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	cancel_delayed_work_sync(&hctx->run_work);
 40
 41	if (hctx->flags & BLK_MQ_F_BLOCKING)
 42		cleanup_srcu_struct(hctx->srcu);
 43	blk_free_flush_queue(hctx->fq);
 44	sbitmap_free(&hctx->ctx_map);
 45	free_cpumask_var(hctx->cpumask);
 46	kfree(hctx->ctxs);
 47	kfree(hctx);
 48}
 49
 50struct blk_mq_ctx_sysfs_entry {
 51	struct attribute attr;
 52	ssize_t (*show)(struct blk_mq_ctx *, char *);
 53	ssize_t (*store)(struct blk_mq_ctx *, const char *, size_t);
 54};
 55
 56struct blk_mq_hw_ctx_sysfs_entry {
 57	struct attribute attr;
 58	ssize_t (*show)(struct blk_mq_hw_ctx *, char *);
 59	ssize_t (*store)(struct blk_mq_hw_ctx *, const char *, size_t);
 60};
 61
 62static ssize_t blk_mq_sysfs_show(struct kobject *kobj, struct attribute *attr,
 63				 char *page)
 64{
 65	struct blk_mq_ctx_sysfs_entry *entry;
 66	struct blk_mq_ctx *ctx;
 67	struct request_queue *q;
 68	ssize_t res;
 69
 70	entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
 71	ctx = container_of(kobj, struct blk_mq_ctx, kobj);
 72	q = ctx->queue;
 73
 74	if (!entry->show)
 75		return -EIO;
 76
 77	mutex_lock(&q->sysfs_lock);
 78	res = entry->show(ctx, page);
 79	mutex_unlock(&q->sysfs_lock);
 80	return res;
 81}
 82
 83static ssize_t blk_mq_sysfs_store(struct kobject *kobj, struct attribute *attr,
 84				  const char *page, size_t length)
 85{
 86	struct blk_mq_ctx_sysfs_entry *entry;
 87	struct blk_mq_ctx *ctx;
 88	struct request_queue *q;
 89	ssize_t res;
 90
 91	entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
 92	ctx = container_of(kobj, struct blk_mq_ctx, kobj);
 93	q = ctx->queue;
 94
 95	if (!entry->store)
 96		return -EIO;
 97
 98	mutex_lock(&q->sysfs_lock);
 99	res = entry->store(ctx, page, length);
100	mutex_unlock(&q->sysfs_lock);
101	return res;
102}
103
104static ssize_t blk_mq_hw_sysfs_show(struct kobject *kobj,
105				    struct attribute *attr, char *page)
106{
107	struct blk_mq_hw_ctx_sysfs_entry *entry;
108	struct blk_mq_hw_ctx *hctx;
109	struct request_queue *q;
110	ssize_t res;
111
112	entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
113	hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
114	q = hctx->queue;
115
116	if (!entry->show)
117		return -EIO;
118
119	mutex_lock(&q->sysfs_lock);
120	res = entry->show(hctx, page);
121	mutex_unlock(&q->sysfs_lock);
122	return res;
123}
124
125static ssize_t blk_mq_hw_sysfs_store(struct kobject *kobj,
126				     struct attribute *attr, const char *page,
127				     size_t length)
128{
129	struct blk_mq_hw_ctx_sysfs_entry *entry;
130	struct blk_mq_hw_ctx *hctx;
131	struct request_queue *q;
132	ssize_t res;
133
134	entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
135	hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
136	q = hctx->queue;
137
138	if (!entry->store)
139		return -EIO;
140
141	mutex_lock(&q->sysfs_lock);
142	res = entry->store(hctx, page, length);
143	mutex_unlock(&q->sysfs_lock);
144	return res;
145}
146
147static ssize_t blk_mq_hw_sysfs_nr_tags_show(struct blk_mq_hw_ctx *hctx,
148					    char *page)
149{
150	return sprintf(page, "%u\n", hctx->tags->nr_tags);
151}
152
153static ssize_t blk_mq_hw_sysfs_nr_reserved_tags_show(struct blk_mq_hw_ctx *hctx,
154						     char *page)
155{
156	return sprintf(page, "%u\n", hctx->tags->nr_reserved_tags);
157}
158
159static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
160{
161	const size_t size = PAGE_SIZE - 1;
162	unsigned int i, first = 1;
163	int ret = 0, pos = 0;
164
165	for_each_cpu(i, hctx->cpumask) {
166		if (first)
167			ret = snprintf(pos + page, size - pos, "%u", i);
168		else
169			ret = snprintf(pos + page, size - pos, ", %u", i);
170
171		if (ret >= size - pos)
172			break;
173
174		first = 0;
175		pos += ret;
176	}
177
178	ret = snprintf(pos + page, size + 1 - pos, "\n");
179	return pos + ret;
180}
181
182static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_nr_tags = {
183	.attr = {.name = "nr_tags", .mode = 0444 },
184	.show = blk_mq_hw_sysfs_nr_tags_show,
185};
186static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_nr_reserved_tags = {
187	.attr = {.name = "nr_reserved_tags", .mode = 0444 },
188	.show = blk_mq_hw_sysfs_nr_reserved_tags_show,
189};
190static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
191	.attr = {.name = "cpu_list", .mode = 0444 },
192	.show = blk_mq_hw_sysfs_cpus_show,
193};
194
195static struct attribute *default_hw_ctx_attrs[] = {
196	&blk_mq_hw_sysfs_nr_tags.attr,
197	&blk_mq_hw_sysfs_nr_reserved_tags.attr,
198	&blk_mq_hw_sysfs_cpus.attr,
199	NULL,
200};
201ATTRIBUTE_GROUPS(default_hw_ctx);
202
203static const struct sysfs_ops blk_mq_sysfs_ops = {
204	.show	= blk_mq_sysfs_show,
205	.store	= blk_mq_sysfs_store,
206};
207
208static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
209	.show	= blk_mq_hw_sysfs_show,
210	.store	= blk_mq_hw_sysfs_store,
211};
212
213static struct kobj_type blk_mq_ktype = {
214	.sysfs_ops	= &blk_mq_sysfs_ops,
215	.release	= blk_mq_sysfs_release,
216};
217
218static struct kobj_type blk_mq_ctx_ktype = {
219	.sysfs_ops	= &blk_mq_sysfs_ops,
220	.release	= blk_mq_ctx_sysfs_release,
221};
222
223static struct kobj_type blk_mq_hw_ktype = {
224	.sysfs_ops	= &blk_mq_hw_sysfs_ops,
225	.default_groups = default_hw_ctx_groups,
226	.release	= blk_mq_hw_sysfs_release,
227};
228
229static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx)
230{
231	struct blk_mq_ctx *ctx;
232	int i;
233
234	if (!hctx->nr_ctx)
235		return;
236
237	hctx_for_each_ctx(hctx, ctx, i)
238		kobject_del(&ctx->kobj);
239
240	kobject_del(&hctx->kobj);
241}
242
243static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx)
244{
245	struct request_queue *q = hctx->queue;
246	struct blk_mq_ctx *ctx;
247	int i, ret;
248
249	if (!hctx->nr_ctx)
250		return 0;
251
252	ret = kobject_add(&hctx->kobj, q->mq_kobj, "%u", hctx->queue_num);
253	if (ret)
254		return ret;
255
256	hctx_for_each_ctx(hctx, ctx, i) {
257		ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
258		if (ret)
259			break;
260	}
261
262	return ret;
263}
264
265void blk_mq_unregister_dev(struct device *dev, struct request_queue *q)
266{
267	struct blk_mq_hw_ctx *hctx;
268	int i;
269
270	lockdep_assert_held(&q->sysfs_dir_lock);
271
272	queue_for_each_hw_ctx(q, hctx, i)
273		blk_mq_unregister_hctx(hctx);
274
275	kobject_uevent(q->mq_kobj, KOBJ_REMOVE);
276	kobject_del(q->mq_kobj);
277	kobject_put(&dev->kobj);
278
279	q->mq_sysfs_init_done = false;
280}
281
282void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx)
283{
284	kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
285}
286
287void blk_mq_sysfs_deinit(struct request_queue *q)
288{
289	struct blk_mq_ctx *ctx;
290	int cpu;
291
292	for_each_possible_cpu(cpu) {
293		ctx = per_cpu_ptr(q->queue_ctx, cpu);
294		kobject_put(&ctx->kobj);
295	}
296	kobject_put(q->mq_kobj);
297}
298
299void blk_mq_sysfs_init(struct request_queue *q)
300{
301	struct blk_mq_ctx *ctx;
302	int cpu;
303
304	kobject_init(q->mq_kobj, &blk_mq_ktype);
305
306	for_each_possible_cpu(cpu) {
307		ctx = per_cpu_ptr(q->queue_ctx, cpu);
308
309		kobject_get(q->mq_kobj);
310		kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
311	}
312}
313
314int __blk_mq_register_dev(struct device *dev, struct request_queue *q)
315{
316	struct blk_mq_hw_ctx *hctx;
317	int ret, i;
318
319	WARN_ON_ONCE(!q->kobj.parent);
320	lockdep_assert_held(&q->sysfs_dir_lock);
321
322	ret = kobject_add(q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
323	if (ret < 0)
324		goto out;
325
326	kobject_uevent(q->mq_kobj, KOBJ_ADD);
327
328	queue_for_each_hw_ctx(q, hctx, i) {
329		ret = blk_mq_register_hctx(hctx);
330		if (ret)
331			goto unreg;
332	}
333
334	q->mq_sysfs_init_done = true;
335
336out:
337	return ret;
338
339unreg:
340	while (--i >= 0)
341		blk_mq_unregister_hctx(q->queue_hw_ctx[i]);
342
343	kobject_uevent(q->mq_kobj, KOBJ_REMOVE);
344	kobject_del(q->mq_kobj);
345	kobject_put(&dev->kobj);
346	return ret;
347}
348
349void blk_mq_sysfs_unregister(struct request_queue *q)
350{
351	struct blk_mq_hw_ctx *hctx;
352	int i;
353
354	mutex_lock(&q->sysfs_dir_lock);
355	if (!q->mq_sysfs_init_done)
356		goto unlock;
357
358	queue_for_each_hw_ctx(q, hctx, i)
359		blk_mq_unregister_hctx(hctx);
360
361unlock:
362	mutex_unlock(&q->sysfs_dir_lock);
363}
364
365int blk_mq_sysfs_register(struct request_queue *q)
366{
367	struct blk_mq_hw_ctx *hctx;
368	int i, ret = 0;
369
370	mutex_lock(&q->sysfs_dir_lock);
371	if (!q->mq_sysfs_init_done)
372		goto unlock;
373
374	queue_for_each_hw_ctx(q, hctx, i) {
375		ret = blk_mq_register_hctx(hctx);
376		if (ret)
377			break;
378	}
379
380unlock:
381	mutex_unlock(&q->sysfs_dir_lock);
382
383	return ret;
384}