1// SPDX-License-Identifier: GPL-2.0
2#include <linux/kernel.h>
3#include <linux/module.h>
4#include <linux/backing-dev.h>
5#include <linux/bio.h>
6#include <linux/blkdev.h>
7#include <linux/mm.h>
8#include <linux/init.h>
9#include <linux/slab.h>
10#include <linux/workqueue.h>
11#include <linux/smp.h>
12
13#include <linux/blk-mq.h>
14#include "blk.h"
15#include "blk-mq.h"
16#include "blk-mq-tag.h"
17
18static void blk_mq_sysfs_release(struct kobject *kobj)
19{
20	struct blk_mq_ctxs *ctxs = container_of(kobj, struct blk_mq_ctxs, kobj);
21
22	free_percpu(ctxs->queue_ctx);
23	kfree(ctxs);
24}
25
26static void blk_mq_ctx_sysfs_release(struct kobject *kobj)
27{
28	struct blk_mq_ctx *ctx = container_of(kobj, struct blk_mq_ctx, kobj);
29
30	/* ctx->ctxs won't be released until all ctx are freed */
31	kobject_put(&ctx->ctxs->kobj);
32}
33
34static void blk_mq_hw_sysfs_release(struct kobject *kobj)
35{
36	struct blk_mq_hw_ctx *hctx = container_of(kobj, struct blk_mq_hw_ctx,
37						  kobj);
38
39	if (hctx->flags & BLK_MQ_F_BLOCKING)
40		cleanup_srcu_struct(hctx->srcu);
41	blk_free_flush_queue(hctx->fq);
42	sbitmap_free(&hctx->ctx_map);
43	free_cpumask_var(hctx->cpumask);
44	kfree(hctx->ctxs);
45	kfree(hctx);
46}
47
48struct blk_mq_ctx_sysfs_entry {
49	struct attribute attr;
50	ssize_t (*show)(struct blk_mq_ctx *, char *);
51	ssize_t (*store)(struct blk_mq_ctx *, const char *, size_t);
52};
53
54struct blk_mq_hw_ctx_sysfs_entry {
55	struct attribute attr;
56	ssize_t (*show)(struct blk_mq_hw_ctx *, char *);
57	ssize_t (*store)(struct blk_mq_hw_ctx *, const char *, size_t);
58};
59
60static ssize_t blk_mq_sysfs_show(struct kobject *kobj, struct attribute *attr,
61				 char *page)
62{
63	struct blk_mq_ctx_sysfs_entry *entry;
64	struct blk_mq_ctx *ctx;
65	struct request_queue *q;
66	ssize_t res;
67
68	entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
69	ctx = container_of(kobj, struct blk_mq_ctx, kobj);
70	q = ctx->queue;
71
72	if (!entry->show)
73		return -EIO;
74
75	mutex_lock(&q->sysfs_lock);
76	res = entry->show(ctx, page);
77	mutex_unlock(&q->sysfs_lock);
78	return res;
79}
80
81static ssize_t blk_mq_sysfs_store(struct kobject *kobj, struct attribute *attr,
82				  const char *page, size_t length)
83{
84	struct blk_mq_ctx_sysfs_entry *entry;
85	struct blk_mq_ctx *ctx;
86	struct request_queue *q;
87	ssize_t res;
88
89	entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
90	ctx = container_of(kobj, struct blk_mq_ctx, kobj);
91	q = ctx->queue;
92
93	if (!entry->store)
94		return -EIO;
95
96	mutex_lock(&q->sysfs_lock);
97	res = entry->store(ctx, page, length);
98	mutex_unlock(&q->sysfs_lock);
99	return res;
100}
101
102static ssize_t blk_mq_hw_sysfs_show(struct kobject *kobj,
103				    struct attribute *attr, char *page)
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->show)
115		return -EIO;
116
117	mutex_lock(&q->sysfs_lock);
118	res = entry->show(hctx, page);
119	mutex_unlock(&q->sysfs_lock);
120	return res;
121}
122
123static ssize_t blk_mq_hw_sysfs_store(struct kobject *kobj,
124				     struct attribute *attr, const char *page,
125				     size_t length)
126{
