1 .. SPDX-License-Identifier: GPL-2.0 1 .. SPDX-License-Identifier: GPL-2.0
2 .. include:: <isonum.txt> 2 .. include:: <isonum.txt>
3 3
4 ============================================== 4 ===============================================
5 ``amd-pstate`` CPU Performance Scaling Driver 5 ``amd-pstate`` CPU Performance Scaling Driver
6 ============================================== 6 ===============================================
7 7
8 :Copyright: |copy| 2021 Advanced Micro Devices 8 :Copyright: |copy| 2021 Advanced Micro Devices, Inc.
9 9
10 :Author: Huang Rui <ray.huang@amd.com> 10 :Author: Huang Rui <ray.huang@amd.com>
11 11
12 12
13 Introduction 13 Introduction
14 =================== 14 ===================
15 15
16 ``amd-pstate`` is the AMD CPU performance scal 16 ``amd-pstate`` is the AMD CPU performance scaling driver that introduces a
17 new CPU frequency control mechanism on modern 17 new CPU frequency control mechanism on modern AMD APU and CPU series in
18 Linux kernel. The new mechanism is based on Co 18 Linux kernel. The new mechanism is based on Collaborative Processor
19 Performance Control (CPPC) which provides fine 19 Performance Control (CPPC) which provides finer grain frequency management
20 than legacy ACPI hardware P-States. Current AM 20 than legacy ACPI hardware P-States. Current AMD CPU/APU platforms are using
21 the ACPI P-states driver to manage CPU frequen 21 the ACPI P-states driver to manage CPU frequency and clocks with switching
22 only in 3 P-states. CPPC replaces the ACPI P-s 22 only in 3 P-states. CPPC replaces the ACPI P-states controls and allows a
23 flexible, low-latency interface for the Linux 23 flexible, low-latency interface for the Linux kernel to directly
24 communicate the performance hints to hardware. 24 communicate the performance hints to hardware.
25 25
26 ``amd-pstate`` leverages the Linux kernel gove 26 ``amd-pstate`` leverages the Linux kernel governors such as ``schedutil``,
27 ``ondemand``, etc. to manage the performance h 27 ``ondemand``, etc. to manage the performance hints which are provided by
28 CPPC hardware functionality that internally fo 28 CPPC hardware functionality that internally follows the hardware
29 specification (for details refer to AMD64 Arch 29 specification (for details refer to AMD64 Architecture Programmer's Manual
30 Volume 2: System Programming [1]_). Currently, 30 Volume 2: System Programming [1]_). Currently, ``amd-pstate`` supports basic
31 frequency control function according to kernel 31 frequency control function according to kernel governors on some of the
32 Zen2 and Zen3 processors, and we will implemen 32 Zen2 and Zen3 processors, and we will implement more AMD specific functions
33 in future after we verify them on the hardware 33 in future after we verify them on the hardware and SBIOS.
34 34
35 35
36 AMD CPPC Overview 36 AMD CPPC Overview
37 ======================= 37 =======================
38 38
39 Collaborative Processor Performance Control (C 39 Collaborative Processor Performance Control (CPPC) interface enumerates a
40 continuous, abstract, and unit-less performanc 40 continuous, abstract, and unit-less performance value in a scale that is
41 not tied to a specific performance state / fre 41 not tied to a specific performance state / frequency. This is an ACPI
42 standard [2]_ which software can specify appli 42 standard [2]_ which software can specify application performance goals and
43 hints as a relative target to the infrastructu 43 hints as a relative target to the infrastructure limits. AMD processors
44 provide the low latency register model (MSR) i 44 provide the low latency register model (MSR) instead of an AML code
45 interpreter for performance adjustments. ``amd 45 interpreter for performance adjustments. ``amd-pstate`` will initialize a
46 ``struct cpufreq_driver`` instance, ``amd_psta 46 ``struct cpufreq_driver`` instance, ``amd_pstate_driver``, with the callbacks
47 to manage each performance update behavior. :: 47 to manage each performance update behavior. ::
48 48
49 Highest Perf ------>+-----------------------+ 49 Highest Perf ------>+-----------------------+ +-----------------------+
50 | | 50 | | | |
51 | | 51 | | | |
52 | | 52 | | Max Perf ---->| |
53 | | 53 | | | |
54 | | 54 | | | |
55 Nominal Perf ------>+-----------------------+ 55 Nominal Perf ------>+-----------------------+ +-----------------------+
56 | | 56 | | | |
57 | | 57 | | | |
58 | | 58 | | | |
59 | | 59 | | | |
60 | | 60 | | | |
61 | | 61 | | | |
62 | | 62 | | Desired Perf ---->| |
63 | | 63 | | | |
64 | | 64 | | | |
65 | | 65 | | | |
66 | | 66 | | | |
67 | | 67 | | | |
68 | | 68 | | | |
69 | | 69 | | | |
70 | | 70 | | | |
71 | | 71 | | | |
72 Lowest non- | | 72 Lowest non- | | | |
73 linear perf ------>+-----------------------+ 73 linear perf ------>+-----------------------+ +-----------------------+
74 | | 74 | | | |
75 | | 75 | | Lowest perf ---->| |
76 | | 76 | | | |
77 Lowest perf ------>+-----------------------+ 77 Lowest perf ------>+-----------------------+ +-----------------------+
78 | | 78 | | | |
79 | | 79 | | | |
80 | | 80 | | | |
81 0 ------>+-----------------------+ 81 0 ------>+-----------------------+ +-----------------------+
82 82
83 AMD P-Sta 83 AMD P-States Performance Scale
84 84
85 85
86 .. _perf_cap: 86 .. _perf_cap:
87 87
88 AMD CPPC Performance Capability 88 AMD CPPC Performance Capability
89 -------------------------------- 89 --------------------------------
90 90
91 Highest Performance (RO) 91 Highest Performance (RO)
92 ......................... 92 .........................
