lkml.org 
[lkml]   [2018]   [Oct]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [BUG -next 20181008] list corruption with "mm/slub: remove useless condition in deactivate_slab"
On Tue, Oct 16, 2018 at 3:36 PM Heiko Carstens
<heiko.carstens@de.ibm.com> wrote:
>
> On Tue, Oct 16, 2018 at 02:29:28PM +0800, Pingfan Liu wrote:
> > > I think it is caused by the uinon page->lru and page->next. It can be fixed by:
> > > diff --git a/include/linux/slub_def.h b/include/linux/slub_def.h
> > > index 3a1a1db..4aa0fb5 100644
> > > --- a/include/linux/slub_def.h
> > > +++ b/include/linux/slub_def.h
> > > @@ -56,6 +56,7 @@ struct kmem_cache_cpu {
> > > #define slub_set_percpu_partial(c, p) \
> > > ({ \
> > > slub_percpu_partial(c) = (p)->next; \
> > > + p->next = NULL; \
> > > })
> > >
> > > I will do some test and post the fix.
> > >
> > Please ignore the above comment. And after re-check the code, I am
> > sure that all callers of deactivate_slab(), pass c->page, which means
> > that page should not be on any list. But your test result "list_add
> > double add: new=000003d1029ecc08,
> > prev=000000008ff846d0,next=000003d1029ecc08" indicates that
> > page(new) is already on a list. I think that maybe something else is
> > wrong which is covered.
> > I can not reproduce this bug on x86. Could you share your config and
> > cmdline? Any do you turn on any debug option of slub?
>
> You can re-create the config with "make ARCH=s390 debug_defconfig".
>
> Not sure which machine I used to reproduce this but most likely it was
> a machine with these command line options:
>
> dasd=e12d root=/dev/dasda1 userprocess_debug numa_debug sched_debug
> ignore_loglevel sclp_con_drop=1 sclp_con_pages=32 audit=0
> crashkernel=128M ignore_rlimit_data
>
> You can ignore the dasd and sclp* command line options. These are
> s390 specific. The rest should be available on any architecture.
>
Thank you for the info. I can reproduce the bug, and find that this
bug is caused by this commit. In deactivate_slab(), page is firstly
add_full(), then hit the redo condition, hence it should be
remove_full(). This wrong commit erases the related code.

Regards,
Pingfan

\
 
 \ /
  Last update: 2018-10-19 03:19    [W:0.061 / U:0.980 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site