acpi: Tasks: Precisely report the maximum number of slots used
Needs ReviewPublic
Actions

Authored by olce on Mon, Oct 5, 5:04 PM.

Details

Reviewers

obiwac
jkim
emaste

Summary

To this end, fix races when updating 'acpi_tasks_hiwater'.

In particular, one thread could enqueue first at a lower but
still-greater-than-the-current-maximum index, another thread then
enqueuing at some higher index, and then the first thread would update
'acpi_tasks_hiwater' while the second was exactly after the 'if (i >
acpi_tasks_hiwater)' but just before attempting the atomic_cmpset_int()
that updates 'acpi_task_hiwater', which then would fail (and the return
value was ignored), leading to the transient maximum not being recorded.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 77729
Build 74612: arc lint + arc unit

Event Timeline

olce created this revision.Mon, Oct 5, 5:04 PM

Herald added a subscriber: imp. · View Herald TranscriptMon, Oct 5, 5:04 PM

olce requested review of this revision.Mon, Oct 5, 5:04 PM

Harbormaster completed remote builds in B77729: Diff 188737.Mon, Oct 5, 5:04 PM

olce added a parent revision: D60374: acpi: Tasks: Size the number of tasks with the actual number of CPUs.Mon, Oct 5, 5:04 PM

olce added a child revision: D60379: acpi: Tasks: Use M_WAITOK to allocate the taskqueue.Mon, Oct 5, 8:00 PM

emaste added inline comments.Wed, Oct 7, 3:56 PM

sys/dev/acpica/Osd/OsdSchedule.c
169	can we avoid the atomic in the common case?

olce marked an inline comment as done.Thu, Oct 8, 9:34 AM

olce added inline comments.

sys/dev/acpica/Osd/OsdSchedule.c
169	I don't think so, but don't think that makes a real difference either. The access in the `while()` guard is just an explicitly "atomic" load of an int, but in the end is just compiled to an actual load (AFAIK; we actually implicitly depend on the loads of ints being atomic in a lot of places). The `atomic_cmpset_int()` below is only executed while the high water value needs updating, so at most the number that it finally reports. Once it's stabilized, the `i > hiwater` test is always false and the `atomic_cmpset_int()` never executed again.

olce marked an inline comment as done.Thu, Oct 8, 9:34 AM

Revision Contents
Changeset List

Path

Size

sys/

dev/

acpica/

Osd/

OsdSchedule.c

9 lines

Diff 188737

View Options

sys/dev/acpica/Osd/OsdSchedule.c

acpi: Tasks: Precisely report the maximum number of slots usedNeeds ReviewPublicActions