Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
1da177e4 LT |
14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
320ae51f | 19 | #include <linux/blk-mq.h> |
1da177e4 LT |
20 | #include <linux/highmem.h> |
21 | #include <linux/mm.h> | |
22 | #include <linux/kernel_stat.h> | |
23 | #include <linux/string.h> | |
24 | #include <linux/init.h> | |
1da177e4 LT |
25 | #include <linux/completion.h> |
26 | #include <linux/slab.h> | |
27 | #include <linux/swap.h> | |
28 | #include <linux/writeback.h> | |
faccbd4b | 29 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 30 | #include <linux/fault-inject.h> |
73c10101 | 31 | #include <linux/list_sort.h> |
e3c78ca5 | 32 | #include <linux/delay.h> |
aaf7c680 | 33 | #include <linux/ratelimit.h> |
6c954667 | 34 | #include <linux/pm_runtime.h> |
eea8f41c | 35 | #include <linux/blk-cgroup.h> |
18fbda91 | 36 | #include <linux/debugfs.h> |
30abb3a6 | 37 | #include <linux/bpf.h> |
55782138 LZ |
38 | |
39 | #define CREATE_TRACE_POINTS | |
40 | #include <trace/events/block.h> | |
1da177e4 | 41 | |
8324aa91 | 42 | #include "blk.h" |
43a5e4e2 | 43 | #include "blk-mq.h" |
bd166ef1 | 44 | #include "blk-mq-sched.h" |
c1c80384 | 45 | #include "blk-rq-qos.h" |
8324aa91 | 46 | |
18fbda91 OS |
47 | #ifdef CONFIG_DEBUG_FS |
48 | struct dentry *blk_debugfs_root; | |
49 | #endif | |
50 | ||
d07335e5 | 51 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 52 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 53 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 54 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 55 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
0bfc2455 | 56 | |
a73f730d TH |
57 | DEFINE_IDA(blk_queue_ida); |
58 | ||
1da177e4 LT |
59 | /* |
60 | * For the allocated request tables | |
61 | */ | |
d674d414 | 62 | struct kmem_cache *request_cachep; |
1da177e4 LT |
63 | |
64 | /* | |
65 | * For queue allocation | |
66 | */ | |
6728cb0e | 67 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 68 | |
1da177e4 LT |
69 | /* |
70 | * Controlling structure to kblockd | |
71 | */ | |
ff856bad | 72 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 73 | |
8814ce8a BVA |
74 | /** |
75 | * blk_queue_flag_set - atomically set a queue flag | |
76 | * @flag: flag to be set | |
77 | * @q: request queue | |
78 | */ | |
79 | void blk_queue_flag_set(unsigned int flag, struct request_queue *q) | |
80 | { | |
81 | unsigned long flags; | |
82 | ||
83 | spin_lock_irqsave(q->queue_lock, flags); | |
84 | queue_flag_set(flag, q); | |
85 | spin_unlock_irqrestore(q->queue_lock, flags); | |
86 | } | |
87 | EXPORT_SYMBOL(blk_queue_flag_set); | |
88 | ||
89 | /** | |
90 | * blk_queue_flag_clear - atomically clear a queue flag | |
91 | * @flag: flag to be cleared | |
92 | * @q: request queue | |
93 | */ | |
94 | void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) | |
95 | { | |
96 | unsigned long flags; | |
97 | ||
98 | spin_lock_irqsave(q->queue_lock, flags); | |
99 | queue_flag_clear(flag, q); | |
100 | spin_unlock_irqrestore(q->queue_lock, flags); | |
101 | } | |
102 | EXPORT_SYMBOL(blk_queue_flag_clear); | |
103 | ||
104 | /** | |
105 | * blk_queue_flag_test_and_set - atomically test and set a queue flag | |
106 | * @flag: flag to be set | |
107 | * @q: request queue | |
108 | * | |
109 | * Returns the previous value of @flag - 0 if the flag was not set and 1 if | |
110 | * the flag was already set. | |
111 | */ | |
112 | bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) | |
113 | { | |
114 | unsigned long flags; | |
115 | bool res; | |
116 | ||
117 | spin_lock_irqsave(q->queue_lock, flags); | |
118 | res = queue_flag_test_and_set(flag, q); | |
119 | spin_unlock_irqrestore(q->queue_lock, flags); | |
120 | ||
121 | return res; | |
122 | } | |
123 | EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); | |
124 | ||
125 | /** | |
126 | * blk_queue_flag_test_and_clear - atomically test and clear a queue flag | |
127 | * @flag: flag to be cleared | |
128 | * @q: request queue | |
129 | * | |
130 | * Returns the previous value of @flag - 0 if the flag was not set and 1 if | |
131 | * the flag was set. | |
132 | */ | |
133 | bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q) | |
134 | { | |
135 | unsigned long flags; | |
136 | bool res; | |
137 | ||
138 | spin_lock_irqsave(q->queue_lock, flags); | |
139 | res = queue_flag_test_and_clear(flag, q); | |
140 | spin_unlock_irqrestore(q->queue_lock, flags); | |
141 | ||
142 | return res; | |
143 | } | |
144 | EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_clear); | |
145 | ||
d40f75a0 TH |
146 | static void blk_clear_congested(struct request_list *rl, int sync) |
147 | { | |
d40f75a0 TH |
148 | #ifdef CONFIG_CGROUP_WRITEBACK |
149 | clear_wb_congested(rl->blkg->wb_congested, sync); | |
150 | #else | |
482cf79c TH |
151 | /* |
152 | * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't | |
153 | * flip its congestion state for events on other blkcgs. | |
154 | */ | |
155 | if (rl == &rl->q->root_rl) | |
dc3b17cc | 156 | clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync); |
d40f75a0 TH |
157 | #endif |
158 | } | |
159 | ||
160 | static void blk_set_congested(struct request_list *rl, int sync) | |
161 | { | |
d40f75a0 TH |
162 | #ifdef CONFIG_CGROUP_WRITEBACK |
163 | set_wb_congested(rl->blkg->wb_congested, sync); | |
164 | #else | |
482cf79c TH |
165 | /* see blk_clear_congested() */ |
166 | if (rl == &rl->q->root_rl) | |
dc3b17cc | 167 | set_wb_congested(rl->q->backing_dev_info->wb.congested, sync); |
d40f75a0 TH |
168 | #endif |
169 | } | |
170 | ||
8324aa91 | 171 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
172 | { |
173 | int nr; | |
174 | ||
175 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
176 | if (nr > q->nr_requests) | |
177 | nr = q->nr_requests; | |
178 | q->nr_congestion_on = nr; | |
179 | ||
180 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
181 | if (nr < 1) | |
182 | nr = 1; | |
183 | q->nr_congestion_off = nr; | |
184 | } | |
185 | ||
2a4aa30c | 186 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 187 | { |
1afb20f3 FT |
188 | memset(rq, 0, sizeof(*rq)); |
189 | ||
1da177e4 | 190 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 191 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 192 | rq->cpu = -1; |
63a71386 | 193 | rq->q = q; |
a2dec7b3 | 194 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
195 | INIT_HLIST_NODE(&rq->hash); |
196 | RB_CLEAR_NODE(&rq->rb_node); | |
63a71386 | 197 | rq->tag = -1; |
bd166ef1 | 198 | rq->internal_tag = -1; |
522a7775 | 199 | rq->start_time_ns = ktime_get_ns(); |
09e099d4 | 200 | rq->part = NULL; |
1da177e4 | 201 | } |
2a4aa30c | 202 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 203 | |
2a842aca CH |
204 | static const struct { |
205 | int errno; | |
206 | const char *name; | |
207 | } blk_errors[] = { | |
208 | [BLK_STS_OK] = { 0, "" }, | |
209 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
210 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
211 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
212 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
213 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
214 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
215 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
216 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
217 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 218 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 219 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
2a842aca | 220 | |
4e4cbee9 CH |
221 | /* device mapper special case, should not leak out: */ |
222 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
223 | ||
2a842aca CH |
224 | /* everything else not covered above: */ |
225 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
226 | }; | |
227 | ||
228 | blk_status_t errno_to_blk_status(int errno) | |
229 | { | |
230 | int i; | |
231 | ||
232 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
233 | if (blk_errors[i].errno == errno) | |
234 | return (__force blk_status_t)i; | |
235 | } | |
236 | ||
237 | return BLK_STS_IOERR; | |
238 | } | |
239 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
240 | ||
241 | int blk_status_to_errno(blk_status_t status) | |
242 | { | |
243 | int idx = (__force int)status; | |
244 | ||
34bd9c1c | 245 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
246 | return -EIO; |
247 | return blk_errors[idx].errno; | |
248 | } | |
249 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
250 | ||
251 | static void print_req_error(struct request *req, blk_status_t status) | |
252 | { | |
253 | int idx = (__force int)status; | |
254 | ||
34bd9c1c | 255 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
256 | return; |
257 | ||
258 | printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n", | |
259 | __func__, blk_errors[idx].name, req->rq_disk ? | |
260 | req->rq_disk->disk_name : "?", | |
261 | (unsigned long long)blk_rq_pos(req)); | |
262 | } | |
263 | ||
5bb23a68 | 264 | static void req_bio_endio(struct request *rq, struct bio *bio, |
2a842aca | 265 | unsigned int nbytes, blk_status_t error) |
1da177e4 | 266 | { |
78d8e58a | 267 | if (error) |
4e4cbee9 | 268 | bio->bi_status = error; |
797e7dbb | 269 | |
e8064021 | 270 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 271 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 272 | |
f79ea416 | 273 | bio_advance(bio, nbytes); |
7ba1ba12 | 274 | |
143a87f4 | 275 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 276 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 277 | bio_endio(bio); |
1da177e4 | 278 | } |
1da177e4 | 279 | |
1da177e4 LT |
280 | void blk_dump_rq_flags(struct request *rq, char *msg) |
281 | { | |
aebf526b CH |
282 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
283 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 284 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 285 | |
83096ebf TH |
286 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
287 | (unsigned long long)blk_rq_pos(rq), | |
288 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
289 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
290 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 291 | } |
1da177e4 LT |
292 | EXPORT_SYMBOL(blk_dump_rq_flags); |
293 | ||
3cca6dc1 | 294 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 295 | { |
3cca6dc1 | 296 | struct request_queue *q; |
1da177e4 | 297 | |
3cca6dc1 JA |
298 | q = container_of(work, struct request_queue, delay_work.work); |
299 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 300 | __blk_run_queue(q); |
3cca6dc1 | 301 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 302 | } |
1da177e4 LT |
303 | |
304 | /** | |
3cca6dc1 JA |
305 | * blk_delay_queue - restart queueing after defined interval |
306 | * @q: The &struct request_queue in question | |
307 | * @msecs: Delay in msecs | |
1da177e4 LT |
308 | * |
309 | * Description: | |
3cca6dc1 JA |
310 | * Sometimes queueing needs to be postponed for a little while, to allow |
311 | * resources to come back. This function will make sure that queueing is | |
2fff8a92 | 312 | * restarted around the specified time. |
3cca6dc1 JA |
313 | */ |
314 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 315 | { |
2fff8a92 | 316 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 317 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 318 | |
70460571 BVA |
319 | if (likely(!blk_queue_dead(q))) |
320 | queue_delayed_work(kblockd_workqueue, &q->delay_work, | |
321 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 322 | } |
3cca6dc1 | 323 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 324 | |
21491412 JA |
325 | /** |
326 | * blk_start_queue_async - asynchronously restart a previously stopped queue | |
327 | * @q: The &struct request_queue in question | |
328 | * | |
329 | * Description: | |
330 | * blk_start_queue_async() will clear the stop flag on the queue, and | |
331 | * ensure that the request_fn for the queue is run from an async | |
332 | * context. | |
333 | **/ | |
334 | void blk_start_queue_async(struct request_queue *q) | |
335 | { | |
2fff8a92 | 336 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 337 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 338 | |
21491412 JA |
339 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
340 | blk_run_queue_async(q); | |
341 | } | |
342 | EXPORT_SYMBOL(blk_start_queue_async); | |
343 | ||
1da177e4 LT |
344 | /** |
345 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 346 | * @q: The &struct request_queue in question |
1da177e4 LT |
347 | * |
348 | * Description: | |
349 | * blk_start_queue() will clear the stop flag on the queue, and call | |
350 | * the request_fn for the queue if it was in a stopped state when | |
2fff8a92 | 351 | * entered. Also see blk_stop_queue(). |
1da177e4 | 352 | **/ |
165125e1 | 353 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 354 | { |
2fff8a92 | 355 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 356 | WARN_ON_ONCE(q->mq_ops); |
a038e253 | 357 | |
75ad23bc | 358 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 359 | __blk_run_queue(q); |
1da177e4 | 360 | } |
1da177e4 LT |
361 | EXPORT_SYMBOL(blk_start_queue); |
362 | ||
363 | /** | |
364 | * blk_stop_queue - stop a queue | |
165125e1 | 365 | * @q: The &struct request_queue in question |
1da177e4 LT |
366 | * |
367 | * Description: | |
368 | * The Linux block layer assumes that a block driver will consume all | |
369 | * entries on the request queue when the request_fn strategy is called. | |
370 | * Often this will not happen, because of hardware limitations (queue | |
371 | * depth settings). If a device driver gets a 'queue full' response, | |
372 | * or if it simply chooses not to queue more I/O at one point, it can | |
373 | * call this function to prevent the request_fn from being called until | |
374 | * the driver has signalled it's ready to go again. This happens by calling | |
2fff8a92 | 375 | * blk_start_queue() to restart queue operations. |
1da177e4 | 376 | **/ |
165125e1 | 377 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 378 | { |
2fff8a92 | 379 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 380 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 381 | |
136b5721 | 382 | cancel_delayed_work(&q->delay_work); |
75ad23bc | 383 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
384 | } |
385 | EXPORT_SYMBOL(blk_stop_queue); | |
386 | ||
387 | /** | |
388 | * blk_sync_queue - cancel any pending callbacks on a queue | |
389 | * @q: the queue | |
390 | * | |
391 | * Description: | |
392 | * The block layer may perform asynchronous callback activity | |
393 | * on a queue, such as calling the unplug function after a timeout. | |
394 | * A block device may call blk_sync_queue to ensure that any | |
395 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 396 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
397 | * that its ->make_request_fn will not re-add plugging prior to calling |
398 | * this function. | |
399 | * | |
da527770 | 400 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 401 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 402 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 403 | * |
1da177e4 LT |
404 | */ |
405 | void blk_sync_queue(struct request_queue *q) | |
406 | { | |
70ed28b9 | 407 | del_timer_sync(&q->timeout); |
4e9b6f20 | 408 | cancel_work_sync(&q->timeout_work); |
f04c1fe7 ML |
409 | |
410 | if (q->mq_ops) { | |
411 | struct blk_mq_hw_ctx *hctx; | |
412 | int i; | |
413 | ||
aba7afc5 | 414 | cancel_delayed_work_sync(&q->requeue_work); |
21c6e939 | 415 | queue_for_each_hw_ctx(q, hctx, i) |
9f993737 | 416 | cancel_delayed_work_sync(&hctx->run_work); |
f04c1fe7 ML |
417 | } else { |
418 | cancel_delayed_work_sync(&q->delay_work); | |
419 | } | |
1da177e4 LT |
420 | } |
421 | EXPORT_SYMBOL(blk_sync_queue); | |
422 | ||
c9254f2d BVA |
423 | /** |
424 | * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY | |
425 | * @q: request queue pointer | |
426 | * | |
427 | * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not | |
428 | * set and 1 if the flag was already set. | |
429 | */ | |
430 | int blk_set_preempt_only(struct request_queue *q) | |
431 | { | |
8814ce8a | 432 | return blk_queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q); |
c9254f2d BVA |
433 | } |
434 | EXPORT_SYMBOL_GPL(blk_set_preempt_only); | |
435 | ||
436 | void blk_clear_preempt_only(struct request_queue *q) | |
437 | { | |
8814ce8a | 438 | blk_queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q); |
3a0a5299 | 439 | wake_up_all(&q->mq_freeze_wq); |
c9254f2d BVA |
440 | } |
441 | EXPORT_SYMBOL_GPL(blk_clear_preempt_only); | |
442 | ||
c246e80d BVA |
443 | /** |
444 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
445 | * @q: The queue to run | |
446 | * | |
447 | * Description: | |
448 | * Invoke request handling on a queue if there are any pending requests. | |
449 | * May be used to restart request handling after a request has completed. | |
450 | * This variant runs the queue whether or not the queue has been | |
451 | * stopped. Must be called with the queue lock held and interrupts | |
452 | * disabled. See also @blk_run_queue. | |
453 | */ | |
454 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
455 | { | |
2fff8a92 | 456 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 457 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 458 | |
c246e80d BVA |
459 | if (unlikely(blk_queue_dead(q))) |
460 | return; | |
461 | ||
24faf6f6 BVA |
462 | /* |
463 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
464 | * the queue lock internally. As a result multiple threads may be | |
465 | * running such a request function concurrently. Keep track of the | |
466 | * number of active request_fn invocations such that blk_drain_queue() | |
467 | * can wait until all these request_fn calls have finished. | |
468 | */ | |
469 | q->request_fn_active++; | |
c246e80d | 470 | q->request_fn(q); |
24faf6f6 | 471 | q->request_fn_active--; |
c246e80d | 472 | } |
a7928c15 | 473 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); |
c246e80d | 474 | |
1da177e4 | 475 | /** |
80a4b58e | 476 | * __blk_run_queue - run a single device queue |
1da177e4 | 477 | * @q: The queue to run |
80a4b58e JA |
478 | * |
479 | * Description: | |
2fff8a92 | 480 | * See @blk_run_queue. |
1da177e4 | 481 | */ |
24ecfbe2 | 482 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 483 | { |
2fff8a92 | 484 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 485 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 486 | |
a538cd03 TH |
487 | if (unlikely(blk_queue_stopped(q))) |
488 | return; | |
489 | ||
c246e80d | 490 | __blk_run_queue_uncond(q); |
75ad23bc NP |
491 | } |
492 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 493 | |
24ecfbe2 CH |
494 | /** |
495 | * blk_run_queue_async - run a single device queue in workqueue context | |
496 | * @q: The queue to run | |
497 | * | |
498 | * Description: | |
499 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
2fff8a92 BVA |
500 | * of us. |
501 | * | |
502 | * Note: | |
503 | * Since it is not allowed to run q->delay_work after blk_cleanup_queue() | |
504 | * has canceled q->delay_work, callers must hold the queue lock to avoid | |
505 | * race conditions between blk_cleanup_queue() and blk_run_queue_async(). | |
24ecfbe2 CH |
506 | */ |
507 | void blk_run_queue_async(struct request_queue *q) | |
508 | { | |
