drivers/scsi/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 <linux/mempool.h>
  36 #include <scsi/scsi.h>
  37 #include <scsi/scsi_cmnd.h>
  38 #include <scsi/scsi_host.h>
  39 #include <scsi/scsi_device.h>
  40 #include <scsi/scsi_tcq.h>
  41 #include <scsi/scsi_eh.h>
  42 #include <scsi/scsi_devinfo.h>
  43 #include <scsi/scsi_dbg.h>
  44
  45 /*
  46  * All wire protocol details (storage protocol between the guest and the host)
  47  * are consolidated here.
  48  *
  49  * Begin protocol definitions.
  50  */
  51
  52 /*
  53  * Version history:
  54  * V1 Beta: 0.1
  55  * V1 RC < 2008/1/31: 1.0
  56  * V1 RC > 2008/1/31:  2.0
  57  * Win7: 4.2
  58  * Win8: 5.1
  59  */
  60
  61
  62 #define VMSTOR_WIN7_MAJOR 4
  63 #define VMSTOR_WIN7_MINOR 2
  64
  65 #define VMSTOR_WIN8_MAJOR 5
  66 #define VMSTOR_WIN8_MINOR 1
  67
  68
  69 /*  Packet structure describing virtual storage requests. */
  70 enum vstor_packet_operation {
  71         VSTOR_OPERATION_COMPLETE_IO             = 1,
  72         VSTOR_OPERATION_REMOVE_DEVICE           = 2,
  73         VSTOR_OPERATION_EXECUTE_SRB             = 3,
  74         VSTOR_OPERATION_RESET_LUN               = 4,
  75         VSTOR_OPERATION_RESET_ADAPTER           = 5,
  76         VSTOR_OPERATION_RESET_BUS               = 6,
  77         VSTOR_OPERATION_BEGIN_INITIALIZATION    = 7,
  78         VSTOR_OPERATION_END_INITIALIZATION      = 8,
  79         VSTOR_OPERATION_QUERY_PROTOCOL_VERSION  = 9,
  80         VSTOR_OPERATION_QUERY_PROPERTIES        = 10,
  81         VSTOR_OPERATION_ENUMERATE_BUS           = 11,
  82         VSTOR_OPERATION_FCHBA_DATA              = 12,
  83         VSTOR_OPERATION_CREATE_SUB_CHANNELS     = 13,
  84         VSTOR_OPERATION_MAXIMUM                 = 13
  85 };
  86
  87 /*
  88  * WWN packet for Fibre Channel HBA
  89  */
  90
  91 struct hv_fc_wwn_packet {
  92         bool    primary_active;
  93         u8      reserved1;
  94         u8      reserved2;
  95         u8      primary_port_wwn[8];
  96         u8      primary_node_wwn[8];
  97         u8      secondary_port_wwn[8];
  98         u8      secondary_node_wwn[8];
  99 };
 100
 101
 102
 103 /*
 104  * SRB Flag Bits
 105  */
 106
 107 #define SRB_FLAGS_QUEUE_ACTION_ENABLE           0x00000002
 108 #define SRB_FLAGS_DISABLE_DISCONNECT            0x00000004
 109 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER        0x00000008
 110 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE           0x00000010
 111 #define SRB_FLAGS_DISABLE_AUTOSENSE             0x00000020
 112 #define SRB_FLAGS_DATA_IN                       0x00000040
 113 #define SRB_FLAGS_DATA_OUT                      0x00000080
 114 #define SRB_FLAGS_NO_DATA_TRANSFER              0x00000000
 115 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
 116 #define SRB_FLAGS_NO_QUEUE_FREEZE               0x00000100
 117 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE          0x00000200
 118 #define SRB_FLAGS_FREE_SENSE_BUFFER             0x00000400
 119
 120 /*
 121  * This flag indicates the request is part of the workflow for processing a D3.
 122  */
 123 #define SRB_FLAGS_D3_PROCESSING                 0x00000800
 124 #define SRB_FLAGS_IS_ACTIVE                     0x00010000
 125 #define SRB_FLAGS_ALLOCATED_FROM_ZONE           0x00020000
 126 #define SRB_FLAGS_SGLIST_FROM_POOL              0x00040000
 127 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE           0x00080000
 128 #define SRB_FLAGS_NO_KEEP_AWAKE                 0x00100000
 129 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE        0x00200000
 130 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT      0x00400000
 131 #define SRB_FLAGS_DONT_START_NEXT_PACKET        0x00800000
 132 #define SRB_FLAGS_PORT_DRIVER_RESERVED          0x0F000000
 133 #define SRB_FLAGS_CLASS_DRIVER_RESERVED         0xF0000000
 134
 135
 136 /*
 137  * Platform neutral description of a scsi request -
 138  * this remains the same across the write regardless of 32/64 bit
 139  * note: it's patterned off the SCSI_PASS_THROUGH structure
 140  */
 141 #define STORVSC_MAX_CMD_LEN                     0x10
 142
 143 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE     0x14
 144 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE      0x12
 145
 146 #define STORVSC_SENSE_BUFFER_SIZE               0x14
 147 #define STORVSC_MAX_BUF_LEN_WITH_PADDING        0x14
 148
 149 /*
 150  * Sense buffer size changed in win8; have a run-time
 151  * variable to track the size we should use.
 152  */
 153 static int sense_buffer_size;
 154
 155 /*
 156  * The size of the vmscsi_request has changed in win8. The
 157  * additional size is because of new elements added to the
 158  * structure. These elements are valid only when we are talking
 159  * to a win8 host.
 160  * Track the correction to size we need to apply.
 161  */
 162
 163 static int vmscsi_size_delta;
 164 static int vmstor_current_major;
 165 static int vmstor_current_minor;
 166
 167 struct vmscsi_win8_extension {
 168         /*
 169          * The following were added in Windows 8
 170          */
 171         u16 reserve;
 172         u8  queue_tag;
 173         u8  queue_action;
 174         u32 srb_flags;
 175         u32 time_out_value;
 176         u32 queue_sort_ey;
 177 } __packed;
 178
 179 struct vmscsi_request {
 180         u16 length;
 181         u8 srb_status;
 182         u8 scsi_status;
 183
 184         u8  port_number;
 185         u8  path_id;
 186         u8  target_id;
 187         u8  lun;
 188
 189         u8  cdb_length;
 190         u8  sense_info_length;
 191         u8  data_in;
 192         u8  reserved;
 193
 194         u32 data_transfer_length;
 195
 196         union {
 197                 u8 cdb[STORVSC_MAX_CMD_LEN];
 198                 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
 199                 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
 200         };
 201         /*
 202          * The following was added in win8.
 203          */
 204         struct vmscsi_win8_extension win8_extension;
 205
 206 } __attribute((packed));
 207
 208
 209 /*
 210  * This structure is sent during the intialization phase to get the different
 211  * properties of the channel.
