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         for (i = 0; i < num_pages; i++) {
 471                 page_buf = alloc_page(GFP_ATOMIC);
 472                 if (!page_buf)
 473                         goto cleanup;
 474                 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0);
 475         }
 476
 477         return bounce_sgl;
 478
 479 cleanup:
 480         destroy_bounce_buffer(bounce_sgl, num_pages);
 481         return NULL;
 482 }
 483
 484 /* Assume the original sgl has enough room */
 485 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
 486                                             struct scatterlist *bounce_sgl,
 487                                             unsigned int orig_sgl_count,
 488                                             unsigned int bounce_sgl_count)
 489 {
 490         int i;
 491         int j = 0;
 492         unsigned long src, dest;
 493         unsigned int srclen, destlen, copylen;
 494         unsigned int total_copied = 0;
 495         unsigned long bounce_addr = 0;
 496         unsigned long dest_addr = 0;
 497         unsigned long flags;
 498
 499         local_irq_save(flags);
 500
 501         for (i = 0; i < orig_sgl_count; i++) {
 502                 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
 503                                         KM_IRQ0) + orig_sgl[i].offset;
 504                 dest = dest_addr;
 505                 destlen = orig_sgl[i].length;
 506
 507                 if (bounce_addr == 0)
 508                         bounce_addr =
 509                         (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
 510                                                         KM_IRQ0);
 511
 512                 while (destlen) {
 513                         src = bounce_addr + bounce_sgl[j].offset;
 514                         srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
 515
 516                         copylen = min(srclen, destlen);
 517                         memcpy((void *)dest, (void *)src, copylen);
 518
 519                         total_copied += copylen;
 520                         bounce_sgl[j].offset += copylen;
 521                         destlen -= copylen;
 522                         dest += copylen;
 523
 524                         if (bounce_sgl[j].offset == bounce_sgl[j].length) {
 525                                 /* full */
 526                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
 527                                 j++;
 528
 529                                 /*
 530                                  * It is possible that the number of elements
 531                                  * in the bounce buffer may not be equal to
 532                                  * the number of elements in the original
 533                                  * scatter list. Handle this correctly.
 534                                  */
 535
 536                                 if (j == bounce_sgl_count) {
 537                                         /*
 538                                          * We are done; cleanup and return.
 539                                          */
 540                                         kunmap_atomic((void *)(dest_addr -
 541                                                         orig_sgl[i].offset),
 542                                                         KM_IRQ0);
 543                                         local_irq_restore(flags);
 544                                         return total_copied;
 545                                 }
 546
 547                                 /* if we need to use another bounce buffer */
 548                                 if (destlen || i != orig_sgl_count - 1)
 549                                         bounce_addr =
 550                                         (unsigned long)kmap_atomic(
 551                                         sg_page((&bounce_sgl[j])), KM_IRQ0);
 552                         } else if (destlen == 0 && i == orig_sgl_count - 1) {
 553                                 /* unmap the last bounce that is < PAGE_SIZE */
 554                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
 555                         }
 556                 }
 557
 558                 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
 559                               KM_IRQ0);
 560         }
 561
 562         local_irq_restore(flags);
 563
 564         return total_copied;
 565 }
 566
 567 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
 568 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
 569                                           struct scatterlist *bounce_sgl,
 570                                           unsigned int orig_sgl_count)
 571 {
 572         int i;
 573         int j = 0;
 574         unsigned long src, dest;
 575         unsigned int srclen, destlen, copylen;
 576         unsigned int total_copied = 0;
 577         unsigned long bounce_addr = 0;
 578         unsigned long src_addr = 0;
 579         unsigned long flags;
 580
 581         local_irq_save(flags);
 582
 583         for (i = 0; i < orig_sgl_count; i++) {
 584                 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
 585                                 KM_IRQ0) + orig_sgl[i].offset;
 586                 src = src_addr;
 587                 srclen = orig_sgl[i].length;
 588
 589                 if (bounce_addr == 0)
 590                         bounce_addr =
 591                         (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
 592                                                 KM_IRQ0);
 593
