1 #include <linux/kernel.h>
2 #include <linux/module.h>
3 #include <linux/backing-dev.h>
5 #include <linux/blkdev.h>
7 #include <linux/init.h>
8 #include <linux/slab.h>
9 #include <linux/workqueue.h>
10 #include <linux/smp.h>
12 #include <linux/blk-mq.h>
14 #include "blk-mq-tag.h"
16 static void blk_mq_sysfs_release(struct kobject *kobj)
20 struct 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);
26 struct 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);
32 static ssize_t blk_mq_sysfs_show(struct kobject *kobj, struct attribute *attr,
35 struct blk_mq_ctx_sysfs_entry *entry;
36 struct blk_mq_ctx *ctx;
37 struct request_queue *q;
40 entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
41 ctx = container_of(kobj, struct blk_mq_ctx, kobj);
48 mutex_lock(&q->sysfs_lock);
49 if (!blk_queue_dying(q))
50 res = entry->show(ctx, page);
51 mutex_unlock(&q->sysfs_lock);
55 static ssize_t blk_mq_sysfs_store(struct kobject *kobj, struct attribute *attr,
56 const char *page, size_t length)
58 struct blk_mq_ctx_sysfs_entry *entry;
59 struct blk_mq_ctx *ctx;
60 struct request_queue *q;
63 entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
64 ctx = container_of(kobj, struct blk_mq_ctx, kobj);
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);
78 static ssize_t blk_mq_hw_sysfs_show(struct kobject *kobj,
79 struct attribute *attr, char *page)
81 struct blk_mq_hw_ctx_sysfs_entry *entry;
82 struct blk_mq_hw_ctx *hctx;
83 struct request_queue *q;
86 entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
87 hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
94 mutex_lock(&q->sysfs_lock);
95 if (!blk_queue_dying(q))
96 res = entry->show(hctx, page);
97 mutex_unlock(&q->sysfs_lock);
101 static ssize_t blk_mq_hw_sysfs_store(struct kobject *kobj,
102 struct attribute *attr, const char *page,
105 struct blk_mq_hw_ctx_sysfs_entry *entry;
106 struct blk_mq_hw_ctx *hctx;
107 struct request_queue *q;
110 entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
111 hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
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);
125 static ssize_t blk_mq_sysfs_dispatched_show(struct blk_mq_ctx *ctx, char *page)
127 return sprintf(page, "%lu %lu\n", ctx->rq_dispatched[1],
128 ctx->rq_dispatched[0]);
131 static ssize_t blk_mq_sysfs_merged_show(struct blk_mq_ctx *ctx, char *page)
133 return sprintf(page, "%lu\n", ctx->rq_merged);
136 static ssize_t blk_mq_sysfs_completed_show(struct blk_mq_ctx *ctx, char *page)
138 return sprintf(page, "%lu %lu\n", ctx->rq_completed[1],
139 ctx->rq_completed[0]);
142 static ssize_t sysfs_list_show(char *page, struct list_head *list, char *msg)
145 int len = snprintf(page, PAGE_SIZE - 1, "%s:\n", msg);
147 list_for_each_entry(rq, list, queuelist) {
148 const int rq_len = 2 * sizeof(rq) + 2;
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)
155 len += snprintf(page + len, PAGE_SIZE - 1 - len,
159 len += snprintf(page + len, PAGE_SIZE - 1 - len,
166 static ssize_t blk_mq_sysfs_rq_list_show(struct blk_mq_ctx *ctx, char *page)
170 spin_lock(&ctx->lock);
171 ret = sysfs_list_show(page, &ctx->rq_list, "CTX pending");
172 spin_unlock(&ctx->lock);
177 static ssize_t blk_mq_hw_sysfs_poll_show(struct blk_mq_hw_ctx *hctx, char *page)
179 return sprintf(page, "considered=%lu, invoked=%lu, success=%lu\n",
180 hctx->poll_considered, hctx->poll_invoked,
184 static ssize_t blk_mq_hw_sysfs_poll_store(struct blk_mq_hw_ctx *hctx,
185 const char *page, size_t size)
187 hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
192 static ssize_t blk_mq_hw_sysfs_queued_show(struct blk_mq_hw_ctx *hctx,
195 return sprintf(page, "%lu\n", hctx->queued);
198 static ssize_t blk_mq_hw_sysfs_run_show(struct blk_mq_hw_ctx *hctx, char *page)
200 return sprintf(page, "%lu\n", hctx->run);
203 static ssize_t blk_mq_hw_sysfs_dispatched_show(struct blk_mq_hw_ctx *hctx,
206 char *start_page = page;
209 page += sprintf(page, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
211 for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
212 unsigned int d = 1U << (i - 1);
214 page += sprintf(page, "%8u\t%lu\n", d, hctx->dispatched[i]);
217 page += sprintf(page, "%8u+\t%lu\n", 1U << (i - 1),
218 hctx->dispatched[i]);
219 return page - start_page;
222 static ssize_t blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx *hctx,
227 spin_lock(&hctx->lock);
228 ret = sysfs_list_show(page, &hctx->dispatch, "HCTX pending");
229 spin_unlock(&hctx->lock);
234 static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
236 return blk_mq_tag_sysfs_show(hctx->tags, page);
239 static ssize_t blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx *hctx, char *page)
241 return sprintf(page, "%u\n", atomic_read(&hctx->nr_active));
244 static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
246 unsigned int i, first = 1;
249 for_each_cpu(i, hctx->cpumask) {
251 ret += sprintf(ret + page, "%u", i);
253 ret += sprintf(ret + page, ", %u", i);
258 ret += sprintf(ret + page, "\n");
