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_ERROR        0x04
 205
 206 /*
 207  * This is the end of Protocol specific defines.
 208  */
 209
 210
 211 /*
 212  * We setup a mempool to allocate request structures for this driver
 213  * on a per-lun basis. The following define specifies the number of
 214  * elements in the pool.
 215  */
 216
 217 #define STORVSC_MIN_BUF_NR                              64
 218 static int storvsc_ringbuffer_size = (20 * PAGE_SIZE);
 219
 220 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
 221 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
 222
 223 #define STORVSC_MAX_IO_REQUESTS                         128
 224
 225 /*
 226  * In Hyper-V, each port/path/target maps to 1 scsi host adapter.  In
 227  * reality, the path/target is not used (ie always set to 0) so our
 228  * scsi host adapter essentially has 1 bus with 1 target that contains
 229  * up to 256 luns.
 230  */
 231 #define STORVSC_MAX_LUNS_PER_TARGET                     64
 232 #define STORVSC_MAX_TARGETS                             1
 233 #define STORVSC_MAX_CHANNELS                            1
 234
 235
 236
 237 struct storvsc_cmd_request {
 238         struct list_head entry;
 239         struct scsi_cmnd *cmd;
 240
 241         unsigned int bounce_sgl_count;
 242         struct scatterlist *bounce_sgl;
 243
 244         struct hv_device *device;
 245
 246         /* Synchronize the request/response if needed */
 247         struct completion wait_event;
 248
 249         unsigned char *sense_buffer;
 250         struct hv_multipage_buffer data_buffer;
 251         struct vstor_packet vstor_packet;
 252 };
 253
 254
 255 /* A storvsc device is a device object that contains a vmbus channel */
 256 struct storvsc_device {
 257         struct hv_device *device;
 258
 259         bool     destroy;
 260         bool     drain_notify;
 261         atomic_t num_outstanding_req;
 262         struct Scsi_Host *host;
 263
 264         wait_queue_head_t waiting_to_drain;
 265
 266         /*
 267          * Each unique Port/Path/Target represents 1 channel ie scsi
 268          * controller. In reality, the pathid, targetid is always 0
 269          * and the port is set by us
 270          */
 271         unsigned int port_number;
 272         unsigned char path_id;
 273         unsigned char target_id;
 274
 275         /* Used for vsc/vsp channel reset process */
 276         struct storvsc_cmd_request init_request;
 277         struct storvsc_cmd_request reset_request;
 278 };
 279
 280 struct stor_mem_pools {
 281         struct kmem_cache *request_pool;
 282         mempool_t *request_mempool;
 283 };
 284
 285 struct hv_host_device {
 286         struct hv_device *dev;
 287         unsigned int port;
 288         unsigned char path;
 289         unsigned char target;
 290 };
 291
 292 struct storvsc_scan_work {
 293         struct work_struct work;
 294         struct Scsi_Host *host;
 295         uint lun;
 296 };
 297
 298 static void storvsc_bus_scan(struct work_struct *work)
 299 {
 300         struct storvsc_scan_work *wrk;
 301         int id, order_id;
 302
 303         wrk = container_of(work, struct storvsc_scan_work, work);
 304         for (id = 0; id < wrk->host->max_id; ++id) {
 305                 if (wrk->host->reverse_ordering)
 306                         order_id = wrk->host->max_id - id - 1;
 307                 else
 308                         order_id = id;
 309
 310                 scsi_scan_target(&wrk->host->shost_gendev, 0,
 311                                 order_id, SCAN_WILD_CARD, 1);
 312         }
 313         kfree(wrk);
 314 }
 315
 316 static void storvsc_remove_lun(struct work_struct *work)
 317 {
 318         struct storvsc_scan_work *wrk;
 319         struct scsi_device *sdev;
 320
 321         wrk = container_of(work, struct storvsc_scan_work, work);
 322         if (!scsi_host_get(wrk->host))
 323                 goto done;
 324
 325         sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun);
 326
 327         if (sdev) {
 328                 scsi_remove_device(sdev);
 329                 scsi_device_put(sdev);
 330         }
 331         scsi_host_put(wrk->host);
 332
 333 done:
 334         kfree(wrk);
 335 }
 336
 337 /*
 338  * Major/minor macros.  Minor version is in LSB, meaning that earlier flat
 339  * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1).
 340  */
 341
 342 static inline u16 storvsc_get_version(u8 major, u8 minor)
 343 {
 344         u16 version;
 345
 346         version = ((major << 8) | minor);
 347         return version;
 348 }
 349
 350 /*
 351  * We can get incoming messages from the host that are not in response to
 352  * messages that we have sent out. An example of this would be messages
 353  * received by the guest to notify dynamic addition/removal of LUNs. To
 354  * deal with potential race conditions where the driver may be in the
 355  * midst of being unloaded when we might receive an unsolicited message
 356  * from the host, we have implemented a mechanism to gurantee sequential
 357  * consistency:
 358  *
 359  * 1) Once the device is marked as being destroyed, we will fail all
 360  *    outgoing messages.
 361  * 2) We permit incoming messages when the device is being destroyed,
 362  *    only to properly account for messages already sent out.
