feat(starboard): Add prctl wrapper for VMA naming

[sherryzy]

This change introduces a wrapper for the prctl system call to support naming Virtual Memory Areas (VMAs) using PR_SET_VMA_ANON_NAME.

When the underlying kernel does not support this feature, the wrapper provides a fallback mechanism that writes the VMA information to a temporary file (/tmp/cobalt_vma_tags_<pid>.txt). This allows for memory debugging and attribution even on systems without the latest kernel features.

A new NPLB test, PosixPrctlTest.SetVmaAnonName, has been added to verify both the direct prctl call and the fallback mechanism.

The prctl wrapper is now used in various parts of the codebase, including the partition allocator and persistent memory allocator, under the IS_STARBOARD build flag.

Bug: 443798603, 444471844

[kaidokert]

This seems like a magic number - what's the origin of it ? Does upstream musl have this defined ?

EDIT: Actually i checked, it's not there. I'd recommend making a pull request to musl to add it: https://github.com/kraj/musl/blob/kraj/master/include/sys/prctl.h

[sherryzy]

I get this number from "linux/prctl.h", and it should be consistent to all Linux kernel systems.

[sherryzy]

@kaidokert I created this PR to update upstream code: kraj/musl#9, though I'm not sure how is the reviewing process going, and I can not add reviewers.

[kaidokert]

They replied - they need patches emailed to mailing list. Different contribution workflow

[sherryzy]

Friendly ping..

[sherryzy]

/generate-commit-message

[github-actions]

🤖 Gemini Suggested Commit Message

---

starboard: Add prctl wrapper for VMA naming

This change introduces a wrapper for the prctl system call within the
Starboard abstraction layer to support naming Virtual Memory Areas (VMAs)
using PR_SET_VMA_ANON_NAME. This improves memory debugging and attribution
on Starboard platforms.

For kernels that do not support PR_SET_VMA_ANON_NAME, a fallback mechanism
is provided. It writes VMA information to a temporary file, ensuring memory
debugging capabilities are available universally. The wrapper is now utilized
in various parts of the codebase, including the partition and persistent
memory allocators, under the IS_STARBOARD build flag.

A new NPLB test, PosixPrctlTest.SetVmaAnonName, has been added to verify
both the direct prctl call and the fallback mechanism.

[niranjanyardi]

We would like to get this in so that smaps works on evergreen-rdk platform : https://issuetracker.google.com/454095852 RDK - Memory leak during endurance tests on 26.eap

talked to @johnedocampo - john already added multiple functions under prctl - so vma(being added in this PR) would be an addition to that . John also mentioned some changes will be needed on this PR to make it work.

@sherryzy when you are back - it'll be good to co-ordinate with John and get this in.

[jellefoks]

I find a few things at issue with this:

• The base name of the file should be a const string so that it doesn't have to be repeated in the different places where it's needed.
• The path should not be hardcoded to /tmp because tempdir may not be /tmp on all devices,
• IMHO this logic should live above Starboard in the Cobalt layer wrapper.
• The file should also not be just stored in tmpdir because tmpdir system-wide readable for all processes, and we don't know if the umask is set to block others from reading the file, and this information should probably no tbe exposed to every process on the device. A good alternative is probably mkstemps (we can add the implementation from musl).