Bug 109170 - GPU HANG: ecode 9:0:0x85dffffb (esp. chrome with hardware-acceleration, also elsewhere) - UHD Graphics 620
Summary: GPU HANG: ecode 9:0:0x85dffffb (esp. chrome with hardware-acceleration, also ...
Status: NEEDINFO
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/DRI/i965 (show other bugs)
Version: unspecified
Hardware: x86-64 (AMD64) Linux (All)
: medium normal
Assignee: Intel 3D Bugs Mailing List
QA Contact: Intel 3D Bugs Mailing List
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2018-12-29 12:27 UTC by Christian Lohmaier
Modified: 2019-01-15 08:38 UTC (History)
1 user (show)

See Also:
i915 platform:
i915 features:


Attachments
/sys/class/drm/card0/error , hang triggered by chrome w/hw-accel on youtube (56.12 KB, text/plain)
2018-12-29 12:27 UTC, Christian Lohmaier
Details
/sys/class/drm/card0/error , when running liveiso with chromium 70.0.3538.77 (69.77 KB, text/plain)
2019-01-08 17:00 UTC, Christian Lohmaier
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Christian Lohmaier 2018-12-29 12:27:47 UTC
Created attachment 142908 [details]
/sys/class/drm/card0/error , hang triggered by chrome w/hw-accel on youtube

On a laptop with Intel Core i7-8550U and integrated UHD Graphics 620, I experience frequent hangs when using chrome browser watching youtube or using mpv to playback video.
If I disable use of hardware acceleration in chromium browser or force software decoding/rendering in mpv, the crashes are way less frequent.

With hardware-acceleration enabled I almost immediately get a gpu hang in chromebrowser when visiting youtube (starting a video is guaranteed to trigger)

Symptoms of hang differ from frozen image for a few seconds after which it recovers (although next error then only is a few minutes away, until it reaches a point where it cannot recover anymore), to complete freezes (still image shown, audio loops the last couple seconds, no reaction to keyboard whatsoever, cannot even use magic sysRq anymore) to crashes of Gnome3/wayland (can switch to virtual console and login/trigger reboot)

Distro: Mageia (Cauldron/development)
kernel: 4.19.12
mesa:  18.3.1
chromium-browsser: 70.0.3538.77
Comment 1 Denis 2019-01-03 12:38:06 UTC
Hi, I checked your issue in my configuration (close, but not the same):
- Manjaro 18.0.2
- Gnome desktop
- Linux manjaro-pc 4.19.13-1-MANJARO #1 SMP PREEMPT Sat Dec 29 15:43:56 UTC 2018 x86_64 GNU/Linux
- Mesa DRI Intel(R) UHD Graphics 630 (Coffeelake 3x8 GT2)
- 3.0 Mesa 18.3.1
- Chrome version 71.0.3578.98 (Official Build) (64-bit)
- Chromium	71.0.3578.98 (Official Build) Arch Linux

In both browsers I have these settings:

Graphics Feature Status
Canvas: Hardware accelerated
Flash: Hardware accelerated
Flash Stage3D: Hardware accelerated
Flash Stage3D Baseline profile: Hardware accelerated
Compositing: Hardware accelerated
Multiple Raster Threads: Enabled
Native GpuMemoryBuffers: Software only. Hardware acceleration disabled
Out-of-process Rasterization: Disabled
Hardware Protected Video Decode: Hardware accelerated
Rasterization: Hardware accelerated
Skia Deferred Display List: Disabled
Skia Renderer: Disabled
Surface Control: Disabled
Surface Synchronization: Enabled
Video Decode: Hardware accelerated
Viz Service Display Compositor: Disabled
WebGL: Hardware accelerated
WebGL2: Hardware accelerated

__________

I didn't see any hangs, played full HD videos and some 3d animations (found them from manual, how to enable h/w-accel => https://codepen.io/eltonkamami/full/ECrKd, as example

A bit later I will check the same on GPU the same with your's - UHD620 exactly. Also I see that you are using wayland, in my case I have x11. So I will try with wayland also
Comment 2 Andrii 2019-01-03 16:34:20 UTC
Tried to watch FullHD videos, with plugged external display and with laptop's only, on both X11 and Wayland, but hang wasn't reproduced. Laptop configuration:
________________________________________

