RE: Kernel lock-up with NPTL

> From: 'Carlo Wood' [mailto:[email protected]]
> 
> On Fri, Jun 06, 2003 at 01:11:41PM -0700, Perez-Gonzalez, Inaky wrote:
> > It is similar, but not the same. This is a TSC based timer; if
> > the interrupts are not too hosed, it will still work.
> 
> Ok, I turned on all debugging that I could possibly
> turned on - and upgraded to 2.5.70.

Ok, now, so this makes a bug report.

> Jun  9 01:01:03 ansset kernel: CPU:    0
> Jun  9 01:01:03 ansset kernel: EIP:    0060:[<c0120582>]    Not tainted
> Jun  9 01:01:03 ansset kernel: EFLAGS: 00010202
> Jun  9 01:01:03 ansset kernel: EIP is at do_fork+0xf2/0x1c0
> Jun  9 01:01:03 ansset kernel: eax: 6b6b6b6b   ebx: 00000000   ecx:
d5256a00   edx: 00000002
> Jun  9 01:01:03 ansset kernel: esi: 6b6b6b6b   edi: 00000000   ebp:
d3b37f98   esp: d3b37f24
> ....
> Jun  9 01:01:03 ansset kernel: Call Trace:
> Jun  9 01:01:03 ansset kernel:  [<c015a807>] mprotect_fixup+0x177/0x280
> Jun  9 01:01:03 ansset kernel:  [<c015acbb>] sys_mprotect+0x3ab/0x3bf
> Jun  9 01:01:03 ansset kernel:  [<c0107cb5>] sys_clone+0x45/0x50
> Jun  9 01:01:03 ansset kernel:  [<c010a24b>] syscall_call+0x7/0xb
> Jun  9 01:01:03 ansset kernel:
> Jun  9 01:01:03 ansset kernel: Code: 0f ab 50 08 89 c8 e8 63 98 ff ff ff
05 ac 4a 43 c0 85 ff 75
> Jun  9 01:01:09 ansset kernel:  <3>Slab corruption: start=d5256a00,
expend=d525707f,
> problemat=d5256fee

I am not that sure this is an NPTL problem - or specific to NPTL,
to be more precise. This is showing that there is corruption of
the memory caches inside the kernel - however, who is causing that
requires more knowledge of the VM than I have (or more time, that
I lack). I suggest you try it again with the latest kernel. It'd
be also interesting if you could isolate the code that causes
as to provide an "exploit".

[maybe an 'strace -vf' dump of running the program can help to 
determine how to provoke it]. 

Iñaky Pérez-González -- Not speaking for Intel -- all opinions are my own
(and my fault)