drivers/block/virtio_blk.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 //#define DEBUG
   3 #include <linux/spinlock.h>
   4 #include <linux/slab.h>
   5 #include <linux/blkdev.h>
   6 #include <linux/hdreg.h>
   7 #include <linux/module.h>
   8 #include <linux/mutex.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/virtio.h>
  11 #include <linux/virtio_blk.h>
  12 #include <linux/scatterlist.h>
  13 #include <linux/string_helpers.h>
  14 #include <linux/idr.h>
  15 #include <linux/blk-mq.h>
  16 #include <linux/blk-mq-virtio.h>
  17 #include <linux/numa.h>
  18 #include <uapi/linux/virtio_ring.h>
  19
  20 #define PART_BITS 4
  21 #define VQ_NAME_LEN 16
  22 #define MAX_DISCARD_SEGMENTS 256u
  23
  24 /* The maximum number of sg elements that fit into a virtqueue */
  25 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
  26
  27 #ifdef CONFIG_ARCH_NO_SG_CHAIN
  28 #define VIRTIO_BLK_INLINE_SG_CNT        0
  29 #else
  30 #define VIRTIO_BLK_INLINE_SG_CNT        2
  31 #endif
  32
  33 static unsigned int num_request_queues;
  34 module_param(num_request_queues, uint, 0644);
  35 MODULE_PARM_DESC(num_request_queues,
  36                  "Limit the number of request queues to use for blk device. "
  37                  "0 for no limit. "
  38                  "Values > nr_cpu_ids truncated to nr_cpu_ids.");
  39
  40 static unsigned int poll_queues;
  41 module_param(poll_queues, uint, 0644);
  42 MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/O");
  43
  44 static int major;
  45 static DEFINE_IDA(vd_index_ida);
  46
  47 static struct workqueue_struct *virtblk_wq;
  48
  49 struct virtio_blk_vq {
  50         struct virtqueue *vq;
  51         spinlock_t lock;
  52         char name[VQ_NAME_LEN];
  53 } ____cacheline_aligned_in_smp;
  54
  55 struct virtio_blk {
  56         /*
  57          * This mutex must be held by anything that may run after
  58          * virtblk_remove() sets vblk->vdev to NULL.
  59          *
  60          * blk-mq, virtqueue processing, and sysfs attribute code paths are
  61          * shut down before vblk->vdev is set to NULL and therefore do not need
  62          * to hold this mutex.
  63          */
  64         struct mutex vdev_mutex;
  65         struct virtio_device *vdev;
  66
  67         /* The disk structure for the kernel. */
  68         struct gendisk *disk;
  69
  70         /* Block layer tags. */
  71         struct blk_mq_tag_set tag_set;
  72
  73         /* Process context for config space updates */
  74         struct work_struct config_work;
  75
  76         /* Ida index - used to track minor number allocations. */
  77         int index;
  78
  79         /* num of vqs */
  80         int num_vqs;
  81         int io_queues[HCTX_MAX_TYPES];
  82         struct virtio_blk_vq *vqs;
  83 };
  84
  85 struct virtblk_req {
  86         struct virtio_blk_outhdr out_hdr;
  87         u8 status;
  88         struct sg_table sg_table;
  89         struct scatterlist sg[];
  90 };
  91
  92 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
  93 {
  94         switch (vbr->status) {
  95         case VIRTIO_BLK_S_OK:
  96                 return BLK_STS_OK;
  97         case VIRTIO_BLK_S_UNSUPP:
  98                 return BLK_STS_NOTSUPP;
  99         default:
 100                 return BLK_STS_IOERR;
 101         }
 102 }
 103
 104 static inline struct virtio_blk_vq *get_virtio_blk_vq(struct blk_mq_hw_ctx *hctx)
 105 {
 106         struct virtio_blk *vblk = hctx->queue->queuedata;
 107         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 108
 109         return vq;
 110 }
 111
 112 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr)
 113 {
 114         struct scatterlist hdr, status, *sgs[3];
 115         unsigned int num_out = 0, num_in = 0;
 116
 117         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
 118         sgs[num_out++] = &hdr;
 119
 120         if (vbr->sg_table.nents) {
 121                 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
 122                         sgs[num_out++] = vbr->sg_table.sgl;
 123                 else
 124                         sgs[num_out + num_in++] = vbr->sg_table.sgl;
 125         }
 126
 127         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
 128         sgs[num_out + num_in++] = &status;
 129
 130         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 131 }
 132
 133 static int virtblk_setup_discard_write_zeroes_erase(struct request *req, bool unmap)
 134 {
 135         unsigned short segments = blk_rq_nr_discard_segments(req);
 136         unsigned short n = 0;
 137         struct virtio_blk_discard_write_zeroes *range;
 138         struct bio *bio;
 139         u32 flags = 0;
 140
 141         if (unmap)
 142                 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
 143
 144         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 145         if (!range)
 146                 return -ENOMEM;
 147
 148         /*
 149          * Single max discard segment means multi-range discard isn't
 150          * supported, and block layer only runs contiguity merge like
 151          * normal RW request. So we can't reply on bio for retrieving
 152          * each range info.
