Re: Infinite loop in devfs_create()

On Fri, Aug 12, 2005 at 08:43:26PM -0400, Kris Kennaway wrote:
> I have an SMP amd64 package machine that has deadlocked, apparently
> here:
> 
> --- trap 0x13, rip = 0xffffffff8034b2a1, rsp = 0xffffffffbe6f2a10, rbp = 0xffffffffbe6f2a40 ---
> devfs_create() at devfs_create+0x41
> make_dev_credv() at make_dev_credv+0x148
> make_dev() at make_dev+0x97
> g_dev_taste() at g_dev_taste+0x130
> g_new_provider_event() at g_new_provider_event+0x8a
> one_event() at one_event+0x19b
> g_run_events() at g_run_events+0x9
> g_event_procbody() at g_event_procbody+0x7d
> fork_exit() at fork_exit+0xdf
> fork_trampoline() at fork_trampoline+0xe
> --- trap 0, rip = 0, rsp = 0xffffffffbe6f2d00, rbp = 0 ---
> 
> If I break/continue and examine devfs_nextino, it appears to be
> looping infinitely in devfs_create:
> 
> db> x/d devfs_nextino
> devfs_nextino:  746
> [...]
> db> x/d devfs_nextino
> devfs_nextino:  339
> 
> This machine has a number of md mounts active (they are created and
> destroyed dynamically and tend to get up to high unit numbers, around
> ~600 at the time of deadlock).  Also devfs mounts come and go, but I
> don't think anything else unusual is going on on the system.

This could explain some mysterious deadlocks I've been getting on
sparc machines too, which also cycle through md devices in a similar
way.  I bet something is not being cleaned up properly when the
devices are unconfigured, and eventually devfs is filling up all
available inodes (1024) and failing gracelessly.

Kris