 363  */
 364
 365 static inline struct storvsc_device *get_out_stor_device(
 366                                         struct hv_device *device)
 367 {
 368         struct storvsc_device *stor_device;
 369
 370         stor_device = hv_get_drvdata(device);
 371
 372         if (stor_device && stor_device->destroy)
 373                 stor_device = NULL;
 374
 375         return stor_device;
 376 }
 377
 378
 379 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
 380 {
 381         dev->drain_notify = true;
 382         wait_event(dev->waiting_to_drain,
 383                    atomic_read(&dev->num_outstanding_req) == 0);
 384         dev->drain_notify = false;
 385 }
 386
 387 static inline struct storvsc_device *get_in_stor_device(
 388                                         struct hv_device *device)
 389 {
 390         struct storvsc_device *stor_device;
 391
 392         stor_device = hv_get_drvdata(device);
 393
 394         if (!stor_device)
 395                 goto get_in_err;
 396
 397         /*
 398          * If the device is being destroyed; allow incoming
 399          * traffic only to cleanup outstanding requests.
 400          */
 401
 402         if (stor_device->destroy  &&
 403                 (atomic_read(&stor_device->num_outstanding_req) == 0))
 404                 stor_device = NULL;
 405
 406 get_in_err:
 407         return stor_device;
 408
 409 }
 410
 411 static void destroy_bounce_buffer(struct scatterlist *sgl,
 412                                   unsigned int sg_count)
 413 {
 414         int i;
 415         struct page *page_buf;
 416
 417         for (i = 0; i < sg_count; i++) {
 418                 page_buf = sg_page((&sgl[i]));
 419                 if (page_buf != NULL)
 420                         __free_page(page_buf);
 421         }
 422
 423         kfree(sgl);
 424 }
 425
 426 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
 427 {
 428         int i;
 429
 430         /* No need to check */
 431         if (sg_count < 2)
 432                 return -1;
 433
 434         /* We have at least 2 sg entries */
 435         for (i = 0; i < sg_count; i++) {
 436                 if (i == 0) {
 437                         /* make sure 1st one does not have hole */
 438                         if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
 439                                 return i;
 440                 } else if (i == sg_count - 1) {
 441                         /* make sure last one does not have hole */
 442                         if (sgl[i].offset != 0)
 443                                 return i;
 444                 } else {
 445                         /* make sure no hole in the middle */
 446                         if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
 447                                 return i;
 448                 }
 449         }
 450         return -1;
 451 }
 452
 453 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
 454                                                 unsigned int sg_count,
 455                                                 unsigned int len,
 456                                                 int write)
 457 {
 458         int i;
 459         int num_pages;
 460         struct scatterlist *bounce_sgl;
 461         struct page *page_buf;
 462         unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
 463
 464         num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
 465
 466         bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
 467         if (!bounce_sgl)
 468                 return NULL;
 469
 470         sg_init_table(bounce_sgl, num_pages);
 471         for (i = 0; i < num_pages; i++) {
 472                 page_buf = alloc_page(GFP_ATOMIC);
 473                 if (!page_buf)
 474                         goto cleanup;
 475                 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0);
 476         }
 477
 478         return bounce_sgl;
 479
 480 cleanup:
 481         destroy_bounce_buffer(bounce_sgl, num_pages);
 482         return NULL;
 483 }
 484
 485 /* Disgusting wrapper functions */
 486 static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx)
 487 {
 488         void *addr = kmap_atomic(sg_page(sgl + idx));
 489         return (unsigned long)addr;
 490 }
 491
 492 static inline void sg_kunmap_atomic(unsigned long addr)
 493 {
 494         kunmap_atomic((void *)addr);
 495 }
 496
 497
 498 /* Assume the original sgl has enough room */
 499 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
 500                                             struct scatterlist *bounce_sgl,
 501                                             unsigned int orig_sgl_count,
 502                                             unsigned int bounce_sgl_count)
 503 {
 504         int i;
 505         int j = 0;
 506         unsigned long src, dest;
 507         unsigned int srclen, destlen, copylen;
 508         unsigned int total_copied = 0;
 509         unsigned long bounce_addr = 0;
 510         unsigned long dest_addr = 0;
 511         unsigned long flags;
 512
 513         local_irq_save(flags);
 514
 515         for (i = 0; i < orig_sgl_count; i++) {
 516                 dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
 517                 dest = dest_addr;
 518                 destlen = orig_sgl[i].length;
 519
 520                 if (bounce_addr == 0)
 521                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 522
 523                 while (destlen) {
 524                         src = bounce_addr + bounce_sgl[j].offset;
 525                         srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
 526
 527                         copylen = min(srclen, destlen);
 528                         memcpy((void *)dest, (void *)src, copylen);
 529
 530                         total_copied += copylen;
 531                         bounce_sgl[j].offset += copylen;
 532                         destlen -= copylen;
 533                         dest += copylen;
 534
 535                         if (bounce_sgl[j].offset == bounce_sgl[j].length) {
 536                                 /* full */
 537                                 sg_kunmap_atomic(bounce_addr);
 538                                 j++;
 539
 540                                 /*
 541                                  * It is possible that the number of elements
 542                                  * in the bounce buffer may not be equal to
 543                                  * the number of elements in the original
 544                                  * scatter list. Handle this correctly.
 545                                  */
 546
 547                                 if (j == bounce_sgl_count) {
 548                                         /*
 549                                          * We are done; cleanup and return.
