History

Merge tag 'thermal-5.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull thermal control updates from Rafael Wysocki: "As far as new functionality is concerned, there is a new thermal driver for the Intel Hardware Feedback Interface (HFI) along with some intel-speed-select utility changes to support it. There are also new DT compatible strings for a couple of platforms, and thermal zones on some platforms will be registered as HWmon sensors now. Apart from the above, some drivers are updated (fixes mostly) and there is a new piece of documentation for the Intel DPTF (Dynamic Power and Thermal Framework) sysfs interface. Specifics: - Add a new thermal driver for the Intel Hardware Feedback Interface (HFI) including the HFI initialization, HFI notification interrupt handling and sending CPU capabilities change messages to user space via the thermal netlink interface (Ricardo Neri, Srinivas Pandruvada, Nathan Chancellor, Randy Dunlap). - Extend the intel-speed-select utility to handle out-of-band CPU configuration changes and add support for the CPU capabilities change messages sent over the thermal netlink interface by the new HFI thermal driver to it (Srinivas Pandruvada). - Convert the DT bindings to yaml format for the Exynos platform and fix and update the MAINTAINERS file for this driver (Krzysztof Kozlowski). - Register the thermal zones as HWmon sensors for the QCom's Tsens driver and TI thermal platforms (Dmitry Baryshkov, Romain Naour). - Add the msm8953 compatible documentation in the bindings (Luca Weiss). - Add the sm8150 platform support to the QCom LMh driver's DT binding (Thara Gopinath). - Check the command result from the IPC command to the BPMP in the Tegra driver (Mikko Perttunen). - Silence the error for normal configuration where the interrupt is optionnal in the Broadcom thermal driver (Florian Fainelli). - Remove remaining dead code from the TI thermal driver (Yue Haibing). - Don't use bitmap_weight() in end_power_clamp() in the powerclamp driver (Yury Norov). - Update the OS policy capabilities handshake in the int340x thermal driver (Srinivas Pandruvada). - Increase the policies bitmap size in int340x (Srinivas Pandruvada). - Replace acpi_bus_get_device() with acpi_fetch_acpi_dev() in the int340x thermal driver (Rafael Wysocki). - Check for NULL after calling kmemdup() in int340x (Jiasheng Jiang). - Add Intel Dynamic Power and Thermal Framework (DPTF) kernel interface documentation (Srinivas Pandruvada). - Fix bullet list warning in the thermal documentation (Randy Dunlap)" * tag 'thermal-5.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (30 commits) thermal: int340x: Update OS policy capability handshake thermal: int340x: Increase bitmap size Documentation: thermal: DPTF Documentation MAINTAINERS: thermal: samsung: update Krzysztof Kozlowski's email thermal/drivers/ti-soc-thermal: Remove unused function ti_thermal_get_temp() thermal/drivers/brcmstb_thermal: Interrupt is optional thermal: tegra-bpmp: Handle errors in BPMP response drivers/thermal/ti-soc-thermal: Add hwmon support dt-bindings: thermal: tsens: Add msm8953 compatible dt-bindings: thermal: Add sm8150 compatible string for LMh thermal/drivers/qcom/lmh: Add support for sm8150 thermal/drivers/tsens: register thermal zones as hwmon sensors MAINTAINERS: thermal: samsung: Drop obsolete properties dt-bindings: thermal: samsung: Convert to dtschema tools/power/x86/intel-speed-select: v1.12 release tools/power/x86/intel-speed-select: HFI support tools/power/x86/intel-speed-select: OOB daemon mode thermal: intel: hfi: INTEL_HFI_THERMAL depends on NET thermal: netlink: Fix parameter type of thermal_genl_cpu_capability_event() stub thermal: Replace acpi_bus_get_device() ...

Linus Torvalds committed 4 years ago

Commit f6483722 parents 02b82b0 31035f334 files · +2026 −235Browse files

Unified · Split

34 files
  1. Documentation/ABI/testing/sysfs-class-thermal +1 −1
  2. Documentation/devicetree/bindings/thermal/exynos-thermal.txt +0 −106
  3. Documentation/devicetree/bindings/thermal/qcom-lmh.yaml +1 −0
  4. Documentation/devicetree/bindings/thermal/qcom-tsens.yaml +1 −0
  5. Documentation/devicetree/bindings/thermal/samsung,exynos-thermal.yaml +184 −0
  6. Documentation/driver-api/thermal/index.rst +1 −0
  7. Documentation/driver-api/thermal/intel_dptf.rst +272 −0
  8. Documentation/x86/index.rst +1 −0
  9. Documentation/x86/intel-hfi.rst +72 −0
  10. MAINTAINERS +4 −3
  11. arch/x86/include/asm/cpufeatures.h +1 −0
  12. arch/x86/include/asm/msr-index.h +6 −0
  13. drivers/thermal/broadcom/brcmstb_thermal.c +1 −1
  14. drivers/thermal/intel/Kconfig +14 −0
  15. drivers/thermal/intel/Makefile +1 −0
  16. drivers/thermal/intel/int340x_thermal/acpi_thermal_rel.c +7 −16
  17. drivers/thermal/intel/int340x_thermal/int3400_thermal.c +103 −50
  18. drivers/thermal/intel/intel_hfi.c +569 −0
  19. drivers/thermal/intel/intel_hfi.h +17 −0
  20. drivers/thermal/intel/intel_powerclamp.c +3 −6
  21. drivers/thermal/intel/therm_throt.c +22 −0
  22. drivers/thermal/qcom/lmh.c +35 −27
  23. drivers/thermal/qcom/tsens.c +5 −0
  24. drivers/thermal/tegra/tegra-bpmp-thermal.c +12 −1
  25. drivers/thermal/thermal_netlink.c +53 −0
  26. drivers/thermal/thermal_netlink.h +14 −0
  27. drivers/thermal/ti-soc-thermal/ti-thermal-common.c +4 −8
  28. include/uapi/linux/thermal.h +5 −1
  29. tools/power/x86/intel-speed-select/Build +1 −1
  30. tools/power/x86/intel-speed-select/Makefile +7 −3
  31. tools/power/x86/intel-speed-select/hfi-events.c +309 −0
  32. tools/power/x86/intel-speed-select/isst-config.c +43 −11
  33. tools/power/x86/intel-speed-select/isst-daemon.c +244 −0
  34. tools/power/x86/intel-speed-select/isst.h +13 −0
Documentation/ABI/testing/sysfs-class-thermal+1−1
@@ -203,7 +203,7 @@ Description: · 202 unmodified lines
203203
204204 - for generic ACPI: should be "Fan", "Processor" or "LCD"
205205 - for memory controller device on intel_menlow platform:
206− should be "Memory controller".
206+ should be "Memory controller".
207207
208208 RO, Required
209209
Documentation/devicetree/bindings/thermal/exynos-thermal.txt+0−106
@@ -1,106 +0,0 @@
1−* Exynos Thermal Management Unit (TMU)
2−
3−** Required properties:
4−
5−- compatible : One of the following:
6− "samsung,exynos3250-tmu"
7− "samsung,exynos4412-tmu"
8− "samsung,exynos4210-tmu"
9− "samsung,exynos5250-tmu"
10− "samsung,exynos5260-tmu"
11− "samsung,exynos5420-tmu" for TMU channel 0, 1 on Exynos5420
12− "samsung,exynos5420-tmu-ext-triminfo" for TMU channels 2, 3 and 4
13− Exynos5420 (Must pass triminfo base and triminfo clock)
14− "samsung,exynos5433-tmu"
15− "samsung,exynos7-tmu"
16−- reg : Address range of the thermal registers. For soc's which has multiple
17− instances of TMU and some registers are shared across all TMU's like
18− interrupt related then 2 set of register has to supplied. First set
19− belongs to register set of TMU instance and second set belongs to
20− registers shared with the TMU instance.
21−
22− NOTE: On Exynos5420, the TRIMINFO register is misplaced for TMU
23− channels 2, 3 and 4
24− Use "samsung,exynos5420-tmu-ext-triminfo" in cases, there is a misplaced
25− register, also provide clock to access that base.
26−
27− TRIMINFO at 0x1006c000 contains data for TMU channel 3
28− TRIMINFO at 0x100a0000 contains data for TMU channel 4
29− TRIMINFO at 0x10068000 contains data for TMU channel 2
30−
31−- interrupts : Should contain interrupt for thermal system
32−- clocks : The main clocks for TMU device
33− -- 1. operational clock for TMU channel
34− -- 2. optional clock to access the shared registers of TMU channel
35− -- 3. optional special clock for functional operation
36−- clock-names : Thermal system clock name
37− -- "tmu_apbif" operational clock for current TMU channel
38− -- "tmu_triminfo_apbif" clock to access the shared triminfo register
39− for current TMU channel
40− -- "tmu_sclk" clock for functional operation of the current TMU
41− channel
42−
43−The Exynos TMU supports generating interrupts when reaching given
44−temperature thresholds. Number of supported thermal trip points depends
45−on the SoC (only first trip points defined in DT will be configured):
46− - most of SoC: 4
47− - samsung,exynos5433-tmu: 8
48− - samsung,exynos7-tmu: 8
49−
50−** Optional properties:
51−
52−- vtmu-supply: This entry is optional and provides the regulator node supplying
53− voltage to TMU. If needed this entry can be placed inside
54− board/platform specific dts file.
55−
56−Example 1):
57−
58− tmu@100c0000 {
59− compatible = "samsung,exynos4412-tmu";
60− interrupt-parent = <&combiner>;
61− reg = <0x100C0000 0x100>;
62− interrupts = <2 4>;
63− clocks = <&clock 383>;
64− clock-names = "tmu_apbif";
65− vtmu-supply = <&tmu_regulator_node>;
66− #thermal-sensor-cells = <0>;
67− };
68−
69−Example 2): (In case of Exynos5420 "with misplaced TRIMINFO register")
70− tmu_cpu2: tmu@10068000 {
71− compatible = "samsung,exynos5420-tmu-ext-triminfo";
72− reg = <0x10068000 0x100>, <0x1006c000 0x4>;
73− interrupts = <0 184 0>;
74− clocks = <&clock 318>, <&clock 318>;
75− clock-names = "tmu_apbif", "tmu_triminfo_apbif";
76− #thermal-sensor-cells = <0>;
77− };
78−
79− tmu_cpu3: tmu@1006c000 {
80− compatible = "samsung,exynos5420-tmu-ext-triminfo";
81− reg = <0x1006c000 0x100>, <0x100a0000 0x4>;
82− interrupts = <0 185 0>;
83− clocks = <&clock 318>, <&clock 319>;
84− clock-names = "tmu_apbif", "tmu_triminfo_apbif";
85− #thermal-sensor-cells = <0>;
86− };
87−
88− tmu_gpu: tmu@100a0000 {
89− compatible = "samsung,exynos5420-tmu-ext-triminfo";
90− reg = <0x100a0000 0x100>, <0x10068000 0x4>;
91− interrupts = <0 215 0>;
92− clocks = <&clock 319>, <&clock 318>;
93− clock-names = "tmu_apbif", "tmu_triminfo_apbif";
94− #thermal-sensor-cells = <0>;
95− };
96−
97−Note: For multi-instance tmu each instance should have an alias correctly
98−numbered in "aliases" node.
99−
100−Example:
101−
102−aliases {
103− tmuctrl0 = &tmuctrl_0;
104− tmuctrl1 = &tmuctrl_1;
105− tmuctrl2 = &tmuctrl_2;
106−};
Documentation/devicetree/bindings/thermal/qcom-lmh.yaml+1−0
@@ -19,6 +19,7 @@ properties: · 18 unmodified lines
1919 compatible:
2020 enum:
2121 - qcom,sdm845-lmh
22+ - qcom,sm8150-lmh
2223
2324 reg:
2425 items:
Documentation/devicetree/bindings/thermal/qcom-tsens.yaml+1−0
@@ -43,6 +43,7 @@ properties: · 42 unmodified lines
4343 - description: v2 of TSENS
4444 items:
4545 - enum:
46+ - qcom,msm8953-tsens
4647 - qcom,msm8996-tsens
4748 - qcom,msm8998-tsens
4849 - qcom,sc7180-tsens
Documentation/devicetree/bindings/thermal/samsung,exynos-thermal.yaml+184−0
@@ -0,0 +1,184 @@
1+# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2+%YAML 1.2
3+---
4+$id: http://devicetree.org/schemas/thermal/samsung,exynos-thermal.yaml#
5+$schema: http://devicetree.org/meta-schemas/core.yaml#
6+
7+title: Samsung Exynos SoC Thermal Management Unit (TMU)
8+
9+maintainers:
10+ - Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
11+
12+description: |
13+ For multi-instance tmu each instance should have an alias correctly numbered
14+ in "aliases" node.
15+
16+properties:
17+ compatible:
18+ enum:
19+ - samsung,exynos3250-tmu
20+ - samsung,exynos4412-tmu
21+ - samsung,exynos4210-tmu
22+ - samsung,exynos5250-tmu
23+ - samsung,exynos5260-tmu
24+ # For TMU channel 0, 1 on Exynos5420:
25+ - samsung,exynos5420-tmu
26+ # For TMU channels 2, 3 and 4 of Exynos5420:
27+ - samsung,exynos5420-tmu-ext-triminfo
28+ - samsung,exynos5433-tmu
29+ - samsung,exynos7-tmu
30+
31+ clocks:
32+ minItems: 1
33+ maxItems: 3
34+
35+ clock-names:
36+ minItems: 1
37+ maxItems: 3
38+
39+ interrupts:
40+ description: |
41+ The Exynos TMU supports generating interrupts when reaching given
42+ temperature thresholds. Number of supported thermal trip points depends
43+ on the SoC (only first trip points defined in DT will be configured)::
44+ - most of SoC: 4
45+ - samsung,exynos5433-tmu: 8
46+ - samsung,exynos7-tmu: 8
47+ maxItems: 1
48+
49+ reg:
50+ items:
51+ - description: TMU instance registers.
52+ - description: |
53+ Shared TMU registers.
54+
55+ Note:: On Exynos5420, the TRIMINFO register is misplaced for TMU
56+ channels 2, 3 and 4 Use "samsung,exynos5420-tmu-ext-triminfo" in
57+ cases, there is a misplaced register, also provide clock to access
58+ that base.
59+ TRIMINFO at 0x1006c000 contains data for TMU channel 3
60+ TRIMINFO at 0x100a0000 contains data for TMU channel 4
61+ TRIMINFO at 0x10068000 contains data for TMU channel 2
62+ minItems: 1
63+
64+ '#thermal-sensor-cells': true
65+
66+ vtmu-supply:
67+ description: The regulator node supplying voltage to TMU.
68+
69+required:
70+ - compatible
71+ - clocks
72+ - clock-names
73+ - interrupts
74+ - reg
75+
76+allOf:
77+ - $ref: /schemas/thermal/thermal-sensor.yaml
78+ - if:
79+ properties:
80+ compatible:
81+ contains:
82+ const: samsung,exynos5420-tmu-ext-triminfo
83+ then:
84+ properties:
85+ clocks:
86+ items:
87+ - description:
88+ Operational clock for TMU channel.
