1bd166ef1SJens Axboe /* 2bd166ef1SJens Axboe * blk-mq scheduling framework 3bd166ef1SJens Axboe * 4bd166ef1SJens Axboe * Copyright (C) 2016 Jens Axboe 5bd166ef1SJens Axboe */ 6bd166ef1SJens Axboe #include <linux/kernel.h> 7bd166ef1SJens Axboe #include <linux/module.h> 8bd166ef1SJens Axboe #include <linux/blk-mq.h> 9bd166ef1SJens Axboe 10bd166ef1SJens Axboe #include <trace/events/block.h> 11bd166ef1SJens Axboe 12bd166ef1SJens Axboe #include "blk.h" 13bd166ef1SJens Axboe #include "blk-mq.h" 14d332ce09SOmar Sandoval #include "blk-mq-debugfs.h" 15bd166ef1SJens Axboe #include "blk-mq-sched.h" 16bd166ef1SJens Axboe #include "blk-mq-tag.h" 17bd166ef1SJens Axboe #include "blk-wbt.h" 18bd166ef1SJens Axboe 19bd166ef1SJens Axboe void blk_mq_sched_free_hctx_data(struct request_queue *q, 20bd166ef1SJens Axboe void (*exit)(struct blk_mq_hw_ctx *)) 21bd166ef1SJens Axboe { 22bd166ef1SJens Axboe struct blk_mq_hw_ctx *hctx; 23bd166ef1SJens Axboe int i; 24bd166ef1SJens Axboe 25bd166ef1SJens Axboe queue_for_each_hw_ctx(q, hctx, i) { 26bd166ef1SJens Axboe if (exit && hctx->sched_data) 27bd166ef1SJens Axboe exit(hctx); 28bd166ef1SJens Axboe kfree(hctx->sched_data); 29bd166ef1SJens Axboe hctx->sched_data = NULL; 30bd166ef1SJens Axboe } 31bd166ef1SJens Axboe } 32bd166ef1SJens Axboe EXPORT_SYMBOL_GPL(blk_mq_sched_free_hctx_data); 33bd166ef1SJens Axboe 34e2b3fa5aSDamien Le Moal void blk_mq_sched_assign_ioc(struct request *rq) 35bd166ef1SJens Axboe { 3644e8c2bfSChristoph Hellwig struct request_queue *q = rq->q; 370c62bff1SJens Axboe struct io_context *ioc; 38bd166ef1SJens Axboe struct io_cq *icq; 39bd166ef1SJens Axboe 400c62bff1SJens Axboe /* 410c62bff1SJens Axboe * May not have an IO context if it's a passthrough request 420c62bff1SJens Axboe */ 430c62bff1SJens Axboe ioc = current->io_context; 440c62bff1SJens Axboe if (!ioc) 450c62bff1SJens Axboe return; 460c62bff1SJens Axboe 470d945c1fSChristoph Hellwig spin_lock_irq(&q->queue_lock); 48bd166ef1SJens Axboe icq = ioc_lookup_icq(ioc, q); 490d945c1fSChristoph Hellwig spin_unlock_irq(&q->queue_lock); 50bd166ef1SJens Axboe 51bd166ef1SJens Axboe if (!icq) { 52bd166ef1SJens Axboe icq = ioc_create_icq(ioc, q, GFP_ATOMIC); 53bd166ef1SJens Axboe if (!icq) 54bd166ef1SJens Axboe return; 55bd166ef1SJens Axboe } 56ea511e3cSChristoph Hellwig get_io_context(icq->ioc); 5744e8c2bfSChristoph Hellwig rq->elv.icq = icq; 58bd166ef1SJens Axboe } 59bd166ef1SJens Axboe 608e8320c9SJens Axboe /* 618e8320c9SJens Axboe * Mark a hardware queue as needing a restart. For shared queues, maintain 628e8320c9SJens Axboe * a count of how many hardware queues are marked for restart. 638e8320c9SJens Axboe */ 647211aef8SDamien Le Moal void blk_mq_sched_mark_restart_hctx(struct blk_mq_hw_ctx *hctx) 658e8320c9SJens Axboe { 668e8320c9SJens Axboe if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) 678e8320c9SJens Axboe return; 688e8320c9SJens Axboe 698e8320c9SJens Axboe set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); 708e8320c9SJens Axboe } 717211aef8SDamien Le Moal EXPORT_SYMBOL_GPL(blk_mq_sched_mark_restart_hctx); 728e8320c9SJens Axboe 7397889f9aSMing Lei void blk_mq_sched_restart(struct