93 93
94 This is the absolute maximum performance an in 94 This is the absolute maximum performance an individual processor may reach,
95 assuming ideal conditions. This performance le 95 assuming ideal conditions. This performance level may not be sustainable
96 for long durations and may only be achievable 96 for long durations and may only be achievable if other platform components
97 are in a specific state; for example, it may r 97 are in a specific state; for example, it may require other processors to be in
98 an idle state. This would be equivalent to the 98 an idle state. This would be equivalent to the highest frequencies
99 supported by the processor. 99 supported by the processor.
100 100
101 Nominal (Guaranteed) Performance (RO) 101 Nominal (Guaranteed) Performance (RO)
102 ...................................... 102 ......................................
103 103
104 This is the maximum sustained performance leve 104 This is the maximum sustained performance level of the processor, assuming
105 ideal operating conditions. In the absence of 105 ideal operating conditions. In the absence of an external constraint (power,
106 thermal, etc.), this is the performance level 106 thermal, etc.), this is the performance level the processor is expected to
107 be able to maintain continuously. All cores/pr 107 be able to maintain continuously. All cores/processors are expected to be
108 able to sustain their nominal performance stat 108 able to sustain their nominal performance state simultaneously.
109 109
110 Lowest non-linear Performance (RO) 110 Lowest non-linear Performance (RO)
111 ................................... 111 ...................................
112 112
113 This is the lowest performance level at which 113 This is the lowest performance level at which nonlinear power savings are
114 achieved, for example, due to the combined eff 114 achieved, for example, due to the combined effects of voltage and frequency
115 scaling. Above this threshold, lower performan 115 scaling. Above this threshold, lower performance levels should be generally
116 more energy efficient than higher performance 116 more energy efficient than higher performance levels. This register
117 effectively conveys the most efficient perform 117 effectively conveys the most efficient performance level to ``amd-pstate``.
118 118
119 Lowest Performance (RO) 119 Lowest Performance (RO)
120 ........................ 120 ........................
121 121
122 This is the absolute lowest performance level 122 This is the absolute lowest performance level of the processor. Selecting a
123 performance level lower than the lowest nonlin 123 performance level lower than the lowest nonlinear performance level may
124 cause an efficiency penalty but should reduce 124 cause an efficiency penalty but should reduce the instantaneous power
125 consumption of the processor. 125 consumption of the processor.
126 126
127 AMD CPPC Performance Control 127 AMD CPPC Performance Control
128 ------------------------------ 128 ------------------------------
129 129
130 ``amd-pstate`` passes performance goals throug 130 ``amd-pstate`` passes performance goals through these registers. The
131 register drives the behavior of the desired pe 131 register drives the behavior of the desired performance target.
132 132
133 Minimum requested performance (RW) 133 Minimum requested performance (RW)
134 ................................... 134 ...................................
135 135
136 ``amd-pstate`` specifies the minimum allowed p 136 ``amd-pstate`` specifies the minimum allowed performance level.
137 137
138 Maximum requested performance (RW) 138 Maximum requested performance (RW)
139 ................................... 139 ...................................
140 140
141 ``amd-pstate`` specifies a limit the maximum p 141 ``amd-pstate`` specifies a limit the maximum performance that is expected
142 to be supplied by the hardware. 142 to be supplied by the hardware.
143 143
144 Desired performance target (RW) 144 Desired performance target (RW)
145 ................................... 145 ...................................
146 146
147 ``amd-pstate`` specifies a desired target in t 147 ``amd-pstate`` specifies a desired target in the CPPC performance scale as
148 a relative number. This can be expressed as pe 148 a relative number. This can be expressed as percentage of nominal
149 performance (infrastructure max). Below the no 149 performance (infrastructure max). Below the nominal sustained performance
150 level, desired performance expresses the avera 150 level, desired performance expresses the average performance level of the
151 processor subject to hardware. Above the nomin 151 processor subject to hardware. Above the nominal performance level,
152 the processor must provide at least nominal pe 152 the processor must provide at least nominal performance requested and go higher
153 if current operating conditions allow. 153 if current operating conditions allow.
154 154
155 Energy Performance Preference (EPP) (RW) 155 Energy Performance Preference (EPP) (RW)
156 ......................................... 156 .........................................
157 157
158 This attribute provides a hint to the hardware 158 This attribute provides a hint to the hardware if software wants to bias
159 toward performance (0x0) or energy efficiency 159 toward performance (0x0) or energy efficiency (0xff).
160 160
161 161
162 Key Governors Support 162 Key Governors Support
163 ======================= 163 =======================
164 164
165 ``amd-pstate`` can be used with all the (gener 165 ``amd-pstate`` can be used with all the (generic) scaling governors listed
166 by the ``scaling_available_governors`` policy 166 by the ``scaling_available_governors`` policy attribute in ``sysfs``. Then,
167 it is responsible for the configuration of pol 167 it is responsible for the configuration of policy objects corresponding to
168 CPUs and provides the ``CPUFreq`` core (and th 168 CPUs and provides the ``CPUFreq`` core (and the scaling governors attached
169 to the policy objects) with accurate informati 169 to the policy objects) with accurate information on the maximum and minimum
170 operating frequencies supported by the hardwar 170 operating frequencies supported by the hardware. Users can check the
171 ``scaling_cur_freq`` information comes from th 171 ``scaling_cur_freq`` information comes from the ``CPUFreq`` core.
