[SCSI] vmw_pvscsi: SCSI driver for VMware's virtual HBA.
Alok Kataria [Tue, 13 Oct 2009 21:51:05 +0000 (14:51 -0700)]
This is a driver for VMware's paravirtualized SCSI device,
which should improve disk performance for guests running
under control of VMware hypervisors that support such devices.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>

MAINTAINERS
drivers/scsi/Kconfig
drivers/scsi/Makefile
drivers/scsi/vmw_pvscsi.c [new file with mode: 0644]
drivers/scsi/vmw_pvscsi.h [new file with mode: 0644]

index 016411c..d1a5cfd 100644 (file)
@@ -5696,6 +5696,14 @@ L:     netdev@vger.kernel.org
 S:     Maintained
 F:     drivers/net/vmxnet3/
 
+VMware PVSCSI driver
+M:     Alok Kataria <akataria@vmware.com>
+M:     VMware PV-Drivers <pv-drivers@vmware.com>
+L:     linux-scsi@vger.kernel.org
+S:     Maintained
+F:     drivers/scsi/vmw_pvscsi.c
+F:     drivers/scsi/vmw_pvscsi.h
+
 VOLTAGE AND CURRENT REGULATOR FRAMEWORK
 M:     Liam Girdwood <lrg@slimlogic.co.uk>
 M:     Mark Brown <broonie@opensource.wolfsonmicro.com>
index 2e4f7d0..1895259 100644 (file)
@@ -621,6 +621,14 @@ config SCSI_FLASHPOINT
          substantial, so users of MultiMaster Host Adapters may not
          wish to include it.
 
+config VMWARE_PVSCSI
+       tristate "VMware PVSCSI driver support"
+       depends on PCI && SCSI && X86
+       help
+         This driver supports VMware's para virtualized SCSI HBA.
+         To compile this driver as a module, choose M here: the
+         module will be called vmw_pvscsi.
+
 config LIBFC
        tristate "LibFC module"
        select SCSI_FC_ATTRS
index 53b1dac..5026bdc 100644 (file)
@@ -134,6 +134,7 @@ obj-$(CONFIG_SCSI_CXGB3_ISCSI)      += libiscsi.o libiscsi_tcp.o cxgb3i/
 obj-$(CONFIG_SCSI_BNX2_ISCSI)  += libiscsi.o bnx2i/
 obj-$(CONFIG_BE2ISCSI)         += libiscsi.o be2iscsi/
 obj-$(CONFIG_SCSI_PMCRAID)     += pmcraid.o
+obj-$(CONFIG_VMWARE_PVSCSI)    += vmw_pvscsi.o
 
 obj-$(CONFIG_ARM)              += arm/
 
diff --git a/drivers/scsi/vmw_pvscsi.c b/drivers/scsi/vmw_pvscsi.c
new file mode 100644 (file)
index 0000000..d2604c8
--- /dev/null
@@ -0,0 +1,1407 @@
+/*
+ * Linux driver for VMware's para-virtualized SCSI HBA.
+ *
+ * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the
+ * Free Software Foundation; version 2 of the License and no later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ * NON INFRINGEMENT.  See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Maintained by: Alok N Kataria <akataria@vmware.com>
+ *
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/interrupt.h>
+#include <linux/workqueue.h>
+#include <linux/pci.h>
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_host.h>
+#include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_device.h>
+
+#include "vmw_pvscsi.h"
+
+#define PVSCSI_LINUX_DRIVER_DESC "VMware PVSCSI driver"
+
+MODULE_DESCRIPTION(PVSCSI_LINUX_DRIVER_DESC);
+MODULE_AUTHOR("VMware, Inc.");
+MODULE_LICENSE("GPL");
+MODULE_VERSION(PVSCSI_DRIVER_VERSION_STRING);
+
+#define PVSCSI_DEFAULT_NUM_PAGES_PER_RING      8
+#define PVSCSI_DEFAULT_NUM_PAGES_MSG_RING      1
+#define PVSCSI_DEFAULT_QUEUE_DEPTH             64
+#define SGL_SIZE                               PAGE_SIZE
+
+struct pvscsi_sg_list {
+       struct PVSCSISGElement sge[PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT];
+};
+
+struct pvscsi_ctx {
+       /*
+        * The index of the context in cmd_map serves as the context ID for a
+        * 1-to-1 mapping completions back to requests.
+        */
+       struct scsi_cmnd        *cmd;
+       struct pvscsi_sg_list   *sgl;
+       struct list_head        list;
+       dma_addr_t              dataPA;
+       dma_addr_t              sensePA;
+       dma_addr_t              sglPA;
+};
+
+struct pvscsi_adapter {
+       char                            *mmioBase;
+       unsigned int                    irq;
+       u8                              rev;
+       bool                            use_msi;
+       bool                            use_msix;
+       bool                            use_msg;
+
+       spinlock_t                      hw_lock;
+
+       struct workqueue_struct         *workqueue;
+       struct work_struct              work;
+
+       struct PVSCSIRingReqDesc        *req_ring;
+       unsigned                        req_pages;
+       unsigned                        req_depth;
+       dma_addr_t                      reqRingPA;
+
+       struct PVSCSIRingCmpDesc        *cmp_ring;
+       unsigned                        cmp_pages;
+       dma_addr_t                      cmpRingPA;
+
+       struct PVSCSIRingMsgDesc        *msg_ring;
+       unsigned                        msg_pages;
+       dma_addr_t                      msgRingPA;
+
+       struct PVSCSIRingsState         *rings_state;
+       dma_addr_t                      ringStatePA;
+
+       struct pci_dev                  *dev;
+       struct Scsi_Host                *host;
+
+       struct list_head                cmd_pool;
+       struct pvscsi_ctx               *cmd_map;
+};
+
+
+/* Command line parameters */
+static int pvscsi_ring_pages     = PVSCSI_DEFAULT_NUM_PAGES_PER_RING;
+static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING;
+static int pvscsi_cmd_per_lun    = PVSCSI_DEFAULT_QUEUE_DEPTH;
+static bool pvscsi_disable_msi;
+static bool pvscsi_disable_msix;
+static bool pvscsi_use_msg       = true;
+
+#define PVSCSI_RW (S_IRUSR | S_IWUSR)
+
+module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW);
+MODULE_PARM_DESC(ring_pages, "Number of pages per req/cmp ring - (default="
+                __stringify(PVSCSI_DEFAULT_NUM_PAGES_PER_RING) ")");
+
+module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW);
+MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default="
+                __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")");
+
+module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW);
+MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default="
+                __stringify(PVSCSI_MAX_REQ_QUEUE_DEPTH) ")");
+
+module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW);
+MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)");
+
+module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW);
+MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)");
+
+module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW);
+MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)");
+
+static const struct pci_device_id pvscsi_pci_tbl[] = {
+       { PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) },
+       { 0 }
+};
+
+MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl);
+
+static struct device *
+pvscsi_dev(const struct pvscsi_adapter *adapter)
+{
+       return &(adapter->dev->dev);
+}
+
+static struct pvscsi_ctx *
+pvscsi_find_context(const struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
+{
+       struct pvscsi_ctx *ctx, *end;
+
+       end = &adapter->cmd_map[adapter->req_depth];
+       for (ctx = adapter->cmd_map; ctx < end; ctx++)
+               if (ctx->cmd == cmd)
+                       return ctx;
+
+       return NULL;
+}
+
+static struct pvscsi_ctx *
+pvscsi_acquire_context(struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
+{
+       struct pvscsi_ctx *ctx;
+
+       if (list_empty(&adapter->cmd_pool))
+               return NULL;
+
+       ctx = list_first_entry(&adapter->cmd_pool, struct pvscsi_ctx, list);
+       ctx->cmd = cmd;
+       list_del(&ctx->list);
+
+       return ctx;
+}
+
+static void pvscsi_release_context(struct pvscsi_adapter *adapter,
+                                  struct pvscsi_ctx *ctx)
+{
+       ctx->cmd = NULL;
+       list_add(&ctx->list, &adapter->cmd_pool);
+}
+
+/*
+ * Map a pvscsi_ctx struct to a context ID field value; we map to a simple
+ * non-zero integer. ctx always points to an entry in cmd_map array, hence
+ * the return value is always >=1.
