[ofa-general] Re: [PATCH 2/3] mcast loopback block

[Roland Dreier]

 > (A) from application point of view, setting this property per QP and not per
 > QP per attach to group makes enough sense, specifically, IPoIB wants this
 > flag to be set for all its attaches since the network stack does loopback in
 > memory.

Yes, it's adequate for the current use but I'm wondering why we want to
choose a more complex implementation that hides hardware capabilities?

 > (B) moving forward with the patches, we'd like to export this feature to
 > user space, and from the previous threads on "special QP features" (eg XRC,
 > Low Latency) it sounds like adding some sort of creation flags (even in the
 > price of new verb such that the existing ABI for qp create still works)
 > might be the way to go.

It's just as easy to add a flag to the multicast attach verb -- maybe
easier, in fact, since there are 16 reserved bits in struct
ib_uverbs_attach_mcast.