1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Virtio SCSI HBA driver
4 *
5 * Copyright IBM Corp. 2010
6 * Copyright Red Hat, Inc. 2011
7 *
8 * Authors:
9 *  Stefan Hajnoczi   <stefanha@linux.vnet.ibm.com>
10 *  Paolo Bonzini   <pbonzini@redhat.com>
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/module.h>
16#include <linux/slab.h>
17#include <linux/mempool.h>
18#include <linux/interrupt.h>
19#include <linux/virtio.h>
20#include <linux/virtio_ids.h>
21#include <linux/virtio_config.h>
22#include <linux/virtio_scsi.h>
23#include <linux/cpu.h>
24#include <linux/blkdev.h>
25#include <linux/blk-integrity.h>
26#include <scsi/scsi_host.h>
27#include <scsi/scsi_device.h>
28#include <scsi/scsi_cmnd.h>
29#include <scsi/scsi_tcq.h>
30#include <scsi/scsi_devinfo.h>
31#include <linux/seqlock.h>
32#include <linux/blk-mq-virtio.h>
33
34#include "sd.h"
35
36#define VIRTIO_SCSI_MEMPOOL_SZ 64
37#define VIRTIO_SCSI_EVENT_LEN 8
38#define VIRTIO_SCSI_VQ_BASE 2
39
40static unsigned int virtscsi_poll_queues;
41module_param(virtscsi_poll_queues, uint, 0644);
42MODULE_PARM_DESC(virtscsi_poll_queues,
43		 "The number of dedicated virtqueues for polling I/O");
44
45/* Command queue element */
46struct virtio_scsi_cmd {
47	struct scsi_cmnd *sc;
48	struct completion *comp;
49	union {
50		struct virtio_scsi_cmd_req       cmd;
51		struct virtio_scsi_cmd_req_pi    cmd_pi;
52		struct virtio_scsi_ctrl_tmf_req  tmf;
53		struct virtio_scsi_ctrl_an_req   an;
54	} req;
55	union {
56		struct virtio_scsi_cmd_resp      cmd;
57		struct virtio_scsi_ctrl_tmf_resp tmf;
58		struct virtio_scsi_ctrl_an_resp  an;
59		struct virtio_scsi_event         evt;
60	} resp;
61} ____cacheline_aligned_in_smp;
62
63struct virtio_scsi_event_node {
64	struct virtio_scsi *vscsi;
65	struct virtio_scsi_event event;
66	struct work_struct work;
67};
68
69struct virtio_scsi_vq {
70	/* Protects vq */
71	spinlock_t vq_lock;
72
73	struct virtqueue *vq;
74};
75
76/* Driver instance state */
77struct virtio_scsi {
78	struct virtio_device *vdev;
79
80	/* Get some buffers ready for event vq */
81	struct virtio_scsi_event_node event_list[VIRTIO_SCSI_EVENT_LEN];
82
83	u32 num_queues;
84	int io_queues[HCTX_MAX_TYPES];
85
86	struct hlist_node node;
87
88	/* Protected by event_vq lock */
89	bool stop_events;
90
91	struct virtio_scsi_vq ctrl_vq;
92	struct virtio_scsi_vq event_vq;
93	struct virtio_scsi_vq req_vqs[];
94};
95
96static struct kmem_cache *virtscsi_cmd_cache;
97static mempool_t *virtscsi_cmd_pool;
98
99static inline struct Scsi_Host *virtio_scsi_host(struct virtio_device *vdev)
100{
101	return vdev->priv;
102}
103
104static void virtscsi_compute_resid(struct scsi_cmnd *sc, u32 resid)
105{
106	if (resid)
107		scsi_set_resid(sc, min(resid, scsi_bufflen(sc)));
108}
109
110/*
111 * virtscsi_complete_cmd - finish a scsi_cmd and invoke scsi_done
112 *
113 * Called with vq_lock held.
114 */
115static void virtscsi_complete_cmd(struct virtio_scsi *vscsi, void *buf)
116{
117	struct virtio_scsi_cmd *cmd = buf;
118	struct scsi_cmnd *sc = cmd->sc;
119	struct virtio_scsi_cmd_resp *resp = &cmd->resp.cmd;
120
121	dev_dbg(&sc->device->sdev_gendev,
122		"cmd %p response %u status %#02x sense_len %u\n",
123		sc, resp->response, resp->status, resp->sense_len);
124
125	sc->result = resp->status;
126	virtscsi_compute_resid(sc, virtio32_to_cpu(vscsi->vdev, resp->resid));
127	switch (resp->response) {
128	case VIRTIO_SCSI_S_OK:
129		set_host_byte(sc, DID_OK);
130		break;
131	case VIRTIO_SCSI_S_OVERRUN:
132		set_host_byte(sc, DID_ERROR);
133		break;
134	case VIRTIO_SCSI_S_ABORTED:
135		set_host_byte(sc, DID_ABORT);
136		break;
137	case VIRTIO_SCSI_S_BAD_TARGET:
138		set_host_byte(sc, DID_BAD_TARGET);
139		break;
140	case VIRTIO_SCSI_S_RESET:
141		set_host_byte(sc, DID_RESET);
142		break;
143	case VIRTIO_SCSI_S_BUSY:
144		set_host_byte(sc, DID_BUS_BUSY);
145		break;
146	case VIRTIO_SCSI_S_TRANSPORT_FAILURE:
147		set_host_byte(sc, DID_TRANSPORT_DISRUPTED);
148		break;
149	case VIRTIO_SCSI_S_TARGET_FAILURE:
150		set_host_byte(sc, DID_BAD_TARGET);
151		break;
152	case VIRTIO_SCSI_S_NEXUS_FAILURE:
153		set_status_byte(sc, SAM_STAT_RESERVATION_CONFLICT);
154		break;
155	default:
156		scmd_printk(KERN_WARNING, sc, "Unknown response %d",
157			    resp->response);
158		fallthrough;
159	case VIRTIO_SCSI_S_FAILURE:
160		set_host_byte(sc, DID_ERROR);
161		break;
162	}
163
164	WARN_ON(virtio32_to_cpu(vscsi->vdev, resp->sense_len) >
165		VIRTIO_SCSI_SENSE_SIZE);
166	if (resp->sense_len) {
167		memcpy(sc->sense_buffer, resp->sense,
168		       min_t(u32,
169			     virtio32_to_cpu(vscsi->vdev, resp->sense_len),
170			     VIRTIO_SCSI_SENSE_SIZE));
171	}
172
173	scsi_done(sc);
174}
175
176static void virtscsi_vq_done(struct virtio_scsi *vscsi,
177			     struct virtio_scsi_vq *virtscsi_vq,
178			     void (*fn)(struct virtio_scsi *vscsi, void *buf))
179{
180	void *buf;
181	unsigned int len;
182	unsigned long flags;
183	struct virtqueue *vq = virtscsi_vq->vq;
184
185	spin_lock_irqsave(&virtscsi_vq->vq_lock, flags);
186	do {
187		virtqueue_disable_cb(vq);
188		while ((buf = virtqueue_get_buf(vq, &len)) != NULL)
189			fn(vscsi, buf);
190
191	} while (!virtqueue_enable_cb(vq));
192	spin_unlock_irqrestore(&virtscsi_vq->vq_lock, flags);
193}
194
195static void virtscsi_req_done(struct virtqueue *vq)
196{
197	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
198	struct virtio_scsi *vscsi = shost_priv(sh);
199	int index = vq->index - VIRTIO_SCSI_VQ_BASE;
200	struct virtio_scsi_vq *req_vq = &vscsi->req_vqs[index];
201
202	virtscsi_vq_done(vscsi, req_vq, virtscsi_complete_cmd);
203};
204
205static void virtscsi_poll_requests(struct virtio_scsi *vscsi)
206{
207	int i, num_vqs;
208
209	num_vqs = vscsi->num_queues;
210	for (i = 0; i < num_vqs; i++)
211		virtscsi_vq_done(vscsi, &vscsi->req_vqs[i],
212				 virtscsi_complete_cmd);
213}
214
215static void virtscsi_complete_free(struct virtio_scsi *vscsi, void *buf)
216{
217	struct virtio_scsi_cmd *cmd = buf;
218
219	if (cmd->comp)
220		complete(cmd->comp);
221}
222
223static void virtscsi_ctrl_done(struct virtqueue *vq)
224{
225	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
226	struct virtio_scsi *vscsi = shost_priv(sh);
227
228	virtscsi_vq_done(vscsi, &vscsi->ctrl_vq, virtscsi_complete_free);
229};
230
231static void virtscsi_handle_event(struct work_struct *work);
232
233static int virtscsi_kick_event(struct virtio_scsi *vscsi,
234			       struct virtio_scsi_event_node *event_node)
235{
236	int err;
237	struct scatterlist sg;
238	unsigned long flags;
239
240	INIT_WORK(&event_node->work, virtscsi_handle_event);
241	sg_init_one(&sg, &event_node->event, sizeof(struct virtio_scsi_event));
242
243	spin_lock_irqsave(&vscsi->event_vq.vq_lock, flags);
244
245	err = virtqueue_add_inbuf(vscsi->event_vq.vq, &sg, 1, event_node,
246				  GFP_ATOMIC);
247	if (!err)
248		virtqueue_kick(vscsi->event_vq.vq);
249
250	spin_unlock_irqrestore(&vscsi->event_vq.vq_lock, flags);
251
252	return err;
253}
254
255static int virtscsi_kick_event_all(struct virtio_scsi *vscsi)
256{
257	int i;
258
259	for (i = 0; i < VIRTIO_SCSI_EVENT_LEN; i++) {
260		vscsi->event_list[i].vscsi = vscsi;
261		virtscsi_kick_event(vscsi, &vscsi->event_list[i]);
262	}
263
264	return 0;
265}
266
267static void virtscsi_cancel_event_work(struct virtio_scsi *vscsi)
268{
269	int i;
270
271	/* Stop scheduling work before calling cancel_work_sync.  */
272	spin_lock_irq(&vscsi->event_vq.vq_lock);
273	vscsi->stop_events = true;
274	spin_unlock_irq(&vscsi->event_vq.vq_lock);
275
276	for (i = 0; i < VIRTIO_SCSI_EVENT_LEN; i++)
277		cancel_work_sync(&vscsi->event_list[i].work);
278}
279
280static void virtscsi_handle_transport_reset(struct virtio_scsi *vscsi,
281					    struct virtio_scsi_event *event)
282{
283	struct scsi_device *sdev;
284	struct Scsi_Host *shost = virtio_scsi_host(vscsi->vdev);
285	unsigned int target = event->lun[1];
286	unsigned int lun = (event->lun[2] << 8) | event->lun[3];
287
288	switch (virtio32_to_cpu(vscsi->vdev, event->reason)) {
289	case VIRTIO_SCSI_EVT_RESET_RESCAN:
290		if (lun == 0) {
291			scsi_scan_target(&shost->shost_gendev, 0, target,
292					 SCAN_WILD_CARD, SCSI_SCAN_INITIAL);
293		} else {
294			scsi_add_device(shost, 0, target, lun);
295		}
296		break;
297	case VIRTIO_SCSI_EVT_RESET_REMOVED:
298		sdev = scsi_device_lookup(shost, 0, target, lun);
299		if (sdev) {
300			scsi_remove_device(sdev);
301			scsi_device_put(sdev);
302		} else {
303			pr_err("SCSI device %d 0 %d %d not found\n",
304				shost->host_no, target, lun);
305		}
306		break;
307	default:
308		pr_info("Unsupported virtio scsi event reason %x\n", event->reason);
309	}
310}
311
312static void virtscsi_handle_param_change(struct virtio_scsi *vscsi,
313					 struct virtio_scsi_event *event)
314{
315	struct scsi_device *sdev;
316	struct Scsi_Host *shost = virtio_scsi_host(vscsi->vdev);
317	unsigned int target = event->lun[1];
318	unsigned int lun = (event->lun[2] << 8) | event->lun[3];
319	u8 asc = virtio32_to_cpu(vscsi->vdev, event->reason) & 255;
320	u8 ascq = virtio32_to_cpu(vscsi->vdev, event->reason) >> 8;
321
322	sdev = scsi_device_lookup(shost, 0, target, lun);
323	if (!sdev) {
324		pr_err("SCSI device %d 0 %d %d not found\n",
325			shost->host_no, target, lun);
326		return;
327	}
328
329	/* Handle "Parameters changed", "Mode parameters changed", and
330	   "Capacity data has changed".  */
331	if (asc == 0x2a && (ascq == 0x00 || ascq == 0x01 || ascq == 0x09))
332		scsi_rescan_device(sdev);
333
334	scsi_device_put(sdev);
335}
336
337static int virtscsi_rescan_hotunplug(struct virtio_scsi *vscsi)
338{
339	struct scsi_device *sdev;
340	struct Scsi_Host *shost = virtio_scsi_host(vscsi->vdev);
341	unsigned char scsi_cmd[MAX_COMMAND_SIZE];
342	int result, inquiry_len, inq_result_len = 256;
343	char *inq_result = kmalloc(inq_result_len, GFP_KERNEL);
344
345	if (!inq_result)
346		return -ENOMEM;
347
348	shost_for_each_device(sdev, shost) {
349		inquiry_len = sdev->inquiry_len ? sdev->inquiry_len : 36;
350
351		memset(scsi_cmd, 0, sizeof(scsi_cmd));
352		scsi_cmd[0] = INQUIRY;
353		scsi_cmd[4] = (unsigned char) inquiry_len;
354
355		memset(inq_result, 0, inq_result_len);
356
357		result = scsi_execute_cmd(sdev, scsi_cmd, REQ_OP_DRV_IN,
358					  inq_result, inquiry_len,
359					  SD_TIMEOUT, SD_MAX_RETRIES, NULL);
360
361		if (result == 0 && inq_result[0] >> 5) {
362			/* PQ indicates the LUN is not attached */
363			scsi_remove_device(sdev);
364		} else if (result > 0 && host_byte(result) == DID_BAD_TARGET) {
365			/*
366			 * If all LUNs of a virtio-scsi device are unplugged
367			 * it will respond with BAD TARGET on any INQUIRY
368			 * command.
369			 * Remove the device in this case as well.
370			 */
371			scsi_remove_device(sdev);
372		}
373	}
374
375	kfree(inq_result);
376	return 0;
377}
378
379static void virtscsi_handle_event(struct work_struct *work)
380{
381	struct virtio_scsi_event_node *event_node =
382		container_of(work, struct virtio_scsi_event_node, work);
383	struct virtio_scsi *vscsi = event_node->vscsi;
384	struct virtio_scsi_event *event = &event_node->event;
385
386	if (event->event &
387	    cpu_to_virtio32(vscsi->vdev, VIRTIO_SCSI_T_EVENTS_MISSED)) {
388		int ret;
389
390		event->event &= ~cpu_to_virtio32(vscsi->vdev,
391						   VIRTIO_SCSI_T_EVENTS_MISSED);
392		ret = virtscsi_rescan_hotunplug(vscsi);
393		if (ret)
394			return;
395		scsi_scan_host(virtio_scsi_host(vscsi->vdev));
396	}
397
398	switch (virtio32_to_cpu(vscsi->vdev, event->event)) {
399	case VIRTIO_SCSI_T_NO_EVENT:
400		break;
401	case VIRTIO_SCSI_T_TRANSPORT_RESET:
402		virtscsi_handle_transport_reset(vscsi, event);
403		break;
404	case VIRTIO_SCSI_T_PARAM_CHANGE:
405		virtscsi_handle_param_change(vscsi, event);
406		break;
407	default:
408		pr_err("Unsupported virtio scsi event %x\n", event->event);
409	}
410	virtscsi_kick_event(vscsi, event_node);
411}
412
413static void virtscsi_complete_event(struct virtio_scsi *vscsi, void *buf)
414{
415	struct virtio_scsi_event_node *event_node = buf;
416
417	if (!vscsi->stop_events)
418		queue_work(system_freezable_wq, &event_node->work);
419}
420
421static void virtscsi_event_done(struct virtqueue *vq)
422{
