xref: /openbmc/linux/drivers/scsi/storvsc_drv.c (revision 42e22cac)
1 /*
2  * Copyright (c) 2009, Microsoft Corporation.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * You should have received a copy of the GNU General Public License along with
14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15  * Place - Suite 330, Boston, MA 02111-1307 USA.
16  *
17  * Authors:
18  *   Haiyang Zhang <haiyangz@microsoft.com>
19  *   Hank Janssen  <hjanssen@microsoft.com>
20  *   K. Y. Srinivasan <kys@microsoft.com>
21  */
22 
23 #include <linux/kernel.h>
24 #include <linux/wait.h>
25 #include <linux/sched.h>
26 #include <linux/completion.h>
27 #include <linux/string.h>
28 #include <linux/mm.h>
29 #include <linux/delay.h>
30 #include <linux/init.h>
31 #include <linux/slab.h>
32 #include <linux/module.h>
33 #include <linux/device.h>
34 #include <linux/hyperv.h>
35 #include <linux/mempool.h>
36 #include <scsi/scsi.h>
37 #include <scsi/scsi_cmnd.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_device.h>
40 #include <scsi/scsi_tcq.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_devinfo.h>
43 #include <scsi/scsi_dbg.h>
44 
45 /*
46  * All wire protocol details (storage protocol between the guest and the host)
47  * are consolidated here.
48  *
49  * Begin protocol definitions.
50  */
51 
52 /*
53  * Version history:
54  * V1 Beta: 0.1
55  * V1 RC < 2008/1/31: 1.0
56  * V1 RC > 2008/1/31:  2.0
57  * Win7: 4.2
58  */
59 
60 #define VMSTOR_CURRENT_MAJOR  4
61 #define VMSTOR_CURRENT_MINOR  2
62 
63 
64 /*  Packet structure describing virtual storage requests. */
65 enum vstor_packet_operation {
66 	VSTOR_OPERATION_COMPLETE_IO		= 1,
67 	VSTOR_OPERATION_REMOVE_DEVICE		= 2,
68 	VSTOR_OPERATION_EXECUTE_SRB		= 3,
69 	VSTOR_OPERATION_RESET_LUN		= 4,
70 	VSTOR_OPERATION_RESET_ADAPTER		= 5,
71 	VSTOR_OPERATION_RESET_BUS		= 6,
72 	VSTOR_OPERATION_BEGIN_INITIALIZATION	= 7,
73 	VSTOR_OPERATION_END_INITIALIZATION	= 8,
74 	VSTOR_OPERATION_QUERY_PROTOCOL_VERSION	= 9,
75 	VSTOR_OPERATION_QUERY_PROPERTIES	= 10,
76 	VSTOR_OPERATION_ENUMERATE_BUS		= 11,
77 	VSTOR_OPERATION_MAXIMUM			= 11
78 };
79 
80 /*
81  * Platform neutral description of a scsi request -
82  * this remains the same across the write regardless of 32/64 bit
83  * note: it's patterned off the SCSI_PASS_THROUGH structure
84  */
85 #define STORVSC_MAX_CMD_LEN			0x10
86 #define STORVSC_SENSE_BUFFER_SIZE		0x12
87 #define STORVSC_MAX_BUF_LEN_WITH_PADDING	0x14
88 
89 struct vmscsi_request {
90 	u16 length;
91 	u8 srb_status;
92 	u8 scsi_status;
93 
94 	u8  port_number;
95 	u8  path_id;
96 	u8  target_id;
97 	u8  lun;
98 
99 	u8  cdb_length;
100 	u8  sense_info_length;
101 	u8  data_in;
102 	u8  reserved;
103 
104 	u32 data_transfer_length;
105 
106 	union {
107 		u8 cdb[STORVSC_MAX_CMD_LEN];
108 		u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
109 		u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
110 	};
111 } __attribute((packed));
112 
113 
114 /*
115  * This structure is sent during the intialization phase to get the different
116  * properties of the channel.
117  */
118 struct vmstorage_channel_properties {
119 	u16 protocol_version;
120 	u8  path_id;
121 	u8 target_id;
122 
123 	/* Note: port number is only really known on the client side */
124 	u32  port_number;
125 	u32  flags;
126 	u32   max_transfer_bytes;
127 
128 	/*
129 	 * This id is unique for each channel and will correspond with
130 	 * vendor specific data in the inquiry data.
131 	 */
132 
133 	u64  unique_id;
134 } __packed;
135 
136 /*  This structure is sent during the storage protocol negotiations. */
137 struct vmstorage_protocol_version {
138 	/* Major (MSW) and minor (LSW) version numbers. */
139 	u16 major_minor;
140 
141 	/*
142 	 * Revision number is auto-incremented whenever this file is changed
143 	 * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
144 	 * definitely indicate incompatibility--but it does indicate mismatched
145 	 * builds.
146 	 * This is only used on the windows side. Just set it to 0.
147 	 */
148 	u16 revision;
149 } __packed;
150 
151 /* Channel Property Flags */
152 #define STORAGE_CHANNEL_REMOVABLE_FLAG		0x1
153 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG	0x2
154 
155 struct vstor_packet {
156 	/* Requested operation type */
157 	enum vstor_packet_operation operation;
158 
159 	/*  Flags - see below for values */
160 	u32 flags;
161 
162 	/* Status of the request returned from the server side. */
163 	u32 status;
164 
165 	/* Data payload area */
166 	union {
167 		/*
168 		 * Structure used to forward SCSI commands from the
169 		 * client to the server.
170 		 */
171 		struct vmscsi_request vm_srb;
172 
173 		/* Structure used to query channel properties. */
174 		struct vmstorage_channel_properties storage_channel_properties;
175 
176 		/* Used during version negotiations. */
177 		struct vmstorage_protocol_version version;
178 	};
179 } __packed;
180 
181 /*
182  * Packet Flags:
183  *
184  * This flag indicates that the server should send back a completion for this
185  * packet.
186  */
187 
188 #define REQUEST_COMPLETION_FLAG	0x1
189 
190 /* Matches Windows-end */
191 enum storvsc_request_type {
192 	WRITE_TYPE = 0,
193 	READ_TYPE,
194 	UNKNOWN_TYPE,
195 };
196 
197 /*
198  * SRB status codes and masks; a subset of the codes used here.
