Hi Rik,
today I stressed my current 2.5.31-bk489 a bit, and kswapd
was killed by kernel bug in spinlock.h at line 123, where
spinlock signature is checked. Stack trace was
try_to_free_buffers
try_to_release_page
shrink_list
shrink_dcache
kmem_cache_reap
shrink_caches
try_to_free_pages
kswap_balance_pgdat
kswap_balance
In try_to_free_buffers it died in spin_lock(&mapping->private_lock)
(fs/buffers.c, line 2487). Because of I did not found any code
path which does not initalize mapping's private_lock, I assume
that bug was triggered by asm-generic/rmap.h:pgtable_add_rmap - this
code assigns mm_struct into page's mapping field, and maybe that
such page was passed down to try_to_release_page and everything
went downhill...
Unfortunately my kernel does not have kdb, and 'gdb vmlinux /proc/kcore'
after some time revealed that mapping contents is neither address_space
nor mm_struct (first two fields of structure contained pointer to the
beginning of structure itself, like if structure begins with empty list_t,
and neither of address_space nor mm_struct begins with such field).
Of course it is also possible that mapping was simple released and
memory was reused before all pages using it were destroyed, but I hope that
we do not have such problem in kernel.
System was running XFree, bk export -tplain and diff -urN kernel-tree-1 kernel-tree-2
when BUG happened. Filesystem is ext2, but XFree use also shm. 384MB RAM,
UP system running non-preemptible SMP kernel. After BUG system worked fine
(for hours), only kswapd was zombie.
BTW, mapping was 0xd7c6ecdc.
Thanks,
Petr Vandrovec
vandrove@vc.cvut.cz
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Aug 23 2002 - 22:00:21 EST