423	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
424	struct virtio_scsi *vscsi = shost_priv(sh);
425
426	virtscsi_vq_done(vscsi, &vscsi->event_vq, virtscsi_complete_event);
427};
428
429static int __virtscsi_add_cmd(struct virtqueue *vq,
430			    struct virtio_scsi_cmd *cmd,
431			    size_t req_size, size_t resp_size)
432{
433	struct scsi_cmnd *sc = cmd->sc;
434	struct scatterlist *sgs[6], req, resp;
435	struct sg_table *out, *in;
436	unsigned out_num = 0, in_num = 0;
437
438	out = in = NULL;
439
440	if (sc && sc->sc_data_direction != DMA_NONE) {
441		if (sc->sc_data_direction != DMA_FROM_DEVICE)
442			out = &sc->sdb.table;
443		if (sc->sc_data_direction != DMA_TO_DEVICE)
444			in = &sc->sdb.table;
445	}
446
447	/* Request header.  */
448	sg_init_one(&req, &cmd->req, req_size);
449	sgs[out_num++] = &req;
450
451	/* Data-out buffer.  */
452	if (out) {
453		/* Place WRITE protection SGLs before Data OUT payload */
454		if (scsi_prot_sg_count(sc))
455			sgs[out_num++] = scsi_prot_sglist(sc);
456		sgs[out_num++] = out->sgl;
457	}
458
459	/* Response header.  */
460	sg_init_one(&resp, &cmd->resp, resp_size);
461	sgs[out_num + in_num++] = &resp;
462
463	/* Data-in buffer */
464	if (in) {
465		/* Place READ protection SGLs before Data IN payload */
466		if (scsi_prot_sg_count(sc))
467			sgs[out_num + in_num++] = scsi_prot_sglist(sc);
468		sgs[out_num + in_num++] = in->sgl;
469	}
470
471	return virtqueue_add_sgs(vq, sgs, out_num, in_num, cmd, GFP_ATOMIC);
472}
473
474static void virtscsi_kick_vq(struct virtio_scsi_vq *vq)
475{
476	bool needs_kick;
477	unsigned long flags;
478
479	spin_lock_irqsave(&vq->vq_lock, flags);
480	needs_kick = virtqueue_kick_prepare(vq->vq);
481	spin_unlock_irqrestore(&vq->vq_lock, flags);
482
483	if (needs_kick)
484		virtqueue_notify(vq->vq);
485}
486
487/**
488 * virtscsi_add_cmd - add a virtio_scsi_cmd to a virtqueue, optionally kick it
489 * @vq		: the struct virtqueue we're talking about
490 * @cmd		: command structure
491 * @req_size	: size of the request buffer
492 * @resp_size	: size of the response buffer
493 * @kick	: whether to kick the virtqueue immediately
494 */
495static int virtscsi_add_cmd(struct virtio_scsi_vq *vq,
496			     struct virtio_scsi_cmd *cmd,
497			     size_t req_size, size_t resp_size,
498			     bool kick)
499{
500	unsigned long flags;
501	int err;
502	bool needs_kick = false;
503
504	spin_lock_irqsave(&vq->vq_lock, flags);
505	err = __virtscsi_add_cmd(vq->vq, cmd, req_size, resp_size);
506	if (!err && kick)
507		needs_kick = virtqueue_kick_prepare(vq->vq);
508
509	spin_unlock_irqrestore(&vq->vq_lock, flags);
510
511	if (needs_kick)
512		virtqueue_notify(vq->vq);
513	return err;
514}
515
516static void virtio_scsi_init_hdr(struct virtio_device *vdev,
517				 struct virtio_scsi_cmd_req *cmd,
518				 struct scsi_cmnd *sc)
519{
520	cmd->lun[0] = 1;
521	cmd->lun[1] = sc->device->id;
522	cmd->lun[2] = (sc->device->lun >> 8) | 0x40;
523	cmd->lun[3] = sc->device->lun & 0xff;
524	cmd->tag = cpu_to_virtio64(vdev, (unsigned long)sc);
525	cmd->task_attr = VIRTIO_SCSI_S_SIMPLE;
526	cmd->prio = 0;
527	cmd->crn = 0;
528}
529
530#ifdef CONFIG_BLK_DEV_INTEGRITY
531static void virtio_scsi_init_hdr_pi(struct virtio_device *vdev,
532				    struct virtio_scsi_cmd_req_pi *cmd_pi,
533				    struct scsi_cmnd *sc)
534{
535	struct request *rq = scsi_cmd_to_rq(sc);
536	struct blk_integrity *bi;
537
538	virtio_scsi_init_hdr(vdev, (struct virtio_scsi_cmd_req *)cmd_pi, sc);
539
540	if (!rq || !scsi_prot_sg_count(sc))
541		return;
542
543	bi = blk_get_integrity(rq->q->disk);
544
545	if (sc->sc_data_direction == DMA_TO_DEVICE)
546		cmd_pi->pi_bytesout = cpu_to_virtio32(vdev,
547						      bio_integrity_bytes(bi,
548							blk_rq_sectors(rq)));
549	else if (sc->sc_data_direction == DMA_FROM_DEVICE)
550		cmd_pi->pi_bytesin = cpu_to_virtio32(vdev,
551						     bio_integrity_bytes(bi,
552							blk_rq_sectors(rq)));
553}
554#endif
555
556static struct virtio_scsi_vq *virtscsi_pick_vq_mq(struct virtio_scsi *vscsi,
557						  struct scsi_cmnd *sc)
558{
559	u32 tag = blk_mq_unique_tag(scsi_cmd_to_rq(sc));
