xref: /openbmc/linux/drivers/scsi/storvsc_drv.c (revision 36fe4655)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2009, Microsoft Corporation.
4  *
5  * Authors:
6  *   Haiyang Zhang <haiyangz@microsoft.com>
7  *   Hank Janssen  <hjanssen@microsoft.com>
8  *   K. Y. Srinivasan <kys@microsoft.com>
9  */
10 
11 #include <linux/kernel.h>
12 #include <linux/wait.h>
13 #include <linux/sched.h>
14 #include <linux/completion.h>
15 #include <linux/string.h>
16 #include <linux/mm.h>
17 #include <linux/delay.h>
18 #include <linux/init.h>
19 #include <linux/slab.h>
20 #include <linux/module.h>
21 #include <linux/device.h>
22 #include <linux/hyperv.h>
23 #include <linux/blkdev.h>
24 #include <scsi/scsi.h>
25 #include <scsi/scsi_cmnd.h>
26 #include <scsi/scsi_host.h>
27 #include <scsi/scsi_device.h>
28 #include <scsi/scsi_tcq.h>
29 #include <scsi/scsi_eh.h>
30 #include <scsi/scsi_devinfo.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_transport_fc.h>
33 #include <scsi/scsi_transport.h>
34 
35 /*
36  * All wire protocol details (storage protocol between the guest and the host)
37  * are consolidated here.
38  *
39  * Begin protocol definitions.
40  */
41 
42 /*
43  * Version history:
44  * V1 Beta: 0.1
45  * V1 RC < 2008/1/31: 1.0
46  * V1 RC > 2008/1/31:  2.0
47  * Win7: 4.2
48  * Win8: 5.1
49  * Win8.1: 6.0
50  * Win10: 6.2
51  */
52 
53 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_)	((((MAJOR_) & 0xff) << 8) | \
54 						(((MINOR_) & 0xff)))
55 
56 #define VMSTOR_PROTO_VERSION_WIN6	VMSTOR_PROTO_VERSION(2, 0)
57 #define VMSTOR_PROTO_VERSION_WIN7	VMSTOR_PROTO_VERSION(4, 2)
58 #define VMSTOR_PROTO_VERSION_WIN8	VMSTOR_PROTO_VERSION(5, 1)
59 #define VMSTOR_PROTO_VERSION_WIN8_1	VMSTOR_PROTO_VERSION(6, 0)
60 #define VMSTOR_PROTO_VERSION_WIN10	VMSTOR_PROTO_VERSION(6, 2)
61 
62 /*  Packet structure describing virtual storage requests. */
63 enum vstor_packet_operation {
64 	VSTOR_OPERATION_COMPLETE_IO		= 1,
65 	VSTOR_OPERATION_REMOVE_DEVICE		= 2,
66 	VSTOR_OPERATION_EXECUTE_SRB		= 3,
67 	VSTOR_OPERATION_RESET_LUN		= 4,
68 	VSTOR_OPERATION_RESET_ADAPTER		= 5,
69 	VSTOR_OPERATION_RESET_BUS		= 6,
70 	VSTOR_OPERATION_BEGIN_INITIALIZATION	= 7,
71 	VSTOR_OPERATION_END_INITIALIZATION	= 8,
72 	VSTOR_OPERATION_QUERY_PROTOCOL_VERSION	= 9,
73 	VSTOR_OPERATION_QUERY_PROPERTIES	= 10,
74 	VSTOR_OPERATION_ENUMERATE_BUS		= 11,
75 	VSTOR_OPERATION_FCHBA_DATA              = 12,
76 	VSTOR_OPERATION_CREATE_SUB_CHANNELS     = 13,
77 	VSTOR_OPERATION_MAXIMUM                 = 13
78 };
79 
80 /*
81  * WWN packet for Fibre Channel HBA
82  */
83 
84 struct hv_fc_wwn_packet {
85 	u8	primary_active;
86 	u8	reserved1[3];
87 	u8	primary_port_wwn[8];
88 	u8	primary_node_wwn[8];
89 	u8	secondary_port_wwn[8];
90 	u8	secondary_node_wwn[8];
91 };
92 
93 
94 
95 /*
96  * SRB Flag Bits
97  */
98 
99 #define SRB_FLAGS_QUEUE_ACTION_ENABLE		0x00000002
100 #define SRB_FLAGS_DISABLE_DISCONNECT		0x00000004
101 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER	0x00000008
102 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE		0x00000010
103 #define SRB_FLAGS_DISABLE_AUTOSENSE		0x00000020
104 #define SRB_FLAGS_DATA_IN			0x00000040
105 #define SRB_FLAGS_DATA_OUT			0x00000080
106 #define SRB_FLAGS_NO_DATA_TRANSFER		0x00000000
107 #define SRB_FLAGS_UNSPECIFIED_DIRECTION	(SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
108 #define SRB_FLAGS_NO_QUEUE_FREEZE		0x00000100
109 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE		0x00000200
110 #define SRB_FLAGS_FREE_SENSE_BUFFER		0x00000400
111 
112 /*
113  * This flag indicates the request is part of the workflow for processing a D3.
114  */
115 #define SRB_FLAGS_D3_PROCESSING			0x00000800
116 #define SRB_FLAGS_IS_ACTIVE			0x00010000
117 #define SRB_FLAGS_ALLOCATED_FROM_ZONE		0x00020000
118 #define SRB_FLAGS_SGLIST_FROM_POOL		0x00040000
119 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE		0x00080000
120 #define SRB_FLAGS_NO_KEEP_AWAKE			0x00100000
121 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE	0x00200000
122 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT	0x00400000
123 #define SRB_FLAGS_DONT_START_NEXT_PACKET	0x00800000
124 #define SRB_FLAGS_PORT_DRIVER_RESERVED		0x0F000000
125 #define SRB_FLAGS_CLASS_DRIVER_RESERVED		0xF0000000
126 
127 #define SP_UNTAGGED			((unsigned char) ~0)
128 #define SRB_SIMPLE_TAG_REQUEST		0x20
129 
130 /*
131  * Platform neutral description of a scsi request -
132  * this remains the same across the write regardless of 32/64 bit
133  * note: it's patterned off the SCSI_PASS_THROUGH structure
134  */
135 #define STORVSC_MAX_CMD_LEN			0x10
136 
137 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE	0x14
138 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE	0x12
139 
140 #define STORVSC_SENSE_BUFFER_SIZE		0x14
141 #define STORVSC_MAX_BUF_LEN_WITH_PADDING	0x14
142 
143 /*
144  * Sense buffer size changed in win8; have a run-time
145  * variable to track the size we should use.  This value will
146  * likely change during protocol negotiation but it is valid
147  * to start by assuming pre-Win8.
148  */
149 static int sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE;
150 
151 /*
152  * The storage protocol version is determined during the
153  * initial exchange with the host.  It will indicate which
154  * storage functionality is available in the host.
155 */
156 static int vmstor_proto_version;
157 
158 #define STORVSC_LOGGING_NONE	0
159 #define STORVSC_LOGGING_ERROR	1
160 #define STORVSC_LOGGING_WARN	2
161 
162 static int logging_level = STORVSC_LOGGING_ERROR;
163 module_param(logging_level, int, S_IRUGO|S_IWUSR);
164 MODULE_PARM_DESC(logging_level,
165 	"Logging level, 0 - None, 1 - Error (default), 2 - Warning.");
166 
167 static inline bool do_logging(int level)
168 {
169 	return logging_level >= level;
170 }
171 
172 #define storvsc_log(dev, level, fmt, ...)			\
173 do {								\
174 	if (do_logging(level))					\
175 		dev_warn(&(dev)->device, fmt, ##__VA_ARGS__);	\
176 } while (0)
177 
178 struct vmscsi_win8_extension {
179 	/*
180 	 * The following were added in Windows 8
181 	 */
182 	u16 reserve;
183 	u8  queue_tag;
184 	u8  queue_action;
185 	u32 srb_flags;
186 	u32 time_out_value;
187 	u32 queue_sort_ey;
188 } __packed;
189 
190 struct vmscsi_request {
191 	u16 length;
192 	u8 srb_status;
193 	u8 scsi_status;
194 
195 	u8  port_number;
196 	u8  path_id;
197 	u8  target_id;
198 	u8  lun;
199 
200 	u8  cdb_length;
201 	u8  sense_info_length;
202 	u8  data_in;
203 	u8  reserved;
204 
205 	u32 data_transfer_length;
206 
207 	union {
208 		u8 cdb[STORVSC_MAX_CMD_LEN];
209 		u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
210 		u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
211 	};
212 	/*
213 	 * The following was added in win8.
214 	 */
215 	struct vmscsi_win8_extension win8_extension;
216 
217 } __attribute((packed));
218 
219 
220 /*
221  * The size of the vmscsi_request has changed in win8. The
222  * additional size is because of new elements added to the
223  * structure. These elements are valid only when we are talking
224  * to a win8 host.
225  * Track the correction to size we need to apply. This value
226  * will likely change during protocol negotiation but it is
227  * valid to start by assuming pre-Win8.
228  */
229 static int vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
230 
231 /*
232  * The list of storage protocols in order of preference.
