xref: /openbmc/linux/drivers/scsi/storvsc_drv.c (revision 0352f880)
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; a subset of the codes used here.
307  */
308 
309 #define SRB_STATUS_AUTOSENSE_VALID	0x80
310 #define SRB_STATUS_QUEUE_FROZEN		0x40
311 #define SRB_STATUS_INVALID_LUN	0x20
312 #define SRB_STATUS_SUCCESS	0x01
313 #define SRB_STATUS_ABORTED	0x02
314 #define SRB_STATUS_ERROR	0x04
315 #define SRB_STATUS_DATA_OVERRUN	0x12
316 
317 #define SRB_STATUS(status) \
318 	(status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
319 /*
320  * This is the end of Protocol specific defines.
321  */
322 
323 static int storvsc_ringbuffer_size = (128 * 1024);
324 static u32 max_outstanding_req_per_channel;
325 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
326 
327 static int storvsc_vcpus_per_sub_channel = 4;
328 static unsigned int storvsc_max_hw_queues;
329 
330 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
331 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
332 
333 module_param(storvsc_max_hw_queues, uint, 0644);
334 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
335 
336 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
337 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
338 
339 static int ring_avail_percent_lowater = 10;
340 module_param(ring_avail_percent_lowater, int, S_IRUGO);
341 MODULE_PARM_DESC(ring_avail_percent_lowater,
342 		"Select a channel if available ring size > this in percent");
343 
344 /*
345  * Timeout in seconds for all devices managed by this driver.
346  */
347 static int storvsc_timeout = 180;
348 
349 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
350 static struct scsi_transport_template *fc_transport_template;
351 #endif
352 
353 static struct scsi_host_template scsi_driver;
354 static void storvsc_on_channel_callback(void *context);
355 
356 #define STORVSC_MAX_LUNS_PER_TARGET			255
357 #define STORVSC_MAX_TARGETS				2
358 #define STORVSC_MAX_CHANNELS				8
359 
360 #define STORVSC_FC_MAX_LUNS_PER_TARGET			255
361 #define STORVSC_FC_MAX_TARGETS				128
362 #define STORVSC_FC_MAX_CHANNELS				8
363 
364 #define STORVSC_IDE_MAX_LUNS_PER_TARGET			64
365 #define STORVSC_IDE_MAX_TARGETS				1
366 #define STORVSC_IDE_MAX_CHANNELS			1
367 
368 /*
369  * Upper bound on the size of a storvsc packet.
370  */
371 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
372 			      sizeof(struct vstor_packet))
373 
374 struct storvsc_cmd_request {
375 	struct scsi_cmnd *cmd;
376 
377 	struct hv_device *device;
378 
379 	/* Synchronize the request/response if needed */
380 	struct completion wait_event;
381 
382 	struct vmbus_channel_packet_multipage_buffer mpb;
383 	struct vmbus_packet_mpb_array *payload;
384 	u32 payload_sz;
385 
386 	struct vstor_packet vstor_packet;
387 };
388 
389 
390 /* A storvsc device is a device object that contains a vmbus channel */
391 struct storvsc_device {
392 	struct hv_device *device;
393 
394 	bool	 destroy;
395 	bool	 drain_notify;
396 	atomic_t num_outstanding_req;
397 	struct Scsi_Host *host;
398 
399 	wait_queue_head_t waiting_to_drain;
400 
401 	/*
402 	 * Each unique Port/Path/Target represents 1 channel ie scsi
403 	 * controller. In reality, the pathid, targetid is always 0
404 	 * and the port is set by us
405 	 */
406 	unsigned int port_number;
407 	unsigned char path_id;
408 	unsigned char target_id;
409 
410 	/*
411 	 * Max I/O, the device can support.
412 	 */
413 	u32   max_transfer_bytes;
414 	/*
415 	 * Number of sub-channels we will open.
416 	 */
417 	u16 num_sc;
418 	struct vmbus_channel **stor_chns;
419 	/*
420 	 * Mask of CPUs bound to subchannels.
421 	 */
422 	struct cpumask alloced_cpus;
423 	/*
424 	 * Serializes modifications of stor_chns[] from storvsc_do_io()
425 	 * and storvsc_change_target_cpu().
426 	 */
427 	spinlock_t lock;
428 	/* Used for vsc/vsp channel reset process */
429 	struct storvsc_cmd_request init_request;
430 	struct storvsc_cmd_request reset_request;
431 	/*
432 	 * Currently active port and node names for FC devices.
433 	 */
434 	u64 node_name;
435 	u64 port_name;
436 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
437 	struct fc_rport *rport;
438 #endif
439 };
440 
441 struct hv_host_device {
442 	struct hv_device *dev;
443 	unsigned int port;
444 	unsigned char path;
445 	unsigned char target;
446 	struct workqueue_struct *handle_error_wq;
447 	struct work_struct host_scan_work;
448 	struct Scsi_Host *host;
449 };
450 
451 struct storvsc_scan_work {
452 	struct work_struct work;
453 	struct Scsi_Host *host;
454 	u8 lun;
455 	u8 tgt_id;
456 };
457 
458 static void storvsc_device_scan(struct work_struct *work)
459 {
460 	struct storvsc_scan_work *wrk;
461 	struct scsi_device *sdev;
462 
463 	wrk = container_of(work, struct storvsc_scan_work, work);
464 
465 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
466 	if (!sdev)
467 		goto done;
468 	scsi_rescan_device(&sdev->sdev_gendev);
469 	scsi_device_put(sdev);
470 
471 done:
472 	kfree(wrk);
473 }
474 
475 static void storvsc_host_scan(struct work_struct *work)
476 {
477 	struct Scsi_Host *host;
478 	struct scsi_device *sdev;
479 	struct hv_host_device *host_device =
480 		container_of(work, struct hv_host_device, host_scan_work);
481 
482 	host = host_device->host;
483 	/*
484 	 * Before scanning the host, first check to see if any of the
485 	 * currently known devices have been hot removed. We issue a
486 	 * "unit ready" command against all currently known devices.
487 	 * This I/O will result in an error for devices that have been
488 	 * removed. As part of handling the I/O error, we remove the device.
489 	 *
490 	 * When a LUN is added or removed, the host sends us a signal to
491 	 * scan the host. Thus we are forced to discover the LUNs that
492 	 * may have been removed this way.
493 	 */
494 	mutex_lock(&host->scan_mutex);
495 	shost_for_each_device(sdev, host)
496 		scsi_test_unit_ready(sdev, 1, 1, NULL);
497 	mutex_unlock(&host->scan_mutex);
498 	/*
499 	 * Now scan the host to discover LUNs that may have been added.
500 	 */
501 	scsi_scan_host(host);
502 }
503 
504 static void storvsc_remove_lun(struct work_struct *work)
505 {
506 	struct storvsc_scan_work *wrk;
507 	struct scsi_device *sdev;
508 
509 	wrk = container_of(work, struct storvsc_scan_work, work);
510 	if (!scsi_host_get(wrk->host))
511 		goto done;
512 
513 	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
514 
515 	if (sdev) {
516 		scsi_remove_device(sdev);
517 		scsi_device_put(sdev);
518 	}
519 	scsi_host_put(wrk->host);
520 
521 done:
522 	kfree(wrk);
523 }
524 
525 
526 /*
527  * We can get incoming messages from the host that are not in response to
528  * messages that we have sent out. An example of this would be messages
529  * received by the guest to notify dynamic addition/removal of LUNs. To
530  * deal with potential race conditions where the driver may be in the
531  * midst of being unloaded when we might receive an unsolicited message
532  * from the host, we have implemented a mechanism to gurantee sequential
533  * consistency:
534  *
535  * 1) Once the device is marked as being destroyed, we will fail all
536  *    outgoing messages.
537  * 2) We permit incoming messages when the device is being destroyed,
538  *    only to properly account for messages already sent out.
