vt: drop the lock before calling vd_postswitch
AcceptedPublic
Actions

Authored by kevans on Fri, Jan 10, 4:34 AM.

Details

Reviewers

emaste
manu
jhb

Summary

Some postswitch methods, e.g., from i915kms, don't currently expect to
be called under a lock and may try to sleep. The expectation seems OK,
we usually do drop the lock prior to calling vd_postswitch, so let's
avoid an unnecessary panic and drop the lock earlier..

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 61638
Build 58522: arc lint + arc unit

Event Timeline

kevans created this revision.Fri, Jan 10, 4:34 AM

Herald added a subscriber: imp. · View Herald TranscriptFri, Jan 10, 4:34 AM

kevans requested review of this revision.Fri, Jan 10, 4:34 AM

Harbormaster completed remote builds in B61638: Diff 149036.Fri, Jan 10, 4:34 AM

kevans added a child revision: D48414: vt: avoid grabbing locks for KERNEL_PANICKED() as well.Fri, Jan 10, 4:34 AM

jhb accepted this revision.Fri, Jan 10, 12:15 PM

jhb added inline comments.

sys/dev/vt/vt_core.c
625	Notably, we drop the VT lock before postswitch here already. Might be worth calling out that you are fixing a special case to follow the same behavior as the more general case. Probably the special cal to vd_postswitch was added later and that later commit just got the locking wrong.

This revision is now accepted and ready to land.Fri, Jan 10, 12:15 PM

manu accepted this revision.Sat, Jan 11, 10:03 AM

kevans mentioned this in D48414: vt: avoid grabbing locks for KERNEL_PANICKED() as well.Tue, Jan 14, 4:56 AM