902 | 890 | } |
| 903 | 891 | |
| 904 | 892 | static bool intel_fbc_reg_params_equal(struct intel_fbc_reg_params *params1, |
| @@ -1296,7 +1284,7 @@ void intel_fbc_init_pipe_state(struct drm_i915_private *dev_priv) · 391 unmodified lines | ||
| 1296 | 1284 | |
| 1297 | 1285 | for_each_intel_crtc(&dev_priv->drm, crtc) |
| 1298 | 1286 | if (intel_crtc_active(crtc) && |
| 1299 | − to_intel_plane_state(crtc->base.primary->state)->base.visible) | |
| 1287 | + crtc->base.primary->state->visible) | |
| 1300 | 1288 | dev_priv->fbc.visible_pipes_mask |= (1 << crtc->pipe); |
| 1301 | 1289 | } |
| 1302 | 1290 | |
| @@ -284,7 +284,7 @@ static int intelfb_create(struct drm_fb_helper *helper, · 283 unmodified lines | ||
| 284 | 284 | out_destroy_fbi: |
| 285 | 285 | drm_fb_helper_release_fbi(helper); |
| 286 | 286 | out_unpin: |
| 287 | − intel_unpin_fb_obj(&ifbdev->fb->base, DRM_ROTATE_0); | |
| 287 | + intel_unpin_fb_vma(vma); | |
| 288 | 288 | out_unlock: |
| 289 | 289 | mutex_unlock(&dev->struct_mutex); |
| 290 | 290 | return ret; |
| @@ -549,7 +549,7 @@ static void intel_fbdev_destroy(struct intel_fbdev *ifbdev) · 258 unmodified lines | ||
| 549 | 549 | |
| 550 | 550 | if (ifbdev->fb) { |
| 551 | 551 | mutex_lock(&ifbdev->helper.dev->struct_mutex); |
| 552 | − intel_unpin_fb_obj(&ifbdev->fb->base, DRM_ROTATE_0); | |
| 552 | + intel_unpin_fb_vma(ifbdev->vma); | |
| 553 | 553 | mutex_unlock(&ifbdev->helper.dev->struct_mutex); |
| 554 | 554 | |
| 555 | 555 | drm_framebuffer_remove(&ifbdev->fb->base); |
| @@ -742,6 +742,9 @@ void intel_fbdev_initial_config_async(struct drm_device *dev) · 186 unmodified lines | ||
| 742 | 742 | { |
| 743 | 743 | struct intel_fbdev *ifbdev = to_i915(dev)->fbdev; |
| 744 | 744 | |
| 745 | + if (!ifbdev) | |
| 746 | + return; | |
| 747 | + | |
| 745 | 748 | ifbdev->cookie = async_schedule(intel_fbdev_initial_config, ifbdev); |
| 746 | 749 | } |
| 747 | 750 | |
| @@ -145,7 +145,7 @@ · 144 unmodified lines | ||
| 145 | 145 | * The GuC firmware layout looks like this: |
| 146 | 146 | * |
| 147 | 147 | * +-------------------------------+ |
| 148 | − * | guc_css_header | | |
| 148 | + * | uc_css_header | | |
| 149 | 149 | * | | |
| 150 | 150 | * | contains major/minor version | |
| 151 | 151 | * +-------------------------------+ |
| @@ -172,9 +172,16 @@ · 20 unmodified lines | ||
| 172 | 172 | * 3. Length info of each component can be found in header, in dwords. |
| 173 | 173 | * 4. Modulus and exponent key are not required by driver. They may not appear |
| 174 | 174 | * in fw. So driver will load a truncated firmware in this case. |
| 175 | + * | |
| 176 | + * HuC firmware layout is same as GuC firmware. | |
| 177 | + * | |
| 178 | + * HuC firmware css header is different. However, the only difference is where | |
| 179 | + * the version information is saved. The uc_css_header is unified to support | |
| 180 | + * both. Driver should get HuC version from uc_css_header.huc_sw_version, while | |
| 181 | + * uc_css_header.guc_sw_version for GuC. | |
| 175 | 182 | */ |
| 176 | 183 | |
| 177 | −struct guc_css_header { | |
| 184 | +struct uc_css_header { | |
| 178 | 185 | uint32_t module_type; |
| 179 | 186 | /* header_size includes all non-uCode bits, including css_header, rsa |
| 180 | 187 | * key, modulus key and exponent data. */ |
| @@ -205,8 +212,16 @@ struct guc_css_header { · 24 unmodified lines | ||
| 205 | 212 | |
| 206 | 213 | char username[8]; |
| 207 | 214 | char buildnumber[12]; |
| 208 | − uint32_t device_id; | |
| 209 | − uint32_t guc_sw_version; | |
| 215 | + union { | |
| 216 | + struct { | |
| 217 | + uint32_t branch_client_version; | |
| 218 | + uint32_t sw_version; | |
| 219 | + } guc; | |
| 220 | + struct { | |
| 221 | + uint32_t sw_version; | |
| 222 | + uint32_t reserved; | |
| 223 | + } huc; | |
| 224 | + }; | |
| 210 | 225 | uint32_t prod_preprod_fw; |
| 211 | 226 | uint32_t reserved[12]; |
| 212 | 227 | uint32_t header_info; |
| @@ -490,6 +505,7 @@ enum intel_guc_action { · 277 unmodified lines | ||
| 490 | 505 | INTEL_GUC_ACTION_ENTER_S_STATE = 0x501, |
| 491 | 506 | INTEL_GUC_ACTION_EXIT_S_STATE = 0x502, |
| 492 | 507 | INTEL_GUC_ACTION_SLPC_REQUEST = 0x3003, |
| 508 | + INTEL_GUC_ACTION_AUTHENTICATE_HUC = 0x4000, | |
| 493 | 509 | INTEL_GUC_ACTION_UK_LOG_ENABLE_LOGGING = 0x0E000, |
| 494 | 510 | INTEL_GUC_ACTION_LIMIT |
| 495 | 511 | }; |
| @@ -51,12 +51,6 @@ · 50 unmodified lines | ||
| 51 | 51 | * 512K. In order to exclude 0-512K address space from GGTT, all gfx objects |
| 52 | 52 | * used by GuC is pinned with PIN_OFFSET_BIAS along with size of WOPCM. |
| 53 | 53 | * |
| 54 | − * Firmware log: | |
| 55 | − * Firmware log is enabled by setting i915.guc_log_level to non-negative level. | |
| 56 | − * Log data is printed out via reading debugfs i915_guc_log_dump. Reading from | |
| 57 | − * i915_guc_load_status will print out firmware loading status and scratch | |
| 58 | − * registers value. | |
| 59 | − * | |
| 60 | 54 | */ |
| 61 | 55 | |
| 62 | 56 | #define SKL_FW_MAJOR 6 |
| @@ -81,16 +75,16 @@ MODULE_FIRMWARE(I915_BXT_GUC_UCODE); · 18 unmodified lines | ||
| 81 | 75 | MODULE_FIRMWARE(I915_KBL_GUC_UCODE); |
| 82 | 76 | |
| 83 | 77 | /* User-friendly representation of an enum */ |
| 84 | −const char *intel_guc_fw_status_repr(enum intel_guc_fw_status status) | |
| 78 | +const char *intel_uc_fw_status_repr(enum intel_uc_fw_status status) | |
| 85 | 79 | { |
| 86 | 80 | switch (status) { |
| 87 | − case GUC_FIRMWARE_FAIL: | |
| 81 | + case INTEL_UC_FIRMWARE_FAIL: | |
| 88 | 82 | return "FAIL"; |
| 89 | − case GUC_FIRMWARE_NONE: | |
| 83 | + case INTEL_UC_FIRMWARE_NONE: | |
| 90 | 84 | return "NONE"; |
| 91 | − case GUC_FIRMWARE_PENDING: | |
| 85 | + case INTEL_UC_FIRMWARE_PENDING: | |
| 92 | 86 | return "PENDING"; |
| 93 | − case GUC_FIRMWARE_SUCCESS: | |
| 87 | + case INTEL_UC_FIRMWARE_SUCCESS: | |
| 94 | 88 | return "SUCCESS"; |
| 95 | 89 | default: |
| 96 | 90 | return "UNKNOWN!"; |
| @@ -278,7 +272,7 @@ static inline bool guc_ucode_response(struct drm_i915_private *dev_priv, · 181 unmodified lines | ||
| 278 | 272 | static int guc_ucode_xfer_dma(struct drm_i915_private *dev_priv, |
| 279 | 273 | struct i915_vma *vma) |
| 280 | 274 | { |
| 281 | − struct intel_guc_fw *guc_fw = &dev_priv->guc.guc_fw; | |
| 275 | + struct intel_uc_fw *guc_fw = &dev_priv->guc.fw; | |
| 282 | 276 | unsigned long offset; |
| 283 | 277 | struct sg_table *sg = vma->pages; |
| 284 | 278 | u32 status, rsa[UOS_RSA_SCRATCH_MAX_COUNT]; |
| @@ -334,12 +328,12 @@ static int guc_ucode_xfer_dma(struct drm_i915_private *dev_priv, · 49 unmodified lines | ||
| 334 | 328 | return ret; |
| 335 | 329 | } |
| 336 | 330 | |
| 337 | −static u32 guc_wopcm_size(struct drm_i915_private *dev_priv) | |
| 331 | +u32 intel_guc_wopcm_size(struct drm_i915_private *dev_priv) | |
| 338 | 332 | { |
| 339 | 333 | u32 wopcm_size = GUC_WOPCM_TOP; |
| 340 | 334 | |
| 341 | 335 | /* On BXT, the top of WOPCM is reserved for RC6 context */ |
| 342 | − if (IS_BROXTON(dev_priv)) | |
| 336 | + if (IS_GEN9_LP(dev_priv)) | |
| 343 | 337 | wopcm_size -= BXT_GUC_WOPCM_RC6_RESERVED; |
| 344 | 338 | |
| 345 | 339 | return wopcm_size; |
| @@ -350,29 +344,27 @@ static u32 guc_wopcm_size(struct drm_i915_private *dev_priv) · 4 unmodified lines | ||
| 350 | 344 | */ |
| 351 | 345 | static int guc_ucode_xfer(struct drm_i915_private *dev_priv) |
| 352 | 346 | { |
| 353 | − struct intel_guc_fw *guc_fw = &dev_priv->guc.guc_fw; | |
| 347 | + struct intel_uc_fw *guc_fw = &dev_priv->guc.fw; | |
| 354 | 348 | struct i915_vma *vma; |
| 355 | 349 | int ret; |
| 356 | 350 | |
| 357 | − ret = i915_gem_object_set_to_gtt_domain(guc_fw->guc_fw_obj, false); | |
| 351 | + ret = i915_gem_object_set_to_gtt_domain(guc_fw->obj, false); | |
| 358 | 352 | if (ret) { |
| 359 | 353 | DRM_DEBUG_DRIVER("set-domain failed %d\n", ret); |
| 360 | 354 | return ret; |
| 361 | 355 | } |
| 362 | 356 | |
| 363 | − vma = i915_gem_object_ggtt_pin(guc_fw->guc_fw_obj, NULL, 0, 0, 0); | |
| 357 | + vma = i915_gem_object_ggtt_pin(guc_fw->obj, NULL, 0, 0, | |
| 358 | + PIN_OFFSET_BIAS | GUC_WOPCM_TOP); | |
| 364 | 359 | if (IS_ERR(vma)) { |
| 365 | 360 | DRM_DEBUG_DRIVER("pin failed %d\n", (int)PTR_ERR(vma)); |
| 366 | 361 | return PTR_ERR(vma); |
| 367 | 362 | } |
| 368 | 363 | |
| 369 | − /* Invalidate GuC TLB to let GuC take the latest updates to GTT. */ | |
| 370 | − I915_WRITE(GEN8_GTCR, GEN8_GTCR_INVALIDATE); | |
| 371 | − | |
| 372 | 364 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
| 373 | 365 | |
| 374 | 366 | /* init WOPCM */ |
| 375 | − I915_WRITE(GUC_WOPCM_SIZE, guc_wopcm_size(dev_priv)); | |
| 367 | + I915_WRITE(GUC_WOPCM_SIZE, intel_guc_wopcm_size(dev_priv)); | |
| 376 | 368 | I915_WRITE(DMA_GUC_WOPCM_OFFSET, GUC_WOPCM_OFFSET_VALUE); |
| 377 | 369 | |
| 378 | 370 | /* Enable MIA caching. GuC clock gating is disabled. */ |
| @@ -388,7 +380,7 @@ static int guc_ucode_xfer(struct drm_i915_private *dev_priv) · 9 unmodified lines | ||
| 388 | 380 | if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_B0)) |
| 389 | 381 | I915_WRITE(GEN6_GFXPAUSE, 0x30FFF); |
| 390 | 382 | |
| 391 | − if (IS_BROXTON(dev_priv)) | |
| 383 | + if (IS_GEN9_LP(dev_priv)) | |
| 392 | 384 | I915_WRITE(GEN9LP_GT_PM_CONFIG, GT_DOORBELL_ENABLE); |
| 393 | 385 | else |
| 394 | 386 | I915_WRITE(GEN9_GT_PM_CONFIG, GT_DOORBELL_ENABLE); |
| @@ -450,14 +442,14 @@ static int guc_hw_reset(struct drm_i915_private *dev_priv) · 55 unmodified lines | ||
| 450 | 442 | */ |
| 451 | 443 | int intel_guc_setup(struct drm_i915_private *dev_priv) |
| 452 | 444 | { |
| 453 | − struct intel_guc_fw *guc_fw = &dev_priv->guc.guc_fw; | |
| 454 | − const char *fw_path = guc_fw->guc_fw_path; | |
| 445 | + struct intel_uc_fw *guc_fw = &dev_priv->guc.fw; | |
| 446 | + const char *fw_path = guc_fw->path; | |
| 455 | 447 | int retries, ret, err; |
| 456 | 448 | |
| 457 | 449 | DRM_DEBUG_DRIVER("GuC fw status: path %s, fetch %s, load %s\n", |
| 458 | 450 | fw_path, |
| 459 | − intel_guc_fw_status_repr(guc_fw->guc_fw_fetch_status), | |
| 460 | − intel_guc_fw_status_repr(guc_fw->guc_fw_load_status)); | |
| 451 | + intel_uc_fw_status_repr(guc_fw->fetch_status), | |
| 452 | + intel_uc_fw_status_repr(guc_fw->load_status)); | |
| 461 | 453 | |
| 462 | 454 | /* Loading forbidden, or no firmware to load? */ |
| 463 | 455 | if (!i915.enable_guc_loading) { |
| @@ -475,10 +467,10 @@ int intel_guc_setup(struct drm_i915_private *dev_priv) · 11 unmodified lines | ||
| 475 | 467 | } |
| 476 | 468 | |
| 477 | 469 | /* Fetch failed, or already fetched but failed to load? */ |
| 478 | − if (guc_fw->guc_fw_fetch_status != GUC_FIRMWARE_SUCCESS) { | |
| 470 | + if (guc_fw->fetch_status != INTEL_UC_FIRMWARE_SUCCESS) { | |
| 479 | 471 | err = -EIO; |
| 480 | 472 | goto fail; |
| 481 | − } else if (guc_fw->guc_fw_load_status == GUC_FIRMWARE_FAIL) { | |
| 473 | + } else if (guc_fw->load_status == INTEL_UC_FIRMWARE_FAIL) { | |
| 482 | 474 | err = -ENOEXEC; |
| 483 | 475 | goto fail; |
| 484 | 476 | } |
| @@ -486,11 +478,14 @@ int intel_guc_setup(struct drm_i915_private *dev_priv) · 1 unmodified line | ||
| 486 | 478 | guc_interrupts_release(dev_priv); |
| 487 | 479 | gen9_reset_guc_interrupts(dev_priv); |
| 488 | 480 | |
| 489 | − guc_fw->guc_fw_load_status = GUC_FIRMWARE_PENDING; | |
| 481 | + /* We need to notify the guc whenever we change the GGTT */ | |
| 482 | + i915_ggtt_enable_guc(dev_priv); | |
| 483 | + | |
| 484 | + guc_fw->load_status = INTEL_UC_FIRMWARE_PENDING; | |
| 490 | 485 | |
| 491 | 486 | DRM_DEBUG_DRIVER("GuC fw status: fetch %s, load %s\n", |
| 492 | − intel_guc_fw_status_repr(guc_fw->guc_fw_fetch_status), | |
| 493 | − intel_guc_fw_status_repr(guc_fw->guc_fw_load_status)); | |
| 487 | + intel_uc_fw_status_repr(guc_fw->fetch_status), | |
| 488 | + intel_uc_fw_status_repr(guc_fw->load_status)); | |
| 494 | 489 | |
| 495 | 490 | err = i915_guc_submission_init(dev_priv); |
| 496 | 491 | if (err) |
| @@ -511,6 +506,7 @@ int intel_guc_setup(struct drm_i915_private *dev_priv) · 14 unmodified lines | ||
| 511 | 506 | if (err) |
| 512 | 507 | goto fail; |
| 513 | 508 | |
| 509 | + intel_huc_load(dev_priv); | |
| 514 | 510 | err = guc_ucode_xfer(dev_priv); |
| 515 | 511 | if (!err) |
| 516 | 512 | break; |
| @@ -522,11 +518,13 @@ int intel_guc_setup(struct drm_i915_private *dev_priv) · 5 unmodified lines | ||
| 522 | 518 | "retry %d more time(s)\n", err, retries); |
| 523 | 519 | } |
| 524 | 520 | |
| 525 | − guc_fw->guc_fw_load_status = GUC_FIRMWARE_SUCCESS; | |
| 521 | + guc_fw->load_status = INTEL_UC_FIRMWARE_SUCCESS; | |
| 526 | 522 | |
| 527 | 523 | DRM_DEBUG_DRIVER("GuC fw status: fetch %s, load %s\n", |
| 528 | − intel_guc_fw_status_repr(guc_fw->guc_fw_fetch_status), | |
| 529 | − intel_guc_fw_status_repr(guc_fw->guc_fw_load_status)); | |
| 524 | + intel_uc_fw_status_repr(guc_fw->fetch_status), | |
| 525 | + intel_uc_fw_status_repr(guc_fw->load_status)); | |
| 526 | + | |
| 527 | + intel_guc_auth_huc(dev_priv); | |
| 530 | 528 | |
| 531 | 529 | if (i915.enable_guc_submission) { |
| 532 | 530 | if (i915.guc_log_level >= 0) |
| @@ -541,12 +539,13 @@ int intel_guc_setup(struct drm_i915_private *dev_priv) · 8 unmodified lines | ||
| 541 | 539 | return 0; |
| 542 | 540 | |
| 543 | 541 | fail: |
| 544 | − if (guc_fw->guc_fw_load_status == GUC_FIRMWARE_PENDING) | |
| 545 | − guc_fw->guc_fw_load_status = GUC_FIRMWARE_FAIL; | |
| 542 | + if (guc_fw->load_status == INTEL_UC_FIRMWARE_PENDING) | |
| 543 | + guc_fw->load_status = INTEL_UC_FIRMWARE_FAIL; | |
| 546 | 544 | |
| 547 | 545 | guc_interrupts_release(dev_priv); |
| 548 | 546 | i915_guc_submission_disable(dev_priv); |
| 549 | 547 | i915_guc_submission_fini(dev_priv); |
| 548 | + i915_ggtt_disable_guc(dev_priv); | |
| 550 | 549 | |
| 551 | 550 | /* |
| 552 | 551 | * We've failed to load the firmware :( |
| @@ -587,93 +586,108 @@ fail: · 34 unmodified lines | ||
| 587 | 586 | return ret; |
| 588 | 587 | } |
| 589 | 588 | |
| 590 | −static void guc_fw_fetch(struct drm_i915_private *dev_priv, | |
| 591 | − struct intel_guc_fw *guc_fw) | |
| 589 | +void intel_uc_fw_fetch(struct drm_i915_private *dev_priv, | |
| 590 | + struct intel_uc_fw *uc_fw) | |
| 592 | 591 | { |
| 593 | 592 | struct pci_dev *pdev = dev_priv->drm.pdev; |
| 594 | 593 | struct drm_i915_gem_object *obj; |
| 595 | 594 | const struct firmware *fw = NULL; |
| 596 | − struct guc_css_header *css; | |
| 595 | + struct uc_css_header *css; | |
| 597 | 596 | size_t size; |
| 598 | 597 | int err; |
| 599 | 598 | |
| 600 | − DRM_DEBUG_DRIVER("before requesting firmware: GuC fw fetch status %s\n", | |
| 601 | − intel_guc_fw_status_repr(guc_fw->guc_fw_fetch_status)); | |
| 599 | + DRM_DEBUG_DRIVER("before requesting firmware: uC fw fetch status %s\n", | |
| 600 | + intel_uc_fw_status_repr(uc_fw->fetch_status)); | |
| 602 | 601 | |
| 603 | − err = request_firmware(&fw, guc_fw->guc_fw_path, &pdev->dev); | |
| 602 | + err = request_firmware(&fw, uc_fw->path, &pdev->dev); | |
| 604 | 603 | if (err) |
| 605 | 604 | goto fail; |
| 606 | 605 | if (!fw) |
| 607 | 606 | goto fail; |
| 608 | 607 | |
| 609 | − DRM_DEBUG_DRIVER("fetch GuC fw from %s succeeded, fw %p\n", | |
| 610 | − guc_fw->guc_fw_path, fw); | |
| 608 | + DRM_DEBUG_DRIVER("fetch uC fw from %s succeeded, fw %p\n", | |
| 609 | + uc_fw->path, fw); | |
| 611 | 610 | |
| 612 | 611 | /* Check the size of the blob before examining buffer contents */ |
| 613 | − if (fw->size < sizeof(struct guc_css_header)) { | |
| 612 | + if (fw->size < sizeof(struct uc_css_header)) { | |
| 614 | 613 | DRM_NOTE("Firmware header is missing\n"); |
| 615 | 614 | goto fail; |
| 616 | 615 | } |
| 617 | 616 | |
| 618 | − css = (struct guc_css_header *)fw->data; | |
| 617 | + css = (struct uc_css_header *)fw->data; | |
| 619 | 618 | |
| 620 | 619 | /* Firmware bits always start from header */ |
| 621 | − guc_fw->header_offset = 0; | |
| 622 | − guc_fw->header_size = (css->header_size_dw - css->modulus_size_dw - | |
| 620 | + uc_fw->header_offset = 0; | |
| 621 | + uc_fw->header_size = (css->header_size_dw - css->modulus_size_dw - | |
| 623 | 622 | css->key_size_dw - css->exponent_size_dw) * sizeof(u32); |
| 624 | 623 | |
| 625 | − if (guc_fw->header_size != sizeof(struct guc_css_header)) { | |
| 624 | + if (uc_fw->header_size != sizeof(struct uc_css_header)) { | |
| 626 | 625 | DRM_NOTE("CSS header definition mismatch\n"); |
| 627 | 626 | goto fail; |
| 628 | 627 | } |
| 629 | 628 | |
| 630 | 629 | /* then, uCode */ |
| 631 | − guc_fw->ucode_offset = guc_fw->header_offset + guc_fw->header_size; | |
| 632 | − guc_fw->ucode_size = (css->size_dw - css->header_size_dw) * sizeof(u32); | |
| 630 | + uc_fw->ucode_offset = uc_fw->header_offset + uc_fw->header_size; | |
| 631 | + uc_fw->ucode_size = (css->size_dw - css->header_size_dw) * sizeof(u32); | |
| 633 | 632 | |
| 634 | 633 | /* now RSA */ |
