drivers/block/null_blk.c

   1 /*
   2  * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
   3  * Shaohua Li <shli@fb.com>
   4  */
   5 #include <linux/module.h>
   6
   7 #include <linux/moduleparam.h>
   8 #include <linux/sched.h>
   9 #include <linux/fs.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/init.h>
  12 #include <linux/slab.h>
  13 #include <linux/blk-mq.h>
  14 #include <linux/hrtimer.h>
  15 #include <linux/lightnvm.h>
  16 #include <linux/configfs.h>
  17 #include <linux/badblocks.h>
  18
  19 #define SECTOR_SHIFT            9
  20 #define PAGE_SECTORS_SHIFT      (PAGE_SHIFT - SECTOR_SHIFT)
  21 #define PAGE_SECTORS            (1 << PAGE_SECTORS_SHIFT)
  22 #define SECTOR_SIZE             (1 << SECTOR_SHIFT)
  23 #define SECTOR_MASK             (PAGE_SECTORS - 1)
  24
  25 #define FREE_BATCH              16
  26
  27 #define TICKS_PER_SEC           50ULL
  28 #define TIMER_INTERVAL          (NSEC_PER_SEC / TICKS_PER_SEC)
  29
  30 static inline u64 mb_per_tick(int mbps)
  31 {
  32         return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
  33 }
  34
  35 struct nullb_cmd {
  36         struct list_head list;
  37         struct llist_node ll_list;
  38         call_single_data_t csd;
  39         struct request *rq;
  40         struct bio *bio;
  41         unsigned int tag;
  42         struct nullb_queue *nq;
  43         struct hrtimer timer;
  44         blk_status_t error;
  45 };
  46
  47 struct nullb_queue {
  48         unsigned long *tag_map;
  49         wait_queue_head_t wait;
  50         unsigned int queue_depth;
  51         struct nullb_device *dev;
  52
  53         struct nullb_cmd *cmds;
  54 };
  55
  56 /*
  57  * Status flags for nullb_device.
  58  *
  59  * CONFIGURED:  Device has been configured and turned on. Cannot reconfigure.
  60  * UP:          Device is currently on and visible in userspace.
  61  * THROTTLED:   Device is being throttled.
  62  * CACHE:       Device is using a write-back cache.
  63  */
  64 enum nullb_device_flags {
  65         NULLB_DEV_FL_CONFIGURED = 0,
  66         NULLB_DEV_FL_UP         = 1,
  67         NULLB_DEV_FL_THROTTLED  = 2,
  68         NULLB_DEV_FL_CACHE      = 3,
  69 };
  70
  71 /*
  72  * nullb_page is a page in memory for nullb devices.
  73  *
  74  * @page:       The page holding the data.
  75  * @bitmap:     The bitmap represents which sector in the page has data.
  76  *              Each bit represents one block size. For example, sector 8
  77  *              will use the 7th bit
  78  * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
  79  * page is being flushing to storage. FREE means the cache page is freed and
  80  * should be skipped from flushing to storage. Please see
  81  * null_make_cache_space
  82  */
  83 struct nullb_page {
  84         struct page *page;
  85         unsigned long bitmap;
  86 };
  87 #define NULLB_PAGE_LOCK (sizeof(unsigned long) * 8 - 1)
  88 #define NULLB_PAGE_FREE (sizeof(unsigned long) * 8 - 2)
  89
  90 struct nullb_device {
  91         struct nullb *nullb;
  92         struct config_item item;
  93         struct radix_tree_root data; /* data stored in the disk */
  94         struct radix_tree_root cache; /* disk cache data */
  95         unsigned long flags; /* device flags */
  96         unsigned int curr_cache;
  97         struct badblocks badblocks;
  98
  99         unsigned long size; /* device size in MB */
 100         unsigned long completion_nsec; /* time in ns to complete a request */
 101         unsigned long cache_size; /* disk cache size in MB */
 102         unsigned int submit_queues; /* number of submission queues */
 103         unsigned int home_node; /* home node for the device */
 104         unsigned int queue_mode; /* block interface */
 105         unsigned int blocksize; /* block size */
 106         unsigned int irqmode; /* IRQ completion handler */
 107         unsigned int hw_queue_depth; /* queue depth */
 108         unsigned int index; /* index of the disk, only valid with a disk */
 109         unsigned int mbps; /* Bandwidth throttle cap (in MB/s) */
 110         bool use_lightnvm; /* register as a LightNVM device */
 111         bool blocking; /* blocking blk-mq device */
 112         bool use_per_node_hctx; /* use per-node allocation for hardware context */
 113         bool power; /* power on/off the device */
 114         bool memory_backed; /* if data is stored in memory */
 115         bool discard; /* if support discard */
 116 };
 117
 118 struct nullb {
 119         struct nullb_device *dev;
 120         struct list_head list;
 121         unsigned int index;
 122         struct request_queue *q;
 123         struct gendisk *disk;
 124         struct nvm_dev *ndev;
 125         struct blk_mq_tag_set *tag_set;
 126         struct blk_mq_tag_set __tag_set;
 127         unsigned int queue_depth;
 128         atomic_long_t cur_bytes;
 129         struct hrtimer bw_timer;
 130         unsigned long cache_flush_pos;
 131         spinlock_t lock;
 132
 133         struct nullb_queue *queues;
 134         unsigned int nr_queues;
 135         char disk_name[DISK_NAME_LEN];
 136 };
 137
 138 static LIST_HEAD(nullb_list);
 139 static struct mutex lock;
 140 static int null_major;
 141 static DEFINE_IDA(nullb_indexes);
 142 static struct kmem_cache *ppa_cache;
 143 static struct blk_mq_tag_set tag_set;
 144
 145 enum {
 146         NULL_IRQ_NONE           = 0,
 147         NULL_IRQ_SOFTIRQ        = 1,
 148         NULL_IRQ_TIMER          = 2,
 149 };
 150
 151 enum {
 152         NULL_Q_BIO              = 0,
 153         NULL_Q_RQ               = 1,
 154         NULL_Q_MQ               = 2,
 155 };
 156
 157 static int g_submit_queues = 1;
 158 module_param_named(submit_queues, g_submit_queues, int, S_IRUGO);
 159 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
 160
 161 static int g_home_node = NUMA_NO_NODE;
 162 module_param_named(home_node, g_home_node, int, S_IRUGO);
 163 MODULE_PARM_DESC(home_node, "Home node for the device");
 164
 165 static int g_queue_mode = NULL_Q_MQ;
 166
 167 static int null_param_store_val(const char *str, int *val, int min, int max)
 168 {
 169         int ret, new_val;
 170
 171         ret = kstrtoint(str, 10, &new_val);
 172         if (ret)
 173                 return -EINVAL;
 174
 175         if (new_val < min || new_val > max)
 176                 return -EINVAL;
 177
 178         *val = new_val;
 179         return 0;
 180 }
 181
 182 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
 183 {
 184         return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
 185 }
 186
 187 static const struct kernel_param_ops null_queue_mode_param_ops = {
 188         .set    = null_set_queue_mode,
 189         .get    = param_get_int,
 190 };
 191
 192 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, S_IRUGO);
 193 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
 194
 195 static int g_gb = 250;
 196 module_param_named(gb, g_gb, int, S_IRUGO);
 197 MODULE_PARM_DESC(gb, "Size in GB");
 198
 199 static int g_bs = 512;
 200 module_param_named(bs, g_bs, int, S_IRUGO);
 201 MODULE_PARM_DESC(bs, "Block size (in bytes)");
 202
 203 static int nr_devices = 1;
 204 module_param(nr_devices, int, S_IRUGO);
 205 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
 206
 207 static bool g_use_lightnvm;
 208 module_param_named(use_lightnvm, g_use_lightnvm, bool, S_IRUGO);
 209 MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
 210
 211 static bool g_blocking;
 212 module_param_named(blocking, g_blocking, bool, S_IRUGO);
 213 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
 214
 215 static bool shared_tags;
 216 module_param(shared_tags, bool, S_IRUGO);
 217 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
 218
 219 static int g_irqmode = NULL_IRQ_SOFTIRQ;
 220
 221 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
 222 {
 223         return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
 224                                         NULL_IRQ_TIMER);
 225 }
 226
 227 static const struct kernel_param_ops null_irqmode_param_ops = {
 228         .set    = null_set_irqmode,
 229         .get    = param_get_int,
 230 };
 231
 232 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, S_IRUGO);
 233 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
 234
 235 static unsigned long g_completion_nsec = 10000;
 236 module_param_named(completion_nsec, g_completion_nsec, ulong, S_IRUGO);
 237 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
 238
 239 static int g_hw_queue_depth = 64;
 240 module_param_named(hw_queue_depth, g_hw_queue_depth, int, S_IRUGO);
 241 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
 242
 243 static bool g_use_per_node_hctx;
 244 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, S_IRUGO);
 245 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
 246
 247 static struct nullb_device *null_alloc_dev(void);
 248 static void null_free_dev(struct nullb_device *dev);
 249 static void null_del_dev(struct nullb *nullb);
 250 static int null_add_dev(struct nullb_device *dev);
 251 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
