drivers/scsi/vmw_pvscsi.c

   1 /*
   2  * Linux driver for VMware's para-virtualized SCSI HBA.
   3  *
   4  * Copyright (C) 2008-2014, VMware, Inc. All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; version 2 of the License and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  13  * NON INFRINGEMENT.  See the GNU General Public License for more
  14  * details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Maintained by: Arvind Kumar <arvindkumar@vmware.com>
  21  *
  22  */
  23
  24 #include <linux/kernel.h>
  25 #include <linux/module.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/slab.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/pci.h>
  30
  31 #include <scsi/scsi.h>
  32 #include <scsi/scsi_host.h>
  33 #include <scsi/scsi_cmnd.h>
  34 #include <scsi/scsi_device.h>
  35 #include <scsi/scsi_tcq.h>
  36
  37 #include "vmw_pvscsi.h"
  38
  39 #define PVSCSI_LINUX_DRIVER_DESC "VMware PVSCSI driver"
  40
  41 MODULE_DESCRIPTION(PVSCSI_LINUX_DRIVER_DESC);
  42 MODULE_AUTHOR("VMware, Inc.");
  43 MODULE_LICENSE("GPL");
  44 MODULE_VERSION(PVSCSI_DRIVER_VERSION_STRING);
  45
  46 #define PVSCSI_DEFAULT_NUM_PAGES_PER_RING       8
  47 #define PVSCSI_DEFAULT_NUM_PAGES_MSG_RING       1
  48 #define PVSCSI_DEFAULT_QUEUE_DEPTH              254
  49 #define SGL_SIZE                                PAGE_SIZE
  50
  51 struct pvscsi_sg_list {
  52         struct PVSCSISGElement sge[PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT];
  53 };
  54
  55 struct pvscsi_ctx {
  56         /*
  57          * The index of the context in cmd_map serves as the context ID for a
  58          * 1-to-1 mapping completions back to requests.
  59          */
  60         struct scsi_cmnd        *cmd;
  61         struct pvscsi_sg_list   *sgl;
  62         struct list_head        list;
  63         dma_addr_t              dataPA;
  64         dma_addr_t              sensePA;
  65         dma_addr_t              sglPA;
  66         struct completion       *abort_cmp;
  67 };
  68
  69 struct pvscsi_adapter {
  70         char                            *mmioBase;
  71         unsigned int                    irq;
  72         u8                              rev;
  73         bool                            use_msi;
  74         bool                            use_msix;
  75         bool                            use_msg;
  76         bool                            use_req_threshold;
  77
  78         spinlock_t                      hw_lock;
  79
  80         struct workqueue_struct         *workqueue;
  81         struct work_struct              work;
  82
  83         struct PVSCSIRingReqDesc        *req_ring;
  84         unsigned                        req_pages;
  85         unsigned                        req_depth;
  86         dma_addr_t                      reqRingPA;
  87
  88         struct PVSCSIRingCmpDesc        *cmp_ring;
  89         unsigned                        cmp_pages;
  90         dma_addr_t                      cmpRingPA;
  91
  92         struct PVSCSIRingMsgDesc        *msg_ring;
  93         unsigned                        msg_pages;
  94         dma_addr_t                      msgRingPA;
  95
  96         struct PVSCSIRingsState         *rings_state;
  97         dma_addr_t                      ringStatePA;
  98
  99         struct pci_dev                  *dev;
 100         struct Scsi_Host                *host;
 101
 102         struct list_head                cmd_pool;
 103         struct pvscsi_ctx               *cmd_map;
 104 };
 105
 106
 107 /* Command line parameters */
 108 static int pvscsi_ring_pages;
 109 static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING;
 110 static int pvscsi_cmd_per_lun    = PVSCSI_DEFAULT_QUEUE_DEPTH;
 111 static bool pvscsi_disable_msi;
 112 static bool pvscsi_disable_msix;
 113 static bool pvscsi_use_msg       = true;
 114 static bool pvscsi_use_req_threshold = true;
 115
 116 #define PVSCSI_RW (S_IRUSR | S_IWUSR)
 117
 118 module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW);
 119 MODULE_PARM_DESC(ring_pages, "Number of pages per req/cmp ring - (default="
 120                  __stringify(PVSCSI_DEFAULT_NUM_PAGES_PER_RING)
 121                  "[up to 16 targets],"
 122                  __stringify(PVSCSI_SETUP_RINGS_MAX_NUM_PAGES)
 123                  "[for 16+ targets])");
 124
 125 module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW);
 126 MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default="
 127                  __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")");
 128
 129 module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW);
 130 MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default="
 131                  __stringify(PVSCSI_DEFAULT_QUEUE_DEPTH) ")");
 132
 133 module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW);
 134 MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)");
 135
 136 module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW);
 137 MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)");
 138
 139 module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW);
 140 MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)");
 141
 142 module_param_named(use_req_threshold, pvscsi_use_req_threshold,
 143                    bool, PVSCSI_RW);
 144 MODULE_PARM_DESC(use_req_threshold, "Use driver-based request coalescing if configured - (default=1)");
 145
 146 static const struct pci_device_id pvscsi_pci_tbl[] = {
 147         { PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) },
 148         { 0 }
 149 };
 150
 151 MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl);
 152
 153 static struct device *
 154 pvscsi_dev(const struct pvscsi_adapter *adapter)
 155 {
 156         return &(adapter->dev->dev);
 157 }
 158
 159 static struct pvscsi_ctx *
 160 pvscsi_find_context(const struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
 161 {
 162         struct pvscsi_ctx *ctx, *end;
 163
 164         end = &adapter->cmd_map[adapter->req_depth];
 165         for (ctx = adapter->cmd_map; ctx < end; ctx++)
 166                 if (ctx->cmd == cmd)
 167                         return ctx;
 168
 169         return NULL;
 170 }
 171
 172 static struct pvscsi_ctx *
 173 pvscsi_acquire_context(struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
 174 {
 175         struct pvscsi_ctx *ctx;
 176
 177         if (list_empty(&adapter->cmd_pool))
 178                 return NULL;
 179
 180         ctx = list_first_entry(&adapter->cmd_pool, struct pvscsi_ctx, list);
 181         ctx->cmd = cmd;
 182         list_del(&ctx->list);
 183
 184         return ctx;
 185 }
 186
 187 static void pvscsi_release_context(struct pvscsi_adapter *adapter,
 188                                    struct pvscsi_ctx *ctx)
 189 {
 190         ctx->cmd = NULL;
 191         ctx->abort_cmp = NULL;
 192         list_add(&ctx->list, &adapter->cmd_pool);
 193 }
 194
 195 /*
 196  * Map a pvscsi_ctx struct to a context ID field value; we map to a simple
 197  * non-zero integer. ctx always points to an entry in cmd_map array, hence
 198  * the return value is always >=1.
