Commit | Line | Data |
---|---|---|
09c434b8 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
e467cde2 RR |
2 | //#define DEBUG |
3 | #include <linux/spinlock.h> | |
5a0e3ad6 | 4 | #include <linux/slab.h> |
e467cde2 RR |
5 | #include <linux/blkdev.h> |
6 | #include <linux/hdreg.h> | |
0c8d44f2 | 7 | #include <linux/module.h> |
4678d6f9 | 8 | #include <linux/mutex.h> |
ad71473d | 9 | #include <linux/interrupt.h> |
e467cde2 RR |
10 | #include <linux/virtio.h> |
11 | #include <linux/virtio_blk.h> | |
3d1266c7 | 12 | #include <linux/scatterlist.h> |
7a7c924c | 13 | #include <linux/string_helpers.h> |
5087a50e | 14 | #include <linux/idr.h> |
1cf7e9c6 | 15 | #include <linux/blk-mq.h> |
ad71473d | 16 | #include <linux/blk-mq-virtio.h> |
1cf7e9c6 | 17 | #include <linux/numa.h> |
95bfec41 | 18 | #include <linux/vmalloc.h> |
55a2415b | 19 | #include <uapi/linux/virtio_ring.h> |
3d1266c7 | 20 | |
4f3bf19c | 21 | #define PART_BITS 4 |
6a27b656 | 22 | #define VQ_NAME_LEN 16 |
1f23816b | 23 | #define MAX_DISCARD_SEGMENTS 256u |
e467cde2 | 24 | |
63947b34 SH |
25 | /* The maximum number of sg elements that fit into a virtqueue */ |
26 | #define VIRTIO_BLK_MAX_SG_ELEMS 32768 | |
27 | ||
02746e26 MG |
28 | #ifdef CONFIG_ARCH_NO_SG_CHAIN |
29 | #define VIRTIO_BLK_INLINE_SG_CNT 0 | |
30 | #else | |
31 | #define VIRTIO_BLK_INLINE_SG_CNT 2 | |
32 | #endif | |
33 | ||
0989c41b | 34 | static unsigned int num_request_queues; |
ead65f76 | 35 | module_param(num_request_queues, uint, 0644); |
0989c41b MG |
36 | MODULE_PARM_DESC(num_request_queues, |
37 | "Limit the number of request queues to use for blk device. " | |
38 | "0 for no limit. " | |
39 | "Values > nr_cpu_ids truncated to nr_cpu_ids."); | |
40 | ||
4e040052 SK |
41 | static unsigned int poll_queues; |
42 | module_param(poll_queues, uint, 0644); | |
43 | MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/O"); | |
44 | ||
5087a50e MT |
45 | static int major; |
46 | static DEFINE_IDA(vd_index_ida); | |
47 | ||
2a647bfe | 48 | static struct workqueue_struct *virtblk_wq; |
4f3bf19c | 49 | |
6a27b656 ML |
50 | struct virtio_blk_vq { |
51 | struct virtqueue *vq; | |
52 | spinlock_t lock; | |
53 | char name[VQ_NAME_LEN]; | |
54 | } ____cacheline_aligned_in_smp; | |
55 | ||
bb6ec576 | 56 | struct virtio_blk { |
90b5feb8 SH |
57 | /* |
58 | * This mutex must be held by anything that may run after | |
59 | * virtblk_remove() sets vblk->vdev to NULL. | |
60 | * | |
61 | * blk-mq, virtqueue processing, and sysfs attribute code paths are | |
62 | * shut down before vblk->vdev is set to NULL and therefore do not need | |
63 | * to hold this mutex. | |
64 | */ | |
65 | struct mutex vdev_mutex; | |
e467cde2 | 66 | struct virtio_device *vdev; |
e467cde2 RR |
67 | |
68 | /* The disk structure for the kernel. */ | |
69 | struct gendisk *disk; | |
70 | ||
24d2f903 CH |
71 | /* Block layer tags. */ |
72 | struct blk_mq_tag_set tag_set; | |
73 | ||
7a7c924c CH |
74 | /* Process context for config space updates */ |
75 | struct work_struct config_work; | |
76 | ||
5087a50e MT |
77 | /* Ida index - used to track minor number allocations. */ |
78 | int index; | |
6a27b656 ML |
79 | |
80 | /* num of vqs */ | |
81 | int num_vqs; | |
4e040052 | 82 | int io_queues[HCTX_MAX_TYPES]; |
6a27b656 | 83 | struct virtio_blk_vq *vqs; |
95bfec41 DF |
84 | |
85 | /* For zoned device */ | |
86 | unsigned int zone_sectors; | |
e467cde2 RR |
87 | }; |
88 | ||
bb6ec576 | 89 | struct virtblk_req { |
95bfec41 | 90 | /* Out header */ |
97b50a65 | 91 | struct virtio_blk_outhdr out_hdr; |
95bfec41 DF |
92 | |
93 | /* In header */ | |
94 | union { | |
95 | u8 status; | |
96 | ||
97 | /* | |
98 | * The zone append command has an extended in header. | |
f1ba4e67 DF |
99 | * The status field in zone_append_in_hdr must always |
100 | * be the last byte. | |
95bfec41 DF |
101 | */ |
102 | struct { | |
f1ba4e67 | 103 | __virtio64 sector; |
95bfec41 | 104 | u8 status; |
f1ba4e67 DF |
105 | } zone_append; |
106 | } in_hdr; | |
95bfec41 DF |
107 | |
108 | size_t in_hdr_len; | |
109 | ||
02746e26 | 110 | struct sg_table sg_table; |
a98755c5 | 111 | struct scatterlist sg[]; |
e467cde2 RR |
112 | }; |
113 | ||
95bfec41 | 114 | static inline blk_status_t virtblk_result(u8 status) |
a98755c5 | 115 | { |
95bfec41 | 116 | switch (status) { |
a98755c5 | 117 | case VIRTIO_BLK_S_OK: |
2a842aca | 118 | return BLK_STS_OK; |
a98755c5 | 119 | case VIRTIO_BLK_S_UNSUPP: |
2a842aca | 120 | return BLK_STS_NOTSUPP; |
95bfec41 DF |
121 | case VIRTIO_BLK_S_ZONE_OPEN_RESOURCE: |
122 | return BLK_STS_ZONE_OPEN_RESOURCE; | |
123 | case VIRTIO_BLK_S_ZONE_ACTIVE_RESOURCE: | |
124 | return BLK_STS_ZONE_ACTIVE_RESOURCE; | |
125 | case VIRTIO_BLK_S_IOERR: | |
126 | case VIRTIO_BLK_S_ZONE_UNALIGNED_WP: | |
a98755c5 | 127 | default: |
2a842aca | 128 | return BLK_STS_IOERR; |
a98755c5 AH |
129 | } |
130 | } | |
131 | ||
8d12ec10 SY |
132 | static inline struct virtio_blk_vq *get_virtio_blk_vq(struct blk_mq_hw_ctx *hctx) |
133 | { | |
134 | struct virtio_blk *vblk = hctx->queue->queuedata; | |
135 | struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num]; | |
136 | ||
137 | return vq; | |
138 | } | |
139 | ||
0e9911fa | 140 | static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr) |
97b50a65 | 141 | { |
95bfec41 | 142 | struct scatterlist out_hdr, in_hdr, *sgs[3]; |
97b50a65 CH |
143 | unsigned int num_out = 0, num_in = 0; |
144 | ||
95bfec41 DF |
145 | sg_init_one(&out_hdr, &vbr->out_hdr, sizeof(vbr->out_hdr)); |
146 | sgs[num_out++] = &out_hdr; | |
20af3cfd | 147 | |
0e9911fa | 148 | if (vbr->sg_table.nents) { |
19c1c5a6 | 149 | if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT)) |
0e9911fa | 150 | sgs[num_out++] = vbr->sg_table.sgl; |
8f39db9d | 151 | else |
0e9911fa | 152 | sgs[num_out + num_in++] = vbr->sg_table.sgl; |
20af3cfd PB |
153 | } |
154 | ||
f1ba4e67 | 155 | sg_init_one(&in_hdr, &vbr->in_hdr.status, vbr->in_hdr_len); |
95bfec41 | 156 | sgs[num_out + num_in++] = &in_hdr; |
8f39db9d PB |
157 | |
158 | return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC); | |
5ee21a52 PB |
159 | } |
160 | ||
e60d6407 | 161 | static int virtblk_setup_discard_write_zeroes_erase(struct request *req, bool unmap) |
1f23816b CL |
162 | { |
163 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
164 | unsigned short n = 0; | |
165 | struct virtio_blk_discard_write_zeroes *range; | |
166 | struct bio *bio; | |
167 | u32 flags = 0; | |
168 | ||
169 | if (unmap) | |
170 | flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP; | |
171 | ||
172 | range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC); | |
173 | if (!range) | |
174 | return -ENOMEM; | |
175 | ||
af822aa6 ML |
176 | /* |
177 | * Single max discard segment means multi-range discard isn't | |
178 | * supported, and block layer only runs contiguity merge like | |
179 | * normal RW request. So we can't reply on bio for retrieving | |
180 | * each range info. | |
181 | */ | |
182 | if (queue_max_discard_segments(req->q) == 1) { | |
183 | range[0].flags = cpu_to_le32(flags); | |
184 | range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req)); | |
185 | range[0].sector = cpu_to_le64(blk_rq_pos(req)); | |
186 | n = 1; | |
187 | } else { | |
188 | __rq_for_each_bio(bio, req) { | |
189 | u64 sector = bio->bi_iter.bi_sector; | |
190 | u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
191 | ||
192 | range[n].flags = cpu_to_le32(flags); | |
193 | range[n].num_sectors = cpu_to_le32(num_sectors); | |
194 | range[n].sector = cpu_to_le64(sector); | |
195 | n++; | |
196 | } | |
1f23816b CL |
197 | } |
198 | ||
af822aa6 ML |
199 | WARN_ON_ONCE(n != segments); |
200 | ||
b831f3a1 | 201 | bvec_set_virt(&req->special_vec, range, sizeof(*range) * segments); |
1f23816b CL |
202 | req->rq_flags |= RQF_SPECIAL_PAYLOAD; |
203 | ||
204 | return 0; | |
205 | } | |
206 | ||
02746e26 | 207 | static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr) |
a98755c5 | 208 | { |
02746e26 MG |
209 | if (blk_rq_nr_phys_segments(req)) |
210 | sg_free_table_chained(&vbr->sg_table, | |
211 | VIRTIO_BLK_INLINE_SG_CNT); | |
212 | } | |
213 | ||
214 | static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req, | |
215 | struct virtblk_req *vbr) | |
216 | { | |
217 | int err; | |
218 | ||
219 | if (!blk_rq_nr_phys_segments(req)) | |
220 | return 0; | |
a98755c5 | 221 | |
02746e26 MG |
222 | vbr->sg_table.sgl = vbr->sg; |
223 | err = sg_alloc_table_chained(&vbr->sg_table, | |
224 | blk_rq_nr_phys_segments(req), | |
225 | vbr->sg_table.sgl, | |
226 | VIRTIO_BLK_INLINE_SG_CNT); | |
227 | if (unlikely(err)) | |
228 | return -ENOMEM; | |
a98755c5 | 229 | |
02746e26 MG |
230 | return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl); |
231 | } | |
232 | ||
233 | static void virtblk_cleanup_cmd(struct request *req) | |
234 | { | |
358b348b CH |
235 | if (req->rq_flags & RQF_SPECIAL_PAYLOAD) |
236 | kfree(bvec_virt(&req->special_vec)); | |
02746e26 MG |
237 | } |
238 | ||
f0839372 MT |
239 | static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev, |
240 | struct request *req, | |
241 | struct virtblk_req *vbr) | |
02746e26 | 242 | { |
f1ba4e67 | 243 | size_t in_hdr_len = sizeof(vbr->in_hdr.status); |
02746e26 MG |
244 | bool unmap = false; |
245 | u32 type; | |
95bfec41 | 246 | u64 sector = 0; |
02746e26 | 247 | |
f1ba4e67 DF |
248 | if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && op_is_zone_mgmt(req_op(req))) |
249 | return BLK_STS_NOTSUPP; | |
250 | ||
95bfec41 DF |
251 | /* Set fields for all request types */ |
252 | vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req)); | |
02746e26 MG |
253 | |
254 | switch (req_op(req)) { | |
255 | case REQ_OP_READ: | |
256 | type = VIRTIO_BLK_T_IN; | |
95bfec41 | 257 | sector = blk_rq_pos(req); |
02746e26 MG |
258 | break; |
259 | case REQ_OP_WRITE: | |
260 | type = VIRTIO_BLK_T_OUT; | |
95bfec41 | 261 | sector = blk_rq_pos(req); |
02746e26 MG |
262 | break; |
263 | case REQ_OP_FLUSH: | |
264 | type = VIRTIO_BLK_T_FLUSH; | |
265 | break; | |
266 | case REQ_OP_DISCARD: | |
267 | type = VIRTIO_BLK_T_DISCARD; | |
268 | break; | |
269 | case REQ_OP_WRITE_ZEROES: | |
270 | type = VIRTIO_BLK_T_WRITE_ZEROES; | |
271 | unmap = !(req->cmd_flags & REQ_NOUNMAP); | |
272 | break; | |
e60d6407 AK |
273 | case REQ_OP_SECURE_ERASE: |
274 | type = VIRTIO_BLK_T_SECURE_ERASE; | |
275 | break; | |
95bfec41 DF |
276 | case REQ_OP_ZONE_OPEN: |
277 | type = VIRTIO_BLK_T_ZONE_OPEN; | |
278 | sector = blk_rq_pos(req); | |
279 | break; | |
280 | case REQ_OP_ZONE_CLOSE: | |
281 | type = VIRTIO_BLK_T_ZONE_CLOSE; | |
282 | sector = blk_rq_pos(req); | |
283 | break; | |
284 | case REQ_OP_ZONE_FINISH: | |
285 | type = VIRTIO_BLK_T_ZONE_FINISH; | |
286 | sector = blk_rq_pos(req); | |
02746e26 | 287 | break; |
95bfec41 DF |
288 | case REQ_OP_ZONE_APPEND: |
289 | type = VIRTIO_BLK_T_ZONE_APPEND; | |
290 | sector = blk_rq_pos(req); | |
f1ba4e67 | 291 | in_hdr_len = sizeof(vbr->in_hdr.zone_append); |
02746e26 | 292 | break; |
95bfec41 DF |
293 | case REQ_OP_ZONE_RESET: |
294 | type = VIRTIO_BLK_T_ZONE_RESET; | |
295 | sector = blk_rq_pos(req); | |
296 | break; | |
297 | case REQ_OP_ZONE_RESET_ALL: | |
298 | type = VIRTIO_BLK_T_ZONE_RESET_ALL; | |
299 | break; | |
300 | case REQ_OP_DRV_IN: | |
f1ba4e67 DF |
301 | /* |
302 | * Out header has already been prepared by the caller (virtblk_get_id() | |
303 | * or virtblk_submit_zone_report()), nothing to do here. | |
304 | */ | |
95bfec41 | 305 | return 0; |
02746e26 MG |
306 | default: |
307 | WARN_ON_ONCE(1); | |
308 | return BLK_STS_IOERR; | |
309 | } | |
310 | ||
95bfec41 DF |
311 | /* Set fields for non-REQ_OP_DRV_IN request types */ |
312 | vbr->in_hdr_len = in_hdr_len; | |
02746e26 | 313 | vbr->out_hdr.type = cpu_to_virtio32(vdev, type); |
95bfec41 | 314 | vbr->out_hdr.sector = cpu_to_virtio64(vdev, sector); |
02746e26 | 315 | |
e60d6407 AK |
316 | if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES || |
317 | type == VIRTIO_BLK_T_SECURE_ERASE) { | |
318 | if (virtblk_setup_discard_write_zeroes_erase(req, unmap)) | |
02746e26 MG |
319 | return BLK_STS_RESOURCE; |
320 | } | |
321 | ||
322 | return 0; | |
323 | } | |
324 | ||
f1ba4e67 DF |
325 | /* |
326 | * The status byte is always the last byte of the virtblk request | |
327 | * in-header. This helper fetches its value for all in-header formats | |
328 | * that are currently defined. | |
329 | */ | |
330 | static inline u8 virtblk_vbr_status(struct virtblk_req *vbr) | |
331 | { | |
332 | return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1); | |
333 | } | |
334 | ||
02746e26 MG |
335 | static inline void virtblk_request_done(struct request *req) |
336 | { | |
337 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); | |
f1ba4e67 DF |
338 | blk_status_t status = virtblk_result(virtblk_vbr_status(vbr)); |
339 | struct virtio_blk *vblk = req->mq_hctx->queue->queuedata; | |
02746e26 MG |
340 | |
341 | virtblk_unmap_data(req, vbr); | |
342 | virtblk_cleanup_cmd(req); | |
95bfec41 DF |
343 | |
344 | if (req_op(req) == REQ_OP_ZONE_APPEND) | |
f1ba4e67 DF |
345 | req->__sector = virtio64_to_cpu(vblk->vdev, |
346 | vbr->in_hdr.zone_append.sector); | |
95bfec41 DF |
347 | |
348 | blk_mq_end_request(req, status); | |
a98755c5 AH |
349 | } |
350 | ||
351 | static void virtblk_done(struct virtqueue *vq) | |
e467cde2 RR |
352 | { |
353 | struct virtio_blk *vblk = vq->vdev->priv; | |
afd384f0 MT |
354 | bool req_done = false; |
355 | int qid = vq->index; | |
356 | struct virtblk_req *vbr; | |
e467cde2 | 357 | unsigned long flags; |
afd384f0 | 358 | unsigned int len; |
e467cde2 | 359 | |
afd384f0 | 360 | spin_lock_irqsave(&vblk->vqs[qid].lock, flags); |
bb811108 AH |
361 | do { |
362 | virtqueue_disable_cb(vq); | |
afd384f0 MT |
363 | while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) { |
364 | struct request *req = blk_mq_rq_from_pdu(vbr); | |
85dada09 | 365 | |
afd384f0 MT |
366 | if (likely(!blk_should_fake_timeout(req->q))) |
367 | blk_mq_complete_request(req); | |
368 | req_done = true; | |
369 | } | |
7f03b17d HG |
370 | if (unlikely(virtqueue_is_broken(vq))) |
371 | break; | |
bb811108 | 372 | } while (!virtqueue_enable_cb(vq)); |
1cf7e9c6 | 373 | |
afd384f0 MT |
374 | /* In case queue is stopped waiting for more buffers. */ |
375 | if (req_done) | |
1b4a3258 | 376 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); |
afd384f0 | 377 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
a98755c5 AH |
378 | } |
379 | ||
944e7c87 JA |
380 | static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx) |
381 | { | |
382 | struct virtio_blk *vblk = hctx->queue->queuedata; | |
383 | struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num]; | |
384 | bool kick; | |
385 | ||
386 | spin_lock_irq(&vq->lock); | |
387 | kick = virtqueue_kick_prepare(vq->vq); | |
388 | spin_unlock_irq(&vq->lock); | |
389 | ||
390 | if (kick) | |
391 | virtqueue_notify(vq->vq); | |
392 | } | |
393 | ||
258896fc DF |
394 | static blk_status_t virtblk_fail_to_queue(struct request *req, int rc) |
395 | { | |
396 | virtblk_cleanup_cmd(req); | |
397 | switch (rc) { | |
398 | case -ENOSPC: | |
399 | return BLK_STS_DEV_RESOURCE; | |
400 | case -ENOMEM: | |
401 | return BLK_STS_RESOURCE; | |
402 | default: | |
403 | return BLK_STS_IOERR; | |
404 | } | |
405 | } | |
406 | ||
0e9911fa SK |
407 | static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx, |
408 | struct virtio_blk *vblk, | |
409 | struct request *req, | |
410 | struct virtblk_req *vbr) | |
411 | { | |
412 | blk_status_t status; | |
a26116c1 | 413 | int num; |
0e9911fa SK |
414 | |
415 | status = virtblk_setup_cmd(vblk->vdev, req, vbr); | |
416 | if (unlikely(status)) | |
417 | return status; | |
418 | ||
a26116c1 RM |
419 | num = virtblk_map_data(hctx, req, vbr); |
420 | if (unlikely(num < 0)) | |
258896fc | 421 | return virtblk_fail_to_queue(req, -ENOMEM); |
a26116c1 | 422 | vbr->sg_table.nents = num; |
0e9911fa | 423 | |
37fafe6b SK |
424 | blk_mq_start_request(req); |
425 | ||
0e9911fa SK |
426 | return BLK_STS_OK; |
427 | } | |
428 | ||
fc17b653 | 429 | static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx, |
74c45052 | 430 | const struct blk_mq_queue_data *bd) |
e467cde2 | 431 | { |
1cf7e9c6 | 432 | struct virtio_blk *vblk = hctx->queue->queuedata; |
74c45052 | 433 | struct request *req = bd->rq; |
9d74e257 | 434 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
1cf7e9c6 | 435 | unsigned long flags; |
6a27b656 | 436 | int qid = hctx->queue_num; |
e8edca6f | 437 | bool notify = false; |
