Fix KVA handling in kmem_bootstrap_free().
ClosedPublic
Actions

Authored by markj on Jul 24 2018, 5:01 PM.

Details

Reviewers

alc
kib
royger

Commits

rS336764: Fix handling of KVA in kmem_bootstrap_free().

Summary

The kmem_* layer must handle pmap updates itself, since vm_map_remove()
won't update the kernel pmap for us. Thus, modify kmem_bootstrap_free()
to remove the KVA range from kernel_pmap and add it to kernel_arena.
For this to work, we must support demotion of the bootstrapped kernel
page tables. It suffices to ensure that PG_A is set on the PDEs (i386
already does this) and to insert the shadowed PTPs into the kernel
pmap's radix tree when initializing them in pmap_init().

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

markj created this revision.Jul 24 2018, 5:01 PM

Harbormaster completed remote builds in B18303: Diff 45790.Jul 24 2018, 5:01 PM

markj added reviewers: alc, kib, royger.Jul 24 2018, 5:01 PM

markj mentioned this in D16330: Add support for freeing preloaded data..

markj added a subscriber: emaste.

kib accepted this revision.Jul 24 2018, 5:30 PM

This revision is now accepted and ready to land.Jul 24 2018, 5:30 PM

markj added inline comments.Jul 24 2018, 5:43 PM

sys/i386/i386/pmap.c
665 ↗	(On Diff #45790)	Oops, this change isn't needed.

alc added inline comments.Jul 25 2018, 6:35 AM

sys/amd64/amd64/pmap.c
1339 ↗	(On Diff #45790)	KERNend could be in the midst of a 2 MB page mapping. Are you sure that you want to use trunc here?

Don't set PG_A on PTEs.
Fix test in pmap_init() on amd64.

This revision now requires review to proceed.Jul 25 2018, 3:17 PM

Harbormaster completed remote builds in B18324: Diff 45826.Jul 25 2018, 3:17 PM

markj added inline comments.Jul 25 2018, 3:18 PM

sys/amd64/amd64/pmap.c
1339 ↗	(On Diff #45790)	Indeed, this doesn't match the loop in create_pagetables() which creates the 2MB mappings of the kernel.

alc added inline comments.Jul 25 2018, 5:45 PM

sys/amd64/amd64/pmap.c
983 ↗	(On Diff #45826)	Is there a reason for placing this assignment here? At this point, create_pagetables() has allocated some, but not all, of the pages that it's going to allocate. Why not do this assignment in pmap_bootstrap(), before create_pagetables() is called?
1013 ↗	(On Diff #45826)	As an aside, I wish there was a way to easily use the up to 1.99... MB of memory that we are wasting here as UMA boot pages.
sys/i386/i386/pmap.c
941 ↗	(On Diff #45826)	While we are here, can you please merge r329171 from amd64.