 252
 253 static inline struct nullb_device *to_nullb_device(struct config_item *item)
 254 {
 255         return item ? container_of(item, struct nullb_device, item) : NULL;
 256 }
 257
 258 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
 259 {
 260         return snprintf(page, PAGE_SIZE, "%u\n", val);
 261 }
 262
 263 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
 264         char *page)
 265 {
 266         return snprintf(page, PAGE_SIZE, "%lu\n", val);
 267 }
 268
 269 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
 270 {
 271         return snprintf(page, PAGE_SIZE, "%u\n", val);
 272 }
 273
 274 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
 275         const char *page, size_t count)
 276 {
 277         unsigned int tmp;
 278         int result;
 279
 280         result = kstrtouint(page, 0, &tmp);
 281         if (result)
 282                 return result;
 283
 284         *val = tmp;
 285         return count;
 286 }
 287
 288 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
 289         const char *page, size_t count)
 290 {
 291         int result;
 292         unsigned long tmp;
 293
 294         result = kstrtoul(page, 0, &tmp);
 295         if (result)
 296                 return result;
 297
 298         *val = tmp;
 299         return count;
 300 }
 301
 302 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
 303         size_t count)
 304 {
 305         bool tmp;
 306         int result;
 307
 308         result = kstrtobool(page,  &tmp);
 309         if (result)
 310                 return result;
 311
 312         *val = tmp;
 313         return count;
 314 }
 315
 316 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
 317 #define NULLB_DEVICE_ATTR(NAME, TYPE)                                           \
 318 static ssize_t                                                                  \
 319 nullb_device_##NAME##_show(struct config_item *item, char *page)                \
 320 {                                                                               \
 321         return nullb_device_##TYPE##_attr_show(                                 \
 322                                 to_nullb_device(item)->NAME, page);             \
 323 }                                                                               \
 324 static ssize_t                                                                  \
 325 nullb_device_##NAME##_store(struct config_item *item, const char *page,         \
 326                             size_t count)                                       \
 327 {                                                                               \
 328         if (test_bit(NULLB_DEV_FL_CONFIGURED, &to_nullb_device(item)->flags))   \
 329                 return -EBUSY;                                                  \
 330         return nullb_device_##TYPE##_attr_store(                                \
 331                         &to_nullb_device(item)->NAME, page, count);             \
 332 }                                                                               \
 333 CONFIGFS_ATTR(nullb_device_, NAME);
 334
 335 NULLB_DEVICE_ATTR(size, ulong);
 336 NULLB_DEVICE_ATTR(completion_nsec, ulong);
 337 NULLB_DEVICE_ATTR(submit_queues, uint);
 338 NULLB_DEVICE_ATTR(home_node, uint);
 339 NULLB_DEVICE_ATTR(queue_mode, uint);
 340 NULLB_DEVICE_ATTR(blocksize, uint);
 341 NULLB_DEVICE_ATTR(irqmode, uint);
 342 NULLB_DEVICE_ATTR(hw_queue_depth, uint);
 343 NULLB_DEVICE_ATTR(index, uint);
 344 NULLB_DEVICE_ATTR(use_lightnvm, bool);
 345 NULLB_DEVICE_ATTR(blocking, bool);
 346 NULLB_DEVICE_ATTR(use_per_node_hctx, bool);
 347 NULLB_DEVICE_ATTR(memory_backed, bool);
 348 NULLB_DEVICE_ATTR(discard, bool);
 349 NULLB_DEVICE_ATTR(mbps, uint);
 350 NULLB_DEVICE_ATTR(cache_size, ulong);
 351
 352 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
 353 {
 354         return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
 355 }
 356
 357 static ssize_t nullb_device_power_store(struct config_item *item,
 358                                      const char *page, size_t count)
 359 {
 360         struct nullb_device *dev = to_nullb_device(item);
 361         bool newp = false;
 362         ssize_t ret;
 363
 364         ret = nullb_device_bool_attr_store(&newp, page, count);
 365         if (ret < 0)
 366                 return ret;
 367
 368         if (!dev->power && newp) {
 369                 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
 370                         return count;
 371                 if (null_add_dev(dev)) {
 372                         clear_bit(NULLB_DEV_FL_UP, &dev->flags);
 373                         return -ENOMEM;
 374                 }
 375
 376                 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 377                 dev->power = newp;
 378         } else if (dev->power && !newp) {
 379                 mutex_lock(&lock);
 380                 dev->power = newp;
 381                 null_del_dev(dev->nullb);
 382                 mutex_unlock(&lock);
 383                 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
 384         }
 385
 386         return count;
 387 }
 388
 389 CONFIGFS_ATTR(nullb_device_, power);
 390
 391 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
 392 {
 393         struct nullb_device *t_dev = to_nullb_device(item);
 394
 395         return badblocks_show(&t_dev->badblocks, page, 0);
 396 }
 397
 398 static ssize_t nullb_device_badblocks_store(struct config_item *item,
 399                                      const char *page, size_t count)
 400 {
 401         struct nullb_device *t_dev = to_nullb_device(item);
 402         char *orig, *buf, *tmp;
 403         u64 start, end;
 404         int ret;
 405
 406         orig = kstrndup(page, count, GFP_KERNEL);
 407         if (!orig)
 408                 return -ENOMEM;
 409
 410         buf = strstrip(orig);
 411
 412         ret = -EINVAL;
 413         if (buf[0] != '+' && buf[0] != '-')
 414                 goto out;
 415         tmp = strchr(&buf[1], '-');
 416         if (!tmp)
 417                 goto out;
 418         *tmp = '\0';
 419         ret = kstrtoull(buf + 1, 0, &start);
 420         if (ret)
 421                 goto out;
 422         ret = kstrtoull(tmp + 1, 0, &end);
 423         if (ret)
 424                 goto out;
 425         ret = -EINVAL;
 426         if (start > end)
 427                 goto out;
 428         /* enable badblocks */
 429         cmpxchg(&t_dev->badblocks.shift, -1, 0);
 430         if (buf[0] == '+')
 431                 ret = badblocks_set(&t_dev->badblocks, start,
 432                         end - start + 1, 1);
 433         else
 434                 ret = badblocks_clear(&t_dev->badblocks, start,
 435                         end - start + 1);
 436         if (ret == 0)
 437                 ret = count;
 438 out:
 439         kfree(orig);
 440         return ret;
 441 }
 442 CONFIGFS_ATTR(nullb_device_, badblocks);
 443
 444 static struct configfs_attribute *nullb_device_attrs[] = {
 445         &nullb_device_attr_size,
 446         &nullb_device_attr_completion_nsec,
 447         &nullb_device_attr_submit_queues,
 448         &nullb_device_attr_home_node,
 449         &nullb_device_attr_queue_mode,
 450         &nullb_device_attr_blocksize,
 451         &nullb_device_attr_irqmode,
 452         &nullb_device_attr_hw_queue_depth,
 453         &nullb_device_attr_index,
 454         &nullb_device_attr_use_lightnvm,
 455         &nullb_device_attr_blocking,
 456         &nullb_device_attr_use_per_node_hctx,
 457         &nullb_device_attr_power,
 458         &nullb_device_attr_memory_backed,
 459         &nullb_device_attr_discard,
 460         &nullb_device_attr_mbps,
 461         &nullb_device_attr_cache_size,
 462         &nullb_device_attr_badblocks,
 463         NULL,
 464 };
 465
 466 static void nullb_device_release(struct config_item *item)
 467 {
 468         struct nullb_device *dev = to_nullb_device(item);
 469
 470         badblocks_exit(&dev->badblocks);
 471         null_free_device_storage(dev, false);
 472         null_free_dev(dev);
 473 }
 474
 475 static struct configfs_item_operations nullb_device_ops = {
 476         .release        = nullb_device_release,
 477 };
 478
 479 static struct config_item_type nullb_device_type = {
 480         .ct_item_ops    = &nullb_device_ops,
 481         .ct_attrs       = nullb_device_attrs,
 482         .ct_owner       = THIS_MODULE,
 483 };
 484
 485 static struct
 486 config_item *nullb_group_make_item(struct config_group *group, const char *name)
 487 {
 488         struct nullb_device *dev;
 489
 490         dev = null_alloc_dev();
 491         if (!dev)
 492                 return ERR_PTR(-ENOMEM);
 493
 494         config_item_init_type_name(&dev->item, name, &nullb_device_type);
 495
 496         return &dev->item;
 497 }
 498
 499 static void
 500 nullb_group_drop_item(struct config_group *group, struct config_item *item)
 501 {
 502         struct nullb_device *dev = to_nullb_device(item);
 503
 504         if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 505                 mutex_lock(&lock);
