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_ipi_show(struct blk_mq_hw_ctx *hctx, char *page)
207{
208	ssize_t ret;
209
210	spin_lock(&hctx->lock);
211	ret = sprintf(page, "%u\n", !!(hctx->flags & BLK_MQ_F_SHOULD_IPI));
212	spin_unlock(&hctx->lock);
213
214	return ret;
215}
216
217static ssize_t blk_mq_hw_sysfs_ipi_store(struct blk_mq_hw_ctx *hctx,
218					 const char *page, size_t len)
219{
220	struct blk_mq_ctx *ctx;
221	unsigned long ret;
222	unsigned int i;
223
224	if (kstrtoul(page, 10, &ret)) {
225		pr_err("blk-mq-sysfs: invalid input '%s'\n", page);
226		return -EINVAL;
227	}
228
229	spin_lock(&hctx->lock);
230	if (ret)
231		hctx->flags |= BLK_MQ_F_SHOULD_IPI;
232	else
233		hctx->flags &= ~BLK_MQ_F_SHOULD_IPI;
234	spin_unlock(&hctx->lock);
235
236	hctx_for_each_ctx(hctx, ctx, i)
237		ctx->ipi_redirect = !!ret;
238
239	return len;
240}
241
242static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
243{
244	return blk_mq_tag_sysfs_show(hctx->tags, page);
245}
246
247static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
248{
249	unsigned int i, queue_num, first = 1;
250	ssize_t ret = 0;
251
252	blk_mq_disable_hotplug();
253
254	for_each_online_cpu(i) {
255		queue_num = hctx->queue->mq_map[i];
256		if (queue_num != hctx->queue_num)
257			continue;
258
259		if (first)
260			ret += sprintf(ret + page, "%u", i);
261		else
262			ret += sprintf(ret + page, ", %u", i);
263
264		first = 0;
265	}
266
267	blk_mq_enable_hotplug();
268
269	ret += sprintf(ret + page, "\n");
270	return ret;
271}
272
273static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = {
274	.attr = {.name = "dispatched", .mode = S_IRUGO },
275	.show = blk_mq_sysfs_dispatched_show,
276};
277static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = {
278	.attr = {.name = "merged", .mode = S_IRUGO },
279	.show = blk_mq_sysfs_merged_show,
280};
281static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = {
282	.attr = {.name = "completed", .mode = S_IRUGO },
283	.show = blk_mq_sysfs_completed_show,
284};
285static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = {
286	.attr = {.name = "rq_list", .mode = S_IRUGO },
287	.show = blk_mq_sysfs_rq_list_show,
288};
289
290static struct attribute *default_ctx_attrs[] = {
291	&blk_mq_sysfs_dispatched.attr,
292	&blk_mq_sysfs_merged.attr,
293	&blk_mq_sysfs_completed.attr,
294	&blk_mq_sysfs_rq_list.attr,
295	NULL,
296};
297
298static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = {
299	.attr = {.name = "queued", .mode = S_IRUGO },
300	.show = blk_mq_hw_sysfs_queued_show,
301};
302static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = {
303	.attr = {.name = "run", .mode = S_IRUGO },
304	.show = blk_mq_hw_sysfs_run_show,
305};
306static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = {
307	.attr = {.name = "dispatched", .mode = S_IRUGO },
308	.show = blk_mq_hw_sysfs_dispatched_show,
309};
310static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = {
311	.attr = {.name = "pending", .mode = S_IRUGO },
312	.show = blk_mq_hw_sysfs_rq_list_show,
313};
314static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_ipi = {
315	.attr = {.name = "ipi_redirect", .mode = S_IRUGO | S_IWUSR},
316	.show = blk_mq_hw_sysfs_ipi_show,
317	.store = blk_mq_hw_sysfs_ipi_store,
318};
319static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = {
320	.attr = {.name = "tags", .mode = S_IRUGO },
321	.show = blk_mq_hw_sysfs_tags_show,
322};
323static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
324	.attr = {.name = "cpu_list", .mode = S_IRUGO },
325	.show = blk_mq_hw_sysfs_cpus_show,
326};
327
328static struct attribute *default_hw_ctx_attrs[] = {
329	&blk_mq_hw_sysfs_queued.attr,
330	&blk_mq_hw_sysfs_run.attr,
331	&blk_mq_hw_sysfs_dispatched.attr,
332	&blk_mq_hw_sysfs_pending.attr,
333	&blk_mq_hw_sysfs_ipi.attr,
334	&blk_mq_hw_sysfs_tags.attr,
335	&blk_mq_hw_sysfs_cpus.attr,
336	NULL,
337};
338
339static const struct sysfs_ops blk_mq_sysfs_ops = {
340	.show	= blk_mq_sysfs_show,
341	.store	= blk_mq_sysfs_store,
342};
343
344static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
345	.show	= blk_mq_hw_sysfs_show,
346	.store	= blk_mq_hw_sysfs_store,
347};
348
349static struct kobj_type blk_mq_ktype = {
350	.sysfs_ops	= &blk_mq_sysfs_ops,
351	.release	= blk_mq_sysfs_release,
352};
353
354static struct kobj_type blk_mq_ctx_ktype = {
355	.sysfs_ops	= &blk_mq_sysfs_ops,
356	.default_attrs	= default_ctx_attrs,
357	.release	= blk_mq_sysfs_release,
358};
359
360static struct kobj_type blk_mq_hw_ktype = {
361	.sysfs_ops	= &blk_mq_hw_sysfs_ops,
362	.default_attrs	= default_hw_ctx_attrs,
363	.release	= blk_mq_sysfs_release,
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		hctx_for_each_ctx(hctx, ctx, j) {
375			kobject_del(&ctx->kobj);
376			kobject_put(&ctx->kobj);
377		}
378		kobject_del(&hctx->kobj);
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
389int blk_mq_register_disk(struct gendisk *disk)
390{
391	struct device *dev = disk_to_dev(disk);
392	struct request_queue *q = disk->queue;
393	struct blk_mq_hw_ctx *hctx;
394	struct blk_mq_ctx *ctx;
395	int ret, i, j;
396
397	kobject_init(&q->mq_kobj, &blk_mq_ktype);
398
399	ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
400	if (ret < 0)
401		return ret;
402
403	kobject_uevent(&q->mq_kobj, KOBJ_ADD);
404
405	queue_for_each_hw_ctx(q, hctx, i) {
406		kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
407		ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", i);
408		if (ret)
409			break;
410
411		if (!hctx->nr_ctx)
412			continue;
413
414		hctx_for_each_ctx(hctx, ctx, j) {
415			kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
416			ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
417			if (ret)
418				break;
419		}
420	}
421
422	if (ret) {
423		blk_mq_unregister_disk(disk);
424		return ret;
425	}
426
427	return 0;
428}