+ */
+static u64 pvscsi_map_context(const struct pvscsi_adapter *adapter,
+                             const struct pvscsi_ctx *ctx)
+{
+       return ctx - adapter->cmd_map + 1;
+}
+
+static struct pvscsi_ctx *
+pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context)
+{
+       return &adapter->cmd_map[context - 1];
+}
+
+static void pvscsi_reg_write(const struct pvscsi_adapter *adapter,
+                            u32 offset, u32 val)
+{
+       writel(val, adapter->mmioBase + offset);
+}
+
+static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset)
+{
+       return readl(adapter->mmioBase + offset);
+}
+
+static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter)
+{
+       return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS);
+}
+
+static void pvscsi_write_intr_status(const struct pvscsi_adapter *adapter,
+                                    u32 val)
+{
+       pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_STATUS, val);
+}
+
+static void pvscsi_unmask_intr(const struct pvscsi_adapter *adapter)
+{
+       u32 intr_bits;
+
+       intr_bits = PVSCSI_INTR_CMPL_MASK;
+       if (adapter->use_msg)
+               intr_bits |= PVSCSI_INTR_MSG_MASK;
+
+       pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, intr_bits);
+}
+
+static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter)
+{
+       pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0);
+}
+
+static void pvscsi_write_cmd_desc(const struct pvscsi_adapter *adapter,
+                                 u32 cmd, const void *desc, size_t len)
+{
+       const u32 *ptr = desc;
+       size_t i;
+
+       len /= sizeof(*ptr);
+       pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, cmd);
+       for (i = 0; i < len; i++)
+               pvscsi_reg_write(adapter,
+                                PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]);
+}
+
+static void pvscsi_abort_cmd(const struct pvscsi_adapter *adapter,
+                            const struct pvscsi_ctx *ctx)
+{
+       struct PVSCSICmdDescAbortCmd cmd = { 0 };
+
+       cmd.target = ctx->cmd->device->id;
+       cmd.context = pvscsi_map_context(adapter, ctx);
+
+       pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ABORT_CMD, &cmd, sizeof(cmd));
+}
+
+static void pvscsi_kick_rw_io(const struct pvscsi_adapter *adapter)
+{
+       pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_RW_IO, 0);
+}
+
+static void pvscsi_process_request_ring(const struct pvscsi_adapter *adapter)
+{
+       pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_NON_RW_IO, 0);
+}
+
+static int scsi_is_rw(unsigned char op)
+{
+       return op == READ_6  || op == WRITE_6 ||
+              op == READ_10 || op == WRITE_10 ||
+              op == READ_12 || op == WRITE_12 ||
+              op == READ_16 || op == WRITE_16;
+}
+
+static void pvscsi_kick_io(const struct pvscsi_adapter *adapter,
+                          unsigned char op)
+{
+       if (scsi_is_rw(op))
+               pvscsi_kick_rw_io(adapter);
+       else
+               pvscsi_process_request_ring(adapter);
+}
+
+static void ll_adapter_reset(const struct pvscsi_adapter *adapter)
+{
+       dev_dbg(pvscsi_dev(adapter), "Adapter Reset on %p\n", adapter);
+
+       pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ADAPTER_RESET, NULL, 0);
+}
+
+static void ll_bus_reset(const struct pvscsi_adapter *adapter)
+{
+       dev_dbg(pvscsi_dev(adapter), "Reseting bus on %p\n", adapter);
+
+       pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_BUS, NULL, 0);
+}
+
+static void ll_device_reset(const struct pvscsi_adapter *adapter, u32 target)
+{
+       struct PVSCSICmdDescResetDevice cmd = { 0 };
+
+       dev_dbg(pvscsi_dev(adapter), "Reseting device: target=%u\n", target);
+
+       cmd.target = target;
+
+       pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_DEVICE,
+                             &cmd, sizeof(cmd));
+}
+
+static void pvscsi_create_sg(struct pvscsi_ctx *ctx,
+                            struct scatterlist *sg, unsigned count)
+{
+       unsigned i;
+       struct PVSCSISGElement *sge;
+
+       BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT);
+
+       sge = &ctx->sgl->sge[0];
+       for (i = 0; i < count; i++, sg++) {
+               sge[i].addr   = sg_dma_address(sg);
+               sge[i].length = sg_dma_len(sg);
+               sge[i].flags  = 0;
+       }
+}
+
+/*
+ * Map all data buffers for a command into PCI space and
+ * setup the scatter/gather list if needed.
+ */
+static void pvscsi_map_buffers(struct pvscsi_adapter *adapter,
+                              struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd,
+                              struct PVSCSIRingReqDesc *e)
+{
+       unsigned count;
+       unsigned bufflen = scsi_bufflen(cmd);
+       struct scatterlist *sg;
+
+       e->dataLen = bufflen;
+       e->dataAddr = 0;
+       if (bufflen == 0)
+               return;
+
+       sg = scsi_sglist(cmd);
+       count = scsi_sg_count(cmd);
+       if (count != 0) {
+               int segs = scsi_dma_map(cmd);
+               if (segs > 1) {
+                       pvscsi_create_sg(ctx, sg, segs);
+
+                       e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST;
+                       ctx->sglPA = pci_map_single(adapter->dev, ctx->sgl,
+                                                   SGL_SIZE, PCI_DMA_TODEVICE);
+                       e->dataAddr = ctx->sglPA;
+               } else
+                       e->dataAddr = sg_dma_address(sg);
+       } else {
+               /*
+                * In case there is no S/G list, scsi_sglist points
+                * directly to the buffer.
+                */
+               ctx->dataPA = pci_map_single(adapter->dev, sg, bufflen,
+                                            cmd->sc_data_direction);
+               e->dataAddr = ctx->dataPA;
+       }
+}
+
+static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
+                                struct pvscsi_ctx *ctx)
+{
+       struct scsi_cmnd *cmd;
+       unsigned bufflen;
+
+       cmd = ctx->cmd;
+       bufflen = scsi_bufflen(cmd);
+
+       if (bufflen != 0) {
+               unsigned count = scsi_sg_count(cmd);
+
+               if (count != 0) {
+                       scsi_dma_unmap(cmd);
+                       if (ctx->sglPA) {
+                               pci_unmap_single(adapter->dev, ctx->sglPA,
+                                                SGL_SIZE, PCI_DMA_TODEVICE);
+                               ctx->sglPA = 0;
+                       }
+               } else
+                       pci_unmap_single(adapter->dev, ctx->dataPA, bufflen,
+                                        cmd->sc_data_direction);
+       }
+       if (cmd->sense_buffer)
+               pci_unmap_single(adapter->dev, ctx->sensePA,
+                                SCSI_SENSE_BUFFERSIZE, PCI_DMA_FROMDEVICE);
+}
+
+static int __devinit pvscsi_allocate_rings(struct pvscsi_adapter *adapter)
+{
+       adapter->rings_state = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
+                                                   &adapter->ringStatePA);
+       if (!adapter->rings_state)
+               return -ENOMEM;
+
+       adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING,
+                                pvscsi_ring_pages);
+       adapter->req_depth = adapter->req_pages
+                                       * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
+       adapter->req_ring = pci_alloc_consistent(adapter->dev,
+                                                adapter->req_pages * PAGE_SIZE,
+                                                &adapter->reqRingPA);
+       if (!adapter->req_ring)
+               return -ENOMEM;
+
+       adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING,
+                                pvscsi_ring_pages);
+       adapter->cmp_ring = pci_alloc_consistent(adapter->dev,
+                                                adapter->cmp_pages * PAGE_SIZE,
+                                                &adapter->cmpRingPA);
+       if (!adapter->cmp_ring)
+               return -ENOMEM;
+
+       BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE));
+       BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE));
+       BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE));
+
+       if (!adapter->use_msg)
+               return 0;
+
+       adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING,
+                                pvscsi_msg_ring_pages);
+       adapter->msg_ring = pci_alloc_consistent(adapter->dev,
+                                                adapter->msg_pages * PAGE_SIZE,
+                                                &adapter->msgRingPA);
+       if (!adapter->msg_ring)
+               return -ENOMEM;
+       BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE));
+
+       return 0;
+}
+
+static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter)
+{
+       struct PVSCSICmdDescSetupRings cmd = { 0 };
+       dma_addr_t base;
+       unsigned i;
+
+       cmd.ringsStatePPN   = adapter->ringStatePA >> PAGE_SHIFT;
+       cmd.reqRingNumPages = adapter->req_pages;
+       cmd.cmpRingNumPages = adapter->cmp_pages;
+
+       base = adapter->reqRingPA;
+       for (i = 0; i < adapter->req_pages; i++) {
+               cmd.reqRingPPNs[i] = base >> PAGE_SHIFT;
+               base += PAGE_SIZE;
+       }
+
+       base = adapter->cmpRingPA;
+       for (i = 0; i < adapter->cmp_pages; i++) {
+               cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT;
+               base += PAGE_SIZE;
+       }
+
+       memset(adapter->rings_state, 0, PAGE_SIZE);
+       memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE);
+       memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE);
+
+       pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS,
+                             &cmd, sizeof(cmd));
+
+       if (adapter->use_msg) {
+               struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 };
+
+               cmd_msg.numPages = adapter->msg_pages;
+
+               base = adapter->msgRingPA;
+               for (i = 0; i < adapter->msg_pages; i++) {
+                       cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT;
+                       base += PAGE_SIZE;
+               }
+               memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE);
+
+               pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING,
+                                     &cmd_msg, sizeof(cmd_msg));
+       }
+}
+
+/*
+ * Pull a completion descriptor off and pass the completion back
+ * to the SCSI mid layer.
