2.1.56 freeze - detailed report

Eugene Crosser (crosser@average.org)
24 Sep 1997 23:01:14 GMT


First, I hate to say it but none of 2.1.x series ran for me without
freezing within 24 hours. (2.0 and all previous where running for many
weeks without a glitch). But, most freezes happend while in X so I
could not gather any information. This time, I caught it while in text
mode. Here goes.

Vanilla 2.1.56 compiled with SMP (sic) is running on a uniprocessor
486dx4 16M RAM, SCSI disks (no IDE), TCP/IP with PPP and SLIP interfaces
only. Swap partition is ~40M.

Booted it in the morning, and when I returned from work, getty did not
respond to the keyboard and nothing was running. Using magic SysRq
revealed a lot of processes, mainly inews and in.nnrpd (which is no
wonder: they where putting linux-kernel messages into a local
newsgroup;), and EIP flipping around a tight region of addresses in the
kernel. When I did a SysRq-K (kill), the system revived! I could log
in and gracefully reboot it.

Now, here is where the EIP was pointing when I hit SysRq-P several
times:

=============
co12306b kmem_cache_grow(c0123060) +b

mm/slab.c:1136
/* prologue of kmem_cache_grow */

c0123116 kmem_cache_grow(c0123060) +b6

mm/slab.c:1178
spin_lock_irqsave(&cachep->c_spinlock, save_flags);

c012317d kmem_cache_grow(c0123060) +11d

mm/slab.c:513
if (!*dma && addr) {

c012392e kmalloc(c0123800) +12e

mm/slab.c:1432
if (kmem_cache_grow(cachep, flags)) {

c0123988 kmalloc(c0123800) +188

mm/slab.c:1449
return NULL; /* err_exit: from kmalloc */

c01257cd __get_free_pages(c01257a0) +2d

mm/page_alloc.c:205
if (in_interrupt() && priority != GFP_ATOMIC) {

c0125939 __get_free_pages(c01257a0) +199 *****

mm/page_alloc.c:222
spin_unlock_irqrestore(&page_alloc_lock, flags);

c015d9e0 tcp_do_sendmsg(c015d760) +280
/* did not dig in */
=============
(***** this is what I got many times)

I also have other register values and SysRq-M display but they are in
the form of jpeg files: I took snapshots of the screen with a digital
camera. If you need this information please write me and I will type it
down (or send you jpegs if you like:).

>From my previous experience, it seems that 2.1.x freezes when (or several
seconds after) many memory-extensive applications run. And yes it is a
real freeze not just a slowdown. All of a sudden, the mouse pointer in
X stops moving and nothing happens anymore.

I hope that someone knowledgeble in memory management will investigate
this... Thanks in advance.

-- 
Eugene Crosser; 2:5020/230@fidonet; http://www.average.org/~crosser/