Re: PROBLEM: 2.4.26 OOPS in __kmem_cache_alloc

[Hank Leininger <hlein@progressive-comp.com>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Sat, 14 Aug 2004, Denis Vlasenko wrote:

> On Saturday 14 August 2004 06:43, Hank Leininger wrote:
> > The primary MARC server has been out of production for, oh, about
> >
> > [horror story abridged]
>
> Oh no.
>
> > Unable to handle kernel NULL pointer dereference at virtual address
> > 00000021 *pde = 00000000
[snip]
> > Code;  c01f20fb <__kmem_cache_alloc+b/130>
> > 00000000 <_EIP>:
> > Code;  c01f20fb <__kmem_cache_alloc+b/130>   <=====
> >    0:   8b 5c 24 20               mov    0x20(%esp,1),%ebx   <=====
> > Code;  c01f20ff <__kmem_cache_alloc+f/130>
> >    4:   f7 c7 01 00 00 00         test   $0x1,%edi
> > Code;  c01f2105 <__kmem_cache_alloc+15/130>
> >    a:   0f 84 f5 00 00 00         je     105 <_EIP+0x105>
> > Code;  c01f210b <__kmem_cache_alloc+1b/130>
> >   10:   f6 43 30 01               testb  $0x1,0x30(%ebx)
>
> Strange. "Unable to handle kernel NULL pointer dereference
> at virtual address 00000021". But %esp == 0xd426ddf8
> and 0x20(%esp,1) == (%esp+0x20) != (0x00000021)....
> Hmmm. But _%ebp_ is equal to 00000001!
>
> I just checked: "8b 5c 24 20" is indeed "mov 0x20(%esp,1),%ebx",
> it's not a ksymoops bug...
>
> What kind of CPU do you use?

When that OOPS was taken, an Athlon XP 2500+ Barton.  The kernel is
still compiled for SMP (for when it was running with dual Athlon MP
2400+'s; that's where the second, old, less reliable OOPS occurred).

> Please send me your mm/slab.c and mm/slab.o (from
> the kernel build directory).

Sure (sent in a separate email off-list, to spare l-k, unless there is
more interest).

Thanks,

Hank Leininger <hlein@progressive-comp.com>
E407 AEF4 761E D39C D401  D4F4 22F8 EF11 861A A6F1
-----BEGIN PGP SIGNATURE-----

iD8DBQFBHcddIvjvEYYapvERAufTAJwInznww51XBpPi0/evrPgzinfBNQCfQqRY
mcAOrBxNWrs6nPPL/38Oei8=
=wevd
-----END PGP SIGNATURE-----