| 635 | 634 | if (css->key_size_dw != UOS_RSA_SCRATCH_MAX_COUNT) { |
| 636 | 635 | DRM_NOTE("RSA key size is bad\n"); |
| 637 | 636 | goto fail; |
| 638 | 637 | } |
| 639 | − guc_fw->rsa_offset = guc_fw->ucode_offset + guc_fw->ucode_size; | |
| 640 | − guc_fw->rsa_size = css->key_size_dw * sizeof(u32); | |
| 638 | + uc_fw->rsa_offset = uc_fw->ucode_offset + uc_fw->ucode_size; | |
| 639 | + uc_fw->rsa_size = css->key_size_dw * sizeof(u32); | |
| 641 | 640 | |
| 642 | 641 | /* At least, it should have header, uCode and RSA. Size of all three. */ |
| 643 | − size = guc_fw->header_size + guc_fw->ucode_size + guc_fw->rsa_size; | |
| 642 | + size = uc_fw->header_size + uc_fw->ucode_size + uc_fw->rsa_size; | |
| 644 | 643 | if (fw->size < size) { |
| 645 | 644 | DRM_NOTE("Missing firmware components\n"); |
| 646 | 645 | goto fail; |
| 647 | 646 | } |
| 648 | 647 | |
| 649 | − /* Header and uCode will be loaded to WOPCM. Size of the two. */ | |
| 650 | − size = guc_fw->header_size + guc_fw->ucode_size; | |
| 651 | − if (size > guc_wopcm_size(dev_priv)) { | |
| 652 | − DRM_NOTE("Firmware is too large to fit in WOPCM\n"); | |
| 653 | − goto fail; | |
| 654 | − } | |
| 655 | − | |
| 656 | 648 | /* |
| 657 | 649 | * The GuC firmware image has the version number embedded at a well-known |
| 658 | 650 | * offset within the firmware blob; note that major / minor version are |
| 659 | 651 | * TWO bytes each (i.e. u16), although all pointers and offsets are defined |
| 660 | 652 | * in terms of bytes (u8). |
| 661 | 653 | */ |
| 662 | − guc_fw->guc_fw_major_found = css->guc_sw_version >> 16; | |
| 663 | − guc_fw->guc_fw_minor_found = css->guc_sw_version & 0xFFFF; | |
| 664 | − | |
| 665 | − if (guc_fw->guc_fw_major_found != guc_fw->guc_fw_major_wanted || | |
| 666 | − guc_fw->guc_fw_minor_found < guc_fw->guc_fw_minor_wanted) { | |
| 667 | − DRM_NOTE("GuC firmware version %d.%d, required %d.%d\n", | |
| 668 | − guc_fw->guc_fw_major_found, guc_fw->guc_fw_minor_found, | |
| 669 | − guc_fw->guc_fw_major_wanted, guc_fw->guc_fw_minor_wanted); | |
| 654 | + switch (uc_fw->fw) { | |
| 655 | + case INTEL_UC_FW_TYPE_GUC: | |
| 656 | + /* Header and uCode will be loaded to WOPCM. Size of the two. */ | |
| 657 | + size = uc_fw->header_size + uc_fw->ucode_size; | |
| 658 | + | |
| 659 | + /* Top 32k of WOPCM is reserved (8K stack + 24k RC6 context). */ | |
| 660 | + if (size > intel_guc_wopcm_size(dev_priv)) { | |
| 661 | + DRM_ERROR("Firmware is too large to fit in WOPCM\n"); | |
| 662 | + goto fail; | |
| 663 | + } | |
| 664 | + uc_fw->major_ver_found = css->guc.sw_version >> 16; | |
| 665 | + uc_fw->minor_ver_found = css->guc.sw_version & 0xFFFF; | |
| 666 | + break; | |
| 667 | + | |
| 668 | + case INTEL_UC_FW_TYPE_HUC: | |
| 669 | + uc_fw->major_ver_found = css->huc.sw_version >> 16; | |
| 670 | + uc_fw->minor_ver_found = css->huc.sw_version & 0xFFFF; | |
| 671 | + break; | |
| 672 | + | |
| 673 | + default: | |
| 674 | + DRM_ERROR("Unknown firmware type %d\n", uc_fw->fw); | |
| 675 | + err = -ENOEXEC; | |
| 676 | + goto fail; | |
| 677 | + } | |
| 678 | + | |
| 679 | + if (uc_fw->major_ver_found != uc_fw->major_ver_wanted || | |
| 680 | + uc_fw->minor_ver_found < uc_fw->minor_ver_wanted) { | |
| 681 | + DRM_NOTE("uC firmware version %d.%d, required %d.%d\n", | |
| 682 | + uc_fw->major_ver_found, uc_fw->minor_ver_found, | |
| 683 | + uc_fw->major_ver_wanted, uc_fw->minor_ver_wanted); | |
| 670 | 684 | err = -ENOEXEC; |
| 671 | 685 | goto fail; |
| 672 | 686 | } |
| 673 | 687 | |
| 674 | 688 | DRM_DEBUG_DRIVER("firmware version %d.%d OK (minimum %d.%d)\n", |
| 675 | − guc_fw->guc_fw_major_found, guc_fw->guc_fw_minor_found, | |
| 676 | − guc_fw->guc_fw_major_wanted, guc_fw->guc_fw_minor_wanted); | |
| 689 | + uc_fw->major_ver_found, uc_fw->minor_ver_found, | |
| 690 | + uc_fw->major_ver_wanted, uc_fw->minor_ver_wanted); | |
| 677 | 691 | |
| 678 | 692 | mutex_lock(&dev_priv->drm.struct_mutex); |
| 679 | 693 | obj = i915_gem_object_create_from_data(dev_priv, fw->data, fw->size); |
| @@ -683,31 +697,31 @@ static void guc_fw_fetch(struct drm_i915_private *dev_priv, · 3 unmodified lines | ||
| 683 | 697 | goto fail; |
| 684 | 698 | } |
| 685 | 699 | |
| 686 | − guc_fw->guc_fw_obj = obj; | |
| 687 | − guc_fw->guc_fw_size = fw->size; | |
| 700 | + uc_fw->obj = obj; | |
| 701 | + uc_fw->size = fw->size; | |
| 688 | 702 | |
| 689 | − DRM_DEBUG_DRIVER("GuC fw fetch status SUCCESS, obj %p\n", | |
| 690 | − guc_fw->guc_fw_obj); | |
| 703 | + DRM_DEBUG_DRIVER("uC fw fetch status SUCCESS, obj %p\n", | |
| 704 | + uc_fw->obj); | |
| 691 | 705 | |
| 692 | 706 | release_firmware(fw); |
| 693 | − guc_fw->guc_fw_fetch_status = GUC_FIRMWARE_SUCCESS; | |
| 707 | + uc_fw->fetch_status = INTEL_UC_FIRMWARE_SUCCESS; | |
| 694 | 708 | return; |
| 695 | 709 | |
| 696 | 710 | fail: |
| 697 | − DRM_WARN("Failed to fetch valid GuC firmware from %s (error %d)\n", | |
| 698 | − guc_fw->guc_fw_path, err); | |
| 699 | − DRM_DEBUG_DRIVER("GuC fw fetch status FAIL; err %d, fw %p, obj %p\n", | |
| 700 | − err, fw, guc_fw->guc_fw_obj); | |
| 711 | + DRM_WARN("Failed to fetch valid uC firmware from %s (error %d)\n", | |
| 712 | + uc_fw->path, err); | |
| 713 | + DRM_DEBUG_DRIVER("uC fw fetch status FAIL; err %d, fw %p, obj %p\n", | |
| 714 | + err, fw, uc_fw->obj); | |
| 701 | 715 | |
| 702 | 716 | mutex_lock(&dev_priv->drm.struct_mutex); |
| 703 | − obj = guc_fw->guc_fw_obj; | |
| 717 | + obj = uc_fw->obj; | |
| 704 | 718 | if (obj) |
| 705 | 719 | i915_gem_object_put(obj); |
| 706 | − guc_fw->guc_fw_obj = NULL; | |
| 720 | + uc_fw->obj = NULL; | |
| 707 | 721 | mutex_unlock(&dev_priv->drm.struct_mutex); |
| 708 | 722 | |
| 709 | 723 | release_firmware(fw); /* OK even if fw is NULL */ |
| 710 | − guc_fw->guc_fw_fetch_status = GUC_FIRMWARE_FAIL; | |
| 724 | + uc_fw->fetch_status = INTEL_UC_FIRMWARE_FAIL; | |
| 711 | 725 | } |
| 712 | 726 | |
| 713 | 727 | /** |
| @@ -721,7 +735,7 @@ fail: · 7 unmodified lines | ||
| 721 | 735 | */ |
| 722 | 736 | void intel_guc_init(struct drm_i915_private *dev_priv) |
| 723 | 737 | { |
| 724 | − struct intel_guc_fw *guc_fw = &dev_priv->guc.guc_fw; | |
| 738 | + struct intel_uc_fw *guc_fw = &dev_priv->guc.fw; | |
| 725 | 739 | const char *fw_path; |
| 726 | 740 | |
| 727 | 741 | if (!HAS_GUC(dev_priv)) { |
| @@ -739,23 +753,23 @@ void intel_guc_init(struct drm_i915_private *dev_priv) · 11 unmodified lines | ||
| 739 | 753 | fw_path = NULL; |
| 740 | 754 | } else if (IS_SKYLAKE(dev_priv)) { |
| 741 | 755 | fw_path = I915_SKL_GUC_UCODE; |
| 742 | − guc_fw->guc_fw_major_wanted = SKL_FW_MAJOR; | |
| 743 | − guc_fw->guc_fw_minor_wanted = SKL_FW_MINOR; | |
| 756 | + guc_fw->major_ver_wanted = SKL_FW_MAJOR; | |
| 757 | + guc_fw->minor_ver_wanted = SKL_FW_MINOR; | |
| 744 | 758 | } else if (IS_BROXTON(dev_priv)) { |
| 745 | 759 | fw_path = I915_BXT_GUC_UCODE; |
| 746 | − guc_fw->guc_fw_major_wanted = BXT_FW_MAJOR; | |
| 747 | − guc_fw->guc_fw_minor_wanted = BXT_FW_MINOR; | |
| 760 | + guc_fw->major_ver_wanted = BXT_FW_MAJOR; | |
| 761 | + guc_fw->minor_ver_wanted = BXT_FW_MINOR; | |
| 748 | 762 | } else if (IS_KABYLAKE(dev_priv)) { |
| 749 | 763 | fw_path = I915_KBL_GUC_UCODE; |
| 750 | − guc_fw->guc_fw_major_wanted = KBL_FW_MAJOR; | |
| 751 | − guc_fw->guc_fw_minor_wanted = KBL_FW_MINOR; | |
| 764 | + guc_fw->major_ver_wanted = KBL_FW_MAJOR; | |
| 765 | + guc_fw->minor_ver_wanted = KBL_FW_MINOR; | |
| 752 | 766 | } else { |
| 753 | 767 | fw_path = ""; /* unknown device */ |
| 754 | 768 | } |
| 755 | 769 | |
| 756 | − guc_fw->guc_fw_path = fw_path; | |
| 757 | − guc_fw->guc_fw_fetch_status = GUC_FIRMWARE_NONE; | |
| 758 | − guc_fw->guc_fw_load_status = GUC_FIRMWARE_NONE; | |
| 770 | + guc_fw->path = fw_path; | |
| 771 | + guc_fw->fetch_status = INTEL_UC_FIRMWARE_NONE; | |
| 772 | + guc_fw->load_status = INTEL_UC_FIRMWARE_NONE; | |
| 759 | 773 | |
| 760 | 774 | /* Early (and silent) return if GuC loading is disabled */ |
| 761 | 775 | if (!i915.enable_guc_loading) |
| @@ -765,9 +779,9 @@ void intel_guc_init(struct drm_i915_private *dev_priv) · 3 unmodified lines | ||
| 765 | 779 | if (*fw_path == '\0') |
| 766 | 780 | return; |
| 767 | 781 | |
| 768 | − guc_fw->guc_fw_fetch_status = GUC_FIRMWARE_PENDING; | |
| 782 | + guc_fw->fetch_status = INTEL_UC_FIRMWARE_PENDING; | |
| 769 | 783 | DRM_DEBUG_DRIVER("GuC firmware pending, path %s\n", fw_path); |
| 770 | − guc_fw_fetch(dev_priv, guc_fw); | |
| 784 | + intel_uc_fw_fetch(dev_priv, guc_fw); | |
| 771 | 785 | /* status must now be FAIL or SUCCESS */ |
| 772 | 786 | } |
| 773 | 787 | |
| @@ -777,17 +791,17 @@ void intel_guc_init(struct drm_i915_private *dev_priv) · 3 unmodified lines | ||
| 777 | 791 | */ |
| 778 | 792 | void intel_guc_fini(struct drm_i915_private *dev_priv) |
| 779 | 793 | { |
| 780 | − struct intel_guc_fw *guc_fw = &dev_priv->guc.guc_fw; | |
| 794 | + struct intel_uc_fw *guc_fw = &dev_priv->guc.fw; | |
| 781 | 795 | |
| 782 | 796 | mutex_lock(&dev_priv->drm.struct_mutex); |
| 783 | 797 | guc_interrupts_release(dev_priv); |
| 784 | 798 | i915_guc_submission_disable(dev_priv); |
| 785 | 799 | i915_guc_submission_fini(dev_priv); |
| 786 | 800 | |
| 787 | − if (guc_fw->guc_fw_obj) | |
| 788 | − i915_gem_object_put(guc_fw->guc_fw_obj); | |
| 789 | − guc_fw->guc_fw_obj = NULL; | |
| 801 | + if (guc_fw->obj) | |
| 802 | + i915_gem_object_put(guc_fw->obj); | |
| 803 | + guc_fw->obj = NULL; | |
| 790 | 804 | mutex_unlock(&dev_priv->drm.struct_mutex); |
| 791 | 805 | |
| 792 | − guc_fw->guc_fw_fetch_status = GUC_FIRMWARE_NONE; | |
| 806 | + guc_fw->fetch_status = INTEL_UC_FIRMWARE_NONE; | |
| 793 | 807 | } |
| @@ -0,0 +1,658 @@ | ||
| 1 | +/* | |
| 2 | + * Copyright © 2014-2017 Intel Corporation | |
| 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 (including the next | |
| 12 | + * paragraph) shall be included in all copies or substantial portions of the | |
| 13 | + * 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 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
| 19 | + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
| 20 | + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
| 21 | + * IN THE SOFTWARE. | |
| 22 | + * | |
| 23 | + */ | |
| 24 | +#include <linux/debugfs.h> | |
| 25 | +#include <linux/relay.h> | |
| 26 | +#include "i915_drv.h" | |
| 27 | + | |
| 28 | +static void guc_log_capture_logs(struct intel_guc *guc); | |
| 29 | + | |
| 30 | +/** | |
| 31 | + * DOC: GuC firmware log | |
| 32 | + * | |
| 33 | + * Firmware log is enabled by setting i915.guc_log_level to non-negative level. | |
| 34 | + * Log data is printed out via reading debugfs i915_guc_log_dump. Reading from | |
| 35 | + * i915_guc_load_status will print out firmware loading status and scratch | |
| 36 | + * registers value. | |
| 37 | + * | |
| 38 | + */ | |
| 39 | + | |
| 40 | +static int guc_log_flush_complete(struct intel_guc *guc) | |
| 41 | +{ | |
| 42 | + u32 action[] = { | |
| 43 | + INTEL_GUC_ACTION_LOG_BUFFER_FILE_FLUSH_COMPLETE | |
| 44 | + }; | |
| 45 | + | |
| 46 | + return intel_guc_send(guc, action, ARRAY_SIZE(action)); | |
| 47 | +} | |
| 48 | + | |
| 49 | +static int guc_log_flush(struct intel_guc *guc) | |
| 50 | +{ | |
| 51 | + u32 action[] = { | |
| 52 | + INTEL_GUC_ACTION_FORCE_LOG_BUFFER_FLUSH, | |
| 53 | + 0 | |
| 54 | + }; | |
| 55 | + | |
| 56 | + return intel_guc_send(guc, action, ARRAY_SIZE(action)); | |
| 57 | +} | |
| 58 | + | |
| 59 | +static int guc_log_control(struct intel_guc *guc, u32 control_val) | |
| 60 | +{ | |
| 61 | + u32 action[] = { | |
| 62 | + INTEL_GUC_ACTION_UK_LOG_ENABLE_LOGGING, | |
| 63 | + control_val | |
| 64 | + }; | |
| 65 | + | |
| 66 | + return intel_guc_send(guc, action, ARRAY_SIZE(action)); | |
| 67 | +} | |
| 68 | + | |
| 69 | + | |
| 70 | +/* | |
| 71 | + * Sub buffer switch callback. Called whenever relay has to switch to a new | |
| 72 | + * sub buffer, relay stays on the same sub buffer if 0 is returned. | |
| 73 | + */ | |
| 74 | +static int subbuf_start_callback(struct rchan_buf *buf, | |
| 75 | + void *subbuf, | |
| 76 | + void *prev_subbuf, | |
| 77 | + size_t prev_padding) | |
| 78 | +{ | |
| 79 | + /* Use no-overwrite mode by default, where relay will stop accepting | |
| 80 | + * new data if there are no empty sub buffers left. | |
| 81 | + * There is no strict synchronization enforced by relay between Consumer | |
| 82 | + * and Producer. In overwrite mode, there is a possibility of getting | |
| 83 | + * inconsistent/garbled data, the producer could be writing on to the | |
| 84 | + * same sub buffer from which Consumer is reading. This can't be avoided | |
| 85 | + * unless Consumer is fast enough and can always run in tandem with | |
| 86 | + * Producer. | |
| 87 | + */ | |
| 88 | + if (relay_buf_full(buf)) | |
| 89 | + return 0; | |
| 90 | + | |
| 91 | + return 1; | |
| 92 | +} | |
| 93 | + | |
| 94 | +/* | |
| 95 | + * file_create() callback. Creates relay file in debugfs. | |
| 96 | + */ | |
| 97 | +static struct dentry *create_buf_file_callback(const char *filename, | |
| 98 | + struct dentry *parent, | |
| 99 | + umode_t mode, | |
| 100 | + struct rchan_buf *buf, | |
| 101 | + int *is_global) | |
| 102 | +{ | |
| 103 | + struct dentry *buf_file; | |
| 104 | + | |
| 105 | + /* This to enable the use of a single buffer for the relay channel and | |
| 106 | + * correspondingly have a single file exposed to User, through which | |
| 107 | + * it can collect the logs in order without any post-processing. | |
| 108 | + * Need to set 'is_global' even if parent is NULL for early logging. | |
| 109 | + */ | |
| 110 | + *is_global = 1; | |
| 111 | + | |
| 112 | + if (!parent) | |
| 113 | + return NULL; | |
| 114 | + | |
| 115 | + /* Not using the channel filename passed as an argument, since for each | |
| 116 | + * channel relay appends the corresponding CPU number to the filename | |
| 117 | + * passed in relay_open(). This should be fine as relay just needs a | |
| 118 | + * dentry of the file associated with the channel buffer and that file's | |
| 119 | + * name need not be same as the filename passed as an argument. | |
| 120 | + */ | |
| 121 | + buf_file = debugfs_create_file("guc_log", mode, | |
| 122 | + parent, buf, &relay_file_operations); | |
| 123 | + return buf_file; | |
| 124 | +} | |
| 125 | + | |
| 126 | +/* | |
| 127 | + * file_remove() default callback. Removes relay file in debugfs. | |
| 128 | + */ | |
| 129 | +static int remove_buf_file_callback(struct dentry *dentry) | |
| 130 | +{ | |
| 131 | + debugfs_remove(dentry); | |
| 132 | + return 0; | |
| 133 | +} | |
| 134 | + | |
| 135 | +/* relay channel callbacks */ | |
| 136 | +static struct rchan_callbacks relay_callbacks = { | |
| 137 | + .subbuf_start = subbuf_start_callback, | |
| 138 | + .create_buf_file = create_buf_file_callback, | |
| 139 | + .remove_buf_file = remove_buf_file_callback, | |
| 140 | +}; | |
| 141 | + | |
| 142 | +static void guc_log_remove_relay_file(struct intel_guc *guc) | |
| 143 | +{ | |
| 144 | + relay_close(guc->log.relay_chan); | |
| 145 | +} | |
| 146 | + | |
| 147 | +static int guc_log_create_relay_channel(struct intel_guc *guc) | |
| 148 | +{ | |
| 149 | + struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 150 | + struct rchan *guc_log_relay_chan; | |
| 151 | + size_t n_subbufs, subbuf_size; | |
| 152 | + | |
| 153 | + /* Keep the size of sub buffers same as shared log buffer */ | |
| 154 | + subbuf_size = guc->log.vma->obj->base.size; | |
| 155 | + | |
| 156 | + /* Store up to 8 snapshots, which is large enough to buffer sufficient | |
| 157 | + * boot time logs and provides enough leeway to User, in terms of | |
| 158 | + * latency, for consuming the logs from relay. Also doesn't take | |
| 159 | + * up too much memory. | |
| 160 | + */ | |
| 161 | + n_subbufs = 8; | |
| 162 | + | |
| 163 | + guc_log_relay_chan = relay_open(NULL, NULL, subbuf_size, | |
| 164 | + n_subbufs, &relay_callbacks, dev_priv); | |
| 165 | + if (!guc_log_relay_chan) { | |
| 166 | + DRM_ERROR("Couldn't create relay chan for GuC logging\n"); | |
| 167 | + return -ENOMEM; | |
| 168 | + } | |
| 169 | + | |
| 170 | + GEM_BUG_ON(guc_log_relay_chan->subbuf_size < subbuf_size); | |
| 171 | + guc->log.relay_chan = guc_log_relay_chan; | |
| 172 | + return 0; | |
| 173 | +} | |
| 174 | + | |
| 175 | +static int guc_log_create_relay_file(struct intel_guc *guc) | |
| 176 | +{ | |
| 177 | + struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 178 | + struct dentry *log_dir; | |
| 179 | + int ret; | |
| 180 | + | |
| 181 | + /* For now create the log file in /sys/kernel/debug/dri/0 dir */ | |
| 182 | + log_dir = dev_priv->drm.primary->debugfs_root; | |
| 183 | + | |
| 184 | + /* If /sys/kernel/debug/dri/0 location do not exist, then debugfs is | |
| 185 | + * not mounted and so can't create the relay file. | |
| 186 | + * The relay API seems to fit well with debugfs only, for availing relay | |