 212  */
 213
 214 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL          0x1
 215
 216 struct vmstorage_channel_properties {
 217         u32 reserved;
 218         u16 max_channel_cnt;
 219         u16 reserved1;
 220
 221         u32 flags;
 222         u32   max_transfer_bytes;
 223
 224         u64  reserved2;
 225 } __packed;
 226
 227 /*  This structure is sent during the storage protocol negotiations. */
 228 struct vmstorage_protocol_version {
 229         /* Major (MSW) and minor (LSW) version numbers. */
 230         u16 major_minor;
 231
 232         /*
 233          * Revision number is auto-incremented whenever this file is changed
 234          * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
 235          * definitely indicate incompatibility--but it does indicate mismatched
 236          * builds.
 237          * This is only used on the windows side. Just set it to 0.
 238          */
 239         u16 revision;
 240 } __packed;
 241
 242 /* Channel Property Flags */
 243 #define STORAGE_CHANNEL_REMOVABLE_FLAG          0x1
 244 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG       0x2
 245
 246 struct vstor_packet {
 247         /* Requested operation type */
 248         enum vstor_packet_operation operation;
 249
 250         /*  Flags - see below for values */
 251         u32 flags;
 252
 253         /* Status of the request returned from the server side. */
 254         u32 status;
 255
 256         /* Data payload area */
 257         union {
 258                 /*
 259                  * Structure used to forward SCSI commands from the
 260                  * client to the server.
 261                  */
 262                 struct vmscsi_request vm_srb;
 263
 264                 /* Structure used to query channel properties. */
 265                 struct vmstorage_channel_properties storage_channel_properties;
 266
 267                 /* Used during version negotiations. */
 268                 struct vmstorage_protocol_version version;
 269
 270                 /* Fibre channel address packet */
 271                 struct hv_fc_wwn_packet wwn_packet;
 272
 273                 /* Number of sub-channels to create */
 274                 u16 sub_channel_count;
 275
 276                 /* This will be the maximum of the union members */
 277                 u8  buffer[0x34];
 278         };
 279 } __packed;
 280
 281 /*
 282  * Packet Flags:
 283  *
 284  * This flag indicates that the server should send back a completion for this
 285  * packet.
 286  */
 287
 288 #define REQUEST_COMPLETION_FLAG 0x1
 289
 290 /* Matches Windows-end */
 291 enum storvsc_request_type {
 292         WRITE_TYPE = 0,
 293         READ_TYPE,
 294         UNKNOWN_TYPE,
 295 };
 296
 297 /*
 298  * SRB status codes and masks; a subset of the codes used here.
 299  */
 300
 301 #define SRB_STATUS_AUTOSENSE_VALID      0x80
 302 #define SRB_STATUS_INVALID_LUN  0x20
 303 #define SRB_STATUS_SUCCESS      0x01
 304 #define SRB_STATUS_ABORTED      0x02
 305 #define SRB_STATUS_ERROR        0x04
 306
 307 /*
 308  * This is the end of Protocol specific defines.
 309  */
 310
 311
 312 /*
 313  * We setup a mempool to allocate request structures for this driver
 314  * on a per-lun basis. The following define specifies the number of
 315  * elements in the pool.
 316  */
 317
 318 #define STORVSC_MIN_BUF_NR                              64
 319 static int storvsc_ringbuffer_size = (20 * PAGE_SIZE);
 320
 321 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
 322 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
 323
 324 /*
 325  * Timeout in seconds for all devices managed by this driver.
 326  */
 327 static int storvsc_timeout = 180;
 328
 329 #define STORVSC_MAX_IO_REQUESTS                         128
 330
 331 /*
 332  * In Hyper-V, each port/path/target maps to 1 scsi host adapter.  In
 333  * reality, the path/target is not used (ie always set to 0) so our
 334  * scsi host adapter essentially has 1 bus with 1 target that contains
 335  * up to 256 luns.
 336  */
 337 #define STORVSC_MAX_LUNS_PER_TARGET                     64
 338 #define STORVSC_MAX_TARGETS                             1
 339 #define STORVSC_MAX_CHANNELS                            1
 340
 341
 342
 343 struct storvsc_cmd_request {
 344         struct list_head entry;
 345         struct scsi_cmnd *cmd;
 346
 347         unsigned int bounce_sgl_count;
 348         struct scatterlist *bounce_sgl;
 349
 350         struct hv_device *device;
 351
 352         /* Synchronize the request/response if needed */
 353         struct completion wait_event;
 354
 355         unsigned char *sense_buffer;
 356         struct hv_multipage_buffer data_buffer;
 357         struct vstor_packet vstor_packet;
 358 };
 359
 360
 361 /* A storvsc device is a device object that contains a vmbus channel */
 362 struct storvsc_device {
 363         struct hv_device *device;
 364
 365         bool     destroy;
 366         bool     drain_notify;
 367         atomic_t num_outstanding_req;
 368         struct Scsi_Host *host;
 369
 370         wait_queue_head_t waiting_to_drain;
 371
 372         /*
 373          * Each unique Port/Path/Target represents 1 channel ie scsi
 374          * controller. In reality, the pathid, targetid is always 0
 375          * and the port is set by us
 376          */
 377         unsigned int port_number;
 378         unsigned char path_id;
 379         unsigned char target_id;
 380
 381         /* Used for vsc/vsp channel reset process */
 382         struct storvsc_cmd_request init_request;
 383         struct storvsc_cmd_request reset_request;
 384 };
 385
 386 struct stor_mem_pools {
 387         struct kmem_cache *request_pool;
 388         mempool_t *request_mempool;
 389 };
 390
 391 struct hv_host_device {
 392         struct hv_device *dev;
 393         unsigned int port;
 394         unsigned char path;
 395         unsigned char target;
 396 };
 397
 398 struct storvsc_scan_work {
 399         struct work_struct work;
 400         struct Scsi_Host *host;
 401         uint lun;
 402 };
 403
 404 static void storvsc_device_scan(struct work_struct *work)
 405 {
 406         struct storvsc_scan_work *wrk;
 407         uint lun;
 408         struct scsi_device *sdev;
 409
 410         wrk = container_of(work, struct storvsc_scan_work, work);
 411         lun = wrk->lun;
 412
 413         sdev = scsi_device_lookup(wrk->host, 0, 0, lun);
 414         if (!sdev)
 415                 goto done;
 416         scsi_rescan_device(&sdev->sdev_gendev);
 417         scsi_device_put(sdev);
 418
 419 done:
 420         kfree(wrk);
 421 }
 422
 423 static void storvsc_bus_scan(struct work_struct *work)
 424 {
 425         struct storvsc_scan_work *wrk;
 426         int id, order_id;
 427
 428         wrk = container_of(work, struct storvsc_scan_work, work);
 429         for (id = 0; id < wrk->host->max_id; ++id) {
 430                 if (wrk->host->reverse_ordering)
 431                         order_id = wrk->host->max_id - id - 1;
 432                 else
 433                         order_id = id;
 434
 435                 scsi_scan_target(&wrk->host->shost_gendev, 0,
 436                                 order_id, SCAN_WILD_CARD, 1);
 437         }
 438         kfree(wrk);
 439 }
 440
 441 static void storvsc_remove_lun(struct work_struct *work)
 442 {
 443         struct storvsc_scan_work *wrk;
 444         struct scsi_device *sdev;
 445
 446         wrk = container_of(work, struct storvsc_scan_work, work);
 447         if (!scsi_host_get(wrk->host))
 448                 goto done;
 449
 450         sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun);
 451
 452         if (sdev) {
 453                 scsi_remove_device(sdev);
 454                 scsi_device_put(sdev);
 455         }
 456         scsi_host_put(wrk->host);
 457
 458 done:
 459         kfree(wrk);
 460 }
 461
 462 /*
 463  * Major/minor macros.  Minor version is in LSB, meaning that earlier flat
 464  * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1).
