x86: New (un)idle IPIs for setting idle flag in scheduler
AbandonedPublic
Actions

Authored by obiwac on Dec 29 2025, 8:59 PM.

Details

Reviewers

olce
markj

Summary

For s2idle, we need a reliable way to force the scheduler into an idle loop. This adds IPIs we can send to specific CPU sets to idle or unidle them.

If I were 100% sure the CPUs wouldn't get any other kinds of IPIs during the s2idle loop, we could probably do away with the extra unidle IPI and just "kick" the CPUs with a preempt or something to clear their scheduler idle flags. But for now this guarantees things will work.

This is a bit heavy-handed and not necessarily the right approach - revision opened mostly so @olce can have a look :)

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 69513
Build 66396: arc lint + arc unit

Event Timeline

obiwac created this revision.Dec 29 2025, 8:59 PM

Herald added a subscriber: imp. · View Herald TranscriptDec 29 2025, 8:59 PM

obiwac requested review of this revision.Dec 29 2025, 8:59 PM

Harbormaster completed remote builds in B69513: Diff 168742.Dec 29 2025, 8:59 PM

obiwac added reviewers: olce, markj.Dec 29 2025, 9:00 PM

obiwac added a parent revision: D54407: sched: Flag to force scheduler to always choose idle thread.

obiwac added a child revision: D54410: acpi: Implement s2idle loop.Dec 29 2025, 9:04 PM

emaste added a subscriber: emaste.Jan 6 2026, 4:49 PM

obiwac planned changes to this revision.Jan 14 2026, 4:22 PM

Can totally use smp_rendezvous_cpus() for this instead of adding 2 new IPIs. I thought I had run into a problem using rendezvous when I first implemented this, but I just tried it again and see no problem doing it this way.

In D54409#1250752, @obiwac wrote:

Can totally use smp_rendezvous_cpus() for this instead of adding 2 new IPIs. I thought I had run into a problem using rendezvous when I first implemented this, but I just tried it again and see no problem doing it this way.

So this is as we discussed. Great.