1943874 - [wayland] Wayland events don't stop even when the window is in the background

Status: RESOLVED DUPLICATE of bug 1940924

(Core :: Widget: Gtk, defect)

Description

[Emilio Cobos Álvarez (:emilio)]

[Tracking Requested - why for this release]: This affects power usage negatively.

STR (I tested on KDE plasma):

• Run Firefox with WAYLAND_DEBUG=1, navigate to about:blank. Make sure urlbar doesn't have focus so nothing moves.

ER:

• Log spew stops after a while.

AR:

• Wayland events keep happening at full speed, even when the window is in the background.

Mozregression points to bug 1934497. Martin, can you take a look?

Comment 1

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

Set release status flags based on info from the regressing bug 1934497

Comment 2

[Shmerl]

Could it be the same as #1879271? See https://bugzilla.mozilla.org/show_bug.cgi?id=1879271#c15

Comment 3

[Emilio Cobos Álvarez (:emilio)]

This seems to be because WaylandVSyncSource::Init always adds a persistent frame callback so we just request frame callbacks over and over ad infinitum. Which is expected, I think, but we seem to still be doing more work than we used to.

Comment 4

[Shmerl]

It doesn't look like it should be happening though in general? I.e. GPU shouldn't be constantly loaded if the window is not actually visible.

Comment 5

[Shmerl]

But even if visible, it doesn't sound right to have an increased GPU usage for basically no gain?

Comment 6

[Emilio Cobos Álvarez (:emilio)]

(In reply to Shmerl from comment #2)

Could it be the same as #1879271? See https://bugzilla.mozilla.org/show_bug.cgi?id=1879271#c15

It seems maybe related but at least not quite the same. You're seeing something in 134, while the regressor here landed in 135. Still, linking the bugs for convenience.

(In reply to Shmerl from comment #5)

But even if visible, it doesn't sound right to have an increased GPU usage for basically no gain?

Requesting the frame callback shouldn't increase GPU usage per se, it's just a signal from the compositor. Ideally we do very little or no work off it if nothing is changing on the screen tho, and that seems to have regressed.

Comment 7

[Emilio Cobos Álvarez (:emilio)]

It seems WaylandVSyncSource::DisableVSync should actually remove the persistent callback, or something.

Comment 8

[Martin Stránský [:stransky] (ni? me) (PTO, back on Feb 17)]

Sure, it's dupe of Bug 1940924. Will look at it ASAP.