[Qemu-devel] [PATCH V9 8/8] Introduce Xen PCI Passthrough, MSI (3/3)

[Anthony PERARD <anthony.perard@citrix.com>]

From: Jiang Yunhong <yunhong.jiang@intel.com>

A more complete history can be found here:
git://xenbits.xensource.com/qemu-xen-unstable.git

Signed-off-by: Jiang Yunhong <yunhong.jiang@intel.com>
Signed-off-by: Shan Haitao <haitao.shan@intel.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
---
 Makefile.target                      |    1 +
 hw/xen_pci_passthrough.c             |   31 ++-
 hw/xen_pci_passthrough.h             |   51 +++
 hw/xen_pci_passthrough_config_init.c |  471 ++++++++++++++++++++++++++
 hw/xen_pci_passthrough_msi.c         |  615 ++++++++++++++++++++++++++++++++++
 5 files changed, 1168 insertions(+), 1 deletions(-)
 create mode 100644 hw/xen_pci_passthrough_msi.c

diff --git a/Makefile.target b/Makefile.target
index 1869f0d..76be50c 100644
--- a/Makefile.target
+++ b/Makefile.target
@@ -236,6 +236,7 @@ obj-i386-$(CONFIG_XEN) += xen_platform.o
 obj-i386-$(CONFIG_XEN_PCI_PASSTHROUGH) += host-pci-device.o
 obj-i386-$(CONFIG_XEN_PCI_PASSTHROUGH) += xen_pci_passthrough.o
 obj-i386-$(CONFIG_XEN_PCI_PASSTHROUGH) += xen_pci_passthrough_config_init.o
+obj-i386-$(CONFIG_XEN_PCI_PASSTHROUGH) += xen_pci_passthrough_msi.o
 
 # Inter-VM PCI shared memory
 CONFIG_IVSHMEM =
diff --git a/hw/xen_pci_passthrough.c b/hw/xen_pci_passthrough.c
index 7380bf5..03b736d 100644
--- a/hw/xen_pci_passthrough.c
+++ b/hw/xen_pci_passthrough.c
@@ -36,6 +36,20 @@
  *
  *     Write '1'
  *       - Set real bit to '1'.
+ *
+ * MSI interrupt:
+ *   Initialize MSI register(pt_msi_setup, pt_msi_update)
+ *     Bind MSI(xc_domain_update_msi_irq)
+ *       <fail>
+ *         - Unmap MSI.
+ *         - Set dev->msi->pirq to '-1'.
+ *
+ * MSI-X interrupt:
+ *   Initialize MSI-X register(pt_msix_update_one)
+ *     Bind MSI-X(xc_domain_update_msi_irq)
+ *       <fail>
+ *         - Unmap MSI-X.
+ *         - Set entry->pirq to '-1'.
  */
 
 #include <sys/ioctl.h>
@@ -529,7 +543,15 @@ static void pt_region_update(XenPCIPassthroughState *s,
     };
 
     bar = pt_bar_from_region(s, mr);
-    if (bar == -1) {
+    if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) {
+        return;
+    }
+
+    if (s->msix && &s->msix->mmio == mr) {
+        if (adding) {
+            s->msix->mmio_base_addr = sec->offset_within_address_space;
+            rc = pt_msix_update_remap(s, s->msix->bar_index);
+        }
         return;
     }
 
@@ -760,6 +782,13 @@ static int pt_unregister_device(PCIDevice *d)
         }
     }
 
+    if (s->msi) {
+        pt_msi_disable(s);
+    }
+    if (s->msix) {
+        pt_msix_disable(s);
+    }
+
     if (machine_irq) {
         mapped_machine_irq[machine_irq]--;
 
diff --git a/hw/xen_pci_passthrough.h b/hw/xen_pci_passthrough.h
index 6a4d416..7bd38df 100644
--- a/hw/xen_pci_passthrough.h
+++ b/hw/xen_pci_passthrough.h
@@ -162,6 +162,36 @@ typedef struct XenPTRegGroup {
 
 
 #define PT_UNASSIGNED_PIRQ (-1)
+typedef struct XenPTMSI {
+    uint16_t flags;
+    uint32_t addr_lo;  /* guest message address */
+    uint32_t addr_hi;  /* guest message upper address */
+    uint16_t data;     /* guest message data */
+    uint32_t ctrl_offset; /* saved control offset */
+    int pirq;          /* guest pirq corresponding */
+    bool initialized;  /* when guest MSI is initialized */
+    bool mapped;       /* when pirq is mapped */
+} XenPTMSI;
+
+typedef struct XenPTMSIXEntry {
+    int pirq;
+    uint64_t addr;
+    uint32_t data;
+    uint32_t vector_ctrl;
+    bool updated; /* indicate whether MSI ADDR or DATA is updated */
+} XenPTMSIXEntry;
+typedef struct XenPTMSIX {
+    uint32_t ctrl_offset;
+    bool enabled;
+    int total_entries;
+    int bar_index;
+    uint64_t table_base;
+    uint32_t table_offset_adjust; /* page align mmap */
+    uint64_t mmio_base_addr;
+    MemoryRegion mmio;
+    void *phys_iomem_base;
+    XenPTMSIXEntry msix_entry[0];
+} XenPTMSIX;
 
 struct XenPCIPassthroughState {
     PCIDevice dev;
@@ -174,6 +204,9 @@ struct XenPCIPassthroughState {
 
     uint32_t machine_irq;
 
+    XenPTMSI *msi;
+    XenPTMSIX *msix;
+
     MemoryRegion bar[PCI_NUM_REGIONS - 1];
     MemoryRegion rom;
 
@@ -249,4 +282,22 @@ static inline uint8_t pci_intx(XenPCIPassthroughState *s)
     return r_val;
 }
 