 199  */
 200 static u64 pvscsi_map_context(const struct pvscsi_adapter *adapter,
 201                               const struct pvscsi_ctx *ctx)
 202 {
 203         return ctx - adapter->cmd_map + 1;
 204 }
 205
 206 static struct pvscsi_ctx *
 207 pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context)
 208 {
 209         return &adapter->cmd_map[context - 1];
 210 }
 211
 212 static void pvscsi_reg_write(const struct pvscsi_adapter *adapter,
 213                              u32 offset, u32 val)
 214 {
 215         writel(val, adapter->mmioBase + offset);
 216 }
 217
 218 static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset)
 219 {
 220         return readl(adapter->mmioBase + offset);
 221 }
 222
 223 static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter)
 224 {
 225         return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS);
 226 }
 227
 228 static void pvscsi_write_intr_status(const struct pvscsi_adapter *adapter,
 229                                      u32 val)
 230 {
 231         pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_STATUS, val);
 232 }
 233
 234 static void pvscsi_unmask_intr(const struct pvscsi_adapter *adapter)
 235 {
 236         u32 intr_bits;
 237
 238         intr_bits = PVSCSI_INTR_CMPL_MASK;
 239         if (adapter->use_msg)
 240                 intr_bits |= PVSCSI_INTR_MSG_MASK;
 241
 242         pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, intr_bits);
 243 }
 244
 245 static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter)
 246 {
 247         pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0);
 248 }
 249
 250 static void pvscsi_write_cmd_desc(const struct pvscsi_adapter *adapter,
 251                                   u32 cmd, const void *desc, size_t len)
 252 {
 253         const u32 *ptr = desc;
 254         size_t i;
 255
 256         len /= sizeof(*ptr);
 257         pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, cmd);
 258         for (i = 0; i < len; i++)
 259                 pvscsi_reg_write(adapter,
 260                                  PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]);
 261 }
 262
 263 static void pvscsi_abort_cmd(const struct pvscsi_adapter *adapter,
 264                              const struct pvscsi_ctx *ctx)
 265 {
 266         struct PVSCSICmdDescAbortCmd cmd = { 0 };
 267
 268         cmd.target = ctx->cmd->device->id;
 269         cmd.context = pvscsi_map_context(adapter, ctx);
 270
 271         pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ABORT_CMD, &cmd, sizeof(cmd));
 272 }
 273
 274 static void pvscsi_kick_rw_io(const struct pvscsi_adapter *adapter)
 275 {
 276         pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_RW_IO, 0);
 277 }
 278
 279 static void pvscsi_process_request_ring(const struct pvscsi_adapter *adapter)
 280 {
 281         pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_NON_RW_IO, 0);
 282 }
 283
 284 static int scsi_is_rw(unsigned char op)
 285 {
 286         return op == READ_6  || op == WRITE_6 ||
 287                op == READ_10 || op == WRITE_10 ||
 288                op == READ_12 || op == WRITE_12 ||
 289                op == READ_16 || op == WRITE_16;
 290 }
 291
 292 static void pvscsi_kick_io(const struct pvscsi_adapter *adapter,
 293                            unsigned char op)
 294 {
 295         if (scsi_is_rw(op)) {
 296                 struct PVSCSIRingsState *s = adapter->rings_state;
 297
 298                 if (!adapter->use_req_threshold ||
 299                     s->reqProdIdx - s->reqConsIdx >= s->reqCallThreshold)
 300                         pvscsi_kick_rw_io(adapter);
 301         } else {
 302                 pvscsi_process_request_ring(adapter);
 303         }
 304 }
 305
 306 static void ll_adapter_reset(const struct pvscsi_adapter *adapter)
 307 {
 308         dev_dbg(pvscsi_dev(adapter), "Adapter Reset on %p\n", adapter);
 309
 310         pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ADAPTER_RESET, NULL, 0);
 311 }
 312
 313 static void ll_bus_reset(const struct pvscsi_adapter *adapter)
 314 {
 315         dev_dbg(pvscsi_dev(adapter), "Resetting bus on %p\n", adapter);
 316
 317         pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_BUS, NULL, 0);
 318 }
 319
 320 static void ll_device_reset(const struct pvscsi_adapter *adapter, u32 target)
 321 {
 322         struct PVSCSICmdDescResetDevice cmd = { 0 };
 323
 324         dev_dbg(pvscsi_dev(adapter), "Resetting device: target=%u\n", target);
 325
 326         cmd.target = target;
 327
 328         pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_DEVICE,
 329                               &cmd, sizeof(cmd));
 330 }
 331
 332 static void pvscsi_create_sg(struct pvscsi_ctx *ctx,
 333                              struct scatterlist *sg, unsigned count)
 334 {
 335         unsigned i;
 336         struct PVSCSISGElement *sge;
 337
 338         BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT);
 339
 340         sge = &ctx->sgl->sge[0];
 341         for (i = 0; i < count; i++, sg++) {
 342                 sge[i].addr   = sg_dma_address(sg);
 343                 sge[i].length = sg_dma_len(sg);
 344                 sge[i].flags  = 0;
 345         }
 346 }
 347
 348 /*
 349  * Map all data buffers for a command into PCI space and
 350  * setup the scatter/gather list if needed.
 351  */
 352 static void pvscsi_map_buffers(struct pvscsi_adapter *adapter,
 353                                struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd,
 354                                struct PVSCSIRingReqDesc *e)
 355 {
 356         unsigned count;
 357         unsigned bufflen = scsi_bufflen(cmd);
 358         struct scatterlist *sg;
 359
 360         e->dataLen = bufflen;
 361         e->dataAddr = 0;
 362         if (bufflen == 0)
 363                 return;
 364
 365         sg = scsi_sglist(cmd);
 366         count = scsi_sg_count(cmd);
 367         if (count != 0) {
 368                 int segs = scsi_dma_map(cmd);
 369                 if (segs > 1) {
 370                         pvscsi_create_sg(ctx, sg, segs);
 371
 372                         e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST;
 373                         ctx->sglPA = pci_map_single(adapter->dev, ctx->sgl,
 374                                                     SGL_SIZE, PCI_DMA_TODEVICE);
 375                         e->dataAddr = ctx->sglPA;
 376                 } else
 377                         e->dataAddr = sg_dma_address(sg);
 378         } else {
 379                 /*
 380                  * In case there is no S/G list, scsi_sglist points
 381                  * directly to the buffer.
 382                  */
 383                 ctx->dataPA = pci_map_single(adapter->dev, sg, bufflen,
 384                                              cmd->sc_data_direction);
 385                 e->dataAddr = ctx->dataPA;
 386         }
 387 }
 388
 389 static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
 390                                  struct pvscsi_ctx *ctx)
 391 {
 392         struct scsi_cmnd *cmd;
 393         unsigned bufflen;
 394
 395         cmd = ctx->cmd;
 396         bufflen = scsi_bufflen(cmd);
 397
 398         if (bufflen != 0) {
 399                 unsigned count = scsi_sg_count(cmd);
 400
 401                 if (count != 0) {
 402                         scsi_dma_unmap(cmd);
 403                         if (ctx->sglPA) {
 404                                 pci_unmap_single(adapter->dev, ctx->sglPA,
 405                                                  SGL_SIZE, PCI_DMA_TODEVICE);
 406                                 ctx->sglPA = 0;
 407                         }
 408                 } else
 409                         pci_unmap_single(adapter->dev, ctx->dataPA, bufflen,
 410                                          cmd->sc_data_direction);
 411         }
 412         if (cmd->sense_buffer)
 413                 pci_unmap_single(adapter->dev, ctx->sensePA,
 414                                  SCSI_SENSE_BUFFERSIZE, PCI_DMA_FROMDEVICE);
 415 }
 416
 417 static int pvscsi_allocate_rings(struct pvscsi_adapter *adapter)
 418 {
 419         adapter->rings_state = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
 420                                                     &adapter->ringStatePA);
 421         if (!adapter->rings_state)
 422                 return -ENOMEM;
 423
 424         adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING,
 425                                  pvscsi_ring_pages);
 426         adapter->req_depth = adapter->req_pages
 427                                         * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
 428         adapter->req_ring = pci_alloc_consistent(adapter->dev,
 429                                                  adapter->req_pages * PAGE_SIZE,
 430                                                  &adapter->reqRingPA);
 431         if (!adapter->req_ring)
 432                 return -ENOMEM;
 433
 434         adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING,
 435                                  pvscsi_ring_pages);
 436         adapter->cmp_ring = pci_alloc_consistent(adapter->dev,
 437                                                  adapter->cmp_pages * PAGE_SIZE,
 438                                                  &adapter->cmpRingPA);
 439         if (!adapter->cmp_ring)
 440                 return -ENOMEM;
 441
 442         BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE));
 443         BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE));
 444         BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE));
 445
 446         if (!adapter->use_msg)
 447                 return 0;
 448
 449         adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING,
 450                                  pvscsi_msg_ring_pages);
 451         adapter->msg_ring = pci_alloc_consistent(adapter->dev,
 452                                                  adapter->msg_pages * PAGE_SIZE,
 453                                                  &adapter->msgRingPA);
 454         if (!adapter->msg_ring)
 455                 return -ENOMEM;
 456         BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE));
 457
 458         return 0;
 459 }
 460
 461 static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter)
 462 {
 463         struct PVSCSICmdDescSetupRings cmd = { 0 };
 464         dma_addr_t base;
 465         unsigned i;
 466
 467         cmd.ringsStatePPN   = adapter->ringStatePA >> PAGE_SHIFT;
 468         cmd.reqRingNumPages = adapter->req_pages;
 469         cmd.cmpRingNumPages = adapter->cmp_pages;
 470
 471         base = adapter->reqRingPA;
 472         for (i = 0; i < adapter->req_pages; i++) {
 473                 cmd.reqRingPPNs[i] = base >> PAGE_SHIFT;
 474                 base += PAGE_SIZE;
 475         }
 476
 477         base = adapter->cmpRingPA;
 478         for (i = 0; i < adapter->cmp_pages; i++) {
 479                 cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT;
 480                 base += PAGE_SIZE;
 481         }
 482
 483         memset(adapter->rings_state, 0, PAGE_SIZE);
 484         memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE);
 485         memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE);
 486
 487         pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS,
 488                               &cmd, sizeof(cmd));
 489
 490         if (adapter->use_msg) {
 491                 struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 };
 492
 493                 cmd_msg.numPages = adapter->msg_pages;
 494
 495                 base = adapter->msgRingPA;
 496                 for (i = 0; i < adapter->msg_pages; i++) {
 497                         cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT;
 498                         base += PAGE_SIZE;
 499                 }
 500                 memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE);
 501
 502                 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING,
 503                                       &cmd_msg, sizeof(cmd_msg));
 504         }
 505 }
 506
 507 static int pvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
 508 {
 509         if (!sdev->tagged_supported)
 510                 qdepth = 1;
 511         return scsi_change_queue_depth(sdev, qdepth);
 512 }
 513
 514 /*
 515  * Pull a completion descriptor off and pass the completion back
 516  * to the SCSI mid layer.
 517  */
 518 static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
 519                                     const struct PVSCSIRingCmpDesc *e)
 520 {
 521         struct pvscsi_ctx *ctx;
 522         struct scsi_cmnd *cmd;
 523         struct completion *abort_cmp;
 524         u32 btstat = e->hostStatus;
 525         u32 sdstat = e->scsiStatus;
 526
 527         ctx = pvscsi_get_context(adapter, e->context);
 528         cmd = ctx->cmd;
 529         abort_cmp = ctx->abort_cmp;
 530         pvscsi_unmap_buffers(adapter, ctx);
 531         pvscsi_release_context(adapter, ctx);
 532         if (abort_cmp) {
 533                 /*
 534                  * The command was requested to be aborted. Just signal that
 535                  * the request completed and swallow the actual cmd completion
 536                  * here. The abort handler will post a completion for this
 537                  * command indicating that it got successfully aborted.
 538                  */
 539                 complete(abort_cmp);
 540                 return;
 541         }
 542
 543         cmd->result = 0;
 544         if (sdstat != SAM_STAT_GOOD &&
 545             (btstat == BTSTAT_SUCCESS ||
 546              btstat == BTSTAT_LINKED_COMMAND_COMPLETED ||
 547              btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) {
 548                 if (sdstat == SAM_STAT_COMMAND_TERMINATED) {
 549                         cmd->result = (DID_RESET << 16);
 550                 } else {
 551                         cmd->result = (DID_OK << 16) | sdstat;
 552                         if (sdstat == SAM_STAT_CHECK_CONDITION &&
 553                             cmd->sense_buffer)
 554                                 cmd->result |= (DRIVER_SENSE << 24);
 555                 }
 556         } else
 557                 switch (btstat) {
 558                 case BTSTAT_SUCCESS:
 559                 case BTSTAT_LINKED_COMMAND_COMPLETED:
 560                 case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG:
 561                         /*
 562                          * Commands like INQUIRY may transfer less data than
 563                          * requested by the initiator via bufflen. Set residual
 564                          * count to make upper layer aware of the actual amount
 565                          * of data returned.
 566                          */
 567                         scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
 568                         cmd->result = (DID_OK << 16);
 569                         break;
 570
 571                 case BTSTAT_DATARUN:
 572                 case BTSTAT_DATA_UNDERRUN:
 573                         /* Report residual data in underruns */
 574                         scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
 575                         cmd->result = (DID_ERROR << 16);
 576                         break;
 577
 578                 case BTSTAT_SELTIMEO:
 579                         /* Our emulation returns this for non-connected devs */
 580                         cmd->result = (DID_BAD_TARGET << 16);
 581                         break;
 582
 583                 case BTSTAT_LUNMISMATCH:
 584                 case BTSTAT_TAGREJECT:
 585                 case BTSTAT_BADMSG:
 586                         cmd->result = (DRIVER_INVALID << 24);
 587                         /* fall through */
 588
 589                 case BTSTAT_HAHARDWARE:
 590                 case BTSTAT_INVPHASE:
 591                 case BTSTAT_HATIMEOUT:
 592                 case BTSTAT_NORESPONSE:
 593                 case BTSTAT_DISCONNECT:
 594                 case BTSTAT_HASOFTWARE:
 595                 case BTSTAT_BUSFREE:
 596                 case BTSTAT_SENSFAILED:
 597                         cmd->result |= (DID_ERROR << 16);
 598                         break;
 599
 600                 case BTSTAT_SENTRST:
 601                 case BTSTAT_RECVRST:
 602                 case BTSTAT_BUSRESET:
 603                         cmd->result = (DID_RESET << 16);
 604                         break;
 605
 606                 case BTSTAT_ABORTQUEUE:
 607                         cmd->result = (DID_ABORT << 16);
 608                         break;
 609
 610                 case BTSTAT_SCSIPARITY:
 611                         cmd->result = (DID_PARITY << 16);
 612                         break;
 613
 614                 default:
 615                         cmd->result = (DID_ERROR << 16);
 616                         scmd_printk(KERN_DEBUG, cmd,
 617                                     "Unknown completion status: 0x%x\n",
 618                                     btstat);
 619         }
 620
 621         dev_dbg(&cmd->device->sdev_gendev,
 622                 "cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n",
 623                 cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat);
 624
 625         cmd->scsi_done(cmd);
 626 }
 627
 628 /*
 629  * barrier usage : Since the PVSCSI device is emulated, there could be cases
 630  * where we may want to serialize some accesses between the driver and the
 631  * emulation layer. We use compiler barriers instead of the more expensive
 632  * memory barriers because PVSCSI is only supported on X86 which has strong
 633  * memory access ordering.
 634  */
 635 static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter)
 636 {
 637         struct PVSCSIRingsState *s = adapter->rings_state;
 638         struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring;
 639         u32 cmp_entries = s->cmpNumEntriesLog2;
 640
 641         while (s->cmpConsIdx != s->cmpProdIdx) {
 642                 struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx &
 643                                                       MASK(cmp_entries));
 644                 /*
 645                  * This barrier() ensures that *e is not dereferenced while
 646                  * the device emulation still writes data into the slot.
 647                  * Since the device emulation advances s->cmpProdIdx only after
 648                  * updating the slot we want to check it first.
 649                  */
 650                 barrier();
 651                 pvscsi_complete_request(adapter, e);
 652                 /*
 653                  * This barrier() ensures that compiler doesn't reorder write
 654                  * to s->cmpConsIdx before the read of (*e) inside
 655                  * pvscsi_complete_request. Otherwise, device emulation may
 656                  * overwrite *e before we had a chance to read it.
 657                  */
 658                 barrier();
 659                 s->cmpConsIdx++;
 660         }
 661 }
 662
 663 /*
 664  * Translate a Linux SCSI request into a request ring entry.
 665  */
 666 static int pvscsi_queue_ring(struct pvscsi_adapter *adapter,
 667                              struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd)
 668 {
 669         struct PVSCSIRingsState *s;
 670         struct PVSCSIRingReqDesc *e;
 671         struct scsi_device *sdev;
 672         u32 req_entries;
 673
 674         s = adapter->rings_state;
 675         sdev = cmd->device;
 676         req_entries = s->reqNumEntriesLog2;
 677
 678         /*
 679          * If this condition holds, we might have room on the request ring, but
 680          * we might not have room on the completion ring for the response.
 681          * However, we have already ruled out this possibility - we would not
 682          * have successfully allocated a context if it were true, since we only
 683          * have one context per request entry.  Check for it anyway, since it
 684          * would be a serious bug.
 685          */
 686         if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) {
 687                 scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: "
 688                             "ring full: reqProdIdx=%d cmpConsIdx=%d\n",
 689                             s->reqProdIdx, s->cmpConsIdx);
 690                 return -1;
 691         }
 692
 693         e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries));
 694
 695         e->bus    = sdev->channel;
 696         e->target = sdev->id;
 697         memset(e->lun, 0, sizeof(e->lun));
 698         e->lun[1] = sdev->lun;
 699
 700         if (cmd->sense_buffer) {
 701                 ctx->sensePA = pci_map_single(adapter->dev, cmd->sense_buffer,
 702                                               SCSI_SENSE_BUFFERSIZE,
 703                                               PCI_DMA_FROMDEVICE);
 704                 e->senseAddr = ctx->sensePA;
 705                 e->senseLen = SCSI_SENSE_BUFFERSIZE;
 706         } else {
 707                 e->senseLen  = 0;
 708                 e->senseAddr = 0;
 709         }
 710         e->cdbLen   = cmd->cmd_len;
 711         e->vcpuHint = smp_processor_id();
 712         memcpy(e->cdb, cmd->cmnd, e->cdbLen);
 713
 714         e->tag = SIMPLE_QUEUE_TAG;
 715
 716         if (cmd->sc_data_direction == DMA_FROM_DEVICE)
 717                 e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST;
 718         else if (cmd->sc_data_direction == DMA_TO_DEVICE)
 719                 e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE;
 720         else if (cmd->sc_data_direction == DMA_NONE)
 721                 e->flags = PVSCSI_FLAG_CMD_DIR_NONE;
 722         else
 723                 e->flags = 0;
 724
 725         pvscsi_map_buffers(adapter, ctx, cmd, e);
 726
 727         e->context = pvscsi_map_context(adapter, ctx);
 728
 729         barrier();
 730
 731         s->reqProdIdx++;
 732
 733         return 0;
 734 }
 735
 736 static int pvscsi_queue_lck(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
 737 {
 738         struct Scsi_Host *host = cmd->device->host;
 739         struct pvscsi_adapter *adapter = shost_priv(host);
 740         struct pvscsi_ctx *ctx;
 741         unsigned long flags;
 742         unsigned char op;
 743
 744         spin_lock_irqsave(&adapter->hw_lock, flags);
 745
 746         ctx = pvscsi_acquire_context(adapter, cmd);
 747         if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) {
 748                 if (ctx)
 749                         pvscsi_release_context(adapter, ctx);
 750                 spin_unlock_irqrestore(&adapter->hw_lock, flags);
 751                 return SCSI_MLQUEUE_HOST_BUSY;
 752         }
 753
 754         cmd->scsi_done = done;
 755         op = cmd->cmnd[0];
 756
 757         dev_dbg(&cmd->device->sdev_gendev,
 758                 "queued cmd %p, ctx %p, op=%x\n", cmd, ctx, op);
 759
 760         spin_unlock_irqrestore(&adapter->hw_lock, flags);
 761
 762         pvscsi_kick_io(adapter, op);
 763
 764         return 0;
 765 }
 766
 767 static DEF_SCSI_QCMD(pvscsi_queue)
 768
 769 static int pvscsi_abort(struct scsi_cmnd *cmd)
 770 {
 771         struct pvscsi_adapter *adapter = shost_priv(cmd->device->host);
 772         struct pvscsi_ctx *ctx;
 773         unsigned long flags;
 774         int result = SUCCESS;
 775         DECLARE_COMPLETION_ONSTACK(abort_cmp);
 776
 777         scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n",
 778                     adapter->host->host_no, cmd);
 779
 780         spin_lock_irqsave(&adapter->hw_lock, flags);
 781
 782         /*
 783          * Poll the completion ring first - we might be trying to abort
 784          * a command that is waiting to be dispatched in the completion ring.
 785          */
 786         pvscsi_process_completion_ring(adapter);
 787
 788         /*
 789          * If there is no context for the command, it either already succeeded
 790          * or else was never properly issued.  Not our problem.
 791          */
 792         ctx = pvscsi_find_context(adapter, cmd);
 793         if (!ctx) {
 794                 scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd);
 795                 goto out;
 796         }
 797
 798         /*
 799          * Mark that the command has been requested to be aborted and issue
 800          * the abort.
 801          */
 802         ctx->abort_cmp = &abort_cmp;
 803
 804         pvscsi_abort_cmd(adapter, ctx);
 805         spin_unlock_irqrestore(&adapter->hw_lock, flags);
 806         /* Wait for 2 secs for the completion. */
 807         wait_for_completion_timeout(&abort_cmp, msecs_to_jiffies(2000));
 808         spin_lock_irqsave(&adapter->hw_lock, flags);
 809
 810         if (!completion_done(&abort_cmp)) {
 811                 /*
 812                  * Failed to abort the command, unmark the fact that it
 813                  * was requested to be aborted.
 814                  */
 815                 ctx->abort_cmp = NULL;
 816                 result = FAILED;
 817                 scmd_printk(KERN_DEBUG, cmd,
 818                             "Failed to get completion for aborted cmd %p\n",
 819                             cmd);
 820                 goto out;
 821         }
 822
 823         /*
 824          * Successfully aborted the command.
 825          */
 826         cmd->result = (DID_ABORT << 16);
 827         cmd->scsi_done(cmd);
 828
 829 out:
 830         spin_unlock_irqrestore(&adapter->hw_lock, flags);
 831         return result;
 832 }
 833
 834 /*
 835  * Abort all outstanding requests.  This is only safe to use if the completion
 836  * ring will never be walked again or the device has been reset, because it
 837  * destroys the 1-1 mapping between context field passed to emulation and our
 838  * request structure.
 839  */
 840 static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
 841 {
 842         unsigned i;
 843
 844         for (i = 0; i < adapter->req_depth; i++) {
 845                 struct pvscsi_ctx *ctx = &adapter->cmd_map[i];
 846                 struct scsi_cmnd *cmd = ctx->cmd;
 847                 if (cmd) {
 848                         scmd_printk(KERN_ERR, cmd,
 849                                     "Forced reset on cmd %p\n", cmd);
 850                         pvscsi_unmap_buffers(adapter, ctx);
 851                         pvscsi_release_context(adapter, ctx);
 852                         cmd->result = (DID_RESET << 16);
 853                         cmd->scsi_done(cmd);
 854                 }
 855         }
 856 }
 857
 858 static int pvscsi_host_reset(struct scsi_cmnd *cmd)
 859 {
 860         struct Scsi_Host *host = cmd->device->host;
 861         struct pvscsi_adapter *adapter = shost_priv(host);
 862         unsigned long flags;
 863         bool use_msg;
 864
 865         scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n");
 866
 867         spin_lock_irqsave(&adapter->hw_lock, flags);
 868
 869         use_msg = adapter->use_msg;
 870
 871         if (use_msg) {
 872                 adapter->use_msg = 0;
 873                 spin_unlock_irqrestore(&adapter->hw_lock, flags);
 874
 875                 /*
 876                  * Now that we know that the ISR won't add more work on the
 877                  * workqueue we can safely flush any outstanding work.
 878                  */
 879                 flush_workqueue(adapter->workqueue);
 880                 spin_lock_irqsave(&adapter->hw_lock, flags);
 881         }
 882
 883         /*
 884          * We're going to tear down the entire ring structure and set it back
 885          * up, so stalling new requests until all completions are flushed and
 886          * the rings are back in place.
 887          */
 888
 889         pvscsi_process_request_ring(adapter);
 890
 891         ll_adapter_reset(adapter);
 892
 893         /*
 894          * Now process any completions.  Note we do this AFTER adapter reset,
 895          * which is strange, but stops races where completions get posted
 896          * between processing the ring and issuing the reset.  The backend will
 897          * not touch the ring memory after reset, so the immediately pre-reset
 898          * completion ring state is still valid.
 899          */
 900         pvscsi_process_completion_ring(adapter);
 901
 902         pvscsi_reset_all(adapter);
 903         adapter->use_msg = use_msg;
 904         pvscsi_setup_all_rings(adapter);
 905         pvscsi_unmask_intr(adapter);
 906
 907         spin_unlock_irqrestore(&adapter->hw_lock, flags);
 908
 909         return SUCCESS;
 910 }
 911
 912 static int pvscsi_bus_reset(struct scsi_cmnd *cmd)
 913 {
 914         struct Scsi_Host *host = cmd->device->host;
 915         struct pvscsi_adapter *adapter = shost_priv(host);
 916         unsigned long flags;
 917
 918         scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n");
 919
 920         /*
 921          * We don't want to queue new requests for this bus after
 922          * flushing all pending requests to emulation, since new
 923          * requests could then sneak in during this bus reset phase,
 924          * so take the lock now.
 925          */
 926         spin_lock_irqsave(&adapter->hw_lock, flags);
 927
 928         pvscsi_process_request_ring(adapter);
 929         ll_bus_reset(adapter);
 930         pvscsi_process_completion_ring(adapter);
 931
 932         spin_unlock_irqrestore(&adapter->hw_lock, flags);
 933
 934         return SUCCESS;
 935 }
 936
 937 static int pvscsi_device_reset(struct scsi_cmnd *cmd)
 938 {
 939         struct Scsi_Host *host = cmd->device->host;
 940         struct pvscsi_adapter *adapter = shost_priv(host);
 941         unsigned long flags;
 942
 943         scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n",
 944                     host->host_no, cmd->device->id);
 945
 946         /*
 947          * We don't want to queue new requests for this device after flushing
 948          * all pending requests to emulation, since new requests could then
 949          * sneak in during this device reset phase, so take the lock now.
 950          */
 951         spin_lock_irqsave(&adapter->hw_lock, flags);
 952
 953         pvscsi_process_request_ring(adapter);
 954         ll_device_reset(adapter, cmd->device->id);
 955         pvscsi_process_completion_ring(adapter);
 956
 957         spin_unlock_irqrestore(&adapter->hw_lock, flags);
 958
 959         return SUCCESS;
 960 }
 961
 962 static struct scsi_host_template pvscsi_template;
 963
 964 static const char *pvscsi_info(struct Scsi_Host *host)
 965 {
 966         struct pvscsi_adapter *adapter = shost_priv(host);
 967         static char buf[256];
 968
 969         sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: "
 970                 "%u/%u/%u pages, cmd_per_lun=%u", adapter->rev,
 971                 adapter->req_pages, adapter->cmp_pages, adapter->msg_pages,
 972                 pvscsi_template.cmd_per_lun);
 973
 974         return buf;
 975 }
 976
 977 static struct scsi_host_template pvscsi_template = {
 978         .module                         = THIS_MODULE,
 979         .name                           = "VMware PVSCSI Host Adapter",
 980         .proc_name                      = "vmw_pvscsi",
 981         .info                           = pvscsi_info,
 982         .queuecommand                   = pvscsi_queue,
 983         .this_id                        = -1,
 984         .sg_tablesize                   = PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT,
 985         .dma_boundary                   = UINT_MAX,
 986         .max_sectors                    = 0xffff,
 987         .use_clustering                 = ENABLE_CLUSTERING,
 988         .change_queue_depth             = pvscsi_change_queue_depth,
 989         .eh_abort_handler               = pvscsi_abort,
 990         .eh_device_reset_handler        = pvscsi_device_reset,
 991         .eh_bus_reset_handler           = pvscsi_bus_reset,
 992         .eh_host_reset_handler          = pvscsi_host_reset,
 993 };
 994
 995 static void pvscsi_process_msg(const struct pvscsi_adapter *adapter,
 996                                const struct PVSCSIRingMsgDesc *e)
 997 {
 998         struct PVSCSIRingsState *s = adapter->rings_state;
 999         struct Scsi_Host *host = adapter->host;
1000         struct scsi_device *sdev;
1001
1002         printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n",
1003                e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2);
1004
1005         BUILD_BUG_ON(PVSCSI_MSG_LAST != 2);
1006
1007         if (e->type == PVSCSI_MSG_DEV_ADDED) {
1008                 struct PVSCSIMsgDescDevStatusChanged *desc;
1009                 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1010
1011                 printk(KERN_INFO
1012                        "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n",
1013                        desc->bus, desc->target, desc->lun[1]);
1014
1015                 if (!scsi_host_get(host))
1016                         return;
1017
1018                 sdev = scsi_device_lookup(host, desc->bus, desc->target,
1019                                           desc->lun[1]);
1020                 if (sdev) {
1021                         printk(KERN_INFO "vmw_pvscsi: device already exists\n");
1022                         scsi_device_put(sdev);
1023                 } else
1024                         scsi_add_device(adapter->host, desc->bus,
1025                                         desc->target, desc->lun[1]);
1026
1027                 scsi_host_put(host);
1028         } else if (e->type == PVSCSI_MSG_DEV_REMOVED) {
1029                 struct PVSCSIMsgDescDevStatusChanged *desc;
1030                 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1031
1032                 printk(KERN_INFO
1033                        "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n",
1034                        desc->bus, desc->target, desc->lun[1]);
1035
1036                 if (!scsi_host_get(host))
1037                         return;
1038
1039                 sdev = scsi_device_lookup(host, desc->bus, desc->target,
1040                                           desc->lun[1]);
1041                 if (sdev) {
1042                         scsi_remove_device(sdev);
1043                         scsi_device_put(sdev);
1044                 } else
1045                         printk(KERN_INFO
1046                                "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n",
1047                                desc->bus, desc->target, desc->lun[1]);
1048
1049                 scsi_host_put(host);
1050         }
1051 }
1052
1053 static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter)
1054 {
1055         struct PVSCSIRingsState *s = adapter->rings_state;
1056
1057         return s->msgProdIdx != s->msgConsIdx;
1058 }
1059
1060 static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter)
1061 {
1062         struct PVSCSIRingsState *s = adapter->rings_state;
1063         struct PVSCSIRingMsgDesc *ring = adapter->msg_ring;
1064         u32 msg_entries = s->msgNumEntriesLog2;
1065
1066         while (pvscsi_msg_pending(adapter)) {
1067                 struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx &
1068                                                       MASK(msg_entries));
1069
1070                 barrier();
1071                 pvscsi_process_msg(adapter, e);
1072                 barrier();
1073                 s->msgConsIdx++;
1074         }
1075 }
1076
1077 static void pvscsi_msg_workqueue_handler(struct work_struct *data)
1078 {
1079         struct pvscsi_adapter *adapter;
1080
1081         adapter = container_of(data, struct pvscsi_adapter, work);
1082
1083         pvscsi_process_msg_ring(adapter);
1084 }
1085
1086 static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter)
1087 {
1088         char name[32];
1089
1090         if (!pvscsi_use_msg)
1091                 return 0;
1092
1093         pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1094                          PVSCSI_CMD_SETUP_MSG_RING);
1095
1096         if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1)
1097                 return 0;
1098
1099         snprintf(name, sizeof(name),
1100                  "vmw_pvscsi_wq_%u", adapter->host->host_no);
1101
1102         adapter->workqueue = create_singlethread_workqueue(name);
1103         if (!adapter->workqueue) {
1104                 printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n");
1105                 return 0;
1106         }
1107         INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler);
1108
1109         return 1;
1110 }
1111
1112 static bool pvscsi_setup_req_threshold(struct pvscsi_adapter *adapter,
1113                                       bool enable)
1114 {
1115         u32 val;
1116
1117         if (!pvscsi_use_req_threshold)
1118                 return false;
1119
1120         pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1121                          PVSCSI_CMD_SETUP_REQCALLTHRESHOLD);
1122         val = pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS);
1123         if (val == -1) {
1124                 printk(KERN_INFO "vmw_pvscsi: device does not support req_threshold\n");
1125                 return false;
1126         } else {
1127                 struct PVSCSICmdDescSetupReqCall cmd_msg = { 0 };
1128                 cmd_msg.enable = enable;
1129                 printk(KERN_INFO
1130                        "vmw_pvscsi: %sabling reqCallThreshold\n",
1131                         enable ? "en" : "dis");
1132                 pvscsi_write_cmd_desc(adapter,
1133                                       PVSCSI_CMD_SETUP_REQCALLTHRESHOLD,
1134                                       &cmd_msg, sizeof(cmd_msg));
1135                 return pvscsi_reg_read(adapter,
1136                                        PVSCSI_REG_OFFSET_COMMAND_STATUS) != 0;
1137         }
1138 }
1139
1140 static irqreturn_t pvscsi_isr(int irq, void *devp)
1141 {
1142         struct pvscsi_adapter *adapter = devp;
1143         int handled;
1144
1145         if (adapter->use_msi || adapter->use_msix)
1146                 handled = true;
1147         else {
1148                 u32 val = pvscsi_read_intr_status(adapter);
1149                 handled = (val & PVSCSI_INTR_ALL_SUPPORTED) != 0;
1150                 if (handled)
1151                         pvscsi_write_intr_status(devp, val);
1152         }
1153
1154         if (handled) {
1155                 unsigned long flags;
1156
1157                 spin_lock_irqsave(&adapter->hw_lock, flags);
1158
1159                 pvscsi_process_completion_ring(adapter);
1160                 if (adapter->use_msg && pvscsi_msg_pending(adapter))
1161                         queue_work(adapter->workqueue, &adapter->work);
1162
1163                 spin_unlock_irqrestore(&adapter->hw_lock, flags);
1164         }
1165
1166         return IRQ_RETVAL(handled);
1167 }
1168
1169 static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter)
1170 {
1171         struct pvscsi_ctx *ctx = adapter->cmd_map;
1172         unsigned i;
1173
1174         for (i = 0; i < adapter->req_depth; ++i, ++ctx)
1175                 free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE));
1176 }
1177
1178 static int pvscsi_setup_msix(const struct pvscsi_adapter *adapter,
1179                              unsigned int *irq)
1180 {
1181         struct msix_entry entry = { 0, PVSCSI_VECTOR_COMPLETION };
1182         int ret;
1183
1184         ret = pci_enable_msix_exact(adapter->dev, &entry, 1);
1185         if (ret)
1186                 return ret;
1187
1188         *irq = entry.vector;
1189
1190         return 0;
1191 }
1192
1193 static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter)
1194 {
1195         if (adapter->irq) {
1196                 free_irq(adapter->irq, adapter);
1197                 adapter->irq = 0;
1198         }
1199         if (adapter->use_msi) {
1200                 pci_disable_msi(adapter->dev);
1201                 adapter->use_msi = 0;
1202         } else if (adapter->use_msix) {
1203                 pci_disable_msix(adapter->dev);
1204                 adapter->use_msix = 0;
1205         }
1206 }
1207
1208 static void pvscsi_release_resources(struct pvscsi_adapter *adapter)
1209 {
1210         if (adapter->workqueue)
1211                 destroy_workqueue(adapter->workqueue);
1212
1213         if (adapter->mmioBase)
1214                 pci_iounmap(adapter->dev, adapter->mmioBase);
1215
1216         pci_release_regions(adapter->dev);
1217
1218         if (adapter->cmd_map) {
1219                 pvscsi_free_sgls(adapter);
1220                 kfree(adapter->cmd_map);
1221         }
1222
1223         if (adapter->rings_state)
1224                 pci_free_consistent(adapter->dev, PAGE_SIZE,
1225                                     adapter->rings_state, adapter->ringStatePA);
1226
1227         if (adapter->req_ring)
1228                 pci_free_consistent(adapter->dev,
1229                                     adapter->req_pages * PAGE_SIZE,
1230                                     adapter->req_ring, adapter->reqRingPA);
1231
1232         if (adapter->cmp_ring)
1233                 pci_free_consistent(adapter->dev,
1234                                     adapter->cmp_pages * PAGE_SIZE,
1235                                     adapter->cmp_ring, adapter->cmpRingPA);
1236
1237         if (adapter->msg_ring)
1238                 pci_free_consistent(adapter->dev,
1239                                     adapter->msg_pages * PAGE_SIZE,
1240                                     adapter->msg_ring, adapter->msgRingPA);
1241 }
1242
1243 /*
1244  * Allocate scatter gather lists.
