Re: [RFC v3 7/8] vhost-vdpa: introduce vhost-vdpa backend

[Jason Wang <jasowang@redhat.com>]

On 2020/5/29 下午10:06, Cindy Lu wrote:
> From: Tiwei Bie <tiwei.bie@intel.com>
>
> Currently we have 2 types of vhost backends in QEMU: vhost kernel and
> vhost-user. The above patch provides a generic device for vDPA purpose,
> this vDPA device exposes to user space a non-vendor-specific configuration
> interface for setting up a vhost HW accelerator, this patch set introduces
> a third vhost backend called vhost-vdpa based on the vDPA interface.
>
> Vhost-vdpa usage:
>
>    qemu-system-x86_64 -cpu host -enable-kvm \
>      ......
>    -netdev type=vhost-vdpa,vhostdev=/dev/vhost-vdpa-id,id=vhost-vdpa0 \
>    -device virtio-net-pci,netdev=vhost-vdpa0,page-per-vq=on \
>
> Co-Authored-By: Lingshan zhu <lingshan.zhu@intel.com>
> Signed-off-by: Cindy Lu <lulu@redhat.com>
> ---
>   configure                         |  21 ++
>   hw/net/vhost_net-stub.c           |   5 +
>   hw/net/vhost_net.c                |  47 +++-
>   hw/virtio/Makefile.objs           |   1 +
>   hw/virtio/vhost-backend.c         |   5 +
>   hw/virtio/vhost-vdpa.c            | 399 ++++++++++++++++++++++++++++++
>   hw/virtio/vhost.c                 |  37 ++-
>   include/hw/virtio/vhost-backend.h |  10 +-
>   include/hw/virtio/vhost-vdpa.h    |  26 ++
>   include/hw/virtio/vhost.h         |   2 +
>   include/net/vhost_net.h           |   4 +-
>   qemu-options.hx                   |  15 ++
>   12 files changed, 566 insertions(+), 6 deletions(-)
>   create mode 100644 hw/virtio/vhost-vdpa.c
>   create mode 100644 include/hw/virtio/vhost-vdpa.h
>
> diff --git a/configure b/configure
> index 23b5e93752..53679ee57f 100755
> --- a/configure
> +++ b/configure
> @@ -1557,6 +1557,10 @@ for opt do
>     ;;
>     --enable-vhost-user) vhost_user="yes"
>     ;;
> +  --disable-vhost-vdpa) vhost_vdpa="no"
> +  ;;
> +  --enable-vhost-vdpa) vhost_vdpa="yes"
> +  ;;
>     --disable-vhost-kernel) vhost_kernel="no"
>     ;;
>     --enable-vhost-kernel) vhost_kernel="yes"
> @@ -1846,6 +1850,7 @@ disabled with --disable-FEATURE, default is enabled if available:
>     vhost-crypto    vhost-user-crypto backend support
>     vhost-kernel    vhost kernel backend support
>     vhost-user      vhost-user backend support
> +  vhost-vdpa      vhost-vdpa kernel backend support
>     spice           spice
>     rbd             rados block device (rbd)
>     libiscsi        iscsi support
> @@ -2336,6 +2341,10 @@ test "$vhost_user" = "" && vhost_user=yes
>   if test "$vhost_user" = "yes" && test "$mingw32" = "yes"; then
>     error_exit "vhost-user isn't available on win32"
>   fi
> +test "$vhost_vdpa" = "" && vhost_vdpa=$linux
> +if test "$vhost_vdpa" = "yes" && test "$linux" != "yes"; then
> +  error_exit "vhost-vdpa is only available on Linux"
> +fi
>   test "$vhost_kernel" = "" && vhost_kernel=$linux
>   if test "$vhost_kernel" = "yes" && test "$linux" != "yes"; then
>     error_exit "vhost-kernel is only available on Linux"
> @@ -2364,6 +2373,11 @@ test "$vhost_user_fs" = "" && vhost_user_fs=$vhost_user
>   if test "$vhost_user_fs" = "yes" && test "$vhost_user" = "no"; then
>     error_exit "--enable-vhost-user-fs requires --enable-vhost-user"
>   fi
> +#vhost-vdpa backends
> +test "$vhost_net_vdpa" = "" && vhost_net_vdpa=$vhost_vdpa
> +if test "$vhost_net_vdpa" = "yes" && test "$vhost_vdpa" = "no"; then
> +  error_exit "--enable-vhost-net-vdpa requires --enable-vhost-vdpa"
> +fi
>   
>   # OR the vhost-kernel and vhost-user values for simplicity
>   if test "$vhost_net" = ""; then
> @@ -6673,6 +6687,7 @@ echo "vhost-scsi support $vhost_scsi"
>   echo "vhost-vsock support $vhost_vsock"
>   echo "vhost-user support $vhost_user"
>   echo "vhost-user-fs support $vhost_user_fs"
> +echo "vhost-vdpa support $vhost_vdpa"
>   echo "Trace backends    $trace_backends"
>   if have_backend "simple"; then
>   echo "Trace output file $trace_file-<pid>"
> @@ -7170,6 +7185,9 @@ fi
>   if test "$vhost_net_user" = "yes" ; then
>     echo "CONFIG_VHOST_NET_USER=y" >> $config_host_mak
>   fi
> +if test "$vhost_net_vdpa" = "yes" ; then
> +  echo "CONFIG_VHOST_NET_VDPA=y" >> $config_host_mak
> +fi
>   if test "$vhost_crypto" = "yes" ; then
>     echo "CONFIG_VHOST_CRYPTO=y" >> $config_host_mak
>   fi
> @@ -7182,6 +7200,9 @@ fi
>   if test "$vhost_user" = "yes" ; then
>     echo "CONFIG_VHOST_USER=y" >> $config_host_mak
>   fi
> +if test "$vhost_vdpa" = "yes" ; then
> +  echo "CONFIG_VHOST_VDPA=y" >> $config_host_mak
> +fi
>   if test "$vhost_user_fs" = "yes" ; then
>     echo "CONFIG_VHOST_USER_FS=y" >> $config_host_mak
>   fi
> diff --git a/hw/net/vhost_net-stub.c b/hw/net/vhost_net-stub.c
> index 43e93e1a9a..ab77a92a7d 100644
> --- a/hw/net/vhost_net-stub.c
> +++ b/hw/net/vhost_net-stub.c
> @@ -94,3 +94,8 @@ int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
>   {
>       return 0;
>   }
> +int vhost_net_get_device_id(struct vhost_net *net, uint32_t * device_id)
> +{
> +    return 0;
> +}
> +
> diff --git a/hw/net/vhost_net.c b/hw/net/vhost_net.c
> index e2bc7de2eb..25045cff59 100644
> --- a/hw/net/vhost_net.c
> +++ b/hw/net/vhost_net.c
> @@ -17,8 +17,10 @@
>   #include "net/net.h"
>   #include "net/tap.h"
>   #include "net/vhost-user.h"
> +#include "net/vhost-vdpa.h"
>   
>   #include "standard-headers/linux/vhost_types.h"
> +#include "linux-headers/linux/vhost.h"
>   #include "hw/virtio/virtio-net.h"
>   #include "net/vhost_net.h"
>   #include "qemu/error-report.h"
> @@ -85,6 +87,30 @@ static const int user_feature_bits[] = {
>       VHOST_INVALID_FEATURE_BIT
>   };
>   
> +static const int vdpa_feature_bits[] = {
> +    VIRTIO_F_NOTIFY_ON_EMPTY,
> +    VIRTIO_RING_F_INDIRECT_DESC,
> +    VIRTIO_RING_F_EVENT_IDX,
> +    VIRTIO_F_ANY_LAYOUT,
> +    VIRTIO_F_VERSION_1,
> +    VIRTIO_NET_F_CSUM,
> +    VIRTIO_NET_F_GUEST_CSUM,
> +    VIRTIO_NET_F_GSO,
> +    VIRTIO_NET_F_GUEST_TSO4,
> +    VIRTIO_NET_F_GUEST_TSO6,
> +    VIRTIO_NET_F_GUEST_ECN,
> +    VIRTIO_NET_F_GUEST_UFO,
> +    VIRTIO_NET_F_HOST_TSO4,
> +    VIRTIO_NET_F_HOST_TSO6,
> +    VIRTIO_NET_F_HOST_ECN,
> +    VIRTIO_NET_F_HOST_UFO,
> +    VIRTIO_NET_F_MRG_RXBUF,
> +    VIRTIO_NET_F_MTU,
> +    VIRTIO_F_IOMMU_PLATFORM,
> +    VIRTIO_F_RING_PACKED,
> +    VIRTIO_NET_F_GUEST_ANNOUNCE,
> +    VHOST_INVALID_FEATURE_BIT
> +};


