Kernel 5.7.8 GPU hang when using OpenGL

When starting a program using OpenGL (ie. glxgears or mpv) the screen hangs on the first frame. After killing the programm the screen flashes between two old contents. To get a working output I have to restart Xorg, but using OpenGL still fails.

This bug is always reproduceable with glxgears, which i used for this report.

-- system architecture: x86_64

-- kernel version: 5.7.8

-- Linux distribution: debian unstable (xorg: 1:7.7+20, xserver-xorg-video-intel: 2:2.99.917+git20200226-1, mesa: 20.1.2-1)

-- Machine or mother board model: ASRock H97 Anniversary, Intel HD Grafik (Haswell, Pentium(R) CPU G3450)

-- Display connector: HDMI

Kernel log (GPU hang at 10:43:09) kern-10_43_09.log.bz2

GPU Crash dump card0-glxgears.error.bz2

OpenGL works without problems on Linux 5.6.9 on the same system.

I could bisect the bug one commit in the kernel:

commit 47f8253d2b8947d79fd3196bf96c1959c0f25f20 Author: Prathap Kumar Valsan prathap.kumar.valsan@intel.com Date: Fri Mar 6 00:09:57 2020 +0000

drm/i915/gen7: Clear all EU/L3 residual contexts

Reverting this commit fixes the hang in Linux 5.7.8.

Additional, on drm-tip HEAD (commit 6802049b80a49f5f45c2bc2dd3e6d189204dc2bb) glxgears does not crash the GPU (no GPU error is recorded) but the graphics output still hangs and does not recover after killing glxgears remotely - only the mouse pointer moves.

I still have attached the kern.log in case it might be useful; I started glxgears at 14:53:00.

kern.log.bz2

I rebuild drm-tip HEAD (commit 9780545cd4109baff8c6eb1cb1060a29b7ab919f ) with the patch from @ickle in issue #2024 (closed) (disablbing the error check from gen7_ctx_switch_bb_init()) and this also fixes the GPU hang. So this might be the same bug as in issue #2024 (closed).

Update with latest drm-tip as of Wed Aug 19 13:52:30 2020 +0000 commit 03a59fdcea84502e199bf7fa55fd117dcd161dab:

Graphics completely hangs after starting glxgears, only mouse pointer moves. Killing glxgears does not help.

GPU crash dump: error.bz2 kern.log: kern.log.bz2

I also attached the output of intel_reg_checker before intel_reg_checker.before-hang.txt.bz2 and after intel_reg_checker.after-hang.txt.bz2 the GPU crash.

Disabling the gen7_ctx_switch_bb_init() as in #2024 (closed) fixes the GPU crash—I'm currently running the drm-tip kernel with this patch without problems:

diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
index a3b10f3c83eb..324f909fcd30 100644
--- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
@@ -1281,7 +1281,7 @@ int intel_ring_submission_setup(struct intel_engine_cs *engine)
 
        GEM_BUG_ON(timeline->hwsp_ggtt != engine->status_page.vma);
 
-       if (IS_HASWELL(engine->i915) && engine->class == RENDER_CLASS) {
+       if (IS_HASWELL(engine->i915) && engine->class == RENDER_CLASS && 0) {
                err = gen7_ctx_switch_bb_init(engine);
                if (err)
                        goto err_ring_unpin;

Update with latest drm-tip as of Wed Oct 7 18:20:42 2020 -0300 commit 536bc47c749da99776bf4080031390299798015b "drm-tip: 2020y-10m-07d-21h-19m-13s UTC integration manifest":

Still not fixed. Current state:

• After booting into xfce4 (with composing enabled in xfwm4) and starting glxgears, the graphics hang and start slowly switching between 2 states before the start of glxgears. "GPU HANG: ecode 7:1:8edcfc79, in glxgears" in kern.log

• card0.error.bz2

• kern.log.bz2

• Linux distribution: debian unstable (xserver-xorg-core: 2:1.20.9-2, xserver-xorg-video-intel: 2:2.99.917+git20200714-1, mesa: 20.1.9-1)

• drm2177-suggested-fix.diff Workaround suggested by #2024 (closed) with module parameter to disable it on bootup. I can use all programs utilizing drm with this patch.

mentioned in issue #2024 (closed)

@ickle posted a fix for 5.7+ here: https://patchwork.freedesktop.org/patch/395580/?series=82783&rev=1

I can confirm that this patch fixes my GPU HANG, too.

Hi Reporter, can you try to reproduce this issue on latest drm-tip.

https://cgit.freedesktop.org/drm-tip

Please post your observations on it and any steps followed to reproduce the issue.

If fix for 5.7+ here: https://patchwork.freedesktop.org/patch/395580/?series=82783&rev=1 serves your purpose, please close the ticket.

The issue is not fixed on latest drm-tip (b3d976d7d7851bcfcd64b5d6c29672a9e2462442 drm-tip: 2020y-11m-27d-15h-48m-06s UTC integration manifest)

After starting glxgears the first frame of the demo is displayed and the whole screen hangs, after a few seconds it alternates between two different frames like it did since the first post. I have attached a recent kern.log.bz2 and error.bz2.

The patch from https://patchwork.freedesktop.org/patch/395580/?series=82783&rev=1 works at least on vanilla 5.9.1, 5.9.7, 5.9.9 and 5.9.11, but it is NOT included in drm-tip, so I prefer to keep this ticket open.

assigned to @tejaskux

unassigned @tejaskux

marked this issue as a duplicate of #2024 (closed)

closed

marked this issue as related to #2024 (closed)