560	u16 hwq = blk_mq_unique_tag_to_hwq(tag);
561
562	return &vscsi->req_vqs[hwq];
563}
564
565static int virtscsi_queuecommand(struct Scsi_Host *shost,
566				 struct scsi_cmnd *sc)
567{
568	struct virtio_scsi *vscsi = shost_priv(shost);
569	struct virtio_scsi_vq *req_vq = virtscsi_pick_vq_mq(vscsi, sc);
570	struct virtio_scsi_cmd *cmd = scsi_cmd_priv(sc);
571	bool kick;
572	unsigned long flags;
573	int req_size;
574	int ret;
575
576	BUG_ON(scsi_sg_count(sc) > shost->sg_tablesize);
577
578	/* TODO: check feature bit and fail if unsupported?  */
579	BUG_ON(sc->sc_data_direction == DMA_BIDIRECTIONAL);
580
581	dev_dbg(&sc->device->sdev_gendev,
582		"cmd %p CDB: %#02x\n", sc, sc->cmnd[0]);
583
584	cmd->sc = sc;
585
586	BUG_ON(sc->cmd_len > VIRTIO_SCSI_CDB_SIZE);
587
588#ifdef CONFIG_BLK_DEV_INTEGRITY
589	if (virtio_has_feature(vscsi->vdev, VIRTIO_SCSI_F_T10_PI)) {
590		virtio_scsi_init_hdr_pi(vscsi->vdev, &cmd->req.cmd_pi, sc);
591		memcpy(cmd->req.cmd_pi.cdb, sc->cmnd, sc->cmd_len);
592		req_size = sizeof(cmd->req.cmd_pi);
593	} else
594#endif
595	{
596		virtio_scsi_init_hdr(vscsi->vdev, &cmd->req.cmd, sc);
597		memcpy(cmd->req.cmd.cdb, sc->cmnd, sc->cmd_len);
598		req_size = sizeof(cmd->req.cmd);
599	}
600
601	kick = (sc->flags & SCMD_LAST) != 0;
602	ret = virtscsi_add_cmd(req_vq, cmd, req_size, sizeof(cmd->resp.cmd), kick);
603	if (ret == -EIO) {
604		cmd->resp.cmd.response = VIRTIO_SCSI_S_BAD_TARGET;
605		spin_lock_irqsave(&req_vq->vq_lock, flags);
606		virtscsi_complete_cmd(vscsi, cmd);
607		spin_unlock_irqrestore(&req_vq->vq_lock, flags);
608	} else if (ret != 0) {
609		return SCSI_MLQUEUE_HOST_BUSY;
610	}
611	return 0;
612}
613
614static int virtscsi_tmf(struct virtio_scsi *vscsi, struct virtio_scsi_cmd *cmd)
615{
616	DECLARE_COMPLETION_ONSTACK(comp);
617	int ret = FAILED;
618
619	cmd->comp = &comp;
620	if (virtscsi_add_cmd(&vscsi->ctrl_vq, cmd,
621			      sizeof cmd->req.tmf, sizeof cmd->resp.tmf, true) < 0)
622		goto out;
623
624	wait_for_completion(&comp);
625	if (cmd->resp.tmf.response == VIRTIO_SCSI_S_OK ||
626	    cmd->resp.tmf.response == VIRTIO_SCSI_S_FUNCTION_SUCCEEDED)
627		ret = SUCCESS;
628
629	/*
630	 * The spec guarantees that all requests related to the TMF have
631	 * been completed, but the callback might not have run yet if
632	 * we're using independent interrupts (e.g. MSI).  Poll the
633	 * virtqueues once.
634	 *
635	 * In the abort case, scsi_done() will do nothing, because the
636	 * command timed out and hence SCMD_STATE_COMPLETE has been set.
637	 */
638	virtscsi_poll_requests(vscsi);
639
640out:
641	mempool_free(cmd, virtscsi_cmd_pool);
642	return ret;
643}
644
645static int virtscsi_device_reset(struct scsi_cmnd *sc)
646{
647	struct virtio_scsi *vscsi = shost_priv(sc->device->host);
648	struct virtio_scsi_cmd *cmd;
649
650	sdev_printk(KERN_INFO, sc->device, "device reset\n");
651	cmd = mempool_alloc(virtscsi_cmd_pool, GFP_NOIO);
652	if (!cmd)
653		return FAILED;
654
655	memset(cmd, 0, sizeof(*cmd));
656	cmd->req.tmf = (struct virtio_scsi_ctrl_tmf_req){
657		.type = VIRTIO_SCSI_T_TMF,
658		.subtype = cpu_to_virtio32(vscsi->vdev,
659					     VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET),
660		.lun[0] = 1,
661		.lun[1] = sc->device->id,
662		.lun[2] = (sc->device->lun >> 8) | 0x40,
663		.lun[3] = sc->device->lun & 0xff,
664	};
665	return virtscsi_tmf(vscsi, cmd);
666}
667
668static int virtscsi_device_alloc(struct scsi_device *sdevice)
669{
670	/*
671	 * Passed through SCSI targets (e.g. with qemu's 'scsi-block')
672	 * may have transfer limits which come from the host SCSI
673	 * controller or something on the host side other than the
674	 * target itself.
675	 *
676	 * To make this work properly, the hypervisor can adjust the
677	 * target's VPD information to advertise these limits.  But
678	 * for that to work, the guest has to look at the VPD pages,
679	 * which we won't do by default if it is an SPC-2 device, even
680	 * if it does actually support it.
681	 *
682	 * So, set the blist to always try to read the VPD pages.
