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
blk_mq_sysfs_release(struct kobject * kobj)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
blk_mq_sysfs_show(struct kobject * kobj,struct attribute * attr,char * page)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
blk_mq_sysfs_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)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
blk_mq_hw_sysfs_show(struct kobject * kobj,struct attribute * attr,char * page)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
blk_mq_hw_sysfs_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)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
blk_mq_sysfs_dispatched_show(struct blk_mq_ctx * ctx,char * page)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
blk_mq_sysfs_merged_show(struct blk_mq_ctx * ctx,char * page)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
blk_mq_sysfs_completed_show(struct blk_mq_ctx * ctx,char * page)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
sysfs_list_show(char * page,struct list_head * list,char * msg)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
blk_mq_sysfs_rq_list_show(struct blk_mq_ctx * ctx,char * page)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
blk_mq_hw_sysfs_poll_show(struct blk_mq_hw_ctx * hctx,char * page)177 static ssize_t blk_mq_hw_sysfs_poll_show(struct blk_mq_hw_ctx *hctx, char *page)
178 {
179 return sprintf(page, "invoked=%lu, success=%lu\n", hctx->poll_invoked, hctx->poll_success);
180 }
181
blk_mq_hw_sysfs_queued_show(struct blk_mq_hw_ctx * hctx,char * page)182 static ssize_t blk_mq_hw_sysfs_queued_show(struct blk_mq_hw_ctx *hctx,
183 char *page)
184 {
185 return sprintf(page, "%lu\n", hctx->queued);
186 }
187
blk_mq_hw_sysfs_run_show(struct blk_mq_hw_ctx * hctx,char * page)188 static ssize_t blk_mq_hw_sysfs_run_show(struct blk_mq_hw_ctx *hctx, char *page)
189 {
190 return sprintf(page, "%lu\n", hctx->run);
191 }
192
blk_mq_hw_sysfs_dispatched_show(struct blk_mq_hw_ctx * hctx,char * page)193 static ssize_t blk_mq_hw_sysfs_dispatched_show(struct blk_mq_hw_ctx *hctx,
194 char *page)
195 {
196 char *start_page = page;
197 int i;
198
199 page += sprintf(page, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
200
201 for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER; i++) {
202 unsigned long d = 1U << (i - 1);
203
204 page += sprintf(page, "%8lu\t%lu\n", d, hctx->dispatched[i]);
205 }
206
207 return page - start_page;
208 }
209
blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx * hctx,char * page)210 static ssize_t blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx *hctx,
211 char *page)
212 {
213 ssize_t ret;
214
215 spin_lock(&hctx->lock);
216 ret = sysfs_list_show(page, &hctx->dispatch, "HCTX pending");
217 spin_unlock(&hctx->lock);
218
219 return ret;
220 }
221
blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx * hctx,char * page)222 static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
223 {
224 return blk_mq_tag_sysfs_show(hctx->tags, page);
225 }
226
blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx * hctx,char * page)227 static ssize_t blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx *hctx, char *page)
228 {
229 return sprintf(page, "%u\n", atomic_read(&hctx->nr_active));
230 }
231
blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx * hctx,char * page)232 static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
233 {
234 const size_t size = PAGE_SIZE - 1;
235 unsigned int i, first = 1;