 153          */
 154         if (queue_max_discard_segments(req->q) == 1) {
 155                 range[0].flags = cpu_to_le32(flags);
 156                 range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
 157                 range[0].sector = cpu_to_le64(blk_rq_pos(req));
 158                 n = 1;
 159         } else {
 160                 __rq_for_each_bio(bio, req) {
 161                         u64 sector = bio->bi_iter.bi_sector;
 162                         u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
 163
 164                         range[n].flags = cpu_to_le32(flags);
 165                         range[n].num_sectors = cpu_to_le32(num_sectors);
 166                         range[n].sector = cpu_to_le64(sector);
 167                         n++;
 168                 }
 169         }
 170
 171         WARN_ON_ONCE(n != segments);
 172
 173         req->special_vec.bv_page = virt_to_page(range);
 174         req->special_vec.bv_offset = offset_in_page(range);
 175         req->special_vec.bv_len = sizeof(*range) * segments;
 176         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 177
 178         return 0;
 179 }
 180
 181 static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
 182 {
 183         if (blk_rq_nr_phys_segments(req))
 184                 sg_free_table_chained(&vbr->sg_table,
 185                                       VIRTIO_BLK_INLINE_SG_CNT);
 186 }
 187
 188 static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
 189                 struct virtblk_req *vbr)
 190 {
 191         int err;
 192
 193         if (!blk_rq_nr_phys_segments(req))
 194                 return 0;
 195
 196         vbr->sg_table.sgl = vbr->sg;
 197         err = sg_alloc_table_chained(&vbr->sg_table,
 198                                      blk_rq_nr_phys_segments(req),
 199                                      vbr->sg_table.sgl,
 200                                      VIRTIO_BLK_INLINE_SG_CNT);
 201         if (unlikely(err))
 202                 return -ENOMEM;
 203
 204         return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl);
 205 }
 206
 207 static void virtblk_cleanup_cmd(struct request *req)
 208 {
 209         if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
 210                 kfree(bvec_virt(&req->special_vec));
 211 }
 212
 213 static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 214                                       struct request *req,
 215                                       struct virtblk_req *vbr)
 216 {
 217         bool unmap = false;
 218         u32 type;
 219
 220         vbr->out_hdr.sector = 0;
 221
 222         switch (req_op(req)) {
 223         case REQ_OP_READ:
 224                 type = VIRTIO_BLK_T_IN;
 225                 vbr->out_hdr.sector = cpu_to_virtio64(vdev,
 226                                                       blk_rq_pos(req));
 227                 break;
 228         case REQ_OP_WRITE:
 229                 type = VIRTIO_BLK_T_OUT;
 230                 vbr->out_hdr.sector = cpu_to_virtio64(vdev,
 231                                                       blk_rq_pos(req));
 232                 break;
 233         case REQ_OP_FLUSH:
 234                 type = VIRTIO_BLK_T_FLUSH;
 235                 break;
 236         case REQ_OP_DISCARD:
 237                 type = VIRTIO_BLK_T_DISCARD;
 238                 break;
 239         case REQ_OP_WRITE_ZEROES:
 240                 type = VIRTIO_BLK_T_WRITE_ZEROES;
 241                 unmap = !(req->cmd_flags & REQ_NOUNMAP);
 242                 break;
 243         case REQ_OP_SECURE_ERASE:
 244                 type = VIRTIO_BLK_T_SECURE_ERASE;
 245                 break;
 246         case REQ_OP_DRV_IN:
 247                 type = VIRTIO_BLK_T_GET_ID;
 248                 break;
 249         default:
 250                 WARN_ON_ONCE(1);
 251                 return BLK_STS_IOERR;
 252         }
 253
 254         vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
 255         vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
 256
 257         if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
 258             type == VIRTIO_BLK_T_SECURE_ERASE) {
 259                 if (virtblk_setup_discard_write_zeroes_erase(req, unmap))
 260                         return BLK_STS_RESOURCE;
 261         }
 262
 263         return 0;
 264 }
 265
 266 static inline void virtblk_request_done(struct request *req)
 267 {
 268         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 269
 270         virtblk_unmap_data(req, vbr);
 271         virtblk_cleanup_cmd(req);
 272         blk_mq_end_request(req, virtblk_result(vbr));
 273 }
 274
 275 static void virtblk_done(struct virtqueue *vq)
 276 {
 277         struct virtio_blk *vblk = vq->vdev->priv;
 278         bool req_done = false;
 279         int qid = vq->index;
 280         struct virtblk_req *vbr;
 281         unsigned long flags;
 282         unsigned int len;
 283
 284         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 285         do {
 286                 virtqueue_disable_cb(vq);
 287                 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
 288                         struct request *req = blk_mq_rq_from_pdu(vbr);
 289
 290                         if (likely(!blk_should_fake_timeout(req->q)))
 291                                 blk_mq_complete_request(req);
 292                         req_done = true;
 293                 }
 294                 if (unlikely(virtqueue_is_broken(vq)))
 295                         break;
 296         } while (!virtqueue_enable_cb(vq));
 297
 298         /* In case queue is stopped waiting for more buffers. */
 299         if (req_done)
 300                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 301         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 302 }
 303
 304 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 305 {
 306         struct virtio_blk *vblk = hctx->queue->queuedata;
 307         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 308         bool kick;
 309
 310         spin_lock_irq(&vq->lock);
 311         kick = virtqueue_kick_prepare(vq->vq);
 312         spin_unlock_irq(&vq->lock);
 313
 314         if (kick)
 315                 virtqueue_notify(vq->vq);
 316 }
 317
 318 static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
 319                                         struct virtio_blk *vblk,
 320                                         struct request *req,
 321                                         struct virtblk_req *vbr)
 322 {
 323         blk_status_t status;
 324
