per-cpu free page caching
ClosedPublic
Actions

Authored by jeff on Mar 30 2018, 3:47 AM.

Details

Reviewers

alc
markj
kib

Commits

rS331863: Add a uma cache of free pages in the DEFAULT freepool. This gives us

Summary

This implements per-cpu free page caching only for FREEPOOL_DEFAULT. I haven't seen enough contention yet from FREEPOOL_DIRECT that it is an issue. This also brings in vm_phys_alloc_npages() which gives you as many as 'count' pages. This is an optimization to reduce round-trips to the backend phys allocator. Contrary to other instances of this patch, this is now completely compatible with reservations.

It might be more elegant to push this into vm_phys.c along with all of the vm_domain_free_lock() calls. However, pushing it there would mean the cached pages were reflected in the free count. I believe this will make us more prone to out of memory deadlocks.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

jeff created this revision.Mar 30 2018, 3:47 AM

Herald added a subscriber: imp. · View Herald TranscriptMar 30 2018, 3:47 AM

kib added inline comments.Mar 30 2018, 3:04 PM

vm/vm_page.c
224 ↗	(On Diff #40906)	Should it be sizeof(struct vm_page *) ?

jeff added inline comments.Mar 30 2018, 11:24 PM

vm/vm_page.c
224 ↗	(On Diff #40906)	actually in this case uma uses it as a cost function to limit the bucket size. But we override with MAXBUCKET. So it is correct to intent but unused.