RFR: 8332362: Implement os::committed_in_range for MacOS and AIX [v4]

[Robert Toyonaga]

On Thu, 13 Jun 2024 06:31:33 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:

>>> This version builds on AIX. I can retry our nightly tests.
>> 
>> Thank you!
>
> @roberttoyonaga @TheRealMDoerr Thanks, Martin, for testing. 
> 
> Pity, I was hoping we could finally have real thread stack sizes in NMT on AIX. To see how much thread stack space is actually used.
> 
> My assumption is that mincore just does not work on thread stacks on AIX, since the manpage states "The mincore subroutine returns the primary-memory residency status for regions created from calls made to the mmap subroutine. " and I think the pthread lib does not use mmap but (I guess) brk. 
> 
> @roberttoyonaga Okay, I would still keep the unified version for all POSIX platforms, but exclude AIX (just return). Then, exclude the test for AIX again.

@tstuefe Ok, I've kept the unified version for POSIX platforms and reverted back to the old behaviour for AIX. 

The IBM docs also say mincore will fail if: "The addr parameter is not a multiple of the page size as returned by the sysconf subroutine using the _SC_PAGE_SIZE ".  I wasn't explicitly checking the start address. Perhaps it needed to be rounded up to 64K boundaries in some cases at the cost of some accuracy. Or maybe this isn't an issue because the start address always ends up being correctly aligned on AIX anyway. But I don't have a good way of confirming this.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/19455#issuecomment-2166125867