lkml.org 
[lkml]   [2014]   [Jan]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2 1/1] mm: fix the theoretical compound_lock() vs prep_new_page() race
On Thu, Jan 09, 2014 at 03:04:47PM +0100, Oleg Nesterov wrote:
>
> > #ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > page_head = page;
> > if (unlikely(PageTail(page))) {
> > put_page(page);
> >
> >
> > so I'm still not seeing how a tail page racing with a split ends up with
> > mayhem.
>
> But get/put(page_tail) plays with page_head which can be freed/reallocated,
> it does compound_lock(page_head).
>
> > I could also still be stuck in a "la la la, everything is fine" mode.
>
> More likely it is me who tries to deny the fact I missed something ;)
>

My hangup was that this was related to futex and I was focusing it as
a specific example that made the patch necessary. However, this is a
therotical case that potentially impacts a put_page if it mistakenly
believes it is still a tail page when it's not due a a parallel split. I
see and understand that race and while I think the patch is overkill, I
have no problem with including it at the start of a series that reexamines
the locking in that area. It makes for a suitable -stable backport and
I hope/expect the reworked locking would then remove the barrier again
for upstream.

I haven't looked at the reworked locking but understand there is a v3 on
the way so I'll wait until that happens and work my way through it.

Thanks and sorry for the noise.

--
Mel Gorman
SUSE Labs


\
 
 \ /
  Last update: 2014-01-10 18:01    [W:0.114 / U:0.116 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site