683	 */
684	sdevice->sdev_bflags = BLIST_TRY_VPD_PAGES;
685
686	return 0;
687}
688
689
690/**
691 * virtscsi_change_queue_depth() - Change a virtscsi target's queue depth
692 * @sdev:	Virtscsi target whose queue depth to change
693 * @qdepth:	New queue depth
694 */
695static int virtscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
696{
697	struct Scsi_Host *shost = sdev->host;
698	int max_depth = shost->cmd_per_lun;
699
700	return scsi_change_queue_depth(sdev, min(max_depth, qdepth));
701}
702
703static int virtscsi_abort(struct scsi_cmnd *sc)
704{
705	struct virtio_scsi *vscsi = shost_priv(sc->device->host);
706	struct virtio_scsi_cmd *cmd;
707
708	scmd_printk(KERN_INFO, sc, "abort\n");
709	cmd = mempool_alloc(virtscsi_cmd_pool, GFP_NOIO);
710	if (!cmd)
711		return FAILED;
712
713	memset(cmd, 0, sizeof(*cmd));
714	cmd->req.tmf = (struct virtio_scsi_ctrl_tmf_req){
715		.type = VIRTIO_SCSI_T_TMF,
716		.subtype = VIRTIO_SCSI_T_TMF_ABORT_TASK,
717		.lun[0] = 1,
718		.lun[1] = sc->device->id,
719		.lun[2] = (sc->device->lun >> 8) | 0x40,
720		.lun[3] = sc->device->lun & 0xff,
721		.tag = cpu_to_virtio64(vscsi->vdev, (unsigned long)sc),
722	};
723	return virtscsi_tmf(vscsi, cmd);
724}
725
726static void virtscsi_map_queues(struct Scsi_Host *shost)
727{
728	struct virtio_scsi *vscsi = shost_priv(shost);
729	int i, qoff;
730
731	for (i = 0, qoff = 0; i < shost->nr_maps; i++) {
732		struct blk_mq_queue_map *map = &shost->tag_set.map[i];
733
734		map->nr_queues = vscsi->io_queues[i];
735		map->queue_offset = qoff;
736		qoff += map->nr_queues;
737
738		if (map->nr_queues == 0)
739			continue;
740
741		/*
742		 * Regular queues have interrupts and hence CPU affinity is
743		 * defined by the core virtio code, but polling queues have
744		 * no interrupts so we let the block layer assign CPU affinity.
745		 */
746		if (i == HCTX_TYPE_POLL)
747			blk_mq_map_queues(map);
748		else
749			blk_mq_virtio_map_queues(map, vscsi->vdev, 2);
750	}
751}
752
753static int virtscsi_mq_poll(struct Scsi_Host *shost, unsigned int queue_num)
754{
755	struct virtio_scsi *vscsi = shost_priv(shost);
756	struct virtio_scsi_vq *virtscsi_vq = &vscsi->req_vqs[queue_num];
757	unsigned long flags;
758	unsigned int len;
759	int found = 0;
760	void *buf;
761
762	spin_lock_irqsave(&virtscsi_vq->vq_lock, flags);
763
764	while ((buf = virtqueue_get_buf(virtscsi_vq->vq, &len)) != NULL) {
765		virtscsi_complete_cmd(vscsi, buf);
766		found++;
767	}
768
769	spin_unlock_irqrestore(&virtscsi_vq->vq_lock, flags);
770
771	return found;
772}
773
774static void virtscsi_commit_rqs(struct Scsi_Host *shost, u16 hwq)
775{
776	struct virtio_scsi *vscsi = shost_priv(shost);
777
778	virtscsi_kick_vq(&vscsi->req_vqs[hwq]);
779}
780
781/*
782 * The host guarantees to respond to each command, although I/O
783 * latencies might be higher than on bare metal.  Reset the timer
784 * unconditionally to give the host a chance to perform EH.
785 */
786static enum scsi_timeout_action virtscsi_eh_timed_out(struct scsi_cmnd *scmnd)
787{
788	return SCSI_EH_RESET_TIMER;
789}
790
791static const struct scsi_host_template virtscsi_host_template = {
792	.module = THIS_MODULE,
793	.name = "Virtio SCSI HBA",
794	.proc_name = "virtio_scsi",
795	.this_id = -1,
796	.cmd_size = sizeof(struct virtio_scsi_cmd),
797	.queuecommand = virtscsi_queuecommand,
798	.mq_poll = virtscsi_mq_poll,
799	.commit_rqs = virtscsi_commit_rqs,
800	.change_queue_depth = virtscsi_change_queue_depth,
801	.eh_abort_handler = virtscsi_abort,
802	.eh_device_reset_handler = virtscsi_device_reset,
803	.eh_timed_out = virtscsi_eh_timed_out,
804	.slave_alloc = virtscsi_device_alloc,
805
806	.dma_boundary = UINT_MAX,
807	.map_queues = virtscsi_map_queues,
808	.track_queue_depth = 1,
809};
810
811#define virtscsi_config_get(vdev, fld) \
812	({ \
813		__virtio_native_type(struct virtio_scsi_config, fld) __val; \
814		virtio_cread(vdev, struct virtio_scsi_config, fld, &__val); \
815		__val; \
816	})
817
818#define virtscsi_config_set(vdev, fld, val) \
819	do { \
820		__virtio_native_type(struct virtio_scsi_config, fld) __val = (val); \
821		virtio_cwrite(vdev, struct virtio_scsi_config, fld, &__val); \