127	struct blk_mq_hw_ctx_sysfs_entry *entry;
128	struct blk_mq_hw_ctx *hctx;
129	struct request_queue *q;
130	ssize_t res;
131
132	entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
133	hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
134	q = hctx->queue;
135
136	if (!entry->store)
137		return -EIO;
138
139	mutex_lock(&q->sysfs_lock);
140	res = entry->store(hctx, page, length);
141	mutex_unlock(&q->sysfs_lock);
142	return res;
143}
144
145static ssize_t blk_mq_hw_sysfs_nr_tags_show(struct blk_mq_hw_ctx *hctx,
146					    char *page)
147{
148	return sprintf(page, "%u\n", hctx->tags->nr_tags);
149}
150
151static ssize_t blk_mq_hw_sysfs_nr_reserved_tags_show(struct blk_mq_hw_ctx *hctx,
152						     char *page)
153{
154	return sprintf(page, "%u\n", hctx->tags->nr_reserved_tags);
155}
156
157static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
158{
159	const size_t size = PAGE_SIZE - 1;
160	unsigned int i, first = 1;
161	int ret = 0, pos = 0;
162
163	for_each_cpu(i, hctx->cpumask) {
164		if (first)
165			ret = snprintf(pos + page, size - pos, "%u", i);
166		else
167			ret = snprintf(pos + page, size - pos, ", %u", i);
168
169		if (ret >= size - pos)
170			break;
171
172		first = 0;
173		pos += ret;
174	}
175
176	ret = snprintf(pos + page, size + 1 - pos, "\n");
177	return pos + ret;
178}
179
180static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_nr_tags = {
181	.attr = {.name = "nr_tags", .mode = 0444 },
182	.show = blk_mq_hw_sysfs_nr_tags_show,
183};
184static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_nr_reserved_tags = {
185	.attr = {.name = "nr_reserved_tags", .mode = 0444 },
186	.show = blk_mq_hw_sysfs_nr_reserved_tags_show,
187};
188static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
189	.attr = {.name = "cpu_list", .mode = 0444 },
190	.show = blk_mq_hw_sysfs_cpus_show,
191};
192
193static struct attribute *default_hw_ctx_attrs[] = {
194	&blk_mq_hw_sysfs_nr_tags.attr,
195	&blk_mq_hw_sysfs_nr_reserved_tags.attr,
196	&blk_mq_hw_sysfs_cpus.attr,
197	NULL,
198};
199ATTRIBUTE_GROUPS(default_hw_ctx);
200
201static const struct sysfs_ops blk_mq_sysfs_ops = {
202	.show	= blk_mq_sysfs_show,
203	.store	= blk_mq_sysfs_store,
204};
205
206static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
207	.show	= blk_mq_hw_sysfs_show,
208	.store	= blk_mq_hw_sysfs_store,
209};
210
211static struct kobj_type blk_mq_ktype = {
212	.sysfs_ops	= &blk_mq_sysfs_ops,
213	.release	= blk_mq_sysfs_release,
214};
215
216static struct kobj_type blk_mq_ctx_ktype = {
217	.sysfs_ops	= &blk_mq_sysfs_ops,
218	.release	= blk_mq_ctx_sysfs_release,
219};
220
221static struct kobj_type blk_mq_hw_ktype = {
222	.sysfs_ops	= &blk_mq_hw_sysfs_ops,
223	.default_groups = default_hw_ctx_groups,
224	.release	= blk_mq_hw_sysfs_release,
225};
226
227static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx)
228{
229	struct blk_mq_ctx *ctx;
230	int i;
231
232	if (!hctx->nr_ctx)
233		return;
234
235	hctx_for_each_ctx(hctx, ctx, i)
236		kobject_del(&ctx->kobj);
237
238	kobject_del(&hctx->kobj);
239}
240
241static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx)
242{
243	struct request_queue *q = hctx->queue;