+/* MSI/MSI-X */
+int pt_msi_set_enable(XenPCIPassthroughState *s, bool en);
+int pt_msi_setup(XenPCIPassthroughState *s);
+int pt_msi_update(XenPCIPassthroughState *d);
+void pt_msi_disable(XenPCIPassthroughState *s);
+
+int pt_msix_init(XenPCIPassthroughState *s, uint32_t base);
+void pt_msix_delete(XenPCIPassthroughState *s);
+int pt_msix_update(XenPCIPassthroughState *s);
+int pt_msix_update_remap(XenPCIPassthroughState *s, int bar_index);
+void pt_msix_disable(XenPCIPassthroughState *s);
+
+static inline bool pt_has_msix_mapping(XenPCIPassthroughState *s, int bar)
+{
+    return s->msix && s->msix->bar_index == bar;
+}
+
+
 #endif /* !QEMU_HW_XEN_PCI_PASSTHROUGH_H */
diff --git a/hw/xen_pci_passthrough_config_init.c b/hw/xen_pci_passthrough_config_init.c
index c897965..bf98e8c 100644
--- a/hw/xen_pci_passthrough_config_init.c
+++ b/hw/xen_pci_passthrough_config_init.c
@@ -1020,6 +1020,410 @@ static XenPTRegInfo pt_emu_reg_pm_tbl[] = {
 };
 
