[i965] xorg intel crashes on console switch with UXA enabled (UXA bug)

Confirmed. Most people will experience this as being presented with the GDM login screen upon resume.

Marking as wishlist since we've decided not to ship UXA for Jaunty due to stability concerns such as this bug.
I would also encourage you to forward this bug upstream to bugzilla.freedesktop.org so it gets investigated.

OK, forwarded upstream. However I would note that if UXA is wishlist, then performance regression on 965-class hardware should be a blocker for Jaunty. The graphics on Jaunty are completely useless without UXA.

Behavior has now changed somewhat - with UXA on, it no longer crashes on VT switch, but still does not survive suspend/resume, presumably for another reason.

On chvt it logs the following but doesn't crash:
[ 675.321065] [drm:i915_get_vblank_counter] *ERROR* trying to get vblank count for disabled pipe 0

On suspend/resume, the X server crashes without logging anything.

Hi epv, thanks for forwarding this issue upstream.

I notice you originally tested this against 2.6.1, however we've released with 2.6.3 which has numerous UXA bug fixes. Would you mind re-testing against the Jaunty release and see if the issue still exists there?

On Fri, May 01, 2009 at 01:46:15AM -0000, Bryce Harrington wrote:
> Hi epv, thanks for forwarding this issue upstream.
>
> I notice you originally tested this against 2.6.1, however we've
> released with 2.6.3 which has numerous UXA bug fixes. Would you mind
> re-testing against the Jaunty release and see if the issue still exists
> there?

Sure - I'll give it a try when I get to work in the morning.

eric

Okay thanks, let us know how it goes.

On Fri, May 01, 2009 at 01:46:15AM -0000, Bryce Harrington wrote:
> Hi epv, thanks for forwarding this issue upstream.
>
> I notice you originally tested this against 2.6.1, however we've
> released with 2.6.3 which has numerous UXA bug fixes. Would you mind
> re-testing against the Jaunty release and see if the issue still exists
> there?

Hello, sorry for the delay.

Trying to test this actually had me kind of confused for a bit, because I
discovered that X with UXA enabled will not work if the server has been started
without UXA since the last reboot.

Switching consoles does not crash the server now with 2.6.3, though. In
addition, it's no longer unusably slow when rotated, though it's still a
little stairsteppy and slower than without UXA.

Unfortunately, I can't test chvt switching with UXA enabled now - when I start
the X server with UXA turned on:

Section "Device"
        Identifier "ATI Technologies Inc M24GL [Mobility FireGL V3200]"
        Driver "intel"
        Option "DRI" "true"
        Option "AccelMethod" "UXA"
EndSection

If I try to start with UXA enable without rebooting first, I get lots of this:

[ 625.864076] [drm:i915_gem_idle] *ERROR* hardware wedged
[ 625.864238] [drm:i915_get_vblank_counter] *ERROR* trying to get vblank count for disabled pipe 0
[ 629.772764] [drm:i915_setparam] *ERROR* unknown parameter 4
[ 629.795155] [drm:i915_gem_entervt_ioctl] *ERROR* Reenabling wedged hardware, good luck
[ 633.628113] [drm:i915_wait_irq] *ERROR* EBUSY -- rec: 1147 emitted: 1156
[ 636.628116] [drm:i915_wait_irq] *ERROR* EBUSY -- rec: 1147 emitted: 1157
[ 638.288062] [drm:i915_gem_idle] *ERROR* hardware wedged
[ 638.288210] [drm:i915_get_vblank_counter] *ERROR* trying to get vblank count for disabled pipe 0
[ 642.187486] [drm:i915_setparam] *ERROR* unknown parameter 4
[ 642.219444] [drm:i915_gem_entervt_ioctl] *ERROR* Reenabling wedged hardware, good luck
[ 646.116113] [drm:i915_wait_irq] *ERROR* EBUSY -- rec: 1147 emitted: 1158
[ 649.116081] [drm:i915_wait_irq] *ERROR* EBUSY -- rec: 1147 emitted: 1159
[ 650.772105] [drm:i915_gem_idle] *ERROR* hardware wedged
[ 650.772271] [drm:i915_get_vblank_counter] *ERROR* trying to get vblank count for disabled pipe 0
[ 662.725419] mtrr: no MTRR for e0000000,770000 found
[ 688.460665] [drm:i915_setparam] *ERROR* unknown parameter 4
[ 688.507581] [drm:i915_gem_entervt_ioctl] *ERROR* Reenabling wedged hardware, good luck

but it never manages to draw the display again until I reboot.

Thanks,
eric

Am I understanding your response correctly, that the original issue (crash on console switch) is resolved now, if UXA is set in xorg.conf from boot?

On Tue, May 05, 2009 at 05:15:20AM -0000, Bryce Harrington wrote:
> Am I understanding your response correctly, that the original issue
> (crash on console switch) is resolved now, if UXA is set in xorg.conf
> from boot?
>
> --
> [i965] xorg intel crashes on console switch with UXA enabled (UXA bug)
> https://bugs.launchpad.net/bugs/329860
> You received this bug notification because you are a direct subscriber
> of the bug.

Correct. Console switching with UXA enabled causes
"[drm:i915_get_vblank_counter] *ERROR* trying to get vblank count for disabled pipe 0"
to be logged, but succeeds regardless.

eric

Okay, the use case for UXA going forward (esp. starting 2.8) will be to have UXA set by default, so while this may continue to be a problem in the short term, by the Karmic release it should no longer be a concern.