drivers/block/null_blk.c

   1 #include <linux/module.h>
   2
   3 #include <linux/moduleparam.h>
   4 #include <linux/sched.h>
   5 #include <linux/fs.h>
   6 #include <linux/blkdev.h>
   7 #include <linux/init.h>
   8 #include <linux/slab.h>
   9 #include <linux/blk-mq.h>
  10 #include <linux/hrtimer.h>
  11 #include <linux/lightnvm.h>
  12
  13 struct nullb_cmd {
  14         struct list_head list;
  15         struct llist_node ll_list;
  16         struct call_single_data csd;
  17         struct request *rq;
  18         struct bio *bio;
  19         unsigned int tag;
  20         struct nullb_queue *nq;
  21         struct hrtimer timer;
  22 };
  23
  24 struct nullb_queue {
  25         unsigned long *tag_map;
  26         wait_queue_head_t wait;
  27         unsigned int queue_depth;
  28
  29         struct nullb_cmd *cmds;
  30 };
  31
  32 struct nullb {
  33         struct list_head list;
  34         unsigned int index;
  35         struct request_queue *q;
  36         struct gendisk *disk;
  37         struct nvm_dev *ndev;
  38         struct blk_mq_tag_set tag_set;
  39         struct hrtimer timer;
  40         unsigned int queue_depth;
  41         spinlock_t lock;
  42
  43         struct nullb_queue *queues;
  44         unsigned int nr_queues;
  45         char disk_name[DISK_NAME_LEN];
  46 };
  47
  48 static LIST_HEAD(nullb_list);
  49 static struct mutex lock;
  50 static int null_major;
  51 static int nullb_indexes;
  52 static struct kmem_cache *ppa_cache;
  53
  54 enum {
  55         NULL_IRQ_NONE           = 0,
  56         NULL_IRQ_SOFTIRQ        = 1,
  57         NULL_IRQ_TIMER          = 2,
  58 };
  59
  60 enum {
  61         NULL_Q_BIO              = 0,
  62         NULL_Q_RQ               = 1,
  63         NULL_Q_MQ               = 2,
  64 };
  65
  66 static int submit_queues;
  67 module_param(submit_queues, int, S_IRUGO);
  68 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
  69
  70 static int home_node = NUMA_NO_NODE;
  71 module_param(home_node, int, S_IRUGO);
  72 MODULE_PARM_DESC(home_node, "Home node for the device");
  73
  74 static int queue_mode = NULL_Q_MQ;
  75
  76 static int null_param_store_val(const char *str, int *val, int min, int max)
  77 {
  78         int ret, new_val;
  79
  80         ret = kstrtoint(str, 10, &new_val);
  81         if (ret)
  82                 return -EINVAL;
  83
  84         if (new_val < min || new_val > max)
  85                 return -EINVAL;
  86
  87         *val = new_val;
  88         return 0;
  89 }
  90
  91 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
  92 {
  93         return null_param_store_val(str, &queue_mode, NULL_Q_BIO, NULL_Q_MQ);
  94 }
  95
  96 static const struct kernel_param_ops null_queue_mode_param_ops = {
  97         .set    = null_set_queue_mode,
  98         .get    = param_get_int,
  99 };
 100
 101 device_param_cb(queue_mode, &null_queue_mode_param_ops, &queue_mode, S_IRUGO);
 102 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
 103
 104 static int gb = 250;
 105 module_param(gb, int, S_IRUGO);
 106 MODULE_PARM_DESC(gb, "Size in GB");
 107
 108 static int bs = 512;
 109 module_param(bs, int, S_IRUGO);
 110 MODULE_PARM_DESC(bs, "Block size (in bytes)");
 111
 112 static int nr_devices = 2;
 113 module_param(nr_devices, int, S_IRUGO);
 114 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
 115
 116 static bool use_lightnvm;
 117 module_param(use_lightnvm, bool, S_IRUGO);
 118 MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
 119
 120 static int irqmode = NULL_IRQ_SOFTIRQ;
 121
 122 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
 123 {
 124         return null_param_store_val(str, &irqmode, NULL_IRQ_NONE,
 125                                         NULL_IRQ_TIMER);
 126 }
 127
 128 static const struct kernel_param_ops null_irqmode_param_ops = {
 129         .set    = null_set_irqmode,
 130         .get    = param_get_int,
 131 };
 132
 133 device_param_cb(irqmode, &null_irqmode_param_ops, &irqmode, S_IRUGO);
 134 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
 135
 136 static unsigned long completion_nsec = 10000;
 137 module_param(completion_nsec, ulong, S_IRUGO);
 138 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
 139
 140 static int hw_queue_depth = 64;
 141 module_param(hw_queue_depth, int, S_IRUGO);
 142 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
 143
 144 static bool use_per_node_hctx = false;
 145 module_param(use_per_node_hctx, bool, S_IRUGO);
