treat hermit like wasm32

[m-mueller678]

This implements stacker on hermit by treating it the same as wasm32

[nagisa]

Could you please elaborate on why this implementation precisely is the right thing to do for the hermit targets and why e.g. psm-based implementation does not work? Is hermit running exclusively on esoteric architectures of some sort?

[m-mueller678]

I forgot to mention, this intends to fix #108. If I can get an implementation that just compiles and does nothing, I would be happy too.

I chose the wasm implementation simply because I am not familiar enough with psm to adapt the psm implementation to do the right thing for Hermit.

From what I could gather from the source code, they differ in how they allocate memory and how they query page size. Hermit supports neither mmap nor sysconf, so I went with the wasm implementation.

Hermit usually runs on x86_64 and aarch64, though there might also be other supported architectures.

[nagisa]

Ah so my bad. I didn't check the PR closely enough, and it looks like psm continues to be used. What psm does is entirely just switching stack pointers without worrying about how the stacks have been allocated. For all it cares, the stack could've been allocated with malloc. For systems like Linux malloc isn't super appropriate as there is no way to add a guard page to protect against stack overflows. If that's fine for hermit then I have no problem with such approach.

That said, using wasm implementation everywhere isn't always appropriate. For instance WASM has 64k pages by requirement in the specification. On code running on x86_64/arm the page size is most likely going to be smaller, though it probably won't hurt anybody if too large of a page size is specified. If hermit has no way to query the page size, then 4k is likely to be the more correct choice (though for malloc-based implementation it probably matters little either way -- might as well say the page size is 1.)

[nagisa]

If there is absolutely no way to obtain the page size on hermit then lets hardcode it to something more appropriate for platforms hermit runs on (probably 4kB.)

[m-mueller678]

Is page_size used for anything but controlling the size of allocations? If not, renaming to allocation_granularity or such might be appropriate.

Hermit uses Talc as an Allocator. Here is how Talc rounds up allocation sizes. As Talc stores some metadata along with each allocation, picking a power of two will have it just barely not fit within the size category. I could not find any special treatment for page size allocations, though maybe I missed it. Given this, an allocation granularity of 1 would make sense.