From mboxrd@z Thu Jan 1 00:00:00 1970 From: james.smart@broadcom.com (James Smart) Date: Mon, 10 Jul 2017 16:49:47 -0700 Subject: [PATCH 2/7] nvme/fc: Fix admin queue depth setup In-Reply-To: References: <1499444581-28268-1-git-send-email-keith.busch@intel.com> <1499444581-28268-3-git-send-email-keith.busch@intel.com> Message-ID: <12688ebf-3874-5c91-a93b-10c46ecbf8f7@broadcom.com> On 7/10/2017 4:40 PM, James Smart wrote: > On 7/7/2017 9:22 AM, Keith Busch wrote: >> The NVME_FC_AQ_BLKMQ_DEPTH is one less than what we actually want the >> admin queue's depth to be since it subtracts the AEN entry. >> >> We also need to subtract 1 from the blk-mq depth to always leave an >> empty queue entry, and to not collide command id with the AEN request. > > I don't have an issue with what changed, but have a few comments. > > Why not kill the NVME_FC_NR_AEN_COMMANDS define and correct the > NVME_FC_AQ_BLKMQ_DEPTH define with NVME_NR_AERS as well ? > > Given AEN_CMDID_BASE is NVME_FC_AQ_BLKMQ_DEPTH + 1 there never should > have been a collision regardless. > > Agree with the -1 for empty queue levels. Although the reserved tag > for Connect assured it was never an issue. > > Note: the logic for the above are in the rdma transport as well. > > Reviewed-by: James Smart > > -- james > ah - patch 4 did what I suggested. Looks good.