With ffvpx enabled, system freeze/crash/hang on scrolling through YT shorts for a long time
Categories
(Core :: Audio/Video: Playback, defect, P2)
Tracking
()
Tracking | Status | |
---|---|---|
firefox-esr115 | --- | unaffected |
firefox-esr128 | --- | unaffected |
firefox-esr140 | --- | disabled |
firefox137 | --- | disabled |
firefox138 | --- | disabled |
firefox139 | --- | disabled |
firefox140 | --- | disabled |
firefox141 | --- | disabled |
firefox142 | --- | disabled |
firefox143 | --- | fixed |
People
(Reporter: mayankleoboy1, Assigned: alwu)
References
(Blocks 3 open bugs, Regression)
Details
(Keywords: regression)
Crash Data
Attachments
(4 files)
With ffvpx enabled, i was seeing some crashes/OOMs/system freezes. So i gave chatgpt some prompts which (i think) simulates that using a local VP9 file.
And I started getting crashes.
STR:
- Set media.ffvpx-hw.enabled = TRUE if not already enabled.
- download the VP9 big-buck-bunny video from here: https://test-videos.co.uk/vids/bigbuckbunny/webm/vp9/1080/Big_Buck_Bunny_1080_10s_30MB.webm
- Open the attached testcase.
- Browse to the VP9 video
- Let the testcase run for 1-2 minutes.
AR: crash/OOM/system freeze
ER: Not so
https://crash-stats.mozilla.org/report/index/d3c96719-77c8-479d-9177-aaa870250404
https://crash-stats.mozilla.org/report/index/94df6704-5d42-4730-b3ea-e9f220250404
Repros with media.wmf.zero-copy-nv12-textures enabled/disabled.
Repros with gfx.direct3d11.reuse-decoder-device enabled/disabled.
The other thing i noticed was that while the testcase is running, (with ffvpx enabled) Microsoft Defender is running at continuous full speed (i.e. using 1 core 100% for the full time).
Reporter | ||
Updated•5 months ago
|
Reporter | ||
Comment 1•5 months ago
|
||
Comment 2•5 months ago
|
||
Set release status flags based on info from the regressing bug 1932772
Comment hidden (obsolete) |
Reporter | ||
Comment 5•5 months ago
|
||
Reporter | ||
Updated•5 months ago
|
Updated•5 months ago
|
Updated•4 months ago
|
Updated•4 months ago
|
Updated•4 months ago
|
Updated•4 months ago
|
Updated•3 months ago
|
Updated•3 months ago
|
Reporter | ||
Comment 6•3 months ago
|
||
Fwiw, i have manually disabled ffvpx on my nightly profile due to this issue. There is a reproducible testcase here to help with diagnosis.
Assignee | ||
Comment 7•3 months ago
|
||
Thanks! I should be able to get back to this shortly—I've been tied up with other tasks recently. I'm hoping the fix for bug 1952340 will also address this issue.
Comment 8•2 months ago
•
|
||
Here's a profile showing 9GB allocated and retained at d3d11va_frames_init
: https://share.firefox.dev/40np7SB
Assignee | ||
Comment 9•2 months ago
|
||
Okay I finally found the leak, the memory is leaking due to this, an additional reference will be added via QueryInterface()
everytime when creating an image, and those ref count would never be decreased.
Assignee | ||
Comment 10•2 months ago
|
||
A successful QueryInterface call increments the reference count, and failure to
properly decrement it leads to memory leaks in the GPU process.
Comment 11•2 months ago
|
||
Comment 12•2 months ago
|
||
bugherder |
Assignee | ||
Comment 13•2 months ago
|
||
Hi Mayank,
Would you mind helping me verify whether this issue is resolved on your end with the latest Nightly? Thanks!
Reporter | ||
Comment 14•2 months ago
|
||
yes, this is fixed. I tested and it is working great. Forgot to update here.
Reporter | ||
Updated•2 months ago
|
Updated•2 months ago
|
Updated•2 months ago
|
Updated•27 days ago
|
Description
•