+ */
+static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
+                                   const struct PVSCSIRingCmpDesc *e)
+{
+       struct pvscsi_ctx *ctx;
+       struct scsi_cmnd *cmd;
+       u32 btstat = e->hostStatus;
+       u32 sdstat = e->scsiStatus;
+
+       ctx = pvscsi_get_context(adapter, e->context);
+       cmd = ctx->cmd;
+       pvscsi_unmap_buffers(adapter, ctx);
+       pvscsi_release_context(adapter, ctx);
+       cmd->result = 0;
+
+       if (sdstat != SAM_STAT_GOOD &&
+           (btstat == BTSTAT_SUCCESS ||
+            btstat == BTSTAT_LINKED_COMMAND_COMPLETED ||
+            btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) {
+               cmd->result = (DID_OK << 16) | sdstat;
+               if (sdstat == SAM_STAT_CHECK_CONDITION && cmd->sense_buffer)
+                       cmd->result |= (DRIVER_SENSE << 24);
+       } else
+               switch (btstat) {
+               case BTSTAT_SUCCESS:
+               case BTSTAT_LINKED_COMMAND_COMPLETED:
+               case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG:
+                       /* If everything went fine, let's move on..  */
+                       cmd->result = (DID_OK << 16);
+                       break;
+
+               case BTSTAT_DATARUN:
+               case BTSTAT_DATA_UNDERRUN:
+                       /* Report residual data in underruns */
+                       scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
+                       cmd->result = (DID_ERROR << 16);
+                       break;
+
+               case BTSTAT_SELTIMEO:
+                       /* Our emulation returns this for non-connected devs */
+                       cmd->result = (DID_BAD_TARGET << 16);
+                       break;
+
+               case BTSTAT_LUNMISMATCH:
+               case BTSTAT_TAGREJECT:
+               case BTSTAT_BADMSG:
+                       cmd->result = (DRIVER_INVALID << 24);
+                       /* fall through */
+
+               case BTSTAT_HAHARDWARE:
+               case BTSTAT_INVPHASE:
+               case BTSTAT_HATIMEOUT:
+               case BTSTAT_NORESPONSE:
+               case BTSTAT_DISCONNECT:
+               case BTSTAT_HASOFTWARE:
+               case BTSTAT_BUSFREE:
+               case BTSTAT_SENSFAILED:
+                       cmd->result |= (DID_ERROR << 16);
+                       break;
+
+               case BTSTAT_SENTRST:
+               case BTSTAT_RECVRST:
+               case BTSTAT_BUSRESET:
+                       cmd->result = (DID_RESET << 16);
+                       break;
+
+               case BTSTAT_ABORTQUEUE:
+                       cmd->result = (DID_ABORT << 16);
+                       break;
+
+               case BTSTAT_SCSIPARITY:
+                       cmd->result = (DID_PARITY << 16);
+                       break;
+
+               default:
+                       cmd->result = (DID_ERROR << 16);
+                       scmd_printk(KERN_DEBUG, cmd,
+                                   "Unknown completion status: 0x%x\n",
+                                   btstat);
+       }
+
+       dev_dbg(&cmd->device->sdev_gendev,
+               "cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n",
+               cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat);
+
+       cmd->scsi_done(cmd);
+}
+
+/*
+ * barrier usage : Since the PVSCSI device is emulated, there could be cases
+ * where we may want to serialize some accesses between the driver and the
+ * emulation layer. We use compiler barriers instead of the more expensive
+ * memory barriers because PVSCSI is only supported on X86 which has strong
+ * memory access ordering.
+ */
+static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter)
+{
+       struct PVSCSIRingsState *s = adapter->rings_state;
+       struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring;
+       u32 cmp_entries = s->cmpNumEntriesLog2;
+
+       while (s->cmpConsIdx != s->cmpProdIdx) {
+               struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx &
+                                                     MASK(cmp_entries));
+               /*
+                * This barrier() ensures that *e is not dereferenced while
+                * the device emulation still writes data into the slot.
+                * Since the device emulation advances s->cmpProdIdx only after
+                * updating the slot we want to check it first.
+                */
+               barrier();
+               pvscsi_complete_request(adapter, e);
+               /*
+                * This barrier() ensures that compiler doesn't reorder write
+                * to s->cmpConsIdx before the read of (*e) inside
+                * pvscsi_complete_request. Otherwise, device emulation may
+                * overwrite *e before we had a chance to read it.
+                */
+               barrier();
+               s->cmpConsIdx++;
+       }
+}
+
+/*
+ * Translate a Linux SCSI request into a request ring entry.
+ */
+static int pvscsi_queue_ring(struct pvscsi_adapter *adapter,
+                            struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd)
+{
+       struct PVSCSIRingsState *s;
+       struct PVSCSIRingReqDesc *e;
+       struct scsi_device *sdev;
+       u32 req_entries;
+
+       s = adapter->rings_state;
+       sdev = cmd->device;
+       req_entries = s->reqNumEntriesLog2;
+
+       /*
+        * If this condition holds, we might have room on the request ring, but
+        * we might not have room on the completion ring for the response.
+        * However, we have already ruled out this possibility - we would not
+        * have successfully allocated a context if it were true, since we only
+        * have one context per request entry.  Check for it anyway, since it
+        * would be a serious bug.