| 187 | + * there are 3 requirements which can be met for debugfs file only in a | |
| 188 | + * straightforward/clean manner :- | |
| 189 | + * i) Need the associated dentry pointer of the file, while opening the | |
| 190 | + * relay channel. | |
| 191 | + * ii) Should be able to use 'relay_file_operations' fops for the file. | |
| 192 | + * iii) Set the 'i_private' field of file's inode to the pointer of | |
| 193 | + * relay channel buffer. | |
| 194 | + */ | |
| 195 | + if (!log_dir) { | |
| 196 | + DRM_ERROR("Debugfs dir not available yet for GuC log file\n"); | |
| 197 | + return -ENODEV; | |
| 198 | + } | |
| 199 | + | |
| 200 | + ret = relay_late_setup_files(guc->log.relay_chan, "guc_log", log_dir); | |
| 201 | + if (ret) { | |
| 202 | + DRM_ERROR("Couldn't associate relay chan with file %d\n", ret); | |
| 203 | + return ret; | |
| 204 | + } | |
| 205 | + | |
| 206 | + return 0; | |
| 207 | +} | |
| 208 | + | |
| 209 | +static void guc_move_to_next_buf(struct intel_guc *guc) | |
| 210 | +{ | |
| 211 | + /* Make sure the updates made in the sub buffer are visible when | |
| 212 | + * Consumer sees the following update to offset inside the sub buffer. | |
| 213 | + */ | |
| 214 | + smp_wmb(); | |
| 215 | + | |
| 216 | + /* All data has been written, so now move the offset of sub buffer. */ | |
| 217 | + relay_reserve(guc->log.relay_chan, guc->log.vma->obj->base.size); | |
| 218 | + | |
| 219 | + /* Switch to the next sub buffer */ | |
| 220 | + relay_flush(guc->log.relay_chan); | |
| 221 | +} | |
| 222 | + | |
| 223 | +static void *guc_get_write_buffer(struct intel_guc *guc) | |
| 224 | +{ | |
| 225 | + if (!guc->log.relay_chan) | |
| 226 | + return NULL; | |
| 227 | + | |
| 228 | + /* Just get the base address of a new sub buffer and copy data into it | |
| 229 | + * ourselves. NULL will be returned in no-overwrite mode, if all sub | |
| 230 | + * buffers are full. Could have used the relay_write() to indirectly | |
| 231 | + * copy the data, but that would have been bit convoluted, as we need to | |
| 232 | + * write to only certain locations inside a sub buffer which cannot be | |
| 233 | + * done without using relay_reserve() along with relay_write(). So its | |
| 234 | + * better to use relay_reserve() alone. | |
| 235 | + */ | |
| 236 | + return relay_reserve(guc->log.relay_chan, 0); | |
| 237 | +} | |
| 238 | + | |
| 239 | +static bool guc_check_log_buf_overflow(struct intel_guc *guc, | |
| 240 | + enum guc_log_buffer_type type, | |
| 241 | + unsigned int full_cnt) | |
| 242 | +{ | |
| 243 | + unsigned int prev_full_cnt = guc->log.prev_overflow_count[type]; | |
| 244 | + bool overflow = false; | |
| 245 | + | |
| 246 | + if (full_cnt != prev_full_cnt) { | |
| 247 | + overflow = true; | |
| 248 | + | |
| 249 | + guc->log.prev_overflow_count[type] = full_cnt; | |
| 250 | + guc->log.total_overflow_count[type] += full_cnt - prev_full_cnt; | |
| 251 | + | |
| 252 | + if (full_cnt < prev_full_cnt) { | |
| 253 | + /* buffer_full_cnt is a 4 bit counter */ | |
| 254 | + guc->log.total_overflow_count[type] += 16; | |
| 255 | + } | |
| 256 | + DRM_ERROR_RATELIMITED("GuC log buffer overflow\n"); | |
| 257 | + } | |
| 258 | + | |
| 259 | + return overflow; | |
| 260 | +} | |
| 261 | + | |
| 262 | +static unsigned int guc_get_log_buffer_size(enum guc_log_buffer_type type) | |
| 263 | +{ | |
| 264 | + switch (type) { | |
| 265 | + case GUC_ISR_LOG_BUFFER: | |
| 266 | + return (GUC_LOG_ISR_PAGES + 1) * PAGE_SIZE; | |
| 267 | + case GUC_DPC_LOG_BUFFER: | |
| 268 | + return (GUC_LOG_DPC_PAGES + 1) * PAGE_SIZE; | |
| 269 | + case GUC_CRASH_DUMP_LOG_BUFFER: | |
| 270 | + return (GUC_LOG_CRASH_PAGES + 1) * PAGE_SIZE; | |
| 271 | + default: | |
| 272 | + MISSING_CASE(type); | |
| 273 | + } | |
| 274 | + | |
| 275 | + return 0; | |
| 276 | +} | |
| 277 | + | |
| 278 | +static void guc_read_update_log_buffer(struct intel_guc *guc) | |
| 279 | +{ | |
| 280 | + unsigned int buffer_size, read_offset, write_offset, bytes_to_copy, full_cnt; | |
| 281 | + struct guc_log_buffer_state *log_buf_state, *log_buf_snapshot_state; | |
| 282 | + struct guc_log_buffer_state log_buf_state_local; | |
| 283 | + enum guc_log_buffer_type type; | |
| 284 | + void *src_data, *dst_data; | |
| 285 | + bool new_overflow; | |
| 286 | + | |
| 287 | + if (WARN_ON(!guc->log.buf_addr)) | |
| 288 | + return; | |
| 289 | + | |
| 290 | + /* Get the pointer to shared GuC log buffer */ | |
| 291 | + log_buf_state = src_data = guc->log.buf_addr; | |
| 292 | + | |
| 293 | + /* Get the pointer to local buffer to store the logs */ | |
| 294 | + log_buf_snapshot_state = dst_data = guc_get_write_buffer(guc); | |
| 295 | + | |
| 296 | + /* Actual logs are present from the 2nd page */ | |
| 297 | + src_data += PAGE_SIZE; | |
| 298 | + dst_data += PAGE_SIZE; | |
| 299 | + | |
| 300 | + for (type = GUC_ISR_LOG_BUFFER; type < GUC_MAX_LOG_BUFFER; type++) { | |
| 301 | + /* Make a copy of the state structure, inside GuC log buffer | |
| 302 | + * (which is uncached mapped), on the stack to avoid reading | |
| 303 | + * from it multiple times. | |
| 304 | + */ | |
| 305 | + memcpy(&log_buf_state_local, log_buf_state, | |
| 306 | + sizeof(struct guc_log_buffer_state)); | |
| 307 | + buffer_size = guc_get_log_buffer_size(type); | |
| 308 | + read_offset = log_buf_state_local.read_ptr; | |
| 309 | + write_offset = log_buf_state_local.sampled_write_ptr; | |
| 310 | + full_cnt = log_buf_state_local.buffer_full_cnt; | |
| 311 | + | |
| 312 | + /* Bookkeeping stuff */ | |
| 313 | + guc->log.flush_count[type] += log_buf_state_local.flush_to_file; | |
| 314 | + new_overflow = guc_check_log_buf_overflow(guc, type, full_cnt); | |
| 315 | + | |
| 316 | + /* Update the state of shared log buffer */ | |
| 317 | + log_buf_state->read_ptr = write_offset; | |
| 318 | + log_buf_state->flush_to_file = 0; | |
| 319 | + log_buf_state++; | |
| 320 | + | |
| 321 | + if (unlikely(!log_buf_snapshot_state)) | |
| 322 | + continue; | |
| 323 | + | |
| 324 | + /* First copy the state structure in snapshot buffer */ | |
| 325 | + memcpy(log_buf_snapshot_state, &log_buf_state_local, | |
| 326 | + sizeof(struct guc_log_buffer_state)); | |
| 327 | + | |
| 328 | + /* The write pointer could have been updated by GuC firmware, | |
| 329 | + * after sending the flush interrupt to Host, for consistency | |
| 330 | + * set write pointer value to same value of sampled_write_ptr | |
| 331 | + * in the snapshot buffer. | |
| 332 | + */ | |
| 333 | + log_buf_snapshot_state->write_ptr = write_offset; | |
| 334 | + log_buf_snapshot_state++; | |
| 335 | + | |
| 336 | + /* Now copy the actual logs. */ | |
| 337 | + if (unlikely(new_overflow)) { | |
| 338 | + /* copy the whole buffer in case of overflow */ | |
| 339 | + read_offset = 0; | |
| 340 | + write_offset = buffer_size; | |
| 341 | + } else if (unlikely((read_offset > buffer_size) || | |
| 342 | + (write_offset > buffer_size))) { | |
| 343 | + DRM_ERROR("invalid log buffer state\n"); | |
| 344 | + /* copy whole buffer as offsets are unreliable */ | |
| 345 | + read_offset = 0; | |
| 346 | + write_offset = buffer_size; | |
| 347 | + } | |
| 348 | + | |
| 349 | + /* Just copy the newly written data */ | |
| 350 | + if (read_offset > write_offset) { | |
| 351 | + i915_memcpy_from_wc(dst_data, src_data, write_offset); | |
| 352 | + bytes_to_copy = buffer_size - read_offset; | |
| 353 | + } else { | |
| 354 | + bytes_to_copy = write_offset - read_offset; | |
| 355 | + } | |
| 356 | + i915_memcpy_from_wc(dst_data + read_offset, | |
| 357 | + src_data + read_offset, bytes_to_copy); | |
| 358 | + | |
| 359 | + src_data += buffer_size; | |
| 360 | + dst_data += buffer_size; | |
| 361 | + } | |
| 362 | + | |
| 363 | + if (log_buf_snapshot_state) | |
| 364 | + guc_move_to_next_buf(guc); | |
| 365 | + else { | |
| 366 | + /* Used rate limited to avoid deluge of messages, logs might be | |
| 367 | + * getting consumed by User at a slow rate. | |
| 368 | + */ | |
| 369 | + DRM_ERROR_RATELIMITED("no sub-buffer to capture logs\n"); | |
| 370 | + guc->log.capture_miss_count++; | |
| 371 | + } | |
| 372 | +} | |
| 373 | + | |
| 374 | +static void guc_log_cleanup(struct intel_guc *guc) | |
| 375 | +{ | |
| 376 | + struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 377 | + | |
| 378 | + lockdep_assert_held(&dev_priv->drm.struct_mutex); | |
| 379 | + | |
| 380 | + /* First disable the flush interrupt */ | |
| 381 | + gen9_disable_guc_interrupts(dev_priv); | |
| 382 | + | |
| 383 | + if (guc->log.flush_wq) | |
| 384 | + destroy_workqueue(guc->log.flush_wq); | |
| 385 | + | |
| 386 | + guc->log.flush_wq = NULL; | |
| 387 | + | |
| 388 | + if (guc->log.relay_chan) | |
| 389 | + guc_log_remove_relay_file(guc); | |
| 390 | + | |
| 391 | + guc->log.relay_chan = NULL; | |
| 392 | + | |
| 393 | + if (guc->log.buf_addr) | |
| 394 | + i915_gem_object_unpin_map(guc->log.vma->obj); | |
| 395 | + | |
| 396 | + guc->log.buf_addr = NULL; | |
| 397 | +} | |
| 398 | + | |
| 399 | +static void capture_logs_work(struct work_struct *work) | |
| 400 | +{ | |
| 401 | + struct intel_guc *guc = | |
| 402 | + container_of(work, struct intel_guc, log.flush_work); | |
| 403 | + | |
| 404 | + guc_log_capture_logs(guc); | |
| 405 | +} | |
| 406 | + | |
| 407 | +static int guc_log_create_extras(struct intel_guc *guc) | |
| 408 | +{ | |
| 409 | + struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 410 | + void *vaddr; | |
| 411 | + int ret; | |
| 412 | + | |
| 413 | + lockdep_assert_held(&dev_priv->drm.struct_mutex); | |
| 414 | + | |
| 415 | + /* Nothing to do */ | |
| 416 | + if (i915.guc_log_level < 0) | |
| 417 | + return 0; | |
| 418 | + | |
| 419 | + if (!guc->log.buf_addr) { | |
| 420 | + /* Create a WC (Uncached for read) vmalloc mapping of log | |
| 421 | + * buffer pages, so that we can directly get the data | |
| 422 | + * (up-to-date) from memory. | |
| 423 | + */ | |
| 424 | + vaddr = i915_gem_object_pin_map(guc->log.vma->obj, I915_MAP_WC); | |
| 425 | + if (IS_ERR(vaddr)) { | |
| 426 | + ret = PTR_ERR(vaddr); | |
| 427 | + DRM_ERROR("Couldn't map log buffer pages %d\n", ret); | |
| 428 | + return ret; | |
| 429 | + } | |
| 430 | + | |
| 431 | + guc->log.buf_addr = vaddr; | |
| 432 | + } | |
| 433 | + | |
| 434 | + if (!guc->log.relay_chan) { | |
| 435 | + /* Create a relay channel, so that we have buffers for storing | |
| 436 | + * the GuC firmware logs, the channel will be linked with a file | |
| 437 | + * later on when debugfs is registered. | |
| 438 | + */ | |
| 439 | + ret = guc_log_create_relay_channel(guc); | |
| 440 | + if (ret) | |
| 441 | + return ret; | |
| 442 | + } | |
| 443 | + | |
| 444 | + if (!guc->log.flush_wq) { | |
| 445 | + INIT_WORK(&guc->log.flush_work, capture_logs_work); | |
| 446 | + | |
| 447 | + /* | |
| 448 | + * GuC log buffer flush work item has to do register access to | |
| 449 | + * send the ack to GuC and this work item, if not synced before | |
| 450 | + * suspend, can potentially get executed after the GFX device is | |
| 451 | + * suspended. | |
| 452 | + * By marking the WQ as freezable, we don't have to bother about | |
| 453 | + * flushing of this work item from the suspend hooks, the pending | |
| 454 | + * work item if any will be either executed before the suspend | |
| 455 | + * or scheduled later on resume. This way the handling of work | |
| 456 | + * item can be kept same between system suspend & rpm suspend. | |
| 457 | + */ | |
| 458 | + guc->log.flush_wq = alloc_ordered_workqueue("i915-guc_log", | |
| 459 | + WQ_HIGHPRI | WQ_FREEZABLE); | |
| 460 | + if (guc->log.flush_wq == NULL) { | |
| 461 | + DRM_ERROR("Couldn't allocate the wq for GuC logging\n"); | |
| 462 | + return -ENOMEM; | |
| 463 | + } | |
| 464 | + } | |
| 465 | + | |
| 466 | + return 0; | |
| 467 | +} | |
| 468 | + | |
| 469 | +void intel_guc_log_create(struct intel_guc *guc) | |
| 470 | +{ | |
| 471 | + struct i915_vma *vma; | |
| 472 | + unsigned long offset; | |
| 473 | + uint32_t size, flags; | |
| 474 | + | |
| 475 | + if (i915.guc_log_level > GUC_LOG_VERBOSITY_MAX) | |
| 476 | + i915.guc_log_level = GUC_LOG_VERBOSITY_MAX; | |
| 477 | + | |
| 478 | + /* The first page is to save log buffer state. Allocate one | |
| 479 | + * extra page for others in case for overlap */ | |
| 480 | + size = (1 + GUC_LOG_DPC_PAGES + 1 + | |
| 481 | + GUC_LOG_ISR_PAGES + 1 + | |
| 482 | + GUC_LOG_CRASH_PAGES + 1) << PAGE_SHIFT; | |
| 483 | + | |
| 484 | + vma = guc->log.vma; | |
| 485 | + if (!vma) { | |
| 486 | + /* We require SSE 4.1 for fast reads from the GuC log buffer and | |
| 487 | + * it should be present on the chipsets supporting GuC based | |
| 488 | + * submisssions. | |
| 489 | + */ | |
| 490 | + if (WARN_ON(!i915_has_memcpy_from_wc())) { | |
| 491 | + /* logging will not be enabled */ | |
| 492 | + i915.guc_log_level = -1; | |
| 493 | + return; | |
| 494 | + } | |
| 495 | + | |
| 496 | + vma = intel_guc_allocate_vma(guc, size); | |
| 497 | + if (IS_ERR(vma)) { | |
| 498 | + /* logging will be off */ | |
| 499 | + i915.guc_log_level = -1; | |
| 500 | + return; | |
| 501 | + } | |
| 502 | + | |
| 503 | + guc->log.vma = vma; | |
| 504 | + | |
| 505 | + if (guc_log_create_extras(guc)) { | |
| 506 | + guc_log_cleanup(guc); | |
| 507 | + i915_vma_unpin_and_release(&guc->log.vma); | |
| 508 | + i915.guc_log_level = -1; | |
| 509 | + return; | |
| 510 | + } | |
| 511 | + } | |
| 512 | + | |
| 513 | + /* each allocated unit is a page */ | |
| 514 | + flags = GUC_LOG_VALID | GUC_LOG_NOTIFY_ON_HALF_FULL | | |
| 515 | + (GUC_LOG_DPC_PAGES << GUC_LOG_DPC_SHIFT) | | |
| 516 | + (GUC_LOG_ISR_PAGES << GUC_LOG_ISR_SHIFT) | | |
| 517 | + (GUC_LOG_CRASH_PAGES << GUC_LOG_CRASH_SHIFT); | |
| 518 | + | |
| 519 | + offset = guc_ggtt_offset(vma) >> PAGE_SHIFT; /* in pages */ | |
| 520 | + guc->log.flags = (offset << GUC_LOG_BUF_ADDR_SHIFT) | flags; | |
| 521 | +} | |
| 522 | + | |
| 523 | +static int guc_log_late_setup(struct intel_guc *guc) | |
| 524 | +{ | |
| 525 | + struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 526 | + int ret; | |
| 527 | + | |
| 528 | + lockdep_assert_held(&dev_priv->drm.struct_mutex); | |
| 529 | + | |
| 530 | + if (i915.guc_log_level < 0) | |
| 531 | + return -EINVAL; | |
| 532 | + | |
| 533 | + /* If log_level was set as -1 at boot time, then setup needed to | |
| 534 | + * handle log buffer flush interrupts would not have been done yet, | |
| 535 | + * so do that now. | |
| 536 | + */ | |
| 537 | + ret = guc_log_create_extras(guc); | |
| 538 | + if (ret) | |
| 539 | + goto err; | |
| 540 | + | |
| 541 | + ret = guc_log_create_relay_file(guc); | |
| 542 | + if (ret) | |
| 543 | + goto err; | |
| 544 | + | |
| 545 | + return 0; | |
| 546 | +err: | |
| 547 | + guc_log_cleanup(guc); | |
| 548 | + /* logging will remain off */ | |
| 549 | + i915.guc_log_level = -1; | |
| 550 | + return ret; | |
| 551 | +} | |
| 552 | + | |
| 553 | +static void guc_log_capture_logs(struct intel_guc *guc) | |
| 554 | +{ | |
| 555 | + struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 556 | + | |
| 557 | + guc_read_update_log_buffer(guc); | |
| 558 | + | |
| 559 | + /* Generally device is expected to be active only at this | |
| 560 | + * time, so get/put should be really quick. | |
| 561 | + */ | |
| 562 | + intel_runtime_pm_get(dev_priv); | |
| 563 | + guc_log_flush_complete(guc); | |
| 564 | + intel_runtime_pm_put(dev_priv); | |
| 565 | +} | |
| 566 | + | |
| 567 | +static void guc_flush_logs(struct intel_guc *guc) | |
| 568 | +{ | |
| 569 | + struct drm_i915_private *dev_priv = guc_to_i915(guc); | |
| 570 | + | |
| 571 | + if (!i915.enable_guc_submission || (i915.guc_log_level < 0)) | |
| 572 | + return; | |
| 573 | + | |
| 574 | + /* First disable the interrupts, will be renabled afterwards */ | |
| 575 | + gen9_disable_guc_interrupts(dev_priv); | |
| 576 | + | |
| 577 | + /* Before initiating the forceful flush, wait for any pending/ongoing | |
| 578 | + * flush to complete otherwise forceful flush may not actually happen. | |
| 579 | + */ | |
| 580 | + flush_work(&guc->log.flush_work); | |
| 581 | + | |
| 582 | + /* Ask GuC to update the log buffer state */ | |
| 583 | + guc_log_flush(guc); | |
| 584 | + | |
| 585 | + /* GuC would have updated log buffer by now, so capture it */ | |
| 586 | + guc_log_capture_logs(guc); | |
| 587 | +} | |
| 588 | + | |
| 589 | +int i915_guc_log_control(struct drm_i915_private *dev_priv, u64 control_val) | |
| 590 | +{ | |
| 591 | + struct intel_guc *guc = &dev_priv->guc; | |
| 592 | + | |
| 593 | + union guc_log_control log_param; | |
| 594 | + int ret; | |
| 595 | + | |
| 596 | + log_param.value = control_val; | |
| 597 | + | |
| 598 | + if (log_param.verbosity < GUC_LOG_VERBOSITY_MIN || | |
| 599 | + log_param.verbosity > GUC_LOG_VERBOSITY_MAX) | |
| 600 | + return -EINVAL; | |
| 601 | + | |
