2 #include <linux/spinlock.h>
3 #include <linux/slab.h>
4 #include <linux/blkdev.h>
5 #include <linux/hdreg.h>
6 #include <linux/module.h>
7 #include <linux/mutex.h>
8 #include <linux/interrupt.h>
9 #include <linux/virtio.h>
10 #include <linux/virtio_blk.h>
11 #include <linux/scatterlist.h>
12 #include <linux/string_helpers.h>
13 #include <scsi/scsi_cmnd.h>
14 #include <linux/idr.h>
15 #include <linux/blk-mq.h>
16 #include <linux/blk-mq-virtio.h>
17 #include <linux/numa.h>
20 #define VQ_NAME_LEN 16
23 static DEFINE_IDA(vd_index_ida);
25 static struct workqueue_struct *virtblk_wq;
27 struct virtio_blk_vq {
30 char name[VQ_NAME_LEN];
31 } ____cacheline_aligned_in_smp;
35 * This mutex must be held by anything that may run after
36 * virtblk_remove() sets vblk->vdev to NULL.
38 * blk-mq, virtqueue processing, and sysfs attribute code paths are
39 * shut down before vblk->vdev is set to NULL and therefore do not need
42 struct mutex vdev_mutex;
43 struct virtio_device *vdev;
45 /* The disk structure for the kernel. */
48 /* Block layer tags. */
49 struct blk_mq_tag_set tag_set;
51 /* Process context for config space updates */
52 struct work_struct config_work;
55 * Tracks references from block_device_operations open/release and
56 * virtio_driver probe/remove so this object can be freed once no
61 /* What host tells us, plus 2 for header & tailer. */
62 unsigned int sg_elems;
64 /* Ida index - used to track minor number allocations. */
69 struct virtio_blk_vq *vqs;
73 #ifdef CONFIG_VIRTIO_BLK_SCSI
74 struct scsi_request sreq; /* for SCSI passthrough, must be first */
75 u8 sense[SCSI_SENSE_BUFFERSIZE];
76 struct virtio_scsi_inhdr in_hdr;
78 struct virtio_blk_outhdr out_hdr;
80 struct scatterlist sg[];
83 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
85 switch (vbr->status) {
88 case VIRTIO_BLK_S_UNSUPP:
89 return BLK_STS_NOTSUPP;
96 * If this is a packet command we need a couple of additional headers. Behind
97 * the normal outhdr we put a segment with the scsi command block, and before
98 * the normal inhdr we put the sense data and the inhdr with additional status
101 #ifdef CONFIG_VIRTIO_BLK_SCSI
102 static int virtblk_add_req_scsi(struct virtqueue *vq, struct virtblk_req *vbr,
103 struct scatterlist *data_sg, bool have_data)
105 struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
106 unsigned int num_out = 0, num_in = 0;
108 sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
109 sgs[num_out++] = &hdr;
110 sg_init_one(&cmd, vbr->sreq.cmd, vbr->sreq.cmd_len);
111 sgs[num_out++] = &cmd;
114 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
115 sgs[num_out++] = data_sg;
117 sgs[num_out + num_in++] = data_sg;
120 sg_init_one(&sense, vbr->sense, SCSI_SENSE_BUFFERSIZE);
121 sgs[num_out + num_in++] = &sense;
122 sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
123 sgs[num_out + num_in++] = &inhdr;
124 sg_init_one(&status, &vbr->status, sizeof(vbr->status));
125 sgs[num_out + num_in++] = &status;
127 return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
130 static inline void virtblk_scsi_request_done(struct request *req)
132 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
133 struct virtio_blk *vblk = req->q->queuedata;
134 struct scsi_request *sreq = &vbr->sreq;
136 sreq->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
137 sreq->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
138 sreq->result = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
141 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
142 unsigned int cmd, unsigned long data)
144 struct gendisk *disk = bdev->bd_disk;
145 struct virtio_blk *vblk = disk->private_data;
148 * Only allow the generic SCSI ioctls if the host can support it.
