Hi Greg, Thanks for quick reply. Please see reply inline. Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> 于2018年7月23日周一 下午3:34写道： > > On Mon, Jul 23, 2018 at 03:24:22PM +0200, Jack Wang wrote: > > Hi Greg, > > > > Please consider this patchset, which include block/scsi multiqueue performance > > enhancement and bugfix. > > What exactly is the performance enhancement? How can you measure it? > How did you measure it? I'm testing on SRP/IBNBD using fio, I've seen +10% with mix IO load, and 50% improvement on small IO (bs=512.) with the patchset. > > Which one(s) are the bugfixes in this series? Why not just submit those > separately as they should be "obvious fixes", right? We had queue stall problem on shared hctx restart, my colleague Roman reported to upstream: https://lkml.org/lkml/2017/10/18/263 the patch wasn't taken by upstream, and a bigger patchset from Lei Ming was accepted instread with revert and further bugfix later on. following patches meant for improving performance on scsi-mq, and also contains the attempt to fix the queue stall problem 358a3a6bccb7 ("blk-mq: don't handle TAG_SHARED in restart") blk-mq-sched: move actual dispatching into one helper blk-mq: introduce .get_budget and .put_budget in blk_mq_ops sbitmap: introduce __sbitmap_for_each_set() blk-mq-sched: improve dispatching from sw queue scsi: allow passing in null rq to scsi_prep_state_check() scsi: implement .get_budget and .put_budget for blk-mq SCSI: don't get target/host busy_count in scsi_mq_get_budget() blk-mq: don't handle TAG_SHARED in restart blk-mq: don't restart queue when .get_budget returns BLK_STS_RESOURCE blk-mq: don't handle failure in .get_budget Following is independent fix: blk-mq: quiesce queue during switching io sched and updating nr_requests blk-mq: fix issue with shared tag queue re-running These are clearup for flush request handling: blk-flush: don't run queue for requests bypassing flush block: pass 'run_queue' to blk_mq_request_bypass_insert blk-flush: use blk_mq_request_bypass_insert() blk-mq-sched: decide how to handle flush rq via RQF_FLUSH_SEQ blk-mq: move blk_mq_put_driver_tag*() into blk-mq.h blk-mq: don't allocate driver tag upfront for flush rq The left are following up fixes. > > And why didn't you cc: the subsystem maintainer and mailing list here? > I need their ack before I could take these. Sorry, my mistake, now add Jens, linux-block, ilnux-scsi in reply. Sorry for confusion. > > thanks, > > greg k-h Regards, Jack Wang
