Re: [PATCH RFC v2 08/13] hw/virtio/vhost-user: send isolation regions to device

On 2026/08/18 14:12, Connor Kite wrote:
> Adds features to fill a vhost_user_set_mem_table message with the
> addresses of isolation memory regions corresponding to bounce buffers
> and vrings.
> 
> Signed-off-by: Connor Kite <[email protected]>
> ---
>   hw/virtio/vhost-user.c | 118 ++++++++++++++++++++++++++++++++++++++++++++++++-
>   1 file changed, 117 insertions(+), 1 deletion(-)
> 
> diff --git a/hw/virtio/vhost-user.c b/hw/virtio/vhost-user.c
> index 7e9233e174..e1e5cba53d 100644
> --- a/hw/virtio/vhost-user.c
> +++ b/hw/virtio/vhost-user.c
> @@ -1140,7 +1140,77 @@ static void cleanup_isolation_regions(struct vhost_dev *dev)
>       }
>   }
>   
> -__attribute__((unused))
> +typedef struct {
> +    VhostUserMsg *msg;
> +    struct vhost_user *u;
> +    int *fds;
> +    size_t fds_size;
> +    uint64_t vring_iova;
> +    size_t vring_size;
> +    bool vring_node_visited;
> +    size_t *idx;
> +} IOVATreeTraversalArgs;
> +
> +static gboolean vhost_user_fill_msg_reg_from_tree(gpointer key,
> +                                                  gpointer value,
> +                                                  gpointer data)
> +{
> +    IOVATreeTraversalArgs *args = data;
> +    struct vhost_user *u;
> +    VhostUserMsg *msg = args->msg;
> +    DMAMap *map = key;
> +    uint64_t offset;
> +
> +    assert(args && key && value);
> +    if (!args->vring_node_visited) {
> +        args->vring_node_visited = true;
> +        return false;
> +    }
> +
> +    u = args->u;
> +    assert(*args->idx < args->fds_size);
> +
> +    args->fds[*args->idx] = args->u->iso_mem_ctx.fd;
> +
> +    /*
> +     * If the number of regions is fixed, it would be wasteful to use one for
> +     * only the vrings. The first vhost_iova_tree element is always
> +     * reserved for the vrings, so we can simply combine the first and second
> +     * elements, which are contiguous in IOVA space, when sending regions to
> +     * the backend.
> +     */
> +    if (*args->idx == 0) {
> +        offset = u->iso_mem_ctx.vring_hva_addr - u->iso_mem_ctx.shared_mem_addr;
> +        msg->payload.memory.regions[*args->idx].userspace_addr =
> +            args->vring_iova;
> +        /*
> +         * The size from the iova tree is inclusive, so 1 is added to it.
> +         * args->vring_size is exclusive, so no addition is required.
> +         */
> +        msg->payload.memory.regions[*args->idx].memory_size =
> +            args->vring_size + map->size + 1;
> +        msg->payload.memory.regions[*args->idx].guest_phys_addr =
> +            args->vring_iova;
> +        msg->payload.memory.regions[*args->idx].mmap_offset = offset;
> +    } else {
> +        /* Use 128 bit operation in unlikely case of negative iso_iova_offset */
> +        offset = int128_get64(int128_add(int128_make64(map->iova),
> +                                        u->iso_mem_ctx.iso_iova_offset)) -
> +                (uint64_t)u->iso_mem_ctx.shared_mem_addr;
> +
> +        msg->payload.memory.regions[*args->idx].userspace_addr = map->iova;
> +        msg->payload.memory.regions[*args->idx].memory_size = map->size + 1;
> +        msg->payload.memory.regions[*args->idx].guest_phys_addr = map->iova;
> +        msg->payload.memory.regions[*args->idx].mmap_offset = offset;
> +    }
> +
> +    assert(offset + msg->payload.memory.regions[*args->idx].memory_size <=
> +        u->iso_mem_ctx.size);
> +    (*args->idx)++;
> +
> +    return false;
> +}
> +