I think those feature bits should belong to net/vhost-vdpa.c, since it 
contains bits that are net specific.


>   static const int *vhost_net_get_feature_bits(struct vhost_net *net)
>   {
>       const int *feature_bits = 0;
> @@ -96,6 +122,9 @@ static const int *vhost_net_get_feature_bits(struct vhost_net *net)
>       case NET_CLIENT_DRIVER_VHOST_USER:
>           feature_bits = user_feature_bits;
>           break;
> +    case NET_CLIENT_DRIVER_VHOST_VDPA:
> +        feature_bits = vdpa_feature_bits;
> +        break;
>       default:
>           error_report("Feature bits not defined for this type: %d",
>                   net->nc->info->type);
> @@ -110,7 +139,10 @@ uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
>       return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
>               features);
>   }
> -
> +int vhost_net_get_device_id(struct vhost_net *net, uint32_t * device_id)
> +{
> +    return vhost_dev_get_device_id(&net->dev, device_id);
> +}
>   void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
>   {
>       net->dev.acked_features = net->dev.backend_features;
> @@ -337,6 +369,11 @@ int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
>       }
>   
>       for (i = 0; i < total_queues; i++) {
> +
> +        if (virtio_queue_enabled(dev, i)) {
> +            vhost_set_vring_ready(peer);
> +        }


So this may break vpda_sim since it call set_vq_ready() before 
set_vring_addr.

I think maybe it's better not introduce vhost_set_vring_ready() but 
enable virtqueue in vhost_vdpa_set_state() before setting DRIVER_OK.


> +
>           peer = qemu_get_peer(ncs, i);
>           r = vhost_net_start_one(get_vhost_net(peer), dev);
>   
> @@ -433,6 +470,12 @@ VHostNetState *get_vhost_net(NetClientState *nc)
>           vhost_net = vhost_user_get_vhost_net(nc);
>           assert(vhost_net);
>           break;
> +#endif
> +#ifdef CONFIG_VHOST_NET_VDPA
> +    case NET_CLIENT_DRIVER_VHOST_VDPA:
> +        vhost_net = vhost_vdpa_get_vhost_net(nc);
> +        assert(vhost_net);
> +        break;
>   #endif
>       default:
>           break;
> @@ -474,3 +517,5 @@ int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
>   
>       return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
>   }
> +
> +


One extra newline.


> diff --git a/hw/virtio/Makefile.objs b/hw/virtio/Makefile.objs
> index 4e4d39a0a4..6b1b1a5fce 100644
> --- a/hw/virtio/Makefile.objs
> +++ b/hw/virtio/Makefile.objs
> @@ -5,6 +5,7 @@ obj-y += virtio.o
>   obj-$(CONFIG_VHOST) += vhost.o vhost-backend.o
>   common-obj-$(call lnot,$(CONFIG_VHOST)) += vhost-stub.o
>   obj-$(CONFIG_VHOST_USER) += vhost-user.o
> +obj-$(CONFIG_VHOST_VDPA) += vhost-vdpa.o
>   
>   common-obj-$(CONFIG_VIRTIO_RNG) += virtio-rng.o
>   common-obj-$(CONFIG_VIRTIO_PCI) += virtio-pci.o
> diff --git a/hw/virtio/vhost-backend.c b/hw/virtio/vhost-backend.c
> index 42efb4967b..420341e8c5 100644
> --- a/hw/virtio/vhost-backend.c
> +++ b/hw/virtio/vhost-backend.c
> @@ -291,6 +291,11 @@ int vhost_set_backend_type(struct vhost_dev *dev, VhostBackendType backend_type)
>       case VHOST_BACKEND_TYPE_USER:
>           dev->vhost_ops = &user_ops;
>           break;
> +#endif
> +#ifdef CONFIG_VHOST_VDPA
> +    case VHOST_BACKEND_TYPE_VDPA:
> +        dev->vhost_ops = &vdpa_ops;
> +        break;
>   #endif
>       default:
>           error_report("Unknown vhost backend type");
> diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
> new file mode 100644
> index 0000000000..2d136a8565
> --- /dev/null
> +++ b/hw/virtio/vhost-vdpa.c
> @@ -0,0 +1,399 @@
> +/*
> + * vhost-vdpa
> + *
> + *  Copyright(c) 2017-2018 Intel Corporation.
> + *  Copyright(c) 2020 Red Hat, Inc.
> + *
> + * 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 "qemu/osdep.h"
> +#include <linux/vhost.h>
> +#include <linux/vfio.h>
> +#include <sys/eventfd.h>
> +#include <sys/ioctl.h>
> +#include "hw/virtio/vhost.h"
> +#include "hw/virtio/vhost-backend.h"
> +#include "hw/virtio/virtio-net.h"
> +#include "hw/virtio/vhost-vdpa.h"
> +#include "qemu/main-loop.h"
> +#include <linux/kvm.h>
> +#include "sysemu/kvm.h"
> +
> +
> +static bool vhost_vdpa_listener_skipped_section(MemoryRegionSection *section)
> +{
> +    return (!memory_region_is_ram(section->mr) &&
> +            !memory_region_is_iommu(section->mr)) ||
> +           /*
> +            * Sizing an enabled 64-bit BAR can cause spurious mappings to
> +            * addresses in the upper part of the 64-bit address space.  These
> +            * are never accessed by the CPU and beyond the address width of
> +            * some IOMMU hardware.  TODO: VDPA should tell us the IOMMU width.
> +            */
> +           section->offset_within_address_space & (1ULL << 63);
> +}
> +
> +static int vhost_vdpa_dma_map(struct vhost_vdpa *v, hwaddr iova, hwaddr size,
> +                              void *vaddr, bool readonly)
> +{
> +    struct vhost_msg_v2 msg;
> +    int fd = v->device_fd;
> +    int ret = 0;
> +
> +    msg.type =  v->msg_type;
> +    msg.iotlb.iova = iova;
> +    msg.iotlb.size = size;
> +    msg.iotlb.uaddr = (uint64_t)vaddr;
> +    msg.iotlb.perm = readonly ? VHOST_ACCESS_RO : VHOST_ACCESS_RW;
> +    msg.iotlb.type = VHOST_IOTLB_UPDATE;
> +
> +    if (write(fd, &msg, sizeof(msg)) != sizeof(msg)) {
> +        error_report("failed to write, fd=%d, errno=%d (%s)",
> +            fd, errno, strerror(errno));
> +        return -EIO ;
> +    }
> +
> +    return ret;
> +}
> +
> +static int vhost_vdpa_dma_unmap(struct vhost_vdpa *v, hwaddr iova,
> +                                hwaddr size)