822	} while(0)
823
824static void virtscsi_init_vq(struct virtio_scsi_vq *virtscsi_vq,
825			     struct virtqueue *vq)
826{
827	spin_lock_init(&virtscsi_vq->vq_lock);
828	virtscsi_vq->vq = vq;
829}
830
831static void virtscsi_remove_vqs(struct virtio_device *vdev)
832{
833	/* Stop all the virtqueues. */
834	virtio_reset_device(vdev);
835	vdev->config->del_vqs(vdev);
836}
837
838static int virtscsi_init(struct virtio_device *vdev,
839			 struct virtio_scsi *vscsi)
840{
841	int err;
842	u32 i;
843	u32 num_vqs, num_poll_vqs, num_req_vqs;
844	vq_callback_t **callbacks;
845	const char **names;
846	struct virtqueue **vqs;
847	struct irq_affinity desc = { .pre_vectors = 2 };
848
849	num_req_vqs = vscsi->num_queues;
850	num_vqs = num_req_vqs + VIRTIO_SCSI_VQ_BASE;
851	vqs = kmalloc_array(num_vqs, sizeof(struct virtqueue *), GFP_KERNEL);
852	callbacks = kmalloc_array(num_vqs, sizeof(vq_callback_t *),
853				  GFP_KERNEL);
854	names = kmalloc_array(num_vqs, sizeof(char *), GFP_KERNEL);
855
856	if (!callbacks || !vqs || !names) {
857		err = -ENOMEM;
858		goto out;
859	}
860
861	num_poll_vqs = min_t(unsigned int, virtscsi_poll_queues,
862			     num_req_vqs - 1);
863	vscsi->io_queues[HCTX_TYPE_DEFAULT] = num_req_vqs - num_poll_vqs;
864	vscsi->io_queues[HCTX_TYPE_READ] = 0;
865	vscsi->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
866
867	dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
868		 vscsi->io_queues[HCTX_TYPE_DEFAULT],
869		 vscsi->io_queues[HCTX_TYPE_READ],
870		 vscsi->io_queues[HCTX_TYPE_POLL]);
871
872	callbacks[0] = virtscsi_ctrl_done;
873	callbacks[1] = virtscsi_event_done;
874	names[0] = "control";
875	names[1] = "event";
876	for (i = VIRTIO_SCSI_VQ_BASE; i < num_vqs - num_poll_vqs; i++) {
877		callbacks[i] = virtscsi_req_done;
878		names[i] = "request";
879	}
880
881	for (; i < num_vqs; i++) {
882		callbacks[i] = NULL;
883		names[i] = "request_poll";
884	}
885
886	/* Discover virtqueues and write information to configuration.  */
887	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
888	if (err)
889		goto out;
890
891	virtscsi_init_vq(&vscsi->ctrl_vq, vqs[0]);
892	virtscsi_init_vq(&vscsi->event_vq, vqs[1]);
893	for (i = VIRTIO_SCSI_VQ_BASE; i < num_vqs; i++)
894		virtscsi_init_vq(&vscsi->req_vqs[i - VIRTIO_SCSI_VQ_BASE],
895				 vqs[i]);
896
897	virtscsi_config_set(vdev, cdb_size, VIRTIO_SCSI_CDB_SIZE);
898	virtscsi_config_set(vdev, sense_size, VIRTIO_SCSI_SENSE_SIZE);
899
900	err = 0;
901
902out:
903	kfree(names);
904	kfree(callbacks);
905	kfree(vqs);
906	if (err)
907		virtscsi_remove_vqs(vdev);
908	return err;
909}
910
911static int virtscsi_probe(struct virtio_device *vdev)
912{
913	struct Scsi_Host *shost;
914	struct virtio_scsi *vscsi;
915	int err;
916	u32 sg_elems, num_targets;
917	u32 cmd_per_lun;
918	u32 num_queues;
919
920	if (!vdev->config->get) {
921		dev_err(&vdev->dev, "%s failure: config access disabled\n",
922			__func__);
923		return -EINVAL;
924	}
925
926	/* We need to know how many queues before we allocate. */
927	num_queues = virtscsi_config_get(vdev, num_queues) ? : 1;
928	num_queues = min_t(unsigned int, nr_cpu_ids, num_queues);
929
930	num_targets = virtscsi_config_get(vdev, max_target) + 1;
931
932	shost = scsi_host_alloc(&virtscsi_host_template,
933				struct_size(vscsi, req_vqs, num_queues));
934	if (!shost)
935		return -ENOMEM;
936
937	sg_elems = virtscsi_config_get(vdev, seg_max) ?: 1;
938	shost->sg_tablesize = sg_elems;
939	shost->nr_maps = 1;
940	vscsi = shost_priv(shost);
941	vscsi->vdev = vdev;
942	vscsi->num_queues = num_queues;
943	vdev->priv = shost;
944
945	err = virtscsi_init(vdev, vscsi);
946	if (err)
947		goto virtscsi_init_failed;
948
949	if (vscsi->io_queues[HCTX_TYPE_POLL])
950		shost->nr_maps = HCTX_TYPE_POLL + 1;
951
952	shost->can_queue = virtqueue_get_vring_size(vscsi->req_vqs[0].vq);
953
954	cmd_per_lun = virtscsi_config_get(vdev, cmd_per_lun) ?: 1;
955	shost->cmd_per_lun = min_t(u32, cmd_per_lun, shost->can_queue);
956	shost->max_sectors = virtscsi_config_get(vdev, max_sectors) ?: 0xFFFF;
957
958	/* LUNs > 256 are reported with format 1, so they go in the range
959	 * 16640-32767.