 550                                          */
 551                                         sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
 552                                         local_irq_restore(flags);
 553                                         return total_copied;
 554                                 }
 555
 556                                 /* if we need to use another bounce buffer */
 557                                 if (destlen || i != orig_sgl_count - 1)
 558                                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 559                         } else if (destlen == 0 && i == orig_sgl_count - 1) {
 560                                 /* unmap the last bounce that is < PAGE_SIZE */
 561                                 sg_kunmap_atomic(bounce_addr);
 562                         }
 563                 }
 564
 565                 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
 566         }
 567
 568         local_irq_restore(flags);
 569
 570         return total_copied;
 571 }
 572
 573 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
 574 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
 575                                           struct scatterlist *bounce_sgl,
 576                                           unsigned int orig_sgl_count)
 577 {
 578         int i;
 579         int j = 0;
 580         unsigned long src, dest;
 581         unsigned int srclen, destlen, copylen;
 582         unsigned int total_copied = 0;
 583         unsigned long bounce_addr = 0;
 584         unsigned long src_addr = 0;
 585         unsigned long flags;
 586
 587         local_irq_save(flags);
 588
 589         for (i = 0; i < orig_sgl_count; i++) {
 590                 src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
 591                 src = src_addr;
 592                 srclen = orig_sgl[i].length;
 593
 594                 if (bounce_addr == 0)
 595                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 596
 597                 while (srclen) {
 598                         /* assume bounce offset always == 0 */
 599                         dest = bounce_addr + bounce_sgl[j].length;
 600                         destlen = PAGE_SIZE - bounce_sgl[j].length;
 601
 602                         copylen = min(srclen, destlen);
 603                         memcpy((void *)dest, (void *)src, copylen);
 604
 605                         total_copied += copylen;
 606                         bounce_sgl[j].length += copylen;
 607                         srclen -= copylen;
 608                         src += copylen;
 609
 610                         if (bounce_sgl[j].length == PAGE_SIZE) {
 611                                 /* full..move to next entry */
 612                                 sg_kunmap_atomic(bounce_addr);
 613                                 j++;
 614
 615                                 /* if we need to use another bounce buffer */
 616                                 if (srclen || i != orig_sgl_count - 1)
 617                                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 618
 619                         } else if (srclen == 0 && i == orig_sgl_count - 1) {
 620                                 /* unmap the last bounce that is < PAGE_SIZE */
 621                                 sg_kunmap_atomic(bounce_addr);
 622                         }
 623                 }
 624
 625                 sg_kunmap_atomic(src_addr - orig_sgl[i].offset);
 626         }
 627
 628         local_irq_restore(flags);
 629
 630         return total_copied;
 631 }
 632
 633 static int storvsc_channel_init(struct hv_device *device)
 634 {
 635         struct storvsc_device *stor_device;
 636         struct storvsc_cmd_request *request;
 637         struct vstor_packet *vstor_packet;
 638         int ret, t;
 639
 640         stor_device = get_out_stor_device(device);
 641         if (!stor_device)
 642                 return -ENODEV;
 643
 644         request = &stor_device->init_request;
 645         vstor_packet = &request->vstor_packet;
 646
 647         /*
 648          * Now, initiate the vsc/vsp initialization protocol on the open
 649          * channel
 650          */
 651         memset(request, 0, sizeof(struct storvsc_cmd_request));
 652         init_completion(&request->wait_event);
 653         vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
 654         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 655
 656         ret = vmbus_sendpacket(device->channel, vstor_packet,
 657                                sizeof(struct vstor_packet),
 658                                (unsigned long)request,
 659                                VM_PKT_DATA_INBAND,
 660                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 661         if (ret != 0)
 662                 goto cleanup;
 663
 664         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 665         if (t == 0) {
 666                 ret = -ETIMEDOUT;
 667                 goto cleanup;
 668         }
 669
 670         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 671             vstor_packet->status != 0)
 672                 goto cleanup;
 673
 674
 675         /* reuse the packet for version range supported */
 676         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 677         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
 678         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 679
 680         vstor_packet->version.major_minor =
 681                 storvsc_get_version(VMSTOR_CURRENT_MAJOR, VMSTOR_CURRENT_MINOR);
 682
 683         /*
 684          * The revision number is only used in Windows; set it to 0.