2fff8a92 | 509 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 510 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 511 | |
70460571 | 512 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) |
e7c2f967 | 513 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
24ecfbe2 | 514 | } |
c21e6beb | 515 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 516 | |
75ad23bc NP |
517 | /** |
518 | * blk_run_queue - run a single device queue | |
519 | * @q: The queue to run | |
80a4b58e JA |
520 | * |
521 | * Description: | |
522 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 523 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
524 | */ |
525 | void blk_run_queue(struct request_queue *q) | |
526 | { | |
527 | unsigned long flags; | |
528 | ||
332ebbf7 BVA |
529 | WARN_ON_ONCE(q->mq_ops); |
530 | ||
75ad23bc | 531 | spin_lock_irqsave(q->queue_lock, flags); |
24ecfbe2 | 532 | __blk_run_queue(q); |
1da177e4 LT |
533 | spin_unlock_irqrestore(q->queue_lock, flags); |
534 | } | |
535 | EXPORT_SYMBOL(blk_run_queue); | |
536 | ||
165125e1 | 537 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
538 | { |
539 | kobject_put(&q->kobj); | |
540 | } | |
d86e0e83 | 541 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 542 | |
e3c78ca5 | 543 | /** |
807592a4 | 544 | * __blk_drain_queue - drain requests from request_queue |
e3c78ca5 | 545 | * @q: queue to drain |
c9a929dd | 546 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 547 | * |
c9a929dd TH |
548 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
549 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
550 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 551 | */ |
807592a4 BVA |
552 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) |
553 | __releases(q->queue_lock) | |
554 | __acquires(q->queue_lock) | |
e3c78ca5 | 555 | { |
458f27a9 AH |
556 | int i; |
557 | ||
807592a4 | 558 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 559 | WARN_ON_ONCE(q->mq_ops); |
807592a4 | 560 | |
e3c78ca5 | 561 | while (true) { |
481a7d64 | 562 | bool drain = false; |
e3c78ca5 | 563 | |
b855b04a TH |
564 | /* |
565 | * The caller might be trying to drain @q before its | |
566 | * elevator is initialized. | |
567 | */ | |
568 | if (q->elevator) | |
569 | elv_drain_elevator(q); | |
570 | ||
5efd6113 | 571 | blkcg_drain_queue(q); |
e3c78ca5 | 572 | |
4eabc941 TH |
573 | /* |
574 | * This function might be called on a queue which failed | |
b855b04a TH |
575 | * driver init after queue creation or is not yet fully |
576 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
577 | * in such cases. Kick queue iff dispatch queue has | |
578 | * something on it and @q has request_fn set. | |
4eabc941 | 579 | */ |
b855b04a | 580 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 581 | __blk_run_queue(q); |
c9a929dd | 582 | |
8a5ecdd4 | 583 | drain |= q->nr_rqs_elvpriv; |
24faf6f6 | 584 | drain |= q->request_fn_active; |
481a7d64 TH |
585 | |
586 | /* | |
587 | * Unfortunately, requests are queued at and tracked from | |
588 | * multiple places and there's no single counter which can | |
589 | * be drained. Check all the queues and counters. | |
590 | */ | |
591 | if (drain_all) { | |
e97c293c | 592 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); |
481a7d64 TH |
593 | drain |= !list_empty(&q->queue_head); |
594 | for (i = 0; i < 2; i++) { | |
8a5ecdd4 | 595 | drain |= q->nr_rqs[i]; |
481a7d64 | 596 | drain |= q->in_flight[i]; |
7c94e1c1 ML |
597 | if (fq) |
598 | drain |= !list_empty(&fq->flush_queue[i]); | |
481a7d64 TH |
599 | } |
600 | } | |
e3c78ca5 | 601 | |
481a7d64 | 602 | if (!drain) |
e3c78ca5 | 603 | break; |
807592a4 BVA |
604 | |
605 | spin_unlock_irq(q->queue_lock); | |
606 | ||
e3c78ca5 | 607 | msleep(10); |
807592a4 BVA |
608 | |
609 | spin_lock_irq(q->queue_lock); | |
e3c78ca5 | 610 | } |
458f27a9 AH |
611 | |
612 | /* | |
613 | * With queue marked dead, any woken up waiter will fail the | |
614 | * allocation path, so the wakeup chaining is lost and we're | |
615 | * left with hung waiters. We need to wake up those waiters. | |
616 | */ | |
617 | if (q->request_fn) { | |
a051661c TH |
618 | struct request_list *rl; |
619 | ||
a051661c TH |
620 | blk_queue_for_each_rl(rl, q) |
621 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
622 | wake_up_all(&rl->wait[i]); | |
458f27a9 | 623 | } |
e3c78ca5 TH |
624 | } |
625 | ||
454be724 ML |
626 | void blk_drain_queue(struct request_queue *q) |
627 | { | |
628 | spin_lock_irq(q->queue_lock); | |
629 | __blk_drain_queue(q, true); | |
630 | spin_unlock_irq(q->queue_lock); | |
631 | } | |
632 | ||
d732580b TH |
633 | /** |
634 | * blk_queue_bypass_start - enter queue bypass mode | |
635 | * @q: queue of interest | |
636 | * | |
637 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
638 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 639 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
640 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
641 | * inside queue or RCU read lock. | |
d732580b TH |
642 | */ |
643 | void blk_queue_bypass_start(struct request_queue *q) | |
644 | { | |
332ebbf7 BVA |
645 | WARN_ON_ONCE(q->mq_ops); |
646 | ||
d732580b | 647 | spin_lock_irq(q->queue_lock); |
776687bc | 648 | q->bypass_depth++; |
d732580b TH |
649 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
650 | spin_unlock_irq(q->queue_lock); | |
651 | ||
776687bc TH |
652 | /* |
653 | * Queues start drained. Skip actual draining till init is | |
654 | * complete. This avoids lenghty delays during queue init which | |
655 | * can happen many times during boot. | |
656 | */ | |
657 | if (blk_queue_init_done(q)) { | |
807592a4 BVA |
658 | spin_lock_irq(q->queue_lock); |
659 | __blk_drain_queue(q, false); | |
660 | spin_unlock_irq(q->queue_lock); | |
661 | ||
b82d4b19 TH |
662 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ |
663 | synchronize_rcu(); | |
664 | } | |
d732580b TH |
665 | } |
666 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
667 | ||
668 | /** | |
669 | * blk_queue_bypass_end - leave queue bypass mode | |
670 | * @q: queue of interest | |
671 | * | |
672 | * Leave bypass mode and restore the normal queueing behavior. | |
332ebbf7 BVA |
673 | * |
674 | * Note: although blk_queue_bypass_start() is only called for blk-sq queues, | |
675 | * this function is called for both blk-sq and blk-mq queues. | |
d732580b TH |
676 | */ |
677 | void blk_queue_bypass_end(struct request_queue *q) | |
678 | { | |
679 | spin_lock_irq(q->queue_lock); | |
680 | if (!--q->bypass_depth) | |
681 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
682 | WARN_ON_ONCE(q->bypass_depth < 0); | |
683 | spin_unlock_irq(q->queue_lock); | |
684 | } | |
685 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
686 | ||
aed3ea94 JA |
687 | void blk_set_queue_dying(struct request_queue *q) |
688 | { | |
8814ce8a | 689 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); |
aed3ea94 | 690 | |
d3cfb2a0 ML |
691 | /* |
692 | * When queue DYING flag is set, we need to block new req | |
693 | * entering queue, so we call blk_freeze_queue_start() to | |
694 | * prevent I/O from crossing blk_queue_enter(). | |
695 | */ | |
696 | blk_freeze_queue_start(q); | |
697 | ||
aed3ea94 JA |
698 | if (q->mq_ops) |
699 | blk_mq_wake_waiters(q); | |
700 | else { | |
701 | struct request_list *rl; | |
702 | ||
bbfc3c5d | 703 | spin_lock_irq(q->queue_lock); |
aed3ea94 JA |
704 | blk_queue_for_each_rl(rl, q) { |
705 | if (rl->rq_pool) { | |
34d9715a ML |
706 | wake_up_all(&rl->wait[BLK_RW_SYNC]); |
707 | wake_up_all(&rl->wait[BLK_RW_ASYNC]); | |
aed3ea94 JA |
708 | } |
709 | } | |
bbfc3c5d | 710 | spin_unlock_irq(q->queue_lock); |
aed3ea94 | 711 | } |
055f6e18 ML |
712 | |
713 | /* Make blk_queue_enter() reexamine the DYING flag. */ | |
714 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 JA |
715 | } |
716 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
717 | ||
4cf6324b BVA |
718 | /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */ |
719 | void blk_exit_queue(struct request_queue *q) | |
720 | { | |
721 | /* | |
722 | * Since the I/O scheduler exit code may access cgroup information, | |
723 | * perform I/O scheduler exit before disassociating from the block | |
724 | * cgroup controller. | |
725 | */ | |
726 | if (q->elevator) { | |
727 | ioc_clear_queue(q); | |
728 | elevator_exit(q, q->elevator); | |
729 | q->elevator = NULL; | |
730 | } | |
731 | ||
732 | /* | |
733 | * Remove all references to @q from the block cgroup controller before | |
734 | * restoring @q->queue_lock to avoid that restoring this pointer causes | |
735 | * e.g. blkcg_print_blkgs() to crash. | |
736 | */ | |
737 | blkcg_exit_queue(q); | |
738 | ||
739 | /* | |
740 | * Since the cgroup code may dereference the @q->backing_dev_info | |
741 | * pointer, only decrease its reference count after having removed the | |
742 | * association with the block cgroup controller. | |
743 | */ | |
744 | bdi_put(q->backing_dev_info); | |
745 | } | |
746 | ||
c9a929dd TH |
747 | /** |
748 | * blk_cleanup_queue - shutdown a request queue | |
749 | * @q: request queue to shutdown | |
750 | * | |
c246e80d BVA |
751 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
752 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 753 | */ |
6728cb0e | 754 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 755 | { |
c9a929dd | 756 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 757 | |
3f3299d5 | 758 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
483f4afc | 759 | mutex_lock(&q->sysfs_lock); |
aed3ea94 | 760 | blk_set_queue_dying(q); |
c9a929dd | 761 | spin_lock_irq(lock); |
6ecf23af | 762 | |
80fd9979 | 763 | /* |
3f3299d5 | 764 | * A dying queue is permanently in bypass mode till released. Note |
80fd9979 TH |
765 | * that, unlike blk_queue_bypass_start(), we aren't performing |
766 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
767 | * as some drivers create and destroy a lot of queues while | |
768 | * probing. This is still safe because blk_release_queue() will be | |
769 | * called only after the queue refcnt drops to zero and nothing, | |
770 | * RCU or not, would be traversing the queue by then. | |
771 | */ | |
6ecf23af TH |
772 | q->bypass_depth++; |
773 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
774 | ||
c9a929dd TH |
775 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
776 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
3f3299d5 | 777 | queue_flag_set(QUEUE_FLAG_DYING, q); |
c9a929dd TH |
778 | spin_unlock_irq(lock); |
779 | mutex_unlock(&q->sysfs_lock); | |
780 | ||
c246e80d BVA |
781 | /* |
782 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
783 | * prevent that q->request_fn() gets invoked after draining finished. | |
784 | */ | |
3ef28e83 | 785 | blk_freeze_queue(q); |
9c1051aa | 786 | spin_lock_irq(lock); |
c246e80d | 787 | queue_flag_set(QUEUE_FLAG_DEAD, q); |
807592a4 | 788 | spin_unlock_irq(lock); |
c9a929dd | 789 | |
c2856ae2 ML |
790 | /* |
791 | * make sure all in-progress dispatch are completed because | |
792 | * blk_freeze_queue() can only complete all requests, and | |
793 | * dispatch may still be in-progress since we dispatch requests | |
1311326c ML |
794 | * from more than one contexts. |
795 | * | |
796 | * No need to quiesce queue if it isn't initialized yet since | |
797 | * blk_freeze_queue() should be enough for cases of passthrough | |
798 | * request. | |
c2856ae2 | 799 | */ |
1311326c | 800 | if (q->mq_ops && blk_queue_init_done(q)) |
c2856ae2 ML |
801 | blk_mq_quiesce_queue(q); |
802 | ||
5a48fc14 DW |
803 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
804 | blk_flush_integrity(); | |
805 | ||
c9a929dd | 806 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 807 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
808 | blk_sync_queue(q); |
809 | ||
a063057d BVA |
810 | /* |
811 | * I/O scheduler exit is only safe after the sysfs scheduler attribute | |
812 | * has been removed. | |
813 | */ | |
814 | WARN_ON_ONCE(q->kobj.state_in_sysfs); | |
815 | ||
4cf6324b | 816 | blk_exit_queue(q); |
a063057d | 817 | |
45a9c9d9 BVA |
818 | if (q->mq_ops) |
819 | blk_mq_free_queue(q); | |
3ef28e83 | 820 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 821 | |
5e5cfac0 AH |
822 | spin_lock_irq(lock); |
823 | if (q->queue_lock != &q->__queue_lock) | |
824 | q->queue_lock = &q->__queue_lock; | |
825 | spin_unlock_irq(lock); | |
826 | ||
c9a929dd | 827 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
828 | blk_put_queue(q); |
829 | } | |
1da177e4 LT |
830 | EXPORT_SYMBOL(blk_cleanup_queue); |
831 | ||
271508db | 832 | /* Allocate memory local to the request queue */ |
6d247d7f | 833 | static void *alloc_request_simple(gfp_t gfp_mask, void *data) |
271508db | 834 | { |
6d247d7f CH |
835 | struct request_queue *q = data; |
836 | ||
837 | return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node); | |
271508db DR |
838 | } |
839 | ||
6d247d7f | 840 | static void free_request_simple(void *element, void *data) |
271508db DR |
841 | { |
842 | kmem_cache_free(request_cachep, element); | |
843 | } | |
844 | ||
6d247d7f CH |
845 | static void *alloc_request_size(gfp_t gfp_mask, void *data) |
846 | { | |
847 | struct request_queue *q = data; | |
848 | struct request *rq; | |
849 | ||
850 | rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask, | |
851 | q->node); | |
852 | if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) { | |
853 | kfree(rq); | |
854 | rq = NULL; | |
855 | } | |
856 | return rq; | |
857 | } | |
858 | ||
859 | static void free_request_size(void *element, void *data) | |
860 | { | |
861 | struct request_queue *q = data; | |
862 | ||
863 | if (q->exit_rq_fn) | |
864 | q->exit_rq_fn(q, element); | |
865 | kfree(element); | |
866 | } | |
867 | ||
5b788ce3 TH |
868 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
869 | gfp_t gfp_mask) | |
1da177e4 | 870 | { |
85acb3ba | 871 | if (unlikely(rl->rq_pool) || q->mq_ops) |
1abec4fd MS |
872 | return 0; |
873 | ||
5b788ce3 | 874 | rl->q = q; |
1faa16d2 JA |
875 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
876 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
1faa16d2 JA |
877 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
878 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 879 | |
6d247d7f CH |
880 | if (q->cmd_size) { |
881 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
882 | alloc_request_size, free_request_size, | |
883 | q, gfp_mask, q->node); | |
884 | } else { | |
885 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
886 | alloc_request_simple, free_request_simple, | |
887 | q, gfp_mask, q->node); | |
888 | } | |
1da177e4 LT |
889 | if (!rl->rq_pool) |
890 | return -ENOMEM; | |
891 | ||
b425e504 BVA |
892 | if (rl != &q->root_rl) |
893 | WARN_ON_ONCE(!blk_get_queue(q)); | |
894 | ||
1da177e4 LT |
895 | return 0; |
896 | } | |
897 | ||
b425e504 | 898 | void blk_exit_rl(struct request_queue *q, struct request_list *rl) |
5b788ce3 | 899 | { |
b425e504 | 900 | if (rl->rq_pool) { |
5b788ce3 | 901 | mempool_destroy(rl->rq_pool); |
b425e504 BVA |
902 | if (rl != &q->root_rl) |
903 | blk_put_queue(q); | |
904 | } | |
5b788ce3 TH |
905 | } |
906 | ||
165125e1 | 907 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 908 | { |
5ee0524b | 909 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE, NULL); |
1946089a CL |
910 | } |
911 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 912 | |
3a0a5299 BVA |
913 | /** |
914 | * blk_queue_enter() - try to increase q->q_usage_counter | |
915 | * @q: request queue pointer | |
916 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT | |
917 | */ | |
9a95e4ef | 918 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 919 | { |
3a0a5299 BVA |
920 | const bool preempt = flags & BLK_MQ_REQ_PREEMPT; |
921 | ||
3ef28e83 | 922 | while (true) { |
3a0a5299 | 923 | bool success = false; |
3ef28e83 | 924 | |
818e0fa2 | 925 | rcu_read_lock(); |
3a0a5299 BVA |
926 | if (percpu_ref_tryget_live(&q->q_usage_counter)) { |
927 | /* | |
928 | * The code that sets the PREEMPT_ONLY flag is | |
929 | * responsible for ensuring that that flag is globally | |
930 | * visible before the queue is unfrozen. | |
931 | */ | |
932 | if (preempt || !blk_queue_preempt_only(q)) { | |
933 | success = true; | |
934 | } else { | |
935 | percpu_ref_put(&q->q_usage_counter); | |
936 | } | |
937 | } | |
818e0fa2 | 938 | rcu_read_unlock(); |
3a0a5299 BVA |
939 | |
940 | if (success) | |
3ef28e83 DW |
941 | return 0; |
942 | ||
3a0a5299 | 943 | if (flags & BLK_MQ_REQ_NOWAIT) |
3ef28e83 DW |
944 | return -EBUSY; |
945 | ||
5ed61d3f | 946 | /* |
1671d522 | 947 | * read pair of barrier in blk_freeze_queue_start(), |
5ed61d3f | 948 | * we need to order reading __PERCPU_REF_DEAD flag of |
d3cfb2a0 ML |
949 | * .q_usage_counter and reading .mq_freeze_depth or |
950 | * queue dying flag, otherwise the following wait may | |
951 | * never return if the two reads are reordered. | |
5ed61d3f ML |
952 | */ |
953 | smp_rmb(); | |
954 | ||
1dc3039b AJ |
955 | wait_event(q->mq_freeze_wq, |
956 | (atomic_read(&q->mq_freeze_depth) == 0 && | |
957 | (preempt || !blk_queue_preempt_only(q))) || | |
958 | blk_queue_dying(q)); | |
3ef28e83 DW |
959 | if (blk_queue_dying(q)) |
960 | return -ENODEV; | |
3ef28e83 DW |
961 | } |
962 | } | |
963 | ||
964 | void blk_queue_exit(struct request_queue *q) | |
965 | { | |
966 | percpu_ref_put(&q->q_usage_counter); | |
967 | } | |
968 | ||
969 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
970 | { | |
971 | struct request_queue *q = | |
972 | container_of(ref, struct request_queue, q_usage_counter); | |
973 | ||
974 | wake_up_all(&q->mq_freeze_wq); | |
975 | } | |
976 | ||
bca237a5 | 977 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 978 | { |
bca237a5 | 979 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
980 | |
981 | kblockd_schedule_work(&q->timeout_work); | |
982 | } | |
983 | ||
498f6650 BVA |
984 | /** |
985 | * blk_alloc_queue_node - allocate a request queue | |
986 | * @gfp_mask: memory allocation flags | |
987 | * @node_id: NUMA node to allocate memory from | |
988 | * @lock: For legacy queues, pointer to a spinlock that will be used to e.g. | |
989 | * serialize calls to the legacy .request_fn() callback. Ignored for | |
990 | * blk-mq request queues. | |
991 | * | |
992 | * Note: pass the queue lock as the third argument to this function instead of | |
993 | * setting the queue lock pointer explicitly to avoid triggering a sporadic | |
994 | * crash in the blkcg code. This function namely calls blkcg_init_queue() and | |
995 | * the queue lock pointer must be set before blkcg_init_queue() is called. | |
996 | */ | |
5ee0524b BVA |
997 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id, |
998 | spinlock_t *lock) | |
1946089a | 999 | { |
165125e1 | 1000 | struct request_queue *q; |
338aa96d | 1001 | int ret; |
1946089a | 1002 | |
8324aa91 | 1003 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 1004 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
1005 | if (!q) |
1006 | return NULL; | |
1007 | ||
cbf62af3 CH |
1008 | INIT_LIST_HEAD(&q->queue_head); |
1009 | q->last_merge = NULL; | |
1010 | q->end_sector = 0; | |
1011 | q->boundary_rq = NULL; | |
1012 | ||
00380a40 | 1013 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 1014 | if (q->id < 0) |
3d2936f4 | 1015 | goto fail_q; |
a73f730d | 1016 | |
338aa96d KO |
1017 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
1018 | if (ret) | |
54efd50b KO |
1019 | goto fail_id; |