 465  */
 466
 467 static inline u16 storvsc_get_version(u8 major, u8 minor)
 468 {
 469         u16 version;
 470
 471         version = ((major << 8) | minor);
 472         return version;
 473 }
 474
 475 /*
 476  * We can get incoming messages from the host that are not in response to
 477  * messages that we have sent out. An example of this would be messages
 478  * received by the guest to notify dynamic addition/removal of LUNs. To
 479  * deal with potential race conditions where the driver may be in the
 480  * midst of being unloaded when we might receive an unsolicited message
 481  * from the host, we have implemented a mechanism to gurantee sequential
 482  * consistency:
 483  *
 484  * 1) Once the device is marked as being destroyed, we will fail all
 485  *    outgoing messages.
 486  * 2) We permit incoming messages when the device is being destroyed,
 487  *    only to properly account for messages already sent out.
 488  */
 489
 490 static inline struct storvsc_device *get_out_stor_device(
 491                                         struct hv_device *device)
 492 {
 493         struct storvsc_device *stor_device;
 494
 495         stor_device = hv_get_drvdata(device);
 496
 497         if (stor_device && stor_device->destroy)
 498                 stor_device = NULL;
 499
 500         return stor_device;
 501 }
 502
 503
 504 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
 505 {
 506         dev->drain_notify = true;
 507         wait_event(dev->waiting_to_drain,
 508                    atomic_read(&dev->num_outstanding_req) == 0);
 509         dev->drain_notify = false;
 510 }
 511
 512 static inline struct storvsc_device *get_in_stor_device(
 513                                         struct hv_device *device)
 514 {
 515         struct storvsc_device *stor_device;
 516
 517         stor_device = hv_get_drvdata(device);
 518
 519         if (!stor_device)
 520                 goto get_in_err;
 521
 522         /*
 523          * If the device is being destroyed; allow incoming
 524          * traffic only to cleanup outstanding requests.
 525          */
 526
 527         if (stor_device->destroy  &&
 528                 (atomic_read(&stor_device->num_outstanding_req) == 0))
 529                 stor_device = NULL;
 530
 531 get_in_err:
 532         return stor_device;
 533
 534 }
 535
 536 static void destroy_bounce_buffer(struct scatterlist *sgl,
 537                                   unsigned int sg_count)
 538 {
 539         int i;
 540         struct page *page_buf;
 541
 542         for (i = 0; i < sg_count; i++) {
 543                 page_buf = sg_page((&sgl[i]));
 544                 if (page_buf != NULL)
 545                         __free_page(page_buf);
 546         }
 547
 548         kfree(sgl);
 549 }
 550
 551 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
 552 {
 553         int i;
 554
 555         /* No need to check */
 556         if (sg_count < 2)
 557                 return -1;
 558
 559         /* We have at least 2 sg entries */
 560         for (i = 0; i < sg_count; i++) {
 561                 if (i == 0) {
 562                         /* make sure 1st one does not have hole */
 563                         if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
 564                                 return i;
 565                 } else if (i == sg_count - 1) {
 566                         /* make sure last one does not have hole */
 567                         if (sgl[i].offset != 0)
 568                                 return i;
 569                 } else {
 570                         /* make sure no hole in the middle */
 571                         if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
 572                                 return i;
 573                 }
 574         }
 575         return -1;
 576 }
 577
 578 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
 579                                                 unsigned int sg_count,
 580                                                 unsigned int len,
 581                                                 int write)
 582 {
 583         int i;
 584         int num_pages;
 585         struct scatterlist *bounce_sgl;
 586         struct page *page_buf;
 587         unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
 588
 589         num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
 590
 591         bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
 592         if (!bounce_sgl)
 593                 return NULL;
 594
 595         sg_init_table(bounce_sgl, num_pages);
 596         for (i = 0; i < num_pages; i++) {
 597                 page_buf = alloc_page(GFP_ATOMIC);
 598                 if (!page_buf)
 599                         goto cleanup;
 600                 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0);
 601         }
 602
 603         return bounce_sgl;
 604
 605 cleanup:
 606         destroy_bounce_buffer(bounce_sgl, num_pages);
 607         return NULL;
 608 }
 609
 610 /* Disgusting wrapper functions */
 611 static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx)
 612 {
 613         void *addr = kmap_atomic(sg_page(sgl + idx));
 614         return (unsigned long)addr;
 615 }
 616
 617 static inline void sg_kunmap_atomic(unsigned long addr)
 618 {
 619         kunmap_atomic((void *)addr);
 620 }
 621
 622
 623 /* Assume the original sgl has enough room */
 624 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
 625                                             struct scatterlist *bounce_sgl,
 626                                             unsigned int orig_sgl_count,
 627                                             unsigned int bounce_sgl_count)
 628 {
 629         int i;
 630         int j = 0;
 631         unsigned long src, dest;
 632         unsigned int srclen, destlen, copylen;
 633         unsigned int total_copied = 0;
 634         unsigned long bounce_addr = 0;
 635         unsigned long dest_addr = 0;
 636         unsigned long flags;
 637
 638         local_irq_save(flags);
 639
 640         for (i = 0; i < orig_sgl_count; i++) {
 641                 dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
 642                 dest = dest_addr;
 643                 destlen = orig_sgl[i].length;
 644
 645                 if (bounce_addr == 0)
 646                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 647
 648                 while (destlen) {
 649                         src = bounce_addr + bounce_sgl[j].offset;
 650                         srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
 651
 652                         copylen = min(srclen, destlen);
 653                         memcpy((void *)dest, (void *)src, copylen);
 654
 655                         total_copied += copylen;
 656                         bounce_sgl[j].offset += copylen;
 657                         destlen -= copylen;
 658                         dest += copylen;
 659
 660                         if (bounce_sgl[j].offset == bounce_sgl[j].length) {
 661                                 /* full */
 662                                 sg_kunmap_atomic(bounce_addr);
 663                                 j++;
 664
 665                                 /*
 666                                  * It is possible that the number of elements
 667                                  * in the bounce buffer may not be equal to
 668                                  * the number of elements in the original
 669                                  * scatter list. Handle this correctly.
 670                                  */
 671
 672                                 if (j == bounce_sgl_count) {
 673                                         /*
 674                                          * We are done; cleanup and return.
