[Qemu-devel] [PATCH v3 2/2] Inter-VM shared memory PCI device

[Cam Macdonell]

Support an inter-vm shared memory device that maps a shared-memory object as a
PCI device in the guest.  This patch also supports interrupts between guest by
communicating over a unix domain socket.  This patch applies to the qemu-kvm
repository.

    -device ivshmem,size=<size in MB>[,shm=<shm name>]

Interrupts are supported between multiple VMs by using a shared memory server
by using a chardev socket.

    -device ivshmem,size=<size in MB>[,shm=<shm name>][,chardev=<id>][,msi=on]
            [,irqfd=on][,nvectors=n]
    -chardev socket,path=<path>,id=<id>

Sample programs, init scripts and the shared memory server are available in a
git repo here:

    www.gitorious.org/nahanni
---
 Makefile.target |    3 +
 hw/ivshmem.c    |  622 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 qemu-char.c     |    6 +
 qemu-char.h     |    3 +
 4 files changed, 634 insertions(+), 0 deletions(-)
 create mode 100644 hw/ivshmem.c

diff --git a/Makefile.target b/Makefile.target
index 4d88543..15edf19 100644
--- a/Makefile.target
+++ b/Makefile.target
@@ -219,6 +219,9 @@ obj-y += pcnet.o
 obj-y += rtl8139.o
 obj-y += e1000.o
 
