diff options
| -rw-r--r-- | src/plugins/dpdk/CMakeLists.txt | 1 | ||||
| -rw-r--r-- | src/plugins/dpdk/device/device.c | 2 | ||||
| -rw-r--r-- | src/plugins/dpdk/device/dpdk.h | 107 | ||||
| -rw-r--r-- | src/plugins/dpdk/device/dpdk_priv.h | 2 | ||||
| -rw-r--r-- | src/plugins/dpdk/device/driver.c | 145 | ||||
| -rw-r--r-- | src/plugins/dpdk/device/format.c | 159 | ||||
| -rw-r--r-- | src/plugins/dpdk/device/init.c | 452 |
7 files changed, 328 insertions, 540 deletions
diff --git a/src/plugins/dpdk/CMakeLists.txt b/src/plugins/dpdk/CMakeLists.txt index b53f4ad2375..91cf236ea05 100644 --- a/src/plugins/dpdk/CMakeLists.txt +++ b/src/plugins/dpdk/CMakeLists.txt @@ -134,6 +134,7 @@ add_vpp_plugin(dpdk device/cli.c device/common.c device/device.c + device/driver.c device/flow.c device/format.c device/init.c diff --git a/src/plugins/dpdk/device/device.c b/src/plugins/dpdk/device/device.c index b8418278fba..c9061a9949a 100644 --- a/src/plugins/dpdk/device/device.c +++ b/src/plugins/dpdk/device/device.c @@ -547,7 +547,7 @@ dpdk_subif_add_del_function (vnet_main_t * vnm, xd->num_subifs--; /* currently we program VLANS only for IXGBE VF */ - if (xd->pmd != VNET_DPDK_PMD_IXGBEVF) + if (xd->driver->program_vlans == 0) goto done; if (t->sub.eth.flags.no_tags == 1) diff --git a/src/plugins/dpdk/device/dpdk.h b/src/plugins/dpdk/device/dpdk.h index 1ebda6c5677..f1cd17da1cd 100644 --- a/src/plugins/dpdk/device/dpdk.h +++ b/src/plugins/dpdk/device/dpdk.h @@ -45,79 +45,6 @@ extern vnet_device_class_t dpdk_device_class; extern vlib_node_registration_t dpdk_input_node; extern vlib_node_registration_t admin_up_down_process_node; -#if RTE_VERSION < RTE_VERSION_NUM(20, 8, 0, 0) -#define DPDK_MLX5_PMD_NAME "net_mlx5" -#else -#define DPDK_MLX5_PMD_NAME "mlx5_pci" -#endif - -#define foreach_dpdk_pmd \ - _ ("net_thunderx", THUNDERX) \ - _ ("net_e1000_em", E1000EM) \ - _ ("net_e1000_igb", IGB) \ - _ ("net_e1000_igb_vf", IGBVF) \ - _ ("net_ixgbe", IXGBE) \ - _ ("net_ixgbe_vf", IXGBEVF) \ - _ ("net_i40e", I40E) \ - _ ("net_i40e_vf", I40EVF) \ - _ ("net_ice", ICE) \ - _ ("net_iavf", IAVF) \ - _ ("net_igc", IGC) \ - _ ("net_virtio", VIRTIO) \ - _ ("net_enic", ENIC) \ - _ ("net_vmxnet3", VMXNET3) \ - _ ("AF_PACKET PMD", AF_PACKET) \ - _ ("net_fm10k", FM10K) \ - _ ("net_cxgbe", CXGBE) \ - _ ("net_mlx4", MLX4) \ - _ (DPDK_MLX5_PMD_NAME, MLX5) \ - _ ("net_dpaa2", DPAA2) \ - _ ("net_virtio_user", VIRTIO_USER) \ - _ ("net_vhost", VHOST_ETHER) \ - _ ("net_ena", ENA) \ - _ ("net_failsafe", FAILSAFE) \ - _ ("net_liovf", LIOVF_ETHER) \ - _ ("net_qede", QEDE) \ - _ ("net_netvsc", NETVSC) \ - _ ("net_bnxt", BNXT) - -typedef enum -{ - VNET_DPDK_PMD_NONE, -#define _(s,f) VNET_DPDK_PMD_##f, - foreach_dpdk_pmd -#undef _ - VNET_DPDK_PMD_UNKNOWN, /* must be last */ -} dpdk_pmd_t; - -#define forach_dpdk_port_type \ - _ (ETH_1G, "GigabitEthernet") \ - _ (ETH_2_5G, "Two_FiveGigabitEthernet") \ - _ (ETH_5G, "FiveGigabitEthernet") \ - _ (ETH_10G, "TenGigabitEthernet") \ - _ (ETH_20G, "TwentyGigabitEthernet") \ - _ (ETH_25G, "TwentyFiveGigabitEthernet") \ - _ (ETH_40G, "FortyGigabitEthernet") \ - _ (ETH_50G, "FiftyGigabitEthernet") \ - _ (ETH_56G, "FiftySixGigabitEthernet") \ - _ (ETH_100G, "HundredGigabitEthernet") \ - _ (ETH_200G, "TwoHundredGigabitEthernet") \ - _ (ETH_SWITCH, "EthernetSwitch") \ - _ (ETH_VF, "VirtualFunctionEthernet") \ - _ (AF_PACKET, "af_packet") \ - _ (VIRTIO_USER, "VirtioUser") \ - _ (VHOST_ETHER, "VhostEthernet") \ - _ (FAILSAFE, "FailsafeEthernet") \ - _ (NETVSC, "NetVSC") - -typedef enum -{ - VNET_DPDK_PORT_TYPE_UNKNOWN = 0, -#define _(n, s) VNET_DPDK_PORT_TYPE_##n, - forach_dpdk_port_type -#undef _ -} dpdk_port_type_t; - typedef uint16_t dpdk_portid_t; #define foreach_dpdk_device_flags \ @@ -133,12 +60,12 @@ typedef uint16_t dpdk_portid_t; _ (13, INT_SUPPORTED, "int-supported") \ _ (14, INT_UNMASKABLE, "int-unmaskable") -enum +typedef enum { #define _(a, b, c) DPDK_DEVICE_FLAG_##b = (1 << a), foreach_dpdk_device_flags #undef _ -}; +} dpdk_device_flag_t; typedef struct { @@ -169,6 +96,31 @@ typedef struct clib_spinlock_t lock; } dpdk_tx_queue_t; +typedef struct +{ + const char *name; + const char *desc; +} dpdk_driver_name_t; + +typedef struct +{ + dpdk_driver_name_t *drivers; + const char *interface_name_prefix; + u16 n_rx_desc; + u16 n_tx_desc; + u32 supported_flow_actions; + i32 enable_lsc_int : 1; + i32 enable_rxq_int : 1; + i32 disable_rx_scatter : 1; + i32 program_vlans : 1; + i32 mq_mode_none : 1; + i32 interface_number_from_port_id : 1; + i32 use_intel_phdr_cksum : 1; + i32 int_unmaskable : 1; +} dpdk_driver_t; + +dpdk_driver_t *dpdk_driver_find (const char *name, const char **desc); + typedef union { struct @@ -216,15 +168,15 @@ typedef struct /* DPDK device port number */ dpdk_portid_t port_id; - dpdk_pmd_t pmd:8; i8 cpu_socket; CLIB_CACHE_LINE_ALIGN_MARK (cacheline1); u64 enabled_tx_off; u64 enabled_rx_off; + dpdk_driver_t *driver; u8 *name; - u8 *interface_name_suffix; + const char *if_desc; /* number of sub-interfaces */ u16 num_subifs; @@ -244,7 +196,6 @@ typedef struct struct rte_eth_stats last_stats; struct rte_eth_xstat *xstats; f64 time_last_stats_update; - dpdk_port_type_t port_type; /* mac address */ u8 *default_mac_address; diff --git a/src/plugins/dpdk/device/dpdk_priv.h b/src/plugins/dpdk/device/dpdk_priv.h index e170134052d..4e4438a007e 100644 --- a/src/plugins/dpdk/device/dpdk_priv.h +++ b/src/plugins/dpdk/device/dpdk_priv.h @@ -16,8 +16,6 @@ #define DPDK_NB_RX_DESC_DEFAULT 1024 #define DPDK_NB_TX_DESC_DEFAULT 1024 #define DPDK_MAX_LRO_SIZE_DEFAULT 65536 -#define DPDK_NB_RX_DESC_VIRTIO 256 -#define DPDK_NB_TX_DESC_VIRTIO 256 /* These args appear by themselves */ #define foreach_eal_double_hyphen_predicate_arg \ diff --git a/src/plugins/dpdk/device/driver.c b/src/plugins/dpdk/device/driver.c new file mode 100644 index 00000000000..11d65f7de94 --- /dev/null +++ b/src/plugins/dpdk/device/driver.c @@ -0,0 +1,145 @@ +/* SPDX-License-Identifier: Apache-2.0 + * Copyright(c) 2021 Cisco Systems, Inc. + */ + +#include <vlib/vlib.h> + +#include <dpdk/device/dpdk.h> + +static const u32 supported_flow_actions_intel = + (VNET_FLOW_ACTION_MARK | VNET_FLOW_ACTION_REDIRECT_TO_NODE | + VNET_FLOW_ACTION_REDIRECT_TO_QUEUE | VNET_FLOW_ACTION_BUFFER_ADVANCE | + VNET_FLOW_ACTION_COUNT | VNET_FLOW_ACTION_DROP | VNET_FLOW_ACTION_RSS); + +#define DPDK_DRIVERS(...) \ + (dpdk_driver_name_t[]) \ + { \ + __VA_ARGS__, {} \ + } + +static dpdk_driver_t dpdk_drivers[] = { + { + .drivers = DPDK_DRIVERS ({ "net_ice", "Intel E810 Family" }, + { "net_igc", "Intel I225 2.5G Family" }, + { "net_e1000_igb", "Intel e1000" }, + { "net_e1000_em", "Intel 82540EM (e1000)" }), + .enable_rxq_int = 1, + .supported_flow_actions = supported_flow_actions_intel, + .use_intel_phdr_cksum = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_ixgbe", "Intel 82599" }), + .enable_rxq_int = 1, + .program_vlans = 1, + .supported_flow_actions = supported_flow_actions_intel, + .use_intel_phdr_cksum = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_i40e", "Intel X710/XL710 Family" }), + .enable_rxq_int = 1, + .supported_flow_actions = supported_flow_actions_intel, + .use_intel_phdr_cksum = 1, + .int_unmaskable = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_liovf", "Cavium Lio VF" }, + { "net_thunderx", "Cavium ThunderX" }), + .interface_name_prefix = "VirtualFunctionEthernet", + }, + { + .drivers = DPDK_DRIVERS ({ "net_iavf", "Intel iAVF" }, + { "net_i40e_vf", "Intel X710/XL710 Family VF" }), + .interface_name_prefix = "VirtualFunctionEthernet", + .supported_flow_actions = supported_flow_actions_intel, + .use_intel_phdr_cksum = 1, + .int_unmaskable = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_e1000_igb_vf", "Intel e1000 VF" }, + { "net_ixgbe_vf", "Intel 82599 VF" }), + .interface_name_prefix = "VirtualFunctionEthernet", + .use_intel_phdr_cksum = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_dpaa2", "NXP