 675                                          */
 676                                         sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
 677                                         local_irq_restore(flags);
 678                                         return total_copied;
 679                                 }
 680
 681                                 /* if we need to use another bounce buffer */
 682                                 if (destlen || i != orig_sgl_count - 1)
 683                                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 684                         } else if (destlen == 0 && i == orig_sgl_count - 1) {
 685                                 /* unmap the last bounce that is < PAGE_SIZE */
 686                                 sg_kunmap_atomic(bounce_addr);
 687                         }
 688                 }
 689
 690                 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
 691         }
 692
 693         local_irq_restore(flags);
 694
 695         return total_copied;
 696 }
 697
 698 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
 699 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
 700                                           struct scatterlist *bounce_sgl,
 701                                           unsigned int orig_sgl_count)
 702 {
 703         int i;
 704         int j = 0;
 705         unsigned long src, dest;
 706         unsigned int srclen, destlen, copylen;
 707         unsigned int total_copied = 0;
 708         unsigned long bounce_addr = 0;
 709         unsigned long src_addr = 0;
 710         unsigned long flags;
 711
 712         local_irq_save(flags);
 713
 714         for (i = 0; i < orig_sgl_count; i++) {
 715                 src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
 716                 src = src_addr;
 717                 srclen = orig_sgl[i].length;
 718
 719                 if (bounce_addr == 0)
 720                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 721
 722                 while (srclen) {
 723                         /* assume bounce offset always == 0 */
 724                         dest = bounce_addr + bounce_sgl[j].length;
 725                         destlen = PAGE_SIZE - bounce_sgl[j].length;
 726
 727                         copylen = min(srclen, destlen);
 728                         memcpy((void *)dest, (void *)src, copylen);
 729
 730                         total_copied += copylen;
 731                         bounce_sgl[j].length += copylen;
 732                         srclen -= copylen;
 733                         src += copylen;
 734
 735                         if (bounce_sgl[j].length == PAGE_SIZE) {
 736                                 /* full..move to next entry */
 737                                 sg_kunmap_atomic(bounce_addr);
 738                                 j++;
 739
 740                                 /* if we need to use another bounce buffer */
 741                                 if (srclen || i != orig_sgl_count - 1)
 742                                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 743
 744                         } else if (srclen == 0 && i == orig_sgl_count - 1) {
 745                                 /* unmap the last bounce that is < PAGE_SIZE */
 746                                 sg_kunmap_atomic(bounce_addr);
 747                         }
 748                 }
 749
 750                 sg_kunmap_atomic(src_addr - orig_sgl[i].offset);
 751         }
 752
 753         local_irq_restore(flags);
 754
 755         return total_copied;
 756 }
 757
 758 static int storvsc_channel_init(struct hv_device *device)
 759 {
 760         struct storvsc_device *stor_device;
 761         struct storvsc_cmd_request *request;
 762         struct vstor_packet *vstor_packet;
 763         int ret, t;
 764
 765         stor_device = get_out_stor_device(device);
 766         if (!stor_device)
 767                 return -ENODEV;
 768
 769         request = &stor_device->init_request;
 770         vstor_packet = &request->vstor_packet;
 771
 772         /*
 773          * Now, initiate the vsc/vsp initialization protocol on the open
 774          * channel
 775          */
 776         memset(request, 0, sizeof(struct storvsc_cmd_request));
 777         init_completion(&request->wait_event);
 778         vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
 779         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 780
 781         ret = vmbus_sendpacket(device->channel, vstor_packet,
 782                                (sizeof(struct vstor_packet) -
 783                                vmscsi_size_delta),
 784                                (unsigned long)request,
 785                                VM_PKT_DATA_INBAND,
 786                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 787         if (ret != 0)
 788                 goto cleanup;
 789
 790         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 791         if (t == 0) {
 792                 ret = -ETIMEDOUT;
 793                 goto cleanup;
 794         }
 795
 796         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 797             vstor_packet->status != 0)
 798                 goto cleanup;
 799
 800
 801         /* reuse the packet for version range supported */
 802         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 803         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
 804         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 805
 806         vstor_packet->version.major_minor =
 807                 storvsc_get_version(vmstor_current_major, vmstor_current_minor);
 808
 809         /*
 810          * The revision number is only used in Windows; set it to 0.
 811          */
 812         vstor_packet->version.revision = 0;
 813
 814         ret = vmbus_sendpacket(device->channel, vstor_packet,
 815                                (sizeof(struct vstor_packet) -
 816                                 vmscsi_size_delta),
 817                                (unsigned long)request,
 818                                VM_PKT_DATA_INBAND,
 819                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 820         if (ret != 0)
 821                 goto cleanup;
 822
 823         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 824         if (t == 0) {
 825                 ret = -ETIMEDOUT;
 826                 goto cleanup;
 827         }
 828
 829         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 830             vstor_packet->status != 0)
 831                 goto cleanup;
 832
 833
 834         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 835         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
 836         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 837
 838         ret = vmbus_sendpacket(device->channel, vstor_packet,
 839                                (sizeof(struct vstor_packet) -
 840                                 vmscsi_size_delta),
 841                                (unsigned long)request,
 842                                VM_PKT_DATA_INBAND,
 843                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 844
 845         if (ret != 0)
 846                 goto cleanup;
 847
 848         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 849         if (t == 0) {
 850                 ret = -ETIMEDOUT;
 851                 goto cleanup;
 852         }
 853
 854         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 855             vstor_packet->status != 0)
 856                 goto cleanup;
 857
 858         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 859         vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
 860         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 861
 862         ret = vmbus_sendpacket(device->channel, vstor_packet,
 863                                (sizeof(struct vstor_packet) -
 864                                 vmscsi_size_delta),
 865                                (unsigned long)request,
 866                                VM_PKT_DATA_INBAND,
 867                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 868
 869         if (ret != 0)
 870                 goto cleanup;
 871
 872         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 873         if (t == 0) {
 874                 ret = -ETIMEDOUT;
 875                 goto cleanup;
 876         }
 877
 878         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 879             vstor_packet->status != 0)
 880                 goto cleanup;
 881
 882
 883 cleanup:
 884         return ret;
 885 }
 886
 887 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
 888                                 struct scsi_cmnd *scmnd,
 889                                 struct Scsi_Host *host,
 890                                 u8 asc, u8 ascq)
 891 {
 892         struct storvsc_scan_work *wrk;
 893         void (*process_err_fn)(struct work_struct *work);
 894         bool do_work = false;
 895
 896         switch (vm_srb->srb_status) {
 897         case SRB_STATUS_ERROR:
 898                 /*
 899                  * If there is an error; offline the device since all
 900                  * error recovery strategies would have already been
 901                  * deployed on the host side. However, if the command
 902                  * were a pass-through command deal with it appropriately.
 903                  */
 904                 switch (scmnd->cmnd[0]) {
 905                 case ATA_16:
 906                 case ATA_12:
 907                         set_host_byte(scmnd, DID_PASSTHROUGH);
 908                         break;
 909                 default:
 910                         set_host_byte(scmnd, DID_TARGET_FAILURE);
 911                 }
 912                 break;
 913         case SRB_STATUS_INVALID_LUN:
 914                 do_work = true;
 915                 process_err_fn = storvsc_remove_lun;
 916                 break;
 917         case (SRB_STATUS_ABORTED | SRB_STATUS_AUTOSENSE_VALID):
 918                 if ((asc == 0x2a) && (ascq == 0x9)) {
 919                         do_work = true;
 920                         process_err_fn = storvsc_device_scan;
 921                         /*
 922                          * Retry the I/O that trigerred this.
 923                          */
 924                         set_host_byte(scmnd, DID_REQUEUE);
 925                 }
 926                 break;
 927         }
 928
 929         if (!do_work)
 930                 return;
 931
 932         /*
 933          * We need to schedule work to process this error; schedule it.
 934          */
 935         wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
 936         if (!wrk) {
 937                 set_host_byte(scmnd, DID_TARGET_FAILURE);
 938                 return;
 939         }
 940
 941         wrk->host = host;
 942         wrk->lun = vm_srb->lun;
 943         INIT_WORK(&wrk->work, process_err_fn);
 944         schedule_work(&wrk->work);
 945 }
 946
 947
 948 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
 949 {
 950         struct scsi_cmnd *scmnd = cmd_request->cmd;
 951         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
 952         void (*scsi_done_fn)(struct scsi_cmnd *);
 953         struct scsi_sense_hdr sense_hdr;
 954         struct vmscsi_request *vm_srb;
 955         struct stor_mem_pools *memp = scmnd->device->hostdata;
 956         struct Scsi_Host *host;
 957         struct storvsc_device *stor_dev;
 958         struct hv_device *dev = host_dev->dev;
 959
 960         stor_dev = get_in_stor_device(dev);
 961         host = stor_dev->host;
 962
 963         vm_srb = &cmd_request->vstor_packet.vm_srb;
 964         if (cmd_request->bounce_sgl_count) {
 965                 if (vm_srb->data_in == READ_TYPE)
 966                         copy_from_bounce_buffer(scsi_sglist(scmnd),
 967                                         cmd_request->bounce_sgl,
 968                                         scsi_sg_count(scmnd),
 969                                         cmd_request->bounce_sgl_count);
 970                 destroy_bounce_buffer(cmd_request->bounce_sgl,
 971                                         cmd_request->bounce_sgl_count);
 972         }
 973
 974         scmnd->result = vm_srb->scsi_status;
 975
 976         if (scmnd->result) {
 977                 if (scsi_normalize_sense(scmnd->sense_buffer,
 978                                 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
 979                         scsi_print_sense_hdr("storvsc", &sense_hdr);
 980         }
 981
 982         if (vm_srb->srb_status != SRB_STATUS_SUCCESS)
 983                 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
 984                                          sense_hdr.ascq);
 985
 986         scsi_set_resid(scmnd,
 987                 cmd_request->data_buffer.len -
 988                 vm_srb->data_transfer_length);
 989
 990         scsi_done_fn = scmnd->scsi_done;
 991
 992         scmnd->host_scribble = NULL;
 993         scmnd->scsi_done = NULL;
 994
 995         scsi_done_fn(scmnd);
 996
 997         mempool_free(cmd_request, memp->request_mempool);
 998 }
 999
1000 static void storvsc_on_io_completion(struct hv_device *device,
1001                                   struct vstor_packet *vstor_packet,
1002                                   struct storvsc_cmd_request *request)
1003 {
1004         struct storvsc_device *stor_device;
1005         struct vstor_packet *stor_pkt;
1006
1007         stor_device = hv_get_drvdata(device);
1008         stor_pkt = &request->vstor_packet;
1009
1010         /*
1011          * The current SCSI handling on the host side does
1012          * not correctly handle:
1013          * INQUIRY command with page code parameter set to 0x80
1014          * MODE_SENSE command with cmd[2] == 0x1c
1015          *
1016          * Setup srb and scsi status so this won't be fatal.
1017          * We do this so we can distinguish truly fatal failues
1018          * (srb status == 0x4) and off-line the device in that case.
1019          */
1020
1021         if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1022            (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1023                 vstor_packet->vm_srb.scsi_status = 0;
1024                 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1025         }
1026
1027
1028         /* Copy over the status...etc */
1029         stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1030         stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1031         stor_pkt->vm_srb.sense_info_length =
1032         vstor_packet->vm_srb.sense_info_length;
1033
1034         if (vstor_packet->vm_srb.scsi_status != 0 ||
1035                 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){
1036                 dev_warn(&device->device,
1037                          "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
1038                          stor_pkt->vm_srb.cdb[0],
1039                          vstor_packet->vm_srb.scsi_status,
1040                          vstor_packet->vm_srb.srb_status);
1041         }
1042
1043         if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
1044                 /* CHECK_CONDITION */
1045                 if (vstor_packet->vm_srb.srb_status &
1046                         SRB_STATUS_AUTOSENSE_VALID) {
1047                         /* autosense data available */
1048                         dev_warn(&device->device,
1049                                  "stor pkt %p autosense data valid - len %d\n",
1050                                  request,
1051                                  vstor_packet->vm_srb.sense_info_length);
1052
1053                         memcpy(request->sense_buffer,
1054                                vstor_packet->vm_srb.sense_data,
1055                                vstor_packet->vm_srb.sense_info_length);
1056
1057                 }
1058         }
1059
1060         stor_pkt->vm_srb.data_transfer_length =
1061         vstor_packet->vm_srb.data_transfer_length;
1062
1063         storvsc_command_completion(request);
1064
1065         if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1066                 stor_device->drain_notify)
1067                 wake_up(&stor_device->waiting_to_drain);
1068
1069
1070 }
1071
1072 static void storvsc_on_receive(struct hv_device *device,
1073                              struct vstor_packet *vstor_packet,
1074                              struct storvsc_cmd_request *request)
1075 {
1076         struct storvsc_scan_work *work;
1077         struct storvsc_device *stor_device;
1078
1079         switch (vstor_packet->operation) {
1080         case VSTOR_OPERATION_COMPLETE_IO:
1081                 storvsc_on_io_completion(device, vstor_packet, request);
1082                 break;
1083
1084         case VSTOR_OPERATION_REMOVE_DEVICE:
1085         case VSTOR_OPERATION_ENUMERATE_BUS:
1086                 stor_device = get_in_stor_device(device);
1087                 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1088                 if (!work)
1089                         return;
1090
1091                 INIT_WORK(&work->work, storvsc_bus_scan);
1092                 work->host = stor_device->host;
1093                 schedule_work(&work->work);
1094                 break;
1095
1096         default:
1097                 break;
1098         }
1099 }
1100
1101 static void storvsc_on_channel_callback(void *context)
1102 {
1103         struct hv_device *device = (struct hv_device *)context;