 506                 dev->power = false;
 507                 null_del_dev(dev->nullb);
 508                 mutex_unlock(&lock);
 509         }
 510
 511         config_item_put(item);
 512 }
 513
 514 static ssize_t memb_group_features_show(struct config_item *item, char *page)
 515 {
 516         return snprintf(page, PAGE_SIZE, "memory_backed,discard,bandwidth,cache,badblocks\n");
 517 }
 518
 519 CONFIGFS_ATTR_RO(memb_group_, features);
 520
 521 static struct configfs_attribute *nullb_group_attrs[] = {
 522         &memb_group_attr_features,
 523         NULL,
 524 };
 525
 526 static struct configfs_group_operations nullb_group_ops = {
 527         .make_item      = nullb_group_make_item,
 528         .drop_item      = nullb_group_drop_item,
 529 };
 530
 531 static struct config_item_type nullb_group_type = {
 532         .ct_group_ops   = &nullb_group_ops,
 533         .ct_attrs       = nullb_group_attrs,
 534         .ct_owner       = THIS_MODULE,
 535 };
 536
 537 static struct configfs_subsystem nullb_subsys = {
 538         .su_group = {
 539                 .cg_item = {
 540                         .ci_namebuf = "nullb",
 541                         .ci_type = &nullb_group_type,
 542                 },
 543         },
 544 };
 545
 546 static inline int null_cache_active(struct nullb *nullb)
 547 {
 548         return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
 549 }
 550
 551 static struct nullb_device *null_alloc_dev(void)
 552 {
 553         struct nullb_device *dev;
 554
 555         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 556         if (!dev)
 557                 return NULL;
 558         INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
 559         INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
 560         if (badblocks_init(&dev->badblocks, 0)) {
 561                 kfree(dev);
 562                 return NULL;
 563         }
 564
 565         dev->size = g_gb * 1024;
 566         dev->completion_nsec = g_completion_nsec;
 567         dev->submit_queues = g_submit_queues;
 568         dev->home_node = g_home_node;
 569         dev->queue_mode = g_queue_mode;
 570         dev->blocksize = g_bs;
 571         dev->irqmode = g_irqmode;
 572         dev->hw_queue_depth = g_hw_queue_depth;
 573         dev->use_lightnvm = g_use_lightnvm;
 574         dev->blocking = g_blocking;
 575         dev->use_per_node_hctx = g_use_per_node_hctx;
 576         return dev;
 577 }
 578
 579 static void null_free_dev(struct nullb_device *dev)
 580 {
 581         kfree(dev);
 582 }
 583
 584 static void put_tag(struct nullb_queue *nq, unsigned int tag)
 585 {
 586         clear_bit_unlock(tag, nq->tag_map);
 587
 588         if (waitqueue_active(&nq->wait))
 589                 wake_up(&nq->wait);
 590 }
 591
 592 static unsigned int get_tag(struct nullb_queue *nq)
 593 {
 594         unsigned int tag;
 595
 596         do {
 597                 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
 598                 if (tag >= nq->queue_depth)
 599                         return -1U;
 600         } while (test_and_set_bit_lock(tag, nq->tag_map));
 601
 602         return tag;
 603 }
 604
 605 static void free_cmd(struct nullb_cmd *cmd)
 606 {
 607         put_tag(cmd->nq, cmd->tag);
 608 }
 609
 610 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
 611
 612 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 613 {
 614         struct nullb_cmd *cmd;
 615         unsigned int tag;
 616
 617         tag = get_tag(nq);
 618         if (tag != -1U) {
 619                 cmd = &nq->cmds[tag];
 620                 cmd->tag = tag;
 621                 cmd->nq = nq;
 622                 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
 623                         hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
 624                                      HRTIMER_MODE_REL);
 625                         cmd->timer.function = null_cmd_timer_expired;
 626                 }
 627                 return cmd;
 628         }
 629
 630         return NULL;
 631 }
 632
 633 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
 634 {
 635         struct nullb_cmd *cmd;
 636         DEFINE_WAIT(wait);
 637
 638         cmd = __alloc_cmd(nq);
 639         if (cmd || !can_wait)
 640                 return cmd;
 641
 642         do {
 643                 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
 644                 cmd = __alloc_cmd(nq);
 645                 if (cmd)
 646                         break;
 647
 648                 io_schedule();
 649         } while (1);
 650
 651         finish_wait(&nq->wait, &wait);
 652         return cmd;
 653 }
 654
 655 static void end_cmd(struct nullb_cmd *cmd)
 656 {
 657         struct request_queue *q = NULL;
 658         int queue_mode = cmd->nq->dev->queue_mode;
 659
 660         if (cmd->rq)
 661                 q = cmd->rq->q;
 662
 663         switch (queue_mode)  {
 664         case NULL_Q_MQ:
 665                 blk_mq_end_request(cmd->rq, cmd->error);
 666                 return;
 667         case NULL_Q_RQ:
 668                 INIT_LIST_HEAD(&cmd->rq->queuelist);
 669                 blk_end_request_all(cmd->rq, cmd->error);
 670                 break;
 671         case NULL_Q_BIO:
 672                 cmd->bio->bi_status = cmd->error;
 673                 bio_endio(cmd->bio);
 674                 break;
 675         }
 676
 677         free_cmd(cmd);
 678
 679         /* Restart queue if needed, as we are freeing a tag */
 680         if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
 681                 unsigned long flags;
 682
 683                 spin_lock_irqsave(q->queue_lock, flags);
 684                 blk_start_queue_async(q);
 685                 spin_unlock_irqrestore(q->queue_lock, flags);
 686         }
 687 }
 688
 689 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
 690 {
 691         end_cmd(container_of(timer, struct nullb_cmd, timer));
 692
 693         return HRTIMER_NORESTART;
 694 }
 695
 696 static void null_cmd_end_timer(struct nullb_cmd *cmd)
 697 {
 698         ktime_t kt = cmd->nq->dev->completion_nsec;
 699
 700         hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
 701 }
 702
 703 static void null_softirq_done_fn(struct request *rq)
 704 {
 705         struct nullb *nullb = rq->q->queuedata;
 706
 707         if (nullb->dev->queue_mode == NULL_Q_MQ)
 708                 end_cmd(blk_mq_rq_to_pdu(rq));
 709         else
 710                 end_cmd(rq->special);
 711 }
 712
 713 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
 714 {
 715         struct nullb_page *t_page;
 716
 717         t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
 718         if (!t_page)
 719                 goto out;
 720
 721         t_page->page = alloc_pages(gfp_flags, 0);
 722         if (!t_page->page)
 723                 goto out_freepage;
 724
 725         t_page->bitmap = 0;
 726         return t_page;
 727 out_freepage:
 728         kfree(t_page);
 729 out:
 730         return NULL;
 731 }
 732
 733 static void null_free_page(struct nullb_page *t_page)
 734 {
 735         __set_bit(NULLB_PAGE_FREE, &t_page->bitmap);
 736         if (test_bit(NULLB_PAGE_LOCK, &t_page->bitmap))
 737                 return;
 738         __free_page(t_page->page);
 739         kfree(t_page);
 740 }
 741
 742 static void null_free_sector(struct nullb *nullb, sector_t sector,
 743         bool is_cache)
 744 {
 745         unsigned int sector_bit;
 746         u64 idx;
 747         struct nullb_page *t_page, *ret;
 748         struct radix_tree_root *root;
 749
 750         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 751         idx = sector >> PAGE_SECTORS_SHIFT;
 752         sector_bit = (sector & SECTOR_MASK);
 753
 754         t_page = radix_tree_lookup(root, idx);
 755         if (t_page) {
 756                 __clear_bit(sector_bit, &t_page->bitmap);
 757
 758                 if (!t_page->bitmap) {
 759                         ret = radix_tree_delete_item(root, idx, t_page);
 760                         WARN_ON(ret != t_page);
 761                         null_free_page(ret);
 762                         if (is_cache)
 763                                 nullb->dev->curr_cache -= PAGE_SIZE;
 764                 }
 765         }
 766 }
 767
 768 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
 769         struct nullb_page *t_page, bool is_cache)
 770 {
 771         struct radix_tree_root *root;
 772
 773         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 774
 775         if (radix_tree_insert(root, idx, t_page)) {
 776                 null_free_page(t_page);
 777                 t_page = radix_tree_lookup(root, idx);
 778                 WARN_ON(!t_page || t_page->page->index != idx);
 779         } else if (is_cache)
 780                 nullb->dev->curr_cache += PAGE_SIZE;
 781
 782         return t_page;
 783 }
 784
 785 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
 786 {
 787         unsigned long pos = 0;
 788         int nr_pages;
 789         struct nullb_page *ret, *t_pages[FREE_BATCH];
 790         struct radix_tree_root *root;
 791
