Add qemu 2.4.0

Change-Id: Ic99cbad4b61f8b127b7dc74d04576c0bcbaaf4f5
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>

17 files changed, 17689 insertions, 0 deletions

diff --git a/qemu/hw/scsi/Makefile.objs b/qemu/hw/scsi/Makefile.objs
new file mode 100644
index 000000000..40c79d34c
--- /dev/null
+++ b/qemu/hw/scsi/Makefile.objs
@@ -0,0 +1,13 @@
+common-obj-y += scsi-disk.o
+common-obj-y += scsi-generic.o scsi-bus.o
+common-obj-$(CONFIG_LSI_SCSI_PCI) += lsi53c895a.o
+common-obj-$(CONFIG_MEGASAS_SCSI_PCI) += megasas.o
+common-obj-$(CONFIG_VMW_PVSCSI_SCSI_PCI) += vmw_pvscsi.o
+common-obj-$(CONFIG_ESP) += esp.o
+common-obj-$(CONFIG_ESP_PCI) += esp-pci.o
+obj-$(CONFIG_PSERIES) += spapr_vscsi.o
+
+ifeq ($(CONFIG_VIRTIO),y)
+obj-y += virtio-scsi.o virtio-scsi-dataplane.o
+obj-$(CONFIG_VHOST_SCSI) += vhost-scsi.o
+endif
diff --git a/qemu/hw/scsi/esp-pci.c b/qemu/hw/scsi/esp-pci.c
new file mode 100644
index 000000000..8d2242d0a
--- /dev/null
+++ b/qemu/hw/scsi/esp-pci.c
@@ -0,0 +1,529 @@
+/*
+ * QEMU ESP/NCR53C9x emulation
+ *
+ * Copyright (c) 2005-2006 Fabrice Bellard
+ * Copyright (c) 2012 Herve Poussineau
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include "hw/pci/pci.h"
+#include "hw/nvram/eeprom93xx.h"
+#include "hw/scsi/esp.h"
+#include "trace.h"
+#include "qemu/log.h"
+
+#define TYPE_AM53C974_DEVICE "am53c974"
+
+#define PCI_ESP(obj) \
+    OBJECT_CHECK(PCIESPState, (obj), TYPE_AM53C974_DEVICE)
+
+#define DMA_CMD   0x0
+#define DMA_STC   0x1
+#define DMA_SPA   0x2
+#define DMA_WBC   0x3
+#define DMA_WAC   0x4
+#define DMA_STAT  0x5
+#define DMA_SMDLA 0x6
+#define DMA_WMAC  0x7
+
+#define DMA_CMD_MASK   0x03
+#define DMA_CMD_DIAG   0x04
+#define DMA_CMD_MDL    0x10
+#define DMA_CMD_INTE_P 0x20
+#define DMA_CMD_INTE_D 0x40
+#define DMA_CMD_DIR    0x80
+
+#define DMA_STAT_PWDN    0x01
+#define DMA_STAT_ERROR   0x02
+#define DMA_STAT_ABORT   0x04
+#define DMA_STAT_DONE    0x08
+#define DMA_STAT_SCSIINT 0x10
+#define DMA_STAT_BCMBLT  0x20
+
+#define SBAC_STATUS 0x1000
+
+typedef struct PCIESPState {
+    /*< private >*/
+    PCIDevice parent_obj;
+    /*< public >*/
+
+    MemoryRegion io;
+    uint32_t dma_regs[8];
+    uint32_t sbac;
+    ESPState esp;
+} PCIESPState;
+
+static void esp_pci_handle_idle(PCIESPState *pci, uint32_t val)
+{
+    trace_esp_pci_dma_idle(val);
+    esp_dma_enable(&pci->esp, 0, 0);
+}
+
+static void esp_pci_handle_blast(PCIESPState *pci, uint32_t val)
+{
+    trace_esp_pci_dma_blast(val);
+    qemu_log_mask(LOG_UNIMP, "am53c974: cmd BLAST not implemented\n");
+}
+
+static void esp_pci_handle_abort(PCIESPState *pci, uint32_t val)
+{
+    trace_esp_pci_dma_abort(val);
+    if (pci->esp.current_req) {
+        scsi_req_cancel(pci->esp.current_req);
+    }
+}
+
+static void esp_pci_handle_start(PCIESPState *pci, uint32_t val)
+{
+    trace_esp_pci_dma_start(val);
+
+    pci->dma_regs[DMA_WBC] = pci->dma_regs[DMA_STC];
+    pci->dma_regs[DMA_WAC] = pci->dma_regs[DMA_SPA];
+    pci->dma_regs[DMA_WMAC] = pci->dma_regs[DMA_SMDLA];
+
+    pci->dma_regs[DMA_STAT] &= ~(DMA_STAT_BCMBLT | DMA_STAT_SCSIINT
+                               | DMA_STAT_DONE | DMA_STAT_ABORT
+                               | DMA_STAT_ERROR | DMA_STAT_PWDN);
+
+    esp_dma_enable(&pci->esp, 0, 1);
+}
+
+static void esp_pci_dma_write(PCIESPState *pci, uint32_t saddr, uint32_t val)
+{
+    trace_esp_pci_dma_write(saddr, pci->dma_regs[saddr], val);
+    switch (saddr) {
+    case DMA_CMD:
+        pci->dma_regs[saddr] = val;
+        switch (val & DMA_CMD_MASK) {
+        case 0x0: /* IDLE */
+            esp_pci_handle_idle(pci, val);
+            break;
+        case 0x1: /* BLAST */
+            esp_pci_handle_blast(pci, val);
+            break;
+        case 0x2: /* ABORT */
+            esp_pci_handle_abort(pci, val);
+            break;
+        case 0x3: /* START */
+            esp_pci_handle_start(pci, val);
+            break;
+        default: /* can't happen */
+            abort();
+        }
+        break;
+    case DMA_STC:
+    case DMA_SPA:
+    case DMA_SMDLA:
+        pci->dma_regs[saddr] = val;
+        break;
+    case DMA_STAT:
+        if (!(pci->sbac & SBAC_STATUS)) {
+            /* clear some bits on write */
+            uint32_t mask = DMA_STAT_ERROR | DMA_STAT_ABORT | DMA_STAT_DONE;
+            pci->dma_regs[DMA_STAT] &= ~(val & mask);
+        }
+        break;
+    default:
+        trace_esp_pci_error_invalid_write_dma(val, saddr);
+        return;
+    }
+}
+
+static uint32_t esp_pci_dma_read(PCIESPState *pci, uint32_t saddr)
+{
+    uint32_t val;
+
+    val = pci->dma_regs[saddr];
+    if (saddr == DMA_STAT) {
+        if (pci->esp.rregs[ESP_RSTAT] & STAT_INT) {
+            val |= DMA_STAT_SCSIINT;
+        }
+        if (pci->sbac & SBAC_STATUS) {
+            pci->dma_regs[DMA_STAT] &= ~(DMA_STAT_ERROR | DMA_STAT_ABORT |
+                                         DMA_STAT_DONE);
+        }
+    }
+
+    trace_esp_pci_dma_read(saddr, val);
+    return val;
+}
+
+static void esp_pci_io_write(void *opaque, hwaddr addr,
+                             uint64_t val, unsigned int size)
+{
+    PCIESPState *pci = opaque;
+
+    if (size < 4 || addr & 3) {
+        /* need to upgrade request: we only support 4-bytes accesses */
+        uint32_t current = 0, mask;
+        int shift;
+
+        if (addr < 0x40) {
+            current = pci->esp.wregs[addr >> 2];
+        } else if (addr < 0x60) {
+            current = pci->dma_regs[(addr - 0x40) >> 2];
+        } else if (addr < 0x74) {
+            current = pci->sbac;
+        }
+
+        shift = (4 - size) * 8;
+        mask = (~(uint32_t)0 << shift) >> shift;
+
+        shift = ((4 - (addr & 3)) & 3) * 8;
+        val <<= shift;
+        val |= current & ~(mask << shift);
+        addr &= ~3;
+        size = 4;
+    }
+
+    if (addr < 0x40) {
+        /* SCSI core reg */
+        esp_reg_write(&pci->esp, addr >> 2, val);
+    } else if (addr < 0x60) {
+        /* PCI DMA CCB */
+        esp_pci_dma_write(pci, (addr - 0x40) >> 2, val);
+    } else if (addr == 0x70) {
+        /* DMA SCSI Bus and control */
+        trace_esp_pci_sbac_write(pci->sbac, val);
+        pci->sbac = val;
+    } else {
+        trace_esp_pci_error_invalid_write((int)addr);
+    }
+}
+
+static uint64_t esp_pci_io_read(void *opaque, hwaddr addr,
+                                unsigned int size)
+{
+    PCIESPState *pci = opaque;
+    uint32_t ret;
+
+    if (addr < 0x40) {
+        /* SCSI core reg */
+        ret = esp_reg_read(&pci->esp, addr >> 2);
+    } else if (addr < 0x60) {
+        /* PCI DMA CCB */
+        ret = esp_pci_dma_read(pci, (addr - 0x40) >> 2);
+    } else if (addr == 0x70) {
+        /* DMA SCSI Bus and control */
+        trace_esp_pci_sbac_read(pci->sbac);
+        ret = pci->sbac;
+    } else {
+        /* Invalid region */
+        trace_esp_pci_error_invalid_read((int)addr);
+        ret = 0;
+    }
+
+    /* give only requested data */
+    ret >>= (addr & 3) * 8;
+    ret &= ~(~(uint64_t)0 << (8 * size));
+
+    return ret;
+}
+
+static void esp_pci_dma_memory_rw(PCIESPState *pci, uint8_t *buf, int len,
+                                  DMADirection dir)
+{
+    dma_addr_t addr;
+    DMADirection expected_dir;
+
+    if (pci->dma_regs[DMA_CMD] & DMA_CMD_DIR) {
+        expected_dir = DMA_DIRECTION_FROM_DEVICE;
+    } else {
+        expected_dir = DMA_DIRECTION_TO_DEVICE;
+    }
+
+    if (dir != expected_dir) {
+        trace_esp_pci_error_invalid_dma_direction();
+        return;
+    }
+
+    if (pci->dma_regs[DMA_STAT] & DMA_CMD_MDL) {
+        qemu_log_mask(LOG_UNIMP, "am53c974: MDL transfer not implemented\n");
+    }
+
+    addr = pci->dma_regs[DMA_SPA];
+    if (pci->dma_regs[DMA_WBC] < len) {
+        len = pci->dma_regs[DMA_WBC];
+    }
+
+    pci_dma_rw(PCI_DEVICE(pci), addr, buf, len, dir);
+
+    /* update status registers */
+    pci->dma_regs[DMA_WBC] -= len;
+    pci->dma_regs[DMA_WAC] += len;
+    if (pci->dma_regs[DMA_WBC] == 0) {
+        pci->dma_regs[DMA_STAT] |= DMA_STAT_DONE;
+    }
+}
+
+static void esp_pci_dma_memory_read(void *opaque, uint8_t *buf, int len)
+{
+    PCIESPState *pci = opaque;
+    esp_pci_dma_memory_rw(pci, buf, len, DMA_DIRECTION_TO_DEVICE);
+}
+
+static void esp_pci_dma_memory_write(void *opaque, uint8_t *buf, int len)
+{
+    PCIESPState *pci = opaque;
+    esp_pci_dma_memory_rw(pci, buf, len, DMA_DIRECTION_FROM_DEVICE);
+}
+
+static const MemoryRegionOps esp_pci_io_ops = {
+    .read = esp_pci_io_read,
+    .write = esp_pci_io_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .impl = {
+        .min_access_size = 1,
+        .max_access_size = 4,
+    },
+};
+
+static void esp_pci_hard_reset(DeviceState *dev)
+{
+    PCIESPState *pci = PCI_ESP(dev);
+    esp_hard_reset(&pci->esp);
+    pci->dma_regs[DMA_CMD] &= ~(DMA_CMD_DIR | DMA_CMD_INTE_D | DMA_CMD_INTE_P
+                              | DMA_CMD_MDL | DMA_CMD_DIAG | DMA_CMD_MASK);
+    pci->dma_regs[DMA_WBC] &= ~0xffff;
+    pci->dma_regs[DMA_WAC] = 0xffffffff;
+    pci->dma_regs[DMA_STAT] &= ~(DMA_STAT_BCMBLT | DMA_STAT_SCSIINT
+                               | DMA_STAT_DONE | DMA_STAT_ABORT
+                               | DMA_STAT_ERROR);
+    pci->dma_regs[DMA_WMAC] = 0xfffffffd;
+}
+
+static const VMStateDescription vmstate_esp_pci_scsi = {
+    .name = "pciespscsi",
+    .version_id = 0,
+    .minimum_version_id = 0,
+    .fields = (VMStateField[]) {
+        VMSTATE_PCI_DEVICE(parent_obj, PCIESPState),
+        VMSTATE_BUFFER_UNSAFE(dma_regs, PCIESPState, 0, 8 * sizeof(uint32_t)),
+        VMSTATE_STRUCT(esp, PCIESPState, 0, vmstate_esp, ESPState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static void esp_pci_command_complete(SCSIRequest *req, uint32_t status,
+                                     size_t resid)
+{
+    ESPState *s = req->hba_private;
+    PCIESPState *pci = container_of(s, PCIESPState, esp);
+
+    esp_command_complete(req, status, resid);
+    pci->dma_regs[DMA_WBC] = 0;
+    pci->dma_regs[DMA_STAT] |= DMA_STAT_DONE;
+}
+
+static const struct SCSIBusInfo esp_pci_scsi_info = {
+    .tcq = false,
+    .max_target = ESP_MAX_DEVS,
+    .max_lun = 7,
+
+    .transfer_data = esp_transfer_data,
+    .complete = esp_pci_command_complete,
+    .cancel = esp_request_cancelled,
+};
+
+static void esp_pci_scsi_realize(PCIDevice *dev, Error **errp)
+{
+    PCIESPState *pci = PCI_ESP(dev);
+    DeviceState *d = DEVICE(dev);
+    ESPState *s = &pci->esp;
+    uint8_t *pci_conf;
+
+    pci_conf = dev->config;
+
+    /* Interrupt pin A */
+    pci_conf[PCI_INTERRUPT_PIN] = 0x01;
+
+    s->dma_memory_read = esp_pci_dma_memory_read;
+    s->dma_memory_write = esp_pci_dma_memory_write;
+    s->dma_opaque = pci;
+    s->chip_id = TCHI_AM53C974;
+    memory_region_init_io(&pci->io, OBJECT(pci), &esp_pci_io_ops, pci,
+                          "esp-io", 0x80);
+
+    pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_IO, &pci->io);
+    s->irq = pci_allocate_irq(dev);
+
+    scsi_bus_new(&s->bus, sizeof(s->bus), d, &esp_pci_scsi_info, NULL);
+    if (!d->hotplugged) {
+        scsi_bus_legacy_handle_cmdline(&s->bus, errp);
+    }
+}
+
+static void esp_pci_scsi_uninit(PCIDevice *d)
+{
+    PCIESPState *pci = PCI_ESP(d);
+
+    qemu_free_irq(pci->esp.irq);
+}
+
+static void esp_pci_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+
+    k->realize = esp_pci_scsi_realize;
+    k->exit = esp_pci_scsi_uninit;
+    k->vendor_id = PCI_VENDOR_ID_AMD;
+    k->device_id = PCI_DEVICE_ID_AMD_SCSI;
+    k->revision = 0x10;
+    k->class_id = PCI_CLASS_STORAGE_SCSI;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+    dc->desc = "AMD Am53c974 PCscsi-PCI SCSI adapter";
+    dc->reset = esp_pci_hard_reset;
+    dc->vmsd = &vmstate_esp_pci_scsi;
+}
+
+static const TypeInfo esp_pci_info = {
+    .name = TYPE_AM53C974_DEVICE,
+    .parent = TYPE_PCI_DEVICE,
+    .instance_size = sizeof(PCIESPState),
+    .class_init = esp_pci_class_init,
+};
+
+typedef struct {
+    PCIESPState pci;
+    eeprom_t *eeprom;
+} DC390State;
+
+#define TYPE_DC390_DEVICE "dc390"
+#define DC390(obj) \
+    OBJECT_CHECK(DC390State, obj, TYPE_DC390_DEVICE)
+
+#define EE_ADAPT_SCSI_ID 64
+#define EE_MODE2         65
+#define EE_DELAY         66
+#define EE_TAG_CMD_NUM   67
+#define EE_ADAPT_OPTIONS 68
+#define EE_BOOT_SCSI_ID  69
+#define EE_BOOT_SCSI_LUN 70
+#define EE_CHKSUM1       126
+#define EE_CHKSUM2       127
+
+#define EE_ADAPT_OPTION_F6_F8_AT_BOOT   0x01
+#define EE_ADAPT_OPTION_BOOT_FROM_CDROM 0x02
+#define EE_ADAPT_OPTION_INT13           0x04
+#define EE_ADAPT_OPTION_SCAM_SUPPORT    0x08
+
+
+static uint32_t dc390_read_config(PCIDevice *dev, uint32_t addr, int l)
+{
+    DC390State *pci = DC390(dev);
+    uint32_t val;
+
+    val = pci_default_read_config(dev, addr, l);
+
+    if (addr == 0x00 && l == 1) {
+        /* First byte of address space is AND-ed with EEPROM DO line */
+        if (!eeprom93xx_read(pci->eeprom)) {
+            val &= ~0xff;
+        }
+    }
+
+    return val;
+}
+
+static void dc390_write_config(PCIDevice *dev,
+                               uint32_t addr, uint32_t val, int l)
+{
+    DC390State *pci = DC390(dev);
+    if (addr == 0x80) {
+        /* EEPROM write */
+        int eesk = val & 0x80 ? 1 : 0;
+        int eedi = val & 0x40 ? 1 : 0;
+        eeprom93xx_write(pci->eeprom, 1, eesk, eedi);
+    } else if (addr == 0xc0) {
+        /* EEPROM CS low */
+        eeprom93xx_write(pci->eeprom, 0, 0, 0);
+    } else {
+        pci_default_write_config(dev, addr, val, l);
+    }
+}
+
+static void dc390_scsi_realize(PCIDevice *dev, Error **errp)
+{
+    DC390State *pci = DC390(dev);
+    Error *err = NULL;
+    uint8_t *contents;
+    uint16_t chksum = 0;
+    int i;
+
+    /* init base class */
+    esp_pci_scsi_realize(dev, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
+
+    /* EEPROM */
+    pci->eeprom = eeprom93xx_new(DEVICE(dev), 64);
+
+    /* set default eeprom values */
+    contents = (uint8_t *)eeprom93xx_data(pci->eeprom);
+
+    for (i = 0; i < 16; i++) {
+        contents[i * 2] = 0x57;
+        contents[i * 2 + 1] = 0x00;
+    }
+    contents[EE_ADAPT_SCSI_ID] = 7;
+    contents[EE_MODE2] = 0x0f;
+    contents[EE_TAG_CMD_NUM] = 0x04;
+    contents[EE_ADAPT_OPTIONS] = EE_ADAPT_OPTION_F6_F8_AT_BOOT
+                               | EE_ADAPT_OPTION_BOOT_FROM_CDROM
+                               | EE_ADAPT_OPTION_INT13;
+
+    /* update eeprom checksum */
+    for (i = 0; i < EE_CHKSUM1; i += 2) {
+        chksum += contents[i] + (((uint16_t)contents[i + 1]) << 8);
+    }
+    chksum = 0x1234 - chksum;
+    contents[EE_CHKSUM1] = chksum & 0xff;
+    contents[EE_CHKSUM2] = chksum >> 8;
+}
+
+static void dc390_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+
+    k->realize = dc390_scsi_realize;
+    k->config_read = dc390_read_config;
+    k->config_write = dc390_write_config;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+    dc->desc = "Tekram DC-390 SCSI adapter";
+}
+
+static const TypeInfo dc390_info = {
+    .name = "dc390",
+    .parent = TYPE_AM53C974_DEVICE,
+    .instance_size = sizeof(DC390State),
+    .class_init = dc390_class_init,
+};
+
+static void esp_pci_register_types(void)
+{
+    type_register_static(&esp_pci_info);
+    type_register_static(&dc390_info);
+}
+
+type_init(esp_pci_register_types)
diff --git a/qemu/hw/scsi/esp.c b/qemu/hw/scsi/esp.c
new file mode 100644
index 000000000..272d13d63
--- /dev/null
+++ b/qemu/hw/scsi/esp.c
@@ -0,0 +1,743 @@
+/*
+ * QEMU ESP/NCR53C9x emulation
+ *
+ * Copyright (c) 2005-2006 Fabrice Bellard
+ * Copyright (c) 2012 Herve Poussineau
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include "hw/sysbus.h"
+#include "hw/scsi/esp.h"
+#include "trace.h"
+#include "qemu/log.h"
+
+/*
+ * On Sparc32, this is the ESP (NCR53C90) part of chip STP2000 (Master I/O),
+ * also produced as NCR89C100. See
+ * http://www.ibiblio.org/pub/historic-linux/early-ports/Sparc/NCR/NCR89C100.txt
+ * and
+ * http://www.ibiblio.org/pub/historic-linux/early-ports/Sparc/NCR/NCR53C9X.txt
+ */
+
+static void esp_raise_irq(ESPState *s)
+{
+    if (!(s->rregs[ESP_RSTAT] & STAT_INT)) {
+        s->rregs[ESP_RSTAT] |= STAT_INT;
+        qemu_irq_raise(s->irq);
+        trace_esp_raise_irq();
+    }
+}
+
+static void esp_lower_irq(ESPState *s)
+{
+    if (s->rregs[ESP_RSTAT] & STAT_INT) {
+        s->rregs[ESP_RSTAT] &= ~STAT_INT;
+        qemu_irq_lower(s->irq);
+        trace_esp_lower_irq();
+    }
+}
+
+void esp_dma_enable(ESPState *s, int irq, int level)
+{
+    if (level) {
+        s->dma_enabled = 1;
+        trace_esp_dma_enable();
+        if (s->dma_cb) {
+            s->dma_cb(s);
+            s->dma_cb = NULL;
+        }
+    } else {
+        trace_esp_dma_disable();
+        s->dma_enabled = 0;
+    }
+}
+
+void esp_request_cancelled(SCSIRequest *req)
+{
+    ESPState *s = req->hba_private;
+
+    if (req == s->current_req) {
+        scsi_req_unref(s->current_req);
+        s->current_req = NULL;
+        s->current_dev = NULL;
+    }
+}
+
+static uint32_t get_cmd(ESPState *s, uint8_t *buf)
+{
+    uint32_t dmalen;
+    int target;
+
+    target = s->wregs[ESP_WBUSID] & BUSID_DID;
+    if (s->dma) {
+        dmalen = s->rregs[ESP_TCLO];
+        dmalen |= s->rregs[ESP_TCMID] << 8;
+        dmalen |= s->rregs[ESP_TCHI] << 16;
+        s->dma_memory_read(s->dma_opaque, buf, dmalen);
+    } else {
+        dmalen = s->ti_size;
+        memcpy(buf, s->ti_buf, dmalen);
+        buf[0] = buf[2] >> 5;
+    }
+    trace_esp_get_cmd(dmalen, target);
+
+    s->ti_size = 0;
+    s->ti_rptr = 0;
+    s->ti_wptr = 0;
+
+    if (s->current_req) {
+        /* Started a new command before the old one finished.  Cancel it.  */
+        scsi_req_cancel(s->current_req);
+        s->async_len = 0;
+    }
+
+    s->current_dev = scsi_device_find(&s->bus, 0, target, 0);
+    if (!s->current_dev) {
+        // No such drive
+        s->rregs[ESP_RSTAT] = 0;
+        s->rregs[ESP_RINTR] = INTR_DC;
+        s->rregs[ESP_RSEQ] = SEQ_0;
+        esp_raise_irq(s);
+        return 0;
+    }
+    return dmalen;
+}
+
+static void do_busid_cmd(ESPState *s, uint8_t *buf, uint8_t busid)
+{
+    int32_t datalen;
+    int lun;
+    SCSIDevice *current_lun;
+
+    trace_esp_do_busid_cmd(busid);
+    lun = busid & 7;
+    current_lun = scsi_device_find(&s->bus, 0, s->current_dev->id, lun);
+    s->current_req = scsi_req_new(current_lun, 0, lun, buf, s);
+    datalen = scsi_req_enqueue(s->current_req);
+    s->ti_size = datalen;
+    if (datalen != 0) {
+        s->rregs[ESP_RSTAT] = STAT_TC;
+        s->dma_left = 0;
+        s->dma_counter = 0;
+        if (datalen > 0) {
+            s->rregs[ESP_RSTAT] |= STAT_DI;
+        } else {
+            s->rregs[ESP_RSTAT] |= STAT_DO;
+        }
+        scsi_req_continue(s->current_req);
+    }
+    s->rregs[ESP_RINTR] = INTR_BS | INTR_FC;
+    s->rregs[ESP_RSEQ] = SEQ_CD;
+    esp_raise_irq(s);
+}
+
+static void do_cmd(ESPState *s, uint8_t *buf)
+{
+    uint8_t busid = buf[0];
+
+    do_busid_cmd(s, &buf[1], busid);
+}
+
+static void handle_satn(ESPState *s)
+{
+    uint8_t buf[32];
+    int len;
+
+    if (s->dma && !s->dma_enabled) {
+        s->dma_cb = handle_satn;
+        return;
+    }
+    len = get_cmd(s, buf);
+    if (len)
+        do_cmd(s, buf);
+}
+
+static void handle_s_without_atn(ESPState *s)
+{
+    uint8_t buf[32];
+    int len;
+
+    if (s->dma && !s->dma_enabled) {
+        s->dma_cb = handle_s_without_atn;
+        return;
+    }
+    len = get_cmd(s, buf);
+    if (len) {
+        do_busid_cmd(s, buf, 0);
+    }
+}
+
+static void handle_satn_stop(ESPState *s)
+{
+    if (s->dma && !s->dma_enabled) {
+        s->dma_cb = handle_satn_stop;
+        return;
+    }
+    s->cmdlen = get_cmd(s, s->cmdbuf);
+    if (s->cmdlen) {
+        trace_esp_handle_satn_stop(s->cmdlen);
+        s->do_cmd = 1;
+        s->rregs[ESP_RSTAT] = STAT_TC | STAT_CD;
+        s->rregs[ESP_RINTR] = INTR_BS | INTR_FC;
+        s->rregs[ESP_RSEQ] = SEQ_CD;
+        esp_raise_irq(s);
+    }
+}
+
+static void write_response(ESPState *s)
+{
+    trace_esp_write_response(s->status);
+    s->ti_buf[0] = s->status;
+    s->ti_buf[1] = 0;
+    if (s->dma) {
+        s->dma_memory_write(s->dma_opaque, s->ti_buf, 2);
+        s->rregs[ESP_RSTAT] = STAT_TC | STAT_ST;
+        s->rregs[ESP_RINTR] = INTR_BS | INTR_FC;
+        s->rregs[ESP_RSEQ] = SEQ_CD;
+    } else {
+        s->ti_size = 2;
+        s->ti_rptr = 0;
+        s->ti_wptr = 0;
+        s->rregs[ESP_RFLAGS] = 2;
+    }
+    esp_raise_irq(s);
+}
+
+static void esp_dma_done(ESPState *s)
+{
+    s->rregs[ESP_RSTAT] |= STAT_TC;
+    s->rregs[ESP_RINTR] = INTR_BS;
+    s->rregs[ESP_RSEQ] = 0;
+    s->rregs[ESP_RFLAGS] = 0;
+    s->rregs[ESP_TCLO] = 0;
+    s->rregs[ESP_TCMID] = 0;
+    s->rregs[ESP_TCHI] = 0;
+    esp_raise_irq(s);
+}
+
+static void esp_do_dma(ESPState *s)
+{
+    uint32_t len;
+    int to_device;
+
+    to_device = (s->ti_size < 0);
+    len = s->dma_left;
+    if (s->do_cmd) {
+        trace_esp_do_dma(s->cmdlen, len);
+        s->dma_memory_read(s->dma_opaque, &s->cmdbuf[s->cmdlen], len);
+        s->ti_size = 0;
+        s->cmdlen = 0;
+        s->do_cmd = 0;
+        do_cmd(s, s->cmdbuf);
+        return;
+    }
+    if (s->async_len == 0) {
+        /* Defer until data is available.  */
+        return;
+    }
+    if (len > s->async_len) {
+        len = s->async_len;
+    }
+    if (to_device) {
+        s->dma_memory_read(s->dma_opaque, s->async_buf, len);
+    } else {
+        s->dma_memory_write(s->dma_opaque, s->async_buf, len);
+    }
+    s->dma_left -= len;
+    s->async_buf += len;
+    s->async_len -= len;
+    if (to_device)
+        s->ti_size += len;
+    else
+        s->ti_size -= len;
+    if (s->async_len == 0) {
+        scsi_req_continue(s->current_req);
+        /* If there is still data to be read from the device then
+           complete the DMA operation immediately.  Otherwise defer
+           until the scsi layer has completed.  */
+        if (to_device || s->dma_left != 0 || s->ti_size == 0) {
+            return;
+        }
+    }
+
+    /* Partially filled a scsi buffer. Complete immediately.  */
+    esp_dma_done(s);
+}
+
+void esp_command_complete(SCSIRequest *req, uint32_t status,
+                                 size_t resid)
+{
+    ESPState *s = req->hba_private;
+
+    trace_esp_command_complete();
+    if (s->ti_size != 0) {
+        trace_esp_command_complete_unexpected();
+    }
+    s->ti_size = 0;
+    s->dma_left = 0;
+    s->async_len = 0;
+    if (status) {
+        trace_esp_command_complete_fail();
+    }
+    s->status = status;
+    s->rregs[ESP_RSTAT] = STAT_ST;
+    esp_dma_done(s);
+    if (s->current_req) {
+        scsi_req_unref(s->current_req);
+        s->current_req = NULL;
+        s->current_dev = NULL;
+    }
+}
+
+void esp_transfer_data(SCSIRequest *req, uint32_t len)
+{
+    ESPState *s = req->hba_private;
+
+    trace_esp_transfer_data(s->dma_left, s->ti_size);
+    s->async_len = len;
+    s->async_buf = scsi_req_get_buf(req);
+    if (s->dma_left) {
+        esp_do_dma(s);
+    } else if (s->dma_counter != 0 && s->ti_size <= 0) {
+        /* If this was the last part of a DMA transfer then the
+           completion interrupt is deferred to here.  */
+        esp_dma_done(s);
+    }
+}
+
+static void handle_ti(ESPState *s)
+{
+    uint32_t dmalen, minlen;
+
+    if (s->dma && !s->dma_enabled) {
+        s->dma_cb = handle_ti;
+        return;
+    }
+
+    dmalen = s->rregs[ESP_TCLO];
+    dmalen |= s->rregs[ESP_TCMID] << 8;
+    dmalen |= s->rregs[ESP_TCHI] << 16;
+    if (dmalen==0) {
+      dmalen=0x10000;
+    }
+    s->dma_counter = dmalen;
+
+    if (s->do_cmd)
+        minlen = (dmalen < 32) ? dmalen : 32;
+    else if (s->ti_size < 0)
+        minlen = (dmalen < -s->ti_size) ? dmalen : -s->ti_size;
+    else
+        minlen = (dmalen < s->ti_size) ? dmalen : s->ti_size;
+    trace_esp_handle_ti(minlen);
+    if (s->dma) {
+        s->dma_left = minlen;
+        s->rregs[ESP_RSTAT] &= ~STAT_TC;
+        esp_do_dma(s);
+    } else if (s->do_cmd) {
+        trace_esp_handle_ti_cmd(s->cmdlen);
+        s->ti_size = 0;
+        s->cmdlen = 0;
+        s->do_cmd = 0;
+        do_cmd(s, s->cmdbuf);
+        return;
+    }
+}
+
+void esp_hard_reset(ESPState *s)
+{
+    memset(s->rregs, 0, ESP_REGS);
+    memset(s->wregs, 0, ESP_REGS);
+    s->tchi_written = 0;
+    s->ti_size = 0;
+    s->ti_rptr = 0;
+    s->ti_wptr = 0;
+    s->dma = 0;
+    s->do_cmd = 0;
+    s->dma_cb = NULL;
+
+    s->rregs[ESP_CFG1] = 7;
+}
+
+static void esp_soft_reset(ESPState *s)
+{
+    qemu_irq_lower(s->irq);
+    esp_hard_reset(s);
+}
+
+static void parent_esp_reset(ESPState *s, int irq, int level)
+{
+    if (level) {
+        esp_soft_reset(s);
+    }
+}
+
+uint64_t esp_reg_read(ESPState *s, uint32_t saddr)
+{
+    uint32_t old_val;
+
+    trace_esp_mem_readb(saddr, s->rregs[saddr]);
+    switch (saddr) {
+    case ESP_FIFO:
+        if (s->ti_size > 0) {
+            s->ti_size--;
+            if ((s->rregs[ESP_RSTAT] & STAT_PIO_MASK) == 0) {
+                /* Data out.  */
+                qemu_log_mask(LOG_UNIMP,
+                              "esp: PIO data read not implemented\n");
+                s->rregs[ESP_FIFO] = 0;
+            } else {
+                s->rregs[ESP_FIFO] = s->ti_buf[s->ti_rptr++];
+            }
+            esp_raise_irq(s);
+        }
+        if (s->ti_size == 0) {
+            s->ti_rptr = 0;
+            s->ti_wptr = 0;
+        }
+        break;
+    case ESP_RINTR:
+        /* Clear sequence step, interrupt register and all status bits
+           except TC */
+        old_val = s->rregs[ESP_RINTR];
+        s->rregs[ESP_RINTR] = 0;
+        s->rregs[ESP_RSTAT] &= ~STAT_TC;
+        s->rregs[ESP_RSEQ] = SEQ_CD;
+        esp_lower_irq(s);
+
+        return old_val;
+    case ESP_TCHI:
+        /* Return the unique id if the value has never been written */
+        if (!s->tchi_written) {
+            return s->chip_id;
+        }
+    default:
+        break;
+    }
+    return s->rregs[saddr];
+}
+
+void esp_reg_write(ESPState *s, uint32_t saddr, uint64_t val)
+{
+    trace_esp_mem_writeb(saddr, s->wregs[saddr], val);
+    switch (saddr) {
+    case ESP_TCHI:
+        s->tchi_written = true;
+        /* fall through */
+    case ESP_TCLO:
+    case ESP_TCMID:
+        s->rregs[ESP_RSTAT] &= ~STAT_TC;
+        break;
+    case ESP_FIFO:
+        if (s->do_cmd) {
+            s->cmdbuf[s->cmdlen++] = val & 0xff;
+        } else if (s->ti_size == TI_BUFSZ - 1) {
+            trace_esp_error_fifo_overrun();
+        } else {
+            s->ti_size++;
+            s->ti_buf[s->ti_wptr++] = val & 0xff;
+        }
+        break;
+    case ESP_CMD:
+        s->rregs[saddr] = val;
+        if (val & CMD_DMA) {
+            s->dma = 1;
+            /* Reload DMA counter.  */
+            s->rregs[ESP_TCLO] = s->wregs[ESP_TCLO];
+            s->rregs[ESP_TCMID] = s->wregs[ESP_TCMID];
+            s->rregs[ESP_TCHI] = s->wregs[ESP_TCHI];
+        } else {
+            s->dma = 0;
+        }
+        switch(val & CMD_CMD) {
+        case CMD_NOP:
+            trace_esp_mem_writeb_cmd_nop(val);
+            break;
+        case CMD_FLUSH:
+            trace_esp_mem_writeb_cmd_flush(val);
+            //s->ti_size = 0;
+            s->rregs[ESP_RINTR] = INTR_FC;
+            s->rregs[ESP_RSEQ] = 0;
+            s->rregs[ESP_RFLAGS] = 0;
+            break;
+        case CMD_RESET:
+            trace_esp_mem_writeb_cmd_reset(val);
+            esp_soft_reset(s);
+            break;
+        case CMD_BUSRESET:
+            trace_esp_mem_writeb_cmd_bus_reset(val);
+            s->rregs[ESP_RINTR] = INTR_RST;
+            if (!(s->wregs[ESP_CFG1] & CFG1_RESREPT)) {
+                esp_raise_irq(s);
+            }
+            break;
+        case CMD_TI:
+            handle_ti(s);
+            break;
+        case CMD_ICCS:
+            trace_esp_mem_writeb_cmd_iccs(val);
+            write_response(s);
+            s->rregs[ESP_RINTR] = INTR_FC;
+            s->rregs[ESP_RSTAT] |= STAT_MI;
+            break;
+        case CMD_MSGACC:
+            trace_esp_mem_writeb_cmd_msgacc(val);
+            s->rregs[ESP_RINTR] = INTR_DC;
+            s->rregs[ESP_RSEQ] = 0;
+            s->rregs[ESP_RFLAGS] = 0;
+            esp_raise_irq(s);
+            break;
+        case CMD_PAD:
+            trace_esp_mem_writeb_cmd_pad(val);
+            s->rregs[ESP_RSTAT] = STAT_TC;
+            s->rregs[ESP_RINTR] = INTR_FC;
+            s->rregs[ESP_RSEQ] = 0;
+            break;
+        case CMD_SATN:
+            trace_esp_mem_writeb_cmd_satn(val);
+            break;
+        case CMD_RSTATN:
+            trace_esp_mem_writeb_cmd_rstatn(val);
+            break;
+        case CMD_SEL:
+            trace_esp_mem_writeb_cmd_sel(val);
+            handle_s_without_atn(s);
+            break;
+        case CMD_SELATN:
+            trace_esp_mem_writeb_cmd_selatn(val);
+            handle_satn(s);
+            break;
+        case CMD_SELATNS:
+            trace_esp_mem_writeb_cmd_selatns(val);
+            handle_satn_stop(s);
+            break;
+        case CMD_ENSEL:
+            trace_esp_mem_writeb_cmd_ensel(val);
+            s->rregs[ESP_RINTR] = 0;
+            break;
+        case CMD_DISSEL:
+            trace_esp_mem_writeb_cmd_dissel(val);
+            s->rregs[ESP_RINTR] = 0;
+            esp_raise_irq(s);
+            break;
+        default:
+            trace_esp_error_unhandled_command(val);
+            break;
+        }
+        break;
+    case ESP_WBUSID ... ESP_WSYNO:
+        break;
+    case ESP_CFG1:
+    case ESP_CFG2: case ESP_CFG3:
+    case ESP_RES3: case ESP_RES4:
+        s->rregs[saddr] = val;
+        break;
+    case ESP_WCCF ... ESP_WTEST:
+        break;
+    default:
+        trace_esp_error_invalid_write(val, saddr);
+        return;
+    }
+    s->wregs[saddr] = val;
+}
+
+static bool esp_mem_accepts(void *opaque, hwaddr addr,
+                            unsigned size, bool is_write)
+{
+    return (size == 1) || (is_write && size == 4);
+}
+
+const VMStateDescription vmstate_esp = {
+    .name ="esp",
+    .version_id = 3,
+    .minimum_version_id = 3,
+    .fields = (VMStateField[]) {
+        VMSTATE_BUFFER(rregs, ESPState),
+        VMSTATE_BUFFER(wregs, ESPState),
+        VMSTATE_INT32(ti_size, ESPState),
+        VMSTATE_UINT32(ti_rptr, ESPState),
+        VMSTATE_UINT32(ti_wptr, ESPState),
+        VMSTATE_BUFFER(ti_buf, ESPState),
+        VMSTATE_UINT32(status, ESPState),
+        VMSTATE_UINT32(dma, ESPState),
+        VMSTATE_BUFFER(cmdbuf, ESPState),
+        VMSTATE_UINT32(cmdlen, ESPState),
+        VMSTATE_UINT32(do_cmd, ESPState),
+        VMSTATE_UINT32(dma_left, ESPState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+#define TYPE_ESP "esp"
+#define ESP(obj) OBJECT_CHECK(SysBusESPState, (obj), TYPE_ESP)
+
+typedef struct {
+    /*< private >*/
+    SysBusDevice parent_obj;
+    /*< public >*/
+
+    MemoryRegion iomem;
+    uint32_t it_shift;
+    ESPState esp;
+} SysBusESPState;
+
+static void sysbus_esp_mem_write(void *opaque, hwaddr addr,
+                                 uint64_t val, unsigned int size)
+{
+    SysBusESPState *sysbus = opaque;
+    uint32_t saddr;
+
+    saddr = addr >> sysbus->it_shift;
+    esp_reg_write(&sysbus->esp, saddr, val);
+}
+
+static uint64_t sysbus_esp_mem_read(void *opaque, hwaddr addr,
+                                    unsigned int size)
+{
+    SysBusESPState *sysbus = opaque;
+    uint32_t saddr;
+
+    saddr = addr >> sysbus->it_shift;
+    return esp_reg_read(&sysbus->esp, saddr);
+}
+
+static const MemoryRegionOps sysbus_esp_mem_ops = {
+    .read = sysbus_esp_mem_read,
+    .write = sysbus_esp_mem_write,
+    .endianness = DEVICE_NATIVE_ENDIAN,
+    .valid.accepts = esp_mem_accepts,
+};
+
+void esp_init(hwaddr espaddr, int it_shift,
+              ESPDMAMemoryReadWriteFunc dma_memory_read,
+              ESPDMAMemoryReadWriteFunc dma_memory_write,
+              void *dma_opaque, qemu_irq irq, qemu_irq *reset,
+              qemu_irq *dma_enable)
+{
+    DeviceState *dev;
+    SysBusDevice *s;
+    SysBusESPState *sysbus;
+    ESPState *esp;
+
+    dev = qdev_create(NULL, TYPE_ESP);
+    sysbus = ESP(dev);
+    esp = &sysbus->esp;
+    esp->dma_memory_read = dma_memory_read;
+    esp->dma_memory_write = dma_memory_write;
+    esp->dma_opaque = dma_opaque;
+    sysbus->it_shift = it_shift;
+    /* XXX for now until rc4030 has been changed to use DMA enable signal */
+    esp->dma_enabled = 1;
+    qdev_init_nofail(dev);
+    s = SYS_BUS_DEVICE(dev);
+    sysbus_connect_irq(s, 0, irq);
+    sysbus_mmio_map(s, 0, espaddr);
+    *reset = qdev_get_gpio_in(dev, 0);
+    *dma_enable = qdev_get_gpio_in(dev, 1);
+}
+
+static const struct SCSIBusInfo esp_scsi_info = {
+    .tcq = false,
+    .max_target = ESP_MAX_DEVS,
+    .max_lun = 7,
+
+    .transfer_data = esp_transfer_data,
+    .complete = esp_command_complete,
+    .cancel = esp_request_cancelled
+};
+
+static void sysbus_esp_gpio_demux(void *opaque, int irq, int level)
+{
+    SysBusESPState *sysbus = ESP(opaque);
+    ESPState *s = &sysbus->esp;
+
+    switch (irq) {
+    case 0:
+        parent_esp_reset(s, irq, level);
+        break;
+    case 1:
+        esp_dma_enable(opaque, irq, level);
+        break;
+    }
+}
+
+static void sysbus_esp_realize(DeviceState *dev, Error **errp)
+{
+    SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
+    SysBusESPState *sysbus = ESP(dev);
+    ESPState *s = &sysbus->esp;
+    Error *err = NULL;
+
+    sysbus_init_irq(sbd, &s->irq);
+    assert(sysbus->it_shift != -1);
+
+    s->chip_id = TCHI_FAS100A;
+    memory_region_init_io(&sysbus->iomem, OBJECT(sysbus), &sysbus_esp_mem_ops,
+                          sysbus, "esp", ESP_REGS << sysbus->it_shift);
+    sysbus_init_mmio(sbd, &sysbus->iomem);
+
+    qdev_init_gpio_in(dev, sysbus_esp_gpio_demux, 2);
+
+    scsi_bus_new(&s->bus, sizeof(s->bus), dev, &esp_scsi_info, NULL);
+    scsi_bus_legacy_handle_cmdline(&s->bus, &err);
+    if (err != NULL) {
+        error_propagate(errp, err);
+        return;
+    }
+}
+
+static void sysbus_esp_hard_reset(DeviceState *dev)
+{
+    SysBusESPState *sysbus = ESP(dev);
+    esp_hard_reset(&sysbus->esp);
+}
+
+static const VMStateDescription vmstate_sysbus_esp_scsi = {
+    .name = "sysbusespscsi",
+    .version_id = 0,
+    .minimum_version_id = 0,
+    .fields = (VMStateField[]) {
+        VMSTATE_STRUCT(esp, SysBusESPState, 0, vmstate_esp, ESPState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static void sysbus_esp_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+
+    dc->realize = sysbus_esp_realize;
+    dc->reset = sysbus_esp_hard_reset;
+    dc->vmsd = &vmstate_sysbus_esp_scsi;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+}
+
+static const TypeInfo sysbus_esp_info = {
+    .name          = TYPE_ESP,
+    .parent        = TYPE_SYS_BUS_DEVICE,
+    .instance_size = sizeof(SysBusESPState),
+    .class_init    = sysbus_esp_class_init,
+};
+
+static void esp_register_types(void)
+{
+    type_register_static(&sysbus_esp_info);
+}
+
+type_init(esp_register_types)
diff --git a/qemu/hw/scsi/lsi53c895a.c b/qemu/hw/scsi/lsi53c895a.c
new file mode 100644
index 000000000..c5b0cc5ca
--- /dev/null
+++ b/qemu/hw/scsi/lsi53c895a.c
@@ -0,0 +1,2163 @@
+/*
+ * QEMU LSI53C895A SCSI Host Bus Adapter emulation
+ *
+ * Copyright (c) 2006 CodeSourcery.
+ * Written by Paul Brook
+ *
+ * This code is licensed under the LGPL.
+ */
+
+/* Note:
+ * LSI53C810 emulation is incorrect, in the sense that it supports
+ * features added in later evolutions. This should not be a problem,
+ * as well-behaved operating systems will not try to use them.
+ */
+
+#include <assert.h>
+
+#include "hw/hw.h"
+#include "hw/pci/pci.h"
+#include "hw/scsi/scsi.h"
+#include "sysemu/dma.h"
+
+//#define DEBUG_LSI
+//#define DEBUG_LSI_REG
+
+#ifdef DEBUG_LSI
+#define DPRINTF(fmt, ...) \
+do { printf("lsi_scsi: " fmt , ## __VA_ARGS__); } while (0)
+#define BADF(fmt, ...) \
+do { fprintf(stderr, "lsi_scsi: error: " fmt , ## __VA_ARGS__); exit(1);} while (0)
+#else
+#define DPRINTF(fmt, ...) do {} while(0)
+#define BADF(fmt, ...) \
+do { fprintf(stderr, "lsi_scsi: error: " fmt , ## __VA_ARGS__);} while (0)
+#endif
+
+#define LSI_MAX_DEVS 7
+
+#define LSI_SCNTL0_TRG    0x01
+#define LSI_SCNTL0_AAP    0x02
+#define LSI_SCNTL0_EPC    0x08
+#define LSI_SCNTL0_WATN   0x10
+#define LSI_SCNTL0_START  0x20
+
+#define LSI_SCNTL1_SST    0x01
+#define LSI_SCNTL1_IARB   0x02
+#define LSI_SCNTL1_AESP   0x04
+#define LSI_SCNTL1_RST    0x08
+#define LSI_SCNTL1_CON    0x10
+#define LSI_SCNTL1_DHP    0x20
+#define LSI_SCNTL1_ADB    0x40
+#define LSI_SCNTL1_EXC    0x80
+
+#define LSI_SCNTL2_WSR    0x01
+#define LSI_SCNTL2_VUE0   0x02
+#define LSI_SCNTL2_VUE1   0x04
+#define LSI_SCNTL2_WSS    0x08
+#define LSI_SCNTL2_SLPHBEN 0x10
+#define LSI_SCNTL2_SLPMD  0x20
+#define LSI_SCNTL2_CHM    0x40
+#define LSI_SCNTL2_SDU    0x80
+
+#define LSI_ISTAT0_DIP    0x01
+#define LSI_ISTAT0_SIP    0x02
+#define LSI_ISTAT0_INTF   0x04
+#define LSI_ISTAT0_CON    0x08
+#define LSI_ISTAT0_SEM    0x10
+#define LSI_ISTAT0_SIGP   0x20
+#define LSI_ISTAT0_SRST   0x40
+#define LSI_ISTAT0_ABRT   0x80
+
+#define LSI_ISTAT1_SI     0x01
+#define LSI_ISTAT1_SRUN   0x02
+#define LSI_ISTAT1_FLSH   0x04
+
+#define LSI_SSTAT0_SDP0   0x01
+#define LSI_SSTAT0_RST    0x02
+#define LSI_SSTAT0_WOA    0x04
+#define LSI_SSTAT0_LOA    0x08
+#define LSI_SSTAT0_AIP    0x10
+#define LSI_SSTAT0_OLF    0x20
+#define LSI_SSTAT0_ORF    0x40
+#define LSI_SSTAT0_ILF    0x80
+
+#define LSI_SIST0_PAR     0x01
+#define LSI_SIST0_RST     0x02
+#define LSI_SIST0_UDC     0x04
+#define LSI_SIST0_SGE     0x08
+#define LSI_SIST0_RSL     0x10
+#define LSI_SIST0_SEL     0x20
+#define LSI_SIST0_CMP     0x40
+#define LSI_SIST0_MA      0x80
+
+#define LSI_SIST1_HTH     0x01
+#define LSI_SIST1_GEN     0x02
+#define LSI_SIST1_STO     0x04
+#define LSI_SIST1_SBMC    0x10
+
+#define LSI_SOCL_IO       0x01
+#define LSI_SOCL_CD       0x02
+#define LSI_SOCL_MSG      0x04
+#define LSI_SOCL_ATN      0x08
+#define LSI_SOCL_SEL      0x10
+#define LSI_SOCL_BSY      0x20
+#define LSI_SOCL_ACK      0x40
+#define LSI_SOCL_REQ      0x80
+
+#define LSI_DSTAT_IID     0x01
+#define LSI_DSTAT_SIR     0x04
+#define LSI_DSTAT_SSI     0x08
+#define LSI_DSTAT_ABRT    0x10
+#define LSI_DSTAT_BF      0x20
+#define LSI_DSTAT_MDPE    0x40
+#define LSI_DSTAT_DFE     0x80
+
+#define LSI_DCNTL_COM     0x01
+#define LSI_DCNTL_IRQD    0x02
+#define LSI_DCNTL_STD     0x04
+#define LSI_DCNTL_IRQM    0x08
+#define LSI_DCNTL_SSM     0x10
+#define LSI_DCNTL_PFEN    0x20
+#define LSI_DCNTL_PFF     0x40
+#define LSI_DCNTL_CLSE    0x80
+
+#define LSI_DMODE_MAN     0x01
+#define LSI_DMODE_BOF     0x02
+#define LSI_DMODE_ERMP    0x04
+#define LSI_DMODE_ERL     0x08
+#define LSI_DMODE_DIOM    0x10
+#define LSI_DMODE_SIOM    0x20
+
+#define LSI_CTEST2_DACK   0x01
+#define LSI_CTEST2_DREQ   0x02
+#define LSI_CTEST2_TEOP   0x04
+#define LSI_CTEST2_PCICIE 0x08
+#define LSI_CTEST2_CM     0x10
+#define LSI_CTEST2_CIO    0x20
+#define LSI_CTEST2_SIGP   0x40
+#define LSI_CTEST2_DDIR   0x80
+
+#define LSI_CTEST5_BL2    0x04
+#define LSI_CTEST5_DDIR   0x08
+#define LSI_CTEST5_MASR   0x10
+#define LSI_CTEST5_DFSN   0x20
+#define LSI_CTEST5_BBCK   0x40
+#define LSI_CTEST5_ADCK   0x80
+
+#define LSI_CCNTL0_DILS   0x01
+#define LSI_CCNTL0_DISFC  0x10
+#define LSI_CCNTL0_ENNDJ  0x20
+#define LSI_CCNTL0_PMJCTL 0x40
+#define LSI_CCNTL0_ENPMJ  0x80
+
+#define LSI_CCNTL1_EN64DBMV  0x01
+#define LSI_CCNTL1_EN64TIBMV 0x02
+#define LSI_CCNTL1_64TIMOD   0x04
+#define LSI_CCNTL1_DDAC      0x08
+#define LSI_CCNTL1_ZMOD      0x80
+
+/* Enable Response to Reselection */
+#define LSI_SCID_RRE      0x60
+
+#define LSI_CCNTL1_40BIT (LSI_CCNTL1_EN64TIBMV|LSI_CCNTL1_64TIMOD)
+
+#define PHASE_DO          0
+#define PHASE_DI          1
+#define PHASE_CMD         2
+#define PHASE_ST          3
+#define PHASE_MO          6
+#define PHASE_MI          7
+#define PHASE_MASK        7
+
+/* Maximum length of MSG IN data.  */
+#define LSI_MAX_MSGIN_LEN 8
+
+/* Flag set if this is a tagged command.  */
+#define LSI_TAG_VALID     (1 << 16)
+
+typedef struct lsi_request {
+    SCSIRequest *req;
+    uint32_t tag;
+    uint32_t dma_len;
+    uint8_t *dma_buf;
+    uint32_t pending;
+    int out;
+    QTAILQ_ENTRY(lsi_request) next;
+} lsi_request;
+
+typedef struct {
+    /*< private >*/
+    PCIDevice parent_obj;
+    /*< public >*/
+
+    MemoryRegion mmio_io;
+    MemoryRegion ram_io;
+    MemoryRegion io_io;
+
+    int carry; /* ??? Should this be an a visible register somewhere?  */
+    int status;
+    /* Action to take at the end of a MSG IN phase.
+       0 = COMMAND, 1 = disconnect, 2 = DATA OUT, 3 = DATA IN.  */
+    int msg_action;
+    int msg_len;
+    uint8_t msg[LSI_MAX_MSGIN_LEN];
+    /* 0 if SCRIPTS are running or stopped.
+     * 1 if a Wait Reselect instruction has been issued.
+     * 2 if processing DMA from lsi_execute_script.
+     * 3 if a DMA operation is in progress.  */
+    int waiting;
+    SCSIBus bus;
+    int current_lun;
+    /* The tag is a combination of the device ID and the SCSI tag.  */
+    uint32_t select_tag;
+    int command_complete;
+    QTAILQ_HEAD(, lsi_request) queue;
+    lsi_request *current;
+
+    uint32_t dsa;
+    uint32_t temp;
+    uint32_t dnad;
+    uint32_t dbc;
+    uint8_t istat0;
+    uint8_t istat1;
+    uint8_t dcmd;
+    uint8_t dstat;
+    uint8_t dien;
+    uint8_t sist0;
+    uint8_t sist1;
+    uint8_t sien0;
+    uint8_t sien1;
+    uint8_t mbox0;
+    uint8_t mbox1;
+    uint8_t dfifo;
+    uint8_t ctest2;
+    uint8_t ctest3;
+    uint8_t ctest4;
+    uint8_t ctest5;
+    uint8_t ccntl0;
+    uint8_t ccntl1;
+    uint32_t dsp;
+    uint32_t dsps;
+    uint8_t dmode;
+    uint8_t dcntl;
+    uint8_t scntl0;
+    uint8_t scntl1;
+    uint8_t scntl2;
+    uint8_t scntl3;
+    uint8_t sstat0;
+    uint8_t sstat1;
+    uint8_t scid;
+    uint8_t sxfer;
+    uint8_t socl;
+    uint8_t sdid;
+    uint8_t ssid;
+    uint8_t sfbr;
+    uint8_t stest1;
+    uint8_t stest2;
+    uint8_t stest3;
+    uint8_t sidl;
+    uint8_t stime0;
+    uint8_t respid0;
+    uint8_t respid1;
+    uint32_t mmrs;
+    uint32_t mmws;
+    uint32_t sfs;
+    uint32_t drs;
+    uint32_t sbms;
+    uint32_t dbms;
+    uint32_t dnad64;
+    uint32_t pmjad1;
+    uint32_t pmjad2;
+    uint32_t rbc;
+    uint32_t ua;
+    uint32_t ia;
+    uint32_t sbc;
+    uint32_t csbc;
+    uint32_t scratch[18]; /* SCRATCHA-SCRATCHR */
+    uint8_t sbr;
+    uint32_t adder;
+
+    /* Script ram is stored as 32-bit words in host byteorder.  */
+    uint32_t script_ram[2048];
+} LSIState;
+
+#define TYPE_LSI53C810  "lsi53c810"
+#define TYPE_LSI53C895A "lsi53c895a"
+
+#define LSI53C895A(obj) \
+    OBJECT_CHECK(LSIState, (obj), TYPE_LSI53C895A)
+
+static inline int lsi_irq_on_rsl(LSIState *s)
+{
+    return (s->sien0 & LSI_SIST0_RSL) && (s->scid & LSI_SCID_RRE);
+}
+
+static void lsi_soft_reset(LSIState *s)
+{
+    DPRINTF("Reset\n");
+    s->carry = 0;
+
+    s->msg_action = 0;
+    s->msg_len = 0;
+    s->waiting = 0;
+    s->dsa = 0;
+    s->dnad = 0;
+    s->dbc = 0;
+    s->temp = 0;
+    memset(s->scratch, 0, sizeof(s->scratch));
+    s->istat0 = 0;
+    s->istat1 = 0;
+    s->dcmd = 0x40;
+    s->dstat = LSI_DSTAT_DFE;
+    s->dien = 0;
+    s->sist0 = 0;
+    s->sist1 = 0;
+    s->sien0 = 0;
+    s->sien1 = 0;
+    s->mbox0 = 0;
+    s->mbox1 = 0;
+    s->dfifo = 0;
+    s->ctest2 = LSI_CTEST2_DACK;
+    s->ctest3 = 0;
+    s->ctest4 = 0;
+    s->ctest5 = 0;
+    s->ccntl0 = 0;
+    s->ccntl1 = 0;
+    s->dsp = 0;
+    s->dsps = 0;
+    s->dmode = 0;
+    s->dcntl = 0;
+    s->scntl0 = 0xc0;
+    s->scntl1 = 0;
+    s->scntl2 = 0;
+    s->scntl3 = 0;
+    s->sstat0 = 0;
+    s->sstat1 = 0;
+    s->scid = 7;
+    s->sxfer = 0;
+    s->socl = 0;
+    s->sdid = 0;
+    s->ssid = 0;
+    s->stest1 = 0;
+    s->stest2 = 0;
+    s->stest3 = 0;
+    s->sidl = 0;
+    s->stime0 = 0;
+    s->respid0 = 0x80;
+    s->respid1 = 0;
+    s->mmrs = 0;
+    s->mmws = 0;
+    s->sfs = 0;
+    s->drs = 0;
+    s->sbms = 0;
+    s->dbms = 0;
+    s->dnad64 = 0;
+    s->pmjad1 = 0;
+    s->pmjad2 = 0;
+    s->rbc = 0;
+    s->ua = 0;
+    s->ia = 0;
+    s->sbc = 0;
+    s->csbc = 0;
+    s->sbr = 0;
+    assert(QTAILQ_EMPTY(&s->queue));
+    assert(!s->current);
+}
+
+static int lsi_dma_40bit(LSIState *s)
+{
+    if ((s->ccntl1 & LSI_CCNTL1_40BIT) == LSI_CCNTL1_40BIT)
+        return 1;
+    return 0;
+}
+
+static int lsi_dma_ti64bit(LSIState *s)
+{
+    if ((s->ccntl1 & LSI_CCNTL1_EN64TIBMV) == LSI_CCNTL1_EN64TIBMV)
+        return 1;
+    return 0;
+}
+
+static int lsi_dma_64bit(LSIState *s)
+{
+    if ((s->ccntl1 & LSI_CCNTL1_EN64DBMV) == LSI_CCNTL1_EN64DBMV)
+        return 1;
+    return 0;
+}
+
+static uint8_t lsi_reg_readb(LSIState *s, int offset);
+static void lsi_reg_writeb(LSIState *s, int offset, uint8_t val);
+static void lsi_execute_script(LSIState *s);
+static void lsi_reselect(LSIState *s, lsi_request *p);
+
+static inline uint32_t read_dword(LSIState *s, uint32_t addr)
+{
+    uint32_t buf;
+
+    pci_dma_read(PCI_DEVICE(s), addr, &buf, 4);
+    return cpu_to_le32(buf);
+}
+
+static void lsi_stop_script(LSIState *s)
+{
+    s->istat1 &= ~LSI_ISTAT1_SRUN;
+}
+
+static void lsi_update_irq(LSIState *s)
+{
+    PCIDevice *d = PCI_DEVICE(s);
+    int level;
+    static int last_level;
+    lsi_request *p;
+
+    /* It's unclear whether the DIP/SIP bits should be cleared when the
+       Interrupt Status Registers are cleared or when istat0 is read.
+       We currently do the formwer, which seems to work.  */
+    level = 0;
+    if (s->dstat) {
+        if (s->dstat & s->dien)
+            level = 1;
+        s->istat0 |= LSI_ISTAT0_DIP;
+    } else {
+        s->istat0 &= ~LSI_ISTAT0_DIP;
+    }
+
+    if (s->sist0 || s->sist1) {
+        if ((s->sist0 & s->sien0) || (s->sist1 & s->sien1))
+            level = 1;
+        s->istat0 |= LSI_ISTAT0_SIP;
+    } else {
+        s->istat0 &= ~LSI_ISTAT0_SIP;
+    }
+    if (s->istat0 & LSI_ISTAT0_INTF)
+        level = 1;
+
+    if (level != last_level) {
+        DPRINTF("Update IRQ level %d dstat %02x sist %02x%02x\n",
+                level, s->dstat, s->sist1, s->sist0);
+        last_level = level;
+    }
+    pci_set_irq(d, level);
+
+    if (!level && lsi_irq_on_rsl(s) && !(s->scntl1 & LSI_SCNTL1_CON)) {
+        DPRINTF("Handled IRQs & disconnected, looking for pending "
+                "processes\n");
+        QTAILQ_FOREACH(p, &s->queue, next) {
+            if (p->pending) {
+                lsi_reselect(s, p);
+                break;
+            }
+        }
+    }
+}
+
+/* Stop SCRIPTS execution and raise a SCSI interrupt.  */
+static void lsi_script_scsi_interrupt(LSIState *s, int stat0, int stat1)
+{
+    uint32_t mask0;
+    uint32_t mask1;
+
+    DPRINTF("SCSI Interrupt 0x%02x%02x prev 0x%02x%02x\n",
+            stat1, stat0, s->sist1, s->sist0);
+    s->sist0 |= stat0;
+    s->sist1 |= stat1;
+    /* Stop processor on fatal or unmasked interrupt.  As a special hack
+       we don't stop processing when raising STO.  Instead continue
+       execution and stop at the next insn that accesses the SCSI bus.  */
+    mask0 = s->sien0 | ~(LSI_SIST0_CMP | LSI_SIST0_SEL | LSI_SIST0_RSL);
+    mask1 = s->sien1 | ~(LSI_SIST1_GEN | LSI_SIST1_HTH);
+    mask1 &= ~LSI_SIST1_STO;
+    if (s->sist0 & mask0 || s->sist1 & mask1) {
+        lsi_stop_script(s);
+    }
+    lsi_update_irq(s);
+}
+
+/* Stop SCRIPTS execution and raise a DMA interrupt.  */
+static void lsi_script_dma_interrupt(LSIState *s, int stat)
+{
+    DPRINTF("DMA Interrupt 0x%x prev 0x%x\n", stat, s->dstat);
+    s->dstat |= stat;
+    lsi_update_irq(s);
+    lsi_stop_script(s);
+}
+
+static inline void lsi_set_phase(LSIState *s, int phase)
+{
+    s->sstat1 = (s->sstat1 & ~PHASE_MASK) | phase;
+}
+
+static void lsi_bad_phase(LSIState *s, int out, int new_phase)
+{
+    /* Trigger a phase mismatch.  */
+    if (s->ccntl0 & LSI_CCNTL0_ENPMJ) {
+        if ((s->ccntl0 & LSI_CCNTL0_PMJCTL)) {
+            s->dsp = out ? s->pmjad1 : s->pmjad2;
+        } else {
+            s->dsp = (s->scntl2 & LSI_SCNTL2_WSR ? s->pmjad2 : s->pmjad1);
+        }
+        DPRINTF("Data phase mismatch jump to %08x\n", s->dsp);
+    } else {
+        DPRINTF("Phase mismatch interrupt\n");
+        lsi_script_scsi_interrupt(s, LSI_SIST0_MA, 0);
+        lsi_stop_script(s);
+    }
+    lsi_set_phase(s, new_phase);
+}
+
+
+/* Resume SCRIPTS execution after a DMA operation.  */
+static void lsi_resume_script(LSIState *s)
+{
+    if (s->waiting != 2) {
+        s->waiting = 0;
+        lsi_execute_script(s);
+    } else {
+        s->waiting = 0;
+    }
+}
+
+static void lsi_disconnect(LSIState *s)
+{
+    s->scntl1 &= ~LSI_SCNTL1_CON;
+    s->sstat1 &= ~PHASE_MASK;
+}
+
+static void lsi_bad_selection(LSIState *s, uint32_t id)
+{
+    DPRINTF("Selected absent target %d\n", id);
+    lsi_script_scsi_interrupt(s, 0, LSI_SIST1_STO);
+    lsi_disconnect(s);
+}
+
+/* Initiate a SCSI layer data transfer.  */
+static void lsi_do_dma(LSIState *s, int out)
+{
+    PCIDevice *pci_dev;
+    uint32_t count;
+    dma_addr_t addr;
+    SCSIDevice *dev;
+
+    assert(s->current);
+    if (!s->current->dma_len) {
+        /* Wait until data is available.  */
+        DPRINTF("DMA no data available\n");
+        return;
+    }
+
+    pci_dev = PCI_DEVICE(s);
+    dev = s->current->req->dev;
+    assert(dev);
+
+    count = s->dbc;
+    if (count > s->current->dma_len)
+        count = s->current->dma_len;
+
+    addr = s->dnad;
+    /* both 40 and Table Indirect 64-bit DMAs store upper bits in dnad64 */
+    if (lsi_dma_40bit(s) || lsi_dma_ti64bit(s))
+        addr |= ((uint64_t)s->dnad64 << 32);
+    else if (s->dbms)
+        addr |= ((uint64_t)s->dbms << 32);
+    else if (s->sbms)
+        addr |= ((uint64_t)s->sbms << 32);
+
+    DPRINTF("DMA addr=0x" DMA_ADDR_FMT " len=%d\n", addr, count);
+    s->csbc += count;
+    s->dnad += count;
+    s->dbc -= count;
+     if (s->current->dma_buf == NULL) {
+        s->current->dma_buf = scsi_req_get_buf(s->current->req);
+    }
+    /* ??? Set SFBR to first data byte.  */
+    if (out) {
+        pci_dma_read(pci_dev, addr, s->current->dma_buf, count);
+    } else {
+        pci_dma_write(pci_dev, addr, s->current->dma_buf, count);
+    }
+    s->current->dma_len -= count;
+    if (s->current->dma_len == 0) {
+        s->current->dma_buf = NULL;
+        scsi_req_continue(s->current->req);
+    } else {
+        s->current->dma_buf += count;
+        lsi_resume_script(s);
+    }
+}
+
+
+/* Add a command to the queue.  */
+static void lsi_queue_command(LSIState *s)
+{
+    lsi_request *p = s->current;
+
+    DPRINTF("Queueing tag=0x%x\n", p->tag);
+    assert(s->current != NULL);
+    assert(s->current->dma_len == 0);
+    QTAILQ_INSERT_TAIL(&s->queue, s->current, next);
+    s->current = NULL;
+
+    p->pending = 0;
+    p->out = (s->sstat1 & PHASE_MASK) == PHASE_DO;
+}
+
+/* Queue a byte for a MSG IN phase.  */
+static void lsi_add_msg_byte(LSIState *s, uint8_t data)
+{
+    if (s->msg_len >= LSI_MAX_MSGIN_LEN) {
+        BADF("MSG IN data too long\n");
+    } else {
+        DPRINTF("MSG IN 0x%02x\n", data);
+        s->msg[s->msg_len++] = data;
+    }
+}
+
+/* Perform reselection to continue a command.  */
+static void lsi_reselect(LSIState *s, lsi_request *p)
+{
+    int id;
+
+    assert(s->current == NULL);
+    QTAILQ_REMOVE(&s->queue, p, next);
+    s->current = p;
+
+    id = (p->tag >> 8) & 0xf;
+    s->ssid = id | 0x80;
+    /* LSI53C700 Family Compatibility, see LSI53C895A 4-73 */
+    if (!(s->dcntl & LSI_DCNTL_COM)) {
+        s->sfbr = 1 << (id & 0x7);
+    }
+    DPRINTF("Reselected target %d\n", id);
+    s->scntl1 |= LSI_SCNTL1_CON;
+    lsi_set_phase(s, PHASE_MI);
+    s->msg_action = p->out ? 2 : 3;
+    s->current->dma_len = p->pending;
+    lsi_add_msg_byte(s, 0x80);
+    if (s->current->tag & LSI_TAG_VALID) {
+        lsi_add_msg_byte(s, 0x20);
+        lsi_add_msg_byte(s, p->tag & 0xff);
+    }
+
+    if (lsi_irq_on_rsl(s)) {
+        lsi_script_scsi_interrupt(s, LSI_SIST0_RSL, 0);
+    }
+}
+
+static lsi_request *lsi_find_by_tag(LSIState *s, uint32_t tag)
+{
+    lsi_request *p;
+
+    QTAILQ_FOREACH(p, &s->queue, next) {
+        if (p->tag == tag) {
+            return p;
+        }
+    }
+
+    return NULL;
+}
+
+static void lsi_request_free(LSIState *s, lsi_request *p)
+{
+    if (p == s->current) {
+        s->current = NULL;
+    } else {
+        QTAILQ_REMOVE(&s->queue, p, next);
+    }
+    g_free(p);
+}
+
+static void lsi_request_cancelled(SCSIRequest *req)
+{
+    LSIState *s = LSI53C895A(req->bus->qbus.parent);
+    lsi_request *p = req->hba_private;
+
+    req->hba_private = NULL;
+    lsi_request_free(s, p);
+    scsi_req_unref(req);
+}
+
+/* Record that data is available for a queued command.  Returns zero if
+   the device was reselected, nonzero if the IO is deferred.  */
+static int lsi_queue_req(LSIState *s, SCSIRequest *req, uint32_t len)
+{
+    lsi_request *p = req->hba_private;
+
+    if (p->pending) {
+        BADF("Multiple IO pending for request %p\n", p);
+    }
+    p->pending = len;
+    /* Reselect if waiting for it, or if reselection triggers an IRQ
+       and the bus is free.
+       Since no interrupt stacking is implemented in the emulation, it
+       is also required that there are no pending interrupts waiting
+       for service from the device driver. */
+    if (s->waiting == 1 ||
+        (lsi_irq_on_rsl(s) && !(s->scntl1 & LSI_SCNTL1_CON) &&
+         !(s->istat0 & (LSI_ISTAT0_SIP | LSI_ISTAT0_DIP)))) {
+        /* Reselect device.  */
+        lsi_reselect(s, p);
+        return 0;
+    } else {
+        DPRINTF("Queueing IO tag=0x%x\n", p->tag);
+        p->pending = len;
+        return 1;
+    }
+}
+
+ /* Callback to indicate that the SCSI layer has completed a command.  */
+static void lsi_command_complete(SCSIRequest *req, uint32_t status, size_t resid)
+{
+    LSIState *s = LSI53C895A(req->bus->qbus.parent);
+    int out;
+
+    out = (s->sstat1 & PHASE_MASK) == PHASE_DO;
+    DPRINTF("Command complete status=%d\n", (int)status);
+    s->status = status;
+    s->command_complete = 2;
+    if (s->waiting && s->dbc != 0) {
+        /* Raise phase mismatch for short transfers.  */
+        lsi_bad_phase(s, out, PHASE_ST);
+    } else {
+        lsi_set_phase(s, PHASE_ST);
+    }
+
+    if (req->hba_private == s->current) {
+        req->hba_private = NULL;
+        lsi_request_free(s, s->current);
+        scsi_req_unref(req);
+    }
+    lsi_resume_script(s);
+}
+
+ /* Callback to indicate that the SCSI layer has completed a transfer.  */
+static void lsi_transfer_data(SCSIRequest *req, uint32_t len)
+{
+    LSIState *s = LSI53C895A(req->bus->qbus.parent);
+    int out;
+
+    assert(req->hba_private);
+    if (s->waiting == 1 || req->hba_private != s->current ||
+        (lsi_irq_on_rsl(s) && !(s->scntl1 & LSI_SCNTL1_CON))) {
+        if (lsi_queue_req(s, req, len)) {
+            return;
+        }
+    }
+
+    out = (s->sstat1 & PHASE_MASK) == PHASE_DO;
+
+    /* host adapter (re)connected */
+    DPRINTF("Data ready tag=0x%x len=%d\n", req->tag, len);
+    s->current->dma_len = len;
+    s->command_complete = 1;
+    if (s->waiting) {
+        if (s->waiting == 1 || s->dbc == 0) {
+            lsi_resume_script(s);
+        } else {
+            lsi_do_dma(s, out);
+        }
+    }
+}
+
+static void lsi_do_command(LSIState *s)
+{
+    SCSIDevice *dev;
+    uint8_t buf[16];
+    uint32_t id;
+    int n;
+
+    DPRINTF("Send command len=%d\n", s->dbc);
+    if (s->dbc > 16)
+        s->dbc = 16;
+    pci_dma_read(PCI_DEVICE(s), s->dnad, buf, s->dbc);
+    s->sfbr = buf[0];
+    s->command_complete = 0;
+
+    id = (s->select_tag >> 8) & 0xf;
+    dev = scsi_device_find(&s->bus, 0, id, s->current_lun);
+    if (!dev) {
+        lsi_bad_selection(s, id);
+        return;
+    }
+
+    assert(s->current == NULL);
+    s->current = g_new0(lsi_request, 1);
+    s->current->tag = s->select_tag;
+    s->current->req = scsi_req_new(dev, s->current->tag, s->current_lun, buf,
+                                   s->current);
+
+    n = scsi_req_enqueue(s->current->req);
+    if (n) {
+        if (n > 0) {
+            lsi_set_phase(s, PHASE_DI);
+        } else if (n < 0) {
+            lsi_set_phase(s, PHASE_DO);
+        }
+        scsi_req_continue(s->current->req);
+    }
+    if (!s->command_complete) {
+        if (n) {
+            /* Command did not complete immediately so disconnect.  */
+            lsi_add_msg_byte(s, 2); /* SAVE DATA POINTER */
+            lsi_add_msg_byte(s, 4); /* DISCONNECT */
+            /* wait data */
+            lsi_set_phase(s, PHASE_MI);
+            s->msg_action = 1;
+            lsi_queue_command(s);
+        } else {
+            /* wait command complete */
+            lsi_set_phase(s, PHASE_DI);
+        }
+    }
+}
+
+static void lsi_do_status(LSIState *s)
+{
+    uint8_t status;
+    DPRINTF("Get status len=%d status=%d\n", s->dbc, s->status);
+    if (s->dbc != 1)
+        BADF("Bad Status move\n");
+    s->dbc = 1;
+    status = s->status;
+    s->sfbr = status;
+    pci_dma_write(PCI_DEVICE(s), s->dnad, &status, 1);
+    lsi_set_phase(s, PHASE_MI);
+    s->msg_action = 1;
+    lsi_add_msg_byte(s, 0); /* COMMAND COMPLETE */
+}
+
+static void lsi_do_msgin(LSIState *s)
+{
+    int len;
+    DPRINTF("Message in len=%d/%d\n", s->dbc, s->msg_len);
+    s->sfbr = s->msg[0];
+    len = s->msg_len;
+    if (len > s->dbc)
+        len = s->dbc;
+    pci_dma_write(PCI_DEVICE(s), s->dnad, s->msg, len);
+    /* Linux drivers rely on the last byte being in the SIDL.  */
+    s->sidl = s->msg[len - 1];
+    s->msg_len -= len;
+    if (s->msg_len) {
+        memmove(s->msg, s->msg + len, s->msg_len);
+    } else {
+        /* ??? Check if ATN (not yet implemented) is asserted and maybe
+           switch to PHASE_MO.  */
+        switch (s->msg_action) {
+        case 0:
+            lsi_set_phase(s, PHASE_CMD);
+            break;
+        case 1:
+            lsi_disconnect(s);
+            break;
+        case 2:
+            lsi_set_phase(s, PHASE_DO);
+            break;
+        case 3:
+            lsi_set_phase(s, PHASE_DI);
+            break;
+        default:
+            abort();
+        }
+    }
+}
+
+/* Read the next byte during a MSGOUT phase.  */
+static uint8_t lsi_get_msgbyte(LSIState *s)
+{
+    uint8_t data;
+    pci_dma_read(PCI_DEVICE(s), s->dnad, &data, 1);
+    s->dnad++;
+    s->dbc--;
+    return data;
+}
+
+/* Skip the next n bytes during a MSGOUT phase. */
+static void lsi_skip_msgbytes(LSIState *s, unsigned int n)
+{
+    s->dnad += n;
+    s->dbc  -= n;
+}
+
+static void lsi_do_msgout(LSIState *s)
+{
+    uint8_t msg;
+    int len;
+    uint32_t current_tag;
+    lsi_request *current_req, *p, *p_next;
+
+    if (s->current) {
+        current_tag = s->current->tag;
+        current_req = s->current;
+    } else {
+        current_tag = s->select_tag;
+        current_req = lsi_find_by_tag(s, current_tag);
+    }
+
+    DPRINTF("MSG out len=%d\n", s->dbc);
+    while (s->dbc) {
+        msg = lsi_get_msgbyte(s);
+        s->sfbr = msg;
+
+        switch (msg) {
+        case 0x04:
+            DPRINTF("MSG: Disconnect\n");
+            lsi_disconnect(s);
+            break;
+        case 0x08:
+            DPRINTF("MSG: No Operation\n");
+            lsi_set_phase(s, PHASE_CMD);
+            break;
+        case 0x01:
+            len = lsi_get_msgbyte(s);
+            msg = lsi_get_msgbyte(s);
+            (void)len; /* avoid a warning about unused variable*/
+            DPRINTF("Extended message 0x%x (len %d)\n", msg, len);
+            switch (msg) {
+            case 1:
+                DPRINTF("SDTR (ignored)\n");
+                lsi_skip_msgbytes(s, 2);
+                break;
+            case 3:
+                DPRINTF("WDTR (ignored)\n");
+                lsi_skip_msgbytes(s, 1);
+                break;
+            default:
+                goto bad;
+            }
+            break;
+        case 0x20: /* SIMPLE queue */
+            s->select_tag |= lsi_get_msgbyte(s) | LSI_TAG_VALID;
+            DPRINTF("SIMPLE queue tag=0x%x\n", s->select_tag & 0xff);
+            break;
+        case 0x21: /* HEAD of queue */
+            BADF("HEAD queue not implemented\n");
+            s->select_tag |= lsi_get_msgbyte(s) | LSI_TAG_VALID;
+            break;
+        case 0x22: /* ORDERED queue */
+            BADF("ORDERED queue not implemented\n");
+            s->select_tag |= lsi_get_msgbyte(s) | LSI_TAG_VALID;
+            break;
+        case 0x0d:
+            /* The ABORT TAG message clears the current I/O process only. */
+            DPRINTF("MSG: ABORT TAG tag=0x%x\n", current_tag);
+            if (current_req) {
+                scsi_req_cancel(current_req->req);
+            }
+            lsi_disconnect(s);
+            break;
+        case 0x06:
+        case 0x0e:
+        case 0x0c:
+            /* The ABORT message clears all I/O processes for the selecting
+               initiator on the specified logical unit of the target. */
+            if (msg == 0x06) {
+                DPRINTF("MSG: ABORT tag=0x%x\n", current_tag);
+            }
+            /* The CLEAR QUEUE message clears all I/O processes for all
+               initiators on the specified logical unit of the target. */
+            if (msg == 0x0e) {
+                DPRINTF("MSG: CLEAR QUEUE tag=0x%x\n", current_tag);
+            }
+            /* The BUS DEVICE RESET message clears all I/O processes for all
+               initiators on all logical units of the target. */
+            if (msg == 0x0c) {
+                DPRINTF("MSG: BUS DEVICE RESET tag=0x%x\n", current_tag);
+            }
+
+            /* clear the current I/O process */
+            if (s->current) {
+                scsi_req_cancel(s->current->req);
+            }
+
+            /* As the current implemented devices scsi_disk and scsi_generic
+               only support one LUN, we don't need to keep track of LUNs.
