Christoph Hellwig | 3dcf60bc | 2019-04-30 14:42:43 -0400 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Jens Axboe | 75bb462 | 2014-05-28 10:15:41 -0600 | [diff] [blame] | 2 | /* |
| 3 | * Block multiqueue core code |
| 4 | * |
| 5 | * Copyright (C) 2013-2014 Jens Axboe |
| 6 | * Copyright (C) 2013-2014 Christoph Hellwig |
| 7 | */ |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 8 | #include <linux/kernel.h> |
| 9 | #include <linux/module.h> |
| 10 | #include <linux/backing-dev.h> |
| 11 | #include <linux/bio.h> |
| 12 | #include <linux/blkdev.h> |
Christoph Hellwig | fe45e63 | 2021-09-20 14:33:27 +0200 | [diff] [blame] | 13 | #include <linux/blk-integrity.h> |
Catalin Marinas | f75782e | 2015-09-14 18:16:02 +0100 | [diff] [blame] | 14 | #include <linux/kmemleak.h> |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 15 | #include <linux/mm.h> |
| 16 | #include <linux/init.h> |
| 17 | #include <linux/slab.h> |
| 18 | #include <linux/workqueue.h> |
| 19 | #include <linux/smp.h> |
Christoph Hellwig | e41d12f | 2021-09-20 14:33:13 +0200 | [diff] [blame] | 20 | #include <linux/interrupt.h> |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 21 | #include <linux/llist.h> |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 22 | #include <linux/cpu.h> |
| 23 | #include <linux/cache.h> |
| 24 | #include <linux/sched/sysctl.h> |
Ingo Molnar | 105ab3d8 | 2017-02-01 16:36:40 +0100 | [diff] [blame] | 25 | #include <linux/sched/topology.h> |
Ingo Molnar | 174cd4b | 2017-02-02 19:15:33 +0100 | [diff] [blame] | 26 | #include <linux/sched/signal.h> |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 27 | #include <linux/delay.h> |
Jens Axboe | aedcd72 | 2014-09-17 08:27:03 -0600 | [diff] [blame] | 28 | #include <linux/crash_dump.h> |
Jens Axboe | 88c7b2b | 2016-08-25 08:07:30 -0600 | [diff] [blame] | 29 | #include <linux/prefetch.h> |
Satya Tangirala | a892c8d | 2020-05-14 00:37:18 +0000 | [diff] [blame] | 30 | #include <linux/blk-crypto.h> |
Christoph Hellwig | 82d981d | 2021-11-23 19:53:12 +0100 | [diff] [blame] | 31 | #include <linux/part_stat.h> |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 32 | |
| 33 | #include <trace/events/block.h> |
| 34 | |
| 35 | #include <linux/blk-mq.h> |
Max Gurtovoy | 54d4e6a | 2019-09-16 18:44:29 +0300 | [diff] [blame] | 36 | #include <linux/t10-pi.h> |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 37 | #include "blk.h" |
| 38 | #include "blk-mq.h" |
Omar Sandoval | 9c1051a | 2017-05-04 08:17:21 -0600 | [diff] [blame] | 39 | #include "blk-mq-debugfs.h" |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 40 | #include "blk-mq-tag.h" |
Bart Van Assche | 986d413 | 2018-09-26 14:01:10 -0700 | [diff] [blame] | 41 | #include "blk-pm.h" |
Jens Axboe | cf43e6b | 2016-11-07 21:32:37 -0700 | [diff] [blame] | 42 | #include "blk-stat.h" |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 43 | #include "blk-mq-sched.h" |
Josef Bacik | c1c8038 | 2018-07-03 11:14:59 -0400 | [diff] [blame] | 44 | #include "blk-rq-qos.h" |
Jan Kara | 82b74ca | 2022-06-23 09:48:32 +0200 | [diff] [blame] | 45 | #include "blk-ioprio.h" |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 46 | |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 47 | static DEFINE_PER_CPU(struct llist_head, blk_cpu_done); |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 48 | |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 49 | static void blk_mq_poll_stats_start(struct request_queue *q); |
| 50 | static void blk_mq_poll_stats_fn(struct blk_stat_callback *cb); |
| 51 | |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 52 | static int blk_mq_poll_stats_bkt(const struct request *rq) |
| 53 | { |
Hou Tao | 3d24430 | 2019-05-21 15:59:03 +0800 | [diff] [blame] | 54 | int ddir, sectors, bucket; |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 55 | |
Jens Axboe | 99c749a | 2017-04-21 07:55:42 -0600 | [diff] [blame] | 56 | ddir = rq_data_dir(rq); |
Hou Tao | 3d24430 | 2019-05-21 15:59:03 +0800 | [diff] [blame] | 57 | sectors = blk_rq_stats_sectors(rq); |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 58 | |
Hou Tao | 3d24430 | 2019-05-21 15:59:03 +0800 | [diff] [blame] | 59 | bucket = ddir + 2 * ilog2(sectors); |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 60 | |
| 61 | if (bucket < 0) |
| 62 | return -1; |
| 63 | else if (bucket >= BLK_MQ_POLL_STATS_BKTS) |
| 64 | return ddir + BLK_MQ_POLL_STATS_BKTS - 2; |
| 65 | |
| 66 | return bucket; |
| 67 | } |
| 68 | |
Christoph Hellwig | 3e08773 | 2021-10-12 13:12:24 +0200 | [diff] [blame] | 69 | #define BLK_QC_T_SHIFT 16 |
| 70 | #define BLK_QC_T_INTERNAL (1U << 31) |
| 71 | |
Christoph Hellwig | f70299f | 2021-10-12 13:12:15 +0200 | [diff] [blame] | 72 | static inline struct blk_mq_hw_ctx *blk_qc_to_hctx(struct request_queue *q, |
| 73 | blk_qc_t qc) |
| 74 | { |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 75 | return xa_load(&q->hctx_table, |
| 76 | (qc & ~BLK_QC_T_INTERNAL) >> BLK_QC_T_SHIFT); |
Christoph Hellwig | f70299f | 2021-10-12 13:12:15 +0200 | [diff] [blame] | 77 | } |
| 78 | |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 79 | static inline struct request *blk_qc_to_rq(struct blk_mq_hw_ctx *hctx, |
| 80 | blk_qc_t qc) |
| 81 | { |
Christoph Hellwig | efbabbe | 2021-10-12 13:12:17 +0200 | [diff] [blame] | 82 | unsigned int tag = qc & ((1U << BLK_QC_T_SHIFT) - 1); |
| 83 | |
| 84 | if (qc & BLK_QC_T_INTERNAL) |
| 85 | return blk_mq_tag_to_rq(hctx->sched_tags, tag); |
| 86 | return blk_mq_tag_to_rq(hctx->tags, tag); |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 87 | } |
| 88 | |
Christoph Hellwig | 3e08773 | 2021-10-12 13:12:24 +0200 | [diff] [blame] | 89 | static inline blk_qc_t blk_rq_to_qc(struct request *rq) |
| 90 | { |
| 91 | return (rq->mq_hctx->queue_num << BLK_QC_T_SHIFT) | |
| 92 | (rq->tag != -1 ? |
| 93 | rq->tag : (rq->internal_tag | BLK_QC_T_INTERNAL)); |
| 94 | } |
| 95 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 96 | /* |
Yufen Yu | 85fae29 | 2019-03-24 17:57:08 +0800 | [diff] [blame] | 97 | * Check if any of the ctx, dispatch list or elevator |
| 98 | * have pending work in this hardware queue. |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 99 | */ |
Jens Axboe | 79f720a | 2017-11-10 09:13:21 -0700 | [diff] [blame] | 100 | static bool blk_mq_hctx_has_pending(struct blk_mq_hw_ctx *hctx) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 101 | { |
Jens Axboe | 79f720a | 2017-11-10 09:13:21 -0700 | [diff] [blame] | 102 | return !list_empty_careful(&hctx->dispatch) || |
| 103 | sbitmap_any_bit_set(&hctx->ctx_map) || |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 104 | blk_mq_sched_has_work(hctx); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 105 | } |
| 106 | |
| 107 | /* |
| 108 | * Mark this ctx as having pending work in this hardware queue |
| 109 | */ |
| 110 | static void blk_mq_hctx_mark_pending(struct blk_mq_hw_ctx *hctx, |
| 111 | struct blk_mq_ctx *ctx) |
| 112 | { |
Jens Axboe | f31967f | 2018-10-29 13:13:29 -0600 | [diff] [blame] | 113 | const int bit = ctx->index_hw[hctx->type]; |
| 114 | |
| 115 | if (!sbitmap_test_bit(&hctx->ctx_map, bit)) |
| 116 | sbitmap_set_bit(&hctx->ctx_map, bit); |
Jens Axboe | 1429d7c | 2014-05-19 09:23:55 -0600 | [diff] [blame] | 117 | } |
| 118 | |
| 119 | static void blk_mq_hctx_clear_pending(struct blk_mq_hw_ctx *hctx, |
| 120 | struct blk_mq_ctx *ctx) |
| 121 | { |
Jens Axboe | f31967f | 2018-10-29 13:13:29 -0600 | [diff] [blame] | 122 | const int bit = ctx->index_hw[hctx->type]; |
| 123 | |
| 124 | sbitmap_clear_bit(&hctx->ctx_map, bit); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 125 | } |
| 126 | |
Jens Axboe | f299b7c | 2017-08-08 17:51:45 -0600 | [diff] [blame] | 127 | struct mq_inflight { |
Christoph Hellwig | 8446fe9 | 2020-11-24 09:36:54 +0100 | [diff] [blame] | 128 | struct block_device *part; |
Pavel Begunkov | a2e80f6 | 2019-09-30 21:55:34 +0300 | [diff] [blame] | 129 | unsigned int inflight[2]; |
Jens Axboe | f299b7c | 2017-08-08 17:51:45 -0600 | [diff] [blame] | 130 | }; |
| 131 | |
John Garry | 2dd6532 | 2022-07-06 20:03:53 +0800 | [diff] [blame] | 132 | static bool blk_mq_check_inflight(struct request *rq, void *priv) |
Jens Axboe | f299b7c | 2017-08-08 17:51:45 -0600 | [diff] [blame] | 133 | { |
| 134 | struct mq_inflight *mi = priv; |
| 135 | |
Haisu Wang | b81c14c | 2022-05-30 14:40:59 +0800 | [diff] [blame] | 136 | if (rq->part && blk_do_io_stat(rq) && |
| 137 | (!mi->part->bd_partno || rq->part == mi->part) && |
Jeffle Xu | b0d9755 | 2020-12-02 19:11:45 +0800 | [diff] [blame] | 138 | blk_mq_rq_state(rq) == MQ_RQ_IN_FLIGHT) |
Pavel Begunkov | bb4e6b1 | 2019-09-30 21:55:33 +0300 | [diff] [blame] | 139 | mi->inflight[rq_data_dir(rq)]++; |
Jens Axboe | 7baa857 | 2018-11-08 10:24:07 -0700 | [diff] [blame] | 140 | |
| 141 | return true; |
Jens Axboe | f299b7c | 2017-08-08 17:51:45 -0600 | [diff] [blame] | 142 | } |
| 143 | |
Christoph Hellwig | 8446fe9 | 2020-11-24 09:36:54 +0100 | [diff] [blame] | 144 | unsigned int blk_mq_in_flight(struct request_queue *q, |
| 145 | struct block_device *part) |
Jens Axboe | f299b7c | 2017-08-08 17:51:45 -0600 | [diff] [blame] | 146 | { |
Pavel Begunkov | a2e80f6 | 2019-09-30 21:55:34 +0300 | [diff] [blame] | 147 | struct mq_inflight mi = { .part = part }; |
Jens Axboe | f299b7c | 2017-08-08 17:51:45 -0600 | [diff] [blame] | 148 | |
Jens Axboe | f299b7c | 2017-08-08 17:51:45 -0600 | [diff] [blame] | 149 | blk_mq_queue_tag_busy_iter(q, blk_mq_check_inflight, &mi); |
Mikulas Patocka | e016b78 | 2018-12-06 11:41:21 -0500 | [diff] [blame] | 150 | |
Pavel Begunkov | a2e80f6 | 2019-09-30 21:55:34 +0300 | [diff] [blame] | 151 | return mi.inflight[0] + mi.inflight[1]; |
Omar Sandoval | bf0ddab | 2018-04-26 00:21:59 -0700 | [diff] [blame] | 152 | } |
| 153 | |
Christoph Hellwig | 8446fe9 | 2020-11-24 09:36:54 +0100 | [diff] [blame] | 154 | void blk_mq_in_flight_rw(struct request_queue *q, struct block_device *part, |
| 155 | unsigned int inflight[2]) |
Omar Sandoval | bf0ddab | 2018-04-26 00:21:59 -0700 | [diff] [blame] | 156 | { |
Pavel Begunkov | a2e80f6 | 2019-09-30 21:55:34 +0300 | [diff] [blame] | 157 | struct mq_inflight mi = { .part = part }; |
Omar Sandoval | bf0ddab | 2018-04-26 00:21:59 -0700 | [diff] [blame] | 158 | |
Pavel Begunkov | bb4e6b1 | 2019-09-30 21:55:33 +0300 | [diff] [blame] | 159 | blk_mq_queue_tag_busy_iter(q, blk_mq_check_inflight, &mi); |
Pavel Begunkov | a2e80f6 | 2019-09-30 21:55:34 +0300 | [diff] [blame] | 160 | inflight[0] = mi.inflight[0]; |
| 161 | inflight[1] = mi.inflight[1]; |
Omar Sandoval | bf0ddab | 2018-04-26 00:21:59 -0700 | [diff] [blame] | 162 | } |
| 163 | |
Ming Lei | 1671d52 | 2017-03-27 20:06:57 +0800 | [diff] [blame] | 164 | void blk_freeze_queue_start(struct request_queue *q) |
Ming Lei | 43a5e4e | 2013-12-26 21:31:35 +0800 | [diff] [blame] | 165 | { |
Bob Liu | 7996a8b | 2019-05-21 11:25:55 +0800 | [diff] [blame] | 166 | mutex_lock(&q->mq_freeze_lock); |
| 167 | if (++q->mq_freeze_depth == 1) { |
Dan Williams | 3ef28e8 | 2015-10-21 13:20:12 -0400 | [diff] [blame] | 168 | percpu_ref_kill(&q->q_usage_counter); |
Bob Liu | 7996a8b | 2019-05-21 11:25:55 +0800 | [diff] [blame] | 169 | mutex_unlock(&q->mq_freeze_lock); |
Jens Axboe | 344e9ff | 2018-11-15 12:22:51 -0700 | [diff] [blame] | 170 | if (queue_is_mq(q)) |
Ming Lei | 055f6e1 | 2017-11-09 10:49:53 -0800 | [diff] [blame] | 171 | blk_mq_run_hw_queues(q, false); |
Bob Liu | 7996a8b | 2019-05-21 11:25:55 +0800 | [diff] [blame] | 172 | } else { |
| 173 | mutex_unlock(&q->mq_freeze_lock); |
Tejun Heo | cddd5d1 | 2014-08-16 08:02:24 -0400 | [diff] [blame] | 174 | } |
Tejun Heo | f3af020 | 2014-11-04 13:52:27 -0500 | [diff] [blame] | 175 | } |
Ming Lei | 1671d52 | 2017-03-27 20:06:57 +0800 | [diff] [blame] | 176 | EXPORT_SYMBOL_GPL(blk_freeze_queue_start); |
Tejun Heo | f3af020 | 2014-11-04 13:52:27 -0500 | [diff] [blame] | 177 | |
Keith Busch | 6bae363e | 2017-03-01 14:22:10 -0500 | [diff] [blame] | 178 | void blk_mq_freeze_queue_wait(struct request_queue *q) |
Tejun Heo | f3af020 | 2014-11-04 13:52:27 -0500 | [diff] [blame] | 179 | { |
Dan Williams | 3ef28e8 | 2015-10-21 13:20:12 -0400 | [diff] [blame] | 180 | wait_event(q->mq_freeze_wq, percpu_ref_is_zero(&q->q_usage_counter)); |
Ming Lei | 43a5e4e | 2013-12-26 21:31:35 +0800 | [diff] [blame] | 181 | } |
Keith Busch | 6bae363e | 2017-03-01 14:22:10 -0500 | [diff] [blame] | 182 | EXPORT_SYMBOL_GPL(blk_mq_freeze_queue_wait); |
Ming Lei | 43a5e4e | 2013-12-26 21:31:35 +0800 | [diff] [blame] | 183 | |
Keith Busch | f91328c | 2017-03-01 14:22:11 -0500 | [diff] [blame] | 184 | int blk_mq_freeze_queue_wait_timeout(struct request_queue *q, |
| 185 | unsigned long timeout) |
| 186 | { |
| 187 | return wait_event_timeout(q->mq_freeze_wq, |
| 188 | percpu_ref_is_zero(&q->q_usage_counter), |
| 189 | timeout); |
| 190 | } |
| 191 | EXPORT_SYMBOL_GPL(blk_mq_freeze_queue_wait_timeout); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 192 | |
Tejun Heo | f3af020 | 2014-11-04 13:52:27 -0500 | [diff] [blame] | 193 | /* |
| 194 | * Guarantee no request is in use, so we can change any data structure of |
| 195 | * the queue afterward. |
| 196 | */ |
Dan Williams | 3ef28e8 | 2015-10-21 13:20:12 -0400 | [diff] [blame] | 197 | void blk_freeze_queue(struct request_queue *q) |
Tejun Heo | f3af020 | 2014-11-04 13:52:27 -0500 | [diff] [blame] | 198 | { |
Dan Williams | 3ef28e8 | 2015-10-21 13:20:12 -0400 | [diff] [blame] | 199 | /* |
| 200 | * In the !blk_mq case we are only calling this to kill the |
| 201 | * q_usage_counter, otherwise this increases the freeze depth |
| 202 | * and waits for it to return to zero. For this reason there is |
| 203 | * no blk_unfreeze_queue(), and blk_freeze_queue() is not |
| 204 | * exported to drivers as the only user for unfreeze is blk_mq. |
| 205 | */ |
Ming Lei | 1671d52 | 2017-03-27 20:06:57 +0800 | [diff] [blame] | 206 | blk_freeze_queue_start(q); |
Tejun Heo | f3af020 | 2014-11-04 13:52:27 -0500 | [diff] [blame] | 207 | blk_mq_freeze_queue_wait(q); |
| 208 | } |
Dan Williams | 3ef28e8 | 2015-10-21 13:20:12 -0400 | [diff] [blame] | 209 | |
| 210 | void blk_mq_freeze_queue(struct request_queue *q) |
| 211 | { |
| 212 | /* |
| 213 | * ...just an alias to keep freeze and unfreeze actions balanced |
| 214 | * in the blk_mq_* namespace |
| 215 | */ |
| 216 | blk_freeze_queue(q); |
| 217 | } |
Jens Axboe | c761d96 | 2015-01-02 15:05:12 -0700 | [diff] [blame] | 218 | EXPORT_SYMBOL_GPL(blk_mq_freeze_queue); |
Tejun Heo | f3af020 | 2014-11-04 13:52:27 -0500 | [diff] [blame] | 219 | |
Christoph Hellwig | aec89dc | 2021-09-29 09:12:41 +0200 | [diff] [blame] | 220 | void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 221 | { |
Bob Liu | 7996a8b | 2019-05-21 11:25:55 +0800 | [diff] [blame] | 222 | mutex_lock(&q->mq_freeze_lock); |
Christoph Hellwig | aec89dc | 2021-09-29 09:12:41 +0200 | [diff] [blame] | 223 | if (force_atomic) |
| 224 | q->q_usage_counter.data->force_atomic = true; |
Bob Liu | 7996a8b | 2019-05-21 11:25:55 +0800 | [diff] [blame] | 225 | q->mq_freeze_depth--; |
| 226 | WARN_ON_ONCE(q->mq_freeze_depth < 0); |
| 227 | if (!q->mq_freeze_depth) { |
Bart Van Assche | bdd6316 | 2018-09-26 14:01:08 -0700 | [diff] [blame] | 228 | percpu_ref_resurrect(&q->q_usage_counter); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 229 | wake_up_all(&q->mq_freeze_wq); |
Tejun Heo | add703f | 2014-07-01 10:34:38 -0600 | [diff] [blame] | 230 | } |
Bob Liu | 7996a8b | 2019-05-21 11:25:55 +0800 | [diff] [blame] | 231 | mutex_unlock(&q->mq_freeze_lock); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 232 | } |
Christoph Hellwig | aec89dc | 2021-09-29 09:12:41 +0200 | [diff] [blame] | 233 | |
| 234 | void blk_mq_unfreeze_queue(struct request_queue *q) |
| 235 | { |
| 236 | __blk_mq_unfreeze_queue(q, false); |
| 237 | } |
Keith Busch | b4c6a02 | 2014-12-19 17:54:14 -0700 | [diff] [blame] | 238 | EXPORT_SYMBOL_GPL(blk_mq_unfreeze_queue); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 239 | |
Bart Van Assche | 852ec80 | 2017-06-21 10:55:47 -0700 | [diff] [blame] | 240 | /* |
| 241 | * FIXME: replace the scsi_internal_device_*block_nowait() calls in the |
| 242 | * mpt3sas driver such that this function can be removed. |
| 243 | */ |
| 244 | void blk_mq_quiesce_queue_nowait(struct request_queue *q) |
| 245 | { |
Ming Lei | e70feb8 | 2021-10-14 16:17:10 +0800 | [diff] [blame] | 246 | unsigned long flags; |
| 247 | |
| 248 | spin_lock_irqsave(&q->queue_lock, flags); |
| 249 | if (!q->quiesce_depth++) |
| 250 | blk_queue_flag_set(QUEUE_FLAG_QUIESCED, q); |
| 251 | spin_unlock_irqrestore(&q->queue_lock, flags); |
Bart Van Assche | 852ec80 | 2017-06-21 10:55:47 -0700 | [diff] [blame] | 252 | } |
| 253 | EXPORT_SYMBOL_GPL(blk_mq_quiesce_queue_nowait); |
| 254 | |
Bart Van Assche | 6a83e74 | 2016-11-02 10:09:51 -0600 | [diff] [blame] | 255 | /** |
Ming Lei | 9ef4d02 | 2021-11-09 15:11:41 +0800 | [diff] [blame] | 256 | * blk_mq_wait_quiesce_done() - wait until in-progress quiesce is done |
Christoph Hellwig | 483239c | 2022-11-01 16:00:48 +0100 | [diff] [blame] | 257 | * @set: tag_set to wait on |
Ming Lei | 9ef4d02 | 2021-11-09 15:11:41 +0800 | [diff] [blame] | 258 | * |
| 259 | * Note: it is driver's responsibility for making sure that quiesce has |
Christoph Hellwig | 483239c | 2022-11-01 16:00:48 +0100 | [diff] [blame] | 260 | * been started on or more of the request_queues of the tag_set. This |
| 261 | * function only waits for the quiesce on those request_queues that had |
| 262 | * the quiesce flag set using blk_mq_quiesce_queue_nowait. |
Ming Lei | 9ef4d02 | 2021-11-09 15:11:41 +0800 | [diff] [blame] | 263 | */ |
Christoph Hellwig | 483239c | 2022-11-01 16:00:48 +0100 | [diff] [blame] | 264 | void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set) |
Ming Lei | 9ef4d02 | 2021-11-09 15:11:41 +0800 | [diff] [blame] | 265 | { |
Christoph Hellwig | 483239c | 2022-11-01 16:00:48 +0100 | [diff] [blame] | 266 | if (set->flags & BLK_MQ_F_BLOCKING) |
| 267 | synchronize_srcu(set->srcu); |
Ming Lei | 704b914 | 2021-12-03 21:15:32 +0800 | [diff] [blame] | 268 | else |
Ming Lei | 9ef4d02 | 2021-11-09 15:11:41 +0800 | [diff] [blame] | 269 | synchronize_rcu(); |
| 270 | } |
| 271 | EXPORT_SYMBOL_GPL(blk_mq_wait_quiesce_done); |
| 272 | |
| 273 | /** |
Ming Lei | 69e07c4 | 2017-06-06 23:22:07 +0800 | [diff] [blame] | 274 | * blk_mq_quiesce_queue() - wait until all ongoing dispatches have finished |
Bart Van Assche | 6a83e74 | 2016-11-02 10:09:51 -0600 | [diff] [blame] | 275 | * @q: request queue. |
| 276 | * |
| 277 | * Note: this function does not prevent that the struct request end_io() |
Ming Lei | 69e07c4 | 2017-06-06 23:22:07 +0800 | [diff] [blame] | 278 | * callback function is invoked. Once this function is returned, we make |
| 279 | * sure no dispatch can happen until the queue is unquiesced via |
| 280 | * blk_mq_unquiesce_queue(). |
Bart Van Assche | 6a83e74 | 2016-11-02 10:09:51 -0600 | [diff] [blame] | 281 | */ |
| 282 | void blk_mq_quiesce_queue(struct request_queue *q) |
| 283 | { |
Ming Lei | 1d9e9bc | 2017-06-06 23:22:08 +0800 | [diff] [blame] | 284 | blk_mq_quiesce_queue_nowait(q); |
Christoph Hellwig | 8537380 | 2022-11-01 16:00:46 +0100 | [diff] [blame] | 285 | /* nothing to wait for non-mq queues */ |
| 286 | if (queue_is_mq(q)) |
Christoph Hellwig | 483239c | 2022-11-01 16:00:48 +0100 | [diff] [blame] | 287 | blk_mq_wait_quiesce_done(q->tag_set); |
Bart Van Assche | 6a83e74 | 2016-11-02 10:09:51 -0600 | [diff] [blame] | 288 | } |
| 289 | EXPORT_SYMBOL_GPL(blk_mq_quiesce_queue); |
| 290 | |
Ming Lei | e4e7391 | 2017-06-06 23:22:03 +0800 | [diff] [blame] | 291 | /* |
| 292 | * blk_mq_unquiesce_queue() - counterpart of blk_mq_quiesce_queue() |
| 293 | * @q: request queue. |
| 294 | * |
| 295 | * This function recovers queue into the state before quiescing |
| 296 | * which is done by blk_mq_quiesce_queue. |
| 297 | */ |
| 298 | void blk_mq_unquiesce_queue(struct request_queue *q) |
| 299 | { |
Ming Lei | e70feb8 | 2021-10-14 16:17:10 +0800 | [diff] [blame] | 300 | unsigned long flags; |
| 301 | bool run_queue = false; |
| 302 | |
| 303 | spin_lock_irqsave(&q->queue_lock, flags); |
| 304 | if (WARN_ON_ONCE(q->quiesce_depth <= 0)) { |
| 305 | ; |
| 306 | } else if (!--q->quiesce_depth) { |
| 307 | blk_queue_flag_clear(QUEUE_FLAG_QUIESCED, q); |
| 308 | run_queue = true; |
| 309 | } |
| 310 | spin_unlock_irqrestore(&q->queue_lock, flags); |
Ming Lei | f4560ff | 2017-06-18 14:24:27 -0600 | [diff] [blame] | 311 | |
Ming Lei | 1d9e9bc | 2017-06-06 23:22:08 +0800 | [diff] [blame] | 312 | /* dispatch requests which are inserted during quiescing */ |
Ming Lei | e70feb8 | 2021-10-14 16:17:10 +0800 | [diff] [blame] | 313 | if (run_queue) |
| 314 | blk_mq_run_hw_queues(q, true); |
Ming Lei | e4e7391 | 2017-06-06 23:22:03 +0800 | [diff] [blame] | 315 | } |
| 316 | EXPORT_SYMBOL_GPL(blk_mq_unquiesce_queue); |
| 317 | |
Chao Leng | 414dd48 | 2022-11-01 16:00:49 +0100 | [diff] [blame] | 318 | void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set) |
| 319 | { |
| 320 | struct request_queue *q; |
| 321 | |
| 322 | mutex_lock(&set->tag_list_lock); |
| 323 | list_for_each_entry(q, &set->tag_list, tag_set_list) { |
| 324 | if (!blk_queue_skip_tagset_quiesce(q)) |
| 325 | blk_mq_quiesce_queue_nowait(q); |
| 326 | } |
| 327 | blk_mq_wait_quiesce_done(set); |
| 328 | mutex_unlock(&set->tag_list_lock); |
| 329 | } |
| 330 | EXPORT_SYMBOL_GPL(blk_mq_quiesce_tagset); |
| 331 | |
| 332 | void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set) |
| 333 | { |
| 334 | struct request_queue *q; |
| 335 | |
| 336 | mutex_lock(&set->tag_list_lock); |
| 337 | list_for_each_entry(q, &set->tag_list, tag_set_list) { |
| 338 | if (!blk_queue_skip_tagset_quiesce(q)) |
| 339 | blk_mq_unquiesce_queue(q); |
| 340 | } |
| 341 | mutex_unlock(&set->tag_list_lock); |
| 342 | } |
| 343 | EXPORT_SYMBOL_GPL(blk_mq_unquiesce_tagset); |
| 344 | |
Jens Axboe | aed3ea9 | 2014-12-22 14:04:42 -0700 | [diff] [blame] | 345 | void blk_mq_wake_waiters(struct request_queue *q) |
| 346 | { |
| 347 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 348 | unsigned long i; |
Jens Axboe | aed3ea9 | 2014-12-22 14:04:42 -0700 | [diff] [blame] | 349 | |
| 350 | queue_for_each_hw_ctx(q, hctx, i) |
| 351 | if (blk_mq_hw_queue_mapped(hctx)) |
| 352 | blk_mq_tag_wakeup_all(hctx->tags, true); |
| 353 | } |
| 354 | |
Christoph Hellwig | 52fdbbc | 2021-11-17 07:13:59 +0100 | [diff] [blame] | 355 | void blk_rq_init(struct request_queue *q, struct request *rq) |
| 356 | { |
| 357 | memset(rq, 0, sizeof(*rq)); |
| 358 | |
| 359 | INIT_LIST_HEAD(&rq->queuelist); |
| 360 | rq->q = q; |
| 361 | rq->__sector = (sector_t) -1; |
| 362 | INIT_HLIST_NODE(&rq->hash); |
| 363 | RB_CLEAR_NODE(&rq->rb_node); |
| 364 | rq->tag = BLK_MQ_NO_TAG; |
| 365 | rq->internal_tag = BLK_MQ_NO_TAG; |
| 366 | rq->start_time_ns = ktime_get_ns(); |
| 367 | rq->part = NULL; |
| 368 | blk_crypto_rq_set_defaults(rq); |
| 369 | } |
| 370 | EXPORT_SYMBOL(blk_rq_init); |
| 371 | |
Christoph Hellwig | e4cdf1a | 2017-06-16 18:15:27 +0200 | [diff] [blame] | 372 | static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data, |
Jens Axboe | fe6134f | 2021-10-19 09:32:58 -0600 | [diff] [blame] | 373 | struct blk_mq_tags *tags, unsigned int tag, u64 alloc_time_ns) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 374 | { |
Pavel Begunkov | 605f784 | 2021-10-18 21:37:28 +0100 | [diff] [blame] | 375 | struct blk_mq_ctx *ctx = data->ctx; |
| 376 | struct blk_mq_hw_ctx *hctx = data->hctx; |
| 377 | struct request_queue *q = data->q; |
Christoph Hellwig | e4cdf1a | 2017-06-16 18:15:27 +0200 | [diff] [blame] | 378 | struct request *rq = tags->static_rqs[tag]; |
Bart Van Assche | c3a148d | 2017-06-20 11:15:43 -0700 | [diff] [blame] | 379 | |
Jens Axboe | c7b84d4 | 2021-10-19 09:33:00 -0600 | [diff] [blame] | 380 | rq->q = q; |
| 381 | rq->mq_ctx = ctx; |
| 382 | rq->mq_hctx = hctx; |
| 383 | rq->cmd_flags = data->cmd_flags; |
Christoph Hellwig | e4cdf1a | 2017-06-16 18:15:27 +0200 | [diff] [blame] | 384 | |
Pavel Begunkov | 1284590 | 2021-10-18 21:37:29 +0100 | [diff] [blame] | 385 | if (data->flags & BLK_MQ_REQ_PM) |
Jens Axboe | 56f8da6 | 2021-10-19 09:32:57 -0600 | [diff] [blame] | 386 | data->rq_flags |= RQF_PM; |
Pavel Begunkov | 1284590 | 2021-10-18 21:37:29 +0100 | [diff] [blame] | 387 | if (blk_queue_io_stat(q)) |
Jens Axboe | 56f8da6 | 2021-10-19 09:32:57 -0600 | [diff] [blame] | 388 | data->rq_flags |= RQF_IO_STAT; |
| 389 | rq->rq_flags = data->rq_flags; |
Pavel Begunkov | 1284590 | 2021-10-18 21:37:29 +0100 | [diff] [blame] | 390 | |
Jens Axboe | c7b84d4 | 2021-10-19 09:33:00 -0600 | [diff] [blame] | 391 | if (!(data->rq_flags & RQF_ELV)) { |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 392 | rq->tag = tag; |
| 393 | rq->internal_tag = BLK_MQ_NO_TAG; |
Jens Axboe | c7b84d4 | 2021-10-19 09:33:00 -0600 | [diff] [blame] | 394 | } else { |
| 395 | rq->tag = BLK_MQ_NO_TAG; |
| 396 | rq->internal_tag = tag; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 397 | } |
Jens Axboe | c7b84d4 | 2021-10-19 09:33:00 -0600 | [diff] [blame] | 398 | rq->timeout = 0; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 399 | |
Pavel Begunkov | 4f266f2 | 2021-10-18 21:37:27 +0100 | [diff] [blame] | 400 | if (blk_mq_need_time_stamp(rq)) |
| 401 | rq->start_time_ns = ktime_get_ns(); |
| 402 | else |
| 403 | rq->start_time_ns = 0; |
Christoph Hellwig | af76e55 | 2014-05-06 12:12:45 +0200 | [diff] [blame] | 404 | rq->part = NULL; |
Tejun Heo | 6f816b4 | 2019-08-28 15:05:57 -0700 | [diff] [blame] | 405 | #ifdef CONFIG_BLK_RQ_ALLOC_TIME |
| 406 | rq->alloc_time_ns = alloc_time_ns; |
| 407 | #endif |
Omar Sandoval | 544ccc8d | 2018-05-09 02:08:50 -0700 | [diff] [blame] | 408 | rq->io_start_time_ns = 0; |
Hou Tao | 3d24430 | 2019-05-21 15:59:03 +0800 | [diff] [blame] | 409 | rq->stats_sectors = 0; |
Christoph Hellwig | af76e55 | 2014-05-06 12:12:45 +0200 | [diff] [blame] | 410 | rq->nr_phys_segments = 0; |
| 411 | #if defined(CONFIG_BLK_DEV_INTEGRITY) |
| 412 | rq->nr_integrity_segments = 0; |
| 413 | #endif |
Christoph Hellwig | af76e55 | 2014-05-06 12:12:45 +0200 | [diff] [blame] | 414 | rq->end_io = NULL; |
| 415 | rq->end_io_data = NULL; |
Christoph Hellwig | af76e55 | 2014-05-06 12:12:45 +0200 | [diff] [blame] | 416 | |
Pavel Begunkov | 4f266f2 | 2021-10-18 21:37:27 +0100 | [diff] [blame] | 417 | blk_crypto_rq_set_defaults(rq); |
| 418 | INIT_LIST_HEAD(&rq->queuelist); |
| 419 | /* tag was already set */ |
| 420 | WRITE_ONCE(rq->deadline, 0); |
Jens Axboe | 0a467d0 | 2021-10-14 14:39:59 -0600 | [diff] [blame] | 421 | req_ref_set(rq, 1); |
Christoph Hellwig | 7ea4d8a | 2020-05-29 15:53:10 +0200 | [diff] [blame] | 422 | |
Pavel Begunkov | 4f266f2 | 2021-10-18 21:37:27 +0100 | [diff] [blame] | 423 | if (rq->rq_flags & RQF_ELV) { |
Christoph Hellwig | 7ea4d8a | 2020-05-29 15:53:10 +0200 | [diff] [blame] | 424 | struct elevator_queue *e = data->q->elevator; |
| 425 | |
Pavel Begunkov | 4f266f2 | 2021-10-18 21:37:27 +0100 | [diff] [blame] | 426 | INIT_HLIST_NODE(&rq->hash); |
| 427 | RB_CLEAR_NODE(&rq->rb_node); |
| 428 | |
| 429 | if (!op_is_flush(data->cmd_flags) && |
| 430 | e->type->ops.prepare_request) { |
Christoph Hellwig | 7ea4d8a | 2020-05-29 15:53:10 +0200 | [diff] [blame] | 431 | e->type->ops.prepare_request(rq); |
| 432 | rq->rq_flags |= RQF_ELVPRIV; |
| 433 | } |
| 434 | } |
| 435 | |
Christoph Hellwig | e4cdf1a | 2017-06-16 18:15:27 +0200 | [diff] [blame] | 436 | return rq; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 437 | } |
| 438 | |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 439 | static inline struct request * |
| 440 | __blk_mq_alloc_requests_batch(struct blk_mq_alloc_data *data, |
| 441 | u64 alloc_time_ns) |
| 442 | { |
| 443 | unsigned int tag, tag_offset; |
Jens Axboe | fe6134f | 2021-10-19 09:32:58 -0600 | [diff] [blame] | 444 | struct blk_mq_tags *tags; |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 445 | struct request *rq; |
Jens Axboe | fe6134f | 2021-10-19 09:32:58 -0600 | [diff] [blame] | 446 | unsigned long tag_mask; |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 447 | int i, nr = 0; |
| 448 | |
Jens Axboe | fe6134f | 2021-10-19 09:32:58 -0600 | [diff] [blame] | 449 | tag_mask = blk_mq_get_tags(data, data->nr_tags, &tag_offset); |
| 450 | if (unlikely(!tag_mask)) |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 451 | return NULL; |
| 452 | |
Jens Axboe | fe6134f | 2021-10-19 09:32:58 -0600 | [diff] [blame] | 453 | tags = blk_mq_tags_from_data(data); |
| 454 | for (i = 0; tag_mask; i++) { |
| 455 | if (!(tag_mask & (1UL << i))) |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 456 | continue; |
| 457 | tag = tag_offset + i; |
Jens Axboe | a22c00b | 2021-11-01 06:56:09 -0600 | [diff] [blame] | 458 | prefetch(tags->static_rqs[tag]); |
Jens Axboe | fe6134f | 2021-10-19 09:32:58 -0600 | [diff] [blame] | 459 | tag_mask &= ~(1UL << i); |
| 460 | rq = blk_mq_rq_ctx_init(data, tags, tag, alloc_time_ns); |
Jens Axboe | 013a7f9 | 2021-10-13 07:58:52 -0600 | [diff] [blame] | 461 | rq_list_add(data->cached_rq, rq); |
Jens Axboe | c5fc7b9 | 2021-11-03 05:49:07 -0600 | [diff] [blame] | 462 | nr++; |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 463 | } |
Jens Axboe | c5fc7b9 | 2021-11-03 05:49:07 -0600 | [diff] [blame] | 464 | /* caller already holds a reference, add for remainder */ |
| 465 | percpu_ref_get_many(&data->q->q_usage_counter, nr - 1); |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 466 | data->nr_tags -= nr; |
| 467 | |
Jens Axboe | 013a7f9 | 2021-10-13 07:58:52 -0600 | [diff] [blame] | 468 | return rq_list_pop(data->cached_rq); |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 469 | } |
| 470 | |
Christoph Hellwig | b90cfae | 2021-10-12 12:40:44 +0200 | [diff] [blame] | 471 | static struct request *__blk_mq_alloc_requests(struct blk_mq_alloc_data *data) |
Christoph Hellwig | d2c0d38 | 2017-06-16 18:15:19 +0200 | [diff] [blame] | 472 | { |
Christoph Hellwig | e6e7abf | 2020-05-29 15:53:09 +0200 | [diff] [blame] | 473 | struct request_queue *q = data->q; |
Tejun Heo | 6f816b4 | 2019-08-28 15:05:57 -0700 | [diff] [blame] | 474 | u64 alloc_time_ns = 0; |
Jens Axboe | 47c122e | 2021-10-06 06:34:11 -0600 | [diff] [blame] | 475 | struct request *rq; |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 476 | unsigned int tag; |
Christoph Hellwig | d2c0d38 | 2017-06-16 18:15:19 +0200 | [diff] [blame] | 477 | |
Tejun Heo | 6f816b4 | 2019-08-28 15:05:57 -0700 | [diff] [blame] | 478 | /* alloc_time includes depth and tag waits */ |
| 479 | if (blk_queue_rq_alloc_time(q)) |
| 480 | alloc_time_ns = ktime_get_ns(); |
| 481 | |
Jens Axboe | f9afca4 | 2018-10-29 13:11:38 -0600 | [diff] [blame] | 482 | if (data->cmd_flags & REQ_NOWAIT) |
Goldwyn Rodrigues | 03a07c9 | 2017-06-20 07:05:46 -0500 | [diff] [blame] | 483 | data->flags |= BLK_MQ_REQ_NOWAIT; |
Christoph Hellwig | d2c0d38 | 2017-06-16 18:15:19 +0200 | [diff] [blame] | 484 | |
Jens Axboe | 781dd83 | 2021-11-02 08:34:09 -0600 | [diff] [blame] | 485 | if (q->elevator) { |
| 486 | struct elevator_queue *e = q->elevator; |
| 487 | |
| 488 | data->rq_flags |= RQF_ELV; |
| 489 | |
Christoph Hellwig | d2c0d38 | 2017-06-16 18:15:19 +0200 | [diff] [blame] | 490 | /* |
Lin Feng | 8d663f3 | 2021-04-15 11:39:20 +0800 | [diff] [blame] | 491 | * Flush/passthrough requests are special and go directly to the |
Jens Axboe | 17a5119 | 2018-05-09 13:28:50 -0600 | [diff] [blame] | 492 | * dispatch list. Don't include reserved tags in the |
| 493 | * limiting, as it isn't useful. |
Christoph Hellwig | d2c0d38 | 2017-06-16 18:15:19 +0200 | [diff] [blame] | 494 | */ |
Jens Axboe | f9afca4 | 2018-10-29 13:11:38 -0600 | [diff] [blame] | 495 | if (!op_is_flush(data->cmd_flags) && |
Lin Feng | 8d663f3 | 2021-04-15 11:39:20 +0800 | [diff] [blame] | 496 | !blk_op_is_passthrough(data->cmd_flags) && |
Jens Axboe | f9afca4 | 2018-10-29 13:11:38 -0600 | [diff] [blame] | 497 | e->type->ops.limit_depth && |
Jens Axboe | 17a5119 | 2018-05-09 13:28:50 -0600 | [diff] [blame] | 498 | !(data->flags & BLK_MQ_REQ_RESERVED)) |
Jens Axboe | f9afca4 | 2018-10-29 13:11:38 -0600 | [diff] [blame] | 499 | e->type->ops.limit_depth(data->cmd_flags, data); |
Christoph Hellwig | d2c0d38 | 2017-06-16 18:15:19 +0200 | [diff] [blame] | 500 | } |
| 501 | |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 502 | retry: |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 503 | data->ctx = blk_mq_get_ctx(q); |
| 504 | data->hctx = blk_mq_map_queue(q, data->cmd_flags, data->ctx); |
Jens Axboe | 781dd83 | 2021-11-02 08:34:09 -0600 | [diff] [blame] | 505 | if (!(data->rq_flags & RQF_ELV)) |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 506 | blk_mq_tag_busy(data->hctx); |
| 507 | |
John Garry | 99e48cd | 2022-07-06 20:03:50 +0800 | [diff] [blame] | 508 | if (data->flags & BLK_MQ_REQ_RESERVED) |
| 509 | data->rq_flags |= RQF_RESV; |
| 510 | |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 511 | /* |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 512 | * Try batched alloc if we want more than 1 tag. |
| 513 | */ |
| 514 | if (data->nr_tags > 1) { |
| 515 | rq = __blk_mq_alloc_requests_batch(data, alloc_time_ns); |
| 516 | if (rq) |
| 517 | return rq; |
| 518 | data->nr_tags = 1; |
| 519 | } |
| 520 | |
| 521 | /* |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 522 | * Waiting allocations only fail because of an inactive hctx. In that |
| 523 | * case just retry the hctx assignment and tag allocation as CPU hotplug |
| 524 | * should have migrated us to an online CPU by now. |
| 525 | */ |
Christoph Hellwig | e4cdf1a | 2017-06-16 18:15:27 +0200 | [diff] [blame] | 526 | tag = blk_mq_get_tag(data); |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 527 | if (tag == BLK_MQ_NO_TAG) { |
| 528 | if (data->flags & BLK_MQ_REQ_NOWAIT) |
| 529 | return NULL; |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 530 | /* |
Jens Axboe | 349302d | 2021-10-09 13:10:39 -0600 | [diff] [blame] | 531 | * Give up the CPU and sleep for a random short time to |
| 532 | * ensure that thread using a realtime scheduling class |
| 533 | * are migrated off the CPU, and thus off the hctx that |
| 534 | * is going away. |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 535 | */ |
| 536 | msleep(3); |
| 537 | goto retry; |
| 538 | } |
Christoph Hellwig | d2c0d38 | 2017-06-16 18:15:19 +0200 | [diff] [blame] | 539 | |
Jens Axboe | fe6134f | 2021-10-19 09:32:58 -0600 | [diff] [blame] | 540 | return blk_mq_rq_ctx_init(data, blk_mq_tags_from_data(data), tag, |
| 541 | alloc_time_ns); |
Christoph Hellwig | d2c0d38 | 2017-06-16 18:15:19 +0200 | [diff] [blame] | 542 | } |
| 543 | |
Jens Axboe | 4b6a5d9 | 2022-09-21 08:22:09 -0600 | [diff] [blame] | 544 | static struct request *blk_mq_rq_cache_fill(struct request_queue *q, |
| 545 | struct blk_plug *plug, |
| 546 | blk_opf_t opf, |
| 547 | blk_mq_req_flags_t flags) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 548 | { |
Christoph Hellwig | e6e7abf | 2020-05-29 15:53:09 +0200 | [diff] [blame] | 549 | struct blk_mq_alloc_data data = { |
| 550 | .q = q, |
| 551 | .flags = flags, |
Bart Van Assche | 16458cf | 2022-07-14 11:06:32 -0700 | [diff] [blame] | 552 | .cmd_flags = opf, |
Jens Axboe | 4b6a5d9 | 2022-09-21 08:22:09 -0600 | [diff] [blame] | 553 | .nr_tags = plug->nr_ios, |
| 554 | .cached_rq = &plug->cached_rq, |
Christoph Hellwig | e6e7abf | 2020-05-29 15:53:09 +0200 | [diff] [blame] | 555 | }; |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 556 | struct request *rq; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 557 | |
Jens Axboe | 4b6a5d9 | 2022-09-21 08:22:09 -0600 | [diff] [blame] | 558 | if (blk_queue_enter(q, flags)) |
| 559 | return NULL; |
| 560 | |
| 561 | plug->nr_ios = 1; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 562 | |
Christoph Hellwig | b90cfae | 2021-10-12 12:40:44 +0200 | [diff] [blame] | 563 | rq = __blk_mq_alloc_requests(&data); |
Jens Axboe | 4b6a5d9 | 2022-09-21 08:22:09 -0600 | [diff] [blame] | 564 | if (unlikely(!rq)) |
| 565 | blk_queue_exit(q); |
| 566 | return rq; |
| 567 | } |
| 568 | |
| 569 | static struct request *blk_mq_alloc_cached_request(struct request_queue *q, |
| 570 | blk_opf_t opf, |
| 571 | blk_mq_req_flags_t flags) |
| 572 | { |
| 573 | struct blk_plug *plug = current->plug; |
| 574 | struct request *rq; |
| 575 | |
| 576 | if (!plug) |
| 577 | return NULL; |
Jinlong Chen | 4046728 | 2022-11-02 10:52:30 +0800 | [diff] [blame] | 578 | |
Jens Axboe | 4b6a5d9 | 2022-09-21 08:22:09 -0600 | [diff] [blame] | 579 | if (rq_list_empty(plug->cached_rq)) { |
| 580 | if (plug->nr_ios == 1) |
| 581 | return NULL; |
| 582 | rq = blk_mq_rq_cache_fill(q, plug, opf, flags); |
Jinlong Chen | 4046728 | 2022-11-02 10:52:30 +0800 | [diff] [blame] | 583 | if (!rq) |
| 584 | return NULL; |
| 585 | } else { |
| 586 | rq = rq_list_peek(&plug->cached_rq); |
| 587 | if (!rq || rq->q != q) |
| 588 | return NULL; |
| 589 | |
| 590 | if (blk_mq_get_hctx_type(opf) != rq->mq_hctx->type) |
| 591 | return NULL; |
| 592 | if (op_is_flush(rq->cmd_flags) != op_is_flush(opf)) |
| 593 | return NULL; |
| 594 | |
| 595 | plug->cached_rq = rq_list_next(rq); |
Jens Axboe | 4b6a5d9 | 2022-09-21 08:22:09 -0600 | [diff] [blame] | 596 | } |