89+ - description:
90+ Optional clock to access the shared registers (e.g. TRIMINFO) of TMU
91+ channel.
92+ clock-names:
93+ items:
94+ - const: tmu_apbif
95+ - const: tmu_triminfo_apbif
96+ reg:
97+ minItems: 2
98+ maxItems: 2
99+ - if:
100+ properties:
101+ compatible:
102+ contains:
103+ enum:
104+ - samsung,exynos5433-tmu
105+ - samsung,exynos7-tmu
106+ then:
107+ properties:
108+ clocks:
109+ items:
110+ - description:
111+ Operational clock for TMU channel.
112+ - description:
113+ Optional special clock for functional operation of TMU channel.
114+ clock-names:
115+ items:
116+ - const: tmu_apbif
117+ - const: tmu_sclk
118+ reg:
119+ minItems: 1
120+ maxItems: 1
121+
122+ - if:
123+ properties:
124+ compatible:
125+ contains:
126+ enum:
127+ - samsung,exynos3250-tmu
128+ - samsung,exynos4412-tmu
129+ - samsung,exynos4210-tmu
130+ - samsung,exynos5250-tmu
131+ - samsung,exynos5260-tmu
132+ - samsung,exynos5420-tmu
133+ then:
134+ properties:
135+ clocks:
136+ minItems: 1
137+ maxItems: 1
138+ reg:
139+ minItems: 1
140+ maxItems: 1
141+
142+additionalProperties: false
143+
144+examples:
145+ - |
146+ #include <dt-bindings/clock/exynos4.h>
147+
148+ tmu@100c0000 {
149+ compatible = "samsung,exynos4412-tmu";
150+ reg = <0x100C0000 0x100>;
151+ interrupt-parent = <&combiner>;
152+ interrupts = <2 4>;
153+ #thermal-sensor-cells = <0>;
154+ clocks = <&clock CLK_TMU_APBIF>;
155+ clock-names = "tmu_apbif";
156+ vtmu-supply = <&ldo10_reg>;
157+ };
158+
159+ - |
160+ #include <dt-bindings/interrupt-controller/arm-gic.h>
161+
162+ tmu@10068000 {
163+ compatible = "samsung,exynos5420-tmu-ext-triminfo";
164+ reg = <0x10068000 0x100>, <0x1006c000 0x4>;
165+ interrupts = <GIC_SPI 184 IRQ_TYPE_LEVEL_HIGH>;
166+ #thermal-sensor-cells = <0>;
167+ clocks = <&clock 318>, <&clock 318>; /* CLK_TMU */
168+ clock-names = "tmu_apbif", "tmu_triminfo_apbif";
169+ vtmu-supply = <&ldo7_reg>;
170+ };
171+
172+ - |
173+ #include <dt-bindings/interrupt-controller/arm-gic.h>
174+
175+ tmu@10060000 {
176+ compatible = "samsung,exynos5433-tmu";
177+ reg = <0x10060000 0x200>;
178+ interrupts = <GIC_SPI 95 IRQ_TYPE_LEVEL_HIGH>;
179+ #thermal-sensor-cells = <0>;
180+ clocks = <&cmu_peris 3>, /* CLK_PCLK_TMU0_APBIF */
181+ <&cmu_peris 35>; /* CLK_SCLK_TMU0 */
182+ clock-names = "tmu_apbif", "tmu_sclk";
183+ vtmu-supply = <&ldo3_reg>;
184+ };
Documentation/driver-api/thermal/index.rst+1−0
@@ -17,3 +17,4 @@ Thermal · 16 unmodified lines
1717 intel_powerclamp
1818 nouveau_thermal
1919 x86_pkg_temperature_thermal
20+ intel_dptf
Documentation/driver-api/thermal/intel_dptf.rst+272−0
@@ -0,0 +1,272 @@
1+.. SPDX-License-Identifier: GPL-2.0
2+
3+===============================================================
4+Intel(R) Dynamic Platform and Thermal Framework Sysfs Interface
5+===============================================================
6+
7+:Copyright: |copy| 2022 Intel Corporation
8+
9+:Author: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
10+
11+Introduction
12+------------
13+
14+Intel(R) Dynamic Platform and Thermal Framework (DPTF) is a platform
15+level hardware/software solution for power and thermal management.
16+
17+As a container for multiple power/thermal technologies, DPTF provides
18+a coordinated approach for different policies to effect the hardware
19+state of a system.
20+
21+Since it is a platform level framework, this has several components.
22+Some parts of the technology is implemented in the firmware and uses
23+ACPI and PCI devices to expose various features for monitoring and
24+control. Linux has a set of kernel drivers exposing hardware interface
25+to user space. This allows user space thermal solutions like
26+"Linux Thermal Daemon" to read platform specific thermal and power
27+tables to deliver adequate performance while keeping the system under
28+thermal limits.
29+
30+DPTF ACPI Drivers interface
31+----------------------------
32+
33+:file:`/sys/bus/platform/devices/<N>/uuids`, where <N>
34+=INT3400|INTC1040|INTC1041|INTC10A0
35+
36+``available_uuids`` (RO)
37+ A set of UUIDs strings presenting available policies
38+ which should be notified to the firmware when the
39+ user space can support those policies.
40+
41+ UUID strings:
42+
43+ "42A441D6-AE6A-462b-A84B-4A8CE79027D3" : Passive 1
44+
45+ "3A95C389-E4B8-4629-A526-C52C88626BAE" : Active
46+
47+ "97C68AE7-15FA-499c-B8C9-5DA81D606E0A" : Critical
48+
49+ "63BE270F-1C11-48FD-A6F7-3AF253FF3E2D" : Adaptive performance
50+
51+ "5349962F-71E6-431D-9AE8-0A635B710AEE" : Emergency call
52+
53+ "9E04115A-AE87-4D1C-9500-0F3E340BFE75" : Passive 2
54+
55+ "F5A35014-C209-46A4-993A-EB56DE7530A1" : Power Boss
56+
57+ "6ED722A7-9240-48A5-B479-31EEF723D7CF" : Virtual Sensor
58+
59+ "16CAF1B7-DD38-40ED-B1C1-1B8A1913D531" : Cooling mode
60+
61+ "BE84BABF-C4D4-403D-B495-3128FD44dAC1" : HDC
62+
63+``current_uuid`` (RW)
64+ User space can write strings from available UUIDs, one at a
65+ time.
66+
67+:file:`/sys/bus/platform/devices/<N>/`, where <N>
68+=INT3400|INTC1040|INTC1041|INTC10A0
69+
70+``imok`` (WO)
71+ User space daemon write 1 to respond to firmware event
72+ for sending keep alive notification. User space receives
73+ THERMAL_EVENT_KEEP_ALIVE kobject uevent notification when
74+ firmware calls for user space to respond with imok ACPI
75+ method.
76+
77+``odvp*`` (RO)
78+ Firmware thermal status variable values. Thermal tables
79+ calls for different processing based on these variable
80+ values.
81+
82+``data_vault`` (RO)
83+ Binary thermal table. Refer to
84+ https:/github.com/intel/thermal_daemon for decoding
85+ thermal table.
86+
87+
88+ACPI Thermal Relationship table interface
89+------------------------------------------
90+
91+:file:`/dev/acpi_thermal_rel`
92+
93+ This device provides IOCTL interface to read standard ACPI
94+ thermal relationship tables via ACPI methods _TRT and _ART.
95+ These IOCTLs are defined in
96+ drivers/thermal/intel/int340x_thermal/acpi_thermal_rel.h
97+
98+ IOCTLs:
99+
100+ ACPI_THERMAL_GET_TRT_LEN: Get length of TRT table
101+
102+ ACPI_THERMAL_GET_ART_LEN: Get length of ART table
103+
104+ ACPI_THERMAL_GET_TRT_COUNT: Number of records in TRT table
105+
106+ ACPI_THERMAL_GET_ART_COUNT: Number of records in ART table
107+
108+ ACPI_THERMAL_GET_TRT: Read binary TRT table, length to read is
109+ provided via argument to ioctl().
110+
111+ ACPI_THERMAL_GET_ART: Read binary ART table, length to read is
112+ provided via argument to ioctl().
113+
114+DPTF ACPI Sensor drivers
115+-------------------------
116+
117+DPTF Sensor drivers are presented as standard thermal sysfs thermal_zone.
118+
119+
120+DPTF ACPI Cooling drivers
121+--------------------------
122+
123+DPTF cooling drivers are presented as standard thermal sysfs cooling_device.
124+
125+
126+DPTF Processor thermal PCI Driver interface
127+--------------------------------------------
128+
129+:file:`/sys/bus/pci/devices/0000\:00\:04.0/power_limits/`
130+
131+Refer to Documentation/power/powercap/powercap.rst for powercap
132+ABI.
133+
134+``power_limit_0_max_uw`` (RO)
135+ Maximum powercap sysfs constraint_0_power_limit_uw for Intel RAPL
136+
137+``power_limit_0_step_uw`` (RO)
138+ Power limit increment/decrements for Intel RAPL constraint 0 power limit
139+
140+``power_limit_0_min_uw`` (RO)
141+ Minimum powercap sysfs constraint_0_power_limit_uw for Intel RAPL
142+
143+``power_limit_0_tmin_us`` (RO)
144+ Minimum powercap sysfs constraint_0_time_window_us for Intel RAPL
145+
146+``power_limit_0_tmax_us`` (RO)
147+ Maximum powercap sysfs constraint_0_time_window_us for Intel RAPL
148+
149+``power_limit_1_max_uw`` (RO)
150+ Maximum powercap sysfs constraint_1_power_limit_uw for Intel RAPL
151+
152+``power_limit_1_step_uw`` (RO)
153+ Power limit increment/decrements for Intel RAPL constraint 1 power limit
154+
155+``power_limit_1_min_uw`` (RO)
156+ Minimum powercap sysfs constraint_1_power_limit_uw for Intel RAPL
157+
158+``power_limit_1_tmin_us`` (RO)
159+ Minimum powercap sysfs constraint_1_time_window_us for Intel RAPL
160+
161+``power_limit_1_tmax_us`` (RO)
162+ Maximum powercap sysfs constraint_1_time_window_us for Intel RAPL
163+
164+:file:`/sys/bus/pci/devices/0000\:00\:04.0/`
165+
166+``tcc_offset_degree_celsius`` (RW)
167+ TCC offset from the critical temperature where hardware will throttle
168+ CPU.
169+
170+:file:`/sys/bus/pci/devices/0000\:00\:04.0/workload_request`
171+
172+``workload_available_types`` (RO)
173+ Available workload types. User space can specify one of the workload type
174+ it is currently executing via workload_type. For example: idle, bursty,
175+ sustained etc.
176+
177+``workload_type`` (RW)
178+ User space can specify any one of the available workload type using
179+ this interface.
180+
181+DPTF Processor thermal RFIM interface
182+--------------------------------------------
183+
184+RFIM interface allows adjustment of FIVR (Fully Integrated Voltage Regulator)
185+and DDR (Double Data Rate)frequencies to avoid RF interference with WiFi and 5G.
186+
187+Switching voltage regulators (VR) generate radiated EMI or RFI at the
188+fundamental frequency and its harmonics. Some harmonics may interfere
189+with very sensitive wireless receivers such as Wi-Fi and cellular that
190+are integrated into host systems like notebook PCs. One of mitigation
191+methods is requesting SOC integrated VR (IVR) switching frequency to a
192+small % and shift away the switching noise harmonic interference from
193+radio channels. OEM or ODMs can use the driver to control SOC IVR
194+operation within the range where it does not impact IVR performance.
195+
196+DRAM devices of DDR IO interface and their power plane can generate EMI
197+at the data rates. Similar to IVR control mechanism, Intel offers a
198+mechanism by which DDR data rates can be changed if several conditions
199+are met: there is strong RFI interference because of DDR; CPU power
200+management has no other restriction in changing DDR data rates;
201+PC ODMs enable this feature (real time DDR RFI Mitigation referred to as
202+DDR-RFIM) for Wi-Fi from BIOS.
203+
204+
205+FIVR attributes
206+
207+:file:`/sys/bus/pci/devices/0000\:00\:04.0/fivr/`
208+
209+``vco_ref_code_lo`` (RW)
210+ The VCO reference code is an 11-bit field and controls the FIVR
211+ switching frequency. This is the 3-bit LSB field.
212+
213+``vco_ref_code_hi`` (RW)
214+ The VCO reference code is an 11-bit field and controls the FIVR
215+ switching frequency. This is the 8-bit MSB field.
216+
217+``spread_spectrum_pct`` (RW)
218+ Set the FIVR spread spectrum clocking percentage
219+
220+``spread_spectrum_clk_enable`` (RW)
221+ Enable/disable of the FIVR spread spectrum clocking feature
222+
223+``rfi_vco_ref_code`` (RW)
224+ This field is a read only status register which reflects the
225+ current FIVR switching frequency
226+
227+``fivr_fffc_rev`` (RW)
228+ This field indicated the revision of the FIVR HW.
229+
230+
231+DVFS attributes
232+
233+:file:`/sys/bus/pci/devices/0000\:00\:04.0/dvfs/`
234+
235+``rfi_restriction_run_busy`` (RW)
236+ Request the restriction of specific DDR data rate and set this
237+ value 1. Self reset to 0 after operation.
238+
239+``rfi_restriction_err_code`` (RW)
240+ 0 :Request is accepted, 1:Feature disabled,
241+ 2: the request restricts more points than it is allowed
242+
243+``rfi_restriction_data_rate_Delta`` (RW)
244+ Restricted DDR data rate for RFI protection: Lower Limit
245+
246+``rfi_restriction_data_rate_Base`` (RW)
247+ Restricted DDR data rate for RFI protection: Upper Limit
248+
249+``ddr_data_rate_point_0`` (RO)
250+ DDR data rate selection 1st point
251+
252+``ddr_data_rate_point_1`` (RO)
253+ DDR data rate selection 2nd point
254+
255+``ddr_data_rate_point_2`` (RO)
256+ DDR data rate selection 3rd point
257+
258+``ddr_data_rate_point_3`` (RO)
259+ DDR data rate selection 4th point
260+
261+``rfi_disable (RW)``
262+ Disable DDR rate change feature
263+
264+DPTF Power supply and Battery Interface
265+----------------------------------------
266+
267+Refer to Documentation/ABI/testing/sysfs-platform-dptf
268+
269+DPTF Fan Control
270+----------------------------------------
271+
272+Refer to Documentation/admin-guide/acpi/fan_performance_states.rst
Documentation/x86/index.rst+1−0
@@ -21,6 +21,7 @@ x86-specific Documentation · 20 unmodified lines
2121 tlb
2222 mtrr
2323 pat
24+ intel-hfi
2425 intel-iommu
2526 intel_txt
2627 amd-memory-encryption
Documentation/x86/intel-hfi.rst+72−0
@@ -0,0 +1,72 @@
1+.. SPDX-License-Identifier: GPL-2.0
2+
3+============================================================
4+Hardware-Feedback Interface for scheduling on Intel Hardware
5+============================================================
6+
7+Overview
8+--------
9+
10+Intel has described the Hardware Feedback Interface (HFI) in the Intel 64 and
11+IA-32 Architectures Software Developer's Manual (Intel SDM) Volume 3 Section
12+14.6 [1]_.