 594                 while (srclen) {
 595                         /* assume bounce offset always == 0 */
 596                         dest = bounce_addr + bounce_sgl[j].length;
 597                         destlen = PAGE_SIZE - bounce_sgl[j].length;
 598
 599                         copylen = min(srclen, destlen);
 600                         memcpy((void *)dest, (void *)src, copylen);
 601
 602                         total_copied += copylen;
 603                         bounce_sgl[j].length += copylen;
 604                         srclen -= copylen;
 605                         src += copylen;
 606
 607                         if (bounce_sgl[j].length == PAGE_SIZE) {
 608                                 /* full..move to next entry */
 609                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
 610                                 j++;
 611
 612                                 /* if we need to use another bounce buffer */
 613                                 if (srclen || i != orig_sgl_count - 1)
 614                                         bounce_addr =
 615                                         (unsigned long)kmap_atomic(
 616                                         sg_page((&bounce_sgl[j])), KM_IRQ0);
 617
 618                         } else if (srclen == 0 && i == orig_sgl_count - 1) {
 619                                 /* unmap the last bounce that is < PAGE_SIZE */
 620                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
 621                         }
 622                 }
 623
 624                 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
 625         }
 626
 627         local_irq_restore(flags);
 628
 629         return total_copied;
 630 }
 631
 632 static int storvsc_channel_init(struct hv_device *device)
 633 {
 634         struct storvsc_device *stor_device;
 635         struct storvsc_cmd_request *request;
 636         struct vstor_packet *vstor_packet;
 637         int ret, t;
 638
 639         stor_device = get_out_stor_device(device);
 640         if (!stor_device)
 641                 return -ENODEV;
 642
 643         request = &stor_device->init_request;
 644         vstor_packet = &request->vstor_packet;
 645
 646         /*
 647          * Now, initiate the vsc/vsp initialization protocol on the open
 648          * channel
 649          */
 650         memset(request, 0, sizeof(struct storvsc_cmd_request));
 651         init_completion(&request->wait_event);
 652         vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
 653         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 654
 655         ret = vmbus_sendpacket(device->channel, vstor_packet,
 656                                sizeof(struct vstor_packet),
 657                                (unsigned long)request,
 658                                VM_PKT_DATA_INBAND,
 659                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 660         if (ret != 0)
 661                 goto cleanup;
 662
 663         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 664         if (t == 0) {
 665                 ret = -ETIMEDOUT;
 666                 goto cleanup;
 667         }
 668
 669         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 670             vstor_packet->status != 0)
 671                 goto cleanup;
 672
 673
 674         /* reuse the packet for version range supported */
 675         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 676         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
 677         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 678
 679         vstor_packet->version.major_minor =
 680                 storvsc_get_version(VMSTOR_CURRENT_MAJOR, VMSTOR_CURRENT_MINOR);
 681
 682         /*
 683          * The revision number is only used in Windows; set it to 0.
 684          */
 685         vstor_packet->version.revision = 0;
 686
 687         ret = vmbus_sendpacket(device->channel, vstor_packet,
 688                                sizeof(struct vstor_packet),
 689                                (unsigned long)request,
 690                                VM_PKT_DATA_INBAND,
 691                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 692         if (ret != 0)
 693                 goto cleanup;
 694
 695         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 696         if (t == 0) {
 697                 ret = -ETIMEDOUT;
 698                 goto cleanup;
 699         }
 700
 701         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 702             vstor_packet->status != 0)
 703                 goto cleanup;
 704
 705
 706         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 707         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
 708         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 709         vstor_packet->storage_channel_properties.port_number =
 710                                         stor_device->port_number;
 711
 712         ret = vmbus_sendpacket(device->channel, vstor_packet,
 713                                sizeof(struct vstor_packet),
 714                                (unsigned long)request,
 715                                VM_PKT_DATA_INBAND,
 716                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 717
 718         if (ret != 0)
 719                 goto cleanup;
 720
 721         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 722         if (t == 0) {
 723                 ret = -ETIMEDOUT;
 724                 goto cleanup;
 725         }
 726
 727         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 728             vstor_packet->status != 0)