blk_mq_hw_ctx *hctx) 748e8320c9SJens Axboe { 758e8320c9SJens Axboe if (!test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) 7697889f9aSMing Lei return; 778e8320c9SJens Axboe clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); 788e8320c9SJens Axboe 7997889f9aSMing Lei blk_mq_run_hw_queue(hctx, true); 808e8320c9SJens Axboe } 818e8320c9SJens Axboe 821f460b63SMing Lei /* 831f460b63SMing Lei * Only SCSI implements .get_budget and .put_budget, and SCSI restarts 841f460b63SMing Lei * its queue by itself in its completion handler, so we don't need to 851f460b63SMing Lei * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE. 861f460b63SMing Lei */ 871f460b63SMing Lei static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx) 88caf8eb0dSMing Lei { 89caf8eb0dSMing Lei struct request_queue *q = hctx->queue; 90caf8eb0dSMing Lei struct elevator_queue *e = q->elevator; 91caf8eb0dSMing Lei LIST_HEAD(rq_list); 92caf8eb0dSMing Lei 93caf8eb0dSMing Lei do { 94de148297SMing Lei struct request *rq; 95caf8eb0dSMing Lei 96f9cd4bfeSJens Axboe if (e->type->ops.has_work && !e->type->ops.has_work(hctx)) 97caf8eb0dSMing Lei break; 98de148297SMing Lei 9988022d72SMing Lei if (!blk_mq_get_dispatch_budget(hctx)) 1001f460b63SMing Lei break; 101de148297SMing Lei 102f9cd4bfeSJens Axboe rq = e->type->ops.dispatch_request(hctx); 103de148297SMing Lei if (!rq) { 104de148297SMing Lei blk_mq_put_dispatch_budget(hctx); 105de148297SMing Lei break; 106caf8eb0dSMing Lei } 107caf8eb0dSMing Lei 108de148297SMing Lei /* 109de148297SMing Lei * Now this rq owns the budget which has to be released 110de148297SMing Lei * if this rq won't be queued to driver via .queue_rq() 111de148297SMing Lei * in blk_mq_dispatch_rq_list(). 112de148297SMing Lei */ 113de148297SMing Lei list_add(&rq->queuelist, &rq_list); 114de148297SMing Lei } while (blk_mq_dispatch_rq_list(q, &rq_list, true)); 115de148297SMing Lei } 116de148297SMing Lei 117b347689fSMing Lei static struct blk_mq_ctx *blk_mq_next_ctx(struct blk_mq_hw_ctx *hctx, 118b347689fSMing Lei struct blk_mq_ctx *ctx) 119b347689fSMing Lei { 120f31967f0SJens Axboe unsigned short idx = ctx->index_hw[hctx->type]; 121b347689fSMing Lei 122b347689fSMing Lei if (++idx == hctx->nr_ctx) 123b347689fSMing Lei idx = 0; 124b347689fSMing Lei 125b347689fSMing Lei return hctx->ctxs[idx]; 126b347689fSMing Lei } 127b347689fSMing Lei 1281f460b63SMing Lei /* 1291f460b63SMing Lei * Only SCSI implements .get_budget and .put_budget, and SCSI restarts 1301f460b63SMing Lei * its queue by itself in its completion handler, so we don't need to 1311f460b63SMing Lei * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE. 1321f460b63SMing Lei */ 1331f460b63SMing Lei static void blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx) 134b347689fSMing Lei { 135b347689fSMing Lei struct request_queue *q = hctx->queue; 136b347689fSMing Lei LIST_HEAD(rq_list); 137b347689fSMing Lei struct blk_mq_ctx *ctx = READ_ONCE(hctx->dispatch_from); 138b347689fSMing Lei 139b347689fSMing Lei do { 140b347689fSMing Lei struct request *rq; 141b347689fSMing Lei 142b347689fSMing Lei if (!sbitmap_any_bit_set(&hctx->ctx_map)) 