"automatic va_arg instruction only works on Darwin" when running AArch64 tests

[pietroalbini]

When running the rustc test suite on AArch64, the ui/abi/variadic-ffi.rs test fails due to the following LLVM assertion:

rustc: /checkout/src/llvm-project/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp:5726: llvm::SDValue llvm::AArch64TargetLowering::LowerVAARG(llvm::SDValue, llvm::SelectionDAG&) const: Assertion `Subtarget->isTargetDarwin() && "automatic va_arg instruction only works on Darwin"' failed.

[jhass]

Here's how Clang implements va_arg on AArch64: https://github.com/llvm/llvm-project/blob/a574edbba2b24fcfb733aa2d82308131f5b7d2d6/clang/lib/CodeGen/TargetInfo.cpp#L5677-L5921

I think it's best to just disable va_arg on AArch64 (and quite possibly more targets) until LLVM decides to properly implement it. Don't get your hopes up on that though, when I asked in IRC why clang would do this rather than fixing va_arg on AArch64, I got comments in the vein of:

<jhass> it almost feels like va_arg is just broken on aarch64 and rather than fixing that somebody went through the effort of reimplemnting it like this for clang
<nbjoerg> no
<nbjoerg> the IR stuff is really only appropiate and meant for the easy cases
<jhass> a bunch of i32's is an easy case, no? I can't get that to work
<nbjoerg> I don't know the aarch64 ABI from memory
<nbjoerg> in general, the IR is only really applicable if you have a nice stack based calling convention
<jhass> well, the call instruction seems to be implemented fine for the aarch64 variadic calling convention?
<jhass> why is it inappropriate for the va_arg instruction to handle it as well?
<jhass> why does it exist then?
<nbjoerg> jhass: legacy
<nbjoerg> jhass: noone bothered enough to remove it

[pietroalbini]

cc @Stammark @JamieCunliffe @vigoux @joaopaulocarreiro @raw-bin, this is one of the tests that are failing on AArch64.

[pietroalbini]

This was fixed!