 792         root = is_cache ? &dev->cache : &dev->data;
 793
 794         do {
 795                 int i;
 796
 797                 nr_pages = radix_tree_gang_lookup(root,
 798                                 (void **)t_pages, pos, FREE_BATCH);
 799
 800                 for (i = 0; i < nr_pages; i++) {
 801                         pos = t_pages[i]->page->index;
 802                         ret = radix_tree_delete_item(root, pos, t_pages[i]);
 803                         WARN_ON(ret != t_pages[i]);
 804                         null_free_page(ret);
 805                 }
 806
 807                 pos++;
 808         } while (nr_pages == FREE_BATCH);
 809
 810         if (is_cache)
 811                 dev->curr_cache = 0;
 812 }
 813
 814 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
 815         sector_t sector, bool for_write, bool is_cache)
 816 {
 817         unsigned int sector_bit;
 818         u64 idx;
 819         struct nullb_page *t_page;
 820         struct radix_tree_root *root;
 821
 822         idx = sector >> PAGE_SECTORS_SHIFT;
 823         sector_bit = (sector & SECTOR_MASK);
 824
 825         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 826         t_page = radix_tree_lookup(root, idx);
 827         WARN_ON(t_page && t_page->page->index != idx);
 828
 829         if (t_page && (for_write || test_bit(sector_bit, &t_page->bitmap)))
 830                 return t_page;
 831
 832         return NULL;
 833 }
 834
 835 static struct nullb_page *null_lookup_page(struct nullb *nullb,
 836         sector_t sector, bool for_write, bool ignore_cache)
 837 {
 838         struct nullb_page *page = NULL;
 839
 840         if (!ignore_cache)
 841                 page = __null_lookup_page(nullb, sector, for_write, true);
 842         if (page)
 843                 return page;
 844         return __null_lookup_page(nullb, sector, for_write, false);
 845 }
 846
 847 static struct nullb_page *null_insert_page(struct nullb *nullb,
 848         sector_t sector, bool ignore_cache)
 849 {
 850         u64 idx;
 851         struct nullb_page *t_page;
 852
 853         t_page = null_lookup_page(nullb, sector, true, ignore_cache);
 854         if (t_page)
 855                 return t_page;
 856
 857         spin_unlock_irq(&nullb->lock);
 858
 859         t_page = null_alloc_page(GFP_NOIO);
 860         if (!t_page)
 861                 goto out_lock;
 862
 863         if (radix_tree_preload(GFP_NOIO))
 864                 goto out_freepage;
 865
 866         spin_lock_irq(&nullb->lock);
 867         idx = sector >> PAGE_SECTORS_SHIFT;
 868         t_page->page->index = idx;
 869         t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
 870         radix_tree_preload_end();
 871
 872         return t_page;
 873 out_freepage:
 874         null_free_page(t_page);
 875 out_lock:
 876         spin_lock_irq(&nullb->lock);
 877         return null_lookup_page(nullb, sector, true, ignore_cache);
 878 }
 879
 880 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
 881 {
 882         int i;
 883         unsigned int offset;
 884         u64 idx;
 885         struct nullb_page *t_page, *ret;
 886         void *dst, *src;
 887
 888         idx = c_page->page->index;
 889
 890         t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
 891
 892         __clear_bit(NULLB_PAGE_LOCK, &c_page->bitmap);
 893         if (test_bit(NULLB_PAGE_FREE, &c_page->bitmap)) {
 894                 null_free_page(c_page);
 895                 if (t_page && t_page->bitmap == 0) {
 896                         ret = radix_tree_delete_item(&nullb->dev->data,
 897                                 idx, t_page);
 898                         null_free_page(t_page);
 899                 }
 900                 return 0;
 901         }
 902
 903         if (!t_page)
 904                 return -ENOMEM;
 905
 906         src = kmap_atomic(c_page->page);
 907         dst = kmap_atomic(t_page->page);
 908
 909         for (i = 0; i < PAGE_SECTORS;
 910                         i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
 911                 if (test_bit(i, &c_page->bitmap)) {
 912                         offset = (i << SECTOR_SHIFT);
 913                         memcpy(dst + offset, src + offset,
 914                                 nullb->dev->blocksize);
 915                         __set_bit(i, &t_page->bitmap);
 916                 }
 917         }
 918
 919         kunmap_atomic(dst);
 920         kunmap_atomic(src);
 921
 922         ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
 923         null_free_page(ret);
 924         nullb->dev->curr_cache -= PAGE_SIZE;
 925
 926         return 0;
 927 }
 928
 929 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
 930 {
 931         int i, err, nr_pages;
 932         struct nullb_page *c_pages[FREE_BATCH];
 933         unsigned long flushed = 0, one_round;
 934
 935 again:
 936         if ((nullb->dev->cache_size * 1024 * 1024) >
 937              nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
 938                 return 0;
 939
 940         nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
 941                         (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
 942         /*
 943          * nullb_flush_cache_page could unlock before using the c_pages. To
 944          * avoid race, we don't allow page free
 945          */
 946         for (i = 0; i < nr_pages; i++) {
 947                 nullb->cache_flush_pos = c_pages[i]->page->index;
 948                 /*
 949                  * We found the page which is being flushed to disk by other
 950                  * threads
 951                  */
 952                 if (test_bit(NULLB_PAGE_LOCK, &c_pages[i]->bitmap))
 953                         c_pages[i] = NULL;
 954                 else
 955                         __set_bit(NULLB_PAGE_LOCK, &c_pages[i]->bitmap);
 956         }
 957
 958         one_round = 0;
 959         for (i = 0; i < nr_pages; i++) {
 960                 if (c_pages[i] == NULL)
 961                         continue;
 962                 err = null_flush_cache_page(nullb, c_pages[i]);
 963                 if (err)
 964                         return err;
 965                 one_round++;
 966         }
 967         flushed += one_round << PAGE_SHIFT;
 968
 969         if (n > flushed) {
 970                 if (nr_pages == 0)
 971                         nullb->cache_flush_pos = 0;
 972                 if (one_round == 0) {
 973                         /* give other threads a chance */
 974                         spin_unlock_irq(&nullb->lock);
 975                         spin_lock_irq(&nullb->lock);
 976                 }
 977                 goto again;
 978         }
 979         return 0;
 980 }
 981
 982 static int copy_to_nullb(struct nullb *nullb, struct page *source,
 983         unsigned int off, sector_t sector, size_t n, bool is_fua)
 984 {
 985         size_t temp, count = 0;
 986         unsigned int offset;
 987         struct nullb_page *t_page;
 988         void *dst, *src;
 989
 990         while (count < n) {
 991                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
 992
 993                 if (null_cache_active(nullb) && !is_fua)
 994                         null_make_cache_space(nullb, PAGE_SIZE);
 995
 996                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
 997                 t_page = null_insert_page(nullb, sector,
 998                         !null_cache_active(nullb) || is_fua);
 999                 if (!t_page)
1000                         return -ENOSPC;
1001
1002                 src = kmap_atomic(source);
1003                 dst = kmap_atomic(t_page->page);
1004                 memcpy(dst + offset, src + off + count, temp);
1005                 kunmap_atomic(dst);
1006                 kunmap_atomic(src);
1007
1008                 __set_bit(sector & SECTOR_MASK, &t_page->bitmap);
1009
1010                 if (is_fua)
1011                         null_free_sector(nullb, sector, true);
1012
1013                 count += temp;
1014                 sector += temp >> SECTOR_SHIFT;
1015         }
1016         return 0;
1017 }
1018
1019 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1020         unsigned int off, sector_t sector, size_t n)
1021 {
1022         size_t temp, count = 0;
1023         unsigned int offset;
1024         struct nullb_page *t_page;
1025         void *dst, *src;
1026
1027         while (count < n) {
1028                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1029
1030                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1031                 t_page = null_lookup_page(nullb, sector, false,
1032                         !null_cache_active(nullb));
1033
1034                 dst = kmap_atomic(dest);
1035                 if (!t_page) {
1036                         memset(dst + off + count, 0, temp);
1037                         goto next;
1038                 }
1039                 src = kmap_atomic(t_page->page);
1040                 memcpy(dst + off + count, src + offset, temp);
1041                 kunmap_atomic(src);
1042 next:
1043                 kunmap_atomic(dst);
