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
16static void blk_mq_sysfs_release(struct kobject *kobj)
17{
18}
19
20struct 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
26struct 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
32static 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
55static 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
78static 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
101static 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
125static 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
131static 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
136static 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
142static ssize_t sysfs_list_show(char *page, struct list_head *list, char *msg)
143{
144	char *start_page = page;
145	struct request *rq;
146
147	page += sprintf(page, "%s:\n", msg);
148
149	list_for_each_entry(rq, list, queuelist)
150		page += sprintf(page, "\t%p\n", rq);
151
152	return page - start_page;
153}
154
155static ssize_t blk_mq_sysfs_rq_list_show(struct blk_mq_ctx *ctx, char *page)
156{
157	ssize_t ret;
158
159	spin_lock(&ctx->lock);
160	ret = sysfs_list_show(page, &ctx->rq_list, "CTX pending");
161	spin_unlock(&ctx->lock);
162
163	return ret;
164}
165
166static ssize_t blk_mq_hw_sysfs_queued_show(struct blk_mq_hw_ctx *hctx,
167					   char *page)
168{
169	return sprintf(page, "%lu\n", hctx->queued);
170}
171
172static ssize_t blk_mq_hw_sysfs_run_show(struct blk_mq_hw_ctx *hctx, char *page)
173{
174	return sprintf(page, "%lu\n", hctx->run);
175}
176
177static ssize_t blk_mq_hw_sysfs_dispatched_show(struct blk_mq_hw_ctx *hctx,
178					       char *page)
179{
180	char *start_page = page;
181	int i;
182
183	page += sprintf(page, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
184
185	for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER; i++) {
186		unsigned long d = 1U << (i - 1);
187
188		page += sprintf(page, "%8lu\t%lu\n", d, hctx->dispatched[i]);
189	}
190
191	return page - start_page;
192}
193
194static ssize_t blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx *hctx,
195					    char *page)
196{
197	ssize_t ret;
198
199	spin_lock(&hctx->lock);
200	ret = sysfs_list_show(page, &hctx->dispatch, "HCTX pending");
201	spin_unlock(&hctx->lock);
202
203	return ret;
204}
205
206static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
207{
208	return blk_mq_tag_sysfs_show(hctx->tags, page);
209}
210
211static ssize_t blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx *hctx, char *page)
212{
213	return sprintf(page, "%u\n", atomic_read(&hctx->nr_active));
214}
215
216static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
217{
218	unsigned int i, first = 1;
219	ssize_t ret = 0;
220
221	blk_mq_disable_hotplug();
222
223	for_each_cpu(i, hctx->cpumask) {
224		if (first)
225			ret += sprintf(ret + page, "%u", i);
226		else
227			ret += sprintf(ret + page, ", %u", i);
228
229		first = 0;
230	}
231
232	blk_mq_enable_hotplug();
233
234	ret += sprintf(ret + page, "\n");
235	return ret;
236}
237
238static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = {
239	.attr = {.name = "dispatched", .mode = S_IRUGO },
240	.show = blk_mq_sysfs_dispatched_show,
241};
242static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = {
243	.attr = {.name = "merged", .mode = S_IRUGO },
244	.show = blk_mq_sysfs_merged_show,
245};
246static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = {
247	.attr = {.name = "completed", .mode = S_IRUGO },
248	.show = blk_mq_sysfs_completed_show,
249};
250static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = {
251	.attr = {.name = "rq_list", .mode = S_IRUGO },
252	.show = blk_mq_sysfs_rq_list_show,
253};
254
255static struct attribute *default_ctx_attrs[] = {
256	&blk_mq_sysfs_dispatched.attr,
257	&blk_mq_sysfs_merged.attr,
258	&blk_mq_sysfs_completed.attr,
259	&blk_mq_sysfs_rq_list.attr,
260	NULL,
261};
262
263static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = {
264	.attr = {.name = "queued", .mode = S_IRUGO },
265	.show = blk_mq_hw_sysfs_queued_show,
266};
267static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = {
268	.attr = {.name = "run", .mode = S_IRUGO },
269	.show = blk_mq_hw_sysfs_run_show,
270};
271static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = {
272	.attr = {.name = "dispatched", .mode = S_IRUGO },
273	.show = blk_mq_hw_sysfs_dispatched_show,
274};
275static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_active = {
276	.attr = {.name = "active", .mode = S_IRUGO },
