Re: [PATCH V2 5/6] cpufreq: governor: replace per-cpu delayed work with timers

[Rafael J. Wysocki]

On Friday, December 04, 2015 11:41:01 AM Viresh Kumar wrote:
> On 04-12-15, 02:18, Rafael J. Wysocki wrote:
> > > +	shared->skip_work--;
> > 
> > Is there any reason for incrementing and decrementing this instead of setting
> > it to either 0 or 1 (or maybe either 'true' or 'false' for that matter)?
> > 
> > If my reading of the patch is correct, it can only be either 0 or 1 anyway, right?
> 
> No. It can be 0, 1 or 2.
> 
> If the timer handler is running on any CPU, we increment skip_work, so
> its value is 1. If at the same time, we try to stop the governor, we
> increment it again and its value is 2 now.
> 
> Once timer-handler finishes, it decrements it and its value become 1.
> Which guarantees that no other timer handler starts executing at this
> point of time and we can safely do gov_cancel_timers(). And once we
> are sure that we don't have any work/timer left, we make it 0 (as we
> aren't sure of the current value, which can be 0 (if the timer handler
> wasn't running when we stopped the governor) or 1 (if the timer
> handler was running while stopping the governor)).
> 
> Hope this clarifies it.

Well, almost, but not quite yet, because now the question is what prevents
gov_cancel_work() from racing with dbs_work_handler().

If you can guarantee that they'll never run in parallel with each other,
you probably don't need the whole counter dance.  Otherwise, dbs_work_handler()
should decrement the counter under the spinlock after all I suppose.

Thanks,
Rafael

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/