 
+/********************************
+ * MSI Capability
+ */
+
+/* Helper */
+static bool pt_msgdata_check_type(uint32_t offset, uint16_t flags)
+{
+    /* check the offset whether matches the type or not */
+    bool is_32 = (offset == PCI_MSI_DATA_32) && !(flags & PCI_MSI_FLAGS_64BIT);
+    bool is_64 = (offset == PCI_MSI_DATA_64) &&  (flags & PCI_MSI_FLAGS_64BIT);
+    return is_32 || is_64;
+}
+
+/* Message Control register */
+static int pt_msgctrl_reg_init(XenPCIPassthroughState *s,
+                               XenPTRegInfo *reg, uint32_t real_offset,
+                               uint32_t *data)
+{
+    PCIDevice *d = &s->dev;
+    XenPTMSI *msi = s->msi;
+    uint16_t reg_field = 0;
+
+    /* use I/O device register's value as initial value */
+    reg_field = pci_get_word(d->config + real_offset);
+
+    if (reg_field & PCI_MSI_FLAGS_ENABLE) {
+        PT_LOG(&s->dev, "MSI already enabled, disabling it first\n");
+        host_pci_set_word(s->real_device, real_offset,
+                          reg_field & ~PCI_MSI_FLAGS_ENABLE);
+    }
+    msi->flags |= reg_field;
+    msi->ctrl_offset = real_offset;
+    msi->initialized = false;
+    msi->mapped = false;
+
+    *data = reg->init_val;
+    return 0;
+}
+static int pt_msgctrl_reg_write(XenPCIPassthroughState *s, XenPTReg *cfg_entry,
+                                uint16_t *value, uint16_t dev_value,
+                                uint16_t valid_mask)
+{
+    XenPTRegInfo *reg = cfg_entry->reg;
+    XenPTMSI *msi = s->msi;
+    uint16_t writable_mask = 0;
+    uint16_t throughable_mask = 0;
+    uint16_t val;
+
+    /* Currently no support for multi-vector */
+    if (*value & PCI_MSI_FLAGS_QSIZE) {
+        PT_WARN(&s->dev, "Tries to set more than 1 vector ctrl %x\n", *value);
+    }
+
+    /* modify emulate register */
+    writable_mask = reg->emu_mask & ~reg->ro_mask & valid_mask;
+    cfg_entry->data = PT_MERGE_VALUE(*value, cfg_entry->data, writable_mask);
+    msi->flags |= cfg_entry->data & ~PCI_MSI_FLAGS_ENABLE;
+
+    /* create value for writing to I/O device register */
+    val = *value;
+    throughable_mask = ~reg->emu_mask & valid_mask;
+    *value = PT_MERGE_VALUE(*value, dev_value, throughable_mask);
+
+    /* update MSI */
+    if (val & PCI_MSI_FLAGS_ENABLE) {
+        /* setup MSI pirq for the first time */
+        if (!msi->initialized) {
+            /* Init physical one */
+            PT_LOG(&s->dev, "setup MSI\n");
+            if (pt_msi_setup(s)) {
+                /* We do not broadcast the error to the framework code, so
+                 * that MSI errors are contained in MSI emulation code and
+                 * QEMU can go on running.
+                 * Guest MSI would be actually not working.
+                 */
+                *value &= ~PCI_MSI_FLAGS_ENABLE;
+                PT_WARN(&s->dev, "Can not map MSI.\n");
+                return 0;
+            }
+            if (pt_msi_update(s)) {
+                *value &= ~PCI_MSI_FLAGS_ENABLE;
+                PT_WARN(&s->dev, "Can not bind MSI\n");
+                return 0;
+            }
+            msi->initialized = true;
+            msi->mapped = true;
+        }
+        msi->flags |= PCI_MSI_FLAGS_ENABLE;
+    } else {
+        msi->flags &= ~PCI_MSI_FLAGS_ENABLE;
+    }
+
+    /* pass through MSI_ENABLE bit */
+    *value &= ~PCI_MSI_FLAGS_ENABLE;
+    *value |= val & PCI_MSI_FLAGS_ENABLE;
+
+    return 0;
+}
+
+/* initialize Message Upper Address register */
+static int pt_msgaddr64_reg_init(XenPCIPassthroughState *s,
+                                 XenPTRegInfo *reg, uint32_t real_offset,
+                                 uint32_t *data)
+{
+    /* no need to initialize in case of 32 bit type */
+    if (!(s->msi->flags & PCI_MSI_FLAGS_64BIT)) {
+        *data = PT_INVALID_REG;
+    } else {
+        *data = reg->init_val;
+    }
+
+    return 0;
+}
+/* this function will be called twice (for 32 bit and 64 bit type) */
+/* initialize Message Data register */
+static int pt_msgdata_reg_init(XenPCIPassthroughState *s,
+                               XenPTRegInfo *reg, uint32_t real_offset,
+                               uint32_t *data)
+{
+    uint32_t flags = s->msi->flags;
+    uint32_t offset = reg->offset;
+
+    /* check the offset whether matches the type or not */
+    if (pt_msgdata_check_type(offset, flags)) {
+        *data = reg->init_val;
+    } else {
+        *data = PT_INVALID_REG;
+    }
+    return 0;
+}
+
+/* write Message Address register */
+static int pt_msgaddr32_reg_write(XenPCIPassthroughState *s,
+                                  XenPTReg *cfg_entry, uint32_t *value,
+                                  uint32_t dev_value, uint32_t valid_mask)
+{
+    XenPTRegInfo *reg = cfg_entry->reg;
+    uint32_t writable_mask = 0;
+    uint32_t throughable_mask = 0;
+    uint32_t old_addr = cfg_entry->data;
+
+    /* modify emulate register */
+    writable_mask = reg->emu_mask & ~reg->ro_mask & valid_mask;
+    cfg_entry->data = PT_MERGE_VALUE(*value, cfg_entry->data, writable_mask);
+    s->msi->addr_lo = cfg_entry->data;
+
+    /* create value for writing to I/O device register */
+    throughable_mask = ~reg->emu_mask & valid_mask;
+    *value = PT_MERGE_VALUE(*value, dev_value, throughable_mask);
+
+    /* update MSI */
+    if (cfg_entry->data != old_addr) {
+        if (s->msi->mapped) {
+            pt_msi_update(s);
+        }
+    }
+
+    return 0;
+}
+/* write Message Upper Address register */
+static int pt_msgaddr64_reg_write(XenPCIPassthroughState *s,
+                                  XenPTReg *cfg_entry, uint32_t *value,
+                                  uint32_t dev_value, uint32_t valid_mask)
+{
+    XenPTRegInfo *reg = cfg_entry->reg;
+    uint32_t writable_mask = 0;
+    uint32_t throughable_mask = 0;
+    uint32_t old_addr = cfg_entry->data;
+
+    /* check whether the type is 64 bit or not */
+    if (!(s->msi->flags & PCI_MSI_FLAGS_64BIT)) {
+        PT_ERR(&s->dev,
+               "Can't write to the upper address without 64 bit support\n");
+        return -1;
+    }
+
+    /* modify emulate register */
+    writable_mask = reg->emu_mask & ~reg->ro_mask & valid_mask;
+    cfg_entry->data = PT_MERGE_VALUE(*value, cfg_entry->data, writable_mask);
+    /* update the msi_info too */
+    s->msi->addr_hi = cfg_entry->data;
+
+    /* create value for writing to I/O device register */
+    throughable_mask = ~reg->emu_mask & valid_mask;
+    *value = PT_MERGE_VALUE(*value, dev_value, throughable_mask);
+
+    /* update MSI */
+    if (cfg_entry->data != old_addr) {
+        if (s->msi->mapped) {
+            pt_msi_update(s);
+        }
+    }
+
+    return 0;
+}
+
+
+/* this function will be called twice (for 32 bit and 64 bit type) */
+/* write Message Data register */
+static int pt_msgdata_reg_write(XenPCIPassthroughState *s, XenPTReg *cfg_entry,
+                                uint16_t *value, uint16_t dev_value,
+                                uint16_t valid_mask)
+{
+    XenPTRegInfo *reg = cfg_entry->reg;
+    XenPTMSI *msi = s->msi;
+    uint16_t writable_mask = 0;
+    uint16_t throughable_mask = 0;
+    uint16_t old_data = cfg_entry->data;
+    uint32_t offset = reg->offset;
+
+    /* check the offset whether matches the type or not */
+    if (!pt_msgdata_check_type(offset, msi->flags)) {