 729                 goto cleanup;
 730
 731         stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
 732         stor_device->target_id
 733                 = vstor_packet->storage_channel_properties.target_id;
 734
 735         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 736         vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
 737         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 738
 739         ret = vmbus_sendpacket(device->channel, vstor_packet,
 740                                sizeof(struct vstor_packet),
 741                                (unsigned long)request,
 742                                VM_PKT_DATA_INBAND,
 743                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 744
 745         if (ret != 0)
 746                 goto cleanup;
 747
 748         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 749         if (t == 0) {
 750                 ret = -ETIMEDOUT;
 751                 goto cleanup;
 752         }
 753
 754         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 755             vstor_packet->status != 0)
 756                 goto cleanup;
 757
 758
 759 cleanup:
 760         return ret;
 761 }
 762
 763
 764 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
 765 {
 766         struct scsi_cmnd *scmnd = cmd_request->cmd;
 767         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
 768         void (*scsi_done_fn)(struct scsi_cmnd *);
 769         struct scsi_sense_hdr sense_hdr;
 770         struct vmscsi_request *vm_srb;
 771         struct storvsc_scan_work *wrk;
 772         struct stor_mem_pools *memp = scmnd->device->hostdata;
 773
 774         vm_srb = &cmd_request->vstor_packet.vm_srb;
 775         if (cmd_request->bounce_sgl_count) {
 776                 if (vm_srb->data_in == READ_TYPE)
 777                         copy_from_bounce_buffer(scsi_sglist(scmnd),
 778                                         cmd_request->bounce_sgl,
 779                                         scsi_sg_count(scmnd),
 780                                         cmd_request->bounce_sgl_count);
 781                 destroy_bounce_buffer(cmd_request->bounce_sgl,
 782                                         cmd_request->bounce_sgl_count);
 783         }
 784
 785         /*
 786          * If there is an error; offline the device since all
 787          * error recovery strategies would have already been
 788          * deployed on the host side. However, if the command
 789          * were a pass-through command deal with it appropriately.
 790          */
 791         switch (vm_srb->srb_status) {
 792         case SRB_STATUS_ERROR:
 793                 switch (scmnd->cmnd[0]) {
 794                 case ATA_16:
 795                 case ATA_12:
 796                         scmnd->result = DID_PASSTHROUGH << 16;
 797                         break;
 798                 default:
 799                         scmnd->result = DID_TARGET_FAILURE << 16;
 800                 }
 801                 break;
 802         default:
 803                 scmnd->result = vm_srb->scsi_status;
 804         }
 805
 806
 807         /*
 808          * If the LUN is invalid; remove the device.
 809          */
 810         if (vm_srb->srb_status == SRB_STATUS_INVALID_LUN) {
 811                 struct storvsc_device *stor_dev;
 812                 struct hv_device *dev = host_dev->dev;
 813                 struct Scsi_Host *host;
 814
 815                 stor_dev = get_in_stor_device(dev);
 816                 host = stor_dev->host;
 817
 818                 wrk = kmalloc(sizeof(struct storvsc_scan_work),
 819                                 GFP_ATOMIC);
 820                 if (!wrk) {
 821                         scmnd->result = DID_TARGET_FAILURE << 16;
 822                 } else {
 823                         wrk->host = host;
 824                         wrk->lun = vm_srb->lun;
 825                         INIT_WORK(&wrk->work, storvsc_remove_lun);
 826                         schedule_work(&wrk->work);
 827                 }
 828         }
 829
 830         if (scmnd->result) {
 831                 if (scsi_normalize_sense(scmnd->sense_buffer,
 832                                 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
 833                         scsi_print_sense_hdr("storvsc", &sense_hdr);
 834         }
 835
 836         scsi_set_resid(scmnd,
 837                 cmd_request->data_buffer.len -
 838                 vm_srb->data_transfer_length);
 839
 840         scsi_done_fn = scmnd->scsi_done;
 841
 842         scmnd->host_scribble = NULL;
 843         scmnd->scsi_done = NULL;
 844
 845         scsi_done_fn(scmnd);
 846
 847         mempool_free(cmd_request, memp->request_mempool);
 848 }
 849
 850 static void storvsc_on_io_completion(struct hv_device *device,
 851                                   struct vstor_packet *vstor_packet,
 852                                   struct storvsc_cmd_request *request)
 853 {
 854         struct storvsc_device *stor_device;
 855         struct vstor_packet *stor_pkt;
 856
 857         stor_device = hv_get_drvdata(device);
 858         stor_pkt = &request->vstor_packet;
 859
 860         /*
 861          * The current SCSI handling on the host side does
 862          * not correctly handle:
 863          * INQUIRY command with page code parameter set to 0x80
 864          * MODE_SENSE command with cmd[2] == 0x1c
 865          *
 866          * Setup srb and scsi status so this won't be fatal.
 867          * We do this so we can distinguish truly fatal failues
 868          * (srb status == 0x4) and off-line the device in that case.