1020 | ||
d03f6cdc JK |
1021 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); |
1022 | if (!q->backing_dev_info) | |
1023 | goto fail_split; | |
1024 | ||
a83b576c JA |
1025 | q->stats = blk_alloc_queue_stats(); |
1026 | if (!q->stats) | |
1027 | goto fail_stats; | |
1028 | ||
dc3b17cc | 1029 | q->backing_dev_info->ra_pages = |
09cbfeaf | 1030 | (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; |
dc3b17cc JK |
1031 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; |
1032 | q->backing_dev_info->name = "block"; | |
5151412d | 1033 | q->node = node_id; |
0989a025 | 1034 | |
bca237a5 KC |
1035 | timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, |
1036 | laptop_mode_timer_fn, 0); | |
1037 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); | |
4e9b6f20 | 1038 | INIT_WORK(&q->timeout_work, NULL); |
242f9dcb | 1039 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 1040 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 1041 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 1042 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 1043 | #endif |
3cca6dc1 | 1044 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 1045 | |
8324aa91 | 1046 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 1047 | |
5acb3cc2 WL |
1048 | #ifdef CONFIG_BLK_DEV_IO_TRACE |
1049 | mutex_init(&q->blk_trace_mutex); | |
1050 | #endif | |
483f4afc | 1051 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 1052 | spin_lock_init(&q->__queue_lock); |
483f4afc | 1053 | |
498f6650 BVA |
1054 | if (!q->mq_ops) |
1055 | q->queue_lock = lock ? : &q->__queue_lock; | |
c94a96ac | 1056 | |
b82d4b19 TH |
1057 | /* |
1058 | * A queue starts its life with bypass turned on to avoid | |
1059 | * unnecessary bypass on/off overhead and nasty surprises during | |
749fefe6 TH |
1060 | * init. The initial bypass will be finished when the queue is |
1061 | * registered by blk_register_queue(). | |
b82d4b19 TH |
1062 | */ |
1063 | q->bypass_depth = 1; | |
f78bac2c | 1064 | queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q); |
b82d4b19 | 1065 | |
320ae51f JA |
1066 | init_waitqueue_head(&q->mq_freeze_wq); |
1067 | ||
3ef28e83 DW |
1068 | /* |
1069 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
1070 | * See blk_register_queue() for details. | |
1071 | */ | |
1072 | if (percpu_ref_init(&q->q_usage_counter, | |
1073 | blk_queue_usage_counter_release, | |
1074 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 1075 | goto fail_bdi; |
f51b802c | 1076 | |
3ef28e83 DW |
1077 | if (blkcg_init_queue(q)) |
1078 | goto fail_ref; | |
1079 | ||
1da177e4 | 1080 | return q; |
a73f730d | 1081 | |
3ef28e83 DW |
1082 | fail_ref: |
1083 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 1084 | fail_bdi: |
a83b576c JA |
1085 | blk_free_queue_stats(q->stats); |
1086 | fail_stats: | |
d03f6cdc | 1087 | bdi_put(q->backing_dev_info); |
54efd50b | 1088 | fail_split: |
338aa96d | 1089 | bioset_exit(&q->bio_split); |
a73f730d TH |
1090 | fail_id: |
1091 | ida_simple_remove(&blk_queue_ida, q->id); | |
1092 | fail_q: | |
1093 | kmem_cache_free(blk_requestq_cachep, q); | |
1094 | return NULL; | |
1da177e4 | 1095 | } |
1946089a | 1096 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
1097 | |
1098 | /** | |
1099 | * blk_init_queue - prepare a request queue for use with a block device | |
1100 | * @rfn: The function to be called to process requests that have been | |
1101 | * placed on the queue. | |
1102 | * @lock: Request queue spin lock | |
1103 | * | |
1104 | * Description: | |
1105 | * If a block device wishes to use the standard request handling procedures, | |
1106 | * which sorts requests and coalesces adjacent requests, then it must | |
1107 | * call blk_init_queue(). The function @rfn will be called when there | |
1108 | * are requests on the queue that need to be processed. If the device | |
1109 | * supports plugging, then @rfn may not be called immediately when requests | |
1110 | * are available on the queue, but may be called at some time later instead. | |
1111 | * Plugged queues are generally unplugged when a buffer belonging to one | |
1112 | * of the requests on the queue is needed, or due to memory pressure. | |
1113 | * | |
1114 | * @rfn is not required, or even expected, to remove all requests off the | |
1115 | * queue, but only as many as it can handle at a time. If it does leave | |
1116 | * requests on the queue, it is responsible for arranging that the requests | |
1117 | * get dealt with eventually. | |
1118 | * | |
1119 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
1120 | * request queue; this lock will be taken also from interrupt context, so irq |
1121 | * disabling is needed for it. | |
1da177e4 | 1122 | * |
710027a4 | 1123 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
1124 | * it didn't succeed. |
1125 | * | |
1126 | * Note: | |
1127 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
1128 | * when the block device is deactivated (such as at module unload). | |
1129 | **/ | |
1946089a | 1130 | |
165125e1 | 1131 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 1132 | { |
c304a51b | 1133 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); |
1946089a CL |
1134 | } |
1135 | EXPORT_SYMBOL(blk_init_queue); | |
1136 | ||
165125e1 | 1137 | struct request_queue * |
1946089a CL |
1138 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
1139 | { | |
5ea708d1 | 1140 | struct request_queue *q; |
1da177e4 | 1141 | |
498f6650 | 1142 | q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock); |
5ea708d1 | 1143 | if (!q) |
c86d1b8a MS |
1144 | return NULL; |
1145 | ||
5ea708d1 | 1146 | q->request_fn = rfn; |
5ea708d1 CH |
1147 | if (blk_init_allocated_queue(q) < 0) { |
1148 | blk_cleanup_queue(q); | |
1149 | return NULL; | |
1150 | } | |
18741986 | 1151 | |
7982e90c | 1152 | return q; |
01effb0d MS |
1153 | } |
1154 | EXPORT_SYMBOL(blk_init_queue_node); | |
1155 | ||
dece1635 | 1156 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio); |
336b7e1f | 1157 | |
1da177e4 | 1158 | |
5ea708d1 CH |
1159 | int blk_init_allocated_queue(struct request_queue *q) |
1160 | { | |
332ebbf7 BVA |
1161 | WARN_ON_ONCE(q->mq_ops); |
1162 | ||
6d247d7f | 1163 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size); |
ba483388 | 1164 | if (!q->fq) |
5ea708d1 | 1165 | return -ENOMEM; |
7982e90c | 1166 | |
6d247d7f CH |
1167 | if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL)) |
1168 | goto out_free_flush_queue; | |
7982e90c | 1169 | |
a051661c | 1170 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
6d247d7f | 1171 | goto out_exit_flush_rq; |
1da177e4 | 1172 | |
287922eb | 1173 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
60ea8226 | 1174 | q->queue_flags |= QUEUE_FLAG_DEFAULT; |
c94a96ac | 1175 | |
f3b144aa JA |
1176 | /* |
1177 | * This also sets hw/phys segments, boundary and size | |
1178 | */ | |
c20e8de2 | 1179 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 1180 | |
44ec9542 AS |
1181 | q->sg_reserved_size = INT_MAX; |
1182 | ||
acddf3b3 | 1183 | if (elevator_init(q)) |
6d247d7f | 1184 | goto out_exit_flush_rq; |
5ea708d1 | 1185 | return 0; |
eb1c160b | 1186 | |
6d247d7f CH |
1187 | out_exit_flush_rq: |
1188 | if (q->exit_rq_fn) | |
1189 | q->exit_rq_fn(q, q->fq->flush_rq); | |
1190 | out_free_flush_queue: | |
ba483388 | 1191 | blk_free_flush_queue(q->fq); |
54648cf1 | 1192 | q->fq = NULL; |
5ea708d1 | 1193 | return -ENOMEM; |
1da177e4 | 1194 | } |
5151412d | 1195 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 1196 | |
09ac46c4 | 1197 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 1198 | { |
3f3299d5 | 1199 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
1200 | __blk_get_queue(q); |
1201 | return true; | |
1da177e4 LT |
1202 | } |
1203 | ||
09ac46c4 | 1204 | return false; |
1da177e4 | 1205 | } |
d86e0e83 | 1206 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 1207 | |
5b788ce3 | 1208 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
1da177e4 | 1209 | { |
e8064021 | 1210 | if (rq->rq_flags & RQF_ELVPRIV) { |
5b788ce3 | 1211 | elv_put_request(rl->q, rq); |
f1f8cc94 | 1212 | if (rq->elv.icq) |
11a3122f | 1213 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
1214 | } |
1215 | ||
5b788ce3 | 1216 | mempool_free(rq, rl->rq_pool); |
1da177e4 LT |
1217 | } |
1218 | ||
1da177e4 LT |
1219 | /* |
1220 | * ioc_batching returns true if the ioc is a valid batching request and | |
1221 | * should be given priority access to a request. | |
1222 | */ | |
165125e1 | 1223 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1224 | { |
1225 | if (!ioc) | |
1226 | return 0; | |
1227 | ||
1228 | /* | |
1229 | * Make sure the process is able to allocate at least 1 request | |
1230 | * even if the batch times out, otherwise we could theoretically | |
1231 | * lose wakeups. | |
1232 | */ | |
1233 | return ioc->nr_batch_requests == q->nr_batching || | |
1234 | (ioc->nr_batch_requests > 0 | |
1235 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
1236 | } | |
1237 | ||
1238 | /* | |
1239 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
1240 | * will cause the process to be a "batcher" on all queues in the system. This | |
1241 | * is the behaviour we want though - once it gets a wakeup it should be given | |
1242 | * a nice run. | |
1243 | */ | |
165125e1 | 1244 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1245 | { |
1246 | if (!ioc || ioc_batching(q, ioc)) | |
1247 | return; | |
1248 | ||
1249 | ioc->nr_batch_requests = q->nr_batching; | |
1250 | ioc->last_waited = jiffies; | |
1251 | } | |
1252 | ||
5b788ce3 | 1253 | static void __freed_request(struct request_list *rl, int sync) |
1da177e4 | 1254 | { |
5b788ce3 | 1255 | struct request_queue *q = rl->q; |
1da177e4 | 1256 | |
d40f75a0 TH |
1257 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
1258 | blk_clear_congested(rl, sync); | |
1da177e4 | 1259 | |
1faa16d2 JA |
1260 | if (rl->count[sync] + 1 <= q->nr_requests) { |
1261 | if (waitqueue_active(&rl->wait[sync])) | |
1262 | wake_up(&rl->wait[sync]); | |
1da177e4 | 1263 | |
5b788ce3 | 1264 | blk_clear_rl_full(rl, sync); |
1da177e4 LT |
1265 | } |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * A request has just been released. Account for it, update the full and | |
1270 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
1271 | */ | |
e8064021 CH |
1272 | static void freed_request(struct request_list *rl, bool sync, |
1273 | req_flags_t rq_flags) | |
1da177e4 | 1274 | { |
5b788ce3 | 1275 | struct request_queue *q = rl->q; |
1da177e4 | 1276 | |
8a5ecdd4 | 1277 | q->nr_rqs[sync]--; |
1faa16d2 | 1278 | rl->count[sync]--; |
e8064021 | 1279 | if (rq_flags & RQF_ELVPRIV) |
8a5ecdd4 | 1280 | q->nr_rqs_elvpriv--; |
1da177e4 | 1281 | |
5b788ce3 | 1282 | __freed_request(rl, sync); |
1da177e4 | 1283 | |
1faa16d2 | 1284 | if (unlikely(rl->starved[sync ^ 1])) |
5b788ce3 | 1285 | __freed_request(rl, sync ^ 1); |
1da177e4 LT |
1286 | } |
1287 | ||
e3a2b3f9 JA |
1288 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) |
1289 | { | |
1290 | struct request_list *rl; | |
d40f75a0 | 1291 | int on_thresh, off_thresh; |
e3a2b3f9 | 1292 | |
332ebbf7 BVA |
1293 | WARN_ON_ONCE(q->mq_ops); |
1294 | ||
e3a2b3f9 JA |
1295 | spin_lock_irq(q->queue_lock); |
1296 | q->nr_requests = nr; | |
1297 | blk_queue_congestion_threshold(q); | |
d40f75a0 TH |
1298 | on_thresh = queue_congestion_on_threshold(q); |
1299 | off_thresh = queue_congestion_off_threshold(q); | |
e3a2b3f9 | 1300 | |
d40f75a0 TH |
1301 | blk_queue_for_each_rl(rl, q) { |
1302 | if (rl->count[BLK_RW_SYNC] >= on_thresh) | |
1303 | blk_set_congested(rl, BLK_RW_SYNC); | |
1304 | else if (rl->count[BLK_RW_SYNC] < off_thresh) | |
1305 | blk_clear_congested(rl, BLK_RW_SYNC); | |
e3a2b3f9 | 1306 | |
d40f75a0 TH |
1307 | if (rl->count[BLK_RW_ASYNC] >= on_thresh) |
1308 | blk_set_congested(rl, BLK_RW_ASYNC); | |
1309 | else if (rl->count[BLK_RW_ASYNC] < off_thresh) | |
1310 | blk_clear_congested(rl, BLK_RW_ASYNC); | |
e3a2b3f9 | 1311 | |
e3a2b3f9 JA |
1312 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { |
1313 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
1314 | } else { | |
1315 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
1316 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
1317 | } | |
1318 | ||
1319 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
1320 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
1321 | } else { | |
1322 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
1323 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
1324 | } | |
1325 | } | |
1326 | ||
1327 | spin_unlock_irq(q->queue_lock); | |
1328 | return 0; | |
1329 | } | |
1330 | ||
da8303c6 | 1331 | /** |
a06e05e6 | 1332 | * __get_request - get a free request |
5b788ce3 | 1333 | * @rl: request list to allocate from |
ef295ecf | 1334 | * @op: operation and flags |
da8303c6 | 1335 | * @bio: bio to allocate request for (can be %NULL) |
6a15674d | 1336 | * @flags: BLQ_MQ_REQ_* flags |
4accf5fc | 1337 | * @gfp_mask: allocator flags |
da8303c6 TH |
1338 | * |
1339 | * Get a free request from @q. This function may fail under memory | |
1340 | * pressure or if @q is dead. | |
1341 | * | |
da3dae54 | 1342 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1343 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1344 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1345 | */ |
ef295ecf | 1346 | static struct request *__get_request(struct request_list *rl, unsigned int op, |
4accf5fc | 1347 | struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp_mask) |
1da177e4 | 1348 | { |
5b788ce3 | 1349 | struct request_queue *q = rl->q; |
b679281a | 1350 | struct request *rq; |
7f4b35d1 TH |
1351 | struct elevator_type *et = q->elevator->type; |
1352 | struct io_context *ioc = rq_ioc(bio); | |
f1f8cc94 | 1353 | struct io_cq *icq = NULL; |
ef295ecf | 1354 | const bool is_sync = op_is_sync(op); |
75eb6c37 | 1355 | int may_queue; |
e8064021 | 1356 | req_flags_t rq_flags = RQF_ALLOCED; |
88ee5ef1 | 1357 | |
2fff8a92 BVA |
1358 | lockdep_assert_held(q->queue_lock); |
1359 | ||
3f3299d5 | 1360 | if (unlikely(blk_queue_dying(q))) |
a492f075 | 1361 | return ERR_PTR(-ENODEV); |
da8303c6 | 1362 | |
ef295ecf | 1363 | may_queue = elv_may_queue(q, op); |
88ee5ef1 JA |
1364 | if (may_queue == ELV_MQUEUE_NO) |
1365 | goto rq_starved; | |
1366 | ||
1faa16d2 JA |
1367 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
1368 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
88ee5ef1 JA |
1369 | /* |
1370 | * The queue will fill after this allocation, so set | |
1371 | * it as full, and mark this process as "batching". | |
1372 | * This process will be allowed to complete a batch of | |
1373 | * requests, others will be blocked. | |
1374 | */ | |
5b788ce3 | 1375 | if (!blk_rl_full(rl, is_sync)) { |
88ee5ef1 | 1376 | ioc_set_batching(q, ioc); |
5b788ce3 | 1377 | blk_set_rl_full(rl, is_sync); |
88ee5ef1 JA |
1378 | } else { |
1379 | if (may_queue != ELV_MQUEUE_MUST | |
1380 | && !ioc_batching(q, ioc)) { | |
1381 | /* | |
1382 | * The queue is full and the allocating | |
1383 | * process is not a "batcher", and not | |
1384 | * exempted by the IO scheduler | |
1385 | */ | |
a492f075 | 1386 | return ERR_PTR(-ENOMEM); |
88ee5ef1 JA |
1387 | } |
1388 | } | |
1da177e4 | 1389 | } |
d40f75a0 | 1390 | blk_set_congested(rl, is_sync); |
1da177e4 LT |
1391 | } |
1392 | ||
082cf69e JA |
1393 | /* |
1394 | * Only allow batching queuers to allocate up to 50% over the defined | |
1395 | * limit of requests, otherwise we could have thousands of requests | |
1396 | * allocated with any setting of ->nr_requests | |
1397 | */ | |
1faa16d2 | 1398 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
a492f075 | 1399 | return ERR_PTR(-ENOMEM); |
fd782a4a | 1400 | |
8a5ecdd4 | 1401 | q->nr_rqs[is_sync]++; |
1faa16d2 JA |
1402 | rl->count[is_sync]++; |
1403 | rl->starved[is_sync] = 0; | |
cb98fc8b | 1404 | |
f1f8cc94 TH |
1405 | /* |
1406 | * Decide whether the new request will be managed by elevator. If | |
e8064021 | 1407 | * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will |
f1f8cc94 TH |
1408 | * prevent the current elevator from being destroyed until the new |
1409 | * request is freed. This guarantees icq's won't be destroyed and | |
1410 | * makes creating new ones safe. | |
1411 | * | |
e6f7f93d CH |
1412 | * Flush requests do not use the elevator so skip initialization. |
1413 | * This allows a request to share the flush and elevator data. | |
1414 | * | |
f1f8cc94 TH |
1415 | * Also, lookup icq while holding queue_lock. If it doesn't exist, |
1416 | * it will be created after releasing queue_lock. | |
1417 | */ | |
e6f7f93d | 1418 | if (!op_is_flush(op) && !blk_queue_bypass(q)) { |
e8064021 | 1419 | rq_flags |= RQF_ELVPRIV; |
8a5ecdd4 | 1420 | q->nr_rqs_elvpriv++; |
f1f8cc94 TH |
1421 | if (et->icq_cache && ioc) |
1422 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 1423 | } |
cb98fc8b | 1424 | |
f253b86b | 1425 | if (blk_queue_io_stat(q)) |
e8064021 | 1426 | rq_flags |= RQF_IO_STAT; |
1da177e4 LT |
1427 | spin_unlock_irq(q->queue_lock); |
1428 | ||
29e2b09a | 1429 | /* allocate and init request */ |
5b788ce3 | 1430 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
29e2b09a | 1431 | if (!rq) |
b679281a | 1432 | goto fail_alloc; |
1da177e4 | 1433 | |
29e2b09a | 1434 | blk_rq_init(q, rq); |
a051661c | 1435 | blk_rq_set_rl(rq, rl); |
ef295ecf | 1436 | rq->cmd_flags = op; |
e8064021 | 1437 | rq->rq_flags = rq_flags; |
1b6d65a0 BVA |
1438 | if (flags & BLK_MQ_REQ_PREEMPT) |
1439 | rq->rq_flags |= RQF_PREEMPT; | |
29e2b09a | 1440 | |
aaf7c680 | 1441 | /* init elvpriv */ |
e8064021 | 1442 | if (rq_flags & RQF_ELVPRIV) { |
aaf7c680 | 1443 | if (unlikely(et->icq_cache && !icq)) { |
7f4b35d1 TH |
1444 | if (ioc) |
1445 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
aaf7c680 TH |
1446 | if (!icq) |
1447 | goto fail_elvpriv; | |
29e2b09a | 1448 | } |
aaf7c680 TH |
1449 | |
1450 | rq->elv.icq = icq; | |
1451 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1452 | goto fail_elvpriv; | |
1453 | ||
1454 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
1455 | if (icq) |
1456 | get_io_context(icq->ioc); | |
1457 | } | |
aaf7c680 | 1458 | out: |
88ee5ef1 JA |
1459 | /* |
1460 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1461 | * OK, if the queue is under the request limit then requests need | |
1462 | * not count toward the nr_batch_requests limit. There will always | |
1463 | * be some limit enforced by BLK_BATCH_TIME. | |
1464 | */ | |
1da177e4 LT |
1465 | if (ioc_batching(q, ioc)) |
1466 | ioc->nr_batch_requests--; | |
6728cb0e | 1467 | |
e6a40b09 | 1468 | trace_block_getrq(q, bio, op); |
1da177e4 | 1469 | return rq; |
b679281a | 1470 | |
aaf7c680 TH |
1471 | fail_elvpriv: |
1472 | /* | |
1473 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1474 | * and may fail indefinitely under memory pressure and thus | |
1475 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1476 | * disturb iosched and blkcg but weird is bettern than dead. | |
1477 | */ | |
7b2b10e0 | 1478 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", |
dc3b17cc | 1479 | __func__, dev_name(q->backing_dev_info->dev)); |
aaf7c680 | 1480 | |
e8064021 | 1481 | rq->rq_flags &= ~RQF_ELVPRIV; |
aaf7c680 TH |
1482 | rq->elv.icq = NULL; |
1483 | ||
1484 | spin_lock_irq(q->queue_lock); | |
8a5ecdd4 | 1485 | q->nr_rqs_elvpriv--; |
aaf7c680 TH |
1486 | spin_unlock_irq(q->queue_lock); |
1487 | goto out; | |
1488 | ||
b679281a TH |
1489 | fail_alloc: |
1490 | /* | |
1491 | * Allocation failed presumably due to memory. Undo anything we | |
1492 | * might have messed up. | |
1493 | * | |
1494 | * Allocating task should really be put onto the front of the wait | |
1495 | * queue, but this is pretty rare. | |
1496 | */ | |
1497 | spin_lock_irq(q->queue_lock); | |
e8064021 | 1498 | freed_request(rl, is_sync, rq_flags); |
b679281a TH |
1499 | |
1500 | /* | |
1501 | * in the very unlikely event that allocation failed and no | |
1502 | * requests for this direction was pending, mark us starved so that | |
1503 | * freeing of a request in the other direction will notice | |
1504 | * us. another possible fix would be to split the rq mempool into | |
1505 | * READ and WRITE | |
1506 | */ | |
1507 | rq_starved: | |
1508 | if (unlikely(rl->count[is_sync] == 0)) | |
1509 | rl->starved[is_sync] = 1; | |
a492f075 | 1510 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1511 | } |
1512 | ||
da8303c6 | 1513 | /** |
a06e05e6 | 1514 | * get_request - get a free request |
da8303c6 | 1515 | * @q: request_queue to allocate request from |
ef295ecf | 1516 | * @op: operation and flags |
da8303c6 | 1517 | * @bio: bio to allocate request for (can be %NULL) |
6a15674d | 1518 | * @flags: BLK_MQ_REQ_* flags. |
4accf5fc | 1519 | * @gfp: allocator flags |
da8303c6 | 1520 | * |
a9a14d36 | 1521 | * Get a free request from @q. If %BLK_MQ_REQ_NOWAIT is set in @flags, |
d0164adc | 1522 | * this function keeps retrying under memory pressure and fails iff @q is dead. |
d6344532 | 1523 | * |
da3dae54 | 1524 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1525 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1526 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1527 | */ |
ef295ecf | 1528 | static struct request *get_request(struct request_queue *q, unsigned int op, |
4accf5fc | 1529 | struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp) |
1da177e4 | 1530 | { |
ef295ecf | 1531 | const bool is_sync = op_is_sync(op); |
a06e05e6 | 1532 | DEFINE_WAIT(wait); |
a051661c | 1533 | struct request_list *rl; |
1da177e4 | 1534 | struct request *rq; |
a051661c | 1535 | |
2fff8a92 | 1536 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 1537 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 1538 | |
a051661c | 1539 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ |
a06e05e6 | 1540 | retry: |
4accf5fc | 1541 | rq = __get_request(rl, op, bio, flags, gfp); |
a492f075 | 1542 | if (!IS_ERR(rq)) |
a06e05e6 | 1543 | return rq; |
1da177e4 | 1544 | |
03a07c92 GR |
1545 | if (op & REQ_NOWAIT) { |
1546 | blk_put_rl(rl); | |
1547 | return ERR_PTR(-EAGAIN); | |
1548 | } | |
1549 | ||
6a15674d | 1550 | if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) { |
a051661c | 1551 | blk_put_rl(rl); |
a492f075 | 1552 | return rq; |
a051661c | 1553 | } |
1da177e4 | 1554 | |
a06e05e6 TH |
1555 | /* wait on @rl and retry */ |
1556 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1557 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 1558 | |
e6a40b09 | 1559 | trace_block_sleeprq(q, bio, op); |
1da177e4 | 1560 | |
a06e05e6 TH |
1561 | spin_unlock_irq(q->queue_lock); |
1562 | io_schedule(); | |
d6344532 | 1563 | |
a06e05e6 TH |
1564 | /* |
1565 | * After sleeping, we become a "batching" process and will be able | |
1566 | * to allocate at least one request, and up to a big batch of them | |
1567 | * for a small period time. See ioc_batching, ioc_set_batching | |
1568 | */ | |
a06e05e6 | 1569 | ioc_set_batching(q, current->io_context); |
05caf8db | 1570 | |
a06e05e6 TH |
1571 | spin_lock_irq(q->queue_lock); |
1572 | finish_wait(&rl->wait[is_sync], &wait); | |
1da177e4 | 1573 | |
a06e05e6 | 1574 | goto retry; |
1da177e4 LT |
1575 | } |
1576 | ||
6a15674d | 1577 | /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */ |
cd6ce148 | 1578 | static struct request *blk_old_get_request(struct request_queue *q, |
9a95e4ef | 1579 | unsigned int op, blk_mq_req_flags_t flags) |
1da177e4 LT |
1580 | { |
1581 | struct request *rq; | |
c3036021 | 1582 | gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : GFP_NOIO; |
055f6e18 | 1583 | int ret = 0; |
1da177e4 | 1584 | |
332ebbf7 BVA |
1585 | WARN_ON_ONCE(q->mq_ops); |
1586 | ||
7f4b35d1 TH |
1587 | /* create ioc upfront */ |
1588 | create_io_context(gfp_mask, q->node); | |
1589 | ||
3a0a5299 | 1590 | ret = blk_queue_enter(q, flags); |
055f6e18 ML |
1591 | if (ret) |
1592 | return ERR_PTR(ret); | |
d6344532 | 1593 | spin_lock_irq(q->queue_lock); |
4accf5fc | 1594 | rq = get_request(q, op, NULL, flags, gfp_mask); |
0c4de0f3 | 1595 | if (IS_ERR(rq)) { |
da8303c6 | 1596 | spin_unlock_irq(q->queue_lock); |
055f6e18 | 1597 | blk_queue_exit(q); |
0c4de0f3 CH |
1598 | return rq; |
1599 | } | |
1da177e4 | 1600 | |
0c4de0f3 CH |
1601 | /* q->queue_lock is unlocked at this point */ |
1602 | rq->__data_len = 0; | |
1603 | rq->__sector = (sector_t) -1; | |
1604 | rq->bio = rq->biotail = NULL; | |
1da177e4 LT |
1605 | return rq; |
1606 | } | |
320ae51f | 1607 | |
6a15674d | 1608 | /** |
ff005a06 | 1609 | * blk_get_request - allocate a request |
6a15674d BVA |
1610 | * @q: request queue to allocate a request for |
1611 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
1612 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
1613 | */ | |
ff005a06 CH |
1614 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
1615 | blk_mq_req_flags_t flags) | |
320ae51f | 1616 | { |
d280bab3 BVA |
1617 | struct request *req; |
1618 | ||
6a15674d | 1619 | WARN_ON_ONCE(op & REQ_NOWAIT); |
1b6d65a0 | 1620 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT)); |
6a15674d | 1621 | |
d280bab3 | 1622 | if (q->mq_ops) { |
6a15674d | 1623 | req = blk_mq_alloc_request(q, op, flags); |
d280bab3 BVA |
1624 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) |
1625 | q->mq_ops->initialize_rq_fn(req); | |
1626 | } else { | |
6a15674d | 1627 | req = blk_old_get_request(q, op, flags); |
d280bab3 BVA |
1628 | if (!IS_ERR(req) && q->initialize_rq_fn) |
1629 | q->initialize_rq_fn(req); | |
1630 | } | |
1631 | ||
1632 | return req; | |
320ae51f | 1633 | } |
1da177e4 LT |
1634 | EXPORT_SYMBOL(blk_get_request); |
1635 | ||
1636 | /** | |
1637 | * blk_requeue_request - put a request back on queue | |
1638 | * @q: request queue where request should be inserted | |
1639 | * @rq: request to be inserted | |
1640 | * | |
1641 | * Description: | |
1642 | * Drivers often keep queueing requests until the hardware cannot accept | |
1643 | * more, when that condition happens we need to put the request back | |
1644 | * on the queue. Must be called with queue lock held. | |
1645 | */ | |
165125e1 | 1646 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1647 | { |
2fff8a92 | 1648 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 1649 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 1650 | |
242f9dcb JA |
1651 | blk_delete_timer(rq); |
1652 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1653 | trace_block_rq_requeue(q, rq); |
a7905043 | 1654 | rq_qos_requeue(q, rq); |
2056a782 | 1655 | |
e8064021 | 1656 | if (rq->rq_flags & RQF_QUEUED) |
1da177e4 LT |
1657 | blk_queue_end_tag(q, rq); |
1658 | ||
ba396a6c JB |
1659 | BUG_ON(blk_queued_rq(rq)); |
1660 | ||
1da177e4 LT |
1661 | elv_requeue_request(q, rq); |
1662 | } | |
1da177e4 LT |
1663 | EXPORT_SYMBOL(blk_requeue_request); |
1664 | ||
73c10101 JA |
1665 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1666 | int where) | |
1667 | { | |
320ae51f | 1668 | blk_account_io_start(rq, true); |
7eaceacc | 1669 | __elv_add_request(q, rq, where); |
73c10101 JA |
1670 | } |
1671 | ||
d62e26b3 | 1672 | static void part_round_stats_single(struct request_queue *q, int cpu, |
b8d62b3a JA |
1673 | struct hd_struct *part, unsigned long now, |
1674 | unsigned int inflight) | |
074a7aca | 1675 | { |
b8d62b3a | 1676 | if (inflight) { |
074a7aca | 1677 | __part_stat_add(cpu, part, time_in_queue, |
b8d62b3a | 1678 | inflight * (now - part->stamp)); |
074a7aca TH |
1679 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1680 | } | |
1681 | part->stamp = now; | |
1682 | } | |
1683 | ||
1684 | /** | |
496aa8a9 | 1685 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
d62e26b3 | 1686 | * @q: target block queue |
496aa8a9 RD |
1687 | * @cpu: cpu number for stats access |
1688 | * @part: target partition | |
1da177e4 LT |
1689 | * |
1690 | * The average IO queue length and utilisation statistics are maintained | |
1691 | * by observing the current state of the queue length and the amount of | |
1692 | * time it has been in this state for. | |
1693 | * | |
1694 | * Normally, that accounting is done on IO completion, but that can result | |
1695 | * in more than a second's worth of IO being accounted for within any one | |
1696 | * second, leading to >100% utilisation. To deal with that, we call this | |
1697 | * function to do a round-off before returning the results when reading | |
1698 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1699 | * the current jiffies and restarts the counters again. | |
1700 | */ | |
d62e26b3 | 1701 | void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part) |
6f2576af | 1702 | { |
b8d62b3a | 1703 | struct hd_struct *part2 = NULL; |
6f2576af | 1704 | unsigned long now = jiffies; |
b8d62b3a JA |
1705 | unsigned int inflight[2]; |
1706 | int stats = 0; | |
1707 | ||
1708 | if (part->stamp != now) | |
1709 | stats |= 1; | |
1710 | ||
1711 | if (part->partno) { | |
1712 | part2 = &part_to_disk(part)->part0; | |
1713 | if (part2->stamp != now) | |
1714 | stats |= 2; | |
1715 | } | |
1716 | ||
1717 | if (!stats) | |
1718 | return; | |
1719 | ||
1720 | part_in_flight(q, part, inflight); | |
6f2576af | 1721 | |
b8d62b3a JA |
1722 | if (stats & 2) |
1723 | part_round_stats_single(q, cpu, part2, now, inflight[1]); | |
1724 | if (stats & 1) | |
1725 | part_round_stats_single(q, cpu, part, now, inflight[0]); | |
6f2576af | 1726 | } |
074a7aca | 1727 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1728 | |
47fafbc7 | 1729 | #ifdef CONFIG_PM |
c8158819 LM |
1730 | static void blk_pm_put_request(struct request *rq) |
1731 | { | |
e8064021 | 1732 | if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending) |
c8158819 LM |
1733 | pm_runtime_mark_last_busy(rq->q->dev); |
1734 | } | |
1735 | #else | |
1736 | static inline void blk_pm_put_request(struct request *rq) {} | |
1737 | #endif | |
1738 | ||
165125e1 | 1739 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1740 | { |
e8064021 CH |
1741 | req_flags_t rq_flags = req->rq_flags; |
1742 | ||
1da177e4 LT |
1743 | if (unlikely(!q)) |
1744 | return; | |
1da177e4 | 1745 | |
6f5ba581 CH |
1746 | if (q->mq_ops) { |
1747 | blk_mq_free_request(req); | |
1748 | return; | |
1749 | } | |
1750 | ||
2fff8a92 BVA |
1751 | lockdep_assert_held(q->queue_lock); |
1752 | ||
6cc77e9c | 1753 | blk_req_zone_write_unlock(req); |
c8158819 LM |
1754 | blk_pm_put_request(req); |
1755 | ||
8922e16c TH |
1756 | elv_completed_request(q, req); |
1757 | ||
1cd96c24 BH |
1758 | /* this is a bio leak */ |
1759 | WARN_ON(req->bio != NULL); | |
1760 | ||
a7905043 | 1761 | rq_qos_done(q, req); |
87760e5e | 1762 | |
1da177e4 LT |
1763 | /* |
1764 | * Request may not have originated from ll_rw_blk. if not, | |
1765 | * it didn't come out of our reserved rq pools | |
1766 | */ | |
e8064021 | 1767 | if (rq_flags & RQF_ALLOCED) { |
a051661c | 1768 | struct request_list *rl = blk_rq_rl(req); |
ef295ecf | 1769 | bool sync = op_is_sync(req->cmd_flags); |
1da177e4 | 1770 | |
1da177e4 | 1771 | BUG_ON(!list_empty(&req->queuelist)); |
360f92c2 | 1772 | BUG_ON(ELV_ON_HASH(req)); |
1da177e4 | 1773 | |
a051661c | 1774 | blk_free_request(rl, req); |
e8064021 | 1775 | freed_request(rl, sync, rq_flags); |
a051661c | 1776 | blk_put_rl(rl); |
055f6e18 | 1777 | blk_queue_exit(q); |
1da177e4 LT |
1778 | } |
1779 | } | |
6e39b69e MC |
1780 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1781 | ||
1da177e4 LT |
1782 | void blk_put_request(struct request *req) |
1783 | { | |
165125e1 | 1784 | struct request_queue *q = req->q; |
8922e16c | 1785 | |
320ae51f JA |
1786 | if (q->mq_ops) |
1787 | blk_mq_free_request(req); | |
1788 | else { | |
1789 | unsigned long flags; | |
1790 | ||
1791 | spin_lock_irqsave(q->queue_lock, flags); | |
1792 | __blk_put_request(q, req); | |
1793 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1794 | } | |
1da177e4 | 1795 | } |
1da177e4 LT |
1796 | EXPORT_SYMBOL(blk_put_request); |
1797 | ||
320ae51f JA |
1798 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1799 | struct bio *bio) | |
73c10101 | 1800 | { |
1eff9d32 | 1801 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1802 | |
73c10101 JA |
1803 | if (!ll_back_merge_fn(q, req, bio)) |
1804 | return false; | |
1805 | ||
8c1cf6bb | 1806 | trace_block_bio_backmerge(q, req, bio); |
73c10101 JA |
1807 | |
1808 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1809 | blk_rq_set_mixed_merge(req); | |
1810 | ||
1811 | req->biotail->bi_next = bio; | |
1812 | req->biotail = bio; | |
4f024f37 | 1813 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 JA |
1814 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1815 | ||
320ae51f | 1816 | blk_account_io_start(req, false); |
73c10101 JA |
1817 | return true; |
1818 | } | |
1819 | ||
320ae51f JA |
1820 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, |
1821 | struct bio *bio) | |
73c10101 | 1822 | { |
1eff9d32 | 1823 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1824 | |
73c10101 JA |
1825 | if (!ll_front_merge_fn(q, req, bio)) |
1826 | return false; | |
1827 | ||
8c1cf6bb | 1828 | trace_block_bio_frontmerge(q, req, bio); |
73c10101 JA |
1829 | |
1830 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1831 | blk_rq_set_mixed_merge(req); | |
1832 | ||
73c10101 JA |
1833 | bio->bi_next = req->bio; |
1834 | req->bio = bio; | |
1835 | ||
4f024f37 KO |
1836 | req->__sector = bio->bi_iter.bi_sector; |
1837 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 JA |
1838 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1839 | ||
320ae51f | 1840 | blk_account_io_start(req, false); |
73c10101 JA |
1841 | return true; |
1842 | } | |
1843 | ||
1e739730 CH |
1844 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, |
1845 | struct bio *bio) | |
1846 | { | |
1847 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1848 | ||
1849 | if (segments >= queue_max_discard_segments(q)) | |
1850 | goto no_merge; | |
1851 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1852 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1853 | goto no_merge; | |
1854 | ||
1855 | req->biotail->bi_next = bio; | |
1856 | req->biotail = bio; | |
1857 | req->__data_len += bio->bi_iter.bi_size; | |
1858 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1859 | req->nr_phys_segments = segments + 1; | |
1860 | ||
1861 | blk_account_io_start(req, false); | |
1862 | return true; | |
1863 | no_merge: | |
1864 | req_set_nomerge(q, req); | |
1865 | return false; | |
1866 | } | |
1867 | ||
bd87b589 | 1868 | /** |
320ae51f | 1869 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
1870 | * @q: request_queue new bio is being queued at |
1871 | * @bio: new bio being queued | |
1872 | * @request_count: out parameter for number of traversed plugged requests | |
ccc2600b RD |
1873 | * @same_queue_rq: pointer to &struct request that gets filled in when |
1874 | * another request associated with @q is found on the plug list | |
1875 | * (optional, may be %NULL) | |
bd87b589 TH |
1876 | * |
1877 | * Determine whether @bio being queued on @q can be merged with a request | |
1878 | * on %current's plugged list. Returns %true if merge was successful, | |
1879 | * otherwise %false. | |
1880 | * | |
07c2bd37 TH |
1881 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1882 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1883 | * than scheduling, and the request, while may have elvpriv data, is not | |
1884 | * added on the elevator at this point. In addition, we don't have | |
1885 | * reliable access to the elevator outside queue lock. Only check basic | |
1886 | * merging parameters without querying the elevator. | |
da41a589 RE |
1887 | * |
1888 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 1889 | */ |
320ae51f | 1890 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
5b3f341f SL |
1891 | unsigned int *request_count, |
1892 | struct request **same_queue_rq) | |
73c10101 JA |
1893 | { |
1894 | struct blk_plug *plug; | |
1895 | struct request *rq; | |
92f399c7 | 1896 | struct list_head *plug_list; |
73c10101 | 1897 | |
bd87b589 | 1898 | plug = current->plug; |
73c10101 | 1899 | if (!plug) |
34fe7c05 | 1900 | return false; |
56ebdaf2 | 1901 | *request_count = 0; |
73c10101 | 1902 | |
92f399c7 SL |
1903 | if (q->mq_ops) |
1904 | plug_list = &plug->mq_list; | |
1905 | else | |
1906 | plug_list = &plug->list; | |
1907 | ||
1908 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
34fe7c05 | 1909 | bool merged = false; |
73c10101 | 1910 | |
5b3f341f | 1911 | if (rq->q == q) { |
1b2e19f1 | 1912 | (*request_count)++; |
5b3f341f SL |
1913 | /* |
1914 | * Only blk-mq multiple hardware queues case checks the | |
1915 | * rq in the same queue, there should be only one such | |
1916 | * rq in a queue | |
1917 | **/ | |
1918 | if (same_queue_rq) | |
1919 | *same_queue_rq = rq; | |
1920 | } | |
56ebdaf2 | 1921 | |
07c2bd37 | 1922 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1923 | continue; |
1924 | ||
34fe7c05 CH |
1925 | switch (blk_try_merge(rq, bio)) { |
1926 | case ELEVATOR_BACK_MERGE: | |
1927 | merged = bio_attempt_back_merge(q, rq, bio); | |
1928 | break; | |
1929 | case ELEVATOR_FRONT_MERGE: | |
1930 | merged = bio_attempt_front_merge(q, rq, bio); | |
1931 | break; | |
1e739730 CH |
1932 | case ELEVATOR_DISCARD_MERGE: |
1933 | merged = bio_attempt_discard_merge(q, rq, bio); | |
1934 | break; | |
34fe7c05 CH |
1935 | default: |
1936 | break; | |
73c10101 | 1937 | } |
34fe7c05 CH |
1938 | |
1939 | if (merged) | |
1940 | return true; | |
73c10101 | 1941 | } |
34fe7c05 CH |
1942 | |
1943 | return false; | |
73c10101 JA |
1944 | } |
1945 | ||
0809e3ac JM |
1946 | unsigned int blk_plug_queued_count(struct request_queue *q) |
1947 | { | |
1948 | struct blk_plug *plug; | |
1949 | struct request *rq; | |
1950 | struct list_head *plug_list; | |
1951 | unsigned int ret = 0; | |
1952 | ||
1953 | plug = current->plug; | |
1954 | if (!plug) | |
1955 | goto out; | |
1956 | ||
1957 | if (q->mq_ops) | |
1958 | plug_list = &plug->mq_list; | |
1959 | else | |
1960 | plug_list = &plug->list; | |
1961 | ||
1962 | list_for_each_entry(rq, plug_list, queuelist) { | |
1963 | if (rq->q == q) | |
1964 | ret++; | |
1965 | } | |
1966 | out: | |
1967 | return ret; | |
1968 | } | |
1969 | ||
da8d7f07 | 1970 | void blk_init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1971 | { |
0be0dee6 BVA |
1972 | struct io_context *ioc = rq_ioc(bio); |
1973 | ||
1eff9d32 | 1974 | if (bio->bi_opf & REQ_RAHEAD) |
a82afdfc | 1975 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1976 | |
4f024f37 | 1977 | req->__sector = bio->bi_iter.bi_sector; |
5dc8b362 AM |
1978 | if (ioprio_valid(bio_prio(bio))) |
1979 | req->ioprio = bio_prio(bio); | |
0be0dee6 BVA |
1980 | else if (ioc) |
1981 | req->ioprio = ioc->ioprio; | |
1982 | else | |
1983 | req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); | |
cb6934f8 | 1984 | req->write_hint = bio->bi_write_hint; |
bc1c56fd | 1985 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 | 1986 | } |
da8d7f07 | 1987 | EXPORT_SYMBOL_GPL(blk_init_request_from_bio); |
52d9e675 | 1988 | |
dece1635 | 1989 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1990 | { |
73c10101 | 1991 | struct blk_plug *plug; |
34fe7c05 | 1992 | int where = ELEVATOR_INSERT_SORT; |
e4d750c9 | 1993 | struct request *req, *free; |
56ebdaf2 | 1994 | unsigned int request_count = 0; |
1da177e4 | 1995 | |
1da177e4 LT |
1996 | /* |
1997 | * low level driver can indicate that it wants pages above a | |
1998 | * certain limit bounced to low memory (ie for highmem, or even | |
1999 | * ISA dma in theory) | |
2000 | */ | |
2001 | blk_queue_bounce(q, &bio); | |
2002 | ||
af67c31f | 2003 | blk_queue_split(q, &bio); |
23688bf4 | 2004 | |
e23947bd | 2005 | if (!bio_integrity_prep(bio)) |
dece1635 | 2006 | return BLK_QC_T_NONE; |
ffecfd1a | 2007 | |
f73f44eb | 2008 | if (op_is_flush(bio->bi_opf)) { |
73c10101 | 2009 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 2010 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
2011 | goto get_rq; |
2012 | } | |
2013 | ||
73c10101 JA |
2014 | /* |
2015 | * Check if we can merge with the plugged list before grabbing | |
2016 | * any locks. | |
2017 | */ | |
0809e3ac JM |
2018 | if (!blk_queue_nomerges(q)) { |
2019 | if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) | |
dece1635 | 2020 | return BLK_QC_T_NONE; |
0809e3ac JM |
2021 | } else |
2022 | request_count = blk_plug_queued_count(q); | |
1da177e4 | 2023 | |
73c10101 | 2024 | spin_lock_irq(q->queue_lock); |
2056a782 | 2025 | |
34fe7c05 CH |
2026 | switch (elv_merge(q, &req, bio)) { |
2027 | case ELEVATOR_BACK_MERGE: | |
2028 | if (!bio_attempt_back_merge(q, req, bio)) | |
2029 | break; | |
2030 | elv_bio_merged(q, req, bio); | |
2031 | free = attempt_back_merge(q, req); | |
2032 | if (free) | |
2033 | __blk_put_request(q, free); | |
2034 | else | |
2035 | elv_merged_request(q, req, ELEVATOR_BACK_MERGE); | |
2036 | goto out_unlock; | |
2037 | case ELEVATOR_FRONT_MERGE: | |
2038 | if (!bio_attempt_front_merge(q, req, bio)) | |
2039 | break; | |
2040 | elv_bio_merged(q, req, bio); | |
2041 | free = attempt_front_merge(q, req); | |
2042 | if (free) | |
2043 | __blk_put_request(q, free); | |
2044 | else | |
2045 | elv_merged_request(q, req, ELEVATOR_FRONT_MERGE); | |
2046 | goto out_unlock; | |
2047 | default: | |
2048 | break; | |
1da177e4 LT |
2049 | } |
2050 | ||
450991bc | 2051 | get_rq: |
c1c80384 | 2052 | rq_qos_throttle(q, bio, q->queue_lock); |
87760e5e | 2053 | |
1da177e4 | 2054 | /* |
450991bc | 2055 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 2056 | * Returns with the queue unlocked. |
450991bc | 2057 | */ |
055f6e18 | 2058 | blk_queue_enter_live(q); |
c3036021 | 2059 | req = get_request(q, bio->bi_opf, bio, 0, GFP_NOIO); |
a492f075 | 2060 | if (IS_ERR(req)) { |
055f6e18 | 2061 | blk_queue_exit(q); |
c1c80384 | 2062 | rq_qos_cleanup(q, bio); |
4e4cbee9 CH |
2063 | if (PTR_ERR(req) == -ENOMEM) |
2064 | bio->bi_status = BLK_STS_RESOURCE; | |
2065 | else | |
2066 | bio->bi_status = BLK_STS_IOERR; | |
4246a0b6 | 2067 | bio_endio(bio); |
da8303c6 TH |
2068 | goto out_unlock; |
2069 | } | |
d6344532 | 2070 | |
c1c80384 | 2071 | rq_qos_track(q, req, bio); |
87760e5e | 2072 | |
450991bc NP |
2073 | /* |
2074 | * After dropping the lock and possibly sleeping here, our request | |
2075 | * may now be mergeable after it had proven unmergeable (above). | |
2076 | * We don't worry about that case for efficiency. It won't happen | |
2077 | * often, and the elevators are able to handle it. | |
1da177e4 | 2078 | */ |
da8d7f07 | 2079 | blk_init_request_from_bio(req, bio); |
1da177e4 | 2080 | |
9562ad9a | 2081 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 2082 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
2083 | |
2084 | plug = current->plug; | |
721a9602 | 2085 | if (plug) { |
dc6d36c9 JA |
2086 | /* |
2087 | * If this is the first request added after a plug, fire | |
7aef2e78 | 2088 | * of a plug trace. |
0a6219a9 ML |
2089 | * |
2090 | * @request_count may become stale because of schedule | |
2091 | * out, so check plug list again. | |
dc6d36c9 | 2092 | */ |
0a6219a9 | 2093 | if (!request_count || list_empty(&plug->list)) |
dc6d36c9 | 2094 | trace_block_plug(q); |
3540d5e8 | 2095 | else { |
50d24c34 SL |
2096 | struct request *last = list_entry_rq(plug->list.prev); |
2097 | if (request_count >= BLK_MAX_REQUEST_COUNT || | |
2098 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) { | |
3540d5e8 | 2099 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
2100 | trace_block_plug(q); |
2101 | } | |
73c10101 | 2102 | } |
73c10101 | 2103 | list_add_tail(&req->queuelist, &plug->list); |
320ae51f | 2104 | blk_account_io_start(req, true); |
73c10101 JA |
2105 | } else { |
2106 | spin_lock_irq(q->queue_lock); | |
2107 | add_acct_request(q, req, where); | |
24ecfbe2 | 2108 | __blk_run_queue(q); |
73c10101 JA |
2109 | out_unlock: |
2110 | spin_unlock_irq(q->queue_lock); | |
2111 | } | |
dece1635 JA |
2112 | |
2113 | return BLK_QC_T_NONE; | |
1da177e4 LT |
2114 | } |
2115 | ||
52c5e62d | 2116 | static void handle_bad_sector(struct bio *bio, sector_t maxsector) |
1da177e4 LT |
2117 | { |
2118 | char b[BDEVNAME_SIZE]; | |
2119 | ||
2120 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
6296b960 | 2121 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
74d46992 | 2122 | bio_devname(bio, b), bio->bi_opf, |
f73a1c7d | 2123 | (unsigned long long)bio_end_sector(bio), |
52c5e62d | 2124 | (long long)maxsector); |
1da177e4 LT |
2125 | } |
2126 | ||
c17bb495 AM |
2127 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
2128 | ||
2129 | static DECLARE_FAULT_ATTR(fail_make_request); | |
2130 | ||
2131 | static int __init setup_fail_make_request(char *str) | |
2132 | { | |
2133 | return setup_fault_attr(&fail_make_request, str); | |
2134 | } | |
2135 | __setup("fail_make_request=", setup_fail_make_request); | |
2136 | ||
b2c9cd37 | 2137 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 2138 | { |
b2c9cd37 | 2139 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
2140 | } |
2141 | ||
2142 | static int __init fail_make_request_debugfs(void) | |
2143 | { | |
dd48c085 AM |
2144 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
2145 | NULL, &fail_make_request); | |
2146 | ||
21f9fcd8 | 2147 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
2148 | } |
2149 | ||
2150 | late_initcall(fail_make_request_debugfs); | |
2151 | ||
2152 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
2153 | ||
b2c9cd37 AM |
2154 | static inline bool should_fail_request(struct hd_struct *part, |
2155 | unsigned int bytes) | |
c17bb495 | 2156 | { |
b2c9cd37 | 2157 | return false; |
c17bb495 AM |
2158 | } |
2159 | ||
2160 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
2161 | ||
721c7fc7 ID |
2162 | static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) |
2163 | { | |
b089cfd9 JA |
2164 | const int op = bio_op(bio); |
2165 | ||
8b2ded1c | 2166 | if (part->policy && op_is_write(op)) { |
721c7fc7 ID |
2167 | char b[BDEVNAME_SIZE]; |
2168 | ||
8b2ded1c MP |
2169 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
2170 | return false; | |
2171 | ||
a32e236e | 2172 | WARN_ONCE(1, |
721c7fc7 ID |
2173 | "generic_make_request: Trying to write " |
2174 | "to read-only block-device %s (partno %d)\n", | |
2175 | bio_devname(bio, b), part->partno); | |
a32e236e LT |
2176 | /* Older lvm-tools actually trigger this */ |
2177 | return false; | |
721c7fc7 ID |
2178 | } |
2179 | ||
2180 | return false; | |
2181 | } | |
2182 | ||
30abb3a6 HM |
2183 | static noinline int should_fail_bio(struct bio *bio) |
2184 | { | |
2185 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) | |
2186 | return -EIO; | |
2187 | return 0; | |
2188 | } | |
2189 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
2190 | ||
52c5e62d CH |
2191 | /* |
2192 | * Check whether this bio extends beyond the end of the device or partition. | |
2193 | * This may well happen - the kernel calls bread() without checking the size of | |
2194 | * the device, e.g., when mounting a file system. | |
2195 | */ | |
2196 | static inline int bio_check_eod(struct bio *bio, sector_t maxsector) | |
2197 | { | |
2198 | unsigned int nr_sectors = bio_sectors(bio); | |
2199 | ||
2200 | if (nr_sectors && maxsector && | |
2201 | (nr_sectors > maxsector || | |
2202 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { | |
2203 | handle_bad_sector(bio, maxsector); | |
2204 | return -EIO; | |
2205 | } | |
2206 | return 0; | |
2207 | } | |
2208 | ||
74d46992 CH |
2209 | /* |
2210 | * Remap block n of partition p to block n+start(p) of the disk. | |
2211 | */ | |
2212 | static inline int blk_partition_remap(struct bio *bio) | |
2213 | { | |
2214 | struct hd_struct *p; | |
52c5e62d | 2215 | int ret = -EIO; |
74d46992 | 2216 | |
721c7fc7 ID |
2217 | rcu_read_lock(); |
2218 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
52c5e62d CH |
2219 | if (unlikely(!p)) |
2220 | goto out; | |
2221 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) | |
2222 | goto out; | |
2223 | if (unlikely(bio_check_ro(bio, p))) | |
721c7fc7 | 2224 | goto out; |
721c7fc7 | 2225 | |
74d46992 CH |
2226 | /* |
2227 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
2228 | * Include a test for the reset op code and perform the remap if needed. | |
2229 | */ | |
52c5e62d CH |
2230 | if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) { |
2231 | if (bio_check_eod(bio, part_nr_sects_read(p))) | |
2232 | goto out; | |
2233 | bio->bi_iter.bi_sector += p->start_sect; | |
52c5e62d CH |
2234 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), |
2235 | bio->bi_iter.bi_sector - p->start_sect); | |
2236 | } | |
c04fa44b | 2237 | bio->bi_partno = 0; |
52c5e62d | 2238 | ret = 0; |
721c7fc7 ID |
2239 | out: |
2240 | rcu_read_unlock(); | |
74d46992 CH |
2241 | return ret; |
2242 | } | |
2243 | ||
27a84d54 CH |
2244 | static noinline_for_stack bool |
2245 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 2246 | { |
165125e1 | 2247 | struct request_queue *q; |
5a7bbad2 | 2248 | int nr_sectors = bio_sectors(bio); |
4e4cbee9 | 2249 | blk_status_t status = BLK_STS_IOERR; |
5a7bbad2 | 2250 | char b[BDEVNAME_SIZE]; |
1da177e4 LT |
2251 | |
2252 | might_sleep(); | |
1da177e4 | 2253 | |
74d46992 | 2254 | q = bio->bi_disk->queue; |
5a7bbad2 CH |
2255 | if (unlikely(!q)) { |
2256 | printk(KERN_ERR | |
2257 | "generic_make_request: Trying to access " | |
2258 | "nonexistent block-device %s (%Lu)\n", | |
74d46992 | 2259 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); |
5a7bbad2 CH |
2260 | goto end_io; |
2261 | } | |
c17bb495 | 2262 | |
03a07c92 GR |
2263 | /* |
2264 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
2265 | * if queue is not a request based queue. | |
2266 | */ | |
03a07c92 GR |
2267 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q)) |
2268 | goto not_supported; | |
2269 | ||
30abb3a6 | 2270 | if (should_fail_bio(bio)) |
5a7bbad2 | 2271 | goto end_io; |
2056a782 | 2272 | |
52c5e62d CH |
2273 | if (bio->bi_partno) { |
2274 | if (unlikely(blk_partition_remap(bio))) | |
721c7fc7 ID |
2275 | goto end_io; |
2276 | } else { | |
52c5e62d CH |
2277 | if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0))) |
2278 | goto end_io; | |
2279 | if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk)))) | |
721c7fc7 ID |
2280 | goto end_io; |
2281 | } | |
2056a782 | 2282 | |
5a7bbad2 CH |
2283 | /* |
2284 | * Filter flush bio's early so that make_request based | |
2285 | * drivers without flush support don't have to worry | |
2286 | * about them. | |
2287 | */ | |
f3a8ab7d | 2288 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 2289 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 2290 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
5a7bbad2 | 2291 | if (!nr_sectors) { |
4e4cbee9 | 2292 | status = BLK_STS_OK; |
51fd77bd JA |
2293 | goto end_io; |
2294 | } | |
5a7bbad2 | 2295 | } |
5ddfe969 | 2296 | |
288dab8a CH |
2297 | switch (bio_op(bio)) { |
2298 | case REQ_OP_DISCARD: | |
2299 | if (!blk_queue_discard(q)) | |
2300 | goto not_supported; | |
2301 | break; | |
2302 | case REQ_OP_SECURE_ERASE: | |
2303 | if (!blk_queue_secure_erase(q)) | |
2304 | goto not_supported; | |
2305 | break; | |
2306 | case REQ_OP_WRITE_SAME: | |
74d46992 | 2307 | if (!q->limits.max_write_same_sectors) |
288dab8a | 2308 | goto not_supported; |
58886785 | 2309 | break; |
2d253440 ST |
2310 | case REQ_OP_ZONE_REPORT: |
2311 | case REQ_OP_ZONE_RESET: | |
74d46992 | 2312 | if (!blk_queue_is_zoned(q)) |
2d253440 | 2313 | goto not_supported; |
288dab8a | 2314 | break; |
a6f0788e | 2315 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 2316 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
2317 | goto not_supported; |
2318 | break; | |
288dab8a CH |
2319 | default: |
2320 | break; | |
5a7bbad2 | 2321 | } |
01edede4 | 2322 | |
7f4b35d1 TH |
2323 | /* |
2324 | * Various block parts want %current->io_context and lazy ioc | |
2325 | * allocation ends up trading a lot of pain for a small amount of | |
2326 | * memory. Just allocate it upfront. This may fail and block | |
2327 | * layer knows how to live with it. | |
2328 | */ | |
2329 | create_io_context(GFP_ATOMIC, q->node); | |
2330 | ||
ae118896 TH |
2331 | if (!blkcg_bio_issue_check(q, bio)) |
2332 | return false; | |
27a84d54 | 2333 | |
fbbaf700 N |
2334 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
2335 | trace_block_bio_queue(q, bio); | |
2336 | /* Now that enqueuing has been traced, we need to trace | |
2337 | * completion as well. | |
2338 | */ | |
2339 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
2340 | } | |
27a84d54 | 2341 | return true; |
a7384677 | 2342 | |
288dab8a | 2343 | not_supported: |
4e4cbee9 | 2344 | status = BLK_STS_NOTSUPP; |
a7384677 | 2345 | end_io: |
4e4cbee9 | 2346 | bio->bi_status = status; |
4246a0b6 | 2347 | bio_endio(bio); |
27a84d54 | 2348 | return false; |
1da177e4 LT |
2349 | } |
2350 | ||
27a84d54 CH |
2351 | /** |
2352 | * generic_make_request - hand a buffer to its device driver for I/O | |
2353 | * @bio: The bio describing the location in memory and on the device. | |
2354 | * | |
2355 | * generic_make_request() is used to make I/O requests of block | |
2356 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
2357 | * to be done. | |
2358 | * | |
2359 | * generic_make_request() does not return any status. The | |
2360 | * success/failure status of the request, along with notification of | |
2361 | * completion, is delivered asynchronously through the bio->bi_end_io | |
2362 | * function described (one day) else where. | |
2363 | * | |
2364 | * The caller of generic_make_request must make sure that bi_io_vec | |
2365 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
2366 | * set to describe the device address, and the | |
2367 | * bi_end_io and optionally bi_private are set to describe how | |
2368 | * completion notification should be signaled. | |
2369 | * | |
2370 | * generic_make_request and the drivers it calls may use bi_next if this | |
2371 | * bio happens to be merged with someone else, and may resubmit the bio to | |
2372 | * a lower device by calling into generic_make_request recursively, which | |
2373 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 | 2374 | */ |
dece1635 | 2375 | blk_qc_t generic_make_request(struct bio *bio) |
d89d8796 | 2376 | { |
f5fe1b51 N |
2377 | /* |
2378 | * bio_list_on_stack[0] contains bios submitted by the current | |
2379 | * make_request_fn. | |
2380 | * bio_list_on_stack[1] contains bios that were submitted before | |
2381 | * the current make_request_fn, but that haven't been processed | |
2382 | * yet. | |
2383 | */ | |
2384 | struct bio_list bio_list_on_stack[2]; | |
37f9579f BVA |
2385 | blk_mq_req_flags_t flags = 0; |
2386 | struct request_queue *q = bio->bi_disk->queue; | |
dece1635 | 2387 | blk_qc_t ret = BLK_QC_T_NONE; |
bddd87c7 | 2388 | |
37f9579f BVA |
2389 | if (bio->bi_opf & REQ_NOWAIT) |
2390 | flags = BLK_MQ_REQ_NOWAIT; | |
cd4a4ae4 JA |
2391 | if (bio_flagged(bio, BIO_QUEUE_ENTERED)) |
2392 | blk_queue_enter_live(q); | |
2393 | else if (blk_queue_enter(q, flags) < 0) { | |
37f9579f BVA |
2394 | if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT)) |
2395 | bio_wouldblock_error(bio); | |
2396 | else | |
2397 | bio_io_error(bio); | |
2398 | return ret; | |
2399 | } | |
2400 | ||
27a84d54 | 2401 | if (!generic_make_request_checks(bio)) |
dece1635 | 2402 | goto out; |
27a84d54 CH |
2403 | |
2404 | /* | |
2405 | * We only want one ->make_request_fn to be active at a time, else | |
2406 | * stack usage with stacked devices could be a problem. So use | |
2407 | * current->bio_list to keep a list of requests submited by a | |
2408 | * make_request_fn function. current->bio_list is also used as a | |
2409 | * flag to say if generic_make_request is currently active in this | |
2410 | * task or not. If it is NULL, then no make_request is active. If | |
2411 | * it is non-NULL, then a make_request is active, and new requests | |
2412 | * should be added at the tail | |
2413 | */ | |
bddd87c7 | 2414 | if (current->bio_list) { |
f5fe1b51 | 2415 | bio_list_add(¤t->bio_list[0], bio); |
dece1635 | 2416 | goto out; |
d89d8796 | 2417 | } |
27a84d54 | 2418 | |
d89d8796 NB |
2419 | /* following loop may be a bit non-obvious, and so deserves some |
2420 | * explanation. | |
2421 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
2422 | * ensure that) so we have a list with a single bio. | |
2423 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
2424 | * we assign bio_list to a pointer to the bio_list_on_stack, |
2425 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 2426 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
2427 | * through a recursive call to generic_make_request. If it |
2428 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
2429 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 2430 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 2431 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
2432 | */ |
2433 | BUG_ON(bio->bi_next); | |
f5fe1b51 N |
2434 | bio_list_init(&bio_list_on_stack[0]); |
2435 | current->bio_list = bio_list_on_stack; | |
d89d8796 | 2436 | do { |
37f9579f BVA |
2437 | bool enter_succeeded = true; |
2438 | ||
2439 | if (unlikely(q != bio->bi_disk->queue)) { | |
2440 | if (q) | |
2441 | blk_queue_exit(q); | |
2442 | q = bio->bi_disk->queue; | |
2443 | flags = 0; | |
2444 | if (bio->bi_opf & REQ_NOWAIT) | |
2445 | flags = BLK_MQ_REQ_NOWAIT; | |
2446 | if (blk_queue_enter(q, flags) < 0) { | |
2447 | enter_succeeded = false; | |
2448 | q = NULL; | |
2449 | } | |
2450 | } | |
27a84d54 | 2451 | |
37f9579f | 2452 | if (enter_succeeded) { |
79bd9959 N |
2453 | struct bio_list lower, same; |
2454 | ||
2455 | /* Create a fresh bio_list for all subordinate requests */ | |
f5fe1b51 N |
2456 | bio_list_on_stack[1] = bio_list_on_stack[0]; |
2457 | bio_list_init(&bio_list_on_stack[0]); | |
dece1635 | 2458 | ret = q->make_request_fn(q, bio); |
3ef28e83 | 2459 | |
79bd9959 N |
2460 | /* sort new bios into those for a lower level |
2461 | * and those for the same level | |
2462 | */ | |
2463 | bio_list_init(&lower); | |
2464 | bio_list_init(&same); | |
f5fe1b51 | 2465 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) |
74d46992 | 2466 | if (q == bio->bi_disk->queue) |
79bd9959 N |
2467 | bio_list_add(&same, bio); |
2468 | else | |
2469 | bio_list_add(&lower, bio); | |
2470 | /* now assemble so we handle the lowest level first */ | |
f5fe1b51 N |
2471 | bio_list_merge(&bio_list_on_stack[0], &lower); |
