drivers/staging/hv/storvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  *   K. Y. Srinivasan <kys@microsoft.com>
  21  */
  22
  23 #include <linux/kernel.h>
  24 #include <linux/wait.h>
  25 #include <linux/sched.h>
  26 #include <linux/completion.h>
  27 #include <linux/string.h>
  28 #include <linux/mm.h>
  29 #include <linux/delay.h>
  30 #include <linux/init.h>
  31 #include <linux/slab.h>
  32 #include <linux/module.h>
  33 #include <linux/device.h>
  34 #include <linux/hyperv.h>
  35 #include <scsi/scsi.h>
  36 #include <scsi/scsi_cmnd.h>
  37 #include <scsi/scsi_host.h>
  38 #include <scsi/scsi_device.h>
  39 #include <scsi/scsi_tcq.h>
  40 #include <scsi/scsi_eh.h>
  41 #include <scsi/scsi_devinfo.h>
  42 #include <scsi/scsi_dbg.h>
  43
  44
  45 #define STORVSC_RING_BUFFER_SIZE                        (20*PAGE_SIZE)
  46 static int storvsc_ringbuffer_size = STORVSC_RING_BUFFER_SIZE;
  47
  48 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
  49 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
  50
  51 /* to alert the user that structure sizes may be mismatched even though the */
  52 /* protocol versions match. */
  53
  54
  55 #define REVISION_STRING(REVISION_) #REVISION_
  56 #define FILL_VMSTOR_REVISION(RESULT_LVALUE_)                            \
  57         do {                                                            \
  58                 char *revision_string                                   \
  59                         = REVISION_STRING($Rev : 6 $) + 6;              \
  60                 RESULT_LVALUE_ = 0;                                     \
  61                 while (*revision_string >= '0'                          \
  62                         && *revision_string <= '9') {                   \
  63                         RESULT_LVALUE_ *= 10;                           \
  64                         RESULT_LVALUE_ += *revision_string - '0';       \
  65                         revision_string++;                              \
  66                 }                                                       \
  67         } while (0)
  68
  69 /* Major/minor macros.  Minor version is in LSB, meaning that earlier flat */
  70 /* version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1). */
  71 #define VMSTOR_PROTOCOL_MAJOR(VERSION_)         (((VERSION_) >> 8) & 0xff)
  72 #define VMSTOR_PROTOCOL_MINOR(VERSION_)         (((VERSION_))      & 0xff)
  73 #define VMSTOR_PROTOCOL_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \
  74                                                  (((MINOR_) & 0xff)))
  75 #define VMSTOR_INVALID_PROTOCOL_VERSION         (-1)
  76
  77 /* Version history: */
  78 /* V1 Beta                    0.1 */
  79 /* V1 RC < 2008/1/31          1.0 */
  80 /* V1 RC > 2008/1/31          2.0 */
  81 #define VMSTOR_PROTOCOL_VERSION_CURRENT VMSTOR_PROTOCOL_VERSION(2, 0)
  82
  83
  84
  85
  86 /*  This will get replaced with the max transfer length that is possible on */
  87 /*  the host adapter. */
  88 /*  The max transfer length will be published when we offer a vmbus channel. */
  89 #define MAX_TRANSFER_LENGTH     0x40000
  90 #define DEFAULT_PACKET_SIZE (sizeof(struct vmdata_gpa_direct) + \
  91                         sizeof(struct vstor_packet) +           \
  92                         sizesizeof(u64) * (MAX_TRANSFER_LENGTH / PAGE_SIZE)))
  93
  94
  95 /*  Packet structure describing virtual storage requests. */
  96 enum vstor_packet_operation {
  97         VSTOR_OPERATION_COMPLETE_IO             = 1,
  98         VSTOR_OPERATION_REMOVE_DEVICE           = 2,
  99         VSTOR_OPERATION_EXECUTE_SRB             = 3,
 100         VSTOR_OPERATION_RESET_LUN               = 4,
 101         VSTOR_OPERATION_RESET_ADAPTER           = 5,
 102         VSTOR_OPERATION_RESET_BUS               = 6,
 103         VSTOR_OPERATION_BEGIN_INITIALIZATION    = 7,
 104         VSTOR_OPERATION_END_INITIALIZATION      = 8,
 105         VSTOR_OPERATION_QUERY_PROTOCOL_VERSION  = 9,
 106         VSTOR_OPERATION_QUERY_PROPERTIES        = 10,
 107         VSTOR_OPERATION_MAXIMUM                 = 10
 108 };
 109
 110 /*
 111  * Platform neutral description of a scsi request -
 112  * this remains the same across the write regardless of 32/64 bit
 113  * note: it's patterned off the SCSI_PASS_THROUGH structure
 114  */
 115 #define CDB16GENERIC_LENGTH                     0x10
 116
 117 #ifndef SENSE_BUFFER_SIZE
 118 #define SENSE_BUFFER_SIZE                       0x12
 119 #endif
 120
 121 #define MAX_DATA_BUF_LEN_WITH_PADDING           0x14
 122
 123 struct vmscsi_request {
 124         unsigned short length;
 125         unsigned char srb_status;
 126         unsigned char scsi_status;
 127
 128         unsigned char port_number;
 129         unsigned char path_id;
 130         unsigned char target_id;
 131         unsigned char lun;
 132
 133         unsigned char cdb_length;
 134         unsigned char sense_info_length;
 135         unsigned char data_in;
 136         unsigned char reserved;
 137
 138         unsigned int data_transfer_length;
 139
 140         union {
 141                 unsigned char cdb[CDB16GENERIC_LENGTH];
 142                 unsigned char sense_data[SENSE_BUFFER_SIZE];
 143                 unsigned char reserved_array[MAX_DATA_BUF_LEN_WITH_PADDING];
 144         };
 145 } __attribute((packed));
 146
 147
 148 /*
 149  * This structure is sent during the intialization phase to get the different
 150  * properties of the channel.
