None of the cpufreq governors currently in the tree will ever fail
an invocation of the ->governor() callback with the event argument
equal to CPUFREQ_GOV_STOP (unless invoked with incorrect arguments
which doesn't matter anyway) and it is rather difficult to imagine
a valid reason for such a failure.
Accordingly, rearrange the code in the core to make it clear that
this call never fails.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
None of the cpufreq governors currently in the tree will ever fail
an invocation of the ->governor() callback with the event argument
equal to CPUFREQ_GOV_POLICY_EXIT (unless invoked with incorrect
arguments which doesn't matter anyway) and it wouldn't really
make sense to fail it, because the caller won't be able to handle
that failure in a meaningful way.
Accordingly, rearrange the code in the core to make it clear that
this call never fails.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
One of the if () statements in intel_pstate_set_policy() causes
another if () to be evaluated if the condition is true and it
doesn't do anything else, so merge the two if () statements into
one.
No functional changes.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
The comments and the core_busy variable name in
get_target_pstate_use_performance() are totally confusing,
so modify them to reflect what's going on.
The results of the computations should be the same as before.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Notice that get_avg_pstate() can use sample.core_avg_perf instead of
carrying the same division again, so make it do that.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The core_pct_busy field of struct sample actually contains the
average performace during the last sampling period (in percent)
and not the utilization of the core as suggested by its name
which is confusing.
For this reason, change the name of that field to core_avg_perf
and rename the function that computes its value accordingly.
Also notice that storing this value as percentage requires a costly
integer multiplication to be carried out in a hot path, so instead
store it as an "extended fixed point" value with more fraction bits
and update the code using it accordingly (it is better to change the
name of the field along with its meaning in one go than to make those
two changes separately, as that would likely lead to more
confusion).
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently, in intel_pstate_clear_update_util_hook(), after
clearing the utilization update hook, we leverage
synchronize_sched() to deal with synchronization, which
is a little bit time-costly because synchronize_sched()
has to wait for all the CPUs to go through a grace period.
Actually, the synchronize_sched() is not necessary if the utilization
update hook has not been set for the given CPU yet, so make the driver
check if that's the case and avoid the synchronize_sched() call then.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=116371
Tested-by: Tian Ye <yex.tian@intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
[ rjw : Rebase ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
CPU_FREQ_GOV_SCHEDUTIL gained a dependency on SMP, so now we
get a warning if it gets selected by CPU_FREQ_DEFAULT_GOV_SCHEDUTIL
without SMP:
warning: (CPU_FREQ_DEFAULT_GOV_SCHEDUTIL) selects CPU_FREQ_GOV_SCHEDUTIL which has unmet direct dependencies (CPU_FREQ && SMP)
This adds another dependency to avoid the problem.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: bf7cdff194 (cpufreq: schedutil: Make it depend on CONFIG_SMP)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When global and local pstate are equal in a powernv_target_index() call,
we don't queue a timer. But we may have timer already queued for future.
This could cause the timer to fire one additional time for no use.
Signed-off-by: Akshay Adiga <akshay.adiga@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Fix a WARN_ON caused by smp_call_function_any() when irq is disabled,
because of changes made in the patch ('cpufreq: powernv: Ramp-down
global pstate slower than local-pstate')
https://patchwork.ozlabs.org/patch/612058/
WARNING: CPU: 0 PID: 4 at kernel/smp.c:291
smp_call_function_single+0x170/0x180
Call Trace:
[c0000007f648f9f0] [c0000007f648fa90] 0xc0000007f648fa90 (unreliable)
[c0000007f648fa30] [c0000000001430e0] smp_call_function_any+0x170/0x1c0
[c0000007f648fa90] [c0000000007b4b00]
powernv_cpufreq_target_index+0xe0/0x250
[c0000007f648fb00] [c0000000007ac9dc]
__cpufreq_driver_target+0x20c/0x3d0
[c0000007f648fbc0] [c0000000007b1b4c] od_dbs_timer+0xcc/0x260
[c0000007f648fc10] [c0000000007b3024] dbs_work_handler+0x54/0xa0
[c0000007f648fc50] [c0000000000c49a8] process_one_work+0x1d8/0x590
[c0000007f648fce0] [c0000000000c4e08] worker_thread+0xa8/0x660
[c0000007f648fd80] [c0000000000cca88] kthread+0x108/0x130
[c0000007f648fe30] [c0000000000095e8] ret_from_kernel_thread+0x5c/0x74
- Calling smp_call_function_any() with interrupt disabled (through
spin_lock_irqsave) could cause a deadlock, as smp_call_function_any()
relies on the IPI to complete. This is detected in the
smp_call_function_any() call and hence the WARN_ON.
