[PowerPC] Kernel 64-bit atomics emulation for PPC32

[bdragon]

Summary

Here is the kernel side of the 64-bit atomics emulation needed for D22827 and other optimizations.

jhibbits did the main work, this is just a cleanup and split to kernel-side only.

Tested on an iBook G4 with D22827 applied.

Test Plan

• Excercise all atomics from a test module?

[jhibbits]

sys/kern/subr_atomic64.c

1

Since this is being done only for powerpc, should this not be moved to sys/powerpc/powerpc instead?

sys/sys/_atomic64e.h

1

Same as above for this. And then it could be moved into sys/powerpc/include/atomic.h even.

[jeff]

sys/sys/_atomic64e.h

1

I think it's not unlikely that some other 32bit platform may find it useful.

[jhibbits]

If it makes sense to move them to sys/powerpc later, we can do that.

sys/sys/_atomic64e.h

1

I agree. But if none others will make use of them (I've heard mips32 is UP only, so would probably take a much different route for atomic64), it doesn't make sense to keep it generic.

But, I won't block the review on that.

[imp]

I'd add Kyle Evans to this review.
Mips32 does have SMP, but the specific core has 64bit atomics if anybody wants to revive it.

[imp]

sys/kern/subr_atomic64.c

1

Kyle is working on this for mips32 too...

[kevans]

LGTM, added it to mips o32 with this diff: https://people.freebsd.org/~kevans/mips-atomic64.diff

[kevans]

sys/sys/_atomic64e.h

39

Maybe rename this to HAS_EMULATED_ATOMIC64 or something like that since we'll use it for mips, too.

[jeff]

sys/kern/subr_atomic64.c

57

There is alright a macro for this. CACHE_LINE_SIZE

[kevans]

sys/sys/_atomic64e.h

44

These declarations should also be included in the above _KERNEL check to hide them from userland

[imp]

sys/kern/subr_atomic64.c

57

We need to make sure this can be option in kernel config file... unless CACHE LINE SIZE already does that...