Re: [Bugme-new] [Bug 8778] New: Ocotea board: kernel reports access of bad area during boot with DEBUG_SLAB=y

Re: [Bugme-new] [Bug 8778] New: Ocotea board: kernel reports access of bad area during boot with DEBUG_SLAB=y

[Andrew Morton]

On Wed, 18 Jul 2007 08:59:40 -0700 Eugene Surovegin <ebs@ebshome.net> wrote:

> On Wed, Jul 18, 2007 at 08:41:10AM -0500, Josh Boyer wrote:
> > On Wed, 2007-07-18 at 01:34 -0700, Eugene Surovegin wrote:
> > > On Wed, Jul 18, 2007 at 12:52:53AM -0700, Andrew Morton wrote:
> > > > On Wed, 18 Jul 2007 00:07:50 -0700 (PDT) bugme-daemon@bugzilla.kernel.org wrote:
> > > > 
> > > > > http://bugzilla.kernel.org/show_bug.cgi?id=8778
> > > > > 
> > > > >            Summary: Ocotea board: kernel reports access of bad area during
> > > > >                     boot with DEBUG_SLAB=y
> > > 
> > > Slab debugging is probably the culprit here. I had similar problem 
> > > couple of years ago, not sure something has changed since then, 
> > > haven't checked.
> > > 
> > > When slab debugging was enabled it made memory allocations non L1 
> > > cache line aligned. This is very bad for DMA on non-coherent cache 
> > > arches (PPC440 is one of those archs).
> > > 
> > > I have a hack for EMAC which tries to "workaround" this problem:
> > > 	http://kernel.ebshome.net/emac_slab_debug.diff
> > > which might help.
> > 
> > Would you be opposed to including that patch in mainline?
> 
> Yes. I don't think it's the right way to fix this issue. IMO, the 
> right one is to fix slab allocator. You cannot change all drivers to 
> do this kind of cache flushing, and yes, I saw the same problem with 
> PCI based NIC I tried on Ocotea at the time.
> 

hm.  It should be the case that providing SLAB_HWCACHE_ALIGN at
kmem_cache_create() time will override slab-debugging's offsetting
of the returned addresses.

Or is the problem occurring with memory which is returned from kmalloc(),
rather than from kmem_cache_alloc()?

A complete description of the problem would help here, please.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [Bugme-new] [Bug 8778] New: Ocotea board: kernel reports access of bad area during boot with DEBUG_SLAB=y

[Eugene Surovegin]

On Wed, Jul 18, 2007 at 09:55:37AM -0700, Andrew Morton wrote:
> On Wed, 18 Jul 2007 08:59:40 -0700 Eugene Surovegin <ebs@ebshome.net> wrote:
> 
> > On Wed, Jul 18, 2007 at 08:41:10AM -0500, Josh Boyer wrote:
> > > On Wed, 2007-07-18 at 01:34 -0700, Eugene Surovegin wrote:
> > > > On Wed, Jul 18, 2007 at 12:52:53AM -0700, Andrew Morton wrote:
> > > > > On Wed, 18 Jul 2007 00:07:50 -0700 (PDT) bugme-daemon@bugzilla.kernel.org wrote:
> > > > > 
> > > > > > http://bugzilla.kernel.org/show_bug.cgi?id=8778
> > > > > > 
> > > > > >            Summary: Ocotea board: kernel reports access of bad area during
> > > > > >                     boot with DEBUG_SLAB=y
> > > > 
> > > > Slab debugging is probably the culprit here. I had similar problem 
> > > > couple of years ago, not sure something has changed since then, 
> > > > haven't checked.
> > > > 
> > > > When slab debugging was enabled it made memory allocations non L1 
> > > > cache line aligned. This is very bad for DMA on non-coherent cache 
> > > > arches (PPC440 is one of those archs).
> > > > 
> > > > I have a hack for EMAC which tries to "workaround" this problem:
> > > > 	http://kernel.ebshome.net/emac_slab_debug.diff
> > > > which might help.
> > > 
> > > Would you be opposed to including that patch in mainline?
> > 
> > Yes. I don't think it's the right way to fix this issue. IMO, the 
> > right one is to fix slab allocator. You cannot change all drivers to 
> > do this kind of cache flushing, and yes, I saw the same problem with 
> > PCI based NIC I tried on Ocotea at the time.
> > 
> 
> hm.  It should be the case that providing SLAB_HWCACHE_ALIGN at
> kmem_cache_create() time will override slab-debugging's offsetting
> of the returned addresses.
> 
> Or is the problem occurring with memory which is returned from kmalloc(),
> rather than from kmem_cache_alloc()?

It's kmalloc, at least this is how I think skbs are allocated.

Andrew, I don't have access to PPC hw right now (doing MIPS 
development these days), so I cannot quickly check that my theory is 
still correct for the latest kernel. I'd wait for the reporter to try 
my hack and then we can decide what to do. IIRC there was some 
provision in slab allocator to enforce alignment, when I was debugging 
this problem more then a year ago, that option didn't work.

BTW, I think slob allocator had the same issue with alignment as slab 
with enabled debugging (at least at the time I looked at it).

-- 
Eugene

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [Bugme-new] [Bug 8778] New: Ocotea board: kernel reports access of bad area during boot with DEBUG_SLAB=y

[Bart Van Assche]

[-- Attachment #1.1: Type: text/plain, Size: 1205 bytes --]

On 7/18/07, Eugene Surovegin <ebs@ebshome.net> wrote:
>
>
> It's kmalloc, at least this is how I think skbs are allocated.
>
> Andrew, I don't have access to PPC hw right now (doing MIPS
> development these days), so I cannot quickly check that my theory is
> still correct for the latest kernel. I'd wait for the reporter to try
> my hack and then we can decide what to do. IIRC there was some
> provision in slab allocator to enforce alignment, when I was debugging
> this problem more then a year ago, that option didn't work.
>
> BTW, I think slob allocator had the same issue with alignment as slab
> with enabled debugging (at least at the time I looked at it).



Hello Eugene,

In case you didn't notice yet, I have added the following comment to the
kernel bugzilla item:


------- *Comment #5
<http://bugzilla.kernel.org/show_bug.cgi?id=8778#c5>From Bart
Van Assche <bart.vanassche@gmail.com> 2007-07-18 07:12:49 *
[reply<http://bugzilla.kernel.org/show_bug.cgi?id=8778#add_comment>]
-------

I have downloaded the patch from
http://kernel.ebshome.net/emac_slab_debug.diff, and I have tried it. Hereby I
confirm that this patch solves the reported kernel oops.



-- 
Regards,

Bart Van Assche.

[-- Attachment #1.2: Type: text/html, Size: 1884 bytes --]

[-- Attachment #2: Type: text/plain, Size: 161 bytes --]

_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded

Re: [Bugme-new] [Bug 8778] New: Ocotea board: kernel reports access of bad area during boot with DEBUG_SLAB=y

[Christoph Lameter]

On Wed, 18 Jul 2007 09:55:37 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:

> hm.  It should be the case that providing SLAB_HWCACHE_ALIGN at
> kmem_cache_create() time will override slab-debugging's offsetting
> of the returned addresses.


That is true for SLUB but not in SLAB. SLAB has always ignored
SLAB_HWCACHE_ALIGN when debugging is on because of the issues involved
in placing the redzone values etc.  Could be fun to fix.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>