Jens Axboe | 4b6a5d9 | 2022-09-21 08:22:09 -0600 | [diff] [blame] | 597 | |
Jens Axboe | 4b6a5d9 | 2022-09-21 08:22:09 -0600 | [diff] [blame] | 598 | rq->cmd_flags = opf; |
| 599 | INIT_LIST_HEAD(&rq->queuelist); |
| 600 | return rq; |
| 601 | } |
| 602 | |
| 603 | struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf, |
| 604 | blk_mq_req_flags_t flags) |
| 605 | { |
| 606 | struct request *rq; |
| 607 | |
| 608 | rq = blk_mq_alloc_cached_request(q, opf, flags); |
| 609 | if (!rq) { |
| 610 | struct blk_mq_alloc_data data = { |
| 611 | .q = q, |
| 612 | .flags = flags, |
| 613 | .cmd_flags = opf, |
| 614 | .nr_tags = 1, |
| 615 | }; |
| 616 | int ret; |
| 617 | |
| 618 | ret = blk_queue_enter(q, flags); |
| 619 | if (ret) |
| 620 | return ERR_PTR(ret); |
| 621 | |
| 622 | rq = __blk_mq_alloc_requests(&data); |
| 623 | if (!rq) |
| 624 | goto out_queue_exit; |
| 625 | } |
Christoph Hellwig | 0c4de0f | 2016-07-19 11:31:50 +0200 | [diff] [blame] | 626 | rq->__data_len = 0; |
| 627 | rq->__sector = (sector_t) -1; |
| 628 | rq->bio = rq->biotail = NULL; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 629 | return rq; |
Christoph Hellwig | a5ea58110 | 2020-05-16 20:27:58 +0200 | [diff] [blame] | 630 | out_queue_exit: |
| 631 | blk_queue_exit(q); |
| 632 | return ERR_PTR(-EWOULDBLOCK); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 633 | } |
Jens Axboe | 4bb659b | 2014-05-09 09:36:49 -0600 | [diff] [blame] | 634 | EXPORT_SYMBOL(blk_mq_alloc_request); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 635 | |
Bart Van Assche | cd6ce14 | 2017-06-20 11:15:39 -0700 | [diff] [blame] | 636 | struct request *blk_mq_alloc_request_hctx(struct request_queue *q, |
Bart Van Assche | 16458cf | 2022-07-14 11:06:32 -0700 | [diff] [blame] | 637 | blk_opf_t opf, blk_mq_req_flags_t flags, unsigned int hctx_idx) |
Ming Lin | 1f5bd33 | 2016-06-13 16:45:21 +0200 | [diff] [blame] | 638 | { |
Christoph Hellwig | e6e7abf | 2020-05-29 15:53:09 +0200 | [diff] [blame] | 639 | struct blk_mq_alloc_data data = { |
| 640 | .q = q, |
| 641 | .flags = flags, |
Bart Van Assche | 16458cf | 2022-07-14 11:06:32 -0700 | [diff] [blame] | 642 | .cmd_flags = opf, |
Jens Axboe | 47c122e | 2021-10-06 06:34:11 -0600 | [diff] [blame] | 643 | .nr_tags = 1, |
Christoph Hellwig | e6e7abf | 2020-05-29 15:53:09 +0200 | [diff] [blame] | 644 | }; |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 645 | u64 alloc_time_ns = 0; |
John Garry | e3c5a78 | 2022-10-26 18:35:13 +0800 | [diff] [blame] | 646 | struct request *rq; |
Omar Sandoval | 6d2809d | 2017-02-27 10:28:27 -0800 | [diff] [blame] | 647 | unsigned int cpu; |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 648 | unsigned int tag; |
Ming Lin | 1f5bd33 | 2016-06-13 16:45:21 +0200 | [diff] [blame] | 649 | int ret; |
| 650 | |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 651 | /* alloc_time includes depth and tag waits */ |
| 652 | if (blk_queue_rq_alloc_time(q)) |
| 653 | alloc_time_ns = ktime_get_ns(); |
| 654 | |
Ming Lin | 1f5bd33 | 2016-06-13 16:45:21 +0200 | [diff] [blame] | 655 | /* |
| 656 | * If the tag allocator sleeps we could get an allocation for a |
| 657 | * different hardware context. No need to complicate the low level |
| 658 | * allocator for this for the rare use case of a command tied to |
| 659 | * a specific queue. |
| 660 | */ |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 661 | if (WARN_ON_ONCE(!(flags & (BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_RESERVED)))) |
Ming Lin | 1f5bd33 | 2016-06-13 16:45:21 +0200 | [diff] [blame] | 662 | return ERR_PTR(-EINVAL); |
| 663 | |
| 664 | if (hctx_idx >= q->nr_hw_queues) |
| 665 | return ERR_PTR(-EIO); |
| 666 | |
Bart Van Assche | 3a0a529 | 2017-11-09 10:49:58 -0800 | [diff] [blame] | 667 | ret = blk_queue_enter(q, flags); |
Ming Lin | 1f5bd33 | 2016-06-13 16:45:21 +0200 | [diff] [blame] | 668 | if (ret) |
| 669 | return ERR_PTR(ret); |
| 670 | |
Christoph Hellwig | c8712c6 | 2016-09-23 10:25:48 -0600 | [diff] [blame] | 671 | /* |
| 672 | * Check if the hardware context is actually mapped to anything. |
| 673 | * If not tell the caller that it should skip this queue. |
| 674 | */ |
Christoph Hellwig | a5ea58110 | 2020-05-16 20:27:58 +0200 | [diff] [blame] | 675 | ret = -EXDEV; |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 676 | data.hctx = xa_load(&q->hctx_table, hctx_idx); |
Christoph Hellwig | e6e7abf | 2020-05-29 15:53:09 +0200 | [diff] [blame] | 677 | if (!blk_mq_hw_queue_mapped(data.hctx)) |
Christoph Hellwig | a5ea58110 | 2020-05-16 20:27:58 +0200 | [diff] [blame] | 678 | goto out_queue_exit; |
Christoph Hellwig | e6e7abf | 2020-05-29 15:53:09 +0200 | [diff] [blame] | 679 | cpu = cpumask_first_and(data.hctx->cpumask, cpu_online_mask); |
Bart Van Assche | 14dc7a1 | 2022-06-15 14:00:04 -0700 | [diff] [blame] | 680 | if (cpu >= nr_cpu_ids) |
| 681 | goto out_queue_exit; |
Christoph Hellwig | e6e7abf | 2020-05-29 15:53:09 +0200 | [diff] [blame] | 682 | data.ctx = __blk_mq_get_ctx(q, cpu); |
Ming Lin | 1f5bd33 | 2016-06-13 16:45:21 +0200 | [diff] [blame] | 683 | |
Christoph Hellwig | 42fdc5e | 2020-06-29 17:08:34 +0200 | [diff] [blame] | 684 | if (!q->elevator) |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 685 | blk_mq_tag_busy(data.hctx); |
Jens Axboe | 781dd83 | 2021-11-02 08:34:09 -0600 | [diff] [blame] | 686 | else |
| 687 | data.rq_flags |= RQF_ELV; |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 688 | |
John Garry | 99e48cd | 2022-07-06 20:03:50 +0800 | [diff] [blame] | 689 | if (flags & BLK_MQ_REQ_RESERVED) |
| 690 | data.rq_flags |= RQF_RESV; |
| 691 | |
Christoph Hellwig | a5ea58110 | 2020-05-16 20:27:58 +0200 | [diff] [blame] | 692 | ret = -EWOULDBLOCK; |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 693 | tag = blk_mq_get_tag(&data); |
| 694 | if (tag == BLK_MQ_NO_TAG) |
Christoph Hellwig | a5ea58110 | 2020-05-16 20:27:58 +0200 | [diff] [blame] | 695 | goto out_queue_exit; |
John Garry | e3c5a78 | 2022-10-26 18:35:13 +0800 | [diff] [blame] | 696 | rq = blk_mq_rq_ctx_init(&data, blk_mq_tags_from_data(&data), tag, |
Jens Axboe | fe6134f | 2021-10-19 09:32:58 -0600 | [diff] [blame] | 697 | alloc_time_ns); |
John Garry | e3c5a78 | 2022-10-26 18:35:13 +0800 | [diff] [blame] | 698 | rq->__data_len = 0; |
| 699 | rq->__sector = (sector_t) -1; |
| 700 | rq->bio = rq->biotail = NULL; |
| 701 | return rq; |
Christoph Hellwig | 600c3b0 | 2020-05-29 15:53:13 +0200 | [diff] [blame] | 702 | |
Christoph Hellwig | a5ea58110 | 2020-05-16 20:27:58 +0200 | [diff] [blame] | 703 | out_queue_exit: |
| 704 | blk_queue_exit(q); |
| 705 | return ERR_PTR(ret); |
Ming Lin | 1f5bd33 | 2016-06-13 16:45:21 +0200 | [diff] [blame] | 706 | } |
| 707 | EXPORT_SYMBOL_GPL(blk_mq_alloc_request_hctx); |
| 708 | |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 709 | static void __blk_mq_free_request(struct request *rq) |
| 710 | { |
| 711 | struct request_queue *q = rq->q; |
| 712 | struct blk_mq_ctx *ctx = rq->mq_ctx; |
Jens Axboe | ea4f995 | 2018-10-29 15:06:13 -0600 | [diff] [blame] | 713 | struct blk_mq_hw_ctx *hctx = rq->mq_hctx; |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 714 | const int sched_tag = rq->internal_tag; |
| 715 | |
Satya Tangirala | a892c8d | 2020-05-14 00:37:18 +0000 | [diff] [blame] | 716 | blk_crypto_free_request(rq); |
Bart Van Assche | 986d413 | 2018-09-26 14:01:10 -0700 | [diff] [blame] | 717 | blk_pm_mark_last_busy(rq); |
Jens Axboe | ea4f995 | 2018-10-29 15:06:13 -0600 | [diff] [blame] | 718 | rq->mq_hctx = NULL; |
Christoph Hellwig | 76647368 | 2020-05-29 15:53:12 +0200 | [diff] [blame] | 719 | if (rq->tag != BLK_MQ_NO_TAG) |
John Garry | cae740a | 2020-02-26 20:10:15 +0800 | [diff] [blame] | 720 | blk_mq_put_tag(hctx->tags, ctx, rq->tag); |
Christoph Hellwig | 76647368 | 2020-05-29 15:53:12 +0200 | [diff] [blame] | 721 | if (sched_tag != BLK_MQ_NO_TAG) |
John Garry | cae740a | 2020-02-26 20:10:15 +0800 | [diff] [blame] | 722 | blk_mq_put_tag(hctx->sched_tags, ctx, sched_tag); |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 723 | blk_mq_sched_restart(hctx); |
| 724 | blk_queue_exit(q); |
| 725 | } |
| 726 | |
Christoph Hellwig | 6af5405 | 2017-06-16 18:15:22 +0200 | [diff] [blame] | 727 | void blk_mq_free_request(struct request *rq) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 728 | { |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 729 | struct request_queue *q = rq->q; |
Jens Axboe | ea4f995 | 2018-10-29 15:06:13 -0600 | [diff] [blame] | 730 | struct blk_mq_hw_ctx *hctx = rq->mq_hctx; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 731 | |
Christoph Hellwig | 222ee58 | 2021-11-26 12:58:11 +0100 | [diff] [blame] | 732 | if ((rq->rq_flags & RQF_ELVPRIV) && |
| 733 | q->elevator->type->ops.finish_request) |
| 734 | q->elevator->type->ops.finish_request(rq); |
Christoph Hellwig | 6af5405 | 2017-06-16 18:15:22 +0200 | [diff] [blame] | 735 | |
Christoph Hellwig | e806402 | 2016-10-20 15:12:13 +0200 | [diff] [blame] | 736 | if (rq->rq_flags & RQF_MQ_INFLIGHT) |
John Garry | bccf5e2 | 2020-08-19 23:20:26 +0800 | [diff] [blame] | 737 | __blk_mq_dec_active_requests(hctx); |
Jens Axboe | 87760e5 | 2016-11-09 12:38:14 -0700 | [diff] [blame] | 738 | |
Jens Axboe | 7beb2f8 | 2017-09-30 02:08:24 -0600 | [diff] [blame] | 739 | if (unlikely(laptop_mode && !blk_rq_is_passthrough(rq))) |
Christoph Hellwig | d152c68 | 2021-08-16 15:46:24 +0200 | [diff] [blame] | 740 | laptop_io_completion(q->disk->bdi); |
Jens Axboe | 7beb2f8 | 2017-09-30 02:08:24 -0600 | [diff] [blame] | 741 | |
Josef Bacik | a790504 | 2018-07-03 09:32:35 -0600 | [diff] [blame] | 742 | rq_qos_done(q, rq); |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 743 | |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 744 | WRITE_ONCE(rq->state, MQ_RQ_IDLE); |
Jens Axboe | 0a467d0 | 2021-10-14 14:39:59 -0600 | [diff] [blame] | 745 | if (req_ref_put_and_test(rq)) |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 746 | __blk_mq_free_request(rq); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 747 | } |
Jens Axboe | 1a3b595 | 2014-11-17 10:40:48 -0700 | [diff] [blame] | 748 | EXPORT_SYMBOL_GPL(blk_mq_free_request); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 749 | |
Jens Axboe | 47c122e | 2021-10-06 06:34:11 -0600 | [diff] [blame] | 750 | void blk_mq_free_plug_rqs(struct blk_plug *plug) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 751 | { |
Jens Axboe | 013a7f9 | 2021-10-13 07:58:52 -0600 | [diff] [blame] | 752 | struct request *rq; |
Jens Axboe | fe1f452 | 2018-11-28 10:50:07 -0700 | [diff] [blame] | 753 | |
Jens Axboe | c5fc7b9 | 2021-11-03 05:49:07 -0600 | [diff] [blame] | 754 | while ((rq = rq_list_pop(&plug->cached_rq)) != NULL) |
Jens Axboe | 47c122e | 2021-10-06 06:34:11 -0600 | [diff] [blame] | 755 | blk_mq_free_request(rq); |
Jens Axboe | 47c122e | 2021-10-06 06:34:11 -0600 | [diff] [blame] | 756 | } |
Omar Sandoval | 522a777 | 2018-05-09 02:08:53 -0700 | [diff] [blame] | 757 | |
Christoph Hellwig | 22350ad | 2021-11-17 07:14:02 +0100 | [diff] [blame] | 758 | void blk_dump_rq_flags(struct request *rq, char *msg) |
| 759 | { |
| 760 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
Christoph Hellwig | f3fa33a | 2021-11-26 13:18:00 +0100 | [diff] [blame] | 761 | rq->q->disk ? rq->q->disk->disk_name : "?", |
Bart Van Assche | 16458cf | 2022-07-14 11:06:32 -0700 | [diff] [blame] | 762 | (__force unsigned long long) rq->cmd_flags); |
Christoph Hellwig | 22350ad | 2021-11-17 07:14:02 +0100 | [diff] [blame] | 763 | |
| 764 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
| 765 | (unsigned long long)blk_rq_pos(rq), |
| 766 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); |
| 767 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
| 768 | rq->bio, rq->biotail, blk_rq_bytes(rq)); |
| 769 | } |
| 770 | EXPORT_SYMBOL(blk_dump_rq_flags); |
| 771 | |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 772 | static void req_bio_endio(struct request *rq, struct bio *bio, |
| 773 | unsigned int nbytes, blk_status_t error) |
| 774 | { |
Pavel Begunkov | 478eb72 | 2021-10-19 22:24:12 +0100 | [diff] [blame] | 775 | if (unlikely(error)) { |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 776 | bio->bi_status = error; |
Pavel Begunkov | 478eb72 | 2021-10-19 22:24:12 +0100 | [diff] [blame] | 777 | } else if (req_op(rq) == REQ_OP_ZONE_APPEND) { |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 778 | /* |
| 779 | * Partial zone append completions cannot be supported as the |
| 780 | * BIO fragments may end up not being written sequentially. |
| 781 | */ |
Pavel Begunkov | 297db73 | 2021-10-22 16:01:44 +0100 | [diff] [blame] | 782 | if (bio->bi_iter.bi_size != nbytes) |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 783 | bio->bi_status = BLK_STS_IOERR; |
| 784 | else |
| 785 | bio->bi_iter.bi_sector = rq->__sector; |
| 786 | } |
| 787 | |
Pavel Begunkov | 478eb72 | 2021-10-19 22:24:12 +0100 | [diff] [blame] | 788 | bio_advance(bio, nbytes); |
| 789 | |
| 790 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
| 791 | bio_set_flag(bio, BIO_QUIET); |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 792 | /* don't actually finish bio if it's part of flush sequence */ |
| 793 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
| 794 | bio_endio(bio); |
| 795 | } |
| 796 | |
| 797 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
| 798 | { |
| 799 | if (req->part && blk_do_io_stat(req)) { |
| 800 | const int sgrp = op_stat_group(req_op(req)); |
| 801 | |
| 802 | part_stat_lock(); |
| 803 | part_stat_add(req->part, sectors[sgrp], bytes >> 9); |
| 804 | part_stat_unlock(); |
| 805 | } |
| 806 | } |
| 807 | |
Christoph Hellwig | 0d7a29a | 2021-11-17 07:14:03 +0100 | [diff] [blame] | 808 | static void blk_print_req_error(struct request *req, blk_status_t status) |
| 809 | { |
| 810 | printk_ratelimited(KERN_ERR |
| 811 | "%s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " |
| 812 | "phys_seg %u prio class %u\n", |
| 813 | blk_status_to_str(status), |
Christoph Hellwig | f3fa33a | 2021-11-26 13:18:00 +0100 | [diff] [blame] | 814 | req->q->disk ? req->q->disk->disk_name : "?", |
Bart Van Assche | 16458cf | 2022-07-14 11:06:32 -0700 | [diff] [blame] | 815 | blk_rq_pos(req), (__force u32)req_op(req), |
| 816 | blk_op_str(req_op(req)), |
| 817 | (__force u32)(req->cmd_flags & ~REQ_OP_MASK), |
Christoph Hellwig | 0d7a29a | 2021-11-17 07:14:03 +0100 | [diff] [blame] | 818 | req->nr_phys_segments, |
| 819 | IOPRIO_PRIO_CLASS(req->ioprio)); |
| 820 | } |
| 821 | |
Jens Axboe | 5581a5d | 2021-12-01 15:01:51 -0700 | [diff] [blame] | 822 | /* |
| 823 | * Fully end IO on a request. Does not support partial completions, or |
| 824 | * errors. |
| 825 | */ |
| 826 | static void blk_complete_request(struct request *req) |
| 827 | { |
| 828 | const bool is_flush = (req->rq_flags & RQF_FLUSH_SEQ) != 0; |
| 829 | int total_bytes = blk_rq_bytes(req); |
| 830 | struct bio *bio = req->bio; |
| 831 | |
| 832 | trace_block_rq_complete(req, BLK_STS_OK, total_bytes); |
| 833 | |
| 834 | if (!bio) |
| 835 | return; |
| 836 | |
| 837 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
| 838 | if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ) |
| 839 | req->q->integrity.profile->complete_fn(req, total_bytes); |
| 840 | #endif |
| 841 | |
| 842 | blk_account_io_completion(req, total_bytes); |
| 843 | |
| 844 | do { |
| 845 | struct bio *next = bio->bi_next; |
| 846 | |
| 847 | /* Completion has already been traced */ |
| 848 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
Pankaj Raghav | a12821d | 2022-02-11 10:34:25 +0100 | [diff] [blame] | 849 | |
| 850 | if (req_op(req) == REQ_OP_ZONE_APPEND) |
| 851 | bio->bi_iter.bi_sector = req->__sector; |
| 852 | |
Jens Axboe | 5581a5d | 2021-12-01 15:01:51 -0700 | [diff] [blame] | 853 | if (!is_flush) |
| 854 | bio_endio(bio); |
| 855 | bio = next; |
| 856 | } while (bio); |
| 857 | |
| 858 | /* |
| 859 | * Reset counters so that the request stacking driver |
| 860 | * can find how many bytes remain in the request |
| 861 | * later. |
| 862 | */ |
Jens Axboe | ab3e1d3 | 2022-09-21 08:24:16 -0600 | [diff] [blame] | 863 | if (!req->end_io) { |
| 864 | req->bio = NULL; |
| 865 | req->__data_len = 0; |
| 866 | } |
Jens Axboe | 5581a5d | 2021-12-01 15:01:51 -0700 | [diff] [blame] | 867 | } |
| 868 | |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 869 | /** |
| 870 | * blk_update_request - Complete multiple bytes without completing the request |
| 871 | * @req: the request being processed |
| 872 | * @error: block status code |
| 873 | * @nr_bytes: number of bytes to complete for @req |
| 874 | * |
| 875 | * Description: |
| 876 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
| 877 | * the request structure even if @req doesn't have leftover. |
| 878 | * If @req has leftover, sets it up for the next range of segments. |
| 879 | * |
| 880 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees |
| 881 | * %false return from this function. |
| 882 | * |
| 883 | * Note: |
| 884 | * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in this function |
| 885 | * except in the consistency check at the end of this function. |
| 886 | * |
| 887 | * Return: |
| 888 | * %false - this request doesn't have any more data |
| 889 | * %true - this request has more data |
| 890 | **/ |
| 891 | bool blk_update_request(struct request *req, blk_status_t error, |
| 892 | unsigned int nr_bytes) |
| 893 | { |
| 894 | int total_bytes; |
| 895 | |
Christoph Hellwig | 8a7d267 | 2021-10-18 10:45:18 +0200 | [diff] [blame] | 896 | trace_block_rq_complete(req, error, nr_bytes); |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 897 | |
| 898 | if (!req->bio) |
| 899 | return false; |
| 900 | |
| 901 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
| 902 | if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ && |
| 903 | error == BLK_STS_OK) |
| 904 | req->q->integrity.profile->complete_fn(req, nr_bytes); |
| 905 | #endif |
| 906 | |
| 907 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
Christoph Hellwig | 3d973a7 | 2022-03-23 17:38:15 +0100 | [diff] [blame] | 908 | !(req->rq_flags & RQF_QUIET)) && |
| 909 | !test_bit(GD_DEAD, &req->q->disk->state)) { |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 910 | blk_print_req_error(req, error); |
Yang Shi | d5869fd | 2022-02-10 14:52:22 -0800 | [diff] [blame] | 911 | trace_block_rq_error(req, error, nr_bytes); |
| 912 | } |
Jens Axboe | 9be3e06 | 2021-10-14 09:17:01 -0600 | [diff] [blame] | 913 | |
| 914 | blk_account_io_completion(req, nr_bytes); |
| 915 | |
| 916 | total_bytes = 0; |
| 917 | while (req->bio) { |
| 918 | struct bio *bio = req->bio; |
| 919 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
| 920 | |
| 921 | if (bio_bytes == bio->bi_iter.bi_size) |
| 922 | req->bio = bio->bi_next; |
| 923 | |
| 924 | /* Completion has already been traced */ |
| 925 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
| 926 | req_bio_endio(req, bio, bio_bytes, error); |
| 927 | |
| 928 | total_bytes += bio_bytes; |
| 929 | nr_bytes -= bio_bytes; |
| 930 | |
| 931 | if (!nr_bytes) |
| 932 | break; |
| 933 | } |
| 934 | |
| 935 | /* |
| 936 | * completely done |
| 937 | */ |
| 938 | if (!req->bio) { |
| 939 | /* |
| 940 | * Reset counters so that the request stacking driver |
| 941 | * can find how many bytes remain in the request |
| 942 | * later. |
| 943 | */ |
| 944 | req->__data_len = 0; |
| 945 | return false; |
| 946 | } |
| 947 | |
| 948 | req->__data_len -= total_bytes; |
| 949 | |
| 950 | /* update sector only for requests with clear definition of sector */ |
| 951 | if (!blk_rq_is_passthrough(req)) |
| 952 | req->__sector += total_bytes >> 9; |
| 953 | |
| 954 | /* mixed attributes always follow the first bio */ |
| 955 | if (req->rq_flags & RQF_MIXED_MERGE) { |
| 956 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
| 957 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
| 958 | } |
| 959 | |
| 960 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
| 961 | /* |
| 962 | * If total number of sectors is less than the first segment |
| 963 | * size, something has gone terribly wrong. |
| 964 | */ |
| 965 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { |
| 966 | blk_dump_rq_flags(req, "request botched"); |
| 967 | req->__data_len = blk_rq_cur_bytes(req); |
| 968 | } |
| 969 | |
| 970 | /* recalculate the number of segments */ |
| 971 | req->nr_phys_segments = blk_recalc_rq_segments(req); |
| 972 | } |
| 973 | |
| 974 | return true; |
| 975 | } |
| 976 | EXPORT_SYMBOL_GPL(blk_update_request); |
| 977 | |
Christoph Hellwig | 450b787 | 2021-11-17 07:14:01 +0100 | [diff] [blame] | 978 | static void __blk_account_io_done(struct request *req, u64 now) |
| 979 | { |
| 980 | const int sgrp = op_stat_group(req_op(req)); |
| 981 | |
| 982 | part_stat_lock(); |
| 983 | update_io_ticks(req->part, jiffies, true); |
| 984 | part_stat_inc(req->part, ios[sgrp]); |
| 985 | part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns); |
| 986 | part_stat_unlock(); |
| 987 | } |
| 988 | |
| 989 | static inline void blk_account_io_done(struct request *req, u64 now) |
| 990 | { |
| 991 | /* |
| 992 | * Account IO completion. flush_rq isn't accounted as a |
| 993 | * normal IO on queueing nor completion. Accounting the |
| 994 | * containing request is enough. |
| 995 | */ |
| 996 | if (blk_do_io_stat(req) && req->part && |
| 997 | !(req->rq_flags & RQF_FLUSH_SEQ)) |
| 998 | __blk_account_io_done(req, now); |
| 999 | } |
| 1000 | |
| 1001 | static void __blk_account_io_start(struct request *rq) |
| 1002 | { |
Christoph Hellwig | 41fa722 | 2022-03-08 06:51:47 +0100 | [diff] [blame] | 1003 | /* |
| 1004 | * All non-passthrough requests are created from a bio with one |
| 1005 | * exception: when a flush command that is part of a flush sequence |
| 1006 | * generated by the state machine in blk-flush.c is cloned onto the |
| 1007 | * lower device by dm-multipath we can get here without a bio. |
| 1008 | */ |
| 1009 | if (rq->bio) |
Christoph Hellwig | 450b787 | 2021-11-17 07:14:01 +0100 | [diff] [blame] | 1010 | rq->part = rq->bio->bi_bdev; |
Christoph Hellwig | 41fa722 | 2022-03-08 06:51:47 +0100 | [diff] [blame] | 1011 | else |
Christoph Hellwig | f3fa33a | 2021-11-26 13:18:00 +0100 | [diff] [blame] | 1012 | rq->part = rq->q->disk->part0; |
Christoph Hellwig | 450b787 | 2021-11-17 07:14:01 +0100 | [diff] [blame] | 1013 | |
| 1014 | part_stat_lock(); |
| 1015 | update_io_ticks(rq->part, jiffies, false); |
| 1016 | part_stat_unlock(); |
| 1017 | } |
| 1018 | |
| 1019 | static inline void blk_account_io_start(struct request *req) |
| 1020 | { |
| 1021 | if (blk_do_io_stat(req)) |
| 1022 | __blk_account_io_start(req); |
| 1023 | } |
| 1024 | |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1025 | static inline void __blk_mq_end_request_acct(struct request *rq, u64 now) |
| 1026 | { |
Omar Sandoval | 4bc6339 | 2018-05-09 02:08:52 -0700 | [diff] [blame] | 1027 | if (rq->rq_flags & RQF_STATS) { |
| 1028 | blk_mq_poll_stats_start(rq->q); |
Omar Sandoval | 522a777 | 2018-05-09 02:08:53 -0700 | [diff] [blame] | 1029 | blk_stat_add(rq, now); |
Omar Sandoval | 4bc6339 | 2018-05-09 02:08:52 -0700 | [diff] [blame] | 1030 | } |
| 1031 | |
Baolin Wang | 8789009 | 2020-07-04 15:28:21 +0800 | [diff] [blame] | 1032 | blk_mq_sched_completed_request(rq, now); |
Omar Sandoval | 522a777 | 2018-05-09 02:08:53 -0700 | [diff] [blame] | 1033 | blk_account_io_done(rq, now); |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1034 | } |
| 1035 | |
Ming Lei | 0d11e6a | 2013-12-05 10:50:39 -0700 | [diff] [blame] | 1036 | inline void __blk_mq_end_request(struct request *rq, blk_status_t error) |
Christoph Hellwig | 91b6363 | 2014-04-16 09:44:53 +0200 | [diff] [blame] | 1037 | { |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1038 | if (blk_mq_need_time_stamp(rq)) |
| 1039 | __blk_mq_end_request_acct(rq, ktime_get_ns()); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1040 | |
| 1041 | if (rq->end_io) { |
Josef Bacik | a790504 | 2018-07-03 09:32:35 -0600 | [diff] [blame] | 1042 | rq_qos_done(rq->q, rq); |
Jens Axboe | de671d6 | 2022-09-21 15:19:54 -0600 | [diff] [blame] | 1043 | if (rq->end_io(rq, error) == RQ_END_IO_FREE) |
| 1044 | blk_mq_free_request(rq); |
Christoph Hellwig | 91b6363 | 2014-04-16 09:44:53 +0200 | [diff] [blame] | 1045 | } else { |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1046 | blk_mq_free_request(rq); |
Christoph Hellwig | 91b6363 | 2014-04-16 09:44:53 +0200 | [diff] [blame] | 1047 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1048 | } |
Christoph Hellwig | c8a446a | 2014-09-13 16:40:10 -0700 | [diff] [blame] | 1049 | EXPORT_SYMBOL(__blk_mq_end_request); |
Christoph Hellwig | 63151a4 | 2014-04-16 09:44:52 +0200 | [diff] [blame] | 1050 | |
Christoph Hellwig | 2a842ac | 2017-06-03 09:38:04 +0200 | [diff] [blame] | 1051 | void blk_mq_end_request(struct request *rq, blk_status_t error) |
Christoph Hellwig | 63151a4 | 2014-04-16 09:44:52 +0200 | [diff] [blame] | 1052 | { |
| 1053 | if (blk_update_request(rq, error, blk_rq_bytes(rq))) |
| 1054 | BUG(); |
Christoph Hellwig | c8a446a | 2014-09-13 16:40:10 -0700 | [diff] [blame] | 1055 | __blk_mq_end_request(rq, error); |
Christoph Hellwig | 63151a4 | 2014-04-16 09:44:52 +0200 | [diff] [blame] | 1056 | } |
Christoph Hellwig | c8a446a | 2014-09-13 16:40:10 -0700 | [diff] [blame] | 1057 | EXPORT_SYMBOL(blk_mq_end_request); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1058 | |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1059 | #define TAG_COMP_BATCH 32 |
| 1060 | |
| 1061 | static inline void blk_mq_flush_tag_batch(struct blk_mq_hw_ctx *hctx, |
| 1062 | int *tag_array, int nr_tags) |
| 1063 | { |
| 1064 | struct request_queue *q = hctx->queue; |
| 1065 | |
Ming Lei | 3b87c6e | 2021-11-02 23:36:19 +0800 | [diff] [blame] | 1066 | /* |
| 1067 | * All requests should have been marked as RQF_MQ_INFLIGHT, so |
| 1068 | * update hctx->nr_active in batch |
| 1069 | */ |
| 1070 | if (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED) |
| 1071 | __blk_mq_sub_active_requests(hctx, nr_tags); |
| 1072 | |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1073 | blk_mq_put_tags(hctx->tags, tag_array, nr_tags); |
| 1074 | percpu_ref_put_many(&q->q_usage_counter, nr_tags); |
| 1075 | } |
| 1076 | |
| 1077 | void blk_mq_end_request_batch(struct io_comp_batch *iob) |
| 1078 | { |
| 1079 | int tags[TAG_COMP_BATCH], nr_tags = 0; |
Jens Axboe | 02f7eab | 2021-10-28 12:08:34 -0600 | [diff] [blame] | 1080 | struct blk_mq_hw_ctx *cur_hctx = NULL; |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1081 | struct request *rq; |
| 1082 | u64 now = 0; |
| 1083 | |
| 1084 | if (iob->need_ts) |
| 1085 | now = ktime_get_ns(); |
| 1086 | |
| 1087 | while ((rq = rq_list_pop(&iob->req_list)) != NULL) { |
| 1088 | prefetch(rq->bio); |
| 1089 | prefetch(rq->rq_next); |
| 1090 | |
Jens Axboe | 5581a5d | 2021-12-01 15:01:51 -0700 | [diff] [blame] | 1091 | blk_complete_request(rq); |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1092 | if (iob->need_ts) |
| 1093 | __blk_mq_end_request_acct(rq, now); |
| 1094 | |
Jens Axboe | 98b26a0 | 2021-11-26 09:53:23 -0700 | [diff] [blame] | 1095 | rq_qos_done(rq->q, rq); |
| 1096 | |
Jens Axboe | ab3e1d3 | 2022-09-21 08:24:16 -0600 | [diff] [blame] | 1097 | /* |
| 1098 | * If end_io handler returns NONE, then it still has |
| 1099 | * ownership of the request. |
| 1100 | */ |
| 1101 | if (rq->end_io && rq->end_io(rq, 0) == RQ_END_IO_NONE) |
| 1102 | continue; |
| 1103 | |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1104 | WRITE_ONCE(rq->state, MQ_RQ_IDLE); |
Jens Axboe | 0a467d0 | 2021-10-14 14:39:59 -0600 | [diff] [blame] | 1105 | if (!req_ref_put_and_test(rq)) |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1106 | continue; |
| 1107 | |
| 1108 | blk_crypto_free_request(rq); |
| 1109 | blk_pm_mark_last_busy(rq); |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1110 | |
Jens Axboe | 02f7eab | 2021-10-28 12:08:34 -0600 | [diff] [blame] | 1111 | if (nr_tags == TAG_COMP_BATCH || cur_hctx != rq->mq_hctx) { |
| 1112 | if (cur_hctx) |
| 1113 | blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags); |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1114 | nr_tags = 0; |
Jens Axboe | 02f7eab | 2021-10-28 12:08:34 -0600 | [diff] [blame] | 1115 | cur_hctx = rq->mq_hctx; |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1116 | } |
| 1117 | tags[nr_tags++] = rq->tag; |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1118 | } |
| 1119 | |
| 1120 | if (nr_tags) |
Jens Axboe | 02f7eab | 2021-10-28 12:08:34 -0600 | [diff] [blame] | 1121 | blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags); |
Jens Axboe | f794f33 | 2021-10-08 05:50:46 -0600 | [diff] [blame] | 1122 | } |
| 1123 | EXPORT_SYMBOL_GPL(blk_mq_end_request_batch); |
| 1124 | |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1125 | static void blk_complete_reqs(struct llist_head *list) |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 1126 | { |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1127 | struct llist_node *entry = llist_reverse_order(llist_del_all(list)); |
| 1128 | struct request *rq, *next; |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 1129 | |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1130 | llist_for_each_entry_safe(rq, next, entry, ipi_list) |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 1131 | rq->q->mq_ops->complete(rq); |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 1132 | } |
| 1133 | |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1134 | static __latent_entropy void blk_done_softirq(struct softirq_action *h) |
Christoph Hellwig | 115243f | 2020-06-11 08:44:42 +0200 | [diff] [blame] | 1135 | { |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1136 | blk_complete_reqs(this_cpu_ptr(&blk_cpu_done)); |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 1137 | } |
| 1138 | |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 1139 | static int blk_softirq_cpu_dead(unsigned int cpu) |
| 1140 | { |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1141 | blk_complete_reqs(&per_cpu(blk_cpu_done, cpu)); |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 1142 | return 0; |
| 1143 | } |
| 1144 | |
Christoph Hellwig | 30a91cb | 2014-02-10 03:24:38 -0800 | [diff] [blame] | 1145 | static void __blk_mq_complete_request_remote(void *data) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1146 | { |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1147 | __raise_softirq_irqoff(BLOCK_SOFTIRQ); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1148 | } |
| 1149 | |
Christoph Hellwig | 96339526 | 2020-06-11 08:44:49 +0200 | [diff] [blame] | 1150 | static inline bool blk_mq_complete_need_ipi(struct request *rq) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1151 | { |
Christoph Hellwig | 96339526 | 2020-06-11 08:44:49 +0200 | [diff] [blame] | 1152 | int cpu = raw_smp_processor_id(); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1153 | |
Christoph Hellwig | 96339526 | 2020-06-11 08:44:49 +0200 | [diff] [blame] | 1154 | if (!IS_ENABLED(CONFIG_SMP) || |
| 1155 | !test_bit(QUEUE_FLAG_SAME_COMP, &rq->q->queue_flags)) |
| 1156 | return false; |
Sebastian Andrzej Siewior | 7142518 | 2020-12-04 20:13:54 +0100 | [diff] [blame] | 1157 | /* |
| 1158 | * With force threaded interrupts enabled, raising softirq from an SMP |
| 1159 | * function call will always result in waking the ksoftirqd thread. |
| 1160 | * This is probably worse than completing the request on a different |
| 1161 | * cache domain. |
| 1162 | */ |
Tanner Love | 91cc470 | 2021-06-02 14:03:38 -0400 | [diff] [blame] | 1163 | if (force_irqthreads()) |
Sebastian Andrzej Siewior | 7142518 | 2020-12-04 20:13:54 +0100 | [diff] [blame] | 1164 | return false; |
Christoph Hellwig | 96339526 | 2020-06-11 08:44:49 +0200 | [diff] [blame] | 1165 | |
| 1166 | /* same CPU or cache domain? Complete locally */ |
| 1167 | if (cpu == rq->mq_ctx->cpu || |
| 1168 | (!test_bit(QUEUE_FLAG_SAME_FORCE, &rq->q->queue_flags) && |
| 1169 | cpus_share_cache(cpu, rq->mq_ctx->cpu))) |
| 1170 | return false; |
| 1171 | |
| 1172 | /* don't try to IPI to an offline CPU */ |
| 1173 | return cpu_online(rq->mq_ctx->cpu); |
| 1174 | } |
| 1175 | |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1176 | static void blk_mq_complete_send_ipi(struct request *rq) |
| 1177 | { |
| 1178 | struct llist_head *list; |
| 1179 | unsigned int cpu; |
| 1180 | |
| 1181 | cpu = rq->mq_ctx->cpu; |
| 1182 | list = &per_cpu(blk_cpu_done, cpu); |
| 1183 | if (llist_add(&rq->ipi_list, list)) { |
| 1184 | INIT_CSD(&rq->csd, __blk_mq_complete_request_remote, rq); |
| 1185 | smp_call_function_single_async(cpu, &rq->csd); |
| 1186 | } |
| 1187 | } |
| 1188 | |
| 1189 | static void blk_mq_raise_softirq(struct request *rq) |
| 1190 | { |
| 1191 | struct llist_head *list; |
| 1192 | |
| 1193 | preempt_disable(); |
| 1194 | list = this_cpu_ptr(&blk_cpu_done); |
| 1195 | if (llist_add(&rq->ipi_list, list)) |
| 1196 | raise_softirq(BLOCK_SOFTIRQ); |
| 1197 | preempt_enable(); |
| 1198 | } |
| 1199 | |
Christoph Hellwig | 40d09b5 | 2020-06-11 08:44:50 +0200 | [diff] [blame] | 1200 | bool blk_mq_complete_request_remote(struct request *rq) |
| 1201 | { |
Keith Busch | af78ff7 | 2018-11-26 09:54:30 -0700 | [diff] [blame] | 1202 | WRITE_ONCE(rq->state, MQ_RQ_COMPLETE); |
Ming Lei | 36e7653 | 2018-09-28 16:42:20 +0800 | [diff] [blame] | 1203 | |
Jens Axboe | 4ab32bf | 2018-11-18 16:15:35 -0700 | [diff] [blame] | 1204 | /* |
Liu Song | f168420 | 2022-09-21 11:32:03 +0800 | [diff] [blame] | 1205 | * For request which hctx has only one ctx mapping, |
| 1206 | * or a polled request, always complete locally, |
| 1207 | * it's pointless to redirect the completion. |
Jens Axboe | 4ab32bf | 2018-11-18 16:15:35 -0700 | [diff] [blame] | 1208 | */ |
Liu Song | f168420 | 2022-09-21 11:32:03 +0800 | [diff] [blame] | 1209 | if (rq->mq_hctx->nr_ctx == 1 || |
| 1210 | rq->cmd_flags & REQ_POLLED) |
Christoph Hellwig | 40d09b5 | 2020-06-11 08:44:50 +0200 | [diff] [blame] | 1211 | return false; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1212 | |
Christoph Hellwig | 40d09b5 | 2020-06-11 08:44:50 +0200 | [diff] [blame] | 1213 | if (blk_mq_complete_need_ipi(rq)) { |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1214 | blk_mq_complete_send_ipi(rq); |
| 1215 | return true; |
Christoph Hellwig | 3d6efbf | 2014-01-08 09:33:37 -0800 | [diff] [blame] | 1216 | } |
Christoph Hellwig | 40d09b5 | 2020-06-11 08:44:50 +0200 | [diff] [blame] | 1217 | |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 1218 | if (rq->q->nr_hw_queues == 1) { |
| 1219 | blk_mq_raise_softirq(rq); |
| 1220 | return true; |
| 1221 | } |
| 1222 | return false; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1223 | } |
Christoph Hellwig | 40d09b5 | 2020-06-11 08:44:50 +0200 | [diff] [blame] | 1224 | EXPORT_SYMBOL_GPL(blk_mq_complete_request_remote); |
| 1225 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1226 | /** |
Christoph Hellwig | 15f73f5 | 2020-06-11 08:44:47 +0200 | [diff] [blame] | 1227 | * blk_mq_complete_request - end I/O on a request |
| 1228 | * @rq: the request being processed |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1229 | * |
Christoph Hellwig | 15f73f5 | 2020-06-11 08:44:47 +0200 | [diff] [blame] | 1230 | * Description: |
| 1231 | * Complete a request by scheduling the ->complete_rq operation. |
| 1232 | **/ |
| 1233 | void blk_mq_complete_request(struct request *rq) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1234 | { |
Christoph Hellwig | 40d09b5 | 2020-06-11 08:44:50 +0200 | [diff] [blame] | 1235 | if (!blk_mq_complete_request_remote(rq)) |
Christoph Hellwig | 96339526 | 2020-06-11 08:44:49 +0200 | [diff] [blame] | 1236 | rq->q->mq_ops->complete(rq); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1237 | } |
Christoph Hellwig | 15f73f5 | 2020-06-11 08:44:47 +0200 | [diff] [blame] | 1238 | EXPORT_SYMBOL(blk_mq_complete_request); |
Christoph Hellwig | 30a91cb | 2014-02-10 03:24:38 -0800 | [diff] [blame] | 1239 | |
| 1240 | /** |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 1241 | * blk_mq_start_request - Start processing a request |
| 1242 | * @rq: Pointer to request to be started |
| 1243 | * |
| 1244 | * Function used by device drivers to notify the block layer that a request |
| 1245 | * is going to be processed now, so blk layer can do proper initializations |
| 1246 | * such as starting the timeout timer. |
| 1247 | */ |
Christoph Hellwig | e249007 | 2014-09-13 16:40:09 -0700 | [diff] [blame] | 1248 | void blk_mq_start_request(struct request *rq) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1249 | { |
| 1250 | struct request_queue *q = rq->q; |
| 1251 | |
Christoph Hellwig | a54895f | 2020-12-03 17:21:39 +0100 | [diff] [blame] | 1252 | trace_block_rq_issue(rq); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1253 | |
Jens Axboe | cf43e6b | 2016-11-07 21:32:37 -0700 | [diff] [blame] | 1254 | if (test_bit(QUEUE_FLAG_STATS, &q->queue_flags)) { |
Tejun Heo | 4cddeac | 2022-04-27 09:49:12 -1000 | [diff] [blame] | 1255 | rq->io_start_time_ns = ktime_get_ns(); |
