RFR (S): 8020155: PSR:PERF G1 not collecting old regions when humongous allocations interfer

Thomas Schatzl thomas.schatzl at oracle.com
Thu Jul 11 14:10:41 UTC 2013


On Thu, 2013-07-11 at 15:27 +0200, Bengt Rutisson wrote:
> Hi everyone,
> Could I get a couple of reviews for this small change?
> http://cr.openjdk.java.net/~brutisso/8020155/webrev.00/
> When the marking phase finishes we signal that we should start doing 
> mixed collections by setting _last_young_gc=true. The next GC will be a 
> normal young GC but after that we will start doing mixed GCs.
> G1CollectorPolicy::need_to_start_conc_mark() will return false if a 
> marking cycle is already in progress or if we are doing mixed GCs. But 
> it did not check the intermediate state represented by _last_young_gc. 
> Instead this was explicitly checked by 
> G1CollectorPolicy::record_collection_pause_end() for young GCs. However, 
> need_to_start_conc_mark() is also used by humongous object allocations 
> in two places. So, if a humongous allocation happens after marking has 
> finished but before the following young GC has happened, we might 
> trigger a new marking cycle and skip the mixed GCs.
> The fix is to move the check of _last_young_gc in to 
> need_to_start_conc_mark() to make sure that we don't skip mixed GCs.

Looks good.


More information about the hotspot-gc-dev mailing list