236 int ret = 0, pos = 0;
237
238 for_each_cpu(i, hctx->cpumask) {
239 if (first)
240 ret = snprintf(pos + page, size - pos, "%u", i);
241 else
242 ret = snprintf(pos + page, size - pos, ", %u", i);
243
244 if (ret >= size - pos)
245 break;
246
247 first = 0;
248 pos += ret;
249 }
250
251 ret = snprintf(pos + page, size + 1 - pos, "\n");
252 return pos + ret;
253 }
254
255 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = {
256 .attr = {.name = "dispatched", .mode = S_IRUGO },
257 .show = blk_mq_sysfs_dispatched_show,
258 };
259 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = {
260 .attr = {.name = "merged", .mode = S_IRUGO },
261 .show = blk_mq_sysfs_merged_show,
262 };
263 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = {
264 .attr = {.name = "completed", .mode = S_IRUGO },
265 .show = blk_mq_sysfs_completed_show,
266 };
267 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = {
268 .attr = {.name = "rq_list", .mode = S_IRUGO },
269 .show = blk_mq_sysfs_rq_list_show,
270 };
271
272 static struct attribute *default_ctx_attrs[] = {
273 &blk_mq_sysfs_dispatched.attr,
274 &blk_mq_sysfs_merged.attr,
275 &blk_mq_sysfs_completed.attr,
276 &blk_mq_sysfs_rq_list.attr,
277 NULL,
278 };
279
280 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = {
281 .attr = {.name = "queued", .mode = S_IRUGO },
282 .show = blk_mq_hw_sysfs_queued_show,
283 };
284 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = {
285 .attr = {.name = "run", .mode = S_IRUGO },
286 .show = blk_mq_hw_sysfs_run_show,
287 };
288 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = {
289 .attr = {.name = "dispatched", .mode = S_IRUGO },
290 .show = blk_mq_hw_sysfs_dispatched_show,
291 };
292 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_active = {
293 .attr = {.name = "active", .mode = S_IRUGO },
294 .show = blk_mq_hw_sysfs_active_show,
295 };
296 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = {
297 .attr = {.name = "pending", .mode = S_IRUGO },
298 .show = blk_mq_hw_sysfs_rq_list_show,
299 };
300 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = {
301 .attr = {.name = "tags", .mode = S_IRUGO },
302 .show = blk_mq_hw_sysfs_tags_show,
303 };
304 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
305 .attr = {.name = "cpu_list", .mode = S_IRUGO },
306 .show = blk_mq_hw_sysfs_cpus_show,
307 };
308 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_poll = {
309 .attr = {.name = "io_poll", .mode = S_IRUGO },
310 .show = blk_mq_hw_sysfs_poll_show,
311 };
312
313 static struct attribute *default_hw_ctx_attrs[] = {
314 &blk_mq_hw_sysfs_queued.attr,
315 &blk_mq_hw_sysfs_run.attr,
316 &blk_mq_hw_sysfs_dispatched.attr,
317 &blk_mq_hw_sysfs_pending.attr,
318 &blk_mq_hw_sysfs_tags.attr,
319 &blk_mq_hw_sysfs_cpus.attr,
320 &blk_mq_hw_sysfs_active.attr,
321 &blk_mq_hw_sysfs_poll.attr,
322 NULL,
323 };
324
325 static const struct sysfs_ops blk_mq_sysfs_ops = {
326 .show = blk_mq_sysfs_show,
327 .store = blk_mq_sysfs_store,
328 };
329
330 static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
331 .show = blk_mq_hw_sysfs_show,
332 .store = blk_mq_hw_sysfs_store,
333 };
334
335 static struct kobj_type blk_mq_ktype = {
336 .sysfs_ops = &blk_mq_sysfs_ops,
337 .release = blk_mq_sysfs_release,
338 };
339
340 static struct kobj_type blk_mq_ctx_ktype = {
341 .sysfs_ops = &blk_mq_sysfs_ops,
342 .default_attrs = default_ctx_attrs,
343 .release = blk_mq_sysfs_release,
344 };
345
346 static struct kobj_type blk_mq_hw_ktype = {