Hou Tao | 3d24430 | 2019-05-21 15:59:03 +0800 | [diff] [blame] | 1256 | rq->stats_sectors = blk_rq_sectors(rq); |
Jens Axboe | cf43e6b | 2016-11-07 21:32:37 -0700 | [diff] [blame] | 1257 | rq->rq_flags |= RQF_STATS; |
Josef Bacik | a790504 | 2018-07-03 09:32:35 -0600 | [diff] [blame] | 1258 | rq_qos_issue(q, rq); |
Jens Axboe | cf43e6b | 2016-11-07 21:32:37 -0700 | [diff] [blame] | 1259 | } |
| 1260 | |
Tejun Heo | 1d9bd51 | 2018-01-09 08:29:48 -0800 | [diff] [blame] | 1261 | WARN_ON_ONCE(blk_mq_rq_state(rq) != MQ_RQ_IDLE); |
Jens Axboe | 538b753 | 2014-09-16 10:37:37 -0600 | [diff] [blame] | 1262 | |
Tejun Heo | 1d9bd51 | 2018-01-09 08:29:48 -0800 | [diff] [blame] | 1263 | blk_add_timer(rq); |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 1264 | WRITE_ONCE(rq->state, MQ_RQ_IN_FLIGHT); |
Christoph Hellwig | 49f5baa | 2014-02-11 08:27:14 -0800 | [diff] [blame] | 1265 | |
Max Gurtovoy | 54d4e6a | 2019-09-16 18:44:29 +0300 | [diff] [blame] | 1266 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
| 1267 | if (blk_integrity_rq(rq) && req_op(rq) == REQ_OP_WRITE) |
| 1268 | q->integrity.profile->prepare_fn(rq); |
| 1269 | #endif |
Christoph Hellwig | 3e08773 | 2021-10-12 13:12:24 +0200 | [diff] [blame] | 1270 | if (rq->bio && rq->bio->bi_opf & REQ_POLLED) |
| 1271 | WRITE_ONCE(rq->bio->bi_cookie, blk_rq_to_qc(rq)); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1272 | } |
Christoph Hellwig | e249007 | 2014-09-13 16:40:09 -0700 | [diff] [blame] | 1273 | EXPORT_SYMBOL(blk_mq_start_request); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1274 | |
Ming Lei | a327c34 | 2022-05-12 22:00:10 +0800 | [diff] [blame] | 1275 | /* |
| 1276 | * Allow 2x BLK_MAX_REQUEST_COUNT requests on plug queue for multiple |
| 1277 | * queues. This is important for md arrays to benefit from merging |
| 1278 | * requests. |
| 1279 | */ |
| 1280 | static inline unsigned short blk_plug_max_rq_count(struct blk_plug *plug) |
| 1281 | { |
| 1282 | if (plug->multiple_queues) |
| 1283 | return BLK_MAX_REQUEST_COUNT * 2; |
| 1284 | return BLK_MAX_REQUEST_COUNT; |
| 1285 | } |
| 1286 | |
| 1287 | static void blk_add_rq_to_plug(struct blk_plug *plug, struct request *rq) |
| 1288 | { |
| 1289 | struct request *last = rq_list_peek(&plug->mq_list); |
| 1290 | |
| 1291 | if (!plug->rq_count) { |
| 1292 | trace_block_plug(rq->q); |
| 1293 | } else if (plug->rq_count >= blk_plug_max_rq_count(plug) || |
| 1294 | (!blk_queue_nomerges(rq->q) && |
| 1295 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE)) { |
| 1296 | blk_mq_flush_plug_list(plug, false); |
Al Viro | 878eb6e | 2022-11-01 00:54:13 +0000 | [diff] [blame] | 1297 | last = NULL; |
Ming Lei | a327c34 | 2022-05-12 22:00:10 +0800 | [diff] [blame] | 1298 | trace_block_plug(rq->q); |
| 1299 | } |
| 1300 | |
| 1301 | if (!plug->multiple_queues && last && last->q != rq->q) |
| 1302 | plug->multiple_queues = true; |
| 1303 | if (!plug->has_elevator && (rq->rq_flags & RQF_ELV)) |
| 1304 | plug->has_elevator = true; |
| 1305 | rq->rq_next = NULL; |
| 1306 | rq_list_add(&plug->mq_list, rq); |
| 1307 | plug->rq_count++; |
| 1308 | } |
| 1309 | |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1310 | /** |
| 1311 | * blk_execute_rq_nowait - insert a request to I/O scheduler for execution |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1312 | * @rq: request to insert |
| 1313 | * @at_head: insert request at head or tail of queue |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1314 | * |
| 1315 | * Description: |
| 1316 | * Insert a fully prepared request at the back of the I/O scheduler queue |
| 1317 | * for execution. Don't wait for completion. |
| 1318 | * |
| 1319 | * Note: |
| 1320 | * This function will invoke @done directly if the queue is dead. |
| 1321 | */ |
Christoph Hellwig | e2e53086 | 2022-05-24 14:15:30 +0200 | [diff] [blame] | 1322 | void blk_execute_rq_nowait(struct request *rq, bool at_head) |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1323 | { |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1324 | WARN_ON(irqs_disabled()); |
| 1325 | WARN_ON(!blk_rq_is_passthrough(rq)); |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1326 | |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1327 | blk_account_io_start(rq); |
Pankaj Raghav | 110fdb4 | 2022-09-29 16:41:41 +0200 | [diff] [blame] | 1328 | |
| 1329 | /* |
| 1330 | * As plugging can be enabled for passthrough requests on a zoned |
| 1331 | * device, directly accessing the plug instead of using blk_mq_plug() |
| 1332 | * should not have any consequences. |
| 1333 | */ |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1334 | if (current->plug) |
| 1335 | blk_add_rq_to_plug(current->plug, rq); |
| 1336 | else |
| 1337 | blk_mq_sched_insert_request(rq, at_head, true, false); |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1338 | } |
| 1339 | EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); |
| 1340 | |
Christoph Hellwig | 32ac5a9 | 2022-05-24 14:15:29 +0200 | [diff] [blame] | 1341 | struct blk_rq_wait { |
| 1342 | struct completion done; |
| 1343 | blk_status_t ret; |
| 1344 | }; |
| 1345 | |
Jens Axboe | de671d6 | 2022-09-21 15:19:54 -0600 | [diff] [blame] | 1346 | static enum rq_end_io_ret blk_end_sync_rq(struct request *rq, blk_status_t ret) |
Christoph Hellwig | 32ac5a9 | 2022-05-24 14:15:29 +0200 | [diff] [blame] | 1347 | { |
| 1348 | struct blk_rq_wait *wait = rq->end_io_data; |
| 1349 | |
| 1350 | wait->ret = ret; |
| 1351 | complete(&wait->done); |
Jens Axboe | de671d6 | 2022-09-21 15:19:54 -0600 | [diff] [blame] | 1352 | return RQ_END_IO_NONE; |
Christoph Hellwig | 32ac5a9 | 2022-05-24 14:15:29 +0200 | [diff] [blame] | 1353 | } |
| 1354 | |
Kanchan Joshi | c6e99ea | 2022-08-23 21:44:42 +0530 | [diff] [blame] | 1355 | bool blk_rq_is_poll(struct request *rq) |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1356 | { |
| 1357 | if (!rq->mq_hctx) |
| 1358 | return false; |
| 1359 | if (rq->mq_hctx->type != HCTX_TYPE_POLL) |
| 1360 | return false; |
| 1361 | if (WARN_ON_ONCE(!rq->bio)) |
| 1362 | return false; |
| 1363 | return true; |
| 1364 | } |
Kanchan Joshi | c6e99ea | 2022-08-23 21:44:42 +0530 | [diff] [blame] | 1365 | EXPORT_SYMBOL_GPL(blk_rq_is_poll); |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1366 | |
| 1367 | static void blk_rq_poll_completion(struct request *rq, struct completion *wait) |
| 1368 | { |
| 1369 | do { |
| 1370 | bio_poll(rq->bio, NULL, 0); |
| 1371 | cond_resched(); |
| 1372 | } while (!completion_done(wait)); |
| 1373 | } |
| 1374 | |
| 1375 | /** |
| 1376 | * blk_execute_rq - insert a request into queue for execution |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1377 | * @rq: request to insert |
| 1378 | * @at_head: insert request at head or tail of queue |
| 1379 | * |
| 1380 | * Description: |
| 1381 | * Insert a fully prepared request at the back of the I/O scheduler queue |
| 1382 | * for execution and wait for completion. |
| 1383 | * Return: The blk_status_t result provided to blk_mq_end_request(). |
| 1384 | */ |
Christoph Hellwig | b84ba30 | 2021-11-26 13:18:01 +0100 | [diff] [blame] | 1385 | blk_status_t blk_execute_rq(struct request *rq, bool at_head) |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1386 | { |
Christoph Hellwig | 32ac5a9 | 2022-05-24 14:15:29 +0200 | [diff] [blame] | 1387 | struct blk_rq_wait wait = { |
| 1388 | .done = COMPLETION_INITIALIZER_ONSTACK(wait.done), |
| 1389 | }; |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1390 | |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1391 | WARN_ON(irqs_disabled()); |
| 1392 | WARN_ON(!blk_rq_is_passthrough(rq)); |
| 1393 | |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1394 | rq->end_io_data = &wait; |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1395 | rq->end_io = blk_end_sync_rq; |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1396 | |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1397 | blk_account_io_start(rq); |
| 1398 | blk_mq_sched_insert_request(rq, at_head, true, false); |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1399 | |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1400 | if (blk_rq_is_poll(rq)) { |
Christoph Hellwig | 32ac5a9 | 2022-05-24 14:15:29 +0200 | [diff] [blame] | 1401 | blk_rq_poll_completion(rq, &wait.done); |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1402 | } else { |
| 1403 | /* |
| 1404 | * Prevent hang_check timer from firing at us during very long |
| 1405 | * I/O |
| 1406 | */ |
| 1407 | unsigned long hang_check = sysctl_hung_task_timeout_secs; |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1408 | |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1409 | if (hang_check) |
Christoph Hellwig | 32ac5a9 | 2022-05-24 14:15:29 +0200 | [diff] [blame] | 1410 | while (!wait_for_completion_io_timeout(&wait.done, |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1411 | hang_check * (HZ/2))) |
| 1412 | ; |
| 1413 | else |
Christoph Hellwig | 32ac5a9 | 2022-05-24 14:15:29 +0200 | [diff] [blame] | 1414 | wait_for_completion_io(&wait.done); |
Christoph Hellwig | ae948fd | 2022-05-24 14:15:28 +0200 | [diff] [blame] | 1415 | } |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1416 | |
Christoph Hellwig | 32ac5a9 | 2022-05-24 14:15:29 +0200 | [diff] [blame] | 1417 | return wait.ret; |
Christoph Hellwig | 4054cff | 2021-11-17 07:13:56 +0100 | [diff] [blame] | 1418 | } |
| 1419 | EXPORT_SYMBOL(blk_execute_rq); |
| 1420 | |
Christoph Hellwig | ed0791b | 2014-04-16 09:44:57 +0200 | [diff] [blame] | 1421 | static void __blk_mq_requeue_request(struct request *rq) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1422 | { |
| 1423 | struct request_queue *q = rq->q; |
| 1424 | |
Ming Lei | 923218f | 2017-11-02 23:24:38 +0800 | [diff] [blame] | 1425 | blk_mq_put_driver_tag(rq); |
| 1426 | |
Christoph Hellwig | a54895f | 2020-12-03 17:21:39 +0100 | [diff] [blame] | 1427 | trace_block_rq_requeue(rq); |
Josef Bacik | a790504 | 2018-07-03 09:32:35 -0600 | [diff] [blame] | 1428 | rq_qos_requeue(q, rq); |
Christoph Hellwig | 49f5baa | 2014-02-11 08:27:14 -0800 | [diff] [blame] | 1429 | |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 1430 | if (blk_mq_request_started(rq)) { |
| 1431 | WRITE_ONCE(rq->state, MQ_RQ_IDLE); |
Christoph Hellwig | da66126 | 2018-06-14 13:58:45 +0200 | [diff] [blame] | 1432 | rq->rq_flags &= ~RQF_TIMED_OUT; |
Christoph Hellwig | e249007 | 2014-09-13 16:40:09 -0700 | [diff] [blame] | 1433 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1434 | } |
| 1435 | |
Bart Van Assche | 2b053ac | 2016-10-28 17:21:41 -0700 | [diff] [blame] | 1436 | void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list) |
Christoph Hellwig | ed0791b | 2014-04-16 09:44:57 +0200 | [diff] [blame] | 1437 | { |
Christoph Hellwig | ed0791b | 2014-04-16 09:44:57 +0200 | [diff] [blame] | 1438 | __blk_mq_requeue_request(rq); |
Christoph Hellwig | ed0791b | 2014-04-16 09:44:57 +0200 | [diff] [blame] | 1439 | |
Ming Lei | 105976f | 2018-02-23 23:36:56 +0800 | [diff] [blame] | 1440 | /* this request will be re-inserted to io scheduler queue */ |
| 1441 | blk_mq_sched_requeue_request(rq); |
| 1442 | |
Bart Van Assche | 2b053ac | 2016-10-28 17:21:41 -0700 | [diff] [blame] | 1443 | blk_mq_add_to_requeue_list(rq, true, kick_requeue_list); |
Christoph Hellwig | ed0791b | 2014-04-16 09:44:57 +0200 | [diff] [blame] | 1444 | } |
| 1445 | EXPORT_SYMBOL(blk_mq_requeue_request); |
| 1446 | |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1447 | static void blk_mq_requeue_work(struct work_struct *work) |
| 1448 | { |
| 1449 | struct request_queue *q = |
Mike Snitzer | 2849450 | 2016-09-14 13:28:30 -0400 | [diff] [blame] | 1450 | container_of(work, struct request_queue, requeue_work.work); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1451 | LIST_HEAD(rq_list); |
| 1452 | struct request *rq, *next; |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1453 | |
Jens Axboe | 18e9781 | 2017-07-27 08:03:57 -0600 | [diff] [blame] | 1454 | spin_lock_irq(&q->requeue_lock); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1455 | list_splice_init(&q->requeue_list, &rq_list); |
Jens Axboe | 18e9781 | 2017-07-27 08:03:57 -0600 | [diff] [blame] | 1456 | spin_unlock_irq(&q->requeue_lock); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1457 | |
| 1458 | list_for_each_entry_safe(rq, next, &rq_list, queuelist) { |
Jianchao Wang | aef1897 | 2019-02-12 09:56:25 +0800 | [diff] [blame] | 1459 | if (!(rq->rq_flags & (RQF_SOFTBARRIER | RQF_DONTPREP))) |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1460 | continue; |
| 1461 | |
Christoph Hellwig | e806402 | 2016-10-20 15:12:13 +0200 | [diff] [blame] | 1462 | rq->rq_flags &= ~RQF_SOFTBARRIER; |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1463 | list_del_init(&rq->queuelist); |
Jianchao Wang | aef1897 | 2019-02-12 09:56:25 +0800 | [diff] [blame] | 1464 | /* |
| 1465 | * If RQF_DONTPREP, rq has contained some driver specific |
| 1466 | * data, so insert it to hctx dispatch list to avoid any |
| 1467 | * merge. |
| 1468 | */ |
| 1469 | if (rq->rq_flags & RQF_DONTPREP) |
Ming Lei | 01e99ae | 2020-02-25 09:04:32 +0800 | [diff] [blame] | 1470 | blk_mq_request_bypass_insert(rq, false, false); |
Jianchao Wang | aef1897 | 2019-02-12 09:56:25 +0800 | [diff] [blame] | 1471 | else |
| 1472 | blk_mq_sched_insert_request(rq, true, false, false); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1473 | } |
| 1474 | |
| 1475 | while (!list_empty(&rq_list)) { |
| 1476 | rq = list_entry(rq_list.next, struct request, queuelist); |
| 1477 | list_del_init(&rq->queuelist); |
Mike Snitzer | 9e97d29 | 2018-01-17 11:25:58 -0500 | [diff] [blame] | 1478 | blk_mq_sched_insert_request(rq, false, false, false); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1479 | } |
| 1480 | |
Bart Van Assche | 52d7f1b | 2016-10-28 17:20:32 -0700 | [diff] [blame] | 1481 | blk_mq_run_hw_queues(q, false); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1482 | } |
| 1483 | |
Bart Van Assche | 2b053ac | 2016-10-28 17:21:41 -0700 | [diff] [blame] | 1484 | void blk_mq_add_to_requeue_list(struct request *rq, bool at_head, |
| 1485 | bool kick_requeue_list) |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1486 | { |
| 1487 | struct request_queue *q = rq->q; |
| 1488 | unsigned long flags; |
| 1489 | |
| 1490 | /* |
| 1491 | * We abuse this flag that is otherwise used by the I/O scheduler to |
Jens Axboe | ff821d2 | 2017-11-10 22:05:12 -0700 | [diff] [blame] | 1492 | * request head insertion from the workqueue. |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1493 | */ |
Christoph Hellwig | e806402 | 2016-10-20 15:12:13 +0200 | [diff] [blame] | 1494 | BUG_ON(rq->rq_flags & RQF_SOFTBARRIER); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1495 | |
| 1496 | spin_lock_irqsave(&q->requeue_lock, flags); |
| 1497 | if (at_head) { |
Christoph Hellwig | e806402 | 2016-10-20 15:12:13 +0200 | [diff] [blame] | 1498 | rq->rq_flags |= RQF_SOFTBARRIER; |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1499 | list_add(&rq->queuelist, &q->requeue_list); |
| 1500 | } else { |
| 1501 | list_add_tail(&rq->queuelist, &q->requeue_list); |
| 1502 | } |
| 1503 | spin_unlock_irqrestore(&q->requeue_lock, flags); |
Bart Van Assche | 2b053ac | 2016-10-28 17:21:41 -0700 | [diff] [blame] | 1504 | |
| 1505 | if (kick_requeue_list) |
| 1506 | blk_mq_kick_requeue_list(q); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1507 | } |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1508 | |
| 1509 | void blk_mq_kick_requeue_list(struct request_queue *q) |
| 1510 | { |
Bart Van Assche | ae943d2 | 2018-01-19 08:58:55 -0800 | [diff] [blame] | 1511 | kblockd_mod_delayed_work_on(WORK_CPU_UNBOUND, &q->requeue_work, 0); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 1512 | } |
| 1513 | EXPORT_SYMBOL(blk_mq_kick_requeue_list); |
| 1514 | |
Mike Snitzer | 2849450 | 2016-09-14 13:28:30 -0400 | [diff] [blame] | 1515 | void blk_mq_delay_kick_requeue_list(struct request_queue *q, |
| 1516 | unsigned long msecs) |
| 1517 | { |
Bart Van Assche | d4acf36 | 2017-08-09 11:28:06 -0700 | [diff] [blame] | 1518 | kblockd_mod_delayed_work_on(WORK_CPU_UNBOUND, &q->requeue_work, |
| 1519 | msecs_to_jiffies(msecs)); |
Mike Snitzer | 2849450 | 2016-09-14 13:28:30 -0400 | [diff] [blame] | 1520 | } |
| 1521 | EXPORT_SYMBOL(blk_mq_delay_kick_requeue_list); |
| 1522 | |
John Garry | 2dd6532 | 2022-07-06 20:03:53 +0800 | [diff] [blame] | 1523 | static bool blk_mq_rq_inflight(struct request *rq, void *priv) |
Jens Axboe | ae87991 | 2018-11-08 09:03:51 -0700 | [diff] [blame] | 1524 | { |
| 1525 | /* |
John Garry | 8ab30a3 | 2021-12-06 20:49:48 +0800 | [diff] [blame] | 1526 | * If we find a request that isn't idle we know the queue is busy |
| 1527 | * as it's checked in the iter. |
| 1528 | * Return false to stop the iteration. |
Jens Axboe | ae87991 | 2018-11-08 09:03:51 -0700 | [diff] [blame] | 1529 | */ |
John Garry | 8ab30a3 | 2021-12-06 20:49:48 +0800 | [diff] [blame] | 1530 | if (blk_mq_request_started(rq)) { |
Jens Axboe | ae87991 | 2018-11-08 09:03:51 -0700 | [diff] [blame] | 1531 | bool *busy = priv; |
| 1532 | |
| 1533 | *busy = true; |
| 1534 | return false; |
| 1535 | } |
| 1536 | |
| 1537 | return true; |
| 1538 | } |
| 1539 | |
Jens Axboe | 3c94d83 | 2018-12-17 21:11:17 -0700 | [diff] [blame] | 1540 | bool blk_mq_queue_inflight(struct request_queue *q) |
Jens Axboe | ae87991 | 2018-11-08 09:03:51 -0700 | [diff] [blame] | 1541 | { |
| 1542 | bool busy = false; |
| 1543 | |
Jens Axboe | 3c94d83 | 2018-12-17 21:11:17 -0700 | [diff] [blame] | 1544 | blk_mq_queue_tag_busy_iter(q, blk_mq_rq_inflight, &busy); |
Jens Axboe | ae87991 | 2018-11-08 09:03:51 -0700 | [diff] [blame] | 1545 | return busy; |
| 1546 | } |
Jens Axboe | 3c94d83 | 2018-12-17 21:11:17 -0700 | [diff] [blame] | 1547 | EXPORT_SYMBOL_GPL(blk_mq_queue_inflight); |
Jens Axboe | ae87991 | 2018-11-08 09:03:51 -0700 | [diff] [blame] | 1548 | |
John Garry | 9bdb483 | 2022-07-06 20:03:51 +0800 | [diff] [blame] | 1549 | static void blk_mq_rq_timed_out(struct request *req) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1550 | { |
Christoph Hellwig | da66126 | 2018-06-14 13:58:45 +0200 | [diff] [blame] | 1551 | req->rq_flags |= RQF_TIMED_OUT; |
Christoph Hellwig | d1210d5 | 2018-05-29 15:52:39 +0200 | [diff] [blame] | 1552 | if (req->q->mq_ops->timeout) { |
| 1553 | enum blk_eh_timer_return ret; |
Jens Axboe | 87ee7b1 | 2014-04-24 08:51:47 -0600 | [diff] [blame] | 1554 | |
John Garry | 9bdb483 | 2022-07-06 20:03:51 +0800 | [diff] [blame] | 1555 | ret = req->q->mq_ops->timeout(req); |
Christoph Hellwig | d1210d5 | 2018-05-29 15:52:39 +0200 | [diff] [blame] | 1556 | if (ret == BLK_EH_DONE) |
| 1557 | return; |
| 1558 | WARN_ON_ONCE(ret != BLK_EH_RESET_TIMER); |
Christoph Hellwig | 46f92d4 | 2014-09-13 16:40:12 -0700 | [diff] [blame] | 1559 | } |
Christoph Hellwig | d1210d5 | 2018-05-29 15:52:39 +0200 | [diff] [blame] | 1560 | |
| 1561 | blk_add_timer(req); |
Jens Axboe | 87ee7b1 | 2014-04-24 08:51:47 -0600 | [diff] [blame] | 1562 | } |
Keith Busch | 5b3f25f | 2015-01-07 18:55:46 -0700 | [diff] [blame] | 1563 | |
David Jeffery | 82c2294 | 2022-10-26 13:19:57 +0800 | [diff] [blame] | 1564 | struct blk_expired_data { |
| 1565 | bool has_timedout_rq; |
| 1566 | unsigned long next; |
| 1567 | unsigned long timeout_start; |
| 1568 | }; |
| 1569 | |
| 1570 | static bool blk_mq_req_expired(struct request *rq, struct blk_expired_data *expired) |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 1571 | { |
| 1572 | unsigned long deadline; |
| 1573 | |
| 1574 | if (blk_mq_rq_state(rq) != MQ_RQ_IN_FLIGHT) |
| 1575 | return false; |
Christoph Hellwig | da66126 | 2018-06-14 13:58:45 +0200 | [diff] [blame] | 1576 | if (rq->rq_flags & RQF_TIMED_OUT) |
| 1577 | return false; |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 1578 | |
Christoph Hellwig | 079076b | 2018-11-14 17:02:05 +0100 | [diff] [blame] | 1579 | deadline = READ_ONCE(rq->deadline); |
David Jeffery | 82c2294 | 2022-10-26 13:19:57 +0800 | [diff] [blame] | 1580 | if (time_after_eq(expired->timeout_start, deadline)) |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 1581 | return true; |
| 1582 | |
David Jeffery | 82c2294 | 2022-10-26 13:19:57 +0800 | [diff] [blame] | 1583 | if (expired->next == 0) |
| 1584 | expired->next = deadline; |
| 1585 | else if (time_after(expired->next, deadline)) |
| 1586 | expired->next = deadline; |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 1587 | return false; |
| 1588 | } |
| 1589 | |
Ming Lei | 2e315dc | 2021-05-11 23:22:34 +0800 | [diff] [blame] | 1590 | void blk_mq_put_rq_ref(struct request *rq) |
| 1591 | { |
Jens Axboe | de671d6 | 2022-09-21 15:19:54 -0600 | [diff] [blame] | 1592 | if (is_flush_rq(rq)) { |
| 1593 | if (rq->end_io(rq, 0) == RQ_END_IO_FREE) |
| 1594 | blk_mq_free_request(rq); |
| 1595 | } else if (req_ref_put_and_test(rq)) { |
Ming Lei | 2e315dc | 2021-05-11 23:22:34 +0800 | [diff] [blame] | 1596 | __blk_mq_free_request(rq); |
Jens Axboe | de671d6 | 2022-09-21 15:19:54 -0600 | [diff] [blame] | 1597 | } |
Ming Lei | 2e315dc | 2021-05-11 23:22:34 +0800 | [diff] [blame] | 1598 | } |
| 1599 | |
John Garry | 2dd6532 | 2022-07-06 20:03:53 +0800 | [diff] [blame] | 1600 | static bool blk_mq_check_expired(struct request *rq, void *priv) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1601 | { |
David Jeffery | 82c2294 | 2022-10-26 13:19:57 +0800 | [diff] [blame] | 1602 | struct blk_expired_data *expired = priv; |
Christoph Hellwig | 81481eb | 2014-09-13 16:40:11 -0700 | [diff] [blame] | 1603 | |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 1604 | /* |
Ming Lei | c797b40 | 2021-08-11 23:52:02 +0800 | [diff] [blame] | 1605 | * blk_mq_queue_tag_busy_iter() has locked the request, so it cannot |
| 1606 | * be reallocated underneath the timeout handler's processing, then |
| 1607 | * the expire check is reliable. If the request is not expired, then |
| 1608 | * it was completed and reallocated as a new request after returning |
| 1609 | * from blk_mq_check_expired(). |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 1610 | */ |
David Jeffery | 82c2294 | 2022-10-26 13:19:57 +0800 | [diff] [blame] | 1611 | if (blk_mq_req_expired(rq, expired)) { |
| 1612 | expired->has_timedout_rq = true; |
| 1613 | return false; |
| 1614 | } |
| 1615 | return true; |
| 1616 | } |
| 1617 | |
| 1618 | static bool blk_mq_handle_expired(struct request *rq, void *priv) |
| 1619 | { |
| 1620 | struct blk_expired_data *expired = priv; |
| 1621 | |
| 1622 | if (blk_mq_req_expired(rq, expired)) |
John Garry | 9bdb483 | 2022-07-06 20:03:51 +0800 | [diff] [blame] | 1623 | blk_mq_rq_timed_out(rq); |
Jens Axboe | 7baa857 | 2018-11-08 10:24:07 -0700 | [diff] [blame] | 1624 | return true; |
Tejun Heo | 1d9bd51 | 2018-01-09 08:29:48 -0800 | [diff] [blame] | 1625 | } |
| 1626 | |
Christoph Hellwig | 287922e | 2015-10-30 20:57:30 +0800 | [diff] [blame] | 1627 | static void blk_mq_timeout_work(struct work_struct *work) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1628 | { |
Christoph Hellwig | 287922e | 2015-10-30 20:57:30 +0800 | [diff] [blame] | 1629 | struct request_queue *q = |
| 1630 | container_of(work, struct request_queue, timeout_work); |
David Jeffery | 82c2294 | 2022-10-26 13:19:57 +0800 | [diff] [blame] | 1631 | struct blk_expired_data expired = { |
| 1632 | .timeout_start = jiffies, |
| 1633 | }; |
Tejun Heo | 1d9bd51 | 2018-01-09 08:29:48 -0800 | [diff] [blame] | 1634 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 1635 | unsigned long i; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1636 | |
Gabriel Krisman Bertazi | 71f79fb | 2016-08-01 08:23:39 -0600 | [diff] [blame] | 1637 | /* A deadlock might occur if a request is stuck requiring a |
| 1638 | * timeout at the same time a queue freeze is waiting |
| 1639 | * completion, since the timeout code would not be able to |
| 1640 | * acquire the queue reference here. |
| 1641 | * |
| 1642 | * That's why we don't use blk_queue_enter here; instead, we use |
| 1643 | * percpu_ref_tryget directly, because we need to be able to |
| 1644 | * obtain a reference even in the short window between the queue |
| 1645 | * starting to freeze, by dropping the first reference in |
Ming Lei | 1671d52 | 2017-03-27 20:06:57 +0800 | [diff] [blame] | 1646 | * blk_freeze_queue_start, and the moment the last request is |
Gabriel Krisman Bertazi | 71f79fb | 2016-08-01 08:23:39 -0600 | [diff] [blame] | 1647 | * consumed, marked by the instant q_usage_counter reaches |
| 1648 | * zero. |
| 1649 | */ |
| 1650 | if (!percpu_ref_tryget(&q->q_usage_counter)) |
Christoph Hellwig | 287922e | 2015-10-30 20:57:30 +0800 | [diff] [blame] | 1651 | return; |
| 1652 | |
David Jeffery | 82c2294 | 2022-10-26 13:19:57 +0800 | [diff] [blame] | 1653 | /* check if there is any timed-out request */ |
| 1654 | blk_mq_queue_tag_busy_iter(q, blk_mq_check_expired, &expired); |
| 1655 | if (expired.has_timedout_rq) { |
| 1656 | /* |
| 1657 | * Before walking tags, we must ensure any submit started |
| 1658 | * before the current time has finished. Since the submit |
| 1659 | * uses srcu or rcu, wait for a synchronization point to |
| 1660 | * ensure all running submits have finished |
| 1661 | */ |
Christoph Hellwig | 483239c | 2022-11-01 16:00:48 +0100 | [diff] [blame] | 1662 | blk_mq_wait_quiesce_done(q->tag_set); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1663 | |
David Jeffery | 82c2294 | 2022-10-26 13:19:57 +0800 | [diff] [blame] | 1664 | expired.next = 0; |
| 1665 | blk_mq_queue_tag_busy_iter(q, blk_mq_handle_expired, &expired); |
| 1666 | } |
| 1667 | |
| 1668 | if (expired.next != 0) { |
| 1669 | mod_timer(&q->timeout, expired.next); |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 1670 | } else { |
Bart Van Assche | fcd36c3 | 2018-01-10 08:33:33 -0800 | [diff] [blame] | 1671 | /* |
| 1672 | * Request timeouts are handled as a forward rolling timer. If |
| 1673 | * we end up here it means that no requests are pending and |
| 1674 | * also that no request has been pending for a while. Mark |
| 1675 | * each hctx as idle. |
| 1676 | */ |
Ming Lei | f054b56 | 2015-04-21 10:00:19 +0800 | [diff] [blame] | 1677 | queue_for_each_hw_ctx(q, hctx, i) { |
| 1678 | /* the hctx may be unmapped, so check it here */ |
| 1679 | if (blk_mq_hw_queue_mapped(hctx)) |
| 1680 | blk_mq_tag_idle(hctx); |
| 1681 | } |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 1682 | } |
Christoph Hellwig | 287922e | 2015-10-30 20:57:30 +0800 | [diff] [blame] | 1683 | blk_queue_exit(q); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1684 | } |
| 1685 | |
Omar Sandoval | 8845964 | 2016-09-17 08:38:44 -0600 | [diff] [blame] | 1686 | struct flush_busy_ctx_data { |
| 1687 | struct blk_mq_hw_ctx *hctx; |
| 1688 | struct list_head *list; |
| 1689 | }; |
| 1690 | |
| 1691 | static bool flush_busy_ctx(struct sbitmap *sb, unsigned int bitnr, void *data) |
| 1692 | { |
| 1693 | struct flush_busy_ctx_data *flush_data = data; |
| 1694 | struct blk_mq_hw_ctx *hctx = flush_data->hctx; |
| 1695 | struct blk_mq_ctx *ctx = hctx->ctxs[bitnr]; |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 1696 | enum hctx_type type = hctx->type; |
Omar Sandoval | 8845964 | 2016-09-17 08:38:44 -0600 | [diff] [blame] | 1697 | |
Omar Sandoval | 8845964 | 2016-09-17 08:38:44 -0600 | [diff] [blame] | 1698 | spin_lock(&ctx->lock); |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 1699 | list_splice_tail_init(&ctx->rq_lists[type], flush_data->list); |
Omar Sandoval | e9a99a6 | 2018-02-27 16:56:42 -0800 | [diff] [blame] | 1700 | sbitmap_clear_bit(sb, bitnr); |
Omar Sandoval | 8845964 | 2016-09-17 08:38:44 -0600 | [diff] [blame] | 1701 | spin_unlock(&ctx->lock); |
| 1702 | return true; |
| 1703 | } |
| 1704 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1705 | /* |
Jens Axboe | 1429d7c | 2014-05-19 09:23:55 -0600 | [diff] [blame] | 1706 | * Process software queues that have been marked busy, splicing them |
| 1707 | * to the for-dispatch |
| 1708 | */ |
Jens Axboe | 2c3ad66 | 2016-12-14 14:34:47 -0700 | [diff] [blame] | 1709 | void blk_mq_flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list) |
Jens Axboe | 1429d7c | 2014-05-19 09:23:55 -0600 | [diff] [blame] | 1710 | { |
Omar Sandoval | 8845964 | 2016-09-17 08:38:44 -0600 | [diff] [blame] | 1711 | struct flush_busy_ctx_data data = { |
| 1712 | .hctx = hctx, |
| 1713 | .list = list, |
| 1714 | }; |
Jens Axboe | 1429d7c | 2014-05-19 09:23:55 -0600 | [diff] [blame] | 1715 | |
Omar Sandoval | 8845964 | 2016-09-17 08:38:44 -0600 | [diff] [blame] | 1716 | sbitmap_for_each_set(&hctx->ctx_map, flush_busy_ctx, &data); |
Jens Axboe | 1429d7c | 2014-05-19 09:23:55 -0600 | [diff] [blame] | 1717 | } |
Jens Axboe | 2c3ad66 | 2016-12-14 14:34:47 -0700 | [diff] [blame] | 1718 | EXPORT_SYMBOL_GPL(blk_mq_flush_busy_ctxs); |
Jens Axboe | 1429d7c | 2014-05-19 09:23:55 -0600 | [diff] [blame] | 1719 | |
Ming Lei | b347689 | 2017-10-14 17:22:30 +0800 | [diff] [blame] | 1720 | struct dispatch_rq_data { |
| 1721 | struct blk_mq_hw_ctx *hctx; |
| 1722 | struct request *rq; |
| 1723 | }; |
| 1724 | |
| 1725 | static bool dispatch_rq_from_ctx(struct sbitmap *sb, unsigned int bitnr, |
| 1726 | void *data) |
| 1727 | { |
| 1728 | struct dispatch_rq_data *dispatch_data = data; |
| 1729 | struct blk_mq_hw_ctx *hctx = dispatch_data->hctx; |
| 1730 | struct blk_mq_ctx *ctx = hctx->ctxs[bitnr]; |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 1731 | enum hctx_type type = hctx->type; |
Ming Lei | b347689 | 2017-10-14 17:22:30 +0800 | [diff] [blame] | 1732 | |
| 1733 | spin_lock(&ctx->lock); |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 1734 | if (!list_empty(&ctx->rq_lists[type])) { |
| 1735 | dispatch_data->rq = list_entry_rq(ctx->rq_lists[type].next); |
Ming Lei | b347689 | 2017-10-14 17:22:30 +0800 | [diff] [blame] | 1736 | list_del_init(&dispatch_data->rq->queuelist); |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 1737 | if (list_empty(&ctx->rq_lists[type])) |
Ming Lei | b347689 | 2017-10-14 17:22:30 +0800 | [diff] [blame] | 1738 | sbitmap_clear_bit(sb, bitnr); |
| 1739 | } |
| 1740 | spin_unlock(&ctx->lock); |
| 1741 | |
| 1742 | return !dispatch_data->rq; |
| 1743 | } |
| 1744 | |
| 1745 | struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx, |
| 1746 | struct blk_mq_ctx *start) |
| 1747 | { |
Jens Axboe | f31967f | 2018-10-29 13:13:29 -0600 | [diff] [blame] | 1748 | unsigned off = start ? start->index_hw[hctx->type] : 0; |
Ming Lei | b347689 | 2017-10-14 17:22:30 +0800 | [diff] [blame] | 1749 | struct dispatch_rq_data data = { |
| 1750 | .hctx = hctx, |
| 1751 | .rq = NULL, |
| 1752 | }; |
| 1753 | |
| 1754 | __sbitmap_for_each_set(&hctx->ctx_map, off, |
| 1755 | dispatch_rq_from_ctx, &data); |
| 1756 | |
| 1757 | return data.rq; |
| 1758 | } |
| 1759 | |
Jens Axboe | a808a9d | 2021-10-13 08:28:14 -0600 | [diff] [blame] | 1760 | static bool __blk_mq_alloc_driver_tag(struct request *rq) |
Jens Axboe | 703fd1c | 2016-09-16 13:59:14 -0600 | [diff] [blame] | 1761 | { |
John Garry | ae0f1a7 | 2021-10-05 18:23:38 +0800 | [diff] [blame] | 1762 | struct sbitmap_queue *bt = &rq->mq_hctx->tags->bitmap_tags; |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1763 | unsigned int tag_offset = rq->mq_hctx->tags->nr_reserved_tags; |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1764 | int tag; |
| 1765 | |
Ming Lei | 568f270 | 2020-07-06 22:41:11 +0800 | [diff] [blame] | 1766 | blk_mq_tag_busy(rq->mq_hctx); |
| 1767 | |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1768 | if (blk_mq_tag_is_reserved(rq->mq_hctx->sched_tags, rq->internal_tag)) { |
John Garry | ae0f1a7 | 2021-10-05 18:23:38 +0800 | [diff] [blame] | 1769 | bt = &rq->mq_hctx->tags->breserved_tags; |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1770 | tag_offset = 0; |
Ming Lei | 2850085 | 2020-09-11 18:41:14 +0800 | [diff] [blame] | 1771 | } else { |
| 1772 | if (!hctx_may_queue(rq->mq_hctx, bt)) |
| 1773 | return false; |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1774 | } |
| 1775 | |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1776 | tag = __sbitmap_queue_get(bt); |
| 1777 | if (tag == BLK_MQ_NO_TAG) |
| 1778 | return false; |
| 1779 | |
| 1780 | rq->tag = tag + tag_offset; |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1781 | return true; |
| 1782 | } |
| 1783 | |
Jens Axboe | a808a9d | 2021-10-13 08:28:14 -0600 | [diff] [blame] | 1784 | bool __blk_mq_get_driver_tag(struct blk_mq_hw_ctx *hctx, struct request *rq) |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1785 | { |
Jens Axboe | a808a9d | 2021-10-13 08:28:14 -0600 | [diff] [blame] | 1786 | if (rq->tag == BLK_MQ_NO_TAG && !__blk_mq_alloc_driver_tag(rq)) |
Ming Lei | 568f270 | 2020-07-06 22:41:11 +0800 | [diff] [blame] | 1787 | return false; |
| 1788 | |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 1789 | if ((hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED) && |
Ming Lei | 568f270 | 2020-07-06 22:41:11 +0800 | [diff] [blame] | 1790 | !(rq->rq_flags & RQF_MQ_INFLIGHT)) { |
| 1791 | rq->rq_flags |= RQF_MQ_INFLIGHT; |
John Garry | bccf5e2 | 2020-08-19 23:20:26 +0800 | [diff] [blame] | 1792 | __blk_mq_inc_active_requests(hctx); |
Ming Lei | 568f270 | 2020-07-06 22:41:11 +0800 | [diff] [blame] | 1793 | } |
| 1794 | hctx->tags->rqs[rq->tag] = rq; |
| 1795 | return true; |
Ming Lei | 570e9b7 | 2020-06-30 22:03:55 +0800 | [diff] [blame] | 1796 | } |
| 1797 | |
Jens Axboe | eb619fd | 2017-11-09 08:32:43 -0700 | [diff] [blame] | 1798 | static int blk_mq_dispatch_wake(wait_queue_entry_t *wait, unsigned mode, |
| 1799 | int flags, void *key) |
Omar Sandoval | da55f2c | 2017-02-22 10:58:29 -0800 | [diff] [blame] | 1800 | { |
| 1801 | struct blk_mq_hw_ctx *hctx; |
| 1802 | |
| 1803 | hctx = container_of(wait, struct blk_mq_hw_ctx, dispatch_wait); |
| 1804 | |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 1805 | spin_lock(&hctx->dispatch_wait_lock); |
Jens Axboe | e861857 | 2019-03-25 12:34:10 -0600 | [diff] [blame] | 1806 | if (!list_empty(&wait->entry)) { |
| 1807 | struct sbitmap_queue *sbq; |
| 1808 | |
| 1809 | list_del_init(&wait->entry); |
John Garry | ae0f1a7 | 2021-10-05 18:23:38 +0800 | [diff] [blame] | 1810 | sbq = &hctx->tags->bitmap_tags; |
Jens Axboe | e861857 | 2019-03-25 12:34:10 -0600 | [diff] [blame] | 1811 | atomic_dec(&sbq->ws_active); |
| 1812 | } |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 1813 | spin_unlock(&hctx->dispatch_wait_lock); |
| 1814 | |
Omar Sandoval | da55f2c | 2017-02-22 10:58:29 -0800 | [diff] [blame] | 1815 | blk_mq_run_hw_queue(hctx, true); |
| 1816 | return 1; |
| 1817 | } |
| 1818 | |
Jens Axboe | f906a6a | 2017-11-09 16:10:13 -0700 | [diff] [blame] | 1819 | /* |
| 1820 | * Mark us waiting for a tag. For shared tags, this involves hooking us into |
Bart Van Assche | ee3e4de | 2018-01-09 10:09:15 -0800 | [diff] [blame] | 1821 | * the tag wakeups. For non-shared tags, we can simply mark us needing a |
| 1822 | * restart. For both cases, take care to check the condition again after |
Jens Axboe | f906a6a | 2017-11-09 16:10:13 -0700 | [diff] [blame] | 1823 | * marking us as waiting. |
| 1824 | */ |
Ming Lei | 2278d69 | 2018-06-25 19:31:46 +0800 | [diff] [blame] | 1825 | static bool blk_mq_mark_tag_wait(struct blk_mq_hw_ctx *hctx, |
Jens Axboe | f906a6a | 2017-11-09 16:10:13 -0700 | [diff] [blame] | 1826 | struct request *rq) |
Omar Sandoval | da55f2c | 2017-02-22 10:58:29 -0800 | [diff] [blame] | 1827 | { |
John Garry | ae0f1a7 | 2021-10-05 18:23:38 +0800 | [diff] [blame] | 1828 | struct sbitmap_queue *sbq = &hctx->tags->bitmap_tags; |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 1829 | struct wait_queue_head *wq; |
Jens Axboe | f906a6a | 2017-11-09 16:10:13 -0700 | [diff] [blame] | 1830 | wait_queue_entry_t *wait; |
| 1831 | bool ret; |
Omar Sandoval | da55f2c | 2017-02-22 10:58:29 -0800 | [diff] [blame] | 1832 | |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 1833 | if (!(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) { |
Yufen Yu | 684b732 | 2019-03-15 11:05:10 +0800 | [diff] [blame] | 1834 | blk_mq_sched_mark_restart_hctx(hctx); |
Omar Sandoval | da55f2c | 2017-02-22 10:58:29 -0800 | [diff] [blame] | 1835 | |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1836 | /* |
| 1837 | * It's possible that a tag was freed in the window between the |
| 1838 | * allocation failure and adding the hardware queue to the wait |
| 1839 | * queue. |
| 1840 | * |
| 1841 | * Don't clear RESTART here, someone else could have set it. |
| 1842 | * At most this will cost an extra queue run. |
| 1843 | */ |
Ming Lei | 8ab6bb9e | 2018-06-25 19:31:45 +0800 | [diff] [blame] | 1844 | return blk_mq_get_driver_tag(rq); |
Jens Axboe | eb619fd | 2017-11-09 08:32:43 -0700 | [diff] [blame] | 1845 | } |
| 1846 | |
Ming Lei | 2278d69 | 2018-06-25 19:31:46 +0800 | [diff] [blame] | 1847 | wait = &hctx->dispatch_wait; |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1848 | if (!list_empty_careful(&wait->entry)) |
| 1849 | return false; |
| 1850 | |
Jens Axboe | e861857 | 2019-03-25 12:34:10 -0600 | [diff] [blame] | 1851 | wq = &bt_wait_ptr(sbq, hctx)->wait; |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 1852 | |
| 1853 | spin_lock_irq(&wq->lock); |
| 1854 | spin_lock(&hctx->dispatch_wait_lock); |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1855 | if (!list_empty(&wait->entry)) { |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 1856 | spin_unlock(&hctx->dispatch_wait_lock); |
| 1857 | spin_unlock_irq(&wq->lock); |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1858 | return false; |
| 1859 | } |
| 1860 | |
Jens Axboe | e861857 | 2019-03-25 12:34:10 -0600 | [diff] [blame] | 1861 | atomic_inc(&sbq->ws_active); |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 1862 | wait->flags &= ~WQ_FLAG_EXCLUSIVE; |
| 1863 | __add_wait_queue(wq, wait); |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1864 | |
Omar Sandoval | da55f2c | 2017-02-22 10:58:29 -0800 | [diff] [blame] | 1865 | /* |
Jens Axboe | eb619fd | 2017-11-09 08:32:43 -0700 | [diff] [blame] | 1866 | * It's possible that a tag was freed in the window between the |
| 1867 | * allocation failure and adding the hardware queue to the wait |
| 1868 | * queue. |
Omar Sandoval | da55f2c | 2017-02-22 10:58:29 -0800 | [diff] [blame] | 1869 | */ |
Ming Lei | 8ab6bb9e | 2018-06-25 19:31:45 +0800 | [diff] [blame] | 1870 | ret = blk_mq_get_driver_tag(rq); |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1871 | if (!ret) { |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 1872 | spin_unlock(&hctx->dispatch_wait_lock); |
| 1873 | spin_unlock_irq(&wq->lock); |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1874 | return false; |
Jens Axboe | f906a6a | 2017-11-09 16:10:13 -0700 | [diff] [blame] | 1875 | } |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1876 | |
| 1877 | /* |
| 1878 | * We got a tag, remove ourselves from the wait queue to ensure |
| 1879 | * someone else gets the wakeup. |
| 1880 | */ |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1881 | list_del_init(&wait->entry); |
Jens Axboe | e861857 | 2019-03-25 12:34:10 -0600 | [diff] [blame] | 1882 | atomic_dec(&sbq->ws_active); |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 1883 | spin_unlock(&hctx->dispatch_wait_lock); |
| 1884 | spin_unlock_irq(&wq->lock); |
Bart Van Assche | c27d53f | 2018-01-10 13:41:21 -0800 | [diff] [blame] | 1885 | |
| 1886 | return true; |
Omar Sandoval | da55f2c | 2017-02-22 10:58:29 -0800 | [diff] [blame] | 1887 | } |
| 1888 | |
Ming Lei | 6e768717 | 2018-07-03 09:03:16 -0600 | [diff] [blame] | 1889 | #define BLK_MQ_DISPATCH_BUSY_EWMA_WEIGHT 8 |
| 1890 | #define BLK_MQ_DISPATCH_BUSY_EWMA_FACTOR 4 |
| 1891 | /* |
| 1892 | * Update dispatch busy with the Exponential Weighted Moving Average(EWMA): |
| 1893 | * - EWMA is one simple way to compute running average value |
| 1894 | * - weight(7/8 and 1/8) is applied so that it can decrease exponentially |
| 1895 | * - take 4 as factor for avoiding to get too small(0) result, and this |
| 1896 | * factor doesn't matter because EWMA decreases exponentially |
| 1897 | */ |
| 1898 | static void blk_mq_update_dispatch_busy(struct blk_mq_hw_ctx *hctx, bool busy) |
| 1899 | { |
| 1900 | unsigned int ewma; |
| 1901 | |
Ming Lei | 6e768717 | 2018-07-03 09:03:16 -0600 | [diff] [blame] | 1902 | ewma = hctx->dispatch_busy; |
| 1903 | |
| 1904 | if (!ewma && !busy) |
| 1905 | return; |
| 1906 | |
| 1907 | ewma *= BLK_MQ_DISPATCH_BUSY_EWMA_WEIGHT - 1; |
| 1908 | if (busy) |
| 1909 | ewma += 1 << BLK_MQ_DISPATCH_BUSY_EWMA_FACTOR; |
| 1910 | ewma /= BLK_MQ_DISPATCH_BUSY_EWMA_WEIGHT; |
| 1911 | |
| 1912 | hctx->dispatch_busy = ewma; |
| 1913 | } |
| 1914 | |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 1915 | #define BLK_MQ_RESOURCE_DELAY 3 /* ms units */ |
| 1916 | |
Johannes Thumshirn | c92a410 | 2020-03-25 00:24:44 +0900 | [diff] [blame] | 1917 | static void blk_mq_handle_dev_resource(struct request *rq, |
| 1918 | struct list_head *list) |
| 1919 | { |
| 1920 | struct request *next = |
| 1921 | list_first_entry_or_null(list, struct request, queuelist); |
| 1922 | |
| 1923 | /* |
| 1924 | * If an I/O scheduler has been configured and we got a driver tag for |
| 1925 | * the next request already, free it. |
| 1926 | */ |
| 1927 | if (next) |
| 1928 | blk_mq_put_driver_tag(next); |
| 1929 | |
| 1930 | list_add(&rq->queuelist, list); |
| 1931 | __blk_mq_requeue_request(rq); |
| 1932 | } |
| 1933 | |
Keith Busch | 0512a75 | 2020-05-12 17:55:47 +0900 | [diff] [blame] | 1934 | static void blk_mq_handle_zone_resource(struct request *rq, |
| 1935 | struct list_head *zone_list) |
| 1936 | { |
| 1937 | /* |
| 1938 | * If we end up here it is because we cannot dispatch a request to a |
| 1939 | * specific zone due to LLD level zone-write locking or other zone |
| 1940 | * related resource not being available. In this case, set the request |
| 1941 | * aside in zone_list for retrying it later. |
| 1942 | */ |
| 1943 | list_add(&rq->queuelist, zone_list); |
| 1944 | __blk_mq_requeue_request(rq); |
| 1945 | } |
| 1946 | |
Ming Lei | 7538352 | 2020-06-30 18:24:58 +0800 | [diff] [blame] | 1947 | enum prep_dispatch { |
| 1948 | PREP_DISPATCH_OK, |
| 1949 | PREP_DISPATCH_NO_TAG, |
| 1950 | PREP_DISPATCH_NO_BUDGET, |
| 1951 | }; |
| 1952 | |
| 1953 | static enum prep_dispatch blk_mq_prep_dispatch_rq(struct request *rq, |
| 1954 | bool need_budget) |
| 1955 | { |
| 1956 | struct blk_mq_hw_ctx *hctx = rq->mq_hctx; |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 1957 | int budget_token = -1; |
Ming Lei | 7538352 | 2020-06-30 18:24:58 +0800 | [diff] [blame] | 1958 | |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 1959 | if (need_budget) { |
| 1960 | budget_token = blk_mq_get_dispatch_budget(rq->q); |
| 1961 | if (budget_token < 0) { |
| 1962 | blk_mq_put_driver_tag(rq); |
| 1963 | return PREP_DISPATCH_NO_BUDGET; |
| 1964 | } |
| 1965 | blk_mq_set_rq_budget_token(rq, budget_token); |
Ming Lei | 7538352 | 2020-06-30 18:24:58 +0800 | [diff] [blame] | 1966 | } |
| 1967 | |
| 1968 | if (!blk_mq_get_driver_tag(rq)) { |
| 1969 | /* |
| 1970 | * The initial allocation attempt failed, so we need to |
| 1971 | * rerun the hardware queue when a tag is freed. The |
| 1972 | * waitqueue takes care of that. If the queue is run |
| 1973 | * before we add this entry back on the dispatch list, |
| 1974 | * we'll re-run it below. |
| 1975 | */ |
| 1976 | if (!blk_mq_mark_tag_wait(hctx, rq)) { |
Ming Lei | 1fd40b5 | 2020-06-30 18:25:00 +0800 | [diff] [blame] | 1977 | /* |
| 1978 | * All budgets not got from this function will be put |
| 1979 | * together during handling partial dispatch |
| 1980 | */ |
| 1981 | if (need_budget) |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 1982 | blk_mq_put_dispatch_budget(rq->q, budget_token); |
Ming Lei | 7538352 | 2020-06-30 18:24:58 +0800 | [diff] [blame] | 1983 | return PREP_DISPATCH_NO_TAG; |
| 1984 | } |
| 1985 | } |
| 1986 | |
| 1987 | return PREP_DISPATCH_OK; |
| 1988 | } |
| 1989 | |
Ming Lei | 1fd40b5 | 2020-06-30 18:25:00 +0800 | [diff] [blame] | 1990 | /* release all allocated budgets before calling to blk_mq_dispatch_rq_list */ |
| 1991 | static void blk_mq_release_budgets(struct request_queue *q, |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 1992 | struct list_head *list) |
Ming Lei | 1fd40b5 | 2020-06-30 18:25:00 +0800 | [diff] [blame] | 1993 | { |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 1994 | struct request *rq; |
Ming Lei | 1fd40b5 | 2020-06-30 18:25:00 +0800 | [diff] [blame] | 1995 | |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 1996 | list_for_each_entry(rq, list, queuelist) { |
| 1997 | int budget_token = blk_mq_get_rq_budget_token(rq); |
| 1998 | |
| 1999 | if (budget_token >= 0) |
| 2000 | blk_mq_put_dispatch_budget(q, budget_token); |
| 2001 | } |
Ming Lei | 1fd40b5 | 2020-06-30 18:25:00 +0800 | [diff] [blame] | 2002 | } |
| 2003 | |
Jens Axboe | 1f57f8d | 2018-06-28 11:54:01 -0600 | [diff] [blame] | 2004 | /* |
| 2005 | * Returns true if we did some work AND can potentially do more. |
| 2006 | */ |
Ming Lei | 445874e | 2020-06-30 18:24:57 +0800 | [diff] [blame] | 2007 | bool blk_mq_dispatch_rq_list(struct blk_mq_hw_ctx *hctx, struct list_head *list, |
Ming Lei | 1fd40b5 | 2020-06-30 18:25:00 +0800 | [diff] [blame] | 2008 | unsigned int nr_budgets) |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2009 | { |
Ming Lei | 7538352 | 2020-06-30 18:24:58 +0800 | [diff] [blame] | 2010 | enum prep_dispatch prep; |
Ming Lei | 445874e | 2020-06-30 18:24:57 +0800 | [diff] [blame] | 2011 | struct request_queue *q = hctx->queue; |
Jianchao Wang | 6d6f167c | 2017-11-02 23:24:32 +0800 | [diff] [blame] | 2012 | struct request *rq, *nxt; |
Christoph Hellwig | fc17b65 | 2017-06-03 09:38:05 +0200 | [diff] [blame] | 2013 | int errors, queued; |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 2014 | blk_status_t ret = BLK_STS_OK; |
Keith Busch | 0512a75 | 2020-05-12 17:55:47 +0900 | [diff] [blame] | 2015 | LIST_HEAD(zone_list); |
Naohiro Aota | 9586e67 | 2021-10-27 01:51:27 +0900 | [diff] [blame] | 2016 | bool needs_resource = false; |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2017 | |
Omar Sandoval | 81380ca | 2017-04-07 08:56:26 -0600 | [diff] [blame] | 2018 | if (list_empty(list)) |
| 2019 | return false; |
| 2020 | |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2021 | /* |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2022 | * Now process all the entries, sending them to the driver. |
| 2023 | */ |
Jens Axboe | 93efe98 | 2017-03-24 12:04:19 -0600 | [diff] [blame] | 2024 | errors = queued = 0; |
Omar Sandoval | 81380ca | 2017-04-07 08:56:26 -0600 | [diff] [blame] | 2025 | do { |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2026 | struct blk_mq_queue_data bd; |
| 2027 | |
| 2028 | rq = list_first_entry(list, struct request, queuelist); |
Ming Lei | 0bca799 | 2018-04-05 00:35:21 +0800 | [diff] [blame] | 2029 | |
Ming Lei | 445874e | 2020-06-30 18:24:57 +0800 | [diff] [blame] | 2030 | WARN_ON_ONCE(hctx != rq->mq_hctx); |
Ming Lei | 1fd40b5 | 2020-06-30 18:25:00 +0800 | [diff] [blame] | 2031 | prep = blk_mq_prep_dispatch_rq(rq, !nr_budgets); |
Ming Lei | 7538352 | 2020-06-30 18:24:58 +0800 | [diff] [blame] | 2032 | if (prep != PREP_DISPATCH_OK) |
Ming Lei | 0bca799 | 2018-04-05 00:35:21 +0800 | [diff] [blame] | 2033 | break; |
Ming Lei | de14829 | 2017-10-14 17:22:29 +0800 | [diff] [blame] | 2034 | |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2035 | list_del_init(&rq->queuelist); |
| 2036 | |
| 2037 | bd.rq = rq; |
Jens Axboe | 113285b | 2017-03-02 13:26:04 -0700 | [diff] [blame] | 2038 | |
| 2039 | /* |
| 2040 | * Flag last if we have no more requests, or if we have more |
| 2041 | * but can't assign a driver tag to it. |
| 2042 | */ |
| 2043 | if (list_empty(list)) |
| 2044 | bd.last = true; |
| 2045 | else { |
Jens Axboe | 113285b | 2017-03-02 13:26:04 -0700 | [diff] [blame] | 2046 | nxt = list_first_entry(list, struct request, queuelist); |
Ming Lei | 8ab6bb9e | 2018-06-25 19:31:45 +0800 | [diff] [blame] | 2047 | bd.last = !blk_mq_get_driver_tag(nxt); |
Jens Axboe | 113285b | 2017-03-02 13:26:04 -0700 | [diff] [blame] | 2048 | } |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2049 | |
Ming Lei | 1fd40b5 | 2020-06-30 18:25:00 +0800 | [diff] [blame] | 2050 | /* |
| 2051 | * once the request is queued to lld, no need to cover the |
| 2052 | * budget any more |
| 2053 | */ |
| 2054 | if (nr_budgets) |
| 2055 | nr_budgets--; |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2056 | ret = q->mq_ops->queue_rq(hctx, &bd); |
Ming Lei | 7bf1372 | 2020-07-01 21:58:57 +0800 | [diff] [blame] | 2057 | switch (ret) { |
| 2058 | case BLK_STS_OK: |
| 2059 | queued++; |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2060 | break; |
Ming Lei | 7bf1372 | 2020-07-01 21:58:57 +0800 | [diff] [blame] | 2061 | case BLK_STS_RESOURCE: |
Naohiro Aota | 9586e67 | 2021-10-27 01:51:27 +0900 | [diff] [blame] | 2062 | needs_resource = true; |
| 2063 | fallthrough; |
Ming Lei | 7bf1372 | 2020-07-01 21:58:57 +0800 | [diff] [blame] | 2064 | case BLK_STS_DEV_RESOURCE: |
| 2065 | blk_mq_handle_dev_resource(rq, list); |
| 2066 | goto out; |
| 2067 | case BLK_STS_ZONE_RESOURCE: |
Keith Busch | 0512a75 | 2020-05-12 17:55:47 +0900 | [diff] [blame] | 2068 | /* |
| 2069 | * Move the request to zone_list and keep going through |
| 2070 | * the dispatch list to find more requests the drive can |
| 2071 | * accept. |
| 2072 | */ |
| 2073 | blk_mq_handle_zone_resource(rq, &zone_list); |
Naohiro Aota | 9586e67 | 2021-10-27 01:51:27 +0900 | [diff] [blame] | 2074 | needs_resource = true; |
Ming Lei | 7bf1372 | 2020-07-01 21:58:57 +0800 | [diff] [blame] | 2075 | break; |
| 2076 | default: |
Christoph Hellwig | fc17b65 | 2017-06-03 09:38:05 +0200 | [diff] [blame] | 2077 | errors++; |
Hannes Reinecke | e21ee5a | 2020-09-30 10:02:53 +0200 | [diff] [blame] | 2078 | blk_mq_end_request(rq, ret); |
Christoph Hellwig | fc17b65 | 2017-06-03 09:38:05 +0200 | [diff] [blame] | 2079 | } |
Omar Sandoval | 81380ca | 2017-04-07 08:56:26 -0600 | [diff] [blame] | 2080 | } while (!list_empty(list)); |
Ming Lei | 7bf1372 | 2020-07-01 21:58:57 +0800 | [diff] [blame] | 2081 | out: |
Keith Busch | 0512a75 | 2020-05-12 17:55:47 +0900 | [diff] [blame] | 2082 | if (!list_empty(&zone_list)) |
| 2083 | list_splice_tail_init(&zone_list, list); |
| 2084 | |
yangerkun | 632bfb6 | 2020-09-05 19:25:56 +0800 | [diff] [blame] | 2085 | /* If we didn't flush the entire list, we could have told the driver |
| 2086 | * there was more coming, but that turned out to be a lie. |
| 2087 | */ |
Yu Kuai | 65fac0d | 2022-07-26 20:22:24 +0800 | [diff] [blame] | 2088 | if ((!list_empty(list) || errors || needs_resource || |
| 2089 | ret == BLK_STS_DEV_RESOURCE) && q->mq_ops->commit_rqs && queued) |
yangerkun | 632bfb6 | 2020-09-05 19:25:56 +0800 | [diff] [blame] | 2090 | q->mq_ops->commit_rqs(hctx); |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2091 | /* |
| 2092 | * Any items that need requeuing? Stuff them into hctx->dispatch, |
| 2093 | * that is where we will continue on next queue run. |
| 2094 | */ |
| 2095 | if (!list_empty(list)) { |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 2096 | bool needs_restart; |
Ming Lei | 7538352 | 2020-06-30 18:24:58 +0800 | [diff] [blame] | 2097 | /* For non-shared tags, the RESTART check will suffice */ |
| 2098 | bool no_tag = prep == PREP_DISPATCH_NO_TAG && |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 2099 | (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED); |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 2100 | |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 2101 | if (nr_budgets) |
| 2102 | blk_mq_release_budgets(q, list); |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2103 | |
| 2104 | spin_lock(&hctx->lock); |
Ming Lei | 01e99ae | 2020-02-25 09:04:32 +0800 | [diff] [blame] | 2105 | list_splice_tail_init(list, &hctx->dispatch); |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2106 | spin_unlock(&hctx->lock); |
| 2107 | |
| 2108 | /* |
Ming Lei | d7d8535 | 2020-08-17 18:01:15 +0800 | [diff] [blame] | 2109 | * Order adding requests to hctx->dispatch and checking |
| 2110 | * SCHED_RESTART flag. The pair of this smp_mb() is the one |
| 2111 | * in blk_mq_sched_restart(). Avoid restart code path to |
| 2112 | * miss the new added requests to hctx->dispatch, meantime |
| 2113 | * SCHED_RESTART is observed here. |
| 2114 | */ |
| 2115 | smp_mb(); |
| 2116 | |
| 2117 | /* |
Bart Van Assche | 710c785 | 2017-04-07 11:16:51 -0700 | [diff] [blame] | 2118 | * If SCHED_RESTART was set by the caller of this function and |
| 2119 | * it is no longer set that means that it was cleared by another |
| 2120 | * thread and hence that a queue rerun is needed. |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2121 | * |
Jens Axboe | eb619fd | 2017-11-09 08:32:43 -0700 | [diff] [blame] | 2122 | * If 'no_tag' is set, that means that we failed getting |
| 2123 | * a driver tag with an I/O scheduler attached. If our dispatch |
| 2124 | * waitqueue is no longer active, ensure that we run the queue |
| 2125 | * AFTER adding our entries back to the list. |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 2126 | * |
Bart Van Assche | 710c785 | 2017-04-07 11:16:51 -0700 | [diff] [blame] | 2127 | * If no I/O scheduler has been configured it is possible that |
| 2128 | * the hardware queue got stopped and restarted before requests |
| 2129 | * were pushed back onto the dispatch list. Rerun the queue to |
| 2130 | * avoid starvation. Notes: |
| 2131 | * - blk_mq_run_hw_queue() checks whether or not a queue has |
| 2132 | * been stopped before rerunning a queue. |
| 2133 | * - Some but not all block drivers stop a queue before |
Christoph Hellwig | fc17b65 | 2017-06-03 09:38:05 +0200 | [diff] [blame] | 2134 | * returning BLK_STS_RESOURCE. Two exceptions are scsi-mq |
Bart Van Assche | 710c785 | 2017-04-07 11:16:51 -0700 | [diff] [blame] | 2135 | * and dm-rq. |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 2136 | * |
| 2137 | * If driver returns BLK_STS_RESOURCE and SCHED_RESTART |
| 2138 | * bit is set, run queue after a delay to avoid IO stalls |
Douglas Anderson | ab3cee3 | 2020-04-20 09:24:51 -0700 | [diff] [blame] | 2139 | * that could otherwise occur if the queue is idle. We'll do |
Naohiro Aota | 9586e67 | 2021-10-27 01:51:27 +0900 | [diff] [blame] | 2140 | * similar if we couldn't get budget or couldn't lock a zone |
| 2141 | * and SCHED_RESTART is set. |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 2142 | */ |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 2143 | needs_restart = blk_mq_sched_needs_restart(hctx); |
Naohiro Aota | 9586e67 | 2021-10-27 01:51:27 +0900 | [diff] [blame] | 2144 | if (prep == PREP_DISPATCH_NO_BUDGET) |
| 2145 | needs_resource = true; |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 2146 | if (!needs_restart || |
Jens Axboe | eb619fd | 2017-11-09 08:32:43 -0700 | [diff] [blame] | 2147 | (no_tag && list_empty_careful(&hctx->dispatch_wait.entry))) |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 2148 | blk_mq_run_hw_queue(hctx, true); |
Miaohe Lin | 6d5e8d2 | 2022-09-05 18:19:50 +0800 | [diff] [blame] | 2149 | else if (needs_resource) |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 2150 | blk_mq_delay_run_hw_queue(hctx, BLK_MQ_RESOURCE_DELAY); |
Jens Axboe | 1f57f8d | 2018-06-28 11:54:01 -0600 | [diff] [blame] | 2151 | |
Ming Lei | 6e768717 | 2018-07-03 09:03:16 -0600 | [diff] [blame] | 2152 | blk_mq_update_dispatch_busy(hctx, true); |
Jens Axboe | 1f57f8d | 2018-06-28 11:54:01 -0600 | [diff] [blame] | 2153 | return false; |
Ming Lei | 6e768717 | 2018-07-03 09:03:16 -0600 | [diff] [blame] | 2154 | } else |
| 2155 | blk_mq_update_dispatch_busy(hctx, false); |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2156 | |
Jens Axboe | 93efe98 | 2017-03-24 12:04:19 -0600 | [diff] [blame] | 2157 | return (queued + errors) != 0; |
Jens Axboe | f04c3df | 2016-12-07 08:41:17 -0700 | [diff] [blame] | 2158 | } |
| 2159 | |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2160 | /** |
| 2161 | * __blk_mq_run_hw_queue - Run a hardware queue. |
| 2162 | * @hctx: Pointer to the hardware queue to run. |
| 2163 | * |
| 2164 | * Send pending requests to the hardware. |
| 2165 | */ |
Bart Van Assche | 6a83e74 | 2016-11-02 10:09:51 -0600 | [diff] [blame] | 2166 | static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx) |
| 2167 | { |
Jens Axboe | b7a71e6 | 2017-08-01 09:28:24 -0600 | [diff] [blame] | 2168 | /* |
Jens Axboe | b7a71e6 | 2017-08-01 09:28:24 -0600 | [diff] [blame] | 2169 | * We can't run the queue inline with ints disabled. Ensure that |
| 2170 | * we catch bad users of this early. |
| 2171 | */ |
| 2172 | WARN_ON_ONCE(in_interrupt()); |
| 2173 | |
Ming Lei | bcc330f | 2021-12-03 21:15:33 +0800 | [diff] [blame] | 2174 | blk_mq_run_dispatch_ops(hctx->queue, |
| 2175 | blk_mq_sched_dispatch_requests(hctx)); |
Bart Van Assche | 6a83e74 | 2016-11-02 10:09:51 -0600 | [diff] [blame] | 2176 | } |
| 2177 | |
Ming Lei | f82ddf1 | 2018-04-08 17:48:10 +0800 | [diff] [blame] | 2178 | static inline int blk_mq_first_mapped_cpu(struct blk_mq_hw_ctx *hctx) |
| 2179 | { |
| 2180 | int cpu = cpumask_first_and(hctx->cpumask, cpu_online_mask); |
| 2181 | |
| 2182 | if (cpu >= nr_cpu_ids) |
| 2183 | cpu = cpumask_first(hctx->cpumask); |
| 2184 | return cpu; |
| 2185 | } |
| 2186 | |
Jens Axboe | 506e931 | 2014-05-07 10:26:44 -0600 | [diff] [blame] | 2187 | /* |
| 2188 | * It'd be great if the workqueue API had a way to pass |
| 2189 | * in a mask and had some smarts for more clever placement. |
| 2190 | * For now we just round-robin here, switching for every |
| 2191 | * BLK_MQ_CPU_WORK_BATCH queued items. |
| 2192 | */ |
| 2193 | static int blk_mq_hctx_next_cpu(struct blk_mq_hw_ctx *hctx) |
| 2194 | { |
Ming Lei | 7bed459 | 2018-01-18 00:41:51 +0800 | [diff] [blame] | 2195 | bool tried = false; |
Ming Lei | 476f8c9 | 2018-04-08 17:48:09 +0800 | [diff] [blame] | 2196 | int next_cpu = hctx->next_cpu; |
Ming Lei | 7bed459 | 2018-01-18 00:41:51 +0800 | [diff] [blame] | 2197 | |
Christoph Hellwig | b657d7e | 2014-11-24 09:27:23 +0100 | [diff] [blame] | 2198 | if (hctx->queue->nr_hw_queues == 1) |
| 2199 | return WORK_CPU_UNBOUND; |
Jens Axboe | 506e931 | 2014-05-07 10:26:44 -0600 | [diff] [blame] | 2200 | |
| 2201 | if (--hctx->next_cpu_batch <= 0) { |
Ming Lei | 7bed459 | 2018-01-18 00:41:51 +0800 | [diff] [blame] | 2202 | select_cpu: |
Ming Lei | 476f8c9 | 2018-04-08 17:48:09 +0800 | [diff] [blame] | 2203 | next_cpu = cpumask_next_and(next_cpu, hctx->cpumask, |
Christoph Hellwig | 20e4d813 | 2018-01-12 10:53:06 +0800 | [diff] [blame] | 2204 | cpu_online_mask); |
Jens Axboe | 506e931 | 2014-05-07 10:26:44 -0600 | [diff] [blame] | 2205 | if (next_cpu >= nr_cpu_ids) |
Ming Lei | f82ddf1 | 2018-04-08 17:48:10 +0800 | [diff] [blame] | 2206 | next_cpu = blk_mq_first_mapped_cpu(hctx); |
Jens Axboe | 506e931 | 2014-05-07 10:26:44 -0600 | [diff] [blame] | 2207 | hctx->next_cpu_batch = BLK_MQ_CPU_WORK_BATCH; |
| 2208 | } |
| 2209 | |
Ming Lei | 7bed459 | 2018-01-18 00:41:51 +0800 | [diff] [blame] | 2210 | /* |
| 2211 | * Do unbound schedule if we can't find a online CPU for this hctx, |
| 2212 | * and it should only happen in the path of handling CPU DEAD. |
| 2213 | */ |
Ming Lei | 476f8c9 | 2018-04-08 17:48:09 +0800 | [diff] [blame] | 2214 | if (!cpu_online(next_cpu)) { |
Ming Lei | 7bed459 | 2018-01-18 00:41:51 +0800 | [diff] [blame] | 2215 | if (!tried) { |
| 2216 | tried = true; |
| 2217 | goto select_cpu; |
| 2218 | } |
| 2219 | |
| 2220 | /* |
| 2221 | * Make sure to re-select CPU next time once after CPUs |
| 2222 | * in hctx->cpumask become online again. |
| 2223 | */ |
Ming Lei | 476f8c9 | 2018-04-08 17:48:09 +0800 | [diff] [blame] | 2224 | hctx->next_cpu = next_cpu; |
Ming Lei | 7bed459 | 2018-01-18 00:41:51 +0800 | [diff] [blame] | 2225 | hctx->next_cpu_batch = 1; |
| 2226 | return WORK_CPU_UNBOUND; |
| 2227 | } |
Ming Lei | 476f8c9 | 2018-04-08 17:48:09 +0800 | [diff] [blame] | 2228 | |
| 2229 | hctx->next_cpu = next_cpu; |
| 2230 | return next_cpu; |
Jens Axboe | 506e931 | 2014-05-07 10:26:44 -0600 | [diff] [blame] | 2231 | } |
| 2232 | |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2233 | /** |
| 2234 | * __blk_mq_delay_run_hw_queue - Run (or schedule to run) a hardware queue. |
| 2235 | * @hctx: Pointer to the hardware queue to run. |
| 2236 | * @async: If we want to run the queue asynchronously. |
Minwoo Im | fa94ba8 | 2020-12-05 00:20:55 +0900 | [diff] [blame] | 2237 | * @msecs: Milliseconds of delay to wait before running the queue. |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2238 | * |
| 2239 | * If !@async, try to run the queue now. Else, run the queue asynchronously and |
| 2240 | * with a delay of @msecs. |
| 2241 | */ |
Bart Van Assche | 7587a5a | 2017-04-07 11:16:52 -0700 | [diff] [blame] | 2242 | static void __blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async, |
| 2243 | unsigned long msecs) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2244 | { |
Bart Van Assche | 5435c02 | 2017-06-20 11:15:49 -0700 | [diff] [blame] | 2245 | if (unlikely(blk_mq_hctx_stopped(hctx))) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2246 | return; |
| 2247 | |
Jens Axboe | 1b792f2 | 2016-09-21 10:12:13 -0600 | [diff] [blame] | 2248 | if (!async && !(hctx->flags & BLK_MQ_F_BLOCKING)) { |
Sebastian Andrzej Siewior | 3c8f9da | 2022-06-22 10:25:54 +0200 | [diff] [blame] | 2249 | if (cpumask_test_cpu(raw_smp_processor_id(), hctx->cpumask)) { |
Paolo Bonzini | 398205b | 2014-11-07 23:03:59 +0100 | [diff] [blame] | 2250 | __blk_mq_run_hw_queue(hctx); |
Paolo Bonzini | 398205b | 2014-11-07 23:03:59 +0100 | [diff] [blame] | 2251 | return; |
| 2252 | } |
Jens Axboe | e4043dc | 2014-04-09 10:18:23 -0600 | [diff] [blame] | 2253 | } |
Paolo Bonzini | 398205b | 2014-11-07 23:03:59 +0100 | [diff] [blame] | 2254 | |
Bart Van Assche | ae943d2 | 2018-01-19 08:58:55 -0800 | [diff] [blame] | 2255 | kblockd_mod_delayed_work_on(blk_mq_hctx_next_cpu(hctx), &hctx->run_work, |
| 2256 | msecs_to_jiffies(msecs)); |
Bart Van Assche | 7587a5a | 2017-04-07 11:16:52 -0700 | [diff] [blame] | 2257 | } |
| 2258 | |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2259 | /** |
| 2260 | * blk_mq_delay_run_hw_queue - Run a hardware queue asynchronously. |
| 2261 | * @hctx: Pointer to the hardware queue to run. |
Minwoo Im | fa94ba8 | 2020-12-05 00:20:55 +0900 | [diff] [blame] | 2262 | * @msecs: Milliseconds of delay to wait before running the queue. |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2263 | * |
| 2264 | * Run a hardware queue asynchronously with a delay of @msecs. |
| 2265 | */ |
Bart Van Assche | 7587a5a | 2017-04-07 11:16:52 -0700 | [diff] [blame] | 2266 | void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs) |
| 2267 | { |
| 2268 | __blk_mq_delay_run_hw_queue(hctx, true, msecs); |
| 2269 | } |
| 2270 | EXPORT_SYMBOL(blk_mq_delay_run_hw_queue); |
| 2271 | |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2272 | /** |
| 2273 | * blk_mq_run_hw_queue - Start to run a hardware queue. |
| 2274 | * @hctx: Pointer to the hardware queue to run. |
| 2275 | * @async: If we want to run the queue asynchronously. |
| 2276 | * |
| 2277 | * Check if the request queue is not in a quiesced state and if there are |
| 2278 | * pending requests to be sent. If this is true, run the queue to send requests |
| 2279 | * to hardware. |
| 2280 | */ |
John Garry | 626fb73 | 2019-10-30 00:59:30 +0800 | [diff] [blame] | 2281 | void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async) |
Bart Van Assche | 7587a5a | 2017-04-07 11:16:52 -0700 | [diff] [blame] | 2282 | { |
Ming Lei | 24f5a90 | 2018-01-06 16:27:38 +0800 | [diff] [blame] | 2283 | bool need_run; |
| 2284 | |
| 2285 | /* |
| 2286 | * When queue is quiesced, we may be switching io scheduler, or |
| 2287 | * updating nr_hw_queues, or other things, and we can't run queue |
| 2288 | * any more, even __blk_mq_hctx_has_pending() can't be called safely. |
| 2289 | * |
| 2290 | * And queue will be rerun in blk_mq_unquiesce_queue() if it is |
| 2291 | * quiesced. |
| 2292 | */ |
Ming Lei | 41adf53 | 2021-12-06 19:12:13 +0800 | [diff] [blame] | 2293 | __blk_mq_run_dispatch_ops(hctx->queue, false, |
Ming Lei | 2a904d0 | 2021-12-03 21:15:31 +0800 | [diff] [blame] | 2294 | need_run = !blk_queue_quiesced(hctx->queue) && |
| 2295 | blk_mq_hctx_has_pending(hctx)); |
Ming Lei | 24f5a90 | 2018-01-06 16:27:38 +0800 | [diff] [blame] | 2296 | |
John Garry | 626fb73 | 2019-10-30 00:59:30 +0800 | [diff] [blame] | 2297 | if (need_run) |
Jens Axboe | 79f720a | 2017-11-10 09:13:21 -0700 | [diff] [blame] | 2298 | __blk_mq_delay_run_hw_queue(hctx, async, 0); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2299 | } |
Omar Sandoval | 5b72727 | 2017-04-14 01:00:00 -0700 | [diff] [blame] | 2300 | EXPORT_SYMBOL(blk_mq_run_hw_queue); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2301 | |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2302 | /* |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2303 | * Return prefered queue to dispatch from (if any) for non-mq aware IO |
| 2304 | * scheduler. |
| 2305 | */ |
| 2306 | static struct blk_mq_hw_ctx *blk_mq_get_sq_hctx(struct request_queue *q) |
| 2307 | { |
Ming Lei | 5d05426 | 2022-05-22 20:23:50 +0800 | [diff] [blame] | 2308 | struct blk_mq_ctx *ctx = blk_mq_get_ctx(q); |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2309 | /* |
| 2310 | * If the IO scheduler does not respect hardware queues when |
| 2311 | * dispatching, we just don't bother with multiple HW queues and |
| 2312 | * dispatch from hctx for the current CPU since running multiple queues |
| 2313 | * just causes lock contention inside the scheduler and pointless cache |
| 2314 | * bouncing. |
| 2315 | */ |
Bart Van Assche | 51ab80f | 2022-06-15 15:55:49 -0700 | [diff] [blame] | 2316 | struct blk_mq_hw_ctx *hctx = ctx->hctxs[HCTX_TYPE_DEFAULT]; |
Ming Lei | 5d05426 | 2022-05-22 20:23:50 +0800 | [diff] [blame] | 2317 | |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2318 | if (!blk_mq_hctx_stopped(hctx)) |
| 2319 | return hctx; |
| 2320 | return NULL; |
| 2321 | } |
| 2322 | |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2323 | /** |
Mauro Carvalho Chehab | 24f7bb8 | 2020-10-23 18:32:54 +0200 | [diff] [blame] | 2324 | * blk_mq_run_hw_queues - Run all hardware queues in a request queue. |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2325 | * @q: Pointer to the request queue to run. |
| 2326 | * @async: If we want to run the queue asynchronously. |
| 2327 | */ |
Mike Snitzer | b94ec29 | 2015-03-11 23:56:38 -0400 | [diff] [blame] | 2328 | void blk_mq_run_hw_queues(struct request_queue *q, bool async) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2329 | { |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2330 | struct blk_mq_hw_ctx *hctx, *sq_hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 2331 | unsigned long i; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2332 | |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2333 | sq_hctx = NULL; |
Ming Lei | 4d337ceb | 2022-06-16 09:44:00 +0800 | [diff] [blame] | 2334 | if (blk_queue_sq_sched(q)) |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2335 | sq_hctx = blk_mq_get_sq_hctx(q); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2336 | queue_for_each_hw_ctx(q, hctx, i) { |
Jens Axboe | 79f720a | 2017-11-10 09:13:21 -0700 | [diff] [blame] | 2337 | if (blk_mq_hctx_stopped(hctx)) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2338 | continue; |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2339 | /* |
| 2340 | * Dispatch from this hctx either if there's no hctx preferred |
| 2341 | * by IO scheduler or if it has requests that bypass the |
| 2342 | * scheduler. |
| 2343 | */ |
| 2344 | if (!sq_hctx || sq_hctx == hctx || |
| 2345 | !list_empty_careful(&hctx->dispatch)) |
| 2346 | blk_mq_run_hw_queue(hctx, async); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2347 | } |
| 2348 | } |
Mike Snitzer | b94ec29 | 2015-03-11 23:56:38 -0400 | [diff] [blame] | 2349 | EXPORT_SYMBOL(blk_mq_run_hw_queues); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2350 | |
Bart Van Assche | fd00144 | 2016-10-28 17:19:37 -0700 | [diff] [blame] | 2351 | /** |
Douglas Anderson | b9151e7 | 2020-04-20 09:24:52 -0700 | [diff] [blame] | 2352 | * blk_mq_delay_run_hw_queues - Run all hardware queues asynchronously. |
| 2353 | * @q: Pointer to the request queue to run. |
Minwoo Im | fa94ba8 | 2020-12-05 00:20:55 +0900 | [diff] [blame] | 2354 | * @msecs: Milliseconds of delay to wait before running the queues. |
Douglas Anderson | b9151e7 | 2020-04-20 09:24:52 -0700 | [diff] [blame] | 2355 | */ |
| 2356 | void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs) |
| 2357 | { |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2358 | struct blk_mq_hw_ctx *hctx, *sq_hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 2359 | unsigned long i; |
Douglas Anderson | b9151e7 | 2020-04-20 09:24:52 -0700 | [diff] [blame] | 2360 | |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2361 | sq_hctx = NULL; |
Ming Lei | 4d337ceb | 2022-06-16 09:44:00 +0800 | [diff] [blame] | 2362 | if (blk_queue_sq_sched(q)) |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2363 | sq_hctx = blk_mq_get_sq_hctx(q); |
Douglas Anderson | b9151e7 | 2020-04-20 09:24:52 -0700 | [diff] [blame] | 2364 | queue_for_each_hw_ctx(q, hctx, i) { |
| 2365 | if (blk_mq_hctx_stopped(hctx)) |
| 2366 | continue; |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2367 | /* |
David Jeffery | 8f5fea6 | 2022-01-31 15:33:37 -0500 | [diff] [blame] | 2368 | * If there is already a run_work pending, leave the |
| 2369 | * pending delay untouched. Otherwise, a hctx can stall |
| 2370 | * if another hctx is re-delaying the other's work |
| 2371 | * before the work executes. |
| 2372 | */ |
| 2373 | if (delayed_work_pending(&hctx->run_work)) |
| 2374 | continue; |
| 2375 | /* |
Jan Kara | b6e68ee | 2021-01-11 17:47:17 +0100 | [diff] [blame] | 2376 | * Dispatch from this hctx either if there's no hctx preferred |
| 2377 | * by IO scheduler or if it has requests that bypass the |
| 2378 | * scheduler. |
| 2379 | */ |
| 2380 | if (!sq_hctx || sq_hctx == hctx || |
| 2381 | !list_empty_careful(&hctx->dispatch)) |
| 2382 | blk_mq_delay_run_hw_queue(hctx, msecs); |
Douglas Anderson | b9151e7 | 2020-04-20 09:24:52 -0700 | [diff] [blame] | 2383 | } |
| 2384 | } |
| 2385 | EXPORT_SYMBOL(blk_mq_delay_run_hw_queues); |
| 2386 | |
Ming Lei | 39a70c7 | 2017-06-06 23:22:09 +0800 | [diff] [blame] | 2387 | /* |
| 2388 | * This function is often used for pausing .queue_rq() by driver when |
| 2389 | * there isn't enough resource or some conditions aren't satisfied, and |
Bart Van Assche | 4d60621 | 2017-08-17 16:23:00 -0700 | [diff] [blame] | 2390 | * BLK_STS_RESOURCE is usually returned. |
Ming Lei | 39a70c7 | 2017-06-06 23:22:09 +0800 | [diff] [blame] | 2391 | * |
| 2392 | * We do not guarantee that dispatch can be drained or blocked |
| 2393 | * after blk_mq_stop_hw_queue() returns. Please use |
| 2394 | * blk_mq_quiesce_queue() for that requirement. |
| 2395 | */ |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2396 | void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx) |
| 2397 | { |
Ming Lei | 641a9ed | 2017-06-06 23:22:10 +0800 | [diff] [blame] | 2398 | cancel_delayed_work(&hctx->run_work); |
| 2399 | |
| 2400 | set_bit(BLK_MQ_S_STOPPED, &hctx->state); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2401 | } |
| 2402 | EXPORT_SYMBOL(blk_mq_stop_hw_queue); |
| 2403 | |
Ming Lei | 39a70c7 | 2017-06-06 23:22:09 +0800 | [diff] [blame] | 2404 | /* |
| 2405 | * This function is often used for pausing .queue_rq() by driver when |
| 2406 | * there isn't enough resource or some conditions aren't satisfied, and |
Bart Van Assche | 4d60621 | 2017-08-17 16:23:00 -0700 | [diff] [blame] | 2407 | * BLK_STS_RESOURCE is usually returned. |
Ming Lei | 39a70c7 | 2017-06-06 23:22:09 +0800 | [diff] [blame] | 2408 | * |
| 2409 | * We do not guarantee that dispatch can be drained or blocked |
| 2410 | * after blk_mq_stop_hw_queues() returns. Please use |
| 2411 | * blk_mq_quiesce_queue() for that requirement. |
| 2412 | */ |
Jens Axboe | 2719aa2 | 2017-05-03 11:08:14 -0600 | [diff] [blame] | 2413 | void blk_mq_stop_hw_queues(struct request_queue *q) |
| 2414 | { |
Ming Lei | 641a9ed | 2017-06-06 23:22:10 +0800 | [diff] [blame] | 2415 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 2416 | unsigned long i; |
Ming Lei | 641a9ed | 2017-06-06 23:22:10 +0800 | [diff] [blame] | 2417 | |
| 2418 | queue_for_each_hw_ctx(q, hctx, i) |
| 2419 | blk_mq_stop_hw_queue(hctx); |
Christoph Hellwig | 280d45f | 2013-10-25 14:45:58 +0100 | [diff] [blame] | 2420 | } |
| 2421 | EXPORT_SYMBOL(blk_mq_stop_hw_queues); |
| 2422 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2423 | void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx) |
| 2424 | { |
| 2425 | clear_bit(BLK_MQ_S_STOPPED, &hctx->state); |
Jens Axboe | e4043dc | 2014-04-09 10:18:23 -0600 | [diff] [blame] | 2426 | |
Jens Axboe | 0ffbce8 | 2014-06-25 08:22:34 -0600 | [diff] [blame] | 2427 | blk_mq_run_hw_queue(hctx, false); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2428 | } |
| 2429 | EXPORT_SYMBOL(blk_mq_start_hw_queue); |
| 2430 | |
Christoph Hellwig | 2f26855 | 2014-04-16 09:44:56 +0200 | [diff] [blame] | 2431 | void blk_mq_start_hw_queues(struct request_queue *q) |
| 2432 | { |
| 2433 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 2434 | unsigned long i; |
Christoph Hellwig | 2f26855 | 2014-04-16 09:44:56 +0200 | [diff] [blame] | 2435 | |
| 2436 | queue_for_each_hw_ctx(q, hctx, i) |
| 2437 | blk_mq_start_hw_queue(hctx); |
| 2438 | } |
| 2439 | EXPORT_SYMBOL(blk_mq_start_hw_queues); |
| 2440 | |
Jens Axboe | ae911c5 | 2016-12-08 13:19:30 -0700 | [diff] [blame] | 2441 | void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async) |
| 2442 | { |
| 2443 | if (!blk_mq_hctx_stopped(hctx)) |
| 2444 | return; |
| 2445 | |
| 2446 | clear_bit(BLK_MQ_S_STOPPED, &hctx->state); |
| 2447 | blk_mq_run_hw_queue(hctx, async); |
| 2448 | } |
| 2449 | EXPORT_SYMBOL_GPL(blk_mq_start_stopped_hw_queue); |
| 2450 | |
Christoph Hellwig | 1b4a325 | 2014-04-16 09:44:54 +0200 | [diff] [blame] | 2451 | void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2452 | { |