+               Clearing I/O processes for other initiators could be possible
+               for scsi_generic by sending a SG_SCSI_RESET to the /dev/sgX
+               device, but this is currently not implemented (and seems not
+               to be really necessary). So let's simply clear all queued
+               commands for the current device: */
+            QTAILQ_FOREACH_SAFE(p, &s->queue, next, p_next) {
+                if ((p->tag & 0x0000ff00) == (current_tag & 0x0000ff00)) {
+                    scsi_req_cancel(p->req);
+                }
+            }
+
+            lsi_disconnect(s);
+            break;
+        default:
+            if ((msg & 0x80) == 0) {
+                goto bad;
+            }
+            s->current_lun = msg & 7;
+            DPRINTF("Select LUN %d\n", s->current_lun);
+            lsi_set_phase(s, PHASE_CMD);
+            break;
+        }
+    }
+    return;
+bad:
+    BADF("Unimplemented message 0x%02x\n", msg);
+    lsi_set_phase(s, PHASE_MI);
+    lsi_add_msg_byte(s, 7); /* MESSAGE REJECT */
+    s->msg_action = 0;
+}
+
+#define LSI_BUF_SIZE 4096
+static void lsi_memcpy(LSIState *s, uint32_t dest, uint32_t src, int count)
+{
+    PCIDevice *d = PCI_DEVICE(s);
+    int n;
+    uint8_t buf[LSI_BUF_SIZE];
+
+    DPRINTF("memcpy dest 0x%08x src 0x%08x count %d\n", dest, src, count);
+    while (count) {
+        n = (count > LSI_BUF_SIZE) ? LSI_BUF_SIZE : count;
+        pci_dma_read(d, src, buf, n);
+        pci_dma_write(d, dest, buf, n);
+        src += n;
+        dest += n;
+        count -= n;
+    }
+}
+
+static void lsi_wait_reselect(LSIState *s)
+{
+    lsi_request *p;
+
+    DPRINTF("Wait Reselect\n");
+
+    QTAILQ_FOREACH(p, &s->queue, next) {
+        if (p->pending) {
+            lsi_reselect(s, p);
+            break;
+        }
+    }
+    if (s->current == NULL) {
+        s->waiting = 1;
+    }
+}
+
+static void lsi_execute_script(LSIState *s)
+{
+    PCIDevice *pci_dev = PCI_DEVICE(s);
+    uint32_t insn;
+    uint32_t addr, addr_high;
+    int opcode;
+    int insn_processed = 0;
+
+    s->istat1 |= LSI_ISTAT1_SRUN;
+again:
+    insn_processed++;
+    insn = read_dword(s, s->dsp);
+    if (!insn) {
+        /* If we receive an empty opcode increment the DSP by 4 bytes
+           instead of 8 and execute the next opcode at that location */
+        s->dsp += 4;
+        goto again;
+    }
+    addr = read_dword(s, s->dsp + 4);
+    addr_high = 0;
+    DPRINTF("SCRIPTS dsp=%08x opcode %08x arg %08x\n", s->dsp, insn, addr);
+    s->dsps = addr;
+    s->dcmd = insn >> 24;
+    s->dsp += 8;
+    switch (insn >> 30) {
+    case 0: /* Block move.  */
+        if (s->sist1 & LSI_SIST1_STO) {
+            DPRINTF("Delayed select timeout\n");
+            lsi_stop_script(s);
+            break;
+        }
+        s->dbc = insn & 0xffffff;
+        s->rbc = s->dbc;
+        /* ??? Set ESA.  */
+        s->ia = s->dsp - 8;
+        if (insn & (1 << 29)) {
+            /* Indirect addressing.  */
+            addr = read_dword(s, addr);
+        } else if (insn & (1 << 28)) {
+            uint32_t buf[2];
+            int32_t offset;
+            /* Table indirect addressing.  */
+
+            /* 32-bit Table indirect */
+            offset = sextract32(addr, 0, 24);
+            pci_dma_read(pci_dev, s->dsa + offset, buf, 8);
+            /* byte count is stored in bits 0:23 only */
+            s->dbc = cpu_to_le32(buf[0]) & 0xffffff;
+            s->rbc = s->dbc;
+            addr = cpu_to_le32(buf[1]);
+
+            /* 40-bit DMA, upper addr bits [39:32] stored in first DWORD of
+             * table, bits [31:24] */
+            if (lsi_dma_40bit(s))
+                addr_high = cpu_to_le32(buf[0]) >> 24;
+            else if (lsi_dma_ti64bit(s)) {
+                int selector = (cpu_to_le32(buf[0]) >> 24) & 0x1f;
+                switch (selector) {
+                case 0 ... 0x0f:
+                    /* offset index into scratch registers since
+                     * TI64 mode can use registers C to R */
+                    addr_high = s->scratch[2 + selector];
+                    break;
+                case 0x10:
+                    addr_high = s->mmrs;
+                    break;
+                case 0x11:
+                    addr_high = s->mmws;
+                    break;
+                case 0x12:
+                    addr_high = s->sfs;
+                    break;
+                case 0x13:
+                    addr_high = s->drs;
+                    break;
+                case 0x14:
+                    addr_high = s->sbms;
+                    break;
+                case 0x15:
+                    addr_high = s->dbms;
+                    break;
+                default:
+                    BADF("Illegal selector specified (0x%x > 0x15)"
+                         " for 64-bit DMA block move", selector);
+                    break;
+                }
+            }
+        } else if (lsi_dma_64bit(s)) {
+            /* fetch a 3rd dword if 64-bit direct move is enabled and
+               only if we're not doing table indirect or indirect addressing */
+            s->dbms = read_dword(s, s->dsp);
+            s->dsp += 4;
+            s->ia = s->dsp - 12;
+        }
+        if ((s->sstat1 & PHASE_MASK) != ((insn >> 24) & 7)) {
+            DPRINTF("Wrong phase got %d expected %d\n",
+                    s->sstat1 & PHASE_MASK, (insn >> 24) & 7);
+            lsi_script_scsi_interrupt(s, LSI_SIST0_MA, 0);
+            break;
+        }
+        s->dnad = addr;
+        s->dnad64 = addr_high;
+        switch (s->sstat1 & 0x7) {
+        case PHASE_DO:
+            s->waiting = 2;
+            lsi_do_dma(s, 1);
+            if (s->waiting)
+                s->waiting = 3;
+            break;
+        case PHASE_DI:
+            s->waiting = 2;
+            lsi_do_dma(s, 0);
+            if (s->waiting)
+                s->waiting = 3;
+            break;
+        case PHASE_CMD:
+            lsi_do_command(s);
+            break;
+        case PHASE_ST:
+            lsi_do_status(s);
+            break;
+        case PHASE_MO:
+            lsi_do_msgout(s);
+            break;
+        case PHASE_MI:
+            lsi_do_msgin(s);
+            break;
+        default:
+            BADF("Unimplemented phase %d\n", s->sstat1 & PHASE_MASK);
+            exit(1);
+        }
+        s->dfifo = s->dbc & 0xff;
+        s->ctest5 = (s->ctest5 & 0xfc) | ((s->dbc >> 8) & 3);
+        s->sbc = s->dbc;
+        s->rbc -= s->dbc;
+        s->ua = addr + s->dbc;
+        break;
+
+    case 1: /* IO or Read/Write instruction.  */
+        opcode = (insn >> 27) & 7;
+        if (opcode < 5) {
+            uint32_t id;
+
+            if (insn & (1 << 25)) {
+                id = read_dword(s, s->dsa + sextract32(insn, 0, 24));
+            } else {
+                id = insn;
+            }
+            id = (id >> 16) & 0xf;
+            if (insn & (1 << 26)) {
+                addr = s->dsp + sextract32(addr, 0, 24);
+            }
+            s->dnad = addr;
+            switch (opcode) {
+            case 0: /* Select */
+                s->sdid = id;
+                if (s->scntl1 & LSI_SCNTL1_CON) {
+                    DPRINTF("Already reselected, jumping to alternative address\n");
+                    s->dsp = s->dnad;
+                    break;
+                }
+                s->sstat0 |= LSI_SSTAT0_WOA;
+                s->scntl1 &= ~LSI_SCNTL1_IARB;
+                if (!scsi_device_find(&s->bus, 0, id, 0)) {
+                    lsi_bad_selection(s, id);
+                    break;
+                }
+                DPRINTF("Selected target %d%s\n",
+                        id, insn & (1 << 3) ? " ATN" : "");
+                /* ??? Linux drivers compain when this is set.  Maybe
+                   it only applies in low-level mode (unimplemented).
+                lsi_script_scsi_interrupt(s, LSI_SIST0_CMP, 0); */
+                s->select_tag = id << 8;
+                s->scntl1 |= LSI_SCNTL1_CON;
+                if (insn & (1 << 3)) {
+                    s->socl |= LSI_SOCL_ATN;
+                }
+                lsi_set_phase(s, PHASE_MO);
+                break;
+            case 1: /* Disconnect */
+                DPRINTF("Wait Disconnect\n");
+                s->scntl1 &= ~LSI_SCNTL1_CON;
+                break;
+            case 2: /* Wait Reselect */
+                if (!lsi_irq_on_rsl(s)) {
+                    lsi_wait_reselect(s);
+                }
+                break;
+            case 3: /* Set */
+                DPRINTF("Set%s%s%s%s\n",
+                        insn & (1 << 3) ? " ATN" : "",
+                        insn & (1 << 6) ? " ACK" : "",
+                        insn & (1 << 9) ? " TM" : "",
+                        insn & (1 << 10) ? " CC" : "");
+                if (insn & (1 << 3)) {
+                    s->socl |= LSI_SOCL_ATN;
+                    lsi_set_phase(s, PHASE_MO);
+                }
+                if (insn & (1 << 9)) {
+                    BADF("Target mode not implemented\n");
+                    exit(1);
+                }
+                if (insn & (1 << 10))
+                    s->carry = 1;
+                break;
+            case 4: /* Clear */
+                DPRINTF("Clear%s%s%s%s\n",
+                        insn & (1 << 3) ? " ATN" : "",
+                        insn & (1 << 6) ? " ACK" : "",
+                        insn & (1 << 9) ? " TM" : "",
+                        insn & (1 << 10) ? " CC" : "");
+                if (insn & (1 << 3)) {
+                    s->socl &= ~LSI_SOCL_ATN;
+                }
+                if (insn & (1 << 10))
+                    s->carry = 0;
+                break;
+            }
+        } else {
+            uint8_t op0;
+            uint8_t op1;
+            uint8_t data8;
+            int reg;
+            int operator;
+#ifdef DEBUG_LSI
+            static const char *opcode_names[3] =
+                {"Write", "Read", "Read-Modify-Write"};
+            static const char *operator_names[8] =
+                {"MOV", "SHL", "OR", "XOR", "AND", "SHR", "ADD", "ADC"};
+#endif
+
+            reg = ((insn >> 16) & 0x7f) | (insn & 0x80);
+            data8 = (insn >> 8) & 0xff;
+            opcode = (insn >> 27) & 7;
+            operator = (insn >> 24) & 7;
+            DPRINTF("%s reg 0x%x %s data8=0x%02x sfbr=0x%02x%s\n",
+                    opcode_names[opcode - 5], reg,
+                    operator_names[operator], data8, s->sfbr,
+                    (insn & (1 << 23)) ? " SFBR" : "");
+            op0 = op1 = 0;
+            switch (opcode) {
+            case 5: /* From SFBR */
+                op0 = s->sfbr;
+                op1 = data8;
+                break;
+            case 6: /* To SFBR */
+                if (operator)
+                    op0 = lsi_reg_readb(s, reg);
+                op1 = data8;
+                break;
+            case 7: /* Read-modify-write */
+                if (operator)
+                    op0 = lsi_reg_readb(s, reg);
+                if (insn & (1 << 23)) {
+                    op1 = s->sfbr;
+                } else {
+                    op1 = data8;
+                }
+                break;
+            }
+
+            switch (operator) {
+            case 0: /* move */
+                op0 = op1;
+                break;
+            case 1: /* Shift left */
+                op1 = op0 >> 7;
+                op0 = (op0 << 1) | s->carry;
+                s->carry = op1;
+                break;
+            case 2: /* OR */
+                op0 |= op1;
+                break;
+            case 3: /* XOR */
+                op0 ^= op1;
+                break;
+            case 4: /* AND */
+                op0 &= op1;
+                break;
+            case 5: /* SHR */
+                op1 = op0 & 1;
+                op0 = (op0 >> 1) | (s->carry << 7);
+                s->carry = op1;
+                break;
+            case 6: /* ADD */
+                op0 += op1;
+                s->carry = op0 < op1;
+                break;
+            case 7: /* ADC */
+                op0 += op1 + s->carry;
+                if (s->carry)
+                    s->carry = op0 <= op1;
+                else
+                    s->carry = op0 < op1;
+                break;
+            }
+
+            switch (opcode) {
+            case 5: /* From SFBR */
+            case 7: /* Read-modify-write */
+                lsi_reg_writeb(s, reg, op0);
+                break;
+            case 6: /* To SFBR */
+                s->sfbr = op0;
+                break;
+            }
+        }
+        break;
+
+    case 2: /* Transfer Control.  */
+        {
+            int cond;
+            int jmp;
+
+            if ((insn & 0x002e0000) == 0) {
+                DPRINTF("NOP\n");
+                break;
+            }
+            if (s->sist1 & LSI_SIST1_STO) {
+                DPRINTF("Delayed select timeout\n");
+                lsi_stop_script(s);
+                break;
+            }
+            cond = jmp = (insn & (1 << 19)) != 0;
+            if (cond == jmp && (insn & (1 << 21))) {
+                DPRINTF("Compare carry %d\n", s->carry == jmp);
+                cond = s->carry != 0;
+            }
+            if (cond == jmp && (insn & (1 << 17))) {
+                DPRINTF("Compare phase %d %c= %d\n",
+                        (s->sstat1 & PHASE_MASK),
+                        jmp ? '=' : '!',
+                        ((insn >> 24) & 7));
+                cond = (s->sstat1 & PHASE_MASK) == ((insn >> 24) & 7);
+            }
+            if (cond == jmp && (insn & (1 << 18))) {
+                uint8_t mask;
+
+                mask = (~insn >> 8) & 0xff;
+                DPRINTF("Compare data 0x%x & 0x%x %c= 0x%x\n",
+                        s->sfbr, mask, jmp ? '=' : '!', insn & mask);
+                cond = (s->sfbr & mask) == (insn & mask);
+            }
+            if (cond == jmp) {
+                if (insn & (1 << 23)) {
+                    /* Relative address.  */
+                    addr = s->dsp + sextract32(addr, 0, 24);
+                }
+                switch ((insn >> 27) & 7) {
+                case 0: /* Jump */
+                    DPRINTF("Jump to 0x%08x\n", addr);
+                    s->adder = addr;
+                    s->dsp = addr;
+                    break;
+                case 1: /* Call */
+                    DPRINTF("Call 0x%08x\n", addr);
+                    s->temp = s->dsp;
+                    s->dsp = addr;
+                    break;
+                case 2: /* Return */
+                    DPRINTF("Return to 0x%08x\n", s->temp);
+                    s->dsp = s->temp;
+                    break;
+                case 3: /* Interrupt */
+                    DPRINTF("Interrupt 0x%08x\n", s->dsps);
+                    if ((insn & (1 << 20)) != 0) {
+                        s->istat0 |= LSI_ISTAT0_INTF;
+                        lsi_update_irq(s);
+                    } else {
+                        lsi_script_dma_interrupt(s, LSI_DSTAT_SIR);
+                    }
+                    break;
+                default:
+                    DPRINTF("Illegal transfer control\n");
+                    lsi_script_dma_interrupt(s, LSI_DSTAT_IID);
+                    break;
+                }
+            } else {
+                DPRINTF("Control condition failed\n");
+            }
+        }
+        break;
+
+    case 3:
+        if ((insn & (1 << 29)) == 0) {
+            /* Memory move.  */
+            uint32_t dest;
+            /* ??? The docs imply the destination address is loaded into
+               the TEMP register.  However the Linux drivers rely on
+               the value being presrved.  */
+            dest = read_dword(s, s->dsp);
+            s->dsp += 4;
+            lsi_memcpy(s, dest, addr, insn & 0xffffff);
+        } else {
+            uint8_t data[7];
+            int reg;
+            int n;
+            int i;
+
+            if (insn & (1 << 28)) {
+                addr = s->dsa + sextract32(addr, 0, 24);
+            }
+            n = (insn & 7);
+            reg = (insn >> 16) & 0xff;
+            if (insn & (1 << 24)) {
+                pci_dma_read(pci_dev, addr, data, n);
+                DPRINTF("Load reg 0x%x size %d addr 0x%08x = %08x\n", reg, n,
+                        addr, *(int *)data);
+                for (i = 0; i < n; i++) {
+                    lsi_reg_writeb(s, reg + i, data[i]);
+                }
+            } else {
+                DPRINTF("Store reg 0x%x size %d addr 0x%08x\n", reg, n, addr);
+                for (i = 0; i < n; i++) {
+                    data[i] = lsi_reg_readb(s, reg + i);
+                }
+                pci_dma_write(pci_dev, addr, data, n);
+            }
+        }
+    }
+    if (insn_processed > 10000 && !s->waiting) {
+        /* Some windows drivers make the device spin waiting for a memory
+           location to change.  If we have been executed a lot of code then
+           assume this is the case and force an unexpected device disconnect.
+           This is apparently sufficient to beat the drivers into submission.
+         */
+        if (!(s->sien0 & LSI_SIST0_UDC))
+            fprintf(stderr, "inf. loop with UDC masked\n");
+        lsi_script_scsi_interrupt(s, LSI_SIST0_UDC, 0);
+        lsi_disconnect(s);
+    } else if (s->istat1 & LSI_ISTAT1_SRUN && !s->waiting) {
+        if (s->dcntl & LSI_DCNTL_SSM) {
+            lsi_script_dma_interrupt(s, LSI_DSTAT_SSI);
+        } else {
+            goto again;
+        }
+    }
+    DPRINTF("SCRIPTS execution stopped\n");
+}
+
+static uint8_t lsi_reg_readb(LSIState *s, int offset)
+{
+    uint8_t tmp;
+#define CASE_GET_REG24(name, addr) \
+    case addr: return s->name & 0xff; \
+    case addr + 1: return (s->name >> 8) & 0xff; \
+    case addr + 2: return (s->name >> 16) & 0xff;
+
+#define CASE_GET_REG32(name, addr) \
+    case addr: return s->name & 0xff; \
+    case addr + 1: return (s->name >> 8) & 0xff; \
+    case addr + 2: return (s->name >> 16) & 0xff; \
+    case addr + 3: return (s->name >> 24) & 0xff;
+
+#ifdef DEBUG_LSI_REG
+    DPRINTF("Read reg %x\n", offset);
+#endif
+    switch (offset) {
+    case 0x00: /* SCNTL0 */
+        return s->scntl0;
+    case 0x01: /* SCNTL1 */
+        return s->scntl1;
+    case 0x02: /* SCNTL2 */
+        return s->scntl2;
+    case 0x03: /* SCNTL3 */
+        return s->scntl3;
+    case 0x04: /* SCID */
+        return s->scid;
+    case 0x05: /* SXFER */
+        return s->sxfer;
+    case 0x06: /* SDID */
+        return s->sdid;
+    case 0x07: /* GPREG0 */
+        return 0x7f;
+    case 0x08: /* Revision ID */
+        return 0x00;
+    case 0x09: /* SOCL */
+        return s->socl;
+    case 0xa: /* SSID */
+        return s->ssid;
+    case 0xb: /* SBCL */
+        /* ??? This is not correct. However it's (hopefully) only
+           used for diagnostics, so should be ok.  */
+        return 0;
+    case 0xc: /* DSTAT */
+        tmp = s->dstat | LSI_DSTAT_DFE;
+        if ((s->istat0 & LSI_ISTAT0_INTF) == 0)
+            s->dstat = 0;
+        lsi_update_irq(s);
+        return tmp;
+    case 0x0d: /* SSTAT0 */
+        return s->sstat0;
+    case 0x0e: /* SSTAT1 */
+        return s->sstat1;
+    case 0x0f: /* SSTAT2 */
+        return s->scntl1 & LSI_SCNTL1_CON ? 0 : 2;
+    CASE_GET_REG32(dsa, 0x10)
+    case 0x14: /* ISTAT0 */
+        return s->istat0;
+    case 0x15: /* ISTAT1 */
+        return s->istat1;
+    case 0x16: /* MBOX0 */
+        return s->mbox0;
+    case 0x17: /* MBOX1 */
+        return s->mbox1;
+    case 0x18: /* CTEST0 */
+        return 0xff;
+    case 0x19: /* CTEST1 */
+        return 0;
+    case 0x1a: /* CTEST2 */
+        tmp = s->ctest2 | LSI_CTEST2_DACK | LSI_CTEST2_CM;
+        if (s->istat0 & LSI_ISTAT0_SIGP) {
+            s->istat0 &= ~LSI_ISTAT0_SIGP;
+            tmp |= LSI_CTEST2_SIGP;
+        }
+        return tmp;
+    case 0x1b: /* CTEST3 */
+        return s->ctest3;
+    CASE_GET_REG32(temp, 0x1c)
+    case 0x20: /* DFIFO */
+        return 0;
+    case 0x21: /* CTEST4 */
+        return s->ctest4;
+    case 0x22: /* CTEST5 */
+        return s->ctest5;
+    case 0x23: /* CTEST6 */
+         return 0;
+    CASE_GET_REG24(dbc, 0x24)
+    case 0x27: /* DCMD */
+        return s->dcmd;
+    CASE_GET_REG32(dnad, 0x28)
+    CASE_GET_REG32(dsp, 0x2c)
+    CASE_GET_REG32(dsps, 0x30)
+    CASE_GET_REG32(scratch[0], 0x34)
+    case 0x38: /* DMODE */
+        return s->dmode;
+    case 0x39: /* DIEN */
+        return s->dien;
+    case 0x3a: /* SBR */
+        return s->sbr;
+    case 0x3b: /* DCNTL */
+        return s->dcntl;
+    /* ADDER Output (Debug of relative jump address) */
+    CASE_GET_REG32(adder, 0x3c)
+    case 0x40: /* SIEN0 */
+        return s->sien0;
+    case 0x41: /* SIEN1 */
+        return s->sien1;
+    case 0x42: /* SIST0 */
+        tmp = s->sist0;
+        s->sist0 = 0;
+        lsi_update_irq(s);
+        return tmp;
+    case 0x43: /* SIST1 */
+        tmp = s->sist1;
+        s->sist1 = 0;
+        lsi_update_irq(s);
+        return tmp;
+    case 0x46: /* MACNTL */
+        return 0x0f;
+    case 0x47: /* GPCNTL0 */
+        return 0x0f;
+    case 0x48: /* STIME0 */
+        return s->stime0;
+    case 0x4a: /* RESPID0 */
+        return s->respid0;
+    case 0x4b: /* RESPID1 */
+        return s->respid1;
+    case 0x4d: /* STEST1 */
+        return s->stest1;
+    case 0x4e: /* STEST2 */
+        return s->stest2;
+    case 0x4f: /* STEST3 */
+        return s->stest3;
+    case 0x50: /* SIDL */
+        /* This is needed by the linux drivers.  We currently only update it
+           during the MSG IN phase.  */
+        return s->sidl;
+    case 0x52: /* STEST4 */
+        return 0xe0;
+    case 0x56: /* CCNTL0 */
+        return s->ccntl0;
+    case 0x57: /* CCNTL1 */
+        return s->ccntl1;
+    case 0x58: /* SBDL */
+        /* Some drivers peek at the data bus during the MSG IN phase.  */
+        if ((s->sstat1 & PHASE_MASK) == PHASE_MI)
+            return s->msg[0];
+        return 0;
+    case 0x59: /* SBDL high */
+        return 0;
+    CASE_GET_REG32(mmrs, 0xa0)
+    CASE_GET_REG32(mmws, 0xa4)
+    CASE_GET_REG32(sfs, 0xa8)
+    CASE_GET_REG32(drs, 0xac)
+    CASE_GET_REG32(sbms, 0xb0)
+    CASE_GET_REG32(dbms, 0xb4)
+    CASE_GET_REG32(dnad64, 0xb8)
+    CASE_GET_REG32(pmjad1, 0xc0)
+    CASE_GET_REG32(pmjad2, 0xc4)
+    CASE_GET_REG32(rbc, 0xc8)
+    CASE_GET_REG32(ua, 0xcc)
+    CASE_GET_REG32(ia, 0xd4)
+    CASE_GET_REG32(sbc, 0xd8)
+    CASE_GET_REG32(csbc, 0xdc)
+    }
+    if (offset >= 0x5c && offset < 0xa0) {
+        int n;
+        int shift;
+        n = (offset - 0x58) >> 2;
+        shift = (offset & 3) * 8;
+        return (s->scratch[n] >> shift) & 0xff;
+    }
+    BADF("readb 0x%x\n", offset);
+    exit(1);
+#undef CASE_GET_REG24
+#undef CASE_GET_REG32
+}
+
+static void lsi_reg_writeb(LSIState *s, int offset, uint8_t val)
+{
+#define CASE_SET_REG24(name, addr) \
+    case addr    : s->name &= 0xffffff00; s->name |= val;       break; \
+    case addr + 1: s->name &= 0xffff00ff; s->name |= val << 8;  break; \
+    case addr + 2: s->name &= 0xff00ffff; s->name |= val << 16; break;
+
+#define CASE_SET_REG32(name, addr) \
+    case addr    : s->name &= 0xffffff00; s->name |= val;       break; \
+    case addr + 1: s->name &= 0xffff00ff; s->name |= val << 8;  break; \
+    case addr + 2: s->name &= 0xff00ffff; s->name |= val << 16; break; \
+    case addr + 3: s->name &= 0x00ffffff; s->name |= val << 24; break;
+
+#ifdef DEBUG_LSI_REG
+    DPRINTF("Write reg %x = %02x\n", offset, val);
+#endif
+    switch (offset) {
+    case 0x00: /* SCNTL0 */
+        s->scntl0 = val;
+        if (val & LSI_SCNTL0_START) {
+            BADF("Start sequence not implemented\n");
+        }
+        break;
+    case 0x01: /* SCNTL1 */
+        s->scntl1 = val & ~LSI_SCNTL1_SST;
+        if (val & LSI_SCNTL1_IARB) {
+            BADF("Immediate Arbritration not implemented\n");
+        }
+        if (val & LSI_SCNTL1_RST) {
+            if (!(s->sstat0 & LSI_SSTAT0_RST)) {
+                qbus_reset_all(&s->bus.qbus);
+                s->sstat0 |= LSI_SSTAT0_RST;
+                lsi_script_scsi_interrupt(s, LSI_SIST0_RST, 0);
+            }
+        } else {
+            s->sstat0 &= ~LSI_SSTAT0_RST;
+        }
+        break;
+    case 0x02: /* SCNTL2 */
+        val &= ~(LSI_SCNTL2_WSR | LSI_SCNTL2_WSS);
+        s->scntl2 = val;
+        break;
+    case 0x03: /* SCNTL3 */
+        s->scntl3 = val;
+        break;
+    case 0x04: /* SCID */
+        s->scid = val;
+        break;
+    case 0x05: /* SXFER */
+        s->sxfer = val;
+        break;
+    case 0x06: /* SDID */
+        if ((s->ssid & 0x80) && (val & 0xf) != (s->ssid & 0xf)) {
+            BADF("Destination ID does not match SSID\n");
+        }
+        s->sdid = val & 0xf;
+        break;
+    case 0x07: /* GPREG0 */
+        break;
+    case 0x08: /* SFBR */
+        /* The CPU is not allowed to write to this register.  However the
+           SCRIPTS register move instructions are.  */
+        s->sfbr = val;
+        break;
+    case 0x0a: case 0x0b:
+        /* Openserver writes to these readonly registers on startup */
+	return;
+    case 0x0c: case 0x0d: case 0x0e: case 0x0f:
+        /* Linux writes to these readonly registers on startup.  */
+        return;
+    CASE_SET_REG32(dsa, 0x10)
+    case 0x14: /* ISTAT0 */
+        s->istat0 = (s->istat0 & 0x0f) | (val & 0xf0);
+        if (val & LSI_ISTAT0_ABRT) {
+            lsi_script_dma_interrupt(s, LSI_DSTAT_ABRT);
+        }
+        if (val & LSI_ISTAT0_INTF) {
+            s->istat0 &= ~LSI_ISTAT0_INTF;
+            lsi_update_irq(s);
+        }
+        if (s->waiting == 1 && val & LSI_ISTAT0_SIGP) {
+            DPRINTF("Woken by SIGP\n");
+            s->waiting = 0;
+            s->dsp = s->dnad;
+            lsi_execute_script(s);
+        }
+        if (val & LSI_ISTAT0_SRST) {
+            qdev_reset_all(DEVICE(s));
+        }
+        break;
+    case 0x16: /* MBOX0 */
+        s->mbox0 = val;
+        break;
+    case 0x17: /* MBOX1 */
+        s->mbox1 = val;
+        break;
+    case 0x18: /* CTEST0 */
+        /* nothing to do */
+        break;
+    case 0x1a: /* CTEST2 */
+	s->ctest2 = val & LSI_CTEST2_PCICIE;
+	break;
+    case 0x1b: /* CTEST3 */
+        s->ctest3 = val & 0x0f;
+        break;
+    CASE_SET_REG32(temp, 0x1c)
+    case 0x21: /* CTEST4 */
+        if (val & 7) {
+           BADF("Unimplemented CTEST4-FBL 0x%x\n", val);
+        }
+        s->ctest4 = val;
+        break;
+    case 0x22: /* CTEST5 */
+        if (val & (LSI_CTEST5_ADCK | LSI_CTEST5_BBCK)) {
+            BADF("CTEST5 DMA increment not implemented\n");
+        }
+        s->ctest5 = val;
+        break;
+    CASE_SET_REG24(dbc, 0x24)
+    CASE_SET_REG32(dnad, 0x28)
+    case 0x2c: /* DSP[0:7] */
+        s->dsp &= 0xffffff00;
+        s->dsp |= val;
+        break;
+    case 0x2d: /* DSP[8:15] */
+        s->dsp &= 0xffff00ff;
+        s->dsp |= val << 8;
+        break;
+    case 0x2e: /* DSP[16:23] */
+        s->dsp &= 0xff00ffff;
+        s->dsp |= val << 16;
+        break;
+    case 0x2f: /* DSP[24:31] */
+        s->dsp &= 0x00ffffff;
+        s->dsp |= val << 24;
+        if ((s->dmode & LSI_DMODE_MAN) == 0
+            && (s->istat1 & LSI_ISTAT1_SRUN) == 0)
+            lsi_execute_script(s);
+        break;
+    CASE_SET_REG32(dsps, 0x30)
+    CASE_SET_REG32(scratch[0], 0x34)
+    case 0x38: /* DMODE */
+        if (val & (LSI_DMODE_SIOM | LSI_DMODE_DIOM)) {
+            BADF("IO mappings not implemented\n");
+        }
+        s->dmode = val;
+        break;
+    case 0x39: /* DIEN */
+        s->dien = val;
+        lsi_update_irq(s);
+        break;
+    case 0x3a: /* SBR */
+        s->sbr = val;
+        break;
+    case 0x3b: /* DCNTL */
+        s->dcntl = val & ~(LSI_DCNTL_PFF | LSI_DCNTL_STD);
+        if ((val & LSI_DCNTL_STD) && (s->istat1 & LSI_ISTAT1_SRUN) == 0)
+            lsi_execute_script(s);
+        break;
+    case 0x40: /* SIEN0 */
+        s->sien0 = val;
+        lsi_update_irq(s);
+        break;
+    case 0x41: /* SIEN1 */
+        s->sien1 = val;
+        lsi_update_irq(s);
+        break;
+    case 0x47: /* GPCNTL0 */
+        break;
+    case 0x48: /* STIME0 */
+        s->stime0 = val;
+        break;
+    case 0x49: /* STIME1 */
+        if (val & 0xf) {
+            DPRINTF("General purpose timer not implemented\n");
+            /* ??? Raising the interrupt immediately seems to be sufficient
+               to keep the FreeBSD driver happy.  */
+            lsi_script_scsi_interrupt(s, 0, LSI_SIST1_GEN);
+        }
+        break;
+    case 0x4a: /* RESPID0 */
+        s->respid0 = val;
+        break;
+    case 0x4b: /* RESPID1 */
+        s->respid1 = val;
+        break;
+    case 0x4d: /* STEST1 */
+        s->stest1 = val;
+        break;
+    case 0x4e: /* STEST2 */
+        if (val & 1) {
+            BADF("Low level mode not implemented\n");
+        }
+        s->stest2 = val;
+        break;
+    case 0x4f: /* STEST3 */
+        if (val & 0x41) {
+            BADF("SCSI FIFO test mode not implemented\n");
+        }
+        s->stest3 = val;
+        break;
+    case 0x56: /* CCNTL0 */
+        s->ccntl0 = val;
+        break;
+    case 0x57: /* CCNTL1 */
+        s->ccntl1 = val;
+        break;
+    CASE_SET_REG32(mmrs, 0xa0)
+    CASE_SET_REG32(mmws, 0xa4)
+    CASE_SET_REG32(sfs, 0xa8)
+    CASE_SET_REG32(drs, 0xac)
+    CASE_SET_REG32(sbms, 0xb0)
+    CASE_SET_REG32(dbms, 0xb4)
+    CASE_SET_REG32(dnad64, 0xb8)
+    CASE_SET_REG32(pmjad1, 0xc0)
+    CASE_SET_REG32(pmjad2, 0xc4)
+    CASE_SET_REG32(rbc, 0xc8)
+    CASE_SET_REG32(ua, 0xcc)
+    CASE_SET_REG32(ia, 0xd4)
+    CASE_SET_REG32(sbc, 0xd8)
+    CASE_SET_REG32(csbc, 0xdc)
+    default:
+        if (offset >= 0x5c && offset < 0xa0) {
+            int n;
+            int shift;
+            n = (offset - 0x58) >> 2;
+            shift = (offset & 3) * 8;
+            s->scratch[n] = deposit32(s->scratch[n], shift, 8, val);
+        } else {
+            BADF("Unhandled writeb 0x%x = 0x%x\n", offset, val);
+        }
+    }
+#undef CASE_SET_REG24
+#undef CASE_SET_REG32
+}
+
+static void lsi_mmio_write(void *opaque, hwaddr addr,
+                           uint64_t val, unsigned size)
+{
+    LSIState *s = opaque;
+
+    lsi_reg_writeb(s, addr & 0xff, val);
+}
+
+static uint64_t lsi_mmio_read(void *opaque, hwaddr addr,
+                              unsigned size)
+{
+    LSIState *s = opaque;
+
+    return lsi_reg_readb(s, addr & 0xff);
+}
+
+static const MemoryRegionOps lsi_mmio_ops = {
+    .read = lsi_mmio_read,
+    .write = lsi_mmio_write,
+    .endianness = DEVICE_NATIVE_ENDIAN,
+    .impl = {
+        .min_access_size = 1,
+        .max_access_size = 1,
+    },
+};
+
+static void lsi_ram_write(void *opaque, hwaddr addr,
+                          uint64_t val, unsigned size)
+{
+    LSIState *s = opaque;
+    uint32_t newval;
+    uint32_t mask;
+    int shift;
+
+    newval = s->script_ram[addr >> 2];
+    shift = (addr & 3) * 8;
+    mask = ((uint64_t)1 << (size * 8)) - 1;
+    newval &= ~(mask << shift);
+    newval |= val << shift;
+    s->script_ram[addr >> 2] = newval;
+}
+
+static uint64_t lsi_ram_read(void *opaque, hwaddr addr,
+                             unsigned size)
+{
+    LSIState *s = opaque;
+    uint32_t val;
+    uint32_t mask;
+
+    val = s->script_ram[addr >> 2];
+    mask = ((uint64_t)1 << (size * 8)) - 1;
+    val >>= (addr & 3) * 8;
+    return val & mask;
+}
+
+static const MemoryRegionOps lsi_ram_ops = {
+    .read = lsi_ram_read,
+    .write = lsi_ram_write,
+    .endianness = DEVICE_NATIVE_ENDIAN,
+};
+
+static uint64_t lsi_io_read(void *opaque, hwaddr addr,
+                            unsigned size)
+{
+    LSIState *s = opaque;
+    return lsi_reg_readb(s, addr & 0xff);
+}
+
+static void lsi_io_write(void *opaque, hwaddr addr,
+                         uint64_t val, unsigned size)
+{
+    LSIState *s = opaque;
+    lsi_reg_writeb(s, addr & 0xff, val);
+}
+
+static const MemoryRegionOps lsi_io_ops = {
+    .read = lsi_io_read,
+    .write = lsi_io_write,
+    .endianness = DEVICE_NATIVE_ENDIAN,
+    .impl = {
+        .min_access_size = 1,
+        .max_access_size = 1,
+    },
+};
+
+static void lsi_scsi_reset(DeviceState *dev)
+{
+    LSIState *s = LSI53C895A(dev);
+
+    lsi_soft_reset(s);
+}
+
+static void lsi_pre_save(void *opaque)
+{
+    LSIState *s = opaque;
+
+    if (s->current) {
+        assert(s->current->dma_buf == NULL);
+        assert(s->current->dma_len == 0);
+    }
+    assert(QTAILQ_EMPTY(&s->queue));
+}
+
+static const VMStateDescription vmstate_lsi_scsi = {
+    .name = "lsiscsi",
+    .version_id = 0,
+    .minimum_version_id = 0,
+    .pre_save = lsi_pre_save,
+    .fields = (VMStateField[]) {
+        VMSTATE_PCI_DEVICE(parent_obj, LSIState),
+
+        VMSTATE_INT32(carry, LSIState),
+        VMSTATE_INT32(status, LSIState),
+        VMSTATE_INT32(msg_action, LSIState),
+        VMSTATE_INT32(msg_len, LSIState),
+        VMSTATE_BUFFER(msg, LSIState),
+        VMSTATE_INT32(waiting, LSIState),
+
+        VMSTATE_UINT32(dsa, LSIState),
+        VMSTATE_UINT32(temp, LSIState),
+        VMSTATE_UINT32(dnad, LSIState),
+        VMSTATE_UINT32(dbc, LSIState),
+        VMSTATE_UINT8(istat0, LSIState),
+        VMSTATE_UINT8(istat1, LSIState),
+        VMSTATE_UINT8(dcmd, LSIState),
+        VMSTATE_UINT8(dstat, LSIState),
+        VMSTATE_UINT8(dien, LSIState),
+        VMSTATE_UINT8(sist0, LSIState),
+        VMSTATE_UINT8(sist1, LSIState),
+        VMSTATE_UINT8(sien0, LSIState),
+        VMSTATE_UINT8(sien1, LSIState),
+        VMSTATE_UINT8(mbox0, LSIState),
+        VMSTATE_UINT8(mbox1, LSIState),
+        VMSTATE_UINT8(dfifo, LSIState),
+        VMSTATE_UINT8(ctest2, LSIState),
+        VMSTATE_UINT8(ctest3, LSIState),
+        VMSTATE_UINT8(ctest4, LSIState),
+        VMSTATE_UINT8(ctest5, LSIState),
+        VMSTATE_UINT8(ccntl0, LSIState),
+        VMSTATE_UINT8(ccntl1, LSIState),
+        VMSTATE_UINT32(dsp, LSIState),
+        VMSTATE_UINT32(dsps, LSIState),
+        VMSTATE_UINT8(dmode, LSIState),
+        VMSTATE_UINT8(dcntl, LSIState),
+        VMSTATE_UINT8(scntl0, LSIState),
+        VMSTATE_UINT8(scntl1, LSIState),
+        VMSTATE_UINT8(scntl2, LSIState),
+        VMSTATE_UINT8(scntl3, LSIState),
+        VMSTATE_UINT8(sstat0, LSIState),
+        VMSTATE_UINT8(sstat1, LSIState),
+        VMSTATE_UINT8(scid, LSIState),
+        VMSTATE_UINT8(sxfer, LSIState),
+        VMSTATE_UINT8(socl, LSIState),
+        VMSTATE_UINT8(sdid, LSIState),
+        VMSTATE_UINT8(ssid, LSIState),
+        VMSTATE_UINT8(sfbr, LSIState),
+        VMSTATE_UINT8(stest1, LSIState),
+        VMSTATE_UINT8(stest2, LSIState),
+        VMSTATE_UINT8(stest3, LSIState),
+        VMSTATE_UINT8(sidl, LSIState),
+        VMSTATE_UINT8(stime0, LSIState),
+        VMSTATE_UINT8(respid0, LSIState),
+        VMSTATE_UINT8(respid1, LSIState),
+        VMSTATE_UINT32(mmrs, LSIState),
+        VMSTATE_UINT32(mmws, LSIState),
+        VMSTATE_UINT32(sfs, LSIState),
+        VMSTATE_UINT32(drs, LSIState),
+        VMSTATE_UINT32(sbms, LSIState),
+        VMSTATE_UINT32(dbms, LSIState),
+        VMSTATE_UINT32(dnad64, LSIState),
+        VMSTATE_UINT32(pmjad1, LSIState),
+        VMSTATE_UINT32(pmjad2, LSIState),
+        VMSTATE_UINT32(rbc, LSIState),
+        VMSTATE_UINT32(ua, LSIState),
+        VMSTATE_UINT32(ia, LSIState),
+        VMSTATE_UINT32(sbc, LSIState),
+        VMSTATE_UINT32(csbc, LSIState),
+        VMSTATE_BUFFER_UNSAFE(scratch, LSIState, 0, 18 * sizeof(uint32_t)),
+        VMSTATE_UINT8(sbr, LSIState),
+
+        VMSTATE_BUFFER_UNSAFE(script_ram, LSIState, 0, 2048 * sizeof(uint32_t)),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static const struct SCSIBusInfo lsi_scsi_info = {
+    .tcq = true,
+    .max_target = LSI_MAX_DEVS,
+    .max_lun = 0,  /* LUN support is buggy */
+
+    .transfer_data = lsi_transfer_data,
+    .complete = lsi_command_complete,
+    .cancel = lsi_request_cancelled
+};
+
+static void lsi_scsi_realize(PCIDevice *dev, Error **errp)
+{
+    LSIState *s = LSI53C895A(dev);
+    DeviceState *d = DEVICE(dev);
+    uint8_t *pci_conf;
+
+    pci_conf = dev->config;
+
+    /* PCI latency timer = 255 */
+    pci_conf[PCI_LATENCY_TIMER] = 0xff;
+    /* Interrupt pin A */
+    pci_conf[PCI_INTERRUPT_PIN] = 0x01;
+
+    memory_region_init_io(&s->mmio_io, OBJECT(s), &lsi_mmio_ops, s,
+                          "lsi-mmio", 0x400);
+    memory_region_init_io(&s->ram_io, OBJECT(s), &lsi_ram_ops, s,
+                          "lsi-ram", 0x2000);
+    memory_region_init_io(&s->io_io, OBJECT(s), &lsi_io_ops, s,
+                          "lsi-io", 256);
+
+    pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_IO, &s->io_io);
+    pci_register_bar(dev, 1, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->mmio_io);
+    pci_register_bar(dev, 2, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->ram_io);
+    QTAILQ_INIT(&s->queue);
+
+    scsi_bus_new(&s->bus, sizeof(s->bus), d, &lsi_scsi_info, NULL);
+    if (!d->hotplugged) {
+        scsi_bus_legacy_handle_cmdline(&s->bus, errp);
+    }
+}
+
+static void lsi_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+
+    k->realize = lsi_scsi_realize;
+    k->vendor_id = PCI_VENDOR_ID_LSI_LOGIC;
+    k->device_id = PCI_DEVICE_ID_LSI_53C895A;
+    k->class_id = PCI_CLASS_STORAGE_SCSI;
+    k->subsystem_id = 0x1000;
+    dc->reset = lsi_scsi_reset;
+    dc->vmsd = &vmstate_lsi_scsi;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+}
+
+static const TypeInfo lsi_info = {
+    .name          = TYPE_LSI53C895A,
+    .parent        = TYPE_PCI_DEVICE,
+    .instance_size = sizeof(LSIState),
+    .class_init    = lsi_class_init,
+};
+
+static void lsi53c810_class_init(ObjectClass *klass, void *data)
+{
+    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+
+    k->device_id = PCI_DEVICE_ID_LSI_53C810;
+}
+
+static TypeInfo lsi53c810_info = {
+    .name          = TYPE_LSI53C810,
+    .parent        = TYPE_LSI53C895A,
+    .class_init    = lsi53c810_class_init,
+};
+
+static void lsi53c895a_register_types(void)
+{
+    type_register_static(&lsi_info);
+    type_register_static(&lsi53c810_info);
+}
+
+type_init(lsi53c895a_register_types)
diff --git a/qemu/hw/scsi/megasas.c b/qemu/hw/scsi/megasas.c
new file mode 100644
index 000000000..a04369c5a
--- /dev/null
+++ b/qemu/hw/scsi/megasas.c
@@ -0,0 +1,2547 @@
+/*
+ * QEMU MegaRAID SAS 8708EM2 Host Bus Adapter emulation
+ * Based on the linux driver code at drivers/scsi/megaraid
+ *
+ * Copyright (c) 2009-2012 Hannes Reinecke, SUSE Labs
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library 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.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "hw/hw.h"
+#include "hw/pci/pci.h"
+#include "sysemu/dma.h"
+#include "sysemu/block-backend.h"
+#include "hw/pci/msi.h"
+#include "hw/pci/msix.h"
+#include "qemu/iov.h"
+#include "hw/scsi/scsi.h"
+#include "block/scsi.h"
+#include "trace.h"
+
+#include "mfi.h"
+
+#define MEGASAS_VERSION_GEN1 "1.70"
+#define MEGASAS_VERSION_GEN2 "1.80"
+#define MEGASAS_MAX_FRAMES 2048         /* Firmware limit at 65535 */
+#define MEGASAS_DEFAULT_FRAMES 1000     /* Windows requires this */
+#define MEGASAS_GEN2_DEFAULT_FRAMES 1008     /* Windows requires this */
+#define MEGASAS_MAX_SGE 128             /* Firmware limit */
+#define MEGASAS_DEFAULT_SGE 80
+#define MEGASAS_MAX_SECTORS 0xFFFF      /* No real limit */
+#define MEGASAS_MAX_ARRAYS 128
+
+#define MEGASAS_HBA_SERIAL "QEMU123456"
+#define NAA_LOCALLY_ASSIGNED_ID 0x3ULL
+#define IEEE_COMPANY_LOCALLY_ASSIGNED 0x525400
+
+#define MEGASAS_FLAG_USE_JBOD      0
+#define MEGASAS_MASK_USE_JBOD      (1 << MEGASAS_FLAG_USE_JBOD)
+#define MEGASAS_FLAG_USE_MSI       1
+#define MEGASAS_MASK_USE_MSI       (1 << MEGASAS_FLAG_USE_MSI)
+#define MEGASAS_FLAG_USE_MSIX      2
+#define MEGASAS_MASK_USE_MSIX      (1 << MEGASAS_FLAG_USE_MSIX)
+#define MEGASAS_FLAG_USE_QUEUE64   3
+#define MEGASAS_MASK_USE_QUEUE64   (1 << MEGASAS_FLAG_USE_QUEUE64)
+
+static const char *mfi_frame_desc[] = {
+    "MFI init", "LD Read", "LD Write", "LD SCSI", "PD SCSI",
+    "MFI Doorbell", "MFI Abort", "MFI SMP", "MFI Stop"};
+
+typedef struct MegasasCmd {
+    uint32_t index;
+    uint16_t flags;
+    uint16_t count;
+    uint64_t context;
+
+    hwaddr pa;
+    hwaddr pa_size;
+    union mfi_frame *frame;
+    SCSIRequest *req;
+    QEMUSGList qsg;
+    void *iov_buf;
+    size_t iov_size;
+    size_t iov_offset;
+    struct MegasasState *state;
+} MegasasCmd;
+
+typedef struct MegasasState {
+    /*< private >*/
+    PCIDevice parent_obj;
+    /*< public >*/
+
+    MemoryRegion mmio_io;
+    MemoryRegion port_io;
+    MemoryRegion queue_io;
+    uint32_t frame_hi;
+
+    int fw_state;
+    uint32_t fw_sge;
+    uint32_t fw_cmds;
+    uint32_t flags;
+    int fw_luns;
+    int intr_mask;
+    int doorbell;
+    int busy;
+    int diag;
+    int adp_reset;
+
+    MegasasCmd *event_cmd;
+    int event_locale;
+    int event_class;
+    int event_count;
+    int shutdown_event;
+    int boot_event;
+
+    uint64_t sas_addr;
+    char *hba_serial;
+
+    uint64_t reply_queue_pa;
+    void *reply_queue;
+    int reply_queue_len;
+    int reply_queue_head;
+    int reply_queue_tail;
+    uint64_t consumer_pa;
+    uint64_t producer_pa;
+
+    MegasasCmd frames[MEGASAS_MAX_FRAMES];
+    DECLARE_BITMAP(frame_map, MEGASAS_MAX_FRAMES);
+    SCSIBus bus;
+} MegasasState;
+
+typedef struct MegasasBaseClass {
+    PCIDeviceClass parent_class;
+    const char *product_name;
+    const char *product_version;
+    int mmio_bar;
+    int ioport_bar;
+    int osts;
+} MegasasBaseClass;
+
+#define TYPE_MEGASAS_BASE "megasas-base"
+#define TYPE_MEGASAS_GEN1 "megasas"
+#define TYPE_MEGASAS_GEN2 "megasas-gen2"
+
+#define MEGASAS(obj) \
+    OBJECT_CHECK(MegasasState, (obj), TYPE_MEGASAS_BASE)
+
+#define MEGASAS_DEVICE_CLASS(oc) \
+    OBJECT_CLASS_CHECK(MegasasBaseClass, (oc), TYPE_MEGASAS_BASE)
+#define MEGASAS_DEVICE_GET_CLASS(oc) \
+    OBJECT_GET_CLASS(MegasasBaseClass, (oc), TYPE_MEGASAS_BASE)
+
+#define MEGASAS_INTR_DISABLED_MASK 0xFFFFFFFF
+
+static bool megasas_intr_enabled(MegasasState *s)
+{
+    if ((s->intr_mask & MEGASAS_INTR_DISABLED_MASK) !=
+        MEGASAS_INTR_DISABLED_MASK) {
+        return true;
+    }
+    return false;
+}
+
+static bool megasas_use_queue64(MegasasState *s)
+{
+    return s->flags & MEGASAS_MASK_USE_QUEUE64;
+}
+
+static bool megasas_use_msi(MegasasState *s)
+{
+    return s->flags & MEGASAS_MASK_USE_MSI;
+}
+
+static bool megasas_use_msix(MegasasState *s)
+{
+    return s->flags & MEGASAS_MASK_USE_MSIX;
+}
+
+static bool megasas_is_jbod(MegasasState *s)
+{
+    return s->flags & MEGASAS_MASK_USE_JBOD;
+}
+
+static void megasas_frame_set_cmd_status(MegasasState *s,
+                                         unsigned long frame, uint8_t v)
+{
+    PCIDevice *pci = &s->parent_obj;
+    stb_pci_dma(pci, frame + offsetof(struct mfi_frame_header, cmd_status), v);
+}
+
+static void megasas_frame_set_scsi_status(MegasasState *s,
+                                          unsigned long frame, uint8_t v)
+{
+    PCIDevice *pci = &s->parent_obj;
+    stb_pci_dma(pci, frame + offsetof(struct mfi_frame_header, scsi_status), v);
+}
+
+/*
+ * Context is considered opaque, but the HBA firmware is running
+ * in little endian mode. So convert it to little endian, too.
+ */
+static uint64_t megasas_frame_get_context(MegasasState *s,
+                                          unsigned long frame)
+{
+    PCIDevice *pci = &s->parent_obj;
+    return ldq_le_pci_dma(pci, frame + offsetof(struct mfi_frame_header, context));
+}
+
+static bool megasas_frame_is_ieee_sgl(MegasasCmd *cmd)
+{
+    return cmd->flags & MFI_FRAME_IEEE_SGL;
+}
+
+static bool megasas_frame_is_sgl64(MegasasCmd *cmd)
+{
+    return cmd->flags & MFI_FRAME_SGL64;
+}
+
+static bool megasas_frame_is_sense64(MegasasCmd *cmd)
+{
+    return cmd->flags & MFI_FRAME_SENSE64;
+}
+
+static uint64_t megasas_sgl_get_addr(MegasasCmd *cmd,
+                                     union mfi_sgl *sgl)
+{
+    uint64_t addr;
+
+    if (megasas_frame_is_ieee_sgl(cmd)) {
+        addr = le64_to_cpu(sgl->sg_skinny->addr);
+    } else if (megasas_frame_is_sgl64(cmd)) {
+        addr = le64_to_cpu(sgl->sg64->addr);
+    } else {
+        addr = le32_to_cpu(sgl->sg32->addr);
+    }
+    return addr;
+}
+
+static uint32_t megasas_sgl_get_len(MegasasCmd *cmd,
+                                    union mfi_sgl *sgl)
+{
+    uint32_t len;
+
+    if (megasas_frame_is_ieee_sgl(cmd)) {
+        len = le32_to_cpu(sgl->sg_skinny->len);
+    } else if (megasas_frame_is_sgl64(cmd)) {
+        len = le32_to_cpu(sgl->sg64->len);
+    } else {
+        len = le32_to_cpu(sgl->sg32->len);
+    }
+    return len;
+}
+
+static union mfi_sgl *megasas_sgl_next(MegasasCmd *cmd,
+                                       union mfi_sgl *sgl)
+{
+    uint8_t *next = (uint8_t *)sgl;
+
+    if (megasas_frame_is_ieee_sgl(cmd)) {
+        next += sizeof(struct mfi_sg_skinny);
+    } else if (megasas_frame_is_sgl64(cmd)) {
+        next += sizeof(struct mfi_sg64);
+    } else {
+        next += sizeof(struct mfi_sg32);
+    }
+
+    if (next >= (uint8_t *)cmd->frame + cmd->pa_size) {
+        return NULL;
+    }
+    return (union mfi_sgl *)next;
+}
+
+static void megasas_soft_reset(MegasasState *s);
+
+static int megasas_map_sgl(MegasasState *s, MegasasCmd *cmd, union mfi_sgl *sgl)
+{
+    int i;
+    int iov_count = 0;
+    size_t iov_size = 0;
+
+    cmd->flags = le16_to_cpu(cmd->frame->header.flags);
+    iov_count = cmd->frame->header.sge_count;
+    if (iov_count > MEGASAS_MAX_SGE) {
+        trace_megasas_iovec_sgl_overflow(cmd->index, iov_count,
+                                         MEGASAS_MAX_SGE);
+        return iov_count;
+    }
+    pci_dma_sglist_init(&cmd->qsg, PCI_DEVICE(s), iov_count);
+    for (i = 0; i < iov_count; i++) {
+        dma_addr_t iov_pa, iov_size_p;
+
+        if (!sgl) {
+            trace_megasas_iovec_sgl_underflow(cmd->index, i);
+            goto unmap;
+        }
+        iov_pa = megasas_sgl_get_addr(cmd, sgl);
+        iov_size_p = megasas_sgl_get_len(cmd, sgl);
+        if (!iov_pa || !iov_size_p) {
+            trace_megasas_iovec_sgl_invalid(cmd->index, i,
+                                            iov_pa, iov_size_p);
+            goto unmap;
+        }
+        qemu_sglist_add(&cmd->qsg, iov_pa, iov_size_p);
+        sgl = megasas_sgl_next(cmd, sgl);
+        iov_size += (size_t)iov_size_p;
+    }
+    if (cmd->iov_size > iov_size) {
+        trace_megasas_iovec_overflow(cmd->index, iov_size, cmd->iov_size);
+    } else if (cmd->iov_size < iov_size) {
+        trace_megasas_iovec_underflow(cmd->iov_size, iov_size, cmd->iov_size);
+    }
+    cmd->iov_offset = 0;
+    return 0;
+unmap:
+    qemu_sglist_destroy(&cmd->qsg);
+    return iov_count - i;
+}
+
+static void megasas_unmap_sgl(MegasasCmd *cmd)
+{
+    qemu_sglist_destroy(&cmd->qsg);
+    cmd->iov_offset = 0;
+}
+
+/*
+ * passthrough sense and io sense are at the same offset
+ */
+static int megasas_build_sense(MegasasCmd *cmd, uint8_t *sense_ptr,
+    uint8_t sense_len)
+{
+    PCIDevice *pcid = PCI_DEVICE(cmd->state);
+    uint32_t pa_hi = 0, pa_lo;
+    hwaddr pa;
+
+    if (sense_len > cmd->frame->header.sense_len) {
+        sense_len = cmd->frame->header.sense_len;
+    }
+    if (sense_len) {
+        pa_lo = le32_to_cpu(cmd->frame->pass.sense_addr_lo);
+        if (megasas_frame_is_sense64(cmd)) {
+            pa_hi = le32_to_cpu(cmd->frame->pass.sense_addr_hi);
+        }
+        pa = ((uint64_t) pa_hi << 32) | pa_lo;
+        pci_dma_write(pcid, pa, sense_ptr, sense_len);
+        cmd->frame->header.sense_len = sense_len;
+    }
+    return sense_len;
+}
+
+static void megasas_write_sense(MegasasCmd *cmd, SCSISense sense)
+{
+    uint8_t sense_buf[SCSI_SENSE_BUF_SIZE];
+    uint8_t sense_len = 18;
+
+    memset(sense_buf, 0, sense_len);
+    sense_buf[0] = 0xf0;
+    sense_buf[2] = sense.key;
+    sense_buf[7] = 10;
+    sense_buf[12] = sense.asc;
+    sense_buf[13] = sense.ascq;
+    megasas_build_sense(cmd, sense_buf, sense_len);
+}
+
+static void megasas_copy_sense(MegasasCmd *cmd)
+{
+    uint8_t sense_buf[SCSI_SENSE_BUF_SIZE];
+    uint8_t sense_len;
+
+    sense_len = scsi_req_get_sense(cmd->req, sense_buf,
+                                   SCSI_SENSE_BUF_SIZE);
+    megasas_build_sense(cmd, sense_buf, sense_len);
+}
+
+/*
+ * Format an INQUIRY CDB
+ */
+static int megasas_setup_inquiry(uint8_t *cdb, int pg, int len)
+{
+    memset(cdb, 0, 6);
+    cdb[0] = INQUIRY;
+    if (pg > 0) {
+        cdb[1] = 0x1;
+        cdb[2] = pg;
+    }
+    cdb[3] = (len >> 8) & 0xff;
+    cdb[4] = (len & 0xff);
+    return len;
+}
+
+/*
+ * Encode lba and len into a READ_16/WRITE_16 CDB
+ */
+static void megasas_encode_lba(uint8_t *cdb, uint64_t lba,
+                               uint32_t len, bool is_write)
+{
+    memset(cdb, 0x0, 16);
+    if (is_write) {
+        cdb[0] = WRITE_16;
+    } else {
+        cdb[0] = READ_16;
+    }
+    cdb[2] = (lba >> 56) & 0xff;
+    cdb[3] = (lba >> 48) & 0xff;
+    cdb[4] = (lba >> 40) & 0xff;
+    cdb[5] = (lba >> 32) & 0xff;
+    cdb[6] = (lba >> 24) & 0xff;
+    cdb[7] = (lba >> 16) & 0xff;
+    cdb[8] = (lba >> 8) & 0xff;
+    cdb[9] = (lba) & 0xff;
+    cdb[10] = (len >> 24) & 0xff;
+    cdb[11] = (len >> 16) & 0xff;
+    cdb[12] = (len >> 8) & 0xff;
+    cdb[13] = (len) & 0xff;
+}
+
+/*
+ * Utility functions
+ */
+static uint64_t megasas_fw_time(void)
+{
+    struct tm curtime;
+    uint64_t bcd_time;
+
+    qemu_get_timedate(&curtime, 0);
+    bcd_time = ((uint64_t)curtime.tm_sec & 0xff) << 48 |
+        ((uint64_t)curtime.tm_min & 0xff)  << 40 |
+        ((uint64_t)curtime.tm_hour & 0xff) << 32 |
+        ((uint64_t)curtime.tm_mday & 0xff) << 24 |
+        ((uint64_t)curtime.tm_mon & 0xff)  << 16 |
+        ((uint64_t)(curtime.tm_year + 1900) & 0xffff);
+
+    return bcd_time;
+}
+
+/*
+ * Default disk sata address
+ * 0x1221 is the magic number as
+ * present in real hardware,
+ * so use it here, too.
+ */
+static uint64_t megasas_get_sata_addr(uint16_t id)
+{
+    uint64_t addr = (0x1221ULL << 48);
+    return addr & (id << 24);
+}
+
+/*
+ * Frame handling
+ */
+static int megasas_next_index(MegasasState *s, int index, int limit)
+{
+    index++;
+    if (index == limit) {
+        index = 0;
+    }
+    return index;
+}
+
+static MegasasCmd *megasas_lookup_frame(MegasasState *s,
+    hwaddr frame)
+{
+    MegasasCmd *cmd = NULL;
+    int num = 0, index;
+
+    index = s->reply_queue_head;
+
+    while (num < s->fw_cmds) {
+        if (s->frames[index].pa && s->frames[index].pa == frame) {
+            cmd = &s->frames[index];
+            break;
+        }
+        index = megasas_next_index(s, index, s->fw_cmds);
+        num++;
+    }
+
+    return cmd;
+}
+
+static void megasas_unmap_frame(MegasasState *s, MegasasCmd *cmd)
+{
+    PCIDevice *p = PCI_DEVICE(s);
+
+    pci_dma_unmap(p, cmd->frame, cmd->pa_size, 0, 0);
+    cmd->frame = NULL;
+    cmd->pa = 0;
+    clear_bit(cmd->index, s->frame_map);
+}
+
+/*
+ * This absolutely needs to be locked if
+ * qemu ever goes multithreaded.
+ */
+static MegasasCmd *megasas_enqueue_frame(MegasasState *s,
+    hwaddr frame, uint64_t context, int count)
+{
+    PCIDevice *pcid = PCI_DEVICE(s);
+    MegasasCmd *cmd = NULL;
+    int frame_size = MFI_FRAME_SIZE * 16;
+    hwaddr frame_size_p = frame_size;
+    unsigned long index;
+
+    index = 0;
+    while (index < s->fw_cmds) {
+        index = find_next_zero_bit(s->frame_map, s->fw_cmds, index);
+        if (!s->frames[index].pa)
+            break;
+        /* Busy frame found */
+        trace_megasas_qf_mapped(index);
+    }
+    if (index >= s->fw_cmds) {
+        /* All frames busy */
+        trace_megasas_qf_busy(frame);
+        return NULL;
+    }
+    cmd = &s->frames[index];
+    set_bit(index, s->frame_map);
+    trace_megasas_qf_new(index, frame);
+
+    cmd->pa = frame;
+    /* Map all possible frames */
+    cmd->frame = pci_dma_map(pcid, frame, &frame_size_p, 0);
+    if (frame_size_p != frame_size) {
+        trace_megasas_qf_map_failed(cmd->index, (unsigned long)frame);
+        if (cmd->frame) {
+            megasas_unmap_frame(s, cmd);
+        }
+        s->event_count++;
+        return NULL;
+    }
+    cmd->pa_size = frame_size_p;
+    cmd->context = context;
+    if (!megasas_use_queue64(s)) {
+        cmd->context &= (uint64_t)0xFFFFFFFF;
+    }
+    cmd->count = count;
+    s->busy++;
+
+    if (s->consumer_pa) {
+        s->reply_queue_tail = ldl_le_pci_dma(pcid, s->consumer_pa);
+    }
+    trace_megasas_qf_enqueue(cmd->index, cmd->count, cmd->context,
+                             s->reply_queue_head, s->reply_queue_tail, s->busy);
+
+    return cmd;
+}
+
+static void megasas_complete_frame(MegasasState *s, uint64_t context)
+{
+    PCIDevice *pci_dev = PCI_DEVICE(s);
+    int tail, queue_offset;
+
+    /* Decrement busy count */
+    s->busy--;
+    if (s->reply_queue_pa) {
+        /*
+         * Put command on the reply queue.
+         * Context is opaque, but emulation is running in
+         * little endian. So convert it.
+         */
+        if (megasas_use_queue64(s)) {
+            queue_offset = s->reply_queue_head * sizeof(uint64_t);
+            stq_le_pci_dma(pci_dev, s->reply_queue_pa + queue_offset, context);
+        } else {
+            queue_offset = s->reply_queue_head * sizeof(uint32_t);
+            stl_le_pci_dma(pci_dev, s->reply_queue_pa + queue_offset, context);
+        }
+        s->reply_queue_tail = ldl_le_pci_dma(pci_dev, s->consumer_pa);
+        trace_megasas_qf_complete(context, s->reply_queue_head,
+                                  s->reply_queue_tail, s->busy);
+    }
+
+    if (megasas_intr_enabled(s)) {
+        /* Update reply queue pointer */
+        s->reply_queue_tail = ldl_le_pci_dma(pci_dev, s->consumer_pa);
+        tail = s->reply_queue_head;
+        s->reply_queue_head = megasas_next_index(s, tail, s->fw_cmds);
+        trace_megasas_qf_update(s->reply_queue_head, s->reply_queue_tail,
+                                s->busy);
+        stl_le_pci_dma(pci_dev, s->producer_pa, s->reply_queue_head);
+        /* Notify HBA */
+        if (msix_enabled(pci_dev)) {
+            trace_megasas_msix_raise(0);
+            msix_notify(pci_dev, 0);
+        } else if (msi_enabled(pci_dev)) {
+            trace_megasas_msi_raise(0);
+            msi_notify(pci_dev, 0);
+        } else {
+            s->doorbell++;
+            if (s->doorbell == 1) {
+                trace_megasas_irq_raise();
+                pci_irq_assert(pci_dev);
+            }
+        }
+    } else {
+        trace_megasas_qf_complete_noirq(context);
+    }
+}
+
+static void megasas_reset_frames(MegasasState *s)
+{
+    int i;
+    MegasasCmd *cmd;
+
+    for (i = 0; i < s->fw_cmds; i++) {
+        cmd = &s->frames[i];
+        if (cmd->pa) {
+            megasas_unmap_frame(s, cmd);
+        }
+    }
+    bitmap_zero(s->frame_map, MEGASAS_MAX_FRAMES);
+}
+
+static void megasas_abort_command(MegasasCmd *cmd)
+{
+    if (cmd->req) {
+        scsi_req_cancel(cmd->req);
+        cmd->req = NULL;
+    }
+}
+
+static int megasas_init_firmware(MegasasState *s, MegasasCmd *cmd)
+{
+    PCIDevice *pcid = PCI_DEVICE(s);
+    uint32_t pa_hi, pa_lo;
+    hwaddr iq_pa, initq_size = sizeof(struct mfi_init_qinfo);
+    struct mfi_init_qinfo *initq = NULL;
+    uint32_t flags;
+    int ret = MFI_STAT_OK;
+
+    if (s->reply_queue_pa) {
+        trace_megasas_initq_mapped(s->reply_queue_pa);
+        goto out;
+    }
+    pa_lo = le32_to_cpu(cmd->frame->init.qinfo_new_addr_lo);
+    pa_hi = le32_to_cpu(cmd->frame->init.qinfo_new_addr_hi);
+    iq_pa = (((uint64_t) pa_hi << 32) | pa_lo);
+    trace_megasas_init_firmware((uint64_t)iq_pa);
+    initq = pci_dma_map(pcid, iq_pa, &initq_size, 0);
+    if (!initq || initq_size != sizeof(*initq)) {
+        trace_megasas_initq_map_failed(cmd->index);
+        s->event_count++;
+        ret = MFI_STAT_MEMORY_NOT_AVAILABLE;
+        goto out;
+    }
+    s->reply_queue_len = le32_to_cpu(initq->rq_entries) & 0xFFFF;
+    if (s->reply_queue_len > s->fw_cmds) {
+        trace_megasas_initq_mismatch(s->reply_queue_len, s->fw_cmds);
+        s->event_count++;
+        ret = MFI_STAT_INVALID_PARAMETER;
+        goto out;
+    }
+    pa_lo = le32_to_cpu(initq->rq_addr_lo);
+    pa_hi = le32_to_cpu(initq->rq_addr_hi);
+    s->reply_queue_pa = ((uint64_t) pa_hi << 32) | pa_lo;
+    pa_lo = le32_to_cpu(initq->ci_addr_lo);
+    pa_hi = le32_to_cpu(initq->ci_addr_hi);
+    s->consumer_pa = ((uint64_t) pa_hi << 32) | pa_lo;
+    pa_lo = le32_to_cpu(initq->pi_addr_lo);
+    pa_hi = le32_to_cpu(initq->pi_addr_hi);
+    s->producer_pa = ((uint64_t) pa_hi << 32) | pa_lo;
+    s->reply_queue_head = ldl_le_pci_dma(pcid, s->producer_pa);
+    s->reply_queue_tail = ldl_le_pci_dma(pcid, s->consumer_pa);
+    flags = le32_to_cpu(initq->flags);
+    if (flags & MFI_QUEUE_FLAG_CONTEXT64) {
+        s->flags |= MEGASAS_MASK_USE_QUEUE64;
+    }
+    trace_megasas_init_queue((unsigned long)s->reply_queue_pa,
+                             s->reply_queue_len, s->reply_queue_head,
+                             s->reply_queue_tail, flags);
+    megasas_reset_frames(s);
+    s->fw_state = MFI_FWSTATE_OPERATIONAL;
+out:
+    if (initq) {
+        pci_dma_unmap(pcid, initq, initq_size, 0, 0);
+    }
+    return ret;
+}
+
+static int megasas_map_dcmd(MegasasState *s, MegasasCmd *cmd)
+{
+    dma_addr_t iov_pa, iov_size;
+
+    cmd->flags = le16_to_cpu(cmd->frame->header.flags);
+    if (!cmd->frame->header.sge_count) {
+        trace_megasas_dcmd_zero_sge(cmd->index);
+        cmd->iov_size = 0;
+        return 0;
+    } else if (cmd->frame->header.sge_count > 1) {
+        trace_megasas_dcmd_invalid_sge(cmd->index,
+                                       cmd->frame->header.sge_count);
+        cmd->iov_size = 0;
+        return -1;
+    }
+    iov_pa = megasas_sgl_get_addr(cmd, &cmd->frame->dcmd.sgl);
+    iov_size = megasas_sgl_get_len(cmd, &cmd->frame->dcmd.sgl);
+    pci_dma_sglist_init(&cmd->qsg, PCI_DEVICE(s), 1);
+    qemu_sglist_add(&cmd->qsg, iov_pa, iov_size);
+    cmd->iov_size = iov_size;
+    return cmd->iov_size;
+}
+
+static void megasas_finish_dcmd(MegasasCmd *cmd, uint32_t iov_size)
+{
+    trace_megasas_finish_dcmd(cmd->index, iov_size);
+
+    if (cmd->frame->header.sge_count) {
+        qemu_sglist_destroy(&cmd->qsg);
+    }
+    if (iov_size > cmd->iov_size) {
+        if (megasas_frame_is_ieee_sgl(cmd)) {
+            cmd->frame->dcmd.sgl.sg_skinny->len = cpu_to_le32(iov_size);
+        } else if (megasas_frame_is_sgl64(cmd)) {
+            cmd->frame->dcmd.sgl.sg64->len = cpu_to_le32(iov_size);
+        } else {
+            cmd->frame->dcmd.sgl.sg32->len = cpu_to_le32(iov_size);
+        }
+    }
+    cmd->iov_size = 0;
+}
+
+static int megasas_ctrl_get_info(MegasasState *s, MegasasCmd *cmd)
+{
+    PCIDevice *pci_dev = PCI_DEVICE(s);
+    PCIDeviceClass *pci_class = PCI_DEVICE_GET_CLASS(pci_dev);
+    MegasasBaseClass *base_class = MEGASAS_DEVICE_GET_CLASS(s);
+    struct mfi_ctrl_info info;
+    size_t dcmd_size = sizeof(info);
+    BusChild *kid;
+    int num_pd_disks = 0;
+
+    memset(&info, 0x0, cmd->iov_size);
+    if (cmd->iov_size < dcmd_size) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            dcmd_size);
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+
+    info.pci.vendor = cpu_to_le16(pci_class->vendor_id);
+    info.pci.device = cpu_to_le16(pci_class->device_id);
+    info.pci.subvendor = cpu_to_le16(pci_class->subsystem_vendor_id);
+    info.pci.subdevice = cpu_to_le16(pci_class->subsystem_id);
+
+    /*
+     * For some reason the firmware supports
+     * only up to 8 device ports.
+     * Despite supporting a far larger number
+     * of devices for the physical devices.
+     * So just display the first 8 devices
+     * in the device port list, independent
+     * of how many logical devices are actually
+     * present.
+     */
+    info.host.type = MFI_INFO_HOST_PCIE;
+    info.device.type = MFI_INFO_DEV_SAS3G;
+    info.device.port_count = 8;
+    QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+        SCSIDevice *sdev = DO_UPCAST(SCSIDevice, qdev, kid->child);
+        uint16_t pd_id;
+
+        if (num_pd_disks < 8) {
+            pd_id = ((sdev->id & 0xFF) << 8) | (sdev->lun & 0xFF);
+            info.device.port_addr[num_pd_disks] =
+                cpu_to_le64(megasas_get_sata_addr(pd_id));
+        }
+        num_pd_disks++;
+    }
+
+    memcpy(info.product_name, base_class->product_name, 24);
+    snprintf(info.serial_number, 32, "%s", s->hba_serial);
+    snprintf(info.package_version, 0x60, "%s-QEMU", QEMU_VERSION);
+    memcpy(info.image_component[0].name, "APP", 3);
+    snprintf(info.image_component[0].version, 10, "%s-QEMU",
+             base_class->product_version);
+    memcpy(info.image_component[0].build_date, "Apr  1 2014", 11);
+    memcpy(info.image_component[0].build_time, "12:34:56", 8);
+    info.image_component_count = 1;
+    if (pci_dev->has_rom) {
+        uint8_t biosver[32];
+        uint8_t *ptr;
+
+        ptr = memory_region_get_ram_ptr(&pci_dev->rom);
+        memcpy(biosver, ptr + 0x41, 31);
+        memcpy(info.image_component[1].name, "BIOS", 4);
+        memcpy(info.image_component[1].version, biosver,
+               strlen((const char *)biosver));
+        info.image_component_count++;
+    }
+    info.current_fw_time = cpu_to_le32(megasas_fw_time());
+    info.max_arms = 32;
+    info.max_spans = 8;
+    info.max_arrays = MEGASAS_MAX_ARRAYS;
+    info.max_lds = MFI_MAX_LD;
+    info.max_cmds = cpu_to_le16(s->fw_cmds);
+    info.max_sg_elements = cpu_to_le16(s->fw_sge);
+    info.max_request_size = cpu_to_le32(MEGASAS_MAX_SECTORS);
+    if (!megasas_is_jbod(s))
+        info.lds_present = cpu_to_le16(num_pd_disks);
+    info.pd_present = cpu_to_le16(num_pd_disks);
+    info.pd_disks_present = cpu_to_le16(num_pd_disks);
+    info.hw_present = cpu_to_le32(MFI_INFO_HW_NVRAM |
+                                   MFI_INFO_HW_MEM |
+                                   MFI_INFO_HW_FLASH);
+    info.memory_size = cpu_to_le16(512);
+    info.nvram_size = cpu_to_le16(32);
+    info.flash_size = cpu_to_le16(16);
+    info.raid_levels = cpu_to_le32(MFI_INFO_RAID_0);
+    info.adapter_ops = cpu_to_le32(MFI_INFO_AOPS_RBLD_RATE |
+                                    MFI_INFO_AOPS_SELF_DIAGNOSTIC |
+                                    MFI_INFO_AOPS_MIXED_ARRAY);
+    info.ld_ops = cpu_to_le32(MFI_INFO_LDOPS_DISK_CACHE_POLICY |
+                               MFI_INFO_LDOPS_ACCESS_POLICY |
+                               MFI_INFO_LDOPS_IO_POLICY |
+                               MFI_INFO_LDOPS_WRITE_POLICY |
+                               MFI_INFO_LDOPS_READ_POLICY);
+    info.max_strips_per_io = cpu_to_le16(s->fw_sge);
+    info.stripe_sz_ops.min = 3;
+    info.stripe_sz_ops.max = ctz32(MEGASAS_MAX_SECTORS + 1);
+    info.properties.pred_fail_poll_interval = cpu_to_le16(300);
+    info.properties.intr_throttle_cnt = cpu_to_le16(16);
+    info.properties.intr_throttle_timeout = cpu_to_le16(50);
+    info.properties.rebuild_rate = 30;
+    info.properties.patrol_read_rate = 30;
+    info.properties.bgi_rate = 30;
+    info.properties.cc_rate = 30;
+    info.properties.recon_rate = 30;
+    info.properties.cache_flush_interval = 4;
+    info.properties.spinup_drv_cnt = 2;
+    info.properties.spinup_delay = 6;
+    info.properties.ecc_bucket_size = 15;
+    info.properties.ecc_bucket_leak_rate = cpu_to_le16(1440);
+    info.properties.expose_encl_devices = 1;
+    info.properties.OnOffProperties = cpu_to_le32(MFI_CTRL_PROP_EnableJBOD);
+    info.pd_ops = cpu_to_le32(MFI_INFO_PDOPS_FORCE_ONLINE |
+                               MFI_INFO_PDOPS_FORCE_OFFLINE);
+    info.pd_mix_support = cpu_to_le32(MFI_INFO_PDMIX_SAS |
+                                       MFI_INFO_PDMIX_SATA |
+                                       MFI_INFO_PDMIX_LD);
+
+    cmd->iov_size -= dma_buf_read((uint8_t *)&info, dcmd_size, &cmd->qsg);
+    return MFI_STAT_OK;
+}
+
+static int megasas_mfc_get_defaults(MegasasState *s, MegasasCmd *cmd)
+{
+    struct mfi_defaults info;
+    size_t dcmd_size = sizeof(struct mfi_defaults);
+
+    memset(&info, 0x0, dcmd_size);
+    if (cmd->iov_size < dcmd_size) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            dcmd_size);
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+
+    info.sas_addr = cpu_to_le64(s->sas_addr);
+    info.stripe_size = 3;
+    info.flush_time = 4;
+    info.background_rate = 30;
+    info.allow_mix_in_enclosure = 1;
+    info.allow_mix_in_ld = 1;
+    info.direct_pd_mapping = 1;
+    /* Enable for BIOS support */
+    info.bios_enumerate_lds = 1;
+    info.disable_ctrl_r = 1;
+    info.expose_enclosure_devices = 1;
+    info.disable_preboot_cli = 1;
+    info.cluster_disable = 1;
+
+    cmd->iov_size -= dma_buf_read((uint8_t *)&info, dcmd_size, &cmd->qsg);
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_get_bios_info(MegasasState *s, MegasasCmd *cmd)
+{
+    struct mfi_bios_data info;
+    size_t dcmd_size = sizeof(info);
+
+    memset(&info, 0x0, dcmd_size);
+    if (cmd->iov_size < dcmd_size) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            dcmd_size);
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+    info.continue_on_error = 1;
+    info.verbose = 1;
+    if (megasas_is_jbod(s)) {
+        info.expose_all_drives = 1;
+    }
+
+    cmd->iov_size -= dma_buf_read((uint8_t *)&info, dcmd_size, &cmd->qsg);
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_get_fw_time(MegasasState *s, MegasasCmd *cmd)
+{
+    uint64_t fw_time;
+    size_t dcmd_size = sizeof(fw_time);
+
+    fw_time = cpu_to_le64(megasas_fw_time());
+
+    cmd->iov_size -= dma_buf_read((uint8_t *)&fw_time, dcmd_size, &cmd->qsg);
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_set_fw_time(MegasasState *s, MegasasCmd *cmd)
+{
+    uint64_t fw_time;
+
+    /* This is a dummy; setting of firmware time is not allowed */
+    memcpy(&fw_time, cmd->frame->dcmd.mbox, sizeof(fw_time));
+
+    trace_megasas_dcmd_set_fw_time(cmd->index, fw_time);
+    fw_time = cpu_to_le64(megasas_fw_time());
+    return MFI_STAT_OK;
+}
+
+static int megasas_event_info(MegasasState *s, MegasasCmd *cmd)
+{
+    struct mfi_evt_log_state info;
+    size_t dcmd_size = sizeof(info);
+
+    memset(&info, 0, dcmd_size);
+
+    info.newest_seq_num = cpu_to_le32(s->event_count);
+    info.shutdown_seq_num = cpu_to_le32(s->shutdown_event);
+    info.boot_seq_num = cpu_to_le32(s->boot_event);
+
+    cmd->iov_size -= dma_buf_read((uint8_t *)&info, dcmd_size, &cmd->qsg);
+    return MFI_STAT_OK;
+}
+
+static int megasas_event_wait(MegasasState *s, MegasasCmd *cmd)
+{
+    union mfi_evt event;
+
+    if (cmd->iov_size < sizeof(struct mfi_evt_detail)) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            sizeof(struct mfi_evt_detail));
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+    s->event_count = cpu_to_le32(cmd->frame->dcmd.mbox[0]);
+    event.word = cpu_to_le32(cmd->frame->dcmd.mbox[4]);
+    s->event_locale = event.members.locale;
+    s->event_class = event.members.class;
+    s->event_cmd = cmd;
+    /* Decrease busy count; event frame doesn't count here */
+    s->busy--;
+    cmd->iov_size = sizeof(struct mfi_evt_detail);
+    return MFI_STAT_INVALID_STATUS;
+}
+
+static int megasas_dcmd_pd_get_list(MegasasState *s, MegasasCmd *cmd)
+{
+    struct mfi_pd_list info;
+    size_t dcmd_size = sizeof(info);
+    BusChild *kid;
+    uint32_t offset, dcmd_limit, num_pd_disks = 0, max_pd_disks;
+
+    memset(&info, 0, dcmd_size);
+    offset = 8;
+    dcmd_limit = offset + sizeof(struct mfi_pd_address);
+    if (cmd->iov_size < dcmd_limit) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            dcmd_limit);
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+
+    max_pd_disks = (cmd->iov_size - offset) / sizeof(struct mfi_pd_address);
+    if (max_pd_disks > MFI_MAX_SYS_PDS) {
+        max_pd_disks = MFI_MAX_SYS_PDS;
+    }
+    QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+        SCSIDevice *sdev = DO_UPCAST(SCSIDevice, qdev, kid->child);
+        uint16_t pd_id;
+
+        if (num_pd_disks >= max_pd_disks)
+            break;
+
+        pd_id = ((sdev->id & 0xFF) << 8) | (sdev->lun & 0xFF);
+        info.addr[num_pd_disks].device_id = cpu_to_le16(pd_id);
+        info.addr[num_pd_disks].encl_device_id = 0xFFFF;
+        info.addr[num_pd_disks].encl_index = 0;
+        info.addr[num_pd_disks].slot_number = sdev->id & 0xFF;
+        info.addr[num_pd_disks].scsi_dev_type = sdev->type;
+        info.addr[num_pd_disks].connect_port_bitmap = 0x1;
+        info.addr[num_pd_disks].sas_addr[0] =
+            cpu_to_le64(megasas_get_sata_addr(pd_id));
+        num_pd_disks++;
+        offset += sizeof(struct mfi_pd_address);
+    }
+    trace_megasas_dcmd_pd_get_list(cmd->index, num_pd_disks,
+                                   max_pd_disks, offset);
+
+    info.size = cpu_to_le32(offset);
+    info.count = cpu_to_le32(num_pd_disks);
+
+    cmd->iov_size -= dma_buf_read((uint8_t *)&info, offset, &cmd->qsg);
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_pd_list_query(MegasasState *s, MegasasCmd *cmd)
+{
+    uint16_t flags;
+
+    /* mbox0 contains flags */
+    flags = le16_to_cpu(cmd->frame->dcmd.mbox[0]);
+    trace_megasas_dcmd_pd_list_query(cmd->index, flags);
+    if (flags == MR_PD_QUERY_TYPE_ALL ||
+        megasas_is_jbod(s)) {
+        return megasas_dcmd_pd_get_list(s, cmd);
+    }
+
+    return MFI_STAT_OK;
+}
+
+static int megasas_pd_get_info_submit(SCSIDevice *sdev, int lun,
+                                      MegasasCmd *cmd)
+{
+    struct mfi_pd_info *info = cmd->iov_buf;
+    size_t dcmd_size = sizeof(struct mfi_pd_info);
+    uint64_t pd_size;
+    uint16_t pd_id = ((sdev->id & 0xFF) << 8) | (lun & 0xFF);
+    uint8_t cmdbuf[6];
+    SCSIRequest *req;
+    size_t len, resid;
+
+    if (!cmd->iov_buf) {
+        cmd->iov_buf = g_malloc0(dcmd_size);
+        info = cmd->iov_buf;
+        info->inquiry_data[0] = 0x7f; /* Force PQual 0x3, PType 0x1f */
+        info->vpd_page83[0] = 0x7f;
+        megasas_setup_inquiry(cmdbuf, 0, sizeof(info->inquiry_data));
+        req = scsi_req_new(sdev, cmd->index, lun, cmdbuf, cmd);
+        if (!req) {
+            trace_megasas_dcmd_req_alloc_failed(cmd->index,
+                                                "PD get info std inquiry");
+            g_free(cmd->iov_buf);
+            cmd->iov_buf = NULL;
+            return MFI_STAT_FLASH_ALLOC_FAIL;
+        }
+        trace_megasas_dcmd_internal_submit(cmd->index,
+                                           "PD get info std inquiry", lun);
+        len = scsi_req_enqueue(req);
+        if (len > 0) {
+            cmd->iov_size = len;
+            scsi_req_continue(req);
+        }
+        return MFI_STAT_INVALID_STATUS;
+    } else if (info->inquiry_data[0] != 0x7f && info->vpd_page83[0] == 0x7f) {
+        megasas_setup_inquiry(cmdbuf, 0x83, sizeof(info->vpd_page83));
+        req = scsi_req_new(sdev, cmd->index, lun, cmdbuf, cmd);
+        if (!req) {
+            trace_megasas_dcmd_req_alloc_failed(cmd->index,
+                                                "PD get info vpd inquiry");
+            return MFI_STAT_FLASH_ALLOC_FAIL;
+        }
+        trace_megasas_dcmd_internal_submit(cmd->index,
+                                           "PD get info vpd inquiry", lun);
+        len = scsi_req_enqueue(req);
+        if (len > 0) {
+            cmd->iov_size = len;
+            scsi_req_continue(req);
+        }
+        return MFI_STAT_INVALID_STATUS;
+    }
+    /* Finished, set FW state */
+    if ((info->inquiry_data[0] >> 5) == 0) {
+        if (megasas_is_jbod(cmd->state)) {
+            info->fw_state = cpu_to_le16(MFI_PD_STATE_SYSTEM);
+        } else {
+            info->fw_state = cpu_to_le16(MFI_PD_STATE_ONLINE);
+        }
+    } else {
+        info->fw_state = cpu_to_le16(MFI_PD_STATE_OFFLINE);
+    }
+
+    info->ref.v.device_id = cpu_to_le16(pd_id);
+    info->state.ddf.pd_type = cpu_to_le16(MFI_PD_DDF_TYPE_IN_VD|
+                                          MFI_PD_DDF_TYPE_INTF_SAS);
+    blk_get_geometry(sdev->conf.blk, &pd_size);
+    info->raw_size = cpu_to_le64(pd_size);
+    info->non_coerced_size = cpu_to_le64(pd_size);
+    info->coerced_size = cpu_to_le64(pd_size);
+    info->encl_device_id = 0xFFFF;
+    info->slot_number = (sdev->id & 0xFF);
+    info->path_info.count = 1;
+    info->path_info.sas_addr[0] =
+        cpu_to_le64(megasas_get_sata_addr(pd_id));
+    info->connected_port_bitmap = 0x1;
+    info->device_speed = 1;
+    info->link_speed = 1;
+    resid = dma_buf_read(cmd->iov_buf, dcmd_size, &cmd->qsg);
+    g_free(cmd->iov_buf);
+    cmd->iov_size = dcmd_size - resid;
+    cmd->iov_buf = NULL;
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_pd_get_info(MegasasState *s, MegasasCmd *cmd)
+{
+    size_t dcmd_size = sizeof(struct mfi_pd_info);
+    uint16_t pd_id;
+    uint8_t target_id, lun_id;
+    SCSIDevice *sdev = NULL;
+    int retval = MFI_STAT_DEVICE_NOT_FOUND;
+
+    if (cmd->iov_size < dcmd_size) {
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+
+    /* mbox0 has the ID */
+    pd_id = le16_to_cpu(cmd->frame->dcmd.mbox[0]);
+    target_id = (pd_id >> 8) & 0xFF;
+    lun_id = pd_id & 0xFF;
+    sdev = scsi_device_find(&s->bus, 0, target_id, lun_id);
+    trace_megasas_dcmd_pd_get_info(cmd->index, pd_id);
+
+    if (sdev) {
+        /* Submit inquiry */
+        retval = megasas_pd_get_info_submit(sdev, pd_id, cmd);
+    }
+
+    return retval;
+}
+
+static int megasas_dcmd_ld_get_list(MegasasState *s, MegasasCmd *cmd)
+{
+    struct mfi_ld_list info;
+    size_t dcmd_size = sizeof(info), resid;
+    uint32_t num_ld_disks = 0, max_ld_disks;
+    uint64_t ld_size;
+    BusChild *kid;
+
+    memset(&info, 0, dcmd_size);
+    if (cmd->iov_size > dcmd_size) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            dcmd_size);
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+
+    max_ld_disks = (cmd->iov_size - 8) / 16;
+    if (megasas_is_jbod(s)) {
+        max_ld_disks = 0;
+    }
+    if (max_ld_disks > MFI_MAX_LD) {
+        max_ld_disks = MFI_MAX_LD;
+    }
+    QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+        SCSIDevice *sdev = DO_UPCAST(SCSIDevice, qdev, kid->child);
+
+        if (num_ld_disks >= max_ld_disks) {
+            break;
+        }
+        /* Logical device size is in blocks */
+        blk_get_geometry(sdev->conf.blk, &ld_size);
+        info.ld_list[num_ld_disks].ld.v.target_id = sdev->id;
+        info.ld_list[num_ld_disks].state = MFI_LD_STATE_OPTIMAL;
+        info.ld_list[num_ld_disks].size = cpu_to_le64(ld_size);
+        num_ld_disks++;
+    }
+    info.ld_count = cpu_to_le32(num_ld_disks);
+    trace_megasas_dcmd_ld_get_list(cmd->index, num_ld_disks, max_ld_disks);
+
+    resid = dma_buf_read((uint8_t *)&info, dcmd_size, &cmd->qsg);
+    cmd->iov_size = dcmd_size - resid;
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_ld_list_query(MegasasState *s, MegasasCmd *cmd)
+{
+    uint16_t flags;
+    struct mfi_ld_targetid_list info;
+    size_t dcmd_size = sizeof(info), resid;
+    uint32_t num_ld_disks = 0, max_ld_disks = s->fw_luns;
+    BusChild *kid;
+
+    /* mbox0 contains flags */
+    flags = le16_to_cpu(cmd->frame->dcmd.mbox[0]);
+    trace_megasas_dcmd_ld_list_query(cmd->index, flags);
+    if (flags != MR_LD_QUERY_TYPE_ALL &&
+        flags != MR_LD_QUERY_TYPE_EXPOSED_TO_HOST) {
+        max_ld_disks = 0;
+    }
+
+    memset(&info, 0, dcmd_size);
+    if (cmd->iov_size < 12) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            dcmd_size);
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+    dcmd_size = sizeof(uint32_t) * 2 + 3;
+    max_ld_disks = cmd->iov_size - dcmd_size;
+    if (megasas_is_jbod(s)) {
+        max_ld_disks = 0;
+    }
+    if (max_ld_disks > MFI_MAX_LD) {
+        max_ld_disks = MFI_MAX_LD;
+    }
+    QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+        SCSIDevice *sdev = DO_UPCAST(SCSIDevice, qdev, kid->child);
+
+        if (num_ld_disks >= max_ld_disks) {
+            break;
+        }
+        info.targetid[num_ld_disks] = sdev->lun;
+        num_ld_disks++;
+        dcmd_size++;
+    }
+    info.ld_count = cpu_to_le32(num_ld_disks);
+    info.size = dcmd_size;
+    trace_megasas_dcmd_ld_get_list(cmd->index, num_ld_disks, max_ld_disks);
+
+    resid = dma_buf_read((uint8_t *)&info, dcmd_size, &cmd->qsg);
+    cmd->iov_size = dcmd_size - resid;
+    return MFI_STAT_OK;
+}
+
+static int megasas_ld_get_info_submit(SCSIDevice *sdev, int lun,
+                                      MegasasCmd *cmd)
+{
+    struct mfi_ld_info *info = cmd->iov_buf;
+    size_t dcmd_size = sizeof(struct mfi_ld_info);
+    uint8_t cdb[6];
+    SCSIRequest *req;
+    ssize_t len, resid;
+    uint16_t sdev_id = ((sdev->id & 0xFF) << 8) | (lun & 0xFF);
+    uint64_t ld_size;
+
+    if (!cmd->iov_buf) {
+        cmd->iov_buf = g_malloc0(dcmd_size);
+        info = cmd->iov_buf;
+        megasas_setup_inquiry(cdb, 0x83, sizeof(info->vpd_page83));
+        req = scsi_req_new(sdev, cmd->index, lun, cdb, cmd);
+        if (!req) {
+            trace_megasas_dcmd_req_alloc_failed(cmd->index,
+                                                "LD get info vpd inquiry");
+            g_free(cmd->iov_buf);
+            cmd->iov_buf = NULL;
+            return MFI_STAT_FLASH_ALLOC_FAIL;
+        }
+        trace_megasas_dcmd_internal_submit(cmd->index,
+                                           "LD get info vpd inquiry", lun);
+        len = scsi_req_enqueue(req);
+        if (len > 0) {
+            cmd->iov_size = len;
+            scsi_req_continue(req);
+        }
+        return MFI_STAT_INVALID_STATUS;
+    }
+
+    info->ld_config.params.state = MFI_LD_STATE_OPTIMAL;
+    info->ld_config.properties.ld.v.target_id = lun;
+    info->ld_config.params.stripe_size = 3;
+    info->ld_config.params.num_drives = 1;
+    info->ld_config.params.is_consistent = 1;
+    /* Logical device size is in blocks */
+    blk_get_geometry(sdev->conf.blk, &ld_size);
+    info->size = cpu_to_le64(ld_size);
+    memset(info->ld_config.span, 0, sizeof(info->ld_config.span));
+    info->ld_config.span[0].start_block = 0;
+    info->ld_config.span[0].num_blocks = info->size;
+    info->ld_config.span[0].array_ref = cpu_to_le16(sdev_id);
+
+    resid = dma_buf_read(cmd->iov_buf, dcmd_size, &cmd->qsg);
+    g_free(cmd->iov_buf);
+    cmd->iov_size = dcmd_size - resid;
+    cmd->iov_buf = NULL;
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_ld_get_info(MegasasState *s, MegasasCmd *cmd)
+{
+    struct mfi_ld_info info;
+    size_t dcmd_size = sizeof(info);
+    uint16_t ld_id;
+    uint32_t max_ld_disks = s->fw_luns;
+    SCSIDevice *sdev = NULL;
+    int retval = MFI_STAT_DEVICE_NOT_FOUND;
+
+    if (cmd->iov_size < dcmd_size) {
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+
+    /* mbox0 has the ID */
+    ld_id = le16_to_cpu(cmd->frame->dcmd.mbox[0]);
+    trace_megasas_dcmd_ld_get_info(cmd->index, ld_id);
+
+    if (megasas_is_jbod(s)) {
+        return MFI_STAT_DEVICE_NOT_FOUND;
+    }
+
+    if (ld_id < max_ld_disks) {
+        sdev = scsi_device_find(&s->bus, 0, ld_id, 0);
+    }
+
+    if (sdev) {
+        retval = megasas_ld_get_info_submit(sdev, ld_id, cmd);
+    }
+
+    return retval;
+}
+
+static int megasas_dcmd_cfg_read(MegasasState *s, MegasasCmd *cmd)
+{
+    uint8_t data[4096];
+    struct mfi_config_data *info;
+    int num_pd_disks = 0, array_offset, ld_offset;
+    BusChild *kid;
+
+    if (cmd->iov_size > 4096) {
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+
+    QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+        num_pd_disks++;
+    }
+    info = (struct mfi_config_data *)&data;
+    /*
+     * Array mapping:
+     * - One array per SCSI device
+     * - One logical drive per SCSI device
+     *   spanning the entire device
+     */
+    info->array_count = num_pd_disks;
+    info->array_size = sizeof(struct mfi_array) * num_pd_disks;
+    info->log_drv_count = num_pd_disks;
+    info->log_drv_size = sizeof(struct mfi_ld_config) * num_pd_disks;
+    info->spares_count = 0;
+    info->spares_size = sizeof(struct mfi_spare);
+    info->size = sizeof(struct mfi_config_data) + info->array_size +
+        info->log_drv_size;
+    if (info->size > 4096) {
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+
+    array_offset = sizeof(struct mfi_config_data);
+    ld_offset = array_offset + sizeof(struct mfi_array) * num_pd_disks;
+
+    QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+        SCSIDevice *sdev = DO_UPCAST(SCSIDevice, qdev, kid->child);
+        uint16_t sdev_id = ((sdev->id & 0xFF) << 8) | (sdev->lun & 0xFF);
+        struct mfi_array *array;
+        struct mfi_ld_config *ld;
+        uint64_t pd_size;
+        int i;
+
+        array = (struct mfi_array *)(data + array_offset);
+        blk_get_geometry(sdev->conf.blk, &pd_size);
+        array->size = cpu_to_le64(pd_size);
+        array->num_drives = 1;
+        array->array_ref = cpu_to_le16(sdev_id);
+        array->pd[0].ref.v.device_id = cpu_to_le16(sdev_id);
+        array->pd[0].ref.v.seq_num = 0;
+        array->pd[0].fw_state = MFI_PD_STATE_ONLINE;
+        array->pd[0].encl.pd = 0xFF;
+        array->pd[0].encl.slot = (sdev->id & 0xFF);
+        for (i = 1; i < MFI_MAX_ROW_SIZE; i++) {
+            array->pd[i].ref.v.device_id = 0xFFFF;
+            array->pd[i].ref.v.seq_num = 0;
+            array->pd[i].fw_state = MFI_PD_STATE_UNCONFIGURED_GOOD;
+            array->pd[i].encl.pd = 0xFF;
+            array->pd[i].encl.slot = 0xFF;
+        }
+        array_offset += sizeof(struct mfi_array);
+        ld = (struct mfi_ld_config *)(data + ld_offset);
+        memset(ld, 0, sizeof(struct mfi_ld_config));
+        ld->properties.ld.v.target_id = sdev->id;
+        ld->properties.default_cache_policy = MR_LD_CACHE_READ_AHEAD |
+            MR_LD_CACHE_READ_ADAPTIVE;
+        ld->properties.current_cache_policy = MR_LD_CACHE_READ_AHEAD |
+            MR_LD_CACHE_READ_ADAPTIVE;
+        ld->params.state = MFI_LD_STATE_OPTIMAL;
+        ld->params.stripe_size = 3;
+        ld->params.num_drives = 1;
+        ld->params.span_depth = 1;
+        ld->params.is_consistent = 1;
+        ld->span[0].start_block = 0;
+        ld->span[0].num_blocks = cpu_to_le64(pd_size);
+        ld->span[0].array_ref = cpu_to_le16(sdev_id);
+        ld_offset += sizeof(struct mfi_ld_config);
+    }
+
+    cmd->iov_size -= dma_buf_read((uint8_t *)data, info->size, &cmd->qsg);
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_get_properties(MegasasState *s, MegasasCmd *cmd)
+{
+    struct mfi_ctrl_props info;
+    size_t dcmd_size = sizeof(info);
+
+    memset(&info, 0x0, dcmd_size);
+    if (cmd->iov_size < dcmd_size) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            dcmd_size);
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+    info.pred_fail_poll_interval = cpu_to_le16(300);
+    info.intr_throttle_cnt = cpu_to_le16(16);
+    info.intr_throttle_timeout = cpu_to_le16(50);
+    info.rebuild_rate = 30;
+    info.patrol_read_rate = 30;
+    info.bgi_rate = 30;
+    info.cc_rate = 30;
+    info.recon_rate = 30;
+    info.cache_flush_interval = 4;
+    info.spinup_drv_cnt = 2;
+    info.spinup_delay = 6;
+    info.ecc_bucket_size = 15;
+    info.ecc_bucket_leak_rate = cpu_to_le16(1440);
+    info.expose_encl_devices = 1;
+
+    cmd->iov_size -= dma_buf_read((uint8_t *)&info, dcmd_size, &cmd->qsg);
+    return MFI_STAT_OK;
+}
+
+static int megasas_cache_flush(MegasasState *s, MegasasCmd *cmd)
+{
+    blk_drain_all();
+    return MFI_STAT_OK;
+}
+
+static int megasas_ctrl_shutdown(MegasasState *s, MegasasCmd *cmd)
+{
+    s->fw_state = MFI_FWSTATE_READY;
+    return MFI_STAT_OK;
+}
+
+/* Some implementations use CLUSTER RESET LD to simulate a device reset */
+static int megasas_cluster_reset_ld(MegasasState *s, MegasasCmd *cmd)
+{
+    uint16_t target_id;
+    int i;
+
+    /* mbox0 contains the device index */
+    target_id = le16_to_cpu(cmd->frame->dcmd.mbox[0]);
+    trace_megasas_dcmd_reset_ld(cmd->index, target_id);
+    for (i = 0; i < s->fw_cmds; i++) {
+        MegasasCmd *tmp_cmd = &s->frames[i];
+        if (tmp_cmd->req && tmp_cmd->req->dev->id == target_id) {
+            SCSIDevice *d = tmp_cmd->req->dev;
+            qdev_reset_all(&d->qdev);
+        }
+    }
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_set_properties(MegasasState *s, MegasasCmd *cmd)
+{
+    struct mfi_ctrl_props info;
+    size_t dcmd_size = sizeof(info);
+
+    if (cmd->iov_size < dcmd_size) {
+        trace_megasas_dcmd_invalid_xfer_len(cmd->index, cmd->iov_size,
+                                            dcmd_size);
+        return MFI_STAT_INVALID_PARAMETER;
+    }
+    dma_buf_write((uint8_t *)&info, cmd->iov_size, &cmd->qsg);
+    trace_megasas_dcmd_unsupported(cmd->index, cmd->iov_size);
+    return MFI_STAT_OK;
+}
+
+static int megasas_dcmd_dummy(MegasasState *s, MegasasCmd *cmd)
+{
+    trace_megasas_dcmd_dummy(cmd->index, cmd->iov_size);
+    return MFI_STAT_OK;
+}
+
+static const struct dcmd_cmd_tbl_t {
+    int opcode;
+    const char *desc;
+    int (*func)(MegasasState *s, MegasasCmd *cmd);
+} dcmd_cmd_tbl[] = {
+    { MFI_DCMD_CTRL_MFI_HOST_MEM_ALLOC, "CTRL_HOST_MEM_ALLOC",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_GET_INFO, "CTRL_GET_INFO",
+      megasas_ctrl_get_info },
+    { MFI_DCMD_CTRL_GET_PROPERTIES, "CTRL_GET_PROPERTIES",
+      megasas_dcmd_get_properties },
+    { MFI_DCMD_CTRL_SET_PROPERTIES, "CTRL_SET_PROPERTIES",
+      megasas_dcmd_set_properties },
+    { MFI_DCMD_CTRL_ALARM_GET, "CTRL_ALARM_GET",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_ALARM_ENABLE, "CTRL_ALARM_ENABLE",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_ALARM_DISABLE, "CTRL_ALARM_DISABLE",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_ALARM_SILENCE, "CTRL_ALARM_SILENCE",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_ALARM_TEST, "CTRL_ALARM_TEST",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_EVENT_GETINFO, "CTRL_EVENT_GETINFO",
+      megasas_event_info },
+    { MFI_DCMD_CTRL_EVENT_GET, "CTRL_EVENT_GET",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_EVENT_WAIT, "CTRL_EVENT_WAIT",
+      megasas_event_wait },
+    { MFI_DCMD_CTRL_SHUTDOWN, "CTRL_SHUTDOWN",
+      megasas_ctrl_shutdown },
+    { MFI_DCMD_HIBERNATE_STANDBY, "CTRL_STANDBY",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_GET_TIME, "CTRL_GET_TIME",
+      megasas_dcmd_get_fw_time },
+    { MFI_DCMD_CTRL_SET_TIME, "CTRL_SET_TIME",
+      megasas_dcmd_set_fw_time },
+    { MFI_DCMD_CTRL_BIOS_DATA_GET, "CTRL_BIOS_DATA_GET",
+      megasas_dcmd_get_bios_info },
+    { MFI_DCMD_CTRL_FACTORY_DEFAULTS, "CTRL_FACTORY_DEFAULTS",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_MFC_DEFAULTS_GET, "CTRL_MFC_DEFAULTS_GET",
+      megasas_mfc_get_defaults },
+    { MFI_DCMD_CTRL_MFC_DEFAULTS_SET, "CTRL_MFC_DEFAULTS_SET",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CTRL_CACHE_FLUSH, "CTRL_CACHE_FLUSH",
+      megasas_cache_flush },
+    { MFI_DCMD_PD_GET_LIST, "PD_GET_LIST",
+      megasas_dcmd_pd_get_list },
+    { MFI_DCMD_PD_LIST_QUERY, "PD_LIST_QUERY",
+      megasas_dcmd_pd_list_query },
+    { MFI_DCMD_PD_GET_INFO, "PD_GET_INFO",
+      megasas_dcmd_pd_get_info },
+    { MFI_DCMD_PD_STATE_SET, "PD_STATE_SET",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_PD_REBUILD, "PD_REBUILD",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_PD_BLINK, "PD_BLINK",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_PD_UNBLINK, "PD_UNBLINK",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_LD_GET_LIST, "LD_GET_LIST",
+      megasas_dcmd_ld_get_list},
+    { MFI_DCMD_LD_LIST_QUERY, "LD_LIST_QUERY",
+      megasas_dcmd_ld_list_query },
+    { MFI_DCMD_LD_GET_INFO, "LD_GET_INFO",
+      megasas_dcmd_ld_get_info },
+    { MFI_DCMD_LD_GET_PROP, "LD_GET_PROP",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_LD_SET_PROP, "LD_SET_PROP",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_LD_DELETE, "LD_DELETE",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CFG_READ, "CFG_READ",
+      megasas_dcmd_cfg_read },
+    { MFI_DCMD_CFG_ADD, "CFG_ADD",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CFG_CLEAR, "CFG_CLEAR",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CFG_FOREIGN_READ, "CFG_FOREIGN_READ",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CFG_FOREIGN_IMPORT, "CFG_FOREIGN_IMPORT",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_BBU_STATUS, "BBU_STATUS",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_BBU_CAPACITY_INFO, "BBU_CAPACITY_INFO",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_BBU_DESIGN_INFO, "BBU_DESIGN_INFO",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_BBU_PROP_GET, "BBU_PROP_GET",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CLUSTER, "CLUSTER",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CLUSTER_RESET_ALL, "CLUSTER_RESET_ALL",
+      megasas_dcmd_dummy },
+    { MFI_DCMD_CLUSTER_RESET_LD, "CLUSTER_RESET_LD",
+      megasas_cluster_reset_ld },
+    { -1, NULL, NULL }
+};
+
+static int megasas_handle_dcmd(MegasasState *s, MegasasCmd *cmd)
+{
+    int opcode, len;
+    int retval = 0;
+    const struct dcmd_cmd_tbl_t *cmdptr = dcmd_cmd_tbl;
+
+    opcode = le32_to_cpu(cmd->frame->dcmd.opcode);
+    trace_megasas_handle_dcmd(cmd->index, opcode);
+    len = megasas_map_dcmd(s, cmd);
+    if (len < 0) {
+        return MFI_STAT_MEMORY_NOT_AVAILABLE;
+    }
+    while (cmdptr->opcode != -1 && cmdptr->opcode != opcode) {
+        cmdptr++;
+    }
+    if (cmdptr->opcode == -1) {
+        trace_megasas_dcmd_unhandled(cmd->index, opcode, len);
+        retval = megasas_dcmd_dummy(s, cmd);
+    } else {
+        trace_megasas_dcmd_enter(cmd->index, cmdptr->desc, len);
+        retval = cmdptr->func(s, cmd);
+    }
+    if (retval != MFI_STAT_INVALID_STATUS) {
+        megasas_finish_dcmd(cmd, len);
+    }
+    return retval;
+}
+
+static int megasas_finish_internal_dcmd(MegasasCmd *cmd,
+                                        SCSIRequest *req)
+{
+    int opcode;
+    int retval = MFI_STAT_OK;
+    int lun = req->lun;
+
+    opcode = le32_to_cpu(cmd->frame->dcmd.opcode);
+    scsi_req_unref(req);
+    trace_megasas_dcmd_internal_finish(cmd->index, opcode, lun);
+    switch (opcode) {
+    case MFI_DCMD_PD_GET_INFO:
+        retval = megasas_pd_get_info_submit(req->dev, lun, cmd);
+        break;
+    case MFI_DCMD_LD_GET_INFO:
+        retval = megasas_ld_get_info_submit(req->dev, lun, cmd);
+        break;
+    default:
+        trace_megasas_dcmd_internal_invalid(cmd->index, opcode);
+        retval = MFI_STAT_INVALID_DCMD;
+        break;
+    }
+    if (retval != MFI_STAT_INVALID_STATUS) {
+        megasas_finish_dcmd(cmd, cmd->iov_size);
+    }
+    return retval;
+}
+
+static int megasas_enqueue_req(MegasasCmd *cmd, bool is_write)
+{
+    int len;
+
+    len = scsi_req_enqueue(cmd->req);
+    if (len < 0) {
+        len = -len;
+    }
+    if (len > 0) {
+        if (len > cmd->iov_size) {
+            if (is_write) {
+                trace_megasas_iov_write_overflow(cmd->index, len,
+                                                 cmd->iov_size);
+            } else {
+                trace_megasas_iov_read_overflow(cmd->index, len,
+                                                cmd->iov_size);
+            }
+        }
+        if (len < cmd->iov_size) {
+            if (is_write) {
+                trace_megasas_iov_write_underflow(cmd->index, len,
+                                                  cmd->iov_size);
+            } else {
+                trace_megasas_iov_read_underflow(cmd->index, len,
+                                                 cmd->iov_size);
+            }
+            cmd->iov_size = len;
+        }
+        scsi_req_continue(cmd->req);
+    }
+    return len;
+}
+
+static int megasas_handle_scsi(MegasasState *s, MegasasCmd *cmd,
+                               bool is_logical)
+{
+    uint8_t *cdb;
+    bool is_write;
+    struct SCSIDevice *sdev = NULL;
+
+    cdb = cmd->frame->pass.cdb;
+
+    if (is_logical) {
+        if (cmd->frame->header.target_id >= MFI_MAX_LD ||
+            cmd->frame->header.lun_id != 0) {
+            trace_megasas_scsi_target_not_present(
+                mfi_frame_desc[cmd->frame->header.frame_cmd], is_logical,
+                cmd->frame->header.target_id, cmd->frame->header.lun_id);
+            return MFI_STAT_DEVICE_NOT_FOUND;
+        }
+    }
+    sdev = scsi_device_find(&s->bus, 0, cmd->frame->header.target_id,
+                            cmd->frame->header.lun_id);
+
+    cmd->iov_size = le32_to_cpu(cmd->frame->header.data_len);
+    trace_megasas_handle_scsi(mfi_frame_desc[cmd->frame->header.frame_cmd],
+                              is_logical, cmd->frame->header.target_id,
+                              cmd->frame->header.lun_id, sdev, cmd->iov_size);
+
+    if (!sdev || (megasas_is_jbod(s) && is_logical)) {
+        trace_megasas_scsi_target_not_present(
+            mfi_frame_desc[cmd->frame->header.frame_cmd], is_logical,
+            cmd->frame->header.target_id, cmd->frame->header.lun_id);
+        return MFI_STAT_DEVICE_NOT_FOUND;
+    }
+
+    if (cmd->frame->header.cdb_len > 16) {
+        trace_megasas_scsi_invalid_cdb_len(
+                mfi_frame_desc[cmd->frame->header.frame_cmd], is_logical,
+                cmd->frame->header.target_id, cmd->frame->header.lun_id,
+                cmd->frame->header.cdb_len);
+        megasas_write_sense(cmd, SENSE_CODE(INVALID_OPCODE));
+        cmd->frame->header.scsi_status = CHECK_CONDITION;
+        s->event_count++;
+        return MFI_STAT_SCSI_DONE_WITH_ERROR;
+    }
+
+    if (megasas_map_sgl(s, cmd, &cmd->frame->pass.sgl)) {
+        megasas_write_sense(cmd, SENSE_CODE(TARGET_FAILURE));
+        cmd->frame->header.scsi_status = CHECK_CONDITION;
+        s->event_count++;
+        return MFI_STAT_SCSI_DONE_WITH_ERROR;
+    }
+
+    cmd->req = scsi_req_new(sdev, cmd->index,
+                            cmd->frame->header.lun_id, cdb, cmd);
+    if (!cmd->req) {
+        trace_megasas_scsi_req_alloc_failed(
+                mfi_frame_desc[cmd->frame->header.frame_cmd],
+                cmd->frame->header.target_id, cmd->frame->header.lun_id);
+        megasas_write_sense(cmd, SENSE_CODE(NO_SENSE));
+        cmd->frame->header.scsi_status = BUSY;
+        s->event_count++;
+        return MFI_STAT_SCSI_DONE_WITH_ERROR;
+    }
+
+    is_write = (cmd->req->cmd.mode == SCSI_XFER_TO_DEV);
+    if (cmd->iov_size) {
+        if (is_write) {
+            trace_megasas_scsi_write_start(cmd->index, cmd->iov_size);
+        } else {
+            trace_megasas_scsi_read_start(cmd->index, cmd->iov_size);
+        }
+    } else {
+        trace_megasas_scsi_nodata(cmd->index);
+    }
+    megasas_enqueue_req(cmd, is_write);
+    return MFI_STAT_INVALID_STATUS;
+}
+
+static int megasas_handle_io(MegasasState *s, MegasasCmd *cmd)
+{
+    uint32_t lba_count, lba_start_hi, lba_start_lo;
+    uint64_t lba_start;
+    bool is_write = (cmd->frame->header.frame_cmd == MFI_CMD_LD_WRITE);
+    uint8_t cdb[16];
+    int len;
+    struct SCSIDevice *sdev = NULL;
+
+    lba_count = le32_to_cpu(cmd->frame->io.header.data_len);
+    lba_start_lo = le32_to_cpu(cmd->frame->io.lba_lo);
+    lba_start_hi = le32_to_cpu(cmd->frame->io.lba_hi);
+    lba_start = ((uint64_t)lba_start_hi << 32) | lba_start_lo;
+
+    if (cmd->frame->header.target_id < MFI_MAX_LD &&
+        cmd->frame->header.lun_id == 0) {
+        sdev = scsi_device_find(&s->bus, 0, cmd->frame->header.target_id,
+                                cmd->frame->header.lun_id);
+    }
+
+    trace_megasas_handle_io(cmd->index,
+                            mfi_frame_desc[cmd->frame->header.frame_cmd],
+                            cmd->frame->header.target_id,
+                            cmd->frame->header.lun_id,
+                            (unsigned long)lba_start, (unsigned long)lba_count);
+    if (!sdev) {
+        trace_megasas_io_target_not_present(cmd->index,
+            mfi_frame_desc[cmd->frame->header.frame_cmd],
+            cmd->frame->header.target_id, cmd->frame->header.lun_id);
+        return MFI_STAT_DEVICE_NOT_FOUND;
+    }
+
+    if (cmd->frame->header.cdb_len > 16) {
+        trace_megasas_scsi_invalid_cdb_len(
+            mfi_frame_desc[cmd->frame->header.frame_cmd], 1,
+            cmd->frame->header.target_id, cmd->frame->header.lun_id,
+            cmd->frame->header.cdb_len);
+        megasas_write_sense(cmd, SENSE_CODE(INVALID_OPCODE));
+        cmd->frame->header.scsi_status = CHECK_CONDITION;
+        s->event_count++;
+        return MFI_STAT_SCSI_DONE_WITH_ERROR;
+    }
+
+    cmd->iov_size = lba_count * sdev->blocksize;
+    if (megasas_map_sgl(s, cmd, &cmd->frame->io.sgl)) {
+        megasas_write_sense(cmd, SENSE_CODE(TARGET_FAILURE));
+        cmd->frame->header.scsi_status = CHECK_CONDITION;
+        s->event_count++;
+        return MFI_STAT_SCSI_DONE_WITH_ERROR;
+    }
+
+    megasas_encode_lba(cdb, lba_start, lba_count, is_write);
+    cmd->req = scsi_req_new(sdev, cmd->index,
+                            cmd->frame->header.lun_id, cdb, cmd);
+    if (!cmd->req) {
+        trace_megasas_scsi_req_alloc_failed(
+            mfi_frame_desc[cmd->frame->header.frame_cmd],
+            cmd->frame->header.target_id, cmd->frame->header.lun_id);
+        megasas_write_sense(cmd, SENSE_CODE(NO_SENSE));
+        cmd->frame->header.scsi_status = BUSY;
+        s->event_count++;
+        return MFI_STAT_SCSI_DONE_WITH_ERROR;
+    }
+    len = megasas_enqueue_req(cmd, is_write);
+    if (len > 0) {
+        if (is_write) {
+            trace_megasas_io_write_start(cmd->index, lba_start, lba_count, len);
+        } else {
+            trace_megasas_io_read_start(cmd->index, lba_start, lba_count, len);
+        }
+    }
+    return MFI_STAT_INVALID_STATUS;
+}
+
+static int megasas_finish_internal_command(MegasasCmd *cmd,
+                                           SCSIRequest *req, size_t resid)
+{
+    int retval = MFI_STAT_INVALID_CMD;
+
+    if (cmd->frame->header.frame_cmd == MFI_CMD_DCMD) {
+        cmd->iov_size -= resid;
+        retval = megasas_finish_internal_dcmd(cmd, req);
+    }
+    return retval;
+}
+
+static QEMUSGList *megasas_get_sg_list(SCSIRequest *req)
+{
+    MegasasCmd *cmd = req->hba_private;
+
+    if (cmd->frame->header.frame_cmd == MFI_CMD_DCMD) {
+        return NULL;
+    } else {
+        return &cmd->qsg;
+    }
+}
+
+static void megasas_xfer_complete(SCSIRequest *req, uint32_t len)
+{
+    MegasasCmd *cmd = req->hba_private;
+    uint8_t *buf;
+    uint32_t opcode;
+
+    trace_megasas_io_complete(cmd->index, len);
+
+    if (cmd->frame->header.frame_cmd != MFI_CMD_DCMD) {
+        scsi_req_continue(req);
+        return;
+    }
+
+    buf = scsi_req_get_buf(req);
+    opcode = le32_to_cpu(cmd->frame->dcmd.opcode);
+    if (opcode == MFI_DCMD_PD_GET_INFO && cmd->iov_buf) {
+        struct mfi_pd_info *info = cmd->iov_buf;
+
+        if (info->inquiry_data[0] == 0x7f) {
+            memset(info->inquiry_data, 0, sizeof(info->inquiry_data));
+            memcpy(info->inquiry_data, buf, len);
+        } else if (info->vpd_page83[0] == 0x7f) {
+            memset(info->vpd_page83, 0, sizeof(info->vpd_page83));
+            memcpy(info->vpd_page83, buf, len);
+        }
+        scsi_req_continue(req);
+    } else if (opcode == MFI_DCMD_LD_GET_INFO) {
+        struct mfi_ld_info *info = cmd->iov_buf;
+
+        if (cmd->iov_buf) {
+            memcpy(info->vpd_page83, buf, sizeof(info->vpd_page83));
+            scsi_req_continue(req);
+        }
+    }
+}
+
+static void megasas_command_complete(SCSIRequest *req, uint32_t status,
+                                     size_t resid)
+{
+    MegasasCmd *cmd = req->hba_private;
+    uint8_t cmd_status = MFI_STAT_OK;
+
+    trace_megasas_command_complete(cmd->index, status, resid);
+
+    if (cmd->req != req) {
+        /*
+         * Internal command complete
+         */
+        cmd_status = megasas_finish_internal_command(cmd, req, resid);
+        if (cmd_status == MFI_STAT_INVALID_STATUS) {
+            return;
+        }
+    } else {
+        req->status = status;
+        trace_megasas_scsi_complete(cmd->index, req->status,
+                                    cmd->iov_size, req->cmd.xfer);
+        if (req->status != GOOD) {
+            cmd_status = MFI_STAT_SCSI_DONE_WITH_ERROR;
+        }
+        if (req->status == CHECK_CONDITION) {
+            megasas_copy_sense(cmd);
+        }
+
+        megasas_unmap_sgl(cmd);
+        cmd->frame->header.scsi_status = req->status;
+        scsi_req_unref(cmd->req);
+        cmd->req = NULL;
+    }
+    cmd->frame->header.cmd_status = cmd_status;
+    megasas_unmap_frame(cmd->state, cmd);
+    megasas_complete_frame(cmd->state, cmd->context);
+}
+
+static void megasas_command_cancel(SCSIRequest *req)
+{
+    MegasasCmd *cmd = req->hba_private;
+
+    if (cmd) {
+        megasas_abort_command(cmd);
+    } else {
+        scsi_req_unref(req);
+    }
+}
+
+static int megasas_handle_abort(MegasasState *s, MegasasCmd *cmd)
+{
+    uint64_t abort_ctx = le64_to_cpu(cmd->frame->abort.abort_context);
+    hwaddr abort_addr, addr_hi, addr_lo;
+    MegasasCmd *abort_cmd;
+
+    addr_hi = le32_to_cpu(cmd->frame->abort.abort_mfi_addr_hi);
+    addr_lo = le32_to_cpu(cmd->frame->abort.abort_mfi_addr_lo);
+    abort_addr = ((uint64_t)addr_hi << 32) | addr_lo;
+
+    abort_cmd = megasas_lookup_frame(s, abort_addr);
+    if (!abort_cmd) {
+        trace_megasas_abort_no_cmd(cmd->index, abort_ctx);
+        s->event_count++;
+        return MFI_STAT_OK;
+    }
+    if (!megasas_use_queue64(s)) {
+        abort_ctx &= (uint64_t)0xFFFFFFFF;
+    }
+    if (abort_cmd->context != abort_ctx) {
+        trace_megasas_abort_invalid_context(cmd->index, abort_cmd->index,
+                                            abort_cmd->context);
+        s->event_count++;
+        return MFI_STAT_ABORT_NOT_POSSIBLE;
+    }
+    trace_megasas_abort_frame(cmd->index, abort_cmd->index);
+    megasas_abort_command(abort_cmd);
+    if (!s->event_cmd || abort_cmd != s->event_cmd) {
+        s->event_cmd = NULL;
+    }
+    s->event_count++;
+    return MFI_STAT_OK;
+}
+
+static void megasas_handle_frame(MegasasState *s, uint64_t frame_addr,
+                                 uint32_t frame_count)
+{
+    uint8_t frame_status = MFI_STAT_INVALID_CMD;
+    uint64_t frame_context;
+    MegasasCmd *cmd;
+
+    /*
+     * Always read 64bit context, top bits will be
+     * masked out if required in megasas_enqueue_frame()
+     */
+    frame_context = megasas_frame_get_context(s, frame_addr);
+
+    cmd = megasas_enqueue_frame(s, frame_addr, frame_context, frame_count);
+    if (!cmd) {
+        /* reply queue full */
+        trace_megasas_frame_busy(frame_addr);
+        megasas_frame_set_scsi_status(s, frame_addr, BUSY);
+        megasas_frame_set_cmd_status(s, frame_addr, MFI_STAT_SCSI_DONE_WITH_ERROR);
+        megasas_complete_frame(s, frame_context);
+        s->event_count++;
+        return;
+    }
+    switch (cmd->frame->header.frame_cmd) {
+    case MFI_CMD_INIT:
+        frame_status = megasas_init_firmware(s, cmd);
+        break;
+    case MFI_CMD_DCMD:
+        frame_status = megasas_handle_dcmd(s, cmd);
+        break;
+    case MFI_CMD_ABORT:
+        frame_status = megasas_handle_abort(s, cmd);
+        break;
+    case MFI_CMD_PD_SCSI_IO:
+        frame_status = megasas_handle_scsi(s, cmd, 0);
+        break;
+    case MFI_CMD_LD_SCSI_IO:
+        frame_status = megasas_handle_scsi(s, cmd, 1);
+        break;
+    case MFI_CMD_LD_READ:
+    case MFI_CMD_LD_WRITE:
+        frame_status = megasas_handle_io(s, cmd);
+        break;
+    default:
+        trace_megasas_unhandled_frame_cmd(cmd->index,
+                                          cmd->frame->header.frame_cmd);
+        s->event_count++;
+        break;
+    }
+    if (frame_status != MFI_STAT_INVALID_STATUS) {
+        if (cmd->frame) {
+            cmd->frame->header.cmd_status = frame_status;
+        } else {
+            megasas_frame_set_cmd_status(s, frame_addr, frame_status);
+        }
+        megasas_unmap_frame(s, cmd);
+        megasas_complete_frame(s, cmd->context);
+    }
+}
+
+static uint64_t megasas_mmio_read(void *opaque, hwaddr addr,
+                                  unsigned size)
+{
+    MegasasState *s = opaque;
+    PCIDevice *pci_dev = PCI_DEVICE(s);
+    MegasasBaseClass *base_class = MEGASAS_DEVICE_GET_CLASS(s);
+    uint32_t retval = 0;
+
+    switch (addr) {
+    case MFI_IDB:
+        retval = 0;
+        trace_megasas_mmio_readl("MFI_IDB", retval);
+        break;
+    case MFI_OMSG0:
+    case MFI_OSP0:
+        retval = (msix_present(pci_dev) ? MFI_FWSTATE_MSIX_SUPPORTED : 0) |
+            (s->fw_state & MFI_FWSTATE_MASK) |
+            ((s->fw_sge & 0xff) << 16) |
+            (s->fw_cmds & 0xFFFF);
+        trace_megasas_mmio_readl(addr == MFI_OMSG0 ? "MFI_OMSG0" : "MFI_OSP0",
+                                 retval);
+        break;
+    case MFI_OSTS:
+        if (megasas_intr_enabled(s) && s->doorbell) {
+            retval = base_class->osts;
+        }
+        trace_megasas_mmio_readl("MFI_OSTS", retval);
+        break;
+    case MFI_OMSK:
+        retval = s->intr_mask;
+        trace_megasas_mmio_readl("MFI_OMSK", retval);
+        break;
+    case MFI_ODCR0:
+        retval = s->doorbell ? 1 : 0;
+        trace_megasas_mmio_readl("MFI_ODCR0", retval);
+        break;
+    case MFI_DIAG:
+        retval = s->diag;
+        trace_megasas_mmio_readl("MFI_DIAG", retval);
+        break;
+    case MFI_OSP1:
+        retval = 15;
+        trace_megasas_mmio_readl("MFI_OSP1", retval);
+        break;
+    default:
+        trace_megasas_mmio_invalid_readl(addr);
+        break;
+    }
+    return retval;
+}
+
+static int adp_reset_seq[] = {0x00, 0x04, 0x0b, 0x02, 0x07, 0x0d};
+
+static void megasas_mmio_write(void *opaque, hwaddr addr,
+                               uint64_t val, unsigned size)
+{
+    MegasasState *s = opaque;
+    PCIDevice *pci_dev = PCI_DEVICE(s);
+    uint64_t frame_addr;
+    uint32_t frame_count;
+    int i;
+
+    switch (addr) {
+    case MFI_IDB:
+        trace_megasas_mmio_writel("MFI_IDB", val);
+        if (val & MFI_FWINIT_ABORT) {
+            /* Abort all pending cmds */
+            for (i = 0; i < s->fw_cmds; i++) {
+                megasas_abort_command(&s->frames[i]);
+            }
+        }
+        if (val & MFI_FWINIT_READY) {
+            /* move to FW READY */
+            megasas_soft_reset(s);
+        }
+        if (val & MFI_FWINIT_MFIMODE) {
+            /* discard MFIs */
+        }
+        if (val & MFI_FWINIT_STOP_ADP) {
+            /* Terminal error, stop processing */
+            s->fw_state = MFI_FWSTATE_FAULT;
+        }
+        break;
+    case MFI_OMSK:
+        trace_megasas_mmio_writel("MFI_OMSK", val);
+        s->intr_mask = val;
+        if (!megasas_intr_enabled(s) &&
+            !msi_enabled(pci_dev) &&
+            !msix_enabled(pci_dev)) {
+            trace_megasas_irq_lower();
+            pci_irq_deassert(pci_dev);
+        }
+        if (megasas_intr_enabled(s)) {
+            if (msix_enabled(pci_dev)) {
+                trace_megasas_msix_enabled(0);
+            } else if (msi_enabled(pci_dev)) {
+                trace_megasas_msi_enabled(0);
+            } else {
+                trace_megasas_intr_enabled();
+            }
+        } else {
+            trace_megasas_intr_disabled();
+            megasas_soft_reset(s);
+        }
+        break;
+    case MFI_ODCR0:
+        trace_megasas_mmio_writel("MFI_ODCR0", val);
+        s->doorbell = 0;
+        if (megasas_intr_enabled(s)) {
+            if (!msix_enabled(pci_dev) && !msi_enabled(pci_dev)) {
+                trace_megasas_irq_lower();
+                pci_irq_deassert(pci_dev);
+            }
+        }
+        break;
+    case MFI_IQPH:
+        trace_megasas_mmio_writel("MFI_IQPH", val);
+        /* Received high 32 bits of a 64 bit MFI frame address */
+        s->frame_hi = val;
+        break;
+    case MFI_IQPL:
+        trace_megasas_mmio_writel("MFI_IQPL", val);
+        /* Received low 32 bits of a 64 bit MFI frame address */
+        /* Fallthrough */
+    case MFI_IQP:
+        if (addr == MFI_IQP) {
+            trace_megasas_mmio_writel("MFI_IQP", val);
+            /* Received 64 bit MFI frame address */
+            s->frame_hi = 0;
+        }
+        frame_addr = (val & ~0x1F);
+        /* Add possible 64 bit offset */
+        frame_addr |= ((uint64_t)s->frame_hi << 32);
+        s->frame_hi = 0;
+        frame_count = (val >> 1) & 0xF;
+        megasas_handle_frame(s, frame_addr, frame_count);
+        break;
+    case MFI_SEQ:
+        trace_megasas_mmio_writel("MFI_SEQ", val);
+        /* Magic sequence to start ADP reset */
+        if (adp_reset_seq[s->adp_reset] == val) {
+            s->adp_reset++;
+        } else {
+            s->adp_reset = 0;
+            s->diag = 0;
+        }
+        if (s->adp_reset == 6) {
+            s->diag = MFI_DIAG_WRITE_ENABLE;
+        }
+        break;
+    case MFI_DIAG:
+        trace_megasas_mmio_writel("MFI_DIAG", val);
+        /* ADP reset */
+        if ((s->diag & MFI_DIAG_WRITE_ENABLE) &&
+            (val & MFI_DIAG_RESET_ADP)) {
+            s->diag |= MFI_DIAG_RESET_ADP;
+            megasas_soft_reset(s);
+            s->adp_reset = 0;
+            s->diag = 0;
+        }
+        break;
+    default:
+        trace_megasas_mmio_invalid_writel(addr, val);
+        break;
+    }
+}
+
+static const MemoryRegionOps megasas_mmio_ops = {
+    .read = megasas_mmio_read,
+    .write = megasas_mmio_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .impl = {
+        .min_access_size = 8,
+        .max_access_size = 8,
+    }
+};
+
+static uint64_t megasas_port_read(void *opaque, hwaddr addr,
+                                  unsigned size)
+{
+    return megasas_mmio_read(opaque, addr & 0xff, size);
+}
+
+static void megasas_port_write(void *opaque, hwaddr addr,
+                               uint64_t val, unsigned size)
+{
+    megasas_mmio_write(opaque, addr & 0xff, val, size);
+}
+
+static const MemoryRegionOps megasas_port_ops = {
+    .read = megasas_port_read,
+    .write = megasas_port_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    }
+};
+
+static uint64_t megasas_queue_read(void *opaque, hwaddr addr,
+                                   unsigned size)
+{
+    return 0;
+}
+
+static void megasas_queue_write(void *opaque, hwaddr addr,
+                               uint64_t val, unsigned size)
+{
+    return;
+}
+
+static const MemoryRegionOps megasas_queue_ops = {
+    .read = megasas_queue_read,
+    .write = megasas_queue_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .impl = {
+        .min_access_size = 8,
+        .max_access_size = 8,
+    }
+};
+
+static void megasas_soft_reset(MegasasState *s)
+{
+    int i;
+    MegasasCmd *cmd;
+
+    trace_megasas_reset(s->fw_state);
+    for (i = 0; i < s->fw_cmds; i++) {
+        cmd = &s->frames[i];
+        megasas_abort_command(cmd);
+    }
+    if (s->fw_state == MFI_FWSTATE_READY) {
+        BusChild *kid;
+
+        /*
+         * The EFI firmware doesn't handle UA,
+         * so we need to clear the Power On/Reset UA
+         * after the initial reset.
+         */
+        QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+            SCSIDevice *sdev = DO_UPCAST(SCSIDevice, qdev, kid->child);
+
+            sdev->unit_attention = SENSE_CODE(NO_SENSE);
+            scsi_device_unit_attention_reported(sdev);
+        }
+    }
+    megasas_reset_frames(s);
+    s->reply_queue_len = s->fw_cmds;
+    s->reply_queue_pa = 0;
+    s->consumer_pa = 0;
+    s->producer_pa = 0;
+    s->fw_state = MFI_FWSTATE_READY;
+    s->doorbell = 0;
+    s->intr_mask = MEGASAS_INTR_DISABLED_MASK;
+    s->frame_hi = 0;
+    s->flags &= ~MEGASAS_MASK_USE_QUEUE64;
+    s->event_count++;
+    s->boot_event = s->event_count;
+}
+
+static void megasas_scsi_reset(DeviceState *dev)
+{
+    MegasasState *s = MEGASAS(dev);
+
+    megasas_soft_reset(s);
+}
+
+static const VMStateDescription vmstate_megasas_gen1 = {
+    .name = "megasas",
+    .version_id = 0,
+    .minimum_version_id = 0,
+    .fields = (VMStateField[]) {
+        VMSTATE_PCI_DEVICE(parent_obj, MegasasState),
+        VMSTATE_MSIX(parent_obj, MegasasState),
+
+        VMSTATE_INT32(fw_state, MegasasState),
+        VMSTATE_INT32(intr_mask, MegasasState),
+        VMSTATE_INT32(doorbell, MegasasState),
+        VMSTATE_UINT64(reply_queue_pa, MegasasState),
+        VMSTATE_UINT64(consumer_pa, MegasasState),
+        VMSTATE_UINT64(producer_pa, MegasasState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static const VMStateDescription vmstate_megasas_gen2 = {
+    .name = "megasas-gen2",
+    .version_id = 0,
+    .minimum_version_id = 0,
+    .minimum_version_id_old = 0,
+    .fields      = (VMStateField[]) {
+        VMSTATE_PCIE_DEVICE(parent_obj, MegasasState),
+        VMSTATE_MSIX(parent_obj, MegasasState),
+
+        VMSTATE_INT32(fw_state, MegasasState),
+        VMSTATE_INT32(intr_mask, MegasasState),
+        VMSTATE_INT32(doorbell, MegasasState),
+        VMSTATE_UINT64(reply_queue_pa, MegasasState),
+        VMSTATE_UINT64(consumer_pa, MegasasState),
+        VMSTATE_UINT64(producer_pa, MegasasState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static void megasas_scsi_uninit(PCIDevice *d)
+{
+    MegasasState *s = MEGASAS(d);
+
+    if (megasas_use_msix(s)) {
+        msix_uninit(d, &s->mmio_io, &s->mmio_io);
+    }
+    if (megasas_use_msi(s)) {
+        msi_uninit(d);
+    }
+}
+
+static const struct SCSIBusInfo megasas_scsi_info = {
+    .tcq = true,
+    .max_target = MFI_MAX_LD,
+    .max_lun = 255,
+
+    .transfer_data = megasas_xfer_complete,
+    .get_sg_list = megasas_get_sg_list,
+    .complete = megasas_command_complete,
+    .cancel = megasas_command_cancel,
+};
+
+static void megasas_scsi_realize(PCIDevice *dev, Error **errp)
+{
+    DeviceState *d = DEVICE(dev);
+    MegasasState *s = MEGASAS(dev);
+    MegasasBaseClass *b = MEGASAS_DEVICE_GET_CLASS(s);
+    uint8_t *pci_conf;
+    int i, bar_type;
+
+    pci_conf = dev->config;
+
+    /* PCI latency timer = 0 */
+    pci_conf[PCI_LATENCY_TIMER] = 0;
+    /* Interrupt pin 1 */
+    pci_conf[PCI_INTERRUPT_PIN] = 0x01;
+
+    memory_region_init_io(&s->mmio_io, OBJECT(s), &megasas_mmio_ops, s,
+                          "megasas-mmio", 0x4000);
+    memory_region_init_io(&s->port_io, OBJECT(s), &megasas_port_ops, s,
+                          "megasas-io", 256);
+    memory_region_init_io(&s->queue_io, OBJECT(s), &megasas_queue_ops, s,
+                          "megasas-queue", 0x40000);
+
+    if (megasas_use_msi(s) &&
+        msi_init(dev, 0x50, 1, true, false)) {
+        s->flags &= ~MEGASAS_MASK_USE_MSI;
+    }
+    if (megasas_use_msix(s) &&
+        msix_init(dev, 15, &s->mmio_io, b->mmio_bar, 0x2000,
+                  &s->mmio_io, b->mmio_bar, 0x3800, 0x68)) {
+        s->flags &= ~MEGASAS_MASK_USE_MSIX;
+    }
+    if (pci_is_express(dev)) {
+        pcie_endpoint_cap_init(dev, 0xa0);
+    }
+
+    bar_type = PCI_BASE_ADDRESS_SPACE_MEMORY | PCI_BASE_ADDRESS_MEM_TYPE_64;
+    pci_register_bar(dev, b->ioport_bar,
+                     PCI_BASE_ADDRESS_SPACE_IO, &s->port_io);
+    pci_register_bar(dev, b->mmio_bar, bar_type, &s->mmio_io);
+    pci_register_bar(dev, 3, bar_type, &s->queue_io);
+
+    if (megasas_use_msix(s)) {
+        msix_vector_use(dev, 0);
+    }
+
+    s->fw_state = MFI_FWSTATE_READY;
+    if (!s->sas_addr) {
+        s->sas_addr = ((NAA_LOCALLY_ASSIGNED_ID << 24) |
+                       IEEE_COMPANY_LOCALLY_ASSIGNED) << 36;
+        s->sas_addr |= (pci_bus_num(dev->bus) << 16);
+        s->sas_addr |= (PCI_SLOT(dev->devfn) << 8);
+        s->sas_addr |= PCI_FUNC(dev->devfn);
+    }
+    if (!s->hba_serial) {
+        s->hba_serial = g_strdup(MEGASAS_HBA_SERIAL);
+    }
+    if (s->fw_sge >= MEGASAS_MAX_SGE - MFI_PASS_FRAME_SIZE) {
+        s->fw_sge = MEGASAS_MAX_SGE - MFI_PASS_FRAME_SIZE;
+    } else if (s->fw_sge >= 128 - MFI_PASS_FRAME_SIZE) {
+        s->fw_sge = 128 - MFI_PASS_FRAME_SIZE;
+    } else {
+        s->fw_sge = 64 - MFI_PASS_FRAME_SIZE;
+    }
+    if (s->fw_cmds > MEGASAS_MAX_FRAMES) {
+        s->fw_cmds = MEGASAS_MAX_FRAMES;
+    }
+    trace_megasas_init(s->fw_sge, s->fw_cmds,
+                       megasas_is_jbod(s) ? "jbod" : "raid");
+
+    if (megasas_is_jbod(s)) {
+        s->fw_luns = MFI_MAX_SYS_PDS;
+    } else {
+        s->fw_luns = MFI_MAX_LD;
+    }
+    s->producer_pa = 0;
+    s->consumer_pa = 0;
+    for (i = 0; i < s->fw_cmds; i++) {
+        s->frames[i].index = i;
+        s->frames[i].context = -1;
+        s->frames[i].pa = 0;
+        s->frames[i].state = s;
+    }
+
+    scsi_bus_new(&s->bus, sizeof(s->bus), DEVICE(dev),
+                 &megasas_scsi_info, NULL);
+    if (!d->hotplugged) {
+        scsi_bus_legacy_handle_cmdline(&s->bus, errp);
+    }
+}
+
+static Property megasas_properties_gen1[] = {
+    DEFINE_PROP_UINT32("max_sge", MegasasState, fw_sge,
+                       MEGASAS_DEFAULT_SGE),
+    DEFINE_PROP_UINT32("max_cmds", MegasasState, fw_cmds,
+                       MEGASAS_DEFAULT_FRAMES),
+    DEFINE_PROP_STRING("hba_serial", MegasasState, hba_serial),
+    DEFINE_PROP_UINT64("sas_address", MegasasState, sas_addr, 0),
+    DEFINE_PROP_BIT("use_msi", MegasasState, flags,
+                    MEGASAS_FLAG_USE_MSI, false),
+    DEFINE_PROP_BIT("use_msix", MegasasState, flags,
+                    MEGASAS_FLAG_USE_MSIX, false),
+    DEFINE_PROP_BIT("use_jbod", MegasasState, flags,
+                    MEGASAS_FLAG_USE_JBOD, false),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static Property megasas_properties_gen2[] = {
+    DEFINE_PROP_UINT32("max_sge", MegasasState, fw_sge,
+                       MEGASAS_DEFAULT_SGE),
+    DEFINE_PROP_UINT32("max_cmds", MegasasState, fw_cmds,
+                       MEGASAS_GEN2_DEFAULT_FRAMES),
+    DEFINE_PROP_STRING("hba_serial", MegasasState, hba_serial),
+    DEFINE_PROP_UINT64("sas_address", MegasasState, sas_addr, 0),
+    DEFINE_PROP_BIT("use_msi", MegasasState, flags,
+                    MEGASAS_FLAG_USE_MSI, true),
+    DEFINE_PROP_BIT("use_msix", MegasasState, flags,
+                    MEGASAS_FLAG_USE_MSIX, true),
+    DEFINE_PROP_BIT("use_jbod", MegasasState, flags,
+                    MEGASAS_FLAG_USE_JBOD, false),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+typedef struct MegasasInfo {
+    const char *name;
+    const char *desc;
+    const char *product_name;
+    const char *product_version;
+    uint16_t device_id;
+    uint16_t subsystem_id;
+    int ioport_bar;
+    int mmio_bar;
+    bool is_express;
+    int osts;
+    const VMStateDescription *vmsd;
+    Property *props;
+} MegasasInfo;
+
+static struct MegasasInfo megasas_devices[] = {
+    {
+        .name = TYPE_MEGASAS_GEN1,
+        .desc = "LSI MegaRAID SAS 1078",
+        .product_name = "LSI MegaRAID SAS 8708EM2",
+        .product_version = MEGASAS_VERSION_GEN1,
+        .device_id = PCI_DEVICE_ID_LSI_SAS1078,
+        .subsystem_id = 0x1013,
+        .ioport_bar = 2,
+        .mmio_bar = 0,
+        .osts = MFI_1078_RM | 1,
+        .is_express = false,
+        .vmsd = &vmstate_megasas_gen1,
+        .props = megasas_properties_gen1,
+    },{
+        .name = TYPE_MEGASAS_GEN2,
+        .desc = "LSI MegaRAID SAS 2108",
+        .product_name = "LSI MegaRAID SAS 9260-8i",
+        .product_version = MEGASAS_VERSION_GEN2,
+        .device_id = PCI_DEVICE_ID_LSI_SAS0079,
+        .subsystem_id = 0x9261,
+        .ioport_bar = 0,
+        .mmio_bar = 1,
+        .osts = MFI_GEN2_RM,
+        .is_express = true,
+        .vmsd = &vmstate_megasas_gen2,
+        .props = megasas_properties_gen2,
+    }
+};
+
+static void megasas_class_init(ObjectClass *oc, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(oc);
+    PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
+    MegasasBaseClass *e = MEGASAS_DEVICE_CLASS(oc);
+    const MegasasInfo *info = data;
+
+    pc->realize = megasas_scsi_realize;
+    pc->exit = megasas_scsi_uninit;
+    pc->vendor_id = PCI_VENDOR_ID_LSI_LOGIC;
+    pc->device_id = info->device_id;
+    pc->subsystem_vendor_id = PCI_VENDOR_ID_LSI_LOGIC;
+    pc->subsystem_id = info->subsystem_id;
+    pc->class_id = PCI_CLASS_STORAGE_RAID;
+    pc->is_express = info->is_express;
+    e->mmio_bar = info->mmio_bar;
+    e->ioport_bar = info->ioport_bar;
+    e->osts = info->osts;
+    e->product_name = info->product_name;
+    e->product_version = info->product_version;
+    dc->props = info->props;
+    dc->reset = megasas_scsi_reset;
+    dc->vmsd = info->vmsd;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+    dc->desc = info->desc;
+}
+
+static const TypeInfo megasas_info = {
+    .name  = TYPE_MEGASAS_BASE,
+    .parent = TYPE_PCI_DEVICE,
+    .instance_size = sizeof(MegasasState),
+    .class_size = sizeof(MegasasBaseClass),
+    .abstract = true,
+};
+
+static void megasas_register_types(void)
+{
+    int i;
+
+    type_register_static(&megasas_info);
+    for (i = 0; i < ARRAY_SIZE(megasas_devices); i++) {
+        const MegasasInfo *info = &megasas_devices[i];
+        TypeInfo type_info = {};
+
+        type_info.name = info->name;
+        type_info.parent = TYPE_MEGASAS_BASE;
+        type_info.class_data = (void *)info;
+        type_info.class_init = megasas_class_init;
+
+        type_register(&type_info);
+    }
+}
+
+type_init(megasas_register_types)
diff --git a/qemu/hw/scsi/mfi.h b/qemu/hw/scsi/mfi.h
new file mode 100644
index 000000000..29d41775d
--- /dev/null
+++ b/qemu/hw/scsi/mfi.h
@@ -0,0 +1,1272 @@
+/*
+ * NetBSD header file, copied from
+ * http://gitorious.org/freebsd/freebsd/blobs/HEAD/sys/dev/mfi/mfireg.h
+ */
+/*-
+ * Copyright (c) 2006 IronPort Systems
+ * Copyright (c) 2007 LSI Corp.
+ * Copyright (c) 2007 Rajesh Prabhakaran.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#ifndef MFI_REG_H
+#define MFI_REG_H
+
+/*
+ * MegaRAID SAS MFI firmware definitions
+ */
+
+/*
+ * Start with the register set.  All registers are 32 bits wide.
+ * The usual Intel IOP style setup.
+ */
+#define MFI_IMSG0 0x10    /* Inbound message 0 */
+#define MFI_IMSG1 0x14    /* Inbound message 1 */
+#define MFI_OMSG0 0x18    /* Outbound message 0 */
+#define MFI_OMSG1 0x1c    /* Outbound message 1 */
+#define MFI_IDB   0x20    /* Inbound doorbell */
+#define MFI_ISTS  0x24    /* Inbound interrupt status */
+#define MFI_IMSK  0x28    /* Inbound interrupt mask */
+#define MFI_ODB   0x2c    /* Outbound doorbell */
+#define MFI_OSTS  0x30    /* Outbound interrupt status */
+#define MFI_OMSK  0x34    /* Outbound interrupt mask */
+#define MFI_IQP   0x40    /* Inbound queue port */
+#define MFI_OQP   0x44    /* Outbound queue port */
+
+/*
+ * 1078 specific related register
+ */
+#define MFI_ODR0        0x9c            /* outbound doorbell register0 */
+#define MFI_ODCR0       0xa0            /* outbound doorbell clear register0  */
+#define MFI_OSP0        0xb0            /* outbound scratch pad0  */
+#define MFI_OSP1        0xb4            /* outbound scratch pad1  */
+#define MFI_IQPL        0xc0            /* Inbound queue port (low bytes)  */
+#define MFI_IQPH        0xc4            /* Inbound queue port (high bytes)  */
+#define MFI_DIAG        0xf8            /* Host diag */
+#define MFI_SEQ         0xfc            /* Sequencer offset */
+#define MFI_1078_EIM    0x80000004      /* 1078 enable intrrupt mask  */
+#define MFI_RMI         0x2             /* reply message interrupt  */
+#define MFI_1078_RM     0x80000000      /* reply 1078 message interrupt  */
+#define MFI_ODC         0x4             /* outbound doorbell change interrupt */
+
+/*
+ * gen2 specific changes
+ */
+#define MFI_GEN2_EIM    0x00000005      /* gen2 enable interrupt mask */
+#define MFI_GEN2_RM     0x00000001      /* reply gen2 message interrupt */
+
+/*
+ * skinny specific changes
+ */
+#define MFI_SKINNY_IDB  0x00    /* Inbound doorbell is at 0x00 for skinny */
+#define MFI_SKINNY_RM   0x00000001      /* reply skinny message interrupt */
+
+/* Bits for MFI_OSTS */
+#define MFI_OSTS_INTR_VALID     0x00000002
+
+/*
+ * Firmware state values.  Found in OMSG0 during initialization.
+ */
+#define MFI_FWSTATE_MASK                0xf0000000
+#define MFI_FWSTATE_UNDEFINED           0x00000000
+#define MFI_FWSTATE_BB_INIT             0x10000000
+#define MFI_FWSTATE_FW_INIT             0x40000000
+#define MFI_FWSTATE_WAIT_HANDSHAKE      0x60000000
+#define MFI_FWSTATE_FW_INIT_2           0x70000000
+#define MFI_FWSTATE_DEVICE_SCAN         0x80000000
+#define MFI_FWSTATE_BOOT_MSG_PENDING    0x90000000
+#define MFI_FWSTATE_FLUSH_CACHE         0xa0000000
+#define MFI_FWSTATE_READY               0xb0000000
+#define MFI_FWSTATE_OPERATIONAL         0xc0000000
+#define MFI_FWSTATE_FAULT               0xf0000000
+#define MFI_FWSTATE_MAXSGL_MASK         0x00ff0000
+#define MFI_FWSTATE_MAXCMD_MASK         0x0000ffff
+#define MFI_FWSTATE_MSIX_SUPPORTED      0x04000000
+#define MFI_FWSTATE_HOSTMEMREQD_MASK    0x08000000
+
+/*
+ * Control bits to drive the card to ready state.  These go into the IDB
+ * register.
+ */
+#define MFI_FWINIT_ABORT        0x00000001 /* Abort all pending commands */
+#define MFI_FWINIT_READY        0x00000002 /* Move from operational to ready */
+#define MFI_FWINIT_MFIMODE      0x00000004 /* unknown */
+#define MFI_FWINIT_CLEAR_HANDSHAKE 0x00000008 /* Respond to WAIT_HANDSHAKE */
+#define MFI_FWINIT_HOTPLUG      0x00000010
+#define MFI_FWINIT_STOP_ADP     0x00000020 /* Move to operational, stop */
+#define MFI_FWINIT_ADP_RESET    0x00000040 /* Reset ADP */
+
+/*
+ * Control bits for the DIAG register
+ */
+#define MFI_DIAG_WRITE_ENABLE 0x00000080
+#define MFI_DIAG_RESET_ADP    0x00000004
+
+/* MFI Commands */
+typedef enum {
+    MFI_CMD_INIT = 0x00,
+    MFI_CMD_LD_READ,
+    MFI_CMD_LD_WRITE,
+    MFI_CMD_LD_SCSI_IO,
+    MFI_CMD_PD_SCSI_IO,
+    MFI_CMD_DCMD,
+    MFI_CMD_ABORT,
+    MFI_CMD_SMP,
+    MFI_CMD_STP
+} mfi_cmd_t;
+
+/* Direct commands */
+typedef enum {
+    MFI_DCMD_CTRL_MFI_HOST_MEM_ALLOC =  0x0100e100,
+    MFI_DCMD_CTRL_GET_INFO =            0x01010000,
+    MFI_DCMD_CTRL_GET_PROPERTIES =      0x01020100,
+    MFI_DCMD_CTRL_SET_PROPERTIES =      0x01020200,
+    MFI_DCMD_CTRL_ALARM =               0x01030000,
+    MFI_DCMD_CTRL_ALARM_GET =           0x01030100,
+    MFI_DCMD_CTRL_ALARM_ENABLE =        0x01030200,
+    MFI_DCMD_CTRL_ALARM_DISABLE =       0x01030300,
+    MFI_DCMD_CTRL_ALARM_SILENCE =       0x01030400,
+    MFI_DCMD_CTRL_ALARM_TEST =          0x01030500,
+    MFI_DCMD_CTRL_EVENT_GETINFO =       0x01040100,
+    MFI_DCMD_CTRL_EVENT_CLEAR =         0x01040200,
+    MFI_DCMD_CTRL_EVENT_GET =           0x01040300,
+    MFI_DCMD_CTRL_EVENT_COUNT =         0x01040400,
+    MFI_DCMD_CTRL_EVENT_WAIT =          0x01040500,
+    MFI_DCMD_CTRL_SHUTDOWN =            0x01050000,
+    MFI_DCMD_HIBERNATE_STANDBY =        0x01060000,
+    MFI_DCMD_CTRL_GET_TIME =            0x01080101,
+    MFI_DCMD_CTRL_SET_TIME =            0x01080102,
+    MFI_DCMD_CTRL_BIOS_DATA_GET =       0x010c0100,
+    MFI_DCMD_CTRL_BIOS_DATA_SET =       0x010c0200,
+    MFI_DCMD_CTRL_FACTORY_DEFAULTS =    0x010d0000,
+    MFI_DCMD_CTRL_MFC_DEFAULTS_GET =    0x010e0201,
+    MFI_DCMD_CTRL_MFC_DEFAULTS_SET =    0x010e0202,
+    MFI_DCMD_CTRL_CACHE_FLUSH =         0x01101000,
+    MFI_DCMD_PD_GET_LIST =              0x02010000,
+    MFI_DCMD_PD_LIST_QUERY =            0x02010100,
+    MFI_DCMD_PD_GET_INFO =              0x02020000,
+    MFI_DCMD_PD_STATE_SET =             0x02030100,
+    MFI_DCMD_PD_REBUILD =               0x02040100,
+    MFI_DCMD_PD_BLINK =                 0x02070100,
+    MFI_DCMD_PD_UNBLINK =               0x02070200,
+    MFI_DCMD_LD_GET_LIST =              0x03010000,
+    MFI_DCMD_LD_LIST_QUERY =            0x03010100,
+    MFI_DCMD_LD_GET_INFO =              0x03020000,
+    MFI_DCMD_LD_GET_PROP =              0x03030000,
+    MFI_DCMD_LD_SET_PROP =              0x03040000,
+    MFI_DCMD_LD_DELETE =                0x03090000,
+    MFI_DCMD_CFG_READ =                 0x04010000,
+    MFI_DCMD_CFG_ADD =                  0x04020000,
+    MFI_DCMD_CFG_CLEAR =                0x04030000,
+    MFI_DCMD_CFG_FOREIGN_READ =         0x04060100,
+    MFI_DCMD_CFG_FOREIGN_IMPORT =       0x04060400,
+    MFI_DCMD_BBU_STATUS =               0x05010000,
+    MFI_DCMD_BBU_CAPACITY_INFO =        0x05020000,
+    MFI_DCMD_BBU_DESIGN_INFO =          0x05030000,
+    MFI_DCMD_BBU_PROP_GET =             0x05050100,
+    MFI_DCMD_CLUSTER =                  0x08000000,
+    MFI_DCMD_CLUSTER_RESET_ALL =        0x08010100,
+    MFI_DCMD_CLUSTER_RESET_LD =         0x08010200
+} mfi_dcmd_t;
+
+/* Modifiers for MFI_DCMD_CTRL_FLUSHCACHE */
+#define MFI_FLUSHCACHE_CTRL     0x01
+#define MFI_FLUSHCACHE_DISK     0x02
+
+/* Modifiers for MFI_DCMD_CTRL_SHUTDOWN */
+#define MFI_SHUTDOWN_SPINDOWN   0x01
+
+/*
+ * MFI Frame flags
+ */
+typedef enum {
+    MFI_FRAME_DONT_POST_IN_REPLY_QUEUE =        0x0001,
+    MFI_FRAME_SGL64 =                           0x0002,
+    MFI_FRAME_SENSE64 =                         0x0004,
+    MFI_FRAME_DIR_WRITE =                       0x0008,
+    MFI_FRAME_DIR_READ =                        0x0010,
+    MFI_FRAME_IEEE_SGL =                        0x0020,
+} mfi_frame_flags;
+
+/* MFI Status codes */
+typedef enum {
+    MFI_STAT_OK =                       0x00,
+    MFI_STAT_INVALID_CMD,
+    MFI_STAT_INVALID_DCMD,
+    MFI_STAT_INVALID_PARAMETER,
+    MFI_STAT_INVALID_SEQUENCE_NUMBER,
+    MFI_STAT_ABORT_NOT_POSSIBLE,
+    MFI_STAT_APP_HOST_CODE_NOT_FOUND,
+    MFI_STAT_APP_IN_USE,
+    MFI_STAT_APP_NOT_INITIALIZED,
+    MFI_STAT_ARRAY_INDEX_INVALID,
+    MFI_STAT_ARRAY_ROW_NOT_EMPTY,
+    MFI_STAT_CONFIG_RESOURCE_CONFLICT,
+    MFI_STAT_DEVICE_NOT_FOUND,
+    MFI_STAT_DRIVE_TOO_SMALL,
+    MFI_STAT_FLASH_ALLOC_FAIL,
+    MFI_STAT_FLASH_BUSY,
+    MFI_STAT_FLASH_ERROR =              0x10,
+    MFI_STAT_FLASH_IMAGE_BAD,
+    MFI_STAT_FLASH_IMAGE_INCOMPLETE,
+    MFI_STAT_FLASH_NOT_OPEN,
+    MFI_STAT_FLASH_NOT_STARTED,
+    MFI_STAT_FLUSH_FAILED,
+    MFI_STAT_HOST_CODE_NOT_FOUNT,
+    MFI_STAT_LD_CC_IN_PROGRESS,
+    MFI_STAT_LD_INIT_IN_PROGRESS,
+    MFI_STAT_LD_LBA_OUT_OF_RANGE,
+    MFI_STAT_LD_MAX_CONFIGURED,
+    MFI_STAT_LD_NOT_OPTIMAL,
+    MFI_STAT_LD_RBLD_IN_PROGRESS,
+    MFI_STAT_LD_RECON_IN_PROGRESS,
+    MFI_STAT_LD_WRONG_RAID_LEVEL,
+    MFI_STAT_MAX_SPARES_EXCEEDED,
+    MFI_STAT_MEMORY_NOT_AVAILABLE =     0x20,
+    MFI_STAT_MFC_HW_ERROR,
+    MFI_STAT_NO_HW_PRESENT,
+    MFI_STAT_NOT_FOUND,
+    MFI_STAT_NOT_IN_ENCL,
+    MFI_STAT_PD_CLEAR_IN_PROGRESS,
+    MFI_STAT_PD_TYPE_WRONG,
+    MFI_STAT_PR_DISABLED,
+    MFI_STAT_ROW_INDEX_INVALID,
+    MFI_STAT_SAS_CONFIG_INVALID_ACTION,
+    MFI_STAT_SAS_CONFIG_INVALID_DATA,
+    MFI_STAT_SAS_CONFIG_INVALID_PAGE,
+    MFI_STAT_SAS_CONFIG_INVALID_TYPE,
+    MFI_STAT_SCSI_DONE_WITH_ERROR,
+    MFI_STAT_SCSI_IO_FAILED,
+    MFI_STAT_SCSI_RESERVATION_CONFLICT,
+    MFI_STAT_SHUTDOWN_FAILED =          0x30,
+    MFI_STAT_TIME_NOT_SET,
+    MFI_STAT_WRONG_STATE,
+    MFI_STAT_LD_OFFLINE,
+    MFI_STAT_PEER_NOTIFICATION_REJECTED,
+    MFI_STAT_PEER_NOTIFICATION_FAILED,
+    MFI_STAT_RESERVATION_IN_PROGRESS,
+    MFI_STAT_I2C_ERRORS_DETECTED,
+    MFI_STAT_PCI_ERRORS_DETECTED,
+    MFI_STAT_DIAG_FAILED,
+    MFI_STAT_BOOT_MSG_PENDING,
+    MFI_STAT_FOREIGN_CONFIG_INCOMPLETE,
+    MFI_STAT_INVALID_SGL,
+    MFI_STAT_UNSUPPORTED_HW,
+    MFI_STAT_CC_SCHEDULE_DISABLED,
+    MFI_STAT_PD_COPYBACK_IN_PROGRESS,
+    MFI_STAT_MULTIPLE_PDS_IN_ARRAY =    0x40,
+    MFI_STAT_FW_DOWNLOAD_ERROR,
+    MFI_STAT_FEATURE_SECURITY_NOT_ENABLED,
+    MFI_STAT_LOCK_KEY_ALREADY_EXISTS,
+    MFI_STAT_LOCK_KEY_BACKUP_NOT_ALLOWED,
+    MFI_STAT_LOCK_KEY_VERIFY_NOT_ALLOWED,
+    MFI_STAT_LOCK_KEY_VERIFY_FAILED,
+    MFI_STAT_LOCK_KEY_REKEY_NOT_ALLOWED,
+    MFI_STAT_LOCK_KEY_INVALID,
+    MFI_STAT_LOCK_KEY_ESCROW_INVALID,
+    MFI_STAT_LOCK_KEY_BACKUP_REQUIRED,
+    MFI_STAT_SECURE_LD_EXISTS,
+    MFI_STAT_LD_SECURE_NOT_ALLOWED,
+    MFI_STAT_REPROVISION_NOT_ALLOWED,
+    MFI_STAT_PD_SECURITY_TYPE_WRONG,
+    MFI_STAT_LD_ENCRYPTION_TYPE_INVALID,
+    MFI_STAT_CONFIG_FDE_NON_FDE_MIX_NOT_ALLOWED = 0x50,
+    MFI_STAT_CONFIG_LD_ENCRYPTION_TYPE_MIX_NOT_ALLOWED,
+    MFI_STAT_SECRET_KEY_NOT_ALLOWED,
+    MFI_STAT_PD_HW_ERRORS_DETECTED,
+    MFI_STAT_LD_CACHE_PINNED,
+    MFI_STAT_POWER_STATE_SET_IN_PROGRESS,
+    MFI_STAT_POWER_STATE_SET_BUSY,
+    MFI_STAT_POWER_STATE_WRONG,
+    MFI_STAT_PR_NO_AVAILABLE_PD_FOUND,
+    MFI_STAT_CTRL_RESET_REQUIRED,
+    MFI_STAT_LOCK_KEY_EKM_NO_BOOT_AGENT,
+    MFI_STAT_SNAP_NO_SPACE,
+    MFI_STAT_SNAP_PARTIAL_FAILURE,
+    MFI_STAT_UPGRADE_KEY_INCOMPATIBLE,
+    MFI_STAT_PFK_INCOMPATIBLE,
+    MFI_STAT_PD_MAX_UNCONFIGURED,
+    MFI_STAT_IO_METRICS_DISABLED =      0x60,
+    MFI_STAT_AEC_NOT_STOPPED,
+    MFI_STAT_PI_TYPE_WRONG,
+    MFI_STAT_LD_PD_PI_INCOMPATIBLE,
+    MFI_STAT_PI_NOT_ENABLED,
+    MFI_STAT_LD_BLOCK_SIZE_MISMATCH,
+    MFI_STAT_INVALID_STATUS =           0xFF
+} mfi_status_t;
+
+/* Event classes */
+typedef enum {
+    MFI_EVT_CLASS_DEBUG =      -2,
+    MFI_EVT_CLASS_PROGRESS =   -1,
+    MFI_EVT_CLASS_INFO =        0,
+    MFI_EVT_CLASS_WARNING =     1,
+    MFI_EVT_CLASS_CRITICAL =    2,
+    MFI_EVT_CLASS_FATAL =       3,
+    MFI_EVT_CLASS_DEAD =        4
+} mfi_evt_class_t;
+
+/* Event locales */
+typedef enum {
+    MFI_EVT_LOCALE_LD =         0x0001,
+    MFI_EVT_LOCALE_PD =         0x0002,
+    MFI_EVT_LOCALE_ENCL =       0x0004,
+    MFI_EVT_LOCALE_BBU =        0x0008,
+    MFI_EVT_LOCALE_SAS =        0x0010,
+    MFI_EVT_LOCALE_CTRL =       0x0020,
+    MFI_EVT_LOCALE_CONFIG =     0x0040,
+    MFI_EVT_LOCALE_CLUSTER =    0x0080,
+    MFI_EVT_LOCALE_ALL =        0xffff
+} mfi_evt_locale_t;
+
+/* Event args */
+typedef enum {
+    MR_EVT_ARGS_NONE =          0x00,
+    MR_EVT_ARGS_CDB_SENSE,
+    MR_EVT_ARGS_LD,
+    MR_EVT_ARGS_LD_COUNT,
+    MR_EVT_ARGS_LD_LBA,
+    MR_EVT_ARGS_LD_OWNER,
+    MR_EVT_ARGS_LD_LBA_PD_LBA,
+    MR_EVT_ARGS_LD_PROG,
+    MR_EVT_ARGS_LD_STATE,
+    MR_EVT_ARGS_LD_STRIP,
+    MR_EVT_ARGS_PD,
+    MR_EVT_ARGS_PD_ERR,
+    MR_EVT_ARGS_PD_LBA,
+    MR_EVT_ARGS_PD_LBA_LD,
+    MR_EVT_ARGS_PD_PROG,
+    MR_EVT_ARGS_PD_STATE,
+    MR_EVT_ARGS_PCI,
+    MR_EVT_ARGS_RATE,
+    MR_EVT_ARGS_STR,
+    MR_EVT_ARGS_TIME,
+    MR_EVT_ARGS_ECC,
+    MR_EVT_ARGS_LD_PROP,
+    MR_EVT_ARGS_PD_SPARE,
+    MR_EVT_ARGS_PD_INDEX,
+    MR_EVT_ARGS_DIAG_PASS,
+    MR_EVT_ARGS_DIAG_FAIL,
+    MR_EVT_ARGS_PD_LBA_LBA,
+    MR_EVT_ARGS_PORT_PHY,
+    MR_EVT_ARGS_PD_MISSING,
+    MR_EVT_ARGS_PD_ADDRESS,
+    MR_EVT_ARGS_BITMAP,
+    MR_EVT_ARGS_CONNECTOR,
+    MR_EVT_ARGS_PD_PD,
+    MR_EVT_ARGS_PD_FRU,
+    MR_EVT_ARGS_PD_PATHINFO,
+    MR_EVT_ARGS_PD_POWER_STATE,
+    MR_EVT_ARGS_GENERIC,
+} mfi_evt_args;
+
+/* Event codes */
+#define MR_EVT_CFG_CLEARED                          0x0004
+#define MR_EVT_CTRL_SHUTDOWN                        0x002a
+#define MR_EVT_LD_STATE_CHANGE                      0x0051
+#define MR_EVT_PD_INSERTED                          0x005b
+#define MR_EVT_PD_REMOVED                           0x0070
+#define MR_EVT_PD_STATE_CHANGED                     0x0072
+#define MR_EVT_LD_CREATED                           0x008a
+#define MR_EVT_LD_DELETED                           0x008b
+#define MR_EVT_FOREIGN_CFG_IMPORTED                 0x00db
+#define MR_EVT_LD_OFFLINE                           0x00fc
+#define MR_EVT_CTRL_HOST_BUS_SCAN_REQUESTED         0x0152
+
+typedef enum {
+    MR_LD_CACHE_WRITE_BACK =            0x01,
+    MR_LD_CACHE_WRITE_ADAPTIVE =        0x02,
+    MR_LD_CACHE_READ_AHEAD =            0x04,
+    MR_LD_CACHE_READ_ADAPTIVE =         0x08,
+    MR_LD_CACHE_WRITE_CACHE_BAD_BBU =   0x10,
+    MR_LD_CACHE_ALLOW_WRITE_CACHE =     0x20,
+    MR_LD_CACHE_ALLOW_READ_CACHE =      0x40
+} mfi_ld_cache;
+
+typedef enum {
+    MR_PD_CACHE_UNCHANGED  =    0,
+    MR_PD_CACHE_ENABLE =        1,
+    MR_PD_CACHE_DISABLE =       2
+} mfi_pd_cache;
+
+typedef enum {
+    MR_PD_QUERY_TYPE_ALL =              0,
+    MR_PD_QUERY_TYPE_STATE =            1,
+    MR_PD_QUERY_TYPE_POWER_STATE =      2,
+    MR_PD_QUERY_TYPE_MEDIA_TYPE =       3,
+    MR_PD_QUERY_TYPE_SPEED =            4,
+    MR_PD_QUERY_TYPE_EXPOSED_TO_HOST =  5, /*query for system drives */
+} mfi_pd_query_type;
+
+typedef enum {
+    MR_LD_QUERY_TYPE_ALL =              0,
+    MR_LD_QUERY_TYPE_EXPOSED_TO_HOST =  1,
+    MR_LD_QUERY_TYPE_USED_TGT_IDS =     2,
+    MR_LD_QUERY_TYPE_CLUSTER_ACCESS =   3,
+    MR_LD_QUERY_TYPE_CLUSTER_LOCALE =   4,
+} mfi_ld_query_type;
+
+/*
+ * Other propertities and definitions
+ */
+#define MFI_MAX_PD_CHANNELS     2
+#define MFI_MAX_LD_CHANNELS     2
+#define MFI_MAX_CHANNELS        (MFI_MAX_PD_CHANNELS + MFI_MAX_LD_CHANNELS)
+#define MFI_MAX_CHANNEL_DEVS  128
+#define MFI_DEFAULT_ID         -1
+#define MFI_MAX_LUN             8
+#define MFI_MAX_LD             64
+
+#define MFI_FRAME_SIZE         64
+#define MFI_MBOX_SIZE          12
+
+/* Firmware flashing can take 40s */
+#define MFI_POLL_TIMEOUT_SECS  50
+
+/* Allow for speedier math calculations */
+#define MFI_SECTOR_LEN        512
+
+/* Scatter Gather elements */
+struct mfi_sg32 {
+    uint32_t addr;
+    uint32_t len;
+} QEMU_PACKED;
+
+struct mfi_sg64 {
+    uint64_t addr;
+    uint32_t len;
+} QEMU_PACKED;
+
+struct mfi_sg_skinny {
+    uint64_t addr;
+    uint32_t len;
+    uint32_t flag;
+} QEMU_PACKED;
+
+union mfi_sgl {
+    struct mfi_sg32 sg32[1];
+    struct mfi_sg64 sg64[1];
+    struct mfi_sg_skinny sg_skinny[1];
+} QEMU_PACKED;
+
+/* Message frames.  All messages have a common header */
+struct mfi_frame_header {
+    uint8_t frame_cmd;
+    uint8_t sense_len;
+    uint8_t cmd_status;
+    uint8_t scsi_status;
+    uint8_t target_id;
+    uint8_t lun_id;
+    uint8_t cdb_len;
+    uint8_t sge_count;
+    uint64_t context;
+    uint16_t flags;
+    uint16_t timeout;
+    uint32_t data_len;
+} QEMU_PACKED;
+
+struct mfi_init_frame {
+    struct mfi_frame_header header;
+    uint32_t qinfo_new_addr_lo;
+    uint32_t qinfo_new_addr_hi;
+    uint32_t qinfo_old_addr_lo;
+    uint32_t qinfo_old_addr_hi;
+    uint32_t reserved[6];
+};
+
+#define MFI_IO_FRAME_SIZE 40
+struct mfi_io_frame {
+    struct mfi_frame_header header;
+    uint32_t sense_addr_lo;
+    uint32_t sense_addr_hi;
+    uint32_t lba_lo;
+    uint32_t lba_hi;
+    union mfi_sgl sgl;
+} QEMU_PACKED;
+
+#define MFI_PASS_FRAME_SIZE 48
+struct mfi_pass_frame {
+    struct mfi_frame_header header;
+    uint32_t sense_addr_lo;
+    uint32_t sense_addr_hi;
+    uint8_t cdb[16];
+    union mfi_sgl sgl;
+} QEMU_PACKED;
+
+#define MFI_DCMD_FRAME_SIZE 40
+struct mfi_dcmd_frame {
+    struct mfi_frame_header header;
+    uint32_t opcode;
+    uint8_t mbox[MFI_MBOX_SIZE];
+    union mfi_sgl sgl;
+} QEMU_PACKED;
+
+struct mfi_abort_frame {
+    struct mfi_frame_header header;
+    uint64_t abort_context;
+    uint32_t abort_mfi_addr_lo;
+    uint32_t abort_mfi_addr_hi;
+    uint32_t reserved1[6];
+} QEMU_PACKED;
+
+struct mfi_smp_frame {
+    struct mfi_frame_header header;
+    uint64_t sas_addr;
+    union {
+        struct mfi_sg32 sg32[2];
+        struct mfi_sg64 sg64[2];
+    } sgl;
+} QEMU_PACKED;
+
+struct mfi_stp_frame {
+    struct mfi_frame_header header;
+    uint16_t fis[10];
+    uint32_t stp_flags;
+    union {
+        struct mfi_sg32 sg32[2];
+        struct mfi_sg64 sg64[2];
+    } sgl;
+} QEMU_PACKED;
+
+union mfi_frame {
+    struct mfi_frame_header header;
+    struct mfi_init_frame init;
+    struct mfi_io_frame io;
+    struct mfi_pass_frame pass;
+    struct mfi_dcmd_frame dcmd;
+    struct mfi_abort_frame abort;
+    struct mfi_smp_frame smp;
+    struct mfi_stp_frame stp;
+    uint64_t raw[8];
+    uint8_t bytes[MFI_FRAME_SIZE];
+};
+
+#define MFI_SENSE_LEN 128
+struct mfi_sense {
+    uint8_t     data[MFI_SENSE_LEN];
+};
+
+#define MFI_QUEUE_FLAG_CONTEXT64 0x00000002
+
+/* The queue init structure that is passed with the init message */
+struct mfi_init_qinfo {
+    uint32_t flags;
+    uint32_t rq_entries;
+    uint32_t rq_addr_lo;
+    uint32_t rq_addr_hi;
+    uint32_t pi_addr_lo;
+    uint32_t pi_addr_hi;
+    uint32_t ci_addr_lo;
+    uint32_t ci_addr_hi;
+} QEMU_PACKED;
+
+/* Controller properties */
+struct mfi_ctrl_props {
+    uint16_t seq_num;
+    uint16_t pred_fail_poll_interval;
+    uint16_t intr_throttle_cnt;
+    uint16_t intr_throttle_timeout;
+    uint8_t rebuild_rate;
+    uint8_t patrol_read_rate;
+    uint8_t bgi_rate;
+    uint8_t cc_rate;
+    uint8_t recon_rate;
+    uint8_t cache_flush_interval;
+    uint8_t spinup_drv_cnt;
+    uint8_t spinup_delay;
+    uint8_t cluster_enable;
+    uint8_t coercion_mode;
+    uint8_t alarm_enable;
+    uint8_t disable_auto_rebuild;
+    uint8_t disable_battery_warn;
+    uint8_t ecc_bucket_size;
+    uint16_t ecc_bucket_leak_rate;
+    uint8_t restore_hotspare_on_insertion;
+    uint8_t expose_encl_devices;
+    uint8_t maintainPdFailHistory;
+    uint8_t disallowHostRequestReordering;
+    uint8_t abortCCOnError;
+    uint8_t loadBalanceMode;
+    uint8_t disableAutoDetectBackplane;
+    uint8_t snapVDSpace;
+    uint32_t OnOffProperties;
+/* set TRUE to disable copyBack (0=copyback enabled) */
+#define MFI_CTRL_PROP_CopyBackDisabled           (1 << 0)
+#define MFI_CTRL_PROP_SMARTerEnabled             (1 << 1)
+#define MFI_CTRL_PROP_PRCorrectUnconfiguredAreas (1 << 2)
+#define MFI_CTRL_PROP_UseFdeOnly                 (1 << 3)
+#define MFI_CTRL_PROP_DisableNCQ                 (1 << 4)
+#define MFI_CTRL_PROP_SSDSMARTerEnabled          (1 << 5)
+#define MFI_CTRL_PROP_SSDPatrolReadEnabled       (1 << 6)
+#define MFI_CTRL_PROP_EnableSpinDownUnconfigured (1 << 7)
+#define MFI_CTRL_PROP_AutoEnhancedImport         (1 << 8)
+#define MFI_CTRL_PROP_EnableSecretKeyControl     (1 << 9)
+#define MFI_CTRL_PROP_DisableOnlineCtrlReset     (1 << 10)
+#define MFI_CTRL_PROP_AllowBootWithPinnedCache   (1 << 11)
+#define MFI_CTRL_PROP_DisableSpinDownHS          (1 << 12)
+#define MFI_CTRL_PROP_EnableJBOD                 (1 << 13)
+
+    uint8_t autoSnapVDSpace; /* % of source LD to be
+                              * reserved for auto snapshot
+                              * in snapshot repository, for
+                              * metadata and user data
+                              * 1=5%, 2=10%, 3=15% and so on
+                              */
+    uint8_t viewSpace;       /* snapshot writeable VIEWs
+                              * capacity as a % of source LD
+                              * capacity. 0=READ only
+                              * 1=5%, 2=10%, 3=15% and so on
+                              */
+    uint16_t spinDownTime;    /* # of idle minutes before device
+                               * is spun down (0=use FW defaults)
+                               */
+    uint8_t reserved[24];
+} QEMU_PACKED;
+
+/* PCI information about the card. */
+struct mfi_info_pci {
+    uint16_t vendor;
+    uint16_t device;
+    uint16_t subvendor;
+    uint16_t subdevice;
+    uint8_t reserved[24];
+} QEMU_PACKED;
+
+/* Host (front end) interface information */
+struct mfi_info_host {
+    uint8_t type;
+#define MFI_INFO_HOST_PCIX      0x01
+#define MFI_INFO_HOST_PCIE      0x02
+#define MFI_INFO_HOST_ISCSI     0x04
+#define MFI_INFO_HOST_SAS3G     0x08
+    uint8_t reserved[6];
+    uint8_t port_count;
+    uint64_t port_addr[8];
+} QEMU_PACKED;
+
+/* Device (back end) interface information */
+struct mfi_info_device {
+    uint8_t type;
+#define MFI_INFO_DEV_SPI        0x01
+#define MFI_INFO_DEV_SAS3G      0x02
+#define MFI_INFO_DEV_SATA1      0x04
+#define MFI_INFO_DEV_SATA3G     0x08
+#define MFI_INFO_DEV_PCIE       0x10
+    uint8_t reserved[6];
+    uint8_t port_count;
+    uint64_t port_addr[8];
+} QEMU_PACKED;
+
+/* Firmware component information */
+struct mfi_info_component {
+    char name[8];
+    char version[32];
+    char build_date[16];
+    char build_time[16];
+} QEMU_PACKED;
+
+/* Controller default settings */
+struct mfi_defaults {
+    uint64_t sas_addr;
+    uint8_t phy_polarity;
+    uint8_t background_rate;
+    uint8_t stripe_size;
+    uint8_t flush_time;
+    uint8_t write_back;
+    uint8_t read_ahead;
+    uint8_t cache_when_bbu_bad;
+    uint8_t cached_io;
+    uint8_t smart_mode;
+    uint8_t alarm_disable;
+    uint8_t coercion;
+    uint8_t zrc_config;
+    uint8_t dirty_led_shows_drive_activity;
+    uint8_t bios_continue_on_error;
+    uint8_t spindown_mode;
+    uint8_t allowed_device_types;
+    uint8_t allow_mix_in_enclosure;
+    uint8_t allow_mix_in_ld;
+    uint8_t allow_sata_in_cluster;
+    uint8_t max_chained_enclosures;
+    uint8_t disable_ctrl_r;
+    uint8_t enable_web_bios;
+    uint8_t phy_polarity_split;
+    uint8_t direct_pd_mapping;
+    uint8_t bios_enumerate_lds;
+    uint8_t restored_hot_spare_on_insertion;
+    uint8_t expose_enclosure_devices;
+    uint8_t maintain_pd_fail_history;
+    uint8_t disable_puncture;
+    uint8_t zero_based_enumeration;
+    uint8_t disable_preboot_cli;
+    uint8_t show_drive_led_on_activity;
+    uint8_t cluster_disable;
+    uint8_t sas_disable;
+    uint8_t auto_detect_backplane;
+    uint8_t fde_only;
+    uint8_t delay_during_post;
+    uint8_t resv[19];
+} QEMU_PACKED;
+
+/* Controller default settings */
+struct mfi_bios_data {
+    uint16_t boot_target_id;
+    uint8_t do_not_int_13;
+    uint8_t continue_on_error;
+    uint8_t verbose;
+    uint8_t geometry;
+    uint8_t expose_all_drives;
+    uint8_t reserved[56];
+    uint8_t check_sum;
+} QEMU_PACKED;
+
+/* SAS (?) controller info, returned from MFI_DCMD_CTRL_GETINFO. */
+struct mfi_ctrl_info {
+    struct mfi_info_pci pci;
+    struct mfi_info_host host;
+    struct mfi_info_device device;
+
+    /* Firmware components that are present and active. */
+    uint32_t image_check_word;
+    uint32_t image_component_count;
+    struct mfi_info_component image_component[8];
+
+    /* Firmware components that have been flashed but are inactive */
+    uint32_t pending_image_component_count;
+    struct mfi_info_component pending_image_component[8];
+
+    uint8_t max_arms;
+    uint8_t max_spans;
+    uint8_t max_arrays;
+    uint8_t max_lds;
+    char product_name[80];
+    char serial_number[32];
+    uint32_t hw_present;
+#define MFI_INFO_HW_BBU         0x01
+#define MFI_INFO_HW_ALARM       0x02
+#define MFI_INFO_HW_NVRAM       0x04
+#define MFI_INFO_HW_UART        0x08
+#define MFI_INFO_HW_MEM         0x10
+#define MFI_INFO_HW_FLASH       0x20
+    uint32_t current_fw_time;
+    uint16_t max_cmds;
+    uint16_t max_sg_elements;
+    uint32_t max_request_size;
+    uint16_t lds_present;
+    uint16_t lds_degraded;
+    uint16_t lds_offline;
+    uint16_t pd_present;
+    uint16_t pd_disks_present;
+    uint16_t pd_disks_pred_failure;
+    uint16_t pd_disks_failed;
+    uint16_t nvram_size;
+    uint16_t memory_size;
+    uint16_t flash_size;
+    uint16_t ram_correctable_errors;
+    uint16_t ram_uncorrectable_errors;
+    uint8_t cluster_allowed;
+    uint8_t cluster_active;
+    uint16_t max_strips_per_io;
+
+    uint32_t raid_levels;
+#define MFI_INFO_RAID_0         0x01
+#define MFI_INFO_RAID_1         0x02
+#define MFI_INFO_RAID_5         0x04
+#define MFI_INFO_RAID_1E        0x08
+#define MFI_INFO_RAID_6         0x10
+
+    uint32_t adapter_ops;
+#define MFI_INFO_AOPS_RBLD_RATE         0x0001
+#define MFI_INFO_AOPS_CC_RATE           0x0002
+#define MFI_INFO_AOPS_BGI_RATE          0x0004
+#define MFI_INFO_AOPS_RECON_RATE        0x0008
+#define MFI_INFO_AOPS_PATROL_RATE       0x0010
+#define MFI_INFO_AOPS_ALARM_CONTROL     0x0020
+#define MFI_INFO_AOPS_CLUSTER_SUPPORTED 0x0040
+#define MFI_INFO_AOPS_BBU               0x0080
+#define MFI_INFO_AOPS_SPANNING_ALLOWED  0x0100
+#define MFI_INFO_AOPS_DEDICATED_SPARES  0x0200
+#define MFI_INFO_AOPS_REVERTIBLE_SPARES 0x0400
+#define MFI_INFO_AOPS_FOREIGN_IMPORT    0x0800
+#define MFI_INFO_AOPS_SELF_DIAGNOSTIC   0x1000
+#define MFI_INFO_AOPS_MIXED_ARRAY       0x2000
+#define MFI_INFO_AOPS_GLOBAL_SPARES     0x4000
+
+    uint32_t ld_ops;
+#define MFI_INFO_LDOPS_READ_POLICY      0x01
+#define MFI_INFO_LDOPS_WRITE_POLICY     0x02
+#define MFI_INFO_LDOPS_IO_POLICY        0x04
+#define MFI_INFO_LDOPS_ACCESS_POLICY    0x08
+#define MFI_INFO_LDOPS_DISK_CACHE_POLICY 0x10
+
+    struct {
+        uint8_t min;
+        uint8_t max;
+        uint8_t reserved[2];
+    } QEMU_PACKED stripe_sz_ops;
+
+    uint32_t pd_ops;
+#define MFI_INFO_PDOPS_FORCE_ONLINE     0x01
+#define MFI_INFO_PDOPS_FORCE_OFFLINE    0x02
+#define MFI_INFO_PDOPS_FORCE_REBUILD    0x04
+
+    uint32_t pd_mix_support;
+#define MFI_INFO_PDMIX_SAS              0x01
+#define MFI_INFO_PDMIX_SATA             0x02
+#define MFI_INFO_PDMIX_ENCL             0x04
+#define MFI_INFO_PDMIX_LD               0x08
+#define MFI_INFO_PDMIX_SATA_CLUSTER     0x10
+
+    uint8_t ecc_bucket_count;
+    uint8_t reserved2[11];
+    struct mfi_ctrl_props properties;
+    char package_version[0x60];
+    uint8_t pad[0x800 - 0x6a0];
+} QEMU_PACKED;
+
+/* keep track of an event. */
+union mfi_evt {
+    struct {
+        uint16_t locale;
+        uint8_t reserved;
+        int8_t class;
+    } members;
+    uint32_t word;
+} QEMU_PACKED;
+
+/* event log state. */
+struct mfi_evt_log_state {
+    uint32_t newest_seq_num;
+    uint32_t oldest_seq_num;
+    uint32_t clear_seq_num;
+    uint32_t shutdown_seq_num;
+    uint32_t boot_seq_num;
+} QEMU_PACKED;
+
+struct mfi_progress {
+    uint16_t progress;
+    uint16_t elapsed_seconds;
+} QEMU_PACKED;
+
+struct mfi_evt_ld {
+    uint16_t target_id;
+    uint8_t ld_index;
+    uint8_t reserved;
+} QEMU_PACKED;
+
+struct mfi_evt_pd {
+    uint16_t device_id;
+    uint8_t enclosure_index;
+    uint8_t slot_number;
+} QEMU_PACKED;
+
+/* event detail, returned from MFI_DCMD_CTRL_EVENT_WAIT. */
+struct mfi_evt_detail {
+    uint32_t seq;
+    uint32_t time;
+    uint32_t code;
+    union mfi_evt class;
+    uint8_t arg_type;
+    uint8_t reserved1[15];
+
+    union {
+        struct {
+            struct mfi_evt_pd pd;
+            uint8_t cdb_len;
+            uint8_t sense_len;
+            uint8_t reserved[2];
+            uint8_t cdb[16];
+            uint8_t sense[64];
+        } cdb_sense;
+
+        struct mfi_evt_ld ld;
+
+        struct {
+            struct mfi_evt_ld ld;
+            uint64_t count;
+        } ld_count;
+
+        struct {
+            uint64_t lba;
+            struct mfi_evt_ld ld;
+        } ld_lba;
+
+        struct {
+            struct mfi_evt_ld ld;
+            uint32_t pre_owner;
+            uint32_t new_owner;
+        } ld_owner;
+
+        struct {
+            uint64_t ld_lba;
+            uint64_t pd_lba;
+            struct mfi_evt_ld ld;
+            struct mfi_evt_pd pd;
+        } ld_lba_pd_lba;
+
+        struct {
+            struct mfi_evt_ld ld;
+            struct mfi_progress prog;
+        } ld_prog;
+
+        struct {
+            struct mfi_evt_ld ld;
+            uint32_t prev_state;
+            uint32_t new_state;
+        } ld_state;
+
+        struct {
+            uint64_t strip;
+            struct mfi_evt_ld ld;
+        } ld_strip;
+
+        struct mfi_evt_pd pd;
+
+        struct {
+            struct mfi_evt_pd pd;
+            uint32_t err;
+        } pd_err;
+
+        struct {
+            uint64_t lba;
+            struct mfi_evt_pd pd;
+        } pd_lba;
+
+        struct {
+            uint64_t lba;
+            struct mfi_evt_pd pd;
+            struct mfi_evt_ld ld;
+        } pd_lba_ld;
+
+        struct {
+            struct mfi_evt_pd pd;
+            struct mfi_progress prog;
+        } pd_prog;
+
+        struct {
+            struct mfi_evt_pd ld;
+            uint32_t prev_state;
+            uint32_t new_state;
+        } pd_state;
+
+        struct {
+            uint16_t venderId;
+            uint16_t deviceId;
+            uint16_t subVenderId;
+            uint16_t subDeviceId;
+        } pci;
+
+        uint32_t rate;
+
+        char str[96];
+
+        struct {
+            uint32_t rtc;
+            uint16_t elapsedSeconds;
+        } time;
+
+        struct {
+            uint32_t ecar;
+            uint32_t elog;
+            char str[64];
+        } ecc;
+
+        uint8_t b[96];
+        uint16_t s[48];
+        uint32_t w[24];
+        uint64_t d[12];
+    } args;
+
+    char description[128];
+} QEMU_PACKED;
+
+struct mfi_evt_list {
+    uint32_t count;
+    uint32_t reserved;
+    struct mfi_evt_detail event[1];
+} QEMU_PACKED;
+
+union mfi_pd_ref {
+    struct {
+        uint16_t device_id;
+        uint16_t seq_num;
+    } v;
+    uint32_t ref;
+} QEMU_PACKED;
+
+union mfi_pd_ddf_type {
+    struct {
+        uint16_t pd_type;
+#define MFI_PD_DDF_TYPE_FORCED_PD_GUID (1 << 0)
+#define MFI_PD_DDF_TYPE_IN_VD          (1 << 1)
+#define MFI_PD_DDF_TYPE_IS_GLOBAL_SPARE (1 << 2)
+#define MFI_PD_DDF_TYPE_IS_SPARE        (1 << 3)
+#define MFI_PD_DDF_TYPE_IS_FOREIGN      (1 << 4)
+#define MFI_PD_DDF_TYPE_INTF_SPI        (1 << 12)
+#define MFI_PD_DDF_TYPE_INTF_SAS        (1 << 13)
+#define MFI_PD_DDF_TYPE_INTF_SATA1      (1 << 14)
+#define MFI_PD_DDF_TYPE_INTF_SATA3G     (1 << 15)
+        uint16_t reserved;
+    } ddf;
+    struct {
+        uint32_t reserved;
+    } non_disk;
+    uint32_t type;
+} QEMU_PACKED;
+
+struct mfi_pd_progress {
+    uint32_t active;
+#define PD_PROGRESS_ACTIVE_REBUILD (1 << 0)
+#define PD_PROGRESS_ACTIVE_PATROL  (1 << 1)
+#define PD_PROGRESS_ACTIVE_CLEAR   (1 << 2)
+    struct mfi_progress rbld;
+    struct mfi_progress patrol;
+    struct mfi_progress clear;
+    struct mfi_progress reserved[4];
+} QEMU_PACKED;
+
+struct mfi_pd_info {
+    union mfi_pd_ref ref;
+    uint8_t inquiry_data[96];
+    uint8_t vpd_page83[64];
+    uint8_t not_supported;
+    uint8_t scsi_dev_type;
+    uint8_t connected_port_bitmap;
+    uint8_t device_speed;
+    uint32_t media_err_count;
+    uint32_t other_err_count;
+    uint32_t pred_fail_count;
+    uint32_t last_pred_fail_event_seq_num;
+    uint16_t fw_state;
+    uint8_t disable_for_removal;
+    uint8_t link_speed;
+    union mfi_pd_ddf_type state;
+    struct {
+        uint8_t count;
+        uint8_t is_path_broken;
+        uint8_t reserved[6];
+        uint64_t sas_addr[4];
+    } path_info;
+    uint64_t raw_size;
+    uint64_t non_coerced_size;
+    uint64_t coerced_size;
+    uint16_t encl_device_id;
+    uint8_t encl_index;
+    uint8_t slot_number;
+    struct mfi_pd_progress prog_info;
+    uint8_t bad_block_table_full;
+    uint8_t unusable_in_current_config;
+    uint8_t vpd_page83_ext[64];
+    uint8_t reserved[512-358];
+} QEMU_PACKED;
+
+struct mfi_pd_address {
+    uint16_t device_id;
+    uint16_t encl_device_id;
+    uint8_t encl_index;
+    uint8_t slot_number;
+    uint8_t scsi_dev_type;
+    uint8_t connect_port_bitmap;
+    uint64_t sas_addr[2];
+} QEMU_PACKED;
+
+#define MFI_MAX_SYS_PDS 240
+struct mfi_pd_list {
+    uint32_t size;
+    uint32_t count;
+    struct mfi_pd_address addr[MFI_MAX_SYS_PDS];
+} QEMU_PACKED;
+
+union mfi_ld_ref {
+    struct {
+        uint8_t target_id;
+        uint8_t reserved;
+        uint16_t seq;
+    } v;
+    uint32_t ref;
+} QEMU_PACKED;
+
+struct mfi_ld_list {
+    uint32_t ld_count;
+    uint32_t reserved1;
+    struct {
+        union mfi_ld_ref ld;
+        uint8_t state;
+        uint8_t reserved2[3];
+        uint64_t size;
+    } ld_list[MFI_MAX_LD];
+} QEMU_PACKED;
+
+struct mfi_ld_targetid_list {
+    uint32_t size;
+    uint32_t ld_count;
+    uint8_t pad[3];
+    uint8_t targetid[MFI_MAX_LD];
+} QEMU_PACKED;
+
+enum mfi_ld_access {
+    MFI_LD_ACCESS_RW =          0,
+    MFI_LD_ACCSSS_RO =          2,
+    MFI_LD_ACCESS_BLOCKED =     3,
+};
+#define MFI_LD_ACCESS_MASK      3
+
+enum mfi_ld_state {
+    MFI_LD_STATE_OFFLINE =              0,
+    MFI_LD_STATE_PARTIALLY_DEGRADED =   1,
+    MFI_LD_STATE_DEGRADED =             2,
+    MFI_LD_STATE_OPTIMAL =              3
+};
+
+enum mfi_syspd_state {
+    MFI_PD_STATE_UNCONFIGURED_GOOD =    0x00,
+    MFI_PD_STATE_UNCONFIGURED_BAD =     0x01,
+    MFI_PD_STATE_HOT_SPARE =            0x02,
+    MFI_PD_STATE_OFFLINE =              0x10,
+    MFI_PD_STATE_FAILED =               0x11,
+    MFI_PD_STATE_REBUILD =              0x14,
+    MFI_PD_STATE_ONLINE =               0x18,
+    MFI_PD_STATE_COPYBACK =             0x20,
+    MFI_PD_STATE_SYSTEM =               0x40
+};
+
+struct mfi_ld_props {
+    union mfi_ld_ref ld;
+    char name[16];
+    uint8_t default_cache_policy;
+    uint8_t access_policy;
+    uint8_t disk_cache_policy;
+    uint8_t current_cache_policy;
+    uint8_t no_bgi;
+    uint8_t reserved[7];
+} QEMU_PACKED;
+
+struct mfi_ld_params {
+    uint8_t primary_raid_level;
+    uint8_t raid_level_qualifier;
+    uint8_t secondary_raid_level;
+    uint8_t stripe_size;
+    uint8_t num_drives;
+    uint8_t span_depth;
+    uint8_t state;
+    uint8_t init_state;
+    uint8_t is_consistent;
+    uint8_t reserved[23];
+} QEMU_PACKED;
+
+struct mfi_ld_progress {
+    uint32_t            active;
+#define MFI_LD_PROGRESS_CC      (1<<0)
+#define MFI_LD_PROGRESS_BGI     (1<<1)
+#define MFI_LD_PROGRESS_FGI     (1<<2)
+#define MFI_LD_PORGRESS_RECON   (1<<3)
+    struct mfi_progress cc;
+    struct mfi_progress bgi;
+    struct mfi_progress fgi;
+    struct mfi_progress recon;
+    struct mfi_progress reserved[4];
+} QEMU_PACKED;
+
+struct mfi_span {
+    uint64_t start_block;
+    uint64_t num_blocks;
+    uint16_t array_ref;
+    uint8_t reserved[6];
+} QEMU_PACKED;
+
+#define MFI_MAX_SPAN_DEPTH      8
+struct mfi_ld_config {
+    struct mfi_ld_props properties;
+    struct mfi_ld_params params;
+    struct mfi_span span[MFI_MAX_SPAN_DEPTH];
+} QEMU_PACKED;
+
+struct mfi_ld_info {
+    struct mfi_ld_config ld_config;
+    uint64_t size;
+    struct mfi_ld_progress progress;
+    uint16_t cluster_owner;
+    uint8_t reconstruct_active;
+    uint8_t reserved1[1];
+    uint8_t vpd_page83[64];
+    uint8_t reserved2[16];
+} QEMU_PACKED;
+
+union mfi_spare_type {
+    uint8_t flags;
+#define MFI_SPARE_IS_DEDICATED (1 << 0)
+#define MFI_SPARE_IS_REVERTABLE (1 << 1)
+#define MFI_SPARE_IS_ENCL_AFFINITY (1 << 2)
+    uint8_t type;
+} QEMU_PACKED;
+
+#define MFI_MAX_ARRAYS 16
+struct mfi_spare {
+    union mfi_pd_ref ref;
+    union mfi_spare_type spare_type;
+    uint8_t reserved[2];
+    uint8_t array_count;
+    uint16_t array_refd[MFI_MAX_ARRAYS];
+} QEMU_PACKED;
+
+#define MFI_MAX_ROW_SIZE 32
+struct mfi_array {
+    uint64_t size;
+    uint8_t num_drives;
+    uint8_t reserved;
+    uint16_t array_ref;
+    uint8_t pad[20];
+    struct {
+        union mfi_pd_ref ref;
+        uint16_t fw_state; /* enum mfi_syspd_state */
+        struct {
+            uint8_t pd;
+            uint8_t slot;
+        } encl;
+    } pd[MFI_MAX_ROW_SIZE];
+} QEMU_PACKED;
+
+struct mfi_config_data {
+    uint32_t size;
+    uint16_t array_count;
+    uint16_t array_size;
+    uint16_t log_drv_count;
+    uint16_t log_drv_size;
+    uint16_t spares_count;
+    uint16_t spares_size;
+    uint8_t reserved[16];
+    /*
+      struct mfi_array  array[];
+      struct mfi_ld_config ld[];
+      struct mfi_spare  spare[];
+    */
+} QEMU_PACKED;
+
+#define MFI_SCSI_MAX_TARGETS  128
+#define MFI_SCSI_MAX_LUNS       8
+#define MFI_SCSI_INITIATOR_ID 255
+#define MFI_SCSI_MAX_CMDS       8
+#define MFI_SCSI_MAX_CDB_LEN   16
+
+#endif /* MFI_REG_H */
diff --git a/qemu/hw/scsi/scsi-bus.c b/qemu/hw/scsi/scsi-bus.c
new file mode 100644
index 000000000..f0ae4625f
--- /dev/null
+++ b/qemu/hw/scsi/scsi-bus.c
@@ -0,0 +1,2049 @@
+#include "hw/hw.h"
+#include "qemu/error-report.h"
+#include "hw/scsi/scsi.h"
+#include "block/scsi.h"
+#include "hw/qdev.h"
+#include "sysemu/block-backend.h"
+#include "sysemu/blockdev.h"
+#include "trace.h"
+#include "sysemu/dma.h"
+
+static char *scsibus_get_dev_path(DeviceState *dev);
+static char *scsibus_get_fw_dev_path(DeviceState *dev);
+static void scsi_req_dequeue(SCSIRequest *req);
+static uint8_t *scsi_target_alloc_buf(SCSIRequest *req, size_t len);
+static void scsi_target_free_buf(SCSIRequest *req);
+
+static Property scsi_props[] = {
+    DEFINE_PROP_UINT32("channel", SCSIDevice, channel, 0),
+    DEFINE_PROP_UINT32("scsi-id", SCSIDevice, id, -1),
+    DEFINE_PROP_UINT32("lun", SCSIDevice, lun, -1),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void scsi_bus_class_init(ObjectClass *klass, void *data)
+{
+    BusClass *k = BUS_CLASS(klass);
+    HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
+
+    k->get_dev_path = scsibus_get_dev_path;
+    k->get_fw_dev_path = scsibus_get_fw_dev_path;
+    hc->unplug = qdev_simple_device_unplug_cb;
+}
+
+static const TypeInfo scsi_bus_info = {
+    .name = TYPE_SCSI_BUS,
+    .parent = TYPE_BUS,
+    .instance_size = sizeof(SCSIBus),
+    .class_init = scsi_bus_class_init,
+    .interfaces = (InterfaceInfo[]) {
+        { TYPE_HOTPLUG_HANDLER },
+        { }
+    }
+};
+static int next_scsi_bus;
+
+static void scsi_device_realize(SCSIDevice *s, Error **errp)
+{
+    SCSIDeviceClass *sc = SCSI_DEVICE_GET_CLASS(s);
+    if (sc->realize) {
+        sc->realize(s, errp);
+    }
+}
+
+int scsi_bus_parse_cdb(SCSIDevice *dev, SCSICommand *cmd, uint8_t *buf,
+                       void *hba_private)
+{
+    SCSIBus *bus = DO_UPCAST(SCSIBus, qbus, dev->qdev.parent_bus);
+    int rc;
+
+    assert(cmd->len == 0);
+    rc = scsi_req_parse_cdb(dev, cmd, buf);
+    if (bus->info->parse_cdb) {
+        rc = bus->info->parse_cdb(dev, cmd, buf, hba_private);
+    }
+    return rc;
+}
+
+static SCSIRequest *scsi_device_alloc_req(SCSIDevice *s, uint32_t tag, uint32_t lun,
+                                          uint8_t *buf, void *hba_private)
+{
+    SCSIDeviceClass *sc = SCSI_DEVICE_GET_CLASS(s);
+    if (sc->alloc_req) {
+        return sc->alloc_req(s, tag, lun, buf, hba_private);
+    }
+
+    return NULL;
+}
+
+void scsi_device_unit_attention_reported(SCSIDevice *s)
+{
+    SCSIDeviceClass *sc = SCSI_DEVICE_GET_CLASS(s);
+    if (sc->unit_attention_reported) {
+        sc->unit_attention_reported(s);
+    }
+}
+
+/* Create a scsi bus, and attach devices to it.  */
+void scsi_bus_new(SCSIBus *bus, size_t bus_size, DeviceState *host,
+                  const SCSIBusInfo *info, const char *bus_name)
+{
+    qbus_create_inplace(bus, bus_size, TYPE_SCSI_BUS, host, bus_name);
+    bus->busnr = next_scsi_bus++;
+    bus->info = info;
+    qbus_set_bus_hotplug_handler(BUS(bus), &error_abort);
+}
+
+static void scsi_dma_restart_bh(void *opaque)
+{
+    SCSIDevice *s = opaque;
+    SCSIRequest *req, *next;
+
+    qemu_bh_delete(s->bh);
+    s->bh = NULL;
+
+    QTAILQ_FOREACH_SAFE(req, &s->requests, next, next) {
+        scsi_req_ref(req);
+        if (req->retry) {
+            req->retry = false;
+            switch (req->cmd.mode) {
+            case SCSI_XFER_FROM_DEV:
+            case SCSI_XFER_TO_DEV:
+                scsi_req_continue(req);
+                break;
+            case SCSI_XFER_NONE:
+                scsi_req_dequeue(req);
+                scsi_req_enqueue(req);
+                break;
+            }
+        }
+        scsi_req_unref(req);
+    }
+}
+
+void scsi_req_retry(SCSIRequest *req)
+{
+    /* No need to save a reference, because scsi_dma_restart_bh just
+     * looks at the request list.  */
+    req->retry = true;
+}
+
+static void scsi_dma_restart_cb(void *opaque, int running, RunState state)
+{
+    SCSIDevice *s = opaque;
+
+    if (!running) {
+        return;
+    }
+    if (!s->bh) {
+        s->bh = qemu_bh_new(scsi_dma_restart_bh, s);
+        qemu_bh_schedule(s->bh);
+    }
+}
+
+static void scsi_qdev_realize(DeviceState *qdev, Error **errp)
+{
+    SCSIDevice *dev = SCSI_DEVICE(qdev);
+    SCSIBus *bus = DO_UPCAST(SCSIBus, qbus, dev->qdev.parent_bus);
+    SCSIDevice *d;
+    Error *local_err = NULL;
+
+    if (dev->channel > bus->info->max_channel) {
+        error_setg(errp, "bad scsi channel id: %d", dev->channel);
+        return;
+    }
+    if (dev->id != -1 && dev->id > bus->info->max_target) {
+        error_setg(errp, "bad scsi device id: %d", dev->id);
+        return;
+    }
+    if (dev->lun != -1 && dev->lun > bus->info->max_lun) {
+        error_setg(errp, "bad scsi device lun: %d", dev->lun);
+        return;
+    }
+
+    if (dev->id == -1) {
+        int id = -1;
+        if (dev->lun == -1) {
+            dev->lun = 0;
+        }
+        do {
+            d = scsi_device_find(bus, dev->channel, ++id, dev->lun);
+        } while (d && d->lun == dev->lun && id < bus->info->max_target);
+        if (d && d->lun == dev->lun) {
+            error_setg(errp, "no free target");
+            return;
+        }
+        dev->id = id;
+    } else if (dev->lun == -1) {
+        int lun = -1;
+        do {
+            d = scsi_device_find(bus, dev->channel, dev->id, ++lun);
+        } while (d && d->lun == lun && lun < bus->info->max_lun);
+        if (d && d->lun == lun) {
+            error_setg(errp, "no free lun");
+            return;
+        }
+        dev->lun = lun;
+    } else {
+        d = scsi_device_find(bus, dev->channel, dev->id, dev->lun);
+        assert(d);
+        if (d->lun == dev->lun && dev != d) {
+            error_setg(errp, "lun already used by '%s'", d->qdev.id);
+            return;
+        }
+    }
+
+    QTAILQ_INIT(&dev->requests);
+    scsi_device_realize(dev, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+    dev->vmsentry = qemu_add_vm_change_state_handler(scsi_dma_restart_cb,
+                                                     dev);
+}
+
+static void scsi_qdev_unrealize(DeviceState *qdev, Error **errp)
+{
+    SCSIDevice *dev = SCSI_DEVICE(qdev);
+
+    if (dev->vmsentry) {
+        qemu_del_vm_change_state_handler(dev->vmsentry);
+    }
+
+    scsi_device_purge_requests(dev, SENSE_CODE(NO_SENSE));
+    blockdev_mark_auto_del(dev->conf.blk);
+}
+
+/* handle legacy '-drive if=scsi,...' cmd line args */
+SCSIDevice *scsi_bus_legacy_add_drive(SCSIBus *bus, BlockBackend *blk,
+                                      int unit, bool removable, int bootindex,
+                                      const char *serial, Error **errp)
+{
+    const char *driver;
+    char *name;
+    DeviceState *dev;
+    Error *err = NULL;
+
+    driver = blk_is_sg(blk) ? "scsi-generic" : "scsi-disk";
+    dev = qdev_create(&bus->qbus, driver);
+    name = g_strdup_printf("legacy[%d]", unit);
+    object_property_add_child(OBJECT(bus), name, OBJECT(dev), NULL);
+    g_free(name);
+
+    qdev_prop_set_uint32(dev, "scsi-id", unit);
+    if (bootindex >= 0) {
+        object_property_set_int(OBJECT(dev), bootindex, "bootindex",
+                                &error_abort);
+    }
+    if (object_property_find(OBJECT(dev), "removable", NULL)) {
+        qdev_prop_set_bit(dev, "removable", removable);
+    }
+    if (serial && object_property_find(OBJECT(dev), "serial", NULL)) {
+        qdev_prop_set_string(dev, "serial", serial);
+    }
+    qdev_prop_set_drive(dev, "drive", blk, &err);
+    if (err) {
+        error_propagate(errp, err);
+        object_unparent(OBJECT(dev));
+        return NULL;
+    }
+    object_property_set_bool(OBJECT(dev), true, "realized", &err);
+    if (err != NULL) {
+        error_propagate(errp, err);
+        object_unparent(OBJECT(dev));
+        return NULL;
+    }
+    return SCSI_DEVICE(dev);
+}
+
+void scsi_bus_legacy_handle_cmdline(SCSIBus *bus, Error **errp)
+{
+    Location loc;
+    DriveInfo *dinfo;
+    int unit;
+    Error *err = NULL;
+
+    loc_push_none(&loc);
+    for (unit = 0; unit <= bus->info->max_target; unit++) {
+        dinfo = drive_get(IF_SCSI, bus->busnr, unit);
+        if (dinfo == NULL) {
+            continue;
+        }
+        qemu_opts_loc_restore(dinfo->opts);
+        scsi_bus_legacy_add_drive(bus, blk_by_legacy_dinfo(dinfo),
+                                  unit, false, -1, NULL, &err);
+        if (err != NULL) {
+            error_propagate(errp, err);
+            break;
+        }
+    }
+    loc_pop(&loc);
+}
+
+static int32_t scsi_invalid_field(SCSIRequest *req, uint8_t *buf)
+{
+    scsi_req_build_sense(req, SENSE_CODE(INVALID_FIELD));
+    scsi_req_complete(req, CHECK_CONDITION);
+    return 0;
+}
+
+static const struct SCSIReqOps reqops_invalid_field = {
+    .size         = sizeof(SCSIRequest),
+    .send_command = scsi_invalid_field
+};
+
+/* SCSIReqOps implementation for invalid commands.  */
+
+static int32_t scsi_invalid_command(SCSIRequest *req, uint8_t *buf)
+{
+    scsi_req_build_sense(req, SENSE_CODE(INVALID_OPCODE));
+    scsi_req_complete(req, CHECK_CONDITION);
+    return 0;
+}
+
+static const struct SCSIReqOps reqops_invalid_opcode = {
+    .size         = sizeof(SCSIRequest),
+    .send_command = scsi_invalid_command
+};
+
+/* SCSIReqOps implementation for unit attention conditions.  */
+
+static int32_t scsi_unit_attention(SCSIRequest *req, uint8_t *buf)
+{
+    if (req->dev->unit_attention.key == UNIT_ATTENTION) {
+        scsi_req_build_sense(req, req->dev->unit_attention);
+    } else if (req->bus->unit_attention.key == UNIT_ATTENTION) {
+        scsi_req_build_sense(req, req->bus->unit_attention);
+    }
+    scsi_req_complete(req, CHECK_CONDITION);
+    return 0;
+}
+
+static const struct SCSIReqOps reqops_unit_attention = {
+    .size         = sizeof(SCSIRequest),
+    .send_command = scsi_unit_attention
+};
+
+/* SCSIReqOps implementation for REPORT LUNS and for commands sent to
+   an invalid LUN.  */
+
+typedef struct SCSITargetReq SCSITargetReq;
+
+struct SCSITargetReq {
+    SCSIRequest req;
+    int len;
+    uint8_t *buf;
+    int buf_len;
+};
+
+static void store_lun(uint8_t *outbuf, int lun)
+{
+    if (lun < 256) {
+        outbuf[1] = lun;
+        return;
+    }
+    outbuf[1] = (lun & 255);
+    outbuf[0] = (lun >> 8) | 0x40;
+}
+
+static bool scsi_target_emulate_report_luns(SCSITargetReq *r)
+{
+    BusChild *kid;
+    int i, len, n;
+    int channel, id;
+    bool found_lun0;
+
+    if (r->req.cmd.xfer < 16) {
+        return false;
+    }
+    if (r->req.cmd.buf[2] > 2) {
+        return false;
+    }
+    channel = r->req.dev->channel;
+    id = r->req.dev->id;
+    found_lun0 = false;
+    n = 0;
+    QTAILQ_FOREACH(kid, &r->req.bus->qbus.children, sibling) {
+        DeviceState *qdev = kid->child;
+        SCSIDevice *dev = SCSI_DEVICE(qdev);
+
+        if (dev->channel == channel && dev->id == id) {
+            if (dev->lun == 0) {
+                found_lun0 = true;
+            }
+            n += 8;
+        }
+    }
+    if (!found_lun0) {
+        n += 8;
+    }
+
+    scsi_target_alloc_buf(&r->req, n + 8);
+
+    len = MIN(n + 8, r->req.cmd.xfer & ~7);
+    memset(r->buf, 0, len);
+    stl_be_p(&r->buf[0], n);
+    i = found_lun0 ? 8 : 16;
+    QTAILQ_FOREACH(kid, &r->req.bus->qbus.children, sibling) {
+        DeviceState *qdev = kid->child;
+        SCSIDevice *dev = SCSI_DEVICE(qdev);
+
+        if (dev->channel == channel && dev->id == id) {
+            store_lun(&r->buf[i], dev->lun);
+            i += 8;
+        }
+    }
+    assert(i == n + 8);
+    r->len = len;
+    return true;
+}
+
+static bool scsi_target_emulate_inquiry(SCSITargetReq *r)
+{
+    assert(r->req.dev->lun != r->req.lun);
+
+    scsi_target_alloc_buf(&r->req, SCSI_INQUIRY_LEN);
+
+    if (r->req.cmd.buf[1] & 0x2) {
+        /* Command support data - optional, not implemented */
+        return false;
+    }
+
+    if (r->req.cmd.buf[1] & 0x1) {
+        /* Vital product data */
+        uint8_t page_code = r->req.cmd.buf[2];
+        r->buf[r->len++] = page_code ; /* this page */
+        r->buf[r->len++] = 0x00;
+
+        switch (page_code) {
+        case 0x00: /* Supported page codes, mandatory */
+        {
+            int pages;
+            pages = r->len++;
+            r->buf[r->len++] = 0x00; /* list of supported pages (this page) */
+            r->buf[pages] = r->len - pages - 1; /* number of pages */
+            break;
+        }
+        default:
+            return false;
+        }
+        /* done with EVPD */
+        assert(r->len < r->buf_len);
+        r->len = MIN(r->req.cmd.xfer, r->len);
+        return true;
+    }
+
+    /* Standard INQUIRY data */
+    if (r->req.cmd.buf[2] != 0) {
+        return false;
+    }
+
+    /* PAGE CODE == 0 */
+    r->len = MIN(r->req.cmd.xfer, SCSI_INQUIRY_LEN);
+    memset(r->buf, 0, r->len);
+    if (r->req.lun != 0) {
+        r->buf[0] = TYPE_NO_LUN;
+    } else {
+        r->buf[0] = TYPE_NOT_PRESENT | TYPE_INACTIVE;
+        r->buf[2] = 5; /* Version */
+        r->buf[3] = 2 | 0x10; /* HiSup, response data format */
+        r->buf[4] = r->len - 5; /* Additional Length = (Len - 1) - 4 */
+        r->buf[7] = 0x10 | (r->req.bus->info->tcq ? 0x02 : 0); /* Sync, TCQ.  */
+        memcpy(&r->buf[8], "QEMU    ", 8);
+        memcpy(&r->buf[16], "QEMU TARGET     ", 16);
+        pstrcpy((char *) &r->buf[32], 4, qemu_get_version());
+    }
+    return true;
+}
+
+static int32_t scsi_target_send_command(SCSIRequest *req, uint8_t *buf)
+{
+    SCSITargetReq *r = DO_UPCAST(SCSITargetReq, req, req);
+
+    switch (buf[0]) {
+    case REPORT_LUNS:
+        if (!scsi_target_emulate_report_luns(r)) {
+            goto illegal_request;
+        }
+        break;
+    case INQUIRY:
+        if (!scsi_target_emulate_inquiry(r)) {
+            goto illegal_request;
+        }
+        break;
+    case REQUEST_SENSE:
+        scsi_target_alloc_buf(&r->req, SCSI_SENSE_LEN);
+        r->len = scsi_device_get_sense(r->req.dev, r->buf,
+                                       MIN(req->cmd.xfer, r->buf_len),
+                                       (req->cmd.buf[1] & 1) == 0);
+        if (r->req.dev->sense_is_ua) {
+            scsi_device_unit_attention_reported(req->dev);
+            r->req.dev->sense_len = 0;
+            r->req.dev->sense_is_ua = false;
+        }
+        break;
+    case TEST_UNIT_READY:
+        break;
+    default:
+        scsi_req_build_sense(req, SENSE_CODE(LUN_NOT_SUPPORTED));
+        scsi_req_complete(req, CHECK_CONDITION);
+        return 0;
+    illegal_request:
+        scsi_req_build_sense(req, SENSE_CODE(INVALID_FIELD));
+        scsi_req_complete(req, CHECK_CONDITION);
+        return 0;
+    }
+
+    if (!r->len) {
+        scsi_req_complete(req, GOOD);
+    }
+    return r->len;
+}
+
+static void scsi_target_read_data(SCSIRequest *req)
+{
+    SCSITargetReq *r = DO_UPCAST(SCSITargetReq, req, req);
+    uint32_t n;
+
+    n = r->len;
+    if (n > 0) {
+        r->len = 0;
+        scsi_req_data(&r->req, n);
+    } else {
+        scsi_req_complete(&r->req, GOOD);
+    }
+}
+
+static uint8_t *scsi_target_get_buf(SCSIRequest *req)
+{
+    SCSITargetReq *r = DO_UPCAST(SCSITargetReq, req, req);
+
+    return r->buf;
+}
+
+static uint8_t *scsi_target_alloc_buf(SCSIRequest *req, size_t len)
+{
+    SCSITargetReq *r = DO_UPCAST(SCSITargetReq, req, req);
+
+    r->buf = g_malloc(len);
+    r->buf_len = len;
+
+    return r->buf;
+}
+
+static void scsi_target_free_buf(SCSIRequest *req)
+{
+    SCSITargetReq *r = DO_UPCAST(SCSITargetReq, req, req);
+
+    g_free(r->buf);
+}
+
+static const struct SCSIReqOps reqops_target_command = {
+    .size         = sizeof(SCSITargetReq),
+    .send_command = scsi_target_send_command,
+    .read_data    = scsi_target_read_data,
+    .get_buf      = scsi_target_get_buf,
+    .free_req     = scsi_target_free_buf,
+};
+
+
+SCSIRequest *scsi_req_alloc(const SCSIReqOps *reqops, SCSIDevice *d,
+                            uint32_t tag, uint32_t lun, void *hba_private)
+{
+    SCSIRequest *req;
+    SCSIBus *bus = scsi_bus_from_device(d);
+    BusState *qbus = BUS(bus);
+    const int memset_off = offsetof(SCSIRequest, sense)
+                           + sizeof(req->sense);
+
+    req = g_slice_alloc(reqops->size);
+    memset((uint8_t *)req + memset_off, 0, reqops->size - memset_off);
+    req->refcount = 1;
+    req->bus = bus;
+    req->dev = d;
+    req->tag = tag;
+    req->lun = lun;
+    req->hba_private = hba_private;
+    req->status = -1;
+    req->ops = reqops;
+    object_ref(OBJECT(d));
+    object_ref(OBJECT(qbus->parent));
+    notifier_list_init(&req->cancel_notifiers);
+    trace_scsi_req_alloc(req->dev->id, req->lun, req->tag);
+    return req;
+}
+
+SCSIRequest *scsi_req_new(SCSIDevice *d, uint32_t tag, uint32_t lun,
+                          uint8_t *buf, void *hba_private)
+{
+    SCSIBus *bus = DO_UPCAST(SCSIBus, qbus, d->qdev.parent_bus);
+    const SCSIReqOps *ops;
+    SCSIDeviceClass *sc = SCSI_DEVICE_GET_CLASS(d);
+    SCSIRequest *req;
+    SCSICommand cmd = { .len = 0 };
+    int ret;
+
+    if ((d->unit_attention.key == UNIT_ATTENTION ||
+         bus->unit_attention.key == UNIT_ATTENTION) &&
+        (buf[0] != INQUIRY &&
+         buf[0] != REPORT_LUNS &&
+         buf[0] != GET_CONFIGURATION &&
+         buf[0] != GET_EVENT_STATUS_NOTIFICATION &&
+
+         /*
+          * If we already have a pending unit attention condition,
+          * report this one before triggering another one.
+          */
+         !(buf[0] == REQUEST_SENSE && d->sense_is_ua))) {
+        ops = &reqops_unit_attention;
+    } else if (lun != d->lun ||
+               buf[0] == REPORT_LUNS ||
+               (buf[0] == REQUEST_SENSE && d->sense_len)) {
+        ops = &reqops_target_command;
+    } else {
+        ops = NULL;
+    }
+
+    if (ops != NULL || !sc->parse_cdb) {
+        ret = scsi_req_parse_cdb(d, &cmd, buf);
+    } else {
+        ret = sc->parse_cdb(d, &cmd, buf, hba_private);
+    }
+
+    if (ret != 0) {
+        trace_scsi_req_parse_bad(d->id, lun, tag, buf[0]);
+        req = scsi_req_alloc(&reqops_invalid_opcode, d, tag, lun, hba_private);
+    } else {
+        assert(cmd.len != 0);
+        trace_scsi_req_parsed(d->id, lun, tag, buf[0],
+                              cmd.mode, cmd.xfer);
+        if (cmd.lba != -1) {
+            trace_scsi_req_parsed_lba(d->id, lun, tag, buf[0],
+                                      cmd.lba);
+        }
+
+        if (cmd.xfer > INT32_MAX) {
+            req = scsi_req_alloc(&reqops_invalid_field, d, tag, lun, hba_private);
+        } else if (ops) {
+            req = scsi_req_alloc(ops, d, tag, lun, hba_private);
+        } else {
+            req = scsi_device_alloc_req(d, tag, lun, buf, hba_private);
+        }
+    }
+
+    req->cmd = cmd;
+    req->resid = req->cmd.xfer;
+
+    switch (buf[0]) {
+    case INQUIRY:
+        trace_scsi_inquiry(d->id, lun, tag, cmd.buf[1], cmd.buf[2]);
+        break;
+    case TEST_UNIT_READY:
+        trace_scsi_test_unit_ready(d->id, lun, tag);
+        break;
+    case REPORT_LUNS:
+        trace_scsi_report_luns(d->id, lun, tag);
+        break;
+    case REQUEST_SENSE:
+        trace_scsi_request_sense(d->id, lun, tag);
+        break;
+    default:
+        break;
+    }
+
+    return req;
+}
+
+uint8_t *scsi_req_get_buf(SCSIRequest *req)
+{
+    return req->ops->get_buf(req);
+}
+
+static void scsi_clear_unit_attention(SCSIRequest *req)
+{
+    SCSISense *ua;
+    if (req->dev->unit_attention.key != UNIT_ATTENTION &&
+        req->bus->unit_attention.key != UNIT_ATTENTION) {
+        return;
+    }
+
+    /*
+     * If an INQUIRY command enters the enabled command state,
+     * the device server shall [not] clear any unit attention condition;
+     * See also MMC-6, paragraphs 6.5 and 6.6.2.
+     */
+    if (req->cmd.buf[0] == INQUIRY ||
+        req->cmd.buf[0] == GET_CONFIGURATION ||
+        req->cmd.buf[0] == GET_EVENT_STATUS_NOTIFICATION) {
+        return;
+    }
+
+    if (req->dev->unit_attention.key == UNIT_ATTENTION) {
+        ua = &req->dev->unit_attention;
+    } else {
+        ua = &req->bus->unit_attention;
+    }
+
+    /*
+     * If a REPORT LUNS command enters the enabled command state, [...]
+     * the device server shall clear any pending unit attention condition
+     * with an additional sense code of REPORTED LUNS DATA HAS CHANGED.
+     */
+    if (req->cmd.buf[0] == REPORT_LUNS &&
+        !(ua->asc == SENSE_CODE(REPORTED_LUNS_CHANGED).asc &&
+          ua->ascq == SENSE_CODE(REPORTED_LUNS_CHANGED).ascq)) {
+        return;
+    }
+
+    *ua = SENSE_CODE(NO_SENSE);
+}
+
+int scsi_req_get_sense(SCSIRequest *req, uint8_t *buf, int len)
+{
+    int ret;
+
+    assert(len >= 14);
+    if (!req->sense_len) {
+        return 0;
+    }
+
+    ret = scsi_build_sense(req->sense, req->sense_len, buf, len, true);
+
+    /*
+     * FIXME: clearing unit attention conditions upon autosense should be done
+     * only if the UA_INTLCK_CTRL field in the Control mode page is set to 00b
+     * (SAM-5, 5.14).
+     *
+     * We assume UA_INTLCK_CTRL to be 00b for HBAs that support autosense, and
+     * 10b for HBAs that do not support it (do not call scsi_req_get_sense).
+     * Here we handle unit attention clearing for UA_INTLCK_CTRL == 00b.
+     */
+    if (req->dev->sense_is_ua) {
+        scsi_device_unit_attention_reported(req->dev);
+        req->dev->sense_len = 0;
+        req->dev->sense_is_ua = false;
+    }
+    return ret;
+}
+
+int scsi_device_get_sense(SCSIDevice *dev, uint8_t *buf, int len, bool fixed)
+{
+    return scsi_build_sense(dev->sense, dev->sense_len, buf, len, fixed);
+}
+
+void scsi_req_build_sense(SCSIRequest *req, SCSISense sense)
+{
+    trace_scsi_req_build_sense(req->dev->id, req->lun, req->tag,
+                               sense.key, sense.asc, sense.ascq);
+    memset(req->sense, 0, 18);
+    req->sense[0] = 0x70;
+    req->sense[2] = sense.key;
+    req->sense[7] = 10;
+    req->sense[12] = sense.asc;
+    req->sense[13] = sense.ascq;
+    req->sense_len = 18;
+}
+
+static void scsi_req_enqueue_internal(SCSIRequest *req)
+{
+    assert(!req->enqueued);
+    scsi_req_ref(req);
+    if (req->bus->info->get_sg_list) {
+        req->sg = req->bus->info->get_sg_list(req);
+    } else {
+        req->sg = NULL;
+    }
+    req->enqueued = true;
+    QTAILQ_INSERT_TAIL(&req->dev->requests, req, next);
+}
+
+int32_t scsi_req_enqueue(SCSIRequest *req)
+{
+    int32_t rc;
+
+    assert(!req->retry);
+    scsi_req_enqueue_internal(req);
+    scsi_req_ref(req);
+    rc = req->ops->send_command(req, req->cmd.buf);
+    scsi_req_unref(req);
+    return rc;
+}
+
+static void scsi_req_dequeue(SCSIRequest *req)
+{
+    trace_scsi_req_dequeue(req->dev->id, req->lun, req->tag);
+    req->retry = false;
+    if (req->enqueued) {
+        QTAILQ_REMOVE(&req->dev->requests, req, next);
+        req->enqueued = false;
+        scsi_req_unref(req);
+    }
+}
+
+static int scsi_get_performance_length(int num_desc, int type, int data_type)
+{
+    /* MMC-6, paragraph 6.7.  */
+    switch (type) {
+    case 0:
+        if ((data_type & 3) == 0) {
+            /* Each descriptor is as in Table 295 - Nominal performance.  */
+            return 16 * num_desc + 8;
+        } else {
+            /* Each descriptor is as in Table 296 - Exceptions.  */
+            return 6 * num_desc + 8;
+        }
+    case 1:
+    case 4:
+    case 5:
+        return 8 * num_desc + 8;
+    case 2:
+        return 2048 * num_desc + 8;
+    case 3:
+        return 16 * num_desc + 8;
+    default:
+        return 8;
+    }
+}
+
+static int ata_passthrough_xfer_unit(SCSIDevice *dev, uint8_t *buf)
+{
+    int byte_block = (buf[2] >> 2) & 0x1;
+    int type = (buf[2] >> 4) & 0x1;
+    int xfer_unit;
+
+    if (byte_block) {
+        if (type) {
+            xfer_unit = dev->blocksize;
+        } else {
+            xfer_unit = 512;
+        }
+    } else {
+        xfer_unit = 1;
+    }
+
+    return xfer_unit;
+}
+
+static int ata_passthrough_12_xfer(SCSIDevice *dev, uint8_t *buf)
+{
+    int length = buf[2] & 0x3;
+    int xfer;
+    int unit = ata_passthrough_xfer_unit(dev, buf);
+
+    switch (length) {
+    case 0:
+    case 3: /* USB-specific.  */
+    default:
+        xfer = 0;
+        break;
+    case 1:
+        xfer = buf[3];
+        break;
+    case 2:
+        xfer = buf[4];
+        break;
+    }
+
+    return xfer * unit;
+}
+
+static int ata_passthrough_16_xfer(SCSIDevice *dev, uint8_t *buf)
+{
+    int extend = buf[1] & 0x1;
+    int length = buf[2] & 0x3;
+    int xfer;
+    int unit = ata_passthrough_xfer_unit(dev, buf);
+
+    switch (length) {
+    case 0:
+    case 3: /* USB-specific.  */
+    default:
+        xfer = 0;
+        break;
+    case 1:
+        xfer = buf[4];
+        xfer |= (extend ? buf[3] << 8 : 0);
+        break;
+    case 2:
+        xfer = buf[6];
+        xfer |= (extend ? buf[5] << 8 : 0);
+        break;
+    }
+
+    return xfer * unit;
+}
+
+uint32_t scsi_data_cdb_xfer(uint8_t *buf)
+{
+    if ((buf[0] >> 5) == 0 && buf[4] == 0) {
+        return 256;
+    } else {
+        return scsi_cdb_xfer(buf);
+    }
+}
+
+uint32_t scsi_cdb_xfer(uint8_t *buf)
+{
+    switch (buf[0] >> 5) {
+    case 0:
+        return buf[4];
+        break;
+    case 1:
+    case 2:
+        return lduw_be_p(&buf[7]);
+        break;
+    case 4:
+        return ldl_be_p(&buf[10]) & 0xffffffffULL;
+        break;
+    case 5:
+        return ldl_be_p(&buf[6]) & 0xffffffffULL;
+        break;
+    default:
+        return -1;
+    }
+}
+
+static int scsi_req_xfer(SCSICommand *cmd, SCSIDevice *dev, uint8_t *buf)
+{
+    cmd->xfer = scsi_cdb_xfer(buf);
+    switch (buf[0]) {
+    case TEST_UNIT_READY:
+    case REWIND:
+    case START_STOP:
+    case SET_CAPACITY:
+    case WRITE_FILEMARKS:
+    case WRITE_FILEMARKS_16:
+    case SPACE:
+    case RESERVE:
+    case RELEASE:
+    case ERASE:
+    case ALLOW_MEDIUM_REMOVAL:
+    case SEEK_10:
+    case SYNCHRONIZE_CACHE:
+    case SYNCHRONIZE_CACHE_16:
+    case LOCATE_16:
+    case LOCK_UNLOCK_CACHE:
+    case SET_CD_SPEED:
+    case SET_LIMITS:
+    case WRITE_LONG_10:
+    case UPDATE_BLOCK:
+    case RESERVE_TRACK:
+    case SET_READ_AHEAD:
+    case PRE_FETCH:
+    case PRE_FETCH_16:
+    case ALLOW_OVERWRITE:
+        cmd->xfer = 0;
+        break;
+    case VERIFY_10:
+    case VERIFY_12:
+    case VERIFY_16:
+        if ((buf[1] & 2) == 0) {
+            cmd->xfer = 0;
+        } else if ((buf[1] & 4) != 0) {
+            cmd->xfer = 1;
+        }
+        cmd->xfer *= dev->blocksize;
+        break;
+    case MODE_SENSE:
+        break;
+    case WRITE_SAME_10:
+    case WRITE_SAME_16:
+        cmd->xfer = dev->blocksize;
+        break;
+    case READ_CAPACITY_10:
+        cmd->xfer = 8;
+        break;
+    case READ_BLOCK_LIMITS:
+        cmd->xfer = 6;
+        break;
+    case SEND_VOLUME_TAG:
+        /* GPCMD_SET_STREAMING from multimedia commands.  */
+        if (dev->type == TYPE_ROM) {
+            cmd->xfer = buf[10] | (buf[9] << 8);
+        } else {
+            cmd->xfer = buf[9] | (buf[8] << 8);
+        }
+        break;
+    case WRITE_6:
+        /* length 0 means 256 blocks */
+        if (cmd->xfer == 0) {
+            cmd->xfer = 256;
+        }
+        /* fall through */
+    case WRITE_10:
+    case WRITE_VERIFY_10:
+    case WRITE_12:
+    case WRITE_VERIFY_12:
+    case WRITE_16:
+    case WRITE_VERIFY_16:
+        cmd->xfer *= dev->blocksize;
+        break;
+    case READ_6:
+    case READ_REVERSE:
+        /* length 0 means 256 blocks */
+        if (cmd->xfer == 0) {
+            cmd->xfer = 256;
+        }
+        /* fall through */
+    case READ_10:
+    case RECOVER_BUFFERED_DATA:
+    case READ_12:
+    case READ_16:
+        cmd->xfer *= dev->blocksize;
+        break;
+    case FORMAT_UNIT:
+        /* MMC mandates the parameter list to be 12-bytes long.  Parameters
+         * for block devices are restricted to the header right now.  */
+        if (dev->type == TYPE_ROM && (buf[1] & 16)) {
+            cmd->xfer = 12;
+        } else {
+            cmd->xfer = (buf[1] & 16) == 0 ? 0 : (buf[1] & 32 ? 8 : 4);
+        }
+        break;
+    case INQUIRY:
+    case RECEIVE_DIAGNOSTIC:
+    case SEND_DIAGNOSTIC:
+        cmd->xfer = buf[4] | (buf[3] << 8);
+        break;
+    case READ_CD:
+    case READ_BUFFER:
+    case WRITE_BUFFER:
+    case SEND_CUE_SHEET:
+        cmd->xfer = buf[8] | (buf[7] << 8) | (buf[6] << 16);
+        break;
+    case PERSISTENT_RESERVE_OUT:
+        cmd->xfer = ldl_be_p(&buf[5]) & 0xffffffffULL;
+        break;
+    case ERASE_12:
+        if (dev->type == TYPE_ROM) {
+            /* MMC command GET PERFORMANCE.  */
+            cmd->xfer = scsi_get_performance_length(buf[9] | (buf[8] << 8),
+                                                    buf[10], buf[1] & 0x1f);
+        }
+        break;
+    case MECHANISM_STATUS:
+    case READ_DVD_STRUCTURE:
+    case SEND_DVD_STRUCTURE:
+    case MAINTENANCE_OUT:
+    case MAINTENANCE_IN:
+        if (dev->type == TYPE_ROM) {
+            /* GPCMD_REPORT_KEY and GPCMD_SEND_KEY from multi media commands */
+            cmd->xfer = buf[9] | (buf[8] << 8);
+        }
+        break;
+    case ATA_PASSTHROUGH_12:
+        if (dev->type == TYPE_ROM) {
+            /* BLANK command of MMC */
+            cmd->xfer = 0;
+        } else {
+            cmd->xfer = ata_passthrough_12_xfer(dev, buf);
+        }
+        break;
+    case ATA_PASSTHROUGH_16:
+        cmd->xfer = ata_passthrough_16_xfer(dev, buf);
+        break;
+    }
+    return 0;
+}
+
+static int scsi_req_stream_xfer(SCSICommand *cmd, SCSIDevice *dev, uint8_t *buf)
+{
+    switch (buf[0]) {
+    /* stream commands */
+    case ERASE_12:
+    case ERASE_16:
+        cmd->xfer = 0;
+        break;
+    case READ_6:
+    case READ_REVERSE:
+    case RECOVER_BUFFERED_DATA:
+    case WRITE_6:
+        cmd->xfer = buf[4] | (buf[3] << 8) | (buf[2] << 16);
+        if (buf[1] & 0x01) { /* fixed */
+            cmd->xfer *= dev->blocksize;
+        }
+        break;
+    case READ_16:
+    case READ_REVERSE_16:
+    case VERIFY_16:
+    case WRITE_16:
+        cmd->xfer = buf[14] | (buf[13] << 8) | (buf[12] << 16);
+        if (buf[1] & 0x01) { /* fixed */
+            cmd->xfer *= dev->blocksize;
+        }
+        break;
+    case REWIND:
+    case LOAD_UNLOAD:
+        cmd->xfer = 0;
+        break;
+    case SPACE_16:
+        cmd->xfer = buf[13] | (buf[12] << 8);
+        break;
+    case READ_POSITION:
+        switch (buf[1] & 0x1f) /* operation code */ {
+        case SHORT_FORM_BLOCK_ID:
+        case SHORT_FORM_VENDOR_SPECIFIC:
+            cmd->xfer = 20;
+            break;
+        case LONG_FORM:
+            cmd->xfer = 32;
+            break;
+        case EXTENDED_FORM:
+            cmd->xfer = buf[8] | (buf[7] << 8);
+            break;
+        default:
+            return -1;
+        }
+
+        break;
+    case FORMAT_UNIT:
+        cmd->xfer = buf[4] | (buf[3] << 8);
+        break;
+    /* generic commands */
+    default:
+        return scsi_req_xfer(cmd, dev, buf);
+    }
+    return 0;
+}
+
+static int scsi_req_medium_changer_xfer(SCSICommand *cmd, SCSIDevice *dev, uint8_t *buf)
+{
+    switch (buf[0]) {
+    /* medium changer commands */
+    case EXCHANGE_MEDIUM:
+    case INITIALIZE_ELEMENT_STATUS:
+    case INITIALIZE_ELEMENT_STATUS_WITH_RANGE:
+    case MOVE_MEDIUM:
+    case POSITION_TO_ELEMENT:
+        cmd->xfer = 0;
+        break;
+    case READ_ELEMENT_STATUS:
+        cmd->xfer = buf[9] | (buf[8] << 8) | (buf[7] << 16);
+        break;
+
+    /* generic commands */
+    default:
+        return scsi_req_xfer(cmd, dev, buf);
+    }
+    return 0;
+}
+
+
+static void scsi_cmd_xfer_mode(SCSICommand *cmd)
+{
+    if (!cmd->xfer) {
+        cmd->mode = SCSI_XFER_NONE;
+        return;
+    }
+    switch (cmd->buf[0]) {
+    case WRITE_6:
+    case WRITE_10:
+    case WRITE_VERIFY_10:
+    case WRITE_12:
+    case WRITE_VERIFY_12:
+    case WRITE_16:
+    case WRITE_VERIFY_16:
+    case VERIFY_10:
+    case VERIFY_12:
+    case VERIFY_16:
+    case COPY:
+    case COPY_VERIFY:
+    case COMPARE:
+    case CHANGE_DEFINITION:
+    case LOG_SELECT:
+    case MODE_SELECT:
+    case MODE_SELECT_10:
+    case SEND_DIAGNOSTIC:
+    case WRITE_BUFFER:
+    case FORMAT_UNIT:
+    case REASSIGN_BLOCKS:
+    case SEARCH_EQUAL:
+    case SEARCH_HIGH:
+    case SEARCH_LOW:
+    case UPDATE_BLOCK:
+    case WRITE_LONG_10:
+    case WRITE_SAME_10:
+    case WRITE_SAME_16:
+    case UNMAP:
+    case SEARCH_HIGH_12:
+    case SEARCH_EQUAL_12:
+    case SEARCH_LOW_12:
+    case MEDIUM_SCAN:
+    case SEND_VOLUME_TAG:
+    case SEND_CUE_SHEET:
+    case SEND_DVD_STRUCTURE:
+    case PERSISTENT_RESERVE_OUT:
+    case MAINTENANCE_OUT:
+        cmd->mode = SCSI_XFER_TO_DEV;
+        break;
+    case ATA_PASSTHROUGH_12:
+    case ATA_PASSTHROUGH_16:
+        /* T_DIR */
+        cmd->mode = (cmd->buf[2] & 0x8) ?
+                   SCSI_XFER_FROM_DEV : SCSI_XFER_TO_DEV;
+        break;
+    default:
+        cmd->mode = SCSI_XFER_FROM_DEV;
+        break;
+    }
+}
+
+static uint64_t scsi_cmd_lba(SCSICommand *cmd)
+{
+    uint8_t *buf = cmd->buf;
+    uint64_t lba;
+
+    switch (buf[0] >> 5) {
+    case 0:
+        lba = ldl_be_p(&buf[0]) & 0x1fffff;
+        break;
+    case 1:
+    case 2:
+    case 5:
+        lba = ldl_be_p(&buf[2]) & 0xffffffffULL;
+        break;
+    case 4:
+        lba = ldq_be_p(&buf[2]);
+        break;
+    default:
+        lba = -1;
+
+    }
+    return lba;
+}
+
+int scsi_cdb_length(uint8_t *buf) {
+    int cdb_len;
+
+    switch (buf[0] >> 5) {
+    case 0:
+        cdb_len = 6;
+        break;
+    case 1:
+    case 2:
+        cdb_len = 10;
+        break;
+    case 4:
+        cdb_len = 16;
+        break;
+    case 5:
+        cdb_len = 12;
+        break;
+    default:
+        cdb_len = -1;
+    }
+    return cdb_len;
+}
+
+int scsi_req_parse_cdb(SCSIDevice *dev, SCSICommand *cmd, uint8_t *buf)
+{
+    int rc;
+    int len;
+
+    cmd->lba = -1;
+    len = scsi_cdb_length(buf);
+    if (len < 0) {
+        return -1;
+    }
+
+    cmd->len = len;
+    switch (dev->type) {
+    case TYPE_TAPE:
+        rc = scsi_req_stream_xfer(cmd, dev, buf);
+        break;
+    case TYPE_MEDIUM_CHANGER:
+        rc = scsi_req_medium_changer_xfer(cmd, dev, buf);
+        break;
+    default:
+        rc = scsi_req_xfer(cmd, dev, buf);
+        break;
+    }
+
+    if (rc != 0)
+        return rc;
+
+    memcpy(cmd->buf, buf, cmd->len);
+    scsi_cmd_xfer_mode(cmd);
+    cmd->lba = scsi_cmd_lba(cmd);
+    return 0;
+}
+
+void scsi_device_report_change(SCSIDevice *dev, SCSISense sense)
+{
+    SCSIBus *bus = DO_UPCAST(SCSIBus, qbus, dev->qdev.parent_bus);
+
+    scsi_device_set_ua(dev, sense);
+    if (bus->info->change) {
+        bus->info->change(bus, dev, sense);
+    }
+}
+
+/*
+ * Predefined sense codes
+ */
+
+/* No sense data available */
+const struct SCSISense sense_code_NO_SENSE = {
+    .key = NO_SENSE , .asc = 0x00 , .ascq = 0x00
+};
+
+/* LUN not ready, Manual intervention required */
+const struct SCSISense sense_code_LUN_NOT_READY = {
+    .key = NOT_READY, .asc = 0x04, .ascq = 0x03
+};
+
+/* LUN not ready, Medium not present */
+const struct SCSISense sense_code_NO_MEDIUM = {
+    .key = NOT_READY, .asc = 0x3a, .ascq = 0x00
+};
+
+/* LUN not ready, medium removal prevented */
+const struct SCSISense sense_code_NOT_READY_REMOVAL_PREVENTED = {
+    .key = NOT_READY, .asc = 0x53, .ascq = 0x02
+};
+
+/* Hardware error, internal target failure */
+const struct SCSISense sense_code_TARGET_FAILURE = {
+    .key = HARDWARE_ERROR, .asc = 0x44, .ascq = 0x00
+};
+
+/* Illegal request, invalid command operation code */
+const struct SCSISense sense_code_INVALID_OPCODE = {
+    .key = ILLEGAL_REQUEST, .asc = 0x20, .ascq = 0x00
+};
+
+/* Illegal request, LBA out of range */
+const struct SCSISense sense_code_LBA_OUT_OF_RANGE = {
+    .key = ILLEGAL_REQUEST, .asc = 0x21, .ascq = 0x00
+};
+
+/* Illegal request, Invalid field in CDB */
+const struct SCSISense sense_code_INVALID_FIELD = {
+    .key = ILLEGAL_REQUEST, .asc = 0x24, .ascq = 0x00
+};
+
+/* Illegal request, Invalid field in parameter list */
+const struct SCSISense sense_code_INVALID_PARAM = {
+    .key = ILLEGAL_REQUEST, .asc = 0x26, .ascq = 0x00
+};
+
+/* Illegal request, Parameter list length error */
+const struct SCSISense sense_code_INVALID_PARAM_LEN = {
+    .key = ILLEGAL_REQUEST, .asc = 0x1a, .ascq = 0x00
+};
+
+/* Illegal request, LUN not supported */
+const struct SCSISense sense_code_LUN_NOT_SUPPORTED = {
+    .key = ILLEGAL_REQUEST, .asc = 0x25, .ascq = 0x00
+};
+
+/* Illegal request, Saving parameters not supported */
+const struct SCSISense sense_code_SAVING_PARAMS_NOT_SUPPORTED = {
+    .key = ILLEGAL_REQUEST, .asc = 0x39, .ascq = 0x00
+};
+
+/* Illegal request, Incompatible medium installed */
+const struct SCSISense sense_code_INCOMPATIBLE_FORMAT = {
+    .key = ILLEGAL_REQUEST, .asc = 0x30, .ascq = 0x00
+};
+
+/* Illegal request, medium removal prevented */
+const struct SCSISense sense_code_ILLEGAL_REQ_REMOVAL_PREVENTED = {
+    .key = ILLEGAL_REQUEST, .asc = 0x53, .ascq = 0x02
+};
+
+/* Illegal request, Invalid Transfer Tag */
+const struct SCSISense sense_code_INVALID_TAG = {
+    .key = ILLEGAL_REQUEST, .asc = 0x4b, .ascq = 0x01
+};
+
+/* Command aborted, I/O process terminated */
+const struct SCSISense sense_code_IO_ERROR = {
+    .key = ABORTED_COMMAND, .asc = 0x00, .ascq = 0x06
+};
+
+/* Command aborted, I_T Nexus loss occurred */
+const struct SCSISense sense_code_I_T_NEXUS_LOSS = {
+    .key = ABORTED_COMMAND, .asc = 0x29, .ascq = 0x07
+};
+
+/* Command aborted, Logical Unit failure */
+const struct SCSISense sense_code_LUN_FAILURE = {
+    .key = ABORTED_COMMAND, .asc = 0x3e, .ascq = 0x01
+};
+
+/* Command aborted, Overlapped Commands Attempted */
+const struct SCSISense sense_code_OVERLAPPED_COMMANDS = {
+    .key = ABORTED_COMMAND, .asc = 0x4e, .ascq = 0x00
+};
+
+/* Unit attention, Capacity data has changed */
+const struct SCSISense sense_code_CAPACITY_CHANGED = {
+    .key = UNIT_ATTENTION, .asc = 0x2a, .ascq = 0x09
+};
+
+/* Unit attention, Power on, reset or bus device reset occurred */
+const struct SCSISense sense_code_RESET = {
+    .key = UNIT_ATTENTION, .asc = 0x29, .ascq = 0x00
+};
+
+/* Unit attention, No medium */
+const struct SCSISense sense_code_UNIT_ATTENTION_NO_MEDIUM = {
+    .key = UNIT_ATTENTION, .asc = 0x3a, .ascq = 0x00
+};
+
+/* Unit attention, Medium may have changed */
+const struct SCSISense sense_code_MEDIUM_CHANGED = {
+    .key = UNIT_ATTENTION, .asc = 0x28, .ascq = 0x00
+};
+
+/* Unit attention, Reported LUNs data has changed */
+const struct SCSISense sense_code_REPORTED_LUNS_CHANGED = {
+    .key = UNIT_ATTENTION, .asc = 0x3f, .ascq = 0x0e
+};
+
+/* Unit attention, Device internal reset */
+const struct SCSISense sense_code_DEVICE_INTERNAL_RESET = {
+    .key = UNIT_ATTENTION, .asc = 0x29, .ascq = 0x04
+};
+
+/* Data Protection, Write Protected */
+const struct SCSISense sense_code_WRITE_PROTECTED = {
+    .key = DATA_PROTECT, .asc = 0x27, .ascq = 0x00
+};
+
+/* Data Protection, Space Allocation Failed Write Protect */
+const struct SCSISense sense_code_SPACE_ALLOC_FAILED = {
+    .key = DATA_PROTECT, .asc = 0x27, .ascq = 0x07
+};
+
+/*
+ * scsi_build_sense
+ *
+ * Convert between fixed and descriptor sense buffers
+ */
+int scsi_build_sense(uint8_t *in_buf, int in_len,
+                     uint8_t *buf, int len, bool fixed)
+{
+    bool fixed_in;
+    SCSISense sense;
+    if (!fixed && len < 8) {
+        return 0;
+    }
+
+    if (in_len == 0) {
+        sense.key = NO_SENSE;
+        sense.asc = 0;
+        sense.ascq = 0;
+    } else {
+        fixed_in = (in_buf[0] & 2) == 0;
+
+        if (fixed == fixed_in) {
+            memcpy(buf, in_buf, MIN(len, in_len));
+            return MIN(len, in_len);
+        }
+
+        if (fixed_in) {
+            sense.key = in_buf[2];
+            sense.asc = in_buf[12];
+            sense.ascq = in_buf[13];
+        } else {
+            sense.key = in_buf[1];
+            sense.asc = in_buf[2];
+            sense.ascq = in_buf[3];
+        }
+    }
+
+    memset(buf, 0, len);
+    if (fixed) {
+        /* Return fixed format sense buffer */
+        buf[0] = 0x70;
+        buf[2] = sense.key;
+        buf[7] = 10;
+        buf[12] = sense.asc;
+        buf[13] = sense.ascq;
+        return MIN(len, SCSI_SENSE_LEN);
+    } else {
+        /* Return descriptor format sense buffer */
+        buf[0] = 0x72;
+        buf[1] = sense.key;
+        buf[2] = sense.asc;
+        buf[3] = sense.ascq;
+        return 8;
+    }
+}
+
+const char *scsi_command_name(uint8_t cmd)
+{
+    static const char *names[] = {
+        [ TEST_UNIT_READY          ] = "TEST_UNIT_READY",
+        [ REWIND                   ] = "REWIND",
+        [ REQUEST_SENSE            ] = "REQUEST_SENSE",
+        [ FORMAT_UNIT              ] = "FORMAT_UNIT",
+        [ READ_BLOCK_LIMITS        ] = "READ_BLOCK_LIMITS",
+        [ REASSIGN_BLOCKS          ] = "REASSIGN_BLOCKS/INITIALIZE ELEMENT STATUS",
+        /* LOAD_UNLOAD and INITIALIZE_ELEMENT_STATUS use the same operation code */
+        [ READ_6                   ] = "READ_6",
+        [ WRITE_6                  ] = "WRITE_6",
+        [ SET_CAPACITY             ] = "SET_CAPACITY",
+        [ READ_REVERSE             ] = "READ_REVERSE",
+        [ WRITE_FILEMARKS          ] = "WRITE_FILEMARKS",
+        [ SPACE                    ] = "SPACE",
+        [ INQUIRY                  ] = "INQUIRY",
+        [ RECOVER_BUFFERED_DATA    ] = "RECOVER_BUFFERED_DATA",
+        [ MAINTENANCE_IN           ] = "MAINTENANCE_IN",
+        [ MAINTENANCE_OUT          ] = "MAINTENANCE_OUT",
+        [ MODE_SELECT              ] = "MODE_SELECT",
+        [ RESERVE                  ] = "RESERVE",
+        [ RELEASE                  ] = "RELEASE",
+        [ COPY                     ] = "COPY",
+        [ ERASE                    ] = "ERASE",
+        [ MODE_SENSE               ] = "MODE_SENSE",
+        [ START_STOP               ] = "START_STOP/LOAD_UNLOAD",
+        /* LOAD_UNLOAD and START_STOP use the same operation code */
+        [ RECEIVE_DIAGNOSTIC       ] = "RECEIVE_DIAGNOSTIC",
+        [ SEND_DIAGNOSTIC          ] = "SEND_DIAGNOSTIC",
+        [ ALLOW_MEDIUM_REMOVAL     ] = "ALLOW_MEDIUM_REMOVAL",
+        [ READ_CAPACITY_10         ] = "READ_CAPACITY_10",
+        [ READ_10                  ] = "READ_10",
+        [ WRITE_10                 ] = "WRITE_10",
+        [ SEEK_10                  ] = "SEEK_10/POSITION_TO_ELEMENT",
+        /* SEEK_10 and POSITION_TO_ELEMENT use the same operation code */
+        [ WRITE_VERIFY_10          ] = "WRITE_VERIFY_10",
+        [ VERIFY_10                ] = "VERIFY_10",
+        [ SEARCH_HIGH              ] = "SEARCH_HIGH",
+        [ SEARCH_EQUAL             ] = "SEARCH_EQUAL",
+        [ SEARCH_LOW               ] = "SEARCH_LOW",
+        [ SET_LIMITS               ] = "SET_LIMITS",
+        [ PRE_FETCH                ] = "PRE_FETCH/READ_POSITION",
+        /* READ_POSITION and PRE_FETCH use the same operation code */
+        [ SYNCHRONIZE_CACHE        ] = "SYNCHRONIZE_CACHE",
+        [ LOCK_UNLOCK_CACHE        ] = "LOCK_UNLOCK_CACHE",
+        [ READ_DEFECT_DATA         ] = "READ_DEFECT_DATA/INITIALIZE_ELEMENT_STATUS_WITH_RANGE",
+        /* READ_DEFECT_DATA and INITIALIZE_ELEMENT_STATUS_WITH_RANGE use the same operation code */
+        [ MEDIUM_SCAN              ] = "MEDIUM_SCAN",
+        [ COMPARE                  ] = "COMPARE",
+        [ COPY_VERIFY              ] = "COPY_VERIFY",
+        [ WRITE_BUFFER             ] = "WRITE_BUFFER",
+        [ READ_BUFFER              ] = "READ_BUFFER",
+        [ UPDATE_BLOCK             ] = "UPDATE_BLOCK",
+        [ READ_LONG_10             ] = "READ_LONG_10",
+        [ WRITE_LONG_10            ] = "WRITE_LONG_10",
+        [ CHANGE_DEFINITION        ] = "CHANGE_DEFINITION",
+        [ WRITE_SAME_10            ] = "WRITE_SAME_10",
+        [ UNMAP                    ] = "UNMAP",
+        [ READ_TOC                 ] = "READ_TOC",
+        [ REPORT_DENSITY_SUPPORT   ] = "REPORT_DENSITY_SUPPORT",
+        [ SANITIZE                 ] = "SANITIZE",
+        [ GET_CONFIGURATION        ] = "GET_CONFIGURATION",
+        [ LOG_SELECT               ] = "LOG_SELECT",
+        [ LOG_SENSE                ] = "LOG_SENSE",
+        [ MODE_SELECT_10           ] = "MODE_SELECT_10",
+        [ RESERVE_10               ] = "RESERVE_10",
+        [ RELEASE_10               ] = "RELEASE_10",
+        [ MODE_SENSE_10            ] = "MODE_SENSE_10",
+        [ PERSISTENT_RESERVE_IN    ] = "PERSISTENT_RESERVE_IN",
+        [ PERSISTENT_RESERVE_OUT   ] = "PERSISTENT_RESERVE_OUT",
+        [ WRITE_FILEMARKS_16       ] = "WRITE_FILEMARKS_16",
+        [ EXTENDED_COPY            ] = "EXTENDED_COPY",
+        [ ATA_PASSTHROUGH_16       ] = "ATA_PASSTHROUGH_16",
+        [ ACCESS_CONTROL_IN        ] = "ACCESS_CONTROL_IN",
+        [ ACCESS_CONTROL_OUT       ] = "ACCESS_CONTROL_OUT",
+        [ READ_16                  ] = "READ_16",
+        [ COMPARE_AND_WRITE        ] = "COMPARE_AND_WRITE",
+        [ WRITE_16                 ] = "WRITE_16",
+        [ WRITE_VERIFY_16          ] = "WRITE_VERIFY_16",
+        [ VERIFY_16                ] = "VERIFY_16",
+        [ PRE_FETCH_16             ] = "PRE_FETCH_16",
+        [ SYNCHRONIZE_CACHE_16     ] = "SPACE_16/SYNCHRONIZE_CACHE_16",
+        /* SPACE_16 and SYNCHRONIZE_CACHE_16 use the same operation code */
+        [ LOCATE_16                ] = "LOCATE_16",
+        [ WRITE_SAME_16            ] = "ERASE_16/WRITE_SAME_16",
+        /* ERASE_16 and WRITE_SAME_16 use the same operation code */
+        [ SERVICE_ACTION_IN_16     ] = "SERVICE_ACTION_IN_16",
+        [ WRITE_LONG_16            ] = "WRITE_LONG_16",
+        [ REPORT_LUNS              ] = "REPORT_LUNS",
+        [ ATA_PASSTHROUGH_12       ] = "BLANK/ATA_PASSTHROUGH_12",
+        [ MOVE_MEDIUM              ] = "MOVE_MEDIUM",
+        [ EXCHANGE_MEDIUM          ] = "EXCHANGE MEDIUM",
+        [ READ_12                  ] = "READ_12",
+        [ WRITE_12                 ] = "WRITE_12",
+        [ ERASE_12                 ] = "ERASE_12/GET_PERFORMANCE",
+        /* ERASE_12 and GET_PERFORMANCE use the same operation code */
+        [ SERVICE_ACTION_IN_12     ] = "SERVICE_ACTION_IN_12",
+        [ WRITE_VERIFY_12          ] = "WRITE_VERIFY_12",
+        [ VERIFY_12                ] = "VERIFY_12",
+        [ SEARCH_HIGH_12           ] = "SEARCH_HIGH_12",
+        [ SEARCH_EQUAL_12          ] = "SEARCH_EQUAL_12",
+        [ SEARCH_LOW_12            ] = "SEARCH_LOW_12",
+        [ READ_ELEMENT_STATUS      ] = "READ_ELEMENT_STATUS",
+        [ SEND_VOLUME_TAG          ] = "SEND_VOLUME_TAG/SET_STREAMING",
+        /* SEND_VOLUME_TAG and SET_STREAMING use the same operation code */
+        [ READ_CD                  ] = "READ_CD",
+        [ READ_DEFECT_DATA_12      ] = "READ_DEFECT_DATA_12",
+        [ READ_DVD_STRUCTURE       ] = "READ_DVD_STRUCTURE",
+        [ RESERVE_TRACK            ] = "RESERVE_TRACK",
+        [ SEND_CUE_SHEET           ] = "SEND_CUE_SHEET",
+        [ SEND_DVD_STRUCTURE       ] = "SEND_DVD_STRUCTURE",
+        [ SET_CD_SPEED             ] = "SET_CD_SPEED",
+        [ SET_READ_AHEAD           ] = "SET_READ_AHEAD",
+        [ ALLOW_OVERWRITE          ] = "ALLOW_OVERWRITE",
+        [ MECHANISM_STATUS         ] = "MECHANISM_STATUS",
+        [ GET_EVENT_STATUS_NOTIFICATION ] = "GET_EVENT_STATUS_NOTIFICATION",
+        [ READ_DISC_INFORMATION    ] = "READ_DISC_INFORMATION",
+    };
+
+    if (cmd >= ARRAY_SIZE(names) || names[cmd] == NULL)
+        return "*UNKNOWN*";
+    return names[cmd];
+}
+
+SCSIRequest *scsi_req_ref(SCSIRequest *req)
+{
+    assert(req->refcount > 0);
+    req->refcount++;
+    return req;
+}
+
+void scsi_req_unref(SCSIRequest *req)
+{
+    assert(req->refcount > 0);
+    if (--req->refcount == 0) {
+        BusState *qbus = req->dev->qdev.parent_bus;
+        SCSIBus *bus = DO_UPCAST(SCSIBus, qbus, qbus);
+
+        if (bus->info->free_request && req->hba_private) {
+            bus->info->free_request(bus, req->hba_private);
+        }
+        if (req->ops->free_req) {
+            req->ops->free_req(req);
+        }
+        object_unref(OBJECT(req->dev));
+        object_unref(OBJECT(qbus->parent));
+        g_slice_free1(req->ops->size, req);
+    }
+}
+
+/* Tell the device that we finished processing this chunk of I/O.  It
+   will start the next chunk or complete the command.  */
+void scsi_req_continue(SCSIRequest *req)
+{
+    if (req->io_canceled) {
+        trace_scsi_req_continue_canceled(req->dev->id, req->lun, req->tag);
+        return;
+    }
+    trace_scsi_req_continue(req->dev->id, req->lun, req->tag);
+    if (req->cmd.mode == SCSI_XFER_TO_DEV) {
+        req->ops->write_data(req);
+    } else {
+        req->ops->read_data(req);
+    }
+}
+
+/* Called by the devices when data is ready for the HBA.  The HBA should
+   start a DMA operation to read or fill the device's data buffer.
+   Once it completes, calling scsi_req_continue will restart I/O.  */
+void scsi_req_data(SCSIRequest *req, int len)
+{
+    uint8_t *buf;
+    if (req->io_canceled) {
+        trace_scsi_req_data_canceled(req->dev->id, req->lun, req->tag, len);
+        return;
+    }
+    trace_scsi_req_data(req->dev->id, req->lun, req->tag, len);
+    assert(req->cmd.mode != SCSI_XFER_NONE);
+    if (!req->sg) {
+        req->resid -= len;
+        req->bus->info->transfer_data(req, len);
+        return;
+    }
+
+    /* If the device calls scsi_req_data and the HBA specified a
+     * scatter/gather list, the transfer has to happen in a single
+     * step.  */
+    assert(!req->dma_started);
+    req->dma_started = true;
+
+    buf = scsi_req_get_buf(req);
+    if (req->cmd.mode == SCSI_XFER_FROM_DEV) {
+        req->resid = dma_buf_read(buf, len, req->sg);
+    } else {
+        req->resid = dma_buf_write(buf, len, req->sg);
+    }
+    scsi_req_continue(req);
+}
+
+void scsi_req_print(SCSIRequest *req)
+{
+    FILE *fp = stderr;
+    int i;
+
+    fprintf(fp, "[%s id=%d] %s",
+            req->dev->qdev.parent_bus->name,
+            req->dev->id,
+            scsi_command_name(req->cmd.buf[0]));
+    for (i = 1; i < req->cmd.len; i++) {
+        fprintf(fp, " 0x%02x", req->cmd.buf[i]);
+    }
+    switch (req->cmd.mode) {
+    case SCSI_XFER_NONE:
+        fprintf(fp, " - none\n");
+        break;
+    case SCSI_XFER_FROM_DEV:
+        fprintf(fp, " - from-dev len=%zd\n", req->cmd.xfer);
+        break;
+    case SCSI_XFER_TO_DEV:
+        fprintf(fp, " - to-dev len=%zd\n", req->cmd.xfer);
+        break;
+    default:
+        fprintf(fp, " - Oops\n");
+        break;
+    }
+}
+
+void scsi_req_complete(SCSIRequest *req, int status)
+{
+    assert(req->status == -1);
+    req->status = status;
+
+    assert(req->sense_len <= sizeof(req->sense));
+    if (status == GOOD) {
+        req->sense_len = 0;
+    }
+
+    if (req->sense_len) {
+        memcpy(req->dev->sense, req->sense, req->sense_len);
+        req->dev->sense_len = req->sense_len;
+        req->dev->sense_is_ua = (req->ops == &reqops_unit_attention);
+    } else {
+        req->dev->sense_len = 0;
+        req->dev->sense_is_ua = false;
+    }
+
+    /*
+     * Unit attention state is now stored in the device's sense buffer
+     * if the HBA didn't do autosense.  Clear the pending unit attention
+     * flags.
+     */
+    scsi_clear_unit_attention(req);
+
+    scsi_req_ref(req);
+    scsi_req_dequeue(req);
+    req->bus->info->complete(req, req->status, req->resid);
+
+    /* Cancelled requests might end up being completed instead of cancelled */
+    notifier_list_notify(&req->cancel_notifiers, req);
+    scsi_req_unref(req);
+}
+
+/* Called by the devices when the request is canceled. */
+void scsi_req_cancel_complete(SCSIRequest *req)
+{
+    assert(req->io_canceled);
+    if (req->bus->info->cancel) {
+        req->bus->info->cancel(req);
+    }
+    notifier_list_notify(&req->cancel_notifiers, req);
+    scsi_req_unref(req);
+}
+
+/* Cancel @req asynchronously. @notifier is added to @req's cancellation
+ * notifier list, the bus will be notified the requests cancellation is
+ * completed.
+ * */
+void scsi_req_cancel_async(SCSIRequest *req, Notifier *notifier)
+{
+    trace_scsi_req_cancel(req->dev->id, req->lun, req->tag);
+    if (notifier) {
+        notifier_list_add(&req->cancel_notifiers, notifier);
+    }
+    if (req->io_canceled) {
+        return;
+    }
+    scsi_req_ref(req);
+    scsi_req_dequeue(req);
+    req->io_canceled = true;
+    if (req->aiocb) {
+        blk_aio_cancel_async(req->aiocb);
+    } else {
+        scsi_req_cancel_complete(req);
+    }
+}
+
+void scsi_req_cancel(SCSIRequest *req)
+{
+    trace_scsi_req_cancel(req->dev->id, req->lun, req->tag);
+    if (!req->enqueued) {
+        return;
+    }
+    scsi_req_ref(req);
+    scsi_req_dequeue(req);
+    req->io_canceled = true;
+    if (req->aiocb) {
+        blk_aio_cancel(req->aiocb);
+    } else {
+        scsi_req_cancel_complete(req);
+    }
+}
+
+static int scsi_ua_precedence(SCSISense sense)
+{
+    if (sense.key != UNIT_ATTENTION) {
+        return INT_MAX;
+    }
+    if (sense.asc == 0x29 && sense.ascq == 0x04) {
+        /* DEVICE INTERNAL RESET goes with POWER ON OCCURRED */
+        return 1;
+    } else if (sense.asc == 0x3F && sense.ascq == 0x01) {
+        /* MICROCODE HAS BEEN CHANGED goes with SCSI BUS RESET OCCURRED */
+        return 2;
+    } else if (sense.asc == 0x29 && (sense.ascq == 0x05 || sense.ascq == 0x06)) {
+        /* These two go with "all others". */
+        ;
+    } else if (sense.asc == 0x29 && sense.ascq <= 0x07) {
+        /* POWER ON, RESET OR BUS DEVICE RESET OCCURRED = 0
+         * POWER ON OCCURRED = 1
+         * SCSI BUS RESET OCCURRED = 2
+         * BUS DEVICE RESET FUNCTION OCCURRED = 3
+         * I_T NEXUS LOSS OCCURRED = 7
+         */
+        return sense.ascq;
+    } else if (sense.asc == 0x2F && sense.ascq == 0x01) {
+        /* COMMANDS CLEARED BY POWER LOSS NOTIFICATION  */
+        return 8;
+    }
+    return (sense.asc << 8) | sense.ascq;
+}
+
+void scsi_device_set_ua(SCSIDevice *sdev, SCSISense sense)
+{
+    int prec1, prec2;
+    if (sense.key != UNIT_ATTENTION) {
+        return;
+    }
+    trace_scsi_device_set_ua(sdev->id, sdev->lun, sense.key,
+                             sense.asc, sense.ascq);
+
+    /*
+     * Override a pre-existing unit attention condition, except for a more
+     * important reset condition.
+    */
+    prec1 = scsi_ua_precedence(sdev->unit_attention);
+    prec2 = scsi_ua_precedence(sense);
+    if (prec2 < prec1) {
+        sdev->unit_attention = sense;
+    }
+}
+
+void scsi_device_purge_requests(SCSIDevice *sdev, SCSISense sense)
+{
+    SCSIRequest *req;
+
+    while (!QTAILQ_EMPTY(&sdev->requests)) {
+        req = QTAILQ_FIRST(&sdev->requests);
+        scsi_req_cancel(req);
+    }
+
+    scsi_device_set_ua(sdev, sense);
+}
+
+static char *scsibus_get_dev_path(DeviceState *dev)
+{
+    SCSIDevice *d = DO_UPCAST(SCSIDevice, qdev, dev);
+    DeviceState *hba = dev->parent_bus->parent;
+    char *id;
+    char *path;
+
+    id = qdev_get_dev_path(hba);
+    if (id) {
+        path = g_strdup_printf("%s/%d:%d:%d", id, d->channel, d->id, d->lun);
+    } else {
+        path = g_strdup_printf("%d:%d:%d", d->channel, d->id, d->lun);
+    }
+    g_free(id);
+    return path;
+}
+
+static char *scsibus_get_fw_dev_path(DeviceState *dev)
+{
+    SCSIDevice *d = SCSI_DEVICE(dev);
+    return g_strdup_printf("channel@%x/%s@%x,%x", d->channel,
+                           qdev_fw_name(dev), d->id, d->lun);
+}
+
+SCSIDevice *scsi_device_find(SCSIBus *bus, int channel, int id, int lun)
+{
+    BusChild *kid;
+    SCSIDevice *target_dev = NULL;
+
+    QTAILQ_FOREACH_REVERSE(kid, &bus->qbus.children, ChildrenHead, sibling) {
+        DeviceState *qdev = kid->child;
+        SCSIDevice *dev = SCSI_DEVICE(qdev);
+
+        if (dev->channel == channel && dev->id == id) {
+            if (dev->lun == lun) {
+                return dev;
+            }
+            target_dev = dev;
+        }
+    }
+    return target_dev;
+}
+
+/* SCSI request list.  For simplicity, pv points to the whole device */
+
+static void put_scsi_requests(QEMUFile *f, void *pv, size_t size)
+{
+    SCSIDevice *s = pv;
+    SCSIBus *bus = DO_UPCAST(SCSIBus, qbus, s->qdev.parent_bus);
+    SCSIRequest *req;
+
+    QTAILQ_FOREACH(req, &s->requests, next) {
+        assert(!req->io_canceled);
+        assert(req->status == -1);
+        assert(req->enqueued);
+
+        qemu_put_sbyte(f, req->retry ? 1 : 2);
+        qemu_put_buffer(f, req->cmd.buf, sizeof(req->cmd.buf));
+        qemu_put_be32s(f, &req->tag);
+        qemu_put_be32s(f, &req->lun);
+        if (bus->info->save_request) {
+            bus->info->save_request(f, req);
+        }
+        if (req->ops->save_request) {
+            req->ops->save_request(f, req);
+        }
+    }
+    qemu_put_sbyte(f, 0);
+}
+
+static int get_scsi_requests(QEMUFile *f, void *pv, size_t size)
+{
+    SCSIDevice *s = pv;
+    SCSIBus *bus = DO_UPCAST(SCSIBus, qbus, s->qdev.parent_bus);
+    int8_t sbyte;
+
+    while ((sbyte = qemu_get_sbyte(f)) > 0) {
+        uint8_t buf[SCSI_CMD_BUF_SIZE];
+        uint32_t tag;
+        uint32_t lun;
+        SCSIRequest *req;
+
+        qemu_get_buffer(f, buf, sizeof(buf));
+        qemu_get_be32s(f, &tag);
+        qemu_get_be32s(f, &lun);
+        req = scsi_req_new(s, tag, lun, buf, NULL);
+        req->retry = (sbyte == 1);
+        if (bus->info->load_request) {
+            req->hba_private = bus->info->load_request(f, req);
+        }
+        if (req->ops->load_request) {
+            req->ops->load_request(f, req);
+        }
+
+        /* Just restart it later.  */
+        scsi_req_enqueue_internal(req);
+
+        /* At this point, the request will be kept alive by the reference
+         * added by scsi_req_enqueue_internal, so we can release our reference.
+         * The HBA of course will add its own reference in the load_request
+         * callback if it needs to hold on the SCSIRequest.
+         */
+        scsi_req_unref(req);
+    }
+
+    return 0;
+}
+
+static const VMStateInfo vmstate_info_scsi_requests = {
+    .name = "scsi-requests",
+    .get  = get_scsi_requests,
+    .put  = put_scsi_requests,
+};
+
+static bool scsi_sense_state_needed(void *opaque)
+{
+    SCSIDevice *s = opaque;
+
+    return s->sense_len > SCSI_SENSE_BUF_SIZE_OLD;
+}
+
+static const VMStateDescription vmstate_scsi_sense_state = {
+    .name = "SCSIDevice/sense",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = scsi_sense_state_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT8_SUB_ARRAY(sense, SCSIDevice,
+                                SCSI_SENSE_BUF_SIZE_OLD,
+                                SCSI_SENSE_BUF_SIZE - SCSI_SENSE_BUF_SIZE_OLD),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+const VMStateDescription vmstate_scsi_device = {
+    .name = "SCSIDevice",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT8(unit_attention.key, SCSIDevice),
+        VMSTATE_UINT8(unit_attention.asc, SCSIDevice),
+        VMSTATE_UINT8(unit_attention.ascq, SCSIDevice),
+        VMSTATE_BOOL(sense_is_ua, SCSIDevice),
+        VMSTATE_UINT8_SUB_ARRAY(sense, SCSIDevice, 0, SCSI_SENSE_BUF_SIZE_OLD),
+        VMSTATE_UINT32(sense_len, SCSIDevice),
+        {
+            .name         = "requests",
+            .version_id   = 0,
+            .field_exists = NULL,
+            .size         = 0,   /* ouch */
+            .info         = &vmstate_info_scsi_requests,
+            .flags        = VMS_SINGLE,
+            .offset       = 0,
+        },
+        VMSTATE_END_OF_LIST()
+    },
+    .subsections = (const VMStateDescription*[]) {
+        &vmstate_scsi_sense_state,
+        NULL
+    }
+};
+
+static void scsi_device_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *k = DEVICE_CLASS(klass);
+    set_bit(DEVICE_CATEGORY_STORAGE, k->categories);
+    k->bus_type  = TYPE_SCSI_BUS;
+    k->realize   = scsi_qdev_realize;
+    k->unrealize = scsi_qdev_unrealize;
+    k->props     = scsi_props;
+}
+
+static void scsi_dev_instance_init(Object *obj)
+{
+    DeviceState *dev = DEVICE(obj);
+    SCSIDevice *s = DO_UPCAST(SCSIDevice, qdev, dev);
+
+    device_add_bootindex_property(obj, &s->conf.bootindex,
+                                  "bootindex", NULL,
+                                  &s->qdev, NULL);
+}
+
+static const TypeInfo scsi_device_type_info = {
+    .name = TYPE_SCSI_DEVICE,
+    .parent = TYPE_DEVICE,
+    .instance_size = sizeof(SCSIDevice),
+    .abstract = true,
+    .class_size = sizeof(SCSIDeviceClass),
+    .class_init = scsi_device_class_init,
+    .instance_init = scsi_dev_instance_init,
+};
+
+static void scsi_register_types(void)
+{
+    type_register_static(&scsi_bus_info);
+    type_register_static(&scsi_device_type_info);
+}
+
+type_init(scsi_register_types)
diff --git a/qemu/hw/scsi/scsi-disk.c b/qemu/hw/scsi/scsi-disk.c
new file mode 100644
index 000000000..64f069473
--- /dev/null
+++ b/qemu/hw/scsi/scsi-disk.c
@@ -0,0 +1,2776 @@
+/*
+ * SCSI Device emulation
+ *
+ * Copyright (c) 2006 CodeSourcery.
+ * Based on code by Fabrice Bellard
+ *
+ * Written by Paul Brook
+ * Modifications:
+ *  2009-Dec-12 Artyom Tarasenko : implemented stamdard inquiry for the case
+ *                                 when the allocation length of CDB is smaller
+ *                                 than 36.
+ *  2009-Oct-13 Artyom Tarasenko : implemented the block descriptor in the
+ *                                 MODE SENSE response.
+ *
+ * This code is licensed under the LGPL.
+ *
+ * Note that this file only handles the SCSI architecture model and device
+ * commands.  Emulation of interface/link layer protocols is handled by
+ * the host adapter emulator.
+ */
+
+//#define DEBUG_SCSI
+
+#ifdef DEBUG_SCSI
+#define DPRINTF(fmt, ...) \
+do { printf("scsi-disk: " fmt , ## __VA_ARGS__); } while (0)
+#else
+#define DPRINTF(fmt, ...) do {} while(0)
+#endif
+
+#include "qemu-common.h"
+#include "qemu/error-report.h"
+#include "hw/scsi/scsi.h"
+#include "block/scsi.h"
+#include "sysemu/sysemu.h"
+#include "sysemu/block-backend.h"
+#include "sysemu/blockdev.h"
+#include "hw/block/block.h"
+#include "sysemu/dma.h"
+
+#ifdef __linux
+#include <scsi/sg.h>
+#endif
+
+#define SCSI_WRITE_SAME_MAX         524288
+#define SCSI_DMA_BUF_SIZE           131072
+#define SCSI_MAX_INQUIRY_LEN        256
+#define SCSI_MAX_MODE_LEN           256
+
+#define DEFAULT_DISCARD_GRANULARITY 4096
+#define DEFAULT_MAX_UNMAP_SIZE      (1 << 30)   /* 1 GB */
+#define DEFAULT_MAX_IO_SIZE         INT_MAX     /* 2 GB - 1 block */
+
+typedef struct SCSIDiskState SCSIDiskState;
+
+typedef struct SCSIDiskReq {
+    SCSIRequest req;
+    /* Both sector and sector_count are in terms of qemu 512 byte blocks.  */
+    uint64_t sector;
+    uint32_t sector_count;
+    uint32_t buflen;
+    bool started;
+    struct iovec iov;
+    QEMUIOVector qiov;
+    BlockAcctCookie acct;
+} SCSIDiskReq;
+
+#define SCSI_DISK_F_REMOVABLE             0
+#define SCSI_DISK_F_DPOFUA                1
+#define SCSI_DISK_F_NO_REMOVABLE_DEVOPS   2
+
+struct SCSIDiskState
+{
+    SCSIDevice qdev;
+    uint32_t features;
+    bool media_changed;
+    bool media_event;
+    bool eject_request;
+    uint64_t wwn;
+    uint64_t port_wwn;
+    uint16_t port_index;
+    uint64_t max_unmap_size;
+    uint64_t max_io_size;
+    QEMUBH *bh;
+    char *version;
+    char *serial;
+    char *vendor;
+    char *product;
+    bool tray_open;
+    bool tray_locked;
+};
+
+static int scsi_handle_rw_error(SCSIDiskReq *r, int error);
+
+static void scsi_free_request(SCSIRequest *req)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+
+    qemu_vfree(r->iov.iov_base);
+}
+
+/* Helper function for command completion with sense.  */
+static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
+{
+    DPRINTF("Command complete tag=0x%x sense=%d/%d/%d\n",
+            r->req.tag, sense.key, sense.asc, sense.ascq);
+    scsi_req_build_sense(&r->req, sense);
+    scsi_req_complete(&r->req, CHECK_CONDITION);
+}
+
+static uint32_t scsi_init_iovec(SCSIDiskReq *r, size_t size)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+
+    if (!r->iov.iov_base) {
+        r->buflen = size;
+        r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
+    }
+    r->iov.iov_len = MIN(r->sector_count * 512, r->buflen);
+    qemu_iovec_init_external(&r->qiov, &r->iov, 1);
+    return r->qiov.size / 512;
+}
+
+static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+
+    qemu_put_be64s(f, &r->sector);
+    qemu_put_be32s(f, &r->sector_count);
+    qemu_put_be32s(f, &r->buflen);
+    if (r->buflen) {
+        if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+            qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
+        } else if (!req->retry) {
+            uint32_t len = r->iov.iov_len;
+            qemu_put_be32s(f, &len);
+            qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
+        }
+    }
+}
+
+static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+
+    qemu_get_be64s(f, &r->sector);
+    qemu_get_be32s(f, &r->sector_count);
+    qemu_get_be32s(f, &r->buflen);
+    if (r->buflen) {
+        scsi_init_iovec(r, r->buflen);
+        if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+            qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
+        } else if (!r->req.retry) {
+            uint32_t len;
+            qemu_get_be32s(f, &len);
+            r->iov.iov_len = len;
+            assert(r->iov.iov_len <= r->buflen);
+            qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
+        }
+    }
+
+    qemu_iovec_init_external(&r->qiov, &r->iov, 1);
+}
+
+static void scsi_aio_complete(void *opaque, int ret)
+{
+    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+
+    assert(r->req.aiocb != NULL);
+    r->req.aiocb = NULL;
+    block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+
+    if (ret < 0) {
+        if (scsi_handle_rw_error(r, -ret)) {
+            goto done;
+        }
+    }
+
+    scsi_req_complete(&r->req, GOOD);
+
+done:
+    scsi_req_unref(&r->req);
+}
+
+static bool scsi_is_cmd_fua(SCSICommand *cmd)
+{
+    switch (cmd->buf[0]) {
+    case READ_10:
+    case READ_12:
+    case READ_16:
+    case WRITE_10:
+    case WRITE_12:
+    case WRITE_16:
+        return (cmd->buf[1] & 8) != 0;
+
+    case VERIFY_10:
+    case VERIFY_12:
+    case VERIFY_16:
+    case WRITE_VERIFY_10:
+    case WRITE_VERIFY_12:
+    case WRITE_VERIFY_16:
+        return true;
+
+    case READ_6:
+    case WRITE_6:
+    default:
+        return false;
+    }
+}
+
+static void scsi_write_do_fua(SCSIDiskReq *r)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+
+    if (scsi_is_cmd_fua(&r->req.cmd)) {
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
+                         BLOCK_ACCT_FLUSH);
+        r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
+        return;
+    }
+
+    scsi_req_complete(&r->req, GOOD);
+
+done:
+    scsi_req_unref(&r->req);
+}
+
+static void scsi_dma_complete_noio(void *opaque, int ret)
+{
+    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+
+    if (r->req.aiocb != NULL) {
+        r->req.aiocb = NULL;
+        block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    }
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+
+    if (ret < 0) {
+        if (scsi_handle_rw_error(r, -ret)) {
+            goto done;
+        }
+    }
+
+    r->sector += r->sector_count;
+    r->sector_count = 0;
+    if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+        scsi_write_do_fua(r);
+        return;
+    } else {
+        scsi_req_complete(&r->req, GOOD);
+    }
+
+done:
+    scsi_req_unref(&r->req);
+}
+
+static void scsi_dma_complete(void *opaque, int ret)
+{
+    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
+
+    assert(r->req.aiocb != NULL);
+    scsi_dma_complete_noio(opaque, ret);
+}
+
+static void scsi_read_complete(void * opaque, int ret)
+{
+    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    int n;
+
+    assert(r->req.aiocb != NULL);
+    r->req.aiocb = NULL;
+    block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+
+    if (ret < 0) {
+        if (scsi_handle_rw_error(r, -ret)) {
+            goto done;
+        }
+    }
+
+    DPRINTF("Data ready tag=0x%x len=%zd\n", r->req.tag, r->qiov.size);
+
+    n = r->qiov.size / 512;
+    r->sector += n;
+    r->sector_count -= n;
+    scsi_req_data(&r->req, r->qiov.size);
+
+done:
+    scsi_req_unref(&r->req);
+}
+
+/* Actually issue a read to the block device.  */
+static void scsi_do_read(void *opaque, int ret)
+{
+    SCSIDiskReq *r = opaque;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    uint32_t n;
+
+    if (r->req.aiocb != NULL) {
+        r->req.aiocb = NULL;
+        block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    }
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+
+    if (ret < 0) {
+        if (scsi_handle_rw_error(r, -ret)) {
+            goto done;
+        }
+    }
+
+    /* The request is used as the AIO opaque value, so add a ref.  */
+    scsi_req_ref(&r->req);
+
+    if (r->req.sg) {
+        dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
+        r->req.resid -= r->req.sg->size;
+        r->req.aiocb = dma_blk_read(s->qdev.conf.blk, r->req.sg, r->sector,
+                                    scsi_dma_complete, r);
+    } else {
+        n = scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
+                         n * BDRV_SECTOR_SIZE, BLOCK_ACCT_READ);
+        r->req.aiocb = blk_aio_readv(s->qdev.conf.blk, r->sector, &r->qiov, n,
+                                     scsi_read_complete, r);
+    }
+
+done:
+    scsi_req_unref(&r->req);
+}
+
+/* Read more data from scsi device into buffer.  */
+static void scsi_read_data(SCSIRequest *req)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    bool first;
+
+    DPRINTF("Read sector_count=%d\n", r->sector_count);
+    if (r->sector_count == 0) {
+        /* This also clears the sense buffer for REQUEST SENSE.  */
+        scsi_req_complete(&r->req, GOOD);
+        return;
+    }
+
+    /* No data transfer may already be in progress */
+    assert(r->req.aiocb == NULL);
+
+    /* The request is used as the AIO opaque value, so add a ref.  */
+    scsi_req_ref(&r->req);
+    if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+        DPRINTF("Data transfer direction invalid\n");
+        scsi_read_complete(r, -EINVAL);
+        return;
+    }
+
+    if (s->tray_open) {
+        scsi_read_complete(r, -ENOMEDIUM);
+        return;
+    }
+
+    first = !r->started;
+    r->started = true;
+    if (first && scsi_is_cmd_fua(&r->req.cmd)) {
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
+                         BLOCK_ACCT_FLUSH);
+        r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read, r);
+    } else {
+        scsi_do_read(r, 0);
+    }
+}
+
+/*
+ * scsi_handle_rw_error has two return values.  0 means that the error
+ * must be ignored, 1 means that the error has been processed and the
+ * caller should not do anything else for this request.  Note that
+ * scsi_handle_rw_error always manages its reference counts, independent
+ * of the return value.
+ */
+static int scsi_handle_rw_error(SCSIDiskReq *r, int error)
+{
+    bool is_read = (r->req.cmd.xfer == SCSI_XFER_FROM_DEV);
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    BlockErrorAction action = blk_get_error_action(s->qdev.conf.blk,
+                                                   is_read, error);
+
+    if (action == BLOCK_ERROR_ACTION_REPORT) {
+        switch (error) {
+        case ENOMEDIUM:
+            scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
+            break;
+        case ENOMEM:
+            scsi_check_condition(r, SENSE_CODE(TARGET_FAILURE));
+            break;
+        case EINVAL:
+            scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+            break;
+        case ENOSPC:
+            scsi_check_condition(r, SENSE_CODE(SPACE_ALLOC_FAILED));
+            break;
+        default:
+            scsi_check_condition(r, SENSE_CODE(IO_ERROR));
+            break;
+        }
+    }
+    blk_error_action(s->qdev.conf.blk, action, is_read, error);
+    if (action == BLOCK_ERROR_ACTION_STOP) {
+        scsi_req_retry(&r->req);
+    }
+    return action != BLOCK_ERROR_ACTION_IGNORE;
+}
+
+static void scsi_write_complete(void * opaque, int ret)
+{
+    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    uint32_t n;
+
+    if (r->req.aiocb != NULL) {
+        r->req.aiocb = NULL;
+        block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    }
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+
+    if (ret < 0) {
+        if (scsi_handle_rw_error(r, -ret)) {
+            goto done;
+        }
+    }
+
+    n = r->qiov.size / 512;
+    r->sector += n;
+    r->sector_count -= n;
+    if (r->sector_count == 0) {
+        scsi_write_do_fua(r);
+        return;
+    } else {
+        scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
+        DPRINTF("Write complete tag=0x%x more=%zd\n", r->req.tag, r->qiov.size);
+        scsi_req_data(&r->req, r->qiov.size);
+    }
+
+done:
+    scsi_req_unref(&r->req);
+}
+
+static void scsi_write_data(SCSIRequest *req)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    uint32_t n;
+
+    /* No data transfer may already be in progress */
+    assert(r->req.aiocb == NULL);
+
+    /* The request is used as the AIO opaque value, so add a ref.  */
+    scsi_req_ref(&r->req);
+    if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
+        DPRINTF("Data transfer direction invalid\n");
+        scsi_write_complete(r, -EINVAL);
+        return;
+    }
+
+    if (!r->req.sg && !r->qiov.size) {
+        /* Called for the first time.  Ask the driver to send us more data.  */
+        r->started = true;
+        scsi_write_complete(r, 0);
+        return;
+    }
+    if (s->tray_open) {
+        scsi_write_complete(r, -ENOMEDIUM);
+        return;
+    }
+
+    if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
+        r->req.cmd.buf[0] == VERIFY_16) {
+        if (r->req.sg) {
+            scsi_dma_complete_noio(r, 0);
+        } else {
+            scsi_write_complete(r, 0);
+        }
+        return;
+    }
+
+    if (r->req.sg) {
+        dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
+        r->req.resid -= r->req.sg->size;
+        r->req.aiocb = dma_blk_write(s->qdev.conf.blk, r->req.sg, r->sector,
+                                     scsi_dma_complete, r);
+    } else {
+        n = r->qiov.size / 512;
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
+                         n * BDRV_SECTOR_SIZE, BLOCK_ACCT_WRITE);
+        r->req.aiocb = blk_aio_writev(s->qdev.conf.blk, r->sector, &r->qiov, n,
+                                      scsi_write_complete, r);
+    }
+}
+
+/* Return a pointer to the data buffer.  */
+static uint8_t *scsi_get_buf(SCSIRequest *req)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+
+    return (uint8_t *)r->iov.iov_base;
+}
+
+static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
+    int buflen = 0;
+    int start;
+
+    if (req->cmd.buf[1] & 0x1) {
+        /* Vital product data */
+        uint8_t page_code = req->cmd.buf[2];
+
+        outbuf[buflen++] = s->qdev.type & 0x1f;
+        outbuf[buflen++] = page_code ; // this page
+        outbuf[buflen++] = 0x00;
+        outbuf[buflen++] = 0x00;
+        start = buflen;
+
+        switch (page_code) {
+        case 0x00: /* Supported page codes, mandatory */
+        {
+            DPRINTF("Inquiry EVPD[Supported pages] "
+                    "buffer size %zd\n", req->cmd.xfer);
+            outbuf[buflen++] = 0x00; // list of supported pages (this page)
+            if (s->serial) {
+                outbuf[buflen++] = 0x80; // unit serial number
+            }
+            outbuf[buflen++] = 0x83; // device identification
+            if (s->qdev.type == TYPE_DISK) {
+                outbuf[buflen++] = 0xb0; // block limits
+                outbuf[buflen++] = 0xb2; // thin provisioning
+            }
+            break;
+        }
+        case 0x80: /* Device serial number, optional */
+        {
+            int l;
+
+            if (!s->serial) {
+                DPRINTF("Inquiry (EVPD[Serial number] not supported\n");
+                return -1;
+            }
+
+            l = strlen(s->serial);
+            if (l > 20) {
+                l = 20;
+            }
+
+            DPRINTF("Inquiry EVPD[Serial number] "
+                    "buffer size %zd\n", req->cmd.xfer);
+            memcpy(outbuf+buflen, s->serial, l);
+            buflen += l;
+            break;
+        }
+
+        case 0x83: /* Device identification page, mandatory */
+        {
+            const char *str = s->serial ?: blk_name(s->qdev.conf.blk);
+            int max_len = s->serial ? 20 : 255 - 8;
+            int id_len = strlen(str);
+
+            if (id_len > max_len) {
+                id_len = max_len;
+            }
+            DPRINTF("Inquiry EVPD[Device identification] "
+                    "buffer size %zd\n", req->cmd.xfer);
+
+            outbuf[buflen++] = 0x2; // ASCII
+            outbuf[buflen++] = 0;   // not officially assigned
+            outbuf[buflen++] = 0;   // reserved
+            outbuf[buflen++] = id_len; // length of data following
+            memcpy(outbuf+buflen, str, id_len);
+            buflen += id_len;
+
+            if (s->wwn) {
+                outbuf[buflen++] = 0x1; // Binary
+                outbuf[buflen++] = 0x3; // NAA
+                outbuf[buflen++] = 0;   // reserved
+                outbuf[buflen++] = 8;
+                stq_be_p(&outbuf[buflen], s->wwn);
+                buflen += 8;
+            }
+
+            if (s->port_wwn) {
+                outbuf[buflen++] = 0x61; // SAS / Binary
+                outbuf[buflen++] = 0x93; // PIV / Target port / NAA
+                outbuf[buflen++] = 0;    // reserved
+                outbuf[buflen++] = 8;
+                stq_be_p(&outbuf[buflen], s->port_wwn);
+                buflen += 8;
+            }
+
+            if (s->port_index) {
+                outbuf[buflen++] = 0x61; // SAS / Binary
+                outbuf[buflen++] = 0x94; // PIV / Target port / relative target port
+                outbuf[buflen++] = 0;    // reserved
+                outbuf[buflen++] = 4;
+                stw_be_p(&outbuf[buflen + 2], s->port_index);
+                buflen += 4;
+            }
+            break;
+        }
+        case 0xb0: /* block limits */
+        {
+            unsigned int unmap_sectors =
+                    s->qdev.conf.discard_granularity / s->qdev.blocksize;
+            unsigned int min_io_size =
+                    s->qdev.conf.min_io_size / s->qdev.blocksize;
+            unsigned int opt_io_size =
+                    s->qdev.conf.opt_io_size / s->qdev.blocksize;
+            unsigned int max_unmap_sectors =
+                    s->max_unmap_size / s->qdev.blocksize;
+            unsigned int max_io_sectors =
+                    s->max_io_size / s->qdev.blocksize;
+
+            if (s->qdev.type == TYPE_ROM) {
+                DPRINTF("Inquiry (EVPD[%02X] not supported for CDROM\n",
+                        page_code);
+                return -1;
+            }
+            /* required VPD size with unmap support */
+            buflen = 0x40;
+            memset(outbuf + 4, 0, buflen - 4);
+
+            outbuf[4] = 0x1; /* wsnz */
+
+            /* optimal transfer length granularity */
+            outbuf[6] = (min_io_size >> 8) & 0xff;
+            outbuf[7] = min_io_size & 0xff;
+
+            /* maximum transfer length */
+            outbuf[8] = (max_io_sectors >> 24) & 0xff;
+            outbuf[9] = (max_io_sectors >> 16) & 0xff;
+            outbuf[10] = (max_io_sectors >> 8) & 0xff;
+            outbuf[11] = max_io_sectors & 0xff;
+
+            /* optimal transfer length */
+            outbuf[12] = (opt_io_size >> 24) & 0xff;
+            outbuf[13] = (opt_io_size >> 16) & 0xff;
+            outbuf[14] = (opt_io_size >> 8) & 0xff;
+            outbuf[15] = opt_io_size & 0xff;
+
+            /* max unmap LBA count, default is 1GB */
+            outbuf[20] = (max_unmap_sectors >> 24) & 0xff;
+            outbuf[21] = (max_unmap_sectors >> 16) & 0xff;
+            outbuf[22] = (max_unmap_sectors >> 8) & 0xff;
+            outbuf[23] = max_unmap_sectors & 0xff;
+
+            /* max unmap descriptors, 255 fit in 4 kb with an 8-byte header.  */
+            outbuf[24] = 0;
+            outbuf[25] = 0;
+            outbuf[26] = 0;
+            outbuf[27] = 255;
+
+            /* optimal unmap granularity */
+            outbuf[28] = (unmap_sectors >> 24) & 0xff;
+            outbuf[29] = (unmap_sectors >> 16) & 0xff;
+            outbuf[30] = (unmap_sectors >> 8) & 0xff;
+            outbuf[31] = unmap_sectors & 0xff;
+
+            /* max write same size */
+            outbuf[36] = 0;
+            outbuf[37] = 0;
+            outbuf[38] = 0;
+            outbuf[39] = 0;
+
+            outbuf[40] = (max_io_sectors >> 24) & 0xff;
+            outbuf[41] = (max_io_sectors >> 16) & 0xff;
+            outbuf[42] = (max_io_sectors >> 8) & 0xff;
+            outbuf[43] = max_io_sectors & 0xff;
+            break;
+        }
+        case 0xb2: /* thin provisioning */
+        {
+            buflen = 8;
+            outbuf[4] = 0;
+            outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
+            outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
+            outbuf[7] = 0;
+            break;
+        }
+        default:
+            return -1;
+        }
+        /* done with EVPD */
+        assert(buflen - start <= 255);
+        outbuf[start - 1] = buflen - start;
+        return buflen;
+    }
+
+    /* Standard INQUIRY data */
+    if (req->cmd.buf[2] != 0) {
+        return -1;
+    }
+
+    /* PAGE CODE == 0 */
+    buflen = req->cmd.xfer;
+    if (buflen > SCSI_MAX_INQUIRY_LEN) {
+        buflen = SCSI_MAX_INQUIRY_LEN;
+    }
+
+    outbuf[0] = s->qdev.type & 0x1f;
+    outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
+
+    strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
+    strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
+
+    memset(&outbuf[32], 0, 4);
+    memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
+    /*
+     * We claim conformance to SPC-3, which is required for guests
+     * to ask for modern features like READ CAPACITY(16) or the
+     * block characteristics VPD page by default.  Not all of SPC-3
+     * is actually implemented, but we're good enough.
+     */
+    outbuf[2] = 5;
+    outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
+
+    if (buflen > 36) {
+        outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
+    } else {
+        /* If the allocation length of CDB is too small,
+               the additional length is not adjusted */
+        outbuf[4] = 36 - 5;
+    }
+
+    /* Sync data transfer and TCQ.  */
+    outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
+    return buflen;
+}
+
+static inline bool media_is_dvd(SCSIDiskState *s)
+{
+    uint64_t nb_sectors;
+    if (s->qdev.type != TYPE_ROM) {
+        return false;
+    }
+    if (!blk_is_inserted(s->qdev.conf.blk)) {
+        return false;
+    }
+    if (s->tray_open) {
+        return false;
+    }
+    blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
+    return nb_sectors > CD_MAX_SECTORS;
+}
+
+static inline bool media_is_cd(SCSIDiskState *s)
+{
+    uint64_t nb_sectors;
+    if (s->qdev.type != TYPE_ROM) {
+        return false;
+    }
+    if (!blk_is_inserted(s->qdev.conf.blk)) {
+        return false;
+    }
+    if (s->tray_open) {
+        return false;
+    }
+    blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
+    return nb_sectors <= CD_MAX_SECTORS;
+}
+
+static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
+                                      uint8_t *outbuf)
+{
+    uint8_t type = r->req.cmd.buf[1] & 7;
+
+    if (s->qdev.type != TYPE_ROM) {
+        return -1;
+    }
+
+    /* Types 1/2 are only defined for Blu-Ray.  */
+    if (type != 0) {
+        scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+        return -1;
+    }
+
+    memset(outbuf, 0, 34);
+    outbuf[1] = 32;
+    outbuf[2] = 0xe; /* last session complete, disc finalized */
+    outbuf[3] = 1;   /* first track on disc */
+    outbuf[4] = 1;   /* # of sessions */
+    outbuf[5] = 1;   /* first track of last session */
+    outbuf[6] = 1;   /* last track of last session */
+    outbuf[7] = 0x20; /* unrestricted use */
+    outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
+    /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
+    /* 12-23: not meaningful for CD-ROM or DVD-ROM */
+    /* 24-31: disc bar code */
+    /* 32: disc application code */
+    /* 33: number of OPC tables */
+
+    return 34;
+}
+
+static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
+                                   uint8_t *outbuf)
+{
+    static const int rds_caps_size[5] = {
+        [0] = 2048 + 4,
+        [1] = 4 + 4,
+        [3] = 188 + 4,
+        [4] = 2048 + 4,
+    };
+
+    uint8_t media = r->req.cmd.buf[1];
+    uint8_t layer = r->req.cmd.buf[6];
+    uint8_t format = r->req.cmd.buf[7];
+    int size = -1;
+
+    if (s->qdev.type != TYPE_ROM) {
+        return -1;
+    }
+    if (media != 0) {
+        scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+        return -1;
+    }
+
+    if (format != 0xff) {
+        if (s->tray_open || !blk_is_inserted(s->qdev.conf.blk)) {
+            scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
+            return -1;
+        }
+        if (media_is_cd(s)) {
+            scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
+            return -1;
+        }
+        if (format >= ARRAY_SIZE(rds_caps_size)) {
+            return -1;
+        }
+        size = rds_caps_size[format];
+        memset(outbuf, 0, size);
+    }
+
+    switch (format) {
+    case 0x00: {
+        /* Physical format information */
+        uint64_t nb_sectors;
+        if (layer != 0) {
+            goto fail;
+        }
+        blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
+
+        outbuf[4] = 1;   /* DVD-ROM, part version 1 */
+        outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
+        outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
+        outbuf[7] = 0;   /* default densities */
+
+        stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
+        stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
+        break;
+    }
+
+    case 0x01: /* DVD copyright information, all zeros */
+        break;
+
+    case 0x03: /* BCA information - invalid field for no BCA info */
+        return -1;
+
+    case 0x04: /* DVD disc manufacturing information, all zeros */
+        break;
+
+    case 0xff: { /* List capabilities */
+        int i;
+        size = 4;
+        for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
+            if (!rds_caps_size[i]) {
+                continue;
+            }
+            outbuf[size] = i;
+            outbuf[size + 1] = 0x40; /* Not writable, readable */
+            stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
+            size += 4;
+        }
+        break;
+     }
+
+    default:
+        return -1;
+    }
+
+    /* Size of buffer, not including 2 byte size field */
+    stw_be_p(outbuf, size - 2);
+    return size;
+
+fail:
+    return -1;
+}
+
+static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
+{
+    uint8_t event_code, media_status;
+
+    media_status = 0;
+    if (s->tray_open) {
+        media_status = MS_TRAY_OPEN;
+    } else if (blk_is_inserted(s->qdev.conf.blk)) {
+        media_status = MS_MEDIA_PRESENT;
+    }
+
+    /* Event notification descriptor */
+    event_code = MEC_NO_CHANGE;
+    if (media_status != MS_TRAY_OPEN) {
+        if (s->media_event) {
+            event_code = MEC_NEW_MEDIA;
+            s->media_event = false;
+        } else if (s->eject_request) {
+            event_code = MEC_EJECT_REQUESTED;
+            s->eject_request = false;
+        }
+    }
+
+    outbuf[0] = event_code;
+    outbuf[1] = media_status;
+
+    /* These fields are reserved, just clear them. */
+    outbuf[2] = 0;
+    outbuf[3] = 0;
+    return 4;
+}
+
+static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
+                                              uint8_t *outbuf)
+{
+    int size;
+    uint8_t *buf = r->req.cmd.buf;
+    uint8_t notification_class_request = buf[4];
+    if (s->qdev.type != TYPE_ROM) {
+        return -1;
+    }
+    if ((buf[1] & 1) == 0) {
+        /* asynchronous */
+        return -1;
+    }
+
+    size = 4;
+    outbuf[0] = outbuf[1] = 0;
+    outbuf[3] = 1 << GESN_MEDIA; /* supported events */
+    if (notification_class_request & (1 << GESN_MEDIA)) {
+        outbuf[2] = GESN_MEDIA;
+        size += scsi_event_status_media(s, &outbuf[size]);
+    } else {
+        outbuf[2] = 0x80;
+    }
+    stw_be_p(outbuf, size - 4);
+    return size;
+}
+
+static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
+{
+    int current;
+
+    if (s->qdev.type != TYPE_ROM) {
+        return -1;
+    }
+
+    if (media_is_dvd(s)) {
+        current = MMC_PROFILE_DVD_ROM;
+    } else if (media_is_cd(s)) {
+        current = MMC_PROFILE_CD_ROM;
+    } else {
+        current = MMC_PROFILE_NONE;
+    }
+
+    memset(outbuf, 0, 40);
+    stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
+    stw_be_p(&outbuf[6], current);
+    /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
+    outbuf[10] = 0x03; /* persistent, current */
+    outbuf[11] = 8; /* two profiles */
+    stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
+    outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
+    stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
+    outbuf[18] = (current == MMC_PROFILE_CD_ROM);
+    /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
+    stw_be_p(&outbuf[20], 1);
+    outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
+    outbuf[23] = 8;
+    stl_be_p(&outbuf[24], 1); /* SCSI */
+    outbuf[28] = 1; /* DBE = 1, mandatory */
+    /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
+    stw_be_p(&outbuf[32], 3);
+    outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
+    outbuf[35] = 4;
+    outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
+    /* TODO: Random readable, CD read, DVD read, drive serial number,
+       power management */
+    return 40;
+}
+
+static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
+{
+    if (s->qdev.type != TYPE_ROM) {
+        return -1;
+    }
+    memset(outbuf, 0, 8);
+    outbuf[5] = 1; /* CD-ROM */
+    return 8;
+}
+
+static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
+                           int page_control)
+{
+    static const int mode_sense_valid[0x3f] = {
+        [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
+        [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
+        [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
+        [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
+        [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
+        [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
+    };
+
+    uint8_t *p = *p_outbuf + 2;
+    int length;
+
+    if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
+        return -1;
+    }
+
+    /*
+     * If Changeable Values are requested, a mask denoting those mode parameters
+     * that are changeable shall be returned. As we currently don't support
+     * parameter changes via MODE_SELECT all bits are returned set to zero.
+     * The buffer was already menset to zero by the caller of this function.
+     *
+     * The offsets here are off by two compared to the descriptions in the
+     * SCSI specs, because those include a 2-byte header.  This is unfortunate,
+     * but it is done so that offsets are consistent within our implementation
+     * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
+     * 2-byte and 4-byte headers.
+     */
+    switch (page) {
+    case MODE_PAGE_HD_GEOMETRY:
+        length = 0x16;
+        if (page_control == 1) { /* Changeable Values */
+            break;
+        }
+        /* if a geometry hint is available, use it */
+        p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
+        p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
+        p[2] = s->qdev.conf.cyls & 0xff;
+        p[3] = s->qdev.conf.heads & 0xff;
+        /* Write precomp start cylinder, disabled */
+        p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
+        p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
+        p[6] = s->qdev.conf.cyls & 0xff;
+        /* Reduced current start cylinder, disabled */
+        p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
+        p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
+        p[9] = s->qdev.conf.cyls & 0xff;
+        /* Device step rate [ns], 200ns */
+        p[10] = 0;
+        p[11] = 200;
+        /* Landing zone cylinder */
+        p[12] = 0xff;
+        p[13] =  0xff;
+        p[14] = 0xff;
+        /* Medium rotation rate [rpm], 5400 rpm */
+        p[18] = (5400 >> 8) & 0xff;
+        p[19] = 5400 & 0xff;
+        break;
+
+    case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
+        length = 0x1e;
+        if (page_control == 1) { /* Changeable Values */
+            break;
+        }
+        /* Transfer rate [kbit/s], 5Mbit/s */
+        p[0] = 5000 >> 8;
+        p[1] = 5000 & 0xff;
+        /* if a geometry hint is available, use it */
+        p[2] = s->qdev.conf.heads & 0xff;
+        p[3] = s->qdev.conf.secs & 0xff;
+        p[4] = s->qdev.blocksize >> 8;
+        p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
+        p[7] = s->qdev.conf.cyls & 0xff;
+        /* Write precomp start cylinder, disabled */
+        p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
+        p[9] = s->qdev.conf.cyls & 0xff;
+        /* Reduced current start cylinder, disabled */
+        p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
+        p[11] = s->qdev.conf.cyls & 0xff;
+        /* Device step rate [100us], 100us */
+        p[12] = 0;
+        p[13] = 1;
+        /* Device step pulse width [us], 1us */
+        p[14] = 1;
+        /* Device head settle delay [100us], 100us */
+        p[15] = 0;
+        p[16] = 1;
+        /* Motor on delay [0.1s], 0.1s */
+        p[17] = 1;
+        /* Motor off delay [0.1s], 0.1s */
+        p[18] = 1;
+        /* Medium rotation rate [rpm], 5400 rpm */
+        p[26] = (5400 >> 8) & 0xff;
+        p[27] = 5400 & 0xff;
+        break;
+
+    case MODE_PAGE_CACHING:
+        length = 0x12;
+        if (page_control == 1 || /* Changeable Values */
+            blk_enable_write_cache(s->qdev.conf.blk)) {
+            p[0] = 4; /* WCE */
+        }
+        break;
+
+    case MODE_PAGE_R_W_ERROR:
+        length = 10;
+        if (page_control == 1) { /* Changeable Values */
+            break;
+        }
+        p[0] = 0x80; /* Automatic Write Reallocation Enabled */
+        if (s->qdev.type == TYPE_ROM) {
+            p[1] = 0x20; /* Read Retry Count */
+        }
+        break;
+
+    case MODE_PAGE_AUDIO_CTL:
+        length = 14;
+        break;
+
+    case MODE_PAGE_CAPABILITIES:
+        length = 0x14;
+        if (page_control == 1) { /* Changeable Values */
+            break;
+        }
+
+        p[0] = 0x3b; /* CD-R & CD-RW read */
+        p[1] = 0; /* Writing not supported */
+        p[2] = 0x7f; /* Audio, composite, digital out,
+                        mode 2 form 1&2, multi session */
+        p[3] = 0xff; /* CD DA, DA accurate, RW supported,
+                        RW corrected, C2 errors, ISRC,
+                        UPC, Bar code */
+        p[4] = 0x2d | (s->tray_locked ? 2 : 0);
+        /* Locking supported, jumper present, eject, tray */
+        p[5] = 0; /* no volume & mute control, no
+                     changer */
+        p[6] = (50 * 176) >> 8; /* 50x read speed */
+        p[7] = (50 * 176) & 0xff;
+        p[8] = 2 >> 8; /* Two volume levels */
+        p[9] = 2 & 0xff;
+        p[10] = 2048 >> 8; /* 2M buffer */
+        p[11] = 2048 & 0xff;
+        p[12] = (16 * 176) >> 8; /* 16x read speed current */
+        p[13] = (16 * 176) & 0xff;
+        p[16] = (16 * 176) >> 8; /* 16x write speed */
+        p[17] = (16 * 176) & 0xff;
+        p[18] = (16 * 176) >> 8; /* 16x write speed current */
+        p[19] = (16 * 176) & 0xff;
+        break;
+
+    default:
+        return -1;
+    }
+
+    assert(length < 256);
+    (*p_outbuf)[0] = page;
+    (*p_outbuf)[1] = length;
+    *p_outbuf += length + 2;
+    return length + 2;
+}
+
+static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    uint64_t nb_sectors;
+    bool dbd;
+    int page, buflen, ret, page_control;
+    uint8_t *p;
+    uint8_t dev_specific_param;
+
+    dbd = (r->req.cmd.buf[1] & 0x8) != 0;
+    page = r->req.cmd.buf[2] & 0x3f;
+    page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
+    DPRINTF("Mode Sense(%d) (page %d, xfer %zd, page_control %d)\n",
+        (r->req.cmd.buf[0] == MODE_SENSE) ? 6 : 10, page, r->req.cmd.xfer, page_control);
+    memset(outbuf, 0, r->req.cmd.xfer);
+    p = outbuf;
+
+    if (s->qdev.type == TYPE_DISK) {
+        dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
+        if (blk_is_read_only(s->qdev.conf.blk)) {
+            dev_specific_param |= 0x80; /* Readonly.  */
+        }
+    } else {
+        /* MMC prescribes that CD/DVD drives have no block descriptors,
+         * and defines no device-specific parameter.  */
+        dev_specific_param = 0x00;
+        dbd = true;
+    }
+
+    if (r->req.cmd.buf[0] == MODE_SENSE) {
+        p[1] = 0; /* Default media type.  */
+        p[2] = dev_specific_param;
+        p[3] = 0; /* Block descriptor length.  */
+        p += 4;
+    } else { /* MODE_SENSE_10 */
+        p[2] = 0; /* Default media type.  */
+        p[3] = dev_specific_param;
+        p[6] = p[7] = 0; /* Block descriptor length.  */
+        p += 8;
+    }
+
+    blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
+    if (!dbd && nb_sectors) {
+        if (r->req.cmd.buf[0] == MODE_SENSE) {
+            outbuf[3] = 8; /* Block descriptor length  */
+        } else { /* MODE_SENSE_10 */
+            outbuf[7] = 8; /* Block descriptor length  */
+        }
+        nb_sectors /= (s->qdev.blocksize / 512);
+        if (nb_sectors > 0xffffff) {
+            nb_sectors = 0;
+        }
+        p[0] = 0; /* media density code */
+        p[1] = (nb_sectors >> 16) & 0xff;
+        p[2] = (nb_sectors >> 8) & 0xff;
+        p[3] = nb_sectors & 0xff;
+        p[4] = 0; /* reserved */
+        p[5] = 0; /* bytes 5-7 are the sector size in bytes */
+        p[6] = s->qdev.blocksize >> 8;
+        p[7] = 0;
+        p += 8;
+    }
+
+    if (page_control == 3) {
+        /* Saved Values */
+        scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
+        return -1;
+    }
+
+    if (page == 0x3f) {
+        for (page = 0; page <= 0x3e; page++) {
+            mode_sense_page(s, page, &p, page_control);
+        }
+    } else {
+        ret = mode_sense_page(s, page, &p, page_control);
+        if (ret == -1) {
+            return -1;
+        }
+    }
+
+    buflen = p - outbuf;
+    /*
+     * The mode data length field specifies the length in bytes of the
+     * following data that is available to be transferred. The mode data
+     * length does not include itself.
+     */
+    if (r->req.cmd.buf[0] == MODE_SENSE) {
+        outbuf[0] = buflen - 1;
+    } else { /* MODE_SENSE_10 */
+        outbuf[0] = ((buflen - 2) >> 8) & 0xff;
+        outbuf[1] = (buflen - 2) & 0xff;
+    }
+    return buflen;
+}
+
+static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
+    int start_track, format, msf, toclen;
+    uint64_t nb_sectors;
+
+    msf = req->cmd.buf[1] & 2;
+    format = req->cmd.buf[2] & 0xf;
+    start_track = req->cmd.buf[6];
+    blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
+    DPRINTF("Read TOC (track %d format %d msf %d)\n", start_track, format, msf >> 1);
+    nb_sectors /= s->qdev.blocksize / 512;
+    switch (format) {
+    case 0:
+        toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
+        break;
+    case 1:
+        /* multi session : only a single session defined */
+        toclen = 12;
+        memset(outbuf, 0, 12);
+        outbuf[1] = 0x0a;
+        outbuf[2] = 0x01;
+        outbuf[3] = 0x01;
+        break;
+    case 2:
+        toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
+        break;
+    default:
+        return -1;
+    }
+    return toclen;
+}
+
+static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
+{
+    SCSIRequest *req = &r->req;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
+    bool start = req->cmd.buf[4] & 1;
+    bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
+    int pwrcnd = req->cmd.buf[4] & 0xf0;
+
+    if (pwrcnd) {
+        /* eject/load only happens for power condition == 0 */
+        return 0;
+    }
+
+    if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
+        if (!start && !s->tray_open && s->tray_locked) {
+            scsi_check_condition(r,
+                                 blk_is_inserted(s->qdev.conf.blk)
+                                 ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
+                                 : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
+            return -1;
+        }
+
+        if (s->tray_open != !start) {
+            blk_eject(s->qdev.conf.blk, !start);
+            s->tray_open = !start;
+        }
+    }
+    return 0;
+}
+
+static void scsi_disk_emulate_read_data(SCSIRequest *req)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+    int buflen = r->iov.iov_len;
+
+    if (buflen) {
+        DPRINTF("Read buf_len=%d\n", buflen);
+        r->iov.iov_len = 0;
+        r->started = true;
+        scsi_req_data(&r->req, buflen);
+        return;
+    }
+
+    /* This also clears the sense buffer for REQUEST SENSE.  */
+    scsi_req_complete(&r->req, GOOD);
+}
+
+static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
+                                       uint8_t *inbuf, int inlen)
+{
+    uint8_t mode_current[SCSI_MAX_MODE_LEN];
+    uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
+    uint8_t *p;
+    int len, expected_len, changeable_len, i;
+
+    /* The input buffer does not include the page header, so it is
+     * off by 2 bytes.
+     */
+    expected_len = inlen + 2;
+    if (expected_len > SCSI_MAX_MODE_LEN) {
+        return -1;
+    }
+
+    p = mode_current;
+    memset(mode_current, 0, inlen + 2);
+    len = mode_sense_page(s, page, &p, 0);
+    if (len < 0 || len != expected_len) {
+        return -1;
+    }
+
+    p = mode_changeable;
+    memset(mode_changeable, 0, inlen + 2);
+    changeable_len = mode_sense_page(s, page, &p, 1);
+    assert(changeable_len == len);
+
+    /* Check that unchangeable bits are the same as what MODE SENSE
+     * would return.
+     */
+    for (i = 2; i < len; i++) {
+        if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
+            return -1;
+        }
+    }
+    return 0;
+}
+
+static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
+{
+    switch (page) {
+    case MODE_PAGE_CACHING:
+        blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
+        break;
+
+    default:
+        break;
+    }
+}
+
+static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+
+    while (len > 0) {
+        int page, subpage, page_len;
+
+        /* Parse both possible formats for the mode page headers.  */
+        page = p[0] & 0x3f;
+        if (p[0] & 0x40) {
+            if (len < 4) {
+                goto invalid_param_len;
+            }
+            subpage = p[1];
+            page_len = lduw_be_p(&p[2]);
+            p += 4;
+            len -= 4;
+        } else {
+            if (len < 2) {
+                goto invalid_param_len;
+            }
+            subpage = 0;
+            page_len = p[1];
+            p += 2;
+            len -= 2;
+        }
+
+        if (subpage) {
+            goto invalid_param;
+        }
+        if (page_len > len) {
+            goto invalid_param_len;
+        }
+
+        if (!change) {
+            if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
+                goto invalid_param;
+            }
+        } else {
+            scsi_disk_apply_mode_select(s, page, p);
+        }
+
+        p += page_len;
+        len -= page_len;
+    }
+    return 0;
+
+invalid_param:
+    scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
+    return -1;
+
+invalid_param_len:
+    scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
+    return -1;
+}
+
+static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    uint8_t *p = inbuf;
+    int cmd = r->req.cmd.buf[0];
+    int len = r->req.cmd.xfer;
+    int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
+    int bd_len;
+    int pass;
+
+    /* We only support PF=1, SP=0.  */
+    if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
+        goto invalid_field;
+    }
+
+    if (len < hdr_len) {
+        goto invalid_param_len;
+    }
+
+    bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
+    len -= hdr_len;
+    p += hdr_len;
+    if (len < bd_len) {
+        goto invalid_param_len;
+    }
+    if (bd_len != 0 && bd_len != 8) {
+        goto invalid_param;
+    }
+
+    len -= bd_len;
+    p += bd_len;
+
+    /* Ensure no change is made if there is an error!  */
+    for (pass = 0; pass < 2; pass++) {
+        if (mode_select_pages(r, p, len, pass == 1) < 0) {
+            assert(pass == 0);
+            return;
+        }
+    }
+    if (!blk_enable_write_cache(s->qdev.conf.blk)) {
+        /* The request is used as the AIO opaque value, so add a ref.  */
+        scsi_req_ref(&r->req);
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
+                         BLOCK_ACCT_FLUSH);
+        r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
+        return;
+    }
+
+    scsi_req_complete(&r->req, GOOD);
+    return;
+
+invalid_param:
+    scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
+    return;
+
+invalid_param_len:
+    scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
+    return;
+
+invalid_field:
+    scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+}
+
+static inline bool check_lba_range(SCSIDiskState *s,
+                                   uint64_t sector_num, uint32_t nb_sectors)
+{
+    /*
+     * The first line tests that no overflow happens when computing the last
+     * sector.  The second line tests that the last accessed sector is in
+     * range.
+     *
+     * Careful, the computations should not underflow for nb_sectors == 0,
+     * and a 0-block read to the first LBA beyond the end of device is
+     * valid.
+     */
+    return (sector_num <= sector_num + nb_sectors &&
+            sector_num + nb_sectors <= s->qdev.max_lba + 1);
+}
+
+typedef struct UnmapCBData {
+    SCSIDiskReq *r;
+    uint8_t *inbuf;
+    int count;
+} UnmapCBData;
+
+static void scsi_unmap_complete(void *opaque, int ret)
+{
+    UnmapCBData *data = opaque;
+    SCSIDiskReq *r = data->r;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    uint64_t sector_num;
+    uint32_t nb_sectors;
+
+    r->req.aiocb = NULL;
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+
+    if (ret < 0) {
+        if (scsi_handle_rw_error(r, -ret)) {
+            goto done;
+        }
+    }
+
+    if (data->count > 0) {
+        sector_num = ldq_be_p(&data->inbuf[0]);
+        nb_sectors = ldl_be_p(&data->inbuf[8]) & 0xffffffffULL;
+        if (!check_lba_range(s, sector_num, nb_sectors)) {
+            scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
+            goto done;
+        }
+
+        r->req.aiocb = blk_aio_discard(s->qdev.conf.blk,
+                                       sector_num * (s->qdev.blocksize / 512),
+                                       nb_sectors * (s->qdev.blocksize / 512),
+                                       scsi_unmap_complete, data);
+        data->count--;
+        data->inbuf += 16;
+        return;
+    }
+
+    scsi_req_complete(&r->req, GOOD);
+
+done:
+    scsi_req_unref(&r->req);
+    g_free(data);
+}
+
+static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    uint8_t *p = inbuf;
+    int len = r->req.cmd.xfer;
+    UnmapCBData *data;
+
+    /* Reject ANCHOR=1.  */
+    if (r->req.cmd.buf[1] & 0x1) {
+        goto invalid_field;
+    }
+
+    if (len < 8) {
+        goto invalid_param_len;
+    }
+    if (len < lduw_be_p(&p[0]) + 2) {
+        goto invalid_param_len;
+    }
+    if (len < lduw_be_p(&p[2]) + 8) {
+        goto invalid_param_len;
+    }
+    if (lduw_be_p(&p[2]) & 15) {
+        goto invalid_param_len;
+    }
+
+    if (blk_is_read_only(s->qdev.conf.blk)) {
+        scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
+        return;
+    }
+
+    data = g_new0(UnmapCBData, 1);
+    data->r = r;
+    data->inbuf = &p[8];
+    data->count = lduw_be_p(&p[2]) >> 4;
+
+    /* The matching unref is in scsi_unmap_complete, before data is freed.  */
+    scsi_req_ref(&r->req);
+    scsi_unmap_complete(data, 0);
+    return;
+
+invalid_param_len:
+    scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
+    return;
+
+invalid_field:
+    scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+}
+
+typedef struct WriteSameCBData {
+    SCSIDiskReq *r;
+    int64_t sector;
+    int nb_sectors;
+    QEMUIOVector qiov;
+    struct iovec iov;
+} WriteSameCBData;
+
+static void scsi_write_same_complete(void *opaque, int ret)
+{
+    WriteSameCBData *data = opaque;
+    SCSIDiskReq *r = data->r;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+
+    assert(r->req.aiocb != NULL);
+    r->req.aiocb = NULL;
+    block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+
+    if (ret < 0) {
+        if (scsi_handle_rw_error(r, -ret)) {
+            goto done;
+        }
+    }
+
+    data->nb_sectors -= data->iov.iov_len / 512;
+    data->sector += data->iov.iov_len / 512;
+    data->iov.iov_len = MIN(data->nb_sectors * 512, data->iov.iov_len);
+    if (data->iov.iov_len) {
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
+                         data->iov.iov_len, BLOCK_ACCT_WRITE);
+        r->req.aiocb = blk_aio_writev(s->qdev.conf.blk, data->sector,
+                                      &data->qiov, data->iov.iov_len / 512,
+                                      scsi_write_same_complete, data);
+        return;
+    }
+
+    scsi_req_complete(&r->req, GOOD);
+
+done:
+    scsi_req_unref(&r->req);
+    qemu_vfree(data->iov.iov_base);
+    g_free(data);
+}
+
+static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
+{
+    SCSIRequest *req = &r->req;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
+    uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
+    WriteSameCBData *data;
+    uint8_t *buf;
+    int i;
+
+    /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
+    if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
+        scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+        return;
+    }
+
+    if (blk_is_read_only(s->qdev.conf.blk)) {
+        scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
+        return;
+    }
+    if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
+        scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
+        return;
+    }
+
+    if (buffer_is_zero(inbuf, s->qdev.blocksize)) {
+        int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
+
+        /* The request is used as the AIO opaque value, so add a ref.  */
+        scsi_req_ref(&r->req);
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
+                         nb_sectors * s->qdev.blocksize,
+                        BLOCK_ACCT_WRITE);
+        r->req.aiocb = blk_aio_write_zeroes(s->qdev.conf.blk,
+                                r->req.cmd.lba * (s->qdev.blocksize / 512),
+                                nb_sectors * (s->qdev.blocksize / 512),
+                                flags, scsi_aio_complete, r);
+        return;
+    }
+
+    data = g_new0(WriteSameCBData, 1);
+    data->r = r;
+    data->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
+    data->nb_sectors = nb_sectors * (s->qdev.blocksize / 512);
+    data->iov.iov_len = MIN(data->nb_sectors * 512, SCSI_WRITE_SAME_MAX);
+    data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
+                                              data->iov.iov_len);
+    qemu_iovec_init_external(&data->qiov, &data->iov, 1);
+
+    for (i = 0; i < data->iov.iov_len; i += s->qdev.blocksize) {
+        memcpy(&buf[i], inbuf, s->qdev.blocksize);
+    }
+
+    scsi_req_ref(&r->req);
+    block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
+                     data->iov.iov_len, BLOCK_ACCT_WRITE);
+    r->req.aiocb = blk_aio_writev(s->qdev.conf.blk, data->sector,
+                                  &data->qiov, data->iov.iov_len / 512,
+                                  scsi_write_same_complete, data);
+}
+
+static void scsi_disk_emulate_write_data(SCSIRequest *req)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+
+    if (r->iov.iov_len) {
+        int buflen = r->iov.iov_len;
+        DPRINTF("Write buf_len=%d\n", buflen);
+        r->iov.iov_len = 0;
+        scsi_req_data(&r->req, buflen);
+        return;
+    }
+
+    switch (req->cmd.buf[0]) {
+    case MODE_SELECT:
+    case MODE_SELECT_10:
+        /* This also clears the sense buffer for REQUEST SENSE.  */
+        scsi_disk_emulate_mode_select(r, r->iov.iov_base);
+        break;
+
+    case UNMAP:
+        scsi_disk_emulate_unmap(r, r->iov.iov_base);
+        break;
+
+    case VERIFY_10:
+    case VERIFY_12:
+    case VERIFY_16:
+        if (r->req.status == -1) {
+            scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+        }
+        break;
+
+    case WRITE_SAME_10:
+    case WRITE_SAME_16:
+        scsi_disk_emulate_write_same(r, r->iov.iov_base);
+        break;
+
+    default:
+        abort();
+    }
+}
+
+static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
+    uint64_t nb_sectors;
+    uint8_t *outbuf;
+    int buflen;
+
+    switch (req->cmd.buf[0]) {
+    case INQUIRY:
+    case MODE_SENSE:
+    case MODE_SENSE_10:
+    case RESERVE:
+    case RESERVE_10:
+    case RELEASE:
+    case RELEASE_10:
+    case START_STOP:
+    case ALLOW_MEDIUM_REMOVAL:
+    case GET_CONFIGURATION:
+    case GET_EVENT_STATUS_NOTIFICATION:
+    case MECHANISM_STATUS:
+    case REQUEST_SENSE:
+        break;
+
+    default:
+        if (s->tray_open || !blk_is_inserted(s->qdev.conf.blk)) {
+            scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
+            return 0;
+        }
+        break;
+    }
+
+    /*
+     * FIXME: we shouldn't return anything bigger than 4k, but the code
+     * requires the buffer to be as big as req->cmd.xfer in several
+     * places.  So, do not allow CDBs with a very large ALLOCATION
+     * LENGTH.  The real fix would be to modify scsi_read_data and
+     * dma_buf_read, so that they return data beyond the buflen
+     * as all zeros.
+     */
+    if (req->cmd.xfer > 65536) {
+        goto illegal_request;
+    }
+    r->buflen = MAX(4096, req->cmd.xfer);
+
+    if (!r->iov.iov_base) {
+        r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
+    }
+
+    buflen = req->cmd.xfer;
+    outbuf = r->iov.iov_base;
+    memset(outbuf, 0, r->buflen);
+    switch (req->cmd.buf[0]) {
+    case TEST_UNIT_READY:
+        assert(!s->tray_open && blk_is_inserted(s->qdev.conf.blk));
+        break;
+    case INQUIRY:
+        buflen = scsi_disk_emulate_inquiry(req, outbuf);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case MODE_SENSE:
+    case MODE_SENSE_10:
+        buflen = scsi_disk_emulate_mode_sense(r, outbuf);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case READ_TOC:
+        buflen = scsi_disk_emulate_read_toc(req, outbuf);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case RESERVE:
+        if (req->cmd.buf[1] & 1) {
+            goto illegal_request;
+        }
+        break;
+    case RESERVE_10:
+        if (req->cmd.buf[1] & 3) {
+            goto illegal_request;
+        }
+        break;
+    case RELEASE:
+        if (req->cmd.buf[1] & 1) {
+            goto illegal_request;
+        }
+        break;
+    case RELEASE_10:
+        if (req->cmd.buf[1] & 3) {
+            goto illegal_request;
+        }
+        break;
+    case START_STOP:
+        if (scsi_disk_emulate_start_stop(r) < 0) {
+            return 0;
+        }
+        break;
+    case ALLOW_MEDIUM_REMOVAL:
+        s->tray_locked = req->cmd.buf[4] & 1;
+        blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
+        break;
+    case READ_CAPACITY_10:
+        /* The normal LEN field for this command is zero.  */
+        memset(outbuf, 0, 8);
+        blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
+        if (!nb_sectors) {
+            scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
+            return 0;
+        }
+        if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
+            goto illegal_request;
+        }
+        nb_sectors /= s->qdev.blocksize / 512;
+        /* Returned value is the address of the last sector.  */
+        nb_sectors--;
+        /* Remember the new size for read/write sanity checking. */
+        s->qdev.max_lba = nb_sectors;
+        /* Clip to 2TB, instead of returning capacity modulo 2TB. */
+        if (nb_sectors > UINT32_MAX) {
+            nb_sectors = UINT32_MAX;
+        }
+        outbuf[0] = (nb_sectors >> 24) & 0xff;
+        outbuf[1] = (nb_sectors >> 16) & 0xff;
+        outbuf[2] = (nb_sectors >> 8) & 0xff;
+        outbuf[3] = nb_sectors & 0xff;
+        outbuf[4] = 0;
+        outbuf[5] = 0;
+        outbuf[6] = s->qdev.blocksize >> 8;
+        outbuf[7] = 0;
+        break;
+    case REQUEST_SENSE:
+        /* Just return "NO SENSE".  */
+        buflen = scsi_build_sense(NULL, 0, outbuf, r->buflen,
+                                  (req->cmd.buf[1] & 1) == 0);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case MECHANISM_STATUS:
+        buflen = scsi_emulate_mechanism_status(s, outbuf);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case GET_CONFIGURATION:
+        buflen = scsi_get_configuration(s, outbuf);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case GET_EVENT_STATUS_NOTIFICATION:
+        buflen = scsi_get_event_status_notification(s, r, outbuf);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case READ_DISC_INFORMATION:
+        buflen = scsi_read_disc_information(s, r, outbuf);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case READ_DVD_STRUCTURE:
+        buflen = scsi_read_dvd_structure(s, r, outbuf);
+        if (buflen < 0) {
+            goto illegal_request;
+        }
+        break;
+    case SERVICE_ACTION_IN_16:
+        /* Service Action In subcommands. */
+        if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
+            DPRINTF("SAI READ CAPACITY(16)\n");
+            memset(outbuf, 0, req->cmd.xfer);
+            blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
+            if (!nb_sectors) {
+                scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
+                return 0;
+            }
+            if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
+                goto illegal_request;
+            }
+            nb_sectors /= s->qdev.blocksize / 512;
+            /* Returned value is the address of the last sector.  */
+            nb_sectors--;
+            /* Remember the new size for read/write sanity checking. */
+            s->qdev.max_lba = nb_sectors;
+            outbuf[0] = (nb_sectors >> 56) & 0xff;
+            outbuf[1] = (nb_sectors >> 48) & 0xff;
+            outbuf[2] = (nb_sectors >> 40) & 0xff;
+            outbuf[3] = (nb_sectors >> 32) & 0xff;
+            outbuf[4] = (nb_sectors >> 24) & 0xff;
+            outbuf[5] = (nb_sectors >> 16) & 0xff;
+            outbuf[6] = (nb_sectors >> 8) & 0xff;
+            outbuf[7] = nb_sectors & 0xff;
+            outbuf[8] = 0;
+            outbuf[9] = 0;
+            outbuf[10] = s->qdev.blocksize >> 8;
+            outbuf[11] = 0;
+            outbuf[12] = 0;
+            outbuf[13] = get_physical_block_exp(&s->qdev.conf);
+
+            /* set TPE bit if the format supports discard */
+            if (s->qdev.conf.discard_granularity) {
+                outbuf[14] = 0x80;
+            }
+
+            /* Protection, exponent and lowest lba field left blank. */
+            break;
+        }
+        DPRINTF("Unsupported Service Action In\n");
+        goto illegal_request;
+    case SYNCHRONIZE_CACHE:
+        /* The request is used as the AIO opaque value, so add a ref.  */
+        scsi_req_ref(&r->req);
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
+                         BLOCK_ACCT_FLUSH);
+        r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
+        return 0;
+    case SEEK_10:
+        DPRINTF("Seek(10) (sector %" PRId64 ")\n", r->req.cmd.lba);
+        if (r->req.cmd.lba > s->qdev.max_lba) {
+            goto illegal_lba;
+        }
+        break;
+    case MODE_SELECT:
+        DPRINTF("Mode Select(6) (len %lu)\n", (long)r->req.cmd.xfer);
+        break;
+    case MODE_SELECT_10:
+        DPRINTF("Mode Select(10) (len %lu)\n", (long)r->req.cmd.xfer);
+        break;
+    case UNMAP:
+        DPRINTF("Unmap (len %lu)\n", (long)r->req.cmd.xfer);
+        break;
+    case VERIFY_10:
+    case VERIFY_12:
+    case VERIFY_16:
+        DPRINTF("Verify (bytchk %d)\n", (req->cmd.buf[1] >> 1) & 3);
+        if (req->cmd.buf[1] & 6) {
+            goto illegal_request;
+        }
+        break;
+    case WRITE_SAME_10:
+    case WRITE_SAME_16:
+        DPRINTF("WRITE SAME %d (len %lu)\n",
+                req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16,
+                (long)r->req.cmd.xfer);
+        break;
+    default:
+        DPRINTF("Unknown SCSI command (%2.2x=%s)\n", buf[0],
+                scsi_command_name(buf[0]));
+        scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
+        return 0;
+    }
+    assert(!r->req.aiocb);
+    r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
+    if (r->iov.iov_len == 0) {
+        scsi_req_complete(&r->req, GOOD);
+    }
+    if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+        assert(r->iov.iov_len == req->cmd.xfer);
+        return -r->iov.iov_len;
+    } else {
+        return r->iov.iov_len;
+    }
+
+illegal_request:
+    if (r->req.status == -1) {
+        scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+    }
+    return 0;
+
+illegal_lba:
+    scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
+    return 0;
+}
+
+/* Execute a scsi command.  Returns the length of the data expected by the
+   command.  This will be Positive for data transfers from the device
+   (eg. disk reads), negative for transfers to the device (eg. disk writes),
+   and zero if the command does not transfer any data.  */
+
+static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
+{
+    SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
+    uint32_t len;
+    uint8_t command;
+
+    command = buf[0];
+
+    if (s->tray_open || !blk_is_inserted(s->qdev.conf.blk)) {
+        scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
+        return 0;
+    }
+
+    len = scsi_data_cdb_xfer(r->req.cmd.buf);
+    switch (command) {
+    case READ_6:
+    case READ_10:
+    case READ_12:
+    case READ_16:
+        DPRINTF("Read (sector %" PRId64 ", count %u)\n", r->req.cmd.lba, len);
+        if (r->req.cmd.buf[1] & 0xe0) {
+            goto illegal_request;
+        }
+        if (!check_lba_range(s, r->req.cmd.lba, len)) {
+            goto illegal_lba;
+        }
+        r->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
+        r->sector_count = len * (s->qdev.blocksize / 512);
+        break;
+    case WRITE_6:
+    case WRITE_10:
+    case WRITE_12:
+    case WRITE_16:
+    case WRITE_VERIFY_10:
+    case WRITE_VERIFY_12:
+    case WRITE_VERIFY_16:
+        if (blk_is_read_only(s->qdev.conf.blk)) {
+            scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
+            return 0;
+        }
+        DPRINTF("Write %s(sector %" PRId64 ", count %u)\n",
+                (command & 0xe) == 0xe ? "And Verify " : "",
+                r->req.cmd.lba, len);
+        if (r->req.cmd.buf[1] & 0xe0) {
+            goto illegal_request;
+        }
+        if (!check_lba_range(s, r->req.cmd.lba, len)) {
+            goto illegal_lba;
+        }
+        r->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
+        r->sector_count = len * (s->qdev.blocksize / 512);
+        break;
+    default:
+        abort();
+    illegal_request:
+        scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
+        return 0;
+    illegal_lba:
+        scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
+        return 0;
+    }
+    if (r->sector_count == 0) {
+        scsi_req_complete(&r->req, GOOD);
+    }
+    assert(r->iov.iov_len == 0);
+    if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+        return -r->sector_count * 512;
+    } else {
+        return r->sector_count * 512;
+    }
+}
+
+static void scsi_disk_reset(DeviceState *dev)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
+    uint64_t nb_sectors;
+
+    scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
+
+    blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
+    nb_sectors /= s->qdev.blocksize / 512;
+    if (nb_sectors) {
+        nb_sectors--;
+    }
+    s->qdev.max_lba = nb_sectors;
+    /* reset tray statuses */
+    s->tray_locked = 0;
+    s->tray_open = 0;
+}
+
+static void scsi_disk_resize_cb(void *opaque)
+{
+    SCSIDiskState *s = opaque;
+
+    /* SPC lists this sense code as available only for
+     * direct-access devices.
+     */
+    if (s->qdev.type == TYPE_DISK) {
+        scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
+    }
+}
+
+static void scsi_cd_change_media_cb(void *opaque, bool load)
+{
+    SCSIDiskState *s = opaque;
+
+    /*
+     * When a CD gets changed, we have to report an ejected state and
+     * then a loaded state to guests so that they detect tray
+     * open/close and media change events.  Guests that do not use
+     * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
+     * states rely on this behavior.
+     *
+     * media_changed governs the state machine used for unit attention
+     * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
+     */
+    s->media_changed = load;
+    s->tray_open = !load;
+    scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
+    s->media_event = true;
+    s->eject_request = false;
+}
+
+static void scsi_cd_eject_request_cb(void *opaque, bool force)
+{
+    SCSIDiskState *s = opaque;
+
+    s->eject_request = true;
+    if (force) {
+        s->tray_locked = false;
+    }
+}
+
+static bool scsi_cd_is_tray_open(void *opaque)
+{
+    return ((SCSIDiskState *)opaque)->tray_open;
+}
+
+static bool scsi_cd_is_medium_locked(void *opaque)
+{
+    return ((SCSIDiskState *)opaque)->tray_locked;
+}
+
+static const BlockDevOps scsi_disk_removable_block_ops = {
+    .change_media_cb = scsi_cd_change_media_cb,
+    .eject_request_cb = scsi_cd_eject_request_cb,
+    .is_tray_open = scsi_cd_is_tray_open,
+    .is_medium_locked = scsi_cd_is_medium_locked,
+
+    .resize_cb = scsi_disk_resize_cb,
+};
+
+static const BlockDevOps scsi_disk_block_ops = {
+    .resize_cb = scsi_disk_resize_cb,
+};
+
+static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
+    if (s->media_changed) {
+        s->media_changed = false;
+        scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
+    }
+}
+
+static void scsi_realize(SCSIDevice *dev, Error **errp)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
+    Error *err = NULL;
+
+    if (!s->qdev.conf.blk) {
+        error_setg(errp, "drive property not set");
+        return;
+    }
+
+    if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
+        !blk_is_inserted(s->qdev.conf.blk)) {
+        error_setg(errp, "Device needs media, but drive is empty");
+        return;
+    }
+
+    blkconf_serial(&s->qdev.conf, &s->serial);
+    blkconf_blocksizes(&s->qdev.conf);
+    if (dev->type == TYPE_DISK) {
+        blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, &err);
+        if (err) {
+            error_propagate(errp, err);
+            return;
+        }
+    }
+
+    if (s->qdev.conf.discard_granularity == -1) {
+        s->qdev.conf.discard_granularity =
+            MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
+    }
+
+    if (!s->version) {
+        s->version = g_strdup(qemu_get_version());
+    }
+    if (!s->vendor) {
+        s->vendor = g_strdup("QEMU");
+    }
+
+    if (blk_is_sg(s->qdev.conf.blk)) {
+        error_setg(errp, "unwanted /dev/sg*");
+        return;
+    }
+
+    if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
+            !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
+        blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
+    } else {
+        blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
+    }
+    blk_set_guest_block_size(s->qdev.conf.blk, s->qdev.blocksize);
+
+    blk_iostatus_enable(s->qdev.conf.blk);
+}
+
+static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
+    /* can happen for devices without drive. The error message for missing
+     * backend will be issued in scsi_realize
+     */
+    if (s->qdev.conf.blk) {
+        blkconf_blocksizes(&s->qdev.conf);
+    }
+    s->qdev.blocksize = s->qdev.conf.logical_block_size;
+    s->qdev.type = TYPE_DISK;
+    if (!s->product) {
+        s->product = g_strdup("QEMU HARDDISK");
+    }
+    scsi_realize(&s->qdev, errp);
+}
+
+static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
+    s->qdev.blocksize = 2048;
+    s->qdev.type = TYPE_ROM;
+    s->features |= 1 << SCSI_DISK_F_REMOVABLE;
+    if (!s->product) {
+        s->product = g_strdup("QEMU CD-ROM");
+    }
+    scsi_realize(&s->qdev, errp);
+}
+
+static void scsi_disk_realize(SCSIDevice *dev, Error **errp)
+{
+    DriveInfo *dinfo;
+    Error *local_err = NULL;
+
+    if (!dev->conf.blk) {
+        scsi_realize(dev, &local_err);
+        assert(local_err);
+        error_propagate(errp, local_err);
+        return;
+    }
+
+    dinfo = blk_legacy_dinfo(dev->conf.blk);
+    if (dinfo && dinfo->media_cd) {
+        scsi_cd_realize(dev, errp);
+    } else {
+        scsi_hd_realize(dev, errp);
+    }
+}
+
+static const SCSIReqOps scsi_disk_emulate_reqops = {
+    .size         = sizeof(SCSIDiskReq),
+    .free_req     = scsi_free_request,
+    .send_command = scsi_disk_emulate_command,
+    .read_data    = scsi_disk_emulate_read_data,
+    .write_data   = scsi_disk_emulate_write_data,
+    .get_buf      = scsi_get_buf,
+};
+
+static const SCSIReqOps scsi_disk_dma_reqops = {
+    .size         = sizeof(SCSIDiskReq),
+    .free_req     = scsi_free_request,
+    .send_command = scsi_disk_dma_command,
+    .read_data    = scsi_read_data,
+    .write_data   = scsi_write_data,
+    .get_buf      = scsi_get_buf,
+    .load_request = scsi_disk_load_request,
+    .save_request = scsi_disk_save_request,
+};
+
+static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
+    [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
+    [INQUIRY]                         = &scsi_disk_emulate_reqops,
+    [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
+    [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
+    [START_STOP]                      = &scsi_disk_emulate_reqops,
+    [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
+    [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
+    [READ_TOC]                        = &scsi_disk_emulate_reqops,
+    [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
+    [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
+    [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
+    [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
+    [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
+    [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
+    [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
+    [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
+    [SEEK_10]                         = &scsi_disk_emulate_reqops,
+    [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
+    [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
+    [UNMAP]                           = &scsi_disk_emulate_reqops,
+    [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
+    [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
+    [VERIFY_10]                       = &scsi_disk_emulate_reqops,
+    [VERIFY_12]                       = &scsi_disk_emulate_reqops,
+    [VERIFY_16]                       = &scsi_disk_emulate_reqops,
+
+    [READ_6]                          = &scsi_disk_dma_reqops,
+    [READ_10]                         = &scsi_disk_dma_reqops,
+    [READ_12]                         = &scsi_disk_dma_reqops,
+    [READ_16]                         = &scsi_disk_dma_reqops,
+    [WRITE_6]                         = &scsi_disk_dma_reqops,
+    [WRITE_10]                        = &scsi_disk_dma_reqops,
+    [WRITE_12]                        = &scsi_disk_dma_reqops,
+    [WRITE_16]                        = &scsi_disk_dma_reqops,
+    [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
+    [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
+    [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
+};
+
+static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
+                                     uint8_t *buf, void *hba_private)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
+    SCSIRequest *req;
+    const SCSIReqOps *ops;
+    uint8_t command;
+
+    command = buf[0];
+    ops = scsi_disk_reqops_dispatch[command];
+    if (!ops) {
+        ops = &scsi_disk_emulate_reqops;
+    }
+    req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
+
+#ifdef DEBUG_SCSI
+    DPRINTF("Command: lun=%d tag=0x%x data=0x%02x", lun, tag, buf[0]);
+    {
+        int i;
+        for (i = 1; i < scsi_cdb_length(buf); i++) {
+            printf(" 0x%02x", buf[i]);
+        }
+        printf("\n");
+    }
+#endif
+
+    return req;
+}
+
+#ifdef __linux__
+static int get_device_type(SCSIDiskState *s)
+{
+    uint8_t cmd[16];
+    uint8_t buf[36];
+    uint8_t sensebuf[8];
+    sg_io_hdr_t io_header;
+    int ret;
+
+    memset(cmd, 0, sizeof(cmd));
+    memset(buf, 0, sizeof(buf));
+    cmd[0] = INQUIRY;
+    cmd[4] = sizeof(buf);
+
+    memset(&io_header, 0, sizeof(io_header));
+    io_header.interface_id = 'S';
+    io_header.dxfer_direction = SG_DXFER_FROM_DEV;
+    io_header.dxfer_len = sizeof(buf);
+    io_header.dxferp = buf;
+    io_header.cmdp = cmd;
+    io_header.cmd_len = sizeof(cmd);
+    io_header.mx_sb_len = sizeof(sensebuf);
+    io_header.sbp = sensebuf;
+    io_header.timeout = 6000; /* XXX */
+
+    ret = blk_ioctl(s->qdev.conf.blk, SG_IO, &io_header);
+    if (ret < 0 || io_header.driver_status || io_header.host_status) {
+        return -1;
+    }
+    s->qdev.type = buf[0];
+    if (buf[1] & 0x80) {
+        s->features |= 1 << SCSI_DISK_F_REMOVABLE;
+    }
+    return 0;
+}
+
+static void scsi_block_realize(SCSIDevice *dev, Error **errp)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
+    int sg_version;
+    int rc;
+
+    if (!s->qdev.conf.blk) {
+        error_setg(errp, "drive property not set");
+        return;
+    }
+
+    /* check we are using a driver managing SG_IO (version 3 and after) */
+    rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
+    if (rc < 0) {
+        error_setg(errp, "cannot get SG_IO version number: %s.  "
+                     "Is this a SCSI device?",
+                     strerror(-rc));
+        return;
+    }
+    if (sg_version < 30000) {
+        error_setg(errp, "scsi generic interface too old");
+        return;
+    }
+
+    /* get device type from INQUIRY data */
+    rc = get_device_type(s);
+    if (rc < 0) {
+        error_setg(errp, "INQUIRY failed");
+        return;
+    }
+
+    /* Make a guess for the block size, we'll fix it when the guest sends.
+     * READ CAPACITY.  If they don't, they likely would assume these sizes
+     * anyway. (TODO: check in /sys).
+     */
+    if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
+        s->qdev.blocksize = 2048;
+    } else {
+        s->qdev.blocksize = 512;
+    }
+
+    /* Makes the scsi-block device not removable by using HMP and QMP eject
+     * command.
+     */
+    s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
+
+    scsi_realize(&s->qdev, errp);
+}
+
+static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
+{
+    switch (buf[0]) {
+    case READ_6:
+    case READ_10:
+    case READ_12:
+    case READ_16:
+    case VERIFY_10:
+    case VERIFY_12:
+    case VERIFY_16:
+    case WRITE_6:
+    case WRITE_10:
+    case WRITE_12:
+    case WRITE_16:
+    case WRITE_VERIFY_10:
+    case WRITE_VERIFY_12:
+    case WRITE_VERIFY_16:
+        /* If we are not using O_DIRECT, we might read stale data from the
+         * host cache if writes were made using other commands than these
+         * ones (such as WRITE SAME or EXTENDED COPY, etc.).  So, without
+         * O_DIRECT everything must go through SG_IO.
+         */
+        if (!(blk_get_flags(s->qdev.conf.blk) & BDRV_O_NOCACHE)) {
+            break;
+        }
+
+        /* MMC writing cannot be done via pread/pwrite, because it sometimes
+         * involves writing beyond the maximum LBA or to negative LBA (lead-in).
+         * And once you do these writes, reading from the block device is
+         * unreliable, too.  It is even possible that reads deliver random data
+         * from the host page cache (this is probably a Linux bug).
+         *
+         * We might use scsi_disk_dma_reqops as long as no writing commands are
+         * seen, but performance usually isn't paramount on optical media.  So,
+         * just make scsi-block operate the same as scsi-generic for them.
+         */
+        if (s->qdev.type != TYPE_ROM) {
+            return false;
+        }
+        break;
+
+    default:
+        break;
+    }
+
+    return true;
+}
+
+
+static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
+                                           uint32_t lun, uint8_t *buf,
+                                           void *hba_private)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
+
+    if (scsi_block_is_passthrough(s, buf)) {
+        return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
+                              hba_private);
+    } else {
+        return scsi_req_alloc(&scsi_disk_dma_reqops, &s->qdev, tag, lun,
+                              hba_private);
+    }
+}
+
+static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
+                                  uint8_t *buf, void *hba_private)
+{
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
+
+    if (scsi_block_is_passthrough(s, buf)) {
+        return scsi_bus_parse_cdb(&s->qdev, cmd, buf, hba_private);
+    } else {
+        return scsi_req_parse_cdb(&s->qdev, cmd, buf);
+    }
+}
+
+#endif
+
+#define DEFINE_SCSI_DISK_PROPERTIES()                                \
+    DEFINE_BLOCK_PROPERTIES(SCSIDiskState, qdev.conf),               \
+    DEFINE_PROP_STRING("ver", SCSIDiskState, version),               \
+    DEFINE_PROP_STRING("serial", SCSIDiskState, serial),             \
+    DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),             \
+    DEFINE_PROP_STRING("product", SCSIDiskState, product)
+
+static Property scsi_hd_properties[] = {
+    DEFINE_SCSI_DISK_PROPERTIES(),
+    DEFINE_PROP_BIT("removable", SCSIDiskState, features,
+                    SCSI_DISK_F_REMOVABLE, false),
+    DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
+                    SCSI_DISK_F_DPOFUA, false),
+    DEFINE_PROP_UINT64("wwn", SCSIDiskState, wwn, 0),
+    DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, port_wwn, 0),
+    DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
+    DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
+                       DEFAULT_MAX_UNMAP_SIZE),
+    DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
+                       DEFAULT_MAX_IO_SIZE),
+    DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static const VMStateDescription vmstate_scsi_disk_state = {
+    .name = "scsi-disk",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
+        VMSTATE_BOOL(media_changed, SCSIDiskState),
+        VMSTATE_BOOL(media_event, SCSIDiskState),
+        VMSTATE_BOOL(eject_request, SCSIDiskState),
+        VMSTATE_BOOL(tray_open, SCSIDiskState),
+        VMSTATE_BOOL(tray_locked, SCSIDiskState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static void scsi_hd_class_initfn(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
+
+    sc->realize      = scsi_hd_realize;
+    sc->alloc_req    = scsi_new_request;
+    sc->unit_attention_reported = scsi_disk_unit_attention_reported;
+    dc->fw_name = "disk";
+    dc->desc = "virtual SCSI disk";
+    dc->reset = scsi_disk_reset;
+    dc->props = scsi_hd_properties;
+    dc->vmsd  = &vmstate_scsi_disk_state;
+}
+
+static const TypeInfo scsi_hd_info = {
+    .name          = "scsi-hd",
+    .parent        = TYPE_SCSI_DEVICE,
+    .instance_size = sizeof(SCSIDiskState),
+    .class_init    = scsi_hd_class_initfn,
+};
+
+static Property scsi_cd_properties[] = {
+    DEFINE_SCSI_DISK_PROPERTIES(),
+    DEFINE_PROP_UINT64("wwn", SCSIDiskState, wwn, 0),
+    DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, port_wwn, 0),
+    DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
+    DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
+                       DEFAULT_MAX_IO_SIZE),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void scsi_cd_class_initfn(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
+
+    sc->realize      = scsi_cd_realize;
+    sc->alloc_req    = scsi_new_request;
+    sc->unit_attention_reported = scsi_disk_unit_attention_reported;
+    dc->fw_name = "disk";
+    dc->desc = "virtual SCSI CD-ROM";
+    dc->reset = scsi_disk_reset;
+    dc->props = scsi_cd_properties;
+    dc->vmsd  = &vmstate_scsi_disk_state;
+}
+
+static const TypeInfo scsi_cd_info = {
+    .name          = "scsi-cd",
+    .parent        = TYPE_SCSI_DEVICE,
+    .instance_size = sizeof(SCSIDiskState),
+    .class_init    = scsi_cd_class_initfn,
+};
+
+#ifdef __linux__
+static Property scsi_block_properties[] = {
+    DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void scsi_block_class_initfn(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
+
+    sc->realize      = scsi_block_realize;
+    sc->alloc_req    = scsi_block_new_request;
+    sc->parse_cdb    = scsi_block_parse_cdb;
+    dc->fw_name = "disk";
+    dc->desc = "SCSI block device passthrough";
+    dc->reset = scsi_disk_reset;
+    dc->props = scsi_block_properties;
+    dc->vmsd  = &vmstate_scsi_disk_state;
+}
+
+static const TypeInfo scsi_block_info = {
+    .name          = "scsi-block",
+    .parent        = TYPE_SCSI_DEVICE,
+    .instance_size = sizeof(SCSIDiskState),
+    .class_init    = scsi_block_class_initfn,
+};
+#endif
+
+static Property scsi_disk_properties[] = {
+    DEFINE_SCSI_DISK_PROPERTIES(),
+    DEFINE_PROP_BIT("removable", SCSIDiskState, features,
+                    SCSI_DISK_F_REMOVABLE, false),
+    DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
+                    SCSI_DISK_F_DPOFUA, false),
+    DEFINE_PROP_UINT64("wwn", SCSIDiskState, wwn, 0),
+    DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, port_wwn, 0),
+    DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
+    DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
+                       DEFAULT_MAX_UNMAP_SIZE),
+    DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
+                       DEFAULT_MAX_IO_SIZE),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void scsi_disk_class_initfn(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
+
+    sc->realize      = scsi_disk_realize;
+    sc->alloc_req    = scsi_new_request;
+    sc->unit_attention_reported = scsi_disk_unit_attention_reported;
+    dc->fw_name = "disk";
+    dc->desc = "virtual SCSI disk or CD-ROM (legacy)";
+    dc->reset = scsi_disk_reset;
+    dc->props = scsi_disk_properties;
+    dc->vmsd  = &vmstate_scsi_disk_state;
+}
+
+static const TypeInfo scsi_disk_info = {
+    .name          = "scsi-disk",
+    .parent        = TYPE_SCSI_DEVICE,
+    .instance_size = sizeof(SCSIDiskState),
+    .class_init    = scsi_disk_class_initfn,
+};
+
+static void scsi_disk_register_types(void)
+{
+    type_register_static(&scsi_hd_info);
+    type_register_static(&scsi_cd_info);
+#ifdef __linux__
+    type_register_static(&scsi_block_info);
+#endif
+    type_register_static(&scsi_disk_info);
+}
+
+type_init(scsi_disk_register_types)
diff --git a/qemu/hw/scsi/scsi-generic.c b/qemu/hw/scsi/scsi-generic.c
new file mode 100644
index 000000000..e53470f85
--- /dev/null
+++ b/qemu/hw/scsi/scsi-generic.c
@@ -0,0 +1,496 @@
+/*
+ * Generic SCSI Device support
+ *
+ * Copyright (c) 2007 Bull S.A.S.
+ * Based on code by Paul Brook
+ * Based on code by Fabrice Bellard
+ *
+ * Written by Laurent Vivier <Laurent.Vivier@bull.net>
+ *
+ * This code is licensed under the LGPL.
+ *
+ */
+
+#include "qemu-common.h"
+#include "qemu/error-report.h"
+#include "hw/scsi/scsi.h"
+#include "sysemu/block-backend.h"
+#include "sysemu/blockdev.h"
+
+#ifdef __linux__
+
+//#define DEBUG_SCSI
+
+#ifdef DEBUG_SCSI
+#define DPRINTF(fmt, ...) \
+do { printf("scsi-generic: " fmt , ## __VA_ARGS__); } while (0)
+#else
+#define DPRINTF(fmt, ...) do {} while(0)
+#endif
+
+#define BADF(fmt, ...) \
+do { fprintf(stderr, "scsi-generic: " fmt , ## __VA_ARGS__); } while (0)
+
+#include <stdio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <scsi/sg.h>
+#include "block/scsi.h"
+
+#define SG_ERR_DRIVER_TIMEOUT  0x06
+#define SG_ERR_DRIVER_SENSE    0x08
+
+#define SG_ERR_DID_OK          0x00
+#define SG_ERR_DID_NO_CONNECT  0x01
+#define SG_ERR_DID_BUS_BUSY    0x02
+#define SG_ERR_DID_TIME_OUT    0x03
+
+#ifndef MAX_UINT
+#define MAX_UINT ((unsigned int)-1)
+#endif
+
+typedef struct SCSIGenericReq {
+    SCSIRequest req;
+    uint8_t *buf;
+    int buflen;
+    int len;
+    sg_io_hdr_t io_header;
+} SCSIGenericReq;
+
+static void scsi_generic_save_request(QEMUFile *f, SCSIRequest *req)
+{
+    SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
+
+    qemu_put_sbe32s(f, &r->buflen);
+    if (r->buflen && r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+        assert(!r->req.sg);
+        qemu_put_buffer(f, r->buf, r->req.cmd.xfer);
+    }
+}
+
+static void scsi_generic_load_request(QEMUFile *f, SCSIRequest *req)
+{
+    SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
+
+    qemu_get_sbe32s(f, &r->buflen);
+    if (r->buflen && r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+        assert(!r->req.sg);
+        qemu_get_buffer(f, r->buf, r->req.cmd.xfer);
+    }
+}
+
+static void scsi_free_request(SCSIRequest *req)
+{
+    SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
+
+    g_free(r->buf);
+}
+
+/* Helper function for command completion.  */
+static void scsi_command_complete(void *opaque, int ret)
+{
+    int status;
+    SCSIGenericReq *r = (SCSIGenericReq *)opaque;
+
+    r->req.aiocb = NULL;
+    if (r->req.io_canceled) {
+        scsi_req_cancel_complete(&r->req);
+        goto done;
+    }
+    if (r->io_header.driver_status & SG_ERR_DRIVER_SENSE) {
+        r->req.sense_len = r->io_header.sb_len_wr;
+    }
+
+    if (ret != 0) {
+        switch (ret) {
+        case -EDOM:
+            status = TASK_SET_FULL;
+            break;
+        case -ENOMEM:
+            status = CHECK_CONDITION;
+            scsi_req_build_sense(&r->req, SENSE_CODE(TARGET_FAILURE));
+            break;
+        default:
+            status = CHECK_CONDITION;
+            scsi_req_build_sense(&r->req, SENSE_CODE(IO_ERROR));
+            break;
+        }
+    } else {
+        if (r->io_header.host_status == SG_ERR_DID_NO_CONNECT ||
+            r->io_header.host_status == SG_ERR_DID_BUS_BUSY ||
+            r->io_header.host_status == SG_ERR_DID_TIME_OUT ||
+            (r->io_header.driver_status & SG_ERR_DRIVER_TIMEOUT)) {
+            status = BUSY;
+            BADF("Driver Timeout\n");
+        } else if (r->io_header.host_status) {
+            status = CHECK_CONDITION;
+            scsi_req_build_sense(&r->req, SENSE_CODE(I_T_NEXUS_LOSS));
+        } else if (r->io_header.status) {
+            status = r->io_header.status;
+        } else if (r->io_header.driver_status & SG_ERR_DRIVER_SENSE) {
+            status = CHECK_CONDITION;
+        } else {
+            status = GOOD;
+        }
+    }
+    DPRINTF("Command complete 0x%p tag=0x%x status=%d\n",
+            r, r->req.tag, status);
+
+    scsi_req_complete(&r->req, status);
+done:
+    scsi_req_unref(&r->req);
+}
+
+static int execute_command(BlockBackend *blk,
+                           SCSIGenericReq *r, int direction,
+                           BlockCompletionFunc *complete)
+{
+    r->io_header.interface_id = 'S';
+    r->io_header.dxfer_direction = direction;
+    r->io_header.dxferp = r->buf;
+    r->io_header.dxfer_len = r->buflen;
+    r->io_header.cmdp = r->req.cmd.buf;
+    r->io_header.cmd_len = r->req.cmd.len;
+    r->io_header.mx_sb_len = sizeof(r->req.sense);
+    r->io_header.sbp = r->req.sense;
+    r->io_header.timeout = MAX_UINT;
+    r->io_header.usr_ptr = r;
+    r->io_header.flags |= SG_FLAG_DIRECT_IO;
+
+    r->req.aiocb = blk_aio_ioctl(blk, SG_IO, &r->io_header, complete, r);
+    if (r->req.aiocb == NULL) {
+        return -EIO;
+    }
+
+    return 0;
+}
+
+static void scsi_read_complete(void * opaque, int ret)
+{
+    SCSIGenericReq *r = (SCSIGenericReq *)opaque;
+    SCSIDevice *s = r->req.dev;
+    int len;
+
+    r->req.aiocb = NULL;
+    if (ret || r->req.io_canceled) {
+        scsi_command_complete(r, ret);
+        return;
+    }
+    len = r->io_header.dxfer_len - r->io_header.resid;
+    DPRINTF("Data ready tag=0x%x len=%d\n", r->req.tag, len);
+
+    r->len = -1;
+    if (len == 0) {
+        scsi_command_complete(r, 0);
+    } else {
+        /* Snoop READ CAPACITY output to set the blocksize.  */
+        if (r->req.cmd.buf[0] == READ_CAPACITY_10 &&
+            (ldl_be_p(&r->buf[0]) != 0xffffffffU || s->max_lba == 0)) {
+            s->blocksize = ldl_be_p(&r->buf[4]);
+            s->max_lba = ldl_be_p(&r->buf[0]) & 0xffffffffULL;
+        } else if (r->req.cmd.buf[0] == SERVICE_ACTION_IN_16 &&
+                   (r->req.cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
+            s->blocksize = ldl_be_p(&r->buf[8]);
+            s->max_lba = ldq_be_p(&r->buf[0]);
+        }
+        blk_set_guest_block_size(s->conf.blk, s->blocksize);
+
+        scsi_req_data(&r->req, len);
+        scsi_req_unref(&r->req);
+    }
+}
+
+/* Read more data from scsi device into buffer.  */
+static void scsi_read_data(SCSIRequest *req)
+{
+    SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
+    SCSIDevice *s = r->req.dev;
+    int ret;
+
+    DPRINTF("scsi_read_data 0x%x\n", req->tag);
+
+    /* The request is used as the AIO opaque value, so add a ref.  */
+    scsi_req_ref(&r->req);
+    if (r->len == -1) {
+        scsi_command_complete(r, 0);
+        return;
+    }
+
+    ret = execute_command(s->conf.blk, r, SG_DXFER_FROM_DEV,
+                          scsi_read_complete);
+    if (ret < 0) {
+        scsi_command_complete(r, ret);
+    }
+}
+
+static void scsi_write_complete(void * opaque, int ret)
+{
+    SCSIGenericReq *r = (SCSIGenericReq *)opaque;
+    SCSIDevice *s = r->req.dev;
+
+    DPRINTF("scsi_write_complete() ret = %d\n", ret);
+    r->req.aiocb = NULL;
+    if (ret || r->req.io_canceled) {
+        scsi_command_complete(r, ret);
+        return;
+    }
+
+    if (r->req.cmd.buf[0] == MODE_SELECT && r->req.cmd.buf[4] == 12 &&
+        s->type == TYPE_TAPE) {
+        s->blocksize = (r->buf[9] << 16) | (r->buf[10] << 8) | r->buf[11];
+        DPRINTF("block size %d\n", s->blocksize);
+    }
+
+    scsi_command_complete(r, ret);
+}
+
+/* Write data to a scsi device.  Returns nonzero on failure.
+   The transfer may complete asynchronously.  */
+static void scsi_write_data(SCSIRequest *req)
+{
+    SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
+    SCSIDevice *s = r->req.dev;
+    int ret;
+
+    DPRINTF("scsi_write_data 0x%x\n", req->tag);
+    if (r->len == 0) {
+        r->len = r->buflen;
+        scsi_req_data(&r->req, r->len);
+        return;
+    }
+
+    /* The request is used as the AIO opaque value, so add a ref.  */
+    scsi_req_ref(&r->req);
+    ret = execute_command(s->conf.blk, r, SG_DXFER_TO_DEV, scsi_write_complete);
+    if (ret < 0) {
+        scsi_command_complete(r, ret);
+    }
+}
+
+/* Return a pointer to the data buffer.  */
+static uint8_t *scsi_get_buf(SCSIRequest *req)
+{
+    SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
+
+    return r->buf;
+}
+
+/* Execute a scsi command.  Returns the length of the data expected by the
+   command.  This will be Positive for data transfers from the device
+   (eg. disk reads), negative for transfers to the device (eg. disk writes),
+   and zero if the command does not transfer any data.  */
+
+static int32_t scsi_send_command(SCSIRequest *req, uint8_t *cmd)
+{
+    SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
+    SCSIDevice *s = r->req.dev;
+    int ret;
+
+#ifdef DEBUG_SCSI
+    {
+        int i;
+        for (i = 1; i < r->req.cmd.len; i++) {
+            printf(" 0x%02x", cmd[i]);
+        }
+        printf("\n");
+    }
+#endif
+
+    if (r->req.cmd.xfer == 0) {
+        g_free(r->buf);
+        r->buflen = 0;
+        r->buf = NULL;
+        /* The request is used as the AIO opaque value, so add a ref.  */
+        scsi_req_ref(&r->req);
+        ret = execute_command(s->conf.blk, r, SG_DXFER_NONE,
+                              scsi_command_complete);
+        if (ret < 0) {
+            scsi_command_complete(r, ret);
+            return 0;
+        }
+        return 0;
+    }
+
+    if (r->buflen != r->req.cmd.xfer) {
+        g_free(r->buf);
+        r->buf = g_malloc(r->req.cmd.xfer);
+        r->buflen = r->req.cmd.xfer;
+    }
+
+    memset(r->buf, 0, r->buflen);
+    r->len = r->req.cmd.xfer;
+    if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
+        r->len = 0;
+        return -r->req.cmd.xfer;
+    } else {
+        return r->req.cmd.xfer;
+    }
+}
+
+static int get_stream_blocksize(BlockBackend *blk)
+{
+    uint8_t cmd[6];
+    uint8_t buf[12];
+    uint8_t sensebuf[8];
+    sg_io_hdr_t io_header;
+    int ret;
+
+    memset(cmd, 0, sizeof(cmd));
+    memset(buf, 0, sizeof(buf));
+    cmd[0] = MODE_SENSE;
+    cmd[4] = sizeof(buf);
+
+    memset(&io_header, 0, sizeof(io_header));
+    io_header.interface_id = 'S';
+    io_header.dxfer_direction = SG_DXFER_FROM_DEV;
+    io_header.dxfer_len = sizeof(buf);
+    io_header.dxferp = buf;
+    io_header.cmdp = cmd;
+    io_header.cmd_len = sizeof(cmd);
+    io_header.mx_sb_len = sizeof(sensebuf);
+    io_header.sbp = sensebuf;
+    io_header.timeout = 6000; /* XXX */
+
+    ret = blk_ioctl(blk, SG_IO, &io_header);
+    if (ret < 0 || io_header.driver_status || io_header.host_status) {
+        return -1;
+    }
+    return (buf[9] << 16) | (buf[10] << 8) | buf[11];
+}
+
+static void scsi_generic_reset(DeviceState *dev)
+{
+    SCSIDevice *s = SCSI_DEVICE(dev);
+
+    scsi_device_purge_requests(s, SENSE_CODE(RESET));
+}
+
+static void scsi_generic_realize(SCSIDevice *s, Error **errp)
+{
+    int rc;
+    int sg_version;
+    struct sg_scsi_id scsiid;
+
+    if (!s->conf.blk) {
+        error_setg(errp, "drive property not set");
+        return;
+    }
+
+    if (blk_get_on_error(s->conf.blk, 0) != BLOCKDEV_ON_ERROR_ENOSPC) {
+        error_setg(errp, "Device doesn't support drive option werror");
+        return;
+    }
+    if (blk_get_on_error(s->conf.blk, 1) != BLOCKDEV_ON_ERROR_REPORT) {
+        error_setg(errp, "Device doesn't support drive option rerror");
+        return;
+    }
+
+    /* check we are using a driver managing SG_IO (version 3 and after */
+    rc = blk_ioctl(s->conf.blk, SG_GET_VERSION_NUM, &sg_version);
+    if (rc < 0) {
+        error_setg(errp, "cannot get SG_IO version number: %s.  "
+                         "Is this a SCSI device?",
+                         strerror(-rc));
+        return;
+    }
+    if (sg_version < 30000) {
+        error_setg(errp, "scsi generic interface too old");
+        return;
+    }
+
+    /* get LUN of the /dev/sg? */
+    if (blk_ioctl(s->conf.blk, SG_GET_SCSI_ID, &scsiid)) {
+        error_setg(errp, "SG_GET_SCSI_ID ioctl failed");
+        return;
+    }
+
+    /* define device state */
+    s->type = scsiid.scsi_type;
+    DPRINTF("device type %d\n", s->type);
+
+    switch (s->type) {
+    case TYPE_TAPE:
+        s->blocksize = get_stream_blocksize(s->conf.blk);
+        if (s->blocksize == -1) {
+            s->blocksize = 0;
+        }
+        break;
+
+        /* Make a guess for block devices, we'll fix it when the guest sends.
+         * READ CAPACITY.  If they don't, they likely would assume these sizes
+         * anyway. (TODO: they could also send MODE SENSE).
+         */
+    case TYPE_ROM:
+    case TYPE_WORM:
+        s->blocksize = 2048;
+        break;
+    default:
+        s->blocksize = 512;
+        break;
+    }
+
+    DPRINTF("block size %d\n", s->blocksize);
+}
+
+const SCSIReqOps scsi_generic_req_ops = {
+    .size         = sizeof(SCSIGenericReq),
+    .free_req     = scsi_free_request,
+    .send_command = scsi_send_command,
+    .read_data    = scsi_read_data,
+    .write_data   = scsi_write_data,
+    .get_buf      = scsi_get_buf,
+    .load_request = scsi_generic_load_request,
+    .save_request = scsi_generic_save_request,
+};
+
+static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
+                                     uint8_t *buf, void *hba_private)
+{
+    SCSIRequest *req;
+
+    req = scsi_req_alloc(&scsi_generic_req_ops, d, tag, lun, hba_private);
+    return req;
+}
+
+static Property scsi_generic_properties[] = {
+    DEFINE_PROP_DRIVE("drive", SCSIDevice, conf.blk),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static int scsi_generic_parse_cdb(SCSIDevice *dev, SCSICommand *cmd,
+                                  uint8_t *buf, void *hba_private)
+{
+    return scsi_bus_parse_cdb(dev, cmd, buf, hba_private);
+}
+
+static void scsi_generic_class_initfn(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
+
+    sc->realize      = scsi_generic_realize;
+    sc->alloc_req    = scsi_new_request;
+    sc->parse_cdb    = scsi_generic_parse_cdb;
+    dc->fw_name = "disk";
+    dc->desc = "pass through generic scsi device (/dev/sg*)";
+    dc->reset = scsi_generic_reset;
+    dc->props = scsi_generic_properties;
+    dc->vmsd  = &vmstate_scsi_device;
+}
+
+static const TypeInfo scsi_generic_info = {
+    .name          = "scsi-generic",
+    .parent        = TYPE_SCSI_DEVICE,
+    .instance_size = sizeof(SCSIDevice),
+    .class_init    = scsi_generic_class_initfn,
+};
+
+static void scsi_generic_register_types(void)
+{
+    type_register_static(&scsi_generic_info);
+}
+
+type_init(scsi_generic_register_types)
+
+#endif /* __linux__ */
diff --git a/qemu/hw/scsi/spapr_vscsi.c b/qemu/hw/scsi/spapr_vscsi.c
new file mode 100644
index 000000000..891424fae
--- /dev/null
+++ b/qemu/hw/scsi/spapr_vscsi.c
@@ -0,0 +1,1302 @@
+/*
+ * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
+ *
+ * PAPR Virtual SCSI, aka ibmvscsi
+ *
+ * Copyright (c) 2010,2011 Benjamin Herrenschmidt, IBM Corporation.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ * TODO:
+ *
+ *  - Cleanups :-)
+ *  - Sort out better how to assign devices to VSCSI instances
+ *  - Fix residual counts
+ *  - Add indirect descriptors support
+ *  - Maybe do autosense (PAPR seems to mandate it, linux doesn't care)
+ */
+#include "hw/hw.h"
+#include "hw/scsi/scsi.h"
+#include "block/scsi.h"
+#include "srp.h"
+#include "hw/qdev.h"
+#include "hw/ppc/spapr.h"
+#include "hw/ppc/spapr_vio.h"
+#include "viosrp.h"
+
+#include <libfdt.h>
+
+/*#define DEBUG_VSCSI*/
+
+#ifdef DEBUG_VSCSI
+#define DPRINTF(fmt, ...) \
+    do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
+#else
+#define DPRINTF(fmt, ...) \
+    do { } while (0)
+#endif
+
+/*
+ * Virtual SCSI device
+ */
+
+/* Random numbers */
+#define VSCSI_MAX_SECTORS       4096
+#define VSCSI_REQ_LIMIT         24
+
+#define SRP_RSP_SENSE_DATA_LEN  18
+
+#define SRP_REPORT_LUNS_WLUN    0xc10100000000000ULL
+
+typedef union vscsi_crq {
+    struct viosrp_crq s;
+    uint8_t raw[16];
+} vscsi_crq;
+
+typedef struct vscsi_req {
+    vscsi_crq               crq;
+    union viosrp_iu         iu;
+
+    /* SCSI request tracking */
+    SCSIRequest             *sreq;
+    uint32_t                qtag; /* qemu tag != srp tag */
+    bool                    active;
+    bool                    writing;
+    bool                    dma_error;
+    uint32_t                data_len;
+    uint32_t                senselen;
+    uint8_t                 sense[SCSI_SENSE_BUF_SIZE];
+
+    /* RDMA related bits */
+    uint8_t                 dma_fmt;
+    uint16_t                local_desc;
+    uint16_t                total_desc;
+    uint16_t                cdb_offset;
+    uint16_t                cur_desc_num;
+    uint16_t                cur_desc_offset;
+} vscsi_req;
+
+#define TYPE_VIO_SPAPR_VSCSI_DEVICE "spapr-vscsi"
+#define VIO_SPAPR_VSCSI_DEVICE(obj) \
+     OBJECT_CHECK(VSCSIState, (obj), TYPE_VIO_SPAPR_VSCSI_DEVICE)
+
+typedef struct {
+    VIOsPAPRDevice vdev;
+    SCSIBus bus;
+    vscsi_req reqs[VSCSI_REQ_LIMIT];
+} VSCSIState;
+
+static struct vscsi_req *vscsi_get_req(VSCSIState *s)
+{
+    vscsi_req *req;
+    int i;
+
+    for (i = 0; i < VSCSI_REQ_LIMIT; i++) {
+        req = &s->reqs[i];
+        if (!req->active) {
+            memset(req, 0, sizeof(*req));
+            req->qtag = i;
+            req->active = 1;
+            return req;
+        }
+    }
+    return NULL;
+}
+
+static struct vscsi_req *vscsi_find_req(VSCSIState *s, uint64_t srp_tag)
+{
+    vscsi_req *req;
+    int i;
+
+    for (i = 0; i < VSCSI_REQ_LIMIT; i++) {
+        req = &s->reqs[i];
+        if (req->iu.srp.cmd.tag == srp_tag) {
+            return req;
+        }
+    }
+    return NULL;
+}
+
+static void vscsi_put_req(vscsi_req *req)
+{
+    if (req->sreq != NULL) {
+        scsi_req_unref(req->sreq);
+    }
+    req->sreq = NULL;
+    req->active = 0;
+}
+
+static SCSIDevice *vscsi_device_find(SCSIBus *bus, uint64_t srp_lun, int *lun)
+{
+    int channel = 0, id = 0;
+
+retry:
+    switch (srp_lun >> 62) {
+    case 0:
+        if ((srp_lun >> 56) != 0) {
+            channel = (srp_lun >> 56) & 0x3f;
+            id = (srp_lun >> 48) & 0xff;
+            srp_lun <<= 16;
+            goto retry;
+        }
+        *lun = (srp_lun >> 48) & 0xff;
+        break;
+
+    case 1:
+        *lun = (srp_lun >> 48) & 0x3fff;
+        break;
+    case 2:
+        channel = (srp_lun >> 53) & 0x7;
+        id = (srp_lun >> 56) & 0x3f;
+        *lun = (srp_lun >> 48) & 0x1f;
+        break;
+    case 3:
+        *lun = -1;
+        return NULL;
+    default:
+        abort();
+    }
+
+    return scsi_device_find(bus, channel, id, *lun);
+}
+
+static int vscsi_send_iu(VSCSIState *s, vscsi_req *req,
+                         uint64_t length, uint8_t format)
+{
+    long rc, rc1;
+
+    /* First copy the SRP */
+    rc = spapr_vio_dma_write(&s->vdev, req->crq.s.IU_data_ptr,
+                             &req->iu, length);
+    if (rc) {
+        fprintf(stderr, "vscsi_send_iu: DMA write failure !\n");
+    }
+
+    req->crq.s.valid = 0x80;
+    req->crq.s.format = format;
+    req->crq.s.reserved = 0x00;
+    req->crq.s.timeout = cpu_to_be16(0x0000);
+    req->crq.s.IU_length = cpu_to_be16(length);
+    req->crq.s.IU_data_ptr = req->iu.srp.rsp.tag; /* right byte order */
+
+    if (rc == 0) {
+        req->crq.s.status = VIOSRP_OK;
+    } else {
+        req->crq.s.status = VIOSRP_ADAPTER_FAIL;
+    }
+
+    rc1 = spapr_vio_send_crq(&s->vdev, req->crq.raw);
+    if (rc1) {
+        fprintf(stderr, "vscsi_send_iu: Error sending response\n");
+        return rc1;
+    }
+
+    return rc;
+}
+
+static void vscsi_makeup_sense(VSCSIState *s, vscsi_req *req,
+                               uint8_t key, uint8_t asc, uint8_t ascq)
+{
+    req->senselen = SRP_RSP_SENSE_DATA_LEN;
+
+    /* Valid bit and 'current errors' */
+    req->sense[0] = (0x1 << 7 | 0x70);
+    /* Sense key */
+    req->sense[2] = key;
+    /* Additional sense length */
+    req->sense[7] = 0xa; /* 10 bytes */
+    /* Additional sense code */
+    req->sense[12] = asc;
+    req->sense[13] = ascq;
+}
+
+static int vscsi_send_rsp(VSCSIState *s, vscsi_req *req,
+                          uint8_t status, int32_t res_in, int32_t res_out)
+{
+    union viosrp_iu *iu = &req->iu;
+    uint64_t tag = iu->srp.rsp.tag;
+    int total_len = sizeof(iu->srp.rsp);
+    uint8_t sol_not = iu->srp.cmd.sol_not;
+
+    DPRINTF("VSCSI: Sending resp status: 0x%x, "
+            "res_in: %d, res_out: %d\n", status, res_in, res_out);
+
+    memset(iu, 0, sizeof(struct srp_rsp));
+    iu->srp.rsp.opcode = SRP_RSP;
+    iu->srp.rsp.req_lim_delta = cpu_to_be32(1);
+    iu->srp.rsp.tag = tag;
+
+    /* Handle residuals */
+    if (res_in < 0) {
+        iu->srp.rsp.flags |= SRP_RSP_FLAG_DIUNDER;
+        res_in = -res_in;
+    } else if (res_in) {
+        iu->srp.rsp.flags |= SRP_RSP_FLAG_DIOVER;
+    }
+    if (res_out < 0) {
+        iu->srp.rsp.flags |= SRP_RSP_FLAG_DOUNDER;
+        res_out = -res_out;
+    } else if (res_out) {
+        iu->srp.rsp.flags |= SRP_RSP_FLAG_DOOVER;
+    }
+    iu->srp.rsp.data_in_res_cnt = cpu_to_be32(res_in);
+    iu->srp.rsp.data_out_res_cnt = cpu_to_be32(res_out);
+
+    /* We don't do response data */
+    /* iu->srp.rsp.flags &= ~SRP_RSP_FLAG_RSPVALID; */
+    iu->srp.rsp.resp_data_len = cpu_to_be32(0);
+
+    /* Handle success vs. failure */
+    iu->srp.rsp.status = status;
+    if (status) {
+        iu->srp.rsp.sol_not = (sol_not & 0x04) >> 2;
+        if (req->senselen) {
+            req->iu.srp.rsp.flags |= SRP_RSP_FLAG_SNSVALID;
+            req->iu.srp.rsp.sense_data_len = cpu_to_be32(req->senselen);
+            memcpy(req->iu.srp.rsp.data, req->sense, req->senselen);
+            total_len += req->senselen;
+        }
+    } else {
+        iu->srp.rsp.sol_not = (sol_not & 0x02) >> 1;
+    }
+
+    vscsi_send_iu(s, req, total_len, VIOSRP_SRP_FORMAT);
+    return 0;
+}
+
+static inline struct srp_direct_buf vscsi_swap_desc(struct srp_direct_buf desc)
+{
+    desc.va = be64_to_cpu(desc.va);
+    desc.len = be32_to_cpu(desc.len);
+    return desc;
+}
+
+static int vscsi_fetch_desc(VSCSIState *s, struct vscsi_req *req,
+                            unsigned n, unsigned buf_offset,
+                            struct srp_direct_buf *ret)
+{
+    struct srp_cmd *cmd = &req->iu.srp.cmd;
+
+    switch (req->dma_fmt) {
+    case SRP_NO_DATA_DESC: {
+        DPRINTF("VSCSI: no data descriptor\n");
+        return 0;
+    }
+    case SRP_DATA_DESC_DIRECT: {
+        memcpy(ret, cmd->add_data + req->cdb_offset, sizeof(*ret));
+        assert(req->cur_desc_num == 0);
+        DPRINTF("VSCSI: direct segment\n");
+        break;
+    }
+    case SRP_DATA_DESC_INDIRECT: {
+        struct srp_indirect_buf *tmp = (struct srp_indirect_buf *)
+                                       (cmd->add_data + req->cdb_offset);
+        if (n < req->local_desc) {
+            *ret = tmp->desc_list[n];
+            DPRINTF("VSCSI: indirect segment local tag=0x%x desc#%d/%d\n",
+                    req->qtag, n, req->local_desc);
+
+        } else if (n < req->total_desc) {
+            int rc;
+            struct srp_direct_buf tbl_desc = vscsi_swap_desc(tmp->table_desc);
+            unsigned desc_offset = n * sizeof(struct srp_direct_buf);
+
+            if (desc_offset >= tbl_desc.len) {
+                DPRINTF("VSCSI:   #%d is ouf of range (%d bytes)\n",
+                        n, desc_offset);
+                return -1;
+            }
+            rc = spapr_vio_dma_read(&s->vdev, tbl_desc.va + desc_offset,
+                                    ret, sizeof(struct srp_direct_buf));
+            if (rc) {
+                DPRINTF("VSCSI: spapr_vio_dma_read -> %d reading ext_desc\n",
+                        rc);
+                return -1;
+            }
+            DPRINTF("VSCSI: indirect segment ext. tag=0x%x desc#%d/%d { va=%"PRIx64" len=%x }\n",
+                    req->qtag, n, req->total_desc, tbl_desc.va, tbl_desc.len);
+        } else {
+            DPRINTF("VSCSI:   Out of descriptors !\n");
+            return 0;
+        }
+        break;
+    }
+    default:
+        fprintf(stderr, "VSCSI:   Unknown format %x\n", req->dma_fmt);
+        return -1;
+    }
+
+    *ret = vscsi_swap_desc(*ret);
+    if (buf_offset > ret->len) {
+        DPRINTF("   offset=%x is out of a descriptor #%d boundary=%x\n",
+                buf_offset, req->cur_desc_num, ret->len);
+        return -1;
+    }
+    ret->va += buf_offset;
+    ret->len -= buf_offset;
+
+    DPRINTF("   cur=%d offs=%x ret { va=%"PRIx64" len=%x }\n",
+            req->cur_desc_num, req->cur_desc_offset, ret->va, ret->len);
+
+    return ret->len ? 1 : 0;
+}
+
+static int vscsi_srp_direct_data(VSCSIState *s, vscsi_req *req,
+                                 uint8_t *buf, uint32_t len)
+{
+    struct srp_direct_buf md;
+    uint32_t llen;
+    int rc = 0;
+
+    rc = vscsi_fetch_desc(s, req, req->cur_desc_num, req->cur_desc_offset, &md);
+    if (rc < 0) {
+        return -1;
+    } else if (rc == 0) {
+        return 0;
+    }
+
+    llen = MIN(len, md.len);
+    if (llen) {
+        if (req->writing) { /* writing = to device = reading from memory */
+            rc = spapr_vio_dma_read(&s->vdev, md.va, buf, llen);
+        } else {
+            rc = spapr_vio_dma_write(&s->vdev, md.va, buf, llen);
+        }
+    }
+
+    if (rc) {
+        return -1;
+    }
+    req->cur_desc_offset += llen;
+
+    return llen;
+}
+
+static int vscsi_srp_indirect_data(VSCSIState *s, vscsi_req *req,
+                                   uint8_t *buf, uint32_t len)
+{
+    struct srp_direct_buf md;
+    int rc = 0;
+    uint32_t llen, total = 0;
+
+    DPRINTF("VSCSI: indirect segment 0x%x bytes\n", len);
+
+    /* While we have data ... */
+    while (len) {
+        rc = vscsi_fetch_desc(s, req, req->cur_desc_num, req->cur_desc_offset, &md);
+        if (rc < 0) {
+            return -1;
+        } else if (rc == 0) {
+            break;
+        }
+
+        /* Perform transfer */
+        llen = MIN(len, md.len);
+        if (req->writing) { /* writing = to device = reading from memory */
+            rc = spapr_vio_dma_read(&s->vdev, md.va, buf, llen);
+        } else {
+            rc = spapr_vio_dma_write(&s->vdev, md.va, buf, llen);
+        }
+        if (rc) {
+            DPRINTF("VSCSI: spapr_vio_dma_r/w(%d) -> %d\n", req->writing, rc);
+            break;
+        }
+        DPRINTF("VSCSI:     data: %02x %02x %02x %02x...\n",
+                buf[0], buf[1], buf[2], buf[3]);
+
+        len -= llen;
+        buf += llen;
+
+        total += llen;
+
+        /* Update current position in the current descriptor */
+        req->cur_desc_offset += llen;
+        if (md.len == llen) {
+            /* Go to the next descriptor if the current one finished */
+            ++req->cur_desc_num;
+            req->cur_desc_offset = 0;
+        }
+    }
+
+    return rc ? -1 : total;
+}
+
+static int vscsi_srp_transfer_data(VSCSIState *s, vscsi_req *req,
+                                   int writing, uint8_t *buf, uint32_t len)
+{
+    int err = 0;
+
+    switch (req->dma_fmt) {
+    case SRP_NO_DATA_DESC:
+        DPRINTF("VSCSI: no data desc transfer, skipping 0x%x bytes\n", len);
+        break;
+    case SRP_DATA_DESC_DIRECT:
+        err = vscsi_srp_direct_data(s, req, buf, len);
+        break;
+    case SRP_DATA_DESC_INDIRECT:
+        err = vscsi_srp_indirect_data(s, req, buf, len);
+        break;
+    }
+    return err;
+}
+
+/* Bits from linux srp */
+static int data_out_desc_size(struct srp_cmd *cmd)
+{
+    int size = 0;
+    uint8_t fmt = cmd->buf_fmt >> 4;
+
+    switch (fmt) {
+    case SRP_NO_DATA_DESC:
+        break;
+    case SRP_DATA_DESC_DIRECT:
+        size = sizeof(struct srp_direct_buf);
+        break;
+    case SRP_DATA_DESC_INDIRECT:
+        size = sizeof(struct srp_indirect_buf) +
+            sizeof(struct srp_direct_buf)*cmd->data_out_desc_cnt;
+        break;
+    default:
+        break;
+    }
+    return size;
+}
+
+static int vscsi_preprocess_desc(vscsi_req *req)
+{
+    struct srp_cmd *cmd = &req->iu.srp.cmd;
+
+    req->cdb_offset = cmd->add_cdb_len & ~3;
+
+    if (req->writing) {
+        req->dma_fmt = cmd->buf_fmt >> 4;
+    } else {
+        req->cdb_offset += data_out_desc_size(cmd);
+        req->dma_fmt = cmd->buf_fmt & ((1U << 4) - 1);
+    }
+
+    switch (req->dma_fmt) {
+    case SRP_NO_DATA_DESC:
+        break;
+    case SRP_DATA_DESC_DIRECT:
+        req->total_desc = req->local_desc = 1;
+        break;
+    case SRP_DATA_DESC_INDIRECT: {
+        struct srp_indirect_buf *ind_tmp = (struct srp_indirect_buf *)
+                (cmd->add_data + req->cdb_offset);
+
+        req->total_desc = be32_to_cpu(ind_tmp->table_desc.len) /
+                          sizeof(struct srp_direct_buf);
+        req->local_desc = req->writing ? cmd->data_out_desc_cnt :
+                          cmd->data_in_desc_cnt;
+        break;
+    }
+    default:
+        fprintf(stderr,
+                "vscsi_preprocess_desc: Unknown format %x\n", req->dma_fmt);
+        return -1;
+    }
+
+    return 0;
+}
+
+/* Callback to indicate that the SCSI layer has completed a transfer.  */
+static void vscsi_transfer_data(SCSIRequest *sreq, uint32_t len)
+{
+    VSCSIState *s = VIO_SPAPR_VSCSI_DEVICE(sreq->bus->qbus.parent);
+    vscsi_req *req = sreq->hba_private;
+    uint8_t *buf;
+    int rc = 0;
+
+    DPRINTF("VSCSI: SCSI xfer complete tag=0x%x len=0x%x, req=%p\n",
+            sreq->tag, len, req);
+    if (req == NULL) {
+        fprintf(stderr, "VSCSI: Can't find request for tag 0x%x\n", sreq->tag);
+        return;
+    }
+
+    if (len) {
+        buf = scsi_req_get_buf(sreq);
+        rc = vscsi_srp_transfer_data(s, req, req->writing, buf, len);
+    }
+    if (rc < 0) {
+        fprintf(stderr, "VSCSI: RDMA error rc=%d!\n", rc);
+        req->dma_error = true;
+        scsi_req_cancel(req->sreq);
+        return;
+    }
+
+    /* Start next chunk */
+    req->data_len -= rc;
+    scsi_req_continue(sreq);
+}
+
+/* Callback to indicate that the SCSI layer has completed a transfer.  */
+static void vscsi_command_complete(SCSIRequest *sreq, uint32_t status, size_t resid)
+{
+    VSCSIState *s = VIO_SPAPR_VSCSI_DEVICE(sreq->bus->qbus.parent);
+    vscsi_req *req = sreq->hba_private;
+    int32_t res_in = 0, res_out = 0;
+
+    DPRINTF("VSCSI: SCSI cmd complete, tag=0x%x status=0x%x, req=%p\n",
+            sreq->tag, status, req);
+    if (req == NULL) {
+        fprintf(stderr, "VSCSI: Can't find request for tag 0x%x\n", sreq->tag);
+        return;
+    }
+
+    if (status == CHECK_CONDITION) {
+        req->senselen = scsi_req_get_sense(req->sreq, req->sense,
+                                           sizeof(req->sense));
+        DPRINTF("VSCSI: Sense data, %d bytes:\n", req->senselen);
+        DPRINTF("       %02x  %02x  %02x  %02x  %02x  %02x  %02x  %02x\n",
+                req->sense[0], req->sense[1], req->sense[2], req->sense[3],
+                req->sense[4], req->sense[5], req->sense[6], req->sense[7]);
+        DPRINTF("       %02x  %02x  %02x  %02x  %02x  %02x  %02x  %02x\n",
+                req->sense[8], req->sense[9], req->sense[10], req->sense[11],
+                req->sense[12], req->sense[13], req->sense[14], req->sense[15]);
+    }
+
+    DPRINTF("VSCSI: Command complete err=%d\n", status);
+    if (status == 0) {
+        /* We handle overflows, not underflows for normal commands,
+         * but hopefully nobody cares
+         */
+        if (req->writing) {
+            res_out = req->data_len;
+        } else {
+            res_in = req->data_len;
+        }
+    }
+    vscsi_send_rsp(s, req, status, res_in, res_out);
+    vscsi_put_req(req);
+}
+
+static void vscsi_request_cancelled(SCSIRequest *sreq)
+{
+    vscsi_req *req = sreq->hba_private;
+
+    if (req->dma_error) {
+        VSCSIState *s = VIO_SPAPR_VSCSI_DEVICE(sreq->bus->qbus.parent);
+
+        vscsi_makeup_sense(s, req, HARDWARE_ERROR, 0, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+    }
+    vscsi_put_req(req);
+}
+
+static const VMStateDescription vmstate_spapr_vscsi_req = {
+    .name = "spapr_vscsi_req",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_BUFFER(crq.raw, vscsi_req),
+        VMSTATE_BUFFER(iu.srp.reserved, vscsi_req),
+        VMSTATE_UINT32(qtag, vscsi_req),
+        VMSTATE_BOOL(active, vscsi_req),
+        VMSTATE_UINT32(data_len, vscsi_req),
+        VMSTATE_BOOL(writing, vscsi_req),
+        VMSTATE_UINT32(senselen, vscsi_req),
+        VMSTATE_BUFFER(sense, vscsi_req),
+        VMSTATE_UINT8(dma_fmt, vscsi_req),
+        VMSTATE_UINT16(local_desc, vscsi_req),
+        VMSTATE_UINT16(total_desc, vscsi_req),
+        VMSTATE_UINT16(cdb_offset, vscsi_req),
+      /*Restart SCSI request from the beginning for now */
+      /*VMSTATE_UINT16(cur_desc_num, vscsi_req),
+        VMSTATE_UINT16(cur_desc_offset, vscsi_req),*/
+        VMSTATE_END_OF_LIST()
+    },
+};
+
+static void vscsi_save_request(QEMUFile *f, SCSIRequest *sreq)
+{
+    vscsi_req *req = sreq->hba_private;
+    assert(req->active);
+
+    vmstate_save_state(f, &vmstate_spapr_vscsi_req, req, NULL);
+
+    DPRINTF("VSCSI: saving tag=%u, current desc#%d, offset=%x\n",
+            req->qtag, req->cur_desc_num, req->cur_desc_offset);
+}
+
+static void *vscsi_load_request(QEMUFile *f, SCSIRequest *sreq)
+{
+    SCSIBus *bus = sreq->bus;
+    VSCSIState *s = VIO_SPAPR_VSCSI_DEVICE(bus->qbus.parent);
+    vscsi_req *req;
+    int rc;
+
+    assert(sreq->tag < VSCSI_REQ_LIMIT);
+    req = &s->reqs[sreq->tag];
+    assert(!req->active);
+
+    memset(req, 0, sizeof(*req));
+    rc = vmstate_load_state(f, &vmstate_spapr_vscsi_req, req, 1);
+    if (rc) {
+        fprintf(stderr, "VSCSI: failed loading request tag#%u\n", sreq->tag);
+        return NULL;
+    }
+    assert(req->active);
+
+    req->sreq = scsi_req_ref(sreq);
+
+    DPRINTF("VSCSI: restoring tag=%u, current desc#%d, offset=%x\n",
+            req->qtag, req->cur_desc_num, req->cur_desc_offset);
+
+    return req;
+}
+
+static void vscsi_process_login(VSCSIState *s, vscsi_req *req)
+{
+    union viosrp_iu *iu = &req->iu;
+    struct srp_login_rsp *rsp = &iu->srp.login_rsp;
+    uint64_t tag = iu->srp.rsp.tag;
+
+    DPRINTF("VSCSI: Got login, sendin response !\n");
+
+    /* TODO handle case that requested size is wrong and
+     * buffer format is wrong
+     */
+    memset(iu, 0, sizeof(struct srp_login_rsp));
+    rsp->opcode = SRP_LOGIN_RSP;
+    /* Don't advertise quite as many request as we support to
+     * keep room for management stuff etc...
