]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - block/blk-mq-sysfs.c
blk-mq-sched: add framework for MQ capable IO schedulers
[mirror_ubuntu-bionic-kernel.git] / block / blk-mq-sysfs.c
1 #include <linux/kernel.h>
2 #include <linux/module.h>
3 #include <linux/backing-dev.h>
4 #include <linux/bio.h>
5 #include <linux/blkdev.h>
6 #include <linux/mm.h>
7 #include <linux/init.h>
8 #include <linux/slab.h>
9 #include <linux/workqueue.h>
10 #include <linux/smp.h>
11
12 #include <linux/blk-mq.h>
13 #include "blk-mq.h"
14 #include "blk-mq-tag.h"
15
16 static void blk_mq_sysfs_release(struct kobject *kobj)
17 {
18 }
19
20 struct blk_mq_ctx_sysfs_entry {
21 struct attribute attr;
22 ssize_t (*show)(struct blk_mq_ctx *, char *);
23 ssize_t (*store)(struct blk_mq_ctx *, const char *, size_t);
24 };
25
26 struct blk_mq_hw_ctx_sysfs_entry {
27 struct attribute attr;
28 ssize_t (*show)(struct blk_mq_hw_ctx *, char *);
29 ssize_t (*store)(struct blk_mq_hw_ctx *, const char *, size_t);
30 };
31
32 static ssize_t blk_mq_sysfs_show(struct kobject *kobj, struct attribute *attr,
33 char *page)
34 {
35 struct blk_mq_ctx_sysfs_entry *entry;
36 struct blk_mq_ctx *ctx;
37 struct request_queue *q;
38 ssize_t res;
39
40 entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
41 ctx = container_of(kobj, struct blk_mq_ctx, kobj);
42 q = ctx->queue;
43
44 if (!entry->show)
45 return -EIO;
46
47 res = -ENOENT;
48 mutex_lock(&q->sysfs_lock);
49 if (!blk_queue_dying(q))
50 res = entry->show(ctx, page);
51 mutex_unlock(&q->sysfs_lock);
52 return res;
53 }
54
55 static ssize_t blk_mq_sysfs_store(struct kobject *kobj, struct attribute *attr,
56 const char *page, size_t length)
57 {
58 struct blk_mq_ctx_sysfs_entry *entry;
59 struct blk_mq_ctx *ctx;
60 struct request_queue *q;
61 ssize_t res;
62
63 entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
64 ctx = container_of(kobj, struct blk_mq_ctx, kobj);
65 q = ctx->queue;
66
67 if (!entry->store)
68 return -EIO;
69
70 res = -ENOENT;
71 mutex_lock(&q->sysfs_lock);
72 if (!blk_queue_dying(q))
73 res = entry->store(ctx, page, length);
74 mutex_unlock(&q->sysfs_lock);
75 return res;
76 }
77
78 static ssize_t blk_mq_hw_sysfs_show(struct kobject *kobj,
79 struct attribute *attr, char *page)
80 {
81 struct blk_mq_hw_ctx_sysfs_entry *entry;
82 struct blk_mq_hw_ctx *hctx;
83 struct request_queue *q;
84 ssize_t res;
85
86 entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
87 hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
88 q = hctx->queue;
89
90 if (!entry->show)
91 return -EIO;
92
93 res = -ENOENT;
94 mutex_lock(&q->sysfs_lock);
95 if (!blk_queue_dying(q))
96 res = entry->show(hctx, page);
97 mutex_unlock(&q->sysfs_lock);
98 return res;
99 }
100
101 static ssize_t blk_mq_hw_sysfs_store(struct kobject *kobj,
102 struct attribute *attr, const char *page,
103 size_t length)
104 {
105 struct blk_mq_hw_ctx_sysfs_entry *entry;
106 struct blk_mq_hw_ctx *hctx;
107 struct request_queue *q;
108 ssize_t res;
109
110 entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
111 hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
112 q = hctx->queue;
113
114 if (!entry->store)
115 return -EIO;
116
117 res = -ENOENT;
118 mutex_lock(&q->sysfs_lock);
119 if (!blk_queue_dying(q))
120 res = entry->store(hctx, page, length);
121 mutex_unlock(&q->sysfs_lock);
122 return res;
123 }
124
125 static ssize_t blk_mq_sysfs_dispatched_show(struct blk_mq_ctx *ctx, char *page)
126 {
127 return sprintf(page, "%lu %lu\n", ctx->rq_dispatched[1],
128 ctx->rq_dispatched[0]);
129 }
130
131 static ssize_t blk_mq_sysfs_merged_show(struct blk_mq_ctx *ctx, char *page)
132 {
133 return sprintf(page, "%lu\n", ctx->rq_merged);
134 }
135
136 static ssize_t blk_mq_sysfs_completed_show(struct blk_mq_ctx *ctx, char *page)
137 {
