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 <scsi/scsi.h>
  25 #include <scsi/scsi_cmnd.h>
  26 #include <scsi/scsi_host.h>
  27 #include <scsi/scsi_device.h>
  28 #include <scsi/scsi_tcq.h>
  29 #include <scsi/scsi_eh.h>
  30 #include <scsi/scsi_devinfo.h>
  31 #include <scsi/scsi_dbg.h>
  32 #include <scsi/scsi_transport_fc.h>
  33 #include <scsi/scsi_transport.h>
  34
  35 /*
  36  * All wire protocol details (storage protocol between the guest and the host)
  37  * are consolidated here.
  38  *
  39  * Begin protocol definitions.
  40  */
  41
  42 /*
  43  * Version history:
  44  * V1 Beta: 0.1
  45  * V1 RC < 2008/1/31: 1.0
  46  * V1 RC > 2008/1/31:  2.0
  47  * Win7: 4.2
  48  * Win8: 5.1
  49  * Win8.1: 6.0
  50  * Win10: 6.2
  51  */
  52
  53 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_)    ((((MAJOR_) & 0xff) << 8) | \
  54                                                 (((MINOR_) & 0xff)))
  55
  56 #define VMSTOR_PROTO_VERSION_WIN6       VMSTOR_PROTO_VERSION(2, 0)
  57 #define VMSTOR_PROTO_VERSION_WIN7       VMSTOR_PROTO_VERSION(4, 2)
  58 #define VMSTOR_PROTO_VERSION_WIN8       VMSTOR_PROTO_VERSION(5, 1)
  59 #define VMSTOR_PROTO_VERSION_WIN8_1     VMSTOR_PROTO_VERSION(6, 0)
  60 #define VMSTOR_PROTO_VERSION_WIN10      VMSTOR_PROTO_VERSION(6, 2)
  61
  62 /*  Packet structure describing virtual storage requests. */
  63 enum vstor_packet_operation {
  64         VSTOR_OPERATION_COMPLETE_IO             = 1,
  65         VSTOR_OPERATION_REMOVE_DEVICE           = 2,
  66         VSTOR_OPERATION_EXECUTE_SRB             = 3,
  67         VSTOR_OPERATION_RESET_LUN               = 4,
  68         VSTOR_OPERATION_RESET_ADAPTER           = 5,
  69         VSTOR_OPERATION_RESET_BUS               = 6,
  70         VSTOR_OPERATION_BEGIN_INITIALIZATION    = 7,
  71         VSTOR_OPERATION_END_INITIALIZATION      = 8,
  72         VSTOR_OPERATION_QUERY_PROTOCOL_VERSION  = 9,
  73         VSTOR_OPERATION_QUERY_PROPERTIES        = 10,
  74         VSTOR_OPERATION_ENUMERATE_BUS           = 11,
  75         VSTOR_OPERATION_FCHBA_DATA              = 12,
  76         VSTOR_OPERATION_CREATE_SUB_CHANNELS     = 13,
  77         VSTOR_OPERATION_MAXIMUM                 = 13
  78 };
  79
  80 /*
  81  * WWN packet for Fibre Channel HBA
  82  */
  83
  84 struct hv_fc_wwn_packet {
  85         u8      primary_active;
  86         u8      reserved1[3];
  87         u8      primary_port_wwn[8];
  88         u8      primary_node_wwn[8];
  89         u8      secondary_port_wwn[8];
  90         u8      secondary_node_wwn[8];
  91 };
  92
  93
  94
  95 /*
  96  * SRB Flag Bits
  97  */
  98
  99 #define SRB_FLAGS_QUEUE_ACTION_ENABLE           0x00000002
 100 #define SRB_FLAGS_DISABLE_DISCONNECT            0x00000004
 101 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER        0x00000008
 102 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE           0x00000010
 103 #define SRB_FLAGS_DISABLE_AUTOSENSE             0x00000020
 104 #define SRB_FLAGS_DATA_IN                       0x00000040
 105 #define SRB_FLAGS_DATA_OUT                      0x00000080
 106 #define SRB_FLAGS_NO_DATA_TRANSFER              0x00000000
 107 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
 108 #define SRB_FLAGS_NO_QUEUE_FREEZE               0x00000100
 109 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE          0x00000200
 110 #define SRB_FLAGS_FREE_SENSE_BUFFER             0x00000400
 111
 112 /*
 113  * This flag indicates the request is part of the workflow for processing a D3.
 114  */
 115 #define SRB_FLAGS_D3_PROCESSING                 0x00000800
 116 #define SRB_FLAGS_IS_ACTIVE                     0x00010000
 117 #define SRB_FLAGS_ALLOCATED_FROM_ZONE           0x00020000
 118 #define SRB_FLAGS_SGLIST_FROM_POOL              0x00040000
 119 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE           0x00080000
 120 #define SRB_FLAGS_NO_KEEP_AWAKE                 0x00100000
 121 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE        0x00200000
 122 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT      0x00400000
 123 #define SRB_FLAGS_DONT_START_NEXT_PACKET        0x00800000
 124 #define SRB_FLAGS_PORT_DRIVER_RESERVED          0x0F000000
 125 #define SRB_FLAGS_CLASS_DRIVER_RESERVED         0xF0000000
 126
 127 #define SP_UNTAGGED                     ((unsigned char) ~0)
 128 #define SRB_SIMPLE_TAG_REQUEST          0x20
 129
 130 /*
 131  * Platform neutral description of a scsi request -
 132  * this remains the same across the write regardless of 32/64 bit
 133  * note: it's patterned off the SCSI_PASS_THROUGH structure
 134  */
 135 #define STORVSC_MAX_CMD_LEN                     0x10
 136
 137 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE     0x14
 138 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE      0x12
 139
 140 #define STORVSC_SENSE_BUFFER_SIZE               0x14
 141 #define STORVSC_MAX_BUF_LEN_WITH_PADDING        0x14
 142
 143 /*
 144  * Sense buffer size changed in win8; have a run-time
 145  * variable to track the size we should use.  This value will
 146  * likely change during protocol negotiation but it is valid
 147  * to start by assuming pre-Win8.
 148  */
 149 static int sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE;
 150
 151 /*
 152  * The storage protocol version is determined during the
 153  * initial exchange with the host.  It will indicate which
 154  * storage functionality is available in the host.
 155 */
 156 static int vmstor_proto_version;
 157
 158 #define STORVSC_LOGGING_NONE    0
 159 #define STORVSC_LOGGING_ERROR   1
 160 #define STORVSC_LOGGING_WARN    2
 161
 162 static int logging_level = STORVSC_LOGGING_ERROR;
 163 module_param(logging_level, int, S_IRUGO|S_IWUSR);
 164 MODULE_PARM_DESC(logging_level,
 165         "Logging level, 0 - None, 1 - Error (default), 2 - Warning.");
 166
 167 static inline bool do_logging(int level)
 168 {
 169         return logging_level >= level;
 170 }
 171
 172 #define storvsc_log(dev, level, fmt, ...)                       \
 173 do {                                                            \
 174         if (do_logging(level))                                  \
 175                 dev_warn(&(dev)->device, fmt, ##__VA_ARGS__);   \
 176 } while (0)
 177
 178 struct vmscsi_win8_extension {
 179         /*
 180          * The following were added in Windows 8
 181          */
 182         u16 reserve;
 183         u8  queue_tag;
 184         u8  queue_action;
 185         u32 srb_flags;
 186         u32 time_out_value;
 187         u32 queue_sort_ey;
 188 } __packed;
 189
 190 struct vmscsi_request {
 191         u16 length;
 192         u8 srb_status;
 193         u8 scsi_status;
 194
 195         u8  port_number;
 196         u8  path_id;
 197         u8  target_id;
 198         u8  lun;
 199
 200         u8  cdb_length;
 201         u8  sense_info_length;
 202         u8  data_in;
 203         u8  reserved;
 204
 205         u32 data_transfer_length;
 206
 207         union {
 208                 u8 cdb[STORVSC_MAX_CMD_LEN];
 209                 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
 210                 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
 211         };
 212         /*
 213          * The following was added in win8.
 214          */
 215         struct vmscsi_win8_extension win8_extension;
 216
 217 } __attribute((packed));
 218
 219 /*
 220  * The list of storage protocols in order of preference.
 221  */
 222 struct vmstor_protocol {
 223         int protocol_version;
 224         int sense_buffer_size;
 225         int vmscsi_size_delta;
 226 };
 227
 228
 229 static const struct vmstor_protocol vmstor_protocols[] = {
 230         {
 231                 VMSTOR_PROTO_VERSION_WIN10,
 232                 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
 233                 0
 234         },
 235         {
 236                 VMSTOR_PROTO_VERSION_WIN8_1,
 237                 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
 238                 0
 239         },
 240         {
 241                 VMSTOR_PROTO_VERSION_WIN8,
 242                 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
 243                 0
 244         },
 245         {
 246                 VMSTOR_PROTO_VERSION_WIN7,
 247                 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
 248                 sizeof(struct vmscsi_win8_extension),
 249         },
 250         {
 251                 VMSTOR_PROTO_VERSION_WIN6,
 252                 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
 253                 sizeof(struct vmscsi_win8_extension),
 254         }
 255 };
 256
 257
 258 /*
 259  * This structure is sent during the initialization phase to get the different
 260  * properties of the channel.
 261  */
 262
 263 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL          0x1
 264
 265 struct vmstorage_channel_properties {
 266         u32 reserved;
 267         u16 max_channel_cnt;
 268         u16 reserved1;
 269
 270         u32 flags;
 271         u32   max_transfer_bytes;
 272
 273         u64  reserved2;
 274 } __packed;
 275
 276 /*  This structure is sent during the storage protocol negotiations. */
 277 struct vmstorage_protocol_version {
 278         /* Major (MSW) and minor (LSW) version numbers. */
 279         u16 major_minor;
 280
 281         /*
 282          * Revision number is auto-incremented whenever this file is changed
 283          * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
 284          * definitely indicate incompatibility--but it does indicate mismatched
 285          * builds.
 286          * This is only used on the windows side. Just set it to 0.
 287          */
 288         u16 revision;
 289 } __packed;
 290
 291 /* Channel Property Flags */
 292 #define STORAGE_CHANNEL_REMOVABLE_FLAG          0x1
 293 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG       0x2
 294
 295 struct vstor_packet {
 296         /* Requested operation type */
 297         enum vstor_packet_operation operation;
 298
 299         /*  Flags - see below for values */
 300         u32 flags;
 301
 302         /* Status of the request returned from the server side. */
 303         u32 status;
 304
 305         /* Data payload area */
 306         union {
 307                 /*
 308                  * Structure used to forward SCSI commands from the
 309                  * client to the server.
 310                  */
 311                 struct vmscsi_request vm_srb;
 312
 313                 /* Structure used to query channel properties. */
 314                 struct vmstorage_channel_properties storage_channel_properties;
 315
 316                 /* Used during version negotiations. */
 317                 struct vmstorage_protocol_version version;
 318
 319                 /* Fibre channel address packet */
 320                 struct hv_fc_wwn_packet wwn_packet;
 321
 322                 /* Number of sub-channels to create */
 323                 u16 sub_channel_count;
 324
 325                 /* This will be the maximum of the union members */
 326                 u8  buffer[0x34];
 327         };
 328 } __packed;
 329
 330 /*
 331  * Packet Flags:
 332  *
 333  * This flag indicates that the server should send back a completion for this
 334  * packet.
 335  */
 336
 337 #define REQUEST_COMPLETION_FLAG 0x1
 338
 339 /* Matches Windows-end */
 340 enum storvsc_request_type {
 341         WRITE_TYPE = 0,
 342         READ_TYPE,
 343         UNKNOWN_TYPE,
 344 };
 345
 346 /*
 347  * SRB status codes and masks; a subset of the codes used here.
 348  */
 349
 350 #define SRB_STATUS_AUTOSENSE_VALID      0x80
 351 #define SRB_STATUS_QUEUE_FROZEN         0x40
 352 #define SRB_STATUS_INVALID_LUN  0x20
 353 #define SRB_STATUS_SUCCESS      0x01
 354 #define SRB_STATUS_ABORTED      0x02
 355 #define SRB_STATUS_ERROR        0x04
 356 #define SRB_STATUS_DATA_OVERRUN 0x12
 357
 358 #define SRB_STATUS(status) \
 359         (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
 360 /*
 361  * This is the end of Protocol specific defines.
 362  */
 363
 364 static int storvsc_ringbuffer_size = (128 * 1024);
 365 static u32 max_outstanding_req_per_channel;
 366 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
 367
 368 static int storvsc_vcpus_per_sub_channel = 4;
 369 static unsigned int storvsc_max_hw_queues;
 370
 371 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
 372 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
 373
 374 module_param(storvsc_max_hw_queues, uint, 0644);
 375 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
 376
 377 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
 378 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
 379
 380 static int ring_avail_percent_lowater = 10;
 381 module_param(ring_avail_percent_lowater, int, S_IRUGO);
 382 MODULE_PARM_DESC(ring_avail_percent_lowater,
 383                 "Select a channel if available ring size > this in percent");
 384
 385 /*
 386  * Timeout in seconds for all devices managed by this driver.
 387  */
 388 static int storvsc_timeout = 180;
 389
 390 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
 391 static struct scsi_transport_template *fc_transport_template;
 392 #endif
 393
 394 static struct scsi_host_template scsi_driver;
 395 static void storvsc_on_channel_callback(void *context);
 396
 397 #define STORVSC_MAX_LUNS_PER_TARGET                     255
 398 #define STORVSC_MAX_TARGETS                             2
 399 #define STORVSC_MAX_CHANNELS                            8
 400
 401 #define STORVSC_FC_MAX_LUNS_PER_TARGET                  255
 402 #define STORVSC_FC_MAX_TARGETS                          128
 403 #define STORVSC_FC_MAX_CHANNELS                         8
 404
 405 #define STORVSC_IDE_MAX_LUNS_PER_TARGET                 64
 406 #define STORVSC_IDE_MAX_TARGETS                         1
 407 #define STORVSC_IDE_MAX_CHANNELS                        1
 408
 409 /*
 410  * Upper bound on the size of a storvsc packet. vmscsi_size_delta is not
 411  * included in the calculation because it is set after STORVSC_MAX_PKT_SIZE
 412  * is used in storvsc_connect_to_vsp
 413  */
 414 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
 415                               sizeof(struct vstor_packet))
 416
 417 struct storvsc_cmd_request {
 418         struct scsi_cmnd *cmd;
 419
 420         struct hv_device *device;
 421
 422         /* Synchronize the request/response if needed */
 423         struct completion wait_event;
 424
 425         struct vmbus_channel_packet_multipage_buffer mpb;
 426         struct vmbus_packet_mpb_array *payload;
 427         u32 payload_sz;
 428
 429         struct vstor_packet vstor_packet;
 430 };
 431
 432
 433 /* A storvsc device is a device object that contains a vmbus channel */
 434 struct storvsc_device {
 435         struct hv_device *device;
 436
 437         bool     destroy;
 438         bool     drain_notify;
 439         atomic_t num_outstanding_req;
 440         struct Scsi_Host *host;
 441
 442         wait_queue_head_t waiting_to_drain;
 443
 444         /*
 445          * Each unique Port/Path/Target represents 1 channel ie scsi
 446          * controller. In reality, the pathid, targetid is always 0
 447          * and the port is set by us
 448          */
 449         unsigned int port_number;
 450         unsigned char path_id;
 451         unsigned char target_id;
 452
 453         /*
 454          * The size of the vmscsi_request has changed in win8. The
 455          * additional size is because of new elements added to the
 456          * structure. These elements are valid only when we are talking
 457          * to a win8 host.
 458          * Track the correction to size we need to apply. This value
 459          * will likely change during protocol negotiation but it is
 460          * valid to start by assuming pre-Win8.
 461          */
 462         int vmscsi_size_delta;
 463
 464         /*
 465          * Max I/O, the device can support.
 466          */
 467         u32   max_transfer_bytes;
 468         /*
 469          * Number of sub-channels we will open.
 470          */
 471         u16 num_sc;
 472         struct vmbus_channel **stor_chns;
 473         /*
 474          * Mask of CPUs bound to subchannels.
 475          */
 476         struct cpumask alloced_cpus;
 477         /*
 478          * Serializes modifications of stor_chns[] from storvsc_do_io()
 479          * and storvsc_change_target_cpu().
 480          */
 481         spinlock_t lock;
 482         /* Used for vsc/vsp channel reset process */
 483         struct storvsc_cmd_request init_request;
 484         struct storvsc_cmd_request reset_request;
 485         /*
 486          * Currently active port and node names for FC devices.
 487          */
 488         u64 node_name;
 489         u64 port_name;
 490 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
 491         struct fc_rport *rport;
 492 #endif
 493 };
 494
 495 struct hv_host_device {
 496         struct hv_device *dev;
 497         unsigned int port;
 498         unsigned char path;
 499         unsigned char target;
 500         struct workqueue_struct *handle_error_wq;
 501         struct work_struct host_scan_work;
 502         struct Scsi_Host *host;
 503 };
 504
 505 struct storvsc_scan_work {
 506         struct work_struct work;
 507         struct Scsi_Host *host;
 508         u8 lun;
 509         u8 tgt_id;
 510 };
 511
 512 static void storvsc_device_scan(struct work_struct *work)
 513 {
 514         struct storvsc_scan_work *wrk;
 515         struct scsi_device *sdev;
 516
 517         wrk = container_of(work, struct storvsc_scan_work, work);
 518
 519         sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
 520         if (!sdev)
 521                 goto done;
 522         scsi_rescan_device(&sdev->sdev_gendev);
 523         scsi_device_put(sdev);
 524
 525 done:
 526         kfree(wrk);
 527 }
 528
 529 static void storvsc_host_scan(struct work_struct *work)
 530 {
 531         struct Scsi_Host *host;
 532         struct scsi_device *sdev;
 533         struct hv_host_device *host_device =
 534                 container_of(work, struct hv_host_device, host_scan_work);
 535
 536         host = host_device->host;
 537         /*
 538          * Before scanning the host, first check to see if any of the
 539          * currrently known devices have been hot removed. We issue a
 540          * "unit ready" command against all currently known devices.
 541          * This I/O will result in an error for devices that have been
 542          * removed. As part of handling the I/O error, we remove the device.
 543          *
 544          * When a LUN is added or removed, the host sends us a signal to
 545          * scan the host. Thus we are forced to discover the LUNs that
 546          * may have been removed this way.
 547          */
 548         mutex_lock(&host->scan_mutex);
 549         shost_for_each_device(sdev, host)
 550                 scsi_test_unit_ready(sdev, 1, 1, NULL);
 551         mutex_unlock(&host->scan_mutex);
 552         /*
 553          * Now scan the host to discover LUNs that may have been added.
 554          */
 555         scsi_scan_host(host);
 556 }
 557
 558 static void storvsc_remove_lun(struct work_struct *work)
 559 {
 560         struct storvsc_scan_work *wrk;
 561         struct scsi_device *sdev;
 562
 563         wrk = container_of(work, struct storvsc_scan_work, work);
 564         if (!scsi_host_get(wrk->host))
 565                 goto done;
 566
 567         sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
 568
 569         if (sdev) {
 570                 scsi_remove_device(sdev);
 571                 scsi_device_put(sdev);
 572         }
 573         scsi_host_put(wrk->host);
 574
 575 done:
 576         kfree(wrk);
 577 }
 578
 579
 580 /*
 581  * We can get incoming messages from the host that are not in response to
 582  * messages that we have sent out. An example of this would be messages
 583  * received by the guest to notify dynamic addition/removal of LUNs. To
 584  * deal with potential race conditions where the driver may be in the
 585  * midst of being unloaded when we might receive an unsolicited message
 586  * from the host, we have implemented a mechanism to gurantee sequential
 587  * consistency:
 588  *
 589  * 1) Once the device is marked as being destroyed, we will fail all
 590  *    outgoing messages.
 591  * 2) We permit incoming messages when the device is being destroyed,
 592  *    only to properly account for messages already sent out.
 593  */
 594
 595 static inline struct storvsc_device *get_out_stor_device(
 596                                         struct hv_device *device)
 597 {
 598         struct storvsc_device *stor_device;
 599
 600         stor_device = hv_get_drvdata(device);
 601
 602         if (stor_device && stor_device->destroy)
 603                 stor_device = NULL;
 604
 605         return stor_device;
 606 }
 607
 608
 609 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
 610 {
 611         dev->drain_notify = true;
 612         wait_event(dev->waiting_to_drain,
 613                    atomic_read(&dev->num_outstanding_req) == 0);
 614         dev->drain_notify = false;
 615 }
 616
 617 static inline struct storvsc_device *get_in_stor_device(
 618                                         struct hv_device *device)
 619 {
 620         struct storvsc_device *stor_device;
 621
 622         stor_device = hv_get_drvdata(device);
 623
 624         if (!stor_device)
 625                 goto get_in_err;
 626
 627         /*
 628          * If the device is being destroyed; allow incoming
 629          * traffic only to cleanup outstanding requests.
 630          */
 631
 632         if (stor_device->destroy  &&
 633                 (atomic_read(&stor_device->num_outstanding_req) == 0))
 634                 stor_device = NULL;
 635
 636 get_in_err:
 637         return stor_device;
 638
 639 }
 640
 641 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
 642                                       u32 new)
 643 {
 644         struct storvsc_device *stor_device;
 645         struct vmbus_channel *cur_chn;
 646         bool old_is_alloced = false;
 647         struct hv_device *device;
 648         unsigned long flags;
 649         int cpu;
 650
 651         device = channel->primary_channel ?
 652                         channel->primary_channel->device_obj
 653                                 : channel->device_obj;
 654         stor_device = get_out_stor_device(device);
 655         if (!stor_device)
 656                 return;
 657
 658         /* See storvsc_do_io() -> get_og_chn(). */
 659         spin_lock_irqsave(&stor_device->lock, flags);
 660
 661         /*
 662          * Determines if the storvsc device has other channels assigned to
 663          * the "old" CPU to update the alloced_cpus mask and the stor_chns
 664          * array.
 665          */
 666         if (device->channel != channel && device->channel->target_cpu == old) {
 667                 cur_chn = device->channel;
 668                 old_is_alloced = true;
 669                 goto old_is_alloced;
 670         }
 671         list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
 672                 if (cur_chn == channel)
 673                         continue;
 674                 if (cur_chn->target_cpu == old) {
 675                         old_is_alloced = true;
 676                         goto old_is_alloced;
 677                 }
 678         }
 679
 680 old_is_alloced:
 681         if (old_is_alloced)
 682                 WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
 683         else
 684                 cpumask_clear_cpu(old, &stor_device->alloced_cpus);
 685
 686         /* "Flush" the stor_chns array. */
 687         for_each_possible_cpu(cpu) {
 688                 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
 689                                         cpu, &stor_device->alloced_cpus))
 690                         WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
 691         }
 692
 693         WRITE_ONCE(stor_device->stor_chns[new], channel);
 694         cpumask_set_cpu(new, &stor_device->alloced_cpus);
 695
 696         spin_unlock_irqrestore(&stor_device->lock, flags);
 697 }
 698
 699 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
 700 {
 701         struct storvsc_cmd_request *request =
 702                 (struct storvsc_cmd_request *)(unsigned long)rqst_addr;
 703
 704         if (rqst_addr == VMBUS_RQST_INIT)
 705                 return VMBUS_RQST_INIT;
 706         if (rqst_addr == VMBUS_RQST_RESET)
 707                 return VMBUS_RQST_RESET;
 708
 709         /*
 710          * Cannot return an ID of 0, which is reserved for an unsolicited
 711          * message from Hyper-V.
 712          */
 713         return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
 714 }
 715
 716 static void handle_sc_creation(struct vmbus_channel *new_sc)
 717 {
 718         struct hv_device *device = new_sc->primary_channel->device_obj;
 719         struct device *dev = &device->device;
 720         struct storvsc_device *stor_device;
 721         struct vmstorage_channel_properties props;
 722         int ret;
 723
 724         stor_device = get_out_stor_device(device);
 725         if (!stor_device)
 726                 return;
 727
 728         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
 729         new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
 730
 731         new_sc->next_request_id_callback = storvsc_next_request_id;
 732
 733         ret = vmbus_open(new_sc,
 734                          storvsc_ringbuffer_size,
 735                          storvsc_ringbuffer_size,
 736                          (void *)&props,
 737                          sizeof(struct vmstorage_channel_properties),
 738                          storvsc_on_channel_callback, new_sc);
 739
 740         /* In case vmbus_open() fails, we don't use the sub-channel. */
 741         if (ret != 0) {
 742                 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
 743                 return;
 744         }
 745
 746         new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
 747
 748         /* Add the sub-channel to the array of available channels. */
 749         stor_device->stor_chns[new_sc->target_cpu] = new_sc;
 750         cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
 751 }
 752
 753 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
 754 {
 755         struct device *dev = &device->device;
 756         struct storvsc_device *stor_device;
 757         int num_sc;
 758         struct storvsc_cmd_request *request;
 759         struct vstor_packet *vstor_packet;
 760         int ret, t;
 761
 762         /*
 763          * If the number of CPUs is artificially restricted, such as
 764          * with maxcpus=1 on the kernel boot line, Hyper-V could offer
 765          * sub-channels >= the number of CPUs. These sub-channels
 766          * should not be created. The primary channel is already created
 767          * and assigned to one CPU, so check against # CPUs - 1.
 768          */
 769         num_sc = min((int)(num_online_cpus() - 1), max_chns);
 770         if (!num_sc)
 771                 return;
 772
 773         stor_device = get_out_stor_device(device);
 774         if (!stor_device)
 775                 return;
 776
 777         stor_device->num_sc = num_sc;
 778         request = &stor_device->init_request;
 779         vstor_packet = &request->vstor_packet;
 780
 781         /*
 782          * Establish a handler for dealing with subchannels.
 783          */
 784         vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
 785
 786         /*
 787          * Request the host to create sub-channels.
 788          */
 789         memset(request, 0, sizeof(struct storvsc_cmd_request));
 790         init_completion(&request->wait_event);
 791         vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
 792         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 793         vstor_packet->sub_channel_count = num_sc;
 794
 795         ret = vmbus_sendpacket(device->channel, vstor_packet,
 796                                (sizeof(struct vstor_packet) -
 797                                stor_device->vmscsi_size_delta),
 798                                VMBUS_RQST_INIT,
 799                                VM_PKT_DATA_INBAND,
 800                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 801
 802         if (ret != 0) {
 803                 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
 804                 return;
 805         }
 806
 807         t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
 808         if (t == 0) {
 809                 dev_err(dev, "Failed to create sub-channel: timed out\n");
 810                 return;
 811         }
 812
 813         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 814             vstor_packet->status != 0) {
 815                 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
 816                         vstor_packet->operation, vstor_packet->status);
 817                 return;
 818         }
 819
 820         /*
 821          * We need to do nothing here, because vmbus_process_offer()
 822          * invokes channel->sc_creation_callback, which will open and use
 823          * the sub-channel(s).
 824          */
 825 }
 826
 827 static void cache_wwn(struct storvsc_device *stor_device,
 828                       struct vstor_packet *vstor_packet)
 829 {
 830         /*
 831          * Cache the currently active port and node ww names.
 832          */
 833         if (vstor_packet->wwn_packet.primary_active) {
 834                 stor_device->node_name =
 835                         wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
 836                 stor_device->port_name =
 837                         wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
 838         } else {
 839                 stor_device->node_name =
 840                         wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
 841                 stor_device->port_name =
 842                         wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
 843         }
 844 }
 845
 846
 847 static int storvsc_execute_vstor_op(struct hv_device *device,
 848                                     struct storvsc_cmd_request *request,
 849                                     bool status_check)
 850 {
 851         struct storvsc_device *stor_device;
 852         struct vstor_packet *vstor_packet;
 853         int ret, t;
 854
 855         stor_device = get_out_stor_device(device);
 856         if (!stor_device)
 857                 return -ENODEV;
 858
 859         vstor_packet = &request->vstor_packet;
 860
 861         init_completion(&request->wait_event);
 862         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 863
 864         ret = vmbus_sendpacket(device->channel, vstor_packet,
 865                                (sizeof(struct vstor_packet) -
 866                                stor_device->vmscsi_size_delta),
 867                                VMBUS_RQST_INIT,
 868                                VM_PKT_DATA_INBAND,
 869                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 870         if (ret != 0)
 871                 return ret;
 872
 873         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 874         if (t == 0)
 875                 return -ETIMEDOUT;
 876
 877         if (!status_check)
 878                 return ret;
 879
 880         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 881             vstor_packet->status != 0)
 882                 return -EINVAL;
 883
 884         return ret;
 885 }
 886
 887 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
 888 {
 889         struct storvsc_device *stor_device;
 890         struct storvsc_cmd_request *request;
 891         struct vstor_packet *vstor_packet;
 892         int ret, i;
 893         int max_chns;
 894         bool process_sub_channels = false;
 895
 896         stor_device = get_out_stor_device(device);
 897         if (!stor_device)
 898                 return -ENODEV;
 899
 900         request = &stor_device->init_request;
 901         vstor_packet = &request->vstor_packet;
 902
 903         /*
 904          * Now, initiate the vsc/vsp initialization protocol on the open
 905          * channel
 906          */
 907         memset(request, 0, sizeof(struct storvsc_cmd_request));
 908         vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
 909         ret = storvsc_execute_vstor_op(device, request, true);
 910         if (ret)
 911                 return ret;
 912         /*
 913          * Query host supported protocol version.
 914          */
 915
 916         for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) {
 917                 /* reuse the packet for version range supported */
 918                 memset(vstor_packet, 0, sizeof(struct vstor_packet));
 919                 vstor_packet->operation =
 920                         VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
 921
 922                 vstor_packet->version.major_minor =
 923                         vmstor_protocols[i].protocol_version;
 924
 925                 /*
 926                  * The revision number is only used in Windows; set it to 0.
 927                  */
 928                 vstor_packet->version.revision = 0;
 929                 ret = storvsc_execute_vstor_op(device, request, false);
 930                 if (ret != 0)
 931                         return ret;
 932
 933                 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
 934                         return -EINVAL;
 935
 936                 if (vstor_packet->status == 0) {
 937                         vmstor_proto_version =
 938                                 vmstor_protocols[i].protocol_version;
 939
 940                         sense_buffer_size =
 941                                 vmstor_protocols[i].sense_buffer_size;
 942
 943                         stor_device->vmscsi_size_delta =
 944                                 vmstor_protocols[i].vmscsi_size_delta;
 945
 946                         break;
 947                 }
 948         }
 949
 950         if (vstor_packet->status != 0)
 951                 return -EINVAL;
 952
 953
 954         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 955         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
 956         ret = storvsc_execute_vstor_op(device, request, true);
 957         if (ret != 0)
 958                 return ret;
 959
 960         /*
 961          * Check to see if multi-channel support is there.
 962          * Hosts that implement protocol version of 5.1 and above
 963          * support multi-channel.
 964          */
 965         max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
 966
 967         /*
 968          * Allocate state to manage the sub-channels.
 969          * We allocate an array based on the numbers of possible CPUs
 970          * (Hyper-V does not support cpu online/offline).
 971          * This Array will be sparseley populated with unique
 972          * channels - primary + sub-channels.
 973          * We will however populate all the slots to evenly distribute
 974          * the load.
 975          */
 976         stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
 977                                          GFP_KERNEL);
 978         if (stor_device->stor_chns == NULL)
 979                 return -ENOMEM;
 980
 981         device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
 982
 983         stor_device->stor_chns[device->channel->target_cpu] = device->channel;
 984         cpumask_set_cpu(device->channel->target_cpu,
 985                         &stor_device->alloced_cpus);
 986
 987         if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) {
 988                 if (vstor_packet->storage_channel_properties.flags &
 989                     STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
 990                         process_sub_channels = true;
 991         }
 992         stor_device->max_transfer_bytes =
 993                 vstor_packet->storage_channel_properties.max_transfer_bytes;
 994
 995         if (!is_fc)
 996                 goto done;
 997
 998         /*
 999          * For FC devices retrieve FC HBA data.
1000          */
1001         memset(vstor_packet, 0, sizeof(struct vstor_packet));
1002         vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
1003         ret = storvsc_execute_vstor_op(device, request, true);
1004         if (ret != 0)
1005                 return ret;
1006
1007         /*
1008          * Cache the currently active port and node ww names.
1009          */
1010         cache_wwn(stor_device, vstor_packet);
1011
1012 done:
1013
1014         memset(vstor_packet, 0, sizeof(struct vstor_packet));
1015         vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
1016         ret = storvsc_execute_vstor_op(device, request, true);
1017         if (ret != 0)
1018                 return ret;
1019
1020         if (process_sub_channels)
1021                 handle_multichannel_storage(device, max_chns);
1022
1023         return ret;
1024 }
1025
1026 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
1027                                 struct scsi_cmnd *scmnd,
1028                                 struct Scsi_Host *host,
1029                                 u8 asc, u8 ascq)
1030 {
1031         struct storvsc_scan_work *wrk;
1032         void (*process_err_fn)(struct work_struct *work);
1033         struct hv_host_device *host_dev = shost_priv(host);
1034
1035         /*
1036          * In some situations, Hyper-V sets multiple bits in the
1037          * srb_status, such as ABORTED and ERROR. So process them
1038          * individually, with the most specific bits first.
1039          */
1040
1041         if (vm_srb->srb_status & SRB_STATUS_INVALID_LUN) {
1042                 set_host_byte(scmnd, DID_NO_CONNECT);
1043                 process_err_fn = storvsc_remove_lun;
1044                 goto do_work;
1045         }
1046
1047         if (vm_srb->srb_status & SRB_STATUS_ABORTED) {
1048                 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
1049                     /* Capacity data has changed */
1050                     (asc == 0x2a) && (ascq == 0x9)) {
1051                         process_err_fn = storvsc_device_scan;
1052                         /*
1053                          * Retry the I/O that triggered this.
1054                          */
1055                         set_host_byte(scmnd, DID_REQUEUE);
1056                         goto do_work;
1057                 }
1058         }
1059
1060         if (vm_srb->srb_status & SRB_STATUS_ERROR) {
1061                 /*
1062                  * Let upper layer deal with error when
1063                  * sense message is present.
1064                  */
1065                 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
1066                         return;
1067
1068                 /*
1069                  * If there is an error; offline the device since all
1070                  * error recovery strategies would have already been
1071                  * deployed on the host side. However, if the command
1072                  * were a pass-through command deal with it appropriately.
1073                  */
1074                 switch (scmnd->cmnd[0]) {
1075                 case ATA_16:
1076                 case ATA_12:
1077                         set_host_byte(scmnd, DID_PASSTHROUGH);
1078                         break;
1079                 /*
1080                  * On some Hyper-V hosts TEST_UNIT_READY command can
1081                  * return SRB_STATUS_ERROR. Let the upper level code
1082                  * deal with it based on the sense information.
1083                  */
1084                 case TEST_UNIT_READY:
1085                         break;
1086                 default:
1087                         set_host_byte(scmnd, DID_ERROR);
1088                 }
1089         }
1090         return;
1091
1092 do_work:
1093         /*
1094          * We need to schedule work to process this error; schedule it.
1095          */
1096         wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1097         if (!wrk) {
1098                 set_host_byte(scmnd, DID_TARGET_FAILURE);
1099                 return;
1100         }
1101
1102         wrk->host = host;
1103         wrk->lun = vm_srb->lun;
1104         wrk->tgt_id = vm_srb->target_id;
1105         INIT_WORK(&wrk->work, process_err_fn);
1106         queue_work(host_dev->handle_error_wq, &wrk->work);
1107 }
1108
1109
1110 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1111                                        struct storvsc_device *stor_dev)
1112 {
1113         struct scsi_cmnd *scmnd = cmd_request->cmd;
1114         struct scsi_sense_hdr sense_hdr;
1115         struct vmscsi_request *vm_srb;
1116         u32 data_transfer_length;
1117         struct Scsi_Host *host;
1118         u32 payload_sz = cmd_request->payload_sz;
1119         void *payload = cmd_request->payload;
1120         bool sense_ok;
1121
1122         host = stor_dev->host;
1123
1124         vm_srb = &cmd_request->vstor_packet.vm_srb;
1125         data_transfer_length = vm_srb->data_transfer_length;
1126
1127         scmnd->result = vm_srb->scsi_status;
1128
1129         if (scmnd->result) {
1130                 sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1131                                 SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1132
1133                 if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1134                         scsi_print_sense_hdr(scmnd->device, "storvsc",
1135                                              &sense_hdr);
1136         }
1137
1138         if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1139                 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1140                                          sense_hdr.ascq);
1141                 /*
1142                  * The Windows driver set data_transfer_length on
1143                  * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1144                  * is untouched.  In these cases we set it to 0.
1145                  */
1146                 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1147                         data_transfer_length = 0;
1148         }
1149
1150         /* Validate data_transfer_length (from Hyper-V) */
1151         if (data_transfer_length > cmd_request->payload->range.len)
1152                 data_transfer_length = cmd_request->payload->range.len;
1153
1154         scsi_set_resid(scmnd,
1155                 cmd_request->payload->range.len - data_transfer_length);
1156
1157         scmnd->scsi_done(scmnd);
1158
1159         if (payload_sz >
1160                 sizeof(struct vmbus_channel_packet_multipage_buffer))
1161                 kfree(payload);
1162 }
1163
1164 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1165                                   struct vstor_packet *vstor_packet,
1166                                   struct storvsc_cmd_request *request)
1167 {
1168         struct vstor_packet *stor_pkt;
1169         struct hv_device *device = stor_device->device;
1170
1171         stor_pkt = &request->vstor_packet;
1172
1173         /*
1174          * The current SCSI handling on the host side does
1175          * not correctly handle:
1176          * INQUIRY command with page code parameter set to 0x80
1177          * MODE_SENSE command with cmd[2] == 0x1c
1178          *
1179          * Setup srb and scsi status so this won't be fatal.
1180          * We do this so we can distinguish truly fatal failues
1181          * (srb status == 0x4) and off-line the device in that case.
1182          */
1183
1184         if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1185            (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1186                 vstor_packet->vm_srb.scsi_status = 0;
1187                 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1188         }
1189
1190         /* Copy over the status...etc */
1191         stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1192         stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1193
1194         /*
1195          * Copy over the sense_info_length, but limit to the known max
1196          * size if Hyper-V returns a bad value.
1197          */
1198         stor_pkt->vm_srb.sense_info_length = min_t(u8, sense_buffer_size,
1199                 vstor_packet->vm_srb.sense_info_length);
1200
1201         if (vstor_packet->vm_srb.scsi_status != 0 ||
1202             vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS)
1203                 storvsc_log(device, STORVSC_LOGGING_ERROR,
1204                         "tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1205                         scsi_cmd_to_rq(request->cmd)->tag,
1206                         stor_pkt->vm_srb.cdb[0],
1207                         vstor_packet->vm_srb.scsi_status,
1208                         vstor_packet->vm_srb.srb_status,
1209                         vstor_packet->status);
1210
1211         if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1212             (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1213                 memcpy(request->cmd->sense_buffer,
1214                        vstor_packet->vm_srb.sense_data,
1215                        stor_pkt->vm_srb.sense_info_length);
1216
1217         stor_pkt->vm_srb.data_transfer_length =
1218                 vstor_packet->vm_srb.data_transfer_length;
1219
1220         storvsc_command_completion(request, stor_device);
1221
1222         if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1223                 stor_device->drain_notify)
1224                 wake_up(&stor_device->waiting_to_drain);
1225 }
1226
1227 static void storvsc_on_receive(struct storvsc_device *stor_device,
1228                              struct vstor_packet *vstor_packet,
1229                              struct storvsc_cmd_request *request)
1230 {
1231         struct hv_host_device *host_dev;
1232         switch (vstor_packet->operation) {
1233         case VSTOR_OPERATION_COMPLETE_IO:
1234                 storvsc_on_io_completion(stor_device, vstor_packet, request);
1235                 break;
1236
1237         case VSTOR_OPERATION_REMOVE_DEVICE:
1238         case VSTOR_OPERATION_ENUMERATE_BUS:
1239                 host_dev = shost_priv(stor_device->host);
1240                 queue_work(
1241                         host_dev->handle_error_wq, &host_dev->host_scan_work);
1242                 break;
1243
1244         case VSTOR_OPERATION_FCHBA_DATA:
1245                 cache_wwn(stor_device, vstor_packet);
1246 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1247                 fc_host_node_name(stor_device->host) = stor_device->node_name;
1248                 fc_host_port_name(stor_device->host) = stor_device->port_name;
1249 #endif
1250                 break;
1251         default:
1252                 break;
1253         }
1254 }
1255
1256 static void storvsc_on_channel_callback(void *context)
1257 {
1258         struct vmbus_channel *channel = (struct vmbus_channel *)context;
1259         const struct vmpacket_descriptor *desc;
1260         struct hv_device *device;
1261         struct storvsc_device *stor_device;
1262         struct Scsi_Host *shost;
1263
1264         if (channel->primary_channel != NULL)
1265                 device = channel->primary_channel->device_obj;
1266         else
1267                 device = channel->device_obj;
1268
1269         stor_device = get_in_stor_device(device);
1270         if (!stor_device)
1271                 return;
1272
1273         shost = stor_device->host;
1274
1275         foreach_vmbus_pkt(desc, channel) {
1276                 struct vstor_packet *packet = hv_pkt_data(desc);
1277                 struct storvsc_cmd_request *request = NULL;
1278                 u64 rqst_id = desc->trans_id;
1279
1280                 if (hv_pkt_datalen(desc) < sizeof(struct vstor_packet) -
1281                                 stor_device->vmscsi_size_delta) {
1282                         dev_err(&device->device, "Invalid packet len\n");
1283                         continue;
1284                 }
1285
1286                 if (rqst_id == VMBUS_RQST_INIT) {
1287                         request = &stor_device->init_request;
1288                 } else if (rqst_id == VMBUS_RQST_RESET) {
1289                         request = &stor_device->reset_request;
1290                 } else {
1291                         /* Hyper-V can send an unsolicited message with ID of 0 */
1292                         if (rqst_id == 0) {
1293                                 /*
1294                                  * storvsc_on_receive() looks at the vstor_packet in the message
1295                                  * from the ring buffer.  If the operation in the vstor_packet is
1296                                  * COMPLETE_IO, then we call storvsc_on_io_completion(), and
1297                                  * dereference the guest memory address.  Make sure we don't call
1298                                  * storvsc_on_io_completion() with a guest memory address that is
1299                                  * zero if Hyper-V were to construct and send such a bogus packet.
1300                                  */
1301                                 if (packet->operation == VSTOR_OPERATION_COMPLETE_IO) {
1302                                         dev_err(&device->device, "Invalid packet with ID of 0\n");
1303                                         continue;
1304                                 }
1305                         } else {
1306                                 struct scsi_cmnd *scmnd;
1307
1308                                 /* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1309                                 scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1310                                 if (scmnd == NULL) {
1311                                         dev_err(&device->device, "Incorrect transaction ID\n");
1312                                         continue;
1313                                 }
1314                                 request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1315                         }
1316
1317                         storvsc_on_receive(stor_device, packet, request);
1318                         continue;
1319                 }
1320
1321                 memcpy(&request->vstor_packet, packet,
1322                        (sizeof(struct vstor_packet) - stor_device->vmscsi_size_delta));
1323                 complete(&request->wait_event);
1324         }
1325 }
1326
1327 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1328                                   bool is_fc)
1329 {
1330         struct vmstorage_channel_properties props;
1331         int ret;
1332
1333         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1334
1335         device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1336         device->channel->next_request_id_callback = storvsc_next_request_id;
1337
1338         ret = vmbus_open(device->channel,
1339                          ring_size,
1340                          ring_size,
1341                          (void *)&props,
1342                          sizeof(struct vmstorage_channel_properties),
1343                          storvsc_on_channel_callback, device->channel);
1344
1345         if (ret != 0)
1346                 return ret;
1347
1348         ret = storvsc_channel_init(device, is_fc);
1349
1350         return ret;
1351 }
1352
1353 static int storvsc_dev_remove(struct hv_device *device)
1354 {
1355         struct storvsc_device *stor_device;
1356
1357         stor_device = hv_get_drvdata(device);
1358
1359         stor_device->destroy = true;
1360
1361         /* Make sure flag is set before waiting */
1362         wmb();
1363
1364         /*
1365          * At this point, all outbound traffic should be disable. We
1366          * only allow inbound traffic (responses) to proceed so that
1367          * outstanding requests can be completed.
1368          */
1369
1370         storvsc_wait_to_drain(stor_device);
1371
1372         /*
1373          * Since we have already drained, we don't need to busy wait
1374          * as was done in final_release_stor_device()
1375          * Note that we cannot set the ext pointer to NULL until
1376          * we have drained - to drain the outgoing packets, we need to
1377          * allow incoming packets.
1378          */
1379         hv_set_drvdata(device, NULL);
1380
1381         /* Close the channel */
1382         vmbus_close(device->channel);
1383
1384         kfree(stor_device->stor_chns);
1385         kfree(stor_device);
1386         return 0;
1387 }
1388
1389 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1390                                         u16 q_num)
1391 {
1392         u16 slot = 0;
1393         u16 hash_qnum;
1394         const struct cpumask *node_mask;
1395         int num_channels, tgt_cpu;
1396
1397         if (stor_device->num_sc == 0) {
1398                 stor_device->stor_chns[q_num] = stor_device->device->channel;
1399                 return stor_device->device->channel;
1400         }
1401
1402         /*
1403          * Our channel array is sparsley populated and we
1404          * initiated I/O on a processor/hw-q that does not
1405          * currently have a designated channel. Fix this.
1406          * The strategy is simple:
1407          * I. Ensure NUMA locality
1408          * II. Distribute evenly (best effort)
1409          */
1410
1411         node_mask = cpumask_of_node(cpu_to_node(q_num));
1412
1413         num_channels = 0;
1414         for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1415                 if (cpumask_test_cpu(tgt_cpu, node_mask))
1416                         num_channels++;
1417         }
1418         if (num_channels == 0) {
1419                 stor_device->stor_chns[q_num] = stor_device->device->channel;
1420                 return stor_device->device->channel;
1421         }
1422
1423         hash_qnum = q_num;
1424         while (hash_qnum >= num_channels)
1425                 hash_qnum -= num_channels;
1426
1427         for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1428                 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1429                         continue;
1430                 if (slot == hash_qnum)
1431                         break;
1432                 slot++;
1433         }
1434
1435         stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1436
1437         return stor_device->stor_chns[q_num];
1438 }
1439
1440
1441 static int storvsc_do_io(struct hv_device *device,
1442                          struct storvsc_cmd_request *request, u16 q_num)
1443 {
1444         struct storvsc_device *stor_device;
1445         struct vstor_packet *vstor_packet;
1446         struct vmbus_channel *outgoing_channel, *channel;
1447         unsigned long flags;
1448         int ret = 0;
1449         const struct cpumask *node_mask;
1450         int tgt_cpu;
1451
1452         vstor_packet = &request->vstor_packet;
1453         stor_device = get_out_stor_device(device);
1454
1455         if (!stor_device)
1456                 return -ENODEV;
1457
1458
1459         request->device  = device;
1460         /*
1461          * Select an appropriate channel to send the request out.
1462          */
1463         /* See storvsc_change_target_cpu(). */
1464         outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1465         if (outgoing_channel != NULL) {
1466                 if (outgoing_channel->target_cpu == q_num) {
1467                         /*
1468                          * Ideally, we want to pick a different channel if
1469                          * available on the same NUMA node.
1470                          */
1471                         node_mask = cpumask_of_node(cpu_to_node(q_num));
1472                         for_each_cpu_wrap(tgt_cpu,
1473                                  &stor_device->alloced_cpus, q_num + 1) {
1474                                 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1475                                         continue;
1476                                 if (tgt_cpu == q_num)
1477                                         continue;
1478                                 channel = READ_ONCE(
1479                                         stor_device->stor_chns[tgt_cpu]);
1480                                 if (channel == NULL)
1481                                         continue;
1482                                 if (hv_get_avail_to_write_percent(
1483                                                         &channel->outbound)
1484                                                 > ring_avail_percent_lowater) {
1485                                         outgoing_channel = channel;
1486                                         goto found_channel;
1487                                 }
1488                         }
1489
1490                         /*
1491                          * All the other channels on the same NUMA node are
1492                          * busy. Try to use the channel on the current CPU
1493                          */
1494                         if (hv_get_avail_to_write_percent(
1495                                                 &outgoing_channel->outbound)
1496                                         > ring_avail_percent_lowater)
1497                                 goto found_channel;
1498
1499                         /*
1500                          * If we reach here, all the channels on the current
1501                          * NUMA node are busy. Try to find a channel in
1502                          * other NUMA nodes
1503                          */
1504                         for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1505                                 if (cpumask_test_cpu(tgt_cpu, node_mask))
1506                                         continue;
1507                                 channel = READ_ONCE(
1508                                         stor_device->stor_chns[tgt_cpu]);
1509                                 if (channel == NULL)
1510                                         continue;
1511                                 if (hv_get_avail_to_write_percent(
1512                                                         &channel->outbound)
1513                                                 > ring_avail_percent_lowater) {
1514                                         outgoing_channel = channel;
1515                                         goto found_channel;
1516                                 }
1517                         }
1518                 }
1519         } else {
1520                 spin_lock_irqsave(&stor_device->lock, flags);
1521                 outgoing_channel = stor_device->stor_chns[q_num];
1522                 if (outgoing_channel != NULL) {
1523                         spin_unlock_irqrestore(&stor_device->lock, flags);
1524                         goto found_channel;
1525                 }
1526                 outgoing_channel = get_og_chn(stor_device, q_num);
1527                 spin_unlock_irqrestore(&stor_device->lock, flags);
1528         }
1529
1530 found_channel:
1531         vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1532
1533         vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1534                                         stor_device->vmscsi_size_delta);
1535
1536
1537         vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1538
1539
1540         vstor_packet->vm_srb.data_transfer_length =
1541         request->payload->range.len;
1542
1543         vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1544
1545         if (request->payload->range.len) {
1546
1547                 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1548                                 request->payload, request->payload_sz,
1549                                 vstor_packet,
1550                                 (sizeof(struct vstor_packet) -
1551                                 stor_device->vmscsi_size_delta),
1552                                 (unsigned long)request);
1553         } else {
1554                 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1555                                (sizeof(struct vstor_packet) -
1556                                 stor_device->vmscsi_size_delta),
1557                                (unsigned long)request,
1558                                VM_PKT_DATA_INBAND,
1559                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1560         }
1561
1562         if (ret != 0)
1563                 return ret;
1564
1565         atomic_inc(&stor_device->num_outstanding_req);
1566
1567         return ret;
1568 }
1569
1570 static int storvsc_device_alloc(struct scsi_device *sdevice)
1571 {
1572         /*
1573          * Set blist flag to permit the reading of the VPD pages even when
1574          * the target may claim SPC-2 compliance. MSFT targets currently
1575          * claim SPC-2 compliance while they implement post SPC-2 features.
1576          * With this flag we can correctly handle WRITE_SAME_16 issues.
1577          *
1578          * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1579          * still supports REPORT LUN.
1580          */
1581         sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1582
1583         return 0;
1584 }
1585
1586 static int storvsc_device_configure(struct scsi_device *sdevice)
1587 {
1588         blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1589
1590         sdevice->no_write_same = 1;
1591
1592         /*
1593          * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1594          * if the device is a MSFT virtual device.  If the host is
1595          * WIN10 or newer, allow write_same.
1596          */
1597         if (!strncmp(sdevice->vendor, "Msft", 4)) {
1598                 switch (vmstor_proto_version) {
1599                 case VMSTOR_PROTO_VERSION_WIN8:
1600                 case VMSTOR_PROTO_VERSION_WIN8_1:
1601                         sdevice->scsi_level = SCSI_SPC_3;
1602                         break;
1603                 }
1604
1605                 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1606                         sdevice->no_write_same = 0;
1607         }
1608
1609         return 0;
1610 }
1611
1612 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1613                            sector_t capacity, int *info)
1614 {
1615         sector_t nsect = capacity;
1616         sector_t cylinders = nsect;
1617         int heads, sectors_pt;
1618
1619         /*
1620          * We are making up these values; let us keep it simple.
1621          */
1622         heads = 0xff;
1623         sectors_pt = 0x3f;      /* Sectors per track */
1624         sector_div(cylinders, heads * sectors_pt);
1625         if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1626                 cylinders = 0xffff;
1627
1628         info[0] = heads;
1629         info[1] = sectors_pt;
1630         info[2] = (int)cylinders;
1631
1632         return 0;
1633 }
1634
1635 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1636 {
1637         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1638         struct hv_device *device = host_dev->dev;
1639
1640         struct storvsc_device *stor_device;
1641         struct storvsc_cmd_request *request;
1642         struct vstor_packet *vstor_packet;
1643         int ret, t;
1644
1645         stor_device = get_out_stor_device(device);
1646         if (!stor_device)
1647                 return FAILED;
1648
1649         request = &stor_device->reset_request;
1650         vstor_packet = &request->vstor_packet;
1651         memset(vstor_packet, 0, sizeof(struct vstor_packet));
1652
1653         init_completion(&request->wait_event);
1654
1655         vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1656         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1657         vstor_packet->vm_srb.path_id = stor_device->path_id;
1658
1659         ret = vmbus_sendpacket(device->channel, vstor_packet,
1660                                (sizeof(struct vstor_packet) -
1661                                 stor_device->vmscsi_size_delta),
1662                                VMBUS_RQST_RESET,
1663                                VM_PKT_DATA_INBAND,
1664                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1665         if (ret != 0)
1666                 return FAILED;
1667
1668         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1669         if (t == 0)
1670                 return TIMEOUT_ERROR;
1671
1672
1673         /*
1674          * At this point, all outstanding requests in the adapter
1675          * should have been flushed out and return to us
1676          * There is a potential race here where the host may be in
1677          * the process of responding when we return from here.
1678          * Just wait for all in-transit packets to be accounted for
1679          * before we return from here.
1680          */
1681         storvsc_wait_to_drain(stor_device);
1682
1683         return SUCCESS;
1684 }
1685
1686 /*
1687  * The host guarantees to respond to each command, although I/O latencies might
1688  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1689  * chance to perform EH.
1690  */
1691 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1692 {
1693 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1694         if (scmnd->device->host->transportt == fc_transport_template)
1695                 return fc_eh_timed_out(scmnd);
1696 #endif
1697         return BLK_EH_RESET_TIMER;
1698 }
1699
1700 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1701 {
1702         bool allowed = true;
1703         u8 scsi_op = scmnd->cmnd[0];
1704
1705         switch (scsi_op) {
1706         /* the host does not handle WRITE_SAME, log accident usage */
1707         case WRITE_SAME:
1708         /*
1709          * smartd sends this command and the host does not handle
1710          * this. So, don't send it.
1711          */
1712         case SET_WINDOW:
1713                 set_host_byte(scmnd, DID_ERROR);
1714                 allowed = false;
1715                 break;
1716         default:
1717                 break;
1718         }
1719         return allowed;
1720 }
1721
1722 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1723 {
1724         int ret;
1725         struct hv_host_device *host_dev = shost_priv(host);
1726         struct hv_device *dev = host_dev->dev;
1727         struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1728         int i;
1729         struct scatterlist *sgl;
1730         unsigned int sg_count;
1731         struct vmscsi_request *vm_srb;
1732         struct vmbus_packet_mpb_array  *payload;
1733         u32 payload_sz;
1734         u32 length;
1735
1736         if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1737                 /*
1738                  * On legacy hosts filter unimplemented commands.
1739                  * Future hosts are expected to correctly handle
1740                  * unsupported commands. Furthermore, it is
1741                  * possible that some of the currently
1742                  * unsupported commands maybe supported in
1743                  * future versions of the host.
1744                  */
1745                 if (!storvsc_scsi_cmd_ok(scmnd)) {
1746                         scmnd->scsi_done(scmnd);
1747                         return 0;
1748                 }
1749         }
1750
1751         /* Setup the cmd request */
1752         cmd_request->cmd = scmnd;
1753
1754         memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1755         vm_srb = &cmd_request->vstor_packet.vm_srb;
1756         vm_srb->win8_extension.time_out_value = 60;
1757
1758         vm_srb->win8_extension.srb_flags |=
1759                 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1760
1761         if (scmnd->device->tagged_supported) {
1762                 vm_srb->win8_extension.srb_flags |=
1763                 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1764                 vm_srb->win8_extension.queue_tag = SP_UNTAGGED;
1765                 vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST;
1766         }
1767
1768         /* Build the SRB */
1769         switch (scmnd->sc_data_direction) {
1770         case DMA_TO_DEVICE:
1771                 vm_srb->data_in = WRITE_TYPE;
1772                 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1773                 break;
1774         case DMA_FROM_DEVICE:
1775                 vm_srb->data_in = READ_TYPE;
1776                 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1777                 break;
1778         case DMA_NONE:
1779                 vm_srb->data_in = UNKNOWN_TYPE;
1780                 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1781                 break;
1782         default:
1783                 /*
1784                  * This is DMA_BIDIRECTIONAL or something else we are never
1785                  * supposed to see here.
1786                  */
1787                 WARN(1, "Unexpected data direction: %d\n",
1788                      scmnd->sc_data_direction);
1789                 return -EINVAL;
1790         }
1791
1792
1793         vm_srb->port_number = host_dev->port;
1794         vm_srb->path_id = scmnd->device->channel;
1795         vm_srb->target_id = scmnd->device->id;
1796         vm_srb->lun = scmnd->device->lun;
1797
1798         vm_srb->cdb_length = scmnd->cmd_len;
1799
1800         memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1801
1802         sgl = (struct scatterlist *)scsi_sglist(scmnd);
1803         sg_count = scsi_sg_count(scmnd);
1804
1805         length = scsi_bufflen(scmnd);
1806         payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1807         payload_sz = sizeof(cmd_request->mpb);
1808
1809         if (sg_count) {
1810                 unsigned int hvpgoff, hvpfns_to_add;
1811                 unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1812                 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1813                 u64 hvpfn;
1814
1815                 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1816
1817                         payload_sz = (hvpg_count * sizeof(u64) +
1818                                       sizeof(struct vmbus_packet_mpb_array));
1819                         payload = kzalloc(payload_sz, GFP_ATOMIC);
1820                         if (!payload)
1821                                 return SCSI_MLQUEUE_DEVICE_BUSY;
1822                 }
1823
1824                 payload->range.len = length;
1825                 payload->range.offset = offset_in_hvpg;
1826
1827
1828                 for (i = 0; sgl != NULL; sgl = sg_next(sgl)) {
1829                         /*
1830                          * Init values for the current sgl entry. hvpgoff
1831                          * and hvpfns_to_add are in units of Hyper-V size
1832                          * pages. Handling the PAGE_SIZE != HV_HYP_PAGE_SIZE
1833                          * case also handles values of sgl->offset that are
1834                          * larger than PAGE_SIZE. Such offsets are handled
1835                          * even on other than the first sgl entry, provided
1836                          * they are a multiple of PAGE_SIZE.
1837                          */
1838                         hvpgoff = HVPFN_DOWN(sgl->offset);
1839                         hvpfn = page_to_hvpfn(sg_page(sgl)) + hvpgoff;
1840                         hvpfns_to_add = HVPFN_UP(sgl->offset + sgl->length) -
1841                                                 hvpgoff;
1842
1843                         /*
1844                          * Fill the next portion of the PFN array with
1845                          * sequential Hyper-V PFNs for the continguous physical
1846                          * memory described by the sgl entry. The end of the
1847                          * last sgl should be reached at the same time that
1848                          * the PFN array is filled.
1849                          */
1850                         while (hvpfns_to_add--)
1851                                 payload->range.pfn_array[i++] = hvpfn++;
1852                 }
1853         }
1854
1855         cmd_request->payload = payload;
1856         cmd_request->payload_sz = payload_sz;
1857
1858         /* Invokes the vsc to start an IO */
1859         ret = storvsc_do_io(dev, cmd_request, get_cpu());
1860         put_cpu();
1861
1862         if (ret == -EAGAIN) {
1863                 if (payload_sz > sizeof(cmd_request->mpb))
1864                         kfree(payload);
1865                 /* no more space */
1866                 return SCSI_MLQUEUE_DEVICE_BUSY;
1867         }
1868
1869         return 0;
1870 }
1871
1872 static struct scsi_host_template scsi_driver = {
1873         .module =               THIS_MODULE,
1874         .name =                 "storvsc_host_t",
1875         .cmd_size =             sizeof(struct storvsc_cmd_request),
1876         .bios_param =           storvsc_get_chs,
1877         .queuecommand =         storvsc_queuecommand,
1878         .eh_host_reset_handler =        storvsc_host_reset_handler,
1879         .proc_name =            "storvsc_host",
1880         .eh_timed_out =         storvsc_eh_timed_out,
1881         .slave_alloc =          storvsc_device_alloc,
1882         .slave_configure =      storvsc_device_configure,
1883         .cmd_per_lun =          2048,
1884         .this_id =              -1,
1885         /* Ensure there are no gaps in presented sgls */
1886         .virt_boundary_mask =   PAGE_SIZE-1,
1887         .no_write_same =        1,
1888         .track_queue_depth =    1,
1889         .change_queue_depth =   storvsc_change_queue_depth,
1890 };
1891
1892 enum {
1893         SCSI_GUID,
1894         IDE_GUID,
1895         SFC_GUID,
1896 };
1897
1898 static const struct hv_vmbus_device_id id_table[] = {
1899         /* SCSI guid */
1900         { HV_SCSI_GUID,
1901           .driver_data = SCSI_GUID
1902         },
1903         /* IDE guid */
1904         { HV_IDE_GUID,
1905           .driver_data = IDE_GUID
1906         },
1907         /* Fibre Channel GUID */
1908         {
1909           HV_SYNTHFC_GUID,
1910           .driver_data = SFC_GUID
1911         },
1912         { },
1913 };
1914
1915 MODULE_DEVICE_TABLE(vmbus, id_table);
1916
1917 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1918
1919 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1920 {
1921         return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1922 }
1923
1924 static int storvsc_probe(struct hv_device *device,
1925                         const struct hv_vmbus_device_id *dev_id)
1926 {
1927         int ret;
1928         int num_cpus = num_online_cpus();
1929         int num_present_cpus = num_present_cpus();
1930         struct Scsi_Host *host;
1931         struct hv_host_device *host_dev;
1932         bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1933         bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1934         int target = 0;
1935         struct storvsc_device *stor_device;
1936         int max_luns_per_target;
1937         int max_targets;
1938         int max_channels;
1939         int max_sub_channels = 0;
1940
1941         /*
1942          * Based on the windows host we are running on,
1943          * set state to properly communicate with the host.
1944          */
1945
1946         if (vmbus_proto_version < VERSION_WIN8) {
1947                 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1948                 max_targets = STORVSC_IDE_MAX_TARGETS;
1949                 max_channels = STORVSC_IDE_MAX_CHANNELS;
1950         } else {
1951                 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
1952                 max_targets = STORVSC_MAX_TARGETS;
1953                 max_channels = STORVSC_MAX_CHANNELS;
1954                 /*
1955                  * On Windows8 and above, we support sub-channels for storage
1956                  * on SCSI and FC controllers.
1957                  * The number of sub-channels offerred is based on the number of
1958                  * VCPUs in the guest.
1959                  */
1960                 if (!dev_is_ide)
1961                         max_sub_channels =
1962                                 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1963         }
1964
1965         scsi_driver.can_queue = max_outstanding_req_per_channel *
1966                                 (max_sub_channels + 1) *
1967                                 (100 - ring_avail_percent_lowater) / 100;
1968
1969         host = scsi_host_alloc(&scsi_driver,
1970                                sizeof(struct hv_host_device));
1971         if (!host)
1972                 return -ENOMEM;
1973
1974         host_dev = shost_priv(host);
1975         memset(host_dev, 0, sizeof(struct hv_host_device));
1976
1977         host_dev->port = host->host_no;
1978         host_dev->dev = device;
1979         host_dev->host = host;
1980
1981
1982         stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1983         if (!stor_device) {
1984                 ret = -ENOMEM;
1985                 goto err_out0;
1986         }
1987
1988         stor_device->destroy = false;
1989         init_waitqueue_head(&stor_device->waiting_to_drain);
1990         stor_device->device = device;
1991         stor_device->host = host;
1992         stor_device->vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
1993         spin_lock_init(&stor_device->lock);
1994         hv_set_drvdata(device, stor_device);
1995
1996         stor_device->port_number = host->host_no;
1997         ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1998         if (ret)
1999                 goto err_out1;
2000
2001         host_dev->path = stor_device->path_id;
2002         host_dev->target = stor_device->target_id;
2003
2004         switch (dev_id->driver_data) {
2005         case SFC_GUID:
2006                 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
2007                 host->max_id = STORVSC_FC_MAX_TARGETS;
2008                 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
2009 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2010                 host->transportt = fc_transport_template;
2011 #endif
2012                 break;
2013
2014         case SCSI_GUID:
2015                 host->max_lun = max_luns_per_target;
2016                 host->max_id = max_targets;
2017                 host->max_channel = max_channels - 1;
2018                 break;
2019
2020         default:
2021                 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
2022                 host->max_id = STORVSC_IDE_MAX_TARGETS;
2023                 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2024                 break;
2025         }
2026         /* max cmd length */
2027         host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2028
2029         /*
2030          * set the table size based on the info we got
2031          * from the host.
2032          */
2033         host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
2034         /*
2035          * For non-IDE disks, the host supports multiple channels.
2036          * Set the number of HW queues we are supporting.
2037          */
2038         if (!dev_is_ide) {
2039                 if (storvsc_max_hw_queues > num_present_cpus) {
2040                         storvsc_max_hw_queues = 0;
2041                         storvsc_log(device, STORVSC_LOGGING_WARN,
2042                                 "Resetting invalid storvsc_max_hw_queues value to default.\n");
2043                 }
2044                 if (storvsc_max_hw_queues)
2045                         host->nr_hw_queues = storvsc_max_hw_queues;
2046                 else
2047                         host->nr_hw_queues = num_present_cpus;
2048         }
2049
2050         /*
2051          * Set the error handler work queue.
2052          */
2053         host_dev->handle_error_wq =
2054                         alloc_ordered_workqueue("storvsc_error_wq_%d",
2055                                                 WQ_MEM_RECLAIM,
2056                                                 host->host_no);
2057         if (!host_dev->handle_error_wq) {
2058                 ret = -ENOMEM;
2059                 goto err_out2;
2060         }
2061         INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2062         /* Register the HBA and start the scsi bus scan */
2063         ret = scsi_add_host(host, &device->device);
2064         if (ret != 0)
2065                 goto err_out3;
2066
2067         if (!dev_is_ide) {
2068                 scsi_scan_host(host);
2069         } else {
2070                 target = (device->dev_instance.b[5] << 8 |
2071                          device->dev_instance.b[4]);
2072                 ret = scsi_add_device(host, 0, target, 0);
2073                 if (ret)
2074                         goto err_out4;
2075         }
2076 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2077         if (host->transportt == fc_transport_template) {
2078                 struct fc_rport_identifiers ids = {
2079                         .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2080                 };
2081
2082                 fc_host_node_name(host) = stor_device->node_name;
2083                 fc_host_port_name(host) = stor_device->port_name;
2084                 stor_device->rport = fc_remote_port_add(host, 0, &ids);
2085                 if (!stor_device->rport) {
2086                         ret = -ENOMEM;
2087                         goto err_out4;
2088                 }
2089         }
2090 #endif
2091         return 0;
2092
2093 err_out4:
2094         scsi_remove_host(host);
2095
2096 err_out3:
2097         destroy_workqueue(host_dev->handle_error_wq);
2098
2099 err_out2:
2100         /*
2101          * Once we have connected with the host, we would need to
2102          * to invoke storvsc_dev_remove() to rollback this state and
2103          * this call also frees up the stor_device; hence the jump around
2104          * err_out1 label.
2105          */
2106         storvsc_dev_remove(device);
2107         goto err_out0;
2108
2109 err_out1:
2110         kfree(stor_device->stor_chns);
2111         kfree(stor_device);
2112
2113 err_out0:
2114         scsi_host_put(host);
2115         return ret;
2116 }
2117
2118 /* Change a scsi target's queue depth */
2119 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2120 {
2121         if (queue_depth > scsi_driver.can_queue)
2122                 queue_depth = scsi_driver.can_queue;
2123
2124         return scsi_change_queue_depth(sdev, queue_depth);
2125 }
2126
2127 static int storvsc_remove(struct hv_device *dev)
2128 {
2129         struct storvsc_device *stor_device = hv_get_drvdata(dev);
2130         struct Scsi_Host *host = stor_device->host;
2131         struct hv_host_device *host_dev = shost_priv(host);
2132
2133 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2134         if (host->transportt == fc_transport_template) {
2135                 fc_remote_port_delete(stor_device->rport);
2136                 fc_remove_host(host);
2137         }
2138 #endif
2139         destroy_workqueue(host_dev->handle_error_wq);
2140         scsi_remove_host(host);
2141         storvsc_dev_remove(dev);
2142         scsi_host_put(host);
2143
2144         return 0;
2145 }
2146
2147 static int storvsc_suspend(struct hv_device *hv_dev)
2148 {
2149         struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2150         struct Scsi_Host *host = stor_device->host;
2151         struct hv_host_device *host_dev = shost_priv(host);
2152
2153         storvsc_wait_to_drain(stor_device);
2154
2155         drain_workqueue(host_dev->handle_error_wq);
2156
2157         vmbus_close(hv_dev->channel);
2158
2159         kfree(stor_device->stor_chns);
2160         stor_device->stor_chns = NULL;
2161
2162         cpumask_clear(&stor_device->alloced_cpus);
2163
2164         return 0;
2165 }
2166
2167 static int storvsc_resume(struct hv_device *hv_dev)
2168 {
2169         int ret;
2170
2171         ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2172                                      hv_dev_is_fc(hv_dev));
2173         return ret;
2174 }
2175
2176 static struct hv_driver storvsc_drv = {
2177         .name = KBUILD_MODNAME,
2178         .id_table = id_table,
2179         .probe = storvsc_probe,
2180         .remove = storvsc_remove,
2181         .suspend = storvsc_suspend,
2182         .resume = storvsc_resume,
2183         .driver = {
2184                 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2185         },
2186 };
2187
2188 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2189 static struct fc_function_template fc_transport_functions = {
2190         .show_host_node_name = 1,
2191         .show_host_port_name = 1,
2192 };
2193 #endif
2194
2195 static int __init storvsc_drv_init(void)
2196 {
2197         int ret;
2198
2199         /*
2200          * Divide the ring buffer data size (which is 1 page less
2201          * than the ring buffer size since that page is reserved for
2202          * the ring buffer indices) by the max request size (which is
2203          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2204          *
2205          * The computation underestimates max_outstanding_req_per_channel
2206          * for Win7 and older hosts because it does not take into account
2207          * the vmscsi_size_delta correction to the max request size.
2208          */
2209         max_outstanding_req_per_channel =
2210                 ((storvsc_ringbuffer_size - PAGE_SIZE) /
2211                 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2212                 sizeof(struct vstor_packet) + sizeof(u64),
2213                 sizeof(u64)));
2214
2215 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2216         fc_transport_template = fc_attach_transport(&fc_transport_functions);
2217         if (!fc_transport_template)
2218                 return -ENODEV;
2219 #endif
2220
2221         ret = vmbus_driver_register(&storvsc_drv);
2222
2223 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2224         if (ret)
2225                 fc_release_transport(fc_transport_template);
2226 #endif
2227
2228         return ret;
2229 }
2230
2231 static void __exit storvsc_drv_exit(void)
2232 {
2233         vmbus_driver_unregister(&storvsc_drv);
2234 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2235         fc_release_transport(fc_transport_template);
2236 #endif
2237 }
2238
2239 MODULE_LICENSE("GPL");
2240 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2241 module_init(storvsc_drv_init);
2242 module_exit(storvsc_drv_exit);