[Intel-gfx] [PATCH] drm/i915: avoid struct mutex output_poll mutex lock loop on unload

[Dave Airlie]

On Mon, Oct 4, 2010 at 6:49 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> On Sun,  3 Oct 2010 19:36:26 -0700, Keith Packard <keithp at keithp.com> wrote:
>> Cancel the output polling work proc before acquiring the struct mutex
>> to avoid acquiring the work proc mutex with the struct mutex
>> held. This avoids inverting the lock order seen when the work proc
>> runs.
>
> I thought this was part of Daniel's reordering to avoid race conditions
> between interrupts and module unload. Fortunately not. The lack of locking
> during hotplug is worrisome should we ever do dynamic connectors.

We don't do dynamic connectors now, so adding locking with no way of
actually really testing it would just mean you'd probably have just as
much pain when you do add dyanamic connectors.

I also tried to think of a reason to support dynamic connectors, maybe
displayport daisychaining, but none of the userspace drivers are ready
for it.

Dave.