244	struct blk_mq_ctx *ctx;
245	int i, j, ret;
246
247	if (!hctx->nr_ctx)
248		return 0;
249
250	ret = kobject_add(&hctx->kobj, q->mq_kobj, "%u", hctx->queue_num);
251	if (ret)
252		return ret;
253
254	hctx_for_each_ctx(hctx, ctx, i) {
255		ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
256		if (ret)
257			goto out;
258	}
259
260	return 0;
261out:
262	hctx_for_each_ctx(hctx, ctx, j) {
263		if (j < i)
264			kobject_del(&ctx->kobj);
265	}
266	kobject_del(&hctx->kobj);
267	return ret;
268}
269
270void blk_mq_unregister_dev(struct device *dev, struct request_queue *q)
271{
272	struct blk_mq_hw_ctx *hctx;
273	int i;
274
275	lockdep_assert_held(&q->sysfs_dir_lock);
276
277	queue_for_each_hw_ctx(q, hctx, i)
278		blk_mq_unregister_hctx(hctx);
279
280	kobject_uevent(q->mq_kobj, KOBJ_REMOVE);
281	kobject_del(q->mq_kobj);
282	kobject_put(&dev->kobj);
283
284	q->mq_sysfs_init_done = false;
285}
286
287void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx)
288{
289	kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
290}
291
292void blk_mq_sysfs_deinit(struct request_queue *q)
293{
294	struct blk_mq_ctx *ctx;
295	int cpu;
296
297	for_each_possible_cpu(cpu) {
298		ctx = per_cpu_ptr(q->queue_ctx, cpu);
299		kobject_put(&ctx->kobj);
300	}
301	kobject_put(q->mq_kobj);
302}
303
304void blk_mq_sysfs_init(struct request_queue *q)
305{
306	struct blk_mq_ctx *ctx;
307	int cpu;
308
309	kobject_init(q->mq_kobj, &blk_mq_ktype);
310
311	for_each_possible_cpu(cpu) {
312		ctx = per_cpu_ptr(q->queue_ctx, cpu);
313
314		kobject_get(q->mq_kobj);
315		kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
316	}
317}
318
319int __blk_mq_register_dev(struct device *dev, struct request_queue *q)
320{
321	struct blk_mq_hw_ctx *hctx;
322	int ret, i;
323
324	WARN_ON_ONCE(!q->kobj.parent);
325	lockdep_assert_held(&q->sysfs_dir_lock);
326
327	ret = kobject_add(q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
328	if (ret < 0)
329		goto out;
330
331	kobject_uevent(q->mq_kobj, KOBJ_ADD);
332
333	queue_for_each_hw_ctx(q, hctx, i) {
334		ret = blk_mq_register_hctx(hctx);
335		if (ret)
336			goto unreg;
337	}
338
339	q->mq_sysfs_init_done = true;
340
341out:
342	return ret;
343
344unreg:
345	while (--i >= 0)
346		blk_mq_unregister_hctx(q->queue_hw_ctx[i]);
347
348	kobject_uevent(q->mq_kobj, KOBJ_REMOVE);
349	kobject_del(q->mq_kobj);
350	kobject_put(&dev->kobj);
351	return ret;
352}
353
354void blk_mq_sysfs_unregister(struct request_queue *q)
355{
356	struct blk_mq_hw_ctx *hctx;
357	int i;
358
359	mutex_lock(&q->sysfs_dir_lock);
360	if (!q->mq_sysfs_init_done)
361		goto unlock;
362
363	queue_for_each_hw_ctx(q, hctx, i)
364		blk_mq_unregister_hctx(hctx);
365
366unlock:
367	mutex_unlock(&q->sysfs_dir_lock);
368}
369
370int blk_mq_sysfs_register(struct request_queue *q)
371{
372	struct blk_mq_hw_ctx *hctx;
373	int i, ret = 0;
374
375	mutex_lock(&q->sysfs_dir_lock);
376	if (!q->mq_sysfs_init_done)
377		goto unlock;
378
379	queue_for_each_hw_ctx(q, hctx, i) {
380		ret = blk_mq_register_hctx(hctx);
381		if (ret)
382			break;
383	}
384
385unlock:
386	mutex_unlock(&q->sysfs_dir_lock);
387
388	return ret;
389}
390