+        /* exit I/O emulator */
+        PT_ERR(&s->dev, "the offset does not match the 32/64 bit type!\n");
+        return -1;
+    }
+
+    /* modify emulate register */
+    writable_mask = reg->emu_mask & ~reg->ro_mask & valid_mask;
+    cfg_entry->data = PT_MERGE_VALUE(*value, cfg_entry->data, writable_mask);
+    /* update the msi_info too */
+    msi->data = cfg_entry->data;
+
+    /* create value for writing to I/O device register */
+    throughable_mask = ~reg->emu_mask & valid_mask;
+    *value = PT_MERGE_VALUE(*value, dev_value, throughable_mask);
+
+    /* update MSI */
+    if (cfg_entry->data != old_data) {
+        if (msi->mapped) {
+            pt_msi_update(s);
+        }
+    }
+
+    return 0;
+}
+
+/* MSI Capability Structure reg static infomation table */
+static XenPTRegInfo pt_emu_reg_msi_tbl[] = {
+    /* Next Pointer reg */
+    {
+        .offset     = PCI_CAP_LIST_NEXT,
+        .size       = 1,
+        .init_val   = 0x00,
+        .ro_mask    = 0xFF,
+        .emu_mask   = 0xFF,
+        .init       = pt_ptr_reg_init,
+        .u.b.read   = pt_byte_reg_read,
+        .u.b.write  = pt_byte_reg_write,
+    },
+    /* Message Control reg */
+    {
+        .offset     = PCI_MSI_FLAGS,
+        .size       = 2,
+        .init_val   = 0x0000,
+        .ro_mask    = 0xFF8E,
+        .emu_mask   = 0x007F,
+        .init       = pt_msgctrl_reg_init,
+        .u.w.read   = pt_word_reg_read,
+        .u.w.write  = pt_msgctrl_reg_write,
+    },
+    /* Message Address reg */
+    {
+        .offset     = PCI_MSI_ADDRESS_LO,
+        .size       = 4,
+        .init_val   = 0x00000000,
+        .ro_mask    = 0x00000003,
+        .emu_mask   = 0xFFFFFFFF,
+        .no_wb      = 1,
+        .init       = pt_common_reg_init,
+        .u.dw.read  = pt_long_reg_read,
+        .u.dw.write = pt_msgaddr32_reg_write,
+    },
+    /* Message Upper Address reg (if PCI_MSI_FLAGS_64BIT set) */
+    {
+        .offset     = PCI_MSI_ADDRESS_HI,
+        .size       = 4,
+        .init_val   = 0x00000000,
+        .ro_mask    = 0x00000000,
+        .emu_mask   = 0xFFFFFFFF,
+        .no_wb      = 1,
+        .init       = pt_msgaddr64_reg_init,
+        .u.dw.read  = pt_long_reg_read,
+        .u.dw.write = pt_msgaddr64_reg_write,
+    },
+    /* Message Data reg (16 bits of data for 32-bit devices) */
+    {
+        .offset     = PCI_MSI_DATA_32,
+        .size       = 2,
+        .init_val   = 0x0000,
+        .ro_mask    = 0x0000,
+        .emu_mask   = 0xFFFF,
+        .no_wb      = 1,
+        .init       = pt_msgdata_reg_init,
+        .u.w.read   = pt_word_reg_read,
+        .u.w.write  = pt_msgdata_reg_write,
+    },
+    /* Message Data reg (16 bits of data for 64-bit devices) */
+    {
+        .offset     = PCI_MSI_DATA_64,
+        .size       = 2,
+        .init_val   = 0x0000,
+        .ro_mask    = 0x0000,
+        .emu_mask   = 0xFFFF,
+        .no_wb      = 1,
+        .init       = pt_msgdata_reg_init,
+        .u.w.read   = pt_word_reg_read,
+        .u.w.write  = pt_msgdata_reg_write,
+    },
+    {
+        .size = 0,
+    },
+};
+
+
+/**************************************
+ * MSI-X Capability
+ */
+
+/* Message Control register for MSI-X */
+static int pt_msixctrl_reg_init(XenPCIPassthroughState *s,
+                                XenPTRegInfo *reg, uint32_t real_offset,
+                                uint32_t *data)
+{
+    PCIDevice *d = &s->dev;
+    uint16_t reg_field = 0;
+
+    /* use I/O device register's value as initial value */
+    reg_field = pci_get_word(d->config + real_offset);
+
+    if (reg_field & PCI_MSIX_FLAGS_ENABLE) {
+        PT_LOG(d, "MSIX already enabled, disabling it first\n");
+        host_pci_set_word(s->real_device, real_offset,
+                          reg_field & ~PCI_MSIX_FLAGS_ENABLE);
+    }
+
+    s->msix->ctrl_offset = real_offset;
+
+    *data = reg->init_val;
+    return 0;
+}
+static int pt_msixctrl_reg_write(XenPCIPassthroughState *s,
+                                 XenPTReg *cfg_entry, uint16_t *value,
+                                 uint16_t dev_value, uint16_t valid_mask)
+{
+    XenPTRegInfo *reg = cfg_entry->reg;
+    uint16_t writable_mask = 0;
+    uint16_t throughable_mask = 0;
+    int debug_msix_enabled_old;
+
+    /* modify emulate register */
+    writable_mask = reg->emu_mask & ~reg->ro_mask & valid_mask;
+    cfg_entry->data = PT_MERGE_VALUE(*value, cfg_entry->data, writable_mask);
+
+    /* create value for writing to I/O device register */
+    throughable_mask = ~reg->emu_mask & valid_mask;
+    *value = PT_MERGE_VALUE(*value, dev_value, throughable_mask);
+
+    /* update MSI-X */
+    if ((*value & PCI_MSIX_FLAGS_ENABLE)
+        && !(*value & PCI_MSIX_FLAGS_MASKALL)) {
+        pt_msix_update(s);
+    }
+
+    debug_msix_enabled_old = s->msix->enabled;
+    s->msix->enabled = !!(*value & PCI_MSIX_FLAGS_ENABLE);
+    if (s->msix->enabled != debug_msix_enabled_old) {
+        PT_LOG(&s->dev, "%s MSI-X\n",
+               s->msix->enabled ? "enable" : "disable");
+    }
+
+    return 0;
+}
+
+/* MSI-X Capability Structure reg static infomation table */
+static XenPTRegInfo pt_emu_reg_msix_tbl[] = {
+    /* Next Pointer reg */
+    {
+        .offset     = PCI_CAP_LIST_NEXT,
+        .size       = 1,
+        .init_val   = 0x00,
+        .ro_mask    = 0xFF,
+        .emu_mask   = 0xFF,
+        .init       = pt_ptr_reg_init,
+        .u.b.read   = pt_byte_reg_read,
+        .u.b.write  = pt_byte_reg_write,
+    },
+    /* Message Control reg */
+    {
+        .offset     = PCI_MSI_FLAGS,
+        .size       = 2,
+        .init_val   = 0x0000,
+        .ro_mask    = 0x3FFF,
+        .emu_mask   = 0x0000,
+        .init       = pt_msixctrl_reg_init,
+        .u.w.read   = pt_word_reg_read,
+        .u.w.write  = pt_msixctrl_reg_write,
+    },
+    {
+        .size = 0,
+    },
+};
+
+
 /****************************
  * Capabilities
  */
@@ -1113,6 +1517,49 @@ static int pt_pcie_size_init(XenPCIPassthroughState *s,
     *size = pcie_size;
     return 0;
 }
+/* get MSI Capability Structure register group size */
+static int pt_msi_size_init(XenPCIPassthroughState *s,
+                            const XenPTRegGroupInfo *grp_reg,
+                            uint32_t base_offset, uint8_t *size)
+{
+    PCIDevice *d = &s->dev;
+    uint16_t msg_ctrl = 0;
+    uint8_t msi_size = 0xa;
+
+    msg_ctrl = pci_get_word(d->config + (base_offset + PCI_MSI_FLAGS));
+
+    /* check if 64-bit address is capable of per-vector masking */
+    if (msg_ctrl & PCI_MSI_FLAGS_64BIT) {
+        msi_size += 4;
+    }
+    if (msg_ctrl & PCI_MSI_FLAGS_MASKBIT) {
+        msi_size += 10;
+    }
+
+    s->msi = g_new0(XenPTMSI, 1);
+    s->msi->pirq = PT_UNASSIGNED_PIRQ;
+
+    *size = msi_size;
+    return 0;
+}
+/* get MSI-X Capability Structure register group size */
+static int pt_msix_size_init(XenPCIPassthroughState *s,
+                             const XenPTRegGroupInfo *grp_reg,
+                             uint32_t base_offset, uint8_t *size)
+{
+    int rc = 0;
+
+    rc = pt_msix_init(s, base_offset);
+
+    if (rc < 0) {
+        PT_ERR(&s->dev, "Internal error: Invalid pt_msix_init.\n");
+        return rc;
+    }
+
+    *size = grp_reg->grp_size;
+    return 0;
+}
+
 