150 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
153 return scsi_cmd_blk_ioctl(bdev, mode, cmd,
154 (void __user *)data);
157 static inline int virtblk_add_req_scsi(struct virtqueue *vq,
158 struct virtblk_req *vbr, struct scatterlist *data_sg,
163 static inline void virtblk_scsi_request_done(struct request *req)
166 #define virtblk_ioctl NULL
167 #endif /* CONFIG_VIRTIO_BLK_SCSI */
169 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
170 struct scatterlist *data_sg, bool have_data)
172 struct scatterlist hdr, status, *sgs[3];
173 unsigned int num_out = 0, num_in = 0;
175 sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
176 sgs[num_out++] = &hdr;
179 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
180 sgs[num_out++] = data_sg;
182 sgs[num_out + num_in++] = data_sg;
185 sg_init_one(&status, &vbr->status, sizeof(vbr->status));
186 sgs[num_out + num_in++] = &status;
188 return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
191 static inline void virtblk_request_done(struct request *req)
193 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
195 switch (req_op(req)) {
197 case REQ_OP_SCSI_OUT:
198 virtblk_scsi_request_done(req);
202 blk_mq_end_request(req, virtblk_result(vbr));
205 static void virtblk_done(struct virtqueue *vq)
207 struct virtio_blk *vblk = vq->vdev->priv;
208 bool req_done = false;
210 struct virtblk_req *vbr;
214 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
216 virtqueue_disable_cb(vq);
217 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
218 struct request *req = blk_mq_rq_from_pdu(vbr);
220 blk_mq_complete_request(req);
223 if (unlikely(virtqueue_is_broken(vq)))
225 } while (!virtqueue_enable_cb(vq));
227 /* In case queue is stopped waiting for more buffers. */
229 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
230 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
233 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
234 const struct blk_mq_queue_data *bd)
236 struct virtio_blk *vblk = hctx->queue->queuedata;
237 struct request *req = bd->rq;
238 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
241 int qid = hctx->queue_num;
246 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
248 switch (req_op(req)) {
254 type = VIRTIO_BLK_T_FLUSH;
257 case REQ_OP_SCSI_OUT:
258 type = VIRTIO_BLK_T_SCSI_CMD;
261 type = VIRTIO_BLK_T_GET_ID;
265 return BLK_STS_IOERR;
268 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
269 vbr->out_hdr.sector = type ?
270 0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
271 vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
273 blk_mq_start_request(req);
275 num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
277 if (rq_data_dir(req) == WRITE)
278 vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
280 vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
283 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
284 if (blk_rq_is_scsi(req))
285 err = virtblk_add_req_scsi(vblk->vqs[qid].vq, vbr, vbr->sg, num);
287 err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
289 virtqueue_kick(vblk->vqs[qid].vq);
290 /* Don't stop the queue if -ENOMEM: we may have failed to
291 * bounce the buffer due to global resource outage.
294 blk_mq_stop_hw_queue(hctx);
295 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
296 if (err == -ENOMEM || err == -ENOSPC)
297 return BLK_STS_RESOURCE;
298 return BLK_STS_IOERR;
301 if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
303 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
306 virtqueue_notify(vblk->vqs[qid].vq);
310 /* return id (s/n) string for *disk to *id_str
312 static int virtblk_get_id(struct gendisk *disk, char *id_str)
314 struct virtio_blk *vblk = disk->private_data;
315 struct request_queue *q = vblk->disk->queue;
319 req = blk_get_request(q, REQ_OP_DRV_IN, GFP_KERNEL);
323 err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
327 blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
328 err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
330 blk_put_request(req);
334 static void virtblk_get(struct virtio_blk *vblk)
336 refcount_inc(&vblk->refs);
339 static void virtblk_put(struct virtio_blk *vblk)
341 if (refcount_dec_and_test(&vblk->refs)) {
342 ida_simple_remove(&vd_index_ida, vblk->index);
343 mutex_destroy(&vblk->vdev_mutex);
348 static int virtblk_open(struct block_device *bd, fmode_t mode)
350 struct virtio_blk *vblk = bd->bd_disk->private_data;
353 mutex_lock(&vblk->vdev_mutex);
360 mutex_unlock(&vblk->vdev_mutex);
364 static void virtblk_release(struct gendisk *disk, fmode_t mode)
366 struct virtio_blk *vblk = disk->private_data;
371 /* We provide getgeo only to please some old bootloader/partitioning tools */
372 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
374 struct virtio_blk *vblk = bd->bd_disk->private_data;
377 mutex_lock(&vblk->vdev_mutex);
384 /* see if the host passed in geometry config */
385 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
386 virtio_cread(vblk->vdev, struct virtio_blk_config,
387 geometry.cylinders, &geo->cylinders);
388 virtio_cread(vblk->vdev, struct virtio_blk_config,
389 geometry.heads, &geo->heads);
390 virtio_cread(vblk->vdev, struct virtio_blk_config,
391 geometry.sectors, &geo->sectors);
393 /* some standard values, similar to sd */
395 geo->sectors = 1 << 5;
396 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
399 mutex_unlock(&vblk->vdev_mutex);
403 static const struct block_device_operations virtblk_fops = {
404 .ioctl = virtblk_ioctl,
405 .owner = THIS_MODULE,
406 .open = virtblk_open,
407 .release = virtblk_release,
408 .getgeo = virtblk_getgeo,
411 static int index_to_minor(int index)
413 return index << PART_BITS;
416 static int minor_to_index(int minor)
418 return minor >> PART_BITS;
421 static ssize_t virtblk_serial_show(struct device *dev,
422 struct device_attribute *attr, char *buf)
424 struct gendisk *disk = dev_to_disk(dev);
427 /* sysfs gives us a PAGE_SIZE buffer */
428 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
430 buf[VIRTIO_BLK_ID_BYTES] = '\0';
431 err = virtblk_get_id(disk, buf);
435 if (err == -EIO) /* Unsupported? Make it empty. */
441 static DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
443 static void virtblk_config_changed_work(struct work_struct *work)
445 struct virtio_blk *vblk =
446 container_of(work, struct virtio_blk, config_work);
447 struct virtio_device *vdev = vblk->vdev;
448 struct request_queue *q = vblk->disk->queue;
449 char cap_str_2[10], cap_str_10[10];
450 char *envp[] = { "RESIZE=1", NULL };
451 unsigned long long nblocks;
454 /* Host must always specify the capacity. */
455 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
457 /* If capacity is too big, truncate with warning. */
458 if ((sector_t)capacity != capacity) {
459 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
460 (unsigned long long)capacity);
461 capacity = (sector_t)-1;
464 nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
466 string_get_size(nblocks, queue_logical_block_size(q),
467 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
468 string_get_size(nblocks, queue_logical_block_size(q),
469 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
471 dev_notice(&vdev->dev,
472 "new size: %llu %d-byte logical blocks (%s/%s)\n",
474 queue_logical_block_size(q),
478 set_capacity(vblk->disk, capacity);
479 revalidate_disk(vblk->disk);
480 kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
483 static void virtblk_config_changed(struct virtio_device *vdev)
485 struct virtio_blk *vblk = vdev->priv;
487 queue_work(virtblk_wq, &vblk->config_work);
490 static int init_vq(struct virtio_blk *vblk)
494 vq_callback_t **callbacks;
496 struct virtqueue **vqs;
497 unsigned short num_vqs;
498 struct virtio_device *vdev = vblk->vdev;
499 struct irq_affinity desc = { 0, };
501 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
502 struct virtio_blk_config, num_queues,
507 num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
509 vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
513 names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
514 callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
515 vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
516 if (!names || !callbacks || !vqs) {
521 for (i = 0; i < num_vqs; i++) {
522 callbacks[i] = virtblk_done;
523 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
524 names[i] = vblk->vqs[i].name;
527 /* Discover virtqueues and write information to configuration. */
528 err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
532 for (i = 0; i < num_vqs; i++) {
533 spin_lock_init(&vblk->vqs[i].lock);
534 vblk->vqs[i].vq = vqs[i];
536 vblk->num_vqs = num_vqs;
548 * Legacy naming scheme used for virtio devices. We are stuck with it for
549 * virtio blk but don't ever use it for any new driver.
551 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
553 const int base = 'z' - 'a' + 1;
554 char *begin = buf + strlen(prefix);
555 char *end = buf + buflen;
565 *--p = 'a' + (index % unit);
566 index = (index / unit) - 1;
567 } while (index >= 0);
569 memmove(begin, p, end - p);
570 memcpy(buf, prefix, strlen(prefix));
575 static int virtblk_get_cache_mode(struct virtio_device *vdev)
580 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
581 struct virtio_blk_config, wce,
585 * If WCE is not configurable and flush is not available,
586 * assume no writeback cache is in use.
589 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
594 static void virtblk_update_cache_mode(struct virtio_device *vdev)
596 u8 writeback = virtblk_get_cache_mode(vdev);
597 struct virtio_blk *vblk = vdev->priv;
599 blk_queue_write_cache(vblk->disk->queue, writeback, false);
600 revalidate_disk(vblk->disk);
603 static const char *const virtblk_cache_types[] = {
604 "write through", "write back"
608 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
609 const char *buf, size_t count)
611 struct gendisk *disk = dev_to_disk(dev);
612 struct virtio_blk *vblk = disk->private_data;
613 struct virtio_device *vdev = vblk->vdev;
616 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
617 i = sysfs_match_string(virtblk_cache_types, buf);
621 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
622 virtblk_update_cache_mode(vdev);
627 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
630 struct gendisk *disk = dev_to_disk(dev);
631 struct virtio_blk *vblk = disk->private_data;
632 u8 writeback = virtblk_get_cache_mode(vblk->vdev);
634 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
635 return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
638 static const struct device_attribute dev_attr_cache_type_ro =
639 __ATTR(cache_type, S_IRUGO,
640 virtblk_cache_type_show, NULL);
641 static const struct device_attribute dev_attr_cache_type_rw =
642 __ATTR(cache_type, S_IRUGO|S_IWUSR,
643 virtblk_cache_type_show, virtblk_cache_type_store);
645 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
646 unsigned int hctx_idx, unsigned int numa_node)
648 struct virtio_blk *vblk = set->driver_data;
649 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
651 #ifdef CONFIG_VIRTIO_BLK_SCSI
652 vbr->sreq.sense = vbr->sense;
654 sg_init_table(vbr->sg, vblk->sg_elems);
658 static int virtblk_map_queues(struct blk_mq_tag_set *set)
660 struct virtio_blk *vblk = set->driver_data;
662 return blk_mq_virtio_map_queues(set, vblk->vdev, 0);
665 #ifdef CONFIG_VIRTIO_BLK_SCSI
666 static void virtblk_initialize_rq(struct request *req)
668 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
670 scsi_req_init(&vbr->sreq);
674 static const struct blk_mq_ops virtio_mq_ops = {
675 .queue_rq = virtio_queue_rq,
676 .complete = virtblk_request_done,
677 .init_request = virtblk_init_request,
678 #ifdef CONFIG_VIRTIO_BLK_SCSI
679 .initialize_rq_fn = virtblk_initialize_rq,
681 .map_queues = virtblk_map_queues,
684 static unsigned int virtblk_queue_depth;
685 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
687 static int virtblk_probe(struct virtio_device *vdev)
689 struct virtio_blk *vblk;
690 struct request_queue *q;
694 u32 v, blk_size, sg_elems, opt_io_size;
696 u8 physical_block_exp, alignment_offset;
698 if (!vdev->config->get) {
699 dev_err(&vdev->dev, "%s failure: config access disabled\n",
704 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
710 /* We need to know how many segments before we allocate. */
711 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
712 struct virtio_blk_config, seg_max,
715 /* We need at least one SG element, whatever they say. */
716 if (err || !sg_elems)
719 /* We need an extra sg elements at head and tail. */
721 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
727 /* This reference is dropped in virtblk_remove(). */
728 refcount_set(&vblk->refs, 1);
729 mutex_init(&vblk->vdev_mutex);
732 vblk->sg_elems = sg_elems;
734 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
740 /* FIXME: How many partitions? How long is a piece of string? */
741 vblk->disk = alloc_disk(1 << PART_BITS);
747 /* Default queue sizing is to fill the ring. */
748 if (!virtblk_queue_depth) {
749 virtblk_queue_depth = vblk->vqs[0].vq->num_free;
750 /* ... but without indirect descs, we use 2 descs per req */
751 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
752 virtblk_queue_depth /= 2;
755 memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
756 vblk->tag_set.ops = &virtio_mq_ops;
757 vblk->tag_set.queue_depth = virtblk_queue_depth;
758 vblk->tag_set.numa_node = NUMA_NO_NODE;
759 vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
760 vblk->tag_set.cmd_size =
761 sizeof(struct virtblk_req) +
762 sizeof(struct scatterlist) * sg_elems;
763 vblk->tag_set.driver_data = vblk;
764 vblk->tag_set.nr_hw_queues = vblk->num_vqs;
766 err = blk_mq_alloc_tag_set(&vblk->tag_set);
770 q = blk_mq_init_queue(&vblk->tag_set);
775 vblk->disk->queue = q;
779 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
781 vblk->disk->major = major;
782 vblk->disk->first_minor = index_to_minor(index);
783 vblk->disk->private_data = vblk;
784 vblk->disk->fops = &virtblk_fops;
785 vblk->disk->flags |= GENHD_FL_EXT_DEVT;
788 /* configure queue flush support */
789 virtblk_update_cache_mode(vdev);
791 /* If disk is read-only in the host, the guest should obey */
792 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
793 set_disk_ro(vblk->disk, 1);
795 /* Host must always specify the capacity. */
796 virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
798 /* If capacity is too big, truncate with warning. */
799 if ((sector_t)cap != cap) {
800 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
801 (unsigned long long)cap);
804 set_capacity(vblk->disk, cap);
806 /* We can handle whatever the host told us to handle. */
807 blk_queue_max_segments(q, vblk->sg_elems-2);
809 /* No real sector limit. */
810 blk_queue_max_hw_sectors(q, -1U);
812 /* Host can optionally specify maximum segment size and number of
814 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
815 struct virtio_blk_config, size_max, &v);
817 blk_queue_max_segment_size(q, v);
819 blk_queue_max_segment_size(q, -1U);
821 /* Host can optionally specify the block size of the device */
822 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
823 struct virtio_blk_config, blk_size,
826 err = blk_validate_block_size(blk_size);
829 "virtio_blk: invalid block size: 0x%x\n",
834 blk_queue_logical_block_size(q, blk_size);
836 blk_size = queue_logical_block_size(q);
838 /* Use topology information if available */
839 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
840 struct virtio_blk_config, physical_block_exp,
841 &physical_block_exp);
842 if (!err && physical_block_exp)
843 blk_queue_physical_block_size(q,
844 blk_size * (1 << physical_block_exp));
846 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
847 struct virtio_blk_config, alignment_offset,
849 if (!err && alignment_offset)
850 blk_queue_alignment_offset(q, blk_size * alignment_offset);
852 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
853 struct virtio_blk_config, min_io_size,
855 if (!err && min_io_size)
856 blk_queue_io_min(q, blk_size * min_io_size);
858 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
859 struct virtio_blk_config, opt_io_size,
861 if (!err && opt_io_size)
862 blk_queue_io_opt(q, blk_size * opt_io_size);
864 virtio_device_ready(vdev);
866 device_add_disk(&vdev->dev, vblk->disk);
867 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
871 if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
872 err = device_create_file(disk_to_dev(vblk->disk),
873 &dev_attr_cache_type_rw);
875 err = device_create_file(disk_to_dev(vblk->disk),
876 &dev_attr_cache_type_ro);
882 del_gendisk(vblk->disk);
883 blk_cleanup_queue(vblk->disk->queue);
885 blk_mq_free_tag_set(&vblk->tag_set);
887 put_disk(vblk->disk);
889 vdev->config->del_vqs(vdev);
894 ida_simple_remove(&vd_index_ida, index);
899 static void virtblk_remove(struct virtio_device *vdev)
901 struct virtio_blk *vblk = vdev->priv;
903 /* Make sure no work handler is accessing the device. */
904 flush_work(&vblk->config_work);
906 del_gendisk(vblk->disk);
907 blk_cleanup_queue(vblk->disk->queue);
909 blk_mq_free_tag_set(&vblk->tag_set);
911 mutex_lock(&vblk->vdev_mutex);
913 /* Stop all the virtqueues. */
914 vdev->config->reset(vdev);
916 /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
919 put_disk(vblk->disk);
920 vdev->config->del_vqs(vdev);
923 mutex_unlock(&vblk->vdev_mutex);
928 #ifdef CONFIG_PM_SLEEP
929 static int virtblk_freeze(struct virtio_device *vdev)
931 struct virtio_blk *vblk = vdev->priv;
933 /* Ensure we don't receive any more interrupts */
934 vdev->config->reset(vdev);
936 /* Make sure no work handler is accessing the device. */
937 flush_work(&vblk->config_work);
939 blk_mq_quiesce_queue(vblk->disk->queue);
941 vdev->config->del_vqs(vdev);
947 static int virtblk_restore(struct virtio_device *vdev)
949 struct virtio_blk *vblk = vdev->priv;
952 ret = init_vq(vdev->priv);
956 virtio_device_ready(vdev);
958 blk_mq_unquiesce_queue(vblk->disk->queue);
963 static const struct virtio_device_id id_table[] = {
964 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
968 static unsigned int features_legacy[] = {
969 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
970 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
971 #ifdef CONFIG_VIRTIO_BLK_SCSI
974 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
978 static unsigned int features[] = {
979 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
980 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
981 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
985 static struct virtio_driver virtio_blk = {
986 .feature_table = features,
987 .feature_table_size = ARRAY_SIZE(features),
988 .feature_table_legacy = features_legacy,
989 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
990 .driver.name = KBUILD_MODNAME,
991 .driver.owner = THIS_MODULE,
992 .id_table = id_table,
993 .probe = virtblk_probe,
994 .remove = virtblk_remove,
995 .config_changed = virtblk_config_changed,
996 #ifdef CONFIG_PM_SLEEP
997 .freeze = virtblk_freeze,
998 .restore = virtblk_restore,
1002 static int __init init(void)
1006 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1010 major = register_blkdev(0, "virtblk");
1013 goto out_destroy_workqueue;
1016 error = register_virtio_driver(&virtio_blk);
1018 goto out_unregister_blkdev;
1021 out_unregister_blkdev:
1022 unregister_blkdev(major, "virtblk");
1023 out_destroy_workqueue:
1024 destroy_workqueue(virtblk_wq);
1028 static void __exit fini(void)
1030 unregister_virtio_driver(&virtio_blk);
1031 unregister_blkdev(major, "virtblk");
1032 destroy_workqueue(virtblk_wq);
1037 MODULE_DEVICE_TABLE(virtio, id_table);
1038 MODULE_DESCRIPTION("Virtio block driver");
1039 MODULE_LICENSE("GPL");