1044
1045                 count += temp;
1046                 sector += temp >> SECTOR_SHIFT;
1047         }
1048         return 0;
1049 }
1050
1051 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1052 {
1053         size_t temp;
1054
1055         spin_lock_irq(&nullb->lock);
1056         while (n > 0) {
1057                 temp = min_t(size_t, n, nullb->dev->blocksize);
1058                 null_free_sector(nullb, sector, false);
1059                 if (null_cache_active(nullb))
1060                         null_free_sector(nullb, sector, true);
1061                 sector += temp >> SECTOR_SHIFT;
1062                 n -= temp;
1063         }
1064         spin_unlock_irq(&nullb->lock);
1065 }
1066
1067 static int null_handle_flush(struct nullb *nullb)
1068 {
1069         int err;
1070
1071         if (!null_cache_active(nullb))
1072                 return 0;
1073
1074         spin_lock_irq(&nullb->lock);
1075         while (true) {
1076                 err = null_make_cache_space(nullb,
1077                         nullb->dev->cache_size * 1024 * 1024);
1078                 if (err || nullb->dev->curr_cache == 0)
1079                         break;
1080         }
1081
1082         WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1083         spin_unlock_irq(&nullb->lock);
1084         return err;
1085 }
1086
1087 static int null_transfer(struct nullb *nullb, struct page *page,
1088         unsigned int len, unsigned int off, bool is_write, sector_t sector,
1089         bool is_fua)
1090 {
1091         int err = 0;
1092
1093         if (!is_write) {
1094                 err = copy_from_nullb(nullb, page, off, sector, len);
1095                 flush_dcache_page(page);
1096         } else {
1097                 flush_dcache_page(page);
1098                 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1099         }
1100
1101         return err;
1102 }
1103
1104 static int null_handle_rq(struct nullb_cmd *cmd)
1105 {
1106         struct request *rq = cmd->rq;
1107         struct nullb *nullb = cmd->nq->dev->nullb;
1108         int err;
1109         unsigned int len;
1110         sector_t sector;
1111         struct req_iterator iter;
1112         struct bio_vec bvec;
1113
1114         sector = blk_rq_pos(rq);
1115
1116         if (req_op(rq) == REQ_OP_DISCARD) {
1117                 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1118                 return 0;
1119         }
1120
1121         spin_lock_irq(&nullb->lock);
1122         rq_for_each_segment(bvec, rq, iter) {
1123                 len = bvec.bv_len;
1124                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1125                                      op_is_write(req_op(rq)), sector,
1126                                      req_op(rq) & REQ_FUA);
1127                 if (err) {
1128                         spin_unlock_irq(&nullb->lock);
1129                         return err;
1130                 }
1131                 sector += len >> SECTOR_SHIFT;
1132         }
1133         spin_unlock_irq(&nullb->lock);
1134
1135         return 0;
1136 }
1137
1138 static int null_handle_bio(struct nullb_cmd *cmd)
1139 {
1140         struct bio *bio = cmd->bio;
1141         struct nullb *nullb = cmd->nq->dev->nullb;
1142         int err;
1143         unsigned int len;
1144         sector_t sector;
1145         struct bio_vec bvec;
1146         struct bvec_iter iter;
1147
1148         sector = bio->bi_iter.bi_sector;
1149
1150         if (bio_op(bio) == REQ_OP_DISCARD) {
1151                 null_handle_discard(nullb, sector,
1152                         bio_sectors(bio) << SECTOR_SHIFT);
1153                 return 0;
1154         }
1155
1156         spin_lock_irq(&nullb->lock);
1157         bio_for_each_segment(bvec, bio, iter) {
1158                 len = bvec.bv_len;
1159                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1160                                      op_is_write(bio_op(bio)), sector,
1161                                      bio_op(bio) & REQ_FUA);
1162                 if (err) {
1163                         spin_unlock_irq(&nullb->lock);
1164                         return err;
1165                 }
1166                 sector += len >> SECTOR_SHIFT;
1167         }
1168         spin_unlock_irq(&nullb->lock);
1169         return 0;
1170 }
1171
1172 static void null_stop_queue(struct nullb *nullb)
1173 {
1174         struct request_queue *q = nullb->q;
1175
1176         if (nullb->dev->queue_mode == NULL_Q_MQ)
1177                 blk_mq_stop_hw_queues(q);
1178         else {
1179                 spin_lock_irq(q->queue_lock);
1180                 blk_stop_queue(q);
1181                 spin_unlock_irq(q->queue_lock);
1182         }
1183 }
1184
1185 static void null_restart_queue_async(struct nullb *nullb)
1186 {
1187         struct request_queue *q = nullb->q;
1188         unsigned long flags;
1189
1190         if (nullb->dev->queue_mode == NULL_Q_MQ)
1191                 blk_mq_start_stopped_hw_queues(q, true);
1192         else {
1193                 spin_lock_irqsave(q->queue_lock, flags);
1194                 blk_start_queue_async(q);
1195                 spin_unlock_irqrestore(q->queue_lock, flags);
1196         }
1197 }
1198
1199 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd)
1200 {
1201         struct nullb_device *dev = cmd->nq->dev;
1202         struct nullb *nullb = dev->nullb;
1203         int err = 0;
1204
1205         if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1206                 struct request *rq = cmd->rq;
1207
1208                 if (!hrtimer_active(&nullb->bw_timer))
1209                         hrtimer_restart(&nullb->bw_timer);
1210
1211                 if (atomic_long_sub_return(blk_rq_bytes(rq),
1212                                 &nullb->cur_bytes) < 0) {
1213                         null_stop_queue(nullb);
1214                         /* race with timer */
1215                         if (atomic_long_read(&nullb->cur_bytes) > 0)
1216                                 null_restart_queue_async(nullb);
1217                         if (dev->queue_mode == NULL_Q_RQ) {
1218                                 struct request_queue *q = nullb->q;
1219
1220                                 spin_lock_irq(q->queue_lock);
1221                                 rq->rq_flags |= RQF_DONTPREP;
1222                                 blk_requeue_request(q, rq);
1223                                 spin_unlock_irq(q->queue_lock);
1224                                 return BLK_STS_OK;
1225                         } else
1226                                 /* requeue request */
1227                                 return BLK_STS_RESOURCE;
1228                 }
1229         }
1230
1231         if (nullb->dev->badblocks.shift != -1) {
1232                 int bad_sectors;
1233                 sector_t sector, size, first_bad;
1234                 bool is_flush = true;
1235
1236                 if (dev->queue_mode == NULL_Q_BIO &&
1237                                 bio_op(cmd->bio) != REQ_OP_FLUSH) {
1238                         is_flush = false;
1239                         sector = cmd->bio->bi_iter.bi_sector;
1240                         size = bio_sectors(cmd->bio);
1241                 }
1242                 if (dev->queue_mode != NULL_Q_BIO &&
1243                                 req_op(cmd->rq) != REQ_OP_FLUSH) {
1244                         is_flush = false;
1245                         sector = blk_rq_pos(cmd->rq);
1246                         size = blk_rq_sectors(cmd->rq);
1247                 }
1248                 if (!is_flush && badblocks_check(&nullb->dev->badblocks, sector,
1249                                 size, &first_bad, &bad_sectors)) {
1250                         cmd->error = BLK_STS_IOERR;
1251                         goto out;
1252                 }
1253         }
1254
1255         if (dev->memory_backed) {
1256                 if (dev->queue_mode == NULL_Q_BIO) {
1257                         if (bio_op(cmd->bio) == REQ_OP_FLUSH)
1258                                 err = null_handle_flush(nullb);
1259                         else
1260                                 err = null_handle_bio(cmd);
1261                 } else {
1262                         if (req_op(cmd->rq) == REQ_OP_FLUSH)
1263                                 err = null_handle_flush(nullb);
1264                         else
1265                                 err = null_handle_rq(cmd);
1266                 }
1267         }
1268         cmd->error = errno_to_blk_status(err);
1269 out:
1270         /* Complete IO by inline, softirq or timer */
1271         switch (dev->irqmode) {
1272         case NULL_IRQ_SOFTIRQ:
1273                 switch (dev->queue_mode)  {
1274                 case NULL_Q_MQ:
1275                         blk_mq_complete_request(cmd->rq);
1276                         break;
1277                 case NULL_Q_RQ:
1278                         blk_complete_request(cmd->rq);
1279                         break;
1280                 case NULL_Q_BIO:
1281                         /*
1282                          * XXX: no proper submitting cpu information available.
