xref: /openbmc/linux/drivers/scsi/xen-scsifront.c (revision a9ebfc40)
1 /*
2  * Xen SCSI frontend driver
3  *
4  * Copyright (c) 2008, FUJITSU Limited
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License version 2
8  * as published by the Free Software Foundation; or, when distributed
9  * separately from the Linux kernel or incorporated into other
10  * software packages, subject to the following license:
11  *
12  * Permission is hereby granted, free of charge, to any person obtaining a copy
13  * of this source file (the "Software"), to deal in the Software without
14  * restriction, including without limitation the rights to use, copy, modify,
15  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
16  * and to permit persons to whom the Software is furnished to do so, subject to
17  * the following conditions:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28  * IN THE SOFTWARE.
29  */
30 
31 #include <linux/module.h>
32 #include <linux/kernel.h>
33 #include <linux/device.h>
34 #include <linux/wait.h>
35 #include <linux/interrupt.h>
36 #include <linux/mutex.h>
37 #include <linux/spinlock.h>
38 #include <linux/sched.h>
39 #include <linux/blkdev.h>
40 #include <linux/pfn.h>
41 #include <linux/slab.h>
42 #include <linux/bitops.h>
43 
44 #include <scsi/scsi_cmnd.h>
45 #include <scsi/scsi_device.h>
46 #include <scsi/scsi.h>
47 #include <scsi/scsi_host.h>
48 
49 #include <xen/xen.h>
50 #include <xen/xenbus.h>
51 #include <xen/grant_table.h>
52 #include <xen/events.h>
53 #include <xen/page.h>
54 
55 #include <xen/interface/grant_table.h>
56 #include <xen/interface/io/vscsiif.h>
57 #include <xen/interface/io/protocols.h>
58 
59 #include <asm/xen/hypervisor.h>
60 
61 #define VSCSIFRONT_OP_ADD_LUN	1
62 #define VSCSIFRONT_OP_DEL_LUN	2
63 #define VSCSIFRONT_OP_READD_LUN	3
64 
65 /* Tuning point. */
66 #define VSCSIIF_DEFAULT_CMD_PER_LUN 10
67 #define VSCSIIF_MAX_TARGET          64
68 #define VSCSIIF_MAX_LUN             255
69 
70 #define VSCSIIF_RING_SIZE	__CONST_RING_SIZE(vscsiif, PAGE_SIZE)
71 #define VSCSIIF_MAX_REQS	VSCSIIF_RING_SIZE
72 
73 #define vscsiif_grants_sg(_sg)	(PFN_UP((_sg) *		\
74 				sizeof(struct scsiif_request_segment)))
75 
76 struct vscsifrnt_shadow {
77 	/* command between backend and frontend */
78 	unsigned char act;
79 	uint8_t nr_segments;
80 	uint16_t rqid;
81 	uint16_t ref_rqid;
82 
83 	bool inflight;
84 
85 	unsigned int nr_grants;		/* number of grants in gref[] */
86 	struct scsiif_request_segment *sg;	/* scatter/gather elements */
87 	struct scsiif_request_segment seg[VSCSIIF_SG_TABLESIZE];
88 
89 	/* Do reset or abort function. */
90 	wait_queue_head_t wq_reset;	/* reset work queue           */
91 	int wait_reset;			/* reset work queue condition */
92 	int32_t rslt_reset;		/* reset response status:     */
93 					/* SUCCESS or FAILED or:      */
94 #define RSLT_RESET_WAITING	0
95 #define RSLT_RESET_ERR		-1
96 
97 	/* Requested struct scsi_cmnd is stored from kernel. */
98 	struct scsi_cmnd *sc;
99 	int gref[vscsiif_grants_sg(SG_ALL) + SG_ALL];
100 };
101 
102 struct vscsifrnt_info {
103 	struct xenbus_device *dev;
104 
105 	struct Scsi_Host *host;
106 	enum {
107 		STATE_INACTIVE,
108 		STATE_ACTIVE,
109 		STATE_ERROR
110 	}  host_active;
111 
112 	unsigned int evtchn;
113 	unsigned int irq;
114 
115 	grant_ref_t ring_ref;
116 	struct vscsiif_front_ring ring;
117 	struct vscsiif_response	ring_rsp;
118 
119 	spinlock_t shadow_lock;
120 	DECLARE_BITMAP(shadow_free_bitmap, VSCSIIF_MAX_REQS);
121 	struct vscsifrnt_shadow *shadow[VSCSIIF_MAX_REQS];
122 
123 	/* Following items are protected by the host lock. */
124 	wait_queue_head_t wq_sync;
125 	wait_queue_head_t wq_pause;
126 	unsigned int wait_ring_available:1;
127 	unsigned int waiting_pause:1;
128 	unsigned int pause:1;
129 	unsigned callers;
130 
131 	char dev_state_path[64];
132 	struct task_struct *curr;
133 };
134 
135 static DEFINE_MUTEX(scsifront_mutex);
136 
137 static void scsifront_wake_up(struct vscsifrnt_info *info)
138 {
139 	info->wait_ring_available = 0;
140 	wake_up(&info->wq_sync);
141 }
142 
143 static int scsifront_get_rqid(struct vscsifrnt_info *info)
144 {
145 	unsigned long flags;
146 	int free;
147 
148 	spin_lock_irqsave(&info->shadow_lock, flags);
149 
150 	free = find_first_bit(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
151 	__clear_bit(free, info->shadow_free_bitmap);
152 
