lkml.org 
[lkml]   [2013]   [Nov]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 1/4] mm/rmap: per anon_vma lock
On Fri, Nov 01, 2013 at 01:07:45PM +0100, Peter Zijlstra wrote:
> On Fri, Nov 01, 2013 at 07:44:29PM +0800, Yuanhan Liu wrote:
> > commit 012f18004da33ba672e3c60838cc4898126174d3
> > Author: Rik van Riel <riel@redhat.com>
> > Date: Mon Aug 9 17:18:40 2010 -0700
> >
> > mm: always lock the root (oldest) anon_vma
> >
> > Always (and only) lock the root (oldest) anon_vma whenever we do something
> > in an anon_vma. The recently introduced anon_vma scalability is due to
> > the rmap code scanning only the VMAs that need to be scanned. Many common
> > operations still took the anon_vma lock on the root anon_vma, so always
> > taking that lock is not expected to introduce any scalability issues.
> >
> > However, always taking the same lock does mean we only need to take one
> > lock, which means rmap_walk on pages from any anon_vma in the vma is
> > excluded from occurring during an munmap, expand_stack or other operation
> > that needs to exclude rmap_walk and similar functions.
> >
> > Also add the proper locking to vma_adjust.
> >
> > Signed-off-by: Rik van Riel <riel@redhat.com>
> > Tested-by: Larry Woodman <lwoodman@redhat.com>
> > Acked-by: Larry Woodman <lwoodman@redhat.com>
> > Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
> > Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > Acked-by: Mel Gorman <mel@csn.ul.ie>
> > Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
> > Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> > Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
>
> Right that commit did.

Sorry again for that! I was jusy being brain dead :(

> I'm still not sure why you change both the
> locking proper and the locking primitive used in one patch set.

convert rwsem to rwlock silightly depends on per anon_vma lock, as it's
a bad idea to do avc allocation inside a spin lock.

Without converting rwsem to rwlock, it's not that useful to introduce
per anon_vma lock, or worse, it may introduce regressions.

>
> Also, changing the locking proper requires a very detailed explanation
> on why it is correct;

Thanks for the tip. And yes, this patch really lacks of some explanation.
I tried to find some potentional races. I then digged the git history
and found it was per anon_lock at the first time avc was introduced.
It was changed to root locking not for fixing race, thus I think we
can changed it back, and this time, for performance boost.

anon_vma lock owns biggest lock contention on our many-core(say 120)servers
from /proc/lock_stat. I found Ingo's patch makes it better, and since
it's a spin lock, I then tried to narrow down the lock range. Hence,
I wrote this patch.

This patch may be wrong, but I guess it's somehow worthy sending out
for comments.

> we've had far too many 'fun' issues with the
> anon_vma locking in the past.

Yeah, I know. Say, http://lwn.net/Articles/383162/ ;)

Thanks.

--yliu


\
 
 \ /
  Last update: 2013-11-01 15:21    [W:0.049 / U:0.388 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site