On Mon, May 31, 2004 at 03:15:05PM -0700, Andrew Morton wrote:
> Marcelo Tosatti <[email protected]> wrote:
> >
> > Proc; loop2
> > >>EIP; e5d145c0 <END_OF_CODE+2597a348/????> <=====
> > Trace; c0189ec8 <journal_alloc_journal_head+18/80>
> > Trace; c0189fa2 <journal_add_journal_head+52/140>
> > Trace; c0107b92 <__down+82/d0>
> > Trace; c0107d2c <__down_failed+8/c>
> > Trace; c01845fe <.text.lock.transaction+4/246>
> > Trace; c018204a <new_handle+4a/70>
> > Trace; c0182114 <journal_start+a4/c0>
> > Trace; c017bc3c <ext3_writepage_trans_blocks+1c/a0>
> > Trace; c01795ec <ext3_prepare_write+24c/260>
> > Trace; c013349e <find_or_create_page+5e/150>
> > Trace; c01d6f84 <lo_send+124/2e0>
> > Trace; c01d7408 <do_bh_filebacked+b8/c0>
> > Trace; c01d7b84 <loop_thread+224/250>
> > Trace; c0109172 <ret_from_fork+6/20>
> > Trace; c01d7960 <loop_thread+0/250>
> > Trace; c010740e <arch_kernel_thread+2e/40>
> > Trace; c01d7960 <loop_thread+0/250>
>
> You've run out of memory and the loop thread is looping in
> journal_alloc_journal_head(), waiting for memory to come free.
>
> Meanwhile, kswapd and bdflush are blocked waiting for I/O which requires
> loop thread services to complete. Deadlock.
>
> A proper fix for this might be fairly complex. A kludgey fix like the
> below might work though.
Andrew,
Attached is output of task backtraces, decoded by ksymoops, if you care.
We are in the process of moving this box to v2.6.