1283                          */
1284                         end_cmd(cmd);
1285                         break;
1286                 }
1287                 break;
1288         case NULL_IRQ_NONE:
1289                 end_cmd(cmd);
1290                 break;
1291         case NULL_IRQ_TIMER:
1292                 null_cmd_end_timer(cmd);
1293                 break;
1294         }
1295         return BLK_STS_OK;
1296 }
1297
1298 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1299 {
1300         struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1301         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1302         unsigned int mbps = nullb->dev->mbps;
1303
1304         if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1305                 return HRTIMER_NORESTART;
1306
1307         atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1308         null_restart_queue_async(nullb);
1309
1310         hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1311
1312         return HRTIMER_RESTART;
1313 }
1314
1315 static void nullb_setup_bwtimer(struct nullb *nullb)
1316 {
1317         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1318
1319         hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1320         nullb->bw_timer.function = nullb_bwtimer_fn;
1321         atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1322         hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1323 }
1324
1325 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1326 {
1327         int index = 0;
1328
1329         if (nullb->nr_queues != 1)
1330                 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1331
1332         return &nullb->queues[index];
1333 }
1334
1335 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
1336 {
1337         struct nullb *nullb = q->queuedata;
1338         struct nullb_queue *nq = nullb_to_queue(nullb);
1339         struct nullb_cmd *cmd;
1340
1341         cmd = alloc_cmd(nq, 1);
1342         cmd->bio = bio;
1343
1344         null_handle_cmd(cmd);
1345         return BLK_QC_T_NONE;
1346 }
1347
1348 static int null_rq_prep_fn(struct request_queue *q, struct request *req)
1349 {
1350         struct nullb *nullb = q->queuedata;
1351         struct nullb_queue *nq = nullb_to_queue(nullb);
1352         struct nullb_cmd *cmd;
1353
1354         cmd = alloc_cmd(nq, 0);
1355         if (cmd) {
1356                 cmd->rq = req;
1357                 req->special = cmd;
1358                 return BLKPREP_OK;
1359         }
1360         blk_stop_queue(q);
1361
1362         return BLKPREP_DEFER;
1363 }
1364
1365 static void null_request_fn(struct request_queue *q)
1366 {
1367         struct request *rq;
1368
1369         while ((rq = blk_fetch_request(q)) != NULL) {
1370                 struct nullb_cmd *cmd = rq->special;
1371
1372                 spin_unlock_irq(q->queue_lock);
1373                 null_handle_cmd(cmd);
1374                 spin_lock_irq(q->queue_lock);
1375         }
1376 }
1377
1378 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1379                          const struct blk_mq_queue_data *bd)
1380 {
1381         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1382         struct nullb_queue *nq = hctx->driver_data;
1383
1384         might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1385
1386         if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1387                 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1388                 cmd->timer.function = null_cmd_timer_expired;
1389         }
1390         cmd->rq = bd->rq;
1391         cmd->nq = nq;
1392
1393         blk_mq_start_request(bd->rq);
1394
1395         return null_handle_cmd(cmd);
1396 }
1397
1398 static const struct blk_mq_ops null_mq_ops = {
1399         .queue_rq       = null_queue_rq,
1400         .complete       = null_softirq_done_fn,
1401 };
1402
1403 static void cleanup_queue(struct nullb_queue *nq)
1404 {
1405         kfree(nq->tag_map);
1406         kfree(nq->cmds);
1407 }
1408
1409 static void cleanup_queues(struct nullb *nullb)
1410 {
1411         int i;
1412
1413         for (i = 0; i < nullb->nr_queues; i++)
1414                 cleanup_queue(&nullb->queues[i]);
1415
1416         kfree(nullb->queues);
1417 }
1418
1419 #ifdef CONFIG_NVM
1420
1421 static void null_lnvm_end_io(struct request *rq, blk_status_t status)
1422 {
1423         struct nvm_rq *rqd = rq->end_io_data;
1424
1425         /* XXX: lighnvm core seems to expect NVM_RSP_* values here.. */
1426         rqd->error = status ? -EIO : 0;
1427         nvm_end_io(rqd);
1428
1429         blk_put_request(rq);
1430 }
1431
1432 static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
1433 {
1434         struct request_queue *q = dev->q;
1435         struct request *rq;
1436         struct bio *bio = rqd->bio;
1437
1438         rq = blk_mq_alloc_request(q,
1439                 op_is_write(bio_op(bio)) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1440         if (IS_ERR(rq))
1441                 return -ENOMEM;
1442
1443         blk_init_request_from_bio(rq, bio);
1444
1445         rq->end_io_data = rqd;
1446
1447         blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
1448
1449         return 0;
1450 }
1451
1452 static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
1453 {
1454         struct nullb *nullb = dev->q->queuedata;
1455         sector_t size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1456         sector_t blksize;
1457         struct nvm_id_group *grp;
1458
1459         id->ver_id = 0x1;
1460         id->vmnt = 0;
1461         id->cap = 0x2;
1462         id->dom = 0x1;
1463
1464         id->ppaf.blk_offset = 0;
1465         id->ppaf.blk_len = 16;
1466         id->ppaf.pg_offset = 16;
1467         id->ppaf.pg_len = 16;
1468         id->ppaf.sect_offset = 32;
1469         id->ppaf.sect_len = 8;
1470         id->ppaf.pln_offset = 40;
1471         id->ppaf.pln_len = 8;
1472         id->ppaf.lun_offset = 48;
1473         id->ppaf.lun_len = 8;
1474         id->ppaf.ch_offset = 56;
1475         id->ppaf.ch_len = 8;
1476
1477         sector_div(size, nullb->dev->blocksize); /* convert size to pages */
1478         size >>= 8; /* concert size to pgs pr blk */
1479         grp = &id->grp;
1480         grp->mtype = 0;
1481         grp->fmtype = 0;
1482         grp->num_ch = 1;
1483         grp->num_pg = 256;
1484         blksize = size;
1485         size >>= 16;
1486         grp->num_lun = size + 1;
1487         sector_div(blksize, grp->num_lun);
1488         grp->num_blk = blksize;
1489         grp->num_pln = 1;
1490
1491         grp->fpg_sz = nullb->dev->blocksize;
1492         grp->csecs = nullb->dev->blocksize;
1493         grp->trdt = 25000;
1494         grp->trdm = 25000;
1495         grp->tprt = 500000;
1496         grp->tprm = 500000;
1497         grp->tbet = 1500000;
1498         grp->tbem = 1500000;
1499         grp->mpos = 0x010101; /* single plane rwe */
1500         grp->cpar = nullb->dev->hw_queue_depth;
1501
1502         return 0;
1503 }
1504
1505 static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
1506 {
1507         mempool_t *virtmem_pool;
1508
1509         virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
1510         if (!virtmem_pool) {
1511                 pr_err("null_blk: Unable to create virtual memory pool\n");
1512                 return NULL;
1513         }
1514
1515         return virtmem_pool;
1516 }
1517
1518 static void null_lnvm_destroy_dma_pool(void *pool)
1519 {
1520         mempool_destroy(pool);
1521 }
1522
1523 static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
1524                                 gfp_t mem_flags, dma_addr_t *dma_handler)
1525 {
1526         return mempool_alloc(pool, mem_flags);
1527 }
1528
1529 static void null_lnvm_dev_dma_free(void *pool, void *entry,
1530                                                         dma_addr_t dma_handler)
1531 {
1532         mempool_free(entry, pool);
1533 }
1534
1535 static struct nvm_dev_ops null_lnvm_dev_ops = {
1536         .identity               = null_lnvm_id,
1537         .submit_io              = null_lnvm_submit_io,
1538
1539         .create_dma_pool        = null_lnvm_create_dma_pool,
1540         .destroy_dma_pool       = null_lnvm_destroy_dma_pool,
1541         .dev_dma_alloc          = null_lnvm_dev_dma_alloc,
1542         .dev_dma_free           = null_lnvm_dev_dma_free,
1543
1544         /* Simulate nvme protocol restriction */
1545         .max_phys_sect          = 64,
1546 };
1547
1548 static int null_nvm_register(struct nullb *nullb)
1549 {
1550         struct nvm_dev *dev;
1551         int rv;
1552
1553         dev = nvm_alloc_dev(0);
1554         if (!dev)
1555                 return -ENOMEM;
1556