- Kubuntu 18.04.1 LTS (Bionic Beaver)
- KDE Plasma desktop
- Intel Corporation UHD Graphics 620
- Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz
- Mesa 18.3.1
- Chrome version 7.1.302.31 (Official Build) (64-bit)
________________________________________

Browser settings:

Canvas: Hardware accelerated
Flash: Hardware accelerated
Flash Stage3D: Hardware accelerated
Flash Stage3D Baseline profile: Hardware accelerated
Compositing: Hardware accelerated
Multiple Raster Threads: Enabled
Native GpuMemoryBuffers: Software only. Hardware acceleration disabled
Out-of-process Rasterization: Disabled
Hardware Protected Video Decode: Hardware accelerated
Rasterization: Hardware accelerated
Skia Deferred Display List: Disabled
Skia Renderer: Disabled
Surface Control: Disabled
Surface Synchronization: Enabled
Video Decode: Hardware accelerated
Viz Service Display Compositor: Disabled
WebGL: Hardware accelerated
WebGL2: Hardware accelerated
________________________________________

Сhristian is this issue stable for you? If so, I will try to setup your distributive.
Comment 3 Christian Lohmaier 2019-01-03 17:47:55 UTC
> Сhristian is this issue stable for you? If so, I will try to setup your
> distributive.

Thanks for trying to reproduce, and yes, the issue is still reproducible 100% despite chromium browser and kernel update in the meantime (although it seems to be able to recover more often until a freezes completely) - but trying to watch a youtube-video with hw-acceleration enabled in chrome settings will trigger rsc hang / will be completely unusable. Same for the particles demo - fine with hw-accel disabled, and triggers the issue with hw-accel enabled.

Very likely you get rsc hang just for restarting chrome after enabling hw-acceleration.
Here's chrome://gpu on my system with hw-accel enabled:

Canvas: Hardware accelerated
Flash: Hardware accelerated
Flash Stage3D: Hardware accelerated
Flash Stage3D Baseline profile: Hardware accelerated
Compositing: Hardware accelerated
Multiple Raster Threads: Enabled
Native GpuMemoryBuffers: Software only. Hardware acceleration disabled
Out-of-process Rasterization: Disabled
Hardware Protected Video Decode: Unavailable
Rasterization: Software only. Hardware acceleration disabled
Skia Deferred Display List: Disabled
Skia Renderer: Disabled
Surface Synchronization: Enabled
Video Decode: Unavailable
Viz Service Display Compositor: Enabled
WebGL: Hardware accelerated
WebGL2: Hardware accelerated

differences to your output is that "Hardware Protected Video Decode" is listed as unavaliable on my system, and "Rasterization" is software only, "Video decode" also is unavailable here.

Thanks for offering to test with same distro as I have - Magiea released a beta1 of the testing version recently, and has live iso (and classical installers) available from https://www.mageia.org/en/downloads/prerelease/ that you can dd to a usb-thumbdrive
Comment 4 Christian Lohmaier 2019-01-03 18:41:27 UTC
quickstart for mageia: after you boot from GNOME live-iso, setup wifi if necessary, then open a terminal, type 
$ su                                    # passwordless in live iso
# urpmi.addmedia --distrib --mirrorlist # setup media repositories
# urpmi chromium-browser                # install chromium browser

That's all - since hw-acceleration is enabled by default in chrome, I get hang when I launch chromium for the first time already (that also triggers gnome-keyring prompt), and trying to playback yt videos triggers more hangs/is unusable.

Also maybe worthwhile to note: the laptop has a QHD+ 3200x1800 resolution / is running at that with the default 200% scaling option in Gnome
Comment 5 Andrii 2019-01-04 22:31:32 UTC
I tried to boot from thumbrive, Mageia 7 beta1, installed Chrome but issue wasn`t reproduced. Mesa wasn`t updated, used build in version from distrib.

Chrome version - Chrome version 7.1.302.31 (Official Build) (64-bit).
Browser settings - same to previous.

