Skip to contentSkip to conversation

i915 fixes

GIT PULL Chris Wilson 4 participants 7 messages Mar 7, 2011 – Mar 8, 2011

lore
Share & tools
lore
#0 lore raw

Dave, hopefully this completes the regression fixes for .38. There is still some doubt over eDP, with 011b9910bdaf failing for Jim Gettys but seeming required for other (currently working in .37) machines. And despite Linus's intervention, backlight controls are still broken for some users.

I've taken the conservative approach and disabled GPU semaphores by default since that prevents one cause of hangs, even though I think it is fixed by the earlier patch to prevent the FIFO overflow. The other patch addresses a bug introduced by the combination of relaxed fencing and fence pipelining infrastructure where to forgot to teardown a bound buffer if it invalidated its new alignment constraints. Hopefully this will fix the reported corruption for gen2/3 devices, though there is the possibility of further underlying bugs.
-Chris

The following changes since commit fb62c00a6d8942775abc23d1621db1252e2d93d1:

Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client (2011-03-05 10:43:22 -0800)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/ickle/drm-intel.git drm-intel-fixes

Chris Wilson (4):
Revert "drm/i915: fix corruptions on i8xx due to relaxed fencing" drm/i915: Do not overflow the MMADDR write FIFO
drm/i915: Disable GPU semaphores by default
drm/i915: Rebind the buffer if its alignment constraints changes with tiling

 drivers/gpu/drm/i915/i915_debugfs.c        |    4 +-
 drivers/gpu/drm/i915/i915_drv.c            |   17 +++++++++++-
 drivers/gpu/drm/i915/i915_drv.h            |   24 ++++++++++++++----
 drivers/gpu/drm/i915/i915_gem.c            |    2 +-
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |    4 +-
 drivers/gpu/drm/i915/i915_gem_tiling.c     |   37 +++++++++++++--------------
 drivers/gpu/drm/i915/i915_reg.h            |    2 +
 drivers/gpu/drm/i915/intel_display.c       |    8 +++---
 drivers/gpu/drm/i915/intel_ringbuffer.h    |   13 +++++----
 9 files changed, 70 insertions(+), 41 deletions(-)
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#1 lore raw
Quoted 5 lines · Chris Wilson

Dave, hopefully this completes the regression fixes for .38. There is still some doubt over eDP, with 011b9910bdaf failing for Jim Gettys but seeming required for other (currently working in .37) machines. And despite Linus's intervention, backlight controls are still broken for some users.

FWIW:

The failure is not so common for me as to claim the change should not go in; and my work around is to suspend/resume the system again.

It's mostly niggling discomfort that there is still something about eDP we still don't understand...

The remaining question, of course, is whether the other fixes in the final release will continue the fine year long tradition of completely breaking my laptop in Linus' releases yet again ;-).
- Jim

Quoted 36 lines

I've taken the conservative approach and disabled GPU semaphores by default since that prevents one cause of hangs, even though I think it is fixed by the earlier patch to prevent the FIFO overflow. The other patch addresses a bug introduced by the combination of relaxed fencing and fence pipelining infrastructure where to forgot to teardown a bound buffer if it invalidated its new alignment constraints. Hopefully this will fix the reported corruption for gen2/3 devices, though there is the possibility of further underlying bugs.
-Chris

The following changes since commit fb62c00a6d8942775abc23d1621db1252e2d93d1:

   Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client (2011-03-05 10:43:22 -0800)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/ickle/drm-intel.git drm-intel-fixes

Chris Wilson (4):
Revert "drm/i915: fix corruptions on i8xx due to relaxed fencing" drm/i915: Do not overflow the MMADDR write FIFO
drm/i915: Disable GPU semaphores by default
drm/i915: Rebind the buffer if its alignment constraints changes with tiling

  drivers/gpu/drm/i915/i915_debugfs.c        |    4 +-
  drivers/gpu/drm/i915/i915_drv.c            |   17 +++++++++++-
  drivers/gpu/drm/i915/i915_drv.h            |   24 ++++++++++++++----
  drivers/gpu/drm/i915/i915_gem.c            |    2 +-
  drivers/gpu/drm/i915/i915_gem_execbuffer.c |    4 +-
  drivers/gpu/drm/i915/i915_gem_tiling.c     |   37 +++++++++++++--------------
  drivers/gpu/drm/i915/i915_reg.h            |    2 +
  drivers/gpu/drm/i915/intel_display.c       |    8 +++---
  drivers/gpu/drm/i915/intel_ringbuffer.h    |   13 +++++----
  9 files changed, 70 insertions(+), 41 deletions(-)
#2 lore raw
Quoted 8 lines · Jim Gettys
Quoted 6 lines · Chris Wilson

Dave, hopefully this completes the regression fixes for .38. There is still some doubt over eDP, with 011b9910bdaf failing for Jim Gettys but seeming required for other (currently working in .37) machines. And despite Linus's intervention, backlight controls are still broken for some users.

Quoted 4 lines

FWIW:

The failure is not so common for me as to claim the change should not go in; and my work around is to suspend/resume the system again.

The thing with this patch is the only think I can think it could do different on your system is timing related.

I'm a bit worried that I haven't seen much attempt at debugging this from the Intel folks,
since its an 80 line patch it really wouldn't harm if they could suggest something.

It might be interetsing to print out the is_pch_port to make sure it never changes,
and also to see if bypassing the whole new code section added in to turn off the CPU FDI and PCH FDI
helps at all.

Dave.

#4 lore raw
In reply to Jim Gettys
Quoted 13 lines · Jim Gettys
Quoted 6 lines · Chris Wilson

Dave, hopefully this completes the regression fixes for .38. There is still some doubt over eDP, with 011b9910bdaf failing for Jim Gettys but seeming required for other (currently working in .37) machines. And despite Linus's intervention, backlight controls are still broken for some
users.

