]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/commit
fabrics: define admin sqsize min default, per spec
authorJay Freyensee <james_p_freyensee@linux.intel.com>
Wed, 17 Aug 2016 22:00:26 +0000 (15:00 -0700)
committerSagi Grimberg <sagi@grimberg.me>
Thu, 18 Aug 2016 06:58:05 +0000 (09:58 +0300)
commitf994d9dc28bc27353acde2caaf718222d92a3e24
treef54079f734752cc79dc3269a2dd3de2370200d9f
parentb825b44c4ef4dabfdaf4e82db2263d377ac45d67
fabrics: define admin sqsize min default, per spec

Upon admin queue connect(), the rdma qp was being
set based on NVMF_AQ_DEPTH.  However, the fabrics layer was
using the sqsize field value set for I/O queues for the admin
queue, which threw the nvme layer and rdma layer off-whack:

root@fedora23-fabrics-host1 nvmf]# dmesg
[ 3507.798642] nvme_fabrics: nvmf_connect_admin_queue():admin sqsize
being sent is: 128
[ 3507.798858] nvme nvme0: creating 16 I/O queues.
[ 3507.896407] nvme nvme0: new ctrl: NQN "nullside-nqn", addr
192.168.1.3:4420

Thus, to have a different admin queue value, we use
NVMF_AQ_DEPTH for connect() and RDMA private data
as the minimum depth specified in the NVMe-over-Fabrics 1.0 spec
(and in that RDMA private data we treat hrqsize as 1's-based
value, per current understanding of the fabrics spec).

Reported-by: Daniel Verkamp <daniel.verkamp@intel.com>
Signed-off-by: Jay Freyensee <james_p_freyensee@linux.intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
drivers/nvme/host/fabrics.c
drivers/nvme/host/rdma.c