153 	spin_unlock_irqrestore(&info->shadow_lock, flags);
154 
155 	return free;
156 }
157 
158 static int _scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
159 {
160 	int empty = bitmap_empty(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
161 
162 	__set_bit(id, info->shadow_free_bitmap);
163 	info->shadow[id] = NULL;
164 
165 	return empty || info->wait_ring_available;
166 }
167 
168 static void scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
169 {
170 	unsigned long flags;
171 	int kick;
172 
173 	spin_lock_irqsave(&info->shadow_lock, flags);
174 	kick = _scsifront_put_rqid(info, id);
175 	spin_unlock_irqrestore(&info->shadow_lock, flags);
176 
177 	if (kick)
178 		scsifront_wake_up(info);
179 }
180 
181 static int scsifront_do_request(struct vscsifrnt_info *info,
182 				struct vscsifrnt_shadow *shadow)
183 {
184 	struct vscsiif_front_ring *ring = &(info->ring);
185 	struct vscsiif_request *ring_req;
186 	struct scsi_cmnd *sc = shadow->sc;
187 	uint32_t id;
188 	int i, notify;
189 
190 	if (RING_FULL(&info->ring))
191 		return -EBUSY;
192 
193 	id = scsifront_get_rqid(info);	/* use id in response */
194 	if (id >= VSCSIIF_MAX_REQS)
195 		return -EBUSY;
196 
197 	info->shadow[id] = shadow;
198 	shadow->rqid = id;
199 
200 	ring_req = RING_GET_REQUEST(&(info->ring), ring->req_prod_pvt);
201 	ring->req_prod_pvt++;
202 
203 	ring_req->rqid        = id;
204 	ring_req->act         = shadow->act;
205 	ring_req->ref_rqid    = shadow->ref_rqid;
206 	ring_req->nr_segments = shadow->nr_segments;
207 
208 	ring_req->id      = sc->device->id;
209 	ring_req->lun     = sc->device->lun;
210 	ring_req->channel = sc->device->channel;
211 	ring_req->cmd_len = sc->cmd_len;
212 
213 	BUG_ON(sc->cmd_len > VSCSIIF_MAX_COMMAND_SIZE);
214 
215 	memcpy(ring_req->cmnd, sc->cmnd, sc->cmd_len);
216 
217 	ring_req->sc_data_direction   = (uint8_t)sc->sc_data_direction;
218 	ring_req->timeout_per_command = scsi_cmd_to_rq(sc)->timeout / HZ;
219 
220 	for (i = 0; i < (shadow->nr_segments & ~VSCSIIF_SG_GRANT); i++)
221 		ring_req->seg[i] = shadow->seg[i];
222 
223 	shadow->inflight = true;
224 
225 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(ring, notify);
226 	if (notify)
227 		notify_remote_via_irq(info->irq);
228 
229 	return 0;
230 }
231 
232 static void scsifront_set_error(struct vscsifrnt_info *info, const char *msg)
233 {
234 	shost_printk(KERN_ERR, info->host, KBUILD_MODNAME "%s\n"
235 		     "Disabling device for further use\n", msg);
236 	info->host_active = STATE_ERROR;
237 }
238 
239 static void scsifront_gnttab_done(struct vscsifrnt_info *info,
240 				  struct vscsifrnt_shadow *shadow)
241 {
242 	int i;
243 
244 	if (shadow->sc->sc_data_direction == DMA_NONE)
245 		return;
246 
247 	for (i = 0; i < shadow->nr_grants; i++) {
248 		if (unlikely(!gnttab_try_end_foreign_access(shadow->gref[i]))) {
249 			scsifront_set_error(info, "grant still in use by backend");
250 			return;
251 		}
252 	}
253 
254 	kfree(shadow->sg);
255 }
256 
257 static unsigned int scsifront_host_byte(int32_t rslt)
258 {
259 	switch (XEN_VSCSIIF_RSLT_HOST(rslt)) {
260 	case XEN_VSCSIIF_RSLT_HOST_OK:
261 		return DID_OK;
262 	case XEN_VSCSIIF_RSLT_HOST_NO_CONNECT:
263 		return DID_NO_CONNECT;
264 	case XEN_VSCSIIF_RSLT_HOST_BUS_BUSY:
265 		return DID_BUS_BUSY;
266 	case XEN_VSCSIIF_RSLT_HOST_TIME_OUT:
267 		return DID_TIME_OUT;
268 	case XEN_VSCSIIF_RSLT_HOST_BAD_TARGET:
269 		return DID_BAD_TARGET;
270 	case XEN_VSCSIIF_RSLT_HOST_ABORT:
271 		return DID_ABORT;
272 	case XEN_VSCSIIF_RSLT_HOST_PARITY:
273 		return DID_PARITY;
274 	case XEN_VSCSIIF_RSLT_HOST_ERROR:
275 		return DID_ERROR;
276 	case XEN_VSCSIIF_RSLT_HOST_RESET:
277 		return DID_RESET;
278 	case XEN_VSCSIIF_RSLT_HOST_BAD_INTR:
279 		return DID_BAD_INTR;
280 	case XEN_VSCSIIF_RSLT_HOST_PASSTHROUGH:
281 		return DID_PASSTHROUGH;
282 	case XEN_VSCSIIF_RSLT_HOST_SOFT_ERROR:
283 		return DID_SOFT_ERROR;
284 	case XEN_VSCSIIF_RSLT_HOST_IMM_RETRY:
285 		return DID_IMM_RETRY;
286 	case XEN_VSCSIIF_RSLT_HOST_REQUEUE:
287 		return DID_REQUEUE;
288 	case XEN_VSCSIIF_RSLT_HOST_TRANSPORT_DISRUPTED:
289 		return DID_TRANSPORT_DISRUPTED;
290 	case XEN_VSCSIIF_RSLT_HOST_TRANSPORT_FAILFAST:
291 		return DID_TRANSPORT_FAILFAST;
292 	case XEN_VSCSIIF_RSLT_HOST_TRANSPORT_MARGINAL:
293 		return DID_TRANSPORT_MARGINAL;
294 	default:
295 		return DID_ERROR;
296 	}
297 }
298 