+        */
+       if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) {
+               scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: "
+                           "ring full: reqProdIdx=%d cmpConsIdx=%d\n",
+                           s->reqProdIdx, s->cmpConsIdx);
+               return -1;
+       }
+
+       e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries));
+
+       e->bus    = sdev->channel;
+       e->target = sdev->id;
+       memset(e->lun, 0, sizeof(e->lun));
+       e->lun[1] = sdev->lun;
+
+       if (cmd->sense_buffer) {
+               ctx->sensePA = pci_map_single(adapter->dev, cmd->sense_buffer,
+                                             SCSI_SENSE_BUFFERSIZE,
+                                             PCI_DMA_FROMDEVICE);
+               e->senseAddr = ctx->sensePA;
+               e->senseLen = SCSI_SENSE_BUFFERSIZE;
+       } else {
+               e->senseLen  = 0;
+               e->senseAddr = 0;
+       }
+       e->cdbLen   = cmd->cmd_len;
+       e->vcpuHint = smp_processor_id();
+       memcpy(e->cdb, cmd->cmnd, e->cdbLen);
+
+       e->tag = SIMPLE_QUEUE_TAG;
+       if (sdev->tagged_supported &&
+           (cmd->tag == HEAD_OF_QUEUE_TAG ||
+            cmd->tag == ORDERED_QUEUE_TAG))
+               e->tag = cmd->tag;
+
+       if (cmd->sc_data_direction == DMA_FROM_DEVICE)
+               e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST;
+       else if (cmd->sc_data_direction == DMA_TO_DEVICE)
+               e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE;
+       else if (cmd->sc_data_direction == DMA_NONE)
+               e->flags = PVSCSI_FLAG_CMD_DIR_NONE;
+       else
+               e->flags = 0;
+
+       pvscsi_map_buffers(adapter, ctx, cmd, e);
+
+       e->context = pvscsi_map_context(adapter, ctx);
+
+       barrier();
+
+       s->reqProdIdx++;
+
+       return 0;
+}
+
+static int pvscsi_queue(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
+{
+       struct Scsi_Host *host = cmd->device->host;
+       struct pvscsi_adapter *adapter = shost_priv(host);
+       struct pvscsi_ctx *ctx;
+       unsigned long flags;
+
+       spin_lock_irqsave(&adapter->hw_lock, flags);
+
+       ctx = pvscsi_acquire_context(adapter, cmd);
+       if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) {
+               if (ctx)
+                       pvscsi_release_context(adapter, ctx);
+               spin_unlock_irqrestore(&adapter->hw_lock, flags);
+               return SCSI_MLQUEUE_HOST_BUSY;
+       }
+
+       cmd->scsi_done = done;
+
+       dev_dbg(&cmd->device->sdev_gendev,
+               "queued cmd %p, ctx %p, op=%x\n", cmd, ctx, cmd->cmnd[0]);
+
+       spin_unlock_irqrestore(&adapter->hw_lock, flags);
+
+       pvscsi_kick_io(adapter, cmd->cmnd[0]);
+
+       return 0;
+}
+
+static int pvscsi_abort(struct scsi_cmnd *cmd)
+{
+       struct pvscsi_adapter *adapter = shost_priv(cmd->device->host);
+       struct pvscsi_ctx *ctx;
+       unsigned long flags;
+
+       scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n",
+                   adapter->host->host_no, cmd);
+
+       spin_lock_irqsave(&adapter->hw_lock, flags);
+
+       /*
+        * Poll the completion ring first - we might be trying to abort
+        * a command that is waiting to be dispatched in the completion ring.
+        */
+       pvscsi_process_completion_ring(adapter);
+
+       /*
+        * If there is no context for the command, it either already succeeded
+        * or else was never properly issued.  Not our problem.
+        */
+       ctx = pvscsi_find_context(adapter, cmd);
+       if (!ctx) {
+               scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd);
+               goto out;
+       }
+
+       pvscsi_abort_cmd(adapter, ctx);
+
+       pvscsi_process_completion_ring(adapter);
+
+out:
+       spin_unlock_irqrestore(&adapter->hw_lock, flags);
+       return SUCCESS;
+}
+
+/*
+ * Abort all outstanding requests.  This is only safe to use if the completion
+ * ring will never be walked again or the device has been reset, because it
+ * destroys the 1-1 mapping between context field passed to emulation and our
+ * request structure.
+ */
+static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
+{
+       unsigned i;
+
+       for (i = 0; i < adapter->req_depth; i++) {
+               struct pvscsi_ctx *ctx = &adapter->cmd_map[i];
+               struct scsi_cmnd *cmd = ctx->cmd;
+               if (cmd) {
+                       scmd_printk(KERN_ERR, cmd,
+                                   "Forced reset on cmd %p\n", cmd);
+                       pvscsi_unmap_buffers(adapter, ctx);
+                       pvscsi_release_context(adapter, ctx);
+                       cmd->result = (DID_RESET << 16);
+                       cmd->scsi_done(cmd);
+               }
+       }
+}
+
+static int pvscsi_host_reset(struct scsi_cmnd *cmd)
+{
+       struct Scsi_Host *host = cmd->device->host;
+       struct pvscsi_adapter *adapter = shost_priv(host);
+       unsigned long flags;
+       bool use_msg;
+
+       scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n");
+
+       spin_lock_irqsave(&adapter->hw_lock, flags);
+
+       use_msg = adapter->use_msg;
+
+       if (use_msg) {
+               adapter->use_msg = 0;
+               spin_unlock_irqrestore(&adapter->hw_lock, flags);
+
+               /*
+                * Now that we know that the ISR won't add more work on the
+                * workqueue we can safely flush any outstanding work.
+                */
+               flush_workqueue(adapter->workqueue);
+               spin_lock_irqsave(&adapter->hw_lock, flags);
+       }
+
+       /*
+        * We're going to tear down the entire ring structure and set it back
+        * up, so stalling new requests until all completions are flushed and
+        * the rings are back in place.
+        */
+
+       pvscsi_process_request_ring(adapter);
+
+       ll_adapter_reset(adapter);
+
+       /*
+        * Now process any completions.  Note we do this AFTER adapter reset,
+        * which is strange, but stops races where completions get posted
+        * between processing the ring and issuing the reset.  The backend will
+        * not touch the ring memory after reset, so the immediately pre-reset
+        * completion ring state is still valid.
+        */
+       pvscsi_process_completion_ring(adapter);
+
+       pvscsi_reset_all(adapter);
+       adapter->use_msg = use_msg;
+       pvscsi_setup_all_rings(adapter);
+       pvscsi_unmask_intr(adapter);
+
+       spin_unlock_irqrestore(&adapter->hw_lock, flags);
+
+       return SUCCESS;
+}
+
+static int pvscsi_bus_reset(struct scsi_cmnd *cmd)
+{
+       struct Scsi_Host *host = cmd->device->host;
+       struct pvscsi_adapter *adapter = shost_priv(host);
+       unsigned long flags;
+
+       scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n");
+
+       /*
+        * We don't want to queue new requests for this bus after
+        * flushing all pending requests to emulation, since new
+        * requests could then sneak in during this bus reset phase,
+        * so take the lock now.
+        */
+       spin_lock_irqsave(&adapter->hw_lock, flags);
+
+       pvscsi_process_request_ring(adapter);
+       ll_bus_reset(adapter);
+       pvscsi_process_completion_ring(adapter);
+
+       spin_unlock_irqrestore(&adapter->hw_lock, flags);
+
+       return SUCCESS;
+}
+
+static int pvscsi_device_reset(struct scsi_cmnd *cmd)
+{
+       struct Scsi_Host *host = cmd->device->host;
+       struct pvscsi_adapter *adapter = shost_priv(host);
+       unsigned long flags;
+
+       scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n",
+                   host->host_no, cmd->device->id);
+
+       /*
+        * We don't want to queue new requests for this device after flushing
+        * all pending requests to emulation, since new requests could then
+        * sneak in during this device reset phase, so take the lock now.