143b347689fSMing Lei break; 144b347689fSMing Lei 14588022d72SMing Lei if (!blk_mq_get_dispatch_budget(hctx)) 1461f460b63SMing Lei break; 147b347689fSMing Lei 148b347689fSMing Lei rq = blk_mq_dequeue_from_ctx(hctx, ctx); 149b347689fSMing Lei if (!rq) { 150b347689fSMing Lei blk_mq_put_dispatch_budget(hctx); 151b347689fSMing Lei break; 152b347689fSMing Lei } 153b347689fSMing Lei 154b347689fSMing Lei /* 155b347689fSMing Lei * Now this rq owns the budget which has to be released 156b347689fSMing Lei * if this rq won't be queued to driver via .queue_rq() 157b347689fSMing Lei * in blk_mq_dispatch_rq_list(). 158b347689fSMing Lei */ 159b347689fSMing Lei list_add(&rq->queuelist, &rq_list); 160b347689fSMing Lei 161b347689fSMing Lei /* round robin for fair dispatch */ 162b347689fSMing Lei ctx = blk_mq_next_ctx(hctx, rq->mq_ctx); 163b347689fSMing Lei 164b347689fSMing Lei } while (blk_mq_dispatch_rq_list(q, &rq_list, true)); 165b347689fSMing Lei 166b347689fSMing Lei WRITE_ONCE(hctx->dispatch_from, ctx); 167b347689fSMing Lei } 168b347689fSMing Lei 1691f460b63SMing Lei void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx) 170bd166ef1SJens Axboe { 17181380ca1SOmar Sandoval struct request_queue *q = hctx->queue; 17281380ca1SOmar Sandoval struct elevator_queue *e = q->elevator; 173f9cd4bfeSJens Axboe const bool has_sched_dispatch = e && e->type->ops.dispatch_request; 174bd166ef1SJens Axboe LIST_HEAD(rq_list); 175bd166ef1SJens Axboe 176f4560ffeSMing Lei /* RCU or SRCU read lock is needed before checking quiesced flag */ 177f4560ffeSMing Lei if (unlikely(blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q))) 1781f460b63SMing Lei return; 179bd166ef1SJens Axboe 180bd166ef1SJens Axboe hctx->run++; 181bd166ef1SJens Axboe 182bd166ef1SJens Axboe /* 183bd166ef1SJens Axboe * If we have previous entries on our dispatch list, grab them first for 184bd166ef1SJens Axboe * more fair dispatch. 185bd166ef1SJens Axboe */ 186bd166ef1SJens Axboe if (!list_empty_careful(&hctx->dispatch)) { 187bd166ef1SJens Axboe spin_lock(&hctx->lock); 188bd166ef1SJens Axboe if (!list_empty(&hctx->dispatch)) 189bd166ef1SJens Axboe list_splice_init(&hctx->dispatch, &rq_list); 190bd166ef1SJens Axboe spin_unlock(&hctx->lock); 191bd166ef1SJens Axboe } 192bd166ef1SJens Axboe 193bd166ef1SJens Axboe /* 194bd166ef1SJens Axboe * Only ask the scheduler for requests, if we didn't have residual 195bd166ef1SJens Axboe * requests from the dispatch list. This is to avoid the case where 196bd166ef1SJens Axboe * we only ever dispatch a fraction of the requests available because 197bd166ef1SJens Axboe * of low device queue depth. Once we pull requests out of the IO 198bd166ef1SJens Axboe * scheduler, we can no longer merge or sort them. So it's best to 199bd166ef1SJens Axboe * leave them there for as long as we can. Mark the hw queue as 200bd166ef1SJens Axboe * needing a restart in that case. 201caf8eb0dSMing Lei * 2025e3d02bbSMing Lei * We want to dispatch from the scheduler if there was nothing 2035e3d02bbSMing Lei * on the dispatch list or we were able to dispatch from the 2045e3d02bbSMing Lei * dispatch list. 20564765a75SJens Axboe */ 206caf8eb0dSMing Lei if (!list_empty(&rq_list)) { 207caf8eb0dSMing Lei blk_mq_sched_mark_restart_hctx(hctx); 208b347689fSMing Lei if (blk_mq_dispatch_rq_list(q, &rq_list, false)) { 209b347689fSMing Lei if (has_sched_dispatch) 2101f460b63SMing Lei blk_mq_do_dispatch_sched(hctx); 211b347689fSMing Lei else 2121f460b63SMing Lei blk_mq_do_dispatch_ctx(hctx); 213b347689fSMing Lei } 214caf8eb0dSMing Lei } else if (has_sched_dispatch) { 2151f460b63SMing Lei blk_mq_do_dispatch_sched(hctx); 2166e768717SMing Lei } else if (hctx->dispatch_busy) { 2176e768717SMing Lei /* dequeue request one by one from sw queue if queue is busy */ 2181f460b63SMing Lei blk_mq_do_dispatch_ctx(hctx); 219caf8eb0dSMing Lei } else { 220caf8eb0dSMing Lei blk_mq_flush_busy_ctxs(hctx, &rq_list); 221de148297SMing Lei blk_mq_dispatch_rq_list(q, &rq_list, false); 222c13660a0SJens Axboe } 223bd166ef1SJens Axboe } 224bd166ef1SJens Axboe 225e4d750c9SJens Axboe bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, 226e4d750c9SJens Axboe struct request **merged_request) 227bd166ef1SJens Axboe { 228bd166ef1SJens Axboe struct request *rq; 229bd166ef1SJens Axboe 23034fe7c05SChristoph Hellwig switch (elv_merge(q, &rq, bio)) { 23134fe7c05SChristoph Hellwig case ELEVATOR_BACK_MERGE: 232bd166ef1SJens Axboe if (!blk_mq_sched_allow_merge(q, rq, bio)) 233bd166ef1SJens Axboe return false; 23434fe7c05SChristoph Hellwig if (!bio_attempt_back_merge(q, rq, bio)) 23534fe7c05SChristoph Hellwig return false; 236e4d750c9SJens Axboe *merged_request = attempt_back_merge(q, rq); 237e4d750c9SJens Axboe if (!*merged_request) 23834fe7c05SChristoph Hellwig elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); 239bd166ef1SJens Axboe return true; 24034fe7c05SChristoph Hellwig case ELEVATOR_FRONT_MERGE: 241bd166ef1SJens Axboe if (!blk_mq_sched_allow_merge(q, rq, bio)) 242bd166ef1SJens Axboe return false; 24334fe7c05SChristoph Hellwig if (!bio_attempt_front_merge(q, rq, bio)) 24434fe7c05SChristoph Hellwig return false; 245e4d750c9SJens Axboe *merged_request = attempt_front_merge(q, rq); 246e4d750c9SJens Axboe if (!*merged_request) 24734fe7c05SChristoph Hellwig elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); 248bd166ef1SJens Axboe return true; 249bea99a50SKeith Busch case ELEVATOR_DISCARD_MERGE: 250bea99a50SKeith Busch return bio_attempt_discard_merge(q, rq, bio); 25134fe7c05SChristoph Hellwig default: 252bd166ef1SJens Axboe return false; 253bd166ef1SJens Axboe } 25434fe7c05SChristoph Hellwig } 255bd166ef1SJens Axboe EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); 256bd166ef1SJens Axboe 2579bddeb2aSMing Lei /* 2589c558734SJens Axboe * Iterate list of requests and see if we can merge this bio with any 2599c558734SJens Axboe * of them. 2609bddeb2aSMing Lei */ 2619c558734SJens Axboe bool blk_mq_bio_list_merge(struct request_queue *q, struct list_head *list, 2629c558734SJens Axboe struct bio *bio) 2639bddeb2aSMing Lei { 2649bddeb2aSMing Lei struct request *rq; 2659bddeb2aSMing Lei int checked = 8; 2669bddeb2aSMing Lei 2679c558734SJens Axboe list_for_each_entry_reverse(rq, list, queuelist) { 2689bddeb2aSMing Lei bool merged = false; 2699bddeb2aSMing Lei 2709bddeb2aSMing Lei if (!checked--) 2719bddeb2aSMing Lei break; 2729bddeb2aSMing Lei 2739bddeb2aSMing Lei if (!blk_rq_merge_ok(rq, bio)) 2749bddeb2aSMing Lei continue; 2759bddeb2aSMing Lei 2769bddeb2aSMing Lei switch (blk_try_merge(rq, bio)) { 2779bddeb2aSMing Lei case ELEVATOR_BACK_MERGE: 2789bddeb2aSMing Lei if (blk_mq_sched_allow_merge(q, rq, bio)) 2799bddeb2aSMing Lei merged = bio_attempt_back_merge(q, rq, bio); 2809bddeb2aSMing Lei break; 2819bddeb2aSMing Lei case ELEVATOR_FRONT_MERGE: 2829bddeb2aSMing Lei if (blk_mq_sched_allow_merge(q, rq, bio)) 2839bddeb2aSMing Lei merged = bio_attempt_front_merge(q, rq, bio); 2849bddeb2aSMing Lei break; 2859bddeb2aSMing Lei case ELEVATOR_DISCARD_MERGE: 2869bddeb2aSMing Lei merged = bio_attempt_discard_merge(q, rq, bio); 2879bddeb2aSMing Lei break; 2889bddeb2aSMing Lei default: 2899bddeb2aSMing Lei continue; 2909bddeb2aSMing Lei } 2919bddeb2aSMing Lei 2929bddeb2aSMing Lei return merged; 2939bddeb2aSMing Lei } 2949bddeb2aSMing Lei 2959bddeb2aSMing Lei return false; 2969bddeb2aSMing Lei } 2979c558734SJens Axboe EXPORT_SYMBOL_GPL(blk_mq_bio_list_merge); 2989c558734SJens Axboe 2999c558734SJens Axboe /* 3009c558734SJens Axboe * Reverse check our software queue for entries that we could potentially 3019c558734SJens Axboe * merge with. Currently includes a hand-wavy stop count of 8, to not spend 3029c558734SJens Axboe * too much time checking for merges. 3039c558734SJens Axboe */ 3049c558734SJens Axboe static bool blk_mq_attempt_merge(struct request_queue *q, 305c16d6b5aSMing Lei struct blk_mq_hw_ctx *hctx, 3069c558734SJens Axboe struct blk_mq_ctx *ctx, struct bio *bio) 3079c558734SJens Axboe { 308c16d6b5aSMing Lei enum hctx_type type = hctx->type; 309c16d6b5aSMing Lei 3109c558734SJens Axboe lockdep_assert_held(&ctx->lock); 3119c558734SJens Axboe 312c16d6b5aSMing Lei if (blk_mq_bio_list_merge(q, &ctx->rq_lists[type], bio)) { 3139c558734SJens Axboe ctx->rq_merged++; 3149c558734SJens Axboe return true; 3159c558734SJens Axboe } 3169c558734SJens Axboe 3179c558734SJens Axboe return false; 3189c558734SJens Axboe } 3199bddeb2aSMing Lei 320bd166ef1SJens Axboe bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio) 321bd166ef1SJens Axboe { 322bd166ef1SJens Axboe struct elevator_queue *e = q->elevator; 323bd166ef1SJens Axboe struct blk_mq_ctx *ctx = blk_mq_get_ctx(q); 3248ccdf4a3SJianchao Wang struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, bio->bi_opf, ctx); 3259bddeb2aSMing Lei bool ret = false; 326c16d6b5aSMing Lei enum hctx_type type; 327bd166ef1SJens Axboe 328f9cd4bfeSJens Axboe if (e && e->type->ops.bio_merge) { 329bd166ef1SJens Axboe blk_mq_put_ctx(ctx); 330f9cd4bfeSJens Axboe return e->type->ops.bio_merge(hctx, bio); 331bd166ef1SJens Axboe } 332bd166ef1SJens Axboe 333c16d6b5aSMing Lei type = hctx->type; 334b04f50abSMing Lei if ((hctx->flags & BLK_MQ_F_SHOULD_MERGE) && 335c16d6b5aSMing Lei !list_empty_careful(&ctx->rq_lists[type])) { 3369bddeb2aSMing Lei /* default per sw-queue merge */ 3379bddeb2aSMing Lei spin_lock(&ctx->lock); 338c16d6b5aSMing Lei ret = blk_mq_attempt_merge(q, hctx, ctx, bio); 3399bddeb2aSMing Lei spin_unlock(&ctx->lock); 