Commit | Line | Data |
---|---|---|
320ae51f JA |
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 | { | |
320ae51f | 144 | struct request *rq; |
596f5aad ML |
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; | |
320ae51f JA |
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 | ||
05229bee JA |
177 | static ssize_t blk_mq_hw_sysfs_poll_show(struct blk_mq_hw_ctx *hctx, char *page) |
178 | { | |
6e219353 SB |
179 | return sprintf(page, "considered=%lu, invoked=%lu, success=%lu\n", |
180 | hctx->poll_considered, hctx->poll_invoked, | |
181 | hctx->poll_success); | |
05229bee JA |
182 | } |
183 | ||
d21ea4bc SB |
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 | ||
320ae51f JA |
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; i++) { | |
212 | unsigned long d = 1U << (i - 1); | |
213 | ||
214 | page += sprintf(page, "%8lu\t%lu\n", d, hctx->dispatched[i]); | |
215 | } | |
216 | ||
217 | return page - start_page; | |
218 | } | |
219 | ||
220 | static ssize_t blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx *hctx, | |
221 | char *page) | |
222 | { | |
223 | ssize_t ret; | |
224 | ||
225 | spin_lock(&hctx->lock); | |
226 | ret = sysfs_list_show(page, &hctx->dispatch, "HCTX pending"); | |
227 | spin_unlock(&hctx->lock); | |
228 | ||
229 | return ret; | |
230 | } | |
231 | ||
320ae51f JA |
232 | static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page) |
233 | { | |
234 | return blk_mq_tag_sysfs_show(hctx->tags, page); | |
235 | } | |
236 | ||
0d2602ca JA |
237 | static ssize_t blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx *hctx, char *page) |
238 | { | |
239 | return sprintf(page, "%u\n", atomic_read(&hctx->nr_active)); | |
240 | } | |
241 | ||
676141e4 JA |
242 | static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page) |
243 | { | |
cb2da43e | 244 | unsigned int i, first = 1; |
676141e4 JA |
245 | ssize_t ret = 0; |
246 | ||
cb2da43e | 247 | for_each_cpu(i, hctx->cpumask) { |
676141e4 JA |
248 | if (first) |
249 | ret += sprintf(ret + page, "%u", i); | |
250 | else | |
251 | ret += sprintf(ret + page, ", %u", i); | |
252 | ||
253 | first = 0; | |
254 | } | |
255 | ||
676141e4 JA |
256 | ret += sprintf(ret + page, "\n"); |
257 | return ret; | |
258 | } | |
259 | ||
320ae51f JA |
260 | static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = { |
261 | .attr = {.name = "dispatched", .mode = S_IRUGO }, | |
262 | .show = blk_mq_sysfs_dispatched_show, | |
263 | }; | |
264 | static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = { | |
265 | .attr = {.name = "merged", .mode = S_IRUGO }, | |
266 | .show = blk_mq_sysfs_merged_show, | |
267 | }; | |
268 | static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = { | |
269 | .attr = {.name = "completed", .mode = S_IRUGO }, | |
270 | .show = blk_mq_sysfs_completed_show, | |
271 | }; | |
272 | static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = { | |
273 | .attr = {.name = "rq_list", .mode = S_IRUGO }, | |
274 | .show = blk_mq_sysfs_rq_list_show, | |
275 | }; | |
276 | ||
277 | static struct attribute *default_ctx_attrs[] = { | |
278 | &blk_mq_sysfs_dispatched.attr, | |
279 | &blk_mq_sysfs_merged.attr, | |
280 | &blk_mq_sysfs_completed.attr, | |
281 | &blk_mq_sysfs_rq_list.attr, | |
282 | NULL, | |
283 | }; | |
284 | ||
285 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = { | |
286 | .attr = {.name = "queued", .mode = S_IRUGO }, | |
287 | .show = blk_mq_hw_sysfs_queued_show, | |
288 | }; | |
289 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = { | |
290 | .attr = {.name = "run", .mode = S_IRUGO }, | |
291 | .show = blk_mq_hw_sysfs_run_show, | |
292 | }; | |
293 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = { | |
294 | .attr = {.name = "dispatched", .mode = S_IRUGO }, | |
295 | .show = blk_mq_hw_sysfs_dispatched_show, | |
296 | }; | |
0d2602ca JA |
297 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_active = { |
298 | .attr = {.name = "active", .mode = S_IRUGO }, | |
299 | .show = blk_mq_hw_sysfs_active_show, | |
300 | }; | |
320ae51f JA |
301 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = { |
302 | .attr = {.name = "pending", .mode = S_IRUGO }, | |
303 | .show = blk_mq_hw_sysfs_rq_list_show, | |
304 | }; | |
320ae51f JA |
305 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = { |
306 | .attr = {.name = "tags", .mode = S_IRUGO }, | |
307 | .show = blk_mq_hw_sysfs_tags_show, | |
308 | }; | |
676141e4 JA |
309 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = { |
310 | .attr = {.name = "cpu_list", .mode = S_IRUGO }, | |
311 | .show = blk_mq_hw_sysfs_cpus_show, | |
312 | }; | |
05229bee | 313 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_poll = { |
d21ea4bc | 314 | .attr = {.name = "io_poll", .mode = S_IWUSR | S_IRUGO }, |
05229bee | 315 | .show = blk_mq_hw_sysfs_poll_show, |
d21ea4bc | 316 | .store = blk_mq_hw_sysfs_poll_store, |
05229bee | 317 | }; |
320ae51f JA |
318 | |
319 | static struct attribute *default_hw_ctx_attrs[] = { | |
320 | &blk_mq_hw_sysfs_queued.attr, | |
321 | &blk_mq_hw_sysfs_run.attr, | |
322 | &blk_mq_hw_sysfs_dispatched.attr, | |
323 | &blk_mq_hw_sysfs_pending.attr, | |
320ae51f | 324 | &blk_mq_hw_sysfs_tags.attr, |
676141e4 | 325 | &blk_mq_hw_sysfs_cpus.attr, |
0d2602ca | 326 | &blk_mq_hw_sysfs_active.attr, |
05229bee | 327 | &blk_mq_hw_sysfs_poll.attr, |
320ae51f JA |
328 | NULL, |
329 | }; | |
330 | ||
331 | static const struct sysfs_ops blk_mq_sysfs_ops = { | |
332 | .show = blk_mq_sysfs_show, | |
333 | .store = blk_mq_sysfs_store, | |
334 | }; | |
335 | ||
336 | static const struct sysfs_ops blk_mq_hw_sysfs_ops = { | |
337 | .show = blk_mq_hw_sysfs_show, | |
338 | .store = blk_mq_hw_sysfs_store, | |
339 | }; | |
340 | ||
341 | static struct kobj_type blk_mq_ktype = { | |
342 | .sysfs_ops = &blk_mq_sysfs_ops, | |
343 | .release = blk_mq_sysfs_release, | |
344 | }; | |
345 | ||
346 | static struct kobj_type blk_mq_ctx_ktype = { | |
347 | .sysfs_ops = &blk_mq_sysfs_ops, | |
348 | .default_attrs = default_ctx_attrs, | |
74170118 | 349 | .release = blk_mq_sysfs_release, |
320ae51f JA |
350 | }; |
351 | ||
352 | static struct kobj_type blk_mq_hw_ktype = { | |
353 | .sysfs_ops = &blk_mq_hw_sysfs_ops, | |
354 | .default_attrs = default_hw_ctx_attrs, | |
74170118 | 355 | .release = blk_mq_sysfs_release, |
320ae51f JA |
356 | }; |
357 | ||
ee3c5db0 | 358 | static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx) |
67aec14c JA |
359 | { |
360 | struct blk_mq_ctx *ctx; | |
361 | int i; | |
362 | ||
4593fdbe | 363 | if (!hctx->nr_ctx) |
67aec14c JA |
364 | return; |
365 | ||
366 | hctx_for_each_ctx(hctx, ctx, i) | |
367 | kobject_del(&ctx->kobj); | |
368 | ||
369 | kobject_del(&hctx->kobj); | |
370 | } | |
371 | ||
ee3c5db0 | 372 | static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx) |
67aec14c JA |
373 | { |
374 | struct request_queue *q = hctx->queue; | |
375 | struct blk_mq_ctx *ctx; | |
376 | int i, ret; | |
377 | ||
4593fdbe | 378 | if (!hctx->nr_ctx) |
67aec14c JA |
379 | return 0; |
380 | ||
381 | ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", hctx->queue_num); | |
382 | if (ret) | |
383 | return ret; | |
384 | ||
385 | hctx_for_each_ctx(hctx, ctx, i) { | |
386 | ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu); | |
387 | if (ret) | |
388 | break; | |
