Closed Bug 1323790 Opened 8 years ago Closed 8 years ago

Linux debug R8 reftest suite leaking objects

Categories

(Core :: Graphics: WebRender, defect, P3)

53 Branch
x86_64
Linux
defect

Tracking

()

RESOLVED INCOMPLETE
Tracking Status
firefox53 --- affected

People

(Reporter: kats, Assigned: kats)

References

Details

(Whiteboard: [gfx-noted])

Attachments

(1 file)

After merging from m-c in https://treeherder.mozilla.org/#/jobs?repo=graphics&revision=05f6c6e2e3d45128c268b6681980a91febabfab2 the Linux-QR R8 debug reftests started showing leaks, in both non-e10s and e10s. It's probably leaking in opt too, but we don't do leakchecking in opt so it's not showing up. I'm doing a local build to chase this down, and will probably do some try pushes to bisect the m-c merge as well.
Attached file Refcount stacks
So after spending half the day debugging this it looks like this isn't even a real problem. The leak only showed up in the TreeHerder snippet because of the test failure that went along with it. The leak is pre-existing - it shows up in the logs of previous green R8 runs but TreeHerder ignores the failure. That being said I did repro this locally and looked at the stacks that added/released refptrs to the leaked WebRenderBridgeChild instance. There were two dangling pointers, from ImageClientSingle::ImageClientSingle and one from TextureClientRecycleAllocator::TextureClientRecycleAllocator. The full stacks along with the refcnt changes are attached.
I'm going to close this for now since it doesn't seem to be an issue on TH.
Status: NEW → RESOLVED
Closed: 8 years ago
Resolution: --- → INCOMPLETE
Looking at the log [1] more closely it looks like we're actually leaking thousands of WebRender*Layer instances too, so we should definitely track this down at some point. [1] https://treeherder.mozilla.org/logviewer.html#?job_id=119871&repo=graphics#L24440
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: