Add phys_avail allocation routines.
ClosedPublic
Actions

Authored by jeff on Aug 13 2019, 5:31 PM.

Details

Reviewers

alc
markj
kib
gallatin

Commits

rS351181: Encapsulate phys_avail manipulation in a set of simple routines. Add a

Summary

This depends on https://reviews.freebsd.org/D21250

To implement a more intelligent early alloc I neded to be able to split phys_avail entries. This really could grow a little more encapsulation so there is less direct list manipulation elsewhere. The primary goal is to enable domain aware early allocation but I also think a little abstraction is nicer.

Some of this code came from Drew at netflix.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

jeff created this revision.Aug 13 2019, 5:31 PM

Harbormaster completed remote builds in B25847: Diff 60747.Aug 13 2019, 5:31 PM

jeff retitled this revision from Add phys_avail processing routines. to Add phys_avail allocation routines..Aug 13 2019, 5:40 PM

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

jeff added reviewers: alc, markj, kib, gallatin.

jeff mentioned this in D21250: Move phys_avail and dump_avail into MI code..Aug 13 2019, 5:53 PM

gallatin accepted this revision.Aug 15 2019, 12:14 AM

gallatin added inline comments.

sys/vm/vm_phys.c
1635 ↗	(On Diff #60747)	Style nit .. I think these locals should be sorted alphabetically
1699 ↗	(On Diff #60747)	In the numa case, would it make sense to iterate to another domain rather than panic?

This revision is now accepted and ready to land.Aug 15 2019, 12:14 AM

kib accepted this revision.Aug 17 2019, 6:49 PM

kib added inline comments.

sys/vm/vm_phys.c
1604 ↗	(On Diff #60747)	`phys_avail[i + 1] != 0`
1642 ↗	(On Diff #60747)	dot at the end of the sentence
1668 ↗	(On Diff #60747)	Again, end the sentence with dot.
1672 ↗	(On Diff #60747)	!= 0

jeff added inline comments.Aug 18 2019, 6:40 AM

sys/vm/vm_phys.c
1699 ↗	(On Diff #60747)	I think that the startup memory we're allocating should be so small relative to domain memory that if we run into this it indicates another bug.