Simplify lazy advance and decouple it somewhat from ticks so that it canbe forwarded after other broadcast interrupts.

[jeff]

Summary

This uses an atomic_cmpset_64 to simplify lazy_advance. This will allow me to advance lazy without advancing ticks which will be useful for blocking polls on lazy smrs.

[jeff]

sys/kern/subr_smr.c

186 ↗

(On Diff #68781)

This would have a bug if you acquired a spinlock and blocked the interrupt for 2 ticks without the processor flushing the store buffer.

This seems really unlikely but perhaps needs an assert or some other protection.

477 ↗

(On Diff #68781)

I think ultimately poll will have some top-level driver function that handles the sleep/advance for deferred and lazy and this function can be simplified slightly.

So many polls don't require actual looping that it may be advantageous to make a small poll pre-check and return fast-path.

sys/sys/smr.h

59 ↗

(On Diff #68781)

I feel like this is a little ugly but ok. If someone has another suggestion to ensure the grouping and alignment I'm all ears.

[rlibby]

Looks good.

sys/kern/subr_smr.c

249 ↗

(On Diff #68781)

(While looking at the generated code, I noticed amd64 atomic_cmpset forces us to populate a register with the return value of cmpset, even though we don't care. I'll see what kib thinks of a possible fix for that...)

sys/sys/smr.h

59 ↗

(On Diff #68781)

I know that gcc explicitly blesses type punning with a union, but from what I was able to find, I think clang does not (e.g. https://reviews.llvm.org/D30538). However, we seem to use -fno-strict-aliasing, so I think we should be okay anyway. I did look at the (amd64) code generation and it looked fine as far as I was able to tell.

An alternative would be to _Static_assert the size, drop the union, and do the read/write like

uint64_t old, new;
old = atomic_load_acq_64(&s->s_wr);
memcpy(&s_wr, &old, sizeof());
...
memcpy(&new, &s_wr, sizeof());
atomic_cmpset_64(&s->s_wr, &old, new);

But one may find this to be even uglier.