Bug ID: JDK-8257041 [aix] os::release_memory may not release the full range

Type: Bug
Component: hotspot
Sub-Component: runtime
Affected Version: 8,11,16

Priority: P5
Status: Resolved
Resolution: Won't Fix
OS: aix

Submitted: 2020-11-25
Updated: 2021-06-30
Resolved: 2021-06-30

JDK-8255978 fixed a bug on Windows where it was not possible to release an address range consisting of multiple mappings, and added a regression gtest (os release_multi_mappings).

Ironically, that test now fails on AIX, where we use SystemV shared memory which has the same semantics as the Windows virtual memory API. See also JDK-8257040.

(Note that we may have a similar problem on Linux where we use System V shm for large pages with +UseSHM - but there, this type of memory is restricted to os::reserve_memory_special, and therefore the os.release_one_mapping_multi_commits_vm gtest does not fail, since it only tests the interplay of os::release_memory() with the standard os::reserve_memory()).

No time to fix this. Maybe IBM wants to step in.
30-06-2021
Sorry, I split the macos_aarch64 failure from this bug into its own separate one. https://bugs.openjdk.java.net/browse/JDK-8262952 and reverted this one to the way it was.
03-03-2021
Hi Gerard, could you investigate on MacOs a bit further? On Mac, we use only mmap, which should not have the problem I describe above.
03-03-2021
Need to raise the priority.
02-03-2021
This seems to happen in the new macosx-aarch64 port as well - https://github.com/openjdk/jdk/pull/2200 open/test/hotspot/gtest/runtime/test_os.cpp:376: Failure Value of: os::commit_memory((char*)q, stripe_len, executable) Actual: false Expected: true
02-03-2021

Relates :	JDK-8262952 - [macos_aarch64] os::commit_memory failure
Relates :	JDK-8257042 - [aix] Disable os.release_one_mapping_multi_commits_vm gtest
Relates :	JDK-8257040 - [posix] os::split_reserved_memory will not work as expected on SysV shm
Relates :	JDK-8255978 - [windows] os::release_memory may not release the full range