[ewg] Re: [PATCH] [RFC] IB/cache: Add ib_cache report for cache in process

[Roland Dreier]

 > +	atomic_t		coherent;

Why atomic_t?  There's nothing magic that protects against races if
you never do anything but atomic_read and atomic_set -- you're just
using an int here.

Also, why do you only add the stale checking to the P_Key cache
methods?  Shouldn't we be consistent and tell the caller of any cache
lookup whether the data is stale or not?

And I don't see anything that handles the ESTALE return in any of the
users of this code...

 - R.