| 602 | + /* This combination doesn't make sense & won't have any effect */ | |
| 603 | + if (!log_param.logging_enabled && (i915.guc_log_level < 0)) | |
| 604 | + return 0; | |
| 605 | + | |
| 606 | + ret = guc_log_control(guc, log_param.value); | |
| 607 | + if (ret < 0) { | |
| 608 | + DRM_DEBUG_DRIVER("guc_logging_control action failed %d\n", ret); | |
| 609 | + return ret; | |
| 610 | + } | |
| 611 | + | |
| 612 | + i915.guc_log_level = log_param.verbosity; | |
| 613 | + | |
| 614 | + /* If log_level was set as -1 at boot time, then the relay channel file | |
| 615 | + * wouldn't have been created by now and interrupts also would not have | |
| 616 | + * been enabled. | |
| 617 | + */ | |
| 618 | + if (!dev_priv->guc.log.relay_chan) { | |
| 619 | + ret = guc_log_late_setup(guc); | |
| 620 | + if (!ret) | |
| 621 | + gen9_enable_guc_interrupts(dev_priv); | |
| 622 | + } else if (!log_param.logging_enabled) { | |
| 623 | + /* Once logging is disabled, GuC won't generate logs & send an | |
| 624 | + * interrupt. But there could be some data in the log buffer | |
| 625 | + * which is yet to be captured. So request GuC to update the log | |
| 626 | + * buffer state and then collect the left over logs. | |
| 627 | + */ | |
| 628 | + guc_flush_logs(guc); | |
| 629 | + | |
| 630 | + /* As logging is disabled, update log level to reflect that */ | |
| 631 | + i915.guc_log_level = -1; | |
| 632 | + } else { | |
| 633 | + /* In case interrupts were disabled, enable them now */ | |
| 634 | + gen9_enable_guc_interrupts(dev_priv); | |
| 635 | + } | |
| 636 | + | |
| 637 | + return ret; | |
| 638 | +} | |
| 639 | + | |
| 640 | +void i915_guc_log_register(struct drm_i915_private *dev_priv) | |
| 641 | +{ | |
| 642 | + if (!i915.enable_guc_submission) | |
| 643 | + return; | |
| 644 | + | |
| 645 | + mutex_lock(&dev_priv->drm.struct_mutex); | |
| 646 | + guc_log_late_setup(&dev_priv->guc); | |
| 647 | + mutex_unlock(&dev_priv->drm.struct_mutex); | |
| 648 | +} | |
| 649 | + | |
| 650 | +void i915_guc_log_unregister(struct drm_i915_private *dev_priv) | |
| 651 | +{ | |
| 652 | + if (!i915.enable_guc_submission) | |
| 653 | + return; | |
| 654 | + | |
| 655 | + mutex_lock(&dev_priv->drm.struct_mutex); | |
| 656 | + guc_log_cleanup(&dev_priv->guc); | |
| 657 | + mutex_unlock(&dev_priv->drm.struct_mutex); | |
| 658 | +} | |
| @@ -0,0 +1,338 @@ | ||
| 1 | +/* | |
| 2 | + * Copyright © 2016-2017 Intel Corporation | |
| 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 (including the next | |
| 12 | + * paragraph) shall be included in all copies or substantial portions of the | |
| 13 | + * 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 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
| 19 | + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
| 20 | + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
| 21 | + * IN THE SOFTWARE. | |
| 22 | + * | |
| 23 | + */ | |
| 24 | +#include <linux/firmware.h> | |
| 25 | +#include "i915_drv.h" | |
| 26 | +#include "intel_uc.h" | |
| 27 | + | |
| 28 | +/** | |
| 29 | + * DOC: HuC Firmware | |
| 30 | + * | |
| 31 | + * Motivation: | |
| 32 | + * GEN9 introduces a new dedicated firmware for usage in media HEVC (High | |
| 33 | + * Efficiency Video Coding) operations. Userspace can use the firmware | |
| 34 | + * capabilities by adding HuC specific commands to batch buffers. | |
| 35 | + * | |
| 36 | + * Implementation: | |
| 37 | + * The same firmware loader is used as the GuC. However, the actual | |
| 38 | + * loading to HW is deferred until GEM initialization is done. | |
| 39 | + * | |
| 40 | + * Note that HuC firmware loading must be done before GuC loading. | |
| 41 | + */ | |
| 42 | + | |
| 43 | +#define BXT_HUC_FW_MAJOR 01 | |
| 44 | +#define BXT_HUC_FW_MINOR 07 | |
| 45 | +#define BXT_BLD_NUM 1398 | |
| 46 | + | |
| 47 | +#define SKL_HUC_FW_MAJOR 01 | |
| 48 | +#define SKL_HUC_FW_MINOR 07 | |
| 49 | +#define SKL_BLD_NUM 1398 | |
| 50 | + | |
| 51 | +#define KBL_HUC_FW_MAJOR 02 | |
| 52 | +#define KBL_HUC_FW_MINOR 00 | |
| 53 | +#define KBL_BLD_NUM 1810 | |
| 54 | + | |
| 55 | +#define HUC_FW_PATH(platform, major, minor, bld_num) \ | |
| 56 | + "i915/" __stringify(platform) "_huc_ver" __stringify(major) "_" \ | |
| 57 | + __stringify(minor) "_" __stringify(bld_num) ".bin" | |
| 58 | + | |
| 59 | +#define I915_SKL_HUC_UCODE HUC_FW_PATH(skl, SKL_HUC_FW_MAJOR, \ | |
| 60 | + SKL_HUC_FW_MINOR, SKL_BLD_NUM) | |
| 61 | +MODULE_FIRMWARE(I915_SKL_HUC_UCODE); | |
| 62 | + | |
| 63 | +#define I915_BXT_HUC_UCODE HUC_FW_PATH(bxt, BXT_HUC_FW_MAJOR, \ | |
| 64 | + BXT_HUC_FW_MINOR, BXT_BLD_NUM) | |
| 65 | +MODULE_FIRMWARE(I915_BXT_HUC_UCODE); | |
| 66 | + | |
| 67 | +#define I915_KBL_HUC_UCODE HUC_FW_PATH(kbl, KBL_HUC_FW_MAJOR, \ | |
| 68 | + KBL_HUC_FW_MINOR, KBL_BLD_NUM) | |
| 69 | +MODULE_FIRMWARE(I915_KBL_HUC_UCODE); | |
| 70 | + | |
| 71 | +/** | |
| 72 | + * huc_ucode_xfer() - DMA's the firmware | |
| 73 | + * @dev_priv: the drm_i915_private device | |
| 74 | + * | |
| 75 | + * Transfer the firmware image to RAM for execution by the microcontroller. | |
| 76 | + * | |
| 77 | + * Return: 0 on success, non-zero on failure | |
| 78 | + */ | |
| 79 | +static int huc_ucode_xfer(struct drm_i915_private *dev_priv) | |
| 80 | +{ | |
| 81 | + struct intel_uc_fw *huc_fw = &dev_priv->huc.fw; | |
| 82 | + struct i915_vma *vma; | |
| 83 | + unsigned long offset = 0; | |
| 84 | + u32 size; | |
| 85 | + int ret; | |
| 86 | + | |
| 87 | + ret = i915_gem_object_set_to_gtt_domain(huc_fw->obj, false); | |
| 88 | + if (ret) { | |
| 89 | + DRM_DEBUG_DRIVER("set-domain failed %d\n", ret); | |
| 90 | + return ret; | |
| 91 | + } | |
| 92 | + | |
| 93 | + vma = i915_gem_object_ggtt_pin(huc_fw->obj, NULL, 0, 0, | |
| 94 | + PIN_OFFSET_BIAS | GUC_WOPCM_TOP); | |
| 95 | + if (IS_ERR(vma)) { | |
| 96 | + DRM_DEBUG_DRIVER("pin failed %d\n", (int)PTR_ERR(vma)); | |
| 97 | + return PTR_ERR(vma); | |
| 98 | + } | |
| 99 | + | |
| 100 | + intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); | |
| 101 | + | |
| 102 | + /* init WOPCM */ | |
| 103 | + I915_WRITE(GUC_WOPCM_SIZE, intel_guc_wopcm_size(dev_priv)); | |
| 104 | + I915_WRITE(DMA_GUC_WOPCM_OFFSET, GUC_WOPCM_OFFSET_VALUE | | |
| 105 | + HUC_LOADING_AGENT_GUC); | |
| 106 | + | |
| 107 | + /* Set the source address for the uCode */ | |
| 108 | + offset = guc_ggtt_offset(vma) + huc_fw->header_offset; | |
| 109 | + I915_WRITE(DMA_ADDR_0_LOW, lower_32_bits(offset)); | |
| 110 | + I915_WRITE(DMA_ADDR_0_HIGH, upper_32_bits(offset) & 0xFFFF); | |
| 111 | + | |
| 112 | + /* Hardware doesn't look at destination address for HuC. Set it to 0, | |
| 113 | + * but still program the correct address space. | |
| 114 | + */ | |
| 115 | + I915_WRITE(DMA_ADDR_1_LOW, 0); | |
| 116 | + I915_WRITE(DMA_ADDR_1_HIGH, DMA_ADDRESS_SPACE_WOPCM); | |
| 117 | + | |
| 118 | + size = huc_fw->header_size + huc_fw->ucode_size; | |
| 119 | + I915_WRITE(DMA_COPY_SIZE, size); | |
| 120 | + | |
| 121 | + /* Start the DMA */ | |
| 122 | + I915_WRITE(DMA_CTRL, _MASKED_BIT_ENABLE(HUC_UKERNEL | START_DMA)); | |
| 123 | + | |
| 124 | + /* Wait for DMA to finish */ | |
| 125 | + ret = wait_for((I915_READ(DMA_CTRL) & START_DMA) == 0, 100); | |
| 126 | + | |
| 127 | + DRM_DEBUG_DRIVER("HuC DMA transfer wait over with ret %d\n", ret); | |
| 128 | + | |
| 129 | + /* Disable the bits once DMA is over */ | |
| 130 | + I915_WRITE(DMA_CTRL, _MASKED_BIT_DISABLE(HUC_UKERNEL)); | |
| 131 | + | |
| 132 | + intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); | |
| 133 | + | |
| 134 | + /* | |
| 135 | + * We keep the object pages for reuse during resume. But we can unpin it | |
| 136 | + * now that DMA has completed, so it doesn't continue to take up space. | |
| 137 | + */ | |
| 138 | + i915_vma_unpin(vma); | |
| 139 | + | |
| 140 | + return ret; | |
| 141 | +} | |
| 142 | + | |
| 143 | +/** | |
| 144 | + * intel_huc_init() - initiate HuC firmware loading request | |
| 145 | + * @dev_priv: the drm_i915_private device | |
| 146 | + * | |
| 147 | + * Called early during driver load, but after GEM is initialised. The loading | |
| 148 | + * will continue only when driver explicitly specify firmware name and version. | |
| 149 | + * All other cases are considered as INTEL_UC_FIRMWARE_NONE either because HW | |
| 150 | + * is not capable or driver yet support it. And there will be no error message | |
| 151 | + * for INTEL_UC_FIRMWARE_NONE cases. | |
| 152 | + * | |
| 153 | + * The DMA-copying to HW is done later when intel_huc_load() is called. | |
| 154 | + */ | |
| 155 | +void intel_huc_init(struct drm_i915_private *dev_priv) | |
| 156 | +{ | |
| 157 | + struct intel_huc *huc = &dev_priv->huc; | |
| 158 | + struct intel_uc_fw *huc_fw = &huc->fw; | |
| 159 | + const char *fw_path = NULL; | |
| 160 | + | |
| 161 | + huc_fw->path = NULL; | |
| 162 | + huc_fw->fetch_status = INTEL_UC_FIRMWARE_NONE; | |
| 163 | + huc_fw->load_status = INTEL_UC_FIRMWARE_NONE; | |
| 164 | + huc_fw->fw = INTEL_UC_FW_TYPE_HUC; | |
| 165 | + | |
| 166 | + if (!HAS_HUC_UCODE(dev_priv)) | |
| 167 | + return; | |
| 168 | + | |
| 169 | + if (IS_SKYLAKE(dev_priv)) { | |
| 170 | + fw_path = I915_SKL_HUC_UCODE; | |
| 171 | + huc_fw->major_ver_wanted = SKL_HUC_FW_MAJOR; | |
| 172 | + huc_fw->minor_ver_wanted = SKL_HUC_FW_MINOR; | |
| 173 | + } else if (IS_BROXTON(dev_priv)) { | |
| 174 | + fw_path = I915_BXT_HUC_UCODE; | |
| 175 | + huc_fw->major_ver_wanted = BXT_HUC_FW_MAJOR; | |
| 176 | + huc_fw->minor_ver_wanted = BXT_HUC_FW_MINOR; | |
| 177 | + } else if (IS_KABYLAKE(dev_priv)) { | |
| 178 | + fw_path = I915_KBL_HUC_UCODE; | |
| 179 | + huc_fw->major_ver_wanted = KBL_HUC_FW_MAJOR; | |
| 180 | + huc_fw->minor_ver_wanted = KBL_HUC_FW_MINOR; | |
| 181 | + } | |
| 182 | + | |
| 183 | + huc_fw->path = fw_path; | |
| 184 | + huc_fw->fetch_status = INTEL_UC_FIRMWARE_PENDING; | |
| 185 | + | |
| 186 | + DRM_DEBUG_DRIVER("HuC firmware pending, path %s\n", fw_path); | |
| 187 | + | |
| 188 | + WARN(huc_fw->path == NULL, "HuC present but no fw path\n"); | |
| 189 | + | |
| 190 | + intel_uc_fw_fetch(dev_priv, huc_fw); | |
| 191 | +} | |
| 192 | + | |
| 193 | +/** | |
| 194 | + * intel_huc_load() - load HuC uCode to device | |
| 195 | + * @dev_priv: the drm_i915_private device | |
| 196 | + * | |
| 197 | + * Called from guc_setup() during driver loading and also after a GPU reset. | |
| 198 | + * Be note that HuC loading must be done before GuC loading. | |
| 199 | + * | |
| 200 | + * The firmware image should have already been fetched into memory by the | |
| 201 | + * earlier call to intel_huc_init(), so here we need only check that | |
| 202 | + * is succeeded, and then transfer the image to the h/w. | |
| 203 | + * | |
| 204 | + * Return: non-zero code on error | |
| 205 | + */ | |
| 206 | +int intel_huc_load(struct drm_i915_private *dev_priv) | |
| 207 | +{ | |
| 208 | + struct intel_uc_fw *huc_fw = &dev_priv->huc.fw; | |
| 209 | + int err; | |
| 210 | + | |
| 211 | + if (huc_fw->fetch_status == INTEL_UC_FIRMWARE_NONE) | |
| 212 | + return 0; | |
| 213 | + | |
| 214 | + DRM_DEBUG_DRIVER("%s fw status: fetch %s, load %s\n", | |
| 215 | + huc_fw->path, | |
| 216 | + intel_uc_fw_status_repr(huc_fw->fetch_status), | |
| 217 | + intel_uc_fw_status_repr(huc_fw->load_status)); | |
| 218 | + | |
| 219 | + if (huc_fw->fetch_status == INTEL_UC_FIRMWARE_SUCCESS && | |
| 220 | + huc_fw->load_status == INTEL_UC_FIRMWARE_FAIL) | |
| 221 | + return -ENOEXEC; | |
| 222 | + | |
| 223 | + huc_fw->load_status = INTEL_UC_FIRMWARE_PENDING; | |
| 224 | + | |
| 225 | + switch (huc_fw->fetch_status) { | |
| 226 | + case INTEL_UC_FIRMWARE_FAIL: | |
| 227 | + /* something went wrong :( */ | |
| 228 | + err = -EIO; | |
| 229 | + goto fail; | |
| 230 | + | |
| 231 | + case INTEL_UC_FIRMWARE_NONE: | |
| 232 | + case INTEL_UC_FIRMWARE_PENDING: | |
| 233 | + default: | |
| 234 | + /* "can't happen" */ | |
| 235 | + WARN_ONCE(1, "HuC fw %s invalid fetch_status %s [%d]\n", | |
| 236 | + huc_fw->path, | |
| 237 | + intel_uc_fw_status_repr(huc_fw->fetch_status), | |
| 238 | + huc_fw->fetch_status); | |
| 239 | + err = -ENXIO; | |
| 240 | + goto fail; | |
| 241 | + | |
| 242 | + case INTEL_UC_FIRMWARE_SUCCESS: | |
| 243 | + break; | |
| 244 | + } | |
| 245 | + | |
| 246 | + err = huc_ucode_xfer(dev_priv); | |
| 247 | + if (err) | |
| 248 | + goto fail; | |
| 249 | + | |
| 250 | + huc_fw->load_status = INTEL_UC_FIRMWARE_SUCCESS; | |
| 251 | + | |
| 252 | + DRM_DEBUG_DRIVER("%s fw status: fetch %s, load %s\n", | |
| 253 | + huc_fw->path, | |
| 254 | + intel_uc_fw_status_repr(huc_fw->fetch_status), | |
| 255 | + intel_uc_fw_status_repr(huc_fw->load_status)); | |
| 256 | + | |
| 257 | + return 0; | |
| 258 | + | |
| 259 | +fail: | |
| 260 | + if (huc_fw->load_status == INTEL_UC_FIRMWARE_PENDING) | |
| 261 | + huc_fw->load_status = INTEL_UC_FIRMWARE_FAIL; | |
| 262 | + | |
| 263 | + DRM_ERROR("Failed to complete HuC uCode load with ret %d\n", err); | |
| 264 | + | |
| 265 | + return err; | |
| 266 | +} | |
| 267 | + | |
| 268 | +/** | |
| 269 | + * intel_huc_fini() - clean up resources allocated for HuC | |
| 270 | + * @dev_priv: the drm_i915_private device | |
| 271 | + * | |
| 272 | + * Cleans up by releasing the huc firmware GEM obj. | |
| 273 | + */ | |
| 274 | +void intel_huc_fini(struct drm_i915_private *dev_priv) | |
| 275 | +{ | |
| 276 | + struct intel_uc_fw *huc_fw = &dev_priv->huc.fw; | |
| 277 | + | |
| 278 | + mutex_lock(&dev_priv->drm.struct_mutex); | |
| 279 | + if (huc_fw->obj) | |
| 280 | + i915_gem_object_put(huc_fw->obj); | |
| 281 | + huc_fw->obj = NULL; | |
| 282 | + mutex_unlock(&dev_priv->drm.struct_mutex); | |
| 283 | + | |
| 284 | + huc_fw->fetch_status = INTEL_UC_FIRMWARE_NONE; | |
| 285 | +} | |
| 286 | + | |
| 287 | +/** | |
| 288 | + * intel_guc_auth_huc() - authenticate ucode | |
| 289 | + * @dev_priv: the drm_i915_device | |
| 290 | + * | |
| 291 | + * Triggers a HuC fw authentication request to the GuC via intel_guc_action_ | |
| 292 | + * authenticate_huc interface. | |
| 293 | + */ | |
| 294 | +void intel_guc_auth_huc(struct drm_i915_private *dev_priv) | |
| 295 | +{ | |
| 296 | + struct intel_guc *guc = &dev_priv->guc; | |
| 297 | + struct intel_huc *huc = &dev_priv->huc; | |
| 298 | + struct i915_vma *vma; | |
| 299 | + int ret; | |
| 300 | + u32 data[2]; | |
| 301 | + | |
| 302 | + if (huc->fw.load_status != INTEL_UC_FIRMWARE_SUCCESS) | |
| 303 | + return; | |
| 304 | + | |
| 305 | + vma = i915_gem_object_ggtt_pin(huc->fw.obj, NULL, 0, 0, | |
| 306 | + PIN_OFFSET_BIAS | GUC_WOPCM_TOP); | |
| 307 | + if (IS_ERR(vma)) { | |
| 308 | + DRM_ERROR("failed to pin huc fw object %d\n", | |
| 309 | + (int)PTR_ERR(vma)); | |
| 310 | + return; | |
| 311 | + } | |
| 312 | + | |
| 313 | + /* Specify auth action and where public signature is. */ | |
| 314 | + data[0] = INTEL_GUC_ACTION_AUTHENTICATE_HUC; | |
| 315 | + data[1] = guc_ggtt_offset(vma) + huc->fw.rsa_offset; | |
| 316 | + | |
| 317 | + ret = intel_guc_send(guc, data, ARRAY_SIZE(data)); | |
| 318 | + if (ret) { | |
| 319 | + DRM_ERROR("HuC: GuC did not ack Auth request %d\n", ret); | |
| 320 | + goto out; | |
| 321 | + } | |
| 322 | + | |
| 323 | + /* Check authentication status, it should be done by now */ | |
| 324 | + ret = intel_wait_for_register(dev_priv, | |
| 325 | + HUC_STATUS2, | |
| 326 | + HUC_FW_VERIFIED, | |
| 327 | + HUC_FW_VERIFIED, | |
| 328 | + 50); | |
| 329 | + | |
| 330 | + if (ret) { | |
| 331 | + DRM_ERROR("HuC: Authentication failed %d\n", ret); | |
| 332 | + goto out; | |
| 333 | + } | |
| 334 | + | |
| 335 | +out: | |
| 336 | + i915_vma_unpin(vma); | |
| 337 | +} | |
| 338 | + | |
| @@ -811,12 +811,6 @@ static int execlists_context_pin(struct intel_engine_cs *engine, · 810 unmodified lines | ||
| 811 | 811 | |
| 812 | 812 | ce->state->obj->mm.dirty = true; |
| 813 | 813 | |
| 814 | − /* Invalidate GuC TLB. */ | |
| 815 | − if (i915.enable_guc_submission) { | |
| 816 | − struct drm_i915_private *dev_priv = ctx->i915; | |
| 817 | − I915_WRITE(GEN8_GTCR, GEN8_GTCR_INVALIDATE); | |
| 818 | − } | |
| 819 | − | |
| 820 | 814 | i915_gem_context_get(ctx); |
| 821 | 815 | return 0; |
| 822 | 816 | |
| @@ -1231,7 +1225,7 @@ static int lrc_setup_wa_ctx_obj(struct intel_engine_cs *engine, u32 size) · 408 unmodified lines | ||
| 1231 | 1225 | if (IS_ERR(obj)) |
| 1232 | 1226 | return PTR_ERR(obj); |
| 1233 | 1227 | |
| 1234 | − vma = i915_vma_create(obj, &engine->i915->ggtt.base, NULL); | |
| 1228 | + vma = i915_vma_instance(obj, &engine->i915->ggtt.base, NULL); | |
| 1235 | 1229 | if (IS_ERR(vma)) { |
| 1236 | 1230 | err = PTR_ERR(vma); |
| 1237 | 1231 | goto err; |
| @@ -1917,7 +1911,7 @@ int logical_render_ring_init(struct intel_engine_cs *engine) · 679 unmodified lines | ||
| 1917 | 1911 | engine->emit_breadcrumb = gen8_emit_breadcrumb_render; |
