Re: [Qemu-devel] [PATCH v2 for-4.0 2/7] vhost-user: Support providing shared memory to backend

[Michael S. Tsirkin]

On Tue, Dec 18, 2018 at 05:59:57PM +0800, address@hidden wrote:
> From: Xie Yongji <address@hidden>
> 
> This patch introduces two new messages VHOST_USER_GET_SHM_SIZE
> and VHOST_USER_SET_SHM_FD to support providing shared
> memory to backend.
> 
> Firstly, qemu uses VHOST_USER_GET_SHM_SIZE to get the
> required size of shared memory from backend. Then, qemu
> allocates memory and sends them back to backend through
> VHOST_USER_SET_SHM_FD.
> 
> Note that the shared memory should be used to record
> inflight I/O by backend. Qemu will clear it when vm reset.

An interesting design choice. Why not let the backend clear it
on start?

> 
> Signed-off-by: Xie Yongji <address@hidden>
> Signed-off-by: Chai Wen <address@hidden>
> Signed-off-by: Zhang Yu <address@hidden>
> ---
>  docs/interop/vhost-user.txt       |  41 +++++++++++
>  hw/virtio/vhost-user.c            |  86 ++++++++++++++++++++++
>  hw/virtio/vhost.c                 | 117 ++++++++++++++++++++++++++++++
>  include/hw/virtio/vhost-backend.h |   9 +++
>  include/hw/virtio/vhost.h         |  19 +++++
>  5 files changed, 272 insertions(+)
> 
> diff --git a/docs/interop/vhost-user.txt b/docs/interop/vhost-user.txt
> index c2194711d9..5ee9c28ab0 100644
> --- a/docs/interop/vhost-user.txt
> +++ b/docs/interop/vhost-user.txt
> @@ -142,6 +142,19 @@ Depending on the request type, payload can be:
>     Offset: a 64-bit offset of this area from the start of the
>         supplied file descriptor
>  
> + * Shm description
> +   -----------------------------------
> +   | mmap_size | mmap_offset | dev_size | vq_size | align | version |
> +   -----------------------------------
> +
> +   Mmap_size: a 64-bit size of the shared memory
> +   Mmap_offset: a 64-bit offset of the shared memory from the start
> +                of the supplied file descriptor
> +   Dev_size: a 32-bit size of device region in shared memory
> +   Vq_size: a 32-bit size of each virtqueue region in shared memory
> +   Align: a 32-bit align of each region in shared memory
> +   Version: a 32-bit version of this shared memory
> +
>  In QEMU the vhost-user message is implemented with the following struct:
>  
>  typedef struct VhostUserMsg {
> @@ -157,6 +170,7 @@ typedef struct VhostUserMsg {
>          struct vhost_iotlb_msg iotlb;
>          VhostUserConfig config;
>          VhostUserVringArea area;
> +        VhostUserShm shm;
>      };
>  } QEMU_PACKED VhostUserMsg;
>  
> @@ -175,6 +189,7 @@ the ones that do:
>   * VHOST_USER_GET_PROTOCOL_FEATURES
>   * VHOST_USER_GET_VRING_BASE
>   * VHOST_USER_SET_LOG_BASE (if VHOST_USER_PROTOCOL_F_LOG_SHMFD)
> + * VHOST_USER_GET_SHM_SIZE (if VHOST_USER_PROTOCOL_F_SLAVE_SHMFD)
>  
>  [ Also see the section on REPLY_ACK protocol extension. ]
>  
> @@ -188,6 +203,7 @@ in the ancillary data:
>   * VHOST_USER_SET_VRING_CALL
>   * VHOST_USER_SET_VRING_ERR
>   * VHOST_USER_SET_SLAVE_REQ_FD
> + * VHOST_USER_SET_SHM_FD (if VHOST_USER_PROTOCOL_F_SLAVE_SHMFD)
>  
>  If Master is unable to send the full message or receives a wrong reply it 
> will
>  close the connection. An optional reconnection mechanism can be implemented.
> @@ -397,6 +413,7 @@ Protocol features
>  #define VHOST_USER_PROTOCOL_F_CONFIG         9
>  #define VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD  10
>  #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER  11
> +#define VHOST_USER_PROTOCOL_F_SLAVE_SHMFD 12
>  
>  Master message types
>  --------------------
> @@ -761,6 +778,30 @@ Master message types
>        was previously sent.
>        The value returned is an error indication; 0 is success.
>  
> + * VHOST_USER_GET_SHM_SIZE
> +      Id: 31
> +      Equivalent ioctl: N/A
> +      Master payload: shm description
> +
> +      When VHOST_USER_PROTOCOL_F_SLAVE_SHMFD protocol feature has been
> +      successfully negotiated, master need to provide a shared memory to
> +      slave. This message is used by master to get required size from slave.