>   static int init_isolation_regions(struct vhost_dev *dev,
>                                     VhostUserMsg *msg,
>                                     int *fds, size_t *fd_num)
> @@ -1159,6 +1229,7 @@ static int init_isolation_regions(struct vhost_dev *dev,
>       DMAMap *map;
>       DMAMap vring_map;
>       int r;
> +    IOVATreeTraversalArgs trav_args;
>   
>       msg->hdr.request = VHOST_USER_SET_MEM_TABLE;
>   
> @@ -1244,6 +1315,27 @@ static int init_isolation_regions(struct vhost_dev *dev,
>           }
>       }
>   
> +    *fd_num = 0;
> +    trav_args.idx = fd_num;
> +    trav_args.fds = fds;
> +    trav_args.msg = msg;
> +    trav_args.u = u;
> +    trav_args.vring_iova = vring_map.iova;
> +    trav_args.vring_size = total_vring_size;
> +    trav_args.fds_size = nregions;
> +    trav_args.vring_node_visited = false;
> +
> +    vhost_iova_tree_foreach(u->iso_mem_ctx.tree,
> +                            vhost_user_fill_msg_reg_from_tree, &trav_args);
> +
> +    msg->payload.memory.nregions = *fd_num;
> +
> +    assert(*fd_num == nregions);
> +
> +    msg->hdr.size = sizeof(msg->payload.memory.nregions);
> +    msg->hdr.size += sizeof(msg->payload.memory.padding);
> +    msg->hdr.size += *fd_num * sizeof(VhostUserMemoryRegion);
> +
>       return 0;
>   }
>   
> @@ -1251,6 +1343,7 @@ static int vhost_user_set_mem_table(struct vhost_dev *dev,
>                                       struct vhost_memory *mem)
>   {
>       struct vhost_user *u = dev->opaque;
> +    bool memory_isolation = u->user->memory_isolation;
>       int fds[VHOST_MEMORY_BASELINE_NREGIONS];
>       size_t fd_num = 0;
>       bool do_postcopy = u->postcopy_listen && u->postcopy_fd.handler;
> @@ -1262,6 +1355,11 @@ static int vhost_user_set_mem_table(struct vhost_dev *dev,
>       int ret;
>   
>       if (do_postcopy) {
> +        /* Postcopy is not supported with memory isolation yet */
> +        if (memory_isolation) {
> +            return -1;
> +        }
> +
>           /*
>            * Postcopy has enough differences that it's best done in it's own
>            * version
> @@ -1278,6 +1376,24 @@ static int vhost_user_set_mem_table(struct vhost_dev *dev,
>           msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
>       }
>   
> +    if (memory_isolation) {
> +        ret = init_isolation_regions(dev, &msg, fds, &fd_num);
> +        if (ret < 0) {
> +            return ret;
> +        }
> +
> +        ret = vhost_user_write(dev, &msg, fds, fd_num);
> +        if (ret < 0) {
> +            return ret;
> +        }
> +
> +        if (reply_supported) {
> +            return process_message_reply(dev, &msg);
> +        }
> +
> +        return 0;
> +    }
> +

This bypasses vhost_user_fill_set_mem_table_msg() below and suppresses 
the error saying: "Failed initializing vhost-user memory map, consider 
using -object memory-backend-file share=on"

However, vhost-user doesn't work even with memory isolation in such a 
case because vhost_user_no_private_memslots() always returns true and 
that makes vhost_section() filter private memory out. This restriction 
is unnecessary with memory isolation and can be removed instead of 
restoring the error.

This also lacks handling of VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS, 
and will overflow fds and VhostUserMemory::regions when 
VHOST_USER_GET_MAX_MEM_SLOTS is larger than VHOST_MEMORY_BASELINE_NREGIONS.

Regards,
Akihiko Odaki

>       if (config_mem_slots) {
>           ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, false);
>           if (ret < 0) {
>