Implement on_each_cpu()
ClosedPublic
Actions

Authored by markj on Apr 30 2017, 6:48 PM.

Details

Reviewers

• hselasky

Commits

rS317651: Add on_each_cpu() and wbinvd_on_all_cpus().

Summary

This is for DRM. We only implement the blocking form (wait == 1),
since that's all our smp_rendezvous implements. But that's what almost
all callers in Linux use anyway.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

markj created this revision.Apr 30 2017, 6:48 PM

Herald added a subscriber: emaste. · View Herald TranscriptApr 30 2017, 6:48 PM

markj retitled this revision from Implement on_each_cpu(). to Implement on_each_cpu().Apr 30 2017, 6:52 PM

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

markj added a reviewer: • hselasky.

emaste added inline comments.Apr 30 2017, 6:59 PM

sys/compat/linuxkpi/common/src/linux_compat.c
1640 ↗	(On Diff #27869)	A stylistic question - this follows Linux APIs in calls from this function; is that preferable to having the Linux API only at the interface, and using FreeBSD routines internally (i.e. here calling smp_rendezvous instead?) For this case it makes little difference of course, but I'm curious about the two approaches in general.

markj added inline comments.Apr 30 2017, 7:05 PM

sys/compat/linuxkpi/common/src/linux_compat.c
1640 ↗	(On Diff #27869)	Oops - I had actually meant to use the internal interface there. I do think it's preferable to avoid using the external interfaces, but in some cases (e.g., spin_lock()) that's not very practical right now.

• hselasky added inline comments.Apr 30 2017, 7:43 PM

sys/compat/linuxkpi/common/include/asm/smp.h
34 ↗	(On Diff #27869)	Please use a function macro here. wbinvd_on_all_cpus()
sys/compat/linuxkpi/common/include/linux/smp.h
33 ↗	(On Diff #27869)	I think CTASSERT() is preferred??

emaste added inline comments.Apr 30 2017, 7:50 PM

sys/compat/linuxkpi/common/include/linux/smp.h
33 ↗	(On Diff #27869)	`_Static_assert` is C11, and when a C11 or better compiler is required for all architectures I'd say we should prefer it. Until then I supposed `CTASSERT` is preferred. (FWIW CTASSERT was first introduced in rS89040)