1 // SPDX-License-Identifier: GPL-2.0-only
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>
21 #define VQ_NAME_LEN 16
22 #define MAX_DISCARD_SEGMENTS 256u
24 /* The maximum number of sg elements that fit into a virtqueue */
25 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
27 #ifdef CONFIG_ARCH_NO_SG_CHAIN
28 #define VIRTIO_BLK_INLINE_SG_CNT 0
30 #define VIRTIO_BLK_INLINE_SG_CNT 2
34 static DEFINE_IDA(vd_index_ida);
36 static struct workqueue_struct *virtblk_wq;
38 struct virtio_blk_vq {
41 char name[VQ_NAME_LEN];
42 } ____cacheline_aligned_in_smp;
46 * This mutex must be held by anything that may run after
47 * virtblk_remove() sets vblk->vdev to NULL.
49 * blk-mq, virtqueue processing, and sysfs attribute code paths are
50 * shut down before vblk->vdev is set to NULL and therefore do not need
53 struct mutex vdev_mutex;
54 struct virtio_device *vdev;
56 /* The disk structure for the kernel. */
59 /* Block layer tags. */
60 struct blk_mq_tag_set tag_set;
62 /* Process context for config space updates */
63 struct work_struct config_work;
66 * Tracks references from block_device_operations open/release and
67 * virtio_driver probe/remove so this object can be freed once no
72 /* What host tells us, plus 2 for header & tailer. */
73 unsigned int sg_elems;
75 /* Ida index - used to track minor number allocations. */
80 struct virtio_blk_vq *vqs;
84 struct virtio_blk_outhdr out_hdr;
86 struct sg_table sg_table;
87 struct scatterlist sg[];
90 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
92 switch (vbr->status) {
95 case VIRTIO_BLK_S_UNSUPP:
96 return BLK_STS_NOTSUPP;
102 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
103 struct scatterlist *data_sg, bool have_data)
105 struct scatterlist hdr, status, *sgs[3];
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;
112 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
113 sgs[num_out++] = data_sg;
115 sgs[num_out + num_in++] = data_sg;
118 sg_init_one(&status, &vbr->status, sizeof(vbr->status));
119 sgs[num_out + num_in++] = &status;
121 return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
124 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
126 unsigned short segments = blk_rq_nr_discard_segments(req);
127 unsigned short n = 0;
128 struct virtio_blk_discard_write_zeroes *range;
133 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
135 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
140 * Single max discard segment means multi-range discard isn't
141 * supported, and block layer only runs contiguity merge like
142 * normal RW request. So we can't reply on bio for retrieving
145 if (queue_max_discard_segments(req->q) == 1) {
146 range[0].flags = cpu_to_le32(flags);
147 range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
148 range[0].sector = cpu_to_le64(blk_rq_pos(req));
151 __rq_for_each_bio(bio, req) {
152 u64 sector = bio->bi_iter.bi_sector;
153 u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
155 range[n].flags = cpu_to_le32(flags);
156 range[n].num_sectors = cpu_to_le32(num_sectors);
157 range[n].sector = cpu_to_le64(sector);
162 WARN_ON_ONCE(n != segments);
164 req->special_vec.bv_page = virt_to_page(range);
165 req->special_vec.bv_offset = offset_in_page(range);
166 req->special_vec.bv_len = sizeof(*range) * segments;
167 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
172 static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
174 if (blk_rq_nr_phys_segments(req))
175 sg_free_table_chained(&vbr->sg_table,
176 VIRTIO_BLK_INLINE_SG_CNT);
179 static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
180 struct virtblk_req *vbr)
184 if (!blk_rq_nr_phys_segments(req))
187 vbr->sg_table.sgl = vbr->sg;
188 err = sg_alloc_table_chained(&vbr->sg_table,
189 blk_rq_nr_phys_segments(req),
191 VIRTIO_BLK_INLINE_SG_CNT);
195 return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl);
198 static void virtblk_cleanup_cmd(struct request *req)
200 if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
201 kfree(bvec_virt(&req->special_vec));
204 static int virtblk_setup_cmd(struct virtio_device *vdev, struct request *req,
205 struct virtblk_req *vbr)
210 vbr->out_hdr.sector = 0;
212 switch (req_op(req)) {
214 type = VIRTIO_BLK_T_IN;
215 vbr->out_hdr.sector = cpu_to_virtio64(vdev,
219 type = VIRTIO_BLK_T_OUT;
220 vbr->out_hdr.sector = cpu_to_virtio64(vdev,
224 type = VIRTIO_BLK_T_FLUSH;
227 type = VIRTIO_BLK_T_DISCARD;
229 case REQ_OP_WRITE_ZEROES:
230 type = VIRTIO_BLK_T_WRITE_ZEROES;
231 unmap = !(req->cmd_flags & REQ_NOUNMAP);
234 type = VIRTIO_BLK_T_GET_ID;
238 return BLK_STS_IOERR;
241 vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
242 vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
244 if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
245 if (virtblk_setup_discard_write_zeroes(req, unmap))
246 return BLK_STS_RESOURCE;
252 static inline void virtblk_request_done(struct request *req)
254 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
256 virtblk_unmap_data(req, vbr);
257 virtblk_cleanup_cmd(req);
258 blk_mq_end_request(req, virtblk_result(vbr));
261 static void virtblk_done(struct virtqueue *vq)
263 struct virtio_blk *vblk = vq->vdev->priv;
264 bool req_done = false;
266 struct virtblk_req *vbr;
270 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
272 virtqueue_disable_cb(vq);
273 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
274 struct request *req = blk_mq_rq_from_pdu(vbr);
276 if (likely(!blk_should_fake_timeout(req->q)))
277 blk_mq_complete_request(req);
280 if (unlikely(virtqueue_is_broken(vq)))
282 } while (!virtqueue_enable_cb(vq));
284 /* In case queue is stopped waiting for more buffers. */
286 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
287 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
290 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
292 struct virtio_blk *vblk = hctx->queue->queuedata;
293 struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
296 spin_lock_irq(&vq->lock);
297 kick = virtqueue_kick_prepare(vq->vq);
298 spin_unlock_irq(&vq->lock);
301 virtqueue_notify(vq->vq);
304 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
305 const struct blk_mq_queue_data *bd)
307 struct virtio_blk *vblk = hctx->queue->queuedata;
308 struct request *req = bd->rq;
309 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
312 int qid = hctx->queue_num;
316 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
318 err = virtblk_setup_cmd(vblk->vdev, req, vbr);
322 blk_mq_start_request(req);
324 num = virtblk_map_data(hctx, req, vbr);
325 if (unlikely(num < 0)) {
326 virtblk_cleanup_cmd(req);
327 return BLK_STS_RESOURCE;
330 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
331 err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg_table.sgl, num);
333 virtqueue_kick(vblk->vqs[qid].vq);
334 /* Don't stop the queue if -ENOMEM: we may have failed to
335 * bounce the buffer due to global resource outage.
338 blk_mq_stop_hw_queue(hctx);
339 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
340 virtblk_unmap_data(req, vbr);
341 virtblk_cleanup_cmd(req);
344 return BLK_STS_DEV_RESOURCE;
346 return BLK_STS_RESOURCE;
348 return BLK_STS_IOERR;
352 if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
354 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
357 virtqueue_notify(vblk->vqs[qid].vq);
361 /* return id (s/n) string for *disk to *id_str
363 static int virtblk_get_id(struct gendisk *disk, char *id_str)
365 struct virtio_blk *vblk = disk->private_data;
366 struct request_queue *q = vblk->disk->queue;
370 req = blk_get_request(q, REQ_OP_DRV_IN, 0);
374 err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
378 blk_execute_rq(vblk->disk, req, false);
379 err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
381 blk_put_request(req);
385 static void virtblk_get(struct virtio_blk *vblk)
387 refcount_inc(&vblk->refs);
390 static void virtblk_put(struct virtio_blk *vblk)
392 if (refcount_dec_and_test(&vblk->refs)) {
393 ida_simple_remove(&vd_index_ida, vblk->index);
394 mutex_destroy(&vblk->vdev_mutex);
399 static int virtblk_open(struct block_device *bd, fmode_t mode)
401 struct virtio_blk *vblk = bd->bd_disk->private_data;
404 mutex_lock(&vblk->vdev_mutex);
411 mutex_unlock(&vblk->vdev_mutex);
415 static void virtblk_release(struct gendisk *disk, fmode_t mode)
417 struct virtio_blk *vblk = disk->private_data;
422 /* We provide getgeo only to please some old bootloader/partitioning tools */
423 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
425 struct virtio_blk *vblk = bd->bd_disk->private_data;
428 mutex_lock(&vblk->vdev_mutex);
435 /* see if the host passed in geometry config */
436 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
437 virtio_cread(vblk->vdev, struct virtio_blk_config,
438 geometry.cylinders, &geo->cylinders);
439 virtio_cread(vblk->vdev, struct virtio_blk_config,
440 geometry.heads, &geo->heads);
441 virtio_cread(vblk->vdev, struct virtio_blk_config,
442 geometry.sectors, &geo->sectors);
444 /* some standard values, similar to sd */
446 geo->sectors = 1 << 5;
447 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
450 mutex_unlock(&vblk->vdev_mutex);
454 static const struct block_device_operations virtblk_fops = {
455 .owner = THIS_MODULE,
456 .open = virtblk_open,
457 .release = virtblk_release,
458 .getgeo = virtblk_getgeo,
461 static int index_to_minor(int index)
463 return index << PART_BITS;
466 static int minor_to_index(int minor)
468 return minor >> PART_BITS;
471 static ssize_t serial_show(struct device *dev,
472 struct device_attribute *attr, char *buf)
474 struct gendisk *disk = dev_to_disk(dev);
477 /* sysfs gives us a PAGE_SIZE buffer */
478 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
480 buf[VIRTIO_BLK_ID_BYTES] = '\0';
481 err = virtblk_get_id(disk, buf);
485 if (err == -EIO) /* Unsupported? Make it empty. */
491 static DEVICE_ATTR_RO(serial);
493 /* The queue's logical block size must be set before calling this */
494 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
496 struct virtio_device *vdev = vblk->vdev;
497 struct request_queue *q = vblk->disk->queue;
498 char cap_str_2[10], cap_str_10[10];
499 unsigned long long nblocks;
502 /* Host must always specify the capacity. */
503 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
505 nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
507 string_get_size(nblocks, queue_logical_block_size(q),
508 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
509 string_get_size(nblocks, queue_logical_block_size(q),
510 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
512 dev_notice(&vdev->dev,
513 "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
514 vblk->disk->disk_name,
515 resize ? "new size: " : "",
517 queue_logical_block_size(q),
521 set_capacity_and_notify(vblk->disk, capacity);
524 static void virtblk_config_changed_work(struct work_struct *work)
526 struct virtio_blk *vblk =
527 container_of(work, struct virtio_blk, config_work);
529 virtblk_update_capacity(vblk, true);
532 static void virtblk_config_changed(struct virtio_device *vdev)
534 struct virtio_blk *vblk = vdev->priv;
536 queue_work(virtblk_wq, &vblk->config_work);
539 static int init_vq(struct virtio_blk *vblk)
543 vq_callback_t **callbacks;
545 struct virtqueue **vqs;
546 unsigned short num_vqs;
547 struct virtio_device *vdev = vblk->vdev;
548 struct irq_affinity desc = { 0, };
550 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
551 struct virtio_blk_config, num_queues,
556 num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
558 vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
562 names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
563 callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
564 vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
565 if (!names || !callbacks || !vqs) {
570 for (i = 0; i < num_vqs; i++) {
571 callbacks[i] = virtblk_done;
572 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
573 names[i] = vblk->vqs[i].name;
576 /* Discover virtqueues and write information to configuration. */
577 err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
581 for (i = 0; i < num_vqs; i++) {
582 spin_lock_init(&vblk->vqs[i].lock);
583 vblk->vqs[i].vq = vqs[i];
585 vblk->num_vqs = num_vqs;
597 * Legacy naming scheme used for virtio devices. We are stuck with it for
598 * virtio blk but don't ever use it for any new driver.
