blackshard83
Posts: 91
Joined: Fri Jan 10, 2014 8:31 am

Multicore bugs in OMX video components fixed lately?

Fri Jul 08, 2016 8:33 am

Hello,

I get strange, sporadic and unreplicable issues when running videos with a modified omxplayer on raspberry pi 2 models.
Practically the video gets stuck on a frame and audio stops, sometimes the time goes on (given by a OMX Scheduler component), sometimes also the time freezes.

The system is responsive: killing the omxplayer process and running it again restores the normal condition. Apparently there were no VRAM leaks, just the videocore got stuck for an unknown reason. I have also ~35 raspberry pi 1 model B that instead never show this problem.

vcdbg log msg doesn't anything valuable, nor dmesg says anything. Sometimes also rendered frames get corrupted (see the attached jpeg). One interesting thing is that starting the whole process with affinity set to a single core seems to mitigate the problem. Also, as long as the problem happens only on raspberry pi2, I think maybe there is something that doesn't work well with multicore systems.

kernel is 3.18.7 and firmware version is:
Feb 14 2015 22:23:03
Copyright (c) 2012 Broadcom
version 7789db485409720b0e523a3d6b86b12ed56fd152 (clean) (release)

I would ask if any bug of the kind I explained above has been corrected lately, so upgrading the firmware of these production machines would help solve this problem.
Thanks
Attachments
corrupt.jpg
corrupt.jpg (58.64 KiB) Viewed 2446 times

6by9
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 7008
Joined: Wed Dec 04, 2013 11:27 am
Location: ZZ9 Plural Z Alpha, aka just outside Cambridge.

Re: Multicore bugs in OMX video components fixed lately?

Fri Jul 08, 2016 10:37 am

Without knowing what the issue is that you're hitting it's near impossible to say if it is fixed. There are a huge number of changes made over the last 16 months.

The issue that resulted in https://github.com/raspberrypi/linux/co ... 421f20d00a rings bells as fixing spontaneous halts in IL stuff, but I can't find that issue quickly. I'd suspect that change could be cherrypicked back to a 3.18 kernel if you felt like trying it.

Your corrupted image looks like it may be related to https://github.com/raspberrypi/firmware/issues/499 where a bug resulted in wrong reference image being used, but that's pure guesswork.

A bit more information needed on how to reproduce your lockups, or just try it.
Software Engineer at Raspberry Pi Trading. Views expressed are still personal views.
I'm not interested in doing contracts for bespoke functionality - please don't ask.

6by9
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 7008
Joined: Wed Dec 04, 2013 11:27 am
Location: ZZ9 Plural Z Alpha, aka just outside Cambridge.

Re: Multicore bugs in OMX video components fixed lately?

Fri Jul 08, 2016 10:38 am

Software Engineer at Raspberry Pi Trading. Views expressed are still personal views.
I'm not interested in doing contracts for bespoke functionality - please don't ask.

blackshard83
Posts: 91
Joined: Fri Jan 10, 2014 8:31 am

Re: Multicore bugs in OMX video components fixed lately?

Fri Jul 08, 2016 2:48 pm

I really appreciate your answer pointing me to the issues, thank you!

Apparently it looks really similar to https://github.com/raspberrypi/firmware/issues/449, but the OP says that his problem is more severe on pi1 than pi2, instead mine happens only on pi2.
I'll double check the discussions.
My plan is to find a way to replicate the problem with the old firmware, kernel and userland code it is in production on my machines, then I will move to newer software and see what happens. If necessary I will post again here.

Return to “OpenMAX”