[Bug 103076] [v4.13 ARCH] GPU HANG: DMAR: DRHD: handling fault status reg 3 (arch reverted the use of intel_iommu=igfx_off)

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=103076

--- Comment #19 from Carsten Mattner <carstenmattner at gmail.com> ---
(In reply to Ansgar Hegerfeld from comment #16)
> Maybe we should close this bug as a duplicate of
> https://bugs.freedesktop.org/show_bug.cgi?id=89360 or did I miss something?
> The workaround "intel_iommu=igfx_off" works for me using Arch Linux, too.

I don't think the problem is fixed. I'm on Sandybridge and have tested
various kernels and configs.

4.9-LTS    OK

4.4-LTS    KINDA-OK because it has atomic modesetting errors
           that got introduced in 4.2 and haven't been fixed
           until 4.9-LTS (or earlier, don't have kernel in-between
           which is maintained). This would be a nice kernel because
           it will be supported until 2020 IIUC, but the new atomic
           modesetting code is buggier than in 4.9.

4.12       OK but EOL already

4.13       BAD. most problematic drm version:
On my Sandybridge machine I've disabled IOMMU in the BIOS and also added
intel_iommu=igfx_off, on top off the 4.13.5+ kernel having
CONFIG_INTEL_IOMMU_DEFAULT_ON not set anymore. Even though it's harder
to trigger now in 4.13, I can still provoke GPU errors not present in
either 4.12 or 4.9. I've been successfully using 4.9 for more than a
day with heavy GPU and CPU utilization and haven't hit the same errors
as in 4.13.

drm-tip from a week ago   No improvement over 4.13.4

I'm able to hit errors in 4.13 by running ffmpeg to encode a video,
utilizing vaapi for decoding the input stream, using the cpu cores
for encoding, and then starting a second VAAPI client or a browser
with a compositor process like Firefox or Chrome. If I just run
ffmpeg as the sole VAAPI client and no browser or mpv with vaapi
decoding, there are no hangs. The minute I fire up a video to
watch via vaapi and rendering with OpenGL or use Firefox/Chrome,
there's a GPU hang with reset.

Firefox:
[drm] GPU HANG: ecode 6:0:0x80202f7b, in Compositor [2620], reason: Hang on
rcs0, action: reset
drm/i915: Resetting chip after gpu hang

Chrome:
drm/i915: Resetting chip after gpu hang
asynchronous wait on fence i915:[global]:a4255 timed out
drm/i915: Resetting chip after gpu hang


Summary: 4.9 stable for days, 4.4 not good, 4.12 good, 4.13 very bad.

-- 
You are receiving this mail because:
You are the assignee for the bug.
You are on the CC list for the bug.
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-gfx-bugs/attachments/20171012/94676cde/attachment-0001.html>