libcxx: use __SIZEOF_LONG__ == 8 instead of __LP64__

[kib]

Summary

for CHERI.

Requested by: arichardson

[emaste]

We should perhaps put a comment about this (__SIZEOF_LONG__ == 8) idiom in arch.7. @arichardson?

[imp]

In D38178#867428, @emaste wrote:

We should perhaps put a comment about this (__SIZEOF_LONG__ == 8) idiom in arch.7. @arichardson?

I think that would be good in general.

I also think that the commit message comment of

Only 64bit architectures can be supported this way, because libcxx
defines __cxx_contention_t to be int64_t for FreeBSD.

or similar, perhaps terser, should be included in this section of code. I puzzled over why 64-bit was special here and while I could find it in the commit logs, it struck me as being important enough to include:

/* libcxx defines __cxx_contention_t as int64_t, so this only works on 64-bit pointers */

maybe?

[kib]

Add comment

[emaste]

contrib/llvm-project/libcxx/src/atomic.cpp

84

sp: synonyms

[jrtc27]

This still isn't quite right if you're MFC'ing it; MIPS is "special" in-tree and uses int32_t, both for 32-bit and 64-bit... see D26774 which broke ABI on both to make it 32-bit and deviated from upstream's ABI

[kib]

In D38178#867839, @jrtc27 wrote:

This still isn't quite right if you're MFC'ing it; MIPS is "special" in-tree and uses int32_t, both for 32-bit and 64-bit... see D26774 which broke ABI on both to make it 32-bit and deviated from upstream's ABI

But it will work, just causing more spurious wakeups?

Anyway, I can add !defined(__mips__) to the MFC. Are you fine with that?