| 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 "blk.h" |
| 14 | #include "blk-mq.h" |
| 15 | |
| 16 | static void blk_mq_sysfs_release(struct kobject *kobj) |
| 17 | { |
| 18 | struct blk_mq_ctxs *ctxs = container_of(kobj, struct blk_mq_ctxs, kobj); |
| 19 | |
| 20 | free_percpu(ctxs->queue_ctx); |
| 21 | kfree(ctxs); |
| 22 | } |
| 23 | |
| 24 | static void blk_mq_ctx_sysfs_release(struct kobject *kobj) |
| 25 | { |
| 26 | struct blk_mq_ctx *ctx = container_of(kobj, struct blk_mq_ctx, kobj); |
| 27 | |
| 28 | /* ctx->ctxs won't be released until all ctx are freed */ |
| 29 | kobject_put(&ctx->ctxs->kobj); |
| 30 | } |
| 31 | |
| 32 | static void blk_mq_hw_sysfs_release(struct kobject *kobj) |
| 33 | { |
| 34 | struct blk_mq_hw_ctx *hctx = container_of(kobj, struct blk_mq_hw_ctx, |
| 35 | kobj); |
| 36 | |
| 37 | blk_free_flush_queue(hctx->fq); |
| 38 | sbitmap_free(&hctx->ctx_map); |
| 39 | free_cpumask_var(hctx->cpumask); |
| 40 | kfree(hctx->ctxs); |
| 41 | kfree(hctx); |
| 42 | } |
| 43 | |
| 44 | struct blk_mq_hw_ctx_sysfs_entry { |
| 45 | struct attribute attr; |
| 46 | ssize_t (*show)(struct blk_mq_hw_ctx *, char *); |
| 47 | }; |
| 48 | |
| 49 | static ssize_t blk_mq_hw_sysfs_show(struct kobject *kobj, |
| 50 | struct attribute *attr, char *page) |
| 51 | { |
| 52 | struct blk_mq_hw_ctx_sysfs_entry *entry; |
| 53 | struct blk_mq_hw_ctx *hctx; |
| 54 | struct request_queue *q; |
| 55 | ssize_t res; |
| 56 | |
| 57 | entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr); |
| 58 | hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj); |
| 59 | q = hctx->queue; |
| 60 | |
| 61 | if (!entry->show) |
| 62 | return -EIO; |
| 63 | |
| 64 | mutex_lock(&q->sysfs_lock); |
| 65 | res = entry->show(hctx, page); |
| 66 | mutex_unlock(&q->sysfs_lock); |
| 67 | return res; |
| 68 | } |
| 69 | |
| 70 | static ssize_t blk_mq_hw_sysfs_nr_tags_show(struct blk_mq_hw_ctx *hctx, |
| 71 | char *page) |
| 72 | { |
| 73 | return sprintf(page, "%u\n", hctx->tags->nr_tags); |
| 74 | } |
| 75 | |
| 76 | static ssize_t blk_mq_hw_sysfs_nr_reserved_tags_show(struct blk_mq_hw_ctx *hctx, |
| 77 | char *page) |
| 78 | { |
| 79 | return sprintf(page, "%u\n", hctx->tags->nr_reserved_tags); |
| 80 | } |
| 81 | |
| 82 | static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page) |
| 83 | { |
| 84 | const size_t size = PAGE_SIZE - 1; |
| 85 | unsigned int i, first = 1; |
| 86 | int ret = 0, pos = 0; |
| 87 | |
| 88 | for_each_cpu(i, hctx->cpumask) { |
| 89 | if (first) |
| 90 | ret = snprintf(pos + page, size - pos, "%u", i); |
| 91 | else |
| 92 | ret = snprintf(pos + page, size - pos, ", %u", i); |
| 93 | |
| 94 | if (ret >= size - pos) |
| 95 | break; |
| 96 | |
| 97 | first = 0; |
| 98 | pos += ret; |
| 99 | } |
| 100 | |
| 101 | ret = snprintf(pos + page, size + 1 - pos, "\n"); |
| 102 | return pos + ret; |
| 103 | } |
| 104 | |
| 105 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_nr_tags = { |
| 106 | .attr = {.name = "nr_tags", .mode = 0444 }, |
| 107 | .show = blk_mq_hw_sysfs_nr_tags_show, |
| 108 | }; |
| 109 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_nr_reserved_tags = { |
| 110 | .attr = {.name = "nr_reserved_tags", .mode = 0444 }, |
| 111 | .show = blk_mq_hw_sysfs_nr_reserved_tags_show, |
| 112 | }; |
| 113 | static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = { |
| 114 | .attr = {.name = "cpu_list", .mode = 0444 }, |
| 115 | .show = blk_mq_hw_sysfs_cpus_show, |
| 116 | }; |
| 117 | |
| 118 | static struct attribute *default_hw_ctx_attrs[] = { |
| 119 | &blk_mq_hw_sysfs_nr_tags.attr, |
| 120 | &blk_mq_hw_sysfs_nr_reserved_tags.attr, |
| 121 | &blk_mq_hw_sysfs_cpus.attr, |
| 122 | NULL, |
| 123 | }; |
| 124 | ATTRIBUTE_GROUPS(default_hw_ctx); |
| 125 | |
| 126 | static const struct sysfs_ops blk_mq_hw_sysfs_ops = { |
| 127 | .show = blk_mq_hw_sysfs_show, |
| 128 | }; |
| 129 | |
| 130 | static const struct kobj_type blk_mq_ktype = { |
| 131 | .release = blk_mq_sysfs_release, |
| 132 | }; |
| 133 | |
| 134 | static const struct kobj_type blk_mq_ctx_ktype = { |
| 135 | .release = blk_mq_ctx_sysfs_release, |
| 136 | }; |
| 137 | |
| 138 | static const struct kobj_type blk_mq_hw_ktype = { |
| 139 | .sysfs_ops = &blk_mq_hw_sysfs_ops, |
| 140 | .default_groups = default_hw_ctx_groups, |
| 141 | .release = blk_mq_hw_sysfs_release, |
| 142 | }; |
| 143 | |
| 144 | static void blk_mq_unregister_hctx(struct blk_mq_hw_ctx *hctx) |
| 145 | { |
| 146 | struct blk_mq_ctx *ctx; |
| 147 | int i; |
| 148 | |
| 149 | if (!hctx->nr_ctx) |
| 150 | return; |
| 151 | |
| 152 | hctx_for_each_ctx(hctx, ctx, i) |
| 153 | kobject_del(&ctx->kobj); |
| 154 | |
| 155 | kobject_del(&hctx->kobj); |
| 156 | } |
| 157 | |
| 158 | static int blk_mq_register_hctx(struct blk_mq_hw_ctx *hctx) |
| 159 | { |
| 160 | struct request_queue *q = hctx->queue; |
| 161 | struct blk_mq_ctx *ctx; |
| 162 | int i, j, ret; |
| 163 | |
| 164 | if (!hctx->nr_ctx) |
| 165 | return 0; |
| 166 | |
| 167 | ret = kobject_add(&hctx->kobj, q->mq_kobj, "%u", hctx->queue_num); |
| 168 | if (ret) |
| 169 | return ret; |
| 170 | |
| 171 | hctx_for_each_ctx(hctx, ctx, i) { |
| 172 | ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu); |
| 173 | if (ret) |
| 174 | goto out; |
| 175 | } |
| 176 | |
| 177 | return 0; |
| 178 | out: |
| 179 | hctx_for_each_ctx(hctx, ctx, j) { |
| 180 | if (j < i) |
| 181 | kobject_del(&ctx->kobj); |
| 182 | } |
| 183 | kobject_del(&hctx->kobj); |
| 184 | return ret; |
| 185 | } |
| 186 | |
| 187 | void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx) |
| 188 | { |
| 189 | kobject_init(&hctx->kobj, &blk_mq_hw_ktype); |
| 190 | } |
| 191 | |
| 192 | void blk_mq_sysfs_deinit(struct request_queue *q) |
| 193 | { |
| 194 | struct blk_mq_ctx *ctx; |
| 195 | int cpu; |
| 196 | |
| 197 | for_each_possible_cpu(cpu) { |
| 198 | ctx = per_cpu_ptr(q->queue_ctx, cpu); |
| 199 | kobject_put(&ctx->kobj); |
| 200 | } |