 151  */
 152 struct vmstorage_channel_properties {
 153         unsigned short protocol_version;
 154         unsigned char path_id;
 155         unsigned char target_id;
 156
 157         /* Note: port number is only really known on the client side */
 158         unsigned int port_number;
 159         unsigned int flags;
 160         unsigned int max_transfer_bytes;
 161
 162         /*  This id is unique for each channel and will correspond with */
 163         /*  vendor specific data in the inquirydata */
 164         unsigned long long unique_id;
 165 } __packed;
 166
 167 /*  This structure is sent during the storage protocol negotiations. */
 168 struct vmstorage_protocol_version {
 169         /* Major (MSW) and minor (LSW) version numbers. */
 170         unsigned short major_minor;
 171
 172         /*
 173          * Revision number is auto-incremented whenever this file is changed
 174          * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
 175          * definitely indicate incompatibility--but it does indicate mismatched
 176          * builds.
 177          */
 178         unsigned short revision;
 179 } __packed;
 180
 181 /* Channel Property Flags */
 182 #define STORAGE_CHANNEL_REMOVABLE_FLAG          0x1
 183 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG       0x2
 184
 185 struct vstor_packet {
 186         /* Requested operation type */
 187         enum vstor_packet_operation operation;
 188
 189         /*  Flags - see below for values */
 190         unsigned int flags;
 191
 192         /* Status of the request returned from the server side. */
 193         unsigned int status;
 194
 195         /* Data payload area */
 196         union {
 197                 /*
 198                  * Structure used to forward SCSI commands from the
 199                  * client to the server.
 200                  */
 201                 struct vmscsi_request vm_srb;
 202
 203                 /* Structure used to query channel properties. */
 204                 struct vmstorage_channel_properties storage_channel_properties;
 205
 206                 /* Used during version negotiations. */
 207                 struct vmstorage_protocol_version version;
 208         };
 209 } __packed;
 210
 211 /* Packet flags */
 212 /*
 213  * This flag indicates that the server should send back a completion for this
 214  * packet.
 215  */
 216 #define REQUEST_COMPLETION_FLAG 0x1
 217
 218 /*  This is the set of flags that the vsc can set in any packets it sends */
 219 #define VSC_LEGAL_FLAGS         (REQUEST_COMPLETION_FLAG)
 220
 221
 222 /* Defines */
 223
 224 #define STORVSC_MAX_IO_REQUESTS                         128
 225
 226 /*
 227  * In Hyper-V, each port/path/target maps to 1 scsi host adapter.  In
 228  * reality, the path/target is not used (ie always set to 0) so our
 229  * scsi host adapter essentially has 1 bus with 1 target that contains
 230  * up to 256 luns.
 231  */
 232 #define STORVSC_MAX_LUNS_PER_TARGET                     64
 233 #define STORVSC_MAX_TARGETS                             1
 234 #define STORVSC_MAX_CHANNELS                            1
 235 #define STORVSC_MAX_CMD_LEN                             16
 236
 237 struct hv_storvsc_request;
 238
 239 /* Matches Windows-end */
 240 enum storvsc_request_type {
 241         WRITE_TYPE,
 242         READ_TYPE,
 243         UNKNOWN_TYPE,
 244 };
 245
 246
 247 struct hv_storvsc_request {
 248         struct hv_device *device;
 249
 250         /* Synchronize the request/response if needed */
 251         struct completion wait_event;
 252
 253         unsigned char *sense_buffer;
 254         void *context;
 255         void (*on_io_completion)(struct hv_storvsc_request *request);
 256         struct hv_multipage_buffer data_buffer;
 257
 258         struct vstor_packet vstor_packet;
 259 };
 260
 261
 262 /* A storvsc device is a device object that contains a vmbus channel */
 263 struct storvsc_device {
 264         struct hv_device *device;
 265
 266         bool     destroy;
 267         bool     drain_notify;
 268         atomic_t num_outstanding_req;
 269         struct Scsi_Host *host;
 270
 271         wait_queue_head_t waiting_to_drain;
 272
 273         /*
 274          * Each unique Port/Path/Target represents 1 channel ie scsi
 275          * controller. In reality, the pathid, targetid is always 0
 276          * and the port is set by us
 277          */
 278         unsigned int port_number;
 279         unsigned char path_id;
 280         unsigned char target_id;
 281
 282         /* Used for vsc/vsp channel reset process */
 283         struct hv_storvsc_request init_request;
 284         struct hv_storvsc_request reset_request;
 285 };
 286
 287 struct hv_host_device {
 288         struct hv_device *dev;
 289         struct kmem_cache *request_pool;
 290         unsigned int port;
 291         unsigned char path;
 292         unsigned char target;
 293 };
 294
 295 struct storvsc_cmd_request {
 296         struct list_head entry;
 297         struct scsi_cmnd *cmd;
 298
 299         unsigned int bounce_sgl_count;
 300         struct scatterlist *bounce_sgl;
 301
 302         struct hv_storvsc_request request;
 303 };
 304
 305 static inline struct storvsc_device *get_out_stor_device(
 306                                         struct hv_device *device)
 307 {
 308         struct storvsc_device *stor_device;
 309
 310         stor_device = hv_get_drvdata(device);
 311
 312         if (stor_device && stor_device->destroy)
 313                 stor_device = NULL;
 314
 315         return stor_device;
 316 }
 317
 318
 319 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
 320 {
 321         dev->drain_notify = true;
 322         wait_event(dev->waiting_to_drain,
 323                    atomic_read(&dev->num_outstanding_req) == 0);
 324         dev->drain_notify = false;
 325 }
 326
 327 static inline struct storvsc_device *get_in_stor_device(
 328                                         struct hv_device *device)
 329 {
 330         struct storvsc_device *stor_device;
 331
 332         stor_device = hv_get_drvdata(device);
 333
 334         if (!stor_device)
 335                 goto get_in_err;
 336
 337         /*
 338          * If the device is being destroyed; allow incoming
 339          * traffic only to cleanup outstanding requests.
