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	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
166static 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
177static 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
184static 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
192static 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
198static 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
203static 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
222static 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
234static 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
239static 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
244static 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
262static void blk_mq_stat_clear(struct blk_mq_hw_ctx *hctx)
263{
264	struct blk_mq_ctx *ctx;
265	unsigned int i;
266
267	hctx_for_each_ctx(hctx, ctx, i) {
268		blk_stat_init(&ctx->stat[BLK_STAT_READ]);
269		blk_stat_init(&ctx->stat[BLK_STAT_WRITE]);
270	}
271}
272
273static ssize_t blk_mq_hw_sysfs_stat_store(struct blk_mq_hw_ctx *hctx,
274					  const char *page, size_t count)
275{
276	blk_mq_stat_clear(hctx);
277	return count;
278}
279
280static ssize_t print_stat(char *page, struct blk_rq_stat *stat, const char *pre)
281{
282	return sprintf(page, "%s samples=%llu, mean=%lld, min=%lld, max=%lld\n",
283			pre, (long long) stat->nr_samples,
284			(long long) stat->mean, (long long) stat->min,
285			(long long) stat->max);
286}
287
288static ssize_t blk_mq_hw_sysfs_stat_show(struct blk_mq_hw_ctx *hctx, char *page)
289{
290	struct blk_rq_stat stat[2];
291	ssize_t ret;
292
293	blk_stat_init(&stat[BLK_STAT_READ]);
294	blk_stat_init(&stat[BLK_STAT_WRITE]);
295
296	blk_hctx_stat_get(hctx, stat);
297
298	ret = print_stat(page, &stat[BLK_STAT_READ], "read :");
299	ret += print_stat(page + ret, &stat[BLK_STAT_WRITE], "write:");
300	return ret;
301}
302
303static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = {
304	.attr = {.name = "dispatched", .mode = S_IRUGO },
305	.show = blk_mq_sysfs_dispatched_show,
306};
307static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = {
308	.attr = {.name = "merged", .mode = S_IRUGO },
309	.show = blk_mq_sysfs_merged_show,
310};
311static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = {
312	.attr = {.name = "completed", .mode = S_IRUGO },
313	.show = blk_mq_sysfs_completed_show,
314};
315static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = {
316	.attr = {.name = "rq_list", .mode = S_IRUGO },
317	.show = blk_mq_sysfs_rq_list_show,
318};
319
320static struct attribute *default_ctx_attrs[] = {
321	&blk_mq_sysfs_dispatched.attr,
322	&blk_mq_sysfs_merged.attr,
323	&blk_mq_sysfs_completed.attr,
324	&blk_mq_sysfs_rq_list.attr,
325	NULL,
326};
327
328static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = {
329	.attr = {.name = "queued", .mode = S_IRUGO },
330	.show = blk_mq_hw_sysfs_queued_show,
331};
332static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = {
333	.attr = {.name = "run", .mode = S_IRUGO },
334	.show = blk_mq_hw_sysfs_run_show,
335};
336static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = {
337	.attr = {.name = "dispatched", .mode = S_IRUGO },
338	.show = blk_mq_hw_sysfs_dispatched_show,
339};
340static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_active = {
341	.attr = {.name = "active", .mode = S_IRUGO },
342	.show = blk_mq_hw_sysfs_active_show,
343};
344static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = {
345	.attr = {.name = "pending", .mode = S_IRUGO },
346	.show = blk_mq_hw_sysfs_rq_list_show,
347};
348static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = {
349	.attr = {.name = "tags", .mode = S_IRUGO },
350	.show = blk_mq_hw_sysfs_tags_show,
351};
352static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
353	.attr = {.name = "cpu_list", .mode = S_IRUGO },
354	.show = blk_mq_hw_sysfs_cpus_show,
355};
356static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_poll = {
357	.attr = {.name = "io_poll", .mode = S_IWUSR | S_IRUGO },
358	.show = blk_mq_hw_sysfs_poll_show,
359	.store = blk_mq_hw_sysfs_poll_store,
360};
361static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_stat = {
362	.attr = {.name = "stats", .mode = S_IRUGO | S_IWUSR },
363	.show = blk_mq_hw_sysfs_stat_show,