960	 */
961	shost->max_lun = virtscsi_config_get(vdev, max_lun) + 1 + 0x4000;
962	shost->max_id = num_targets;
963	shost->max_channel = 0;
964	shost->max_cmd_len = VIRTIO_SCSI_CDB_SIZE;
965	shost->nr_hw_queues = num_queues;
966
967#ifdef CONFIG_BLK_DEV_INTEGRITY
968	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_T10_PI)) {
969		int host_prot;
970
971		host_prot = SHOST_DIF_TYPE1_PROTECTION | SHOST_DIF_TYPE2_PROTECTION |
972			    SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE1_PROTECTION |
973			    SHOST_DIX_TYPE2_PROTECTION | SHOST_DIX_TYPE3_PROTECTION;
974
975		scsi_host_set_prot(shost, host_prot);
976		scsi_host_set_guard(shost, SHOST_DIX_GUARD_CRC);
977	}
978#endif
979
980	err = scsi_add_host(shost, &vdev->dev);
981	if (err)
982		goto scsi_add_host_failed;
983
984	virtio_device_ready(vdev);
985
986	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG))
987		virtscsi_kick_event_all(vscsi);
988
989	scsi_scan_host(shost);
990	return 0;
991
992scsi_add_host_failed:
993	vdev->config->del_vqs(vdev);
994virtscsi_init_failed:
995	scsi_host_put(shost);
996	return err;
997}
998
999static void virtscsi_remove(struct virtio_device *vdev)
1000{
1001	struct Scsi_Host *shost = virtio_scsi_host(vdev);
1002	struct virtio_scsi *vscsi = shost_priv(shost);
1003
1004	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG))
1005		virtscsi_cancel_event_work(vscsi);
1006
1007	scsi_remove_host(shost);
1008	virtscsi_remove_vqs(vdev);
1009	scsi_host_put(shost);
1010}
1011
1012#ifdef CONFIG_PM_SLEEP
1013static int virtscsi_freeze(struct virtio_device *vdev)
1014{
1015	virtscsi_remove_vqs(vdev);
1016	return 0;
1017}
1018
1019static int virtscsi_restore(struct virtio_device *vdev)
1020{
1021	struct Scsi_Host *sh = virtio_scsi_host(vdev);
1022	struct virtio_scsi *vscsi = shost_priv(sh);
1023	int err;
1024
1025	err = virtscsi_init(vdev, vscsi);
1026	if (err)
1027		return err;
1028
1029	virtio_device_ready(vdev);
1030
1031	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG))
1032		virtscsi_kick_event_all(vscsi);
1033
1034	return err;
1035}
1036#endif
1037
1038static struct virtio_device_id id_table[] = {
1039	{ VIRTIO_ID_SCSI, VIRTIO_DEV_ANY_ID },
1040	{ 0 },
1041};
1042
1043static unsigned int features[] = {
1044	VIRTIO_SCSI_F_HOTPLUG,
1045	VIRTIO_SCSI_F_CHANGE,
1046#ifdef CONFIG_BLK_DEV_INTEGRITY
1047	VIRTIO_SCSI_F_T10_PI,
1048#endif
1049};
1050
1051static struct virtio_driver virtio_scsi_driver = {
1052	.feature_table = features,
1053	.feature_table_size = ARRAY_SIZE(features),
1054	.driver.name = KBUILD_MODNAME,
1055	.driver.owner = THIS_MODULE,
1056	.id_table = id_table,
1057	.probe = virtscsi_probe,
1058#ifdef CONFIG_PM_SLEEP
1059	.freeze = virtscsi_freeze,
1060	.restore = virtscsi_restore,
1061#endif
1062	.remove = virtscsi_remove,
1063};
1064
1065static int __init virtio_scsi_init(void)
1066{
1067	int ret = -ENOMEM;
1068
1069	virtscsi_cmd_cache = KMEM_CACHE(virtio_scsi_cmd, 0);
1070	if (!virtscsi_cmd_cache) {
1071		pr_err("kmem_cache_create() for virtscsi_cmd_cache failed\n");
1072		goto error;
1073	}
1074
1075
1076	virtscsi_cmd_pool =
1077		mempool_create_slab_pool(VIRTIO_SCSI_MEMPOOL_SZ,
1078					 virtscsi_cmd_cache);
1079	if (!virtscsi_cmd_pool) {
1080		pr_err("mempool_create() for virtscsi_cmd_pool failed\n");
1081		goto error;
1082	}
1083	ret = register_virtio_driver(&virtio_scsi_driver);
1084	if (ret < 0)
1085		goto error;
1086
1087	return 0;
1088
1089error:
1090	mempool_destroy(virtscsi_cmd_pool);
1091	virtscsi_cmd_pool = NULL;
1092	kmem_cache_destroy(virtscsi_cmd_cache);
1093	virtscsi_cmd_cache = NULL;
1094	return ret;
1095}
1096
1097static void __exit virtio_scsi_fini(void)
1098{
1099	unregister_virtio_driver(&virtio_scsi_driver);
1100	mempool_destroy(virtscsi_cmd_pool);
1101	kmem_cache_destroy(virtscsi_cmd_cache);
1102}
1103module_init(virtio_scsi_init);
1104module_exit(virtio_scsi_fini);
1105
1106MODULE_DEVICE_TABLE(virtio, id_table);
1107MODULE_DESCRIPTION("Virtio SCSI HBA driver");
1108MODULE_LICENSE("GPL");
1109