 325         status = virtblk_setup_cmd(vblk->vdev, req, vbr);
 326         if (unlikely(status))
 327                 return status;
 328
 329         vbr->sg_table.nents = virtblk_map_data(hctx, req, vbr);
 330         if (unlikely(vbr->sg_table.nents < 0)) {
 331                 virtblk_cleanup_cmd(req);
 332                 return BLK_STS_RESOURCE;
 333         }
 334
 335         blk_mq_start_request(req);
 336
 337         return BLK_STS_OK;
 338 }
 339
 340 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 341                            const struct blk_mq_queue_data *bd)
 342 {
 343         struct virtio_blk *vblk = hctx->queue->queuedata;
 344         struct request *req = bd->rq;
 345         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 346         unsigned long flags;
 347         int qid = hctx->queue_num;
 348         bool notify = false;
 349         blk_status_t status;
 350         int err;
 351
 352         status = virtblk_prep_rq(hctx, vblk, req, vbr);
 353         if (unlikely(status))
 354                 return status;
 355
 356         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 357         err = virtblk_add_req(vblk->vqs[qid].vq, vbr);
 358         if (err) {
 359                 virtqueue_kick(vblk->vqs[qid].vq);
 360                 /* Don't stop the queue if -ENOMEM: we may have failed to
 361                  * bounce the buffer due to global resource outage.
 362                  */
 363                 if (err == -ENOSPC)
 364                         blk_mq_stop_hw_queue(hctx);
 365                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 366                 virtblk_unmap_data(req, vbr);
 367                 virtblk_cleanup_cmd(req);
 368                 switch (err) {
 369                 case -ENOSPC:
 370                         return BLK_STS_DEV_RESOURCE;
 371                 case -ENOMEM:
 372                         return BLK_STS_RESOURCE;
 373                 default:
 374                         return BLK_STS_IOERR;
 375                 }
 376         }
 377
 378         if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 379                 notify = true;
 380         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 381
 382         if (notify)
 383                 virtqueue_notify(vblk->vqs[qid].vq);
 384         return BLK_STS_OK;
 385 }
 386
 387 static bool virtblk_prep_rq_batch(struct request *req)
 388 {
 389         struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
 390         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 391
 392         req->mq_hctx->tags->rqs[req->tag] = req;
 393
 394         return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
 395 }
 396
 397 static bool virtblk_add_req_batch(struct virtio_blk_vq *vq,
 398                                         struct request **rqlist)
 399 {
 400         unsigned long flags;
 401         int err;
 402         bool kick;
 403
 404         spin_lock_irqsave(&vq->lock, flags);
 405
 406         while (!rq_list_empty(*rqlist)) {
 407                 struct request *req = rq_list_pop(rqlist);
 408                 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 409
 410                 err = virtblk_add_req(vq->vq, vbr);
 411                 if (err) {
 412                         virtblk_unmap_data(req, vbr);
 413                         virtblk_cleanup_cmd(req);
 414                         blk_mq_requeue_request(req, true);
 415                 }
 416         }
 417
 418         kick = virtqueue_kick_prepare(vq->vq);
 419         spin_unlock_irqrestore(&vq->lock, flags);
 420
 421         return kick;
 422 }
 423
 424 static void virtio_queue_rqs(struct request **rqlist)
 425 {
 426         struct request *req, *next, *prev = NULL;
 427         struct request *requeue_list = NULL;
 428
 429         rq_list_for_each_safe(rqlist, req, next) {
 430                 struct virtio_blk_vq *vq = get_virtio_blk_vq(req->mq_hctx);
 431                 bool kick;
 432
 433                 if (!virtblk_prep_rq_batch(req)) {
 434                         rq_list_move(rqlist, &requeue_list, req, prev);
 435                         req = prev;
 436                         if (!req)
 437                                 continue;
 438                 }
 439
 440                 if (!next || req->mq_hctx != next->mq_hctx) {
 441                         req->rq_next = NULL;
 442                         kick = virtblk_add_req_batch(vq, rqlist);
 443                         if (kick)
 444                                 virtqueue_notify(vq->vq);
 445
 446                         *rqlist = next;
 447                         prev = NULL;
 448                 } else
 449                         prev = req;
 450         }
 451
 452         *rqlist = requeue_list;
 453 }
 454
 455 /* return id (s/n) string for *disk to *id_str
 456  */
 457 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 458 {
 459         struct virtio_blk *vblk = disk->private_data;
 460         struct request_queue *q = vblk->disk->queue;
 461         struct request *req;
 462         int err;
 463
 464         req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
 465         if (IS_ERR(req))
 466                 return PTR_ERR(req);
 467
 468         err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 469         if (err)
 470                 goto out;
 471
 472         blk_execute_rq(req, false);
 473         err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
 474 out:
 475         blk_mq_free_request(req);
 476         return err;
 477 }
 478
 479 /* We provide getgeo only to please some old bootloader/partitioning tools */
 480 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 481 {
 482         struct virtio_blk *vblk = bd->bd_disk->private_data;
 483         int ret = 0;
 484
 485         mutex_lock(&vblk->vdev_mutex);
 486
 487         if (!vblk->vdev) {
 488                 ret = -ENXIO;
 489                 goto out;
 490         }
 491
 492         /* see if the host passed in geometry config */
 493         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 494                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 495                              geometry.cylinders, &geo->cylinders);
 496                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 497                              geometry.heads, &geo->heads);
 498                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 499                              geometry.sectors, &geo->sectors);
 500         } else {
 501                 /* some standard values, similar to sd */
 502                 geo->heads = 1 << 6;