| 2453 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 2454 | unsigned long i; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2455 | |
Jens Axboe | ae911c5 | 2016-12-08 13:19:30 -0700 | [diff] [blame] | 2456 | queue_for_each_hw_ctx(q, hctx, i) |
| 2457 | blk_mq_start_stopped_hw_queue(hctx, async); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2458 | } |
| 2459 | EXPORT_SYMBOL(blk_mq_start_stopped_hw_queues); |
| 2460 | |
Christoph Hellwig | 70f4db6 | 2014-04-16 10:48:08 -0600 | [diff] [blame] | 2461 | static void blk_mq_run_work_fn(struct work_struct *work) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2462 | { |
| 2463 | struct blk_mq_hw_ctx *hctx; |
| 2464 | |
Jens Axboe | 9f99373 | 2017-04-10 09:54:54 -0600 | [diff] [blame] | 2465 | hctx = container_of(work, struct blk_mq_hw_ctx, run_work.work); |
Jens Axboe | 21c6e93 | 2017-04-10 09:54:56 -0600 | [diff] [blame] | 2466 | |
| 2467 | /* |
Ming Lei | 15fe8a90 | 2018-04-08 17:48:11 +0800 | [diff] [blame] | 2468 | * If we are stopped, don't run the queue. |
Jens Axboe | 21c6e93 | 2017-04-10 09:54:56 -0600 | [diff] [blame] | 2469 | */ |
Yufen Yu | 0841031 | 2020-10-08 23:26:30 -0400 | [diff] [blame] | 2470 | if (blk_mq_hctx_stopped(hctx)) |
Jianchao Wang | 0196d6b | 2018-06-04 17:03:55 +0800 | [diff] [blame] | 2471 | return; |
Jens Axboe | e4043dc | 2014-04-09 10:18:23 -0600 | [diff] [blame] | 2472 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2473 | __blk_mq_run_hw_queue(hctx); |
| 2474 | } |
| 2475 | |
Ming Lei | cfd0c55 | 2015-10-20 23:13:57 +0800 | [diff] [blame] | 2476 | static inline void __blk_mq_insert_req_list(struct blk_mq_hw_ctx *hctx, |
Ming Lei | cfd0c55 | 2015-10-20 23:13:57 +0800 | [diff] [blame] | 2477 | struct request *rq, |
| 2478 | bool at_head) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2479 | { |
Jens Axboe | e57690f | 2016-08-24 15:34:35 -0600 | [diff] [blame] | 2480 | struct blk_mq_ctx *ctx = rq->mq_ctx; |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 2481 | enum hctx_type type = hctx->type; |
Jens Axboe | e57690f | 2016-08-24 15:34:35 -0600 | [diff] [blame] | 2482 | |
Bart Van Assche | 7b60781 | 2017-06-20 11:15:47 -0700 | [diff] [blame] | 2483 | lockdep_assert_held(&ctx->lock); |
| 2484 | |
Christoph Hellwig | a54895f | 2020-12-03 17:21:39 +0100 | [diff] [blame] | 2485 | trace_block_rq_insert(rq); |
Jens Axboe | 01b983c | 2013-11-19 18:59:10 -0700 | [diff] [blame] | 2486 | |
Christoph Hellwig | 72a0a36 | 2014-02-07 10:22:36 -0800 | [diff] [blame] | 2487 | if (at_head) |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 2488 | list_add(&rq->queuelist, &ctx->rq_lists[type]); |
Christoph Hellwig | 72a0a36 | 2014-02-07 10:22:36 -0800 | [diff] [blame] | 2489 | else |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 2490 | list_add_tail(&rq->queuelist, &ctx->rq_lists[type]); |
Ming Lei | cfd0c55 | 2015-10-20 23:13:57 +0800 | [diff] [blame] | 2491 | } |
Jens Axboe | 4bb659b | 2014-05-09 09:36:49 -0600 | [diff] [blame] | 2492 | |
Jens Axboe | 2c3ad66 | 2016-12-14 14:34:47 -0700 | [diff] [blame] | 2493 | void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq, |
| 2494 | bool at_head) |
Ming Lei | cfd0c55 | 2015-10-20 23:13:57 +0800 | [diff] [blame] | 2495 | { |
| 2496 | struct blk_mq_ctx *ctx = rq->mq_ctx; |
| 2497 | |
Bart Van Assche | 7b60781 | 2017-06-20 11:15:47 -0700 | [diff] [blame] | 2498 | lockdep_assert_held(&ctx->lock); |
| 2499 | |
Jens Axboe | e57690f | 2016-08-24 15:34:35 -0600 | [diff] [blame] | 2500 | __blk_mq_insert_req_list(hctx, rq, at_head); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2501 | blk_mq_hctx_mark_pending(hctx, ctx); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2502 | } |
| 2503 | |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2504 | /** |
| 2505 | * blk_mq_request_bypass_insert - Insert a request at dispatch list. |
| 2506 | * @rq: Pointer to request to be inserted. |
Randy Dunlap | 26bfeb2 | 2020-08-16 16:39:34 -0700 | [diff] [blame] | 2507 | * @at_head: true if the request should be inserted at the head of the list. |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2508 | * @run_queue: If we should run the hardware queue after inserting the request. |
| 2509 | * |
Jens Axboe | 157f377 | 2017-09-11 16:43:57 -0600 | [diff] [blame] | 2510 | * Should only be used carefully, when the caller knows we want to |
| 2511 | * bypass a potential IO scheduler on the target device. |
| 2512 | */ |
Ming Lei | 01e99ae | 2020-02-25 09:04:32 +0800 | [diff] [blame] | 2513 | void blk_mq_request_bypass_insert(struct request *rq, bool at_head, |
| 2514 | bool run_queue) |
Jens Axboe | 157f377 | 2017-09-11 16:43:57 -0600 | [diff] [blame] | 2515 | { |
Jens Axboe | ea4f995 | 2018-10-29 15:06:13 -0600 | [diff] [blame] | 2516 | struct blk_mq_hw_ctx *hctx = rq->mq_hctx; |
Jens Axboe | 157f377 | 2017-09-11 16:43:57 -0600 | [diff] [blame] | 2517 | |
| 2518 | spin_lock(&hctx->lock); |
Ming Lei | 01e99ae | 2020-02-25 09:04:32 +0800 | [diff] [blame] | 2519 | if (at_head) |
| 2520 | list_add(&rq->queuelist, &hctx->dispatch); |
| 2521 | else |
| 2522 | list_add_tail(&rq->queuelist, &hctx->dispatch); |
Jens Axboe | 157f377 | 2017-09-11 16:43:57 -0600 | [diff] [blame] | 2523 | spin_unlock(&hctx->lock); |
| 2524 | |
Ming Lei | b085029 | 2017-11-02 23:24:34 +0800 | [diff] [blame] | 2525 | if (run_queue) |
| 2526 | blk_mq_run_hw_queue(hctx, false); |
Jens Axboe | 157f377 | 2017-09-11 16:43:57 -0600 | [diff] [blame] | 2527 | } |
| 2528 | |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 2529 | void blk_mq_insert_requests(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx, |
| 2530 | struct list_head *list) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2531 | |
| 2532 | { |
Ming Lei | 3f0cedc | 2018-07-02 17:35:58 +0800 | [diff] [blame] | 2533 | struct request *rq; |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 2534 | enum hctx_type type = hctx->type; |
Ming Lei | 3f0cedc | 2018-07-02 17:35:58 +0800 | [diff] [blame] | 2535 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2536 | /* |
| 2537 | * preemption doesn't flush plug list, so it's possible ctx->cpu is |
| 2538 | * offline now |
| 2539 | */ |
Ming Lei | 3f0cedc | 2018-07-02 17:35:58 +0800 | [diff] [blame] | 2540 | list_for_each_entry(rq, list, queuelist) { |
Jens Axboe | e57690f | 2016-08-24 15:34:35 -0600 | [diff] [blame] | 2541 | BUG_ON(rq->mq_ctx != ctx); |
Christoph Hellwig | a54895f | 2020-12-03 17:21:39 +0100 | [diff] [blame] | 2542 | trace_block_rq_insert(rq); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2543 | } |
Ming Lei | 3f0cedc | 2018-07-02 17:35:58 +0800 | [diff] [blame] | 2544 | |
| 2545 | spin_lock(&ctx->lock); |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 2546 | list_splice_tail_init(list, &ctx->rq_lists[type]); |
Ming Lei | cfd0c55 | 2015-10-20 23:13:57 +0800 | [diff] [blame] | 2547 | blk_mq_hctx_mark_pending(hctx, ctx); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2548 | spin_unlock(&ctx->lock); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2549 | } |
| 2550 | |
Jens Axboe | dc5fc361 | 2021-10-19 06:02:30 -0600 | [diff] [blame] | 2551 | static void blk_mq_commit_rqs(struct blk_mq_hw_ctx *hctx, int *queued, |
| 2552 | bool from_schedule) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2553 | { |
Jens Axboe | dc5fc361 | 2021-10-19 06:02:30 -0600 | [diff] [blame] | 2554 | if (hctx->queue->mq_ops->commit_rqs) { |
| 2555 | trace_block_unplug(hctx->queue, *queued, !from_schedule); |
| 2556 | hctx->queue->mq_ops->commit_rqs(hctx); |
| 2557 | } |
| 2558 | *queued = 0; |
| 2559 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2560 | |
Christoph Hellwig | 14ccb66 | 2019-06-06 12:29:01 +0200 | [diff] [blame] | 2561 | static void blk_mq_bio_to_request(struct request *rq, struct bio *bio, |
| 2562 | unsigned int nr_segs) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2563 | { |
Eric Biggers | 93f221a | 2020-09-15 20:53:14 -0700 | [diff] [blame] | 2564 | int err; |
| 2565 | |
Christoph Hellwig | f924cdd | 2019-06-06 12:29:00 +0200 | [diff] [blame] | 2566 | if (bio->bi_opf & REQ_RAHEAD) |
| 2567 | rq->cmd_flags |= REQ_FAILFAST_MASK; |
| 2568 | |
| 2569 | rq->__sector = bio->bi_iter.bi_sector; |
Christoph Hellwig | 14ccb66 | 2019-06-06 12:29:01 +0200 | [diff] [blame] | 2570 | blk_rq_bio_prep(rq, bio, nr_segs); |
Eric Biggers | 93f221a | 2020-09-15 20:53:14 -0700 | [diff] [blame] | 2571 | |
| 2572 | /* This can't fail, since GFP_NOIO includes __GFP_DIRECT_RECLAIM. */ |
| 2573 | err = blk_crypto_rq_bio_prep(rq, bio, GFP_NOIO); |
| 2574 | WARN_ON_ONCE(err); |
Jens Axboe | 4b57052 | 2014-05-29 11:00:11 -0600 | [diff] [blame] | 2575 | |
Konstantin Khlebnikov | b5af37a | 2020-05-27 07:24:16 +0200 | [diff] [blame] | 2576 | blk_account_io_start(rq); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 2577 | } |
| 2578 | |
Mike Snitzer | 0f95549 | 2018-01-17 11:25:56 -0500 | [diff] [blame] | 2579 | static blk_status_t __blk_mq_issue_directly(struct blk_mq_hw_ctx *hctx, |
Christoph Hellwig | 3e08773 | 2021-10-12 13:12:24 +0200 | [diff] [blame] | 2580 | struct request *rq, bool last) |
Shaohua Li | f984df1 | 2015-05-08 10:51:32 -0700 | [diff] [blame] | 2581 | { |
Shaohua Li | f984df1 | 2015-05-08 10:51:32 -0700 | [diff] [blame] | 2582 | struct request_queue *q = rq->q; |
Shaohua Li | f984df1 | 2015-05-08 10:51:32 -0700 | [diff] [blame] | 2583 | struct blk_mq_queue_data bd = { |
| 2584 | .rq = rq, |
Jens Axboe | be94f05 | 2018-11-24 10:15:46 -0700 | [diff] [blame] | 2585 | .last = last, |
Shaohua Li | f984df1 | 2015-05-08 10:51:32 -0700 | [diff] [blame] | 2586 | }; |
Jens Axboe | f06345a | 2017-06-12 11:22:46 -0600 | [diff] [blame] | 2587 | blk_status_t ret; |
Mike Snitzer | 0f95549 | 2018-01-17 11:25:56 -0500 | [diff] [blame] | 2588 | |
Mike Snitzer | 0f95549 | 2018-01-17 11:25:56 -0500 | [diff] [blame] | 2589 | /* |
| 2590 | * For OK queue, we are done. For error, caller may kill it. |
| 2591 | * Any other error (busy), just add it to our list as we |
| 2592 | * previously would have done. |
| 2593 | */ |
| 2594 | ret = q->mq_ops->queue_rq(hctx, &bd); |
| 2595 | switch (ret) { |
| 2596 | case BLK_STS_OK: |
Ming Lei | 6ce3dd6 | 2018-07-10 09:03:31 +0800 | [diff] [blame] | 2597 | blk_mq_update_dispatch_busy(hctx, false); |
Mike Snitzer | 0f95549 | 2018-01-17 11:25:56 -0500 | [diff] [blame] | 2598 | break; |
| 2599 | case BLK_STS_RESOURCE: |
Ming Lei | 86ff7c2 | 2018-01-30 22:04:57 -0500 | [diff] [blame] | 2600 | case BLK_STS_DEV_RESOURCE: |
Ming Lei | 6ce3dd6 | 2018-07-10 09:03:31 +0800 | [diff] [blame] | 2601 | blk_mq_update_dispatch_busy(hctx, true); |
Mike Snitzer | 0f95549 | 2018-01-17 11:25:56 -0500 | [diff] [blame] | 2602 | __blk_mq_requeue_request(rq); |
| 2603 | break; |
| 2604 | default: |
Ming Lei | 6ce3dd6 | 2018-07-10 09:03:31 +0800 | [diff] [blame] | 2605 | blk_mq_update_dispatch_busy(hctx, false); |
Mike Snitzer | 0f95549 | 2018-01-17 11:25:56 -0500 | [diff] [blame] | 2606 | break; |
| 2607 | } |
| 2608 | |
| 2609 | return ret; |
| 2610 | } |
| 2611 | |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2612 | static blk_status_t __blk_mq_try_issue_directly(struct blk_mq_hw_ctx *hctx, |
Mike Snitzer | 0f95549 | 2018-01-17 11:25:56 -0500 | [diff] [blame] | 2613 | struct request *rq, |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2614 | bool bypass_insert, bool last) |
Mike Snitzer | 0f95549 | 2018-01-17 11:25:56 -0500 | [diff] [blame] | 2615 | { |
| 2616 | struct request_queue *q = rq->q; |
Ming Lei | d964f04 | 2017-06-06 23:22:00 +0800 | [diff] [blame] | 2617 | bool run_queue = true; |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 2618 | int budget_token; |
Ming Lei | d964f04 | 2017-06-06 23:22:00 +0800 | [diff] [blame] | 2619 | |
Ming Lei | 23d4ee1 | 2018-01-18 12:06:59 +0800 | [diff] [blame] | 2620 | /* |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2621 | * RCU or SRCU read lock is needed before checking quiesced flag. |
Ming Lei | 23d4ee1 | 2018-01-18 12:06:59 +0800 | [diff] [blame] | 2622 | * |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2623 | * When queue is stopped or quiesced, ignore 'bypass_insert' from |
| 2624 | * blk_mq_request_issue_directly(), and return BLK_STS_OK to caller, |
| 2625 | * and avoid driver to try to dispatch again. |
Ming Lei | 23d4ee1 | 2018-01-18 12:06:59 +0800 | [diff] [blame] | 2626 | */ |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2627 | if (blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q)) { |
Ming Lei | d964f04 | 2017-06-06 23:22:00 +0800 | [diff] [blame] | 2628 | run_queue = false; |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2629 | bypass_insert = false; |
| 2630 | goto insert; |
Ming Lei | d964f04 | 2017-06-06 23:22:00 +0800 | [diff] [blame] | 2631 | } |
Shaohua Li | f984df1 | 2015-05-08 10:51:32 -0700 | [diff] [blame] | 2632 | |
Jens Axboe | 2ff0682 | 2021-10-15 09:44:38 -0600 | [diff] [blame] | 2633 | if ((rq->rq_flags & RQF_ELV) && !bypass_insert) |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2634 | goto insert; |
Bart Van Assche | 2253efc | 2016-10-28 17:20:02 -0700 | [diff] [blame] | 2635 | |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 2636 | budget_token = blk_mq_get_dispatch_budget(q); |
| 2637 | if (budget_token < 0) |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2638 | goto insert; |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 2639 | |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 2640 | blk_mq_set_rq_budget_token(rq, budget_token); |
| 2641 | |
Ming Lei | 8ab6bb9e | 2018-06-25 19:31:45 +0800 | [diff] [blame] | 2642 | if (!blk_mq_get_driver_tag(rq)) { |
Ming Lei | 2a5a24a | 2021-01-22 10:33:12 +0800 | [diff] [blame] | 2643 | blk_mq_put_dispatch_budget(q, budget_token); |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2644 | goto insert; |
Ming Lei | 88022d7 | 2017-11-05 02:21:12 +0800 | [diff] [blame] | 2645 | } |
Ming Lei | de14829 | 2017-10-14 17:22:29 +0800 | [diff] [blame] | 2646 | |
Christoph Hellwig | 3e08773 | 2021-10-12 13:12:24 +0200 | [diff] [blame] | 2647 | return __blk_mq_issue_directly(hctx, rq, last); |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2648 | insert: |
| 2649 | if (bypass_insert) |
| 2650 | return BLK_STS_RESOURCE; |
| 2651 | |
Ming Lei | db03f88 | 2020-08-18 17:07:28 +0800 | [diff] [blame] | 2652 | blk_mq_sched_insert_request(rq, false, run_queue, false); |
| 2653 | |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2654 | return BLK_STS_OK; |
| 2655 | } |
| 2656 | |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2657 | /** |
| 2658 | * blk_mq_try_issue_directly - Try to send a request directly to device driver. |
| 2659 | * @hctx: Pointer of the associated hardware queue. |
| 2660 | * @rq: Pointer to request to be sent. |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2661 | * |
| 2662 | * If the device has enough resources to accept a new request now, send the |
| 2663 | * request directly to device driver. Else, insert at hctx->dispatch queue, so |
| 2664 | * we can try send it another time in the future. Requests inserted at this |
| 2665 | * queue have higher priority. |
| 2666 | */ |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2667 | static void blk_mq_try_issue_directly(struct blk_mq_hw_ctx *hctx, |
Christoph Hellwig | 3e08773 | 2021-10-12 13:12:24 +0200 | [diff] [blame] | 2668 | struct request *rq) |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2669 | { |
Ming Lei | 2a904d0 | 2021-12-03 21:15:31 +0800 | [diff] [blame] | 2670 | blk_status_t ret = |
| 2671 | __blk_mq_try_issue_directly(hctx, rq, false, true); |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2672 | |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2673 | if (ret == BLK_STS_RESOURCE || ret == BLK_STS_DEV_RESOURCE) |
Ming Lei | 01e99ae | 2020-02-25 09:04:32 +0800 | [diff] [blame] | 2674 | blk_mq_request_bypass_insert(rq, false, true); |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2675 | else if (ret != BLK_STS_OK) |
| 2676 | blk_mq_end_request(rq, ret); |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2677 | } |
| 2678 | |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 2679 | static blk_status_t blk_mq_request_issue_directly(struct request *rq, bool last) |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2680 | { |
Ming Lei | 4cafe86 | 2021-12-03 21:15:34 +0800 | [diff] [blame] | 2681 | return __blk_mq_try_issue_directly(rq->mq_hctx, rq, true, last); |
Christoph Hellwig | 5eb6126 | 2017-03-22 15:01:51 -0400 | [diff] [blame] | 2682 | } |
| 2683 | |
Christoph Hellwig | b84c5b5 | 2021-11-17 07:13:57 +0100 | [diff] [blame] | 2684 | static void blk_mq_plug_issue_direct(struct blk_plug *plug, bool from_schedule) |
| 2685 | { |
| 2686 | struct blk_mq_hw_ctx *hctx = NULL; |
| 2687 | struct request *rq; |
| 2688 | int queued = 0; |
| 2689 | int errors = 0; |
| 2690 | |
| 2691 | while ((rq = rq_list_pop(&plug->mq_list))) { |
| 2692 | bool last = rq_list_empty(plug->mq_list); |
| 2693 | blk_status_t ret; |
| 2694 | |
| 2695 | if (hctx != rq->mq_hctx) { |
| 2696 | if (hctx) |
| 2697 | blk_mq_commit_rqs(hctx, &queued, from_schedule); |
| 2698 | hctx = rq->mq_hctx; |
| 2699 | } |
| 2700 | |
| 2701 | ret = blk_mq_request_issue_directly(rq, last); |
| 2702 | switch (ret) { |
| 2703 | case BLK_STS_OK: |
| 2704 | queued++; |
| 2705 | break; |
| 2706 | case BLK_STS_RESOURCE: |
| 2707 | case BLK_STS_DEV_RESOURCE: |
Yufen Yu | d3b38596 | 2022-08-03 10:33:55 +0800 | [diff] [blame] | 2708 | blk_mq_request_bypass_insert(rq, false, true); |
Christoph Hellwig | b84c5b5 | 2021-11-17 07:13:57 +0100 | [diff] [blame] | 2709 | blk_mq_commit_rqs(hctx, &queued, from_schedule); |
| 2710 | return; |
| 2711 | default: |
| 2712 | blk_mq_end_request(rq, ret); |
| 2713 | errors++; |
| 2714 | break; |
| 2715 | } |
| 2716 | } |
| 2717 | |
| 2718 | /* |
| 2719 | * If we didn't flush the entire list, we could have told the driver |
| 2720 | * there was more coming, but that turned out to be a lie. |
| 2721 | */ |
| 2722 | if (errors) |
| 2723 | blk_mq_commit_rqs(hctx, &queued, from_schedule); |
| 2724 | } |
| 2725 | |
Keith Busch | 518579a | 2021-12-20 12:59:19 -0800 | [diff] [blame] | 2726 | static void __blk_mq_flush_plug_list(struct request_queue *q, |
| 2727 | struct blk_plug *plug) |
| 2728 | { |
| 2729 | if (blk_queue_quiesced(q)) |
| 2730 | return; |
| 2731 | q->mq_ops->queue_rqs(&plug->mq_list); |
| 2732 | } |
| 2733 | |
Jens Axboe | 26fed4a | 2022-03-11 10:24:17 -0700 | [diff] [blame] | 2734 | static void blk_mq_dispatch_plug_list(struct blk_plug *plug, bool from_sched) |
| 2735 | { |
| 2736 | struct blk_mq_hw_ctx *this_hctx = NULL; |
| 2737 | struct blk_mq_ctx *this_ctx = NULL; |
| 2738 | struct request *requeue_list = NULL; |
| 2739 | unsigned int depth = 0; |
| 2740 | LIST_HEAD(list); |
| 2741 | |
| 2742 | do { |
| 2743 | struct request *rq = rq_list_pop(&plug->mq_list); |
| 2744 | |
| 2745 | if (!this_hctx) { |
| 2746 | this_hctx = rq->mq_hctx; |
| 2747 | this_ctx = rq->mq_ctx; |
| 2748 | } else if (this_hctx != rq->mq_hctx || this_ctx != rq->mq_ctx) { |
| 2749 | rq_list_add(&requeue_list, rq); |
| 2750 | continue; |
| 2751 | } |
| 2752 | list_add_tail(&rq->queuelist, &list); |
| 2753 | depth++; |
| 2754 | } while (!rq_list_empty(plug->mq_list)); |
| 2755 | |
| 2756 | plug->mq_list = requeue_list; |
| 2757 | trace_block_unplug(this_hctx->queue, depth, !from_sched); |
| 2758 | blk_mq_sched_insert_requests(this_hctx, this_ctx, &list, from_sched); |
| 2759 | } |
| 2760 | |
Christoph Hellwig | b84c5b5 | 2021-11-17 07:13:57 +0100 | [diff] [blame] | 2761 | void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
| 2762 | { |
Jens Axboe | 3c67d44 | 2021-12-03 06:48:53 -0700 | [diff] [blame] | 2763 | struct request *rq; |
Christoph Hellwig | b84c5b5 | 2021-11-17 07:13:57 +0100 | [diff] [blame] | 2764 | |
| 2765 | if (rq_list_empty(plug->mq_list)) |
| 2766 | return; |
| 2767 | plug->rq_count = 0; |
| 2768 | |
| 2769 | if (!plug->multiple_queues && !plug->has_elevator && !from_schedule) { |
Jens Axboe | 3c67d44 | 2021-12-03 06:48:53 -0700 | [diff] [blame] | 2770 | struct request_queue *q; |
| 2771 | |
| 2772 | rq = rq_list_peek(&plug->mq_list); |
| 2773 | q = rq->q; |
| 2774 | |
| 2775 | /* |
| 2776 | * Peek first request and see if we have a ->queue_rqs() hook. |
| 2777 | * If we do, we can dispatch the whole plug list in one go. We |
| 2778 | * already know at this point that all requests belong to the |
| 2779 | * same queue, caller must ensure that's the case. |
| 2780 | * |
| 2781 | * Since we pass off the full list to the driver at this point, |
| 2782 | * we do not increment the active request count for the queue. |
| 2783 | * Bypass shared tags for now because of that. |
| 2784 | */ |
| 2785 | if (q->mq_ops->queue_rqs && |
| 2786 | !(rq->mq_hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) { |
| 2787 | blk_mq_run_dispatch_ops(q, |
Keith Busch | 518579a | 2021-12-20 12:59:19 -0800 | [diff] [blame] | 2788 | __blk_mq_flush_plug_list(q, plug)); |
Jens Axboe | 3c67d44 | 2021-12-03 06:48:53 -0700 | [diff] [blame] | 2789 | if (rq_list_empty(plug->mq_list)) |
| 2790 | return; |
| 2791 | } |
Ming Lei | 73f3760 | 2021-12-06 11:33:50 +0800 | [diff] [blame] | 2792 | |
| 2793 | blk_mq_run_dispatch_ops(q, |
Ming Lei | 4cafe86 | 2021-12-03 21:15:34 +0800 | [diff] [blame] | 2794 | blk_mq_plug_issue_direct(plug, false)); |
Christoph Hellwig | b84c5b5 | 2021-11-17 07:13:57 +0100 | [diff] [blame] | 2795 | if (rq_list_empty(plug->mq_list)) |
| 2796 | return; |
| 2797 | } |
| 2798 | |
Christoph Hellwig | b84c5b5 | 2021-11-17 07:13:57 +0100 | [diff] [blame] | 2799 | do { |
Jens Axboe | 26fed4a | 2022-03-11 10:24:17 -0700 | [diff] [blame] | 2800 | blk_mq_dispatch_plug_list(plug, from_schedule); |
Christoph Hellwig | b84c5b5 | 2021-11-17 07:13:57 +0100 | [diff] [blame] | 2801 | } while (!rq_list_empty(plug->mq_list)); |
Christoph Hellwig | b84c5b5 | 2021-11-17 07:13:57 +0100 | [diff] [blame] | 2802 | } |
| 2803 | |
Ming Lei | 6ce3dd6 | 2018-07-10 09:03:31 +0800 | [diff] [blame] | 2804 | void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx, |
| 2805 | struct list_head *list) |
| 2806 | { |
Keith Busch | 536167d4 | 2020-04-07 03:13:48 +0900 | [diff] [blame] | 2807 | int queued = 0; |
yangerkun | 632bfb6 | 2020-09-05 19:25:56 +0800 | [diff] [blame] | 2808 | int errors = 0; |
Keith Busch | 536167d4 | 2020-04-07 03:13:48 +0900 | [diff] [blame] | 2809 | |
Ming Lei | 6ce3dd6 | 2018-07-10 09:03:31 +0800 | [diff] [blame] | 2810 | while (!list_empty(list)) { |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2811 | blk_status_t ret; |
Ming Lei | 6ce3dd6 | 2018-07-10 09:03:31 +0800 | [diff] [blame] | 2812 | struct request *rq = list_first_entry(list, struct request, |
| 2813 | queuelist); |
| 2814 | |
| 2815 | list_del_init(&rq->queuelist); |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2816 | ret = blk_mq_request_issue_directly(rq, list_empty(list)); |
| 2817 | if (ret != BLK_STS_OK) { |
Yu Kuai | 65fac0d | 2022-07-26 20:22:24 +0800 | [diff] [blame] | 2818 | errors++; |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2819 | if (ret == BLK_STS_RESOURCE || |
| 2820 | ret == BLK_STS_DEV_RESOURCE) { |
Ming Lei | 01e99ae | 2020-02-25 09:04:32 +0800 | [diff] [blame] | 2821 | blk_mq_request_bypass_insert(rq, false, |
Jens Axboe | c616cbe | 2018-12-06 22:17:44 -0700 | [diff] [blame] | 2822 | list_empty(list)); |
Bart Van Assche | fd9c40f | 2019-04-04 10:08:43 -0700 | [diff] [blame] | 2823 | break; |
| 2824 | } |
| 2825 | blk_mq_end_request(rq, ret); |
Keith Busch | 536167d4 | 2020-04-07 03:13:48 +0900 | [diff] [blame] | 2826 | } else |
| 2827 | queued++; |
Ming Lei | 6ce3dd6 | 2018-07-10 09:03:31 +0800 | [diff] [blame] | 2828 | } |
Jens Axboe | d666ba9 | 2018-11-27 17:02:25 -0700 | [diff] [blame] | 2829 | |
| 2830 | /* |
| 2831 | * If we didn't flush the entire list, we could have told |
| 2832 | * the driver there was more coming, but that turned out to |
| 2833 | * be a lie. |
| 2834 | */ |
yangerkun | 632bfb6 | 2020-09-05 19:25:56 +0800 | [diff] [blame] | 2835 | if ((!list_empty(list) || errors) && |
| 2836 | hctx->queue->mq_ops->commit_rqs && queued) |
Jens Axboe | d666ba9 | 2018-11-27 17:02:25 -0700 | [diff] [blame] | 2837 | hctx->queue->mq_ops->commit_rqs(hctx); |
Ming Lei | 6ce3dd6 | 2018-07-10 09:03:31 +0800 | [diff] [blame] | 2838 | } |
| 2839 | |
Ming Lei | b131f20 | 2021-11-11 16:51:34 +0800 | [diff] [blame] | 2840 | static bool blk_mq_attempt_bio_merge(struct request_queue *q, |
Christoph Hellwig | 0c5bcc9 | 2021-11-23 17:04:41 +0100 | [diff] [blame] | 2841 | struct bio *bio, unsigned int nr_segs) |
Jens Axboe | 900e080 | 2021-11-03 05:47:09 -0600 | [diff] [blame] | 2842 | { |
| 2843 | if (!blk_queue_nomerges(q) && bio_mergeable(bio)) { |
Christoph Hellwig | 0c5bcc9 | 2021-11-23 17:04:41 +0100 | [diff] [blame] | 2844 | if (blk_attempt_plug_merge(q, bio, nr_segs)) |
Jens Axboe | 900e080 | 2021-11-03 05:47:09 -0600 | [diff] [blame] | 2845 | return true; |
| 2846 | if (blk_mq_sched_bio_merge(q, bio, nr_segs)) |
| 2847 | return true; |
| 2848 | } |
| 2849 | return false; |
| 2850 | } |
| 2851 | |
Jens Axboe | 7153971 | 2021-11-03 05:52:45 -0600 | [diff] [blame] | 2852 | static struct request *blk_mq_get_new_requests(struct request_queue *q, |
| 2853 | struct blk_plug *plug, |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2854 | struct bio *bio, |
| 2855 | unsigned int nsegs) |
Jens Axboe | 7153971 | 2021-11-03 05:52:45 -0600 | [diff] [blame] | 2856 | { |
| 2857 | struct blk_mq_alloc_data data = { |
| 2858 | .q = q, |
| 2859 | .nr_tags = 1, |
Ming Lei | 9d497e2 | 2022-01-04 21:42:23 +0800 | [diff] [blame] | 2860 | .cmd_flags = bio->bi_opf, |
Jens Axboe | 7153971 | 2021-11-03 05:52:45 -0600 | [diff] [blame] | 2861 | }; |
| 2862 | struct request *rq; |
| 2863 | |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2864 | if (unlikely(bio_queue_enter(bio))) |
Ming Lei | b637108 | 2021-11-12 20:47:15 +0800 | [diff] [blame] | 2865 | return NULL; |
Jens Axboe | 900e080 | 2021-11-03 05:47:09 -0600 | [diff] [blame] | 2866 | |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2867 | if (blk_mq_attempt_bio_merge(q, bio, nsegs)) |
| 2868 | goto queue_exit; |
| 2869 | |
| 2870 | rq_qos_throttle(q, bio); |
| 2871 | |
Jens Axboe | 7153971 | 2021-11-03 05:52:45 -0600 | [diff] [blame] | 2872 | if (plug) { |
| 2873 | data.nr_tags = plug->nr_ios; |
| 2874 | plug->nr_ios = 1; |
| 2875 | data.cached_rq = &plug->cached_rq; |
| 2876 | } |
| 2877 | |
| 2878 | rq = __blk_mq_alloc_requests(&data); |
Jens Axboe | 373b541 | 2021-12-02 12:42:58 -0700 | [diff] [blame] | 2879 | if (rq) |
| 2880 | return rq; |
Jens Axboe | 7153971 | 2021-11-03 05:52:45 -0600 | [diff] [blame] | 2881 | rq_qos_cleanup(q, bio); |
| 2882 | if (bio->bi_opf & REQ_NOWAIT) |
| 2883 | bio_wouldblock_error(bio); |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2884 | queue_exit: |
Jens Axboe | 95febeb | 2021-11-15 14:23:08 -0700 | [diff] [blame] | 2885 | blk_queue_exit(q); |
| 2886 | return NULL; |
Jens Axboe | 7153971 | 2021-11-03 05:52:45 -0600 | [diff] [blame] | 2887 | } |
| 2888 | |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2889 | static inline struct request *blk_mq_get_cached_request(struct request_queue *q, |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2890 | struct blk_plug *plug, struct bio **bio, unsigned int nsegs) |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2891 | { |
| 2892 | struct request *rq; |
| 2893 | |
| 2894 | if (!plug) |
| 2895 | return NULL; |
| 2896 | rq = rq_list_peek(&plug->cached_rq); |
| 2897 | if (!rq || rq->q != q) |
| 2898 | return NULL; |
| 2899 | |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2900 | if (blk_mq_attempt_bio_merge(q, *bio, nsegs)) { |
| 2901 | *bio = NULL; |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2902 | return NULL; |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2903 | } |
| 2904 | |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2905 | if (blk_mq_get_hctx_type((*bio)->bi_opf) != rq->mq_hctx->type) |
| 2906 | return NULL; |
| 2907 | if (op_is_flush(rq->cmd_flags) != op_is_flush((*bio)->bi_opf)) |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2908 | return NULL; |
| 2909 | |
Jens Axboe | 2645672 | 2022-06-21 10:03:57 -0600 | [diff] [blame] | 2910 | /* |
| 2911 | * If any qos ->throttle() end up blocking, we will have flushed the |
| 2912 | * plug and hence killed the cached_rq list as well. Pop this entry |
| 2913 | * before we throttle. |
| 2914 | */ |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2915 | plug->cached_rq = rq_list_next(rq); |
Jens Axboe | 2645672 | 2022-06-21 10:03:57 -0600 | [diff] [blame] | 2916 | rq_qos_throttle(q, *bio); |
| 2917 | |
| 2918 | rq->cmd_flags = (*bio)->bi_opf; |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2919 | INIT_LIST_HEAD(&rq->queuelist); |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2920 | return rq; |
| 2921 | } |
| 2922 | |
Jan Kara | 82b74ca | 2022-06-23 09:48:32 +0200 | [diff] [blame] | 2923 | static void bio_set_ioprio(struct bio *bio) |
| 2924 | { |
Jan Kara | a78418e | 2022-06-23 09:48:34 +0200 | [diff] [blame] | 2925 | /* Nobody set ioprio so far? Initialize it based on task's nice value */ |
| 2926 | if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE) |
| 2927 | bio->bi_ioprio = get_current_ioprio(); |
Jan Kara | 82b74ca | 2022-06-23 09:48:32 +0200 | [diff] [blame] | 2928 | blkcg_set_ioprio(bio); |
| 2929 | } |
| 2930 | |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2931 | /** |
Christoph Hellwig | c62b37d | 2020-07-01 10:59:43 +0200 | [diff] [blame] | 2932 | * blk_mq_submit_bio - Create and send a request to block device. |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2933 | * @bio: Bio pointer. |
| 2934 | * |
| 2935 | * Builds up a request structure from @q and @bio and send to the device. The |
| 2936 | * request may not be queued directly to hardware if: |
| 2937 | * * This request can be merged with another one |
| 2938 | * * We want to place request at plug queue for possible future merging |
| 2939 | * * There is an IO scheduler active at this queue |
| 2940 | * |
| 2941 | * It will not queue the request if there is an error with the bio, or at the |
| 2942 | * request creation. |
André Almeida | 105663f | 2020-01-06 15:08:18 -0300 | [diff] [blame] | 2943 | */ |
Christoph Hellwig | 3e08773 | 2021-10-12 13:12:24 +0200 | [diff] [blame] | 2944 | void blk_mq_submit_bio(struct bio *bio) |
Jens Axboe | 07068d5 | 2014-05-22 10:40:51 -0600 | [diff] [blame] | 2945 | { |
Pavel Begunkov | ed6cdde | 2021-10-14 15:03:30 +0100 | [diff] [blame] | 2946 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
Christoph Hellwig | 6deacb3 | 2022-07-06 09:03:38 +0200 | [diff] [blame] | 2947 | struct blk_plug *plug = blk_mq_plug(bio); |
Christoph Hellwig | ef295ec | 2016-10-28 08:48:16 -0600 | [diff] [blame] | 2948 | const int is_sync = op_is_sync(bio->bi_opf); |
Jens Axboe | 07068d5 | 2014-05-22 10:40:51 -0600 | [diff] [blame] | 2949 | struct request *rq; |
Jens Axboe | abd45c1 | 2021-10-13 12:43:41 -0600 | [diff] [blame] | 2950 | unsigned int nr_segs = 1; |
Satya Tangirala | a892c8d | 2020-05-14 00:37:18 +0000 | [diff] [blame] | 2951 | blk_status_t ret; |
Jens Axboe | 07068d5 | 2014-05-22 10:40:51 -0600 | [diff] [blame] | 2952 | |
Christoph Hellwig | 51d798c | 2022-07-27 12:22:56 -0400 | [diff] [blame] | 2953 | bio = blk_queue_bounce(bio, q); |
Jens Axboe | 613b148 | 2023-01-04 08:51:19 -0700 | [diff] [blame] | 2954 | if (bio_may_exceed_limits(bio, &q->limits)) { |
Christoph Hellwig | c55ddd9 | 2022-07-27 12:23:00 -0400 | [diff] [blame] | 2955 | bio = __bio_split_to_limits(bio, &q->limits, &nr_segs); |
Jens Axboe | 613b148 | 2023-01-04 08:51:19 -0700 | [diff] [blame] | 2956 | if (!bio) |
| 2957 | return; |
| 2958 | } |
Wen Xiong | f36ea50 | 2017-05-10 08:54:11 -0500 | [diff] [blame] | 2959 | |
Dmitry Monakhov | e23947b | 2017-06-29 11:31:11 -0700 | [diff] [blame] | 2960 | if (!bio_integrity_prep(bio)) |
Jens Axboe | 900e080 | 2021-11-03 05:47:09 -0600 | [diff] [blame] | 2961 | return; |
Jens Axboe | 87760e5 | 2016-11-09 12:38:14 -0700 | [diff] [blame] | 2962 | |
Jan Kara | 9c6227e | 2022-06-23 09:48:33 +0200 | [diff] [blame] | 2963 | bio_set_ioprio(bio); |
| 2964 | |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2965 | rq = blk_mq_get_cached_request(q, plug, &bio, nr_segs); |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2966 | if (!rq) { |
Shin'ichiro Kawasaki | 0a5aa8d | 2022-03-08 17:09:15 +0900 | [diff] [blame] | 2967 | if (!bio) |
| 2968 | return; |
| 2969 | rq = blk_mq_get_new_requests(q, plug, bio, nr_segs); |
Christoph Hellwig | 5b13bc8 | 2021-11-24 07:28:56 +0100 | [diff] [blame] | 2970 | if (unlikely(!rq)) |
| 2971 | return; |
| 2972 | } |
Jens Axboe | 87760e5 | 2016-11-09 12:38:14 -0700 | [diff] [blame] | 2973 | |
Christoph Hellwig | e8a676d | 2020-12-03 17:21:36 +0100 | [diff] [blame] | 2974 | trace_block_getrq(bio); |
Xiaoguang Wang | d6f1dda | 2018-10-23 22:30:50 +0800 | [diff] [blame] | 2975 | |
Josef Bacik | c1c8038 | 2018-07-03 11:14:59 -0400 | [diff] [blame] | 2976 | rq_qos_track(q, rq, bio); |
Jens Axboe | 07068d5 | 2014-05-22 10:40:51 -0600 | [diff] [blame] | 2977 | |
Bart Van Assche | 970d168 | 2019-07-01 08:47:30 -0700 | [diff] [blame] | 2978 | blk_mq_bio_to_request(rq, bio, nr_segs); |
| 2979 | |
Satya Tangirala | a892c8d | 2020-05-14 00:37:18 +0000 | [diff] [blame] | 2980 | ret = blk_crypto_init_request(rq); |
| 2981 | if (ret != BLK_STS_OK) { |
| 2982 | bio->bi_status = ret; |
| 2983 | bio_endio(bio); |
| 2984 | blk_mq_free_request(rq); |
Christoph Hellwig | 3e08773 | 2021-10-12 13:12:24 +0200 | [diff] [blame] | 2985 | return; |
Satya Tangirala | a892c8d | 2020-05-14 00:37:18 +0000 | [diff] [blame] | 2986 | } |
| 2987 | |
Ming Lei | 2b504bd | 2021-11-18 23:30:41 +0800 | [diff] [blame] | 2988 | if (op_is_flush(bio->bi_opf)) { |
| 2989 | blk_insert_flush(rq); |
Christoph Hellwig | d92ca9d8 | 2021-10-19 14:25:53 +0200 | [diff] [blame] | 2990 | return; |
Ming Lei | 2b504bd | 2021-11-18 23:30:41 +0800 | [diff] [blame] | 2991 | } |
Christoph Hellwig | d92ca9d8 | 2021-10-19 14:25:53 +0200 | [diff] [blame] | 2992 | |
Christoph Hellwig | 1e9c230 | 2021-11-23 17:04:42 +0100 | [diff] [blame] | 2993 | if (plug) |
Jens Axboe | ce5b009 | 2018-11-27 17:13:56 -0700 | [diff] [blame] | 2994 | blk_add_rq_to_plug(plug, rq); |
Christoph Hellwig | 1e9c230 | 2021-11-23 17:04:42 +0100 | [diff] [blame] | 2995 | else if ((rq->rq_flags & RQF_ELV) || |
| 2996 | (rq->mq_hctx->dispatch_busy && |
| 2997 | (q->nr_hw_queues == 1 || !is_sync))) |
Ming Lei | a12de1d | 2019-09-27 15:24:30 +0800 | [diff] [blame] | 2998 | blk_mq_sched_insert_request(rq, false, true, true); |
Christoph Hellwig | 1e9c230 | 2021-11-23 17:04:42 +0100 | [diff] [blame] | 2999 | else |
Ming Lei | bcc330f | 2021-12-03 21:15:33 +0800 | [diff] [blame] | 3000 | blk_mq_run_dispatch_ops(rq->q, |
Ming Lei | 2a904d0 | 2021-12-03 21:15:31 +0800 | [diff] [blame] | 3001 | blk_mq_try_issue_directly(rq->mq_hctx, rq)); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3002 | } |
| 3003 | |
Christoph Hellwig | 248c793 | 2022-02-15 11:05:36 +0100 | [diff] [blame] | 3004 | #ifdef CONFIG_BLK_MQ_STACKING |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3005 | /** |
Christoph Hellwig | a5efda3 | 2022-02-15 11:05:37 +0100 | [diff] [blame] | 3006 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request |
Christoph Hellwig | a5efda3 | 2022-02-15 11:05:37 +0100 | [diff] [blame] | 3007 | * @rq: the request being queued |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3008 | */ |
Christoph Hellwig | 28db471 | 2022-02-15 11:05:38 +0100 | [diff] [blame] | 3009 | blk_status_t blk_insert_cloned_request(struct request *rq) |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3010 | { |
Christoph Hellwig | 28db471 | 2022-02-15 11:05:38 +0100 | [diff] [blame] | 3011 | struct request_queue *q = rq->q; |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3012 | unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); |
Christoph Hellwig | a5efda3 | 2022-02-15 11:05:37 +0100 | [diff] [blame] | 3013 | blk_status_t ret; |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3014 | |
| 3015 | if (blk_rq_sectors(rq) > max_sectors) { |
| 3016 | /* |
| 3017 | * SCSI device does not have a good way to return if |
| 3018 | * Write Same/Zero is actually supported. If a device rejects |
| 3019 | * a non-read/write command (discard, write same,etc.) the |
| 3020 | * low-level device driver will set the relevant queue limit to |
| 3021 | * 0 to prevent blk-lib from issuing more of the offending |
| 3022 | * operations. Commands queued prior to the queue limit being |
| 3023 | * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O |
| 3024 | * errors being propagated to upper layers. |
| 3025 | */ |
| 3026 | if (max_sectors == 0) |
| 3027 | return BLK_STS_NOTSUPP; |
| 3028 | |
| 3029 | printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", |
| 3030 | __func__, blk_rq_sectors(rq), max_sectors); |
| 3031 | return BLK_STS_IOERR; |
| 3032 | } |
| 3033 | |
| 3034 | /* |
| 3035 | * The queue settings related to segment counting may differ from the |
| 3036 | * original queue. |
| 3037 | */ |
| 3038 | rq->nr_phys_segments = blk_recalc_rq_segments(rq); |
