1.. SPDX-License-Identifier: GPL-2.0
2.. include:: <isonum.txt>
3
4===============================================
5``intel_pstate`` CPU Performance Scaling Driver
6===============================================
7
8:Copyright: |copy| 2017 Intel Corporation
9
10:Author: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
11
12
13General Information
14===================
15
16``intel_pstate`` is a part of the
17:doc:`CPU performance scaling subsystem <cpufreq>` in the Linux kernel
18(``CPUFreq``).  It is a scaling driver for the Sandy Bridge and later
19generations of Intel processors.  Note, however, that some of those processors
20may not be supported.  [To understand ``intel_pstate`` it is necessary to know
21how ``CPUFreq`` works in general, so this is the time to read :doc:`cpufreq` if
22you have not done that yet.]
23
24For the processors supported by ``intel_pstate``, the P-state concept is broader
25than just an operating frequency or an operating performance point (see the
26LinuxCon Europe 2015 presentation by Kristen Accardi [1]_ for more
27information about that).  For this reason, the representation of P-states used
28by ``intel_pstate`` internally follows the hardware specification (for details
29refer to Intel Software Developer’s Manual [2]_).  However, the ``CPUFreq`` core
30uses frequencies for identifying operating performance points of CPUs and
31frequencies are involved in the user space interface exposed by it, so
32``intel_pstate`` maps its internal representation of P-states to frequencies too
33(fortunately, that mapping is unambiguous).  At the same time, it would not be
34practical for ``intel_pstate`` to supply the ``CPUFreq`` core with a table of
35available frequencies due to the possible size of it, so the driver does not do
36that.  Some functionality of the core is limited by that.
37
38Since the hardware P-state selection interface used by ``intel_pstate`` is
39available at the logical CPU level, the driver always works with individual
40CPUs.  Consequently, if ``intel_pstate`` is in use, every ``CPUFreq`` policy
41object corresponds to one logical CPU and ``CPUFreq`` policies are effectively
42equivalent to CPUs.  In particular, this means that they become "inactive" every
43time the corresponding CPU is taken offline and need to be re-initialized when
44it goes back online.
45
46``intel_pstate`` is not modular, so it cannot be unloaded, which means that the
47only way to pass early-configuration-time parameters to it is via the kernel
48command line.  However, its configuration can be adjusted via ``sysfs`` to a
49great extent.  In some configurations it even is possible to unregister it via
50``sysfs`` which allows another ``CPUFreq`` scaling driver to be loaded and
51registered (see `below <status_attr_>`_).
52
53
54Operation Modes
55===============
56
57``intel_pstate`` can operate in three different modes: in the active mode with
58or without hardware-managed P-states support and in the passive mode.  Which of
59them will be in effect depends on what kernel command line options are used and
60on the capabilities of the processor.
61
62Active Mode
63-----------
64
65This is the default operation mode of ``intel_pstate``.  If it works in this
66mode, the ``scaling_driver`` policy attribute in ``sysfs`` for all ``CPUFreq``
67policies contains the string "intel_pstate".
68
69In this mode the driver bypasses the scaling governors layer of ``CPUFreq`` and
70provides its own scaling algorithms for P-state selection.  Those algorithms
71can be applied to ``CPUFreq`` policies in the same way as generic scaling
72governors (that is, through the ``scaling_governor`` policy attribute in
73``sysfs``).  [Note that different P-state selection algorithms may be chosen for
74different policies, but that is not recommended.]
75
76They are not generic scaling governors, but their names are the same as the
77names of some of those governors.  Moreover, confusingly enough, they generally
78do not work in the same way as the generic governors they share the names with.
79For example, the ``powersave`` P-state selection algorithm provided by
80``intel_pstate`` is not a counterpart of the generic ``powersave`` governor
81(roughly, it corresponds to the ``schedutil`` and ``ondemand`` governors).
82
83There are two P-state selection algorithms provided by ``intel_pstate`` in the
84active mode: ``powersave`` and ``performance``.  The way they both operate
85depends on whether or not the hardware-managed P-states (HWP) feature has been
86enabled in the processor and possibly on the processor model.
87
88Which of the P-state selection algorithms is used by default depends on the
89:c:macro:`CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE` kernel configuration option.