138 return sprintf(page, "%lu %lu\n", ctx->rq_completed[1],
139 ctx->rq_completed[0]);
140 }
141
142 static ssize_t sysfs_list_show(char *page, struct list_head *list, char *msg)
143 {
144 struct request *rq;
145 int len = snprintf(page, PAGE_SIZE - 1, "%s:\n", msg);
146
147 list_for_each_entry(rq, list, queuelist) {
148 const int rq_len = 2 * sizeof(rq) + 2;
149
150 /* if the output will be truncated */
151 if (PAGE_SIZE - 1 < len + rq_len) {
152 /* backspacing if it can't hold '\t...\n' */
153 if (PAGE_SIZE - 1 < len + 5)
154 len -= rq_len;
155 len += snprintf(page + len, PAGE_SIZE - 1 - len,
156 "\t...\n");
157 break;
158 }
159 len += snprintf(page + len, PAGE_SIZE - 1 - len,
160 "\t%p\n", rq);
161 }
162
163 return len;
164 }
165
166 static ssize_t blk_mq_sysfs_rq_list_show(struct blk_mq_ctx *ctx, char *page)
167 {
168 ssize_t ret;
169
170 spin_lock(&ctx->lock);
171 ret = sysfs_list_show(page, &ctx->rq_list, "CTX pending");
172 spin_unlock(&ctx->lock);
173
174 return ret;
175 }
176
177 static ssize_t blk_mq_hw_sysfs_poll_show(struct blk_mq_hw_ctx *hctx, char *page)
178 {
179 return sprintf(page, "considered=%lu, invoked=%lu, success=%lu\n",
180 hctx->poll_considered, hctx->poll_invoked,
181 hctx->poll_success);
182 }
183
184 static ssize_t blk_mq_hw_sysfs_poll_store(struct blk_mq_hw_ctx *hctx,
185 const char *page, size_t size)
186 {
187 hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
188
189 return size;
190 }
191
192 static ssize_t blk_mq_hw_sysfs_queued_show(struct blk_mq_hw_ctx *hctx,
193 char *page)
194 {
195 return sprintf(page, "%lu\n", hctx->queued);
196 }
197
198 static ssize_t blk_mq_hw_sysfs_run_show(struct blk_mq_hw_ctx *hctx, char *page)
199 {
200 return sprintf(page, "%lu\n", hctx->run);
201 }
202
203 static ssize_t blk_mq_hw_sysfs_dispatched_show(struct blk_mq_hw_ctx *hctx,
204 char *page)
205 {
206 char *start_page = page;
207 int i;
208
209 page += sprintf(page, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
210
211 for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
212 unsigned int d = 1U << (i - 1);
213
214 page += sprintf(page, "%8u\t%lu\n", d, hctx->dispatched[i]);
215 }
216
217 page += sprintf(page, "%8u+\t%lu\n", 1U << (i - 1),
218 hctx->dispatched[i]);
219 return page - start_page;
220 }
221
222 static ssize_t blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx *hctx,
223 char *page)
224 {
225 ssize_t ret;
226
227 spin_lock(&hctx->lock);
228 ret = sysfs_list_show(page, &hctx->dispatch, "HCTX pending");
229 spin_unlock(&hctx->lock);
230
231 return ret;
232 }
233
234 static ssize_t blk_mq_hw_sysfs_sched_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
235 {
236 if (hctx->sched_tags)
237 return blk_mq_tag_sysfs_show(hctx->sched_tags, page);
238
239 return 0;
240 }
241
242 static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
243 {
244 return blk_mq_tag_sysfs_show(hctx->tags, page);
245 }
246
247 static ssize_t blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx *hctx, char *page)
248 {
249 return sprintf(page, "%u\n", atomic_read(&hctx->nr_active));
250 }
251
252 static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
253 {
254 unsigned int i, first = 1;
255 ssize_t ret = 0;
256
257 for_each_cpu(i, hctx->cpumask) {
258 if (first)
259 ret += sprintf(ret + page, "%u", i);
260 else
261 ret += sprintf(ret + page, ", %u", i);
262
263 first = 0;
264 }
265
266 ret += sprintf(ret + page, "\n");
267 return ret;
268 }
269
270 static void blk_mq_stat_clear(struct blk_mq_hw_ctx *hctx)
271 {
272 struct blk_mq_ctx *ctx;
273 unsigned int i;
274
275 hctx_for_each_ctx(hctx, ctx, i) {
276 blk_stat_init(&ctx->stat[BLK_STAT_READ]);
277 blk_stat_init(&ctx->stat[BLK_STAT_WRITE]);
278 }
279 }
280
281 static ssize_t blk_mq_hw_sysfs_stat_store(struct blk_mq_hw_ctx *hctx,
282 const char *page, size_t count)
283 {
