Fix bug in semispace shrink-to.

Fix bug in semispace shrink-to.

Committed: http://code.google.com/p/v8/source/detail?r=8463

[Vyacheslav Egorov (Chromium), 2011-06-29 11:16:59]

LGMT

http://codereview.chromium.org/7281006/diff/1001/src/spaces.cc
File src/spaces.cc (right):

http://codereview.chromium.org/7281006/diff/1001/src/spaces.cc#newcode1226
src/spaces.cc:1226: Address old_start = space_end - capacity_;
please add a comment about new space growing backwards from the end of the
reserved space.

http://codereview.chromium.org/7281006/diff/1001/src/spaces.cc#newcode1229
src/spaces.cc:1229: if
(!heap()->isolate()->memory_allocator()->UncommitBlock(old_start, delta)) {
I am a bit concerned about uncommiting piece of the page. It seems possible.

[Lasse Reichstein, 2011-06-29 11:31:13]

http://codereview.chromium.org/7281006/diff/1001/src/spaces.cc
File src/spaces.cc (right):

http://codereview.chromium.org/7281006/diff/1001/src/spaces.cc#newcode1226
src/spaces.cc:1226: Address old_start = space_end - capacity_;
On 2011/06/29 11:16:59, Vyacheslav Egorov wrote:
> please add a comment about new space growing backwards from the end of the
> reserved space.

Done.

http://codereview.chromium.org/7281006/diff/1001/src/spaces.cc#newcode1229
src/spaces.cc:1229: if
(!heap()->isolate()->memory_allocator()->UncommitBlock(old_start, delta)) {
And you are correct to be worried.
It won't happen that we free part of a page that contains content, but that's
due to the interaction of minimum_capacity and the times-2 multiplier in Shrink,
so instead of trying to codify that, we sohuld just round up before calling
ShrinkTo. I'll change my other CL to do that instead of removing the assertion.