Re: [PATCH] cpufreq: respect the min/max settings from user space

[Vince Hsu <vinceh@nvidia.com>]

Hi Viresh,

On 10/06/2014 12:45 PM, Viresh Kumar wrote:
> On 2 October 2014 12:25, Vince Hsu <vinceh@nvidia.com> wrote:
>> When the user space tries to set scaling_(max|min)_freq through
>> sysfs, the cpufreq_set_policy() asks other driver's opinions
>> for the max/min frequencies. Some device drivers, like Tegra
>> CPU EDP which is not upstreamed yet though, may constrain the
>> CPU maximum frequency dynamically because of board design.
>> So if the user space access happens and some driver is capping
>> the cpu frequency at the same time, the user_policy->(max|min)
>> is overridden by the capped value, and that's not expected by
>> the user space. And if the user space is not invoked again,
>> the CPU will always be capped by the user_policy->(max|min)
>> even no drivers limit the CPU frequency any more.
>>
>> This patch preserves the user specified min/max settings, so that
>> every time the cpufreq policy is updated, the new max/min can
>> be re-evaluated correctly based on the user's expection and
>> the present device drivers' status.
>>
>> Signed-off-by: Vince Hsu <vinceh@nvidia.com>
>> ---
>> Hi,
>>
>> I'm not sure if any platform that is supported mainlin might have this
>> issue, and this patch is complie tested only.
> Why only compiled tested? Why haven't you tested it on tegra?
I did test with Chrome kernel on Tegra platform. I can't do that with 
mainline kernel because we haven't had the CPU EDP driver upstream yet.

Thanks,
Vince

>
>>   drivers/cpufreq/cpufreq.c | 6 ++++--
>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
>> index 24bf76fba141..c007cf2a3d2a 100644
>> --- a/drivers/cpufreq/cpufreq.c
>> +++ b/drivers/cpufreq/cpufreq.c
>> @@ -524,7 +524,7 @@ static int cpufreq_set_policy(struct cpufreq_policy *policy,
>>   static ssize_t store_##file_name                                       \
>>   (struct cpufreq_policy *policy, const char *buf, size_t count)         \
>>   {                                                                      \
>> -       int ret;                                                        \
>> +       int ret, temp;                                          \
>>          struct cpufreq_policy new_policy;                               \
>>                                                                          \
>>          ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
>> @@ -535,8 +535,10 @@ static ssize_t store_##file_name                                   \
>>          if (ret != 1)                                                   \
>>                  return -EINVAL;                                         \
>>                                                                          \
>> +       temp = new_policy.object;                                       \
>>          ret = cpufreq_set_policy(policy, &new_policy);          \
>> -       policy->user_policy.object = policy->object;                    \
>> +       if (!ret)                                                       \
>> +               policy->user_policy.object = temp;                      \
>>                                                                          \
>>          return ret ? ret : count;                                       \
>>   }
> Looks fine otherwise.
>
> Acked-by: Viresh Kumar <viresh.kumar@linaro.org>