+        */
+       spin_lock_irqsave(&adapter->hw_lock, flags);
+
+       pvscsi_process_request_ring(adapter);
+       ll_device_reset(adapter, cmd->device->id);
+       pvscsi_process_completion_ring(adapter);
+
+       spin_unlock_irqrestore(&adapter->hw_lock, flags);
+
+       return SUCCESS;
+}
+
+static struct scsi_host_template pvscsi_template;
+
+static const char *pvscsi_info(struct Scsi_Host *host)
+{
+       struct pvscsi_adapter *adapter = shost_priv(host);
+       static char buf[256];
+
+       sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: "
+               "%u/%u/%u pages, cmd_per_lun=%u", adapter->rev,
+               adapter->req_pages, adapter->cmp_pages, adapter->msg_pages,
+               pvscsi_template.cmd_per_lun);
+
+       return buf;
+}
+
+static struct scsi_host_template pvscsi_template = {
+       .module                         = THIS_MODULE,
+       .name                           = "VMware PVSCSI Host Adapter",
+       .proc_name                      = "vmw_pvscsi",
+       .info                           = pvscsi_info,
+       .queuecommand                   = pvscsi_queue,
+       .this_id                        = -1,
+       .sg_tablesize                   = PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT,
+       .dma_boundary                   = UINT_MAX,
+       .max_sectors                    = 0xffff,
+       .use_clustering                 = ENABLE_CLUSTERING,
+       .eh_abort_handler               = pvscsi_abort,
+       .eh_device_reset_handler        = pvscsi_device_reset,
+       .eh_bus_reset_handler           = pvscsi_bus_reset,
+       .eh_host_reset_handler          = pvscsi_host_reset,
+};
+
+static void pvscsi_process_msg(const struct pvscsi_adapter *adapter,
+                              const struct PVSCSIRingMsgDesc *e)
+{
+       struct PVSCSIRingsState *s = adapter->rings_state;
+       struct Scsi_Host *host = adapter->host;
+       struct scsi_device *sdev;
+
+       printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n",
+              e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2);
+
+       BUILD_BUG_ON(PVSCSI_MSG_LAST != 2);
+
+       if (e->type == PVSCSI_MSG_DEV_ADDED) {
+               struct PVSCSIMsgDescDevStatusChanged *desc;
+               desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
+
+               printk(KERN_INFO
+                      "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n",
+                      desc->bus, desc->target, desc->lun[1]);
+
+               if (!scsi_host_get(host))
+                       return;
+
+               sdev = scsi_device_lookup(host, desc->bus, desc->target,
+                                         desc->lun[1]);
+               if (sdev) {
+                       printk(KERN_INFO "vmw_pvscsi: device already exists\n");
+                       scsi_device_put(sdev);
+               } else
+                       scsi_add_device(adapter->host, desc->bus,
+                                       desc->target, desc->lun[1]);
+
+               scsi_host_put(host);
+       } else if (e->type == PVSCSI_MSG_DEV_REMOVED) {
+               struct PVSCSIMsgDescDevStatusChanged *desc;
+               desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
+
+               printk(KERN_INFO
+                      "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n",
+                      desc->bus, desc->target, desc->lun[1]);
+
+               if (!scsi_host_get(host))
+                       return;
+
+               sdev = scsi_device_lookup(host, desc->bus, desc->target,
+                                         desc->lun[1]);
+               if (sdev) {
+                       scsi_remove_device(sdev);
+                       scsi_device_put(sdev);
+               } else
+                       printk(KERN_INFO
+                              "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n",
+                              desc->bus, desc->target, desc->lun[1]);
+
+               scsi_host_put(host);
+       }
+}
+
+static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter)
+{
+       struct PVSCSIRingsState *s = adapter->rings_state;
+
+       return s->msgProdIdx != s->msgConsIdx;
+}
+
+static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter)
+{
+       struct PVSCSIRingsState *s = adapter->rings_state;
+       struct PVSCSIRingMsgDesc *ring = adapter->msg_ring;
+       u32 msg_entries = s->msgNumEntriesLog2;
+
+       while (pvscsi_msg_pending(adapter)) {
+               struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx &
+                                                     MASK(msg_entries));
+
+               barrier();
+               pvscsi_process_msg(adapter, e);
+               barrier();
+               s->msgConsIdx++;
+       }
+}
+
+static void pvscsi_msg_workqueue_handler(struct work_struct *data)
+{
+       struct pvscsi_adapter *adapter;
+
+       adapter = container_of(data, struct pvscsi_adapter, work);
+
+       pvscsi_process_msg_ring(adapter);
+}
+
+static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter)
+{
+       char name[32];
+
+       if (!pvscsi_use_msg)
+               return 0;
+
+       pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
+                        PVSCSI_CMD_SETUP_MSG_RING);
+
+       if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1)
+               return 0;
+
+       snprintf(name, sizeof(name),
+                "vmw_pvscsi_wq_%u", adapter->host->host_no);
+
+       adapter->workqueue = create_singlethread_workqueue(name);
+       if (!adapter->workqueue) {
+               printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n");
+               return 0;
+       }
+       INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler);
+
+       return 1;
+}
+
+static irqreturn_t pvscsi_isr(int irq, void *devp)
+{
+       struct pvscsi_adapter *adapter = devp;
+       int handled;
+
+       if (adapter->use_msi || adapter->use_msix)
+               handled = true;
+       else {
+               u32 val = pvscsi_read_intr_status(adapter);
+               handled = (val & PVSCSI_INTR_ALL_SUPPORTED) != 0;
+               if (handled)
+                       pvscsi_write_intr_status(devp, val);
+       }
+
+       if (handled) {
+               unsigned long flags;
+
+               spin_lock_irqsave(&adapter->hw_lock, flags);
+
+               pvscsi_process_completion_ring(adapter);
+               if (adapter->use_msg && pvscsi_msg_pending(adapter))
+                       queue_work(adapter->workqueue, &adapter->work);
+
+               spin_unlock_irqrestore(&adapter->hw_lock, flags);
+       }
+
+       return IRQ_RETVAL(handled);
+}
+
+static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter)
+{
+       struct pvscsi_ctx *ctx = adapter->cmd_map;
+       unsigned i;
+
+       for (i = 0; i < adapter->req_depth; ++i, ++ctx)
+               free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE));
+}
+
+static int pvscsi_setup_msix(const struct pvscsi_adapter *adapter, int *irq)
+{
+       struct msix_entry entry = { 0, PVSCSI_VECTOR_COMPLETION };
+       int ret;
+
+       ret = pci_enable_msix(adapter->dev, &entry, 1);
+       if (ret)
+               return ret;
+
+       *irq = entry.vector;
+
+       return 0;
+}
+
+static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter)
+{
+       if (adapter->irq) {
+               free_irq(adapter->irq, adapter);
+               adapter->irq = 0;
+       }
+       if (adapter->use_msi) {
+               pci_disable_msi(adapter->dev);
+               adapter->use_msi = 0;
+       } else if (adapter->use_msix) {
+               pci_disable_msix(adapter->dev);
+               adapter->use_msix = 0;
+       }
+}
+
+static void pvscsi_release_resources(struct pvscsi_adapter *adapter)
+{
+       pvscsi_shutdown_intr(adapter);
+
+       if (adapter->workqueue)
+               destroy_workqueue(adapter->workqueue);
+
+       if (adapter->mmioBase)
+               pci_iounmap(adapter->dev, adapter->mmioBase);
+
+       pci_release_regions(adapter->dev);
+
+       if (adapter->cmd_map) {
+               pvscsi_free_sgls(adapter);
+               kfree(adapter->cmd_map);
+       }
+
+       if (adapter->rings_state)
+               pci_free_consistent(adapter->dev, PAGE_SIZE,
+                                   adapter->rings_state, adapter->ringStatePA);
+
+       if (adapter->req_ring)
+               pci_free_consistent(adapter->dev,
+                                   adapter->req_pages * PAGE_SIZE,
+                                   adapter->req_ring, adapter->reqRingPA);
+
+       if (adapter->cmp_ring)
+               pci_free_consistent(adapter->dev,
+                                   adapter->cmp_pages * PAGE_SIZE,
+                                   adapter->cmp_ring, adapter->cmpRingPA);
+
+       if (adapter->msg_ring)
+               pci_free_consistent(adapter->dev,
+                                   adapter->msg_pages * PAGE_SIZE,
+                                   adapter->msg_ring, adapter->msgRingPA);
+}
+
+/*
+ * Allocate scatter gather lists.
+ *
+ * These are statically allocated.  Trying to be clever was not worth it.
+ *
+ * Dynamic allocation can fail, and we can't go deeep into the memory
+ * allocator, since we're a SCSI driver, and trying too hard to allocate
+ * memory might generate disk I/O.  We also don't want to fail disk I/O
+ * in that case because we can't get an allocation - the I/O could be
+ * trying to swap out data to free memory.  Since that is pathological,
+ * just use a statically allocated scatter list.
