RFR(T) [13] : 8066173 : compiler/types/correctness/OffTest.java failed with assert
Igor Ignatyev
igor.ignatyev at oracle.com
Tue Jun 11 21:13:58 UTC 2019
http://cr.openjdk.java.net/~iignatyev//8066173/webrev.00/index.html
> 3 lines changed: 1 ins; 0 del; 2 mod;
Hi all,
could you please review this trivial fix removes compiler/types/correctness from problem list on all platform and problem list them only on solaris-sparcv9 due to 8225620[1]?
from JBS:
> Patric Hedlin added a comment:
> The implementation in *load_extra_data()* has changed over the last few months (for JDK12/13) and may/should not manifest the problem any more. However, there might still be a potential problem (window) in *prepare_metadata()*, in the current JDK13 code base, where _PrepareExtraDataClosure_ will use the (now) fragile/raced (_SafepointSynchronize_) *safepoint_counter()*. The scenario would be that a safepoint has indeed started during *prepare_metadata()* but finishes after, without being detected.
>
> The test-case also uses the WhiteBox API to clean method data, via (_WhiteBox_) *clearMethodState()*. This would explain the (mal-) interaction with the previous implementation of *load_extra_data()*.
>
> 1. I would suggest that the issue is closed until manifested anew.
>
> 2. The Safepoint code/state should perhaps be scrutinized further to close the potential problem window. (Robbin will file a new report.)
webrev: http://cr.openjdk.java.net/~iignatyev//8066173/webrev.00/index.html
testing: compiler/types/correctness on {linux-x64,windows-x64,mac-x64} x {product,fastdebug}
JBS: https://bugs.openjdk.java.net/browse/JDK-8066173
[1] https://bugs.openjdk.java.net/browse/JDK-8225620
Thanks,
-- Igor
More information about the hotspot-compiler-dev
mailing list