Re: i915 X lockup

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 27 Feb 2009 10:28:51 +0100 Jiri Slaby <jirislaby@gmail.com> wrote:

> everytime I run X, it gets stuck. Currently running on mmotm 
> 2009-02-26-16-58, but I think this is wider problem. I had i915 disabled 
> for a long time (until I noticed today).
> 
> SysRq : Show Locks Held
> 
> Showing all locks held in the system:
> 3 locks held by events/0/10:
>   #0:  (events){+.+.+.}, at: [<ffffffff8025223d>] worker_thread+0x19d/0x340
>   #1:  (&(&dev_priv->mm.retire_work)->work){+.+...}, at: 
> [<ffffffff8025223d>] worker_thread+0x19d/0x340
>   #2:  (&dev->struct_mutex){+.+.+.}, at: [<ffffffff804057ba>] 
> i915_gem_retire_work_handler+0x3a/0x90
> 1 lock held by mingetty/3899:
>   #0:  (&tty->atomic_read_lock){+.+.+.}, at: [<ffffffff803cb5de>] 
> n_tty_read+0x48e/0x8e0
> 1 lock held by mingetty/3900:
>   #0:  (&tty->atomic_read_lock){+.+.+.}, at: [<ffffffff803cb5de>] 
> n_tty_read+0x48e/0x8e0
> 1 lock held by mingetty/3901:
>   #0:  (&tty->atomic_read_lock){+.+.+.}, at: [<ffffffff803cb5de>] 
> n_tty_read+0x48e/0x8e0
> 1 lock held by X/4007:
>   #0:  (&dev->struct_mutex){+.+.+.}, at: [<ffffffff8040563c>] 
> i915_gem_throttle_ioctl+0x2c/0x60
> 2 locks held by bash/4105:
>   #0:  (sysrq_key_table_lock){......}, at: [<ffffffff803de366>] 
> __handle_sysrq+0x26/0x190
>   #1:  (tasklist_lock){.+.+..}, at: [<ffffffff80266c1f>] 
> debug_show_all_locks+0x3f/0x1c0

I assume that i915_gem_throttle_ioctl->i915_gem_ring_throttle is stuck
in i915_wait_request(), holding struct_mutex.  That of course makes
keventd block.

Perhaps i915_wait_request() is waiting for keventd to do something,
which is the deadlock.  That "something" could be to simply finish its
current call to i915_gem_retire_work_handler().

But worse, it could be some completely other keventd handler which
isn't getting run, because that keventd instance is stuck over in
i915_gem_retire_work_handler().

IOW, the usual keventd problem.