FWIW:

The failure is not so common for me as to claim the change should not go in; and my work around is to suspend/resume the system again.

Quoted 6 lines

It's mostly niggling discomfort that there is still something about eDP we still don't understand...

The remaining question, of course, is whether the other fixes in the final release will continue the fine year long tradition of completely breaking my laptop in Linus' releases yet again ;-).

Well, Chris' tree with the last set of patches works about the same as Linus' rc7; still exhibits the flashing bug, but suspending again can get it to go away. So maybe we can avoid the completely broken Linus release this time (knock on wooden head); certainly we're well beyond the previous state where installation required serious gymnastics.
- Jim

Quoted 51 lines
Quoted 38 lines

I've taken the conservative approach and disabled GPU semaphores by default since that prevents one cause of hangs, even though I think it is fixed by the earlier patch to prevent the FIFO overflow. The other patch addresses a bug introduced by the combination of relaxed fencing and fence
pipelining infrastructure where to forgot to teardown a bound buffer if it
invalidated its new alignment constraints. Hopefully this will fix the reported corruption for gen2/3 devices, though there is the possibility of
further underlying bugs.
-Chris

The following changes since commit
fb62c00a6d8942775abc23d1621db1252e2d93d1:

Merge branch 'for-linus' of
git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client
(2011-03-05 10:43:22 -0800)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/ickle/drm-intel.git
drm-intel-fixes

Chris Wilson (4):
Revert "drm/i915: fix corruptions on i8xx due to relaxed fencing" drm/i915: Do not overflow the MMADDR write FIFO
drm/i915: Disable GPU semaphores by default
drm/i915: Rebind the buffer if its alignment constraints changes with tiling

drivers/gpu/drm/i915/i915_debugfs.c | 4 +-

drivers/gpu/drm/i915/i915_drv.c | 17 +++++++++++-
drivers/gpu/drm/i915/i915_drv.h | 24 ++++++++++++++----
drivers/gpu/drm/i915/i915_gem.c | 2 +-
Quoted 7 lines

drivers/gpu/drm/i915/i915_gem_execbuffer.c | 4 +- drivers/gpu/drm/i915/i915_gem_tiling.c | 37 +++++++++++++-------------- drivers/gpu/drm/i915/i915_reg.h | 2 + drivers/gpu/drm/i915/intel_display.c | 8 +++--- drivers/gpu/drm/i915/intel_ringbuffer.h | 13 +++++----

9 files changed, 70 insertions(+), 41 deletions(-)

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#5 lore raw
Quoted 23 lines · Jim Gettys
Quoted 15 lines · Jim Gettys
Quoted 7 lines · Chris Wilson

Dave, hopefully this completes the regression fixes for .38. There is still some doubt over eDP, with 011b9910bdaf failing for Jim Gettys but seeming required for other (currently working in .37) machines. And despite Linus's intervention, backlight controls are still broken for some
users.

FWIW:

The failure is not so common for me as to claim the change should not go in; and my work around is to suspend/resume the system again.

Quoted 6 lines

It's mostly niggling discomfort that there is still something about eDP we still don't understand...

The remaining question, of course, is whether the other fixes in the final release will continue the fine year long tradition of completely breaking my laptop in Linus' releases yet again ;-).

Quoted 5 lines

Well, Chris' tree with the last set of patches works about the same as Linus' rc7; still exhibits the flashing bug, but suspending again can get it to go away.  So maybe we can avoid the completely broken Linus release this time (knock on wooden head); certainly we're well beyond the previous state where installation required serious gymnastics.

drm-intel-next on my HP laptop fails on boot with some GMBUS error and I have to s/r to see gdm on it ;-)

We have a week more with rc8 so maybe we can get some movement or ideas.

Dave.

#6 lore raw
Quoted 4 lines · Dave Airlie

drm-intel-next on my HP laptop fails on boot with some GMBUS error and I have to s/r to see gdm on it ;-)

We have a week more with rc8 so maybe we can get some movement or ideas.

Unfortunately, the US team, Jesse included, will be preoccupied by the same event as Linus. Maybe they'll find time to fix a bug or two, but I doubt Jesse managed to take his full array of laptops. ;-)

Hmm, that leaves me...
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#3 lore raw
In reply to Dave Airlie
Quoted 22 lines · On Tue, 8 Mar 2011 14:00:45 +1000 Dave Airlie <[email protected]> wrote:
Quoted 8 lines · Jim Gettys
Quoted 6 lines · Chris Wilson

Dave, hopefully this completes the regression fixes for .38. There is still some doubt over eDP, with 011b9910bdaf failing for Jim Gettys but seeming required for other (currently working in .37) machines. And despite Linus's intervention, backlight controls are still broken for some users.

Quoted 4 lines

FWIW:

The failure is not so common for me as to claim the change should not go in; and my work around is to suspend/resume the system again.

The thing with this patch is the only think I can think it could do different on your system is timing related.

I'm a bit worried that I haven't seen much attempt at debugging this from the Intel folks,
since its an 80 line patch it really wouldn't harm if they could suggest something.

Quoted 5 lines

It might be interetsing to print out the is_pch_port to make sure it never changes,
and also to see if bypassing the whole new code section added in to turn off the CPU FDI and PCH FDI
helps at all.

I did check out the reg dumps, and the only theory I have is that we're re-training DP at resume and coming up with a different vswing and/or pre-emphasis, which could cause trouble. The patch should only affect things in a timing sense though like you say. I'll try to find time this week to come up with a patch to allow suspend/resume to use fast link training, and see if that makes things better.

I don't think this is a regression relative to 2.6.37 though? If not, we probably shouldn't revert, since the patch referenced fixes so many other machines...

Jesse