[Intel-gfx] high number of wakeups/second shown with 2.5 branch

[Tino Keitel]

On Mon, Oct 20, 2008 at 09:41:22 -0700, Eric Anholt wrote:
> On Mon, 2008-10-20 at 08:29 +0200, Tino Keitel wrote:
> > Hi,
> > 
> > after I switched to the pre-2.5 driver, I often see a high number of
> > wakeups/second like this:
> > 
> > Top causes for wakeups:
> >   76.7% (1268.1)       <interrupt> : i915 at pci:0000:00:02.0 
> >   16.3% (269.4)      <kernel IPI> : Rescheduling interrupts 
> >    5.2% ( 85.5)       <interrupt> : PS/2 keyboard/mouse/touchpad 
> >    0.8% ( 13.6)       <interrupt> : extra timer interrupt 
> >    0.2% (  3.1)      <kernel IPI> : function call interrupts
> >    0.2% (  3.1)       firefox-bin : futex_wait (hrtimer_wakeup)
> >    0.1% (  1.5)              Xorg : schedule_timeout (process_timeout)
> > 
> > This wasn't the case with the 2.4 driver. Currently, I use the version
> > 2.4.98 on the ThinkPad X61s with i965.
> > 
> > Is there a specific reason for this behaviour with the 2.5 branch?
> 
> Without knowing your specific workload it's hard to tell if there's a
> problem or not, but we've replaced busy-waiting with using interrupts in
> the 2d driver.

The workload is a plain 2D desktop (no compositing), with clicking
around a bit, scrolling, changing window focus etc. A completely idle
desktop with no moving parts doesn't produce any wakeups in the i915
driver.

Regards,
Tino