> +      The shared memory contains one region for device and several regions
> +      for virtqueue. The size of those two kinds of regions is specified
> +      by dev_size field and vq_size filed. The align field specify the 
> alignment
> +      of those regions.
> +
> + * VHOST_USER_SET_SHM_FD
> +      Id: 32
> +      Equivalent ioctl: N/A
> +      Master payload: shm description
> +
> +      When VHOST_USER_PROTOCOL_F_SLAVE_SHMFD protocol feature has been
> +      successfully negotiated, master uses this message to set shared memory
> +      for slave. The memory fd is passed in the ancillary data. The shared
> +      memory should be used to record inflight I/O by slave. And master will
> +      clear it when vm reset.
> +
>  Slave message types
>  -------------------
>  
> diff --git a/hw/virtio/vhost-user.c b/hw/virtio/vhost-user.c
> index e09bed0e4a..8cdf3b5121 100644
> --- a/hw/virtio/vhost-user.c
> +++ b/hw/virtio/vhost-user.c
> @@ -52,6 +52,7 @@ enum VhostUserProtocolFeature {
>      VHOST_USER_PROTOCOL_F_CONFIG = 9,
>      VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD = 10,
>      VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 11,
> +    VHOST_USER_PROTOCOL_F_SLAVE_SHMFD = 12,
>      VHOST_USER_PROTOCOL_F_MAX
>  };
>  
> @@ -89,6 +90,8 @@ typedef enum VhostUserRequest {
>      VHOST_USER_POSTCOPY_ADVISE  = 28,
>      VHOST_USER_POSTCOPY_LISTEN  = 29,
>      VHOST_USER_POSTCOPY_END     = 30,
> +    VHOST_USER_GET_SHM_SIZE = 31,
> +    VHOST_USER_SET_SHM_FD = 32,
>      VHOST_USER_MAX
>  } VhostUserRequest;
>  
> @@ -147,6 +150,15 @@ typedef struct VhostUserVringArea {
>      uint64_t offset;
>  } VhostUserVringArea;
>  
> +typedef struct VhostUserShm {
> +    uint64_t mmap_size;
> +    uint64_t mmap_offset;
> +    uint32_t dev_size;
> +    uint32_t vq_size;
> +    uint32_t align;
> +    uint32_t version;
> +} VhostUserShm;
> +
>  typedef struct {
>      VhostUserRequest request;
>  
> @@ -169,6 +181,7 @@ typedef union {
>          VhostUserConfig config;
>          VhostUserCryptoSession session;
>          VhostUserVringArea area;
> +        VhostUserShm shm;
>  } VhostUserPayload;
>  
>  typedef struct VhostUserMsg {
> @@ -1739,6 +1752,77 @@ static bool vhost_user_mem_section_filter(struct 
> vhost_dev *dev,
>      return result;
>  }
>  
> +static int vhost_user_get_shm_size(struct vhost_dev *dev,
> +                                   struct vhost_shm *shm)
> +{
> +    VhostUserMsg msg = {
> +        .hdr.request = VHOST_USER_GET_SHM_SIZE,
> +        .hdr.flags = VHOST_USER_VERSION,
> +        .hdr.size = sizeof(msg.payload.shm),
> +    };
> +
> +    if (!virtio_has_feature(dev->protocol_features,
> +                            VHOST_USER_PROTOCOL_F_SLAVE_SHMFD)) {
> +        shm->dev_size = 0;
> +        shm->vq_size = 0;
> +        return 0;
> +    }
> +
> +    if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
> +        return -1;
> +    }
> +
> +    if (vhost_user_read(dev, &msg) < 0) {
> +        return -1;
> +    }
> +
> +    if (msg.hdr.request != VHOST_USER_GET_SHM_SIZE) {
> +        error_report("Received unexpected msg type. "
> +                     "Expected %d received %d",
> +                     VHOST_USER_GET_SHM_SIZE, msg.hdr.request);
> +        return -1;
> +    }
> +
> +    if (msg.hdr.size != sizeof(msg.payload.shm)) {
> +        error_report("Received bad msg size.");
> +        return -1;
> +    }
> +
> +    shm->dev_size = msg.payload.shm.dev_size;
> +    shm->vq_size = msg.payload.shm.vq_size;
> +    shm->align = msg.payload.shm.align;
> +    shm->version = msg.payload.shm.version;
> +
> +    return 0;
> +}
> +
> +static int vhost_user_set_shm_fd(struct vhost_dev *dev,