90Namely, if that option is set, the ``performance`` algorithm will be used by
91default, and the other one will be used by default if it is not set.
92
93Active Mode With HWP
94~~~~~~~~~~~~~~~~~~~~
95
96If the processor supports the HWP feature, it will be enabled during the
97processor initialization and cannot be disabled after that.  It is possible
98to avoid enabling it by passing the ``intel_pstate=no_hwp`` argument to the
99kernel in the command line.
100
101If the HWP feature has been enabled, ``intel_pstate`` relies on the processor to
102select P-states by itself, but still it can give hints to the processor's
103internal P-state selection logic.  What those hints are depends on which P-state
104selection algorithm has been applied to the given policy (or to the CPU it
105corresponds to).
106
107Even though the P-state selection is carried out by the processor automatically,
108``intel_pstate`` registers utilization update callbacks with the CPU scheduler
109in this mode.  However, they are not used for running a P-state selection
110algorithm, but for periodic updates of the current CPU frequency information to
111be made available from the ``scaling_cur_freq`` policy attribute in ``sysfs``.
112
113HWP + ``performance``
114.....................
115
116In this configuration ``intel_pstate`` will write 0 to the processor's
117Energy-Performance Preference (EPP) knob (if supported) or its
118Energy-Performance Bias (EPB) knob (otherwise), which means that the processor's
119internal P-state selection logic is expected to focus entirely on performance.
120
121This will override the EPP/EPB setting coming from the ``sysfs`` interface
122(see `Energy vs Performance Hints`_ below).
123
124Also, in this configuration the range of P-states available to the processor's
125internal P-state selection logic is always restricted to the upper boundary
126(that is, the maximum P-state that the driver is allowed to use).
127
128HWP + ``powersave``
129...................
130
131In this configuration ``intel_pstate`` will set the processor's
132Energy-Performance Preference (EPP) knob (if supported) or its
133Energy-Performance Bias (EPB) knob (otherwise) to whatever value it was
134previously set to via ``sysfs`` (or whatever default value it was
135set to by the platform firmware).  This usually causes the processor's
136internal P-state selection logic to be less performance-focused.
137
138Active Mode Without HWP
139~~~~~~~~~~~~~~~~~~~~~~~
140
141This is the default operation mode for processors that do not support the HWP
142feature.  It also is used by default with the ``intel_pstate=no_hwp`` argument
143in the kernel command line.  However, in this mode ``intel_pstate`` may refuse
144to work with the given processor if it does not recognize it.  [Note that
145``intel_pstate`` will never refuse to work with any processor with the HWP
146feature enabled.]
147
148In this mode ``intel_pstate`` registers utilization update callbacks with the
149CPU scheduler in order to run a P-state selection algorithm, either
150``powersave`` or ``performance``, depending on the ``scaling_governor`` policy
151setting in ``sysfs``.  The current CPU frequency information to be made
152available from the ``scaling_cur_freq`` policy attribute in ``sysfs`` is
153periodically updated by those utilization update callbacks too.
154
155``performance``
156...............
157
158Without HWP, this P-state selection algorithm is always the same regardless of
159the processor model and platform configuration.
160
161It selects the maximum P-state it is allowed to use, subject to limits set via
162``sysfs``, every time the driver configuration for the given CPU is updated
163(e.g. via ``sysfs``).
164
165This is the default P-state selection algorithm if the
166:c:macro:`CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE` kernel configuration option
167is set.
168
169``powersave``
170.............
171
172Without HWP, this P-state selection algorithm is similar to the algorithm
173implemented by the generic ``schedutil`` scaling governor except that the
174utilization metric used by it is based on numbers coming from feedback
175registers of the CPU.  It generally selects P-states proportional to the
176current CPU utilization.
177
178This algorithm is run by the driver's utilization update callback for the
179given CPU when it is invoked by the CPU scheduler, but not more often than
180every 10 ms.  Like in the ``performance`` case, the hardware configuration
181is not touched if the new P-state turns out to be the same as the current
182one.
183
184This is the default P-state selection algorithm if the
185:c:macro:`CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE` kernel configuration option
186is not set.
187
188Passive Mode
189------------
190
191This mode is used if the ``intel_pstate=passive`` argument is passed to the
192kernel in the command line (it implies the ``intel_pstate=no_hwp`` setting too).
