commit 1a2bf5560168a4816ec194ab44cae2010d9173b2
author David Stevens <stevensd@google.com> Tue May 21 18:07:25 2019 +0000
committer CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Tue May 21 18:07:25 2019 +0000
tree 3dd08ab3c2441ed98ccf665e763c1add296a1b65
parent c0a853e9fa85185a5531fbfaff3e1c220c12c30f

[kernel][vm] Improve efficiency of bidirectional clones

This change reduces the number of page allocations done by bidirectional
clones so that they are never less efficient than simply memcpying into
a new VMO. This is done by moving pages out of hidden VMOs when writing
to one of their children if the page is not accessible by the other
child.

Whether a page in a hidden VMO is 'accessible' to a particular child is
determined by a combination of two factors. First, a page is not
accessible if it lies outside the range of the hidden VMO which the
child can see. This is tracked by the existing parent offset and limit
values in the child. Second, if a page in the hidden VMO has already
been copied into a child, that page in the hidden VMO is no longer
accessible to that child. This is tracked by new 'split' bits in the
vm_page_t structure.

Test: existing vmo-clone2 core tests were written to cover this code,
and kstress has a test for hammering COW clones
ZX-1268 #comment [kernel][vm] Improve efficiency of bidirectional clones

Change-Id: Idbfcab1b684c12faa29499e3c9c78e478989a1ef