 340          */
 341
 342         if (stor_device->destroy  &&
 343                 (atomic_read(&stor_device->num_outstanding_req) == 0))
 344                 stor_device = NULL;
 345
 346 get_in_err:
 347         return stor_device;
 348
 349 }
 350
 351 static int storvsc_channel_init(struct hv_device *device)
 352 {
 353         struct storvsc_device *stor_device;
 354         struct hv_storvsc_request *request;
 355         struct vstor_packet *vstor_packet;
 356         int ret, t;
 357
 358         stor_device = get_out_stor_device(device);
 359         if (!stor_device)
 360                 return -ENODEV;
 361
 362         request = &stor_device->init_request;
 363         vstor_packet = &request->vstor_packet;
 364
 365         /*
 366          * Now, initiate the vsc/vsp initialization protocol on the open
 367          * channel
 368          */
 369         memset(request, 0, sizeof(struct hv_storvsc_request));
 370         init_completion(&request->wait_event);
 371         vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
 372         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 373
 374         ret = vmbus_sendpacket(device->channel, vstor_packet,
 375                                sizeof(struct vstor_packet),
 376                                (unsigned long)request,
 377                                VM_PKT_DATA_INBAND,
 378                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 379         if (ret != 0)
 380                 goto cleanup;
 381
 382         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 383         if (t == 0) {
 384                 ret = -ETIMEDOUT;
 385                 goto cleanup;
 386         }
 387
 388         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 389             vstor_packet->status != 0)
 390                 goto cleanup;
 391
 392
 393         /* reuse the packet for version range supported */
 394         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 395         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
 396         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 397
 398         vstor_packet->version.major_minor = VMSTOR_PROTOCOL_VERSION_CURRENT;
 399         FILL_VMSTOR_REVISION(vstor_packet->version.revision);
 400
 401         ret = vmbus_sendpacket(device->channel, vstor_packet,
 402                                sizeof(struct vstor_packet),
 403                                (unsigned long)request,
 404                                VM_PKT_DATA_INBAND,
 405                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 406         if (ret != 0)
 407                 goto cleanup;
 408
 409         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 410         if (t == 0) {
 411                 ret = -ETIMEDOUT;
 412                 goto cleanup;
 413         }
 414
 415         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 416             vstor_packet->status != 0)
 417                 goto cleanup;
 418
 419
 420         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 421         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
 422         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 423         vstor_packet->storage_channel_properties.port_number =
 424                                         stor_device->port_number;
 425
 426         ret = vmbus_sendpacket(device->channel, vstor_packet,
 427                                sizeof(struct vstor_packet),
 428                                (unsigned long)request,
 429                                VM_PKT_DATA_INBAND,
 430                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 431
 432         if (ret != 0)
 433                 goto cleanup;
 434
 435         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 436         if (t == 0) {
 437                 ret = -ETIMEDOUT;
 438                 goto cleanup;
 439         }
 440
 441         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 442             vstor_packet->status != 0)
 443                 goto cleanup;
 444
 445         stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
 446         stor_device->target_id
 447                 = vstor_packet->storage_channel_properties.target_id;
 448
 449         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 450         vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
 451         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 452
 453         ret = vmbus_sendpacket(device->channel, vstor_packet,
 454                                sizeof(struct vstor_packet),
 455                                (unsigned long)request,
 456                                VM_PKT_DATA_INBAND,
 457                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 458
 459         if (ret != 0)
 460                 goto cleanup;
 461
 462         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 463         if (t == 0) {
 464                 ret = -ETIMEDOUT;
 465                 goto cleanup;
 466         }
 467
 468         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 469             vstor_packet->status != 0)
 470                 goto cleanup;
 471
 472
 473 cleanup:
 474         return ret;
 475 }
 476
 477 static void storvsc_on_io_completion(struct hv_device *device,
 478                                   struct vstor_packet *vstor_packet,
 479                                   struct hv_storvsc_request *request)
 480 {
 481         struct storvsc_device *stor_device;
 482         struct vstor_packet *stor_pkt;
 483
 484         stor_device = hv_get_drvdata(device);
 485         stor_pkt = &request->vstor_packet;
 486
 487         /*
 488          * The current SCSI handling on the host side does
 489          * not correctly handle:
 490          * INQUIRY command with page code parameter set to 0x80
 491          * MODE_SENSE command with cmd[2] == 0x1c
 492          *
 493          * Setup srb and scsi status so this won't be fatal.
 494          * We do this so we can distinguish truly fatal failues
 495          * (srb status == 0x4) and off-line the device in that case.