> +{
> +    struct vhost_msg_v2 msg;
> +    int fd = v->device_fd;
> +    int ret = 0;
> +
> +    msg.type =  v->msg_type;
> +    msg.iotlb.iova = iova;
> +    msg.iotlb.size = size;
> +    msg.iotlb.type = VHOST_IOTLB_INVALIDATE;
> +
> +    if (write(fd, &msg, sizeof(msg)) != sizeof(msg)) {
> +        error_report("failed to write, fd=%d, errno=%d (%s)",
> +            fd, errno, strerror(errno));
> +        return -EIO ;
> +    }
> +
> +    return ret;
> +}
> +
> +static void vhost_vdpa_listener_region_add(MemoryListener *listener,
> +                                           MemoryRegionSection *section)
> +{
> +    struct vhost_vdpa *v = container_of(listener, struct vhost_vdpa, listener);
> +    hwaddr iova;
> +    Int128 llend, llsize;
> +    void *vaddr;
> +    int ret;
> +
> +    if (vhost_vdpa_listener_skipped_section(section)) {
> +        return;
> +    }
> +
> +    if (unlikely((section->offset_within_address_space & ~TARGET_PAGE_MASK) !=
> +                 (section->offset_within_region & ~TARGET_PAGE_MASK))) {
> +        error_report("%s received unaligned region", __func__);
> +        return;
> +    }
> +
> +    iova = TARGET_PAGE_ALIGN(section->offset_within_address_space);
> +    llend = int128_make64(section->offset_within_address_space);
> +    llend = int128_add(llend, section->size);
> +    llend = int128_and(llend, int128_exts64(TARGET_PAGE_MASK));
> +
> +    if (int128_ge(int128_make64(iova), llend)) {
> +        return;
> +    }
> +
> +    memory_region_ref(section->mr);
> +
> +    /* Here we assume that memory_region_is_ram(section->mr)==true */
> +
> +    vaddr = memory_region_get_ram_ptr(section->mr) +
> +            section->offset_within_region +
> +            (iova - section->offset_within_address_space);
> +
> +    llsize = int128_sub(llend, int128_make64(iova));
> +
> +    ret = vhost_vdpa_dma_map(v, iova, int128_get64(llsize),
> +                             vaddr, section->readonly);
> +    if (ret) {
> +        error_report("vhost vdpa map fail!");
> +        if (memory_region_is_ram_device(section->mr)) {
> +            /* Allow unexpected mappings not to be fatal for RAM devices */
> +            error_report("map ram fail!");
> +          return ;
> +        }
> +        goto fail;
> +    }
> +
> +    return;
> +
> +fail:
> +    if (memory_region_is_ram_device(section->mr)) {
> +        error_report("failed to vdpa_dma_map. pci p2p may not work");
> +        return;
> +
> +    }
> +    /*
> +     * On the initfn path, store the first error in the container so we
> +     * can gracefully fail.  Runtime, there's not much we can do other
> +     * than throw a hardware error.
> +     */
> +    error_report("vhost-vdpa: DMA mapping failed, unable to continue");
> +    return;
> +
> +}
> +
> +static void vhost_vdpa_listener_region_del(MemoryListener *listener,
> +                                           MemoryRegionSection *section)
> +{
> +    struct vhost_vdpa *v = container_of(listener, struct vhost_vdpa, listener);
> +    hwaddr iova;
> +    Int128 llend, llsize;
> +    int ret;
> +    bool try_unmap = true;
> +
> +    if (vhost_vdpa_listener_skipped_section(section)) {
> +        return;
> +    }
> +
> +    if (unlikely((section->offset_within_address_space & ~TARGET_PAGE_MASK) !=
> +                 (section->offset_within_region & ~TARGET_PAGE_MASK))) {
> +        error_report("%s received unaligned region", __func__);
> +        return;
> +    }
> +
> +    iova = TARGET_PAGE_ALIGN(section->offset_within_address_space);
> +    llend = int128_make64(section->offset_within_address_space);
> +    llend = int128_add(llend, section->size);
> +    llend = int128_and(llend, int128_exts64(TARGET_PAGE_MASK));
> +
> +    if (int128_ge(int128_make64(iova), llend)) {
> +        return;
> +    }
> +
> +    llsize = int128_sub(llend, int128_make64(iova));
> +
> +    if (try_unmap) {
> +        ret = vhost_vdpa_dma_unmap(v, iova, int128_get64(llsize));
> +        if (ret) {
> +            error_report("vhost_vdpa dma unmap error!");
> +        }
> +    }
> +
> +    memory_region_unref(section->mr);
> +}


newline is needed here.