13+
14+The HFI gives the operating system a performance and energy efficiency
15+capability data for each CPU in the system. Linux can use the information from
16+the HFI to influence task placement decisions.
17+
18+The Hardware Feedback Interface
19+-------------------------------
20+
21+The Hardware Feedback Interface provides to the operating system information
22+about the performance and energy efficiency of each CPU in the system. Each
23+capability is given as a unit-less quantity in the range [0-255]. Higher values
24+indicate higher capability. Energy efficiency and performance are reported in
25+separate capabilities. Even though on some systems these two metrics may be
26+related, they are specified as independent capabilities in the Intel SDM.
27+
28+These capabilities may change at runtime as a result of changes in the
29+operating conditions of the system or the action of external factors. The rate
30+at which these capabilities are updated is specific to each processor model. On
31+some models, capabilities are set at boot time and never change. On others,
32+capabilities may change every tens of milliseconds. For instance, a remote
33+mechanism may be used to lower Thermal Design Power. Such change can be
34+reflected in the HFI. Likewise, if the system needs to be throttled due to
35+excessive heat, the HFI may reflect reduced performance on specific CPUs.
36+
37+The kernel or a userspace policy daemon can use these capabilities to modify
38+task placement decisions. For instance, if either the performance or energy
39+capabilities of a given logical processor becomes zero, it is an indication that
40+the hardware recommends to the operating system to not schedule any tasks on
41+that processor for performance or energy efficiency reasons, respectively.
42+
43+Implementation details for Linux
44+--------------------------------
45+
46+The infrastructure to handle thermal event interrupts has two parts. In the
47+Local Vector Table of a CPU's local APIC, there exists a register for the
48+Thermal Monitor Register. This register controls how interrupts are delivered
49+to a CPU when the thermal monitor generates and interrupt. Further details
50+can be found in the Intel SDM Vol. 3 Section 10.5 [1]_.
51+
52+The thermal monitor may generate interrupts per CPU or per package. The HFI
53+generates package-level interrupts. This monitor is configured and initialized
54+via a set of machine-specific registers. Specifically, the HFI interrupt and
55+status are controlled via designated bits in the IA32_PACKAGE_THERM_INTERRUPT
56+and IA32_PACKAGE_THERM_STATUS registers, respectively. There exists one HFI
57+table per package. Further details can be found in the Intel SDM Vol. 3
58+Section 14.9 [1]_.
59+
60+The hardware issues an HFI interrupt after updating the HFI table and is ready
61+for the operating system to consume it. CPUs receive such interrupt via the
62+thermal entry in the Local APIC's Local Vector Table.
63+
64+When servicing such interrupt, the HFI driver parses the updated table and
65+relays the update to userspace using the thermal notification framework. Given
66+that there may be many HFI updates every second, the updates relayed to
67+userspace are throttled at a rate of CONFIG_HZ jiffies.
68+
69+References
70+----------
71+
72+.. [1] https://www.intel.com/sdm
MAINTAINERS+4−3
@@ -17143,11 +17143,12 @@ S: Supported · 17142 unmodified lines
1714317143 F: drivers/net/ethernet/samsung/sxgbe/
1714417144
1714517145 SAMSUNG THERMAL DRIVER
17146−M: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
17146+M: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17147+M: Krzysztof Kozlowski <krzk@kernel.org>
1714717148 L: linux-pm@vger.kernel.org
1714817149 L: linux-samsung-soc@vger.kernel.org
17149−S: Supported
17150−T: git https://github.com/lmajewski/linux-samsung-thermal.git
17150+S: Maintained
17151+F: Documentation/devicetree/bindings/thermal/samsung,exynos-thermal.yaml
1715117152 F: drivers/thermal/samsung/
1715217153
1715317154 SAMSUNG USB2 PHY DRIVER
arch/x86/include/asm/cpufeatures.h+1−0
@@ -327,6 +327,7 @@ · 326 unmodified lines
327327 #define X86_FEATURE_HWP_ACT_WINDOW (14*32+ 9) /* HWP Activity Window */
328328 #define X86_FEATURE_HWP_EPP (14*32+10) /* HWP Energy Perf. Preference */
329329 #define X86_FEATURE_HWP_PKG_REQ (14*32+11) /* HWP Package Level Request */
330+#define X86_FEATURE_HFI (14*32+19) /* Hardware Feedback Interface */
330331
331332 /* AMD SVM Feature Identification, CPUID level 0x8000000a (EDX), word 15 */
332333 #define X86_FEATURE_NPT (15*32+ 0) /* Nested Page Table support */
arch/x86/include/asm/msr-index.h+6−0
@@ -705,12 +705,14 @@ · 704 unmodified lines
705705
706706 #define PACKAGE_THERM_STATUS_PROCHOT (1 << 0)
707707 #define PACKAGE_THERM_STATUS_POWER_LIMIT (1 << 10)
708+#define PACKAGE_THERM_STATUS_HFI_UPDATED (1 << 26)
708709
709710 #define MSR_IA32_PACKAGE_THERM_INTERRUPT 0x000001b2
710711
711712 #define PACKAGE_THERM_INT_HIGH_ENABLE (1 << 0)
712713 #define PACKAGE_THERM_INT_LOW_ENABLE (1 << 1)
713714 #define PACKAGE_THERM_INT_PLN_ENABLE (1 << 24)
715+#define PACKAGE_THERM_INT_HFI_ENABLE (1 << 25)
714716
715717 /* Thermal Thresholds Support */
716718 #define THERM_INT_THRESHOLD0_ENABLE (1 << 15)
@@ -959,4 +961,8 @@ · 242 unmodified lines
959961 #define MSR_VM_IGNNE 0xc0010115
960962 #define MSR_VM_HSAVE_PA 0xc0010117
961963
964+/* Hardware Feedback Interface */
965+#define MSR_IA32_HW_FEEDBACK_PTR 0x17d0
966+#define MSR_IA32_HW_FEEDBACK_CONFIG 0x17d1
967+
962968 #endif /* _ASM_X86_MSR_INDEX_H */
drivers/thermal/broadcom/brcmstb_thermal.c+1−1
@@ -351,7 +351,7 @@ static int brcmstb_thermal_probe(struct platform_device *pdev) · 350 unmodified lines
351351
352352 priv->thermal = thermal;
353353
354− irq = platform_get_irq(pdev, 0);
354+ irq = platform_get_irq_optional(pdev, 0);
355355 if (irq >= 0) {
356356 ret = devm_request_threaded_irq(&pdev->dev, irq, NULL,
357357 brcmstb_tmon_irq_thread,
drivers/thermal/intel/Kconfig+14−0
@@ -99,3 +99,17 @@ config INTEL_MENLOW · 98 unmodified lines
9999 Intel Menlow platform.
100100
101101 If unsure, say N.
102+
103+config INTEL_HFI_THERMAL
104+ bool "Intel Hardware Feedback Interface"
105+ depends on NET
106+ depends on CPU_SUP_INTEL
107+ depends on X86_THERMAL_VECTOR
108+ select THERMAL_NETLINK
109+ help
110+ Select this option to enable the Hardware Feedback Interface. If
111+ selected, hardware provides guidance to the operating system on
112+ the performance and energy efficiency capabilities of each CPU.
113+ These capabilities may change as a result of changes in the operating
114+ conditions of the system such power and thermal limits. If selected,
115+ the kernel relays updates in CPUs' capabilities to userspace.
drivers/thermal/intel/Makefile+1−0
@@ -13,3 +13,4 @@ obj-$(CONFIG_INTEL_PCH_THERMAL) += intel_pch_thermal.o · 12 unmodified lines
1313 obj-$(CONFIG_INTEL_TCC_COOLING) += intel_tcc_cooling.o
1414 obj-$(CONFIG_X86_THERMAL_VECTOR) += therm_throt.o
1515 obj-$(CONFIG_INTEL_MENLOW) += intel_menlow.o
16+obj-$(CONFIG_INTEL_HFI_THERMAL) += intel_hfi.o
drivers/thermal/intel/int340x_thermal/acpi_thermal_rel.c+7−16
@@ -72,7 +72,6 @@ int acpi_parse_trt(acpi_handle handle, int *trt_count, struct trt **trtp, · 71 unmodified lines
7272 int i;
7373 int nr_bad_entries = 0;
7474 struct trt *trts;
75− struct acpi_device *adev;
7675 union acpi_object *p;
7776 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
7877 struct acpi_buffer element = { 0, NULL };
@@ -112,12 +111,10 @@ int acpi_parse_trt(acpi_handle handle, int *trt_count, struct trt **trtp, · 33 unmodified lines
112111 if (!create_dev)
113112 continue;
114113
115− result = acpi_bus_get_device(trt->source, &adev);
116− if (result)
114+ if (!acpi_fetch_acpi_dev(trt->source))
117115 pr_warn("Failed to get source ACPI device\n");
118116
119− result = acpi_bus_get_device(trt->target, &adev);
120− if (result)
117+ if (!acpi_fetch_acpi_dev(trt->target))
121118 pr_warn("Failed to get target ACPI device\n");
122119 }
123120
@@ -149,7 +146,6 @@ int acpi_parse_art(acpi_handle handle, int *art_count, struct art **artp, · 25 unmodified lines
149146 int i;
150147 int nr_bad_entries = 0;
151148 struct art *arts;
152− struct acpi_device *adev;
153149 union acpi_object *p;
154150 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
155151 struct acpi_buffer element = { 0, NULL };
@@ -191,16 +187,11 @@ int acpi_parse_art(acpi_handle handle, int *art_count, struct art **artp, · 35 unmodified lines
191187 if (!create_dev)
192188 continue;
193189
194− if (art->source) {
195− result = acpi_bus_get_device(art->source, &adev);
196− if (result)
197− pr_warn("Failed to get source ACPI device\n");
198− }
199− if (art->target) {
200− result = acpi_bus_get_device(art->target, &adev);
201− if (result)
202− pr_warn("Failed to get target ACPI device\n");
203− }
190+ if (!acpi_fetch_acpi_dev(art->source))
191+ pr_warn("Failed to get source ACPI device\n");
192+
193+ if (!acpi_fetch_acpi_dev(art->target))
194+ pr_warn("Failed to get target ACPI device\n");
204195 }
205196
206197 *artp = arts;
drivers/thermal/intel/int340x_thermal/int3400_thermal.c+103−50
@@ -17,8 +17,8 @@ · 16 unmodified lines
1717 #define INT3400_KEEP_ALIVE 0xA0
1818
1919 enum int3400_thermal_uuid {
20+ INT3400_THERMAL_ACTIVE = 0,
2021 INT3400_THERMAL_PASSIVE_1,
21− INT3400_THERMAL_ACTIVE,
2222 INT3400_THERMAL_CRITICAL,
2323 INT3400_THERMAL_ADAPTIVE_PERFORMANCE,
2424 INT3400_THERMAL_EMERGENCY_CALL_MODE,
@@ -31,8 +31,8 @@ enum int3400_thermal_uuid { · 6 unmodified lines
3131 };
3232
3333 static char *int3400_thermal_uuids[INT3400_THERMAL_MAXIMUM_UUID] = {
34− "42A441D6-AE6A-462b-A84B-4A8CE79027D3",
3534 "3A95C389-E4B8-4629-A526-C52C88626BAE",
35+ "42A441D6-AE6A-462b-A84B-4A8CE79027D3",
3636 "97C68AE7-15FA-499c-B8C9-5DA81D606E0A",
3737 "63BE270F-1C11-48FD-A6F7-3AF253FF3E2D",
3838 "5349962F-71E6-431D-9AE8-0A635B710AEE",
@@ -53,12 +53,13 @@ struct int3400_thermal_priv { · 14 unmodified lines
5353 struct art *arts;
5454 int trt_count;
5555 struct trt *trts;
56− u8 uuid_bitmap;
56+ u32 uuid_bitmap;
5757 int rel_misc_dev_res;
5858 int current_uuid_index;
5959 char *data_vault;
6060 int odvp_count;
6161 int *odvp;
62+ u32 os_uuid_mask;
6263 struct odvp_attr *odvp_attrs;
6364 };
6465
@@ -142,12 +143,55 @@ static ssize_t current_uuid_show(struct device *dev, · 77 unmodified lines
142143 struct device_attribute *devattr, char *buf)
143144 {
144145 struct int3400_thermal_priv *priv = dev_get_drvdata(dev);
146+ int i, length = 0;
145147
146− if (priv->current_uuid_index == -1)
147− return sprintf(buf, "INVALID\n");
148+ if (priv->current_uuid_index > 0)
149+ return sprintf(buf, "%s\n",
150+ int3400_thermal_uuids[priv->current_uuid_index]);
148151
149− return sprintf(buf, "%s\n",
150− int3400_thermal_uuids[priv->current_uuid_index]);
152+ for (i = 0; i <= INT3400_THERMAL_CRITICAL; i++) {
153+ if (priv->os_uuid_mask & BIT(i))
154+ length += scnprintf(&buf[length],
155+ PAGE_SIZE - length,
156+ "%s\n",
157+ int3400_thermal_uuids[i]);
158+ }
159+
160+ if (length)
161+ return length;
162+
163+ return sprintf(buf, "INVALID\n");
164+}
165+
166+static int int3400_thermal_run_osc(acpi_handle handle, char *uuid_str, int *enable)
167+{
168+ u32 ret, buf[2];
169+ acpi_status status;
170+ int result = 0;
171+ struct acpi_osc_context context = {
172+ .uuid_str = NULL,
173+ .rev = 1,
174+ .cap.length = 8,
175+ };
176+
177+ context.uuid_str = uuid_str;
178+
179+ buf[OSC_QUERY_DWORD] = 0;
180+ buf[OSC_SUPPORT_DWORD] = *enable;
181+
182+ context.cap.pointer = buf;
183+
184+ status = acpi_run_osc(handle, &context);
185+ if (ACPI_SUCCESS(status)) {
186+ ret = *((u32 *)(context.ret.pointer + 4));
187+ if (ret != *enable)
188+ result = -EPERM;
189+ } else
190+ result = -EPERM;
191+
192+ kfree(context.ret.pointer);
193+
194+ return result;
151195 }
152196
153197 static ssize_t current_uuid_store(struct device *dev,
@@ -164,16 +208,47 @@ static ssize_t current_uuid_store(struct device *dev, · 10 unmodified lines
164208 * If we have a list of supported UUIDs, make sure
165209 * this one is supported.