 496          */
 497
 498         if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
 499                 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
 500                 vstor_packet->vm_srb.scsi_status = 0;
 501                 vstor_packet->vm_srb.srb_status = 0x1;
 502         }
 503
 504
 505         /* Copy over the status...etc */
 506         stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
 507         stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
 508         stor_pkt->vm_srb.sense_info_length =
 509         vstor_packet->vm_srb.sense_info_length;
 510
 511         if (vstor_packet->vm_srb.scsi_status != 0 ||
 512                 vstor_packet->vm_srb.srb_status != 1){
 513                 dev_warn(&device->device,
 514                          "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
 515                          stor_pkt->vm_srb.cdb[0],
 516                          vstor_packet->vm_srb.scsi_status,
 517                          vstor_packet->vm_srb.srb_status);
 518         }
 519
 520         if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
 521                 /* CHECK_CONDITION */
 522                 if (vstor_packet->vm_srb.srb_status & 0x80) {
 523                         /* autosense data available */
 524                         dev_warn(&device->device,
 525                                  "stor pkt %p autosense data valid - len %d\n",
 526                                  request,
 527                                  vstor_packet->vm_srb.sense_info_length);
 528
 529                         memcpy(request->sense_buffer,
 530                                vstor_packet->vm_srb.sense_data,
 531                                vstor_packet->vm_srb.sense_info_length);
 532
 533                 }
 534         }
 535
 536         stor_pkt->vm_srb.data_transfer_length =
 537         vstor_packet->vm_srb.data_transfer_length;
 538
 539         request->on_io_completion(request);
 540
 541         if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
 542                 stor_device->drain_notify)
 543                 wake_up(&stor_device->waiting_to_drain);
 544
 545
 546 }
 547
 548 static void storvsc_on_receive(struct hv_device *device,
 549                              struct vstor_packet *vstor_packet,
 550                              struct hv_storvsc_request *request)
 551 {
 552         switch (vstor_packet->operation) {
 553         case VSTOR_OPERATION_COMPLETE_IO:
 554                 storvsc_on_io_completion(device, vstor_packet, request);
 555                 break;
 556         case VSTOR_OPERATION_REMOVE_DEVICE:
 557
 558         default:
 559                 break;
 560         }
 561 }
 562
 563 static void storvsc_on_channel_callback(void *context)
 564 {
 565         struct hv_device *device = (struct hv_device *)context;
 566         struct storvsc_device *stor_device;
 567         u32 bytes_recvd;
 568         u64 request_id;
 569         unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
 570         struct hv_storvsc_request *request;
 571         int ret;
 572
 573
 574         stor_device = get_in_stor_device(device);
 575         if (!stor_device)
 576                 return;
 577
 578         do {
 579                 ret = vmbus_recvpacket(device->channel, packet,
 580                                        ALIGN(sizeof(struct vstor_packet), 8),
 581                                        &bytes_recvd, &request_id);
 582                 if (ret == 0 && bytes_recvd > 0) {
 583
 584                         request = (struct hv_storvsc_request *)
 585                                         (unsigned long)request_id;
 586
 587                         if ((request == &stor_device->init_request) ||
 588                             (request == &stor_device->reset_request)) {
 589
 590                                 memcpy(&request->vstor_packet, packet,
 591                                        sizeof(struct vstor_packet));
 592                                 complete(&request->wait_event);
 593                         } else {
 594                                 storvsc_on_receive(device,
 595                                                 (struct vstor_packet *)packet,
 596                                                 request);
 597                         }
 598                 } else {
 599                         break;
 600                 }
 601         } while (1);
 602
 603         return;
 604 }
 605
 606 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
 607 {
 608         struct vmstorage_channel_properties props;
 609         int ret;
 610
 611         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
 612
 613         /* Open the channel */
 614         ret = vmbus_open(device->channel,
 615                          ring_size,
 616                          ring_size,
 617                          (void *)&props,
 618                          sizeof(struct vmstorage_channel_properties),
 619                          storvsc_on_channel_callback, device);
 620
 621         if (ret != 0)
 622                 return ret;
 623
 624         ret = storvsc_channel_init(device);
 625
 626         return ret;
 627 }
 628
 629 static int storvsc_dev_remove(struct hv_device *device)
 630 {
 631         struct storvsc_device *stor_device;
 632         unsigned long flags;
 633
 634         stor_device = hv_get_drvdata(device);
 635
 636         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 637         stor_device->destroy = true;
 638         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 639
 640         /*
 641          * At this point, all outbound traffic should be disable. We
 642          * only allow inbound traffic (responses) to proceed so that
 643          * outstanding requests can be completed.
 644          */
 645
 646         storvsc_wait_to_drain(stor_device);
 647
 648         /*
 649          * Since we have already drained, we don't need to busy wait
 650          * as was done in final_release_stor_device()
 651          * Note that we cannot set the ext pointer to NULL until
 652          * we have drained - to drain the outgoing packets, we need to
 653          * allow incoming packets.
 654          */
 655         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 656         hv_set_drvdata(device, NULL);
 657         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 658
 659         /* Close the channel */
 660         vmbus_close(device->channel);
 661
 662         kfree(stor_device);
 663         return 0;
 664 }
 665
 666 static int storvsc_do_io(struct hv_device *device,
 667                               struct hv_storvsc_request *request)
 668 {
 669         struct storvsc_device *stor_device;
 670         struct vstor_packet *vstor_packet;
 671         int ret = 0;
 672
 673         vstor_packet = &request->vstor_packet;
 674         stor_device = get_out_stor_device(device);
 675
 676         if (!stor_device)
 677                 return -ENODEV;
 678
 679
 680         request->device  = device;
 681
 682
 683         vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
 684
 685         vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
 686
 687
 688         vstor_packet->vm_srb.sense_info_length = SENSE_BUFFER_SIZE;
 689
 690
 691         vstor_packet->vm_srb.data_transfer_length =
 692         request->data_buffer.len;
 693
 694         vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
 695
 696         if (request->data_buffer.len) {
 697                 ret = vmbus_sendpacket_multipagebuffer(device->channel,
 698                                 &request->data_buffer,
 699                                 vstor_packet,
 700                                 sizeof(struct vstor_packet),
 701                                 (unsigned long)request);
 702         } else {
 703                 ret = vmbus_sendpacket(device->channel, vstor_packet,
 704                                sizeof(struct vstor_packet),
 705                                (unsigned long)request,
 706                                VM_PKT_DATA_INBAND,
 707                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 708         }
 709
 710         if (ret != 0)
 711                 return ret;
 712
 713         atomic_inc(&stor_device->num_outstanding_req);
 714
 715         return ret;
 716 }
 717
 718 static void storvsc_get_ide_info(struct hv_device *dev, int *target, int *path)
 719 {
 720         *target =
 721                 dev->dev_instance.b[5] << 8 | dev->dev_instance.b[4];
 722
 723         *path =
 724                 dev->dev_instance.b[3] << 24 |
 725                 dev->dev_instance.b[2] << 16 |
 726                 dev->dev_instance.b[1] << 8  | dev->dev_instance.b[0];
 727 }
 728
 729
 730 static int storvsc_device_alloc(struct scsi_device *sdevice)
 731 {
 732         /*
 733          * This enables luns to be located sparsely. Otherwise, we may not
 734          * discovered them.
