Commit 02c3de11 authored by Linus Torvalds's avatar Linus Torvalds
Browse files

Merge tag 'pm-4.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull power management updates from Rafael Wysocki:
 "The majority of changes go into the Operating Performance Points (OPP)
  framework and cpufreq this time, followed by devfreq and some
  scattered updates all over.

  The OPP changes are mostly related to switching over from RCU-based
  synchronization, that turned out to be overly complicated and
  problematic, to reference counting using krefs.

  In the cpufreq land there are core cleanups, documentation updates, a
  new driver for Broadcom BMIPS SoCs, a new cpufreq-dt sub-driver for TI
  SoCs that require special handling, ARM64 SoCs support for the qoriq
  driver, intel_pstate updates, powernv driver update and assorted
  fixes.

  The devfreq changes are mostly fixes related to the sysfs interface
  and some Exynos drivers updates.

  Apart from that, the cpuidle menu governor will support per-CPU PM QoS
  constraints for the wakeup latency now, some bugs in the wakeup IRQs
  framework are fixed, the generic power domains framework should handle
  asynchronous invocations of *noirq suspend/resume callbacks from now
  on, the analyze_suspend.py script is updated and there is a new tool
  for intel_pstate diagnostics.

  Specifics:

   - Operating Performance Points (OPP) framework fixes, cleanups and
     switch over from RCU-based synchronization to reference counting
     using krefs (Viresh Kumar, Wei Yongjun, Dave Gerlach)

   - cpufreq core cleanups and documentation updates (Viresh Kumar,
     Rafael Wysocki)

   - New cpufreq driver for Broadcom BMIPS SoCs (Markus Mayer)

   - New cpufreq-dt sub-driver for TI SoCs requiring special handling,
     like in the AM335x, AM437x, DRA7x, and AM57x families, along with
     new DT bindings for it (Dave Gerlach, Paul Gortmaker)

   - ARM64 SoCs support for the qoriq cpufreq driver (Tang Yuantian)

   - intel_pstate driver updates including a new sysfs knob to control
     the driver's operation mode and fixes related to the no_turbo sysfs
     knob and the hardware-managed P-states feature support (Rafael
     Wysocki, Srinivas Pandruvada)

   - New interface to export ultra-turbo frequencies for the powernv
     cpufreq driver (Shilpasri Bhat)

   - Assorted fixes for cpufreq drivers (Arnd Bergmann, Dan Carpenter,
     Wei Yongjun)

   - devfreq core fixes, mostly related to the sysfs interface exported
     by it (Chanwoo Choi, Chris Diamand)

   - Updates of the exynos-bus and exynos-ppmu devfreq drivers (Chanwoo
     Choi)

   - Device PM QoS extension to support CPUs and support for per-CPU
     wakeup (device resume) latency constraints in the cpuidle menu
     governor (Alex Shi)

   - Wakeup IRQs framework fixes (Grygorii Strashko)

   - Generic power domains framework update including a fix to make it
     handle asynchronous invocations of *noirq suspend/resume callbacks
     correctly (Ulf Hansson, Geert Uytterhoeven)

   - Assorted fixes and cleanups in the core suspend/hibernate code, PM
     QoS framework and x86 ACPI idle support code (Corentin Labbe, Geert
     Uytterhoeven, Geliang Tang, John Keeping, Nick Desaulniers)

   - Update of the analyze_suspend.py script is updated to version 4.5
     offering multiple improvements (Todd Brandt)

   - New tool for intel_pstate diagnostics using the pstate_sample
     tracepoint (Doug Smythies)"

* tag 'pm-4.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (85 commits)
  MAINTAINERS: cpufreq: add bmips-cpufreq.c
  PM / QoS: Fix memory leak on resume_latency.notifiers
  PM / Documentation: Spelling s/wrtie/write/
  PM / sleep: Fix test_suspend after sleep state rework
  cpufreq: CPPC: add ACPI_PROCESSOR dependency
  cpufreq: make ti-cpufreq explicitly non-modular
  cpufreq: Do not clear real_cpus mask on policy init
  tools/power/x86: Debug utility for intel_pstate driver
  AnalyzeSuspend: fix drag and zoom bug in javascript
  PM / wakeirq: report a wakeup_event on dedicated wekup irq
  PM / wakeirq: Fix spurious wake-up events for dedicated wakeirqs
  PM / wakeirq: Enable dedicated wakeirq for suspend
  cpufreq: dt: Don't use generic platdev driver for ti-cpufreq platforms
  cpufreq: ti: Add cpufreq driver to determine available OPPs at runtime
  Documentation: dt: add bindings for ti-cpufreq
  PM / OPP: Expose _of_get_opp_desc_node as dev_pm_opp API
  cpufreq: qoriq: Don't look at clock implementation details
  cpufreq: qoriq: add ARM64 SoCs support
  PM / Domains: Provide dummy governors if CONFIG_PM_GENERIC_DOMAINS=n
  cpufreq: brcmstb-avs-cpufreq: remove unnecessary platform_set_drvdata()
  ...
parents 7aa7d608 eee77a8a
What: /sys/class/devfreq-event/event(x)/
Date: January 2017
Contact: Chanwoo Choi <cw00.choi@samsung.com>
Description:
Provide a place in sysfs for the devfreq-event objects.
This allows accessing various devfreq-event specific variables.
The name of devfreq-event object denoted as 'event(x)' which
includes the unique number of 'x' for each devfreq-event object.
What: /sys/class/devfreq-event/event(x)/name
Date: January 2017
Contact: Chanwoo Choi <cw00.choi@samsung.com>
Description:
The /sys/class/devfreq-event/event(x)/name attribute contains
the name of the devfreq-event object. This attribute is
read-only.
What: /sys/class/devfreq-event/event(x)/enable_count
Date: January 2017
Contact: Chanwoo Choi <cw00.choi@samsung.com>
Description:
The /sys/class/devfreq-event/event(x)/enable_count attribute
contains the reference count to enable the devfreq-event
object. If the device is enabled, the value of attribute is
greater than zero.
......@@ -8,6 +8,8 @@
Dominik Brodowski <linux@brodo.de>
David Kimdon <dwhedon@debian.org>
Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Viresh Kumar <viresh.kumar@linaro.org>
......@@ -36,10 +38,11 @@ speed limits (like LCD drivers on ARM architecture). Additionally, the
kernel "constant" loops_per_jiffy is updated on frequency changes
here.
Reference counting is done by cpufreq_get_cpu and cpufreq_put_cpu,
which make sure that the cpufreq processor driver is correctly
registered with the core, and will not be unloaded until
cpufreq_put_cpu is called.
Reference counting of the cpufreq policies is done by cpufreq_cpu_get
and cpufreq_cpu_put, which make sure that the cpufreq driver is
correctly registered with the core, and will not be unloaded until
cpufreq_put_cpu is called. That also ensures that the respective cpufreq
policy doesn't get freed while being used.
2. CPUFreq notifiers
====================
......@@ -69,18 +72,16 @@ CPUFreq policy notifier is called twice for a policy transition:
The phase is specified in the second argument to the notifier.
The third argument, a void *pointer, points to a struct cpufreq_policy
consisting of five values: cpu, min, max, policy and max_cpu_freq. min
and max are the lower and upper frequencies (in kHz) of the new
policy, policy the new policy, cpu the number of the affected CPU; and
max_cpu_freq the maximum supported CPU frequency. This value is given
for informational purposes only.
consisting of several values, including min, max (the lower and upper
frequencies (in kHz) of the new policy).
2.2 CPUFreq transition notifiers
--------------------------------
These are notified twice when the CPUfreq driver switches the CPU core
frequency and this change has any external implications.
These are notified twice for each online CPU in the policy, when the
CPUfreq driver switches the CPU core frequency and this change has no
any external implications.
The second argument specifies the phase - CPUFREQ_PRECHANGE or
CPUFREQ_POSTCHANGE.
......@@ -90,6 +91,7 @@ values:
cpu - number of the affected CPU
old - old frequency
new - new frequency
flags - flags of the cpufreq driver
3. CPUFreq Table Generation with Operating Performance Point (OPP)
==================================================================
......
......@@ -9,6 +9,8 @@
Dominik Brodowski <linux@brodo.de>
Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Viresh Kumar <viresh.kumar@linaro.org>
......@@ -49,49 +51,65 @@ using cpufreq_register_driver()
What shall this struct cpufreq_driver contain?
cpufreq_driver.name - The name of this driver.
.name - The name of this driver.
cpufreq_driver.init - A pointer to the per-CPU initialization
function.
.init - A pointer to the per-policy initialization function.
cpufreq_driver.verify - A pointer to a "verification" function.
.verify - A pointer to a "verification" function.
cpufreq_driver.setpolicy _or_
cpufreq_driver.target/
target_index - See below on the differences.
.setpolicy _or_ .fast_switch _or_ .target _or_ .target_index - See
below on the differences.
And optionally
cpufreq_driver.exit - A pointer to a per-CPU cleanup
function called during CPU_POST_DEAD
phase of cpu hotplug process.
.flags - Hints for the cpufreq core.
cpufreq_driver.stop_cpu - A pointer to a per-CPU stop function
called during CPU_DOWN_PREPARE phase of
cpu hotplug process.
.driver_data - cpufreq driver specific data.
cpufreq_driver.resume - A pointer to a per-CPU resume function
which is called with interrupts disabled
and _before_ the pre-suspend frequency
and/or policy is restored by a call to
->target/target_index or ->setpolicy.
.resolve_freq - Returns the most appropriate frequency for a target
frequency. Doesn't change the frequency though.
cpufreq_driver.attr - A pointer to a NULL-terminated list of
"struct freq_attr" which allow to
export values to sysfs.
.get_intermediate and target_intermediate - Used to switch to stable
frequency while changing CPU frequency.
cpufreq_driver.get_intermediate
and target_intermediate Used to switch to stable frequency while
changing CPU frequency.
.get - Returns current frequency of the CPU.
.bios_limit - Returns HW/BIOS max frequency limitations for the CPU.
.exit - A pointer to a per-policy cleanup function called during
CPU_POST_DEAD phase of cpu hotplug process.
.stop_cpu - A pointer to a per-policy stop function called during
CPU_DOWN_PREPARE phase of cpu hotplug process.
.suspend - A pointer to a per-policy suspend function which is called
with interrupts disabled and _after_ the governor is stopped for the
policy.
.resume - A pointer to a per-policy resume function which is called
with interrupts disabled and _before_ the governor is started again.
.ready - A pointer to a per-policy ready function which is called after
the policy is fully initialized.
.attr - A pointer to a NULL-terminated list of "struct freq_attr" which
allow to export values to sysfs.
.boost_enabled - If set, boost frequencies are enabled.
.set_boost - A pointer to a per-policy function to enable/disable boost
frequencies.
1.2 Per-CPU Initialization
--------------------------
Whenever a new CPU is registered with the device model, or after the
cpufreq driver registers itself, the per-CPU initialization function
cpufreq_driver.init is called. It takes a struct cpufreq_policy
*policy as argument. What to do now?
cpufreq driver registers itself, the per-policy initialization function
cpufreq_driver.init is called if no cpufreq policy existed for the CPU.
Note that the .init() and .exit() routines are called only once for the
policy and not for each CPU managed by the policy. It takes a struct
cpufreq_policy *policy as argument. What to do now?
If necessary, activate the CPUfreq support on your CPU.
......@@ -117,47 +135,45 @@ policy->governor must contain the "default policy" for
cpufreq_driver.setpolicy or
cpufreq_driver.target/target_index is called
with these values.
policy->cpus Update this with the masks of the
(online + offline) CPUs that do DVFS
along with this CPU (i.e. that share
clock/voltage rails with it).
For setting some of these values (cpuinfo.min[max]_freq, policy->min[max]), the
frequency table helpers might be helpful. See the section 2 for more information
on them.
SMP systems normally have same clock source for a group of cpus. For these the
.init() would be called only once for the first online cpu. Here the .init()
routine must initialize policy->cpus with mask of all possible cpus (Online +
Offline) that share the clock. Then the core would copy this mask onto
policy->related_cpus and will reset policy->cpus to carry only online cpus.
1.3 verify
------------
----------
When the user decides a new policy (consisting of
"policy,governor,min,max") shall be set, this policy must be validated
so that incompatible values can be corrected. For verifying these
values, a frequency table helper and/or the
cpufreq_verify_within_limits(struct cpufreq_policy *policy, unsigned
int min_freq, unsigned int max_freq) function might be helpful. See
section 2 for details on frequency table helpers.
values cpufreq_verify_within_limits(struct cpufreq_policy *policy,
unsigned int min_freq, unsigned int max_freq) function might be helpful.
See section 2 for details on frequency table helpers.
You need to make sure that at least one valid frequency (or operating