 503                 geo->sectors = 1 << 5;
 504                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 505         }
 506 out:
 507         mutex_unlock(&vblk->vdev_mutex);
 508         return ret;
 509 }
 510
 511 static void virtblk_free_disk(struct gendisk *disk)
 512 {
 513         struct virtio_blk *vblk = disk->private_data;
 514
 515         ida_simple_remove(&vd_index_ida, vblk->index);
 516         mutex_destroy(&vblk->vdev_mutex);
 517         kfree(vblk);
 518 }
 519
 520 static const struct block_device_operations virtblk_fops = {
 521         .owner          = THIS_MODULE,
 522         .getgeo         = virtblk_getgeo,
 523         .free_disk      = virtblk_free_disk,
 524 };
 525
 526 static int index_to_minor(int index)
 527 {
 528         return index << PART_BITS;
 529 }
 530
 531 static int minor_to_index(int minor)
 532 {
 533         return minor >> PART_BITS;
 534 }
 535
 536 static ssize_t serial_show(struct device *dev,
 537                            struct device_attribute *attr, char *buf)
 538 {
 539         struct gendisk *disk = dev_to_disk(dev);
 540         int err;
 541
 542         /* sysfs gives us a PAGE_SIZE buffer */
 543         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 544
 545         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 546         err = virtblk_get_id(disk, buf);
 547         if (!err)
 548                 return strlen(buf);
 549
 550         if (err == -EIO) /* Unsupported? Make it empty. */
 551                 return 0;
 552
 553         return err;
 554 }
 555
 556 static DEVICE_ATTR_RO(serial);
 557
 558 /* The queue's logical block size must be set before calling this */
 559 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 560 {
 561         struct virtio_device *vdev = vblk->vdev;
 562         struct request_queue *q = vblk->disk->queue;
 563         char cap_str_2[10], cap_str_10[10];
 564         unsigned long long nblocks;
 565         u64 capacity;
 566
 567         /* Host must always specify the capacity. */
 568         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 569
 570         nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
 571
 572         string_get_size(nblocks, queue_logical_block_size(q),
 573                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 574         string_get_size(nblocks, queue_logical_block_size(q),
 575                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 576
 577         dev_notice(&vdev->dev,
 578                    "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
 579                    vblk->disk->disk_name,
 580                    resize ? "new size: " : "",
 581                    nblocks,
 582                    queue_logical_block_size(q),
 583                    cap_str_10,
 584                    cap_str_2);
 585
 586         set_capacity_and_notify(vblk->disk, capacity);
 587 }
 588
 589 static void virtblk_config_changed_work(struct work_struct *work)
 590 {
 591         struct virtio_blk *vblk =
 592                 container_of(work, struct virtio_blk, config_work);
 593
 594         virtblk_update_capacity(vblk, true);
 595 }
 596
 597 static void virtblk_config_changed(struct virtio_device *vdev)
 598 {
 599         struct virtio_blk *vblk = vdev->priv;
 600
 601         queue_work(virtblk_wq, &vblk->config_work);
 602 }
 603
 604 static int init_vq(struct virtio_blk *vblk)
 605 {
 606         int err;
 607         int i;
 608         vq_callback_t **callbacks;
 609         const char **names;
 610         struct virtqueue **vqs;
 611         unsigned short num_vqs;
 612         unsigned int num_poll_vqs;
 613         struct virtio_device *vdev = vblk->vdev;
 614         struct irq_affinity desc = { 0, };
 615
 616         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 617                                    struct virtio_blk_config, num_queues,
 618                                    &num_vqs);
 619         if (err)
 620                 num_vqs = 1;
 621
 622         if (!err && !num_vqs) {
 623                 dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
 624                 return -EINVAL;
 625         }
 626
 627         num_vqs = min_t(unsigned int,
 628                         min_not_zero(num_request_queues, nr_cpu_ids),
 629                         num_vqs);
 630
 631         num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
 632
 633         vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
 634         vblk->io_queues[HCTX_TYPE_READ] = 0;
 635         vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
 636
 637         dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
 638                                 vblk->io_queues[HCTX_TYPE_DEFAULT],
 639                                 vblk->io_queues[HCTX_TYPE_READ],
 640                                 vblk->io_queues[HCTX_TYPE_POLL]);
 641
 642         vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
 643         if (!vblk->vqs)
 644                 return -ENOMEM;
 645
 646         names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
 647         callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
 648         vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
 649         if (!names || !callbacks || !vqs) {
 650                 err = -ENOMEM;
 651                 goto out;
 652         }
 653
 654         for (i = 0; i < num_vqs - num_poll_vqs; i++) {
 655                 callbacks[i] = virtblk_done;
 656                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
 657                 names[i] = vblk->vqs[i].name;
 658         }
 659
 660         for (; i < num_vqs; i++) {
 661                 callbacks[i] = NULL;
 662                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i);
 663                 names[i] = vblk->vqs[i].name;
 664         }
 665
 666         /* Discover virtqueues and write information to configuration.  */
 667         err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
 668         if (err)
 669                 goto out;
 670
 671         for (i = 0; i < num_vqs; i++) {
 672                 spin_lock_init(&vblk->vqs[i].lock);
 673                 vblk->vqs[i].vq = vqs[i];
 674         }
 675         vblk->num_vqs = num_vqs;
 676
 677 out:
 678         kfree(vqs);
 679         kfree(callbacks);
 680         kfree(names);
 681         if (err)
 682                 kfree(vblk->vqs);
 683         return err;
 684 }
 685
 686 /*
 687  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 688  * virtio blk but don't ever use it for any new driver.
 689  */
 690 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 691 {
 692         const int base = 'z' - 'a' + 1;
 693         char *begin = buf + strlen(prefix);
 694         char *end = buf + buflen;
 695         char *p;
 696         int unit;
 697
 698         p = end - 1;
 699         *p = '\0';
 700         unit = base;
 701         do {
 702                 if (p == begin)
 703                         return -EINVAL;
 704                 *--p = 'a' + (index % unit);
 705                 index = (index / unit) - 1;
 706         } while (index >= 0);
 707
 708         memmove(begin, p, end - p);
 709         memcpy(buf, prefix, strlen(prefix));
 710
 711         return 0;
 712 }
 713
 714 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 715 {
 716         u8 writeback;
 717         int err;
 718
 719         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 720                                    struct virtio_blk_config, wce,
 721                                    &writeback);
 722
 723         /*
 724          * If WCE is not configurable and flush is not available,
 725          * assume no writeback cache is in use.
 726          */
 727         if (err)
 728                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
 729
 730         return writeback;
 731 }
 732
 733 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 734 {
 735         u8 writeback = virtblk_get_cache_mode(vdev);
 736         struct virtio_blk *vblk = vdev->priv;
 737
 738         blk_queue_write_cache(vblk->disk->queue, writeback, false);
 739 }
 740
 741 static const char *const virtblk_cache_types[] = {
 742         "write through", "write back"
 743 };
 744
 745 static ssize_t
 746 cache_type_store(struct device *dev, struct device_attribute *attr,
 747                  const char *buf, size_t count)
 748 {
 749         struct gendisk *disk = dev_to_disk(dev);
 750         struct virtio_blk *vblk = disk->private_data;
 751         struct virtio_device *vdev = vblk->vdev;
 752         int i;
 753
 754         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 755         i = sysfs_match_string(virtblk_cache_types, buf);
 756         if (i < 0)
 757                 return i;
 758
 759         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
 760         virtblk_update_cache_mode(vdev);
 761         return count;
 762 }
 763
 764 static ssize_t
 765 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 766 {
 767         struct gendisk *disk = dev_to_disk(dev);
 768         struct virtio_blk *vblk = disk->private_data;
 769         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 770
 771         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 772         return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
 773 }
 774
 775 static DEVICE_ATTR_RW(cache_type);
 776
 777 static struct attribute *virtblk_attrs[] = {
 778         &dev_attr_serial.attr,
 779         &dev_attr_cache_type.attr,
 780         NULL,
 781 };
 782
 783 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
 784                 struct attribute *a, int n)
 785 {
 786         struct device *dev = kobj_to_dev(kobj);
 787         struct gendisk *disk = dev_to_disk(dev);
 788         struct virtio_blk *vblk = disk->private_data;
 789         struct virtio_device *vdev = vblk->vdev;
 790
 791         if (a == &dev_attr_cache_type.attr &&
 792             !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 793                 return S_IRUGO;
 794
 795         return a->mode;
 796 }
 797
 798 static const struct attribute_group virtblk_attr_group = {
 799         .attrs = virtblk_attrs,
 800         .is_visible = virtblk_attrs_are_visible,
 801 };
 802
 803 static const struct attribute_group *virtblk_attr_groups[] = {
 804         &virtblk_attr_group,
 805         NULL,
 806 };
 807
 808 static void virtblk_map_queues(struct blk_mq_tag_set *set)
 809 {
 810         struct virtio_blk *vblk = set->driver_data;
 811         int i, qoff;
 812
 813         for (i = 0, qoff = 0; i < set->nr_maps; i++) {
 814                 struct blk_mq_queue_map *map = &set->map[i];
 815
 816                 map->nr_queues = vblk->io_queues[i];
 817                 map->queue_offset = qoff;
 818                 qoff += map->nr_queues;
 819
 820                 if (map->nr_queues == 0)
 821                         continue;
 822
 823                 /*
 824                  * Regular queues have interrupts and hence CPU affinity is
 825                  * defined by the core virtio code, but polling queues have
 826                  * no interrupts so we let the block layer assign CPU affinity.
