xref: /linux/drivers/scsi/virtio_scsi.c (revision 791d3ef2e11100449837dc0b6fe884e60ca3a484)
1 /*
2  * Virtio SCSI HBA driver
3  *
4  * Copyright IBM Corp. 2010
5  * Copyright Red Hat, Inc. 2011
6  *
7  * Authors:
8  *  Stefan Hajnoczi   <stefanha@linux.vnet.ibm.com>
9  *  Paolo Bonzini   <pbonzini@redhat.com>
10  *
11  * This work is licensed under the terms of the GNU GPL, version 2 or later.
12  * See the COPYING file in the top-level directory.
13  *
14  */
15 
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 
18 #include <linux/module.h>
19 #include <linux/slab.h>
20 #include <linux/mempool.h>
21 #include <linux/interrupt.h>
22 #include <linux/virtio.h>
23 #include <linux/virtio_ids.h>
24 #include <linux/virtio_config.h>
25 #include <linux/virtio_scsi.h>
26 #include <linux/cpu.h>
27 #include <linux/blkdev.h>
28 #include <scsi/scsi_host.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_tcq.h>
32 #include <scsi/scsi_devinfo.h>
33 #include <linux/seqlock.h>
34 #include <linux/blk-mq-virtio.h>
35 
36 #define VIRTIO_SCSI_MEMPOOL_SZ 64
37 #define VIRTIO_SCSI_EVENT_LEN 8
38 #define VIRTIO_SCSI_VQ_BASE 2
39 
40 /* Command queue element */
41 struct virtio_scsi_cmd {
42 	struct scsi_cmnd *sc;
43 	struct completion *comp;
44 	union {
45 		struct virtio_scsi_cmd_req       cmd;
46 		struct virtio_scsi_cmd_req_pi    cmd_pi;
47 		struct virtio_scsi_ctrl_tmf_req  tmf;
48 		struct virtio_scsi_ctrl_an_req   an;
49 	} req;
50 	union {
51 		struct virtio_scsi_cmd_resp      cmd;
52 		struct virtio_scsi_ctrl_tmf_resp tmf;
53 		struct virtio_scsi_ctrl_an_resp  an;
54 		struct virtio_scsi_event         evt;
55 	} resp;
56 } ____cacheline_aligned_in_smp;
57 
58 struct virtio_scsi_event_node {
59 	struct virtio_scsi *vscsi;
60 	struct virtio_scsi_event event;
61 	struct work_struct work;
62 };
63 
64 struct virtio_scsi_vq {
65 	/* Protects vq */
66 	spinlock_t vq_lock;
67 
68 	struct virtqueue *vq;
69 };
70 
71 /*
72  * Per-target queue state.
73  *
74  * This struct holds the data needed by the queue steering policy.  When a
75  * target is sent multiple requests, we need to drive them to the same queue so
76  * that FIFO processing order is kept.  However, if a target was idle, we can
77  * choose a queue arbitrarily.  In this case the queue is chosen according to
78  * the current VCPU, so the driver expects the number of request queues to be
79  * equal to the number of VCPUs.  This makes it easy and fast to select the
80  * queue, and also lets the driver optimize the IRQ affinity for the virtqueues
81  * (each virtqueue's affinity is set to the CPU that "owns" the queue).
82  *
83  * tgt_seq is held to serialize reading and writing req_vq.
84  *
85  * Decrements of reqs are never concurrent with writes of req_vq: before the
86  * decrement reqs will be != 0; after the decrement the virtqueue completion
87  * routine will not use the req_vq so it can be changed by a new request.
88  * Thus they can happen outside the tgt_seq, provided of course we make reqs
89  * an atomic_t.
90  */
91 struct virtio_scsi_target_state {
92 	seqcount_t tgt_seq;
93 
94 	/* Currently active virtqueue for requests sent to this target. */
95 	struct virtio_scsi_vq *req_vq;
96 };
97 
98 /* Driver instance state */
99 struct virtio_scsi {
100 	struct virtio_device *vdev;
101 
102 	/* Get some buffers ready for event vq */
103 	struct virtio_scsi_event_node event_list[VIRTIO_SCSI_EVENT_LEN];
104 
105 	u32 num_queues;
106 
107 	/* If the affinity hint is set for virtqueues */
108 	bool affinity_hint_set;
109 
110 	struct hlist_node node;
111 
112 	/* Protected by event_vq lock */
113 	bool stop_events;
114 
115 	struct virtio_scsi_vq ctrl_vq;
116 	struct virtio_scsi_vq event_vq;
117 	struct virtio_scsi_vq req_vqs[];
118 };
119 
120 static struct kmem_cache *virtscsi_cmd_cache;
121 static mempool_t *virtscsi_cmd_pool;
122 
123 static inline struct Scsi_Host *virtio_scsi_host(struct virtio_device *vdev)
124 {
125 	return vdev->priv;
126 }
127 
128 static void virtscsi_compute_resid(struct scsi_cmnd *sc, u32 resid)
129 {
130 	if (!resid)
131 		return;
132 
133 	if (!scsi_bidi_cmnd(sc)) {
134 		scsi_set_resid(sc, resid);
135 		return;
136 	}
137 
138 	scsi_in(sc)->resid = min(resid, scsi_in(sc)->length);
139 	scsi_out(sc)->resid = resid - scsi_in(sc)->resid;
140 }
141 
142 /**
143  * virtscsi_complete_cmd - finish a scsi_cmd and invoke scsi_done
144  *
145  * Called with vq_lock held.
146  */
147 static void virtscsi_complete_cmd(struct virtio_scsi *vscsi, void *buf)
148 {
149 	struct virtio_scsi_cmd *cmd = buf;
150 	struct scsi_cmnd *sc = cmd->sc;
151 	struct virtio_scsi_cmd_resp *resp = &cmd->resp.cmd;
152 
153 	dev_dbg(&sc->device->sdev_gendev,
154 		"cmd %p response %u status %#02x sense_len %u\n",
155 		sc, resp->response, resp->status, resp->sense_len);
156 
157 	sc->result = resp->status;
158 	virtscsi_compute_resid(sc, virtio32_to_cpu(vscsi->vdev, resp->resid));
159 	switch (resp->response) {
160 	case VIRTIO_SCSI_S_OK:
161 		set_host_byte(sc, DID_OK);
162 		break;
163 	case VIRTIO_SCSI_S_OVERRUN:
164 		set_host_byte(sc, DID_ERROR);
165 		break;
166 	case VIRTIO_SCSI_S_ABORTED:
167 		set_host_byte(sc, DID_ABORT);
168 		break;
169 	case VIRTIO_SCSI_S_BAD_TARGET:
170 		set_host_byte(sc, DID_BAD_TARGET);
171 		break;
172 	case VIRTIO_SCSI_S_RESET:
173 		set_host_byte(sc, DID_RESET);
174 		break;
175 	case VIRTIO_SCSI_S_BUSY:
176 		set_host_byte(sc, DID_BUS_BUSY);
177 		break;
178 	case VIRTIO_SCSI_S_TRANSPORT_FAILURE:
179 		set_host_byte(sc, DID_TRANSPORT_DISRUPTED);
180 		break;
181 	case VIRTIO_SCSI_S_TARGET_FAILURE:
182 		set_host_byte(sc, DID_TARGET_FAILURE);
183 		break;
184 	case VIRTIO_SCSI_S_NEXUS_FAILURE:
185 		set_host_byte(sc, DID_NEXUS_FAILURE);
186 		break;
187 	default:
188 		scmd_printk(KERN_WARNING, sc, "Unknown response %d",
189 			    resp->response);
190 		/* fall through */
191 	case VIRTIO_SCSI_S_FAILURE:
192 		set_host_byte(sc, DID_ERROR);
193 		break;
194 	}
195 
196 	WARN_ON(virtio32_to_cpu(vscsi->vdev, resp->sense_len) >
197 		VIRTIO_SCSI_SENSE_SIZE);
198 	if (sc->sense_buffer) {
199 		memcpy(sc->sense_buffer, resp->sense,
200 		       min_t(u32,
201 			     virtio32_to_cpu(vscsi->vdev, resp->sense_len),
202 			     VIRTIO_SCSI_SENSE_SIZE));
203 		if (resp->sense_len)
204 			set_driver_byte(sc, DRIVER_SENSE);
205 	}
206 
207 	sc->scsi_done(sc);
208 }
209 
210 static void virtscsi_vq_done(struct virtio_scsi *vscsi,
211 			     struct virtio_scsi_vq *virtscsi_vq,
212 			     void (*fn)(struct virtio_scsi *vscsi, void *buf))
213 {
214 	void *buf;
215 	unsigned int len;
216 	unsigned long flags;
217 	struct virtqueue *vq = virtscsi_vq->vq;
218 
219 	spin_lock_irqsave(&virtscsi_vq->vq_lock, flags);
220 	do {
221 		virtqueue_disable_cb(vq);
222 		while ((buf = virtqueue_get_buf(vq, &len)) != NULL)
223 			fn(vscsi, buf);
224 
225 		if (unlikely(virtqueue_is_broken(vq)))
226 			break;
227 	} while (!virtqueue_enable_cb(vq));
228 	spin_unlock_irqrestore(&virtscsi_vq->vq_lock, flags);
229 }
230 
231 static void virtscsi_req_done(struct virtqueue *vq)
232 {
233 	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
234 	struct virtio_scsi *vscsi = shost_priv(sh);
235 	int index = vq->index - VIRTIO_SCSI_VQ_BASE;
236 	struct virtio_scsi_vq *req_vq = &vscsi->req_vqs[index];
237 
238 	virtscsi_vq_done(vscsi, req_vq, virtscsi_complete_cmd);
239 };
240 
241 static void virtscsi_poll_requests(struct virtio_scsi *vscsi)
242 {
243 	int i, num_vqs;
244 
245 	num_vqs = vscsi->num_queues;
246 	for (i = 0; i < num_vqs; i++)
247 		virtscsi_vq_done(vscsi, &vscsi->req_vqs[i],
248 				 virtscsi_complete_cmd);
249 }
250 
251 static void virtscsi_complete_free(struct virtio_scsi *vscsi, void *buf)
252 {
253 	struct virtio_scsi_cmd *cmd = buf;
254 
255 	if (cmd->comp)
256 		complete(cmd->comp);
257 }
258 
259 static void virtscsi_ctrl_done(struct virtqueue *vq)
260 {
261 	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
262 	struct virtio_scsi *vscsi = shost_priv(sh);
263 
264 	virtscsi_vq_done(vscsi, &vscsi->ctrl_vq, virtscsi_complete_free);
265 };
266 
267 static void virtscsi_handle_event(struct work_struct *work);
268 
269 static int virtscsi_kick_event(struct virtio_scsi *vscsi,
270 			       struct virtio_scsi_event_node *event_node)
271 {
272 	int err;
273 	struct scatterlist sg;
274 	unsigned long flags;
275 
276 	INIT_WORK(&event_node->work, virtscsi_handle_event);
277 	sg_init_one(&sg, &event_node->event, sizeof(struct virtio_scsi_event));
278 
279 	spin_lock_irqsave(&vscsi->event_vq.vq_lock, flags);
280 
281 	err = virtqueue_add_inbuf(vscsi->event_vq.vq, &sg, 1, event_node,
282 				  GFP_ATOMIC);
283 	if (!err)
284 		virtqueue_kick(vscsi->event_vq.vq);
285 
286 	spin_unlock_irqrestore(&vscsi->event_vq.vq_lock, flags);
287 
288 	return err;
289 }
290 
291 static int virtscsi_kick_event_all(struct virtio_scsi *vscsi)
292 {
293 	int i;
294 
295 	for (i = 0; i < VIRTIO_SCSI_EVENT_LEN; i++) {
296 		vscsi->event_list[i].vscsi = vscsi;
297 		virtscsi_kick_event(vscsi, &vscsi->event_list[i]);
298 	}
299 
300 	return 0;
301 }
302 
303 static void virtscsi_cancel_event_work(struct virtio_scsi *vscsi)
304 {
305 	int i;
306 
307 	/* Stop scheduling work before calling cancel_work_sync.  */
308 	spin_lock_irq(&vscsi->event_vq.vq_lock);
309 	vscsi->stop_events = true;
310 	spin_unlock_irq(&vscsi->event_vq.vq_lock);
311 
312 	for (i = 0; i < VIRTIO_SCSI_EVENT_LEN; i++)
313 		cancel_work_sync(&vscsi->event_list[i].work);
314 }
315 
316 static void virtscsi_handle_transport_reset(struct virtio_scsi *vscsi,
317 					    struct virtio_scsi_event *event)
318 {
319 	struct scsi_device *sdev;
320 	struct Scsi_Host *shost = virtio_scsi_host(vscsi->vdev);
321 	unsigned int target = event->lun[1];
322 	unsigned int lun = (event->lun[2] << 8) | event->lun[3];
323 
324 	switch (virtio32_to_cpu(vscsi->vdev, event->reason)) {
325 	case VIRTIO_SCSI_EVT_RESET_RESCAN:
326 		scsi_add_device(shost, 0, target, lun);
327 		break;
328 	case VIRTIO_SCSI_EVT_RESET_REMOVED:
329 		sdev = scsi_device_lookup(shost, 0, target, lun);
330 		if (sdev) {
331 			scsi_remove_device(sdev);
332 			scsi_device_put(sdev);
333 		} else {
334 			pr_err("SCSI device %d 0 %d %d not found\n",
335 				shost->host_no, target, lun);
336 		}
337 		break;
338 	default:
339 		pr_info("Unsupport virtio scsi event reason %x\n", event->reason);
340 	}
341 }
342 
343 static void virtscsi_handle_param_change(struct virtio_scsi *vscsi,
344 					 struct virtio_scsi_event *event)
345 {
346 	struct scsi_device *sdev;
347 	struct Scsi_Host *shost = virtio_scsi_host(vscsi->vdev);
348 	unsigned int target = event->lun[1];
349 	unsigned int lun = (event->lun[2] << 8) | event->lun[3];
350 	u8 asc = virtio32_to_cpu(vscsi->vdev, event->reason) & 255;
351 	u8 ascq = virtio32_to_cpu(vscsi->vdev, event->reason) >> 8;
352 
353 	sdev = scsi_device_lookup(shost, 0, target, lun);
354 	if (!sdev) {
355 		pr_err("SCSI device %d 0 %d %d not found\n",
356 			shost->host_no, target, lun);
357 		return;
358 	}
359 
360 	/* Handle "Parameters changed", "Mode parameters changed", and
361 	   "Capacity data has changed".  */
362 	if (asc == 0x2a && (ascq == 0x00 || ascq == 0x01 || ascq == 0x09))
363 		scsi_rescan_device(&sdev->sdev_gendev);
364 
365 	scsi_device_put(sdev);
366 }
367 
368 static void virtscsi_handle_event(struct work_struct *work)
369 {
370 	struct virtio_scsi_event_node *event_node =
371 		container_of(work, struct virtio_scsi_event_node, work);
372 	struct virtio_scsi *vscsi = event_node->vscsi;
373 	struct virtio_scsi_event *event = &event_node->event;
374 
375 	if (event->event &
376 	    cpu_to_virtio32(vscsi->vdev, VIRTIO_SCSI_T_EVENTS_MISSED)) {
377 		event->event &= ~cpu_to_virtio32(vscsi->vdev,
378 						   VIRTIO_SCSI_T_EVENTS_MISSED);
379 		scsi_scan_host(virtio_scsi_host(vscsi->vdev));
380 	}
381 
382 	switch (virtio32_to_cpu(vscsi->vdev, event->event)) {
383 	case VIRTIO_SCSI_T_NO_EVENT:
384 		break;
385 	case VIRTIO_SCSI_T_TRANSPORT_RESET:
386 		virtscsi_handle_transport_reset(vscsi, event);
387 		break;
388 	case VIRTIO_SCSI_T_PARAM_CHANGE:
389 		virtscsi_handle_param_change(vscsi, event);
390 		break;
391 	default:
392 		pr_err("Unsupport virtio scsi event %x\n", event->event);
393 	}
394 	virtscsi_kick_event(vscsi, event_node);
395 }
396 