 146 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
 147
 148 static void put_tag(struct nullb_queue *nq, unsigned int tag)
 149 {
 150         clear_bit_unlock(tag, nq->tag_map);
 151
 152         if (waitqueue_active(&nq->wait))
 153                 wake_up(&nq->wait);
 154 }
 155
 156 static unsigned int get_tag(struct nullb_queue *nq)
 157 {
 158         unsigned int tag;
 159
 160         do {
 161                 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
 162                 if (tag >= nq->queue_depth)
 163                         return -1U;
 164         } while (test_and_set_bit_lock(tag, nq->tag_map));
 165
 166         return tag;
 167 }
 168
 169 static void free_cmd(struct nullb_cmd *cmd)
 170 {
 171         put_tag(cmd->nq, cmd->tag);
 172 }
 173
 174 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
 175
 176 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 177 {
 178         struct nullb_cmd *cmd;
 179         unsigned int tag;
 180
 181         tag = get_tag(nq);
 182         if (tag != -1U) {
 183                 cmd = &nq->cmds[tag];
 184                 cmd->tag = tag;
 185                 cmd->nq = nq;
 186                 if (irqmode == NULL_IRQ_TIMER) {
 187                         hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
 188                                      HRTIMER_MODE_REL);
 189                         cmd->timer.function = null_cmd_timer_expired;
 190                 }
 191                 return cmd;
 192         }
 193
 194         return NULL;
 195 }
 196
 197 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
 198 {
 199         struct nullb_cmd *cmd;
 200         DEFINE_WAIT(wait);
 201
 202         cmd = __alloc_cmd(nq);
 203         if (cmd || !can_wait)
 204                 return cmd;
 205
 206         do {
 207                 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
 208                 cmd = __alloc_cmd(nq);
 209                 if (cmd)
 210                         break;
 211
 212                 io_schedule();
 213         } while (1);
 214
 215         finish_wait(&nq->wait, &wait);
 216         return cmd;
 217 }
 218
 219 static void end_cmd(struct nullb_cmd *cmd)
 220 {
 221         struct request_queue *q = NULL;
 222
 223         if (cmd->rq)
 224                 q = cmd->rq->q;
 225
 226         switch (queue_mode)  {
 227         case NULL_Q_MQ:
 228                 blk_mq_end_request(cmd->rq, 0);
 229                 return;
 230         case NULL_Q_RQ:
 231                 INIT_LIST_HEAD(&cmd->rq->queuelist);
 232                 blk_end_request_all(cmd->rq, 0);
 233                 break;
 234         case NULL_Q_BIO:
 235                 bio_endio(cmd->bio);
 236                 break;
 237         }
 238
 239         free_cmd(cmd);
 240
 241         /* Restart queue if needed, as we are freeing a tag */
 242         if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
 243                 unsigned long flags;
 244
 245                 spin_lock_irqsave(q->queue_lock, flags);
 246                 blk_start_queue_async(q);
 247                 spin_unlock_irqrestore(q->queue_lock, flags);
 248         }
 249 }
 250
 251 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
 252 {
 253         end_cmd(container_of(timer, struct nullb_cmd, timer));
 254
 255         return HRTIMER_NORESTART;
 256 }
 257
 258 static void null_cmd_end_timer(struct nullb_cmd *cmd)
 259 {
 260         ktime_t kt = ktime_set(0, completion_nsec);
 261
 262         hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
 263 }
 264
 265 static void null_softirq_done_fn(struct request *rq)
 266 {
 267         if (queue_mode == NULL_Q_MQ)
 268                 end_cmd(blk_mq_rq_to_pdu(rq));
 269         else
 270                 end_cmd(rq->special);
 271 }
 272
 273 static inline void null_handle_cmd(struct nullb_cmd *cmd)
 274 {
 275         /* Complete IO by inline, softirq or timer */
 276         switch (irqmode) {
 277         case NULL_IRQ_SOFTIRQ:
 278                 switch (queue_mode)  {
 279                 case NULL_Q_MQ:
 280                         blk_mq_complete_request(cmd->rq, cmd->rq->errors);
 281                         break;
 282                 case NULL_Q_RQ:
 283                         blk_complete_request(cmd->rq);
 284                         break;
 285                 case NULL_Q_BIO:
 286                         /*
 287                          * XXX: no proper submitting cpu information available.