> +/* Register a new memory listener, only to get diffs from qemu,
> + * this help to reduce the tricky codes in vhost
> + * (e.g generating diffs of two rbtree as usnic did).*/


This comment needs some improvement. How about:

/* IOTLB API is used by vhost-vpda which requires incremental updating 
of the mapping. So we can not use generic vhost memory listener which 
depends on the addnop(). */


> +static const MemoryListener vhost_vdpa_memory_listener = {
> +    .region_add = vhost_vdpa_listener_region_add,
> +    .region_del = vhost_vdpa_listener_region_del,
> +};
> +
> +static int vhost_vdpa_call(struct vhost_dev *dev, unsigned long int request,
> +                             void *arg)
> +{
> +    struct vhost_vdpa *v = dev->opaque;
> +    int fd = v->device_fd;
> +
> +    assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_VDPA);
> +
> +    return ioctl(fd, request, arg);
> +}
> +
> +static void vhost_vdpa_add_status(struct vhost_dev *dev, uint8_t status)
> +{
> +    uint8_t s;
> +
> +    if (vhost_vdpa_call(dev, VHOST_VDPA_GET_STATUS, &s)) {
> +        return;
> +    }
> +
> +    s |= status;
> +
> +    vhost_vdpa_call(dev, VHOST_VDPA_SET_STATUS, &s);
> +}
> +
> +static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque)
> +{
> +    struct vhost_vdpa *v;
> +
> +    assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_VDPA);
> +
> +    v = opaque;
> +    dev->opaque =  opaque ;
> +
> +    v->listener = vhost_vdpa_memory_listener;
> +    v->msg_type = VHOST_IOTLB_MSG_V2;
> +    memory_listener_register(&v->listener, &address_space_memory);


Let's move the memory listener register/unregister to 
vhost_vdpa_set_state(). Then we can avoid lots of unnecessary vhost 
IOTLB transactions before DRIVER_OK which vhost-vDPA doesn't care.


> +
> +    vhost_vdpa_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE |
> +                               VIRTIO_CONFIG_S_DRIVER);
> +
> +    return 0;
> +}
> +
> +static int vhost_vdpa_cleanup(struct vhost_dev *dev)
> +{
> +    struct vhost_vdpa *v;
> +    assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_VDPA);
> +
> +    v = dev->opaque;
> +    memory_listener_unregister(&v->listener);
> +
> +    dev->opaque = NULL;
> +    return 0;
> +}
> +
> +static int vhost_vdpa_memslots_limit(struct vhost_dev *dev)
> +{
> +    return INT_MAX;
> +}
> +
> +static int vhost_vdpa_set_mem_table(struct vhost_dev *dev,
> +                                    struct vhost_memory *mem)
> +{
> +
> +    if (mem->padding) {
> +        return -1;
> +    }
> +
> +    return 0;
> +}
> +
> +static int vhost_vdpa_set_features(struct vhost_dev *dev,
> +                                   uint64_t features)
> +{
> +    int ret = vhost_vdpa_call(dev, VHOST_SET_FEATURES, &features);
> +    uint8_t status = 0;
> +
> +    if (ret) {
> +        return ret;
> +    }
> +    vhost_vdpa_add_status(dev, VIRTIO_CONFIG_S_FEATURES_OK);
> +    vhost_vdpa_call(dev, VHOST_VDPA_GET_STATUS, &status);
> +
> +    return !(status & VIRTIO_CONFIG_S_FEATURES_OK);
> +}
> +
> +int vhost_vdpa_get_device_id(struct vhost_dev *dev,
> +                                   uint32_t *device_id)
> +{
> +    return vhost_vdpa_call(dev, VHOST_VDPA_GET_DEVICE_ID, device_id);
> +}
> +
> +static int vhost_vdpa_reset_device(struct vhost_dev *dev)
> +{
> +    uint8_t status = 0;
> +
> +    return vhost_vdpa_call(dev, VHOST_VDPA_SET_STATUS, &status);
> +}
> +
> +static int vhost_vdpa_get_vq_index(struct vhost_dev *dev, int idx)
> +{
> +    assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
> +
> +    return idx - dev->vq_index;
> +}
> +
> +static int vhost_vdpa_set_vring_ready(struct vhost_dev *dev)
> +{
> +    int i;
> +    for (i = 0; i < dev->nvqs; ++i) {
> +        struct vhost_vring_state state = {
> +            .index = dev->vq_index + i,
> +            .num = 1,
> +        };
> +        vhost_vdpa_call(dev, VHOST_VDPA_SET_VRING_ENABLE, &state);
> +    }
> +    return 0;
> +}
> +
> +static int vhost_vdpa_set_config(struct vhost_dev *dev, const uint8_t *data,
> +                                   uint32_t offset, uint32_t size,
> +                                   uint32_t flags)
> +{
> +    struct vhost_vdpa_config config;
> +    int ret;
> +    if ((size > VHOST_VDPA_MAX_CONFIG_SIZE) || (data == NULL)) {
> +        return -1;
> +    }
> +    memset(&config, 0, sizeof(struct vhost_vdpa_config));
> +    config.off = 0;
> +    config.len = size;
> +    memcpy(&config.buf, data, size);
> +    ret = vhost_vdpa_call(dev, VHOST_VDPA_SET_CONFIG, &config);
> +    return ret;
> +}
> +
> +static int vhost_vdpa_get_config(struct vhost_dev *dev, uint8_t *config,
> +                                   uint32_t config_len)
> +{
> +    struct vhost_vdpa_config v_config;
> +    int ret;
> +
> +    memset(&v_config, 0, sizeof(struct vhost_vdpa_config));
> +    if (config == NULL) {
> +        return -1;
> +    }
> +    ret = vhost_vdpa_call(dev, VHOST_VDPA_GET_CONFIG, &v_config);
> +    if ((v_config.len > config_len) || (v_config.len == 0)) {
> +        return -EINVAL;
> +    }
> +    memcpy(config, &v_config.buf, config_len);
> +    return ret;
> + }
> +
> +static int vhost_vdpa_set_state(struct vhost_dev *dev, bool started)


We probably need a better name, e.g vhost_vdpa_start()?