397 static void virtscsi_complete_event(struct virtio_scsi *vscsi, void *buf)
398 {
399 	struct virtio_scsi_event_node *event_node = buf;
400 
401 	if (!vscsi->stop_events)
402 		queue_work(system_freezable_wq, &event_node->work);
403 }
404 
405 static void virtscsi_event_done(struct virtqueue *vq)
406 {
407 	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
408 	struct virtio_scsi *vscsi = shost_priv(sh);
409 
410 	virtscsi_vq_done(vscsi, &vscsi->event_vq, virtscsi_complete_event);
411 };
412 
413 /**
414  * virtscsi_add_cmd - add a virtio_scsi_cmd to a virtqueue
415  * @vq		: the struct virtqueue we're talking about
416  * @cmd		: command structure
417  * @req_size	: size of the request buffer
418  * @resp_size	: size of the response buffer
419  */
420 static int virtscsi_add_cmd(struct virtqueue *vq,
421 			    struct virtio_scsi_cmd *cmd,
422 			    size_t req_size, size_t resp_size)
423 {
424 	struct scsi_cmnd *sc = cmd->sc;
425 	struct scatterlist *sgs[6], req, resp;
426 	struct sg_table *out, *in;
427 	unsigned out_num = 0, in_num = 0;
428 
429 	out = in = NULL;
430 
431 	if (sc && sc->sc_data_direction != DMA_NONE) {
432 		if (sc->sc_data_direction != DMA_FROM_DEVICE)
433 			out = &scsi_out(sc)->table;
434 		if (sc->sc_data_direction != DMA_TO_DEVICE)
435 			in = &scsi_in(sc)->table;
436 	}
437 
438 	/* Request header.  */
439 	sg_init_one(&req, &cmd->req, req_size);
440 	sgs[out_num++] = &req;
441 
442 	/* Data-out buffer.  */
443 	if (out) {
444 		/* Place WRITE protection SGLs before Data OUT payload */
445 		if (scsi_prot_sg_count(sc))
446 			sgs[out_num++] = scsi_prot_sglist(sc);
447 		sgs[out_num++] = out->sgl;
448 	}
449 
450 	/* Response header.  */
451 	sg_init_one(&resp, &cmd->resp, resp_size);
452 	sgs[out_num + in_num++] = &resp;
453 
454 	/* Data-in buffer */
455 	if (in) {
456 		/* Place READ protection SGLs before Data IN payload */
457 		if (scsi_prot_sg_count(sc))
458 			sgs[out_num + in_num++] = scsi_prot_sglist(sc);
459 		sgs[out_num + in_num++] = in->sgl;
460 	}
461 
462 	return virtqueue_add_sgs(vq, sgs, out_num, in_num, cmd, GFP_ATOMIC);
463 }
464 
465 static int virtscsi_kick_cmd(struct virtio_scsi_vq *vq,
466 			     struct virtio_scsi_cmd *cmd,
467 			     size_t req_size, size_t resp_size)
468 {
469 	unsigned long flags;
470 	int err;
471 	bool needs_kick = false;
472 
473 	spin_lock_irqsave(&vq->vq_lock, flags);
474 	err = virtscsi_add_cmd(vq->vq, cmd, req_size, resp_size);
475 	if (!err)
476 		needs_kick = virtqueue_kick_prepare(vq->vq);
477 
478 	spin_unlock_irqrestore(&vq->vq_lock, flags);
479 
480 	if (needs_kick)
481 		virtqueue_notify(vq->vq);
482 	return err;
483 }
484 
485 static void virtio_scsi_init_hdr(struct virtio_device *vdev,
486 				 struct virtio_scsi_cmd_req *cmd,
487 				 struct scsi_cmnd *sc)
488 {
489 	cmd->lun[0] = 1;
490 	cmd->lun[1] = sc->device->id;
491 	cmd->lun[2] = (sc->device->lun >> 8) | 0x40;
492 	cmd->lun[3] = sc->device->lun & 0xff;
493 	cmd->tag = cpu_to_virtio64(vdev, (unsigned long)sc);
494 	cmd->task_attr = VIRTIO_SCSI_S_SIMPLE;
495 	cmd->prio = 0;
496 	cmd->crn = 0;
497 }
498 
499 #ifdef CONFIG_BLK_DEV_INTEGRITY
500 static void virtio_scsi_init_hdr_pi(struct virtio_device *vdev,
501 				    struct virtio_scsi_cmd_req_pi *cmd_pi,
502 				    struct scsi_cmnd *sc)
503 {
504 	struct request *rq = sc->request;
505 	struct blk_integrity *bi;
506 
507 	virtio_scsi_init_hdr(vdev, (struct virtio_scsi_cmd_req *)cmd_pi, sc);
508 
509 	if (!rq || !scsi_prot_sg_count(sc))
510 		return;
511 
512 	bi = blk_get_integrity(rq->rq_disk);
513 
514 	if (sc->sc_data_direction == DMA_TO_DEVICE)
515 		cmd_pi->pi_bytesout = cpu_to_virtio32(vdev,
516 							blk_rq_sectors(rq) *
517 							bi->tuple_size);
518 	else if (sc->sc_data_direction == DMA_FROM_DEVICE)
519 		cmd_pi->pi_bytesin = cpu_to_virtio32(vdev,
520 						       blk_rq_sectors(rq) *
521 						       bi->tuple_size);
522 }
523 #endif
524 
525 static struct virtio_scsi_vq *virtscsi_pick_vq_mq(struct virtio_scsi *vscsi,
526 						  struct scsi_cmnd *sc)
527 {
528 	u32 tag = blk_mq_unique_tag(sc->request);
529 	u16 hwq = blk_mq_unique_tag_to_hwq(tag);
530 
531 	return &vscsi->req_vqs[hwq];
532 }
533 
534 static int virtscsi_queuecommand(struct Scsi_Host *shost,
535 				 struct scsi_cmnd *sc)
536 {
537 	struct virtio_scsi *vscsi = shost_priv(shost);
538 	struct virtio_scsi_vq *req_vq = virtscsi_pick_vq_mq(vscsi, sc);
539 	struct virtio_scsi_cmd *cmd = scsi_cmd_priv(sc);
540 	unsigned long flags;
541 	int req_size;
542 	int ret;
543 
544 	BUG_ON(scsi_sg_count(sc) > shost->sg_tablesize);
545 
546 	/* TODO: check feature bit and fail if unsupported?  */
547 	BUG_ON(sc->sc_data_direction == DMA_BIDIRECTIONAL);
548 
549 	dev_dbg(&sc->device->sdev_gendev,
550 		"cmd %p CDB: %#02x\n", sc, sc->cmnd[0]);
551 
552 	cmd->sc = sc;
553 
554 	BUG_ON(sc->cmd_len > VIRTIO_SCSI_CDB_SIZE);
555 
556 #ifdef CONFIG_BLK_DEV_INTEGRITY
557 	if (virtio_has_feature(vscsi->vdev, VIRTIO_SCSI_F_T10_PI)) {
558 		virtio_scsi_init_hdr_pi(vscsi->vdev, &cmd->req.cmd_pi, sc);
559 		memcpy(cmd->req.cmd_pi.cdb, sc->cmnd, sc->cmd_len);
560 		req_size = sizeof(cmd->req.cmd_pi);
561 	} else
562 #endif
563 	{
564 		virtio_scsi_init_hdr(vscsi->vdev, &cmd->req.cmd, sc);
565 		memcpy(cmd->req.cmd.cdb, sc->cmnd, sc->cmd_len);
566 		req_size = sizeof(cmd->req.cmd);
567 	}
568 
569 	ret = virtscsi_kick_cmd(req_vq, cmd, req_size, sizeof(cmd->resp.cmd));
570 	if (ret == -EIO) {
571 		cmd->resp.cmd.response = VIRTIO_SCSI_S_BAD_TARGET;
572 		spin_lock_irqsave(&req_vq->vq_lock, flags);
573 		virtscsi_complete_cmd(vscsi, cmd);
574 		spin_unlock_irqrestore(&req_vq->vq_lock, flags);
575 	} else if (ret != 0) {
576 		return SCSI_MLQUEUE_HOST_BUSY;
577 	}
578 	return 0;
579 }
580 
581 static int virtscsi_tmf(struct virtio_scsi *vscsi, struct virtio_scsi_cmd *cmd)
582 {
583 	DECLARE_COMPLETION_ONSTACK(comp);
584 	int ret = FAILED;
585 
586 	cmd->comp = &comp;
587 	if (virtscsi_kick_cmd(&vscsi->ctrl_vq, cmd,
588 			      sizeof cmd->req.tmf, sizeof cmd->resp.tmf) < 0)
589 		goto out;
590 
591 	wait_for_completion(&comp);
592 	if (cmd->resp.tmf.response == VIRTIO_SCSI_S_OK ||
593 	    cmd->resp.tmf.response == VIRTIO_SCSI_S_FUNCTION_SUCCEEDED)
594 		ret = SUCCESS;
595 
596 	/*
597 	 * The spec guarantees that all requests related to the TMF have
598 	 * been completed, but the callback might not have run yet if
599 	 * we're using independent interrupts (e.g. MSI).  Poll the
600 	 * virtqueues once.
601 	 *
602 	 * In the abort case, sc->scsi_done will do nothing, because
603 	 * the block layer must have detected a timeout and as a result
604 	 * REQ_ATOM_COMPLETE has been set.
605 	 */
606 	virtscsi_poll_requests(vscsi);
607 
608 out:
609 	mempool_free(cmd, virtscsi_cmd_pool);
610 	return ret;
611 }
612 
613 static int virtscsi_device_reset(struct scsi_cmnd *sc)
614 {
615 	struct virtio_scsi *vscsi = shost_priv(sc->device->host);
616 	struct virtio_scsi_cmd *cmd;
617 
618 	sdev_printk(KERN_INFO, sc->device, "device reset\n");
619 	cmd = mempool_alloc(virtscsi_cmd_pool, GFP_NOIO);
620 	if (!cmd)
621 		return FAILED;
622 
623 	memset(cmd, 0, sizeof(*cmd));
624 	cmd->sc = sc;
625 	cmd->req.tmf = (struct virtio_scsi_ctrl_tmf_req){
626 		.type = VIRTIO_SCSI_T_TMF,
627 		.subtype = cpu_to_virtio32(vscsi->vdev,
628 					     VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET),
629 		.lun[0] = 1,
630 		.lun[1] = sc->device->id,
631 		.lun[2] = (sc->device->lun >> 8) | 0x40,
632 		.lun[3] = sc->device->lun & 0xff,
633 	};
634 	return virtscsi_tmf(vscsi, cmd);
635 }
636 
637 static int virtscsi_device_alloc(struct scsi_device *sdevice)
638 {
639 	/*
640 	 * Passed through SCSI targets (e.g. with qemu's 'scsi-block')
641 	 * may have transfer limits which come from the host SCSI
642 	 * controller or something on the host side other than the
643 	 * target itself.
644 	 *
645 	 * To make this work properly, the hypervisor can adjust the
646 	 * target's VPD information to advertise these limits.  But
647 	 * for that to work, the guest has to look at the VPD pages,
648 	 * which we won't do by default if it is an SPC-2 device, even
649 	 * if it does actually support it.
650 	 *
651 	 * So, set the blist to always try to read the VPD pages.
652 	 */
653 	sdevice->sdev_bflags = BLIST_TRY_VPD_PAGES;
654 
655 	return 0;
656 }
657 
658 
659 /**
660  * virtscsi_change_queue_depth() - Change a virtscsi target's queue depth
661  * @sdev:	Virtscsi target whose queue depth to change
662  * @qdepth:	New queue depth
663  */
664 static int virtscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
665 {
666 	struct Scsi_Host *shost = sdev->host;
667 	int max_depth = shost->cmd_per_lun;
668 
669 	return scsi_change_queue_depth(sdev, min(max_depth, qdepth));
670 }
671 
672 static int virtscsi_abort(struct scsi_cmnd *sc)
673 {
674 	struct virtio_scsi *vscsi = shost_priv(sc->device->host);
675 	struct virtio_scsi_cmd *cmd;
676 
677 	scmd_printk(KERN_INFO, sc, "abort\n");
678 	cmd = mempool_alloc(virtscsi_cmd_pool, GFP_NOIO);
679 	if (!cmd)
680 		return FAILED;
681 
682 	memset(cmd, 0, sizeof(*cmd));
683 	cmd->sc = sc;
684 	cmd->req.tmf = (struct virtio_scsi_ctrl_tmf_req){
685 		.type = VIRTIO_SCSI_T_TMF,
686 		.subtype = VIRTIO_SCSI_T_TMF_ABORT_TASK,
687 		.lun[0] = 1,
688 		.lun[1] = sc->device->id,
689 		.lun[2] = (sc->device->lun >> 8) | 0x40,
690 		.lun[3] = sc->device->lun & 0xff,
691 		.tag = cpu_to_virtio64(vscsi->vdev, (unsigned long)sc),
692 	};
693 	return virtscsi_tmf(vscsi, cmd);
694 }
695 
696 static int virtscsi_target_alloc(struct scsi_target *starget)
697 {
698 	struct Scsi_Host *sh = dev_to_shost(starget->dev.parent);
699 	struct virtio_scsi *vscsi = shost_priv(sh);
700 
701 	struct virtio_scsi_target_state *tgt =
702 				kmalloc(sizeof(*tgt), GFP_KERNEL);
703 	if (!tgt)
704 		return -ENOMEM;
705 
706 	seqcount_init(&tgt->tgt_seq);
707 	tgt->req_vq = &vscsi->req_vqs[0];
708 
709 	starget->hostdata = tgt;
710 	return 0;
711 }
712 
713 static void virtscsi_target_destroy(struct scsi_target *starget)
714 {
715 	struct virtio_scsi_target_state *tgt = starget->hostdata;
716 	kfree(tgt);
717 }
718 
719 static int virtscsi_map_queues(struct Scsi_Host *shost)
720 {
721 	struct virtio_scsi *vscsi = shost_priv(shost);
722 
723 	return blk_mq_virtio_map_queues(&shost->tag_set, vscsi->vdev, 2);
724 }
725 
726 /*
727  * The host guarantees to respond to each command, although I/O
728  * latencies might be higher than on bare metal.  Reset the timer
729  * unconditionally to give the host a chance to perform EH.
