Make uma_int.h friendlier to userspace.
ClosedPublic
Actions

Authored by markj on Mar 5 2020, 7:12 PM.

Details

Reviewers

jeff
rlibby

Commits

rS358732: Clean up uma_int.h a bit.

Summary

I have a program that uses libkvm to dump the address of each slab
allocated to a zone. The intent is to measure fragmentation caused by
_NOFREE zones. When updating the program to catch up with all of the
UMA changes, I ended up simplifying uma_int.h a little bit:

Make slab_sizeof(), slab_space() and slab_ipers() local to uma_core.c. I can't see why they need to be exported, and they don't follow the UMA namespace conventions. Note that a couple of these functions are unused.
Group KEG_ASSERT_COLD with other keg macros.
Stop defining the slab structures under _KERNEL. There's no real reason they can't be visible to userspace like the rest of UMA's structures are.
Remove an assertion about MAXMEMDOM - there are tons of places in the tree that assume that a domain index fits in 8 bits.

No functional change intended.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

markj created this revision.Mar 5 2020, 7:12 PM

Harbormaster completed remote builds in B29817: Diff 69237.Mar 5 2020, 7:12 PM

markj added reviewers: jeff, rlibby.Mar 5 2020, 7:13 PM

jeff added inline comments.Mar 5 2020, 9:10 PM

sys/vm/uma_core.c
1962–1966 ↗	(On Diff #69237)	If this is no longer used and libby does not plan on using it let's remove it.
sys/vm/uma_int.h
396 ↗	(On Diff #69237)	What does removing this get us? I added it so that you could find failures explicitly. Otherwise it may be hard to know.

markj added inline comments.Mar 5 2020, 9:37 PM

sys/vm/uma_core.c
1962–1966 ↗	(On Diff #69237)	Ok, I just wanted to make sure you were both ok with that first.
sys/vm/uma_int.h
396 ↗	(On Diff #69237)	I can re-add it, I think it just needs to be kernel-only. We make this assumption all over the place, FWIW, and there are pad bytes following the domain field on 64-bit platforms. Maybe it should be domainid_t instead.

Looks good except SLAB_BITSETS and Jeff's comments.

sys/vm/uma_core.c
1962–1966 ↗	(On Diff #69237)	Yep, go ahead. I think this got left behind with the startup alloc changes.
sys/vm/uma_int.h
394–402 ↗	(On Diff #69237)	This can't make sense to user space and should either remain under _KERNEL or move to uma_core.c.
396 ↗	(On Diff #69237)	Incidentally, if you are changing it, it can be written with _Static_assert.

This revision is now accepted and ready to land.Mar 6 2020, 10:20 PM

Remove dead functions.
Move SLAB_BITSETS to uma_core.c.
Restore the MAXMEMDOM assertion, I missed that it is checking MAXMEMDOM < 255, not MAXMEMDOM <= 255.

This revision now requires review to proceed.Mar 7 2020, 12:44 AM

Harbormaster completed remote builds in B29836: Diff 69282.Mar 7 2020, 12:44 AM

Fixups LGTM.

This revision is now accepted and ready to land.Mar 7 2020, 12:52 AM

Closed by commit rS358732: Clean up uma_int.h a bit. (authored by markj). · Explain WhyMar 7 2020, 3:37 PM

This revision was automatically updated to reflect the committed changes.

markj added a commit: rS358732: Clean up uma_int.h a bit..

Herald added a subscriber: imp. · View Herald TranscriptMar 7 2020, 3:37 PM