 288                          */
 289                         end_cmd(cmd);
 290                         break;
 291                 }
 292                 break;
 293         case NULL_IRQ_NONE:
 294                 end_cmd(cmd);
 295                 break;
 296         case NULL_IRQ_TIMER:
 297                 null_cmd_end_timer(cmd);
 298                 break;
 299         }
 300 }
 301
 302 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
 303 {
 304         int index = 0;
 305
 306         if (nullb->nr_queues != 1)
 307                 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
 308
 309         return &nullb->queues[index];
 310 }
 311
 312 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
 313 {
 314         struct nullb *nullb = q->queuedata;
 315         struct nullb_queue *nq = nullb_to_queue(nullb);
 316         struct nullb_cmd *cmd;
 317
 318         cmd = alloc_cmd(nq, 1);
 319         cmd->bio = bio;
 320
 321         null_handle_cmd(cmd);
 322         return BLK_QC_T_NONE;
 323 }
 324
 325 static int null_rq_prep_fn(struct request_queue *q, struct request *req)
 326 {
 327         struct nullb *nullb = q->queuedata;
 328         struct nullb_queue *nq = nullb_to_queue(nullb);
 329         struct nullb_cmd *cmd;
 330
 331         cmd = alloc_cmd(nq, 0);
 332         if (cmd) {
 333                 cmd->rq = req;
 334                 req->special = cmd;
 335                 return BLKPREP_OK;
 336         }
 337         blk_stop_queue(q);
 338
 339         return BLKPREP_DEFER;
 340 }
 341
 342 static void null_request_fn(struct request_queue *q)
 343 {
 344         struct request *rq;
 345
 346         while ((rq = blk_fetch_request(q)) != NULL) {
 347                 struct nullb_cmd *cmd = rq->special;
 348
 349                 spin_unlock_irq(q->queue_lock);
 350                 null_handle_cmd(cmd);
 351                 spin_lock_irq(q->queue_lock);
 352         }
 353 }
 354
 355 static int null_queue_rq(struct blk_mq_hw_ctx *hctx,
 356                          const struct blk_mq_queue_data *bd)
 357 {
 358         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
 359
 360         if (irqmode == NULL_IRQ_TIMER) {
 361                 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 362                 cmd->timer.function = null_cmd_timer_expired;
 363         }
 364         cmd->rq = bd->rq;
 365         cmd->nq = hctx->driver_data;
 366
 367         blk_mq_start_request(bd->rq);
 368
 369         null_handle_cmd(cmd);
 370         return BLK_MQ_RQ_QUEUE_OK;
 371 }
 372
 373 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
 374 {
 375         BUG_ON(!nullb);
 376         BUG_ON(!nq);
 377
 378         init_waitqueue_head(&nq->wait);
 379         nq->queue_depth = nullb->queue_depth;
 380 }
 381
 382 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
 383                           unsigned int index)
 384 {
 385         struct nullb *nullb = data;
 386         struct nullb_queue *nq = &nullb->queues[index];
 387
 388         hctx->driver_data = nq;
 389         null_init_queue(nullb, nq);
 390         nullb->nr_queues++;
 391
 392         return 0;
 393 }
 394
 395 static struct blk_mq_ops null_mq_ops = {
 396         .queue_rq       = null_queue_rq,
 397         .init_hctx      = null_init_hctx,
 398         .complete       = null_softirq_done_fn,
 399 };
 400
 401 static void cleanup_queue(struct nullb_queue *nq)
 402 {
 403         kfree(nq->tag_map);
 404         kfree(nq->cmds);
 405 }
 406
 407 static void cleanup_queues(struct nullb *nullb)
 408 {
 409         int i;
 410
 411         for (i = 0; i < nullb->nr_queues; i++)
 412                 cleanup_queue(&nullb->queues[i]);
 413
 414         kfree(nullb->queues);
 415 }
 416
 417 #ifdef CONFIG_NVM
 418
 419 static void null_lnvm_end_io(struct request *rq, int error)
 420 {
