Re: [PATCH -next 3/3] cpufreq: conservative: Fix relation when decreasing frequency

[Namhyung Kim <namhyung@kernel.org>]

Hi Viresh,

On Thu, 28 Feb 2013 11:17:03 +0530, Viresh Kumar wrote:
> On 28 February 2013 11:08, Namhyung Kim <namhyung@kernel.org> wrote:
>> From: Namhyung Kim <namhyung.kim@lge.com>
>>
>> The relation should be CPUFREQ_RELATION_L to find optimal frequency
>> when decreasing.
>>
>> Cc: Viresh Kumar <viresh.kumar@linaro.org>
>> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
>> ---
>>  drivers/cpufreq/cpufreq_conservative.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/cpufreq/cpufreq_conservative.c b/drivers/cpufreq/cpufreq_conservative.c
>> index dd2fd9094819..0d582811d66c 100644
>> --- a/drivers/cpufreq/cpufreq_conservative.c
>> +++ b/drivers/cpufreq/cpufreq_conservative.c
>> @@ -106,7 +106,7 @@ static void cs_check_cpu(int cpu, unsigned int load)
>>                         dbs_info->requested_freq = policy->min;
>>
>>                 __cpufreq_driver_target(policy, dbs_info->requested_freq,
>> -                               CPUFREQ_RELATION_H);
>> +                               CPUFREQ_RELATION_L);
>
> Other two patches are fine but really not sure about this one.
> When decreasing freq, what do we want:
> - lowest frequency at or above target, i.e. >= requested_freq
> - highest frequency below or at target, i.e. <= requested_freq
>
> I thought second option was better and so CPUFREQ_RELATION_H
> suits more. What made you do this change?

When decreasing, we were on a higher frequency than target so selecting
above or equal to the target frequency seems to be "conservative".  And
AFAICS the ondemance governor also uses RELATION_L for decreasing.

Thanks,
Namhyung