Re: [PATCH v5]: Improve performance of LZO hibernation

[Pekka Enberg]

On Wed, Sep 28, 2011 at 10:29 AM, Bojan Smojver <bojan@xxxxxxxxxxxxx> wrote:
> On Wed, 2011-09-28 at 10:22 +0300, Pekka Enberg wrote:
>
>> > +       while(1) {
>> > +               wait_event(d->go, atomic_read(&d->ready) ||
>> > +                                 kthread_should_stop());
>> > +               if (kthread_should_stop())
>> > +                       break;
>>
>> So... what happens to the hibernation process when 'kthread_should_stop()'
>> returns true?
>
> The compression/decompression threads stop by breaking out of the loop.
> At least they should, right? Did I misread some docs here?

Yes, the threads are stopped. What happens after that? Will the
hibernation process be aborted? How can this be tested?

>> What's going on here anyway? Why "num_online_cpus() - 1"? What's wrong with
>>
>>   nr_threads = num_online_cpus();
>>   if (nr_threads > LZO_THREADS)
>>     nr_threads = LZO_THREADS;
>
> We want to keep at least one CPU free for that I/O and for pulling the
> other threads into sync when they are done (that is if we have more than
> one), right?

Well, dunno if it matters much. Did you see performance improvement
with that? Is the CPU binding really needed?

Anyway, if you want to keep the existing behavior, maybe something like

nr_other_cpus = min(1, num_online_cpus()-1);

nr_threads = min(nr_other_cpus, LZO_THREADS);

would do the trick?

                        Pekka
--
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/