1104         struct storvsc_device *stor_device;
1105         u32 bytes_recvd;
1106         u64 request_id;
1107         unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
1108         struct storvsc_cmd_request *request;
1109         int ret;
1110
1111
1112         stor_device = get_in_stor_device(device);
1113         if (!stor_device)
1114                 return;
1115
1116         do {
1117                 ret = vmbus_recvpacket(device->channel, packet,
1118                                        ALIGN((sizeof(struct vstor_packet) -
1119                                              vmscsi_size_delta), 8),
1120                                        &bytes_recvd, &request_id);
1121                 if (ret == 0 && bytes_recvd > 0) {
1122
1123                         request = (struct storvsc_cmd_request *)
1124                                         (unsigned long)request_id;
1125
1126                         if ((request == &stor_device->init_request) ||
1127                             (request == &stor_device->reset_request)) {
1128
1129                                 memcpy(&request->vstor_packet, packet,
1130                                        (sizeof(struct vstor_packet) -
1131                                         vmscsi_size_delta));
1132                                 complete(&request->wait_event);
1133                         } else {
1134                                 storvsc_on_receive(device,
1135                                                 (struct vstor_packet *)packet,
1136                                                 request);
1137                         }
1138                 } else {
1139                         break;
1140                 }
1141         } while (1);
1142
1143         return;
1144 }
1145
1146 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
1147 {
1148         struct vmstorage_channel_properties props;
1149         int ret;
1150
1151         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1152
1153         ret = vmbus_open(device->channel,
1154                          ring_size,
1155                          ring_size,
1156                          (void *)&props,
1157                          sizeof(struct vmstorage_channel_properties),
1158                          storvsc_on_channel_callback, device);
1159
1160         if (ret != 0)
1161                 return ret;
1162
1163         ret = storvsc_channel_init(device);
1164
1165         return ret;
1166 }
1167
1168 static int storvsc_dev_remove(struct hv_device *device)
1169 {
1170         struct storvsc_device *stor_device;
1171         unsigned long flags;
1172
1173         stor_device = hv_get_drvdata(device);
1174
1175         spin_lock_irqsave(&device->channel->inbound_lock, flags);
1176         stor_device->destroy = true;
1177         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1178
1179         /*
1180          * At this point, all outbound traffic should be disable. We
1181          * only allow inbound traffic (responses) to proceed so that
1182          * outstanding requests can be completed.
1183          */
1184
1185         storvsc_wait_to_drain(stor_device);
1186
1187         /*
1188          * Since we have already drained, we don't need to busy wait
1189          * as was done in final_release_stor_device()
1190          * Note that we cannot set the ext pointer to NULL until
1191          * we have drained - to drain the outgoing packets, we need to
1192          * allow incoming packets.
1193          */
1194         spin_lock_irqsave(&device->channel->inbound_lock, flags);
1195         hv_set_drvdata(device, NULL);
1196         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1197
1198         /* Close the channel */
1199         vmbus_close(device->channel);
1200
1201         kfree(stor_device);
1202         return 0;
1203 }
1204
1205 static int storvsc_do_io(struct hv_device *device,
1206                               struct storvsc_cmd_request *request)
1207 {
1208         struct storvsc_device *stor_device;
1209         struct vstor_packet *vstor_packet;
1210         int ret = 0;
1211
1212         vstor_packet = &request->vstor_packet;
1213         stor_device = get_out_stor_device(device);
1214
1215         if (!stor_device)
1216                 return -ENODEV;
1217
1218
1219         request->device  = device;
1220
1221
1222         vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1223
1224         vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1225                                         vmscsi_size_delta);
1226
1227
1228         vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1229
1230
1231         vstor_packet->vm_srb.data_transfer_length =
1232         request->data_buffer.len;
1233
1234         vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1235
1236         if (request->data_buffer.len) {
1237                 ret = vmbus_sendpacket_multipagebuffer(device->channel,
1238                                 &request->data_buffer,
1239                                 vstor_packet,
1240                                 (sizeof(struct vstor_packet) -
1241                                 vmscsi_size_delta),
1242                                 (unsigned long)request);
1243         } else {
1244                 ret = vmbus_sendpacket(device->channel, vstor_packet,
1245                                (sizeof(struct vstor_packet) -
1246                                 vmscsi_size_delta),
1247                                (unsigned long)request,
1248                                VM_PKT_DATA_INBAND,
1249                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1250         }
1251
1252         if (ret != 0)
1253                 return ret;
1254
1255         atomic_inc(&stor_device->num_outstanding_req);
1256
1257         return ret;
1258 }
1259
1260 static int storvsc_device_alloc(struct scsi_device *sdevice)
1261 {
1262         struct stor_mem_pools *memp;
1263         int number = STORVSC_MIN_BUF_NR;
1264
1265         memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
1266         if (!memp)
1267                 return -ENOMEM;
1268
1269         memp->request_pool =
1270                 kmem_cache_create(dev_name(&sdevice->sdev_dev),
1271                                 sizeof(struct storvsc_cmd_request), 0,
1272                                 SLAB_HWCACHE_ALIGN, NULL);
1273
1274         if (!memp->request_pool)
1275                 goto err0;
1276
1277         memp->request_mempool = mempool_create(number, mempool_alloc_slab,
1278                                                 mempool_free_slab,
1279                                                 memp->request_pool);
1280
1281         if (!memp->request_mempool)
1282                 goto err1;
1283
1284         sdevice->hostdata = memp;
1285
1286         return 0;
1287
1288 err1:
1289         kmem_cache_destroy(memp->request_pool);
1290
1291 err0:
1292         kfree(memp);
1293         return -ENOMEM;
1294 }
1295
1296 static void storvsc_device_destroy(struct scsi_device *sdevice)
1297 {
1298         struct stor_mem_pools *memp = sdevice->hostdata;
1299
1300         mempool_destroy(memp->request_mempool);
1301         kmem_cache_destroy(memp->request_pool);
1302         kfree(memp);
1303         sdevice->hostdata = NULL;
1304 }
1305
1306 static int storvsc_device_configure(struct scsi_device *sdevice)
1307 {
1308         scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
1309                                 STORVSC_MAX_IO_REQUESTS);
1310
1311         blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
1312
1313         blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
1314
1315         blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1316
1317         sdevice->no_write_same = 1;
1318
1319         return 0;
1320 }
1321
1322 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1323                            sector_t capacity, int *info)
1324 {
1325         sector_t nsect = capacity;
1326         sector_t cylinders = nsect;
1327         int heads, sectors_pt;
1328
1329         /*
1330          * We are making up these values; let us keep it simple.