299 static void scsifront_cdb_cmd_done(struct vscsifrnt_info *info,
300 				   struct vscsiif_response *ring_rsp)
301 {
302 	struct vscsifrnt_shadow *shadow;
303 	struct scsi_cmnd *sc;
304 	uint32_t id;
305 	uint8_t sense_len;
306 
307 	id = ring_rsp->rqid;
308 	shadow = info->shadow[id];
309 	sc = shadow->sc;
310 
311 	BUG_ON(sc == NULL);
312 
313 	scsifront_gnttab_done(info, shadow);
314 	if (info->host_active == STATE_ERROR)
315 		return;
316 	scsifront_put_rqid(info, id);
317 
318 	set_host_byte(sc, scsifront_host_byte(ring_rsp->rslt));
319 	set_status_byte(sc, XEN_VSCSIIF_RSLT_STATUS(ring_rsp->rslt));
320 	scsi_set_resid(sc, ring_rsp->residual_len);
321 
322 	sense_len = min_t(uint8_t, VSCSIIF_SENSE_BUFFERSIZE,
323 			  ring_rsp->sense_len);
324 
325 	if (sense_len)
326 		memcpy(sc->sense_buffer, ring_rsp->sense_buffer, sense_len);
327 
328 	scsi_done(sc);
329 }
330 
331 static void scsifront_sync_cmd_done(struct vscsifrnt_info *info,
332 				    struct vscsiif_response *ring_rsp)
333 {
334 	uint16_t id = ring_rsp->rqid;
335 	unsigned long flags;
336 	struct vscsifrnt_shadow *shadow = info->shadow[id];
337 	int kick;
338 
339 	spin_lock_irqsave(&info->shadow_lock, flags);
340 	shadow->wait_reset = 1;
341 	switch (shadow->rslt_reset) {
342 	case RSLT_RESET_WAITING:
343 		if (ring_rsp->rslt == XEN_VSCSIIF_RSLT_RESET_SUCCESS)
344 			shadow->rslt_reset = SUCCESS;
345 		else
346 			shadow->rslt_reset = FAILED;
347 		break;
348 	case RSLT_RESET_ERR:
349 		kick = _scsifront_put_rqid(info, id);
350 		spin_unlock_irqrestore(&info->shadow_lock, flags);
351 		kfree(shadow);
352 		if (kick)
353 			scsifront_wake_up(info);
354 		return;
355 	default:
356 		scsifront_set_error(info, "bad reset state");
357 		break;
358 	}
359 	spin_unlock_irqrestore(&info->shadow_lock, flags);
360 
361 	wake_up(&shadow->wq_reset);
362 }
363 
364 static void scsifront_do_response(struct vscsifrnt_info *info,
365 				  struct vscsiif_response *ring_rsp)
366 {
367 	struct vscsifrnt_shadow *shadow;
368 
369 	if (ring_rsp->rqid >= VSCSIIF_MAX_REQS ||
370 	    !info->shadow[ring_rsp->rqid]->inflight) {
371 		scsifront_set_error(info, "illegal rqid returned by backend!");
372 		return;
373 	}
374 	shadow = info->shadow[ring_rsp->rqid];
375 	shadow->inflight = false;
376 
377 	if (shadow->act == VSCSIIF_ACT_SCSI_CDB)
378 		scsifront_cdb_cmd_done(info, ring_rsp);
379 	else
380 		scsifront_sync_cmd_done(info, ring_rsp);
381 }
382 
383 static int scsifront_ring_drain(struct vscsifrnt_info *info,
384 				unsigned int *eoiflag)
385 {
386 	struct vscsiif_response ring_rsp;
387 	RING_IDX i, rp;
388 	int more_to_do = 0;
389 
390 	rp = READ_ONCE(info->ring.sring->rsp_prod);
391 	virt_rmb();	/* ordering required respective to backend */
392 	if (RING_RESPONSE_PROD_OVERFLOW(&info->ring, rp)) {
393 		scsifront_set_error(info, "illegal number of responses");
394 		return 0;
395 	}
396 	for (i = info->ring.rsp_cons; i != rp; i++) {
397 		RING_COPY_RESPONSE(&info->ring, i, &ring_rsp);
398 		scsifront_do_response(info, &ring_rsp);
399 		if (info->host_active == STATE_ERROR)
400 			return 0;
401 		*eoiflag &= ~XEN_EOI_FLAG_SPURIOUS;
402 	}
403 
404 	info->ring.rsp_cons = i;
405 
406 	if (i != info->ring.req_prod_pvt)
407 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
408 	else
409 		info->ring.sring->rsp_event = i + 1;
410 
411 	return more_to_do;
412 }
413 
414 static int scsifront_cmd_done(struct vscsifrnt_info *info,
415 			      unsigned int *eoiflag)
416 {
417 	int more_to_do;
418 	unsigned long flags;
419 
420 	spin_lock_irqsave(info->host->host_lock, flags);
421 
422 	more_to_do = scsifront_ring_drain(info, eoiflag);
423 
424 	info->wait_ring_available = 0;
425 
426 	spin_unlock_irqrestore(info->host->host_lock, flags);
427 
428 	wake_up(&info->wq_sync);
429 
430 	return more_to_do;
431 }
432 
433 static irqreturn_t scsifront_irq_fn(int irq, void *dev_id)
434 {
435 	struct vscsifrnt_info *info = dev_id;
436 	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
437 
438 	if (info->host_active == STATE_ERROR) {
439 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
440 		return IRQ_HANDLED;
441 	}
442 
443 	while (scsifront_cmd_done(info, &eoiflag))
444 		/* Yield point for this unbounded loop. */
445 		cond_resched();
446 
447 	xen_irq_lateeoi(irq, eoiflag);
448 
449 	return IRQ_HANDLED;
450 }
451 
452 static void scsifront_finish_all(struct vscsifrnt_info *info)
453 {
