Add aarch64 support

[rickvantooster]

No description provided.

[0dminnimda]

Nice! That works on termux aarch64, except their gcc reports an error for (void)fd, a guard will fix it:

#ifndef __ANDROID__
    (void) fd;
#endif

---

I really don't know arm, but this (https://github.com/ARM-software/abi-aa/blob/main/aapcs64/aapcs64.rst#611general-purpose-registers) document on aarch64 suggests that:

A subroutine invocation must preserve the contents of the registers r19-r29 and SP. All 64 bits of each value stored in r19-r29 must be preserved.

And for SIMD/floating-point:

Registers v8-v15 must be preserved by a callee across subroutine calls; [...]. Additionally, only the bottom 64 bits of each value stored in v8-v15 need to be preserved; it is the responsibility of the caller to preserve larger values.

And if I read and understood the code correctly, you are saving r19-r30, but what about v8-v15?

[rickvantooster]

And if I read and understood the code correctly, you are saving r19-r30, but what about v8-v15?

Thank you for pointing it out. They are now being saved.
I've also added the guard around the fd void cast. However shouldn't

(void)rsp;

in coroutine_restore_context also have the same guard?

I also added the inline assembly to move fd to the correct registers in sleep_write and sleep_read as I forgot to add those in my initial commit 🥲.

[0dminnimda]

in coroutine_restore_context also have the same guard?

yes, all of the cases with (void)... inside naked functions. It's like with clang, compilers don't want to see any code except inline assembly inside naked functions.

[0dminnimda]

Don't you also need to reserve space for r30 here, since the load will read both r29 and r30?

[rickvantooster]

Don't you also need to reserve space for r30 here, since the load will read both r29 and r30?

r30 is the same as the link register so the line

*(--rsp) = coroutine__finish_current;

will already reserve space for it.

[0dminnimda]

Ok, yeah, we don't need to preserve it's value, so we can just read some garbage into it, instead of the previous value. Now but what about storing to the stack, wouldn't it override *(--rsp) = f; // push r0?

[rickvantooster]

I am not sure if I am following you. So far everything runs perfectly fine on the raspberry pi 4 model B.