172 172
173 ``amd-pstate`` mainly supports ``schedutil`` a 173 ``amd-pstate`` mainly supports ``schedutil`` and ``ondemand`` for dynamic
174 frequency control. It is to fine tune the proc 174 frequency control. It is to fine tune the processor configuration on
175 ``amd-pstate`` to the ``schedutil`` with CPU C 175 ``amd-pstate`` to the ``schedutil`` with CPU CFS scheduler. ``amd-pstate``
176 registers the adjust_perf callback to implemen 176 registers the adjust_perf callback to implement performance update behavior
177 similar to CPPC. It is initialized by ``sugov_ 177 similar to CPPC. It is initialized by ``sugov_start`` and then populates the
178 CPU's update_util_data pointer to assign ``sug 178 CPU's update_util_data pointer to assign ``sugov_update_single_perf`` as the
179 utilization update callback function in the CP 179 utilization update callback function in the CPU scheduler. The CPU scheduler
180 will call ``cpufreq_update_util`` and assigns 180 will call ``cpufreq_update_util`` and assigns the target performance according
181 to the ``struct sugov_cpu`` that the utilizati 181 to the ``struct sugov_cpu`` that the utilization update belongs to.
182 Then, ``amd-pstate`` updates the desired perfo 182 Then, ``amd-pstate`` updates the desired performance according to the CPU
183 scheduler assigned. 183 scheduler assigned.
184 184
185 .. _processor_support: <<
186 185
187 Processor Support 186 Processor Support
188 ======================= 187 =======================
189 188
190 The ``amd-pstate`` initialization will fail if 189 The ``amd-pstate`` initialization will fail if the ``_CPC`` entry in the ACPI
191 SBIOS does not exist in the detected processor 190 SBIOS does not exist in the detected processor. It uses ``acpi_cpc_valid``
192 to check the existence of ``_CPC``. All Zen ba 191 to check the existence of ``_CPC``. All Zen based processors support the legacy
193 ACPI hardware P-States function, so when ``amd 192 ACPI hardware P-States function, so when ``amd-pstate`` fails initialization,
194 the kernel will fall back to initialize the `` 193 the kernel will fall back to initialize the ``acpi-cpufreq`` driver.
195 194
196 There are two types of hardware implementation 195 There are two types of hardware implementations for ``amd-pstate``: one is
197 `Full MSR Support <perf_cap_>`_ and another is 196 `Full MSR Support <perf_cap_>`_ and another is `Shared Memory Support
198 <perf_cap_>`_. It can use the :c:macro:`X86_FE 197 <perf_cap_>`_. It can use the :c:macro:`X86_FEATURE_CPPC` feature flag to
199 indicate the different types. (For details, re 198 indicate the different types. (For details, refer to the Processor Programming
200 Reference (PPR) for AMD Family 19h Model 51h, 199 Reference (PPR) for AMD Family 19h Model 51h, Revision A1 Processors [3]_.)
201 ``amd-pstate`` is to register different ``stat 200 ``amd-pstate`` is to register different ``static_call`` instances for different
202 hardware implementations. 201 hardware implementations.
203 202
204 Currently, some of the Zen2 and Zen3 processor 203 Currently, some of the Zen2 and Zen3 processors support ``amd-pstate``. In the
205 future, it will be supported on more and more 204 future, it will be supported on more and more AMD processors.
206 205
207 Full MSR Support 206 Full MSR Support
208 ----------------- 207 -----------------
209 208
210 Some new Zen3 processors such as Cezanne provi 209 Some new Zen3 processors such as Cezanne provide the MSR registers directly
211 while the :c:macro:`X86_FEATURE_CPPC` CPU feat 210 while the :c:macro:`X86_FEATURE_CPPC` CPU feature flag is set.
212 ``amd-pstate`` can handle the MSR register to 211 ``amd-pstate`` can handle the MSR register to implement the fast switch
213 function in ``CPUFreq`` that can reduce the la 212 function in ``CPUFreq`` that can reduce the latency of frequency control in
214 interrupt context. The functions with a ``psta 213 interrupt context. The functions with a ``pstate_xxx`` prefix represent the
215 operations on MSR registers. 214 operations on MSR registers.
216 215
217 Shared Memory Support 216 Shared Memory Support
218 ---------------------- 217 ----------------------
219 218
220 If the :c:macro:`X86_FEATURE_CPPC` CPU feature 219 If the :c:macro:`X86_FEATURE_CPPC` CPU feature flag is not set, the
221 processor supports the shared memory solution. 220 processor supports the shared memory solution. In this case, ``amd-pstate``
222 uses the ``cppc_acpi`` helper methods to imple 221 uses the ``cppc_acpi`` helper methods to implement the callback functions
223 that are defined on ``static_call``. The funct 222 that are defined on ``static_call``. The functions with the ``cppc_xxx`` prefix
224 represent the operations of ACPI CPPC helpers 223 represent the operations of ACPI CPPC helpers for the shared memory solution.
225 224
226 225
227 AMD P-States and ACPI hardware P-States always 226 AMD P-States and ACPI hardware P-States always can be supported in one
228 processor. But AMD P-States has the higher pri 227 processor. But AMD P-States has the higher priority and if it is enabled
229 with :c:macro:`MSR_AMD_CPPC_ENABLE` or ``cppc_ 228 with :c:macro:`MSR_AMD_CPPC_ENABLE` or ``cppc_set_enable``, it will respond
230 to the request from AMD P-States. 229 to the request from AMD P-States.
