Re: [PATCH] cpufreq, intel_pstate, set max_sysfs_pct and min_sysfs_pct on governor switch

[Prarit Bhargava <prarit@redhat.com>]

On 10/06/2015 07:06 PM, Rafael J. Wysocki wrote:
> On Wednesday, October 07, 2015 12:43:55 AM Rafael J. Wysocki wrote:
>> On Tuesday, October 06, 2015 05:49:07 PM Prarit Bhargava wrote:
>>> Intel CPUs will not enter higher p-states when after switching from the
>>> performance governor to the powersave governor, until
>>> /sys/devices/system/cpu/intel_pstate/min_perf_pct is set to a low value.
>>> This differs from previous behaviour in which a switch to the powersave
>>> governor would result in a low default value for min_perf_pct.
>>>
>>> The behavior of the powersave governor changed after commit a04759924e25
>>> ("[cpufreq] intel_pstate: honor user space min_perf_pct override on
>>> resume").  The commit introduced tracking of performance percentage
>>> changes via sysfs in order to restore userspace changes during
>>> suspend/resume.  The problem occurs because the global values of the newly
>>> introduced max_sysfs_pct and min_sysfs_pct are not reset on a governor
>>> change and this causes the new governor to inherit the previous governor's
>>> settings.
>>>
>>> This patch sets max_sysfs_pct to 100 and min_sysfs_pct to 0 on a governor
>>> change which fixes the problem with governor switching.  These changes
>>> also make the initial calculations for max_perf_pct and min_perf_pct
>>> slightly simpler.
>>>
>>> Before patch:
>>> [root@intel-skylake-y-01 power]# cpupower frequency-set -g performance
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/min_perf_pct
>>> 100
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/max_perf_pct
>>> 100
>>> [root@intel-skylake-y-01 power]# cpupower frequency-set -g powersave
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/min_perf_pct
>>> 100
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/max_perf_pct
>>> 100
>>>
>>> After patch:
>>> [root@intel-skylake-y-01 power]# cpupower frequency-set -g performance
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/min_perf_pct
>>> 100
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/max_perf_pct
>>> 100
>>> [root@intel-skylake-y-01 power]# cpupower frequency-set -g powersave
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/min_perf_pct
>>> 14
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/max_perf_pct
>>> 100
>>>
>>> Also note that I have tested suspend/resume (using CONFIG_PM_DEBUG):
>>> [root@intel-skylake-y-01 power]# echo 50 > /sys/devices/system/cpu/intel_pstate/min_perf_pct
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/*_perf_pct
>>> 100
>>> 50
>>> [root@intel-skylake-y-01 power]# echo devices > /sys/power/pm_test
>>> [root@intel-skylake-y-01 power]# echo platform > /sys/power/disk
>>> [root@intel-skylake-y-01 power]# echo disk > /sys/power/state
>>> [root@intel-skylake-y-01 power]# cat /sys/devices/system/cpu/intel_pstate/*_perf_pct
>>> 100
>>> 50
>>>
>>> Fixes: a04759924e25 ("[cpufreq] intel_pstate: honor user space min_perf_pct override on resume")
>>> Cc: Kristen Carlson Accardi <kristen@linux.intel.com>
>>> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
>>> Cc: Viresh Kumar <viresh.kumar@linaro.org>
>>> Cc: linux-pm@vger.kernel.org
>>> Signed-off-by: Prarit Bhargava <prarit@redhat.com>
>>> ---
>>>  drivers/cpufreq/intel_pstate.c |    7 +++++--
>>>  1 file changed, 5 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
>>> index 3af9dd7..bb24458 100644
>>> --- a/drivers/cpufreq/intel_pstate.c
>>> +++ b/drivers/cpufreq/intel_pstate.c
>>> @@ -986,6 +986,9 @@ static int intel_pstate_set_policy(struct cpufreq_policy *policy)
>>>  	if (!policy->cpuinfo.max_freq)
>>>  		return -ENODEV;
>>>  
>>> +	limits.min_sysfs_pct = 0;
>>> +	limits.max_sysfs_pct = 100;
>>> +
>>>  	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE &&
>>>  	    policy->max >= policy->cpuinfo.max_freq) {
>>>  		limits.min_policy_pct = 100;
>>> @@ -1004,9 +1007,9 @@ static int intel_pstate_set_policy(struct cpufreq_policy *policy)
>>>  	limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
>>>  
>>>  	/* Normalize user input to [min_policy_pct, max_policy_pct] */
>>> -	limits.min_perf_pct = max(limits.min_policy_pct, limits.min_sysfs_pct);
>>> +	limits.min_perf_pct = limits.min_policy_pct;
>>>  	limits.min_perf_pct = min(limits.max_policy_pct, limits.min_perf_pct);
>>> -	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
>>> +	limits.max_perf_pct = limits.max_sysfs_pct;
> 
> On a second thought, isn't that always 100?  If so, doesn't it basically discard
> limits.max_policy_pct?

Looking at it, yes.  And that's definitely an unintended consequence of this
patch :).  I'll take a closer look.  I thought it should be permissible to set a
range of (min_perf_pct, max_perf_pct) while changing p-states and I thought the
purpose of max_perf_pct was to set the higher percentage limit.

P.