DPAA2 Mac" }), + .interface_name_prefix = "TenGigabitEthernet", + }, + { + .drivers = + DPDK_DRIVERS ({ "net_fm10k", "Intel FM10000 Family Ethernet Switch" }), + .interface_name_prefix = "EthernetSwitch", + }, + { + .drivers = DPDK_DRIVERS ({ "net_cxgbe", "Chelsio T4/T5" }), + .interface_number_from_port_id = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_mlx4", "Mellanox ConnectX-3 Family" }, + { "net_qede", "Cavium QLogic FastLinQ QL4xxxx" }, + { "net_bnxt", "Broadcom NetXtreme E/S-Series" }), + }, + { + .drivers = DPDK_DRIVERS ({ "net_virtio_user", "Virtio User" }), + .interface_name_prefix = "VirtioUser", + }, + { + .drivers = DPDK_DRIVERS ({ "net_vhost", "VhostEthernet" }), + .interface_name_prefix = "VhostEthernet", + }, + { + .drivers = DPDK_DRIVERS ({ "mlx5_pci", "Mellanox ConnectX-4 Family" }, + { "net_enic", "Cisco VIC" }), + .use_intel_phdr_cksum = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_failsafe", "FailsafeEthernet" }), + .interface_name_prefix = "FailsafeEthernet", + .enable_lsc_int = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "AF_PACKET PMD", "af_packet" }), + .interface_name_prefix = "af_packet", + }, + { + .drivers = DPDK_DRIVERS ({ "net_netvsc", "Microsoft Hyper-V Netvsc" }), + .interface_name_prefix = "NetVSC", + }, + { + .drivers = DPDK_DRIVERS ({ "net_ena", "AWS ENA VF" }), + .interface_name_prefix = "VirtualFunctionEthernet", + .enable_rxq_int = 1, + .disable_rx_scatter = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_vmxnet3", "VMware VMXNET3" }), + .interface_name_prefix = "GigabitEthernet", + }, + { + .drivers = DPDK_DRIVERS ({ "net_virtio", "Red Hat Virtio" }), + .interface_name_prefix = "GigabitEthernet", + .n_rx_desc = 256, + .n_tx_desc = 256, + .mq_mode_none = 1, + } +}; + +dpdk_driver_t * +dpdk_driver_find (const char *name, const char **desc) +{ + for (int i = 0; i < ARRAY_LEN (dpdk_drivers); i++) + { + dpdk_driver_t *dr = dpdk_drivers + i; + dpdk_driver_name_t *dn = dr->drivers; + + while (dn->name) + { + if (name && !strcmp (name, dn->name)) + { + *desc = dn->desc; + return dr; + } + dn++; + } + } + return 0; +} diff --git a/src/plugins/dpdk/device/format.c b/src/plugins/dpdk/device/format.c index e8aca247be5..4a4b4110b9f 100644 --- a/src/plugins/dpdk/device/format.c +++ b/src/plugins/dpdk/device/format.c @@ -147,49 +147,14 @@ #define foreach_dpdk_pkt_dyn_rx_offload_flag \ _ (RX_TIMESTAMP, 0, "Timestamp field is valid") -static char *device_name_by_port_type[] = { -#define _(n, s) [VNET_DPDK_PORT_TYPE_##n] = (s), - forach_dpdk_port_type -#undef _ -}; - u8 * format_dpdk_device_name (u8 * s, va_list * args) { dpdk_main_t *dm = &dpdk_main; - char *devname_format; - char *device_name = 0; u32 i = va_arg (*args, u32); dpdk_device_t *xd = vec_elt_at_index (dm->devices, i); - struct rte_eth_dev_info dev_info; - struct rte_pci_device *pci_dev; - u8 *ret; - - if (xd->name) - return format (s, "%s", xd->name); - - if (dm->conf->interface_name_format_decimal) - devname_format = "%s%d/%d/%d"; - else - devname_format = "%s%x/%x/%x"; - - if (xd->port_type < ARRAY_LEN (device_name_by_port_type)) - device_name = device_name_by_port_type[xd->port_type]; - - device_name = device_name ? device_name : "UnknownEthernet"; - - rte_eth_dev_info_get (xd->port_id, &dev_info); - pci_dev = dpdk_get_pci_device (&dev_info); - - if (pci_dev && xd->port_type != VNET_DPDK_PORT_TYPE_FAILSAFE) - ret = format (s, devname_format, device_name, pci_dev->addr.bus, - pci_dev->addr.devid, pci_dev->addr.function); - else - ret = format (s, "%s%d", device_name, xd->port_id); - if (xd->interface_name_suffix) - return format (ret, "/%s", xd->interface_name_suffix); - return ret; + return format (s, "%v", xd->name); } u8 * @@ -211,126 +176,12 @@ static u8 * format_dpdk_device_type (u8 * s, va_list * args) { dpdk_main_t *dm = &dpdk_main; - char *dev_type; u32 i = va_arg (*args, u32); - switch (dm->devices[i].pmd) - { - case VNET_DPDK_PMD_E1000EM: - dev_type = "Intel 82540EM (e1000)"; - break; - - case VNET_DPDK_PMD_IGB: - dev_type = "Intel e1000"; - break; - - case VNET_DPDK_PMD_I40E: - dev_type = "Intel X710/XL710 Family"; - break; - - case VNET_DPDK_PMD_I40EVF: - dev_type = "Intel X710/XL710 Family VF"; - break; - - case VNET_DPDK_PMD_ICE: - dev_type = "Intel E810 Family"; - break; - - case VNET_DPDK_PMD_IAVF: - dev_type = "Intel iAVF"; - break; - - case VNET_DPDK_PMD_FM10K: - dev_type = "Intel FM10000 Family Ethernet Switch"; - break; - - case VNET_DPDK_PMD_IGBVF: - dev_type = "Intel e1000 VF"; - break; - - case VNET_DPDK_PMD_VIRTIO: - dev_type = "Red Hat Virtio"; - break; - - case VNET_DPDK_PMD_IXGBEVF: - dev_type = "Intel 82599 VF"; - break; - - case VNET_DPDK_PMD_IXGBE: - dev_type = "Intel 82599"; - break; - - case VNET_DPDK_PMD_ENIC: - dev_type = "Cisco VIC"; - break; - - case VNET_DPDK_PMD_CXGBE: - dev_type = "Chelsio T4/T5"; - break; - - case VNET_DPDK_PMD_MLX4: - dev_type = "Mellanox ConnectX-3 Family"; - break; - - case VNET_DPDK_PMD_MLX5: - dev_type = "Mellanox ConnectX-4 Family"; - break; - - case VNET_DPDK_PMD_VMXNET3: - dev_type = "VMware VMXNET3"; - break; - - case VNET_DPDK_PMD_AF_PACKET: - dev_type = "af_packet"; - break; - - case VNET_DPDK_PMD_DPAA2: - dev_type = "NXP DPAA2 Mac"; - break; - - case VNET_DPDK_PMD_VIRTIO_USER: - dev_type = "Virtio User"; - break; - - case VNET_DPDK_PMD_THUNDERX: - dev_type = "Cavium ThunderX"; - break; - - case VNET_DPDK_PMD_VHOST_ETHER: - dev_type = "VhostEthernet"; - break; - - case VNET_DPDK_PMD_ENA: - dev_type = "AWS ENA VF"; - break; - - case VNET_DPDK_PMD_FAILSAFE: - dev_type = "FailsafeEthernet"; - break; - - case VNET_DPDK_PMD_LIOVF_ETHER: - dev_type = "Cavium Lio VF"; - break; - - case VNET_DPDK_PMD_QEDE: - dev_type = "Cavium QLogic FastLinQ QL4xxxx"; - break; - - case VNET_DPDK_PMD_NETVSC: - dev_type = "Microsoft Hyper-V Netvsc"; - break; - - case VNET_DPDK_PMD_BNXT: - dev_type = "Broadcom NetXtreme E/S-Series"; - break; - - default: - case VNET_DPDK_PMD_UNKNOWN: - dev_type = "### UNKNOWN ###"; - break; - } - - return format (s, dev_type); + if (dm->devices[i].if_desc) + return format (s, dm->devices[i].if_desc); + else + return format (s, "### UNKNOWN ###"); } static u8 * diff --git a/src/plugins/dpdk/device/init.c b/src/plugins/dpdk/device/init.c index 1d3e061aaaf..5e06e48d37e 100644 --- a/src/plugins/dpdk/device/init.c +++ b/src/plugins/dpdk/device/init.c @@ -43,42 +43,30 @@ #include <dpdk/device/dpdk_priv.h> -#define ETHER_MAX_LEN 1518 /**< Maximum frame len, including CRC. */ - dpdk_main_t dpdk_main; dpdk_config_main_t dpdk_config_main; #define LINK_STATE_ELOGS 0 -/* Port configuration, mildly modified Intel app values */ - -static dpdk_port_type_t -port_type_from_speed_capa (struct rte_eth_dev_info *dev_info) +/* dev_info.speed_capa -> interface name mapppings */ +const struct { - - if (dev_info->speed_capa & ETH_LINK_SPEED_100G) - return VNET_DPDK_PORT_TYPE_ETH_100G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_56G) - return VNET_DPDK_PORT_TYPE_ETH_56G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_50G) - return VNET_DPDK_PORT_TYPE_ETH_50G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_40G) - return VNET_DPDK_PORT_TYPE_ETH_40G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_25G) - return VNET_DPDK_PORT_TYPE_ETH_25G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_20G) - return VNET_DPDK_PORT_TYPE_ETH_20G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_10G) - return VNET_DPDK_PORT_TYPE_ETH_10G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_5G) - return VNET_DPDK_PORT_TYPE_ETH_5G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_2_5G) - return VNET_DPDK_PORT_TYPE_ETH_2_5G; - else if (dev_info->speed_capa & ETH_LINK_SPEED_1G) - return VNET_DPDK_PORT_TYPE_ETH_1G; - - return VNET_DPDK_PORT_TYPE_UNKNOWN; -} + u32 link_speed; + const char *pfx; +} if_name_prefixes[] = { + /* sorted, higher speed first */ + { ETH_LINK_SPEED_200G, "TwoHundredGigabitEthernet" }, + { ETH_LINK_SPEED_100G, "HundredGigabitEthernet" }, + { ETH_LINK_SPEED_56G, "FiftySixGigabitEthernet" }, + { ETH_LINK_SPEED_50G, "FiftyGigabitEthernet" }, + { ETH_LINK_SPEED_40G, "FortyGigabitEthernet" }, + { ETH_LINK_SPEED_25G, "TwentyFiveGigabitEthernet" }, + { ETH_LINK_SPEED_20G, "TwentyGigabitEthernet" }, + { ETH_LINK_SPEED_10G, "TenGigabitEthernet" }, + { ETH_LINK_SPEED_5G, "FiveGigabitEthernet" }, + { ETH_LINK_SPEED_2_5G, "TwoDotFiveGigabitEthernet" }, + { ETH_LINK_SPEED_1G, "GigabitEthernet" }, +}; static u32 dpdk_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi, u32 flags) @@ -161,31 +149,53 @@ check_l3cache () return 0; } +static dpdk_device_config_t * +dpdk_find_startup_config (struct rte_eth_dev_info *di) +{ + dpdk_main_t *dm = &dpdk_main; + struct rte_pci_device *pci_dev; + struct rte_vmbus_device *vmbus_dev; + vlib_pci_addr_t pci_addr; + vlib_vmbus_addr_t vmbus_addr; + uword *p = 0; + + if ((pci_dev = dpdk_get_pci_device (di))) + { + pci_addr.domain = pci_dev->addr.domain; + pci_addr.bus = pci_dev->addr.bus; + pci_addr.slot = pci_dev->addr.devid; + pci_addr.function = pci_dev->addr.function; + p = + hash_get (dm->conf->device_config_index_by_pci_addr, pci_addr.as_u32); + } + + if ((vmbus_dev = dpdk_get_vmbus_device (di))) + { + unformat_input_t input_vmbus; + unformat_init_string (&input_vmbus, di->device->name, + strlen (di->device->name)); + if (unformat (&input_vmbus, "%U", unformat_vlib_vmbus_addr, &vmbus_addr)) + p = mhash_get (&dm->conf->device_config_index_by_vmbus_addr, + &vmbus_addr); + unformat_free (&input_vmbus); + } + + if (p) + return pool_elt_at_index (dm->conf->dev_confs, p[0]); + return &dm->conf->default_devconf; +} + static clib_error_t * dpdk_lib_init (dpdk_main_t * dm) { vnet_main_t *vnm = vnet_get_main (); - u32 nports; u16 port_id; - vlib_main_t *vm = vlib_get_main (); vlib_thread_main_t *tm = vlib_get_thread_main (); vnet_device_main_t *vdm = &vnet_device_main; vnet_sw_interface_t *sw; vnet_hw_interface_t *hi; dpdk_device_t *xd; - vlib_pci_addr_t last_pci_addr; - u32 last_pci_addr_port = 0; - last_pci_addr.as_u32 = ~0; - - nports = rte_eth_dev_count_avail (); - - if (nports < 1) - { - dpdk_log_notice ("DPDK drivers found no Ethernet devices..."); - } - - if (CLIB_DEBUG > 0) - dpdk_log_notice ("DPDK drivers found %d ports...", nports); + char *if_num_fmt; /* vlib_buffer_t template */ vec_validate_aligned (dm->per_thread_data, tm->n_vlib_mains - 1, @@ -197,6 +207,9 @@ dpdk_lib_init (dpdk_main_t * dm) vnet_buffer (&ptd->buffer_template)->sw_if_index[VLIB_TX] = (u32) ~ 0; } + if_num_fmt = + dm->conf->interface_name_format_decimal ? "%d/%d/%d" : "%x/%x/%x"; + /* device config defaults */ dm->default_port_conf.n_rx_desc = DPDK_NB_RX_DESC_DEFAULT; dm->default_port_conf.n_tx_desc = DPDK_NB_TX_DESC_DEFAULT; @@ -212,121 +225,109 @@ dpdk_lib_init (dpdk_main_t * dm) RTE_ETH_FOREACH_DEV (port_id) { u8 addr[6]; + int rv, q; struct rte_eth_dev_info di; - struct rte_pci_device *pci_dev; - struct rte_vmbus_device *vmbus_dev; - dpdk_portid_t next_port_id; dpdk_device_config_t *devconf = 0; vnet_eth_interface_registration_t eir = {}; - vlib_pci_addr_t pci_addr; - vlib_vmbus_addr_t vmbus_addr; - uword *p = 0; + dpdk_driver_t *dr; if (!rte_eth_dev_is_valid_port (port_id)) continue; - rte_eth_dev_info_get (port_id, &di); - - if (di.device == 0) + if ((rv = rte_eth_dev_info_get (port_id, &di)) != 0) { - dpdk_log_notice ("DPDK bug: missing device info. Skipping %s device", - di.driver_name); + dpdk_log_warn ("[%u] failed to get device info. skipping device.", + port_id); continue; } - pci_dev = dpdk_get_pci_device (&di); - - if (pci_dev) + if (di.device == 0) { - pci_addr.domain = pci_dev->addr.domain; - pci_addr.bus = pci_dev->addr.bus; - pci_addr.slot = pci_dev->addr.devid; - pci_addr.function = pci_dev->addr.function; - p = hash_get (dm->conf->device_config_index_by_pci_addr, - pci_addr.as_u32); + dpdk_log_warn ("[%u] missing