1331          */
1332         heads = 0xff;
1333         sectors_pt = 0x3f;      /* Sectors per track */
1334         sector_div(cylinders, heads * sectors_pt);
1335         if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1336                 cylinders = 0xffff;
1337
1338         info[0] = heads;
1339         info[1] = sectors_pt;
1340         info[2] = (int)cylinders;
1341
1342         return 0;
1343 }
1344
1345 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1346 {
1347         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1348         struct hv_device *device = host_dev->dev;
1349
1350         struct storvsc_device *stor_device;
1351         struct storvsc_cmd_request *request;
1352         struct vstor_packet *vstor_packet;
1353         int ret, t;
1354
1355
1356         stor_device = get_out_stor_device(device);
1357         if (!stor_device)
1358                 return FAILED;
1359
1360         request = &stor_device->reset_request;
1361         vstor_packet = &request->vstor_packet;
1362
1363         init_completion(&request->wait_event);
1364
1365         vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1366         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1367         vstor_packet->vm_srb.path_id = stor_device->path_id;
1368
1369         ret = vmbus_sendpacket(device->channel, vstor_packet,
1370                                (sizeof(struct vstor_packet) -
1371                                 vmscsi_size_delta),
1372                                (unsigned long)&stor_device->reset_request,
1373                                VM_PKT_DATA_INBAND,
1374                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1375         if (ret != 0)
1376                 return FAILED;
1377
1378         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1379         if (t == 0)
1380                 return TIMEOUT_ERROR;
1381
1382
1383         /*
1384          * At this point, all outstanding requests in the adapter
1385          * should have been flushed out and return to us
1386          * There is a potential race here where the host may be in
1387          * the process of responding when we return from here.
1388          * Just wait for all in-transit packets to be accounted for
1389          * before we return from here.
1390          */
1391         storvsc_wait_to_drain(stor_device);
1392
1393         return SUCCESS;
1394 }
1395
1396 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1397 {
1398         bool allowed = true;
1399         u8 scsi_op = scmnd->cmnd[0];
1400
1401         switch (scsi_op) {
1402         /* the host does not handle WRITE_SAME, log accident usage */
1403         case WRITE_SAME:
1404         /*
1405          * smartd sends this command and the host does not handle
1406          * this. So, don't send it.
1407          */
1408         case SET_WINDOW:
1409                 scmnd->result = ILLEGAL_REQUEST << 16;
1410                 allowed = false;
1411                 break;
1412         default:
1413                 break;
1414         }
1415         return allowed;
1416 }
1417
1418 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1419 {
1420         int ret;
1421         struct hv_host_device *host_dev = shost_priv(host);
1422         struct hv_device *dev = host_dev->dev;
1423         struct storvsc_cmd_request *cmd_request;
1424         unsigned int request_size = 0;
1425         int i;
1426         struct scatterlist *sgl;
1427         unsigned int sg_count = 0;
1428         struct vmscsi_request *vm_srb;
1429         struct stor_mem_pools *memp = scmnd->device->hostdata;
1430
1431         if (!storvsc_scsi_cmd_ok(scmnd)) {
1432                 scmnd->scsi_done(scmnd);
1433                 return 0;
1434         }
1435
1436         request_size = sizeof(struct storvsc_cmd_request);
1437
1438         cmd_request = mempool_alloc(memp->request_mempool,
1439                                        GFP_ATOMIC);
1440
1441         /*
1442          * We might be invoked in an interrupt context; hence
1443          * mempool_alloc() can fail.
1444          */
1445         if (!cmd_request)
1446                 return SCSI_MLQUEUE_DEVICE_BUSY;
1447
1448         memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1449
1450         /* Setup the cmd request */
1451         cmd_request->cmd = scmnd;
1452
1453         scmnd->host_scribble = (unsigned char *)cmd_request;
1454
1455         vm_srb = &cmd_request->vstor_packet.vm_srb;
1456         vm_srb->win8_extension.time_out_value = 60;
1457
1458
1459         /* Build the SRB */
1460         switch (scmnd->sc_data_direction) {
1461         case DMA_TO_DEVICE:
1462                 vm_srb->data_in = WRITE_TYPE;
1463                 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1464                 vm_srb->win8_extension.srb_flags |=
1465                         (SRB_FLAGS_QUEUE_ACTION_ENABLE |
1466                         SRB_FLAGS_DISABLE_SYNCH_TRANSFER);
1467                 break;
1468         case DMA_FROM_DEVICE:
1469                 vm_srb->data_in = READ_TYPE;
1470                 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1471                 vm_srb->win8_extension.srb_flags |=
1472                         (SRB_FLAGS_QUEUE_ACTION_ENABLE |
1473                         SRB_FLAGS_DISABLE_SYNCH_TRANSFER);
1474                 break;
1475         default:
1476                 vm_srb->data_in = UNKNOWN_TYPE;
1477                 vm_srb->win8_extension.srb_flags = 0;
1478                 break;
1479         }
1480
1481
1482         vm_srb->port_number = host_dev->port;
1483         vm_srb->path_id = scmnd->device->channel;
1484         vm_srb->target_id = scmnd->device->id;
1485         vm_srb->lun = scmnd->device->lun;
1486
1487         vm_srb->cdb_length = scmnd->cmd_len;
1488
1489         memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1490
1491         cmd_request->sense_buffer = scmnd->sense_buffer;
1492
1493
1494         cmd_request->data_buffer.len = scsi_bufflen(scmnd);
1495         if (scsi_sg_count(scmnd)) {
1496                 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1497                 sg_count = scsi_sg_count(scmnd);
1498
1499                 /* check if we need to bounce the sgl */
1500                 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1501                         cmd_request->bounce_sgl =
1502                                 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1503                                                      scsi_bufflen(scmnd),
1504                                                      vm_srb->data_in);
1505                         if (!cmd_request->bounce_sgl) {
1506                                 ret = SCSI_MLQUEUE_HOST_BUSY;
1507                                 goto queue_error;
1508                         }
1509
1510                         cmd_request->bounce_sgl_count =
1511                                 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1512                                         PAGE_SHIFT;
1513
1514                         if (vm_srb->data_in == WRITE_TYPE)
1515                                 copy_to_bounce_buffer(sgl,
1516                                         cmd_request->bounce_sgl,
1517                                         scsi_sg_count(scmnd));
1518
1519                         sgl = cmd_request->bounce_sgl;
1520                         sg_count = cmd_request->bounce_sgl_count;
1521                 }
1522
1523                 cmd_request->data_buffer.offset = sgl[0].offset;
1524
1525                 for (i = 0; i < sg_count; i++)
1526                         cmd_request->data_buffer.pfn_array[i] =
1527                                 page_to_pfn(sg_page((&sgl[i])));
1528
1529         } else if (scsi_sglist(scmnd)) {
1530                 cmd_request->data_buffer.offset =
1531                         virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1532                 cmd_request->data_buffer.pfn_array[0] =
1533                         virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1534         }
1535
1536         /* Invokes the vsc to start an IO */
1537         ret = storvsc_do_io(dev, cmd_request);
1538
1539         if (ret == -EAGAIN) {
1540                 /* no more space */
1541
1542                 if (cmd_request->bounce_sgl_count) {
1543                         destroy_bounce_buffer(cmd_request->bounce_sgl,
1544                                         cmd_request->bounce_sgl_count);
1545
1546                         ret = SCSI_MLQUEUE_DEVICE_BUSY;
1547                         goto queue_error;
1548                 }
1549         }
1550
1551         return 0;
1552
1553 queue_error:
1554         mempool_free(cmd_request, memp->request_mempool);
1555         scmnd->host_scribble = NULL;
1556         return ret;
1557 }
1558
1559 static struct scsi_host_template scsi_driver = {
1560         .module =               THIS_MODULE,
1561         .name =                 "storvsc_host_t",
1562         .bios_param =           storvsc_get_chs,
1563         .queuecommand =         storvsc_queuecommand,
1564         .eh_host_reset_handler =        storvsc_host_reset_handler,
1565         .slave_alloc =          storvsc_device_alloc,
1566         .slave_destroy =        storvsc_device_destroy,
1567         .slave_configure =      storvsc_device_configure,
1568         .cmd_per_lun =          1,
1569         /* 64 max_queue * 1 target */
1570         .can_queue =            STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1571         .this_id =              -1,
1572         /* no use setting to 0 since ll_blk_rw reset it to 1 */
1573         /* currently 32 */
1574         .sg_tablesize =         MAX_MULTIPAGE_BUFFER_COUNT,
1575         .use_clustering =       DISABLE_CLUSTERING,
1576         /* Make sure we dont get a sg segment crosses a page boundary */
1577         .dma_boundary =         PAGE_SIZE-1,
1578 };
1579
1580 enum {
1581         SCSI_GUID,
1582         IDE_GUID,
1583 };
1584
1585 static const struct hv_vmbus_device_id id_table[] = {
1586         /* SCSI guid */
1587         { HV_SCSI_GUID,
1588           .driver_data = SCSI_GUID
1589         },
1590         /* IDE guid */
1591         { HV_IDE_GUID,
1592           .driver_data = IDE_GUID
1593         },
1594         { },
1595 };
1596
1597 MODULE_DEVICE_TABLE(vmbus, id_table);
1598
1599 static int storvsc_probe(struct hv_device *device,
1600                         const struct hv_vmbus_device_id *dev_id)
1601 {
1602         int ret;
1603         struct Scsi_Host *host;
1604         struct hv_host_device *host_dev;
1605         bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1606         int target = 0;
1607         struct storvsc_device *stor_device;
1608
1609         /*
1610          * Based on the windows host we are running on,
1611          * set state to properly communicate with the host.
1612          */
1613
1614         if (vmbus_proto_version == VERSION_WIN8) {
1615                 sense_buffer_size = POST_WIN7_STORVSC_SENSE_BUFFER_SIZE;
1616                 vmscsi_size_delta = 0;
1617                 vmstor_current_major = VMSTOR_WIN8_MAJOR;
1618                 vmstor_current_minor = VMSTOR_WIN8_MINOR;
1619         } else {
1620                 sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE;
1621                 vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
1622                 vmstor_current_major = VMSTOR_WIN7_MAJOR;
1623                 vmstor_current_minor = VMSTOR_WIN7_MINOR;
1624         }
1625
1626
1627         host = scsi_host_alloc(&scsi_driver,
1628                                sizeof(struct hv_host_device));
1629         if (!host)
1630                 return -ENOMEM;
1631
1632         host_dev = shost_priv(host);
1633         memset(host_dev, 0, sizeof(struct hv_host_device));
1634
1635         host_dev->port = host->host_no;
1636         host_dev->dev = device;
1637
1638
1639         stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1640         if (!stor_device) {
1641                 ret = -ENOMEM;
1642                 goto err_out0;
1643         }
1644
1645         stor_device->destroy = false;
1646         init_waitqueue_head(&stor_device->waiting_to_drain);
1647         stor_device->device = device;
1648         stor_device->host = host;
1649         hv_set_drvdata(device, stor_device);
1650
1651         stor_device->port_number = host->host_no;
1652         ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1653         if (ret)
1654                 goto err_out1;
1655
1656         host_dev->path = stor_device->path_id;
1657         host_dev->target = stor_device->target_id;
1658
1659         /* max # of devices per target */
1660         host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1661         /* max # of targets per channel */
1662         host->max_id = STORVSC_MAX_TARGETS;
1663         /* max # of channels */
1664         host->max_channel = STORVSC_MAX_CHANNELS - 1;
1665         /* max cmd length */
1666         host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1667
1668         /* Register the HBA and start the scsi bus scan */
1669         ret = scsi_add_host(host, &device->device);
1670         if (ret != 0)
1671                 goto err_out2;
1672
1673         if (!dev_is_ide) {
1674                 scsi_scan_host(host);
1675         } else {
1676                 target = (device->dev_instance.b[5] << 8 |
1677                          device->dev_instance.b[4]);
1678                 ret = scsi_add_device(host, 0, target, 0);
1679                 if (ret) {
1680                         scsi_remove_host(host);
1681                         goto err_out2;
1682                 }
1683         }
1684         return 0;
1685
1686 err_out2:
1687         /*
1688          * Once we have connected with the host, we would need to
1689          * to invoke storvsc_dev_remove() to rollback this state and
1690          * this call also frees up the stor_device; hence the jump around
1691          * err_out1 label.
1692          */
1693         storvsc_dev_remove(device);
1694         goto err_out0;
1695
1696 err_out1:
1697         kfree(stor_device);
1698
1699 err_out0:
1700         scsi_host_put(host);
1701         return ret;
1702 }
1703
1704 static int storvsc_remove(struct hv_device *dev)
1705 {
1706         struct storvsc_device *stor_device = hv_get_drvdata(dev);
1707         struct Scsi_Host *host = stor_device->host;
1708
1709         scsi_remove_host(host);
1710         storvsc_dev_remove(dev);
1711         scsi_host_put(host);
1712
1713         return 0;
1714 }
1715
1716 static struct hv_driver storvsc_drv = {
1717         .name = KBUILD_MODNAME,
1718         .id_table = id_table,
1719         .probe = storvsc_probe,
1720         .remove = storvsc_remove,
1721 };
1722
1723 static int __init storvsc_drv_init(void)
1724 {
1725         u32 max_outstanding_req_per_channel;
1726
1727         /*
1728          * Divide the ring buffer data size (which is 1 page less
1729          * than the ring buffer size since that page is reserved for
1730          * the ring buffer indices) by the max request size (which is
1731          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1732          */
1733         max_outstanding_req_per_channel =
1734                 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1735                 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1736                 sizeof(struct vstor_packet) + sizeof(u64) -
1737                 vmscsi_size_delta,
1738                 sizeof(u64)));
1739
1740         if (max_outstanding_req_per_channel <
1741             STORVSC_MAX_IO_REQUESTS)
1742                 return -EINVAL;
1743
1744         return vmbus_driver_register(&storvsc_drv);
1745 }
1746
1747 static void __exit storvsc_drv_exit(void)
1748 {
1749         vmbus_driver_unregister(&storvsc_drv);
1750 }
1751
1752 MODULE_LICENSE("GPL");
1753 MODULE_VERSION(HV_DRV_VERSION);
1754 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1755 module_init(storvsc_drv_init);
1756 module_exit(storvsc_drv_exit);