[Intel-gfx] [PATCH] [drm/i915] Handle swap buffers in work proc instead of softirq

[Jesse Barnes]

On Saturday, October 18, 2008 3:27 pm Keith Packard wrote:
> On Sat, 2008-10-18 at 15:05 -0700, Jesse Barnes wrote:
> > No it's previous chips that have the LVDS on pipe B, FBC only on plane A
> > limitation.  But the flipping only ever worked with drm.git master bits,
> > which never went upstream.  So if you want to redefine the APIs that's
> > still possible.
>
> Ok, that's what Eric said.
>
> > It doesn't matter which way we go as long as we're consistent.  There are
> > only two places where you need to know the mapping:  FBC setup (and
> > therefore 2D mode programming) and vblank interrupt time.  The rest of
> > the code in Mesa and elsewhere can just deal with either planes or pipes
> > all the time.
>
> Given that using pipes eliminates piles of ugly plane<->pipe conversions
> in vblank, and that the mode setting code doesn't really care all that
> much, I'd say using pipes is a better plan.
>
> I've got a patch sitting here that does this; the result is a net
> reduction of code.

Cool, sounds good, and is probably less confusing since CRTC is generally 
equated to pipes.

-- 
Jesse Barnes, Intel Open Source Technology Center