231 230
232 231
233 User Space Interface in ``sysfs`` - Per-policy !! 232 User Space Interface in ``sysfs``
234 ============================================== !! 233 ==================================
235 234
236 ``amd-pstate`` exposes several global attribut 235 ``amd-pstate`` exposes several global attributes (files) in ``sysfs`` to
237 control its functionality at the system level. 236 control its functionality at the system level. They are located in the
238 ``/sys/devices/system/cpu/cpufreq/policyX/`` d 237 ``/sys/devices/system/cpu/cpufreq/policyX/`` directory and affect all CPUs. ::
239 238
240 root@hr-test1:/home/ray# ls /sys/devices/syst 239 root@hr-test1:/home/ray# ls /sys/devices/system/cpu/cpufreq/policy0/*amd*
241 /sys/devices/system/cpu/cpufreq/policy0/amd_p 240 /sys/devices/system/cpu/cpufreq/policy0/amd_pstate_highest_perf
242 /sys/devices/system/cpu/cpufreq/policy0/amd_p 241 /sys/devices/system/cpu/cpufreq/policy0/amd_pstate_lowest_nonlinear_freq
243 /sys/devices/system/cpu/cpufreq/policy0/amd_p 242 /sys/devices/system/cpu/cpufreq/policy0/amd_pstate_max_freq
244 243
245 244
246 ``amd_pstate_highest_perf / amd_pstate_max_fre 245 ``amd_pstate_highest_perf / amd_pstate_max_freq``
247 246
248 Maximum CPPC performance and CPU frequency tha 247 Maximum CPPC performance and CPU frequency that the driver is allowed to
249 set, in percent of the maximum supported CPPC 248 set, in percent of the maximum supported CPPC performance level (the highest
250 performance supported in `AMD CPPC Performance 249 performance supported in `AMD CPPC Performance Capability <perf_cap_>`_).
251 In some ASICs, the highest CPPC performance is 250 In some ASICs, the highest CPPC performance is not the one in the ``_CPC``
252 table, so we need to expose it to sysfs. If bo 251 table, so we need to expose it to sysfs. If boost is not active, but
253 still supported, this maximum frequency will b 252 still supported, this maximum frequency will be larger than the one in
254 ``cpuinfo``. On systems that support preferred !! 253 ``cpuinfo``.
255 different values for some cores than others an <<
256 advertised by the platform at bootup. <<
257 This attribute is read-only. 254 This attribute is read-only.
258 255
259 ``amd_pstate_lowest_nonlinear_freq`` 256 ``amd_pstate_lowest_nonlinear_freq``
260 257
261 The lowest non-linear CPPC CPU frequency that 258 The lowest non-linear CPPC CPU frequency that the driver is allowed to set,
262 in percent of the maximum supported CPPC perfo 259 in percent of the maximum supported CPPC performance level. (Please see the
263 lowest non-linear performance in `AMD CPPC Per 260 lowest non-linear performance in `AMD CPPC Performance Capability
264 <perf_cap_>`_.) 261 <perf_cap_>`_.)
265 This attribute is read-only. 262 This attribute is read-only.
266 263
267 ``amd_pstate_hw_prefcore`` <<
268 <<
269 Whether the platform supports the preferred co <<
270 enabled. This attribute is read-only. <<
271 <<
272 ``amd_pstate_prefcore_ranking`` <<
273 <<
274 The performance ranking of the core. This numb <<
275 larger numbers are preferred at the time of re <<
276 runtime based on platform conditions. This att <<
277 <<
278 ``energy_performance_available_preferences`` <<
279 <<
280 A list of all the supported EPP preferences th <<
281 ``energy_performance_preference`` on this syst <<
282 These profiles represent different hints that <<
283 to the low-level firmware about the user's des <<
284 tradeoff. ``default`` represents the epp valu <<
285 firmware. This attribute is read-only. <<
286 <<
287 ``energy_performance_preference`` <<
288 <<
289 The current energy performance preference can <<
290 and user can change current preference accordi <<
291 Please get all support profiles list from <<
292 ``energy_performance_available_preferences`` a <<
293 integer values defined between 0 to 255 when E <<
294 firmware, if EPP feature is disabled, driver w <<
295 This attribute is read-write. <<
296 <<
297 ``boost`` <<
298 The `boost` sysfs attribute provides control o <<
299 performance boost, allowing users to manage th <<
300 of the CPU. This attribute can be used to enab <<
301 on individual CPUs. <<
302 <<
303 When the boost feature is enabled, the CPU can <<
304 beyond the base frequency, providing enhanced <<
305 On the other hand, disabling the boost feature <<
306 base frequency, which may be desirable in cert <<
307 efficiency or manage temperature. <<
308 <<
309 To manipulate the `boost` attribute, users can <<
310 boost or `1` to enable it, for the respective <<
311 `/sys/devices/system/cpu/cpuX/cpufreq/boost`, <<
312 <<
313 Other performance and frequency values can be 264 Other performance and frequency values can be read back from
314 ``/sys/devices/system/cpu/cpuX/acpi_cppc/``, s 265 ``/sys/devices/system/cpu/cpuX/acpi_cppc/``, see :ref:`cppc_sysfs`.
315 266
316 267
317 ``amd-pstate`` vs ``acpi-cpufreq`` 268 ``amd-pstate`` vs ``acpi-cpufreq``
318 ====================================== 269 ======================================
319 270
320 On the majority of AMD platforms supported by 271 On the majority of AMD platforms supported by ``acpi-cpufreq``, the ACPI tables
321 provided by the platform firmware are used for 272 provided by the platform firmware are used for CPU performance scaling, but
322 only provide 3 P-states on AMD processors. 273 only provide 3 P-states on AMD processors.
323 However, on modern AMD APU and CPU series, har 274 However, on modern AMD APU and CPU series, hardware provides the Collaborative
324 Processor Performance Control according to the 275 Processor Performance Control according to the ACPI protocol and customizes this
325 for AMD platforms. That is, fine-grained and c 276 for AMD platforms. That is, fine-grained and continuous frequency ranges
326 instead of the legacy hardware P-states. ``amd 277 instead of the legacy hardware P-states. ``amd-pstate`` is the kernel
327 module which supports the new AMD P-States mec 278 module which supports the new AMD P-States mechanism on most of the future AMD
328 platforms. The AMD P-States mechanism is the m 279 platforms. The AMD P-States mechanism is the more performance and energy
329 efficiency frequency management method on AMD 280 efficiency frequency management method on AMD processors.
330 281
>> 282 Kernel Module Options for ``amd-pstate``
>> 283 =========================================
331 284
332 ``amd-pstate`` Driver Operation Modes !! 285 ``shared_mem``
333 ====================================== !! 286 Use a module param (shared_mem) to enable related processors manually with
334 !! 287 **amd_pstate.shared_mem=1**.