166210 */
167− if (priv->uuid_bitmap &&
168− !(priv->uuid_bitmap & (1 << i)))
211+ if (priv->uuid_bitmap & BIT(i)) {
212+ priv->current_uuid_index = i;
213+ return count;
214+ }
215+
216+ /*
217+ * There is support of only 3 policies via the new
218+ * _OSC to inform OS capability:
219+ * INT3400_THERMAL_ACTIVE
220+ * INT3400_THERMAL_PASSIVE_1
221+ * INT3400_THERMAL_CRITICAL
222+ */
223+
224+ if (i > INT3400_THERMAL_CRITICAL)
169225 return -EINVAL;
170226
171− priv->current_uuid_index = i;
172− return count;
227+ priv->os_uuid_mask |= BIT(i);
228+
229+ break;
173230 }
174231 }
175232
176− return -EINVAL;
233+ if (priv->os_uuid_mask) {
234+ int cap, ret;
235+
236+ /*
237+ * Capability bits:
238+ * Bit 0: set to 1 to indicate DPTF is active
239+ * Bi1 1: set to 1 to active cooling is supported by user space daemon
240+ * Bit 2: set to 1 to passive cooling is supported by user space daemon
241+ * Bit 3: set to 1 to critical trip is handled by user space daemon
242+ */
243+ cap = ((priv->os_uuid_mask << 1) | 0x01);
244+ ret = int3400_thermal_run_osc(priv->adev->handle,
245+ "b23ba85d-c8b7-3542-88de-8de2ffcfd698",
246+ &cap);
247+ if (ret)
248+ return ret;
249+ }
250+
251+ return count;
177252 }
178253
179254 static DEVICE_ATTR_RW(current_uuid);
@@ -236,41 +311,6 @@ end: · 56 unmodified lines
236311 return result;
237312 }
238313
239−static int int3400_thermal_run_osc(acpi_handle handle,
240− enum int3400_thermal_uuid uuid, bool enable)
241−{
242− u32 ret, buf[2];
243− acpi_status status;
244− int result = 0;
245− struct acpi_osc_context context = {
246− .uuid_str = NULL,
247− .rev = 1,
248− .cap.length = 8,
249− };
250−
251− if (uuid < 0 || uuid >= INT3400_THERMAL_MAXIMUM_UUID)
252− return -EINVAL;
253−
254− context.uuid_str = int3400_thermal_uuids[uuid];
255−
256− buf[OSC_QUERY_DWORD] = 0;
257− buf[OSC_SUPPORT_DWORD] = enable;
258−
259− context.cap.pointer = buf;
260−
261− status = acpi_run_osc(handle, &context);
262− if (ACPI_SUCCESS(status)) {
263− ret = *((u32 *)(context.ret.pointer + 4));
264− if (ret != enable)
265− result = -EPERM;
266− } else
267− result = -EPERM;
268−
269− kfree(context.ret.pointer);
270−
271− return result;
272−}
273−
274314 static ssize_t odvp_show(struct kobject *kobj, struct kobj_attribute *attr,
275315 char *buf)
276316 {
@@ -426,10 +466,18 @@ static int int3400_thermal_change_mode(struct thermal_zone_device *thermal, · 149 unmodified lines
426466 if (!priv)
427467 return -EINVAL;
428468
429− if (mode != thermal->mode)
469+ if (mode != thermal->mode) {
470+ int enabled;
471+
472+ if (priv->current_uuid_index < 0 ||
473+ priv->current_uuid_index >= INT3400_THERMAL_MAXIMUM_UUID)
474+ return -EINVAL;
475+
476+ enabled = (mode == THERMAL_DEVICE_ENABLED);
430477 result = int3400_thermal_run_osc(priv->adev->handle,
431− priv->current_uuid_index,
432− mode == THERMAL_DEVICE_ENABLED);
478+ int3400_thermal_uuids[priv->current_uuid_index],
479+ &enabled);
480+ }
433481
434482
435483 evaluate_odvp(priv);
@@ -468,6 +516,11 @@ static void int3400_setup_gddv(struct int3400_thermal_priv *priv) · 32 unmodified lines
468516 priv->data_vault = kmemdup(obj->package.elements[0].buffer.pointer,
469517 obj->package.elements[0].buffer.length,
470518 GFP_KERNEL);
519+ if (!priv->data_vault) {
520+ kfree(buffer.pointer);
521+ return;
522+ }
523+
471524 bin_attr_data_vault.private = priv->data_vault;
472525 bin_attr_data_vault.size = obj->package.elements[0].buffer.length;
473526 kfree(buffer.pointer);
drivers/thermal/intel/intel_hfi.c+569−0
@@ -0,0 +1,569 @@
1+// SPDX-License-Identifier: GPL-2.0-only
2+/*
3+ * Hardware Feedback Interface Driver
4+ *
5+ * Copyright (c) 2021, Intel Corporation.
6+ *
7+ * Authors: Aubrey Li <aubrey.li@linux.intel.com>
8+ * Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
9+ *
10+ *
11+ * The Hardware Feedback Interface provides a performance and energy efficiency
12+ * capability information for each CPU in the system. Depending on the processor
13+ * model, hardware may periodically update these capabilities as a result of
14+ * changes in the operating conditions (e.g., power limits or thermal
15+ * constraints). On other processor models, there is a single HFI update
16+ * at boot.
17+ *
18+ * This file provides functionality to process HFI updates and relay these
19+ * updates to userspace.
20+ */
21+
22+#define pr_fmt(fmt) "intel-hfi: " fmt
23+
24+#include <linux/bitops.h>
25+#include <linux/cpufeature.h>
26+#include <linux/cpumask.h>
27+#include <linux/gfp.h>
28+#include <linux/io.h>
29+#include <linux/kernel.h>
30+#include <linux/math.h>
31+#include <linux/mutex.h>
32+#include <linux/percpu-defs.h>
33+#include <linux/printk.h>
34+#include <linux/processor.h>
35+#include <linux/slab.h>
36+#include <linux/spinlock.h>
37+#include <linux/string.h>
38+#include <linux/topology.h>
39+#include <linux/workqueue.h>
40+
41+#include <asm/msr.h>
42+
43+#include "../thermal_core.h"
44+#include "intel_hfi.h"
45+
46+#define THERM_STATUS_CLEAR_PKG_MASK (BIT(1) | BIT(3) | BIT(5) | BIT(7) | \
47+ BIT(9) | BIT(11) | BIT(26))
48+
49+/* Hardware Feedback Interface MSR configuration bits */
50+#define HW_FEEDBACK_PTR_VALID_BIT BIT(0)
51+#define HW_FEEDBACK_CONFIG_HFI_ENABLE_BIT BIT(0)
52+
53+/* CPUID detection and enumeration definitions for HFI */
54+
55+#define CPUID_HFI_LEAF 6
56+
57+union hfi_capabilities {
58+ struct {
59+ u8 performance:1;
60+ u8 energy_efficiency:1;
61+ u8 __reserved:6;
62+ } split;
63+ u8 bits;
64+};
65+
66+union cpuid6_edx {
67+ struct {
68+ union hfi_capabilities capabilities;
69+ u32 table_pages:4;
70+ u32 __reserved:4;
71+ s32 index:16;
72+ } split;
73+ u32 full;
74+};
75+
76+/**
77+ * struct hfi_cpu_data - HFI capabilities per CPU
78+ * @perf_cap: Performance capability
79+ * @ee_cap: Energy efficiency capability
80+ *
81+ * Capabilities of a logical processor in the HFI table. These capabilities are
82+ * unitless.
83+ */
84+struct hfi_cpu_data {
85+ u8 perf_cap;
86+ u8 ee_cap;
87+} __packed;
88+
89+/**
90+ * struct hfi_hdr - Header of the HFI table
91+ * @perf_updated: Hardware updated performance capabilities
92+ * @ee_updated: Hardware updated energy efficiency capabilities
93+ *
94+ * Properties of the data in an HFI table.
95+ */
96+struct hfi_hdr {
97+ u8 perf_updated;
98+ u8 ee_updated;
99+} __packed;
100+
101+/**
102+ * struct hfi_instance - Representation of an HFI instance (i.e., a table)
103+ * @local_table: Base of the local copy of the HFI table
104+ * @timestamp: Timestamp of the last update of the local table.
105+ * Located at the base of the local table.
106+ * @hdr: Base address of the header of the local table
107+ * @data: Base address of the data of the local table
108+ * @cpus: CPUs represented in this HFI table instance
109+ * @hw_table: Pointer to the HFI table of this instance
110+ * @update_work: Delayed work to process HFI updates
111+ * @table_lock: Lock to protect acceses to the table of this instance
112+ * @event_lock: Lock to process HFI interrupts
113+ *
114+ * A set of parameters to parse and navigate a specific HFI table.
115+ */
116+struct hfi_instance {
117+ union {
118+ void *local_table;
119+ u64 *timestamp;
120+ };
121+ void *hdr;
122+ void *data;
123+ cpumask_var_t cpus;
124+ void *hw_table;
125+ struct delayed_work update_work;
126+ raw_spinlock_t table_lock;
127+ raw_spinlock_t event_lock;
128+};
129+
130+/**
131+ * struct hfi_features - Supported HFI features
132+ * @nr_table_pages: Size of the HFI table in 4KB pages
133+ * @cpu_stride: Stride size to locate the capability data of a logical
134+ * processor within the table (i.e., row stride)
135+ * @hdr_size: Size of the table header
136+ *
137+ * Parameters and supported features that are common to all HFI instances
138+ */
139+struct hfi_features {
140+ unsigned int nr_table_pages;
141+ unsigned int cpu_stride;
142+ unsigned int hdr_size;
143+};
144+
145+/**
146+ * struct hfi_cpu_info - Per-CPU attributes to consume HFI data
147+ * @index: Row of this CPU in its HFI table
148+ * @hfi_instance: Attributes of the HFI table to which this CPU belongs
149+ *
150+ * Parameters to link a logical processor to an HFI table and a row within it.
151+ */
152+struct hfi_cpu_info {
153+ s16 index;
154+ struct hfi_instance *hfi_instance;
155+};
156+
157+static DEFINE_PER_CPU(struct hfi_cpu_info, hfi_cpu_info) = { .index = -1 };
158+
159+static int max_hfi_instances;
160+static struct hfi_instance *hfi_instances;
161+
162+static struct hfi_features hfi_features;
163+static DEFINE_MUTEX(hfi_instance_lock);
164+
165+static struct workqueue_struct *hfi_updates_wq;
166+#define HFI_UPDATE_INTERVAL HZ
167+#define HFI_MAX_THERM_NOTIFY_COUNT 16
168+
169+static void get_hfi_caps(struct hfi_instance *hfi_instance,
170+ struct thermal_genl_cpu_caps *cpu_caps)
171+{
172+ int cpu, i = 0;
173+
174+ raw_spin_lock_irq(&hfi_instance->table_lock);
175+ for_each_cpu(cpu, hfi_instance->cpus) {
176+ struct hfi_cpu_data *caps;
177+ s16 index;
178+
179+ index = per_cpu(hfi_cpu_info, cpu).index;
180+ caps = hfi_instance->data + index * hfi_features.cpu_stride;
181+ cpu_caps[i].cpu = cpu;
182+
183+ /*
184+ * Scale performance and energy efficiency to
185+ * the [0, 1023] interval that thermal netlink uses.
186+ */
187+ cpu_caps[i].performance = caps->perf_cap << 2;
188+ cpu_caps[i].efficiency = caps->ee_cap << 2;
189+
190+ ++i;
191+ }
192+ raw_spin_unlock_irq(&hfi_instance->table_lock);
193+}
194+
195+/*
196+ * Call update_capabilities() when there are changes in the HFI table.
197+ */
198+static void update_capabilities(struct hfi_instance *hfi_instance)
199+{
200+ struct thermal_genl_cpu_caps *cpu_caps;
201+ int i = 0, cpu_count;
202+
203+ /* CPUs may come online/offline while processing an HFI update. */
204+ mutex_lock(&hfi_instance_lock);
205+
206+ cpu_count = cpumask_weight(hfi_instance->cpus);
207+
208+ /* No CPUs to report in this hfi_instance. */
209+ if (!cpu_count)
210+ goto out;
211+
212+ cpu_caps = kcalloc(cpu_count, sizeof(*cpu_caps), GFP_KERNEL);
213+ if (!cpu_caps)
214+ goto out;
215+
216+ get_hfi_caps(hfi_instance, cpu_caps);
217+
218+ if (cpu_count < HFI_MAX_THERM_NOTIFY_COUNT)
219+ goto last_cmd;
220+
221+ /* Process complete chunks of HFI_MAX_THERM_NOTIFY_COUNT capabilities. */
222+ for (i = 0;
223+ (i + HFI_MAX_THERM_NOTIFY_COUNT) <= cpu_count;
224+ i += HFI_MAX_THERM_NOTIFY_COUNT)
225+ thermal_genl_cpu_capability_event(HFI_MAX_THERM_NOTIFY_COUNT,
226+ &cpu_caps[i]);
227+
228+ cpu_count = cpu_count - i;
229+
230+last_cmd:
231+ /* Process the remaining capabilities if any. */
232+ if (cpu_count)
233+ thermal_genl_cpu_capability_event(cpu_count, &cpu_caps[i]);
234+
235+ kfree(cpu_caps);
236+out:
237+ mutex_unlock(&hfi_instance_lock);
238+}
239+
240+static void hfi_update_work_fn(struct work_struct *work)
241+{
242+ struct hfi_instance *hfi_instance;
243+
244+ hfi_instance = container_of(to_delayed_work(work), struct hfi_instance,
245+ update_work);
246+ if (!hfi_instance)
247+ return;
248+
249+ update_capabilities(hfi_instance);
250+}
251+
252+void intel_hfi_process_event(__u64 pkg_therm_status_msr_val)
253+{
254+ struct hfi_instance *hfi_instance;
255+ int cpu = smp_processor_id();
256+ struct hfi_cpu_info *info;
257+ u64 new_timestamp;
258+
259+ if (!pkg_therm_status_msr_val)
260+ return;
261+
262+ info = &per_cpu(hfi_cpu_info, cpu);
263+ if (!info)
264+ return;
265+
266+ /*
267+ * A CPU is linked to its HFI instance before the thermal vector in the
268+ * local APIC is unmasked. Hence, info->hfi_instance cannot be NULL
269+ * when receiving an HFI event.
270+ */
271+ hfi_instance = info->hfi_instance;
272+ if (unlikely(!hfi_instance)) {
273+ pr_debug("Received event on CPU %d but instance was null", cpu);
274+ return;
275+ }
276+
277+ /*
278+ * On most systems, all CPUs in the package receive a package-level
279+ * thermal interrupt when there is an HFI update. It is sufficient to
280+ * let a single CPU to acknowledge the update and queue work to
281+ * process it. The remaining CPUs can resume their work.
282+ */
283+ if (!raw_spin_trylock(&hfi_instance->event_lock))
284+ return;
285+
286+ /* Skip duplicated updates. */
287+ new_timestamp = *(u64 *)hfi_instance->hw_table;
288+ if (*hfi_instance->timestamp == new_timestamp) {
289+ raw_spin_unlock(&hfi_instance->event_lock);
290+ return;
291+ }
292+
293+ raw_spin_lock(&hfi_instance->table_lock);
294+
295+ /*
296+ * Copy the updated table into our local copy. This includes the new
297+ * timestamp.
