nvme-rdma: assign completion vector correctly
authorMax Gurtovoy <maxg@mellanox.com>
Tue, 23 Jun 2020 14:55:25 +0000 (17:55 +0300)
committerChristoph Hellwig <hch@lst.de>
Wed, 24 Jun 2020 16:41:19 +0000 (18:41 +0200)
commit032a9966a22a3596addf81dacf0c1736dfedc32a
tree9eb5d953a75f620045ec84b092ec45eef09826c4
parent1b4ad7a50ab06573aa8841217d6a472dc1db2d85
nvme-rdma: assign completion vector correctly

The completion vector index that is given during CQ creation can't
exceed the number of support vectors by the underlying RDMA device. This
violation currently can accure, for example, in case one will try to
connect with N regular read/write queues and M poll queues and the sum
of N + M > num_supported_vectors. This will lead to failure in establish
a connection to remote target. Instead, in that case, share a completion
vector between queues.

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
drivers/nvme/host/rdma.c