On Mon, 11 May 2026 16:00:48 GMT, Robert Toyonaga <[email protected]> wrote:
> ### Summary > `os::committed_in_range` is used by NMT to account for thread stacks. The > name is misleading, it actually is meant to find **live** pages not just > **committed** pages. On POSIX (except AIX) this works correctly using > `mincore`. On Windows this worked incorrectly using `VirtualQuery` (only > finds committed, not live regions). This means that the values reported on > Windows were inaccurate. > > This PR fixes the Windows implementation by using `QueryWorkingSetEx` instead > of `VirtualQuery`. This properly identifies pages that are truly live in the > working set, not just committed. I also renamed `committed_in_range` to > `resident_in_range` so that the intention is clearer. I have tried to > structure the Windows implementation as similarly to Posix as possible to > help with maintainability. > > ### Testing > - Tested on windows and linux and64 > - `make test TEST=gtest:NMTCommittedVirtualMemory` (this tests > `live_in_range` directly) > - `make test > TEST=test/hotspot/jtreg/runtime/Thread/TestAlwaysPreTouchStacks.java` (this > tests `resident_in_range` indirectly through NMT thread stack accounting) > - `make test TEST=gtest:os` > - `make test TEST=hotspot_nmt` > - tier 1 > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). This pull request has now been integrated. Changeset: 79923019 Author: Robert Toyonaga <[email protected]> Committer: Thomas Stuefe <[email protected]> URL: https://git.openjdk.org/jdk/commit/79923019f58af0cba44becfc22975452a128b52a Stats: 106 lines in 5 files changed: 32 ins; 5 del; 69 mod 8334403: Unify behavior of os::committed_in_range across OSes Reviewed-by: stuefe, jsjolen ------------- PR: https://git.openjdk.org/jdk/pull/31124