298+ */
299+ memcpy(hfi_instance->local_table, hfi_instance->hw_table,
300+ hfi_features.nr_table_pages << PAGE_SHIFT);
301+
302+ raw_spin_unlock(&hfi_instance->table_lock);
303+ raw_spin_unlock(&hfi_instance->event_lock);
304+
305+ /*
306+ * Let hardware know that we are done reading the HFI table and it is
307+ * free to update it again.
308+ */
309+ pkg_therm_status_msr_val &= THERM_STATUS_CLEAR_PKG_MASK &
310+ ~PACKAGE_THERM_STATUS_HFI_UPDATED;
311+ wrmsrl(MSR_IA32_PACKAGE_THERM_STATUS, pkg_therm_status_msr_val);
312+
313+ queue_delayed_work(hfi_updates_wq, &hfi_instance->update_work,
314+ HFI_UPDATE_INTERVAL);
315+}
316+
317+static void init_hfi_cpu_index(struct hfi_cpu_info *info)
318+{
319+ union cpuid6_edx edx;
320+
321+ /* Do not re-read @cpu's index if it has already been initialized. */
322+ if (info->index > -1)
323+ return;
324+
325+ edx.full = cpuid_edx(CPUID_HFI_LEAF);
326+ info->index = edx.split.index;
327+}
328+
329+/*
330+ * The format of the HFI table depends on the number of capabilities that the
331+ * hardware supports. Keep a data structure to navigate the table.
332+ */
333+static void init_hfi_instance(struct hfi_instance *hfi_instance)
334+{
335+ /* The HFI header is below the time-stamp. */
336+ hfi_instance->hdr = hfi_instance->local_table +
337+ sizeof(*hfi_instance->timestamp);
338+
339+ /* The HFI data starts below the header. */
340+ hfi_instance->data = hfi_instance->hdr + hfi_features.hdr_size;
341+}
342+
343+/**
344+ * intel_hfi_online() - Enable HFI on @cpu
345+ * @cpu: CPU in which the HFI will be enabled
346+ *
347+ * Enable the HFI to be used in @cpu. The HFI is enabled at the die/package
348+ * level. The first CPU in the die/package to come online does the full HFI
349+ * initialization. Subsequent CPUs will just link themselves to the HFI
350+ * instance of their die/package.
351+ *
352+ * This function is called before enabling the thermal vector in the local APIC
353+ * in order to ensure that @cpu has an associated HFI instance when it receives
354+ * an HFI event.
355+ */
356+void intel_hfi_online(unsigned int cpu)
357+{
358+ struct hfi_instance *hfi_instance;
359+ struct hfi_cpu_info *info;
360+ phys_addr_t hw_table_pa;
361+ u64 msr_val;
362+ u16 die_id;
363+
364+ /* Nothing to do if hfi_instances are missing. */
365+ if (!hfi_instances)
366+ return;
367+
368+ /*
369+ * Link @cpu to the HFI instance of its package/die. It does not
370+ * matter whether the instance has been initialized.
371+ */
372+ info = &per_cpu(hfi_cpu_info, cpu);
373+ die_id = topology_logical_die_id(cpu);
374+ hfi_instance = info->hfi_instance;
375+ if (!hfi_instance) {
376+ if (die_id < 0 || die_id >= max_hfi_instances)
377+ return;
378+
379+ hfi_instance = &hfi_instances[die_id];
380+ info->hfi_instance = hfi_instance;
381+ }
382+
383+ init_hfi_cpu_index(info);
384+
385+ /*
386+ * Now check if the HFI instance of the package/die of @cpu has been
387+ * initialized (by checking its header). In such case, all we have to
388+ * do is to add @cpu to this instance's cpumask.
389+ */
390+ mutex_lock(&hfi_instance_lock);
391+ if (hfi_instance->hdr) {
392+ cpumask_set_cpu(cpu, hfi_instance->cpus);
393+ goto unlock;
394+ }
395+
396+ /*
397+ * Hardware is programmed with the physical address of the first page
398+ * frame of the table. Hence, the allocated memory must be page-aligned.
399+ */
400+ hfi_instance->hw_table = alloc_pages_exact(hfi_features.nr_table_pages,
401+ GFP_KERNEL | __GFP_ZERO);
402+ if (!hfi_instance->hw_table)
403+ goto unlock;
404+
405+ hw_table_pa = virt_to_phys(hfi_instance->hw_table);
406+
407+ /*
408+ * Allocate memory to keep a local copy of the table that
409+ * hardware generates.
410+ */
411+ hfi_instance->local_table = kzalloc(hfi_features.nr_table_pages << PAGE_SHIFT,
412+ GFP_KERNEL);
413+ if (!hfi_instance->local_table)
414+ goto free_hw_table;
415+
416+ /*
417+ * Program the address of the feedback table of this die/package. On
418+ * some processors, hardware remembers the old address of the HFI table
419+ * even after having been reprogrammed and re-enabled. Thus, do not free
420+ * the pages allocated for the table or reprogram the hardware with a
421+ * new base address. Namely, program the hardware only once.
422+ */
423+ msr_val = hw_table_pa | HW_FEEDBACK_PTR_VALID_BIT;
424+ wrmsrl(MSR_IA32_HW_FEEDBACK_PTR, msr_val);
425+
426+ init_hfi_instance(hfi_instance);
427+
428+ INIT_DELAYED_WORK(&hfi_instance->update_work, hfi_update_work_fn);
429+ raw_spin_lock_init(&hfi_instance->table_lock);
430+ raw_spin_lock_init(&hfi_instance->event_lock);
431+
432+ cpumask_set_cpu(cpu, hfi_instance->cpus);
433+
434+ /*
435+ * Enable the hardware feedback interface and never disable it. See
436+ * comment on programming the address of the table.
437+ */
438+ rdmsrl(MSR_IA32_HW_FEEDBACK_CONFIG, msr_val);
439+ msr_val |= HW_FEEDBACK_CONFIG_HFI_ENABLE_BIT;
440+ wrmsrl(MSR_IA32_HW_FEEDBACK_CONFIG, msr_val);
441+
442+unlock:
443+ mutex_unlock(&hfi_instance_lock);
444+ return;
445+
446+free_hw_table:
447+ free_pages_exact(hfi_instance->hw_table, hfi_features.nr_table_pages);
448+ goto unlock;
449+}
450+
451+/**
452+ * intel_hfi_offline() - Disable HFI on @cpu
453+ * @cpu: CPU in which the HFI will be disabled
454+ *
455+ * Remove @cpu from those covered by its HFI instance.
456+ *
457+ * On some processors, hardware remembers previous programming settings even
458+ * after being reprogrammed. Thus, keep HFI enabled even if all CPUs in the
459+ * die/package of @cpu are offline. See note in intel_hfi_online().
460+ */
461+void intel_hfi_offline(unsigned int cpu)
462+{
463+ struct hfi_cpu_info *info = &per_cpu(hfi_cpu_info, cpu);
464+ struct hfi_instance *hfi_instance;
465+
466+ /*
467+ * Check if @cpu as an associated, initialized (i.e., with a non-NULL
468+ * header). Also, HFI instances are only initialized if X86_FEATURE_HFI
469+ * is present.
470+ */
471+ hfi_instance = info->hfi_instance;
472+ if (!hfi_instance)
473+ return;
474+
475+ if (!hfi_instance->hdr)
476+ return;
477+
478+ mutex_lock(&hfi_instance_lock);
479+ cpumask_clear_cpu(cpu, hfi_instance->cpus);
480+ mutex_unlock(&hfi_instance_lock);
481+}
482+
483+static __init int hfi_parse_features(void)
484+{
485+ unsigned int nr_capabilities;
486+ union cpuid6_edx edx;
487+
488+ if (!boot_cpu_has(X86_FEATURE_HFI))
489+ return -ENODEV;
490+
491+ /*
492+ * If we are here we know that CPUID_HFI_LEAF exists. Parse the
493+ * supported capabilities and the size of the HFI table.
494+ */
495+ edx.full = cpuid_edx(CPUID_HFI_LEAF);
496+
497+ if (!edx.split.capabilities.split.performance) {
498+ pr_debug("Performance reporting not supported! Not using HFI\n");
499+ return -ENODEV;
500+ }
501+
502+ /*
503+ * The number of supported capabilities determines the number of
504+ * columns in the HFI table. Exclude the reserved bits.
505+ */
506+ edx.split.capabilities.split.__reserved = 0;
507+ nr_capabilities = hweight8(edx.split.capabilities.bits);
508+
509+ /* The number of 4KB pages required by the table */
510+ hfi_features.nr_table_pages = edx.split.table_pages + 1;
511+
512+ /*
513+ * The header contains change indications for each supported feature.
514+ * The size of the table header is rounded up to be a multiple of 8
515+ * bytes.
516+ */
517+ hfi_features.hdr_size = DIV_ROUND_UP(nr_capabilities, 8) * 8;
518+
519+ /*
520+ * Data of each logical processor is also rounded up to be a multiple
521+ * of 8 bytes.
522+ */
523+ hfi_features.cpu_stride = DIV_ROUND_UP(nr_capabilities, 8) * 8;
524+
525+ return 0;
526+}
527+
528+void __init intel_hfi_init(void)
529+{
530+ struct hfi_instance *hfi_instance;
531+ int i, j;
532+
533+ if (hfi_parse_features())
534+ return;
535+
536+ /* There is one HFI instance per die/package. */
537+ max_hfi_instances = topology_max_packages() *
538+ topology_max_die_per_package();
539+
540+ /*
541+ * This allocation may fail. CPU hotplug callbacks must check
542+ * for a null pointer.
543+ */
544+ hfi_instances = kcalloc(max_hfi_instances, sizeof(*hfi_instances),
545+ GFP_KERNEL);
546+ if (!hfi_instances)
547+ return;
548+
549+ for (i = 0; i < max_hfi_instances; i++) {
550+ hfi_instance = &hfi_instances[i];
551+ if (!zalloc_cpumask_var(&hfi_instance->cpus, GFP_KERNEL))
552+ goto err_nomem;
553+ }
554+
555+ hfi_updates_wq = create_singlethread_workqueue("hfi-updates");
556+ if (!hfi_updates_wq)
557+ goto err_nomem;
558+
559+ return;
560+
561+err_nomem:
562+ for (j = 0; j < i; ++j) {
563+ hfi_instance = &hfi_instances[j];
564+ free_cpumask_var(hfi_instance->cpus);
565+ }
566+
567+ kfree(hfi_instances);
568+ hfi_instances = NULL;
569+}
drivers/thermal/intel/intel_hfi.h+17−0
@@ -0,0 +1,17 @@
1+/* SPDX-License-Identifier: GPL-2.0 */
2+#ifndef _INTEL_HFI_H
3+#define _INTEL_HFI_H
4+
5+#if defined(CONFIG_INTEL_HFI_THERMAL)
6+void __init intel_hfi_init(void);
7+void intel_hfi_online(unsigned int cpu);
8+void intel_hfi_offline(unsigned int cpu);
9+void intel_hfi_process_event(__u64 pkg_therm_status_msr_val);
10+#else
11+static inline void intel_hfi_init(void) { }
12+static inline void intel_hfi_online(unsigned int cpu) { }
13+static inline void intel_hfi_offline(unsigned int cpu) { }
14+static inline void intel_hfi_process_event(__u64 pkg_therm_status_msr_val) { }
15+#endif /* CONFIG_INTEL_HFI_THERMAL */
16+
17+#endif /* _INTEL_HFI_H */
drivers/thermal/intel/intel_powerclamp.c+3−6
@@ -556,12 +556,9 @@ static void end_power_clamp(void) · 555 unmodified lines
556556 * stop faster.
557557 */
558558 clamping = false;
559− if (bitmap_weight(cpu_clamping_mask, num_possible_cpus())) {
560− for_each_set_bit(i, cpu_clamping_mask, num_possible_cpus()) {
561− pr_debug("clamping worker for cpu %d alive, destroy\n",
562− i);
563− stop_power_clamp_worker(i);
564− }
559+ for_each_set_bit(i, cpu_clamping_mask, num_possible_cpus()) {
560+ pr_debug("clamping worker for cpu %d alive, destroy\n", i);
561+ stop_power_clamp_worker(i);
565562 }
566563 }
567564
drivers/thermal/intel/therm_throt.c+22−0
@@ -32,6 +32,7 @@ · 31 unmodified lines
3232 #include <asm/irq.h>
3333 #include <asm/msr.h>
3434
35+#include "intel_hfi.h"
3536 #include "thermal_interrupt.h"
3637
3738 /* How long to wait between reporting thermal events */
@@ -475,6 +476,13 @@ static int thermal_throttle_online(unsigned int cpu) · 437 unmodified lines
475476 INIT_DELAYED_WORK(&state->package_throttle.therm_work, throttle_active_work);
476477 INIT_DELAYED_WORK(&state->core_throttle.therm_work, throttle_active_work);
477478
479+ /*
480+ * The first CPU coming online will enable the HFI. Usually this causes
481+ * hardware to issue an HFI thermal interrupt. Such interrupt will reach
482+ * the CPU once we enable the thermal vector in the local APIC.