600 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
602 const int base = 'z' - 'a' + 1;
603 char *begin = buf + strlen(prefix);
604 char *end = buf + buflen;
614 *--p = 'a' + (index % unit);
615 index = (index / unit) - 1;
616 } while (index >= 0);
618 memmove(begin, p, end - p);
619 memcpy(buf, prefix, strlen(prefix));
624 static int virtblk_get_cache_mode(struct virtio_device *vdev)
629 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
630 struct virtio_blk_config, wce,
634 * If WCE is not configurable and flush is not available,
635 * assume no writeback cache is in use.
638 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
643 static void virtblk_update_cache_mode(struct virtio_device *vdev)
645 u8 writeback = virtblk_get_cache_mode(vdev);
646 struct virtio_blk *vblk = vdev->priv;
648 blk_queue_write_cache(vblk->disk->queue, writeback, false);
651 static const char *const virtblk_cache_types[] = {
652 "write through", "write back"
656 cache_type_store(struct device *dev, struct device_attribute *attr,
657 const char *buf, size_t count)
659 struct gendisk *disk = dev_to_disk(dev);
660 struct virtio_blk *vblk = disk->private_data;
661 struct virtio_device *vdev = vblk->vdev;
664 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
665 i = sysfs_match_string(virtblk_cache_types, buf);
669 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
670 virtblk_update_cache_mode(vdev);
675 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
677 struct gendisk *disk = dev_to_disk(dev);
678 struct virtio_blk *vblk = disk->private_data;
679 u8 writeback = virtblk_get_cache_mode(vblk->vdev);
681 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
682 return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
685 static DEVICE_ATTR_RW(cache_type);
687 static struct attribute *virtblk_attrs[] = {
688 &dev_attr_serial.attr,
689 &dev_attr_cache_type.attr,
693 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
694 struct attribute *a, int n)
696 struct device *dev = kobj_to_dev(kobj);
697 struct gendisk *disk = dev_to_disk(dev);
698 struct virtio_blk *vblk = disk->private_data;
699 struct virtio_device *vdev = vblk->vdev;
701 if (a == &dev_attr_cache_type.attr &&
702 !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
708 static const struct attribute_group virtblk_attr_group = {
709 .attrs = virtblk_attrs,
710 .is_visible = virtblk_attrs_are_visible,
713 static const struct attribute_group *virtblk_attr_groups[] = {
718 static int virtblk_map_queues(struct blk_mq_tag_set *set)
720 struct virtio_blk *vblk = set->driver_data;
722 return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
726 static const struct blk_mq_ops virtio_mq_ops = {
727 .queue_rq = virtio_queue_rq,
728 .commit_rqs = virtio_commit_rqs,
729 .complete = virtblk_request_done,
730 .map_queues = virtblk_map_queues,
733 static unsigned int virtblk_queue_depth;
734 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
736 static int virtblk_probe(struct virtio_device *vdev)
738 struct virtio_blk *vblk;
739 struct request_queue *q;
742 u32 v, blk_size, max_size, sg_elems, opt_io_size;
744 u8 physical_block_exp, alignment_offset;
745 unsigned int queue_depth;
747 if (!vdev->config->get) {
748 dev_err(&vdev->dev, "%s failure: config access disabled\n",
753 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
759 /* We need to know how many segments before we allocate. */
760 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
761 struct virtio_blk_config, seg_max,
764 /* We need at least one SG element, whatever they say. */
765 if (err || !sg_elems)
768 /* Prevent integer overflows and honor max vq size */
769 sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
771 /* We need extra sg elements at head and tail. */
773 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
779 /* This reference is dropped in virtblk_remove(). */
780 refcount_set(&vblk->refs, 1);
781 mutex_init(&vblk->vdev_mutex);
784 vblk->sg_elems = sg_elems;
786 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
792 /* Default queue sizing is to fill the ring. */
793 if (!virtblk_queue_depth) {
794 queue_depth = vblk->vqs[0].vq->num_free;
795 /* ... but without indirect descs, we use 2 descs per req */
796 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
799 queue_depth = virtblk_queue_depth;
802 memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
803 vblk->tag_set.ops = &virtio_mq_ops;
804 vblk->tag_set.queue_depth = queue_depth;
805 vblk->tag_set.numa_node = NUMA_NO_NODE;
806 vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
807 vblk->tag_set.cmd_size =
808 sizeof(struct virtblk_req) +
809 sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
810 vblk->tag_set.driver_data = vblk;
811 vblk->tag_set.nr_hw_queues = vblk->num_vqs;
813 err = blk_mq_alloc_tag_set(&vblk->tag_set);
817 vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
818 if (IS_ERR(vblk->disk)) {
819 err = PTR_ERR(vblk->disk);
822 q = vblk->disk->queue;
824 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
826 vblk->disk->major = major;
827 vblk->disk->first_minor = index_to_minor(index);
828 vblk->disk->minors = 1 << PART_BITS;
829 vblk->disk->private_data = vblk;
830 vblk->disk->fops = &virtblk_fops;
831 vblk->disk->flags |= GENHD_FL_EXT_DEVT;
834 /* configure queue flush support */
835 virtblk_update_cache_mode(vdev);
837 /* If disk is read-only in the host, the guest should obey */
838 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
839 set_disk_ro(vblk->disk, 1);
841 /* We can handle whatever the host told us to handle. */
842 blk_queue_max_segments(q, vblk->sg_elems-2);
844 /* No real sector limit. */
845 blk_queue_max_hw_sectors(q, -1U);
847 max_size = virtio_max_dma_size(vdev);
849 /* Host can optionally specify maximum segment size and number of
851 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
852 struct virtio_blk_config, size_max, &v);
854 max_size = min(max_size, v);
856 blk_queue_max_segment_size(q, max_size);
858 /* Host can optionally specify the block size of the device */
859 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
860 struct virtio_blk_config, blk_size,
863 err = blk_validate_block_size(blk_size);
866 "virtio_blk: invalid block size: 0x%x\n",
868 goto out_cleanup_disk;
871 blk_queue_logical_block_size(q, blk_size);
873 blk_size = queue_logical_block_size(q);
875 /* Use topology information if available */
876 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
877 struct virtio_blk_config, physical_block_exp,
878 &physical_block_exp);
879 if (!err && physical_block_exp)
880 blk_queue_physical_block_size(q,
881 blk_size * (1 << physical_block_exp));
883 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
884 struct virtio_blk_config, alignment_offset,
886 if (!err && alignment_offset)
887 blk_queue_alignment_offset(q, blk_size * alignment_offset);
889 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
890 struct virtio_blk_config, min_io_size,
892 if (!err && min_io_size)
893 blk_queue_io_min(q, blk_size * min_io_size);
895 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
896 struct virtio_blk_config, opt_io_size,
898 if (!err && opt_io_size)
899 blk_queue_io_opt(q, blk_size * opt_io_size);
901 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
902 virtio_cread(vdev, struct virtio_blk_config,
903 discard_sector_alignment, &v);
905 q->limits.discard_granularity = v << SECTOR_SHIFT;
907 q->limits.discard_granularity = blk_size;
909 virtio_cread(vdev, struct virtio_blk_config,
910 max_discard_sectors, &v);
911 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
913 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
917 * max_discard_seg == 0 is out of spec but we always
922 blk_queue_max_discard_segments(q,
923 min(v, MAX_DISCARD_SEGMENTS));
925 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
928 if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
929 virtio_cread(vdev, struct virtio_blk_config,
930 max_write_zeroes_sectors, &v);
931 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
934 virtblk_update_capacity(vblk, false);
935 virtio_device_ready(vdev);
937 err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
939 goto out_cleanup_disk;
944 blk_cleanup_disk(vblk->disk);
946 blk_mq_free_tag_set(&vblk->tag_set);
948 vdev->config->del_vqs(vdev);
953 ida_simple_remove(&vd_index_ida, index);
958 static void virtblk_remove(struct virtio_device *vdev)
960 struct virtio_blk *vblk = vdev->priv;
962 /* Make sure no work handler is accessing the device. */
963 flush_work(&vblk->config_work);
965 del_gendisk(vblk->disk);
966 blk_cleanup_disk(vblk->disk);
967 blk_mq_free_tag_set(&vblk->tag_set);
969 mutex_lock(&vblk->vdev_mutex);
971 /* Stop all the virtqueues. */
972 vdev->config->reset(vdev);
974 /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
977 vdev->config->del_vqs(vdev);
980 mutex_unlock(&vblk->vdev_mutex);
985 #ifdef CONFIG_PM_SLEEP
986 static int virtblk_freeze(struct virtio_device *vdev)
988 struct virtio_blk *vblk = vdev->priv;
990 /* Ensure we don't receive any more interrupts */
991 vdev->config->reset(vdev);
993 /* Make sure no work handler is accessing the device. */
994 flush_work(&vblk->config_work);
996 blk_mq_quiesce_queue(vblk->disk->queue);
998 vdev->config->del_vqs(vdev);
1004 static int virtblk_restore(struct virtio_device *vdev)
1006 struct virtio_blk *vblk = vdev->priv;
1009 ret = init_vq(vdev->priv);
1013 virtio_device_ready(vdev);
1015 blk_mq_unquiesce_queue(vblk->disk->queue);
1020 static const struct virtio_device_id id_table[] = {
1021 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1025 static unsigned int features_legacy[] = {
1026 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1027 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1028 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1029 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1032 static unsigned int features[] = {
1033 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1034 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1035 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1036 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1039 static struct virtio_driver virtio_blk = {
1040 .feature_table = features,
1041 .feature_table_size = ARRAY_SIZE(features),
1042 .feature_table_legacy = features_legacy,
1043 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
1044 .driver.name = KBUILD_MODNAME,
1045 .driver.owner = THIS_MODULE,
1046 .id_table = id_table,
1047 .probe = virtblk_probe,
1048 .remove = virtblk_remove,
1049 .config_changed = virtblk_config_changed,
1050 #ifdef CONFIG_PM_SLEEP
1051 .freeze = virtblk_freeze,
1052 .restore = virtblk_restore,
1056 static int __init init(void)
1060 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1064 major = register_blkdev(0, "virtblk");
1067 goto out_destroy_workqueue;
1070 error = register_virtio_driver(&virtio_blk);
1072 goto out_unregister_blkdev;
1075 out_unregister_blkdev:
1076 unregister_blkdev(major, "virtblk");
1077 out_destroy_workqueue:
1078 destroy_workqueue(virtblk_wq);
1082 static void __exit fini(void)
1084 unregister_virtio_driver(&virtio_blk);
1085 unregister_blkdev(major, "virtblk");
1086 destroy_workqueue(virtblk_wq);
1091 MODULE_DEVICE_TABLE(virtio, id_table);
1092 MODULE_DESCRIPTION("Virtio block driver");
1093 MODULE_LICENSE("GPL");