Remove the deprecated timeout(9) interface.
ClosedPublic
Actions

Authored by jhb on Nov 28 2019, 8:49 PM.

Details

Reviewers

kib
markj

Group Reviewers

manpages

Commits

rS355732: Remove the deprecated timeout(9) interface.

Summary

All in-tree consumers have been converted to callout(9).

Test Plan

make tinderbox
booted amd64 in a VM

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Passed

Unit

No Test Coverage

Build Status

Buildable 27845
Build 26022: arc lint + arc unit

Event Timeline

jhb created this revision.Nov 28 2019, 8:49 PM

Herald added a reviewer: manpages. · View Herald TranscriptNov 28 2019, 8:49 PM

Herald added subscribers: donner, imp. · View Herald Transcript

Harbormaster completed remote builds in B27845: Diff 65026.Nov 28 2019, 8:49 PM

jhb added parent revisions: D22601: Remove uses of the deprecated timeout_t typedef., D22599: Use callout(9) instead of deprecated timeout(9)., D22598: Use callout(9) instead of deprecated timeout(9)., D22597: Use a callout instead of timeout(9) for delayed zio's..Nov 28 2019, 8:49 PM

Will need a __FreeBSD_version bump probably.

sys/kern/kern_timeout.c
386	It is not clear if we want to really treat the soft clock thread on CPU 0 different going forward. If not, we would retire cc_default_cpu using PCPU_GET(cpuid) instead and merge these two tunable to be a single tunable perhaps.
1327	For this commit I should drop the XXX, however, there is an open question I think perhaps as a followup if we should not force timeouts onto CPU 0 but just let them use whatever CPU they are created on as the default?

Drop unrelated crypto.9 diff.

Harbormaster completed remote builds in B27846: Diff 65027.Nov 28 2019, 8:53 PM

kib added inline comments.Nov 28 2019, 10:06 PM

share/man/man9/callout.9
92	This looks very much as a material for other commit ('remove timeout_t typedef'), but again it raises a question why remove this useful typedef.
sys/kern/kern_timeout.c
386	I do not see much sense in cc_default_cpu. We use curcpu for callouts otherwise, and I believe we should just extend the practice there.

cem added a subscriber: cem.Nov 29 2019, 3:01 AM

cem added inline comments.

share/man/man9/callout.9
804–806	Maybe throw a reference in a .Sh HISTORY section? Not a big deal.
sys/kern/kern_timeout.c
1327	There may be some power saving benefit on (relatively common) 2-4? 8? core mobile hardware of coalescing callouts onto a single core, because you can wake fewer cores less often. On the other hand, on big machines with a lot of callouts you can make a big queue on cpu 0. I might coalesce nonspecific callouts to cpu % 4 or cpu % 8 to sort of meet halfway? I haven't thought that through thoroughly.

jhb added a parent revision: D22684: Use a void * argument to callout handlers instead of timeout_t casts..Dec 4 2019, 10:11 PM

jhb added a parent revision: D22751: Add a callout_func_t typedef for functions used with callout_*()..Dec 10 2019, 7:02 PM

jhb added a parent revision: D22752: Use callout_func_t instead of the deprecated timeout_t..Dec 10 2019, 7:04 PM

jhb marked an inline comment as done.Dec 11 2019, 12:25 AM

Rebase, flesh out HISTORY.

Harbormaster completed remote builds in B28070: Diff 65481.Dec 11 2019, 12:28 AM

kib accepted this revision.Dec 11 2019, 11:08 AM

This revision is now accepted and ready to land.Dec 11 2019, 11:08 AM

markj accepted this revision.Dec 11 2019, 2:26 PM

markj added inline comments.

share/man/man9/callout.9
33	The date change looks wrong.
sys/kern/kern_timeout.c
940	"allow migration"?

emaste added a subscriber: emaste.Dec 13 2019, 7:57 PM

jhb marked 2 inline comments as done.Dec 13 2019, 8:52 PM

jhb added inline comments.

share/man/man9/callout.9
33	Oops, that's due to how I rebased this tree (by copying the old one off and pulling in the updated timeout.9 after the callout_func_t commit then copying this back over). I will bump it to the real date at commit time.