483+ */
484+ intel_hfi_online(cpu);
485+
478486 /* Unmask the thermal vector after the above workqueues are initialized. */
479487 l = apic_read(APIC_LVTTHMR);
480488 apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
@@ -492,6 +500,8 @@ static int thermal_throttle_offline(unsigned int cpu) · 11 unmodified lines
492500 l = apic_read(APIC_LVTTHMR);
493501 apic_write(APIC_LVTTHMR, l | APIC_LVT_MASKED);
494502
503+ intel_hfi_offline(cpu);
504+
495505 cancel_delayed_work_sync(&state->package_throttle.therm_work);
496506 cancel_delayed_work_sync(&state->core_throttle.therm_work);
497507
@@ -509,6 +519,8 @@ static __init int thermal_throttle_init_device(void) · 11 unmodified lines
509519 if (!atomic_read(&therm_throt_en))
510520 return 0;
511521
522+ intel_hfi_init();
523+
512524 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/therm:online",
513525 thermal_throttle_online,
514526 thermal_throttle_offline);
@@ -608,6 +620,10 @@ void intel_thermal_interrupt(void) · 93 unmodified lines
608620 PACKAGE_THERM_STATUS_POWER_LIMIT,
609621 POWER_LIMIT_EVENT,
610622 PACKAGE_LEVEL);
623+
624+ if (this_cpu_has(X86_FEATURE_HFI))
625+ intel_hfi_process_event(msr_val &
626+ PACKAGE_THERM_STATUS_HFI_UPDATED);
611627 }
612628 }
613629
@@ -717,6 +733,12 @@ void intel_init_thermal(struct cpuinfo_x86 *c) · 103 unmodified lines
717733 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
718734 l | (PACKAGE_THERM_INT_LOW_ENABLE
719735 | PACKAGE_THERM_INT_HIGH_ENABLE), h);
736+
737+ if (cpu_has(c, X86_FEATURE_HFI)) {
738+ rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
739+ wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
740+ l | PACKAGE_THERM_INT_HFI_ENABLE, h);
741+ }
720742 }
721743
722744 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
drivers/thermal/qcom/lmh.c+35−27
@@ -28,6 +28,8 @@ · 27 unmodified lines
2828
2929 #define LMH_REG_DCVS_INTR_CLR 0x8
3030
31+#define LMH_ENABLE_ALGOS 1
32+
3133 struct lmh_hw_data {
3234 void __iomem *base;
3335 struct irq_domain *domain;
@@ -90,6 +92,7 @@ static int lmh_probe(struct platform_device *pdev) · 56 unmodified lines
9092 struct device_node *cpu_node;
9193 struct lmh_hw_data *lmh_data;
9294 int temp_low, temp_high, temp_arm, cpu_id, ret;
95+ unsigned int enable_alg;
9396 u32 node_id;
9497
9598 lmh_data = devm_kzalloc(dev, sizeof(*lmh_data), GFP_KERNEL);
@@ -141,32 +144,36 @@ static int lmh_probe(struct platform_device *pdev) · 45 unmodified lines
141144 if (!qcom_scm_lmh_dcvsh_available())
142145 return -EINVAL;
143146
144− ret = qcom_scm_lmh_dcvsh(LMH_SUB_FN_CRNT, LMH_ALGO_MODE_ENABLE, 1,
145− LMH_NODE_DCVS, node_id, 0);
146− if (ret)
147− dev_err(dev, "Error %d enabling current subfunction\n", ret);
148−
149− ret = qcom_scm_lmh_dcvsh(LMH_SUB_FN_REL, LMH_ALGO_MODE_ENABLE, 1,
150− LMH_NODE_DCVS, node_id, 0);
151− if (ret)
152− dev_err(dev, "Error %d enabling reliability subfunction\n", ret);
153−
154− ret = qcom_scm_lmh_dcvsh(LMH_SUB_FN_BCL, LMH_ALGO_MODE_ENABLE, 1,
155− LMH_NODE_DCVS, node_id, 0);
156− if (ret)
157− dev_err(dev, "Error %d enabling BCL subfunction\n", ret);
158−
159− ret = qcom_scm_lmh_dcvsh(LMH_SUB_FN_THERMAL, LMH_ALGO_MODE_ENABLE, 1,
160− LMH_NODE_DCVS, node_id, 0);
161− if (ret) {
162− dev_err(dev, "Error %d enabling thermal subfunction\n", ret);
163− return ret;
164− }
165−
166− ret = qcom_scm_lmh_profile_change(0x1);
167− if (ret) {
168− dev_err(dev, "Error %d changing profile\n", ret);
169− return ret;
147+ enable_alg = (uintptr_t)of_device_get_match_data(dev);
148+
149+ if (enable_alg) {
150+ ret = qcom_scm_lmh_dcvsh(LMH_SUB_FN_CRNT, LMH_ALGO_MODE_ENABLE, 1,
151+ LMH_NODE_DCVS, node_id, 0);
152+ if (ret)
153+ dev_err(dev, "Error %d enabling current subfunction\n", ret);
154+
155+ ret = qcom_scm_lmh_dcvsh(LMH_SUB_FN_REL, LMH_ALGO_MODE_ENABLE, 1,
156+ LMH_NODE_DCVS, node_id, 0);
157+ if (ret)
158+ dev_err(dev, "Error %d enabling reliability subfunction\n", ret);
159+
160+ ret = qcom_scm_lmh_dcvsh(LMH_SUB_FN_BCL, LMH_ALGO_MODE_ENABLE, 1,
161+ LMH_NODE_DCVS, node_id, 0);
162+ if (ret)
163+ dev_err(dev, "Error %d enabling BCL subfunction\n", ret);
164+
165+ ret = qcom_scm_lmh_dcvsh(LMH_SUB_FN_THERMAL, LMH_ALGO_MODE_ENABLE, 1,
166+ LMH_NODE_DCVS, node_id, 0);
167+ if (ret) {
168+ dev_err(dev, "Error %d enabling thermal subfunction\n", ret);
169+ return ret;
170+ }
171+
172+ ret = qcom_scm_lmh_profile_change(0x1);
173+ if (ret) {
174+ dev_err(dev, "Error %d changing profile\n", ret);
175+ return ret;
176+ }
170177 }
171178
172179 /* Set default thermal trips */
@@ -213,7 +220,8 @@ static int lmh_probe(struct platform_device *pdev) · 40 unmodified lines
213220 }
214221
215222 static const struct of_device_id lmh_table[] = {
216− { .compatible = "qcom,sdm845-lmh", },
223+ { .compatible = "qcom,sdm845-lmh", .data = (void *)LMH_ENABLE_ALGOS},
224+ { .compatible = "qcom,sm8150-lmh", },
217225 {}
218226 };
219227 MODULE_DEVICE_TABLE(of, lmh_table);
drivers/thermal/qcom/tsens.c+5−0
@@ -18,6 +18,7 @@ · 17 unmodified lines
1818 #include <linux/regmap.h>
1919 #include <linux/slab.h>
2020 #include <linux/thermal.h>
21+#include "../thermal_hwmon.h"
2122 #include "tsens.h"
2223
2324 /**
@@ -1060,6 +1061,10 @@ static int tsens_register(struct tsens_priv *priv) · 1036 unmodified lines
10601061 priv->sensor[i].tzd = tzd;
10611062 if (priv->ops->enable)
10621063 priv->ops->enable(priv, i);
1064+
1065+ if (devm_thermal_add_hwmon_sysfs(tzd))
1066+ dev_warn(priv->dev,
1067+ "Failed to add hwmon sysfs attributes\n");
10631068 }
10641069
10651070 /* VER_0 require to set MIN and MAX THRESH
drivers/thermal/tegra/tegra-bpmp-thermal.c+12−1
@@ -52,6 +52,8 @@ static int tegra_bpmp_thermal_get_temp(void *data, int *out_temp) · 51 unmodified lines
5252 err = tegra_bpmp_transfer(zone->tegra->bpmp, &msg);
5353 if (err)
5454 return err;
55+ if (msg.rx.ret)
56+ return -EINVAL;
5557
5658 *out_temp = reply.get_temp.temp;
5759
@@ -63,6 +65,7 @@ static int tegra_bpmp_thermal_set_trips(void *data, int low, int high) · 5 unmodified lines
6365 struct tegra_bpmp_thermal_zone *zone = data;
6466 struct mrq_thermal_host_to_bpmp_request req;
6567 struct tegra_bpmp_message msg;
68+ int err;
6669
6770 memset(&req, 0, sizeof(req));
6871 req.type = CMD_THERMAL_SET_TRIP;
@@ -76,7 +79,13 @@ static int tegra_bpmp_thermal_set_trips(void *data, int low, int high) · 7 unmodified lines
7679 msg.tx.data = &req;
7780 msg.tx.size = sizeof(req);
7881
79− return tegra_bpmp_transfer(zone->tegra->bpmp, &msg);
82+ err = tegra_bpmp_transfer(zone->tegra->bpmp, &msg);
83+ if (err)
84+ return err;
85+ if (msg.rx.ret)
86+ return -EINVAL;
87+
88+ return 0;
8089 }
8190
8291 static void tz_device_update_work_fn(struct work_struct *work)
@@ -140,6 +149,8 @@ static int tegra_bpmp_thermal_get_num_zones(struct tegra_bpmp *bpmp, · 57 unmodified lines
140149 err = tegra_bpmp_transfer(bpmp, &msg);
141150 if (err)
142151 return err;
152+ if (msg.rx.ret)
153+ return -EINVAL;
143154
144155 *num_zones = reply.get_num_zones.num;
145156
drivers/thermal/thermal_netlink.c+53−0
@@ -43,6 +43,11 @@ static const struct nla_policy thermal_genl_policy[THERMAL_GENL_ATTR_MAX + 1] = · 42 unmodified lines
4343 [THERMAL_GENL_ATTR_CDEV_MAX_STATE] = { .type = NLA_U32 },
4444 [THERMAL_GENL_ATTR_CDEV_NAME] = { .type = NLA_STRING,
4545 .len = THERMAL_NAME_LENGTH },
46+ /* CPU capabilities */
47+ [THERMAL_GENL_ATTR_CPU_CAPABILITY] = { .type = NLA_NESTED },
48+ [THERMAL_GENL_ATTR_CPU_CAPABILITY_ID] = { .type = NLA_U32 },
49+ [THERMAL_GENL_ATTR_CPU_CAPABILITY_PERFORMANCE] = { .type = NLA_U32 },
50+ [THERMAL_GENL_ATTR_CPU_CAPABILITY_EFFICIENCY] = { .type = NLA_U32 },
4651 };
4752
4853 struct param {
@@ -58,6 +63,8 @@ struct param { · 9 unmodified lines
5863 int temp;
5964 int cdev_state;
6065 int cdev_max_state;
66+ struct thermal_genl_cpu_caps *cpu_capabilities;
67+ int cpu_capabilities_count;
6168 };
6269
6370 typedef int (*cb_t)(struct param *);
@@ -190,6 +197,42 @@ static int thermal_genl_event_gov_change(struct param *p) · 126 unmodified lines
190197 return 0;
191198 }
192199
200+static int thermal_genl_event_cpu_capability_change(struct param *p)
201+{
202+ struct thermal_genl_cpu_caps *cpu_cap = p->cpu_capabilities;
203+ struct sk_buff *msg = p->msg;
204+ struct nlattr *start_cap;
205+ int i;
206+
207+ start_cap = nla_nest_start(msg, THERMAL_GENL_ATTR_CPU_CAPABILITY);
208+ if (!start_cap)
209+ return -EMSGSIZE;
210+
211+ for (i = 0; i < p->cpu_capabilities_count; ++i) {
212+ if (nla_put_u32(msg, THERMAL_GENL_ATTR_CPU_CAPABILITY_ID,
213+ cpu_cap->cpu))
214+ goto out_cancel_nest;
215+
216+ if (nla_put_u32(msg, THERMAL_GENL_ATTR_CPU_CAPABILITY_PERFORMANCE,
217+ cpu_cap->performance))
218+ goto out_cancel_nest;
219+
220+ if (nla_put_u32(msg, THERMAL_GENL_ATTR_CPU_CAPABILITY_EFFICIENCY,
221+ cpu_cap->efficiency))
222+ goto out_cancel_nest;
223+
224+ ++cpu_cap;
225+ }
226+
227+ nla_nest_end(msg, start_cap);
228+
229+ return 0;
230+out_cancel_nest:
231+ nla_nest_cancel(msg, start_cap);
232+
233+ return -EMSGSIZE;
234+}
235+
193236 int thermal_genl_event_tz_delete(struct param *p)
194237 __attribute__((alias("thermal_genl_event_tz")));
195238
@@ -219,6 +262,7 @@ static cb_t event_cb[] = { · 23 unmodified lines
219262 [THERMAL_GENL_EVENT_CDEV_DELETE] = thermal_genl_event_cdev_delete,
220263 [THERMAL_GENL_EVENT_CDEV_STATE_UPDATE] = thermal_genl_event_cdev_state_update,
221264 [THERMAL_GENL_EVENT_TZ_GOV_CHANGE] = thermal_genl_event_gov_change,
265+ [THERMAL_GENL_EVENT_CPU_CAPABILITY_CHANGE] = thermal_genl_event_cpu_capability_change,
222266 };
223267
224268 /*
@@ -356,6 +400,15 @@ int thermal_notify_tz_gov_change(int tz_id, const char *name) · 131 unmodified lines
356400 return thermal_genl_send_event(THERMAL_GENL_EVENT_TZ_GOV_CHANGE, &p);
357401 }
358402
403+int thermal_genl_cpu_capability_event(int count,
404+ struct thermal_genl_cpu_caps *caps)
405+{
406+ struct param p = { .cpu_capabilities_count = count, .cpu_capabilities = caps };
407+
408+ return thermal_genl_send_event(THERMAL_GENL_EVENT_CPU_CAPABILITY_CHANGE, &p);
409+}
410+EXPORT_SYMBOL_GPL(thermal_genl_cpu_capability_event);
411+
359412 /*************************** Command encoding ********************************/
360413
361414 static int __thermal_genl_cmd_tz_get_id(struct thermal_zone_device *tz,
drivers/thermal/thermal_netlink.h+14−0
@@ -4,6 +4,12 @@ · 3 unmodified lines
44 * Author: Daniel Lezcano <daniel.lezcano@linaro.org>
55 */
66
7+struct thermal_genl_cpu_caps {
8+ int cpu;
9+ int performance;
10+ int efficiency;
11+};
12+
713 /* Netlink notification function */
814 #ifdef CONFIG_THERMAL_NETLINK
915 int __init thermal_netlink_init(void);
@@ -23,6 +29,8 @@ int thermal_notify_cdev_add(int cdev_id, const char *name, int max_state); · 13 unmodified lines
2329 int thermal_notify_cdev_delete(int cdev_id);
2430 int thermal_notify_tz_gov_change(int tz_id, const char *name);
2531 int thermal_genl_sampling_temp(int id, int temp);
32+int thermal_genl_cpu_capability_event(int count,
33+ struct thermal_genl_cpu_caps *caps);
2634 #else
2735 static inline int thermal_netlink_init(void)
2836 {
@@ -101,4 +109,10 @@ static inline int thermal_genl_sampling_temp(int id, int temp) · 72 unmodified lines
101109 {
102110 return 0;
103111 }
112+
113+static inline int thermal_genl_cpu_capability_event(int count, struct thermal_genl_cpu_caps *caps)
114+{
115+ return 0;
116+}
117+
104118 #endif /* CONFIG_THERMAL_NETLINK */
drivers/thermal/ti-soc-thermal/ti-thermal-common.c+4−8
@@ -21,6 +21,7 @@ · 20 unmodified lines
2121
2222 #include "ti-thermal.h"
2323 #include "ti-bandgap.h"
24+#include "../thermal_hwmon.h"
2425
2526 /* common data structures */
2627 struct ti_thermal_data {
@@ -106,14 +107,6 @@ static inline int __ti_thermal_get_temp(void *devdata, int *temp) · 79 unmodified lines
106107 return ret;
107108 }
108109
109−static inline int ti_thermal_get_temp(struct thermal_zone_device *thermal,
110− int *temp)
111−{
112− struct ti_thermal_data *data = thermal->devdata;
113−
114− return __ti_thermal_get_temp(data, temp);
115−}
116−
117110 static int __ti_thermal_get_trend(void *p, int trip, enum thermal_trend *trend)
118111 {
119112 struct ti_thermal_data *data = p;
@@ -189,6 +182,9 @@ int ti_thermal_expose_sensor(struct ti_bandgap *bgp, int id, · 69 unmodified lines
189182 ti_bandgap_set_sensor_data(bgp, id, data);
190183 ti_bandgap_write_update_interval(bgp, data->sensor_id, interval);
191184
185+ if (devm_thermal_add_hwmon_sysfs(data->ti_thermal))
186+ dev_warn(bgp->dev, "failed to add hwmon sysfs attributes\n");
187+
192188 return 0;
193189 }
194190
include/uapi/linux/thermal.h+5−1
@@ -44,7 +44,10 @@ enum thermal_genl_attr { · 43 unmodified lines
4444 THERMAL_GENL_ATTR_CDEV_MAX_STATE,
4545 THERMAL_GENL_ATTR_CDEV_NAME,
4646 THERMAL_GENL_ATTR_GOV_NAME,
47−
47+ THERMAL_GENL_ATTR_CPU_CAPABILITY,
48+ THERMAL_GENL_ATTR_CPU_CAPABILITY_ID,
49+ THERMAL_GENL_ATTR_CPU_CAPABILITY_PERFORMANCE,
50+ THERMAL_GENL_ATTR_CPU_CAPABILITY_EFFICIENCY,
4851 __THERMAL_GENL_ATTR_MAX,
4952 };
5053 #define THERMAL_GENL_ATTR_MAX (__THERMAL_GENL_ATTR_MAX - 1)
@@ -71,6 +74,7 @@ enum thermal_genl_event { · 20 unmodified lines
7174 THERMAL_GENL_EVENT_CDEV_DELETE, /* Cdev unbound */
7275 THERMAL_GENL_EVENT_CDEV_STATE_UPDATE, /* Cdev state updated */
7376 THERMAL_GENL_EVENT_TZ_GOV_CHANGE, /* Governor policy changed */
77+ THERMAL_GENL_EVENT_CPU_CAPABILITY_CHANGE, /* CPU capability changed */
7478 __THERMAL_GENL_EVENT_MAX,
7579 };
7680 #define THERMAL_GENL_EVENT_MAX (__THERMAL_GENL_EVENT_MAX - 1)
tools/power/x86/intel-speed-select/Build+1−1
@@ -1 +1 @@
1−intel-speed-select-y += isst-config.o isst-core.o isst-display.o
1+intel-speed-select-y += isst-config.o isst-core.o isst-display.o isst-daemon.o hfi-events.o
tools/power/x86/intel-speed-select/Makefile+7−3
@@ -13,8 +13,8 @@ endif · 12 unmodified lines
1313 # Do not use make's built-in rules
1414 # (this improves performance and avoids hard-to-debug behaviour);
1515 MAKEFLAGS += -r
16−
17−override CFLAGS += -O2 -Wall -g -D_GNU_SOURCE -I$(OUTPUT)include
16+override CFLAGS += -O2 -Wall -g -D_GNU_SOURCE -I$(OUTPUT)include -I/usr/include/libnl3
17+override LDFLAGS += -lnl-genl-3 -lnl-3
1818
1919 ALL_TARGETS := intel-speed-select
2020 ALL_PROGRAMS := $(patsubst %,$(OUTPUT)%,$(ALL_TARGETS))
@@ -31,7 +31,11 @@ $(OUTPUT)include/linux/isst_if.h: ../../../../include/uapi/linux/isst_if.h · 10 unmodified lines
3131 mkdir -p $(OUTPUT)include/linux 2>&1 || true
3232 ln -sf $(CURDIR)/../../../../include/uapi/linux/isst_if.h $@
3333
34−prepare: $(OUTPUT)include/linux/isst_if.h
34+$(OUTPUT)include/linux/thermal.h: ../../../../include/uapi/linux/thermal.h
35+ mkdir -p $(OUTPUT)include/linux 2>&1 || true
36+ ln -sf $(CURDIR)/../../../../include/uapi/linux/thermal.h $@
37+
38+prepare: $(OUTPUT)include/linux/isst_if.h $(OUTPUT)include/linux/thermal.h
3539
3640 ISST_IN := $(OUTPUT)intel-speed-select-in.o
3741
tools/power/x86/intel-speed-select/hfi-events.c+309−0
@@ -0,0 +1,309 @@
1+// SPDX-License-Identifier: GPL-2.0
2+/*
3+ * Intel Speed Select -- Read HFI events for OOB
4+ * Copyright (c) 2022 Intel Corporation.
