[Intel-gfx] [PATCH] drm/i915: dump even more into the error_state

[Ben Widawsky]

On Sat, Feb 04, 2012 at 10:45:50AM +0000, Chris Wilson wrote:
> On Fri, 3 Feb 2012 18:23:10 -0800, Ben Widawsky <ben at bwidawsk.net> wrote:
> > On Wed, Feb 01, 2012 at 10:26:45PM +0100, Daniel Vetter wrote:
> > > Because the hangs always involved semaphores in one way or another,
> > > we've tried to dump the mbox registers, but couldn't find any
> > > inconsistencies. Still, dump them too.
> > 
> > How about we only do this if semaphores are actually enabled? I think
> > that will eliminate potential confusion.
> 
> I suggested that we add the module parameters to the error-state to
> aide, though even that is not fool proof as they could change between
> the driver doing an undefined op and it fouling up the GPU. Any sign of
> conflicting information inside the error-state is meant to be thought
> provoking. ;-)
> 
> The idea is to dump as much information as we can about the errors and
> leave the complex art of interpretation up to userspace.
> -Chris

I understand your sentiment, but I think this is one of those things
where 18 months from now someone (very likely me) will spend far too
much time chasing weird MBOX values only to realize semaphores weren't
turned on. Dumping the module parameters is good enough for me. The risk
of the situation you describe should be quite low.

Personally, I think it's confusing enough that we put the MBOX_UPDATES
in the ring even when semaphores aren't on.

~Ben