xref: /linux/drivers/scsi/storvsc_drv.c (revision 63ba5b0fb4f54db256ec43b3062b2606b383055d)
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_TIMEOUT		0x09
320 #define SRB_STATUS_SELECTION_TIMEOUT	0x0A
321 #define SRB_STATUS_BUS_RESET		0x0E
322 #define SRB_STATUS_DATA_OVERRUN		0x12
323 #define SRB_STATUS_INVALID_LUN		0x20
324 #define SRB_STATUS_INTERNAL_ERROR	0x30
325 
326 #define SRB_STATUS(status) \
327 	(status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
328 /*
329  * This is the end of Protocol specific defines.
330  */
331 
332 static int storvsc_ringbuffer_size = (128 * 1024);
333 static int aligned_ringbuffer_size;
334 static u32 max_outstanding_req_per_channel;
335 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
336 
337 static int storvsc_vcpus_per_sub_channel = 4;
338 static unsigned int storvsc_max_hw_queues;
339 
340 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
341 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
342 
343 module_param(storvsc_max_hw_queues, uint, 0644);
344 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
345 
346 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
347 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
348 
349 static int ring_avail_percent_lowater = 10;
350 module_param(ring_avail_percent_lowater, int, S_IRUGO);
351 MODULE_PARM_DESC(ring_avail_percent_lowater,
352 		"Select a channel if available ring size > this in percent");
353 
354 /*
355  * Timeout in seconds for all devices managed by this driver.
356  */
357 static int storvsc_timeout = 180;
358 
359 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
360 static struct scsi_transport_template *fc_transport_template;
361 #endif
362 
363 static struct scsi_host_template scsi_driver;
364 static void storvsc_on_channel_callback(void *context);
365 
366 #define STORVSC_MAX_LUNS_PER_TARGET			255
367 #define STORVSC_MAX_TARGETS				2
368 #define STORVSC_MAX_CHANNELS				8
369 
370 #define STORVSC_FC_MAX_LUNS_PER_TARGET			255
371 #define STORVSC_FC_MAX_TARGETS				128
372 #define STORVSC_FC_MAX_CHANNELS				8
373 #define STORVSC_FC_MAX_XFER_SIZE			((u32)(512 * 1024))
374 
375 #define STORVSC_IDE_MAX_LUNS_PER_TARGET			64
376 #define STORVSC_IDE_MAX_TARGETS				1
377 #define STORVSC_IDE_MAX_CHANNELS			1
378 
379 /*
380  * Upper bound on the size of a storvsc packet.
381  */
382 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
383 			      sizeof(struct vstor_packet))
384 
385 struct storvsc_cmd_request {
386 	struct scsi_cmnd *cmd;
387 
388 	struct hv_device *device;
389 
390 	/* Synchronize the request/response if needed */
391 	struct completion wait_event;
392 
393 	struct vmbus_channel_packet_multipage_buffer mpb;
394 	struct vmbus_packet_mpb_array *payload;
395 	u32 payload_sz;
396 
397 	struct vstor_packet vstor_packet;
398 };
399 
400 
401 /* A storvsc device is a device object that contains a vmbus channel */
402 struct storvsc_device {
403 	struct hv_device *device;
404 
405 	bool	 destroy;
406 	bool	 drain_notify;
407 	atomic_t num_outstanding_req;
408 	struct Scsi_Host *host;
409 
410 	wait_queue_head_t waiting_to_drain;
411 
412 	/*
413 	 * Each unique Port/Path/Target represents 1 channel ie scsi
414 	 * controller. In reality, the pathid, targetid is always 0
415 	 * and the port is set by us
416 	 */
417 	unsigned int port_number;
418 	unsigned char path_id;
419 	unsigned char target_id;
420 
421 	/*
422 	 * Max I/O, the device can support.
423 	 */
424 	u32   max_transfer_bytes;
425 	/*
426 	 * Number of sub-channels we will open.
427 	 */
428 	u16 num_sc;
429 	struct vmbus_channel **stor_chns;
430 	/*
431 	 * Mask of CPUs bound to subchannels.
432 	 */
433 	struct cpumask alloced_cpus;
434 	/*
435 	 * Serializes modifications of stor_chns[] from storvsc_do_io()
436 	 * and storvsc_change_target_cpu().
437 	 */
438 	spinlock_t lock;
439 	/* Used for vsc/vsp channel reset process */
440 	struct storvsc_cmd_request init_request;
441 	struct storvsc_cmd_request reset_request;
442 	/*
443 	 * Currently active port and node names for FC devices.
444 	 */
445 	u64 node_name;
446 	u64 port_name;
447 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
448 	struct fc_rport *rport;
449 #endif
450 };
451 
452 struct hv_host_device {
453 	struct hv_device *dev;
454 	unsigned int port;
455 	unsigned char path;
456 	unsigned char target;
457 	struct workqueue_struct *handle_error_wq;
458 	struct work_struct host_scan_work;
459 	struct Scsi_Host *host;
460 };
461 
462 struct storvsc_scan_work {
463 	struct work_struct work;
464 	struct Scsi_Host *host;
465 	u8 lun;
466 	u8 tgt_id;
467 };
468 
469 static void storvsc_device_scan(struct work_struct *work)
470 {
471 	struct storvsc_scan_work *wrk;
472 	struct scsi_device *sdev;
473 
474 	wrk = container_of(work, struct storvsc_scan_work, work);
475 
476 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
477 	if (!sdev)
478 		goto done;
479 	scsi_rescan_device(sdev);
480 	scsi_device_put(sdev);
481 
482 done:
483 	kfree(wrk);
484 }
485 
486 static void storvsc_host_scan(struct work_struct *work)
487 {
488 	struct Scsi_Host *host;
489 	struct scsi_device *sdev;
490 	struct hv_host_device *host_device =
491 		container_of(work, struct hv_host_device, host_scan_work);
492 
493 	host = host_device->host;
494 	/*
495 	 * Before scanning the host, first check to see if any of the
496 	 * currently known devices have been hot removed. We issue a
497 	 * "unit ready" command against all currently known devices.
498 	 * This I/O will result in an error for devices that have been
499 	 * removed. As part of handling the I/O error, we remove the device.
500 	 *
501 	 * When a LUN is added or removed, the host sends us a signal to
502 	 * scan the host. Thus we are forced to discover the LUNs that
503 	 * may have been removed this way.
504 	 */
505 	mutex_lock(&host->scan_mutex);
506 	shost_for_each_device(sdev, host)
507 		scsi_test_unit_ready(sdev, 1, 1, NULL);
508 	mutex_unlock(&host->scan_mutex);
509 	/*
510 	 * Now scan the host to discover LUNs that may have been added.
511 	 */
512 	scsi_scan_host(host);
513 }
514 
515 static void storvsc_remove_lun(struct work_struct *work)
516 {
517 	struct storvsc_scan_work *wrk;
518 	struct scsi_device *sdev;
519 
520 	wrk = container_of(work, struct storvsc_scan_work, work);
521 	if (!scsi_host_get(wrk->host))
522 		goto done;
523 
524 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
525 
526 	if (sdev) {
527 		scsi_remove_device(sdev);
528 		scsi_device_put(sdev);
529 	}
530 	scsi_host_put(wrk->host);
531 
532 done:
533 	kfree(wrk);
534 }
535 
536 
537 /*
538  * We can get incoming messages from the host that are not in response to
539  * messages that we have sent out. An example of this would be messages
540  * received by the guest to notify dynamic addition/removal of LUNs. To
541  * deal with potential race conditions where the driver may be in the
542  * midst of being unloaded when we might receive an unsolicited message
543  * from the host, we have implemented a mechanism to gurantee sequential
544  * consistency:
545  *
546  * 1) Once the device is marked as being destroyed, we will fail all
547  *    outgoing messages.
548  * 2) We permit incoming messages when the device is being destroyed,
549  *    only to properly account for messages already sent out.
550  */
551 
552 static inline struct storvsc_device *get_out_stor_device(
553 					struct hv_device *device)
554 {
555 	struct storvsc_device *stor_device;
556 
557 	stor_device = hv_get_drvdata(device);
558 
559 	if (stor_device && stor_device->destroy)
560 		stor_device = NULL;
561 
562 	return stor_device;
563 }
564 
565 
566 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
567 {
568 	dev->drain_notify = true;
569 	wait_event(dev->waiting_to_drain,
570 		   atomic_read(&dev->num_outstanding_req) == 0);
571 	dev->drain_notify = false;
572 }
573 
574 static inline struct storvsc_device *get_in_stor_device(
575 					struct hv_device *device)
576 {
577 	struct storvsc_device *stor_device;
578 
579 	stor_device = hv_get_drvdata(device);
580 
581 	if (!stor_device)
582 		goto get_in_err;
583 
584 	/*
585 	 * If the device is being destroyed; allow incoming
586 	 * traffic only to cleanup outstanding requests.
587 	 */
588 
589 	if (stor_device->destroy  &&
590 		(atomic_read(&stor_device->num_outstanding_req) == 0))
591 		stor_device = NULL;
592 
593 get_in_err:
594 	return stor_device;
595 
596 }
597 
598 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
599 				      u32 new)
600 {
601 	struct storvsc_device *stor_device;
602 	struct vmbus_channel *cur_chn;
603 	bool old_is_alloced = false;
604 	struct hv_device *device;
605 	unsigned long flags;
606 	int cpu;
607 
608 	device = channel->primary_channel ?
609 			channel->primary_channel->device_obj
610 				: channel->device_obj;
611 	stor_device = get_out_stor_device(device);
612 	if (!stor_device)
613 		return;
614 
615 	/* See storvsc_do_io() -> get_og_chn(). */
616 	spin_lock_irqsave(&stor_device->lock, flags);
617 
618 	/*
619 	 * Determines if the storvsc device has other channels assigned to
620 	 * the "old" CPU to update the alloced_cpus mask and the stor_chns
621 	 * array.
622 	 */
623 	if (device->channel != channel && device->channel->target_cpu == old) {
624 		cur_chn = device->channel;
625 		old_is_alloced = true;
626 		goto old_is_alloced;
627 	}
628 	list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
629 		if (cur_chn == channel)
630 			continue;
631 		if (cur_chn->target_cpu == old) {
632 			old_is_alloced = true;
633 			goto old_is_alloced;
634 		}
635 	}
636 
637 old_is_alloced:
638 	if (old_is_alloced)
639 		WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
640 	else
641 		cpumask_clear_cpu(old, &stor_device->alloced_cpus);
642 
643 	/* "Flush" the stor_chns array. */
644 	for_each_possible_cpu(cpu) {
645 		if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
646 					cpu, &stor_device->alloced_cpus))
647 			WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
648 	}
649 
650 	WRITE_ONCE(stor_device->stor_chns[new], channel);
651 	cpumask_set_cpu(new, &stor_device->alloced_cpus);
652 
653 	spin_unlock_irqrestore(&stor_device->lock, flags);
654 }
655 
656 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
657 {
658 	struct storvsc_cmd_request *request =
659 		(struct storvsc_cmd_request *)(unsigned long)rqst_addr;
660 
661 	if (rqst_addr == VMBUS_RQST_INIT)
662 		return VMBUS_RQST_INIT;
663 	if (rqst_addr == VMBUS_RQST_RESET)
664 		return VMBUS_RQST_RESET;
665 
666 	/*
667 	 * Cannot return an ID of 0, which is reserved for an unsolicited
668 	 * message from Hyper-V.
669 	 */
670 	return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
671 }
672 
673 static void handle_sc_creation(struct vmbus_channel *new_sc)
674 {
675 	struct hv_device *device = new_sc->primary_channel->device_obj;
676 	struct device *dev = &device->device;
677 	struct storvsc_device *stor_device;
678 	struct vmstorage_channel_properties props;
679 	int ret;
680 
681 	stor_device = get_out_stor_device(device);
682 	if (!stor_device)
683 		return;
684 
685 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
686 	new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
687 
688 	new_sc->next_request_id_callback = storvsc_next_request_id;
689 
690 	ret = vmbus_open(new_sc,
691 			 aligned_ringbuffer_size,
692 			 aligned_ringbuffer_size,
693 			 (void *)&props,
694 			 sizeof(struct vmstorage_channel_properties),
695 			 storvsc_on_channel_callback, new_sc);
696 
697 	/* In case vmbus_open() fails, we don't use the sub-channel. */
698 	if (ret != 0) {
699 		dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
700 		return;
701 	}
702 
703 	new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
704 
705 	/* Add the sub-channel to the array of available channels. */
706 	stor_device->stor_chns[new_sc->target_cpu] = new_sc;
707 	cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
708 }
709 
710 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
711 {
712 	struct device *dev = &device->device;
713 	struct storvsc_device *stor_device;
714 	int num_sc;
715 	struct storvsc_cmd_request *request;
716 	struct vstor_packet *vstor_packet;
717 	int ret, t;
718 
719 	/*
720 	 * If the number of CPUs is artificially restricted, such as
721 	 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
722 	 * sub-channels >= the number of CPUs. These sub-channels
723 	 * should not be created. The primary channel is already created
724 	 * and assigned to one CPU, so check against # CPUs - 1.
725 	 */
726 	num_sc = min((int)(num_online_cpus() - 1), max_chns);
727 	if (!num_sc)
728 		return;
729 
730 	stor_device = get_out_stor_device(device);
731 	if (!stor_device)
732 		return;
733 
734 	stor_device->num_sc = num_sc;
735 	request = &stor_device->init_request;
736 	vstor_packet = &request->vstor_packet;
737 
738 	/*
739 	 * Establish a handler for dealing with subchannels.
740 	 */
741 	vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
742 
743 	/*
744 	 * Request the host to create sub-channels.
745 	 */
746 	memset(request, 0, sizeof(struct storvsc_cmd_request));
747 	init_completion(&request->wait_event);
748 	vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
749 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
750 	vstor_packet->sub_channel_count = num_sc;
751 
752 	ret = vmbus_sendpacket(device->channel, vstor_packet,
753 			       sizeof(struct vstor_packet),
754 			       VMBUS_RQST_INIT,
755 			       VM_PKT_DATA_INBAND,
756 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
757 
758 	if (ret != 0) {
759 		dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
760 		return;
761 	}
762 
763 	t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
764 	if (t == 0) {
765 		dev_err(dev, "Failed to create sub-channel: timed out\n");
766 		return;
767 	}
768 
769 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
770 	    vstor_packet->status != 0) {
771 		dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
772 			vstor_packet->operation, vstor_packet->status);
773 		return;
774 	}
775 
776 	/*
777 	 * We need to do nothing here, because vmbus_process_offer()
778 	 * invokes channel->sc_creation_callback, which will open and use
779 	 * the sub-channel(s).
780 	 */
781 }
782 
783 static void cache_wwn(struct storvsc_device *stor_device,
784 		      struct vstor_packet *vstor_packet)
785 {
786 	/*
787 	 * Cache the currently active port and node ww names.
788 	 */
789 	if (vstor_packet->wwn_packet.primary_active) {
790 		stor_device->node_name =
791 			wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
792 		stor_device->port_name =
793 			wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
794 	} else {
795 		stor_device->node_name =
796 			wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
797 		stor_device->port_name =
798 			wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
799 	}
800 }
801 
802 
803 static int storvsc_execute_vstor_op(struct hv_device *device,
804 				    struct storvsc_cmd_request *request,
805 				    bool status_check)
806 {
807 	struct storvsc_device *stor_device;
808 	struct vstor_packet *vstor_packet;
809 	int ret, t;
810 
811 	stor_device = get_out_stor_device(device);
812 	if (!stor_device)
813 		return -ENODEV;
814 
815 	vstor_packet = &request->vstor_packet;
816 
817 	init_completion(&request->wait_event);
818 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
819 
820 	ret = vmbus_sendpacket(device->channel, vstor_packet,
821 			       sizeof(struct vstor_packet),
822 			       VMBUS_RQST_INIT,
823 			       VM_PKT_DATA_INBAND,
824 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
825 	if (ret != 0)
826 		return ret;
827 
828 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
829 	if (t == 0)
830 		return -ETIMEDOUT;
831 
832 	if (!status_check)
833 		return ret;
834 
835 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
836 	    vstor_packet->status != 0)
837 		return -EINVAL;
838 
839 	return ret;
840 }
841 
842 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
843 {
844 	struct storvsc_device *stor_device;
845 	struct storvsc_cmd_request *request;
846 	struct vstor_packet *vstor_packet;
847 	int ret, i;
848 	int max_chns;
849 	bool process_sub_channels = false;
850 
851 	stor_device = get_out_stor_device(device);
852 	if (!stor_device)
853 		return -ENODEV;
854 
855 	request = &stor_device->init_request;
856 	vstor_packet = &request->vstor_packet;
857 
858 	/*
859 	 * Now, initiate the vsc/vsp initialization protocol on the open
860 	 * channel
861 	 */
862 	memset(request, 0, sizeof(struct storvsc_cmd_request));
863 	vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
864 	ret = storvsc_execute_vstor_op(device, request, true);
865 	if (ret)
866 		return ret;
867 	/*
868 	 * Query host supported protocol version.
869 	 */
870 
871 	for (i = 0; i < ARRAY_SIZE(protocol_version); i++) {
872 		/* reuse the packet for version range supported */
873 		memset(vstor_packet, 0, sizeof(struct vstor_packet));
874 		vstor_packet->operation =
875 			VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
876 
877 		vstor_packet->version.major_minor = protocol_version[i];
878 
879 		/*
880 		 * The revision number is only used in Windows; set it to 0.
881 		 */
882 		vstor_packet->version.revision = 0;
883 		ret = storvsc_execute_vstor_op(device, request, false);
884 		if (ret != 0)
885 			return ret;
886 
887 		if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
888 			return -EINVAL;
889 
890 		if (vstor_packet->status == 0) {
891 			vmstor_proto_version = protocol_version[i];
892 
893 			break;
894 		}
895 	}
896 
897 	if (vstor_packet->status != 0) {
898 		dev_err(&device->device, "Obsolete Hyper-V version\n");
899 		return -EINVAL;
900 	}
901 
902 
903 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
904 	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
905 	ret = storvsc_execute_vstor_op(device, request, true);
906 	if (ret != 0)
907 		return ret;
908 
909 	/*
910 	 * Check to see if multi-channel support is there.
911 	 * Hosts that implement protocol version of 5.1 and above
912 	 * support multi-channel.
913 	 */
914 	max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
915 
916 	/*
917 	 * Allocate state to manage the sub-channels.
918 	 * We allocate an array based on the numbers of possible CPUs
919 	 * (Hyper-V does not support cpu online/offline).
920 	 * This Array will be sparseley populated with unique
921 	 * channels - primary + sub-channels.
922 	 * We will however populate all the slots to evenly distribute
923 	 * the load.
924 	 */
925 	stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
926 					 GFP_KERNEL);
927 	if (stor_device->stor_chns == NULL)
928 		return -ENOMEM;
929 
930 	device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
931 
932 	stor_device->stor_chns[device->channel->target_cpu] = device->channel;
933 	cpumask_set_cpu(device->channel->target_cpu,
934 			&stor_device->alloced_cpus);
935 
936 	if (vstor_packet->storage_channel_properties.flags &
937 	    STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
938 		process_sub_channels = true;
939 
940 	stor_device->max_transfer_bytes =
941 		vstor_packet->storage_channel_properties.max_transfer_bytes;
942 
943 	if (!is_fc)
944 		goto done;
945 
946 	/*
947 	 * For FC devices retrieve FC HBA data.
948 	 */
949 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
950 	vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
951 	ret = storvsc_execute_vstor_op(device, request, true);
952 	if (ret != 0)
953 		return ret;
954 
955 	/*
956 	 * Cache the currently active port and node ww names.
957 	 */
958 	cache_wwn(stor_device, vstor_packet);
959 
960 done:
961 
962 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
963 	vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
964 	ret = storvsc_execute_vstor_op(device, request, true);
965 	if (ret != 0)
966 		return ret;
967 
968 	if (process_sub_channels)
969 		handle_multichannel_storage(device, max_chns);
970 
971 	return ret;
972 }
973 
974 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
975 				struct scsi_cmnd *scmnd,
976 				struct Scsi_Host *host,
977 				u8 asc, u8 ascq)
978 {
979 	struct storvsc_scan_work *wrk;
980 	void (*process_err_fn)(struct work_struct *work);
981 	struct hv_host_device *host_dev = shost_priv(host);
982 
983 	switch (SRB_STATUS(vm_srb->srb_status)) {
984 	case SRB_STATUS_ERROR:
985 	case SRB_STATUS_ABORTED:
986 	case SRB_STATUS_INVALID_REQUEST:
987 	case SRB_STATUS_INTERNAL_ERROR:
988 	case SRB_STATUS_TIMEOUT:
989 	case SRB_STATUS_SELECTION_TIMEOUT:
990 	case SRB_STATUS_BUS_RESET:
991 	case SRB_STATUS_DATA_OVERRUN:
992 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
993 			/* Check for capacity change */
994 			if ((asc == 0x2a) && (ascq == 0x9)) {
995 				process_err_fn = storvsc_device_scan;
996 				/* Retry the I/O that triggered this. */
997 				set_host_byte(scmnd, DID_REQUEUE);
998 				goto do_work;
999 			}
1000 
1001 			/*
1002 			 * Check for "Operating parameters have changed"
1003 			 * due to Hyper-V changing the VHD/VHDX BlockSize
1004 			 * when adding/removing a differencing disk. This
1005 			 * causes discard_granularity to change, so do a
1006 			 * rescan to pick up the new granularity. We don't
1007 			 * want scsi_report_sense() to output a message
1008 			 * that a sysadmin wouldn't know what to do with.
1009 			 */
1010 			if ((asc == 0x3f) && (ascq != 0x03) &&
1011 					(ascq != 0x0e)) {
1012 				process_err_fn = storvsc_device_scan;
1013 				set_host_byte(scmnd, DID_REQUEUE);
1014 				goto do_work;
1015 			}
1016 
1017 			/*
1018 			 * Otherwise, let upper layer deal with the
1019 			 * error when sense message is present
1020 			 */
1021 			return;
1022 		}
1023 
1024 		/*
1025 		 * If there is an error; offline the device since all
1026 		 * error recovery strategies would have already been
1027 		 * deployed on the host side. However, if the command
1028 		 * were a pass-through command deal with it appropriately.
1029 		 */
1030 		switch (scmnd->cmnd[0]) {
1031 		case ATA_16:
1032 		case ATA_12:
1033 			set_host_byte(scmnd, DID_PASSTHROUGH);
1034 			break;
1035 		/*
1036 		 * On some Hyper-V hosts TEST_UNIT_READY command can
1037 		 * return SRB_STATUS_ERROR. Let the upper level code
1038 		 * deal with it based on the sense information.
1039 		 */
1040 		case TEST_UNIT_READY:
1041 			break;
1042 		default:
1043 			set_host_byte(scmnd, DID_ERROR);
1044 		}
1045 		return;
1046 
1047 	case SRB_STATUS_INVALID_LUN:
1048 		set_host_byte(scmnd, DID_NO_CONNECT);
1049 		process_err_fn = storvsc_remove_lun;
1050 		goto do_work;
1051 
1052 	}
1053 	return;
1054 
1055 do_work:
1056 	/*
1057 	 * We need to schedule work to process this error; schedule it.
1058 	 */
1059 	wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1060 	if (!wrk) {
1061 		set_host_byte(scmnd, DID_BAD_TARGET);
1062 		return;
1063 	}
1064 
1065 	wrk->host = host;
1066 	wrk->lun = vm_srb->lun;
1067 	wrk->tgt_id = vm_srb->target_id;
1068 	INIT_WORK(&wrk->work, process_err_fn);
1069 	queue_work(host_dev->handle_error_wq, &wrk->work);
1070 }
1071 
1072 
1073 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1074 				       struct storvsc_device *stor_dev)
1075 {
1076 	struct scsi_cmnd *scmnd = cmd_request->cmd;
1077 	struct scsi_sense_hdr sense_hdr;
1078 	struct vmscsi_request *vm_srb;
1079 	u32 data_transfer_length;
1080 	struct Scsi_Host *host;
1081 	u32 payload_sz = cmd_request->payload_sz;
1082 	void *payload = cmd_request->payload;
1083 	bool sense_ok;
1084 
1085 	host = stor_dev->host;
1086 
1087 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1088 	data_transfer_length = vm_srb->data_transfer_length;
1089 
1090 	scmnd->result = vm_srb->scsi_status;
1091 
1092 	if (scmnd->result) {
1093 		sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1094 				SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1095 
1096 		if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1097 			scsi_print_sense_hdr(scmnd->device, "storvsc",
1098 					     &sense_hdr);
1099 	}
1100 
1101 	if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1102 		storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1103 					 sense_hdr.ascq);
1104 		/*
1105 		 * The Windows driver set data_transfer_length on
1106 		 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1107 		 * is untouched.  In these cases we set it to 0.
1108 		 */
1109 		if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1110 			data_transfer_length = 0;
1111 	}
1112 
1113 	/* Validate data_transfer_length (from Hyper-V) */
1114 	if (data_transfer_length > cmd_request->payload->range.len)
1115 		data_transfer_length = cmd_request->payload->range.len;
1116 
1117 	scsi_set_resid(scmnd,
1118 		cmd_request->payload->range.len - data_transfer_length);
1119 
1120 	scsi_done(scmnd);
1121 
1122 	if (payload_sz >
1123 		sizeof(struct vmbus_channel_packet_multipage_buffer))
1124 		kfree(payload);
1125 }
1126 
1127 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1128 				  struct vstor_packet *vstor_packet,
1129 				  struct storvsc_cmd_request *request)
1130 {
1131 	struct vstor_packet *stor_pkt;
1132 	struct hv_device *device = stor_device->device;
1133 
1134 	stor_pkt = &request->vstor_packet;
1135 
1136 	/*
1137 	 * The current SCSI handling on the host side does
1138 	 * not correctly handle:
1139 	 * INQUIRY command with page code parameter set to 0x80
1140 	 * MODE_SENSE command with cmd[2] == 0x1c
1141 	 *
1142 	 * Setup srb and scsi status so this won't be fatal.
1143 	 * We do this so we can distinguish truly fatal failues
1144 	 * (srb status == 0x4) and off-line the device in that case.
1145 	 */
1146 
1147 	if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1148 	   (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1149 		vstor_packet->vm_srb.scsi_status = 0;
1150 		vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1151 	}
1152 
1153 	/* Copy over the status...etc */
1154 	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1155 	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1156 
1157 	/*
1158 	 * Copy over the sense_info_length, but limit to the known max
1159 	 * size if Hyper-V returns a bad value.
1160 	 */
1161 	stor_pkt->vm_srb.sense_info_length = min_t(u8, STORVSC_SENSE_BUFFER_SIZE,
1162 		vstor_packet->vm_srb.sense_info_length);
1163 
1164 	if (vstor_packet->vm_srb.scsi_status != 0 ||
1165 	    vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1166 
1167 		/*
1168 		 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1169 		 * return errors when detecting devices using TEST_UNIT_READY,
1170 		 * and logging these as errors produces unhelpful noise.
1171 		 */
1172 		int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1173 			STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1174 
1175 		storvsc_log(device, loglevel,
1176 			"tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1177 			scsi_cmd_to_rq(request->cmd)->tag,
1178 			stor_pkt->vm_srb.cdb[0],
1179 			vstor_packet->vm_srb.scsi_status,
1180 			vstor_packet->vm_srb.srb_status,
1181 			vstor_packet->status);
1182 	}
1183 
1184 	if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1185 	    (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1186 		memcpy(request->cmd->sense_buffer,
1187 		       vstor_packet->vm_srb.sense_data,
1188 		       stor_pkt->vm_srb.sense_info_length);
1189 
1190 	stor_pkt->vm_srb.data_transfer_length =
1191 		vstor_packet->vm_srb.data_transfer_length;
1192 
1193 	storvsc_command_completion(request, stor_device);
1194 
1195 	if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1196 		stor_device->drain_notify)
1197 		wake_up(&stor_device->waiting_to_drain);
1198 }
1199 
1200 static void storvsc_on_receive(struct storvsc_device *stor_device,
1201 			     struct vstor_packet *vstor_packet,
1202 			     struct storvsc_cmd_request *request)
1203 {
1204 	struct hv_host_device *host_dev;
1205 	switch (vstor_packet->operation) {
1206 	case VSTOR_OPERATION_COMPLETE_IO:
1207 		storvsc_on_io_completion(stor_device, vstor_packet, request);
1208 		break;
1209 
1210 	case VSTOR_OPERATION_REMOVE_DEVICE:
1211 	case VSTOR_OPERATION_ENUMERATE_BUS:
1212 		host_dev = shost_priv(stor_device->host);
1213 		queue_work(
1214 			host_dev->handle_error_wq, &host_dev->host_scan_work);
1215 		break;
1216 
1217 	case VSTOR_OPERATION_FCHBA_DATA:
1218 		cache_wwn(stor_device, vstor_packet);
1219 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1220 		fc_host_node_name(stor_device->host) = stor_device->node_name;
1221 		fc_host_port_name(stor_device->host) = stor_device->port_name;
1222 #endif
1223 		break;
1224 	default:
1225 		break;
1226 	}
1227 }
1228 
1229 static void storvsc_on_channel_callback(void *context)
1230 {
1231 	struct vmbus_channel *channel = (struct vmbus_channel *)context;
1232 	const struct vmpacket_descriptor *desc;
1233 	struct hv_device *device;
1234 	struct storvsc_device *stor_device;
1235 	struct Scsi_Host *shost;
1236 	unsigned long time_limit = jiffies + msecs_to_jiffies(CALLBACK_TIMEOUT);
1237 
1238 	if (channel->primary_channel != NULL)
1239 		device = channel->primary_channel->device_obj;
1240 	else
1241 		device = channel->device_obj;
1242 
1243 	stor_device = get_in_stor_device(device);
1244 	if (!stor_device)
1245 		return;
1246 
1247 	shost = stor_device->host;
1248 
1249 	foreach_vmbus_pkt(desc, channel) {
1250 		struct vstor_packet *packet = hv_pkt_data(desc);
1251 		struct storvsc_cmd_request *request = NULL;
1252 		u32 pktlen = hv_pkt_datalen(desc);
1253 		u64 rqst_id = desc->trans_id;
1254 		u32 minlen = rqst_id ? sizeof(struct vstor_packet) :
1255 			sizeof(enum vstor_packet_operation);
1256 
1257 		if (unlikely(time_after(jiffies, time_limit))) {
1258 			hv_pkt_iter_close(channel);
1259 			return;
1260 		}
1261 
1262 		if (pktlen < minlen) {
1263 			dev_err(&device->device,
1264 				"Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1265 				rqst_id, pktlen, minlen);
1266 			continue;
1267 		}
1268 
1269 		if (rqst_id == VMBUS_RQST_INIT) {
1270 			request = &stor_device->init_request;
1271 		} else if (rqst_id == VMBUS_RQST_RESET) {
1272 			request = &stor_device->reset_request;
1273 		} else {
1274 			/* Hyper-V can send an unsolicited message with ID of 0 */
1275 			if (rqst_id == 0) {
1276 				/*
1277 				 * storvsc_on_receive() looks at the vstor_packet in the message
1278 				 * from the ring buffer.
1279 				 *
1280 				 * - If the operation in the vstor_packet is COMPLETE_IO, then
1281 				 *   we call storvsc_on_io_completion(), and dereference the
1282 				 *   guest memory address.  Make sure we don't call
1283 				 *   storvsc_on_io_completion() with a guest memory address
1284 				 *   that is zero if Hyper-V were to construct and send such
1285 				 *   a bogus packet.
1286 				 *
1287 				 * - If the operation in the vstor_packet is FCHBA_DATA, then
1288 				 *   we call cache_wwn(), and access the data payload area of
1289 				 *   the packet (wwn_packet); however, there is no guarantee
1290 				 *   that the packet is big enough to contain such area.
1291 				 *   Future-proof the code by rejecting such a bogus packet.
1292 				 */
1293 				if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1294 				    packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1295 					dev_err(&device->device, "Invalid packet with ID of 0\n");
1296 					continue;
1297 				}
1298 			} else {
1299 				struct scsi_cmnd *scmnd;
1300 
1301 				/* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1302 				scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1303 				if (scmnd == NULL) {
1304 					dev_err(&device->device, "Incorrect transaction ID\n");
1305 					continue;
1306 				}
1307 				request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1308 				scsi_dma_unmap(scmnd);
1309 			}
1310 
1311 			storvsc_on_receive(stor_device, packet, request);
1312 			continue;
1313 		}
1314 
1315 		memcpy(&request->vstor_packet, packet,
1316 		       sizeof(struct vstor_packet));
1317 		complete(&request->wait_event);
1318 	}
1319 }
1320 
1321 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1322 				  bool is_fc)
1323 {
1324 	struct vmstorage_channel_properties props;
1325 	int ret;
1326 
1327 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1328 
1329 	device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1330 	device->channel->next_request_id_callback = storvsc_next_request_id;
1331 
1332 	ret = vmbus_open(device->channel,
1333 			 ring_size,
1334 			 ring_size,
1335 			 (void *)&props,
1336 			 sizeof(struct vmstorage_channel_properties),
1337 			 storvsc_on_channel_callback, device->channel);
1338 
1339 	if (ret != 0)
1340 		return ret;
1341 
1342 	ret = storvsc_channel_init(device, is_fc);
1343 
1344 	return ret;
1345 }
1346 
1347 static int storvsc_dev_remove(struct hv_device *device)
1348 {
1349 	struct storvsc_device *stor_device;
1350 
1351 	stor_device = hv_get_drvdata(device);
1352 
1353 	stor_device->destroy = true;
1354 
1355 	/* Make sure flag is set before waiting */
1356 	wmb();
1357 
1358 	/*
1359 	 * At this point, all outbound traffic should be disable. We
1360 	 * only allow inbound traffic (responses) to proceed so that
1361 	 * outstanding requests can be completed.
1362 	 */
1363 
1364 	storvsc_wait_to_drain(stor_device);
1365 
1366 	/*
1367 	 * Since we have already drained, we don't need to busy wait
1368 	 * as was done in final_release_stor_device()
1369 	 * Note that we cannot set the ext pointer to NULL until
1370 	 * we have drained - to drain the outgoing packets, we need to
1371 	 * allow incoming packets.
1372 	 */
1373 	hv_set_drvdata(device, NULL);
1374 
1375 	/* Close the channel */
1376 	vmbus_close(device->channel);
1377 
1378 	kfree(stor_device->stor_chns);
1379 	kfree(stor_device);
1380 	return 0;
1381 }
1382 
1383 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1384 					u16 q_num)
1385 {
1386 	u16 slot = 0;
1387 	u16 hash_qnum;
1388 	const struct cpumask *node_mask;
1389 	int num_channels, tgt_cpu;
1390 
1391 	if (stor_device->num_sc == 0) {
1392 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1393 		return stor_device->device->channel;
1394 	}
1395 
1396 	/*
1397 	 * Our channel array is sparsley populated and we
1398 	 * initiated I/O on a processor/hw-q that does not
1399 	 * currently have a designated channel. Fix this.
1400 	 * The strategy is simple:
1401 	 * I. Ensure NUMA locality
1402 	 * II. Distribute evenly (best effort)
1403 	 */
1404 
1405 	node_mask = cpumask_of_node(cpu_to_node(q_num));
1406 
1407 	num_channels = 0;
1408 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1409 		if (cpumask_test_cpu(tgt_cpu, node_mask))
1410 			num_channels++;
1411 	}
1412 	if (num_channels == 0) {
1413 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1414 		return stor_device->device->channel;
1415 	}
1416 
1417 	hash_qnum = q_num;
1418 	while (hash_qnum >= num_channels)
1419 		hash_qnum -= num_channels;
1420 
1421 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1422 		if (!cpumask_test_cpu(tgt_cpu, node_mask))
1423 			continue;
1424 		if (slot == hash_qnum)
1425 			break;
1426 		slot++;
1427 	}
1428 
1429 	stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1430 
1431 	return stor_device->stor_chns[q_num];
1432 }
1433 
1434 
1435 static int storvsc_do_io(struct hv_device *device,
1436 			 struct storvsc_cmd_request *request, u16 q_num)
1437 {
1438 	struct storvsc_device *stor_device;
1439 	struct vstor_packet *vstor_packet;
1440 	struct vmbus_channel *outgoing_channel, *channel;
1441 	unsigned long flags;
1442 	int ret = 0;
1443 	const struct cpumask *node_mask;
1444 	int tgt_cpu;
1445 
1446 	vstor_packet = &request->vstor_packet;
1447 	stor_device = get_out_stor_device(device);
1448 
1449 	if (!stor_device)
1450 		return -ENODEV;
1451 
1452 
1453 	request->device  = device;
1454 	/*
1455 	 * Select an appropriate channel to send the request out.
1456 	 */
1457 	/* See storvsc_change_target_cpu(). */
1458 	outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1459 	if (outgoing_channel != NULL) {
1460 		if (outgoing_channel->target_cpu == q_num) {
1461 			/*
1462 			 * Ideally, we want to pick a different channel if
1463 			 * available on the same NUMA node.
1464 			 */
1465 			node_mask = cpumask_of_node(cpu_to_node(q_num));
1466 			for_each_cpu_wrap(tgt_cpu,
1467 				 &stor_device->alloced_cpus, q_num + 1) {
1468 				if (!cpumask_test_cpu(tgt_cpu, node_mask))
1469 					continue;
1470 				if (tgt_cpu == q_num)
1471 					continue;
1472 				channel = READ_ONCE(
1473 					stor_device->stor_chns[tgt_cpu]);
1474 				if (channel == NULL)
1475 					continue;
1476 				if (hv_get_avail_to_write_percent(
1477 							&channel->outbound)
1478 						> ring_avail_percent_lowater) {
1479 					outgoing_channel = channel;
1480 					goto found_channel;
1481 				}
1482 			}
1483 
1484 			/*
1485 			 * All the other channels on the same NUMA node are
1486 			 * busy. Try to use the channel on the current CPU
1487 			 */
1488 			if (hv_get_avail_to_write_percent(
1489 						&outgoing_channel->outbound)
1490 					> ring_avail_percent_lowater)
1491 				goto found_channel;
1492 
1493 			/*
1494 			 * If we reach here, all the channels on the current
1495 			 * NUMA node are busy. Try to find a channel in
1496 			 * other NUMA nodes
1497 			 */
1498 			for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1499 				if (cpumask_test_cpu(tgt_cpu, node_mask))
1500 					continue;
1501 				channel = READ_ONCE(
1502 					stor_device->stor_chns[tgt_cpu]);
1503 				if (channel == NULL)
1504 					continue;
1505 				if (hv_get_avail_to_write_percent(
1506 							&channel->outbound)
1507 						> ring_avail_percent_lowater) {
1508 					outgoing_channel = channel;
1509 					goto found_channel;
1510 				}
1511 			}
1512 		}
1513 	} else {
1514 		spin_lock_irqsave(&stor_device->lock, flags);
1515 		outgoing_channel = stor_device->stor_chns[q_num];
1516 		if (outgoing_channel != NULL) {
1517 			spin_unlock_irqrestore(&stor_device->lock, flags);
1518 			goto found_channel;
1519 		}
1520 		outgoing_channel = get_og_chn(stor_device, q_num);
1521 		spin_unlock_irqrestore(&stor_device->lock, flags);
1522 	}
1523 
1524 found_channel:
1525 	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1526 
1527 	vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1528 
1529 
1530 	vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1531 
1532 
1533 	vstor_packet->vm_srb.data_transfer_length =
1534 	request->payload->range.len;
1535 
1536 	vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1537 
1538 	if (request->payload->range.len) {
1539 
1540 		ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1541 				request->payload, request->payload_sz,
1542 				vstor_packet,
1543 				sizeof(struct vstor_packet),
1544 				(unsigned long)request);
1545 	} else {
1546 		ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1547 			       sizeof(struct vstor_packet),
1548 			       (unsigned long)request,
1549 			       VM_PKT_DATA_INBAND,
1550 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1551 	}
1552 
1553 	if (ret != 0)
1554 		return ret;
1555 
1556 	atomic_inc(&stor_device->num_outstanding_req);
1557 
1558 	return ret;
1559 }
1560 
1561 static int storvsc_device_alloc(struct scsi_device *sdevice)
1562 {
1563 	/*
1564 	 * Set blist flag to permit the reading of the VPD pages even when
1565 	 * the target may claim SPC-2 compliance. MSFT targets currently
1566 	 * claim SPC-2 compliance while they implement post SPC-2 features.
1567 	 * With this flag we can correctly handle WRITE_SAME_16 issues.
1568 	 *
1569 	 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1570 	 * still supports REPORT LUN.
1571 	 */
1572 	sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1573 
1574 	return 0;
1575 }
1576 
1577 static int storvsc_device_configure(struct scsi_device *sdevice)
1578 {
1579 	blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1580 
1581 	/* storvsc devices don't support MAINTENANCE_IN SCSI cmd */
1582 	sdevice->no_report_opcodes = 1;
1583 	sdevice->no_write_same = 1;
1584 
1585 	/*
1586 	 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1587 	 * if the device is a MSFT virtual device.  If the host is
1588 	 * WIN10 or newer, allow write_same.
1589 	 */
1590 	if (!strncmp(sdevice->vendor, "Msft", 4)) {
1591 		switch (vmstor_proto_version) {
1592 		case VMSTOR_PROTO_VERSION_WIN8:
1593 		case VMSTOR_PROTO_VERSION_WIN8_1:
1594 			sdevice->scsi_level = SCSI_SPC_3;
1595 			break;
1596 		}
1597 
1598 		if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1599 			sdevice->no_write_same = 0;
1600 	}
1601 
1602 	return 0;
1603 }
1604 
1605 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1606 			   sector_t capacity, int *info)
1607 {
1608 	sector_t nsect = capacity;
1609 	sector_t cylinders = nsect;
1610 	int heads, sectors_pt;
1611 
1612 	/*
1613 	 * We are making up these values; let us keep it simple.
1614 	 */
1615 	heads = 0xff;
1616 	sectors_pt = 0x3f;      /* Sectors per track */
1617 	sector_div(cylinders, heads * sectors_pt);
1618 	if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1619 		cylinders = 0xffff;
1620 
1621 	info[0] = heads;
1622 	info[1] = sectors_pt;
1623 	info[2] = (int)cylinders;
1624 
1625 	return 0;
1626 }
1627 
1628 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1629 {
1630 	struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1631 	struct hv_device *device = host_dev->dev;
1632 
1633 	struct storvsc_device *stor_device;
1634 	struct storvsc_cmd_request *request;
1635 	struct vstor_packet *vstor_packet;
1636 	int ret, t;
1637 
1638 	stor_device = get_out_stor_device(device);
1639 	if (!stor_device)
1640 		return FAILED;
1641 
1642 	request = &stor_device->reset_request;
1643 	vstor_packet = &request->vstor_packet;
1644 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1645 
1646 	init_completion(&request->wait_event);
1647 
1648 	vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1649 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1650 	vstor_packet->vm_srb.path_id = stor_device->path_id;
1651 
1652 	ret = vmbus_sendpacket(device->channel, vstor_packet,
1653 			       sizeof(struct vstor_packet),
1654 			       VMBUS_RQST_RESET,
1655 			       VM_PKT_DATA_INBAND,
1656 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1657 	if (ret != 0)
1658 		return FAILED;
1659 
1660 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1661 	if (t == 0)
1662 		return TIMEOUT_ERROR;
1663 
1664 
1665 	/*
1666 	 * At this point, all outstanding requests in the adapter
1667 	 * should have been flushed out and return to us
1668 	 * There is a potential race here where the host may be in
1669 	 * the process of responding when we return from here.
1670 	 * Just wait for all in-transit packets to be accounted for
1671 	 * before we return from here.
1672 	 */
1673 	storvsc_wait_to_drain(stor_device);
1674 
1675 	return SUCCESS;
1676 }
1677 
1678 /*
1679  * The host guarantees to respond to each command, although I/O latencies might
1680  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1681  * chance to perform EH.
1682  */
1683 static enum scsi_timeout_action storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1684 {
1685 	return SCSI_EH_RESET_TIMER;
1686 }
1687 
1688 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1689 {
1690 	bool allowed = true;
1691 	u8 scsi_op = scmnd->cmnd[0];
1692 
1693 	switch (scsi_op) {
1694 	/* the host does not handle WRITE_SAME, log accident usage */
1695 	case WRITE_SAME:
1696 	/*
1697 	 * smartd sends this command and the host does not handle
1698 	 * this. So, don't send it.
1699 	 */
1700 	case SET_WINDOW:
1701 		set_host_byte(scmnd, DID_ERROR);
1702 		allowed = false;
1703 		break;
1704 	default:
1705 		break;
1706 	}
1707 	return allowed;
1708 }
1709 
1710 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1711 {
1712 	int ret;
1713 	struct hv_host_device *host_dev = shost_priv(host);
1714 	struct hv_device *dev = host_dev->dev;
1715 	struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1716 	struct scatterlist *sgl;
1717 	struct vmscsi_request *vm_srb;
1718 	struct vmbus_packet_mpb_array  *payload;
1719 	u32 payload_sz;
1720 	u32 length;
1721 
1722 	if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1723 		/*
1724 		 * On legacy hosts filter unimplemented commands.
1725 		 * Future hosts are expected to correctly handle
1726 		 * unsupported commands. Furthermore, it is
1727 		 * possible that some of the currently
1728 		 * unsupported commands maybe supported in
1729 		 * future versions of the host.
1730 		 */
1731 		if (!storvsc_scsi_cmd_ok(scmnd)) {
1732 			scsi_done(scmnd);
1733 			return 0;
1734 		}
1735 	}
1736 
1737 	/* Setup the cmd request */
1738 	cmd_request->cmd = scmnd;
1739 
1740 	memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1741 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1742 	vm_srb->time_out_value = 60;
1743 
1744 	vm_srb->srb_flags |=
1745 		SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1746 
1747 	if (scmnd->device->tagged_supported) {
1748 		vm_srb->srb_flags |=
1749 		(SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1750 		vm_srb->queue_tag = SP_UNTAGGED;
1751 		vm_srb->queue_action = SRB_SIMPLE_TAG_REQUEST;
1752 	}
1753 
1754 	/* Build the SRB */
1755 	switch (scmnd->sc_data_direction) {
1756 	case DMA_TO_DEVICE:
1757 		vm_srb->data_in = WRITE_TYPE;
1758 		vm_srb->srb_flags |= SRB_FLAGS_DATA_OUT;
1759 		break;
1760 	case DMA_FROM_DEVICE:
1761 		vm_srb->data_in = READ_TYPE;
1762 		vm_srb->srb_flags |= SRB_FLAGS_DATA_IN;
1763 		break;
1764 	case DMA_NONE:
1765 		vm_srb->data_in = UNKNOWN_TYPE;
1766 		vm_srb->srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1767 		break;
1768 	default:
1769 		/*
1770 		 * This is DMA_BIDIRECTIONAL or something else we are never
1771 		 * supposed to see here.
1772 		 */
1773 		WARN(1, "Unexpected data direction: %d\n",
1774 		     scmnd->sc_data_direction);
1775 		return -EINVAL;
1776 	}
1777 
1778 
1779 	vm_srb->port_number = host_dev->port;
1780 	vm_srb->path_id = scmnd->device->channel;
1781 	vm_srb->target_id = scmnd->device->id;
1782 	vm_srb->lun = scmnd->device->lun;
1783 
1784 	vm_srb->cdb_length = scmnd->cmd_len;
1785 
1786 	memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1787 
1788 	sgl = (struct scatterlist *)scsi_sglist(scmnd);
1789 
1790 	length = scsi_bufflen(scmnd);
1791 	payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1792 	payload_sz = 0;
1793 
1794 	if (scsi_sg_count(scmnd)) {
1795 		unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1796 		unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1797 		struct scatterlist *sg;
1798 		unsigned long hvpfn, hvpfns_to_add;
1799 		int j, i = 0, sg_count;
1800 
1801 		payload_sz = (hvpg_count * sizeof(u64) +
1802 			      sizeof(struct vmbus_packet_mpb_array));
1803 
1804 		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1805 			payload = kzalloc(payload_sz, GFP_ATOMIC);
1806 			if (!payload)
1807 				return SCSI_MLQUEUE_DEVICE_BUSY;
1808 		}
1809 
1810 		payload->range.len = length;
1811 		payload->range.offset = offset_in_hvpg;
1812 
1813 		sg_count = scsi_dma_map(scmnd);
1814 		if (sg_count < 0) {
1815 			ret = SCSI_MLQUEUE_DEVICE_BUSY;
1816 			goto err_free_payload;
1817 		}
1818 
1819 		for_each_sg(sgl, sg, sg_count, j) {
1820 			/*
1821 			 * Init values for the current sgl entry. hvpfns_to_add
1822 			 * is in units of Hyper-V size pages. Handling the
1823 			 * PAGE_SIZE != HV_HYP_PAGE_SIZE case also handles
1824 			 * values of sgl->offset that are larger than PAGE_SIZE.
1825 			 * Such offsets are handled even on other than the first
1826 			 * sgl entry, provided they are a multiple of PAGE_SIZE.
1827 			 */
1828 			hvpfn = HVPFN_DOWN(sg_dma_address(sg));
1829 			hvpfns_to_add = HVPFN_UP(sg_dma_address(sg) +
1830 						 sg_dma_len(sg)) - hvpfn;
1831 
1832 			/*
1833 			 * Fill the next portion of the PFN array with
1834 			 * sequential Hyper-V PFNs for the continguous physical
1835 			 * memory described by the sgl entry. The end of the
1836 			 * last sgl should be reached at the same time that
1837 			 * the PFN array is filled.
1838 			 */
1839 			while (hvpfns_to_add--)
1840 				payload->range.pfn_array[i++] = hvpfn++;
1841 		}
1842 	}
1843 
1844 	cmd_request->payload = payload;
1845 	cmd_request->payload_sz = payload_sz;
1846 
1847 	/* Invokes the vsc to start an IO */
1848 	ret = storvsc_do_io(dev, cmd_request, get_cpu());
1849 	put_cpu();
1850 
1851 	if (ret)
1852 		scsi_dma_unmap(scmnd);
1853 
1854 	if (ret == -EAGAIN) {
1855 		/* no more space */
1856 		ret = SCSI_MLQUEUE_DEVICE_BUSY;
1857 		goto err_free_payload;
1858 	}
1859 
1860 	return 0;
1861 
1862 err_free_payload:
1863 	if (payload_sz > sizeof(cmd_request->mpb))
1864 		kfree(payload);
1865 
1866 	return ret;
1867 }
1868 
1869 static struct scsi_host_template scsi_driver = {
1870 	.module	=		THIS_MODULE,
1871 	.name =			"storvsc_host_t",
1872 	.cmd_size =             sizeof(struct storvsc_cmd_request),
1873 	.bios_param =		storvsc_get_chs,
1874 	.queuecommand =		storvsc_queuecommand,
1875 	.eh_host_reset_handler =	storvsc_host_reset_handler,
1876 	.proc_name =		"storvsc_host",
1877 	.eh_timed_out =		storvsc_eh_timed_out,
1878 	.slave_alloc =		storvsc_device_alloc,
1879 	.slave_configure =	storvsc_device_configure,
1880 	.cmd_per_lun =		2048,
1881 	.this_id =		-1,
1882 	/* Ensure there are no gaps in presented sgls */
1883 	.virt_boundary_mask =	HV_HYP_PAGE_SIZE - 1,
1884 	.no_write_same =	1,
1885 	.track_queue_depth =	1,
1886 	.change_queue_depth =	storvsc_change_queue_depth,
1887 };
1888 
1889 enum {
1890 	SCSI_GUID,
1891 	IDE_GUID,
1892 	SFC_GUID,
1893 };
1894 
1895 static const struct hv_vmbus_device_id id_table[] = {
1896 	/* SCSI guid */
1897 	{ HV_SCSI_GUID,
1898 	  .driver_data = SCSI_GUID
1899 	},
1900 	/* IDE guid */
1901 	{ HV_IDE_GUID,
1902 	  .driver_data = IDE_GUID
1903 	},
1904 	/* Fibre Channel GUID */
1905 	{
1906 	  HV_SYNTHFC_GUID,
1907 	  .driver_data = SFC_GUID
1908 	},
1909 	{ },
1910 };
1911 
1912 MODULE_DEVICE_TABLE(vmbus, id_table);
1913 
1914 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1915 
1916 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1917 {
1918 	return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1919 }
1920 
1921 static int storvsc_probe(struct hv_device *device,
1922 			const struct hv_vmbus_device_id *dev_id)
1923 {
1924 	int ret;
1925 	int num_cpus = num_online_cpus();
1926 	int num_present_cpus = num_present_cpus();
1927 	struct Scsi_Host *host;
1928 	struct hv_host_device *host_dev;
1929 	bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1930 	bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1931 	int target = 0;
1932 	struct storvsc_device *stor_device;
1933 	int max_sub_channels = 0;
1934 	u32 max_xfer_bytes;
1935 
1936 	/*
1937 	 * We support sub-channels for storage on SCSI and FC controllers.
1938 	 * The number of sub-channels offerred is based on the number of
1939 	 * VCPUs in the guest.
1940 	 */
1941 	if (!dev_is_ide)
1942 		max_sub_channels =
1943 			(num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1944 
1945 	scsi_driver.can_queue = max_outstanding_req_per_channel *
1946 				(max_sub_channels + 1) *
1947 				(100 - ring_avail_percent_lowater) / 100;
1948 
1949 	host = scsi_host_alloc(&scsi_driver,
1950 			       sizeof(struct hv_host_device));
1951 	if (!host)
1952 		return -ENOMEM;
1953 
1954 	host_dev = shost_priv(host);
1955 	memset(host_dev, 0, sizeof(struct hv_host_device));
1956 
1957 	host_dev->port = host->host_no;
1958 	host_dev->dev = device;
1959 	host_dev->host = host;
1960 
1961 
1962 	stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1963 	if (!stor_device) {
1964 		ret = -ENOMEM;
1965 		goto err_out0;
1966 	}
1967 
1968 	stor_device->destroy = false;
1969 	init_waitqueue_head(&stor_device->waiting_to_drain);
1970 	stor_device->device = device;
1971 	stor_device->host = host;
1972 	spin_lock_init(&stor_device->lock);
1973 	hv_set_drvdata(device, stor_device);
1974 	dma_set_min_align_mask(&device->device, HV_HYP_PAGE_SIZE - 1);
1975 
1976 	stor_device->port_number = host->host_no;
1977 	ret = storvsc_connect_to_vsp(device, aligned_ringbuffer_size, is_fc);
1978 	if (ret)
1979 		goto err_out1;
1980 
1981 	host_dev->path = stor_device->path_id;
1982 	host_dev->target = stor_device->target_id;
1983 
1984 	switch (dev_id->driver_data) {
1985 	case SFC_GUID:
1986 		host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1987 		host->max_id = STORVSC_FC_MAX_TARGETS;
1988 		host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1989 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1990 		host->transportt = fc_transport_template;
1991 #endif
1992 		break;
1993 
1994 	case SCSI_GUID:
1995 		host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1996 		host->max_id = STORVSC_MAX_TARGETS;
1997 		host->max_channel = STORVSC_MAX_CHANNELS - 1;
1998 		break;
1999 
2000 	default:
2001 		host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
2002 		host->max_id = STORVSC_IDE_MAX_TARGETS;
2003 		host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2004 		break;
2005 	}
2006 	/* max cmd length */
2007 	host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2008 	/*
2009 	 * Any reasonable Hyper-V configuration should provide
2010 	 * max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
2011 	 * protecting it from any weird value.
2012 	 */
2013 	max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_SIZE);
2014 	if (is_fc)
2015 		max_xfer_bytes = min(max_xfer_bytes, STORVSC_FC_MAX_XFER_SIZE);
2016 
2017 	/* max_hw_sectors_kb */
2018 	host->max_sectors = max_xfer_bytes >> 9;
2019 	/*
2020 	 * There are 2 requirements for Hyper-V storvsc sgl segments,
2021 	 * based on which the below calculation for max segments is
2022 	 * done:
2023 	 *
2024 	 * 1. Except for the first and last sgl segment, all sgl segments
2025 	 *    should be align to HV_HYP_PAGE_SIZE, that also means the
2026 	 *    maximum number of segments in a sgl can be calculated by
2027 	 *    dividing the total max transfer length by HV_HYP_PAGE_SIZE.
2028 	 *
2029 	 * 2. Except for the first and last, each entry in the SGL must
2030 	 *    have an offset that is a multiple of HV_HYP_PAGE_SIZE.
2031 	 */
2032 	host->sg_tablesize = (max_xfer_bytes >> HV_HYP_PAGE_SHIFT) + 1;
2033 	/*
2034 	 * For non-IDE disks, the host supports multiple channels.
2035 	 * Set the number of HW queues we are supporting.
2036 	 */
2037 	if (!dev_is_ide) {
2038 		if (storvsc_max_hw_queues > num_present_cpus) {
2039 			storvsc_max_hw_queues = 0;
2040 			storvsc_log(device, STORVSC_LOGGING_WARN,
2041 				"Resetting invalid storvsc_max_hw_queues value to default.\n");
2042 		}
2043 		if (storvsc_max_hw_queues)
2044 			host->nr_hw_queues = storvsc_max_hw_queues;
2045 		else
2046 			host->nr_hw_queues = num_present_cpus;
2047 	}
2048 
2049 	/*
2050 	 * Set the error handler work queue.
2051 	 */
2052 	host_dev->handle_error_wq =
2053 			alloc_ordered_workqueue("storvsc_error_wq_%d",
2054 						0,
2055 						host->host_no);
2056 	if (!host_dev->handle_error_wq) {
2057 		ret = -ENOMEM;
2058 		goto err_out2;
2059 	}
2060 	INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2061 	/* Register the HBA and start the scsi bus scan */
2062 	ret = scsi_add_host(host, &device->device);
2063 	if (ret != 0)
2064 		goto err_out3;
2065 
2066 	if (!dev_is_ide) {
2067 		scsi_scan_host(host);
2068 	} else {
2069 		target = (device->dev_instance.b[5] << 8 |
2070 			 device->dev_instance.b[4]);
2071 		ret = scsi_add_device(host, 0, target, 0);
2072 		if (ret)
2073 			goto err_out4;
2074 	}
2075 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2076 	if (host->transportt == fc_transport_template) {
2077 		struct fc_rport_identifiers ids = {
2078 			.roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2079 		};
2080 
2081 		fc_host_node_name(host) = stor_device->node_name;
2082 		fc_host_port_name(host) = stor_device->port_name;
2083 		stor_device->rport = fc_remote_port_add(host, 0, &ids);
2084 		if (!stor_device->rport) {
2085 			ret = -ENOMEM;
2086 			goto err_out4;
2087 		}
2088 	}
2089 #endif
2090 	return 0;
2091 
2092 err_out4:
2093 	scsi_remove_host(host);
2094 
2095 err_out3:
2096 	destroy_workqueue(host_dev->handle_error_wq);
2097 
2098 err_out2:
2099 	/*
2100 	 * Once we have connected with the host, we would need to
2101 	 * invoke storvsc_dev_remove() to rollback this state and
2102 	 * this call also frees up the stor_device; hence the jump around
2103 	 * err_out1 label.
2104 	 */
2105 	storvsc_dev_remove(device);
2106 	goto err_out0;
2107 
2108 err_out1:
2109 	kfree(stor_device->stor_chns);
2110 	kfree(stor_device);
2111 
2112 err_out0:
2113 	scsi_host_put(host);
2114 	return ret;
2115 }
2116 
2117 /* Change a scsi target's queue depth */
2118 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2119 {
2120 	if (queue_depth > scsi_driver.can_queue)
2121 		queue_depth = scsi_driver.can_queue;
2122 
2123 	return scsi_change_queue_depth(sdev, queue_depth);
2124 }
2125 
2126 static void storvsc_remove(struct hv_device *dev)
2127 {
2128 	struct storvsc_device *stor_device = hv_get_drvdata(dev);
2129 	struct Scsi_Host *host = stor_device->host;
2130 	struct hv_host_device *host_dev = shost_priv(host);
2131 
2132 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2133 	if (host->transportt == fc_transport_template) {
2134 		fc_remote_port_delete(stor_device->rport);
2135 		fc_remove_host(host);
2136 	}
2137 #endif
2138 	destroy_workqueue(host_dev->handle_error_wq);
2139 	scsi_remove_host(host);
2140 	storvsc_dev_remove(dev);
2141 	scsi_host_put(host);
2142 }
2143 
2144 static int storvsc_suspend(struct hv_device *hv_dev)
2145 {
2146 	struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2147 	struct Scsi_Host *host = stor_device->host;
2148 	struct hv_host_device *host_dev = shost_priv(host);
2149 
2150 	storvsc_wait_to_drain(stor_device);
2151 
2152 	drain_workqueue(host_dev->handle_error_wq);
2153 
2154 	vmbus_close(hv_dev->channel);
2155 
2156 	kfree(stor_device->stor_chns);
2157 	stor_device->stor_chns = NULL;
2158 
2159 	cpumask_clear(&stor_device->alloced_cpus);
2160 
2161 	return 0;
2162 }
2163 
2164 static int storvsc_resume(struct hv_device *hv_dev)
2165 {
2166 	int ret;
2167 
2168 	ret = storvsc_connect_to_vsp(hv_dev, aligned_ringbuffer_size,
2169 				     hv_dev_is_fc(hv_dev));
2170 	return ret;
2171 }
2172 
2173 static struct hv_driver storvsc_drv = {
2174 	.name = KBUILD_MODNAME,
2175 	.id_table = id_table,
2176 	.probe = storvsc_probe,
2177 	.remove = storvsc_remove,
2178 	.suspend = storvsc_suspend,
2179 	.resume = storvsc_resume,
2180 	.driver = {
2181 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
2182 	},
2183 };
2184 
2185 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2186 static struct fc_function_template fc_transport_functions = {
2187 	.show_host_node_name = 1,
2188 	.show_host_port_name = 1,
2189 };
2190 #endif
2191 
2192 static int __init storvsc_drv_init(void)
2193 {
2194 	int ret;
2195 
2196 	/*
2197 	 * Divide the ring buffer data size (which is 1 page less
2198 	 * than the ring buffer size since that page is reserved for
2199 	 * the ring buffer indices) by the max request size (which is
2200 	 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2201 	 */
2202 	aligned_ringbuffer_size = VMBUS_RING_SIZE(storvsc_ringbuffer_size);
2203 	max_outstanding_req_per_channel =
2204 		((aligned_ringbuffer_size - PAGE_SIZE) /
2205 		ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2206 		sizeof(struct vstor_packet) + sizeof(u64),
2207 		sizeof(u64)));
2208 
2209 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2210 	fc_transport_template = fc_attach_transport(&fc_transport_functions);
2211 	if (!fc_transport_template)
2212 		return -ENODEV;
2213 #endif
2214 
2215 	ret = vmbus_driver_register(&storvsc_drv);
2216 
2217 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2218 	if (ret)
2219 		fc_release_transport(fc_transport_template);
2220 #endif
2221 
2222 	return ret;
2223 }
2224 
2225 static void __exit storvsc_drv_exit(void)
2226 {
2227 	vmbus_driver_unregister(&storvsc_drv);
2228 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2229 	fc_release_transport(fc_transport_template);
2230 #endif
2231 }
2232 
2233 MODULE_LICENSE("GPL");
2234 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2235 module_init(storvsc_drv_init);
2236 module_exit(storvsc_drv_exit);
2237