| 2017-08-11 03:18:05 |
Andrew Montalenti |
description |
I am running a Lenovo X1C 4th Generation on Ubuntu 17.04. With the stock xserver-xorg-video-intel, I am able to reliably reproduce a complete Xorg hang -- that occasionally results in a full system hang, requiring reboot -- simply by opening a 360 photograph (using WebGL under the hood) on Facebook.com running in Chrome with hardware acceleration enabled. This is using modesetting and thus glamor.
A similar hang also happens with this driver using certain software leveraging the GPU or hardware acceleration, for example the proprietary Zoom Video app. But since the Facebook example works very reliably (it crashes without fail), it has been good for testing/reproduction.
On a stock Ubuntu 17.04, this crash would result in a total system hang requiring a reboot. However, by enabling CTRL+ALT+BACKSPACE on my Xorg version using GNOME tweak, I was able to induce the crash, press CTRL+ALT+BACKSPACE, and end up in a virtual terminal. From there, I could inspect dmesg (to find the message reported in the summary, which has also been reported elsewhere on bug trackers in various forms). This is the more precise dmesg entry:
GPU HANG: ecode 9:0:0x86dffffd, in chrome [24053], reason: Hang on render ring, action: reset
I could also capture the state of the GPU/drm error by looking in /sys/class/drm/card0/error. That has been attached to this ticket.
Some interesting notes about this:
- I could make Chrome not crash by turning off hardware acceleration in its advanced settings. In this case, chrome://gpu would show it is no longer using hardware acceleration. It would thus software accelerate WebGL on Facebook.com. And thus, no hang/crash.
- I upgraded to the xserver-xorg-video-intel version that is in this PPA: https://launchpad.net/~oibaf/+archive/ubuntu/graphics-drivers -- after this, the problem went away, definitively. I have only been running this version for a day, but have had no crashes, and the Facebook WebGL reproduction case is no longer a reproduction case.
My suspicion is that this is a bug deep in the intel driver and its interaction with certain GPUs, since the hang has happened in more than one userland program, and |
I am running a Lenovo X1C 4th Generation on Ubuntu 17.04. With the stock xserver-xorg-video-intel, I am able to reliably reproduce a complete Xorg hang -- that occasionally results in a full system hang, requiring reboot -- simply by opening a 360 photograph (using WebGL under the hood) on Facebook.com running in Chrome with hardware acceleration enabled. This is using modesetting and thus glamor.
A similar hang also happens with this driver using certain software leveraging the GPU or hardware acceleration, for example the proprietary Zoom Video app. But since the Facebook example works very reliably (it crashes without fail), it has been good for testing/reproduction.
On a stock Ubuntu 17.04, this crash would result in a total system hang requiring a reboot. However, by enabling CTRL+ALT+BACKSPACE on my Xorg version using GNOME tweak, I was able to induce the crash, press CTRL+ALT+BACKSPACE, and end up in a virtual terminal. From there, I could inspect dmesg (to find the message reported in the summary, which has also been reported elsewhere on bug trackers in various forms). This is the more precise dmesg entry:
GPU HANG: ecode 9:0:0x86dffffd, in chrome [24053], reason: Hang on render ring, action: reset
I could also capture the state of the GPU/drm error by looking in /sys/class/drm/card0/error. That has been attached to this ticket.
Some interesting notes about this:
- I could make Chrome not crash by turning off hardware acceleration in its advanced settings. In this case, chrome://gpu would show it is no longer using hardware acceleration. It would thus software accelerate WebGL on Facebook.com. And thus, no hang/crash.
- I upgraded to the xserver-xorg-video-intel version that is in this PPA: https://launchpad.net/~oibaf/+archive/ubuntu/graphics-drivers -- after this, the problem went away, definitively. I have only been running this version for a day, but have had no crashes, and the Facebook WebGL reproduction case is no longer a reproduction case.
My suspicion is that this is a bug deep in the intel driver and its interaction with certain GPUs, since the hang has happened in more than one userland program, and seems to be "repaired" by recent updates in the above-linked PPA. |
|
