xref: /openbmc/linux/drivers/scsi/storvsc_drv.c (revision 7288dd2f)
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 <linux/dma-mapping.h>
25 
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_host.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_tcq.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_devinfo.h>
33 #include <scsi/scsi_dbg.h>
34 #include <scsi/scsi_transport_fc.h>
35 #include <scsi/scsi_transport.h>
36 
37 /*
38  * All wire protocol details (storage protocol between the guest and the host)
39  * are consolidated here.
40  *
41  * Begin protocol definitions.
42  */
43 
44 /*
45  * Version history:
46  * V1 Beta: 0.1
47  * V1 RC < 2008/1/31: 1.0
48  * V1 RC > 2008/1/31:  2.0
49  * Win7: 4.2
50  * Win8: 5.1
51  * Win8.1: 6.0
52  * Win10: 6.2
53  */
54 
55 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_)	((((MAJOR_) & 0xff) << 8) | \
56 						(((MINOR_) & 0xff)))
57 #define VMSTOR_PROTO_VERSION_WIN6	VMSTOR_PROTO_VERSION(2, 0)
58 #define VMSTOR_PROTO_VERSION_WIN7	VMSTOR_PROTO_VERSION(4, 2)
59 #define VMSTOR_PROTO_VERSION_WIN8	VMSTOR_PROTO_VERSION(5, 1)
60 #define VMSTOR_PROTO_VERSION_WIN8_1	VMSTOR_PROTO_VERSION(6, 0)
61 #define VMSTOR_PROTO_VERSION_WIN10	VMSTOR_PROTO_VERSION(6, 2)
62 
63 /* channel callback timeout in ms */
64 #define CALLBACK_TIMEOUT               2
65 
66 /*  Packet structure describing virtual storage requests. */
67 enum vstor_packet_operation {
68 	VSTOR_OPERATION_COMPLETE_IO		= 1,
69 	VSTOR_OPERATION_REMOVE_DEVICE		= 2,
70 	VSTOR_OPERATION_EXECUTE_SRB		= 3,
71 	VSTOR_OPERATION_RESET_LUN		= 4,
72 	VSTOR_OPERATION_RESET_ADAPTER		= 5,
73 	VSTOR_OPERATION_RESET_BUS		= 6,
74 	VSTOR_OPERATION_BEGIN_INITIALIZATION	= 7,
75 	VSTOR_OPERATION_END_INITIALIZATION	= 8,
76 	VSTOR_OPERATION_QUERY_PROTOCOL_VERSION	= 9,
77 	VSTOR_OPERATION_QUERY_PROPERTIES	= 10,
78 	VSTOR_OPERATION_ENUMERATE_BUS		= 11,
79 	VSTOR_OPERATION_FCHBA_DATA              = 12,
80 	VSTOR_OPERATION_CREATE_SUB_CHANNELS     = 13,
81 	VSTOR_OPERATION_MAXIMUM                 = 13
82 };
83 
84 /*
85  * WWN packet for Fibre Channel HBA
86  */
87 
88 struct hv_fc_wwn_packet {
89 	u8	primary_active;
90 	u8	reserved1[3];
91 	u8	primary_port_wwn[8];
92 	u8	primary_node_wwn[8];
93 	u8	secondary_port_wwn[8];
94 	u8	secondary_node_wwn[8];
95 };
96 
97 
98 
99 /*
100  * SRB Flag Bits
101  */
102 
103 #define SRB_FLAGS_QUEUE_ACTION_ENABLE		0x00000002
104 #define SRB_FLAGS_DISABLE_DISCONNECT		0x00000004
105 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER	0x00000008
106 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE		0x00000010
107 #define SRB_FLAGS_DISABLE_AUTOSENSE		0x00000020
108 #define SRB_FLAGS_DATA_IN			0x00000040
109 #define SRB_FLAGS_DATA_OUT			0x00000080
110 #define SRB_FLAGS_NO_DATA_TRANSFER		0x00000000
111 #define SRB_FLAGS_UNSPECIFIED_DIRECTION	(SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
112 #define SRB_FLAGS_NO_QUEUE_FREEZE		0x00000100
113 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE		0x00000200
114 #define SRB_FLAGS_FREE_SENSE_BUFFER		0x00000400
115 
116 /*
117  * This flag indicates the request is part of the workflow for processing a D3.
118  */
119 #define SRB_FLAGS_D3_PROCESSING			0x00000800
120 #define SRB_FLAGS_IS_ACTIVE			0x00010000
121 #define SRB_FLAGS_ALLOCATED_FROM_ZONE		0x00020000
122 #define SRB_FLAGS_SGLIST_FROM_POOL		0x00040000
123 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE		0x00080000
124 #define SRB_FLAGS_NO_KEEP_AWAKE			0x00100000
125 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE	0x00200000
126 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT	0x00400000
127 #define SRB_FLAGS_DONT_START_NEXT_PACKET	0x00800000
128 #define SRB_FLAGS_PORT_DRIVER_RESERVED		0x0F000000
129 #define SRB_FLAGS_CLASS_DRIVER_RESERVED		0xF0000000
130 
131 #define SP_UNTAGGED			((unsigned char) ~0)
132 #define SRB_SIMPLE_TAG_REQUEST		0x20
133 
134 /*
135  * Platform neutral description of a scsi request -
136  * this remains the same across the write regardless of 32/64 bit
137  * note: it's patterned off the SCSI_PASS_THROUGH structure
138  */
139 #define STORVSC_MAX_CMD_LEN			0x10
140 
141 /* Sense buffer size is the same for all versions since Windows 8 */
142 #define STORVSC_SENSE_BUFFER_SIZE		0x14
143 #define STORVSC_MAX_BUF_LEN_WITH_PADDING	0x14
144 
145 /*
146  * The storage protocol version is determined during the
147  * initial exchange with the host.  It will indicate which
148  * storage functionality is available in the host.
149 */
150 static int vmstor_proto_version;
151 
152 #define STORVSC_LOGGING_NONE	0
153 #define STORVSC_LOGGING_ERROR	1
154 #define STORVSC_LOGGING_WARN	2
155 
156 static int logging_level = STORVSC_LOGGING_ERROR;
157 module_param(logging_level, int, S_IRUGO|S_IWUSR);
158 MODULE_PARM_DESC(logging_level,
159 	"Logging level, 0 - None, 1 - Error (default), 2 - Warning.");
160 
161 static inline bool do_logging(int level)
162 {
163 	return logging_level >= level;
164 }
165 
166 #define storvsc_log(dev, level, fmt, ...)			\
167 do {								\
168 	if (do_logging(level))					\
169 		dev_warn(&(dev)->device, fmt, ##__VA_ARGS__);	\
170 } while (0)
171 
172 struct vmscsi_request {
173 	u16 length;
174 	u8 srb_status;
175 	u8 scsi_status;
176 
177 	u8  port_number;
178 	u8  path_id;
179 	u8  target_id;
180 	u8  lun;
181 
182 	u8  cdb_length;
183 	u8  sense_info_length;
184 	u8  data_in;
185 	u8  reserved;
186 
187 	u32 data_transfer_length;
188 
189 	union {
190 		u8 cdb[STORVSC_MAX_CMD_LEN];
191 		u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
192 		u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
193 	};
194 	/*
195 	 * The following was added in win8.
196 	 */
197 	u16 reserve;
198 	u8  queue_tag;
199 	u8  queue_action;
200 	u32 srb_flags;
201 	u32 time_out_value;
202 	u32 queue_sort_ey;
203 
204 } __attribute((packed));
205 
206 /*
207  * The list of windows version in order of preference.
208  */
209 
210 static const int protocol_version[] = {
211 		VMSTOR_PROTO_VERSION_WIN10,
212 		VMSTOR_PROTO_VERSION_WIN8_1,
213 		VMSTOR_PROTO_VERSION_WIN8,
214 };
215 
216 
217 /*
218  * This structure is sent during the initialization phase to get the different
219  * properties of the channel.
220  */
221 
222 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL		0x1
223 
224 struct vmstorage_channel_properties {
225 	u32 reserved;
226 	u16 max_channel_cnt;
227 	u16 reserved1;
228 
229 	u32 flags;
230 	u32   max_transfer_bytes;
231 
232 	u64  reserved2;
233 } __packed;
234 
235 /*  This structure is sent during the storage protocol negotiations. */
236 struct vmstorage_protocol_version {
237 	/* Major (MSW) and minor (LSW) version numbers. */
238 	u16 major_minor;
239 
240 	/*
241 	 * Revision number is auto-incremented whenever this file is changed
242 	 * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
243 	 * definitely indicate incompatibility--but it does indicate mismatched
244 	 * builds.
245 	 * This is only used on the windows side. Just set it to 0.
246 	 */
247 	u16 revision;
248 } __packed;
249 
250 /* Channel Property Flags */
251 #define STORAGE_CHANNEL_REMOVABLE_FLAG		0x1
252 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG	0x2
253 
254 struct vstor_packet {
255 	/* Requested operation type */
256 	enum vstor_packet_operation operation;
257 
258 	/*  Flags - see below for values */
259 	u32 flags;
260 
261 	/* Status of the request returned from the server side. */
262 	u32 status;
263 
264 	/* Data payload area */
265 	union {
266 		/*
267 		 * Structure used to forward SCSI commands from the
268 		 * client to the server.
