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 16 static void blk_mq_sysfs_release(struct kobject *kobj) 17 { 18 } 19 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); 24 }; 25 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); 30 }; 31 32 static 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 55 static 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 78 static 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 } 100 101 static 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 } 124 125 static 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 } 130 131 static 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 } 135 136 static 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 } 141 142 static 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 } 165 166 static 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 } 176 177 static ssize_t blk_mq_hw_sysfs_poll_show(struct blk_mq_hw_ctx *hctx, char *page) 178 { 179 return sprintf(page, "invoked=%lu, success=%lu\n", hctx->poll_invoked, hctx->poll_success); 180 } 181 182 static ssize_t blk_mq_hw_sysfs_queued_show(struct blk_mq_hw_ctx *hctx, 183 char *page) 184 { 185 return sprintf(page, "%lu\n", hctx->queued); 186 } 187 188 static ssize_t blk_mq_hw_sysfs_run_show(struct blk_mq_hw_ctx *hctx, char *page) 189 { 190 return sprintf(page, "%lu\n", hctx->run); 191 } 192 193 static ssize_t blk_mq_hw_sysfs_dispatched_show(struct blk_mq_hw_ctx *hctx, 194 char *page) 195 { 196 char *start_page = page; 197 int i; 198 199 page += sprintf(page, "%8u\t%lu\n", 0U, hctx->dispatched[0]); 200 201 for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER; i++) { 202 unsigned long d = 1U << (i - 1); 203 204 page += sprintf(page, "%8lu\t%lu\n", d, hctx->dispatched[i]); 205 } 206 207 return page - start_page; 208 } 209 210 static ssize_t blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx *hctx, 211 char *page) 212 { 213 ssize_t ret; 214 215 spin_lock(&hctx->lock); 216 ret = sysfs_list_show(page, &hctx->dispatch, "HCTX pending"); 217 spin_unlock(&hctx->lock); 218 219 return ret; 220 } 221 222 static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page) 223 { 224 return blk_mq_tag_sysfs_show(hctx->tags, page); 225 } 226 227 static ssize_t blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx *hctx, char *page) 228 { 229 return sprintf(page, "%u\n", atomic_read(&hctx->nr_active)); 230 } 231 232 static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page) 233 { 234 unsigned int i, first = 1; 235 ssize_t ret = 0; 236 237 for_each_cpu(i, hctx->cpumask) { 238 if (first) 239 ret += sprintf(ret + page, "%u", i); 240 else 241 ret += sprintf(ret + page, ", %u", i); 242 243 first = 0; 244 } 245 246 ret += sprintf(ret + page, "\n"); 247 return ret; 248 } 249 250 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = { 251 .attr = {.name = "dispatched", .mode = S_IRUGO }, 252 .show = blk_mq_sysfs_dispatched_show, 253 }; 254 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = { 255 .attr = {.name = "merged", .mode = S_IRUGO }, 256 .show = blk_mq_sysfs_merged_show, 257 }; 258 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = { 259 .attr = {.name = "completed", .mode = S_IRUGO }, 260 .show = blk_mq_sysfs_completed_show, 261 }; 262 static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = { 263 .attr = {.name = "rq_list", .mode = S_IRUGO }, 264 .show = blk_mq_sysfs_rq_list_show, 265 }; 266 267 static struct attribute *default_ctx_attrs[] = { 268 &blk_mq_sysfs_dispatched.attr, 269 &blk_mq_sysfs_merged.attr, 270 &blk_mq_sysfs_completed.attr, 271 &blk_mq_sysfs_rq_list.attr, 272 NULL, 273 }; 274 275 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = { 276 .attr = {.name = "queued", .mode = S_IRUGO }, 277 .show = blk_mq_hw_sysfs_queued_show, 278 }; 279 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = { 280 .attr = {.name = "run", .mode = S_IRUGO }, 281 .show = blk_mq_hw_sysfs_run_show, 282 }; 283 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = { 284 .attr = {.name = "dispatched", .mode = S_IRUGO }, 285 .show = blk_mq_hw_sysfs_dispatched_show, 286 }; 287 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_active = { 288 .attr = {.name = "active", .mode = S_IRUGO }, 289 .show = blk_mq_hw_sysfs_active_show, 290 }; 291 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = { 292 .attr = {.name = "pending", .mode = S_IRUGO }, 293 .show = blk_mq_hw_sysfs_rq_list_show, 294 }; 295 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = { 296 .attr = {.name = "tags", .mode = S_IRUGO }, 297 .show = blk_mq_hw_sysfs_tags_show, 298 }; 