The last Mellanox adapters\SoCs support virtio operations to accelerate the vhost data path.
A new net PMD will be added to implement the vdpa driver requirements on top of Mellanox devices support it like ConnectX6 and BlueField. Points: * The mlx5_vdpa PMD run on top of PCI devices VF and PF. * The Mellanox PCI device can be configured ether to support ethdev device or to support vdpa device. * An one physical device can contain ethdev VFs\PF driven by the mlx5 PMD and vdpa VFs\PF driven by the new mlx5_vdpa PMD parallelly. * The decision which driver should be selected to probe Mellanox PCI device should be taken by the user using the PCI device devargs. * mlx5 and mlx5_vdpa depend in rdma-core lib so some code may be shared between them, due that a new mlx5 common directory will be added in drivers/commom for code reusing. * All the guest physical memory of the virtqs will be translated to the host physical memory by the HW.