f0839372 MT |
438 | blk_status_t status; |
439 | int err; | |
e467cde2 | 440 | |
0e9911fa | 441 | status = virtblk_prep_rq(hctx, vblk, req, vbr); |
f0839372 MT |
442 | if (unlikely(status)) |
443 | return status; | |
aebf526b | 444 | |
6a27b656 | 445 | spin_lock_irqsave(&vblk->vqs[qid].lock, flags); |
0e9911fa | 446 | err = virtblk_add_req(vblk->vqs[qid].vq, vbr); |
5261b85e | 447 | if (err) { |
6a27b656 | 448 | virtqueue_kick(vblk->vqs[qid].vq); |
f5f6b95c HP |
449 | /* Don't stop the queue if -ENOMEM: we may have failed to |
450 | * bounce the buffer due to global resource outage. | |
451 | */ | |
452 | if (err == -ENOSPC) | |
453 | blk_mq_stop_hw_queue(hctx); | |
6a27b656 | 454 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
02746e26 | 455 | virtblk_unmap_data(req, vbr); |
258896fc | 456 | return virtblk_fail_to_queue(req, err); |
a98755c5 AH |
457 | } |
458 | ||
74c45052 | 459 | if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq)) |
e8edca6f | 460 | notify = true; |
6a27b656 | 461 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
e8edca6f ML |
462 | |
463 | if (notify) | |
6a27b656 | 464 | virtqueue_notify(vblk->vqs[qid].vq); |
fc17b653 | 465 | return BLK_STS_OK; |
a98755c5 AH |
466 | } |
467 | ||
0e9911fa SK |
468 | static bool virtblk_prep_rq_batch(struct request *req) |
469 | { | |
470 | struct virtio_blk *vblk = req->mq_hctx->queue->queuedata; | |
471 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); | |
472 | ||
473 | req->mq_hctx->tags->rqs[req->tag] = req; | |
474 | ||
475 | return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK; | |
476 | } | |
477 | ||
478 | static bool virtblk_add_req_batch(struct virtio_blk_vq *vq, | |
37fafe6b | 479 | struct request **rqlist) |
0e9911fa SK |
480 | { |
481 | unsigned long flags; | |
482 | int err; | |
483 | bool kick; | |
484 | ||
485 | spin_lock_irqsave(&vq->lock, flags); | |
486 | ||
487 | while (!rq_list_empty(*rqlist)) { | |
488 | struct request *req = rq_list_pop(rqlist); | |
489 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); | |
490 | ||
491 | err = virtblk_add_req(vq->vq, vbr); | |
492 | if (err) { | |
493 | virtblk_unmap_data(req, vbr); | |
494 | virtblk_cleanup_cmd(req); | |
37fafe6b | 495 | blk_mq_requeue_request(req, true); |
0e9911fa SK |
496 | } |
497 | } | |
498 | ||
499 | kick = virtqueue_kick_prepare(vq->vq); | |
500 | spin_unlock_irqrestore(&vq->lock, flags); | |
501 | ||
502 | return kick; | |
503 | } | |
504 | ||
505 | static void virtio_queue_rqs(struct request **rqlist) | |
506 | { | |
507 | struct request *req, *next, *prev = NULL; | |
508 | struct request *requeue_list = NULL; | |
509 | ||
510 | rq_list_for_each_safe(rqlist, req, next) { | |
8d12ec10 | 511 | struct virtio_blk_vq *vq = get_virtio_blk_vq(req->mq_hctx); |
0e9911fa SK |
512 | bool kick; |
513 | ||
514 | if (!virtblk_prep_rq_batch(req)) { | |
515 | rq_list_move(rqlist, &requeue_list, req, prev); | |
516 | req = prev; | |
517 | if (!req) | |
518 | continue; | |
519 | } | |
520 | ||
521 | if (!next || req->mq_hctx != next->mq_hctx) { | |
522 | req->rq_next = NULL; | |
37fafe6b | 523 | kick = virtblk_add_req_batch(vq, rqlist); |
0e9911fa SK |
524 | if (kick) |
525 | virtqueue_notify(vq->vq); | |
526 | ||
527 | *rqlist = next; | |
528 | prev = NULL; | |
529 | } else | |
530 | prev = req; | |
531 | } | |
532 | ||
533 | *rqlist = requeue_list; | |
534 | } | |
535 | ||
95bfec41 DF |
536 | #ifdef CONFIG_BLK_DEV_ZONED |
537 | static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk, | |
538 | unsigned int nr_zones, | |
95bfec41 DF |
539 | size_t *buflen) |
540 | { | |
541 | struct request_queue *q = vblk->disk->queue; | |
542 | size_t bufsize; | |
543 | void *buf; | |
544 | ||
545 | nr_zones = min_t(unsigned int, nr_zones, | |
f1ba4e67 | 546 | get_capacity(vblk->disk) >> ilog2(vblk->zone_sectors)); |
95bfec41 DF |
547 | |
548 | bufsize = sizeof(struct virtio_blk_zone_report) + | |
549 | nr_zones * sizeof(struct virtio_blk_zone_descriptor); | |
550 | bufsize = min_t(size_t, bufsize, | |
551 | queue_max_hw_sectors(q) << SECTOR_SHIFT); | |
552 | bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT); | |
553 | ||
554 | while (bufsize >= sizeof(struct virtio_blk_zone_report)) { | |
555 | buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY); | |
556 | if (buf) { | |
557 | *buflen = bufsize; | |
558 | return buf; | |
559 | } | |
560 | bufsize >>= 1; | |
561 | } | |
562 | ||
563 | return NULL; | |
564 | } | |
565 | ||
566 | static int virtblk_submit_zone_report(struct virtio_blk *vblk, | |
567 | char *report_buf, size_t report_len, | |
568 | sector_t sector) | |
569 | { | |
570 | struct request_queue *q = vblk->disk->queue; | |
571 | struct request *req; | |
572 | struct virtblk_req *vbr; | |
573 | int err; | |
574 | ||
575 | req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0); | |
576 | if (IS_ERR(req)) | |
577 | return PTR_ERR(req); | |
578 | ||
579 | vbr = blk_mq_rq_to_pdu(req); | |
f1ba4e67 | 580 | vbr->in_hdr_len = sizeof(vbr->in_hdr.status); |
95bfec41 DF |
581 | vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT); |
582 | vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector); | |
583 | ||
584 | err = blk_rq_map_kern(q, req, report_buf, report_len, GFP_KERNEL); | |
585 | if (err) | |
586 | goto out; | |
587 | ||
588 | blk_execute_rq(req, false); | |
f1ba4e67 | 589 | err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status)); |
95bfec41 DF |
590 | out: |
591 | blk_mq_free_request(req); | |
592 | return err; | |
593 | } | |
594 | ||
595 | static int virtblk_parse_zone(struct virtio_blk *vblk, | |
596 | struct virtio_blk_zone_descriptor *entry, | |
f1ba4e67 | 597 | unsigned int idx, report_zones_cb cb, void *data) |
95bfec41 DF |
598 | { |
599 | struct blk_zone zone = { }; | |
600 | ||
f1ba4e67 DF |
601 | zone.start = virtio64_to_cpu(vblk->vdev, entry->z_start); |
602 | if (zone.start + vblk->zone_sectors <= get_capacity(vblk->disk)) | |
603 | zone.len = vblk->zone_sectors; | |
604 | else | |
605 | zone.len = get_capacity(vblk->disk) - zone.start; | |
606 | zone.capacity = virtio64_to_cpu(vblk->vdev, entry->z_cap); | |
607 | zone.wp = virtio64_to_cpu(vblk->vdev, entry->z_wp); | |
608 | ||
609 | switch (entry->z_type) { | |
610 | case VIRTIO_BLK_ZT_SWR: | |
611 | zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ; | |
612 | break; | |
613 | case VIRTIO_BLK_ZT_SWP: | |
614 | zone.type = BLK_ZONE_TYPE_SEQWRITE_PREF; | |
615 | break; | |
616 | case VIRTIO_BLK_ZT_CONV: | |
617 | zone.type = BLK_ZONE_TYPE_CONVENTIONAL; | |
618 | break; | |
619 | default: | |
620 | dev_err(&vblk->vdev->dev, "zone %llu: invalid type %#x\n", | |
621 | zone.start, entry->z_type); | |
622 | return -EIO; | |
95bfec41 DF |
623 | } |
624 | ||
f1ba4e67 DF |
625 | switch (entry->z_state) { |
626 | case VIRTIO_BLK_ZS_EMPTY: | |
627 | zone.cond = BLK_ZONE_COND_EMPTY; | |
628 | break; | |
629 | case VIRTIO_BLK_ZS_CLOSED: | |
630 | zone.cond = BLK_ZONE_COND_CLOSED; | |
631 | break; | |
632 | case VIRTIO_BLK_ZS_FULL: | |
633 | zone.cond = BLK_ZONE_COND_FULL; | |
95bfec41 | 634 | zone.wp = zone.start + zone.len; |
f1ba4e67 DF |
635 | break; |
636 | case VIRTIO_BLK_ZS_EOPEN: | |
637 | zone.cond = BLK_ZONE_COND_EXP_OPEN; | |
638 | break; | |
639 | case VIRTIO_BLK_ZS_IOPEN: | |
640 | zone.cond = BLK_ZONE_COND_IMP_OPEN; | |
641 | break; | |
642 | case VIRTIO_BLK_ZS_NOT_WP: | |
643 | zone.cond = BLK_ZONE_COND_NOT_WP; | |
644 | break; | |
645 | case VIRTIO_BLK_ZS_RDONLY: | |
646 | zone.cond = BLK_ZONE_COND_READONLY; | |
647 | zone.wp = ULONG_MAX; | |
648 | break; | |
649 | case VIRTIO_BLK_ZS_OFFLINE: | |
650 | zone.cond = BLK_ZONE_COND_OFFLINE; | |
651 | zone.wp = ULONG_MAX; | |
652 | break; | |
653 | default: | |
654 | dev_err(&vblk->vdev->dev, "zone %llu: invalid condition %#x\n", | |
655 | zone.start, entry->z_state); | |
656 | return -EIO; | |
657 | } | |
95bfec41 | 658 | |
f1ba4e67 DF |
659 | /* |
660 | * The callback below checks the validity of the reported | |
661 | * entry data, no need to further validate it here. | |
662 | */ | |
95bfec41 DF |
663 | return cb(&zone, idx, data); |
664 | } | |
665 | ||
666 | static int virtblk_report_zones(struct gendisk *disk, sector_t sector, | |
667 | unsigned int nr_zones, report_zones_cb cb, | |
668 | void *data) | |
669 | { | |
670 | struct virtio_blk *vblk = disk->private_data; | |
671 | struct virtio_blk_zone_report *report; | |
f1ba4e67 | 672 | unsigned long long nz, i; |
95bfec41 | 673 | size_t buflen; |
f1ba4e67 DF |
674 | unsigned int zone_idx = 0; |
675 | int ret; | |
95bfec41 DF |
676 | |
677 | if (WARN_ON_ONCE(!vblk->zone_sectors)) | |
678 | return -EOPNOTSUPP; | |
679 | ||
f1ba4e67 | 680 | report = virtblk_alloc_report_buffer(vblk, nr_zones, &buflen); |
95bfec41 DF |
681 | if (!report) |
682 | return -ENOMEM; | |
683 | ||
f1ba4e67 DF |
684 | mutex_lock(&vblk->vdev_mutex); |
685 | ||
686 | if (!vblk->vdev) { | |
687 | ret = -ENXIO; | |
688 | goto fail_report; | |
689 | } | |
690 | ||
95bfec41 DF |
691 | while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) { |
692 | memset(report, 0, buflen); | |
693 | ||
694 | ret = virtblk_submit_zone_report(vblk, (char *)report, | |
695 | buflen, sector); | |
f1ba4e67 DF |
696 | if (ret) |
697 | goto fail_report; | |
698 | ||
699 | nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones), | |
700 | nr_zones); | |
95bfec41 DF |
701 | if (!nz) |
702 | break; | |
703 | ||
704 | for (i = 0; i < nz && zone_idx < nr_zones; i++) { | |
705 | ret = virtblk_parse_zone(vblk, &report->zones[i], | |
f1ba4e67 | 706 | zone_idx, cb, data); |
95bfec41 | 707 | if (ret) |
f1ba4e67 DF |
708 | goto fail_report; |
709 | ||
710 | sector = virtio64_to_cpu(vblk->vdev, | |
711 | report->zones[i].z_start) + | |
712 | vblk->zone_sectors; | |
95bfec41 DF |
713 | zone_idx++; |
714 | } | |
715 | } | |
716 | ||
717 | if (zone_idx > 0) | |
718 | ret = zone_idx; | |
719 | else | |
720 | ret = -EINVAL; | |
f1ba4e67 DF |
721 | fail_report: |
722 | mutex_unlock(&vblk->vdev_mutex); | |
95bfec41 DF |
723 | kvfree(report); |
724 | return ret; | |
725 | } | |
726 | ||
727 | static void virtblk_revalidate_zones(struct virtio_blk *vblk) | |
728 | { | |
729 | u8 model; | |
730 | ||
95bfec41 DF |
731 | virtio_cread(vblk->vdev, struct virtio_blk_config, |
732 | zoned.model, &model); | |
f1ba4e67 DF |
733 | switch (model) { |
734 | default: | |
735 | dev_err(&vblk->vdev->dev, "unknown zone model %d\n", model); | |
736 | fallthrough; | |
737 | case VIRTIO_BLK_Z_NONE: | |
738 | case VIRTIO_BLK_Z_HA: | |
739 | disk_set_zoned(vblk->disk, BLK_ZONED_NONE); | |
740 | return; | |
741 | case VIRTIO_BLK_Z_HM: | |
742 | WARN_ON_ONCE(!vblk->zone_sectors); | |
743 | if (!blk_revalidate_disk_zones(vblk->disk, NULL)) | |
744 | set_capacity_and_notify(vblk->disk, 0); | |
745 | } | |
95bfec41 DF |
746 | } |
747 | ||
748 | static int virtblk_probe_zoned_device(struct virtio_device *vdev, | |
749 | struct virtio_blk *vblk, | |
750 | struct request_queue *q) | |
751 | { | |
f1ba4e67 | 752 | u32 v, wg; |
95bfec41 | 753 | u8 model; |
95bfec41 DF |
754 | |
755 | virtio_cread(vdev, struct virtio_blk_config, | |
756 | zoned.model, &model); | |
757 | ||
758 | switch (model) { | |
759 | case VIRTIO_BLK_Z_NONE: | |
f1ba4e67 DF |
760 | case VIRTIO_BLK_Z_HA: |
761 | /* Present the host-aware device as non-zoned */ | |
95bfec41 DF |
762 | return 0; |
763 | case VIRTIO_BLK_Z_HM: | |
764 | break; | |
95bfec41 DF |
765 | default: |
766 | dev_err(&vdev->dev, "unsupported zone model %d\n", model); | |
767 | return -EINVAL; | |
768 | } | |
769 | ||
770 | dev_dbg(&vdev->dev, "probing host-managed zoned device\n"); | |
771 | ||
772 | disk_set_zoned(vblk->disk, BLK_ZONED_HM); | |
773 | blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q); | |
774 | ||
775 | virtio_cread(vdev, struct virtio_blk_config, | |
776 | zoned.max_open_zones, &v); | |
f1ba4e67 DF |
777 | disk_set_max_open_zones(vblk->disk, v); |
778 | dev_dbg(&vdev->dev, "max open zones = %u\n", v); | |
95bfec41 DF |
779 | |
780 | virtio_cread(vdev, struct virtio_blk_config, | |
781 | zoned.max_active_zones, &v); | |
f1ba4e67 DF |
782 | disk_set_max_active_zones(vblk->disk, v); |
783 | dev_dbg(&vdev->dev, "max active zones = %u\n", v); | |
95bfec41 DF |
784 | |
785 | virtio_cread(vdev, struct virtio_blk_config, | |
f1ba4e67 DF |
786 | zoned.write_granularity, &wg); |
787 | if (!wg) { | |
95bfec41 DF |
788 | dev_warn(&vdev->dev, "zero write granularity reported\n"); |
789 | return -ENODEV; | |
790 | } | |
f1ba4e67 DF |
791 | blk_queue_physical_block_size(q, wg); |
792 | blk_queue_io_min(q, wg); | |
95bfec41 | 793 | |
f1ba4e67 | 794 | dev_dbg(&vdev->dev, "write granularity = %u\n", wg); |
95bfec41 DF |
795 | |
796 | /* | |
797 | * virtio ZBD specification doesn't require zones to be a power of | |
798 | * two sectors in size, but the code in this driver expects that. | |
799 | */ | |
f1ba4e67 DF |
800 | virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors, |
801 | &vblk->zone_sectors); | |
95bfec41 DF |
802 | if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) { |
803 | dev_err(&vdev->dev, | |
804 | "zoned device with non power of two zone size %u\n", | |
805 | vblk->zone_sectors); | |
806 | return -ENODEV; | |
807 | } | |
a3d96ed2 | 808 | blk_queue_chunk_sectors(q, vblk->zone_sectors); |
95bfec41 DF |
809 | dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors); |
810 | ||
811 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) { | |
812 | dev_warn(&vblk->vdev->dev, | |
813 | "ignoring negotiated F_DISCARD for zoned device\n"); | |
814 | blk_queue_max_discard_sectors(q, 0); | |
815 | } | |
816 | ||
a3d96ed2 DLM |
817 | virtio_cread(vdev, struct virtio_blk_config, |
818 | zoned.max_append_sectors, &v); | |
819 | if (!v) { | |
820 | dev_warn(&vdev->dev, "zero max_append_sectors reported\n"); | |
821 | return -ENODEV; | |
95bfec41 | 822 | } |
a3d96ed2 DLM |
823 | if ((v << SECTOR_SHIFT) < wg) { |
824 | dev_err(&vdev->dev, | |
825 | "write granularity %u exceeds max_append_sectors %u limit\n", | |
826 | wg, v); | |
827 | return -ENODEV; | |
828 | } | |
829 | blk_queue_max_zone_append_sectors(q, v); | |
830 | dev_dbg(&vdev->dev, "max append sectors = %u\n", v); | |
95bfec41 | 831 | |
a3d96ed2 | 832 | return blk_revalidate_disk_zones(vblk->disk, NULL); |
95bfec41 DF |
833 | } |
834 | ||
95bfec41 DF |
835 | #else |
836 | ||
837 | /* | |
838 | * Zoned block device support is not configured in this kernel. | |
10805eb5 DF |
839 | * Host-managed zoned devices can't be supported, but others are |
840 | * good to go as regular block devices. | |
95bfec41 DF |
841 | */ |
842 | #define virtblk_report_zones NULL | |
f1ba4e67 | 843 | |
95bfec41 DF |
844 | static inline void virtblk_revalidate_zones(struct virtio_blk *vblk) |
845 | { | |
846 | } | |
f1ba4e67 | 847 | |
95bfec41 DF |
848 | static inline int virtblk_probe_zoned_device(struct virtio_device *vdev, |
849 | struct virtio_blk *vblk, struct request_queue *q) | |
850 | { | |
10805eb5 | 851 | u8 model; |
95bfec41 | 852 | |
10805eb5 DF |
853 | virtio_cread(vdev, struct virtio_blk_config, zoned.model, &model); |
854 | if (model == VIRTIO_BLK_Z_HM) { | |
855 | dev_err(&vdev->dev, | |
856 | "virtio_blk: zoned devices are not supported"); | |
857 | return -EOPNOTSUPP; | |
858 | } | |
859 | ||
860 | return 0; | |
95bfec41 DF |
861 | } |
862 | #endif /* CONFIG_BLK_DEV_ZONED */ | |
863 | ||
4cb2ea28 | 864 | /* return id (s/n) string for *disk to *id_str |
865 | */ | |
866 | static int virtblk_get_id(struct gendisk *disk, char *id_str) | |
867 | { | |
868 | struct virtio_blk *vblk = disk->private_data; | |
f9596695 | 869 | struct request_queue *q = vblk->disk->queue; |
4cb2ea28 | 870 | struct request *req; |
95bfec41 | 871 | struct virtblk_req *vbr; |
e4c4776d | 872 | int err; |
4cb2ea28 | 873 | |
0bf6d96c | 874 | req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0); |
f9596695 | 875 | if (IS_ERR(req)) |
4cb2ea28 | 876 | return PTR_ERR(req); |
f9596695 | 877 | |
95bfec41 | 878 | vbr = blk_mq_rq_to_pdu(req); |
f1ba4e67 | 879 | vbr->in_hdr_len = sizeof(vbr->in_hdr.status); |
95bfec41 DF |
880 | vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID); |
881 | vbr->out_hdr.sector = 0; | |
882 | ||
f9596695 CH |
883 | err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL); |
884 | if (err) | |
885 | goto out; | |
886 | ||
b84ba30b | 887 | blk_execute_rq(req, false); |
f1ba4e67 | 888 | err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status)); |
f9596695 | 889 | out: |
0bf6d96c | 890 | blk_mq_free_request(req); |
e4c4776d | 891 | return err; |
4cb2ea28 | 892 | } |
893 | ||
135da0b0 CB |
894 | /* We provide getgeo only to please some old bootloader/partitioning tools */ |
895 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
896 | { | |
48e4043d | 897 | struct virtio_blk *vblk = bd->bd_disk->private_data; |
90b5feb8 SH |
898 | int ret = 0; |
899 | ||
900 | mutex_lock(&vblk->vdev_mutex); | |
901 | ||
902 | if (!vblk->vdev) { | |
903 | ret = -ENXIO; | |
904 | goto out; | |
905 | } | |
48e4043d RH |
906 | |
907 | /* see if the host passed in geometry config */ | |
855e0c52 RR |
908 | if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) { |
909 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
910 | geometry.cylinders, &geo->cylinders); | |
911 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
912 | geometry.heads, &geo->heads); | |
913 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
914 | geometry.sectors, &geo->sectors); | |
48e4043d RH |
915 | } else { |
916 | /* some standard values, similar to sd */ | |
917 | geo->heads = 1 << 6; | |
918 | geo->sectors = 1 << 5; | |
919 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
920 | } | |
90b5feb8 SH |
921 | out: |
922 | mutex_unlock(&vblk->vdev_mutex); | |
923 | return ret; | |
135da0b0 CB |
924 | } |
925 | ||
24b45e6c CH |
926 | static void virtblk_free_disk(struct gendisk *disk) |
927 | { | |
928 | struct virtio_blk *vblk = disk->private_data; | |
929 | ||
92a34c46 | 930 | ida_free(&vd_index_ida, vblk->index); |
24b45e6c CH |
931 | mutex_destroy(&vblk->vdev_mutex); |
932 | kfree(vblk); | |
933 | } | |
934 | ||
83d5cde4 | 935 | static const struct block_device_operations virtblk_fops = { |
24b45e6c CH |
936 | .owner = THIS_MODULE, |
937 | .getgeo = virtblk_getgeo, | |
938 | .free_disk = virtblk_free_disk, | |
95bfec41 | 939 | .report_zones = virtblk_report_zones, |
e467cde2 RR |
940 | }; |
941 | ||
d50ed907 CB |
942 | static int index_to_minor(int index) |
943 | { | |
944 | return index << PART_BITS; | |
945 | } | |
946 | ||
5087a50e MT |
947 | static int minor_to_index(int minor) |
948 | { | |
949 | return minor >> PART_BITS; | |
950 | } | |
951 | ||
e982c4d0 HR |
952 | static ssize_t serial_show(struct device *dev, |
953 | struct device_attribute *attr, char *buf) | |
a5eb9e4f RH |
954 | { |
955 | struct gendisk *disk = dev_to_disk(dev); | |
956 | int err; | |
957 | ||
958 | /* sysfs gives us a PAGE_SIZE buffer */ | |
959 | BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); | |
960 | ||
961 | buf[VIRTIO_BLK_ID_BYTES] = '\0'; | |
962 | err = virtblk_get_id(disk, buf); | |
963 | if (!err) | |
964 | return strlen(buf); | |
965 | ||
966 | if (err == -EIO) /* Unsupported? Make it empty. */ | |
967 | return 0; | |
968 | ||
969 | return err; | |
970 | } | |
393c525b | 971 | |
e982c4d0 | 972 | static DEVICE_ATTR_RO(serial); |
a5eb9e4f | 973 | |
daf2a501 SH |
974 | /* The queue's logical block size must be set before calling this */ |
975 | static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize) | |
7a7c924c | 976 | { |
7a7c924c CH |
977 | struct virtio_device *vdev = vblk->vdev; |
978 | struct request_queue *q = vblk->disk->queue; | |
979 | char cap_str_2[10], cap_str_10[10]; | |
1046d304 | 980 | unsigned long long nblocks; |
b9f28d86 | 981 | u64 capacity; |
7a7c924c CH |
982 | |
983 | /* Host must always specify the capacity. */ | |
855e0c52 | 984 | virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity); |
7a7c924c | 985 | |
1046d304 SH |
986 | nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9); |
987 | ||
988 | string_get_size(nblocks, queue_logical_block_size(q), | |
b9f28d86 | 989 | STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); |
1046d304 | 990 | string_get_size(nblocks, queue_logical_block_size(q), |
b9f28d86 | 991 | STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); |
7a7c924c CH |
992 | |
993 | dev_notice(&vdev->dev, | |
daf2a501 SH |
994 | "[%s] %s%llu %d-byte logical blocks (%s/%s)\n", |
995 | vblk->disk->disk_name, | |
996 | resize ? "new size: " : "", | |
1046d304 SH |
997 | nblocks, |
998 | queue_logical_block_size(q), | |
999 | cap_str_10, | |
1000 | cap_str_2); | |
7a7c924c | 1001 | |
449f4ec9 | 1002 | set_capacity_and_notify(vblk->disk, capacity); |
daf2a501 SH |
1003 | } |
1004 | ||
1005 | static void virtblk_config_changed_work(struct work_struct *work) | |
1006 | { | |
1007 | struct virtio_blk *vblk = | |
1008 | container_of(work, struct virtio_blk, config_work); | |
daf2a501 | 1009 | |
95bfec41 | 1010 | virtblk_revalidate_zones(vblk); |
daf2a501 | 1011 | virtblk_update_capacity(vblk, true); |
7a7c924c CH |
1012 | } |
1013 | ||
1014 | static void virtblk_config_changed(struct virtio_device *vdev) | |
1015 | { | |
1016 | struct virtio_blk *vblk = vdev->priv; | |
1017 | ||
1018 | queue_work(virtblk_wq, &vblk->config_work); | |
1019 | } | |
1020 | ||
6abd6e5a AS |
1021 | static int init_vq(struct virtio_blk *vblk) |
1022 | { | |
2ff98449 | 1023 | int err; |
6a27b656 ML |
1024 | int i; |
1025 | vq_callback_t **callbacks; | |
1026 | const char **names; | |
1027 | struct virtqueue **vqs; | |
1028 | unsigned short num_vqs; | |
4e040052 | 1029 | unsigned int num_poll_vqs; |
6a27b656 | 1030 | struct virtio_device *vdev = vblk->vdev; |
ad71473d | 1031 | struct irq_affinity desc = { 0, }; |
6a27b656 ML |
1032 | |
1033 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ, | |
1034 | struct virtio_blk_config, num_queues, | |
1035 | &num_vqs); | |
1036 | if (err) | |
1037 | num_vqs = 1; | |
4e040052 | 1038 | |
6ae6ff6f | 1039 | if (!err && !num_vqs) { |
63b4ffa4 | 1040 | dev_err(&vdev->dev, "MQ advertised but zero queues reported\n"); |
6ae6ff6f JW |
1041 | return -EINVAL; |
1042 | } | |
6a27b656 | 1043 | |
0989c41b MG |
1044 | num_vqs = min_t(unsigned int, |
1045 | min_not_zero(num_request_queues, nr_cpu_ids), | |
1046 | num_vqs); | |
bf348f9b | 1047 | |
4e040052 SK |
1048 | num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1); |
1049 | ||
1050 | vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs; | |
1051 | vblk->io_queues[HCTX_TYPE_READ] = 0; | |
1052 | vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs; | |
1053 | ||
1054 | dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n", | |
1055 | vblk->io_queues[HCTX_TYPE_DEFAULT], | |
1056 | vblk->io_queues[HCTX_TYPE_READ], | |
1057 | vblk->io_queues[HCTX_TYPE_POLL]); | |
1058 | ||
668866b6 | 1059 | vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL); |
347a5293 MH |
1060 | if (!vblk->vqs) |
1061 | return -ENOMEM; | |
6a27b656 | 1062 | |
668866b6 ME |
1063 | names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL); |
1064 | callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL); | |
1065 | vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL); | |
347a5293 MH |
1066 | if (!names || !callbacks || !vqs) { |
1067 | err = -ENOMEM; | |
1068 | goto out; | |
1069 | } | |
6abd6e5a | 1070 | |
4e040052 | 1071 | for (i = 0; i < num_vqs - num_poll_vqs; i++) { |
6a27b656 ML |
1072 | callbacks[i] = virtblk_done; |
1073 | snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i); | |
1074 | names[i] = vblk->vqs[i].name; | |
1075 | } | |
1076 | ||
4e040052 SK |
1077 | for (; i < num_vqs; i++) { |
1078 | callbacks[i] = NULL; | |
1079 | snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i); | |
1080 | names[i] = vblk->vqs[i].name; | |
1081 | } | |
1082 | ||
6a27b656 | 1083 | /* Discover virtqueues and write information to configuration. */ |
9b2bbdb2 | 1084 | err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc); |
6a27b656 | 1085 | if (err) |
347a5293 | 1086 | goto out; |
6abd6e5a | 1087 | |
6a27b656 ML |
1088 | for (i = 0; i < num_vqs; i++) { |
1089 | spin_lock_init(&vblk->vqs[i].lock); | |
1090 | vblk->vqs[i].vq = vqs[i]; | |
1091 | } | |
1092 | vblk->num_vqs = num_vqs; | |
1093 | ||
347a5293 | 1094 | out: |
6a27b656 | 1095 | kfree(vqs); |
6a27b656 | 1096 | kfree(callbacks); |
6a27b656 | 1097 | kfree(names); |
6a27b656 ML |
1098 | if (err) |
1099 | kfree(vblk->vqs); | |
6abd6e5a AS |
1100 | return err; |
1101 | } | |
1102 | ||
c0aa3e09 RM |
1103 | /* |
1104 | * Legacy naming scheme used for virtio devices. We are stuck with it for | |
1105 | * virtio blk but don't ever use it for any new driver. | |