269 		 */
270 		struct vmscsi_request vm_srb;
271 
272 		/* Structure used to query channel properties. */
273 		struct vmstorage_channel_properties storage_channel_properties;
274 
275 		/* Used during version negotiations. */
276 		struct vmstorage_protocol_version version;
277 
278 		/* Fibre channel address packet */
279 		struct hv_fc_wwn_packet wwn_packet;
280 
281 		/* Number of sub-channels to create */
282 		u16 sub_channel_count;
283 
284 		/* This will be the maximum of the union members */
285 		u8  buffer[0x34];
286 	};
287 } __packed;
288 
289 /*
290  * Packet Flags:
291  *
292  * This flag indicates that the server should send back a completion for this
293  * packet.
294  */
295 
296 #define REQUEST_COMPLETION_FLAG	0x1
297 
298 /* Matches Windows-end */
299 enum storvsc_request_type {
300 	WRITE_TYPE = 0,
301 	READ_TYPE,
302 	UNKNOWN_TYPE,
303 };
304 
305 /*
306  * SRB status codes and masks. In the 8-bit field, the two high order bits
307  * are flags, while the remaining 6 bits are an integer status code.  The
308  * definitions here include only the subset of the integer status codes that
309  * are tested for in this driver.
310  */
311 #define SRB_STATUS_AUTOSENSE_VALID	0x80
312 #define SRB_STATUS_QUEUE_FROZEN		0x40
313 
314 /* SRB status integer codes */
315 #define SRB_STATUS_SUCCESS		0x01
316 #define SRB_STATUS_ABORTED		0x02
317 #define SRB_STATUS_ERROR		0x04
318 #define SRB_STATUS_INVALID_REQUEST	0x06
319 #define SRB_STATUS_DATA_OVERRUN		0x12
320 #define SRB_STATUS_INVALID_LUN		0x20
321 #define SRB_STATUS_INTERNAL_ERROR	0x30
322 
323 #define SRB_STATUS(status) \
324 	(status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
325 /*
326  * This is the end of Protocol specific defines.
327  */
328 
329 static int storvsc_ringbuffer_size = (128 * 1024);
330 static u32 max_outstanding_req_per_channel;
331 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
332 
333 static int storvsc_vcpus_per_sub_channel = 4;
334 static unsigned int storvsc_max_hw_queues;
335 
336 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
337 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
338 
339 module_param(storvsc_max_hw_queues, uint, 0644);
340 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
341 
342 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
343 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
344 
345 static int ring_avail_percent_lowater = 10;
346 module_param(ring_avail_percent_lowater, int, S_IRUGO);
347 MODULE_PARM_DESC(ring_avail_percent_lowater,
348 		"Select a channel if available ring size > this in percent");
349 
350 /*
351  * Timeout in seconds for all devices managed by this driver.
352  */
353 static int storvsc_timeout = 180;
354 
355 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
356 static struct scsi_transport_template *fc_transport_template;
357 #endif
358 
359 static struct scsi_host_template scsi_driver;
360 static void storvsc_on_channel_callback(void *context);
361 
362 #define STORVSC_MAX_LUNS_PER_TARGET			255
363 #define STORVSC_MAX_TARGETS				2
364 #define STORVSC_MAX_CHANNELS				8
365 
366 #define STORVSC_FC_MAX_LUNS_PER_TARGET			255
367 #define STORVSC_FC_MAX_TARGETS				128
368 #define STORVSC_FC_MAX_CHANNELS				8
369 #define STORVSC_FC_MAX_XFER_SIZE			((u32)(512 * 1024))
370 
371 #define STORVSC_IDE_MAX_LUNS_PER_TARGET			64
372 #define STORVSC_IDE_MAX_TARGETS				1
373 #define STORVSC_IDE_MAX_CHANNELS			1
374 
375 /*
376  * Upper bound on the size of a storvsc packet.
377  */
378 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
379 			      sizeof(struct vstor_packet))
380 
381 struct storvsc_cmd_request {
382 	struct scsi_cmnd *cmd;
383 
384 	struct hv_device *device;
385 
386 	/* Synchronize the request/response if needed */
387 	struct completion wait_event;
388 
389 	struct vmbus_channel_packet_multipage_buffer mpb;
390 	struct vmbus_packet_mpb_array *payload;
391 	u32 payload_sz;
392 
393 	struct vstor_packet vstor_packet;
394 };
395 
396 
397 /* A storvsc device is a device object that contains a vmbus channel */
398 struct storvsc_device {
399 	struct hv_device *device;
400 
401 	bool	 destroy;
402 	bool	 drain_notify;
403 	atomic_t num_outstanding_req;
404 	struct Scsi_Host *host;
405 
406 	wait_queue_head_t waiting_to_drain;
407 
408 	/*
409 	 * Each unique Port/Path/Target represents 1 channel ie scsi
410 	 * controller. In reality, the pathid, targetid is always 0
411 	 * and the port is set by us
412 	 */
413 	unsigned int port_number;
414 	unsigned char path_id;
415 	unsigned char target_id;
416 
417 	/*
418 	 * Max I/O, the device can support.
419 	 */
420 	u32   max_transfer_bytes;
421 	/*
422 	 * Number of sub-channels we will open.
423 	 */
424 	u16 num_sc;
425 	struct vmbus_channel **stor_chns;
426 	/*
427 	 * Mask of CPUs bound to subchannels.
428 	 */
429 	struct cpumask alloced_cpus;
430 	/*
431 	 * Serializes modifications of stor_chns[] from storvsc_do_io()
432 	 * and storvsc_change_target_cpu().
433 	 */
434 	spinlock_t lock;
435 	/* Used for vsc/vsp channel reset process */
436 	struct storvsc_cmd_request init_request;
437 	struct storvsc_cmd_request reset_request;
438 	/*
439 	 * Currently active port and node names for FC devices.
440 	 */
441 	u64 node_name;
442 	u64 port_name;
443 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
444 	struct fc_rport *rport;
445 #endif
446 };
447 
448 struct hv_host_device {
449 	struct hv_device *dev;
450 	unsigned int port;
451 	unsigned char path;
452 	unsigned char target;
453 	struct workqueue_struct *handle_error_wq;
454 	struct work_struct host_scan_work;
455 	struct Scsi_Host *host;
456 };
457 
458 struct storvsc_scan_work {
459 	struct work_struct work;
460 	struct Scsi_Host *host;
461 	u8 lun;
462 	u8 tgt_id;
463 };
464 
465 static void storvsc_device_scan(struct work_struct *work)
466 {
467 	struct storvsc_scan_work *wrk;
468 	struct scsi_device *sdev;
469 
470 	wrk = container_of(work, struct storvsc_scan_work, work);
471 
472 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
473 	if (!sdev)
474 		goto done;
475 	scsi_rescan_device(&sdev->sdev_gendev);
476 	scsi_device_put(sdev);
477 
478 done:
479 	kfree(wrk);
480 }
481 
482 static void storvsc_host_scan(struct work_struct *work)
483 {
484 	struct Scsi_Host *host;
485 	struct scsi_device *sdev;
486 	struct hv_host_device *host_device =
487 		container_of(work, struct hv_host_device, host_scan_work);
488 
489 	host = host_device->host;
490 	/*
491 	 * Before scanning the host, first check to see if any of the
492 	 * currently known devices have been hot removed. We issue a
493 	 * "unit ready" command against all currently known devices.
494 	 * This I/O will result in an error for devices that have been
495 	 * removed. As part of handling the I/O error, we remove the device.
496 	 *
497 	 * When a LUN is added or removed, the host sends us a signal to
498 	 * scan the host. Thus we are forced to discover the LUNs that
499 	 * may have been removed this way.
500 	 */
501 	mutex_lock(&host->scan_mutex);
502 	shost_for_each_device(sdev, host)
503 		scsi_test_unit_ready(sdev, 1, 1, NULL);
504 	mutex_unlock(&host->scan_mutex);
505 	/*
506 	 * Now scan the host to discover LUNs that may have been added.
507 	 */
508 	scsi_scan_host(host);
509 }
510 
511 static void storvsc_remove_lun(struct work_struct *work)
512 {
513 	struct storvsc_scan_work *wrk;
514 	struct scsi_device *sdev;
515 
516 	wrk = container_of(work, struct storvsc_scan_work, work);
517 	if (!scsi_host_get(wrk->host))
518 		goto done;
519 
520 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
521 
522 	if (sdev) {
523 		scsi_remove_device(sdev);
524 		scsi_device_put(sdev);
525 	}
526 	scsi_host_put(wrk->host);
527 
528 done:
529 	kfree(wrk);
530 }
531 
532 
533 /*
534  * We can get incoming messages from the host that are not in response to
535  * messages that we have sent out. An example of this would be messages
536  * received by the guest to notify dynamic addition/removal of LUNs. To
537  * deal with potential race conditions where the driver may be in the
538  * midst of being unloaded when we might receive an unsolicited message
539  * from the host, we have implemented a mechanism to gurantee sequential
540  * consistency:
541  *
542  * 1) Once the device is marked as being destroyed, we will fail all
543  *    outgoing messages.
544  * 2) We permit incoming messages when the device is being destroyed,
545  *    only to properly account for messages already sent out.
546  */
547 
548 static inline struct storvsc_device *get_out_stor_device(
549 					struct hv_device *device)
550 {
551 	struct storvsc_device *stor_device;
552 
553 	stor_device = hv_get_drvdata(device);
554 
555 	if (stor_device && stor_device->destroy)
556 		stor_device = NULL;
557 
558 	return stor_device;
559 }
560 
561 
562 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
563 {
564 	dev->drain_notify = true;
565 	wait_event(dev->waiting_to_drain,
566 		   atomic_read(&dev->num_outstanding_req) == 0);
567 	dev->drain_notify = false;
568 }
569 
570 static inline struct storvsc_device *get_in_stor_device(
571 					struct hv_device *device)
572 {
573 	struct storvsc_device *stor_device;
574 
575 	stor_device = hv_get_drvdata(device);
576 
577 	if (!stor_device)
578 		goto get_in_err;
579 
580 	/*
581 	 * If the device is being destroyed; allow incoming
582 	 * traffic only to cleanup outstanding requests.
583 	 */
584 
585 	if (stor_device->destroy  &&
586 		(atomic_read(&stor_device->num_outstanding_req) == 0))
587 		stor_device = NULL;
588 
589 get_in_err:
590 	return stor_device;
591 
592 }
593 
594 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
595 				      u32 new)
596 {
597 	struct storvsc_device *stor_device;
598 	struct vmbus_channel *cur_chn;
599 	bool old_is_alloced = false;
600 	struct hv_device *device;
601 	unsigned long flags;
602 	int cpu;
603 
604 	device = channel->primary_channel ?
605 			channel->primary_channel->device_obj
606 				: channel->device_obj;
607 	stor_device = get_out_stor_device(device);
608 	if (!stor_device)
609 		return;
610 
611 	/* See storvsc_do_io() -> get_og_chn(). */
612 	spin_lock_irqsave(&stor_device->lock, flags);
613 
614 	/*
615 	 * Determines if the storvsc device has other channels assigned to
616 	 * the "old" CPU to update the alloced_cpus mask and the stor_chns
617 	 * array.
618 	 */
619 	if (device->channel != channel && device->channel->target_cpu == old) {
620 		cur_chn = device->channel;
621 		old_is_alloced = true;
622 		goto old_is_alloced;
623 	}
624 	list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
625 		if (cur_chn == channel)
626 			continue;
627 		if (cur_chn->target_cpu == old) {
628 			old_is_alloced = true;
629 			goto old_is_alloced;
630 		}
631 	}
632 
633 old_is_alloced:
634 	if (old_is_alloced)
635 		WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
636 	else
637 		cpumask_clear_cpu(old, &stor_device->alloced_cpus);
638 
639 	/* "Flush" the stor_chns array. */
640 	for_each_possible_cpu(cpu) {
641 		if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
642 					cpu, &stor_device->alloced_cpus))
643 			WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
644 	}
645 
646 	WRITE_ONCE(stor_device->stor_chns[new], channel);
647 	cpumask_set_cpu(new, &stor_device->alloced_cpus);
648 
649 	spin_unlock_irqrestore(&stor_device->lock, flags);
650 }
651 
652 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
653 {
654 	struct storvsc_cmd_request *request =
655 		(struct storvsc_cmd_request *)(unsigned long)rqst_addr;
656 
657 	if (rqst_addr == VMBUS_RQST_INIT)
658 		return VMBUS_RQST_INIT;
659 	if (rqst_addr == VMBUS_RQST_RESET)
660 		return VMBUS_RQST_RESET;
661 
662 	/*
663 	 * Cannot return an ID of 0, which is reserved for an unsolicited
664 	 * message from Hyper-V.
665 	 */
666 	return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
667 }
668 
669 static void handle_sc_creation(struct vmbus_channel *new_sc)
670 {
671 	struct hv_device *device = new_sc->primary_channel->device_obj;
672 	struct device *dev = &device->device;
673 	struct storvsc_device *stor_device;
674 	struct vmstorage_channel_properties props;
675 	int ret;
676 
677 	stor_device = get_out_stor_device(device);
678 	if (!stor_device)
679 		return;
680 
681 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
682 	new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
683 
684 	new_sc->next_request_id_callback = storvsc_next_request_id;
685 
686 	ret = vmbus_open(new_sc,
687 			 storvsc_ringbuffer_size,
688 			 storvsc_ringbuffer_size,
689 			 (void *)&props,
690 			 sizeof(struct vmstorage_channel_properties),
691 			 storvsc_on_channel_callback, new_sc);
692 
693 	/* In case vmbus_open() fails, we don't use the sub-channel. */
694 	if (ret != 0) {
695 		dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
696 		return;
697 	}
698 
699 	new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
700 
701 	/* Add the sub-channel to the array of available channels. */
702 	stor_device->stor_chns[new_sc->target_cpu] = new_sc;
703 	cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
704 }
705 
706 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
707 {
708 	struct device *dev = &device->device;
709 	struct storvsc_device *stor_device;
710 	int num_sc;
711 	struct storvsc_cmd_request *request;
712 	struct vstor_packet *vstor_packet;
713 	int ret, t;
714 
715 	/*
716 	 * If the number of CPUs is artificially restricted, such as
717 	 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
718 	 * sub-channels >= the number of CPUs. These sub-channels
719 	 * should not be created. The primary channel is already created
720 	 * and assigned to one CPU, so check against # CPUs - 1.
721 	 */
722 	num_sc = min((int)(num_online_cpus() - 1), max_chns);
723 	if (!num_sc)
724 		return;
725 
726 	stor_device = get_out_stor_device(device);
727 	if (!stor_device)
728 		return;
729 
730 	stor_device->num_sc = num_sc;
731 	request = &stor_device->init_request;
732 	vstor_packet = &request->vstor_packet;
733 
734 	/*
735 	 * Establish a handler for dealing with subchannels.
736 	 */
737 	vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
738 
739 	/*
740 	 * Request the host to create sub-channels.
741 	 */
742 	memset(request, 0, sizeof(struct storvsc_cmd_request));
743 	init_completion(&request->wait_event);
744 	vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
745 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
746 	vstor_packet->sub_channel_count = num_sc;
747 
748 	ret = vmbus_sendpacket(device->channel, vstor_packet,
749 			       sizeof(struct vstor_packet),
750 			       VMBUS_RQST_INIT,
751 			       VM_PKT_DATA_INBAND,
752 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
753 
754 	if (ret != 0) {
755 		dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
756 		return;
757 	}
758 
759 	t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
760 	if (t == 0) {
761 		dev_err(dev, "Failed to create sub-channel: timed out\n");
762 		return;
763 	}
764 
765 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
766 	    vstor_packet->status != 0) {
767 		dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
768 			vstor_packet->operation, vstor_packet->status);
769 		return;
770 	}
771 
772 	/*
773 	 * We need to do nothing here, because vmbus_process_offer()
774 	 * invokes channel->sc_creation_callback, which will open and use
775 	 * the sub-channel(s).
776 	 */
777 }
778 
779 static void cache_wwn(struct storvsc_device *stor_device,
780 		      struct vstor_packet *vstor_packet)
781 {
782 	/*
783 	 * Cache the currently active port and node ww names.
784 	 */
785 	if (vstor_packet->wwn_packet.primary_active) {
786 		stor_device->node_name =
787 			wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
788 		stor_device->port_name =
789 			wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
790 	} else {
791 		stor_device->node_name =
792 			wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
793 		stor_device->port_name =
794 			wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
795 	}
796 }
797 
798 
799 static int storvsc_execute_vstor_op(struct hv_device *device,
800 				    struct storvsc_cmd_request *request,
801 				    bool status_check)
802 {
803 	struct storvsc_device *stor_device;
804 	struct vstor_packet *vstor_packet;
805 	int ret, t;
806 
807 	stor_device = get_out_stor_device(device);
808 	if (!stor_device)
809 		return -ENODEV;
810 
811 	vstor_packet = &request->vstor_packet;
812 
813 	init_completion(&request->wait_event);
814 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
815 
816 	ret = vmbus_sendpacket(device->channel, vstor_packet,
817 			       sizeof(struct vstor_packet),
818 			       VMBUS_RQST_INIT,
819 			       VM_PKT_DATA_INBAND,
820 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
821 	if (ret != 0)
822 		return ret;
823 
824 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
825 	if (t == 0)
826 		return -ETIMEDOUT;
827 
828 	if (!status_check)
829 		return ret;
830 
831 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
832 	    vstor_packet->status != 0)
833 		return -EINVAL;
834 
835 	return ret;
836 }
837 
838 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
839 {
840 	struct storvsc_device *stor_device;
841 	struct storvsc_cmd_request *request;
842 	struct vstor_packet *vstor_packet;
843 	int ret, i;
844 	int max_chns;
845 	bool process_sub_channels = false;
846 
847 	stor_device = get_out_stor_device(device);
848 	if (!stor_device)
849 		return -ENODEV;
850 
851 	request = &stor_device->init_request;
852 	vstor_packet = &request->vstor_packet;
853 
854 	/*
855 	 * Now, initiate the vsc/vsp initialization protocol on the open
856 	 * channel
857 	 */
858 	memset(request, 0, sizeof(struct storvsc_cmd_request));
859 	vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
860 	ret = storvsc_execute_vstor_op(device, request, true);
861 	if (ret)
862 		return ret;
863 	/*
864 	 * Query host supported protocol version.
865 	 */
866 
867 	for (i = 0; i < ARRAY_SIZE(protocol_version); i++) {
868 		/* reuse the packet for version range supported */
869 		memset(vstor_packet, 0, sizeof(struct vstor_packet));
870 		vstor_packet->operation =
871 			VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
872 
873 		vstor_packet->version.major_minor = protocol_version[i];
874 
875 		/*
876 		 * The revision number is only used in Windows; set it to 0.
877 		 */
878 		vstor_packet->version.revision = 0;
879 		ret = storvsc_execute_vstor_op(device, request, false);
880 		if (ret != 0)
881 			return ret;
882 
883 		if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
884 			return -EINVAL;
885 
886 		if (vstor_packet->status == 0) {
887 			vmstor_proto_version = protocol_version[i];
888 
889 			break;
890 		}
891 	}
892 
893 	if (vstor_packet->status != 0) {
894 		dev_err(&device->device, "Obsolete Hyper-V version\n");
895 		return -EINVAL;
896 	}
897 
898 
899 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
900 	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
901 	ret = storvsc_execute_vstor_op(device, request, true);
902 	if (ret != 0)
903 		return ret;
904 
905 	/*
906 	 * Check to see if multi-channel support is there.
907 	 * Hosts that implement protocol version of 5.1 and above
908 	 * support multi-channel.
909 	 */
910 	max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
911 
912 	/*
913 	 * Allocate state to manage the sub-channels.
914 	 * We allocate an array based on the numbers of possible CPUs
915 	 * (Hyper-V does not support cpu online/offline).
916 	 * This Array will be sparseley populated with unique
917 	 * channels - primary + sub-channels.
918 	 * We will however populate all the slots to evenly distribute
919 	 * the load.
920 	 */
921 	stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
922 					 GFP_KERNEL);
923 	if (stor_device->stor_chns == NULL)
924 		return -ENOMEM;
925 
926 	device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
927 
928 	stor_device->stor_chns[device->channel->target_cpu] = device->channel;
929 	cpumask_set_cpu(device->channel->target_cpu,
930 			&stor_device->alloced_cpus);
931 
932 	if (vstor_packet->storage_channel_properties.flags &
933 	    STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
934 		process_sub_channels = true;
935 
936 	stor_device->max_transfer_bytes =
937 		vstor_packet->storage_channel_properties.max_transfer_bytes;
938 
939 	if (!is_fc)
940 		goto done;
941 
942 	/*
943 	 * For FC devices retrieve FC HBA data.
944 	 */
945 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
946 	vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
947 	ret = storvsc_execute_vstor_op(device, request, true);
948 	if (ret != 0)
949 		return ret;
950 
951 	/*
952 	 * Cache the currently active port and node ww names.
953 	 */
954 	cache_wwn(stor_device, vstor_packet);
955 
956 done:
957 
958 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
959 	vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
960 	ret = storvsc_execute_vstor_op(device, request, true);
961 	if (ret != 0)
962 		return ret;
963 
964 	if (process_sub_channels)
965 		handle_multichannel_storage(device, max_chns);
966 
967 	return ret;
968 }
969 
970 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
971 				struct scsi_cmnd *scmnd,
972 				struct Scsi_Host *host,
973 				u8 asc, u8 ascq)
974 {
975 	struct storvsc_scan_work *wrk;
976 	void (*process_err_fn)(struct work_struct *work);
977 	struct hv_host_device *host_dev = shost_priv(host);
978 
979 	switch (SRB_STATUS(vm_srb->srb_status)) {
980 	case SRB_STATUS_ERROR:
981 	case SRB_STATUS_ABORTED:
982 	case SRB_STATUS_INVALID_REQUEST:
983 	case SRB_STATUS_INTERNAL_ERROR:
984 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
985 			/* Check for capacity change */
986 			if ((asc == 0x2a) && (ascq == 0x9)) {
987 				process_err_fn = storvsc_device_scan;
988 				/* Retry the I/O that triggered this. */
989 				set_host_byte(scmnd, DID_REQUEUE);
990 				goto do_work;
991 			}
992 
993 			/*
994 			 * Check for "Operating parameters have changed"
995 			 * due to Hyper-V changing the VHD/VHDX BlockSize
996 			 * when adding/removing a differencing disk. This
997 			 * causes discard_granularity to change, so do a
998 			 * rescan to pick up the new granularity. We don't
999 			 * want scsi_report_sense() to output a message
1000 			 * that a sysadmin wouldn't know what to do with.
1001 			 */
1002 			if ((asc == 0x3f) && (ascq != 0x03) &&
1003 					(ascq != 0x0e)) {
1004 				process_err_fn = storvsc_device_scan;
1005 				set_host_byte(scmnd, DID_REQUEUE);
1006 				goto do_work;
1007 			}
1008 
1009 			/*
1010 			 * Otherwise, let upper layer deal with the
1011 			 * error when sense message is present
1012 			 */
1013 			return;
1014 		}
1015 
1016 		/*
1017 		 * If there is an error; offline the device since all
1018 		 * error recovery strategies would have already been
1019 		 * deployed on the host side. However, if the command
1020 		 * were a pass-through command deal with it appropriately.
1021 		 */
1022 		switch (scmnd->cmnd[0]) {
1023 		case ATA_16:
1024 		case ATA_12:
1025 			set_host_byte(scmnd, DID_PASSTHROUGH);
1026 			break;
1027 		/*
1028 		 * On some Hyper-V hosts TEST_UNIT_READY command can
1029 		 * return SRB_STATUS_ERROR. Let the upper level code
1030 		 * deal with it based on the sense information.
1031 		 */
1032 		case TEST_UNIT_READY:
1033 			break;
1034 		default:
1035 			set_host_byte(scmnd, DID_ERROR);
1036 		}
1037 		return;
1038 
1039 	case SRB_STATUS_INVALID_LUN:
1040 		set_host_byte(scmnd, DID_NO_CONNECT);
1041 		process_err_fn = storvsc_remove_lun;
1042 		goto do_work;
1043 
1044 	}
1045 	return;
1046 
1047 do_work:
1048 	/*
1049 	 * We need to schedule work to process this error; schedule it.
1050 	 */
1051 	wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1052 	if (!wrk) {
1053 		set_host_byte(scmnd, DID_BAD_TARGET);
1054 		return;
1055 	}
1056 
1057 	wrk->host = host;
1058 	wrk->lun = vm_srb->lun;
1059 	wrk->tgt_id = vm_srb->target_id;
1060 	INIT_WORK(&wrk->work, process_err_fn);
1061 	queue_work(host_dev->handle_error_wq, &wrk->work);
1062 }
1063 
1064 
1065 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1066 				       struct storvsc_device *stor_dev)
1067 {
1068 	struct scsi_cmnd *scmnd = cmd_request->cmd;
1069 	struct scsi_sense_hdr sense_hdr;
1070 	struct vmscsi_request *vm_srb;
1071 	u32 data_transfer_length;
1072 	struct Scsi_Host *host;
1073 	u32 payload_sz = cmd_request->payload_sz;
1074 	void *payload = cmd_request->payload;
1075 	bool sense_ok;
1076 
1077 	host = stor_dev->host;
1078 
1079 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1080 	data_transfer_length = vm_srb->data_transfer_length;
1081 
1082 	scmnd->result = vm_srb->scsi_status;
1083 
1084 	if (scmnd->result) {
1085 		sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1086 				SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1087 
1088 		if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1089 			scsi_print_sense_hdr(scmnd->device, "storvsc",
1090 					     &sense_hdr);
1091 	}
1092 
1093 	if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1094 		storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1095 					 sense_hdr.ascq);
1096 		/*
1097 		 * The Windows driver set data_transfer_length on
1098 		 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1099 		 * is untouched.  In these cases we set it to 0.
1100 		 */
1101 		if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1102 			data_transfer_length = 0;
1103 	}
1104 
1105 	/* Validate data_transfer_length (from Hyper-V) */
1106 	if (data_transfer_length > cmd_request->payload->range.len)
1107 		data_transfer_length = cmd_request->payload->range.len;
1108 
1109 	scsi_set_resid(scmnd,
1110 		cmd_request->payload->range.len - data_transfer_length);
1111 
1112 	scsi_done(scmnd);
1113 
1114 	if (payload_sz >
1115 		sizeof(struct vmbus_channel_packet_multipage_buffer))
1116 		kfree(payload);
1117 }
1118 
1119 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1120 				  struct vstor_packet *vstor_packet,
1121 				  struct storvsc_cmd_request *request)
1122 {
1123 	struct vstor_packet *stor_pkt;
1124 	struct hv_device *device = stor_device->device;
1125 
1126 	stor_pkt = &request->vstor_packet;
1127 
1128 	/*
1129 	 * The current SCSI handling on the host side does
1130 	 * not correctly handle:
1131 	 * INQUIRY command with page code parameter set to 0x80
1132 	 * MODE_SENSE command with cmd[2] == 0x1c
1133 	 *
1134 	 * Setup srb and scsi status so this won't be fatal.
1135 	 * We do this so we can distinguish truly fatal failues
1136 	 * (srb status == 0x4) and off-line the device in that case.
1137 	 */
1138 
1139 	if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1140 	   (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1141 		vstor_packet->vm_srb.scsi_status = 0;
1142 		vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1143 	}
1144 
1145 	/* Copy over the status...etc */
1146 	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1147 	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1148 
1149 	/*
1150 	 * Copy over the sense_info_length, but limit to the known max
1151 	 * size if Hyper-V returns a bad value.
1152 	 */
1153 	stor_pkt->vm_srb.sense_info_length = min_t(u8, STORVSC_SENSE_BUFFER_SIZE,
1154 		vstor_packet->vm_srb.sense_info_length);
1155 
1156 	if (vstor_packet->vm_srb.scsi_status != 0 ||
1157 	    vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1158 
1159 		/*
1160 		 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1161 		 * return errors when detecting devices using TEST_UNIT_READY,
1162 		 * and logging these as errors produces unhelpful noise.
1163 		 */
1164 		int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1165 			STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1166 
1167 		storvsc_log(device, loglevel,
1168 			"tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1169 			scsi_cmd_to_rq(request->cmd)->tag,
1170 			stor_pkt->vm_srb.cdb[0],
1171 			vstor_packet->vm_srb.scsi_status,
1172 			vstor_packet->vm_srb.srb_status,
1173 			vstor_packet->status);
1174 	}
1175 
1176 	if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1177 	    (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1178 		memcpy(request->cmd->sense_buffer,
1179 		       vstor_packet->vm_srb.sense_data,
1180 		       stor_pkt->vm_srb.sense_info_length);
1181 
1182 	stor_pkt->vm_srb.data_transfer_length =
1183 		vstor_packet->vm_srb.data_transfer_length;
1184 
1185 	storvsc_command_completion(request, stor_device);
1186 
1187 	if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1188 		stor_device->drain_notify)
1189 		wake_up(&stor_device->waiting_to_drain);
1190 }
1191 
1192 static void storvsc_on_receive(struct storvsc_device *stor_device,
1193 			     struct vstor_packet *vstor_packet,
1194 			     struct storvsc_cmd_request *request)
1195 {
1196 	struct hv_host_device *host_dev;
1197 	switch (vstor_packet->operation) {
1198 	case VSTOR_OPERATION_COMPLETE_IO:
1199 		storvsc_on_io_completion(stor_device, vstor_packet, request);
1200 		break;
1201 
1202 	case VSTOR_OPERATION_REMOVE_DEVICE:
1203 	case VSTOR_OPERATION_ENUMERATE_BUS:
1204 		host_dev = shost_priv(stor_device->host);
1205 		queue_work(
1206 			host_dev->handle_error_wq, &host_dev->host_scan_work);
1207 		break;
1208 
1209 	case VSTOR_OPERATION_FCHBA_DATA:
1210 		cache_wwn(stor_device, vstor_packet);
1211 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1212 		fc_host_node_name(stor_device->host) = stor_device->node_name;
1213 		fc_host_port_name(stor_device->host) = stor_device->port_name;
1214 #endif
1215 		break;
1216 	default:
1217 		break;
1218 	}
1219 }
1220 
1221 static void storvsc_on_channel_callback(void *context)
1222 {
1223 	struct vmbus_channel *channel = (struct vmbus_channel *)context;
1224 	const struct vmpacket_descriptor *desc;
1225 	struct hv_device *device;
1226 	struct storvsc_device *stor_device;
1227 	struct Scsi_Host *shost;
1228 	unsigned long time_limit = jiffies + msecs_to_jiffies(CALLBACK_TIMEOUT);
1229 
1230 	if (channel->primary_channel != NULL)
1231 		device = channel->primary_channel->device_obj;
1232 	else
1233 		device = channel->device_obj;
1234 
1235 	stor_device = get_in_stor_device(device);
1236 	if (!stor_device)
1237 		return;
1238 
1239 	shost = stor_device->host;
1240 
1241 	foreach_vmbus_pkt(desc, channel) {
1242 		struct vstor_packet *packet = hv_pkt_data(desc);
1243 		struct storvsc_cmd_request *request = NULL;
1244 		u32 pktlen = hv_pkt_datalen(desc);
1245 		u64 rqst_id = desc->trans_id;
1246 		u32 minlen = rqst_id ? sizeof(struct vstor_packet) :
1247 			sizeof(enum vstor_packet_operation);
1248 
1249 		if (unlikely(time_after(jiffies, time_limit))) {
1250 			hv_pkt_iter_close(channel);
1251 			return;
1252 		}
1253 
1254 		if (pktlen < minlen) {
1255 			dev_err(&device->device,
1256 				"Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1257 				rqst_id, pktlen, minlen);
1258 			continue;
1259 		}
1260 
1261 		if (rqst_id == VMBUS_RQST_INIT) {
1262 			request = &stor_device->init_request;
1263 		} else if (rqst_id == VMBUS_RQST_RESET) {
1264 			request = &stor_device->reset_request;
1265 		} else {
1266 			/* Hyper-V can send an unsolicited message with ID of 0 */
1267 			if (rqst_id == 0) {
1268 				/*
1269 				 * storvsc_on_receive() looks at the vstor_packet in the message
1270 				 * from the ring buffer.
1271 				 *
1272 				 * - If the operation in the vstor_packet is COMPLETE_IO, then
1273 				 *   we call storvsc_on_io_completion(), and dereference the
1274 				 *   guest memory address.  Make sure we don't call
1275 				 *   storvsc_on_io_completion() with a guest memory address
1276 				 *   that is zero if Hyper-V were to construct and send such
1277 				 *   a bogus packet.
1278 				 *
1279 				 * - If the operation in the vstor_packet is FCHBA_DATA, then
1280 				 *   we call cache_wwn(), and access the data payload area of
1281 				 *   the packet (wwn_packet); however, there is no guarantee
1282 				 *   that the packet is big enough to contain such area.
1283 				 *   Future-proof the code by rejecting such a bogus packet.
1284 				 */
1285 				if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1286 				    packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1287 					dev_err(&device->device, "Invalid packet with ID of 0\n");
1288 					continue;
1289 				}
1290 			} else {
1291 				struct scsi_cmnd *scmnd;
1292 
1293 				/* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1294 				scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1295 				if (scmnd == NULL) {
1296 					dev_err(&device->device, "Incorrect transaction ID\n");
1297 					continue;
1298 				}
1299 				request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1300 				scsi_dma_unmap(scmnd);
1301 			}
1302 
1303 			storvsc_on_receive(stor_device, packet, request);
1304 			continue;
1305 		}
1306 
1307 		memcpy(&request->vstor_packet, packet,
1308 		       sizeof(struct vstor_packet));
1309 		complete(&request->wait_event);
1310 	}
1311 }
1312 
1313 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1314 				  bool is_fc)
1315 {
1316 	struct vmstorage_channel_properties props;
1317 	int ret;
1318 
1319 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1320 
1321 	device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1322 	device->channel->next_request_id_callback = storvsc_next_request_id;
1323 
1324 	ret = vmbus_open(device->channel,
1325 			 ring_size,
1326 			 ring_size,
1327 			 (void *)&props,
1328 			 sizeof(struct vmstorage_channel_properties),
1329 			 storvsc_on_channel_callback, device->channel);
1330 
1331 	if (ret != 0)
1332 		return ret;
1333 
1334 	ret = storvsc_channel_init(device, is_fc);
1335 
1336 	return ret;
1337 }
1338 
1339 static int storvsc_dev_remove(struct hv_device *device)
1340 {
1341 	struct storvsc_device *stor_device;
1342 
1343 	stor_device = hv_get_drvdata(device);
1344 
1345 	stor_device->destroy = true;
1346 
1347 	/* Make sure flag is set before waiting */
1348 	wmb();
1349 
1350 	/*
1351 	 * At this point, all outbound traffic should be disable. We
1352 	 * only allow inbound traffic (responses) to proceed so that
1353 	 * outstanding requests can be completed.
1354 	 */
1355 
1356 	storvsc_wait_to_drain(stor_device);
1357 
1358 	/*
1359 	 * Since we have already drained, we don't need to busy wait
1360 	 * as was done in final_release_stor_device()
1361 	 * Note that we cannot set the ext pointer to NULL until
1362 	 * we have drained - to drain the outgoing packets, we need to
1363 	 * allow incoming packets.
1364 	 */
1365 	hv_set_drvdata(device, NULL);
1366 
1367 	/* Close the channel */
1368 	vmbus_close(device->channel);
1369 
1370 	kfree(stor_device->stor_chns);
1371 	kfree(stor_device);
1372 	return 0;
1373 }
1374 
1375 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1376 					u16 q_num)
1377 {
1378 	u16 slot = 0;
1379 	u16 hash_qnum;
1380 	const struct cpumask *node_mask;
1381 	int num_channels, tgt_cpu;
1382 
1383 	if (stor_device->num_sc == 0) {
1384 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1385 		return stor_device->device->channel;
1386 	}
1387 
1388 	/*
1389 	 * Our channel array is sparsley populated and we
1390 	 * initiated I/O on a processor/hw-q that does not
1391 	 * currently have a designated channel. Fix this.
1392 	 * The strategy is simple:
1393 	 * I. Ensure NUMA locality
1394 	 * II. Distribute evenly (best effort)
1395 	 */
1396 
1397 	node_mask = cpumask_of_node(cpu_to_node(q_num));
1398 
1399 	num_channels = 0;
1400 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1401 		if (cpumask_test_cpu(tgt_cpu, node_mask))
1402 			num_channels++;
1403 	}
1404 	if (num_channels == 0) {
1405 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1406 		return stor_device->device->channel;
1407 	}
1408 
1409 	hash_qnum = q_num;
1410 	while (hash_qnum >= num_channels)
1411 		hash_qnum -= num_channels;
1412 
1413 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1414 		if (!cpumask_test_cpu(tgt_cpu, node_mask))
1415 			continue;
1416 		if (slot == hash_qnum)
1417 			break;
1418 		slot++;
1419 	}
1420 
1421 	stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1422 
1423 	return stor_device->stor_chns[q_num];
1424 }
1425 
1426 
1427 static int storvsc_do_io(struct hv_device *device,
1428 			 struct storvsc_cmd_request *request, u16 q_num)
1429 {
1430 	struct storvsc_device *stor_device;
1431 	struct vstor_packet *vstor_packet;
1432 	struct vmbus_channel *outgoing_channel, *channel;
1433 	unsigned long flags;
1434 	int ret = 0;
1435 	const struct cpumask *node_mask;
1436 	int tgt_cpu;
1437 
1438 	vstor_packet = &request->vstor_packet;
1439 	stor_device = get_out_stor_device(device);
1440 
1441 	if (!stor_device)
1442 		return -ENODEV;
1443 
1444 
1445 	request->device  = device;
1446 	/*
1447 	 * Select an appropriate channel to send the request out.
1448 	 */
1449 	/* See storvsc_change_target_cpu(). */
1450 	outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1451 	if (outgoing_channel != NULL) {
1452 		if (outgoing_channel->target_cpu == q_num) {
1453 			/*
1454 			 * Ideally, we want to pick a different channel if
1455 			 * available on the same NUMA node.
1456 			 */
1457 			node_mask = cpumask_of_node(cpu_to_node(q_num));
1458 			for_each_cpu_wrap(tgt_cpu,
1459 				 &stor_device->alloced_cpus, q_num + 1) {
1460 				if (!cpumask_test_cpu(tgt_cpu, node_mask))
1461 					continue;
1462 				if (tgt_cpu == q_num)
1463 					continue;
1464 				channel = READ_ONCE(
1465 					stor_device->stor_chns[tgt_cpu]);
1466 				if (channel == NULL)
1467 					continue;
1468 				if (hv_get_avail_to_write_percent(
1469 							&channel->outbound)
1470 						> ring_avail_percent_lowater) {
1471 					outgoing_channel = channel;
1472 					goto found_channel;
1473 				}
1474 			}
1475 
1476 			/*
1477 			 * All the other channels on the same NUMA node are
1478 			 * busy. Try to use the channel on the current CPU
1479 			 */
1480 			if (hv_get_avail_to_write_percent(
1481 						&outgoing_channel->outbound)
1482 					> ring_avail_percent_lowater)
1483 				goto found_channel;
1484 
1485 			/*
1486 			 * If we reach here, all the channels on the current
1487 			 * NUMA node are busy. Try to find a channel in
1488 			 * other NUMA nodes
1489 			 */
1490 			for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1491 				if (cpumask_test_cpu(tgt_cpu, node_mask))
1492 					continue;
1493 				channel = READ_ONCE(
1494 					stor_device->stor_chns[tgt_cpu]);
1495 				if (channel == NULL)
1496 					continue;
1497 				if (hv_get_avail_to_write_percent(
1498 							&channel->outbound)
1499 						> ring_avail_percent_lowater) {
1500 					outgoing_channel = channel;
1501 					goto found_channel;
1502 				}
1503 			}
1504 		}
1505 	} else {
1506 		spin_lock_irqsave(&stor_device->lock, flags);
1507 		outgoing_channel = stor_device->stor_chns[q_num];
1508 		if (outgoing_channel != NULL) {
1509 			spin_unlock_irqrestore(&stor_device->lock, flags);
1510 			goto found_channel;
1511 		}
1512 		outgoing_channel = get_og_chn(stor_device, q_num);
1513 		spin_unlock_irqrestore(&stor_device->lock, flags);
1514 	}
1515 
1516 found_channel:
1517 	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1518 
1519 	vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1520 
1521 
1522 	vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1523 
1524 
1525 	vstor_packet->vm_srb.data_transfer_length =
1526 	request->payload->range.len;
1527 
1528 	vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1529 
1530 	if (request->payload->range.len) {
1531 
1532 		ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1533 				request->payload, request->payload_sz,
1534 				vstor_packet,
1535 				sizeof(struct vstor_packet),
1536 				(unsigned long)request);
1537 	} else {
1538 		ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1539 			       sizeof(struct vstor_packet),
1540 			       (unsigned long)request,
1541 			       VM_PKT_DATA_INBAND,
1542 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1543 	}
1544 
1545 	if (ret != 0)
1546 		return ret;
1547 
1548 	atomic_inc(&stor_device->num_outstanding_req);
1549 
1550 	return ret;
1551 }
1552 
1553 static int storvsc_device_alloc(struct scsi_device *sdevice)
1554 {
1555 	/*
1556 	 * Set blist flag to permit the reading of the VPD pages even when
1557 	 * the target may claim SPC-2 compliance. MSFT targets currently
1558 	 * claim SPC-2 compliance while they implement post SPC-2 features.
1559 	 * With this flag we can correctly handle WRITE_SAME_16 issues.
1560 	 *
1561 	 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1562 	 * still supports REPORT LUN.
1563 	 */
1564 	sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1565 
1566 	return 0;
1567 }
1568 
1569 static int storvsc_device_configure(struct scsi_device *sdevice)
1570 {
1571 	blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1572 
1573 	/* storvsc devices don't support MAINTENANCE_IN SCSI cmd */
1574 	sdevice->no_report_opcodes = 1;
1575 	sdevice->no_write_same = 1;
1576 
1577 	/*
1578 	 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1579 	 * if the device is a MSFT virtual device.  If the host is
1580 	 * WIN10 or newer, allow write_same.
1581 	 */
1582 	if (!strncmp(sdevice->vendor, "Msft", 4)) {
1583 		switch (vmstor_proto_version) {
1584 		case VMSTOR_PROTO_VERSION_WIN8:
1585 		case VMSTOR_PROTO_VERSION_WIN8_1:
1586 			sdevice->scsi_level = SCSI_SPC_3;
1587 			break;
1588 		}
1589 
1590 		if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1591 			sdevice->no_write_same = 0;
1592 	}
1593 
1594 	return 0;
1595 }
1596 
1597 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1598 			   sector_t capacity, int *info)
1599 {
1600 	sector_t nsect = capacity;
1601 	sector_t cylinders = nsect;
1602 	int heads, sectors_pt;
1603 
1604 	/*
1605 	 * We are making up these values; let us keep it simple.
1606 	 */
1607 	heads = 0xff;
1608 	sectors_pt = 0x3f;      /* Sectors per track */
1609 	sector_div(cylinders, heads * sectors_pt);
1610 	if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1611 		cylinders = 0xffff;
1612 
1613 	info[0] = heads;
1614 	info[1] = sectors_pt;
1615 	info[2] = (int)cylinders;
1616 
1617 	return 0;
1618 }
1619 
1620 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1621 {
1622 	struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1623 	struct hv_device *device = host_dev->dev;
1624 
1625 	struct storvsc_device *stor_device;
1626 	struct storvsc_cmd_request *request;
1627 	struct vstor_packet *vstor_packet;
1628 	int ret, t;
1629 
1630 	stor_device = get_out_stor_device(device);
1631 	if (!stor_device)
1632 		return FAILED;
1633 
1634 	request = &stor_device->reset_request;
1635 	vstor_packet = &request->vstor_packet;
1636 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1637 
1638 	init_completion(&request->wait_event);
1639 
1640 	vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1641 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1642 	vstor_packet->vm_srb.path_id = stor_device->path_id;
1643 
1644 	ret = vmbus_sendpacket(device->channel, vstor_packet,
1645 			       sizeof(struct vstor_packet),
1646 			       VMBUS_RQST_RESET,
1647 			       VM_PKT_DATA_INBAND,
1648 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1649 	if (ret != 0)
1650 		return FAILED;
1651 
1652 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1653 	if (t == 0)
1654 		return TIMEOUT_ERROR;
1655 
1656 
1657 	/*
1658 	 * At this point, all outstanding requests in the adapter
1659 	 * should have been flushed out and return to us
1660 	 * There is a potential race here where the host may be in
1661 	 * the process of responding when we return from here.
1662 	 * Just wait for all in-transit packets to be accounted for
1663 	 * before we return from here.
1664 	 */
1665 	storvsc_wait_to_drain(stor_device);
1666 
1667 	return SUCCESS;
1668 }
1669 
1670 /*
1671  * The host guarantees to respond to each command, although I/O latencies might
1672  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1673  * chance to perform EH.
1674  */
1675 static enum scsi_timeout_action storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1676 {
1677 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1678 	if (scmnd->device->host->transportt == fc_transport_template)
1679 		return fc_eh_timed_out(scmnd);
1680 #endif
1681 	return SCSI_EH_RESET_TIMER;
1682 }
1683 
1684 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1685 {
1686 	bool allowed = true;
1687 	u8 scsi_op = scmnd->cmnd[0];
1688 
1689 	switch (scsi_op) {
1690 	/* the host does not handle WRITE_SAME, log accident usage */
1691 	case WRITE_SAME:
1692 	/*
1693 	 * smartd sends this command and the host does not handle
1694 	 * this. So, don't send it.
1695 	 */
1696 	case SET_WINDOW:
1697 		set_host_byte(scmnd, DID_ERROR);
1698 		allowed = false;
1699 		break;
1700 	default:
1701 		break;
1702 	}
1703 	return allowed;
1704 }
1705 
1706 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1707 {
1708 	int ret;
1709 	struct hv_host_device *host_dev = shost_priv(host);
1710 	struct hv_device *dev = host_dev->dev;
1711 	struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1712 	struct scatterlist *sgl;
1713 	struct vmscsi_request *vm_srb;
1714 	struct vmbus_packet_mpb_array  *payload;
1715 	u32 payload_sz;
1716 	u32 length;
1717 
1718 	if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1719 		/*
1720 		 * On legacy hosts filter unimplemented commands.
1721 		 * Future hosts are expected to correctly handle
1722 		 * unsupported commands. Furthermore, it is
1723 		 * possible that some of the currently
1724 		 * unsupported commands maybe supported in
1725 		 * future versions of the host.
1726 		 */
1727 		if (!storvsc_scsi_cmd_ok(scmnd)) {
1728 			scsi_done(scmnd);
1729 			return 0;
1730 		}
1731 	}
1732 
1733 	/* Setup the cmd request */
1734 	cmd_request->cmd = scmnd;
1735 
1736 	memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1737 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1738 	vm_srb->time_out_value = 60;
1739 
1740 	vm_srb->srb_flags |=
1741 		SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1742 
1743 	if (scmnd->device->tagged_supported) {
1744 		vm_srb->srb_flags |=
1745 		(SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1746 		vm_srb->queue_tag = SP_UNTAGGED;
1747 		vm_srb->queue_action = SRB_SIMPLE_TAG_REQUEST;
1748 	}
1749 
1750 	/* Build the SRB */
1751 	switch (scmnd->sc_data_direction) {
1752 	case DMA_TO_DEVICE:
1753 		vm_srb->data_in = WRITE_TYPE;
1754 		vm_srb->srb_flags |= SRB_FLAGS_DATA_OUT;
1755 		break;
1756 	case DMA_FROM_DEVICE:
1757 		vm_srb->data_in = READ_TYPE;
1758 		vm_srb->srb_flags |= SRB_FLAGS_DATA_IN;
1759 		break;
1760 	case DMA_NONE:
1761 		vm_srb->data_in = UNKNOWN_TYPE;
1762 		vm_srb->srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1763 		break;
1764 	default:
1765 		/*
1766 		 * This is DMA_BIDIRECTIONAL or something else we are never
1767 		 * supposed to see here.
1768 		 */
1769 		WARN(1, "Unexpected data direction: %d\n",
1770 		     scmnd->sc_data_direction);
1771 		return -EINVAL;
1772 	}
1773 
1774 
1775 	vm_srb->port_number = host_dev->port;
1776 	vm_srb->path_id = scmnd->device->channel;
1777 	vm_srb->target_id = scmnd->device->id;
1778 	vm_srb->lun = scmnd->device->lun;
1779 
1780 	vm_srb->cdb_length = scmnd->cmd_len;
1781 
1782 	memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1783 
1784 	sgl = (struct scatterlist *)scsi_sglist(scmnd);
1785 
1786 	length = scsi_bufflen(scmnd);
1787 	payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1788 	payload_sz = 0;
1789 
1790 	if (scsi_sg_count(scmnd)) {
1791 		unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1792 		unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1793 		struct scatterlist *sg;
1794 		unsigned long hvpfn, hvpfns_to_add;
1795 		int j, i = 0, sg_count;
1796 
1797 		payload_sz = (hvpg_count * sizeof(u64) +
1798 			      sizeof(struct vmbus_packet_mpb_array));
1799 
1800 		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1801 			payload = kzalloc(payload_sz, GFP_ATOMIC);
1802 			if (!payload)
1803 				return SCSI_MLQUEUE_DEVICE_BUSY;
1804 		}
1805 
1806 		payload->range.len = length;
1807 		payload->range.offset = offset_in_hvpg;
1808 
1809 		sg_count = scsi_dma_map(scmnd);
1810 		if (sg_count < 0) {
1811 			ret = SCSI_MLQUEUE_DEVICE_BUSY;
1812 			goto err_free_payload;
1813 		}
1814 
1815 		for_each_sg(sgl, sg, sg_count, j) {
1816 			/*
1817 			 * Init values for the current sgl entry. hvpfns_to_add
1818 			 * is in units of Hyper-V size pages. Handling the
1819 			 * PAGE_SIZE != HV_HYP_PAGE_SIZE case also handles
1820 			 * values of sgl->offset that are larger than PAGE_SIZE.
1821 			 * Such offsets are handled even on other than the first
1822 			 * sgl entry, provided they are a multiple of PAGE_SIZE.
1823 			 */
1824 			hvpfn = HVPFN_DOWN(sg_dma_address(sg));
1825 			hvpfns_to_add = HVPFN_UP(sg_dma_address(sg) +
1826 						 sg_dma_len(sg)) - hvpfn;
1827 
1828 			/*
1829 			 * Fill the next portion of the PFN array with
1830 			 * sequential Hyper-V PFNs for the continguous physical
1831 			 * memory described by the sgl entry. The end of the
1832 			 * last sgl should be reached at the same time that
1833 			 * the PFN array is filled.
1834 			 */
1835 			while (hvpfns_to_add--)
1836 				payload->range.pfn_array[i++] = hvpfn++;
1837 		}
1838 	}
1839 
1840 	cmd_request->payload = payload;
1841 	cmd_request->payload_sz = payload_sz;
1842 
1843 	/* Invokes the vsc to start an IO */
1844 	ret = storvsc_do_io(dev, cmd_request, get_cpu());
1845 	put_cpu();
1846 
1847 	if (ret)
1848 		scsi_dma_unmap(scmnd);
1849 
1850 	if (ret == -EAGAIN) {
1851 		/* no more space */
1852 		ret = SCSI_MLQUEUE_DEVICE_BUSY;
1853 		goto err_free_payload;
1854 	}
1855 
1856 	return 0;
1857 
1858 err_free_payload:
1859 	if (payload_sz > sizeof(cmd_request->mpb))
1860 		kfree(payload);
1861 
1862 	return ret;
1863 }
1864 
1865 static struct scsi_host_template scsi_driver = {
1866 	.module	=		THIS_MODULE,
1867 	.name =			"storvsc_host_t",
1868 	.cmd_size =             sizeof(struct storvsc_cmd_request),
1869 	.bios_param =		storvsc_get_chs,
1870 	.queuecommand =		storvsc_queuecommand,
1871 	.eh_host_reset_handler =	storvsc_host_reset_handler,
1872 	.proc_name =		"storvsc_host",
1873 	.eh_timed_out =		storvsc_eh_timed_out,
1874 	.slave_alloc =		storvsc_device_alloc,
1875 	.slave_configure =	storvsc_device_configure,
1876 	.cmd_per_lun =		2048,
1877 	.this_id =		-1,
1878 	/* Ensure there are no gaps in presented sgls */
1879 	.virt_boundary_mask =	HV_HYP_PAGE_SIZE - 1,
1880 	.no_write_same =	1,
1881 	.track_queue_depth =	1,
1882 	.change_queue_depth =	storvsc_change_queue_depth,
1883 };
1884 
1885 enum {
1886 	SCSI_GUID,
1887 	IDE_GUID,
1888 	SFC_GUID,
1889 };
1890 
1891 static const struct hv_vmbus_device_id id_table[] = {
1892 	/* SCSI guid */
1893 	{ HV_SCSI_GUID,
1894 	  .driver_data = SCSI_GUID
1895 	},
1896 	/* IDE guid */
1897 	{ HV_IDE_GUID,
1898 	  .driver_data = IDE_GUID
1899 	},
1900 	/* Fibre Channel GUID */
1901 	{
1902 	  HV_SYNTHFC_GUID,
1903 	  .driver_data = SFC_GUID
1904 	},
1905 	{ },
1906 };
1907 
1908 MODULE_DEVICE_TABLE(vmbus, id_table);
1909 
1910 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1911 
1912 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1913 {
1914 	return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1915 }
1916 
1917 static int storvsc_probe(struct hv_device *device,
1918 			const struct hv_vmbus_device_id *dev_id)
1919 {
1920 	int ret;
1921 	int num_cpus = num_online_cpus();
1922 	int num_present_cpus = num_present_cpus();
1923 	struct Scsi_Host *host;
1924 	struct hv_host_device *host_dev;
1925 	bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1926 	bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1927 	int target = 0;
1928 	struct storvsc_device *stor_device;
1929 	int max_sub_channels = 0;
1930 	u32 max_xfer_bytes;
1931 
1932 	/*
1933 	 * We support sub-channels for storage on SCSI and FC controllers.
1934 	 * The number of sub-channels offerred is based on the number of
1935 	 * VCPUs in the guest.
1936 	 */
1937 	if (!dev_is_ide)
1938 		max_sub_channels =
1939 			(num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1940 
1941 	scsi_driver.can_queue = max_outstanding_req_per_channel *
1942 				(max_sub_channels + 1) *
1943 				(100 - ring_avail_percent_lowater) / 100;
1944 
1945 	host = scsi_host_alloc(&scsi_driver,
1946 			       sizeof(struct hv_host_device));
1947 	if (!host)
1948 		return -ENOMEM;
1949 
1950 	host_dev = shost_priv(host);
1951 	memset(host_dev, 0, sizeof(struct hv_host_device));
1952 
1953 	host_dev->port = host->host_no;
1954 	host_dev->dev = device;
1955 	host_dev->host = host;
1956 
1957 
1958 	stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1959 	if (!stor_device) {
1960 		ret = -ENOMEM;
1961 		goto err_out0;
1962 	}
1963 
1964 	stor_device->destroy = false;
1965 	init_waitqueue_head(&stor_device->waiting_to_drain);
1966 	stor_device->device = device;
1967 	stor_device->host = host;
1968 	spin_lock_init(&stor_device->lock);
1969 	hv_set_drvdata(device, stor_device);
1970 	dma_set_min_align_mask(&device->device, HV_HYP_PAGE_SIZE - 1);
1971 
1972 	stor_device->port_number = host->host_no;
1973 	ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1974 	if (ret)
1975 		goto err_out1;
1976 
1977 	host_dev->path = stor_device->path_id;
1978 	host_dev->target = stor_device->target_id;
1979 
1980 	switch (dev_id->driver_data) {
1981 	case SFC_GUID:
1982 		host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1983 		host->max_id = STORVSC_FC_MAX_TARGETS;
1984 		host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1985 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1986 		host->transportt = fc_transport_template;
1987 #endif
1988 		break;
1989 
1990 	case SCSI_GUID:
1991 		host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1992 		host->max_id = STORVSC_MAX_TARGETS;
1993 		host->max_channel = STORVSC_MAX_CHANNELS - 1;
1994 		break;
1995 
1996 	default:
1997 		host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1998 		host->max_id = STORVSC_IDE_MAX_TARGETS;
1999 		host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2000 		break;
2001 	}
2002 	/* max cmd length */
2003 	host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2004 	/*
2005 	 * Any reasonable Hyper-V configuration should provide
2006 	 * max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
2007 	 * protecting it from any weird value.
2008 	 */
2009 	max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_SIZE);
2010 	if (is_fc)
2011 		max_xfer_bytes = min(max_xfer_bytes, STORVSC_FC_MAX_XFER_SIZE);
2012 
2013 	/* max_hw_sectors_kb */
2014 	host->max_sectors = max_xfer_bytes >> 9;
2015 	/*
2016 	 * There are 2 requirements for Hyper-V storvsc sgl segments,
2017 	 * based on which the below calculation for max segments is
2018 	 * done:
2019 	 *
2020 	 * 1. Except for the first and last sgl segment, all sgl segments
2021 	 *    should be align to HV_HYP_PAGE_SIZE, that also means the
2022 	 *    maximum number of segments in a sgl can be calculated by
2023 	 *    dividing the total max transfer length by HV_HYP_PAGE_SIZE.
2024 	 *
2025 	 * 2. Except for the first and last, each entry in the SGL must
2026 	 *    have an offset that is a multiple of HV_HYP_PAGE_SIZE.
2027 	 */
2028 	host->sg_tablesize = (max_xfer_bytes >> HV_HYP_PAGE_SHIFT) + 1;
2029 	/*
2030 	 * For non-IDE disks, the host supports multiple channels.
2031 	 * Set the number of HW queues we are supporting.
2032 	 */
2033 	if (!dev_is_ide) {
2034 		if (storvsc_max_hw_queues > num_present_cpus) {
2035 			storvsc_max_hw_queues = 0;
2036 			storvsc_log(device, STORVSC_LOGGING_WARN,
2037 				"Resetting invalid storvsc_max_hw_queues value to default.\n");
2038 		}
2039 		if (storvsc_max_hw_queues)
2040 			host->nr_hw_queues = storvsc_max_hw_queues;
2041 		else
2042 			host->nr_hw_queues = num_present_cpus;
2043 	}
2044 
2045 	/*
2046 	 * Set the error handler work queue.
2047 	 */
2048 	host_dev->handle_error_wq =
2049 			alloc_ordered_workqueue("storvsc_error_wq_%d",
2050 						0,
2051 						host->host_no);
2052 	if (!host_dev->handle_error_wq) {
2053 		ret = -ENOMEM;
2054 		goto err_out2;
2055 	}
2056 	INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2057 	/* Register the HBA and start the scsi bus scan */
2058 	ret = scsi_add_host(host, &device->device);
2059 	if (ret != 0)
2060 		goto err_out3;
2061 
2062 	if (!dev_is_ide) {
2063 		scsi_scan_host(host);
2064 	} else {
2065 		target = (device->dev_instance.b[5] << 8 |
2066 			 device->dev_instance.b[4]);
2067 		ret = scsi_add_device(host, 0, target, 0);
2068 		if (ret)
2069 			goto err_out4;
2070 	}
2071 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2072 	if (host->transportt == fc_transport_template) {
2073 		struct fc_rport_identifiers ids = {
2074 			.roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2075 		};
2076 
2077 		fc_host_node_name(host) = stor_device->node_name;
2078 		fc_host_port_name(host) = stor_device->port_name;
2079 		stor_device->rport = fc_remote_port_add(host, 0, &ids);
2080 		if (!stor_device->rport) {
2081 			ret = -ENOMEM;
2082 			goto err_out4;
2083 		}
2084 	}
2085 #endif
2086 	return 0;
2087 
2088 err_out4:
2089 	scsi_remove_host(host);
2090 
2091 err_out3:
2092 	destroy_workqueue(host_dev->handle_error_wq);
2093 
2094 err_out2:
2095 	/*
2096 	 * Once we have connected with the host, we would need to
2097 	 * invoke storvsc_dev_remove() to rollback this state and
2098 	 * this call also frees up the stor_device; hence the jump around
2099 	 * err_out1 label.
2100 	 */
2101 	storvsc_dev_remove(device);
2102 	goto err_out0;
2103 
2104 err_out1:
2105 	kfree(stor_device->stor_chns);
2106 	kfree(stor_device);
2107 
2108 err_out0:
2109 	scsi_host_put(host);
2110 	return ret;
2111 }
2112 
2113 /* Change a scsi target's queue depth */
2114 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2115 {
2116 	if (queue_depth > scsi_driver.can_queue)
2117 		queue_depth = scsi_driver.can_queue;
2118 
2119 	return scsi_change_queue_depth(sdev, queue_depth);
2120 }
2121 
2122 static void storvsc_remove(struct hv_device *dev)
2123 {
2124 	struct storvsc_device *stor_device = hv_get_drvdata(dev);
2125 	struct Scsi_Host *host = stor_device->host;
2126 	struct hv_host_device *host_dev = shost_priv(host);
2127 
2128 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2129 	if (host->transportt == fc_transport_template) {
2130 		fc_remote_port_delete(stor_device->rport);
2131 		fc_remove_host(host);
2132 	}
2133 #endif
2134 	destroy_workqueue(host_dev->handle_error_wq);
2135 	scsi_remove_host(host);
2136 	storvsc_dev_remove(dev);
2137 	scsi_host_put(host);
2138 }
2139 
2140 static int storvsc_suspend(struct hv_device *hv_dev)
2141 {
2142 	struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2143 	struct Scsi_Host *host = stor_device->host;
2144 	struct hv_host_device *host_dev = shost_priv(host);
2145 
2146 	storvsc_wait_to_drain(stor_device);
2147 
2148 	drain_workqueue(host_dev->handle_error_wq);
2149 
2150 	vmbus_close(hv_dev->channel);
2151 
2152 	kfree(stor_device->stor_chns);
2153 	stor_device->stor_chns = NULL;
2154 
2155 	cpumask_clear(&stor_device->alloced_cpus);
2156 
2157 	return 0;
2158 }
2159 
2160 static int storvsc_resume(struct hv_device *hv_dev)
2161 {
2162 	int ret;
2163 
2164 	ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2165 				     hv_dev_is_fc(hv_dev));
2166 	return ret;
2167 }
2168 
2169 static struct hv_driver storvsc_drv = {
2170 	.name = KBUILD_MODNAME,
2171 	.id_table = id_table,
2172 	.probe = storvsc_probe,
2173 	.remove = storvsc_remove,
2174 	.suspend = storvsc_suspend,
2175 	.resume = storvsc_resume,
2176 	.driver = {
2177 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
2178 	},
2179 };
2180 
2181 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2182 static struct fc_function_template fc_transport_functions = {
2183 	.show_host_node_name = 1,
2184 	.show_host_port_name = 1,
2185 };
2186 #endif
2187 
2188 static int __init storvsc_drv_init(void)
2189 {
2190 	int ret;
2191 
2192 	/*
2193 	 * Divide the ring buffer data size (which is 1 page less
2194 	 * than the ring buffer size since that page is reserved for
2195 	 * the ring buffer indices) by the max request size (which is
2196 	 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2197 	 */
2198 	max_outstanding_req_per_channel =
2199 		((storvsc_ringbuffer_size - PAGE_SIZE) /
2200 		ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2201 		sizeof(struct vstor_packet) + sizeof(u64),
2202 		sizeof(u64)));
2203 
2204 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2205 	fc_transport_template = fc_attach_transport(&fc_transport_functions);
2206 	if (!fc_transport_template)
2207 		return -ENODEV;
2208 #endif
2209 
2210 	ret = vmbus_driver_register(&storvsc_drv);
2211 
2212 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2213 	if (ret)
2214 		fc_release_transport(fc_transport_template);
2215 #endif
2216 
2217 	return ret;
2218 }
2219 
2220 static void __exit storvsc_drv_exit(void)
2221 {
2222 	vmbus_driver_unregister(&storvsc_drv);
2223 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2224 	fc_release_transport(fc_transport_template);
2225 #endif
2226 }
2227 
2228 MODULE_LICENSE("GPL");
2229 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2230 module_init(storvsc_drv_init);
2231 module_exit(storvsc_drv_exit);
2232