 869          */
 870
 871         if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
 872            (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
 873                 vstor_packet->vm_srb.scsi_status = 0;
 874                 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
 875         }
 876
 877
 878         /* Copy over the status...etc */
 879         stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
 880         stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
 881         stor_pkt->vm_srb.sense_info_length =
 882         vstor_packet->vm_srb.sense_info_length;
 883
 884         if (vstor_packet->vm_srb.scsi_status != 0 ||
 885                 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){
 886                 dev_warn(&device->device,
 887                          "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
 888                          stor_pkt->vm_srb.cdb[0],
 889                          vstor_packet->vm_srb.scsi_status,
 890                          vstor_packet->vm_srb.srb_status);
 891         }
 892
 893         if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
 894                 /* CHECK_CONDITION */
 895                 if (vstor_packet->vm_srb.srb_status &
 896                         SRB_STATUS_AUTOSENSE_VALID) {
 897                         /* autosense data available */
 898                         dev_warn(&device->device,
 899                                  "stor pkt %p autosense data valid - len %d\n",
 900                                  request,
 901                                  vstor_packet->vm_srb.sense_info_length);
 902
 903                         memcpy(request->sense_buffer,
 904                                vstor_packet->vm_srb.sense_data,
 905                                vstor_packet->vm_srb.sense_info_length);
 906
 907                 }
 908         }
 909
 910         stor_pkt->vm_srb.data_transfer_length =
 911         vstor_packet->vm_srb.data_transfer_length;
 912
 913         storvsc_command_completion(request);
 914
 915         if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
 916                 stor_device->drain_notify)
 917                 wake_up(&stor_device->waiting_to_drain);
 918
 919
 920 }
 921
 922 static void storvsc_on_receive(struct hv_device *device,
 923                              struct vstor_packet *vstor_packet,
 924                              struct storvsc_cmd_request *request)
 925 {
 926         struct storvsc_scan_work *work;
 927         struct storvsc_device *stor_device;
 928
 929         switch (vstor_packet->operation) {
 930         case VSTOR_OPERATION_COMPLETE_IO:
 931                 storvsc_on_io_completion(device, vstor_packet, request);
 932                 break;
 933
 934         case VSTOR_OPERATION_REMOVE_DEVICE:
 935         case VSTOR_OPERATION_ENUMERATE_BUS:
 936                 stor_device = get_in_stor_device(device);
 937                 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
 938                 if (!work)
 939                         return;
 940
 941                 INIT_WORK(&work->work, storvsc_bus_scan);
 942                 work->host = stor_device->host;
 943                 schedule_work(&work->work);
 944                 break;
 945
 946         default:
 947                 break;
 948         }
 949 }
 950
 951 static void storvsc_on_channel_callback(void *context)
 952 {
 953         struct hv_device *device = (struct hv_device *)context;
 954         struct storvsc_device *stor_device;
 955         u32 bytes_recvd;
 956         u64 request_id;
 957         unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
 958         struct storvsc_cmd_request *request;
 959         int ret;
 960
 961
 962         stor_device = get_in_stor_device(device);
 963         if (!stor_device)
 964                 return;
 965
 966         do {
 967                 ret = vmbus_recvpacket(device->channel, packet,
 968                                        ALIGN(sizeof(struct vstor_packet), 8),
 969                                        &bytes_recvd, &request_id);
 970                 if (ret == 0 && bytes_recvd > 0) {
 971
 972                         request = (struct storvsc_cmd_request *)
 973                                         (unsigned long)request_id;
 974
 975                         if ((request == &stor_device->init_request) ||
 976                             (request == &stor_device->reset_request)) {
 977
 978                                 memcpy(&request->vstor_packet, packet,
 979                                        sizeof(struct vstor_packet));
 980                                 complete(&request->wait_event);
 981                         } else {
 982                                 storvsc_on_receive(device,
 983                                                 (struct vstor_packet *)packet,
 984                                                 request);
 985                         }
 986                 } else {
 987                         break;
 988                 }
 989         } while (1);
 990
 991         return;
 992 }
 993
 994 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
 995 {
 996         struct vmstorage_channel_properties props;
 997         int ret;
 998
 999         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1000
1001         ret = vmbus_open(device->channel,
1002                          ring_size,
1003                          ring_size,
1004                          (void *)&props,
1005                          sizeof(struct vmstorage_channel_properties),
1006                          storvsc_on_channel_callback, device);
1007
1008         if (ret != 0)
1009                 return ret;
1010
1011         ret = storvsc_channel_init(device);
1012
1013         return ret;
1014 }
1015
1016 static int storvsc_dev_remove(struct hv_device *device)
1017 {
1018         struct storvsc_device *stor_device;
1019         unsigned long flags;
1020
1021         stor_device = hv_get_drvdata(device);
1022
1023         spin_lock_irqsave(&device->channel->inbound_lock, flags);
1024         stor_device->destroy = true;
1025         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1026
1027         /*
1028          * At this point, all outbound traffic should be disable. We
1029          * only allow inbound traffic (responses) to proceed so that
1030          * outstanding requests can be completed.