range) is within policy->min and policy->max. If necessary, increase
policy->max first, and only if this is no solution, decrease policy->min.
1.4 target/target_index or setpolicy?
----------------------------
1.4 target or target_index or setpolicy or fast_switch?
-------------------------------------------------------
Most cpufreq drivers or even most cpu frequency scaling algorithms
only allow the CPU to be set to one frequency. For these, you use the
->target/target_index call.
only allow the CPU frequency to be set to predefined fixed values. For
these, you use the ->target(), ->target_index() or ->fast_switch()
callbacks.
Some cpufreq-capable processors switch the frequency between certain
limits on their own. These shall use the ->setpolicy call
Some cpufreq capable processors switch the frequency between certain
limits on their own. These shall use the ->setpolicy() callback.
1.5. target/target_index
-------------
------------------------
The target_index call has two arguments: struct cpufreq_policy *policy,
and unsigned int index (into the exposed frequency table).
......@@ -186,9 +202,20 @@ actual frequency must be determined using the following rules:
Here again the frequency table helper might assist you - see section 2
for details.
1.6. fast_switch
----------------
1.6 setpolicy
---------------
This function is used for frequency switching from scheduler's context.
Not all drivers are expected to implement it, as sleeping from within
this callback isn't allowed. This callback must be highly optimized to
do switching as fast as possible.
This function has two arguments: struct cpufreq_policy *policy and
unsigned int target_frequency.
1.7 setpolicy
-------------
The setpolicy call only takes a struct cpufreq_policy *policy as
argument. You need to set the lower limit of the in-processor or
......@@ -198,7 +225,7 @@ setting when policy->policy is CPUFREQ_POLICY_PERFORMANCE, and a
powersaving-oriented setting when CPUFREQ_POLICY_POWERSAVE. Also check
the reference implementation in drivers/cpufreq/longrun.c
1.7 get_intermediate and target_intermediate
1.8 get_intermediate and target_intermediate
--------------------------------------------
Only for drivers with target_index() and CPUFREQ_ASYNC_NOTIFICATION unset.
......@@ -222,42 +249,36 @@ failures as core would send notifications for that.
As most cpufreq processors only allow for being set to a few specific
frequencies, a "frequency table" with some functions might assist in
some work of the processor driver. Such a "frequency table" consists
of an array of struct cpufreq_frequency_table entries, with any value in
"driver_data" you want to use, and the corresponding frequency in
"frequency". At the end of the table, you need to add a
cpufreq_frequency_table entry with frequency set to CPUFREQ_TABLE_END. And
if you want to skip one entry in the table, set the frequency to
CPUFREQ_ENTRY_INVALID. The entries don't need to be in ascending
order.
By calling cpufreq_table_validate_and_show(struct cpufreq_policy *policy,
struct cpufreq_frequency_table *table);
the cpuinfo.min_freq and cpuinfo.max_freq values are detected, and
policy->min and policy->max are set to the same values. This is
helpful for the per-CPU initialization stage.
int cpufreq_frequency_table_verify(struct cpufreq_policy *policy,
struct cpufreq_frequency_table *table);
assures that at least one valid frequency is within policy->min and
policy->max, and all other criteria are met. This is helpful for the
->verify call.
int cpufreq_frequency_table_target(struct cpufreq_policy *policy,
unsigned int target_freq,
unsigned int relation);
is the corresponding frequency table helper for the ->target
stage. Just pass the values to this function, and this function
returns the number of the frequency table entry which contains
the frequency the CPU shall be set to.
some work of the processor driver. Such a "frequency table" consists of
an array of struct cpufreq_frequency_table entries, with driver specific
values in "driver_data", the corresponding frequency in "frequency" and
flags set. At the end of the table, you need to add a
cpufreq_frequency_table entry with frequency set to CPUFREQ_TABLE_END.
And if you want to skip one entry in the table, set the frequency to
CPUFREQ_ENTRY_INVALID. The entries don't need to be in sorted in any
particular order, but if they are cpufreq core will do DVFS a bit
quickly for them as search for best match is faster.
By calling cpufreq_table_validate_and_show(), the cpuinfo.min_freq and
cpuinfo.max_freq values are detected, and policy->min and policy->max
are set to the same values. This is helpful for the per-CPU
initialization stage.
cpufreq_frequency_table_verify() assures that at least one valid
frequency is within policy->min and policy->max, and all other criteria