 421         struct nvm_rq *rqd = rq->end_io_data;
 422
 423         nvm_end_io(rqd, error);
 424
 425         blk_put_request(rq);
 426 }
 427
 428 static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
 429 {
 430         struct request_queue *q = dev->q;
 431         struct request *rq;
 432         struct bio *bio = rqd->bio;
 433
 434         rq = blk_mq_alloc_request(q, bio_data_dir(bio), 0);
 435         if (IS_ERR(rq))
 436                 return -ENOMEM;
 437
 438         rq->cmd_type = REQ_TYPE_DRV_PRIV;
 439         rq->__sector = bio->bi_iter.bi_sector;
 440         rq->ioprio = bio_prio(bio);
 441
 442         if (bio_has_data(bio))
 443                 rq->nr_phys_segments = bio_phys_segments(q, bio);
 444
 445         rq->__data_len = bio->bi_iter.bi_size;
 446         rq->bio = rq->biotail = bio;
 447
 448         rq->end_io_data = rqd;
 449
 450         blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
 451
 452         return 0;
 453 }
 454
 455 static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
 456 {
 457         sector_t size = gb * 1024 * 1024 * 1024ULL;
 458         sector_t blksize;
 459         struct nvm_id_group *grp;
 460
 461         id->ver_id = 0x1;
 462         id->vmnt = 0;
 463         id->cgrps = 1;
 464         id->cap = 0x2;
 465         id->dom = 0x1;
 466
 467         id->ppaf.blk_offset = 0;
 468         id->ppaf.blk_len = 16;
 469         id->ppaf.pg_offset = 16;
 470         id->ppaf.pg_len = 16;
 471         id->ppaf.sect_offset = 32;
 472         id->ppaf.sect_len = 8;
 473         id->ppaf.pln_offset = 40;
 474         id->ppaf.pln_len = 8;
 475         id->ppaf.lun_offset = 48;
 476         id->ppaf.lun_len = 8;
 477         id->ppaf.ch_offset = 56;
 478         id->ppaf.ch_len = 8;
 479
 480         sector_div(size, bs); /* convert size to pages */
 481         size >>= 8; /* concert size to pgs pr blk */
 482         grp = &id->groups[0];
 483         grp->mtype = 0;
 484         grp->fmtype = 0;
 485         grp->num_ch = 1;
 486         grp->num_pg = 256;
 487         blksize = size;
 488         size >>= 16;
 489         grp->num_lun = size + 1;
 490         sector_div(blksize, grp->num_lun);
 491         grp->num_blk = blksize;
 492         grp->num_pln = 1;
 493
 494         grp->fpg_sz = bs;
 495         grp->csecs = bs;
 496         grp->trdt = 25000;
 497         grp->trdm = 25000;
 498         grp->tprt = 500000;
 499         grp->tprm = 500000;
 500         grp->tbet = 1500000;
 501         grp->tbem = 1500000;
 502         grp->mpos = 0x010101; /* single plane rwe */
 503         grp->cpar = hw_queue_depth;
 504
 505         return 0;
 506 }
 507
 508 static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
 509 {
 510         mempool_t *virtmem_pool;
 511
 512         virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
 513         if (!virtmem_pool) {
 514                 pr_err("null_blk: Unable to create virtual memory pool\n");
 515                 return NULL;
 516         }
 517
 518         return virtmem_pool;
 519 }
 520
 521 static void null_lnvm_destroy_dma_pool(void *pool)
 522 {
 523         mempool_destroy(pool);
 524 }
 525
 526 static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
 527                                 gfp_t mem_flags, dma_addr_t *dma_handler)
 528 {
 529         return mempool_alloc(pool, mem_flags);
 530 }
 531
 532 static void null_lnvm_dev_dma_free(void *pool, void *entry,
 533                                                         dma_addr_t dma_handler)
 534 {
 535         mempool_free(entry, pool);
 536 }
 537
 538 static struct nvm_dev_ops null_lnvm_dev_ops = {
 539         .identity               = null_lnvm_id,
 540         .submit_io              = null_lnvm_submit_io,
 541
 542         .create_dma_pool        = null_lnvm_create_dma_pool,
 543         .destroy_dma_pool       = null_lnvm_destroy_dma_pool,
