Bug 27698

CCing Eric to see if he has idea which commit might regress.

We bisect out that it was caused by the commit cfecde435dda78248d6fcdc424bed68d5db6be0b in srm-intel-next. 

commit cfecde435dda78248d6fcdc424bed68d5db6be0b
Author: Carl Worth <cworth@cworth.org>
Date:   Thu Apr 8 23:31:57 2010 -0700

    drm/i915: Don't enable pipe/plane/VCO early (wait for DPMS on).

    The existing code handling the DPMS ON event is much more careful to
    ensure that these registers are enabled according to strict sequencing
    requirements. Enabling these early in mode_set simply defeats that.

    Signed-off-by: Carl Worth <cworth@cworth.org>
    Signed-off-by: Eric Anholt <eric@anholt.net>

It seems like the only way this commit could affect 3D performance would be by affecting vblank interrupt frequency.  Can you check the vblank interrupt frequency before and after this patch?  If it's the same, I would expect this bisect is just happening to fail, maybe due to some other latent issue.

(In reply to comment #3)
> It seems like the only way this commit could affect 3D performance would be by
> affecting vblank interrupt frequency.  Can you check the vblank interrupt
> frequency before and after this patch?  If it's the same, I would expect this
> bisect is just happening to fail, maybe due to some other latent issue.

Yes, Jesse, you are right. It really affect the vblank interrupt frequency. I used with the kernel before this commit, I can get about 75 vblank interrupts, and if I change to the culprit commit, I can only get 40 vblank interrupts. 

The command I use is: cat /proc/interrupts | grep i915 > interrupt_before_good(bad) ; sleep 1; cat /proc/interrupts | grep i915 > interrupt_after_good(bad)

And I also notice the absolute amount of the vblank interrupt differs a lot between these two kernels. With the commit cfecde435dda78248d6fcdc424bed68d5db6be0b, after I run openarena, it can only receive 15 vblank interrupts per second. And with commit ea059a1ec4496a10f94ca9d0c1b530faf1b85dce,  after I run openarena, it will receive 60 interrupts per second. 

with kernel commit ea059a1ec4496a10f94ca9d0c1b530faf1b85dce:
[root@x-g45c ~]# cat interrupt_before_good
            CPU0       CPU1       CPU2       CPU3
  28:       1498       1655       1213        905   PCI-MSI-edge      i915
[root@x-g45c ~]# cat interrupt_after_good
  28:       1522       1674       1232        918   PCI-MSI-edge      i915

with kernel cfecde435dda78248d6fcdc424bed68d5db6be0b:
[root@x-g45c ~]# cat interrupt_before_bad
  28:        132        184         43         56   PCI-MSI-edge      i915
[root@x-g45c ~]# cat interrupt_after_bad
  28:        151        192         43         69   PCI-MSI-edge      i915

Thanks for the bug report.

I've recommended reverting the bisected commit.

-Carl

Removing out of Q2 release, since Q2 release won't use this code.

Patch reverted. Closing.

It has been reverted in linux-2.6 master branch, but it is still not reverted in drm-intel-next branch.

Verify detailed description:
-------------------------
With the newest code((drm-intel-next)7b824ec2e5d7d086264ecae51e30e3c5e00cdecc) on drm-intel-next, 3D demo openarena 's performance goes up to about 60fps. 
By contrast, with the bug code before(
(drm-intel-next)7f8a85698f5c8a981641ec0bdf9926768786db9d), its  performance goes
down to about 18fps.

Kernel detailed:
-------------------------
Kernel: (drm-intel-next)7b824ec2e5d7d086264ecae51e30e3c5e00cdecc
Some additional commit info:
Author: Eric Anholt <eric@anholt.net>
Date:   Mon Jul 26 14:49:07 2010 -0700

Kernel: (drm-intel-next)7f8a85698f5c8a981641ec0bdf9926768786db9d
Some additional commit info:
Author: Zhenyu Wang <zhenyuw@linux.intel.com>
Date:   Thu Apr 1 13:07:53 2010 +0800

Closing very old verified+fixed.