Re: [PATCH RFC] schedutil: Address the r/w ordering race in kthread

["Rafael J. Wysocki" <rafael@kernel.org>]

On Wed, May 23, 2018 at 1:50 AM, Joel Fernandes (Google)
<joelaf@google.com> wrote:
> Currently there is a race in schedutil code for slow-switch single-CPU
> systems. Fix it by enforcing ordering the write to work_in_progress to
> happen before the read of next_freq.
>
> Kthread                                       Sched update
>
> sugov_work()                                  sugov_update_single()
>
>       lock();
>       // The CPU is free to rearrange below
>       // two in any order, so it may clear
>       // the flag first and then read next
>       // freq. Lets assume it does.
>       work_in_progress = false
>
>                                                if (work_in_progress)
>                                                      return;
>
>                                                sg_policy->next_freq = 0;
>       freq = sg_policy->next_freq;
>                                                sg_policy->next_freq = real-freq;
>       unlock();
>
> Reported-by: Viresh Kumar <viresh.kumar@linaro.org>
> CC: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> CC: Peter Zijlstra <peterz@infradead.org>
> CC: Ingo Molnar <mingo@redhat.com>
> CC: Patrick Bellasi <patrick.bellasi@arm.com>
> CC: Juri Lelli <juri.lelli@redhat.com>
> Cc: Luca Abeni <luca.abeni@santannapisa.it>
> CC: Todd Kjos <tkjos@google.com>
> CC: claudio@evidence.eu.com
> CC: kernel-team@android.com
> CC: linux-pm@vger.kernel.org
> Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
> ---
> I split this into separate patch, because this race can also happen in
> mainline.
>
>  kernel/sched/cpufreq_schedutil.c | 7 +++++++
>  1 file changed, 7 insertions(+)
>
> diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c
> index 5c482ec38610..ce7749da7a44 100644
> --- a/kernel/sched/cpufreq_schedutil.c
> +++ b/kernel/sched/cpufreq_schedutil.c
> @@ -401,6 +401,13 @@ static void sugov_work(struct kthread_work *work)
>          */
>         raw_spin_lock_irqsave(&sg_policy->update_lock, flags);
>         freq = sg_policy->next_freq;
> +
> +       /*
> +        * sugov_update_single can access work_in_progress without update_lock,
> +        * make sure next_freq is read before work_in_progress is set.
> +        */
> +       smp_mb();
> +

This requires a corresponding barrier somewhere else.

>         sg_policy->work_in_progress = false;
>         raw_spin_unlock_irqrestore(&sg_policy->update_lock, flags);
>
> --

Also, as I said I actually would prefer to use the spinlock in the
one-CPU case when the kthread is used.

I'll have a patch for that shortly.