On 12/12/2011 10:18 PM, Ben Skeggs wrote:
> Are you able to bisect this back to a commit?  I've attempted on a few
> different cards to cause this and I'm not seeing it.

Okay, after one of the most time-consuming git bisects ever, I think
I've found it:

0b1076c4b2a06e517fafbb2b4704f23e69b05386 is the first bad commit
commit 0b1076c4b2a06e517fafbb2b4704f23e69b05386
Author: Julian Andres Klode <jak@jak-linux.org>
Date:   Tue Sep 27 19:00:48 2011 +0200

    staging: nvec: Introduce new internal API for msg alloc/free

    Introduce two new functions nvec_msg_alloc() and nvec_msg_free()
    that allocate and free message buffers from the internal pool
    of messages.

    Signed-off-by: Julian Andres Klode <jak@jak-linux.org>
    Acked-by: Marc Dietrich <marvin24@gmx.de>
    Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

Since you guys have probably forgotten about this since it took so long,
I'll recap:

While watching video with mplayer, the nouveau driver crashes rendering
X totally non-responsive, though the system still behaves normally via
ssh or serial console.  Dmesg attached, and I've cc'ed the author of the
first bad commit.

Berck