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