+ *
+ */
+static int __devinit pvscsi_allocate_sg(struct pvscsi_adapter *adapter)
+{
+       struct pvscsi_ctx *ctx;
+       int i;
+
+       ctx = adapter->cmd_map;
+       BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE);
+
+       for (i = 0; i < adapter->req_depth; ++i, ++ctx) {
+               ctx->sgl = (void *)__get_free_pages(GFP_KERNEL,
+                                                   get_order(SGL_SIZE));
+               ctx->sglPA = 0;
+               BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE));
+               if (!ctx->sgl) {
+                       for (; i >= 0; --i, --ctx) {
+                               free_pages((unsigned long)ctx->sgl,
+                                          get_order(SGL_SIZE));
+                               ctx->sgl = NULL;
+                       }
+                       return -ENOMEM;
+               }
+       }
+
+       return 0;
+}
+
+static int __devinit pvscsi_probe(struct pci_dev *pdev,
+                                 const struct pci_device_id *id)
+{
+       struct pvscsi_adapter *adapter;
+       struct Scsi_Host *host;
+       unsigned int i;
+       unsigned long flags = 0;
+       int error;
+
+       error = -ENODEV;
+
+       if (pci_enable_device(pdev))
+               return error;
+
+       if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0 &&
+           pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
+               printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n");
+       } else if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) == 0 &&
+                  pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)) == 0) {
+               printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n");
+       } else {
+               printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n");
+               goto out_disable_device;
+       }
+
+       pvscsi_template.can_queue =
+               min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) *
+               PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
+       pvscsi_template.cmd_per_lun =
+               min(pvscsi_template.can_queue, pvscsi_cmd_per_lun);
+       host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter));
+       if (!host) {
+               printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n");
+               goto out_disable_device;
+       }
+
+       adapter = shost_priv(host);
+       memset(adapter, 0, sizeof(*adapter));
+       adapter->dev  = pdev;
+       adapter->host = host;
+
+       spin_lock_init(&adapter->hw_lock);
+
+       host->max_channel = 0;
+       host->max_id      = 16;
+       host->max_lun     = 1;
+       host->max_cmd_len = 16;
+
+       adapter->rev = pdev->revision;
+
+       if (pci_request_regions(pdev, "vmw_pvscsi")) {
+               printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n");
+               goto out_free_host;
+       }
+
+       for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
+               if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO))
+                       continue;
+
+               if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE)
+                       continue;
+
+               break;
+       }
+
+       if (i == DEVICE_COUNT_RESOURCE) {
+               printk(KERN_ERR
+                      "vmw_pvscsi: adapter has no suitable MMIO region\n");
+               goto out_release_resources;
+       }
+
+       adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE);
+
+       if (!adapter->mmioBase) {
+               printk(KERN_ERR
+                      "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n",
+                      i, PVSCSI_MEM_SPACE_SIZE);
+               goto out_release_resources;
+       }
+
+       pci_set_master(pdev);
+       pci_set_drvdata(pdev, host);
+
+       ll_adapter_reset(adapter);
+
+       adapter->use_msg = pvscsi_setup_msg_workqueue(adapter);
+
+       error = pvscsi_allocate_rings(adapter);
+       if (error) {
+               printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n");
+               goto out_release_resources;
+       }
+
+       /*
+        * From this point on we should reset the adapter if anything goes
+        * wrong.
+        */
+       pvscsi_setup_all_rings(adapter);
+
+       adapter->cmd_map = kcalloc(adapter->req_depth,
+                                  sizeof(struct pvscsi_ctx), GFP_KERNEL);
+       if (!adapter->cmd_map) {
+               printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n");
+               error = -ENOMEM;
+               goto out_reset_adapter;
+       }
+
+       INIT_LIST_HEAD(&adapter->cmd_pool);
+       for (i = 0; i < adapter->req_depth; i++) {
+               struct pvscsi_ctx *ctx = adapter->cmd_map + i;
+               list_add(&ctx->list, &adapter->cmd_pool);
+       }
+
+       error = pvscsi_allocate_sg(adapter);
+       if (error) {
+               printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n");
+               goto out_reset_adapter;
+       }
+
+       if (!pvscsi_disable_msix &&
+           pvscsi_setup_msix(adapter, &adapter->irq) == 0) {
+               printk(KERN_INFO "vmw_pvscsi: using MSI-X\n");
+               adapter->use_msix = 1;
+       } else if (!pvscsi_disable_msi && pci_enable_msi(pdev) == 0) {
+               printk(KERN_INFO "vmw_pvscsi: using MSI\n");
+               adapter->use_msi = 1;
+               adapter->irq = pdev->irq;
+       } else {
+               printk(KERN_INFO "vmw_pvscsi: using INTx\n");
+               adapter->irq = pdev->irq;
+               flags = IRQF_SHARED;
+       }
+
+       error = request_irq(adapter->irq, pvscsi_isr, flags,
+                           "vmw_pvscsi", adapter);
+       if (error) {
+               printk(KERN_ERR
+                      "vmw_pvscsi: unable to request IRQ: %d\n", error);
+               adapter->irq = 0;
+               goto out_reset_adapter;
+       }
+
+       error = scsi_add_host(host, &pdev->dev);
+       if (error) {
+               printk(KERN_ERR
+                      "vmw_pvscsi: scsi_add_host failed: %d\n", error);
+               goto out_reset_adapter;
+       }
+
+       dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n",
+                adapter->rev, host->host_no);
+
+       pvscsi_unmask_intr(adapter);
+
+       scsi_scan_host(host);
+
+       return 0;
+
+out_reset_adapter:
+       ll_adapter_reset(adapter);
+out_release_resources:
+       pvscsi_release_resources(adapter);
+out_free_host:
+       scsi_host_put(host);
+out_disable_device:
+       pci_set_drvdata(pdev, NULL);
+       pci_disable_device(pdev);
+
+       return error;
+}
+
+static void __pvscsi_shutdown(struct pvscsi_adapter *adapter)
+{
+       pvscsi_mask_intr(adapter);
+
+       if (adapter->workqueue)
+               flush_workqueue(adapter->workqueue);
+
+       pvscsi_shutdown_intr(adapter);
+
+       pvscsi_process_request_ring(adapter);
+       pvscsi_process_completion_ring(adapter);
+       ll_adapter_reset(adapter);
+}
+
+static void pvscsi_shutdown(struct pci_dev *dev)
+{
+       struct Scsi_Host *host = pci_get_drvdata(dev);
+       struct pvscsi_adapter *adapter = shost_priv(host);
+
+       __pvscsi_shutdown(adapter);
+}
+
+static void pvscsi_remove(struct pci_dev *pdev)
+{
+       struct Scsi_Host *host = pci_get_drvdata(pdev);
+       struct pvscsi_adapter *adapter = shost_priv(host);
+
+       scsi_remove_host(host);
+
+       __pvscsi_shutdown(adapter);
+       pvscsi_release_resources(adapter);
+
+       scsi_host_put(host);
+
+       pci_set_drvdata(pdev, NULL);
+       pci_disable_device(pdev);
+}
+
+static struct pci_driver pvscsi_pci_driver = {
+       .name           = "vmw_pvscsi",
+       .id_table       = pvscsi_pci_tbl,
+       .probe          = pvscsi_probe,
+       .remove         = __devexit_p(pvscsi_remove),
+       .shutdown       = pvscsi_shutdown,
+};
+
+static int __init pvscsi_init(void)
+{
+       pr_info("%s - version %s\n",
+               PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING);
+       return pci_register_driver(&pvscsi_pci_driver);
+}
+
+static void __exit pvscsi_exit(void)
+{
+       pci_unregister_driver(&pvscsi_pci_driver);
+}
+
+module_init(pvscsi_init);
+module_exit(pvscsi_exit);
diff --git a/drivers/scsi/vmw_pvscsi.h b/drivers/scsi/vmw_pvscsi.h
new file mode 100644 (file)
index 0000000..62e36e7
--- /dev/null
@@ -0,0 +1,397 @@
+/*
+ * VMware PVSCSI header file
+ *
+ * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the
+ * Free Software Foundation; version 2 of the License and no later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ * NON INFRINGEMENT.  See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Maintained by: Alok N Kataria <akataria@vmware.com>
+ *
+ */
+
+#ifndef _VMW_PVSCSI_H_
+#define _VMW_PVSCSI_H_
+
+#include <linux/types.h>
+
+#define PVSCSI_DRIVER_VERSION_STRING   "1.0.1.0-k"
+
+#define PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT 128
+
+#define MASK(n)        ((1 << (n)) - 1)        /* make an n-bit mask */
+
+#define PCI_VENDOR_ID_VMWARE           0x15AD
+#define PCI_DEVICE_ID_VMWARE_PVSCSI    0x07C0
+
+/*
+ * host adapter status/error codes
+ */
+enum HostBusAdapterStatus {
+   BTSTAT_SUCCESS       = 0x00,  /* CCB complete normally with no errors */
+   BTSTAT_LINKED_COMMAND_COMPLETED           = 0x0a,
+   BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG = 0x0b,
+   BTSTAT_DATA_UNDERRUN = 0x0c,
+   BTSTAT_SELTIMEO      = 0x11,  /* SCSI selection timeout */
+   BTSTAT_DATARUN       = 0x12,  /* data overrun/underrun */
+   BTSTAT_BUSFREE       = 0x13,  /* unexpected bus free */
+   BTSTAT_INVPHASE      = 0x14,  /* invalid bus phase or sequence requested by target */
+   BTSTAT_LUNMISMATCH   = 0x17,  /* linked CCB has different LUN from first CCB */
+   BTSTAT_SENSFAILED    = 0x1b,  /* auto request sense failed */
+   BTSTAT_TAGREJECT     = 0x1c,  /* SCSI II tagged queueing message rejected by target */
+   BTSTAT_BADMSG        = 0x1d,  /* unsupported message received by the host adapter */
+   BTSTAT_HAHARDWARE    = 0x20,  /* host adapter hardware failed */
+   BTSTAT_NORESPONSE    = 0x21,  /* target did not respond to SCSI ATN, sent a SCSI RST */
+   BTSTAT_SENTRST       = 0x22,  /* host adapter asserted a SCSI RST */
+   BTSTAT_RECVRST       = 0x23,  /* other SCSI devices asserted a SCSI RST */
+   BTSTAT_DISCONNECT    = 0x24,  /* target device reconnected improperly (w/o tag) */
+   BTSTAT_BUSRESET      = 0x25,  /* host adapter issued BUS device reset */
+   BTSTAT_ABORTQUEUE    = 0x26,  /* abort queue generated */
+   BTSTAT_HASOFTWARE    = 0x27,  /* host adapter software error */
+   BTSTAT_HATIMEOUT     = 0x30,  /* host adapter hardware timeout error */
+   BTSTAT_SCSIPARITY    = 0x34,  /* SCSI parity error detected */
+};
+
+/*
+ * Register offsets.
+ *
+ * These registers are accessible both via i/o space and mm i/o.
+ */
+
+enum PVSCSIRegOffset {
+       PVSCSI_REG_OFFSET_COMMAND        =    0x0,
+       PVSCSI_REG_OFFSET_COMMAND_DATA   =    0x4,
+       PVSCSI_REG_OFFSET_COMMAND_STATUS =    0x8,
+       PVSCSI_REG_OFFSET_LAST_STS_0     =  0x100,
+       PVSCSI_REG_OFFSET_LAST_STS_1     =  0x104,
+       PVSCSI_REG_OFFSET_LAST_STS_2     =  0x108,
+       PVSCSI_REG_OFFSET_LAST_STS_3     =  0x10c,
+       PVSCSI_REG_OFFSET_INTR_STATUS    = 0x100c,
+       PVSCSI_REG_OFFSET_INTR_MASK      = 0x2010,
+       PVSCSI_REG_OFFSET_KICK_NON_RW_IO = 0x3014,
+       PVSCSI_REG_OFFSET_DEBUG          = 0x3018,
+       PVSCSI_REG_OFFSET_KICK_RW_IO     = 0x4018,
+};
+
+/*
+ * Virtual h/w commands.
+ */
+
+enum PVSCSICommands {
+       PVSCSI_CMD_FIRST             = 0, /* has to be first */
+
+       PVSCSI_CMD_ADAPTER_RESET     = 1,
+       PVSCSI_CMD_ISSUE_SCSI        = 2,
+       PVSCSI_CMD_SETUP_RINGS       = 3,
+       PVSCSI_CMD_RESET_BUS         = 4,
+       PVSCSI_CMD_RESET_DEVICE      = 5,
+       PVSCSI_CMD_ABORT_CMD         = 6,
+       PVSCSI_CMD_CONFIG            = 7,
+       PVSCSI_CMD_SETUP_MSG_RING    = 8,
+       PVSCSI_CMD_DEVICE_UNPLUG     = 9,
+
+       PVSCSI_CMD_LAST              = 10  /* has to be last */
+};
+
+/*
+ * Command descriptor for PVSCSI_CMD_RESET_DEVICE --
+ */
+
+struct PVSCSICmdDescResetDevice {
+       u32     target;
+       u8      lun[8];
+} __packed;
+
+/*
+ * Command descriptor for PVSCSI_CMD_ABORT_CMD --
+ *
+ * - currently does not support specifying the LUN.
+ * - _pad should be 0.
+ */
+
+struct PVSCSICmdDescAbortCmd {
+       u64     context;
+       u32     target;
+       u32     _pad;
+} __packed;
+
+/*
+ * Command descriptor for PVSCSI_CMD_SETUP_RINGS --
+ *
+ * Notes:
+ * - reqRingNumPages and cmpRingNumPages need to be power of two.
+ * - reqRingNumPages and cmpRingNumPages need to be different from 0,
+ * - reqRingNumPages and cmpRingNumPages need to be inferior to
+ *   PVSCSI_SETUP_RINGS_MAX_NUM_PAGES.
+ */
+
+#define PVSCSI_SETUP_RINGS_MAX_NUM_PAGES        32
+struct PVSCSICmdDescSetupRings {
+       u32     reqRingNumPages;
+       u32     cmpRingNumPages;
+       u64     ringsStatePPN;
+       u64     reqRingPPNs[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
+       u64     cmpRingPPNs[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
+} __packed;
+
+/*
+ * Command descriptor for PVSCSI_CMD_SETUP_MSG_RING --
+ *
+ * Notes:
+ * - this command was not supported in the initial revision of the h/w
+ *   interface. Before using it, you need to check that it is supported by
+ *   writing PVSCSI_CMD_SETUP_MSG_RING to the 'command' register, then
+ *   immediately after read the 'command status' register:
+ *       * a value of -1 means that the cmd is NOT supported,
+ *       * a value != -1 means that the cmd IS supported.
+ *   If it's supported the 'command status' register should return:
+ *      sizeof(PVSCSICmdDescSetupMsgRing) / sizeof(u32).
+ * - this command should be issued _after_ the usual SETUP_RINGS so that the
+ *   RingsState page is already setup. If not, the command is a nop.
+ * - numPages needs to be a power of two,
+ * - numPages needs to be different from 0,
+ * - _pad should be zero.
+ */
+
+#define PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES  16
+
+struct PVSCSICmdDescSetupMsgRing {
+       u32     numPages;
+       u32     _pad;
+       u64     ringPPNs[PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES];
+} __packed;
+
+enum PVSCSIMsgType {
+       PVSCSI_MSG_DEV_ADDED          = 0,
+       PVSCSI_MSG_DEV_REMOVED        = 1,
+       PVSCSI_MSG_LAST               = 2,
+};
+
+/*
+ * Msg descriptor.
+ *
+ * sizeof(struct PVSCSIRingMsgDesc) == 128.
+ *
+ * - type is of type enum PVSCSIMsgType.
+ * - the content of args depend on the type of event being delivered.
+ */
+
+struct PVSCSIRingMsgDesc {
+       u32     type;
+       u32     args[31];
+} __packed;
+
+struct PVSCSIMsgDescDevStatusChanged {
+       u32     type;  /* PVSCSI_MSG_DEV _ADDED / _REMOVED */
+       u32     bus;
+       u32     target;
+       u8      lun[8];
+       u32     pad[27];
+} __packed;
+
+/*
+ * Rings state.
+ *
+ * - the fields:
+ *    . msgProdIdx,
+ *    . msgConsIdx,
+ *    . msgNumEntriesLog2,
+ *   .. are only used once the SETUP_MSG_RING cmd has been issued.
+ * - '_pad' helps to ensure that the msg related fields are on their own
+ *   cache-line.