 685          */
 686         vstor_packet->version.revision = 0;
 687
 688         ret = vmbus_sendpacket(device->channel, vstor_packet,
 689                                sizeof(struct vstor_packet),
 690                                (unsigned long)request,
 691                                VM_PKT_DATA_INBAND,
 692                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 693         if (ret != 0)
 694                 goto cleanup;
 695
 696         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 697         if (t == 0) {
 698                 ret = -ETIMEDOUT;
 699                 goto cleanup;
 700         }
 701
 702         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 703             vstor_packet->status != 0)
 704                 goto cleanup;
 705
 706
 707         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 708         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
 709         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 710         vstor_packet->storage_channel_properties.port_number =
 711                                         stor_device->port_number;
 712
 713         ret = vmbus_sendpacket(device->channel, vstor_packet,
 714                                sizeof(struct vstor_packet),
 715                                (unsigned long)request,
 716                                VM_PKT_DATA_INBAND,
 717                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 718
 719         if (ret != 0)
 720                 goto cleanup;
 721
 722         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 723         if (t == 0) {
 724                 ret = -ETIMEDOUT;
 725                 goto cleanup;
 726         }
 727
 728         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 729             vstor_packet->status != 0)
 730                 goto cleanup;
 731
 732         stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
 733         stor_device->target_id
 734                 = vstor_packet->storage_channel_properties.target_id;
 735
 736         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 737         vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
 738         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 739
 740         ret = vmbus_sendpacket(device->channel, vstor_packet,
 741                                sizeof(struct vstor_packet),
 742                                (unsigned long)request,
 743                                VM_PKT_DATA_INBAND,
 744                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 745
 746         if (ret != 0)
 747                 goto cleanup;
 748
 749         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 750         if (t == 0) {
 751                 ret = -ETIMEDOUT;
 752                 goto cleanup;
 753         }
 754
 755         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 756             vstor_packet->status != 0)
 757                 goto cleanup;
 758
 759
 760 cleanup:
 761         return ret;
 762 }
 763
 764
 765 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
 766 {
 767         struct scsi_cmnd *scmnd = cmd_request->cmd;
 768         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
 769         void (*scsi_done_fn)(struct scsi_cmnd *);
 770         struct scsi_sense_hdr sense_hdr;
 771         struct vmscsi_request *vm_srb;
 772         struct storvsc_scan_work *wrk;
 773         struct stor_mem_pools *memp = scmnd->device->hostdata;
 774
 775         vm_srb = &cmd_request->vstor_packet.vm_srb;
 776         if (cmd_request->bounce_sgl_count) {
 777                 if (vm_srb->data_in == READ_TYPE)
 778                         copy_from_bounce_buffer(scsi_sglist(scmnd),
 779                                         cmd_request->bounce_sgl,
 780                                         scsi_sg_count(scmnd),
 781                                         cmd_request->bounce_sgl_count);
 782                 destroy_bounce_buffer(cmd_request->bounce_sgl,
 783                                         cmd_request->bounce_sgl_count);
 784         }
 785
 786         /*
 787          * If there is an error; offline the device since all
 788          * error recovery strategies would have already been
 789          * deployed on the host side. However, if the command
 790          * were a pass-through command deal with it appropriately.
 791          */
 792         scmnd->result = vm_srb->scsi_status;
 793
 794         if (vm_srb->srb_status == SRB_STATUS_ERROR) {
 795                 switch (scmnd->cmnd[0]) {
 796                 case ATA_16:
 797                 case ATA_12:
 798                         set_host_byte(scmnd, DID_PASSTHROUGH);
 799                         break;
 800                 default:
 801                         set_host_byte(scmnd, DID_TARGET_FAILURE);
 802                 }
 803         }
 804
 805
 806         /*
 807          * If the LUN is invalid; remove the device.
 808          */
 809         if (vm_srb->srb_status == SRB_STATUS_INVALID_LUN) {
 810                 struct storvsc_device *stor_dev;
 811                 struct hv_device *dev = host_dev->dev;
 812                 struct Scsi_Host *host;
 813
 814                 stor_dev = get_in_stor_device(dev);
 815                 host = stor_dev->host;
 816
 817                 wrk = kmalloc(sizeof(struct storvsc_scan_work),
 818                                 GFP_ATOMIC);
 819                 if (!wrk) {
 820                         scmnd->result = DID_TARGET_FAILURE << 16;
 821                 } else {
 822                         wrk->host = host;
 823                         wrk->lun = vm_srb->lun;
 824                         INIT_WORK(&wrk->work, storvsc_remove_lun);
 825                         schedule_work(&wrk->work);
 826                 }
 827         }
 828
 829         if (scmnd->result) {
 830                 if (scsi_normalize_sense(scmnd->sense_buffer,
 831                                 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
 832                         scsi_print_sense_hdr("storvsc", &sense_hdr);
 833         }
 834
 835         scsi_set_resid(scmnd,
 836                 cmd_request->data_buffer.len -
 837                 vm_srb->data_transfer_length);
 838
 839         scsi_done_fn = scmnd->scsi_done;
 840
 841         scmnd->host_scribble = NULL;
 842         scmnd->scsi_done = NULL;
 843
 844         scsi_done_fn(scmnd);
 845
 846         mempool_free(cmd_request, memp->request_mempool);
 847 }
 848
 849 static void storvsc_on_io_completion(struct hv_device *device,
 850                                   struct vstor_packet *vstor_packet,
 851                                   struct storvsc_cmd_request *request)
 852 {
 853         struct storvsc_device *stor_device;
 854         struct vstor_packet *stor_pkt;
 855
 856         stor_device = hv_get_drvdata(device);
 857         stor_pkt = &request->vstor_packet;
 858
 859         /*
 860          * The current SCSI handling on the host side does
 861          * not correctly handle:
 862          * INQUIRY command with page code parameter set to 0x80
 863          * MODE_SENSE command with cmd[2] == 0x1c
 864          *
 865          * Setup srb and scsi status so this won't be fatal.
 866          * We do this so we can distinguish truly fatal failues
 867          * (srb status == 0x4) and off-line the device in that case.