1245  *
1246  * These are statically allocated.  Trying to be clever was not worth it.
1247  *
1248  * Dynamic allocation can fail, and we can't go deep into the memory
1249  * allocator, since we're a SCSI driver, and trying too hard to allocate
1250  * memory might generate disk I/O.  We also don't want to fail disk I/O
1251  * in that case because we can't get an allocation - the I/O could be
1252  * trying to swap out data to free memory.  Since that is pathological,
1253  * just use a statically allocated scatter list.
1254  *
1255  */
1256 static int pvscsi_allocate_sg(struct pvscsi_adapter *adapter)
1257 {
1258         struct pvscsi_ctx *ctx;
1259         int i;
1260
1261         ctx = adapter->cmd_map;
1262         BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE);
1263
1264         for (i = 0; i < adapter->req_depth; ++i, ++ctx) {
1265                 ctx->sgl = (void *)__get_free_pages(GFP_KERNEL,
1266                                                     get_order(SGL_SIZE));
1267                 ctx->sglPA = 0;
1268                 BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE));
1269                 if (!ctx->sgl) {
1270                         for (; i >= 0; --i, --ctx) {
1271                                 free_pages((unsigned long)ctx->sgl,
1272                                            get_order(SGL_SIZE));
1273                                 ctx->sgl = NULL;
1274                         }
1275                         return -ENOMEM;
1276                 }
1277         }
1278
1279         return 0;
1280 }
1281
1282 /*
1283  * Query the device, fetch the config info and return the
1284  * maximum number of targets on the adapter. In case of
1285  * failure due to any reason return default i.e. 16.
1286  */
1287 static u32 pvscsi_get_max_targets(struct pvscsi_adapter *adapter)
1288 {
1289         struct PVSCSICmdDescConfigCmd cmd;
1290         struct PVSCSIConfigPageHeader *header;
1291         struct device *dev;
1292         dma_addr_t configPagePA;
1293         void *config_page;
1294         u32 numPhys = 16;
1295
1296         dev = pvscsi_dev(adapter);
1297         config_page = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
1298                                            &configPagePA);
1299         if (!config_page) {
1300                 dev_warn(dev, "vmw_pvscsi: failed to allocate memory for config page\n");
1301                 goto exit;
1302         }
1303         BUG_ON(configPagePA & ~PAGE_MASK);
1304
1305         /* Fetch config info from the device. */
1306         cmd.configPageAddress = ((u64)PVSCSI_CONFIG_CONTROLLER_ADDRESS) << 32;
1307         cmd.configPageNum = PVSCSI_CONFIG_PAGE_CONTROLLER;
1308         cmd.cmpAddr = configPagePA;
1309         cmd._pad = 0;
1310
1311         /*
1312          * Mark the completion page header with error values. If the device
1313          * completes the command successfully, it sets the status values to
1314          * indicate success.
1315          */
1316         header = config_page;
1317         memset(header, 0, sizeof *header);
1318         header->hostStatus = BTSTAT_INVPARAM;
1319         header->scsiStatus = SDSTAT_CHECK;
1320
1321         pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_CONFIG, &cmd, sizeof cmd);
1322
1323         if (header->hostStatus == BTSTAT_SUCCESS &&
1324             header->scsiStatus == SDSTAT_GOOD) {
1325                 struct PVSCSIConfigPageController *config;
1326
1327                 config = config_page;
1328                 numPhys = config->numPhys;
1329         } else
1330                 dev_warn(dev, "vmw_pvscsi: PVSCSI_CMD_CONFIG failed. hostStatus = 0x%x, scsiStatus = 0x%x\n",
1331                          header->hostStatus, header->scsiStatus);
1332         pci_free_consistent(adapter->dev, PAGE_SIZE, config_page, configPagePA);
1333 exit:
1334         return numPhys;
1335 }
1336
1337 static int pvscsi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1338 {
1339         struct pvscsi_adapter *adapter;
1340         struct pvscsi_adapter adapter_temp;
1341         struct Scsi_Host *host = NULL;
1342         unsigned int i;
1343         unsigned long flags = 0;
1344         int error;
1345         u32 max_id;
1346
1347         error = -ENODEV;
1348
1349         if (pci_enable_device(pdev))
1350                 return error;
1351
1352         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0 &&
1353             pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1354                 printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n");
1355         } else if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) == 0 &&
1356                    pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)) == 0) {
1357                 printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n");
1358         } else {
1359                 printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n");
1360                 goto out_disable_device;
1361         }
1362
1363         /*
1364          * Let's use a temp pvscsi_adapter struct until we find the number of
1365          * targets on the adapter, after that we will switch to the real
1366          * allocated struct.
1367          */
1368         adapter = &adapter_temp;
1369         memset(adapter, 0, sizeof(*adapter));
1370         adapter->dev  = pdev;
1371         adapter->rev = pdev->revision;
1372
1373         if (pci_request_regions(pdev, "vmw_pvscsi")) {
1374                 printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n");
1375                 goto out_disable_device;
1376         }
1377
1378         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1379                 if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO))
1380                         continue;
1381
1382                 if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE)
1383                         continue;
1384
1385                 break;
1386         }
1387
1388         if (i == DEVICE_COUNT_RESOURCE) {
1389                 printk(KERN_ERR
1390                        "vmw_pvscsi: adapter has no suitable MMIO region\n");
1391                 goto out_release_resources_and_disable;
1392         }
1393
1394         adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE);
1395
1396         if (!adapter->mmioBase) {
1397                 printk(KERN_ERR
1398                        "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n",
1399                        i, PVSCSI_MEM_SPACE_SIZE);
1400                 goto out_release_resources_and_disable;
1401         }
1402
1403         pci_set_master(pdev);
1404
1405         /*
1406          * Ask the device for max number of targets before deciding the
1407          * default pvscsi_ring_pages value.
1408          */
1409         max_id = pvscsi_get_max_targets(adapter);
1410         printk(KERN_INFO "vmw_pvscsi: max_id: %u\n", max_id);
1411
1412         if (pvscsi_ring_pages == 0)
1413                 /*
1414                  * Set the right default value. Up to 16 it is 8, above it is
1415                  * max.
1416                  */
1417                 pvscsi_ring_pages = (max_id > 16) ?
1418                         PVSCSI_SETUP_RINGS_MAX_NUM_PAGES :
1419                         PVSCSI_DEFAULT_NUM_PAGES_PER_RING;
1420         printk(KERN_INFO
1421                "vmw_pvscsi: setting ring_pages to %d\n",
1422                pvscsi_ring_pages);
1423
1424         pvscsi_template.can_queue =
1425                 min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) *
1426                 PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
1427         pvscsi_template.cmd_per_lun =
1428                 min(pvscsi_template.can_queue, pvscsi_cmd_per_lun);
1429         host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter));
1430         if (!host) {
1431                 printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n");
1432                 goto out_release_resources_and_disable;
1433         }
1434
1435         /*
1436          * Let's use the real pvscsi_adapter struct here onwards.
1437          */
1438         adapter = shost_priv(host);
1439         memset(adapter, 0, sizeof(*adapter));
1440         adapter->dev  = pdev;
1441         adapter->host = host;
1442         /*
1443          * Copy back what we already have to the allocated adapter struct.
1444          */
1445         adapter->rev = adapter_temp.rev;
1446         adapter->mmioBase = adapter_temp.mmioBase;
1447
1448         spin_lock_init(&adapter->hw_lock);
1449         host->max_channel = 0;
1450         host->max_lun     = 1;
1451         host->max_cmd_len = 16;
1452         host->max_id      = max_id;
1453
1454         pci_set_drvdata(pdev, host);
1455
1456         ll_adapter_reset(adapter);
1457
1458         adapter->use_msg = pvscsi_setup_msg_workqueue(adapter);
1459
1460         error = pvscsi_allocate_rings(adapter);
1461         if (error) {
1462                 printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n");
1463                 goto out_release_resources;
1464         }
1465
1466         /*
1467          * From this point on we should reset the adapter if anything goes
1468          * wrong.
1469          */
1470         pvscsi_setup_all_rings(adapter);
1471
1472         adapter->cmd_map = kcalloc(adapter->req_depth,
1473                                    sizeof(struct pvscsi_ctx), GFP_KERNEL);
1474         if (!adapter->cmd_map) {
1475                 printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n");
1476                 error = -ENOMEM;
1477                 goto out_reset_adapter;
1478         }
1479
1480         INIT_LIST_HEAD(&adapter->cmd_pool);
1481         for (i = 0; i < adapter->req_depth; i++) {
1482                 struct pvscsi_ctx *ctx = adapter->cmd_map + i;
1483                 list_add(&ctx->list, &adapter->cmd_pool);
1484         }
1485
1486         error = pvscsi_allocate_sg(adapter);
1487         if (error) {
1488                 printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n");
1489                 goto out_reset_adapter;
1490         }
1491
1492         if (!pvscsi_disable_msix &&
1493             pvscsi_setup_msix(adapter, &adapter->irq) == 0) {
1494                 printk(KERN_INFO "vmw_pvscsi: using MSI-X\n");
1495                 adapter->use_msix = 1;
1496         } else if (!pvscsi_disable_msi && pci_enable_msi(pdev) == 0) {
1497                 printk(KERN_INFO "vmw_pvscsi: using MSI\n");
1498                 adapter->use_msi = 1;
1499                 adapter->irq = pdev->irq;
1500         } else {
1501                 printk(KERN_INFO "vmw_pvscsi: using INTx\n");
1502                 adapter->irq = pdev->irq;
1503                 flags = IRQF_SHARED;
1504         }
1505
1506         adapter->use_req_threshold = pvscsi_setup_req_threshold(adapter, true);
1507         printk(KERN_DEBUG "vmw_pvscsi: driver-based request coalescing %sabled\n",
1508                adapter->use_req_threshold ? "en" : "dis");
1509
1510         error = request_irq(adapter->irq, pvscsi_isr, flags,
1511                             "vmw_pvscsi", adapter);
1512         if (error) {
1513                 printk(KERN_ERR
1514                        "vmw_pvscsi: unable to request IRQ: %d\n", error);
1515                 adapter->irq = 0;
1516                 goto out_reset_adapter;
1517         }
1518
1519         error = scsi_add_host(host, &pdev->dev);
1520         if (error) {
1521                 printk(KERN_ERR
1522                        "vmw_pvscsi: scsi_add_host failed: %d\n", error);
1523                 goto out_reset_adapter;
1524         }
1525
1526         dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n",
1527                  adapter->rev, host->host_no);
1528
1529         pvscsi_unmask_intr(adapter);
1530
1531         scsi_scan_host(host);
1532
1533         return 0;
1534
1535 out_reset_adapter:
1536         ll_adapter_reset(adapter);
1537 out_release_resources:
1538         pvscsi_shutdown_intr(adapter);
1539         pvscsi_release_resources(adapter);
1540         scsi_host_put(host);
1541 out_disable_device:
1542         pci_disable_device(pdev);
1543
1544         return error;
1545
1546 out_release_resources_and_disable:
1547         pvscsi_shutdown_intr(adapter);
1548         pvscsi_release_resources(adapter);
1549         goto out_disable_device;
1550 }
1551
1552 static void __pvscsi_shutdown(struct pvscsi_adapter *adapter)
1553 {
1554         pvscsi_mask_intr(adapter);
1555
1556         if (adapter->workqueue)
1557                 flush_workqueue(adapter->workqueue);
1558
1559         pvscsi_shutdown_intr(adapter);
1560
1561         pvscsi_process_request_ring(adapter);
1562         pvscsi_process_completion_ring(adapter);
1563         ll_adapter_reset(adapter);
1564 }
1565
1566 static void pvscsi_shutdown(struct pci_dev *dev)
1567 {
1568         struct Scsi_Host *host = pci_get_drvdata(dev);
1569         struct pvscsi_adapter *adapter = shost_priv(host);
1570
1571         __pvscsi_shutdown(adapter);
1572 }
1573
1574 static void pvscsi_remove(struct pci_dev *pdev)
1575 {
1576         struct Scsi_Host *host = pci_get_drvdata(pdev);
1577         struct pvscsi_adapter *adapter = shost_priv(host);
1578
1579         scsi_remove_host(host);
1580
1581         __pvscsi_shutdown(adapter);
1582         pvscsi_release_resources(adapter);
1583
1584         scsi_host_put(host);
1585
1586         pci_disable_device(pdev);
1587 }
1588
1589 static struct pci_driver pvscsi_pci_driver = {
1590         .name           = "vmw_pvscsi",
1591         .id_table       = pvscsi_pci_tbl,
1592         .probe          = pvscsi_probe,
1593         .remove         = pvscsi_remove,
1594         .shutdown       = pvscsi_shutdown,
1595 };
1596
1597 static int __init pvscsi_init(void)
1598 {
1599         pr_info("%s - version %s\n",
1600                 PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING);
1601         return pci_register_driver(&pvscsi_pci_driver);
1602 }
1603
1604 static void __exit pvscsi_exit(void)
1605 {
1606         pci_unregister_driver(&pvscsi_pci_driver);
1607 }
1608
1609 module_init(pvscsi_init);
1610 module_exit(pvscsi_exit);