- As the spinlock (gpstates->lock) is only used to synchronize access of
global_pstate_info between timer irq handler and target_index calls. And
the timer irq handler just try_locks() hence it would not cause a
deadlock. Hence could do without making spinlocks irq safe.
- As the smp_call_function_any() is a blocking call and does not access
global_pstates_info, it could reduce the critcal section by moving
smp_call_function_any() after giving up the lock.
Reported-by: Abdul Haleem <abdhalee@linux.vnet.linux.com>
Signed-off-by: Akshay Adiga <akshay.adiga@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
intel_pstate_get() contains a local variable that's initialized but
never used and it can be written in fewer lines of code, so clean
it up.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Make the schedutil cpufreq governor depend on CONFIG_SMP, because
the scheduler-provided utilization numbers used by it are only
available with CONFIG_SMP set.
Fixes: 9bdcb44e39 (cpufreq: schedutil: New governor based on scheduler utilization data)
Reported-by: Steve Muckle <steve.muckle@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
* pm-cpufreq-fixes:
intel_pstate: Fix intel_pstate_get()
cpufreq: intel_pstate: Fix HWP on boot CPU after system resume
cpufreq: st: enable selective initialization based on the platform
cpufreq: intel_pstate: Fix processing for turbo activation ratio
As reported in KBZ 69821:
"With CONFIG_HZ_PERIODIC=y cpu stays at the lowest frequcency 800MHz
even if usage goes to 100%, frequency does not scale up, the governor
in use is ondemand. Neither works conservative. Performance and
userspace governors work as expected.
With CONFIG_NO_HZ_IDLE or CONFIG_NO_HZ_FULL cpu scales up with ondemand
as expected."
Analysis carried out by Chen Yu leads to the conclusion that the
observed issue is due to idle_time in dbs_update() representing a
negative number in which case the function will return 0 as the load
(unless load is greater than 0 for another CPU sharing the policy),
although that need not be the right choice.
Indeed, idle_time representing a negative number means that during
the last sampling interval the CPU was almost 100% busy on the rough
average, so 100 should be returned as the load in that case.
Modify the code accordingly and rearrange it to clarify the handling
of all of the special cases in it. While at it, also avoid returning
zero as the load if time_elapsed is 0 (it doesn't really make sense
to return 0 then).
Link: https://bugzilla.kernel.org/show_bug.cgi?id=69821
Tested-by: Chen Yu <yu.c.chen@intel.com>
Tested-by: Timo Valtoaho <timo.valtoaho@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
When HWP (hardware P states) feature is active, the ACPI _PSS and _PPC
is not used. So ignore processing for _PPC limits.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently when performing random CPU hot-plugs and suspend-to-ram(S2R)
on systems using arm_big_little cpufreq driver, we get warnings similar
to something like below:
cpu cpu1: _opp_add: duplicate OPPs detected. Existing: freq: 600000000,
volt: 800000, enabled: 1. New: freq: 600000000, volt: 800000, enabled: 1
This is mainly because the OPPs for the shared cpus are not set. We can
just use dev_pm_opp_of_cpumask_add_table in case the OPPs are obtained
from DT(arm_big_little_dt.c) or use dev_pm_opp_set_sharing_cpus if the
OPPs are obtained by other means like firmware(e.g. scpi-cpufreq.c)
Also now that the generic dev_pm_opp{,_of}_cpumask_remove_table can
handle removal of opp table and entries for all associated CPUs, we can
re-use dev_pm_opp{,_of}_cpumask_remove_table as free_opp_table in
cpufreq_arm_bL_ops.