364	.store = blk_mq_hw_sysfs_stat_store,
365};
366
367static struct attribute *default_hw_ctx_attrs[] = {
368	&blk_mq_hw_sysfs_queued.attr,
369	&blk_mq_hw_sysfs_run.attr,
370	&blk_mq_hw_sysfs_dispatched.attr,
371	&blk_mq_hw_sysfs_pending.attr,
372	&blk_mq_hw_sysfs_tags.attr,
373	&blk_mq_hw_sysfs_cpus.attr,
374	&blk_mq_hw_sysfs_active.attr,
375	&blk_mq_hw_sysfs_poll.attr,
376	&blk_mq_hw_sysfs_stat.attr,
377	NULL,
378};
379
380static const struct sysfs_ops blk_mq_sysfs_ops = {
381	.show	= blk_mq_sysfs_show,
382	.store	= blk_mq_sysfs_store,
383};
384
385static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
386	.show	= blk_mq_hw_sysfs_show,
387	.store	= blk_mq_hw_sysfs_store,
388};
389
390static struct kobj_type blk_mq_ktype = {
391	.sysfs_ops	= &blk_mq_sysfs_ops,
392	.release	= blk_mq_sysfs_release,
393};
394
395static struct kobj_type blk_mq_ctx_ktype = {
396	.sysfs_ops	= &blk_mq_sysfs_ops,
397	.default_attrs	= default_ctx_attrs,
398	.release	= blk_mq_sysfs_release,
399};
400
401static struct kobj_type blk_mq_hw_ktype = {
402	.sysfs_ops	= &blk_mq_hw_sysfs_ops,
403	.default_attrs	= default_hw_ctx_attrs,
404	.release	= blk_mq_sysfs_release,
405};
406
407static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx)
408{
409	struct blk_mq_ctx *ctx;
410	int i;
411
412	if (!hctx->nr_ctx)
413		return;
414
415	hctx_for_each_ctx(hctx, ctx, i)
416		kobject_del(&ctx->kobj);
417
418	kobject_del(&hctx->kobj);
419}
420
421static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx)
422{
423	struct request_queue *q = hctx->queue;
424	struct blk_mq_ctx *ctx;
425	int i, ret;
426
427	if (!hctx->nr_ctx)
428		return 0;
429
430	ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", hctx->queue_num);
431	if (ret)
432		return ret;
433
434	hctx_for_each_ctx(hctx, ctx, i) {
435		ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
436		if (ret)
437			break;
438	}
439
440	return ret;
441}
442
443static void __blk_mq_unregister_dev(struct device *dev, struct request_queue *q)
444{
445	struct blk_mq_hw_ctx *hctx;
446	struct blk_mq_ctx *ctx;
447	int i, j;
448
449	queue_for_each_hw_ctx(q, hctx, i) {
450		blk_mq_unregister_hctx(hctx);
451
452		hctx_for_each_ctx(hctx, ctx, j)
453			kobject_put(&ctx->kobj);
454
455		kobject_put(&hctx->kobj);
456	}
457
458	kobject_uevent(&q->mq_kobj, KOBJ_REMOVE);
459	kobject_del(&q->mq_kobj);
460	kobject_put(&q->mq_kobj);
461
462	kobject_put(&dev->kobj);
463
464	q->mq_sysfs_init_done = false;
465}
466
467void blk_mq_unregister_dev(struct device *dev, struct request_queue *q)
468{
469	blk_mq_disable_hotplug();
470	__blk_mq_unregister_dev(dev, q);
471	blk_mq_enable_hotplug();
472}
473
474void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx)
475{
476	kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
477}
478
479static void blk_mq_sysfs_init(struct request_queue *q)
480{
481	struct blk_mq_ctx *ctx;
482	int cpu;
483
484	kobject_init(&q->mq_kobj, &blk_mq_ktype);
485
486	for_each_possible_cpu(cpu) {
487		ctx = per_cpu_ptr(q->queue_ctx, cpu);
488		kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
489	}
490}
491
492int blk_mq_register_dev(struct device *dev, struct request_queue *q)
493{
494	struct blk_mq_hw_ctx *hctx;
495	int ret, i;
496
497	blk_mq_disable_hotplug();
498
499	blk_mq_sysfs_init(q);
500
501	ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
502	if (ret < 0)
503		goto out;
504
505	kobject_uevent(&q->mq_kobj, KOBJ_ADD);
506
507	queue_for_each_hw_ctx(q, hctx, i) {
508		ret = blk_mq_register_hctx(hctx);
509		if (ret)
510			break;
511	}
512
513	if (ret)
514		__blk_mq_unregister_dev(dev, q);
515	else
516		q->mq_sysfs_init_done = true;
517out:
518	blk_mq_enable_hotplug();
519
520	return ret;
521}
522EXPORT_SYMBOL_GPL(blk_mq_register_dev);
523
524void blk_mq_sysfs_unregister(struct request_queue *q)
525{
526	struct blk_mq_hw_ctx *hctx;
527	int i;
528
529	if (!q->mq_sysfs_init_done)
530		return;
531
532	queue_for_each_hw_ctx(q, hctx, i)
533		blk_mq_unregister_hctx(hctx);
534}
535
536int blk_mq_sysfs_register(struct request_queue *q)
537{
538	struct blk_mq_hw_ctx *hctx;
539	int i, ret = 0;
540
541	if (!q->mq_sysfs_init_done)
542		return ret;
543
544	queue_for_each_hw_ctx(q, hctx, i) {
545		ret = blk_mq_register_hctx(hctx);
546		if (ret)
547			break;
548	}
549
550	return ret;
551}