Re: [REGRESSION 2.6.35+] crash (maybe kmemleak related)

[Maxim Levitsky <maximlevitsky@gmail.com>]

On Fri, 2010-08-13 at 11:38 +0100, Catalin Marinas wrote: 
> On Fri, 2010-08-13 at 11:28 +0100, Pekka Enberg wrote:
> > On Fri, Aug 13, 2010 at 12:26 PM, Catalin Marinas
> > <catalin.marinas@arm.com> wrote:
> > > This kmemleak warning tells us that the kmemleak_alloc() hook got called
> > > with a pointer (0xffff880079fe6000) that's already registered with
> > > kmemleak. The first trace tells us where the hook gets called from while
> > > the second trace shows the details of the already existing pointer.
> > >
> > > So __getname() allocates the same 4096 bytes block twice via
> > > kmem_cache_alloc() but there is no kmemleak_free() call between them and
> > > kmemleak gives up. It disables itself but does not panic the system.
> > >
> > > Pekka, were there any recent changes in the slab/slob/slub area and
> > > maybe a kmemleak_free() hook is missing? Or maybe something else went
> > > wrong with the slab allocator and returns an already allocated block?
> > >
> > >> <0>[   24.578578] general protection fault: 0000 [#1] PREEMPT SMP
> > >
> > > That's what's causing the fault but it doesn't seem to be related to
> > > kmemleak.
> > 
> > Looking at Maxim's log, slab seems to be corrupted. I don't see
> > anything obviously wrong with SLUB (which he is using) kmemleak hooks
> > so it doesn't look like a slab allocator problem either. Could this be
> > related to the bootmem kmemleak hook changes?
> 
> I don't think in this case since both allocations came via
> kmem_cache_alloc() (and not one from bootmem and the other from slab, as
> it happened to me in the past). But it's worth trying with kmemleak
> disabled.


Thank you.
I will try all the suggestions soon.

Best regards,
Maxim Levitsky