 static const XenPTRegGroupInfo pt_emu_reg_grp_tbl[] = {
     /* Header Type0 reg group */
@@ -1153,6 +1600,14 @@ static const XenPTRegGroupInfo pt_emu_reg_grp_tbl[] = {
         .grp_size   = 0x04,
         .size_init  = pt_reg_grp_size_init,
     },
+    /* MSI Capability Structure reg group */
+    {
+        .grp_id      = PCI_CAP_ID_MSI,
+        .grp_type    = GRP_TYPE_EMU,
+        .grp_size    = 0xFF,
+        .size_init   = pt_msi_size_init,
+        .emu_reg_tbl = pt_emu_reg_msi_tbl,
+    },
     /* PCI-X Capabilities List Item reg group */
     {
         .grp_id     = PCI_CAP_ID_PCIX,
@@ -1197,6 +1652,14 @@ static const XenPTRegGroupInfo pt_emu_reg_grp_tbl[] = {
         .size_init   = pt_pcie_size_init,
         .emu_reg_tbl = pt_emu_reg_pcie_tbl,
     },
+    /* MSI-X Capability Structure reg group */
+    {
+        .grp_id      = PCI_CAP_ID_MSIX,
+        .grp_type    = GRP_TYPE_EMU,
+        .grp_size    = 0x0C,
+        .size_init   = pt_msix_size_init,
+        .emu_reg_tbl = pt_emu_reg_msix_tbl,
+    },
     {
         .grp_size = 0,
     },
@@ -1381,6 +1844,14 @@ void pt_config_delete(XenPCIPassthroughState *s)
     struct XenPTRegGroup *reg_group, *next_grp;
     struct XenPTReg *reg, *next_reg;
 
