diff options
Diffstat (limited to 'src/plugins/dpdk')
-rw-r--r-- | src/plugins/dpdk/CMakeLists.txt | 20 | ||||
-rw-r--r-- | src/plugins/dpdk/buffer.c | 28 | ||||
-rw-r--r-- | src/plugins/dpdk/cryptodev/cryptodev.c | 432 | ||||
-rw-r--r-- | src/plugins/dpdk/cryptodev/cryptodev.h | 204 | ||||
-rw-r--r-- | src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c | 467 | ||||
-rw-r--r-- | src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c | 428 | ||||
-rw-r--r-- | src/plugins/dpdk/device/cli.c | 53 | ||||
-rw-r--r-- | src/plugins/dpdk/device/common.c | 276 | ||||
-rw-r--r-- | src/plugins/dpdk/device/device.c | 137 | ||||
-rw-r--r-- | src/plugins/dpdk/device/dpdk.h | 289 | ||||
-rw-r--r-- | src/plugins/dpdk/device/dpdk_priv.h | 173 | ||||
-rw-r--r-- | src/plugins/dpdk/device/driver.c | 154 | ||||
-rw-r--r-- | src/plugins/dpdk/device/flow.c | 226 | ||||
-rw-r--r-- | src/plugins/dpdk/device/format.c | 601 | ||||
-rw-r--r-- | src/plugins/dpdk/device/init.c | 1428 | ||||
-rw-r--r-- | src/plugins/dpdk/device/node.c | 150 | ||||
-rw-r--r-- | src/plugins/dpdk/main.c | 13 | ||||
-rw-r--r-- | src/plugins/dpdk/thread.c | 85 |
18 files changed, 2895 insertions, 2269 deletions
diff --git a/src/plugins/dpdk/CMakeLists.txt b/src/plugins/dpdk/CMakeLists.txt index 5de75e76289..48c56f35282 100644 --- a/src/plugins/dpdk/CMakeLists.txt +++ b/src/plugins/dpdk/CMakeLists.txt @@ -90,8 +90,10 @@ else() ############################################################################## # libnuma ############################################################################## - vpp_plugin_find_library(dpdk NUMA_LIB "numa") - list(APPEND DPDK_LINK_LIBRARIES ${NUMA_LIB}) + if("${CMAKE_SYSTEM_NAME}" STREQUAL "Linux") + vpp_plugin_find_library(dpdk NUMA_LIB "numa") + list(APPEND DPDK_LINK_LIBRARIES ${NUMA_LIB}) + endif() ############################################################################## # Mellanox libraries @@ -103,12 +105,10 @@ else() list(APPEND DPDK_LINK_LIBRARIES "${MNL_LIB}") else() message(WARNING "EXPERIMENTAL: DPDK plugin without dlopen mode") - vpp_plugin_find_library(dpdk IBVERBS_LIB "libibverbs.a") - vpp_plugin_find_library(dpdk MLX5_LIB "libmlx5.a") - vpp_plugin_find_library(dpdk MLX4_LIB "libmlx4.a") - vpp_plugin_find_library(dpdk CCAN_LIB "libccan.a") - vpp_plugin_find_library(dpdk RDMA_UTIL_LIB "rdma_util") - string_append(DPDK_LINK_FLAGS "-Wl,--whole-archive,${IBVERBS_LIB},${MLX5_LIB},${MLX4_LIB},${CCAN_LIB},${RDMA_UTIL_LIB},--no-whole-archive") + vpp_plugin_find_library(dpdk IBVERBS_LIB "libibverbs.a") + vpp_plugin_find_library(dpdk MLX5_LIB "libmlx5.a") + vpp_plugin_find_library(dpdk MLX4_LIB "libmlx4.a") + string_append(DPDK_LINK_FLAGS "-Wl,--whole-archive,${IBVERBS_LIB},${MLX5_LIB},${MLX4_LIB} -Wl,--no-whole-archive,--exclude-libs,ALL") endif() endif() endif() @@ -131,10 +131,10 @@ add_vpp_plugin(dpdk SOURCES buffer.c main.c - thread.c device/cli.c device/common.c device/device.c + device/driver.c device/flow.c device/format.c device/init.c @@ -158,7 +158,7 @@ add_vpp_plugin(dpdk ${DPDK_LINK_LIBRARIES} LINK_LIBRARIES - ${OPENSSL_LIBRARIES} + ${OPENSSL_CRYPTO_LIBRARIES} COMPONENT vpp-plugin-dpdk diff --git a/src/plugins/dpdk/buffer.c b/src/plugins/dpdk/buffer.c index 97184519428..f3137a996d6 100644 --- a/src/plugins/dpdk/buffer.c +++ b/src/plugins/dpdk/buffer.c @@ -19,6 +19,7 @@ #include <rte_config.h> #include <rte_mbuf.h> #include <rte_ethdev.h> +#include <rte_cryptodev.h> #include <rte_vfio.h> #include <rte_version.h> @@ -115,6 +116,9 @@ dpdk_buffer_pool_init (vlib_main_t * vm, vlib_buffer_pool_t * bp) mp->populated_size++; nmp->populated_size++; } +#if RTE_VERSION >= RTE_VERSION_NUM(22, 3, 0, 0) + mp->flags &= ~RTE_MEMPOOL_F_NON_IO; +#endif /* call the object initializers */ rte_mempool_obj_iter (mp, rte_pktmbuf_init, 0); @@ -131,11 +135,11 @@ dpdk_buffer_pool_init (vlib_main_t * vm, vlib_buffer_pool_t * bp) { vlib_buffer_t *b; b = vlib_buffer_ptr_from_index (buffer_mem_start, bp->buffers[i], 0); - vlib_buffer_copy_template (b, &bp->buffer_template); + b->template = bp->buffer_template; } /* map DMA pages if at least one physical device exists */ - if (rte_eth_dev_count_avail ()) + if (rte_eth_dev_count_avail () || rte_cryptodev_count ()) { uword i; size_t page_sz; @@ -193,7 +197,7 @@ dpdk_ops_vpp_free (struct rte_mempool *mp) #endif static_always_inline void -dpdk_ops_vpp_enqueue_one (vlib_buffer_t * bt, void *obj) +dpdk_ops_vpp_enqueue_one (vlib_buffer_template_t *bt, void *obj) { /* Only non-replicated packets (b->ref_count == 1) expected */ @@ -201,7 +205,7 @@ dpdk_ops_vpp_enqueue_one (vlib_buffer_t * bt, void *obj) vlib_buffer_t *b = vlib_buffer_from_rte_mbuf (mb); ASSERT (b->ref_count == 1); ASSERT (b->buffer_pool_index == bt->buffer_pool_index); - vlib_buffer_copy_template (b, bt); + b->template = *bt; } int @@ -210,14 +214,14 @@ CLIB_MULTIARCH_FN (dpdk_ops_vpp_enqueue) (struct rte_mempool * mp, { const int batch_size = 32; vlib_main_t *vm = vlib_get_main (); - vlib_buffer_t bt; + vlib_buffer_template_t bt; u8 buffer_pool_index = mp->pool_id; vlib_buffer_pool_t *bp = vlib_get_buffer_pool (vm, buffer_pool_index); u32 bufs[batch_size]; u32 n_left = n; void *const *obj = obj_table; - vlib_buffer_copy_template (&bt, &bp->buffer_template); + bt = bp->buffer_template; while (n_left >= 4) { @@ -259,9 +263,9 @@ CLIB_MULTIARCH_FN (dpdk_ops_vpp_enqueue) (struct rte_mempool * mp, CLIB_MARCH_FN_REGISTRATION (dpdk_ops_vpp_enqueue); static_always_inline void -dpdk_ops_vpp_enqueue_no_cache_one (vlib_main_t * vm, struct rte_mempool *old, +dpdk_ops_vpp_enqueue_no_cache_one (vlib_main_t *vm, struct rte_mempool *old, struct rte_mempool *new, void *obj, - vlib_buffer_t * bt) + vlib_buffer_template_t *bt) { struct rte_mbuf *mb = obj; vlib_buffer_t *b = vlib_buffer_from_rte_mbuf (mb); @@ -269,7 +273,7 @@ dpdk_ops_vpp_enqueue_no_cache_one (vlib_main_t * vm, struct rte_mempool *old, if (clib_atomic_sub_fetch (&b->ref_count, 1) == 0) { u32 bi = vlib_get_buffer_index (vm, b); - vlib_buffer_copy_template (b, bt); + b->template = *bt; vlib_buffer_pool_put (vm, bt->buffer_pool_index, &bi, 1); return; } @@ -281,12 +285,12 @@ CLIB_MULTIARCH_FN (dpdk_ops_vpp_enqueue_no_cache) (struct rte_mempool * cmp, unsigned n) { vlib_main_t *vm = vlib_get_main (); - vlib_buffer_t bt; + vlib_buffer_template_t bt; struct rte_mempool *mp; mp = dpdk_mempool_by_buffer_pool_index[cmp->pool_id]; u8 buffer_pool_index = cmp->pool_id; vlib_buffer_pool_t *bp = vlib_get_buffer_pool (vm, buffer_pool_index); - vlib_buffer_copy_template (&bt, &bp->buffer_template); + bt = bp->buffer_template; while (n >= 4) { @@ -456,11 +460,9 @@ dpdk_buffer_pools_create (vlib_main_t * vm) ops.dequeue = dpdk_ops_vpp_dequeue_no_cache; rte_mempool_register_ops (&ops); - /* *INDENT-OFF* */ vec_foreach (bp, vm->buffer_main->buffer_pools) if (bp->start && (err = dpdk_buffer_pool_init (vm, bp))) return err; - /* *INDENT-ON* */ return 0; } diff --git a/src/plugins/dpdk/cryptodev/cryptodev.c b/src/plugins/dpdk/cryptodev/cryptodev.c index d52fa407ec5..43c2c879aab 100644 --- a/src/plugins/dpdk/cryptodev/cryptodev.c +++ b/src/plugins/dpdk/cryptodev/cryptodev.c @@ -29,7 +29,6 @@ #include <rte_cryptodev.h> #include <rte_crypto_sym.h> #include <rte_crypto.h> -#include <rte_cryptodev_pmd.h> #include <rte_config.h> #include "cryptodev.h" @@ -52,12 +51,19 @@ prepare_aead_xform (struct rte_crypto_sym_xform *xform, xform->type = RTE_CRYPTO_SYM_XFORM_AEAD; xform->next = 0; - if (key->alg != VNET_CRYPTO_ALG_AES_128_GCM && - key->alg != VNET_CRYPTO_ALG_AES_192_GCM && - key->alg != VNET_CRYPTO_ALG_AES_256_GCM) + if (key->alg == VNET_CRYPTO_ALG_AES_128_GCM || + key->alg == VNET_CRYPTO_ALG_AES_192_GCM || + key->alg == VNET_CRYPTO_ALG_AES_256_GCM) + { + aead_xform->algo = RTE_CRYPTO_AEAD_AES_GCM; + } + else if (key->alg == VNET_CRYPTO_ALG_CHACHA20_POLY1305) + { + aead_xform->algo = RTE_CRYPTO_AEAD_CHACHA20_POLY1305; + } + else return -1; - aead_xform->algo = RTE_CRYPTO_AEAD_AES_GCM; aead_xform->op = (op_type == CRYPTODEV_OP_TYPE_ENCRYPT) ? RTE_CRYPTO_AEAD_OP_ENCRYPT : RTE_CRYPTO_AEAD_OP_DECRYPT; aead_xform->aad_length = aad_len; @@ -135,7 +141,7 @@ prepare_linked_xform (struct rte_crypto_sym_xform *xforms, } static_always_inline void -cryptodev_session_del (struct rte_cryptodev_sym_session *sess) +cryptodev_session_del (cryptodev_session_t *sess) { u32 n_devs, i; @@ -145,9 +151,14 @@ cryptodev_session_del (struct rte_cryptodev_sym_session *sess) n_devs = rte_cryptodev_count (); for (i = 0; i < n_devs; i++) +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + if (rte_cryptodev_sym_session_free (i, sess) == 0) + break; +#else rte_cryptodev_sym_session_clear (i, sess); rte_cryptodev_sym_session_free (sess); +#endif } static int @@ -310,7 +321,7 @@ cryptodev_sess_handler (vlib_main_t *vm, vnet_crypto_key_op_t kop, if (cryptodev_check_supported_vnet_alg (key) == 0) return; - vec_validate (ckey->keys, idx); + vec_validate (ckey->keys, vec_len (cmt->per_numa_data) - 1); vec_foreach_index (i, ckey->keys) vec_validate (ckey->keys[i], CRYPTODEV_N_OP_TYPES - 1); } @@ -322,6 +333,59 @@ cryptodev_key_handler (vlib_main_t *vm, vnet_crypto_key_op_t kop, cryptodev_sess_handler (vm, kop, idx, 8); } +clib_error_t * +allocate_session_pools (u32 numa_node, + cryptodev_session_pool_t *sess_pools_elt, u32 len) +{ + cryptodev_main_t *cmt = &cryptodev_main; + u8 *name; + clib_error_t *error = NULL; + + name = format (0, "vcrypto_sess_pool_%u_%04x%c", numa_node, len, 0); +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + sess_pools_elt->sess_pool = rte_cryptodev_sym_session_pool_create ( + (char *) name, CRYPTODEV_NB_SESSION, cmt->sess_sz, 0, 0, numa_node); +#else + sess_pools_elt->sess_pool = rte_cryptodev_sym_session_pool_create ( + (char *) name, CRYPTODEV_NB_SESSION, 0, 0, 0, numa_node); +#endif + + if (!sess_pools_elt->sess_pool) + { + error = clib_error_return (0, "Not enough memory for mp %s", name); + goto clear_mempools; + } + vec_free (name); + +#if RTE_VERSION < RTE_VERSION_NUM(22, 11, 0, 0) + name = format (0, "crypto_sess_pool_%u_%04x%c", numa_node, len, 0); + sess_pools_elt->sess_priv_pool = rte_mempool_create ( + (char *) name, CRYPTODEV_NB_SESSION * (cmt->drivers_cnt), cmt->sess_sz, 0, + 0, NULL, NULL, NULL, NULL, numa_node, 0); + + if (!sess_pools_elt->sess_priv_pool) + { + error = clib_error_return (0, "Not enough memory for mp %s", name); + goto clear_mempools; + } + vec_free (name); +#endif + +clear_mempools: + if (error) + { + vec_free (name); + if (sess_pools_elt->sess_pool) + rte_mempool_free (sess_pools_elt->sess_pool); +#if RTE_VERSION < RTE_VERSION_NUM(22, 11, 0, 0) + if (sess_pools_elt->sess_priv_pool) + rte_mempool_free (sess_pools_elt->sess_priv_pool); +#endif + return error; + } + return 0; +} + int cryptodev_session_create (vlib_main_t *vm, vnet_crypto_key_index_t idx, u32 aad_len) @@ -330,52 +394,106 @@ cryptodev_session_create (vlib_main_t *vm, vnet_crypto_key_index_t idx, cryptodev_numa_data_t *numa_data; cryptodev_inst_t *dev_inst; vnet_crypto_key_t *key = vnet_crypto_get_key (idx); - struct rte_mempool *sess_pool, *sess_priv_pool; + struct rte_mempool *sess_pool; + cryptodev_session_pool_t *sess_pools_elt; cryptodev_key_t *ckey = vec_elt_at_index (cmt->keys, idx); struct rte_crypto_sym_xform xforms_enc[2] = { { 0 } }; struct rte_crypto_sym_xform xforms_dec[2] = { { 0 } }; - struct rte_cryptodev_sym_session *sessions[CRYPTODEV_N_OP_TYPES] = { 0 }; + cryptodev_session_t *sessions[CRYPTODEV_N_OP_TYPES] = { 0 }; +#if RTE_VERSION < RTE_VERSION_NUM(22, 11, 0, 0) + struct rte_mempool *sess_priv_pool; + struct rte_cryptodev_info dev_info; +#endif u32 numa_node = vm->numa_node; - int ret; + clib_error_t *error; + int ret = 0; + u8 found = 0; numa_data = vec_elt_at_index (cmt->per_numa_data, numa_node); - sess_pool = numa_data->sess_pool; - sess_priv_pool = numa_data->sess_priv_pool; - sessions[CRYPTODEV_OP_TYPE_ENCRYPT] = - rte_cryptodev_sym_session_create (sess_pool); - if (!sessions[CRYPTODEV_OP_TYPE_ENCRYPT]) + clib_spinlock_lock (&cmt->tlock); + vec_foreach (sess_pools_elt, numa_data->sess_pools) { - ret = -1; - goto clear_key; + if (sess_pools_elt->sess_pool == NULL) + { + error = allocate_session_pools (numa_node, sess_pools_elt, + vec_len (numa_data->sess_pools) - 1); + if (error) + { + ret = -1; + goto clear_key; + } + } + if (rte_mempool_avail_count (sess_pools_elt->sess_pool) >= 2) + { + found = 1; + break; + } } - sessions[CRYPTODEV_OP_TYPE_DECRYPT] = - rte_cryptodev_sym_session_create (sess_pool); - if (!sessions[CRYPTODEV_OP_TYPE_DECRYPT]) + if (found == 0) { - ret = -1; - goto clear_key; + vec_add2 (numa_data->sess_pools, sess_pools_elt, 1); + error = allocate_session_pools (numa_node, sess_pools_elt, + vec_len (numa_data->sess_pools) - 1); + if (error) + { + ret = -1; + goto clear_key; + } } + sess_pool = sess_pools_elt->sess_pool; +#if RTE_VERSION < RTE_VERSION_NUM(22, 11, 0, 0) + sess_priv_pool = sess_pools_elt->sess_priv_pool; + + sessions[CRYPTODEV_OP_TYPE_ENCRYPT] = + rte_cryptodev_sym_session_create (sess_pool); + + sessions[CRYPTODEV_OP_TYPE_DECRYPT] = + rte_cryptodev_sym_session_create (sess_pool); +#endif + if (key->type == VNET_CRYPTO_KEY_TYPE_LINK) ret = prepare_linked_xform (xforms_enc, CRYPTODEV_OP_TYPE_ENCRYPT, key); else ret = prepare_aead_xform (xforms_enc, CRYPTODEV_OP_TYPE_ENCRYPT, key, aad_len); if (ret) - return 0; + { + ret = -1; + goto clear_key; + } if (key->type == VNET_CRYPTO_KEY_TYPE_LINK) prepare_linked_xform (xforms_dec, CRYPTODEV_OP_TYPE_DECRYPT, key); else prepare_aead_xform (xforms_dec, CRYPTODEV_OP_TYPE_DECRYPT, key, aad_len); +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + dev_inst = vec_elt_at_index (cmt->cryptodev_inst, 0); + u32 dev_id = dev_inst->dev_id; + sessions[CRYPTODEV_OP_TYPE_ENCRYPT] = + rte_cryptodev_sym_session_create (dev_id, xforms_enc, sess_pool); + sessions[CRYPTODEV_OP_TYPE_DECRYPT] = + rte_cryptodev_sym_session_create (dev_id, xforms_dec, sess_pool); + if (!sessions[CRYPTODEV_OP_TYPE_ENCRYPT] || + !sessions[CRYPTODEV_OP_TYPE_DECRYPT]) + { + ret = -1; + goto clear_key; + } + + rte_cryptodev_sym_session_opaque_data_set ( + sessions[CRYPTODEV_OP_TYPE_ENCRYPT], aad_len); + rte_cryptodev_sym_session_opaque_data_set ( + sessions[CRYPTODEV_OP_TYPE_DECRYPT], aad_len); +#else vec_foreach (dev_inst, cmt->cryptodev_inst) { u32 dev_id = dev_inst->dev_id; - struct rte_cryptodev *cdev = rte_cryptodev_pmd_get_dev (dev_id); - u32 driver_id = cdev->driver_id; + rte_cryptodev_info_get (dev_id, &dev_info); + u32 driver_id = dev_info.driver_id; /* if the session is already configured for the driver type, avoid configuring it again to increase the session data's refcnt */ @@ -390,11 +508,12 @@ cryptodev_session_create (vlib_main_t *vm, vnet_crypto_key_index_t idx, dev_id, sessions[CRYPTODEV_OP_TYPE_DECRYPT], xforms_dec, sess_priv_pool); if (ret < 0) - return ret; + goto clear_key; } sessions[CRYPTODEV_OP_TYPE_ENCRYPT]->opaque_data = aad_len; sessions[CRYPTODEV_OP_TYPE_DECRYPT]->opaque_data = aad_len; +#endif CLIB_MEMORY_STORE_BARRIER (); ckey->keys[numa_node][CRYPTODEV_OP_TYPE_ENCRYPT] = @@ -408,6 +527,7 @@ clear_key: cryptodev_session_del (sessions[CRYPTODEV_OP_TYPE_ENCRYPT]); cryptodev_session_del (sessions[CRYPTODEV_OP_TYPE_DECRYPT]); } + clib_spinlock_unlock (&cmt->tlock); return ret; } @@ -459,14 +579,14 @@ cryptodev_assign_resource (cryptodev_engine_thread_t * cet, return -EBUSY; vec_foreach_index (idx, cmt->cryptodev_inst) - { - cinst = cmt->cryptodev_inst + idx; - if (cinst->dev_id == cet->cryptodev_id && - cinst->q_id == cet->cryptodev_q) - break; - } + { + cinst = cmt->cryptodev_inst + idx; + if (cinst->dev_id == cet->cryptodev_id && + cinst->q_id == cet->cryptodev_q) + break; + } /* invalid existing worker resource assignment */ - if (idx == vec_len (cmt->cryptodev_inst)) + if (idx >= vec_len (cmt->cryptodev_inst)) return -EINVAL; clib_spinlock_lock (&cmt->tlock); clib_bitmap_set_no_check (cmt->active_cdev_inst_mask, idx, 0); @@ -547,6 +667,90 @@ VLIB_CLI_COMMAND (show_cryptodev_assignment, static) = { }; static clib_error_t * +cryptodev_show_cache_rings_fn (vlib_main_t *vm, unformat_input_t *input, + vlib_cli_command_t *cmd) +{ + cryptodev_main_t *cmt = &cryptodev_main; + u32 thread_index = 0; + u16 i; + vec_foreach_index (thread_index, cmt->per_thread_data) + { + cryptodev_engine_thread_t *cet = cmt->per_thread_data + thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + u16 head = ring->head; + u16 tail = ring->tail; + u16 n_cached = (CRYPTODEV_CACHE_QUEUE_SIZE - tail + head) & + CRYPTODEV_CACHE_QUEUE_MASK; + + u16 enq_head = ring->enq_head; + u16 deq_tail = ring->deq_tail; + u16 n_frames_inflight = + (enq_head == deq_tail) ? + 0 : + ((CRYPTODEV_CACHE_QUEUE_SIZE + enq_head - deq_tail) & + CRYPTODEV_CACHE_QUEUE_MASK); + /* even if some elements of dequeued frame are still pending for deq + * we consider the frame as processed */ + u16 n_frames_processed = + ((tail == deq_tail) && (ring->frames[deq_tail].f == 0)) ? + 0 : + ((CRYPTODEV_CACHE_QUEUE_SIZE - tail + deq_tail) & + CRYPTODEV_CACHE_QUEUE_MASK) + + 1; + /* even if some elements of enqueued frame are still pending for enq + * we consider the frame as enqueued */ + u16 n_frames_pending = + (head == enq_head) ? 