| 3039 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
| 3040 | printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", |
| 3041 | __func__, rq->nr_phys_segments, queue_max_segments(q)); |
| 3042 | return BLK_STS_IOERR; |
| 3043 | } |
| 3044 | |
Christoph Hellwig | 28db471 | 2022-02-15 11:05:38 +0100 | [diff] [blame] | 3045 | if (q->disk && should_fail_request(q->disk->part0, blk_rq_bytes(rq))) |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3046 | return BLK_STS_IOERR; |
| 3047 | |
| 3048 | if (blk_crypto_insert_cloned_request(rq)) |
| 3049 | return BLK_STS_IOERR; |
| 3050 | |
| 3051 | blk_account_io_start(rq); |
| 3052 | |
| 3053 | /* |
| 3054 | * Since we have a scheduler attached on the top device, |
| 3055 | * bypass a potential scheduler on the bottom device for |
| 3056 | * insert. |
| 3057 | */ |
Christoph Hellwig | 28db471 | 2022-02-15 11:05:38 +0100 | [diff] [blame] | 3058 | blk_mq_run_dispatch_ops(q, |
Ming Lei | 4cafe86 | 2021-12-03 21:15:34 +0800 | [diff] [blame] | 3059 | ret = blk_mq_request_issue_directly(rq, true)); |
Yu Kuai | 592ee11 | 2022-01-26 09:21:32 +0800 | [diff] [blame] | 3060 | if (ret) |
| 3061 | blk_account_io_done(rq, ktime_get_ns()); |
Ming Lei | 4cafe86 | 2021-12-03 21:15:34 +0800 | [diff] [blame] | 3062 | return ret; |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3063 | } |
| 3064 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); |
| 3065 | |
| 3066 | /** |
| 3067 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request |
| 3068 | * @rq: the clone request to be cleaned up |
| 3069 | * |
| 3070 | * Description: |
| 3071 | * Free all bios in @rq for a cloned request. |
| 3072 | */ |
| 3073 | void blk_rq_unprep_clone(struct request *rq) |
| 3074 | { |
| 3075 | struct bio *bio; |
| 3076 | |
| 3077 | while ((bio = rq->bio) != NULL) { |
| 3078 | rq->bio = bio->bi_next; |
| 3079 | |
| 3080 | bio_put(bio); |
| 3081 | } |
| 3082 | } |
| 3083 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); |
| 3084 | |
| 3085 | /** |
| 3086 | * blk_rq_prep_clone - Helper function to setup clone request |
| 3087 | * @rq: the request to be setup |
| 3088 | * @rq_src: original request to be cloned |
| 3089 | * @bs: bio_set that bios for clone are allocated from |
| 3090 | * @gfp_mask: memory allocation mask for bio |
| 3091 | * @bio_ctr: setup function to be called for each clone bio. |
| 3092 | * Returns %0 for success, non %0 for failure. |
| 3093 | * @data: private data to be passed to @bio_ctr |
| 3094 | * |
| 3095 | * Description: |
| 3096 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. |
| 3097 | * Also, pages which the original bios are pointing to are not copied |
| 3098 | * and the cloned bios just point same pages. |
| 3099 | * So cloned bios must be completed before original bios, which means |
| 3100 | * the caller must complete @rq before @rq_src. |
| 3101 | */ |
| 3102 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, |
| 3103 | struct bio_set *bs, gfp_t gfp_mask, |
| 3104 | int (*bio_ctr)(struct bio *, struct bio *, void *), |
| 3105 | void *data) |
| 3106 | { |
| 3107 | struct bio *bio, *bio_src; |
| 3108 | |
| 3109 | if (!bs) |
| 3110 | bs = &fs_bio_set; |
| 3111 | |
| 3112 | __rq_for_each_bio(bio_src, rq_src) { |
Christoph Hellwig | abfc426 | 2022-02-02 17:01:09 +0100 | [diff] [blame] | 3113 | bio = bio_alloc_clone(rq->q->disk->part0, bio_src, gfp_mask, |
| 3114 | bs); |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3115 | if (!bio) |
| 3116 | goto free_and_out; |
| 3117 | |
| 3118 | if (bio_ctr && bio_ctr(bio, bio_src, data)) |
| 3119 | goto free_and_out; |
| 3120 | |
| 3121 | if (rq->bio) { |
| 3122 | rq->biotail->bi_next = bio; |
| 3123 | rq->biotail = bio; |
| 3124 | } else { |
| 3125 | rq->bio = rq->biotail = bio; |
| 3126 | } |
| 3127 | bio = NULL; |
| 3128 | } |
| 3129 | |
| 3130 | /* Copy attributes of the original request to the clone request. */ |
| 3131 | rq->__sector = blk_rq_pos(rq_src); |
| 3132 | rq->__data_len = blk_rq_bytes(rq_src); |
| 3133 | if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) { |
| 3134 | rq->rq_flags |= RQF_SPECIAL_PAYLOAD; |
| 3135 | rq->special_vec = rq_src->special_vec; |
| 3136 | } |
| 3137 | rq->nr_phys_segments = rq_src->nr_phys_segments; |
| 3138 | rq->ioprio = rq_src->ioprio; |
| 3139 | |
| 3140 | if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0) |
| 3141 | goto free_and_out; |
| 3142 | |
| 3143 | return 0; |
| 3144 | |
| 3145 | free_and_out: |
| 3146 | if (bio) |
| 3147 | bio_put(bio); |
| 3148 | blk_rq_unprep_clone(rq); |
| 3149 | |
| 3150 | return -ENOMEM; |
| 3151 | } |
| 3152 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); |
Christoph Hellwig | 248c793 | 2022-02-15 11:05:36 +0100 | [diff] [blame] | 3153 | #endif /* CONFIG_BLK_MQ_STACKING */ |
Christoph Hellwig | 06c8c69 | 2021-11-17 07:13:58 +0100 | [diff] [blame] | 3154 | |
Christoph Hellwig | f2b8f3c | 2021-11-17 07:14:00 +0100 | [diff] [blame] | 3155 | /* |
| 3156 | * Steal bios from a request and add them to a bio list. |
| 3157 | * The request must not have been partially completed before. |
| 3158 | */ |
| 3159 | void blk_steal_bios(struct bio_list *list, struct request *rq) |
| 3160 | { |
| 3161 | if (rq->bio) { |
| 3162 | if (list->tail) |
| 3163 | list->tail->bi_next = rq->bio; |
| 3164 | else |
| 3165 | list->head = rq->bio; |
| 3166 | list->tail = rq->biotail; |
| 3167 | |
| 3168 | rq->bio = NULL; |
| 3169 | rq->biotail = NULL; |
| 3170 | } |
| 3171 | |
| 3172 | rq->__data_len = 0; |
| 3173 | } |
| 3174 | EXPORT_SYMBOL_GPL(blk_steal_bios); |
| 3175 | |
Ming Lei | bd63141 | 2021-05-11 23:22:35 +0800 | [diff] [blame] | 3176 | static size_t order_to_size(unsigned int order) |
| 3177 | { |
| 3178 | return (size_t)PAGE_SIZE << order; |
| 3179 | } |
| 3180 | |
| 3181 | /* called before freeing request pool in @tags */ |
John Garry | f32e4ea | 2021-10-05 18:23:32 +0800 | [diff] [blame] | 3182 | static void blk_mq_clear_rq_mapping(struct blk_mq_tags *drv_tags, |
| 3183 | struct blk_mq_tags *tags) |
Ming Lei | bd63141 | 2021-05-11 23:22:35 +0800 | [diff] [blame] | 3184 | { |
Ming Lei | bd63141 | 2021-05-11 23:22:35 +0800 | [diff] [blame] | 3185 | struct page *page; |
| 3186 | unsigned long flags; |
| 3187 | |
Yu Kuai | 76dd298 | 2022-10-11 22:22:53 +0800 | [diff] [blame] | 3188 | /* |
| 3189 | * There is no need to clear mapping if driver tags is not initialized |
| 3190 | * or the mapping belongs to the driver tags. |
| 3191 | */ |
| 3192 | if (!drv_tags || drv_tags == tags) |
John Garry | 4f245d5 | 2021-10-05 18:23:33 +0800 | [diff] [blame] | 3193 | return; |
| 3194 | |
Ming Lei | bd63141 | 2021-05-11 23:22:35 +0800 | [diff] [blame] | 3195 | list_for_each_entry(page, &tags->page_list, lru) { |
| 3196 | unsigned long start = (unsigned long)page_address(page); |
| 3197 | unsigned long end = start + order_to_size(page->private); |
| 3198 | int i; |
| 3199 | |
John Garry | f32e4ea | 2021-10-05 18:23:32 +0800 | [diff] [blame] | 3200 | for (i = 0; i < drv_tags->nr_tags; i++) { |
Ming Lei | bd63141 | 2021-05-11 23:22:35 +0800 | [diff] [blame] | 3201 | struct request *rq = drv_tags->rqs[i]; |
| 3202 | unsigned long rq_addr = (unsigned long)rq; |
| 3203 | |
| 3204 | if (rq_addr >= start && rq_addr < end) { |
Jens Axboe | 0a467d0 | 2021-10-14 14:39:59 -0600 | [diff] [blame] | 3205 | WARN_ON_ONCE(req_ref_read(rq) != 0); |
Ming Lei | bd63141 | 2021-05-11 23:22:35 +0800 | [diff] [blame] | 3206 | cmpxchg(&drv_tags->rqs[i], rq, NULL); |
| 3207 | } |
| 3208 | } |
| 3209 | } |
| 3210 | |
| 3211 | /* |
| 3212 | * Wait until all pending iteration is done. |
| 3213 | * |
| 3214 | * Request reference is cleared and it is guaranteed to be observed |
| 3215 | * after the ->lock is released. |
| 3216 | */ |
| 3217 | spin_lock_irqsave(&drv_tags->lock, flags); |
| 3218 | spin_unlock_irqrestore(&drv_tags->lock, flags); |
| 3219 | } |
| 3220 | |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3221 | void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags, |
| 3222 | unsigned int hctx_idx) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3223 | { |
John Garry | f32e4ea | 2021-10-05 18:23:32 +0800 | [diff] [blame] | 3224 | struct blk_mq_tags *drv_tags; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3225 | struct page *page; |
| 3226 | |
Ming Lei | e02657e | 2022-03-08 06:51:48 +0100 | [diff] [blame] | 3227 | if (list_empty(&tags->page_list)) |
| 3228 | return; |
| 3229 | |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 3230 | if (blk_mq_is_shared_tags(set->flags)) |
| 3231 | drv_tags = set->shared_tags; |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3232 | else |
| 3233 | drv_tags = set->tags[hctx_idx]; |
John Garry | f32e4ea | 2021-10-05 18:23:32 +0800 | [diff] [blame] | 3234 | |
John Garry | 65de57b | 2021-10-05 18:23:26 +0800 | [diff] [blame] | 3235 | if (tags->static_rqs && set->ops->exit_request) { |
Christoph Hellwig | e9b267d | 2014-04-15 13:59:10 -0600 | [diff] [blame] | 3236 | int i; |
| 3237 | |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3238 | for (i = 0; i < tags->nr_tags; i++) { |
Jens Axboe | 2af8cbe | 2017-01-13 14:39:30 -0700 | [diff] [blame] | 3239 | struct request *rq = tags->static_rqs[i]; |
| 3240 | |
| 3241 | if (!rq) |
Christoph Hellwig | e9b267d | 2014-04-15 13:59:10 -0600 | [diff] [blame] | 3242 | continue; |
Christoph Hellwig | d6296d39 | 2017-05-01 10:19:08 -0600 | [diff] [blame] | 3243 | set->ops->exit_request(set, rq, hctx_idx); |
Jens Axboe | 2af8cbe | 2017-01-13 14:39:30 -0700 | [diff] [blame] | 3244 | tags->static_rqs[i] = NULL; |
Christoph Hellwig | e9b267d | 2014-04-15 13:59:10 -0600 | [diff] [blame] | 3245 | } |
| 3246 | } |
| 3247 | |
John Garry | f32e4ea | 2021-10-05 18:23:32 +0800 | [diff] [blame] | 3248 | blk_mq_clear_rq_mapping(drv_tags, tags); |
Ming Lei | bd63141 | 2021-05-11 23:22:35 +0800 | [diff] [blame] | 3249 | |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3250 | while (!list_empty(&tags->page_list)) { |
| 3251 | page = list_first_entry(&tags->page_list, struct page, lru); |
Dave Hansen | 6753471 | 2014-01-08 20:17:46 -0700 | [diff] [blame] | 3252 | list_del_init(&page->lru); |
Catalin Marinas | f75782e | 2015-09-14 18:16:02 +0100 | [diff] [blame] | 3253 | /* |
| 3254 | * Remove kmemleak object previously allocated in |
Raul E Rangel | 273938b | 2019-05-02 13:48:11 -0600 | [diff] [blame] | 3255 | * blk_mq_alloc_rqs(). |
Catalin Marinas | f75782e | 2015-09-14 18:16:02 +0100 | [diff] [blame] | 3256 | */ |
| 3257 | kmemleak_free(page_address(page)); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3258 | __free_pages(page, page->private); |
| 3259 | } |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3260 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3261 | |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3262 | void blk_mq_free_rq_map(struct blk_mq_tags *tags) |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3263 | { |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3264 | kfree(tags->rqs); |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3265 | tags->rqs = NULL; |
Jens Axboe | 2af8cbe | 2017-01-13 14:39:30 -0700 | [diff] [blame] | 3266 | kfree(tags->static_rqs); |
| 3267 | tags->static_rqs = NULL; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3268 | |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3269 | blk_mq_free_tags(tags); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3270 | } |
| 3271 | |
Ming Lei | 4d80513 | 2022-03-08 15:32:14 +0800 | [diff] [blame] | 3272 | static enum hctx_type hctx_idx_to_type(struct blk_mq_tag_set *set, |
| 3273 | unsigned int hctx_idx) |
| 3274 | { |
| 3275 | int i; |
| 3276 | |
| 3277 | for (i = 0; i < set->nr_maps; i++) { |
| 3278 | unsigned int start = set->map[i].queue_offset; |
| 3279 | unsigned int end = start + set->map[i].nr_queues; |
| 3280 | |
| 3281 | if (hctx_idx >= start && hctx_idx < end) |
| 3282 | break; |
| 3283 | } |
| 3284 | |
| 3285 | if (i >= set->nr_maps) |
| 3286 | i = HCTX_TYPE_DEFAULT; |
| 3287 | |
| 3288 | return i; |
| 3289 | } |
| 3290 | |
| 3291 | static int blk_mq_get_hctx_node(struct blk_mq_tag_set *set, |
| 3292 | unsigned int hctx_idx) |
| 3293 | { |
| 3294 | enum hctx_type type = hctx_idx_to_type(set, hctx_idx); |
| 3295 | |
| 3296 | return blk_mq_hw_queue_to_node(&set->map[type], hctx_idx); |
| 3297 | } |
| 3298 | |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3299 | static struct blk_mq_tags *blk_mq_alloc_rq_map(struct blk_mq_tag_set *set, |
| 3300 | unsigned int hctx_idx, |
| 3301 | unsigned int nr_tags, |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3302 | unsigned int reserved_tags) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3303 | { |
Ming Lei | 4d80513 | 2022-03-08 15:32:14 +0800 | [diff] [blame] | 3304 | int node = blk_mq_get_hctx_node(set, hctx_idx); |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3305 | struct blk_mq_tags *tags; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3306 | |
Shaohua Li | 59f082e | 2017-02-01 09:53:14 -0800 | [diff] [blame] | 3307 | if (node == NUMA_NO_NODE) |
| 3308 | node = set->numa_node; |
| 3309 | |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3310 | tags = blk_mq_init_tags(nr_tags, reserved_tags, node, |
| 3311 | BLK_MQ_FLAG_TO_ALLOC_POLICY(set->flags)); |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3312 | if (!tags) |
| 3313 | return NULL; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3314 | |
Kees Cook | 590b5b7 | 2018-06-12 14:04:20 -0700 | [diff] [blame] | 3315 | tags->rqs = kcalloc_node(nr_tags, sizeof(struct request *), |
Gabriel Krisman Bertazi | 36e1f3d1 | 2016-12-06 13:31:44 -0200 | [diff] [blame] | 3316 | GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY, |
Shaohua Li | 59f082e | 2017-02-01 09:53:14 -0800 | [diff] [blame] | 3317 | node); |
Jinlong Chen | 7edfd68 | 2022-11-02 10:52:29 +0800 | [diff] [blame] | 3318 | if (!tags->rqs) |
| 3319 | goto err_free_tags; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3320 | |
Kees Cook | 590b5b7 | 2018-06-12 14:04:20 -0700 | [diff] [blame] | 3321 | tags->static_rqs = kcalloc_node(nr_tags, sizeof(struct request *), |
| 3322 | GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY, |
| 3323 | node); |
Jinlong Chen | 7edfd68 | 2022-11-02 10:52:29 +0800 | [diff] [blame] | 3324 | if (!tags->static_rqs) |
| 3325 | goto err_free_rqs; |
Jens Axboe | 2af8cbe | 2017-01-13 14:39:30 -0700 | [diff] [blame] | 3326 | |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3327 | return tags; |
Jinlong Chen | 7edfd68 | 2022-11-02 10:52:29 +0800 | [diff] [blame] | 3328 | |
| 3329 | err_free_rqs: |
| 3330 | kfree(tags->rqs); |
| 3331 | err_free_tags: |
| 3332 | blk_mq_free_tags(tags); |
| 3333 | return NULL; |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3334 | } |
| 3335 | |
Tejun Heo | 1d9bd51 | 2018-01-09 08:29:48 -0800 | [diff] [blame] | 3336 | static int blk_mq_init_request(struct blk_mq_tag_set *set, struct request *rq, |
| 3337 | unsigned int hctx_idx, int node) |
| 3338 | { |
| 3339 | int ret; |
| 3340 | |
| 3341 | if (set->ops->init_request) { |
| 3342 | ret = set->ops->init_request(set, rq, hctx_idx, node); |
| 3343 | if (ret) |
| 3344 | return ret; |
| 3345 | } |
| 3346 | |
Keith Busch | 12f5b93 | 2018-05-29 15:52:28 +0200 | [diff] [blame] | 3347 | WRITE_ONCE(rq->state, MQ_RQ_IDLE); |
Tejun Heo | 1d9bd51 | 2018-01-09 08:29:48 -0800 | [diff] [blame] | 3348 | return 0; |
| 3349 | } |
| 3350 | |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3351 | static int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, |
| 3352 | struct blk_mq_tags *tags, |
| 3353 | unsigned int hctx_idx, unsigned int depth) |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3354 | { |
| 3355 | unsigned int i, j, entries_per_page, max_order = 4; |
Ming Lei | 4d80513 | 2022-03-08 15:32:14 +0800 | [diff] [blame] | 3356 | int node = blk_mq_get_hctx_node(set, hctx_idx); |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3357 | size_t rq_size, left; |
Shaohua Li | 59f082e | 2017-02-01 09:53:14 -0800 | [diff] [blame] | 3358 | |
Shaohua Li | 59f082e | 2017-02-01 09:53:14 -0800 | [diff] [blame] | 3359 | if (node == NUMA_NO_NODE) |
| 3360 | node = set->numa_node; |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3361 | |
| 3362 | INIT_LIST_HEAD(&tags->page_list); |
| 3363 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3364 | /* |
| 3365 | * rq_size is the size of the request plus driver payload, rounded |
| 3366 | * to the cacheline size |
| 3367 | */ |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3368 | rq_size = round_up(sizeof(struct request) + set->cmd_size, |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3369 | cache_line_size()); |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3370 | left = rq_size * depth; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3371 | |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3372 | for (i = 0; i < depth; ) { |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3373 | int this_order = max_order; |
| 3374 | struct page *page; |
| 3375 | int to_do; |
| 3376 | void *p; |
| 3377 | |
Bartlomiej Zolnierkiewicz | b3a834b | 2016-05-16 09:54:47 -0600 | [diff] [blame] | 3378 | while (this_order && left < order_to_size(this_order - 1)) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3379 | this_order--; |
| 3380 | |
| 3381 | do { |
Shaohua Li | 59f082e | 2017-02-01 09:53:14 -0800 | [diff] [blame] | 3382 | page = alloc_pages_node(node, |
Gabriel Krisman Bertazi | 36e1f3d1 | 2016-12-06 13:31:44 -0200 | [diff] [blame] | 3383 | GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO, |
Jens Axboe | a516440 | 2014-09-10 09:02:03 -0600 | [diff] [blame] | 3384 | this_order); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3385 | if (page) |
| 3386 | break; |
| 3387 | if (!this_order--) |
| 3388 | break; |
| 3389 | if (order_to_size(this_order) < rq_size) |
| 3390 | break; |
| 3391 | } while (1); |
| 3392 | |
| 3393 | if (!page) |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3394 | goto fail; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3395 | |
| 3396 | page->private = this_order; |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3397 | list_add_tail(&page->lru, &tags->page_list); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3398 | |
| 3399 | p = page_address(page); |
Catalin Marinas | f75782e | 2015-09-14 18:16:02 +0100 | [diff] [blame] | 3400 | /* |
| 3401 | * Allow kmemleak to scan these pages as they contain pointers |
| 3402 | * to additional allocations like via ops->init_request(). |
| 3403 | */ |
Gabriel Krisman Bertazi | 36e1f3d1 | 2016-12-06 13:31:44 -0200 | [diff] [blame] | 3404 | kmemleak_alloc(p, order_to_size(this_order), 1, GFP_NOIO); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3405 | entries_per_page = order_to_size(this_order) / rq_size; |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3406 | to_do = min(entries_per_page, depth - i); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3407 | left -= to_do * rq_size; |
| 3408 | for (j = 0; j < to_do; j++) { |
Jens Axboe | 2af8cbe | 2017-01-13 14:39:30 -0700 | [diff] [blame] | 3409 | struct request *rq = p; |
| 3410 | |
| 3411 | tags->static_rqs[i] = rq; |
Tejun Heo | 1d9bd51 | 2018-01-09 08:29:48 -0800 | [diff] [blame] | 3412 | if (blk_mq_init_request(set, rq, hctx_idx, node)) { |
| 3413 | tags->static_rqs[i] = NULL; |
| 3414 | goto fail; |
Christoph Hellwig | e9b267d | 2014-04-15 13:59:10 -0600 | [diff] [blame] | 3415 | } |
| 3416 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3417 | p += rq_size; |
| 3418 | i++; |
| 3419 | } |
| 3420 | } |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3421 | return 0; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3422 | |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 3423 | fail: |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3424 | blk_mq_free_rqs(set, tags, hctx_idx); |
| 3425 | return -ENOMEM; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3426 | } |
| 3427 | |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 3428 | struct rq_iter_data { |
| 3429 | struct blk_mq_hw_ctx *hctx; |
| 3430 | bool has_rq; |
| 3431 | }; |
| 3432 | |
John Garry | 2dd6532 | 2022-07-06 20:03:53 +0800 | [diff] [blame] | 3433 | static bool blk_mq_has_request(struct request *rq, void *data) |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 3434 | { |
| 3435 | struct rq_iter_data *iter_data = data; |
| 3436 | |
| 3437 | if (rq->mq_hctx != iter_data->hctx) |
| 3438 | return true; |
| 3439 | iter_data->has_rq = true; |
| 3440 | return false; |
| 3441 | } |
| 3442 | |
| 3443 | static bool blk_mq_hctx_has_requests(struct blk_mq_hw_ctx *hctx) |
| 3444 | { |
| 3445 | struct blk_mq_tags *tags = hctx->sched_tags ? |
| 3446 | hctx->sched_tags : hctx->tags; |
| 3447 | struct rq_iter_data data = { |
| 3448 | .hctx = hctx, |
| 3449 | }; |
| 3450 | |
| 3451 | blk_mq_all_tag_iter(tags, blk_mq_has_request, &data); |
| 3452 | return data.has_rq; |
| 3453 | } |
| 3454 | |
| 3455 | static inline bool blk_mq_last_cpu_in_hctx(unsigned int cpu, |
| 3456 | struct blk_mq_hw_ctx *hctx) |
| 3457 | { |
Yury Norov | 9b51d9d | 2021-08-14 14:17:05 -0700 | [diff] [blame] | 3458 | if (cpumask_first_and(hctx->cpumask, cpu_online_mask) != cpu) |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 3459 | return false; |
| 3460 | if (cpumask_next_and(cpu, hctx->cpumask, cpu_online_mask) < nr_cpu_ids) |
| 3461 | return false; |
| 3462 | return true; |
| 3463 | } |
| 3464 | |
| 3465 | static int blk_mq_hctx_notify_offline(unsigned int cpu, struct hlist_node *node) |
| 3466 | { |
| 3467 | struct blk_mq_hw_ctx *hctx = hlist_entry_safe(node, |
| 3468 | struct blk_mq_hw_ctx, cpuhp_online); |
| 3469 | |
| 3470 | if (!cpumask_test_cpu(cpu, hctx->cpumask) || |
| 3471 | !blk_mq_last_cpu_in_hctx(cpu, hctx)) |
| 3472 | return 0; |
| 3473 | |
| 3474 | /* |
| 3475 | * Prevent new request from being allocated on the current hctx. |
| 3476 | * |
| 3477 | * The smp_mb__after_atomic() Pairs with the implied barrier in |
| 3478 | * test_and_set_bit_lock in sbitmap_get(). Ensures the inactive flag is |
| 3479 | * seen once we return from the tag allocator. |
| 3480 | */ |
| 3481 | set_bit(BLK_MQ_S_INACTIVE, &hctx->state); |
| 3482 | smp_mb__after_atomic(); |
| 3483 | |
| 3484 | /* |
| 3485 | * Try to grab a reference to the queue and wait for any outstanding |
| 3486 | * requests. If we could not grab a reference the queue has been |
| 3487 | * frozen and there are no requests. |
| 3488 | */ |
| 3489 | if (percpu_ref_tryget(&hctx->queue->q_usage_counter)) { |
| 3490 | while (blk_mq_hctx_has_requests(hctx)) |
| 3491 | msleep(5); |
| 3492 | percpu_ref_put(&hctx->queue->q_usage_counter); |
| 3493 | } |
| 3494 | |
| 3495 | return 0; |
| 3496 | } |
| 3497 | |
| 3498 | static int blk_mq_hctx_notify_online(unsigned int cpu, struct hlist_node *node) |
| 3499 | { |
| 3500 | struct blk_mq_hw_ctx *hctx = hlist_entry_safe(node, |
| 3501 | struct blk_mq_hw_ctx, cpuhp_online); |
| 3502 | |
| 3503 | if (cpumask_test_cpu(cpu, hctx->cpumask)) |
| 3504 | clear_bit(BLK_MQ_S_INACTIVE, &hctx->state); |
| 3505 | return 0; |
| 3506 | } |
| 3507 | |
Jens Axboe | e57690f | 2016-08-24 15:34:35 -0600 | [diff] [blame] | 3508 | /* |
| 3509 | * 'cpu' is going away. splice any existing rq_list entries from this |
| 3510 | * software queue to the hw queue dispatch list, and ensure that it |
| 3511 | * gets run. |
| 3512 | */ |
Thomas Gleixner | 9467f85 | 2016-09-22 08:05:17 -0600 | [diff] [blame] | 3513 | static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node) |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3514 | { |
Thomas Gleixner | 9467f85 | 2016-09-22 08:05:17 -0600 | [diff] [blame] | 3515 | struct blk_mq_hw_ctx *hctx; |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3516 | struct blk_mq_ctx *ctx; |
| 3517 | LIST_HEAD(tmp); |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 3518 | enum hctx_type type; |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3519 | |
Thomas Gleixner | 9467f85 | 2016-09-22 08:05:17 -0600 | [diff] [blame] | 3520 | hctx = hlist_entry_safe(node, struct blk_mq_hw_ctx, cpuhp_dead); |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 3521 | if (!cpumask_test_cpu(cpu, hctx->cpumask)) |
| 3522 | return 0; |
| 3523 | |
Jens Axboe | e57690f | 2016-08-24 15:34:35 -0600 | [diff] [blame] | 3524 | ctx = __blk_mq_get_ctx(hctx->queue, cpu); |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 3525 | type = hctx->type; |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3526 | |
| 3527 | spin_lock(&ctx->lock); |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 3528 | if (!list_empty(&ctx->rq_lists[type])) { |
| 3529 | list_splice_init(&ctx->rq_lists[type], &tmp); |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3530 | blk_mq_hctx_clear_pending(hctx, ctx); |
| 3531 | } |
| 3532 | spin_unlock(&ctx->lock); |
| 3533 | |
| 3534 | if (list_empty(&tmp)) |
Thomas Gleixner | 9467f85 | 2016-09-22 08:05:17 -0600 | [diff] [blame] | 3535 | return 0; |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3536 | |
Jens Axboe | e57690f | 2016-08-24 15:34:35 -0600 | [diff] [blame] | 3537 | spin_lock(&hctx->lock); |
| 3538 | list_splice_tail_init(&tmp, &hctx->dispatch); |
| 3539 | spin_unlock(&hctx->lock); |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3540 | |
| 3541 | blk_mq_run_hw_queue(hctx, true); |
Thomas Gleixner | 9467f85 | 2016-09-22 08:05:17 -0600 | [diff] [blame] | 3542 | return 0; |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3543 | } |
| 3544 | |
Thomas Gleixner | 9467f85 | 2016-09-22 08:05:17 -0600 | [diff] [blame] | 3545 | static void blk_mq_remove_cpuhp(struct blk_mq_hw_ctx *hctx) |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3546 | { |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 3547 | if (!(hctx->flags & BLK_MQ_F_STACKING)) |
| 3548 | cpuhp_state_remove_instance_nocalls(CPUHP_AP_BLK_MQ_ONLINE, |
| 3549 | &hctx->cpuhp_online); |
Thomas Gleixner | 9467f85 | 2016-09-22 08:05:17 -0600 | [diff] [blame] | 3550 | cpuhp_state_remove_instance_nocalls(CPUHP_BLK_MQ_DEAD, |
| 3551 | &hctx->cpuhp_dead); |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3552 | } |
| 3553 | |
Ming Lei | 364b618 | 2021-05-11 23:22:36 +0800 | [diff] [blame] | 3554 | /* |
| 3555 | * Before freeing hw queue, clearing the flush request reference in |
| 3556 | * tags->rqs[] for avoiding potential UAF. |
| 3557 | */ |
| 3558 | static void blk_mq_clear_flush_rq_mapping(struct blk_mq_tags *tags, |
| 3559 | unsigned int queue_depth, struct request *flush_rq) |
| 3560 | { |
| 3561 | int i; |
| 3562 | unsigned long flags; |
| 3563 | |
| 3564 | /* The hw queue may not be mapped yet */ |
| 3565 | if (!tags) |
| 3566 | return; |
| 3567 | |
Jens Axboe | 0a467d0 | 2021-10-14 14:39:59 -0600 | [diff] [blame] | 3568 | WARN_ON_ONCE(req_ref_read(flush_rq) != 0); |
Ming Lei | 364b618 | 2021-05-11 23:22:36 +0800 | [diff] [blame] | 3569 | |
| 3570 | for (i = 0; i < queue_depth; i++) |
| 3571 | cmpxchg(&tags->rqs[i], flush_rq, NULL); |
| 3572 | |
| 3573 | /* |
| 3574 | * Wait until all pending iteration is done. |
| 3575 | * |
| 3576 | * Request reference is cleared and it is guaranteed to be observed |
| 3577 | * after the ->lock is released. |
| 3578 | */ |
| 3579 | spin_lock_irqsave(&tags->lock, flags); |
| 3580 | spin_unlock_irqrestore(&tags->lock, flags); |
| 3581 | } |
| 3582 | |
Ming Lei | c3b4afc | 2015-06-04 22:25:04 +0800 | [diff] [blame] | 3583 | /* hctx->ctxs will be freed in queue's release handler */ |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3584 | static void blk_mq_exit_hctx(struct request_queue *q, |
| 3585 | struct blk_mq_tag_set *set, |
| 3586 | struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) |
| 3587 | { |
Ming Lei | 364b618 | 2021-05-11 23:22:36 +0800 | [diff] [blame] | 3588 | struct request *flush_rq = hctx->fq->flush_rq; |
| 3589 | |
Ming Lei | 8ab0b7d | 2018-01-09 21:28:29 +0800 | [diff] [blame] | 3590 | if (blk_mq_hw_queue_mapped(hctx)) |
| 3591 | blk_mq_tag_idle(hctx); |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3592 | |
Ming Lei | 6cfeadb | 2022-06-16 09:44:01 +0800 | [diff] [blame] | 3593 | if (blk_queue_init_done(q)) |
| 3594 | blk_mq_clear_flush_rq_mapping(set->tags[hctx_idx], |
| 3595 | set->queue_depth, flush_rq); |
Ming Lei | f70ced0 | 2014-09-25 23:23:47 +0800 | [diff] [blame] | 3596 | if (set->ops->exit_request) |
Ming Lei | 364b618 | 2021-05-11 23:22:36 +0800 | [diff] [blame] | 3597 | set->ops->exit_request(set, flush_rq, hctx_idx); |
Ming Lei | f70ced0 | 2014-09-25 23:23:47 +0800 | [diff] [blame] | 3598 | |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3599 | if (set->ops->exit_hctx) |
| 3600 | set->ops->exit_hctx(hctx, hctx_idx); |
| 3601 | |
Thomas Gleixner | 9467f85 | 2016-09-22 08:05:17 -0600 | [diff] [blame] | 3602 | blk_mq_remove_cpuhp(hctx); |
Ming Lei | 2f8f133 | 2019-04-30 09:52:27 +0800 | [diff] [blame] | 3603 | |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 3604 | xa_erase(&q->hctx_table, hctx_idx); |
| 3605 | |
Ming Lei | 2f8f133 | 2019-04-30 09:52:27 +0800 | [diff] [blame] | 3606 | spin_lock(&q->unused_hctx_lock); |
| 3607 | list_add(&hctx->hctx_list, &q->unused_hctx_list); |
| 3608 | spin_unlock(&q->unused_hctx_lock); |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3609 | } |
| 3610 | |
Ming Lei | 624dbe4 | 2014-05-27 23:35:13 +0800 | [diff] [blame] | 3611 | static void blk_mq_exit_hw_queues(struct request_queue *q, |
| 3612 | struct blk_mq_tag_set *set, int nr_queue) |
| 3613 | { |
| 3614 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 3615 | unsigned long i; |
Ming Lei | 624dbe4 | 2014-05-27 23:35:13 +0800 | [diff] [blame] | 3616 | |
| 3617 | queue_for_each_hw_ctx(q, hctx, i) { |
| 3618 | if (i == nr_queue) |
| 3619 | break; |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3620 | blk_mq_exit_hctx(q, set, hctx, i); |
Ming Lei | 624dbe4 | 2014-05-27 23:35:13 +0800 | [diff] [blame] | 3621 | } |
Ming Lei | 624dbe4 | 2014-05-27 23:35:13 +0800 | [diff] [blame] | 3622 | } |
| 3623 | |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3624 | static int blk_mq_init_hctx(struct request_queue *q, |
| 3625 | struct blk_mq_tag_set *set, |
| 3626 | struct blk_mq_hw_ctx *hctx, unsigned hctx_idx) |
| 3627 | { |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3628 | hctx->queue_num = hctx_idx; |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3629 | |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 3630 | if (!(hctx->flags & BLK_MQ_F_STACKING)) |
| 3631 | cpuhp_state_add_instance_nocalls(CPUHP_AP_BLK_MQ_ONLINE, |
| 3632 | &hctx->cpuhp_online); |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3633 | cpuhp_state_add_instance_nocalls(CPUHP_BLK_MQ_DEAD, &hctx->cpuhp_dead); |
| 3634 | |
| 3635 | hctx->tags = set->tags[hctx_idx]; |
| 3636 | |
| 3637 | if (set->ops->init_hctx && |
| 3638 | set->ops->init_hctx(hctx, set->driver_data, hctx_idx)) |
| 3639 | goto unregister_cpu_notifier; |
| 3640 | |
| 3641 | if (blk_mq_init_request(set, hctx->fq->flush_rq, hctx_idx, |
| 3642 | hctx->numa_node)) |
| 3643 | goto exit_hctx; |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 3644 | |
| 3645 | if (xa_insert(&q->hctx_table, hctx_idx, hctx, GFP_KERNEL)) |
| 3646 | goto exit_flush_rq; |
| 3647 | |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3648 | return 0; |
| 3649 | |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 3650 | exit_flush_rq: |
| 3651 | if (set->ops->exit_request) |
| 3652 | set->ops->exit_request(set, hctx->fq->flush_rq, hctx_idx); |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3653 | exit_hctx: |
| 3654 | if (set->ops->exit_hctx) |
| 3655 | set->ops->exit_hctx(hctx, hctx_idx); |
| 3656 | unregister_cpu_notifier: |
| 3657 | blk_mq_remove_cpuhp(hctx); |
| 3658 | return -1; |
| 3659 | } |
| 3660 | |
| 3661 | static struct blk_mq_hw_ctx * |
| 3662 | blk_mq_alloc_hctx(struct request_queue *q, struct blk_mq_tag_set *set, |
| 3663 | int node) |
| 3664 | { |
| 3665 | struct blk_mq_hw_ctx *hctx; |
| 3666 | gfp_t gfp = GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY; |
| 3667 | |
Ming Lei | 704b914 | 2021-12-03 21:15:32 +0800 | [diff] [blame] | 3668 | hctx = kzalloc_node(sizeof(struct blk_mq_hw_ctx), gfp, node); |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3669 | if (!hctx) |
| 3670 | goto fail_alloc_hctx; |
| 3671 | |
| 3672 | if (!zalloc_cpumask_var_node(&hctx->cpumask, gfp, node)) |
| 3673 | goto free_hctx; |
| 3674 | |
| 3675 | atomic_set(&hctx->nr_active, 0); |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3676 | if (node == NUMA_NO_NODE) |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3677 | node = set->numa_node; |
| 3678 | hctx->numa_node = node; |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3679 | |
Jens Axboe | 9f99373 | 2017-04-10 09:54:54 -0600 | [diff] [blame] | 3680 | INIT_DELAYED_WORK(&hctx->run_work, blk_mq_run_work_fn); |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3681 | spin_lock_init(&hctx->lock); |
| 3682 | INIT_LIST_HEAD(&hctx->dispatch); |
| 3683 | hctx->queue = q; |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 3684 | hctx->flags = set->flags & ~BLK_MQ_F_TAG_QUEUE_SHARED; |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3685 | |
Ming Lei | 2f8f133 | 2019-04-30 09:52:27 +0800 | [diff] [blame] | 3686 | INIT_LIST_HEAD(&hctx->hctx_list); |
| 3687 | |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3688 | /* |
| 3689 | * Allocate space for all possible cpus to avoid allocation at |
| 3690 | * runtime |
| 3691 | */ |
Johannes Thumshirn | d904bfa | 2017-11-15 17:32:33 -0800 | [diff] [blame] | 3692 | hctx->ctxs = kmalloc_array_node(nr_cpu_ids, sizeof(void *), |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3693 | gfp, node); |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3694 | if (!hctx->ctxs) |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3695 | goto free_cpumask; |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3696 | |
Jianchao Wang | 5b20285 | 2018-10-12 18:07:26 +0800 | [diff] [blame] | 3697 | if (sbitmap_init_node(&hctx->ctx_map, nr_cpu_ids, ilog2(8), |
Ming Lei | c548e62 | 2021-01-22 10:33:08 +0800 | [diff] [blame] | 3698 | gfp, node, false, false)) |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3699 | goto free_ctxs; |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3700 | hctx->nr_ctx = 0; |
| 3701 | |
Ming Lei | 5815839b | 2018-06-25 19:31:47 +0800 | [diff] [blame] | 3702 | spin_lock_init(&hctx->dispatch_wait_lock); |
Jens Axboe | eb619fd | 2017-11-09 08:32:43 -0700 | [diff] [blame] | 3703 | init_waitqueue_func_entry(&hctx->dispatch_wait, blk_mq_dispatch_wake); |
| 3704 | INIT_LIST_HEAD(&hctx->dispatch_wait.entry); |
| 3705 | |
Guoqing Jiang | 754a157 | 2020-03-09 22:41:37 +0100 | [diff] [blame] | 3706 | hctx->fq = blk_alloc_flush_queue(hctx->numa_node, set->cmd_size, gfp); |
Ming Lei | f70ced0 | 2014-09-25 23:23:47 +0800 | [diff] [blame] | 3707 | if (!hctx->fq) |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3708 | goto free_bitmap; |
Ming Lei | f70ced0 | 2014-09-25 23:23:47 +0800 | [diff] [blame] | 3709 | |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3710 | blk_mq_hctx_kobj_init(hctx); |
Bart Van Assche | 6a83e74 | 2016-11-02 10:09:51 -0600 | [diff] [blame] | 3711 | |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3712 | return hctx; |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3713 | |
| 3714 | free_bitmap: |
Omar Sandoval | 8845964 | 2016-09-17 08:38:44 -0600 | [diff] [blame] | 3715 | sbitmap_free(&hctx->ctx_map); |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3716 | free_ctxs: |
| 3717 | kfree(hctx->ctxs); |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 3718 | free_cpumask: |
| 3719 | free_cpumask_var(hctx->cpumask); |
| 3720 | free_hctx: |
| 3721 | kfree(hctx); |
| 3722 | fail_alloc_hctx: |
| 3723 | return NULL; |
Ming Lei | 08e98fc | 2014-09-25 23:23:38 +0800 | [diff] [blame] | 3724 | } |
| 3725 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3726 | static void blk_mq_init_cpu_queues(struct request_queue *q, |
| 3727 | unsigned int nr_hw_queues) |
| 3728 | { |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 3729 | struct blk_mq_tag_set *set = q->tag_set; |
| 3730 | unsigned int i, j; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3731 | |
| 3732 | for_each_possible_cpu(i) { |
| 3733 | struct blk_mq_ctx *__ctx = per_cpu_ptr(q->queue_ctx, i); |
| 3734 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 3735 | int k; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3736 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3737 | __ctx->cpu = i; |
| 3738 | spin_lock_init(&__ctx->lock); |
Ming Lei | c16d6b5 | 2018-12-17 08:44:05 -0700 | [diff] [blame] | 3739 | for (k = HCTX_TYPE_DEFAULT; k < HCTX_MAX_TYPES; k++) |
| 3740 | INIT_LIST_HEAD(&__ctx->rq_lists[k]); |
| 3741 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3742 | __ctx->queue = q; |
| 3743 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3744 | /* |
| 3745 | * Set local node, IFF we have more than one hw queue. If |
| 3746 | * not, we remain on the home node of the device |
| 3747 | */ |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 3748 | for (j = 0; j < set->nr_maps; j++) { |
| 3749 | hctx = blk_mq_map_queue_type(q, j, i); |