454 	unsigned int i, dummy;
455 	struct vscsiif_response resp;
456 
457 	scsifront_ring_drain(info, &dummy);
458 
459 	for (i = 0; i < VSCSIIF_MAX_REQS; i++) {
460 		if (test_bit(i, info->shadow_free_bitmap))
461 			continue;
462 		resp.rqid = i;
463 		resp.sense_len = 0;
464 		resp.rslt = DID_RESET << 16;
465 		resp.residual_len = 0;
466 		scsifront_do_response(info, &resp);
467 	}
468 }
469 
470 static int map_data_for_request(struct vscsifrnt_info *info,
471 				struct scsi_cmnd *sc,
472 				struct vscsifrnt_shadow *shadow)
473 {
474 	grant_ref_t gref_head;
475 	struct page *page;
476 	int err, ref, ref_cnt = 0;
477 	int grant_ro = (sc->sc_data_direction == DMA_TO_DEVICE);
478 	unsigned int i, off, len, bytes;
479 	unsigned int data_len = scsi_bufflen(sc);
480 	unsigned int data_grants = 0, seg_grants = 0;
481 	struct scatterlist *sg;
482 	struct scsiif_request_segment *seg;
483 
484 	if (sc->sc_data_direction == DMA_NONE || !data_len)
485 		return 0;
486 
487 	scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i)
488 		data_grants += PFN_UP(sg->offset + sg->length);
489 
490 	if (data_grants > VSCSIIF_SG_TABLESIZE) {
491 		if (data_grants > info->host->sg_tablesize) {
492 			shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
493 			     "Unable to map request_buffer for command!\n");
494 			return -E2BIG;
495 		}
496 		seg_grants = vscsiif_grants_sg(data_grants);
497 		shadow->sg = kcalloc(data_grants,
498 			sizeof(struct scsiif_request_segment), GFP_ATOMIC);
499 		if (!shadow->sg)
500 			return -ENOMEM;
501 	}
502 	seg = shadow->sg ? : shadow->seg;
503 
504 	err = gnttab_alloc_grant_references(seg_grants + data_grants,
505 					    &gref_head);
506 	if (err) {
507 		kfree(shadow->sg);
508 		shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
509 			     "gnttab_alloc_grant_references() error\n");
510 		return -ENOMEM;
511 	}
512 
513 	if (seg_grants) {
514 		page = virt_to_page(seg);
515 		off = offset_in_page(seg);
516 		len = sizeof(struct scsiif_request_segment) * data_grants;
517 		while (len > 0) {
518 			bytes = min_t(unsigned int, len, PAGE_SIZE - off);
519 
520 			ref = gnttab_claim_grant_reference(&gref_head);
521 			BUG_ON(ref == -ENOSPC);
522 
523 			gnttab_grant_foreign_access_ref(ref,
524 				info->dev->otherend_id,
525 				xen_page_to_gfn(page), 1);
526 			shadow->gref[ref_cnt] = ref;
527 			shadow->seg[ref_cnt].gref   = ref;
528 			shadow->seg[ref_cnt].offset = (uint16_t)off;
529 			shadow->seg[ref_cnt].length = (uint16_t)bytes;
530 
531 			page++;
532 			len -= bytes;
533 			off = 0;
534 			ref_cnt++;
535 		}
536 		BUG_ON(seg_grants < ref_cnt);
537 		seg_grants = ref_cnt;
538 	}
539 
540 	scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) {
541 		page = sg_page(sg);
542 		off = sg->offset;
543 		len = sg->length;
544 
545 		while (len > 0 && data_len > 0) {
546 			/*
547 			 * sg sends a scatterlist that is larger than
548 			 * the data_len it wants transferred for certain
549 			 * IO sizes.
550 			 */
551 			bytes = min_t(unsigned int, len, PAGE_SIZE - off);
552 			bytes = min(bytes, data_len);
553 
554 			ref = gnttab_claim_grant_reference(&gref_head);
555 			BUG_ON(ref == -ENOSPC);
556 
557 			gnttab_grant_foreign_access_ref(ref,
558 				info->dev->otherend_id,
559 				xen_page_to_gfn(page),
560 				grant_ro);
561 
562 			shadow->gref[ref_cnt] = ref;
563 			seg->gref   = ref;
564 			seg->offset = (uint16_t)off;
565 			seg->length = (uint16_t)bytes;
566 
567 			page++;
568 			seg++;
569 			len -= bytes;
570 			data_len -= bytes;
571 			off = 0;
572 			ref_cnt++;
573 		}
574 	}
575 
576 	if (seg_grants)
577 		shadow->nr_segments = VSCSIIF_SG_GRANT | seg_grants;
578 	else
579 		shadow->nr_segments = (uint8_t)ref_cnt;
580 	shadow->nr_grants = ref_cnt;
581 
582 	return 0;
583 }
584 
585 static int scsifront_enter(struct vscsifrnt_info *info)
586 {
587 	if (info->pause)
588 		return 1;
589 	info->callers++;
590 	return 0;
591 }
592 
593 static void scsifront_return(struct vscsifrnt_info *info)
594 {
595 	info->callers--;
596 	if (info->callers)
597 		return;
598 
599 	if (!info->waiting_pause)
600 		return;
601 
602 	info->waiting_pause = 0;
603 	wake_up(&info->wq_pause);
604 }
605 
606 static int scsifront_queuecommand(struct Scsi_Host *shost,
607 				  struct scsi_cmnd *sc)
608 {
609 	struct vscsifrnt_info *info = shost_priv(shost);
610 	struct vscsifrnt_shadow *shadow = scsi_cmd_priv(sc);
611 	unsigned long flags;
612 	int err;
613 
614 	if (info->host_active == STATE_ERROR)
615 		return SCSI_MLQUEUE_HOST_BUSY;
616 
617 	sc->result = 0;
618 
619 	shadow->sc  = sc;
620 	shadow->act = VSCSIIF_ACT_SCSI_CDB;
621 
622 	spin_lock_irqsave(shost->host_lock, flags);
623 	if (scsifront_enter(info)) {
624 		spin_unlock_irqrestore(shost->host_lock, flags);
625 		return SCSI_MLQUEUE_HOST_BUSY;
626 	}
627 
628 	err = map_data_for_request(info, sc, shadow);
629 	if (err < 0) {
630 		pr_debug("%s: err %d\n", __func__, err);
631 		scsifront_return(info);
632 		spin_unlock_irqrestore(shost->host_lock, flags);
633 		if (err == -ENOMEM)
634 			return SCSI_MLQUEUE_HOST_BUSY;
635 		sc->result = DID_ERROR << 16;
636 		scsi_done(sc);
637 		return 0;
638 	}
639 
640 	if (scsifront_do_request(info, shadow)) {
641 		scsifront_gnttab_done(info, shadow);
642 		goto busy;
643 	}
644 
645 	scsifront_return(info);
646 	spin_unlock_irqrestore(shost->host_lock, flags);
647 
648 	return 0;
649 
650 busy:
651 	scsifront_return(info);
652 	spin_unlock_irqrestore(shost->host_lock, flags);
653 	pr_debug("%s: busy\n", __func__);
654 	return SCSI_MLQUEUE_HOST_BUSY;
655 }
656 
657 /*
658  * Any exception handling (reset or abort) must be forwarded to the backend.
659  * We have to wait until an answer is returned. This answer contains the
660  * result to be returned to the requestor.
661  */
662 static int scsifront_action_handler(struct scsi_cmnd *sc, uint8_t act)
663 {
664 	struct Scsi_Host *host = sc->device->host;
665 	struct vscsifrnt_info *info = shost_priv(host);
666 	struct vscsifrnt_shadow *shadow, *s = scsi_cmd_priv(sc);
667 	int err = 0;
668 
669 	if (info->host_active == STATE_ERROR)
670 		return FAILED;
671 
672 	shadow = kzalloc(sizeof(*shadow), GFP_NOIO);
673 	if (!shadow)
674 		return FAILED;
675 
676 	shadow->act = act;
677 	shadow->rslt_reset = RSLT_RESET_WAITING;
678 	shadow->sc = sc;
679 	shadow->ref_rqid = s->rqid;
680 	init_waitqueue_head(&shadow->wq_reset);
681 
682 	spin_lock_irq(host->host_lock);
683 
684 	for (;;) {
685 		if (scsifront_enter(info))
686 			goto fail;
687 
688 		if (!scsifront_do_request(info, shadow))
689 			break;
690 
691 		scsifront_return(info);
692 		if (err)
693 			goto fail;
694 		info->wait_ring_available = 1;
695 		spin_unlock_irq(host->host_lock);
696 		err = wait_event_interruptible(info->wq_sync,
697 					       !info->wait_ring_available);
698 		spin_lock_irq(host->host_lock);
699 	}
700 
701 	spin_unlock_irq(host->host_lock);
702 	err = wait_event_interruptible(shadow->wq_reset, shadow->wait_reset);
703 	spin_lock_irq(host->host_lock);
704 
705 	if (!err) {
706 		err = shadow->rslt_reset;
707 		scsifront_put_rqid(info, shadow->rqid);
708 		kfree(shadow);
709 	} else {
710 		spin_lock(&info->shadow_lock);
711 		shadow->rslt_reset = RSLT_RESET_ERR;
712 		spin_unlock(&info->shadow_lock);
713 		err = FAILED;
714 	}
715 
716 	scsifront_return(info);
717 	spin_unlock_irq(host->host_lock);
718 	return err;
719 
720 fail:
721 	spin_unlock_irq(host->host_lock);
722 	kfree(shadow);
723 	return FAILED;
724 }
725 
726 static int scsifront_eh_abort_handler(struct scsi_cmnd *sc)
727 {
728 	pr_debug("%s\n", __func__);
729 	return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_ABORT);
730 }
731 
732 static int scsifront_dev_reset_handler(struct scsi_cmnd *sc)
733 {
734 	pr_debug("%s\n", __func__);
735 	return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_RESET);
736 }
737 
738 static int scsifront_sdev_configure(struct scsi_device *sdev)
739 {
740 	struct vscsifrnt_info *info = shost_priv(sdev->host);
741 	int err;
742 
743 	if (info->host_active == STATE_ERROR)
744 		return -EIO;
745 
746 	if (current == info->curr) {
747 		err = xenbus_printf(XBT_NIL, info->dev->nodename,
748 			      info->dev_state_path, "%d", XenbusStateConnected);
749 		if (err) {
750 			xenbus_dev_error(info->dev, err,
751 				"%s: writing dev_state_path", __func__);
752 			return err;
753 		}
754 	}
755 
756 	return 0;
757 }
758 
759 static void scsifront_sdev_destroy(struct scsi_device *sdev)
760 {
761 	struct vscsifrnt_info *info = shost_priv(sdev->host);
762 	int err;
763 
764 	if (current == info->curr) {
765 		err = xenbus_printf(XBT_NIL, info->dev->nodename,
766 			      info->dev_state_path, "%d", XenbusStateClosed);
767 		if (err)
768 			xenbus_dev_error(info->dev, err,
769 				"%s: writing dev_state_path", __func__);
770 	}
771 }
772 
773 static const struct scsi_host_template scsifront_sht = {
774 	.module			= THIS_MODULE,
775 	.name			= "Xen SCSI frontend driver",
776 	.queuecommand		= scsifront_queuecommand,
777 	.eh_abort_handler	= scsifront_eh_abort_handler,
778 	.eh_device_reset_handler = scsifront_dev_reset_handler,
779 	.slave_configure	= scsifront_sdev_configure,
780 	.slave_destroy		= scsifront_sdev_destroy,
781 	.cmd_per_lun		= VSCSIIF_DEFAULT_CMD_PER_LUN,
782 	.can_queue		= VSCSIIF_MAX_REQS,
783 	.this_id		= -1,
784 	.cmd_size		= sizeof(struct vscsifrnt_shadow),
785 	.sg_tablesize		= VSCSIIF_SG_TABLESIZE,
786 	.proc_name		= "scsifront",
787 };
788 
789 static int scsifront_alloc_ring(struct vscsifrnt_info *info)
790 {
791 	struct xenbus_device *dev = info->dev;
792 	struct vscsiif_sring *sring;
793 	int err;
794 
795 	/***** Frontend to Backend ring start *****/
796 	err = xenbus_setup_ring(dev, GFP_KERNEL, (void **)&sring, 1,
797 				&info->ring_ref);
798 	if (err)
799 		return err;
800 
801 	XEN_FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
802 
803 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
804 	if (err) {
805 		xenbus_dev_fatal(dev, err, "xenbus_alloc_evtchn");
806 		goto free_gnttab;
807 	}
808 
809 	err = bind_evtchn_to_irq_lateeoi(info->evtchn);
810 	if (err <= 0) {
811 		xenbus_dev_fatal(dev, err, "bind_evtchn_to_irq");
812 		goto free_gnttab;
813 	}
814 
815 	info->irq = err;
816 
817 	err = request_threaded_irq(info->irq, NULL, scsifront_irq_fn,
818 				   IRQF_ONESHOT, "scsifront", info);
819 	if (err) {
820 		xenbus_dev_fatal(dev, err, "request_threaded_irq");
821 		goto free_irq;
822 	}
823 
824 	return 0;
825 
826 /* free resource */
827 free_irq:
828 	unbind_from_irqhandler(info->irq, info);
829 free_gnttab:
830 	xenbus_teardown_ring((void **)&sring, 1, &info->ring_ref);
831 
832 	return err;
833 }
834 
835 static void scsifront_free_ring(struct vscsifrnt_info *info)
836 {
837 	unbind_from_irqhandler(info->irq, info);
838 	xenbus_teardown_ring((void **)&info->ring.sring, 1, &info->ring_ref);
839 }
840 
841 static int scsifront_init_ring(struct vscsifrnt_info *info)
842 {
843 	struct xenbus_device *dev = info->dev;
844 	struct xenbus_transaction xbt;
845 	int err;
846 
847 	pr_debug("%s\n", __func__);
848 
849 	err = scsifront_alloc_ring(info);
850 	if (err)
851 		return err;
852 	pr_debug("%s: %u %u\n", __func__, info->ring_ref, info->evtchn);
853 
854 again:
855 	err = xenbus_transaction_start(&xbt);
856 	if (err)
857 		xenbus_dev_fatal(dev, err, "starting transaction");
858 
859 	err = xenbus_printf(xbt, dev->nodename, "ring-ref", "%u",
860 			    info->ring_ref);
861 	if (err) {
862 		xenbus_dev_fatal(dev, err, "%s", "writing ring-ref");
863 		goto fail;
864 	}
865 
866 	err = xenbus_printf(xbt, dev->nodename, "event-channel", "%u",
867 			    info->evtchn);
868 
869 	if (err) {
870 		xenbus_dev_fatal(dev, err, "%s", "writing event-channel");
871 		goto fail;
872 	}
873 
874 	err = xenbus_transaction_end(xbt, 0);
875 	if (err) {
876 		if (err == -EAGAIN)
877 			goto again;
878 		xenbus_dev_fatal(dev, err, "completing transaction");
879 		goto free_sring;
880 	}
881 
882 	return 0;
883 
884 fail:
885 	xenbus_transaction_end(xbt, 1);
886 free_sring:
887 	scsifront_free_ring(info);
888 
889 	return err;
890 }
891 
892 
893 static int scsifront_probe(struct xenbus_device *dev,
894 			   const struct xenbus_device_id *id)
895 {
896 	struct vscsifrnt_info *info;
897 	struct Scsi_Host *host;
898 	int err = -ENOMEM;
899 	char name[TASK_COMM_LEN];
900 
901 	host = scsi_host_alloc(&scsifront_sht, sizeof(*info));
902 	if (!host) {
903 		xenbus_dev_fatal(dev, err, "fail to allocate scsi host");
904 		return err;
905 	}
906 	info = shost_priv(host);
907 
908 	dev_set_drvdata(&dev->dev, info);
909 	info->dev = dev;
910 
911 	bitmap_fill(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
912 
913 	err = scsifront_init_ring(info);
914 	if (err) {
915 		scsi_host_put(host);
916 		return err;
917 	}
918 
919 	init_waitqueue_head(&info->wq_sync);
920 	init_waitqueue_head(&info->wq_pause);
921 	spin_lock_init(&info->shadow_lock);
922 
923 	snprintf(name, TASK_COMM_LEN, "vscsiif.%d", host->host_no);
924 
925 	host->max_id      = VSCSIIF_MAX_TARGET;
926 	host->max_channel = 0;
927 	host->max_lun     = VSCSIIF_MAX_LUN;
928 	host->max_sectors = (host->sg_tablesize - 1) * PAGE_SIZE / 512;
929 	host->max_cmd_len = VSCSIIF_MAX_COMMAND_SIZE;
930 
931 	err = scsi_add_host(host, &dev->dev);
932 	if (err) {
933 		dev_err(&dev->dev, "fail to add scsi host %d\n", err);
934 		goto free_sring;
935 	}
936 	info->host = host;
937 	info->host_active = STATE_ACTIVE;
938 
939 	xenbus_switch_state(dev, XenbusStateInitialised);
940 
941 	return 0;
942 
943 free_sring:
944 	scsifront_free_ring(info);
945 	scsi_host_put(host);
946 	return err;
947 }
948 
949 static int scsifront_resume(struct xenbus_device *dev)
950 {
951 	struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
952 	struct Scsi_Host *host = info->host;
953 	int err;
954 
955 	spin_lock_irq(host->host_lock);
956 
957 	/* Finish all still pending commands. */
958 	scsifront_finish_all(info);
959 
960 	spin_unlock_irq(host->host_lock);
961 
962 	/* Reconnect to dom0. */
963 	scsifront_free_ring(info);
964 	err = scsifront_init_ring(info);
965 	if (err) {
966 		dev_err(&dev->dev, "fail to resume %d\n", err);
967 		scsi_host_put(host);
968 		return err;
969 	}
970 
971 	xenbus_switch_state(dev, XenbusStateInitialised);
972 
973 	return 0;
974 }
975 
976 static int scsifront_suspend(struct xenbus_device *dev)
977 {
978 	struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
979 	struct Scsi_Host *host = info->host;
980 	int err = 0;
981 
982 	/* No new commands for the backend. */
983 	spin_lock_irq(host->host_lock);
984 	info->pause = 1;
985 	while (info->callers && !err) {
986 		info->waiting_pause = 1;
987 		info->wait_ring_available = 0;
988 		spin_unlock_irq(host->host_lock);
989 		wake_up(&info->wq_sync);
990 		err = wait_event_interruptible(info->wq_pause,
991 					       !info->waiting_pause);
992 		spin_lock_irq(host->host_lock);
993 	}
994 	spin_unlock_irq(host->host_lock);
995 	return err;
996 }
997 
998 static void scsifront_remove(struct xenbus_device *dev)
999 {
1000 	struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
1001 
1002 	pr_debug("%s: %s removed\n", __func__, dev->nodename);
1003 
1004 	mutex_lock(&scsifront_mutex);
1005 	if (info->host_active != STATE_INACTIVE) {
1006 		/* Scsi_host not yet removed */
1007 		scsi_remove_host(info->host);
1008 		info->host_active = STATE_INACTIVE;
1009 	}
1010 	mutex_unlock(&scsifront_mutex);
1011 
1012 	scsifront_free_ring(info);
1013 	scsi_host_put(info->host);
1014 }
1015 
1016 static void scsifront_disconnect(struct vscsifrnt_info *info)
1017 {
1018 	struct xenbus_device *dev = info->dev;
1019 	struct Scsi_Host *host = info->host;
1020 
1021 	pr_debug("%s: %s disconnect\n", __func__, dev->nodename);
1022 
1023 	/*
1024 	 * When this function is executed, all devices of
1025 	 * Frontend have been deleted.
1026 	 * Therefore, it need not block I/O before remove_host.
1027 	 */
1028 
1029 	mutex_lock(&scsifront_mutex);
1030 	if (info->host_active != STATE_INACTIVE) {
1031 		scsi_remove_host(host);
1032 		info->host_active = STATE_INACTIVE;
1033 	}
1034 	mutex_unlock(&scsifront_mutex);
1035 
1036 	xenbus_frontend_closed(dev);
1037 }
1038 
1039 static void scsifront_do_lun_hotplug(struct vscsifrnt_info *info, int op)
1040 {
1041 	struct xenbus_device *dev = info->dev;
1042 	int i, err = 0;
1043 	char str[64];
1044 	char **dir;
1045 	unsigned int dir_n = 0;
1046 	unsigned int device_state;
1047 	unsigned int hst, chn, tgt, lun;
1048 	struct scsi_device *sdev;
1049 
1050 	if (info->host_active == STATE_ERROR)
1051 		return;
1052 
1053 	dir = xenbus_directory(XBT_NIL, dev->otherend, "vscsi-devs", &dir_n);
1054 	if (IS_ERR(dir))
1055 		return;
1056 
1057 	/* mark current task as the one allowed to modify device states */
1058 	BUG_ON(info->curr);
1059 	info->curr = current;
1060 
1061 	for (i = 0; i < dir_n; i++) {
1062 		/* read status */
1063 		snprintf(str, sizeof(str), "vscsi-devs/%s/state", dir[i]);
1064 		err = xenbus_scanf(XBT_NIL, dev->otherend, str, "%u",
1065 				   &device_state);
1066 		if (XENBUS_EXIST_ERR(err))
1067 			continue;
1068 
1069 		/* virtual SCSI device */
1070 		snprintf(str, sizeof(str), "vscsi-devs/%s/v-dev", dir[i]);
1071 		err = xenbus_scanf(XBT_NIL, dev->otherend, str,
1072 				   "%u:%u:%u:%u", &hst, &chn, &tgt, &lun);
1073 		if (XENBUS_EXIST_ERR(err))
1074 			continue;
1075 
1076 		/*
1077 		 * Front device state path, used in slave_configure called
1078 		 * on successfull scsi_add_device, and in slave_destroy called
1079 		 * on remove of a device.