0 : + ((CRYPTODEV_CACHE_QUEUE_SIZE - enq_head + head) & + CRYPTODEV_CACHE_QUEUE_MASK) - + 1; + + u16 elts_to_enq = + (ring->frames[enq_head].n_elts - ring->frames[enq_head].enq_elts_head); + u16 elts_to_deq = + (ring->frames[deq_tail].n_elts - ring->frames[deq_tail].deq_elts_tail); + + u32 elts_total = 0; + + for (i = 0; i < CRYPTODEV_CACHE_QUEUE_SIZE; i++) + elts_total += ring->frames[i].n_elts; + + if (vlib_num_workers () > 0 && thread_index == 0) + continue; + + vlib_cli_output (vm, "\n\n"); + vlib_cli_output (vm, "Frames cached in the ring: %u", n_cached); + vlib_cli_output (vm, "Frames cached but not processed: %u", + n_frames_pending); + vlib_cli_output (vm, "Frames inflight: %u", n_frames_inflight); + vlib_cli_output (vm, "Frames processed: %u", n_frames_processed); + vlib_cli_output (vm, "Elements total: %u", elts_total); + vlib_cli_output (vm, "Elements inflight: %u", cet->inflight); + vlib_cli_output (vm, "Head index: %u", head); + vlib_cli_output (vm, "Tail index: %u", tail); + vlib_cli_output (vm, "Current frame index beeing enqueued: %u", + enq_head); + vlib_cli_output (vm, "Current frame index being dequeued: %u", deq_tail); + vlib_cli_output (vm, + "Elements in current frame to be enqueued: %u, waiting " + "to be enqueued: %u", + ring->frames[enq_head].n_elts, elts_to_enq); + vlib_cli_output (vm, + "Elements in current frame to be dequeued: %u, waiting " + "to be dequeued: %u", + ring->frames[deq_tail].n_elts, elts_to_deq); + vlib_cli_output (vm, "\n\n"); + } + return 0; +} + +VLIB_CLI_COMMAND (show_cryptodev_sw_rings, static) = { + .path = "show cryptodev cache status", + .short_help = "show status of all cryptodev cache rings", + .function = cryptodev_show_cache_rings_fn, +}; + +static clib_error_t * cryptodev_set_assignment_fn (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd) { @@ -643,6 +847,15 @@ cryptodev_configure (vlib_main_t *vm, u32 cryptodev_id) rte_cryptodev_info_get (cryptodev_id, &info); + /* Starting from DPDK 22.11, VPP does not allow heterogeneous crypto devices + anymore. Only devices that have the same driver type as the first + initialized device can be initialized. + */ +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + if (cmt->drivers_cnt == 1 && cmt->driver_id != info.driver_id) + return -1; +#endif + if (!(info.feature_flags & RTE_CRYPTODEV_FF_SYMMETRIC_CRYPTO)) return -1; @@ -656,7 +869,9 @@ cryptodev_configure (vlib_main_t *vm, u32 cryptodev_id) struct rte_cryptodev_qp_conf qp_cfg; qp_cfg.mp_session = 0; +#if RTE_VERSION < RTE_VERSION_NUM(22, 11, 0, 0) qp_cfg.mp_session_private = 0; +#endif qp_cfg.nb_descriptors = CRYPTODEV_NB_CRYPTO_OPS; ret = rte_cryptodev_queue_pair_setup (cryptodev_id, i, &qp_cfg, @@ -675,16 +890,30 @@ cryptodev_configure (vlib_main_t *vm, u32 cryptodev_id) /* start the device */ rte_cryptodev_start (cryptodev_id); +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + if (cmt->drivers_cnt == 0) + { + cmt->drivers_cnt = 1; + cmt->driver_id = info.driver_id; + cmt->sess_sz = rte_cryptodev_sym_get_private_session_size (cryptodev_id); + } +#endif + for (i = 0; i < info.max_nb_queue_pairs; i++) { cryptodev_inst_t *cdev_inst; +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + const char *dev_name = rte_dev_name (info.device); +#else + const char *dev_name = info.device->name; +#endif vec_add2(cmt->cryptodev_inst, cdev_inst, 1); - cdev_inst->desc = vec_new (char, strlen (info.device->name) + 10); + cdev_inst->desc = vec_new (char, strlen (dev_name) + 10); cdev_inst->dev_id = cryptodev_id; cdev_inst->q_id = i; - snprintf (cdev_inst->desc, strlen (info.device->name) + 9, - "%s_q%u", info.device->name, i); + snprintf (cdev_inst->desc, strlen (dev_name) + 9, "%s_q%u", + info.device->name, i); } return 0; @@ -1016,46 +1245,26 @@ cryptodev_probe (vlib_main_t *vm, u32 n_workers) return 0; } +#if RTE_VERSION < RTE_VERSION_NUM(22, 11, 0, 0) static void -cryptodev_get_max_sz (u32 *max_sess_sz, u32 *max_dp_sz) +is_drv_unique (u32 driver_id, u32 **unique_drivers) { - cryptodev_main_t *cmt = &cryptodev_main; - cryptodev_inst_t *cinst; - u32 max_sess = 0, max_dp = 0; + u32 *unique_elt; + u8 found = 0; - vec_foreach (cinst, cmt->cryptodev_inst) + vec_foreach (unique_elt, *unique_drivers) { - u32 sess_sz = rte_cryptodev_sym_get_private_session_size (cinst->dev_id); - u32 dp_sz = rte_cryptodev_get_raw_dp_ctx_size (cinst->dev_id); - - max_sess = clib_max (sess_sz, max_sess); - max_dp = clib_max (dp_sz, max_dp); + if (*unique_elt == driver_id) + { + found = 1; + break; + } } - *max_sess_sz = max_sess; - *max_dp_sz = max_dp; -} - -static void -dpdk_disable_cryptodev_engine (vlib_main_t *vm) -{ - vlib_thread_main_t *tm = vlib_get_thread_main (); - cryptodev_main_t *cmt = &cryptodev_main; - u32 i; - - for (i = (vlib_num_workers () > 0); i < tm->n_vlib_mains; i++) - { - u32 numa = vlib_get_main_by_index (i)->numa_node; - cryptodev_numa_data_t *numa_data; - - vec_validate (cmt->per_numa_data, numa); - numa_data = cmt->per_numa_data + numa; - if (numa_data->sess_pool) - rte_mempool_free (numa_data->sess_pool); - if (numa_data->sess_priv_pool) - rte_mempool_free (numa_data->sess_priv_pool); - } + if (!found) + vec_add1 (*unique_drivers, driver_id); } +#endif clib_error_t * dpdk_cryptodev_init (vlib_main_t * vm) @@ -1064,30 +1273,53 @@ dpdk_cryptodev_init (vlib_main_t * vm) vlib_thread_main_t *tm = vlib_get_thread_main (); cryptodev_engine_thread_t *cet; cryptodev_numa_data_t *numa_data; - struct rte_mempool *mp; + u32 node; + u8 nodes = 0; u32 skip_master = vlib_num_workers () > 0; u32 n_workers = tm->n_vlib_mains - skip_master; - u32 numa = vm->numa_node; - u32 sess_sz, dp_sz; u32 eidx; u32 i; - u8 *name = 0; clib_error_t *error; cmt->iova_mode = rte_eal_iova_mode (); - vec_validate (cmt->per_numa_data, vm->numa_node); + clib_bitmap_foreach (node, tm->cpu_socket_bitmap) + { + if (node >= nodes) + nodes = node; + } + + vec_validate (cmt->per_numa_data, nodes); + vec_foreach (numa_data, cmt->per_numa_data) + { + vec_validate (numa_data->sess_pools, 0); + } /* probe all cryptodev devices and get queue info */ if (cryptodev_probe (vm, n_workers) < 0) + return 0; + +#if RTE_VERSION < RTE_VERSION_NUM(22, 11, 0, 0) + struct rte_cryptodev_info dev_info; + cryptodev_inst_t *dev_inst; + u32 *unique_drivers = 0; + vec_foreach (dev_inst, cmt->cryptodev_inst) { - error = clib_error_return (0, "Failed to configure cryptodev"); - goto err_handling; + u32 dev_id = dev_inst->dev_id; + rte_cryptodev_info_get (dev_id, &dev_info); + u32 driver_id = dev_info.driver_id; + is_drv_unique (driver_id, &unique_drivers); + + u32 sess_sz = + rte_cryptodev_sym_get_private_session_size (dev_inst->dev_id); + cmt->sess_sz = clib_max (cmt->sess_sz, sess_sz); } - cryptodev_get_max_sz (&sess_sz, &dp_sz); + cmt->drivers_cnt = vec_len (unique_drivers); + vec_free (unique_drivers); +#endif - clib_bitmap_vec_validate (cmt->active_cdev_inst_mask, tm->n_vlib_mains); + clib_bitmap_vec_validate (cmt->active_cdev_inst_mask, n_workers); clib_spinlock_init (&cmt->tlock); vec_validate_aligned(cmt->per_thread_data, tm->n_vlib_mains - 1, @@ -1095,46 +1327,13 @@ dpdk_cryptodev_init (vlib_main_t * vm) for (i = skip_master; i < tm->n_vlib_mains; i++) { cet = cmt->per_thread_data + i; - numa = vlib_get_main_by_index (i)->numa_node; - vec_validate (cmt->per_numa_data, numa); - numa_data = vec_elt_at_index (cmt->per_numa_data, numa); - - if (!numa_data->sess_pool) + if (cryptodev_assign_resource (cet, 0, CRYPTODEV_RESOURCE_ASSIGN_AUTO) < + 0) { - /* create session pool for the numa node */ - name = format (0, "vcryptodev_sess_pool_%u%c", numa, 0); - mp = rte_cryptodev_sym_session_pool_create ( - (char *) name, CRYPTODEV_NB_SESSION, 0, 0, 0, numa); - if (!mp) - { - error = - clib_error_return (0, "Not enough memory for mp %s", name); - goto err_handling; - } - vec_free (name); - - numa_data->sess_pool = mp; - - /* create session private pool for the numa node */ - name = format (0, "cryptodev_sess_pool_%u%c", numa, 0); - mp = - rte_mempool_create ((char *) name, CRYPTODEV_NB_SESSION, sess_sz, - 0, 0, NULL, NULL, NULL, NULL, numa, 0); - if (!mp) - { - error = - clib_error_return (0, "Not enough memory for mp %s", name); - vec_free (name); - goto err_handling; - } - - vec_free (name); - - numa_data->sess_priv_pool = mp; + error = clib_error_return (0, "Failed to configure cryptodev"); + goto err_handling; } - - cryptodev_assign_resource (cet, 0, CRYPTODEV_RESOURCE_ASSIGN_AUTO); } /* register handler */ @@ -1154,13 +1353,10 @@ dpdk_cryptodev_init (vlib_main_t * vm) /* this engine is only enabled when cryptodev device(s) are presented in * startup.conf. Assume it is wanted to be used, turn on async mode here. */ - vnet_crypto_request_async_mode (1); ipsec_set_async_mode (1); return 0; err_handling: - dpdk_disable_cryptodev_engine (vm); - return error; } diff --git a/src/plugins/dpdk/cryptodev/cryptodev.h b/src/plugins/dpdk/cryptodev/cryptodev.h index 3b47b43f538..7cd525dac56 100644 --- a/src/plugins/dpdk/cryptodev/cryptodev.h +++ b/src/plugins/dpdk/cryptodev/cryptodev.h @@ -26,11 +26,13 @@ #define CRYPTODEV_CACHE_QUEUE_MASK (VNET_CRYPTO_FRAME_POOL_SIZE - 1) #define CRYPTODEV_MAX_INFLIGHT (CRYPTODEV_NB_CRYPTO_OPS - 1) #define CRYPTODEV_AAD_MASK (CRYPTODEV_NB_CRYPTO_OPS - 1) -#define CRYPTODEV_DEQ_CACHE_SZ 32 -#define CRYPTODEV_NB_SESSION 10240 +#define CRYPTODE_ENQ_MAX 64 +#define CRYPTODE_DEQ_MAX 64 +#define CRYPTODEV_NB_SESSION 4096 #define CRYPTODEV_MAX_IV_SIZE 16 #define CRYPTODEV_MAX_AAD_SIZE 16 #define CRYPTODEV_MAX_N_SGL 8 /**< maximum number of segments */ +#define CRYPTODEV_MAX_PROCESED_IN_CACHE_QUEUE 8 #define CRYPTODEV_IV_OFFSET (offsetof (cryptodev_op_t, iv)) #define CRYPTODEV_AAD_OFFSET (offsetof (cryptodev_op_t, aad)) @@ -43,7 +45,10 @@ _ (AES_192_GCM, AEAD, AES_GCM, 12, 16, 8, 24) \ _ (AES_192_GCM, AEAD, AES_GCM, 12, 16, 12, 24) \ _ (AES_256_GCM, AEAD, AES_GCM, 12, 16, 8, 32) \ - _ (AES_256_GCM, AEAD, AES_GCM, 12, 16, 12, 32) + _ (AES_256_GCM, AEAD, AES_GCM, 12, 16, 12, 32) \ + _ (CHACHA20_POLY1305, AEAD, CHACHA20_POLY1305, 12, 16, 0, 32) \ + _ (CHACHA20_POLY1305, AEAD, CHACHA20_POLY1305, 12, 16, 8, 32) \ + _ (CHACHA20_POLY1305, AEAD, CHACHA20_POLY1305, 12, 16, 12, 32) /** * crypto (alg, cryptodev_alg, key_size), hash (alg, digest-size) @@ -66,7 +71,10 @@ _ (AES_256_CBC, AES_CBC, 32, SHA384, 24) \ _ (AES_128_CBC, AES_CBC, 16, SHA512, 32) \ _ (AES_192_CBC, AES_CBC, 24, SHA512, 32) \ - _ (AES_256_CBC, AES_CBC, 32, SHA512, 32) + _ (AES_256_CBC, AES_CBC, 32, SHA512, 32) \ + _ (AES_128_CTR, AES_CTR, 16, SHA1, 12) \ + _ (AES_192_CTR, AES_CTR, 24, SHA1, 12) \ + _ (AES_256_CTR, AES_CTR, 32, SHA1, 12) typedef enum { @@ -75,10 +83,16 @@ typedef enum CRYPTODEV_N_OP_TYPES, } cryptodev_op_type_t; +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) +typedef void cryptodev_session_t; +#else +typedef struct rte_cryptodev_sym_session cryptodev_session_t; +#endif + /* Cryptodev session data, one data per direction per numa */ typedef struct { - struct rte_cryptodev_sym_session ***keys; + cryptodev_session_t ***keys; } cryptodev_key_t; /* Replicate DPDK rte_cryptodev_sym_capability structure with key size ranges @@ -119,7 +133,14 @@ typedef struct typedef struct { struct rte_mempool *sess_pool; +#if RTE_VERSION < RTE_VERSION_NUM(22, 11, 0, 0) struct rte_mempool *sess_priv_pool; +#endif +} cryptodev_session_pool_t; + +typedef struct +{ + cryptodev_session_pool_t *sess_pools; } cryptodev_numa_data_t; typedef struct @@ -135,26 +156,71 @@ typedef struct typedef struct { - CLIB_CACHE_LINE_ALIGN_MARK (cacheline0); - vlib_buffer_t *b[VNET_CRYPTO_FRAME_SIZE]; + vnet_crypto_async_frame_t *f; union { struct { - cryptodev_op_t **cops; - struct rte_mempool *cop_pool; - struct rte_ring *ring; + /* index of frame elt where enque to + * the crypto engine is happening */ + u8 enq_elts_head; + /* index of the frame elt where dequeue + * from the crypto engine is happening */ + u8 deq_elts_tail; + u8 elts_inflight; + + u8 op_type; + u8 aad_len; + u8 n_elts; + u16 reserved; }; + u64 raw; + }; + + u64 frame_elts_errs_mask; +} cryptodev_cache_ring_elt_t; + +typedef struct +{ + cryptodev_cache_ring_elt_t frames[VNET_CRYPTO_FRAME_POOL_SIZE]; + + union + { + struct + { + /* head of the cache ring */ + u16 head; + /* tail of the cache ring */ + u16 tail; + /* index of the frame where enqueue + * to the crypto engine is happening */ + u16 enq_head; + /* index of the frame where dequeue + * from the crypto engine is happening */ + u16 deq_tail; + }; + u64 raw; + }; +} cryptodev_cache_ring_t; + +typedef struct +{ + CLIB_CACHE_LINE_ALIGN_MARK (cacheline0); + vlib_buffer_t *b[VNET_CRYPTO_FRAME_SIZE]; + union + { + struct rte_mempool *cop_pool; struct { struct rte_crypto_raw_dp_ctx *ctx; - struct rte_ring *cached_frame; u16 aad_index; u8 *aad_buf; u64 aad_phy_addr; - struct rte_cryptodev_sym_session *reset_sess; + cryptodev_session_t *reset_sess; }; }; + + cryptodev_cache_ring_t cache_ring; u16 cryptodev_id; u16 cryptodev_q; u16 inflight; @@ -170,20 +236,122 @@ typedef struct clib_bitmap_t *active_cdev_inst_mask; clib_spinlock_t tlock; cryptodev_capability_t *supported_caps; + u32 sess_sz; + u32 drivers_cnt; u8 is_raw_api; +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + u8 driver_id; +#endif } cryptodev_main_t; extern cryptodev_main_t cryptodev_main; +#define CRYPTODEV_CACHE_RING_GET_FRAME(r, i) \ + ((r)->frames[(i) &CRYPTODEV_CACHE_QUEUE_MASK].f) + +#define CRYPTODEV_CACHE_RING_GET_ERR_MASK(r, i) \ + ((r)->frames[(i) &CRYPTODEV_CACHE_QUEUE_MASK].frame_elts_errs_mask) + +#define CRYPTODEV_CACHE_RING_GET_FRAME_ELTS_INFLIGHT(r, i) \ + (((r)->frames[(i) &CRYPTODEV_CACHE_QUEUE_MASK].enq_elts_head) - \ + ((r)->frames[(i) &CRYPTODEV_CACHE_QUEUE_MASK].deq_elts_tail)) + static_always_inline void -cryptodev_mark_frame_err_status (vnet_crypto_async_frame_t *f, - vnet_crypto_op_status_t s) +cryptodev_cache_ring_update_enq_head (cryptodev_cache_ring_t *r, + vnet_crypto_async_frame_t *f) +{ + if (r->frames[r->enq_head].enq_elts_head == f->n_elts) + { + r->enq_head++; + r->enq_head &= CRYPTODEV_CACHE_QUEUE_MASK; + f->state = VNET_CRYPTO_FRAME_STATE_NOT_PROCESSED; + } +} + +static_always_inline bool +cryptodev_cache_ring_update_deq_tail (cryptodev_cache_ring_t *r, + u16 *const deq) +{ + if (r->frames[*deq].deq_elts_tail == r->frames[*deq].n_elts) + { + *deq += 1; + *deq &= CRYPTODEV_CACHE_QUEUE_MASK; + return 1; + } + + return 0; +} +static_always_inline u64 +cryptodev_mark_frame_fill_err (vnet_crypto_async_frame_t *f, u64 current_err, + u16 index, u16 n, vnet_crypto_op_status_t op_s) +{ + u64 err = current_err; + u16 i; + + ERROR_ASSERT (index + n <= VNET_CRYPTO_FRAME_SIZE); + ERROR_ASSERT (op_s != VNET_CRYPTO_OP_STATUS_COMPLETED); + + for (i = index; i < (index + n); i++) + f->elts[i].status = op_s; + + err |= (~(~(0ull) << n) << index); + + return err; +} + +static_always_inline cryptodev_cache_ring_elt_t * +cryptodev_cache_ring_push (cryptodev_cache_ring_t *r, + vnet_crypto_async_frame_t *f) +{ + u16 head = r->head; + u16 tail = r->tail; + + cryptodev_cache_ring_elt_t *ring_elt = &r->frames[head]; + /** + * in debug mode we do the ring sanity test when a frame is enqueued to + * the ring. + **/ +#if CLIB_DEBUG > 0 + u16 n_cached = (head >= tail) ? (head - tail) : + (CRYPTODEV_CACHE_QUEUE_MASK - tail + head); + ERROR_ASSERT (n_cached < CRYPTODEV_CACHE_QUEUE_SIZE); + ERROR_ASSERT (r->raw == 0 && r->frames[head].raw == 0 && + r->frames[head].f == 0); +#endif + /*the ring capacity is CRYPTODEV_CACHE_QUEUE_SIZE - 1*/ + if (PREDICT_FALSE (head + 1) == tail) + return 0; + + ring_elt->f = f; + ring_elt->n_elts = f->n_elts; + /* update head */ + r->head++; + r->head &= CRYPTODEV_CACHE_QUEUE_MASK; + return ring_elt; +} + +static_always_inline vnet_crypto_async_frame_t * +cryptodev_cache_ring_pop (cryptodev_cache_ring_t *r) { - u32 n_elts = f->n_elts, i; + vnet_crypto_async_frame_t *f; + u16 tail = r->tail; + cryptodev_cache_ring_elt_t *ring_elt = &r->frames[tail]; + + ERROR_ASSERT (r->frames[r->head].raw == 0 ? r->head != tail : 1); + ERROR_ASSERT (r->frames[tail].raw != 0); + ERROR_ASSERT (ring_elt->deq_elts_tail == ring_elt->enq_elts_head && + ring_elt->deq_elts_tail == ring_elt->n_elts); + + f = CRYPTODEV_CACHE_RING_GET_FRAME (r, tail); + f->state = CRYPTODEV_CACHE_RING_GET_ERR_MASK (r, r->tail) == 0 ? + VNET_CRYPTO_FRAME_STATE_SUCCESS : + VNET_CRYPTO_FRAME_STATE_ELT_ERROR; + + clib_memset (ring_elt, 0, sizeof (*ring_elt)); + r->tail++; + r->tail &= CRYPTODEV_CACHE_QUEUE_MASK; - for (i = 0; i < n_elts; i++) - f->elts[i].status = s; - f->state = VNET_CRYPTO_FRAME_STATE_NOT_PROCESSED; + return f; } int cryptodev_session_create (vlib_main_t *vm, vnet_crypto_key_index_t idx, diff --git a/src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c b/src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c index 4545e24fc83..8d55e4fbf0f 100644 --- a/src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c +++ b/src/plugins/dpdk/cryptodev/cryptodev_op_data_path.c @@ -27,7 +27,6 @@ #include <rte_cryptodev.h> #include <rte_crypto_sym.h> #include <rte_crypto.h> -#include <rte_cryptodev_pmd.h> #include <rte_ring_peek_zc.h> #include <rte_config.h> @@ -68,6 +67,23 @@ cryptodev_get_iova (clib_pmalloc_main_t *pm, enum rte_iova_mode mode, } static_always_inline void +cryptodev_validate_mbuf (struct rte_mbuf *mb, vlib_buffer_t *b) +{ + /* on vnet side vlib_buffer current_length is updated by cipher padding and + * icv_sh. mbuf needs to be sync with these changes */ + u16 data_len = b->current_length + + (b->data + b->current_data - rte_pktmbuf_mtod (mb, u8 *)); + + /* for input nodes that are not dpdk-input, it is possible the mbuf + * was updated before as one of the chained mbufs. Setting nb_segs + * to 1 here to prevent the cryptodev PMD to access potentially + * invalid m_src->next pointers. + */ + mb->nb_segs = 1; + mb->pkt_len = mb->data_len = data_len; +} + +static_always_inline void cryptodev_validate_mbuf_chain (vlib_main_t *vm, struct rte_mbuf *mb, vlib_buffer_t *b) { @@ -125,39 +141,66 @@ cryptodev_frame_linked_algs_enqueue (vlib_main_t *vm, cryptodev_op_type_t op_type) { cryptodev_main_t *cmt = &cryptodev_main; + cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + ERROR_ASSERT (frame != 0); + ERROR_ASSERT (frame->n_elts > 0); + cryptodev_cache_ring_elt_t *ring_elt = + cryptodev_cache_ring_push (ring, frame); + + if (PREDICT_FALSE (ring_elt == NULL)) + return -1; + + ring_elt->aad_len = 1; + ring_elt->op_type = (u8) op_type; + return 0; +} + +static_always_inline void +cryptodev_frame_linked_algs_enqueue_internal (vlib_main_t *vm, + vnet_crypto_async_frame_t *frame, + cryptodev_op_type_t op_type) +{ + cryptodev_main_t *cmt = &cryptodev_main; clib_pmalloc_main_t *pm = vm->physmem_main.pmalloc_main; cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + u16 *const enq = &ring->enq_head; vnet_crypto_async_frame_elt_t *fe; - struct rte_cryptodev_sym_session *sess = 0; - cryptodev_op_t **cop; - u32 *bi; + cryptodev_session_t *sess = 0; + cryptodev_op_t *cops[CRYPTODE_ENQ_MAX] = {}; + cryptodev_op_t **cop = cops; + u32 *bi = 0; u32 n_enqueue, n_elts; u32 last_key_index = ~0; + u32 max_to_enq; if (PREDICT_FALSE (frame == 0 || frame->n_elts == 0)) - return -1; - n_elts = frame->n_elts; + return; - if (PREDICT_FALSE (CRYPTODEV_NB_CRYPTO_OPS - cet->inflight < n_elts)) - { - cryptodev_mark_frame_err_status (frame, - VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); - return -1; - } + max_to_enq = clib_min (CRYPTODE_ENQ_MAX, + frame->n_elts - ring->frames[*enq].enq_elts_head); + + if (cet->inflight + max_to_enq > CRYPTODEV_MAX_INFLIGHT) + return; + + n_elts = max_to_enq; if (PREDICT_FALSE ( - rte_mempool_get_bulk (cet->cop_pool, (void **) cet->cops, n_elts) < 0)) + rte_mempool_get_bulk (cet->cop_pool, (void **) cops, n_elts) < 0)) { - cryptodev_mark_frame_err_status (frame, - VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); - return -1; + cryptodev_mark_frame_fill_err ( + frame, ring->frames[*enq].frame_elts_errs_mask, + ring->frames[*enq].enq_elts_head, max_to_enq, + VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + ring->frames[*enq].enq_elts_head += max_to_enq; + ring->frames[*enq].deq_elts_tail += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); + return; } - cop = cet->cops; - fe = frame->elts; - bi = frame->buffer_indices; - cop[0]->frame = frame; - cop[0]->n_elts = n_elts; + fe = frame->elts + ring->frames[*enq].enq_elts_head; + bi = frame->buffer_indices + ring->frames[*enq].enq_elts_head; while (n_elts) { @@ -169,8 +212,8 @@ cryptodev_frame_linked_algs_enqueue (vlib_main_t *vm, if (n_elts > 2) { - CLIB_PREFETCH (cop[1], CLIB_CACHE_LINE_BYTES * 3, STORE); - CLIB_PREFETCH (cop[2], CLIB_CACHE_LINE_BYTES * 3, STORE); + CLIB_PREFETCH (cop[1], sizeof (*cop[1]), STORE); + CLIB_PREFETCH (cop[2], sizeof (*cop[2]), STORE); clib_prefetch_load (&fe[1]); clib_prefetch_load (&fe[2]); } @@ -184,9 +227,11 @@ cryptodev_frame_linked_algs_enqueue (vlib_main_t *vm, if (PREDICT_FALSE ( cryptodev_session_create (vm, last_key_index, 0) < 0)) { - cryptodev_mark_frame_err_status ( - frame, VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); - return -1; + cryptodev_mark_frame_fill_err ( + frame, ring->frames[*enq].frame_elts_errs_mask, + ring->frames[*enq].enq_elts_head, max_to_enq, + VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + goto error_exit; } } sess = key->keys[vm->numa_node][op_type]; @@ -216,26 +261,29 @@ cryptodev_frame_linked_algs_enqueue (vlib_main_t *vm, if (PREDICT_FALSE (fe->flags & VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS)) cryptodev_validate_mbuf_chain (vm, sop->m_src, b); else - /* for input nodes that are not dpdk-input, it is possible the mbuf - * was updated before as one of the chained mbufs. Setting nb_segs - * to 1 here to prevent the cryptodev PMD to access potentially - * invalid m_src->next pointers. - */ - sop->m_src->nb_segs = 1; + cryptodev_validate_mbuf (sop->m_src, b); + clib_memcpy_fast (cop[0]->iv, fe->iv, 16); + ring->frames[*enq].enq_elts_head++; cop++; bi++; fe++; n_elts--; } - n_enqueue = rte_cryptodev_enqueue_burst (cet->cryptodev_id, cet->cryptodev_q, - (struct rte_crypto_op **) cet->cops, - frame->n_elts); - ASSERT (n_enqueue == frame->n_elts); - cet->inflight += n_enqueue; + n_enqueue = + rte_cryptodev_enqueue_burst (cet->cryptodev_id, cet->cryptodev_q, + (struct rte_crypto_op **) cops, max_to_enq); + ERROR_ASSERT (n_enqueue == max_to_enq); + cet->inflight += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); + return; - return 0; +error_exit: + ring->frames[*enq].enq_elts_head += max_to_enq; + ring->frames[*enq].deq_elts_tail += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); + rte_mempool_put_bulk (cet->cop_pool, (void **) cops, max_to_enq); } static_always_inline int @@ -244,39 +292,64 @@ cryptodev_frame_aead_enqueue (vlib_main_t *vm, cryptodev_op_type_t op_type, u8 aad_len) { cryptodev_main_t *cmt = &cryptodev_main; - clib_pmalloc_main_t *pm = vm->physmem_main.pmalloc_main; cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + ERROR_ASSERT (frame != 0); + ERROR_ASSERT (frame->n_elts > 0); + cryptodev_cache_ring_elt_t *ring_elt = + cryptodev_cache_ring_push (ring, frame); + + if (PREDICT_FALSE (ring_elt == NULL)) + return -1; + + ring_elt->aad_len = aad_len; + ring_elt->op_type = (u8) op_type; + return 0; +} + +static_always_inline int +cryptodev_aead_enqueue_internal (vlib_main_t *vm, + vnet_crypto_async_frame_t *frame, + cryptodev_op_type_t op_type, u8 aad_len) +{ + cryptodev_main_t *cmt = &cryptodev_main; + cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + u16 *const enq = &ring->enq_head; + clib_pmalloc_main_t *pm = vm->physmem_main.pmalloc_main; vnet_crypto_async_frame_elt_t *fe; - struct rte_cryptodev_sym_session *sess = 0; - cryptodev_op_t **cop; - u32 *bi; + cryptodev_session_t *sess = 0; + cryptodev_op_t *cops[CRYPTODE_ENQ_MAX] = {}; + cryptodev_op_t **cop = cops; + u32 *bi = 0; u32 n_enqueue = 0, n_elts; u32 last_key_index = ~0; + u16 left_to_enq = frame->n_elts - ring->frames[*enq].enq_elts_head; + const u16 max_to_enq = clib_min (CRYPTODE_ENQ_MAX, left_to_enq); if (PREDICT_FALSE (frame == 0 || frame->n_elts == 0)) return -1; - n_elts = frame->n_elts; - if (PREDICT_FALSE (CRYPTODEV_MAX_INFLIGHT - cet->inflight < n_elts)) - { - cryptodev_mark_frame_err_status (frame, - VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); - return -1; - } + if (cet->inflight + max_to_enq > CRYPTODEV_MAX_INFLIGHT) + return -1; + + n_elts = max_to_enq; if (PREDICT_FALSE ( - rte_mempool_get_bulk (cet->cop_pool, (void **) cet->cops, n_elts) < 0)) + rte_mempool_get_bulk (cet->cop_pool, (void **) cops, n_elts) < 0)) { - cryptodev_mark_frame_err_status (frame, - VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + cryptodev_mark_frame_fill_err ( + frame, ring->frames[*enq].frame_elts_errs_mask, + ring->frames[*enq].enq_elts_head, max_to_enq, + VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + ring->frames[*enq].enq_elts_head += max_to_enq; + ring->frames[*enq].deq_elts_tail += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); return -1; } - cop = cet->cops; - fe = frame->elts; - bi = frame->buffer_indices; - cop[0]->frame = frame; - cop[0]->n_elts = n_elts; + fe = frame->elts + ring->frames[*enq].enq_elts_head; + bi = frame->buffer_indices + ring->frames[*enq].enq_elts_head; while (n_elts) { @@ -286,8 +359,8 @@ cryptodev_frame_aead_enqueue (vlib_main_t *vm, if (n_elts > 2) { - CLIB_PREFETCH (cop[1], CLIB_CACHE_LINE_BYTES * 3, STORE); - CLIB_PREFETCH (cop[2], CLIB_CACHE_LINE_BYTES * 3, STORE); + CLIB_PREFETCH (cop[1], sizeof (*cop[1]), STORE); + CLIB_PREFETCH (cop[2], sizeof (*cop[2]), STORE); clib_prefetch_load (&fe[1]); clib_prefetch_load (&fe[2]); } @@ -301,23 +374,32 @@ cryptodev_frame_aead_enqueue (vlib_main_t *vm, if (PREDICT_FALSE (cryptodev_session_create (vm, last_key_index, aad_len) < 0)) { - cryptodev_mark_frame_err_status ( - frame, VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); - return -1; + cryptodev_mark_frame_fill_err ( + frame, ring->frames[*enq].frame_elts_errs_mask, + ring->frames[*enq].enq_elts_head, max_to_enq, + VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + goto error_exit; } } else if (PREDICT_FALSE ( - key->keys[vm->numa_node][op_type]->opaque_data != - aad_len)) +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + rte_cryptodev_sym_session_opaque_data_get ( + key->keys[vm->numa_node][op_type]) != (u64) aad_len +#else + key->keys[vm->numa_node][op_type]->opaque_data != aad_len +#endif + )) { cryptodev_sess_handler (vm, VNET_CRYPTO_KEY_OP_DEL, fe->key_index, aad_len); if (PREDICT_FALSE (cryptodev_session_create (vm, last_key_index, aad_len) < 0)) { - cryptodev_mark_frame_err_status ( - frame, VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); - return -1; + cryptodev_mark_frame_fill_err ( + frame, ring->frames[*enq].frame_elts_errs_mask, + ring->frames[*enq].enq_elts_head, max_to_enq, + VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + goto error_exit; } } @@ -348,117 +430,179 @@ cryptodev_frame_aead_enqueue (vlib_main_t *vm, if (PREDICT_FALSE (fe->flags & VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS)) cryptodev_validate_mbuf_chain (vm, sop->m_src, b); else - /* for input nodes that are not dpdk-input, it is possible the mbuf - * was updated before as one of the chained mbufs. Setting nb_segs - * to 1 here to prevent the cryptodev PMD to access potentially - * invalid m_src->next pointers. - */ - sop->m_src->nb_segs = 1; + cryptodev_validate_mbuf (sop->m_src, b); + clib_memcpy_fast (cop[0]->iv, fe->iv, 12); clib_memcpy_fast (cop[0]->aad, fe->aad, aad_len); + cop++; bi++; fe++; n_elts--; } - n_enqueue = rte_cryptodev_enqueue_burst (cet->cryptodev_id, cet->cryptodev_q, - (struct rte_crypto_op **) cet->cops, - frame->n_elts); - ASSERT (n_enqueue == frame->n_elts); - cet->inflight += n_enqueue; + n_enqueue = + rte_cryptodev_enqueue_burst (cet->cryptodev_id, cet->cryptodev_q, + (struct rte_crypto_op **) cops, max_to_enq); + ERROR_ASSERT (n_enqueue == max_to_enq); + cet->inflight += max_to_enq; + ring->frames[*enq].enq_elts_head += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); return 0; -} - -static_always_inline u16 -cryptodev_ring_deq (struct rte_ring *r, cryptodev_op_t **cops) -{ - u16 n, n_elts = 0; - n = rte_ring_dequeue_bulk_start (r, (void **) cops, 1, 0); - rte_ring_dequeue_finish (r, 0); - if (!n) - return 0; - - n = cops[0]->n_elts; - if (rte_ring_count (r) < n) - return 0; - - n_elts = rte_ring_sc_dequeue_bulk (r, (void **) cops, n, 0); - ASSERT (n_elts == n); +error_exit: + ring->frames[*enq].enq_elts_head += max_to_enq; + ring->frames[*enq].deq_elts_tail += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); + rte_mempool_put_bulk (cet->cop_pool, (void **) cops, max_to_enq); - return n_elts; + return -1; } -static_always_inline vnet_crypto_async_frame_t * -cryptodev_frame_dequeue (vlib_main_t *vm, u32 *nb_elts_processed, - u32 *enqueue_thread_idx) +static_always_inline u8 +cryptodev_frame_dequeue_internal (vlib_main_t *vm, u32 *enqueue_thread_idx) { cryptodev_main_t *cmt = &cryptodev_main; cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; - cryptodev_op_t **cop = cet->cops; + vnet_crypto_async_frame_t *frame = NULL; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + u16 *const deq = &ring->deq_tail; + u16 n_deq, left_to_deq; + u16 max_to_deq = 0; + u16 inflight = cet->inflight; + u8 dequeue_more = 0; + cryptodev_op_t *cops[CRYPTODE_DEQ_MAX] = {}; + cryptodev_op_t **cop = cops; vnet_crypto_async_frame_elt_t *fe; - vnet_crypto_async_frame_t *frame; - u32 n_elts, n_completed_ops = rte_ring_count (cet->ring); - u32 ss0 = 0, ss1 = 0, ss2 = 0, ss3 = 0; /* sum of status */ - - if (cet->inflight) - { - n_elts = rte_cryptodev_dequeue_burst ( - cet->cryptodev_id, cet->cryptodev_q, - (struct rte_crypto_op **) cet->cops, VNET_CRYPTO_FRAME_SIZE); + u32 n_elts, n; + u64 err0 = 0, err1 = 0, err2 = 0, err3 = 0; /* partial errors mask */ - if (n_elts) - { - cet->inflight -= n_elts; - n_completed_ops += n_elts; + left_to_deq = + ring->frames[*deq].f->n_elts - ring->frames[*deq].deq_elts_tail; + max_to_deq = clib_min (left_to_deq, CRYPTODE_DEQ_MAX); - rte_ring_sp_enqueue_burst (cet->ring, (void **) cet->cops, n_elts, - NULL); - } - } + /* deq field can be used to track frame that is currently dequeued + based on that you can specify the amount of elements to deq for the frame */ + n_deq = + rte_cryptodev_dequeue_burst (cet->cryptodev_id, cet->cryptodev_q, + (struct rte_crypto_op **) cops, max_to_deq); - if (PREDICT_FALSE (n_completed_ops == 0)) - return 0; + if (n_deq == 0) + return dequeue_more; - n_elts = cryptodev_ring_deq (cet->ring, cop); - if (!n_elts) - return 0; + frame = ring->frames[*deq].f; + fe = frame->elts + ring->frames[*deq].deq_elts_tail; - frame = cop[0]->frame; - fe = frame->elts; + n_elts = n_deq; + n = ring->frames[*deq].deq_elts_tail; while (n_elts > 4) { - ss0 |= fe[0].status = cryptodev_status_conversion[cop[0]->op.status]; - ss1 |= fe[1].status = cryptodev_status_conversion[cop[1]->op.status]; - ss2 |= fe[2].status = cryptodev_status_conversion[cop[2]->op.status]; - ss3 |= fe[3].status = cryptodev_status_conversion[cop[3]->op.status]; + fe[0].status = cryptodev_status_conversion[cop[0]->op.status]; + fe[1].status = cryptodev_status_conversion[cop[1]->op.status]; + fe[2].status = cryptodev_status_conversion[cop[2]->op.status]; + fe[3].status = cryptodev_status_conversion[cop[3]->op.status]; + + err0 |= ((u64) (fe[0].status == VNET_CRYPTO_OP_STATUS_COMPLETED)) << n; + err1 |= ((u64) (fe[1].status == VNET_CRYPTO_OP_STATUS_COMPLETED)) + << (n + 1); + err2 |= ((u64) (fe[2].status == VNET_CRYPTO_OP_STATUS_COMPLETED)) + << (n + 2); + err3 |= ((u64) (fe[3].status == VNET_CRYPTO_OP_STATUS_COMPLETED)) + << (n + 3); cop += 4; fe += 4; n_elts -= 4; + n += 4; } while (n_elts) { - ss0 |= fe[0].status = cryptodev_status_conversion[cop[0]->op.status]; + fe[0].status = cryptodev_status_conversion[cop[0]->op.status]; + err0 |= ((u64) (fe[0].status == VNET_CRYPTO_OP_STATUS_COMPLETED)) << n; + n++; fe++; cop++; n_elts--; } - frame->state = (ss0 | ss1 | ss2 | ss3) == VNET_CRYPTO_OP_STATUS_COMPLETED ? - VNET_CRYPTO_FRAME_STATE_SUCCESS : - VNET_CRYPTO_FRAME_STATE_ELT_ERROR; + ring->frames[*deq].frame_elts_errs_mask |= (err0 | err1 | err2 | err3); - rte_mempool_put_bulk (cet->cop_pool, (void **) cet->cops, frame->n_elts); - *nb_elts_processed = frame->n_elts; - *enqueue_thread_idx = frame->enqueue_thread_index; - return frame; + rte_mempool_put_bulk (cet->cop_pool, (void **) cops, n_deq); + + inflight -= n_deq; + ring->frames[*deq].deq_elts_tail += n_deq; + if (cryptodev_cache_ring_update_deq_tail (ring, deq)) + { + u32 fr_processed = + (CRYPTODEV_CACHE_QUEUE_SIZE - ring->tail + ring->deq_tail) & + CRYPTODEV_CACHE_QUEUE_MASK; + + *enqueue_thread_idx = frame->enqueue_thread_index; + dequeue_more = (fr_processed < CRYPTODEV_MAX_PROCESED_IN_CACHE_QUEUE); + } + + cet->inflight = inflight; + return dequeue_more; } +static_always_inline void +cryptodev_enqueue_frame (vlib_main_t *vm, cryptodev_cache_ring_elt_t *ring_elt) +{ + cryptodev_op_type_t op_type = (cryptodev_op_type_t) ring_elt->op_type; + u8 linked_or_aad_len = ring_elt->aad_len; + + if (linked_or_aad_len == 1) + cryptodev_frame_linked_algs_enqueue_internal (vm, ring_elt->f, op_type); + else + cryptodev_aead_enqueue_internal (vm, ring_elt->f, op_type, + linked_or_aad_len); +} + +static_always_inline vnet_crypto_async_frame_t * +cryptodev_frame_dequeue (vlib_main_t *vm, u32 *nb_elts_processed, + u32 *enqueue_thread_idx) +{ + cryptodev_main_t *cmt = &cryptodev_main; + vnet_crypto_main_t *cm = &crypto_main; + cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + cryptodev_cache_ring_elt_t *ring_elt = &ring->frames[ring->tail]; + + vnet_crypto_async_frame_t *ret_frame = 0; + u8 dequeue_more = 1; + + while (cet->inflight > 0 && dequeue_more) + { + dequeue_more = cryptodev_frame_dequeue_internal (vm, enqueue_thread_idx); + } + + if (PREDICT_TRUE (ring->frames[ring->enq_head].f != 0)) + cryptodev_enqueue_frame (vm, &ring->frames[ring->enq_head]); + + if (PREDICT_TRUE (ring_elt->f != 0)) + { + if (ring_elt->n_elts == ring_elt->deq_elts_tail) + { + *nb_elts_processed = ring_elt->n_elts; + vlib_node_set_interrupt_pending ( + vlib_get_main_by_index (vm->thread_index), cm->crypto_node_index); + ret_frame = cryptodev_cache_ring_pop (ring); + return ret_frame; + } + } + + return ret_frame; +} +static_always_inline int +cryptodev_enqueue_aead_aad_0_enc (vlib_main_t *vm, + vnet_crypto_async_frame_t *frame) +{ + return cryptodev_frame_aead_enqueue (vm, frame, CRYPTODEV_OP_TYPE_ENCRYPT, + 0); +} static_always_inline int cryptodev_enqueue_aead_aad_8_enc (vlib_main_t *vm, vnet_crypto_async_frame_t *frame) @@ -475,6 +619,13 @@ cryptodev_enqueue_aead_aad_12_enc (vlib_main_t *vm, } static_always_inline int +cryptodev_enqueue_aead_aad_0_dec (vlib_main_t *vm, + vnet_crypto_async_frame_t *frame) +{ + return cryptodev_frame_aead_enqueue (vm, frame, CRYPTODEV_OP_TYPE_DECRYPT, + 0); +} +static_always_inline int cryptodev_enqueue_aead_aad_8_dec (vlib_main_t *vm, vnet_crypto_async_frame_t *frame) { @@ -515,6 +666,7 @@ cryptodev_register_cop_hdl (vlib_main_t *vm, u32 eidx) struct rte_cryptodev_sym_capability_idx cap_aead_idx; u8 *name; clib_error_t *error = 0; + u32 ref_cnt = 0; vec_foreach (cet, cmt->per_thread_data) { @@ -525,43 +677,28 @@ cryptodev_register_cop_hdl (vlib_main_t *vm, u32 eidx) (char *) name, CRYPTODEV_NB_CRYPTO_OPS, sizeof (cryptodev_op_t), 0, sizeof (struct rte_crypto_op_pool_private), NULL, NULL, crypto_op_init, NULL, vm->numa_node, 0); - if (!cet->cop_pool) - { - error = clib_error_return ( - 0, "Failed to create cryptodev op pool %s", name); - - goto error_exit; - } vec_free (name); - - name = format (0, "frames_ring_%u_%u", numa, thread_index); - cet->ring = - rte_ring_create ((char *) name, CRYPTODEV_NB_CRYPTO_OPS, vm->numa_node, - RING_F_SP_ENQ | RING_F_SC_DEQ); - if (!cet->ring) + if (!cet->cop_pool) { error = clib_error_return ( 0, "Failed to create cryptodev op pool %s", name); goto error_exit; } - vec_free (name); - - vec_validate (cet->cops, VNET_CRYPTO_FRAME_SIZE - 1); } - /** INDENT-OFF **/ #define _(a, b, c, d, e, f, g) \ cap_aead_idx.type = RTE_CRYPTO_SYM_XFORM_AEAD; \ cap_aead_idx.algo.aead = RTE_CRYPTO_##b##_##c; \ if (cryptodev_check_cap_support (&cap_aead_idx, g, e, f)) \ { \ - vnet_crypto_register_async_handler ( \ + vnet_crypto_register_enqueue_handler ( \ vm, eidx, VNET_CRYPTO_OP_##a##_TAG##e##_AAD##f##_ENC, \ - cryptodev_enqueue_aead_aad_##f##_enc, cryptodev_frame_dequeue); \ - vnet_crypto_register_async_handler ( \ + cryptodev_enqueue_aead_aad_##f##_enc); \ + vnet_crypto_register_enqueue_handler ( \ vm, eidx, VNET_CRYPTO_OP_##a##_TAG##e##_AAD##f##_DEC, \ - cryptodev_enqueue_aead_aad_##f##_dec, cryptodev_frame_dequeue); \ + cryptodev_enqueue_aead_aad_##f##_dec); \ + ref_cnt++; \ } foreach_vnet_aead_crypto_conversion #undef _ @@ -574,25 +711,25 @@ cryptodev_register_cop_hdl (vlib_main_t *vm, u32 eidx) if (cryptodev_check_cap_support (&cap_cipher_idx, c, -1, -1) && \ cryptodev_check_cap_support (&cap_auth_idx, -1, e, -1)) \ { \ - vnet_crypto_register_async_handler ( \ + vnet_crypto_register_enqueue_handler ( \ vm, eidx, VNET_CRYPTO_OP_##a##_##d##_TAG##e##_ENC, \ - cryptodev_enqueue_linked_alg_enc, cryptodev_frame_dequeue); \ - vnet_crypto_register_async_handler ( \ + cryptodev_enqueue_linked_alg_enc); \ + vnet_crypto_register_enqueue_handler ( \ vm, eidx, VNET_CRYPTO_OP_##a##_##d##_TAG##e##_DEC, \ - cryptodev_enqueue_linked_alg_dec, cryptodev_frame_dequeue); \ + cryptodev_enqueue_linked_alg_dec); \ + ref_cnt++; \ } foreach_cryptodev_link_async_alg #undef _ - /** INDENT-ON **/ + + if (ref_cnt) + vnet_crypto_register_dequeue_handler (vm, eidx, cryptodev_frame_dequeue); return 0; error_exit: vec_foreach (cet, cmt->per_thread_data) { - if (cet->ring) - rte_ring_free (cet->ring); - if (cet->cop_pool) rte_mempool_free (cet->cop_pool); } diff --git a/src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c b/src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c index 41a1e0c2a09..67ab9c89e67 100644 --- a/src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c +++ b/src/plugins/dpdk/cryptodev/cryptodev_raw_data_path.c @@ -29,7 +29,7 @@ #include <rte_cryptodev.h> #include <rte_crypto_sym.h> #include <rte_crypto.h> -#include <rte_cryptodev_pmd.h> +#include <rte_malloc.h> #include <rte_config.h> #include "cryptodev.h" @@ -96,7 +96,7 @@ cryptodev_reset_ctx (cryptodev_engine_thread_t *cet) { union rte_cryptodev_session_ctx sess_ctx; - ASSERT (cet->reset_sess != 0); + ERROR_ASSERT (cet->reset_sess != 0); sess_ctx.crypto_sess = cet->reset_sess; @@ -112,30 +112,51 @@ cryptodev_frame_linked_algs_enqueue (vlib_main_t *vm, { cryptodev_main_t *cmt = &cryptodev_main; cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + ERROR_ASSERT (frame != 0); + ERROR_ASSERT (frame->n_elts > 0); + cryptodev_cache_ring_elt_t *ring_elt = + cryptodev_cache_ring_push (ring, frame); + + if (PREDICT_FALSE (ring_elt == NULL)) + return -1; + + ring_elt->aad_len = 1; + ring_elt->op_type = (u8) op_type; + return 0; +} + +static_always_inline void +cryptodev_frame_linked_algs_enqueue_internal (vlib_main_t *vm, + vnet_crypto_async_frame_t *frame, + cryptodev_op_type_t op_type) +{ + cryptodev_main_t *cmt = &cryptodev_main; + cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; vnet_crypto_async_frame_elt_t *fe; vlib_buffer_t **b; struct rte_crypto_vec vec[CRYPTODEV_MAX_N_SGL]; struct rte_crypto_va_iova_ptr iv_vec, digest_vec; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + u16 *const enq = &ring->enq_head; u32 n_elts; u32 last_key_index = ~0; i16 min_ofs; u32 max_end; + u32 max_to_enq = clib_min (CRYPTODE_ENQ_MAX, + frame->n_elts - ring->frames[*enq].enq_elts_head); u8 is_update = 0; int status; - n_elts = frame->n_elts; + if (cet->inflight + max_to_enq > CRYPTODEV_MAX_INFLIGHT) + return; - if (PREDICT_FALSE (CRYPTODEV_MAX_INFLIGHT - cet->inflight < n_elts)) - { - cryptodev_mark_frame_err_status (frame, - VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); - return -1; - } + n_elts = max_to_enq; vlib_get_buffers (vm, frame->buffer_indices, cet->b, frame->n_elts); - b = cet->b; - fe = frame->elts; + b = cet->b + ring->frames[*enq].enq_elts_head; + fe = frame->elts + ring->frames[*enq].enq_elts_head; while (n_elts) { @@ -215,26 +236,31 @@ cryptodev_frame_linked_algs_enqueue (vlib_main_t *vm, if (PREDICT_FALSE (status < 0)) goto error_exit; + ring->frames[*enq].enq_elts_head += 1; b++; fe++; n_elts--; } - status = rte_cryptodev_raw_enqueue_done (cet->ctx, frame->n_elts); + status = rte_cryptodev_raw_enqueue_done (cet->ctx, max_to_enq); if (PREDICT_FALSE (status < 0)) - { - cryptodev_reset_ctx (cet); - return -1; - } + goto error_exit; - cet->inflight += frame->n_elts; - return 0; + cet->inflight += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); + return; error_exit: - cryptodev_mark_frame_err_status (frame, - VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + cryptodev_mark_frame_fill_err (frame, + ring->frames[*enq].frame_elts_errs_mask, + ring->frames[*enq].enq_elts_head, max_to_enq, + VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + ring->frames[*enq].enq_elts_head += max_to_enq; + ring->frames[*enq].deq_elts_tail += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); cryptodev_reset_ctx (cet); - return -1; + + return; } static_always_inline int @@ -243,6 +269,28 @@ cryptodev_raw_aead_enqueue (vlib_main_t *vm, vnet_crypto_async_frame_t *frame, { cryptodev_main_t *cmt = &cryptodev_main; cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + ERROR_ASSERT (frame != 0); + ERROR_ASSERT (frame->n_elts > 0); + cryptodev_cache_ring_elt_t *ring_elt = + cryptodev_cache_ring_push (ring, frame); + + if (PREDICT_FALSE (ring_elt == NULL)) + return -1; + + ring_elt->aad_len = aad_len; + ring_elt->op_type = (u8) op_type; + return 0; +} + +static_always_inline void +cryptodev_raw_aead_enqueue_internal (vlib_main_t *vm, + vnet_crypto_async_frame_t *frame, + cryptodev_op_type_t op_type, u8 aad_len) +{ + cryptodev_main_t *cmt = &cryptodev_main; + cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; vnet_crypto_async_frame_elt_t *fe; vlib_buffer_t **b; u32 n_elts; @@ -250,22 +298,23 @@ cryptodev_raw_aead_enqueue (vlib_main_t *vm, vnet_crypto_async_frame_t *frame, struct rte_crypto_vec vec[CRYPTODEV_MAX_N_SGL]; struct rte_crypto_va_iova_ptr iv_vec, digest_vec, aad_vec; u32 last_key_index = ~0; + u16 *const enq = &ring->enq_head; + u16 left_to_enq = frame->n_elts - ring->frames[*enq].enq_elts_head; + u16 max_to_enq = clib_min (CRYPTODE_ENQ_MAX, left_to_enq); u8 is_update = 0; int status; - n_elts = frame->n_elts; - - if (PREDICT_FALSE (CRYPTODEV_MAX_INFLIGHT - cet->inflight < n_elts)) + if (cet->inflight + max_to_enq > CRYPTODEV_MAX_INFLIGHT) { - cryptodev_mark_frame_err_status (frame, - VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); - return -1; + return; } + n_elts = max_to_enq; + vlib_get_buffers (vm, frame->buffer_indices, cet->b, frame->n_elts); - fe = frame->elts; - b = cet->b; + fe = frame->elts + ring->frames[*enq].enq_elts_head; + b = cet->b + ring->frames[*enq].enq_elts_head; cofs.raw = 0; while (n_elts) @@ -292,8 +341,13 @@ cryptodev_raw_aead_enqueue (vlib_main_t *vm, vnet_crypto_async_frame_t *frame, } if (PREDICT_FALSE ( - (u8) key->keys[vm->numa_node][op_type]->opaque_data != - aad_len)) +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + rte_cryptodev_sym_session_opaque_data_get ( + key->keys[vm->numa_node][op_type]) != (u64) aad_len +#else + (u8) key->keys[vm->numa_node][op_type]->opaque_data != aad_len +#endif + )) { cryptodev_sess_handler (vm, VNET_CRYPTO_KEY_OP_DEL, fe->key_index, aad_len); @@ -349,7 +403,7 @@ cryptodev_raw_aead_enqueue (vlib_main_t *vm, vnet_crypto_async_frame_t *frame, if (aad_len == 8) *(u64 *) (cet->aad_buf + aad_offset) = *(u64 *) fe->aad; - else + else if (aad_len != 0) { /* aad_len == 12 */ *(u64 *) (cet->aad_buf + aad_offset) = *(u64 *) fe->aad; @@ -373,31 +427,30 @@ cryptodev_raw_aead_enqueue (vlib_main_t *vm, vnet_crypto_async_frame_t *frame, if (PREDICT_FALSE (status < 0)) goto error_exit; + ring->frames[*enq].enq_elts_head += 1; fe++; b++; n_elts--; } - status = rte_cryptodev_raw_enqueue_done (cet->ctx, frame->n_elts); + status = rte_cryptodev_raw_enqueue_done (cet->ctx, max_to_enq); if (PREDICT_FALSE (status < 0)) goto error_exit; - cet->inflight += frame->n_elts; - - return 0; + cet->inflight += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); + return; error_exit: - cryptodev_mark_frame_err_status (frame, - VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + cryptodev_mark_frame_fill_err (frame, + ring->frames[*enq].frame_elts_errs_mask, + ring->frames[*enq].enq_elts_head, max_to_enq, + VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR); + ring->frames[*enq].enq_elts_head += max_to_enq; + ring->frames[*enq].deq_elts_tail += max_to_enq; + cryptodev_cache_ring_update_enq_head (ring, frame); cryptodev_reset_ctx (cet); - return -1; -} - -static_always_inline u32 -cryptodev_get_frame_n_elts (void *frame) -{ - vnet_crypto_async_frame_t *f = (vnet_crypto_async_frame_t *) frame; - return f->n_elts; + return; } static_always_inline void @@ -409,180 +462,114 @@ cryptodev_post_dequeue (void *frame, u32 index, u8 is_op_success) VNET_CRYPTO_OP_STATUS_FAIL_BAD_HMAC; } -#define GET_RING_OBJ(r, pos, f) \ - do \ - { \ - vnet_crypto_async_frame_t **ring = (void *) &r[1]; \ - f = ring[(r->cons.head + pos) & r->mask]; \ - } \ - while (0) - -static_always_inline vnet_crypto_async_frame_t * -cryptodev_raw_dequeue (vlib_main_t *vm, u32 *nb_elts_processed, - u32 *enqueue_thread_idx) +static_always_inline u8 +cryptodev_raw_dequeue_internal (vlib_main_t *vm, u32 *enqueue_thread_idx) { cryptodev_main_t *cmt = &cryptodev_main; cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; - vnet_crypto_async_frame_t *frame, *frame_ret = 0; - u32 n_deq, n_success; - u32 n_cached_frame = rte_ring_count (cet->cached_frame), n_room_left; - u8 no_job_to_deq = 0; + vnet_crypto_async_frame_t *frame; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + u16 *const deq = &ring->deq_tail; + u32 n_success; + u16 n_deq, i, left_to_deq; + u16 max_to_deq = 0; u16 inflight = cet->inflight; + u8 dequeue_more = 0; int dequeue_status; - n_room_left = CRYPTODEV_DEQ_CACHE_SZ - n_cached_frame - 1; + left_to_deq = ring->frames[*deq].n_elts - ring->frames[*deq].deq_elts_tail; + max_to_deq = clib_min (left_to_deq, CRYPTODE_DEQ_MAX); - if (n_cached_frame) - { - u32 i; - for (i = 0; i < n_cached_frame; i++) - { - vnet_crypto_async_frame_t *f; - void *f_ret; - enum rte_crypto_op_status op_status; - u8 n_left, err, j; + /* deq field can be used to track frame that is currently dequeued */ + /* based on thatthe amount of elements to deq for the frame can be specified + */ - GET_RING_OBJ (cet->cached_frame, i, f); - - if (i < n_cached_frame - 2) - { - vnet_crypto_async_frame_t *f1, *f2; - GET_RING_OBJ (cet->cached_frame, i + 1, f1); - GET_RING_OBJ (cet->cached_frame, i + 2, f2); - clib_prefetch_load (f1); - clib_prefetch_load (f2); - } - - n_left = f->state & 0x7f; - err = f->state & 0x80; - - for (j = f->n_elts - n_left; j < f->n_elts && inflight; j++) - { - int ret; - f_ret = rte_cryptodev_raw_dequeue (cet->ctx, &ret, &op_status); - - if (!f_ret) - break; - - switch (op_status) - { - case RTE_CRYPTO_OP_STATUS_SUCCESS: - f->elts[j].status = VNET_CRYPTO_OP_STATUS_COMPLETED; - break; - default: - f->elts[j].status = VNET_CRYPTO_OP_STATUS_FAIL_ENGINE_ERR; - err |= 1 << 7; - } - - inflight--; - } + n_deq = rte_cryptodev_raw_dequeue_burst ( + cet->ctx, NULL, max_to_deq, cryptodev_post_dequeue, (void **) &frame, 0, + &n_success, &dequeue_status); - if (j == f->n_elts) - { - if (i == 0) - { - frame_ret = f; - f->state = err ? VNET_CRYPTO_FRAME_STATE_ELT_ERROR : - VNET_CRYPTO_FRAME_STATE_SUCCESS; - } - else - { - f->state = f->n_elts - j; - f->state |= err; - } - if (inflight) - continue; - } + if (n_deq == 0) + return dequeue_more; - /* to here f is not completed dequeued and no more job can be - * dequeued - */ - f->state = f->n_elts - j; - f->state |= err; - no_job_to_deq = 1; - break; - } + inflight -= n_deq; + if (PREDICT_FALSE (n_success < n_deq)) + { + u16 idx = ring->frames[*deq].deq_elts_tail; - if (frame_ret) + for (i = 0; i < n_deq; i++) { - rte_ring_sc_dequeue (cet->cached_frame, (void **) &frame_ret); - n_room_left++; + if (frame->elts[idx + i].status != VNET_CRYPTO_OP_STATUS_COMPLETED) + ring->frames[*deq].frame_elts_errs_mask |= 1 << (idx + i); } } + ring->frames[*deq].deq_elts_tail += n_deq; - /* no point to dequeue further */ - if (!inflight || no_job_to_deq || !n_room_left) - goto end_deq; + if (cryptodev_cache_ring_update_deq_tail (ring, deq)) + { + u32 fr_processed = + (CRYPTODEV_CACHE_QUEUE_SIZE - ring->tail + ring->deq_tail) & + CRYPTODEV_CACHE_QUEUE_MASK; -#if RTE_VERSION >= RTE_VERSION_NUM(21, 5, 0, 0) - n_deq = rte_cryptodev_raw_dequeue_burst ( - cet->ctx, cryptodev_get_frame_n_elts, 0, cryptodev_post_dequeue, - (void **) &frame, 0, &n_success, &dequeue_status); -#else - n_deq = rte_cryptodev_raw_dequeue_burst ( - cet->ctx, cryptodev_get_frame_n_elts, cryptodev_post_dequeue, - (void **) &frame, 0, &n_success, &dequeue_status); -#endif + *enqueue_thread_idx = frame->enqueue_thread_index; + dequeue_more = (fr_processed < CRYPTODEV_MAX_PROCESED_IN_CACHE_QUEUE); + } - if (!n_deq) - goto end_deq; + int res = + rte_cryptodev_raw_dequeue_done (cet->ctx, cet->inflight - inflight); + ERROR_ASSERT (res == 0); + cet->inflight = inflight; + return dequeue_more; +} - inflight -= n_deq; - no_job_to_deq = n_deq < frame->n_elts; - /* we have to cache the frame */ - if (frame_ret || n_cached_frame || no_job_to_deq) - { - frame->state = frame->n_elts - n_deq; - frame->state |= ((n_success < n_deq) << 7); - rte_ring_sp_enqueue (cet->cached_frame, (void *) frame); - n_room_left--; - } +static_always_inline void +cryptodev_enqueue_frame_to_qat (vlib_main_t *vm, + cryptodev_cache_ring_elt_t *ring_elt) +{ + cryptodev_op_type_t op_type = (cryptodev_op_type_t) ring_elt->op_type; + u8 linked_or_aad_len = ring_elt->aad_len; + + if (linked_or_aad_len == 1) + cryptodev_frame_linked_algs_enqueue_internal (vm, ring_elt->f, op_type); else - { - frame->state = n_success == frame->n_elts ? - VNET_CRYPTO_FRAME_STATE_SUCCESS : - VNET_CRYPTO_FRAME_STATE_ELT_ERROR; - frame_ret = frame; - } + cryptodev_raw_aead_enqueue_internal (vm, ring_elt->f, op_type, + linked_or_aad_len); +} - /* see if we can dequeue more */ - while (inflight && n_room_left && !no_job_to_deq) - { -#if RTE_VERSION >= RTE_VERSION_NUM(21, 5, 0, 0) - n_deq = rte_cryptodev_raw_dequeue_burst ( - cet->ctx, cryptodev_get_frame_n_elts, 0, cryptodev_post_dequeue, - (void **) &frame, 0, &n_success, &dequeue_status); -#else - n_deq = rte_cryptodev_raw_dequeue_burst ( - cet->ctx, cryptodev_get_frame_n_elts, cryptodev_post_dequeue, - (void **) &frame, 0, &n_success, &dequeue_status); -#endif - if (!n_deq) - break; - inflight -= n_deq; - no_job_to_deq = n_deq < frame->n_elts; - frame->state = frame->n_elts - n_deq; - frame->state |= ((n_success < n_deq) << 7); - rte_ring_sp_enqueue (cet->cached_frame, (void *) frame); - n_room_left--; - } +static_always_inline vnet_crypto_async_frame_t * +cryptodev_raw_dequeue (vlib_main_t *vm, u32 *nb_elts_processed, + u32 *enqueue_thread_idx) +{ + cryptodev_main_t *cmt = &cryptodev_main; + vnet_crypto_main_t *cm = &crypto_main; + cryptodev_engine_thread_t *cet = cmt->per_thread_data + vm->thread_index; + cryptodev_cache_ring_t *ring = &cet->cache_ring; + cryptodev_cache_ring_elt_t *ring_elt = &ring->frames[ring->tail]; + vnet_crypto_async_frame_t *ret_frame = 0; + u8 dequeue_more = 1; -end_deq: - if (inflight < cet->inflight) - { - int res = - rte_cryptodev_raw_dequeue_done (cet->ctx, cet->inflight - inflight); - ASSERT (res == 0); - cet->inflight = inflight; - } + while (cet->inflight > 0 && dequeue_more) + dequeue_more = cryptodev_raw_dequeue_internal (vm, enqueue_thread_idx); + + if (PREDICT_TRUE (ring->frames[ring->enq_head].f != 0)) + cryptodev_enqueue_frame_to_qat (vm, &ring->frames[ring->enq_head]); - if (frame_ret) + if (PREDICT_TRUE (ring_elt->f != 0) && + (ring_elt->n_elts == ring_elt->deq_elts_tail)) { - *nb_elts_processed = frame_ret->n_elts; - *enqueue_thread_idx = frame_ret->enqueue_thread_index; + *nb_elts_processed = ring_elt->n_elts; + vlib_node_set_interrupt_pending ( + vlib_get_main_by_index (vm->thread_index), cm->crypto_node_index); + ret_frame = cryptodev_cache_ring_pop (ring); } - return frame_ret; + return ret_frame; +} + +static_always_inline int +cryptodev_raw_enq_aead_aad_0_enc (vlib_main_t *vm, + vnet_crypto_async_frame_t *frame) +{ + return cryptodev_raw_aead_enqueue (vm, frame, CRYPTODEV_OP_TYPE_ENCRYPT, 0); } static_always_inline int @@ -599,6 +586,13 @@ cryptodev_raw_enq_aead_aad_12_enc (vlib_main_t *vm, } static_always_inline int +cryptodev_raw_enq_aead_aad_0_dec (vlib_main_t *vm, + vnet_crypto_async_frame_t *frame) +{ + return cryptodev_raw_aead_enqueue (vm, frame, CRYPTODEV_OP_TYPE_DECRYPT, 0); +} + +static_always_inline int cryptodev_raw_enq_aead_aad_8_dec (vlib_main_t *vm, vnet_crypto_async_frame_t *frame) { @@ -639,6 +633,7 @@ cryptodev_register_raw_hdl (vlib_main_t *vm, u32 eidx) struct rte_cryptodev_sym_capability_idx cap_aead_idx; u32 support_raw_api = 1, max_ctx_size = 0; clib_error_t *error = 0; + u8 ref_cnt = 0; vec_foreach (cinst, cmt->cryptodev_inst) { @@ -661,11 +656,7 @@ cryptodev_register_raw_hdl (vlib_main_t *vm, u32 eidx) { u32 thread_id = cet - cmt->per_thread_data; u32 numa = vlib_get_main_by_index (thread_id)->numa_node; - u8 *name = format (0, "cache_frame_ring_%u_%u", numa, thread_id); - - cet->cached_frame = - rte_ring_create ((char *) name, CRYPTODEV_DEQ_CACHE_SZ, numa, - RING_F_SC_DEQ | RING_F_SP_ENQ); + u8 *name = format (0, "cache_cache_ring_%u_%u", numa, thread_id); cet->aad_buf = rte_zmalloc_socket ( 0, CRYPTODEV_NB_CRYPTO_OPS * CRYPTODEV_MAX_AAD_SIZE, @@ -684,28 +675,21 @@ cryptodev_register_raw_hdl (vlib_main_t *vm, u32 eidx) error = clib_error_return (0, "Failed to alloc raw dp ctx"); goto err_handling; } - - if (cet->cached_frame == 0) - { - error = clib_error_return (0, "Failed to alloc frame ring %s", name); - goto err_handling; - } - vec_free (name); } -/** INDENT-OFF **/ #define _(a, b, c, d, e, f, g) \ cap_aead_idx.type = RTE_CRYPTO_SYM_XFORM_AEAD; \ cap_aead_idx.algo.aead = RTE_CRYPTO_##b##_##c; \ if (cryptodev_check_cap_support (&cap_aead_idx, g, e, f)) \ { \ - vnet_crypto_register_async_handler ( \ + vnet_crypto_register_enqueue_handler ( \ vm, eidx, VNET_CRYPTO_OP_##a##_TAG##e##_AAD##f##_ENC, \ - cryptodev_raw_enq_aead_aad_##f##_enc, cryptodev_raw_dequeue); \ - vnet_crypto_register_async_handler ( \ + cryptodev_raw_enq_aead_aad_##f##_enc); \ + vnet_crypto_register_enqueue_handler ( \ vm, eidx, VNET_CRYPTO_OP_##a##_TAG##e##_AAD##f##_DEC, \ - cryptodev_raw_enq_aead_aad_##f##_dec, cryptodev_raw_dequeue); \ + cryptodev_raw_enq_aead_aad_##f##_dec); \ + ref_cnt++; \ } foreach_vnet_aead_crypto_conversion #undef _ @@ -718,26 +702,24 @@ cryptodev_register_raw_hdl (vlib_main_t *vm, u32 eidx) if (cryptodev_check_cap_support (&cap_cipher_idx, c, -1, -1) && \ cryptodev_check_cap_support (&cap_auth_idx, -1, e, -1)) \ { \ - vnet_crypto_register_async_handler ( \ + vnet_crypto_register_enqueue_handler ( \ vm, eidx, VNET_CRYPTO_OP_##a##_##d##_TAG##e##_ENC, \ - cryptodev_raw_enq_linked_alg_enc, cryptodev_raw_dequeue); \ - vnet_crypto_register_async_handler ( \ + cryptodev_raw_enq_linked_alg_enc); \ + vnet_crypto_register_enqueue_handler ( \ vm, eidx, VNET_CRYPTO_OP_##a##_##d##_TAG##e##_DEC, \ - cryptodev_raw_enq_linked_alg_dec, cryptodev_raw_dequeue); \ + cryptodev_raw_enq_linked_alg_dec); \ + ref_cnt++; \ } foreach_cryptodev_link_async_alg #undef _ - cmt->is_raw_api = 1; + if (ref_cnt) + vnet_crypto_register_dequeue_handler (vm, eidx, cryptodev_raw_dequeue); + + cmt->is_raw_api = 1; return 0; err_handling: - vec_foreach (cet, cmt->per_thread_data) - { - if (cet->cached_frame) - rte_ring_free (cet->cached_frame); - } - return error; } diff --git a/src/plugins/dpdk/device/cli.c b/src/plugins/dpdk/device/cli.c index 0f771c6ba77..77f9a27f97b 100644 --- a/src/plugins/dpdk/device/cli.c +++ b/src/plugins/dpdk/device/cli.c @@ -77,26 +77,30 @@ show_dpdk_buffer (vlib_main_t * vm, unformat_input_t * input, * name="mbuf_pool_socket0" available = 15104 allocated = 1280 total = 16384 * @cliexend ?*/ -/* *INDENT-OFF* */ VLIB_CLI_COMMAND (cmd_show_dpdk_buffer,static) = { .path = "show dpdk buffer", .short_help = "show dpdk buffer", .function = show_dpdk_buffer, .is_mp_safe = 1, }; -/* *INDENT-ON* */ static clib_error_t * show_dpdk_physmem (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd) { clib_error_t *err = 0; - u32 pipe_max_size; int fds[2]; u8 *s = 0; int n, n_try; FILE *f; + /* + * XXX: Pipes on FreeBSD grow dynamically up to 64KB (FreeBSD 15), don't + * manually tweak this value on FreeBSD at the moment. + */ +#ifdef __linux__ + u32 pipe_max_size; + err = clib_sysfs_read ("/proc/sys/fs/pipe-max-size", "%u", &pipe_max_size); if (err) @@ -114,6 +118,7 @@ show_dpdk_physmem (vlib_main_t * vm, unformat_input_t * input, err = clib_error_return_unix (0, "fcntl(F_SETPIPE_SZ)"); goto error; } +#endif /* __linux__ */ if (fcntl (fds[0], F_SETFL, O_NONBLOCK) == -1) { @@ -142,7 +147,7 @@ show_dpdk_physmem (vlib_main_t * vm, unformat_input_t * input, err = clib_error_return_unix (0, "read"); goto error; } - _vec_len (s) = len + (n < 0 ? 0 : n); + vec_set_len (s, len + (n < 0 ? 0 : n)); } vlib_cli_output (vm, "%v", s); @@ -162,14 +167,12 @@ error: * @cliexstart{show dpdk physmem} * @cliexend ?*/ -/* *INDENT-OFF* */ VLIB_CLI_COMMAND (cmd_show_dpdk_physmem,static) = { .path = "show dpdk physmem", .short_help = "show dpdk physmem", .function = show_dpdk_physmem, .is_mp_safe = 1, }; -/* *INDENT-ON* */ static clib_error_t * test_dpdk_buffer (vlib_main_t * vm, unformat_input_t * input, @@ -198,7 +201,7 @@ test_dpdk_buffer (vlib_main_t * vm, unformat_input_t * input, first = vec_len (allocated_buffers) - n_free; vlib_buffer_free (vm, allocated_buffers + first, n_free); - _vec_len (allocated_buffers) = first; + vec_set_len (allocated_buffers, first); } if (n_alloc) { @@ -208,7 +211,7 @@ test_dpdk_buffer (vlib_main_t * vm, unformat_input_t * input, actual_alloc = vlib_buffer_alloc (vm, allocated_buffers + first, n_alloc); - _vec_len (allocated_buffers) = first + actual_alloc; + vec_set_len (allocated_buffers, first + actual_alloc); if (actual_alloc < n_alloc) vlib_cli_output (vm, "WARNING: only allocated %d buffers", @@ -250,14 +253,12 @@ test_dpdk_buffer (vlib_main_t * vm, unformat_input_t * input, * @cliexend * @endparblock ?