Re: [PATCH v2] schedutil: Allow cpufreq requests to be made even when kthread kicked

[Joel Fernandes]

On Tue, May 22, 2018 at 11:52:46PM +0200, Rafael J. Wysocki wrote:
> On Tue, May 22, 2018 at 11:41 PM, Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote:
> > On Tue, May 22, 2018 at 05:27:11PM +0200, Rafael J. Wysocki wrote:
> >> On Tue, May 22, 2018 at 2:22 PM, Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> wrote:
> >> > On Tuesday, May 22, 2018 1:42:05 PM CEST Rafael J. Wysocki wrote:
> >> >> On Tue, May 22, 2018 at 1:38 PM, Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
> >> >> > On 22-05-18, 13:31, Rafael J. Wysocki wrote:
> >> >> >> So below is my (compiled-only) version of the $subject patch, obviously based
> >> >> >> on the Joel's work.
> >> >> >>
> >> >> >> Roughly, what it does is to move the fast_switch_enabled path entirely to
> >> >> >> sugov_update_single() and take the spinlock around sugov_update_commit()
> >> >> >> in the one-CPU case too.
> >> >
> >> > [cut]
> >> >
> >> >> >
> >> >> > Why do you assume that fast switch isn't possible in shared policy
> >> >> > cases ? It infact is already enabled for few drivers.
> >> >
> >> > I hope that fast_switch is not used with devfs_possible_from_any_cpu set in the
> >> > one-CPU policy case, as that looks racy even without any patching.
> >>
> >> Which would be the only case in which sugov_update_single() would run
> >> on a CPU that is not the target.
> >>
> >> And running sugov_update_single() concurrently on two different CPUs
> >> for the same target is a no-no, as we don't prevent concurrent updates
> >> from occurring in that path.
> >>
> >> Which means that the original patch from Joel will be sufficient as
> >> long as we ensure that sugov_update_single() can only run on one CPU
> >> at a time.
> >
> > Since target CPU's runqueue lock is held, I don't see how we can run
> > sugov_update_single concurrently with any other CPU for single policy, so
> > protecting such race shouldn't be necessary.
> 
> If dvfs_possible_from_any_cpu is set, any CPU can run
> sugov_update_single(), but the kthread will only run on the target
> itself.  So another CPU running sugov_update_single() for the target
> may be racing with the target's kthread.
> 
Yes, I agree. I thought you meant the case of sugov_update_single running
currently with other sugov_update_single. So just to be on the same page,
I'll fix the commit log and repost this one as is.

And then I'll post the smp_rmb() patch separately to address the memory order
issue (which I believe is in mainline as well). Basically I was thinking to
address Viresh's issue, there should be an smp_mb() after the next_freq is
read, but before the write to work_in_progress.

thanks,

 - Joel