IB/IPoIB: Set ah valid flag in multicast send flow
authorDenis Drozdov <denisd@mellanox.com>
Sun, 29 Jul 2018 08:42:28 +0000 (11:42 +0300)
committerJason Gunthorpe <jgg@mellanox.com>
Wed, 1 Aug 2018 21:23:03 +0000 (15:23 -0600)
commit75da96067ade4e7854379ec2f7834f3497652b1a
treec2d53306b5adefede056c4e352dceac8d6714097
parent0f50d88a6e9ae6d9dd14ed1a7d6b309280a9c23b
IB/IPoIB: Set ah valid flag in multicast send flow

The change of ipoib_ah data structure with adding "valid" flag and
checks of ah->valid in ipoib_start_xmit affected multicast packet flow.

Since the multicast flow doesn't invoke path_rec_start, "ah->valid" flag
remains unset, so that ipoib_start_xmit end up with neigh_refresh_path
instead of sending the packet using neigh.

"ah->valid" has to be set in multicast send flow. As a result IPoIB
starts sending packets via neigh immediately and eliminates 60sec delay
of neigh keep alive interval.

The typical example of this issue are two sequential arpings:

arping 11.134.208.9 -> got response (mcast_send)
arping 11.134.208.9 -> no response  (ah->valid = 0)

Fixes: fa9391dbad4b ("RDMA/ipoib: Update paths on CLIENT_REREG/SM_CHANGE events")
Signed-off-by: Denis Drozdov <denisd@mellanox.com>
Reviewed-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Feras Daoud <ferasda@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
drivers/infiniband/ulp/ipoib/ipoib_multicast.c