299 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = { 300 .attr = {.name = "cpu_list", .mode = S_IRUGO }, 301 .show = blk_mq_hw_sysfs_cpus_show, 302 }; 303 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_poll = { 304 .attr = {.name = "io_poll", .mode = S_IRUGO }, 305 .show = blk_mq_hw_sysfs_poll_show, 306 }; 307 308 static struct attribute *default_hw_ctx_attrs[] = { 309 &blk_mq_hw_sysfs_queued.attr, 310 &blk_mq_hw_sysfs_run.attr, 311 &blk_mq_hw_sysfs_dispatched.attr, 312 &blk_mq_hw_sysfs_pending.attr, 313 &blk_mq_hw_sysfs_tags.attr, 314 &blk_mq_hw_sysfs_cpus.attr, 315 &blk_mq_hw_sysfs_active.attr, 316 &blk_mq_hw_sysfs_poll.attr, 317 NULL, 318 }; 319 320 static const struct sysfs_ops blk_mq_sysfs_ops = { 321 .show = blk_mq_sysfs_show, 322 .store = blk_mq_sysfs_store, 323 }; 324 325 static const struct sysfs_ops blk_mq_hw_sysfs_ops = { 326 .show = blk_mq_hw_sysfs_show, 327 .store = blk_mq_hw_sysfs_store, 328 }; 329 330 static struct kobj_type blk_mq_ktype = { 331 .sysfs_ops = &blk_mq_sysfs_ops, 332 .release = blk_mq_sysfs_release, 333 }; 334 335 static struct kobj_type blk_mq_ctx_ktype = { 336 .sysfs_ops = &blk_mq_sysfs_ops, 337 .default_attrs = default_ctx_attrs, 338 .release = blk_mq_sysfs_release, 339 }; 340 341 static struct kobj_type blk_mq_hw_ktype = { 342 .sysfs_ops = &blk_mq_hw_sysfs_ops, 343 .default_attrs = default_hw_ctx_attrs, 344 .release = blk_mq_sysfs_release, 345 }; 346 347 static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx) 348 { 349 struct blk_mq_ctx *ctx; 350 int i; 351 352 if (!hctx->nr_ctx) 353 return; 354 355 hctx_for_each_ctx(hctx, ctx, i) 356 kobject_del(&ctx->kobj); 357 358 kobject_del(&hctx->kobj); 359 } 360 361 static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx) 362 { 363 struct request_queue *q = hctx->queue; 364 struct blk_mq_ctx *ctx; 365 int i, ret; 366 367 if (!hctx->nr_ctx) 368 return 0; 369 370 ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", hctx->queue_num); 371 if (ret) 372 return ret; 373 374 hctx_for_each_ctx(hctx, ctx, i) { 375 ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu); 376 if (ret) 377 break; 378 } 379 380 return ret; 381 } 382 383 static void __blk_mq_unregister_disk(struct gendisk *disk) 384 { 385 struct request_queue *q = disk->queue; 386 struct blk_mq_hw_ctx *hctx; 387 struct blk_mq_ctx *ctx; 388 int i, j; 389 390 queue_for_each_hw_ctx(q, hctx, i) { 391 blk_mq_unregister_hctx(hctx); 392 393 hctx_for_each_ctx(hctx, ctx, j) 394 kobject_put(&ctx->kobj); 395 396 kobject_put(&hctx->kobj); 397 } 398 399 kobject_uevent(&q->mq_kobj, KOBJ_REMOVE); 400 kobject_del(&q->mq_kobj); 401 kobject_put(&q->mq_kobj); 402 403 kobject_put(&disk_to_dev(disk)->kobj); 404 405 q->mq_sysfs_init_done = false; 406 } 407 408 void blk_mq_unregister_disk(struct gendisk *disk) 409 { 410 blk_mq_disable_hotplug(); 411 __blk_mq_unregister_disk(disk); 412 blk_mq_enable_hotplug(); 413 } 414 415 void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx) 416 { 417 kobject_init(&hctx->kobj, &blk_mq_hw_ktype); 418 } 419 420 static void blk_mq_sysfs_init(struct request_queue *q) 421 { 422 struct blk_mq_ctx *ctx; 423 int cpu; 424 425 kobject_init(&q->mq_kobj, &blk_mq_ktype); 426 427 for_each_possible_cpu(cpu) { 428 ctx = per_cpu_ptr(q->queue_ctx, cpu); 429 kobject_init(&ctx->kobj, &blk_mq_ctx_ktype); 430 } 431 } 432 433 int blk_mq_register_disk(struct gendisk *disk) 434 { 435 struct device *dev = disk_to_dev(disk); 436 struct request_queue *q = disk->queue; 437 struct blk_mq_hw_ctx *hctx; 438 int ret, i; 439 440 blk_mq_disable_hotplug(); 441 442 blk_mq_sysfs_init(q); 443 444 ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq"); 445 if (ret < 0) 446 goto out; 447 448 kobject_uevent(&q->mq_kobj, KOBJ_ADD); 449 450 queue_for_each_hw_ctx(q, hctx, i) { 451 ret = blk_mq_register_hctx(hctx); 452 if (ret) 453 break; 454 } 455 456 if (ret) 457 __blk_mq_unregister_disk(disk); 458 else 459 q->mq_sysfs_init_done = true; 460 out: 461 blk_mq_enable_hotplug(); 462 463 return ret; 464 } 465 EXPORT_SYMBOL_GPL(blk_mq_register_disk); 466 467 void blk_mq_sysfs_unregister(struct request_queue *q) 468 { 469 struct blk_mq_hw_ctx *hctx; 470 int i; 471 472 if (!q->mq_sysfs_init_done) 473 return; 474 475 queue_for_each_hw_ctx(q, hctx, i) 476 blk_mq_unregister_hctx(hctx); 477 } 478 479 int blk_mq_sysfs_register(struct request_queue *q) 480 { 481 struct blk_mq_hw_ctx *hctx; 482 int i, ret = 0; 483 484 if (!q->mq_sysfs_init_done) 485 return ret; 486 487 queue_for_each_hw_ctx(q, hctx, i) { 488 ret = blk_mq_register_hctx(hctx); 489 if (ret) 490 break; 491 } 492 493 return ret; 494 } 495