1106 | */ | |
1107 | static int virtblk_name_format(char *prefix, int index, char *buf, int buflen) | |
1108 | { | |
1109 | const int base = 'z' - 'a' + 1; | |
1110 | char *begin = buf + strlen(prefix); | |
1111 | char *end = buf + buflen; | |
1112 | char *p; | |
1113 | int unit; | |
1114 | ||
1115 | p = end - 1; | |
1116 | *p = '\0'; | |
1117 | unit = base; | |
1118 | do { | |
1119 | if (p == begin) | |
1120 | return -EINVAL; | |
1121 | *--p = 'a' + (index % unit); | |
1122 | index = (index / unit) - 1; | |
1123 | } while (index >= 0); | |
1124 | ||
1125 | memmove(begin, p, end - p); | |
1126 | memcpy(buf, prefix, strlen(prefix)); | |
1127 | ||
1128 | return 0; | |
1129 | } | |
1130 | ||
cd5d5038 PB |
1131 | static int virtblk_get_cache_mode(struct virtio_device *vdev) |
1132 | { | |
1133 | u8 writeback; | |
1134 | int err; | |
1135 | ||
855e0c52 RR |
1136 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE, |
1137 | struct virtio_blk_config, wce, | |
1138 | &writeback); | |
592002f5 MT |
1139 | |
1140 | /* | |
1141 | * If WCE is not configurable and flush is not available, | |
1142 | * assume no writeback cache is in use. | |
1143 | */ | |
cd5d5038 | 1144 | if (err) |
592002f5 | 1145 | writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH); |
cd5d5038 PB |
1146 | |
1147 | return writeback; | |
1148 | } | |
1149 | ||
1150 | static void virtblk_update_cache_mode(struct virtio_device *vdev) | |
1151 | { | |
1152 | u8 writeback = virtblk_get_cache_mode(vdev); | |
1153 | struct virtio_blk *vblk = vdev->priv; | |
1154 | ||
ad9126ac | 1155 | blk_queue_write_cache(vblk->disk->queue, writeback, false); |
cd5d5038 PB |
1156 | } |
1157 | ||
1158 | static const char *const virtblk_cache_types[] = { | |
1159 | "write through", "write back" | |
1160 | }; | |
1161 | ||
1162 | static ssize_t | |
e982c4d0 HR |
1163 | cache_type_store(struct device *dev, struct device_attribute *attr, |
1164 | const char *buf, size_t count) | |
cd5d5038 PB |
1165 | { |
1166 | struct gendisk *disk = dev_to_disk(dev); | |
1167 | struct virtio_blk *vblk = disk->private_data; | |
1168 | struct virtio_device *vdev = vblk->vdev; | |
1169 | int i; | |
cd5d5038 PB |
1170 | |
1171 | BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE)); | |
f53d5aa0 | 1172 | i = sysfs_match_string(virtblk_cache_types, buf); |
cd5d5038 | 1173 | if (i < 0) |
f53d5aa0 | 1174 | return i; |
cd5d5038 | 1175 | |
855e0c52 | 1176 | virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i); |
cd5d5038 PB |
1177 | virtblk_update_cache_mode(vdev); |
1178 | return count; | |
1179 | } | |
1180 | ||
1181 | static ssize_t | |
e982c4d0 | 1182 | cache_type_show(struct device *dev, struct device_attribute *attr, char *buf) |
cd5d5038 PB |
1183 | { |
1184 | struct gendisk *disk = dev_to_disk(dev); | |
1185 | struct virtio_blk *vblk = disk->private_data; | |
1186 | u8 writeback = virtblk_get_cache_mode(vblk->vdev); | |
1187 | ||
1188 | BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types)); | |
f1aa12f5 | 1189 | return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]); |
cd5d5038 PB |
1190 | } |
1191 | ||
e982c4d0 HR |
1192 | static DEVICE_ATTR_RW(cache_type); |
1193 | ||
1194 | static struct attribute *virtblk_attrs[] = { | |
1195 | &dev_attr_serial.attr, | |
1196 | &dev_attr_cache_type.attr, | |
1197 | NULL, | |
1198 | }; | |
1199 | ||
1200 | static umode_t virtblk_attrs_are_visible(struct kobject *kobj, | |
1201 | struct attribute *a, int n) | |
1202 | { | |
4ce79063 | 1203 | struct device *dev = kobj_to_dev(kobj); |
e982c4d0 HR |
1204 | struct gendisk *disk = dev_to_disk(dev); |
1205 | struct virtio_blk *vblk = disk->private_data; | |
1206 | struct virtio_device *vdev = vblk->vdev; | |
1207 | ||
1208 | if (a == &dev_attr_cache_type.attr && | |
1209 | !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) | |
1210 | return S_IRUGO; | |
1211 | ||
1212 | return a->mode; | |
1213 | } | |
1214 | ||
1215 | static const struct attribute_group virtblk_attr_group = { | |
1216 | .attrs = virtblk_attrs, | |
1217 | .is_visible = virtblk_attrs_are_visible, | |
1218 | }; | |
1219 | ||
1220 | static const struct attribute_group *virtblk_attr_groups[] = { | |
1221 | &virtblk_attr_group, | |
1222 | NULL, | |
1223 | }; | |
cd5d5038 | 1224 | |
a4e1d0b7 | 1225 | static void virtblk_map_queues(struct blk_mq_tag_set *set) |
ad71473d CH |
1226 | { |
1227 | struct virtio_blk *vblk = set->driver_data; | |
4e040052 SK |
1228 | int i, qoff; |
1229 | ||
1230 | for (i = 0, qoff = 0; i < set->nr_maps; i++) { | |
1231 | struct blk_mq_queue_map *map = &set->map[i]; | |
1232 | ||
1233 | map->nr_queues = vblk->io_queues[i]; | |
1234 | map->queue_offset = qoff; | |
1235 | qoff += map->nr_queues; | |
1236 | ||
1237 | if (map->nr_queues == 0) | |
1238 | continue; | |
1239 | ||
1240 | /* | |
1241 | * Regular queues have interrupts and hence CPU affinity is | |
1242 | * defined by the core virtio code, but polling queues have | |
1243 | * no interrupts so we let the block layer assign CPU affinity. | |
1244 | */ | |
1245 | if (i == HCTX_TYPE_POLL) | |
1246 | blk_mq_map_queues(&set->map[i]); | |
1247 | else | |
1248 | blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0); | |
1249 | } | |
4e040052 SK |
1250 | } |
1251 | ||
afd384f0 MT |
1252 | static void virtblk_complete_batch(struct io_comp_batch *iob) |
1253 | { | |
1254 | struct request *req; | |
1255 | ||
1256 | rq_list_for_each(&iob->req_list, req) { | |
1257 | virtblk_unmap_data(req, blk_mq_rq_to_pdu(req)); | |
1258 | virtblk_cleanup_cmd(req); | |
1259 | } | |
1260 | blk_mq_end_request_batch(iob); | |
1261 | } | |
1262 | ||
4e040052 SK |
1263 | static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob) |
1264 | { | |
1265 | struct virtio_blk *vblk = hctx->queue->queuedata; | |
8d12ec10 | 1266 | struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx); |
afd384f0 | 1267 | struct virtblk_req *vbr; |
4e040052 | 1268 | unsigned long flags; |
afd384f0 | 1269 | unsigned int len; |
4e040052 SK |
1270 | int found = 0; |
1271 | ||
1272 | spin_lock_irqsave(&vq->lock, flags); | |
afd384f0 MT |
1273 | |
1274 | while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) { | |
1275 | struct request *req = blk_mq_rq_from_pdu(vbr); | |
1276 | ||
1277 | found++; | |
1278 | if (!blk_mq_complete_request_remote(req) && | |
1279 | !blk_mq_add_to_batch(req, iob, virtblk_vbr_status(vbr), | |
1280 | virtblk_complete_batch)) | |
1281 | virtblk_request_done(req); | |
1282 | } | |
4e040052 SK |
1283 | |
1284 | if (found) | |
1285 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); | |
1286 | ||
1287 | spin_unlock_irqrestore(&vq->lock, flags); | |
1288 | ||
1289 | return found; | |
1290 | } | |
1291 | ||
f363b089 | 1292 | static const struct blk_mq_ops virtio_mq_ops = { |
1cf7e9c6 | 1293 | .queue_rq = virtio_queue_rq, |
0e9911fa | 1294 | .queue_rqs = virtio_queue_rqs, |
944e7c87 | 1295 | .commit_rqs = virtio_commit_rqs, |
5124c285 | 1296 | .complete = virtblk_request_done, |
ad71473d | 1297 | .map_queues = virtblk_map_queues, |
4e040052 | 1298 | .poll = virtblk_poll, |
1cf7e9c6 JA |
1299 | }; |
1300 | ||
24d2f903 CH |
1301 | static unsigned int virtblk_queue_depth; |
1302 | module_param_named(queue_depth, virtblk_queue_depth, uint, 0444); | |
1cf7e9c6 | 1303 | |
8d85fce7 | 1304 | static int virtblk_probe(struct virtio_device *vdev) |
e467cde2 RR |
1305 | { |
1306 | struct virtio_blk *vblk; | |
69740c8b | 1307 | struct request_queue *q; |
5087a50e | 1308 | int err, index; |
a98755c5 | 1309 | |
fd1068e1 | 1310 | u32 v, blk_size, max_size, sg_elems, opt_io_size; |
e60d6407 AK |
1311 | u32 max_discard_segs = 0; |
1312 | u32 discard_granularity = 0; | |
69740c8b CH |
1313 | u16 min_io_size; |
1314 | u8 physical_block_exp, alignment_offset; | |
d1e9aa9c | 1315 | unsigned int queue_depth; |
e467cde2 | 1316 | |
ff631988 MT |
1317 | if (!vdev->config->get) { |
1318 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
1319 | __func__); | |
1320 | return -EINVAL; | |
1321 | } | |
1322 | ||
92a34c46 PR |
1323 | err = ida_alloc_range(&vd_index_ida, 0, |
1324 | minor_to_index(1 << MINORBITS) - 1, GFP_KERNEL); | |
5087a50e MT |
1325 | if (err < 0) |
1326 | goto out; | |
1327 | index = err; | |
4f3bf19c | 1328 | |
0864b79a | 1329 | /* We need to know how many segments before we allocate. */ |
855e0c52 RR |
1330 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX, |
1331 | struct virtio_blk_config, seg_max, | |
1332 | &sg_elems); | |
a5b365a6 CH |
1333 | |
1334 | /* We need at least one SG element, whatever they say. */ | |
1335 | if (err || !sg_elems) | |
0864b79a RR |
1336 | sg_elems = 1; |
1337 | ||
63947b34 SH |
1338 | /* Prevent integer overflows and honor max vq size */ |
1339 | sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2); | |
1340 | ||
1cf7e9c6 | 1341 | vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); |
e467cde2 RR |
1342 | if (!vblk) { |
1343 | err = -ENOMEM; | |
5087a50e | 1344 | goto out_free_index; |
e467cde2 RR |
1345 | } |
1346 | ||
90b5feb8 SH |
1347 | mutex_init(&vblk->vdev_mutex); |
1348 | ||
e467cde2 | 1349 | vblk->vdev = vdev; |
a98755c5 | 1350 | |
7a7c924c | 1351 | INIT_WORK(&vblk->config_work, virtblk_config_changed_work); |
e467cde2 | 1352 | |
6abd6e5a AS |
1353 | err = init_vq(vblk); |
1354 | if (err) | |
e467cde2 | 1355 | goto out_free_vblk; |
e467cde2 | 1356 | |
fc4324b4 | 1357 | /* Default queue sizing is to fill the ring. */ |
6105d1fe | 1358 | if (!virtblk_queue_depth) { |
d1e9aa9c | 1359 | queue_depth = vblk->vqs[0].vq->num_free; |
fc4324b4 RR |
1360 | /* ... but without indirect descs, we use 2 descs per req */ |
1361 | if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC)) | |
d1e9aa9c JQ |
1362 | queue_depth /= 2; |
1363 | } else { | |
1364 | queue_depth = virtblk_queue_depth; | |
fc4324b4 | 1365 | } |
24d2f903 CH |
1366 | |
1367 | memset(&vblk->tag_set, 0, sizeof(vblk->tag_set)); | |
1368 | vblk->tag_set.ops = &virtio_mq_ops; | |
d1e9aa9c | 1369 | vblk->tag_set.queue_depth = queue_depth; |
24d2f903 CH |
1370 | vblk->tag_set.numa_node = NUMA_NO_NODE; |
1371 | vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; | |
1372 | vblk->tag_set.cmd_size = | |
1cf7e9c6 | 1373 | sizeof(struct virtblk_req) + |
02746e26 | 1374 | sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT; |
24d2f903 | 1375 | vblk->tag_set.driver_data = vblk; |
6a27b656 | 1376 | vblk->tag_set.nr_hw_queues = vblk->num_vqs; |
4e040052 SK |
1377 | vblk->tag_set.nr_maps = 1; |
1378 | if (vblk->io_queues[HCTX_TYPE_POLL]) | |
1379 | vblk->tag_set.nr_maps = 3; | |
1cf7e9c6 | 1380 | |
24d2f903 CH |
1381 | err = blk_mq_alloc_tag_set(&vblk->tag_set); |
1382 | if (err) | |
89a5f065 | 1383 | goto out_free_vq; |
24d2f903 | 1384 | |
89a5f065 CH |
1385 | vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk); |
1386 | if (IS_ERR(vblk->disk)) { | |
1387 | err = PTR_ERR(vblk->disk); | |
24d2f903 | 1388 | goto out_free_tags; |
e467cde2 | 1389 | } |
89a5f065 | 1390 | q = vblk->disk->queue; |
7d116b62 | 1391 | |
c0aa3e09 | 1392 | virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN); |
d50ed907 | 1393 | |
e467cde2 | 1394 | vblk->disk->major = major; |
d50ed907 | 1395 | vblk->disk->first_minor = index_to_minor(index); |
89a5f065 | 1396 | vblk->disk->minors = 1 << PART_BITS; |
e467cde2 RR |
1397 | vblk->disk->private_data = vblk; |
1398 | vblk->disk->fops = &virtblk_fops; | |
5087a50e | 1399 | vblk->index = index; |
4f3bf19c | 1400 | |
02c42b7a | 1401 | /* configure queue flush support */ |
cd5d5038 | 1402 | virtblk_update_cache_mode(vdev); |
e467cde2 | 1403 | |
3ef53609 CB |
1404 | /* If disk is read-only in the host, the guest should obey */ |
1405 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
1406 | set_disk_ro(vblk->disk, 1); | |
1407 | ||
0864b79a | 1408 | /* We can handle whatever the host told us to handle. */ |
e030759a | 1409 | blk_queue_max_segments(q, sg_elems); |
0864b79a | 1410 | |
4b7f7e20 | 1411 | /* No real sector limit. */ |
f4e468f7 | 1412 | blk_queue_max_hw_sectors(q, UINT_MAX); |
4b7f7e20 | 1413 | |
fd1068e1 JR |
1414 | max_size = virtio_max_dma_size(vdev); |
1415 | ||
a586d4f6 RR |
1416 | /* Host can optionally specify maximum segment size and number of |
1417 | * segments. */ | |
855e0c52 RR |
1418 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX, |
1419 | struct virtio_blk_config, size_max, &v); | |
e467cde2 | 1420 | if (!err) |
fd1068e1 JR |
1421 | max_size = min(max_size, v); |
1422 | ||
1423 | blk_queue_max_segment_size(q, max_size); | |
e467cde2 | 1424 | |
066f4d82 | 1425 | /* Host can optionally specify the block size of the device */ |
855e0c52 RR |
1426 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE, |
1427 | struct virtio_blk_config, blk_size, | |
1428 | &blk_size); | |
57a13a5b XY |
1429 | if (!err) { |
1430 | err = blk_validate_block_size(blk_size); | |
1431 | if (err) { | |
1432 | dev_err(&vdev->dev, | |
1433 | "virtio_blk: invalid block size: 0x%x\n", | |
1434 | blk_size); | |
1435 | goto out_cleanup_disk; | |
1436 | } | |
1437 | ||
69740c8b | 1438 | blk_queue_logical_block_size(q, blk_size); |
57a13a5b | 1439 | } else |
69740c8b CH |
1440 | blk_size = queue_logical_block_size(q); |
1441 | ||
1442 | /* Use topology information if available */ | |
855e0c52 RR |
1443 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
1444 | struct virtio_blk_config, physical_block_exp, | |
1445 | &physical_block_exp); | |
69740c8b CH |
1446 | if (!err && physical_block_exp) |
1447 | blk_queue_physical_block_size(q, | |
1448 | blk_size * (1 << physical_block_exp)); | |
1449 | ||
855e0c52 RR |
1450 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
1451 | struct virtio_blk_config, alignment_offset, | |
1452 | &alignment_offset); | |
69740c8b CH |
1453 | if (!err && alignment_offset) |
1454 | blk_queue_alignment_offset(q, blk_size * alignment_offset); | |
1455 | ||
855e0c52 RR |
1456 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
1457 | struct virtio_blk_config, min_io_size, | |
1458 | &min_io_size); | |
69740c8b CH |
1459 | if (!err && min_io_size) |
1460 | blk_queue_io_min(q, blk_size * min_io_size); | |
1461 | ||
855e0c52 RR |
1462 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
1463 | struct virtio_blk_config, opt_io_size, | |
1464 | &opt_io_size); | |
69740c8b CH |
1465 | if (!err && opt_io_size) |
1466 | blk_queue_io_opt(q, blk_size * opt_io_size); | |
1467 | ||
1f23816b | 1468 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) { |
1f23816b | 1469 | virtio_cread(vdev, struct virtio_blk_config, |
e60d6407 | 1470 | discard_sector_alignment, &discard_granularity); |
1f23816b CL |
1471 | |
1472 | virtio_cread(vdev, struct virtio_blk_config, | |
1473 | max_discard_sectors, &v); | |
1474 | blk_queue_max_discard_sectors(q, v ? v : UINT_MAX); | |
1475 | ||
1476 | virtio_cread(vdev, struct virtio_blk_config, max_discard_seg, | |
e60d6407 | 1477 | &max_discard_segs); |
1f23816b CL |
1478 | } |
1479 | ||
1480 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) { | |
1481 | virtio_cread(vdev, struct virtio_blk_config, | |
1482 | max_write_zeroes_sectors, &v); | |
1483 | blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX); | |
1484 | } | |
1485 | ||
e60d6407 AK |
1486 | /* The discard and secure erase limits are combined since the Linux |
1487 | * block layer uses the same limit for both commands. | |
1488 | * | |
1489 | * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features | |
1490 | * are negotiated, we will use the minimum between the limits. | |
1491 | * | |
1492 | * discard sector alignment is set to the minimum between discard_sector_alignment | |
1493 | * and secure_erase_sector_alignment. | |
1494 | * | |
1495 | * max discard sectors is set to the minimum between max_discard_seg and | |
1496 | * max_secure_erase_seg. | |
1497 | */ | |
1498 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) { | |
1499 | ||
1500 | virtio_cread(vdev, struct virtio_blk_config, | |
1501 | secure_erase_sector_alignment, &v); | |
1502 | ||
1503 | /* secure_erase_sector_alignment should not be zero, the device should set a | |
1504 | * valid number of sectors. | |
1505 | */ | |
1506 | if (!v) { | |
1507 | dev_err(&vdev->dev, | |
1508 | "virtio_blk: secure_erase_sector_alignment can't be 0\n"); | |
1509 | err = -EINVAL; | |
1510 | goto out_cleanup_disk; | |
1511 | } | |
1512 | ||
1513 | discard_granularity = min_not_zero(discard_granularity, v); | |
1514 | ||
1515 | virtio_cread(vdev, struct virtio_blk_config, | |
1516 | max_secure_erase_sectors, &v); | |
