Request for reviews (S): 6987703 iCMS: Intermittent hang with gc/gctests/CallGC/CallGC01 and +ExplicitGCInvokesConcurrent
John Coomes
John.Coomes at oracle.com
Tue Mar 15 01:15:39 UTC 2011
Y. Srinivas Ramakrishna (y.s.ramakrishna at oracle.com) wrote:
>
> Summary: In the existing implementation of ExplicitGCInvokesConcurrent
> two concurrent System.gc()'s could intefere when using CMSIncremetalMode,
> leaving the CMS thread stranded in incremental mode with no foreground
> allocation happening to act as a trigger to get the (i)CMS thread going.
>
> Solution: enable_icms() and disable_icms(), which respectively enable and
> disable the incremental mode of CMS, should be counting requests
> so as not to intefere. Also added some related asserts.
> (Removed the obsolete and unused _next field in the CMST class.)
>
> webrev: http://cr.openjdk.java.net/~ysr/6987703/webrev.00/
>
> Testing: The failing test (and JPRT) with
> +/-ExplicitGCInvokesConcurrent +/-ExplicitGCInvokesConcurrent
> Prior to the fix, on a specific machine, the test would
> hang about once in three runs or so. After the fix it's been
> running without hanging for a whole day (many thousands of iterations).
Looks good to me.
-John
More information about the hotspot-gc-dev
mailing list