From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ming Lei Subject: Re: [PATCH 0/7] hisi_sas: Misc bugfixes and an optimisation patch Date: Thu, 11 Oct 2018 21:32:00 +0800 Message-ID: <20181011133159.GA31355@ming.t460p> References: <1537801594-207139-1-git-send-email-john.garry@huawei.com> <20181011063603.GA3456@infradead.org> <118d01db-05f4-a85b-850c-c4e5a5616a16@huawei.com> <20181011101534.GA31802@infradead.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org To: John Garry Cc: Christoph Hellwig , "Martin K. Petersen" , jejb@linux.vnet.ibm.com, linuxarm@huawei.com, linux-scsi@vger.kernel.org, linux-kernel@vger.kernel.org, chenxiang List-Id: linux-scsi@vger.kernel.org On Thu, Oct 11, 2018 at 02:12:11PM +0100, John Garry wrote: > On 11/10/2018 11:15, Christoph Hellwig wrote: > > On Thu, Oct 11, 2018 at 10:59:11AM +0100, John Garry wrote: > > > > > > > blk-mq tags are always per-host (which has actually caused problems for > > > > ATA, which is now using its own per-device tags). > > > > > > > > > > So, for example, if Scsi_host.can_queue = 2048 and Scsi_host.nr_hw_queues = > > > 16, then rq tags are still in range [0, 2048) for that HBA, i.e. invariant > > > on queue count? > > > > Yes, if can_queue is 2048 you will gets tags from 0..2047. > > > > I should be clear about some things before discussing this further. Our > device has 16 hw queues. And each command we send to any queue in the device > must have a unique tag across all hw queues for that device, and should be > in the range [0, 2048) - it's called an IPTT. So Scsi_host.can_queue = 2048. Could you describe a bit about IPTT? Looks like the 16 hw queues are like reply queues in other drivers, such as megara_sas, but given all the 16 reply queues share one tagset, so the hw queue number has to be 1 from blk-mq's view. > > However today we only expose a single queue to upper layer (for unrelated > LLDD error handling restriction). We hope to expose all 16 queues in future, > which is what I meant by "enabling SCSI MQ in the driver". However, with > 6/7, this creates a problem, below. If the tag of each request from all hw queues has to be unique, you can't expose all 16 queues. > > > IFF you device needs different tags for different queues it can use > > the blk_mq_unique_tag heper to generate unique global tag. > > So this helper can't help, as fundamentially the issue is "the tag field in > struct request is unique per hardware queue but not all all hw queues". > Indeed blk_mq_unique_tag() does give a unique global tag, but cannot be used > for the IPTT. > > OTOH, We could expose 16 queues to upper layer, and drop 6/7, but we found > it performs worse. We discussed this issue before, but not found a good solution yet for exposing multiple hw queues to blk-mq. However, we still get good performance in case of none scheduler by the following patches: 8824f62246be blk-mq: fail the request in case issue failure 6ce3dd6eec11 blk-mq: issue directly if hw queue isn't busy in case of 'none' Thanks, Ming