335 ``amd_pstate`` CPPC has 3 operation modes: aut !! 288 Due to the performance issue on the processors with `Shared Memory Support
336 non-autonomous (passive) mode and guided auton !! 289 <perf_cap_>`_, we disable it presently and will re-enable this by default
337 Active/passive/guided mode can be chosen by di !! 290 once we address performance issue with this solution.
338 <<
339 - In autonomous mode, platform ignores the des <<
340 and takes into account only the values set t <<
341 performance preference registers. <<
342 - In non-autonomous mode, platform gets desire <<
343 from OS directly through Desired Performance <<
344 - In guided-autonomous mode, platform sets ope <<
345 autonomously according to the current worklo <<
346 OS through min and max performance registers <<
347 <<
348 Active Mode <<
349 ------------ <<
350 <<
351 ``amd_pstate=active`` <<
352 <<
353 This is the low-level firmware control mode wh <<
354 driver with ``amd_pstate=active`` passed to th <<
355 In this mode, ``amd_pstate_epp`` driver provid <<
356 wants to bias toward performance (0x0) or ener <<
357 then CPPC power algorithm will calculate the r <<
358 cores frequency according to the power supply <<
359 hardware conditions. <<
360 <<
361 Passive Mode <<
362 ------------ <<
363 <<
364 ``amd_pstate=passive`` <<
365 <<
366 It will be enabled if the ``amd_pstate=passive <<
367 In this mode, ``amd_pstate`` driver software s <<
368 performance scale as a relative number. This c <<
369 performance (infrastructure max). Below the no <<
370 desired performance expresses the average perf <<
371 to the Performance Reduction Tolerance registe <<
372 processor must provide at least nominal perfor <<
373 operating conditions allow. <<
374 <<
375 Guided Mode <<
376 ----------- <<
377 <<
378 ``amd_pstate=guided`` <<
379 <<
380 If ``amd_pstate=guided`` is passed to kernel c <<
381 is activated. In this mode, driver requests m <<
382 level and the platform autonomously selects a <<
383 and appropriate to the current workload. <<
384 <<
385 ``amd-pstate`` Preferred Core <<
386 ================================= <<
387 <<
388 The core frequency is subjected to the process <<
389 Not all cores are able to reach the maximum fr <<
390 infrastructure limits. Consequently, AMD has r <<
391 maximum frequency of a part. This means that a <<
392 maximum frequency. To find the best process sc <<
393 scenario, OS needs to know the core ordering i <<
394 highest performance capability register of the <<
395 <<
396 ``amd-pstate`` preferred core enables the sche <<
397 cores that can achieve a higher frequency with <<
398 core rankings can dynamically change based on <<
399 thermals and ageing. <<
400 <<
401 The priority metric will be initialized by the <<
402 driver will also determine whether or not ``am <<
403 supported by the platform. <<
404 <<
405 ``amd-pstate`` driver will provide an initial <<
406 The platform uses the CPPC interfaces to commu <<
407 operating system and scheduler to make sure th <<
408 with highest performance firstly for schedulin <<
409 driver receives a message with the highest per <<
410 update the core ranking and set the cpu's prio <<
411 <<
412 ``amd-pstate`` Preferred Core Switch <<
413 ===================================== <<
414 Kernel Parameters <<
415 ----------------- <<
416 <<
417 ``amd-pstate`` peferred core`` has two states: <<
418 Enable/disable states can be chosen by differe <<
419 Default enable ``amd-pstate`` preferred core. <<
420 <<
421 ``amd_prefcore=disable`` <<
422 <<
423 For systems that support ``amd-pstate`` prefer <<
424 always be advertised by the platform. But OS c <<
425 kernel parameter ``amd_prefcore=disable``. <<
426 <<
427 User Space Interface in ``sysfs`` - General <<
428 =========================================== <<
429 291
430 Global Attributes !! 292 To check whether the current processor is using `Full MSR Support <perf_cap_>`_
431 ----------------- !! 293 or `Shared Memory Support <perf_cap_>`_ : ::
432 <<
433 ``amd-pstate`` exposes several global attribut <<
434 control its functionality at the system level. <<
435 ``/sys/devices/system/cpu/amd_pstate/`` direct <<
436 294
437 ``status`` !! 295 ray@hr-test1:~$ lscpu | grep cppc
438 Operation mode of the driver: "active" !! 296 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm
439 297
440 "active" !! 298 If the CPU flags have ``cppc``, then this processor supports `Full MSR Support
441 The driver is functional and i !! 299 <perf_cap_>`_. Otherwise, it supports `Shared Memory Support <perf_cap_>`_.
442 300
443 "passive" <<
444 The driver is functional and i <<
445 <<
446 "guided" <<
447 The driver is functional and i <<
448 <<
449 "disable" <<
450 The driver is unregistered and <<
451 <<
452 This attribute can be written to in or <<
453 operation mode or to unregister it. T <<
454 one of the possible values of it and, <<
455 these values to the sysfs file will ca <<
456 to the operation mode represented by t <<
457 unregistered in the "disable" case. <<
458 <<
459 ``prefcore`` <<
460 Preferred core state of the driver: "e <<
461 <<
462 "enabled" <<
463 Enable the ``amd-pstate`` pref <<
464 <<
465 "disabled" <<
466 Disable the ``amd-pstate`` pre <<
467 <<
468 <<
469 This attribute is read-only to check t <<
470 by the kernel parameter. <<
471 301
472 ``cpupower`` tool support for ``amd-pstate`` 302 ``cpupower`` tool support for ``amd-pstate``
473 ============================================== 303 ===============================================
474 304
475 ``amd-pstate`` is supported by the ``cpupower` 305 ``amd-pstate`` is supported by the ``cpupower`` tool, which can be used to dump
476 frequency information. Development is in progr 306 frequency information. Development is in progress to support more and more
477 operations for the new ``amd-pstate`` module w 307 operations for the new ``amd-pstate`` module with this tool. ::
478 308
479 root@hr-test1:/home/ray# cpupower frequency-i 309 root@hr-test1:/home/ray# cpupower frequency-info
480 analyzing CPU 0: 310 analyzing CPU 0:
481 driver: amd-pstate 311 driver: amd-pstate
482 CPUs which run at the same hardware frequen 312 CPUs which run at the same hardware frequency: 0
483 CPUs which need to have their frequency coo 313 CPUs which need to have their frequency coordinated by software: 0
484 maximum transition latency: 131 us 314 maximum transition latency: 131 us
485 hardware limits: 400 MHz - 4.68 GHz 315 hardware limits: 400 MHz - 4.68 GHz
486 available cpufreq governors: ondemand conse 316 available cpufreq governors: ondemand conservative powersave userspace performance schedutil
487 current policy: frequency should be within 317 current policy: frequency should be within 400 MHz and 4.68 GHz.