| 1918 | 1912 | engine->emit_breadcrumb_sz = gen8_emit_breadcrumb_render_sz; |
| 1919 | 1913 | |
| 1920 | − ret = intel_engine_create_scratch(engine, 4096); | |
| 1914 | + ret = intel_engine_create_scratch(engine, PAGE_SIZE); | |
| 1921 | 1915 | if (ret) |
| 1922 | 1916 | return ret; |
| 1923 | 1917 | |
| @@ -2093,19 +2087,12 @@ static void execlists_init_reg_state(u32 *reg_state, · 169 unmodified lines | ||
| 2093 | 2087 | ASSIGN_CTX_REG(reg_state, CTX_PDP0_LDW, GEN8_RING_PDP_LDW(engine, 0), |
| 2094 | 2088 | 0); |
| 2095 | 2089 | |
| 2096 | − if (USES_FULL_48BIT_PPGTT(ppgtt->base.dev)) { | |
| 2090 | + if (ppgtt && USES_FULL_48BIT_PPGTT(ppgtt->base.dev)) { | |
| 2097 | 2091 | /* 64b PPGTT (48bit canonical) |
| 2098 | 2092 | * PDP0_DESCRIPTOR contains the base address to PML4 and |
| 2099 | 2093 | * other PDP Descriptors are ignored. |
| 2100 | 2094 | */ |
| 2101 | 2095 | ASSIGN_CTX_PML4(ppgtt, reg_state); |
| 2102 | − } else { | |
| 2103 | − /* 32b PPGTT | |
| 2104 | − * PDP*_DESCRIPTOR contains the base address of space supported. | |
| 2105 | − * With dynamic page allocation, PDPs may not be allocated at | |
| 2106 | − * this point. Point the unallocated PDPs to the scratch page | |
| 2107 | − */ | |
| 2108 | − execlists_update_context_pdps(ppgtt, reg_state); | |
| 2109 | 2096 | } |
| 2110 | 2097 | |
| 2111 | 2098 | if (engine->id == RCS) { |
| @@ -2199,7 +2186,8 @@ static int execlists_context_deferred_alloc(struct i915_gem_context *ctx, · 87 unmodified lines | ||
| 2199 | 2186 | |
| 2200 | 2187 | WARN_ON(ce->state); |
| 2201 | 2188 | |
| 2202 | − context_size = round_up(intel_lr_context_size(engine), 4096); | |
| 2189 | + context_size = round_up(intel_lr_context_size(engine), | |
| 2190 | + I915_GTT_PAGE_SIZE); | |
| 2203 | 2191 | |
| 2204 | 2192 | /* One extra page as the sharing data between driver and GuC */ |
| 2205 | 2193 | context_size += PAGE_SIZE * LRC_PPHWSP_PN; |
| @@ -2210,7 +2198,7 @@ static int execlists_context_deferred_alloc(struct i915_gem_context *ctx, · 4 unmodified lines | ||
| 2210 | 2198 | return PTR_ERR(ctx_obj); |
| 2211 | 2199 | } |
| 2212 | 2200 | |
| 2213 | − vma = i915_vma_create(ctx_obj, &ctx->i915->ggtt.base, NULL); | |
| 2201 | + vma = i915_vma_instance(ctx_obj, &ctx->i915->ggtt.base, NULL); | |
| 2214 | 2202 | if (IS_ERR(vma)) { |
| 2215 | 2203 | ret = PTR_ERR(vma); |
| 2216 | 2204 | goto error_deref_obj; |
| @@ -26,7 +26,7 @@ · 25 unmodified lines | ||
| 26 | 26 | |
| 27 | 27 | #include "intel_ringbuffer.h" |
| 28 | 28 | |
| 29 | −#define GEN8_LR_CONTEXT_ALIGN 4096 | |
| 29 | +#define GEN8_LR_CONTEXT_ALIGN I915_GTT_MIN_ALIGNMENT | |
| 30 | 30 | |
| 31 | 31 | /* Execlists regs */ |
| 32 | 32 | #define RING_ELSP(engine) _MMIO((engine)->mmio_base + 0x230) |
| @@ -811,8 +811,7 @@ static int intel_overlay_do_put_image(struct intel_overlay *overlay, · 810 unmodified lines | ||
| 811 | 811 | if (ret != 0) |
| 812 | 812 | return ret; |
| 813 | 813 | |
| 814 | − vma = i915_gem_object_pin_to_display_plane(new_bo, 0, | |
| 815 | − &i915_ggtt_view_normal); | |
| 814 | + vma = i915_gem_object_pin_to_display_plane(new_bo, 0, NULL); | |
| 816 | 815 | if (IS_ERR(vma)) |
| 817 | 816 | return PTR_ERR(vma); |
| 818 | 817 | |
| @@ -560,14 +560,14 @@ static void hsw_trans_edp_pipe_A_crc_wa(struct drm_i915_private *dev_priv, · 559 unmodified lines | ||
| 560 | 560 | state = drm_atomic_state_alloc(dev); |
| 561 | 561 | if (!state) { |
| 562 | 562 | ret = -ENOMEM; |
| 563 | − goto out; | |
| 563 | + goto unlock; | |
| 564 | 564 | } |
| 565 | 565 | |
| 566 | 566 | state->acquire_ctx = drm_modeset_legacy_acquire_ctx(&crtc->base); |
| 567 | 567 | pipe_config = intel_atomic_get_crtc_state(state, crtc); |
| 568 | 568 | if (IS_ERR(pipe_config)) { |
| 569 | 569 | ret = PTR_ERR(pipe_config); |
| 570 | − goto out; | |
| 570 | + goto put_state; | |
| 571 | 571 | } |
| 572 | 572 | |
| 573 | 573 | pipe_config->pch_pfit.force_thru = enable; |
| @@ -576,10 +576,12 @@ static void hsw_trans_edp_pipe_A_crc_wa(struct drm_i915_private *dev_priv, · 2 unmodified lines | ||
| 576 | 576 | pipe_config->base.connectors_changed = true; |
| 577 | 577 | |
| 578 | 578 | ret = drm_atomic_commit(state); |
| 579 | −out: | |
| 579 | + | |
| 580 | +put_state: | |
| 581 | + drm_atomic_state_put(state); | |
| 582 | +unlock: | |
| 580 | 583 | WARN(ret, "Toggling workaround to %i returns %i\n", enable, ret); |
| 581 | 584 | drm_modeset_unlock_all(dev); |
| 582 | − drm_atomic_state_put(state); | |
| 583 | 585 | } |
| 584 | 586 | |
| 585 | 587 | static int ivb_pipe_crc_ctl_reg(struct drm_i915_private *dev_priv, |
| @@ -613,6 +615,22 @@ static int ivb_pipe_crc_ctl_reg(struct drm_i915_private *dev_priv, · 27 unmodified lines | ||
| 613 | 615 | return 0; |
| 614 | 616 | } |
| 615 | 617 | |
| 618 | +static int get_new_crc_ctl_reg(struct drm_i915_private *dev_priv, | |
| 619 | + enum pipe pipe, | |
| 620 | + enum intel_pipe_crc_source *source, u32 *val) | |
| 621 | +{ | |
| 622 | + if (IS_GEN2(dev_priv)) | |
| 623 | + return i8xx_pipe_crc_ctl_reg(source, val); | |
| 624 | + else if (INTEL_GEN(dev_priv) < 5) | |
| 625 | + return i9xx_pipe_crc_ctl_reg(dev_priv, pipe, source, val); | |
| 626 | + else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) | |
| 627 | + return vlv_pipe_crc_ctl_reg(dev_priv, pipe, source, val); | |
| 628 | + else if (IS_GEN5(dev_priv) || IS_GEN6(dev_priv)) | |
| 629 | + return ilk_pipe_crc_ctl_reg(source, val); | |
| 630 | + else | |
| 631 | + return ivb_pipe_crc_ctl_reg(dev_priv, pipe, source, val); | |
| 632 | +} | |
| 633 | + | |
| 616 | 634 | static int pipe_crc_set_source(struct drm_i915_private *dev_priv, |
| 617 | 635 | enum pipe pipe, |
| 618 | 636 | enum intel_pipe_crc_source source) |
| @@ -636,17 +654,7 @@ static int pipe_crc_set_source(struct drm_i915_private *dev_priv, · 17 unmodified lines | ||
| 636 | 654 | return -EIO; |
| 637 | 655 | } |
| 638 | 656 | |
| 639 | − if (IS_GEN2(dev_priv)) | |
| 640 | − ret = i8xx_pipe_crc_ctl_reg(&source, &val); | |
| 641 | − else if (INTEL_GEN(dev_priv) < 5) | |
| 642 | − ret = i9xx_pipe_crc_ctl_reg(dev_priv, pipe, &source, &val); | |
| 643 | − else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) | |
| 644 | − ret = vlv_pipe_crc_ctl_reg(dev_priv, pipe, &source, &val); | |
| 645 | − else if (IS_GEN5(dev_priv) || IS_GEN6(dev_priv)) | |
| 646 | − ret = ilk_pipe_crc_ctl_reg(&source, &val); | |
| 647 | − else | |
| 648 | − ret = ivb_pipe_crc_ctl_reg(dev_priv, pipe, &source, &val); | |
| 649 | − | |
| 657 | + ret = get_new_crc_ctl_reg(dev_priv, pipe, &source, &val); | |
| 650 | 658 | if (ret != 0) |
| 651 | 659 | goto out; |
| 652 | 660 | |
| @@ -687,7 +695,7 @@ static int pipe_crc_set_source(struct drm_i915_private *dev_priv, · 34 unmodified lines | ||
| 687 | 695 | POSTING_READ(PIPE_CRC_CTL(pipe)); |
| 688 | 696 | |
| 689 | 697 | /* real source -> none transition */ |
| 690 | − if (source == INTEL_PIPE_CRC_SOURCE_NONE) { | |
| 698 | + if (!source) { | |
| 691 | 699 | struct intel_pipe_crc_entry *entries; |
| 692 | 700 | struct intel_crtc *crtc = intel_get_crtc_for_pipe(dev_priv, |
| 693 | 701 | pipe); |
| @@ -809,6 +817,11 @@ display_crc_ctl_parse_source(const char *buf, enum intel_pipe_crc_source *s) · 115 unmodified lines | ||
| 809 | 817 | { |
| 810 | 818 | int i; |
| 811 | 819 | |
| 820 | + if (!buf) { | |
| 821 | + *s = INTEL_PIPE_CRC_SOURCE_NONE; | |
| 822 | + return 0; | |
| 823 | + } | |
| 824 | + | |
| 812 | 825 | for (i = 0; i < ARRAY_SIZE(pipe_crc_sources); i++) |
| 813 | 826 | if (!strcmp(buf, pipe_crc_sources[i])) { |
| 814 | 827 | *s = i; |
| @@ -937,3 +950,62 @@ void intel_pipe_crc_cleanup(struct drm_minor *minor) · 122 unmodified lines | ||
| 937 | 950 | drm_debugfs_remove_files(info_list, 1, minor); |
| 938 | 951 | } |
| 939 | 952 | } |
| 953 | + | |
| 954 | +int intel_crtc_set_crc_source(struct drm_crtc *crtc, const char *source_name, | |
| 955 | + size_t *values_cnt) | |
| 956 | +{ | |
| 957 | + struct drm_i915_private *dev_priv = crtc->dev->dev_private; | |
| 958 | + struct intel_pipe_crc *pipe_crc = &dev_priv->pipe_crc[crtc->index]; | |
| 959 | + struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
| 960 | + enum intel_display_power_domain power_domain; | |
| 961 | + enum intel_pipe_crc_source source; | |
| 962 | + u32 val = 0; /* shut up gcc */ | |
| 963 | + int ret = 0; | |
| 964 | + | |
| 965 | + if (display_crc_ctl_parse_source(source_name, &source) < 0) { | |
| 966 | + DRM_DEBUG_DRIVER("unknown source %s\n", source_name); | |
| 967 | + return -EINVAL; | |
| 968 | + } | |
| 969 | + | |
| 970 | + power_domain = POWER_DOMAIN_PIPE(crtc->index); | |
| 971 | + if (!intel_display_power_get_if_enabled(dev_priv, power_domain)) { | |
| 972 | + DRM_DEBUG_KMS("Trying to capture CRC while pipe is off\n"); | |
| 973 | + return -EIO; | |
| 974 | + } | |
| 975 | + | |
| 976 | + ret = get_new_crc_ctl_reg(dev_priv, crtc->index, &source, &val); | |
| 977 | + if (ret != 0) | |
| 978 | + goto out; | |
| 979 | + | |
| 980 | + if (source) { | |
| 981 | + /* | |
| 982 | + * When IPS gets enabled, the pipe CRC changes. Since IPS gets | |
| 983 | + * enabled and disabled dynamically based on package C states, | |
| 984 | + * user space can't make reliable use of the CRCs, so let's just | |
| 985 | + * completely disable it. | |
| 986 | + */ | |
| 987 | + hsw_disable_ips(intel_crtc); | |
| 988 | + } | |
| 989 | + | |
| 990 | + I915_WRITE(PIPE_CRC_CTL(crtc->index), val); | |
| 991 | + POSTING_READ(PIPE_CRC_CTL(crtc->index)); | |
| 992 | + | |
| 993 | + if (!source) { | |
| 994 | + if (IS_G4X(dev_priv)) | |
| 995 | + g4x_undo_pipe_scramble_reset(dev_priv, crtc->index); | |
| 996 | + else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) | |
| 997 | + vlv_undo_pipe_scramble_reset(dev_priv, crtc->index); | |
| 998 | + else if (IS_HASWELL(dev_priv) && crtc->index == PIPE_A) | |
| 999 | + hsw_trans_edp_pipe_A_crc_wa(dev_priv, false); | |
| 1000 | + | |
| 1001 | + hsw_enable_ips(intel_crtc); | |
| 1002 | + } | |
| 1003 | + | |
| 1004 | + pipe_crc->skipped = 0; | |
| 1005 | + *values_cnt = 5; | |
| 1006 | + | |
| 1007 | +out: | |
| 1008 | + intel_display_power_put(dev_priv, power_domain); | |
| 1009 | + | |
| 1010 | + return ret; | |
| 1011 | +} | |
| @@ -122,13 +122,26 @@ static void vlv_psr_setup_vsc(struct intel_dp *intel_dp) · 121 unmodified lines | ||
| 122 | 122 | static void skl_psr_setup_su_vsc(struct intel_dp *intel_dp) |
| 123 | 123 | { |
| 124 | 124 | struct edp_vsc_psr psr_vsc; |
| 125 | + struct intel_digital_port *intel_dig_port = dp_to_dig_port(intel_dp); | |
| 126 | + struct drm_device *dev = intel_dig_port->base.base.dev; | |
| 127 | + struct drm_i915_private *dev_priv = to_i915(dev); | |
| 125 | 128 | |
| 126 | 129 | /* Prepare VSC Header for SU as per EDP 1.4 spec, Table 6.11 */ |
| 127 | 130 | memset(&psr_vsc, 0, sizeof(psr_vsc)); |
| 128 | 131 | psr_vsc.sdp_header.HB0 = 0; |
| 129 | 132 | psr_vsc.sdp_header.HB1 = 0x7; |
| 130 | − psr_vsc.sdp_header.HB2 = 0x3; | |
| 131 | − psr_vsc.sdp_header.HB3 = 0xb; | |
| 133 | + if (dev_priv->psr.colorimetry_support && | |
| 134 | + dev_priv->psr.y_cord_support) { | |
| 135 | + psr_vsc.sdp_header.HB2 = 0x5; | |
| 136 | + psr_vsc.sdp_header.HB3 = 0x13; | |
| 137 | + } else if (dev_priv->psr.y_cord_support) { | |
| 138 | + psr_vsc.sdp_header.HB2 = 0x4; | |
| 139 | + psr_vsc.sdp_header.HB3 = 0xe; | |
| 140 | + } else { | |
| 141 | + psr_vsc.sdp_header.HB2 = 0x3; | |
| 142 | + psr_vsc.sdp_header.HB3 = 0xc; | |
| 143 | + } | |
| 144 | + | |
| 132 | 145 | intel_psr_write_vsc(intel_dp, &psr_vsc); |
| 133 | 146 | } |
| 134 | 147 | |
| @@ -196,7 +209,11 @@ static void hsw_psr_enable_sink(struct intel_dp *intel_dp) · 61 unmodified lines | ||
| 196 | 209 | drm_dp_dpcd_writeb(&intel_dp->aux, |
| 197 | 210 | DP_SINK_DEVICE_AUX_FRAME_SYNC_CONF, |
| 198 | 211 | DP_AUX_FRAME_SYNC_ENABLE); |
| 199 | − | |
| 212 | + /* Enable ALPM at sink for psr2 */ | |
| 213 | + if (dev_priv->psr.psr2_support && dev_priv->psr.alpm) | |
| 214 | + drm_dp_dpcd_writeb(&intel_dp->aux, | |
| 215 | + DP_RECEIVER_ALPM_CONFIG, | |
| 216 | + DP_ALPM_ENABLE); | |
| 200 | 217 | if (dev_priv->psr.link_standby) |
| 201 | 218 | drm_dp_dpcd_writeb(&intel_dp->aux, DP_PSR_EN_CFG, |
| 202 | 219 | DP_PSR_ENABLE | DP_PSR_MAIN_LINK_ACTIVE); |
| @@ -248,7 +265,7 @@ static void vlv_psr_activate(struct intel_dp *intel_dp) · 45 unmodified lines | ||
| 248 | 265 | VLV_EDP_PSR_ACTIVE_ENTRY); |
| 249 | 266 | } |
| 250 | 267 | |
| 251 | −static void hsw_psr_enable_source(struct intel_dp *intel_dp) | |
| 268 | +static void intel_enable_source_psr1(struct intel_dp *intel_dp) | |
| 252 | 269 | { |
| 253 | 270 | struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp); |
| 254 | 271 | struct drm_device *dev = dig_port->base.base.dev; |
| @@ -299,14 +316,31 @@ static void hsw_psr_enable_source(struct intel_dp *intel_dp) · 44 unmodified lines | ||
| 299 | 316 | val |= EDP_PSR_TP1_TP2_SEL; |
| 300 | 317 | |
| 301 | 318 | I915_WRITE(EDP_PSR_CTL, val); |
| 319 | +} | |
| 302 | 320 | |
| 303 | − if (!dev_priv->psr.psr2_support) | |
| 304 | − return; | |
| 321 | +static void intel_enable_source_psr2(struct intel_dp *intel_dp) | |
| 322 | +{ | |
| 323 | + struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp); | |
| 324 | + struct drm_device *dev = dig_port->base.base.dev; | |
| 325 | + struct drm_i915_private *dev_priv = to_i915(dev); | |
| 326 | + /* | |
| 327 | + * Let's respect VBT in case VBT asks a higher idle_frame value. | |
| 328 | + * Let's use 6 as the minimum to cover all known cases including | |
| 329 | + * the off-by-one issue that HW has in some cases. Also there are | |
| 330 | + * cases where sink should be able to train | |
| 331 | + * with the 5 or 6 idle patterns. | |
| 332 | + */ | |
| 333 | + uint32_t idle_frames = max(6, dev_priv->vbt.psr.idle_frames); | |
| 334 | + uint32_t val; | |
| 335 | + | |
| 336 | + val = idle_frames << EDP_PSR_IDLE_FRAME_SHIFT; | |
| 305 | 337 | |
| 306 | 338 | /* FIXME: selective update is probably totally broken because it doesn't |
| 307 | 339 | * mesh at all with our frontbuffer tracking. And the hw alone isn't |
| 308 | 340 | * good enough. */ |
| 309 | − val = EDP_PSR2_ENABLE | EDP_SU_TRACK_ENABLE; | |
| 341 | + val |= EDP_PSR2_ENABLE | | |
| 342 | + EDP_SU_TRACK_ENABLE | | |
| 343 | + EDP_FRAMES_BEFORE_SU_ENTRY; | |
| 310 | 344 | |
| 311 | 345 | if (dev_priv->vbt.psr.tp2_tp3_wakeup_time > 5) |
| 312 | 346 | val |= EDP_PSR2_TP2_TIME_2500; |
| @@ -320,6 +354,19 @@ static void hsw_psr_enable_source(struct intel_dp *intel_dp) · 7 unmodified lines | ||
| 320 | 354 | I915_WRITE(EDP_PSR2_CTL, val); |
| 321 | 355 | } |
| 322 | 356 | |
| 357 | +static void hsw_psr_enable_source(struct intel_dp *intel_dp) | |
| 358 | +{ | |
| 359 | + struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp); | |
| 360 | + struct drm_device *dev = dig_port->base.base.dev; | |
| 361 | + struct drm_i915_private *dev_priv = to_i915(dev); | |
| 362 | + | |
| 363 | + /* psr1 and psr2 are mutually exclusive.*/ | |
| 364 | + if (dev_priv->psr.psr2_support) | |
| 365 | + intel_enable_source_psr2(intel_dp); | |
| 366 | + else | |
| 367 | + intel_enable_source_psr1(intel_dp); | |
| 368 | +} | |
| 369 | + | |
| 323 | 370 | static bool intel_psr_match_conditions(struct intel_dp *intel_dp) |
| 324 | 371 | { |
| 325 | 372 | struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp); |
| @@ -387,6 +434,22 @@ static bool intel_psr_match_conditions(struct intel_dp *intel_dp) · 61 unmodified lines | ||
| 387 | 434 | return false; |
| 388 | 435 | } |
| 389 | 436 | |
| 437 | + /* PSR2 is restricted to work with panel resolutions upto 3200x2000 */ | |
| 438 | + if (intel_crtc->config->pipe_src_w > 3200 || | |
| 439 | + intel_crtc->config->pipe_src_h > 2000) { | |
| 440 | + dev_priv->psr.psr2_support = false; | |
| 441 | + return false; | |
| 442 | + } | |
| 443 | + | |
| 444 | + /* | |
| 445 | + * FIXME:enable psr2 only for y-cordinate psr2 panels | |
| 446 | + * After gtc implementation , remove this restriction. | |
| 447 | + */ | |
| 448 | + if (!dev_priv->psr.y_cord_support && dev_priv->psr.psr2_support) { | |
| 449 | + DRM_DEBUG_KMS("PSR2 disabled, panel does not support Y coordinate\n"); | |
| 450 | + return false; | |
| 451 | + } | |
| 452 | + | |
| 390 | 453 | dev_priv->psr.source_ok = true; |
| 391 | 454 | return true; |
| 392 | 455 | } |
| @@ -397,7 +460,10 @@ static void intel_psr_activate(struct intel_dp *intel_dp) · 4 unmodified lines | ||
| 397 | 460 | struct drm_device *dev = intel_dig_port->base.base.dev; |
| 398 | 461 | struct drm_i915_private *dev_priv = to_i915(dev); |
| 399 | 462 | |
| 400 | − WARN_ON(I915_READ(EDP_PSR_CTL) & EDP_PSR_ENABLE); | |
| 463 | + if (dev_priv->psr.psr2_support) | |
| 464 | + WARN_ON(I915_READ(EDP_PSR2_CTL) & EDP_PSR2_ENABLE); | |
| 465 | + else | |
| 466 | + WARN_ON(I915_READ(EDP_PSR_CTL) & EDP_PSR_ENABLE); | |
| 401 | 467 | WARN_ON(dev_priv->psr.active); |
