[BUG] __write_lock_failed() oops

[William Lee Irwin III <wli@holomorphy.com>]

After running 64 simultaneous tiobench 256's a few times,
I get the following oops, which I've been seeing intermittently
for a number of 2.5.x releases (since 2.5.x booted on NUMA-Q):


Program received signal SIGSEGV, Segmentation fault.
0xc0106693 in __write_lock_failed () at semaphore.c:176
176     semaphore.c: No such file or directory.
        in semaphore.c

for some reason, I'm unable to get a backtrace:

(gdb) bt
#0  0xc0106693 in __write_lock_failed () at semaphore.c:176
Reply contains invalid hex digit 36

This one's relatively painful to reproduce. My attempts to debug it
have more or less flopped thus far.


(gdb) disassemble __write_lock_failed
Dump of assembler code for function __write_lock_failed:
0xc0106684 <__write_lock_failed>:       lock addl $0x1000000,(%eax)
0xc010668b <__write_lock_failed+7>:     repz nop 
0xc010668d <__write_lock_failed+9>:     cmpl   $0x1000000,(%eax)
0xc0106693 <__write_lock_failed+15>:    
    jne    0xc010668b <__write_lock_failed+7>
0xc0106695 <__write_lock_failed+17>:    lock subl $0x1000000,(%eax)
0xc010669c <__write_lock_failed+24>:    jne    0xc0106684 <__write_lock_failed>
0xc01066a2 <__write_lock_failed+30>:    ret    
0xc01066a3 <__write_lock_failed+31>:    nop    
End of assembler dump.

(gdb) p/x $eax
$3 = 0xc0331ca0
(gdb) p/x *(unsigned long *)$eax
$4 = 0xffffff



Cheers,
Bill