488 The governor "schedutil" ma 318 The governor "schedutil" may decide which speed to use
489 within this range. 319 within this range.
490 current CPU frequency: Unable to call hardw 320 current CPU frequency: Unable to call hardware
491 current CPU frequency: 4.02 GHz (asserted b 321 current CPU frequency: 4.02 GHz (asserted by call to kernel)
492 boost state support: 322 boost state support:
493 Supported: yes 323 Supported: yes
494 Active: yes 324 Active: yes
495 AMD PSTATE Highest Performance: 166. Maxi 325 AMD PSTATE Highest Performance: 166. Maximum Frequency: 4.68 GHz.
496 AMD PSTATE Nominal Performance: 117. Nomi 326 AMD PSTATE Nominal Performance: 117. Nominal Frequency: 3.30 GHz.
497 AMD PSTATE Lowest Non-linear Performance: 327 AMD PSTATE Lowest Non-linear Performance: 39. Lowest Non-linear Frequency: 1.10 GHz.
498 AMD PSTATE Lowest Performance: 15. Lowest 328 AMD PSTATE Lowest Performance: 15. Lowest Frequency: 400 MHz.
499 329
500 330
501 Diagnostics and Tuning 331 Diagnostics and Tuning
502 ======================= 332 =======================
503 333
504 Trace Events 334 Trace Events
505 -------------- 335 --------------
506 336
507 There are two static trace events that can be 337 There are two static trace events that can be used for ``amd-pstate``
508 diagnostics. One of them is the ``cpu_frequenc 338 diagnostics. One of them is the ``cpu_frequency`` trace event generally used
509 by ``CPUFreq``, and the other one is the ``amd 339 by ``CPUFreq``, and the other one is the ``amd_pstate_perf`` trace event
510 specific to ``amd-pstate``. The following seq 340 specific to ``amd-pstate``. The following sequence of shell commands can
511 be used to enable them and see their output (i 341 be used to enable them and see their output (if the kernel is
512 configured to support event tracing). :: 342 configured to support event tracing). ::
513 343
514 root@hr-test1:/home/ray# cd /sys/kernel/traci 344 root@hr-test1:/home/ray# cd /sys/kernel/tracing/
515 root@hr-test1:/sys/kernel/tracing# echo 1 > e 345 root@hr-test1:/sys/kernel/tracing# echo 1 > events/amd_cpu/enable
516 root@hr-test1:/sys/kernel/tracing# cat trace 346 root@hr-test1:/sys/kernel/tracing# cat trace
517 # tracer: nop 347 # tracer: nop
518 # 348 #
519 # entries-in-buffer/entries-written: 47827/42 349 # entries-in-buffer/entries-written: 47827/42233061 #P:2
520 # 350 #
521 # _-----=> irq 351 # _-----=> irqs-off
522 # / _----=> nee 352 # / _----=> need-resched
523 # | / _---=> har 353 # | / _---=> hardirq/softirq
524 # || / _--=> pre 354 # || / _--=> preempt-depth
525 # ||| / dela 355 # ||| / delay
526 # TASK-PID CPU# |||| TIMESTA 356 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
527 # | | | |||| | 357 # | | | |||| | |
528 <idle>-0 [015] dN... 4995.979 358 <idle>-0 [015] dN... 4995.979886: amd_pstate_perf: amd_min_perf=85 amd_des_perf=85 amd_max_perf=166 cpu_id=15 changed=false fast_switch=true
529 <idle>-0 [007] d.h.. 4995.979 359 <idle>-0 [007] d.h.. 4995.979893: amd_pstate_perf: amd_min_perf=85 amd_des_perf=85 amd_max_perf=166 cpu_id=7 changed=false fast_switch=true
530 cat-2161 [000] d.... 4995.980 360 cat-2161 [000] d.... 4995.980841: amd_pstate_perf: amd_min_perf=85 amd_des_perf=85 amd_max_perf=166 cpu_id=0 changed=false fast_switch=true
531 sshd-2125 [004] d.s.. 4995.980 361 sshd-2125 [004] d.s.. 4995.980968: amd_pstate_perf: amd_min_perf=85 amd_des_perf=85 amd_max_perf=166 cpu_id=4 changed=false fast_switch=true
532 <idle>-0 [007] d.s.. 4995.980 362 <idle>-0 [007] d.s.. 4995.980968: amd_pstate_perf: amd_min_perf=85 amd_des_perf=85 amd_max_perf=166 cpu_id=7 changed=false fast_switch=true
533 <idle>-0 [003] d.s.. 4995.980 363 <idle>-0 [003] d.s.. 4995.980971: amd_pstate_perf: amd_min_perf=85 amd_des_perf=85 amd_max_perf=166 cpu_id=3 changed=false fast_switch=true
534 <idle>-0 [011] d.s.. 4995.980 364 <idle>-0 [011] d.s.. 4995.980996: amd_pstate_perf: amd_min_perf=85 amd_des_perf=85 amd_max_perf=166 cpu_id=11 changed=false fast_switch=true
535 365
536 The ``cpu_frequency`` trace event will be trig 366 The ``cpu_frequency`` trace event will be triggered either by the ``schedutil`` scaling
537 governor (for the policies it is attached to), 367 governor (for the policies it is attached to), or by the ``CPUFreq`` core (for the
538 policies with other scaling governors). 368 policies with other scaling governors).