 827                  */
 828                 if (i == HCTX_TYPE_POLL)
 829                         blk_mq_map_queues(&set->map[i]);
 830                 else
 831                         blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
 832         }
 833 }
 834
 835 static void virtblk_complete_batch(struct io_comp_batch *iob)
 836 {
 837         struct request *req;
 838
 839         rq_list_for_each(&iob->req_list, req) {
 840                 virtblk_unmap_data(req, blk_mq_rq_to_pdu(req));
 841                 virtblk_cleanup_cmd(req);
 842         }
 843         blk_mq_end_request_batch(iob);
 844 }
 845
 846 static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
 847 {
 848         struct virtio_blk *vblk = hctx->queue->queuedata;
 849         struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
 850         struct virtblk_req *vbr;
 851         unsigned long flags;
 852         unsigned int len;
 853         int found = 0;
 854
 855         spin_lock_irqsave(&vq->lock, flags);
 856
 857         while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
 858                 struct request *req = blk_mq_rq_from_pdu(vbr);
 859
 860                 found++;
 861                 if (!blk_mq_add_to_batch(req, iob, vbr->status,
 862                                                 virtblk_complete_batch))
 863                         blk_mq_complete_request(req);
 864         }
 865
 866         if (found)
 867                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 868
 869         spin_unlock_irqrestore(&vq->lock, flags);
 870
 871         return found;
 872 }
 873
 874 static const struct blk_mq_ops virtio_mq_ops = {
 875         .queue_rq       = virtio_queue_rq,
 876         .queue_rqs      = virtio_queue_rqs,
 877         .commit_rqs     = virtio_commit_rqs,
 878         .complete       = virtblk_request_done,
 879         .map_queues     = virtblk_map_queues,
 880         .poll           = virtblk_poll,
 881 };
 882
 883 static unsigned int virtblk_queue_depth;
 884 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 885
 886 static int virtblk_probe(struct virtio_device *vdev)
 887 {
 888         struct virtio_blk *vblk;
 889         struct request_queue *q;
 890         int err, index;
 891
 892         u32 v, blk_size, max_size, sg_elems, opt_io_size;
 893         u32 max_discard_segs = 0;
 894         u32 discard_granularity = 0;
 895         u16 min_io_size;
 896         u8 physical_block_exp, alignment_offset;
 897         unsigned int queue_depth;
 898
 899         if (!vdev->config->get) {
 900                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
 901                         __func__);
 902                 return -EINVAL;
 903         }
 904
 905         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 906                              GFP_KERNEL);
 907         if (err < 0)
 908                 goto out;
 909         index = err;
 910
 911         /* We need to know how many segments before we allocate. */
 912         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
 913                                    struct virtio_blk_config, seg_max,
 914                                    &sg_elems);
 915
 916         /* We need at least one SG element, whatever they say. */
 917         if (err || !sg_elems)
 918                 sg_elems = 1;
 919
 920         /* Prevent integer overflows and honor max vq size */
 921         sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
 922
 923         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 924         if (!vblk) {
 925                 err = -ENOMEM;
 926                 goto out_free_index;
 927         }
 928
 929         mutex_init(&vblk->vdev_mutex);
 930
 931         vblk->vdev = vdev;
 932
 933         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 934
 935         err = init_vq(vblk);
 936         if (err)
 937                 goto out_free_vblk;
 938
 939         /* Default queue sizing is to fill the ring. */
 940         if (!virtblk_queue_depth) {
 941                 queue_depth = vblk->vqs[0].vq->num_free;
 942                 /* ... but without indirect descs, we use 2 descs per req */
 943                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
 944                         queue_depth /= 2;
 945         } else {
 946                 queue_depth = virtblk_queue_depth;
 947         }
 948
 949         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
 950         vblk->tag_set.ops = &virtio_mq_ops;
 951         vblk->tag_set.queue_depth = queue_depth;
 952         vblk->tag_set.numa_node = NUMA_NO_NODE;
 953         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 954         vblk->tag_set.cmd_size =
 955                 sizeof(struct virtblk_req) +
 956                 sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
 957         vblk->tag_set.driver_data = vblk;
 958         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
 959         vblk->tag_set.nr_maps = 1;
 960         if (vblk->io_queues[HCTX_TYPE_POLL])
 961                 vblk->tag_set.nr_maps = 3;
 962
 963         err = blk_mq_alloc_tag_set(&vblk->tag_set);
 964         if (err)
 965                 goto out_free_vq;
 966
 967         vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
 968         if (IS_ERR(vblk->disk)) {
 969                 err = PTR_ERR(vblk->disk);
 970                 goto out_free_tags;
 971         }
 972         q = vblk->disk->queue;
 973
 974         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 975
 976         vblk->disk->major = major;
 977         vblk->disk->first_minor = index_to_minor(index);
 978         vblk->disk->minors = 1 << PART_BITS;
 979         vblk->disk->private_data = vblk;
 980         vblk->disk->fops = &virtblk_fops;
 981         vblk->index = index;
 982
 983         /* configure queue flush support */
 984         virtblk_update_cache_mode(vdev);
 985
 986         /* If disk is read-only in the host, the guest should obey */
 987         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 988                 set_disk_ro(vblk->disk, 1);
 989
 990         /* We can handle whatever the host told us to handle. */
 991         blk_queue_max_segments(q, sg_elems);
 992
 993         /* No real sector limit. */
 994         blk_queue_max_hw_sectors(q, -1U);
 995
 996         max_size = virtio_max_dma_size(vdev);
 997
 998         /* Host can optionally specify maximum segment size and number of
 999          * segments. */
