RFR: 8280056: gtest/LargePageGtests.java#use-large-pages failed "os.release_one_mapping_multi_commits_vm" [v2]
Afshin Zafari
azafari at openjdk.org
Fri Jan 12 08:24:31 UTC 2024
On Mon, 23 Oct 2023 14:47:47 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:
>>> release the middle part of
>>
>> It is not possible to release middle of a region, since NMT rejects it. The `release_multi_mappings` won't run if NMT is not off. Same as this `release_one_mapping_multi_commits` test.
>
>> > release the middle part of
>>
>> It is not possible to release middle of a region, since NMT rejects it. The `release_multi_mappings` won't run if NMT is not off. Same as this `release_one_mapping_multi_commits` test.
>
> Yes, having NMT on is a big part of the test.
>
> You could reserve the border mapping separately. Even just a trailing mapping would be sufficient to prevent intermittent errors.
>
> For example:
>
> address p = reserve_one_commit_multiple(num_stripes, stripe_len);
> ASSERT_NE(p, (address)NULL);
>
> // this can fail; if it does, skip test. At the very end of the test, release border separately.
> address border = os::attempt_res_at (p + num_stripes*stripe_len);
>
> --
>
> That, and / or the repeat-20times-idea, I would try
Thank you @tstuefe, @dholmes-ora and @gerard-ziemski for your comments.
-------------
PR Comment: https://git.openjdk.org/jdk/pull/16240#issuecomment-1888637830
More information about the hotspot-runtime-dev
mailing list