Safepointing in GC

Add safepointing to the GC


BUG=
R=iposva@google.com

Committed: https://github.com/dart-lang/sdk/commit/cab003366785773ace16b5305ac1f33c228cac54

[koda, 2015-08-14 17:10:08]

koda@google.com changed reviewers:
+ iposva@google.com

[Ivan Posva, 2015-08-19 08:00:04]

https://codereview.chromium.org/1292353004/diff/60001/runtime/vm/gc_sweeper.cc
File runtime/vm/gc_sweeper.cc (right):

https://codereview.chromium.org/1292353004/diff/60001/runtime/vm/gc_sweeper.c...
runtime/vm/gc_sweeper.cc:192: MonitorLocker ml(&started_monitor);
I don't expect it to be needed to wait for the sweeper threads to start running
before the marker can make further progress.

[koda, 2015-08-19 20:55:48]

Simplified the CL considerably:

- reverted "wait for sweeper to start" changes; we will continue using the task
count for the cases where we specifically care about GC/oldspace state

- added safepoint checks to sweeper betweeen pages

- updated unit test to busy-wait with safepoint checks

(The issue of threads started during a safepoint will still come up for the
parallel marking tasks, but their lifetime will be simpler to manage.)

PTAL

[Ivan Posva, 2015-08-25 23:27:15]

LGTM with comments added. -ip

https://codereview.chromium.org/1292353004/diff/120001/runtime/vm/pages.cc
File runtime/vm/pages.cc (right):

https://codereview.chromium.org/1292353004/diff/120001/runtime/vm/pages.cc#ne...
runtime/vm/pages.cc:772: isolate->thread_registry()->SafepointThreads();
// Ensure that all threads for this isolate are at a safepoint (either stopped
or in native code).

https://codereview.chromium.org/1292353004/diff/120001/runtime/vm/scavenger.cc
File runtime/vm/scavenger.cc (right):

https://codereview.chromium.org/1292353004/diff/120001/runtime/vm/scavenger.c...
runtime/vm/scavenger.cc:808: isolate->thread_registry()->SafepointThreads();
ditto about the comment.

Also please explain what will happen if two different threads decide to race for
triggering a GC at the same time.

[koda, 2015-08-26 01:09:15]

https://codereview.chromium.org/1292353004/diff/120001/runtime/vm/pages.cc
File runtime/vm/pages.cc (right):

https://codereview.chromium.org/1292353004/diff/120001/runtime/vm/pages.cc#ne...
runtime/vm/pages.cc:772: isolate->thread_registry()->SafepointThreads();
On 2015/08/25 23:27:15, Ivan Posva wrote:
> // Ensure that all threads for this isolate are at a safepoint (either stopped
> or in native code).

Done.

https://codereview.chromium.org/1292353004/diff/120001/runtime/vm/scavenger.cc
File runtime/vm/scavenger.cc (right):

https://codereview.chromium.org/1292353004/diff/120001/runtime/vm/scavenger.c...
runtime/vm/scavenger.cc:808: isolate->thread_registry()->SafepointThreads();
On 2015/08/25 23:27:15, Ivan Posva wrote:
> ditto about the comment.
> 
> Also please explain what will happen if two different threads decide to race
for
> triggering a GC at the same time.

Done.

Added TODO about avoiding the possibility of redundant collections by moving
safepointing into allocation failure/retry logic.

[koda, 2015-08-26 01:20:31]

Committed patchset #11 (id:190001) manually as
cab003366785773ace16b5305ac1f33c228cac54 (presubmit successful).