drivers/scsi/storvsc_drv.c

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