389 | } | |
390 | ||
391 | return ret; | |
392 | } | |
393 | ||
c0f3fd2b | 394 | static void __blk_mq_unregister_disk(struct gendisk *disk) |
320ae51f JA |
395 | { |
396 | struct request_queue *q = disk->queue; | |
85157366 AV |
397 | struct blk_mq_hw_ctx *hctx; |
398 | struct blk_mq_ctx *ctx; | |
399 | int i, j; | |
400 | ||
401 | queue_for_each_hw_ctx(q, hctx, i) { | |
67aec14c JA |
402 | blk_mq_unregister_hctx(hctx); |
403 | ||
404 | hctx_for_each_ctx(hctx, ctx, j) | |
85157366 | 405 | kobject_put(&ctx->kobj); |
67aec14c | 406 | |
85157366 AV |
407 | kobject_put(&hctx->kobj); |
408 | } | |
320ae51f JA |
409 | |
410 | kobject_uevent(&q->mq_kobj, KOBJ_REMOVE); | |
411 | kobject_del(&q->mq_kobj); | |
85157366 | 412 | kobject_put(&q->mq_kobj); |
320ae51f JA |
413 | |
414 | kobject_put(&disk_to_dev(disk)->kobj); | |
4593fdbe AM |
415 | |
416 | q->mq_sysfs_init_done = false; | |
c0f3fd2b JA |
417 | } |
418 | ||
419 | void blk_mq_unregister_disk(struct gendisk *disk) | |
420 | { | |
421 | blk_mq_disable_hotplug(); | |
422 | __blk_mq_unregister_disk(disk); | |
4593fdbe | 423 | blk_mq_enable_hotplug(); |
320ae51f JA |
424 | } |
425 | ||
868f2f0b KB |
426 | void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx) |
427 | { | |
428 | kobject_init(&hctx->kobj, &blk_mq_hw_ktype); | |
429 | } | |
430 | ||
67aec14c JA |
431 | static void blk_mq_sysfs_init(struct request_queue *q) |
432 | { | |
67aec14c | 433 | struct blk_mq_ctx *ctx; |
897bb0c7 | 434 | int cpu; |
67aec14c JA |
435 | |
436 | kobject_init(&q->mq_kobj, &blk_mq_ktype); | |
437 | ||
897bb0c7 TG |
438 | for_each_possible_cpu(cpu) { |
439 | ctx = per_cpu_ptr(q->queue_ctx, cpu); | |
06a41a99 | 440 | kobject_init(&ctx->kobj, &blk_mq_ctx_ktype); |
897bb0c7 | 441 | } |
67aec14c JA |
442 | } |
443 | ||
320ae51f JA |
444 | int blk_mq_register_disk(struct gendisk *disk) |
445 | { | |
446 | struct device *dev = disk_to_dev(disk); | |
447 | struct request_queue *q = disk->queue; | |
448 | struct blk_mq_hw_ctx *hctx; | |
67aec14c | 449 | int ret, i; |
320ae51f | 450 | |
4593fdbe AM |
451 | blk_mq_disable_hotplug(); |
452 | ||
67aec14c | 453 | blk_mq_sysfs_init(q); |
320ae51f JA |
454 | |
455 | ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq"); | |
456 | if (ret < 0) | |
4593fdbe | 457 | goto out; |
320ae51f JA |
458 | |
459 | kobject_uevent(&q->mq_kobj, KOBJ_ADD); | |
460 | ||
461 | queue_for_each_hw_ctx(q, hctx, i) { | |
67aec14c | 462 | ret = blk_mq_register_hctx(hctx); |
320ae51f JA |
463 | if (ret) |
464 | break; | |
320ae51f JA |
465 | } |
466 | ||
4593fdbe | 467 | if (ret) |
c0f3fd2b | 468 | __blk_mq_unregister_disk(disk); |
4593fdbe AM |
469 | else |
470 | q->mq_sysfs_init_done = true; | |
471 | out: | |
472 | blk_mq_enable_hotplug(); | |
320ae51f | 473 | |
4593fdbe | 474 | return ret; |
320ae51f | 475 | } |
b62c21b7 | 476 | EXPORT_SYMBOL_GPL(blk_mq_register_disk); |
67aec14c JA |
477 | |
478 | void blk_mq_sysfs_unregister(struct request_queue *q) | |
479 | { | |
480 | struct blk_mq_hw_ctx *hctx; | |
481 | int i; | |
482 | ||
4593fdbe AM |
483 | if (!q->mq_sysfs_init_done) |
484 | return; | |
485 | ||
67aec14c JA |
486 | queue_for_each_hw_ctx(q, hctx, i) |
487 | blk_mq_unregister_hctx(hctx); | |
488 | } | |
489 | ||
490 | int blk_mq_sysfs_register(struct request_queue *q) | |
491 | { | |
492 | struct blk_mq_hw_ctx *hctx; | |
493 | int i, ret = 0; | |
494 | ||
4593fdbe AM |
495 | if (!q->mq_sysfs_init_done) |
496 | return ret; | |
497 | ||
67aec14c JA |
498 | queue_for_each_hw_ctx(q, hctx, i) { |
499 | ret = blk_mq_register_hctx(hctx); | |
500 | if (ret) | |
501 | break; | |
502 | } | |
503 | ||
504 | return ret; | |
505 | } |