acpi: Make taskqueue only run on BSP
ClosedPublic
Actions

Authored by obiwac on Mon, Dec 29, 8:45 PM.

Details

Reviewers

olce
jkim
kib

Commits

rGc0df8f6f0e6a: acpi: Make taskqueue only run on BSP

Summary

This is needed because when we add the suspend-to-idle loop, we only want to break the BSP out of idle to process the taskqueue while keeping all the others idled.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

obiwac created this revision.Mon, Dec 29, 8:45 PM

Herald added a subscriber: imp. · View Herald TranscriptMon, Dec 29, 8:46 PM

obiwac requested review of this revision.Mon, Dec 29, 8:46 PM

Harbormaster completed remote builds in B69510: Diff 168739.Mon, Dec 29, 8:46 PM

Simplify cpuset initialization.

Harbormaster completed remote builds in B69516: Diff 168745.Mon, Dec 29, 9:44 PM

emaste added a subscriber: kib.Tue, Jan 6, 4:47 PM

kib added inline comments.Wed, Jan 7, 2:14 AM

sys/dev/acpica/Osd/OsdSchedule.c
44	Includes should be ordered alphabetically (except sys/param.h, sys/systm.h)
113	There is no MI guarantee that BSP is CPU with cpuid 0. I suggest to add MD KPI like GET_BSP_CPUID() and use it there. It is also a good moment to make IS_BSP() available on all arches, not only x86. If you need help, I might prototype the patch.

Reorder includes.

Harbormaster completed remote builds in B69680: Diff 169235.Wed, Jan 7, 12:35 PM

obiwac added a child revision: D54410: acpi: Implement s2idle loop.Wed, Jan 7, 12:38 PM

obiwac marked an inline comment as done.Wed, Jan 7, 1:00 PM

obiwac added inline comments.

sys/dev/acpica/Osd/OsdSchedule.c
113	Sure thing. Currently digging a little deeper for getting BSP on non-x86 arches. There probably are some other places we'll need to change this in our ACPI code as at least the shutdown code seems to assume BSP is cpuid 0. I also see the HPET device is binding its interrupt to the BSP by calling `CPU_FIRST()` - is this guaranteed to be the BSP? In which case we wouldn't need a `GET_BSP_CPUID()` I guess and we could make `IS_BSP()` MI easily.

obiwac edited the summary of this revision. (Show Details)Wed, Jan 7, 1:07 PM

obiwac retitled this revision from acpi: Make taskqueue only run on CPU0 to acpi: Make taskqueue only run on BSP.

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

In fact, if you plan to work on proper get_bsp() etc I think it would be useful to commit this change as is now since it eliminates the real issue. Then you would fix it in due course of the the bsp changes.

Might be, mark the cpuset calculation with XXX comment.

sys/dev/acpica/Osd/OsdSchedule.c
113	No, CPU_FIRST() is not guaranteed to be BSP, there is no guarantee for it at all. Look at the implementation. It happens so far that on x86 BSP == CPUID0 == First cpu since all CPUIDs are dense. But this is only so happen.