2472 | bio_list_merge(&bio_list_on_stack[0], &same); | |
2473 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
3ef28e83 | 2474 | } else { |
03a07c92 GR |
2475 | if (unlikely(!blk_queue_dying(q) && |
2476 | (bio->bi_opf & REQ_NOWAIT))) | |
2477 | bio_wouldblock_error(bio); | |
2478 | else | |
2479 | bio_io_error(bio); | |
3ef28e83 | 2480 | } |
f5fe1b51 | 2481 | bio = bio_list_pop(&bio_list_on_stack[0]); |
d89d8796 | 2482 | } while (bio); |
bddd87c7 | 2483 | current->bio_list = NULL; /* deactivate */ |
dece1635 JA |
2484 | |
2485 | out: | |
37f9579f BVA |
2486 | if (q) |
2487 | blk_queue_exit(q); | |
dece1635 | 2488 | return ret; |
d89d8796 | 2489 | } |
1da177e4 LT |
2490 | EXPORT_SYMBOL(generic_make_request); |
2491 | ||
f421e1d9 CH |
2492 | /** |
2493 | * direct_make_request - hand a buffer directly to its device driver for I/O | |
2494 | * @bio: The bio describing the location in memory and on the device. | |
2495 | * | |
2496 | * This function behaves like generic_make_request(), but does not protect | |
2497 | * against recursion. Must only be used if the called driver is known | |
2498 | * to not call generic_make_request (or direct_make_request) again from | |
2499 | * its make_request function. (Calling direct_make_request again from | |
2500 | * a workqueue is perfectly fine as that doesn't recurse). | |
2501 | */ | |
2502 | blk_qc_t direct_make_request(struct bio *bio) | |
2503 | { | |
2504 | struct request_queue *q = bio->bi_disk->queue; | |
2505 | bool nowait = bio->bi_opf & REQ_NOWAIT; | |
2506 | blk_qc_t ret; | |
2507 | ||
2508 | if (!generic_make_request_checks(bio)) | |
2509 | return BLK_QC_T_NONE; | |
2510 | ||
3a0a5299 | 2511 | if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) { |
f421e1d9 CH |
2512 | if (nowait && !blk_queue_dying(q)) |
2513 | bio->bi_status = BLK_STS_AGAIN; | |
2514 | else | |
2515 | bio->bi_status = BLK_STS_IOERR; | |
2516 | bio_endio(bio); | |
2517 | return BLK_QC_T_NONE; | |
2518 | } | |
2519 | ||
2520 | ret = q->make_request_fn(q, bio); | |
2521 | blk_queue_exit(q); | |
2522 | return ret; | |
2523 | } | |
2524 | EXPORT_SYMBOL_GPL(direct_make_request); | |
2525 | ||
1da177e4 | 2526 | /** |
710027a4 | 2527 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
2528 | * @bio: The &struct bio which describes the I/O |
2529 | * | |
2530 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
2531 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 2532 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
2533 | * |
2534 | */ | |
4e49ea4a | 2535 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 2536 | { |
bf2de6f5 JA |
2537 | /* |
2538 | * If it's a regular read/write or a barrier with data attached, | |
2539 | * go through the normal accounting stuff before submission. | |
2540 | */ | |
e2a60da7 | 2541 | if (bio_has_data(bio)) { |
4363ac7c MP |
2542 | unsigned int count; |
2543 | ||
95fe6c1a | 2544 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
7c5a0dcf | 2545 | count = queue_logical_block_size(bio->bi_disk->queue) >> 9; |
4363ac7c MP |
2546 | else |
2547 | count = bio_sectors(bio); | |
2548 | ||
a8ebb056 | 2549 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
2550 | count_vm_events(PGPGOUT, count); |
2551 | } else { | |
4f024f37 | 2552 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
2553 | count_vm_events(PGPGIN, count); |
2554 | } | |
2555 | ||
2556 | if (unlikely(block_dump)) { | |
2557 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 2558 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 2559 | current->comm, task_pid_nr(current), |
a8ebb056 | 2560 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 2561 | (unsigned long long)bio->bi_iter.bi_sector, |
74d46992 | 2562 | bio_devname(bio, b), count); |
bf2de6f5 | 2563 | } |
1da177e4 LT |
2564 | } |
2565 | ||
dece1635 | 2566 | return generic_make_request(bio); |
1da177e4 | 2567 | } |
1da177e4 LT |
2568 | EXPORT_SYMBOL(submit_bio); |
2569 | ||
ea435e1b CH |
2570 | bool blk_poll(struct request_queue *q, blk_qc_t cookie) |
2571 | { | |
2572 | if (!q->poll_fn || !blk_qc_t_valid(cookie)) | |
2573 | return false; | |
2574 | ||
2575 | if (current->plug) | |
2576 | blk_flush_plug_list(current->plug, false); | |
2577 | return q->poll_fn(q, cookie); | |
2578 | } | |
2579 | EXPORT_SYMBOL_GPL(blk_poll); | |
2580 | ||
82124d60 | 2581 | /** |
bf4e6b4e HR |
2582 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
2583 | * for new the queue limits | |
82124d60 KU |
2584 | * @q: the queue |
2585 | * @rq: the request being checked | |
2586 | * | |
2587 | * Description: | |
2588 | * @rq may have been made based on weaker limitations of upper-level queues | |
2589 | * in request stacking drivers, and it may violate the limitation of @q. | |
2590 | * Since the block layer and the underlying device driver trust @rq | |
2591 | * after it is inserted to @q, it should be checked against @q before | |
2592 | * the insertion using this generic function. | |
2593 | * | |
82124d60 | 2594 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
2595 | * limits when retrying requests on other queues. Those requests need |
2596 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 2597 | */ |
bf4e6b4e HR |
2598 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
2599 | struct request *rq) | |
82124d60 | 2600 | { |
8fe0d473 | 2601 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
82124d60 KU |
2602 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
2603 | return -EIO; | |
2604 | } | |
2605 | ||
2606 | /* | |
2607 | * queue's settings related to segment counting like q->bounce_pfn | |
2608 | * may differ from that of other stacking queues. | |
2609 | * Recalculate it to check the request correctly on this queue's | |
2610 | * limitation. | |
2611 | */ | |
2612 | blk_recalc_rq_segments(rq); | |
8a78362c | 2613 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
2614 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
2615 | return -EIO; | |
2616 | } | |
2617 | ||
2618 | return 0; | |
2619 | } | |
82124d60 KU |
2620 | |
2621 | /** | |
2622 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2623 | * @q: the queue to submit the request | |
2624 | * @rq: the request being queued | |
2625 | */ | |
2a842aca | 2626 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 KU |
2627 | { |
2628 | unsigned long flags; | |
4853abaa | 2629 | int where = ELEVATOR_INSERT_BACK; |
82124d60 | 2630 | |
bf4e6b4e | 2631 | if (blk_cloned_rq_check_limits(q, rq)) |
2a842aca | 2632 | return BLK_STS_IOERR; |
82124d60 | 2633 | |
b2c9cd37 AM |
2634 | if (rq->rq_disk && |
2635 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
2a842aca | 2636 | return BLK_STS_IOERR; |
82124d60 | 2637 | |
7fb4898e KB |
2638 | if (q->mq_ops) { |
2639 | if (blk_queue_io_stat(q)) | |
2640 | blk_account_io_start(rq, true); | |
157f377b JA |
2641 | /* |
2642 | * Since we have a scheduler attached on the top device, | |
2643 | * bypass a potential scheduler on the bottom device for | |
2644 | * insert. | |
2645 | */ | |
c77ff7fd | 2646 | return blk_mq_request_issue_directly(rq); |
7fb4898e KB |
2647 | } |
2648 | ||
82124d60 | 2649 | spin_lock_irqsave(q->queue_lock, flags); |
3f3299d5 | 2650 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 | 2651 | spin_unlock_irqrestore(q->queue_lock, flags); |
2a842aca | 2652 | return BLK_STS_IOERR; |
8ba61435 | 2653 | } |
82124d60 KU |
2654 | |
2655 | /* | |
2656 | * Submitting request must be dequeued before calling this function | |
2657 | * because it will be linked to another request_queue | |
2658 | */ | |
2659 | BUG_ON(blk_queued_rq(rq)); | |
2660 | ||
f73f44eb | 2661 | if (op_is_flush(rq->cmd_flags)) |
4853abaa JM |
2662 | where = ELEVATOR_INSERT_FLUSH; |
2663 | ||
2664 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
2665 | if (where == ELEVATOR_INSERT_FLUSH) |
2666 | __blk_run_queue(q); | |
82124d60 KU |
2667 | spin_unlock_irqrestore(q->queue_lock, flags); |
2668 | ||
2a842aca | 2669 | return BLK_STS_OK; |
82124d60 KU |
2670 | } |
2671 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2672 | ||
80a761fd TH |
2673 | /** |
2674 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2675 | * @rq: request to examine | |
2676 | * | |
2677 | * Description: | |
2678 | * A request could be merge of IOs which require different failure | |
2679 | * handling. This function determines the number of bytes which | |
2680 | * can be failed from the beginning of the request without | |
2681 | * crossing into area which need to be retried further. | |
2682 | * | |
2683 | * Return: | |
2684 | * The number of bytes to fail. | |
80a761fd TH |
2685 | */ |
2686 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2687 | { | |
2688 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2689 | unsigned int bytes = 0; | |
2690 | struct bio *bio; | |
2691 | ||
e8064021 | 2692 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
2693 | return blk_rq_bytes(rq); |
2694 | ||
2695 | /* | |
2696 | * Currently the only 'mixing' which can happen is between | |
2697 | * different fastfail types. We can safely fail portions | |
2698 | * which have all the failfast bits that the first one has - | |
2699 | * the ones which are at least as eager to fail as the first | |
2700 | * one. | |
2701 | */ | |
2702 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 2703 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 2704 | break; |
4f024f37 | 2705 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
2706 | } |
2707 | ||
2708 | /* this could lead to infinite loop */ | |
2709 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2710 | return bytes; | |
2711 | } | |
2712 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2713 | ||
320ae51f | 2714 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 2715 | { |
c2553b58 | 2716 | if (blk_do_io_stat(req)) { |
ddcf35d3 | 2717 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 JA |
2718 | struct hd_struct *part; |
2719 | int cpu; | |
2720 | ||
2721 | cpu = part_stat_lock(); | |
09e099d4 | 2722 | part = req->part; |
ddcf35d3 | 2723 | part_stat_add(cpu, part, sectors[sgrp], bytes >> 9); |
bc58ba94 JA |
2724 | part_stat_unlock(); |
2725 | } | |
2726 | } | |
2727 | ||
522a7775 | 2728 | void blk_account_io_done(struct request *req, u64 now) |
bc58ba94 | 2729 | { |
bc58ba94 | 2730 | /* |
dd4c133f TH |
2731 | * Account IO completion. flush_rq isn't accounted as a |
2732 | * normal IO on queueing nor completion. Accounting the | |
2733 | * containing request is enough. | |
bc58ba94 | 2734 | */ |
e8064021 | 2735 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
522a7775 | 2736 | unsigned long duration; |
ddcf35d3 | 2737 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 JA |
2738 | struct hd_struct *part; |
2739 | int cpu; | |
2740 | ||
522a7775 | 2741 | duration = nsecs_to_jiffies(now - req->start_time_ns); |
bc58ba94 | 2742 | cpu = part_stat_lock(); |
09e099d4 | 2743 | part = req->part; |
bc58ba94 | 2744 | |
ddcf35d3 MC |
2745 | part_stat_inc(cpu, part, ios[sgrp]); |
2746 | part_stat_add(cpu, part, ticks[sgrp], duration); | |
d62e26b3 | 2747 | part_round_stats(req->q, cpu, part); |
ddcf35d3 | 2748 | part_dec_in_flight(req->q, part, rq_data_dir(req)); |
bc58ba94 | 2749 | |
6c23a968 | 2750 | hd_struct_put(part); |
bc58ba94 JA |
2751 | part_stat_unlock(); |
2752 | } | |
2753 | } | |
2754 | ||
47fafbc7 | 2755 | #ifdef CONFIG_PM |
c8158819 LM |
2756 | /* |
2757 | * Don't process normal requests when queue is suspended | |
2758 | * or in the process of suspending/resuming | |
2759 | */ | |
e4f36b24 | 2760 | static bool blk_pm_allow_request(struct request *rq) |
c8158819 | 2761 | { |
e4f36b24 CH |
2762 | switch (rq->q->rpm_status) { |
2763 | case RPM_RESUMING: | |
2764 | case RPM_SUSPENDING: | |
2765 | return rq->rq_flags & RQF_PM; | |
2766 | case RPM_SUSPENDED: | |
2767 | return false; | |
e9a83853 GU |
2768 | default: |
2769 | return true; | |
e4f36b24 | 2770 | } |
c8158819 LM |
2771 | } |
2772 | #else | |
e4f36b24 | 2773 | static bool blk_pm_allow_request(struct request *rq) |
c8158819 | 2774 | { |
e4f36b24 | 2775 | return true; |
c8158819 LM |
2776 | } |
2777 | #endif | |
2778 | ||
320ae51f JA |
2779 | void blk_account_io_start(struct request *rq, bool new_io) |
2780 | { | |
2781 | struct hd_struct *part; | |
2782 | int rw = rq_data_dir(rq); | |
2783 | int cpu; | |
2784 | ||
2785 | if (!blk_do_io_stat(rq)) | |
2786 | return; | |
2787 | ||
2788 | cpu = part_stat_lock(); | |
2789 | ||
2790 | if (!new_io) { | |
2791 | part = rq->part; | |
2792 | part_stat_inc(cpu, part, merges[rw]); | |
2793 | } else { | |
2794 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2795 | if (!hd_struct_try_get(part)) { | |
2796 | /* | |
2797 | * The partition is already being removed, | |
2798 | * the request will be accounted on the disk only | |
2799 | * | |
2800 | * We take a reference on disk->part0 although that | |
2801 | * partition will never be deleted, so we can treat | |
2802 | * it as any other partition. | |
2803 | */ | |
2804 | part = &rq->rq_disk->part0; | |
2805 | hd_struct_get(part); | |
2806 | } | |
d62e26b3 JA |
2807 | part_round_stats(rq->q, cpu, part); |
2808 | part_inc_in_flight(rq->q, part, rw); | |
320ae51f JA |
2809 | rq->part = part; |
2810 | } | |
2811 | ||
2812 | part_stat_unlock(); | |
2813 | } | |
2814 | ||
9c988374 CH |
2815 | static struct request *elv_next_request(struct request_queue *q) |
2816 | { | |
2817 | struct request *rq; | |
2818 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); | |
2819 | ||
2820 | WARN_ON_ONCE(q->mq_ops); | |
2821 | ||
2822 | while (1) { | |
e4f36b24 CH |
2823 | list_for_each_entry(rq, &q->queue_head, queuelist) { |
2824 | if (blk_pm_allow_request(rq)) | |
2825 | return rq; | |
2826 | ||
2827 | if (rq->rq_flags & RQF_SOFTBARRIER) | |
2828 | break; | |
9c988374 CH |
2829 | } |
2830 | ||
2831 | /* | |
2832 | * Flush request is running and flush request isn't queueable | |
2833 | * in the drive, we can hold the queue till flush request is | |
2834 | * finished. Even we don't do this, driver can't dispatch next | |
2835 | * requests and will requeue them. And this can improve | |
2836 | * throughput too. For example, we have request flush1, write1, | |
2837 | * flush 2. flush1 is dispatched, then queue is hold, write1 | |
2838 | * isn't inserted to queue. After flush1 is finished, flush2 | |
2839 | * will be dispatched. Since disk cache is already clean, | |
2840 | * flush2 will be finished very soon, so looks like flush2 is | |
2841 | * folded to flush1. | |
2842 | * Since the queue is hold, a flag is set to indicate the queue | |
2843 | * should be restarted later. Please see flush_end_io() for | |
2844 | * details. | |
2845 | */ | |
2846 | if (fq->flush_pending_idx != fq->flush_running_idx && | |
2847 | !queue_flush_queueable(q)) { | |
2848 | fq->flush_queue_delayed = 1; | |
2849 | return NULL; | |
2850 | } | |
2851 | if (unlikely(blk_queue_bypass(q)) || | |
2852 | !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0)) | |
2853 | return NULL; | |
2854 | } | |
2855 | } | |
2856 | ||
3bcddeac | 2857 | /** |
9934c8c0 TH |
2858 | * blk_peek_request - peek at the top of a request queue |
2859 | * @q: request queue to peek at | |
2860 | * | |
2861 | * Description: | |
2862 | * Return the request at the top of @q. The returned request | |
2863 | * should be started using blk_start_request() before LLD starts | |
2864 | * processing it. | |
2865 | * | |
2866 | * Return: | |
2867 | * Pointer to the request at the top of @q if available. Null | |
2868 | * otherwise. | |
9934c8c0 TH |
2869 | */ |
2870 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2871 | { |
2872 | struct request *rq; | |
2873 | int ret; | |
2874 | ||
2fff8a92 | 2875 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 2876 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2877 | |
9c988374 | 2878 | while ((rq = elv_next_request(q)) != NULL) { |
e8064021 | 2879 | if (!(rq->rq_flags & RQF_STARTED)) { |
158dbda0 TH |
2880 | /* |
2881 | * This is the first time the device driver | |
2882 | * sees this request (possibly after | |
2883 | * requeueing). Notify IO scheduler. | |
2884 | */ | |
e8064021 | 2885 | if (rq->rq_flags & RQF_SORTED) |
158dbda0 TH |
2886 | elv_activate_rq(q, rq); |
2887 | ||
2888 | /* | |
2889 | * just mark as started even if we don't start | |
2890 | * it, a request that has been delayed should | |
2891 | * not be passed by new incoming requests | |
2892 | */ | |
e8064021 | 2893 | rq->rq_flags |= RQF_STARTED; |
158dbda0 TH |
2894 | trace_block_rq_issue(q, rq); |
2895 | } | |
2896 | ||
2897 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2898 | q->end_sector = rq_end_sector(rq); | |
2899 | q->boundary_rq = NULL; | |
2900 | } | |
2901 | ||
e8064021 | 2902 | if (rq->rq_flags & RQF_DONTPREP) |
158dbda0 TH |
2903 | break; |
2904 | ||
2e46e8b2 | 2905 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2906 | /* |
2907 | * make sure space for the drain appears we | |
2908 | * know we can do this because max_hw_segments | |
2909 | * has been adjusted to be one fewer than the | |
2910 | * device can handle | |
2911 | */ | |
2912 | rq->nr_phys_segments++; | |
2913 | } | |
2914 | ||
2915 | if (!q->prep_rq_fn) | |
2916 | break; | |
2917 | ||
2918 | ret = q->prep_rq_fn(q, rq); | |
2919 | if (ret == BLKPREP_OK) { | |
2920 | break; | |
2921 | } else if (ret == BLKPREP_DEFER) { | |
2922 | /* | |
2923 | * the request may have been (partially) prepped. | |
2924 | * we need to keep this request in the front to | |
e8064021 | 2925 | * avoid resource deadlock. RQF_STARTED will |
158dbda0 TH |
2926 | * prevent other fs requests from passing this one. |
2927 | */ | |
2e46e8b2 | 2928 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
e8064021 | 2929 | !(rq->rq_flags & RQF_DONTPREP)) { |
158dbda0 TH |
2930 | /* |
2931 | * remove the space for the drain we added | |
2932 | * so that we don't add it again | |
2933 | */ | |
2934 | --rq->nr_phys_segments; | |
2935 | } | |
2936 | ||
2937 | rq = NULL; | |
2938 | break; | |
0fb5b1fb | 2939 | } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) { |
e8064021 | 2940 | rq->rq_flags |= RQF_QUIET; |
c143dc90 JB |
2941 | /* |
2942 | * Mark this request as started so we don't trigger | |
2943 | * any debug logic in the end I/O path. | |
2944 | */ | |
2945 | blk_start_request(rq); | |
2a842aca CH |
2946 | __blk_end_request_all(rq, ret == BLKPREP_INVALID ? |
2947 | BLK_STS_TARGET : BLK_STS_IOERR); | |
158dbda0 TH |
2948 | } else { |
2949 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2950 | break; | |
2951 | } | |
2952 | } | |
2953 | ||
2954 | return rq; | |
2955 | } | |
9934c8c0 | 2956 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2957 | |
5034435c | 2958 | static void blk_dequeue_request(struct request *rq) |
158dbda0 | 2959 | { |
9934c8c0 TH |
2960 | struct request_queue *q = rq->q; |
2961 | ||
158dbda0 TH |
2962 | BUG_ON(list_empty(&rq->queuelist)); |
2963 | BUG_ON(ELV_ON_HASH(rq)); | |
2964 | ||
2965 | list_del_init(&rq->queuelist); | |
2966 | ||
2967 | /* | |
2968 | * the time frame between a request being removed from the lists | |
2969 | * and to it is freed is accounted as io that is in progress at | |
2970 | * the driver side. | |
2971 | */ | |
522a7775 | 2972 | if (blk_account_rq(rq)) |
0a7ae2ff | 2973 | q->in_flight[rq_is_sync(rq)]++; |
158dbda0 TH |
2974 | } |
2975 | ||
9934c8c0 TH |
2976 | /** |
2977 | * blk_start_request - start request processing on the driver | |
2978 | * @req: request to dequeue | |
2979 | * | |