284 blk_mq_stat_clear(hctx);
285 return count;
286 }
287
288 static ssize_t print_stat(char *page, struct blk_rq_stat *stat, const char *pre)
289 {
290 return sprintf(page, "%s samples=%llu, mean=%lld, min=%lld, max=%lld\n",
291 pre, (long long) stat->nr_samples,
292 (long long) stat->mean, (long long) stat->min,
293 (long long) stat->max);
294 }
295
296 static ssize_t blk_mq_hw_sysfs_stat_show(struct blk_mq_hw_ctx *hctx, char *page)
297 {
298 struct blk_rq_stat stat[2];
299 ssize_t ret;
300
301 blk_stat_init(&stat[BLK_STAT_READ]);
302 blk_stat_init(&stat[BLK_STAT_WRITE]);
303
304 blk_hctx_stat_get(hctx, stat);
305
306 ret = print_stat(page, &stat[BLK_STAT_READ], "read :");
307 ret += print_stat(page + ret, &stat[BLK_STAT_WRITE], "write:");
308 return ret;
309 }
310
311 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = {
312 .attr = {.name = "dispatched", .mode = S_IRUGO },
313 .show = blk_mq_sysfs_dispatched_show,
314 };
315 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = {
316 .attr = {.name = "merged", .mode = S_IRUGO },
317 .show = blk_mq_sysfs_merged_show,
318 };
319 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = {
320 .attr = {.name = "completed", .mode = S_IRUGO },
321 .show = blk_mq_sysfs_completed_show,
322 };
323 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = {
324 .attr = {.name = "rq_list", .mode = S_IRUGO },
325 .show = blk_mq_sysfs_rq_list_show,
326 };
327
328 static struct attribute *default_ctx_attrs[] = {
329 &blk_mq_sysfs_dispatched.attr,
330 &blk_mq_sysfs_merged.attr,
331 &blk_mq_sysfs_completed.attr,
332 &blk_mq_sysfs_rq_list.attr,
333 NULL,
334 };
335
336 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = {
337 .attr = {.name = "queued", .mode = S_IRUGO },
338 .show = blk_mq_hw_sysfs_queued_show,
339 };
340 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = {
341 .attr = {.name = "run", .mode = S_IRUGO },
342 .show = blk_mq_hw_sysfs_run_show,
343 };
344 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = {
345 .attr = {.name = "dispatched", .mode = S_IRUGO },
346 .show = blk_mq_hw_sysfs_dispatched_show,
347 };
348 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_active = {
349 .attr = {.name = "active", .mode = S_IRUGO },
350 .show = blk_mq_hw_sysfs_active_show,
351 };
352 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = {
353 .attr = {.name = "pending", .mode = S_IRUGO },
354 .show = blk_mq_hw_sysfs_rq_list_show,
355 };
356 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_sched_tags = {
357 .attr = {.name = "sched_tags", .mode = S_IRUGO },
358 .show = blk_mq_hw_sysfs_sched_tags_show,
359 };
360 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = {
361 .attr = {.name = "tags", .mode = S_IRUGO },
362 .show = blk_mq_hw_sysfs_tags_show,
363 };
364 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
365 .attr = {.name = "cpu_list", .mode = S_IRUGO },
366 .show = blk_mq_hw_sysfs_cpus_show,
367 };
368 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_poll = {
369 .attr = {.name = "io_poll", .mode = S_IWUSR | S_IRUGO },
370 .show = blk_mq_hw_sysfs_poll_show,
371 .store = blk_mq_hw_sysfs_poll_store,
372 };
373 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_stat = {
374 .attr = {.name = "stats", .mode = S_IRUGO | S_IWUSR },
375 .show = blk_mq_hw_sysfs_stat_show,
376 .store = blk_mq_hw_sysfs_stat_store,
377 };
378
379 static struct attribute *default_hw_ctx_attrs[] = {
380 &blk_mq_hw_sysfs_queued.attr,
381 &blk_mq_hw_sysfs_run.attr,
382 &blk_mq_hw_sysfs_dispatched.attr,
383 &blk_mq_hw_sysfs_pending.attr,
384 &blk_mq_hw_sysfs_tags.attr,
385 &blk_mq_hw_sysfs_sched_tags.attr,
386 &blk_mq_hw_sysfs_cpus.attr,
387 &blk_mq_hw_sysfs_active.attr,
388 &blk_mq_hw_sysfs_poll.attr,
389 &blk_mq_hw_sysfs_stat.attr,
390 NULL,
391 };
392
393 static const struct sysfs_ops blk_mq_sysfs_ops = {