+    /* free MSI/MSI-X info table */
+    if (s->msix) {
+        pt_msix_delete(s);
+    }
+    if (s->msi) {
+        g_free(s->msi);
+    }
+
     /* free all register group entry */
     QLIST_FOREACH_SAFE(reg_group, &s->reg_grp_tbl, entries, next_grp) {
         /* free all register entry */
diff --git a/hw/xen_pci_passthrough_msi.c b/hw/xen_pci_passthrough_msi.c
new file mode 100644
index 0000000..dc62493
--- /dev/null
+++ b/hw/xen_pci_passthrough_msi.c
@@ -0,0 +1,615 @@
+/*
+ * Copyright (c) 2007, Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ * Jiang Yunhong <yunhong.jiang@intel.com>
+ *
+ * This file implements direct PCI assignment to a HVM guest
+ */
+
+#include <sys/mman.h>
+
+#include "xen_backend.h"
+#include "xen_pci_passthrough.h"
+#include "apic-msidef.h"
+
+
+#define AUTO_ASSIGN -1
+
+/* shift count for gflags */
+#define GFLAGS_SHIFT_DEST_ID        0
+#define GFLAGS_SHIFT_RH             8
+#define GFLAGS_SHIFT_DM             9
+#define GLFAGS_SHIFT_DELIV_MODE     12
+#define GLFAGS_SHIFT_TRG_MODE       15
+
+
+/*
+ * Helpers
+ */
+
+static inline uint8_t msi_vector(uint32_t data)
+{
+    return (data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT;
+}
+
+static inline uint8_t msi_dest_id(uint32_t addr)
+{
+    return (addr & MSI_ADDR_DEST_ID_MASK) >> MSI_ADDR_DEST_ID_SHIFT;
+}
+
+static inline uint32_t msi_ext_dest_id(uint32_t addr_hi)
+{
+    return addr_hi & 0xffffff00;
+}
+
+static uint32_t msi_gflags(uint32_t data, uint64_t addr)
+{
+    uint32_t result = 0;
+    int rh, dm, dest_id, deliv_mode, trig_mode;
+
+    rh = (addr >> MSI_ADDR_REDIRECTION_SHIFT) & 0x1;
+    dm = (addr >> MSI_ADDR_DEST_MODE_SHIFT) & 0x1;
+    dest_id = msi_dest_id(addr);
+    deliv_mode = (data >> MSI_DATA_DELIVERY_MODE_SHIFT) & 0x7;
+    trig_mode = (data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
+
+    result = dest_id | (rh << GFLAGS_SHIFT_RH) | (dm << GFLAGS_SHIFT_DM) |
+        (deliv_mode << GLFAGS_SHIFT_DELIV_MODE) |
+        (trig_mode << GLFAGS_SHIFT_TRG_MODE);
+
+    return result;
+}
+
+static inline uint64_t msi_addr64(XenPTMSI *msi)
+{
+    return (uint64_t)msi->addr_hi << 32 | msi->addr_lo;
+}
+
+static int msi_msix_enable(XenPCIPassthroughState *s,
+                           uint32_t address,
+                           uint16_t flag,
+                           bool enable)
+{
+    uint16_t val = 0;
+
+    if (!address) {
+        return -1;
+    }
+
+    host_pci_get_word(s->real_device, address, &val);
+    if (enable) {
+        val |= flag;
+    } else {
+        val &= ~flag;
+    }
+    host_pci_set_word(s->real_device, address, val);
+    return 0;
+}
+
+static int msi_msix_setup(XenPCIPassthroughState *s,
+                           uint64_t addr,
+                           uint32_t data,
+                           int *ppirq,
+                           bool is_msix,
+                           int msix_entry,
+                           bool is_not_mapped)
+{
+    uint8_t gvec = msi_vector(data);
+    int rc = 0;
+
+    assert((!is_msix && msix_entry == 0) || is_msix);
+
+    if (gvec == 0) {
+        /* if gvec is 0, the guest is asking for a particular pirq that
+         * is passed as dest_id */
+        *ppirq = msi_ext_dest_id(addr >> 32) | msi_dest_id(addr);
+        if (!*ppirq) {
+            /* this probably identifies an misconfiguration of the guest,
+             * try the emulated path */
+            *ppirq = PT_UNASSIGNED_PIRQ;
+        } else {
+            PT_LOG(&s->dev, "requested pirq %d for MSI%s"
+                   " (vec: %#x, entry: %#x)\n",
+                   *ppirq, is_msix ? "-X" : "", gvec, msix_entry);
+        }
+    }
+
+    if (is_not_mapped) {
+        uint64_t table_base = 0;
+
+        if (is_msix) {
+            table_base = s->msix->table_base;
+        }
+
+        rc = xc_physdev_map_pirq_msi(xen_xc, xen_domid, AUTO_ASSIGN, ppirq,
+                                     PCI_DEVFN(s->real_device->dev,
+                                               s->real_device->func),
+                                     s->real_device->bus,
+                                     msix_entry, table_base);
+        if (rc) {
+            PT_ERR(&s->dev, "Mapping of MSI%s (rc: %i, vec: %#x, entry %#x)\n",
+                   is_msix ? "-X" : "", rc, gvec, msix_entry);
+            return rc;
+        }
+    }
+
+    return 0;
+}
+static int msi_msix_update(XenPCIPassthroughState *s,
+                           uint64_t addr,
+                           uint32_t data,
+                           int pirq,
+                           bool is_msix,
+                           int msix_entry,
+                           int *old_pirq)
+{
+    PCIDevice *d = &s->dev;
+    uint8_t gvec = msi_vector(data);
+    uint32_t gflags = msi_gflags(data, addr);
+    int rc = 0;
+    uint64_t table_addr = 0;
+
+    PT_LOG(d, "Updating MSI%s with pirq %d gvec %#x gflags %#x (entry: %#x)\n",
+           is_msix ? "-X" : "", pirq, gvec, gflags, msix_entry);
+
+    if (is_msix) {
+        table_addr = s->msix->mmio_base_addr;
+    }
+
+    rc = xc_domain_update_msi_irq(xen_xc, xen_domid, gvec,
+                                  pirq, gflags, table_addr);
+
+    if (rc) {
+        PT_ERR(d, "Updating of MSI%s failed. (rc: %d)\n",
+               is_msix ? "-X" : "", rc);
+
+        if (xc_physdev_unmap_pirq(xen_xc, xen_domid, *old_pirq)) {
+            PT_ERR(d, "Unmapping of MSI%s pirq %d failed.\n",
+                   is_msix ? "-X" : "", *old_pirq);
+        }
+        *old_pirq = PT_UNASSIGNED_PIRQ;
+    }
+    return rc;
+}
+
+static int msi_msix_disable(XenPCIPassthroughState *s,
+                            uint64_t addr,
+                            uint32_t data,
+                            int pirq,
+                            bool is_msix,
+                            bool is_binded)
+{
+    PCIDevice *d = &s->dev;
+    uint8_t gvec = msi_vector(data);