> +                                 struct vhost_shm *shm)
> +{
> +    VhostUserMsg msg = {
> +        .hdr.request = VHOST_USER_SET_SHM_FD,
> +        .hdr.flags = VHOST_USER_VERSION,
> +        .payload.shm.mmap_size = shm->mmap_size,
> +        .payload.shm.mmap_offset = 0,
> +        .payload.shm.dev_size = shm->dev_size,
> +        .payload.shm.vq_size = shm->vq_size,
> +        .payload.shm.align = shm->align,
> +        .payload.shm.version = shm->version,
> +        .hdr.size = sizeof(msg.payload.shm),
> +    };
> +
> +    if (!virtio_has_feature(dev->protocol_features,
> +                            VHOST_USER_PROTOCOL_F_SLAVE_SHMFD)) {
> +        return 0;
> +    }
> +
> +    if (vhost_user_write(dev, &msg, &shm->fd, 1) < 0) {
> +        return -1;
> +    }
> +
> +    return 0;
> +}
> +
>  VhostUserState *vhost_user_init(void)
>  {
>      VhostUserState *user = g_new0(struct VhostUserState, 1);
> @@ -1790,4 +1874,6 @@ const VhostOps user_ops = {
>          .vhost_crypto_create_session = vhost_user_crypto_create_session,
>          .vhost_crypto_close_session = vhost_user_crypto_close_session,
>          .vhost_backend_mem_section_filter = vhost_user_mem_section_filter,
> +        .vhost_get_shm_size = vhost_user_get_shm_size,
> +        .vhost_set_shm_fd = vhost_user_set_shm_fd,
>  };
> diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
> index 569c4053ea..7a38fed50f 100644
> --- a/hw/virtio/vhost.c
> +++ b/hw/virtio/vhost.c
> @@ -1481,6 +1481,123 @@ void vhost_dev_set_config_notifier(struct vhost_dev 
> *hdev,
>      hdev->config_ops = ops;
>  }
>  
> +void vhost_dev_reset_shm(struct vhost_shm *shm)
> +{
> +    if (shm->addr) {
> +        memset(shm->addr, 0, shm->mmap_size);
> +    }
> +}
> +
> +void vhost_dev_free_shm(struct vhost_shm *shm)
> +{
> +    if (shm->addr) {
> +        qemu_memfd_free(shm->addr, shm->mmap_size, shm->fd);
> +        shm->addr = NULL;
> +        shm->fd = -1;
> +    }
> +}
> +
> +int vhost_dev_alloc_shm(struct vhost_shm *shm)
> +{
> +    Error *err = NULL;
> +    int fd = -1;
> +    void *addr = qemu_memfd_alloc("vhost-shm", shm->mmap_size,
> +                                  F_SEAL_GROW | F_SEAL_SHRINK | F_SEAL_SEAL,
> +                                  &fd, &err);
> +    if (err) {
> +        error_report_err(err);
> +        return -1;
> +    }
> +
> +    shm->addr = addr;
> +    shm->fd = fd;
> +
> +    return 0;
> +}
> +
> +void vhost_dev_save_shm(struct vhost_shm *shm, QEMUFile *f)
> +{
> +    if (shm->addr) {
> +        qemu_put_be64(f, shm->mmap_size);
> +        qemu_put_be32(f, shm->dev_size);
> +        qemu_put_be32(f, shm->vq_size);
> +        qemu_put_be32(f, shm->align);
> +        qemu_put_be32(f, shm->version);
> +        qemu_put_buffer(f, shm->addr, shm->mmap_size);
> +    } else {
> +        qemu_put_be64(f, 0);
> +    }
> +}
> +
> +int vhost_dev_load_shm(struct vhost_shm *shm, QEMUFile *f)
> +{
> +    uint64_t mmap_size;
> +
> +    mmap_size = qemu_get_be64(f);
> +    if (!mmap_size) {
> +        return 0;
> +    }
> +
> +    vhost_dev_free_shm(shm);
> +
> +    shm->mmap_size = mmap_size;
> +    shm->dev_size = qemu_get_be32(f);
> +    shm->vq_size = qemu_get_be32(f);
> +    shm->align = qemu_get_be32(f);
> +    shm->version = qemu_get_be32(f);
> +
> +    if (vhost_dev_alloc_shm(shm)) {
> +        return -ENOMEM;
> +    }
> +
> +    qemu_get_buffer(f, shm->addr, mmap_size);
> +
> +    return 0;
> +}
> +
> +int vhost_dev_set_shm(struct vhost_dev *dev, struct vhost_shm *shm)
> +{
> +    int r;
> +
> +    if (dev->vhost_ops->vhost_set_shm_fd && shm->addr) {
> +        r = dev->vhost_ops->vhost_set_shm_fd(dev, shm);
> +        if (r) {