This patch makes necessary changes to reuse the generic OPP functions for
{init,free}_opp_table and thereby eliminating the warnings.
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Functions dev_pm_opp_of_{cpumask_,}remove_table removes/frees all the
static OPP entries associated with the device and/or all cpus(in case
of cpumask) that are created from DT.
However the OPP entries are populated reading from the firmware or some
different method using dev_pm_opp_add are marked dynamic and can't be
removed using above functions.
This patch adds non DT/OF versions of dev_pm_opp_{cpumask_,}remove_table
to support the above mentioned usecase.
This is in preparation to make use of the same in scpi-cpufreq.c
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Add tango4 compatible string to the list.
Signed-off-by: Marc Gonzalez <marc_gonzalez@sigmadesigns.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The new use of dev_pm_opp_set_sharing_cpus resulted in a harmless compiler
warning with CONFIG_CPUMASK_OFFSTACK=y:
drivers/cpufreq/mvebu-cpufreq.c: In function 'armada_xp_pmsu_cpufreq_init':
include/linux/cpumask.h:550:25: error: passing argument 2 of 'dev_pm_opp_set_sharing_cpus' discards 'const' qualifier from pointer target type [-Werror=discarded-qualifiers]
The problem here is that cpumask_var_t gets passed by reference, but
by declaring a 'const cpumask_var_t' argument, only the pointer is
constant, not the actual mask. This is harmless because the function
does not actually modify the mask.
This patch changes the function prototypes for all of the related functions
to pass a 'struct cpumask *' instead of 'cpumask_var_t', matching what
most other such functions do in the kernel. This lets us mark all the
other similar functions as taking a 'const' mask where possible,
and it avoids the warning without any change in object code.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 947bd567f7 (mvebu: Use dev_pm_opp_set_sharing_cpus() to mark OPP tables as shared)
Acked-by: Pavel Machek <pavel@ucw.cz>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently, the userspace governor only updates frequency on GOV_LIMITS
if policy->cur falls outside policy->{min/max}. However, it is also
necessary to update current frequency on GOV_LIMITS to match the user
requested value if it can be achieved within the new policy->{max/min}.
This was previously the behaviour in the governor until commit d1922f0
("cpufreq: Simplify userspace governor") which incorrectly assumed that
policy->cur == user requested frequency via scaling_setspeed. This won't
be true if the user requested frequency falls outside policy->{min/max}.
Ex: a temporary thermal cap throttled the user requested frequency.
Fix this by storing the user requested frequency in a seperate variable.
The governor will then try to achieve this request on every GOV_LIMITS
change.
Fixes: d1922f0256 (cpufreq: Simplify userspace governor)
Signed-off-by: Sai Gurrappadi <sgurrappadi@nvidia.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
After commit 8fa520af50 "intel_pstate: Remove freq calculation from
intel_pstate_calc_busy()" intel_pstate_get() calls get_avg_frequency()
to compute the average frequency, which is problematic for two reasons.
First, intel_pstate_get() may be invoked before the driver reads the
CPU feedback registers for the first time and if that happens,
get_avg_frequency() will attempt to divide by zero.
Second, the get_avg_frequency() call in intel_pstate_get() is racy
with respect to intel_pstate_sample() and it may end up returning
completely meaningless values for this reason.
Moreover, after commit 7349ec0470 "intel_pstate: Move
intel_pstate_calc_busy() into get_target_pstate_use_performance()"
sample.core_pct_busy is never computed on Atom, but it is used in
intel_pstate_adjust_busy_pstate() in that case too.
To address those problems notice that if sample.core_pct_busy
was used in the average frequency computation carried out by
get_avg_frequency(), both the divide by zero problem and the
race with respect to intel_pstate_sample() would be avoided.
Accordingly, move the invocation of intel_pstate_calc_busy() from
get_target_pstate_use_performance() to intel_pstate_update_util(),
which also will take care of the uninitialized sample.core_pct_busy
on Atom, and modify get_avg_frequency() to use sample.core_pct_busy
as per the above.
Reported-by: kernel test robot <ying.huang@linux.intel.com>
Link: http://marc.info/?l=linux-kernel&m=146226437623173&w=4
Fixes: 8fa520af50 "intel_pstate: Remove freq calculation from intel_pstate_calc_busy()"
Fixes: 7349ec0470 "intel_pstate: Move intel_pstate_calc_busy() into get_target_pstate_use_performance()"
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Commit 41cfd64cf4 "Update frequencies of policy->cpus only from
->set_policy()" changed the way the intel_pstate driver's ->set_policy
callback updates the HWP (hardware-managed P-states) settings.
A side effect of it is that if those settings are modified on the
boot CPU during system suspend and wakeup, they will never be
restored during subsequent system resume.
To address this problem, allow cpufreq drivers that don't provide
->target or ->target_index callbacks to use ->suspend and ->resume
callbacks and add a ->resume callback to intel_pstate to restore
the HWP settings on the CPUs that belong to the given policy.
Fixes: 41cfd64cf4 "Update frequencies of policy->cpus only from ->set_policy()"
Tested-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
The sti-cpufreq does unconditional registration of the cpufreq-dt driver
which causes issue on an multi-platform build. For example, on Vexpress
TC2 platform, we get the following error on boot:
cpu cpu0: OPP-v2 not supported
cpu cpu0: Not doing voltage scaling
cpu: dev_pm_opp_of_cpumask_add_table: couldn't find opp table
for cpu:0, -19
cpu cpu0: dev_pm_opp_get_max_volt_latency: Invalid regulator (-6)
...
arm_big_little: bL_cpufreq_register: Failed registering platform driver:
vexpress-spc, err: -17
The actual driver fails to initialise as cpufreq-dt is probed
successfully, which is incorrect. This issue can happen to any platform
not using cpufreq-dt in a multi-platform build.
This patch adds a check to do selective initialization of the driver.
Fixes: ab0ea257fc (cpufreq: st: Provide runtime initialised driver for ST's platforms)
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Lee Jones <lee.jones@linaro.org>
Cc: 4.5+ <stable@vger.kernel.org> # 4.5+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Move cpufreq bits for mvebu into drivers/cpufreq/ directory, that's
where they really belong to.
Compiled tested only.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There are no more users of platform-data for cpufreq-dt driver, get rid
of it.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
That will allow us to avoid using cpufreq-dt platform data.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Existing platforms, which do not support operating-points-v2, can
explicitly tell the opp core that some of the CPUs share opp tables,
with help of dev_pm_opp_set_sharing_cpus().
For such platforms, explicitly ask the opp core to provide list of CPUs
sharing the opp table with current cpu device, before falling back to
platform data.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
OPP core allows a platform to mark OPP table as shared, when the
platform isn't using operating-points-v2 bindings.
And, so there should be a non DT way of finding out if the OPP table is
shared or not.
This patch adds dev_pm_opp_get_sharing_cpus(), which first tries to get
OPP sharing information from the opp-table (in case it is already marked
as shared), otherwise it uses the existing DT way of finding sharing
information.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
dev_pm_opp_set_sharing_cpus() isn't supposed to update the cpumask
passed as its parameter, and so it should always have been marked
'const'.
Do it now.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Some of the routines have used -ENOSYS for the cases where the
functionality isn't implemented in the kernel. But ENOSYS is supposed to
be used only for syscalls.
Replace that with -ENOTSUPP, which specifically means that the operation
isn't supported.
While at it, replace exiting -EINVAL errors for similar cases to
-ENOTSUPP.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The name of the prev_cpu_wall field in struct cpu_dbs_info is
confusing, because it doesn't represent wall time, but the previous
update time as returned by get_cpu_idle_time() (that may be the
current value of jiffies_64 in some cases, for example).
Moreover, the names of some related variables in dbs_update() take
that confusion further.
Rename all of those things to make their names reflect the purpose
more accurately. While at it, drop unnecessary parens from one of
the updated expressions.
