Misalinged stack on syscall wrapper calls

[difcsi]

I ran into compilation issues with the do_syscall implementations, resulting in runtime SEGfaults.

I drafted a PR fixing this in #22, but that doesn't solve the root cause, just happens to trigger a lucky emission change from GCC.

Instantiation of generic_syscall structs, the first thing in raw_syscall functions, may result in a segfault due to a misaligned stack at the beginning of a function. As per ABI, the compiler expects 16-byte alignment at the start of the function calls and optimises by emitting movaps instructions that fail due to an apparent 8-byte offset.

I suspect the alignment macros at raw_syscalls_impl.h have something to do with this, I'll investigate when I find some time for it.

As for why this worked so far, I suspect an emission change in GCC. Its worth noting that the issue occurred whilst upgrading xed and mbuild submodules in contrib, but I don't immediately see how that may be relevant.

[stephenrkell]

Well done on debugging this. Yes, if modern GCC is using movaps where older versions didn't, it would surface a stack pointer misalignment bug.

Remember that if the stack pointer is misaligned at the very beginning of the function, that indicates a bug in the caller (or its caller, or...).