[ofa-general] [PATCH v2] IB/mlx4: Add send with invalidate support

[Roland Dreier]

 > No reason for them to be different. Roland already suggested to use a
 > union here although he defines the union locally inside the containing
 > struct thus he has two definitions for the same union. Roland do you
 > intend to commit that?

I can if everyone agrees with it.

I can't think of a good way to describe the union independently, so I
think I'll keep it as being duplicated between the WR and completion
structures.

 - R.