Extend cpuset_get/setaffinity() APIs

[cem]

Summary

Add IRQ placement-only and ithread-only API variants. intr_event_bind
has been extended with sibling methods, as it has many more callsites in
existing code.

Test Plan

(I moved the mutually unobjectionable part of D10435 to its own review.)

[adrian]

This is just the API and not any cpuset changes. Guess we can do the userland side of it later!

[cem]

In D10586#219478, @adrian wrote:

This is just the API and not any cpuset changes. Guess we can do the userland side of it later!

Can you imagine the bikeshed over new cpuset(1) argument letters? :-D

[adrian]

Yes I can. Sure, commit it, I'll uhm, "do" the cpuset bikeshed. :)

[kib]

Overall it is fine, see some trivial nits inline.

sys/kern/kern_intr.c

294 ↗

(On Diff #27979)

Use bool for boolean arguments ?

344 ↗

(On Diff #27979)

This is racy beyond any tolerance. Two parallel setaffinity or one set and one get are not supposed to work correctly, it seems. I believe a low-tech but sufficient solution is to add global perhaps sleepable (sx) lock around both intr_setaffinity and intr_getaffinity.

This is a subject for a separate patch, of course.

sys/sys/cpuset.h

86 ↗

(On Diff #27979)

I suggest CPU_WHICH_INTRHANDLER or similar name. IRQ_ONLY is puzzling.

[cem]

sys/kern/kern_intr.c

294 ↗

(On Diff #27979)

I just used the prevailing style in the file. It would be fine to start using bool, but I don't want to convert all of the other uses of int-as-bool in this file.

sys/sys/cpuset.h

86 ↗

(On Diff #27979)

Ok.

[cem]

• Use bool for boolean values
• Change name from IRQ_ONLY to INTRHANDLER

[kib]

Later, you would want to update cpuset(2).

sys/kern/kern_intr.c

328 ↗

(On Diff #27986)

If you write this as

if (error != 0 && bindithread) {

you can de-indent the body by one level.

[cem]

sys/kern/kern_intr.c

328 ↗

(On Diff #27986)

Sort of. The return (error) still needs to happen if error != 0 && !bindithread. So that would have to change into an } else if () {. And there's no compelling reason yet — none of the extra indented lines wrap.

I'll leave it as-is for now.

[kib]

sys/kern/kern_intr.c

328 ↗

(On Diff #27986)

If would be

if (error != 0 && !bindithread) {
...
}
if (error != 0)
  return (error);

Ok.