539  */
540 
541 static inline struct storvsc_device *get_out_stor_device(
542 					struct hv_device *device)
543 {
544 	struct storvsc_device *stor_device;
545 
546 	stor_device = hv_get_drvdata(device);
547 
548 	if (stor_device && stor_device->destroy)
549 		stor_device = NULL;
550 
551 	return stor_device;
552 }
553 
554 
555 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
556 {
557 	dev->drain_notify = true;
558 	wait_event(dev->waiting_to_drain,
559 		   atomic_read(&dev->num_outstanding_req) == 0);
560 	dev->drain_notify = false;
561 }
562 
563 static inline struct storvsc_device *get_in_stor_device(
564 					struct hv_device *device)
565 {
566 	struct storvsc_device *stor_device;
567 
568 	stor_device = hv_get_drvdata(device);
569 
570 	if (!stor_device)
571 		goto get_in_err;
572 
573 	/*
574 	 * If the device is being destroyed; allow incoming
575 	 * traffic only to cleanup outstanding requests.
576 	 */
577 
578 	if (stor_device->destroy  &&
579 		(atomic_read(&stor_device->num_outstanding_req) == 0))
580 		stor_device = NULL;
581 
582 get_in_err:
583 	return stor_device;
584 
585 }
586 
587 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
588 				      u32 new)
589 {
590 	struct storvsc_device *stor_device;
591 	struct vmbus_channel *cur_chn;
592 	bool old_is_alloced = false;
593 	struct hv_device *device;
594 	unsigned long flags;
595 	int cpu;
596 
597 	device = channel->primary_channel ?
598 			channel->primary_channel->device_obj
599 				: channel->device_obj;
600 	stor_device = get_out_stor_device(device);
601 	if (!stor_device)
602 		return;
603 
604 	/* See storvsc_do_io() -> get_og_chn(). */
605 	spin_lock_irqsave(&stor_device->lock, flags);
606 
607 	/*
608 	 * Determines if the storvsc device has other channels assigned to
609 	 * the "old" CPU to update the alloced_cpus mask and the stor_chns
610 	 * array.
611 	 */
612 	if (device->channel != channel && device->channel->target_cpu == old) {
613 		cur_chn = device->channel;
614 		old_is_alloced = true;
615 		goto old_is_alloced;
616 	}
617 	list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
618 		if (cur_chn == channel)
619 			continue;
620 		if (cur_chn->target_cpu == old) {
621 			old_is_alloced = true;
622 			goto old_is_alloced;
623 		}
624 	}
625 
626 old_is_alloced:
627 	if (old_is_alloced)
628 		WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
629 	else
630 		cpumask_clear_cpu(old, &stor_device->alloced_cpus);
631 
632 	/* "Flush" the stor_chns array. */
633 	for_each_possible_cpu(cpu) {
634 		if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
635 					cpu, &stor_device->alloced_cpus))
636 			WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
637 	}
638 
639 	WRITE_ONCE(stor_device->stor_chns[new], channel);
640 	cpumask_set_cpu(new, &stor_device->alloced_cpus);
641 
642 	spin_unlock_irqrestore(&stor_device->lock, flags);
643 }
644 
645 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
646 {
647 	struct storvsc_cmd_request *request =
648 		(struct storvsc_cmd_request *)(unsigned long)rqst_addr;
649 
650 	if (rqst_addr == VMBUS_RQST_INIT)
651 		return VMBUS_RQST_INIT;
652 	if (rqst_addr == VMBUS_RQST_RESET)
653 		return VMBUS_RQST_RESET;
654 
655 	/*
656 	 * Cannot return an ID of 0, which is reserved for an unsolicited
657 	 * message from Hyper-V.
658 	 */
659 	return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
660 }
661 
662 static void handle_sc_creation(struct vmbus_channel *new_sc)
663 {
664 	struct hv_device *device = new_sc->primary_channel->device_obj;
665 	struct device *dev = &device->device;
666 	struct storvsc_device *stor_device;
667 	struct vmstorage_channel_properties props;
668 	int ret;
669 
670 	stor_device = get_out_stor_device(device);
671 	if (!stor_device)
672 		return;
673 
674 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
675 	new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
676 
677 	new_sc->next_request_id_callback = storvsc_next_request_id;
678 
679 	ret = vmbus_open(new_sc,
680 			 storvsc_ringbuffer_size,
681 			 storvsc_ringbuffer_size,
682 			 (void *)&props,
683 			 sizeof(struct vmstorage_channel_properties),
684 			 storvsc_on_channel_callback, new_sc);
685 
686 	/* In case vmbus_open() fails, we don't use the sub-channel. */
687 	if (ret != 0) {
688 		dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
689 		return;
690 	}
691 
692 	new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
693 
694 	/* Add the sub-channel to the array of available channels. */
695 	stor_device->stor_chns[new_sc->target_cpu] = new_sc;
696 	cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
697 }
698 
699 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
700 {
701 	struct device *dev = &device->device;
702 	struct storvsc_device *stor_device;
703 	int num_sc;
704 	struct storvsc_cmd_request *request;
705 	struct vstor_packet *vstor_packet;
706 	int ret, t;
707 
708 	/*
709 	 * If the number of CPUs is artificially restricted, such as
710 	 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
711 	 * sub-channels >= the number of CPUs. These sub-channels
712 	 * should not be created. The primary channel is already created
713 	 * and assigned to one CPU, so check against # CPUs - 1.
714 	 */
715 	num_sc = min((int)(num_online_cpus() - 1), max_chns);
716 	if (!num_sc)
717 		return;
718 
719 	stor_device = get_out_stor_device(device);
720 	if (!stor_device)
721 		return;
722 
723 	stor_device->num_sc = num_sc;
724 	request = &stor_device->init_request;
725 	vstor_packet = &request->vstor_packet;
726 
727 	/*
728 	 * Establish a handler for dealing with subchannels.
729 	 */
730 	vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
731 
732 	/*
733 	 * Request the host to create sub-channels.
734 	 */
735 	memset(request, 0, sizeof(struct storvsc_cmd_request));
736 	init_completion(&request->wait_event);
737 	vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
738 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
739 	vstor_packet->sub_channel_count = num_sc;
740 
741 	ret = vmbus_sendpacket(device->channel, vstor_packet,
742 			       sizeof(struct vstor_packet),
743 			       VMBUS_RQST_INIT,
744 			       VM_PKT_DATA_INBAND,
745 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
746 
747 	if (ret != 0) {
748 		dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
749 		return;
750 	}
751 
752 	t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
753 	if (t == 0) {
754 		dev_err(dev, "Failed to create sub-channel: timed out\n");
755 		return;
756 	}
757 
758 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
759 	    vstor_packet->status != 0) {
760 		dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
761 			vstor_packet->operation, vstor_packet->status);
762 		return;
763 	}
764 
765 	/*
766 	 * We need to do nothing here, because vmbus_process_offer()
767 	 * invokes channel->sc_creation_callback, which will open and use
768 	 * the sub-channel(s).
769 	 */
770 }
771 
772 static void cache_wwn(struct storvsc_device *stor_device,
773 		      struct vstor_packet *vstor_packet)
774 {
775 	/*
776 	 * Cache the currently active port and node ww names.
