Page MenuHome
Create Task
Maniphest T87229

Random hard crashes
Closed, DuplicatePublic
Actions

Description

System Information
Operating system: Linux-5.11.10-100.fc32.x86_64-x86_64-with-glibc2.29 64 Bits
Graphics card: Mesa Intel(R) HD Graphics 620 (KBL GT2) Intel 4.6 (Core Profile) Mesa 20.2.3

Blender Version
Broken: version: 2.83.12, branch: unknown, commit date: 1970-01-01 00:00, hash: rBunknown. 2.83.9 had the same problems, have not tested earlier 2.83 releases.
Worked: blender-2.82a-1

Short description of error
Blender crashes, and takes my whole system down with it. The problem is so severe that closing blender (pkill -9 blender) or restarting the window manager (i3wm) does not resolve it. The problem can only be resolved by a reboot.

Exact steps for others to reproduce the error
Run blender 2.83.12 on Fedora 32 with i3wm (no compositor). Start an empty project, use blender for a while. A random system-crippling crash will happen after between 1 min to 30 min of usage. The crashes seem unrelated to the things I am doing with blender (rotating the default cube could be enough).

Please note that a crash like this is unique for me. Other complex software (Unity, older versions of blender, gimp, firefox) run smoothly. Apart from Blender 2.83, I have never experienced crashes that required a reboot, for any software, in many years of linux usage. I realize that this is a hard one to debug, just hoping my report can help. I'm not sure where I can find any log files after a crash (and submitting those could be challenging with my whole system going down). Thanks :)

Related Objects

Event Timeline

Abe Wits (melonmousegames) created this task.Tue, Apr 6, 9:54 AM
Abe Wits (melonmousegames) updated the task description. (Show Details)
Henrik Dick (weasel) added a subscriber: Henrik Dick (weasel).Tue, Apr 6, 11:34 AM

please check if this is a duplicate of T80458.

Henrik Dick (weasel) changed the task status from Needs Triage to Needs Information from User.Tue, Apr 6, 1:14 PM
Abe Wits (melonmousegames) changed the task status from Needs Information from User to Needs Triage.Tue, Apr 6, 1:52 PM

Testing with gpuhangtest.blend (file from T80458) on 2.82 (sub 7), I do get a hard crash.

Here is the relevant journalctl output. Does that look consistent with T80458?

blender_gpu_crash.log4 KBDownload

However, in day to day use, blender 2.82 has been much more stable for me than blender 2.83. Is that consistent with the experiences in T80458? If not, is it possible that there are multiple bugs at play here?

Henrik Dick (weasel) added a comment.Tue, Apr 6, 6:17 PM

The log looks like its a gpu hang with the intel drivers much like in T80458. I think the issue was raised around the time of 2.82 or 2.83 release, so I think it roughly matches the experience there about 2.82 being more stable. I also remember using 2.82 on my laptop completely stable. I can't say though if it was the driver that was available then, or the blender version, which made it stable. Also, there are definitly multiple bugs at play here! The random crashes look the same as the crash that people are getting with that file in T80458, but it might be something different, as there is no obvious high gpu load on those random crashes, while the gpu hang with the file from T80458 is likely caused by a too big render call.

Abe Wits (melonmousegames) added a comment.Thu, Apr 8, 10:11 AM

Thanks so much Hendrik!

Since the random crashes with low gpu load are hard to reproduce, we could close this bug until T80458 is fixed. Any idea how to figure out what code causes these random crashes? Some options:

  1. git bisect on blender - this is hard because I guess we'd need to [build + test until a crash happens] automatically
  2. run a debug version of blender + try to get a stack trace - this is hard because the bug seems to show itself as an interaction between the driver and blender
  3. run a debug version of blender + do very verbose logging of anything driver related - the hard part is looking through the logs to find what is going wrong
Germano Cavalcante (mano-wii) added a subscriber: Germano Cavalcante (mano-wii).Thu, Apr 8, 5:51 PM

Since the problem is very similar to the T80458, I will merge the reports.
If that problem is solved and this one is not, this report can be reopened as a different issue.

Germano Cavalcante (mano-wii) closed this task as a duplicate of T80458: Reliable GPU hang on common Intel cards (Linux).Thu, Apr 8, 5:51 PM