mlx4_ib: add blue flame support for kernel consumers

mlx4_ib: add blue flame support for kernel consumers

Using blue flame can improve latency by allowing the HW to more
efficiently access the WQE. A consumer who wants to use blue flame,
has to create the QP with inline support. When posting a send WR,
the consumer has to set IB_SEND_INLINE in the send flags. This
approach is similar to that take in userspace; that is, in order
to use blue flame you must use inline. However, if the send WR is
too large for blue flame, it will only use inline.

A kernel consumer that creates a QP with inline support, will be
allocated a UAR and a blue flame register. All QP doorbells will
be set to the UAR and blue flame posts to the blue flame register.
We make use of all available registers in a blue flame page.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
(Ported from Mellanox OFED 2.4)

Signed-off-by: Mukesh Kacker <mukesh.kacker@oracle.com>