RFR: JDK-8311261: [AIX] TestAlwaysPreTouchStacks.java fails due to java.lang.RuntimeException: Did not find expected NMT output
Thomas Stuefe
stuefe at openjdk.org
Fri Jul 21 06:40:43 UTC 2023
On Fri, 21 Jul 2023 06:32:30 GMT, Varada M <duke at openjdk.org> wrote:
>> Okay and trivial.
>
> Thank you @tstuefe @backwaterred
> @varada1110 @tstuefe, I am thinking of this in relation to the discussion on [this PR](https://github.com/openjdk/jdk/pull/14904#issuecomment-1640558149). Are these stacks not-page aligned, or would making the change suggested there affect the validity of this test on AIX?
Good question; I am not sure. It could be that the stacks are 4K page aligned but that our "64K-paged-lie" makes this appear as if the stack were unaligned.
But getting the stack dimensions is tricky. See `AixMisc::query_stack_bounds_for_current_thread`. I dimly remember that the stack had a structure embedded at the end, and the usable stack area started after that. So the dimensions may be unaligned independently from that.
Easy test: run a JVM with `-Xlog:os+thread` and check the thread-started infos, those should show the stack boundaries too. If they are all 4K aligned, run this test with `-XX:-Use64KPages` and check if the test now succeeds.
-------------
PR Comment: https://git.openjdk.org/jdk/pull/14923#issuecomment-1645057554
More information about the hotspot-runtime-dev
mailing list