| 402 | 468 | lockdep_assert_held(&dev_priv->psr.lock); |
| 403 | 469 | |
| @@ -426,6 +492,8 @@ void intel_psr_enable(struct intel_dp *intel_dp) · 22 unmodified lines | ||
| 426 | 492 | struct drm_device *dev = intel_dig_port->base.base.dev; |
| 427 | 493 | struct drm_i915_private *dev_priv = to_i915(dev); |
| 428 | 494 | struct intel_crtc *crtc = to_intel_crtc(intel_dig_port->base.base.crtc); |
| 495 | + enum transcoder cpu_transcoder = crtc->config->cpu_transcoder; | |
| 496 | + u32 chicken; | |
| 429 | 497 | |
| 430 | 498 | if (!HAS_PSR(dev_priv)) { |
| 431 | 499 | DRM_DEBUG_KMS("PSR not supported on this platform\n"); |
| @@ -449,26 +517,34 @@ void intel_psr_enable(struct intel_dp *intel_dp) · 17 unmodified lines | ||
| 449 | 517 | dev_priv->psr.busy_frontbuffer_bits = 0; |
| 450 | 518 | |
| 451 | 519 | if (HAS_DDI(dev_priv)) { |
| 452 | − hsw_psr_setup_vsc(intel_dp); | |
| 453 | − | |
| 454 | 520 | if (dev_priv->psr.psr2_support) { |
| 455 | − /* PSR2 is restricted to work with panel resolutions upto 3200x2000 */ | |
| 456 | − if (crtc->config->pipe_src_w > 3200 || | |
| 457 | − crtc->config->pipe_src_h > 2000) | |
| 458 | − dev_priv->psr.psr2_support = false; | |
| 459 | − else | |
| 460 | − skl_psr_setup_su_vsc(intel_dp); | |
| 521 | + skl_psr_setup_su_vsc(intel_dp); | |
| 522 | + chicken = PSR2_VSC_ENABLE_PROG_HEADER; | |
| 523 | + if (dev_priv->psr.y_cord_support) | |
| 524 | + chicken |= PSR2_ADD_VERTICAL_LINE_COUNT; | |
| 525 | + I915_WRITE(CHICKEN_TRANS(cpu_transcoder), chicken); | |
| 526 | + I915_WRITE(EDP_PSR_DEBUG_CTL, | |
| 527 | + EDP_PSR_DEBUG_MASK_MEMUP | | |
| 528 | + EDP_PSR_DEBUG_MASK_HPD | | |
| 529 | + EDP_PSR_DEBUG_MASK_LPSP | | |
| 530 | + EDP_PSR_DEBUG_MASK_MAX_SLEEP | | |
| 531 | + EDP_PSR_DEBUG_MASK_DISP_REG_WRITE); | |
| 532 | + } else { | |
| 533 | + /* set up vsc header for psr1 */ | |
| 534 | + hsw_psr_setup_vsc(intel_dp); | |
| 535 | + /* | |
| 536 | + * Per Spec: Avoid continuous PSR exit by masking MEMUP | |
| 537 | + * and HPD. also mask LPSP to avoid dependency on other | |
| 538 | + * drivers that might block runtime_pm besides | |
| 539 | + * preventing other hw tracking issues now we can rely | |
| 540 | + * on frontbuffer tracking. | |
| 541 | + */ | |
| 542 | + I915_WRITE(EDP_PSR_DEBUG_CTL, | |
| 543 | + EDP_PSR_DEBUG_MASK_MEMUP | | |
| 544 | + EDP_PSR_DEBUG_MASK_HPD | | |
| 545 | + EDP_PSR_DEBUG_MASK_LPSP); | |
| 461 | 546 | } |
| 462 | 547 | |
| 463 | − /* | |
| 464 | − * Per Spec: Avoid continuous PSR exit by masking MEMUP and HPD. | |
| 465 | − * Also mask LPSP to avoid dependency on other drivers that | |
| 466 | − * might block runtime_pm besides preventing other hw tracking | |
| 467 | − * issues now we can rely on frontbuffer tracking. | |
| 468 | − */ | |
| 469 | − I915_WRITE(EDP_PSR_DEBUG_CTL, EDP_PSR_DEBUG_MASK_MEMUP | | |
| 470 | − EDP_PSR_DEBUG_MASK_HPD | EDP_PSR_DEBUG_MASK_LPSP); | |
| 471 | − | |
| 472 | 548 | /* Enable PSR on the panel */ |
| 473 | 549 | hsw_psr_enable_sink(intel_dp); |
| 474 | 550 | |
| @@ -544,20 +620,42 @@ static void hsw_psr_disable(struct intel_dp *intel_dp) · 69 unmodified lines | ||
| 544 | 620 | struct drm_i915_private *dev_priv = to_i915(dev); |
| 545 | 621 | |
| 546 | 622 | if (dev_priv->psr.active) { |
| 547 | − I915_WRITE(EDP_PSR_CTL, | |
| 548 | − I915_READ(EDP_PSR_CTL) & ~EDP_PSR_ENABLE); | |
| 623 | + i915_reg_t psr_ctl; | |
| 624 | + u32 psr_status_mask; | |
| 625 | + | |
| 626 | + if (dev_priv->psr.aux_frame_sync) | |
| 627 | + drm_dp_dpcd_writeb(&intel_dp->aux, | |
| 628 | + DP_SINK_DEVICE_AUX_FRAME_SYNC_CONF, | |
| 629 | + 0); | |
| 630 | + | |
| 631 | + if (dev_priv->psr.psr2_support) { | |
| 632 | + psr_ctl = EDP_PSR2_CTL; | |
| 633 | + psr_status_mask = EDP_PSR2_STATUS_STATE_MASK; | |
| 634 | + | |
| 635 | + I915_WRITE(psr_ctl, | |
| 636 | + I915_READ(psr_ctl) & | |
| 637 | + ~(EDP_PSR2_ENABLE | EDP_SU_TRACK_ENABLE)); | |
| 638 | + | |
| 639 | + } else { | |
| 640 | + psr_ctl = EDP_PSR_STATUS_CTL; | |
| 641 | + psr_status_mask = EDP_PSR_STATUS_STATE_MASK; | |
| 642 | + | |
| 643 | + I915_WRITE(psr_ctl, | |
| 644 | + I915_READ(psr_ctl) & ~EDP_PSR_ENABLE); | |
| 645 | + } | |
| 549 | 646 | |
| 550 | 647 | /* Wait till PSR is idle */ |
| 551 | 648 | if (intel_wait_for_register(dev_priv, |
| 552 | − EDP_PSR_STATUS_CTL, | |
| 553 | − EDP_PSR_STATUS_STATE_MASK, | |
| 554 | − 0, | |
| 649 | + psr_ctl, psr_status_mask, 0, | |
| 555 | 650 | 2000)) |
| 556 | 651 | DRM_ERROR("Timed out waiting for PSR Idle State\n"); |
| 557 | 652 | |
| 558 | 653 | dev_priv->psr.active = false; |
| 559 | 654 | } else { |
| 560 | − WARN_ON(I915_READ(EDP_PSR_CTL) & EDP_PSR_ENABLE); | |
| 655 | + if (dev_priv->psr.psr2_support) | |
| 656 | + WARN_ON(I915_READ(EDP_PSR2_CTL) & EDP_PSR2_ENABLE); | |
| 657 | + else | |
| 658 | + WARN_ON(I915_READ(EDP_PSR_CTL) & EDP_PSR_ENABLE); | |
| 561 | 659 | } |
| 562 | 660 | } |
| 563 | 661 | |
| @@ -608,13 +706,24 @@ static void intel_psr_work(struct work_struct *work) · 44 unmodified lines | ||
| 608 | 706 | * and be ready for re-enable. |
| 609 | 707 | */ |
| 610 | 708 | if (HAS_DDI(dev_priv)) { |
| 611 | − if (intel_wait_for_register(dev_priv, | |
| 612 | − EDP_PSR_STATUS_CTL, | |
| 613 | − EDP_PSR_STATUS_STATE_MASK, | |
| 614 | − 0, | |
| 615 | − 50)) { | |
| 616 | − DRM_ERROR("Timed out waiting for PSR Idle for re-enable\n"); | |
| 617 | − return; | |
| 709 | + if (dev_priv->psr.psr2_support) { | |
| 710 | + if (intel_wait_for_register(dev_priv, | |
| 711 | + EDP_PSR2_STATUS_CTL, | |
| 712 | + EDP_PSR2_STATUS_STATE_MASK, | |
| 713 | + 0, | |
| 714 | + 50)) { | |
| 715 | + DRM_ERROR("Timed out waiting for PSR2 Idle for re-enable\n"); | |
| 716 | + return; | |
| 717 | + } | |
| 718 | + } else { | |
| 719 | + if (intel_wait_for_register(dev_priv, | |
| 720 | + EDP_PSR_STATUS_CTL, | |
| 721 | + EDP_PSR_STATUS_STATE_MASK, | |
| 722 | + 0, | |
| 723 | + 50)) { | |
| 724 | + DRM_ERROR("Timed out waiting for PSR Idle for re-enable\n"); | |
| 725 | + return; | |
| 726 | + } | |
| 618 | 727 | } |
| 619 | 728 | } else { |
| 620 | 729 | if (intel_wait_for_register(dev_priv, |
| @@ -656,11 +765,19 @@ static void intel_psr_exit(struct drm_i915_private *dev_priv) · 35 unmodified lines | ||
| 656 | 765 | return; |
| 657 | 766 | |
| 658 | 767 | if (HAS_DDI(dev_priv)) { |
| 659 | − val = I915_READ(EDP_PSR_CTL); | |
| 660 | − | |
| 661 | − WARN_ON(!(val & EDP_PSR_ENABLE)); | |
| 662 | − | |
| 663 | − I915_WRITE(EDP_PSR_CTL, val & ~EDP_PSR_ENABLE); | |
| 768 | + if (dev_priv->psr.aux_frame_sync) | |
| 769 | + drm_dp_dpcd_writeb(&intel_dp->aux, | |
| 770 | + DP_SINK_DEVICE_AUX_FRAME_SYNC_CONF, | |
| 771 | + 0); | |
| 772 | + if (dev_priv->psr.psr2_support) { | |
| 773 | + val = I915_READ(EDP_PSR2_CTL); | |
| 774 | + WARN_ON(!(val & EDP_PSR2_ENABLE)); | |
| 775 | + I915_WRITE(EDP_PSR2_CTL, val & ~EDP_PSR2_ENABLE); | |
| 776 | + } else { | |
| 777 | + val = I915_READ(EDP_PSR_CTL); | |
| 778 | + WARN_ON(!(val & EDP_PSR_ENABLE)); | |
| 779 | + I915_WRITE(EDP_PSR_CTL, val & ~EDP_PSR_ENABLE); | |
| 780 | + } | |
| 664 | 781 | } else { |
| 665 | 782 | val = I915_READ(VLV_PSRCTL(pipe)); |
| 666 | 783 | |
| @@ -1728,7 +1728,7 @@ static int init_status_page(struct intel_engine_cs *engine) · 1727 unmodified lines | ||
| 1728 | 1728 | void *vaddr; |
| 1729 | 1729 | int ret; |
| 1730 | 1730 | |
| 1731 | − obj = i915_gem_object_create_internal(engine->i915, 4096); | |
| 1731 | + obj = i915_gem_object_create_internal(engine->i915, PAGE_SIZE); | |
| 1732 | 1732 | if (IS_ERR(obj)) { |
| 1733 | 1733 | DRM_ERROR("Failed to allocate status page\n"); |
| 1734 | 1734 | return PTR_ERR(obj); |
| @@ -1738,7 +1738,7 @@ static int init_status_page(struct intel_engine_cs *engine) · 3 unmodified lines | ||
| 1738 | 1738 | if (ret) |
| 1739 | 1739 | goto err; |
| 1740 | 1740 | |
| 1741 | − vma = i915_vma_create(obj, &engine->i915->ggtt.base, NULL); | |
| 1741 | + vma = i915_vma_instance(obj, &engine->i915->ggtt.base, NULL); | |
| 1742 | 1742 | if (IS_ERR(vma)) { |
| 1743 | 1743 | ret = PTR_ERR(vma); |
| 1744 | 1744 | goto err; |
| @@ -1769,7 +1769,7 @@ static int init_status_page(struct intel_engine_cs *engine) · 24 unmodified lines | ||
| 1769 | 1769 | |
| 1770 | 1770 | engine->status_page.vma = vma; |
| 1771 | 1771 | engine->status_page.ggtt_offset = i915_ggtt_offset(vma); |
| 1772 | − engine->status_page.page_addr = memset(vaddr, 0, 4096); | |
| 1772 | + engine->status_page.page_addr = memset(vaddr, 0, PAGE_SIZE); | |
| 1773 | 1773 | |
| 1774 | 1774 | DRM_DEBUG_DRIVER("%s hws offset: 0x%08x\n", |
| 1775 | 1775 | engine->name, i915_ggtt_offset(vma)); |
| @@ -1872,7 +1872,7 @@ intel_ring_create_vma(struct drm_i915_private *dev_priv, int size) · 96 unmodified lines | ||
| 1872 | 1872 | /* mark ring buffers as read-only from GPU side by default */ |
| 1873 | 1873 | obj->gt_ro = 1; |
| 1874 | 1874 | |
| 1875 | − vma = i915_vma_create(obj, &dev_priv->ggtt.base, NULL); | |
| 1875 | + vma = i915_vma_instance(obj, &dev_priv->ggtt.base, NULL); | |
| 1876 | 1876 | if (IS_ERR(vma)) |
| 1877 | 1877 | goto err; |
| 1878 | 1878 | |
| @@ -2041,7 +2041,7 @@ static int intel_init_ring_buffer(struct intel_engine_cs *engine) · 162 unmodified lines | ||
| 2041 | 2041 | } |
| 2042 | 2042 | |
| 2043 | 2043 | /* Ring wraparound at offset 0 sometimes hangs. No idea why. */ |
| 2044 | − ret = intel_ring_pin(ring, 4096); | |
| 2044 | + ret = intel_ring_pin(ring, I915_GTT_PAGE_SIZE); | |
| 2045 | 2045 | if (ret) { |
| 2046 | 2046 | intel_ring_free(ring); |
| 2047 | 2047 | goto error; |
| @@ -2458,11 +2458,11 @@ static void intel_ring_init_semaphores(struct drm_i915_private *dev_priv, · 410 unmodified lines | ||
| 2458 | 2458 | if (INTEL_GEN(dev_priv) >= 8 && !dev_priv->semaphore) { |
| 2459 | 2459 | struct i915_vma *vma; |
| 2460 | 2460 | |
| 2461 | − obj = i915_gem_object_create(dev_priv, 4096); | |
| 2461 | + obj = i915_gem_object_create(dev_priv, PAGE_SIZE); | |
| 2462 | 2462 | if (IS_ERR(obj)) |
| 2463 | 2463 | goto err; |
| 2464 | 2464 | |
| 2465 | − vma = i915_vma_create(obj, &dev_priv->ggtt.base, NULL); | |
| 2465 | + vma = i915_vma_instance(obj, &dev_priv->ggtt.base, NULL); | |
| 2466 | 2466 | if (IS_ERR(vma)) |
| 2467 | 2467 | goto err_obj; |
| 2468 | 2468 | |
| @@ -2675,7 +2675,7 @@ int intel_init_render_ring_buffer(struct intel_engine_cs *engine) · 206 unmodified lines | ||
| 2675 | 2675 | return ret; |
| 2676 | 2676 | |
| 2677 | 2677 | if (INTEL_GEN(dev_priv) >= 6) { |
| 2678 | − ret = intel_engine_create_scratch(engine, 4096); | |
| 2678 | + ret = intel_engine_create_scratch(engine, PAGE_SIZE); | |
| 2679 | 2679 | if (ret) |
| 2680 | 2680 | return ret; |
| 2681 | 2681 | } else if (HAS_BROKEN_CS_TLB(dev_priv)) { |
| @@ -273,7 +273,7 @@ skl_update_plane(struct drm_plane *drm_plane, · 272 unmodified lines | ||
| 273 | 273 | |
| 274 | 274 | I915_WRITE(PLANE_CTL(pipe, plane_id), plane_ctl); |
| 275 | 275 | I915_WRITE(PLANE_SURF(pipe, plane_id), |
| 276 | − intel_fb_gtt_offset(fb, rotation) + surf_addr); | |
| 276 | + intel_plane_ggtt_offset(plane_state) + surf_addr); | |
| 277 | 277 | POSTING_READ(PLANE_SURF(pipe, plane_id)); |
| 278 | 278 | } |
| 279 | 279 | |
| @@ -458,7 +458,7 @@ vlv_update_plane(struct drm_plane *dplane, · 178 unmodified lines | ||
| 458 | 458 | I915_WRITE(SPSIZE(pipe, plane_id), (crtc_h << 16) | crtc_w); |
| 459 | 459 | I915_WRITE(SPCNTR(pipe, plane_id), sprctl); |
| 460 | 460 | I915_WRITE(SPSURF(pipe, plane_id), |
| 461 | − intel_fb_gtt_offset(fb, rotation) + sprsurf_offset); | |
| 461 | + intel_plane_ggtt_offset(plane_state) + sprsurf_offset); | |
| 462 | 462 | POSTING_READ(SPSURF(pipe, plane_id)); |
| 463 | 463 | } |
| 464 | 464 | |
| @@ -594,7 +594,7 @@ ivb_update_plane(struct drm_plane *plane, · 129 unmodified lines | ||
| 594 | 594 | I915_WRITE(SPRSCALE(pipe), sprscale); |
| 595 | 595 | I915_WRITE(SPRCTL(pipe), sprctl); |
| 596 | 596 | I915_WRITE(SPRSURF(pipe), |
| 597 | − intel_fb_gtt_offset(fb, rotation) + sprsurf_offset); | |
| 597 | + intel_plane_ggtt_offset(plane_state) + sprsurf_offset); | |
| 598 | 598 | POSTING_READ(SPRSURF(pipe)); |
| 599 | 599 | } |
| 600 | 600 | |
| @@ -721,7 +721,7 @@ ilk_update_plane(struct drm_plane *plane, · 120 unmodified lines | ||
| 721 | 721 | I915_WRITE(DVSSCALE(pipe), dvsscale); |
| 722 | 722 | I915_WRITE(DVSCNTR(pipe), dvscntr); |
| 723 | 723 | I915_WRITE(DVSSURF(pipe), |
| 724 | − intel_fb_gtt_offset(fb, rotation) + dvssurf_offset); | |
| 724 | + intel_plane_ggtt_offset(plane_state) + dvssurf_offset); | |
| 725 | 725 | POSTING_READ(DVSSURF(pipe)); |
| 726 | 726 | } |
| 727 | 727 | |
| @@ -114,29 +114,3 @@ int intel_guc_sample_forcewake(struct intel_guc *guc) · 113 unmodified lines | ||
| 114 | 114 | return intel_guc_send(guc, action, ARRAY_SIZE(action)); |
| 115 | 115 | } |
| 116 | 116 | |
| 117 | −int intel_guc_log_flush_complete(struct intel_guc *guc) | |
| 118 | −{ | |
| 119 | − u32 action[] = { INTEL_GUC_ACTION_LOG_BUFFER_FILE_FLUSH_COMPLETE }; | |
| 120 | − | |
| 121 | − return intel_guc_send(guc, action, ARRAY_SIZE(action)); | |
| 122 | −} | |
| 123 | − | |
| 124 | −int intel_guc_log_flush(struct intel_guc *guc) | |
| 125 | −{ | |
| 126 | − u32 action[] = { | |
| 127 | − INTEL_GUC_ACTION_FORCE_LOG_BUFFER_FLUSH, | |
| 128 | − 0 | |
| 129 | − }; | |
| 130 | − | |
| 131 | − return intel_guc_send(guc, action, ARRAY_SIZE(action)); | |
| 132 | −} | |
| 133 | − | |
| 134 | −int intel_guc_log_control(struct intel_guc *guc, u32 control_val) | |
| 135 | −{ | |
| 136 | − u32 action[] = { | |
| 137 | − INTEL_GUC_ACTION_UK_LOG_ENABLE_LOGGING, | |
| 138 | − control_val | |
| 139 | − }; | |
| 140 | − | |
| 141 | − return intel_guc_send(guc, action, ARRAY_SIZE(action)); | |
| 142 | −} | |
| @@ -93,29 +93,35 @@ struct i915_guc_client { · 92 unmodified lines | ||
| 93 | 93 | uint64_t submissions[I915_NUM_ENGINES]; |
| 94 | 94 | }; |
| 95 | 95 | |
| 96 | −enum intel_guc_fw_status { | |
| 97 | − GUC_FIRMWARE_FAIL = -1, | |
| 98 | − GUC_FIRMWARE_NONE = 0, | |
| 99 | − GUC_FIRMWARE_PENDING, | |
| 100 | − GUC_FIRMWARE_SUCCESS | |
| 96 | +enum intel_uc_fw_status { | |
| 97 | + INTEL_UC_FIRMWARE_FAIL = -1, | |
| 98 | + INTEL_UC_FIRMWARE_NONE = 0, | |
| 99 | + INTEL_UC_FIRMWARE_PENDING, | |
| 100 | + INTEL_UC_FIRMWARE_SUCCESS | |
| 101 | +}; | |
| 102 | + | |
| 103 | +enum intel_uc_fw_type { | |
| 104 | + INTEL_UC_FW_TYPE_GUC, | |
| 105 | + INTEL_UC_FW_TYPE_HUC | |
| 101 | 106 | }; |
| 102 | 107 | |
| 103 | 108 | /* |
| 104 | 109 | * This structure encapsulates all the data needed during the process |
| 105 | 110 | * of fetching, caching, and loading the firmware image into the GuC. |
| 106 | 111 | */ |
| 107 | −struct intel_guc_fw { | |
| 108 | − const char * guc_fw_path; | |
| 109 | − size_t guc_fw_size; | |
| 110 | − struct drm_i915_gem_object * guc_fw_obj; | |
| 111 | − enum intel_guc_fw_status guc_fw_fetch_status; | |
| 112 | − enum intel_guc_fw_status guc_fw_load_status; | |
| 113 | − | |
| 114 | − uint16_t guc_fw_major_wanted; | |
| 115 | − uint16_t guc_fw_minor_wanted; | |
| 116 | − uint16_t guc_fw_major_found; | |
| 117 | − uint16_t guc_fw_minor_found; | |
| 118 | − | |
| 112 | +struct intel_uc_fw { | |
| 113 | + const char *path; | |
| 114 | + size_t size; | |
| 115 | + struct drm_i915_gem_object *obj; | |
| 116 | + enum intel_uc_fw_status fetch_status; | |
| 117 | + enum intel_uc_fw_status load_status; | |
| 118 | + | |
| 119 | + uint16_t major_ver_wanted; | |
| 120 | + uint16_t minor_ver_wanted; | |
| 121 | + uint16_t major_ver_found; | |
| 122 | + uint16_t minor_ver_found; | |
| 123 | + | |
| 124 | + enum intel_uc_fw_type fw; | |
| 119 | 125 | uint32_t header_size; |
| 120 | 126 | uint32_t header_offset; |
| 121 | 127 | uint32_t rsa_size; |
| @@ -141,7 +147,7 @@ struct intel_guc_log { · 19 unmodified lines | ||
| 141 | 147 | }; |
| 142 | 148 | |
| 143 | 149 | struct intel_guc { |
| 144 | − struct intel_guc_fw guc_fw; | |
| 150 | + struct intel_uc_fw fw; | |
| 145 | 151 | struct intel_guc_log log; |
| 146 | 152 | |
| 147 | 153 | /* intel_guc_recv interrupt related state */ |
| @@ -170,21 +176,28 @@ struct intel_guc { · 22 unmodified lines | ||
| 170 | 176 | struct mutex send_mutex; |
| 171 | 177 | }; |
| 172 | 178 | |
| 179 | +struct intel_huc { | |
| 180 | + /* Generic uC firmware management */ | |
| 181 | + struct intel_uc_fw fw; | |
| 182 | + | |
| 183 | + /* HuC-specific additions */ | |
| 184 | +}; | |
| 185 | + | |
| 173 | 186 | /* intel_uc.c */ |
| 174 | 187 | void intel_uc_init_early(struct drm_i915_private *dev_priv); |
| 175 | 188 | int intel_guc_send(struct intel_guc *guc, const u32 *action, u32 len); |
| 176 | 189 | int intel_guc_sample_forcewake(struct intel_guc *guc); |
| 177 | −int intel_guc_log_flush_complete(struct intel_guc *guc); | |
| 178 | −int intel_guc_log_flush(struct intel_guc *guc); | |
| 179 | −int intel_guc_log_control(struct intel_guc *guc, u32 control_val); | |