+# Inter-VM PCI shared memory
+obj-y += ivshmem.o
+
 # Hardware support
 obj-i386-y = ide/core.o ide/qdev.o ide/isa.o ide/pci.o ide/piix.o
 obj-i386-y += pckbd.o $(sound-obj-y) dma.o
diff --git a/hw/ivshmem.c b/hw/ivshmem.c
new file mode 100644
index 0000000..c76aae3
--- /dev/null
+++ b/hw/ivshmem.c
@@ -0,0 +1,622 @@
+/*
+ * Inter-VM Shared Memory PCI device.
+ *
+ * Author:
+ *      Cam Macdonell <address@hidden>
+ *
+ * Based On: cirrus_vga.c and rtl8139.c
+ *
+ * This code is licensed under the GNU GPL v2.
+ */
+#include <sys/mman.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/io.h>
+#include <sys/ioctl.h>
+#include <sys/eventfd.h>
+#include "hw.h"
+#include "console.h"
+#include "pc.h"
+#include "pci.h"
+#include "sysemu.h"
+
+#include "msix.h"
+#include "qemu-kvm.h"
+#include "libkvm.h"
+
+#include <sys/eventfd.h>
+#include <sys/mman.h>
+#include <sys/socket.h>
+#include <sys/ioctl.h>
+
+#define PCI_COMMAND_IOACCESS                0x0001
+#define PCI_COMMAND_MEMACCESS               0x0002
+
+#define DEBUG_IVSHMEM
+
+#define IVSHMEM_IRQFD   0
+#define IVSHMEM_MSI     1
+
+#ifdef DEBUG_IVSHMEM
+#define IVSHMEM_DPRINTF(fmt, args...)        \
+    do {printf("IVSHMEM: " fmt, ##args); } while (0)
+#else
+#define IVSHMEM_DPRINTF(fmt, args...)
+#endif
+
+#define NEW_GUEST_VAL UINT_MAX
+
+typedef struct IVShmemState {
+    PCIDevice dev;
+    uint32_t intrmask;
+    uint32_t intrstatus;
+    uint32_t doorbell;
+
+    CharDriverState * chr;
+    CharDriverState * eventfd_chr;
+    int ivshmem_mmio_io_addr;
+
+    pcibus_t mmio_addr;
+    uint8_t *ivshmem_ptr;
+    unsigned long ivshmem_offset;
+    unsigned int ivshmem_size;
+    int shm_fd; /* shared memory file descriptor */
+
+    struct kvm_ioeventfd ioeventfds[16];
+    int eventfds[16]; /* for now we have a limit of 16 inter-connected guests 
*/
+    int eventfd_posn;
+    int num_eventfds;
+    uint32_t vectors;
+    uint32_t features;
+
+    char * shmobj;
+    uint32_t size; /*size of shared memory in MB*/
+} IVShmemState;
+
+/* registers for the Inter-VM shared memory device */
+enum ivshmem_registers {
+    IntrMask = 0,
+    IntrStatus = 4,
+    IVPosition = 8,
+    Doorbell = 12,
+};
+
+static inline uint32_t ivshmem_has_feature(IVShmemState *ivs, int feature) {
+    return (ivs->features & (1 << feature));
+}
+
+static inline int is_power_of_two(int x) {return (x & (x-1)) == 0;}
+
+static void ivshmem_map(PCIDevice *pci_dev, int region_num,
+                    pcibus_t addr, pcibus_t size, int type)
+{
+    IVShmemState *s = DO_UPCAST(IVShmemState, dev, pci_dev);
+
+    IVSHMEM_DPRINTF("addr = %u size = %u\n", (uint32_t)addr, (uint32_t)size);
+    cpu_register_physical_memory(addr, s->ivshmem_size, s->ivshmem_offset);
+
+}
+
+/* accessing registers - based on rtl8139 */
+static void ivshmem_update_irq(IVShmemState *s, int val)
+{
+    int isr;
+    isr = (s->intrstatus & s->intrmask) & 0xffffffff;
+
+    /* don't print ISR resets */
+    if (isr) {
+        IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
+           isr ? 1 : 0, s->intrstatus, s->intrmask);
+    }
+
+    qemu_set_irq(s->dev.irq[0], (isr != 0));
+}
+
+static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
+{
+    IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
+
+    s->intrmask = val;
+
+    ivshmem_update_irq(s, val);
+}
+
+static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
+{
+    uint32_t ret = s->intrmask;
+
+    IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
+
+    return ret;
+}
+
+static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
+{
+    IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
+
+    s->intrstatus = val;
+
+    ivshmem_update_irq(s, val);
+    return;
+}
+
+static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
+{
+    uint32_t ret = s->intrstatus;
+
+    /* reading ISR clears all interrupts */
+    s->intrstatus = 0;
+
+    ivshmem_update_irq(s, 0);
+
+    return ret;
+}
+
+static void ivshmem_io_writew(void *opaque, uint8_t addr, uint32_t val)
+{
+
+    IVSHMEM_DPRINTF("We shouldn't be writing words\n");
+}
+
+static void ivshmem_io_writel(void *opaque, uint8_t addr, uint32_t val)
+{
+    IVShmemState *s = opaque;
+    u_int64_t writelong = val;
+    int dest = (addr - Doorbell)/sizeof(int);
+
+    addr &= 0xfe;
+
+    switch (addr)
+    {
+        case IntrMask:
+            ivshmem_IntrMask_write(s, val);
+            break;
+
+        case IntrStatus:
+            ivshmem_IntrStatus_write(s, val);
+            break;
+
+        default:
+            /* check doorbell range */
+            if ((addr >= Doorbell) &&
+                    (addr <= Doorbell + sizeof(int) * s->num_eventfds)) {
+                IVSHMEM_DPRINTF("Writing %ld to VM %d\n", writelong, dest);
+                if (write(s->eventfds[dest], &(writelong), 8) != 8)
+                    IVSHMEM_DPRINTF("error writing to eventfd\n");
+            } else {
+                IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
+            }
+    }
+}
+
+static void ivshmem_io_writeb(void *opaque, uint8_t addr, uint32_t val)
+{
+    IVSHMEM_DPRINTF("We shouldn't be writing bytes\n");
+}
+
+static uint32_t ivshmem_io_readw(void *opaque, uint8_t addr)
+{
+
+    IVSHMEM_DPRINTF("We shouldn't be reading words\n");
+    return 0;
+}
+
+static uint32_t ivshmem_io_readl(void *opaque, uint8_t addr)
+{
+
+    IVShmemState *s = opaque;
+    uint32_t ret;
+
+    switch (addr)
+    {
+        case IntrMask:
+            ret = ivshmem_IntrMask_read(s);
+            break;
+
+        case IntrStatus:
+            ret = ivshmem_IntrStatus_read(s);
+            break;
+
+        case IVPosition:
+            /* return my id in the ivshmem list */
+            ret = s->eventfd_posn;
+            break;
+
+        default:
+            IVSHMEM_DPRINTF("why are we reading 0x%x\n", addr);
+            ret = 0;
+    }
+
+    return ret;
+
+}
+
+static uint32_t ivshmem_io_readb(void *opaque, uint8_t addr)
+{
+    IVSHMEM_DPRINTF("We shouldn't be reading bytes\n");
+
+    return 0;
+}
+
+static void ivshmem_mmio_writeb(void *opaque,
+                                target_phys_addr_t addr, uint32_t val)
+{
+    ivshmem_io_writeb(opaque, addr & 0xFF, val);
+}
+
+static void ivshmem_mmio_writew(void *opaque,
+                                target_phys_addr_t addr, uint32_t val)
+{
+    ivshmem_io_writew(opaque, addr & 0xFF, val);
+}
+
+static void ivshmem_mmio_writel(void *opaque,
+                                target_phys_addr_t addr, uint32_t val)
+{
+    ivshmem_io_writel(opaque, addr & 0xFF, val);
+}
+
+static uint32_t ivshmem_mmio_readb(void *opaque, target_phys_addr_t addr)
+{
+    return ivshmem_io_readb(opaque, addr & 0xFF);
+}
+
+static uint32_t ivshmem_mmio_readw(void *opaque, target_phys_addr_t addr)
+{
+    uint32_t val = ivshmem_io_readw(opaque, addr & 0xFF);
+    return val;
+}
+
+static uint32_t ivshmem_mmio_readl(void *opaque, target_phys_addr_t addr)
+{
+    uint32_t val = ivshmem_io_readl(opaque, addr & 0xFF);
+    return val;
+}
+
+static CPUReadMemoryFunc *ivshmem_mmio_read[3] = {
+    ivshmem_mmio_readb,
+    ivshmem_mmio_readw,
+    ivshmem_mmio_readl,
+};
+
+static CPUWriteMemoryFunc *ivshmem_mmio_write[3] = {
+    ivshmem_mmio_writeb,
+    ivshmem_mmio_writew,
+    ivshmem_mmio_writel,
+};
+
+static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
+{
+    IVShmemState *s = opaque;
+    uint32_t val;
+
+    val = *buf;
+    if (msix_enabled(&s->dev) && (val < s->vectors))
+        msix_notify(&s->dev, val);
+    else
+        ivshmem_IntrStatus_write(s, val);
+
+    IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
+}
+
+static int ivshmem_can_receive(void * opaque)
+{
+    return 8;
+}
+
+static void ivshmem_event(void *opaque, int event)
+{
+//    IVShmemState *s = opaque;
+    IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
+}
+
+
+static CharDriverState* create_eventfd_chr_device(void * opaque, int eventfd)
+{
+    // create a event character device based on the passed eventfd
+    IVShmemState *s = opaque;
+    CharDriverState * chr;
+
+    chr = qemu_chr_open_eventfd(eventfd);
+
+    if (chr == NULL) {
+        IVSHMEM_DPRINTF("creating eventfd for eventfd %d failed\n", eventfd);
+        exit(-1);
+    }
+
+    qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
+                      ivshmem_event, s);
+
+    return chr;
+
+}
+
+static int check_shm_size(IVShmemState *s, int shmemfd) {
+    /* check that the guest isn't going to try and map more memory than the
+     * card server allocated return -1 to indicate error */
+
+    struct stat buf;
+
+    fstat(shmemfd, &buf);
+
+    if (s->ivshmem_size > buf.st_size) {
+        fprintf(stderr, "IVSHMEM ERROR: Requested memory size greater");
+        fprintf(stderr, " than shared object size (%d > %ld)\n",
+                                          s->ivshmem_size, buf.st_size);
+        return -1;
+    } else {
+        return 0;
+    }
+}
+
+static void create_shared_memory_BAR(IVShmemState *s, int fd) {
+
+    s->shm_fd = fd;
+
+    s->ivshmem_offset = qemu_ram_mmap(s->shm_fd, s->ivshmem_size,
+             MAP_SHARED, 0);
+
+    s->ivshmem_ptr = qemu_get_ram_ptr(s->ivshmem_offset);
+
+    /* region for shared memory */
+    pci_register_bar(&s->dev, 2, s->ivshmem_size,
+                       PCI_BASE_ADDRESS_SPACE_MEMORY, ivshmem_map);
+}
+
+static int ivshmem_irqfd(PCIDevice* pdev, uint16_t vector, int fd)
+{
+    struct kvm_irqfd call = { };
+    int r;
+
+    IVSHMEM_DPRINTF("inside irqfd\n");
+    if (vector >= pdev->msix_entries_nr)
+        return -EINVAL;
+    call.fd = fd;
+    call.gsi = pdev->msix_irq_entries[vector].gsi;
+    r = kvm_vm_ioctl(kvm_state, KVM_IRQFD, &call);
+    if (r < 0)
+        return r;
+    return 0;
+}
+
+static int ivshmem_ioeventfd(IVShmemState* s, int posn, int fd)
+{
+
+    int ret, offset;
+    struct kvm_ioeventfd iofd;
+
+    offset = Doorbell + 4 * posn;
+    iofd.addr = s->mmio_addr + offset;
+    iofd.len = 4;
+    iofd.flags = 0;
+    iofd.fd = fd;
+
+    ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
+
+    if (ret < 0) {
+        fprintf(stderr, "error assigning ioeventfd (%d) - disabling 
ioeventfds\n", ret);
+        perror(strerror(ret));
+    } else IVSHMEM_DPRINTF("success assigning ioeventfd (%d)\n", ret);
+
+    return ret;
+}
+/* notify that a new guest has joined */
+static void new_guest_interrupt(IVShmemState *s)
+{
+    if (msix_enabled(&s->dev)) {
+        msix_notify(&s->dev, 0);
+    } else {
+        ivshmem_IntrStatus_write(s, NEW_GUEST_VAL);
+    }
+}
+
+static void ivshmem_read(void *opaque, const uint8_t * buf, int flags)
+{
+    IVShmemState *s = opaque;
+    int incoming_fd, tmp_fd;
+    long incoming_posn;
+
+    memcpy(&incoming_posn, buf, sizeof(long));
+    /* pick off s->chr->msgfd and store it, posn should accompany msg */
+    tmp_fd = qemu_chr_get_msgfd(s->chr);
+    IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
+
+    if (tmp_fd == -1) {
+        s->eventfd_posn = incoming_posn;
+        return;
+    }
+
+    /* because of the implementation of get_msgfd, we need a dup */
+    incoming_fd = dup(tmp_fd);
+
+    /* if the position is -1, then it's shared memory region fd */
+    if (incoming_posn == -1) {
+
+        s->num_eventfds = 0;
+
+        if (check_shm_size(s, incoming_fd) == -1) {
+            exit(-1);
+        }
+
+        /* creating a BAR in qemu_chr callback may be crazy */
+        create_shared_memory_BAR(s, incoming_fd);
+
+       return;
+    }
+
+    if (ivshmem_has_feature(s, IVSHMEM_IRQFD)) {
+        /* allocate ioeventfd for the new guest */
+        ivshmem_ioeventfd(s, incoming_posn, s->eventfds[incoming_posn]);
+    }
+
+    /* this is an eventfd for a particular guest VM */
+    IVSHMEM_DPRINTF("eventfds[%ld] = %d\n", incoming_posn, incoming_fd);
+    s->eventfds[incoming_posn] = incoming_fd;
+
+    /* keep track of the maximum VM ID */
+    if (incoming_posn > s->num_eventfds) {
+        s->num_eventfds = incoming_posn;
+    }
+
+    if (incoming_posn == s->eventfd_posn) {
+        if (ivshmem_has_feature(s, IVSHMEM_IRQFD)) {
+            /* setup irqfd for this VM's eventfd */
+            /* vector is set to 1 */
+            ivshmem_irqfd(&s->dev, 1, s->eventfds[s->eventfd_posn]);
+        } else {
+            /* initialize char device for callback on my eventfd */
+            s->eventfd_chr = create_eventfd_chr_device(s, 
s->eventfds[s->eventfd_posn]);
+        }
+    } else {
+        new_guest_interrupt(s);
+    }
+
+    return;
+}
+
+static void ivshmem_reset(DeviceState *d)
+{
+    return;
+}
+
+static void ivshmem_mmio_map(PCIDevice *pci_dev, int region_num,
+                       pcibus_t addr, pcibus_t size, int type)
+{
+    IVShmemState *s = DO_UPCAST(IVShmemState, dev, pci_dev);
+
+    s->mmio_addr = addr;
+    cpu_register_physical_memory(addr + 0, 0x400, s->ivshmem_mmio_io_addr);
+
+    /* now that our mmio region has been allocated, we can receive
+     * the file descriptors */
+    qemu_chr_add_handlers(s->chr, ivshmem_can_receive, ivshmem_read,
+                     ivshmem_event, s);
+
+}
+
+static int pci_ivshmem_init(PCIDevice *dev)
+{
+    IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
+    uint8_t *pci_conf;
+    int i;
+
+    /* BARs must be a power of 2 */
+    if (is_power_of_two(s->size))
+        s->ivshmem_size = s->size * 1024* 1024;
+    else {
+        fprintf(stderr, "ivshmem: size must be power of 2\n");
+        exit(1);
+    }
+
+    /* IRQFD requires MSI */
+    if (ivshmem_has_feature(s, IVSHMEM_IRQFD) &&
+        !ivshmem_has_feature(s, IVSHMEM_MSI)) {