device info. Skipping '%s' device", + port_id, di.driver_name); + continue; } - vmbus_dev = dpdk_get_vmbus_device (&di); + devconf = dpdk_find_startup_config (&di); - if (vmbus_dev) + /* If device is blacklisted, we should skip it */ + if (devconf->is_blacklisted) { - unformat_input_t input_vmbus; - unformat_init_string (&input_vmbus, di.device->name, - strlen (di.device->name)); - if (unformat (&input_vmbus, "%U", unformat_vlib_vmbus_addr, - &vmbus_addr)) - { - p = mhash_get (&dm->conf->device_config_index_by_vmbus_addr, - &vmbus_addr); - } - unformat_free (&input_vmbus); - } - - if (p) - { - devconf = pool_elt_at_index (dm->conf->dev_confs, p[0]); - /* If device is blacklisted, we should skip it */ - if (devconf->is_blacklisted) - { - continue; - } + dpdk_log_notice ("[%d] Device %s blacklisted. Skipping...", port_id, + di.driver_name); + continue; } - else - devconf = &dm->conf->default_devconf; - /* Create vnet interface */ vec_add2_aligned (dm->devices, xd, 1, CLIB_CACHE_LINE_BYTES); - xd->cpu_socket = (i8) rte_eth_dev_socket_id (port_id); + xd->port_id = port_id; + xd->device_index = xd - dm->devices; + xd->per_interface_next_index = ~0; + clib_memcpy (&xd->conf, &dm->default_port_conf, sizeof (dpdk_port_conf_t)); - if (p) + /* find driver datea for this PMD */ + if ((dr = dpdk_driver_find (di.driver_name, &xd->if_desc))) { - xd->name = devconf->name; + xd->driver = dr; + xd->supported_flow_actions = dr->supported_flow_actions; + xd->conf.disable_rss = dr->mq_mode_none; + xd->conf.disable_rx_scatter = dr->disable_rx_scatter; + if (dr->use_intel_phdr_cksum) + dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM, 1); + if (dr->int_unmaskable) + dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_INT_UNMASKABLE, 1); } + else + dpdk_log_warn ("[%u] unknown driver '%s'", port_id, di.driver_name); - /* Handle representor devices that share the same PCI ID */ - if (di.switch_info.domain_id != RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) + if (devconf->name) { - if (di.switch_info.port_id != (uint16_t) -1) - xd->interface_name_suffix = - format (0, "%d", di.switch_info.port_id); + xd->name = devconf->name; } - /* Handle interface naming for devices with multiple ports sharing same - * PCI ID */ - else if (pci_dev && ((next_port_id = rte_eth_find_next (port_id + 1)) != - RTE_MAX_ETHPORTS)) + else { - struct rte_eth_dev_info next_di = { 0 }; - struct rte_pci_device *next_pci_dev; - rte_eth_dev_info_get (next_port_id, &next_di); - next_pci_dev = next_di.device ? RTE_DEV_TO_PCI (next_di.device) : 0; - if (next_pci_dev && pci_addr.as_u32 != last_pci_addr.as_u32 && - memcmp (&pci_dev->addr, &next_pci_dev->addr, - sizeof (struct rte_pci_addr)) == 0) - { - xd->interface_name_suffix = format (0, "0"); - last_pci_addr.as_u32 = pci_addr.as_u32; - last_pci_addr_port = port_id; - } - else if (pci_addr.as_u32 == last_pci_addr.as_u32) + struct rte_pci_device *pci_dev; + if (dr && dr->interface_name_prefix) { - xd->interface_name_suffix = - format (0, "%u", port_id - last_pci_addr_port); + /* prefix override by driver */ + xd->name = format (xd->name, "%s", dr->interface_name_prefix); } else { - last_pci_addr.as_u32 = ~0; + /* interface name prefix from speed_capa */ + u64 mask = ~((if_name_prefixes[0].link_speed << 1) - 1); + + if (di.speed_capa & mask) + dpdk_log_warn ("[%u] unknown speed capability 0x%x reported", + xd->port_id, di.speed_capa & mask); + + for (int i = 0; i < ARRAY_LEN (if_name_prefixes); i++) + if (if_name_prefixes[i].link_speed & di.speed_capa) + { + xd->name = + format (xd->name, "%s", if_name_prefixes[i].pfx); + break; + } + if (xd->name == 0) + xd->name = format (xd->name, "Ethernet"); } - } - else - last_pci_addr.as_u32 = ~0; - if (devconf->max_lro_pkt_size) - xd->conf.max_lro_pkt_size = devconf->max_lro_pkt_size; + if (dr && dr->interface_number_from_port_id) + xd->name = format (xd->name, "%u", port_id); + else if ((pci_dev = dpdk_get_pci_device (&di))) + xd->name = format (xd->name, if_num_fmt, pci_dev->addr.bus, + pci_dev->addr.devid, pci_dev->addr.function); + else + xd->name = format (xd->name, "%u", port_id); + } - xd->conf.n_tx_queues = clib_min (di.max_tx_queues, xd->conf.n_tx_queues); + /* Handle representor devices that share the same PCI ID */ + if ((di.switch_info.domain_id != RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) && + (di.switch_info.port_id != (uint16_t) -1)) + xd->name = format (xd->name, "/%d", di.switch_info.port_id); + /* number of RX and TX queues */ if (devconf->num_tx_queues > 0 && devconf->num_tx_queues < xd->conf.n_tx_queues) xd->conf.n_tx_queues = devconf->num_tx_queues; @@ -348,184 +349,39 @@ dpdk_lib_init (dpdk_main_t * dm) xd->conf.rss_hf &= di.flow_type_rss_offloads; } + /* number of RX and TX tescriptors */ if (devconf->num_rx_desc) xd->conf.n_rx_desc = devconf->num_rx_desc; + else if (dr && dr->n_rx_desc) + xd->conf.n_rx_desc = dr->n_rx_desc; if (devconf->num_tx_desc) xd->conf.n_tx_desc = devconf->num_tx_desc; + else if (dr && dr->n_tx_desc) + xd->conf.n_tx_desc = dr->n_tx_desc; vec_validate_aligned (xd->rx_queues, xd->conf.n_rx_queues - 1, CLIB_CACHE_LINE_BYTES); - /* workaround for drivers not setting driver_name */ - if ((!di.driver_name) && (pci_dev)) - di.driver_name = pci_dev->driver->driver.name; - - ASSERT (di.driver_name); - - if (!xd->pmd) - { - -#define _(s, f) \ - else if (di.driver_name && !strcmp (di.driver_name, s)) xd->pmd = \ - VNET_DPDK_PMD_##f; - if (0) - ; - foreach_dpdk_pmd -#undef _ - else xd->pmd = VNET_DPDK_PMD_UNKNOWN; - - xd->port_type = VNET_DPDK_PORT_TYPE_UNKNOWN; - - switch (xd->pmd) - { - /* Drivers with valid speed_capa set */ - case VNET_DPDK_PMD_I40E: - dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_INT_UNMASKABLE, 1); - /* fall through */ - case VNET_DPDK_PMD_E1000EM: - case VNET_DPDK_PMD_IGB: - case VNET_DPDK_PMD_IGC: - case VNET_DPDK_PMD_IXGBE: - case VNET_DPDK_PMD_ICE: - xd->supported_flow_actions = - VNET_FLOW_ACTION_MARK | VNET_FLOW_ACTION_REDIRECT_TO_NODE | - VNET_FLOW_ACTION_REDIRECT_TO_QUEUE | - VNET_FLOW_ACTION_BUFFER_ADVANCE | VNET_FLOW_ACTION_COUNT | - VNET_FLOW_ACTION_DROP | VNET_FLOW_ACTION_RSS; - dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM, 1); - xd->conf.enable_rxq_int = 1; - /* fall through */ - case VNET_DPDK_PMD_MLX5: - case VNET_DPDK_PMD_CXGBE: - case VNET_DPDK_PMD_MLX4: - case VNET_DPDK_PMD_QEDE: - case VNET_DPDK_PMD_BNXT: - case VNET_DPDK_PMD_ENIC: - xd->port_type = port_type_from_speed_capa (&di); - break; - - /* SR-IOV VFs */ - case VNET_DPDK_PMD_I40EVF: - dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_INT_UNMASKABLE, 1); - /* fall through */ - case VNET_DPDK_PMD_IGBVF: - case VNET_DPDK_PMD_IXGBEVF: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM, 1); - /* DPDK bug in multiqueue... */ - /* xd->port_conf.intr_conf.rxq = 1; */ - break; - - /* iAVF */ - case VNET_DPDK_PMD_IAVF: - dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_INT_UNMASKABLE, 1); - dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM, 1); - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - xd->supported_flow_actions = - VNET_FLOW_ACTION_MARK | VNET_FLOW_ACTION_REDIRECT_TO_NODE | - VNET_FLOW_ACTION_REDIRECT_TO_QUEUE | - VNET_FLOW_ACTION_BUFFER_ADVANCE | VNET_FLOW_ACTION_COUNT | - VNET_FLOW_ACTION_DROP | VNET_FLOW_ACTION_RSS; - /* DPDK bug in multiqueue... */ - /* xd->port_conf.intr_conf.rxq = 1; */ - break; - - case VNET_DPDK_PMD_THUNDERX: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - break; - - case VNET_DPDK_PMD_ENA: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - xd->conf.disable_rx_scatter = 1; - xd->conf.enable_rxq_int = 1; - break; - - case VNET_DPDK_PMD_DPAA2: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_10G; - break; - - /* Intel Red Rock Canyon */ - case VNET_DPDK_PMD_FM10K: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_SWITCH; - break; - - /* virtio */ - case VNET_DPDK_PMD_VIRTIO: - xd->conf.disable_rss = 1; - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_1G; - xd->conf.n_rx_desc = DPDK_NB_RX_DESC_VIRTIO; - xd->conf.n_tx_desc = DPDK_NB_TX_DESC_VIRTIO; - xd->conf.enable_rxq_int = 1; - break; - - /* vmxnet3 */ - case VNET_DPDK_PMD_VMXNET3: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_1G; - break; - - case VNET_DPDK_PMD_AF_PACKET: - xd->port_type = VNET_DPDK_PORT_TYPE_AF_PACKET; - break; - - case VNET_DPDK_PMD_VIRTIO_USER: - xd->port_type = VNET_DPDK_PORT_TYPE_VIRTIO_USER; - break; - - case VNET_DPDK_PMD_VHOST_ETHER: - xd->port_type = VNET_DPDK_PORT_TYPE_VHOST_ETHER; - break; - - case VNET_DPDK_PMD_LIOVF_ETHER: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - break; - - case VNET_DPDK_PMD_FAILSAFE: - xd->port_type = VNET_DPDK_PORT_TYPE_FAILSAFE; - xd->conf.enable_lsc_int = 1; - break; - - case VNET_DPDK_PMD_NETVSC: - { - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - } - break; - - default: - xd->port_type = VNET_DPDK_PORT_TYPE_UNKNOWN; - } - } - - if (xd->pmd == VNET_DPDK_PMD_AF_PACKET) - { - f64 now = vlib_time_now (vm); - u32 rnd; - rnd = (u32) (now * 1e6); - rnd = random_u32 (&rnd); - clib_memcpy (addr + 2, &rnd, sizeof (rnd)); - addr[0] = 2; - addr[1] = 0xfe; - } - else - rte_eth_macaddr_get (port_id, (void *) addr); - - xd->port_id = port_id; - xd->device_index = xd - dm->devices; - xd->per_interface_next_index = ~0; - - /* assign interface to input thread */ - int q; + rte_eth_macaddr_get (port_id, (void *) addr); + /* create interface */ eir.dev_class_index = dpdk_device_class.index; eir.dev_instance = xd->device_index; eir.address = addr; eir.cb.flag_change = dpdk_flag_change; xd->hw_if_index = vnet_eth_register_interface (vnm, &eir); - + hi = vnet_get_hw_interface (vnm, xd->hw_if_index); + hi->numa_node = xd->cpu_socket = (i8) rte_eth_dev_socket_id (port_id); sw = vnet_get_hw_sw_interface (vnm, xd->hw_if_index); xd->sw_if_index = sw->sw_if_index; - vnet_hw_if_set_input_node (vnm, xd->hw_if_index, dpdk_input_node.index); + dpdk_log_debug ("[%u] interface %s created", port_id, hi->name); + + ethernet_set_flags (vnm, xd->hw_if_index, + ETHERNET_INTERFACE_FLAG_DEFAULT_L3); + /* assign worker threads */ + vnet_hw_if_set_input_node (vnm, xd->hw_if_index, dpdk_input_node.index); if (devconf->workers) { int j; @@ -545,20 +401,6 @@ dpdk_lib_init (dpdk_main_t * dm) vnm, xd->hw_if_index, q, VNET_HW_IF_RXQ_THREAD_ANY); } - - /*Get vnet hardware interface */ - hi = vnet_get_hw_interface (vnm, xd->hw_if_index); - - if (hi) - { - hi->numa_node = xd->cpu_socket; - - /* Indicate ability to support L3 DMAC filtering and - * initialize interface to L3 non-promisc mode */ - ethernet_set_flags (vnm, xd->hw_if_index, - ETHERNET_INTERFACE_FLAG_DEFAULT_L3); - } - if (devconf->tso == DPDK_DEVICE_TSO_ON) { /*tcp_udp checksum must be enabled*/ @@ -571,21 +413,21 @@ dpdk_lib_init (dpdk_main_t * dm) xd->conf.enable_tso = 1; } + if (devconf->max_lro_pkt_size) + xd->conf.max_lro_pkt_size = devconf->max_lro_pkt_size; + dpdk_device_setup (xd); /* rss queues should be configured after dpdk_device_setup() */ - if ((hi != NULL) && (devconf->rss_queues != NULL)) + if (devconf->rss_queues) { if (vnet_hw_interface_set_rss_queues (vnet_get_main (), hi, devconf->rss_queues)) - { - clib_warning ("%s: Failed to set rss queues", hi->name); - } + dpdk_log_warn ("[%u] Failed to set rss queues", port_id); } if (vec_len (xd->errors)) - dpdk_log_err ("setup failed for device %U. Errors:\n %U", - format_dpdk_device_name, port_id, + dpdk_log_err ("[%u] setup failed Errors:\n %U", port_id, format_dpdk_device_errors, xd); } @@ -944,7 +786,7 @@ dpdk_device_config (dpdk_config_main_t *conf, void *addr, ; else if (unformat (input, "num-tx-desc %u", &devconf->num_tx_desc)) ; - else if (unformat (input, "name %s", &devconf->name)) + else if (unformat (input, "name %v", &devconf->name)) ; else if (unformat (input, "workers %U", unformat_bitmap_list, &devconf->workers)) |