2980 | * Description: | |
2981 | * Dequeue @req and start timeout timer on it. This hands off the | |
2982 | * request to the driver. | |
9934c8c0 TH |
2983 | */ |
2984 | void blk_start_request(struct request *req) | |
2985 | { | |
2fff8a92 | 2986 | lockdep_assert_held(req->q->queue_lock); |
332ebbf7 | 2987 | WARN_ON_ONCE(req->q->mq_ops); |
2fff8a92 | 2988 | |
9934c8c0 TH |
2989 | blk_dequeue_request(req); |
2990 | ||
cf43e6be | 2991 | if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) { |
544ccc8d OS |
2992 | req->io_start_time_ns = ktime_get_ns(); |
2993 | #ifdef CONFIG_BLK_DEV_THROTTLING_LOW | |
2994 | req->throtl_size = blk_rq_sectors(req); | |
2995 | #endif | |
cf43e6be | 2996 | req->rq_flags |= RQF_STATS; |
a7905043 | 2997 | rq_qos_issue(req->q, req); |
cf43e6be JA |
2998 | } |
2999 | ||
e14575b3 | 3000 | BUG_ON(blk_rq_is_complete(req)); |
9934c8c0 TH |
3001 | blk_add_timer(req); |
3002 | } | |
3003 | EXPORT_SYMBOL(blk_start_request); | |
3004 | ||
3005 | /** | |
3006 | * blk_fetch_request - fetch a request from a request queue | |
3007 | * @q: request queue to fetch a request from | |
3008 | * | |
3009 | * Description: | |
3010 | * Return the request at the top of @q. The request is started on | |
3011 | * return and LLD can start processing it immediately. | |
3012 | * | |
3013 | * Return: | |
3014 | * Pointer to the request at the top of @q if available. Null | |
3015 | * otherwise. | |
9934c8c0 TH |
3016 | */ |
3017 | struct request *blk_fetch_request(struct request_queue *q) | |
3018 | { | |
3019 | struct request *rq; | |
3020 | ||
2fff8a92 | 3021 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 3022 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 3023 | |
9934c8c0 TH |
3024 | rq = blk_peek_request(q); |
3025 | if (rq) | |
3026 | blk_start_request(rq); | |
3027 | return rq; | |
3028 | } | |
3029 | EXPORT_SYMBOL(blk_fetch_request); | |
3030 | ||
ef71de8b CH |
3031 | /* |
3032 | * Steal bios from a request and add them to a bio list. | |
3033 | * The request must not have been partially completed before. | |
3034 | */ | |
3035 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
3036 | { | |
3037 | if (rq->bio) { | |
3038 | if (list->tail) | |
3039 | list->tail->bi_next = rq->bio; | |
3040 | else | |
3041 | list->head = rq->bio; | |
3042 | list->tail = rq->biotail; | |
3043 | ||
3044 | rq->bio = NULL; | |
3045 | rq->biotail = NULL; | |
3046 | } | |
3047 | ||
3048 | rq->__data_len = 0; | |
3049 | } | |
3050 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
3051 | ||
3bcddeac | 3052 | /** |
2e60e022 | 3053 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 3054 | * @req: the request being processed |
2a842aca | 3055 | * @error: block status code |
8ebf9756 | 3056 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
3057 | * |
3058 | * Description: | |
8ebf9756 RD |
3059 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
3060 | * the request structure even if @req doesn't have leftover. | |
3061 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
3062 | * |
3063 | * This special helper function is only for request stacking drivers | |
3064 | * (e.g. request-based dm) so that they can handle partial completion. | |
3065 | * Actual device drivers should use blk_end_request instead. | |
3066 | * | |
3067 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
3068 | * %false return from this function. | |
3bcddeac | 3069 | * |
1954e9a9 BVA |
3070 | * Note: |
3071 | * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both | |
3072 | * blk_rq_bytes() and in blk_update_request(). | |
3073 | * | |
3bcddeac | 3074 | * Return: |
2e60e022 TH |
3075 | * %false - this request doesn't have any more data |
3076 | * %true - this request has more data | |
3bcddeac | 3077 | **/ |
2a842aca CH |
3078 | bool blk_update_request(struct request *req, blk_status_t error, |
3079 | unsigned int nr_bytes) | |
1da177e4 | 3080 | { |
f79ea416 | 3081 | int total_bytes; |
1da177e4 | 3082 | |
2a842aca | 3083 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
4a0efdc9 | 3084 | |
2e60e022 TH |
3085 | if (!req->bio) |
3086 | return false; | |
3087 | ||
2a842aca CH |
3088 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
3089 | !(req->rq_flags & RQF_QUIET))) | |
3090 | print_req_error(req, error); | |
1da177e4 | 3091 | |
bc58ba94 | 3092 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 3093 | |
f79ea416 KO |
3094 | total_bytes = 0; |
3095 | while (req->bio) { | |
3096 | struct bio *bio = req->bio; | |
4f024f37 | 3097 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 3098 | |
9c24c10a | 3099 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 3100 | req->bio = bio->bi_next; |
1da177e4 | 3101 | |
fbbaf700 N |
3102 | /* Completion has already been traced */ |
3103 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
f79ea416 | 3104 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 3105 | |
f79ea416 KO |
3106 | total_bytes += bio_bytes; |
3107 | nr_bytes -= bio_bytes; | |
1da177e4 | 3108 | |
f79ea416 KO |
3109 | if (!nr_bytes) |
3110 | break; | |
1da177e4 LT |
3111 | } |
3112 | ||
3113 | /* | |
3114 | * completely done | |
3115 | */ | |
2e60e022 TH |
3116 | if (!req->bio) { |
3117 | /* | |
3118 | * Reset counters so that the request stacking driver | |
3119 | * can find how many bytes remain in the request | |
3120 | * later. | |
3121 | */ | |
a2dec7b3 | 3122 | req->__data_len = 0; |
2e60e022 TH |
3123 | return false; |
3124 | } | |
1da177e4 | 3125 | |
a2dec7b3 | 3126 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
3127 | |
3128 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 3129 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 3130 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 3131 | |
80a761fd | 3132 | /* mixed attributes always follow the first bio */ |
e8064021 | 3133 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 3134 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 3135 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
3136 | } |
3137 | ||
ed6565e7 CH |
3138 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
3139 | /* | |
3140 | * If total number of sectors is less than the first segment | |
3141 | * size, something has gone terribly wrong. | |
3142 | */ | |
3143 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
3144 | blk_dump_rq_flags(req, "request botched"); | |
3145 | req->__data_len = blk_rq_cur_bytes(req); | |
3146 | } | |
2e46e8b2 | 3147 | |
ed6565e7 CH |
3148 | /* recalculate the number of segments */ |
3149 | blk_recalc_rq_segments(req); | |
3150 | } | |
2e46e8b2 | 3151 | |
2e60e022 | 3152 | return true; |
1da177e4 | 3153 | } |
2e60e022 | 3154 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 3155 | |
2a842aca | 3156 | static bool blk_update_bidi_request(struct request *rq, blk_status_t error, |
2e60e022 TH |
3157 | unsigned int nr_bytes, |
3158 | unsigned int bidi_bytes) | |
5efccd17 | 3159 | { |
2e60e022 TH |
3160 | if (blk_update_request(rq, error, nr_bytes)) |
3161 | return true; | |
5efccd17 | 3162 | |
2e60e022 TH |
3163 | /* Bidi request must be completed as a whole */ |
3164 | if (unlikely(blk_bidi_rq(rq)) && | |
3165 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
3166 | return true; | |
5efccd17 | 3167 | |
e2e1a148 JA |
3168 | if (blk_queue_add_random(rq->q)) |
3169 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
3170 | |
3171 | return false; | |
1da177e4 LT |
3172 | } |
3173 | ||
28018c24 JB |
3174 | /** |
3175 | * blk_unprep_request - unprepare a request | |
3176 | * @req: the request | |
3177 | * | |
3178 | * This function makes a request ready for complete resubmission (or | |
3179 | * completion). It happens only after all error handling is complete, | |
3180 | * so represents the appropriate moment to deallocate any resources | |
3181 | * that were allocated to the request in the prep_rq_fn. The queue | |
3182 | * lock is held when calling this. | |
3183 | */ | |
3184 | void blk_unprep_request(struct request *req) | |
3185 | { | |
3186 | struct request_queue *q = req->q; | |
3187 | ||
e8064021 | 3188 | req->rq_flags &= ~RQF_DONTPREP; |
28018c24 JB |
3189 | if (q->unprep_rq_fn) |
3190 | q->unprep_rq_fn(q, req); | |
3191 | } | |
3192 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
3193 | ||
2a842aca | 3194 | void blk_finish_request(struct request *req, blk_status_t error) |
1da177e4 | 3195 | { |
cf43e6be | 3196 | struct request_queue *q = req->q; |
522a7775 | 3197 | u64 now = ktime_get_ns(); |
cf43e6be | 3198 | |
2fff8a92 | 3199 | lockdep_assert_held(req->q->queue_lock); |
332ebbf7 | 3200 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 3201 | |
cf43e6be | 3202 | if (req->rq_flags & RQF_STATS) |
522a7775 | 3203 | blk_stat_add(req, now); |
cf43e6be | 3204 | |
e8064021 | 3205 | if (req->rq_flags & RQF_QUEUED) |
cf43e6be | 3206 | blk_queue_end_tag(q, req); |
b8286239 | 3207 | |
ba396a6c | 3208 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 3209 | |
57292b58 | 3210 | if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req)) |
dc3b17cc | 3211 | laptop_io_completion(req->q->backing_dev_info); |
1da177e4 | 3212 | |
e78042e5 MA |
3213 | blk_delete_timer(req); |
3214 | ||
e8064021 | 3215 | if (req->rq_flags & RQF_DONTPREP) |
28018c24 JB |
3216 | blk_unprep_request(req); |
3217 | ||
522a7775 | 3218 | blk_account_io_done(req, now); |
b8286239 | 3219 | |
87760e5e | 3220 | if (req->end_io) { |
a7905043 | 3221 | rq_qos_done(q, req); |
8ffdc655 | 3222 | req->end_io(req, error); |
87760e5e | 3223 | } else { |
b8286239 KU |
3224 | if (blk_bidi_rq(req)) |
3225 | __blk_put_request(req->next_rq->q, req->next_rq); | |
3226 | ||
cf43e6be | 3227 | __blk_put_request(q, req); |
b8286239 | 3228 | } |
1da177e4 | 3229 | } |
12120077 | 3230 | EXPORT_SYMBOL(blk_finish_request); |
1da177e4 | 3231 | |
3b11313a | 3232 | /** |
2e60e022 TH |
3233 | * blk_end_bidi_request - Complete a bidi request |
3234 | * @rq: the request to complete | |
2a842aca | 3235 | * @error: block status code |
2e60e022 TH |
3236 | * @nr_bytes: number of bytes to complete @rq |
3237 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
3238 | * |
3239 | * Description: | |
e3a04fe3 | 3240 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
3241 | * Drivers that supports bidi can safely call this member for any |
3242 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
3243 | * just ignored. | |
336cdb40 KU |
3244 | * |
3245 | * Return: | |
2e60e022 TH |
3246 | * %false - we are done with this request |
3247 | * %true - still buffers pending for this request | |
a0cd1285 | 3248 | **/ |
2a842aca | 3249 | static bool blk_end_bidi_request(struct request *rq, blk_status_t error, |
32fab448 KU |
3250 | unsigned int nr_bytes, unsigned int bidi_bytes) |
3251 | { | |
336cdb40 | 3252 | struct request_queue *q = rq->q; |
2e60e022 | 3253 | unsigned long flags; |
32fab448 | 3254 | |
332ebbf7 BVA |
3255 | WARN_ON_ONCE(q->mq_ops); |
3256 | ||
2e60e022 TH |
3257 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
3258 | return true; | |
32fab448 | 3259 | |
336cdb40 | 3260 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 3261 | blk_finish_request(rq, error); |
336cdb40 KU |
3262 | spin_unlock_irqrestore(q->queue_lock, flags); |
3263 | ||
2e60e022 | 3264 | return false; |
32fab448 KU |
3265 | } |
3266 | ||
336cdb40 | 3267 | /** |
2e60e022 TH |
3268 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
3269 | * @rq: the request to complete | |
2a842aca | 3270 | * @error: block status code |
e3a04fe3 KU |
3271 | * @nr_bytes: number of bytes to complete @rq |
3272 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
3273 | * |
3274 | * Description: | |
2e60e022 TH |
3275 | * Identical to blk_end_bidi_request() except that queue lock is |
3276 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
3277 | * |
3278 | * Return: | |
2e60e022 TH |
3279 | * %false - we are done with this request |
3280 | * %true - still buffers pending for this request | |
336cdb40 | 3281 | **/ |
2a842aca | 3282 | static bool __blk_end_bidi_request(struct request *rq, blk_status_t error, |
b1f74493 | 3283 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 3284 | { |
2fff8a92 | 3285 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3286 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3287 | |
2e60e022 TH |
3288 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
3289 | return true; | |
336cdb40 | 3290 | |
2e60e022 | 3291 | blk_finish_request(rq, error); |
336cdb40 | 3292 | |
2e60e022 | 3293 | return false; |
336cdb40 | 3294 | } |
e19a3ab0 KU |
3295 | |
3296 | /** | |
3297 | * blk_end_request - Helper function for drivers to complete the request. | |
3298 | * @rq: the request being processed | |
2a842aca | 3299 | * @error: block status code |
e19a3ab0 KU |
3300 | * @nr_bytes: number of bytes to complete |
3301 | * | |
3302 | * Description: | |
3303 | * Ends I/O on a number of bytes attached to @rq. | |
3304 | * If @rq has leftover, sets it up for the next range of segments. | |
3305 | * | |
3306 | * Return: | |
b1f74493 FT |
3307 | * %false - we are done with this request |
3308 | * %true - still buffers pending for this request | |
e19a3ab0 | 3309 | **/ |
2a842aca CH |
3310 | bool blk_end_request(struct request *rq, blk_status_t error, |
3311 | unsigned int nr_bytes) | |
e19a3ab0 | 3312 | { |
332ebbf7 | 3313 | WARN_ON_ONCE(rq->q->mq_ops); |
b1f74493 | 3314 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 3315 | } |
56ad1740 | 3316 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
3317 | |
3318 | /** | |
b1f74493 FT |
3319 | * blk_end_request_all - Helper function for drives to finish the request. |
3320 | * @rq: the request to finish | |
2a842aca | 3321 | * @error: block status code |
336cdb40 KU |
3322 | * |
3323 | * Description: | |
b1f74493 FT |
3324 | * Completely finish @rq. |
3325 | */ | |
2a842aca | 3326 | void blk_end_request_all(struct request *rq, blk_status_t error) |
336cdb40 | 3327 | { |
b1f74493 FT |
3328 | bool pending; |
3329 | unsigned int bidi_bytes = 0; | |
336cdb40 | 3330 | |
b1f74493 FT |
3331 | if (unlikely(blk_bidi_rq(rq))) |
3332 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 3333 | |
b1f74493 FT |
3334 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
3335 | BUG_ON(pending); | |
3336 | } | |
56ad1740 | 3337 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 3338 | |
e3a04fe3 | 3339 | /** |
b1f74493 FT |
3340 | * __blk_end_request - Helper function for drivers to complete the request. |
3341 | * @rq: the request being processed | |
2a842aca | 3342 | * @error: block status code |
b1f74493 | 3343 | * @nr_bytes: number of bytes to complete |
e3a04fe3 KU |
3344 | * |
3345 | * Description: | |
b1f74493 | 3346 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
3347 | * |
3348 | * Return: | |
b1f74493 FT |
3349 | * %false - we are done with this request |
3350 | * %true - still buffers pending for this request | |
e3a04fe3 | 3351 | **/ |
2a842aca CH |
3352 | bool __blk_end_request(struct request *rq, blk_status_t error, |
3353 | unsigned int nr_bytes) | |
e3a04fe3 | 3354 | { |
2fff8a92 | 3355 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3356 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3357 | |
b1f74493 | 3358 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 3359 | } |
56ad1740 | 3360 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 3361 | |
32fab448 | 3362 | /** |
b1f74493 FT |
3363 | * __blk_end_request_all - Helper function for drives to finish the request. |
3364 | * @rq: the request to finish | |
2a842aca | 3365 | * @error: block status code |
32fab448 KU |
3366 | * |
3367 | * Description: | |
b1f74493 | 3368 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 3369 | */ |
2a842aca | 3370 | void __blk_end_request_all(struct request *rq, blk_status_t error) |
32fab448 | 3371 | { |
b1f74493 FT |
3372 | bool pending; |
3373 | unsigned int bidi_bytes = 0; | |
3374 | ||
2fff8a92 | 3375 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3376 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3377 | |
b1f74493 FT |
3378 | if (unlikely(blk_bidi_rq(rq))) |
3379 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
3380 | ||
3381 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
3382 | BUG_ON(pending); | |
32fab448 | 3383 | } |
56ad1740 | 3384 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 3385 | |
e19a3ab0 | 3386 | /** |
b1f74493 FT |
3387 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
3388 | * @rq: the request to finish the current chunk for | |
2a842aca | 3389 | * @error: block status code |
e19a3ab0 KU |
3390 | * |
3391 | * Description: | |
b1f74493 FT |
3392 | * Complete the current consecutively mapped chunk from @rq. Must |
3393 | * be called with queue lock held. | |
e19a3ab0 KU |
3394 | * |
3395 | * Return: | |
b1f74493 FT |
3396 | * %false - we are done with this request |
3397 | * %true - still buffers pending for this request | |
3398 | */ | |
2a842aca | 3399 | bool __blk_end_request_cur(struct request *rq, blk_status_t error) |
e19a3ab0 | 3400 | { |
b1f74493 | 3401 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 3402 | } |
56ad1740 | 3403 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 3404 | |
86db1e29 JA |
3405 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
3406 | struct bio *bio) | |
1da177e4 | 3407 | { |
b4f42e28 | 3408 | if (bio_has_data(bio)) |
fb2dce86 | 3409 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
445251d0 JA |
3410 | else if (bio_op(bio) == REQ_OP_DISCARD) |
3411 | rq->nr_phys_segments = 1; | |
b4f42e28 | 3412 | |
4f024f37 | 3413 | rq->__data_len = bio->bi_iter.bi_size; |
1da177e4 | 3414 | rq->bio = rq->biotail = bio; |
1da177e4 | 3415 | |
74d46992 CH |
3416 | if (bio->bi_disk) |
3417 | rq->rq_disk = bio->bi_disk; | |
66846572 | 3418 | } |
1da177e4 | 3419 | |
2d4dc890 IL |
3420 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
3421 | /** | |
3422 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
3423 | * @rq: the request to be flushed | |
3424 | * | |
3425 | * Description: | |
3426 | * Flush all pages in @rq. | |
3427 | */ | |
3428 | void rq_flush_dcache_pages(struct request *rq) | |
3429 | { | |
3430 | struct req_iterator iter; | |
7988613b | 3431 | struct bio_vec bvec; |
2d4dc890 IL |
3432 | |
3433 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 3434 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
3435 | } |
3436 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
3437 | #endif | |
3438 | ||
ef9e3fac KU |
3439 | /** |
3440 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
3441 | * @q : the queue of the device being checked | |
3442 | * | |
3443 | * Description: | |
3444 | * Check if underlying low-level drivers of a device are busy. | |
3445 | * If the drivers want to export their busy state, they must set own | |
3446 | * exporting function using blk_queue_lld_busy() first. | |
3447 | * | |
3448 | * Basically, this function is used only by request stacking drivers | |
3449 | * to stop dispatching requests to underlying devices when underlying | |
3450 | * devices are busy. This behavior helps more I/O merging on the queue | |
3451 | * of the request stacking driver and prevents I/O throughput regression | |
3452 | * on burst I/O load. | |
3453 | * | |
3454 | * Return: | |
3455 | * 0 - Not busy (The request stacking driver should dispatch request) | |
3456 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
3457 | */ | |
3458 | int blk_lld_busy(struct request_queue *q) | |
3459 | { | |
3460 | if (q->lld_busy_fn) | |