3409bddeb2aSMing Lei } 3419bddeb2aSMing Lei 3429bddeb2aSMing Lei blk_mq_put_ctx(ctx); 3439bddeb2aSMing Lei return ret; 344bd166ef1SJens Axboe } 345bd166ef1SJens Axboe 346bd166ef1SJens Axboe bool blk_mq_sched_try_insert_merge(struct request_queue *q, struct request *rq) 347bd166ef1SJens Axboe { 348bd166ef1SJens Axboe return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq); 349bd166ef1SJens Axboe } 350bd166ef1SJens Axboe EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge); 351bd166ef1SJens Axboe 352bd166ef1SJens Axboe void blk_mq_sched_request_inserted(struct request *rq) 353bd166ef1SJens Axboe { 354bd166ef1SJens Axboe trace_block_rq_insert(rq->q, rq); 355bd166ef1SJens Axboe } 356bd166ef1SJens Axboe EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted); 357bd166ef1SJens Axboe 3580cacba6cSOmar Sandoval static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx, 359a6a252e6SMing Lei bool has_sched, 3600cacba6cSOmar Sandoval struct request *rq) 361bd166ef1SJens Axboe { 362a6a252e6SMing Lei /* dispatch flush rq directly */ 363a6a252e6SMing Lei if (rq->rq_flags & RQF_FLUSH_SEQ) { 364bd166ef1SJens Axboe spin_lock(&hctx->lock); 365bd166ef1SJens Axboe list_add(&rq->queuelist, &hctx->dispatch); 366bd166ef1SJens Axboe spin_unlock(&hctx->lock); 367bd166ef1SJens Axboe return true; 368bd166ef1SJens Axboe } 369bd166ef1SJens Axboe 370923218f6SMing Lei if (has_sched) 371a6a252e6SMing Lei rq->rq_flags |= RQF_SORTED; 372a6a252e6SMing Lei 373a6a252e6SMing Lei return false; 374a6a252e6SMing Lei } 375a6a252e6SMing Lei 376bd6737f1SJens Axboe void blk_mq_sched_insert_request(struct request *rq, bool at_head, 3779e97d295SMike Snitzer bool run_queue, bool async) 378bd6737f1SJens Axboe { 379bd6737f1SJens Axboe struct request_queue *q = rq->q; 380bd6737f1SJens Axboe struct elevator_queue *e = q->elevator; 381bd6737f1SJens Axboe struct blk_mq_ctx *ctx = rq->mq_ctx; 382ea4f995eSJens Axboe struct blk_mq_hw_ctx *hctx = rq->mq_hctx; 383bd6737f1SJens Axboe 384a6a252e6SMing Lei /* flush rq in flush machinery need to be dispatched directly */ 385a6a252e6SMing Lei if (!(rq->rq_flags & RQF_FLUSH_SEQ) && op_is_flush(rq->cmd_flags)) { 386923218f6SMing Lei blk_insert_flush(rq); 387923218f6SMing Lei goto run; 388bd6737f1SJens Axboe } 389bd6737f1SJens Axboe 390923218f6SMing Lei WARN_ON(e && (rq->tag != -1)); 391923218f6SMing Lei 392a6a252e6SMing Lei if (blk_mq_sched_bypass_insert(hctx, !!e, rq)) 3930cacba6cSOmar Sandoval goto run; 3940cacba6cSOmar Sandoval 395f9cd4bfeSJens Axboe if (e && e->type->ops.insert_requests) { 396bd6737f1SJens Axboe LIST_HEAD(list); 397bd6737f1SJens Axboe 398bd6737f1SJens Axboe list_add(&rq->queuelist, &list); 399f9cd4bfeSJens Axboe e->type->ops.insert_requests(hctx, &list, at_head); 400bd6737f1SJens Axboe } else { 401bd6737f1SJens Axboe spin_lock(&ctx->lock); 402bd6737f1SJens Axboe __blk_mq_insert_request(hctx, rq, at_head); 403bd6737f1SJens Axboe spin_unlock(&ctx->lock); 404bd6737f1SJens Axboe } 405bd6737f1SJens Axboe 4060cacba6cSOmar Sandoval run: 407bd6737f1SJens Axboe if (run_queue) 408bd6737f1SJens Axboe blk_mq_run_hw_queue(hctx, async); 409bd6737f1SJens Axboe } 410bd6737f1SJens