199  */
200 
201 #define SRB_STATUS_AUTOSENSE_VALID	0x80
202 #define SRB_STATUS_INVALID_LUN	0x20
203 #define SRB_STATUS_SUCCESS	0x01
204 #define SRB_STATUS_ERROR	0x04
205 
206 /*
207  * This is the end of Protocol specific defines.
208  */
209 
210 
211 /*
212  * We setup a mempool to allocate request structures for this driver
213  * on a per-lun basis. The following define specifies the number of
214  * elements in the pool.
215  */
216 
217 #define STORVSC_MIN_BUF_NR				64
218 static int storvsc_ringbuffer_size = (20 * PAGE_SIZE);
219 
220 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
221 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
222 
223 #define STORVSC_MAX_IO_REQUESTS				128
224 
225 /*
226  * In Hyper-V, each port/path/target maps to 1 scsi host adapter.  In
227  * reality, the path/target is not used (ie always set to 0) so our
228  * scsi host adapter essentially has 1 bus with 1 target that contains
229  * up to 256 luns.
230  */
231 #define STORVSC_MAX_LUNS_PER_TARGET			64
232 #define STORVSC_MAX_TARGETS				1
233 #define STORVSC_MAX_CHANNELS				1
234 
235 
236 
237 struct storvsc_cmd_request {
238 	struct list_head entry;
239 	struct scsi_cmnd *cmd;
240 
241 	unsigned int bounce_sgl_count;
242 	struct scatterlist *bounce_sgl;
243 
244 	struct hv_device *device;
245 
246 	/* Synchronize the request/response if needed */
247 	struct completion wait_event;
248 
249 	unsigned char *sense_buffer;
250 	struct hv_multipage_buffer data_buffer;
251 	struct vstor_packet vstor_packet;
252 };
253 
254 
255 /* A storvsc device is a device object that contains a vmbus channel */
256 struct storvsc_device {
257 	struct hv_device *device;
258 
259 	bool	 destroy;
260 	bool	 drain_notify;
261 	atomic_t num_outstanding_req;
262 	struct Scsi_Host *host;
263 
264 	wait_queue_head_t waiting_to_drain;
265 
266 	/*
267 	 * Each unique Port/Path/Target represents 1 channel ie scsi
268 	 * controller. In reality, the pathid, targetid is always 0
269 	 * and the port is set by us
270 	 */
271 	unsigned int port_number;
272 	unsigned char path_id;
273 	unsigned char target_id;
274 
275 	/* Used for vsc/vsp channel reset process */
276 	struct storvsc_cmd_request init_request;
277 	struct storvsc_cmd_request reset_request;
278 };
279 
280 struct stor_mem_pools {
281 	struct kmem_cache *request_pool;
282 	mempool_t *request_mempool;
283 };
284 
285 struct hv_host_device {
286 	struct hv_device *dev;
287 	unsigned int port;
288 	unsigned char path;
289 	unsigned char target;
290 };
291 
292 struct storvsc_scan_work {
293 	struct work_struct work;
294 	struct Scsi_Host *host;
295 	uint lun;
296 };
297 
298 static void storvsc_bus_scan(struct work_struct *work)
299 {
300 	struct storvsc_scan_work *wrk;
301 	int id, order_id;
302 
303 	wrk = container_of(work, struct storvsc_scan_work, work);
304 	for (id = 0; id < wrk->host->max_id; ++id) {
305 		if (wrk->host->reverse_ordering)
306 			order_id = wrk->host->max_id - id - 1;
307 		else
308 			order_id = id;
309 
310 		scsi_scan_target(&wrk->host->shost_gendev, 0,
311 				order_id, SCAN_WILD_CARD, 1);
312 	}
313 	kfree(wrk);
314 }
315 
316 static void storvsc_remove_lun(struct work_struct *work)
317 {
318 	struct storvsc_scan_work *wrk;
319 	struct scsi_device *sdev;
320 
321 	wrk = container_of(work, struct storvsc_scan_work, work);
322 	if (!scsi_host_get(wrk->host))
323 		goto done;
324 
325 	sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun);
326 
327 	if (sdev) {
328 		scsi_remove_device(sdev);
329 		scsi_device_put(sdev);
330 	}
331 	scsi_host_put(wrk->host);
332 
333 done:
334 	kfree(wrk);
335 }
336 
337 /*
338  * Major/minor macros.  Minor version is in LSB, meaning that earlier flat
339  * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1).
340  */
341 
342 static inline u16 storvsc_get_version(u8 major, u8 minor)
343 {
344 	u16 version;
345 
346 	version = ((major << 8) | minor);
347 	return version;
348 }
349 
350 /*
351  * We can get incoming messages from the host that are not in response to
352  * messages that we have sent out. An example of this would be messages
353  * received by the guest to notify dynamic addition/removal of LUNs. To
354  * deal with potential race conditions where the driver may be in the
355  * midst of being unloaded when we might receive an unsolicited message
356  * from the host, we have implemented a mechanism to gurantee sequential
357  * consistency:
358  *
359  * 1) Once the device is marked as being destroyed, we will fail all
360  *    outgoing messages.
361  * 2) We permit incoming messages when the device is being destroyed,
362  *    only to properly account for messages already sent out.
363  */
364 
365 static inline struct storvsc_device *get_out_stor_device(
366 					struct hv_device *device)
367 {
368 	struct storvsc_device *stor_device;
369 
370 	stor_device = hv_get_drvdata(device);
371 
372 	if (stor_device && stor_device->destroy)
373 		stor_device = NULL;
374 
375 	return stor_device;
376 }
377 
378 
379 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
380 {
381 	dev->drain_notify = true;
382 	wait_event(dev->waiting_to_drain,
383 		   atomic_read(&dev->num_outstanding_req) == 0);
384 	dev->drain_notify = false;
385 }
386 
387 static inline struct storvsc_device *get_in_stor_device(
388 					struct hv_device *device)
389 {
390 	struct storvsc_device *stor_device;
391 
392 	stor_device = hv_get_drvdata(device);
393 
394 	if (!stor_device)
395 		goto get_in_err;
396 
397 	/*
398 	 * If the device is being destroyed; allow incoming
399 	 * traffic only to cleanup outstanding requests.
400 	 */
401 
402 	if (stor_device->destroy  &&
403 		(atomic_read(&stor_device->num_outstanding_req) == 0))
404 		stor_device = NULL;
405 
406 get_in_err:
407 	return stor_device;
408 
409 }
410 
411 static void destroy_bounce_buffer(struct scatterlist *sgl,
412 				  unsigned int sg_count)
413 {
414 	int i;
415 	struct page *page_buf;
416 
417 	for (i = 0; i < sg_count; i++) {
418 		page_buf = sg_page((&sgl[i]));
419 		if (page_buf != NULL)
420 			__free_page(page_buf);
421 	}
422 
423 	kfree(sgl);
424 }
425 
426 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
427 {
428 	int i;
429 
430 	/* No need to check */
431 	if (sg_count < 2)
432 		return -1;
433 
434 	/* We have at least 2 sg entries */
435 	for (i = 0; i < sg_count; i++) {
436 		if (i == 0) {
437 			/* make sure 1st one does not have hole */
438 			if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
439 				return i;
440 		} else if (i == sg_count - 1) {
441 			/* make sure last one does not have hole */
442 			if (sgl[i].offset != 0)
443 				return i;
444 		} else {
445 			/* make sure no hole in the middle */
446 			if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
447 				return i;
448 		}
449 	}
450 	return -1;
451 }
452 
453 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
454 						unsigned int sg_count,
455 						unsigned int len,
456 						int write)
457 {
458 	int i;
459 	int num_pages;
460 	struct scatterlist *bounce_sgl;
461 	struct page *page_buf;
462 	unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
463 
464 	num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
465 
466 	bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
467 	if (!bounce_sgl)
468 		return NULL;
469 
470 	for (i = 0; i < num_pages; i++) {
471 		page_buf = alloc_page(GFP_ATOMIC);
472 		if (!page_buf)
473 			goto cleanup;
474 		sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0);
475 	}
476 
477 	return bounce_sgl;
478 
479 cleanup:
480 	destroy_bounce_buffer(bounce_sgl, num_pages);
481 	return NULL;
482 }
483 
484 /* Disgusting wrapper functions */
485 static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx)
486 {
487 	void *addr = kmap_atomic(sg_page(sgl + idx));
488 	return (unsigned long)addr;
489 }
490 
491 static inline void sg_kunmap_atomic(unsigned long addr)
492 {
493 	kunmap_atomic((void *)addr);
494 }
495 
496 
497 /* Assume the original sgl has enough room */
498 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
499 					    struct scatterlist *bounce_sgl,
500 					    unsigned int orig_sgl_count,
501 					    unsigned int bounce_sgl_count)
502 {
503 	int i;
504 	int j = 0;
505 	unsigned long src, dest;
506 	unsigned int srclen, destlen, copylen;
507 	unsigned int total_copied = 0;
508 	unsigned long bounce_addr = 0;
509 	unsigned long dest_addr = 0;
510 	unsigned long flags;
511 
512 	local_irq_save(flags);
513 
514 	for (i = 0; i < orig_sgl_count; i++) {
515 		dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
516 		dest = dest_addr;
517 		destlen = orig_sgl[i].length;
518 
519 		if (bounce_addr == 0)
520 			bounce_addr = sg_kmap_atomic(bounce_sgl,j);
521 
522 		while (destlen) {
523 			src = bounce_addr + bounce_sgl[j].offset;
524 			srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
525 
526 			copylen = min(srclen, destlen);
527 			memcpy((void *)dest, (void *)src, copylen);
528 
529 			total_copied += copylen;
530 			bounce_sgl[j].offset += copylen;
531 			destlen -= copylen;
532 			dest += copylen;
533 
534 			if (bounce_sgl[j].offset == bounce_sgl[j].length) {
535 				/* full */
536 				sg_kunmap_atomic(bounce_addr);
537 				j++;
538 
539 				/*
540 				 * It is possible that the number of elements
541 				 * in the bounce buffer may not be equal to
542 				 * the number of elements in the original
543 				 * scatter list. Handle this correctly.
544 				 */
545 
546 				if (j == bounce_sgl_count) {
547 					/*
548 					 * We are done; cleanup and return.
549 					 */
550 					sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
551 					local_irq_restore(flags);
552 					return total_copied;
553 				}
554 
555 				/* if we need to use another bounce buffer */
556 				if (destlen || i != orig_sgl_count - 1)
557 					bounce_addr = sg_kmap_atomic(bounce_sgl,j);
558 			} else if (destlen == 0 && i == orig_sgl_count - 1) {
559 				/* unmap the last bounce that is < PAGE_SIZE */
560 				sg_kunmap_atomic(bounce_addr);
561 			}
562 		}
563 
564 		sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
565 	}
566 
567 	local_irq_restore(flags);
568 
569 	return total_copied;
570 }
571 
572 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
573 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
574 					  struct scatterlist *bounce_sgl,
575 					  unsigned int orig_sgl_count)
576 {
577 	int i;
578 	int j = 0;
579 	unsigned long src, dest;
580 	unsigned int srclen, destlen, copylen;
581 	unsigned int total_copied = 0;
582 	unsigned long bounce_addr = 0;
583 	unsigned long src_addr = 0;
584 	unsigned long flags;
585 
586 	local_irq_save(flags);
587 
588 	for (i = 0; i < orig_sgl_count; i++) {
589 		src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
590 		src = src_addr;
591 		srclen = orig_sgl[i].length;
592 
593 		if (bounce_addr == 0)
594 			bounce_addr = sg_kmap_atomic(bounce_sgl,j);
595 
596 		while (srclen) {
597 			/* assume bounce offset always == 0 */
598 			dest = bounce_addr + bounce_sgl[j].length;
599 			destlen = PAGE_SIZE - bounce_sgl[j].length;
600 
601 			copylen = min(srclen, destlen);
602 			memcpy((void *)dest, (void *)src, copylen);
603 
604 			total_copied += copylen;
605 			bounce_sgl[j].length += copylen;
606 			srclen -= copylen;
607 			src += copylen;
608 
609 			if (bounce_sgl[j].length == PAGE_SIZE) {
610 				/* full..move to next entry */
611 				sg_kunmap_atomic(bounce_addr);
612 				j++;
613 
614 				/* if we need to use another bounce buffer */
615 				if (srclen || i != orig_sgl_count - 1)
616 					bounce_addr = sg_kmap_atomic(bounce_sgl,j);
617 
618 			} else if (srclen == 0 && i == orig_sgl_count - 1) {
619 				/* unmap the last bounce that is < PAGE_SIZE */
620 				sg_kunmap_atomic(bounce_addr);
621 			}
622 		}
623 
624 		sg_kunmap_atomic(src_addr - orig_sgl[i].offset);
625 	}
626 
627 	local_irq_restore(flags);
628 
629 	return total_copied;
630 }
631 
632 static int storvsc_channel_init(struct hv_device *device)
633 {
634 	struct storvsc_device *stor_device;
635 	struct storvsc_cmd_request *request;
636 	struct vstor_packet *vstor_packet;
637 	int ret, t;
638 
639 	stor_device = get_out_stor_device(device);
640 	if (!stor_device)
641 		return -ENODEV;
642 
643 	request = &stor_device->init_request;
644 	vstor_packet = &request->vstor_packet;
645 
646 	/*
647 	 * Now, initiate the vsc/vsp initialization protocol on the open
648 	 * channel
649 	 */
650 	memset(request, 0, sizeof(struct storvsc_cmd_request));
651 	init_completion(&request->wait_event);
652 	vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
653 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
654 
655 	ret = vmbus_sendpacket(device->channel, vstor_packet,
656 			       sizeof(struct vstor_packet),
657 			       (unsigned long)request,
658 			       VM_PKT_DATA_INBAND,
659 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
660 	if (ret != 0)
661 		goto cleanup;
662 
663 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
664 	if (t == 0) {
665 		ret = -ETIMEDOUT;
666 		goto cleanup;
667 	}
668 
669 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
670 	    vstor_packet->status != 0)
671 		goto cleanup;
672 
673 
674 	/* reuse the packet for version range supported */
675 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
676 	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
677 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
678 
679 	vstor_packet->version.major_minor =
680 		storvsc_get_version(VMSTOR_CURRENT_MAJOR, VMSTOR_CURRENT_MINOR);
681 
682 	/*
683 	 * The revision number is only used in Windows; set it to 0.
684 	 */
685 	vstor_packet->version.revision = 0;
686 
687 	ret = vmbus_sendpacket(device->channel, vstor_packet,
688 			       sizeof(struct vstor_packet),
689 			       (unsigned long)request,
690 			       VM_PKT_DATA_INBAND,
691 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
692 	if (ret != 0)
693 		goto cleanup;
694 
695 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
696 	if (t == 0) {
697 		ret = -ETIMEDOUT;
698 		goto cleanup;
699 	}
700 
701 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
702 	    vstor_packet->status != 0)
703 		goto cleanup;
704 
705 
706 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
707 	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
708 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
709 	vstor_packet->storage_channel_properties.port_number =
710 					stor_device->port_number;
711 
712 	ret = vmbus_sendpacket(device->channel, vstor_packet,
713 			       sizeof(struct vstor_packet),
714 			       (unsigned long)request,
715 			       VM_PKT_DATA_INBAND,
716 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
717 
718 	if (ret != 0)
719 		goto cleanup;
720 
721 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
722 	if (t == 0) {
723 		ret = -ETIMEDOUT;
724 		goto cleanup;
725 	}
726 
727 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
728 	    vstor_packet->status != 0)
729 		goto cleanup;
730 
731 	stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
732 	stor_device->target_id
733 		= vstor_packet->storage_channel_properties.target_id;
734 
735 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
736 	vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
737 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
738 
739 	ret = vmbus_sendpacket(device->channel, vstor_packet,
740 			       sizeof(struct vstor_packet),
741 			       (unsigned long)request,
742 			       VM_PKT_DATA_INBAND,
743 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
744 
745 	if (ret != 0)
746 		goto cleanup;
747 
748 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
749 	if (t == 0) {
750 		ret = -ETIMEDOUT;
751 		goto cleanup;
752 	}
753 
754 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
755 	    vstor_packet->status != 0)
756 		goto cleanup;
757 
758 
759 cleanup:
760 	return ret;
761 }
762 
763 
764 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
765 {
766 	struct scsi_cmnd *scmnd = cmd_request->cmd;
767 	struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
768 	void (*scsi_done_fn)(struct scsi_cmnd *);
769 	struct scsi_sense_hdr sense_hdr;
770 	struct vmscsi_request *vm_srb;
771 	struct storvsc_scan_work *wrk;
772 	struct stor_mem_pools *memp = scmnd->device->hostdata;
773 
774 	vm_srb = &cmd_request->vstor_packet.vm_srb;
775 	if (cmd_request->bounce_sgl_count) {
776 		if (vm_srb->data_in == READ_TYPE)
777 			copy_from_bounce_buffer(scsi_sglist(scmnd),
778 					cmd_request->bounce_sgl,
779 					scsi_sg_count(scmnd),
780 					cmd_request->bounce_sgl_count);
781 		destroy_bounce_buffer(cmd_request->bounce_sgl,
782 					cmd_request->bounce_sgl_count);
783 	}
784 
785 	/*
786 	 * If there is an error; offline the device since all
787 	 * error recovery strategies would have already been
788 	 * deployed on the host side. However, if the command
789 	 * were a pass-through command deal with it appropriately.
790 	 */
791 	scmnd->result = vm_srb->scsi_status;
792 
793 	if (vm_srb->srb_status == SRB_STATUS_ERROR) {
794 		switch (scmnd->cmnd[0]) {
795 		case ATA_16:
796 		case ATA_12:
797 			set_host_byte(scmnd, DID_PASSTHROUGH);
798 			break;
799 		default:
800 			set_host_byte(scmnd, DID_TARGET_FAILURE);
801 		}
802 	}
803 
804 
805 	/*
806 	 * If the LUN is invalid; remove the device.
807 	 */
808 	if (vm_srb->srb_status == SRB_STATUS_INVALID_LUN) {
809 		struct storvsc_device *stor_dev;
810 		struct hv_device *dev = host_dev->dev;
811 		struct Scsi_Host *host;
812 
813 		stor_dev = get_in_stor_device(dev);
814 		host = stor_dev->host;
815 
816 		wrk = kmalloc(sizeof(struct storvsc_scan_work),
817 				GFP_ATOMIC);
818 		if (!wrk) {
819 			scmnd->result = DID_TARGET_FAILURE << 16;
820 		} else {
821 			wrk->host = host;
822 			wrk->lun = vm_srb->lun;
823 			INIT_WORK(&wrk->work, storvsc_remove_lun);
824 			schedule_work(&wrk->work);
825 		}
826 	}
827 
828 	if (scmnd->result) {
829 		if (scsi_normalize_sense(scmnd->sense_buffer,
830 				SCSI_SENSE_BUFFERSIZE, &sense_hdr))
831 			scsi_print_sense_hdr("storvsc", &sense_hdr);
832 	}
833 
834 	scsi_set_resid(scmnd,
835 		cmd_request->data_buffer.len -
836 		vm_srb->data_transfer_length);
837 
838 	scsi_done_fn = scmnd->scsi_done;
839 
840 	scmnd->host_scribble = NULL;
841 	scmnd->scsi_done = NULL;
842 
843 	scsi_done_fn(scmnd);
844 
845 	mempool_free(cmd_request, memp->request_mempool);
846 }
847 
848 static void storvsc_on_io_completion(struct hv_device *device,
849 				  struct vstor_packet *vstor_packet,
850 				  struct storvsc_cmd_request *request)
851 {
852 	struct storvsc_device *stor_device;
853 	struct vstor_packet *stor_pkt;
854 
855 	stor_device = hv_get_drvdata(device);
856 	stor_pkt = &request->vstor_packet;
857 
858 	/*
859 	 * The current SCSI handling on the host side does
860 	 * not correctly handle:
861 	 * INQUIRY command with page code parameter set to 0x80
862 	 * MODE_SENSE command with cmd[2] == 0x1c
863 	 *
864 	 * Setup srb and scsi status so this won't be fatal.
865 	 * We do this so we can distinguish truly fatal failues
866 	 * (srb status == 0x4) and off-line the device in that case.
867 	 */
868 
869 	if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
870 	   (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
871 		vstor_packet->vm_srb.scsi_status = 0;
872 		vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
873 	}
874 
875 
876 	/* Copy over the status...etc */
877 	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
878 	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
879 	stor_pkt->vm_srb.sense_info_length =
880 	vstor_packet->vm_srb.sense_info_length;
881 
882 	if (vstor_packet->vm_srb.scsi_status != 0 ||
883 		vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){
884 		dev_warn(&device->device,
885 			 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
886 			 stor_pkt->vm_srb.cdb[0],
887 			 vstor_packet->vm_srb.scsi_status,
888 			 vstor_packet->vm_srb.srb_status);
889 	}
890 
891 	if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
892 		/* CHECK_CONDITION */
893 		if (vstor_packet->vm_srb.srb_status &
894 			SRB_STATUS_AUTOSENSE_VALID) {
895 			/* autosense data available */
896 			dev_warn(&device->device,
897 				 "stor pkt %p autosense data valid - len %d\n",
898 				 request,
899 				 vstor_packet->vm_srb.sense_info_length);
900 
901 			memcpy(request->sense_buffer,
902 			       vstor_packet->vm_srb.sense_data,
903 			       vstor_packet->vm_srb.sense_info_length);
904 
905 		}
906 	}
907 
908 	stor_pkt->vm_srb.data_transfer_length =
909 	vstor_packet->vm_srb.data_transfer_length;
910 
911 	storvsc_command_completion(request);
912 
913 	if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
914 		stor_device->drain_notify)
915 		wake_up(&stor_device->waiting_to_drain);
916 
917 
918 }
919 
920 static void storvsc_on_receive(struct hv_device *device,
921 			     struct vstor_packet *vstor_packet,
922 			     struct storvsc_cmd_request *request)
923 {
924 	struct storvsc_scan_work *work;
925 	struct storvsc_device *stor_device;
926 
927 	switch (vstor_packet->operation) {
928 	case VSTOR_OPERATION_COMPLETE_IO:
929 		storvsc_on_io_completion(device, vstor_packet, request);
930 		break;
931 
932 	case VSTOR_OPERATION_REMOVE_DEVICE:
933 	case VSTOR_OPERATION_ENUMERATE_BUS:
934 		stor_device = get_in_stor_device(device);
935 		work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
936 		if (!work)
937 			return;
938 
939 		INIT_WORK(&work->work, storvsc_bus_scan);
940 		work->host = stor_device->host;
941 		schedule_work(&work->work);
942 		break;
943 
944 	default:
945 		break;
946 	}
947 }
948 
949 static void storvsc_on_channel_callback(void *context)
950 {
951 	struct hv_device *device = (struct hv_device *)context;
952 	struct storvsc_device *stor_device;
953 	u32 bytes_recvd;
954 	u64 request_id;
955 	unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
956 	struct storvsc_cmd_request *request;
957 	int ret;
958 
959 
960 	stor_device = get_in_stor_device(device);
961 	if (!stor_device)
962 		return;
963 
964 	do {
965 		ret = vmbus_recvpacket(device->channel, packet,
966 				       ALIGN(sizeof(struct vstor_packet), 8),
967 				       &bytes_recvd, &request_id);
968 		if (ret == 0 && bytes_recvd > 0) {
969 
970 			request = (struct storvsc_cmd_request *)
971 					(unsigned long)request_id;
972 
973 			if ((request == &stor_device->init_request) ||
974 			    (request == &stor_device->reset_request)) {
975 
976 				memcpy(&request->vstor_packet, packet,
977 				       sizeof(struct vstor_packet));
978 				complete(&request->wait_event);
979 			} else {
980 				storvsc_on_receive(device,
981 						(struct vstor_packet *)packet,
982 						request);
983 			}
984 		} else {
985 			break;
986 		}
987 	} while (1);
988 
989 	return;
990 }
991 
992 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
993 {
994 	struct vmstorage_channel_properties props;
995 	int ret;
996 
997 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
998 
999 	ret = vmbus_open(device->channel,
1000 			 ring_size,
1001 			 ring_size,
1002 			 (void *)&props,
1003 			 sizeof(struct vmstorage_channel_properties),
1004 			 storvsc_on_channel_callback, device);
1005 
1006 	if (ret != 0)
1007 		return ret;
1008 
1009 	ret = storvsc_channel_init(device);
1010 
1011 	return ret;
1012 }
1013 
1014 static int storvsc_dev_remove(struct hv_device *device)
1015 {
1016 	struct storvsc_device *stor_device;
1017 	unsigned long flags;
1018 
1019 	stor_device = hv_get_drvdata(device);
1020 
1021 	spin_lock_irqsave(&device->channel->inbound_lock, flags);
1022 	stor_device->destroy = true;
1023 	spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1024 
1025 	/*
1026 	 * At this point, all outbound traffic should be disable. We
1027 	 * only allow inbound traffic (responses) to proceed so that
1028 	 * outstanding requests can be completed.
1029 	 */
1030 
1031 	storvsc_wait_to_drain(stor_device);
1032 
1033 	/*
1034 	 * Since we have already drained, we don't need to busy wait
1035 	 * as was done in final_release_stor_device()
1036 	 * Note that we cannot set the ext pointer to NULL until
1037 	 * we have drained - to drain the outgoing packets, we need to
1038 	 * allow incoming packets.
1039 	 */
1040 	spin_lock_irqsave(&device->channel->inbound_lock, flags);
1041 	hv_set_drvdata(device, NULL);
1042 	spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1043 
1044 	/* Close the channel */
1045 	vmbus_close(device->channel);
1046 
1047 	kfree(stor_device);
1048 	return 0;
1049 }
1050 
1051 static int storvsc_do_io(struct hv_device *device,
1052 			      struct storvsc_cmd_request *request)
1053 {
1054 	struct storvsc_device *stor_device;
1055 	struct vstor_packet *vstor_packet;
1056 	int ret = 0;
1057 
1058 	vstor_packet = &request->vstor_packet;
1059 	stor_device = get_out_stor_device(device);
1060 
1061 	if (!stor_device)
1062 		return -ENODEV;
1063 
1064 
1065 	request->device  = device;
1066 
1067 
1068 	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1069 
1070 	vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1071 
1072 
1073 	vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1074 
1075 
1076 	vstor_packet->vm_srb.data_transfer_length =
1077 	request->data_buffer.len;
1078 
1079 	vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1080 
1081 	if (request->data_buffer.len) {
1082 		ret = vmbus_sendpacket_multipagebuffer(device->channel,
1083 				&request->data_buffer,
1084 				vstor_packet,
1085 				sizeof(struct vstor_packet),
1086 				(unsigned long)request);
1087 	} else {
1088 		ret = vmbus_sendpacket(device->channel, vstor_packet,
1089 			       sizeof(struct vstor_packet),
1090 			       (unsigned long)request,
1091 			       VM_PKT_DATA_INBAND,
1092 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1093 	}
1094 
1095 	if (ret != 0)
1096 		return ret;
1097 
1098 	atomic_inc(&stor_device->num_outstanding_req);
1099 
1100 	return ret;
1101 }
1102 
1103 static int storvsc_device_alloc(struct scsi_device *sdevice)
1104 {
1105 	struct stor_mem_pools *memp;
1106 	int number = STORVSC_MIN_BUF_NR;
1107 
1108 	memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
1109 	if (!memp)
1110 		return -ENOMEM;
1111 
1112 	memp->request_pool =
1113 		kmem_cache_create(dev_name(&sdevice->sdev_dev),
1114 				sizeof(struct storvsc_cmd_request), 0,
1115 				SLAB_HWCACHE_ALIGN, NULL);
1116 
1117 	if (!memp->request_pool)
1118 		goto err0;
1119 
1120 	memp->request_mempool = mempool_create(number, mempool_alloc_slab,
1121 						mempool_free_slab,
1122 						memp->request_pool);
1123 
1124 	if (!memp->request_mempool)
1125 		goto err1;
1126 
1127 	sdevice->hostdata = memp;
1128 
1129 	return 0;
1130 
1131 err1:
1132 	kmem_cache_destroy(memp->request_pool);
1133 
1134 err0:
1135 	kfree(memp);
1136 	return -ENOMEM;
1137 }
1138 
1139 static void storvsc_device_destroy(struct scsi_device *sdevice)
1140 {
1141 	struct stor_mem_pools *memp = sdevice->hostdata;
1142 
1143 	mempool_destroy(memp->request_mempool);
1144 	kmem_cache_destroy(memp->request_pool);
1145 	kfree(memp);
1146 	sdevice->hostdata = NULL;
1147 }
1148 
1149 static int storvsc_device_configure(struct scsi_device *sdevice)
1150 {
1151 	scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
1152 				STORVSC_MAX_IO_REQUESTS);
1153 
1154 	blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
1155 
1156 	blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
1157 
1158 	return 0;
1159 }
1160 
1161 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1162 			   sector_t capacity, int *info)
1163 {
1164 	sector_t nsect = capacity;
1165 	sector_t cylinders = nsect;
1166 	int heads, sectors_pt;
1167 
1168 	/*
1169 	 * We are making up these values; let us keep it simple.
1170 	 */
1171 	heads = 0xff;
1172 	sectors_pt = 0x3f;      /* Sectors per track */
1173 	sector_div(cylinders, heads * sectors_pt);
1174 	if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1175 		cylinders = 0xffff;
1176 
1177 	info[0] = heads;
1178 	info[1] = sectors_pt;
1179 	info[2] = (int)cylinders;
1180 
1181 	return 0;
1182 }
1183 
1184 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1185 {
1186 	struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1187 	struct hv_device *device = host_dev->dev;
1188 
1189 	struct storvsc_device *stor_device;
1190 	struct storvsc_cmd_request *request;
1191 	struct vstor_packet *vstor_packet;
1192 	int ret, t;
1193 
1194 
1195 	stor_device = get_out_stor_device(device);
1196 	if (!stor_device)
1197 		return FAILED;
1198 
1199 	request = &stor_device->reset_request;
1200 	vstor_packet = &request->vstor_packet;
1201 
1202 	init_completion(&request->wait_event);
1203 
1204 	vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1205 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1206 	vstor_packet->vm_srb.path_id = stor_device->path_id;
1207 
1208 	ret = vmbus_sendpacket(device->channel, vstor_packet,
1209 			       sizeof(struct vstor_packet),
1210 			       (unsigned long)&stor_device->reset_request,
1211 			       VM_PKT_DATA_INBAND,
1212 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1213 	if (ret != 0)
1214 		return FAILED;
1215 
1216 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1217 	if (t == 0)
1218 		return TIMEOUT_ERROR;
1219 
1220 
1221 	/*
1222 	 * At this point, all outstanding requests in the adapter
1223 	 * should have been flushed out and return to us
1224 	 */
1225 
1226 	return SUCCESS;
1227 }
1228 
1229 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1230 {
1231 	bool allowed = true;
1232 	u8 scsi_op = scmnd->cmnd[0];
1233 
1234 	switch (scsi_op) {
1235 	/*
1236 	 * smartd sends this command and the host does not handle
1237 	 * this. So, don't send it.
1238 	 */
1239 	case SET_WINDOW:
1240 		scmnd->result = ILLEGAL_REQUEST << 16;
1241 		allowed = false;
1242 		break;
1243 	default:
1244 		break;
1245 	}
1246 	return allowed;
1247 }
1248 
1249 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1250 {
1251 	int ret;
1252 	struct hv_host_device *host_dev = shost_priv(host);
1253 	struct hv_device *dev = host_dev->dev;
1254 	struct storvsc_cmd_request *cmd_request;
1255 	unsigned int request_size = 0;
1256 	int i;
1257 	struct scatterlist *sgl;
1258 	unsigned int sg_count = 0;
1259 	struct vmscsi_request *vm_srb;
1260 	struct stor_mem_pools *memp = scmnd->device->hostdata;
1261 
1262 	if (!storvsc_scsi_cmd_ok(scmnd)) {
1263 		scmnd->scsi_done(scmnd);
1264 		return 0;
1265 	}
1266 
1267 	request_size = sizeof(struct storvsc_cmd_request);
1268 
1269 	cmd_request = mempool_alloc(memp->request_mempool,
1270 				       GFP_ATOMIC);
1271 
1272 	/*
1273 	 * We might be invoked in an interrupt context; hence
1274 	 * mempool_alloc() can fail.
1275 	 */
1276 	if (!cmd_request)
1277 		return SCSI_MLQUEUE_DEVICE_BUSY;
1278 
1279 	memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1280 
1281 	/* Setup the cmd request */
1282 	cmd_request->cmd = scmnd;
1283 
1284 	scmnd->host_scribble = (unsigned char *)cmd_request;
1285 
1286 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1287 
1288 
1289 	/* Build the SRB */
1290 	switch (scmnd->sc_data_direction) {
1291 	case DMA_TO_DEVICE:
1292 		vm_srb->data_in = WRITE_TYPE;
1293 		break;
1294 	case DMA_FROM_DEVICE:
1295 		vm_srb->data_in = READ_TYPE;
1296 		break;
1297 	default:
1298 		vm_srb->data_in = UNKNOWN_TYPE;
1299 		break;
1300 	}
1301 
1302 
1303 	vm_srb->port_number = host_dev->port;
1304 	vm_srb->path_id = scmnd->device->channel;
1305 	vm_srb->target_id = scmnd->device->id;
1306 	vm_srb->lun = scmnd->device->lun;
1307 
1308 	vm_srb->cdb_length = scmnd->cmd_len;
1309 
1310 	memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1311 
1312 	cmd_request->sense_buffer = scmnd->sense_buffer;
1313 
1314 
1315 	cmd_request->data_buffer.len = scsi_bufflen(scmnd);
1316 	if (scsi_sg_count(scmnd)) {
1317 		sgl = (struct scatterlist *)scsi_sglist(scmnd);
1318 		sg_count = scsi_sg_count(scmnd);
1319 
1320 		/* check if we need to bounce the sgl */
1321 		if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1322 			cmd_request->bounce_sgl =
1323 				create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1324 						     scsi_bufflen(scmnd),
1325 						     vm_srb->data_in);
1326 			if (!cmd_request->bounce_sgl) {
1327 				ret = SCSI_MLQUEUE_HOST_BUSY;
1328 				goto queue_error;
1329 			}
1330 
1331 			cmd_request->bounce_sgl_count =
1332 				ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1333 					PAGE_SHIFT;
1334 
1335 			if (vm_srb->data_in == WRITE_TYPE)
1336 				copy_to_bounce_buffer(sgl,
1337 					cmd_request->bounce_sgl,
1338 					scsi_sg_count(scmnd));
1339 
1340 			sgl = cmd_request->bounce_sgl;
1341 			sg_count = cmd_request->bounce_sgl_count;
1342 		}
1343 
1344 		cmd_request->data_buffer.offset = sgl[0].offset;
1345 
1346 		for (i = 0; i < sg_count; i++)
1347 			cmd_request->data_buffer.pfn_array[i] =
1348 				page_to_pfn(sg_page((&sgl[i])));
1349 
1350 	} else if (scsi_sglist(scmnd)) {
1351 		cmd_request->data_buffer.offset =
1352 			virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1353 		cmd_request->data_buffer.pfn_array[0] =
1354 			virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1355 	}
1356 
1357 	/* Invokes the vsc to start an IO */
1358 	ret = storvsc_do_io(dev, cmd_request);
1359 
1360 	if (ret == -EAGAIN) {
1361 		/* no more space */
1362 
1363 		if (cmd_request->bounce_sgl_count) {
1364 			destroy_bounce_buffer(cmd_request->bounce_sgl,
1365 					cmd_request->bounce_sgl_count);
1366 
1367 			ret = SCSI_MLQUEUE_DEVICE_BUSY;
1368 			goto queue_error;
1369 		}
1370 	}
1371 
1372 	return 0;
1373 
1374 queue_error:
1375 	mempool_free(cmd_request, memp->request_mempool);
1376 	scmnd->host_scribble = NULL;
1377 	return ret;
1378 }
1379 
1380 static struct scsi_host_template scsi_driver = {
1381 	.module	=		THIS_MODULE,
1382 	.name =			"storvsc_host_t",
1383 	.bios_param =		storvsc_get_chs,
1384 	.queuecommand =		storvsc_queuecommand,
1385 	.eh_host_reset_handler =	storvsc_host_reset_handler,
1386 	.slave_alloc =		storvsc_device_alloc,
1387 	.slave_destroy =	storvsc_device_destroy,
1388 	.slave_configure =	storvsc_device_configure,
1389 	.cmd_per_lun =		1,
1390 	/* 64 max_queue * 1 target */
1391 	.can_queue =		STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1392 	.this_id =		-1,
1393 	/* no use setting to 0 since ll_blk_rw reset it to 1 */
1394 	/* currently 32 */
1395 	.sg_tablesize =		MAX_MULTIPAGE_BUFFER_COUNT,
1396 	.use_clustering =	DISABLE_CLUSTERING,
1397 	/* Make sure we dont get a sg segment crosses a page boundary */
1398 	.dma_boundary =		PAGE_SIZE-1,
1399 };
1400 
1401 enum {
1402 	SCSI_GUID,
1403 	IDE_GUID,
1404 };
1405 
1406 static const struct hv_vmbus_device_id id_table[] = {
1407 	/* SCSI guid */
1408 	{ VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1409 		       0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1410 	  .driver_data = SCSI_GUID },
1411 	/* IDE guid */
1412 	{ VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1413 		       0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1414 	  .driver_data = IDE_GUID },
1415 	{ },
1416 };
1417 
1418 MODULE_DEVICE_TABLE(vmbus, id_table);
1419 
1420 static int storvsc_probe(struct hv_device *device,
1421 			const struct hv_vmbus_device_id *dev_id)
1422 {
1423 	int ret;
1424 	struct Scsi_Host *host;
1425 	struct hv_host_device *host_dev;
1426 	bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1427 	int target = 0;
1428 	struct storvsc_device *stor_device;
1429 
1430 	host = scsi_host_alloc(&scsi_driver,
1431 			       sizeof(struct hv_host_device));
1432 	if (!host)
1433 		return -ENOMEM;
1434 
1435 	host_dev = shost_priv(host);
1436 	memset(host_dev, 0, sizeof(struct hv_host_device));
1437 
1438 	host_dev->port = host->host_no;
1439 	host_dev->dev = device;
1440 
1441 
1442 	stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1443 	if (!stor_device) {
1444 		ret = -ENOMEM;
1445 		goto err_out0;
1446 	}
1447 
1448 	stor_device->destroy = false;
1449 	init_waitqueue_head(&stor_device->waiting_to_drain);
1450 	stor_device->device = device;
1451 	stor_device->host = host;
1452 	hv_set_drvdata(device, stor_device);
1453 
1454 	stor_device->port_number = host->host_no;
1455 	ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1456 	if (ret)
1457 		goto err_out1;
1458 
1459 	host_dev->path = stor_device->path_id;
1460 	host_dev->target = stor_device->target_id;
1461 
1462 	/* max # of devices per target */
1463 	host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1464 	/* max # of targets per channel */
1465 	host->max_id = STORVSC_MAX_TARGETS;
1466 	/* max # of channels */
1467 	host->max_channel = STORVSC_MAX_CHANNELS - 1;
1468 	/* max cmd length */
1469 	host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1470 
1471 	/* Register the HBA and start the scsi bus scan */
1472 	ret = scsi_add_host(host, &device->device);
1473 	if (ret != 0)
1474 		goto err_out2;
1475 
1476 	if (!dev_is_ide) {
1477 		scsi_scan_host(host);
1478 	} else {
1479 		target = (device->dev_instance.b[5] << 8 |
1480 			 device->dev_instance.b[4]);
1481 		ret = scsi_add_device(host, 0, target, 0);
1482 		if (ret) {
1483 			scsi_remove_host(host);
1484 			goto err_out2;
1485 		}
1486 	}
1487 	return 0;
1488 
1489 err_out2:
1490 	/*
1491 	 * Once we have connected with the host, we would need to
1492 	 * to invoke storvsc_dev_remove() to rollback this state and
1493 	 * this call also frees up the stor_device; hence the jump around
1494 	 * err_out1 label.
1495 	 */
1496 	storvsc_dev_remove(device);
1497 	goto err_out0;
1498 
1499 err_out1:
1500 	kfree(stor_device);
1501 
1502 err_out0:
1503 	scsi_host_put(host);
1504 	return ret;
1505 }
1506 
1507 static int storvsc_remove(struct hv_device *dev)
1508 {
1509 	struct storvsc_device *stor_device = hv_get_drvdata(dev);
1510 	struct Scsi_Host *host = stor_device->host;
1511 
1512 	scsi_remove_host(host);
1513 	storvsc_dev_remove(dev);
1514 	scsi_host_put(host);
1515 
1516 	return 0;
1517 }
1518 
1519 static struct hv_driver storvsc_drv = {
1520 	.name = KBUILD_MODNAME,
1521 	.id_table = id_table,
1522 	.probe = storvsc_probe,
1523 	.remove = storvsc_remove,
1524 };
1525 
1526 static int __init storvsc_drv_init(void)
1527 {
1528 	u32 max_outstanding_req_per_channel;
1529 
1530 	/*
1531 	 * Divide the ring buffer data size (which is 1 page less
1532 	 * than the ring buffer size since that page is reserved for
1533 	 * the ring buffer indices) by the max request size (which is
1534 	 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1535 	 */
1536 	max_outstanding_req_per_channel =
1537 		((storvsc_ringbuffer_size - PAGE_SIZE) /
1538 		ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1539 		sizeof(struct vstor_packet) + sizeof(u64),
1540 		sizeof(u64)));
1541 
1542 	if (max_outstanding_req_per_channel <
1543 	    STORVSC_MAX_IO_REQUESTS)
1544 		return -EINVAL;
1545 
1546 	return vmbus_driver_register(&storvsc_drv);
1547 }
1548 
1549 static void __exit storvsc_drv_exit(void)
1550 {
1551 	vmbus_driver_unregister(&storvsc_drv);
1552 }
1553 
1554 MODULE_LICENSE("GPL");
1555 MODULE_VERSION(HV_DRV_VERSION);
1556 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1557 module_init(storvsc_drv_init);
1558 module_exit(storvsc_drv_exit);
1559