1557         dev->q = nullb->q;
1558         memcpy(dev->name, nullb->disk_name, DISK_NAME_LEN);
1559         dev->ops = &null_lnvm_dev_ops;
1560
1561         rv = nvm_register(dev);
1562         if (rv) {
1563                 kfree(dev);
1564                 return rv;
1565         }
1566         nullb->ndev = dev;
1567         return 0;
1568 }
1569
1570 static void null_nvm_unregister(struct nullb *nullb)
1571 {
1572         nvm_unregister(nullb->ndev);
1573 }
1574 #else
1575 static int null_nvm_register(struct nullb *nullb)
1576 {
1577         pr_err("null_blk: CONFIG_NVM needs to be enabled for LightNVM\n");
1578         return -EINVAL;
1579 }
1580 static void null_nvm_unregister(struct nullb *nullb) {}
1581 #endif /* CONFIG_NVM */
1582
1583 static void null_del_dev(struct nullb *nullb)
1584 {
1585         struct nullb_device *dev = nullb->dev;
1586
1587         ida_simple_remove(&nullb_indexes, nullb->index);
1588
1589         list_del_init(&nullb->list);
1590
1591         if (dev->use_lightnvm)
1592                 null_nvm_unregister(nullb);
1593         else
1594                 del_gendisk(nullb->disk);
1595
1596         if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1597                 hrtimer_cancel(&nullb->bw_timer);
1598                 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1599                 null_restart_queue_async(nullb);
1600         }
1601
1602         blk_cleanup_queue(nullb->q);
1603         if (dev->queue_mode == NULL_Q_MQ &&
1604             nullb->tag_set == &nullb->__tag_set)
1605                 blk_mq_free_tag_set(nullb->tag_set);
1606         if (!dev->use_lightnvm)
1607                 put_disk(nullb->disk);
1608         cleanup_queues(nullb);
1609         if (null_cache_active(nullb))
1610                 null_free_device_storage(nullb->dev, true);
1611         kfree(nullb);
1612         dev->nullb = NULL;
1613 }
1614
1615 static void null_config_discard(struct nullb *nullb)
1616 {
1617         if (nullb->dev->discard == false)
1618                 return;
1619         nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1620         nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1621         blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1622         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nullb->q);
1623 }
1624
1625 static int null_open(struct block_device *bdev, fmode_t mode)
1626 {
1627         return 0;
1628 }
1629
1630 static void null_release(struct gendisk *disk, fmode_t mode)
1631 {
1632 }
1633
1634 static const struct block_device_operations null_fops = {
1635         .owner =        THIS_MODULE,
1636         .open =         null_open,
1637         .release =      null_release,
1638 };
1639
1640 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1641 {
1642         BUG_ON(!nullb);
1643         BUG_ON(!nq);
1644
1645         init_waitqueue_head(&nq->wait);
1646         nq->queue_depth = nullb->queue_depth;
1647         nq->dev = nullb->dev;
1648 }
1649
1650 static void null_init_queues(struct nullb *nullb)
1651 {
1652         struct request_queue *q = nullb->q;
1653         struct blk_mq_hw_ctx *hctx;
1654         struct nullb_queue *nq;
1655         int i;
1656
1657         queue_for_each_hw_ctx(q, hctx, i) {
1658                 if (!hctx->nr_ctx || !hctx->tags)
1659                         continue;
1660                 nq = &nullb->queues[i];
1661                 hctx->driver_data = nq;
1662                 null_init_queue(nullb, nq);
1663                 nullb->nr_queues++;
1664         }
1665 }
1666
1667 static int setup_commands(struct nullb_queue *nq)
1668 {
1669         struct nullb_cmd *cmd;
1670         int i, tag_size;
1671
1672         nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
1673         if (!nq->cmds)
1674                 return -ENOMEM;
1675
1676         tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1677         nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
1678         if (!nq->tag_map) {
1679                 kfree(nq->cmds);
1680                 return -ENOMEM;
1681         }
1682
1683         for (i = 0; i < nq->queue_depth; i++) {
1684                 cmd = &nq->cmds[i];
1685                 INIT_LIST_HEAD(&cmd->list);
1686                 cmd->ll_list.next = NULL;
1687                 cmd->tag = -1U;
1688         }
1689
1690         return 0;
1691 }
1692
1693 static int setup_queues(struct nullb *nullb)
1694 {
1695         nullb->queues = kzalloc(nullb->dev->submit_queues *
1696                 sizeof(struct nullb_queue), GFP_KERNEL);
1697         if (!nullb->queues)
1698                 return -ENOMEM;
1699
1700         nullb->nr_queues = 0;
1701         nullb->queue_depth = nullb->dev->hw_queue_depth;
1702
1703         return 0;
1704 }
1705
1706 static int init_driver_queues(struct nullb *nullb)
1707 {
1708         struct nullb_queue *nq;
1709         int i, ret = 0;
1710
1711         for (i = 0; i < nullb->dev->submit_queues; i++) {
1712                 nq = &nullb->queues[i];
1713
1714                 null_init_queue(nullb, nq);
1715
1716                 ret = setup_commands(nq);
1717                 if (ret)
1718                         return ret;
1719                 nullb->nr_queues++;
1720         }
1721         return 0;
1722 }
1723
1724 static int null_gendisk_register(struct nullb *nullb)
1725 {
1726         struct gendisk *disk;
1727         sector_t size;
1728
1729         disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1730         if (!disk)
1731                 return -ENOMEM;
1732         size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1733         set_capacity(disk, size >> 9);
1734
1735         disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1736         disk->major             = null_major;
1737         disk->first_minor       = nullb->index;
1738         disk->fops              = &null_fops;
1739         disk->private_data      = nullb;
1740         disk->queue             = nullb->q;
1741         strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1742
1743         add_disk(disk);
1744         return 0;
1745 }
1746
1747 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1748 {
1749         set->ops = &null_mq_ops;
1750         set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1751                                                 g_submit_queues;
1752         set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1753                                                 g_hw_queue_depth;
1754         set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1755         set->cmd_size   = sizeof(struct nullb_cmd);
1756         set->flags = BLK_MQ_F_SHOULD_MERGE;
1757         set->driver_data = NULL;
1758
1759         if ((nullb && nullb->dev->blocking) || g_blocking)
1760                 set->flags |= BLK_MQ_F_BLOCKING;
1761
1762         return blk_mq_alloc_tag_set(set);
1763 }
1764
1765 static void null_validate_conf(struct nullb_device *dev)
1766 {
1767         dev->blocksize = round_down(dev->blocksize, 512);
1768         dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1769         if (dev->use_lightnvm && dev->blocksize != 4096)
1770                 dev->blocksize = 4096;
1771
1772         if (dev->use_lightnvm && dev->queue_mode != NULL_Q_MQ)
1773                 dev->queue_mode = NULL_Q_MQ;
1774
1775         if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1776                 if (dev->submit_queues != nr_online_nodes)
1777                         dev->submit_queues = nr_online_nodes;
1778         } else if (dev->submit_queues > nr_cpu_ids)
1779                 dev->submit_queues = nr_cpu_ids;
1780         else if (dev->submit_queues == 0)
1781                 dev->submit_queues = 1;
1782
1783         dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1784         dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1785
1786         /* Do memory allocation, so set blocking */
1787         if (dev->memory_backed)
1788                 dev->blocking = true;
1789         else /* cache is meaningless */
1790                 dev->cache_size = 0;
1791         dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1792                                                 dev->cache_size);
1793         dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1794         /* can not stop a queue */
1795         if (dev->queue_mode == NULL_Q_BIO)
1796                 dev->mbps = 0;
1797 }
1798
1799 static int null_add_dev(struct nullb_device *dev)
1800 {
1801         struct nullb *nullb;
1802         int rv;
1803
1804         null_validate_conf(dev);
1805
1806         nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1807         if (!nullb) {
1808                 rv = -ENOMEM;
1809                 goto out;