| 3750 | if (nr_hw_queues > 1 && hctx->numa_node == NUMA_NO_NODE) |
Xianting Tian | 576e85c | 2020-10-19 16:20:47 +0800 | [diff] [blame] | 3751 | hctx->numa_node = cpu_to_node(i); |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 3752 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3753 | } |
| 3754 | } |
| 3755 | |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3756 | struct blk_mq_tags *blk_mq_alloc_map_and_rqs(struct blk_mq_tag_set *set, |
| 3757 | unsigned int hctx_idx, |
| 3758 | unsigned int depth) |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3759 | { |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3760 | struct blk_mq_tags *tags; |
| 3761 | int ret; |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3762 | |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3763 | tags = blk_mq_alloc_rq_map(set, hctx_idx, depth, set->reserved_tags); |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3764 | if (!tags) |
| 3765 | return NULL; |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3766 | |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3767 | ret = blk_mq_alloc_rqs(set, tags, hctx_idx, depth); |
| 3768 | if (ret) { |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3769 | blk_mq_free_rq_map(tags); |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3770 | return NULL; |
| 3771 | } |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3772 | |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3773 | return tags; |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3774 | } |
| 3775 | |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3776 | static bool __blk_mq_alloc_map_and_rqs(struct blk_mq_tag_set *set, |
| 3777 | int hctx_idx) |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3778 | { |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 3779 | if (blk_mq_is_shared_tags(set->flags)) { |
| 3780 | set->tags[hctx_idx] = set->shared_tags; |
John Garry | 1c0706a | 2020-08-19 23:20:22 +0800 | [diff] [blame] | 3781 | |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3782 | return true; |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 3783 | } |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3784 | |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3785 | set->tags[hctx_idx] = blk_mq_alloc_map_and_rqs(set, hctx_idx, |
| 3786 | set->queue_depth); |
| 3787 | |
| 3788 | return set->tags[hctx_idx]; |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3789 | } |
| 3790 | |
John Garry | 645db34 | 2021-10-05 18:23:36 +0800 | [diff] [blame] | 3791 | void blk_mq_free_map_and_rqs(struct blk_mq_tag_set *set, |
| 3792 | struct blk_mq_tags *tags, |
| 3793 | unsigned int hctx_idx) |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3794 | { |
John Garry | 645db34 | 2021-10-05 18:23:36 +0800 | [diff] [blame] | 3795 | if (tags) { |
| 3796 | blk_mq_free_rqs(set, tags, hctx_idx); |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3797 | blk_mq_free_rq_map(tags); |
Jens Axboe | cc71a6f | 2017-01-11 14:29:56 -0700 | [diff] [blame] | 3798 | } |
| 3799 | } |
| 3800 | |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3801 | static void __blk_mq_free_map_and_rqs(struct blk_mq_tag_set *set, |
| 3802 | unsigned int hctx_idx) |
| 3803 | { |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 3804 | if (!blk_mq_is_shared_tags(set->flags)) |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3805 | blk_mq_free_map_and_rqs(set, set->tags[hctx_idx], hctx_idx); |
| 3806 | |
| 3807 | set->tags[hctx_idx] = NULL; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3808 | } |
| 3809 | |
Christoph Hellwig | 4b855ad | 2017-06-26 12:20:57 +0200 | [diff] [blame] | 3810 | static void blk_mq_map_swqueue(struct request_queue *q) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3811 | { |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 3812 | unsigned int j, hctx_idx; |
| 3813 | unsigned long i; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3814 | struct blk_mq_hw_ctx *hctx; |
| 3815 | struct blk_mq_ctx *ctx; |
Ming Lei | 2a34c08 | 2015-04-21 10:00:20 +0800 | [diff] [blame] | 3816 | struct blk_mq_tag_set *set = q->tag_set; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3817 | |
| 3818 | queue_for_each_hw_ctx(q, hctx, i) { |
Jens Axboe | e4043dc | 2014-04-09 10:18:23 -0600 | [diff] [blame] | 3819 | cpumask_clear(hctx->cpumask); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3820 | hctx->nr_ctx = 0; |
huhai | d416c92 | 2018-05-18 08:32:30 -0600 | [diff] [blame] | 3821 | hctx->dispatch_from = NULL; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3822 | } |
| 3823 | |
| 3824 | /* |
Christoph Hellwig | 4b855ad | 2017-06-26 12:20:57 +0200 | [diff] [blame] | 3825 | * Map software to hardware queues. |
Ming Lei | 4412efe | 2018-04-25 04:01:44 +0800 | [diff] [blame] | 3826 | * |
| 3827 | * If the cpu isn't present, the cpu is mapped to first hctx. |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3828 | */ |
Christoph Hellwig | 20e4d813 | 2018-01-12 10:53:06 +0800 | [diff] [blame] | 3829 | for_each_possible_cpu(i) { |
Ming Lei | 4412efe | 2018-04-25 04:01:44 +0800 | [diff] [blame] | 3830 | |
Thomas Gleixner | 897bb0c | 2016-03-19 11:30:33 +0100 | [diff] [blame] | 3831 | ctx = per_cpu_ptr(q->queue_ctx, i); |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 3832 | for (j = 0; j < set->nr_maps; j++) { |
Jianchao Wang | bb94aea | 2019-01-24 18:25:33 +0800 | [diff] [blame] | 3833 | if (!set->map[j].nr_queues) { |
| 3834 | ctx->hctxs[j] = blk_mq_map_queue_type(q, |
| 3835 | HCTX_TYPE_DEFAULT, i); |
Ming Lei | e5edd5f | 2018-12-18 01:28:56 +0800 | [diff] [blame] | 3836 | continue; |
Jianchao Wang | bb94aea | 2019-01-24 18:25:33 +0800 | [diff] [blame] | 3837 | } |
Ming Lei | fd68987 | 2020-05-07 21:04:08 +0800 | [diff] [blame] | 3838 | hctx_idx = set->map[j].mq_map[i]; |
| 3839 | /* unmapped hw queue can be remapped after CPU topo changed */ |
| 3840 | if (!set->tags[hctx_idx] && |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 3841 | !__blk_mq_alloc_map_and_rqs(set, hctx_idx)) { |
Ming Lei | fd68987 | 2020-05-07 21:04:08 +0800 | [diff] [blame] | 3842 | /* |
| 3843 | * If tags initialization fail for some hctx, |
| 3844 | * that hctx won't be brought online. In this |
| 3845 | * case, remap the current ctx to hctx[0] which |
| 3846 | * is guaranteed to always have tags allocated |
| 3847 | */ |
| 3848 | set->map[j].mq_map[i] = 0; |
| 3849 | } |
Ming Lei | e5edd5f | 2018-12-18 01:28:56 +0800 | [diff] [blame] | 3850 | |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 3851 | hctx = blk_mq_map_queue_type(q, j, i); |
Jianchao Wang | 8ccdf4a | 2019-01-24 18:25:32 +0800 | [diff] [blame] | 3852 | ctx->hctxs[j] = hctx; |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 3853 | /* |
| 3854 | * If the CPU is already set in the mask, then we've |
| 3855 | * mapped this one already. This can happen if |
| 3856 | * devices share queues across queue maps. |
| 3857 | */ |
| 3858 | if (cpumask_test_cpu(i, hctx->cpumask)) |
| 3859 | continue; |
| 3860 | |
| 3861 | cpumask_set_cpu(i, hctx->cpumask); |
| 3862 | hctx->type = j; |
| 3863 | ctx->index_hw[hctx->type] = hctx->nr_ctx; |
| 3864 | hctx->ctxs[hctx->nr_ctx++] = ctx; |
| 3865 | |
| 3866 | /* |
| 3867 | * If the nr_ctx type overflows, we have exceeded the |
| 3868 | * amount of sw queues we can support. |
| 3869 | */ |
| 3870 | BUG_ON(!hctx->nr_ctx); |
| 3871 | } |
Jianchao Wang | bb94aea | 2019-01-24 18:25:33 +0800 | [diff] [blame] | 3872 | |
| 3873 | for (; j < HCTX_MAX_TYPES; j++) |
| 3874 | ctx->hctxs[j] = blk_mq_map_queue_type(q, |
| 3875 | HCTX_TYPE_DEFAULT, i); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3876 | } |
Jens Axboe | 506e931 | 2014-05-07 10:26:44 -0600 | [diff] [blame] | 3877 | |
| 3878 | queue_for_each_hw_ctx(q, hctx, i) { |
Ming Lei | 4412efe | 2018-04-25 04:01:44 +0800 | [diff] [blame] | 3879 | /* |
| 3880 | * If no software queues are mapped to this hardware queue, |
| 3881 | * disable it and free the request entries. |
| 3882 | */ |
| 3883 | if (!hctx->nr_ctx) { |
| 3884 | /* Never unmap queue 0. We need it as a |
| 3885 | * fallback in case of a new remap fails |
| 3886 | * allocation |
| 3887 | */ |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 3888 | if (i) |
| 3889 | __blk_mq_free_map_and_rqs(set, i); |
Ming Lei | 4412efe | 2018-04-25 04:01:44 +0800 | [diff] [blame] | 3890 | |
| 3891 | hctx->tags = NULL; |
| 3892 | continue; |
| 3893 | } |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3894 | |
Ming Lei | 2a34c08 | 2015-04-21 10:00:20 +0800 | [diff] [blame] | 3895 | hctx->tags = set->tags[i]; |
| 3896 | WARN_ON(!hctx->tags); |
| 3897 | |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3898 | /* |
Chong Yuan | 889fa31 | 2015-04-15 11:39:29 -0600 | [diff] [blame] | 3899 | * Set the map size to the number of mapped software queues. |
| 3900 | * This is more accurate and more efficient than looping |
| 3901 | * over all possibly mapped software queues. |
| 3902 | */ |
Omar Sandoval | 8845964 | 2016-09-17 08:38:44 -0600 | [diff] [blame] | 3903 | sbitmap_resize(&hctx->ctx_map, hctx->nr_ctx); |
Chong Yuan | 889fa31 | 2015-04-15 11:39:29 -0600 | [diff] [blame] | 3904 | |
| 3905 | /* |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 3906 | * Initialize batch roundrobin counts |
| 3907 | */ |
Ming Lei | f82ddf1 | 2018-04-08 17:48:10 +0800 | [diff] [blame] | 3908 | hctx->next_cpu = blk_mq_first_mapped_cpu(hctx); |
Jens Axboe | 506e931 | 2014-05-07 10:26:44 -0600 | [diff] [blame] | 3909 | hctx->next_cpu_batch = BLK_MQ_CPU_WORK_BATCH; |
| 3910 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 3911 | } |
| 3912 | |
Jens Axboe | 8e8320c | 2017-06-20 17:56:13 -0600 | [diff] [blame] | 3913 | /* |
| 3914 | * Caller needs to ensure that we're either frozen/quiesced, or that |
| 3915 | * the queue isn't live yet. |
| 3916 | */ |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3917 | static void queue_set_hctx_shared(struct request_queue *q, bool shared) |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3918 | { |
| 3919 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 3920 | unsigned long i; |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3921 | |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3922 | queue_for_each_hw_ctx(q, hctx, i) { |
Yu Kuai | 454bb67 | 2021-07-31 14:21:30 +0800 | [diff] [blame] | 3923 | if (shared) { |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 3924 | hctx->flags |= BLK_MQ_F_TAG_QUEUE_SHARED; |
Yu Kuai | 454bb67 | 2021-07-31 14:21:30 +0800 | [diff] [blame] | 3925 | } else { |
| 3926 | blk_mq_tag_idle(hctx); |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 3927 | hctx->flags &= ~BLK_MQ_F_TAG_QUEUE_SHARED; |
Yu Kuai | 454bb67 | 2021-07-31 14:21:30 +0800 | [diff] [blame] | 3928 | } |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3929 | } |
| 3930 | } |
| 3931 | |
Hannes Reinecke | 655ac30 | 2020-08-19 23:20:20 +0800 | [diff] [blame] | 3932 | static void blk_mq_update_tag_set_shared(struct blk_mq_tag_set *set, |
| 3933 | bool shared) |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3934 | { |
| 3935 | struct request_queue *q; |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3936 | |
Bart Van Assche | 705cda9 | 2017-04-07 11:16:49 -0700 | [diff] [blame] | 3937 | lockdep_assert_held(&set->tag_list_lock); |
| 3938 | |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3939 | list_for_each_entry(q, &set->tag_list, tag_set_list) { |
| 3940 | blk_mq_freeze_queue(q); |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3941 | queue_set_hctx_shared(q, shared); |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3942 | blk_mq_unfreeze_queue(q); |
| 3943 | } |
| 3944 | } |
| 3945 | |
| 3946 | static void blk_mq_del_queue_tag_set(struct request_queue *q) |
| 3947 | { |
| 3948 | struct blk_mq_tag_set *set = q->tag_set; |
| 3949 | |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3950 | mutex_lock(&set->tag_list_lock); |
Daniel Wagner | 08c875c | 2020-07-28 15:29:51 +0200 | [diff] [blame] | 3951 | list_del(&q->tag_set_list); |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3952 | if (list_is_singular(&set->tag_list)) { |
| 3953 | /* just transitioned to unshared */ |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 3954 | set->flags &= ~BLK_MQ_F_TAG_QUEUE_SHARED; |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3955 | /* update existing queue */ |
Hannes Reinecke | 655ac30 | 2020-08-19 23:20:20 +0800 | [diff] [blame] | 3956 | blk_mq_update_tag_set_shared(set, false); |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3957 | } |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3958 | mutex_unlock(&set->tag_list_lock); |
Roman Pen | a347c7a | 2018-06-10 22:38:24 +0200 | [diff] [blame] | 3959 | INIT_LIST_HEAD(&q->tag_set_list); |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3960 | } |
| 3961 | |
| 3962 | static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set, |
| 3963 | struct request_queue *q) |
| 3964 | { |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3965 | mutex_lock(&set->tag_list_lock); |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3966 | |
Jens Axboe | ff821d2 | 2017-11-10 22:05:12 -0700 | [diff] [blame] | 3967 | /* |
| 3968 | * Check to see if we're transitioning to shared (from 1 to 2 queues). |
| 3969 | */ |
| 3970 | if (!list_empty(&set->tag_list) && |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 3971 | !(set->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) { |
| 3972 | set->flags |= BLK_MQ_F_TAG_QUEUE_SHARED; |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3973 | /* update existing queue */ |
Hannes Reinecke | 655ac30 | 2020-08-19 23:20:20 +0800 | [diff] [blame] | 3974 | blk_mq_update_tag_set_shared(set, true); |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3975 | } |
Ming Lei | 51db1c3 | 2020-08-19 23:20:19 +0800 | [diff] [blame] | 3976 | if (set->flags & BLK_MQ_F_TAG_QUEUE_SHARED) |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3977 | queue_set_hctx_shared(q, true); |
Daniel Wagner | 08c875c | 2020-07-28 15:29:51 +0200 | [diff] [blame] | 3978 | list_add_tail(&q->tag_set_list, &set->tag_list); |
Jeff Moyer | 2404e60 | 2015-11-03 10:40:06 -0500 | [diff] [blame] | 3979 | |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 3980 | mutex_unlock(&set->tag_list_lock); |
| 3981 | } |
| 3982 | |
Ming Lei | 1db4909 | 2018-11-20 09:44:35 +0800 | [diff] [blame] | 3983 | /* All allocations will be freed in release handler of q->mq_kobj */ |
| 3984 | static int blk_mq_alloc_ctxs(struct request_queue *q) |
| 3985 | { |
| 3986 | struct blk_mq_ctxs *ctxs; |
| 3987 | int cpu; |
| 3988 | |
| 3989 | ctxs = kzalloc(sizeof(*ctxs), GFP_KERNEL); |
| 3990 | if (!ctxs) |
| 3991 | return -ENOMEM; |
| 3992 | |
| 3993 | ctxs->queue_ctx = alloc_percpu(struct blk_mq_ctx); |
| 3994 | if (!ctxs->queue_ctx) |
| 3995 | goto fail; |
| 3996 | |
| 3997 | for_each_possible_cpu(cpu) { |
| 3998 | struct blk_mq_ctx *ctx = per_cpu_ptr(ctxs->queue_ctx, cpu); |
| 3999 | ctx->ctxs = ctxs; |
| 4000 | } |
| 4001 | |
| 4002 | q->mq_kobj = &ctxs->kobj; |
| 4003 | q->queue_ctx = ctxs->queue_ctx; |
| 4004 | |
| 4005 | return 0; |
| 4006 | fail: |
| 4007 | kfree(ctxs); |
| 4008 | return -ENOMEM; |
| 4009 | } |
| 4010 | |
Ming Lei | e09aae7 | 2015-01-29 20:17:27 +0800 | [diff] [blame] | 4011 | /* |
| 4012 | * It is the actual release handler for mq, but we do it from |
| 4013 | * request queue's release handler for avoiding use-after-free |
| 4014 | * and headache because q->mq_kobj shouldn't have been introduced, |
| 4015 | * but we can't group ctx/kctx kobj without it. |
| 4016 | */ |
| 4017 | void blk_mq_release(struct request_queue *q) |
| 4018 | { |
Ming Lei | 2f8f133 | 2019-04-30 09:52:27 +0800 | [diff] [blame] | 4019 | struct blk_mq_hw_ctx *hctx, *next; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 4020 | unsigned long i; |
Ming Lei | e09aae7 | 2015-01-29 20:17:27 +0800 | [diff] [blame] | 4021 | |
Ming Lei | 2f8f133 | 2019-04-30 09:52:27 +0800 | [diff] [blame] | 4022 | queue_for_each_hw_ctx(q, hctx, i) |
| 4023 | WARN_ON_ONCE(hctx && list_empty(&hctx->hctx_list)); |
| 4024 | |
| 4025 | /* all hctx are in .unused_hctx_list now */ |
| 4026 | list_for_each_entry_safe(hctx, next, &q->unused_hctx_list, hctx_list) { |
| 4027 | list_del_init(&hctx->hctx_list); |
Ming Lei | 6c8b232 | 2017-02-22 18:14:01 +0800 | [diff] [blame] | 4028 | kobject_put(&hctx->kobj); |
Ming Lei | c3b4afc | 2015-06-04 22:25:04 +0800 | [diff] [blame] | 4029 | } |
Ming Lei | e09aae7 | 2015-01-29 20:17:27 +0800 | [diff] [blame] | 4030 | |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 4031 | xa_destroy(&q->hctx_table); |
Ming Lei | e09aae7 | 2015-01-29 20:17:27 +0800 | [diff] [blame] | 4032 | |
Ming Lei | 7ea5fe3 | 2017-02-22 18:14:00 +0800 | [diff] [blame] | 4033 | /* |
| 4034 | * release .mq_kobj and sw queue's kobject now because |
| 4035 | * both share lifetime with request queue. |
| 4036 | */ |
| 4037 | blk_mq_sysfs_deinit(q); |
Ming Lei | e09aae7 | 2015-01-29 20:17:27 +0800 | [diff] [blame] | 4038 | } |
| 4039 | |
Christoph Hellwig | 5ec780a | 2021-06-24 10:10:12 +0200 | [diff] [blame] | 4040 | static struct request_queue *blk_mq_init_queue_data(struct blk_mq_tag_set *set, |
Christoph Hellwig | 2f227bb | 2020-03-27 09:30:08 +0100 | [diff] [blame] | 4041 | void *queuedata) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4042 | { |
Christoph Hellwig | 26a9750 | 2021-06-02 09:53:17 +0300 | [diff] [blame] | 4043 | struct request_queue *q; |
| 4044 | int ret; |
Mike Snitzer | b62c21b | 2015-03-12 23:56:02 -0400 | [diff] [blame] | 4045 | |
Christoph Hellwig | 80bd4a7 | 2022-11-01 16:00:47 +0100 | [diff] [blame] | 4046 | q = blk_alloc_queue(set->numa_node); |
Christoph Hellwig | 26a9750 | 2021-06-02 09:53:17 +0300 | [diff] [blame] | 4047 | if (!q) |
Mike Snitzer | b62c21b | 2015-03-12 23:56:02 -0400 | [diff] [blame] | 4048 | return ERR_PTR(-ENOMEM); |
Christoph Hellwig | 26a9750 | 2021-06-02 09:53:17 +0300 | [diff] [blame] | 4049 | q->queuedata = queuedata; |
| 4050 | ret = blk_mq_init_allocated_queue(set, q); |
| 4051 | if (ret) { |
Christoph Hellwig | 6f8191f | 2022-06-19 08:05:51 +0200 | [diff] [blame] | 4052 | blk_put_queue(q); |
Christoph Hellwig | 26a9750 | 2021-06-02 09:53:17 +0300 | [diff] [blame] | 4053 | return ERR_PTR(ret); |
| 4054 | } |
Mike Snitzer | b62c21b | 2015-03-12 23:56:02 -0400 | [diff] [blame] | 4055 | return q; |
| 4056 | } |
Christoph Hellwig | 2f227bb | 2020-03-27 09:30:08 +0100 | [diff] [blame] | 4057 | |
| 4058 | struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set) |
| 4059 | { |
| 4060 | return blk_mq_init_queue_data(set, NULL); |
| 4061 | } |
Mike Snitzer | b62c21b | 2015-03-12 23:56:02 -0400 | [diff] [blame] | 4062 | EXPORT_SYMBOL(blk_mq_init_queue); |
| 4063 | |
Christoph Hellwig | 6f8191f | 2022-06-19 08:05:51 +0200 | [diff] [blame] | 4064 | /** |
| 4065 | * blk_mq_destroy_queue - shutdown a request queue |
| 4066 | * @q: request queue to shutdown |
| 4067 | * |
| 4068 | * This shuts down a request queue allocated by blk_mq_init_queue() and drops |
| 4069 | * the initial reference. All future requests will failed with -ENODEV. |
| 4070 | * |
| 4071 | * Context: can sleep |
| 4072 | */ |
| 4073 | void blk_mq_destroy_queue(struct request_queue *q) |
| 4074 | { |
| 4075 | WARN_ON_ONCE(!queue_is_mq(q)); |
| 4076 | WARN_ON_ONCE(blk_queue_registered(q)); |
| 4077 | |
| 4078 | might_sleep(); |
| 4079 | |
| 4080 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); |
| 4081 | blk_queue_start_drain(q); |
Jinlong Chen | 56c1ee9 | 2022-10-30 16:32:12 +0800 | [diff] [blame] | 4082 | blk_mq_freeze_queue_wait(q); |
Christoph Hellwig | 6f8191f | 2022-06-19 08:05:51 +0200 | [diff] [blame] | 4083 | |
| 4084 | blk_sync_queue(q); |
| 4085 | blk_mq_cancel_work_sync(q); |
| 4086 | blk_mq_exit_queue(q); |
Christoph Hellwig | 6f8191f | 2022-06-19 08:05:51 +0200 | [diff] [blame] | 4087 | } |
| 4088 | EXPORT_SYMBOL(blk_mq_destroy_queue); |
| 4089 | |
Christoph Hellwig | 4dcc4874 | 2021-08-16 15:19:05 +0200 | [diff] [blame] | 4090 | struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata, |
| 4091 | struct lock_class_key *lkclass) |
Jens Axboe | 9316a9e | 2018-10-15 08:40:37 -0600 | [diff] [blame] | 4092 | { |
| 4093 | struct request_queue *q; |
Christoph Hellwig | b461dfc | 2021-06-02 09:53:18 +0300 | [diff] [blame] | 4094 | struct gendisk *disk; |
Jens Axboe | 9316a9e | 2018-10-15 08:40:37 -0600 | [diff] [blame] | 4095 | |
Christoph Hellwig | b461dfc | 2021-06-02 09:53:18 +0300 | [diff] [blame] | 4096 | q = blk_mq_init_queue_data(set, queuedata); |
| 4097 | if (IS_ERR(q)) |
| 4098 | return ERR_CAST(q); |
Jens Axboe | 9316a9e | 2018-10-15 08:40:37 -0600 | [diff] [blame] | 4099 | |
Christoph Hellwig | 4a1fa41 | 2021-08-16 15:19:08 +0200 | [diff] [blame] | 4100 | disk = __alloc_disk_node(q, set->numa_node, lkclass); |
Christoph Hellwig | b461dfc | 2021-06-02 09:53:18 +0300 | [diff] [blame] | 4101 | if (!disk) { |
Christoph Hellwig | 0a3e5cc | 2022-07-20 15:05:40 +0200 | [diff] [blame] | 4102 | blk_mq_destroy_queue(q); |
Christoph Hellwig | 2b3f056 | 2022-10-18 15:57:17 +0200 | [diff] [blame] | 4103 | blk_put_queue(q); |
Christoph Hellwig | b461dfc | 2021-06-02 09:53:18 +0300 | [diff] [blame] | 4104 | return ERR_PTR(-ENOMEM); |
Jens Axboe | 9316a9e | 2018-10-15 08:40:37 -0600 | [diff] [blame] | 4105 | } |
Christoph Hellwig | 6f8191f | 2022-06-19 08:05:51 +0200 | [diff] [blame] | 4106 | set_bit(GD_OWNS_QUEUE, &disk->state); |
Christoph Hellwig | b461dfc | 2021-06-02 09:53:18 +0300 | [diff] [blame] | 4107 | return disk; |
Jens Axboe | 9316a9e | 2018-10-15 08:40:37 -0600 | [diff] [blame] | 4108 | } |
Christoph Hellwig | b461dfc | 2021-06-02 09:53:18 +0300 | [diff] [blame] | 4109 | EXPORT_SYMBOL(__blk_mq_alloc_disk); |
Jens Axboe | 9316a9e | 2018-10-15 08:40:37 -0600 | [diff] [blame] | 4110 | |
Christoph Hellwig | 6f8191f | 2022-06-19 08:05:51 +0200 | [diff] [blame] | 4111 | struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q, |
| 4112 | struct lock_class_key *lkclass) |
| 4113 | { |
Christoph Hellwig | 22c17e2 | 2022-11-22 08:27:53 +0100 | [diff] [blame] | 4114 | struct gendisk *disk; |
| 4115 | |
Christoph Hellwig | 6f8191f | 2022-06-19 08:05:51 +0200 | [diff] [blame] | 4116 | if (!blk_get_queue(q)) |
| 4117 | return NULL; |
Christoph Hellwig | 22c17e2 | 2022-11-22 08:27:53 +0100 | [diff] [blame] | 4118 | disk = __alloc_disk_node(q, NUMA_NO_NODE, lkclass); |
| 4119 | if (!disk) |
| 4120 | blk_put_queue(q); |
| 4121 | return disk; |
Christoph Hellwig | 6f8191f | 2022-06-19 08:05:51 +0200 | [diff] [blame] | 4122 | } |
| 4123 | EXPORT_SYMBOL(blk_mq_alloc_disk_for_queue); |
| 4124 | |
Jianchao Wang | 34d11ff | 2018-10-12 18:07:27 +0800 | [diff] [blame] | 4125 | static struct blk_mq_hw_ctx *blk_mq_alloc_and_init_hctx( |
| 4126 | struct blk_mq_tag_set *set, struct request_queue *q, |
| 4127 | int hctx_idx, int node) |
| 4128 | { |
Ming Lei | 2f8f133 | 2019-04-30 09:52:27 +0800 | [diff] [blame] | 4129 | struct blk_mq_hw_ctx *hctx = NULL, *tmp; |
Jianchao Wang | 34d11ff | 2018-10-12 18:07:27 +0800 | [diff] [blame] | 4130 | |
Ming Lei | 2f8f133 | 2019-04-30 09:52:27 +0800 | [diff] [blame] | 4131 | /* reuse dead hctx first */ |
| 4132 | spin_lock(&q->unused_hctx_lock); |
| 4133 | list_for_each_entry(tmp, &q->unused_hctx_list, hctx_list) { |
| 4134 | if (tmp->numa_node == node) { |
| 4135 | hctx = tmp; |
| 4136 | break; |
| 4137 | } |
| 4138 | } |
| 4139 | if (hctx) |
| 4140 | list_del_init(&hctx->hctx_list); |
| 4141 | spin_unlock(&q->unused_hctx_lock); |
| 4142 | |
| 4143 | if (!hctx) |
| 4144 | hctx = blk_mq_alloc_hctx(q, set, node); |
Jianchao Wang | 34d11ff | 2018-10-12 18:07:27 +0800 | [diff] [blame] | 4145 | if (!hctx) |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 4146 | goto fail; |
Jianchao Wang | 34d11ff | 2018-10-12 18:07:27 +0800 | [diff] [blame] | 4147 | |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 4148 | if (blk_mq_init_hctx(q, set, hctx, hctx_idx)) |
| 4149 | goto free_hctx; |
Jianchao Wang | 34d11ff | 2018-10-12 18:07:27 +0800 | [diff] [blame] | 4150 | |
| 4151 | return hctx; |
Ming Lei | 7c6c5b7 | 2019-04-30 09:52:26 +0800 | [diff] [blame] | 4152 | |
| 4153 | free_hctx: |
| 4154 | kobject_put(&hctx->kobj); |
| 4155 | fail: |
| 4156 | return NULL; |
Jianchao Wang | 34d11ff | 2018-10-12 18:07:27 +0800 | [diff] [blame] | 4157 | } |
| 4158 | |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4159 | static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set, |
| 4160 | struct request_queue *q) |
Mike Snitzer | b62c21b | 2015-03-12 23:56:02 -0400 | [diff] [blame] | 4161 | { |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 4162 | struct blk_mq_hw_ctx *hctx; |
| 4163 | unsigned long i, j; |
Bart Van Assche | ac0d6b9 | 2019-10-25 09:50:09 -0700 | [diff] [blame] | 4164 | |
Ming Lei | fb350e0 | 2018-01-06 16:27:40 +0800 | [diff] [blame] | 4165 | /* protect against switching io scheduler */ |
| 4166 | mutex_lock(&q->sysfs_lock); |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4167 | for (i = 0; i < set->nr_hw_queues; i++) { |
Ming Lei | 306f13ee | 2022-03-08 15:32:15 +0800 | [diff] [blame] | 4168 | int old_node; |
Ming Lei | 4d80513 | 2022-03-08 15:32:14 +0800 | [diff] [blame] | 4169 | int node = blk_mq_get_hctx_node(set, i); |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 4170 | struct blk_mq_hw_ctx *old_hctx = xa_load(&q->hctx_table, i); |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4171 | |
Ming Lei | 306f13ee | 2022-03-08 15:32:15 +0800 | [diff] [blame] | 4172 | if (old_hctx) { |
| 4173 | old_node = old_hctx->numa_node; |
| 4174 | blk_mq_exit_hctx(q, set, old_hctx, i); |
| 4175 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4176 | |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 4177 | if (!blk_mq_alloc_and_init_hctx(set, q, i, node)) { |
Ming Lei | 306f13ee | 2022-03-08 15:32:15 +0800 | [diff] [blame] | 4178 | if (!old_hctx) |
Jianchao Wang | 34d11ff | 2018-10-12 18:07:27 +0800 | [diff] [blame] | 4179 | break; |
Ming Lei | 306f13ee | 2022-03-08 15:32:15 +0800 | [diff] [blame] | 4180 | pr_warn("Allocate new hctx on node %d fails, fallback to previous one on node %d\n", |
| 4181 | node, old_node); |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 4182 | hctx = blk_mq_alloc_and_init_hctx(set, q, i, old_node); |
| 4183 | WARN_ON_ONCE(!hctx); |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4184 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4185 | } |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4186 | /* |
| 4187 | * Increasing nr_hw_queues fails. Free the newly allocated |
| 4188 | * hctxs and keep the previous q->nr_hw_queues. |
| 4189 | */ |
| 4190 | if (i != set->nr_hw_queues) { |
| 4191 | j = q->nr_hw_queues; |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4192 | } else { |
| 4193 | j = i; |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4194 | q->nr_hw_queues = set->nr_hw_queues; |
| 4195 | } |
Jianchao Wang | 34d11ff | 2018-10-12 18:07:27 +0800 | [diff] [blame] | 4196 | |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 4197 | xa_for_each_start(&q->hctx_table, j, hctx, j) |
| 4198 | blk_mq_exit_hctx(q, set, hctx, j); |
Ming Lei | fb350e0 | 2018-01-06 16:27:40 +0800 | [diff] [blame] | 4199 | mutex_unlock(&q->sysfs_lock); |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4200 | } |
| 4201 | |
Ming Lei | 42ee306 | 2022-03-08 15:32:16 +0800 | [diff] [blame] | 4202 | static void blk_mq_update_poll_flag(struct request_queue *q) |
| 4203 | { |
| 4204 | struct blk_mq_tag_set *set = q->tag_set; |
| 4205 | |
| 4206 | if (set->nr_maps > HCTX_TYPE_POLL && |
| 4207 | set->map[HCTX_TYPE_POLL].nr_queues) |
| 4208 | blk_queue_flag_set(QUEUE_FLAG_POLL, q); |
| 4209 | else |
| 4210 | blk_queue_flag_clear(QUEUE_FLAG_POLL, q); |
| 4211 | } |
| 4212 | |
Christoph Hellwig | 26a9750 | 2021-06-02 09:53:17 +0300 | [diff] [blame] | 4213 | int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, |
| 4214 | struct request_queue *q) |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4215 | { |
Ming Lei | 6684167 | 2016-02-12 15:27:00 +0800 | [diff] [blame] | 4216 | /* mark the queue as mq asap */ |
| 4217 | q->mq_ops = set->ops; |
| 4218 | |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4219 | q->poll_cb = blk_stat_alloc_callback(blk_mq_poll_stats_fn, |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 4220 | blk_mq_poll_stats_bkt, |
| 4221 | BLK_MQ_POLL_STATS_BKTS, q); |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4222 | if (!q->poll_cb) |
| 4223 | goto err_exit; |
| 4224 | |
Ming Lei | 1db4909 | 2018-11-20 09:44:35 +0800 | [diff] [blame] | 4225 | if (blk_mq_alloc_ctxs(q)) |
Jes Sorensen | 41de54c | 2019-04-19 16:35:44 -0400 | [diff] [blame] | 4226 | goto err_poll; |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4227 | |
Ming Lei | 737f98c | 2017-02-22 18:13:59 +0800 | [diff] [blame] | 4228 | /* init q->mq_kobj and sw queues' kobjects */ |
| 4229 | blk_mq_sysfs_init(q); |
| 4230 | |
Ming Lei | 2f8f133 | 2019-04-30 09:52:27 +0800 | [diff] [blame] | 4231 | INIT_LIST_HEAD(&q->unused_hctx_list); |
| 4232 | spin_lock_init(&q->unused_hctx_lock); |
| 4233 | |
Ming Lei | 4e5cc99 | 2022-03-08 15:32:19 +0800 | [diff] [blame] | 4234 | xa_init(&q->hctx_table); |
| 4235 | |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4236 | blk_mq_realloc_hw_ctxs(set, q); |
| 4237 | if (!q->nr_hw_queues) |
| 4238 | goto err_hctxs; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4239 | |
Christoph Hellwig | 287922e | 2015-10-30 20:57:30 +0800 | [diff] [blame] | 4240 | INIT_WORK(&q->timeout_work, blk_mq_timeout_work); |
Ming Lei | e56f698 | 2015-07-16 19:53:22 +0800 | [diff] [blame] | 4241 | blk_queue_rq_timeout(q, set->timeout ? set->timeout : 30 * HZ); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4242 | |
Jens Axboe | a890893 | 2018-10-16 14:23:06 -0600 | [diff] [blame] | 4243 | q->tag_set = set; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4244 | |
Jens Axboe | 94eddfb | 2013-11-19 09:25:07 -0700 | [diff] [blame] | 4245 | q->queue_flags |= QUEUE_FLAG_MQ_DEFAULT; |
Ming Lei | 42ee306 | 2022-03-08 15:32:16 +0800 | [diff] [blame] | 4246 | blk_mq_update_poll_flag(q); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4247 | |
Mike Snitzer | 2849450 | 2016-09-14 13:28:30 -0400 | [diff] [blame] | 4248 | INIT_DELAYED_WORK(&q->requeue_work, blk_mq_requeue_work); |
Christoph Hellwig | 6fca6a6 | 2014-05-28 08:08:02 -0600 | [diff] [blame] | 4249 | INIT_LIST_HEAD(&q->requeue_list); |
| 4250 | spin_lock_init(&q->requeue_lock); |
| 4251 | |
Jens Axboe | eba7176 | 2014-05-20 15:17:27 -0600 | [diff] [blame] | 4252 | q->nr_requests = set->queue_depth; |
| 4253 | |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4254 | /* |
| 4255 | * Default to classic polling |
| 4256 | */ |
Yufen Yu | 29ece8b | 2019-03-18 22:44:41 +0800 | [diff] [blame] | 4257 | q->poll_nsec = BLK_MQ_POLL_CLASSIC; |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4258 | |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4259 | blk_mq_init_cpu_queues(q, set->nr_hw_queues); |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 4260 | blk_mq_add_queue_tag_set(set, q); |
Christoph Hellwig | 4b855ad | 2017-06-26 12:20:57 +0200 | [diff] [blame] | 4261 | blk_mq_map_swqueue(q); |
Christoph Hellwig | 26a9750 | 2021-06-02 09:53:17 +0300 | [diff] [blame] | 4262 | return 0; |
Christoph Hellwig | 1874198 | 2014-02-10 09:29:00 -0700 | [diff] [blame] | 4263 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4264 | err_hctxs: |
Chen Jun | 943f45b | 2022-10-31 03:12:42 +0000 | [diff] [blame] | 4265 | blk_mq_release(q); |
Jes Sorensen | 41de54c | 2019-04-19 16:35:44 -0400 | [diff] [blame] | 4266 | err_poll: |
| 4267 | blk_stat_free_callback(q->poll_cb); |
| 4268 | q->poll_cb = NULL; |
Ming Lin | c7de572 | 2016-05-25 23:23:27 -0700 | [diff] [blame] | 4269 | err_exit: |
| 4270 | q->mq_ops = NULL; |
Christoph Hellwig | 26a9750 | 2021-06-02 09:53:17 +0300 | [diff] [blame] | 4271 | return -ENOMEM; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4272 | } |
Mike Snitzer | b62c21b | 2015-03-12 23:56:02 -0400 | [diff] [blame] | 4273 | EXPORT_SYMBOL(blk_mq_init_allocated_queue); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4274 | |
Ming Lei | c7e2d94 | 2019-04-30 09:52:25 +0800 | [diff] [blame] | 4275 | /* tags can _not_ be used after returning from blk_mq_exit_queue */ |
| 4276 | void blk_mq_exit_queue(struct request_queue *q) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4277 | { |
Bart Van Assche | 630ef62 | 2021-05-13 10:15:29 -0700 | [diff] [blame] | 4278 | struct blk_mq_tag_set *set = q->tag_set; |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4279 | |
Bart Van Assche | 630ef62 | 2021-05-13 10:15:29 -0700 | [diff] [blame] | 4280 | /* Checks hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED. */ |
Ming Lei | 624dbe4 | 2014-05-27 23:35:13 +0800 | [diff] [blame] | 4281 | blk_mq_exit_hw_queues(q, set, set->nr_hw_queues); |
Bart Van Assche | 630ef62 | 2021-05-13 10:15:29 -0700 | [diff] [blame] | 4282 | /* May clear BLK_MQ_F_TAG_QUEUE_SHARED in hctx->flags. */ |
| 4283 | blk_mq_del_queue_tag_set(q); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4284 | } |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4285 | |
Jens Axboe | a516440 | 2014-09-10 09:02:03 -0600 | [diff] [blame] | 4286 | static int __blk_mq_alloc_rq_maps(struct blk_mq_tag_set *set) |
| 4287 | { |
| 4288 | int i; |
| 4289 | |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 4290 | if (blk_mq_is_shared_tags(set->flags)) { |
| 4291 | set->shared_tags = blk_mq_alloc_map_and_rqs(set, |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 4292 | BLK_MQ_NO_HCTX_IDX, |
| 4293 | set->queue_depth); |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 4294 | if (!set->shared_tags) |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 4295 | return -ENOMEM; |
| 4296 | } |
| 4297 | |
Xianting Tian | 8229cca | 2020-09-26 10:39:47 +0800 | [diff] [blame] | 4298 | for (i = 0; i < set->nr_hw_queues; i++) { |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 4299 | if (!__blk_mq_alloc_map_and_rqs(set, i)) |
Jens Axboe | a516440 | 2014-09-10 09:02:03 -0600 | [diff] [blame] | 4300 | goto out_unwind; |
Xianting Tian | 8229cca | 2020-09-26 10:39:47 +0800 | [diff] [blame] | 4301 | cond_resched(); |
| 4302 | } |
Jens Axboe | a516440 | 2014-09-10 09:02:03 -0600 | [diff] [blame] | 4303 | |
| 4304 | return 0; |
| 4305 | |
| 4306 | out_unwind: |
| 4307 | while (--i >= 0) |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 4308 | __blk_mq_free_map_and_rqs(set, i); |
| 4309 | |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 4310 | if (blk_mq_is_shared_tags(set->flags)) { |
| 4311 | blk_mq_free_map_and_rqs(set, set->shared_tags, |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 4312 | BLK_MQ_NO_HCTX_IDX); |
John Garry | 645db34 | 2021-10-05 18:23:36 +0800 | [diff] [blame] | 4313 | } |
Jens Axboe | a516440 | 2014-09-10 09:02:03 -0600 | [diff] [blame] | 4314 | |
Jens Axboe | a516440 | 2014-09-10 09:02:03 -0600 | [diff] [blame] | 4315 | return -ENOMEM; |
| 4316 | } |
| 4317 | |
| 4318 | /* |
| 4319 | * Allocate the request maps associated with this tag_set. Note that this |
| 4320 | * may reduce the depth asked for, if memory is tight. set->queue_depth |
| 4321 | * will be updated to reflect the allocated depth. |
| 4322 | */ |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 4323 | static int blk_mq_alloc_set_map_and_rqs(struct blk_mq_tag_set *set) |
Jens Axboe | a516440 | 2014-09-10 09:02:03 -0600 | [diff] [blame] | 4324 | { |
| 4325 | unsigned int depth; |
| 4326 | int err; |
| 4327 | |
| 4328 | depth = set->queue_depth; |
| 4329 | do { |
| 4330 | err = __blk_mq_alloc_rq_maps(set); |
| 4331 | if (!err) |
| 4332 | break; |
| 4333 | |
| 4334 | set->queue_depth >>= 1; |
| 4335 | if (set->queue_depth < set->reserved_tags + BLK_MQ_TAG_MIN) { |
| 4336 | err = -ENOMEM; |
| 4337 | break; |
| 4338 | } |
| 4339 | } while (set->queue_depth); |
| 4340 | |
| 4341 | if (!set->queue_depth || err) { |
| 4342 | pr_err("blk-mq: failed to allocate request map\n"); |
| 4343 | return -ENOMEM; |
| 4344 | } |
| 4345 | |
| 4346 | if (depth != set->queue_depth) |
| 4347 | pr_info("blk-mq: reduced tag depth (%u -> %u)\n", |
| 4348 | depth, set->queue_depth); |
| 4349 | |
| 4350 | return 0; |
| 4351 | } |
| 4352 | |
Bart Van Assche | a4e1d0b | 2022-08-15 10:00:43 -0700 | [diff] [blame] | 4353 | static void blk_mq_update_queue_map(struct blk_mq_tag_set *set) |
Omar Sandoval | ebe8bdd | 2017-04-07 08:53:11 -0600 | [diff] [blame] | 4354 | { |
Bart Van Assche | 6e66b49 | 2020-03-09 21:26:17 -0700 | [diff] [blame] | 4355 | /* |
| 4356 | * blk_mq_map_queues() and multiple .map_queues() implementations |
| 4357 | * expect that set->map[HCTX_TYPE_DEFAULT].nr_queues is set to the |
| 4358 | * number of hardware queues. |
| 4359 | */ |
| 4360 | if (set->nr_maps == 1) |
| 4361 | set->map[HCTX_TYPE_DEFAULT].nr_queues = set->nr_hw_queues; |
| 4362 | |
Ming Lei | 5938870 | 2018-12-07 11:03:53 +0800 | [diff] [blame] | 4363 | if (set->ops->map_queues && !is_kdump_kernel()) { |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4364 | int i; |
| 4365 | |
Ming Lei | 7d4901a | 2018-01-06 16:27:39 +0800 | [diff] [blame] | 4366 | /* |
