lkml.org 
[lkml]   [2015]   [Jul]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 3/3] zsmalloc: do not take class lock in zs_pages_to_compact()
Hi Sergey,

On Wed, Jul 15, 2015 at 01:07:03PM +0900, Sergey Senozhatsky wrote:
> On (07/11/15 18:45), Sergey Senozhatsky wrote:
> [..]
> > We re-do this calculations during compaction on a per class basis
> > anyway.
> >
> > zs_unregister_shrinker() will not return until we have an active
> > shrinker, so classes won't unexpectedly disappear while
> > zs_pages_to_compact(), invoked by zs_shrinker_count(), iterates
> > them.
> >
> > When called from zram, we are protected by zram's ->init_lock,
> > so, again, classes will be there until zs_pages_to_compact()
> > iterates them.
> >
> > Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
> > ---
> > mm/zsmalloc.c | 2 --
> > 1 file changed, 2 deletions(-)
> >
> > diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
> > index b10a228..824c182 100644
> > --- a/mm/zsmalloc.c
> > +++ b/mm/zsmalloc.c
> > @@ -1811,9 +1811,7 @@ unsigned long zs_pages_to_compact(struct zs_pool *pool)
> > if (class->index != i)
> > continue;
> >
> > - spin_lock(&class->lock);
> > pages_to_free += zs_can_compact(class);
> > - spin_unlock(&class->lock);
> > }
> >
> > return pages_to_free;
>
> This patch still makes sense. Agree?

There is already race window between shrink_count and shrink_slab so
it would be okay if we return stale stat with removing the lock if
the difference is not huge.

Even, now we don't obey nr_to_scan of shrinker in zs_shrinker_scan
so the such accuracy would be pointless.

Please resend the patch and correct zs_can_compact's comment.

Thanks.


\
 
 \ /
  Last update: 2015-07-16 02:01    [W:0.064 / U:1.556 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site