5+ */
6+
7+/*
8+ * This file incorporates work covered by the following copyright and
9+ * permission notice:
10+
11+ * WPA Supplicant - driver interaction with Linux nl80211/cfg80211
12+ * Copyright (c) 2003-2008, Jouni Malinen <j@w1.fi>
13+ *
14+ * This program is free software; you can redistribute it and/or modify
15+ * it under the terms of the GNU General Public License version 2 as
16+ * published by the Free Software Foundation.
17+ *
18+ * Alternatively, this software may be distributed under the terms of
19+ * BSD license.
20+ *
21+ * Requires
22+ * libnl-genl-3-dev
23+ *
24+ * For Fedora/CenOS
25+ * dnf install libnl3-devel
26+ * For Ubuntu
27+ * apt install libnl-3-dev libnl-genl-3-dev
28+ */
29+
30+#include <stdio.h>
31+#include <stdlib.h>
32+#include <stdarg.h>
33+#include <string.h>
34+#include <unistd.h>
35+#include <fcntl.h>
36+#include <sys/file.h>
37+#include <sys/types.h>
38+#include <sys/stat.h>
39+#include <errno.h>
40+#include <getopt.h>
41+#include <signal.h>
42+#include <netlink/genl/genl.h>
43+#include <netlink/genl/family.h>
44+#include <netlink/genl/ctrl.h>
45+
46+#include <linux/thermal.h>
47+#include "isst.h"
48+
49+struct hfi_event_data {
50+ struct nl_sock *nl_handle;
51+ struct nl_cb *nl_cb;
52+};
53+
54+struct hfi_event_data drv;
55+
56+static int ack_handler(struct nl_msg *msg, void *arg)
57+{
58+ int *err = arg;
59+ *err = 0;
60+ return NL_STOP;
61+}
62+
63+static int finish_handler(struct nl_msg *msg, void *arg)
64+{
65+ int *ret = arg;
66+ *ret = 0;
67+ return NL_SKIP;
68+}
69+
70+static int error_handler(struct sockaddr_nl *nla, struct nlmsgerr *err,
71+ void *arg)
72+{
73+ int *ret = arg;
74+ *ret = err->error;
75+ return NL_SKIP;
76+}
77+
78+static int seq_check_handler(struct nl_msg *msg, void *arg)
79+{
80+ return NL_OK;
81+}
82+
83+static int send_and_recv_msgs(struct hfi_event_data *drv,
84+ struct nl_msg *msg,
85+ int (*valid_handler)(struct nl_msg *, void *),
86+ void *valid_data)
87+{
88+ struct nl_cb *cb;
89+ int err = -ENOMEM;
90+
91+ cb = nl_cb_clone(drv->nl_cb);
92+ if (!cb)
93+ goto out;
94+
95+ err = nl_send_auto_complete(drv->nl_handle, msg);
96+ if (err < 0)
97+ goto out;
98+
99+ err = 1;
100+
101+ nl_cb_err(cb, NL_CB_CUSTOM, error_handler, &err);
102+ nl_cb_set(cb, NL_CB_FINISH, NL_CB_CUSTOM, finish_handler, &err);
103+ nl_cb_set(cb, NL_CB_ACK, NL_CB_CUSTOM, ack_handler, &err);
104+
105+ if (valid_handler)
106+ nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM,
107+ valid_handler, valid_data);
108+
109+ while (err > 0)
110+ nl_recvmsgs(drv->nl_handle, cb);
111+ out:
112+ nl_cb_put(cb);
113+ nlmsg_free(msg);
114+ return err;
115+}
116+
117+struct family_data {
118+ const char *group;
119+ int id;
120+};
121+
122+static int family_handler(struct nl_msg *msg, void *arg)
123+{
124+ struct family_data *res = arg;
125+ struct nlattr *tb[CTRL_ATTR_MAX + 1];
126+ struct genlmsghdr *gnlh = nlmsg_data(nlmsg_hdr(msg));
127+ struct nlattr *mcgrp;
128+ int i;
129+
130+ nla_parse(tb, CTRL_ATTR_MAX, genlmsg_attrdata(gnlh, 0),
131+ genlmsg_attrlen(gnlh, 0), NULL);
132+ if (!tb[CTRL_ATTR_MCAST_GROUPS])
133+ return NL_SKIP;
134+
135+ nla_for_each_nested(mcgrp, tb[CTRL_ATTR_MCAST_GROUPS], i) {
136+ struct nlattr *tb2[CTRL_ATTR_MCAST_GRP_MAX + 1];
137+ nla_parse(tb2, CTRL_ATTR_MCAST_GRP_MAX, nla_data(mcgrp),
138+ nla_len(mcgrp), NULL);
139+ if (!tb2[CTRL_ATTR_MCAST_GRP_NAME] ||
140+ !tb2[CTRL_ATTR_MCAST_GRP_ID] ||
141+ strncmp(nla_data(tb2[CTRL_ATTR_MCAST_GRP_NAME]),
142+ res->group,
143+ nla_len(tb2[CTRL_ATTR_MCAST_GRP_NAME])) != 0)
144+ continue;
145+ res->id = nla_get_u32(tb2[CTRL_ATTR_MCAST_GRP_ID]);
146+ break;
147+ };
148+
149+ return 0;
150+}
151+
152+static int nl_get_multicast_id(struct hfi_event_data *drv,
153+ const char *family, const char *group)
154+{
155+ struct nl_msg *msg;
156+ int ret = -1;
157+ struct family_data res = { group, -ENOENT };
158+
159+ msg = nlmsg_alloc();
160+ if (!msg)
161+ return -ENOMEM;
162+ genlmsg_put(msg, 0, 0, genl_ctrl_resolve(drv->nl_handle, "nlctrl"),
163+ 0, 0, CTRL_CMD_GETFAMILY, 0);
164+ NLA_PUT_STRING(msg, CTRL_ATTR_FAMILY_NAME, family);
165+
166+ ret = send_and_recv_msgs(drv, msg, family_handler, &res);
167+ msg = NULL;
168+ if (ret == 0)
169+ ret = res.id;
170+
171+nla_put_failure:
172+ nlmsg_free(msg);
173+ return ret;
174+}
175+
176+struct perf_cap {
177+ int cpu;
178+ int perf;
179+ int eff;
180+};
181+
182+static void process_hfi_event(struct perf_cap *perf_cap)
183+{
184+ process_level_change(perf_cap->cpu);
185+}
186+
187+static int handle_event(struct nl_msg *n, void *arg)
188+{
189+ struct nlmsghdr *nlh = nlmsg_hdr(n);
190+ struct genlmsghdr *genlhdr = genlmsg_hdr(nlh);
191+ struct nlattr *attrs[THERMAL_GENL_ATTR_MAX + 1];
192+ int ret;
193+ struct perf_cap perf_cap;
194+
195+ ret = genlmsg_parse(nlh, 0, attrs, THERMAL_GENL_ATTR_MAX, NULL);
196+
197+ debug_printf("Received event %d parse_rer:%d\n", genlhdr->cmd, ret);
198+ if (genlhdr->cmd == THERMAL_GENL_EVENT_CPU_CAPABILITY_CHANGE) {
199+ struct nlattr *cap;
200+ int j, index = 0;
201+
202+ debug_printf("THERMAL_GENL_EVENT_CPU_CAPABILITY_CHANGE\n");
203+ nla_for_each_nested(cap, attrs[THERMAL_GENL_ATTR_CPU_CAPABILITY], j) {
204+ switch (index) {
205+ case 0:
206+ perf_cap.cpu = nla_get_u32(cap);
207+ break;
208+ case 1:
209+ perf_cap.perf = nla_get_u32(cap);
210+ break;
211+ case 2:
212+ perf_cap.eff = nla_get_u32(cap);
213+ break;
214+ default:
215+ break;
216+ }
217+ ++index;
218+ if (index == 3) {
219+ index = 0;
220+ process_hfi_event(&perf_cap);
221+ }
222+ }
223+ }
224+
225+ return 0;
226+}
227+
228+static int _hfi_exit;
229+
230+static int check_hf_suport(void)
231+{
232+ unsigned int eax = 0, ebx = 0, ecx = 0, edx = 0;
233+
234+ __cpuid(6, eax, ebx, ecx, edx);
235+ if (eax & BIT(19))
236+ return 1;
237+
238+ return 0;
239+}
240+
241+int hfi_main(void)
242+{
243+ struct nl_sock *sock;
244+ struct nl_cb *cb;
245+ int err = 0;
246+ int mcast_id;
247+ int no_block = 0;
248+
249+ if (!check_hf_suport()) {
250+ fprintf(stderr, "CPU Doesn't support HFI\n");
251+ return -1;
252+ }
253+
254+ sock = nl_socket_alloc();
255+ if (!sock) {
256+ fprintf(stderr, "nl_socket_alloc failed\n");
257+ return -1;
258+ }
259+
260+ if (genl_connect(sock)) {
261+ fprintf(stderr, "genl_connect(sk_event) failed\n");
262+ goto free_sock;
263+ }
264+
265+ drv.nl_handle = sock;
266+ drv.nl_cb = cb = nl_cb_alloc(NL_CB_DEFAULT);
267+ if (drv.nl_cb == NULL) {
268+ printf("Failed to allocate netlink callbacks");
269+ goto free_sock;
270+ }
271+
272+ mcast_id = nl_get_multicast_id(&drv, THERMAL_GENL_FAMILY_NAME,
273+ THERMAL_GENL_EVENT_GROUP_NAME);
274+ if (mcast_id < 0) {
275+ fprintf(stderr, "nl_get_multicast_id failed\n");
276+ goto free_sock;
277+ }
278+
279+ if (nl_socket_add_membership(sock, mcast_id)) {
280+ fprintf(stderr, "nl_socket_add_membership failed");
281+ goto free_sock;
282+ }
283+
284+ nl_cb_set(cb, NL_CB_SEQ_CHECK, NL_CB_CUSTOM, seq_check_handler, 0);
285+ nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, handle_event, NULL);
286+
287+ if (no_block)
288+ nl_socket_set_nonblocking(sock);
289+
290+ debug_printf("hfi is initialized\n");
291+
292+ while (!_hfi_exit && !err) {
293+ err = nl_recvmsgs(sock, cb);
294+ debug_printf("nl_recv_message err:%d\n", err);
295+ }
296+
297+ return 0;
298+
299+ /* Netlink library doesn't have calls to dealloc cb or disconnect */
300+free_sock:
301+ nl_socket_free(sock);
302+
303+ return -1;
304+}
305+
306+void hfi_exit(void)
307+{
308+ _hfi_exit = 1;
309+}
tools/power/x86/intel-speed-select/isst-config.c+43−11
@@ -15,7 +15,8 @@ struct process_cmd_struct { · 14 unmodified lines
1515 int arg;
1616 };
1717
18−static const char *version_str = "v1.11";
18+static const char *version_str = "v1.12";
19+
1920 static const int supported_api_ver = 1;
2021 static struct isst_if_platform_info isst_platform_info;
2122 static char *progname;
@@ -368,7 +369,7 @@ int get_topo_max_cpus(void) · 346 unmodified lines
368369 return topo_max_cpus;
369370 }
370371
371−static void set_cpu_online_offline(int cpu, int state)
372+void set_cpu_online_offline(int cpu, int state)
372373 {
373374 char buffer[128];
374375 int fd, ret;
@@ -409,12 +410,10 @@ static void force_all_cpus_online(void) · 34 unmodified lines
409410 unlink("/var/run/isst_cpu_topology.dat");
410411 }
411412
412−#define MAX_PACKAGE_COUNT 8
413−#define MAX_DIE_PER_PACKAGE 2
414−static void for_each_online_package_in_set(void (*callback)(int, void *, void *,
415− void *, void *),
416− void *arg1, void *arg2, void *arg3,
417− void *arg4)
413+void for_each_online_package_in_set(void (*callback)(int, void *, void *,
414+ void *, void *),
415+ void *arg1, void *arg2, void *arg3,
416+ void *arg4)
418417 {
419418 int max_packages[MAX_PACKAGE_COUNT * MAX_PACKAGE_COUNT];
420419 int pkg_index = 0, i;
@@ -2803,7 +2802,9 @@ static void usage(void) · 2382 unmodified lines
28032802 printf("\t[-p|--pause] : Delay between two mail box commands in milliseconds\n");
28042803 printf("\t[-r|--retry] : Retry count for mail box commands on failure, default 3\n");
28052804 printf("\t[-v|--version] : Print version\n");
2806−
2805+ printf("\t[-b|--oob : Start a daemon to process HFI events for perf profile change from Out of Band agent.\n");
2806+ printf("\t[-n|--no-daemon : Don't run as daemon. By default --oob will turn on daemon mode\n");
2807+ printf("\t[-w|--delay : Delay for reading config level state change in OOB poll mode.\n");
28072808 printf("\nResult format\n");
28082809 printf("\tResult display uses a common format for each command:\n");
28092810 printf("\tResults are formatted in text/JSON with\n");