193Like in the active mode without HWP support, in this mode ``intel_pstate`` may
194refuse to work with the given processor if it does not recognize it.
195
196If the driver works in this mode, the ``scaling_driver`` policy attribute in
197``sysfs`` for all ``CPUFreq`` policies contains the string "intel_cpufreq".
198Then, the driver behaves like a regular ``CPUFreq`` scaling driver.  That is,
199it is invoked by generic scaling governors when necessary to talk to the
200hardware in order to change the P-state of a CPU (in particular, the
201``schedutil`` governor can invoke it directly from scheduler context).
202
203While in this mode, ``intel_pstate`` can be used with all of the (generic)
204scaling governors listed by the ``scaling_available_governors`` policy attribute
205in ``sysfs`` (and the P-state selection algorithms described above are not
206used).  Then, it is responsible for the configuration of policy objects
207corresponding to CPUs and provides the ``CPUFreq`` core (and the scaling
208governors attached to the policy objects) with accurate information on the
209maximum and minimum operating frequencies supported by the hardware (including
210the so-called "turbo" frequency ranges).  In other words, in the passive mode
211the entire range of available P-states is exposed by ``intel_pstate`` to the
212``CPUFreq`` core.  However, in this mode the driver does not register
213utilization update callbacks with the CPU scheduler and the ``scaling_cur_freq``
214information comes from the ``CPUFreq`` core (and is the last frequency selected
215by the current scaling governor for the given policy).
216
217
218.. _turbo:
219
220Turbo P-states Support
221======================
222
223In the majority of cases, the entire range of P-states available to
224``intel_pstate`` can be divided into two sub-ranges that correspond to
225different types of processor behavior, above and below a boundary that
226will be referred to as the "turbo threshold" in what follows.
227
228The P-states above the turbo threshold are referred to as "turbo P-states" and
229the whole sub-range of P-states they belong to is referred to as the "turbo
230range".  These names are related to the Turbo Boost technology allowing a
231multicore processor to opportunistically increase the P-state of one or more
232cores if there is enough power to do that and if that is not going to cause the
233thermal envelope of the processor package to be exceeded.
234
235Specifically, if software sets the P-state of a CPU core within the turbo range
236(that is, above the turbo threshold), the processor is permitted to take over
237performance scaling control for that core and put it into turbo P-states of its
238choice going forward.  However, that permission is interpreted differently by
239different processor generations.  Namely, the Sandy Bridge generation of
240processors will never use any P-states above the last one set by software for
241the given core, even if it is within the turbo range, whereas all of the later
242processor generations will take it as a license to use any P-states from the
243turbo range, even above the one set by software.  In other words, on those
244processors setting any P-state from the turbo range will enable the processor
245to put the given core into all turbo P-states up to and including the maximum
246supported one as it sees fit.
247
248One important property of turbo P-states is that they are not sustainable.  More
249precisely, there is no guarantee that any CPUs will be able to stay in any of
250those states indefinitely, because the power distribution within the processor
251package may change over time  or the thermal envelope it was designed for might
252be exceeded if a turbo P-state was used for too long.
253
254In turn, the P-states below the turbo threshold generally are sustainable.  In
255fact, if one of them is set by software, the processor is not expected to change
256it to a lower one unless in a thermal stress or a power limit violation
257situation (a higher P-state may still be used if it is set for another CPU in
258the same package at the same time, for example).
259
260Some processors allow multiple cores to be in turbo P-states at the same time,
261but the maximum P-state that can be set for them generally depends on the number
262of cores running concurrently.  The maximum turbo P-state that can be set for 3
263cores at the same time usually is lower than the analogous maximum P-state for
2642 cores, which in turn usually is lower than the maximum turbo P-state that can
265be set for 1 core.  The one-core maximum turbo P-state is thus the maximum
266supported one overall.
267
268The maximum supported turbo P-state, the turbo threshold (the maximum supported
269non-turbo P-state) and the minimum supported P-state are specific to the
270processor model and can be determined by reading the processor's model-specific
271registers (MSRs).  Moreover, some processors support the Configurable TDP
272(Thermal Design Power) feature and, when that feature is enabled, the turbo
273threshold effectively becomes a configurable value that can be set by the
274platform firmware.