+     */
+    rsp->req_lim_delta = cpu_to_be32(VSCSI_REQ_LIMIT-2);
+    rsp->tag = tag;
+    rsp->max_it_iu_len = cpu_to_be32(sizeof(union srp_iu));
+    rsp->max_ti_iu_len = cpu_to_be32(sizeof(union srp_iu));
+    /* direct and indirect */
+    rsp->buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT);
+
+    vscsi_send_iu(s, req, sizeof(*rsp), VIOSRP_SRP_FORMAT);
+}
+
+static void vscsi_inquiry_no_target(VSCSIState *s, vscsi_req *req)
+{
+    uint8_t *cdb = req->iu.srp.cmd.cdb;
+    uint8_t resp_data[36];
+    int rc, len, alen;
+
+    /* We dont do EVPD. Also check that page_code is 0 */
+    if ((cdb[1] & 0x01) || cdb[2] != 0) {
+        /* Send INVALID FIELD IN CDB */
+        vscsi_makeup_sense(s, req, ILLEGAL_REQUEST, 0x24, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+        return;
+    }
+    alen = cdb[3];
+    alen = (alen << 8) | cdb[4];
+    len = MIN(alen, 36);
+
+    /* Fake up inquiry using PQ=3 */
+    memset(resp_data, 0, 36);
+    resp_data[0] = 0x7f;   /* Not capable of supporting a device here */
+    resp_data[2] = 0x06;   /* SPS-4 */
+    resp_data[3] = 0x02;   /* Resp data format */
+    resp_data[4] = 36 - 5; /* Additional length */
+    resp_data[7] = 0x10;   /* Sync transfers */
+    memcpy(&resp_data[16], "QEMU EMPTY      ", 16);
+    memcpy(&resp_data[8], "QEMU    ", 8);
+
+    req->writing = 0;
+    vscsi_preprocess_desc(req);
+    rc = vscsi_srp_transfer_data(s, req, 0, resp_data, len);
+    if (rc < 0) {
+        vscsi_makeup_sense(s, req, HARDWARE_ERROR, 0, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+    } else {
+        vscsi_send_rsp(s, req, 0, 36 - rc, 0);
+    }
+}
+
+static void vscsi_report_luns(VSCSIState *s, vscsi_req *req)
+{
+    BusChild *kid;
+    int i, len, n, rc;
+    uint8_t *resp_data;
+    bool found_lun0;
+
+    n = 0;
+    found_lun0 = false;
+    QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+        SCSIDevice *dev = SCSI_DEVICE(kid->child);
+
+        n += 8;
+        if (dev->channel == 0 && dev->id == 0 && dev->lun == 0) {
+            found_lun0 = true;
+        }
+    }
+    if (!found_lun0) {
+        n += 8;
+    }
+    len = n+8;
+
+    resp_data = g_malloc0(len);
+    memset(resp_data, 0, len);
+    stl_be_p(resp_data, n);
+    i = found_lun0 ? 8 : 16;
+    QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+        DeviceState *qdev = kid->child;
+        SCSIDevice *dev = SCSI_DEVICE(qdev);
+
+        if (dev->id == 0 && dev->channel == 0) {
+            resp_data[i] = 0;         /* Use simple LUN for 0 (SAM5 4.7.7.1) */
+        } else {
+            resp_data[i] = (2 << 6);  /* Otherwise LUN addressing (4.7.7.4)  */
+        }
+        resp_data[i] |= dev->id;
+        resp_data[i+1] = (dev->channel << 5);
+        resp_data[i+1] |= dev->lun;
+        i += 8;
+    }
+
+    vscsi_preprocess_desc(req);
+    rc = vscsi_srp_transfer_data(s, req, 0, resp_data, len);
+    g_free(resp_data);
+    if (rc < 0) {
+        vscsi_makeup_sense(s, req, HARDWARE_ERROR, 0, 0);
+        vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+    } else {
+        vscsi_send_rsp(s, req, 0, len - rc, 0);
+    }
+}
+
+static int vscsi_queue_cmd(VSCSIState *s, vscsi_req *req)
+{
+    union srp_iu *srp = &req->iu.srp;
+    SCSIDevice *sdev;
+    int n, lun;
+
+    if ((srp->cmd.lun == 0 || be64_to_cpu(srp->cmd.lun) == SRP_REPORT_LUNS_WLUN)
+      && srp->cmd.cdb[0] == REPORT_LUNS) {
+        vscsi_report_luns(s, req);
+        return 0;
+    }
+
+    sdev = vscsi_device_find(&s->bus, be64_to_cpu(srp->cmd.lun), &lun);
+    if (!sdev) {
+        DPRINTF("VSCSI: Command for lun %08" PRIx64 " with no drive\n",
+                be64_to_cpu(srp->cmd.lun));
+        if (srp->cmd.cdb[0] == INQUIRY) {
+            vscsi_inquiry_no_target(s, req);
+        } else {
+            vscsi_makeup_sense(s, req, ILLEGAL_REQUEST, 0x24, 0x00);
+            vscsi_send_rsp(s, req, CHECK_CONDITION, 0, 0);
+        } return 1;
+    }
+
+    req->sreq = scsi_req_new(sdev, req->qtag, lun, srp->cmd.cdb, req);
+    n = scsi_req_enqueue(req->sreq);
+
+    DPRINTF("VSCSI: Queued command tag 0x%x CMD 0x%x=%s LUN %d ret: %d\n",
+            req->qtag, srp->cmd.cdb[0], scsi_command_name(srp->cmd.cdb[0]),
+            lun, n);
+
+    if (n) {
+        /* Transfer direction must be set before preprocessing the
+         * descriptors
+         */
+        req->writing = (n < 1);
+
+        /* Preprocess RDMA descriptors */
+        vscsi_preprocess_desc(req);
+
+        /* Get transfer direction and initiate transfer */
+        if (n > 0) {
+            req->data_len = n;
+        } else if (n < 0) {
+            req->data_len = -n;
+        }
+        scsi_req_continue(req->sreq);
+    }
+    /* Don't touch req here, it may have been recycled already */
+
+    return 0;
+}
+
+static int vscsi_process_tsk_mgmt(VSCSIState *s, vscsi_req *req)
+{
+    union viosrp_iu *iu = &req->iu;
+    vscsi_req *tmpreq;
+    int i, lun = 0, resp = SRP_TSK_MGMT_COMPLETE;
+    SCSIDevice *d;
+    uint64_t tag = iu->srp.rsp.tag;
+    uint8_t sol_not = iu->srp.cmd.sol_not;
+
+    fprintf(stderr, "vscsi_process_tsk_mgmt %02x\n",
+            iu->srp.tsk_mgmt.tsk_mgmt_func);
+
+    d = vscsi_device_find(&s->bus, be64_to_cpu(req->iu.srp.tsk_mgmt.lun), &lun);
+    if (!d) {
+        resp = SRP_TSK_MGMT_FIELDS_INVALID;
+    } else {
+        switch (iu->srp.tsk_mgmt.tsk_mgmt_func) {
+        case SRP_TSK_ABORT_TASK:
+            if (d->lun != lun) {
+                resp = SRP_TSK_MGMT_FIELDS_INVALID;
+                break;
+            }
+
+            tmpreq = vscsi_find_req(s, req->iu.srp.tsk_mgmt.task_tag);
+            if (tmpreq && tmpreq->sreq) {
+                assert(tmpreq->sreq->hba_private);
+                scsi_req_cancel(tmpreq->sreq);
+            }
+            break;
+
+        case SRP_TSK_LUN_RESET:
+            if (d->lun != lun) {
+                resp = SRP_TSK_MGMT_FIELDS_INVALID;
+                break;
+            }
+
+            qdev_reset_all(&d->qdev);
+            break;
+
+        case SRP_TSK_ABORT_TASK_SET:
+        case SRP_TSK_CLEAR_TASK_SET:
+            if (d->lun != lun) {
+                resp = SRP_TSK_MGMT_FIELDS_INVALID;
+                break;
+            }
+
+            for (i = 0; i < VSCSI_REQ_LIMIT; i++) {
+                tmpreq = &s->reqs[i];
+                if (tmpreq->iu.srp.cmd.lun != req->iu.srp.tsk_mgmt.lun) {
+                    continue;
+                }
+                if (!tmpreq->active || !tmpreq->sreq) {
+                    continue;
+                }
+                assert(tmpreq->sreq->hba_private);
+                scsi_req_cancel(tmpreq->sreq);
+            }
+            break;
+
+        case SRP_TSK_CLEAR_ACA:
+            resp = SRP_TSK_MGMT_NOT_SUPPORTED;
+            break;
+
+        default:
+            resp = SRP_TSK_MGMT_FIELDS_INVALID;
+            break;
+        }
+    }
+
+    /* Compose the response here as  */
+    memset(iu, 0, sizeof(struct srp_rsp) + 4);
+    iu->srp.rsp.opcode = SRP_RSP;
+    iu->srp.rsp.req_lim_delta = cpu_to_be32(1);
+    iu->srp.rsp.tag = tag;
+    iu->srp.rsp.flags |= SRP_RSP_FLAG_RSPVALID;
+    iu->srp.rsp.resp_data_len = cpu_to_be32(4);
+    if (resp) {
+        iu->srp.rsp.sol_not = (sol_not & 0x04) >> 2;
+    } else {
+        iu->srp.rsp.sol_not = (sol_not & 0x02) >> 1;
+    }
+
+    iu->srp.rsp.status = GOOD;
+    iu->srp.rsp.data[3] = resp;
+
+    vscsi_send_iu(s, req, sizeof(iu->srp.rsp) + 4, VIOSRP_SRP_FORMAT);
+
+    return 1;
+}
+
+static int vscsi_handle_srp_req(VSCSIState *s, vscsi_req *req)
+{
+    union srp_iu *srp = &req->iu.srp;
+    int done = 1;
+    uint8_t opcode = srp->rsp.opcode;
+
+    switch (opcode) {
+    case SRP_LOGIN_REQ:
+        vscsi_process_login(s, req);
+        break;
+    case SRP_TSK_MGMT:
+        done = vscsi_process_tsk_mgmt(s, req);
+        break;
+    case SRP_CMD:
+        done = vscsi_queue_cmd(s, req);
+        break;
+    case SRP_LOGIN_RSP:
+    case SRP_I_LOGOUT:
+    case SRP_T_LOGOUT:
+    case SRP_RSP:
+    case SRP_CRED_REQ:
+    case SRP_CRED_RSP:
+    case SRP_AER_REQ:
+    case SRP_AER_RSP:
+        fprintf(stderr, "VSCSI: Unsupported opcode %02x\n", opcode);
+        break;
+    default:
+        fprintf(stderr, "VSCSI: Unknown type %02x\n", opcode);
+    }
+
+    return done;
+}
+
+static int vscsi_send_adapter_info(VSCSIState *s, vscsi_req *req)
+{
+    struct viosrp_adapter_info *sinfo;
+    struct mad_adapter_info_data info;
+    int rc;
+
+    sinfo = &req->iu.mad.adapter_info;
+
+#if 0 /* What for ? */
+    rc = spapr_vio_dma_read(&s->vdev, be64_to_cpu(sinfo->buffer),
+                            &info, be16_to_cpu(sinfo->common.length));
+    if (rc) {
+        fprintf(stderr, "vscsi_send_adapter_info: DMA read failure !\n");
+    }
+#endif
+    memset(&info, 0, sizeof(info));
+    strcpy(info.srp_version, SRP_VERSION);
+    memcpy(info.partition_name, "qemu", sizeof("qemu"));
+    info.partition_number = cpu_to_be32(0);
+    info.mad_version = cpu_to_be32(1);
+    info.os_type = cpu_to_be32(2);
+    info.port_max_txu[0] = cpu_to_be32(VSCSI_MAX_SECTORS << 9);
+
+    rc = spapr_vio_dma_write(&s->vdev, be64_to_cpu(sinfo->buffer),
+                             &info, be16_to_cpu(sinfo->common.length));
+    if (rc)  {
+        fprintf(stderr, "vscsi_send_adapter_info: DMA write failure !\n");
+    }
+
+    sinfo->common.status = rc ? cpu_to_be32(1) : 0;
+
+    return vscsi_send_iu(s, req, sizeof(*sinfo), VIOSRP_MAD_FORMAT);
+}
+
+static int vscsi_send_capabilities(VSCSIState *s, vscsi_req *req)
+{
+    struct viosrp_capabilities *vcap;
+    struct capabilities cap = { };
+    uint16_t len, req_len;
+    uint64_t buffer;
+    int rc;
+
+    vcap = &req->iu.mad.capabilities;
+    req_len = len = be16_to_cpu(vcap->common.length);
+    buffer = be64_to_cpu(vcap->buffer);
+    if (len > sizeof(cap)) {
+        fprintf(stderr, "vscsi_send_capabilities: capabilities size mismatch !\n");
+
+        /*
+         * Just read and populate the structure that is known.
+         * Zero rest of the structure.
+         */
+        len = sizeof(cap);
+    }
+    rc = spapr_vio_dma_read(&s->vdev, buffer, &cap, len);
+    if (rc)  {
+        fprintf(stderr, "vscsi_send_capabilities: DMA read failure !\n");
+    }
+
+    /*
+     * Current implementation does not suppport any migration or
+     * reservation capabilities. Construct the response telling the
+     * guest not to use them.
+     */
+    cap.flags = 0;
+    cap.migration.ecl = 0;
+    cap.reserve.type = 0;
+    cap.migration.common.server_support = 0;
+    cap.reserve.common.server_support = 0;
+
+    rc = spapr_vio_dma_write(&s->vdev, buffer, &cap, len);
+    if (rc)  {
+        fprintf(stderr, "vscsi_send_capabilities: DMA write failure !\n");
+    }
+    if (req_len > len) {
+        /*
+         * Being paranoid and lets not worry about the error code
+         * here. Actual write of the cap is done above.
+         */
+        spapr_vio_dma_set(&s->vdev, (buffer + len), 0, (req_len - len));
+    }
+    vcap->common.status = rc ? cpu_to_be32(1) : 0;
+    return vscsi_send_iu(s, req, sizeof(*vcap), VIOSRP_MAD_FORMAT);
+}
+
+static int vscsi_handle_mad_req(VSCSIState *s, vscsi_req *req)
+{
+    union mad_iu *mad = &req->iu.mad;
+    bool request_handled = false;
+    uint64_t retlen = 0;
+
+    switch (be32_to_cpu(mad->empty_iu.common.type)) {
+    case VIOSRP_EMPTY_IU_TYPE:
+        fprintf(stderr, "Unsupported EMPTY MAD IU\n");
+        retlen = sizeof(mad->empty_iu);
+        break;
+    case VIOSRP_ERROR_LOG_TYPE:
+        fprintf(stderr, "Unsupported ERROR LOG MAD IU\n");
+        retlen = sizeof(mad->error_log);
+        break;
+    case VIOSRP_ADAPTER_INFO_TYPE:
+        vscsi_send_adapter_info(s, req);
+        request_handled = true;
+        break;
+    case VIOSRP_HOST_CONFIG_TYPE:
+        retlen = sizeof(mad->host_config);
+        break;
+    case VIOSRP_CAPABILITIES_TYPE:
+        vscsi_send_capabilities(s, req);
+        request_handled = true;
+        break;
+    default:
+        fprintf(stderr, "VSCSI: Unknown MAD type %02x\n",
+                be32_to_cpu(mad->empty_iu.common.type));
+        /*
+         * PAPR+ says that "The length field is set to the length
+         * of the data structure(s) used in the command".
+         * As we did not recognize the request type, put zero there.
+         */
+        retlen = 0;
+    }
+
+    if (!request_handled) {
+        mad->empty_iu.common.status = cpu_to_be16(VIOSRP_MAD_NOT_SUPPORTED);
+        vscsi_send_iu(s, req, retlen, VIOSRP_MAD_FORMAT);
+    }
+
+    return 1;
+}
+
+static void vscsi_got_payload(VSCSIState *s, vscsi_crq *crq)
+{
+    vscsi_req *req;
+    int done;
+
+    req = vscsi_get_req(s);
+    if (req == NULL) {
+        fprintf(stderr, "VSCSI: Failed to get a request !\n");
+        return;
+    }
+
+    /* We only support a limited number of descriptors, we know
+     * the ibmvscsi driver uses up to 10 max, so it should fit
+     * in our 256 bytes IUs. If not we'll have to increase the size
+     * of the structure.
+     */
+    if (crq->s.IU_length > sizeof(union viosrp_iu)) {
+        fprintf(stderr, "VSCSI: SRP IU too long (%d bytes) !\n",
+                crq->s.IU_length);
+        vscsi_put_req(req);
+        return;
+    }
+
+    /* XXX Handle failure differently ? */
+    if (spapr_vio_dma_read(&s->vdev, crq->s.IU_data_ptr, &req->iu,
+                           crq->s.IU_length)) {
+        fprintf(stderr, "vscsi_got_payload: DMA read failure !\n");
+        vscsi_put_req(req);
+        return;
+    }
+    memcpy(&req->crq, crq, sizeof(vscsi_crq));
+
+    if (crq->s.format == VIOSRP_MAD_FORMAT) {
+        done = vscsi_handle_mad_req(s, req);
+    } else {
+        done = vscsi_handle_srp_req(s, req);
+    }
+
+    if (done) {
+        vscsi_put_req(req);
+    }
+}
+
+
+static int vscsi_do_crq(struct VIOsPAPRDevice *dev, uint8_t *crq_data)
+{
+    VSCSIState *s = VIO_SPAPR_VSCSI_DEVICE(dev);
+    vscsi_crq crq;
+
+    memcpy(crq.raw, crq_data, 16);
+    crq.s.timeout = be16_to_cpu(crq.s.timeout);
+    crq.s.IU_length = be16_to_cpu(crq.s.IU_length);
+    crq.s.IU_data_ptr = be64_to_cpu(crq.s.IU_data_ptr);
+
+    DPRINTF("VSCSI: do_crq %02x %02x ...\n", crq.raw[0], crq.raw[1]);
+
+    switch (crq.s.valid) {
+    case 0xc0: /* Init command/response */
+
+        /* Respond to initialization request */
+        if (crq.s.format == 0x01) {
+            memset(crq.raw, 0, 16);
+            crq.s.valid = 0xc0;
+            crq.s.format = 0x02;
+            spapr_vio_send_crq(dev, crq.raw);
+        }
+
+        /* Note that in hotplug cases, we might get a 0x02
+         * as a result of us emitting the init request
+         */
+
+        break;
+    case 0xff: /* Link event */
+
+        /* Not handled for now */
+
+        break;
+    case 0x80: /* Payloads */
+        switch (crq.s.format) {
+        case VIOSRP_SRP_FORMAT: /* AKA VSCSI request */
+        case VIOSRP_MAD_FORMAT: /* AKA VSCSI response */
+            vscsi_got_payload(s, &crq);
+            break;
+        case VIOSRP_OS400_FORMAT:
+        case VIOSRP_AIX_FORMAT:
+        case VIOSRP_LINUX_FORMAT:
+        case VIOSRP_INLINE_FORMAT:
+            fprintf(stderr, "vscsi_do_srq: Unsupported payload format %02x\n",
+                    crq.s.format);
+            break;
+        default:
+            fprintf(stderr, "vscsi_do_srq: Unknown payload format %02x\n",
+                    crq.s.format);
+        }
+        break;
+    default:
+        fprintf(stderr, "vscsi_do_crq: unknown CRQ %02x %02x ...\n",
+                crq.raw[0], crq.raw[1]);
+    };
+
+    return 0;
+}
+
+static const struct SCSIBusInfo vscsi_scsi_info = {
+    .tcq = true,
+    .max_channel = 7, /* logical unit addressing format */
+    .max_target = 63,
+    .max_lun = 31,
+
+    .transfer_data = vscsi_transfer_data,
+    .complete = vscsi_command_complete,
+    .cancel = vscsi_request_cancelled,
+    .save_request = vscsi_save_request,
+    .load_request = vscsi_load_request,
+};
+
+static void spapr_vscsi_reset(VIOsPAPRDevice *dev)
+{
+    VSCSIState *s = VIO_SPAPR_VSCSI_DEVICE(dev);
+    int i;
+
+    memset(s->reqs, 0, sizeof(s->reqs));
+    for (i = 0; i < VSCSI_REQ_LIMIT; i++) {
+        s->reqs[i].qtag = i;
+    }
+}
+
+static void spapr_vscsi_realize(VIOsPAPRDevice *dev, Error **errp)
+{
+    VSCSIState *s = VIO_SPAPR_VSCSI_DEVICE(dev);
+
+    dev->crq.SendFunc = vscsi_do_crq;
+
+    scsi_bus_new(&s->bus, sizeof(s->bus), DEVICE(dev),
+                 &vscsi_scsi_info, NULL);
+    if (!dev->qdev.hotplugged) {
+        scsi_bus_legacy_handle_cmdline(&s->bus, errp);
+    }
+}
+
+void spapr_vscsi_create(VIOsPAPRBus *bus)
+{
+    DeviceState *dev;
+
+    dev = qdev_create(&bus->bus, "spapr-vscsi");
+
+    qdev_init_nofail(dev);
+}
+
+static int spapr_vscsi_devnode(VIOsPAPRDevice *dev, void *fdt, int node_off)
+{
+    int ret;
+
+    ret = fdt_setprop_cell(fdt, node_off, "#address-cells", 2);
+    if (ret < 0) {
+        return ret;
+    }
+
+    ret = fdt_setprop_cell(fdt, node_off, "#size-cells", 0);
+    if (ret < 0) {
+        return ret;
+    }
+
+    return 0;
+}
+
+static Property spapr_vscsi_properties[] = {
+    DEFINE_SPAPR_PROPERTIES(VSCSIState, vdev),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static const VMStateDescription vmstate_spapr_vscsi = {
+    .name = "spapr_vscsi",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_SPAPR_VIO(vdev, VSCSIState),
+        /* VSCSI state */
+        /* ???? */
+
+        VMSTATE_END_OF_LIST()
+    },
+};
+
+static void spapr_vscsi_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    VIOsPAPRDeviceClass *k = VIO_SPAPR_DEVICE_CLASS(klass);
+
+    k->realize = spapr_vscsi_realize;
+    k->reset = spapr_vscsi_reset;
+    k->devnode = spapr_vscsi_devnode;
+    k->dt_name = "v-scsi";
+    k->dt_type = "vscsi";
+    k->dt_compatible = "IBM,v-scsi";
+    k->signal_mask = 0x00000001;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+    dc->props = spapr_vscsi_properties;
+    k->rtce_window_size = 0x10000000;
+    dc->vmsd = &vmstate_spapr_vscsi;
+}
+
+static const TypeInfo spapr_vscsi_info = {
+    .name          = TYPE_VIO_SPAPR_VSCSI_DEVICE,
+    .parent        = TYPE_VIO_SPAPR_DEVICE,
+    .instance_size = sizeof(VSCSIState),
+    .class_init    = spapr_vscsi_class_init,
+};
+
+static void spapr_vscsi_register_types(void)
+{
+    type_register_static(&spapr_vscsi_info);
+}
+
+type_init(spapr_vscsi_register_types)
diff --git a/qemu/hw/scsi/srp.h b/qemu/hw/scsi/srp.h
new file mode 100644
index 000000000..d27f31d2d
--- /dev/null
+++ b/qemu/hw/scsi/srp.h
@@ -0,0 +1,247 @@
+/*
+ * Copyright (c) 2005 Cisco Systems.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef SCSI_SRP_H
+#define SCSI_SRP_H
+
+/*
+ * Structures and constants for the SCSI RDMA Protocol (SRP) as
+ * defined by the INCITS T10 committee.  This file was written using
+ * draft Revision 16a of the SRP standard.
+ */
+
+enum {
+
+    SRP_LOGIN_REQ = 0x00,
+    SRP_TSK_MGMT  = 0x01,
+    SRP_CMD       = 0x02,
+    SRP_I_LOGOUT  = 0x03,
+    SRP_LOGIN_RSP = 0xc0,
+    SRP_RSP       = 0xc1,
+    SRP_LOGIN_REJ = 0xc2,
+    SRP_T_LOGOUT  = 0x80,
+    SRP_CRED_REQ  = 0x81,
+    SRP_AER_REQ   = 0x82,
+    SRP_CRED_RSP  = 0x41,
+    SRP_AER_RSP   = 0x42
+};
+
+enum {
+    SRP_BUF_FORMAT_DIRECT   = 1 << 1,
+    SRP_BUF_FORMAT_INDIRECT = 1 << 2
+};
+
+enum {
+    SRP_NO_DATA_DESC       = 0,
+    SRP_DATA_DESC_DIRECT   = 1,
+    SRP_DATA_DESC_INDIRECT = 2
+};
+
+enum {
+    SRP_TSK_ABORT_TASK     = 0x01,
+    SRP_TSK_ABORT_TASK_SET = 0x02,
+    SRP_TSK_CLEAR_TASK_SET = 0x04,
+    SRP_TSK_LUN_RESET      = 0x08,
+    SRP_TSK_CLEAR_ACA      = 0x40
+};
+
+enum srp_login_rej_reason {
+    SRP_LOGIN_REJ_UNABLE_ESTABLISH_CHANNEL   = 0x00010000,
+    SRP_LOGIN_REJ_INSUFFICIENT_RESOURCES     = 0x00010001,
+    SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE = 0x00010002,
+    SRP_LOGIN_REJ_UNABLE_ASSOCIATE_CHANNEL   = 0x00010003,
+    SRP_LOGIN_REJ_UNSUPPORTED_DESCRIPTOR_FMT = 0x00010004,
+    SRP_LOGIN_REJ_MULTI_CHANNEL_UNSUPPORTED  = 0x00010005,
+    SRP_LOGIN_REJ_CHANNEL_LIMIT_REACHED      = 0x00010006
+};
+
+enum {
+    SRP_REV10_IB_IO_CLASS  = 0xff00,
+    SRP_REV16A_IB_IO_CLASS = 0x0100
+};
+
+enum {
+    SRP_TSK_MGMT_COMPLETE       = 0x00,
+    SRP_TSK_MGMT_FIELDS_INVALID = 0x02,
+    SRP_TSK_MGMT_NOT_SUPPORTED  = 0x04,
+    SRP_TSK_MGMT_FAILED         = 0x05
+};
+
+struct srp_direct_buf {
+    uint64_t    va;
+    uint32_t    key;
+    uint32_t    len;
+};
+
+/*
+ * We need the packed attribute because the SRP spec puts the list of
+ * descriptors at an offset of 20, which is not aligned to the size of
+ * struct srp_direct_buf.  The whole structure must be packed to avoid
+ * having the 20-byte structure padded to 24 bytes on 64-bit architectures.
+ */
+struct srp_indirect_buf {
+    struct srp_direct_buf    table_desc;
+    uint32_t                 len;
+    struct srp_direct_buf    desc_list[0];
+} QEMU_PACKED;
+
+enum {
+    SRP_MULTICHAN_SINGLE = 0,
+    SRP_MULTICHAN_MULTI  = 1
+};
+
+struct srp_login_req {
+    uint8_t    opcode;
+    uint8_t    reserved1[7];
+    uint64_t   tag;
+    uint32_t   req_it_iu_len;
+    uint8_t    reserved2[4];
+    uint16_t   req_buf_fmt;
+    uint8_t    req_flags;
+    uint8_t    reserved3[5];
+    uint8_t    initiator_port_id[16];
+    uint8_t    target_port_id[16];
+};
+
+/*
+ * The SRP spec defines the size of the LOGIN_RSP structure to be 52
+ * bytes, so it needs to be packed to avoid having it padded to 56
+ * bytes on 64-bit architectures.
+ */
+struct srp_login_rsp {
+    uint8_t    opcode;
+    uint8_t    reserved1[3];
+    uint32_t   req_lim_delta;
+    uint64_t   tag;
+    uint32_t   max_it_iu_len;
+    uint32_t   max_ti_iu_len;
+    uint16_t   buf_fmt;
+    uint8_t    rsp_flags;
+    uint8_t    reserved2[25];
+} QEMU_PACKED;
+
+struct srp_login_rej {
+    uint8_t    opcode;
+    uint8_t    reserved1[3];
+    uint32_t   reason;
+    uint64_t   tag;
+    uint8_t    reserved2[8];
+    uint16_t   buf_fmt;
+    uint8_t    reserved3[6];
+};
+
+struct srp_i_logout {
+    uint8_t    opcode;
+    uint8_t    reserved[7];
+    uint64_t   tag;
+};
+
+struct srp_t_logout {
+    uint8_t    opcode;
+    uint8_t    sol_not;
+    uint8_t    reserved[2];
+    uint32_t   reason;
+    uint64_t   tag;
+};
+
+/*
+ * We need the packed attribute because the SRP spec only aligns the
+ * 8-byte LUN field to 4 bytes.
+ */
+struct srp_tsk_mgmt {
+    uint8_t    opcode;
+    uint8_t    sol_not;
+    uint8_t    reserved1[6];
+    uint64_t   tag;
+    uint8_t    reserved2[4];
+    uint64_t   lun;
+    uint8_t    reserved3[2];
+    uint8_t    tsk_mgmt_func;
+    uint8_t    reserved4;
+    uint64_t   task_tag;
+    uint8_t    reserved5[8];
+} QEMU_PACKED;
+
+/*
+ * We need the packed attribute because the SRP spec only aligns the
+ * 8-byte LUN field to 4 bytes.
+ */
+struct srp_cmd {
+    uint8_t    opcode;
+    uint8_t    sol_not;
+    uint8_t    reserved1[3];
+    uint8_t    buf_fmt;
+    uint8_t    data_out_desc_cnt;
+    uint8_t    data_in_desc_cnt;
+    uint64_t   tag;
+    uint8_t    reserved2[4];
+    uint64_t   lun;
+    uint8_t    reserved3;
+    uint8_t    task_attr;
+    uint8_t    reserved4;
+    uint8_t    add_cdb_len;
+    uint8_t    cdb[16];
+    uint8_t    add_data[0];
+} QEMU_PACKED;
+
+enum {
+    SRP_RSP_FLAG_RSPVALID = 1 << 0,
+    SRP_RSP_FLAG_SNSVALID = 1 << 1,
+    SRP_RSP_FLAG_DOOVER   = 1 << 2,
+    SRP_RSP_FLAG_DOUNDER  = 1 << 3,
+    SRP_RSP_FLAG_DIOVER   = 1 << 4,
+    SRP_RSP_FLAG_DIUNDER  = 1 << 5
+};
+
+/*
+ * The SRP spec defines the size of the RSP structure to be 36 bytes,
+ * so it needs to be packed to avoid having it padded to 40 bytes on
+ * 64-bit architectures.
+ */
+struct srp_rsp {
+    uint8_t    opcode;
+    uint8_t    sol_not;
+    uint8_t    reserved1[2];
+    uint32_t   req_lim_delta;
+    uint64_t   tag;
+    uint8_t    reserved2[2];
+    uint8_t    flags;
+    uint8_t    status;
+    uint32_t   data_out_res_cnt;
+    uint32_t   data_in_res_cnt;
+    uint32_t   sense_data_len;
+    uint32_t   resp_data_len;
+    uint8_t    data[0];
+} QEMU_PACKED;
+
+#endif /* SCSI_SRP_H */
diff --git a/qemu/hw/scsi/vhost-scsi.c b/qemu/hw/scsi/vhost-scsi.c
new file mode 100644
index 000000000..a69918bef
--- /dev/null
+++ b/qemu/hw/scsi/vhost-scsi.c
@@ -0,0 +1,351 @@
+/*
+ * vhost_scsi host device
+ *
+ * Copyright IBM, Corp. 2011
+ *
+ * Authors:
+ *  Stefan Hajnoczi   <stefanha@linux.vnet.ibm.com>
+ *
+ * Changes for QEMU mainline + tcm_vhost kernel upstream:
+ *  Nicholas Bellinger <nab@risingtidesystems.com>
+ *
+ * This work is licensed under the terms of the GNU LGPL, version 2 or later.
+ * See the COPYING.LIB file in the top-level directory.
+ *
+ */
+
+#include <sys/ioctl.h>
+#include "config.h"
+#include "qemu/error-report.h"
+#include "qemu/queue.h"
+#include "monitor/monitor.h"
+#include "migration/migration.h"
+#include "hw/virtio/vhost-scsi.h"
+#include "hw/virtio/vhost.h"
+#include "hw/virtio/virtio-scsi.h"
+#include "hw/virtio/virtio-bus.h"
+#include "hw/virtio/virtio-access.h"
+#include "hw/fw-path-provider.h"
+
+/* Features supported by host kernel. */
+static const int kernel_feature_bits[] = {
+    VIRTIO_F_NOTIFY_ON_EMPTY,
+    VIRTIO_RING_F_INDIRECT_DESC,
+    VIRTIO_RING_F_EVENT_IDX,
+    VIRTIO_SCSI_F_HOTPLUG,
+    VHOST_INVALID_FEATURE_BIT
+};
+
+static int vhost_scsi_set_endpoint(VHostSCSI *s)
+{
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+    const VhostOps *vhost_ops = s->dev.vhost_ops;
+    struct vhost_scsi_target backend;
+    int ret;
+
+    memset(&backend, 0, sizeof(backend));
+    pstrcpy(backend.vhost_wwpn, sizeof(backend.vhost_wwpn), vs->conf.wwpn);
+    ret = vhost_ops->vhost_call(&s->dev, VHOST_SCSI_SET_ENDPOINT, &backend);
+    if (ret < 0) {
+        return -errno;
+    }
+    return 0;
+}
+
+static void vhost_scsi_clear_endpoint(VHostSCSI *s)
+{
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+    struct vhost_scsi_target backend;
+    const VhostOps *vhost_ops = s->dev.vhost_ops;
+
+    memset(&backend, 0, sizeof(backend));
+    pstrcpy(backend.vhost_wwpn, sizeof(backend.vhost_wwpn), vs->conf.wwpn);
+    vhost_ops->vhost_call(&s->dev, VHOST_SCSI_CLEAR_ENDPOINT, &backend);
+}
+
+static int vhost_scsi_start(VHostSCSI *s)
+{
+    int ret, abi_version, i;
+    VirtIODevice *vdev = VIRTIO_DEVICE(s);
+    BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
+    VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
+    const VhostOps *vhost_ops = s->dev.vhost_ops;
+
+    if (!k->set_guest_notifiers) {
+        error_report("binding does not support guest notifiers");
+        return -ENOSYS;
+    }
+
+    ret = vhost_ops->vhost_call(&s->dev,
+                                VHOST_SCSI_GET_ABI_VERSION, &abi_version);
+    if (ret < 0) {
+        return -errno;
+    }
+    if (abi_version > VHOST_SCSI_ABI_VERSION) {
+        error_report("vhost-scsi: The running tcm_vhost kernel abi_version:"
+                     " %d is greater than vhost_scsi userspace supports: %d, please"
+                     " upgrade your version of QEMU", abi_version,
+                     VHOST_SCSI_ABI_VERSION);
+        return -ENOSYS;
+    }
+
+    ret = vhost_dev_enable_notifiers(&s->dev, vdev);
+    if (ret < 0) {
+        return ret;
+    }
+
+    s->dev.acked_features = vdev->guest_features;
+    ret = vhost_dev_start(&s->dev, vdev);
+    if (ret < 0) {
+        error_report("Error start vhost dev");
+        goto err_notifiers;
+    }
+
+    ret = vhost_scsi_set_endpoint(s);
+    if (ret < 0) {
+        error_report("Error set vhost-scsi endpoint");
+        goto err_vhost_stop;
+    }
+
+    ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, true);
+    if (ret < 0) {
+        error_report("Error binding guest notifier");
+        goto err_endpoint;
+    }
+
+    /* guest_notifier_mask/pending not used yet, so just unmask
+     * everything here.  virtio-pci will do the right thing by
+     * enabling/disabling irqfd.
+     */
+    for (i = 0; i < s->dev.nvqs; i++) {
+        vhost_virtqueue_mask(&s->dev, vdev, i, false);
+    }
+
+    return ret;
+
+err_endpoint:
+    vhost_scsi_clear_endpoint(s);
+err_vhost_stop:
+    vhost_dev_stop(&s->dev, vdev);
+err_notifiers:
+    vhost_dev_disable_notifiers(&s->dev, vdev);
+    return ret;
+}
+
+static void vhost_scsi_stop(VHostSCSI *s)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(s);
+    BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
+    VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
+    int ret = 0;
+
+    if (k->set_guest_notifiers) {
+        ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false);
+        if (ret < 0) {
+                error_report("vhost guest notifier cleanup failed: %d", ret);
+        }
+    }
+    assert(ret >= 0);
+
+    vhost_scsi_clear_endpoint(s);
+    vhost_dev_stop(&s->dev, vdev);
+    vhost_dev_disable_notifiers(&s->dev, vdev);
+}
+
+static uint64_t vhost_scsi_get_features(VirtIODevice *vdev,
+                                        uint64_t features,
+                                        Error **errp)
+{
+    VHostSCSI *s = VHOST_SCSI(vdev);
+
+    return vhost_get_features(&s->dev, kernel_feature_bits, features);
+}
+
+static void vhost_scsi_set_config(VirtIODevice *vdev,
+                                  const uint8_t *config)
+{
+    VirtIOSCSIConfig *scsiconf = (VirtIOSCSIConfig *)config;
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(vdev);
+
+    if ((uint32_t) virtio_ldl_p(vdev, &scsiconf->sense_size) != vs->sense_size ||
+        (uint32_t) virtio_ldl_p(vdev, &scsiconf->cdb_size) != vs->cdb_size) {
+        error_report("vhost-scsi does not support changing the sense data and CDB sizes");
+        exit(1);
+    }
+}
+
+static void vhost_scsi_set_status(VirtIODevice *vdev, uint8_t val)
+{
+    VHostSCSI *s = (VHostSCSI *)vdev;
+    bool start = (val & VIRTIO_CONFIG_S_DRIVER_OK);
+
+    if (s->dev.started == start) {
+        return;
+    }
+
+    if (start) {
+        int ret;
+
+        ret = vhost_scsi_start(s);
+        if (ret < 0) {
+            error_report("virtio-scsi: unable to start vhost: %s",
+                         strerror(-ret));
+
+            /* There is no userspace virtio-scsi fallback so exit */
+            exit(1);
+        }
+    } else {
+        vhost_scsi_stop(s);
+    }
+}
+
+static void vhost_dummy_handle_output(VirtIODevice *vdev, VirtQueue *vq)
+{
+}
+
+static void vhost_scsi_realize(DeviceState *dev, Error **errp)
+{
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(dev);
+    VHostSCSI *s = VHOST_SCSI(dev);
+    Error *err = NULL;
+    int vhostfd = -1;
+    int ret;
+
+    if (!vs->conf.wwpn) {
+        error_setg(errp, "vhost-scsi: missing wwpn");
+        return;
+    }
+
+    if (vs->conf.vhostfd) {
+        vhostfd = monitor_fd_param(cur_mon, vs->conf.vhostfd, &err);
+        if (vhostfd == -1) {
+            error_setg(errp, "vhost-scsi: unable to parse vhostfd: %s",
+                       error_get_pretty(err));
+            error_free(err);
+            return;
+        }
+    } else {
+        vhostfd = open("/dev/vhost-scsi", O_RDWR);
+        if (vhostfd < 0) {
+            error_setg(errp, "vhost-scsi: open vhost char device failed: %s",
+                       strerror(errno));
+            return;
+        }
+    }
+
+    virtio_scsi_common_realize(dev, &err, vhost_dummy_handle_output,
+                               vhost_dummy_handle_output,
+                               vhost_dummy_handle_output);
+    if (err != NULL) {
+        error_propagate(errp, err);
+        close(vhostfd);
+        return;
+    }
+
+    s->dev.nvqs = VHOST_SCSI_VQ_NUM_FIXED + vs->conf.num_queues;
+    s->dev.vqs = g_new(struct vhost_virtqueue, s->dev.nvqs);
+    s->dev.vq_index = 0;
+    s->dev.backend_features = 0;
+
+    ret = vhost_dev_init(&s->dev, (void *)(uintptr_t)vhostfd,
+                         VHOST_BACKEND_TYPE_KERNEL);
+    if (ret < 0) {
+        error_setg(errp, "vhost-scsi: vhost initialization failed: %s",
+                   strerror(-ret));
+        return;
+    }
+
+    /* At present, channel and lun both are 0 for bootable vhost-scsi disk */
+    s->channel = 0;
+    s->lun = 0;
+    /* Note: we can also get the minimum tpgt from kernel */
+    s->target = vs->conf.boot_tpgt;
+
+    error_setg(&s->migration_blocker,
+            "vhost-scsi does not support migration");
+    migrate_add_blocker(s->migration_blocker);
+}
+
+static void vhost_scsi_unrealize(DeviceState *dev, Error **errp)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
+    VHostSCSI *s = VHOST_SCSI(dev);
+
+    migrate_del_blocker(s->migration_blocker);
+    error_free(s->migration_blocker);
+
+    /* This will stop vhost backend. */
+    vhost_scsi_set_status(vdev, 0);
+
+    g_free(s->dev.vqs);
+
+    virtio_scsi_common_unrealize(dev, errp);
+}
+
+/*
+ * Implementation of an interface to adjust firmware path
+ * for the bootindex property handling.
+ */
+static char *vhost_scsi_get_fw_dev_path(FWPathProvider *p, BusState *bus,
+                                        DeviceState *dev)
+{
+    VHostSCSI *s = VHOST_SCSI(dev);
+    /* format: channel@channel/vhost-scsi@target,lun */
+    return g_strdup_printf("channel@%x/%s@%x,%x", s->channel,
+                           qdev_fw_name(dev), s->target, s->lun);
+}
+
+static Property vhost_scsi_properties[] = {
+    DEFINE_PROP_STRING("vhostfd", VHostSCSI, parent_obj.conf.vhostfd),
+    DEFINE_PROP_STRING("wwpn", VHostSCSI, parent_obj.conf.wwpn),
+    DEFINE_PROP_UINT32("boot_tpgt", VHostSCSI, parent_obj.conf.boot_tpgt, 0),
+    DEFINE_PROP_UINT32("num_queues", VHostSCSI, parent_obj.conf.num_queues, 1),
+    DEFINE_PROP_UINT32("max_sectors", VHostSCSI, parent_obj.conf.max_sectors,
+                                                 0xFFFF),
+    DEFINE_PROP_UINT32("cmd_per_lun", VHostSCSI, parent_obj.conf.cmd_per_lun,
+                                                 128),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void vhost_scsi_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
+    FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(klass);
+
+    dc->props = vhost_scsi_properties;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+    vdc->realize = vhost_scsi_realize;
+    vdc->unrealize = vhost_scsi_unrealize;
+    vdc->get_features = vhost_scsi_get_features;
+    vdc->set_config = vhost_scsi_set_config;
+    vdc->set_status = vhost_scsi_set_status;
+    fwc->get_dev_path = vhost_scsi_get_fw_dev_path;
+}
+
+static void vhost_scsi_instance_init(Object *obj)
+{
+    VHostSCSI *dev = VHOST_SCSI(obj);
+
+    device_add_bootindex_property(obj, &dev->bootindex, "bootindex", NULL,
+                                  DEVICE(dev), NULL);
+}
+
+static const TypeInfo vhost_scsi_info = {
+    .name = TYPE_VHOST_SCSI,
+    .parent = TYPE_VIRTIO_SCSI_COMMON,
+    .instance_size = sizeof(VHostSCSI),
+    .class_init = vhost_scsi_class_init,
+    .instance_init = vhost_scsi_instance_init,
+    .interfaces = (InterfaceInfo[]) {
+        { TYPE_FW_PATH_PROVIDER },
+        { }
+    },
+};
+
+static void virtio_register_types(void)
+{
+    type_register_static(&vhost_scsi_info);
+}
+
+type_init(virtio_register_types)
diff --git a/qemu/hw/scsi/viosrp.h b/qemu/hw/scsi/viosrp.h
new file mode 100644
index 000000000..d8e365db1
--- /dev/null
+++ b/qemu/hw/scsi/viosrp.h
@@ -0,0 +1,216 @@
+/*****************************************************************************/
+/* srp.h -- SCSI RDMA Protocol definitions                                   */
+/*                                                                           */
+/* Written By: Colin Devilbis, IBM Corporation                               */
+/*                                                                           */
+/* Copyright (C) 2003 IBM Corporation                                        */
+/*                                                                           */
+/* 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; either version 2 of the License, or         */
+/* (at your option) any 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.  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., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */
+/*                                                                           */
+/*                                                                           */
+/* This file contains structures and definitions for IBM RPA (RS/6000        */
+/* platform architecture) implementation of the SRP (SCSI RDMA Protocol)     */
+/* standard.  SRP is used on IBM iSeries and pSeries platforms to send SCSI  */
+/* commands between logical partitions.                                      */
+/*                                                                           */
+/* SRP Information Units (IUs) are sent on a "Command/Response Queue" (CRQ)  */
+/* between partitions.  The definitions in this file are architected,        */
+/* and cannot be changed without breaking compatibility with other versions  */
+/* of Linux and other operating systems (AIX, OS/400) that talk this protocol*/
+/* between logical partitions                                                */
+/*****************************************************************************/
+#ifndef PPC_VIOSRP_H
+#define PPC_VIOSRP_H
+
+#define SRP_VERSION "16.a"
+#define SRP_MAX_IU_LEN    256
+#define SRP_MAX_LOC_LEN 32
+
+union srp_iu {
+    struct srp_login_req login_req;
+    struct srp_login_rsp login_rsp;
+    struct srp_login_rej login_rej;
+    struct srp_i_logout i_logout;
+    struct srp_t_logout t_logout;
+    struct srp_tsk_mgmt tsk_mgmt;
+    struct srp_cmd cmd;
+    struct srp_rsp rsp;
+    uint8_t reserved[SRP_MAX_IU_LEN];
+};
+
+enum viosrp_crq_formats {
+    VIOSRP_SRP_FORMAT = 0x01,
+    VIOSRP_MAD_FORMAT = 0x02,
+    VIOSRP_OS400_FORMAT = 0x03,
+    VIOSRP_AIX_FORMAT = 0x04,
+    VIOSRP_LINUX_FORMAT = 0x06,
+    VIOSRP_INLINE_FORMAT = 0x07
+};
+
+enum viosrp_crq_status {
+    VIOSRP_OK = 0x0,
+    VIOSRP_NONRECOVERABLE_ERR = 0x1,
+    VIOSRP_VIOLATES_MAX_XFER = 0x2,
+    VIOSRP_PARTNER_PANIC = 0x3,
+    VIOSRP_DEVICE_BUSY = 0x8,
+    VIOSRP_ADAPTER_FAIL = 0x10,
+    VIOSRP_OK2 = 0x99,
+};
+
+struct viosrp_crq {
+    uint8_t valid;        /* used by RPA */
+    uint8_t format;        /* SCSI vs out-of-band */
+    uint8_t reserved;
+    uint8_t status;        /* non-scsi failure? (e.g. DMA failure) */
+    uint16_t timeout;        /* in seconds */
+    uint16_t IU_length;        /* in bytes */
+    uint64_t IU_data_ptr;    /* the TCE for transferring data */
+};
+
+/* MADs are Management requests above and beyond the IUs defined in the SRP
+ * standard.
+ */
+enum viosrp_mad_types {
+    VIOSRP_EMPTY_IU_TYPE = 0x01,
+    VIOSRP_ERROR_LOG_TYPE = 0x02,
+    VIOSRP_ADAPTER_INFO_TYPE = 0x03,
+    VIOSRP_HOST_CONFIG_TYPE = 0x04,
+    VIOSRP_CAPABILITIES_TYPE = 0x05,
+    VIOSRP_ENABLE_FAST_FAIL = 0x08,
+};
+
+enum viosrp_mad_status {
+    VIOSRP_MAD_SUCCESS = 0x00,
+    VIOSRP_MAD_NOT_SUPPORTED = 0xF1,
+    VIOSRP_MAD_FAILED = 0xF7,
+};
+
+enum viosrp_capability_type {
+    MIGRATION_CAPABILITIES = 0x01,
+    RESERVATION_CAPABILITIES = 0x02,
+};
+
+enum viosrp_capability_support {
+    SERVER_DOES_NOT_SUPPORTS_CAP = 0x0,
+    SERVER_SUPPORTS_CAP = 0x01,
+    SERVER_CAP_DATA = 0x02,
+};
+
+enum viosrp_reserve_type {
+    CLIENT_RESERVE_SCSI_2 = 0x01,
+};
+
+enum viosrp_capability_flag {
+    CLIENT_MIGRATED = 0x01,
+    CLIENT_RECONNECT = 0x02,
+    CAP_LIST_SUPPORTED = 0x04,
+    CAP_LIST_DATA = 0x08,
+};
+
+/*
+ * Common MAD header
+ */
+struct mad_common {
+    uint32_t type;
+    uint16_t status;
+    uint16_t length;
+    uint64_t tag;
+};
+
+/*
+ * All SRP (and MAD) requests normally flow from the
+ * client to the server.  There is no way for the server to send
+ * an asynchronous message back to the client.  The Empty IU is used
+ * to hang out a meaningless request to the server so that it can respond
+ * asynchrouously with something like a SCSI AER
+ */
+struct viosrp_empty_iu {
+    struct mad_common common;
+    uint64_t buffer;
+    uint32_t port;
+};
+
+struct viosrp_error_log {
+    struct mad_common common;
+    uint64_t buffer;
+};
+
+struct viosrp_adapter_info {
+    struct mad_common common;
+    uint64_t buffer;
+};
+
+struct viosrp_host_config {
+    struct mad_common common;
+    uint64_t buffer;
+};
+
+struct viosrp_fast_fail {
+    struct mad_common common;
+};
+
+struct viosrp_capabilities {
+    struct mad_common common;
+    uint64_t buffer;
+};
+
+struct mad_capability_common {
+    uint32_t cap_type;
+    uint16_t length;
+    uint16_t server_support;
+};
+
+struct mad_reserve_cap {
+    struct mad_capability_common common;
+    uint32_t type;
+};
+
+struct mad_migration_cap {
+    struct mad_capability_common common;
+    uint32_t ecl;
+};
+
+struct capabilities {
+    uint32_t flags;
+    char name[SRP_MAX_LOC_LEN];
+    char loc[SRP_MAX_LOC_LEN];
+    struct mad_migration_cap migration;
+    struct mad_reserve_cap reserve;
+};
+
+union mad_iu {
+    struct viosrp_empty_iu empty_iu;
+    struct viosrp_error_log error_log;
+    struct viosrp_adapter_info adapter_info;
+    struct viosrp_host_config host_config;
+    struct viosrp_fast_fail fast_fail;
+    struct viosrp_capabilities capabilities;
+};
+
+union viosrp_iu {
+    union srp_iu srp;
+    union mad_iu mad;
+};
+
+struct mad_adapter_info_data {
+    char srp_version[8];
+    char partition_name[96];
+    uint32_t partition_number;
+    uint32_t mad_version;
+    uint32_t os_type;
+    uint32_t port_max_txu[8];    /* per-port maximum transfer */
+};
+
+#endif
diff --git a/qemu/hw/scsi/virtio-scsi-dataplane.c b/qemu/hw/scsi/virtio-scsi-dataplane.c
new file mode 100644
index 000000000..5575648a9
--- /dev/null
+++ b/qemu/hw/scsi/virtio-scsi-dataplane.c
@@ -0,0 +1,316 @@
+/*
+ * Virtio SCSI dataplane
+ *
+ * Copyright Red Hat, Inc. 2014
+ *
+ * Authors:
+ *   Fam Zheng <famz@redhat.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#include "hw/virtio/virtio-scsi.h"
+#include "qemu/error-report.h"
+#include "sysemu/block-backend.h"
+#include <hw/scsi/scsi.h>
+#include <block/scsi.h>
+#include <hw/virtio/virtio-bus.h>
+#include "hw/virtio/virtio-access.h"
+#include "stdio.h"
+
+/* Context: QEMU global mutex held */
+void virtio_scsi_set_iothread(VirtIOSCSI *s, IOThread *iothread)
+{
+    BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s)));
+    VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+
+    assert(!s->ctx);
+    s->ctx = iothread_get_aio_context(vs->conf.iothread);
+
+    /* Don't try if transport does not support notifiers. */
+    if (!k->set_guest_notifiers || !k->set_host_notifier) {
+        fprintf(stderr, "virtio-scsi: Failed to set iothread "
+                   "(transport does not support notifiers)");
+        exit(1);
+    }
+}
+
+static VirtIOSCSIVring *virtio_scsi_vring_init(VirtIOSCSI *s,
+                                               VirtQueue *vq,
+                                               EventNotifierHandler *handler,
+                                               int n)
+{
+    BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s)));
+    VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
+    VirtIOSCSIVring *r;
+    int rc;
+
+    /* Set up virtqueue notify */
+    rc = k->set_host_notifier(qbus->parent, n, true);
+    if (rc != 0) {
+        fprintf(stderr, "virtio-scsi: Failed to set host notifier (%d)\n",
+                rc);
+        s->dataplane_fenced = true;
+        return NULL;
+    }
+
+    r = g_slice_new(VirtIOSCSIVring);
+    r->host_notifier = *virtio_queue_get_host_notifier(vq);
+    r->guest_notifier = *virtio_queue_get_guest_notifier(vq);
+    aio_set_event_notifier(s->ctx, &r->host_notifier, handler);
+
+    r->parent = s;
+
+    if (!vring_setup(&r->vring, VIRTIO_DEVICE(s), n)) {
+        fprintf(stderr, "virtio-scsi: VRing setup failed\n");
+        goto fail_vring;
+    }
+    return r;
+
+fail_vring:
+    aio_set_event_notifier(s->ctx, &r->host_notifier, NULL);
+    k->set_host_notifier(qbus->parent, n, false);
+    g_slice_free(VirtIOSCSIVring, r);
+    return NULL;
+}
+
+VirtIOSCSIReq *virtio_scsi_pop_req_vring(VirtIOSCSI *s,
+                                         VirtIOSCSIVring *vring)
+{
+    VirtIOSCSIReq *req = virtio_scsi_init_req(s, NULL);
+    int r;
+
+    req->vring = vring;
+    r = vring_pop((VirtIODevice *)s, &vring->vring, &req->elem);
+    if (r < 0) {
+        virtio_scsi_free_req(req);
+        req = NULL;
+    }
+    return req;
+}
+
+void virtio_scsi_vring_push_notify(VirtIOSCSIReq *req)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(req->vring->parent);
+
+    vring_push(vdev, &req->vring->vring, &req->elem,
+               req->qsgl.size + req->resp_iov.size);
+
+    if (vring_should_notify(vdev, &req->vring->vring)) {
+        event_notifier_set(&req->vring->guest_notifier);
+    }
+}
+
+static void virtio_scsi_iothread_handle_ctrl(EventNotifier *notifier)
+{
+    VirtIOSCSIVring *vring = container_of(notifier,
+                                          VirtIOSCSIVring, host_notifier);
+    VirtIOSCSI *s = VIRTIO_SCSI(vring->parent);
+    VirtIOSCSIReq *req;
+
+    event_notifier_test_and_clear(notifier);
+    while ((req = virtio_scsi_pop_req_vring(s, vring))) {
+        virtio_scsi_handle_ctrl_req(s, req);
+    }
+}
+
+static void virtio_scsi_iothread_handle_event(EventNotifier *notifier)
+{
+    VirtIOSCSIVring *vring = container_of(notifier,
+                                          VirtIOSCSIVring, host_notifier);
+    VirtIOSCSI *s = vring->parent;
+    VirtIODevice *vdev = VIRTIO_DEVICE(s);
+
+    event_notifier_test_and_clear(notifier);
+
+    if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
+        return;
+    }
+
+    if (s->events_dropped) {
+        virtio_scsi_push_event(s, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
+    }
+}
+
+static void virtio_scsi_iothread_handle_cmd(EventNotifier *notifier)
+{
+    VirtIOSCSIVring *vring = container_of(notifier,
+                                          VirtIOSCSIVring, host_notifier);
+    VirtIOSCSI *s = (VirtIOSCSI *)vring->parent;
+    VirtIOSCSIReq *req, *next;
+    QTAILQ_HEAD(, VirtIOSCSIReq) reqs = QTAILQ_HEAD_INITIALIZER(reqs);
+
+    event_notifier_test_and_clear(notifier);
+    while ((req = virtio_scsi_pop_req_vring(s, vring))) {
+        if (virtio_scsi_handle_cmd_req_prepare(s, req)) {
+            QTAILQ_INSERT_TAIL(&reqs, req, next);
+        }
+    }
+
+    QTAILQ_FOREACH_SAFE(req, &reqs, next, next) {
+        virtio_scsi_handle_cmd_req_submit(s, req);
+    }
+}
+
+/* assumes s->ctx held */
+static void virtio_scsi_clear_aio(VirtIOSCSI *s)
+{
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+    int i;
+
+    if (s->ctrl_vring) {
+        aio_set_event_notifier(s->ctx, &s->ctrl_vring->host_notifier, NULL);
+    }
+    if (s->event_vring) {
+        aio_set_event_notifier(s->ctx, &s->event_vring->host_notifier, NULL);
+    }
+    if (s->cmd_vrings) {
+        for (i = 0; i < vs->conf.num_queues && s->cmd_vrings[i]; i++) {
+            aio_set_event_notifier(s->ctx, &s->cmd_vrings[i]->host_notifier, NULL);
+        }
+    }
+}
+
+static void virtio_scsi_vring_teardown(VirtIOSCSI *s)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(s);
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+    int i;
+
+    if (s->ctrl_vring) {
+        vring_teardown(&s->ctrl_vring->vring, vdev, 0);
+        g_slice_free(VirtIOSCSIVring, s->ctrl_vring);
+        s->ctrl_vring = NULL;
+    }
+    if (s->event_vring) {
+        vring_teardown(&s->event_vring->vring, vdev, 1);
+        g_slice_free(VirtIOSCSIVring, s->event_vring);
+        s->event_vring = NULL;
+    }
+    if (s->cmd_vrings) {
+        for (i = 0; i < vs->conf.num_queues && s->cmd_vrings[i]; i++) {
+            vring_teardown(&s->cmd_vrings[i]->vring, vdev, 2 + i);
+            g_slice_free(VirtIOSCSIVring, s->cmd_vrings[i]);
+            s->cmd_vrings[i] = NULL;
+        }
+        free(s->cmd_vrings);
+        s->cmd_vrings = NULL;
+    }
+}
+
+/* Context: QEMU global mutex held */
+void virtio_scsi_dataplane_start(VirtIOSCSI *s)
+{
+    int i;
+    int rc;
+    BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s)));
+    VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+
+    if (s->dataplane_started ||
+        s->dataplane_starting ||
+        s->dataplane_fenced ||
+        s->ctx != iothread_get_aio_context(vs->conf.iothread)) {
+        return;
+    }
+
+    s->dataplane_starting = true;
+
+    /* Set up guest notifier (irq) */
+    rc = k->set_guest_notifiers(qbus->parent, vs->conf.num_queues + 2, true);
+    if (rc != 0) {
+        fprintf(stderr, "virtio-scsi: Failed to set guest notifiers (%d), "
+                "ensure -enable-kvm is set\n", rc);
+        s->dataplane_fenced = true;
+        goto fail_guest_notifiers;
+    }
+
+    aio_context_acquire(s->ctx);
+    s->ctrl_vring = virtio_scsi_vring_init(s, vs->ctrl_vq,
+                                           virtio_scsi_iothread_handle_ctrl,
+                                           0);
+    if (!s->ctrl_vring) {
+        goto fail_vrings;
+    }
+    s->event_vring = virtio_scsi_vring_init(s, vs->event_vq,
+                                            virtio_scsi_iothread_handle_event,
+                                            1);
+    if (!s->event_vring) {
+        goto fail_vrings;
+    }
+    s->cmd_vrings = g_new(VirtIOSCSIVring *, vs->conf.num_queues);
+    for (i = 0; i < vs->conf.num_queues; i++) {
+        s->cmd_vrings[i] =
+            virtio_scsi_vring_init(s, vs->cmd_vqs[i],
+                                   virtio_scsi_iothread_handle_cmd,
+                                   i + 2);
+        if (!s->cmd_vrings[i]) {
+            goto fail_vrings;
+        }
+    }
+
+    s->dataplane_starting = false;
+    s->dataplane_started = true;
+    aio_context_release(s->ctx);
+    return;
+
+fail_vrings:
+    virtio_scsi_clear_aio(s);
+    aio_context_release(s->ctx);
+    virtio_scsi_vring_teardown(s);
+    for (i = 0; i < vs->conf.num_queues + 2; i++) {
+        k->set_host_notifier(qbus->parent, i, false);
+    }
+    k->set_guest_notifiers(qbus->parent, vs->conf.num_queues + 2, false);
+fail_guest_notifiers:
+    s->dataplane_starting = false;
+}
+
+/* Context: QEMU global mutex held */
+void virtio_scsi_dataplane_stop(VirtIOSCSI *s)
+{
+    BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s)));
+    VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+    int i;
+
+    /* Better luck next time. */
+    if (s->dataplane_fenced) {
+        s->dataplane_fenced = false;
+        return;
+    }
+    if (!s->dataplane_started || s->dataplane_stopping) {
+        return;
+    }
+    s->dataplane_stopping = true;
+    assert(s->ctx == iothread_get_aio_context(vs->conf.iothread));
+
+    aio_context_acquire(s->ctx);
+
+    aio_set_event_notifier(s->ctx, &s->ctrl_vring->host_notifier, NULL);
+    aio_set_event_notifier(s->ctx, &s->event_vring->host_notifier, NULL);
+    for (i = 0; i < vs->conf.num_queues; i++) {
+        aio_set_event_notifier(s->ctx, &s->cmd_vrings[i]->host_notifier, NULL);
+    }
+
+    blk_drain_all(); /* ensure there are no in-flight requests */
+
+    aio_context_release(s->ctx);
+
+    /* Sync vring state back to virtqueue so that non-dataplane request
+     * processing can continue when we disable the host notifier below.
+     */
+    virtio_scsi_vring_teardown(s);
+
+    for (i = 0; i < vs->conf.num_queues + 2; i++) {
+        k->set_host_notifier(qbus->parent, i, false);
+    }
+
+    /* Clean up guest notifier (irq) */
+    k->set_guest_notifiers(qbus->parent, vs->conf.num_queues + 2, false);
+    s->dataplane_stopping = false;
+    s->dataplane_started = false;
+}
diff --git a/qemu/hw/scsi/virtio-scsi.c b/qemu/hw/scsi/virtio-scsi.c
new file mode 100644
index 000000000..811c3da8b
--- /dev/null
+++ b/qemu/hw/scsi/virtio-scsi.c
@@ -0,0 +1,1016 @@
+/*
+ * Virtio SCSI HBA
+ *
+ * Copyright IBM, Corp. 2010
+ * Copyright Red Hat, Inc. 2011
+ *
+ * Authors:
+ *   Stefan Hajnoczi    <stefanha@linux.vnet.ibm.com>
+ *   Paolo Bonzini      <pbonzini@redhat.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#include "standard-headers/linux/virtio_ids.h"
+#include "hw/virtio/virtio-scsi.h"
+#include "qemu/error-report.h"
+#include "qemu/iov.h"
+#include "sysemu/block-backend.h"
+#include <hw/scsi/scsi.h>
+#include <block/scsi.h>
+#include <hw/virtio/virtio-bus.h>
+#include "hw/virtio/virtio-access.h"
+#include "migration/migration.h"
+
+static inline int virtio_scsi_get_lun(uint8_t *lun)
+{
+    return ((lun[2] << 8) | lun[3]) & 0x3FFF;
+}
+
+static inline SCSIDevice *virtio_scsi_device_find(VirtIOSCSI *s, uint8_t *lun)
+{
+    if (lun[0] != 1) {
+        return NULL;
+    }
+    if (lun[2] != 0 && !(lun[2] >= 0x40 && lun[2] < 0x80)) {
+        return NULL;
+    }
+    return scsi_device_find(&s->bus, 0, lun[1], virtio_scsi_get_lun(lun));
+}
+
+VirtIOSCSIReq *virtio_scsi_init_req(VirtIOSCSI *s, VirtQueue *vq)
+{
+    VirtIOSCSIReq *req;
+    VirtIOSCSICommon *vs = (VirtIOSCSICommon *)s;
+    const size_t zero_skip = offsetof(VirtIOSCSIReq, elem)
+                             + sizeof(VirtQueueElement);
+
+    req = g_slice_alloc(sizeof(*req) + vs->cdb_size);
+    req->vq = vq;
+    req->dev = s;
+    qemu_sglist_init(&req->qsgl, DEVICE(s), 8, &address_space_memory);
+    qemu_iovec_init(&req->resp_iov, 1);
+    memset((uint8_t *)req + zero_skip, 0, sizeof(*req) - zero_skip);
+    return req;
+}
+
+void virtio_scsi_free_req(VirtIOSCSIReq *req)
+{
+    VirtIOSCSICommon *vs = (VirtIOSCSICommon *)req->dev;
+
+    qemu_iovec_destroy(&req->resp_iov);
+    qemu_sglist_destroy(&req->qsgl);
+    g_slice_free1(sizeof(*req) + vs->cdb_size, req);
+}
+
+static void virtio_scsi_complete_req(VirtIOSCSIReq *req)
+{
+    VirtIOSCSI *s = req->dev;
+    VirtQueue *vq = req->vq;
+    VirtIODevice *vdev = VIRTIO_DEVICE(s);
+
+    qemu_iovec_from_buf(&req->resp_iov, 0, &req->resp, req->resp_size);
+    if (req->vring) {
+        assert(req->vq == NULL);
+        virtio_scsi_vring_push_notify(req);
+    } else {
+        virtqueue_push(vq, &req->elem, req->qsgl.size + req->resp_iov.size);
+        virtio_notify(vdev, vq);
+    }
+
+    if (req->sreq) {
+        req->sreq->hba_private = NULL;
+        scsi_req_unref(req->sreq);
+    }
+    virtio_scsi_free_req(req);
+}
+
+static void virtio_scsi_bad_req(void)
+{
+    error_report("wrong size for virtio-scsi headers");
+    exit(1);
+}
+
+static size_t qemu_sgl_concat(VirtIOSCSIReq *req, struct iovec *iov,
+                              hwaddr *addr, int num, size_t skip)
+{
+    QEMUSGList *qsgl = &req->qsgl;
+    size_t copied = 0;
+
+    while (num) {
+        if (skip >= iov->iov_len) {
+            skip -= iov->iov_len;
+        } else {
+            qemu_sglist_add(qsgl, *addr + skip, iov->iov_len - skip);
+            copied += iov->iov_len - skip;
+            skip = 0;
+        }
+        iov++;
+        addr++;
+        num--;
+    }
+
+    assert(skip == 0);
+    return copied;
+}
+
+static int virtio_scsi_parse_req(VirtIOSCSIReq *req,
+                                 unsigned req_size, unsigned resp_size)
+{
+    VirtIODevice *vdev = (VirtIODevice *) req->dev;
+    size_t in_size, out_size;
+
+    if (iov_to_buf(req->elem.out_sg, req->elem.out_num, 0,
+                   &req->req, req_size) < req_size) {
+        return -EINVAL;
+    }
+
+    if (qemu_iovec_concat_iov(&req->resp_iov,
+                              req->elem.in_sg, req->elem.in_num, 0,
+                              resp_size) < resp_size) {
+        return -EINVAL;
+    }
+
+    req->resp_size = resp_size;
+
+    /* Old BIOSes left some padding by mistake after the req_size/resp_size.
+     * As a workaround, always consider the first buffer as the virtio-scsi
+     * request/response, making the payload start at the second element
+     * of the iovec.
+     *
+     * The actual length of the response header, stored in req->resp_size,
+     * does not change.
+     *
+     * TODO: always disable this workaround for virtio 1.0 devices.
+     */
+    if (!virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT)) {
+        if (req->elem.out_num) {
+            req_size = req->elem.out_sg[0].iov_len;
+        }
+        if (req->elem.in_num) {
+            resp_size = req->elem.in_sg[0].iov_len;
+        }
+    }
+
+    out_size = qemu_sgl_concat(req, req->elem.out_sg,
+                               &req->elem.out_addr[0], req->elem.out_num,
+                               req_size);
+    in_size = qemu_sgl_concat(req, req->elem.in_sg,
+                              &req->elem.in_addr[0], req->elem.in_num,
+                              resp_size);
+
+    if (out_size && in_size) {
+        return -ENOTSUP;
+    }
+
+    if (out_size) {
+        req->mode = SCSI_XFER_TO_DEV;
+    } else if (in_size) {
+        req->mode = SCSI_XFER_FROM_DEV;
+    }
+
+    return 0;
+}
+
+static VirtIOSCSIReq *virtio_scsi_pop_req(VirtIOSCSI *s, VirtQueue *vq)
+{
+    VirtIOSCSIReq *req = virtio_scsi_init_req(s, vq);
+    if (!virtqueue_pop(vq, &req->elem)) {
+        virtio_scsi_free_req(req);
+        return NULL;
+    }
+    return req;
+}
+
+static void virtio_scsi_save_request(QEMUFile *f, SCSIRequest *sreq)
+{
+    VirtIOSCSIReq *req = sreq->hba_private;
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(req->dev);
+    uint32_t n = virtio_queue_get_id(req->vq) - 2;
+
+    assert(n < vs->conf.num_queues);
+    qemu_put_be32s(f, &n);
+    qemu_put_buffer(f, (unsigned char *)&req->elem, sizeof(req->elem));
+}
+
+static void *virtio_scsi_load_request(QEMUFile *f, SCSIRequest *sreq)
+{
+    SCSIBus *bus = sreq->bus;
+    VirtIOSCSI *s = container_of(bus, VirtIOSCSI, bus);
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+    VirtIOSCSIReq *req;
+    uint32_t n;
+
+    qemu_get_be32s(f, &n);
+    assert(n < vs->conf.num_queues);
+    req = virtio_scsi_init_req(s, vs->cmd_vqs[n]);
+    qemu_get_buffer(f, (unsigned char *)&req->elem, sizeof(req->elem));
+    /* TODO: add a way for SCSIBusInfo's load_request to fail,
+     * and fail migration instead of asserting here.
+     * When we do, we might be able to re-enable NDEBUG below.
+     */
+#ifdef NDEBUG
+#error building with NDEBUG is not supported
+#endif
+    assert(req->elem.in_num <= ARRAY_SIZE(req->elem.in_sg));
+    assert(req->elem.out_num <= ARRAY_SIZE(req->elem.out_sg));
+
+    if (virtio_scsi_parse_req(req, sizeof(VirtIOSCSICmdReq) + vs->cdb_size,
+                              sizeof(VirtIOSCSICmdResp) + vs->sense_size) < 0) {
+        error_report("invalid SCSI request migration data");
+        exit(1);
+    }
+
+    scsi_req_ref(sreq);
+    req->sreq = sreq;
+    if (req->sreq->cmd.mode != SCSI_XFER_NONE) {
+        assert(req->sreq->cmd.mode == req->mode);
+    }
+    return req;
+}
+
+typedef struct {
+    Notifier        notifier;
+    VirtIOSCSIReq  *tmf_req;
+} VirtIOSCSICancelNotifier;
+
+static void virtio_scsi_cancel_notify(Notifier *notifier, void *data)
+{
+    VirtIOSCSICancelNotifier *n = container_of(notifier,
+                                               VirtIOSCSICancelNotifier,
+                                               notifier);
+
+    if (--n->tmf_req->remaining == 0) {
+        virtio_scsi_complete_req(n->tmf_req);
+    }
+    g_slice_free(VirtIOSCSICancelNotifier, n);
+}
+
+/* Return 0 if the request is ready to be completed and return to guest;
+ * -EINPROGRESS if the request is submitted and will be completed later, in the
+ *  case of async cancellation. */
+static int virtio_scsi_do_tmf(VirtIOSCSI *s, VirtIOSCSIReq *req)
+{
+    SCSIDevice *d = virtio_scsi_device_find(s, req->req.tmf.lun);
+    SCSIRequest *r, *next;
+    BusChild *kid;
+    int target;
+    int ret = 0;
+
+    if (s->dataplane_started) {
+        assert(blk_get_aio_context(d->conf.blk) == s->ctx);
+    }
+    /* Here VIRTIO_SCSI_S_OK means "FUNCTION COMPLETE".  */
+    req->resp.tmf.response = VIRTIO_SCSI_S_OK;
+
+    virtio_tswap32s(VIRTIO_DEVICE(s), &req->req.tmf.subtype);
+    switch (req->req.tmf.subtype) {
+    case VIRTIO_SCSI_T_TMF_ABORT_TASK:
+    case VIRTIO_SCSI_T_TMF_QUERY_TASK:
+        if (!d) {
+            goto fail;
+        }
+        if (d->lun != virtio_scsi_get_lun(req->req.tmf.lun)) {
+            goto incorrect_lun;
+        }
+        QTAILQ_FOREACH_SAFE(r, &d->requests, next, next) {
+            VirtIOSCSIReq *cmd_req = r->hba_private;
+            if (cmd_req && cmd_req->req.cmd.tag == req->req.tmf.tag) {
+                break;
+            }
+        }
+        if (r) {
+            /*
+             * Assert that the request has not been completed yet, we
+             * check for it in the loop above.
