Re: zfs: Fatal trap 12: page fault while in kernel mode

On Jul 29, 2009, at 16:36, Andriy Gapon wrote:

> on 29/07/2009 17:10 Thomas Backman said the following:
> [snip]
>> (kgdb) fr 11
> [snip]
>> (kgdb) p *sx
>> $8 = {lock_object = {lo_name = 0xffffffff80b5634c "zp->z_lock",  
>> lo_flags
>> = 40894464 [0x2700000, btw], lo_data = 0, lo_witness = 0x0},
>>  sx_lock = 6}
>>
>> ... as you might notice, I'm mostly clueless as to what I'm doing  
>> here. :o
>> Hope that helps (a bit), though.
>
> Yes, it does and a lot.
> sx_lock = 6 means that this sx lock is destroyed:
> #define  
> SX_LOCK_DESTROYED                                               \
>        (SX_LOCK_SHARED_WAITERS | SX_LOCK_EXCLUSIVE_WAITERS)
>
> And lo_name tells that this is zp->z_lock.
> This lock is destroyed in zfs_znode_cache_destructor.
> Not enough knowledge for me to proceed further.

Also, FWIW: Without "options SMP", "zpool" simply goes into an  
uninterruptible sleep (state D+) on export. kill -9 has no effect, and  
the backup process just hangs. The rest of the system works great,  
but... yeah.
(The block that causes the panic is wrapped by #ifdef ADAPTIVE_SX,  
which isn't defined without smp AFAIK.)

Regards,
Thomas