No functional changes.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Chen Yu <yu.c.chen@intel.com>
For platforms which are controlled via remove node manager, enable _PPC by
default. These platforms are mostly categorized as enterprise server or
performance servers. These platforms needs to go through some
certifications tests, which tests control via _PPC.
The relative risk of enabling by default is low as this is is less likely
that these systems have broken _PSS table.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When policy->max is changed via _PPC or sysfs and is more than the max non
turbo frequency, it does not really change resulting performance in some
processors. When policy->max results in a P-State ratio more than the
turbo activation ratio, then processor can choose any P-State up to max
turbo. So the user or _PPC setting has no value, but this can cause
undesirable side effects like:
- Showing reduced max percentage in Intel P-State sysfs
- It can cause reduced max performance under certain boundary conditions:
The requested max scaling frequency either via _PPC or via cpufreq-sysfs,
will be converted into a fixed floating point max percent scale. In
majority of the cases this will result in correct max. But not 100% of the
time. If the _PPC is requested at a point where the calculation lead to a
lower max, this can result in a lower P-State then expected and it will
impact performance.
Example of this condition using a Broadwell laptop with config TDP.
ACPI _PSS table from a Broadwell laptop
2301000 2300000 2200000 2000000 1900000 1800000 1700000 1500000 1400000
1300000 1100000 1000000 900000 800000 600000 500000
The actual results by disabling config TDP so that we can get what is
requested on or below 2300000Khz.
scaling_max_freq Max Requested P-State Resultant scaling
max
---------------------------------------- ----------------------
2400000 18 2900000 (max
turbo)
2300000 17 2300000 (max
physical non turbo)
2200000 15 2100000
2100000 15 2100000
2000000 13 1900000
1900000 13 1900000
1800000 12 1800000
1700000 11 1700000
1600000 10 1600000
1500000 f 1500000
1400000 e 1400000
1300000 d 1300000
1200000 c 1200000
1100000 a 1000000
1000000 a 1000000
900000 9 900000
800000 8 800000
700000 7 700000
600000 6 600000
500000 5 500000
------------------------------------------------------------------
Now set the config TDP level 1 ratio as 0x0b (equivalent to 1100000KHz)
in BIOS (not every system will let you adjust this).
The turbo activation ratio will be set to one less than that, which will
be 0x0a (So any request above 1000000KHz should result in turbo region
assuming no thermal limits).
Here _PPC will request max to 1100000KHz (which basically should still
result in turbo as this is more than the turbo activation ratio up to
max allowable turbo frequency), but actual calculation resulted in a max
ceiling P-State which is 0x0a. So under any load condition, this driver
will not request turbo P-States. This will be a huge performance hit.
When config TDP feature is ON, if the _PPC points to a frequency above
turbo activation ratio, the performance can still reach max turbo. In this
case we don't need to treat this as the reduced frequency in set_policy
callback.
In this change when config TDP is active (by checking if the physical max
non turbo ratio is more than the current max non turbo ratio), any request
above current max non turbo is treated as full performance.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
[ rjw : Minor cleanups ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Use ACPI _PPC notification to limit max P state driver will request.
ACPI _PPC change notification is sent by BIOS to limit max P state
in several cases:
- Reduce impact of platform thermal condition
- When Config TDP feature is used, a changed _PPC is sent to
follow TDP change
- Remote node managers in server want to control platform power
via baseboard management controller (BMC)
This change registers with ACPI processor performance lib so that
_PPC changes are notified to cpufreq core, which in turns will
result in call to .setpolicy() callback. Also the way _PSS
table identifies a turbo frequency is not compatible to max turbo
frequency in intel_pstate, so the very first entry in _PSS needs
to be adjusted.
This feature can be turned on by using kernel parameters:
intel_pstate=support_acpi_ppc
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
[ rjw: Minor cleanups ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The frequency transition latency from pmin to pmax is observed to be in
few millisecond granurality. And it usually happens to take a performance
penalty during sudden frequency rampup requests.
This patch set solves this problem by using an entity called "global
pstates". The global pstate is a Chip-level entity, so the global entitiy
(Voltage) is managed across the cores. The local pstate is a Core-level
entity, so the local entity (frequency) is managed across threads.
