Ensure that GC idle notifications either make progress or stop requesting more GCs.

Ensure that GC idle notifications either make progress or stop requesting more GCs.

The V8::IdleNotification will only return 'True' when the gc idle time handler
thinks there is no more GC which can be done. However, the gc idle task handler
can end up repeatedly making no progress (e.g., if it can't finalize a sweep)
which causes idle tasks to be repeatedly scheduled in Chrome which do nothing
but wake up Chrome. Fix this by returning Done if we can't make any progress
within an Idle Round.

BUG=chromium:470615
LOG=Y

Committed: https://crrev.com/00477a5d72e1246d9611966a0b4c708f90a2228e
Cr-Commit-Position: refs/heads/master@{#27529}

[rmcilroy, 2015-03-27 17:41:21]

rmcilroy@chromium.org changed reviewers:
+ hpayer@chromium.org

[rmcilroy, 2015-03-27 17:41:22]

PTAL, thanks.

[Hannes Payer (out of office), 2015-03-27 19:04:19]

What about the following scenario:
We just finished a full gc and it while take a while to start incremental
marking.
We continuously approach the incremental marking start state but are slower then
10 received idle notifications.
In that case we would turn off the idle notification mechanism, just before
being eligible to do incremental marking during idle times. I don't think we
want that.
Have you tried running it on the smoothness benchmarks?

[rmcilroy, 2015-03-27 19:41:19]

On 2015/03/27 19:04:19, Hannes Payer wrote:
> What about the following scenario:
> We just finished a full gc and it while take a while to start incremental
> marking.
> We continuously approach the incremental marking start state but are slower
then
> 10 received idle notifications.
> In that case we would turn off the idle notification mechanism, just before
> being eligible to do incremental marking during idle times. I don't think we
> want that.
> Have you tried running it on the smoothness benchmarks?

This should never affect things when Chrome is not-idle (as in not completely in
the background) - We don't turn off the idle notifications if we return 'Done'
here, we just post the next idle task as 'afterwakeup' instead of as a normal
idle task. This means it can't wake up Chrome to do another 50ms long idle
period, but it will become eligable to run if Chrome runs any other Task
(including any task which causes JS to be run). I think anything which causes us
to approach being eligable for incremental marking would require a task being
run by Chrome so would make the idle task become runnable too.

[Hannes Payer (out of office), 2015-03-30 08:31:17]

OK, sounds good. Please add a unit test.

[rmcilroy, 2015-03-30 12:21:48]

On 2015/03/30 08:31:17, Hannes Payer wrote:
> OK, sounds good. Please add a unit test.

Done

[Hannes Payer (out of office), 2015-03-30 13:46:35]

lgtm

[rmcilroy, 2015-03-30 13:47:02]

The CQ bit was checked by rmcilroy@chromium.org

[commit-bot: I haz the power, 2015-03-30 13:47:13]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1042483002/20001

[commit-bot: I haz the power, 2015-03-30 13:55:21]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-03-30 13:55:22]

Try jobs failed on following builders:
  v8_linux_gcc_compile_rel on tryserver.v8 (JOB_FAILED,
http://build.chromium.org/p/tryserver.v8/builders/v8_linux_gcc_compile_rel/bu...)

[rmcilroy, 2015-03-30 16:22:36]

The CQ bit was checked by rmcilroy@chromium.org

[rmcilroy, 2015-03-30 16:22:37]

The patchset sent to the CQ was uploaded after l-g-t-m from hpayer@chromium.org
Link to the patchset: https://codereview.chromium.org/1042483002/#ps40001
(title: "size_t -> int")

[commit-bot: I haz the power, 2015-03-30 16:22:49]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1042483002/40001

[commit-bot: I haz the power, 2015-03-30 17:04:53]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2015-03-30 17:05:14]

Patchset 3 (id:??) landed as
https://crrev.com/00477a5d72e1246d9611966a0b4c708f90a2228e
Cr-Commit-Position: refs/heads/master@{#27529}