xref: /openbmc/linux/drivers/scsi/storvsc_drv.c (revision 63705da3)
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  * The list of storage protocols in order of preference.
221  */
222 struct vmstor_protocol {
223 	int protocol_version;
224 	int sense_buffer_size;
225 	int vmscsi_size_delta;
226 };
227 
228 
229 static const struct vmstor_protocol vmstor_protocols[] = {
230 	{
231 		VMSTOR_PROTO_VERSION_WIN10,
232 		POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
233 		0
234 	},
235 	{
236 		VMSTOR_PROTO_VERSION_WIN8_1,
237 		POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
238 		0
239 	},
240 	{
241 		VMSTOR_PROTO_VERSION_WIN8,
242 		POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
243 		0
244 	},
245 	{
246 		VMSTOR_PROTO_VERSION_WIN7,
247 		PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
248 		sizeof(struct vmscsi_win8_extension),
249 	},
250 	{
251 		VMSTOR_PROTO_VERSION_WIN6,
252 		PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
253 		sizeof(struct vmscsi_win8_extension),
254 	}
255 };
256 
257 
258 /*
259  * This structure is sent during the initialization phase to get the different
260  * properties of the channel.
261  */
262 
263 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL		0x1
264 
265 struct vmstorage_channel_properties {
266 	u32 reserved;
267 	u16 max_channel_cnt;
268 	u16 reserved1;
269 
270 	u32 flags;
271 	u32   max_transfer_bytes;
272 
273 	u64  reserved2;
274 } __packed;
275 
276 /*  This structure is sent during the storage protocol negotiations. */
277 struct vmstorage_protocol_version {
278 	/* Major (MSW) and minor (LSW) version numbers. */
279 	u16 major_minor;
280 
281 	/*
282 	 * Revision number is auto-incremented whenever this file is changed
283 	 * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
284 	 * definitely indicate incompatibility--but it does indicate mismatched
285 	 * builds.
286 	 * This is only used on the windows side. Just set it to 0.
287 	 */
288 	u16 revision;
289 } __packed;
290 
291 /* Channel Property Flags */
292 #define STORAGE_CHANNEL_REMOVABLE_FLAG		0x1
293 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG	0x2
294 
295 struct vstor_packet {
296 	/* Requested operation type */
297 	enum vstor_packet_operation operation;
298 
299 	/*  Flags - see below for values */
300 	u32 flags;
301 
302 	/* Status of the request returned from the server side. */
303 	u32 status;
304 
305 	/* Data payload area */
306 	union {
307 		/*
308 		 * Structure used to forward SCSI commands from the
309 		 * client to the server.
310 		 */
311 		struct vmscsi_request vm_srb;
312 
313 		/* Structure used to query channel properties. */
314 		struct vmstorage_channel_properties storage_channel_properties;
315 
316 		/* Used during version negotiations. */
317 		struct vmstorage_protocol_version version;
318 
319 		/* Fibre channel address packet */
320 		struct hv_fc_wwn_packet wwn_packet;
321 
322 		/* Number of sub-channels to create */
323 		u16 sub_channel_count;
324 
325 		/* This will be the maximum of the union members */
326 		u8  buffer[0x34];
327 	};
328 } __packed;
329 
330 /*
331  * Packet Flags:
332  *
333  * This flag indicates that the server should send back a completion for this
334  * packet.
335  */
336 
337 #define REQUEST_COMPLETION_FLAG	0x1
338 
339 /* Matches Windows-end */
340 enum storvsc_request_type {
341 	WRITE_TYPE = 0,
342 	READ_TYPE,
343 	UNKNOWN_TYPE,
344 };
345 
346 /*
347  * SRB status codes and masks; a subset of the codes used here.
348  */
349 
350 #define SRB_STATUS_AUTOSENSE_VALID	0x80
351 #define SRB_STATUS_QUEUE_FROZEN		0x40
352 #define SRB_STATUS_INVALID_LUN	0x20
353 #define SRB_STATUS_SUCCESS	0x01
354 #define SRB_STATUS_ABORTED	0x02
355 #define SRB_STATUS_ERROR	0x04
356 #define SRB_STATUS_DATA_OVERRUN	0x12
357 
358 #define SRB_STATUS(status) \
359 	(status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
360 /*
361  * This is the end of Protocol specific defines.
362  */
363 
364 static int storvsc_ringbuffer_size = (128 * 1024);
365 static u32 max_outstanding_req_per_channel;
366 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
367 
368 static int storvsc_vcpus_per_sub_channel = 4;
369 static unsigned int storvsc_max_hw_queues;
370 
371 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
372 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
373 
374 module_param(storvsc_max_hw_queues, uint, 0644);
375 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
376 
377 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
378 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
379 
380 static int ring_avail_percent_lowater = 10;
381 module_param(ring_avail_percent_lowater, int, S_IRUGO);
382 MODULE_PARM_DESC(ring_avail_percent_lowater,
383 		"Select a channel if available ring size > this in percent");
384 
385 /*
386  * Timeout in seconds for all devices managed by this driver.
387  */
388 static int storvsc_timeout = 180;
389 
390 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
391 static struct scsi_transport_template *fc_transport_template;
392 #endif
393 
394 static struct scsi_host_template scsi_driver;
395 static void storvsc_on_channel_callback(void *context);
396 
397 #define STORVSC_MAX_LUNS_PER_TARGET			255
398 #define STORVSC_MAX_TARGETS				2
399 #define STORVSC_MAX_CHANNELS				8
400 
401 #define STORVSC_FC_MAX_LUNS_PER_TARGET			255
402 #define STORVSC_FC_MAX_TARGETS				128
403 #define STORVSC_FC_MAX_CHANNELS				8
404 
405 #define STORVSC_IDE_MAX_LUNS_PER_TARGET			64
406 #define STORVSC_IDE_MAX_TARGETS				1
407 #define STORVSC_IDE_MAX_CHANNELS			1
408 
409 /*
410  * Upper bound on the size of a storvsc packet. vmscsi_size_delta is not
411  * included in the calculation because it is set after STORVSC_MAX_PKT_SIZE
412  * is used in storvsc_connect_to_vsp
413  */
414 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
415 			      sizeof(struct vstor_packet))
416 
417 struct storvsc_cmd_request {
418 	struct scsi_cmnd *cmd;
419 
420 	struct hv_device *device;
421 
422 	/* Synchronize the request/response if needed */
423 	struct completion wait_event;
424 
425 	struct vmbus_channel_packet_multipage_buffer mpb;
426 	struct vmbus_packet_mpb_array *payload;
427 	u32 payload_sz;
428 
429 	struct vstor_packet vstor_packet;
430 };
431 
432 
433 /* A storvsc device is a device object that contains a vmbus channel */
434 struct storvsc_device {
435 	struct hv_device *device;
436 
437 	bool	 destroy;
438 	bool	 drain_notify;
439 	atomic_t num_outstanding_req;
440 	struct Scsi_Host *host;
441 
442 	wait_queue_head_t waiting_to_drain;
443 
444 	/*
445 	 * Each unique Port/Path/Target represents 1 channel ie scsi
446 	 * controller. In reality, the pathid, targetid is always 0
447 	 * and the port is set by us
448 	 */
449 	unsigned int port_number;
450 	unsigned char path_id;
451 	unsigned char target_id;
452 
453 	/*
454 	 * The size of the vmscsi_request has changed in win8. The
455 	 * additional size is because of new elements added to the
456 	 * structure. These elements are valid only when we are talking
457 	 * to a win8 host.
458 	 * Track the correction to size we need to apply. This value
459 	 * will likely change during protocol negotiation but it is
460 	 * valid to start by assuming pre-Win8.
461 	 */
462 	int vmscsi_size_delta;
463 
464 	/*
465 	 * Max I/O, the device can support.
466 	 */
467 	u32   max_transfer_bytes;
468 	/*
469 	 * Number of sub-channels we will open.
470 	 */
471 	u16 num_sc;
472 	struct vmbus_channel **stor_chns;
473 	/*
474 	 * Mask of CPUs bound to subchannels.
475 	 */
476 	struct cpumask alloced_cpus;
477 	/*
478 	 * Serializes modifications of stor_chns[] from storvsc_do_io()
479 	 * and storvsc_change_target_cpu().
480 	 */
481 	spinlock_t lock;
482 	/* Used for vsc/vsp channel reset process */
483 	struct storvsc_cmd_request init_request;
484 	struct storvsc_cmd_request reset_request;
485 	/*
486 	 * Currently active port and node names for FC devices.
