Re: [PATCH] make radeons fire fewer vblank interrupts

[Michel <mi...@tu...>]

On Wed, 2007-05-09 at 08:56 -0700, Eric Anholt wrote:
> On Wed, 2007-05-09 at 12:33 +0200, Michel Dänzer wrote:
> > On Fri, 2007-05-04 at 14:47 -0700, Jesse Barnes wrote:
> > > In playing around yesterday, we found that some drivers will 
> > > unnecessarily enable interrupts for vblank events.  Since these tend to 
> > > happen frequently (60+ Hz), they'll cause your CPU to wake up a lot, 
> > > which will waste power if they're not really in use.
> > > 
> > > This patch hacks the radeon driver to only enable vblank interrupts when 
> > > the user is waiting for one, rather than at IRQ setup time.  I couldn't 
> > > find any code in the DDX that wanted vblank support, so I suppose the 
> > > real users are in the Mesa driver somewhere, so I haven't tested it 
> > > other than to see that my interrupt frequency really does decrease.
> > > 
> > > Comments?
> > 
> > I suspect doing it like this might break userspace expectations about
> > the behaviour of the vblank counter. It would be better to do it
> > similarly to how Eric Anholt did it for i915, i.e. by toggling the
> > vblank interrupt in the 2D driver TransitionTo2/3D hooks. Or possibly
> > even better (as this would e.g. still leave it enabled all the time when
> > using a GLX compositing manager), the 3D driver could tell the DRM
> > whether it needs the vblank interrupt or not.
> 
> Yeah, I wasn't sure if the 3D driver would be able to know easily when
> it might need to wait on an absolute vblank.  Arjan suggested that we
> not turn off the vblank when 3d is running until nobody's waited one one
> for (for example) a second, and then if someone waits on one after that,
> the kernel can re-enable the interrupt and extrapolate the vblank
> counter using the system clock.  Hiding vblank enable/disable knowledge
> in the kernel sounds nicer than what I did on i915.

Agreed. Another possibility would be to do as Jesse suggested but to use
the hardware frame counters instead of just incrementing counters on
each interrupt, which may be prone to missing updates anyway.


-- 
Earthling Michel Dänzer           |          http://tungstengraphics.com
Libre software enthusiast         |          Debian, X and DRI developer