Hi Jens This is the second version patchset to make the kyber io scheduler more firendly with merging. Most of time, kyber io scheduler will not leave any requests in ctx rq_list, this is because even if tokens of one domain is used up, kyber will try to dispatch requests from other domain and flush the rq_list there. So there is not any change to do merge. To improve this, setup kyber_ctx_queue (kcq) which is similar with blk_mq_ctx but has rq_list for every domain, and build mapping between kcq and khd as ctx and hctx. Then we could merge, insert and dispatch on rq_list of different domains separately. At the same time, only flush the rq_list of kcq after get domain token successfully, then the requests could be left in the rq_list of that domain and maybe merged with following io. In my local test on NVMe card, the sequential io performance could be improved a lot, especially on high workload. More details please refer to the second patch. Changes V1 to V2: - Add patch from Jens which abstract out the blk-mq-sched rq list iteration bio merge helper interface and rebase the patch based on it. - merge bio_sched_domain and rq_sched_domain to avoid the duplicated code. - allocate kcqs per khd Jens Axboe (1) 0001-blk-mq-abstract-out-blk-mq-sched-rq-list-iteration-b.patch Jianchao Wang (1) 0002-block-kyber-make-kyber-more-friendly-with-merging.patch block/blk-mq-sched.c | 34 ++++++--- block/kyber-iosched.c | 197 ++++++++++++++++++++++++++++++++++++++++--------- include/linux/blk-mq.h | 3 +- 3 files changed, 188 insertions(+), 46 deletions(-) Thanks Jianchao
