[dpdk-dev] [PATCH v2 09/16] Enable VFIO device binding
Thomas Monjalon
thomas.monjalon at 6wind.com
Thu May 22 14:03:44 CEST 2014
2014-05-19 16:51, Anatoly Burakov:
> Add support for binding VFIO devices if RTE_PCI_DRV_NEED_IGB_UIO
> is set for this driver. Try VFIO first, if not mapped then try
> IGB_UIO too.
You have renamed RTE_PCI_DRV_NEED_IGB_UIO. Please update this log :)
> --- a/lib/librte_eal/linuxapp/eal/eal_pci.c
> +++ b/lib/librte_eal/linuxapp/eal/eal_pci.c
> @@ -401,6 +401,7 @@ rte_eal_pci_probe_one_driver(struct rte_pci_driver *dr,
> struct rte_pci_device *d {
> struct rte_pci_id *id_table;
> int ret = 0;
> + int mapped = 0;
>
> for (id_table = dr->id_table ; id_table->vendor_id != 0; id_table++) {
>
> @@ -435,8 +436,17 @@ rte_eal_pci_probe_one_driver(struct rte_pci_driver *dr,
> struct rte_pci_device *d }
>
> if (dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING) {
> + /* try mapping the NIC resources using VFIO if it exists */
> +#ifdef VFIO_PRESENT
> + if (vfio_cfg.vfio_enabled) {
> + if ((ret = pci_vfio_map_resource(dev)) == 0)
> + mapped = 1;
> + else if (ret < 0)
> + return ret;
> + }
> +#endif
> /* map resources for devices that use igb_uio */
> - if ((ret = pci_uio_map_resource(dev)) != 0)
> + if (!mapped && (ret = pci_uio_map_resource(dev)) != 0)
> return ret;
I think creating a function pci_map_resource() could be cleaner (you won't
need variable mapped).
> +#ifdef VFIO_PRESENT
> + memset(&vfio_cfg, 0, sizeof(vfio_cfg));
> +
> + /* initialize group list */
> + int i, ret;
> +
> + for (i = 0; i < VFIO_MAX_GROUPS; i++) {
> + vfio_cfg.vfio_groups[i].fd = -1;
> + vfio_cfg.vfio_groups[i].group_no = -1;
> + }
> + vfio_cfg.vfio_container_fd = -1;
> +
> + /* check if we have VFIO driver enabled */
> + if (access(VFIO_DIR, F_OK) == 0) {
> + static int socket_fd;
> +
> + vfio_cfg.vfio_enabled = 1;
> +
> + /* if we are primary process, create a thread to communicate with
> + * secondary processes. the thread will use a socket to wait for
> + * requests from secondary process to send open file descriptors,
> + * because VFIO does not allow multiple open descriptors on a group
or
> + * VFIO container.
> + */
> + if (internal_config.process_type == RTE_PROC_PRIMARY) {
> + /* set up local socket */
> + if ((socket_fd = pci_vfio_socket_setup()) < 0) {
> + RTE_LOG(ERR, EAL, "Failed to set up local socket!\n");
> + return -1;
> + }
> + ret = pthread_create(&socket_thread, NULL,
> + pci_vfio_socket_thread, (void*) &socket_fd);
> + if (ret) {
> + RTE_LOG(ERR, EAL,
> + "Failed to create thread for communication with secondary
"
> + "processes!\n");
> + return -1;
> + }
> + }
Also here, it could help to have a dedicated function for vfio init.
> + }
> + else
checkpatch.pl reports an error: "else should follow close brace '}'"
> + RTE_LOG(INFO, EAL, "VFIO driver not loaded or wrong permissions\n");
> +#endif
Thanks
--
Thomas
More information about the dev
mailing list