This patch brings down global pstate at a slower rate than the local
pstate. Hence by holding global pstates higher than local pstate makes
the subsequent rampups faster.
A per policy structure is maintained to keep track of the global and
local pstate changes. The global pstate is brought down using a parabolic
equation. The ramp down time to pmin is set to ~5 seconds. To make sure
that the global pstates are dropped at regular interval , a timer is
queued for every 2 seconds during ramp-down phase, which eventually brings
the pstate down to local pstate.
Iozone results show fairly consistent performance boost.
YCSB on redis shows improved Max latencies in most cases.
Iozone write/rewite test were made with filesizes 200704Kb and 401408Kb
with different record sizes . The following table shows IOoperations/sec
with and without patch.
Iozone Results ( in op/sec) ( mean over 3 iterations )
---------------------------------------------------------------------
file size- with without %
recordsize-IOtype patch patch change
----------------------------------------------------------------------
200704-1-SeqWrite 1616532 1615425 0.06
200704-1-Rewrite 2423195 2303130 5.21
200704-2-SeqWrite 1628577 1602620 1.61
200704-2-Rewrite 2428264 2312154 5.02
200704-4-SeqWrite 1617605 1617182 0.02
200704-4-Rewrite 2430524 2351238 3.37
200704-8-SeqWrite 1629478 1600436 1.81
200704-8-Rewrite 2415308 2298136 5.09
200704-16-SeqWrite 1619632 1618250 0.08
200704-16-Rewrite 2396650 2352591 1.87
200704-32-SeqWrite 1632544 1598083 2.15
200704-32-Rewrite 2425119 2329743 4.09
200704-64-SeqWrite 1617812 1617235 0.03
200704-64-Rewrite 2402021 2321080 3.48
200704-128-SeqWrite 1631998 1600256 1.98
200704-128-Rewrite 2422389 2304954 5.09
200704-256 SeqWrite 1617065 1616962 0.00
200704-256-Rewrite 2432539 2301980 5.67
200704-512-SeqWrite 1632599 1598656 2.12
200704-512-Rewrite 2429270 2323676 4.54
200704-1024-SeqWrite 1618758 1616156 0.16
200704-1024-Rewrite 2431631 2315889 4.99
401408-1-SeqWrite 1631479 1608132 1.45
401408-1-Rewrite 2501550 2459409 1.71
401408-2-SeqWrite 1617095 1626069 -0.55
401408-2-Rewrite 2507557 2443621 2.61
401408-4-SeqWrite 1629601 1611869 1.10
401408-4-Rewrite 2505909 2462098 1.77
401408-8-SeqWrite 1617110 1626968 -0.60
401408-8-Rewrite 2512244 2456827 2.25
401408-16-SeqWrite 1632609 1609603 1.42
401408-16-Rewrite 2500792 2451405 2.01
401408-32-SeqWrite 1619294 1628167 -0.54
401408-32-Rewrite 2510115 2451292 2.39
401408-64-SeqWrite 1632709 1603746 1.80
401408-64-Rewrite 2506692 2433186 3.02
401408-128-SeqWrite 1619284 1627461 -0.50
401408-128-Rewrite 2518698 2453361 2.66
401408-256-SeqWrite 1634022 1610681 1.44
401408-256-Rewrite 2509987 2446328 2.60
401408-512-SeqWrite 1617524 1628016 -0.64
401408-512-Rewrite 2504409 2442899 2.51
401408-1024-SeqWrite 1629812 1611566 1.13
401408-1024-Rewrite 2507620 2442968 2.64
Tested with YCSB workload (50% update + 50% read) over redis for 1 million
records and 1 million operation. Each test was carried out with target
operations per second and persistence disabled.
Max-latency (in us)( mean over 5 iterations )
---------------------------------------------------------------
op/s Operation with patch without patch %change
---------------------------------------------------------------
15000 Read 61480.6 50261.4 22.32
15000 cleanup 215.2 293.6 -26.70
15000 update 25666.2 25163.8 2.00
25000 Read 32626.2 89525.4 -63.56
25000 cleanup 292.2 263.0 11.10
25000 update 32293.4 90255.0 -64.22
35000 Read 34783.0 33119.0 5.02
35000 cleanup 321.2 395.8 -18.8
35000 update 36047.0 38747.8 -6.97
40000 Read 38562.2 42357.4 -8.96
40000 cleanup 371.8 384.6 -3.33
40000 update 27861.4 41547.8 -32.94
45000 Read 42271.0 88120.6 -52.03
45000 cleanup 263.6 383.0 -31.17
45000 update 29755.8 81359.0 -63.43
(test without target op/s)
47659 Read 83061.4 136440.6 -39.12
47659 cleanup 195.8 193.8 1.03
47659 update 73429.4 124971.8 -41.24
Signed-off-by: Akshay Adiga <akshay.adiga@linux.vnet.ibm.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
commit 1b0289848d ("cpufreq: powernv: Add sysfs attributes to show
throttle stats") used policy->driver_data as a flag for one-time creation
of throttle sysfs files. Instead of this use 'kernfs_find_and_get()' to
check if the attribute already exists. This is required as
policy->driver_data is used for other purposes in the later patch.
Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
Signed-off-by: Akshay Adiga <akshay.adiga@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The IS_ENABLED() macro checks if a Kconfig symbol has been enabled either
built-in or as a module, use that macro instead of open coding the same.
Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When the config TDP level is not nominal (level = 0), the MSR values for
reading level 1 and level 2 ratios contain power in low 14 bits and actual
ratio bits are at bits [23:16]. The current processing for level 1 and
level 2 is wrong as there is no shift done to get actual ratio.
Fixes: 6a35fc2d6c (cpufreq: intel_pstate: get P1 from TAR when available)
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: 4.4+ <stable@vger.kernel.org> # 4.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The way cpufreq_governor_start() initializes j_cdbs->prev_load is
questionable.
First off, j_cdbs->prev_cpu_wall used as a denominator in the
computation may be zero. The case this happens is when
get_cpu_idle_time_us() returns -1 and get_cpu_idle_time_jiffy()
used to return that number is called exactly at the jiffies_64
wrap time. It is rather hard to trigger that error, but it is not
impossible and it will just crash the kernel then.
Second, j_cdbs->prev_load is computed as the average load during
the entire time since the system started and it may not reflect the
load in the previous sampling period (as it is expected to).
That doesn't play well with the way dbs_update() uses that value.
Namely, if the update time delta (wall_time) happens do be greater
than twice the sampling rate on the first invocation of it, the
initial value of j_cdbs->prev_load (which may be completely off) will
be returned to the caller as the current load (unless it is equal to
zero and unless another CPU sharing the same policy object has a
greater load value).
For this reason, notice that the prev_load field of struct cpu_dbs_info
is only used by dbs_update() and only in that one place, so if
cpufreq_governor_start() is modified to always initialize it to 0,
it will make dbs_update() always compute the actual load first time
it checks the update time delta against the doubled sampling rate
(after initialization) and there won't be any side effects of it.
Consequently, modify cpufreq_governor_start() as described.
Fixes: 18b46abd00 (cpufreq: governor: Be friendly towards latency-sensitive bursty workloads)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
The cpufreq-dt-platdev driver supports creation of cpufreq-dt platform
device now, reuse that and remove similar code from platform code.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The cpufreq-dt-platdev driver supports creation of cpufreq-dt platform
device now, reuse that and remove similar code from platform code.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The cpufreq-dt-platdev driver supports creation of cpufreq-dt platform
device now, reuse that and remove similar code from platform code.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The cpufreq-dt-platdev driver supports creation of cpufreq-dt platform
device now, reuse that and remove similar code from platform code.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This patch add rockchip's compatible string to the compat list and
remove similar code from platform code for supporting generic platdev
driver.
Signed-off-by: Finley Xiao <finley.xiao@rock-chips.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Heiko Stuebner <heiko@sntech.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The cpufreq-dt-platdev driver supports creation of cpufreq-dt platform
device now, reuse that and remove similar code from platform code.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