275
276Unlike ``_PSS`` objects in the ACPI tables, ``intel_pstate`` always exposes
277the entire range of available P-states, including the whole turbo range, to the
278``CPUFreq`` core and (in the passive mode) to generic scaling governors.  This
279generally causes turbo P-states to be set more often when ``intel_pstate`` is
280used relative to ACPI-based CPU performance scaling (see `below <acpi-cpufreq_>`_
281for more information).
282
283Moreover, since ``intel_pstate`` always knows what the real turbo threshold is
284(even if the Configurable TDP feature is enabled in the processor), its
285``no_turbo`` attribute in ``sysfs`` (described `below <no_turbo_attr_>`_) should
286work as expected in all cases (that is, if set to disable turbo P-states, it
287always should prevent ``intel_pstate`` from using them).
288
289
290Processor Support
291=================
292
293To handle a given processor ``intel_pstate`` requires a number of different
294pieces of information on it to be known, including:
295
296 * The minimum supported P-state.
297
298 * The maximum supported `non-turbo P-state <turbo_>`_.
299
300 * Whether or not turbo P-states are supported at all.
301
302 * The maximum supported `one-core turbo P-state <turbo_>`_ (if turbo P-states
303   are supported).
304
305 * The scaling formula to translate the driver's internal representation
306   of P-states into frequencies and the other way around.
307
308Generally, ways to obtain that information are specific to the processor model
309or family.  Although it often is possible to obtain all of it from the processor
310itself (using model-specific registers), there are cases in which hardware
311manuals need to be consulted to get to it too.
312
313For this reason, there is a list of supported processors in ``intel_pstate`` and
314the driver initialization will fail if the detected processor is not in that
315list, unless it supports the `HWP feature <Active Mode_>`_.  [The interface to
316obtain all of the information listed above is the same for all of the processors
317supporting the HWP feature, which is why they all are supported by
318``intel_pstate``.]
319
320
321User Space Interface in ``sysfs``
322=================================
323
324Global Attributes
325-----------------
326
327``intel_pstate`` exposes several global attributes (files) in ``sysfs`` to
328control its functionality at the system level.  They are located in the
329``/sys/devices/system/cpu/intel_pstate/`` directory and affect all CPUs.
330
331Some of them are not present if the ``intel_pstate=per_cpu_perf_limits``
332argument is passed to the kernel in the command line.
333
334``max_perf_pct``
335	Maximum P-state the driver is allowed to set in percent of the
336	maximum supported performance level (the highest supported `turbo
337	P-state <turbo_>`_).
338
339	This attribute will not be exposed if the
340	``intel_pstate=per_cpu_perf_limits`` argument is present in the kernel
341	command line.
342
343``min_perf_pct``
344	Minimum P-state the driver is allowed to set in percent of the
345	maximum supported performance level (the highest supported `turbo
346	P-state <turbo_>`_).
347
348	This attribute will not be exposed if the
349	``intel_pstate=per_cpu_perf_limits`` argument is present in the kernel
350	command line.
351
352``num_pstates``
353	Number of P-states supported by the processor (between 0 and 255
354	inclusive) including both turbo and non-turbo P-states (see
355	`Turbo P-states Support`_).
356
357	The value of this attribute is not affected by the ``no_turbo``
358	setting described `below <no_turbo_attr_>`_.
359
360	This attribute is read-only.
361
362``turbo_pct``
363	Ratio of the `turbo range <turbo_>`_ size to the size of the entire
364	range of supported P-states, in percent.
365
366	This attribute is read-only.
367
368.. _no_turbo_attr:
369
370``no_turbo``
371	If set (equal to 1), the driver is not allowed to set any turbo P-states
372	(see `Turbo P-states Support`_).  If unset (equalt to 0, which is the
373	default), turbo P-states can be set by the driver.
374	[Note that ``intel_pstate`` does not support the general ``boost``
375	attribute (supported by some other scaling drivers) which is replaced
376	by this one.]
377
378	This attrubute does not affect the maximum supported frequency value
379	supplied to the ``CPUFreq`` core and exposed via the policy interface,
380	but it affects the maximum possible value of per-policy P-state	limits
381	(see `Interpretation of Policy Attributes`_ below for details).