730  */
731 static enum blk_eh_timer_return virtscsi_eh_timed_out(struct scsi_cmnd *scmnd)
732 {
733 	return BLK_EH_RESET_TIMER;
734 }
735 
736 static struct scsi_host_template virtscsi_host_template = {
737 	.module = THIS_MODULE,
738 	.name = "Virtio SCSI HBA",
739 	.proc_name = "virtio_scsi",
740 	.this_id = -1,
741 	.cmd_size = sizeof(struct virtio_scsi_cmd),
742 	.queuecommand = virtscsi_queuecommand,
743 	.change_queue_depth = virtscsi_change_queue_depth,
744 	.eh_abort_handler = virtscsi_abort,
745 	.eh_device_reset_handler = virtscsi_device_reset,
746 	.eh_timed_out = virtscsi_eh_timed_out,
747 	.slave_alloc = virtscsi_device_alloc,
748 
749 	.dma_boundary = UINT_MAX,
750 	.use_clustering = ENABLE_CLUSTERING,
751 	.target_alloc = virtscsi_target_alloc,
752 	.target_destroy = virtscsi_target_destroy,
753 	.map_queues = virtscsi_map_queues,
754 	.track_queue_depth = 1,
755 	.force_blk_mq = 1,
756 };
757 
758 #define virtscsi_config_get(vdev, fld) \
759 	({ \
760 		typeof(((struct virtio_scsi_config *)0)->fld) __val; \
761 		virtio_cread(vdev, struct virtio_scsi_config, fld, &__val); \
762 		__val; \
763 	})
764 
765 #define virtscsi_config_set(vdev, fld, val) \
766 	do { \
767 		typeof(((struct virtio_scsi_config *)0)->fld) __val = (val); \
768 		virtio_cwrite(vdev, struct virtio_scsi_config, fld, &__val); \
769 	} while(0)
770 
771 static void virtscsi_init_vq(struct virtio_scsi_vq *virtscsi_vq,
772 			     struct virtqueue *vq)
773 {
774 	spin_lock_init(&virtscsi_vq->vq_lock);
775 	virtscsi_vq->vq = vq;
776 }
777 
778 static void virtscsi_remove_vqs(struct virtio_device *vdev)
779 {
780 	/* Stop all the virtqueues. */
781 	vdev->config->reset(vdev);
782 	vdev->config->del_vqs(vdev);
783 }
784 
785 static int virtscsi_init(struct virtio_device *vdev,
786 			 struct virtio_scsi *vscsi)
787 {
788 	int err;
789 	u32 i;
790 	u32 num_vqs;
791 	vq_callback_t **callbacks;
792 	const char **names;
793 	struct virtqueue **vqs;
794 	struct irq_affinity desc = { .pre_vectors = 2 };
795 
796 	num_vqs = vscsi->num_queues + VIRTIO_SCSI_VQ_BASE;
797 	vqs = kmalloc_array(num_vqs, sizeof(struct virtqueue *), GFP_KERNEL);
798 	callbacks = kmalloc_array(num_vqs, sizeof(vq_callback_t *),
799 				  GFP_KERNEL);
800 	names = kmalloc_array(num_vqs, sizeof(char *), GFP_KERNEL);
801 
802 	if (!callbacks || !vqs || !names) {
803 		err = -ENOMEM;
804 		goto out;
805 	}
806 
807 	callbacks[0] = virtscsi_ctrl_done;
808 	callbacks[1] = virtscsi_event_done;
809 	names[0] = "control";
810 	names[1] = "event";
811 	for (i = VIRTIO_SCSI_VQ_BASE; i < num_vqs; i++) {
812 		callbacks[i] = virtscsi_req_done;
813 		names[i] = "request";
814 	}
815 
816 	/* Discover virtqueues and write information to configuration.  */
817 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
818 	if (err)
819 		goto out;
820 
821 	virtscsi_init_vq(&vscsi->ctrl_vq, vqs[0]);
822 	virtscsi_init_vq(&vscsi->event_vq, vqs[1]);
823 	for (i = VIRTIO_SCSI_VQ_BASE; i < num_vqs; i++)
824 		virtscsi_init_vq(&vscsi->req_vqs[i - VIRTIO_SCSI_VQ_BASE],
825 				 vqs[i]);
826 
827 	virtscsi_config_set(vdev, cdb_size, VIRTIO_SCSI_CDB_SIZE);
828 	virtscsi_config_set(vdev, sense_size, VIRTIO_SCSI_SENSE_SIZE);
829 
830 	err = 0;
831 
832 out:
833 	kfree(names);
834 	kfree(callbacks);
835 	kfree(vqs);
836 	if (err)
837 		virtscsi_remove_vqs(vdev);
838 	return err;
839 }
840 
841 static int virtscsi_probe(struct virtio_device *vdev)
842 {
843 	struct Scsi_Host *shost;
844 	struct virtio_scsi *vscsi;
845 	int err;
846 	u32 sg_elems, num_targets;
847 	u32 cmd_per_lun;
848 	u32 num_queues;
849 
850 	if (!vdev->config->get) {
851 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
852 			__func__);
853 		return -EINVAL;
854 	}
855 
856 	/* We need to know how many queues before we allocate. */
857 	num_queues = virtscsi_config_get(vdev, num_queues) ? : 1;
858 
859 	num_targets = virtscsi_config_get(vdev, max_target) + 1;
860 
861 	shost = scsi_host_alloc(&virtscsi_host_template,
862 		sizeof(*vscsi) + sizeof(vscsi->req_vqs[0]) * num_queues);
863 	if (!shost)
864 		return -ENOMEM;
865 
866 	sg_elems = virtscsi_config_get(vdev, seg_max) ?: 1;
867 	shost->sg_tablesize = sg_elems;
868 	vscsi = shost_priv(shost);
869 	vscsi->vdev = vdev;
870 	vscsi->num_queues = num_queues;
871 	vdev->priv = shost;
872 
873 	err = virtscsi_init(vdev, vscsi);
874 	if (err)
875 		goto virtscsi_init_failed;
876 
877 	shost->can_queue = virtqueue_get_vring_size(vscsi->req_vqs[0].vq);
878 
879 	cmd_per_lun = virtscsi_config_get(vdev, cmd_per_lun) ?: 1;
880 	shost->cmd_per_lun = min_t(u32, cmd_per_lun, shost->can_queue);
881 	shost->max_sectors = virtscsi_config_get(vdev, max_sectors) ?: 0xFFFF;
882 
883 	/* LUNs > 256 are reported with format 1, so they go in the range
884 	 * 16640-32767.
