I believe that the general preference is to put this kind of asserts into .c file and not in .h. It is enough to get single error from the build, instead of 40+ from each instance of compiler in the parallel build.

Andrew, WRT to new extensions, please look how Intel extensions are handled on x86, in particular getcontextx(3). Indeed we store a pointer to extended part of XSAVE area into existing mcontext(32).

In D32958#744229, @jhb wrote:

I would be fine with moving the assertions to a C file (probably arm64/arm64/machdep.c next to get_mcontext? The 32-bit assertion could perhaps be next to get_mcontext32)

I was going to add several more. My plan was just to add them all to the top of exec_machdep.c on all the architectures, with appropriate ifdefs for 32-bit compat stuff since I was also going to assert on ucontext_t size and sigframe as well.

exec_machdep.c seems to be the place that has all the MS signal handling on all the platforms, so it seems like the right place for these. I did one as an example, but will do the others in a similar manner once this gets signed off on.
I can also add an assert for the UCONTEXT too. bsd-user just cares about mcontext at compile time (though arguably it should care about both).