 868          */
 869
 870         if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
 871            (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
 872                 vstor_packet->vm_srb.scsi_status = 0;
 873                 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
 874         }
 875
 876
 877         /* Copy over the status...etc */
 878         stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
 879         stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
 880         stor_pkt->vm_srb.sense_info_length =
 881         vstor_packet->vm_srb.sense_info_length;
 882
 883         if (vstor_packet->vm_srb.scsi_status != 0 ||
 884                 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){
 885                 dev_warn(&device->device,
 886                          "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
 887                          stor_pkt->vm_srb.cdb[0],
 888                          vstor_packet->vm_srb.scsi_status,
 889                          vstor_packet->vm_srb.srb_status);
 890         }
 891
 892         if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
 893                 /* CHECK_CONDITION */
 894                 if (vstor_packet->vm_srb.srb_status &
 895                         SRB_STATUS_AUTOSENSE_VALID) {
 896                         /* autosense data available */
 897                         dev_warn(&device->device,
 898                                  "stor pkt %p autosense data valid - len %d\n",
 899                                  request,
 900                                  vstor_packet->vm_srb.sense_info_length);
 901
 902                         memcpy(request->sense_buffer,
 903                                vstor_packet->vm_srb.sense_data,
 904                                vstor_packet->vm_srb.sense_info_length);
 905
 906                 }
 907         }
 908
 909         stor_pkt->vm_srb.data_transfer_length =
 910         vstor_packet->vm_srb.data_transfer_length;
 911
 912         storvsc_command_completion(request);
 913
 914         if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
 915                 stor_device->drain_notify)
 916                 wake_up(&stor_device->waiting_to_drain);
 917
 918
 919 }
 920
 921 static void storvsc_on_receive(struct hv_device *device,
 922                              struct vstor_packet *vstor_packet,
 923                              struct storvsc_cmd_request *request)
 924 {
 925         struct storvsc_scan_work *work;
 926         struct storvsc_device *stor_device;
 927
 928         switch (vstor_packet->operation) {
 929         case VSTOR_OPERATION_COMPLETE_IO:
 930                 storvsc_on_io_completion(device, vstor_packet, request);
 931                 break;
 932
 933         case VSTOR_OPERATION_REMOVE_DEVICE:
 934         case VSTOR_OPERATION_ENUMERATE_BUS:
 935                 stor_device = get_in_stor_device(device);
 936                 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
 937                 if (!work)
 938                         return;
 939
 940                 INIT_WORK(&work->work, storvsc_bus_scan);
 941                 work->host = stor_device->host;
 942                 schedule_work(&work->work);
 943                 break;
 944
 945         default:
 946                 break;
 947         }
 948 }
 949
 950 static void storvsc_on_channel_callback(void *context)
 951 {
 952         struct hv_device *device = (struct hv_device *)context;
 953         struct storvsc_device *stor_device;
 954         u32 bytes_recvd;
 955         u64 request_id;
 956         unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
 957         struct storvsc_cmd_request *request;
 958         int ret;
 959
 960
 961         stor_device = get_in_stor_device(device);
 962         if (!stor_device)
 963                 return;
 964
 965         do {
 966                 ret = vmbus_recvpacket(device->channel, packet,
 967                                        ALIGN(sizeof(struct vstor_packet), 8),
 968                                        &bytes_recvd, &request_id);
 969                 if (ret == 0 && bytes_recvd > 0) {
 970
 971                         request = (struct storvsc_cmd_request *)
 972                                         (unsigned long)request_id;
 973
 974                         if ((request == &stor_device->init_request) ||
 975                             (request == &stor_device->reset_request)) {
 976
 977                                 memcpy(&request->vstor_packet, packet,
 978                                        sizeof(struct vstor_packet));
 979                                 complete(&request->wait_event);
 980                         } else {
 981                                 storvsc_on_receive(device,
 982                                                 (struct vstor_packet *)packet,
 983                                                 request);
 984                         }
 985                 } else {
 986                         break;
 987                 }
 988         } while (1);
 989
 990         return;
 991 }
 992
 993 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
 994 {
 995         struct vmstorage_channel_properties props;
 996         int ret;
 997
 998         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
 999
1000         ret = vmbus_open(device->channel,
1001                          ring_size,
1002                          ring_size,
1003                          (void *)&props,
1004                          sizeof(struct vmstorage_channel_properties),
1005                          storvsc_on_channel_callback, device);
1006
1007         if (ret != 0)
1008                 return ret;
1009
1010         ret = storvsc_channel_init(device);
1011
1012         return ret;
1013 }
1014
1015 static int storvsc_dev_remove(struct hv_device *device)
1016 {
1017         struct storvsc_device *stor_device;
1018         unsigned long flags;
1019
1020         stor_device = hv_get_drvdata(device);
1021
1022         spin_lock_irqsave(&device->channel->inbound_lock, flags);
1023         stor_device->destroy = true;
1024         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1025
1026         /*
1027          * At this point, all outbound traffic should be disable. We
1028          * only allow inbound traffic (responses) to proceed so that
1029          * outstanding requests can be completed.
1030          */
1031
1032         storvsc_wait_to_drain(stor_device);
1033
1034         /*
1035          * Since we have already drained, we don't need to busy wait
1036          * as was done in final_release_stor_device()
1037          * Note that we cannot set the ext pointer to NULL until
1038          * we have drained - to drain the outgoing packets, we need to
1039          * allow incoming packets.