> +{
> +    if (started) {
> +        uint8_t status = 0;
> +
> +        vhost_vdpa_add_status(dev, VIRTIO_CONFIG_S_DRIVER_OK);
> +        vhost_vdpa_call(dev, VHOST_VDPA_GET_STATUS, &status);
> +
> +        return !(status & VIRTIO_CONFIG_S_DRIVER_OK);
> +    } else {
> +        vhost_vdpa_reset_device(dev);
> +        vhost_vdpa_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE |
> +                                   VIRTIO_CONFIG_S_DRIVER);
> +        return 0;
> +    }
> +}
> +
> +const VhostOps vdpa_ops = {
> +        .backend_type = VHOST_BACKEND_TYPE_VDPA,
> +        .vhost_backend_init = vhost_vdpa_init,
> +        .vhost_backend_cleanup = vhost_vdpa_cleanup,
> +        .vhost_set_log_base = vhost_kernel_set_log_base,
> +        .vhost_set_vring_addr = vhost_kernel_set_vring_addr,
> +        .vhost_set_vring_num = vhost_kernel_set_vring_num,
> +        .vhost_set_vring_base = vhost_kernel_set_vring_base,
> +        .vhost_get_vring_base = vhost_kernel_get_vring_base,
> +        .vhost_set_vring_kick = vhost_kernel_set_vring_kick,
> +        .vhost_set_vring_call = vhost_kernel_set_vring_call,
> +        .vhost_get_features = vhost_kernel_get_features,
> +        .vhost_set_owner = vhost_kernel_set_owner,
> +        .vhost_set_vring_endian = NULL,
> +        .vhost_backend_memslots_limit = vhost_vdpa_memslots_limit,
> +        .vhost_set_mem_table = vhost_vdpa_set_mem_table,
> +        .vhost_set_features = vhost_vdpa_set_features,
> +        .vhost_reset_device = vhost_vdpa_reset_device,
> +        .vhost_get_vq_index = vhost_vdpa_get_vq_index,
> +        .vhost_set_vring_ready = vhost_vdpa_set_vring_ready,
> +        .vhost_get_config  = vhost_vdpa_get_config,
> +        .vhost_set_config = vhost_vdpa_set_config,
> +        .vhost_requires_shm_log = NULL,
> +        .vhost_migration_done = NULL,
> +        .vhost_backend_can_merge = NULL,
> +        .vhost_net_set_mtu = NULL,
> +        .vhost_set_iotlb_callback = NULL,
> +        .vhost_send_device_iotlb_msg = NULL,
> +        .vhost_set_state = vhost_vdpa_set_state,


Since it only accept boolean parameter I guess vhost_dev_start() is better?


> +        .vhost_get_device_id = vhost_vdpa_get_device_id,
> +};
> diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
> index 01ebe12f28..b97aa02a4c 100644
> --- a/hw/virtio/vhost.c
> +++ b/hw/virtio/vhost.c
> @@ -756,6 +756,12 @@ static int vhost_virtqueue_set_addr(struct vhost_dev *dev,
>           .log_guest_addr = vq->used_phys,
>           .flags = enable_log ? (1 << VHOST_VRING_F_LOG) : 0,
>       };
> +    /*vDPA need to use the phys address here to set to hardware*/


Actually it's "IOVA" instead of "phys address".


> +    if (dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_VDPA) {
> +        addr.desc_user_addr = (uint64_t)(unsigned long)vq->desc_phys;
> +        addr.avail_user_addr = (uint64_t)(unsigned long)vq->avail_phys;
> +        addr.used_user_addr = (uint64_t)(unsigned long)vq->used_phys;
> +    }


Let's introduce a callback here instead of such hard-coded ones.


>       int r = dev->vhost_ops->vhost_set_vring_addr(dev, &addr);
>       if (r < 0) {
>           VHOST_OPS_DEBUG("vhost_set_vring_addr failed");
> @@ -1506,6 +1512,14 @@ int vhost_dev_set_config(struct vhost_dev *hdev, const uint8_t *data,
>       return -1;
>   }
>   
> +int vhost_dev_get_device_id(struct vhost_dev *hdev, uint32_t *device_id)
> +{
> +    assert(hdev->vhost_ops);
> +    if (hdev->vhost_ops->vhost_get_device_id) {
> +        return hdev->vhost_ops->vhost_get_device_id(hdev, device_id);
> +    }
> +    return -1;
> +}
>   void vhost_dev_set_config_notifier(struct vhost_dev *hdev,
>                                      const VhostDevConfigOps *ops)
>   {
> @@ -1661,7 +1675,13 @@ int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev)
>           }
>       }
>   
> -    if (vhost_dev_has_iommu(hdev)) {
> +    r = vhost_set_state(hdev, true);
> +    if (r) {
> +        goto fail_log;
> +    }


Please use a separate patch for introducing vhost_set_state().