> +            VHOST_OPS_DEBUG("vhost_set_vring_shm_fd failed");
> +            return -errno;
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +int vhost_dev_init_shm(struct vhost_dev *dev, struct vhost_shm *shm)
> +{
> +    int r;
> +
> +    if (dev->vhost_ops->vhost_get_shm_size) {
> +        r = dev->vhost_ops->vhost_get_shm_size(dev, shm);
> +        if (r) {
> +            VHOST_OPS_DEBUG("vhost_get_vring_shm_size failed");
> +            return -errno;
> +        }
> +
> +        if (!shm->dev_size && !shm->vq_size) {
> +            return 0;
> +        }
> +
> +        shm->mmap_size = QEMU_ALIGN_UP(shm->dev_size, shm->align) +
> +                         dev->nvqs * QEMU_ALIGN_UP(shm->vq_size, shm->align);
> +
> +        if (vhost_dev_alloc_shm(shm)) {
> +            return -ENOMEM;
> +        }
> +
> +        vhost_dev_reset_shm(shm);
> +    }
> +
> +    return 0;
> +}
> +
>  /* Host notifiers must be enabled at this point. */
>  int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev)
>  {
> diff --git a/include/hw/virtio/vhost-backend.h 
> b/include/hw/virtio/vhost-backend.h
> index 81283ec50f..4e7f13c9e9 100644
> --- a/include/hw/virtio/vhost-backend.h
> +++ b/include/hw/virtio/vhost-backend.h
> @@ -25,6 +25,7 @@ typedef enum VhostSetConfigType {
>      VHOST_SET_CONFIG_TYPE_MIGRATION = 1,
>  } VhostSetConfigType;
>  
> +struct vhost_shm;
>  struct vhost_dev;
>  struct vhost_log;
>  struct vhost_memory;
> @@ -104,6 +105,12 @@ typedef int (*vhost_crypto_close_session_op)(struct 
> vhost_dev *dev,
>  typedef bool (*vhost_backend_mem_section_filter_op)(struct vhost_dev *dev,
>                                                  MemoryRegionSection 
> *section);
>  
> +typedef int (*vhost_get_shm_size_op)(struct vhost_dev *dev,
> +                                     struct vhost_shm *shm);
> +
> +typedef int (*vhost_set_shm_fd_op)(struct vhost_dev *dev,
> +                                   struct vhost_shm *shm);
> +
>  typedef struct VhostOps {
>      VhostBackendType backend_type;
>      vhost_backend_init vhost_backend_init;
> @@ -142,6 +149,8 @@ typedef struct VhostOps {
>      vhost_crypto_create_session_op vhost_crypto_create_session;
>      vhost_crypto_close_session_op vhost_crypto_close_session;
>      vhost_backend_mem_section_filter_op vhost_backend_mem_section_filter;
> +    vhost_get_shm_size_op vhost_get_shm_size;
> +    vhost_set_shm_fd_op vhost_set_shm_fd;
>  } VhostOps;
>  
>  extern const VhostOps user_ops;
> diff --git a/include/hw/virtio/vhost.h b/include/hw/virtio/vhost.h
> index a7f449fa87..b6e3d6ab56 100644
> --- a/include/hw/virtio/vhost.h
> +++ b/include/hw/virtio/vhost.h
> @@ -7,6 +7,17 @@
>  #include "exec/memory.h"
>  
>  /* Generic structures common for any vhost based device. */
> +
> +struct vhost_shm {
> +    void *addr;
> +    uint64_t mmap_size;
> +    uint32_t dev_size;
> +    uint32_t vq_size;
> +    uint32_t align;
> +    uint32_t version;
> +    int fd;
> +};
> +
>  struct vhost_virtqueue {
>      int kick;
>      int call;
> @@ -120,4 +131,12 @@ int vhost_dev_set_config(struct vhost_dev *dev, const 
> uint8_t *data,
>   */
>  void vhost_dev_set_config_notifier(struct vhost_dev *dev,
>                                     const VhostDevConfigOps *ops);
> +
> +void vhost_dev_reset_shm(struct vhost_shm *shm);
> +void vhost_dev_free_shm(struct vhost_shm *shm);
> +int vhost_dev_alloc_shm(struct vhost_shm *shm);
> +void vhost_dev_save_shm(struct vhost_shm *shm, QEMUFile *f);
> +int vhost_dev_load_shm(struct vhost_shm *shm, QEMUFile *f);
> +int vhost_dev_set_shm(struct vhost_dev *dev, struct vhost_shm *shm);
> +int vhost_dev_init_shm(struct vhost_dev *dev, struct vhost_shm *shm);
>  #endif
> -- 
> 2.17.1