382
383``hwp_dynamic_boost``
384	This attribute is only present if ``intel_pstate`` works in the
385	`active mode with the HWP feature enabled <Active Mode With HWP_>`_ in
386	the processor.  If set (equal to 1), it causes the minimum P-state limit
387	to be increased dynamically for a short time whenever a task previously
388	waiting on I/O is selected to run on a given logical CPU (the purpose
389	of this mechanism is to improve performance).
390
391	This setting has no effect on logical CPUs whose minimum P-state limit
392	is directly set to the highest non-turbo P-state or above it.
393
394.. _status_attr:
395
396``status``
397	Operation mode of the driver: "active", "passive" or "off".
398
399	"active"
400		The driver is functional and in the `active mode
401		<Active Mode_>`_.
402
403	"passive"
404		The driver is functional and in the `passive mode
405		<Passive Mode_>`_.
406
407	"off"
408		The driver is not functional (it is not registered as a scaling
409		driver with the ``CPUFreq`` core).
410
411	This attribute can be written to in order to change the driver's
412	operation mode or to unregister it.  The string written to it must be
413	one of the possible values of it and, if successful, the write will
414	cause the driver to switch over to the operation mode represented by
415	that string - or to be unregistered in the "off" case.  [Actually,
416	switching over from the active mode to the passive mode or the other
417	way around causes the driver to be unregistered and registered again
418	with a different set of callbacks, so all of its settings (the global
419	as well as the per-policy ones) are then reset to their default
420	values, possibly depending on the target operation mode.]
421
422	That only is supported in some configurations, though (for example, if
423	the `HWP feature is enabled in the processor <Active Mode With HWP_>`_,
424	the operation mode of the driver cannot be changed), and if it is not
425	supported in the current configuration, writes to this attribute will
426	fail with an appropriate error.
427
428Interpretation of Policy Attributes
429-----------------------------------
430
431The interpretation of some ``CPUFreq`` policy attributes described in
432:doc:`cpufreq` is special with ``intel_pstate`` as the current scaling driver
433and it generally depends on the driver's `operation mode <Operation Modes_>`_.
434
435First of all, the values of the ``cpuinfo_max_freq``, ``cpuinfo_min_freq`` and
436``scaling_cur_freq`` attributes are produced by applying a processor-specific
437multiplier to the internal P-state representation used by ``intel_pstate``.
438Also, the values of the ``scaling_max_freq`` and ``scaling_min_freq``
439attributes are capped by the frequency corresponding to the maximum P-state that
440the driver is allowed to set.
441
442If the ``no_turbo`` `global attribute <no_turbo_attr_>`_ is set, the driver is
443not allowed to use turbo P-states, so the maximum value of ``scaling_max_freq``
444and ``scaling_min_freq`` is limited to the maximum non-turbo P-state frequency.
445Accordingly, setting ``no_turbo`` causes ``scaling_max_freq`` and
446``scaling_min_freq`` to go down to that value if they were above it before.
447However, the old values of ``scaling_max_freq`` and ``scaling_min_freq`` will be
448restored after unsetting ``no_turbo``, unless these attributes have been written
449to after ``no_turbo`` was set.
450
451If ``no_turbo`` is not set, the maximum possible value of ``scaling_max_freq``
452and ``scaling_min_freq`` corresponds to the maximum supported turbo P-state,
453which also is the value of ``cpuinfo_max_freq`` in either case.
454
455Next, the following policy attributes have special meaning if
456``intel_pstate`` works in the `active mode <Active Mode_>`_:
457
458``scaling_available_governors``
459	List of P-state selection algorithms provided by ``intel_pstate``.
460
461``scaling_governor``
462	P-state selection algorithm provided by ``intel_pstate`` currently in
463	use with the given policy.
464
465``scaling_cur_freq``
466	Frequency of the average P-state of the CPU represented by the given
467	policy for the time interval between the last two invocations of the
468	driver's utilization update callback by the CPU scheduler for that CPU.
469
470One more policy attribute is present if the `HWP feature is enabled in the
471processor <Active Mode With HWP_>`_:
472
473``base_frequency``
474	Shows the base frequency of the CPU. Any frequency above this will be
475	in the turbo frequency range.
