add get_cpus interface
AbandonedPublic
Actions

Authored by sbruno on Feb 6 2016, 12:29 AM.

Details

Reviewers

kmacy
gallatin
jhb
glebius

Summary

Taken from one of jhb's branches. Now a dependency for iflib. Determines the CPUs one should bind a device's interrupt handler's to.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 5449
Build 5659: CI src build	Jenkins

Event Timeline

kmacy updated this revision to Diff 13076.Feb 6 2016, 12:29 AM

kmacy retitled this revision from to add get_cpus interface.

kmacy updated this object.

kmacy edited the test plan for this revision. (Show Details)

kmacy added reviewers: jhb, gallatin, glebius.

kmacy set the repository for this revision to rS FreeBSD src repository - subversion.

kmacy added subscribers: benno, scottl.

Herald added a subscriber: imp. · View Herald TranscriptFeb 6 2016, 12:29 AM

kmacy added a child revision: D5211: import iflib.Feb 6 2016, 2:38 AM

Thanks for resurrecting this. I've been meaning to get back to it.

I had a thought last night (after seeing this review pop up yesterday). I think we should adjust bus_get_cpus() to also accept the cpuset size similar to the cpuset system calls. This will allow the API to handle modules compiled with a different MAXCPU (one way or another). This is somewhat related to PR 200802 which points out that even our system calls need to do a better job of handling that, but I do think we need the size parameter.

sys/dev/acpica/acpivar.h
15	This looks like a typo.

I noticed this when trying to backport to stable..

sys/dev/acpica/acpi.c
1099	This hunk seems to be extranious
sys/dev/acpica/acpivar.h
504	I think this is also extranious

remove extraneous changes
add size argument to get_cpus

jhb added inline comments.Feb 13 2016, 8:45 PM

sys/dev/acpica/acpi.c
1099	Actually, it's a bit of a mismerge. Let me check my git branch which I've merged more recently than the patches posted to arch@. However, the idea is to treat non-ACPI devices (no handle) differently than what get_domain() does. Maybe the mis-merge is related to the fact that I also changed get_domain to work with the updated acpi_parse_pxm() and this patch tries to not do that? Hmm, my branch is quite different here actually. Let me rebase that (and drop the Intel NIC driver diffs for now) and post that as an alternative.