487 	 */
488 	u64 node_name;
489 	u64 port_name;
490 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
491 	struct fc_rport *rport;
492 #endif
493 };
494 
495 struct hv_host_device {
496 	struct hv_device *dev;
497 	unsigned int port;
498 	unsigned char path;
499 	unsigned char target;
500 	struct workqueue_struct *handle_error_wq;
501 	struct work_struct host_scan_work;
502 	struct Scsi_Host *host;
503 };
504 
505 struct storvsc_scan_work {
506 	struct work_struct work;
507 	struct Scsi_Host *host;
508 	u8 lun;
509 	u8 tgt_id;
510 };
511 
512 static void storvsc_device_scan(struct work_struct *work)
513 {
514 	struct storvsc_scan_work *wrk;
515 	struct scsi_device *sdev;
516 
517 	wrk = container_of(work, struct storvsc_scan_work, work);
518 
519 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
520 	if (!sdev)
521 		goto done;
522 	scsi_rescan_device(&sdev->sdev_gendev);
523 	scsi_device_put(sdev);
524 
525 done:
526 	kfree(wrk);
527 }
528 
529 static void storvsc_host_scan(struct work_struct *work)
530 {
531 	struct Scsi_Host *host;
532 	struct scsi_device *sdev;
533 	struct hv_host_device *host_device =
534 		container_of(work, struct hv_host_device, host_scan_work);
535 
536 	host = host_device->host;
537 	/*
538 	 * Before scanning the host, first check to see if any of the
539 	 * currrently known devices have been hot removed. We issue a
540 	 * "unit ready" command against all currently known devices.
541 	 * This I/O will result in an error for devices that have been
542 	 * removed. As part of handling the I/O error, we remove the device.
543 	 *
544 	 * When a LUN is added or removed, the host sends us a signal to
545 	 * scan the host. Thus we are forced to discover the LUNs that
546 	 * may have been removed this way.
547 	 */
548 	mutex_lock(&host->scan_mutex);
549 	shost_for_each_device(sdev, host)
550 		scsi_test_unit_ready(sdev, 1, 1, NULL);
551 	mutex_unlock(&host->scan_mutex);
552 	/*
553 	 * Now scan the host to discover LUNs that may have been added.
554 	 */
555 	scsi_scan_host(host);
556 }
557 
558 static void storvsc_remove_lun(struct work_struct *work)
559 {
560 	struct storvsc_scan_work *wrk;
561 	struct scsi_device *sdev;
562 
563 	wrk = container_of(work, struct storvsc_scan_work, work);
564 	if (!scsi_host_get(wrk->host))
565 		goto done;
566 
567 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
568 
569 	if (sdev) {
570 		scsi_remove_device(sdev);
571 		scsi_device_put(sdev);
572 	}
573 	scsi_host_put(wrk->host);
574 
575 done:
576 	kfree(wrk);
577 }
578 
579 
580 /*
581  * We can get incoming messages from the host that are not in response to
582  * messages that we have sent out. An example of this would be messages
583  * received by the guest to notify dynamic addition/removal of LUNs. To
584  * deal with potential race conditions where the driver may be in the
585  * midst of being unloaded when we might receive an unsolicited message
586  * from the host, we have implemented a mechanism to gurantee sequential
587  * consistency:
588  *
589  * 1) Once the device is marked as being destroyed, we will fail all
590  *    outgoing messages.
591  * 2) We permit incoming messages when the device is being destroyed,
592  *    only to properly account for messages already sent out.
593  */
594 
595 static inline struct storvsc_device *get_out_stor_device(
596 					struct hv_device *device)
597 {
598 	struct storvsc_device *stor_device;
599 
600 	stor_device = hv_get_drvdata(device);
601 
602 	if (stor_device && stor_device->destroy)
603 		stor_device = NULL;
604 
605 	return stor_device;
606 }
607 
608 
609 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
610 {
611 	dev->drain_notify = true;
612 	wait_event(dev->waiting_to_drain,
613 		   atomic_read(&dev->num_outstanding_req) == 0);
614 	dev->drain_notify = false;
615 }
616 
617 static inline struct storvsc_device *get_in_stor_device(
618 					struct hv_device *device)
619 {
620 	struct storvsc_device *stor_device;
621 
622 	stor_device = hv_get_drvdata(device);
623 
624 	if (!stor_device)
625 		goto get_in_err;
626 
627 	/*
628 	 * If the device is being destroyed; allow incoming
629 	 * traffic only to cleanup outstanding requests.
630 	 */
631 
632 	if (stor_device->destroy  &&
633 		(atomic_read(&stor_device->num_outstanding_req) == 0))
634 		stor_device = NULL;
635 
636 get_in_err:
637 	return stor_device;
638 
639 }
640 
641 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
642 				      u32 new)
643 {
644 	struct storvsc_device *stor_device;
645 	struct vmbus_channel *cur_chn;
646 	bool old_is_alloced = false;
647 	struct hv_device *device;
648 	unsigned long flags;
649 	int cpu;
650 
651 	device = channel->primary_channel ?
652 			channel->primary_channel->device_obj
653 				: channel->device_obj;
654 	stor_device = get_out_stor_device(device);
655 	if (!stor_device)
656 		return;
657 
658 	/* See storvsc_do_io() -> get_og_chn(). */
659 	spin_lock_irqsave(&stor_device->lock, flags);
660 
661 	/*
662 	 * Determines if the storvsc device has other channels assigned to
663 	 * the "old" CPU to update the alloced_cpus mask and the stor_chns
664 	 * array.
665 	 */
666 	if (device->channel != channel && device->channel->target_cpu == old) {
667 		cur_chn = device->channel;
668 		old_is_alloced = true;
669 		goto old_is_alloced;
670 	}
671 	list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
672 		if (cur_chn == channel)
673 			continue;
674 		if (cur_chn->target_cpu == old) {
675 			old_is_alloced = true;
676 			goto old_is_alloced;
677 		}
678 	}
679 
680 old_is_alloced:
681 	if (old_is_alloced)
682 		WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
683 	else
684 		cpumask_clear_cpu(old, &stor_device->alloced_cpus);
685 
686 	/* "Flush" the stor_chns array. */
687 	for_each_possible_cpu(cpu) {
688 		if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
689 					cpu, &stor_device->alloced_cpus))
690 			WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
691 	}
692 
693 	WRITE_ONCE(stor_device->stor_chns[new], channel);
694 	cpumask_set_cpu(new, &stor_device->alloced_cpus);
695 
696 	spin_unlock_irqrestore(&stor_device->lock, flags);
697 }
698 
699 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
700 {
701 	struct storvsc_cmd_request *request =
702 		(struct storvsc_cmd_request *)(unsigned long)rqst_addr;
703 
704 	if (rqst_addr == VMBUS_RQST_INIT)
705 		return VMBUS_RQST_INIT;
706 	if (rqst_addr == VMBUS_RQST_RESET)
707 		return VMBUS_RQST_RESET;
708 
709 	/*
710 	 * Cannot return an ID of 0, which is reserved for an unsolicited
711 	 * message from Hyper-V.
712 	 */
713 	return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
714 }
715 
716 static void handle_sc_creation(struct vmbus_channel *new_sc)
717 {
718 	struct hv_device *device = new_sc->primary_channel->device_obj;
719 	struct device *dev = &device->device;
720 	struct storvsc_device *stor_device;
721 	struct vmstorage_channel_properties props;
722 	int ret;
723 
724 	stor_device = get_out_stor_device(device);
725 	if (!stor_device)
726 		return;
727 
728 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
729 	new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
730 
731 	new_sc->next_request_id_callback = storvsc_next_request_id;
732 
733 	ret = vmbus_open(new_sc,
734 			 storvsc_ringbuffer_size,
735 			 storvsc_ringbuffer_size,
736 			 (void *)&props,
737 			 sizeof(struct vmstorage_channel_properties),
738 			 storvsc_on_channel_callback, new_sc);
739 
740 	/* In case vmbus_open() fails, we don't use the sub-channel. */
741 	if (ret != 0) {
742 		dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
743 		return;
744 	}
745 
746 	new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
747 
748 	/* Add the sub-channel to the array of available channels. */
749 	stor_device->stor_chns[new_sc->target_cpu] = new_sc;
750 	cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
751 }
752 
753 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
754 {
755 	struct device *dev = &device->device;
756 	struct storvsc_device *stor_device;
757 	int num_sc;
758 	struct storvsc_cmd_request *request;
759 	struct vstor_packet *vstor_packet;
760 	int ret, t;
761 
762 	/*
763 	 * If the number of CPUs is artificially restricted, such as
764 	 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
765 	 * sub-channels >= the number of CPUs. These sub-channels
766 	 * should not be created. The primary channel is already created
767 	 * and assigned to one CPU, so check against # CPUs - 1.
768 	 */
769 	num_sc = min((int)(num_online_cpus() - 1), max_chns);
770 	if (!num_sc)
771 		return;
772 
773 	stor_device = get_out_stor_device(device);
774 	if (!stor_device)
775 		return;
776 
777 	stor_device->num_sc = num_sc;
778 	request = &stor_device->init_request;
779 	vstor_packet = &request->vstor_packet;
780 
781 	/*
782 	 * Establish a handler for dealing with subchannels.
783 	 */
784 	vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
785 
786 	/*
787 	 * Request the host to create sub-channels.
788 	 */
789 	memset(request, 0, sizeof(struct storvsc_cmd_request));
790 	init_completion(&request->wait_event);
791 	vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
792 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
793 	vstor_packet->sub_channel_count = num_sc;
794 
795 	ret = vmbus_sendpacket(device->channel, vstor_packet,
796 			       (sizeof(struct vstor_packet) -
797 			       stor_device->vmscsi_size_delta),
798 			       VMBUS_RQST_INIT,
799 			       VM_PKT_DATA_INBAND,
800 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
801 
802 	if (ret != 0) {
803 		dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
804 		return;
805 	}
806 
807 	t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
808 	if (t == 0) {
809 		dev_err(dev, "Failed to create sub-channel: timed out\n");
810 		return;
811 	}
812 
813 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
814 	    vstor_packet->status != 0) {
815 		dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
816 			vstor_packet->operation, vstor_packet->status);
817 		return;
818 	}
819 
820 	/*
821 	 * We need to do nothing here, because vmbus_process_offer()
822 	 * invokes channel->sc_creation_callback, which will open and use
823 	 * the sub-channel(s).
824 	 */
825 }
826 
827 static void cache_wwn(struct storvsc_device *stor_device,
828 		      struct vstor_packet *vstor_packet)
829 {
830 	/*
831 	 * Cache the currently active port and node ww names.
832 	 */
833 	if (vstor_packet->wwn_packet.primary_active) {
834 		stor_device->node_name =
835 			wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
836 		stor_device->port_name =
837 			wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
838 	} else {
839 		stor_device->node_name =
840 			wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
841 		stor_device->port_name =
842 			wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
843 	}
844 }
845 
846 
847 static int storvsc_execute_vstor_op(struct hv_device *device,
848 				    struct storvsc_cmd_request *request,
849 				    bool status_check)
850 {
851 	struct storvsc_device *stor_device;
852 	struct vstor_packet *vstor_packet;
853 	int ret, t;
854 
855 	stor_device = get_out_stor_device(device);
856 	if (!stor_device)
857 		return -ENODEV;
858 
859 	vstor_packet = &request->vstor_packet;
860 
861 	init_completion(&request->wait_event);
862 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
863 
864 	ret = vmbus_sendpacket(device->channel, vstor_packet,
865 			       (sizeof(struct vstor_packet) -
866 			       stor_device->vmscsi_size_delta),
867 			       VMBUS_RQST_INIT,
868 			       VM_PKT_DATA_INBAND,
869 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
870 	if (ret != 0)
871 		return ret;
872 
873 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
874 	if (t == 0)
875 		return -ETIMEDOUT;
876 
877 	if (!status_check)
878 		return ret;
879 
880 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
881 	    vstor_packet->status != 0)
882 		return -EINVAL;
883 
884 	return ret;
885 }
886 
887 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
888 {
889 	struct storvsc_device *stor_device;
890 	struct storvsc_cmd_request *request;
891 	struct vstor_packet *vstor_packet;
892 	int ret, i;
893 	int max_chns;
894 	bool process_sub_channels = false;
895 
896 	stor_device = get_out_stor_device(device);
897 	if (!stor_device)
898 		return -ENODEV;
899 
900 	request = &stor_device->init_request;
901 	vstor_packet = &request->vstor_packet;
902 
903 	/*
904 	 * Now, initiate the vsc/vsp initialization protocol on the open
905 	 * channel
906 	 */
907 	memset(request, 0, sizeof(struct storvsc_cmd_request));
908 	vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
909 	ret = storvsc_execute_vstor_op(device, request, true);
910 	if (ret)
911 		return ret;
912 	/*
913 	 * Query host supported protocol version.
914 	 */
915 
916 	for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) {
917 		/* reuse the packet for version range supported */
918 		memset(vstor_packet, 0, sizeof(struct vstor_packet));
919 		vstor_packet->operation =
920 			VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
921 
922 		vstor_packet->version.major_minor =
923 			vmstor_protocols[i].protocol_version;
924 
925 		/*
926 		 * The revision number is only used in Windows; set it to 0.
927 		 */
928 		vstor_packet->version.revision = 0;
929 		ret = storvsc_execute_vstor_op(device, request, false);
930 		if (ret != 0)
931 			return ret;
932 
933 		if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
934 			return -EINVAL;
935 
936 		if (vstor_packet->status == 0) {
937 			vmstor_proto_version =
938 				vmstor_protocols[i].protocol_version;
939 
940 			sense_buffer_size =
941 				vmstor_protocols[i].sense_buffer_size;
942 
943 			stor_device->vmscsi_size_delta =
944 				vmstor_protocols[i].vmscsi_size_delta;
945 
946 			break;
947 		}
948 	}
949 
950 	if (vstor_packet->status != 0)
951 		return -EINVAL;
952 
953 
954 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
955 	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
956 	ret = storvsc_execute_vstor_op(device, request, true);
957 	if (ret != 0)
958 		return ret;
959 
960 	/*
961 	 * Check to see if multi-channel support is there.
962 	 * Hosts that implement protocol version of 5.1 and above
963 	 * support multi-channel.
964 	 */
965 	max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
966 
967 	/*
968 	 * Allocate state to manage the sub-channels.
969 	 * We allocate an array based on the numbers of possible CPUs
970 	 * (Hyper-V does not support cpu online/offline).
971 	 * This Array will be sparseley populated with unique
972 	 * channels - primary + sub-channels.
973 	 * We will however populate all the slots to evenly distribute
974 	 * the load.
975 	 */
976 	stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
977 					 GFP_KERNEL);
978 	if (stor_device->stor_chns == NULL)
979 		return -ENOMEM;
980 
981 	device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
982 
983 	stor_device->stor_chns[device->channel->target_cpu] = device->channel;
984 	cpumask_set_cpu(device->channel->target_cpu,
985 			&stor_device->alloced_cpus);
986 
987 	if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) {
988 		if (vstor_packet->storage_channel_properties.flags &
989 		    STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
990 			process_sub_channels = true;
991 	}
992 	stor_device->max_transfer_bytes =
993 		vstor_packet->storage_channel_properties.max_transfer_bytes;
994 
995 	if (!is_fc)
996 		goto done;
997 
998 	/*
999 	 * For FC devices retrieve FC HBA data.
1000 	 */
1001 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1002 	vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
1003 	ret = storvsc_execute_vstor_op(device, request, true);
1004 	if (ret != 0)
1005 		return ret;
1006 
1007 	/*
1008 	 * Cache the currently active port and node ww names.
1009 	 */
1010 	cache_wwn(stor_device, vstor_packet);
1011 
1012 done:
1013 
1014 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1015 	vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
1016 	ret = storvsc_execute_vstor_op(device, request, true);
1017 	if (ret != 0)
1018 		return ret;
1019 
1020 	if (process_sub_channels)
1021 		handle_multichannel_storage(device, max_chns);
1022 
1023 	return ret;
1024 }
1025 
1026 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
1027 				struct scsi_cmnd *scmnd,
1028 				struct Scsi_Host *host,
1029 				u8 asc, u8 ascq)
1030 {
1031 	struct storvsc_scan_work *wrk;
1032 	void (*process_err_fn)(struct work_struct *work);
1033 	struct hv_host_device *host_dev = shost_priv(host);
1034 
1035 	/*
1036 	 * In some situations, Hyper-V sets multiple bits in the
1037 	 * srb_status, such as ABORTED and ERROR. So process them
1038 	 * individually, with the most specific bits first.
1039 	 */
1040 
1041 	if (vm_srb->srb_status & SRB_STATUS_INVALID_LUN) {
1042 		set_host_byte(scmnd, DID_NO_CONNECT);
1043 		process_err_fn = storvsc_remove_lun;
1044 		goto do_work;
1045 	}
1046 
1047 	if (vm_srb->srb_status & SRB_STATUS_ABORTED) {
1048 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
1049 		    /* Capacity data has changed */
1050 		    (asc == 0x2a) && (ascq == 0x9)) {
1051 			process_err_fn = storvsc_device_scan;
1052 			/*
1053 			 * Retry the I/O that triggered this.
1054 			 */
1055 			set_host_byte(scmnd, DID_REQUEUE);
1056 			goto do_work;
1057 		}
1058 	}
1059 
1060 	if (vm_srb->srb_status & SRB_STATUS_ERROR) {
1061 		/*
1062 		 * Let upper layer deal with error when
1063 		 * sense message is present.
1064 		 */
1065 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
1066 			return;
1067 
1068 		/*
1069 		 * If there is an error; offline the device since all
1070 		 * error recovery strategies would have already been
1071 		 * deployed on the host side. However, if the command
1072 		 * were a pass-through command deal with it appropriately.
1073 		 */
1074 		switch (scmnd->cmnd[0]) {
1075 		case ATA_16:
1076 		case ATA_12:
1077 			set_host_byte(scmnd, DID_PASSTHROUGH);
1078 			break;
1079 		/*
1080 		 * On some Hyper-V hosts TEST_UNIT_READY command can
1081 		 * return SRB_STATUS_ERROR. Let the upper level code
1082 		 * deal with it based on the sense information.
1083 		 */
1084 		case TEST_UNIT_READY:
1085 			break;
1086 		default:
1087 			set_host_byte(scmnd, DID_ERROR);
1088 		}
1089 	}
1090 	return;
1091 
1092 do_work:
1093 	/*
1094 	 * We need to schedule work to process this error; schedule it.
1095 	 */
1096 	wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1097 	if (!wrk) {
1098 		set_host_byte(scmnd, DID_TARGET_FAILURE);
1099 		return;
1100 	}
1101 
1102 	wrk->host = host;
1103 	wrk->lun = vm_srb->lun;
1104 	wrk->tgt_id = vm_srb->target_id;
1105 	INIT_WORK(&wrk->work, process_err_fn);
1106 	queue_work(host_dev->handle_error_wq, &wrk->work);
1107 }
1108 
1109 
1110 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1111 				       struct storvsc_device *stor_dev)
1112 {
1113 	struct scsi_cmnd *scmnd = cmd_request->cmd;
1114 	struct scsi_sense_hdr sense_hdr;
1115 	struct vmscsi_request *vm_srb;
1116 	u32 data_transfer_length;
1117 	struct Scsi_Host *host;
1118 	u32 payload_sz = cmd_request->payload_sz;
1119 	void *payload = cmd_request->payload;
1120 	bool sense_ok;
1121 
1122 	host = stor_dev->host;
1123 
1124 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1125 	data_transfer_length = vm_srb->data_transfer_length;
1126 
1127 	scmnd->result = vm_srb->scsi_status;
1128 
1129 	if (scmnd->result) {
1130 		sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1131 				SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1132 
1133 		if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1134 			scsi_print_sense_hdr(scmnd->device, "storvsc",
1135 					     &sense_hdr);
1136 	}
1137 
1138 	if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1139 		storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1140 					 sense_hdr.ascq);
1141 		/*
1142 		 * The Windows driver set data_transfer_length on
1143 		 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1144 		 * is untouched.  In these cases we set it to 0.
1145 		 */
1146 		if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1147 			data_transfer_length = 0;
1148 	}
1149 
1150 	/* Validate data_transfer_length (from Hyper-V) */
1151 	if (data_transfer_length > cmd_request->payload->range.len)
1152 		data_transfer_length = cmd_request->payload->range.len;
1153 
1154 	scsi_set_resid(scmnd,
1155 		cmd_request->payload->range.len - data_transfer_length);
1156 
1157 	scsi_done(scmnd);
1158 
1159 	if (payload_sz >
1160 		sizeof(struct vmbus_channel_packet_multipage_buffer))
1161 		kfree(payload);
1162 }
1163 
1164 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1165 				  struct vstor_packet *vstor_packet,
1166 				  struct storvsc_cmd_request *request)
1167 {
1168 	struct vstor_packet *stor_pkt;
1169 	struct hv_device *device = stor_device->device;
1170 
1171 	stor_pkt = &request->vstor_packet;
1172 
1173 	/*
1174 	 * The current SCSI handling on the host side does
1175 	 * not correctly handle:
1176 	 * INQUIRY command with page code parameter set to 0x80
1177 	 * MODE_SENSE command with cmd[2] == 0x1c
1178 	 *
1179 	 * Setup srb and scsi status so this won't be fatal.
1180 	 * We do this so we can distinguish truly fatal failues
1181 	 * (srb status == 0x4) and off-line the device in that case.
1182 	 */
1183 
1184 	if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1185 	   (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1186 		vstor_packet->vm_srb.scsi_status = 0;
1187 		vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1188 	}
1189 
1190 	/* Copy over the status...etc */
1191 	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1192 	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1193 
1194 	/*
1195 	 * Copy over the sense_info_length, but limit to the known max
1196 	 * size if Hyper-V returns a bad value.
1197 	 */
1198 	stor_pkt->vm_srb.sense_info_length = min_t(u8, sense_buffer_size,
1199 		vstor_packet->vm_srb.sense_info_length);
1200 
1201 	if (vstor_packet->vm_srb.scsi_status != 0 ||
1202 	    vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1203 
1204 		/*
1205 		 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1206 		 * return errors when detecting devices using TEST_UNIT_READY,
1207 		 * and logging these as errors produces unhelpful noise.
1208 		 */
1209 		int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1210 			STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1211 
1212 		storvsc_log(device, loglevel,
1213 			"tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1214 			scsi_cmd_to_rq(request->cmd)->tag,
1215 			stor_pkt->vm_srb.cdb[0],
1216 			vstor_packet->vm_srb.scsi_status,
1217 			vstor_packet->vm_srb.srb_status,
1218 			vstor_packet->status);
1219 	}
1220 
1221 	if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1222 	    (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1223 		memcpy(request->cmd->sense_buffer,
1224 		       vstor_packet->vm_srb.sense_data,
1225 		       stor_pkt->vm_srb.sense_info_length);
1226 
1227 	stor_pkt->vm_srb.data_transfer_length =
1228 		vstor_packet->vm_srb.data_transfer_length;
1229 
1230 	storvsc_command_completion(request, stor_device);
1231 
1232 	if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1233 		stor_device->drain_notify)
1234 		wake_up(&stor_device->waiting_to_drain);
1235 }
1236 
1237 static void storvsc_on_receive(struct storvsc_device *stor_device,
1238 			     struct vstor_packet *vstor_packet,
1239 			     struct storvsc_cmd_request *request)
1240 {
1241 	struct hv_host_device *host_dev;
1242 	switch (vstor_packet->operation) {
1243 	case VSTOR_OPERATION_COMPLETE_IO:
1244 		storvsc_on_io_completion(stor_device, vstor_packet, request);
1245 		break;
1246 
1247 	case VSTOR_OPERATION_REMOVE_DEVICE:
1248 	case VSTOR_OPERATION_ENUMERATE_BUS:
1249 		host_dev = shost_priv(stor_device->host);
1250 		queue_work(
1251 			host_dev->handle_error_wq, &host_dev->host_scan_work);
1252 		break;
1253 
1254 	case VSTOR_OPERATION_FCHBA_DATA:
1255 		cache_wwn(stor_device, vstor_packet);
1256 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1257 		fc_host_node_name(stor_device->host) = stor_device->node_name;
1258 		fc_host_port_name(stor_device->host) = stor_device->port_name;
1259 #endif
1260 		break;
1261 	default:
1262 		break;
1263 	}
1264 }
1265 
1266 static void storvsc_on_channel_callback(void *context)
1267 {
1268 	struct vmbus_channel *channel = (struct vmbus_channel *)context;
1269 	const struct vmpacket_descriptor *desc;
1270 	struct hv_device *device;
1271 	struct storvsc_device *stor_device;
1272 	struct Scsi_Host *shost;
1273 
1274 	if (channel->primary_channel != NULL)
1275 		device = channel->primary_channel->device_obj;
1276 	else
1277 		device = channel->device_obj;
1278 
1279 	stor_device = get_in_stor_device(device);
1280 	if (!stor_device)
1281 		return;
1282 
1283 	shost = stor_device->host;
1284 
1285 	foreach_vmbus_pkt(desc, channel) {
1286 		struct vstor_packet *packet = hv_pkt_data(desc);
1287 		struct storvsc_cmd_request *request = NULL;
1288 		u32 pktlen = hv_pkt_datalen(desc);
1289 		u64 rqst_id = desc->trans_id;
1290 		u32 minlen = rqst_id ? sizeof(struct vstor_packet) -
1291 			stor_device->vmscsi_size_delta : sizeof(enum vstor_packet_operation);
1292 
1293 		if (pktlen < minlen) {
1294 			dev_err(&device->device,
1295 				"Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1296 				rqst_id, pktlen, minlen);
1297 			continue;
1298 		}
1299 
1300 		if (rqst_id == VMBUS_RQST_INIT) {
1301 			request = &stor_device->init_request;
1302 		} else if (rqst_id == VMBUS_RQST_RESET) {
1303 			request = &stor_device->reset_request;
1304 		} else {
1305 			/* Hyper-V can send an unsolicited message with ID of 0 */
1306 			if (rqst_id == 0) {
1307 				/*
1308 				 * storvsc_on_receive() looks at the vstor_packet in the message
1309 				 * from the ring buffer.
1310 				 *
1311 				 * - If the operation in the vstor_packet is COMPLETE_IO, then
1312 				 *   we call storvsc_on_io_completion(), and dereference the
1313 				 *   guest memory address.  Make sure we don't call
1314 				 *   storvsc_on_io_completion() with a guest memory address
1315 				 *   that is zero if Hyper-V were to construct and send such
1316 				 *   a bogus packet.
1317 				 *
1318 				 * - If the operation in the vstor_packet is FCHBA_DATA, then
1319 				 *   we call cache_wwn(), and access the data payload area of
1320 				 *   the packet (wwn_packet); however, there is no guarantee
1321 				 *   that the packet is big enough to contain such area.
1322 				 *   Future-proof the code by rejecting such a bogus packet.
1323 				 */
1324 				if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1325 				    packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1326 					dev_err(&device->device, "Invalid packet with ID of 0\n");
1327 					continue;
1328 				}
1329 			} else {
1330 				struct scsi_cmnd *scmnd;
1331 
1332 				/* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1333 				scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1334 				if (scmnd == NULL) {
1335 					dev_err(&device->device, "Incorrect transaction ID\n");
1336 					continue;
1337 				}
1338 				request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1339 			}
1340 
1341 			storvsc_on_receive(stor_device, packet, request);
1342 			continue;
1343 		}
1344 
1345 		memcpy(&request->vstor_packet, packet,
1346 		       (sizeof(struct vstor_packet) - stor_device->vmscsi_size_delta));
1347 		complete(&request->wait_event);
1348 	}
1349 }
1350 
1351 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1352 				  bool is_fc)
1353 {
1354 	struct vmstorage_channel_properties props;
1355 	int ret;
1356 
1357 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1358 
1359 	device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1360 	device->channel->next_request_id_callback = storvsc_next_request_id;
1361 
1362 	ret = vmbus_open(device->channel,
1363 			 ring_size,
1364 			 ring_size,
1365 			 (void *)&props,
1366 			 sizeof(struct vmstorage_channel_properties),
1367 			 storvsc_on_channel_callback, device->channel);
1368 
1369 	if (ret != 0)
1370 		return ret;
1371 
1372 	ret = storvsc_channel_init(device, is_fc);
1373 
1374 	return ret;
1375 }
1376 
1377 static int storvsc_dev_remove(struct hv_device *device)
1378 {
1379 	struct storvsc_device *stor_device;
1380 
1381 	stor_device = hv_get_drvdata(device);
1382 
1383 	stor_device->destroy = true;
1384 
1385 	/* Make sure flag is set before waiting */
1386 	wmb();
1387 
1388 	/*
1389 	 * At this point, all outbound traffic should be disable. We
1390 	 * only allow inbound traffic (responses) to proceed so that
1391 	 * outstanding requests can be completed.
1392 	 */
1393 
1394 	storvsc_wait_to_drain(stor_device);
1395 
1396 	/*
1397 	 * Since we have already drained, we don't need to busy wait
1398 	 * as was done in final_release_stor_device()
1399 	 * Note that we cannot set the ext pointer to NULL until
1400 	 * we have drained - to drain the outgoing packets, we need to
1401 	 * allow incoming packets.
1402 	 */
1403 	hv_set_drvdata(device, NULL);
1404 
1405 	/* Close the channel */
1406 	vmbus_close(device->channel);
1407 
1408 	kfree(stor_device->stor_chns);
1409 	kfree(stor_device);
1410 	return 0;
1411 }
1412 
1413 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1414 					u16 q_num)
1415 {
1416 	u16 slot = 0;
1417 	u16 hash_qnum;
1418 	const struct cpumask *node_mask;
1419 	int num_channels, tgt_cpu;
1420 
1421 	if (stor_device->num_sc == 0) {
1422 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1423 		return stor_device->device->channel;
1424 	}
1425 
1426 	/*
1427 	 * Our channel array is sparsley populated and we
1428 	 * initiated I/O on a processor/hw-q that does not
1429 	 * currently have a designated channel. Fix this.
1430 	 * The strategy is simple:
1431 	 * I. Ensure NUMA locality
1432 	 * II. Distribute evenly (best effort)
1433 	 */
1434 
1435 	node_mask = cpumask_of_node(cpu_to_node(q_num));
1436 
1437 	num_channels = 0;
1438 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1439 		if (cpumask_test_cpu(tgt_cpu, node_mask))
1440 			num_channels++;
1441 	}
1442 	if (num_channels == 0) {
1443 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1444 		return stor_device->device->channel;
1445 	}
1446 
1447 	hash_qnum = q_num;
1448 	while (hash_qnum >= num_channels)
1449 		hash_qnum -= num_channels;
1450 
1451 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1452 		if (!cpumask_test_cpu(tgt_cpu, node_mask))
1453 			continue;
1454 		if (slot == hash_qnum)
1455 			break;
1456 		slot++;
1457 	}
1458 
1459 	stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1460 
1461 	return stor_device->stor_chns[q_num];
1462 }
1463 
1464 
1465 static int storvsc_do_io(struct hv_device *device,
1466 			 struct storvsc_cmd_request *request, u16 q_num)
1467 {
1468 	struct storvsc_device *stor_device;
1469 	struct vstor_packet *vstor_packet;
1470 	struct vmbus_channel *outgoing_channel, *channel;
1471 	unsigned long flags;
1472 	int ret = 0;
1473 	const struct cpumask *node_mask;
1474 	int tgt_cpu;
1475 
1476 	vstor_packet = &request->vstor_packet;
1477 	stor_device = get_out_stor_device(device);
1478 
1479 	if (!stor_device)
1480 		return -ENODEV;
1481 
1482 
1483 	request->device  = device;
1484 	/*
1485 	 * Select an appropriate channel to send the request out.
1486 	 */
1487 	/* See storvsc_change_target_cpu(). */
1488 	outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1489 	if (outgoing_channel != NULL) {
1490 		if (outgoing_channel->target_cpu == q_num) {
1491 			/*
1492 			 * Ideally, we want to pick a different channel if
1493 			 * available on the same NUMA node.
1494 			 */
1495 			node_mask = cpumask_of_node(cpu_to_node(q_num));
1496 			for_each_cpu_wrap(tgt_cpu,
1497 				 &stor_device->alloced_cpus, q_num + 1) {
1498 				if (!cpumask_test_cpu(tgt_cpu, node_mask))
1499 					continue;
1500 				if (tgt_cpu == q_num)
1501 					continue;
1502 				channel = READ_ONCE(
1503 					stor_device->stor_chns[tgt_cpu]);
1504 				if (channel == NULL)
1505 					continue;
1506 				if (hv_get_avail_to_write_percent(
1507 							&channel->outbound)
1508 						> ring_avail_percent_lowater) {
1509 					outgoing_channel = channel;
1510 					goto found_channel;
1511 				}
1512 			}
1513 
1514 			/*
1515 			 * All the other channels on the same NUMA node are
1516 			 * busy. Try to use the channel on the current CPU
1517 			 */
1518 			if (hv_get_avail_to_write_percent(
1519 						&outgoing_channel->outbound)
1520 					> ring_avail_percent_lowater)
1521 				goto found_channel;
1522 
1523 			/*
1524 			 * If we reach here, all the channels on the current
1525 			 * NUMA node are busy. Try to find a channel in
1526 			 * other NUMA nodes
1527 			 */
1528 			for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1529 				if (cpumask_test_cpu(tgt_cpu, node_mask))
1530 					continue;
1531 				channel = READ_ONCE(
1532 					stor_device->stor_chns[tgt_cpu]);
1533 				if (channel == NULL)
1534 					continue;
1535 				if (hv_get_avail_to_write_percent(
1536 							&channel->outbound)
1537 						> ring_avail_percent_lowater) {
1538 					outgoing_channel = channel;
1539 					goto found_channel;
1540 				}
1541 			}
1542 		}
1543 	} else {
1544 		spin_lock_irqsave(&stor_device->lock, flags);
1545 		outgoing_channel = stor_device->stor_chns[q_num];
1546 		if (outgoing_channel != NULL) {
1547 			spin_unlock_irqrestore(&stor_device->lock, flags);
1548 			goto found_channel;
1549 		}
1550 		outgoing_channel = get_og_chn(stor_device, q_num);
1551 		spin_unlock_irqrestore(&stor_device->lock, flags);
1552 	}
1553 
1554 found_channel:
1555 	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1556 
1557 	vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1558 					stor_device->vmscsi_size_delta);
1559 
1560 
1561 	vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1562 
1563 
1564 	vstor_packet->vm_srb.data_transfer_length =
1565 	request->payload->range.len;
1566 
1567 	vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1568 
1569 	if (request->payload->range.len) {
1570 
1571 		ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1572 				request->payload, request->payload_sz,
1573 				vstor_packet,
1574 				(sizeof(struct vstor_packet) -
1575 				stor_device->vmscsi_size_delta),
1576 				(unsigned long)request);
1577 	} else {
1578 		ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1579 			       (sizeof(struct vstor_packet) -
1580 				stor_device->vmscsi_size_delta),
1581 			       (unsigned long)request,
1582 			       VM_PKT_DATA_INBAND,
1583 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1584 	}
1585 
1586 	if (ret != 0)
1587 		return ret;
1588 
1589 	atomic_inc(&stor_device->num_outstanding_req);
1590 
1591 	return ret;
1592 }
1593 
1594 static int storvsc_device_alloc(struct scsi_device *sdevice)
1595 {
1596 	/*
1597 	 * Set blist flag to permit the reading of the VPD pages even when
1598 	 * the target may claim SPC-2 compliance. MSFT targets currently
1599 	 * claim SPC-2 compliance while they implement post SPC-2 features.
1600 	 * With this flag we can correctly handle WRITE_SAME_16 issues.
1601 	 *
1602 	 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1603 	 * still supports REPORT LUN.
1604 	 */
1605 	sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1606 
1607 	return 0;
1608 }
1609 
1610 static int storvsc_device_configure(struct scsi_device *sdevice)
1611 {
1612 	blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1613 
1614 	sdevice->no_write_same = 1;
1615 
1616 	/*
1617 	 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1618 	 * if the device is a MSFT virtual device.  If the host is
1619 	 * WIN10 or newer, allow write_same.
1620 	 */
1621 	if (!strncmp(sdevice->vendor, "Msft", 4)) {
1622 		switch (vmstor_proto_version) {
1623 		case VMSTOR_PROTO_VERSION_WIN8:
1624 		case VMSTOR_PROTO_VERSION_WIN8_1:
1625 			sdevice->scsi_level = SCSI_SPC_3;
1626 			break;
1627 		}
1628 
1629 		if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1630 			sdevice->no_write_same = 0;
1631 	}
1632 
1633 	return 0;
1634 }
1635 
1636 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1637 			   sector_t capacity, int *info)
1638 {
1639 	sector_t nsect = capacity;
1640 	sector_t cylinders = nsect;
1641 	int heads, sectors_pt;
1642 
1643 	/*
1644 	 * We are making up these values; let us keep it simple.
1645 	 */
1646 	heads = 0xff;
1647 	sectors_pt = 0x3f;      /* Sectors per track */
1648 	sector_div(cylinders, heads * sectors_pt);
1649 	if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1650 		cylinders = 0xffff;
1651 
1652 	info[0] = heads;
1653 	info[1] = sectors_pt;
1654 	info[2] = (int)cylinders;
1655 
1656 	return 0;
1657 }
1658 
1659 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1660 {
1661 	struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1662 	struct hv_device *device = host_dev->dev;
1663 
1664 	struct storvsc_device *stor_device;
1665 	struct storvsc_cmd_request *request;
1666 	struct vstor_packet *vstor_packet;
1667 	int ret, t;
1668 
1669 	stor_device = get_out_stor_device(device);
1670 	if (!stor_device)
1671 		return FAILED;
1672 
1673 	request = &stor_device->reset_request;
1674 	vstor_packet = &request->vstor_packet;
1675 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1676 
1677 	init_completion(&request->wait_event);
1678 
1679 	vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1680 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1681 	vstor_packet->vm_srb.path_id = stor_device->path_id;
1682 
1683 	ret = vmbus_sendpacket(device->channel, vstor_packet,
1684 			       (sizeof(struct vstor_packet) -
1685 				stor_device->vmscsi_size_delta),
1686 			       VMBUS_RQST_RESET,
1687 			       VM_PKT_DATA_INBAND,
1688 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1689 	if (ret != 0)
1690 		return FAILED;
1691 
1692 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1693 	if (t == 0)
1694 		return TIMEOUT_ERROR;
1695 
1696 
1697 	/*
1698 	 * At this point, all outstanding requests in the adapter
1699 	 * should have been flushed out and return to us
1700 	 * There is a potential race here where the host may be in
1701 	 * the process of responding when we return from here.
1702 	 * Just wait for all in-transit packets to be accounted for
1703 	 * before we return from here.
1704 	 */
1705 	storvsc_wait_to_drain(stor_device);
1706 
1707 	return SUCCESS;
1708 }
1709 
1710 /*
1711  * The host guarantees to respond to each command, although I/O latencies might
1712  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1713  * chance to perform EH.
1714  */
1715 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1716 {
1717 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1718 	if (scmnd->device->host->transportt == fc_transport_template)
1719 		return fc_eh_timed_out(scmnd);
1720 #endif
1721 	return BLK_EH_RESET_TIMER;
1722 }
1723 
1724 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1725 {
1726 	bool allowed = true;
1727 	u8 scsi_op = scmnd->cmnd[0];
1728 
1729 	switch (scsi_op) {
1730 	/* the host does not handle WRITE_SAME, log accident usage */
1731 	case WRITE_SAME:
1732 	/*
1733 	 * smartd sends this command and the host does not handle
1734 	 * this. So, don't send it.
1735 	 */
1736 	case SET_WINDOW:
1737 		set_host_byte(scmnd, DID_ERROR);
1738 		allowed = false;
1739 		break;
1740 	default:
1741 		break;
1742 	}
1743 	return allowed;
1744 }
1745 
1746 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1747 {
1748 	int ret;
1749 	struct hv_host_device *host_dev = shost_priv(host);
1750 	struct hv_device *dev = host_dev->dev;
1751 	struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1752 	int i;
1753 	struct scatterlist *sgl;
1754 	unsigned int sg_count;
1755 	struct vmscsi_request *vm_srb;
1756 	struct vmbus_packet_mpb_array  *payload;
1757 	u32 payload_sz;
1758 	u32 length;
1759 
1760 	if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1761 		/*
1762 		 * On legacy hosts filter unimplemented commands.
1763 		 * Future hosts are expected to correctly handle
1764 		 * unsupported commands. Furthermore, it is
1765 		 * possible that some of the currently
1766 		 * unsupported commands maybe supported in
1767 		 * future versions of the host.
1768 		 */
1769 		if (!storvsc_scsi_cmd_ok(scmnd)) {
1770 			scsi_done(scmnd);
1771 			return 0;
1772 		}
1773 	}
1774 
1775 	/* Setup the cmd request */
1776 	cmd_request->cmd = scmnd;
1777 
1778 	memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1779 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1780 	vm_srb->win8_extension.time_out_value = 60;
1781 
1782 	vm_srb->win8_extension.srb_flags |=
1783 		SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1784 
1785 	if (scmnd->device->tagged_supported) {
1786 		vm_srb->win8_extension.srb_flags |=
1787 		(SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1788 		vm_srb->win8_extension.queue_tag = SP_UNTAGGED;
1789 		vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST;
1790 	}
1791 
1792 	/* Build the SRB */
1793 	switch (scmnd->sc_data_direction) {
1794 	case DMA_TO_DEVICE:
1795 		vm_srb->data_in = WRITE_TYPE;
1796 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1797 		break;
1798 	case DMA_FROM_DEVICE:
1799 		vm_srb->data_in = READ_TYPE;
1800 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1801 		break;
1802 	case DMA_NONE:
1803 		vm_srb->data_in = UNKNOWN_TYPE;
1804 		vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1805 		break;
1806 	default:
1807 		/*
1808 		 * This is DMA_BIDIRECTIONAL or something else we are never
1809 		 * supposed to see here.
1810 		 */
1811 		WARN(1, "Unexpected data direction: %d\n",
1812 		     scmnd->sc_data_direction);
1813 		return -EINVAL;
1814 	}
1815 
1816 
1817 	vm_srb->port_number = host_dev->port;
1818 	vm_srb->path_id = scmnd->device->channel;
1819 	vm_srb->target_id = scmnd->device->id;
1820 	vm_srb->lun = scmnd->device->lun;
1821 
1822 	vm_srb->cdb_length = scmnd->cmd_len;
1823 
1824 	memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1825 
1826 	sgl = (struct scatterlist *)scsi_sglist(scmnd);
1827 	sg_count = scsi_sg_count(scmnd);
1828 
1829 	length = scsi_bufflen(scmnd);
1830 	payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1831 	payload_sz = sizeof(cmd_request->mpb);
1832 
1833 	if (sg_count) {
1834 		unsigned int hvpgoff, hvpfns_to_add;
1835 		unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1836 		unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1837 		u64 hvpfn;
1838 
1839 		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1840 
1841 			payload_sz = (hvpg_count * sizeof(u64) +
1842 				      sizeof(struct vmbus_packet_mpb_array));
1843 			payload = kzalloc(payload_sz, GFP_ATOMIC);
1844 			if (!payload)
1845 				return SCSI_MLQUEUE_DEVICE_BUSY;
1846 		}
1847 
1848 		payload->range.len = length;
1849 		payload->range.offset = offset_in_hvpg;
1850 
1851 
1852 		for (i = 0; sgl != NULL; sgl = sg_next(sgl)) {
1853 			/*
1854 			 * Init values for the current sgl entry. hvpgoff
1855 			 * and hvpfns_to_add are in units of Hyper-V size
1856 			 * pages. Handling the PAGE_SIZE != HV_HYP_PAGE_SIZE
1857 			 * case also handles values of sgl->offset that are
1858 			 * larger than PAGE_SIZE. Such offsets are handled
1859 			 * even on other than the first sgl entry, provided
1860 			 * they are a multiple of PAGE_SIZE.
1861 			 */
1862 			hvpgoff = HVPFN_DOWN(sgl->offset);
1863 			hvpfn = page_to_hvpfn(sg_page(sgl)) + hvpgoff;
1864 			hvpfns_to_add =	HVPFN_UP(sgl->offset + sgl->length) -
1865 						hvpgoff;
1866 
1867 			/*
1868 			 * Fill the next portion of the PFN array with
1869 			 * sequential Hyper-V PFNs for the continguous physical
1870 			 * memory described by the sgl entry. The end of the
1871 			 * last sgl should be reached at the same time that
1872 			 * the PFN array is filled.
1873 			 */
1874 			while (hvpfns_to_add--)
1875 				payload->range.pfn_array[i++] =	hvpfn++;
1876 		}
1877 	}
1878 
1879 	cmd_request->payload = payload;
1880 	cmd_request->payload_sz = payload_sz;
1881 
1882 	/* Invokes the vsc to start an IO */
1883 	ret = storvsc_do_io(dev, cmd_request, get_cpu());
1884 	put_cpu();
1885 
1886 	if (ret == -EAGAIN) {
1887 		if (payload_sz > sizeof(cmd_request->mpb))
1888 			kfree(payload);
1889 		/* no more space */
1890 		return SCSI_MLQUEUE_DEVICE_BUSY;
1891 	}
1892 
1893 	return 0;
1894 }
1895 
1896 static struct scsi_host_template scsi_driver = {
1897 	.module	=		THIS_MODULE,
1898 	.name =			"storvsc_host_t",
1899 	.cmd_size =             sizeof(struct storvsc_cmd_request),
1900 	.bios_param =		storvsc_get_chs,
1901 	.queuecommand =		storvsc_queuecommand,
1902 	.eh_host_reset_handler =	storvsc_host_reset_handler,
1903 	.proc_name =		"storvsc_host",
1904 	.eh_timed_out =		storvsc_eh_timed_out,
1905 	.slave_alloc =		storvsc_device_alloc,
1906 	.slave_configure =	storvsc_device_configure,
1907 	.cmd_per_lun =		2048,
1908 	.this_id =		-1,
1909 	/* Ensure there are no gaps in presented sgls */
1910 	.virt_boundary_mask =	PAGE_SIZE-1,
1911 	.no_write_same =	1,
1912 	.track_queue_depth =	1,
1913 	.change_queue_depth =	storvsc_change_queue_depth,
1914 };
1915 
1916 enum {
1917 	SCSI_GUID,
1918 	IDE_GUID,
1919 	SFC_GUID,
1920 };
1921 
1922 static const struct hv_vmbus_device_id id_table[] = {
1923 	/* SCSI guid */
1924 	{ HV_SCSI_GUID,
1925 	  .driver_data = SCSI_GUID
1926 	},
1927 	/* IDE guid */
1928 	{ HV_IDE_GUID,
1929 	  .driver_data = IDE_GUID
1930 	},
1931 	/* Fibre Channel GUID */
1932 	{
1933 	  HV_SYNTHFC_GUID,
1934 	  .driver_data = SFC_GUID
1935 	},
1936 	{ },
1937 };
1938 
1939 MODULE_DEVICE_TABLE(vmbus, id_table);
1940 
1941 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1942 
1943 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1944 {
1945 	return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1946 }
1947 
1948 static int storvsc_probe(struct hv_device *device,
1949 			const struct hv_vmbus_device_id *dev_id)
1950 {
1951 	int ret;
1952 	int num_cpus = num_online_cpus();
1953 	int num_present_cpus = num_present_cpus();
1954 	struct Scsi_Host *host;
1955 	struct hv_host_device *host_dev;
1956 	bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1957 	bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1958 	int target = 0;
1959 	struct storvsc_device *stor_device;
1960 	int max_luns_per_target;
1961 	int max_targets;
1962 	int max_channels;
1963 	int max_sub_channels = 0;
1964 
1965 	/*
1966 	 * Based on the windows host we are running on,
1967 	 * set state to properly communicate with the host.
1968 	 */
1969 
1970 	if (vmbus_proto_version < VERSION_WIN8) {
1971 		max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1972 		max_targets = STORVSC_IDE_MAX_TARGETS;
1973 		max_channels = STORVSC_IDE_MAX_CHANNELS;
1974 	} else {
1975 		max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
1976 		max_targets = STORVSC_MAX_TARGETS;
1977 		max_channels = STORVSC_MAX_CHANNELS;
1978 		/*
1979 		 * On Windows8 and above, we support sub-channels for storage
1980 		 * on SCSI and FC controllers.
1981 		 * The number of sub-channels offerred is based on the number of
1982 		 * VCPUs in the guest.
1983 		 */
1984 		if (!dev_is_ide)
1985 			max_sub_channels =
1986 				(num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1987 	}
1988 
1989 	scsi_driver.can_queue = max_outstanding_req_per_channel *
1990 				(max_sub_channels + 1) *
1991 				(100 - ring_avail_percent_lowater) / 100;
1992 
1993 	host = scsi_host_alloc(&scsi_driver,
1994 			       sizeof(struct hv_host_device));
1995 	if (!host)
1996 		return -ENOMEM;
1997 
1998 	host_dev = shost_priv(host);
1999 	memset(host_dev, 0, sizeof(struct hv_host_device));
2000 
2001 	host_dev->port = host->host_no;
2002 	host_dev->dev = device;
2003 	host_dev->host = host;
2004 
2005 
2006 	stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
2007 	if (!stor_device) {
2008 		ret = -ENOMEM;
2009 		goto err_out0;
2010 	}
2011 
2012 	stor_device->destroy = false;
2013 	init_waitqueue_head(&stor_device->waiting_to_drain);
2014 	stor_device->device = device;
2015 	stor_device->host = host;
2016 	stor_device->vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
2017 	spin_lock_init(&stor_device->lock);
2018 	hv_set_drvdata(device, stor_device);
2019 
2020 	stor_device->port_number = host->host_no;
2021 	ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
2022 	if (ret)
2023 		goto err_out1;
2024 
2025 	host_dev->path = stor_device->path_id;
2026 	host_dev->target = stor_device->target_id;
2027 
2028 	switch (dev_id->driver_data) {
2029 	case SFC_GUID:
2030 		host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
2031 		host->max_id = STORVSC_FC_MAX_TARGETS;
2032 		host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
2033 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2034 		host->transportt = fc_transport_template;
2035 #endif
2036 		break;
2037 
2038 	case SCSI_GUID:
2039 		host->max_lun = max_luns_per_target;
2040 		host->max_id = max_targets;
2041 		host->max_channel = max_channels - 1;
2042 		break;
2043 
2044 	default:
2045 		host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
2046 		host->max_id = STORVSC_IDE_MAX_TARGETS;
2047 		host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2048 		break;
2049 	}
2050 	/* max cmd length */
2051 	host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2052 
2053 	/*
2054 	 * set the table size based on the info we got
2055 	 * from the host.
2056 	 */
2057 	host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
2058 	/*
2059 	 * For non-IDE disks, the host supports multiple channels.
2060 	 * Set the number of HW queues we are supporting.
2061 	 */
2062 	if (!dev_is_ide) {
2063 		if (storvsc_max_hw_queues > num_present_cpus) {
2064 			storvsc_max_hw_queues = 0;
2065 			storvsc_log(device, STORVSC_LOGGING_WARN,
2066 				"Resetting invalid storvsc_max_hw_queues value to default.\n");
2067 		}
2068 		if (storvsc_max_hw_queues)
2069 			host->nr_hw_queues = storvsc_max_hw_queues;
2070 		else
2071 			host->nr_hw_queues = num_present_cpus;
2072 	}
2073 
2074 	/*
2075 	 * Set the error handler work queue.
2076 	 */
2077 	host_dev->handle_error_wq =
2078 			alloc_ordered_workqueue("storvsc_error_wq_%d",
2079 						WQ_MEM_RECLAIM,
2080 						host->host_no);
2081 	if (!host_dev->handle_error_wq) {
2082 		ret = -ENOMEM;
2083 		goto err_out2;
2084 	}
2085 	INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2086 	/* Register the HBA and start the scsi bus scan */
2087 	ret = scsi_add_host(host, &device->device);
2088 	if (ret != 0)
2089 		goto err_out3;
2090 
2091 	if (!dev_is_ide) {
2092 		scsi_scan_host(host);
2093 	} else {
2094 		target = (device->dev_instance.b[5] << 8 |
2095 			 device->dev_instance.b[4]);
2096 		ret = scsi_add_device(host, 0, target, 0);
2097 		if (ret)
2098 			goto err_out4;
2099 	}
2100 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2101 	if (host->transportt == fc_transport_template) {
2102 		struct fc_rport_identifiers ids = {
2103 			.roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2104 		};
2105 
2106 		fc_host_node_name(host) = stor_device->node_name;
2107 		fc_host_port_name(host) = stor_device->port_name;
2108 		stor_device->rport = fc_remote_port_add(host, 0, &ids);
2109 		if (!stor_device->rport) {
2110 			ret = -ENOMEM;
2111 			goto err_out4;
2112 		}
2113 	}
2114 #endif
2115 	return 0;
2116 
2117 err_out4:
2118 	scsi_remove_host(host);
2119 
2120 err_out3:
2121 	destroy_workqueue(host_dev->handle_error_wq);
2122 
2123 err_out2:
2124 	/*
2125 	 * Once we have connected with the host, we would need to
2126 	 * to invoke storvsc_dev_remove() to rollback this state and
2127 	 * this call also frees up the stor_device; hence the jump around
2128 	 * err_out1 label.
2129 	 */
2130 	storvsc_dev_remove(device);
2131 	goto err_out0;
2132 
2133 err_out1:
2134 	kfree(stor_device->stor_chns);
2135 	kfree(stor_device);
2136 
2137 err_out0:
2138 	scsi_host_put(host);
2139 	return ret;
2140 }
2141 
2142 /* Change a scsi target's queue depth */
2143 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2144 {
2145 	if (queue_depth > scsi_driver.can_queue)
2146 		queue_depth = scsi_driver.can_queue;
2147 
2148 	return scsi_change_queue_depth(sdev, queue_depth);
2149 }
2150 
2151 static int storvsc_remove(struct hv_device *dev)
2152 {
2153 	struct storvsc_device *stor_device = hv_get_drvdata(dev);
2154 	struct Scsi_Host *host = stor_device->host;
2155 	struct hv_host_device *host_dev = shost_priv(host);
2156 
2157 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2158 	if (host->transportt == fc_transport_template) {
2159 		fc_remote_port_delete(stor_device->rport);
2160 		fc_remove_host(host);
2161 	}
2162 #endif
2163 	destroy_workqueue(host_dev->handle_error_wq);
2164 	scsi_remove_host(host);
2165 	storvsc_dev_remove(dev);
2166 	scsi_host_put(host);
2167 
2168 	return 0;
2169 }
2170 
2171 static int storvsc_suspend(struct hv_device *hv_dev)
2172 {
2173 	struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2174 	struct Scsi_Host *host = stor_device->host;
2175 	struct hv_host_device *host_dev = shost_priv(host);
2176 
2177 	storvsc_wait_to_drain(stor_device);
2178 
2179 	drain_workqueue(host_dev->handle_error_wq);
2180 
2181 	vmbus_close(hv_dev->channel);
2182 
2183 	kfree(stor_device->stor_chns);
2184 	stor_device->stor_chns = NULL;
2185 
2186 	cpumask_clear(&stor_device->alloced_cpus);
2187 
2188 	return 0;
2189 }
2190 
2191 static int storvsc_resume(struct hv_device *hv_dev)
2192 {
2193 	int ret;
2194 
2195 	ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2196 				     hv_dev_is_fc(hv_dev));
2197 	return ret;
2198 }
2199 
2200 static struct hv_driver storvsc_drv = {
2201 	.name = KBUILD_MODNAME,
2202 	.id_table = id_table,
2203 	.probe = storvsc_probe,
2204 	.remove = storvsc_remove,
2205 	.suspend = storvsc_suspend,
2206 	.resume = storvsc_resume,
2207 	.driver = {
2208 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
2209 	},
2210 };
2211 
2212 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2213 static struct fc_function_template fc_transport_functions = {
2214 	.show_host_node_name = 1,
2215 	.show_host_port_name = 1,
2216 };
2217 #endif
2218 
2219 static int __init storvsc_drv_init(void)
2220 {
2221 	int ret;
2222 
2223 	/*
2224 	 * Divide the ring buffer data size (which is 1 page less
2225 	 * than the ring buffer size since that page is reserved for
2226 	 * the ring buffer indices) by the max request size (which is
2227 	 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2228 	 *
2229 	 * The computation underestimates max_outstanding_req_per_channel
2230 	 * for Win7 and older hosts because it does not take into account
2231 	 * the vmscsi_size_delta correction to the max request size.
2232 	 */
2233 	max_outstanding_req_per_channel =
2234 		((storvsc_ringbuffer_size - PAGE_SIZE) /
2235 		ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2236 		sizeof(struct vstor_packet) + sizeof(u64),
2237 		sizeof(u64)));
2238 
2239 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2240 	fc_transport_template = fc_attach_transport(&fc_transport_functions);
2241 	if (!fc_transport_template)
2242 		return -ENODEV;
2243 #endif
2244 
2245 	ret = vmbus_driver_register(&storvsc_drv);
2246 
2247 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2248 	if (ret)
2249 		fc_release_transport(fc_transport_template);
2250 #endif
2251 
2252 	return ret;
2253 }
2254 
2255 static void __exit storvsc_drv_exit(void)
2256 {
2257 	vmbus_driver_unregister(&storvsc_drv);
2258 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2259 	fc_release_transport(fc_transport_template);
2260 #endif
2261 }
2262 
2263 MODULE_LICENSE("GPL");
2264 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2265 module_init(storvsc_drv_init);
2266 module_exit(storvsc_drv_exit);
2267