Going to try to build newer version of Mesa.
Comment 6 Denis 2019-01-08 09:30:53 UTC
hi Christian, could you please try that image Andriy used?
It has lower chrome and mesa versions as I see. So would be helpful from your side to confirm, that issue is still actual and there (or not)?
It may help in case if it is chrome issue, for example, or if it is something new in mesa, so would be possible to bisect the issue.
Comment 7 Christian Lohmaier 2019-01-08 17:00:10 UTC
Created attachment 143013 [details]
/sys/class/drm/card0/error , when running liveiso with chromium 70.0.3538.77

Sorry if I wasn't clear - I can reproduce in the live ISO as well, see attached file.
The error is nothing recent/it occurs since I have the machine. And it is not limited to chromium browser, but can also be reproduced with mpv and enabled hw-acceleration (but I compiled mpv myself using mpv-build, so using chromium browser as provided by the distro is easiest way to reproduce).

But since you fail to reproduce so far, anything else I could try? Any Windows diagnostics program I can run? Anything else on Linux?
Comment 8 Denis 2019-01-09 08:46:37 UTC
hmm, ok, that's clear. What I can say right now, is that we have quite close HW configuration - with the only 1 exception - your laptop has buit-in 4K monitor, in our case we have regular fullHD monitor AND - external 4K. All other things look the same 8-/

About - what to do next, I see 3 more things to try...
1. Could you please try to make an apitrace of the hang? https://github.com/apitrace/apitrace/wiki/Google-Chrome-Browser
Tricky thing here - chrome should be run in a single-stream, manual higher should help.
From here you can take apitrace https://github.com/apitrace/apitrace/
Who knows, maybe with apitrace we also will reproduce it, or somebody other will try too.

2. Did you try on other distributions? For example ubuntu or manjaro (doesn't matter actually, just - another distro)

3. I am recalling 1 issue, where we had quite similar case - reporter had stable hang, and we couldn't reproduce it on our side. And then he updated bios version, and it helped https://bugs.freedesktop.org/show_bug.cgi?id=107586
I don't say that we have exactly the same case, but - I would check and bios version also (just in case).
Comment 9 Sergii Romantsov 2019-01-09 09:47:38 UTC
Hello, Christian, i wonder if you are able to make an apitrace
You need to install packag apitrace
Kill all processes of google-chrome browser.
And run chrome with exact link that should hang, like:

apitrace trace -o ~/chrome.trace /opt/google/chrome/google-chrome https://www.youtube.com/watch?v=LXb3EKWsInQ

If you got it hung, then, please, recheck with command that it hangs:

glretrace ~/chrome.trace

And give us chrome.trace.
There are seems no guarantee that we will have such hang, but for case.
And have you tried to disable\enable some properties of hw accelerations (maybe some exact one somehow helps...)?
And has it influence if to decrease resolution/scaling?
Comment 10 Andrii 2019-01-09 17:39:51 UTC
Tried to reproduce hang one more time, used video, which was sent by Sergii. Used built in FullHD display and external 4K. Mesa, Chromium and Kernel version are same to Christian's, but hang wasn't reproduced. Current configuration is:
________________________________________

- Mageia 7 beta 1
- GNOME desktop
- Intel Corporation UHD Graphics 620
- Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz
- Mesa 18.3.1
- Chromium version 70.0.3538.77 (64-bit)
________________________________________

Browser settings:

Canvas: Hardware accelerated
Flash: Hardware accelerated
Flash Stage3D: Hardware accelerated
Flash Stage3D Baseline profile: Hardware accelerated
Compositing: Hardware accelerated
Multiple Raster Threads: Enabled
Native GpuMemoryBuffers: Software only. Hardware acceleration disabled
Out-of-process Rasterization: Disabled
Hardware Protected Video Decode: Unavailable
Rasterization: Software only. Hardware acceleration disabled
Skia Deferred Display List: Disabled
Skia Renderer: Disabled
Surface Synchronization: Enabled
Video Decode: Unavailable
Viz Service Display Compositor: Enabled
WebGL: Hardware accelerated
WebGL2: Hardware accelerated
________________________________________
Comment 11 Sergii Romantsov 2019-01-15 08:38:05 UTC
Hello, Christian.
Seems at this moment there is only one way to continue looking on issue: we need an apitrace (see Comment 9 Sergii Romantsov 2019-01-09 09:47:38 UTC)


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.