omapdss DISPC error: GFX_FIFO_UNDERFLOW, disabling GFX

[Mathieu Poirier]

Platform: Panda board, 8 layers.
Processor: TI OMAP4

Symptom: The screen will go black randomly - even when there is activity - and doesn't come back. The following output can be found on the serial console (when enabled) and the log file:

 [ 404.169006] omapdss DISPC error: GFX_FIFO_UNDERFLOW, disabling GFX

To recover the screen, change to another console and get back to tty7: CTRL + ALT + F1 followed by CTRL + ALT + F7

This is a known problem that is currently being addressed by TI.

[Lee Jones (lag)]

This is a known issue.

It comes when the screen resolution is set too high for the system.

Mine does it all the time (6 layers).

Speak to Rob Clark for more information.

[Ricardo Salveti (rsalveti)]

For this fix you also need MLO from x-loader-omap4_L24.9git20100901-0ubuntu3.

And the fix should just decrease it, but it's still happening once in a while.

[Ricardo Salveti (rsalveti)]

Patches target to fix this issue: http://kernel.ubuntu.com/git?p=rsalveti/ubuntu-maverick.git;a=shortlog;h=refs/heads/rsalveti-ti-omap4-fix-gfx-underflow

And you can find the generated kernel deb file with these fixes at: http://people.canonical.com/~rsalveti/maverick/kernel/es2/linux-image-2.6.35-903-omap4_2.6.35-903.18rsalveti2_armel.deb

Tested while using SGX (without these patches you can easily get an underflow while using SGX), and I couldn't reproduce the issue anymore.

Please if someone could also test it'd be very helpful. Then I'll forward the fixes to the kernel team.

[Sebastien JAN (sebjan)]

Was anyone able to test reverting to the previous fifo setting using the buffer_state sysfs entry added by this patch-set?
(I'd like to be sure that we can use this feature to revert back to the standard setting if needed + would be curious of the impact of this fifo trick on the sync issues)

[Vincent Stehlé (vstehle)]

Ricardo, I confirm that I have no "black screen" in 1080p after a few hours. This is using your 18rsalveti2 .deb files.

Thanks!

V.

[Ricardo Salveti (rsalveti)]

On Fri, Nov 12, 2010 at 6:06 AM, Sebastien JAN <email address hidden> wrote:
> Was anyone able to test reverting to the previous fifo setting using the buffer_state sysfs entry added by this patch-set?
> (I'd like to be sure that we can use this feature to revert back to the standard setting if needed + would be curious of the impact of this fifo trick on the sync issues)

I tried with 1 at the buffer_state and I was able to use the PVR the
same way as setting it to 0. Is there any other test that we can do to
check if it actually reverted back to the standard settings?

[Ricardo Salveti (rsalveti)]

2010/11/12 Vincent Stehlé <email address hidden>:
> Ricardo, I confirm that I have no "black screen" in 1080p after a few
> hours. This is using your 18rsalveti2 .deb files.

It really gets better, but it's still not the final fix. You can still
easily get the fifo underflow by running "$ x11perf -copywinwin500"

[Roope Kuisma (meizirkki)]

This problem also affected the Omap3 touchbook. But it's now marked as fixed.
http://bugzilla.alwaysinnovating.com/show_bug.cgi?id=39

Their kernel tree is here:
http://git.alwaysinnovating.com/cgit.cgi/ai.openembedded.dev/tree/recipes/linux/linux-omap-psp-2.6.32

[Ricardo Salveti (rsalveti)]

SRU Justification:

Impact: While using SGX acceleration on Pandaboard, it's quite normal to face
the GFX_FIFO_UNDERFLOW issue, that disable the display entirely. The only
way to get the screen back to work is to reset it, or by changing to a text mode
session and then getting back to X11. This issue makes practically impossible
to use and stress the SGX driver on a Full HD monitor.

Fix: While TI still didn't provide a proper way to fix is, a valid workaround was
proposed to use the writeback buffers for GFX. The proposed patches are all
already applied at Natty's branch, and this is basically a backport of them for
Maverick.

Testcase: Just run Unity using the SGX driver on a Full HD monitor, and try to
use it normally, and your screen will probably get black with the following
message at your dmesg:
[ 404.169006] omapdss DISPC error: GFX_FIFO_UNDERFLOW, disabling GFX
After applying these patches, you should be able to normally use Unity, without
heading into any underflow issue.

[Ricardo Salveti (rsalveti)]

Pull request: https://lists.ubuntu.com/archives/kernel-team/2010-December/013903.html

[Martin Pitt (pitti)]

Accepted linux-ti-omap4 into maverick-proposed, the package will build now and be available in a few hours. Please test and give feedback here. See https://wiki.ubuntu.com/Testing/EnableProposed for documentation how to enable and use -proposed. Thank you in advance!

[Ricardo Salveti (rsalveti)]

Vincent, can you please test this proposed kernel again so we can push it to updates? Thanks in advance.

[Vincent Stehlé (vstehle)]

I added maverick-proposed and tried linux-image-2.6.35-903-omap4 version 2.6.35-903.20 on a Panda. It works fine, I can run Unity on a 1080p monitor with no issue (i.e. no black screen).

Note that the "black screen" can still be obtained "on purpose" with 'x11perf -copywinwin500' even under the recovery console, though. In which case you need to do a CTRL-ALT-F1 / ALT-F8 "switch" to get the display back.

V.

[Ricardo Salveti (rsalveti)]

Thanks for testing it, now we have the same fix we have at Natty.

[agraf (agraf)]

I'm seeing this issue with upstream 3.2 on openSUSE. Has there ever been a real solution to the issue?