> +
> +    if (vhost_dev_has_iommu(hdev) &&
> +        hdev->vhost_ops->vhost_set_iotlb_callback) {


A new patch for checking vhost_set_iotlb_callback(), or just implement 
it for vhost-vpda that just warn about IOTLB miss.


>           hdev->vhost_ops->vhost_set_iotlb_callback(hdev, true);
>   
>           /* Update used ring information for IOTLB to work correctly,
> @@ -1697,6 +1717,8 @@ void vhost_dev_stop(struct vhost_dev *hdev, VirtIODevice *vdev)
>       /* should only be called after backend is connected */
>       assert(hdev->vhost_ops);
>   
> +    vhost_set_state(hdev, false);
> +
>       for (i = 0; i < hdev->nvqs; ++i) {
>           vhost_virtqueue_stop(hdev,
>                                vdev,
> @@ -1705,7 +1727,9 @@ void vhost_dev_stop(struct vhost_dev *hdev, VirtIODevice *vdev)
>       }
>   
>       if (vhost_dev_has_iommu(hdev)) {
> -        hdev->vhost_ops->vhost_set_iotlb_callback(hdev, false);
> +        if (hdev->vhost_ops->vhost_set_iotlb_callback) {
> +            hdev->vhost_ops->vhost_set_iotlb_callback(hdev, false);
> +        }
>           memory_listener_unregister(&hdev->iommu_listener);
>       }
>       vhost_log_put(hdev, true);
> @@ -1722,3 +1746,12 @@ int vhost_net_set_backend(struct vhost_dev *hdev,
>   
>       return -1;
>   }
> +
> +int vhost_set_state(struct vhost_dev *hdev, bool started)
> +{
> +    if (hdev->vhost_ops->vhost_set_state) {
> +        return hdev->vhost_ops->vhost_set_state(hdev, started);
> +    }
> +
> +    return 0;
> +}
> diff --git a/include/hw/virtio/vhost-backend.h b/include/hw/virtio/vhost-backend.h
> index 300b59c172..1ebe3785cf 100644
> --- a/include/hw/virtio/vhost-backend.h
> +++ b/include/hw/virtio/vhost-backend.h
> @@ -17,7 +17,8 @@ typedef enum VhostBackendType {
>       VHOST_BACKEND_TYPE_NONE = 0,
>       VHOST_BACKEND_TYPE_KERNEL = 1,
>       VHOST_BACKEND_TYPE_USER = 2,
> -    VHOST_BACKEND_TYPE_MAX = 3,
> +    VHOST_BACKEND_TYPE_VDPA = 3,
> +    VHOST_BACKEND_TYPE_MAX = 4,
>   } VhostBackendType;
>   
>   typedef enum VhostSetConfigType {
> @@ -77,6 +78,7 @@ typedef int (*vhost_reset_device_op)(struct vhost_dev *dev);
>   typedef int (*vhost_get_vq_index_op)(struct vhost_dev *dev, int idx);
>   typedef int (*vhost_set_vring_enable_op)(struct vhost_dev *dev,
>                                            int enable);
> +typedef int (*vhost_set_vring_ready_op)(struct vhost_dev *dev);
>   typedef bool (*vhost_requires_shm_log_op)(struct vhost_dev *dev);
>   typedef int (*vhost_migration_done_op)(struct vhost_dev *dev,
>                                          char *mac_addr);
> @@ -112,6 +114,8 @@ typedef int (*vhost_get_inflight_fd_op)(struct vhost_dev *dev,
>   typedef int (*vhost_set_inflight_fd_op)(struct vhost_dev *dev,
>                                           struct vhost_inflight *inflight);
>   
> +typedef int (*vhost_set_state_op)(struct vhost_dev *dev, bool started);
> +typedef int (*vhost_get_device_id_op)(struct vhost_dev *dev, uint32_t *dev_id);
>   typedef struct VhostOps {
>       VhostBackendType backend_type;
>       vhost_backend_init vhost_backend_init;
> @@ -138,6 +142,7 @@ typedef struct VhostOps {
>       vhost_reset_device_op vhost_reset_device;
>       vhost_get_vq_index_op vhost_get_vq_index;
>       vhost_set_vring_enable_op vhost_set_vring_enable;
> +    vhost_set_vring_ready_op vhost_set_vring_ready;
>       vhost_requires_shm_log_op vhost_requires_shm_log;
>       vhost_migration_done_op vhost_migration_done;
>       vhost_backend_can_merge_op vhost_backend_can_merge;
> @@ -152,9 +157,12 @@ typedef struct VhostOps {
>       vhost_backend_mem_section_filter_op vhost_backend_mem_section_filter;
>       vhost_get_inflight_fd_op vhost_get_inflight_fd;
>       vhost_set_inflight_fd_op vhost_set_inflight_fd;
> +    vhost_set_state_op vhost_set_state;
> +    vhost_get_device_id_op vhost_get_device_id;
>   } VhostOps;
>   
>   extern const VhostOps user_ops;
> +extern const VhostOps vdpa_ops;
>   
>   int vhost_set_backend_type(struct vhost_dev *dev,
>                              VhostBackendType backend_type);
> diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h
> new file mode 100644
> index 0000000000..6455663388
> --- /dev/null
> +++ b/include/hw/virtio/vhost-vdpa.h
> @@ -0,0 +1,26 @@
> +/*
> + * vhost-vdpa.h
> + *
> + * Copyright(c) 2017-2018 Intel Corporation.
> + * Copyright(c) 2020 Red Hat, Inc.
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + *
> + */
> +
> +#ifndef HW_VIRTIO_VHOST_VDPA_H
> +#define HW_VIRTIO_VHOST_VDPA_H
> +
> +#include "hw/virtio/virtio.h"
> +
> +typedef struct vhost_vdpa {
> +    int device_fd;
> +    uint32_t msg_type;
> +    MemoryListener listener;
> +} VhostVDPA;
> +
> +extern AddressSpace address_space_memory;
> +extern int vhost_vdpa_get_device_id(struct vhost_dev *dev,
> +                                   uint32_t *device_id);
> +#endif
> diff --git a/include/hw/virtio/vhost.h b/include/hw/virtio/vhost.h
> index 085450c6f8..b682545f51 100644
> --- a/include/hw/virtio/vhost.h
> +++ b/include/hw/virtio/vhost.h
> @@ -124,6 +124,7 @@ int vhost_dev_get_config(struct vhost_dev *dev, uint8_t *config,
>                            uint32_t config_len);
>   int vhost_dev_set_config(struct vhost_dev *dev, const uint8_t *data,
>                            uint32_t offset, uint32_t size, uint32_t flags);
> +int vhost_dev_get_device_id(struct vhost_dev *hdev, uint32_t *device_id);
>   /* notifier callback in case vhost device config space changed
>    */
>   void vhost_dev_set_config_notifier(struct vhost_dev *dev,
> @@ -137,4 +138,5 @@ int vhost_dev_set_inflight(struct vhost_dev *dev,
>                              struct vhost_inflight *inflight);
>   int vhost_dev_get_inflight(struct vhost_dev *dev, uint16_t queue_size,
>                              struct vhost_inflight *inflight);
> +int vhost_set_state(struct vhost_dev *dev, bool started);
>   #endif
> diff --git a/include/net/vhost_net.h b/include/net/vhost_net.h
> index 8a6f208189..56e67fe164 100644
> --- a/include/net/vhost_net.h
> +++ b/include/net/vhost_net.h
> @@ -40,5 +40,5 @@ int vhost_set_vring_ready(NetClientState *nc);
>   uint64_t vhost_net_get_acked_features(VHostNetState *net);
>   
>   int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu);
> -
> -#endif
> +int vhost_net_get_device_id(struct vhost_net *net, uint32_t *device_id);


This should belong to vhost-vdpa header.

Thanks


> +endif
> diff --git a/qemu-options.hx b/qemu-options.hx
> index 292d4e7c0c..c19e10ce9c 100644
> --- a/qemu-options.hx
> +++ b/qemu-options.hx
> @@ -2409,6 +2409,10 @@ DEF("netdev", HAS_ARG, QEMU_OPTION_netdev,
>   #ifdef CONFIG_POSIX
>       "-netdev vhost-user,id=str,chardev=dev[,vhostforce=on|off]\n"
>       "                configure a vhost-user network, backed by a chardev 'dev'\n"
> +#endif
> +#ifdef CONFIG_POSIX
> +    "-netdev vhost-vdpa,id=str,vhostdev=/path/to/dev\n"
> +    "                configure a vhost-vdpa network,Establish a vhost-vdpa netdev\n"
>   #endif
>       "-netdev hubport,id=str,hubid=n[,netdev=nd]\n"
>       "                configure a hub port on the hub with ID 'n'\n", QEMU_ARCH_ALL)
> @@ -2428,6 +2432,9 @@ DEF("nic", HAS_ARG, QEMU_OPTION_nic,
>   #endif
>   #ifdef CONFIG_POSIX
>       "vhost-user|"
> +#endif
> +#ifdef CONFIG_POSIX
> +    "vhost-vdpa|"
>   #endif
>       "socket][,option][,...][mac=macaddr]\n"
>       "                initialize an on-board / default host NIC (using MAC address\n"
> @@ -2896,6 +2903,14 @@ SRST
>       hubport to another netdev with ID nd by using the ``netdev=nd``
>       option.
>   
> +``-netdev vhost-vdpa,vhostdev=/path/to/dev ``
> +    Establish a vhost-vdpa netdev.
> +
> +    vDPA device is a device that uses a datapath which complies with
> +    the virtio specifications with a vendor specific control path.
> +    vDPA devices can be both physically located on the hardware or
> +    emulated by software.
> +
>   ``-net nic[,netdev=nd][,macaddr=mac][,model=type] [,name=name][,addr=addr][,vectors=v]``
>       Legacy option to configure or create an on-board (or machine
>       default) Network Interface Card(NIC) and connect it either to the