New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Possible VC4 memory leak introduced in 5.10.47/5.10.48 #4474
Comments
I would be suspecting one of these : #4441 |
Sounds likely. Are you able to confirm memory leak starts with that commit? |
@mripard any thoughts on a memory leak introduced with "drm/vc4: kms: Wait on previous FIFO users before a commit"? |
I think that we have the drm_crtc_commit ref counting wrong somehow and thus never free it for each commit. However, I started to run an application that does a commit each frame, and while I can definitely see the kmalloc-128 slab increase over time, I'm not getting the drm_atomic_helper_setup_commit call, at all, which is fairly weird. |
I have been experiencing out of memory issues on a raspberry pi 4, 8Gb RAM, armv7h system, using kernel 5.10.52. In my case, the reversions in pull request #4478 have not solved the issue. Archlinuxarm linux-raspberrypi4 package at version 5.10.52-5 is built from commit 6495fa3, and therefore includes the reversions made in #4478 that are supposed to have solved this issue according to above comment, but still has the oom behaviour on my system. Change occurred somewhere between 2697f74 (which I'm running now with no oom issues) and 86729e7 (when they started, and were not fixed by #4478). At least one other user has reported a similar situation on the archlinuxarm forum: https://archlinuxarm.org/forum/viewtopic.php?f=60&t=15460&p=67115#p67115 It would seem that on aarch64 systems the reversions did indeed fix the problem though. |
@Marcool04: It would be interesting to see if you have the memory leak in the same place, as I don't believe the code which caused this memory leak is reachable (not 100% confident however…). Could you add |
Indeed @kwirk there would appear to be some error on my part. Either this has been resolved between 5.10.52-5 and 5.10.52-6 (but no commit in between seems to indicate it should have), or I have an issue somewhere else, as I am now running the latest archlinuxarm kernel just fine…
This time, the difference is I did not upgrade my AUR packages (which I usually do at the same time as I upgrade with pacman)… Maybe one of |
It looks like 5.10.52-6 don't invoked oom-killer for docker anymore.
|
Describe the bug
After upgrading from 5.10.46 to 5.10.48, I had issues with oom-killer killing Kodi in about 20mins of starting to watch a video. The system then became unstable, with other processes being randomly killed, and a system reboot was required.
To reproduce
Play video on Kodi. Monitor kmalloc-128 increasing via
slabtop -s c
Expected behaviour
Kodi not be oom-killed
Actual behaviour
Kodi oom-killed, along with other processes. System unstable requiring reboot.
System
Copy and paste the results of the raspinfo command in to this section. Alternatively, copy and paste a pastebin link, or add answers to the following questions:
Model: Raspberry Pi 4
OS: Arch Linux ARM
Firmware: Jul 21 2021 16:21:46 version 6a796bb0062a6c75191c57cba1c13f9300076d02
Kernel: 5.10.52 (issue seen since updating to 5.10.48; but may have also been present on slightly earlier version)
Logs
Noted high number of
drm_atomic_helper_setup_commit
:I downgraded to 5.10.46 and this has resolved the issue:
Additional context
Also report on anholt#156
Forum post: https://www.raspberrypi.org/forums/viewtopic.php?f=53&t=316394
The text was updated successfully, but these errors were encountered: