Why are you using the cpu_hint as the vector? Aren't you suppose
to seach the cq with vector that maps to the cpu?
IIRC my version used ib_get_vector_affinity to locate a cq that
will actually interrupt on the cpu... Otherwise, just call it vector.
You are right, I ended up removing the call to ib_get_vector_affinity
because it wasn't implemented anyway. I think it would be best just to
change it from cpu_hint to comp_vector_hint and then we will get the
desired spread over comp vectors which is what happens anyway without
ib_get_vector_affinity.
Why isn't that supported btw? I see mlx5_comp_irq_get_affinity_mask and
it should be pretty straight forward to have this for all the device
drivers really, isn't it?