+             */
+            assert(r->hba_private);
+            if (req->req.tmf.subtype == VIRTIO_SCSI_T_TMF_QUERY_TASK) {
+                /* "If the specified command is present in the task set, then
+                 * return a service response set to FUNCTION SUCCEEDED".
+                 */
+                req->resp.tmf.response = VIRTIO_SCSI_S_FUNCTION_SUCCEEDED;
+            } else {
+                VirtIOSCSICancelNotifier *notifier;
+
+                req->remaining = 1;
+                notifier = g_slice_new(VirtIOSCSICancelNotifier);
+                notifier->tmf_req = req;
+                notifier->notifier.notify = virtio_scsi_cancel_notify;
+                scsi_req_cancel_async(r, &notifier->notifier);
+                ret = -EINPROGRESS;
+            }
+        }
+        break;
+
+    case VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET:
+        if (!d) {
+            goto fail;
+        }
+        if (d->lun != virtio_scsi_get_lun(req->req.tmf.lun)) {
+            goto incorrect_lun;
+        }
+        s->resetting++;
+        qdev_reset_all(&d->qdev);
+        s->resetting--;
+        break;
+
+    case VIRTIO_SCSI_T_TMF_ABORT_TASK_SET:
+    case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET:
+    case VIRTIO_SCSI_T_TMF_QUERY_TASK_SET:
+        if (!d) {
+            goto fail;
+        }
+        if (d->lun != virtio_scsi_get_lun(req->req.tmf.lun)) {
+            goto incorrect_lun;
+        }
+
+        /* Add 1 to "remaining" until virtio_scsi_do_tmf returns.
+         * This way, if the bus starts calling back to the notifiers
+         * even before we finish the loop, virtio_scsi_cancel_notify
+         * will not complete the TMF too early.
+         */
+        req->remaining = 1;
+        QTAILQ_FOREACH_SAFE(r, &d->requests, next, next) {
+            if (r->hba_private) {
+                if (req->req.tmf.subtype == VIRTIO_SCSI_T_TMF_QUERY_TASK_SET) {
+                    /* "If there is any command present in the task set, then
+                     * return a service response set to FUNCTION SUCCEEDED".
+                     */
+                    req->resp.tmf.response = VIRTIO_SCSI_S_FUNCTION_SUCCEEDED;
+                    break;
+                } else {
+                    VirtIOSCSICancelNotifier *notifier;
+
+                    req->remaining++;
+                    notifier = g_slice_new(VirtIOSCSICancelNotifier);
+                    notifier->notifier.notify = virtio_scsi_cancel_notify;
+                    notifier->tmf_req = req;
+                    scsi_req_cancel_async(r, &notifier->notifier);
+                }
+            }
+        }
+        if (--req->remaining > 0) {
+            ret = -EINPROGRESS;
+        }
+        break;
+
+    case VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET:
+        target = req->req.tmf.lun[1];
+        s->resetting++;
+        QTAILQ_FOREACH(kid, &s->bus.qbus.children, sibling) {
+             d = DO_UPCAST(SCSIDevice, qdev, kid->child);
+             if (d->channel == 0 && d->id == target) {
+                qdev_reset_all(&d->qdev);
+             }
+        }
+        s->resetting--;
+        break;
+
+    case VIRTIO_SCSI_T_TMF_CLEAR_ACA:
+    default:
+        req->resp.tmf.response = VIRTIO_SCSI_S_FUNCTION_REJECTED;
+        break;
+    }
+
+    return ret;
+
+incorrect_lun:
+    req->resp.tmf.response = VIRTIO_SCSI_S_INCORRECT_LUN;
+    return ret;
+
+fail:
+    req->resp.tmf.response = VIRTIO_SCSI_S_BAD_TARGET;
+    return ret;
+}
+
+void virtio_scsi_handle_ctrl_req(VirtIOSCSI *s, VirtIOSCSIReq *req)
+{
+    VirtIODevice *vdev = (VirtIODevice *)s;
+    uint32_t type;
+    int r = 0;
+
+    if (iov_to_buf(req->elem.out_sg, req->elem.out_num, 0,
+                &type, sizeof(type)) < sizeof(type)) {
+        virtio_scsi_bad_req();
+        return;
+    }
+
+    virtio_tswap32s(vdev, &type);
+    if (type == VIRTIO_SCSI_T_TMF) {
+        if (virtio_scsi_parse_req(req, sizeof(VirtIOSCSICtrlTMFReq),
+                    sizeof(VirtIOSCSICtrlTMFResp)) < 0) {
+            virtio_scsi_bad_req();
+        } else {
+            r = virtio_scsi_do_tmf(s, req);
+        }
+
+    } else if (type == VIRTIO_SCSI_T_AN_QUERY ||
+               type == VIRTIO_SCSI_T_AN_SUBSCRIBE) {
+        if (virtio_scsi_parse_req(req, sizeof(VirtIOSCSICtrlANReq),
+                    sizeof(VirtIOSCSICtrlANResp)) < 0) {
+            virtio_scsi_bad_req();
+        } else {
+            req->resp.an.event_actual = 0;
+            req->resp.an.response = VIRTIO_SCSI_S_OK;
+        }
+    }
+    if (r == 0) {
+        virtio_scsi_complete_req(req);
+    } else {
+        assert(r == -EINPROGRESS);
+    }
+}
+
+static void virtio_scsi_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
+{
+    VirtIOSCSI *s = (VirtIOSCSI *)vdev;
+    VirtIOSCSIReq *req;
+
+    if (s->ctx && !s->dataplane_disabled) {
+        virtio_scsi_dataplane_start(s);
+        return;
+    }
+    while ((req = virtio_scsi_pop_req(s, vq))) {
+        virtio_scsi_handle_ctrl_req(s, req);
+    }
+}
+
+static void virtio_scsi_complete_cmd_req(VirtIOSCSIReq *req)
+{
+    /* Sense data is not in req->resp and is copied separately
+     * in virtio_scsi_command_complete.
+     */
+    req->resp_size = sizeof(VirtIOSCSICmdResp);
+    virtio_scsi_complete_req(req);
+}
+
+static void virtio_scsi_command_complete(SCSIRequest *r, uint32_t status,
+                                         size_t resid)
+{
+    VirtIOSCSIReq *req = r->hba_private;
+    uint8_t sense[SCSI_SENSE_BUF_SIZE];
+    uint32_t sense_len;
+    VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
+
+    if (r->io_canceled) {
+        return;
+    }
+
+    req->resp.cmd.response = VIRTIO_SCSI_S_OK;
+    req->resp.cmd.status = status;
+    if (req->resp.cmd.status == GOOD) {
+        req->resp.cmd.resid = virtio_tswap32(vdev, resid);
+    } else {
+        req->resp.cmd.resid = 0;
+        sense_len = scsi_req_get_sense(r, sense, sizeof(sense));
+        sense_len = MIN(sense_len, req->resp_iov.size - sizeof(req->resp.cmd));
+        qemu_iovec_from_buf(&req->resp_iov, sizeof(req->resp.cmd),
+                            sense, sense_len);
+        req->resp.cmd.sense_len = virtio_tswap32(vdev, sense_len);
+    }
+    virtio_scsi_complete_cmd_req(req);
+}
+
+static int virtio_scsi_parse_cdb(SCSIDevice *dev, SCSICommand *cmd,
+                                 uint8_t *buf, void *hba_private)
+{
+    VirtIOSCSIReq *req = hba_private;
+
+    if (cmd->len == 0) {
+        cmd->len = MIN(VIRTIO_SCSI_CDB_DEFAULT_SIZE, SCSI_CMD_BUF_SIZE);
+        memcpy(cmd->buf, buf, cmd->len);
+    }
+
+    /* Extract the direction and mode directly from the request, for
+     * host device passthrough.
+     */
+    cmd->xfer = req->qsgl.size;
+    cmd->mode = req->mode;
+    return 0;
+}
+
+static QEMUSGList *virtio_scsi_get_sg_list(SCSIRequest *r)
+{
+    VirtIOSCSIReq *req = r->hba_private;
+
+    return &req->qsgl;
+}
+
+static void virtio_scsi_request_cancelled(SCSIRequest *r)
+{
+    VirtIOSCSIReq *req = r->hba_private;
+
+    if (!req) {
+        return;
+    }
+    if (req->dev->resetting) {
+        req->resp.cmd.response = VIRTIO_SCSI_S_RESET;
+    } else {
+        req->resp.cmd.response = VIRTIO_SCSI_S_ABORTED;
+    }
+    virtio_scsi_complete_cmd_req(req);
+}
+
+static void virtio_scsi_fail_cmd_req(VirtIOSCSIReq *req)
+{
+    req->resp.cmd.response = VIRTIO_SCSI_S_FAILURE;
+    virtio_scsi_complete_cmd_req(req);
+}
+
+bool virtio_scsi_handle_cmd_req_prepare(VirtIOSCSI *s, VirtIOSCSIReq *req)
+{
+    VirtIOSCSICommon *vs = &s->parent_obj;
+    SCSIDevice *d;
+    int rc;
+
+    rc = virtio_scsi_parse_req(req, sizeof(VirtIOSCSICmdReq) + vs->cdb_size,
+                               sizeof(VirtIOSCSICmdResp) + vs->sense_size);
+    if (rc < 0) {
+        if (rc == -ENOTSUP) {
+            virtio_scsi_fail_cmd_req(req);
+        } else {
+            virtio_scsi_bad_req();
+        }
+        return false;
+    }
+
+    d = virtio_scsi_device_find(s, req->req.cmd.lun);
+    if (!d) {
+        req->resp.cmd.response = VIRTIO_SCSI_S_BAD_TARGET;
+        virtio_scsi_complete_cmd_req(req);
+        return false;
+    }
+    if (s->dataplane_started) {
+        assert(blk_get_aio_context(d->conf.blk) == s->ctx);
+    }
+    req->sreq = scsi_req_new(d, req->req.cmd.tag,
+                             virtio_scsi_get_lun(req->req.cmd.lun),
+                             req->req.cmd.cdb, req);
+
+    if (req->sreq->cmd.mode != SCSI_XFER_NONE
+        && (req->sreq->cmd.mode != req->mode ||
+            req->sreq->cmd.xfer > req->qsgl.size)) {
+        req->resp.cmd.response = VIRTIO_SCSI_S_OVERRUN;
+        virtio_scsi_complete_cmd_req(req);
+        return false;
+    }
+    scsi_req_ref(req->sreq);
+    blk_io_plug(d->conf.blk);
+    return true;
+}
+
+void virtio_scsi_handle_cmd_req_submit(VirtIOSCSI *s, VirtIOSCSIReq *req)
+{
+    SCSIRequest *sreq = req->sreq;
+    if (scsi_req_enqueue(sreq)) {
+        scsi_req_continue(sreq);
+    }
+    blk_io_unplug(sreq->dev->conf.blk);
+    scsi_req_unref(sreq);
+}
+
+static void virtio_scsi_handle_cmd(VirtIODevice *vdev, VirtQueue *vq)
+{
+    /* use non-QOM casts in the data path */
+    VirtIOSCSI *s = (VirtIOSCSI *)vdev;
+    VirtIOSCSIReq *req, *next;
+    QTAILQ_HEAD(, VirtIOSCSIReq) reqs = QTAILQ_HEAD_INITIALIZER(reqs);
+
+    if (s->ctx && !s->dataplane_disabled) {
+        virtio_scsi_dataplane_start(s);
+        return;
+    }
+    while ((req = virtio_scsi_pop_req(s, vq))) {
+        if (virtio_scsi_handle_cmd_req_prepare(s, req)) {
+            QTAILQ_INSERT_TAIL(&reqs, req, next);
+        }
+    }
+
+    QTAILQ_FOREACH_SAFE(req, &reqs, next, next) {
+        virtio_scsi_handle_cmd_req_submit(s, req);
+    }
+}
+
+static void virtio_scsi_get_config(VirtIODevice *vdev,
+                                   uint8_t *config)
+{
+    VirtIOSCSIConfig *scsiconf = (VirtIOSCSIConfig *)config;
+    VirtIOSCSICommon *s = VIRTIO_SCSI_COMMON(vdev);
+
+    virtio_stl_p(vdev, &scsiconf->num_queues, s->conf.num_queues);
+    virtio_stl_p(vdev, &scsiconf->seg_max, 128 - 2);
+    virtio_stl_p(vdev, &scsiconf->max_sectors, s->conf.max_sectors);
+    virtio_stl_p(vdev, &scsiconf->cmd_per_lun, s->conf.cmd_per_lun);
+    virtio_stl_p(vdev, &scsiconf->event_info_size, sizeof(VirtIOSCSIEvent));
+    virtio_stl_p(vdev, &scsiconf->sense_size, s->sense_size);
+    virtio_stl_p(vdev, &scsiconf->cdb_size, s->cdb_size);
+    virtio_stw_p(vdev, &scsiconf->max_channel, VIRTIO_SCSI_MAX_CHANNEL);
+    virtio_stw_p(vdev, &scsiconf->max_target, VIRTIO_SCSI_MAX_TARGET);
+    virtio_stl_p(vdev, &scsiconf->max_lun, VIRTIO_SCSI_MAX_LUN);
+}
+
+static void virtio_scsi_set_config(VirtIODevice *vdev,
+                                   const uint8_t *config)
+{
+    VirtIOSCSIConfig *scsiconf = (VirtIOSCSIConfig *)config;
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(vdev);
+
+    if ((uint32_t) virtio_ldl_p(vdev, &scsiconf->sense_size) >= 65536 ||
+        (uint32_t) virtio_ldl_p(vdev, &scsiconf->cdb_size) >= 256) {
+        error_report("bad data written to virtio-scsi configuration space");
+        exit(1);
+    }
+
+    vs->sense_size = virtio_ldl_p(vdev, &scsiconf->sense_size);
+    vs->cdb_size = virtio_ldl_p(vdev, &scsiconf->cdb_size);
+}
+
+static uint64_t virtio_scsi_get_features(VirtIODevice *vdev,
+                                         uint64_t requested_features,
+                                         Error **errp)
+{
+    VirtIOSCSI *s = VIRTIO_SCSI(vdev);
+
+    /* Firstly sync all virtio-scsi possible supported features */
+    requested_features |= s->host_features;
+    return requested_features;
+}
+
+static void virtio_scsi_reset(VirtIODevice *vdev)
+{
+    VirtIOSCSI *s = VIRTIO_SCSI(vdev);
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(vdev);
+
+    if (s->ctx) {
+        virtio_scsi_dataplane_stop(s);
+    }
+    s->resetting++;
+    qbus_reset_all(&s->bus.qbus);
+    s->resetting--;
+
+    vs->sense_size = VIRTIO_SCSI_SENSE_DEFAULT_SIZE;
+    vs->cdb_size = VIRTIO_SCSI_CDB_DEFAULT_SIZE;
+    s->events_dropped = false;
+}
+
+/* The device does not have anything to save beyond the virtio data.
+ * Request data is saved with callbacks from SCSI devices.
+ */
+static void virtio_scsi_save(QEMUFile *f, void *opaque)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
+    virtio_save(vdev, f);
+}
+
+static int virtio_scsi_load(QEMUFile *f, void *opaque, int version_id)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
+    int ret;
+
+    ret = virtio_load(vdev, f, version_id);
+    if (ret) {
+        return ret;
+    }
+    return 0;
+}
+
+void virtio_scsi_push_event(VirtIOSCSI *s, SCSIDevice *dev,
+                            uint32_t event, uint32_t reason)
+{
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+    VirtIOSCSIReq *req;
+    VirtIOSCSIEvent *evt;
+    VirtIODevice *vdev = VIRTIO_DEVICE(s);
+
+    if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
+        return;
+    }
+
+    if (s->dataplane_started) {
+        assert(s->ctx);
+        aio_context_acquire(s->ctx);
+    }
+
+    if (s->dataplane_started) {
+        req = virtio_scsi_pop_req_vring(s, s->event_vring);
+    } else {
+        req = virtio_scsi_pop_req(s, vs->event_vq);
+    }
+    if (!req) {
+        s->events_dropped = true;
+        goto out;
+    }
+
+    if (s->events_dropped) {
+        event |= VIRTIO_SCSI_T_EVENTS_MISSED;
+        s->events_dropped = false;
+    }
+
+    if (virtio_scsi_parse_req(req, 0, sizeof(VirtIOSCSIEvent))) {
+        virtio_scsi_bad_req();
+    }
+
+    evt = &req->resp.event;
+    memset(evt, 0, sizeof(VirtIOSCSIEvent));
+    evt->event = virtio_tswap32(vdev, event);
+    evt->reason = virtio_tswap32(vdev, reason);
+    if (!dev) {
+        assert(event == VIRTIO_SCSI_T_EVENTS_MISSED);
+    } else {
+        evt->lun[0] = 1;
+        evt->lun[1] = dev->id;
+
+        /* Linux wants us to keep the same encoding we use for REPORT LUNS.  */
+        if (dev->lun >= 256) {
+            evt->lun[2] = (dev->lun >> 8) | 0x40;
+        }
+        evt->lun[3] = dev->lun & 0xFF;
+    }
+    virtio_scsi_complete_req(req);
+out:
+    if (s->dataplane_started) {
+        aio_context_release(s->ctx);
+    }
+}
+
+static void virtio_scsi_handle_event(VirtIODevice *vdev, VirtQueue *vq)
+{
+    VirtIOSCSI *s = VIRTIO_SCSI(vdev);
+
+    if (s->ctx && !s->dataplane_disabled) {
+        virtio_scsi_dataplane_start(s);
+        return;
+    }
+    if (s->events_dropped) {
+        virtio_scsi_push_event(s, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
+    }
+}
+
+static void virtio_scsi_change(SCSIBus *bus, SCSIDevice *dev, SCSISense sense)
+{
+    VirtIOSCSI *s = container_of(bus, VirtIOSCSI, bus);
+    VirtIODevice *vdev = VIRTIO_DEVICE(s);
+
+    if (virtio_has_feature(vdev, VIRTIO_SCSI_F_CHANGE) &&
+        dev->type != TYPE_ROM) {
+        virtio_scsi_push_event(s, dev, VIRTIO_SCSI_T_PARAM_CHANGE,
+                               sense.asc | (sense.ascq << 8));
+    }
+}
+
+static void virtio_scsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev,
+                                Error **errp)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(hotplug_dev);
+    VirtIOSCSI *s = VIRTIO_SCSI(vdev);
+    SCSIDevice *sd = SCSI_DEVICE(dev);
+
+    if (s->ctx && !s->dataplane_disabled) {
+        if (blk_op_is_blocked(sd->conf.blk, BLOCK_OP_TYPE_DATAPLANE, errp)) {
+            return;
+        }
+        blk_op_block_all(sd->conf.blk, s->blocker);
+        aio_context_acquire(s->ctx);
+        blk_set_aio_context(sd->conf.blk, s->ctx);
+        aio_context_release(s->ctx);
+    }
+
+    if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG)) {
+        virtio_scsi_push_event(s, sd,
+                               VIRTIO_SCSI_T_TRANSPORT_RESET,
+                               VIRTIO_SCSI_EVT_RESET_RESCAN);
+    }
+}
+
+static void virtio_scsi_hotunplug(HotplugHandler *hotplug_dev, DeviceState *dev,
+                                  Error **errp)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(hotplug_dev);
+    VirtIOSCSI *s = VIRTIO_SCSI(vdev);
+    SCSIDevice *sd = SCSI_DEVICE(dev);
+
+    if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG)) {
+        virtio_scsi_push_event(s, sd,
+                               VIRTIO_SCSI_T_TRANSPORT_RESET,
+                               VIRTIO_SCSI_EVT_RESET_REMOVED);
+    }
+
+    if (s->ctx) {
+        blk_op_unblock_all(sd->conf.blk, s->blocker);
+    }
+    qdev_simple_device_unplug_cb(hotplug_dev, dev, errp);
+}
+
+static struct SCSIBusInfo virtio_scsi_scsi_info = {
+    .tcq = true,
+    .max_channel = VIRTIO_SCSI_MAX_CHANNEL,
+    .max_target = VIRTIO_SCSI_MAX_TARGET,
+    .max_lun = VIRTIO_SCSI_MAX_LUN,
+
+    .complete = virtio_scsi_command_complete,
+    .cancel = virtio_scsi_request_cancelled,
+    .change = virtio_scsi_change,
+    .parse_cdb = virtio_scsi_parse_cdb,
+    .get_sg_list = virtio_scsi_get_sg_list,
+    .save_request = virtio_scsi_save_request,
+    .load_request = virtio_scsi_load_request,
+};
+
+void virtio_scsi_common_realize(DeviceState *dev, Error **errp,
+                                HandleOutput ctrl, HandleOutput evt,
+                                HandleOutput cmd)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
+    VirtIOSCSICommon *s = VIRTIO_SCSI_COMMON(dev);
+    int i;
+
+    virtio_init(vdev, "virtio-scsi", VIRTIO_ID_SCSI,
+                sizeof(VirtIOSCSIConfig));
+
+    if (s->conf.num_queues == 0 ||
+            s->conf.num_queues > VIRTIO_QUEUE_MAX - 2) {
+        error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
+                         "must be a positive integer less than %d.",
+                   s->conf.num_queues, VIRTIO_QUEUE_MAX - 2);
+        virtio_cleanup(vdev);
+        return;
+    }
+    s->cmd_vqs = g_new0(VirtQueue *, s->conf.num_queues);
+    s->sense_size = VIRTIO_SCSI_SENSE_DEFAULT_SIZE;
+    s->cdb_size = VIRTIO_SCSI_CDB_DEFAULT_SIZE;
+
+    s->ctrl_vq = virtio_add_queue(vdev, VIRTIO_SCSI_VQ_SIZE,
+                                  ctrl);
+    s->event_vq = virtio_add_queue(vdev, VIRTIO_SCSI_VQ_SIZE,
+                                   evt);
+    for (i = 0; i < s->conf.num_queues; i++) {
+        s->cmd_vqs[i] = virtio_add_queue(vdev, VIRTIO_SCSI_VQ_SIZE,
+                                         cmd);
+    }
+
+    if (s->conf.iothread) {
+        virtio_scsi_set_iothread(VIRTIO_SCSI(s), s->conf.iothread);
+    }
+}
+
+/* Disable dataplane thread during live migration since it does not
+ * update the dirty memory bitmap yet.
+ */
+static void virtio_scsi_migration_state_changed(Notifier *notifier, void *data)
+{
+    VirtIOSCSI *s = container_of(notifier, VirtIOSCSI,
+                                 migration_state_notifier);
+    MigrationState *mig = data;
+
+    if (migration_in_setup(mig)) {
+        if (!s->dataplane_started) {
+            return;
+        }
+        virtio_scsi_dataplane_stop(s);
+        s->dataplane_disabled = true;
+    } else if (migration_has_finished(mig) ||
+               migration_has_failed(mig)) {
+        if (s->dataplane_started) {
+            return;
+        }
+        blk_drain_all(); /* complete in-flight non-dataplane requests */
+        s->dataplane_disabled = false;
+    }
+}
+
+static void virtio_scsi_device_realize(DeviceState *dev, Error **errp)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
+    VirtIOSCSI *s = VIRTIO_SCSI(dev);
+    static int virtio_scsi_id;
+    Error *err = NULL;
+
+    virtio_scsi_common_realize(dev, &err, virtio_scsi_handle_ctrl,
+                               virtio_scsi_handle_event,
+                               virtio_scsi_handle_cmd);
+    if (err != NULL) {
+        error_propagate(errp, err);
+        return;
+    }
+
+    scsi_bus_new(&s->bus, sizeof(s->bus), dev,
+                 &virtio_scsi_scsi_info, vdev->bus_name);
+    /* override default SCSI bus hotplug-handler, with virtio-scsi's one */
+    qbus_set_hotplug_handler(BUS(&s->bus), dev, &error_abort);
+
+    if (!dev->hotplugged) {
+        scsi_bus_legacy_handle_cmdline(&s->bus, &err);
+        if (err != NULL) {
+            error_propagate(errp, err);
+            return;
+        }
+    }
+
+    register_savevm(dev, "virtio-scsi", virtio_scsi_id++, 1,
+                    virtio_scsi_save, virtio_scsi_load, s);
+    s->migration_state_notifier.notify = virtio_scsi_migration_state_changed;
+    add_migration_state_change_notifier(&s->migration_state_notifier);
+
+    error_setg(&s->blocker, "block device is in use by data plane");
+}
+
+static void virtio_scsi_instance_init(Object *obj)
+{
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(obj);
+
+    object_property_add_link(obj, "iothread", TYPE_IOTHREAD,
+                             (Object **)&vs->conf.iothread,
+                             qdev_prop_allow_set_link_before_realize,
+                             OBJ_PROP_LINK_UNREF_ON_RELEASE, &error_abort);
+}
+
+void virtio_scsi_common_unrealize(DeviceState *dev, Error **errp)
+{
+    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
+    VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(dev);
+
+    g_free(vs->cmd_vqs);
+    virtio_cleanup(vdev);
+}
+
+static void virtio_scsi_device_unrealize(DeviceState *dev, Error **errp)
+{
+    VirtIOSCSI *s = VIRTIO_SCSI(dev);
+
+    error_free(s->blocker);
+
+    unregister_savevm(dev, "virtio-scsi", s);
+    remove_migration_state_change_notifier(&s->migration_state_notifier);
+
+    virtio_scsi_common_unrealize(dev, errp);
+}
+
+static Property virtio_scsi_properties[] = {
+    DEFINE_PROP_UINT32("num_queues", VirtIOSCSI, parent_obj.conf.num_queues, 1),
+    DEFINE_PROP_UINT32("max_sectors", VirtIOSCSI, parent_obj.conf.max_sectors,
+                                                  0xFFFF),
+    DEFINE_PROP_UINT32("cmd_per_lun", VirtIOSCSI, parent_obj.conf.cmd_per_lun,
+                                                  128),
+    DEFINE_PROP_BIT("hotplug", VirtIOSCSI, host_features,
+                                           VIRTIO_SCSI_F_HOTPLUG, true),
+    DEFINE_PROP_BIT("param_change", VirtIOSCSI, host_features,
+                                                VIRTIO_SCSI_F_CHANGE, true),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void virtio_scsi_common_class_init(ObjectClass *klass, void *data)
+{
+    VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
+    DeviceClass *dc = DEVICE_CLASS(klass);
+
+    vdc->get_config = virtio_scsi_get_config;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+}
+
+static void virtio_scsi_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
+    HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
+
+    dc->props = virtio_scsi_properties;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+    vdc->realize = virtio_scsi_device_realize;
+    vdc->unrealize = virtio_scsi_device_unrealize;
+    vdc->set_config = virtio_scsi_set_config;
+    vdc->get_features = virtio_scsi_get_features;
+    vdc->reset = virtio_scsi_reset;
+    hc->plug = virtio_scsi_hotplug;
+    hc->unplug = virtio_scsi_hotunplug;
+}
+
+static const TypeInfo virtio_scsi_common_info = {
+    .name = TYPE_VIRTIO_SCSI_COMMON,
+    .parent = TYPE_VIRTIO_DEVICE,
+    .instance_size = sizeof(VirtIOSCSICommon),
+    .abstract = true,
+    .class_init = virtio_scsi_common_class_init,
+};
+
+static const TypeInfo virtio_scsi_info = {
+    .name = TYPE_VIRTIO_SCSI,
+    .parent = TYPE_VIRTIO_SCSI_COMMON,
+    .instance_size = sizeof(VirtIOSCSI),
+    .instance_init = virtio_scsi_instance_init,
+    .class_init = virtio_scsi_class_init,
+    .interfaces = (InterfaceInfo[]) {
+        { TYPE_HOTPLUG_HANDLER },
+        { }
+    }
+};
+
+static void virtio_register_types(void)
+{
+    type_register_static(&virtio_scsi_common_info);
+    type_register_static(&virtio_scsi_info);
+}
+
+type_init(virtio_register_types)
diff --git a/qemu/hw/scsi/vmw_pvscsi.c b/qemu/hw/scsi/vmw_pvscsi.c
new file mode 100644
index 000000000..9c71f31fe
--- /dev/null
+++ b/qemu/hw/scsi/vmw_pvscsi.c
@@ -0,0 +1,1219 @@
+/*
+ * QEMU VMWARE PVSCSI paravirtual SCSI bus
+ *
+ * Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
+ *
+ * Developed by Daynix Computing LTD (http://www.daynix.com)
+ *
+ * Based on implementation by Paolo Bonzini
+ * http://lists.gnu.org/archive/html/qemu-devel/2011-08/msg00729.html
+ *
+ * Authors:
+ * Paolo Bonzini <pbonzini@redhat.com>
+ * Dmitry Fleytman <dmitry@daynix.com>
+ * Yan Vugenfirer <yan@daynix.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.
+ * See the COPYING file in the top-level directory.
+ *
+ * NOTE about MSI-X:
+ * MSI-X support has been removed for the moment because it leads Windows OS
+ * to crash on startup. The crash happens because Windows driver requires
+ * MSI-X shared memory to be part of the same BAR used for rings state
+ * registers, etc. This is not supported by QEMU infrastructure so separate
+ * BAR created from MSI-X purposes. Windows driver fails to deal with 2 BARs.
+ *
+ */
+
+#include "hw/scsi/scsi.h"
+#include <block/scsi.h>
+#include "hw/pci/msi.h"
+#include "vmw_pvscsi.h"
+#include "trace.h"
+
+
+#define PVSCSI_MSI_OFFSET        (0x50)
+#define PVSCSI_USE_64BIT         (true)
+#define PVSCSI_PER_VECTOR_MASK   (false)
+
+#define PVSCSI_MAX_DEVS                   (64)
+#define PVSCSI_MSIX_NUM_VECTORS           (1)
+
+#define PVSCSI_MAX_CMD_DATA_WORDS \
+    (sizeof(PVSCSICmdDescSetupRings)/sizeof(uint32_t))
+
+#define RS_GET_FIELD(m, field) \
+    (ldl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
+                 (m)->rs_pa + offsetof(struct PVSCSIRingsState, field)))
+#define RS_SET_FIELD(m, field, val) \
+    (stl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
+                 (m)->rs_pa + offsetof(struct PVSCSIRingsState, field), val))
+
+#define TYPE_PVSCSI "pvscsi"
+#define PVSCSI(obj) OBJECT_CHECK(PVSCSIState, (obj), TYPE_PVSCSI)
+
+typedef struct PVSCSIRingInfo {
+    uint64_t            rs_pa;
+    uint32_t            txr_len_mask;
+    uint32_t            rxr_len_mask;
+    uint32_t            msg_len_mask;
+    uint64_t            req_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
+    uint64_t            cmp_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
+    uint64_t            msg_ring_pages_pa[PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES];
+    uint64_t            consumed_ptr;
+    uint64_t            filled_cmp_ptr;
+    uint64_t            filled_msg_ptr;
+} PVSCSIRingInfo;
+
+typedef struct PVSCSISGState {
+    hwaddr elemAddr;
+    hwaddr dataAddr;
+    uint32_t resid;
+} PVSCSISGState;
+
+typedef QTAILQ_HEAD(, PVSCSIRequest) PVSCSIRequestList;
+
+typedef struct {
+    PCIDevice parent_obj;
+    MemoryRegion io_space;
+    SCSIBus bus;
+    QEMUBH *completion_worker;
+    PVSCSIRequestList pending_queue;
+    PVSCSIRequestList completion_queue;
+
+    uint64_t reg_interrupt_status;        /* Interrupt status register value */
+    uint64_t reg_interrupt_enabled;       /* Interrupt mask register value   */
+    uint64_t reg_command_status;          /* Command status register value   */
+
+    /* Command data adoption mechanism */
+    uint64_t curr_cmd;                   /* Last command arrived             */
+    uint32_t curr_cmd_data_cntr;         /* Amount of data for last command  */
+
+    /* Collector for current command data */
+    uint32_t curr_cmd_data[PVSCSI_MAX_CMD_DATA_WORDS];
+
+    uint8_t rings_info_valid;            /* Whether data rings initialized   */
+    uint8_t msg_ring_info_valid;         /* Whether message ring initialized */
+    uint8_t use_msg;                     /* Whether to use message ring      */
+
+    uint8_t msi_used;    /* Whether MSI support was installed successfully   */
+
+    PVSCSIRingInfo rings;                /* Data transfer rings manager      */
+    uint32_t resetting;                  /* Reset in progress                */
+} PVSCSIState;
+
+typedef struct PVSCSIRequest {
+    SCSIRequest *sreq;
+    PVSCSIState *dev;
+    uint8_t sense_key;
+    uint8_t completed;
+    int lun;
+    QEMUSGList sgl;
+    PVSCSISGState sg;
+    struct PVSCSIRingReqDesc req;
+    struct PVSCSIRingCmpDesc cmp;
+    QTAILQ_ENTRY(PVSCSIRequest) next;
+} PVSCSIRequest;
+
+/* Integer binary logarithm */
+static int
+pvscsi_log2(uint32_t input)
+{
+    int log = 0;
+    assert(input > 0);
+    while (input >> ++log) {
+    }
+    return log;
+}
+
+static void
+pvscsi_ring_init_data(PVSCSIRingInfo *m, PVSCSICmdDescSetupRings *ri)
+{
+    int i;
+    uint32_t txr_len_log2, rxr_len_log2;
+    uint32_t req_ring_size, cmp_ring_size;
+    m->rs_pa = ri->ringsStatePPN << VMW_PAGE_SHIFT;
+
+    req_ring_size = ri->reqRingNumPages * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
+    cmp_ring_size = ri->cmpRingNumPages * PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
+    txr_len_log2 = pvscsi_log2(req_ring_size - 1);
+    rxr_len_log2 = pvscsi_log2(cmp_ring_size - 1);
+
+    m->txr_len_mask = MASK(txr_len_log2);
+    m->rxr_len_mask = MASK(rxr_len_log2);
+
+    m->consumed_ptr = 0;
+    m->filled_cmp_ptr = 0;
+
+    for (i = 0; i < ri->reqRingNumPages; i++) {
+        m->req_ring_pages_pa[i] = ri->reqRingPPNs[i] << VMW_PAGE_SHIFT;
+    }
+
+    for (i = 0; i < ri->cmpRingNumPages; i++) {
+        m->cmp_ring_pages_pa[i] = ri->cmpRingPPNs[i] << VMW_PAGE_SHIFT;
+    }
+
+    RS_SET_FIELD(m, reqProdIdx, 0);
+    RS_SET_FIELD(m, reqConsIdx, 0);
+    RS_SET_FIELD(m, reqNumEntriesLog2, txr_len_log2);
+
+    RS_SET_FIELD(m, cmpProdIdx, 0);
+    RS_SET_FIELD(m, cmpConsIdx, 0);
+    RS_SET_FIELD(m, cmpNumEntriesLog2, rxr_len_log2);
+
+    trace_pvscsi_ring_init_data(txr_len_log2, rxr_len_log2);
+
+    /* Flush ring state page changes */
+    smp_wmb();
+}
+
+static void
+pvscsi_ring_init_msg(PVSCSIRingInfo *m, PVSCSICmdDescSetupMsgRing *ri)
+{
+    int i;
+    uint32_t len_log2;
+    uint32_t ring_size;
+
+    ring_size = ri->numPages * PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
+    len_log2 = pvscsi_log2(ring_size - 1);
+
+    m->msg_len_mask = MASK(len_log2);
+
+    m->filled_msg_ptr = 0;
+
+    for (i = 0; i < ri->numPages; i++) {
+        m->msg_ring_pages_pa[i] = ri->ringPPNs[i] << VMW_PAGE_SHIFT;
+    }
+
+    RS_SET_FIELD(m, msgProdIdx, 0);
+    RS_SET_FIELD(m, msgConsIdx, 0);
+    RS_SET_FIELD(m, msgNumEntriesLog2, len_log2);
+
+    trace_pvscsi_ring_init_msg(len_log2);
+
+    /* Flush ring state page changes */
+    smp_wmb();
+}
+
+static void
+pvscsi_ring_cleanup(PVSCSIRingInfo *mgr)
+{
+    mgr->rs_pa = 0;
+    mgr->txr_len_mask = 0;
+    mgr->rxr_len_mask = 0;
+    mgr->msg_len_mask = 0;
+    mgr->consumed_ptr = 0;
+    mgr->filled_cmp_ptr = 0;
+    mgr->filled_msg_ptr = 0;
+    memset(mgr->req_ring_pages_pa, 0, sizeof(mgr->req_ring_pages_pa));
+    memset(mgr->cmp_ring_pages_pa, 0, sizeof(mgr->cmp_ring_pages_pa));
+    memset(mgr->msg_ring_pages_pa, 0, sizeof(mgr->msg_ring_pages_pa));
+}
+
+static hwaddr
+pvscsi_ring_pop_req_descr(PVSCSIRingInfo *mgr)
+{
+    uint32_t ready_ptr = RS_GET_FIELD(mgr, reqProdIdx);
+
+    if (ready_ptr != mgr->consumed_ptr) {
+        uint32_t next_ready_ptr =
+            mgr->consumed_ptr++ & mgr->txr_len_mask;
+        uint32_t next_ready_page =
+            next_ready_ptr / PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
+        uint32_t inpage_idx =
+            next_ready_ptr % PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
+
+        return mgr->req_ring_pages_pa[next_ready_page] +
+               inpage_idx * sizeof(PVSCSIRingReqDesc);
+    } else {
+        return 0;
+    }
+}
+
+static void
+pvscsi_ring_flush_req(PVSCSIRingInfo *mgr)
+{
+    RS_SET_FIELD(mgr, reqConsIdx, mgr->consumed_ptr);
+}
+
+static hwaddr
+pvscsi_ring_pop_cmp_descr(PVSCSIRingInfo *mgr)
+{
+    /*
+     * According to Linux driver code it explicitly verifies that number
+     * of requests being processed by device is less then the size of
+     * completion queue, so device may omit completion queue overflow
+     * conditions check. We assume that this is true for other (Windows)
+     * drivers as well.
+     */
+
+    uint32_t free_cmp_ptr =
+        mgr->filled_cmp_ptr++ & mgr->rxr_len_mask;
+    uint32_t free_cmp_page =
+        free_cmp_ptr / PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
+    uint32_t inpage_idx =
+        free_cmp_ptr % PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
+    return mgr->cmp_ring_pages_pa[free_cmp_page] +
+           inpage_idx * sizeof(PVSCSIRingCmpDesc);
+}
+
+static hwaddr
+pvscsi_ring_pop_msg_descr(PVSCSIRingInfo *mgr)
+{
+    uint32_t free_msg_ptr =
+        mgr->filled_msg_ptr++ & mgr->msg_len_mask;
+    uint32_t free_msg_page =
+        free_msg_ptr / PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
+    uint32_t inpage_idx =
+        free_msg_ptr % PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
+    return mgr->msg_ring_pages_pa[free_msg_page] +
+           inpage_idx * sizeof(PVSCSIRingMsgDesc);
+}
+
+static void
+pvscsi_ring_flush_cmp(PVSCSIRingInfo *mgr)
+{
+    /* Flush descriptor changes */
+    smp_wmb();
+
+    trace_pvscsi_ring_flush_cmp(mgr->filled_cmp_ptr);
+
+    RS_SET_FIELD(mgr, cmpProdIdx, mgr->filled_cmp_ptr);
+}
+
+static bool
+pvscsi_ring_msg_has_room(PVSCSIRingInfo *mgr)
+{
+    uint32_t prodIdx = RS_GET_FIELD(mgr, msgProdIdx);
+    uint32_t consIdx = RS_GET_FIELD(mgr, msgConsIdx);
+
+    return (prodIdx - consIdx) < (mgr->msg_len_mask + 1);
+}
+
+static void
+pvscsi_ring_flush_msg(PVSCSIRingInfo *mgr)
+{
+    /* Flush descriptor changes */
+    smp_wmb();
+
+    trace_pvscsi_ring_flush_msg(mgr->filled_msg_ptr);
+
+    RS_SET_FIELD(mgr, msgProdIdx, mgr->filled_msg_ptr);
+}
+
+static void
+pvscsi_reset_state(PVSCSIState *s)
+{
+    s->curr_cmd = PVSCSI_CMD_FIRST;
+    s->curr_cmd_data_cntr = 0;
+    s->reg_command_status = PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
+    s->reg_interrupt_status = 0;
+    pvscsi_ring_cleanup(&s->rings);
+    s->rings_info_valid = FALSE;
+    s->msg_ring_info_valid = FALSE;
+    QTAILQ_INIT(&s->pending_queue);
+    QTAILQ_INIT(&s->completion_queue);
+}
+
+static void
+pvscsi_update_irq_status(PVSCSIState *s)
+{
+    PCIDevice *d = PCI_DEVICE(s);
+    bool should_raise = s->reg_interrupt_enabled & s->reg_interrupt_status;
+
+    trace_pvscsi_update_irq_level(should_raise, s->reg_interrupt_enabled,
+                                  s->reg_interrupt_status);
+
+    if (s->msi_used && msi_enabled(d)) {
+        if (should_raise) {
+            trace_pvscsi_update_irq_msi();
+            msi_notify(d, PVSCSI_VECTOR_COMPLETION);
+        }
+        return;
+    }
+
+    pci_set_irq(d, !!should_raise);
+}
+
+static void
+pvscsi_raise_completion_interrupt(PVSCSIState *s)
+{
+    s->reg_interrupt_status |= PVSCSI_INTR_CMPL_0;
+
+    /* Memory barrier to flush interrupt status register changes*/
+    smp_wmb();
+
+    pvscsi_update_irq_status(s);
+}
+
+static void
+pvscsi_raise_message_interrupt(PVSCSIState *s)
+{
+    s->reg_interrupt_status |= PVSCSI_INTR_MSG_0;
+
+    /* Memory barrier to flush interrupt status register changes*/
+    smp_wmb();
+
+    pvscsi_update_irq_status(s);
+}
+
+static void
+pvscsi_cmp_ring_put(PVSCSIState *s, struct PVSCSIRingCmpDesc *cmp_desc)
+{
+    hwaddr cmp_descr_pa;
+
+    cmp_descr_pa = pvscsi_ring_pop_cmp_descr(&s->rings);
+    trace_pvscsi_cmp_ring_put(cmp_descr_pa);
+    cpu_physical_memory_write(cmp_descr_pa, (void *)cmp_desc,
+                              sizeof(*cmp_desc));
+}
+
+static void
+pvscsi_msg_ring_put(PVSCSIState *s, struct PVSCSIRingMsgDesc *msg_desc)
+{
+    hwaddr msg_descr_pa;
+
+    msg_descr_pa = pvscsi_ring_pop_msg_descr(&s->rings);
+    trace_pvscsi_msg_ring_put(msg_descr_pa);
+    cpu_physical_memory_write(msg_descr_pa, (void *)msg_desc,
+                              sizeof(*msg_desc));
+}
+
+static void
+pvscsi_process_completion_queue(void *opaque)
+{
+    PVSCSIState *s = opaque;
+    PVSCSIRequest *pvscsi_req;
+    bool has_completed = false;
+
+    while (!QTAILQ_EMPTY(&s->completion_queue)) {
+        pvscsi_req = QTAILQ_FIRST(&s->completion_queue);
+        QTAILQ_REMOVE(&s->completion_queue, pvscsi_req, next);
+        pvscsi_cmp_ring_put(s, &pvscsi_req->cmp);
+        g_free(pvscsi_req);
+        has_completed = true;
+    }
+
+    if (has_completed) {
+        pvscsi_ring_flush_cmp(&s->rings);
+        pvscsi_raise_completion_interrupt(s);
+    }
+}
+
+static void
+pvscsi_reset_adapter(PVSCSIState *s)
+{
+    s->resetting++;
+    qbus_reset_all_fn(&s->bus);
+    s->resetting--;
+    pvscsi_process_completion_queue(s);
+    assert(QTAILQ_EMPTY(&s->pending_queue));
+    pvscsi_reset_state(s);
+}
+
+static void
+pvscsi_schedule_completion_processing(PVSCSIState *s)
+{
+    /* Try putting more complete requests on the ring. */
+    if (!QTAILQ_EMPTY(&s->completion_queue)) {
+        qemu_bh_schedule(s->completion_worker);
+    }
+}
+
+static void
+pvscsi_complete_request(PVSCSIState *s, PVSCSIRequest *r)
+{
+    assert(!r->completed);
+
+    trace_pvscsi_complete_request(r->cmp.context, r->cmp.dataLen,
+                                  r->sense_key);
+    if (r->sreq != NULL) {
+        scsi_req_unref(r->sreq);
+        r->sreq = NULL;
+    }
+    r->completed = 1;
+    QTAILQ_REMOVE(&s->pending_queue, r, next);
+    QTAILQ_INSERT_TAIL(&s->completion_queue, r, next);
+    pvscsi_schedule_completion_processing(s);
+}
+
+static QEMUSGList *pvscsi_get_sg_list(SCSIRequest *r)
+{
+    PVSCSIRequest *req = r->hba_private;
+
+    trace_pvscsi_get_sg_list(req->sgl.nsg, req->sgl.size);
+
+    return &req->sgl;
+}
+
+static void
+pvscsi_get_next_sg_elem(PVSCSISGState *sg)
+{
+    struct PVSCSISGElement elem;
+
+    cpu_physical_memory_read(sg->elemAddr, (void *)&elem, sizeof(elem));
+    if ((elem.flags & ~PVSCSI_KNOWN_FLAGS) != 0) {
+        /*
+            * There is PVSCSI_SGE_FLAG_CHAIN_ELEMENT flag described in
+            * header file but its value is unknown. This flag requires
+            * additional processing, so we put warning here to catch it
+            * some day and make proper implementation
+            */
+        trace_pvscsi_get_next_sg_elem(elem.flags);
+    }
+
+    sg->elemAddr += sizeof(elem);
+    sg->dataAddr = elem.addr;
+    sg->resid = elem.length;
+}
+
+static void
+pvscsi_write_sense(PVSCSIRequest *r, uint8_t *sense, int len)
+{
+    r->cmp.senseLen = MIN(r->req.senseLen, len);
+    r->sense_key = sense[(sense[0] & 2) ? 1 : 2];
+    cpu_physical_memory_write(r->req.senseAddr, sense, r->cmp.senseLen);
+}
+
+static void
+pvscsi_command_complete(SCSIRequest *req, uint32_t status, size_t resid)
+{
+    PVSCSIRequest *pvscsi_req = req->hba_private;
+    PVSCSIState *s;
+
+    if (!pvscsi_req) {
+        trace_pvscsi_command_complete_not_found(req->tag);
+        return;
+    }
+    s = pvscsi_req->dev;
+
+    if (resid) {
+        /* Short transfer.  */
+        trace_pvscsi_command_complete_data_run();
+        pvscsi_req->cmp.hostStatus = BTSTAT_DATARUN;
+    }
+
+    pvscsi_req->cmp.scsiStatus = status;
+    if (pvscsi_req->cmp.scsiStatus == CHECK_CONDITION) {
+        uint8_t sense[SCSI_SENSE_BUF_SIZE];
+        int sense_len =
+            scsi_req_get_sense(pvscsi_req->sreq, sense, sizeof(sense));
+
+        trace_pvscsi_command_complete_sense_len(sense_len);
+        pvscsi_write_sense(pvscsi_req, sense, sense_len);
+    }
+    qemu_sglist_destroy(&pvscsi_req->sgl);
+    pvscsi_complete_request(s, pvscsi_req);
+}
+
+static void
+pvscsi_send_msg(PVSCSIState *s, SCSIDevice *dev, uint32_t msg_type)
+{
+    if (s->msg_ring_info_valid && pvscsi_ring_msg_has_room(&s->rings)) {
+        PVSCSIMsgDescDevStatusChanged msg = {0};
+
+        msg.type = msg_type;
+        msg.bus = dev->channel;
+        msg.target = dev->id;
+        msg.lun[1] = dev->lun;
+
+        pvscsi_msg_ring_put(s, (PVSCSIRingMsgDesc *)&msg);
+        pvscsi_ring_flush_msg(&s->rings);
+        pvscsi_raise_message_interrupt(s);
+    }
+}
+
+static void
+pvscsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
+{
+    PVSCSIState *s = PVSCSI(hotplug_dev);
+
+    pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_ADDED);
+}
+
+static void
+pvscsi_hot_unplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
+{
+    PVSCSIState *s = PVSCSI(hotplug_dev);
+
+    pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_REMOVED);
+    qdev_simple_device_unplug_cb(hotplug_dev, dev, errp);
+}
+
+static void
+pvscsi_request_cancelled(SCSIRequest *req)
+{
+    PVSCSIRequest *pvscsi_req = req->hba_private;
+    PVSCSIState *s = pvscsi_req->dev;
+
+    if (pvscsi_req->completed) {
+        return;
+    }
+
+   if (pvscsi_req->dev->resetting) {
+       pvscsi_req->cmp.hostStatus = BTSTAT_BUSRESET;
+    } else {
+       pvscsi_req->cmp.hostStatus = BTSTAT_ABORTQUEUE;
+    }
+
+    pvscsi_complete_request(s, pvscsi_req);
+}
+
+static SCSIDevice*
+pvscsi_device_find(PVSCSIState *s, int channel, int target,
+                   uint8_t *requested_lun, uint8_t *target_lun)
+{
+    if (requested_lun[0] || requested_lun[2] || requested_lun[3] ||
+        requested_lun[4] || requested_lun[5] || requested_lun[6] ||
+        requested_lun[7] || (target > PVSCSI_MAX_DEVS)) {
+        return NULL;
+    } else {
+        *target_lun = requested_lun[1];
+        return scsi_device_find(&s->bus, channel, target, *target_lun);
+    }
+}
+
+static PVSCSIRequest *
+pvscsi_queue_pending_descriptor(PVSCSIState *s, SCSIDevice **d,
+                                struct PVSCSIRingReqDesc *descr)
+{
+    PVSCSIRequest *pvscsi_req;
+    uint8_t lun;
+
+    pvscsi_req = g_malloc0(sizeof(*pvscsi_req));
+    pvscsi_req->dev = s;
+    pvscsi_req->req = *descr;
+    pvscsi_req->cmp.context = pvscsi_req->req.context;
+    QTAILQ_INSERT_TAIL(&s->pending_queue, pvscsi_req, next);
+
+    *d = pvscsi_device_find(s, descr->bus, descr->target, descr->lun, &lun);
+    if (*d) {
+        pvscsi_req->lun = lun;
+    }
+
+    return pvscsi_req;
+}
+
+static void
+pvscsi_convert_sglist(PVSCSIRequest *r)
+{
+    int chunk_size;
+    uint64_t data_length = r->req.dataLen;
+    PVSCSISGState sg = r->sg;
+    while (data_length) {
+        while (!sg.resid) {
+            pvscsi_get_next_sg_elem(&sg);
+            trace_pvscsi_convert_sglist(r->req.context, r->sg.dataAddr,
+                                        r->sg.resid);
+        }
+        assert(data_length > 0);
+        chunk_size = MIN((unsigned) data_length, sg.resid);
+        if (chunk_size) {
+            qemu_sglist_add(&r->sgl, sg.dataAddr, chunk_size);
+        }
+
+        sg.dataAddr += chunk_size;
+        data_length -= chunk_size;
+        sg.resid -= chunk_size;
+    }
+}
+
+static void
+pvscsi_build_sglist(PVSCSIState *s, PVSCSIRequest *r)
+{
+    PCIDevice *d = PCI_DEVICE(s);
+
+    pci_dma_sglist_init(&r->sgl, d, 1);
+    if (r->req.flags & PVSCSI_FLAG_CMD_WITH_SG_LIST) {
+        pvscsi_convert_sglist(r);
+    } else {
+        qemu_sglist_add(&r->sgl, r->req.dataAddr, r->req.dataLen);
+    }
+}
+
+static void
+pvscsi_process_request_descriptor(PVSCSIState *s,
+                                  struct PVSCSIRingReqDesc *descr)
+{
+    SCSIDevice *d;
+    PVSCSIRequest *r = pvscsi_queue_pending_descriptor(s, &d, descr);
+    int64_t n;
+
+    trace_pvscsi_process_req_descr(descr->cdb[0], descr->context);
+
+    if (!d) {
+        r->cmp.hostStatus = BTSTAT_SELTIMEO;
+        trace_pvscsi_process_req_descr_unknown_device();
+        pvscsi_complete_request(s, r);
+        return;
+    }
+
+    if (descr->flags & PVSCSI_FLAG_CMD_WITH_SG_LIST) {
+        r->sg.elemAddr = descr->dataAddr;
+    }
+
+    r->sreq = scsi_req_new(d, descr->context, r->lun, descr->cdb, r);
+    if (r->sreq->cmd.mode == SCSI_XFER_FROM_DEV &&
+        (descr->flags & PVSCSI_FLAG_CMD_DIR_TODEVICE)) {
+        r->cmp.hostStatus = BTSTAT_BADMSG;
+        trace_pvscsi_process_req_descr_invalid_dir();
+        scsi_req_cancel(r->sreq);
+        return;
+    }
+    if (r->sreq->cmd.mode == SCSI_XFER_TO_DEV &&
+        (descr->flags & PVSCSI_FLAG_CMD_DIR_TOHOST)) {
+        r->cmp.hostStatus = BTSTAT_BADMSG;
+        trace_pvscsi_process_req_descr_invalid_dir();
+        scsi_req_cancel(r->sreq);
+        return;
+    }
+
+    pvscsi_build_sglist(s, r);
+    n = scsi_req_enqueue(r->sreq);
+
+    if (n) {
+        scsi_req_continue(r->sreq);
+    }
+}
+
+static void
+pvscsi_process_io(PVSCSIState *s)
+{
+    PVSCSIRingReqDesc descr;
+    hwaddr next_descr_pa;
+
+    assert(s->rings_info_valid);
+    while ((next_descr_pa = pvscsi_ring_pop_req_descr(&s->rings)) != 0) {
+
+        /* Only read after production index verification */
+        smp_rmb();
+
+        trace_pvscsi_process_io(next_descr_pa);
+        cpu_physical_memory_read(next_descr_pa, &descr, sizeof(descr));
+        pvscsi_process_request_descriptor(s, &descr);
+    }
+
+    pvscsi_ring_flush_req(&s->rings);
+}
+
+static void
+pvscsi_dbg_dump_tx_rings_config(PVSCSICmdDescSetupRings *rc)
+{
+    int i;
+    trace_pvscsi_tx_rings_ppn("Rings State", rc->ringsStatePPN);
+
+    trace_pvscsi_tx_rings_num_pages("Request Ring", rc->reqRingNumPages);
+    for (i = 0; i < rc->reqRingNumPages; i++) {
+        trace_pvscsi_tx_rings_ppn("Request Ring", rc->reqRingPPNs[i]);
+    }
+
+    trace_pvscsi_tx_rings_num_pages("Confirm Ring", rc->cmpRingNumPages);
+    for (i = 0; i < rc->cmpRingNumPages; i++) {
+        trace_pvscsi_tx_rings_ppn("Confirm Ring", rc->reqRingPPNs[i]);
+    }
+}
+
+static uint64_t
+pvscsi_on_cmd_config(PVSCSIState *s)
+{
+    trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_CONFIG");
+    return PVSCSI_COMMAND_PROCESSING_FAILED;
+}
+
+static uint64_t
+pvscsi_on_cmd_unplug(PVSCSIState *s)
+{
+    trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_DEVICE_UNPLUG");
+    return PVSCSI_COMMAND_PROCESSING_FAILED;
+}
+
+static uint64_t
+pvscsi_on_issue_scsi(PVSCSIState *s)
+{
+    trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_ISSUE_SCSI");
+    return PVSCSI_COMMAND_PROCESSING_FAILED;
+}
+
+static uint64_t
+pvscsi_on_cmd_setup_rings(PVSCSIState *s)
+{
+    PVSCSICmdDescSetupRings *rc =
+        (PVSCSICmdDescSetupRings *) s->curr_cmd_data;
+
+    trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_RINGS");
+
+    pvscsi_dbg_dump_tx_rings_config(rc);
+    pvscsi_ring_init_data(&s->rings, rc);
+    s->rings_info_valid = TRUE;
+    return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
+}
+
+static uint64_t
+pvscsi_on_cmd_abort(PVSCSIState *s)
+{
+    PVSCSICmdDescAbortCmd *cmd = (PVSCSICmdDescAbortCmd *) s->curr_cmd_data;
+    PVSCSIRequest *r, *next;
+
+    trace_pvscsi_on_cmd_abort(cmd->context, cmd->target);
+
+    QTAILQ_FOREACH_SAFE(r, &s->pending_queue, next, next) {
+        if (r->req.context == cmd->context) {
+            break;
+        }
+    }
+    if (r) {
+        assert(!r->completed);
+        r->cmp.hostStatus = BTSTAT_ABORTQUEUE;
+        scsi_req_cancel(r->sreq);
+    }
+
+    return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
+}
+
+static uint64_t
+pvscsi_on_cmd_unknown(PVSCSIState *s)
+{
+    trace_pvscsi_on_cmd_unknown_data(s->curr_cmd_data[0]);
+    return PVSCSI_COMMAND_PROCESSING_FAILED;
+}
+
+static uint64_t
+pvscsi_on_cmd_reset_device(PVSCSIState *s)
+{
+    uint8_t target_lun = 0;
+    struct PVSCSICmdDescResetDevice *cmd =
+        (struct PVSCSICmdDescResetDevice *) s->curr_cmd_data;
+    SCSIDevice *sdev;
+
+    sdev = pvscsi_device_find(s, 0, cmd->target, cmd->lun, &target_lun);
+
+    trace_pvscsi_on_cmd_reset_dev(cmd->target, (int) target_lun, sdev);
+
+    if (sdev != NULL) {
+        s->resetting++;
+        device_reset(&sdev->qdev);
+        s->resetting--;
+        return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
+    }
+
+    return PVSCSI_COMMAND_PROCESSING_FAILED;
+}
+
+static uint64_t
+pvscsi_on_cmd_reset_bus(PVSCSIState *s)
+{
+    trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_RESET_BUS");
+
+    s->resetting++;
+    qbus_reset_all_fn(&s->bus);
+    s->resetting--;
+    return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
+}
+
+static uint64_t
+pvscsi_on_cmd_setup_msg_ring(PVSCSIState *s)
+{
+    PVSCSICmdDescSetupMsgRing *rc =
+        (PVSCSICmdDescSetupMsgRing *) s->curr_cmd_data;
+
+    trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_MSG_RING");
+
+    if (!s->use_msg) {
+        return PVSCSI_COMMAND_PROCESSING_FAILED;
+    }
+
+    if (s->rings_info_valid) {
+        pvscsi_ring_init_msg(&s->rings, rc);
+        s->msg_ring_info_valid = TRUE;
+    }
+    return sizeof(PVSCSICmdDescSetupMsgRing) / sizeof(uint32_t);
+}
+
+static uint64_t
+pvscsi_on_cmd_adapter_reset(PVSCSIState *s)
+{
+    trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_ADAPTER_RESET");
+
+    pvscsi_reset_adapter(s);
+    return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
+}
+
+static const struct {
+    int       data_size;
+    uint64_t  (*handler_fn)(PVSCSIState *s);
+} pvscsi_commands[] = {
+    [PVSCSI_CMD_FIRST] = {
+        .data_size = 0,
+        .handler_fn = pvscsi_on_cmd_unknown,
+    },
+
+    /* Not implemented, data size defined based on what arrives on windows */
+    [PVSCSI_CMD_CONFIG] = {
+        .data_size = 6 * sizeof(uint32_t),
+        .handler_fn = pvscsi_on_cmd_config,
+    },
+
+    /* Command not implemented, data size is unknown */
+    [PVSCSI_CMD_ISSUE_SCSI] = {
+        .data_size = 0,
+        .handler_fn = pvscsi_on_issue_scsi,
+    },
+
+    /* Command not implemented, data size is unknown */
+    [PVSCSI_CMD_DEVICE_UNPLUG] = {
+        .data_size = 0,
+        .handler_fn = pvscsi_on_cmd_unplug,
+    },
+
+    [PVSCSI_CMD_SETUP_RINGS] = {
+        .data_size = sizeof(PVSCSICmdDescSetupRings),
+        .handler_fn = pvscsi_on_cmd_setup_rings,
+    },
+
+    [PVSCSI_CMD_RESET_DEVICE] = {
+        .data_size = sizeof(struct PVSCSICmdDescResetDevice),
+        .handler_fn = pvscsi_on_cmd_reset_device,
+    },
+
+    [PVSCSI_CMD_RESET_BUS] = {
+        .data_size = 0,
+        .handler_fn = pvscsi_on_cmd_reset_bus,
+    },
+
+    [PVSCSI_CMD_SETUP_MSG_RING] = {
+        .data_size = sizeof(PVSCSICmdDescSetupMsgRing),
+        .handler_fn = pvscsi_on_cmd_setup_msg_ring,
+    },
+
+    [PVSCSI_CMD_ADAPTER_RESET] = {
+        .data_size = 0,
+        .handler_fn = pvscsi_on_cmd_adapter_reset,
+    },
+
+    [PVSCSI_CMD_ABORT_CMD] = {
+        .data_size = sizeof(struct PVSCSICmdDescAbortCmd),
+        .handler_fn = pvscsi_on_cmd_abort,
+    },
+};
+
+static void
+pvscsi_do_command_processing(PVSCSIState *s)
+{
+    size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
+
+    assert(s->curr_cmd < PVSCSI_CMD_LAST);
+    if (bytes_arrived >= pvscsi_commands[s->curr_cmd].data_size) {
+        s->reg_command_status = pvscsi_commands[s->curr_cmd].handler_fn(s);
+        s->curr_cmd = PVSCSI_CMD_FIRST;
+        s->curr_cmd_data_cntr   = 0;
+    }
+}
+
+static void
+pvscsi_on_command_data(PVSCSIState *s, uint32_t value)
+{
+    size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
+
+    assert(bytes_arrived < sizeof(s->curr_cmd_data));
+    s->curr_cmd_data[s->curr_cmd_data_cntr++] = value;
+
+    pvscsi_do_command_processing(s);
+}
+
+static void
+pvscsi_on_command(PVSCSIState *s, uint64_t cmd_id)
+{
+    if ((cmd_id > PVSCSI_CMD_FIRST) && (cmd_id < PVSCSI_CMD_LAST)) {
+        s->curr_cmd = cmd_id;
+    } else {
+        s->curr_cmd = PVSCSI_CMD_FIRST;
+        trace_pvscsi_on_cmd_unknown(cmd_id);
+    }
+
+    s->curr_cmd_data_cntr = 0;
+    s->reg_command_status = PVSCSI_COMMAND_NOT_ENOUGH_DATA;
+
+    pvscsi_do_command_processing(s);
+}
+
+static void
+pvscsi_io_write(void *opaque, hwaddr addr,
+                uint64_t val, unsigned size)
+{
+    PVSCSIState *s = opaque;
+
+    switch (addr) {
+    case PVSCSI_REG_OFFSET_COMMAND:
+        pvscsi_on_command(s, val);
+        break;
+
+    case PVSCSI_REG_OFFSET_COMMAND_DATA:
+        pvscsi_on_command_data(s, (uint32_t) val);
+        break;
+
+    case PVSCSI_REG_OFFSET_INTR_STATUS:
+        trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_STATUS", val);
+        s->reg_interrupt_status &= ~val;
+        pvscsi_update_irq_status(s);
+        pvscsi_schedule_completion_processing(s);
+        break;
+
+    case PVSCSI_REG_OFFSET_INTR_MASK:
+        trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_MASK", val);
+        s->reg_interrupt_enabled = val;
+        pvscsi_update_irq_status(s);
+        break;
+
+    case PVSCSI_REG_OFFSET_KICK_NON_RW_IO:
+        trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_NON_RW_IO", val);
+        pvscsi_process_io(s);
+        break;
+
+    case PVSCSI_REG_OFFSET_KICK_RW_IO:
+        trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_RW_IO", val);
+        pvscsi_process_io(s);
+        break;
+
+    case PVSCSI_REG_OFFSET_DEBUG:
+        trace_pvscsi_io_write("PVSCSI_REG_OFFSET_DEBUG", val);
+        break;
+
+    default:
+        trace_pvscsi_io_write_unknown(addr, size, val);
+        break;
+    }
+
+}
+
+static uint64_t
+pvscsi_io_read(void *opaque, hwaddr addr, unsigned size)
+{
+    PVSCSIState *s = opaque;
+
+    switch (addr) {
+    case PVSCSI_REG_OFFSET_INTR_STATUS:
+        trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_STATUS",
+                             s->reg_interrupt_status);
+        return s->reg_interrupt_status;
+
+    case PVSCSI_REG_OFFSET_INTR_MASK:
+        trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_MASK",
+                             s->reg_interrupt_status);
+        return s->reg_interrupt_enabled;
+
+    case PVSCSI_REG_OFFSET_COMMAND_STATUS:
+        trace_pvscsi_io_read("PVSCSI_REG_OFFSET_COMMAND_STATUS",
+                             s->reg_interrupt_status);
+        return s->reg_command_status;
+
+    default:
+        trace_pvscsi_io_read_unknown(addr, size);
+        return 0;
+    }
+}
+
+
+static bool
+pvscsi_init_msi(PVSCSIState *s)
+{
+    int res;
+    PCIDevice *d = PCI_DEVICE(s);
+
+    res = msi_init(d, PVSCSI_MSI_OFFSET, PVSCSI_MSIX_NUM_VECTORS,
+                   PVSCSI_USE_64BIT, PVSCSI_PER_VECTOR_MASK);
+    if (res < 0) {
+        trace_pvscsi_init_msi_fail(res);
+        s->msi_used = false;
+    } else {
+        s->msi_used = true;
+    }
+
+    return s->msi_used;
+}
+
+static void
+pvscsi_cleanup_msi(PVSCSIState *s)
+{
+    PCIDevice *d = PCI_DEVICE(s);
+
+    if (s->msi_used) {
+        msi_uninit(d);
+    }
+}
+
+static const MemoryRegionOps pvscsi_ops = {
+        .read = pvscsi_io_read,
+        .write = pvscsi_io_write,
+        .endianness = DEVICE_LITTLE_ENDIAN,
+        .impl = {
+                .min_access_size = 4,
+                .max_access_size = 4,
+        },
+};
+
+static const struct SCSIBusInfo pvscsi_scsi_info = {
+        .tcq = true,
+        .max_target = PVSCSI_MAX_DEVS,
+        .max_channel = 0,
+        .max_lun = 0,
+
+        .get_sg_list = pvscsi_get_sg_list,
+        .complete = pvscsi_command_complete,
+        .cancel = pvscsi_request_cancelled,
+};
+
+static int
+pvscsi_init(PCIDevice *pci_dev)
+{
+    PVSCSIState *s = PVSCSI(pci_dev);
+
+    trace_pvscsi_state("init");
+
+    /* PCI subsystem ID */
+    pci_dev->config[PCI_SUBSYSTEM_ID] = 0x00;
+    pci_dev->config[PCI_SUBSYSTEM_ID + 1] = 0x10;
+
+    /* PCI latency timer = 255 */
+    pci_dev->config[PCI_LATENCY_TIMER] = 0xff;
+
+    /* Interrupt pin A */
+    pci_config_set_interrupt_pin(pci_dev->config, 1);
+
+    memory_region_init_io(&s->io_space, OBJECT(s), &pvscsi_ops, s,
+                          "pvscsi-io", PVSCSI_MEM_SPACE_SIZE);
+    pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->io_space);
+
+    pvscsi_init_msi(s);
+
+    s->completion_worker = qemu_bh_new(pvscsi_process_completion_queue, s);
+    if (!s->completion_worker) {
+        pvscsi_cleanup_msi(s);
+        return -ENOMEM;
+    }
+
+    scsi_bus_new(&s->bus, sizeof(s->bus), DEVICE(pci_dev),
+                 &pvscsi_scsi_info, NULL);
+    /* override default SCSI bus hotplug-handler, with pvscsi's one */
+    qbus_set_hotplug_handler(BUS(&s->bus), DEVICE(s), &error_abort);
+    pvscsi_reset_state(s);
+
+    return 0;
+}
+
+static void
+pvscsi_uninit(PCIDevice *pci_dev)
+{
+    PVSCSIState *s = PVSCSI(pci_dev);
+
+    trace_pvscsi_state("uninit");
+    qemu_bh_delete(s->completion_worker);
+
+    pvscsi_cleanup_msi(s);
+}
+
+static void
+pvscsi_reset(DeviceState *dev)
+{
+    PCIDevice *d = PCI_DEVICE(dev);
+    PVSCSIState *s = PVSCSI(d);
+
+    trace_pvscsi_state("reset");
+    pvscsi_reset_adapter(s);
+}
+
+static void
+pvscsi_pre_save(void *opaque)
+{
+    PVSCSIState *s = (PVSCSIState *) opaque;
+
+    trace_pvscsi_state("presave");
+
+    assert(QTAILQ_EMPTY(&s->pending_queue));
+    assert(QTAILQ_EMPTY(&s->completion_queue));
+}
+
+static int
+pvscsi_post_load(void *opaque, int version_id)
+{
+    trace_pvscsi_state("postload");
+    return 0;
+}
+
+static const VMStateDescription vmstate_pvscsi = {
+    .name = "pvscsi",
+    .version_id = 0,
+    .minimum_version_id = 0,
+    .pre_save = pvscsi_pre_save,
+    .post_load = pvscsi_post_load,
+    .fields = (VMStateField[]) {
+        VMSTATE_PCI_DEVICE(parent_obj, PVSCSIState),
+        VMSTATE_UINT8(msi_used, PVSCSIState),
+        VMSTATE_UINT32(resetting, PVSCSIState),
+        VMSTATE_UINT64(reg_interrupt_status, PVSCSIState),
+        VMSTATE_UINT64(reg_interrupt_enabled, PVSCSIState),
+        VMSTATE_UINT64(reg_command_status, PVSCSIState),
+        VMSTATE_UINT64(curr_cmd, PVSCSIState),
+        VMSTATE_UINT32(curr_cmd_data_cntr, PVSCSIState),
+        VMSTATE_UINT32_ARRAY(curr_cmd_data, PVSCSIState,
+                             ARRAY_SIZE(((PVSCSIState *)NULL)->curr_cmd_data)),
+        VMSTATE_UINT8(rings_info_valid, PVSCSIState),
+        VMSTATE_UINT8(msg_ring_info_valid, PVSCSIState),
+        VMSTATE_UINT8(use_msg, PVSCSIState),
+
+        VMSTATE_UINT64(rings.rs_pa, PVSCSIState),
+        VMSTATE_UINT32(rings.txr_len_mask, PVSCSIState),
+        VMSTATE_UINT32(rings.rxr_len_mask, PVSCSIState),
+        VMSTATE_UINT64_ARRAY(rings.req_ring_pages_pa, PVSCSIState,
+                             PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
+        VMSTATE_UINT64_ARRAY(rings.cmp_ring_pages_pa, PVSCSIState,
+                             PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
+        VMSTATE_UINT64(rings.consumed_ptr, PVSCSIState),
+        VMSTATE_UINT64(rings.filled_cmp_ptr, PVSCSIState),
+
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static Property pvscsi_properties[] = {
+    DEFINE_PROP_UINT8("use_msg", PVSCSIState, use_msg, 1),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
+static void pvscsi_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+    HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
+
+    k->init = pvscsi_init;
+    k->exit = pvscsi_uninit;
+    k->vendor_id = PCI_VENDOR_ID_VMWARE;
+    k->device_id = PCI_DEVICE_ID_VMWARE_PVSCSI;
+    k->class_id = PCI_CLASS_STORAGE_SCSI;
+    k->subsystem_id = 0x1000;
+    dc->reset = pvscsi_reset;
+    dc->vmsd = &vmstate_pvscsi;
+    dc->props = pvscsi_properties;
+    set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
+    hc->unplug = pvscsi_hot_unplug;
+    hc->plug = pvscsi_hotplug;
+}
+
+static const TypeInfo pvscsi_info = {
+    .name          = TYPE_PVSCSI,
+    .parent        = TYPE_PCI_DEVICE,
+    .instance_size = sizeof(PVSCSIState),
+    .class_init    = pvscsi_class_init,
+    .interfaces = (InterfaceInfo[]) {
+        { TYPE_HOTPLUG_HANDLER },
+        { }
+    }
+};
+
+static void
+pvscsi_register_types(void)
+{
+    type_register_static(&pvscsi_info);
+}
+
+type_init(pvscsi_register_types);
diff --git a/qemu/hw/scsi/vmw_pvscsi.h b/qemu/hw/scsi/vmw_pvscsi.h
new file mode 100644
index 000000000..17fcf6627
--- /dev/null
+++ b/qemu/hw/scsi/vmw_pvscsi.h
@@ -0,0 +1,434 @@
+/*
+ * 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: Arvind Kumar <arvindkumar@vmware.com>
+ *
+ */
+
+#ifndef VMW_PVSCSI_H
+#define VMW_PVSCSI_H
+
+#define VMW_PAGE_SIZE  (4096)
+#define VMW_PAGE_SHIFT (12)
+
+#define MASK(n)        ((1 << (n)) - 1)        /* make an n-bit mask */
+
+/*
+ * 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 */
+};
+
+#define PVSCSI_COMMAND_PROCESSING_SUCCEEDED   (0)
+#define PVSCSI_COMMAND_PROCESSING_FAILED     (-1)
+#define PVSCSI_COMMAND_NOT_ENOUGH_DATA       (-2)
+
+/*
+ * Command descriptor for PVSCSI_CMD_RESET_DEVICE --
+ */
+
+struct PVSCSICmdDescResetDevice {
+    uint32_t    target;
+    uint8_t     lun[8];
+} QEMU_PACKED;
+
+typedef struct PVSCSICmdDescResetDevice PVSCSICmdDescResetDevice;
+
+/*
+ * Command descriptor for PVSCSI_CMD_ABORT_CMD --
+ *
+ * - currently does not support specifying the LUN.
+ * - pad should be 0.
+ */
+
+struct PVSCSICmdDescAbortCmd {
+    uint64_t    context;
+    uint32_t    target;
+    uint32_t    pad;
+} QEMU_PACKED;
+
+typedef struct PVSCSICmdDescAbortCmd PVSCSICmdDescAbortCmd;
+
+/*
+ * 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 {
+    uint32_t    reqRingNumPages;
+    uint32_t    cmpRingNumPages;
+    uint64_t    ringsStatePPN;
+    uint64_t    reqRingPPNs[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
+    uint64_t    cmpRingPPNs[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
+} QEMU_PACKED;
+
+typedef struct PVSCSICmdDescSetupRings PVSCSICmdDescSetupRings;
+
+/*
+ * 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(uint32_t).
+ * - 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 {
+    uint32_t    numPages;
+    uint32_t    pad;
+    uint64_t    ringPPNs[PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES];
+} QEMU_PACKED;
+
+typedef struct PVSCSICmdDescSetupMsgRing PVSCSICmdDescSetupMsgRing;
+
+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 {
+    uint32_t    type;
+    uint32_t    args[31];
+} QEMU_PACKED;
+
+typedef struct PVSCSIRingMsgDesc PVSCSIRingMsgDesc;
+
+struct PVSCSIMsgDescDevStatusChanged {
+    uint32_t    type;  /* PVSCSI_MSG_DEV _ADDED / _REMOVED */
+    uint32_t    bus;
+    uint32_t    target;
+    uint8_t     lun[8];
+    uint32_t    pad[27];
+} QEMU_PACKED;
+
+typedef struct PVSCSIMsgDescDevStatusChanged PVSCSIMsgDescDevStatusChanged;
+
+/*
+ * 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 {
+    uint32_t    reqProdIdx;
+    uint32_t    reqConsIdx;
+    uint32_t    reqNumEntriesLog2;
+
+    uint32_t    cmpProdIdx;
+    uint32_t    cmpConsIdx;
+    uint32_t    cmpNumEntriesLog2;
+
+    uint8_t     pad[104];
+
+    uint32_t    msgProdIdx;
+    uint32_t    msgConsIdx;
+    uint32_t    msgNumEntriesLog2;
+} QEMU_PACKED;
+
+typedef struct PVSCSIRingsState PVSCSIRingsState;
+
+/*
+ * 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 uint32_t.
+ *   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)
+
+#define PVSCSI_KNOWN_FLAGS \
+  (PVSCSI_FLAG_CMD_WITH_SG_LIST     | \
+   PVSCSI_FLAG_CMD_OUT_OF_BAND_CDB  | \
+   PVSCSI_FLAG_CMD_DIR_NONE         | \
+   PVSCSI_FLAG_CMD_DIR_TOHOST       | \
+   PVSCSI_FLAG_CMD_DIR_TODEVICE)
+
+struct PVSCSIRingReqDesc {
+    uint64_t    context;
+    uint64_t    dataAddr;
+    uint64_t    dataLen;
+    uint64_t    senseAddr;
+    uint32_t    senseLen;
+    uint32_t    flags;
+    uint8_t     cdb[16];
+    uint8_t     cdbLen;
+    uint8_t     lun[8];
+    uint8_t     tag;
+    uint8_t     bus;
+    uint8_t     target;
+    uint8_t     vcpuHint;
+    uint8_t     unused[59];
+} QEMU_PACKED;
+
+typedef struct PVSCSIRingReqDesc PVSCSIRingReqDesc;
+
+/*
+ * 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 {
+    uint64_t    addr;
+    uint32_t    length;
+    uint32_t    flags;
+} QEMU_PACKED;
+
+typedef struct PVSCSISGElement PVSCSISGElement;
+
+/*
+ * 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 {
+    uint64_t    context;
+    uint64_t    dataLen;
+    uint32_t    senseLen;
+    uint16_t    hostStatus;
+    uint16_t    scsiStatus;
+    uint32_t    pad[2];
+} QEMU_PACKED;
+
+typedef struct PVSCSIRingCmpDesc PVSCSIRingCmpDesc;
+
+/*
+ * 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 \
+                (VMW_PAGE_SIZE / sizeof(struct PVSCSIRingReqDesc))
+
+#define PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE \
+                (VMW_PAGE_SIZE / sizeof(PVSCSIRingCmpDesc))
+
+#define PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE \
+                (VMW_PAGE_SIZE / sizeof(PVSCSIRingMsgDesc))
+
+#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 * VMW_PAGE_SIZE)
+
+#endif /* VMW_PVSCSI_H */