Re: [PATCH 1/2] cpufreq: governor: Fix race in dbs_update_util_handler()

[Rafael J. Wysocki]

On Mon, Feb 22, 2016 at 2:04 PM, Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
> On 22-02-16, 13:26, Rafael J. Wysocki wrote:
>> On Mon, Feb 22, 2016 at 6:23 AM, Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
>> > On 21-02-16, 03:14, Rafael J. Wysocki wrote:
>> >> From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>

[cut]

>> > I think we should be doing this here:
>> >
>> >         delta_ns = time - ACCESS_ONCE(policy_dbs->last_sample_time);
>> >         if ((s64)delta_ns < policy_dbs->sample_delay_ns) {
>> >                 atomic_set(&policy_dbs->work_count, 0);
>> >                 return;
>> >         }
>> >
>> > There is no point running the routine again, we already have ->work_count
>> > incremented for us, lets do the check right now.
>>
>> No, we need to check work_in_progress too.
>
> Then maybe move first half of this routine into a separate function
> and call it from the beginning and here ?

That won't help.   The time value is still going to be stale.

Thanks,
Rafael