 735          */
 736         sdevice->sdev_bflags |= BLIST_SPARSELUN | BLIST_LARGELUN;
 737         return 0;
 738 }
 739
 740 static int storvsc_merge_bvec(struct request_queue *q,
 741                               struct bvec_merge_data *bmd, struct bio_vec *bvec)
 742 {
 743         /* checking done by caller. */
 744         return bvec->bv_len;
 745 }
 746
 747 static int storvsc_device_configure(struct scsi_device *sdevice)
 748 {
 749         scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
 750                                 STORVSC_MAX_IO_REQUESTS);
 751
 752         blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
 753
 754         blk_queue_merge_bvec(sdevice->request_queue, storvsc_merge_bvec);
 755
 756         blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
 757
 758         return 0;
 759 }
 760
 761 static void destroy_bounce_buffer(struct scatterlist *sgl,
 762                                   unsigned int sg_count)
 763 {
 764         int i;
 765         struct page *page_buf;
 766
 767         for (i = 0; i < sg_count; i++) {
 768                 page_buf = sg_page((&sgl[i]));
 769                 if (page_buf != NULL)
 770                         __free_page(page_buf);
 771         }
 772
 773         kfree(sgl);
 774 }
 775
 776 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
 777 {
 778         int i;
 779
 780         /* No need to check */
 781         if (sg_count < 2)
 782                 return -1;
 783
 784         /* We have at least 2 sg entries */
 785         for (i = 0; i < sg_count; i++) {
 786                 if (i == 0) {
 787                         /* make sure 1st one does not have hole */
 788                         if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
 789                                 return i;
 790                 } else if (i == sg_count - 1) {
 791                         /* make sure last one does not have hole */
 792                         if (sgl[i].offset != 0)
 793                                 return i;
 794                 } else {
 795                         /* make sure no hole in the middle */
 796                         if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
 797                                 return i;
 798                 }
 799         }
 800         return -1;
 801 }
 802
 803 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
 804                                                 unsigned int sg_count,
 805                                                 unsigned int len)
 806 {
 807         int i;
 808         int num_pages;
 809         struct scatterlist *bounce_sgl;
 810         struct page *page_buf;
 811
 812         num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
 813
 814         bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
 815         if (!bounce_sgl)
 816                 return NULL;
 817
 818         for (i = 0; i < num_pages; i++) {
 819                 page_buf = alloc_page(GFP_ATOMIC);
 820                 if (!page_buf)
 821                         goto cleanup;
 822                 sg_set_page(&bounce_sgl[i], page_buf, 0, 0);
 823         }
 824
 825         return bounce_sgl;
 826
 827 cleanup:
 828         destroy_bounce_buffer(bounce_sgl, num_pages);
 829         return NULL;
 830 }
 831
 832
 833 /* Assume the original sgl has enough room */
 834 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
 835                                             struct scatterlist *bounce_sgl,
 836                                             unsigned int orig_sgl_count)
 837 {
 838         int i;
 839         int j = 0;
 840         unsigned long src, dest;
 841         unsigned int srclen, destlen, copylen;
 842         unsigned int total_copied = 0;
 843         unsigned long bounce_addr = 0;
 844         unsigned long dest_addr = 0;
 845         unsigned long flags;
 846
 847         local_irq_save(flags);
 848
 849         for (i = 0; i < orig_sgl_count; i++) {
 850                 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
 851                                         KM_IRQ0) + orig_sgl[i].offset;
 852                 dest = dest_addr;
 853                 destlen = orig_sgl[i].length;
 854
 855                 if (bounce_addr == 0)
 856                         bounce_addr =
 857                         (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
 858                                                         KM_IRQ0);
 859
 860                 while (destlen) {
 861                         src = bounce_addr + bounce_sgl[j].offset;
 862                         srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
 863
 864                         copylen = min(srclen, destlen);
 865                         memcpy((void *)dest, (void *)src, copylen);
 866
 867                         total_copied += copylen;
 868                         bounce_sgl[j].offset += copylen;
 869                         destlen -= copylen;
 870                         dest += copylen;
 871
 872                         if (bounce_sgl[j].offset == bounce_sgl[j].length) {
 873                                 /* full */
 874                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
 875                                 j++;
 876
 877                                 /* if we need to use another bounce buffer */
 878                                 if (destlen || i != orig_sgl_count - 1)
 879                                         bounce_addr =
 880                                         (unsigned long)kmap_atomic(
 881                                         sg_page((&bounce_sgl[j])), KM_IRQ0);
 882                         } else if (destlen == 0 && i == orig_sgl_count - 1) {
 883                                 /* unmap the last bounce that is < PAGE_SIZE */
 884                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
 885                         }
 886                 }
 887
 888                 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
 889                               KM_IRQ0);
 890         }
 891
 892         local_irq_restore(flags);
 893
 894         return total_copied;
 895 }
 896
 897
 898 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
 899 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
 900                                           struct scatterlist *bounce_sgl,
 901                                           unsigned int orig_sgl_count)
 902 {
 903         int i;
 904         int j = 0;
 905         unsigned long src, dest;
 906         unsigned int srclen, destlen, copylen;
 907         unsigned int total_copied = 0;
 908         unsigned long bounce_addr = 0;
 909         unsigned long src_addr = 0;
 910         unsigned long flags;
 911
 912         local_irq_save(flags);
 913
 914         for (i = 0; i < orig_sgl_count; i++) {
 915                 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
 916                                 KM_IRQ0) + orig_sgl[i].offset;
 917                 src = src_addr;
 918                 srclen = orig_sgl[i].length;
 919
 920                 if (bounce_addr == 0)
 921                         bounce_addr =
 922                         (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
 923                                                 KM_IRQ0);
 924
 925                 while (srclen) {
 926                         /* assume bounce offset always == 0 */
 927                         dest = bounce_addr + bounce_sgl[j].length;
 928                         destlen = PAGE_SIZE - bounce_sgl[j].length;
 929
 930                         copylen = min(srclen, destlen);
 931                         memcpy((void *)dest, (void *)src, copylen);
 932
 933                         total_copied += copylen;
 934                         bounce_sgl[j].length += copylen;
 935                         srclen -= copylen;
 936                         src += copylen;
 937
 938                         if (bounce_sgl[j].length == PAGE_SIZE) {
 939                                 /* full..move to next entry */
 940                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
 941                                 j++;
 942
 943                                 /* if we need to use another bounce buffer */
 944                                 if (srclen || i != orig_sgl_count - 1)
 945                                         bounce_addr =
 946                                         (unsigned long)kmap_atomic(
 947                                         sg_page((&bounce_sgl[j])), KM_IRQ0);
 948
 949                         } else if (srclen == 0 && i == orig_sgl_count - 1) {
 950                                 /* unmap the last bounce that is < PAGE_SIZE */
 951                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
 952                         }
 953                 }
 954
 955                 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
 956         }
 957
 958         local_irq_restore(flags);
 959
 960         return total_copied;
 961 }
 962
 963
 964 static int storvsc_remove(struct hv_device *dev)
 965 {
 966         struct storvsc_device *stor_device = hv_get_drvdata(dev);
 967         struct Scsi_Host *host = stor_device->host;
 968         struct hv_host_device *host_dev =
 969                         (struct hv_host_device *)host->hostdata;
 970
 971         scsi_remove_host(host);
 972
 973         scsi_host_put(host);
 974
 975         storvsc_dev_remove(dev);
 976         if (host_dev->request_pool) {
 977                 kmem_cache_destroy(host_dev->request_pool);
 978                 host_dev->request_pool = NULL;
 979         }
 980         return 0;
 981 }
 982
 983
 984 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
 985                            sector_t capacity, int *info)
 986 {
 987         sector_t nsect = capacity;
 988         sector_t cylinders = nsect;
 989         int heads, sectors_pt;
 990
 991         /*
 992          * We are making up these values; let us keep it simple.
