[ofa-general] Re: [PATCH 3/3 v4] ib/ipoib: blocking multicast loopback ipoib packets
Roland Dreier
rdreier at cisco.com
Thu Jul 3 14:54:30 PDT 2008
> yes, this was one idea (earlier one that is a new QP type eg
> IB_QPT_UD_BML) and the thing is that we would be happy to get your
> take on how can this be done with minimum pain regarding ABI issues
> both up towards libibverbs consumers and down towards ib_uverbs.
Sorry, I don't have a magic solution for you. To minimize pain about
ABI I think the only thing to do is minimize ABI changes.
> Actually, looking on the XRC patch set, as it adds new QP type anyway,
> IB_QPT_UD_BML might turn to be the simplest way here, but its not
> general enough as moving forward, people were talking on other create
> flags they want to use from user space (low latency QP, etc)
If a QP creation flag works for XRC then I would prefer not to add more
QP types I guess. One of the ways that the QP creation flags were
originally sold to me was that XRC needed them anyway. So I'm a bit
confused about where things have ended up.
> Among other things, the XRC patch series adds the
> "ibv_create_xrc_rcv_qp" new verb to libibverbs, so my thought here
> was that a new way X to create QPs is added, and we want a new way Y
> for the block multicast loop. So maybe we can form a framework Z for
> creating QPs other then the old one. Alternatively its possible for
> the create_qp_with_create_flags verb (and other verbs we want to add
> such as cq_modify_moderation_params) to just live next to the new
> verbs added to libibverbs in the new ibv_xrc_ops structure which is
> added for the verbs context.
I don't want to add both an XRC-specific mechanism and a more general
mechanism that XRC could have used but doesn't. So yes the "framework
Z" approach that works for both seems like the best way forward.
- R.
More information about the general
mailing list