blob: 8c0894e0713b99b66ae6840f27a85ce776dd77ad
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_tags_show(struct blk_mq_hw_ctx *hctx, char *page) |
235 | { |
236 | return blk_mq_tag_sysfs_show(hctx->tags, page); |
237 | } |
238 | |
239 | static ssize_t blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx *hctx, char *page) |
240 | { |
241 | return sprintf(page, "%u\n", atomic_read(&hctx->nr_active)); |
242 | } |
243 | |
244 | static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page) |
245 | { |
246 | unsigned int i, first = 1; |
247 | ssize_t ret = 0; |
248 | |
249 | for_each_cpu(i, hctx->cpumask) { |
250 | if (first) |
251 | ret += sprintf(ret + page, "%u", i); |
252 | else |
253 | ret += sprintf(ret + page, ", %u", i); |
254 | |
255 | first = 0; |
256 | } |
257 | |
258 | ret += sprintf(ret + page, "\n"); |
259 | return ret; |
260 | } |
261 | |
262 | static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = { |
263 | .attr = {.name = "dispatched", .mode = S_IRUGO }, |
264 | .show = blk_mq_sysfs_dispatched_show, |
265 | }; |
266 | static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = { |
267 | .attr = {.name = "merged", .mode = S_IRUGO }, |
268 | .show = blk_mq_sysfs_merged_show, |
269 | }; |
270 | static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = { |
271 | .attr = {.name = "completed", .mode = S_IRUGO }, |
272 | .show = blk_mq_sysfs_completed_show, |
273 | }; |
274 | static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = { |
275 | .attr = {.name = "rq_list", .mode = S_IRUGO }, |
276 | .show = blk_mq_sysfs_rq_list_show, |
277 | }; |
278 | |
279 | static struct attribute *default_ctx_attrs[] = { |
280 | &blk_mq_sysfs_dispatched.attr, |
281 | &blk_mq_sysfs_merged.attr, |
282 | &blk_mq_sysfs_completed.attr, |
283 | &blk_mq_sysfs_rq_list.attr, |
284 | NULL, |
285 | }; |
286 | |
287 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = { |
288 | .attr = {.name = "queued", .mode = S_IRUGO }, |
289 | .show = blk_mq_hw_sysfs_queued_show, |
290 | }; |
291 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = { |
292 | .attr = {.name = "run", .mode = S_IRUGO }, |
293 | .show = blk_mq_hw_sysfs_run_show, |
294 | }; |
295 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = { |
296 | .attr = {.name = "dispatched", .mode = S_IRUGO }, |
297 | .show = blk_mq_hw_sysfs_dispatched_show, |
298 | }; |
299 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_active = { |
300 | .attr = {.name = "active", .mode = S_IRUGO }, |
301 | .show = blk_mq_hw_sysfs_active_show, |
302 | }; |
303 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = { |
304 | .attr = {.name = "pending", .mode = S_IRUGO }, |
305 | .show = blk_mq_hw_sysfs_rq_list_show, |
306 | }; |
307 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = { |
308 | .attr = {.name = "tags", .mode = S_IRUGO }, |
309 | .show = blk_mq_hw_sysfs_tags_show, |
310 | }; |
311 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = { |
312 | .attr = {.name = "cpu_list", .mode = S_IRUGO }, |
313 | .show = blk_mq_hw_sysfs_cpus_show, |
314 | }; |
315 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_poll = { |
316 | .attr = {.name = "io_poll", .mode = S_IWUSR | S_IRUGO }, |
317 | .show = blk_mq_hw_sysfs_poll_show, |
318 | .store = blk_mq_hw_sysfs_poll_store, |
319 | }; |
320 | |
321 | static struct attribute *default_hw_ctx_attrs[] = { |
322 | &blk_mq_hw_sysfs_queued.attr, |
323 | &blk_mq_hw_sysfs_run.attr, |
324 | &blk_mq_hw_sysfs_dispatched.attr, |
325 | &blk_mq_hw_sysfs_pending.attr, |
326 | &blk_mq_hw_sysfs_tags.attr, |
327 | &blk_mq_hw_sysfs_cpus.attr, |
328 | &blk_mq_hw_sysfs_active.attr, |
329 | &blk_mq_hw_sysfs_poll.attr, |
330 | NULL, |
331 | }; |
332 | |
333 | static const struct sysfs_ops blk_mq_sysfs_ops = { |
334 | .show = blk_mq_sysfs_show, |
335 | .store = blk_mq_sysfs_store, |
336 | }; |
337 | |
338 | static const struct sysfs_ops blk_mq_hw_sysfs_ops = { |
339 | .show = blk_mq_hw_sysfs_show, |
340 | .store = blk_mq_hw_sysfs_store, |
341 | }; |
342 | |
343 | static struct kobj_type blk_mq_ktype = { |
344 | .sysfs_ops = &blk_mq_sysfs_ops, |
345 | .release = blk_mq_sysfs_release, |
346 | }; |
347 | |
348 | static struct kobj_type blk_mq_ctx_ktype = { |
349 | .sysfs_ops = &blk_mq_sysfs_ops, |
350 | .default_attrs = default_ctx_attrs, |
351 | .release = blk_mq_sysfs_release, |
352 | }; |
353 | |
354 | static struct kobj_type blk_mq_hw_ktype = { |
355 | .sysfs_ops = &blk_mq_hw_sysfs_ops, |
356 | .default_attrs = default_hw_ctx_attrs, |
357 | .release = blk_mq_sysfs_release, |
358 | }; |
359 | |
360 | static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx) |
361 | { |
362 | struct blk_mq_ctx *ctx; |
363 | int i; |
364 | |
365 | if (!hctx->nr_ctx) |
366 | return; |
367 | |
368 | hctx_for_each_ctx(hctx, ctx, i) |
369 | kobject_del(&ctx->kobj); |
370 | |
371 | kobject_del(&hctx->kobj); |
372 | } |
373 | |
374 | static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx) |
375 | { |
376 | struct request_queue *q = hctx->queue; |
377 | struct blk_mq_ctx *ctx; |
378 | int i, ret; |
379 | |
380 | if (!hctx->nr_ctx) |
381 | return 0; |
382 | |
383 | ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", hctx->queue_num); |
384 | if (ret) |
385 | return ret; |
386 | |
387 | hctx_for_each_ctx(hctx, ctx, i) { |
388 | ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu); |
389 | if (ret) |
390 | break; |
391 | } |
392 | |
393 | return ret; |
394 | } |
395 | |
396 | static void __blk_mq_unregister_dev(struct device *dev, struct request_queue *q) |
397 | { |
398 | struct blk_mq_hw_ctx *hctx; |
399 | struct blk_mq_ctx *ctx; |
400 | int i, j; |
401 | |
402 | queue_for_each_hw_ctx(q, hctx, i) { |
403 | blk_mq_unregister_hctx(hctx); |
404 | |
405 | hctx_for_each_ctx(hctx, ctx, j) |
406 | kobject_put(&ctx->kobj); |
407 | |
408 | kobject_put(&hctx->kobj); |
409 | } |
410 | |
411 | kobject_uevent(&q->mq_kobj, KOBJ_REMOVE); |
412 | kobject_del(&q->mq_kobj); |
413 | kobject_put(&q->mq_kobj); |
414 | |
415 | kobject_put(&dev->kobj); |
416 | |
417 | q->mq_sysfs_init_done = false; |
418 | } |
419 | |
420 | void blk_mq_unregister_dev(struct device *dev, struct request_queue *q) |
421 | { |
422 | blk_mq_disable_hotplug(); |
423 | __blk_mq_unregister_dev(dev, q); |
424 | blk_mq_enable_hotplug(); |
425 | } |
426 | |
427 | void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx) |
428 | { |
429 | kobject_init(&hctx->kobj, &blk_mq_hw_ktype); |
430 | } |
431 | |
432 | void blk_mq_sysfs_init(struct request_queue *q) |
433 | { |
434 | struct blk_mq_ctx *ctx; |
435 | int cpu; |
436 | |
437 | kobject_init(&q->mq_kobj, &blk_mq_ktype); |
438 | |
439 | for_each_possible_cpu(cpu) { |
440 | ctx = per_cpu_ptr(q->queue_ctx, cpu); |
441 | kobject_init(&ctx->kobj, &blk_mq_ctx_ktype); |
442 | } |
443 | } |
444 | |
445 | int blk_mq_register_dev(struct device *dev, struct request_queue *q) |
446 | { |
447 | struct blk_mq_hw_ctx *hctx; |
448 | int ret, i; |
449 | |
450 | blk_mq_disable_hotplug(); |
451 | |
452 | ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq"); |
453 | if (ret < 0) |
454 | goto out; |
455 | |
456 | kobject_uevent(&q->mq_kobj, KOBJ_ADD); |
457 | |
458 | queue_for_each_hw_ctx(q, hctx, i) { |
459 | ret = blk_mq_register_hctx(hctx); |
460 | if (ret) |
461 | break; |
462 | } |
463 | |
464 | if (ret) |
465 | __blk_mq_unregister_dev(dev, q); |
466 | else |
467 | q->mq_sysfs_init_done = true; |
468 | out: |
469 | blk_mq_enable_hotplug(); |
470 | |
471 | return ret; |
472 | } |
473 | EXPORT_SYMBOL_GPL(blk_mq_register_dev); |
474 | |
475 | void blk_mq_sysfs_unregister(struct request_queue *q) |
476 | { |
477 | struct blk_mq_hw_ctx *hctx; |
478 | int i; |
479 | |
480 | if (!q->mq_sysfs_init_done) |
481 | return; |
482 | |
483 | queue_for_each_hw_ctx(q, hctx, i) |
484 | blk_mq_unregister_hctx(hctx); |
485 | } |
486 | |
487 | int blk_mq_sysfs_register(struct request_queue *q) |
488 | { |
489 | struct blk_mq_hw_ctx *hctx; |
490 | int i, ret = 0; |
491 | |
492 | if (!q->mq_sysfs_init_done) |
493 | return ret; |
494 | |
495 | queue_for_each_hw_ctx(q, hctx, i) { |
496 | ret = blk_mq_register_hctx(hctx); |
497 | if (ret) |
498 | break; |
499 | } |
500 | |
501 | return ret; |
502 | } |
503 |