476
477The meaning of these attributes in the `passive mode <Passive Mode_>`_ is the
478same as for other scaling drivers.
479
480Additionally, the value of the ``scaling_driver`` attribute for ``intel_pstate``
481depends on the operation mode of the driver.  Namely, it is either
482"intel_pstate" (in the `active mode <Active Mode_>`_) or "intel_cpufreq" (in the
483`passive mode <Passive Mode_>`_).
484
485Coordination of P-State Limits
486------------------------------
487
488``intel_pstate`` allows P-state limits to be set in two ways: with the help of
489the ``max_perf_pct`` and ``min_perf_pct`` `global attributes
490<Global Attributes_>`_ or via the ``scaling_max_freq`` and ``scaling_min_freq``
491``CPUFreq`` policy attributes.  The coordination between those limits is based
492on the following rules, regardless of the current operation mode of the driver:
493
494 1. All CPUs are affected by the global limits (that is, none of them can be
495    requested to run faster than the global maximum and none of them can be
496    requested to run slower than the global minimum).
497
498 2. Each individual CPU is affected by its own per-policy limits (that is, it
499    cannot be requested to run faster than its own per-policy maximum and it
500    cannot be requested to run slower than its own per-policy minimum). The
501    effective performance depends on whether the platform supports per core
502    P-states, hyper-threading is enabled and on current performance requests
503    from other CPUs. When platform doesn't support per core P-states, the
504    effective performance can be more than the policy limits set on a CPU, if
505    other CPUs are requesting higher performance at that moment. Even with per
506    core P-states support, when hyper-threading is enabled, if the sibling CPU
507    is requesting higher performance, the other siblings will get higher
508    performance than their policy limits.
509
510 3. The global and per-policy limits can be set independently.
511
512If the `HWP feature is enabled in the processor <Active Mode With HWP_>`_, the
513resulting effective values are written into its registers whenever the limits
514change in order to request its internal P-state selection logic to always set
515P-states within these limits.  Otherwise, the limits are taken into account by
516scaling governors (in the `passive mode <Passive Mode_>`_) and by the driver
517every time before setting a new P-state for a CPU.
518
519Additionally, if the ``intel_pstate=per_cpu_perf_limits`` command line argument
520is passed to the kernel, ``max_perf_pct`` and ``min_perf_pct`` are not exposed
521at all and the only way to set the limits is by using the policy attributes.
522
523
524Energy vs Performance Hints
525---------------------------
526
527If ``intel_pstate`` works in the `active mode with the HWP feature enabled
528<Active Mode With HWP_>`_ in the processor, additional attributes are present
529in every ``CPUFreq`` policy directory in ``sysfs``.  They are intended to allow
530user space to help ``intel_pstate`` to adjust the processor's internal P-state
531selection logic by focusing it on performance or on energy-efficiency, or
532somewhere between the two extremes:
533
534``energy_performance_preference``
535	Current value of the energy vs performance hint for the given policy
536	(or the CPU represented by it).
537
538	The hint can be changed by writing to this attribute.
539
540``energy_performance_available_preferences``
541	List of strings that can be written to the
542	``energy_performance_preference`` attribute.
543
544	They represent different energy vs performance hints and should be
545	self-explanatory, except that ``default`` represents whatever hint
546	value was set by the platform firmware.
547
548Strings written to the ``energy_performance_preference`` attribute are
549internally translated to integer values written to the processor's
550Energy-Performance Preference (EPP) knob (if supported) or its
551Energy-Performance Bias (EPB) knob.
552
553[Note that tasks may by migrated from one CPU to another by the scheduler's
554load-balancing algorithm and if different energy vs performance hints are
555set for those CPUs, that may lead to undesirable outcomes.  To avoid such
556issues it is better to set the same energy vs performance hint for all CPUs
557or to pin every task potentially sensitive to them to a specific CPU.]
558
559.. _acpi-cpufreq:
560
561``intel_pstate`` vs ``acpi-cpufreq``
562====================================
563
564On the majority of systems supported by ``intel_pstate``, the ACPI tables
565provided by the platform firmware contain ``_PSS`` objects returning information
566that can be used for CPU performance scaling (refer to the ACPI specification
567[3]_ for details on the ``_PSS`` objects and the format of the information
568returned by them).