are met. This is helpful for the ->verify call.
cpufreq_frequency_table_target() is the corresponding frequency table
helper for the ->target stage. Just pass the values to this function,
and this function returns the of the frequency table entry which
contains the frequency the CPU shall be set to.
The following macros can be used as iterators over cpufreq_frequency_table:
cpufreq_for_each_entry(pos, table) - iterates over all entries of frequency
table.
cpufreq-for_each_valid_entry(pos, table) - iterates over all entries,
cpufreq_for_each_valid_entry(pos, table) - iterates over all entries,
excluding CPUFREQ_ENTRY_INVALID frequencies.
Use arguments "pos" - a cpufreq_frequency_table * as a loop cursor and
"table" - the cpufreq_frequency_table * you want to iterate over.
......
......@@ -34,10 +34,10 @@ cpufreq stats provides following statistics (explained in detail below).
- total_trans
- trans_table
All the statistics will be from the time the stats driver has been inserted
to the time when a read of a particular statistic is done. Obviously, stats
driver will not have any information about the frequency transitions before
the stats driver insertion.
All the statistics will be from the time the stats driver has been inserted
(or the time the stats were reset) to the time when a read of a particular
statistic is done. Obviously, stats driver will not have any information
about the frequency transitions before the stats driver insertion.
--------------------------------------------------------------------------------
<mysystem>:/sys/devices/system/cpu/cpu0/cpufreq/stats # ls -l
......@@ -110,25 +110,13 @@ Config Main Menu
CPU Frequency scaling --->
[*] CPU Frequency scaling
[*] CPU frequency translation statistics
[*] CPU frequency translation statistics details
"CPU Frequency scaling" (CONFIG_CPU_FREQ) should be enabled to configure
cpufreq-stats.
"CPU frequency translation statistics" (CONFIG_CPU_FREQ_STAT) provides the
basic statistics which includes time_in_state and total_trans.
statistics which includes time_in_state, total_trans and trans_table.
"CPU frequency translation statistics details" (CONFIG_CPU_FREQ_STAT_DETAILS)
provides fine grained cpufreq stats by trans_table. The reason for having a
separate config option for trans_table is:
- trans_table goes against the traditional /sysfs rule of one value per
interface. It provides a whole bunch of value in a 2 dimensional matrix
form.
Once these two options are enabled and your CPU supports cpufrequency, you
Once this option is enabled and your CPU supports cpufrequency, you
will be able to see the CPU frequency statistics in /sysfs.
......@@ -10,6 +10,8 @@
Dominik Brodowski <linux@brodo.de>
some additions and corrections by Nico Golde <nico@ngolde.de>
Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Viresh Kumar <viresh.kumar@linaro.org>
......@@ -28,32 +30,27 @@ Contents:
2.3 Userspace
2.4 Ondemand
2.5 Conservative
2.6 Schedutil
3. The Governor Interface in the CPUfreq Core
4. References
1. What Is A CPUFreq Governor?
==============================
Most cpufreq drivers (except the intel_pstate and longrun) or even most
cpu frequency scaling algorithms only offer the CPU to be set to one
frequency. In order to offer dynamic frequency scaling, the cpufreq
core must be able to tell these drivers of a "target frequency". So
these specific drivers will be transformed to offer a "->target/target_index"
call instead of the existing "->setpolicy" call. For "longrun", all
stays the same, though.
cpu frequency scaling algorithms only allow the CPU frequency to be set
to predefined fixed values. In order to offer dynamic frequency
scaling, the cpufreq core must be able to tell these drivers of a
"target frequency". So these specific drivers will be transformed to
offer a "->target/target_index/fast_switch()" call instead of the
"->setpolicy()" call. For set_policy drivers, all stays the same,
though.
How to decide what frequency within the CPUfreq policy should be used?
That's done using "cpufreq governors". Two are already in this patch
-- they're the already existing "powersave" and "performance" which
set the frequency statically to the lowest or highest frequency,
respectively. At least two more such governors will be ready for
addition in the near future, but likely many more as there are various
different theories and models about dynamic frequency scaling
around. Using such a generic interface as cpufreq offers to scaling
governors, these can be tested extensively, and the best one can be
selected for each specific use.
That's done using "cpufreq governors".
Basically, it's the following flow graph:
......@@ -71,7 +68,7 @@ CPU can be set to switch independently | CPU can only be set
/ the limits of policy->{min,max}
