[Intel-gfx] [PATCH] [RFC] drm/i915: Unmask all render responses for SNB

[Chris Wilson]

On Fri, 23 Sep 2011 08:23:14 -0700, Ben Widawsky <ben at bwidawsk.net> wrote:
> On Fri, 23 Sep 2011 11:21:29 +0100
> Chris Wilson <chris at chris-wilson.co.uk> wrote:
> 
> > On SandyBridge in order to receive an event on the RENDER ring, you
> > first must unmask the corresponding bit in the DERRMR (Display Engine
> > Render Response Mask Register). This register is not accessible from a
> > non-secure MI_LOAD_REGISTER_IMM so resort to unmasking all events upon
> > initialisation. Presumably this leads to an undesirable increase in
> > PCI traffic (possibly 4 additional wakeups per vblank per active
> > pipe).
> > 
> > More tricky solutions include extending EXECBUFFER to include a
> > request to enable the appropriate render response and for the kernel
> > to unmask and reset around the batch. Also we need to include a flag
> > to indicate that it is safe to use MI_WAIT_FOR_EVENT as well.
> > 
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > ---
> 
> Sounds good, but...
> 
> {WA} Do not cause more than one display event to be reported in the render
> response.  Either mask off all but one event, unmasking one bit in this
> register, or never initiate more than one event.

Dear HW Engineers,

I hate you all,

Thanks.

Looks like we need a more tricksy solution then. I'm pretty sure that I
can't write to that register from the ring either, but that is the only
sane idea I've had. But this means I can't vsync output to both pipes
which is also sad.

And to think this morning I was enjoying tear-free video at last on SNB.
:(
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre