On Wed, May 28, 2014 at 09:19:05AM +0100, Chris Wilson wrote: > On Tue, Feb 25, 2014 at 03:13:37PM +0200, ville.syrjala@xxxxxxxxxxxxxxx wrote: > > From: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx> > > > > I tried to fix the C3 vs. vblank interrupt issues reportd in [1], but > > it looks like the AGPBUSY# stuff doesn't help there for some reason. I > > guess either the board isn't wired correctly, or we're missing > > something else. I doubt the BM wakup mechanism itself would be > > broken since then I would expect the machine to lock up when someone > > does DMA while in C3. IIRC I actually had that kind of an issue on > > some old VIA chipset long ago. > > > > Anyways, my 855gm actually supports C3, and on that machine the MI_STATE > > AGPBUSY# stuff is effective. So I'm going to assume that gen3 behaviour > > should match, and so I'm just sticking it all into .init_clock_gating() > > for both gen2 and gen3. > > > > I also found another gen3 C3 bit in i915_gem_load(). I think it would > > be better to collect that into .init_clock_gating() as well. But I left > > it also in i915_gem_load() for UMS. > > > > [1] https://bugs.freedesktop.org/show_bug.cgi?id=30364 > > > > Ville Syrjälä (4): > > drm/i915: Set AGPBUSY# bit in init_clock_gating > > drm/i915: Flip the sense of AGPBUSY_DIS bit > > drm/i915: Enable interrupt-based AGPBUSY# enable on 85x > > drm/i915: Move the C3 LP write bit setup to gen3_init_clock_gating() > > for KMS > > They all look sensible, seem to better match the docs than the > existing code and make the code easier to read (apart from the > UMS frobbing!), so: The UMS frobbing is optional. I just figured that when we eventually rip out UMS it would be easier to spot that register write since gem_load() is a rather weird place for such things. > > Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> > > The slow vblank delivery issue remains iirc though. Yeah supposedly that's the case on gen3. This series does fix my 855 though. Unfortunately I've not come across any gen3 machines that suffe from this. I tried it on Mika's gen3 which IIRC did support C3, but vblank interrupts got delivered in a timely fashion regardless of the AGPBUSY bit. -- Ville Syrjälä Intel OTC _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
