Re: [PATCH V4 2/2] cpufreq: intel_pstate: Implement QoS supported freq constraints

[Viresh Kumar]

On 08-08-19, 23:35, Doug Smythies wrote:
> O.K. While I understand the explanations, I still struggle with
> this scenario:
>  
> doug@s15:~/temp$ cat /sys/devices/system/cpu/intel_pstate/max_perf_pct
> 50    <<< Note: 50% = 1.9 GHz in my system)
> doug@s15:~/temp$ grep . /sys/devices/system/cpu/cpufreq/policy*/scaling_max_freq
> /sys/devices/system/cpu/cpufreq/policy0/scaling_max_freq:1900000
> /sys/devices/system/cpu/cpufreq/policy1/scaling_max_freq:1900000
> /sys/devices/system/cpu/cpufreq/policy2/scaling_max_freq:1900000
> /sys/devices/system/cpu/cpufreq/policy3/scaling_max_freq:1900000
> /sys/devices/system/cpu/cpufreq/policy4/scaling_max_freq:1900000
> /sys/devices/system/cpu/cpufreq/policy5/scaling_max_freq:1900000
> /sys/devices/system/cpu/cpufreq/policy6/scaling_max_freq:1900000
> /sys/devices/system/cpu/cpufreq/policy7/scaling_max_freq:1900000
> 
> At this point I am not certain what I'll get if I try to
> set max_perf_pct to 100%, nor do I know how to find out
> with a user command.
> 
> So, I'll try it:
> 
> doug@s15:~/temp$ echo 100 | sudo tee /sys/devices/system/cpu/intel_pstate/max_perf_pct
> 100
> doug@s15:~/temp$ cat /sys/devices/system/cpu/intel_pstate/max_perf_pct
> 100  <<< Note: 100% = 3.8 GHz in my system)
> doug@s15:~/temp$ grep . /sys/devices/system/cpu/cpufreq/policy*/scaling_max_freq
> /sys/devices/system/cpu/cpufreq/policy0/scaling_max_freq:2200000
> /sys/devices/system/cpu/cpufreq/policy1/scaling_max_freq:2200000
> /sys/devices/system/cpu/cpufreq/policy2/scaling_max_freq:2200000
> /sys/devices/system/cpu/cpufreq/policy3/scaling_max_freq:2200000
> /sys/devices/system/cpu/cpufreq/policy4/scaling_max_freq:2200000
> /sys/devices/system/cpu/cpufreq/policy5/scaling_max_freq:2200000
> /sys/devices/system/cpu/cpufreq/policy6/scaling_max_freq:2200000
> /sys/devices/system/cpu/cpufreq/policy7/scaling_max_freq:2200000
> 
> I guess I had set it sometime earlier, forgot, and then didn't
> get 3.8 Ghz as I had expected via max_perf_pct.

Right, so the problem here (since ages) is that there is no file in
sysfs to show the value earlier set by the user to
scaling_max/min_freq. Rather when you try to read those files, cpufreq
core gives you the current value of policy->min/max, which can come
from so many factors, like userspace, thermal, max_perf_pct, etc.

You saw 2200000 here because you must have set that value to
scaling_max_freq earlier, else maybe thermal or other stuff
constrained that value for you.

Though the cpufreq core stores the value set by user and uses it to
recalculate policy->max using it. To get rid of your constraint, set
scaling_max_freq to the highest frequency and you will be good.

-- 
viresh