diff --git a/devicemodel/hw/pci/virtio/vhost.c b/devicemodel/hw/pci/virtio/vhost.c
index 8dfc59d47..849e91856 100644
--- a/devicemodel/hw/pci/virtio/vhost.c
+++ b/devicemodel/hw/pci/virtio/vhost.c
@@ -153,12 +153,132 @@ vhost_eventfd_test_and_clear(int fd)
 	return -1;
 }
 
+static void
+vhost_vq_notify(int fd __attribute__((unused)),
+		enum ev_type t __attribute__((unused)),
+		void *arg)
+{
+	struct vhost_vq *vq = arg;
+	struct virtio_vq_info *vqi;
+	struct vhost_dev *vdev;
+
+	vdev = vq->dev;
+	vqi = &vdev->base->queues[vdev->vq_idx + vq->idx];
+	vq_interrupt(vdev->base, vqi);
+}
+
 static int
 vhost_vq_register_eventfd(struct vhost_dev *vdev,
 			  int idx, bool is_register)
 {
-	/* to be implemented */
-	return -1;
+	struct acrn_ioeventfd ioeventfd = {0};
+	struct acrn_irqfd irqfd = {0};
+	struct virtio_base *base;
+	struct vhost_vq *vq;
+	struct virtio_vq_info *vqi;
+	struct pcibar *bar;
+	int rc = -1;
+
+	/* this interface is called only by vhost_vq_start,
+	 * parameters have been checked there
+	 */
+	base = vdev->base;
+	vqi = &vdev->base->queues[vdev->vq_idx + idx];
+	vq = &vdev->vqs[idx];
+
+	if (!is_register) {
+		ioeventfd.flags = ACRN_IOEVENTFD_FLAG_DEASSIGN;
+		irqfd.flags = ACRN_IRQFD_FLAG_DEASSIGN;
+	}
+
+	/* register ioeventfd for kick */
+	if (base->device_caps & ACRN_VIRTIO_F_VERSION_1) {
+		/*
+		 * in the current implementation, if virtio 1.0 with pio
+		 * notity, its bar idx should be set to non-zero
+		 */
+		if (base->modern_pio_bar_idx) {
+			bar = &vdev->base->dev->bar[base->modern_pio_bar_idx];
+			ioeventfd.data = vdev->vq_idx + idx;
+			ioeventfd.addr = bar->addr;
+			ioeventfd.len = 2;
+			ioeventfd.flags |= (ACRN_IOEVENTFD_FLAG_DATAMATCH |
+				ACRN_IOEVENTFD_FLAG_PIO);
+		} else if (base->modern_mmio_bar_idx) {
+			bar = &vdev->base->dev->bar[base->modern_mmio_bar_idx];
+			ioeventfd.data = 0;
+			ioeventfd.addr = bar->addr + VIRTIO_CAP_NOTIFY_OFFSET
+				+ (vdev->vq_idx + idx) *
+				VIRTIO_MODERN_NOTIFY_OFF_MULT;
+			ioeventfd.len = 2;
+			/* no additional flag bit should be set for MMIO */
+		} else {
+			WPRINTF("invalid virtio 1.0 parameters, 0x%lx\n",
+				base->device_caps);
+			return -1;
+		}
+	} else {
+		bar = &vdev->base->dev->bar[base->legacy_pio_bar_idx];
+		ioeventfd.data = vdev->vq_idx + idx;
+		ioeventfd.addr = bar->addr + VIRTIO_CR_QNOTIFY;
+		ioeventfd.len = 2;
+		ioeventfd.flags |= (ACRN_IOEVENTFD_FLAG_DATAMATCH |
+			ACRN_IOEVENTFD_FLAG_PIO);
+	}
+
+	ioeventfd.fd = vq->kick_fd;
+	DPRINTF("[ioeventfd: %d][0x%lx@%d][flags: 0x%x][data: 0x%lx]\n",
+		ioeventfd.fd, ioeventfd.addr, ioeventfd.len,
+		ioeventfd.flags, ioeventfd.data);
+	rc = vm_ioeventfd(vdev->base->dev->vmctx, &ioeventfd);
+	if (rc < 0) {
+		WPRINTF("vm_ioeventfd failed rc = %d, errno = %d\n",
+			rc, errno);
+		return -1;
+	}
+
+	if (pci_msix_enabled(base->dev)) {
+		/* register irqfd for notify */
+		struct msix_table_entry *mte;
+		struct acrn_msi_entry msi;
+
+		mte = &vdev->base->dev->msix.table[vqi->msix_idx];
+		msi.msi_addr = mte->addr;
+		msi.msi_data = mte->msg_data;
+		irqfd.fd = vq->call_fd;
+		/* no additional flag bit should be set */
+		irqfd.msi = msi;
+		DPRINTF("[irqfd: %d][MSIX: %d]\n", irqfd.fd, vqi->msix_idx);
+		rc = vm_irqfd(vdev->base->dev->vmctx, &irqfd);
+	} else {
+		/*
+		 * irqfd only supports MSIx now. For non-MSIx, call_fd is polled
+		 * by dm then inject interrupts to guest
+		 */
+		if (is_register) {
+			vq->mevp = mevent_add(vq->call_fd, EVF_READ,
+				vhost_vq_notify, vq);
+			if (!vq->mevp) {
+				WPRINTF("mevent_add failed\n");
+				rc = -1;
+			}
+		} else if (vq->mevp) {
+			mevent_delete(vq->mevp);
+			vq->mevp = NULL;
+		}
+	}
+
+	if (rc < 0) {
+		WPRINTF("vm_irqfd failed rc = %d, errno = %d\n", rc, errno);
+		/* unregister ioeventfd */
+		if (is_register) {
+			ioeventfd.flags |= ACRN_IOEVENTFD_FLAG_DEASSIGN;
+			vm_ioeventfd(vdev->base->dev->vmctx, &ioeventfd);
+		}
+		return -1;
+	}
+
+	return 0;
 }
 
 static int