+    uint32_t gflags = msi_gflags(data, addr);
+    int rc = 0;
+
+    if (pirq == PT_UNASSIGNED_PIRQ) {
+        return 0;
+    }
+
+    if (is_binded) {
+        PT_LOG(d, "Unbind MSI%s with pirq %d, gvec %#x\n",
+               is_msix ? "-X" : "", pirq, gvec);
+        rc = xc_domain_unbind_msi_irq(xen_xc, xen_domid, gvec, pirq, gflags);
+        if (rc) {
+            PT_ERR(d, "Unbinding of MSI%s failed. (pirq: %d, gvec: %#x)\n",
+                   is_msix ? "-X" : "", pirq, gvec);
+            return rc;
+        }
+    }
+
+    PT_LOG(d, "Unmap MSI%s pirq %d\n", is_msix ? "-X" : "", pirq);
+    rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, pirq);
+    if (rc) {
+        PT_ERR(d, "Unmapping of MSI%s pirq %d failed. (rc: %i)\n",
+               is_msix ? "-X" : "", pirq, rc);
+        return rc;
+    }
+
+    return 0;
+}
+
+/*
+ * MSI virtualization functions
+ */
+
+int pt_msi_set_enable(XenPCIPassthroughState *s, bool enable)
+{
+    PT_LOG(&s->dev, "%s MSI.\n", enable ? "enabling" : "disabling");
+
+    if (!s->msi) {
+        return -1;
+    }
+
+    return msi_msix_enable(s, s->msi->ctrl_offset, PCI_MSI_FLAGS_ENABLE,
+                           enable);
+}
+
+/* setup physical msi, but don't enable it */
+int pt_msi_setup(XenPCIPassthroughState *s)
+{
+    int pirq = PT_UNASSIGNED_PIRQ;
+    int rc = 0;
+    XenPTMSI *msi = s->msi;
+
+    if (msi->initialized) {
+        PT_ERR(&s->dev,
+               "Setup physical MSI when it has been properly initialized.\n");
+        return -1;
+    }
+
+    rc = msi_msix_setup(s, msi_addr64(msi), msi->data, &pirq, false, 0, true);
+    if (rc) {
+        return rc;
+    }
+
+    if (pirq < 0) {
+        PT_ERR(&s->dev, "Invalid pirq number: %d.\n", pirq);
+        return -1;
+    }
+
+    msi->pirq = pirq;
+    PT_LOG(&s->dev, "MSI mapped with pirq %d.\n", pirq);
+
+    return 0;
+}
+
+int pt_msi_update(XenPCIPassthroughState *s)
+{
+    XenPTMSI *msi = s->msi;
+    return msi_msix_update(s, msi_addr64(msi), msi->data, msi->pirq,
+                           false, 0, &msi->pirq);
+}
+
+void pt_msi_disable(XenPCIPassthroughState *s)
+{
+    XenPTMSI *msi = s->msi;
+
+    if (!msi) {
+        return;
+    }
+
+    pt_msi_set_enable(s, false);
+
+    msi_msix_disable(s, msi_addr64(msi), msi->data, msi->pirq, false,
+                     msi->initialized);
+
+    /* clear msi info */
+    msi->flags = 0;
+    msi->mapped = false;
+    msi->pirq = PT_UNASSIGNED_PIRQ;
+}
+
+/*
+ * MSI-X virtualization functions
+ */
+
+static int msix_set_enable(XenPCIPassthroughState *s, bool enabled)
+{
+    PT_LOG(&s->dev, "%s MSI-X.\n", enabled ? "enabling" : "disabling");
+
+    if (!s->msix) {
+        return -1;
+    }
+
+    return msi_msix_enable(s, s->msix->ctrl_offset, PCI_MSIX_FLAGS_ENABLE,
+                           enabled);
+}
+
+static int pt_msix_update_one(XenPCIPassthroughState *s, int entry_nr)
+{
+    XenPTMSIXEntry *entry = NULL;
+    int pirq;
+    int rc;
+
+    if (entry_nr < 0 || entry_nr >= s->msix->total_entries) {
+        return -EINVAL;
+    }
+
+    entry = &s->msix->msix_entry[entry_nr];
+
+    if (!entry->updated) {
+        return 0;
+    }
+
+    pirq = entry->pirq;
+
+    rc = msi_msix_setup(s, entry->data, entry->data, &pirq, true, entry_nr,
+                        entry->pirq == PT_UNASSIGNED_PIRQ);
+    if (rc) {
+        return rc;
+    }
+    if (entry->pirq == PT_UNASSIGNED_PIRQ) {
+        entry->pirq = pirq;
+    }
+
+    rc = msi_msix_update(s, entry->addr, entry->data, pirq, true,
+                         entry_nr, &entry->pirq);
+
+    if (!rc) {
+        entry->updated = false;
+    }
+
+    return rc;
+}
+
+int pt_msix_update(XenPCIPassthroughState *s)
+{
+    XenPTMSIX *msix = s->msix;
+    int i;
+
+    for (i = 0; i < msix->total_entries; i++) {
+        pt_msix_update_one(s, i);
+    }
+
+    return 0;
+}
+
+void pt_msix_disable(XenPCIPassthroughState *s)
+{
+    int i = 0;
+
+    msix_set_enable(s, false);
+
+    for (i = 0; i < s->msix->total_entries; i++) {
+        XenPTMSIXEntry *entry = &s->msix->msix_entry[i];
+
+        msi_msix_disable(s, entry->addr, entry->data, entry->pirq, true, true);
+
+        /* clear MSI-X info */
+        entry->pirq = PT_UNASSIGNED_PIRQ;
+        entry->updated = false;
+    }
+}
+
+int pt_msix_update_remap(XenPCIPassthroughState *s, int bar_index)
+{
+    XenPTMSIXEntry *entry;
+    int i, ret;
+
+    if (!(s->msix && s->msix->bar_index == bar_index)) {
+        return 0;
+    }
+
+    for (i = 0; i < s->msix->total_entries; i++) {
+        entry = &s->msix->msix_entry[i];
+        if (entry->pirq != PT_UNASSIGNED_PIRQ) {
+            ret = xc_domain_unbind_pt_irq(xen_xc, xen_domid, entry->pirq,
+                                          PT_IRQ_TYPE_MSI, 0, 0, 0, 0);
+            if (ret) {
+                PT_ERR(&s->dev, "unbind MSI-X entry %d failed\n", entry->pirq);
+            }
+            entry->updated = true;
+        }
+    }
+    return pt_msix_update(s);
+}
+
+static uint32_t get_entry_value(XenPTMSIXEntry *e, int offset)
+{
+    switch (offset) {
+    case PCI_MSIX_ENTRY_LOWER_ADDR:
+        return e->addr & UINT32_MAX;
+    case PCI_MSIX_ENTRY_UPPER_ADDR:
+        return e->addr >> 32;
+    case PCI_MSIX_ENTRY_DATA:
+        return e->data;
+    case PCI_MSIX_ENTRY_VECTOR_CTRL:
+        return e->vector_ctrl;
+    default:
+        return 0;
+    }
+}
+
+static void set_entry_value(XenPTMSIXEntry *e, int offset, uint32_t val)
+{
+    switch (offset) {
+    case PCI_MSIX_ENTRY_LOWER_ADDR:
+        e->addr = (e->addr & ((uint64_t)UINT32_MAX << 32)) | val;
+        break;
+    case PCI_MSIX_ENTRY_UPPER_ADDR:
+        e->addr = (uint64_t)val << 32 | (e->addr & UINT32_MAX);
+        break;
+    case PCI_MSIX_ENTRY_DATA:
+        e->data = val;
+        break;
+    case PCI_MSIX_ENTRY_VECTOR_CTRL:
+        e->vector_ctrl = val;
+        break;
+    }
+}
+
+static void pci_msix_write(void *opaque, target_phys_addr_t addr,
