diff options
| author | Nicolas Geoffray <ngeoffray@google.com> | 2017-07-06 15:30:10 +0100 |
|---|---|---|
| committer | Nicolas Geoffray <ngeoffray@google.com> | 2017-07-10 09:31:46 +0100 |
| commit | 0436bb29ed9f9a2958454a1140259349d1659f2a (patch) | |
| tree | d2578aa31764917980c5f2e96127779ec5017de3 /test/ProfileTestMultiDex/Second.java | |
| parent | 1f894b057b03b04fbb37fbf721bbcf755e809eb9 (diff) | |
Fix region space when used with SetLengthToUsableSizeVisitor.
The region space relies on obj->SizeOf for some of its logic.
By having SetLengthToUsableSizeVisitor "change" the SizeOf
what's being allocated.
The bug happens during RegionSpace::ClearFromSpace: for unevac regions
we iterate over following regions. If LiveBytes != Top() - Begin()
(which happen for large allocations using SetLengthToUsableSizeVisitor),
we break the loop.
The next region to analyze is a large tail, and we see LiveBytes() == 0
(tails apparently always have live bytes == 0), the code is then
happy to release the large tail, even though the large object is still
live.
bug: 37187694
bug: 62889232
Test: 659-unpadded-array
(cherry picked from commit 4b361a87520643c888a3d2c52dffa050fabd7a0b)
Change-Id: Ia5c156a6969aad0b0c2ea2a4b7a0abdfa4088df6
Diffstat (limited to 'test/ProfileTestMultiDex/Second.java')
0 files changed, 0 insertions, 0 deletions