 544         .dev_dma_alloc          = null_lnvm_dev_dma_alloc,
 545         .dev_dma_free           = null_lnvm_dev_dma_free,
 546
 547         /* Simulate nvme protocol restriction */
 548         .max_phys_sect          = 64,
 549 };
 550
 551 static int null_nvm_register(struct nullb *nullb)
 552 {
 553         struct nvm_dev *dev;
 554         int rv;
 555
 556         dev = nvm_alloc_dev(0);
 557         if (!dev)
 558                 return -ENOMEM;
 559
 560         dev->q = nullb->q;
 561         memcpy(dev->name, nullb->disk_name, DISK_NAME_LEN);
 562         dev->ops = &null_lnvm_dev_ops;
 563
 564         rv = nvm_register(dev);
 565         if (rv) {
 566                 kfree(dev);
 567                 return rv;
 568         }
 569         nullb->ndev = dev;
 570         return 0;
 571 }
 572
 573 static void null_nvm_unregister(struct nullb *nullb)
 574 {
 575         nvm_unregister(nullb->ndev);
 576 }
 577 #else
 578 static int null_nvm_register(struct nullb *nullb)
 579 {
 580         return -EINVAL;
 581 }
 582 static void null_nvm_unregister(struct nullb *nullb) {}
 583 #endif /* CONFIG_NVM */
 584
 585 static void null_del_dev(struct nullb *nullb)
 586 {
 587         list_del_init(&nullb->list);
 588
 589         if (use_lightnvm)
 590                 null_nvm_unregister(nullb);
 591         else
 592                 del_gendisk(nullb->disk);
 593         blk_cleanup_queue(nullb->q);
 594         if (queue_mode == NULL_Q_MQ)
 595                 blk_mq_free_tag_set(&nullb->tag_set);
 596         if (!use_lightnvm)
 597                 put_disk(nullb->disk);
 598         cleanup_queues(nullb);
 599         kfree(nullb);
 600 }
 601
 602 static int null_open(struct block_device *bdev, fmode_t mode)
 603 {
 604         return 0;
 605 }
 606
 607 static void null_release(struct gendisk *disk, fmode_t mode)
 608 {
 609 }
 610
 611 static const struct block_device_operations null_fops = {
 612         .owner =        THIS_MODULE,
 613         .open =         null_open,
 614         .release =      null_release,
 615 };
 616
 617 static int setup_commands(struct nullb_queue *nq)
 618 {
 619         struct nullb_cmd *cmd;
 620         int i, tag_size;
 621
 622         nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
 623         if (!nq->cmds)
 624                 return -ENOMEM;
 625
 626         tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
 627         nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
 628         if (!nq->tag_map) {
 629                 kfree(nq->cmds);
 630                 return -ENOMEM;
 631         }
 632
 633         for (i = 0; i < nq->queue_depth; i++) {
 634                 cmd = &nq->cmds[i];
 635                 INIT_LIST_HEAD(&cmd->list);
 636                 cmd->ll_list.next = NULL;
 637                 cmd->tag = -1U;
 638         }
 639
 640         return 0;
 641 }
 642
 643 static int setup_queues(struct nullb *nullb)
 644 {
 645         nullb->queues = kzalloc(submit_queues * sizeof(struct nullb_queue),
 646                                                                 GFP_KERNEL);
 647         if (!nullb->queues)
 648                 return -ENOMEM;
 649
 650         nullb->nr_queues = 0;
 651         nullb->queue_depth = hw_queue_depth;
 652
 653         return 0;
 654 }
 655
 656 static int init_driver_queues(struct nullb *nullb)
 657 {
 658         struct nullb_queue *nq;
 659         int i, ret = 0;
 660
 661         for (i = 0; i < submit_queues; i++) {
 662                 nq = &nullb->queues[i];
 663
 664                 null_init_queue(nullb, nq);
 665
 666                 ret = setup_commands(nq);
 667                 if (ret)
 668                         return ret;
 669                 nullb->nr_queues++;
 670         }
 671         return 0;
 672 }
 673
 674 static int null_gendisk_register(struct nullb *nullb)
 675 {