*/ -/* *INDENT-OFF* */ VLIB_CLI_COMMAND (cmd_test_dpdk_buffer,static) = { .path = "test dpdk buffer", .short_help = "test dpdk buffer [allocate <nn>] [free <nn>]", .function = test_dpdk_buffer, .is_mp_safe = 1, }; -/* *INDENT-ON* */ static clib_error_t * set_dpdk_if_desc (vlib_main_t * vm, unformat_input_t * input, @@ -265,6 +266,7 @@ set_dpdk_if_desc (vlib_main_t * vm, unformat_input_t * input, { unformat_input_t _line_input, *line_input = &_line_input; dpdk_main_t *dm = &dpdk_main; + vnet_main_t *vnm = vnet_get_main (); vnet_hw_interface_t *hw; dpdk_device_t *xd; u32 hw_if_index = (u32) ~ 0; @@ -277,9 +279,8 @@ set_dpdk_if_desc (vlib_main_t * vm, unformat_input_t * input, while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT) { - if (unformat - (line_input, "%U", unformat_vnet_hw_interface, dm->vnet_main, - &hw_if_index)) + if (unformat (line_input, "%U", unformat_vnet_hw_interface, vnm, + &hw_if_index)) ; else if (unformat (line_input, "tx %d", &nb_tx_desc)) ; @@ -299,30 +300,21 @@ set_dpdk_if_desc (vlib_main_t * vm, unformat_input_t * input, goto done; } - hw = vnet_get_hw_interface (dm->vnet_main, hw_if_index); + hw = vnet_get_hw_interface (vnm, hw_if_index); xd = vec_elt_at_index (dm->devices, hw->dev_instance); - if ((xd->flags & DPDK_DEVICE_FLAG_PMD) == 0) - { - error = - clib_error_return (0, - "number of descriptors can be set only for " - "physical devices"); - goto done; - } - - if ((nb_rx_desc == (u32) ~ 0 || nb_rx_desc == xd->nb_rx_desc) && - (nb_tx_desc == (u32) ~ 0 || nb_tx_desc == xd->nb_tx_desc)) + if ((nb_rx_desc == (u32) ~0 || nb_rx_desc == xd->conf.n_rx_desc) && + (nb_tx_desc == (u32) ~0 || nb_tx_desc == xd->conf.n_tx_desc)) { error = clib_error_return (0, "nothing changed"); goto done; } if (nb_rx_desc != (u32) ~ 0) - xd->nb_rx_desc = nb_rx_desc; + xd->conf.n_rx_desc = nb_rx_desc; if (nb_tx_desc != (u32) ~ 0) - xd->nb_tx_desc = nb_tx_desc; + xd->conf.n_tx_desc = nb_tx_desc; dpdk_device_setup (xd); @@ -345,13 +337,11 @@ done: * Example of how to set the DPDK interface descriptors: * @cliexcmd{set dpdk interface descriptors GigabitEthernet0/8/0 rx 512 tx 512} ?*/ -/* *INDENT-OFF* */ VLIB_CLI_COMMAND (cmd_set_dpdk_if_desc,static) = { .path = "set dpdk interface descriptors", .short_help = "set dpdk interface descriptors <interface> [rx <nn>] [tx <nn>]", .function = set_dpdk_if_desc, }; -/* *INDENT-ON* */ static clib_error_t * show_dpdk_version_command_fn (vlib_main_t * vm, @@ -373,16 +363,15 @@ show_dpdk_version_command_fn (vlib_main_t * vm, * Example of how to display how many DPDK buffer test command has allocated: * @cliexstart{show dpdk version} * DPDK Version: DPDK 16.11.0 - * DPDK EAL init args: -c 1 -n 4 --huge-dir /run/vpp/hugepages --file-prefix vpp -w 0000:00:08.0 -w 0000:00:09.0 --master-lcore 0 --socket-mem 256 + * DPDK EAL init args: --in-memory --no-telemetry --file-prefix vpp + * -w 0000:00:08.0 -w 0000:00:09.0 * @cliexend ?*/ -/* *INDENT-OFF* */ VLIB_CLI_COMMAND (show_vpe_version_command, static) = { .path = "show dpdk version", .short_help = "show dpdk version", .function = show_dpdk_version_command_fn, }; -/* *INDENT-ON* */ /* Dummy function to get us linked in. */ void diff --git a/src/plugins/dpdk/device/common.c b/src/plugins/dpdk/device/common.c index 89046d1a8c0..7a49c5aaef2 100644 --- a/src/plugins/dpdk/device/common.c +++ b/src/plugins/dpdk/device/common.c @@ -29,11 +29,26 @@ #include <dpdk/device/dpdk_priv.h> #include <vppinfra/error.h> +/* DPDK TX offload to vnet hw interface caps mapppings */ +static struct +{ + u64 offload; + vnet_hw_if_caps_t caps; +} tx_off_caps_map[] = { + { RTE_ETH_TX_OFFLOAD_IPV4_CKSUM, VNET_HW_IF_CAP_TX_IP4_CKSUM }, + { RTE_ETH_TX_OFFLOAD_TCP_CKSUM, VNET_HW_IF_CAP_TX_TCP_CKSUM }, + { RTE_ETH_TX_OFFLOAD_UDP_CKSUM, VNET_HW_IF_CAP_TX_UDP_CKSUM }, + { RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM, VNET_HW_IF_CAP_TX_IP4_OUTER_CKSUM }, + { RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM, VNET_HW_IF_CAP_TX_UDP_OUTER_CKSUM }, + { RTE_ETH_TX_OFFLOAD_TCP_TSO, VNET_HW_IF_CAP_TCP_GSO }, + { RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO, VNET_HW_IF_CAP_VXLAN_TNL_GSO } +}; + void dpdk_device_error (dpdk_device_t * xd, char *str, int rv) { - dpdk_log_err ("Interface %U error %d: %s", - format_dpdk_device_name, xd->port_id, rv, rte_strerror (rv)); + dpdk_log_err ("Interface %U error %d: %s", format_dpdk_device_name, + xd->device_index, rv, rte_strerror (rv)); xd->errors = clib_error_return (xd->errors, "%s[port:%d, errno:%d]: %s", str, xd->port_id, rv, rte_strerror (rv)); } @@ -41,14 +56,16 @@ dpdk_device_error (dpdk_device_t * xd, char *str, int rv) void dpdk_device_setup (dpdk_device_t * xd) { - dpdk_main_t *dm = &dpdk_main; vlib_main_t *vm = vlib_get_main (); vnet_main_t *vnm = vnet_get_main (); - vlib_thread_main_t *tm = vlib_get_thread_main (); vnet_sw_interface_t *sw = vnet_get_sw_interface (vnm, xd->sw_if_index); vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, xd->hw_if_index); + u16 buf_sz = vlib_buffer_get_default_data_size (vm); + vnet_hw_if_caps_change_t caps = {}; struct rte_eth_dev_info dev_info; - u64 bitmap; + struct rte_eth_conf conf = {}; + u64 rxo, txo; + u32 max_frame_size; int rv; int j; @@ -59,70 +76,152 @@ dpdk_device_setup (dpdk_device_t * xd) if (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP) { - vnet_hw_interface_set_flags (dm->vnet_main, xd->hw_if_index, 0); + vnet_hw_interface_set_flags (vnm, xd->hw_if_index, 0); dpdk_device_stop (xd); } - /* Enable flow director when flows exist */ - if (xd->pmd == VNET_DPDK_PMD_I40E) + rte_eth_dev_info_get (xd->port_id, &dev_info); + + dpdk_log_debug ("[%u] configuring device %U", xd->port_id, + format_dpdk_rte_device, dev_info.device); + + /* create rx and tx offload wishlist */ + rxo = RTE_ETH_RX_OFFLOAD_IPV4_CKSUM; + txo = 0; + + if (xd->conf.enable_tcp_udp_checksum) + rxo |= RTE_ETH_RX_OFFLOAD_UDP_CKSUM | RTE_ETH_RX_OFFLOAD_TCP_CKSUM; + + if (xd->conf.disable_tx_checksum_offload == 0 && + xd->conf.enable_outer_checksum_offload) + txo |= + RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM | RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM; + + if (xd->conf.disable_tx_checksum_offload == 0) + txo |= RTE_ETH_TX_OFFLOAD_IPV4_CKSUM | RTE_ETH_TX_OFFLOAD_TCP_CKSUM | + RTE_ETH_TX_OFFLOAD_UDP_CKSUM; + + if (xd->conf.disable_multi_seg == 0) { - if ((xd->flags & DPDK_DEVICE_FLAG_RX_FLOW_OFFLOAD) != 0) - xd->port_conf.fdir_conf.mode = RTE_FDIR_MODE_PERFECT; - else - xd->port_conf.fdir_conf.mode = RTE_FDIR_MODE_NONE; + txo |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS; + rxo |= RTE_ETH_RX_OFFLOAD_SCATTER; +#if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0) + rxo |= DEV_RX_OFFLOAD_JUMBO_FRAME; +#endif } - rte_eth_dev_info_get (xd->port_id, &dev_info); - - bitmap = xd->port_conf.txmode.offloads & ~dev_info.tx_offload_capa; - if (bitmap) + if (xd->conf.enable_lro) + rxo |= RTE_ETH_RX_OFFLOAD_TCP_LRO; + + /* per-device offload config */ + if (xd->conf.enable_tso) + txo |= RTE_ETH_TX_OFFLOAD_TCP_CKSUM | RTE_ETH_TX_OFFLOAD_TCP_TSO | + RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO; + + if (xd->conf.disable_rx_scatter) + rxo &= ~RTE_ETH_RX_OFFLOAD_SCATTER; + + /* mask unsupported offloads */ + rxo &= dev_info.rx_offload_capa; + txo &= dev_info.tx_offload_capa; + + dpdk_log_debug ("[%u] Supported RX offloads: %U", xd->port_id, + format_dpdk_rx_offload_caps, dev_info.rx_offload_capa); + dpdk_log_debug ("[%u] Configured RX offloads: %U", xd->port_id, + format_dpdk_rx_offload_caps, rxo); + dpdk_log_debug ("[%u] Supported TX offloads: %U", xd->port_id, + format_dpdk_tx_offload_caps, dev_info.tx_offload_capa); + dpdk_log_debug ("[%u] Configured TX offloads: %U", xd->port_id, + format_dpdk_tx_offload_caps, txo); + + /* finalize configuration */ + conf.rxmode.offloads = rxo; + conf.txmode.offloads = txo; + if (rxo & RTE_ETH_RX_OFFLOAD_TCP_LRO) + conf.rxmode.max_lro_pkt_size = xd->conf.max_lro_pkt_size; + + if (xd->conf.enable_lsc_int) + conf.intr_conf.lsc = 1; + if (xd->conf.enable_rxq_int) + conf.intr_conf.rxq = 1; + + conf.rxmode.mq_mode = RTE_ETH_MQ_RX_NONE; + if (xd->conf.n_rx_queues > 1) { - dpdk_log_warn ("unsupported tx offloads requested on port %u: %U", - xd->port_id, format_dpdk_tx_offload_caps, bitmap); - xd->port_conf.txmode.offloads ^= bitmap; + if (xd->conf.disable_rss == 0) + { + conf.rxmode.mq_mode = RTE_ETH_MQ_RX_RSS; + conf.rx_adv_conf.rss_conf.rss_hf = xd->conf.rss_hf; + } } - bitmap = xd->port_conf.rxmode.offloads & ~dev_info.rx_offload_capa; - if (bitmap) +#if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0) + if (rxo & DEV_RX_OFFLOAD_JUMBO_FRAME) + { + conf.rxmode.max_rx_pkt_len = dev_info.max_rx_pktlen; + xd->max_supported_frame_size = dev_info.max_rx_pktlen; + } + else { - dpdk_log_warn ("unsupported rx offloads requested on port %u: %U", - xd->port_id, format_dpdk_rx_offload_caps, bitmap); - xd->port_conf.rxmode.offloads ^= bitmap; + xd->max_supported_frame_size = + clib_min (1500 + xd->driver_frame_overhead, buf_sz); } +#else + if (xd->conf.disable_multi_seg) + xd->max_supported_frame_size = clib_min (dev_info.max_rx_pktlen, buf_sz); + else + xd->max_supported_frame_size = dev_info.max_rx_pktlen; +#endif + + max_frame_size = clib_min (xd->max_supported_frame_size, + ethernet_main.default_mtu + hi->frame_overhead); - rv = rte_eth_dev_configure (xd->port_id, xd->rx_q_used, - xd->tx_q_used, &xd->port_conf); +#if RTE_VERSION >= RTE_VERSION_NUM(21, 11, 0, 0) + conf.rxmode.mtu = max_frame_size - xd->driver_frame_overhead; +#endif - if (rv < 0) +retry: + rv = rte_eth_dev_configure (xd->port_id, xd->conf.n_rx_queues, + xd->conf.n_tx_queues, &conf); + if (rv < 0 && conf.intr_conf.rxq) { - dpdk_device_error (xd, "rte_eth_dev_configure", rv); - goto error; + conf.intr_conf.rxq = 0; + goto retry; } - vec_validate_aligned (xd->tx_queues, xd->tx_q_used - 1, +#if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0) + rte_eth_dev_set_mtu (xd->port_id, + max_frame_size - xd->driver_frame_overhead); +#endif + + hi->max_frame_size = 0; + vnet_hw_interface_set_max_frame_size (vnm, xd->hw_if_index, max_frame_size); + dpdk_log_debug ("[%u] max_frame_size %u max max_frame_size %u " + "driver_frame_overhead %u", + xd->port_id, hi->max_frame_size, + xd->max_supported_frame_size, xd->driver_frame_overhead); + + vec_validate_aligned (xd->tx_queues, xd->conf.n_tx_queues - 1, CLIB_CACHE_LINE_BYTES); - for (j = 0; j < xd->tx_q_used; j++) + for (j = 0; j < xd->conf.n_tx_queues; j++) { - rv = - rte_eth_tx_queue_setup (xd->port_id, j, xd->nb_tx_desc, - xd->cpu_socket, &xd->tx_conf); + rv = rte_eth_tx_queue_setup (xd->port_id, j, xd->conf.n_tx_desc, + xd->cpu_socket, 0); /* retry with any other CPU socket */ if (rv < 0) - rv = - rte_eth_tx_queue_setup (xd->port_id, j, - xd->nb_tx_desc, SOCKET_ID_ANY, - &xd->tx_conf); + rv = rte_eth_tx_queue_setup (xd->port_id, j, xd->conf.n_tx_desc, + SOCKET_ID_ANY, 0); if (rv < 0) dpdk_device_error (xd, "rte_eth_tx_queue_setup", rv); - if (xd->tx_q_used < tm->n_vlib_mains) - clib_spinlock_init (&vec_elt (xd->tx_queues, j).lock); + clib_spinlock_init (&vec_elt (xd->tx_queues, j).lock); } - vec_validate_aligned (xd->rx_queues, xd->rx_q_used - 1, + vec_validate_aligned (xd->rx_queues, xd->conf.n_rx_queues - 1, CLIB_CACHE_LINE_BYTES); - for (j = 0; j < xd->rx_q_used; j++) + + for (j = 0; j < xd->conf.n_rx_queues; j++) { dpdk_rx_queue_t *rxq = vec_elt_at_index (xd->rx_queues, j); u8 bpidx = vlib_buffer_pool_get_default_for_numa ( @@ -130,12 +229,12 @@ dpdk_device_setup (dpdk_device_t * xd) vlib_buffer_pool_t *bp = vlib_get_buffer_pool (vm, bpidx); struct rte_mempool *mp = dpdk_mempool_by_buffer_pool_index[bpidx]; - rv = rte_eth_rx_queue_setup (xd->port_id, j, xd->nb_rx_desc, + rv = rte_eth_rx_queue_setup (xd->port_id, j, xd->conf.n_rx_desc, xd->cpu_socket, 0, mp); /* retry with any other CPU socket */ if (rv < 0) - rv = rte_eth_rx_queue_setup (xd->port_id, j, xd->nb_rx_desc, + rv = rte_eth_rx_queue_setup (xd->port_id, j, xd->conf.n_rx_desc, SOCKET_ID_ANY, 0, mp); rxq->buffer_pool_index = bp->index; @@ -147,7 +246,40 @@ dpdk_device_setup (dpdk_device_t * xd) if (vec_len (xd->errors)) goto error; - rte_eth_dev_set_mtu (xd->port_id, hi->max_packet_bytes); + xd->buffer_flags = + (VLIB_BUFFER_TOTAL_LENGTH_VALID | VLIB_BUFFER_EXT_HDR_VALID); + + if ((rxo & (RTE_ETH_RX_OFFLOAD_TCP_CKSUM | RTE_ETH_RX_OFFLOAD_UDP_CKSUM)) == + (RTE_ETH_RX_OFFLOAD_TCP_CKSUM | RTE_ETH_RX_OFFLOAD_UDP_CKSUM)) + xd->buffer_flags |= + (VNET_BUFFER_F_L4_CHECKSUM_COMPUTED | VNET_BUFFER_F_L4_CHECKSUM_CORRECT); + + dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_RX_IP4_CKSUM, + rxo & RTE_ETH_RX_OFFLOAD_IPV4_CKSUM); + dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_MAYBE_MULTISEG, + rxo & RTE_ETH_RX_OFFLOAD_SCATTER); + dpdk_device_flag_set ( + xd, DPDK_DEVICE_FLAG_TX_OFFLOAD, + (txo & (RTE_ETH_TX_OFFLOAD_TCP_CKSUM | RTE_ETH_TX_OFFLOAD_UDP_CKSUM)) == + (RTE_ETH_TX_OFFLOAD_TCP_CKSUM | RTE_ETH_TX_OFFLOAD_UDP_CKSUM)); + + /* unconditionally set mac filtering cap */ + caps.val = caps.mask = VNET_HW_IF_CAP_MAC_FILTER; + + ethernet_set_flags (vnm, xd->hw_if_index, + ETHERNET_INTERFACE_FLAG_DEFAULT_L3); + + for (int i = 0; i < ARRAY_LEN (tx_off_caps_map); i++) + { + __typeof__ (tx_off_caps_map[0]) *v = tx_off_caps_map + i; + caps.mask |= v->caps; + if ((v->offload & txo) == v->offload) + caps.val |= v->caps; + } + + vnet_hw_if_change_caps (vnm, xd->hw_if_index, &caps); + xd->enabled_rx_off = rxo; + xd->enabled_tx_off = txo; if (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP) dpdk_device_start (xd); @@ -187,17 +319,18 @@ dpdk_setup_interrupts (dpdk_device_t *xd) { vnet_main_t *vnm = vnet_get_main (); vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, xd->hw_if_index); + int int_mode = 0; if (!hi) return; - if (!xd->port_conf.intr_conf.rxq) + if (!xd->conf.enable_rxq_int) return; /* Probe for interrupt support */ if (rte_eth_dev_rx_intr_enable (xd->port_id, 0)) { dpdk_log_info ("probe for interrupt mode for device %U. Failed.\n", - format_dpdk_device_name, xd->port_id); + format_dpdk_device_name, xd->device_index); } else { @@ -205,13 +338,13 @@ dpdk_setup_interrupts (dpdk_device_t *xd) if (!(xd->flags & DPDK_DEVICE_FLAG_INT_UNMASKABLE)) rte_eth_dev_rx_intr_disable (xd->port_id, 0); dpdk_log_info ("Probe for interrupt mode for device %U. Success.