885 	 */
886 	shost->max_lun = virtscsi_config_get(vdev, max_lun) + 1 + 0x4000;
887 	shost->max_id = num_targets;
888 	shost->max_channel = 0;
889 	shost->max_cmd_len = VIRTIO_SCSI_CDB_SIZE;
890 	shost->nr_hw_queues = num_queues;
891 
892 #ifdef CONFIG_BLK_DEV_INTEGRITY
893 	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_T10_PI)) {
894 		int host_prot;
895 
896 		host_prot = SHOST_DIF_TYPE1_PROTECTION | SHOST_DIF_TYPE2_PROTECTION |
897 			    SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE1_PROTECTION |
898 			    SHOST_DIX_TYPE2_PROTECTION | SHOST_DIX_TYPE3_PROTECTION;
899 
900 		scsi_host_set_prot(shost, host_prot);
901 		scsi_host_set_guard(shost, SHOST_DIX_GUARD_CRC);
902 	}
903 #endif
904 
905 	err = scsi_add_host(shost, &vdev->dev);
906 	if (err)
907 		goto scsi_add_host_failed;
908 
909 	virtio_device_ready(vdev);
910 
911 	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG))
912 		virtscsi_kick_event_all(vscsi);
913 
914 	scsi_scan_host(shost);
915 	return 0;
916 
917 scsi_add_host_failed:
918 	vdev->config->del_vqs(vdev);
919 virtscsi_init_failed:
920 	scsi_host_put(shost);
921 	return err;
922 }
923 
924 static void virtscsi_remove(struct virtio_device *vdev)
925 {
926 	struct Scsi_Host *shost = virtio_scsi_host(vdev);
927 	struct virtio_scsi *vscsi = shost_priv(shost);
928 
929 	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG))
930 		virtscsi_cancel_event_work(vscsi);
931 
932 	scsi_remove_host(shost);
933 	virtscsi_remove_vqs(vdev);
934 	scsi_host_put(shost);
935 }
936 
937 #ifdef CONFIG_PM_SLEEP
938 static int virtscsi_freeze(struct virtio_device *vdev)
939 {
940 	virtscsi_remove_vqs(vdev);
941 	return 0;
942 }
943 
944 static int virtscsi_restore(struct virtio_device *vdev)
945 {
946 	struct Scsi_Host *sh = virtio_scsi_host(vdev);
947 	struct virtio_scsi *vscsi = shost_priv(sh);
948 	int err;
949 
950 	err = virtscsi_init(vdev, vscsi);
951 	if (err)
952 		return err;
953 
954 	virtio_device_ready(vdev);
955 
956 	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG))
957 		virtscsi_kick_event_all(vscsi);
958 
959 	return err;
960 }
961 #endif
962 
963 static struct virtio_device_id id_table[] = {
964 	{ VIRTIO_ID_SCSI, VIRTIO_DEV_ANY_ID },
965 	{ 0 },
966 };
967 
968 static unsigned int features[] = {
969 	VIRTIO_SCSI_F_HOTPLUG,
970 	VIRTIO_SCSI_F_CHANGE,
971 #ifdef CONFIG_BLK_DEV_INTEGRITY
972 	VIRTIO_SCSI_F_T10_PI,
973 #endif
974 };
975 
976 static struct virtio_driver virtio_scsi_driver = {
977 	.feature_table = features,
978 	.feature_table_size = ARRAY_SIZE(features),
979 	.driver.name = KBUILD_MODNAME,
980 	.driver.owner = THIS_MODULE,
981 	.id_table = id_table,
982 	.probe = virtscsi_probe,
983 #ifdef CONFIG_PM_SLEEP
984 	.freeze = virtscsi_freeze,
985 	.restore = virtscsi_restore,
986 #endif
987 	.remove = virtscsi_remove,
988 };
989 
990 static int __init init(void)
991 {
992 	int ret = -ENOMEM;
993 
994 	virtscsi_cmd_cache = KMEM_CACHE(virtio_scsi_cmd, 0);
995 	if (!virtscsi_cmd_cache) {
996 		pr_err("kmem_cache_create() for virtscsi_cmd_cache failed\n");
997 		goto error;
998 	}
999 
1000 
1001 	virtscsi_cmd_pool =
1002 		mempool_create_slab_pool(VIRTIO_SCSI_MEMPOOL_SZ,
1003 					 virtscsi_cmd_cache);
1004 	if (!virtscsi_cmd_pool) {
1005 		pr_err("mempool_create() for virtscsi_cmd_pool failed\n");
1006 		goto error;
1007 	}
1008 	ret = register_virtio_driver(&virtio_scsi_driver);
1009 	if (ret < 0)
1010 		goto error;
1011 
1012 	return 0;
1013 
1014 error:
1015 	if (virtscsi_cmd_pool) {
1016 		mempool_destroy(virtscsi_cmd_pool);
1017 		virtscsi_cmd_pool = NULL;
1018 	}
1019 	if (virtscsi_cmd_cache) {
1020 		kmem_cache_destroy(virtscsi_cmd_cache);
1021 		virtscsi_cmd_cache = NULL;
1022 	}
1023 	return ret;
1024 }
1025 
1026 static void __exit fini(void)
1027 {
1028 	unregister_virtio_driver(&virtio_scsi_driver);
1029 	mempool_destroy(virtscsi_cmd_pool);
1030 	kmem_cache_destroy(virtscsi_cmd_cache);
1031 }
1032 module_init(init);
1033 module_exit(fini);
1034 
1035 MODULE_DEVICE_TABLE(virtio, id_table);
1036 MODULE_DESCRIPTION("Virtio SCSI HBA driver");
1037 MODULE_LICENSE("GPL");
1038