1810         }
1811         nullb->dev = dev;
1812         dev->nullb = nullb;
1813
1814         spin_lock_init(&nullb->lock);
1815
1816         rv = setup_queues(nullb);
1817         if (rv)
1818                 goto out_free_nullb;
1819
1820         if (dev->queue_mode == NULL_Q_MQ) {
1821                 if (shared_tags) {
1822                         nullb->tag_set = &tag_set;
1823                         rv = 0;
1824                 } else {
1825                         nullb->tag_set = &nullb->__tag_set;
1826                         rv = null_init_tag_set(nullb, nullb->tag_set);
1827                 }
1828
1829                 if (rv)
1830                         goto out_cleanup_queues;
1831
1832                 nullb->q = blk_mq_init_queue(nullb->tag_set);
1833                 if (IS_ERR(nullb->q)) {
1834                         rv = -ENOMEM;
1835                         goto out_cleanup_tags;
1836                 }
1837                 null_init_queues(nullb);
1838         } else if (dev->queue_mode == NULL_Q_BIO) {
1839                 nullb->q = blk_alloc_queue_node(GFP_KERNEL, dev->home_node);
1840                 if (!nullb->q) {
1841                         rv = -ENOMEM;
1842                         goto out_cleanup_queues;
1843                 }
1844                 blk_queue_make_request(nullb->q, null_queue_bio);
1845                 rv = init_driver_queues(nullb);
1846                 if (rv)
1847                         goto out_cleanup_blk_queue;
1848         } else {
1849                 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock,
1850                                                 dev->home_node);
1851                 if (!nullb->q) {
1852                         rv = -ENOMEM;
1853                         goto out_cleanup_queues;
1854                 }
1855                 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
1856                 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
1857                 rv = init_driver_queues(nullb);
1858                 if (rv)
1859                         goto out_cleanup_blk_queue;
1860         }
1861
1862         if (dev->mbps) {
1863                 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1864                 nullb_setup_bwtimer(nullb);
1865         }
1866
1867         if (dev->cache_size > 0) {
1868                 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1869                 blk_queue_write_cache(nullb->q, true, true);
1870                 blk_queue_flush_queueable(nullb->q, true);
1871         }
1872
1873         nullb->q->queuedata = nullb;
1874         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
1875         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1876
1877         mutex_lock(&lock);
1878         nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1879         dev->index = nullb->index;
1880         mutex_unlock(&lock);
1881
1882         blk_queue_logical_block_size(nullb->q, dev->blocksize);
1883         blk_queue_physical_block_size(nullb->q, dev->blocksize);
1884
1885         null_config_discard(nullb);
1886
1887         sprintf(nullb->disk_name, "nullb%d", nullb->index);
1888
1889         if (dev->use_lightnvm)
1890                 rv = null_nvm_register(nullb);
1891         else
1892                 rv = null_gendisk_register(nullb);
1893
1894         if (rv)
1895                 goto out_cleanup_blk_queue;
1896
1897         mutex_lock(&lock);
1898         list_add_tail(&nullb->list, &nullb_list);
1899         mutex_unlock(&lock);
1900
1901         return 0;
1902 out_cleanup_blk_queue:
1903         blk_cleanup_queue(nullb->q);
1904 out_cleanup_tags:
1905         if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1906                 blk_mq_free_tag_set(nullb->tag_set);
1907 out_cleanup_queues:
1908         cleanup_queues(nullb);
1909 out_free_nullb:
1910         kfree(nullb);
1911 out:
1912         return rv;
1913 }
1914
1915 static int __init null_init(void)
1916 {
1917         int ret = 0;
1918         unsigned int i;
1919         struct nullb *nullb;
1920         struct nullb_device *dev;
1921
1922         /* check for nullb_page.bitmap */
1923         if (sizeof(unsigned long) * 8 - 2 < (PAGE_SIZE >> SECTOR_SHIFT))
1924                 return -EINVAL;
1925
1926         if (g_bs > PAGE_SIZE) {
1927                 pr_warn("null_blk: invalid block size\n");
1928                 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
1929                 g_bs = PAGE_SIZE;
1930         }
1931
1932         if (g_use_lightnvm && g_bs != 4096) {
1933                 pr_warn("null_blk: LightNVM only supports 4k block size\n");
1934                 pr_warn("null_blk: defaults block size to 4k\n");
1935                 g_bs = 4096;
1936         }
1937
1938         if (g_use_lightnvm && g_queue_mode != NULL_Q_MQ) {
1939                 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
1940                 pr_warn("null_blk: defaults queue mode to blk-mq\n");
1941                 g_queue_mode = NULL_Q_MQ;
1942         }
1943
1944         if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1945                 if (g_submit_queues != nr_online_nodes) {
1946                         pr_warn("null_blk: submit_queues param is set to %u.\n",
1947                                                         nr_online_nodes);
1948                         g_submit_queues = nr_online_nodes;
1949                 }
1950         } else if (g_submit_queues > nr_cpu_ids)
1951                 g_submit_queues = nr_cpu_ids;
1952         else if (g_submit_queues <= 0)
1953                 g_submit_queues = 1;
1954
1955         if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1956                 ret = null_init_tag_set(NULL, &tag_set);
1957                 if (ret)
1958                         return ret;
1959         }
1960
1961         config_group_init(&nullb_subsys.su_group);
1962         mutex_init(&nullb_subsys.su_mutex);
1963
1964         ret = configfs_register_subsystem(&nullb_subsys);
1965         if (ret)
1966                 goto err_tagset;
1967
1968         mutex_init(&lock);
1969
1970         null_major = register_blkdev(0, "nullb");
1971         if (null_major < 0) {
1972                 ret = null_major;
1973                 goto err_conf;
1974         }
1975
1976         if (g_use_lightnvm) {
1977                 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
1978                                                                 0, 0, NULL);
1979                 if (!ppa_cache) {
1980                         pr_err("null_blk: unable to create ppa cache\n");
1981                         ret = -ENOMEM;
1982                         goto err_ppa;
1983                 }
1984         }
1985
1986         for (i = 0; i < nr_devices; i++) {
1987                 dev = null_alloc_dev();
1988                 if (!dev)
1989                         goto err_dev;
1990                 ret = null_add_dev(dev);
1991                 if (ret) {
1992                         null_free_dev(dev);
1993                         goto err_dev;
1994                 }
1995         }
1996
1997         pr_info("null: module loaded\n");
1998         return 0;
1999
2000 err_dev:
2001         while (!list_empty(&nullb_list)) {
2002                 nullb = list_entry(nullb_list.next, struct nullb, list);
2003                 dev = nullb->dev;
2004                 null_del_dev(nullb);
2005                 null_free_dev(dev);
2006         }
2007         kmem_cache_destroy(ppa_cache);
2008 err_ppa:
2009         unregister_blkdev(null_major, "nullb");
2010 err_conf:
2011         configfs_unregister_subsystem(&nullb_subsys);
2012 err_tagset:
2013         if (g_queue_mode == NULL_Q_MQ && shared_tags)
2014                 blk_mq_free_tag_set(&tag_set);
2015         return ret;
2016 }
2017
2018 static void __exit null_exit(void)
2019 {
2020         struct nullb *nullb;
2021
2022         configfs_unregister_subsystem(&nullb_subsys);
2023
2024         unregister_blkdev(null_major, "nullb");
2025
2026         mutex_lock(&lock);
2027         while (!list_empty(&nullb_list)) {
2028                 struct nullb_device *dev;
2029
2030                 nullb = list_entry(nullb_list.next, struct nullb, list);
2031                 dev = nullb->dev;
2032                 null_del_dev(nullb);
2033                 null_free_dev(dev);
2034         }
2035         mutex_unlock(&lock);
2036
2037         if (g_queue_mode == NULL_Q_MQ && shared_tags)
2038                 blk_mq_free_tag_set(&tag_set);
2039
2040         kmem_cache_destroy(ppa_cache);
2041 }
2042
2043 module_init(null_init);
2044 module_exit(null_exit);
2045
2046 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2047 MODULE_LICENSE("GPL");