 676         struct gendisk *disk;
 677         sector_t size;
 678
 679         disk = nullb->disk = alloc_disk_node(1, home_node);
 680         if (!disk)
 681                 return -ENOMEM;
 682         size = gb * 1024 * 1024 * 1024ULL;
 683         set_capacity(disk, size >> 9);
 684
 685         disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
 686         disk->major             = null_major;
 687         disk->first_minor       = nullb->index;
 688         disk->fops              = &null_fops;
 689         disk->private_data      = nullb;
 690         disk->queue             = nullb->q;
 691         strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
 692
 693         add_disk(disk);
 694         return 0;
 695 }
 696
 697 static int null_add_dev(void)
 698 {
 699         struct nullb *nullb;
 700         int rv;
 701
 702         nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, home_node);
 703         if (!nullb) {
 704                 rv = -ENOMEM;
 705                 goto out;
 706         }
 707
 708         spin_lock_init(&nullb->lock);
 709
 710         if (queue_mode == NULL_Q_MQ && use_per_node_hctx)
 711                 submit_queues = nr_online_nodes;
 712
 713         rv = setup_queues(nullb);
 714         if (rv)
 715                 goto out_free_nullb;
 716
 717         if (queue_mode == NULL_Q_MQ) {
 718                 nullb->tag_set.ops = &null_mq_ops;
 719                 nullb->tag_set.nr_hw_queues = submit_queues;
 720                 nullb->tag_set.queue_depth = hw_queue_depth;
 721                 nullb->tag_set.numa_node = home_node;
 722                 nullb->tag_set.cmd_size = sizeof(struct nullb_cmd);
 723                 nullb->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 724                 nullb->tag_set.driver_data = nullb;
 725
 726                 rv = blk_mq_alloc_tag_set(&nullb->tag_set);
 727                 if (rv)
 728                         goto out_cleanup_queues;
 729
 730                 nullb->q = blk_mq_init_queue(&nullb->tag_set);
 731                 if (IS_ERR(nullb->q)) {
 732                         rv = -ENOMEM;
 733                         goto out_cleanup_tags;
 734                 }
 735         } else if (queue_mode == NULL_Q_BIO) {
 736                 nullb->q = blk_alloc_queue_node(GFP_KERNEL, home_node);
 737                 if (!nullb->q) {
 738                         rv = -ENOMEM;
 739                         goto out_cleanup_queues;
 740                 }
 741                 blk_queue_make_request(nullb->q, null_queue_bio);
 742                 rv = init_driver_queues(nullb);
 743                 if (rv)
 744                         goto out_cleanup_blk_queue;
 745         } else {
 746                 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock, home_node);
 747                 if (!nullb->q) {
 748                         rv = -ENOMEM;
 749                         goto out_cleanup_queues;
 750                 }
 751                 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
 752                 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
 753                 rv = init_driver_queues(nullb);
 754                 if (rv)
 755                         goto out_cleanup_blk_queue;
 756         }
 757
 758         nullb->q->queuedata = nullb;
 759         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
 760         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
 761
 762         mutex_lock(&lock);
 763         nullb->index = nullb_indexes++;
 764         mutex_unlock(&lock);
 765
 766         blk_queue_logical_block_size(nullb->q, bs);
 767         blk_queue_physical_block_size(nullb->q, bs);
 768
 769         sprintf(nullb->disk_name, "nullb%d", nullb->index);
 770
 771         if (use_lightnvm)
 772                 rv = null_nvm_register(nullb);
 773         else