1031          */
1032
1033         storvsc_wait_to_drain(stor_device);
1034
1035         /*
1036          * Since we have already drained, we don't need to busy wait
1037          * as was done in final_release_stor_device()
1038          * Note that we cannot set the ext pointer to NULL until
1039          * we have drained - to drain the outgoing packets, we need to
1040          * allow incoming packets.
1041          */
1042         spin_lock_irqsave(&device->channel->inbound_lock, flags);
1043         hv_set_drvdata(device, NULL);
1044         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1045
1046         /* Close the channel */
1047         vmbus_close(device->channel);
1048
1049         kfree(stor_device);
1050         return 0;
1051 }
1052
1053 static int storvsc_do_io(struct hv_device *device,
1054                               struct storvsc_cmd_request *request)
1055 {
1056         struct storvsc_device *stor_device;
1057         struct vstor_packet *vstor_packet;
1058         int ret = 0;
1059
1060         vstor_packet = &request->vstor_packet;
1061         stor_device = get_out_stor_device(device);
1062
1063         if (!stor_device)
1064                 return -ENODEV;
1065
1066
1067         request->device  = device;
1068
1069
1070         vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1071
1072         vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1073
1074
1075         vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1076
1077
1078         vstor_packet->vm_srb.data_transfer_length =
1079         request->data_buffer.len;
1080
1081         vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1082
1083         if (request->data_buffer.len) {
1084                 ret = vmbus_sendpacket_multipagebuffer(device->channel,
1085                                 &request->data_buffer,
1086                                 vstor_packet,
1087                                 sizeof(struct vstor_packet),
1088                                 (unsigned long)request);
1089         } else {
1090                 ret = vmbus_sendpacket(device->channel, vstor_packet,
1091                                sizeof(struct vstor_packet),
1092                                (unsigned long)request,
1093                                VM_PKT_DATA_INBAND,
1094                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1095         }
1096
1097         if (ret != 0)
1098                 return ret;
1099
1100         atomic_inc(&stor_device->num_outstanding_req);
1101
1102         return ret;
1103 }
1104
1105 static int storvsc_device_alloc(struct scsi_device *sdevice)
1106 {
1107         struct stor_mem_pools *memp;
1108         int number = STORVSC_MIN_BUF_NR;
1109
1110         memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
1111         if (!memp)
1112                 return -ENOMEM;
1113
1114         memp->request_pool =
1115                 kmem_cache_create(dev_name(&sdevice->sdev_dev),
1116                                 sizeof(struct storvsc_cmd_request), 0,
1117                                 SLAB_HWCACHE_ALIGN, NULL);
1118
1119         if (!memp->request_pool)
1120                 goto err0;
1121
1122         memp->request_mempool = mempool_create(number, mempool_alloc_slab,
1123                                                 mempool_free_slab,
1124                                                 memp->request_pool);
1125
1126         if (!memp->request_mempool)
1127                 goto err1;
1128
1129         sdevice->hostdata = memp;
1130
1131         return 0;
1132
1133 err1:
1134         kmem_cache_destroy(memp->request_pool);
1135
1136 err0:
1137         kfree(memp);
1138         return -ENOMEM;
1139 }
1140
1141 static void storvsc_device_destroy(struct scsi_device *sdevice)
1142 {
1143         struct stor_mem_pools *memp = sdevice->hostdata;
1144
1145         mempool_destroy(memp->request_mempool);
1146         kmem_cache_destroy(memp->request_pool);
1147         kfree(memp);
1148         sdevice->hostdata = NULL;
1149 }
1150
1151 static int storvsc_device_configure(struct scsi_device *sdevice)
1152 {
1153         scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
1154                                 STORVSC_MAX_IO_REQUESTS);
1155
1156         blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
1157
1158         blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
1159
1160         return 0;
1161 }
1162
1163 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1164                            sector_t capacity, int *info)
1165 {
1166         sector_t nsect = capacity;
1167         sector_t cylinders = nsect;
1168         int heads, sectors_pt;
1169
1170         /*
1171          * We are making up these values; let us keep it simple.
1172          */
1173         heads = 0xff;
1174         sectors_pt = 0x3f;      /* Sectors per track */
1175         sector_div(cylinders, heads * sectors_pt);
1176         if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1177                 cylinders = 0xffff;
1178
1179         info[0] = heads;
1180         info[1] = sectors_pt;
1181         info[2] = (int)cylinders;
1182
1183         return 0;
1184 }
1185
1186 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1187 {
1188         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1189         struct hv_device *device = host_dev->dev;
1190
1191         struct storvsc_device *stor_device;
1192         struct storvsc_cmd_request *request;
1193         struct vstor_packet *vstor_packet;
1194         int ret, t;
1195
1196
1197         stor_device = get_out_stor_device(device);
1198         if (!stor_device)
1199                 return FAILED;
1200
1201         request = &stor_device->reset_request;
1202         vstor_packet = &request->vstor_packet;
1203
1204         init_completion(&request->wait_event);
1205
1206         vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1207         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1208         vstor_packet->vm_srb.path_id = stor_device->path_id;
1209
1210         ret = vmbus_sendpacket(device->channel, vstor_packet,
1211                                sizeof(struct vstor_packet),
1212                                (unsigned long)&stor_device->reset_request,
1213                                VM_PKT_DATA_INBAND,
1214                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1215         if (ret != 0)
1216                 return FAILED;
1217
1218         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1219         if (t == 0)
1220                 return TIMEOUT_ERROR;
1221
1222
1223         /*
1224          * At this point, all outstanding requests in the adapter
1225          * should have been flushed out and return to us
1226          */
1227
1228         return SUCCESS;
1229 }
1230
1231 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1232 {
1233         bool allowed = true;
1234         u8 scsi_op = scmnd->cmnd[0];
1235
1236         switch (scsi_op) {
1237         /*
1238          * smartd sends this command and the host does not handle
1239          * this. So, don't send it.
1240          */
1241         case SET_WINDOW:
1242                 scmnd->result = ILLEGAL_REQUEST << 16;
1243                 allowed = false;
1244                 break;
1245         default:
1246                 break;
1247         }
1248         return allowed;
1249 }
1250
1251 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1252 {
1253         int ret;
1254         struct hv_host_device *host_dev = shost_priv(host);
1255         struct hv_device *dev = host_dev->dev;
1256         struct storvsc_cmd_request *cmd_request;
1257         unsigned int request_size = 0;
1258         int i;
1259         struct scatterlist *sgl;
1260         unsigned int sg_count = 0;
1261         struct vmscsi_request *vm_srb;
1262         struct stor_mem_pools *memp = scmnd->device->hostdata;
1263
1264         if (!storvsc_scsi_cmd_ok(scmnd)) {
1265                 scmnd->scsi_done(scmnd);
1266                 return 0;
1267         }
1268
1269         request_size = sizeof(struct storvsc_cmd_request);
1270
1271         cmd_request = mempool_alloc(memp->request_mempool,
1272                                        GFP_ATOMIC);
1273
1274         /*
1275          * We might be invoked in an interrupt context; hence
1276          * mempool_alloc() can fail.
1277          */
1278         if (!cmd_request)
1279                 return SCSI_MLQUEUE_DEVICE_BUSY;
1280
1281         memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1282
1283         /* Setup the cmd request */
1284         cmd_request->cmd = scmnd;
1285
1286         scmnd->host_scribble = (unsigned char *)cmd_request;
1287
1288         vm_srb = &cmd_request->vstor_packet.vm_srb;
1289
1290
1291         /* Build the SRB */
1292         switch (scmnd->sc_data_direction) {
1293         case DMA_TO_DEVICE:
1294                 vm_srb->data_in = WRITE_TYPE;
1295                 break;
1296         case DMA_FROM_DEVICE:
1297                 vm_srb->data_in = READ_TYPE;
1298                 break;
1299         default:
1300                 vm_srb->data_in = UNKNOWN_TYPE;
1301                 break;
1302         }
1303
1304
1305         vm_srb->port_number = host_dev->port;
1306         vm_srb->path_id = scmnd->device->channel;
1307         vm_srb->target_id = scmnd->device->id;
1308         vm_srb->lun = scmnd->device->lun;
1309
1310         vm_srb->cdb_length = scmnd->cmd_len;
1311
1312         memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1313
1314         cmd_request->sense_buffer = scmnd->sense_buffer;
1315
1316
1317         cmd_request->data_buffer.len = scsi_bufflen(scmnd);
1318         if (scsi_sg_count(scmnd)) {
1319                 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1320                 sg_count = scsi_sg_count(scmnd);
1321
1322                 /* check if we need to bounce the sgl */
1323                 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1324                         cmd_request->bounce_sgl =
1325                                 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1326                                                      scsi_bufflen(scmnd),
1327                                                      vm_srb->data_in);
1328                         if (!cmd_request->bounce_sgl) {
1329                                 ret = SCSI_MLQUEUE_HOST_BUSY;
1330                                 goto queue_error;
1331                         }
1332
1333                         cmd_request->bounce_sgl_count =
1334                                 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1335                                         PAGE_SHIFT;
1336
1337                         if (vm_srb->data_in == WRITE_TYPE)
1338                                 copy_to_bounce_buffer(sgl,
1339                                         cmd_request->bounce_sgl,
1340                                         scsi_sg_count(scmnd));
1341
1342                         sgl = cmd_request->bounce_sgl;
1343                         sg_count = cmd_request->bounce_sgl_count;
1344                 }
1345
1346                 cmd_request->data_buffer.offset = sgl[0].offset;
1347
1348                 for (i = 0; i < sg_count; i++)
1349                         cmd_request->data_buffer.pfn_array[i] =
1350                                 page_to_pfn(sg_page((&sgl[i])));
1351
1352         } else if (scsi_sglist(scmnd)) {
1353                 cmd_request->data_buffer.offset =
1354                         virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1355                 cmd_request->data_buffer.pfn_array[0] =
1356                         virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1357         }
1358
1359         /* Invokes the vsc to start an IO */
1360         ret = storvsc_do_io(dev, cmd_request);
1361
1362         if (ret == -EAGAIN) {
1363                 /* no more space */
1364
1365                 if (cmd_request->bounce_sgl_count) {
1366                         destroy_bounce_buffer(cmd_request->bounce_sgl,
1367                                         cmd_request->bounce_sgl_count);
1368
1369                         ret = SCSI_MLQUEUE_DEVICE_BUSY;
1370                         goto queue_error;
1371                 }
1372         }
1373
1374         return 0;
1375
1376 queue_error:
1377         mempool_free(cmd_request, memp->request_mempool);
1378         scmnd->host_scribble = NULL;
1379         return ret;
1380 }
1381
1382 static struct scsi_host_template scsi_driver = {
1383         .module =               THIS_MODULE,
1384         .name =                 "storvsc_host_t",
1385         .bios_param =           storvsc_get_chs,
1386         .queuecommand =         storvsc_queuecommand,
1387         .eh_host_reset_handler =        storvsc_host_reset_handler,
1388         .slave_alloc =          storvsc_device_alloc,
1389         .slave_destroy =        storvsc_device_destroy,
1390         .slave_configure =      storvsc_device_configure,
1391         .cmd_per_lun =          1,
1392         /* 64 max_queue * 1 target */
1393         .can_queue =            STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1394         .this_id =              -1,
1395         /* no use setting to 0 since ll_blk_rw reset it to 1 */
1396         /* currently 32 */
1397         .sg_tablesize =         MAX_MULTIPAGE_BUFFER_COUNT,
1398         .use_clustering =       DISABLE_CLUSTERING,
1399         /* Make sure we dont get a sg segment crosses a page boundary */
1400         .dma_boundary =         PAGE_SIZE-1,
1401 };
1402
1403 enum {
1404         SCSI_GUID,
1405         IDE_GUID,
1406 };
1407
1408 static const struct hv_vmbus_device_id id_table[] = {
1409         /* SCSI guid */