Axboe 41167cae4c9SJens Axboe void blk_mq_sched_insert_requests(struct blk_mq_hw_ctx *hctx, 412bd6737f1SJens Axboe struct blk_mq_ctx *ctx, 413bd6737f1SJens Axboe struct list_head *list, bool run_queue_async) 414bd6737f1SJens Axboe { 415f9afca4dSJens Axboe struct elevator_queue *e; 416f9afca4dSJens Axboe 417f9afca4dSJens Axboe e = hctx->queue->elevator; 418f9cd4bfeSJens Axboe if (e && e->type->ops.insert_requests) 419f9cd4bfeSJens Axboe e->type->ops.insert_requests(hctx, list, false); 4206ce3dd6eSMing Lei else { 4216ce3dd6eSMing Lei /* 4226ce3dd6eSMing Lei * try to issue requests directly if the hw queue isn't 4236ce3dd6eSMing Lei * busy in case of 'none' scheduler, and this way may save 4246ce3dd6eSMing Lei * us one extra enqueue & dequeue to sw queue. 4256ce3dd6eSMing Lei */ 4265b7a6f12SJianchao Wang if (!hctx->dispatch_busy && !e && !run_queue_async) 4276ce3dd6eSMing Lei blk_mq_try_issue_list_directly(hctx, list); 4285b7a6f12SJianchao Wang else 429bd6737f1SJens Axboe blk_mq_insert_requests(hctx, ctx, list); 4306ce3dd6eSMing Lei } 431bd6737f1SJens Axboe 432bd6737f1SJens Axboe blk_mq_run_hw_queue(hctx, run_queue_async); 433bd6737f1SJens Axboe } 434bd6737f1SJens Axboe 435bd166ef1SJens Axboe static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set, 436bd166ef1SJens Axboe struct blk_mq_hw_ctx *hctx, 437bd166ef1SJens Axboe unsigned int hctx_idx) 438bd166ef1SJens Axboe { 439bd166ef1SJens Axboe if (hctx->sched_tags) { 440bd166ef1SJens Axboe blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx); 441bd166ef1SJens Axboe blk_mq_free_rq_map(hctx->sched_tags); 442bd166ef1SJens Axboe hctx->sched_tags = NULL; 443bd166ef1SJens Axboe } 444bd166ef1SJens Axboe } 445bd166ef1SJens Axboe 4466917ff0bSOmar Sandoval static int blk_mq_sched_alloc_tags(struct request_queue *q, 4476917ff0bSOmar Sandoval struct blk_mq_hw_ctx *hctx, 4486917ff0bSOmar Sandoval unsigned int hctx_idx) 449bd166ef1SJens Axboe { 450bd166ef1SJens Axboe struct blk_mq_tag_set *set = q->tag_set; 4516917ff0bSOmar Sandoval int ret; 452bd166ef1SJens Axboe 4536917ff0bSOmar Sandoval hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests, 4546917ff0bSOmar Sandoval set->reserved_tags); 455bd166ef1SJens Axboe if (!hctx->sched_tags) 4566917ff0bSOmar Sandoval return -ENOMEM; 4576917ff0bSOmar Sandoval 4586917ff0bSOmar Sandoval ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests); 4596917ff0bSOmar Sandoval if (ret) 4606917ff0bSOmar Sandoval blk_mq_sched_free_tags(set, hctx, hctx_idx); 461bd166ef1SJens Axboe 462bd166ef1SJens Axboe return ret; 463bd166ef1SJens Axboe } 464bd166ef1SJens Axboe 46554d5329dSOmar Sandoval static void blk_mq_sched_tags_teardown(struct request_queue *q) 466bd166ef1SJens Axboe { 467bd166ef1SJens Axboe struct blk_mq_tag_set *set = q->tag_set; 468bd166ef1SJens Axboe struct blk_mq_hw_ctx *hctx; 469bd166ef1SJens Axboe int i; 470bd166ef1SJens Axboe 471bd166ef1SJens Axboe queue_for_each_hw_ctx(q, hctx, i) 472bd166ef1SJens Axboe blk_mq_sched_free_tags(set, hctx, i); 473bd166ef1SJens Axboe } 474d3484991SJens Axboe 4756917ff0bSOmar Sandoval int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) 4766917ff0bSOmar