| 201 | kobject_put(q->mq_kobj); |
| 202 | } |
| 203 | |
| 204 | void blk_mq_sysfs_init(struct request_queue *q) |
| 205 | { |
| 206 | struct blk_mq_ctx *ctx; |
| 207 | int cpu; |
| 208 | |
| 209 | kobject_init(q->mq_kobj, &blk_mq_ktype); |
| 210 | |
| 211 | for_each_possible_cpu(cpu) { |
| 212 | ctx = per_cpu_ptr(q->queue_ctx, cpu); |
| 213 | |
| 214 | kobject_get(q->mq_kobj); |
| 215 | kobject_init(&ctx->kobj, &blk_mq_ctx_ktype); |
| 216 | } |
| 217 | } |
| 218 | |
| 219 | int blk_mq_sysfs_register(struct gendisk *disk) |
| 220 | { |
| 221 | struct request_queue *q = disk->queue; |
| 222 | struct blk_mq_hw_ctx *hctx; |
| 223 | unsigned long i, j; |
| 224 | int ret; |
| 225 | |
| 226 | lockdep_assert_held(&q->sysfs_dir_lock); |
| 227 | |
| 228 | ret = kobject_add(q->mq_kobj, &disk_to_dev(disk)->kobj, "mq"); |
| 229 | if (ret < 0) |
| 230 | goto out; |
| 231 | |
| 232 | kobject_uevent(q->mq_kobj, KOBJ_ADD); |
| 233 | |
| 234 | queue_for_each_hw_ctx(q, hctx, i) { |
| 235 | ret = blk_mq_register_hctx(hctx); |
| 236 | if (ret) |
| 237 | goto unreg; |
| 238 | } |
| 239 | |
| 240 | q->mq_sysfs_init_done = true; |
| 241 | |
| 242 | out: |
| 243 | return ret; |
| 244 | |
| 245 | unreg: |
| 246 | queue_for_each_hw_ctx(q, hctx, j) { |
| 247 | if (j < i) |
| 248 | blk_mq_unregister_hctx(hctx); |
| 249 | } |
| 250 | |
| 251 | kobject_uevent(q->mq_kobj, KOBJ_REMOVE); |
| 252 | kobject_del(q->mq_kobj); |
| 253 | return ret; |
| 254 | } |
| 255 | |
| 256 | void blk_mq_sysfs_unregister(struct gendisk *disk) |
| 257 | { |
| 258 | struct request_queue *q = disk->queue; |
| 259 | struct blk_mq_hw_ctx *hctx; |
| 260 | unsigned long i; |
| 261 | |
| 262 | lockdep_assert_held(&q->sysfs_dir_lock); |
| 263 | |
| 264 | queue_for_each_hw_ctx(q, hctx, i) |
| 265 | blk_mq_unregister_hctx(hctx); |
| 266 | |
| 267 | kobject_uevent(q->mq_kobj, KOBJ_REMOVE); |
| 268 | kobject_del(q->mq_kobj); |
| 269 | |
| 270 | q->mq_sysfs_init_done = false; |
| 271 | } |
| 272 | |
| 273 | void blk_mq_sysfs_unregister_hctxs(struct request_queue *q) |
| 274 | { |
| 275 | struct blk_mq_hw_ctx *hctx; |
| 276 | unsigned long i; |
| 277 | |
| 278 | mutex_lock(&q->sysfs_dir_lock); |
| 279 | if (!q->mq_sysfs_init_done) |
| 280 | goto unlock; |
| 281 | |
| 282 | queue_for_each_hw_ctx(q, hctx, i) |
| 283 | blk_mq_unregister_hctx(hctx); |
| 284 | |
| 285 | unlock: |
| 286 | mutex_unlock(&q->sysfs_dir_lock); |
| 287 | } |
| 288 | |
| 289 | int blk_mq_sysfs_register_hctxs(struct request_queue *q) |
| 290 | { |
| 291 | struct blk_mq_hw_ctx *hctx; |
| 292 | unsigned long i; |
| 293 | int ret = 0; |
| 294 | |
| 295 | mutex_lock(&q->sysfs_dir_lock); |
| 296 | if (!q->mq_sysfs_init_done) |
| 297 | goto unlock; |
| 298 | |
| 299 | queue_for_each_hw_ctx(q, hctx, i) { |
| 300 | ret = blk_mq_register_hctx(hctx); |
| 301 | if (ret) |
| 302 | break; |
| 303 | } |
| 304 | |
| 305 | unlock: |
| 306 | mutex_unlock(&q->sysfs_dir_lock); |
| 307 | |
| 308 | return ret; |
| 309 | } |