sound: Use M_WAITOK where possible
AcceptedPublic
Actions

Authored by christos on Sun, Sep 29, 1:23 PM.

Details

Reviewers

markj
emaste
dev_submerge.ch
zlei

Summary

These sections do not call malloc(9) with a lock held.

Sponsored by: The FreeBSD Foundation
MFC after: 2 days

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 59652
Build 56539: arc lint + arc unit

Event Timeline

christos created this revision.Sun, Sep 29, 1:23 PM

Herald added a subscriber: imp. · View Herald TranscriptSun, Sep 29, 1:23 PM

christos requested review of this revision.Sun, Sep 29, 1:23 PM

Harbormaster completed remote builds in B59652: Diff 143899.Sun, Sep 29, 1:23 PM

I accidentally posted a draft patch, which contained some errors. Will re-submit the correct one soon.

christos reclaimed this revision.Sun, Sep 29, 1:33 PM

christos edited the summary of this revision. (Show Details)

Fix.

Harbormaster completed remote builds in B59653: Diff 143900.Sun, Sep 29, 1:34 PM

christos mentioned this in D46821: sound: Move root feeder initialization to separate function.Sun, Sep 29, 1:34 PM

Looks good to me.

This revision is now accepted and ready to land.Sun, Sep 29, 2:29 PM

LGTM, I suppose this is save when called through SYSINIT?

In D46845#1067932, @dev_submerge.ch wrote:

LGTM, I suppose this is save when called through SYSINIT?

Yes.

In D46845#1067933, @christos wrote:

In D46845#1067932, @dev_submerge.ch wrote:

LGTM, I suppose this is save when called through SYSINIT?

Yes.

The justification in your commit log is a bit incomplete: not holding a lock is not sufficient to sleep (e.g., interrupt threads cannot sleep in general, even when they're not holding locks). The thread context matters too. It's ok to sleep in SYSINITs or in driver attach routines.

markj accepted this revision.Mon, Sep 30, 8:12 AM

The justification in your commit log is a bit incomplete: not holding a lock is not sufficient to sleep (e.g., interrupt threads cannot sleep in general, even when they're not holding locks). The thread context matters too. It's ok to sleep in SYSINITs or in driver attach routines.

Good point - the change allows the code to be simplified, and you can just state that the calls are in contexts where sleeping is permitted.