Add initial support for ARM MTE

[ayrtonm]

This PR adds IA2_USE_MPK and IA2_USE_MTE macros to libia2 and replaces calls to pkey_mprotect with an arch-independent function for tagging loaded writable ELF segments with a memory key. On ARM this uses addg to increment a pointer through the range of pages passed to mprotect and st2g to tag the memory.

This should already pass tests because the #ifdefs used in CI ignore ARM for now, but I'd like to at least build libia2 for ARM in CI before merging this.

edit: nginx build failure may currently be expected(?)

[ayrtonm]

Looks like CI works as expected. Also I think the addgs in insert_tag may be a security issue but I need to read a bit more about what kind of CFI we expect to require and how it'll be implemented.

[ayrtonm]

I'm still trying to figure out if the addgs are a security issue but either way I think this is lower priority than setting up CI for arm. As a first step we could just build libia2 for aarch64 on donna. I'm not sure if there's a clean way to build for two archs with one cmake config step so we might just do one config per arch.

[ayrtonm]

I got libia2 building in CI with a cross-compiler for aarch64 on x86. I see the preprocessor warnings I added for IA2_USE_MTE so it is building for MTE, but we might want to test a bit more thoroughly even before the ARM call gates are ready. I'm still trying to think if there's a good intermediate test between this and running a compartmentalized program in user-mode qemu though.

[sim-immunant]

We should take a closer look at the behavior of -ffixed-x18. In my armtest repo, even when compiling with that flag, X18 still seems to get saved and restored between function calls. System libraries also frequently alter x18, so we'll have to compile the whole shebang from libc up with this flag as well.

[ayrtonm]

While working on getting CMake working with the correct compilers for each target I realized I missed some #ifdefs in ia2_compartment_init.inc. Also it's probably better to add individual #warnings at each location where we're currently missing ARM64 code.

[ayrtonm]

The issue with CMake auto-adding -I/usr/include to tests was caused by LLVM include dirs and fixed by replacing include_directories in rewriter/CMakeLists.txt with the usual target_include_directories. The whole repo doesn't really support cross-compiling correctly because CMake makes it a pain to use more than one toolchain and the "just set CMAKE_C_COMPILER" hack I used for pad-tls won't work for the rewriter, so instead we need to use ExternalProject_Add for host tools.