1080 		 */
1081 		snprintf(info->dev_state_path, sizeof(info->dev_state_path),
1082 			 "vscsi-devs/%s/state", dir[i]);
1083 
1084 		switch (op) {
1085 		case VSCSIFRONT_OP_ADD_LUN:
1086 			if (device_state != XenbusStateInitialised)
1087 				break;
1088 
1089 			if (scsi_add_device(info->host, chn, tgt, lun)) {
1090 				dev_err(&dev->dev, "scsi_add_device\n");
1091 				err = xenbus_printf(XBT_NIL, dev->nodename,
1092 					      info->dev_state_path,
1093 					      "%d", XenbusStateClosed);
1094 				if (err)
1095 					xenbus_dev_error(dev, err,
1096 						"%s: writing dev_state_path", __func__);
1097 			}
1098 			break;
1099 		case VSCSIFRONT_OP_DEL_LUN:
1100 			if (device_state != XenbusStateClosing)
1101 				break;
1102 
1103 			sdev = scsi_device_lookup(info->host, chn, tgt, lun);
1104 			if (sdev) {
1105 				scsi_remove_device(sdev);
1106 				scsi_device_put(sdev);
1107 			}
1108 			break;
1109 		case VSCSIFRONT_OP_READD_LUN:
1110 			if (device_state == XenbusStateConnected) {
1111 				err = xenbus_printf(XBT_NIL, dev->nodename,
1112 					      info->dev_state_path,
1113 					      "%d", XenbusStateConnected);
1114 				if (err)
1115 					xenbus_dev_error(dev, err,
1116 						"%s: writing dev_state_path", __func__);
1117 			}
1118 			break;
1119 		default:
1120 			break;
1121 		}
1122 	}
1123 
1124 	info->curr = NULL;
1125 
1126 	kfree(dir);
1127 }
1128 
1129 static void scsifront_read_backend_params(struct xenbus_device *dev,
1130 					  struct vscsifrnt_info *info)
1131 {
1132 	unsigned int sg_grant, nr_segs;
1133 	struct Scsi_Host *host = info->host;
1134 
1135 	sg_grant = xenbus_read_unsigned(dev->otherend, "feature-sg-grant", 0);
1136 	nr_segs = min_t(unsigned int, sg_grant, SG_ALL);
1137 	nr_segs = max_t(unsigned int, nr_segs, VSCSIIF_SG_TABLESIZE);
1138 	nr_segs = min_t(unsigned int, nr_segs,
1139 			VSCSIIF_SG_TABLESIZE * PAGE_SIZE /
1140 			sizeof(struct scsiif_request_segment));
1141 
1142 	if (!info->pause && sg_grant)
1143 		dev_info(&dev->dev, "using up to %d SG entries\n", nr_segs);
1144 	else if (info->pause && nr_segs < host->sg_tablesize)
1145 		dev_warn(&dev->dev,
1146 			 "SG entries decreased from %d to %u - device may not work properly anymore\n",
1147 			 host->sg_tablesize, nr_segs);
1148 
1149 	host->sg_tablesize = nr_segs;
1150 	host->max_sectors = (nr_segs - 1) * PAGE_SIZE / 512;
1151 }
1152 
1153 static void scsifront_backend_changed(struct xenbus_device *dev,
1154 				      enum xenbus_state backend_state)
1155 {
1156 	struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
1157 
1158 	pr_debug("%s: %p %u %u\n", __func__, dev, dev->state, backend_state);
1159 
1160 	switch (backend_state) {
1161 	case XenbusStateUnknown:
1162 	case XenbusStateInitialising:
1163 	case XenbusStateInitWait:
1164 	case XenbusStateInitialised:
1165 		break;
1166 
1167 	case XenbusStateConnected:
1168 		scsifront_read_backend_params(dev, info);
1169 
1170 		if (info->pause) {
1171 			scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_READD_LUN);
1172 			xenbus_switch_state(dev, XenbusStateConnected);
1173 			info->pause = 0;
1174 			return;
1175 		}
1176 
1177 		if (xenbus_read_driver_state(dev->nodename) ==
1178 		    XenbusStateInitialised)
1179 			scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
1180 
1181 		if (dev->state != XenbusStateConnected)
1182 			xenbus_switch_state(dev, XenbusStateConnected);
1183 		break;
1184 
1185 	case XenbusStateClosed:
1186 		if (dev->state == XenbusStateClosed)
1187 			break;
1188 		fallthrough;	/* Missed the backend's Closing state */
1189 	case XenbusStateClosing:
1190 		scsifront_disconnect(info);
1191 		break;
1192 
1193 	case XenbusStateReconfiguring:
1194 		scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_DEL_LUN);
1195 		xenbus_switch_state(dev, XenbusStateReconfiguring);
1196 		break;
1197 
1198 	case XenbusStateReconfigured:
1199 		scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
1200 		xenbus_switch_state(dev, XenbusStateConnected);
1201 		break;
1202 	}
1203 }
1204 
1205 static const struct xenbus_device_id scsifront_ids[] = {
1206 	{ "vscsi" },
1207 	{ "" }
1208 };
1209 
1210 static struct xenbus_driver scsifront_driver = {
1211 	.ids			= scsifront_ids,
1212 	.probe			= scsifront_probe,
1213 	.remove			= scsifront_remove,
1214 	.resume			= scsifront_resume,
1215 	.suspend		= scsifront_suspend,
1216 	.otherend_changed	= scsifront_backend_changed,
1217 };
1218 
1219 static int __init scsifront_init(void)
1220 {
1221 	if (!xen_domain())
1222 		return -ENODEV;
1223 
1224 	return xenbus_register_frontend(&scsifront_driver);
1225 }
1226 module_init(scsifront_init);
1227 
1228 static void __exit scsifront_exit(void)
1229 {
1230 	xenbus_unregister_driver(&scsifront_driver);
1231 }
1232 module_exit(scsifront_exit);
1233 
1234 MODULE_DESCRIPTION("Xen SCSI frontend driver");
1235 MODULE_LICENSE("GPL");
1236 MODULE_ALIAS("xen:vscsi");
1237 MODULE_AUTHOR("Juergen Gross <jgross@suse.com>");
1238