| 4367 | * transport .map_queues is usually done in the following |
| 4368 | * way: |
| 4369 | * |
| 4370 | * for (queue = 0; queue < set->nr_hw_queues; queue++) { |
| 4371 | * mask = get_cpu_mask(queue) |
| 4372 | * for_each_cpu(cpu, mask) |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4373 | * set->map[x].mq_map[cpu] = queue; |
Ming Lei | 7d4901a | 2018-01-06 16:27:39 +0800 | [diff] [blame] | 4374 | * } |
| 4375 | * |
| 4376 | * When we need to remap, the table has to be cleared for |
| 4377 | * killing stale mapping since one CPU may not be mapped |
| 4378 | * to any hw queue. |
| 4379 | */ |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4380 | for (i = 0; i < set->nr_maps; i++) |
| 4381 | blk_mq_clear_mq_map(&set->map[i]); |
Ming Lei | 7d4901a | 2018-01-06 16:27:39 +0800 | [diff] [blame] | 4382 | |
Bart Van Assche | a4e1d0b | 2022-08-15 10:00:43 -0700 | [diff] [blame] | 4383 | set->ops->map_queues(set); |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4384 | } else { |
| 4385 | BUG_ON(set->nr_maps > 1); |
Bart Van Assche | a4e1d0b | 2022-08-15 10:00:43 -0700 | [diff] [blame] | 4386 | blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]); |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4387 | } |
Omar Sandoval | ebe8bdd | 2017-04-07 08:53:11 -0600 | [diff] [blame] | 4388 | } |
| 4389 | |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4390 | static int blk_mq_realloc_tag_set_tags(struct blk_mq_tag_set *set, |
Christoph Hellwig | ee9d552 | 2022-11-09 11:08:11 +0100 | [diff] [blame] | 4391 | int new_nr_hw_queues) |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4392 | { |
| 4393 | struct blk_mq_tags **new_tags; |
| 4394 | |
Christoph Hellwig | ee9d552 | 2022-11-09 11:08:11 +0100 | [diff] [blame] | 4395 | if (set->nr_hw_queues >= new_nr_hw_queues) |
Shin'ichiro Kawasaki | d4b2e0d | 2022-11-22 17:49:17 +0900 | [diff] [blame] | 4396 | goto done; |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4397 | |
| 4398 | new_tags = kcalloc_node(new_nr_hw_queues, sizeof(struct blk_mq_tags *), |
| 4399 | GFP_KERNEL, set->numa_node); |
| 4400 | if (!new_tags) |
| 4401 | return -ENOMEM; |
| 4402 | |
| 4403 | if (set->tags) |
Christoph Hellwig | ee9d552 | 2022-11-09 11:08:11 +0100 | [diff] [blame] | 4404 | memcpy(new_tags, set->tags, set->nr_hw_queues * |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4405 | sizeof(*set->tags)); |
| 4406 | kfree(set->tags); |
| 4407 | set->tags = new_tags; |
Shin'ichiro Kawasaki | d4b2e0d | 2022-11-22 17:49:17 +0900 | [diff] [blame] | 4408 | done: |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4409 | set->nr_hw_queues = new_nr_hw_queues; |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4410 | return 0; |
| 4411 | } |
| 4412 | |
Jens Axboe | a4391c6 | 2014-06-05 15:21:56 -0600 | [diff] [blame] | 4413 | /* |
| 4414 | * Alloc a tag set to be associated with one or more request queues. |
| 4415 | * May fail with EINVAL for various error conditions. May adjust the |
Minwoo Im | c018c84 | 2018-06-30 22:12:41 +0900 | [diff] [blame] | 4416 | * requested depth down, if it's too large. In that case, the set |
Jens Axboe | a4391c6 | 2014-06-05 15:21:56 -0600 | [diff] [blame] | 4417 | * value will be stored in set->queue_depth. |
| 4418 | */ |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4419 | int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set) |
| 4420 | { |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4421 | int i, ret; |
Christoph Hellwig | da695ba | 2016-09-14 16:18:55 +0200 | [diff] [blame] | 4422 | |
Bart Van Assche | 205fb5f5 | 2014-10-30 14:45:11 +0100 | [diff] [blame] | 4423 | BUILD_BUG_ON(BLK_MQ_MAX_DEPTH > 1 << BLK_MQ_UNIQUE_TAG_BITS); |
| 4424 | |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4425 | if (!set->nr_hw_queues) |
| 4426 | return -EINVAL; |
Jens Axboe | a4391c6 | 2014-06-05 15:21:56 -0600 | [diff] [blame] | 4427 | if (!set->queue_depth) |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4428 | return -EINVAL; |
| 4429 | if (set->queue_depth < set->reserved_tags + BLK_MQ_TAG_MIN) |
| 4430 | return -EINVAL; |
| 4431 | |
Christoph Hellwig | 7d7e0f9 | 2016-09-14 16:18:54 +0200 | [diff] [blame] | 4432 | if (!set->ops->queue_rq) |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4433 | return -EINVAL; |
| 4434 | |
Ming Lei | de14829 | 2017-10-14 17:22:29 +0800 | [diff] [blame] | 4435 | if (!set->ops->get_budget ^ !set->ops->put_budget) |
| 4436 | return -EINVAL; |
| 4437 | |
Jens Axboe | a4391c6 | 2014-06-05 15:21:56 -0600 | [diff] [blame] | 4438 | if (set->queue_depth > BLK_MQ_MAX_DEPTH) { |
| 4439 | pr_info("blk-mq: reduced tag depth to %u\n", |
| 4440 | BLK_MQ_MAX_DEPTH); |
| 4441 | set->queue_depth = BLK_MQ_MAX_DEPTH; |
| 4442 | } |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4443 | |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4444 | if (!set->nr_maps) |
| 4445 | set->nr_maps = 1; |
| 4446 | else if (set->nr_maps > HCTX_MAX_TYPES) |
| 4447 | return -EINVAL; |
| 4448 | |
Shaohua Li | 6637fad | 2014-11-30 16:00:58 -0800 | [diff] [blame] | 4449 | /* |
| 4450 | * If a crashdump is active, then we are potentially in a very |
| 4451 | * memory constrained environment. Limit us to 1 queue and |
| 4452 | * 64 tags to prevent using too much memory. |
| 4453 | */ |
| 4454 | if (is_kdump_kernel()) { |
| 4455 | set->nr_hw_queues = 1; |
Ming Lei | 5938870 | 2018-12-07 11:03:53 +0800 | [diff] [blame] | 4456 | set->nr_maps = 1; |
Shaohua Li | 6637fad | 2014-11-30 16:00:58 -0800 | [diff] [blame] | 4457 | set->queue_depth = min(64U, set->queue_depth); |
| 4458 | } |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4459 | /* |
Jens Axboe | 392546a | 2018-10-29 13:25:27 -0600 | [diff] [blame] | 4460 | * There is no use for more h/w queues than cpus if we just have |
| 4461 | * a single map |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4462 | */ |
Jens Axboe | 392546a | 2018-10-29 13:25:27 -0600 | [diff] [blame] | 4463 | if (set->nr_maps == 1 && set->nr_hw_queues > nr_cpu_ids) |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4464 | set->nr_hw_queues = nr_cpu_ids; |
Shaohua Li | 6637fad | 2014-11-30 16:00:58 -0800 | [diff] [blame] | 4465 | |
Christoph Hellwig | 80bd4a7 | 2022-11-01 16:00:47 +0100 | [diff] [blame] | 4466 | if (set->flags & BLK_MQ_F_BLOCKING) { |
| 4467 | set->srcu = kmalloc(sizeof(*set->srcu), GFP_KERNEL); |
| 4468 | if (!set->srcu) |
| 4469 | return -ENOMEM; |
| 4470 | ret = init_srcu_struct(set->srcu); |
| 4471 | if (ret) |
| 4472 | goto out_free_srcu; |
| 4473 | } |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4474 | |
Christoph Hellwig | da695ba | 2016-09-14 16:18:55 +0200 | [diff] [blame] | 4475 | ret = -ENOMEM; |
Christoph Hellwig | 5ee2029 | 2022-11-09 11:08:10 +0100 | [diff] [blame] | 4476 | set->tags = kcalloc_node(set->nr_hw_queues, |
| 4477 | sizeof(struct blk_mq_tags *), GFP_KERNEL, |
| 4478 | set->numa_node); |
| 4479 | if (!set->tags) |
Christoph Hellwig | 80bd4a7 | 2022-11-01 16:00:47 +0100 | [diff] [blame] | 4480 | goto out_cleanup_srcu; |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4481 | |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4482 | for (i = 0; i < set->nr_maps; i++) { |
| 4483 | set->map[i].mq_map = kcalloc_node(nr_cpu_ids, |
Ming Lei | 07b35eb | 2018-12-17 18:42:45 +0800 | [diff] [blame] | 4484 | sizeof(set->map[i].mq_map[0]), |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4485 | GFP_KERNEL, set->numa_node); |
| 4486 | if (!set->map[i].mq_map) |
| 4487 | goto out_free_mq_map; |
Ming Lei | 5938870 | 2018-12-07 11:03:53 +0800 | [diff] [blame] | 4488 | set->map[i].nr_queues = is_kdump_kernel() ? 1 : set->nr_hw_queues; |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4489 | } |
Christoph Hellwig | bdd17e7 | 2016-09-14 16:18:53 +0200 | [diff] [blame] | 4490 | |
Bart Van Assche | a4e1d0b | 2022-08-15 10:00:43 -0700 | [diff] [blame] | 4491 | blk_mq_update_queue_map(set); |
Christoph Hellwig | da695ba | 2016-09-14 16:18:55 +0200 | [diff] [blame] | 4492 | |
John Garry | 63064be | 2021-10-05 18:23:35 +0800 | [diff] [blame] | 4493 | ret = blk_mq_alloc_set_map_and_rqs(set); |
Christoph Hellwig | da695ba | 2016-09-14 16:18:55 +0200 | [diff] [blame] | 4494 | if (ret) |
Christoph Hellwig | bdd17e7 | 2016-09-14 16:18:53 +0200 | [diff] [blame] | 4495 | goto out_free_mq_map; |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4496 | |
Jens Axboe | 0d2602c | 2014-05-13 15:10:52 -0600 | [diff] [blame] | 4497 | mutex_init(&set->tag_list_lock); |
| 4498 | INIT_LIST_HEAD(&set->tag_list); |
| 4499 | |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4500 | return 0; |
Christoph Hellwig | bdd17e7 | 2016-09-14 16:18:53 +0200 | [diff] [blame] | 4501 | |
| 4502 | out_free_mq_map: |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4503 | for (i = 0; i < set->nr_maps; i++) { |
| 4504 | kfree(set->map[i].mq_map); |
| 4505 | set->map[i].mq_map = NULL; |
| 4506 | } |
Robert Elliott | 5676e7b | 2014-09-02 11:38:44 -0500 | [diff] [blame] | 4507 | kfree(set->tags); |
| 4508 | set->tags = NULL; |
Christoph Hellwig | 80bd4a7 | 2022-11-01 16:00:47 +0100 | [diff] [blame] | 4509 | out_cleanup_srcu: |
| 4510 | if (set->flags & BLK_MQ_F_BLOCKING) |
| 4511 | cleanup_srcu_struct(set->srcu); |
| 4512 | out_free_srcu: |
| 4513 | if (set->flags & BLK_MQ_F_BLOCKING) |
| 4514 | kfree(set->srcu); |
Christoph Hellwig | da695ba | 2016-09-14 16:18:55 +0200 | [diff] [blame] | 4515 | return ret; |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4516 | } |
| 4517 | EXPORT_SYMBOL(blk_mq_alloc_tag_set); |
| 4518 | |
Christoph Hellwig | cdb14e0 | 2021-06-02 09:53:16 +0300 | [diff] [blame] | 4519 | /* allocate and initialize a tagset for a simple single-queue device */ |
| 4520 | int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set, |
| 4521 | const struct blk_mq_ops *ops, unsigned int queue_depth, |
| 4522 | unsigned int set_flags) |
| 4523 | { |
| 4524 | memset(set, 0, sizeof(*set)); |
| 4525 | set->ops = ops; |
| 4526 | set->nr_hw_queues = 1; |
| 4527 | set->nr_maps = 1; |
| 4528 | set->queue_depth = queue_depth; |
| 4529 | set->numa_node = NUMA_NO_NODE; |
| 4530 | set->flags = set_flags; |
| 4531 | return blk_mq_alloc_tag_set(set); |
| 4532 | } |
| 4533 | EXPORT_SYMBOL_GPL(blk_mq_alloc_sq_tag_set); |
| 4534 | |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4535 | void blk_mq_free_tag_set(struct blk_mq_tag_set *set) |
| 4536 | { |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4537 | int i, j; |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4538 | |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4539 | for (i = 0; i < set->nr_hw_queues; i++) |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 4540 | __blk_mq_free_map_and_rqs(set, i); |
Jens Axboe | 484b406 | 2014-05-21 14:01:15 -0600 | [diff] [blame] | 4541 | |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 4542 | if (blk_mq_is_shared_tags(set->flags)) { |
| 4543 | blk_mq_free_map_and_rqs(set, set->shared_tags, |
John Garry | e155b0c | 2021-10-05 18:23:37 +0800 | [diff] [blame] | 4544 | BLK_MQ_NO_HCTX_IDX); |
| 4545 | } |
John Garry | 32bc15a | 2020-08-19 23:20:24 +0800 | [diff] [blame] | 4546 | |
Jens Axboe | b3c661b | 2018-10-30 10:36:06 -0600 | [diff] [blame] | 4547 | for (j = 0; j < set->nr_maps; j++) { |
| 4548 | kfree(set->map[j].mq_map); |
| 4549 | set->map[j].mq_map = NULL; |
| 4550 | } |
Christoph Hellwig | bdd17e7 | 2016-09-14 16:18:53 +0200 | [diff] [blame] | 4551 | |
Ming Lei | 981bd18 | 2014-04-24 00:07:34 +0800 | [diff] [blame] | 4552 | kfree(set->tags); |
Robert Elliott | 5676e7b | 2014-09-02 11:38:44 -0500 | [diff] [blame] | 4553 | set->tags = NULL; |
Christoph Hellwig | 80bd4a7 | 2022-11-01 16:00:47 +0100 | [diff] [blame] | 4554 | if (set->flags & BLK_MQ_F_BLOCKING) { |
| 4555 | cleanup_srcu_struct(set->srcu); |
| 4556 | kfree(set->srcu); |
| 4557 | } |
Christoph Hellwig | 24d2f90 | 2014-04-15 14:14:00 -0600 | [diff] [blame] | 4558 | } |
| 4559 | EXPORT_SYMBOL(blk_mq_free_tag_set); |
| 4560 | |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4561 | int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr) |
| 4562 | { |
| 4563 | struct blk_mq_tag_set *set = q->tag_set; |
| 4564 | struct blk_mq_hw_ctx *hctx; |
Ming Lei | 4f48120 | 2022-03-08 15:32:18 +0800 | [diff] [blame] | 4565 | int ret; |
| 4566 | unsigned long i; |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4567 | |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 4568 | if (!set) |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4569 | return -EINVAL; |
| 4570 | |
Aleksei Zakharov | e5fa814 | 2019-02-08 19:14:05 +0300 | [diff] [blame] | 4571 | if (q->nr_requests == nr) |
| 4572 | return 0; |
| 4573 | |
Jens Axboe | 70f36b6 | 2017-01-19 10:59:07 -0700 | [diff] [blame] | 4574 | blk_mq_freeze_queue(q); |
Ming Lei | 24f5a90 | 2018-01-06 16:27:38 +0800 | [diff] [blame] | 4575 | blk_mq_quiesce_queue(q); |
Jens Axboe | 70f36b6 | 2017-01-19 10:59:07 -0700 | [diff] [blame] | 4576 | |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4577 | ret = 0; |
| 4578 | queue_for_each_hw_ctx(q, hctx, i) { |
Keith Busch | e9137d4 | 2016-02-18 14:56:35 -0700 | [diff] [blame] | 4579 | if (!hctx->tags) |
| 4580 | continue; |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 4581 | /* |
| 4582 | * If we're using an MQ scheduler, just update the scheduler |
| 4583 | * queue depth. This is similar to what the old code would do. |
| 4584 | */ |
John Garry | f6adcef | 2021-10-05 18:23:29 +0800 | [diff] [blame] | 4585 | if (hctx->sched_tags) { |
Jens Axboe | 70f36b6 | 2017-01-19 10:59:07 -0700 | [diff] [blame] | 4586 | ret = blk_mq_tag_update_depth(hctx, &hctx->sched_tags, |
John Garry | f6adcef | 2021-10-05 18:23:29 +0800 | [diff] [blame] | 4587 | nr, true); |
John Garry | f6adcef | 2021-10-05 18:23:29 +0800 | [diff] [blame] | 4588 | } else { |
| 4589 | ret = blk_mq_tag_update_depth(hctx, &hctx->tags, nr, |
| 4590 | false); |
Jens Axboe | 70f36b6 | 2017-01-19 10:59:07 -0700 | [diff] [blame] | 4591 | } |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4592 | if (ret) |
| 4593 | break; |
Jens Axboe | 77f1e0a | 2019-01-18 10:34:16 -0700 | [diff] [blame] | 4594 | if (q->elevator && q->elevator->type->ops.depth_updated) |
| 4595 | q->elevator->type->ops.depth_updated(hctx); |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4596 | } |
John Garry | d97e594 | 2021-05-13 20:00:58 +0800 | [diff] [blame] | 4597 | if (!ret) { |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4598 | q->nr_requests = nr; |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 4599 | if (blk_mq_is_shared_tags(set->flags)) { |
John Garry | 8fa0446 | 2021-10-05 18:23:28 +0800 | [diff] [blame] | 4600 | if (q->elevator) |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 4601 | blk_mq_tag_update_sched_shared_tags(q); |
John Garry | 8fa0446 | 2021-10-05 18:23:28 +0800 | [diff] [blame] | 4602 | else |
John Garry | 079a2e3 | 2021-10-05 18:23:39 +0800 | [diff] [blame] | 4603 | blk_mq_tag_resize_shared_tags(set, nr); |
John Garry | 8fa0446 | 2021-10-05 18:23:28 +0800 | [diff] [blame] | 4604 | } |
John Garry | d97e594 | 2021-05-13 20:00:58 +0800 | [diff] [blame] | 4605 | } |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4606 | |
Ming Lei | 24f5a90 | 2018-01-06 16:27:38 +0800 | [diff] [blame] | 4607 | blk_mq_unquiesce_queue(q); |
Jens Axboe | 70f36b6 | 2017-01-19 10:59:07 -0700 | [diff] [blame] | 4608 | blk_mq_unfreeze_queue(q); |
Jens Axboe | 70f36b6 | 2017-01-19 10:59:07 -0700 | [diff] [blame] | 4609 | |
Jens Axboe | e3a2b3f | 2014-05-20 11:49:02 -0600 | [diff] [blame] | 4610 | return ret; |
| 4611 | } |
| 4612 | |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4613 | /* |
| 4614 | * request_queue and elevator_type pair. |
| 4615 | * It is just used by __blk_mq_update_nr_hw_queues to cache |
| 4616 | * the elevator_type associated with a request_queue. |
| 4617 | */ |
| 4618 | struct blk_mq_qe_pair { |
| 4619 | struct list_head node; |
| 4620 | struct request_queue *q; |
| 4621 | struct elevator_type *type; |
| 4622 | }; |
| 4623 | |
| 4624 | /* |
| 4625 | * Cache the elevator_type in qe pair list and switch the |
| 4626 | * io scheduler to 'none' |
| 4627 | */ |
| 4628 | static bool blk_mq_elv_switch_none(struct list_head *head, |
| 4629 | struct request_queue *q) |
| 4630 | { |
| 4631 | struct blk_mq_qe_pair *qe; |
| 4632 | |
| 4633 | if (!q->elevator) |
| 4634 | return true; |
| 4635 | |
| 4636 | qe = kmalloc(sizeof(*qe), GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY); |
| 4637 | if (!qe) |
| 4638 | return false; |
| 4639 | |
Ming Lei | 5fd7a84 | 2022-06-16 09:43:59 +0800 | [diff] [blame] | 4640 | /* q->elevator needs protection from ->sysfs_lock */ |
| 4641 | mutex_lock(&q->sysfs_lock); |
| 4642 | |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4643 | INIT_LIST_HEAD(&qe->node); |
| 4644 | qe->q = q; |
| 4645 | qe->type = q->elevator->type; |
Christoph Hellwig | dd6f7f1 | 2022-10-20 08:48:16 +0200 | [diff] [blame] | 4646 | /* keep a reference to the elevator module as we'll switch back */ |
| 4647 | __elevator_get(qe->type); |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4648 | list_add(&qe->node, head); |
Christoph Hellwig | 64b3607 | 2022-10-30 11:07:14 +0100 | [diff] [blame] | 4649 | elevator_disable(q); |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4650 | mutex_unlock(&q->sysfs_lock); |
| 4651 | |
| 4652 | return true; |
| 4653 | } |
| 4654 | |
Jakob Koschel | 4a3b666e | 2022-03-31 11:12:18 +0200 | [diff] [blame] | 4655 | static struct blk_mq_qe_pair *blk_lookup_qe_pair(struct list_head *head, |
| 4656 | struct request_queue *q) |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4657 | { |
| 4658 | struct blk_mq_qe_pair *qe; |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4659 | |
| 4660 | list_for_each_entry(qe, head, node) |
Jakob Koschel | 4a3b666e | 2022-03-31 11:12:18 +0200 | [diff] [blame] | 4661 | if (qe->q == q) |
| 4662 | return qe; |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4663 | |
Jakob Koschel | 4a3b666e | 2022-03-31 11:12:18 +0200 | [diff] [blame] | 4664 | return NULL; |
| 4665 | } |
| 4666 | |
| 4667 | static void blk_mq_elv_switch_back(struct list_head *head, |
| 4668 | struct request_queue *q) |
| 4669 | { |
| 4670 | struct blk_mq_qe_pair *qe; |
| 4671 | struct elevator_type *t; |
| 4672 | |
| 4673 | qe = blk_lookup_qe_pair(head, q); |
| 4674 | if (!qe) |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4675 | return; |
Jakob Koschel | 4a3b666e | 2022-03-31 11:12:18 +0200 | [diff] [blame] | 4676 | t = qe->type; |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4677 | list_del(&qe->node); |
| 4678 | kfree(qe); |
| 4679 | |
| 4680 | mutex_lock(&q->sysfs_lock); |
Keith Busch | 8237c01 | 2022-09-27 08:56:52 -0700 | [diff] [blame] | 4681 | elevator_switch(q, t); |
Jinlong Chen | 8ed40ee | 2022-10-20 08:48:19 +0200 | [diff] [blame] | 4682 | /* drop the reference acquired in blk_mq_elv_switch_none */ |
| 4683 | elevator_put(t); |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4684 | mutex_unlock(&q->sysfs_lock); |
| 4685 | } |
| 4686 | |
Keith Busch | e4dc2b3 | 2017-05-30 14:39:11 -0400 | [diff] [blame] | 4687 | static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, |
| 4688 | int nr_hw_queues) |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4689 | { |
| 4690 | struct request_queue *q; |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4691 | LIST_HEAD(head); |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4692 | int prev_nr_hw_queues; |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4693 | |
Bart Van Assche | 705cda9 | 2017-04-07 11:16:49 -0700 | [diff] [blame] | 4694 | lockdep_assert_held(&set->tag_list_lock); |
| 4695 | |
Jens Axboe | 392546a | 2018-10-29 13:25:27 -0600 | [diff] [blame] | 4696 | if (set->nr_maps == 1 && nr_hw_queues > nr_cpu_ids) |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4697 | nr_hw_queues = nr_cpu_ids; |
Weiping Zhang | fe35ec5 | 2020-06-17 14:18:37 +0800 | [diff] [blame] | 4698 | if (nr_hw_queues < 1) |
| 4699 | return; |
| 4700 | if (set->nr_maps == 1 && nr_hw_queues == set->nr_hw_queues) |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4701 | return; |
| 4702 | |
| 4703 | list_for_each_entry(q, &set->tag_list, tag_set_list) |
| 4704 | blk_mq_freeze_queue(q); |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4705 | /* |
| 4706 | * Switch IO scheduler to 'none', cleaning up the data associated |
| 4707 | * with the previous scheduler. We will switch back once we are done |
| 4708 | * updating the new sw to hw queue mappings. |
| 4709 | */ |
| 4710 | list_for_each_entry(q, &set->tag_list, tag_set_list) |
| 4711 | if (!blk_mq_elv_switch_none(&head, q)) |
| 4712 | goto switch_back; |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4713 | |
Jianchao Wang | 477e19d | 2018-10-12 18:07:25 +0800 | [diff] [blame] | 4714 | list_for_each_entry(q, &set->tag_list, tag_set_list) { |
| 4715 | blk_mq_debugfs_unregister_hctxs(q); |
Christoph Hellwig | eaa870f | 2022-06-28 19:18:49 +0200 | [diff] [blame] | 4716 | blk_mq_sysfs_unregister_hctxs(q); |
Jianchao Wang | 477e19d | 2018-10-12 18:07:25 +0800 | [diff] [blame] | 4717 | } |
| 4718 | |
Weiping Zhang | a2584e4 | 2020-05-07 21:03:56 +0800 | [diff] [blame] | 4719 | prev_nr_hw_queues = set->nr_hw_queues; |
Christoph Hellwig | ee9d552 | 2022-11-09 11:08:11 +0100 | [diff] [blame] | 4720 | if (blk_mq_realloc_tag_set_tags(set, nr_hw_queues) < 0) |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4721 | goto reregister; |
| 4722 | |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4723 | fallback: |
Weiping Zhang | aa880ad | 2020-05-13 08:44:05 +0800 | [diff] [blame] | 4724 | blk_mq_update_queue_map(set); |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4725 | list_for_each_entry(q, &set->tag_list, tag_set_list) { |
| 4726 | blk_mq_realloc_hw_ctxs(set, q); |
Ming Lei | 42ee306 | 2022-03-08 15:32:16 +0800 | [diff] [blame] | 4727 | blk_mq_update_poll_flag(q); |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4728 | if (q->nr_hw_queues != set->nr_hw_queues) { |
Ye Bin | a846a8e | 2021-11-08 15:40:19 +0800 | [diff] [blame] | 4729 | int i = prev_nr_hw_queues; |
| 4730 | |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4731 | pr_warn("Increasing nr_hw_queues to %d fails, fallback to %d\n", |
| 4732 | nr_hw_queues, prev_nr_hw_queues); |
Ye Bin | a846a8e | 2021-11-08 15:40:19 +0800 | [diff] [blame] | 4733 | for (; i < set->nr_hw_queues; i++) |
| 4734 | __blk_mq_free_map_and_rqs(set, i); |
| 4735 | |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4736 | set->nr_hw_queues = prev_nr_hw_queues; |
Dongli Zhang | 7d76f85 | 2019-02-27 21:35:01 +0800 | [diff] [blame] | 4737 | blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]); |
Jianchao Wang | e01ad46 | 2018-10-12 18:07:28 +0800 | [diff] [blame] | 4738 | goto fallback; |
| 4739 | } |
Jianchao Wang | 477e19d | 2018-10-12 18:07:25 +0800 | [diff] [blame] | 4740 | blk_mq_map_swqueue(q); |
| 4741 | } |
| 4742 | |
Bart Van Assche | f7e76db | 2019-10-25 09:50:10 -0700 | [diff] [blame] | 4743 | reregister: |
Jianchao Wang | 477e19d | 2018-10-12 18:07:25 +0800 | [diff] [blame] | 4744 | list_for_each_entry(q, &set->tag_list, tag_set_list) { |
Christoph Hellwig | eaa870f | 2022-06-28 19:18:49 +0200 | [diff] [blame] | 4745 | blk_mq_sysfs_register_hctxs(q); |
Jianchao Wang | 477e19d | 2018-10-12 18:07:25 +0800 | [diff] [blame] | 4746 | blk_mq_debugfs_register_hctxs(q); |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4747 | } |
| 4748 | |
Jianchao Wang | d48ece2 | 2018-08-21 15:15:03 +0800 | [diff] [blame] | 4749 | switch_back: |
| 4750 | list_for_each_entry(q, &set->tag_list, tag_set_list) |
| 4751 | blk_mq_elv_switch_back(&head, q); |
| 4752 | |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4753 | list_for_each_entry(q, &set->tag_list, tag_set_list) |
| 4754 | blk_mq_unfreeze_queue(q); |
| 4755 | } |
Keith Busch | e4dc2b3 | 2017-05-30 14:39:11 -0400 | [diff] [blame] | 4756 | |
| 4757 | void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues) |
| 4758 | { |
| 4759 | mutex_lock(&set->tag_list_lock); |
| 4760 | __blk_mq_update_nr_hw_queues(set, nr_hw_queues); |
| 4761 | mutex_unlock(&set->tag_list_lock); |
| 4762 | } |
Keith Busch | 868f2f0 | 2015-12-17 17:08:14 -0700 | [diff] [blame] | 4763 | EXPORT_SYMBOL_GPL(blk_mq_update_nr_hw_queues); |
| 4764 | |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4765 | /* Enable polling stats and return whether they were already enabled. */ |
| 4766 | static bool blk_poll_stats_enable(struct request_queue *q) |
| 4767 | { |
Jens Axboe | 48b5c1f | 2021-11-13 14:03:26 -0700 | [diff] [blame] | 4768 | if (q->poll_stat) |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4769 | return true; |
Jens Axboe | 48b5c1f | 2021-11-13 14:03:26 -0700 | [diff] [blame] | 4770 | |
| 4771 | return blk_stats_alloc_enable(q); |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4772 | } |
| 4773 | |
| 4774 | static void blk_mq_poll_stats_start(struct request_queue *q) |
| 4775 | { |
| 4776 | /* |
| 4777 | * We don't arm the callback if polling stats are not enabled or the |
| 4778 | * callback is already active. |
| 4779 | */ |
Jens Axboe | 48b5c1f | 2021-11-13 14:03:26 -0700 | [diff] [blame] | 4780 | if (!q->poll_stat || blk_stat_is_active(q->poll_cb)) |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4781 | return; |
| 4782 | |
| 4783 | blk_stat_activate_msecs(q->poll_cb, 100); |
| 4784 | } |
| 4785 | |
| 4786 | static void blk_mq_poll_stats_fn(struct blk_stat_callback *cb) |
| 4787 | { |
| 4788 | struct request_queue *q = cb->data; |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 4789 | int bucket; |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4790 | |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 4791 | for (bucket = 0; bucket < BLK_MQ_POLL_STATS_BKTS; bucket++) { |
| 4792 | if (cb->stat[bucket].nr_samples) |
| 4793 | q->poll_stat[bucket] = cb->stat[bucket]; |
| 4794 | } |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4795 | } |
| 4796 | |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4797 | static unsigned long blk_mq_poll_nsecs(struct request_queue *q, |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4798 | struct request *rq) |
| 4799 | { |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4800 | unsigned long ret = 0; |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 4801 | int bucket; |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4802 | |
| 4803 | /* |
| 4804 | * If stats collection isn't on, don't sleep but turn it on for |
| 4805 | * future users |
| 4806 | */ |
Omar Sandoval | 34dbad5 | 2017-03-21 08:56:08 -0700 | [diff] [blame] | 4807 | if (!blk_poll_stats_enable(q)) |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4808 | return 0; |
| 4809 | |
| 4810 | /* |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4811 | * As an optimistic guess, use half of the mean service time |
| 4812 | * for this type of request. We can (and should) make this smarter. |
| 4813 | * For instance, if the completion latencies are tight, we can |
| 4814 | * get closer than just half the mean. This is especially |
| 4815 | * important on devices where the completion latencies are longer |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 4816 | * than ~10 usec. We do use the stats for the relevant IO size |
| 4817 | * if available which does lead to better estimates. |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4818 | */ |
Stephen Bates | 720b8cc | 2017-04-07 06:24:03 -0600 | [diff] [blame] | 4819 | bucket = blk_mq_poll_stats_bkt(rq); |
| 4820 | if (bucket < 0) |
| 4821 | return ret; |
| 4822 | |
| 4823 | if (q->poll_stat[bucket].nr_samples) |
| 4824 | ret = (q->poll_stat[bucket].mean + 1) / 2; |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4825 | |
| 4826 | return ret; |
| 4827 | } |
| 4828 | |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4829 | static bool blk_mq_poll_hybrid(struct request_queue *q, blk_qc_t qc) |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4830 | { |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4831 | struct blk_mq_hw_ctx *hctx = blk_qc_to_hctx(q, qc); |
| 4832 | struct request *rq = blk_qc_to_rq(hctx, qc); |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4833 | struct hrtimer_sleeper hs; |
| 4834 | enum hrtimer_mode mode; |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4835 | unsigned int nsecs; |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4836 | ktime_t kt; |
| 4837 | |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4838 | /* |
| 4839 | * If a request has completed on queue that uses an I/O scheduler, we |
| 4840 | * won't get back a request from blk_qc_to_rq. |
| 4841 | */ |
| 4842 | if (!rq || (rq->rq_flags & RQF_MQ_POLL_SLEPT)) |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4843 | return false; |
| 4844 | |
| 4845 | /* |
Jens Axboe | 1052b8a | 2018-11-26 08:21:49 -0700 | [diff] [blame] | 4846 | * If we get here, hybrid polling is enabled. Hence poll_nsec can be: |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4847 | * |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4848 | * 0: use half of prev avg |
| 4849 | * >0: use this specific value |
| 4850 | */ |
Jens Axboe | 1052b8a | 2018-11-26 08:21:49 -0700 | [diff] [blame] | 4851 | if (q->poll_nsec > 0) |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4852 | nsecs = q->poll_nsec; |
| 4853 | else |
John Garry | cae740a | 2020-02-26 20:10:15 +0800 | [diff] [blame] | 4854 | nsecs = blk_mq_poll_nsecs(q, rq); |
Jens Axboe | 64f1c21 | 2016-11-14 13:03:03 -0700 | [diff] [blame] | 4855 | |
| 4856 | if (!nsecs) |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4857 | return false; |
| 4858 | |
Jens Axboe | 76a86f9 | 2018-01-10 11:30:56 -0700 | [diff] [blame] | 4859 | rq->rq_flags |= RQF_MQ_POLL_SLEPT; |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4860 | |
| 4861 | /* |
| 4862 | * This will be replaced with the stats tracking code, using |
| 4863 | * 'avg_completion_time / 2' as the pre-sleep target. |
| 4864 | */ |
Thomas Gleixner | 8b0e195 | 2016-12-25 12:30:41 +0100 | [diff] [blame] | 4865 | kt = nsecs; |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4866 | |
| 4867 | mode = HRTIMER_MODE_REL; |
Sebastian Andrzej Siewior | dbc1625 | 2019-07-26 20:30:50 +0200 | [diff] [blame] | 4868 | hrtimer_init_sleeper_on_stack(&hs, CLOCK_MONOTONIC, mode); |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4869 | hrtimer_set_expires(&hs.timer, kt); |
| 4870 | |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4871 | do { |
Tejun Heo | 5a61c36 | 2018-01-09 08:29:52 -0800 | [diff] [blame] | 4872 | if (blk_mq_rq_state(rq) == MQ_RQ_COMPLETE) |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4873 | break; |
| 4874 | set_current_state(TASK_UNINTERRUPTIBLE); |
Thomas Gleixner | 9dd8813 | 2019-07-30 21:16:55 +0200 | [diff] [blame] | 4875 | hrtimer_sleeper_start_expires(&hs, mode); |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4876 | if (hs.task) |
| 4877 | io_schedule(); |
| 4878 | hrtimer_cancel(&hs.timer); |
| 4879 | mode = HRTIMER_MODE_ABS; |
| 4880 | } while (hs.task && !signal_pending(current)); |
| 4881 | |
| 4882 | __set_current_state(TASK_RUNNING); |
| 4883 | destroy_hrtimer_on_stack(&hs.timer); |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4884 | |
| 4885 | /* |
| 4886 | * If we sleep, have the caller restart the poll loop to reset the |
| 4887 | * state. Like for the other success return cases, the caller is |
| 4888 | * responsible for checking if the IO completed. If the IO isn't |
| 4889 | * complete, we'll get called again and will go straight to the busy |
| 4890 | * poll loop. |
| 4891 | */ |
Jens Axboe | 06426ad | 2016-11-14 13:01:59 -0700 | [diff] [blame] | 4892 | return true; |
| 4893 | } |
| 4894 | |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4895 | static int blk_mq_poll_classic(struct request_queue *q, blk_qc_t cookie, |
Jens Axboe | 5a72e89 | 2021-10-12 09:24:29 -0600 | [diff] [blame] | 4896 | struct io_comp_batch *iob, unsigned int flags) |
Jens Axboe | bbd7bb7 | 2016-11-04 09:34:34 -0600 | [diff] [blame] | 4897 | { |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4898 | struct blk_mq_hw_ctx *hctx = blk_qc_to_hctx(q, cookie); |
| 4899 | long state = get_current_state(); |
| 4900 | int ret; |
Jens Axboe | 1052b8a | 2018-11-26 08:21:49 -0700 | [diff] [blame] | 4901 | |
Jens Axboe | aa61bec | 2018-11-13 21:32:10 -0700 | [diff] [blame] | 4902 | do { |
Jens Axboe | 5a72e89 | 2021-10-12 09:24:29 -0600 | [diff] [blame] | 4903 | ret = q->mq_ops->poll(hctx, iob); |
Jens Axboe | bbd7bb7 | 2016-11-04 09:34:34 -0600 | [diff] [blame] | 4904 | if (ret > 0) { |
Jens Axboe | 849a370 | 2018-11-16 08:37:34 -0700 | [diff] [blame] | 4905 | __set_current_state(TASK_RUNNING); |
Jens Axboe | 85f4d4b | 2018-11-06 13:30:55 -0700 | [diff] [blame] | 4906 | return ret; |
Jens Axboe | bbd7bb7 | 2016-11-04 09:34:34 -0600 | [diff] [blame] | 4907 | } |
| 4908 | |
| 4909 | if (signal_pending_state(state, current)) |
Jens Axboe | 849a370 | 2018-11-16 08:37:34 -0700 | [diff] [blame] | 4910 | __set_current_state(TASK_RUNNING); |
Peter Zijlstra | b03fbd4 | 2021-06-11 10:28:12 +0200 | [diff] [blame] | 4911 | if (task_is_running(current)) |
Jens Axboe | 85f4d4b | 2018-11-06 13:30:55 -0700 | [diff] [blame] | 4912 | return 1; |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4913 | |
Christoph Hellwig | ef99b2d | 2021-10-12 13:12:19 +0200 | [diff] [blame] | 4914 | if (ret < 0 || (flags & BLK_POLL_ONESHOT)) |
Jens Axboe | bbd7bb7 | 2016-11-04 09:34:34 -0600 | [diff] [blame] | 4915 | break; |
| 4916 | cpu_relax(); |
Jens Axboe | aa61bec | 2018-11-13 21:32:10 -0700 | [diff] [blame] | 4917 | } while (!need_resched()); |
Jens Axboe | bbd7bb7 | 2016-11-04 09:34:34 -0600 | [diff] [blame] | 4918 | |
Nitesh Shetty | 67b4110 | 2018-02-13 21:18:12 +0530 | [diff] [blame] | 4919 | __set_current_state(TASK_RUNNING); |
Jens Axboe | 85f4d4b | 2018-11-06 13:30:55 -0700 | [diff] [blame] | 4920 | return 0; |
Jens Axboe | bbd7bb7 | 2016-11-04 09:34:34 -0600 | [diff] [blame] | 4921 | } |
Jens Axboe | bbd7bb7 | 2016-11-04 09:34:34 -0600 | [diff] [blame] | 4922 | |
Jens Axboe | 5a72e89 | 2021-10-12 09:24:29 -0600 | [diff] [blame] | 4923 | int blk_mq_poll(struct request_queue *q, blk_qc_t cookie, struct io_comp_batch *iob, |
| 4924 | unsigned int flags) |
Jens Axboe | 676141e | 2014-03-20 13:29:18 -0600 | [diff] [blame] | 4925 | { |
Christoph Hellwig | d729cf9 | 2021-10-12 13:12:20 +0200 | [diff] [blame] | 4926 | if (!(flags & BLK_POLL_NOSLEEP) && |
| 4927 | q->poll_nsec != BLK_MQ_POLL_CLASSIC) { |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4928 | if (blk_mq_poll_hybrid(q, cookie)) |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4929 | return 1; |
Christoph Hellwig | c6699d6 | 2021-10-12 13:12:16 +0200 | [diff] [blame] | 4930 | } |
Jens Axboe | 5a72e89 | 2021-10-12 09:24:29 -0600 | [diff] [blame] | 4931 | return blk_mq_poll_classic(q, cookie, iob, flags); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4932 | } |
| 4933 | |
Jens Axboe | 9cf2bab | 2018-10-31 17:01:22 -0600 | [diff] [blame] | 4934 | unsigned int blk_mq_rq_cpu(struct request *rq) |
| 4935 | { |
| 4936 | return rq->mq_ctx->cpu; |
| 4937 | } |
| 4938 | EXPORT_SYMBOL(blk_mq_rq_cpu); |
| 4939 | |
Ming Lei | 2a19b28 | 2021-11-16 09:43:43 +0800 | [diff] [blame] | 4940 | void blk_mq_cancel_work_sync(struct request_queue *q) |
| 4941 | { |
Jinlong Chen | 219cf43 | 2022-10-30 17:47:30 +0800 | [diff] [blame] | 4942 | struct blk_mq_hw_ctx *hctx; |
| 4943 | unsigned long i; |
Ming Lei | 2a19b28 | 2021-11-16 09:43:43 +0800 | [diff] [blame] | 4944 | |
Jinlong Chen | 219cf43 | 2022-10-30 17:47:30 +0800 | [diff] [blame] | 4945 | cancel_delayed_work_sync(&q->requeue_work); |
Ming Lei | 2a19b28 | 2021-11-16 09:43:43 +0800 | [diff] [blame] | 4946 | |
Jinlong Chen | 219cf43 | 2022-10-30 17:47:30 +0800 | [diff] [blame] | 4947 | queue_for_each_hw_ctx(q, hctx, i) |
| 4948 | cancel_delayed_work_sync(&hctx->run_work); |
Ming Lei | 2a19b28 | 2021-11-16 09:43:43 +0800 | [diff] [blame] | 4949 | } |
| 4950 | |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4951 | static int __init blk_mq_init(void) |
| 4952 | { |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 4953 | int i; |
| 4954 | |
| 4955 | for_each_possible_cpu(i) |
Sebastian Andrzej Siewior | f9ab491 | 2021-01-23 21:10:27 +0100 | [diff] [blame] | 4956 | init_llist_head(&per_cpu(blk_cpu_done, i)); |
Christoph Hellwig | c3077b5 | 2020-06-11 08:44:41 +0200 | [diff] [blame] | 4957 | open_softirq(BLOCK_SOFTIRQ, blk_done_softirq); |
| 4958 | |
| 4959 | cpuhp_setup_state_nocalls(CPUHP_BLOCK_SOFTIRQ_DEAD, |
| 4960 | "block/softirq:dead", NULL, |
| 4961 | blk_softirq_cpu_dead); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4962 | cpuhp_setup_state_multi(CPUHP_BLK_MQ_DEAD, "block/mq:dead", NULL, |
| 4963 | blk_mq_hctx_notify_dead); |
Ming Lei | bf0beec | 2020-05-29 15:53:15 +0200 | [diff] [blame] | 4964 | cpuhp_setup_state_multi(CPUHP_AP_BLK_MQ_ONLINE, "block/mq:online", |
| 4965 | blk_mq_hctx_notify_online, |
| 4966 | blk_mq_hctx_notify_offline); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 4967 | return 0; |
| 4968 | } |
| 4969 | subsys_initcall(blk_mq_init); |