1263725 - All Win64 e10s tests are going to permafail when Gecko 48 merges to Aurora

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Ryan VanderMeulen [:RyanVM]]

[Tracking Requested - why for this release]: Mass win64 e10s test failures when Gecko 48 merges to Aurora.

I'm assuming this is sandboxing-related since it's win64 e10s only and there's sandbox-related stuff on those stacks. Does any of our sandboxing code depend on NIGHTLY_BUILD?

https://treeherder.mozilla.org/#/jobs?repo=try&revision=5b3d47649e85&filter-searchStr=windows%20x64%20e10s&selectedJob=19232603

Sample log:
https://treeherder.mozilla.org/logviewer.html#?job_id=19232603&repo=try

Comment 1

[Bob Owen (:bobowen)]

NIGHTLY_BUILD is used to set MOZ_CONTENT_SANDBOX and prefs.

What specifically is the "sandbox-related stuff"?

Comment 2

[Ryan VanderMeulen [:RyanVM]]

Observations from Bob on IRC:
bobowen: RyanVM: https://dxr.mozilla.org/mozilla-central/source/gfx/src/DriverCrashGuard.cpp#61
bobowen: RyanVM: https://hg.mozilla.org/mozilla-central/rev/b5a0c2bf1bca
bobowen: RyanVM: added D3D11Video, but it's not in https://dxr.mozilla.org/mozilla-central/source/dom/ipc/ContentParent.cpp#5650

Try bisection supports Bob's observations.
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=efe7d026ac64&tochange=6202ade0e6d6

Also, https://treeherder.mozilla.org/logviewer.html#?job_id=19232844&repo=try looks like the same issue with a more clear-cut link to DriverCrashGuard.

Comment 3

[David Anderson [:dvander] - inactive, e-mail if emergency]

Attachment: fix

Good catch, thanks Bob.

Comment 4

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/30aa9ea0ad0c

Comment 5

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/30aa9ea0ad0c