+                           uint64_t val, unsigned size)
+{
+    XenPCIPassthroughState *s = opaque;
+    XenPTMSIX *msix = s->msix;
+    XenPTMSIXEntry *entry;
+    int entry_nr, offset;
+
+    entry_nr = addr / PCI_MSIX_ENTRY_SIZE;
+    if (entry_nr < 0 || entry_nr >= msix->total_entries) {
+        PT_ERR(&s->dev, "asked MSI-X entry '%i' invalid!\n", entry_nr);
+        return;
+    }
+    entry = &msix->msix_entry[entry_nr];
+    offset = addr % PCI_MSIX_ENTRY_SIZE;
+
+    if (offset != PCI_MSIX_ENTRY_VECTOR_CTRL) {
+        const volatile uint32_t *vec_ctrl;
+
+        if (get_entry_value(entry, offset) == val) {
+            return;
+        }
+
+        /*
+         * If Xen intercepts the mask bit access, entry->vec_ctrl may not be
+         * up-to-date. Read from hardware directly.
+         */
+        vec_ctrl = s->msix->phys_iomem_base + entry_nr * PCI_MSIX_ENTRY_SIZE
+            + PCI_MSIX_ENTRY_VECTOR_CTRL;
+
+        if (msix->enabled && !(*vec_ctrl & PCI_MSIX_ENTRY_CTRL_MASKBIT)) {
+            PT_ERR(&s->dev, "Can't update msix entry %d since MSI-X is already "
+                   "enabled.\n", entry_nr);
+            return;
+        }
+
+        entry->updated = true;
+    }
+
+    set_entry_value(entry, offset, val);
+
+    if (offset == PCI_MSIX_ENTRY_VECTOR_CTRL) {
+        if (msix->enabled && !(val & PCI_MSIX_ENTRY_CTRL_MASKBIT)) {
+            pt_msix_update_one(s, entry_nr);
+        }
+    }
+}
+
+static uint64_t pci_msix_read(void *opaque, target_phys_addr_t addr,
+                              unsigned size)
+{
+    XenPCIPassthroughState *s = opaque;
+    XenPTMSIX *msix = s->msix;
+    int entry_nr, offset;
+
+    entry_nr = addr / PCI_MSIX_ENTRY_SIZE;
+    if (entry_nr < 0) {
+        PT_ERR(&s->dev, "asked MSI-X entry '%i' invalid!\n", entry_nr);
+        return 0;
+    }
+
+    offset = addr % PCI_MSIX_ENTRY_SIZE;
+
+    if (addr < msix->total_entries * PCI_MSIX_ENTRY_SIZE) {
+        return get_entry_value(&msix->msix_entry[entry_nr], offset);
+    } else {
+        /* Pending Bit Array (PBA) */
+        return *(uint32_t *)(msix->phys_iomem_base + addr);
+    }
+}
+
+static const MemoryRegionOps pci_msix_ops = {
+    .read = pci_msix_read,
+    .write = pci_msix_write,
+    .endianness = DEVICE_NATIVE_ENDIAN,
+    .valid = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+        .unaligned = false,
+    },
+};
+
+int pt_msix_init(XenPCIPassthroughState *s, uint32_t base)
+{
+    uint8_t id = 0;
+    uint16_t control = 0;
+    uint32_t table_off = 0;
+    int i, total_entries, bar_index;
+    HostPCIDevice *hd = s->real_device;
+    PCIDevice *d = &s->dev;
+    int fd = -1;
+    XenPTMSIX *msix = NULL;
+    int rc = 0;
+
+    rc = host_pci_get_byte(hd, base + PCI_CAP_LIST_ID, &id);
+    if (rc) {
+        return rc;
+    }
+
+    if (id != PCI_CAP_ID_MSIX) {
+        PT_ERR(d, "Invalid id %#x base %#x\n", id, base);
+        return -1;
+    }
+
+    host_pci_get_word(hd, base + PCI_MSIX_FLAGS, &control);
+    total_entries = control & PCI_MSIX_FLAGS_QSIZE;
+    total_entries += 1;
+
+    s->msix = g_malloc0(sizeof (XenPTMSIX)
+                        + total_entries * sizeof (XenPTMSIXEntry));
+    msix = s->msix;
+
+    msix->total_entries = total_entries;
+    for (i = 0; i < total_entries; i++) {
+        msix->msix_entry[i].pirq = PT_UNASSIGNED_PIRQ;
+    }
+
+    memory_region_init_io(&msix->mmio, &pci_msix_ops, s, "xen-pci-pt-msix",
+                          (total_entries * PCI_MSIX_ENTRY_SIZE
+                           + XC_PAGE_SIZE - 1)
+                          & XC_PAGE_MASK);
+
+    host_pci_get_long(hd, base + PCI_MSIX_TABLE, &table_off);
+    bar_index = msix->bar_index = table_off & PCI_MSIX_FLAGS_BIRMASK;
+    table_off = table_off & ~PCI_MSIX_FLAGS_BIRMASK;
+    msix->table_base = s->real_device->io_regions[bar_index].base_addr;
+    PT_LOG(d, "get MSI-X table BAR base 0x%"PRIx64"\n", msix->table_base);
+
+    fd = open("/dev/mem", O_RDWR);
+    if (fd == -1) {
+        rc = -errno;
+        PT_ERR(d, "Can't open /dev/mem: %s\n", strerror(errno));
+        goto error_out;
+    }
+    PT_LOG(d, "table_off = %#x, total_entries = %d\n",
+           table_off, total_entries);
+    msix->table_offset_adjust = table_off & 0x0fff;
+    msix->phys_iomem_base =
+        mmap(NULL,
+             total_entries * PCI_MSIX_ENTRY_SIZE + msix->table_offset_adjust,
+             PROT_READ,
+             MAP_SHARED | MAP_LOCKED,
+             fd,
+             msix->table_base + table_off - msix->table_offset_adjust);
+    close(fd);
+    if (msix->phys_iomem_base == MAP_FAILED) {
+        rc = -errno;
+        PT_ERR(d, "Can't map physical MSI-X table: %s\n", strerror(errno));
+        goto error_out;
+    }
+    msix->phys_iomem_base = (char *)msix->phys_iomem_base
+        + msix->table_offset_adjust;
+
+    PT_LOG(d, "mapping physical MSI-X table to %p\n", msix->phys_iomem_base);
+
+    memory_region_add_subregion_overlap(&s->bar[bar_index], table_off,
+                                        &msix->mmio,
+                                        2); /* Priority: pci default + 1 */
+
+    return 0;
+
+error_out:
+    memory_region_destroy(&msix->mmio);
+    g_free(s->msix);
+    s->msix = NULL;
+    return rc;
+}
+
+void pt_msix_delete(XenPCIPassthroughState *s)
+{
+    XenPTMSIX *msix = s->msix;
+
+    if (!msix) {
+        return;
+    }
+
+    /* unmap the MSI-X memory mapped register area */
+    if (msix->phys_iomem_base) {
+        PT_LOG(&s->dev, "unmapping physical MSI-X table from %p\n",
+               msix->phys_iomem_base);
+        munmap(msix->phys_iomem_base, msix->total_entries * PCI_MSIX_ENTRY_SIZE
+               + msix->table_offset_adjust);
+    }
+
+    memory_region_del_subregion(&s->bar[msix->bar_index], &msix->mmio);
+    memory_region_destroy(&msix->mmio);
+
+    g_free(s->msix);
+    s->msix = NULL;
+}
-- 
Anthony PERARD