Fix lazy sweeping for new marking bit clearing.

Fix lazy sweeping for new marking bit clearing.

This cleans the semantics of the WAS_SWEPT_* page flags. Pages swept
precisely can be iterated, hitting only the live objects. Whereas those
swept conservatively cannot be iterated over. Both flags indicate that
marking bits have been cleared by the sweeper, otherwise marking bits
are still intact.

The actual fix is in SetPagesToSweep, which had a buggy iteration and
set the WAS_SWEPT_CONSERVATIVELY flag on too many pages. Another fix
is in EvacuateLiveObjectsFromPage which clears marking bits now.

R=erik.corry@gmail.com,vegorov@chromium.org
BUG=v8:1463
TEST=cctest/test-api/Threading

Committed: http://code.google.com/p/v8/source/detail?r=9011

[Erik Corry, 2011-08-24 11:33:34]

I don't think it makes sense to say that a page that was not swept at all has
been swept conservatively.  Should we have a completely new flag that indicatest
that it has not been swept?

[Michael Starzinger, 2011-08-24 12:05:41]

Yes, adding something like SHOULD_BE_SWEPT_CONSERVATIVELY makes perfect sense to
me. I'll look into that.

On 2011/08/24 11:33:34, Erik Corry wrote:
> I don't think it makes sense to say that a page that was not swept at all has
> been swept conservatively.  Should we have a completely new flag that
indicatest
> that it has not been swept?

[Michael Starzinger, 2011-08-25 08:07:34]

As discussed yesterday, I cleaned up the semantics of the WAS_SWEPT_* flags.
There are basically three states a page can be in now:

1) WAS_SWEPT_PRECISELY: Page can be iterated over, marking bits are cleared.
2) WAS_SWEPT_CONSERVATIVELY: Page cannot be iterated over, marking bits are
cleared
3) None of the above two: Marking bits are still intact.

Besides that the patch set contains two fixes needed for correct clearing of
marking bits during sweeping, please see the updated description for details
about that.

[Erik Corry, 2011-08-25 09:19:47]

LGTM

http://codereview.chromium.org/7720024/diff/4001/src/mark-compact.cc
File src/mark-compact.cc (right):

http://codereview.chromium.org/7720024/diff/4001/src/mark-compact.cc#newcode2499
src/mark-compact.cc:2499: !p->WasSweptPrecisely() &&
I think it makes sense to have a
  bool WasSwept() { return WasSweptPrecisely() || WasSweptConservatively(); }
I think there are 4 places this could be used.

[Michael Starzinger, 2011-08-25 09:33:02]

Added new patch set.

http://codereview.chromium.org/7720024/diff/4001/src/mark-compact.cc
File src/mark-compact.cc (right):

http://codereview.chromium.org/7720024/diff/4001/src/mark-compact.cc#newcode2499
src/mark-compact.cc:2499: !p->WasSweptPrecisely() &&
On 2011/08/25 09:19:48, Erik Corry wrote:
> I think it makes sense to have a
>   bool WasSwept() { return WasSweptPrecisely() || WasSweptConservatively(); }
> I think there are 4 places this could be used.

Done.