3461 | return q->lld_busy_fn(q); | |
3462 | ||
3463 | return 0; | |
3464 | } | |
3465 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
3466 | ||
78d8e58a MS |
3467 | /** |
3468 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
3469 | * @rq: the clone request to be cleaned up | |
3470 | * | |
3471 | * Description: | |
3472 | * Free all bios in @rq for a cloned request. | |
3473 | */ | |
3474 | void blk_rq_unprep_clone(struct request *rq) | |
3475 | { | |
3476 | struct bio *bio; | |
3477 | ||
3478 | while ((bio = rq->bio) != NULL) { | |
3479 | rq->bio = bio->bi_next; | |
3480 | ||
3481 | bio_put(bio); | |
3482 | } | |
3483 | } | |
3484 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
3485 | ||
3486 | /* | |
3487 | * Copy attributes of the original request to the clone request. | |
3488 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
3489 | */ | |
3490 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
b0fd271d KU |
3491 | { |
3492 | dst->cpu = src->cpu; | |
b0fd271d KU |
3493 | dst->__sector = blk_rq_pos(src); |
3494 | dst->__data_len = blk_rq_bytes(src); | |
297ba57d BVA |
3495 | if (src->rq_flags & RQF_SPECIAL_PAYLOAD) { |
3496 | dst->rq_flags |= RQF_SPECIAL_PAYLOAD; | |
3497 | dst->special_vec = src->special_vec; | |
3498 | } | |
b0fd271d KU |
3499 | dst->nr_phys_segments = src->nr_phys_segments; |
3500 | dst->ioprio = src->ioprio; | |
3501 | dst->extra_len = src->extra_len; | |
78d8e58a MS |
3502 | } |
3503 | ||
3504 | /** | |
3505 | * blk_rq_prep_clone - Helper function to setup clone request | |
3506 | * @rq: the request to be setup | |
3507 | * @rq_src: original request to be cloned | |
3508 | * @bs: bio_set that bios for clone are allocated from | |
3509 | * @gfp_mask: memory allocation mask for bio | |
3510 | * @bio_ctr: setup function to be called for each clone bio. | |
3511 | * Returns %0 for success, non %0 for failure. | |
3512 | * @data: private data to be passed to @bio_ctr | |
3513 | * | |
3514 | * Description: | |
3515 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
3516 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
3517 | * are not copied, and copying such parts is the caller's responsibility. | |
3518 | * Also, pages which the original bios are pointing to are not copied | |
3519 | * and the cloned bios just point same pages. | |
3520 | * So cloned bios must be completed before original bios, which means | |
3521 | * the caller must complete @rq before @rq_src. | |
3522 | */ | |
3523 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
3524 | struct bio_set *bs, gfp_t gfp_mask, | |
3525 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
3526 | void *data) | |
3527 | { | |
3528 | struct bio *bio, *bio_src; | |
3529 | ||
3530 | if (!bs) | |
f4f8154a | 3531 | bs = &fs_bio_set; |
78d8e58a MS |
3532 | |
3533 | __rq_for_each_bio(bio_src, rq_src) { | |
3534 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
3535 | if (!bio) | |
3536 | goto free_and_out; | |
3537 | ||
3538 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
3539 | goto free_and_out; | |
3540 | ||
3541 | if (rq->bio) { | |
3542 | rq->biotail->bi_next = bio; | |
3543 | rq->biotail = bio; | |
3544 | } else | |
3545 | rq->bio = rq->biotail = bio; | |
3546 | } | |
3547 | ||
3548 | __blk_rq_prep_clone(rq, rq_src); | |
3549 | ||
3550 | return 0; | |
3551 | ||
3552 | free_and_out: | |
3553 | if (bio) | |
3554 | bio_put(bio); | |
3555 | blk_rq_unprep_clone(rq); | |
3556 | ||
3557 | return -ENOMEM; | |
b0fd271d KU |
3558 | } |
3559 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
3560 | ||
59c3d45e | 3561 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
3562 | { |
3563 | return queue_work(kblockd_workqueue, work); | |
3564 | } | |
1da177e4 LT |
3565 | EXPORT_SYMBOL(kblockd_schedule_work); |
3566 | ||
ee63cfa7 JA |
3567 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
3568 | { | |
3569 | return queue_work_on(cpu, kblockd_workqueue, work); | |
3570 | } | |
3571 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
3572 | ||
818cd1cb JA |
3573 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
3574 | unsigned long delay) | |
3575 | { | |
3576 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3577 | } | |
3578 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
3579 | ||
75df7136 SJ |
3580 | /** |
3581 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
3582 | * @plug: The &struct blk_plug that needs to be initialized | |
3583 | * | |
3584 | * Description: | |
3585 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
3586 | * pending I/O should the task end up blocking between blk_start_plug() and | |
3587 | * blk_finish_plug(). This is important from a performance perspective, but | |
3588 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
3589 | * for a memory allocation, memory reclaim could end up wanting to free a | |
3590 | * page belonging to that request that is currently residing in our private | |
3591 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
3592 | * this kind of deadlock. | |
3593 | */ | |
73c10101 JA |
3594 | void blk_start_plug(struct blk_plug *plug) |
3595 | { | |
3596 | struct task_struct *tsk = current; | |
3597 | ||
dd6cf3e1 SL |
3598 | /* |
3599 | * If this is a nested plug, don't actually assign it. | |
3600 | */ | |
3601 | if (tsk->plug) | |
3602 | return; | |
3603 | ||
73c10101 | 3604 | INIT_LIST_HEAD(&plug->list); |
320ae51f | 3605 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 3606 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 | 3607 | /* |
dd6cf3e1 SL |
3608 | * Store ordering should not be needed here, since a potential |
3609 | * preempt will imply a full memory barrier | |
73c10101 | 3610 | */ |
dd6cf3e1 | 3611 | tsk->plug = plug; |
73c10101 JA |
3612 | } |
3613 | EXPORT_SYMBOL(blk_start_plug); | |
3614 | ||
3615 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3616 | { | |
3617 | struct request *rqa = container_of(a, struct request, queuelist); | |
3618 | struct request *rqb = container_of(b, struct request, queuelist); | |
3619 | ||
975927b9 JM |
3620 | return !(rqa->q < rqb->q || |
3621 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
73c10101 JA |
3622 | } |
3623 | ||
49cac01e JA |
3624 | /* |
3625 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
3626 | * until a safe kblockd context. We due this to avoid accidental big | |
3627 | * additional stack usage in driver dispatch, in places where the originally | |
3628 | * plugger did not intend it. | |
3629 | */ | |
f6603783 | 3630 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 3631 | bool from_schedule) |
99e22598 | 3632 | __releases(q->queue_lock) |
94b5eb28 | 3633 | { |
2fff8a92 BVA |
3634 | lockdep_assert_held(q->queue_lock); |
3635 | ||
49cac01e | 3636 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 3637 | |
70460571 | 3638 | if (from_schedule) |
24ecfbe2 | 3639 | blk_run_queue_async(q); |
70460571 | 3640 | else |
24ecfbe2 | 3641 | __blk_run_queue(q); |
50864670 | 3642 | spin_unlock_irq(q->queue_lock); |
94b5eb28 JA |
3643 | } |
3644 | ||
74018dc3 | 3645 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
3646 | { |
3647 | LIST_HEAD(callbacks); | |
3648 | ||
2a7d5559 SL |
3649 | while (!list_empty(&plug->cb_list)) { |
3650 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 3651 | |
2a7d5559 SL |
3652 | while (!list_empty(&callbacks)) { |
3653 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
3654 | struct blk_plug_cb, |
3655 | list); | |
2a7d5559 | 3656 | list_del(&cb->list); |
74018dc3 | 3657 | cb->callback(cb, from_schedule); |
2a7d5559 | 3658 | } |
048c9374 N |
3659 | } |
3660 | } | |
3661 | ||
9cbb1750 N |
3662 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
3663 | int size) | |
3664 | { | |
3665 | struct blk_plug *plug = current->plug; | |
3666 | struct blk_plug_cb *cb; | |
3667 | ||
3668 | if (!plug) | |
3669 | return NULL; | |
3670 | ||
3671 | list_for_each_entry(cb, &plug->cb_list, list) | |
3672 | if (cb->callback == unplug && cb->data == data) | |
3673 | return cb; | |
3674 | ||
3675 | /* Not currently on the callback list */ | |
3676 | BUG_ON(size < sizeof(*cb)); | |
3677 | cb = kzalloc(size, GFP_ATOMIC); | |
3678 | if (cb) { | |
3679 | cb->data = data; | |
3680 | cb->callback = unplug; | |
3681 | list_add(&cb->list, &plug->cb_list); | |
3682 | } | |
3683 | return cb; | |
3684 | } | |
3685 | EXPORT_SYMBOL(blk_check_plugged); | |
3686 | ||
49cac01e | 3687 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
3688 | { |
3689 | struct request_queue *q; | |
73c10101 | 3690 | struct request *rq; |
109b8129 | 3691 | LIST_HEAD(list); |
94b5eb28 | 3692 | unsigned int depth; |
73c10101 | 3693 | |
74018dc3 | 3694 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
3695 | |
3696 | if (!list_empty(&plug->mq_list)) | |
3697 | blk_mq_flush_plug_list(plug, from_schedule); | |
3698 | ||
73c10101 JA |
3699 | if (list_empty(&plug->list)) |
3700 | return; | |
3701 | ||
109b8129 N |
3702 | list_splice_init(&plug->list, &list); |
3703 | ||
422765c2 | 3704 | list_sort(NULL, &list, plug_rq_cmp); |
73c10101 JA |
3705 | |
3706 | q = NULL; | |
94b5eb28 | 3707 | depth = 0; |
18811272 | 3708 | |
109b8129 N |
3709 | while (!list_empty(&list)) { |
3710 | rq = list_entry_rq(list.next); | |
73c10101 | 3711 | list_del_init(&rq->queuelist); |
73c10101 JA |
3712 | BUG_ON(!rq->q); |
3713 | if (rq->q != q) { | |
99e22598 JA |
3714 | /* |
3715 | * This drops the queue lock | |
3716 | */ | |
3717 | if (q) | |
49cac01e | 3718 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3719 | q = rq->q; |
94b5eb28 | 3720 | depth = 0; |
50864670 | 3721 | spin_lock_irq(q->queue_lock); |
73c10101 | 3722 | } |
8ba61435 TH |
3723 | |
3724 | /* | |
3725 | * Short-circuit if @q is dead | |
3726 | */ | |
3f3299d5 | 3727 | if (unlikely(blk_queue_dying(q))) { |
2a842aca | 3728 | __blk_end_request_all(rq, BLK_STS_IOERR); |
8ba61435 TH |
3729 | continue; |
3730 | } | |
3731 | ||
73c10101 JA |
3732 | /* |
3733 | * rq is already accounted, so use raw insert | |
3734 | */ | |
f73f44eb | 3735 | if (op_is_flush(rq->cmd_flags)) |
401a18e9 JA |
3736 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); |
3737 | else | |
3738 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
3739 | |
3740 | depth++; | |
73c10101 JA |
3741 | } |
3742 | ||
99e22598 JA |
3743 | /* |
3744 | * This drops the queue lock | |
3745 | */ | |
3746 | if (q) | |
49cac01e | 3747 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3748 | } |
73c10101 JA |
3749 | |
3750 | void blk_finish_plug(struct blk_plug *plug) | |
3751 | { | |
dd6cf3e1 SL |
3752 | if (plug != current->plug) |
3753 | return; | |
f6603783 | 3754 | blk_flush_plug_list(plug, false); |
73c10101 | 3755 | |
dd6cf3e1 | 3756 | current->plug = NULL; |
73c10101 | 3757 | } |
88b996cd | 3758 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 3759 | |
47fafbc7 | 3760 | #ifdef CONFIG_PM |
6c954667 LM |
3761 | /** |
3762 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3763 | * @q: the queue of the device | |
3764 | * @dev: the device the queue belongs to | |
3765 | * | |
3766 | * Description: | |
3767 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3768 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3769 | * should call this function after @dev has been initialized, and its | |
3770 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3771 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3772 | * cases, driver should call this function before any I/O has taken place. | |
3773 | * | |
3774 | * This function takes care of setting up using auto suspend for the device, | |
3775 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3776 | * until an updated value is either set by user or by driver. Drivers do | |
3777 | * not need to touch other autosuspend settings. | |
3778 | * | |
3779 | * The block layer runtime PM is request based, so only works for drivers | |
3780 | * that use request as their IO unit instead of those directly use bio's. | |
3781 | */ | |
3782 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3783 | { | |
b233f127 ML |
3784 | /* Don't enable runtime PM for blk-mq until it is ready */ |
3785 | if (q->mq_ops) { | |
3786 | pm_runtime_disable(dev); | |
765e40b6 | 3787 | return; |
b233f127 | 3788 | } |
765e40b6 | 3789 | |
6c954667 LM |
3790 | q->dev = dev; |
3791 | q->rpm_status = RPM_ACTIVE; | |
3792 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3793 | pm_runtime_use_autosuspend(q->dev); | |
3794 | } | |
3795 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3796 | ||
3797 | /** | |
3798 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3799 | * @q: the queue of the device | |
3800 | * | |
3801 | * Description: | |
3802 | * This function will check if runtime suspend is allowed for the device | |
3803 | * by examining if there are any requests pending in the queue. If there | |
3804 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3805 | * the queue's status will be updated to SUSPENDING and the driver can | |
3806 | * proceed to suspend the device. | |
3807 | * | |
3808 | * For the not allowed case, we mark last busy for the device so that | |
3809 | * runtime PM core will try to autosuspend it some time later. | |
3810 | * | |
3811 | * This function should be called near the start of the device's | |
3812 | * runtime_suspend callback. | |
3813 | * | |
3814 | * Return: | |
3815 | * 0 - OK to runtime suspend the device | |
3816 | * -EBUSY - Device should not be runtime suspended | |
3817 | */ | |
3818 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3819 | { | |
3820 | int ret = 0; | |
3821 | ||
4fd41a85 KX |
3822 | if (!q->dev) |
3823 | return ret; | |
3824 | ||
6c954667 LM |
3825 | spin_lock_irq(q->queue_lock); |
3826 | if (q->nr_pending) { | |
3827 | ret = -EBUSY; | |
3828 | pm_runtime_mark_last_busy(q->dev); | |
3829 | } else { | |
3830 | q->rpm_status = RPM_SUSPENDING; | |
3831 | } | |
3832 | spin_unlock_irq(q->queue_lock); | |
3833 | return ret; | |
3834 | } | |
3835 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3836 | ||
3837 | /** | |
3838 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3839 | * @q: the queue of the device | |
3840 | * @err: return value of the device's runtime_suspend function | |
3841 | * | |
3842 | * Description: | |
3843 | * Update the queue's runtime status according to the return value of the | |
3844 | * device's runtime suspend function and mark last busy for the device so | |
3845 | * that PM core will try to auto suspend the device at a later time. | |
3846 | * | |
3847 | * This function should be called near the end of the device's | |
3848 | * runtime_suspend callback. | |
3849 | */ | |
3850 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3851 | { | |
4fd41a85 KX |
3852 | if (!q->dev) |
3853 | return; | |
3854 | ||
6c954667 LM |
3855 | spin_lock_irq(q->queue_lock); |
3856 | if (!err) { | |
3857 | q->rpm_status = RPM_SUSPENDED; | |
3858 | } else { | |
3859 | q->rpm_status = RPM_ACTIVE; | |
3860 | pm_runtime_mark_last_busy(q->dev); | |
3861 | } | |
3862 | spin_unlock_irq(q->queue_lock); | |
3863 | } | |
3864 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3865 | ||
3866 | /** | |
3867 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3868 | * @q: the queue of the device | |
3869 | * | |
3870 | * Description: | |
3871 | * Update the queue's runtime status to RESUMING in preparation for the | |
3872 | * runtime resume of the device. | |
3873 | * | |
3874 | * This function should be called near the start of the device's | |
3875 | * runtime_resume callback. | |
3876 | */ | |
3877 | void blk_pre_runtime_resume(struct request_queue *q) | |
3878 | { | |
4fd41a85 KX |
3879 | if (!q->dev) |
3880 | return; | |
3881 | ||
6c954667 LM |
3882 | spin_lock_irq(q->queue_lock); |
3883 | q->rpm_status = RPM_RESUMING; | |
3884 | spin_unlock_irq(q->queue_lock); | |
3885 | } | |
3886 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3887 | ||
3888 | /** | |
3889 | * blk_post_runtime_resume - Post runtime resume processing | |
3890 | * @q: the queue of the device | |
3891 | * @err: return value of the device's runtime_resume function | |
3892 | * | |
3893 | * Description: | |
3894 | * Update the queue's runtime status according to the return value of the | |
3895 | * device's runtime_resume function. If it is successfully resumed, process | |
3896 | * the requests that are queued into the device's queue when it is resuming | |
3897 | * and then mark last busy and initiate autosuspend for it. | |
3898 | * | |
3899 | * This function should be called near the end of the device's | |
3900 | * runtime_resume callback. | |
3901 | */ | |
3902 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3903 | { | |
4fd41a85 KX |
3904 | if (!q->dev) |
3905 | return; | |
3906 | ||
6c954667 LM |
3907 | spin_lock_irq(q->queue_lock); |
3908 | if (!err) { | |
3909 | q->rpm_status = RPM_ACTIVE; | |
3910 | __blk_run_queue(q); | |
3911 | pm_runtime_mark_last_busy(q->dev); | |
c60855cd | 3912 | pm_request_autosuspend(q->dev); |
6c954667 LM |
3913 | } else { |
3914 | q->rpm_status = RPM_SUSPENDED; | |
3915 | } | |
3916 | spin_unlock_irq(q->queue_lock); | |
3917 | } | |
3918 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
d07ab6d1 MW |
3919 | |
3920 | /** | |
3921 | * blk_set_runtime_active - Force runtime status of the queue to be active | |
3922 | * @q: the queue of the device | |
3923 | * | |
3924 | * If the device is left runtime suspended during system suspend the resume | |
3925 | * hook typically resumes the device and corrects runtime status | |
3926 | * accordingly. However, that does not affect the queue runtime PM status | |
3927 | * which is still "suspended". This prevents processing requests from the | |
3928 | * queue. | |
3929 | * | |
3930 | * This function can be used in driver's resume hook to correct queue | |
3931 | * runtime PM status and re-enable peeking requests from the queue. It | |
3932 | * should be called before first request is added to the queue. | |
3933 | */ | |
3934 | void blk_set_runtime_active(struct request_queue *q) | |
3935 | { | |
3936 | spin_lock_irq(q->queue_lock); | |
3937 | q->rpm_status = RPM_ACTIVE; | |
3938 | pm_runtime_mark_last_busy(q->dev); | |
3939 | pm_request_autosuspend(q->dev); | |
3940 | spin_unlock_irq(q->queue_lock); | |
3941 | } | |
3942 | EXPORT_SYMBOL(blk_set_runtime_active); | |
6c954667 LM |
3943 | #endif |
3944 | ||
1da177e4 LT |
3945 | int __init blk_dev_init(void) |
3946 | { | |
ef295ecf CH |
3947 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
3948 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
0762b23d | 3949 | FIELD_SIZEOF(struct request, cmd_flags)); |
ef295ecf CH |
3950 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
3951 | FIELD_SIZEOF(struct bio, bi_opf)); | |
9eb55b03 | 3952 | |
89b90be2 TH |
3953 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
3954 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 3955 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
3956 | if (!kblockd_workqueue) |
3957 | panic("Failed to create kblockd\n"); | |
3958 | ||
3959 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3960 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 3961 | |
c2789bd4 | 3962 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 3963 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3964 | |
18fbda91 OS |
3965 | #ifdef CONFIG_DEBUG_FS |
3966 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
3967 | #endif | |
3968 | ||
d38ecf93 | 3969 | return 0; |
1da177e4 | 3970 | } |