[Differential] [Updated] D1833: Add memory barriers to buf_ring

[kmacy (Kip Macy)]

kmacy added a comment.

I'm mostly echoing Ian Lepore's comments here:

- The wmb() is equivalent to an atomic_store_acq_32 - which is not part of the API. It would seem we'd rather have a atomic_load_acq_32 where the value is actually later to be read. 

- I don't know what the rmb() is protecting against.

- This penalizes architectures that don't need it so at the very least a local macro needs to be defined that is a no-op on all other architectures. 

Are you actually seeing problems that is fixed by this change?

REVISION DETAIL
  https://reviews.freebsd.org/D1833

To: zbb, imp, rpaulo, kmacy
Cc: andrew, ian, adrian, freebsd-arm