 993          */
 994         heads = 0xff;
 995         sectors_pt = 0x3f;      /* Sectors per track */
 996         sector_div(cylinders, heads * sectors_pt);
 997         if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
 998                 cylinders = 0xffff;
 999
1000         info[0] = heads;
1001         info[1] = sectors_pt;
1002         info[2] = (int)cylinders;
1003
1004         return 0;
1005 }
1006
1007 static int storvsc_host_reset(struct hv_device *device)
1008 {
1009         struct storvsc_device *stor_device;
1010         struct hv_storvsc_request *request;
1011         struct vstor_packet *vstor_packet;
1012         int ret, t;
1013
1014
1015         stor_device = get_out_stor_device(device);
1016         if (!stor_device)
1017                 return -ENODEV;
1018
1019         request = &stor_device->reset_request;
1020         vstor_packet = &request->vstor_packet;
1021
1022         init_completion(&request->wait_event);
1023
1024         vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1025         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1026         vstor_packet->vm_srb.path_id = stor_device->path_id;
1027
1028         ret = vmbus_sendpacket(device->channel, vstor_packet,
1029                                sizeof(struct vstor_packet),
1030                                (unsigned long)&stor_device->reset_request,
1031                                VM_PKT_DATA_INBAND,
1032                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1033         if (ret != 0)
1034                 goto cleanup;
1035
1036         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1037         if (t == 0) {
1038                 ret = -ETIMEDOUT;
1039                 goto cleanup;
1040         }
1041
1042
1043         /*
1044          * At this point, all outstanding requests in the adapter
1045          * should have been flushed out and return to us
1046          * There is a potential race here where the host may be in
1047          * the process of responding when we return from here.
1048          * Just wait for all in-transit packets to be accounted for
1049          * before we return from here.
1050          */
1051         storvsc_wait_to_drain(stor_device);
1052
1053 cleanup:
1054         return ret;
1055 }
1056
1057
1058 /*
1059  * storvsc_host_reset_handler - Reset the scsi HBA
1060  */
1061 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1062 {
1063         int ret;
1064         struct hv_host_device *host_dev =
1065                 (struct hv_host_device *)scmnd->device->host->hostdata;
1066         struct hv_device *dev = host_dev->dev;
1067
1068         ret = storvsc_host_reset(dev);
1069         if (ret != 0)
1070                 return ret;
1071
1072         return ret;
1073 }
1074
1075
1076 /*
1077  * storvsc_command_completion - Command completion processing
1078  */
1079 static void storvsc_command_completion(struct hv_storvsc_request *request)
1080 {
1081         struct storvsc_cmd_request *cmd_request =
1082                 (struct storvsc_cmd_request *)request->context;
1083         struct scsi_cmnd *scmnd = cmd_request->cmd;
1084         struct hv_host_device *host_dev =
1085                 (struct hv_host_device *)scmnd->device->host->hostdata;
1086         void (*scsi_done_fn)(struct scsi_cmnd *);
1087         struct scsi_sense_hdr sense_hdr;
1088         struct vmscsi_request *vm_srb;
1089
1090         vm_srb = &request->vstor_packet.vm_srb;
1091         if (cmd_request->bounce_sgl_count) {
1092                 if (vm_srb->data_in == READ_TYPE) {
1093                         copy_from_bounce_buffer(scsi_sglist(scmnd),
1094                                         cmd_request->bounce_sgl,
1095                                         scsi_sg_count(scmnd));
1096                         destroy_bounce_buffer(cmd_request->bounce_sgl,
1097                                         cmd_request->bounce_sgl_count);
1098                 }
1099         }
1100
1101         /*
1102          * If there is an error; offline the device since all
1103          * error recovery strategies would have already been
1104          * deployed on the host side.
1105          */
1106         if (vm_srb->srb_status == 0x4)
1107                 scmnd->result = DID_TARGET_FAILURE << 16;
1108         else
1109                 scmnd->result = vm_srb->scsi_status;
1110
1111         if (scmnd->result) {
1112                 if (scsi_normalize_sense(scmnd->sense_buffer,
1113                                 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1114                         scsi_print_sense_hdr("storvsc", &sense_hdr);
1115         }
1116
1117         scsi_set_resid(scmnd,
1118                 request->data_buffer.len -
1119                 vm_srb->data_transfer_length);
1120
1121         scsi_done_fn = scmnd->scsi_done;
1122
1123         scmnd->host_scribble = NULL;
1124         scmnd->scsi_done = NULL;
1125
1126         scsi_done_fn(scmnd);
1127
1128         kmem_cache_free(host_dev->request_pool, cmd_request);
1129 }
1130
1131 static bool storvsc_check_scsi_cmd(struct scsi_cmnd *scmnd)
1132 {
1133         bool allowed = true;
1134         u8 scsi_op = scmnd->cmnd[0];
1135
1136         switch (scsi_op) {
1137         /* smartd sends this command, which will offline the device */
1138         case SET_WINDOW:
1139                 scmnd->result = DID_ERROR << 16;
1140                 allowed = false;
1141                 break;
1142         default:
1143                 break;
1144         }
1145         return allowed;
1146 }
1147
1148 /*
1149  * storvsc_queuecommand - Initiate command processing
1150  */
1151 static int storvsc_queuecommand_lck(struct scsi_cmnd *scmnd,
1152                                 void (*done)(struct scsi_cmnd *))
1153 {
1154         int ret;
1155         struct hv_host_device *host_dev =
1156                 (struct hv_host_device *)scmnd->device->host->hostdata;
1157         struct hv_device *dev = host_dev->dev;
1158         struct hv_storvsc_request *request;
1159         struct storvsc_cmd_request *cmd_request;
1160         unsigned int request_size = 0;
1161         int i;
1162         struct scatterlist *sgl;
1163         unsigned int sg_count = 0;
1164         struct vmscsi_request *vm_srb;
1165
1166         if (storvsc_check_scsi_cmd(scmnd) == false) {
1167                 done(scmnd);
1168                 return 0;
1169         }
1170
1171         /* If retrying, no need to prep the cmd */
1172         if (scmnd->host_scribble) {
1173
1174                 cmd_request =
1175                         (struct storvsc_cmd_request *)scmnd->host_scribble;
1176
1177                 goto retry_request;
1178         }
1179
1180         scmnd->scsi_done = done;
1181
1182         request_size = sizeof(struct storvsc_cmd_request);
1183
1184         cmd_request = kmem_cache_zalloc(host_dev->request_pool,
1185                                        GFP_ATOMIC);
1186         if (!cmd_request) {
1187                 scmnd->scsi_done = NULL;
1188                 return SCSI_MLQUEUE_DEVICE_BUSY;
1189         }
1190
1191         /* Setup the cmd request */
1192         cmd_request->bounce_sgl_count = 0;
1193         cmd_request->bounce_sgl = NULL;
1194         cmd_request->cmd = scmnd;
1195
1196         scmnd->host_scribble = (unsigned char *)cmd_request;
1197
1198         request = &cmd_request->request;
1199         vm_srb = &request->vstor_packet.vm_srb;
1200
1201
1202         /* Build the SRB */
1203         switch (scmnd->sc_data_direction) {
1204         case DMA_TO_DEVICE:
1205                 vm_srb->data_in = WRITE_TYPE;
1206                 break;
1207         case DMA_FROM_DEVICE:
1208                 vm_srb->data_in = READ_TYPE;
1209                 break;
1210         default:
1211                 vm_srb->data_in = UNKNOWN_TYPE;
1212                 break;
1213         }
1214
1215         request->on_io_completion = storvsc_command_completion;
1216         request->context = cmd_request;/* scmnd; */
1217
1218         vm_srb->port_number = host_dev->port;
1219         vm_srb->path_id = scmnd->device->channel;
1220         vm_srb->target_id = scmnd->device->id;
1221         vm_srb->lun = scmnd->device->lun;
1222
1223         vm_srb->cdb_length = scmnd->cmd_len;
1224
1225         memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1226
1227         request->sense_buffer = scmnd->sense_buffer;
1228
1229
1230         request->data_buffer.len = scsi_bufflen(scmnd);
1231         if (scsi_sg_count(scmnd)) {
1232                 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1233                 sg_count = scsi_sg_count(scmnd);
1234
1235                 /* check if we need to bounce the sgl */
1236                 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1237                         cmd_request->bounce_sgl =
1238                                 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1239                                                      scsi_bufflen(scmnd));
1240                         if (!cmd_request->bounce_sgl) {
1241                                 scmnd->scsi_done = NULL;
1242                                 scmnd->host_scribble = NULL;
1243                                 kmem_cache_free(host_dev->request_pool,
1244                                                 cmd_request);
1245
1246                                 return SCSI_MLQUEUE_HOST_BUSY;
1247                         }
1248
1249                         cmd_request->bounce_sgl_count =
1250                                 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1251                                         PAGE_SHIFT;
1252
1253                         if (vm_srb->data_in == WRITE_TYPE)
1254                                 copy_to_bounce_buffer(sgl,
1255                                         cmd_request->bounce_sgl,
1256                                         scsi_sg_count(scmnd));
1257
1258                         sgl = cmd_request->bounce_sgl;
1259                         sg_count = cmd_request->bounce_sgl_count;
1260                 }
1261
1262                 request->data_buffer.offset = sgl[0].offset;
1263
1264                 for (i = 0; i < sg_count; i++)
1265                         request->data_buffer.pfn_array[i] =
1266                                 page_to_pfn(sg_page((&sgl[i])));
1267
1268         } else if (scsi_sglist(scmnd)) {
1269                 request->data_buffer.offset =
1270                         virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1271                 request->data_buffer.pfn_array[0] =
1272                         virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1273         }
1274
1275 retry_request:
1276         /* Invokes the vsc to start an IO */
1277         ret = storvsc_do_io(dev, &cmd_request->request);
1278
1279         if (ret == -EAGAIN) {
1280                 /* no more space */
1281
1282                 if (cmd_request->bounce_sgl_count)
1283                         destroy_bounce_buffer(cmd_request->bounce_sgl,
1284                                         cmd_request->bounce_sgl_count);
1285
1286                 kmem_cache_free(host_dev->request_pool, cmd_request);
1287
1288                 scmnd->scsi_done = NULL;
1289                 scmnd->host_scribble = NULL;
1290
1291                 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1292         }