1000         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1001                                    struct virtio_blk_config, size_max, &v);
1002         if (!err)
1003                 max_size = min(max_size, v);
1004
1005         blk_queue_max_segment_size(q, max_size);
1006
1007         /* Host can optionally specify the block size of the device */
1008         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1009                                    struct virtio_blk_config, blk_size,
1010                                    &blk_size);
1011         if (!err) {
1012                 err = blk_validate_block_size(blk_size);
1013                 if (err) {
1014                         dev_err(&vdev->dev,
1015                                 "virtio_blk: invalid block size: 0x%x\n",
1016                                 blk_size);
1017                         goto out_cleanup_disk;
1018                 }
1019
1020                 blk_queue_logical_block_size(q, blk_size);
1021         } else
1022                 blk_size = queue_logical_block_size(q);
1023
1024         /* Use topology information if available */
1025         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1026                                    struct virtio_blk_config, physical_block_exp,
1027                                    &physical_block_exp);
1028         if (!err && physical_block_exp)
1029                 blk_queue_physical_block_size(q,
1030                                 blk_size * (1 << physical_block_exp));
1031
1032         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1033                                    struct virtio_blk_config, alignment_offset,
1034                                    &alignment_offset);
1035         if (!err && alignment_offset)
1036                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
1037
1038         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1039                                    struct virtio_blk_config, min_io_size,
1040                                    &min_io_size);
1041         if (!err && min_io_size)
1042                 blk_queue_io_min(q, blk_size * min_io_size);
1043
1044         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1045                                    struct virtio_blk_config, opt_io_size,
1046                                    &opt_io_size);
1047         if (!err && opt_io_size)
1048                 blk_queue_io_opt(q, blk_size * opt_io_size);
1049
1050         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1051                 virtio_cread(vdev, struct virtio_blk_config,
1052                              discard_sector_alignment, &discard_granularity);
1053
1054                 virtio_cread(vdev, struct virtio_blk_config,
1055                              max_discard_sectors, &v);
1056                 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
1057
1058                 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1059                              &max_discard_segs);
1060         }
1061
1062         if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1063                 virtio_cread(vdev, struct virtio_blk_config,
1064                              max_write_zeroes_sectors, &v);
1065                 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
1066         }
1067
1068         /* The discard and secure erase limits are combined since the Linux
1069          * block layer uses the same limit for both commands.
1070          *
1071          * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1072          * are negotiated, we will use the minimum between the limits.
1073          *
1074          * discard sector alignment is set to the minimum between discard_sector_alignment
1075          * and secure_erase_sector_alignment.
1076          *
1077          * max discard sectors is set to the minimum between max_discard_seg and
1078          * max_secure_erase_seg.
1079          */
1080         if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1081
1082                 virtio_cread(vdev, struct virtio_blk_config,
1083                              secure_erase_sector_alignment, &v);
1084
1085                 /* secure_erase_sector_alignment should not be zero, the device should set a
1086                  * valid number of sectors.
1087                  */
1088                 if (!v) {
1089                         dev_err(&vdev->dev,
1090                                 "virtio_blk: secure_erase_sector_alignment can't be 0\n");
1091                         err = -EINVAL;
1092                         goto out_cleanup_disk;
1093                 }
1094
1095                 discard_granularity = min_not_zero(discard_granularity, v);
1096
1097                 virtio_cread(vdev, struct virtio_blk_config,
1098                              max_secure_erase_sectors, &v);
1099
1100                 /* max_secure_erase_sectors should not be zero, the device should set a
1101                  * valid number of sectors.
1102                  */
1103                 if (!v) {
1104                         dev_err(&vdev->dev,
1105                                 "virtio_blk: max_secure_erase_sectors can't be 0\n");
1106                         err = -EINVAL;
1107                         goto out_cleanup_disk;
1108                 }
1109
1110                 blk_queue_max_secure_erase_sectors(q, v);
1111
1112                 virtio_cread(vdev, struct virtio_blk_config,
1113                              max_secure_erase_seg, &v);
1114
1115                 /* max_secure_erase_seg should not be zero, the device should set a
1116                  * valid number of segments
1117                  */
1118                 if (!v) {
1119                         dev_err(&vdev->dev,
1120                                 "virtio_blk: max_secure_erase_seg can't be 0\n");
1121                         err = -EINVAL;
1122                         goto out_cleanup_disk;
1123                 }
1124
1125                 max_discard_segs = min_not_zero(max_discard_segs, v);
1126         }
1127
1128         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1129             virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1130                 /* max_discard_seg and discard_granularity will be 0 only
1131                  * if max_discard_seg and discard_sector_alignment fields in the virtio
1132                  * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1133                  * In this case, we use default values.