| 180 | 190 | |
| 181 | 191 | /* intel_guc_loader.c */ |
| 182 | 192 | extern void intel_guc_init(struct drm_i915_private *dev_priv); |
| 183 | 193 | extern int intel_guc_setup(struct drm_i915_private *dev_priv); |
| 184 | 194 | extern void intel_guc_fini(struct drm_i915_private *dev_priv); |
| 185 | −extern const char *intel_guc_fw_status_repr(enum intel_guc_fw_status status); | |
| 195 | +extern const char *intel_uc_fw_status_repr(enum intel_uc_fw_status status); | |
| 186 | 196 | extern int intel_guc_suspend(struct drm_i915_private *dev_priv); |
| 187 | 197 | extern int intel_guc_resume(struct drm_i915_private *dev_priv); |
| 198 | +void intel_uc_fw_fetch(struct drm_i915_private *dev_priv, | |
| 199 | + struct intel_uc_fw *uc_fw); | |
| 200 | +u32 intel_guc_wopcm_size(struct drm_i915_private *dev_priv); | |
| 188 | 201 | |
| 189 | 202 | /* i915_guc_submission.c */ |
| 190 | 203 | int i915_guc_submission_init(struct drm_i915_private *dev_priv); |
| @@ -193,10 +206,12 @@ int i915_guc_wq_reserve(struct drm_i915_gem_request *rq); · 2 unmodified lines | ||
| 193 | 206 | void i915_guc_wq_unreserve(struct drm_i915_gem_request *request); |
| 194 | 207 | void i915_guc_submission_disable(struct drm_i915_private *dev_priv); |
| 195 | 208 | void i915_guc_submission_fini(struct drm_i915_private *dev_priv); |
| 196 | −void i915_guc_capture_logs(struct drm_i915_private *dev_priv); | |
| 197 | −void i915_guc_flush_logs(struct drm_i915_private *dev_priv); | |
| 198 | −void i915_guc_register(struct drm_i915_private *dev_priv); | |
| 199 | −void i915_guc_unregister(struct drm_i915_private *dev_priv); | |
| 209 | +struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size); | |
| 210 | + | |
| 211 | +/* intel_guc_log.c */ | |
| 212 | +void intel_guc_log_create(struct intel_guc *guc); | |
| 213 | +void i915_guc_log_register(struct drm_i915_private *dev_priv); | |
| 214 | +void i915_guc_log_unregister(struct drm_i915_private *dev_priv); | |
| 200 | 215 | int i915_guc_log_control(struct drm_i915_private *dev_priv, u64 control_val); |
| 201 | 216 | |
| 202 | 217 | static inline u32 guc_ggtt_offset(struct i915_vma *vma) |
| @@ -207,4 +222,10 @@ static inline u32 guc_ggtt_offset(struct i915_vma *vma) · 4 unmodified lines | ||
| 207 | 222 | return offset; |
| 208 | 223 | } |
| 209 | 224 | |
| 225 | +/* intel_huc.c */ | |
| 226 | +void intel_huc_init(struct drm_i915_private *dev_priv); | |
| 227 | +void intel_huc_fini(struct drm_i915_private *dev_priv); | |
| 228 | +int intel_huc_load(struct drm_i915_private *dev_priv); | |
| 229 | +void intel_guc_auth_huc(struct drm_i915_private *dev_priv); | |
| 230 | + | |
| 210 | 231 | #endif |
| @@ -194,7 +194,8 @@ · 193 unmodified lines | ||
| 194 | 194 | # define DP_PSR_SETUP_TIME_0 (6 << 1) |
| 195 | 195 | # define DP_PSR_SETUP_TIME_MASK (7 << 1) |
| 196 | 196 | # define DP_PSR_SETUP_TIME_SHIFT 1 |
| 197 | − | |
| 197 | +# define DP_PSR2_SU_Y_COORDINATE_REQUIRED (1 << 4) /* eDP 1.4a */ | |
| 198 | +# define DP_PSR2_SU_GRANULARITY_REQUIRED (1 << 5) /* eDP 1.4b */ | |
| 198 | 199 | /* |
| 199 | 200 | * 0x80-0x8f describe downstream port capabilities, but there are two layouts |
| 200 | 201 | * based on whether DP_DETAILED_CAP_INFO_AVAILABLE was set. If it was not, |
| @@ -568,6 +569,16 @@ · 367 unmodified lines | ||
| 568 | 569 | #define DP_RECEIVER_ALPM_STATUS 0x200b /* eDP 1.4 */ |
| 569 | 570 | # define DP_ALPM_LOCK_TIMEOUT_ERROR (1 << 0) |
| 570 | 571 | |
| 572 | +#define DP_DPRX_FEATURE_ENUMERATION_LIST 0x2210 /* DP 1.3 */ | |
| 573 | +# define DP_GTC_CAP (1 << 0) /* DP 1.3 */ | |
| 574 | +# define DP_SST_SPLIT_SDP_CAP (1 << 1) /* DP 1.4 */ | |
| 575 | +# define DP_AV_SYNC_CAP (1 << 2) /* DP 1.3 */ | |
| 576 | +# define DP_VSC_SDP_EXT_FOR_COLORIMETRY_SUPPORTED (1 << 3) /* DP 1.3 */ | |
| 577 | +# define DP_VSC_EXT_VESA_SDP_SUPPORTED (1 << 4) /* DP 1.4 */ | |
| 578 | +# define DP_VSC_EXT_VESA_SDP_CHAINING_SUPPORTED (1 << 5) /* DP 1.4 */ | |
| 579 | +# define DP_VSC_EXT_CEA_SDP_SUPPORTED (1 << 6) /* DP 1.4 */ | |
| 580 | +# define DP_VSC_EXT_CEA_SDP_CHAINING_SUPPORTED (1 << 7) /* DP 1.4 */ | |
| 581 | + | |
| 571 | 582 | /* DP 1.2 Sideband message defines */ |
| 572 | 583 | /* peer device type - DP 1.2a Table 2-92 */ |
| 573 | 584 | #define DP_PEER_DEVICE_NONE 0x0 |
| @@ -395,6 +395,7 @@ typedef struct drm_i915_irq_wait { · 394 unmodified lines | ||
| 395 | 395 | * priorities and the driver will attempt to execute batches in priority order. |
| 396 | 396 | */ |
| 397 | 397 | #define I915_PARAM_HAS_SCHEDULER 41 |
| 398 | +#define I915_PARAM_HUC_STATUS 42 | |
| 398 | 399 | |
| 399 | 400 | typedef struct drm_i915_getparam { |
| 400 | 401 | __s32 param; |
| @@ -311,9 +311,15 @@ int snd_hda_get_conn_index(struct hda_codec *codec, hda_nid_t mux, · 310 unmodified lines | ||
| 311 | 311 | } |
| 312 | 312 | EXPORT_SYMBOL_GPL(snd_hda_get_conn_index); |
| 313 | 313 | |
| 314 | − | |
| 315 | −/* return DEVLIST_LEN parameter of the given widget */ | |
| 316 | −static unsigned int get_num_devices(struct hda_codec *codec, hda_nid_t nid) | |
| 314 | +/** | |
| 315 | + * snd_hda_get_num_devices - get DEVLIST_LEN parameter of the given widget | |
| 316 | + * @codec: the HDA codec | |
| 317 | + * @nid: NID of the pin to parse | |
| 318 | + * | |
| 319 | + * Get the device entry number on the given widget. This is a feature of | |
| 320 | + * DP MST audio. Each pin can have several device entries in it. | |
| 321 | + */ | |
| 322 | +unsigned int snd_hda_get_num_devices(struct hda_codec *codec, hda_nid_t nid) | |
| 317 | 323 | { |
| 318 | 324 | unsigned int wcaps = get_wcaps(codec, nid); |
| 319 | 325 | unsigned int parm; |
| @@ -327,6 +333,7 @@ static unsigned int get_num_devices(struct hda_codec *codec, hda_nid_t nid) · 7 unmodified lines | ||
| 327 | 333 | parm = 0; |
| 328 | 334 | return parm & AC_DEV_LIST_LEN_MASK; |
| 329 | 335 | } |
| 336 | +EXPORT_SYMBOL_GPL(snd_hda_get_num_devices); | |
| 330 | 337 | |
| 331 | 338 | /** |
| 332 | 339 | * snd_hda_get_devices - copy device list without cache |
| @@ -344,7 +351,7 @@ int snd_hda_get_devices(struct hda_codec *codec, hda_nid_t nid, · 11 unmodified lines | ||
| 344 | 351 | unsigned int parm; |
| 345 | 352 | int i, dev_len, devices; |
| 346 | 353 | |
| 347 | − parm = get_num_devices(codec, nid); | |
| 354 | + parm = snd_hda_get_num_devices(codec, nid); | |
| 348 | 355 | if (!parm) /* not multi-stream capable */ |
| 349 | 356 | return 0; |
| 350 | 357 | |
| @@ -368,6 +375,63 @@ int snd_hda_get_devices(struct hda_codec *codec, hda_nid_t nid, · 17 unmodified lines | ||
| 368 | 375 | return devices; |
| 369 | 376 | } |
| 370 | 377 | |
| 378 | +/** | |
| 379 | + * snd_hda_get_dev_select - get device entry select on the pin | |
| 380 | + * @codec: the HDA codec | |
| 381 | + * @nid: NID of the pin to get device entry select | |
| 382 | + * | |
| 383 | + * Get the devcie entry select on the pin. Return the device entry | |
| 384 | + * id selected on the pin. Return 0 means the first device entry | |
| 385 | + * is selected or MST is not supported. | |
| 386 | + */ | |
| 387 | +int snd_hda_get_dev_select(struct hda_codec *codec, hda_nid_t nid) | |
| 388 | +{ | |
| 389 | + /* not support dp_mst will always return 0, using first dev_entry */ | |
| 390 | + if (!codec->dp_mst) | |
| 391 | + return 0; | |
| 392 | + | |
| 393 | + return snd_hda_codec_read(codec, nid, 0, AC_VERB_GET_DEVICE_SEL, 0); | |
| 394 | +} | |
| 395 | +EXPORT_SYMBOL_GPL(snd_hda_get_dev_select); | |
| 396 | + | |
| 397 | +/** | |
| 398 | + * snd_hda_set_dev_select - set device entry select on the pin | |
| 399 | + * @codec: the HDA codec | |
| 400 | + * @nid: NID of the pin to set device entry select | |
| 401 | + * @dev_id: device entry id to be set | |
| 402 | + * | |
| 403 | + * Set the device entry select on the pin nid. | |
| 404 | + */ | |
| 405 | +int snd_hda_set_dev_select(struct hda_codec *codec, hda_nid_t nid, int dev_id) | |
| 406 | +{ | |
| 407 | + int ret, num_devices; | |
| 408 | + | |
| 409 | + /* not support dp_mst will always return 0, using first dev_entry */ | |
| 410 | + if (!codec->dp_mst) | |
| 411 | + return 0; | |
| 412 | + | |
| 413 | + /* AC_PAR_DEVLIST_LEN is 0 based. */ | |
| 414 | + num_devices = snd_hda_get_num_devices(codec, nid) + 1; | |
| 415 | + /* If Device List Length is 0 (num_device = 1), | |
| 416 | + * the pin is not multi stream capable. | |
| 417 | + * Do nothing in this case. | |
| 418 | + */ | |
| 419 | + if (num_devices == 1) | |
| 420 | + return 0; | |
| 421 | + | |
| 422 | + /* Behavior of setting index being equal to or greater than | |
| 423 | + * Device List Length is not predictable | |
| 424 | + */ | |
| 425 | + if (num_devices <= dev_id) | |
| 426 | + return -EINVAL; | |
| 427 | + | |
| 428 | + ret = snd_hda_codec_write(codec, nid, 0, | |
| 429 | + AC_VERB_SET_DEVICE_SEL, dev_id); | |
| 430 | + | |
| 431 | + return ret; | |
| 432 | +} | |
| 433 | +EXPORT_SYMBOL_GPL(snd_hda_set_dev_select); | |
| 434 | + | |
| 371 | 435 | /* |
| 372 | 436 | * read widget caps for each widget and store in cache |
| 373 | 437 | */ |
| @@ -403,6 +467,10 @@ static int read_pin_defaults(struct hda_codec *codec) · 29 unmodified lines | ||
| 403 | 467 | pin->nid = nid; |
| 404 | 468 | pin->cfg = snd_hda_codec_read(codec, nid, 0, |
| 405 | 469 | AC_VERB_GET_CONFIG_DEFAULT, 0); |
| 470 | + /* | |
| 471 | + * all device entries are the same widget control so far | |
| 472 | + * fixme: if any codec is different, need fix here | |
| 473 | + */ | |
| 406 | 474 | pin->ctrl = snd_hda_codec_read(codec, nid, 0, |
| 407 | 475 | AC_VERB_GET_PIN_WIDGET_CONTROL, |
| 408 | 476 | 0); |
| @@ -347,8 +347,11 @@ int snd_hda_override_conn_list(struct hda_codec *codec, hda_nid_t nid, int nums, · 346 unmodified lines | ||
| 347 | 347 | const hda_nid_t *list); |
| 348 | 348 | int snd_hda_get_conn_index(struct hda_codec *codec, hda_nid_t mux, |
| 349 | 349 | hda_nid_t nid, int recursive); |
| 350 | +unsigned int snd_hda_get_num_devices(struct hda_codec *codec, hda_nid_t nid); | |
| 350 | 351 | int snd_hda_get_devices(struct hda_codec *codec, hda_nid_t nid, |
| 351 | 352 | u8 *dev_list, int max_devices); |
| 353 | +int snd_hda_get_dev_select(struct hda_codec *codec, hda_nid_t nid); | |
| 354 | +int snd_hda_set_dev_select(struct hda_codec *codec, hda_nid_t nid, int dev_id); | |
| 352 | 355 | |
| 353 | 356 | struct hda_verb { |
| 354 | 357 | hda_nid_t nid; |
| @@ -76,6 +76,7 @@ struct hdmi_spec_per_cvt { · 75 unmodified lines | ||
| 76 | 76 | |
| 77 | 77 | struct hdmi_spec_per_pin { |
| 78 | 78 | hda_nid_t pin_nid; |
| 79 | + int dev_id; | |
| 79 | 80 | /* pin idx, different device entries on the same pin use the same idx */ |
| 80 | 81 | int pin_nid_idx; |
| 81 | 82 | int num_mux_nids; |
| @@ -130,7 +131,23 @@ struct hdmi_spec { · 48 unmodified lines | ||
| 130 | 131 | struct snd_array cvts; /* struct hdmi_spec_per_cvt */ |
| 131 | 132 | hda_nid_t cvt_nids[4]; /* only for haswell fix */ |
| 132 | 133 | |
| 134 | + /* | |
| 135 | + * num_pins is the number of virtual pins | |
| 136 | + * for example, there are 3 pins, and each pin | |
| 137 | + * has 4 device entries, then the num_pins is 12 | |
| 138 | + */ | |
| 133 | 139 | int num_pins; |
| 140 | + /* | |
| 141 | + * num_nids is the number of real pins | |
| 142 | + * In the above example, num_nids is 3 | |
| 143 | + */ | |
| 144 | + int num_nids; | |
| 145 | + /* | |
| 146 | + * dev_num is the number of device entries | |
| 147 | + * on each pin. | |
| 148 | + * In the above example, dev_num is 4 | |
| 149 | + */ | |
| 150 | + int dev_num; | |
| 134 | 151 | struct snd_array pins; /* struct hdmi_spec_per_pin */ |
| 135 | 152 | struct hdmi_pcm pcm_rec[16]; |
| 136 | 153 | struct mutex pcm_lock; |
| @@ -217,14 +234,26 @@ union audio_infoframe { · 80 unmodified lines | ||
| 217 | 234 | /* obtain hda_pcm object assigned to idx */ |
| 218 | 235 | #define get_pcm_rec(spec, idx) (get_hdmi_pcm(spec, idx)->pcm) |
| 219 | 236 | |
| 220 | −static int pin_nid_to_pin_index(struct hda_codec *codec, hda_nid_t pin_nid) | |
| 237 | +static int pin_id_to_pin_index(struct hda_codec *codec, | |
| 238 | + hda_nid_t pin_nid, int dev_id) | |
| 221 | 239 | { |
| 222 | 240 | struct hdmi_spec *spec = codec->spec; |
| 223 | 241 | int pin_idx; |
| 242 | + struct hdmi_spec_per_pin *per_pin; | |
| 243 | + | |
| 244 | + /* | |
| 245 | + * (dev_id == -1) means it is NON-MST pin | |
| 246 | + * return the first virtual pin on this port | |
| 247 | + */ | |
| 248 | + if (dev_id == -1) | |
| 249 | + dev_id = 0; | |
| 224 | 250 | |
| 225 | − for (pin_idx = 0; pin_idx < spec->num_pins; pin_idx++) | |
| 226 | − if (get_pin(spec, pin_idx)->pin_nid == pin_nid) | |
| 251 | + for (pin_idx = 0; pin_idx < spec->num_pins; pin_idx++) { | |
| 252 | + per_pin = get_pin(spec, pin_idx); | |
| 253 | + if ((per_pin->pin_nid == pin_nid) && | |
| 254 | + (per_pin->dev_id == dev_id)) | |
| 227 | 255 | return pin_idx; |
| 256 | + } | |
| 228 | 257 | |
| 229 | 258 | codec_warn(codec, "HDMI: pin nid %d not registered\n", pin_nid); |
| 230 | 259 | return -EINVAL; |
| @@ -724,10 +753,11 @@ static void hdmi_setup_audio_infoframe(struct hda_codec *codec, · 493 unmodified lines | ||
| 724 | 753 | |
| 725 | 754 | static bool hdmi_present_sense(struct hdmi_spec_per_pin *per_pin, int repoll); |
| 726 | 755 | |
| 727 | −static void check_presence_and_report(struct hda_codec *codec, hda_nid_t nid) | |
| 756 | +static void check_presence_and_report(struct hda_codec *codec, hda_nid_t nid, | |
| 757 | + int dev_id) | |
| 728 | 758 | { |
| 729 | 759 | struct hdmi_spec *spec = codec->spec; |
| 730 | − int pin_idx = pin_nid_to_pin_index(codec, nid); | |
| 760 | + int pin_idx = pin_id_to_pin_index(codec, nid, dev_id); | |
| 731 | 761 | |
| 732 | 762 | if (pin_idx < 0) |
| 733 | 763 | return; |
| @@ -738,7 +768,8 @@ static void check_presence_and_report(struct hda_codec *codec, hda_nid_t nid) · 4 unmodified lines | ||
| 738 | 768 | static void jack_callback(struct hda_codec *codec, |
| 739 | 769 | struct hda_jack_callback *jack) |
| 740 | 770 | { |
| 741 | − check_presence_and_report(codec, jack->nid); | |
| 771 | + /* hda_jack don't support DP MST */ | |
| 772 | + check_presence_and_report(codec, jack->nid, 0); | |
| 742 | 773 | } |
| 743 | 774 | |
| 744 | 775 | static void hdmi_intrinsic_event(struct hda_codec *codec, unsigned int res) |
| @@ -747,6 +778,12 @@ static void hdmi_intrinsic_event(struct hda_codec *codec, unsigned int res) · 2 unmodified lines | ||
| 747 | 778 | struct hda_jack_tbl *jack; |
| 748 | 779 | int dev_entry = (res & AC_UNSOL_RES_DE) >> AC_UNSOL_RES_DE_SHIFT; |
| 749 | 780 | |
| 781 | + /* | |
| 782 | + * assume DP MST uses dyn_pcm_assign and acomp and | |
| 783 | + * never comes here | |
| 784 | + * if DP MST supports unsol event, below code need | |
| 785 | + * consider dev_entry | |
| 786 | + */ | |
| 750 | 787 | jack = snd_hda_jack_tbl_get_from_tag(codec, tag); |
| 751 | 788 | if (!jack) |
| 752 | 789 | return; |
| @@ -757,7 +794,8 @@ static void hdmi_intrinsic_event(struct hda_codec *codec, unsigned int res) · 4 unmodified lines | ||
| 757 | 794 | codec->addr, jack->nid, dev_entry, !!(res & AC_UNSOL_RES_IA), |
| 758 | 795 | !!(res & AC_UNSOL_RES_PD), !!(res & AC_UNSOL_RES_ELDV)); |
| 759 | 796 | |
| 760 | − check_presence_and_report(codec, jack->nid); | |
| 797 | + /* hda_jack don't support DP MST */ | |
| 798 | + check_presence_and_report(codec, jack->nid, 0); | |
| 761 | 799 | } |
| 762 | 800 | |
| 763 | 801 | static void hdmi_non_intrinsic_event(struct hda_codec *codec, unsigned int res) |
| @@ -970,28 +1008,60 @@ static int intel_cvt_id_to_mux_idx(struct hdmi_spec *spec, · 206 unmodified lines | ||
| 970 | 1008 | * by any other pins. |
| 971 | 1009 | */ |
| 972 | 1010 | static void intel_not_share_assigned_cvt(struct hda_codec *codec, |
| 973 | − hda_nid_t pin_nid, int mux_idx) | |
| 1011 | + hda_nid_t pin_nid, | |
| 1012 | + int dev_id, int mux_idx) | |
| 974 | 1013 | { |
| 975 | 1014 | struct hdmi_spec *spec = codec->spec; |
| 976 | 1015 | hda_nid_t nid; |
| 977 | 1016 | int cvt_idx, curr; |
| 978 | 1017 | struct hdmi_spec_per_cvt *per_cvt; |
| 1018 | + struct hdmi_spec_per_pin *per_pin; | |
| 1019 | + int pin_idx; | |
| 1020 | + | |
| 1021 | + /* configure the pins connections */ | |
| 1022 | + for (pin_idx = 0; pin_idx < spec->num_pins; pin_idx++) { | |
| 1023 | + int dev_id_saved; | |
| 1024 | + int dev_num; | |
| 979 | 1025 | |
| 980 | − /* configure all pins, including "no physical connection" ones */ | |
| 981 | − for_each_hda_codec_node(nid, codec) { | |
| 982 | − unsigned int wid_caps = get_wcaps(codec, nid); | |
| 983 | − unsigned int wid_type = get_wcaps_type(wid_caps); | |
| 1026 | + per_pin = get_pin(spec, pin_idx); | |
| 1027 | + /* | |
| 1028 | + * pin not connected to monitor | |
| 1029 | + * no need to operate on it | |
| 1030 | + */ | |
| 1031 | + if (!per_pin->pcm) | |
