xref: /linux/drivers/scsi/storvsc_drv.c (revision 48dea9a700c8728cc31a1dd44588b97578de86ee)
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 void storvsc_on_channel_callback(void *context);
403 
404 #define STORVSC_MAX_LUNS_PER_TARGET			255
405 #define STORVSC_MAX_TARGETS				2
406 #define STORVSC_MAX_CHANNELS				8
407 
408 #define STORVSC_FC_MAX_LUNS_PER_TARGET			255
409 #define STORVSC_FC_MAX_TARGETS				128
410 #define STORVSC_FC_MAX_CHANNELS				8
411 
412 #define STORVSC_IDE_MAX_LUNS_PER_TARGET			64
413 #define STORVSC_IDE_MAX_TARGETS				1
414 #define STORVSC_IDE_MAX_CHANNELS			1
415 
416 struct storvsc_cmd_request {
417 	struct scsi_cmnd *cmd;
418 
419 	struct hv_device *device;
420 
421 	/* Synchronize the request/response if needed */
422 	struct completion wait_event;
423 
424 	struct vmbus_channel_packet_multipage_buffer mpb;
425 	struct vmbus_packet_mpb_array *payload;
426 	u32 payload_sz;
427 
428 	struct vstor_packet vstor_packet;
429 };
430 
431 
432 /* A storvsc device is a device object that contains a vmbus channel */
433 struct storvsc_device {
434 	struct hv_device *device;
435 
436 	bool	 destroy;
437 	bool	 drain_notify;
438 	atomic_t num_outstanding_req;
439 	struct Scsi_Host *host;
440 
441 	wait_queue_head_t waiting_to_drain;
442 
443 	/*
444 	 * Each unique Port/Path/Target represents 1 channel ie scsi
445 	 * controller. In reality, the pathid, targetid is always 0
446 	 * and the port is set by us
447 	 */
448 	unsigned int port_number;
449 	unsigned char path_id;
450 	unsigned char target_id;
451 
452 	/*
453 	 * Max I/O, the device can support.
454 	 */
455 	u32   max_transfer_bytes;
456 	/*
457 	 * Number of sub-channels we will open.
458 	 */
459 	u16 num_sc;
460 	struct vmbus_channel **stor_chns;
461 	/*
462 	 * Mask of CPUs bound to subchannels.
463 	 */
464 	struct cpumask alloced_cpus;
465 	/*
466 	 * Serializes modifications of stor_chns[] from storvsc_do_io()
467 	 * and storvsc_change_target_cpu().
468 	 */
469 	spinlock_t lock;
470 	/* Used for vsc/vsp channel reset process */
471 	struct storvsc_cmd_request init_request;
472 	struct storvsc_cmd_request reset_request;
473 	/*
474 	 * Currently active port and node names for FC devices.
475 	 */
476 	u64 node_name;
477 	u64 port_name;
478 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
479 	struct fc_rport *rport;
480 #endif
481 };
482 
483 struct hv_host_device {
484 	struct hv_device *dev;
485 	unsigned int port;
486 	unsigned char path;
487 	unsigned char target;
488 	struct workqueue_struct *handle_error_wq;
489 	struct work_struct host_scan_work;
490 	struct Scsi_Host *host;
491 };
492 
493 struct storvsc_scan_work {
494 	struct work_struct work;
495 	struct Scsi_Host *host;
496 	u8 lun;
497 	u8 tgt_id;
498 };
499 
500 static void storvsc_device_scan(struct work_struct *work)
501 {
502 	struct storvsc_scan_work *wrk;
503 	struct scsi_device *sdev;
504 
505 	wrk = container_of(work, struct storvsc_scan_work, work);
506 
507 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
508 	if (!sdev)
509 		goto done;
510 	scsi_rescan_device(&sdev->sdev_gendev);
511 	scsi_device_put(sdev);
512 
513 done:
514 	kfree(wrk);
515 }
516 
517 static void storvsc_host_scan(struct work_struct *work)
518 {
519 	struct Scsi_Host *host;
520 	struct scsi_device *sdev;
521 	struct hv_host_device *host_device =
522 		container_of(work, struct hv_host_device, host_scan_work);
523 
524 	host = host_device->host;
525 	/*
526 	 * Before scanning the host, first check to see if any of the
527 	 * currrently known devices have been hot removed. We issue a
528 	 * "unit ready" command against all currently known devices.
529 	 * This I/O will result in an error for devices that have been
530 	 * removed. As part of handling the I/O error, we remove the device.
531 	 *
532 	 * When a LUN is added or removed, the host sends us a signal to
533 	 * scan the host. Thus we are forced to discover the LUNs that
534 	 * may have been removed this way.
535 	 */
536 	mutex_lock(&host->scan_mutex);
537 	shost_for_each_device(sdev, host)
538 		scsi_test_unit_ready(sdev, 1, 1, NULL);
539 	mutex_unlock(&host->scan_mutex);
540 	/*
541 	 * Now scan the host to discover LUNs that may have been added.
542 	 */
543 	scsi_scan_host(host);
544 }
545 
546 static void storvsc_remove_lun(struct work_struct *work)
547 {
548 	struct storvsc_scan_work *wrk;
549 	struct scsi_device *sdev;
550 
551 	wrk = container_of(work, struct storvsc_scan_work, work);
552 	if (!scsi_host_get(wrk->host))
553 		goto done;
554 
555 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
556 
557 	if (sdev) {
558 		scsi_remove_device(sdev);
559 		scsi_device_put(sdev);
560 	}
561 	scsi_host_put(wrk->host);
562 
563 done:
564 	kfree(wrk);
565 }
566 
567 
568 /*
569  * We can get incoming messages from the host that are not in response to
570  * messages that we have sent out. An example of this would be messages
571  * received by the guest to notify dynamic addition/removal of LUNs. To
572  * deal with potential race conditions where the driver may be in the
573  * midst of being unloaded when we might receive an unsolicited message
574  * from the host, we have implemented a mechanism to gurantee sequential
575  * consistency:
576  *
577  * 1) Once the device is marked as being destroyed, we will fail all
578  *    outgoing messages.
579  * 2) We permit incoming messages when the device is being destroyed,
580  *    only to properly account for messages already sent out.
581  */
582 
583 static inline struct storvsc_device *get_out_stor_device(
584 					struct hv_device *device)
585 {
586 	struct storvsc_device *stor_device;
587 
588 	stor_device = hv_get_drvdata(device);
589 
590 	if (stor_device && stor_device->destroy)
591 		stor_device = NULL;
592 
593 	return stor_device;
594 }
595 
596 
597 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
598 {
599 	dev->drain_notify = true;
600 	wait_event(dev->waiting_to_drain,
601 		   atomic_read(&dev->num_outstanding_req) == 0);
602 	dev->drain_notify = false;
603 }
604 
605 static inline struct storvsc_device *get_in_stor_device(
606 					struct hv_device *device)
607 {
608 	struct storvsc_device *stor_device;
609 
610 	stor_device = hv_get_drvdata(device);
611 
612 	if (!stor_device)
613 		goto get_in_err;
614 
615 	/*
616 	 * If the device is being destroyed; allow incoming
617 	 * traffic only to cleanup outstanding requests.
618 	 */
619 
620 	if (stor_device->destroy  &&
621 		(atomic_read(&stor_device->num_outstanding_req) == 0))
622 		stor_device = NULL;
623 
624 get_in_err:
625 	return stor_device;
626 
627 }
628 
629 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
630 				      u32 new)
631 {
632 	struct storvsc_device *stor_device;
633 	struct vmbus_channel *cur_chn;
634 	bool old_is_alloced = false;
635 	struct hv_device *device;
636 	unsigned long flags;
637 	int cpu;
638 
639 	device = channel->primary_channel ?
640 			channel->primary_channel->device_obj
641 				: channel->device_obj;
642 	stor_device = get_out_stor_device(device);
643 	if (!stor_device)
644 		return;
645 
646 	/* See storvsc_do_io() -> get_og_chn(). */
647 	spin_lock_irqsave(&stor_device->lock, flags);
648 
649 	/*
650 	 * Determines if the storvsc device has other channels assigned to
651 	 * the "old" CPU to update the alloced_cpus mask and the stor_chns
652 	 * array.
653 	 */
654 	if (device->channel != channel && device->channel->target_cpu == old) {
655 		cur_chn = device->channel;
656 		old_is_alloced = true;
657 		goto old_is_alloced;
658 	}
659 	list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
660 		if (cur_chn == channel)
661 			continue;
662 		if (cur_chn->target_cpu == old) {
663 			old_is_alloced = true;
664 			goto old_is_alloced;
665 		}
666 	}
667 
668 old_is_alloced:
669 	if (old_is_alloced)
670 		WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
671 	else
672 		cpumask_clear_cpu(old, &stor_device->alloced_cpus);
673 
674 	/* "Flush" the stor_chns array. */
675 	for_each_possible_cpu(cpu) {
676 		if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
677 					cpu, &stor_device->alloced_cpus))
678 			WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
679 	}
680 
681 	WRITE_ONCE(stor_device->stor_chns[new], channel);
682 	cpumask_set_cpu(new, &stor_device->alloced_cpus);
683 
684 	spin_unlock_irqrestore(&stor_device->lock, flags);
685 }
686 
687 static void handle_sc_creation(struct vmbus_channel *new_sc)
688 {
689 	struct hv_device *device = new_sc->primary_channel->device_obj;
690 	struct device *dev = &device->device;
691 	struct storvsc_device *stor_device;
692 	struct vmstorage_channel_properties props;
693 	int ret;
694 
695 	stor_device = get_out_stor_device(device);
696 	if (!stor_device)
697 		return;
698 
699 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
700 
701 	ret = vmbus_open(new_sc,
702 			 storvsc_ringbuffer_size,
703 			 storvsc_ringbuffer_size,
704 			 (void *)&props,
705 			 sizeof(struct vmstorage_channel_properties),
706 			 storvsc_on_channel_callback, new_sc);
707 
708 	/* In case vmbus_open() fails, we don't use the sub-channel. */
709 	if (ret != 0) {
710 		dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
711 		return;
712 	}
713 
714 	new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
715 
716 	/* Add the sub-channel to the array of available channels. */
717 	stor_device->stor_chns[new_sc->target_cpu] = new_sc;
718 	cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
719 }
720 
721 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
722 {
723 	struct device *dev = &device->device;
724 	struct storvsc_device *stor_device;
725 	int num_sc;
726 	struct storvsc_cmd_request *request;
727 	struct vstor_packet *vstor_packet;
728 	int ret, t;
729 
730 	/*
731 	 * If the number of CPUs is artificially restricted, such as
732 	 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
733 	 * sub-channels >= the number of CPUs. These sub-channels
734 	 * should not be created. The primary channel is already created
735 	 * and assigned to one CPU, so check against # CPUs - 1.
736 	 */
737 	num_sc = min((int)(num_online_cpus() - 1), max_chns);
738 	if (!num_sc)
739 		return;
740 
741 	stor_device = get_out_stor_device(device);
742 	if (!stor_device)
743 		return;
744 
745 	stor_device->num_sc = num_sc;
746 	request = &stor_device->init_request;
747 	vstor_packet = &request->vstor_packet;
748 
749 	/*
750 	 * Establish a handler for dealing with subchannels.
751 	 */
752 	vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
753 
754 	/*
755 	 * Request the host to create sub-channels.
756 	 */
757 	memset(request, 0, sizeof(struct storvsc_cmd_request));
758 	init_completion(&request->wait_event);
759 	vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
760 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
761 	vstor_packet->sub_channel_count = num_sc;
762 
763 	ret = vmbus_sendpacket(device->channel, vstor_packet,
764 			       (sizeof(struct vstor_packet) -
765 			       vmscsi_size_delta),
766 			       (unsigned long)request,
767 			       VM_PKT_DATA_INBAND,
768 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
769 
770 	if (ret != 0) {
771 		dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
772 		return;
773 	}
774 
775 	t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
776 	if (t == 0) {
777 		dev_err(dev, "Failed to create sub-channel: timed out\n");
778 		return;
779 	}
780 
781 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
782 	    vstor_packet->status != 0) {
783 		dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
784 			vstor_packet->operation, vstor_packet->status);
785 		return;
786 	}
787 
788 	/*
789 	 * We need to do nothing here, because vmbus_process_offer()
790 	 * invokes channel->sc_creation_callback, which will open and use
791 	 * the sub-channel(s).
792 	 */
793 }
794 
795 static void cache_wwn(struct storvsc_device *stor_device,
796 		      struct vstor_packet *vstor_packet)
797 {
798 	/*
799 	 * Cache the currently active port and node ww names.
800 	 */
801 	if (vstor_packet->wwn_packet.primary_active) {
802 		stor_device->node_name =
803 			wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
804 		stor_device->port_name =
805 			wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
806 	} else {
807 		stor_device->node_name =
808 			wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
809 		stor_device->port_name =
810 			wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
811 	}
812 }
813 
814 
815 static int storvsc_execute_vstor_op(struct hv_device *device,
816 				    struct storvsc_cmd_request *request,
817 				    bool status_check)
818 {
819 	struct vstor_packet *vstor_packet;
820 	int ret, t;
821 
822 	vstor_packet = &request->vstor_packet;
823 
824 	init_completion(&request->wait_event);
825 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
826 
827 	ret = vmbus_sendpacket(device->channel, vstor_packet,
828 			       (sizeof(struct vstor_packet) -
829 			       vmscsi_size_delta),
830 			       (unsigned long)request,
831 			       VM_PKT_DATA_INBAND,
832 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
833 	if (ret != 0)
834 		return ret;
835 
836 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
837 	if (t == 0)
838 		return -ETIMEDOUT;
839 
840 	if (!status_check)
841 		return ret;
842 
843 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
844 	    vstor_packet->status != 0)
845 		return -EINVAL;
846 
847 	return ret;
848 }
849 
850 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
851 {
852 	struct storvsc_device *stor_device;
853 	struct storvsc_cmd_request *request;
854 	struct vstor_packet *vstor_packet;
855 	int ret, i;
856 	int max_chns;
857 	bool process_sub_channels = false;
858 
859 	stor_device = get_out_stor_device(device);
860 	if (!stor_device)
861 		return -ENODEV;
862 
863 	request = &stor_device->init_request;
864 	vstor_packet = &request->vstor_packet;
865 
866 	/*
867 	 * Now, initiate the vsc/vsp initialization protocol on the open
868 	 * channel
869 	 */
870 	memset(request, 0, sizeof(struct storvsc_cmd_request));
871 	vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
872 	ret = storvsc_execute_vstor_op(device, request, true);
873 	if (ret)
874 		return ret;
875 	/*
876 	 * Query host supported protocol version.
877 	 */
878 
879 	for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) {
880 		/* reuse the packet for version range supported */
881 		memset(vstor_packet, 0, sizeof(struct vstor_packet));
882 		vstor_packet->operation =
883 			VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
884 
885 		vstor_packet->version.major_minor =
886 			vmstor_protocols[i].protocol_version;
887 
888 		/*
889 		 * The revision number is only used in Windows; set it to 0.
890 		 */
891 		vstor_packet->version.revision = 0;
892 		ret = storvsc_execute_vstor_op(device, request, false);
893 		if (ret != 0)
894 			return ret;
895 
896 		if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
897 			return -EINVAL;
898 
899 		if (vstor_packet->status == 0) {
900 			vmstor_proto_version =
901 				vmstor_protocols[i].protocol_version;
902 
903 			sense_buffer_size =
904 				vmstor_protocols[i].sense_buffer_size;
905 
906 			vmscsi_size_delta =
907 				vmstor_protocols[i].vmscsi_size_delta;
908 
909 			break;
910 		}
911 	}
912 
913 	if (vstor_packet->status != 0)
914 		return -EINVAL;
915 
916 
917 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
918 	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
919 	ret = storvsc_execute_vstor_op(device, request, true);
920 	if (ret != 0)
921 		return ret;
922 
923 	/*
924 	 * Check to see if multi-channel support is there.
925 	 * Hosts that implement protocol version of 5.1 and above
926 	 * support multi-channel.
927 	 */
928 	max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
929 
930 	/*
931 	 * Allocate state to manage the sub-channels.
932 	 * We allocate an array based on the numbers of possible CPUs
933 	 * (Hyper-V does not support cpu online/offline).
934 	 * This Array will be sparseley populated with unique
935 	 * channels - primary + sub-channels.
936 	 * We will however populate all the slots to evenly distribute
937 	 * the load.
938 	 */
939 	stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
940 					 GFP_KERNEL);
941 	if (stor_device->stor_chns == NULL)
942 		return -ENOMEM;
943 
944 	device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
945 
946 	stor_device->stor_chns[device->channel->target_cpu] = device->channel;
947 	cpumask_set_cpu(device->channel->target_cpu,
948 			&stor_device->alloced_cpus);
949 
950 	if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) {
951 		if (vstor_packet->storage_channel_properties.flags &
952 		    STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
953 			process_sub_channels = true;
954 	}
955 	stor_device->max_transfer_bytes =
956 		vstor_packet->storage_channel_properties.max_transfer_bytes;
957 
958 	if (!is_fc)
959 		goto done;
960 
961 	/*
962 	 * For FC devices retrieve FC HBA data.
963 	 */
964 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
965 	vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
966 	ret = storvsc_execute_vstor_op(device, request, true);
967 	if (ret != 0)
968 		return ret;
969 
970 	/*
971 	 * Cache the currently active port and node ww names.
972 	 */
973 	cache_wwn(stor_device, vstor_packet);
974 
975 done:
976 
977 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
978 	vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
979 	ret = storvsc_execute_vstor_op(device, request, true);
980 	if (ret != 0)
981 		return ret;
982 
983 	if (process_sub_channels)
984 		handle_multichannel_storage(device, max_chns);
985 
986 	return ret;
987 }
988 
989 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
990 				struct scsi_cmnd *scmnd,
991 				struct Scsi_Host *host,
992 				u8 asc, u8 ascq)
993 {
994 	struct storvsc_scan_work *wrk;
995 	void (*process_err_fn)(struct work_struct *work);
996 	struct hv_host_device *host_dev = shost_priv(host);
997 	bool do_work = false;
998 
999 	switch (SRB_STATUS(vm_srb->srb_status)) {
1000 	case SRB_STATUS_ERROR:
1001 		/*
1002 		 * Let upper layer deal with error when
1003 		 * sense message is present.
1004 		 */
1005 
1006 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
1007 			break;
1008 		/*
1009 		 * If there is an error; offline the device since all
1010 		 * error recovery strategies would have already been
1011 		 * deployed on the host side. However, if the command
1012 		 * were a pass-through command deal with it appropriately.
1013 		 */
1014 		switch (scmnd->cmnd[0]) {
1015 		case ATA_16:
1016 		case ATA_12:
1017 			set_host_byte(scmnd, DID_PASSTHROUGH);
1018 			break;
1019 		/*
1020 		 * On Some Windows hosts TEST_UNIT_READY command can return
1021 		 * SRB_STATUS_ERROR, let the upper level code deal with it
1022 		 * based on the sense information.
1023 		 */
1024 		case TEST_UNIT_READY:
1025 			break;
1026 		default:
1027 			set_host_byte(scmnd, DID_ERROR);
1028 		}
1029 		break;
1030 	case SRB_STATUS_INVALID_LUN:
1031 		set_host_byte(scmnd, DID_NO_CONNECT);
1032 		do_work = true;
1033 		process_err_fn = storvsc_remove_lun;
1034 		break;
1035 	case SRB_STATUS_ABORTED:
1036 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
1037 		    (asc == 0x2a) && (ascq == 0x9)) {
1038 			do_work = true;
1039 			process_err_fn = storvsc_device_scan;
1040 			/*
1041 			 * Retry the I/O that triggered this.
1042 			 */
1043 			set_host_byte(scmnd, DID_REQUEUE);
1044 		}
1045 		break;
1046 	}
1047 
1048 	if (!do_work)
1049 		return;
1050 
1051 	/*
1052 	 * We need to schedule work to process this error; schedule it.
1053 	 */
1054 	wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1055 	if (!wrk) {
1056 		set_host_byte(scmnd, DID_TARGET_FAILURE);
1057 		return;
1058 	}
1059 
1060 	wrk->host = host;
1061 	wrk->lun = vm_srb->lun;
1062 	wrk->tgt_id = vm_srb->target_id;
1063 	INIT_WORK(&wrk->work, process_err_fn);
1064 	queue_work(host_dev->handle_error_wq, &wrk->work);
1065 }
1066 
1067 
1068 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1069 				       struct storvsc_device *stor_dev)
1070 {
1071 	struct scsi_cmnd *scmnd = cmd_request->cmd;
1072 	struct scsi_sense_hdr sense_hdr;
1073 	struct vmscsi_request *vm_srb;
1074 	u32 data_transfer_length;
1075 	struct Scsi_Host *host;
1076 	u32 payload_sz = cmd_request->payload_sz;
1077 	void *payload = cmd_request->payload;
1078 
1079 	host = stor_dev->host;
1080 
1081 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1082 	data_transfer_length = vm_srb->data_transfer_length;
1083 
1084 	scmnd->result = vm_srb->scsi_status;
1085 
1086 	if (scmnd->result) {
1087 		if (scsi_normalize_sense(scmnd->sense_buffer,
1088 				SCSI_SENSE_BUFFERSIZE, &sense_hdr) &&
1089 		    !(sense_hdr.sense_key == NOT_READY &&
1090 				 sense_hdr.asc == 0x03A) &&
1091 		    do_logging(STORVSC_LOGGING_ERROR))
1092 			scsi_print_sense_hdr(scmnd->device, "storvsc",
1093 					     &sense_hdr);
1094 	}
1095 
1096 	if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1097 		storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1098 					 sense_hdr.ascq);
1099 		/*
1100 		 * The Windows driver set data_transfer_length on
1101 		 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1102 		 * is untouched.  In these cases we set it to 0.
1103 		 */
1104 		if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1105 			data_transfer_length = 0;
1106 	}
1107 
1108 	/* Validate data_transfer_length (from Hyper-V) */
1109 	if (data_transfer_length > cmd_request->payload->range.len)
1110 		data_transfer_length = cmd_request->payload->range.len;
1111 
1112 	scsi_set_resid(scmnd,
1113 		cmd_request->payload->range.len - data_transfer_length);
1114 
1115 	scmnd->scsi_done(scmnd);
1116 
1117 	if (payload_sz >
1118 		sizeof(struct vmbus_channel_packet_multipage_buffer))
1119 		kfree(payload);
1120 }
1121 
1122 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1123 				  struct vstor_packet *vstor_packet,
1124 				  struct storvsc_cmd_request *request)
1125 {
1126 	struct vstor_packet *stor_pkt;
1127 	struct hv_device *device = stor_device->device;
1128 
1129 	stor_pkt = &request->vstor_packet;
1130 
1131 	/*
1132 	 * The current SCSI handling on the host side does
1133 	 * not correctly handle:
1134 	 * INQUIRY command with page code parameter set to 0x80
1135 	 * MODE_SENSE command with cmd[2] == 0x1c
1136 	 *
1137 	 * Setup srb and scsi status so this won't be fatal.
1138 	 * We do this so we can distinguish truly fatal failues
1139 	 * (srb status == 0x4) and off-line the device in that case.
1140 	 */
1141 
1142 	if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1143 	   (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1144 		vstor_packet->vm_srb.scsi_status = 0;
1145 		vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1146 	}
1147 
1148 
1149 	/* Copy over the status...etc */
1150 	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1151 	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1152 
1153 	/* Validate sense_info_length (from Hyper-V) */
1154 	if (vstor_packet->vm_srb.sense_info_length > sense_buffer_size)
1155 		vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1156 
1157 	stor_pkt->vm_srb.sense_info_length =
1158 	vstor_packet->vm_srb.sense_info_length;
1159 
1160 	if (vstor_packet->vm_srb.scsi_status != 0 ||
1161 	    vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS)
1162 		storvsc_log(device, STORVSC_LOGGING_WARN,
1163 			"cmd 0x%x scsi status 0x%x srb status 0x%x\n",
1164 			stor_pkt->vm_srb.cdb[0],
1165 			vstor_packet->vm_srb.scsi_status,
1166 			vstor_packet->vm_srb.srb_status);
1167 
1168 	if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
1169 		/* CHECK_CONDITION */
1170 		if (vstor_packet->vm_srb.srb_status &
1171 			SRB_STATUS_AUTOSENSE_VALID) {
1172 			/* autosense data available */
1173 
1174 			storvsc_log(device, STORVSC_LOGGING_WARN,
1175 				"stor pkt %p autosense data valid - len %d\n",
1176 				request, vstor_packet->vm_srb.sense_info_length);
1177 
1178 			memcpy(request->cmd->sense_buffer,
1179 			       vstor_packet->vm_srb.sense_data,
1180 			       vstor_packet->vm_srb.sense_info_length);
1181 
1182 		}
1183 	}
1184 
1185 	stor_pkt->vm_srb.data_transfer_length =
1186 	vstor_packet->vm_srb.data_transfer_length;
1187 
1188 	storvsc_command_completion(request, stor_device);
1189 
1190 	if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1191 		stor_device->drain_notify)
1192 		wake_up(&stor_device->waiting_to_drain);
1193 
1194 
1195 }
1196 
1197 static void storvsc_on_receive(struct storvsc_device *stor_device,
1198 			     struct vstor_packet *vstor_packet,
1199 			     struct storvsc_cmd_request *request)
1200 {
1201 	struct hv_host_device *host_dev;
1202 	switch (vstor_packet->operation) {
1203 	case VSTOR_OPERATION_COMPLETE_IO:
1204 		storvsc_on_io_completion(stor_device, vstor_packet, request);
1205 		break;
1206 
1207 	case VSTOR_OPERATION_REMOVE_DEVICE:
1208 	case VSTOR_OPERATION_ENUMERATE_BUS:
1209 		host_dev = shost_priv(stor_device->host);
1210 		queue_work(
1211 			host_dev->handle_error_wq, &host_dev->host_scan_work);
1212 		break;
1213 
1214 	case VSTOR_OPERATION_FCHBA_DATA:
1215 		cache_wwn(stor_device, vstor_packet);
1216 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1217 		fc_host_node_name(stor_device->host) = stor_device->node_name;
1218 		fc_host_port_name(stor_device->host) = stor_device->port_name;
1219 #endif
1220 		break;
1221 	default:
1222 		break;
1223 	}
1224 }
1225 
1226 static void storvsc_on_channel_callback(void *context)
1227 {
1228 	struct vmbus_channel *channel = (struct vmbus_channel *)context;
1229 	const struct vmpacket_descriptor *desc;
1230 	struct hv_device *device;
1231 	struct storvsc_device *stor_device;
1232 
1233 	if (channel->primary_channel != NULL)
1234 		device = channel->primary_channel->device_obj;
1235 	else
1236 		device = channel->device_obj;
1237 
1238 	stor_device = get_in_stor_device(device);
1239 	if (!stor_device)
1240 		return;
1241 
1242 	foreach_vmbus_pkt(desc, channel) {
1243 		void *packet = hv_pkt_data(desc);
1244 		struct storvsc_cmd_request *request;
1245 
1246 		request = (struct storvsc_cmd_request *)
1247 			((unsigned long)desc->trans_id);
1248 
1249 		if (request == &stor_device->init_request ||
1250 		    request == &stor_device->reset_request) {
1251 			memcpy(&request->vstor_packet, packet,
1252 			       (sizeof(struct vstor_packet) - vmscsi_size_delta));
1253 			complete(&request->wait_event);
1254 		} else {
1255 			storvsc_on_receive(stor_device, packet, request);
1256 		}
1257 	}
1258 }
1259 
1260 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1261 				  bool is_fc)
1262 {
1263 	struct vmstorage_channel_properties props;
1264 	int ret;
1265 
1266 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1267 
1268 	ret = vmbus_open(device->channel,
1269 			 ring_size,
1270 			 ring_size,
1271 			 (void *)&props,
1272 			 sizeof(struct vmstorage_channel_properties),
1273 			 storvsc_on_channel_callback, device->channel);
1274 
1275 	if (ret != 0)
1276 		return ret;
1277 
1278 	ret = storvsc_channel_init(device, is_fc);
1279 
1280 	return ret;
1281 }
1282 
1283 static int storvsc_dev_remove(struct hv_device *device)
1284 {
1285 	struct storvsc_device *stor_device;
1286 
1287 	stor_device = hv_get_drvdata(device);
1288 
1289 	stor_device->destroy = true;
1290 
1291 	/* Make sure flag is set before waiting */
1292 	wmb();
1293 
1294 	/*
1295 	 * At this point, all outbound traffic should be disable. We
1296 	 * only allow inbound traffic (responses) to proceed so that
1297 	 * outstanding requests can be completed.
1298 	 */
1299 
1300 	storvsc_wait_to_drain(stor_device);
1301 
1302 	/*
1303 	 * Since we have already drained, we don't need to busy wait
1304 	 * as was done in final_release_stor_device()
1305 	 * Note that we cannot set the ext pointer to NULL until
1306 	 * we have drained - to drain the outgoing packets, we need to
1307 	 * allow incoming packets.
1308 	 */
1309 	hv_set_drvdata(device, NULL);
1310 
1311 	/* Close the channel */
1312 	vmbus_close(device->channel);
1313 
1314 	kfree(stor_device->stor_chns);
1315 	kfree(stor_device);
1316 	return 0;
1317 }
1318 
1319 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1320 					u16 q_num)
1321 {
1322 	u16 slot = 0;
1323 	u16 hash_qnum;
1324 	const struct cpumask *node_mask;
1325 	int num_channels, tgt_cpu;
1326 
1327 	if (stor_device->num_sc == 0) {
1328 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1329 		return stor_device->device->channel;
1330 	}
1331 
1332 	/*
1333 	 * Our channel array is sparsley populated and we
1334 	 * initiated I/O on a processor/hw-q that does not
1335 	 * currently have a designated channel. Fix this.
1336 	 * The strategy is simple:
1337 	 * I. Ensure NUMA locality
1338 	 * II. Distribute evenly (best effort)
1339 	 */
1340 
1341 	node_mask = cpumask_of_node(cpu_to_node(q_num));
1342 
1343 	num_channels = 0;
1344 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1345 		if (cpumask_test_cpu(tgt_cpu, node_mask))
1346 			num_channels++;
1347 	}
1348 	if (num_channels == 0) {
1349 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1350 		return stor_device->device->channel;
1351 	}
1352 
1353 	hash_qnum = q_num;
1354 	while (hash_qnum >= num_channels)
1355 		hash_qnum -= num_channels;
1356 
1357 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1358 		if (!cpumask_test_cpu(tgt_cpu, node_mask))
1359 			continue;
1360 		if (slot == hash_qnum)
1361 			break;
1362 		slot++;
1363 	}
1364 
1365 	stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1366 
1367 	return stor_device->stor_chns[q_num];
1368 }
1369 
1370 
1371 static int storvsc_do_io(struct hv_device *device,
1372 			 struct storvsc_cmd_request *request, u16 q_num)
1373 {
1374 	struct storvsc_device *stor_device;
1375 	struct vstor_packet *vstor_packet;
1376 	struct vmbus_channel *outgoing_channel, *channel;
1377 	unsigned long flags;
1378 	int ret = 0;
1379 	const struct cpumask *node_mask;
1380 	int tgt_cpu;
1381 
1382 	vstor_packet = &request->vstor_packet;
1383 	stor_device = get_out_stor_device(device);
1384 
1385 	if (!stor_device)
1386 		return -ENODEV;
1387 
1388 
1389 	request->device  = device;
1390 	/*
1391 	 * Select an appropriate channel to send the request out.
1392 	 */
1393 	/* See storvsc_change_target_cpu(). */
1394 	outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1395 	if (outgoing_channel != NULL) {
1396 		if (outgoing_channel->target_cpu == q_num) {
1397 			/*
1398 			 * Ideally, we want to pick a different channel if
1399 			 * available on the same NUMA node.
1400 			 */
1401 			node_mask = cpumask_of_node(cpu_to_node(q_num));
1402 			for_each_cpu_wrap(tgt_cpu,
1403 				 &stor_device->alloced_cpus, q_num + 1) {
1404 				if (!cpumask_test_cpu(tgt_cpu, node_mask))
1405 					continue;
1406 				if (tgt_cpu == q_num)
1407 					continue;
1408 				channel = READ_ONCE(
1409 					stor_device->stor_chns[tgt_cpu]);
1410 				if (channel == NULL)
1411 					continue;
1412 				if (hv_get_avail_to_write_percent(
1413 							&channel->outbound)
1414 						> ring_avail_percent_lowater) {
1415 					outgoing_channel = channel;
1416 					goto found_channel;
1417 				}
1418 			}
1419 
1420 			/*
1421 			 * All the other channels on the same NUMA node are
1422 			 * busy. Try to use the channel on the current CPU
1423 			 */
1424 			if (hv_get_avail_to_write_percent(
1425 						&outgoing_channel->outbound)
1426 					> ring_avail_percent_lowater)
1427 				goto found_channel;
1428 
1429 			/*
1430 			 * If we reach here, all the channels on the current
1431 			 * NUMA node are busy. Try to find a channel in
1432 			 * other NUMA nodes
1433 			 */
1434 			for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1435 				if (cpumask_test_cpu(tgt_cpu, node_mask))
1436 					continue;
1437 				channel = READ_ONCE(
1438 					stor_device->stor_chns[tgt_cpu]);
1439 				if (channel == NULL)
1440 					continue;
1441 				if (hv_get_avail_to_write_percent(
1442 							&channel->outbound)
1443 						> ring_avail_percent_lowater) {
1444 					outgoing_channel = channel;
1445 					goto found_channel;
1446 				}
1447 			}
1448 		}
1449 	} else {
1450 		spin_lock_irqsave(&stor_device->lock, flags);
1451 		outgoing_channel = stor_device->stor_chns[q_num];
1452 		if (outgoing_channel != NULL) {
1453 			spin_unlock_irqrestore(&stor_device->lock, flags);
1454 			goto found_channel;
1455 		}
1456 		outgoing_channel = get_og_chn(stor_device, q_num);
1457 		spin_unlock_irqrestore(&stor_device->lock, flags);
1458 	}
1459 
1460 found_channel:
1461 	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1462 
1463 	vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1464 					vmscsi_size_delta);
1465 
1466 
1467 	vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1468 
1469 
1470 	vstor_packet->vm_srb.data_transfer_length =
1471 	request->payload->range.len;
1472 
1473 	vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1474 
1475 	if (request->payload->range.len) {
1476 
1477 		ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1478 				request->payload, request->payload_sz,
1479 				vstor_packet,
1480 				(sizeof(struct vstor_packet) -
1481 				vmscsi_size_delta),
1482 				(unsigned long)request);
1483 	} else {
1484 		ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1485 			       (sizeof(struct vstor_packet) -
1486 				vmscsi_size_delta),
1487 			       (unsigned long)request,
1488 			       VM_PKT_DATA_INBAND,
1489 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1490 	}
1491 
1492 	if (ret != 0)
1493 		return ret;
1494 
1495 	atomic_inc(&stor_device->num_outstanding_req);
1496 
1497 	return ret;
1498 }
1499 
1500 static int storvsc_device_alloc(struct scsi_device *sdevice)
1501 {
1502 	/*
1503 	 * Set blist flag to permit the reading of the VPD pages even when
1504 	 * the target may claim SPC-2 compliance. MSFT targets currently
1505 	 * claim SPC-2 compliance while they implement post SPC-2 features.
1506 	 * With this flag we can correctly handle WRITE_SAME_16 issues.
1507 	 *
1508 	 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1509 	 * still supports REPORT LUN.
1510 	 */
1511 	sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1512 
1513 	return 0;
1514 }
1515 
1516 static int storvsc_device_configure(struct scsi_device *sdevice)
1517 {
1518 	blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1519 
1520 	sdevice->no_write_same = 1;
1521 
1522 	/*
1523 	 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1524 	 * if the device is a MSFT virtual device.  If the host is
1525 	 * WIN10 or newer, allow write_same.
1526 	 */
1527 	if (!strncmp(sdevice->vendor, "Msft", 4)) {
1528 		switch (vmstor_proto_version) {
1529 		case VMSTOR_PROTO_VERSION_WIN8:
1530 		case VMSTOR_PROTO_VERSION_WIN8_1:
1531 			sdevice->scsi_level = SCSI_SPC_3;
1532 			break;
1533 		}
1534 
1535 		if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1536 			sdevice->no_write_same = 0;
1537 	}
1538 
1539 	return 0;
1540 }
1541 
1542 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1543 			   sector_t capacity, int *info)
1544 {
1545 	sector_t nsect = capacity;
1546 	sector_t cylinders = nsect;
1547 	int heads, sectors_pt;
1548 
1549 	/*
1550 	 * We are making up these values; let us keep it simple.
1551 	 */
1552 	heads = 0xff;
1553 	sectors_pt = 0x3f;      /* Sectors per track */
1554 	sector_div(cylinders, heads * sectors_pt);
1555 	if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1556 		cylinders = 0xffff;
1557 
1558 	info[0] = heads;
1559 	info[1] = sectors_pt;
1560 	info[2] = (int)cylinders;
1561 
1562 	return 0;
1563 }
1564 
1565 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1566 {
1567 	struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1568 	struct hv_device *device = host_dev->dev;
1569 
1570 	struct storvsc_device *stor_device;
1571 	struct storvsc_cmd_request *request;
1572 	struct vstor_packet *vstor_packet;
1573 	int ret, t;
1574 
1575 
1576 	stor_device = get_out_stor_device(device);
1577 	if (!stor_device)
1578 		return FAILED;
1579 
1580 	request = &stor_device->reset_request;
1581 	vstor_packet = &request->vstor_packet;
1582 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1583 
1584 	init_completion(&request->wait_event);
1585 
1586 	vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1587 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1588 	vstor_packet->vm_srb.path_id = stor_device->path_id;
1589 
1590 	ret = vmbus_sendpacket(device->channel, vstor_packet,
1591 			       (sizeof(struct vstor_packet) -
1592 				vmscsi_size_delta),
1593 			       (unsigned long)&stor_device->reset_request,
1594 			       VM_PKT_DATA_INBAND,
1595 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1596 	if (ret != 0)
1597 		return FAILED;
1598 
1599 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1600 	if (t == 0)
1601 		return TIMEOUT_ERROR;
1602 
1603 
1604 	/*
1605 	 * At this point, all outstanding requests in the adapter
1606 	 * should have been flushed out and return to us
1607 	 * There is a potential race here where the host may be in
1608 	 * the process of responding when we return from here.
1609 	 * Just wait for all in-transit packets to be accounted for
1610 	 * before we return from here.
1611 	 */
1612 	storvsc_wait_to_drain(stor_device);
1613 
1614 	return SUCCESS;
1615 }
1616 
1617 /*
1618  * The host guarantees to respond to each command, although I/O latencies might
1619  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1620  * chance to perform EH.
1621  */
1622 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1623 {
1624 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1625 	if (scmnd->device->host->transportt == fc_transport_template)
1626 		return fc_eh_timed_out(scmnd);
1627 #endif
1628 	return BLK_EH_RESET_TIMER;
1629 }
1630 
1631 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1632 {
1633 	bool allowed = true;
1634 	u8 scsi_op = scmnd->cmnd[0];
1635 
1636 	switch (scsi_op) {
1637 	/* the host does not handle WRITE_SAME, log accident usage */
1638 	case WRITE_SAME:
1639 	/*
1640 	 * smartd sends this command and the host does not handle
1641 	 * this. So, don't send it.
1642 	 */
1643 	case SET_WINDOW:
1644 		scmnd->result = ILLEGAL_REQUEST << 16;
1645 		allowed = false;
1646 		break;
1647 	default:
1648 		break;
1649 	}
1650 	return allowed;
1651 }
1652 
1653 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1654 {
1655 	int ret;
1656 	struct hv_host_device *host_dev = shost_priv(host);
1657 	struct hv_device *dev = host_dev->dev;
1658 	struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1659 	int i;
1660 	struct scatterlist *sgl;
1661 	unsigned int sg_count = 0;
1662 	struct vmscsi_request *vm_srb;
1663 	struct scatterlist *cur_sgl;
1664 	struct vmbus_packet_mpb_array  *payload;
1665 	u32 payload_sz;
1666 	u32 length;
1667 
1668 	if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1669 		/*
1670 		 * On legacy hosts filter unimplemented commands.
1671 		 * Future hosts are expected to correctly handle
1672 		 * unsupported commands. Furthermore, it is
1673 		 * possible that some of the currently
1674 		 * unsupported commands maybe supported in
1675 		 * future versions of the host.
1676 		 */
1677 		if (!storvsc_scsi_cmd_ok(scmnd)) {
1678 			scmnd->scsi_done(scmnd);
1679 			return 0;
1680 		}
1681 	}
1682 
1683 	/* Setup the cmd request */
1684 	cmd_request->cmd = scmnd;
1685 
1686 	memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1687 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1688 	vm_srb->win8_extension.time_out_value = 60;
1689 
1690 	vm_srb->win8_extension.srb_flags |=
1691 		SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1692 
1693 	if (scmnd->device->tagged_supported) {
1694 		vm_srb->win8_extension.srb_flags |=
1695 		(SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1696 		vm_srb->win8_extension.queue_tag = SP_UNTAGGED;
1697 		vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST;
1698 	}
1699 
1700 	/* Build the SRB */
1701 	switch (scmnd->sc_data_direction) {
1702 	case DMA_TO_DEVICE:
1703 		vm_srb->data_in = WRITE_TYPE;
1704 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1705 		break;
1706 	case DMA_FROM_DEVICE:
1707 		vm_srb->data_in = READ_TYPE;
1708 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1709 		break;
1710 	case DMA_NONE:
1711 		vm_srb->data_in = UNKNOWN_TYPE;
1712 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1713 		break;
1714 	default:
1715 		/*
1716 		 * This is DMA_BIDIRECTIONAL or something else we are never
1717 		 * supposed to see here.
1718 		 */
1719 		WARN(1, "Unexpected data direction: %d\n",
1720 		     scmnd->sc_data_direction);
1721 		return -EINVAL;
1722 	}
1723 
1724 
1725 	vm_srb->port_number = host_dev->port;
1726 	vm_srb->path_id = scmnd->device->channel;
1727 	vm_srb->target_id = scmnd->device->id;
1728 	vm_srb->lun = scmnd->device->lun;
1729 
1730 	vm_srb->cdb_length = scmnd->cmd_len;
1731 
1732 	memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1733 
1734 	sgl = (struct scatterlist *)scsi_sglist(scmnd);
1735 	sg_count = scsi_sg_count(scmnd);
1736 
1737 	length = scsi_bufflen(scmnd);
1738 	payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1739 	payload_sz = sizeof(cmd_request->mpb);
1740 
1741 	if (sg_count) {
1742 		if (sg_count > MAX_PAGE_BUFFER_COUNT) {
1743 
1744 			payload_sz = (sg_count * sizeof(u64) +
1745 				      sizeof(struct vmbus_packet_mpb_array));
1746 			payload = kzalloc(payload_sz, GFP_ATOMIC);
1747 			if (!payload)
1748 				return SCSI_MLQUEUE_DEVICE_BUSY;
1749 		}
1750 
1751 		payload->range.len = length;
1752 		payload->range.offset = sgl[0].offset;
1753 
1754 		cur_sgl = sgl;
1755 		for (i = 0; i < sg_count; i++) {
1756 			payload->range.pfn_array[i] =
1757 				page_to_pfn(sg_page((cur_sgl)));
1758 			cur_sgl = sg_next(cur_sgl);
1759 		}
1760 	}
1761 
1762 	cmd_request->payload = payload;
1763 	cmd_request->payload_sz = payload_sz;
1764 
1765 	/* Invokes the vsc to start an IO */
1766 	ret = storvsc_do_io(dev, cmd_request, get_cpu());
1767 	put_cpu();
1768 
1769 	if (ret == -EAGAIN) {
1770 		if (payload_sz > sizeof(cmd_request->mpb))
1771 			kfree(payload);
1772 		/* no more space */
1773 		return SCSI_MLQUEUE_DEVICE_BUSY;
1774 	}
1775 
1776 	return 0;
1777 }
1778 
1779 static struct scsi_host_template scsi_driver = {
1780 	.module	=		THIS_MODULE,
1781 	.name =			"storvsc_host_t",
1782 	.cmd_size =             sizeof(struct storvsc_cmd_request),
1783 	.bios_param =		storvsc_get_chs,
1784 	.queuecommand =		storvsc_queuecommand,
1785 	.eh_host_reset_handler =	storvsc_host_reset_handler,
1786 	.proc_name =		"storvsc_host",
1787 	.eh_timed_out =		storvsc_eh_timed_out,
1788 	.slave_alloc =		storvsc_device_alloc,
1789 	.slave_configure =	storvsc_device_configure,
1790 	.cmd_per_lun =		2048,
1791 	.this_id =		-1,
1792 	/* Make sure we dont get a sg segment crosses a page boundary */
1793 	.dma_boundary =		PAGE_SIZE-1,
1794 	/* Ensure there are no gaps in presented sgls */
1795 	.virt_boundary_mask =	PAGE_SIZE-1,
1796 	.no_write_same =	1,
1797 	.track_queue_depth =	1,
1798 	.change_queue_depth =	storvsc_change_queue_depth,
1799 };
1800 
1801 enum {
1802 	SCSI_GUID,
1803 	IDE_GUID,
1804 	SFC_GUID,
1805 };
1806 
1807 static const struct hv_vmbus_device_id id_table[] = {
1808 	/* SCSI guid */
1809 	{ HV_SCSI_GUID,
1810 	  .driver_data = SCSI_GUID
1811 	},
1812 	/* IDE guid */
1813 	{ HV_IDE_GUID,
1814 	  .driver_data = IDE_GUID
1815 	},
1816 	/* Fibre Channel GUID */
1817 	{
1818 	  HV_SYNTHFC_GUID,
1819 	  .driver_data = SFC_GUID
1820 	},
1821 	{ },
1822 };
1823 
1824 MODULE_DEVICE_TABLE(vmbus, id_table);
1825 
1826 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1827 
1828 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1829 {
1830 	return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1831 }
1832 
1833 static int storvsc_probe(struct hv_device *device,
1834 			const struct hv_vmbus_device_id *dev_id)
1835 {
1836 	int ret;
1837 	int num_cpus = num_online_cpus();
1838 	struct Scsi_Host *host;
1839 	struct hv_host_device *host_dev;
1840 	bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1841 	bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1842 	int target = 0;
1843 	struct storvsc_device *stor_device;
1844 	int max_luns_per_target;
1845 	int max_targets;
1846 	int max_channels;
1847 	int max_sub_channels = 0;
1848 
1849 	/*
1850 	 * Based on the windows host we are running on,
1851 	 * set state to properly communicate with the host.
1852 	 */
1853 
1854 	if (vmbus_proto_version < VERSION_WIN8) {
1855 		max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1856 		max_targets = STORVSC_IDE_MAX_TARGETS;
1857 		max_channels = STORVSC_IDE_MAX_CHANNELS;
1858 	} else {
1859 		max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
1860 		max_targets = STORVSC_MAX_TARGETS;
1861 		max_channels = STORVSC_MAX_CHANNELS;
1862 		/*
1863 		 * On Windows8 and above, we support sub-channels for storage
1864 		 * on SCSI and FC controllers.
1865 		 * The number of sub-channels offerred is based on the number of
1866 		 * VCPUs in the guest.
1867 		 */
1868 		if (!dev_is_ide)
1869 			max_sub_channels =
1870 				(num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1871 	}
1872 
1873 	scsi_driver.can_queue = max_outstanding_req_per_channel *
1874 				(max_sub_channels + 1) *
1875 				(100 - ring_avail_percent_lowater) / 100;
1876 
1877 	host = scsi_host_alloc(&scsi_driver,
1878 			       sizeof(struct hv_host_device));
1879 	if (!host)
1880 		return -ENOMEM;
1881 
1882 	host_dev = shost_priv(host);
1883 	memset(host_dev, 0, sizeof(struct hv_host_device));
1884 
1885 	host_dev->port = host->host_no;
1886 	host_dev->dev = device;
1887 	host_dev->host = host;
1888 
1889 
1890 	stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1891 	if (!stor_device) {
1892 		ret = -ENOMEM;
1893 		goto err_out0;
1894 	}
1895 
1896 	stor_device->destroy = false;
1897 	init_waitqueue_head(&stor_device->waiting_to_drain);
1898 	stor_device->device = device;
1899 	stor_device->host = host;
1900 	spin_lock_init(&stor_device->lock);
1901 	hv_set_drvdata(device, stor_device);
1902 
1903 	stor_device->port_number = host->host_no;
1904 	ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1905 	if (ret)
1906 		goto err_out1;
1907 
1908 	host_dev->path = stor_device->path_id;
1909 	host_dev->target = stor_device->target_id;
1910 
1911 	switch (dev_id->driver_data) {
1912 	case SFC_GUID:
1913 		host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1914 		host->max_id = STORVSC_FC_MAX_TARGETS;
1915 		host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1916 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1917 		host->transportt = fc_transport_template;
1918 #endif
1919 		break;
1920 
1921 	case SCSI_GUID:
1922 		host->max_lun = max_luns_per_target;
1923 		host->max_id = max_targets;
1924 		host->max_channel = max_channels - 1;
1925 		break;
1926 
1927 	default:
1928 		host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1929 		host->max_id = STORVSC_IDE_MAX_TARGETS;
1930 		host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
1931 		break;
1932 	}
1933 	/* max cmd length */
1934 	host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1935 
1936 	/*
1937 	 * set the table size based on the info we got
1938 	 * from the host.
1939 	 */
1940 	host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
1941 	/*
1942 	 * For non-IDE disks, the host supports multiple channels.
1943 	 * Set the number of HW queues we are supporting.
1944 	 */
1945 	if (!dev_is_ide)
1946 		host->nr_hw_queues = num_present_cpus();
1947 
1948 	/*
1949 	 * Set the error handler work queue.
1950 	 */
1951 	host_dev->handle_error_wq =
1952 			alloc_ordered_workqueue("storvsc_error_wq_%d",
1953 						WQ_MEM_RECLAIM,
1954 						host->host_no);
1955 	if (!host_dev->handle_error_wq)
1956 		goto err_out2;
1957 	INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
1958 	/* Register the HBA and start the scsi bus scan */
1959 	ret = scsi_add_host(host, &device->device);
1960 	if (ret != 0)
1961 		goto err_out3;
1962 
1963 	if (!dev_is_ide) {
1964 		scsi_scan_host(host);
1965 	} else {
1966 		target = (device->dev_instance.b[5] << 8 |
1967 			 device->dev_instance.b[4]);
1968 		ret = scsi_add_device(host, 0, target, 0);
1969 		if (ret)
1970 			goto err_out4;
1971 	}
1972 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1973 	if (host->transportt == fc_transport_template) {
1974 		struct fc_rport_identifiers ids = {
1975 			.roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
1976 		};
1977 
1978 		fc_host_node_name(host) = stor_device->node_name;
1979 		fc_host_port_name(host) = stor_device->port_name;
1980 		stor_device->rport = fc_remote_port_add(host, 0, &ids);
1981 		if (!stor_device->rport) {
1982 			ret = -ENOMEM;
1983 			goto err_out4;
1984 		}
1985 	}
1986 #endif
1987 	return 0;
1988 
1989 err_out4:
1990 	scsi_remove_host(host);
1991 
1992 err_out3:
1993 	destroy_workqueue(host_dev->handle_error_wq);
1994 
1995 err_out2:
1996 	/*
1997 	 * Once we have connected with the host, we would need to
1998 	 * to invoke storvsc_dev_remove() to rollback this state and
1999 	 * this call also frees up the stor_device; hence the jump around
2000 	 * err_out1 label.
2001 	 */
2002 	storvsc_dev_remove(device);
2003 	goto err_out0;
2004 
2005 err_out1:
2006 	kfree(stor_device->stor_chns);
2007 	kfree(stor_device);
2008 
2009 err_out0:
2010 	scsi_host_put(host);
2011 	return ret;
2012 }
2013 
2014 /* Change a scsi target's queue depth */
2015 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2016 {
2017 	if (queue_depth > scsi_driver.can_queue)
2018 		queue_depth = scsi_driver.can_queue;
2019 
2020 	return scsi_change_queue_depth(sdev, queue_depth);
2021 }
2022 
2023 static int storvsc_remove(struct hv_device *dev)
2024 {
2025 	struct storvsc_device *stor_device = hv_get_drvdata(dev);
2026 	struct Scsi_Host *host = stor_device->host;
2027 	struct hv_host_device *host_dev = shost_priv(host);
2028 
2029 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2030 	if (host->transportt == fc_transport_template) {
2031 		fc_remote_port_delete(stor_device->rport);
2032 		fc_remove_host(host);
2033 	}
2034 #endif
2035 	destroy_workqueue(host_dev->handle_error_wq);
2036 	scsi_remove_host(host);
2037 	storvsc_dev_remove(dev);
2038 	scsi_host_put(host);
2039 
2040 	return 0;
2041 }
2042 
2043 static int storvsc_suspend(struct hv_device *hv_dev)
2044 {
2045 	struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2046 	struct Scsi_Host *host = stor_device->host;
2047 	struct hv_host_device *host_dev = shost_priv(host);
2048 
2049 	storvsc_wait_to_drain(stor_device);
2050 
2051 	drain_workqueue(host_dev->handle_error_wq);
2052 
2053 	vmbus_close(hv_dev->channel);
2054 
2055 	kfree(stor_device->stor_chns);
2056 	stor_device->stor_chns = NULL;
2057 
2058 	cpumask_clear(&stor_device->alloced_cpus);
2059 
2060 	return 0;
2061 }
2062 
2063 static int storvsc_resume(struct hv_device *hv_dev)
2064 {
2065 	int ret;
2066 
2067 	ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2068 				     hv_dev_is_fc(hv_dev));
2069 	return ret;
2070 }
2071 
2072 static struct hv_driver storvsc_drv = {
2073 	.name = KBUILD_MODNAME,
2074 	.id_table = id_table,
2075 	.probe = storvsc_probe,
2076 	.remove = storvsc_remove,
2077 	.suspend = storvsc_suspend,
2078 	.resume = storvsc_resume,
2079 	.driver = {
2080 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
2081 	},
2082 };
2083 
2084 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2085 static struct fc_function_template fc_transport_functions = {
2086 	.show_host_node_name = 1,
2087 	.show_host_port_name = 1,
2088 };
2089 #endif
2090 
2091 static int __init storvsc_drv_init(void)
2092 {
2093 	int ret;
2094 
2095 	/*
2096 	 * Divide the ring buffer data size (which is 1 page less
2097 	 * than the ring buffer size since that page is reserved for
2098 	 * the ring buffer indices) by the max request size (which is
2099 	 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2100 	 */
2101 	max_outstanding_req_per_channel =
2102 		((storvsc_ringbuffer_size - PAGE_SIZE) /
2103 		ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2104 		sizeof(struct vstor_packet) + sizeof(u64) -
2105 		vmscsi_size_delta,
2106 		sizeof(u64)));
2107 
2108 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2109 	fc_transport_template = fc_attach_transport(&fc_transport_functions);
2110 	if (!fc_transport_template)
2111 		return -ENODEV;
2112 #endif
2113 
2114 	ret = vmbus_driver_register(&storvsc_drv);
2115 
2116 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2117 	if (ret)
2118 		fc_release_transport(fc_transport_template);
2119 #endif
2120 
2121 	return ret;
2122 }
2123 
2124 static void __exit storvsc_drv_exit(void)
2125 {
2126 	vmbus_driver_unregister(&storvsc_drv);
2127 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2128 	fc_release_transport(fc_transport_template);
2129 #endif
2130 }
2131 
2132 MODULE_LICENSE("GPL");
2133 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2134 module_init(storvsc_drv_init);
2135 module_exit(storvsc_drv_exit);
2136