569
570The information returned by the ACPI ``_PSS`` objects is used by the
571``acpi-cpufreq`` scaling driver.  On systems supported by ``intel_pstate``
572the ``acpi-cpufreq`` driver uses the same hardware CPU performance scaling
573interface, but the set of P-states it can use is limited by the ``_PSS``
574output.
575
576On those systems each ``_PSS`` object returns a list of P-states supported by
577the corresponding CPU which basically is a subset of the P-states range that can
578be used by ``intel_pstate`` on the same system, with one exception: the whole
579`turbo range <turbo_>`_ is represented by one item in it (the topmost one).  By
580convention, the frequency returned by ``_PSS`` for that item is greater by 1 MHz
581than the frequency of the highest non-turbo P-state listed by it, but the
582corresponding P-state representation (following the hardware specification)
583returned for it matches the maximum supported turbo P-state (or is the
584special value 255 meaning essentially "go as high as you can get").
585
586The list of P-states returned by ``_PSS`` is reflected by the table of
587available frequencies supplied by ``acpi-cpufreq`` to the ``CPUFreq`` core and
588scaling governors and the minimum and maximum supported frequencies reported by
589it come from that list as well.  In particular, given the special representation
590of the turbo range described above, this means that the maximum supported
591frequency reported by ``acpi-cpufreq`` is higher by 1 MHz than the frequency
592of the highest supported non-turbo P-state listed by ``_PSS`` which, of course,
593affects decisions made by the scaling governors, except for ``powersave`` and
594``performance``.
595
596For example, if a given governor attempts to select a frequency proportional to
597estimated CPU load and maps the load of 100% to the maximum supported frequency
598(possibly multiplied by a constant), then it will tend to choose P-states below
599the turbo threshold if ``acpi-cpufreq`` is used as the scaling driver, because
600in that case the turbo range corresponds to a small fraction of the frequency
601band it can use (1 MHz vs 1 GHz or more).  In consequence, it will only go to
602the turbo range for the highest loads and the other loads above 50% that might
603benefit from running at turbo frequencies will be given non-turbo P-states
604instead.
605
606One more issue related to that may appear on systems supporting the
607`Configurable TDP feature <turbo_>`_ allowing the platform firmware to set the
608turbo threshold.  Namely, if that is not coordinated with the lists of P-states
609returned by ``_PSS`` properly, there may be more than one item corresponding to
610a turbo P-state in those lists and there may be a problem with avoiding the
611turbo range (if desirable or necessary).  Usually, to avoid using turbo
612P-states overall, ``acpi-cpufreq`` simply avoids using the topmost state listed
613by ``_PSS``, but that is not sufficient when there are other turbo P-states in
614the list returned by it.
615
616Apart from the above, ``acpi-cpufreq`` works like ``intel_pstate`` in the
617`passive mode <Passive Mode_>`_, except that the number of P-states it can set
618is limited to the ones listed by the ACPI ``_PSS`` objects.
619
620
621Kernel Command Line Options for ``intel_pstate``
622================================================
623
624Several kernel command line options can be used to pass early-configuration-time
625parameters to ``intel_pstate`` in order to enforce specific behavior of it.  All
626of them have to be prepended with the ``intel_pstate=`` prefix.
627
628``disable``
629	Do not register ``intel_pstate`` as the scaling driver even if the
630	processor is supported by it.
631
632``passive``
633	Register ``intel_pstate`` in the `passive mode <Passive Mode_>`_ to
634	start with.
635
636	This option implies the ``no_hwp`` one described below.
637
638``force``
639	Register ``intel_pstate`` as the scaling driver instead of
640	``acpi-cpufreq`` even if the latter is preferred on the given system.
641
642	This may prevent some platform features (such as thermal controls and
643	power capping) that rely on the availability of ACPI P-states
644	information from functioning as expected, so it should be used with
645	caution.
646
647	This option does not work with processors that are not supported by
648	``intel_pstate`` and on platforms where the ``pcc-cpufreq`` scaling
649	driver is used instead of ``acpi-cpufreq``.
650
651``no_hwp``
652	Do not enable the `hardware-managed P-states (HWP) feature
653	<Active Mode With HWP_>`_ even if it is supported by the processor.
