Disable eager tracing for thread-terminated GCs.

Disable eager tracing for thread-terminated GCs.

The Oilpan marking phase enables and sets up a stack limit
that the eager tracing calls should not exceed while visiting the live
object graph. The marking phase is single threaded. The stack limit
is reset and eager marking disabled once the marking phase has
completed -- thread local weak processing (and its slight amount of
tracing) is run with it disabled.

Do not enable such an eager marking stack limit for the GCs that are run
for a thread's heap when the thread is terminated, as this risks conflicting
with any thread-local weak processing that is still underway. The slight
performance gain that eager marking brings is not an issue for such GCs.

R=haraken
BUG=420515
Committed: https://crrev.com/76c2d557fcab24b4d76659301006d0d8f5e196e8
Cr-Commit-Position: refs/heads/master@{#358273}

[sof, 2015-11-05 18:07:50]

sigbjornf@opera.com changed reviewers:
+ oilpan-reviews@chromium.org

[sof, 2015-11-05 18:07:50]

please take a look.

I believe this is what is behind this mystery assert,


https://storage.googleapis.com/chromium-layout-test-archives/WebKit_Win_Oilpa...

i.e., a thread racing ahead and starting to shut down after its (brief)
thread-local processing has completed..but not before all the others have
completed local weak processing.

[haraken, 2015-11-05 23:37:44]

Nice detective work, Sigbjorn!

LGTM

[sof, 2015-11-06 06:22:28]

Description was changed from

==========
Disable eager tracing for thread-terminated GCs.

The Oilpan marking phase enables and sets up a stack limit
that the eager tracing calls should not exceed while visiting the live
object graph. The stack limit is reset and eager marking disabled once
the marking phase has completed -- thread local weak processing (and
its slight amount of tracing) is run with it disabled.

Do not enable such an eager marking stack limit for the GCs that are run
for a thread's heap when the thread is terminated, as this risks conflicting
with any thread-local weak processing that is still underway. The slight
performance gain that eager marking brings is not an issue for such GCs.

R=
BUG=420515
==========

to

==========
Disable eager tracing for thread-terminated GCs.

The Oilpan marking phase enables and sets up a stack limit
that the eager tracing calls should not exceed while visiting the live
object graph. The marking phase is single threaded. The stack limit
is reset and eager marking disabled once the marking phase has
completed -- thread local weak processing (and its slight amount of
tracing) is run with it disabled.

Do not enable such an eager marking stack limit for the GCs that are run
for a thread's heap when the thread is terminated, as this risks conflicting
with any thread-local weak processing that is still underway. The slight
performance gain that eager marking brings is not an issue for such GCs.

R=
BUG=420515
==========

[sof, 2015-11-06 06:22:45]

Description was changed from

==========
Disable eager tracing for thread-terminated GCs.

The Oilpan marking phase enables and sets up a stack limit
that the eager tracing calls should not exceed while visiting the live
object graph. The marking phase is single threaded. The stack limit
is reset and eager marking disabled once the marking phase has
completed -- thread local weak processing (and its slight amount of
tracing) is run with it disabled.

Do not enable such an eager marking stack limit for the GCs that are run
for a thread's heap when the thread is terminated, as this risks conflicting
with any thread-local weak processing that is still underway. The slight
performance gain that eager marking brings is not an issue for such GCs.

R=
BUG=420515
==========

to

==========
Disable eager tracing for thread-terminated GCs.

The Oilpan marking phase enables and sets up a stack limit
that the eager tracing calls should not exceed while visiting the live
object graph. The marking phase is single threaded. The stack limit
is reset and eager marking disabled once the marking phase has
completed -- thread local weak processing (and its slight amount of
tracing) is run with it disabled.

Do not enable such an eager marking stack limit for the GCs that are run
for a thread's heap when the thread is terminated, as this risks conflicting
with any thread-local weak processing that is still underway. The slight
performance gain that eager marking brings is not an issue for such GCs.

R=haraken
BUG=420515
==========

[sof, 2015-11-06 06:22:51]

The CQ bit was checked by sigbjornf@opera.com

[commit-bot: I haz the power, 2015-11-06 06:23:45]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1429273002/1
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1429273002/1

[commit-bot: I haz the power, 2015-11-06 06:40:13]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2015-11-06 06:41:14]

Patchset 1 (id:??) landed as
https://crrev.com/76c2d557fcab24b4d76659301006d0d8f5e196e8
Cr-Commit-Position: refs/heads/master@{#358273}

[sof, 2015-11-06 15:29:33]

On 2015/11/05 23:37:44, haraken wrote:
> Nice detective work, Sigbjorn!
> 
> LGTM

Looks as if that took care of this long standing issue, last two builds are free
of it --

 
https://storage.googleapis.com/chromium-layout-test-archives/WebKit_Win_Oilpa...

I know of no other remaining Oilpan-specific crashing issues on the bots,
currently.