1293
1294         return ret;
1295 }
1296
1297 static DEF_SCSI_QCMD(storvsc_queuecommand)
1298
1299
1300 /* Scsi driver */
1301 static struct scsi_host_template scsi_driver = {
1302         .module =               THIS_MODULE,
1303         .name =                 "storvsc_host_t",
1304         .bios_param =           storvsc_get_chs,
1305         .queuecommand =         storvsc_queuecommand,
1306         .eh_host_reset_handler =        storvsc_host_reset_handler,
1307         .slave_alloc =          storvsc_device_alloc,
1308         .slave_configure =      storvsc_device_configure,
1309         .cmd_per_lun =          1,
1310         /* 64 max_queue * 1 target */
1311         .can_queue =            STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1312         .this_id =              -1,
1313         /* no use setting to 0 since ll_blk_rw reset it to 1 */
1314         /* currently 32 */
1315         .sg_tablesize =         MAX_MULTIPAGE_BUFFER_COUNT,
1316         /*
1317          * ENABLE_CLUSTERING allows mutiple physically contig bio_vecs to merge
1318          * into 1 sg element. If set, we must limit the max_segment_size to
1319          * PAGE_SIZE, otherwise we may get 1 sg element that represents
1320          * multiple
1321          */
1322         /* physically contig pfns (ie sg[x].length > PAGE_SIZE). */
1323         .use_clustering =       ENABLE_CLUSTERING,
1324         /* Make sure we dont get a sg segment crosses a page boundary */
1325         .dma_boundary =         PAGE_SIZE-1,
1326 };
1327
1328 enum {
1329         SCSI_GUID,
1330         IDE_GUID,
1331 };
1332
1333 static const struct hv_vmbus_device_id id_table[] = {
1334         /* SCSI guid */
1335         { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1336                        0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1337           .driver_data = SCSI_GUID },
1338         /* IDE guid */
1339         { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1340                        0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1341           .driver_data = IDE_GUID },
1342         { },
1343 };
1344
1345 MODULE_DEVICE_TABLE(vmbus, id_table);
1346
1347
1348 /*
1349  * storvsc_probe - Add a new device for this driver
1350  */
1351
1352 static int storvsc_probe(struct hv_device *device,
1353                         const struct hv_vmbus_device_id *dev_id)
1354 {
1355         int ret;
1356         struct Scsi_Host *host;
1357         struct hv_host_device *host_dev;
1358         bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1359         int path = 0;
1360         int target = 0;
1361         struct storvsc_device *stor_device;
1362
1363         host = scsi_host_alloc(&scsi_driver,
1364                                sizeof(struct hv_host_device));
1365         if (!host)
1366                 return -ENOMEM;
1367
1368         host_dev = (struct hv_host_device *)host->hostdata;
1369         memset(host_dev, 0, sizeof(struct hv_host_device));
1370
1371         host_dev->port = host->host_no;
1372         host_dev->dev = device;
1373
1374         host_dev->request_pool =
1375                                 kmem_cache_create(dev_name(&device->device),
1376                                         sizeof(struct storvsc_cmd_request), 0,
1377                                         SLAB_HWCACHE_ALIGN, NULL);
1378
1379         if (!host_dev->request_pool) {
1380                 scsi_host_put(host);
1381                 return -ENOMEM;
1382         }
1383
1384         stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1385         if (!stor_device) {
1386                 kmem_cache_destroy(host_dev->request_pool);
1387                 scsi_host_put(host);
1388                 return -ENOMEM;
1389         }
1390
1391         stor_device->destroy = false;
1392         init_waitqueue_head(&stor_device->waiting_to_drain);
1393         stor_device->device = device;
1394         stor_device->host = host;
1395         hv_set_drvdata(device, stor_device);
1396
1397         stor_device->port_number = host->host_no;
1398         ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1399         if (ret) {
1400                 kmem_cache_destroy(host_dev->request_pool);
1401                 scsi_host_put(host);
1402                 kfree(stor_device);
1403                 return ret;
1404         }
1405
1406         if (dev_is_ide)
1407                 storvsc_get_ide_info(device, &target, &path);
1408
1409         host_dev->path = stor_device->path_id;
1410         host_dev->target = stor_device->target_id;
1411
1412         /* max # of devices per target */
1413         host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1414         /* max # of targets per channel */
1415         host->max_id = STORVSC_MAX_TARGETS;
1416         /* max # of channels */
1417         host->max_channel = STORVSC_MAX_CHANNELS - 1;
1418         /* max cmd length */
1419         host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1420
1421         /* Register the HBA and start the scsi bus scan */
1422         ret = scsi_add_host(host, &device->device);
1423         if (ret != 0)
1424                 goto err_out;
1425
1426         if (!dev_is_ide) {
1427                 scsi_scan_host(host);
1428                 return 0;
1429         }
1430         ret = scsi_add_device(host, 0, target, 0);
1431         if (ret) {
1432                 scsi_remove_host(host);
1433                 goto err_out;
1434         }
1435         return 0;
1436
1437 err_out:
1438         storvsc_dev_remove(device);
1439         kmem_cache_destroy(host_dev->request_pool);
1440         scsi_host_put(host);
1441         return -ENODEV;
1442 }
1443
1444 /* The one and only one */
1445
1446 static struct hv_driver storvsc_drv = {
1447         .name = "storvsc",
1448         .id_table = id_table,
1449         .probe = storvsc_probe,
1450         .remove = storvsc_remove,
1451 };
1452
1453 static int __init storvsc_drv_init(void)
1454 {
1455         u32 max_outstanding_req_per_channel;
1456
1457         /*
1458          * Divide the ring buffer data size (which is 1 page less
1459          * than the ring buffer size since that page is reserved for
1460          * the ring buffer indices) by the max request size (which is
1461          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1462          */
1463         max_outstanding_req_per_channel =
1464                 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1465                 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1466                 sizeof(struct vstor_packet) + sizeof(u64),
1467                 sizeof(u64)));
1468
1469         if (max_outstanding_req_per_channel <
1470             STORVSC_MAX_IO_REQUESTS)
1471                 return -EINVAL;
1472
1473         return vmbus_driver_register(&storvsc_drv);
1474 }
1475
1476 static void __exit storvsc_drv_exit(void)
1477 {
1478         vmbus_driver_unregister(&storvsc_drv);
1479 }
1480
1481 MODULE_LICENSE("GPL");
1482 MODULE_VERSION(HV_DRV_VERSION);
1483 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1484 module_init(storvsc_drv_init);
1485 module_exit(storvsc_drv_exit);