654
655``hwp_only``
656	Register ``intel_pstate`` as the scaling driver only if the
657	`hardware-managed P-states (HWP) feature <Active Mode With HWP_>`_ is
658	supported by the processor.
659
660``support_acpi_ppc``
661	Take ACPI ``_PPC`` performance limits into account.
662
663	If the preferred power management profile in the FADT (Fixed ACPI
664	Description Table) is set to "Enterprise Server" or "Performance
665	Server", the ACPI ``_PPC`` limits are taken into account by default
666	and this option has no effect.
667
668``per_cpu_perf_limits``
669	Use per-logical-CPU P-State limits (see `Coordination of P-state
670	Limits`_ for details).
671
672
673Diagnostics and Tuning
674======================
675
676Trace Events
677------------
678
679There are two static trace events that can be used for ``intel_pstate``
680diagnostics.  One of them is the ``cpu_frequency`` trace event generally used
681by ``CPUFreq``, and the other one is the ``pstate_sample`` trace event specific
682to ``intel_pstate``.  Both of them are triggered by ``intel_pstate`` only if
683it works in the `active mode <Active Mode_>`_.
684
685The following sequence of shell commands can be used to enable them and see
686their output (if the kernel is generally configured to support event tracing)::
687
688 # cd /sys/kernel/debug/tracing/
689 # echo 1 > events/power/pstate_sample/enable
690 # echo 1 > events/power/cpu_frequency/enable
691 # cat trace
692 gnome-terminal--4510  [001] ..s.  1177.680733: pstate_sample: core_busy=107 scaled=94 from=26 to=26 mperf=1143818 aperf=1230607 tsc=29838618 freq=2474476
693 cat-5235  [002] ..s.  1177.681723: cpu_frequency: state=2900000 cpu_id=2
694
695If ``intel_pstate`` works in the `passive mode <Passive Mode_>`_, the
696``cpu_frequency`` trace event will be triggered either by the ``schedutil``
697scaling governor (for the policies it is attached to), or by the ``CPUFreq``
698core (for the policies with other scaling governors).
699
700``ftrace``
701----------
702
703The ``ftrace`` interface can be used for low-level diagnostics of
704``intel_pstate``.  For example, to check how often the function to set a
705P-state is called, the ``ftrace`` filter can be set to to
706:c:func:`intel_pstate_set_pstate`::
707
708 # cd /sys/kernel/debug/tracing/
709 # cat available_filter_functions | grep -i pstate
710 intel_pstate_set_pstate
711 intel_pstate_cpu_init
712 ...
713 # echo intel_pstate_set_pstate > set_ftrace_filter
714 # echo function > current_tracer
715 # cat trace | head -15
716 # tracer: function
717 #
718 # entries-in-buffer/entries-written: 80/80   #P:4
719 #
720 #                              _-----=> irqs-off
721 #                             / _----=> need-resched
722 #                            | / _---=> hardirq/softirq
723 #                            || / _--=> preempt-depth
724 #                            ||| /     delay
725 #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
726 #              | |       |   ||||       |         |
727             Xorg-3129  [000] ..s.  2537.644844: intel_pstate_set_pstate <-intel_pstate_timer_func
728  gnome-terminal--4510  [002] ..s.  2537.649844: intel_pstate_set_pstate <-intel_pstate_timer_func
729      gnome-shell-3409  [001] ..s.  2537.650850: intel_pstate_set_pstate <-intel_pstate_timer_func
730           <idle>-0     [000] ..s.  2537.654843: intel_pstate_set_pstate <-intel_pstate_timer_func
731
732
733References
734==========
735
736.. [1] Kristen Accardi, *Balancing Power and Performance in the Linux Kernel*,
737       https://events.static.linuxfound.org/sites/events/files/slides/LinuxConEurope_2015.pdf
738
739.. [2] *Intel® 64 and IA-32 Architectures Software Developer’s Manual Volume 3: System Programming Guide*,
740       https://www.intel.com/content/www/us/en/architecture-and-technology/64-ia-32-architectures-software-developer-system-programming-manual-325384.html
741
742.. [3] *Advanced Configuration and Power Interface Specification*,
743       https://uefi.org/sites/default/files/resources/ACPI_6_3_final_Jan30.pdf
744