/ \
/ \
Using the ->setpolicy call, Using the ->target/target_index call,
Using the ->setpolicy call, Using the ->target/target_index/fast_switch call,
the limits and the the frequency closest
"policy" is set. to target_freq is set.
It is assured that it
......@@ -109,114 +106,159 @@ directory.
2.4 Ondemand
------------
The CPUfreq governor "ondemand" sets the CPU depending on the
current usage. To do this the CPU must have the capability to
switch the frequency very quickly. There are a number of sysfs file
accessible parameters:
sampling_rate: measured in uS (10^-6 seconds), this is how often you
want the kernel to look at the CPU usage and to make decisions on
what to do about the frequency. Typically this is set to values of
around '10000' or more. It's default value is (cmp. with users-guide.txt):
transition_latency * 1000
Be aware that transition latency is in ns and sampling_rate is in us, so you
get the same sysfs value by default.
Sampling rate should always get adjusted considering the transition latency
To set the sampling rate 750 times as high as the transition latency
in the bash (as said, 1000 is default), do:
echo `$(($(cat cpuinfo_transition_latency) * 750 / 1000)) \
>ondemand/sampling_rate
sampling_rate_min:
The sampling rate is limited by the HW transition latency:
transition_latency * 100
Or by kernel restrictions:
If CONFIG_NO_HZ_COMMON is set, the limit is 10ms fixed.
If CONFIG_NO_HZ_COMMON is not set or nohz=off boot parameter is used, the
limits depend on the CONFIG_HZ option:
HZ=1000: min=20000us (20ms)
HZ=250: min=80000us (80ms)
HZ=100: min=200000us (200ms)
The highest value of kernel and HW latency restrictions is shown and
used as the minimum sampling rate.
up_threshold: defines what the average CPU usage between the samplings
of 'sampling_rate' needs to be for the kernel to make a decision on
whether it should increase the frequency. For example when it is set
to its default value of '95' it means that between the checking
intervals the CPU needs to be on average more than 95% in use to then
decide that the CPU frequency needs to be increased.
ignore_nice_load: this parameter takes a value of '0' or '1'. When
set to '0' (its default), all processes are counted towards the
'cpu utilisation' value. When set to '1', the processes that are
run with a 'nice' value will not count (and thus be ignored) in the
overall usage calculation. This is useful if you are running a CPU
intensive calculation on your laptop that you do not care how long it
takes to complete as you can 'nice' it and prevent it from taking part
in the deciding process of whether to increase your CPU frequency.
sampling_down_factor: this parameter controls the rate at which the
kernel makes a decision on when to decrease the frequency while running
at top speed. When set to 1 (the default) decisions to reevaluate load
are made at the same interval regardless of current clock speed. But
when set to greater than 1 (e.g. 100) it acts as a multiplier for the
scheduling interval for reevaluating load when the CPU is at its top
speed due to high load. This improves performance by reducing the overhead
of load evaluation and helping the CPU stay at its top speed when truly
busy, rather than shifting back and forth in speed. This tunable has no
effect on behavior at lower speeds/lower CPU loads.
powersave_bias: this parameter takes a value between 0 to 1000. It
defines the percentage (times 10) value of the target frequency that
will be shaved off of the target. For example, when set to 100 -- 10%,
when ondemand governor would have targeted 1000 MHz, it will target
1000 MHz - (10% of 1000 MHz) = 900 MHz instead. This is set to 0
(disabled) by default.
When AMD frequency sensitivity powersave bias driver --
drivers/cpufreq/amd_freq_sensitivity.c is loaded, this parameter
defines the workload frequency sensitivity threshold in which a lower
frequency is chosen instead of ondemand governor's original target.
The frequency sensitivity is a hardware reported (on AMD Family 16h
Processors and above) value between 0 to 100% that tells software how
the performance of the workload running on a CPU will change when
frequency changes. A workload with sensitivity of 0% (memory/IO-bound)
will not perform any better on higher core frequency, whereas a
workload with sensitivity of 100% (CPU-bound) will perform better
higher the frequency. When the driver is loaded, this is set to 400
by default -- for CPUs running workloads with sensitivity value below
40%, a lower frequency is chosen. Unloading the driver or writing 0
will disable this feature.
The CPUfreq governor "ondemand" sets the CPU frequency depending on the
current system load. Load estimation is triggered by the scheduler
