Re: [PATCH 0/4] CPUFreq: Implement per policy instances of governors

[Viresh Kumar]

On 4 February 2013 19:06, Borislav Petkov <bp@xxxxxxxxx> wrote:
> On Mon, Feb 04, 2013 at 06:55:25PM +0530, Viresh Kumar wrote:
>> That's not completely true. There lies cpufreq directory in cpu/cpu*/
>> too, where we have per policy stuff in cpu/cpu*/, like policy tunables
>> and stats. And the same is true for governor too.
>
> $ tree /sys/devices/system/cpu/cpu0/cpufreq/
> /sys/devices/system/cpu/cpu0/cpufreq/
> âââ affected_cpus
> âââ bios_limit
> âââ cpb
> âââ cpuinfo_cur_freq
> âââ cpuinfo_max_freq
> âââ cpuinfo_min_freq
> âââ cpuinfo_transition_latency
> âââ related_cpus
> âââ scaling_available_frequencies
> âââ scaling_available_governors
> âââ scaling_cur_freq
> âââ scaling_driver
> âââ scaling_governor
> âââ scaling_max_freq
> âââ scaling_min_freq
> âââ scaling_setspeed
> âââ stats
>     âââ time_in_state
>     âââ total_trans
>     âââ trans_table
>
> 1 directory, 19 files
>
> $ grep -r . /sys/devices/system/cpu/cpu0/cpufreq/*
> /sys/devices/system/cpu/cpu0/cpufreq/affected_cpus:0
> /sys/devices/system/cpu/cpu0/cpufreq/bios_limit:4000000
> /sys/devices/system/cpu/cpu0/cpufreq/cpb:1
> /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq:1400000
> /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq:4000000
> /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_min_freq:1400000
> /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_transition_latency:4000
> /sys/devices/system/cpu/cpu0/cpufreq/related_cpus:0 1
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies:4000000 3400000 2800000 2100000 1400000
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors:powersave userspace conservative ondemand perform
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq:1400000
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver:acpi-cpufreq
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor:ondemand
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq:4000000
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq:1400000
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed:<unsupported>
> /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state:4000000 3089328
> /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state:3400000 47448
> /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state:2800000 67185
> /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state:2100000 92731
> /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state:1400000 11416914
> /sys/devices/system/cpu/cpu0/cpufreq/stats/trans_table:   From  :    To
> /sys/devices/system/cpu/cpu0/cpufreq/stats/trans_table:         :   4000000   3400000   2800000   2100000   140000
> /sys/devices/system/cpu/cpu0/cpufreq/stats/trans_table:  4000000:         0     34756     46388     53179    21824
> /sys/devices/system/cpu/cpu0/cpufreq/stats/trans_table:  3400000:     12938         0      3755      3555     1450
> /sys/devices/system/cpu/cpu0/cpufreq/stats/trans_table:  2800000:     19940         0         0      4547     2565
> /sys/devices/system/cpu/cpu0/cpufreq/stats/trans_table:  2100000:     18523         0         0         0     4275
> /sys/devices/system/cpu/cpu0/cpufreq/stats/trans_table:  1400000:    301168         0         0         0
> /sys/devices/system/cpu/cpu0/cpufreq/stats/total_trans:799918
>
> Show me the policy tunables here.

All files which are directly present in cpu/cpu*/cpufreq/ folder. I am not
talking about governor tunables but policy tunables. Things like
scaling_[min]max_freq are policy tunables.

>> That was slightly confusing to me :( The whole governor directory
>> is per policy, i have to keep that in cpu/cpu*/cpufreq instead of
>> cpu/cpufreq.
>
> So make a /sys/devices/system/cpu/cpufreq/policies/ and add
> functionality to assign cpus to policies or whatever the design of this
> thing will be.

Policies don't have a name associated with them and so cpu/cpufreq/policies
doesn't make any sense. Rather one policy is related to multiple cpus and
its tunables are linked in all the cpus that belong to it, like
scaling_[min]max_freq.

>> Its not only for multicluster system, but a system where multiple cpus
>> have separate clock control and hence multiple policy structures.
>
> What are those systems? Examples?

Don't have examples of these, but there can be few. Over that it is
a must for multicluster systems as clusters normally have separate
clock control.

>> Problem with this is it would fail for single image solutions on which
>> everybody is working on. So, with multiple platforms compiled into a
>> single image, this wouldn't work.
>
> Single-image solutions will enable that config option and get built with
> it - no problem at all.

But then we will get governors tunables in cpu/cpu*/cpufreq/ instead of
cpu/cpufreq/ . Will that not break userspace for other systems?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/