1678136 - layers.gpu-process.enabled on Windows seems to be required now, should probably ignore the pref

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Tomislav Jovanovic :zombie]

Attachment: Mozilla Firefox 2020-11-18 19-18-28.mp4

My main profile Beta just updated to 84, and now drawing of the UI (and content to a lesser degree) flickers real bad.

See attached video.

It doesn't reproduce with a clean profile, but I managed to mozregression this push log:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=56302c29923559e7ae607cbb654b5e5d060d9329&tochange=8dbba1ce63e803c5a14bd92ce8f49541760a46f6

From that, bug 1595994 looks like an only graphics-related thing, even though it says it's about media playback.

Jean-Yves, can you please take a look?

Comment 1

[Tomislav Jovanovic :zombie]

A few more notes

• it's a windows laptop with i7 and integrated graphics.
• the profile is old, so this could be some pref I changed long ago, though I don't remember changing anything graphics-related recently.
• I think I managed to disable webrender, but the issue still persists.

Comment 2

[Jean-Yves Avenard [:jya]]

Seeing it doesn't reproduce with a clean profile, I'm not sure what we can action here.

And yes, bug 1595994 is only media playback related, it's about decoding content out of process.
Bug 1595994 introduced a regression on non e10s , which was resolved in 1672420. It's the only thing related to graphics that could explain what you've described.

Comment 3

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Bug 1668875 landed shortly before this. Emilio, please check it didn't regress this.

Comment 4

[Tomislav Jovanovic :zombie]

After a lot of trial and error, I found this is caused by the layers.gpu-process.enabled pref. I think I remember disabling this a year+ ago because of some crashes during initial rollout.

It looks like the check for e10s from bug 1672420 is not correct, that should maybe test if there's a GPU process or not.

Or alternatively, the gpu-process pref should be ignored if it is now required for basic browser functionality at all. What do you think Jean-Yves?

Comment 5

[Jean-Yves Avenard [:jya]]

This isn't my area of expertise, I'm not sure why you ni? me sorry.

Comment 6

[Tomislav Jovanovic :zombie]

It looked to me like your patch in bug 1595994 changed firefox to now require a GPU process, or maybe that your patch in bug 1672420 only removed the requirement for the non-e10s case.

I'll just reopen this to let graphics folks see if/how this needs to be addressed.

Comment 7

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1595994

Comment 8

[Sotaro Ikeda [:sotaro]]

I confirmed the problem by disabling GPU process by changing pref layers.gpu-process.enabled=false and with Direct3D 11 (Advanced Layers).

Comment 9

[Sotaro Ikeda [:sotaro]]

It seems that https://phabricator.services.mozilla.com/D91699 of Bug 1595994 seemed to break the problem. And bug 1672420 did not fix the problem.

Comment 10

[Sotaro Ikeda [:sotaro]]

When GPU process does not exist, parent process initialize D3D devices.
https://searchfox.org/mozilla-central/rev/85b84f82489451362a351b144fe5232a8e46c61c/gfx/thebes/gfxWindowsPlatform.cpp#1451

Comment 11

[Sotaro Ikeda [:sotaro]]

Attachment: Bug 1678136 - Lazily initialize SyncObject

Comment 12

[Sotaro Ikeda [:sotaro]]

https://treeherder.mozilla.org/jobs?repo=try&revision=0f1a0053114781f379891e0caaf7728bfd9c4010

Comment 13

[Sotaro Ikeda [:sotaro]]

https://treeherder.mozilla.org/jobs?repo=try&revision=175cf16b49109d800dd1fc3d3282307e0eda8643

Comment 14

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/f13553c2d926
Lazily initialize SyncObject r=mattwoodrow

Comment 15

[Razvan Maries]

https://hg.mozilla.org/mozilla-central/rev/f13553c2d926

Comment 16

[BugBot [:suhaib / :marco/ :calixte]]

The patch landed in nightly and beta is affected.
:sotaro, is this bug important enough to require an uplift?
If not please set status_beta to wontfix.

For more information, please visit auto_nag documentation.

Comment 17

[Sotaro Ikeda [:sotaro]]

Comment on attachment 9189686 [details]
Bug 1678136 - Lazily initialize SyncObject

Beta/Release Uplift Approval Request

• User impact if declined: Rendering has problems when GPU process does not exist. It could happen with pref layers.gpu-process.enabled=false.
• Is this code covered by automated tests?: No
• Has the fix been verified in Nightly?: Yes
• Needs manual test from QE?: No
• If yes, steps to reproduce:
• List of other uplifts needed: None
• Risk to taking this patch: Low
• Why is the change risky/not risky? (and alternatives if risky): It changes only SyncObjectD3D11Client initialization. And its fix was already confirmed on nightly.
• String changes made/needed: None

Comment 18

[Ryan VanderMeulen [:RyanVM]]

Comment on attachment 9189686 [details]
Bug 1678136 - Lazily initialize SyncObject

Approved for 84.0b7.

Comment 19

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-beta/rev/e0600c01785a