libc/minimal: provide aligned allocators

[cfriedt]

This change adds support for standardized aligned allocation APIs via posix_memalign(3) and aligned_alloc(3) in the minimal libc.

Fixes #29519

[maxbachmann]

The biggest issue that I can see is that sys_mem_pool does not support aligned allocations, and it is non-obvious to me how to make it support aligned allocations efficientlly. The easiest way to support aligned allocations is to use the k_heap API (which uses sys_heap* under the hood.

In #28611 the sys_mem_pool allocator gets removed

[andrewboie]

mis-click sorry

can you set this as a Draft PR since it is still WIP

[cfriedt]

The biggest issue that I can see is that sys_mem_pool does not support aligned allocations, and it is non-obvious to me how to make it support aligned allocations efficientlly. The easiest way to support aligned allocations is to use the k_heap API (which uses sys_heap* under the hood.

In #28611 the sys_mem_pool allocator gets removed

Now it all makes sense! Thanks for pointing that out @maxbachmann

[cfriedt]

I cherry-picked from #30003 #28611, so please wait until after those are merged before reviewing.
cc @maxbachmann @andyross

[cfriedt]

Leaving this as a draft until #30003 and #28611 are merged.

[cfriedt]

Closing this in favour of #30363 for a couple of reasons.

• Newlib does not provide us with aligned_alloc(3) or posix_memalign(3) so it's pointless to offer in the minimal libc
• @andrewboie said I should just provide a k_malloc_aligned() instead and I probably should have just listened to him in the first place 😉 but it's hard to remember random github comments on old PRs when jumping all over the stack.

In any case, #30363 is really just recycling the aligned_alloc(3) implementation and test case from here and applying them to k_malloc() instead, so I didn't really waste time / code... right? 🤷‍♂️

[pfalcon]

Newlib does not provide us with aligned_alloc(3) or posix_memalign(3) so it's pointless to offer in the minimal libc

I don't agree with the "pointless" valuation: POSIX provides them, that's pretty good a reason on its own.

That said, you're designer of this feature.

[cfriedt]

Newlib does not provide us with aligned_alloc(3) or posix_memalign(3) so it's pointless to offer in the minimal libc

I don't agree with the "pointless" valuation: POSIX provides them, that's pretty good a reason on its own.

Sorry, - it's definitely not "pointless". You're right. For my current purposes though (#29029) I unfortunately can't use posix_memalign() because it's not both in newlib and in the minimal libc.

I'll keep the branch around just in case we want to add this in the future. Personally, I think supporting standards are important.

[pfalcon]

I unfortunately can't use posix_memalign() because it's not both in newlib and in the minimal libc.

That at least should lead to traces of ideas for some kind of "polyfills" (curse JavaScript for abusing thru the entire industry), where we can add functionality for both newlib and minlibs. Actually, we do things like that already - newlib doesn't ship e.g. select() implementation, we do.

(That's not call for action, just call for thinking and maybe planning ahead.)