Re: [PATCH] quota: Make quota stat accounting lockless.

[Dmitry Monakhov <dmonakhov@openvz.org>]

Jan Kara <jack@suse.cz> writes:

> On Mon 26-04-10 19:37:29, Jan Kara wrote:
>> On Mon 26-04-10 20:14:24, Dmitry Monakhov wrote:
>> > Jan Kara <jack@suse.cz> writes:
>> > 
>> > > On Mon 26-04-10 17:16:23, Jan Kara wrote:
>> > >> On Fri 23-04-10 09:31:41, Dmitry Monakhov wrote:
>> > >> > 
>> > >> > I've written this patch long time ago, it is pretty simple, and
>> > >> > allow more non obvious locking optimization to be done later.
>> > >>   The patch looks OK to me. Will merge it.
>> > >   Hmm, maybe I spoke to soon. One question:
>> > >
>> > >> > +void dqstats_inc(unsigned int type)
>> > >> > +{
>> > >> > +#ifdef CONFIG_SMP
>> > >> > +	per_cpu_ptr(dqstats_pcpu, smp_processor_id())->stat[type]++;
>> > >> > +#else
>> > >> > +	dqstats[type]++;
>> > >> > +#endif
>> > >> > +}
>> > >> > +
>> > >> > +void dqstats_dec(unsigned int type)
>> > >> > +{
>> > >> > +#ifdef CONFIG_SMP
>> > >> > +	per_cpu_ptr(dqstats_pcpu, smp_processor_id())->stat[type]--;
>> > >> > +#else
>> > >> > +	dqstats[type]--;
>> > >> > +#endif
>> > >> > +}
>> > >   Why not make these two functions inline? Since they are used
>> > > also from quota_tree.c and quota_v1.c, we'd need to move them to a
>> > > quota.h header in that case.
>> > My initial goal was to keep inc/dec/read helpers together.
>> > We still need to export dqstats_pcpu. But if you prefer this style
>> > i'm ok with it. New version attached.
>>   OK, I've merged the new version (with a few whitespace fixes). Thanks.
>   The patch didn't compile for !CONFIG_SMP systems and also broke
> /proc/sys/fs/quota/warning file. Dmitry, could you be more careful next
> time, please? I've fixed that up and tested that it actually works.
> Attached is a resulting patch that I currently carry.
Hmm.. i dont wat to say... it was my unpredictable laziness effect.
Really sorry for wasting your time in that meaner. Will take more
efforts to avoid this in future.