1134                  */
1135                 if (!max_discard_segs)
1136                         max_discard_segs = sg_elems;
1137
1138                 blk_queue_max_discard_segments(q,
1139                                                min(max_discard_segs, MAX_DISCARD_SEGMENTS));
1140
1141                 if (discard_granularity)
1142                         q->limits.discard_granularity = discard_granularity << SECTOR_SHIFT;
1143                 else
1144                         q->limits.discard_granularity = blk_size;
1145         }
1146
1147         virtblk_update_capacity(vblk, false);
1148         virtio_device_ready(vdev);
1149
1150         err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
1151         if (err)
1152                 goto out_cleanup_disk;
1153
1154         return 0;
1155
1156 out_cleanup_disk:
1157         put_disk(vblk->disk);
1158 out_free_tags:
1159         blk_mq_free_tag_set(&vblk->tag_set);
1160 out_free_vq:
1161         vdev->config->del_vqs(vdev);
1162         kfree(vblk->vqs);
1163 out_free_vblk:
1164         kfree(vblk);
1165 out_free_index:
1166         ida_simple_remove(&vd_index_ida, index);
1167 out:
1168         return err;
1169 }
1170
1171 static void virtblk_remove(struct virtio_device *vdev)
1172 {
1173         struct virtio_blk *vblk = vdev->priv;
1174
1175         /* Make sure no work handler is accessing the device. */
1176         flush_work(&vblk->config_work);
1177
1178         del_gendisk(vblk->disk);
1179         blk_mq_free_tag_set(&vblk->tag_set);
1180
1181         mutex_lock(&vblk->vdev_mutex);
1182
1183         /* Stop all the virtqueues. */
1184         virtio_reset_device(vdev);
1185
1186         /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1187         vblk->vdev = NULL;
1188
1189         vdev->config->del_vqs(vdev);
1190         kfree(vblk->vqs);
1191
1192         mutex_unlock(&vblk->vdev_mutex);
1193
1194         put_disk(vblk->disk);
1195 }
1196
1197 #ifdef CONFIG_PM_SLEEP
1198 static int virtblk_freeze(struct virtio_device *vdev)
1199 {
1200         struct virtio_blk *vblk = vdev->priv;
1201
1202         /* Ensure we don't receive any more interrupts */
1203         virtio_reset_device(vdev);
1204
1205         /* Make sure no work handler is accessing the device. */
1206         flush_work(&vblk->config_work);
1207
1208         blk_mq_quiesce_queue(vblk->disk->queue);
1209
1210         vdev->config->del_vqs(vdev);
1211         kfree(vblk->vqs);
1212
1213         return 0;
1214 }
1215
1216 static int virtblk_restore(struct virtio_device *vdev)
1217 {
1218         struct virtio_blk *vblk = vdev->priv;
1219         int ret;
1220
1221         ret = init_vq(vdev->priv);
1222         if (ret)
1223                 return ret;
1224
1225         virtio_device_ready(vdev);
1226
1227         blk_mq_unquiesce_queue(vblk->disk->queue);
1228         return 0;
1229 }
1230 #endif
1231
1232 static const struct virtio_device_id id_table[] = {
1233         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1234         { 0 },
1235 };
1236
1237 static unsigned int features_legacy[] = {
1238         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1239         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1240         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1241         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1242         VIRTIO_BLK_F_SECURE_ERASE,
1243 }
1244 ;
1245 static unsigned int features[] = {
1246         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1247         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1248         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1249         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1250         VIRTIO_BLK_F_SECURE_ERASE,
1251 };
1252
1253 static struct virtio_driver virtio_blk = {
1254         .feature_table                  = features,
1255         .feature_table_size             = ARRAY_SIZE(features),
1256         .feature_table_legacy           = features_legacy,
1257         .feature_table_size_legacy      = ARRAY_SIZE(features_legacy),
1258         .driver.name                    = KBUILD_MODNAME,
1259         .driver.owner                   = THIS_MODULE,
1260         .id_table                       = id_table,
1261         .probe                          = virtblk_probe,
1262         .remove                         = virtblk_remove,
1263         .config_changed                 = virtblk_config_changed,
1264 #ifdef CONFIG_PM_SLEEP
1265         .freeze                         = virtblk_freeze,
1266         .restore                        = virtblk_restore,
1267 #endif
1268 };
1269
1270 static int __init virtio_blk_init(void)
1271 {
1272         int error;
1273
1274         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1275         if (!virtblk_wq)
1276                 return -ENOMEM;
1277
1278         major = register_blkdev(0, "virtblk");
1279         if (major < 0) {
1280                 error = major;
1281                 goto out_destroy_workqueue;
1282         }
1283
1284         error = register_virtio_driver(&virtio_blk);
1285         if (error)
1286                 goto out_unregister_blkdev;
1287         return 0;
1288
1289 out_unregister_blkdev:
1290         unregister_blkdev(major, "virtblk");
1291 out_destroy_workqueue:
1292         destroy_workqueue(virtblk_wq);
1293         return error;
1294 }
1295
1296 static void __exit virtio_blk_fini(void)
1297 {
1298         unregister_virtio_driver(&virtio_blk);
1299         unregister_blkdev(major, "virtblk");
1300         destroy_workqueue(virtblk_wq);
1301 }
1302 module_init(virtio_blk_init);
1303 module_exit(virtio_blk_fini);
1304
1305 MODULE_DEVICE_TABLE(virtio, id_table);
1306 MODULE_DESCRIPTION("Virtio block driver");
1307 MODULE_LICENSE("GPL");