-
Notifications
You must be signed in to change notification settings - Fork 5k
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
VC4 freeze in (atomic,non-atomic) KMS on arm64 #4627
Comments
FYI you have a typo in your config.txt:
|
@pelwell Thanks for the correction! I should stop doing edits over ssh from my phone! Unfortunately, I see the same issue with that fix:
|
Can you test with our 5.10 kernel? |
This is indeed not an issue with the 5.10 kernel from here. Any chance of getting this fix upstreamed, or merged with the 5.15 kernel here to confirm that it isn't some other issue? |
@mripard has been investigating but it wasn't trivial to see what was wrong with the upstream code. |
One of the issue is that it's not clear what the fix is exactly :) But as it turns out, I'm working on it at the moment. One of the reason it's been taking some time to address has been that there's several issues intertwined, and it's hard to isolate and fix them. It could also be that there's been a regression in 5.11 that has never been uncovered since it receives less testing than 5.10 can. I've pushed my current fixes rebased on rpi-5.15.y here: If it doesn't fix the issue you're seeing, could you let me know where I can grab/create an image of the system that triggers it? |
Thanks. I'll do a build from that branch and see what happens. |
@mripard building with your current kernel tree
|
It's a bad merge. Try deleting one of the two |
After patching the rpi top of tree 5.15 with the two last commits in @mripard's tree, namely mripard@061eb3b and mripard@72be6d5, and then running
Building your kernel @mripard I get the same error after the system goes to sleep:
Happy to help debug this further. My configuration may be particular because of the edid file I am using in cmdline.txt... |
Can you give me a way to download or create your image so that I can test? |
@mripard Let me see if I can get a reproducer setup based on the Ubuntu Desktop 21.10 image at https://ubuntu.com/download/raspberry-pi |
Actually, I made minor settings changes to the stock Ubuntu 21.10 image and got this issue running on a RPI4B. I think the only relevant change was to enable automatic login (during setup) and then disable automatic screen lock thus: Settings -> Privacy -> Automatic Screen Lock (set to disabled.) I left the blank screen delay at 5 minutes. I then rebooted and waited. Once the screen went to sleep, I got the error.
|
@satmandu You're running Ubuntu and we are working on this: https://bugs.launchpad.net/bugs/1946368. |
@juergh It was silly of me to assume this discussion had not already gone upstream. Thanks all! |
Indeed, I can reproduce it with the setup you provided, I'll try to figure out what's going on. Thanks! |
Oh. You as in me? I was just about to upload an image for you Maxime... I guess you don't need that anymore? |
I meant you as in @satmandu |
For what it is worth. I built a 5.15-rc7 kernel at 3a3fc22 yesterday, and I noticed that the screen shows up correctly at boot, but that playing videos with |
I've noticed kodi doesn't display video for the first few seconds on 5.15-rc7, but then starts and is fine for rest of file. |
Yes I'm using the ubuntu ffplay. Here's what I see running
Turns out that adding
|
Funnily enough my kodi issue seems to be h264 only. hevc and sw decode are okay. Possibly a v4l2 decoder issue. |
@popcornmix It does look like trying to offload h264 decoding causes the problem here. |
Aside from that, 3a3fc22 does appear to have the sleep -> |
Are you able to revert 56c8fad and test? |
Changing line |
I thought/hoped/prayed that that setting only affected input buffers, but I guess I was too hopeful. At this point I'm not sure what the correct answer is. |
Sounds like this was resolved - okay to close? |
Definitely resolved. I haven't seen this in a long time. |
Background:
Mutter 40 (using in Ubuntu 21.10 and beyond) is using atomic KMS, which can be disabled by setting
MUTTER_DEBUG_ENABLE_ATOMIC_KMS=0
.The issue is this: After mutter sets the display to go to sleep, it never comes up again, reporting this error within minutes of going to sleep:
A reboot is needed to resolve this.
This isn't unique to kernel 5.15. Ubuntu is seeing it in their release 5.13 kernels too, but I have verified that this issue still exists by building the current 5.15-rc5 kernel here.
The issue exists with current firmware.
It has been mentioned at https://bugs.launchpad.net/ubuntu/+source/linux-raspi/+bug/1944397/comments/20 and https://bugs.launchpad.net/ubuntu/+source/linux-raspi/+bug/1944397/comments/45 by people with familiarity with KMS far far above mine.
As @vanvugt suggests, this may be a kernel issue:
To reproduce
dtoverlay=vc4-kms-v3d-pi4
is being used.With a kernel newer than 5.11, install mutter 40, and let the display go to sleep. The display can then not be woken.
Expected behaviour
Display should wake again.
Actual behaviour
Display goes blank.
System
Debug output:
gowat-10_13_2021_1435EDT.txt
cat /etc/rpi-issue
)? Ubuntu 21.10vcgencmd version
)?(This is the release firmware for Ubuntu 21.10)
I have also tried this firmware:
uname -a
)?5.15.0-rc5-v8-gc0acaa66d (built from c0acaa6 )
Logs
The text was updated successfully, but these errors were encountered: