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