| 1032 | + continue; | |
| 984 | 1033 | |
| 985 | − if (wid_type != AC_WID_PIN) | |
| 1034 | + if ((per_pin->pin_nid == pin_nid) && | |
| 1035 | + (per_pin->dev_id == dev_id)) | |
| 986 | 1036 | continue; |
| 987 | 1037 | |
| 988 | − if (nid == pin_nid) | |
| 1038 | + /* | |
| 1039 | + * if per_pin->dev_id >= dev_num, | |
| 1040 | + * snd_hda_get_dev_select() will fail, | |
| 1041 | + * and the following operation is unpredictable. | |
| 1042 | + * So skip this situation. | |
| 1043 | + */ | |
| 1044 | + dev_num = snd_hda_get_num_devices(codec, per_pin->pin_nid) + 1; | |
| 1045 | + if (per_pin->dev_id >= dev_num) | |
| 989 | 1046 | continue; |
| 990 | 1047 | |
| 1048 | + nid = per_pin->pin_nid; | |
| 1049 | + | |
| 1050 | + /* | |
| 1051 | + * Calling this function should not impact | |
| 1052 | + * on the device entry selection | |
| 1053 | + * So let's save the dev id for each pin, | |
| 1054 | + * and restore it when return | |
| 1055 | + */ | |
| 1056 | + dev_id_saved = snd_hda_get_dev_select(codec, nid); | |
| 1057 | + snd_hda_set_dev_select(codec, nid, per_pin->dev_id); | |
| 991 | 1058 | curr = snd_hda_codec_read(codec, nid, 0, |
| 992 | 1059 | AC_VERB_GET_CONNECT_SEL, 0); |
| 993 | − if (curr != mux_idx) | |
| 1060 | + if (curr != mux_idx) { | |
| 1061 | + snd_hda_set_dev_select(codec, nid, dev_id_saved); | |
| 994 | 1062 | continue; |
| 1063 | + } | |
| 1064 | + | |
| 995 | 1065 | |
| 996 | 1066 | /* choose an unassigned converter. The conveters in the |
| 997 | 1067 | * connection list are in the same order as in the codec. |
| @@ -1008,12 +1078,13 @@ static void intel_not_share_assigned_cvt(struct hda_codec *codec, · 10 unmodified lines | ||
| 1008 | 1078 | break; |
| 1009 | 1079 | } |
| 1010 | 1080 | } |
| 1081 | + snd_hda_set_dev_select(codec, nid, dev_id_saved); | |
| 1011 | 1082 | } |
| 1012 | 1083 | } |
| 1013 | 1084 | |
| 1014 | 1085 | /* A wrapper of intel_not_share_asigned_cvt() */ |
| 1015 | 1086 | static void intel_not_share_assigned_cvt_nid(struct hda_codec *codec, |
| 1016 | − hda_nid_t pin_nid, hda_nid_t cvt_nid) | |
| 1087 | + hda_nid_t pin_nid, int dev_id, hda_nid_t cvt_nid) | |
| 1017 | 1088 | { |
| 1018 | 1089 | int mux_idx; |
| 1019 | 1090 | struct hdmi_spec *spec = codec->spec; |
| @@ -1025,7 +1096,7 @@ static void intel_not_share_assigned_cvt_nid(struct hda_codec *codec, · 5 unmodified lines | ||
| 1025 | 1096 | */ |
| 1026 | 1097 | mux_idx = intel_cvt_id_to_mux_idx(spec, cvt_nid); |
| 1027 | 1098 | if (mux_idx >= 0) |
| 1028 | − intel_not_share_assigned_cvt(codec, pin_nid, mux_idx); | |
| 1099 | + intel_not_share_assigned_cvt(codec, pin_nid, dev_id, mux_idx); | |
| 1029 | 1100 | } |
| 1030 | 1101 | |
| 1031 | 1102 | /* skeleton caller of pin_cvt_fixup ops */ |
| @@ -1140,6 +1211,7 @@ static int hdmi_pcm_open(struct hda_pcm_stream *hinfo, · 108 unmodified lines | ||
| 1140 | 1211 | per_pin->cvt_nid = per_cvt->cvt_nid; |
| 1141 | 1212 | hinfo->nid = per_cvt->cvt_nid; |
| 1142 | 1213 | |
| 1214 | + snd_hda_set_dev_select(codec, per_pin->pin_nid, per_pin->dev_id); | |
| 1143 | 1215 | snd_hda_codec_write_cache(codec, per_pin->pin_nid, 0, |
| 1144 | 1216 | AC_VERB_SET_CONNECT_SEL, |
| 1145 | 1217 | per_pin->mux_idx); |
| @@ -1198,6 +1270,7 @@ static int hdmi_read_pin_conn(struct hda_codec *codec, int pin_idx) · 52 unmodified lines | ||
| 1198 | 1270 | return -EINVAL; |
| 1199 | 1271 | } |
| 1200 | 1272 | |
| 1273 | + /* all the device entries on the same pin have the same conn list */ | |
| 1201 | 1274 | per_pin->num_mux_nids = snd_hda_get_connections(codec, pin_nid, |
| 1202 | 1275 | per_pin->mux_nids, |
| 1203 | 1276 | HDA_MAX_CONNECTIONS); |
| @@ -1215,13 +1288,13 @@ static int hdmi_find_pcm_slot(struct hdmi_spec *spec, · 11 unmodified lines | ||
| 1215 | 1288 | return per_pin->pin_nid_idx; |
| 1216 | 1289 | |
| 1217 | 1290 | /* have a second try; check the "reserved area" over num_pins */ |
| 1218 | − for (i = spec->num_pins; i < spec->pcm_used; i++) { | |
| 1291 | + for (i = spec->num_nids; i < spec->pcm_used; i++) { | |
| 1219 | 1292 | if (!test_bit(i, &spec->pcm_bitmap)) |
| 1220 | 1293 | return i; |
| 1221 | 1294 | } |
| 1222 | 1295 | |
| 1223 | 1296 | /* the last try; check the empty slots in pins */ |
| 1224 | − for (i = 0; i < spec->num_pins; i++) { | |
| 1297 | + for (i = 0; i < spec->num_nids; i++) { | |
| 1225 | 1298 | if (!test_bit(i, &spec->pcm_bitmap)) |
| 1226 | 1299 | return i; |
| 1227 | 1300 | } |
| @@ -1296,10 +1369,13 @@ static void hdmi_pcm_setup_pin(struct hdmi_spec *spec, · 68 unmodified lines | ||
| 1296 | 1369 | per_pin->cvt_nid = hinfo->nid; |
| 1297 | 1370 | |
| 1298 | 1371 | mux_idx = hdmi_get_pin_cvt_mux(spec, per_pin, hinfo->nid); |
| 1299 | − if (mux_idx < per_pin->num_mux_nids) | |
| 1372 | + if (mux_idx < per_pin->num_mux_nids) { | |
| 1373 | + snd_hda_set_dev_select(codec, per_pin->pin_nid, | |
| 1374 | + per_pin->dev_id); | |
| 1300 | 1375 | snd_hda_codec_write_cache(codec, per_pin->pin_nid, 0, |
| 1301 | 1376 | AC_VERB_SET_CONNECT_SEL, |
| 1302 | 1377 | mux_idx); |
| 1378 | + } | |
| 1303 | 1379 | snd_hda_spdif_ctls_assign(codec, per_pin->pcm_idx, hinfo->nid); |
| 1304 | 1380 | |
| 1305 | 1381 | non_pcm = check_non_pcm_per_cvt(codec, hinfo->nid); |
| @@ -1467,6 +1543,11 @@ static struct snd_jack *pin_idx_to_jack(struct hda_codec *codec, · 161 unmodified lines | ||
| 1467 | 1543 | if (per_pin->pcm_idx >= 0 && spec->dyn_pcm_assign) |
| 1468 | 1544 | jack = spec->pcm_rec[per_pin->pcm_idx].jack; |
| 1469 | 1545 | else if (!spec->dyn_pcm_assign) { |
| 1546 | + /* | |
| 1547 | + * jack tbl doesn't support DP MST | |
| 1548 | + * DP MST will use dyn_pcm_assign, | |
| 1549 | + * so DP MST will never come here | |
| 1550 | + */ | |
| 1470 | 1551 | jack_tbl = snd_hda_jack_tbl_get(codec, per_pin->pin_nid); |
| 1471 | 1552 | if (jack_tbl) |
| 1472 | 1553 | jack = jack_tbl->jack; |
| @@ -1485,9 +1566,9 @@ static void sync_eld_via_acomp(struct hda_codec *codec, · 12 unmodified lines | ||
| 1485 | 1566 | |
| 1486 | 1567 | mutex_lock(&per_pin->lock); |
| 1487 | 1568 | eld->monitor_present = false; |
| 1488 | − size = snd_hdac_acomp_get_eld(&codec->core, per_pin->pin_nid, -1, | |
| 1489 | − &eld->monitor_present, eld->eld_buffer, | |
| 1490 | − ELD_MAX_SIZE); | |
| 1569 | + size = snd_hdac_acomp_get_eld(&codec->core, per_pin->pin_nid, | |
| 1570 | + per_pin->dev_id, &eld->monitor_present, | |
| 1571 | + eld->eld_buffer, ELD_MAX_SIZE); | |
| 1491 | 1572 | if (size > 0) { |
| 1492 | 1573 | size = min(size, ELD_MAX_SIZE); |
| 1493 | 1574 | if (snd_hdmi_parse_eld(codec, &eld->info, |
| @@ -1565,38 +1646,81 @@ static int hdmi_add_pin(struct hda_codec *codec, hda_nid_t pin_nid) · 71 unmodified lines | ||
| 1565 | 1646 | int pin_idx; |
| 1566 | 1647 | struct hdmi_spec_per_pin *per_pin; |
| 1567 | 1648 | int err; |
| 1649 | + int dev_num, i; | |
| 1568 | 1650 | |
| 1569 | 1651 | caps = snd_hda_query_pin_caps(codec, pin_nid); |
| 1570 | 1652 | if (!(caps & (AC_PINCAP_HDMI | AC_PINCAP_DP))) |
| 1571 | 1653 | return 0; |
| 1572 | 1654 | |
| 1655 | + /* | |
| 1656 | + * For DP MST audio, Configuration Default is the same for | |
| 1657 | + * all device entries on the same pin | |
| 1658 | + */ | |
| 1573 | 1659 | config = snd_hda_codec_get_pincfg(codec, pin_nid); |
| 1574 | 1660 | if (get_defcfg_connect(config) == AC_JACK_PORT_NONE) |
| 1575 | 1661 | return 0; |
| 1576 | 1662 | |
| 1577 | − if (is_haswell_plus(codec)) | |
| 1578 | − intel_haswell_fixup_connect_list(codec, pin_nid); | |
| 1579 | − | |
| 1580 | − pin_idx = spec->num_pins; | |
| 1581 | − per_pin = snd_array_new(&spec->pins); | |
| 1582 | − if (!per_pin) | |
| 1583 | − return -ENOMEM; | |
| 1584 | − | |
| 1585 | − per_pin->pin_nid = pin_nid; | |
| 1586 | − per_pin->non_pcm = false; | |
| 1587 | − if (spec->dyn_pcm_assign) | |
| 1588 | − per_pin->pcm_idx = -1; | |
| 1589 | − else { | |
| 1590 | − per_pin->pcm = get_hdmi_pcm(spec, pin_idx); | |
| 1591 | − per_pin->pcm_idx = pin_idx; | |
| 1663 | + /* | |
| 1664 | + * To simplify the implementation, malloc all | |
| 1665 | + * the virtual pins in the initialization statically | |
| 1666 | + */ | |
| 1667 | + if (is_haswell_plus(codec)) { | |
| 1668 | + /* | |
| 1669 | + * On Intel platforms, device entries number is | |
| 1670 | + * changed dynamically. If there is a DP MST | |
| 1671 | + * hub connected, the device entries number is 3. | |
| 1672 | + * Otherwise, it is 1. | |
| 1673 | + * Here we manually set dev_num to 3, so that | |
| 1674 | + * we can initialize all the device entries when | |
| 1675 | + * bootup statically. | |
| 1676 | + */ | |
| 1677 | + dev_num = 3; | |
| 1678 | + spec->dev_num = 3; | |
| 1679 | + } else if (spec->dyn_pcm_assign && codec->dp_mst) { | |
| 1680 | + dev_num = snd_hda_get_num_devices(codec, pin_nid) + 1; | |
| 1681 | + /* | |
| 1682 | + * spec->dev_num is the maxinum number of device entries | |
| 1683 | + * among all the pins | |
| 1684 | + */ | |
| 1685 | + spec->dev_num = (spec->dev_num > dev_num) ? | |
| 1686 | + spec->dev_num : dev_num; | |
| 1687 | + } else { | |
| 1688 | + /* | |
| 1689 | + * If the platform doesn't support DP MST, | |
| 1690 | + * manually set dev_num to 1. This means | |
| 1691 | + * the pin has only one device entry. | |
| 1692 | + */ | |
| 1693 | + dev_num = 1; | |
| 1694 | + spec->dev_num = 1; | |
| 1592 | 1695 | } |
| 1593 | − per_pin->pin_nid_idx = pin_idx; | |
| 1594 | 1696 | |
| 1595 | − err = hdmi_read_pin_conn(codec, pin_idx); | |
| 1596 | − if (err < 0) | |
| 1597 | − return err; | |
| 1697 | + for (i = 0; i < dev_num; i++) { | |
| 1698 | + pin_idx = spec->num_pins; | |
| 1699 | + per_pin = snd_array_new(&spec->pins); | |
| 1598 | 1700 | |
| 1599 | − spec->num_pins++; | |
| 1701 | + if (!per_pin) | |
| 1702 | + return -ENOMEM; | |
| 1703 | + | |
| 1704 | + if (spec->dyn_pcm_assign) { | |
| 1705 | + per_pin->pcm = NULL; | |
| 1706 | + per_pin->pcm_idx = -1; | |
| 1707 | + } else { | |
| 1708 | + per_pin->pcm = get_hdmi_pcm(spec, pin_idx); | |
| 1709 | + per_pin->pcm_idx = pin_idx; | |
| 1710 | + } | |
| 1711 | + per_pin->pin_nid = pin_nid; | |
| 1712 | + per_pin->pin_nid_idx = spec->num_nids; | |
| 1713 | + per_pin->dev_id = i; | |
| 1714 | + per_pin->non_pcm = false; | |
| 1715 | + snd_hda_set_dev_select(codec, pin_nid, i); | |
| 1716 | + if (is_haswell_plus(codec)) | |
| 1717 | + intel_haswell_fixup_connect_list(codec, pin_nid); | |
| 1718 | + err = hdmi_read_pin_conn(codec, pin_idx); | |
| 1719 | + if (err < 0) | |
| 1720 | + return err; | |
| 1721 | + spec->num_pins++; | |
| 1722 | + } | |
| 1723 | + spec->num_nids++; | |
| 1600 | 1724 | |
| 1601 | 1725 | return 0; |
| 1602 | 1726 | } |
| @@ -1744,7 +1868,7 @@ static int generic_hdmi_playback_pcm_prepare(struct hda_pcm_stream *hinfo, · 141 unmodified lines | ||
| 1744 | 1868 | /* Call sync_audio_rate to set the N/CTS/M manually if necessary */ |
| 1745 | 1869 | /* Todo: add DP1.2 MST audio support later */ |
| 1746 | 1870 | if (codec_has_acomp(codec)) |
| 1747 | − snd_hdac_sync_audio_rate(&codec->core, pin_nid, -1, | |
| 1871 | + snd_hdac_sync_audio_rate(&codec->core, pin_nid, per_pin->dev_id, | |
| 1748 | 1872 | runtime->rate); |
| 1749 | 1873 | |
| 1750 | 1874 | non_pcm = check_non_pcm_per_cvt(codec, cvt_nid); |
| @@ -1762,6 +1886,7 @@ static int generic_hdmi_playback_pcm_prepare(struct hda_pcm_stream *hinfo, · 11 unmodified lines | ||
| 1762 | 1886 | pinctl | PIN_OUT); |
| 1763 | 1887 | } |
| 1764 | 1888 | |
| 1889 | + /* snd_hda_set_dev_select() has been called before */ | |
| 1765 | 1890 | err = spec->ops.setup_stream(codec, cvt_nid, pin_nid, |
| 1766 | 1891 | stream_tag, format); |
| 1767 | 1892 | mutex_unlock(&spec->pcm_lock); |
| @@ -1897,17 +2022,23 @@ static bool is_hdmi_pcm_attached(struct hdac_device *hdac, int pcm_idx) · 129 unmodified lines | ||
| 1897 | 2022 | static int generic_hdmi_build_pcms(struct hda_codec *codec) |
| 1898 | 2023 | { |
| 1899 | 2024 | struct hdmi_spec *spec = codec->spec; |
| 1900 | − int pin_idx; | |
| 2025 | + int idx; | |
| 1901 | 2026 | |
| 1902 | − for (pin_idx = 0; pin_idx < spec->num_pins; pin_idx++) { | |
| 2027 | + /* | |
| 2028 | + * for non-mst mode, pcm number is the same as before | |
| 2029 | + * for DP MST mode, pcm number is (nid number + dev_num - 1) | |
| 2030 | + * dev_num is the device entry number in a pin | |
| 2031 | + * | |
| 2032 | + */ | |
| 2033 | + for (idx = 0; idx < spec->num_nids + spec->dev_num - 1; idx++) { | |
| 1903 | 2034 | struct hda_pcm *info; |
| 1904 | 2035 | struct hda_pcm_stream *pstr; |
| 1905 | 2036 | |
| 1906 | − info = snd_hda_codec_pcm_new(codec, "HDMI %d", pin_idx); | |
| 2037 | + info = snd_hda_codec_pcm_new(codec, "HDMI %d", idx); | |
| 1907 | 2038 | if (!info) |
| 1908 | 2039 | return -ENOMEM; |
| 1909 | 2040 | |
| 1910 | − spec->pcm_rec[pin_idx].pcm = info; | |
| 2041 | + spec->pcm_rec[idx].pcm = info; | |
| 1911 | 2042 | spec->pcm_used++; |
| 1912 | 2043 | info->pcm_type = HDA_PCM_TYPE_HDMI; |
| 1913 | 2044 | info->own_chmap = true; |
| @@ -1915,6 +2046,9 @@ static int generic_hdmi_build_pcms(struct hda_codec *codec) · 1 unmodified line | ||
| 1915 | 2046 | pstr = &info->stream[SNDRV_PCM_STREAM_PLAYBACK]; |
| 1916 | 2047 | pstr->substreams = 1; |
| 1917 | 2048 | pstr->ops = generic_ops; |
| 2049 | + /* pcm number is less than 16 */ | |
| 2050 | + if (spec->pcm_used >= 16) | |
| 2051 | + break; | |
| 1918 | 2052 | /* other pstr fields are set in open */ |
| 1919 | 2053 | } |
| 1920 | 2054 | |
| @@ -2070,7 +2204,9 @@ static int generic_hdmi_init(struct hda_codec *codec) · 149 unmodified lines | ||
| 2070 | 2204 | for (pin_idx = 0; pin_idx < spec->num_pins; pin_idx++) { |
| 2071 | 2205 | struct hdmi_spec_per_pin *per_pin = get_pin(spec, pin_idx); |
| 2072 | 2206 | hda_nid_t pin_nid = per_pin->pin_nid; |
| 2207 | + int dev_id = per_pin->dev_id; | |
| 2073 | 2208 | |
| 2209 | + snd_hda_set_dev_select(codec, pin_nid, dev_id); | |
| 2074 | 2210 | hdmi_init_pin(codec, pin_nid); |
| 2075 | 2211 | if (!codec_has_acomp(codec)) |
| 2076 | 2212 | snd_hda_jack_detect_enable_callback(codec, pin_nid, |
| @@ -2178,6 +2314,7 @@ static int alloc_generic_hdmi(struct hda_codec *codec) · 101 unmodified lines | ||
| 2178 | 2314 | return -ENOMEM; |
| 2179 | 2315 | |
| 2180 | 2316 | spec->ops = generic_standard_hdmi_ops; |
| 2317 | + spec->dev_num = 1; /* initialize to 1 */ | |
| 2181 | 2318 | mutex_init(&spec->pcm_lock); |
| 2182 | 2319 | snd_hdac_register_chmap_ops(&codec->core, &spec->chmap); |
| 2183 | 2320 | |
| @@ -2295,6 +2432,7 @@ static void intel_pin_eld_notify(void *audio_ptr, int port, int pipe) · 111 unmodified lines | ||
| 2295 | 2432 | { |
| 2296 | 2433 | struct hda_codec *codec = audio_ptr; |
| 2297 | 2434 | int pin_nid; |
| 2435 | + int dev_id = pipe; | |
| 2298 | 2436 | |
| 2299 | 2437 | /* we assume only from port-B to port-D */ |
| 2300 | 2438 | if (port < 1 || port > 3) |
| @@ -2321,7 +2459,7 @@ static void intel_pin_eld_notify(void *audio_ptr, int port, int pipe) · 20 unmodified lines | ||
| 2321 | 2459 | return; |
| 2322 | 2460 | |
| 2323 | 2461 | snd_hdac_i915_set_bclk(&codec->bus->core); |
| 2324 | − check_presence_and_report(codec, pin_nid); | |
| 2462 | + check_presence_and_report(codec, pin_nid, dev_id); | |
| 2325 | 2463 | } |
| 2326 | 2464 | |
| 2327 | 2465 | /* register i915 component pin_eld_notify callback */ |
| @@ -2354,11 +2492,13 @@ static void i915_pin_cvt_fixup(struct hda_codec *codec, · 26 unmodified lines | ||
| 2354 | 2492 | hda_nid_t cvt_nid) |
| 2355 | 2493 | { |
| 2356 | 2494 | if (per_pin) { |
| 2495 | + snd_hda_set_dev_select(codec, per_pin->pin_nid, | |
| 2496 | + per_pin->dev_id); | |
| 2357 | 2497 | intel_verify_pin_cvt_connect(codec, per_pin); |
| 2358 | 2498 | intel_not_share_assigned_cvt(codec, per_pin->pin_nid, |
| 2359 | − per_pin->mux_idx); | |
| 2499 | + per_pin->dev_id, per_pin->mux_idx); | |
| 2360 | 2500 | } else { |
| 2361 | − intel_not_share_assigned_cvt_nid(codec, 0, cvt_nid); | |
| 2501 | + intel_not_share_assigned_cvt_nid(codec, 0, 0, cvt_nid); | |
| 2362 | 2502 | } |
| 2363 | 2503 | } |
| 2364 | 2504 | |
| @@ -2378,6 +2518,8 @@ static int patch_i915_hsw_hdmi(struct hda_codec *codec) · 13 unmodified lines | ||
| 2378 | 2518 | if (err < 0) |
| 2379 | 2519 | return err; |
| 2380 | 2520 | spec = codec->spec; |
| 2521 | + codec->dp_mst = true; | |
| 2522 | + spec->dyn_pcm_assign = true; | |
| 2381 | 2523 | |
| 2382 | 2524 | intel_haswell_enable_all_pins(codec, true); |
| 2383 | 2525 | intel_haswell_fixup_enable_dp12(codec); |
| @@ -2389,7 +2531,6 @@ static int patch_i915_hsw_hdmi(struct hda_codec *codec) · 5 unmodified lines | ||
| 2389 | 2531 | codec->core.link_power_control = 1; |
| 2390 | 2532 | |
| 2391 | 2533 | codec->patch_ops.set_power_state = haswell_set_power_state; |
| 2392 | − codec->dp_mst = true; | |
| 2393 | 2534 | codec->depop_delay = 0; |
| 2394 | 2535 | codec->auto_runtime_pm = 1; |
| 2395 | 2536 | |