277	.show = blk_mq_hw_sysfs_active_show,
278};
279static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = {
280	.attr = {.name = "pending", .mode = S_IRUGO },
281	.show = blk_mq_hw_sysfs_rq_list_show,
282};
283static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = {
284	.attr = {.name = "tags", .mode = S_IRUGO },
285	.show = blk_mq_hw_sysfs_tags_show,
286};
287static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
288	.attr = {.name = "cpu_list", .mode = S_IRUGO },
289	.show = blk_mq_hw_sysfs_cpus_show,
290};
291
292static struct attribute *default_hw_ctx_attrs[] = {
293	&blk_mq_hw_sysfs_queued.attr,
294	&blk_mq_hw_sysfs_run.attr,
295	&blk_mq_hw_sysfs_dispatched.attr,
296	&blk_mq_hw_sysfs_pending.attr,
297	&blk_mq_hw_sysfs_tags.attr,
298	&blk_mq_hw_sysfs_cpus.attr,
299	&blk_mq_hw_sysfs_active.attr,
300	NULL,
301};
302
303static const struct sysfs_ops blk_mq_sysfs_ops = {
304	.show	= blk_mq_sysfs_show,
305	.store	= blk_mq_sysfs_store,
306};
307
308static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
309	.show	= blk_mq_hw_sysfs_show,
310	.store	= blk_mq_hw_sysfs_store,
311};
312
313static struct kobj_type blk_mq_ktype = {
314	.sysfs_ops	= &blk_mq_sysfs_ops,
315	.release	= blk_mq_sysfs_release,
316};
317
318static struct kobj_type blk_mq_ctx_ktype = {
319	.sysfs_ops	= &blk_mq_sysfs_ops,
320	.default_attrs	= default_ctx_attrs,
321	.release	= blk_mq_sysfs_release,
322};
323
324static struct kobj_type blk_mq_hw_ktype = {
325	.sysfs_ops	= &blk_mq_hw_sysfs_ops,
326	.default_attrs	= default_hw_ctx_attrs,
327	.release	= blk_mq_sysfs_release,
328};
329
330static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx)
331{
332	struct blk_mq_ctx *ctx;
333	int i;
334
335	if (!hctx->nr_ctx || !(hctx->flags & BLK_MQ_F_SYSFS_UP))
336		return;
337
338	hctx_for_each_ctx(hctx, ctx, i)
339		kobject_del(&ctx->kobj);
340
341	kobject_del(&hctx->kobj);
342}
343
344static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx)
345{
346	struct request_queue *q = hctx->queue;
347	struct blk_mq_ctx *ctx;
348	int i, ret;
349
350	if (!hctx->nr_ctx || !(hctx->flags & BLK_MQ_F_SYSFS_UP))
351		return 0;
352
353	ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", hctx->queue_num);
354	if (ret)
355		return ret;
356
357	hctx_for_each_ctx(hctx, ctx, i) {
358		ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
359		if (ret)
360			break;
361	}
362
363	return ret;
364}
365
366void blk_mq_unregister_disk(struct gendisk *disk)
367{
368	struct request_queue *q = disk->queue;
369	struct blk_mq_hw_ctx *hctx;
370	struct blk_mq_ctx *ctx;
371	int i, j;
372
373	queue_for_each_hw_ctx(q, hctx, i) {
374		blk_mq_unregister_hctx(hctx);
375
376		hctx_for_each_ctx(hctx, ctx, j)
377			kobject_put(&ctx->kobj);
378
379		kobject_put(&hctx->kobj);
380	}
381
382	kobject_uevent(&q->mq_kobj, KOBJ_REMOVE);
383	kobject_del(&q->mq_kobj);
384	kobject_put(&q->mq_kobj);
385
386	kobject_put(&disk_to_dev(disk)->kobj);
387}
388
389static void blk_mq_sysfs_init(struct request_queue *q)
390{
391	struct blk_mq_hw_ctx *hctx;
392	struct blk_mq_ctx *ctx;
393	int i, j;
394
395	kobject_init(&q->mq_kobj, &blk_mq_ktype);
396
397	queue_for_each_hw_ctx(q, hctx, i) {
398		kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
399
400		hctx_for_each_ctx(hctx, ctx, j)
401			kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
402	}
403}
404
405/* see blk_register_queue() */
406void blk_mq_finish_init(struct request_queue *q)
407{
408	percpu_ref_switch_to_percpu(&q->mq_usage_counter);
409}
410
411int blk_mq_register_disk(struct gendisk *disk)
412{
413	struct device *dev = disk_to_dev(disk);
414	struct request_queue *q = disk->queue;
415	struct blk_mq_hw_ctx *hctx;
416	int ret, i;
417
418	blk_mq_sysfs_init(q);
419
420	ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
421	if (ret < 0)
422		return ret;
423
424	kobject_uevent(&q->mq_kobj, KOBJ_ADD);
425
426	queue_for_each_hw_ctx(q, hctx, i) {
427		hctx->flags |= BLK_MQ_F_SYSFS_UP;
428		ret = blk_mq_register_hctx(hctx);
429		if (ret)
430			break;
431	}
432
433	if (ret) {
434		blk_mq_unregister_disk(disk);
435		return ret;
436	}
437
438	return 0;
439}
440
441void blk_mq_sysfs_unregister(struct request_queue *q)
442{
443	struct blk_mq_hw_ctx *hctx;
444	int i;
445
446	queue_for_each_hw_ctx(q, hctx, i)
447		blk_mq_unregister_hctx(hctx);
448}
449
450int blk_mq_sysfs_register(struct request_queue *q)
451{
452	struct blk_mq_hw_ctx *hctx;
453	int i, ret = 0;
454
455	queue_for_each_hw_ctx(q, hctx, i) {
456		ret = blk_mq_register_hctx(hctx);
457		if (ret)
458			break;
459	}
460
461	return ret;
462}
463