539 369
540 370
541 Tracer Tool 371 Tracer Tool
542 ------------- 372 -------------
543 373
544 ``amd_pstate_tracer.py`` can record and parse 374 ``amd_pstate_tracer.py`` can record and parse ``amd-pstate`` trace log, then
545 generate performance plots. This utility can b 375 generate performance plots. This utility can be used to debug and tune the
546 performance of ``amd-pstate`` driver. The trac 376 performance of ``amd-pstate`` driver. The tracer tool needs to import intel
547 pstate tracer. 377 pstate tracer.
548 378
549 Tracer tool located in ``linux/tools/power/x86 379 Tracer tool located in ``linux/tools/power/x86/amd_pstate_tracer``. It can be
550 used in two ways. If trace file is available, 380 used in two ways. If trace file is available, then directly parse the file
551 with command :: 381 with command ::
552 382
553 ./amd_pstate_trace.py [-c cpus] -t <trace_fil 383 ./amd_pstate_trace.py [-c cpus] -t <trace_file> -n <test_name>
554 384
555 Or generate trace file with root privilege, th 385 Or generate trace file with root privilege, then parse and plot with command ::
556 386
557 sudo ./amd_pstate_trace.py [-c cpus] -n <test 387 sudo ./amd_pstate_trace.py [-c cpus] -n <test_name> -i <interval> [-m kbytes]
558 388
559 The test result can be found in ``results/test 389 The test result can be found in ``results/test_name``. Following is the example
560 about part of the output. :: 390 about part of the output. ::
561 391
562 common_cpu common_secs common_usecs min_pe 392 common_cpu common_secs common_usecs min_perf des_perf max_perf freq mperf apef tsc load duration_ms sample_num elapsed_time common_comm
563 CPU_005 712 116384 39 393 CPU_005 712 116384 39 49 166 0.7565 9645075 2214891 38431470 25.1 11.646 469 2.496 kworker/5:0-40
564 CPU_006 712 116408 39 394 CPU_006 712 116408 39 49 166 0.6769 8950227 1839034 37192089 24.06 11.272 470 2.496 kworker/6:0-1264
565 395
566 Unit Tests for amd-pstate <<
567 ------------------------- <<
568 <<
569 ``amd-pstate-ut`` is a test module for testing <<
570 <<
571 * It can help all users to verify their proce <<
572 <<
573 * Kernel can have a basic function test to av <<
574 <<
575 * We can introduce more functional or perform <<
576 <<
577 1. Test case descriptions <<
578 <<
579 1). Basic tests <<
580 <<
581 Test prerequisite and basic functions <<
582 <<
583 +---------+--------------------------- <<
584 | Index | Functions <<
585 +=========+=========================== <<
586 | 1 | amd_pstate_ut_acpi_cpc_val <<
587 | | <<
588 | | <<
589 +---------+--------------------------- <<
590 | 2 | amd_pstate_ut_check_enable <<
591 | | <<
592 | | <<
593 | | <<
594 | | <<
595 | | <<
596 +---------+--------------------------- <<
597 | 3 | amd_pstate_ut_check_perf <<
598 | | <<
599 +---------+--------------------------- <<
600 | 4 | amd_pstate_ut_check_freq <<
601 | | <<
602 | | <<
603 | | <<
604 | | <<
605 +---------+--------------------------- <<
606 <<
607 2). Tbench test <<
608 <<
609 Test and monitor the cpu changes when <<
610 These changes include desire performan <<
611 The specified governor is ondemand or <<
612 Tbench can also be tested on the ``acp <<
613 <<
614 3). Gitsource test <<
615 <<
616 Test and monitor the cpu changes when <<
617 These changes include desire performan <<
618 The specified governor is ondemand or <<
619 Gitsource can also be tested on the `` <<
620 <<
621 #. How to execute the tests <<
622 <<
623 We use test module in the kselftest framewo <<
624 We create ``amd-pstate-ut`` module and tie <<
625 details refer to Linux Kernel Selftests [4] <<
626 <<
627 1). Build <<
628 <<
629 + open the :c:macro:`CONFIG_X86_AMD_PS <<
630 + set the :c:macro:`CONFIG_X86_AMD_PST <<
631 + make project <<
632 + make selftest :: <<
633 <<
634 $ cd linux <<
635 $ make -C tools/testing/selftests <<
636 <<
637 + make perf :: <<
638 <<
639 $ cd tools/perf/ <<
640 $ make <<
641 <<
642 <<
643 2). Installation & Steps :: <<
644 <<
645 $ make -C tools/testing/selftests inst <<
646 $ cp tools/perf/perf /usr/bin/perf <<
647 $ sudo ./kselftest/run_kselftest.sh -c <<
648 <<
649 3). Specified test case :: <<
650 <<
651 $ cd ~/kselftest/amd-pstate <<
652 $ sudo ./run.sh -t basic <<
653 $ sudo ./run.sh -t tbench <<
654 $ sudo ./run.sh -t tbench -m acpi-cpuf <<
655 $ sudo ./run.sh -t gitsource <<
656 $ sudo ./run.sh -t gitsource -m acpi-c <<
657 $ ./run.sh --help <<
658 ./run.sh: illegal option -- - <<
659 Usage: ./run.sh [OPTION...] <<
660 [-h <help>] <<
661 [-o <output-file-for-dump>] <<
662 [-c <all: All testing, <<
663 basic: Basic testing, <<
664 tbench: Tbench testing, <<
665 gitsource: Gitsource test <<
666 [-t <tbench time limit>] <<
667 [-p <tbench process number>] <<
668 [-l <loop times for tbench>] <<
669 [-i <amd tracer interval>] <<
670 [-m <comparative test: acpi-cp <<
671 <<
672 <<
673 4). Results <<
674 <<
675 + basic <<
676 <<
677 When you finish test, you will get th <<
678 <<
679 $ dmesg | grep "amd_pstate_ut" | tee <<
680 [12977.570663] amd_pstate_ut: 1 a <<