+ */
+
+struct PVSCSIRingsState {
+       u32     reqProdIdx;
+       u32     reqConsIdx;
+       u32     reqNumEntriesLog2;
+
+       u32     cmpProdIdx;
+       u32     cmpConsIdx;
+       u32     cmpNumEntriesLog2;
+
+       u8      _pad[104];
+
+       u32     msgProdIdx;
+       u32     msgConsIdx;
+       u32     msgNumEntriesLog2;
+} __packed;
+
+/*
+ * Request descriptor.
+ *
+ * sizeof(RingReqDesc) = 128
+ *
+ * - context: is a unique identifier of a command. It could normally be any
+ *   64bit value, however we currently store it in the serialNumber variable
+ *   of struct SCSI_Command, so we have the following restrictions due to the
+ *   way this field is handled in the vmkernel storage stack:
+ *    * this value can't be 0,
+ *    * the upper 32bit need to be 0 since serialNumber is as a u32.
+ *   Currently tracked as PR 292060.
+ * - dataLen: contains the total number of bytes that need to be transferred.
+ * - dataAddr:
+ *   * if PVSCSI_FLAG_CMD_WITH_SG_LIST is set: dataAddr is the PA of the first
+ *     s/g table segment, each s/g segment is entirely contained on a single
+ *     page of physical memory,
+ *   * if PVSCSI_FLAG_CMD_WITH_SG_LIST is NOT set, then dataAddr is the PA of
+ *     the buffer used for the DMA transfer,
+ * - flags:
+ *   * PVSCSI_FLAG_CMD_WITH_SG_LIST: see dataAddr above,
+ *   * PVSCSI_FLAG_CMD_DIR_NONE: no DMA involved,
+ *   * PVSCSI_FLAG_CMD_DIR_TOHOST: transfer from device to main memory,
+ *   * PVSCSI_FLAG_CMD_DIR_TODEVICE: transfer from main memory to device,
+ *   * PVSCSI_FLAG_CMD_OUT_OF_BAND_CDB: reserved to handle CDBs larger than
+ *     16bytes. To be specified.
+ * - vcpuHint: vcpuId of the processor that will be most likely waiting for the
+ *   completion of the i/o. For guest OSes that use lowest priority message
+ *   delivery mode (such as windows), we use this "hint" to deliver the
+ *   completion action to the proper vcpu. For now, we can use the vcpuId of
+ *   the processor that initiated the i/o as a likely candidate for the vcpu
+ *   that will be waiting for the completion..
+ * - bus should be 0: we currently only support bus 0 for now.
+ * - unused should be zero'd.
+ */
+
+#define PVSCSI_FLAG_CMD_WITH_SG_LIST        (1 << 0)
+#define PVSCSI_FLAG_CMD_OUT_OF_BAND_CDB     (1 << 1)
+#define PVSCSI_FLAG_CMD_DIR_NONE            (1 << 2)
+#define PVSCSI_FLAG_CMD_DIR_TOHOST          (1 << 3)
+#define PVSCSI_FLAG_CMD_DIR_TODEVICE        (1 << 4)
+
+struct PVSCSIRingReqDesc {
+       u64     context;
+       u64     dataAddr;
+       u64     dataLen;
+       u64     senseAddr;
+       u32     senseLen;
+       u32     flags;
+       u8      cdb[16];
+       u8      cdbLen;
+       u8      lun[8];
+       u8      tag;
+       u8      bus;
+       u8      target;
+       u8      vcpuHint;
+       u8      unused[59];
+} __packed;
+
+/*
+ * Scatter-gather list management.
+ *
+ * As described above, when PVSCSI_FLAG_CMD_WITH_SG_LIST is set in the
+ * RingReqDesc.flags, then RingReqDesc.dataAddr is the PA of the first s/g
+ * table segment.
+ *
+ * - each segment of the s/g table contain a succession of struct
+ *   PVSCSISGElement.
+ * - each segment is entirely contained on a single physical page of memory.
+ * - a "chain" s/g element has the flag PVSCSI_SGE_FLAG_CHAIN_ELEMENT set in
+ *   PVSCSISGElement.flags and in this case:
+ *     * addr is the PA of the next s/g segment,
+ *     * length is undefined, assumed to be 0.
+ */
+
+struct PVSCSISGElement {
+       u64     addr;
+       u32     length;
+       u32     flags;
+} __packed;
+
+/*
+ * Completion descriptor.
+ *
+ * sizeof(RingCmpDesc) = 32
+ *
+ * - context: identifier of the command. The same thing that was specified
+ *   under "context" as part of struct RingReqDesc at initiation time,
+ * - dataLen: number of bytes transferred for the actual i/o operation,
+ * - senseLen: number of bytes written into the sense buffer,
+ * - hostStatus: adapter status,
+ * - scsiStatus: device status,
+ * - _pad should be zero.
+ */
+
+struct PVSCSIRingCmpDesc {
+       u64     context;
+       u64     dataLen;
+       u32     senseLen;
+       u16     hostStatus;
+       u16     scsiStatus;
+       u32     _pad[2];
+} __packed;
+
+/*
+ * Interrupt status / IRQ bits.
+ */
+
+#define PVSCSI_INTR_CMPL_0                 (1 << 0)
+#define PVSCSI_INTR_CMPL_1                 (1 << 1)
+#define PVSCSI_INTR_CMPL_MASK              MASK(2)
+
+#define PVSCSI_INTR_MSG_0                  (1 << 2)
+#define PVSCSI_INTR_MSG_1                  (1 << 3)
+#define PVSCSI_INTR_MSG_MASK               (MASK(2) << 2)
+
+#define PVSCSI_INTR_ALL_SUPPORTED          MASK(4)
+
+/*
+ * Number of MSI-X vectors supported.
+ */
+#define PVSCSI_MAX_INTRS        24
+
+/*
+ * Enumeration of supported MSI-X vectors
+ */
+#define PVSCSI_VECTOR_COMPLETION   0
+
+/*
+ * Misc constants for the rings.
+ */
+
+#define PVSCSI_MAX_NUM_PAGES_REQ_RING   PVSCSI_SETUP_RINGS_MAX_NUM_PAGES
+#define PVSCSI_MAX_NUM_PAGES_CMP_RING   PVSCSI_SETUP_RINGS_MAX_NUM_PAGES
+#define PVSCSI_MAX_NUM_PAGES_MSG_RING   PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES
+
+#define PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE \
+                               (PAGE_SIZE / sizeof(struct PVSCSIRingReqDesc))
+
+#define PVSCSI_MAX_REQ_QUEUE_DEPTH \
+       (PVSCSI_MAX_NUM_PAGES_REQ_RING * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE)
+
+#define PVSCSI_MEM_SPACE_COMMAND_NUM_PAGES     1
+#define PVSCSI_MEM_SPACE_INTR_STATUS_NUM_PAGES 1
+#define PVSCSI_MEM_SPACE_MISC_NUM_PAGES        2
+#define PVSCSI_MEM_SPACE_KICK_IO_NUM_PAGES     2
+#define PVSCSI_MEM_SPACE_MSIX_NUM_PAGES        2
+
+enum PVSCSIMemSpace {
+       PVSCSI_MEM_SPACE_COMMAND_PAGE           = 0,
+       PVSCSI_MEM_SPACE_INTR_STATUS_PAGE       = 1,
+       PVSCSI_MEM_SPACE_MISC_PAGE              = 2,
+       PVSCSI_MEM_SPACE_KICK_IO_PAGE           = 4,
+       PVSCSI_MEM_SPACE_MSIX_TABLE_PAGE        = 6,
+       PVSCSI_MEM_SPACE_MSIX_PBA_PAGE          = 7,
+};
+
+#define PVSCSI_MEM_SPACE_NUM_PAGES \
+       (PVSCSI_MEM_SPACE_COMMAND_NUM_PAGES +       \
+        PVSCSI_MEM_SPACE_INTR_STATUS_NUM_PAGES +   \
+        PVSCSI_MEM_SPACE_MISC_NUM_PAGES +          \
+        PVSCSI_MEM_SPACE_KICK_IO_NUM_PAGES +       \
+        PVSCSI_MEM_SPACE_MSIX_NUM_PAGES)
+
+#define PVSCSI_MEM_SPACE_SIZE        (PVSCSI_MEM_SPACE_NUM_PAGES * PAGE_SIZE)
+
+#endif /* _VMW_PVSCSI_H_ */