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