through the update_util_data->func hook; when triggered, cpufreq checks
the CPU-usage statistics over the last period and the governor sets the
CPU accordingly. The CPU must have the capability to switch the
frequency very quickly.
Sysfs files:
* sampling_rate:
Measured in uS (10^-6 seconds), this is how often you want the kernel
to look at the CPU usage and to make decisions on what to do about the
frequency. Typically this is set to values of around '10000' or more.
It's default value is (cmp. with users-guide.txt): transition_latency
* 1000. Be aware that transition latency is in ns and sampling_rate
is in us, so you get the same sysfs value by default. Sampling rate
should always get adjusted considering the transition latency to set
the sampling rate 750 times as high as the transition latency in the
bash (as said, 1000 is default), do:
$ echo `$(($(cat cpuinfo_transition_latency) * 750 / 1000)) > ondemand/sampling_rate
* sampling_rate_min:
The sampling rate is limited by the HW transition latency:
transition_latency * 100
Or by kernel restrictions:
- If CONFIG_NO_HZ_COMMON is set, the limit is 10ms fixed.
- If CONFIG_NO_HZ_COMMON is not set or nohz=off boot parameter is
used, the limits depend on the CONFIG_HZ option:
HZ=1000: min=20000us (20ms)
HZ=250: min=80000us (80ms)
HZ=100: min=200000us (200ms)
The highest value of kernel and HW latency restrictions is shown and
used as the minimum sampling rate.
* up_threshold:
This defines what the average CPU usage between the samplings of
'sampling_rate' needs to be for the kernel to make a decision on
whether it should increase the frequency. For example when it is set
to its default value of '95' it means that between the checking
intervals the CPU needs to be on average more than 95% in use to then
decide that the CPU frequency needs to be increased.
* ignore_nice_load:
This parameter takes a value of '0' or '1'. When set to '0' (its
default), all processes are counted towards the 'cpu utilisation'
value. When set to '1', the processes that are run with a 'nice'
value will not count (and thus be ignored) in the overall usage
calculation. This is useful if you are running a CPU intensive
calculation on your laptop that you do not care how long it takes to
complete as you can 'nice' it and prevent it from taking part in the
deciding process of whether to increase your CPU frequency.
* sampling_down_factor:
This parameter controls the rate at which the kernel makes a decision
on when to decrease the frequency while running at top speed. When set
to 1 (the default) decisions to reevaluate load are made at the same
interval regardless of current clock speed. But when set to greater
than 1 (e.g. 100) it acts as a multiplier for the scheduling interval
for reevaluating load when the CPU is at its top speed due to high
load. This improves performance by reducing the overhead of load
evaluation and helping the CPU stay at its top speed when truly busy,
rather than shifting back and forth in speed. This tunable has no
effect on behavior at lower speeds/lower CPU loads.
* powersave_bias:
This parameter takes a value between 0 to 1000. It defines the
percentage (times 10) value of the target frequency that will be
shaved off of the target. For example, when set to 100 -- 10%, when
ondemand governor would have targeted 1000 MHz, it will target
1000 MHz - (10% of 1000 MHz) = 900 MHz instead. This is set to 0
(disabled) by default.
When AMD frequency sensitivity powersave bias driver --
drivers/cpufreq/amd_freq_sensitivity.c is loaded, this parameter
defines the workload frequency sensitivity threshold in which a lower
frequency is chosen instead of ondemand governor's original target.
The frequency sensitivity is a hardware reported (on AMD Family 16h
Processors and above) value between 0 to 100% that tells software how
the performance of the workload running on a CPU will change when
frequency changes. A workload with sensitivity of 0% (memory/IO-bound)
will not perform any better on higher core frequency, whereas a
workload with sensitivity of 100% (CPU-bound) will perform better
higher the frequency. When the driver is loaded, this is set to 400 by
default -- for CPUs running workloads with sensitivity value below
40%, a lower frequency is chosen. Unloading the driver or writing 0
will disable this feature.
2.5 Conservative
----------------
The CPUfreq governor "conservative", much like the "ondemand"
governor, sets the CPU depending on the current usage. It differs in
behaviour in that it gracefully increases and decreases the CPU speed
rather than jumping to max speed the moment there is any load on the
CPU. This behaviour more suitable in a battery powered environment.
The governor is tweaked in the same manner as the "ondemand" governor
through sysfs with the addition of: