Before write-protecting a PageSpace, return bump allocation block to freelist.

Before write-protecting a PageSpace, return bump allocation block to freelist.

This fixes a potential issue where attempts to iterate over a write-protected
heap might try to ensure it's walkable by writing a header into the bump block.

R=asiva@google.com

Committed: https://code.google.com/p/dart/source/detail?r=45173

[koda, 2015-04-15 01:31:59]

Added regression test; ready for review.

[siva, 2015-04-15 16:11:56]

lgtm

https://codereview.chromium.org/1071393003/diff/120001/runtime/vm/pages.cc
File runtime/vm/pages.cc (right):

https://codereview.chromium.org/1071393003/diff/120001/runtime/vm/pages.cc#ne...
runtime/vm/pages.cc:507: }
Is it ok to still have bump_top_ and bump_end_ point to an element in the
freelist here? I am wondering if bump_top_ and bump_end_ need to be set to 0
here?

[koda, 2015-04-15 16:24:06]

https://codereview.chromium.org/1071393003/diff/120001/runtime/vm/pages.cc
File runtime/vm/pages.cc (right):

https://codereview.chromium.org/1071393003/diff/120001/runtime/vm/pages.cc#ne...
runtime/vm/pages.cc:507: }
On 2015/04/15 16:11:56, siva wrote:
> Is it ok to still have bump_top_ and bump_end_ point to an element in the
> freelist here? I am wondering if bump_top_ and bump_end_ need to be set to 0
> here?

The call only writes a header to the block, but does not actually add it to any
freelist (FreeList::Free does). Writing the header ensures that we can walk the
entire space. We do this lazily to save a couple of writes in bump pointer
allocation.

[koda, 2015-04-15 16:59:08]

Committed patchset #7 (id:120001) manually as r45173 (presubmit successful).