347 .sysfs_ops = &blk_mq_hw_sysfs_ops,
348 .default_attrs = default_hw_ctx_attrs,
349 .release = blk_mq_sysfs_release,
350 };
351
blk_mq_unregister_hctx(struct blk_mq_hw_ctx * hctx)352 static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx)
353 {
354 struct blk_mq_ctx *ctx;
355 int i;
356
357 if (!hctx->nr_ctx)
358 return;
359
360 hctx_for_each_ctx(hctx, ctx, i)
361 kobject_del(&ctx->kobj);
362
363 kobject_del(&hctx->kobj);
364 }
365
blk_mq_register_hctx(struct blk_mq_hw_ctx * hctx)366 static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx)
367 {
368 struct request_queue *q = hctx->queue;
369 struct blk_mq_ctx *ctx;
370 int i, ret;
371
372 if (!hctx->nr_ctx)
373 return 0;
374
375 ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", hctx->queue_num);
376 if (ret)
377 return ret;
378
379 hctx_for_each_ctx(hctx, ctx, i) {
380 ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
381 if (ret)
382 break;
383 }
384
385 return ret;
386 }
387
blk_mq_unregister_disk(struct gendisk * disk)388 void blk_mq_unregister_disk(struct gendisk *disk)
389 {
390 struct request_queue *q = disk->queue;
391 struct blk_mq_hw_ctx *hctx;
392 struct blk_mq_ctx *ctx;
393 int i, j;
394
395 blk_mq_disable_hotplug();
396
397 queue_for_each_hw_ctx(q, hctx, i) {
398 blk_mq_unregister_hctx(hctx);
399
400 hctx_for_each_ctx(hctx, ctx, j)
401 kobject_put(&ctx->kobj);
402
403 kobject_put(&hctx->kobj);
404 }
405
406 kobject_uevent(&q->mq_kobj, KOBJ_REMOVE);
407 kobject_del(&q->mq_kobj);
408 kobject_put(&q->mq_kobj);
409
410 kobject_put(&disk_to_dev(disk)->kobj);
411
412 q->mq_sysfs_init_done = false;
413 blk_mq_enable_hotplug();
414 }
415
blk_mq_sysfs_init(struct request_queue * q)416 static void blk_mq_sysfs_init(struct request_queue *q)
417 {
418 struct blk_mq_hw_ctx *hctx;
419 struct blk_mq_ctx *ctx;
420 int i;
421
422 kobject_init(&q->mq_kobj, &blk_mq_ktype);
423
424 queue_for_each_hw_ctx(q, hctx, i)
425 kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
426
427 queue_for_each_ctx(q, ctx, i)
428 kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
429 }
430
blk_mq_register_disk(struct gendisk * disk)431 int blk_mq_register_disk(struct gendisk *disk)
432 {
433 struct device *dev = disk_to_dev(disk);
434 struct request_queue *q = disk->queue;
435 struct blk_mq_hw_ctx *hctx;
436 int ret, i;
437
438 blk_mq_disable_hotplug();
439
440 blk_mq_sysfs_init(q);
441
442 ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
443 if (ret < 0)
444 goto out;
445
446 kobject_uevent(&q->mq_kobj, KOBJ_ADD);
447
448 queue_for_each_hw_ctx(q, hctx, i) {
449 ret = blk_mq_register_hctx(hctx);
450 if (ret)
451 break;
452 }
453
454 if (ret)
455 blk_mq_unregister_disk(disk);
456 else
457 q->mq_sysfs_init_done = true;
458 out:
459 blk_mq_enable_hotplug();
460
461 return ret;
462 }
463 EXPORT_SYMBOL_GPL(blk_mq_register_disk);
464
blk_mq_sysfs_unregister(struct request_queue * q)465 void blk_mq_sysfs_unregister(struct request_queue *q)
466 {
467 struct blk_mq_hw_ctx *hctx;
468 int i;
469
470 if (!q->mq_sysfs_init_done)
471 return;
472
473 queue_for_each_hw_ctx(q, hctx, i)
474 blk_mq_unregister_hctx(hctx);
475 }
476
blk_mq_sysfs_register(struct request_queue * q)477 int blk_mq_sysfs_register(struct request_queue *q)
478 {
479 struct blk_mq_hw_ctx *hctx;
480 int i, ret = 0;
481
482 if (!q->mq_sysfs_init_done)
483 return ret;
484
485 queue_for_each_hw_ctx(q, hctx, i) {
486 ret = blk_mq_register_hctx(hctx);
487 if (ret)
488 break;
489 }
490
491 return ret;
492 }
493