Sandoval { 4776917ff0bSOmar Sandoval struct blk_mq_hw_ctx *hctx; 478ee056f98SOmar Sandoval struct elevator_queue *eq; 4796917ff0bSOmar Sandoval unsigned int i; 4806917ff0bSOmar Sandoval int ret; 4816917ff0bSOmar Sandoval 4826917ff0bSOmar Sandoval if (!e) { 4836917ff0bSOmar Sandoval q->elevator = NULL; 48432a50fabSMing Lei q->nr_requests = q->tag_set->queue_depth; 4856917ff0bSOmar Sandoval return 0; 4866917ff0bSOmar Sandoval } 4876917ff0bSOmar Sandoval 4886917ff0bSOmar Sandoval /* 48932825c45SMing Lei * Default to double of smaller one between hw queue_depth and 128, 49032825c45SMing Lei * since we don't split into sync/async like the old code did. 49132825c45SMing Lei * Additionally, this is a per-hw queue depth. 4926917ff0bSOmar Sandoval */ 49332825c45SMing Lei q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth, 49432825c45SMing Lei BLKDEV_MAX_RQ); 4956917ff0bSOmar Sandoval 4966917ff0bSOmar Sandoval queue_for_each_hw_ctx(q, hctx, i) { 4976917ff0bSOmar Sandoval ret = blk_mq_sched_alloc_tags(q, hctx, i); 4986917ff0bSOmar Sandoval if (ret) 4996917ff0bSOmar Sandoval goto err; 5006917ff0bSOmar Sandoval } 5016917ff0bSOmar Sandoval 502f9cd4bfeSJens Axboe ret = e->ops.init_sched(q, e); 5036917ff0bSOmar Sandoval if (ret) 5046917ff0bSOmar Sandoval goto err; 5056917ff0bSOmar Sandoval 506d332ce09SOmar Sandoval blk_mq_debugfs_register_sched(q); 507d332ce09SOmar Sandoval 508ee056f98SOmar Sandoval queue_for_each_hw_ctx(q, hctx, i) { 509f9cd4bfeSJens Axboe if (e->ops.init_hctx) { 510f9cd4bfeSJens Axboe ret = e->ops.init_hctx(hctx, i); 511ee056f98SOmar Sandoval if (ret) { 512ee056f98SOmar Sandoval eq = q->elevator; 513ee056f98SOmar Sandoval blk_mq_exit_sched(q, eq); 514ee056f98SOmar Sandoval kobject_put(&eq->kobj); 515ee056f98SOmar Sandoval return ret; 516ee056f98SOmar Sandoval } 517ee056f98SOmar Sandoval } 518d332ce09SOmar Sandoval blk_mq_debugfs_register_sched_hctx(q, hctx); 519ee056f98SOmar Sandoval } 520ee056f98SOmar Sandoval 5216917ff0bSOmar Sandoval return 0; 5226917ff0bSOmar Sandoval 5236917ff0bSOmar Sandoval err: 52454d5329dSOmar Sandoval blk_mq_sched_tags_teardown(q); 52554d5329dSOmar Sandoval q->elevator = NULL; 5266917ff0bSOmar Sandoval return ret; 5276917ff0bSOmar Sandoval } 5286917ff0bSOmar Sandoval 52954d5329dSOmar Sandoval void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) 53054d5329dSOmar Sandoval { 531ee056f98SOmar Sandoval struct blk_mq_hw_ctx *hctx; 532ee056f98SOmar Sandoval unsigned int i; 533ee056f98SOmar Sandoval 534ee056f98SOmar Sandoval queue_for_each_hw_ctx(q, hctx, i) { 535d332ce09SOmar Sandoval blk_mq_debugfs_unregister_sched_hctx(hctx); 536f9cd4bfeSJens Axboe if (e->type->ops.exit_hctx && hctx->sched_data) { 537f9cd4bfeSJens Axboe e->type->ops.exit_hctx(hctx, i); 538ee056f98SOmar Sandoval hctx->sched_data = NULL; 539ee056f98SOmar Sandoval } 540ee056f98SOmar Sandoval } 541d332ce09SOmar Sandoval blk_mq_debugfs_unregister_sched(q); 542f9cd4bfeSJens Axboe if (e->type->ops.exit_sched) 543f9cd4bfeSJens Axboe e->type->ops.exit_sched(e); 54454d5329dSOmar Sandoval blk_mq_sched_tags_teardown(q); 54554d5329dSOmar Sandoval q->elevator = NULL; 54654d5329dSOmar Sandoval } 547