1040          */
1041         spin_lock_irqsave(&device->channel->inbound_lock, flags);
1042         hv_set_drvdata(device, NULL);
1043         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1044
1045         /* Close the channel */
1046         vmbus_close(device->channel);
1047
1048         kfree(stor_device);
1049         return 0;
1050 }
1051
1052 static int storvsc_do_io(struct hv_device *device,
1053                               struct storvsc_cmd_request *request)
1054 {
1055         struct storvsc_device *stor_device;
1056         struct vstor_packet *vstor_packet;
1057         int ret = 0;
1058
1059         vstor_packet = &request->vstor_packet;
1060         stor_device = get_out_stor_device(device);
1061
1062         if (!stor_device)
1063                 return -ENODEV;
1064
1065
1066         request->device  = device;
1067
1068
1069         vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1070
1071         vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1072
1073
1074         vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1075
1076
1077         vstor_packet->vm_srb.data_transfer_length =
1078         request->data_buffer.len;
1079
1080         vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1081
1082         if (request->data_buffer.len) {
1083                 ret = vmbus_sendpacket_multipagebuffer(device->channel,
1084                                 &request->data_buffer,
1085                                 vstor_packet,
1086                                 sizeof(struct vstor_packet),
1087                                 (unsigned long)request);
1088         } else {
1089                 ret = vmbus_sendpacket(device->channel, vstor_packet,
1090                                sizeof(struct vstor_packet),
1091                                (unsigned long)request,
1092                                VM_PKT_DATA_INBAND,
1093                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1094         }
1095
1096         if (ret != 0)
1097                 return ret;
1098
1099         atomic_inc(&stor_device->num_outstanding_req);
1100
1101         return ret;
1102 }
1103
1104 static int storvsc_device_alloc(struct scsi_device *sdevice)
1105 {
1106         struct stor_mem_pools *memp;
1107         int number = STORVSC_MIN_BUF_NR;
1108
1109         memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
1110         if (!memp)
1111                 return -ENOMEM;
1112
1113         memp->request_pool =
1114                 kmem_cache_create(dev_name(&sdevice->sdev_dev),
1115                                 sizeof(struct storvsc_cmd_request), 0,
1116                                 SLAB_HWCACHE_ALIGN, NULL);
1117
1118         if (!memp->request_pool)
1119                 goto err0;
1120
1121         memp->request_mempool = mempool_create(number, mempool_alloc_slab,
1122                                                 mempool_free_slab,
1123                                                 memp->request_pool);
1124
1125         if (!memp->request_mempool)
1126                 goto err1;
1127
1128         sdevice->hostdata = memp;
1129
1130         return 0;
1131
1132 err1:
1133         kmem_cache_destroy(memp->request_pool);
1134
1135 err0:
1136         kfree(memp);
1137         return -ENOMEM;
1138 }
1139
1140 static void storvsc_device_destroy(struct scsi_device *sdevice)
1141 {
1142         struct stor_mem_pools *memp = sdevice->hostdata;
1143
1144         mempool_destroy(memp->request_mempool);
1145         kmem_cache_destroy(memp->request_pool);
1146         kfree(memp);
1147         sdevice->hostdata = NULL;
1148 }
1149
1150 static int storvsc_device_configure(struct scsi_device *sdevice)
1151 {
1152         scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
1153                                 STORVSC_MAX_IO_REQUESTS);
1154
1155         blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
1156
1157         blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
1158
1159         return 0;
1160 }
1161
1162 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1163                            sector_t capacity, int *info)
1164 {
1165         sector_t nsect = capacity;
1166         sector_t cylinders = nsect;
1167         int heads, sectors_pt;
1168
1169         /*
1170          * We are making up these values; let us keep it simple.
1171          */
1172         heads = 0xff;
1173         sectors_pt = 0x3f;      /* Sectors per track */
1174         sector_div(cylinders, heads * sectors_pt);
1175         if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1176                 cylinders = 0xffff;
1177
1178         info[0] = heads;
1179         info[1] = sectors_pt;
1180         info[2] = (int)cylinders;
1181
1182         return 0;
1183 }
1184
1185 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1186 {
1187         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1188         struct hv_device *device = host_dev->dev;
1189
1190         struct storvsc_device *stor_device;
1191         struct storvsc_cmd_request *request;
1192         struct vstor_packet *vstor_packet;
1193         int ret, t;
1194
1195
1196         stor_device = get_out_stor_device(device);
1197         if (!stor_device)
1198                 return FAILED;
1199
1200         request = &stor_device->reset_request;
1201         vstor_packet = &request->vstor_packet;
1202
1203         init_completion(&request->wait_event);
1204
1205         vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1206         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1207         vstor_packet->vm_srb.path_id = stor_device->path_id;
1208
1209         ret = vmbus_sendpacket(device->channel, vstor_packet,
1210                                sizeof(struct vstor_packet),
1211                                (unsigned long)&stor_device->reset_request,
1212                                VM_PKT_DATA_INBAND,
1213                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1214         if (ret != 0)
1215                 return FAILED;
1216
1217         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1218         if (t == 0)
1219                 return TIMEOUT_ERROR;
1220
1221
1222         /*
1223          * At this point, all outstanding requests in the adapter
1224          * should have been flushed out and return to us
1225          * There is a potential race here where the host may be in
1226          * the process of responding when we return from here.
1227          * Just wait for all in-transit packets to be accounted for
1228          * before we return from here.
1229          */
1230         storvsc_wait_to_drain(stor_device);
1231
1232         return SUCCESS;
1233 }
1234
1235 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1236 {
1237         bool allowed = true;
1238         u8 scsi_op = scmnd->cmnd[0];
1239
1240         switch (scsi_op) {
1241         /*
1242          * smartd sends this command and the host does not handle
1243          * this. So, don't send it.
1244          */
1245         case SET_WINDOW:
1246                 scmnd->result = ILLEGAL_REQUEST << 16;
1247                 allowed = false;
1248                 break;
1249         default:
1250                 break;
1251         }
1252         return allowed;
1253 }
1254
1255 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1256 {
1257         int ret;
1258         struct hv_host_device *host_dev = shost_priv(host);
1259         struct hv_device *dev = host_dev->dev;
1260         struct storvsc_cmd_request *cmd_request;
1261         unsigned int request_size = 0;
1262         int i;
1263         struct scatterlist *sgl;
1264         unsigned int sg_count = 0;
1265         struct vmscsi_request *vm_srb;
1266         struct stor_mem_pools *memp = scmnd->device->hostdata;
1267
1268         if (!storvsc_scsi_cmd_ok(scmnd)) {
1269                 scmnd->scsi_done(scmnd);
1270                 return 0;
1271         }
1272
1273         request_size = sizeof(struct storvsc_cmd_request);
1274
1275         cmd_request = mempool_alloc(memp->request_mempool,
1276                                        GFP_ATOMIC);
1277
1278         /*
1279          * We might be invoked in an interrupt context; hence
1280          * mempool_alloc() can fail.
1281          */
1282         if (!cmd_request)
1283                 return SCSI_MLQUEUE_DEVICE_BUSY;
1284
1285         memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1286
1287         /* Setup the cmd request */
1288         cmd_request->cmd = scmnd;
1289
1290         scmnd->host_scribble = (unsigned char *)cmd_request;
1291
1292         vm_srb = &cmd_request->vstor_packet.vm_srb;
1293
1294
1295         /* Build the SRB */
1296         switch (scmnd->sc_data_direction) {
1297         case DMA_TO_DEVICE:
1298                 vm_srb->data_in = WRITE_TYPE;
1299                 break;
1300         case DMA_FROM_DEVICE:
1301                 vm_srb->data_in = READ_TYPE;
1302                 break;
1303         default:
1304                 vm_srb->data_in = UNKNOWN_TYPE;
1305                 break;
1306         }
1307
1308
1309         vm_srb->port_number = host_dev->port;
1310         vm_srb->path_id = scmnd->device->channel;
1311         vm_srb->target_id = scmnd->device->id;
1312         vm_srb->lun = scmnd->device->lun;
1313
1314         vm_srb->cdb_length = scmnd->cmd_len;
1315
1316         memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1317
1318         cmd_request->sense_buffer = scmnd->sense_buffer;
1319
1320
1321         cmd_request->data_buffer.len = scsi_bufflen(scmnd);
1322         if (scsi_sg_count(scmnd)) {
1323                 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1324                 sg_count = scsi_sg_count(scmnd);
1325
1326                 /* check if we need to bounce the sgl */
1327                 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1328                         cmd_request->bounce_sgl =
1329                                 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1330                                                      scsi_bufflen(scmnd),
1331                                                      vm_srb->data_in);
1332                         if (!cmd_request->bounce_sgl) {
1333                                 ret = SCSI_MLQUEUE_HOST_BUSY;
1334                                 goto queue_error;
1335                         }
1336
1337                         cmd_request->bounce_sgl_count =
1338                                 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1339                                         PAGE_SHIFT;
1340
1341                         if (vm_srb->data_in == WRITE_TYPE)
1342                                 copy_to_bounce_buffer(sgl,
1343                                         cmd_request->bounce_sgl,
1344                                         scsi_sg_count(scmnd));
1345
1346                         sgl = cmd_request->bounce_sgl;
1347                         sg_count = cmd_request->bounce_sgl_count;
1348                 }
1349
1350                 cmd_request->data_buffer.offset = sgl[0].offset;
1351
1352                 for (i = 0; i < sg_count; i++)
1353                         cmd_request->data_buffer.pfn_array[i] =
1354                                 page_to_pfn(sg_page((&sgl[i])));
1355
1356         } else if (scsi_sglist(scmnd)) {
1357                 cmd_request->data_buffer.offset =
1358                         virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1359                 cmd_request->data_buffer.pfn_array[0] =
1360                         virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1361         }
1362
1363         /* Invokes the vsc to start an IO */
1364         ret = storvsc_do_io(dev, cmd_request);
1365
1366         if (ret == -EAGAIN) {
1367                 /* no more space */
1368
1369                 if (cmd_request->bounce_sgl_count) {
1370                         destroy_bounce_buffer(cmd_request->bounce_sgl,
1371                                         cmd_request->bounce_sgl_count);
1372
1373                         ret = SCSI_MLQUEUE_DEVICE_BUSY;
1374                         goto queue_error;
1375                 }
1376         }
1377
1378         return 0;
1379
1380 queue_error:
1381         mempool_free(cmd_request, memp->request_mempool);
1382         scmnd->host_scribble = NULL;
1383         return ret;
1384 }
1385
1386 static struct scsi_host_template scsi_driver = {
1387         .module =               THIS_MODULE,
1388         .name =                 "storvsc_host_t",
1389         .bios_param =           storvsc_get_chs,
1390         .queuecommand =         storvsc_queuecommand,
1391         .eh_host_reset_handler =        storvsc_host_reset_handler,
1392         .slave_alloc =          storvsc_device_alloc,
1393         .slave_destroy =        storvsc_device_destroy,
1394         .slave_configure =      storvsc_device_configure,
1395         .cmd_per_lun =          1,
1396         /* 64 max_queue * 1 target */