1517 | ||
1518 | /* max_secure_erase_sectors should not be zero, the device should set a | |
1519 | * valid number of sectors. | |
1520 | */ | |
1521 | if (!v) { | |
1522 | dev_err(&vdev->dev, | |
1523 | "virtio_blk: max_secure_erase_sectors can't be 0\n"); | |
1524 | err = -EINVAL; | |
1525 | goto out_cleanup_disk; | |
1526 | } | |
1527 | ||
1528 | blk_queue_max_secure_erase_sectors(q, v); | |
1529 | ||
1530 | virtio_cread(vdev, struct virtio_blk_config, | |
1531 | max_secure_erase_seg, &v); | |
1532 | ||
1533 | /* max_secure_erase_seg should not be zero, the device should set a | |
1534 | * valid number of segments | |
1535 | */ | |
1536 | if (!v) { | |
1537 | dev_err(&vdev->dev, | |
1538 | "virtio_blk: max_secure_erase_seg can't be 0\n"); | |
1539 | err = -EINVAL; | |
1540 | goto out_cleanup_disk; | |
1541 | } | |
1542 | ||
1543 | max_discard_segs = min_not_zero(max_discard_segs, v); | |
1544 | } | |
1545 | ||
1546 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) || | |
1547 | virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) { | |
1548 | /* max_discard_seg and discard_granularity will be 0 only | |
1549 | * if max_discard_seg and discard_sector_alignment fields in the virtio | |
1550 | * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated. | |
1551 | * In this case, we use default values. | |
1552 | */ | |
1553 | if (!max_discard_segs) | |
1554 | max_discard_segs = sg_elems; | |
1555 | ||
1556 | blk_queue_max_discard_segments(q, | |
1557 | min(max_discard_segs, MAX_DISCARD_SEGMENTS)); | |
1558 | ||
1559 | if (discard_granularity) | |
1560 | q->limits.discard_granularity = discard_granularity << SECTOR_SHIFT; | |
1561 | else | |
1562 | q->limits.discard_granularity = blk_size; | |
1563 | } | |
1564 | ||
daf2a501 | 1565 | virtblk_update_capacity(vblk, false); |
7a11370e MT |
1566 | virtio_device_ready(vdev); |
1567 | ||
10805eb5 DF |
1568 | /* |
1569 | * All steps that follow use the VQs therefore they need to be | |
1570 | * placed after the virtio_device_ready() call above. | |
1571 | */ | |
1572 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED)) { | |
95bfec41 DF |
1573 | err = virtblk_probe_zoned_device(vdev, vblk, q); |
1574 | if (err) | |
1575 | goto out_cleanup_disk; | |
1576 | } | |
1577 | ||
dbb301f9 LC |
1578 | err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups); |
1579 | if (err) | |
1580 | goto out_cleanup_disk; | |
1581 | ||
e467cde2 RR |
1582 | return 0; |
1583 | ||
dbb301f9 | 1584 | out_cleanup_disk: |
8b9ab626 | 1585 | put_disk(vblk->disk); |
24d2f903 CH |
1586 | out_free_tags: |
1587 | blk_mq_free_tag_set(&vblk->tag_set); | |
e467cde2 | 1588 | out_free_vq: |
d2a7ddda | 1589 | vdev->config->del_vqs(vdev); |
e7eea44e | 1590 | kfree(vblk->vqs); |
e467cde2 RR |
1591 | out_free_vblk: |
1592 | kfree(vblk); | |
5087a50e | 1593 | out_free_index: |
92a34c46 | 1594 | ida_free(&vd_index_ida, index); |
e467cde2 RR |
1595 | out: |
1596 | return err; | |
1597 | } | |
1598 | ||
8d85fce7 | 1599 | static void virtblk_remove(struct virtio_device *vdev) |
e467cde2 RR |
1600 | { |
1601 | struct virtio_blk *vblk = vdev->priv; | |
e467cde2 | 1602 | |
cc74f719 MT |
1603 | /* Make sure no work handler is accessing the device. */ |
1604 | flush_work(&vblk->config_work); | |
7a7c924c | 1605 | |
02e2b124 | 1606 | del_gendisk(vblk->disk); |
24d2f903 CH |
1607 | blk_mq_free_tag_set(&vblk->tag_set); |
1608 | ||
90b5feb8 SH |
1609 | mutex_lock(&vblk->vdev_mutex); |
1610 | ||
6e5aa7ef | 1611 | /* Stop all the virtqueues. */ |
d9679d00 | 1612 | virtio_reset_device(vdev); |
6e5aa7ef | 1613 | |
90b5feb8 SH |
1614 | /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */ |
1615 | vblk->vdev = NULL; | |
1616 | ||
d2a7ddda | 1617 | vdev->config->del_vqs(vdev); |
6a27b656 | 1618 | kfree(vblk->vqs); |
f4953fe6 | 1619 | |
90b5feb8 SH |
1620 | mutex_unlock(&vblk->vdev_mutex); |
1621 | ||
24b45e6c | 1622 | put_disk(vblk->disk); |
e467cde2 RR |
1623 | } |
1624 | ||
89107000 | 1625 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
1626 | static int virtblk_freeze(struct virtio_device *vdev) |
1627 | { | |
1628 | struct virtio_blk *vblk = vdev->priv; | |
1629 | ||
1630 | /* Ensure we don't receive any more interrupts */ | |
d9679d00 | 1631 | virtio_reset_device(vdev); |
f8fb5bc2 | 1632 | |
cc74f719 | 1633 | /* Make sure no work handler is accessing the device. */ |
f8fb5bc2 AS |
1634 | flush_work(&vblk->config_work); |
1635 | ||
9b3e9905 | 1636 | blk_mq_quiesce_queue(vblk->disk->queue); |
f8fb5bc2 AS |
1637 | |
1638 | vdev->config->del_vqs(vdev); | |
b71ba22e XY |
1639 | kfree(vblk->vqs); |
1640 | ||
f8fb5bc2 AS |
1641 | return 0; |
1642 | } | |
1643 | ||
1644 | static int virtblk_restore(struct virtio_device *vdev) | |
1645 | { | |
1646 | struct virtio_blk *vblk = vdev->priv; | |
1647 | int ret; | |
1648 | ||
f8fb5bc2 | 1649 | ret = init_vq(vdev->priv); |
6d62c37f MT |
1650 | if (ret) |
1651 | return ret; | |
1652 | ||
1653 | virtio_device_ready(vdev); | |
1cf7e9c6 | 1654 | |
9b3e9905 | 1655 | blk_mq_unquiesce_queue(vblk->disk->queue); |
6d62c37f | 1656 | return 0; |
f8fb5bc2 AS |
1657 | } |
1658 | #endif | |
1659 | ||
47483e25 | 1660 | static const struct virtio_device_id id_table[] = { |
e467cde2 RR |
1661 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, |
1662 | { 0 }, | |
1663 | }; | |
1664 | ||
19c1c5a6 | 1665 | static unsigned int features_legacy[] = { |
02c42b7a | 1666 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, |
97b50a65 | 1667 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, |
592002f5 | 1668 | VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE, |
1f23816b | 1669 | VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES, |
e60d6407 | 1670 | VIRTIO_BLK_F_SECURE_ERASE, |
19c1c5a6 MT |
1671 | } |
1672 | ; | |
1673 | static unsigned int features[] = { | |
1674 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, | |
1675 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, | |
592002f5 | 1676 | VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE, |
1f23816b | 1677 | VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES, |
10805eb5 | 1678 | VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED, |
c45a6816 RR |
1679 | }; |
1680 | ||
8d85fce7 | 1681 | static struct virtio_driver virtio_blk = { |
19c1c5a6 MT |
1682 | .feature_table = features, |
1683 | .feature_table_size = ARRAY_SIZE(features), | |
1684 | .feature_table_legacy = features_legacy, | |
1685 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
1686 | .driver.name = KBUILD_MODNAME, | |
1687 | .driver.owner = THIS_MODULE, | |
1688 | .id_table = id_table, | |
1689 | .probe = virtblk_probe, | |
1690 | .remove = virtblk_remove, | |
1691 | .config_changed = virtblk_config_changed, | |
89107000 | 1692 | #ifdef CONFIG_PM_SLEEP |
19c1c5a6 MT |
1693 | .freeze = virtblk_freeze, |
1694 | .restore = virtblk_restore, | |
f8fb5bc2 | 1695 | #endif |
e467cde2 RR |
1696 | }; |
1697 | ||
bcfe9b6c | 1698 | static int __init virtio_blk_init(void) |
e467cde2 | 1699 | { |
7a7c924c CH |
1700 | int error; |
1701 | ||
1702 | virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); | |
1703 | if (!virtblk_wq) | |
1704 | return -ENOMEM; | |
1705 | ||
4f3bf19c | 1706 | major = register_blkdev(0, "virtblk"); |
7a7c924c CH |
1707 | if (major < 0) { |
1708 | error = major; | |
1709 | goto out_destroy_workqueue; | |
1710 | } | |
1711 | ||
1712 | error = register_virtio_driver(&virtio_blk); | |
1713 | if (error) | |
1714 | goto out_unregister_blkdev; | |
1715 | return 0; | |
1716 | ||
1717 | out_unregister_blkdev: | |
1718 | unregister_blkdev(major, "virtblk"); | |
1719 | out_destroy_workqueue: | |
1720 | destroy_workqueue(virtblk_wq); | |
1721 | return error; | |
e467cde2 RR |
1722 | } |
1723 | ||
bcfe9b6c | 1724 | static void __exit virtio_blk_fini(void) |
e467cde2 RR |
1725 | { |
1726 | unregister_virtio_driver(&virtio_blk); | |
38f37b57 | 1727 | unregister_blkdev(major, "virtblk"); |
7a7c924c | 1728 | destroy_workqueue(virtblk_wq); |
e467cde2 | 1729 | } |
bcfe9b6c RD |
1730 | module_init(virtio_blk_init); |
1731 | module_exit(virtio_blk_fini); | |
e467cde2 RR |
1732 | |
1733 | MODULE_DEVICE_TABLE(virtio, id_table); | |
1734 | MODULE_DESCRIPTION("Virtio block driver"); | |
1735 | MODULE_LICENSE("GPL"); |