 774                 rv = null_gendisk_register(nullb);
 775
 776         if (rv)
 777                 goto out_cleanup_blk_queue;
 778
 779         mutex_lock(&lock);
 780         list_add_tail(&nullb->list, &nullb_list);
 781         mutex_unlock(&lock);
 782
 783         return 0;
 784 out_cleanup_blk_queue:
 785         blk_cleanup_queue(nullb->q);
 786 out_cleanup_tags:
 787         if (queue_mode == NULL_Q_MQ)
 788                 blk_mq_free_tag_set(&nullb->tag_set);
 789 out_cleanup_queues:
 790         cleanup_queues(nullb);
 791 out_free_nullb:
 792         kfree(nullb);
 793 out:
 794         return rv;
 795 }
 796
 797 static int __init null_init(void)
 798 {
 799         int ret = 0;
 800         unsigned int i;
 801         struct nullb *nullb;
 802
 803         if (bs > PAGE_SIZE) {
 804                 pr_warn("null_blk: invalid block size\n");
 805                 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
 806                 bs = PAGE_SIZE;
 807         }
 808
 809         if (use_lightnvm && bs != 4096) {
 810                 pr_warn("null_blk: LightNVM only supports 4k block size\n");
 811                 pr_warn("null_blk: defaults block size to 4k\n");
 812                 bs = 4096;
 813         }
 814
 815         if (use_lightnvm && queue_mode != NULL_Q_MQ) {
 816                 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
 817                 pr_warn("null_blk: defaults queue mode to blk-mq\n");
 818                 queue_mode = NULL_Q_MQ;
 819         }
 820
 821         if (queue_mode == NULL_Q_MQ && use_per_node_hctx) {
 822                 if (submit_queues < nr_online_nodes) {
 823                         pr_warn("null_blk: submit_queues param is set to %u.",
 824                                                         nr_online_nodes);
 825                         submit_queues = nr_online_nodes;
 826                 }
 827         } else if (submit_queues > nr_cpu_ids)
 828                 submit_queues = nr_cpu_ids;
 829         else if (!submit_queues)
 830                 submit_queues = 1;
 831
 832         mutex_init(&lock);
 833
 834         null_major = register_blkdev(0, "nullb");
 835         if (null_major < 0)
 836                 return null_major;
 837
 838         if (use_lightnvm) {
 839                 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
 840                                                                 0, 0, NULL);
 841                 if (!ppa_cache) {
 842                         pr_err("null_blk: unable to create ppa cache\n");
 843                         ret = -ENOMEM;
 844                         goto err_ppa;
 845                 }
 846         }
 847
 848         for (i = 0; i < nr_devices; i++) {
 849                 ret = null_add_dev();
 850                 if (ret)
 851                         goto err_dev;
 852         }
 853
 854         pr_info("null: module loaded\n");
 855         return 0;
 856
 857 err_dev:
 858         while (!list_empty(&nullb_list)) {
 859                 nullb = list_entry(nullb_list.next, struct nullb, list);
 860                 null_del_dev(nullb);
 861         }
 862         kmem_cache_destroy(ppa_cache);
 863 err_ppa:
 864         unregister_blkdev(null_major, "nullb");
 865         return ret;
 866 }
 867
 868 static void __exit null_exit(void)
 869 {
 870         struct nullb *nullb;
 871
 872         unregister_blkdev(null_major, "nullb");
 873
 874         mutex_lock(&lock);
 875         while (!list_empty(&nullb_list)) {
 876                 nullb = list_entry(nullb_list.next, struct nullb, list);
 877                 null_del_dev(nullb);
 878         }
 879         mutex_unlock(&lock);
 880
 881         kmem_cache_destroy(ppa_cache);
 882 }
 883
 884 module_init(null_init);
 885 module_exit(null_exit);
 886
 887 MODULE_AUTHOR("Jens Axboe <jaxboe@fusionio.com>");
 888 MODULE_LICENSE("GPL");