1425788 - Enable AudioIPC on macOS

Status: RESOLVED FIXED

(Core :: Audio/Video: cubeb, enhancement, P2)

Description

[Matthew Gregan [:kinetik]]

This bug tracks enabling libcubeb remoting/AudioIPC support on macOS.

Comment 1

[Matthew Gregan [:kinetik]]

The existing sandboxing on macOS prevents remoting from working as-is because AudioIPC relies on establishing the initial server connection via a Unix domain socket in the temp directory.  Bug 1405877 addresses this by moving AudioIPC to a model where the server connection is established in the chrome process and then passed to the requesting content process via fd passing in Gecko's IPC.

Comment 2

[Matthew Gregan [:kinetik]]

Attachment: Enable AudioIPC on macOS. r?kamidphish

Comment 3

[Alex Gaynor [:Alex_Gaynor]]

Comment on attachment 8937384 [details] [diff] [review]
Enable AudioIPC on macOS.  r?kamidphish

Review of attachment 8937384 [details] [diff] [review]:
-----------------------------------------------------------------

I didn't realize the remoting was basically ready to go on macOS, this is great!

Once this lands, are we good to lock down the sandbox policy for audio on macOS, or would you prefer we wait a bit for this to bake?

::: toolkit/library/rust/gkrust-features.mozbuild
@@ +22,5 @@
>  
>  if CONFIG['MOZ_RUST_SIMD']:
>      gkrust_features += ['simd-accel']
>  
>  # This feature is only supported on Linux and this check needs to

This comment needs to be updated.

Comment 4

[Matthew Gregan [:kinetik]]

(In reply to Alex Gaynor [:Alex_Gaynor] from comment #3)
> I didn't realize the remoting was basically ready to go on macOS, this is
> great!
> 
> Once this lands, are we good to lock down the sandbox policy for audio on
> macOS, or would you prefer we wait a bit for this to bake?

We'd prefer to wait a bit.  The macOS audio backend has much tighter latency (and other) constraints than the Linux/PulseAudio backend, so there's a reasonable chance of issues.  At this stage we're enabling this in nightly to shake it down, there's a chance we'll have to pref it off pretty quickly if there's breakage.

> >  # This feature is only supported on Linux and this check needs to
> 
> This comment needs to be updated.

Good catch, thanks.

Comment 5

[Alex Gaynor [:Alex_Gaynor]]

Great, thanks for the insights.

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/bded93a4777e
Enable audio remoting (AudioIPC for cubeb) on macOS.  r=kamidphish

Comment 7

[Natalia Csoregi [:nataliaCs]]

https://hg.mozilla.org/mozilla-central/rev/bded93a4777e

Comment 8

[Byron Campen [:bwc]]

This has degraded audio quality pretty badly, leaving webrtc calls nearly unusable on OS X.

Comment 9

[Matthew Gregan [:kinetik]]

Per bug 1429847 comment 6, we'll disable remoting while we investigate solutions for the WebRTC issues.

Comment 10

[Matthew Gregan [:kinetik]]

Attachment: bug1425788_disable.patch

Comment 11

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/89c02d32a35c
Disable AudioIPC on macOS while investigating fallout.  r=kamidphish

Comment 12

[Cosmin Sabou [:CosminS]]

https://hg.mozilla.org/mozilla-central/rev/89c02d32a35c

Comment 13

[Matthew Gregan [:kinetik]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=5a1047c8a400ddd47456ff25b8c176a10013452e

Comment 14

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

The leave-open keyword is there and there is no activity for 6 months.
:kinetik, maybe it's time to close this bug?

Comment 15

[Paul Adenot (:padenot)]

I'm working on having another go at this.

Comment 16

[Matthew Gregan [:kinetik]]

(In reply to Release mgmt bot [:sylvestre / :calixte / :marco for bugbug] from comment #14)

The leave-open keyword is there and there is no activity for 6 months.
:kinetik, maybe it's time to close this bug?

There have been several fixes to resolve the issues that resulted in the original backout, so this bug will be landed again. I've cleared leave-open and marked it as ASSIGNED for now.

Comment 17

[Paul Adenot (:padenot)]

I see a number of failures, even with two threads, but I can repro locally so I'll fix soon. Hopefully it's all the same bug.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=44433c9e7372e1b5ba6740be0cf23f473a75c061

Comment 18

[Matthew Gregan [:kinetik]]

Attachment: GitHub Pull Request

Comment 19

[Matthew Gregan [:kinetik]]

Comment on attachment 9253101 [details] [review]
GitHub Pull Request

Oops, intended to attach this to bug 1726275.

Comment 20

[Matthew Gregan [:kinetik]]

Attachment: Bug 1425788 - Enable AudioIPC for macOS. r?chunmin

Comment 21

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/153a98aa1de7
Enable AudioIPC for macOS. r=chunmin

Comment 22

[Norisz Fay [:noriszfay]]

https://hg.mozilla.org/mozilla-central/rev/153a98aa1de7

Comment 23

[Alex Finder]

== Change summary for alert #32848 (as of Thu, 30 Dec 2021 17:18:31 GMT) ==

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
4%	webaudio	macosx1014-64-shippable-qr	webrender	126.17 -> 121.42

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=32848

Comment 24

[Brindusa Tot, DTE]

Considering this feature was disable, based on bug 1750938, shoudn't we reopen this bug and mark it affected on Fx 97 and Fx 98?

Comment 25

[Haik Aftandilian [:haik]]

The continued efforts on this are appreciated. It's important for our sandbox on Mac because it makes us need to allow microphone access in content processes. When the remoting pref is turned on, the sandbox rules don't allow access to the microphone and other audio services. In case it changes the strategy with the implementation, the microphone is the most important restriction and removing access to it is the main benefit driving this with respect to security.

Comment 26

[Matthew Gregan [:kinetik]]

Bug 1817043 finally enabled AudioIPC in macOS for Firefox 113. I'll mark this as resolved.

Comment 27

[Jim Mathies [:jimm]]

\o/ !