233  */
234 struct vmstor_protocol {
235 	int protocol_version;
236 	int sense_buffer_size;
237 	int vmscsi_size_delta;
238 };
239 
240 
241 static const struct vmstor_protocol vmstor_protocols[] = {
242 	{
243 		VMSTOR_PROTO_VERSION_WIN10,
244 		POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
245 		0
246 	},
247 	{
248 		VMSTOR_PROTO_VERSION_WIN8_1,
249 		POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
250 		0
251 	},
252 	{
253 		VMSTOR_PROTO_VERSION_WIN8,
254 		POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
255 		0
256 	},
257 	{
258 		VMSTOR_PROTO_VERSION_WIN7,
259 		PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
260 		sizeof(struct vmscsi_win8_extension),
261 	},
262 	{
263 		VMSTOR_PROTO_VERSION_WIN6,
264 		PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
265 		sizeof(struct vmscsi_win8_extension),
266 	}
267 };
268 
269 
270 /*
271  * This structure is sent during the initialization phase to get the different
272  * properties of the channel.
273  */
274 
275 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL		0x1
276 
277 struct vmstorage_channel_properties {
278 	u32 reserved;
279 	u16 max_channel_cnt;
280 	u16 reserved1;
281 
282 	u32 flags;
283 	u32   max_transfer_bytes;
284 
285 	u64  reserved2;
286 } __packed;
287 
288 /*  This structure is sent during the storage protocol negotiations. */
289 struct vmstorage_protocol_version {
290 	/* Major (MSW) and minor (LSW) version numbers. */
291 	u16 major_minor;
292 
293 	/*
294 	 * Revision number is auto-incremented whenever this file is changed
295 	 * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
296 	 * definitely indicate incompatibility--but it does indicate mismatched
297 	 * builds.
298 	 * This is only used on the windows side. Just set it to 0.
299 	 */
300 	u16 revision;
301 } __packed;
302 
303 /* Channel Property Flags */
304 #define STORAGE_CHANNEL_REMOVABLE_FLAG		0x1
305 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG	0x2
306 
307 struct vstor_packet {
308 	/* Requested operation type */
309 	enum vstor_packet_operation operation;
310 
311 	/*  Flags - see below for values */
312 	u32 flags;
313 
314 	/* Status of the request returned from the server side. */
315 	u32 status;
316 
317 	/* Data payload area */
318 	union {
319 		/*
320 		 * Structure used to forward SCSI commands from the
321 		 * client to the server.
322 		 */
323 		struct vmscsi_request vm_srb;
324 
325 		/* Structure used to query channel properties. */
326 		struct vmstorage_channel_properties storage_channel_properties;
327 
328 		/* Used during version negotiations. */
329 		struct vmstorage_protocol_version version;
330 
331 		/* Fibre channel address packet */
332 		struct hv_fc_wwn_packet wwn_packet;
333 
334 		/* Number of sub-channels to create */
335 		u16 sub_channel_count;
336 
337 		/* This will be the maximum of the union members */
338 		u8  buffer[0x34];
339 	};
340 } __packed;
341 
342 /*
343  * Packet Flags:
344  *
345  * This flag indicates that the server should send back a completion for this
346  * packet.
347  */
348 
349 #define REQUEST_COMPLETION_FLAG	0x1
350 
351 /* Matches Windows-end */
352 enum storvsc_request_type {
353 	WRITE_TYPE = 0,
354 	READ_TYPE,
355 	UNKNOWN_TYPE,
356 };
357 
358 /*
359  * SRB status codes and masks; a subset of the codes used here.
360  */
361 
362 #define SRB_STATUS_AUTOSENSE_VALID	0x80
363 #define SRB_STATUS_QUEUE_FROZEN		0x40
364 #define SRB_STATUS_INVALID_LUN	0x20
365 #define SRB_STATUS_SUCCESS	0x01
366 #define SRB_STATUS_ABORTED	0x02
367 #define SRB_STATUS_ERROR	0x04
368 #define SRB_STATUS_DATA_OVERRUN	0x12
369 
370 #define SRB_STATUS(status) \
371 	(status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
372 /*
373  * This is the end of Protocol specific defines.
374  */
375 
376 static int storvsc_ringbuffer_size = (128 * 1024);
377 static u32 max_outstanding_req_per_channel;
378 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
379 
380 static int storvsc_vcpus_per_sub_channel = 4;
381 
382 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
383 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
384 
385 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
386 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
387 
388 static int ring_avail_percent_lowater = 10;
389 module_param(ring_avail_percent_lowater, int, S_IRUGO);
390 MODULE_PARM_DESC(ring_avail_percent_lowater,
391 		"Select a channel if available ring size > this in percent");
392 
393 /*
394  * Timeout in seconds for all devices managed by this driver.
395  */
396 static int storvsc_timeout = 180;
397 
398 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
399 static struct scsi_transport_template *fc_transport_template;
400 #endif
401 
402 static struct scsi_host_template scsi_driver;
403 static void storvsc_on_channel_callback(void *context);
404 
405 #define STORVSC_MAX_LUNS_PER_TARGET			255
406 #define STORVSC_MAX_TARGETS				2
407 #define STORVSC_MAX_CHANNELS				8
408 
409 #define STORVSC_FC_MAX_LUNS_PER_TARGET			255
410 #define STORVSC_FC_MAX_TARGETS				128
411 #define STORVSC_FC_MAX_CHANNELS				8
412 
413 #define STORVSC_IDE_MAX_LUNS_PER_TARGET			64
414 #define STORVSC_IDE_MAX_TARGETS				1
415 #define STORVSC_IDE_MAX_CHANNELS			1
416 
417 struct storvsc_cmd_request {
418 	struct scsi_cmnd *cmd;
419 
420 	struct hv_device *device;
421 
422 	/* Synchronize the request/response if needed */
423 	struct completion wait_event;
424 
425 	struct vmbus_channel_packet_multipage_buffer mpb;
426 	struct vmbus_packet_mpb_array *payload;
427 	u32 payload_sz;
428 
429 	struct vstor_packet vstor_packet;
430 };
431 
432 
433 /* A storvsc device is a device object that contains a vmbus channel */
434 struct storvsc_device {
435 	struct hv_device *device;
436 
437 	bool	 destroy;
438 	bool	 drain_notify;
439 	atomic_t num_outstanding_req;
440 	struct Scsi_Host *host;
441 
442 	wait_queue_head_t waiting_to_drain;
443 
444 	/*
445 	 * Each unique Port/Path/Target represents 1 channel ie scsi
446 	 * controller. In reality, the pathid, targetid is always 0
447 	 * and the port is set by us
448 	 */
449 	unsigned int port_number;
450 	unsigned char path_id;
451 	unsigned char target_id;
452 
453 	/*
454 	 * Max I/O, the device can support.
455 	 */
456 	u32   max_transfer_bytes;
457 	/*
458 	 * Number of sub-channels we will open.
459 	 */
460 	u16 num_sc;
461 	struct vmbus_channel **stor_chns;
462 	/*
463 	 * Mask of CPUs bound to subchannels.
464 	 */
465 	struct cpumask alloced_cpus;
466 	/*
467 	 * Serializes modifications of stor_chns[] from storvsc_do_io()
468 	 * and storvsc_change_target_cpu().
469 	 */
470 	spinlock_t lock;
471 	/* Used for vsc/vsp channel reset process */
472 	struct storvsc_cmd_request init_request;
473 	struct storvsc_cmd_request reset_request;
474 	/*
475 	 * Currently active port and node names for FC devices.
476 	 */
477 	u64 node_name;
478 	u64 port_name;
479 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
480 	struct fc_rport *rport;
481 #endif
482 };
483 
484 struct hv_host_device {
485 	struct hv_device *dev;
486 	unsigned int port;
487 	unsigned char path;
488 	unsigned char target;
489 	struct workqueue_struct *handle_error_wq;
490 	struct work_struct host_scan_work;
491 	struct Scsi_Host *host;
492 };
493 
494 struct storvsc_scan_work {
495 	struct work_struct work;
496 	struct Scsi_Host *host;
497 	u8 lun;
498 	u8 tgt_id;
499 };
500 
501 static void storvsc_device_scan(struct work_struct *work)
502 {
503 	struct storvsc_scan_work *wrk;
504 	struct scsi_device *sdev;
505 
506 	wrk = container_of(work, struct storvsc_scan_work, work);
507 
508 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
509 	if (!sdev)
510 		goto done;
511 	scsi_rescan_device(&sdev->sdev_gendev);
512 	scsi_device_put(sdev);
513 
514 done:
515 	kfree(wrk);
516 }
517 
518 static void storvsc_host_scan(struct work_struct *work)
519 {
520 	struct Scsi_Host *host;
521 	struct scsi_device *sdev;
522 	struct hv_host_device *host_device =
523 		container_of(work, struct hv_host_device, host_scan_work);
524 
525 	host = host_device->host;
526 	/*
527 	 * Before scanning the host, first check to see if any of the
528 	 * currrently known devices have been hot removed. We issue a
529 	 * "unit ready" command against all currently known devices.
530 	 * This I/O will result in an error for devices that have been
531 	 * removed. As part of handling the I/O error, we remove the device.
532 	 *
533 	 * When a LUN is added or removed, the host sends us a signal to
534 	 * scan the host. Thus we are forced to discover the LUNs that
535 	 * may have been removed this way.
536 	 */
537 	mutex_lock(&host->scan_mutex);
538 	shost_for_each_device(sdev, host)
539 		scsi_test_unit_ready(sdev, 1, 1, NULL);
540 	mutex_unlock(&host->scan_mutex);
541 	/*
542 	 * Now scan the host to discover LUNs that may have been added.
543 	 */
544 	scsi_scan_host(host);
545 }
546 
547 static void storvsc_remove_lun(struct work_struct *work)
548 {
549 	struct storvsc_scan_work *wrk;
550 	struct scsi_device *sdev;
551 
552 	wrk = container_of(work, struct storvsc_scan_work, work);
553 	if (!scsi_host_get(wrk->host))
554 		goto done;
555 
556 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
557 
558 	if (sdev) {
559 		scsi_remove_device(sdev);
560 		scsi_device_put(sdev);
561 	}
562 	scsi_host_put(wrk->host);
563 
564 done:
565 	kfree(wrk);
566 }
567 
568 
569 /*
570  * We can get incoming messages from the host that are not in response to
571  * messages that we have sent out. An example of this would be messages
572  * received by the guest to notify dynamic addition/removal of LUNs. To
573  * deal with potential race conditions where the driver may be in the
574  * midst of being unloaded when we might receive an unsolicited message
575  * from the host, we have implemented a mechanism to gurantee sequential
576  * consistency:
577  *
578  * 1) Once the device is marked as being destroyed, we will fail all
579  *    outgoing messages.
580  * 2) We permit incoming messages when the device is being destroyed,
581  *    only to properly account for messages already sent out.
582  */
583 
584 static inline struct storvsc_device *get_out_stor_device(
585 					struct hv_device *device)
586 {
587 	struct storvsc_device *stor_device;
588 
589 	stor_device = hv_get_drvdata(device);
590 
591 	if (stor_device && stor_device->destroy)
592 		stor_device = NULL;
593 
594 	return stor_device;
595 }
596 
597 
598 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
599 {
600 	dev->drain_notify = true;
601 	wait_event(dev->waiting_to_drain,
602 		   atomic_read(&dev->num_outstanding_req) == 0);
603 	dev->drain_notify = false;
604 }
605 
606 static inline struct storvsc_device *get_in_stor_device(
607 					struct hv_device *device)
608 {
609 	struct storvsc_device *stor_device;
610 
611 	stor_device = hv_get_drvdata(device);
612 
613 	if (!stor_device)
614 		goto get_in_err;
615 
616 	/*
617 	 * If the device is being destroyed; allow incoming
618 	 * traffic only to cleanup outstanding requests.
619 	 */
620 
621 	if (stor_device->destroy  &&
622 		(atomic_read(&stor_device->num_outstanding_req) == 0))
623 		stor_device = NULL;
624 
625 get_in_err:
626 	return stor_device;
627 
628 }
629 
630 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
631 				      u32 new)
632 {
633 	struct storvsc_device *stor_device;
634 	struct vmbus_channel *cur_chn;
635 	bool old_is_alloced = false;
636 	struct hv_device *device;
637 	unsigned long flags;
638 	int cpu;
639 
640 	device = channel->primary_channel ?
641 			channel->primary_channel->device_obj
642 				: channel->device_obj;
643 	stor_device = get_out_stor_device(device);
644 	if (!stor_device)
645 		return;
646 
647 	/* See storvsc_do_io() -> get_og_chn(). */
648 	spin_lock_irqsave(&stor_device->lock, flags);
649 
650 	/*
651 	 * Determines if the storvsc device has other channels assigned to
652 	 * the "old" CPU to update the alloced_cpus mask and the stor_chns
653 	 * array.
654 	 */
655 	if (device->channel != channel && device->channel->target_cpu == old) {
656 		cur_chn = device->channel;
657 		old_is_alloced = true;
658 		goto old_is_alloced;
659 	}
660 	list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
661 		if (cur_chn == channel)
662 			continue;
663 		if (cur_chn->target_cpu == old) {
664 			old_is_alloced = true;
665 			goto old_is_alloced;
666 		}
667 	}
668 
669 old_is_alloced:
670 	if (old_is_alloced)
671 		WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
672 	else
673 		cpumask_clear_cpu(old, &stor_device->alloced_cpus);
674 
675 	/* "Flush" the stor_chns array. */
676 	for_each_possible_cpu(cpu) {
677 		if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
678 					cpu, &stor_device->alloced_cpus))
679 			WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
680 	}
681 
682 	WRITE_ONCE(stor_device->stor_chns[new], channel);
683 	cpumask_set_cpu(new, &stor_device->alloced_cpus);
684 
685 	spin_unlock_irqrestore(&stor_device->lock, flags);
686 }
687 
688 static void handle_sc_creation(struct vmbus_channel *new_sc)
689 {
690 	struct hv_device *device = new_sc->primary_channel->device_obj;
691 	struct device *dev = &device->device;
692 	struct storvsc_device *stor_device;
693 	struct vmstorage_channel_properties props;
694 	int ret;
695 
696 	stor_device = get_out_stor_device(device);
697 	if (!stor_device)
698 		return;
699 
700 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
701 
702 	/*
703 	 * The size of vmbus_requestor is an upper bound on the number of requests
704 	 * that can be in-progress at any one time across all channels.
705 	 */
706 	new_sc->rqstor_size = scsi_driver.can_queue;
707 
708 	ret = vmbus_open(new_sc,
709 			 storvsc_ringbuffer_size,
710 			 storvsc_ringbuffer_size,
711 			 (void *)&props,
712 			 sizeof(struct vmstorage_channel_properties),
713 			 storvsc_on_channel_callback, new_sc);
714 
715 	/* In case vmbus_open() fails, we don't use the sub-channel. */
716 	if (ret != 0) {
717 		dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
718 		return;
719 	}
720 
721 	new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
722 
723 	/* Add the sub-channel to the array of available channels. */
724 	stor_device->stor_chns[new_sc->target_cpu] = new_sc;
725 	cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
726 }
727 
728 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
729 {
730 	struct device *dev = &device->device;
731 	struct storvsc_device *stor_device;
732 	int num_sc;
733 	struct storvsc_cmd_request *request;
734 	struct vstor_packet *vstor_packet;
735 	int ret, t;
736 
737 	/*
738 	 * If the number of CPUs is artificially restricted, such as
739 	 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
740 	 * sub-channels >= the number of CPUs. These sub-channels
741 	 * should not be created. The primary channel is already created
742 	 * and assigned to one CPU, so check against # CPUs - 1.
743 	 */
744 	num_sc = min((int)(num_online_cpus() - 1), max_chns);
745 	if (!num_sc)
746 		return;
747 
748 	stor_device = get_out_stor_device(device);
749 	if (!stor_device)
750 		return;
751 
752 	stor_device->num_sc = num_sc;
753 	request = &stor_device->init_request;
754 	vstor_packet = &request->vstor_packet;
755 
756 	/*
757 	 * Establish a handler for dealing with subchannels.
758 	 */
759 	vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
760 
761 	/*
762 	 * Request the host to create sub-channels.
763 	 */
764 	memset(request, 0, sizeof(struct storvsc_cmd_request));
765 	init_completion(&request->wait_event);
766 	vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
767 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
768 	vstor_packet->sub_channel_count = num_sc;
769 
770 	ret = vmbus_sendpacket(device->channel, vstor_packet,
771 			       (sizeof(struct vstor_packet) -
772 			       vmscsi_size_delta),
773 			       (unsigned long)request,
774 			       VM_PKT_DATA_INBAND,
775 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
776 
777 	if (ret != 0) {
778 		dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
779 		return;
780 	}
781 
782 	t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
783 	if (t == 0) {
784 		dev_err(dev, "Failed to create sub-channel: timed out\n");
785 		return;
786 	}
787 
788 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
789 	    vstor_packet->status != 0) {
790 		dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
791 			vstor_packet->operation, vstor_packet->status);
792 		return;
793 	}
794 
795 	/*
796 	 * We need to do nothing here, because vmbus_process_offer()
797 	 * invokes channel->sc_creation_callback, which will open and use
798 	 * the sub-channel(s).
799 	 */
800 }
801 
802 static void cache_wwn(struct storvsc_device *stor_device,
803 		      struct vstor_packet *vstor_packet)
804 {
805 	/*
806 	 * Cache the currently active port and node ww names.
807 	 */
808 	if (vstor_packet->wwn_packet.primary_active) {
809 		stor_device->node_name =
810 			wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
811 		stor_device->port_name =
812 			wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
813 	} else {
814 		stor_device->node_name =
815 			wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
816 		stor_device->port_name =
817 			wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
818 	}
819 }
820 
821 
822 static int storvsc_execute_vstor_op(struct hv_device *device,
823 				    struct storvsc_cmd_request *request,
824 				    bool status_check)
825 {
826 	struct vstor_packet *vstor_packet;
827 	int ret, t;
828 
829 	vstor_packet = &request->vstor_packet;
830 
831 	init_completion(&request->wait_event);
832 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
833 
834 	ret = vmbus_sendpacket(device->channel, vstor_packet,
835 			       (sizeof(struct vstor_packet) -
836 			       vmscsi_size_delta),
837 			       (unsigned long)request,
838 			       VM_PKT_DATA_INBAND,
839 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
840 	if (ret != 0)
841 		return ret;
842 
843 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
844 	if (t == 0)
845 		return -ETIMEDOUT;
846 
847 	if (!status_check)
848 		return ret;
849 
850 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
851 	    vstor_packet->status != 0)
852 		return -EINVAL;
853 
854 	return ret;
855 }
856 
857 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
858 {
859 	struct storvsc_device *stor_device;
860 	struct storvsc_cmd_request *request;
861 	struct vstor_packet *vstor_packet;
862 	int ret, i;
863 	int max_chns;
864 	bool process_sub_channels = false;
865 
866 	stor_device = get_out_stor_device(device);
867 	if (!stor_device)
868 		return -ENODEV;
869 
870 	request = &stor_device->init_request;
871 	vstor_packet = &request->vstor_packet;
872 
873 	/*
874 	 * Now, initiate the vsc/vsp initialization protocol on the open
875 	 * channel
876 	 */
877 	memset(request, 0, sizeof(struct storvsc_cmd_request));
878 	vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
879 	ret = storvsc_execute_vstor_op(device, request, true);
880 	if (ret)
881 		return ret;
882 	/*
883 	 * Query host supported protocol version.
884 	 */
885 
886 	for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) {
887 		/* reuse the packet for version range supported */
888 		memset(vstor_packet, 0, sizeof(struct vstor_packet));
889 		vstor_packet->operation =
890 			VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
891 
892 		vstor_packet->version.major_minor =
893 			vmstor_protocols[i].protocol_version;
894 
895 		/*
896 		 * The revision number is only used in Windows; set it to 0.
897 		 */
898 		vstor_packet->version.revision = 0;
899 		ret = storvsc_execute_vstor_op(device, request, false);
900 		if (ret != 0)
901 			return ret;
902 
903 		if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
904 			return -EINVAL;
905 
906 		if (vstor_packet->status == 0) {
907 			vmstor_proto_version =
908 				vmstor_protocols[i].protocol_version;
909 
910 			sense_buffer_size =
911 				vmstor_protocols[i].sense_buffer_size;
912 
913 			vmscsi_size_delta =
914 				vmstor_protocols[i].vmscsi_size_delta;
915 
916 			break;
917 		}
918 	}
919 
920 	if (vstor_packet->status != 0)
921 		return -EINVAL;
922 
923 
924 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
925 	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
926 	ret = storvsc_execute_vstor_op(device, request, true);
927 	if (ret != 0)
928 		return ret;
929 
930 	/*
931 	 * Check to see if multi-channel support is there.
932 	 * Hosts that implement protocol version of 5.1 and above
933 	 * support multi-channel.
934 	 */
935 	max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
936 
937 	/*
938 	 * Allocate state to manage the sub-channels.
939 	 * We allocate an array based on the numbers of possible CPUs
940 	 * (Hyper-V does not support cpu online/offline).
941 	 * This Array will be sparseley populated with unique
942 	 * channels - primary + sub-channels.
943 	 * We will however populate all the slots to evenly distribute
944 	 * the load.
945 	 */
946 	stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
947 					 GFP_KERNEL);
948 	if (stor_device->stor_chns == NULL)
949 		return -ENOMEM;
950 
951 	device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
952 
953 	stor_device->stor_chns[device->channel->target_cpu] = device->channel;
954 	cpumask_set_cpu(device->channel->target_cpu,
955 			&stor_device->alloced_cpus);
956 
957 	if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) {
958 		if (vstor_packet->storage_channel_properties.flags &
959 		    STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
960 			process_sub_channels = true;
961 	}
962 	stor_device->max_transfer_bytes =
963 		vstor_packet->storage_channel_properties.max_transfer_bytes;
964 
965 	if (!is_fc)
966 		goto done;
967 
968 	/*
969 	 * For FC devices retrieve FC HBA data.
970 	 */
971 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
972 	vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
973 	ret = storvsc_execute_vstor_op(device, request, true);
974 	if (ret != 0)
975 		return ret;
976 
977 	/*
978 	 * Cache the currently active port and node ww names.
979 	 */
980 	cache_wwn(stor_device, vstor_packet);
981 
982 done:
983 
984 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
985 	vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
986 	ret = storvsc_execute_vstor_op(device, request, true);
987 	if (ret != 0)
988 		return ret;
989 
990 	if (process_sub_channels)
991 		handle_multichannel_storage(device, max_chns);
992 
993 	return ret;
994 }
995 
996 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
997 				struct scsi_cmnd *scmnd,
998 				struct Scsi_Host *host,
999 				u8 asc, u8 ascq)
1000 {
1001 	struct storvsc_scan_work *wrk;
1002 	void (*process_err_fn)(struct work_struct *work);
1003 	struct hv_host_device *host_dev = shost_priv(host);
1004 	bool do_work = false;
1005 
1006 	switch (SRB_STATUS(vm_srb->srb_status)) {
1007 	case SRB_STATUS_ERROR:
1008 		/*
1009 		 * Let upper layer deal with error when
1010 		 * sense message is present.
1011 		 */
1012 
1013 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
1014 			break;
1015 		/*
1016 		 * If there is an error; offline the device since all
1017 		 * error recovery strategies would have already been
1018 		 * deployed on the host side. However, if the command
1019 		 * were a pass-through command deal with it appropriately.
1020 		 */
1021 		switch (scmnd->cmnd[0]) {
1022 		case ATA_16:
1023 		case ATA_12:
1024 			set_host_byte(scmnd, DID_PASSTHROUGH);
1025 			break;
1026 		/*
1027 		 * On Some Windows hosts TEST_UNIT_READY command can return
1028 		 * SRB_STATUS_ERROR, let the upper level code deal with it
1029 		 * based on the sense information.
1030 		 */
1031 		case TEST_UNIT_READY:
1032 			break;
1033 		default:
1034 			set_host_byte(scmnd, DID_ERROR);
1035 		}
1036 		break;
1037 	case SRB_STATUS_INVALID_LUN:
1038 		set_host_byte(scmnd, DID_NO_CONNECT);
1039 		do_work = true;
1040 		process_err_fn = storvsc_remove_lun;
1041 		break;
1042 	case SRB_STATUS_ABORTED:
1043 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
1044 		    (asc == 0x2a) && (ascq == 0x9)) {
1045 			do_work = true;
1046 			process_err_fn = storvsc_device_scan;
1047 			/*
1048 			 * Retry the I/O that triggered this.
1049 			 */
1050 			set_host_byte(scmnd, DID_REQUEUE);
1051 		}
1052 		break;
1053 	}
1054 
1055 	if (!do_work)
1056 		return;
1057 
1058 	/*
1059 	 * We need to schedule work to process this error; schedule it.
1060 	 */
1061 	wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1062 	if (!wrk) {
1063 		set_host_byte(scmnd, DID_TARGET_FAILURE);
1064 		return;
1065 	}
1066 
1067 	wrk->host = host;
1068 	wrk->lun = vm_srb->lun;
1069 	wrk->tgt_id = vm_srb->target_id;
1070 	INIT_WORK(&wrk->work, process_err_fn);
1071 	queue_work(host_dev->handle_error_wq, &wrk->work);
1072 }
1073 
1074 
1075 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1076 				       struct storvsc_device *stor_dev)
1077 {
1078 	struct scsi_cmnd *scmnd = cmd_request->cmd;
1079 	struct scsi_sense_hdr sense_hdr;
1080 	struct vmscsi_request *vm_srb;
1081 	u32 data_transfer_length;
1082 	struct Scsi_Host *host;
1083 	u32 payload_sz = cmd_request->payload_sz;
1084 	void *payload = cmd_request->payload;
1085 
1086 	host = stor_dev->host;
1087 
1088 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1089 	data_transfer_length = vm_srb->data_transfer_length;
1090 
1091 	scmnd->result = vm_srb->scsi_status;
1092 
1093 	if (scmnd->result) {
1094 		if (scsi_normalize_sense(scmnd->sense_buffer,
1095 				SCSI_SENSE_BUFFERSIZE, &sense_hdr) &&
1096 		    !(sense_hdr.sense_key == NOT_READY &&
1097 				 sense_hdr.asc == 0x03A) &&
1098 		    do_logging(STORVSC_LOGGING_ERROR))
1099 			scsi_print_sense_hdr(scmnd->device, "storvsc",
1100 					     &sense_hdr);
1101 	}
1102 
1103 	if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1104 		storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1105 					 sense_hdr.ascq);
1106 		/*
1107 		 * The Windows driver set data_transfer_length on
1108 		 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1109 		 * is untouched.  In these cases we set it to 0.
1110 		 */
1111 		if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1112 			data_transfer_length = 0;
1113 	}
1114 
1115 	/* Validate data_transfer_length (from Hyper-V) */
1116 	if (data_transfer_length > cmd_request->payload->range.len)
1117 		data_transfer_length = cmd_request->payload->range.len;
1118 
1119 	scsi_set_resid(scmnd,
1120 		cmd_request->payload->range.len - data_transfer_length);
1121 
1122 	scmnd->scsi_done(scmnd);
1123 
1124 	if (payload_sz >
1125 		sizeof(struct vmbus_channel_packet_multipage_buffer))
1126 		kfree(payload);
1127 }
1128 
1129 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1130 				  struct vstor_packet *vstor_packet,
1131 				  struct storvsc_cmd_request *request)
1132 {
1133 	struct vstor_packet *stor_pkt;
1134 	struct hv_device *device = stor_device->device;
1135 
1136 	stor_pkt = &request->vstor_packet;
1137 
1138 	/*
1139 	 * The current SCSI handling on the host side does
1140 	 * not correctly handle:
1141 	 * INQUIRY command with page code parameter set to 0x80
1142 	 * MODE_SENSE command with cmd[2] == 0x1c
1143 	 *
1144 	 * Setup srb and scsi status so this won't be fatal.
1145 	 * We do this so we can distinguish truly fatal failues
1146 	 * (srb status == 0x4) and off-line the device in that case.
1147 	 */
1148 
1149 	if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1150 	   (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1151 		vstor_packet->vm_srb.scsi_status = 0;
1152 		vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1153 	}
1154 
1155 
1156 	/* Copy over the status...etc */
1157 	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1158 	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1159 
1160 	/* Validate sense_info_length (from Hyper-V) */
1161 	if (vstor_packet->vm_srb.sense_info_length > sense_buffer_size)
1162 		vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1163 
1164 	stor_pkt->vm_srb.sense_info_length =
1165 	vstor_packet->vm_srb.sense_info_length;
1166 
1167 	if (vstor_packet->vm_srb.scsi_status != 0 ||
1168 	    vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS)
1169 		storvsc_log(device, STORVSC_LOGGING_WARN,
1170 			"cmd 0x%x scsi status 0x%x srb status 0x%x\n",
1171 			stor_pkt->vm_srb.cdb[0],
1172 			vstor_packet->vm_srb.scsi_status,
1173 			vstor_packet->vm_srb.srb_status);
1174 
1175 	if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
1176 		/* CHECK_CONDITION */
1177 		if (vstor_packet->vm_srb.srb_status &
1178 			SRB_STATUS_AUTOSENSE_VALID) {
1179 			/* autosense data available */
1180 
1181 			storvsc_log(device, STORVSC_LOGGING_WARN,
1182 				"stor pkt %p autosense data valid - len %d\n",
1183 				request, vstor_packet->vm_srb.sense_info_length);
1184 
1185 			memcpy(request->cmd->sense_buffer,
1186 			       vstor_packet->vm_srb.sense_data,
1187 			       vstor_packet->vm_srb.sense_info_length);
1188 
1189 		}
1190 	}
1191 
1192 	stor_pkt->vm_srb.data_transfer_length =
1193 	vstor_packet->vm_srb.data_transfer_length;
1194 
1195 	storvsc_command_completion(request, stor_device);
1196 
1197 	if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1198 		stor_device->drain_notify)
1199 		wake_up(&stor_device->waiting_to_drain);
1200 
1201 
1202 }
1203 
1204 static void storvsc_on_receive(struct storvsc_device *stor_device,
1205 			     struct vstor_packet *vstor_packet,
1206 			     struct storvsc_cmd_request *request)
1207 {
1208 	struct hv_host_device *host_dev;
1209 	switch (vstor_packet->operation) {
1210 	case VSTOR_OPERATION_COMPLETE_IO:
1211 		storvsc_on_io_completion(stor_device, vstor_packet, request);
1212 		break;
1213 
1214 	case VSTOR_OPERATION_REMOVE_DEVICE:
1215 	case VSTOR_OPERATION_ENUMERATE_BUS:
1216 		host_dev = shost_priv(stor_device->host);
1217 		queue_work(
1218 			host_dev->handle_error_wq, &host_dev->host_scan_work);
1219 		break;
1220 
1221 	case VSTOR_OPERATION_FCHBA_DATA:
1222 		cache_wwn(stor_device, vstor_packet);
1223 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1224 		fc_host_node_name(stor_device->host) = stor_device->node_name;
1225 		fc_host_port_name(stor_device->host) = stor_device->port_name;
1226 #endif
1227 		break;
1228 	default:
1229 		break;
1230 	}
1231 }
1232 
1233 static void storvsc_on_channel_callback(void *context)
1234 {
1235 	struct vmbus_channel *channel = (struct vmbus_channel *)context;
1236 	const struct vmpacket_descriptor *desc;
1237 	struct hv_device *device;
1238 	struct storvsc_device *stor_device;
1239 
1240 	if (channel->primary_channel != NULL)
1241 		device = channel->primary_channel->device_obj;
1242 	else
1243 		device = channel->device_obj;
1244 
1245 	stor_device = get_in_stor_device(device);
1246 	if (!stor_device)
1247 		return;
1248 
1249 	foreach_vmbus_pkt(desc, channel) {
1250 		void *packet = hv_pkt_data(desc);
1251 		struct storvsc_cmd_request *request;
1252 		u64 cmd_rqst;
1253 
1254 		cmd_rqst = vmbus_request_addr(&channel->requestor,
1255 					      desc->trans_id);
1256 		if (cmd_rqst == VMBUS_RQST_ERROR) {
1257 			dev_err(&device->device,
1258 				"Incorrect transaction id\n");
1259 			continue;
1260 		}
1261 
1262 		request = (struct storvsc_cmd_request *)(unsigned long)cmd_rqst;
1263 
1264 		if (request == &stor_device->init_request ||
1265 		    request == &stor_device->reset_request) {
1266 			memcpy(&request->vstor_packet, packet,
1267 			       (sizeof(struct vstor_packet) - vmscsi_size_delta));
1268 			complete(&request->wait_event);
1269 		} else {
1270 			storvsc_on_receive(stor_device, packet, request);
1271 		}
1272 	}
1273 }
1274 
1275 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1276 				  bool is_fc)
1277 {
1278 	struct vmstorage_channel_properties props;
1279 	int ret;
1280 
1281 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1282 
1283 	/*
1284 	 * The size of vmbus_requestor is an upper bound on the number of requests
1285 	 * that can be in-progress at any one time across all channels.
1286 	 */
1287 	device->channel->rqstor_size = scsi_driver.can_queue;
1288 
1289 	ret = vmbus_open(device->channel,
1290 			 ring_size,
1291 			 ring_size,
1292 			 (void *)&props,
1293 			 sizeof(struct vmstorage_channel_properties),
1294 			 storvsc_on_channel_callback, device->channel);
1295 
1296 	if (ret != 0)
1297 		return ret;
1298 
1299 	ret = storvsc_channel_init(device, is_fc);
1300 
1301 	return ret;
1302 }
1303 
1304 static int storvsc_dev_remove(struct hv_device *device)
1305 {
1306 	struct storvsc_device *stor_device;
1307 
1308 	stor_device = hv_get_drvdata(device);
1309 
1310 	stor_device->destroy = true;
1311 
1312 	/* Make sure flag is set before waiting */
1313 	wmb();
1314 
1315 	/*
1316 	 * At this point, all outbound traffic should be disable. We
1317 	 * only allow inbound traffic (responses) to proceed so that
1318 	 * outstanding requests can be completed.
1319 	 */
1320 
1321 	storvsc_wait_to_drain(stor_device);
1322 
1323 	/*
1324 	 * Since we have already drained, we don't need to busy wait
1325 	 * as was done in final_release_stor_device()
1326 	 * Note that we cannot set the ext pointer to NULL until
1327 	 * we have drained - to drain the outgoing packets, we need to
1328 	 * allow incoming packets.
1329 	 */
1330 	hv_set_drvdata(device, NULL);
1331 
1332 	/* Close the channel */
1333 	vmbus_close(device->channel);
1334 
1335 	kfree(stor_device->stor_chns);
1336 	kfree(stor_device);
1337 	return 0;
1338 }
1339 
1340 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1341 					u16 q_num)
1342 {
1343 	u16 slot = 0;
1344 	u16 hash_qnum;
1345 	const struct cpumask *node_mask;
1346 	int num_channels, tgt_cpu;
1347 
1348 	if (stor_device->num_sc == 0) {
1349 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1350 		return stor_device->device->channel;
1351 	}
1352 
1353 	/*
1354 	 * Our channel array is sparsley populated and we
1355 	 * initiated I/O on a processor/hw-q that does not
1356 	 * currently have a designated channel. Fix this.
1357 	 * The strategy is simple:
1358 	 * I. Ensure NUMA locality
1359 	 * II. Distribute evenly (best effort)
1360 	 */
1361 
1362 	node_mask = cpumask_of_node(cpu_to_node(q_num));
1363 
1364 	num_channels = 0;
1365 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1366 		if (cpumask_test_cpu(tgt_cpu, node_mask))
1367 			num_channels++;
1368 	}
1369 	if (num_channels == 0) {
1370 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1371 		return stor_device->device->channel;
1372 	}
1373 
1374 	hash_qnum = q_num;
1375 	while (hash_qnum >= num_channels)
1376 		hash_qnum -= num_channels;
1377 
1378 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1379 		if (!cpumask_test_cpu(tgt_cpu, node_mask))
1380 			continue;
1381 		if (slot == hash_qnum)
1382 			break;
1383 		slot++;
1384 	}
1385 
1386 	stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1387 
1388 	return stor_device->stor_chns[q_num];
1389 }
1390 
1391 
1392 static int storvsc_do_io(struct hv_device *device,
1393 			 struct storvsc_cmd_request *request, u16 q_num)
1394 {
1395 	struct storvsc_device *stor_device;
1396 	struct vstor_packet *vstor_packet;
1397 	struct vmbus_channel *outgoing_channel, *channel;
1398 	unsigned long flags;
1399 	int ret = 0;
1400 	const struct cpumask *node_mask;
1401 	int tgt_cpu;
1402 
1403 	vstor_packet = &request->vstor_packet;
1404 	stor_device = get_out_stor_device(device);
1405 
1406 	if (!stor_device)
1407 		return -ENODEV;
1408 
1409 
1410 	request->device  = device;
1411 	/*
1412 	 * Select an appropriate channel to send the request out.
1413 	 */
1414 	/* See storvsc_change_target_cpu(). */
1415 	outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1416 	if (outgoing_channel != NULL) {
1417 		if (outgoing_channel->target_cpu == q_num) {
1418 			/*
1419 			 * Ideally, we want to pick a different channel if
1420 			 * available on the same NUMA node.
1421 			 */
1422 			node_mask = cpumask_of_node(cpu_to_node(q_num));
1423 			for_each_cpu_wrap(tgt_cpu,
1424 				 &stor_device->alloced_cpus, q_num + 1) {
1425 				if (!cpumask_test_cpu(tgt_cpu, node_mask))
1426 					continue;
1427 				if (tgt_cpu == q_num)
1428 					continue;
1429 				channel = READ_ONCE(
1430 					stor_device->stor_chns[tgt_cpu]);
1431 				if (channel == NULL)
1432 					continue;
1433 				if (hv_get_avail_to_write_percent(
1434 							&channel->outbound)
1435 						> ring_avail_percent_lowater) {
1436 					outgoing_channel = channel;
1437 					goto found_channel;
1438 				}
1439 			}
1440 
1441 			/*
1442 			 * All the other channels on the same NUMA node are
1443 			 * busy. Try to use the channel on the current CPU
1444 			 */
1445 			if (hv_get_avail_to_write_percent(
1446 						&outgoing_channel->outbound)
1447 					> ring_avail_percent_lowater)
1448 				goto found_channel;
1449 
1450 			/*
1451 			 * If we reach here, all the channels on the current
1452 			 * NUMA node are busy. Try to find a channel in
1453 			 * other NUMA nodes
1454 			 */
1455 			for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1456 				if (cpumask_test_cpu(tgt_cpu, node_mask))
1457 					continue;
1458 				channel = READ_ONCE(
1459 					stor_device->stor_chns[tgt_cpu]);
1460 				if (channel == NULL)
1461 					continue;
1462 				if (hv_get_avail_to_write_percent(
1463 							&channel->outbound)
1464 						> ring_avail_percent_lowater) {
1465 					outgoing_channel = channel;
1466 					goto found_channel;
1467 				}
1468 			}
1469 		}
1470 	} else {
1471 		spin_lock_irqsave(&stor_device->lock, flags);
1472 		outgoing_channel = stor_device->stor_chns[q_num];
1473 		if (outgoing_channel != NULL) {
1474 			spin_unlock_irqrestore(&stor_device->lock, flags);
1475 			goto found_channel;
1476 		}
1477 		outgoing_channel = get_og_chn(stor_device, q_num);
1478 		spin_unlock_irqrestore(&stor_device->lock, flags);
1479 	}
1480 
1481 found_channel:
1482 	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1483 
1484 	vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1485 					vmscsi_size_delta);
1486 
1487 
1488 	vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1489 
1490 
1491 	vstor_packet->vm_srb.data_transfer_length =
1492 	request->payload->range.len;
1493 
1494 	vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1495 
1496 	if (request->payload->range.len) {
1497 
1498 		ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1499 				request->payload, request->payload_sz,
1500 				vstor_packet,
1501 				(sizeof(struct vstor_packet) -
1502 				vmscsi_size_delta),
1503 				(unsigned long)request);
1504 	} else {
1505 		ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1506 			       (sizeof(struct vstor_packet) -
1507 				vmscsi_size_delta),
1508 			       (unsigned long)request,
1509 			       VM_PKT_DATA_INBAND,
1510 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1511 	}
1512 
1513 	if (ret != 0)
1514 		return ret;
1515 
1516 	atomic_inc(&stor_device->num_outstanding_req);
1517 
1518 	return ret;
1519 }
1520 
1521 static int storvsc_device_alloc(struct scsi_device *sdevice)
1522 {
1523 	/*
1524 	 * Set blist flag to permit the reading of the VPD pages even when
1525 	 * the target may claim SPC-2 compliance. MSFT targets currently
1526 	 * claim SPC-2 compliance while they implement post SPC-2 features.
1527 	 * With this flag we can correctly handle WRITE_SAME_16 issues.
1528 	 *
1529 	 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1530 	 * still supports REPORT LUN.
1531 	 */
1532 	sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1533 
1534 	return 0;
1535 }
1536 
1537 static int storvsc_device_configure(struct scsi_device *sdevice)
1538 {
1539 	blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1540 
1541 	sdevice->no_write_same = 1;
1542 
1543 	/*
1544 	 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1545 	 * if the device is a MSFT virtual device.  If the host is
1546 	 * WIN10 or newer, allow write_same.
1547 	 */
1548 	if (!strncmp(sdevice->vendor, "Msft", 4)) {
1549 		switch (vmstor_proto_version) {
1550 		case VMSTOR_PROTO_VERSION_WIN8:
1551 		case VMSTOR_PROTO_VERSION_WIN8_1:
1552 			sdevice->scsi_level = SCSI_SPC_3;
1553 			break;
1554 		}
1555 
1556 		if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1557 			sdevice->no_write_same = 0;
1558 	}
1559 
1560 	return 0;
1561 }
1562 
1563 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1564 			   sector_t capacity, int *info)
1565 {
1566 	sector_t nsect = capacity;
1567 	sector_t cylinders = nsect;
1568 	int heads, sectors_pt;
1569 
1570 	/*
1571 	 * We are making up these values; let us keep it simple.
1572 	 */
1573 	heads = 0xff;
1574 	sectors_pt = 0x3f;      /* Sectors per track */
1575 	sector_div(cylinders, heads * sectors_pt);
1576 	if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1577 		cylinders = 0xffff;
1578 
1579 	info[0] = heads;
1580 	info[1] = sectors_pt;
1581 	info[2] = (int)cylinders;
1582 
1583 	return 0;
1584 }
1585 
1586 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1587 {
1588 	struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1589 	struct hv_device *device = host_dev->dev;
1590 
1591 	struct storvsc_device *stor_device;
1592 	struct storvsc_cmd_request *request;
1593 	struct vstor_packet *vstor_packet;
1594 	int ret, t;
1595 
1596 	stor_device = get_out_stor_device(device);
1597 	if (!stor_device)
1598 		return FAILED;
1599 
1600 	request = &stor_device->reset_request;
1601 	vstor_packet = &request->vstor_packet;
1602 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1603 
1604 	init_completion(&request->wait_event);
1605 
1606 	vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1607 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1608 	vstor_packet->vm_srb.path_id = stor_device->path_id;
1609 
1610 	ret = vmbus_sendpacket(device->channel, vstor_packet,
1611 			       (sizeof(struct vstor_packet) -
1612 				vmscsi_size_delta),
1613 			       (unsigned long)&stor_device->reset_request,
1614 			       VM_PKT_DATA_INBAND,
1615 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1616 	if (ret != 0)
1617 		return FAILED;
1618 
1619 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1620 	if (t == 0)
1621 		return TIMEOUT_ERROR;
1622 
1623 
1624 	/*
1625 	 * At this point, all outstanding requests in the adapter
1626 	 * should have been flushed out and return to us
1627 	 * There is a potential race here where the host may be in
1628 	 * the process of responding when we return from here.
1629 	 * Just wait for all in-transit packets to be accounted for
1630 	 * before we return from here.
1631 	 */
1632 	storvsc_wait_to_drain(stor_device);
1633 
1634 	return SUCCESS;
1635 }
1636 
1637 /*
1638  * The host guarantees to respond to each command, although I/O latencies might
1639  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1640  * chance to perform EH.
1641  */
1642 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1643 {
1644 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1645 	if (scmnd->device->host->transportt == fc_transport_template)
1646 		return fc_eh_timed_out(scmnd);
1647 #endif
1648 	return BLK_EH_RESET_TIMER;
1649 }
1650 
1651 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1652 {
1653 	bool allowed = true;
1654 	u8 scsi_op = scmnd->cmnd[0];
1655 
1656 	switch (scsi_op) {
1657 	/* the host does not handle WRITE_SAME, log accident usage */
1658 	case WRITE_SAME:
1659 	/*
1660 	 * smartd sends this command and the host does not handle
1661 	 * this. So, don't send it.
1662 	 */
1663 	case SET_WINDOW:
1664 		scmnd->result = ILLEGAL_REQUEST << 16;
1665 		allowed = false;
1666 		break;
1667 	default:
1668 		break;
1669 	}
1670 	return allowed;
1671 }
1672 
1673 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1674 {
1675 	int ret;
1676 	struct hv_host_device *host_dev = shost_priv(host);
1677 	struct hv_device *dev = host_dev->dev;
1678 	struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1679 	int i;
1680 	struct scatterlist *sgl;
1681 	unsigned int sg_count = 0;
1682 	struct vmscsi_request *vm_srb;
1683 	struct scatterlist *cur_sgl;
1684 	struct vmbus_packet_mpb_array  *payload;
1685 	u32 payload_sz;
1686 	u32 length;
1687 
1688 	if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1689 		/*
1690 		 * On legacy hosts filter unimplemented commands.
1691 		 * Future hosts are expected to correctly handle
1692 		 * unsupported commands. Furthermore, it is
1693 		 * possible that some of the currently
1694 		 * unsupported commands maybe supported in
1695 		 * future versions of the host.
1696 		 */
1697 		if (!storvsc_scsi_cmd_ok(scmnd)) {
1698 			scmnd->scsi_done(scmnd);
1699 			return 0;
1700 		}
1701 	}
1702 
1703 	/* Setup the cmd request */
1704 	cmd_request->cmd = scmnd;
1705 
1706 	memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1707 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1708 	vm_srb->win8_extension.time_out_value = 60;
1709 
1710 	vm_srb->win8_extension.srb_flags |=
1711 		SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1712 
1713 	if (scmnd->device->tagged_supported) {
1714 		vm_srb->win8_extension.srb_flags |=
1715 		(SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1716 		vm_srb->win8_extension.queue_tag = SP_UNTAGGED;
1717 		vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST;
1718 	}
1719 
1720 	/* Build the SRB */
1721 	switch (scmnd->sc_data_direction) {
1722 	case DMA_TO_DEVICE:
1723 		vm_srb->data_in = WRITE_TYPE;
1724 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1725 		break;
1726 	case DMA_FROM_DEVICE:
1727 		vm_srb->data_in = READ_TYPE;
1728 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1729 		break;
1730 	case DMA_NONE:
1731 		vm_srb->data_in = UNKNOWN_TYPE;
1732 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1733 		break;
1734 	default:
1735 		/*
1736 		 * This is DMA_BIDIRECTIONAL or something else we are never
1737 		 * supposed to see here.
1738 		 */
1739 		WARN(1, "Unexpected data direction: %d\n",
1740 		     scmnd->sc_data_direction);
1741 		return -EINVAL;
1742 	}
1743 
1744 
1745 	vm_srb->port_number = host_dev->port;
1746 	vm_srb->path_id = scmnd->device->channel;
1747 	vm_srb->target_id = scmnd->device->id;
1748 	vm_srb->lun = scmnd->device->lun;
1749 
1750 	vm_srb->cdb_length = scmnd->cmd_len;
1751 
1752 	memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1753 
1754 	sgl = (struct scatterlist *)scsi_sglist(scmnd);
1755 	sg_count = scsi_sg_count(scmnd);
1756 
1757 	length = scsi_bufflen(scmnd);
1758 	payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1759 	payload_sz = sizeof(cmd_request->mpb);
1760 
1761 	if (sg_count) {
1762 		unsigned int hvpgoff = 0;
1763 		unsigned long offset_in_hvpg = sgl->offset & ~HV_HYP_PAGE_MASK;
1764 		unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1765 		u64 hvpfn;
1766 
1767 		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1768 
1769 			payload_sz = (hvpg_count * sizeof(u64) +
1770 				      sizeof(struct vmbus_packet_mpb_array));
1771 			payload = kzalloc(payload_sz, GFP_ATOMIC);
1772 			if (!payload)
1773 				return SCSI_MLQUEUE_DEVICE_BUSY;
1774 		}
1775 
1776 		/*
1777 		 * sgl is a list of PAGEs, and payload->range.pfn_array
1778 		 * expects the page number in the unit of HV_HYP_PAGE_SIZE (the
1779 		 * page size that Hyper-V uses, so here we need to divide PAGEs
1780 		 * into HV_HYP_PAGE in case that PAGE_SIZE > HV_HYP_PAGE_SIZE.
1781 		 * Besides, payload->range.offset should be the offset in one
1782 		 * HV_HYP_PAGE.
1783 		 */
1784 		payload->range.len = length;
1785 		payload->range.offset = offset_in_hvpg;
1786 		hvpgoff = sgl->offset >> HV_HYP_PAGE_SHIFT;
1787 
1788 		cur_sgl = sgl;
1789 		for (i = 0; i < hvpg_count; i++) {
1790 			/*
1791 			 * 'i' is the index of hv pages in the payload and
1792 			 * 'hvpgoff' is the offset (in hv pages) of the first
1793 			 * hv page in the the first page. The relationship
1794 			 * between the sum of 'i' and 'hvpgoff' and the offset
1795 			 * (in hv pages) in a payload page ('hvpgoff_in_page')
1796 			 * is as follow:
1797 			 *
1798 			 * |------------------ PAGE -------------------|
1799 			 * |   NR_HV_HYP_PAGES_IN_PAGE hvpgs in total  |
1800 			 * |hvpg|hvpg| ...              |hvpg|... |hvpg|
1801 			 * ^         ^                                 ^                 ^
1802 			 * +-hvpgoff-+                                 +-hvpgoff_in_page-+
1803 			 *           ^                                                   |
1804 			 *           +--------------------- i ---------------------------+
1805 			 */
1806 			unsigned int hvpgoff_in_page =
1807 				(i + hvpgoff) % NR_HV_HYP_PAGES_IN_PAGE;
1808 
1809 			/*
1810 			 * Two cases that we need to fetch a page:
1811 			 * 1) i == 0, the first step or
1812 			 * 2) hvpgoff_in_page == 0, when we reach the boundary
1813 			 *    of a page.
1814 			 */
1815 			if (hvpgoff_in_page == 0 || i == 0) {
1816 				hvpfn = page_to_hvpfn(sg_page(cur_sgl));
1817 				cur_sgl = sg_next(cur_sgl);
1818 			}
1819 
1820 			payload->range.pfn_array[i] = hvpfn + hvpgoff_in_page;
1821 		}
1822 	}
1823 
1824 	cmd_request->payload = payload;
1825 	cmd_request->payload_sz = payload_sz;
1826 
1827 	/* Invokes the vsc to start an IO */
1828 	ret = storvsc_do_io(dev, cmd_request, get_cpu());
1829 	put_cpu();
1830 
1831 	if (ret == -EAGAIN) {
1832 		if (payload_sz > sizeof(cmd_request->mpb))
1833 			kfree(payload);
1834 		/* no more space */
1835 		return SCSI_MLQUEUE_DEVICE_BUSY;
1836 	}
1837 
1838 	return 0;
1839 }
1840 
1841 static struct scsi_host_template scsi_driver = {
1842 	.module	=		THIS_MODULE,
1843 	.name =			"storvsc_host_t",
1844 	.cmd_size =             sizeof(struct storvsc_cmd_request),
1845 	.bios_param =		storvsc_get_chs,
1846 	.queuecommand =		storvsc_queuecommand,
1847 	.eh_host_reset_handler =	storvsc_host_reset_handler,
1848 	.proc_name =		"storvsc_host",
1849 	.eh_timed_out =		storvsc_eh_timed_out,
1850 	.slave_alloc =		storvsc_device_alloc,
1851 	.slave_configure =	storvsc_device_configure,
1852 	.cmd_per_lun =		2048,
1853 	.this_id =		-1,
1854 	/* Make sure we dont get a sg segment crosses a page boundary */
1855 	.dma_boundary =		PAGE_SIZE-1,
1856 	/* Ensure there are no gaps in presented sgls */
1857 	.virt_boundary_mask =	PAGE_SIZE-1,
1858 	.no_write_same =	1,
1859 	.track_queue_depth =	1,
1860 	.change_queue_depth =	storvsc_change_queue_depth,
1861 };
1862 
1863 enum {
1864 	SCSI_GUID,
1865 	IDE_GUID,
1866 	SFC_GUID,
1867 };
1868 
1869 static const struct hv_vmbus_device_id id_table[] = {
1870 	/* SCSI guid */
1871 	{ HV_SCSI_GUID,
1872 	  .driver_data = SCSI_GUID
1873 	},
1874 	/* IDE guid */
1875 	{ HV_IDE_GUID,
1876 	  .driver_data = IDE_GUID
1877 	},
1878 	/* Fibre Channel GUID */
1879 	{
1880 	  HV_SYNTHFC_GUID,
1881 	  .driver_data = SFC_GUID
1882 	},
1883 	{ },
1884 };
1885 
1886 MODULE_DEVICE_TABLE(vmbus, id_table);
1887 
1888 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1889 
1890 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1891 {
1892 	return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1893 }
1894 
1895 static int storvsc_probe(struct hv_device *device,
1896 			const struct hv_vmbus_device_id *dev_id)
1897 {
1898 	int ret;
1899 	int num_cpus = num_online_cpus();
1900 	struct Scsi_Host *host;
1901 	struct hv_host_device *host_dev;
1902 	bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1903 	bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1904 	int target = 0;
1905 	struct storvsc_device *stor_device;
1906 	int max_luns_per_target;
1907 	int max_targets;
1908 	int max_channels;
1909 	int max_sub_channels = 0;
1910 
1911 	/*
1912 	 * Based on the windows host we are running on,
1913 	 * set state to properly communicate with the host.
1914 	 */
1915 
1916 	if (vmbus_proto_version < VERSION_WIN8) {
1917 		max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1918 		max_targets = STORVSC_IDE_MAX_TARGETS;
1919 		max_channels = STORVSC_IDE_MAX_CHANNELS;
1920 	} else {
1921 		max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
1922 		max_targets = STORVSC_MAX_TARGETS;
1923 		max_channels = STORVSC_MAX_CHANNELS;
1924 		/*
1925 		 * On Windows8 and above, we support sub-channels for storage
1926 		 * on SCSI and FC controllers.
1927 		 * The number of sub-channels offerred is based on the number of
1928 		 * VCPUs in the guest.
1929 		 */
1930 		if (!dev_is_ide)
1931 			max_sub_channels =
1932 				(num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1933 	}
1934 
1935 	scsi_driver.can_queue = max_outstanding_req_per_channel *
1936 				(max_sub_channels + 1) *
1937 				(100 - ring_avail_percent_lowater) / 100;
1938 
1939 	host = scsi_host_alloc(&scsi_driver,
1940 			       sizeof(struct hv_host_device));
1941 	if (!host)
1942 		return -ENOMEM;
1943 
1944 	host_dev = shost_priv(host);
1945 	memset(host_dev, 0, sizeof(struct hv_host_device));
1946 
1947 	host_dev->port = host->host_no;
1948 	host_dev->dev = device;
1949 	host_dev->host = host;
1950 
1951 
1952 	stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1953 	if (!stor_device) {
1954 		ret = -ENOMEM;
1955 		goto err_out0;
1956 	}
1957 
1958 	stor_device->destroy = false;
1959 	init_waitqueue_head(&stor_device->waiting_to_drain);
1960 	stor_device->device = device;
1961 	stor_device->host = host;
1962 	spin_lock_init(&stor_device->lock);
1963 	hv_set_drvdata(device, stor_device);
1964 
1965 	stor_device->port_number = host->host_no;
1966 	ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1967 	if (ret)
1968 		goto err_out1;
1969 
1970 	host_dev->path = stor_device->path_id;
1971 	host_dev->target = stor_device->target_id;
1972 
1973 	switch (dev_id->driver_data) {
1974 	case SFC_GUID:
1975 		host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1976 		host->max_id = STORVSC_FC_MAX_TARGETS;
1977 		host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1978 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1979 		host->transportt = fc_transport_template;
1980 #endif
1981 		break;
1982 
1983 	case SCSI_GUID:
1984 		host->max_lun = max_luns_per_target;
1985 		host->max_id = max_targets;
1986 		host->max_channel = max_channels - 1;
1987 		break;
1988 
1989 	default:
1990 		host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1991 		host->max_id = STORVSC_IDE_MAX_TARGETS;
1992 		host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
1993 		break;
1994 	}
1995 	/* max cmd length */
1996 	host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1997 
1998 	/*
1999 	 * set the table size based on the info we got
2000 	 * from the host.
2001 	 */
2002 	host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
2003 	/*
2004 	 * For non-IDE disks, the host supports multiple channels.
2005 	 * Set the number of HW queues we are supporting.
2006 	 */
2007 	if (!dev_is_ide)
2008 		host->nr_hw_queues = num_present_cpus();
2009 
2010 	/*
2011 	 * Set the error handler work queue.
2012 	 */
2013 	host_dev->handle_error_wq =
2014 			alloc_ordered_workqueue("storvsc_error_wq_%d",
2015 						WQ_MEM_RECLAIM,
2016 						host->host_no);
2017 	if (!host_dev->handle_error_wq) {
2018 		ret = -ENOMEM;
2019 		goto err_out2;
2020 	}
2021 	INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2022 	/* Register the HBA and start the scsi bus scan */
2023 	ret = scsi_add_host(host, &device->device);
2024 	if (ret != 0)
2025 		goto err_out3;
2026 
2027 	if (!dev_is_ide) {
2028 		scsi_scan_host(host);
2029 	} else {
2030 		target = (device->dev_instance.b[5] << 8 |
2031 			 device->dev_instance.b[4]);
2032 		ret = scsi_add_device(host, 0, target, 0);
2033 		if (ret)
2034 			goto err_out4;
2035 	}
2036 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2037 	if (host->transportt == fc_transport_template) {
2038 		struct fc_rport_identifiers ids = {
2039 			.roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2040 		};
2041 
2042 		fc_host_node_name(host) = stor_device->node_name;
2043 		fc_host_port_name(host) = stor_device->port_name;
2044 		stor_device->rport = fc_remote_port_add(host, 0, &ids);
2045 		if (!stor_device->rport) {
2046 			ret = -ENOMEM;
2047 			goto err_out4;
2048 		}
2049 	}
2050 #endif
2051 	return 0;
2052 
2053 err_out4:
2054 	scsi_remove_host(host);
2055 
2056 err_out3:
2057 	destroy_workqueue(host_dev->handle_error_wq);
2058 
2059 err_out2:
2060 	/*
2061 	 * Once we have connected with the host, we would need to
2062 	 * to invoke storvsc_dev_remove() to rollback this state and
2063 	 * this call also frees up the stor_device; hence the jump around
2064 	 * err_out1 label.
2065 	 */
2066 	storvsc_dev_remove(device);
2067 	goto err_out0;
2068 
2069 err_out1:
2070 	kfree(stor_device->stor_chns);
2071 	kfree(stor_device);
2072 
2073 err_out0:
2074 	scsi_host_put(host);
2075 	return ret;
2076 }
2077 
2078 /* Change a scsi target's queue depth */
2079 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2080 {
2081 	if (queue_depth > scsi_driver.can_queue)
2082 		queue_depth = scsi_driver.can_queue;
2083 
2084 	return scsi_change_queue_depth(sdev, queue_depth);
2085 }
2086 
2087 static int storvsc_remove(struct hv_device *dev)
2088 {
2089 	struct storvsc_device *stor_device = hv_get_drvdata(dev);
2090 	struct Scsi_Host *host = stor_device->host;
2091 	struct hv_host_device *host_dev = shost_priv(host);
2092 
2093 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2094 	if (host->transportt == fc_transport_template) {
2095 		fc_remote_port_delete(stor_device->rport);
2096 		fc_remove_host(host);
2097 	}
2098 #endif
2099 	destroy_workqueue(host_dev->handle_error_wq);
2100 	scsi_remove_host(host);
2101 	storvsc_dev_remove(dev);
2102 	scsi_host_put(host);
2103 
2104 	return 0;
2105 }
2106 
2107 static int storvsc_suspend(struct hv_device *hv_dev)
2108 {
2109 	struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2110 	struct Scsi_Host *host = stor_device->host;
2111 	struct hv_host_device *host_dev = shost_priv(host);
2112 
2113 	storvsc_wait_to_drain(stor_device);
2114 
2115 	drain_workqueue(host_dev->handle_error_wq);
2116 
2117 	vmbus_close(hv_dev->channel);
2118 
2119 	kfree(stor_device->stor_chns);
2120 	stor_device->stor_chns = NULL;
2121 
2122 	cpumask_clear(&stor_device->alloced_cpus);
2123 
2124 	return 0;
2125 }
2126 
2127 static int storvsc_resume(struct hv_device *hv_dev)
2128 {
2129 	int ret;
2130 
2131 	ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2132 				     hv_dev_is_fc(hv_dev));
2133 	return ret;
2134 }
2135 
2136 static struct hv_driver storvsc_drv = {
2137 	.name = KBUILD_MODNAME,
2138 	.id_table = id_table,
2139 	.probe = storvsc_probe,
2140 	.remove = storvsc_remove,
2141 	.suspend = storvsc_suspend,
2142 	.resume = storvsc_resume,
2143 	.driver = {
2144 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
2145 	},
2146 };
2147 
2148 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2149 static struct fc_function_template fc_transport_functions = {
2150 	.show_host_node_name = 1,
2151 	.show_host_port_name = 1,
2152 };
2153 #endif
2154 
2155 static int __init storvsc_drv_init(void)
2156 {
2157 	int ret;
2158 
2159 	/*
2160 	 * Divide the ring buffer data size (which is 1 page less
2161 	 * than the ring buffer size since that page is reserved for
2162 	 * the ring buffer indices) by the max request size (which is
2163 	 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2164 	 */
2165 	max_outstanding_req_per_channel =
2166 		((storvsc_ringbuffer_size - PAGE_SIZE) /
2167 		ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2168 		sizeof(struct vstor_packet) + sizeof(u64) -
2169 		vmscsi_size_delta,
2170 		sizeof(u64)));
2171 
2172 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2173 	fc_transport_template = fc_attach_transport(&fc_transport_functions);
2174 	if (!fc_transport_template)
2175 		return -ENODEV;
2176 #endif
2177 
2178 	ret = vmbus_driver_register(&storvsc_drv);
2179 
2180 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2181 	if (ret)
2182 		fc_release_transport(fc_transport_template);
2183 #endif
2184 
2185 	return ret;
2186 }
2187 
2188 static void __exit storvsc_drv_exit(void)
2189 {
2190 	vmbus_driver_unregister(&storvsc_drv);
2191 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2192 	fc_release_transport(fc_transport_template);
2193 #endif
2194 }
2195 
2196 MODULE_LICENSE("GPL");
2197 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2198 module_init(storvsc_drv_init);
2199 module_exit(storvsc_drv_exit);
2200