Code review request: CR 7181995 NMT ON: NMT assertion failure assert(cur_vm->is_uncommit_record() || cur_vm->is_deallocation_record
karen.kinnear at oracle.com
Tue Sep 11 15:40:14 PDT 2012
Thank you for fixing something so complicated so quickly.
The code looks good, ok to check it. And many thanks.
Couple of minor questions/comments:
line 114: "exiting" -> "existing"
compare actually checks for "contains" - RFE for future: any chance you could either change the name
or add a comment?
3. couple of questions on promote_virtual_memory_records
1. what happens with a thread stack commit for non-guard pages?
2. does this logic assume you have ordered the records by sequence number? If so, could
you please add a comment. I saw you appended them during the merge, but
I wasn't sure what order they were in originally.
3. - if so, why would you get commit, uncommit, release, tag after a release?
4. For the multiple reserve case, wasn't that already handled in the first "if" or did I
read this incorrectly?
On Sep 11, 2012, at 2:23 PM, Zhengyu Gu wrote:
> This is a pretty significant changes on how NMT deal with virtual memory records. The old logic processed virtual memory records based on memory address order, then sequence number, which was fault, since virtual memory operations do not have to base on the ranges' base addresses. The corrected logic is to process virtual memory records purely based on the sequence number.
> CR: http://bugs.sun.com/view_bug.do?bug_id=7181995
> Webrev: http://cr.openjdk.java.net/~zgu/7181995/webrev.00/
> The changes passed following tests:
> - JPRT tests
> - vm quick tests
> - Parallel classloading tests
> - nsk stress tests
> - jck60 stress tests
> - quick jdi tests
> - quick jvmti tests
More information about the hotspot-dev