1410         { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1411                        0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1412           .driver_data = SCSI_GUID },
1413         /* IDE guid */
1414         { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1415                        0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1416           .driver_data = IDE_GUID },
1417         { },
1418 };
1419
1420 MODULE_DEVICE_TABLE(vmbus, id_table);
1421
1422 static int storvsc_probe(struct hv_device *device,
1423                         const struct hv_vmbus_device_id *dev_id)
1424 {
1425         int ret;
1426         struct Scsi_Host *host;
1427         struct hv_host_device *host_dev;
1428         bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1429         int target = 0;
1430         struct storvsc_device *stor_device;
1431
1432         host = scsi_host_alloc(&scsi_driver,
1433                                sizeof(struct hv_host_device));
1434         if (!host)
1435                 return -ENOMEM;
1436
1437         host_dev = shost_priv(host);
1438         memset(host_dev, 0, sizeof(struct hv_host_device));
1439
1440         host_dev->port = host->host_no;
1441         host_dev->dev = device;
1442
1443
1444         stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1445         if (!stor_device) {
1446                 ret = -ENOMEM;
1447                 goto err_out0;
1448         }
1449
1450         stor_device->destroy = false;
1451         init_waitqueue_head(&stor_device->waiting_to_drain);
1452         stor_device->device = device;
1453         stor_device->host = host;
1454         hv_set_drvdata(device, stor_device);
1455
1456         stor_device->port_number = host->host_no;
1457         ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1458         if (ret)
1459                 goto err_out1;
1460
1461         host_dev->path = stor_device->path_id;
1462         host_dev->target = stor_device->target_id;
1463
1464         /* max # of devices per target */
1465         host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1466         /* max # of targets per channel */
1467         host->max_id = STORVSC_MAX_TARGETS;
1468         /* max # of channels */
1469         host->max_channel = STORVSC_MAX_CHANNELS - 1;
1470         /* max cmd length */
1471         host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1472
1473         /* Register the HBA and start the scsi bus scan */
1474         ret = scsi_add_host(host, &device->device);
1475         if (ret != 0)
1476                 goto err_out2;
1477
1478         if (!dev_is_ide) {
1479                 scsi_scan_host(host);
1480         } else {
1481                 target = (device->dev_instance.b[5] << 8 |
1482                          device->dev_instance.b[4]);
1483                 ret = scsi_add_device(host, 0, target, 0);
1484                 if (ret) {
1485                         scsi_remove_host(host);
1486                         goto err_out2;
1487                 }
1488         }
1489         return 0;
1490
1491 err_out2:
1492         /*
1493          * Once we have connected with the host, we would need to
1494          * to invoke storvsc_dev_remove() to rollback this state and
1495          * this call also frees up the stor_device; hence the jump around
1496          * err_out1 label.
1497          */
1498         storvsc_dev_remove(device);
1499         goto err_out0;
1500
1501 err_out1:
1502         kfree(stor_device);
1503
1504 err_out0:
1505         scsi_host_put(host);
1506         return ret;
1507 }
1508
1509 static int storvsc_remove(struct hv_device *dev)
1510 {
1511         struct storvsc_device *stor_device = hv_get_drvdata(dev);
1512         struct Scsi_Host *host = stor_device->host;
1513
1514         scsi_remove_host(host);
1515         storvsc_dev_remove(dev);
1516         scsi_host_put(host);
1517
1518         return 0;
1519 }
1520
1521 static struct hv_driver storvsc_drv = {
1522         .name = KBUILD_MODNAME,
1523         .id_table = id_table,
1524         .probe = storvsc_probe,
1525         .remove = storvsc_remove,
1526 };
1527
1528 static int __init storvsc_drv_init(void)
1529 {
1530         u32 max_outstanding_req_per_channel;
1531
1532         /*
1533          * Divide the ring buffer data size (which is 1 page less
1534          * than the ring buffer size since that page is reserved for
1535          * the ring buffer indices) by the max request size (which is
1536          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1537          */
1538         max_outstanding_req_per_channel =
1539                 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1540                 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1541                 sizeof(struct vstor_packet) + sizeof(u64),
1542                 sizeof(u64)));
1543
1544         if (max_outstanding_req_per_channel <
1545             STORVSC_MAX_IO_REQUESTS)
1546                 return -EINVAL;
1547
1548         return vmbus_driver_register(&storvsc_drv);
1549 }
1550
1551 static void __exit storvsc_drv_exit(void)
1552 {
1553         vmbus_driver_unregister(&storvsc_drv);
1554 }
1555
1556 MODULE_LICENSE("GPL");
1557 MODULE_VERSION(HV_DRV_VERSION);
1558 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1559 module_init(storvsc_drv_init);
1560 module_exit(storvsc_drv_exit);