1397         .can_queue =            STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1398         .this_id =              -1,
1399         /* no use setting to 0 since ll_blk_rw reset it to 1 */
1400         /* currently 32 */
1401         .sg_tablesize =         MAX_MULTIPAGE_BUFFER_COUNT,
1402         .use_clustering =       DISABLE_CLUSTERING,
1403         /* Make sure we dont get a sg segment crosses a page boundary */
1404         .dma_boundary =         PAGE_SIZE-1,
1405 };
1406
1407 enum {
1408         SCSI_GUID,
1409         IDE_GUID,
1410 };
1411
1412 static const struct hv_vmbus_device_id id_table[] = {
1413         /* SCSI guid */
1414         { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1415                        0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1416           .driver_data = SCSI_GUID },
1417         /* IDE guid */
1418         { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1419                        0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1420           .driver_data = IDE_GUID },
1421         { },
1422 };
1423
1424 MODULE_DEVICE_TABLE(vmbus, id_table);
1425
1426 static int storvsc_probe(struct hv_device *device,
1427                         const struct hv_vmbus_device_id *dev_id)
1428 {
1429         int ret;
1430         struct Scsi_Host *host;
1431         struct hv_host_device *host_dev;
1432         bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1433         int target = 0;
1434         struct storvsc_device *stor_device;
1435
1436         host = scsi_host_alloc(&scsi_driver,
1437                                sizeof(struct hv_host_device));
1438         if (!host)
1439                 return -ENOMEM;
1440
1441         host_dev = shost_priv(host);
1442         memset(host_dev, 0, sizeof(struct hv_host_device));
1443
1444         host_dev->port = host->host_no;
1445         host_dev->dev = device;
1446
1447
1448         stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1449         if (!stor_device) {
1450                 ret = -ENOMEM;
1451                 goto err_out0;
1452         }
1453
1454         stor_device->destroy = false;
1455         init_waitqueue_head(&stor_device->waiting_to_drain);
1456         stor_device->device = device;
1457         stor_device->host = host;
1458         hv_set_drvdata(device, stor_device);
1459
1460         stor_device->port_number = host->host_no;
1461         ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1462         if (ret)
1463                 goto err_out1;
1464
1465         host_dev->path = stor_device->path_id;
1466         host_dev->target = stor_device->target_id;
1467
1468         /* max # of devices per target */
1469         host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1470         /* max # of targets per channel */
1471         host->max_id = STORVSC_MAX_TARGETS;
1472         /* max # of channels */
1473         host->max_channel = STORVSC_MAX_CHANNELS - 1;
1474         /* max cmd length */
1475         host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1476
1477         /* Register the HBA and start the scsi bus scan */
1478         ret = scsi_add_host(host, &device->device);
1479         if (ret != 0)
1480                 goto err_out2;
1481
1482         if (!dev_is_ide) {
1483                 scsi_scan_host(host);
1484         } else {
1485                 target = (device->dev_instance.b[5] << 8 |
1486                          device->dev_instance.b[4]);
1487                 ret = scsi_add_device(host, 0, target, 0);
1488                 if (ret) {
1489                         scsi_remove_host(host);
1490                         goto err_out2;
1491                 }
1492         }
1493         return 0;
1494
1495 err_out2:
1496         /*
1497          * Once we have connected with the host, we would need to
1498          * to invoke storvsc_dev_remove() to rollback this state and
1499          * this call also frees up the stor_device; hence the jump around
1500          * err_out1 label.
1501          */
1502         storvsc_dev_remove(device);
1503         goto err_out0;
1504
1505 err_out1:
1506         kfree(stor_device);
1507
1508 err_out0:
1509         scsi_host_put(host);
1510         return ret;
1511 }
1512
1513 static int storvsc_remove(struct hv_device *dev)
1514 {
1515         struct storvsc_device *stor_device = hv_get_drvdata(dev);
1516         struct Scsi_Host *host = stor_device->host;
1517
1518         scsi_remove_host(host);
1519         storvsc_dev_remove(dev);
1520         scsi_host_put(host);
1521
1522         return 0;
1523 }
1524
1525 static struct hv_driver storvsc_drv = {
1526         .name = KBUILD_MODNAME,
1527         .id_table = id_table,
1528         .probe = storvsc_probe,
1529         .remove = storvsc_remove,
1530 };
1531
1532 static int __init storvsc_drv_init(void)
1533 {
1534         u32 max_outstanding_req_per_channel;
1535
1536         /*
1537          * Divide the ring buffer data size (which is 1 page less
1538          * than the ring buffer size since that page is reserved for
1539          * the ring buffer indices) by the max request size (which is
1540          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1541          */
1542         max_outstanding_req_per_channel =
1543                 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1544                 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1545                 sizeof(struct vstor_packet) + sizeof(u64),
1546                 sizeof(u64)));
1547
1548         if (max_outstanding_req_per_channel <
1549             STORVSC_MAX_IO_REQUESTS)
1550                 return -EINVAL;
1551
1552         return vmbus_driver_register(&storvsc_drv);
1553 }
1554
1555 static void __exit storvsc_drv_exit(void)
1556 {
1557         vmbus_driver_unregister(&storvsc_drv);
1558 }
1559
1560 MODULE_LICENSE("GPL");
1561 MODULE_VERSION(HV_DRV_VERSION);
1562 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1563 module_init(storvsc_drv_init);
1564 module_exit(storvsc_drv_exit);