@@ -2837,6 +2838,9 @@ static void cmdline(int argc, char **argv) · 27 unmodified lines
28372838 int opt, force_cpus_online = 0;
28382839 int option_index = 0;
28392840 int ret;
2841+ int oob_mode = 0;
2842+ int poll_interval = -1;
2843+ int no_daemon = 0;
28402844
28412845 static struct option long_options[] = {
28422846 { "all-cpus-online", no_argument, 0, 'a' },
@@ -2849,6 +2853,9 @@ static void cmdline(int argc, char **argv) · 6 unmodified lines
28492853 { "out", required_argument, 0, 'o' },
28502854 { "retry", required_argument, 0, 'r' },
28512855 { "version", no_argument, 0, 'v' },
2856+ { "oob", no_argument, 0, 'b' },
2857+ { "no-daemon", no_argument, 0, 'n' },
2858+ { "poll-interval", required_argument, 0, 'w' },
28522859 { 0, 0, 0, 0 }
28532860 };
28542861
@@ -2875,7 +2882,7 @@ static void cmdline(int argc, char **argv) · 20 unmodified lines
28752882 }
28762883
28772884 progname = argv[0];
2878− while ((opt = getopt_long_only(argc, argv, "+c:df:hio:va", long_options,
2885+ while ((opt = getopt_long_only(argc, argv, "+c:df:hio:vabw:n", long_options,
28792886 &option_index)) != -1) {
28802887 switch (opt) {
28812888 case 'a':
@@ -2920,12 +2927,26 @@ static void cmdline(int argc, char **argv) · 38 unmodified lines
29202927 case 'v':
29212928 print_version();
29222929 break;
2930+ case 'b':
2931+ oob_mode = 1;
2932+ break;
2933+ case 'n':
2934+ no_daemon = 1;
2935+ break;
2936+ case 'w':
2937+ ret = strtol(optarg, &ptr, 10);
2938+ if (!ret) {
2939+ fprintf(stderr, "Invalid poll interval count\n");
2940+ exit(0);
2941+ }
2942+ poll_interval = ret;
2943+ break;
29232944 default:
29242945 usage();
29252946 }
29262947 }
29272948
2928− if (optind > (argc - 2)) {
2949+ if (optind > (argc - 2) && !oob_mode) {
29292950 usage();
29302951 exit(0);
29312952 }
@@ -2936,6 +2957,17 @@ static void cmdline(int argc, char **argv) · 4 unmodified lines
29362957 set_cpu_present_cpu_mask();
29372958 set_cpu_target_cpu_mask();
29382959
2960+ if (oob_mode) {
2961+ create_cpu_map();
2962+ if (debug_flag)
2963+ fprintf(stderr, "OOB mode is enabled in debug mode\n");
2964+
2965+ ret = isst_daemon(debug_flag, poll_interval, no_daemon);
2966+ if (ret)
2967+ fprintf(stderr, "OOB mode enable failed\n");
2968+ goto out;
2969+ }
2970+
29392971 if (!is_clx_n_platform()) {
29402972 ret = isst_fill_platform_info();
29412973 if (ret)
tools/power/x86/intel-speed-select/isst-daemon.c+244−0
@@ -0,0 +1,244 @@
1+// SPDX-License-Identifier: GPL-2.0
2+/*
3+ * Intel Speed Select -- Allow speed select to daemonize
4+ * Copyright (c) 2022 Intel Corporation.
5+ */
6+
7+#include <stdio.h>
8+#include <stdlib.h>
9+#include <stdarg.h>
10+#include <string.h>
11+#include <unistd.h>
12+#include <fcntl.h>
13+#include <sys/file.h>
14+#include <sys/types.h>
15+#include <sys/stat.h>
16+#include <errno.h>
17+#include <getopt.h>
18+#include <signal.h>
19+#include <time.h>
20+
21+#include "isst.h"
22+
23+static int per_package_levels_info[MAX_PACKAGE_COUNT][MAX_DIE_PER_PACKAGE];
24+static time_t per_package_levels_tm[MAX_PACKAGE_COUNT][MAX_DIE_PER_PACKAGE];
25+
26+static void init_levels(void)
27+{
28+ int i, j;
29+
30+ for (i = 0; i < MAX_PACKAGE_COUNT; ++i)
31+ for (j = 0; j < MAX_DIE_PER_PACKAGE; ++j)
32+ per_package_levels_info[i][j] = -1;
33+}
34+
35+void process_level_change(int cpu)
36+{
37+ struct isst_pkg_ctdp_level_info ctdp_level;
38+ int pkg_id = get_physical_package_id(cpu);
39+ int die_id = get_physical_die_id(cpu);
40+ struct isst_pkg_ctdp pkg_dev;
41+ time_t tm;
42+ int ret;
43+
44+ if (pkg_id >= MAX_PACKAGE_COUNT || die_id > MAX_DIE_PER_PACKAGE) {
45+ debug_printf("Invalid package/die info for cpu:%d\n", cpu);
46+ return;
47+ }
48+
49+ tm = time(NULL);
50+ if (tm - per_package_levels_tm[pkg_id][die_id] < 2 )
51+ return;
52+
53+ per_package_levels_tm[pkg_id][die_id] = tm;
54+
55+ ret = isst_get_ctdp_levels(cpu, &pkg_dev);
56+ if (ret) {
57+ debug_printf("Can't get tdp levels for cpu:%d\n", cpu);
58+ return;
59+ }
60+
61+ debug_printf("Get Config level %d pkg:%d die:%d current_level:%d \n", cpu,
62+ pkg_id, die_id, pkg_dev.current_level);
63+
64+ if (pkg_dev.locked) {
65+ debug_printf("config TDP s locked \n");
66+ return;
67+ }
68+
69+ if (per_package_levels_info[pkg_id][die_id] == pkg_dev.current_level)
70+ return;
71+
72+ debug_printf("**Config level change for cpu:%d pkg:%d die:%d from %d to %d\n",
73+ cpu, pkg_id, die_id, per_package_levels_info[pkg_id][die_id],
74+ pkg_dev.current_level);
75+
76+ per_package_levels_info[pkg_id][die_id] = pkg_dev.current_level;
77+
78+ ctdp_level.core_cpumask_size =
79+ alloc_cpu_set(&ctdp_level.core_cpumask);
80+ ret = isst_get_coremask_info(cpu, pkg_dev.current_level, &ctdp_level);
81+ if (ret) {
82+ free_cpu_set(ctdp_level.core_cpumask);
83+ debug_printf("Can't get core_mask:%d\n", cpu);
84+ return;
85+ }
86+
87+ if (ctdp_level.cpu_count) {
88+ int i, max_cpus = get_topo_max_cpus();
89+ for (i = 0; i < max_cpus; ++i) {
90+ if (pkg_id != get_physical_package_id(i) || die_id != get_physical_die_id(i))
91+ continue;
92+ if (CPU_ISSET_S(i, ctdp_level.core_cpumask_size, ctdp_level.core_cpumask)) {
93+ fprintf(stderr, "online cpu %d\n", i);
94+ set_cpu_online_offline(i, 1);
95+ } else {
96+ fprintf(stderr, "offline cpu %d\n", i);
97+ set_cpu_online_offline(i, 0);
98+ }
99+ }
100+ }
101+
102+ free_cpu_set(ctdp_level.core_cpumask);
103+}
104+
105+static void _poll_for_config_change(int cpu, void *arg1, void *arg2,
106+ void *arg3, void *arg4)
107+{
108+ process_level_change(cpu);
109+}
110+
111+static void poll_for_config_change(void)
112+{
113+ for_each_online_package_in_set(_poll_for_config_change, NULL, NULL,
114+ NULL, NULL);
115+}
116+
117+static int done = 0;
118+static int pid_file_handle;
119+
120+static void signal_handler(int sig)
121+{
122+ switch (sig) {
123+ case SIGINT:
124+ case SIGTERM:
125+ done = 1;
126+ hfi_exit();
127+ exit(0);
128+ break;
129+ default:
130+ break;
131+ }
132+}
133+
134+static void daemonize(char *rundir, char *pidfile)
135+{
136+ int pid, sid, i;
137+ char str[10];
138+ struct sigaction sig_actions;
139+ sigset_t sig_set;
140+ int ret;
141+
142+ if (getppid() == 1)
143+ return;
144+
145+ sigemptyset(&sig_set);
146+ sigaddset(&sig_set, SIGCHLD);
147+ sigaddset(&sig_set, SIGTSTP);
148+ sigaddset(&sig_set, SIGTTOU);
149+ sigaddset(&sig_set, SIGTTIN);
150+ sigprocmask(SIG_BLOCK, &sig_set, NULL);
151+
152+ sig_actions.sa_handler = signal_handler;
153+ sigemptyset(&sig_actions.sa_mask);
154+ sig_actions.sa_flags = 0;
155+
156+ sigaction(SIGHUP, &sig_actions, NULL);
157+ sigaction(SIGTERM, &sig_actions, NULL);
158+ sigaction(SIGINT, &sig_actions, NULL);
159+
160+ pid = fork();
161+ if (pid < 0) {
162+ /* Could not fork */
163+ exit(EXIT_FAILURE);
164+ }
165+ if (pid > 0)
166+ exit(EXIT_SUCCESS);
167+
168+ umask(027);
169+
170+ sid = setsid();
171+ if (sid < 0)
172+ exit(EXIT_FAILURE);
173+
174+ /* close all descriptors */
175+ for (i = getdtablesize(); i >= 0; --i)
176+ close(i);
177+
178+ i = open("/dev/null", O_RDWR);
179+ ret = dup(i);
180+ if (ret == -1)
181+ exit(EXIT_FAILURE);
182+
183+ ret = dup(i);
184+ if (ret == -1)
185+ exit(EXIT_FAILURE);
186+
187+ ret = chdir(rundir);
188+ if (ret == -1)
189+ exit(EXIT_FAILURE);
190+
191+ pid_file_handle = open(pidfile, O_RDWR | O_CREAT, 0600);
192+ if (pid_file_handle == -1) {
193+ /* Couldn't open lock file */
194+ exit(1);
195+ }
196+ /* Try to lock file */
197+#ifdef LOCKF_SUPPORT
198+ if (lockf(pid_file_handle, F_TLOCK, 0) == -1) {
199+#else
200+ if (flock(pid_file_handle, LOCK_EX|LOCK_NB) < 0) {
201+#endif
202+ /* Couldn't get lock on lock file */
203+ fprintf(stderr, "Couldn't get lock file %d\n", getpid());
204+ exit(1);
205+ }
206+ snprintf(str, sizeof(str), "%d\n", getpid());
207+ ret = write(pid_file_handle, str, strlen(str));
208+ if (ret == -1)
209+ exit(EXIT_FAILURE);
210+
211+ close(i);
212+}
213+
214+int isst_daemon(int debug_mode, int poll_interval, int no_daemon)
215+{
216+ int ret;
217+
218+ if (!no_daemon && poll_interval < 0 && !debug_mode) {
219+ fprintf(stderr, "OOB mode is enabled and will run as daemon\n");
220+ daemonize((char *) "/tmp/",
221+ (char *)"/tmp/hfi-events.pid");
222+ } else {
223+ signal(SIGINT, signal_handler);
224+ }
225+
226+ init_levels();
227+
228+ if (poll_interval < 0) {
229+ ret = hfi_main();
230+ if (ret) {
231+ fprintf(stderr, "HFI initialization failed\n");
232+ }
233+ fprintf(stderr, "Must specify poll-interval\n");
234+ return ret;
235+ }
236+
237+ debug_printf("Starting loop\n");
238+ while (!done) {
239+ sleep(poll_interval);
240+ poll_for_config_change();
241+ }
242+
243+ return 0;
244+}
tools/power/x86/intel-speed-select/isst.h+13−0
@@ -76,6 +76,9 @@ · 75 unmodified lines
7676
7777 #define DISP_FREQ_MULTIPLIER 100
7878
79+#define MAX_PACKAGE_COUNT 8
80+#define MAX_DIE_PER_PACKAGE 2
81+
7982 struct isst_clos_config {
8083 int pkg_id;
8184 int die_id;
@@ -260,4 +263,14 @@ extern int is_skx_based_platform(void); · 178 unmodified lines
260263 extern int is_spr_platform(void);
261264 extern int is_icx_platform(void);
262265 extern void isst_trl_display_information(int cpu, FILE *outf, unsigned long long trl);
266+
267+extern void set_cpu_online_offline(int cpu, int state);
268+extern void for_each_online_package_in_set(void (*callback)(int, void *, void *,
269+ void *, void *),
270+ void *arg1, void *arg2, void *arg3,
271+ void *arg4);
272+extern int isst_daemon(int debug_mode, int poll_interval, int no_daemon);
273+extern void process_level_change(int cpu);
274+extern int hfi_main(void);
275+extern void hfi_exit(void);
263276 #endif