262 static void blk_mq_stat_clear(struct blk_mq_hw_ctx *hctx)
264 struct blk_mq_ctx *ctx;
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]);
273 static ssize_t blk_mq_hw_sysfs_stat_store(struct blk_mq_hw_ctx *hctx,
274 const char *page, size_t count)
276 blk_mq_stat_clear(hctx);
280 static ssize_t print_stat(char *page, struct blk_rq_stat *stat, const char *pre)
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);
288 static ssize_t blk_mq_hw_sysfs_stat_show(struct blk_mq_hw_ctx *hctx, char *page)
290 struct blk_rq_stat stat[2];
293 blk_stat_init(&stat[BLK_STAT_READ]);
294 blk_stat_init(&stat[BLK_STAT_WRITE]);
296 blk_hctx_stat_get(hctx, stat);
298 ret = print_stat(page, &stat[BLK_STAT_READ], "read :");
299 ret += print_stat(page + ret, &stat[BLK_STAT_WRITE], "write:");
303 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = {
304 .attr = {.name = "dispatched", .mode = S_IRUGO },
305 .show = blk_mq_sysfs_dispatched_show,
307 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = {
308 .attr = {.name = "merged", .mode = S_IRUGO },
309 .show = blk_mq_sysfs_merged_show,
311 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = {
312 .attr = {.name = "completed", .mode = S_IRUGO },
313 .show = blk_mq_sysfs_completed_show,
315 static 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,
320 static 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,
328 static 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,
332 static 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,
336 static 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,
340 static 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,
344 static 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,
348 static 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,
352 static 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,
356 static 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,
361 static 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,
367 static 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,
380 static const struct sysfs_ops blk_mq_sysfs_ops = {
381 .show = blk_mq_sysfs_show,
382 .store = blk_mq_sysfs_store,
385 static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
386 .show = blk_mq_hw_sysfs_show,
387 .store = blk_mq_hw_sysfs_store,
390 static struct kobj_type blk_mq_ktype = {
391 .sysfs_ops = &blk_mq_sysfs_ops,
392 .release = blk_mq_sysfs_release,
395 static 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,
401 static 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,
407 static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx)
409 struct blk_mq_ctx *ctx;
415 hctx_for_each_ctx(hctx, ctx, i)
416 kobject_del(&ctx->kobj);
418 kobject_del(&hctx->kobj);
421 static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx)
423 struct request_queue *q = hctx->queue;
424 struct blk_mq_ctx *ctx;
430 ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", hctx->queue_num);
434 hctx_for_each_ctx(hctx, ctx, i) {
435 ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
443 static void __blk_mq_unregister_dev(struct device *dev, struct request_queue *q)
445 struct blk_mq_hw_ctx *hctx;
446 struct blk_mq_ctx *ctx;
449 queue_for_each_hw_ctx(q, hctx, i) {
450 blk_mq_unregister_hctx(hctx);
452 hctx_for_each_ctx(hctx, ctx, j)
453 kobject_put(&ctx->kobj);
455 kobject_put(&hctx->kobj);
458 kobject_uevent(&q->mq_kobj, KOBJ_REMOVE);
459 kobject_del(&q->mq_kobj);
460 kobject_put(&q->mq_kobj);
462 kobject_put(&dev->kobj);
464 q->mq_sysfs_init_done = false;
467 void blk_mq_unregister_dev(struct device *dev, struct request_queue *q)
469 blk_mq_disable_hotplug();
470 __blk_mq_unregister_dev(dev, q);
471 blk_mq_enable_hotplug();
474 void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx)
476 kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
479 static void blk_mq_sysfs_init(struct request_queue *q)
481 struct blk_mq_ctx *ctx;
484 kobject_init(&q->mq_kobj, &blk_mq_ktype);
486 for_each_possible_cpu(cpu) {
487 ctx = per_cpu_ptr(q->queue_ctx, cpu);
488 kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
492 int blk_mq_register_dev(struct device *dev, struct request_queue *q)
494 struct blk_mq_hw_ctx *hctx;
497 blk_mq_disable_hotplug();
499 blk_mq_sysfs_init(q);
501 ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
505 kobject_uevent(&q->mq_kobj, KOBJ_ADD);
507 queue_for_each_hw_ctx(q, hctx, i) {
508 ret = blk_mq_register_hctx(hctx);
514 __blk_mq_unregister_dev(dev, q);
516 q->mq_sysfs_init_done = true;
518 blk_mq_enable_hotplug();
522 EXPORT_SYMBOL_GPL(blk_mq_register_dev);
524 void blk_mq_sysfs_unregister(struct request_queue *q)
526 struct blk_mq_hw_ctx *hctx;
529 if (!q->mq_sysfs_init_done)
532 queue_for_each_hw_ctx(q, hctx, i)
533 blk_mq_unregister_hctx(hctx);
536 int blk_mq_sysfs_register(struct request_queue *q)
538 struct blk_mq_hw_ctx *hctx;
541 if (!q->mq_sysfs_init_done)
544 queue_for_each_hw_ctx(q, hctx, i) {
545 ret = blk_mq_register_hctx(hctx);