+        fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n");
+        exit(1);
+    }
+
+    pci_conf = s->dev.config;
+    pci_conf[0x00] = 0xf4; // Qumranet vendor ID 0x5002
+    pci_conf[0x01] = 0x1a;
+    pci_conf[0x02] = 0x10;
+    pci_conf[0x03] = 0x11;
+    pci_conf[0x04] = PCI_COMMAND_IOACCESS | PCI_COMMAND_MEMACCESS;
+    pci_conf[0x0a] = 0x00; // RAM controller
+    pci_conf[0x0b] = 0x05;
+    pci_conf[0x0e] = 0x00; // header_type
+
+    s->ivshmem_mmio_io_addr = cpu_register_io_memory(ivshmem_mmio_read,
+                                    ivshmem_mmio_write, s);
+    /* region for registers*/
+    pci_register_bar(&s->dev, 0, 0x400,
+                           PCI_BASE_ADDRESS_SPACE_MEMORY, ivshmem_mmio_map);
+
+    if ((s->chr != NULL) && (strncmp(s->chr->filename, "unix:", 5) == 0)) {
+        /* if we get a UNIX socket as the parameter we will talk
+         * to the ivshmem server later once the MMIO BAR is actually
+         * allocated (see ivshmem_mmio_map) */
+
+        IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
+                                                            s->chr->filename);
+
+        s->eventfd_posn = -1;
+
+    } else {
+        /* just map the file immediately, we're not using a server */
+        int fd;
+
+        IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
+
+        if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
+                        S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
+            fprintf(stderr, "kvm_ivshmem: could not open shared file\n");
+            exit(-1);
+        }
+
+        /* mmap onto PCI device's memory */
+        if (ftruncate(fd, s->ivshmem_size) != 0) {
+            fprintf(stderr, "kvm_ivshmem: could not truncate shared file\n");
+        }
+
+        create_shared_memory_BAR(s, fd);
+
+    }
+
+    IVSHMEM_DPRINTF("shared memory size is = %d\n", s->size);
+
+    pci_conf[PCI_INTERRUPT_PIN] = 1; // we are going to support interrupts
+
+    /* allocate the MSI-X vectors */
+    if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
+
+        if (!msix_init(&s->dev, s->vectors, 1, 0)) {
+            pci_register_bar(&s->dev, 1,
+                             msix_bar_size(&s->dev),
+                             PCI_BASE_ADDRESS_SPACE_MEMORY,
+                             msix_mmio_map);
+            IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
+        } else {
+            IVSHMEM_DPRINTF("msix initialization failed\n");
+        }
+
+        /* 'activate' the vectors */
+        for (i = 0; i < s->vectors; i++) {
+            msix_vector_use(&s->dev, i);
+        }
+    }
+
+    return 0;
+}
+
+static int pci_ivshmem_uninit(PCIDevice *dev)
+{
+    IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
+
+    cpu_unregister_io_memory(s->ivshmem_mmio_io_addr);
+
+    return 0;
+}
+
+static PCIDeviceInfo ivshmem_info = {
+    .qdev.name  = "ivshmem",
+    .qdev.size  = sizeof(IVShmemState),
+    .qdev.reset = ivshmem_reset,
+    .init       = pci_ivshmem_init,
+    .exit       = pci_ivshmem_uninit,
+    .qdev.props = (Property[]) {
+        DEFINE_PROP_CHR("chardev", IVShmemState, chr),
+        DEFINE_PROP_UINT32("size", IVShmemState, size, 0),
+        DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 2),
+        DEFINE_PROP_BIT("irqfd", IVShmemState, features, IVSHMEM_IRQFD, false),
+        DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
+        DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
+        DEFINE_PROP_END_OF_LIST(),
+    }
+};
+
+static void ivshmem_register_devices(void)
+{
+    pci_qdev_register(&ivshmem_info);
+}
+
+device_init(ivshmem_register_devices)
diff --git a/qemu-char.c b/qemu-char.c
index 40cfefa..6533395 100644
--- a/qemu-char.c
+++ b/qemu-char.c
@@ -2074,6 +2074,12 @@ static void tcp_chr_read(void *opaque)
     }
 }
 
+CharDriverState *qemu_chr_open_eventfd(int eventfd){
+
+    return qemu_chr_open_fd(eventfd, eventfd);
+
+}
+
 static void tcp_chr_connect(void *opaque)
 {
     CharDriverState *chr = opaque;
diff --git a/qemu-char.h b/qemu-char.h
index bcc0766..9a0d2c0 100644
--- a/qemu-char.h
+++ b/qemu-char.h
@@ -93,6 +93,9 @@ void qemu_chr_info_print(Monitor *mon, const QObject 
*ret_data);
 void qemu_chr_info(Monitor *mon, QObject **ret_data);
 CharDriverState *qemu_chr_find(const char *name);
 
+/* add an eventfd to the qemu devices that are polled */
+CharDriverState *qemu_chr_open_eventfd(int eventfd);
+
 extern int term_escape_char;
 
 /* async I/O support */
-- 
1.6.6.1