394 .show = blk_mq_sysfs_show,
395 .store = blk_mq_sysfs_store,
396 };
397
398 static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
399 .show = blk_mq_hw_sysfs_show,
400 .store = blk_mq_hw_sysfs_store,
401 };
402
403 static struct kobj_type blk_mq_ktype = {
404 .sysfs_ops = &blk_mq_sysfs_ops,
405 .release = blk_mq_sysfs_release,
406 };
407
408 static struct kobj_type blk_mq_ctx_ktype = {
409 .sysfs_ops = &blk_mq_sysfs_ops,
410 .default_attrs = default_ctx_attrs,
411 .release = blk_mq_sysfs_release,
412 };
413
414 static struct kobj_type blk_mq_hw_ktype = {
415 .sysfs_ops = &blk_mq_hw_sysfs_ops,
416 .default_attrs = default_hw_ctx_attrs,
417 .release = blk_mq_sysfs_release,
418 };
419
420 static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx)
421 {
422 struct blk_mq_ctx *ctx;
423 int i;
424
425 if (!hctx->nr_ctx)
426 return;
427
428 hctx_for_each_ctx(hctx, ctx, i)
429 kobject_del(&ctx->kobj);
430
431 kobject_del(&hctx->kobj);
432 }
433
434 static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx)
435 {
436 struct request_queue *q = hctx->queue;
437 struct blk_mq_ctx *ctx;
438 int i, ret;
439
440 if (!hctx->nr_ctx)
441 return 0;
442
443 ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", hctx->queue_num);
444 if (ret)
445 return ret;
446
447 hctx_for_each_ctx(hctx, ctx, i) {
448 ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
449 if (ret)
450 break;
451 }
452
453 return ret;
454 }
455
456 static void __blk_mq_unregister_dev(struct device *dev, struct request_queue *q)
457 {
458 struct blk_mq_hw_ctx *hctx;
459 struct blk_mq_ctx *ctx;
460 int i, j;
461
462 queue_for_each_hw_ctx(q, hctx, i) {
463 blk_mq_unregister_hctx(hctx);
464
465 hctx_for_each_ctx(hctx, ctx, j)
466 kobject_put(&ctx->kobj);
467
468 kobject_put(&hctx->kobj);
469 }
470
471 kobject_uevent(&q->mq_kobj, KOBJ_REMOVE);
472 kobject_del(&q->mq_kobj);
473 kobject_put(&q->mq_kobj);
474
475 kobject_put(&dev->kobj);
476
477 q->mq_sysfs_init_done = false;
478 }
479
480 void blk_mq_unregister_dev(struct device *dev, struct request_queue *q)
481 {
482 blk_mq_disable_hotplug();
483 __blk_mq_unregister_dev(dev, q);
484 blk_mq_enable_hotplug();
485 }
486
487 void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx)
488 {
489 kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
490 }
491
492 static void blk_mq_sysfs_init(struct request_queue *q)
493 {
494 struct blk_mq_ctx *ctx;
495 int cpu;
496
497 kobject_init(&q->mq_kobj, &blk_mq_ktype);
498
499 for_each_possible_cpu(cpu) {
500 ctx = per_cpu_ptr(q->queue_ctx, cpu);
501 kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
502 }
503 }
504
505 int blk_mq_register_dev(struct device *dev, struct request_queue *q)
506 {
507 struct blk_mq_hw_ctx *hctx;
508 int ret, i;
509
510 blk_mq_disable_hotplug();
511
512 blk_mq_sysfs_init(q);
513
514 ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
515 if (ret < 0)
516 goto out;
517
518 kobject_uevent(&q->mq_kobj, KOBJ_ADD);
519
520 queue_for_each_hw_ctx(q, hctx, i) {
521 ret = blk_mq_register_hctx(hctx);
522 if (ret)
523 break;
524 }
525
526 if (ret)
527 __blk_mq_unregister_dev(dev, q);
528 else
529 q->mq_sysfs_init_done = true;
530 out:
531 blk_mq_enable_hotplug();
532
533 return ret;
534 }
535 EXPORT_SYMBOL_GPL(blk_mq_register_dev);
536
537 void blk_mq_sysfs_unregister(struct request_queue *q)
538 {
539 struct blk_mq_hw_ctx *hctx;
540 int i;
541
542 if (!q->mq_sysfs_init_done)
543 return;
544
545 queue_for_each_hw_ctx(q, hctx, i)
546 blk_mq_unregister_hctx(hctx);
547 }
548
549 int blk_mq_sysfs_register(struct request_queue *q)
550 {
551 struct blk_mq_hw_ctx *hctx;
552 int i, ret = 0;
553
554 if (!q->mq_sysfs_init_done)
555 return ret;
556
557 queue_for_each_hw_ctx(q, hctx, i) {
558 ret = blk_mq_register_hctx(hctx);
559 if (ret)
560 break;
561 }
562
563 return ret;
564 }