777 	 */
778 	if (vstor_packet->wwn_packet.primary_active) {
779 		stor_device->node_name =
780 			wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
781 		stor_device->port_name =
782 			wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
783 	} else {
784 		stor_device->node_name =
785 			wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
786 		stor_device->port_name =
787 			wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
788 	}
789 }
790 
791 
792 static int storvsc_execute_vstor_op(struct hv_device *device,
793 				    struct storvsc_cmd_request *request,
794 				    bool status_check)
795 {
796 	struct storvsc_device *stor_device;
797 	struct vstor_packet *vstor_packet;
798 	int ret, t;
799 
800 	stor_device = get_out_stor_device(device);
801 	if (!stor_device)
802 		return -ENODEV;
803 
804 	vstor_packet = &request->vstor_packet;
805 
806 	init_completion(&request->wait_event);
807 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
808 
809 	ret = vmbus_sendpacket(device->channel, vstor_packet,
810 			       sizeof(struct vstor_packet),
811 			       VMBUS_RQST_INIT,
812 			       VM_PKT_DATA_INBAND,
813 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
814 	if (ret != 0)
815 		return ret;
816 
817 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
818 	if (t == 0)
819 		return -ETIMEDOUT;
820 
821 	if (!status_check)
822 		return ret;
823 
824 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
825 	    vstor_packet->status != 0)
826 		return -EINVAL;
827 
828 	return ret;
829 }
830 
831 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
832 {
833 	struct storvsc_device *stor_device;
834 	struct storvsc_cmd_request *request;
835 	struct vstor_packet *vstor_packet;
836 	int ret, i;
837 	int max_chns;
838 	bool process_sub_channels = false;
839 
840 	stor_device = get_out_stor_device(device);
841 	if (!stor_device)
842 		return -ENODEV;
843 
844 	request = &stor_device->init_request;
845 	vstor_packet = &request->vstor_packet;
846 
847 	/*
848 	 * Now, initiate the vsc/vsp initialization protocol on the open
849 	 * channel
850 	 */
851 	memset(request, 0, sizeof(struct storvsc_cmd_request));
852 	vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
853 	ret = storvsc_execute_vstor_op(device, request, true);
854 	if (ret)
855 		return ret;
856 	/*
857 	 * Query host supported protocol version.
858 	 */
859 
860 	for (i = 0; i < ARRAY_SIZE(protocol_version); i++) {
861 		/* reuse the packet for version range supported */
862 		memset(vstor_packet, 0, sizeof(struct vstor_packet));
863 		vstor_packet->operation =
864 			VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
865 
866 		vstor_packet->version.major_minor = protocol_version[i];
867 
868 		/*
869 		 * The revision number is only used in Windows; set it to 0.
870 		 */
871 		vstor_packet->version.revision = 0;
872 		ret = storvsc_execute_vstor_op(device, request, false);
873 		if (ret != 0)
874 			return ret;
875 
876 		if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
877 			return -EINVAL;
878 
879 		if (vstor_packet->status == 0) {
880 			vmstor_proto_version = protocol_version[i];
881 
882 			break;
883 		}
884 	}
885 
886 	if (vstor_packet->status != 0) {
887 		dev_err(&device->device, "Obsolete Hyper-V version\n");
888 		return -EINVAL;
889 	}
890 
891 
892 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
893 	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
894 	ret = storvsc_execute_vstor_op(device, request, true);
895 	if (ret != 0)
896 		return ret;
897 
898 	/*
899 	 * Check to see if multi-channel support is there.
900 	 * Hosts that implement protocol version of 5.1 and above
901 	 * support multi-channel.
902 	 */
903 	max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
904 
905 	/*
906 	 * Allocate state to manage the sub-channels.
907 	 * We allocate an array based on the numbers of possible CPUs
908 	 * (Hyper-V does not support cpu online/offline).
909 	 * This Array will be sparseley populated with unique
910 	 * channels - primary + sub-channels.
911 	 * We will however populate all the slots to evenly distribute
912 	 * the load.
913 	 */
914 	stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
915 					 GFP_KERNEL);
916 	if (stor_device->stor_chns == NULL)
917 		return -ENOMEM;
918 
919 	device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
920 
921 	stor_device->stor_chns[device->channel->target_cpu] = device->channel;
922 	cpumask_set_cpu(device->channel->target_cpu,
923 			&stor_device->alloced_cpus);
924 
925 	if (vstor_packet->storage_channel_properties.flags &
926 	    STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
927 		process_sub_channels = true;
928 
929 	stor_device->max_transfer_bytes =
930 		vstor_packet->storage_channel_properties.max_transfer_bytes;
931 
932 	if (!is_fc)
933 		goto done;
934 
935 	/*
936 	 * For FC devices retrieve FC HBA data.
937 	 */
938 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
939 	vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
940 	ret = storvsc_execute_vstor_op(device, request, true);
941 	if (ret != 0)
942 		return ret;
943 
944 	/*
945 	 * Cache the currently active port and node ww names.
946 	 */
947 	cache_wwn(stor_device, vstor_packet);
948 
949 done:
950 
951 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
952 	vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
953 	ret = storvsc_execute_vstor_op(device, request, true);
954 	if (ret != 0)
955 		return ret;
956 
957 	if (process_sub_channels)
958 		handle_multichannel_storage(device, max_chns);
959 
960 	return ret;
961 }
962 
963 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
964 				struct scsi_cmnd *scmnd,
965 				struct Scsi_Host *host,
966 				u8 asc, u8 ascq)
967 {
968 	struct storvsc_scan_work *wrk;
969 	void (*process_err_fn)(struct work_struct *work);
970 	struct hv_host_device *host_dev = shost_priv(host);
971 
972 	/*
973 	 * In some situations, Hyper-V sets multiple bits in the
974 	 * srb_status, such as ABORTED and ERROR. So process them
975 	 * individually, with the most specific bits first.
976 	 */
977 
978 	if (vm_srb->srb_status & SRB_STATUS_INVALID_LUN) {
979 		set_host_byte(scmnd, DID_NO_CONNECT);
980 		process_err_fn = storvsc_remove_lun;
981 		goto do_work;
982 	}
983 
984 	if (vm_srb->srb_status & SRB_STATUS_ABORTED) {
985 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
986 		    /* Capacity data has changed */
987 		    (asc == 0x2a) && (ascq == 0x9)) {
988 			process_err_fn = storvsc_device_scan;
989 			/*
990 			 * Retry the I/O that triggered this.
991 			 */
992 			set_host_byte(scmnd, DID_REQUEUE);
993 			goto do_work;
994 		}
995 	}
996 
997 	if (vm_srb->srb_status & SRB_STATUS_ERROR) {
998 		/*
999 		 * Let upper layer deal with error when
1000 		 * sense message is present.
1001 		 */
1002 		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
1003 			return;
1004 
1005 		/*
1006 		 * If there is an error; offline the device since all
1007 		 * error recovery strategies would have already been
1008 		 * deployed on the host side. However, if the command
1009 		 * were a pass-through command deal with it appropriately.
1010 		 */
1011 		switch (scmnd->cmnd[0]) {
1012 		case ATA_16:
1013 		case ATA_12:
1014 			set_host_byte(scmnd, DID_PASSTHROUGH);
1015 			break;
1016 		/*
1017 		 * On some Hyper-V hosts TEST_UNIT_READY command can
1018 		 * return SRB_STATUS_ERROR. Let the upper level code
1019 		 * deal with it based on the sense information.
1020 		 */
1021 		case TEST_UNIT_READY:
1022 			break;
1023 		default:
1024 			set_host_byte(scmnd, DID_ERROR);
1025 		}
1026 	}
1027 	return;
1028 
1029 do_work:
1030 	/*
1031 	 * We need to schedule work to process this error; schedule it.
1032 	 */
1033 	wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1034 	if (!wrk) {
1035 		set_host_byte(scmnd, DID_BAD_TARGET);
1036 		return;
1037 	}
1038 
1039 	wrk->host = host;
1040 	wrk->lun = vm_srb->lun;
1041 	wrk->tgt_id = vm_srb->target_id;
1042 	INIT_WORK(&wrk->work, process_err_fn);
1043 	queue_work(host_dev->handle_error_wq, &wrk->work);
1044 }
1045 
1046 
1047 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1048 				       struct storvsc_device *stor_dev)
1049 {
1050 	struct scsi_cmnd *scmnd = cmd_request->cmd;
1051 	struct scsi_sense_hdr sense_hdr;
1052 	struct vmscsi_request *vm_srb;
1053 	u32 data_transfer_length;
1054 	struct Scsi_Host *host;
1055 	u32 payload_sz = cmd_request->payload_sz;
1056 	void *payload = cmd_request->payload;
1057 	bool sense_ok;
1058 
1059 	host = stor_dev->host;
1060 
1061 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1062 	data_transfer_length = vm_srb->data_transfer_length;
1063 
1064 	scmnd->result = vm_srb->scsi_status;
1065 
1066 	if (scmnd->result) {
1067 		sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1068 				SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1069 
1070 		if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1071 			scsi_print_sense_hdr(scmnd->device, "storvsc",
1072 					     &sense_hdr);
1073 	}
1074 
1075 	if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1076 		storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1077 					 sense_hdr.ascq);
1078 		/*
1079 		 * The Windows driver set data_transfer_length on
1080 		 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1081 		 * is untouched.  In these cases we set it to 0.
1082 		 */
1083 		if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1084 			data_transfer_length = 0;
1085 	}
1086 
1087 	/* Validate data_transfer_length (from Hyper-V) */
1088 	if (data_transfer_length > cmd_request->payload->range.len)
1089 		data_transfer_length = cmd_request->payload->range.len;
1090 
1091 	scsi_set_resid(scmnd,
1092 		cmd_request->payload->range.len - data_transfer_length);
1093 
1094 	scsi_done(scmnd);
1095 
1096 	if (payload_sz >
1097 		sizeof(struct vmbus_channel_packet_multipage_buffer))
1098 		kfree(payload);
1099 }
1100 
1101 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1102 				  struct vstor_packet *vstor_packet,
1103 				  struct storvsc_cmd_request *request)
1104 {
1105 	struct vstor_packet *stor_pkt;
1106 	struct hv_device *device = stor_device->device;
1107 
1108 	stor_pkt = &request->vstor_packet;
1109 
1110 	/*
1111 	 * The current SCSI handling on the host side does
1112 	 * not correctly handle:
1113 	 * INQUIRY command with page code parameter set to 0x80
1114 	 * MODE_SENSE command with cmd[2] == 0x1c
1115 	 *
1116 	 * Setup srb and scsi status so this won't be fatal.
1117 	 * We do this so we can distinguish truly fatal failues
1118 	 * (srb status == 0x4) and off-line the device in that case.
1119 	 */
1120 
1121 	if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1122 	   (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1123 		vstor_packet->vm_srb.scsi_status = 0;
1124 		vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1125 	}
1126 
1127 	/* Copy over the status...etc */
1128 	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1129 	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1130 
1131 	/*
1132 	 * Copy over the sense_info_length, but limit to the known max
1133 	 * size if Hyper-V returns a bad value.
1134 	 */
1135 	stor_pkt->vm_srb.sense_info_length = min_t(u8, STORVSC_SENSE_BUFFER_SIZE,
1136 		vstor_packet->vm_srb.sense_info_length);
1137 
1138 	if (vstor_packet->vm_srb.scsi_status != 0 ||
1139 	    vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1140 
1141 		/*
1142 		 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1143 		 * return errors when detecting devices using TEST_UNIT_READY,
1144 		 * and logging these as errors produces unhelpful noise.
1145 		 */
1146 		int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1147 			STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1148 
1149 		storvsc_log(device, loglevel,
1150 			"tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1151 			scsi_cmd_to_rq(request->cmd)->tag,
1152 			stor_pkt->vm_srb.cdb[0],
1153 			vstor_packet->vm_srb.scsi_status,
1154 			vstor_packet->vm_srb.srb_status,
1155 			vstor_packet->status);
1156 	}
1157 
1158 	if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1159 	    (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1160 		memcpy(request->cmd->sense_buffer,
1161 		       vstor_packet->vm_srb.sense_data,
1162 		       stor_pkt->vm_srb.sense_info_length);
1163 
1164 	stor_pkt->vm_srb.data_transfer_length =
1165 		vstor_packet->vm_srb.data_transfer_length;
1166 
1167 	storvsc_command_completion(request, stor_device);
1168 
1169 	if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1170 		stor_device->drain_notify)
1171 		wake_up(&stor_device->waiting_to_drain);
1172 }
1173 
1174 static void storvsc_on_receive(struct storvsc_device *stor_device,
1175 			     struct vstor_packet *vstor_packet,
1176 			     struct storvsc_cmd_request *request)
1177 {
1178 	struct hv_host_device *host_dev;
1179 	switch (vstor_packet->operation) {
1180 	case VSTOR_OPERATION_COMPLETE_IO:
1181 		storvsc_on_io_completion(stor_device, vstor_packet, request);
1182 		break;
1183 
1184 	case VSTOR_OPERATION_REMOVE_DEVICE:
1185 	case VSTOR_OPERATION_ENUMERATE_BUS:
1186 		host_dev = shost_priv(stor_device->host);
1187 		queue_work(
1188 			host_dev->handle_error_wq, &host_dev->host_scan_work);
1189 		break;
1190 
1191 	case VSTOR_OPERATION_FCHBA_DATA:
1192 		cache_wwn(stor_device, vstor_packet);
1193 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1194 		fc_host_node_name(stor_device->host) = stor_device->node_name;
1195 		fc_host_port_name(stor_device->host) = stor_device->port_name;
1196 #endif
1197 		break;
1198 	default:
1199 		break;
1200 	}
1201 }
1202 
1203 static void storvsc_on_channel_callback(void *context)
1204 {
1205 	struct vmbus_channel *channel = (struct vmbus_channel *)context;
1206 	const struct vmpacket_descriptor *desc;
1207 	struct hv_device *device;
1208 	struct storvsc_device *stor_device;
1209 	struct Scsi_Host *shost;
1210 	unsigned long time_limit = jiffies + msecs_to_jiffies(CALLBACK_TIMEOUT);
1211 
1212 	if (channel->primary_channel != NULL)
1213 		device = channel->primary_channel->device_obj;
1214 	else
1215 		device = channel->device_obj;
1216 
1217 	stor_device = get_in_stor_device(device);
1218 	if (!stor_device)
1219 		return;
1220 
1221 	shost = stor_device->host;
1222 
1223 	foreach_vmbus_pkt(desc, channel) {
1224 		struct vstor_packet *packet = hv_pkt_data(desc);
1225 		struct storvsc_cmd_request *request = NULL;
1226 		u32 pktlen = hv_pkt_datalen(desc);
1227 		u64 rqst_id = desc->trans_id;
1228 		u32 minlen = rqst_id ? sizeof(struct vstor_packet) :
1229 			sizeof(enum vstor_packet_operation);
1230 
1231 		if (unlikely(time_after(jiffies, time_limit))) {
1232 			hv_pkt_iter_close(channel);
1233 			return;
1234 		}
1235 
1236 		if (pktlen < minlen) {
1237 			dev_err(&device->device,
1238 				"Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1239 				rqst_id, pktlen, minlen);
1240 			continue;
1241 		}
1242 
1243 		if (rqst_id == VMBUS_RQST_INIT) {
1244 			request = &stor_device->init_request;
1245 		} else if (rqst_id == VMBUS_RQST_RESET) {
1246 			request = &stor_device->reset_request;
1247 		} else {
1248 			/* Hyper-V can send an unsolicited message with ID of 0 */
1249 			if (rqst_id == 0) {
1250 				/*
1251 				 * storvsc_on_receive() looks at the vstor_packet in the message
1252 				 * from the ring buffer.
1253 				 *
1254 				 * - If the operation in the vstor_packet is COMPLETE_IO, then
1255 				 *   we call storvsc_on_io_completion(), and dereference the
1256 				 *   guest memory address.  Make sure we don't call
1257 				 *   storvsc_on_io_completion() with a guest memory address
1258 				 *   that is zero if Hyper-V were to construct and send such
1259 				 *   a bogus packet.
1260 				 *
1261 				 * - If the operation in the vstor_packet is FCHBA_DATA, then
1262 				 *   we call cache_wwn(), and access the data payload area of
1263 				 *   the packet (wwn_packet); however, there is no guarantee
1264 				 *   that the packet is big enough to contain such area.
1265 				 *   Future-proof the code by rejecting such a bogus packet.
1266 				 */
1267 				if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1268 				    packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1269 					dev_err(&device->device, "Invalid packet with ID of 0\n");
1270 					continue;
1271 				}
1272 			} else {
1273 				struct scsi_cmnd *scmnd;
1274 
1275 				/* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1276 				scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1277 				if (scmnd == NULL) {
1278 					dev_err(&device->device, "Incorrect transaction ID\n");
1279 					continue;
1280 				}
1281 				request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1282 				scsi_dma_unmap(scmnd);
1283 			}
1284 
1285 			storvsc_on_receive(stor_device, packet, request);
1286 			continue;
1287 		}
1288 
1289 		memcpy(&request->vstor_packet, packet,
1290 		       sizeof(struct vstor_packet));
1291 		complete(&request->wait_event);
1292 	}
1293 }
1294 
1295 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1296 				  bool is_fc)
1297 {
1298 	struct vmstorage_channel_properties props;
1299 	int ret;
1300 
1301 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1302 
1303 	device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1304 	device->channel->next_request_id_callback = storvsc_next_request_id;
1305 
1306 	ret = vmbus_open(device->channel,
1307 			 ring_size,
1308 			 ring_size,
1309 			 (void *)&props,
1310 			 sizeof(struct vmstorage_channel_properties),
1311 			 storvsc_on_channel_callback, device->channel);
1312 
1313 	if (ret != 0)
1314 		return ret;
1315 
1316 	ret = storvsc_channel_init(device, is_fc);
1317 
1318 	return ret;
1319 }
1320 
1321 static int storvsc_dev_remove(struct hv_device *device)
1322 {
1323 	struct storvsc_device *stor_device;
1324 
1325 	stor_device = hv_get_drvdata(device);
1326 
1327 	stor_device->destroy = true;
1328 
1329 	/* Make sure flag is set before waiting */
1330 	wmb();
1331 
1332 	/*
1333 	 * At this point, all outbound traffic should be disable. We
1334 	 * only allow inbound traffic (responses) to proceed so that
1335 	 * outstanding requests can be completed.
1336 	 */
1337 
1338 	storvsc_wait_to_drain(stor_device);
1339 
1340 	/*
1341 	 * Since we have already drained, we don't need to busy wait
1342 	 * as was done in final_release_stor_device()
1343 	 * Note that we cannot set the ext pointer to NULL until
1344 	 * we have drained - to drain the outgoing packets, we need to
1345 	 * allow incoming packets.
1346 	 */
1347 	hv_set_drvdata(device, NULL);
1348 
1349 	/* Close the channel */
1350 	vmbus_close(device->channel);
1351 
1352 	kfree(stor_device->stor_chns);
1353 	kfree(stor_device);
1354 	return 0;
1355 }
1356 
1357 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1358 					u16 q_num)
1359 {
1360 	u16 slot = 0;
1361 	u16 hash_qnum;
1362 	const struct cpumask *node_mask;
1363 	int num_channels, tgt_cpu;
1364 
1365 	if (stor_device->num_sc == 0) {
1366 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1367 		return stor_device->device->channel;
1368 	}
1369 
1370 	/*
1371 	 * Our channel array is sparsley populated and we
1372 	 * initiated I/O on a processor/hw-q that does not
1373 	 * currently have a designated channel. Fix this.
1374 	 * The strategy is simple:
1375 	 * I. Ensure NUMA locality
1376 	 * II. Distribute evenly (best effort)
1377 	 */
1378 
1379 	node_mask = cpumask_of_node(cpu_to_node(q_num));
1380 
1381 	num_channels = 0;
1382 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1383 		if (cpumask_test_cpu(tgt_cpu, node_mask))
1384 			num_channels++;
1385 	}
1386 	if (num_channels == 0) {
1387 		stor_device->stor_chns[q_num] = stor_device->device->channel;
1388 		return stor_device->device->channel;
1389 	}
1390 
1391 	hash_qnum = q_num;
1392 	while (hash_qnum >= num_channels)
1393 		hash_qnum -= num_channels;
1394 
1395 	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1396 		if (!cpumask_test_cpu(tgt_cpu, node_mask))
1397 			continue;
1398 		if (slot == hash_qnum)
1399 			break;
1400 		slot++;
1401 	}
1402 
1403 	stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1404 
1405 	return stor_device->stor_chns[q_num];
1406 }
1407 
1408 
1409 static int storvsc_do_io(struct hv_device *device,
1410 			 struct storvsc_cmd_request *request, u16 q_num)
1411 {
1412 	struct storvsc_device *stor_device;
1413 	struct vstor_packet *vstor_packet;
1414 	struct vmbus_channel *outgoing_channel, *channel;
1415 	unsigned long flags;
1416 	int ret = 0;
1417 	const struct cpumask *node_mask;
1418 	int tgt_cpu;
1419 
1420 	vstor_packet = &request->vstor_packet;
1421 	stor_device = get_out_stor_device(device);
1422 
1423 	if (!stor_device)
1424 		return -ENODEV;
1425 
1426 
1427 	request->device  = device;
1428 	/*
1429 	 * Select an appropriate channel to send the request out.
1430 	 */
1431 	/* See storvsc_change_target_cpu(). */
1432 	outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1433 	if (outgoing_channel != NULL) {
1434 		if (outgoing_channel->target_cpu == q_num) {
1435 			/*
1436 			 * Ideally, we want to pick a different channel if
1437 			 * available on the same NUMA node.
1438 			 */
1439 			node_mask = cpumask_of_node(cpu_to_node(q_num));
1440 			for_each_cpu_wrap(tgt_cpu,
1441 				 &stor_device->alloced_cpus, q_num + 1) {
1442 				if (!cpumask_test_cpu(tgt_cpu, node_mask))
1443 					continue;
1444 				if (tgt_cpu == q_num)
1445 					continue;
1446 				channel = READ_ONCE(
1447 					stor_device->stor_chns[tgt_cpu]);
1448 				if (channel == NULL)
1449 					continue;
1450 				if (hv_get_avail_to_write_percent(
1451 							&channel->outbound)
1452 						> ring_avail_percent_lowater) {
1453 					outgoing_channel = channel;
1454 					goto found_channel;
1455 				}
1456 			}
1457 
1458 			/*
1459 			 * All the other channels on the same NUMA node are
1460 			 * busy. Try to use the channel on the current CPU
1461 			 */
1462 			if (hv_get_avail_to_write_percent(
1463 						&outgoing_channel->outbound)
1464 					> ring_avail_percent_lowater)
1465 				goto found_channel;
1466 
1467 			/*
1468 			 * If we reach here, all the channels on the current
1469 			 * NUMA node are busy. Try to find a channel in
1470 			 * other NUMA nodes
1471 			 */
1472 			for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1473 				if (cpumask_test_cpu(tgt_cpu, node_mask))
1474 					continue;
1475 				channel = READ_ONCE(
1476 					stor_device->stor_chns[tgt_cpu]);
1477 				if (channel == NULL)
1478 					continue;
1479 				if (hv_get_avail_to_write_percent(
1480 							&channel->outbound)
1481 						> ring_avail_percent_lowater) {
1482 					outgoing_channel = channel;
1483 					goto found_channel;
1484 				}
1485 			}
1486 		}
1487 	} else {
1488 		spin_lock_irqsave(&stor_device->lock, flags);
1489 		outgoing_channel = stor_device->stor_chns[q_num];
1490 		if (outgoing_channel != NULL) {
1491 			spin_unlock_irqrestore(&stor_device->lock, flags);
1492 			goto found_channel;
1493 		}
1494 		outgoing_channel = get_og_chn(stor_device, q_num);
1495 		spin_unlock_irqrestore(&stor_device->lock, flags);
1496 	}
1497 
1498 found_channel:
1499 	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1500 
1501 	vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1502 
1503 
1504 	vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1505 
1506 
1507 	vstor_packet->vm_srb.data_transfer_length =
1508 	request->payload->range.len;
1509 
1510 	vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1511 
1512 	if (request->payload->range.len) {
1513 
1514 		ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1515 				request->payload, request->payload_sz,
1516 				vstor_packet,
1517 				sizeof(struct vstor_packet),
1518 				(unsigned long)request);
1519 	} else {
1520 		ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1521 			       sizeof(struct vstor_packet),
1522 			       (unsigned long)request,
1523 			       VM_PKT_DATA_INBAND,
1524 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1525 	}
1526 
1527 	if (ret != 0)
1528 		return ret;
1529 
1530 	atomic_inc(&stor_device->num_outstanding_req);
1531 
1532 	return ret;
1533 }
1534 
1535 static int storvsc_device_alloc(struct scsi_device *sdevice)
1536 {
1537 	/*
1538 	 * Set blist flag to permit the reading of the VPD pages even when
1539 	 * the target may claim SPC-2 compliance. MSFT targets currently
1540 	 * claim SPC-2 compliance while they implement post SPC-2 features.
1541 	 * With this flag we can correctly handle WRITE_SAME_16 issues.
1542 	 *
1543 	 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1544 	 * still supports REPORT LUN.
1545 	 */
1546 	sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1547 
1548 	return 0;
1549 }
1550 
1551 static int storvsc_device_configure(struct scsi_device *sdevice)
1552 {
1553 	blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1554 
1555 	sdevice->no_write_same = 1;
1556 
1557 	/*
1558 	 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1559 	 * if the device is a MSFT virtual device.  If the host is
1560 	 * WIN10 or newer, allow write_same.
1561 	 */
1562 	if (!strncmp(sdevice->vendor, "Msft", 4)) {
1563 		switch (vmstor_proto_version) {
1564 		case VMSTOR_PROTO_VERSION_WIN8:
1565 		case VMSTOR_PROTO_VERSION_WIN8_1:
1566 			sdevice->scsi_level = SCSI_SPC_3;
1567 			break;
1568 		}
1569 
1570 		if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1571 			sdevice->no_write_same = 0;
1572 	}
1573 
1574 	return 0;
1575 }
1576 
1577 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1578 			   sector_t capacity, int *info)
1579 {
1580 	sector_t nsect = capacity;
1581 	sector_t cylinders = nsect;
1582 	int heads, sectors_pt;
1583 
1584 	/*
1585 	 * We are making up these values; let us keep it simple.
1586 	 */
1587 	heads = 0xff;
1588 	sectors_pt = 0x3f;      /* Sectors per track */
1589 	sector_div(cylinders, heads * sectors_pt);
1590 	if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1591 		cylinders = 0xffff;
1592 
1593 	info[0] = heads;
1594 	info[1] = sectors_pt;
1595 	info[2] = (int)cylinders;
1596 
1597 	return 0;
1598 }
1599 
1600 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1601 {
1602 	struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1603 	struct hv_device *device = host_dev->dev;
1604 
1605 	struct storvsc_device *stor_device;
1606 	struct storvsc_cmd_request *request;
1607 	struct vstor_packet *vstor_packet;
1608 	int ret, t;
1609 
1610 	stor_device = get_out_stor_device(device);
1611 	if (!stor_device)
1612 		return FAILED;
1613 
1614 	request = &stor_device->reset_request;
1615 	vstor_packet = &request->vstor_packet;
1616 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1617 
1618 	init_completion(&request->wait_event);
1619 
1620 	vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1621 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1622 	vstor_packet->vm_srb.path_id = stor_device->path_id;
1623 
1624 	ret = vmbus_sendpacket(device->channel, vstor_packet,
1625 			       sizeof(struct vstor_packet),
1626 			       VMBUS_RQST_RESET,
1627 			       VM_PKT_DATA_INBAND,
1628 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1629 	if (ret != 0)
1630 		return FAILED;
1631 
1632 	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1633 	if (t == 0)
1634 		return TIMEOUT_ERROR;
1635 
1636 
1637 	/*
1638 	 * At this point, all outstanding requests in the adapter
1639 	 * should have been flushed out and return to us
1640 	 * There is a potential race here where the host may be in
1641 	 * the process of responding when we return from here.
1642 	 * Just wait for all in-transit packets to be accounted for
1643 	 * before we return from here.
1644 	 */
1645 	storvsc_wait_to_drain(stor_device);
1646 
1647 	return SUCCESS;
1648 }
1649 
1650 /*
1651  * The host guarantees to respond to each command, although I/O latencies might
1652  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1653  * chance to perform EH.
1654  */
1655 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1656 {
1657 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1658 	if (scmnd->device->host->transportt == fc_transport_template)
1659 		return fc_eh_timed_out(scmnd);
1660 #endif
1661 	return BLK_EH_RESET_TIMER;
1662 }
1663 
1664 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1665 {
1666 	bool allowed = true;
1667 	u8 scsi_op = scmnd->cmnd[0];
1668 
1669 	switch (scsi_op) {
1670 	/* the host does not handle WRITE_SAME, log accident usage */
1671 	case WRITE_SAME:
1672 	/*
1673 	 * smartd sends this command and the host does not handle
1674 	 * this. So, don't send it.
1675 	 */
1676 	case SET_WINDOW:
1677 		set_host_byte(scmnd, DID_ERROR);
1678 		allowed = false;
1679 		break;
1680 	default:
1681 		break;
1682 	}
1683 	return allowed;
1684 }
1685 
1686 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1687 {
1688 	int ret;
1689 	struct hv_host_device *host_dev = shost_priv(host);
1690 	struct hv_device *dev = host_dev->dev;
1691 	struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1692 	struct scatterlist *sgl;
1693 	struct vmscsi_request *vm_srb;
1694 	struct vmbus_packet_mpb_array  *payload;
1695 	u32 payload_sz;
1696 	u32 length;
1697 
1698 	if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1699 		/*
1700 		 * On legacy hosts filter unimplemented commands.
1701 		 * Future hosts are expected to correctly handle
1702 		 * unsupported commands. Furthermore, it is
1703 		 * possible that some of the currently
1704 		 * unsupported commands maybe supported in
1705 		 * future versions of the host.
1706 		 */
1707 		if (!storvsc_scsi_cmd_ok(scmnd)) {
1708 			scsi_done(scmnd);
1709 			return 0;
1710 		}
1711 	}
1712 
1713 	/* Setup the cmd request */
1714 	cmd_request->cmd = scmnd;
1715 
1716 	memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1717 	vm_srb = &cmd_request->vstor_packet.vm_srb;
1718 	vm_srb->time_out_value = 60;
1719 
1720 	vm_srb->srb_flags |=
1721 		SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1722 
1723 	if (scmnd->device->tagged_supported) {
1724 		vm_srb->srb_flags |=
1725 		(SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1726 		vm_srb->queue_tag = SP_UNTAGGED;
1727 		vm_srb->queue_action = SRB_SIMPLE_TAG_REQUEST;
1728 	}
1729 
1730 	/* Build the SRB */
1731 	switch (scmnd->sc_data_direction) {
1732 	case DMA_TO_DEVICE:
1733 		vm_srb->data_in = WRITE_TYPE;
1734 		vm_srb->srb_flags |= SRB_FLAGS_DATA_OUT;
1735 		break;
1736 	case DMA_FROM_DEVICE:
1737 		vm_srb->data_in = READ_TYPE;
1738 		vm_srb->srb_flags |= SRB_FLAGS_DATA_IN;
1739 		break;
1740 	case DMA_NONE:
1741 		vm_srb->data_in = UNKNOWN_TYPE;
1742 		vm_srb->srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1743 		break;
1744 	default:
1745 		/*
1746 		 * This is DMA_BIDIRECTIONAL or something else we are never
1747 		 * supposed to see here.
1748 		 */
1749 		WARN(1, "Unexpected data direction: %d\n",
1750 		     scmnd->sc_data_direction);
1751 		return -EINVAL;
1752 	}
1753 
1754 
1755 	vm_srb->port_number = host_dev->port;
1756 	vm_srb->path_id = scmnd->device->channel;
1757 	vm_srb->target_id = scmnd->device->id;
1758 	vm_srb->lun = scmnd->device->lun;
1759 
1760 	vm_srb->cdb_length = scmnd->cmd_len;
1761 
1762 	memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1763 
1764 	sgl = (struct scatterlist *)scsi_sglist(scmnd);
1765 
1766 	length = scsi_bufflen(scmnd);
1767 	payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1768 	payload_sz = sizeof(cmd_request->mpb);
1769 
1770 	if (scsi_sg_count(scmnd)) {
1771 		unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1772 		unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1773 		struct scatterlist *sg;
1774 		unsigned long hvpfn, hvpfns_to_add;
1775 		int j, i = 0, sg_count;
1776 
1777 		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1778 
1779 			payload_sz = (hvpg_count * sizeof(u64) +
1780 				      sizeof(struct vmbus_packet_mpb_array));
1781 			payload = kzalloc(payload_sz, GFP_ATOMIC);
1782 			if (!payload)
1783 				return SCSI_MLQUEUE_DEVICE_BUSY;
1784 		}
1785 
1786 		payload->range.len = length;
1787 		payload->range.offset = offset_in_hvpg;
1788 
1789 		sg_count = scsi_dma_map(scmnd);
1790 		if (sg_count < 0) {
1791 			ret = SCSI_MLQUEUE_DEVICE_BUSY;
1792 			goto err_free_payload;
1793 		}
1794 
1795 		for_each_sg(sgl, sg, sg_count, j) {
1796 			/*
1797 			 * Init values for the current sgl entry. hvpfns_to_add
1798 			 * is in units of Hyper-V size pages. Handling the
1799 			 * PAGE_SIZE != HV_HYP_PAGE_SIZE case also handles
1800 			 * values of sgl->offset that are larger than PAGE_SIZE.
1801 			 * Such offsets are handled even on other than the first
1802 			 * sgl entry, provided they are a multiple of PAGE_SIZE.
1803 			 */
1804 			hvpfn = HVPFN_DOWN(sg_dma_address(sg));
1805 			hvpfns_to_add = HVPFN_UP(sg_dma_address(sg) +
1806 						 sg_dma_len(sg)) - hvpfn;
1807 
1808 			/*
1809 			 * Fill the next portion of the PFN array with
1810 			 * sequential Hyper-V PFNs for the continguous physical
1811 			 * memory described by the sgl entry. The end of the
1812 			 * last sgl should be reached at the same time that
1813 			 * the PFN array is filled.
1814 			 */
1815 			while (hvpfns_to_add--)
1816 				payload->range.pfn_array[i++] = hvpfn++;
1817 		}
1818 	}
1819 
1820 	cmd_request->payload = payload;
1821 	cmd_request->payload_sz = payload_sz;
1822 
1823 	/* Invokes the vsc to start an IO */
1824 	ret = storvsc_do_io(dev, cmd_request, get_cpu());
1825 	put_cpu();
1826 
1827 	if (ret == -EAGAIN) {
1828 		/* no more space */
1829 		ret = SCSI_MLQUEUE_DEVICE_BUSY;
1830 		goto err_free_payload;
1831 	}
1832 
1833 	return 0;
1834 
1835 err_free_payload:
1836 	if (payload_sz > sizeof(cmd_request->mpb))
1837 		kfree(payload);
1838 
1839 	return ret;
1840 }
1841 
1842 static struct scsi_host_template scsi_driver = {
1843 	.module	=		THIS_MODULE,
1844 	.name =			"storvsc_host_t",
1845 	.cmd_size =             sizeof(struct storvsc_cmd_request),
1846 	.bios_param =		storvsc_get_chs,
1847 	.queuecommand =		storvsc_queuecommand,
1848 	.eh_host_reset_handler =	storvsc_host_reset_handler,
1849 	.proc_name =		"storvsc_host",
1850 	.eh_timed_out =		storvsc_eh_timed_out,
1851 	.slave_alloc =		storvsc_device_alloc,
1852 	.slave_configure =	storvsc_device_configure,
1853 	.cmd_per_lun =		2048,
1854 	.this_id =		-1,
1855 	/* Ensure there are no gaps in presented sgls */
1856 	.virt_boundary_mask =	HV_HYP_PAGE_SIZE - 1,
1857 	.no_write_same =	1,
1858 	.track_queue_depth =	1,
1859 	.change_queue_depth =	storvsc_change_queue_depth,
1860 };
1861 
1862 enum {
1863 	SCSI_GUID,
1864 	IDE_GUID,
1865 	SFC_GUID,
1866 };
1867 
1868 static const struct hv_vmbus_device_id id_table[] = {
1869 	/* SCSI guid */
1870 	{ HV_SCSI_GUID,
1871 	  .driver_data = SCSI_GUID
1872 	},
1873 	/* IDE guid */
1874 	{ HV_IDE_GUID,
1875 	  .driver_data = IDE_GUID
1876 	},
1877 	/* Fibre Channel GUID */
1878 	{
1879 	  HV_SYNTHFC_GUID,
1880 	  .driver_data = SFC_GUID
1881 	},
1882 	{ },
1883 };
1884 
1885 MODULE_DEVICE_TABLE(vmbus, id_table);
1886 
1887 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1888 
1889 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1890 {
1891 	return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1892 }
1893 
1894 static int storvsc_probe(struct hv_device *device,
1895 			const struct hv_vmbus_device_id *dev_id)
1896 {
1897 	int ret;
1898 	int num_cpus = num_online_cpus();
1899 	int num_present_cpus = num_present_cpus();
1900 	struct Scsi_Host *host;
1901 	struct hv_host_device *host_dev;
1902 	bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1903 	bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1904 	int target = 0;
1905 	struct storvsc_device *stor_device;
1906 	int max_sub_channels = 0;
1907 	u32 max_xfer_bytes;
1908 
1909 	/*
1910 	 * We support sub-channels for storage on SCSI and FC controllers.
1911 	 * The number of sub-channels offerred is based on the number of
1912 	 * VCPUs in the guest.
1913 	 */
1914 	if (!dev_is_ide)
1915 		max_sub_channels =
1916 			(num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1917 
1918 	scsi_driver.can_queue = max_outstanding_req_per_channel *
1919 				(max_sub_channels + 1) *
1920 				(100 - ring_avail_percent_lowater) / 100;
1921 
1922 	host = scsi_host_alloc(&scsi_driver,
1923 			       sizeof(struct hv_host_device));
1924 	if (!host)
1925 		return -ENOMEM;
1926 
1927 	host_dev = shost_priv(host);
1928 	memset(host_dev, 0, sizeof(struct hv_host_device));
1929 
1930 	host_dev->port = host->host_no;
1931 	host_dev->dev = device;
1932 	host_dev->host = host;
1933 
1934 
1935 	stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1936 	if (!stor_device) {
1937 		ret = -ENOMEM;
1938 		goto err_out0;
1939 	}
1940 
1941 	stor_device->destroy = false;
1942 	init_waitqueue_head(&stor_device->waiting_to_drain);
1943 	stor_device->device = device;
1944 	stor_device->host = host;
1945 	spin_lock_init(&stor_device->lock);
1946 	hv_set_drvdata(device, stor_device);
1947 	dma_set_min_align_mask(&device->device, HV_HYP_PAGE_SIZE - 1);
1948 
1949 	stor_device->port_number = host->host_no;
1950 	ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1951 	if (ret)
1952 		goto err_out1;
1953 
1954 	host_dev->path = stor_device->path_id;
1955 	host_dev->target = stor_device->target_id;
1956 
1957 	switch (dev_id->driver_data) {
1958 	case SFC_GUID:
1959 		host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1960 		host->max_id = STORVSC_FC_MAX_TARGETS;
1961 		host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1962 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1963 		host->transportt = fc_transport_template;
1964 #endif
1965 		break;
1966 
1967 	case SCSI_GUID:
1968 		host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1969 		host->max_id = STORVSC_MAX_TARGETS;
1970 		host->max_channel = STORVSC_MAX_CHANNELS - 1;
1971 		break;
1972 
1973 	default:
1974 		host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1975 		host->max_id = STORVSC_IDE_MAX_TARGETS;
1976 		host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
1977 		break;
1978 	}
1979 	/* max cmd length */
1980 	host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1981 	/*
1982 	 * Any reasonable Hyper-V configuration should provide
1983 	 * max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
1984 	 * protecting it from any weird value.
1985 	 */
1986 	max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_SIZE);
1987 	/* max_hw_sectors_kb */
1988 	host->max_sectors = max_xfer_bytes >> 9;
1989 	/*
1990 	 * There are 2 requirements for Hyper-V storvsc sgl segments,
1991 	 * based on which the below calculation for max segments is
1992 	 * done:
1993 	 *
1994 	 * 1. Except for the first and last sgl segment, all sgl segments
1995 	 *    should be align to HV_HYP_PAGE_SIZE, that also means the
1996 	 *    maximum number of segments in a sgl can be calculated by
1997 	 *    dividing the total max transfer length by HV_HYP_PAGE_SIZE.
1998 	 *
1999 	 * 2. Except for the first and last, each entry in the SGL must
2000 	 *    have an offset that is a multiple of HV_HYP_PAGE_SIZE.
2001 	 */
2002 	host->sg_tablesize = (max_xfer_bytes >> HV_HYP_PAGE_SHIFT) + 1;
2003 	/*
2004 	 * For non-IDE disks, the host supports multiple channels.
2005 	 * Set the number of HW queues we are supporting.
2006 	 */
2007 	if (!dev_is_ide) {
2008 		if (storvsc_max_hw_queues > num_present_cpus) {
2009 			storvsc_max_hw_queues = 0;
2010 			storvsc_log(device, STORVSC_LOGGING_WARN,
2011 				"Resetting invalid storvsc_max_hw_queues value to default.\n");
2012 		}
2013 		if (storvsc_max_hw_queues)
2014 			host->nr_hw_queues = storvsc_max_hw_queues;
2015 		else
2016 			host->nr_hw_queues = num_present_cpus;
2017 	}
2018 
2019 	/*
2020 	 * Set the error handler work queue.
2021 	 */
2022 	host_dev->handle_error_wq =
2023 			alloc_ordered_workqueue("storvsc_error_wq_%d",
2024 						0,
2025 						host->host_no);
2026 	if (!host_dev->handle_error_wq) {
2027 		ret = -ENOMEM;
2028 		goto err_out2;
2029 	}
2030 	INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2031 	/* Register the HBA and start the scsi bus scan */
2032 	ret = scsi_add_host(host, &device->device);
2033 	if (ret != 0)
2034 		goto err_out3;
2035 
2036 	if (!dev_is_ide) {
2037 		scsi_scan_host(host);
2038 	} else {
2039 		target = (device->dev_instance.b[5] << 8 |
2040 			 device->dev_instance.b[4]);
2041 		ret = scsi_add_device(host, 0, target, 0);
2042 		if (ret)
2043 			goto err_out4;
2044 	}
2045 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2046 	if (host->transportt == fc_transport_template) {
2047 		struct fc_rport_identifiers ids = {
2048 			.roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2049 		};
2050 
2051 		fc_host_node_name(host) = stor_device->node_name;
2052 		fc_host_port_name(host) = stor_device->port_name;
2053 		stor_device->rport = fc_remote_port_add(host, 0, &ids);
2054 		if (!stor_device->rport) {
2055 			ret = -ENOMEM;
2056 			goto err_out4;
2057 		}
2058 	}
2059 #endif
2060 	return 0;
2061 
2062 err_out4:
2063 	scsi_remove_host(host);
2064 
2065 err_out3:
2066 	destroy_workqueue(host_dev->handle_error_wq);
2067 
2068 err_out2:
2069 	/*
2070 	 * Once we have connected with the host, we would need to
2071 	 * invoke storvsc_dev_remove() to rollback this state and
2072 	 * this call also frees up the stor_device; hence the jump around
2073 	 * err_out1 label.
2074 	 */
2075 	storvsc_dev_remove(device);
2076 	goto err_out0;
2077 
2078 err_out1:
2079 	kfree(stor_device->stor_chns);
2080 	kfree(stor_device);
2081 
2082 err_out0:
2083 	scsi_host_put(host);
2084 	return ret;
2085 }
2086 
2087 /* Change a scsi target's queue depth */
2088 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2089 {
2090 	if (queue_depth > scsi_driver.can_queue)
2091 		queue_depth = scsi_driver.can_queue;
2092 
2093 	return scsi_change_queue_depth(sdev, queue_depth);
2094 }
2095 
2096 static int storvsc_remove(struct hv_device *dev)
2097 {
2098 	struct storvsc_device *stor_device = hv_get_drvdata(dev);
2099 	struct Scsi_Host *host = stor_device->host;
2100 	struct hv_host_device *host_dev = shost_priv(host);
2101 
2102 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2103 	if (host->transportt == fc_transport_template) {
2104 		fc_remote_port_delete(stor_device->rport);
2105 		fc_remove_host(host);
2106 	}
2107 #endif
2108 	destroy_workqueue(host_dev->handle_error_wq);
2109 	scsi_remove_host(host);
2110 	storvsc_dev_remove(dev);
2111 	scsi_host_put(host);
2112 
2113 	return 0;
2114 }
2115 
2116 static int storvsc_suspend(struct hv_device *hv_dev)
2117 {
2118 	struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2119 	struct Scsi_Host *host = stor_device->host;
2120 	struct hv_host_device *host_dev = shost_priv(host);
2121 
2122 	storvsc_wait_to_drain(stor_device);
2123 
2124 	drain_workqueue(host_dev->handle_error_wq);
2125 
2126 	vmbus_close(hv_dev->channel);
2127 
2128 	kfree(stor_device->stor_chns);
2129 	stor_device->stor_chns = NULL;
2130 
2131 	cpumask_clear(&stor_device->alloced_cpus);
2132 
2133 	return 0;
2134 }
2135 
2136 static int storvsc_resume(struct hv_device *hv_dev)
2137 {
2138 	int ret;
2139 
2140 	ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2141 				     hv_dev_is_fc(hv_dev));
2142 	return ret;
2143 }
2144 
2145 static struct hv_driver storvsc_drv = {
2146 	.name = KBUILD_MODNAME,
2147 	.id_table = id_table,
2148 	.probe = storvsc_probe,
2149 	.remove = storvsc_remove,
2150 	.suspend = storvsc_suspend,
2151 	.resume = storvsc_resume,
2152 	.driver = {
2153 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
2154 	},
2155 };
2156 
2157 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2158 static struct fc_function_template fc_transport_functions = {
2159 	.show_host_node_name = 1,
2160 	.show_host_port_name = 1,
2161 };
2162 #endif
2163 
2164 static int __init storvsc_drv_init(void)
2165 {
2166 	int ret;
2167 
2168 	/*
2169 	 * Divide the ring buffer data size (which is 1 page less
2170 	 * than the ring buffer size since that page is reserved for
2171 	 * the ring buffer indices) by the max request size (which is
2172 	 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2173 	 */
2174 	max_outstanding_req_per_channel =
2175 		((storvsc_ringbuffer_size - PAGE_SIZE) /
2176 		ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2177 		sizeof(struct vstor_packet) + sizeof(u64),
2178 		sizeof(u64)));
2179 
2180 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2181 	fc_transport_template = fc_attach_transport(&fc_transport_functions);
2182 	if (!fc_transport_template)
2183 		return -ENODEV;
2184 #endif
2185 
2186 	ret = vmbus_driver_register(&storvsc_drv);
2187 
2188 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2189 	if (ret)
2190 		fc_release_transport(fc_transport_template);
2191 #endif
2192 
2193 	return ret;
2194 }
2195 
2196 static void __exit storvsc_drv_exit(void)
2197 {
2198 	vmbus_driver_unregister(&storvsc_drv);
2199 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2200 	fc_release_transport(fc_transport_template);
2201 #endif
2202 }
2203 
2204 MODULE_LICENSE("GPL");
2205 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2206 module_init(storvsc_drv_init);
2207 module_exit(storvsc_drv_exit);
2208