[openib-general] Re: mthca FMR correctness (and memory windows)

[Michael S. Tsirkin]

Quoting Roland Dreier <rdreier at cisco.com>:
> If you want to invalidate each FMR individually, then there's not much
> point in using FMRs at all.  Just register and unregister memory regions.

At least the way its implemented currently, invalidating FMRs individually
is likely to be faster than regular MRs, especially on memfree silicon:
you need at least 2 roundtrips to hardware to create an MR, FMR can be mapped
with memory writes. Invalidating both an MR and FMR needs one command in
both cases.

-- 
Michael S. Tsirkin
Staff Engineer, Mellanox Technologies