681 [12977.570673] amd_pstate_ut: 2 a <<
682 [12977.571207] amd_pstate_ut: 3 a <<
683 [12977.571212] amd_pstate_ut: 4 a <<
684 <<
685 + tbench <<
686 <<
687 When you finish test, you will get se <<
688 The selftest.tbench.csv file contains <<
689 The png images shows the performance, <<
690 Open selftest.tbench.csv : <<
691 <<
692 +------------------------------------ <<
693 + Governor <<
694 +------------------------------------ <<
695 + Unit <<
696 +==================================== <<
697 + amd-pstate-ondemand <<
698 +------------------------------------ <<
699 + amd-pstate-ondemand <<
700 +------------------------------------ <<
701 + amd-pstate-ondemand <<
702 +------------------------------------ <<
703 + amd-pstate-ondemand <<
704 +------------------------------------ <<
705 + amd-pstate-schedutil <<
706 +------------------------------------ <<
707 + amd-pstate-schedutil <<
708 +------------------------------------ <<
709 + amd-pstate-schedutil <<
710 +------------------------------------ <<
711 + amd-pstate-schedutil <<
712 +------------------------------------ <<
713 + acpi-cpufreq-ondemand <<
714 +------------------------------------ <<
715 + acpi-cpufreq-ondemand <<
716 +------------------------------------ <<
717 + acpi-cpufreq-ondemand <<
718 +------------------------------------ <<
719 + acpi-cpufreq-ondemand <<
720 +------------------------------------ <<
721 + acpi-cpufreq-schedutil <<
722 +------------------------------------ <<
723 + acpi-cpufreq-schedutil <<
724 +------------------------------------ <<
725 + acpi-cpufreq-schedutil <<
726 +------------------------------------ <<
727 + acpi-cpufreq-schedutil <<
728 +------------------------------------ <<
729 + acpi-cpufreq-ondemand VS acpi-cpufr <<
730 +------------------------------------ <<
731 + amd-pstate-ondemand VS amd-pstate-s <<
732 +------------------------------------ <<
733 + acpi-cpufreq-ondemand VS amd-pstate <<
734 +------------------------------------ <<
735 + acpi-cpufreq-schedutil VS amd-pstat <<
736 +------------------------------------ <<
737 <<
738 + gitsource <<
739 <<
740 When you finish test, you will get se <<
741 The selftest.gitsource.csv file conta <<
742 The png images shows the performance, <<
743 Open selftest.gitsource.csv : <<
744 <<
745 +------------------------------------ <<
746 + Governor <<
747 +------------------------------------ <<
748 + Unit <<
749 +==================================== <<
750 + amd-pstate-ondemand <<
751 +------------------------------------ <<
752 + amd-pstate-ondemand <<
753 +------------------------------------ <<
754 + amd-pstate-ondemand <<
755 +------------------------------------ <<
756 + amd-pstate-ondemand <<
757 +------------------------------------ <<
758 + amd-pstate-schedutil <<
759 +------------------------------------ <<
760 + amd-pstate-schedutil <<
761 +------------------------------------ <<
762 + amd-pstate-schedutil <<
763 +------------------------------------ <<
764 + amd-pstate-schedutil <<
765 +------------------------------------ <<
766 + acpi-cpufreq-ondemand <<
767 +------------------------------------ <<
768 + acpi-cpufreq-ondemand <<
769 +------------------------------------ <<
770 + acpi-cpufreq-ondemand <<
771 +------------------------------------ <<
772 + acpi-cpufreq-ondemand <<
773 +------------------------------------ <<
774 + acpi-cpufreq-schedutil <<
775 +------------------------------------ <<
776 + acpi-cpufreq-schedutil <<
777 +------------------------------------ <<
778 + acpi-cpufreq-schedutil <<
779 +------------------------------------ <<
780 + acpi-cpufreq-schedutil <<
781 +------------------------------------ <<
782 + acpi-cpufreq-ondemand VS acpi-cpufr <<
783 +------------------------------------ <<
784 + amd-pstate-ondemand VS amd-pstate-s <<
785 +------------------------------------ <<
786 + acpi-cpufreq-ondemand VS amd-pstate <<
787 +------------------------------------ <<
788 + acpi-cpufreq-schedutil VS amd-pstat <<
789 +------------------------------------ <<
790 396
791 Reference 397 Reference
792 =========== 398 ===========
793 399
794 .. [1] AMD64 Architecture Programmer's Manual 400 .. [1] AMD64 Architecture Programmer's Manual Volume 2: System Programming,
795 https://www.amd.com/system/files/TechDo 401 https://www.amd.com/system/files/TechDocs/24593.pdf
796 402
797 .. [2] Advanced Configuration and Power Interf 403 .. [2] Advanced Configuration and Power Interface Specification,
798 https://uefi.org/sites/default/files/re 404 https://uefi.org/sites/default/files/resources/ACPI_Spec_6_4_Jan22.pdf
799 405
800 .. [3] Processor Programming Reference (PPR) f 406 .. [3] Processor Programming Reference (PPR) for AMD Family 19h Model 51h, Revision A1 Processors
801 https://www.amd.com/system/files/TechDo 407 https://www.amd.com/system/files/TechDocs/56569-A1-PUB.zip
802 <<
803 .. [4] Linux Kernel Selftests, <<
804 https://www.kernel.org/doc/html/latest/ <<
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.