\n", - format_dpdk_device_name, xd->port_id); + format_dpdk_device_name, xd->device_index); } if (xd->flags & DPDK_DEVICE_FLAG_INT_SUPPORTED) { - hi->caps |= VNET_HW_INTERFACE_CAP_SUPPORTS_INT_MODE; - for (int q = 0; q < xd->rx_q_used; q++) + int_mode = 1; + for (int q = 0; q < xd->conf.n_rx_queues; q++) { dpdk_rx_queue_t *rxq = vec_elt_at_index (xd->rx_queues, q); clib_file_t f = { 0 }; @@ -219,15 +352,15 @@ dpdk_setup_interrupts (dpdk_device_t *xd) if (rxq->efd < 0) { xd->flags &= ~DPDK_DEVICE_FLAG_INT_SUPPORTED; - hi->caps &= ~VNET_HW_INTERFACE_CAP_SUPPORTS_INT_MODE; + int_mode = 0; break; } f.read_function = dpdk_rx_read_ready; f.flags = UNIX_FILE_EVENT_EDGE_TRIGGERED; f.file_descriptor = rxq->efd; f.private_data = rxq->queue_index; - f.description = - format (0, "%U queue %u", format_dpdk_device_name, xd->port_id, q); + f.description = format (0, "%U queue %u", format_dpdk_device_name, + xd->device_index, q); rxq->clib_file_index = clib_file_add (&file_main, &f); vnet_hw_if_set_rx_queue_file_index (vnm, rxq->queue_index, rxq->clib_file_index); @@ -240,6 +373,11 @@ dpdk_setup_interrupts (dpdk_device_t *xd) } } } + + if (int_mode) + vnet_hw_if_set_caps (vnm, hi->hw_if_index, VNET_HW_IF_CAP_INT_MODE); + else + vnet_hw_if_unset_caps (vnm, hi->hw_if_index, VNET_HW_IF_CAP_INT_MODE); vnet_hw_if_update_runtime_data (vnm, xd->hw_if_index); } @@ -259,6 +397,11 @@ dpdk_device_start (dpdk_device_t * xd) return; } + dpdk_log_debug ("[%u] RX burst function: %U", xd->port_id, + format_dpdk_burst_fn, xd, VLIB_RX); + dpdk_log_debug ("[%u] TX burst function: %U", xd->port_id, + format_dpdk_burst_fn, xd, VLIB_TX); + dpdk_setup_interrupts (xd); if (xd->default_mac_address) @@ -275,8 +418,8 @@ dpdk_device_start (dpdk_device_t * xd) rte_eth_allmulticast_enable (xd->port_id); - dpdk_log_info ("Interface %U started", - format_dpdk_device_name, xd->port_id); + dpdk_log_info ("Interface %U started", format_dpdk_device_name, + xd->device_index); } void @@ -289,8 +432,8 @@ dpdk_device_stop (dpdk_device_t * xd) rte_eth_dev_stop (xd->port_id); clib_memset (&xd->link, 0, sizeof (struct rte_eth_link)); - dpdk_log_info ("Interface %U stopped", - format_dpdk_device_name, xd->port_id); + dpdk_log_info ("Interface %U stopped", format_dpdk_device_name, + xd->device_index); } void vl_api_force_rpc_call_main_thread (void *fp, u8 * data, u32 data_length); @@ -311,10 +454,11 @@ dpdk_port_state_callback_inline (dpdk_portid_t port_id, rte_eth_link_get_nowait (port_id, &link); u8 link_up = link.link_status; if (link_up) - dpdk_log_info ("Port %d Link Up - speed %u Mbps - %s", - port_id, (unsigned) link.link_speed, - (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? - "full-duplex" : "half-duplex"); + dpdk_log_info ("Port %d Link Up - speed %u Mbps - %s", port_id, + (unsigned) link.link_speed, + (link.link_duplex == RTE_ETH_LINK_FULL_DUPLEX) ? + "full-duplex" : + "half-duplex"); else dpdk_log_info ("Port %d Link Down\n\n", port_id); @@ -337,12 +481,17 @@ dpdk_get_pci_device (const struct rte_eth_dev_info *info) const struct rte_bus *bus; bus = rte_bus_find_by_device (info->device); +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + if (bus && !strcmp (rte_bus_name (bus), "pci")) +#else if (bus && !strcmp (bus->name, "pci")) +#endif return RTE_DEV_TO_PCI (info->device); else return NULL; } +#ifdef __linux__ /* If this device is VMBUS return pointer to info, otherwise NULL */ struct rte_vmbus_device * dpdk_get_vmbus_device (const struct rte_eth_dev_info *info) @@ -350,11 +499,16 @@ dpdk_get_vmbus_device (const struct rte_eth_dev_info *info) const struct rte_bus *bus; bus = rte_bus_find_by_device (info->device); +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + if (bus && !strcmp (rte_bus_name (bus), "vmbus")) +#else if (bus && !strcmp (bus->name, "vmbus")) +#endif return container_of (info->device, struct rte_vmbus_device, device); else return NULL; } +#endif /* __linux__ */ /* * fd.io coding-style-patch-verification: ON diff --git a/src/plugins/dpdk/device/device.c b/src/plugins/dpdk/device/device.c index 7c083e1dcf4..0ba59562838 100644 --- a/src/plugins/dpdk/device/device.c +++ b/src/plugins/dpdk/device/device.c @@ -25,7 +25,6 @@ #include <vlib/unix/unix.h> #define foreach_dpdk_tx_func_error \ - _(BAD_RETVAL, "DPDK tx function returned an error") \ _(PKT_DROP, "Tx packet drops (dpdk tx failure)") typedef enum @@ -153,52 +152,30 @@ dpdk_validate_rte_mbuf (vlib_main_t * vm, vlib_buffer_t * b, * support multiple queues. It returns the number of packets untransmitted * If all packets are transmitted (the normal case), the function returns 0. */ -static_always_inline - u32 tx_burst_vector_internal (vlib_main_t * vm, - dpdk_device_t * xd, - struct rte_mbuf **mb, u32 n_left) +static_always_inline u32 +tx_burst_vector_internal (vlib_main_t *vm, dpdk_device_t *xd, + struct rte_mbuf **mb, u32 n_left, int queue_id, + u8 is_shared) { - dpdk_main_t *dm = &dpdk_main; dpdk_tx_queue_t *txq; u32 n_retry; int n_sent = 0; - int queue_id; n_retry = 16; - queue_id = vm->thread_index % xd->tx_q_used; txq = vec_elt_at_index (xd->tx_queues, queue_id); do { - clib_spinlock_lock_if_init (&txq->lock); + if (is_shared) + clib_spinlock_lock (&txq->lock); - if (PREDICT_TRUE (xd->flags & DPDK_DEVICE_FLAG_PMD)) - { - /* no wrap, transmit in one burst */ - n_sent = rte_eth_tx_burst (xd->port_id, queue_id, mb, n_left); - n_retry--; - } - else - { - ASSERT (0); - n_sent = 0; - } - - clib_spinlock_unlock_if_init (&txq->lock); - - if (PREDICT_FALSE (n_sent < 0)) - { - // emit non-fatal message, bump counter - vnet_main_t *vnm = dm->vnet_main; - vnet_interface_main_t *im = &vnm->interface_main; - u32 node_index; + /* no wrap, transmit in one burst */ + n_sent = rte_eth_tx_burst (xd->port_id, queue_id, mb, n_left); - node_index = vec_elt_at_index (im->hw_interfaces, - xd->hw_if_index)->tx_node_index; + if (is_shared) + clib_spinlock_unlock (&txq->lock); - vlib_error_count (vm, node_index, DPDK_TX_FUNC_ERROR_BAD_RETVAL, 1); - return n_left; // untransmitted packets - } + n_retry--; n_left -= n_sent; mb += n_sent; } @@ -221,7 +198,8 @@ dpdk_buffer_tx_offload (dpdk_device_t * xd, vlib_buffer_t * b, { int is_ip4 = b->flags & VNET_BUFFER_F_IS_IP4; u32 tso = b->flags & VNET_BUFFER_F_GSO, max_pkt_len; - u32 ip_cksum, tcp_cksum, udp_cksum; + u32 ip_cksum, tcp_cksum, udp_cksum, outer_hdr_len = 0; + u32 outer_ip_cksum, vxlan_tunnel; u64 ol_flags; vnet_buffer_oflags_t oflags = 0; @@ -233,25 +211,49 @@ dpdk_buffer_tx_offload (dpdk_device_t * xd, vlib_buffer_t * b, ip_cksum = oflags & VNET_BUFFER_OFFLOAD_F_IP_CKSUM; tcp_cksum = oflags & VNET_BUFFER_OFFLOAD_F_TCP_CKSUM; udp_cksum = oflags & VNET_BUFFER_OFFLOAD_F_UDP_CKSUM; + outer_ip_cksum = oflags & VNET_BUFFER_OFFLOAD_F_OUTER_IP_CKSUM; + vxlan_tunnel = oflags & VNET_BUFFER_OFFLOAD_F_TNL_VXLAN; - mb->l2_len = vnet_buffer (b)->l3_hdr_offset - b->current_data; - mb->l3_len = vnet_buffer (b)->l4_hdr_offset - - vnet_buffer (b)->l3_hdr_offset; - mb->outer_l3_len = 0; - mb->outer_l2_len = 0; - ol_flags = is_ip4 ? PKT_TX_IPV4 : PKT_TX_IPV6; - ol_flags |= ip_cksum ? PKT_TX_IP_CKSUM : 0; - ol_flags |= tcp_cksum ? PKT_TX_TCP_CKSUM : 0; - ol_flags |= udp_cksum ? PKT_TX_UDP_CKSUM : 0; + ol_flags = is_ip4 ? RTE_MBUF_F_TX_IPV4 : RTE_MBUF_F_TX_IPV6; + ol_flags |= ip_cksum ? RTE_MBUF_F_TX_IP_CKSUM : 0; + ol_flags |= tcp_cksum ? RTE_MBUF_F_TX_TCP_CKSUM : 0; + ol_flags |= udp_cksum ? RTE_MBUF_F_TX_UDP_CKSUM : 0; + + if (vxlan_tunnel) + { + ol_flags |= outer_ip_cksum ? + RTE_MBUF_F_TX_OUTER_IPV4 | RTE_MBUF_F_TX_OUTER_IP_CKSUM : + RTE_MBUF_F_TX_OUTER_IPV6; + ol_flags |= RTE_MBUF_F_TX_TUNNEL_VXLAN; + mb->l2_len = + vnet_buffer (b)->l3_hdr_offset - vnet_buffer2 (b)->outer_l4_hdr_offset; + mb->l3_len = + vnet_buffer (b)->l4_hdr_offset - vnet_buffer (b)->l3_hdr_offset; + mb->outer_l2_len = + vnet_buffer2 (b)->outer_l3_hdr_offset - b->current_data; + mb->outer_l3_len = vnet_buffer2 (b)->outer_l4_hdr_offset - + vnet_buffer2 (b)->outer_l3_hdr_offset; + outer_hdr_len = mb->outer_l2_len + mb->outer_l3_len; + } + else + { + mb->l2_len = vnet_buffer (b)->l3_hdr_offset - b->current_data; + mb->l3_len = + vnet_buffer (b)->l4_hdr_offset - vnet_buffer (b)->l3_hdr_offset; + mb->outer_l2_len = 0; + mb->outer_l3_len = 0; + } if (tso) { mb->l4_len = vnet_buffer2 (b)->gso_l4_hdr_sz; mb->tso_segsz = vnet_buffer2 (b)->gso_size; /* ensure packet is large enough to require tso */ - max_pkt_len = mb->l2_len + mb->l3_len + mb->l4_len + mb->tso_segsz; + max_pkt_len = + outer_hdr_len + mb->l2_len + mb->l3_len + mb->l4_len + mb->tso_segsz; if (mb->tso_segsz != 0 && mb->pkt_len > max_pkt_len) - ol_flags |= (tcp_cksum ? PKT_TX_TCP_SEG : PKT_TX_UDP_SEG); + ol_flags |= + (tcp_cksum ? RTE_MBUF_F_TX_TCP_SEG : RTE_MBUF_F_TX_UDP_SEG); } mb->ol_flags |= ol_flags; @@ -274,11 +276,13 @@ VNET_DEVICE_CLASS_TX_FN (dpdk_device_class) (vlib_main_t * vm, dpdk_main_t *dm = &dpdk_main; vnet_interface_output_runtime_t *rd = (void *) node->runtime_data; dpdk_device_t *xd = vec_elt_at_index (dm->devices, rd->dev_instance); + vnet_hw_if_tx_frame_t *tf = vlib_frame_scalar_args (f); u32 n_packets = f->n_vectors; u32 n_left; u32 thread_index = vm->thread_index; - int queue_id = thread_index; - u32 tx_pkts = 0, all_or_flags = 0; + int queue_id = tf->queue_id; + u8 is_shared = tf->shared_queue; + u32 tx_pkts = 0; dpdk_per_thread_data_t *ptd = vec_elt_at_index (dm->per_thread_data, thread_index); struct rte_mbuf **mb; @@ -310,7 +314,6 @@ VNET_DEVICE_CLASS_TX_FN (dpdk_device_class) (vlib_main_t * vm, b[3] = vlib_buffer_from_rte_mbuf (mb[3]); or_flags = b[0]->flags | b[1]->flags | b[2]->flags | b[3]->flags; - all_or_flags |= or_flags; if (or_flags & VLIB_BUFFER_NEXT_PRESENT) { @@ -368,7 +371,6 @@ VNET_DEVICE_CLASS_TX_FN (dpdk_device_class) (vlib_main_t * vm, b[1] = vlib_buffer_from_rte_mbuf (mb[1]); or_flags = b[0]->flags | b[1]->flags; - all_or_flags |= or_flags; if (or_flags & VLIB_BUFFER_NEXT_PRESENT) { @@ -404,7 +406,6 @@ VNET_DEVICE_CLASS_TX_FN (dpdk_device_class) (vlib_main_t * vm, while (n_left > 0) { b[0] = vlib_buffer_from_rte_mbuf (mb[0]); - all_or_flags |= b[0]->flags; dpdk_validate_rte_mbuf (vm, b[0], 1); dpdk_buffer_tx_offload (xd, b[0], mb[0]); @@ -419,7 +420,8 @@ VNET_DEVICE_CLASS_TX_FN (dpdk_device_class) (vlib_main_t * vm, /* transmit as many packets as possible */ tx_pkts = n_packets = mb - ptd->mbufs; - n_left = tx_burst_vector_internal (vm, xd, ptd->mbufs, n_packets); + n_left = tx_burst_vector_internal (vm, xd, ptd->mbufs, n_packets, queue_id, + is_shared); { /* If there is no callback then drop any non-transmitted packets */ @@ -475,7 +477,7 @@ dpdk_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags) if (vec_len (xd->errors)) return clib_error_create ("Interface start failed"); xd->flags |= DPDK_DEVICE_FLAG_ADMIN_UP; - f64 now = vlib_time_now (dm->vlib_main); + f64 now = vlib_time_now (vlib_get_main ()); dpdk_update_counters (xd, now); dpdk_update_link_state (xd, now); } @@ -511,7 +513,7 @@ dpdk_set_interface_next_node (vnet_main_t * vnm, u32 hw_if_index, } xd->per_interface_next_index = - vlib_node_add_next (xm->vlib_main, dpdk_input_node.index, node_index); + vlib_node_add_next (vlib_get_main (), dpdk_input_node.index, node_index); } @@ -533,11 +535,8 @@ dpdk_subif_add_del_function (vnet_main_t * vnm, else if (xd->num_subifs) xd->num_subifs--; - if ((xd->flags & DPDK_DEVICE_FLAG_PMD) == 0) - goto done; - /* 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) @@ -551,7 +550,7 @@ dpdk_subif_add_del_function (vnet_main_t * vnm, } vlan_offload = rte_eth_dev_get_vlan_offload (xd->port_id); - vlan_offload |= ETH_VLAN_FILTER_OFFLOAD; + vlan_offload |= RTE_ETH_VLAN_FILTER_OFFLOAD; if ((r = rte_eth_dev_set_vlan_offload (xd->port_id, vlan_offload))) { @@ -625,7 +624,6 @@ dpdk_interface_set_rss_queues (struct vnet_main_t *vnm, clib_memset (reta, 0, dev_info.reta_size * sizeof (*reta)); valid_queue_count = 0; - /* *INDENT-OFF* */ clib_bitmap_foreach (i, bitmap) { if (i >= dev_info.nb_rx_queues) { @@ -634,7 +632,6 @@ dpdk_interface_set_rss_queues (struct vnet_main_t *vnm, } reta[valid_queue_count++] = i; } - /* *INDENT-ON* */ /* check valid_queue_count not zero, make coverity happy */ if (valid_queue_count == 0) @@ -651,10 +648,8 @@ dpdk_interface_set_rss_queues (struct vnet_main_t *vnm, } /* update reta table */ - reta_conf = - (struct rte_eth_rss_reta_entry64 *) clib_mem_alloc (dev_info.reta_size / - RTE_RETA_GROUP_SIZE * - sizeof (*reta_conf)); + reta_conf = (struct rte_eth_rss_reta_entry64 *) clib_mem_alloc ( + dev_info.reta_size / RTE_ETH_RETA_GROUP_SIZE * sizeof (*reta_conf)); if (reta_conf == NULL) { err = clib_error_return (0, "clib_mem_alloc failed"); @@ -662,13 +657,13 @@ dpdk_interface_set_rss_queues (struct vnet_main_t *vnm, } clib_memset (reta_conf, 0, - dev_info.reta_size / RTE_RETA_GROUP_SIZE * - sizeof (*reta_conf)); + dev_info.reta_size / RTE_ETH_RETA_GROUP_SIZE * + sizeof (*reta_conf)); for (i = 0; i < dev_info.reta_size; i++) { - uint32_t reta_id = i / RTE_RETA_GROUP_SIZE; - uint32_t reta_pos = i % RTE_RETA_GROUP_SIZE; + uint32_t reta_id = i / RTE_ETH_RETA_GROUP_SIZE; + uint32_t reta_pos = i % RTE_ETH_RETA_GROUP_SIZE; reta_conf[reta_id].mask = UINT64_MAX; reta_conf[reta_id].reta[reta_pos] = reta[i]; @@ -726,7 +721,6 @@ dpdk_interface_rx_mode_change (vnet_main_t *vnm, u32 hw_if_index, u32 qid, return 0; } -/* *INDENT-OFF* */ VNET_DEVICE_CLASS (dpdk_device_class) = { .name = "dpdk", .tx_function_n_errors = DPDK_TX_FUNC_N_ERROR, @@ -745,7 +739,6 @@ VNET_DEVICE_CLASS (dpdk_device_class) = { .set_rss_queues_function = dpdk_interface_set_rss_queues, .rx_mode_change_function = dpdk_interface_rx_mode_change, }; -/* *INDENT-ON* */ #define UP_DOWN_FLAG_EVENT 1 @@ -792,14 +785,12 @@ admin_up_down_process (vlib_main_t * vm, return 0; /* or not */ } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (admin_up_down_process_node) = { .function = admin_up_down_process, .type = VLIB_NODE_TYPE_PROCESS, .name = "admin-up-down-process", .process_log2_n_stack_bytes = 17, // 256KB }; -/* *INDENT-ON* */ /* * fd.io coding-style-patch-verification: ON diff --git a/src/plugins/dpdk/device/dpdk.h b/src/plugins/dpdk/device/dpdk.h index 66794a4e67c..a069fbe3818 100644 --- a/src/plugins/dpdk/device/dpdk.h +++ b/src/plugins/dpdk/device/dpdk.h @@ -22,27 +22,25 @@ #include <rte_config.h> -#include <rte_common.h> -#include <rte_dev.h> -#include <rte_memory.h> #include <rte_eal.h> -#include <rte_per_lcore.h> -#include <rte_cycles.h> -#include <rte_lcore.h> -#include <rte_per_lcore.h> -#include <rte_interrupts.h> -#include <rte_pci.h> +#include <rte_bus_pci.h> +#ifdef __linux__ #include <rte_bus_vmbus.h> -#include <rte_ether.h> +#endif /* __linux__ */ #include <rte_ethdev.h> -#include <rte_ring.h> -#include <rte_mempool.h> -#include <rte_mbuf.h> #include <rte_version.h> -#include <rte_sched.h> #include <rte_net.h> -#include <rte_bus_pci.h> -#include <rte_flow.h> +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) +#include <rte_bus.h> +#include <rte_pci.h> +#include <ctype.h> + +#include <bus_driver.h> +#include <bus_pci_driver.h> +#ifdef __linux__ +#include <bus_vmbus_driver.h> +#endif /* __linux__ */ +#endif #include <vnet/devices/devices.h> @@ -60,79 +58,11 @@ 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; - -typedef enum -{ - VNET_DPDK_PORT_TYPE_ETH_1G, - VNET_DPDK_PORT_TYPE_ETH_2_5G, - VNET_DPDK_PORT_TYPE_ETH_5G, - VNET_DPDK_PORT_TYPE_ETH_10G, - VNET_DPDK_PORT_TYPE_ETH_20G, - VNET_DPDK_PORT_TYPE_ETH_25G, - VNET_DPDK_PORT_TYPE_ETH_40G, - VNET_DPDK_PORT_TYPE_ETH_50G, - VNET_DPDK_PORT_TYPE_ETH_56G, - VNET_DPDK_PORT_TYPE_ETH_100G, - VNET_DPDK_PORT_TYPE_ETH_SWITCH, - VNET_DPDK_PORT_TYPE_AF_PACKET, - VNET_DPDK_PORT_TYPE_ETH_VF, - VNET_DPDK_PORT_TYPE_VIRTIO_USER, - VNET_DPDK_PORT_TYPE_VHOST_ETHER, - VNET_DPDK_PORT_TYPE_FAILSAFE, - VNET_DPDK_PORT_TYPE_NETVSC, - VNET_DPDK_PORT_TYPE_UNKNOWN, -} dpdk_port_type_t; - typedef uint16_t dpdk_portid_t; #define foreach_dpdk_device_flags \ _ (0, ADMIN_UP, "admin-up") \ _ (1, PROMISC, "promisc") \ - _ (2, PMD, "pmd") \ _ (3, PMD_INIT_FAIL, "pmd-init-fail") \ _ (4, MAYBE_MULTISEG, "maybe-multiseg") \ _ (5, HAVE_SUBIF, "subif") \ @@ -143,12 +73,12 @@ 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 { @@ -177,10 +107,63 @@ typedef struct { CLIB_CACHE_LINE_ALIGN_MARK (cacheline0); clib_spinlock_t lock; + u32 queue_index; } 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; + u32 enable_lsc_int : 1; + u32 enable_rxq_int : 1; + u32 disable_rx_scatter : 1; + u32 program_vlans : 1; + u32 mq_mode_none : 1; + u32 interface_number_from_port_id : 1; + u32 use_intel_phdr_cksum : 1; + u32 int_unmaskable : 1; + vlib_simple_counter_main_t *xstats_counters; +} dpdk_driver_t; + +dpdk_driver_t *dpdk_driver_find (const char *name, const char **desc); + +typedef union +{ + struct + { + u16 disable_multi_seg : 1; + u16 enable_lro : 1; + u16 enable_tso : 1; + u16 enable_tcp_udp_checksum : 1; + u16 enable_outer_checksum_offload : 1; + u16 enable_lsc_int : 1; + u16 enable_rxq_int : 1; + u16 disable_tx_checksum_offload : 1; + u16 disable_rss : 1; + u16 disable_rx_scatter : 1; + u16 n_rx_queues; + u16 n_tx_queues; + u16 n_rx_desc; + u16 n_tx_desc; + u32 max_lro_pkt_size; + u64 rss_hf; + }; + u64 as_u64[3]; +} dpdk_port_conf_t; + +STATIC_ASSERT_SIZEOF (dpdk_port_conf_t, 24); + +typedef struct +{ CLIB_CACHE_LINE_ALIGN_MARK (cacheline0); dpdk_rx_queue_t *rx_queues; @@ -191,33 +174,28 @@ typedef struct u32 hw_if_index; u32 sw_if_index; + u32 buffer_flags; /* next node index if we decide to steal the rx graph arc */ u32 per_interface_next_index; - u16 rx_q_used; - u16 tx_q_used; u16 flags; /* DPDK device port number */ dpdk_portid_t port_id; - dpdk_pmd_t pmd:8; i8 cpu_socket; - CLIB_CACHE_LINE_ALIGN_MARK (cacheline1); - u16 nb_tx_desc; - u16 nb_rx_desc; + 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; - /* PMD related */ - struct rte_eth_conf port_conf; - struct rte_eth_txconf tx_conf; - /* flow related */ u32 supported_flow_actions; dpdk_flow_entry_t *flow_entries; /* pool */ @@ -226,9 +204,6 @@ typedef struct u32 parked_loop_count; struct rte_flow_error last_flow_error; - /* af_packet instance number */ - u16 af_packet_instance_num; - struct rte_eth_link link; f64 time_last_link_update; @@ -236,26 +211,37 @@ 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; + /* maximum supported max frame size */ + u32 max_supported_frame_size; + + /* due to lack of API to get ethernet max_frame_size we store information + * deducted from device info */ + u8 driver_frame_overhead; + /* error string */ clib_error_t *errors; + dpdk_port_conf_t conf; } dpdk_device_t; +#define DPDK_MIN_POLL_INTERVAL (0.001) /* 1msec */ + #define DPDK_STATS_POLL_INTERVAL (10.0) -#define DPDK_MIN_STATS_POLL_INTERVAL (0.001) /* 1msec */ +#define DPDK_MIN_STATS_POLL_INTERVAL DPDK_MIN_POLL_INTERVAL #define DPDK_LINK_POLL_INTERVAL (3.0) -#define DPDK_MIN_LINK_POLL_INTERVAL (0.001) /* 1msec */ - -#define foreach_dpdk_device_config_item \ - _ (num_rx_queues) \ - _ (num_tx_queues) \ - _ (num_rx_desc) \ - _ (num_tx_desc) \ +#define DPDK_MIN_LINK_POLL_INTERVAL DPDK_MIN_POLL_INTERVAL + +#define foreach_dpdk_device_config_item \ + _ (num_rx_queues) \ + _ (num_tx_queues) \ + _ (num_rx_desc) \ + _ (num_tx_desc) \ + _ (max_lro_pkt_size) \ + _ (disable_rxq_int) \ _ (rss_fn) typedef enum @@ -274,11 +260,8 @@ typedef struct }; dpdk_device_addr_type_t dev_addr_type; u8 *name; + u8 *tag; u8 is_blacklisted; - u8 vlan_strip_offload; -#define DPDK_DEVICE_VLAN_STRIP_DEFAULT 0 -#define DPDK_DEVICE_VLAN_STRIP_OFF 1 -#define DPDK_DEVICE_VLAN_STRIP_ON 2 #define _(x) uword x; foreach_dpdk_device_config_item @@ -300,9 +283,7 @@ typedef struct u8 **eal_init_args; u8 *eal_init_args_str; u8 *uio_driver_name; - u8 no_multi_seg; - u8 enable_tcp_udp_checksum; - u8 no_tx_checksum_offload; + u8 uio_bind_force; u8 enable_telemetry; u16 max_simd_bitwidth; @@ -310,13 +291,6 @@ typedef struct #define DPDK_MAX_SIMD_BITWIDTH_256 256 #define DPDK_MAX_SIMD_BITWIDTH_512 512 - /* Required config parameters */ - u8 coremask_set_manually; - u8 nchannels_set_manually; - u32 coremask; - u32 nchannels; - u32 num_crypto_mbufs; - /* * format interface names ala xxxEthernet%d/%d/%d instead of * xxxEthernet%x/%x/%x. @@ -347,20 +321,16 @@ typedef struct u32 buffers[DPDK_RX_BURST_SZ]; u16 next[DPDK_RX_BURST_SZ]; u16 etype[DPDK_RX_BURST_SZ]; - u16 flags[DPDK_RX_BURST_SZ]; + u32 flags[DPDK_RX_BURST_SZ]; vlib_buffer_t buffer_template; } dpdk_per_thread_data_t; typedef struct { - /* Devices */ dpdk_device_t *devices; dpdk_per_thread_data_t *per_thread_data; - /* buffer flags template, configurable to enable/disable tcp / udp cksum */ - u32 buffer_flags_template; - /* * flag indicating that a posted admin up/down * (via post_sw_interface_set_flags) is in progress @@ -371,10 +341,8 @@ typedef struct f64 link_state_poll_interval; f64 stat_poll_interval; - /* convenience */ - vlib_main_t *vlib_main; - vnet_main_t *vnet_main; dpdk_config_main_t *conf; + dpdk_port_conf_t default_port_conf; /* API message ID base */ u16 msg_id_base; @@ -382,7 +350,6 @@ typedef struct /* logging */ vlib_log_class_t log_default; vlib_log_class_t log_cryptodev; - vlib_log_class_t log_ipsec; } dpdk_main_t; extern dpdk_main_t dpdk_main; @@ -440,35 +407,39 @@ typedef enum vlib_log(VLIB_LOG_LEVEL_NOTICE, dpdk_main.log_default, __VA_ARGS__) #define dpdk_log_info(...) \ vlib_log(VLIB_LOG_LEVEL_INFO, dpdk_main.log_default, __VA_ARGS__) +#define dpdk_log_debug(...) \ + vlib_log (VLIB_LOG_LEVEL_DEBUG, dpdk_main.log_default, __VA_ARGS__) void dpdk_update_link_state (dpdk_device_t * xd, f64 now); -#define foreach_dpdk_rss_hf \ - _(0, ETH_RSS_FRAG_IPV4, "ipv4-frag") \ - _(1, ETH_RSS_NONFRAG_IPV4_TCP, "ipv4-tcp") \ - _(2, ETH_RSS_NONFRAG_IPV4_UDP, "ipv4-udp") \ - _(3, ETH_RSS_NONFRAG_IPV4_SCTP, "ipv4-sctp") \ - _(4, ETH_RSS_NONFRAG_IPV4_OTHER, "ipv4-other") \ - _(5, ETH_RSS_IPV4, "ipv4") \ - _(6, ETH_RSS_IPV6_TCP_EX, "ipv6-tcp-ex") \ - _(7, ETH_RSS_IPV6_UDP_EX, "ipv6-udp-ex") \ - _(8, ETH_RSS_FRAG_IPV6, "ipv6-frag") \ - _(9, ETH_RSS_NONFRAG_IPV6_TCP, "ipv6-tcp") \ - _(10, ETH_RSS_NONFRAG_IPV6_UDP, "ipv6-udp") \ - _(11, ETH_RSS_NONFRAG_IPV6_SCTP, "ipv6-sctp") \ - _(12, ETH_RSS_NONFRAG_IPV6_OTHER, "ipv6-other") \ - _(13, ETH_RSS_IPV6_EX, "ipv6-ex") \ - _(14, ETH_RSS_IPV6, "ipv6") \ - _(15, ETH_RSS_L2_PAYLOAD, "l2-payload") \ - _(16, ETH_RSS_PORT, "port") \ - _(17, ETH_RSS_VXLAN, "vxlan") \ - _(18, ETH_RSS_GENEVE, "geneve") \ - _(19, ETH_RSS_NVGRE, "nvgre") \ - _(20, ETH_RSS_GTPU, "gtpu") \ - _(60, ETH_RSS_L4_DST_ONLY, "l4-dst-only") \ - _(61, ETH_RSS_L4_SRC_ONLY, "l4-src-only") \ - _(62, ETH_RSS_L3_DST_ONLY, "l3-dst-only") \ - _(63, ETH_RSS_L3_SRC_ONLY, "l3-src-only") +#define foreach_dpdk_rss_hf \ + _ (0, RTE_ETH_RSS_FRAG_IPV4, "ipv4-frag") \ + _ (1, RTE_ETH_RSS_NONFRAG_IPV4_TCP, "ipv4-tcp") \ + _ (2, RTE_ETH_RSS_NONFRAG_IPV4_UDP, "ipv4-udp") \ + _ (3, RTE_ETH_RSS_NONFRAG_IPV4_SCTP, "ipv4-sctp") \ + _ (4, RTE_ETH_RSS_NONFRAG_IPV4_OTHER, "ipv4-other") \ + _ (5, RTE_ETH_RSS_IPV4, "ipv4") \ + _ (6, RTE_ETH_RSS_IPV6_TCP_EX, "ipv6-tcp-ex") \ + _ (7, RTE_ETH_RSS_IPV6_UDP_EX, "ipv6-udp-ex") \ + _ (8, RTE_ETH_RSS_FRAG_IPV6, "ipv6-frag") \ + _ (9, RTE_ETH_RSS_NONFRAG_IPV6_TCP, "ipv6-tcp") \ + _ (10, RTE_ETH_RSS_NONFRAG_IPV6_UDP, "ipv6-udp") \ + _ (11, RTE_ETH_RSS_NONFRAG_IPV6_SCTP, "ipv6-sctp") \ + _ (12, RTE_ETH_RSS_NONFRAG_IPV6_OTHER, "ipv6-other") \ + _ (13, RTE_ETH_RSS_IPV6_EX, "ipv6-ex") \ + _ (14, RTE_ETH_RSS_IPV6, "ipv6") \ + _ (15, RTE_ETH_RSS_L2_PAYLOAD, "l2-payload") \ + _ (16, RTE_ETH_RSS_PORT, "port") \ + _ (17, RTE_ETH_RSS_VXLAN, "vxlan") \ + _ (18, RTE_ETH_RSS_GENEVE, "geneve") \ + _ (19, RTE_ETH_RSS_NVGRE, "nvgre") \ + _ (20, RTE_ETH_RSS_GTPU, "gtpu") \ + _ (21, RTE_ETH_RSS_ESP, "esp") \ + _ (22, RTE_ETH_RSS_L2TPV3, "l2tpv3") \ + _ (60, RTE_ETH_RSS_L4_DST_ONLY, "l4-dst-only") \ + _ (61, RTE_ETH_RSS_L4_SRC_ONLY, "l4-src-only") \ + _ (62, RTE_ETH_RSS_L3_DST_ONLY, "l3-dst-only") \ + _ (63, RTE_ETH_RSS_L3_SRC_ONLY, "l3-src-only") format_function_t format_dpdk_device_name; format_function_t format_dpdk_device; @@ -481,6 +452,8 @@ format_function_t format_dpdk_flow; format_function_t format_dpdk_rss_hf_name; format_function_t format_dpdk_rx_offload_caps; format_function_t format_dpdk_tx_offload_caps; +format_function_t format_dpdk_burst_fn; +format_function_t format_dpdk_rte_device; vnet_flow_dev_ops_function_t dpdk_flow_ops_fn; clib_error_t *unformat_rss_fn (unformat_input_t * input, uword * rss_fn); diff --git a/src/plugins/dpdk/device/dpdk_priv.h b/src/plugins/dpdk/device/dpdk_priv.h index a5a8a2ad57d..e5b5a35df80 100644 --- a/src/plugins/dpdk/device/dpdk_priv.h +++ b/src/plugins/dpdk/device/dpdk_priv.h @@ -15,15 +15,7 @@ #define DPDK_NB_RX_DESC_DEFAULT 1024 #define DPDK_NB_TX_DESC_DEFAULT 1024 -#define DPDK_NB_RX_DESC_VIRTIO 256 -#define DPDK_NB_TX_DESC_VIRTIO 256 - -#define I40E_DEV_ID_SFP_XL710 0x1572 -#define I40E_DEV_ID_QSFP_A 0x1583 -#define I40E_DEV_ID_QSFP_B 0x1584 -#define I40E_DEV_ID_QSFP_C 0x1585 -#define I40E_DEV_ID_10G_BASE_T 0x1586 -#define I40E_DEV_ID_VF 0x154C +#define DPDK_MAX_LRO_SIZE_DEFAULT 65536 /* These args appear by themselves */ #define foreach_eal_double_hyphen_predicate_arg \ @@ -32,10 +24,6 @@ _(no-hpet) \ _(no-huge) \ _(vmware-tsc-map) -#define foreach_eal_single_hyphen_mandatory_arg \ -_(coremask, c) \ -_(nchannels, n) \ - #define foreach_eal_single_hyphen_arg \ _(mem-alloc-request, m) \ _(force-ranks, r) @@ -48,18 +36,29 @@ _(proc-type) \ _(file-prefix) \ _(vdev) \ _(log-level) \ +_(block) \ _(iova-mode) \ _(base-virtaddr) /* clang-format on */ -static inline void -dpdk_get_xstats (dpdk_device_t * xd) +static_always_inline void +dpdk_device_flag_set (dpdk_device_t *xd, __typeof__ (xd->flags) flag, int val) { - int len, ret; + xd->flags = val ? xd->flags | flag : xd->flags & ~flag; +} + +void dpdk_counters_xstats_init (dpdk_device_t *xd); +static inline void +dpdk_get_xstats (dpdk_device_t *xd, u32 thread_index) +{ + int ret; + int i; + int len; if (!(xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP)) return; - + if (xd->driver == 0) + return; len = rte_eth_xstats_get (xd->port_id, NULL, 0); if (len < 0) return; @@ -69,11 +68,26 @@ dpdk_get_xstats (dpdk_device_t * xd) ret = rte_eth_xstats_get (xd->port_id, xd->xstats, len); if (ret < 0 || ret > len) { - _vec_len (xd->xstats) = 0; + /* Failed, expand vector and try again on next time around the track. */ + vec_validate (xd->xstats, ret - 1); + vec_set_len (xd->xstats, 0); + dpdk_log_warn ("rte_eth_xstats_get(%d) failed: %d", xd->port_id, ret); return; } - - _vec_len (xd->xstats) = len; + if (len == vec_len (xd->driver->xstats_counters)) + { + vec_foreach_index (i, xd->xstats) + { + vlib_set_simple_counter (&xd->driver->xstats_counters[i], + thread_index, xd->sw_if_index, + xd->xstats[i].value); + } + } + else + { + dpdk_log_warn ("rte_eth_xstats_get vector size mismatch (%d/%d", len, + vec_len (xd->driver->xstats_counters)); + } } #define DPDK_UPDATE_COUNTER(vnm, tidx, xd, stat, cnt) \ @@ -100,10 +114,6 @@ dpdk_update_counters (dpdk_device_t * xd, f64 now) vnet_main_t *vnm = vnet_get_main (); u32 thread_index = vlib_get_thread_index (); - /* only update counters for PMD interfaces */ - if ((xd->flags & DPDK_DEVICE_FLAG_PMD) == 0) - return; - xd->time_last_stats_update = now ? now : xd->time_last_stats_update; clib_memcpy_fast (&xd->last_stats, &xd->stats, sizeof (xd->last_stats)); rte_eth_stats_get (xd->port_id, &xd->stats); @@ -116,9 +126,122 @@ dpdk_update_counters (dpdk_device_t * xd, f64 now) DPDK_UPDATE_COUNTER (vnm, thread_index, xd, ierrors, VNET_INTERFACE_COUNTER_RX_ERROR); - dpdk_get_xstats (xd); + dpdk_get_xstats (xd, thread_index); } +#if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0) +#define RTE_MBUF_F_RX_FDIR PKT_RX_FDIR +#define RTE_MBUF_F_RX_FDIR_FLX PKT_RX_FDIR_FLX +#define RTE_MBUF_F_RX_FDIR_ID PKT_RX_FDIR_ID +#define RTE_MBUF_F_RX_IEEE1588_PTP PKT_RX_IEEE1588_PTP +#define RTE_MBUF_F_RX_IEEE1588_TMST PKT_RX_IEEE1588_TMST +#define RTE_MBUF_F_RX_IP_CKSUM_BAD PKT_RX_IP_CKSUM_BAD +#define RTE_MBUF_F_RX_IP_CKSUM_GOOD PKT_RX_IP_CKSUM_GOOD +#define RTE_MBUF_F_RX_IP_CKSUM_NONE PKT_RX_IP_CKSUM_GOOD +#define RTE_MBUF_F_RX_L4_CKSUM_BAD PKT_RX_L4_CKSUM_BAD +#define RTE_MBUF_F_RX_L4_CKSUM_GOOD PKT_RX_L4_CKSUM_GOOD +#define RTE_MBUF_F_RX_L4_CKSUM_NONE PKT_RX_L4_CKSUM_GOOD +#define RTE_MBUF_F_RX_LRO PKT_RX_LRO +#define RTE_MBUF_F_RX_OUTER_IP_CKSUM_BAD PKT_RX_OUTER_IP_CKSUM_BAD +#define RTE_MBUF_F_RX_OUTER_L4_CKSUM_BAD PKT_RX_OUTER_L4_CKSUM_GOOD +#define RTE_MBUF_F_RX_OUTER_L4_CKSUM_GOOD PKT_RX_OUTER_L4_CKSUM_GOOD +#define RTE_MBUF_F_RX_OUTER_L4_CKSUM_NONE PKT_RX_OUTER_L4_CKSUM_GOOD +#define RTE_MBUF_F_RX_QINQ PKT_RX_QINQ +#define RTE_MBUF_F_RX_QINQ_STRIPPED PKT_RX_QINQ_STRIPPED +#define RTE_MBUF_F_RX_RSS_HASH PKT_RX_RSS_HASH +#define RTE_MBUF_F_RX_SEC_OFFLOAD PKT_RX_SEC_OFFLOAD +#define RTE_MBUF_F_RX_SEC_OFFLOAD_FAILED PKT_RX_SEC_OFFLOAD_FAILED +#define RTE_MBUF_F_RX_VLAN PKT_RX_VLAN +#define RTE_MBUF_F_RX_VLAN_STRIPPED PKT_RX_VLAN_STRIPPED +#define RTE_MBUF_F_TX_IEEE1588_TMST PKT_TX_IEEE1588_TMST +#define RTE_MBUF_F_TX_IPV4 PKT_TX_IPV4 +#define RTE_MBUF_F_TX_IPV6 PKT_TX_IPV6 +#define RTE_MBUF_F_TX_IP_CKSUM PKT_TX_IP_CKSUM +#define RTE_MBUF_F_TX_MACSEC PKT_TX_MACSEC +#define RTE_MBUF_F_TX_OUTER_IPV4 PKT_TX_OUTER_IPV4 +#define RTE_MBUF_F_TX_OUTER_IPV6 PKT_TX_OUTER_IPV6 +#define RTE_MBUF_F_TX_OUTER_IP_CKSUM PKT_TX_OUTER_IP_CKSUM +#define RTE_MBUF_F_TX_OUTER_UDP_CKSUM PKT_TX_OUTER_UDP_CKSUM +#define RTE_MBUF_F_TX_QINQ PKT_TX_QINQ +#define RTE_MBUF_F_TX_SCTP_CKSUM PKT_TX_SCTP_CKSUM +#define RTE_MBUF_F_TX_SEC_OFFLOAD PKT_TX_SEC_OFFLOAD +#define RTE_MBUF_F_TX_TCP_CKSUM PKT_TX_TCP_CKSUM +#define RTE_MBUF_F_TX_TCP_SEG PKT_TX_TCP_SEG +#define RTE_MBUF_F_TX_TUNNEL_GENEVE PKT_TX_TUNNEL_GENEVE +#define RTE_MBUF_F_TX_TUNNEL_GRE PKT_TX_TUNNEL_GRE +#define RTE_MBUF_F_TX_TUNNEL_GTP PKT_TX_TUNNEL_GTP +#define RTE_MBUF_F_TX_TUNNEL_IP PKT_TX_TUNNEL_IP +#define RTE_MBUF_F_TX_TUNNEL_IPIP PKT_TX_TUNNEL_IPIP +#define RTE_MBUF_F_TX_TUNNEL_MPLSINUDP PKT_TX_TUNNEL_MPLSINUDP +#define RTE_MBUF_F_TX_TUNNEL_UDP PKT_TX_TUNNEL_UDP +#define RTE_MBUF_F_TX_TUNNEL_VXLAN PKT_TX_TUNNEL_VXLAN +#define RTE_MBUF_F_TX_TUNNEL_VXLAN_GPE PKT_TX_TUNNEL_VXLAN_GPE +#define RTE_MBUF_F_TX_UDP_CKSUM PKT_TX_UDP_CKSUM +#define RTE_MBUF_F_TX_UDP_SEG PKT_TX_UDP_SEG +#define RTE_MBUF_F_TX_VLAN PKT_TX_VLAN +#define RTE_ETH_RSS_FRAG_IPV4 ETH_RSS_FRAG_IPV4 +#define RTE_ETH_RSS_NONFRAG_IPV4_TCP ETH_RSS_NONFRAG_IPV4_TCP +#define RTE_ETH_RSS_NONFRAG_IPV4_UDP ETH_RSS_NONFRAG_IPV4_UDP +#define RTE_ETH_RSS_NONFRAG_IPV4_SCTP ETH_RSS_NONFRAG_IPV4_SCTP +#define RTE_ETH_RSS_NONFRAG_IPV4_OTHER ETH_RSS_NONFRAG_IPV4_OTHER +#define RTE_ETH_RSS_IPV4 ETH_RSS_IPV4 +#define RTE_ETH_RSS_IPV6_TCP_EX ETH_RSS_IPV6_TCP_EX +#define RTE_ETH_RSS_IPV6_UDP_EX ETH_RSS_IPV6_UDP_EX +#define RTE_ETH_RSS_FRAG_IPV6 ETH_RSS_FRAG_IPV6 +#define RTE_ETH_RSS_NONFRAG_IPV6_TCP ETH_RSS_NONFRAG_IPV6_TCP +#define RTE_ETH_RSS_NONFRAG_IPV6_UDP ETH_RSS_NONFRAG_IPV6_UDP +#define RTE_ETH_RSS_NONFRAG_IPV6_SCTP ETH_RSS_NONFRAG_IPV6_SCTP +#define RTE_ETH_RSS_NONFRAG_IPV6_OTHER ETH_RSS_NONFRAG_IPV6_OTHER +#define RTE_ETH_RSS_IPV6_EX ETH_RSS_IPV6_EX +#define RTE_ETH_RSS_IPV6 ETH_RSS_IPV6 +#define RTE_ETH_RSS_L2_PAYLOAD ETH_RSS_L2_PAYLOAD +#define RTE_ETH_RSS_PORT ETH_RSS_PORT +#define RTE_ETH_RSS_VXLAN ETH_RSS_VXLAN +#define RTE_ETH_RSS_GENEVE ETH_RSS_GENEVE +#define RTE_ETH_RSS_NVGRE ETH_RSS_NVGRE +#define RTE_ETH_RSS_GTPU ETH_RSS_GTPU +#define RTE_ETH_RSS_ESP ETH_RSS_ESP +#define RTE_ETH_RSS_L4_DST_ONLY ETH_RSS_L4_DST_ONLY +#define RTE_ETH_RSS_L4_SRC_ONLY ETH_RSS_L4_SRC_ONLY +#define RTE_ETH_RSS_L3_DST_ONLY ETH_RSS_L3_DST_ONLY +#define RTE_ETH_RSS_L3_SRC_ONLY ETH_RSS_L3_SRC_ONLY +#define RTE_ETH_RETA_GROUP_SIZE RTE_RETA_GROUP_SIZE +#define RTE_ETH_TX_OFFLOAD_IPV4_CKSUM DEV_TX_OFFLOAD_IPV4_CKSUM +#define RTE_ETH_TX_OFFLOAD_TCP_CKSUM DEV_TX_OFFLOAD_TCP_CKSUM +#define RTE_ETH_TX_OFFLOAD_UDP_CKSUM DEV_TX_OFFLOAD_UDP_CKSUM +#define RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM +#define RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM DEV_TX_OFFLOAD_OUTER_UDP_CKSUM +#define RTE_ETH_TX_OFFLOAD_TCP_TSO DEV_TX_OFFLOAD_TCP_TSO +#define RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO DEV_TX_OFFLOAD_VXLAN_TNL_TSO +#define RTE_ETH_TX_OFFLOAD_MULTI_SEGS DEV_TX_OFFLOAD_MULTI_SEGS +#define RTE_ETH_RX_OFFLOAD_IPV4_CKSUM DEV_RX_OFFLOAD_IPV4_CKSUM +#define RTE_ETH_RX_OFFLOAD_SCATTER DEV_RX_OFFLOAD_SCATTER +#define RTE_ETH_RX_OFFLOAD_TCP_LRO DEV_RX_OFFLOAD_TCP_LRO +#define RTE_ETH_MQ_RX_RSS ETH_MQ_RX_RSS +#define RTE_ETH_RX_OFFLOAD_TCP_CKSUM DEV_RX_OFFLOAD_TCP_CKSUM +#define RTE_ETH_RX_OFFLOAD_UDP_CKSUM DEV_RX_OFFLOAD_UDP_CKSUM +#define RTE_ETH_MQ_RX_NONE ETH_MQ_RX_NONE +#define RTE_ETH_LINK_FULL_DUPLEX ETH_LINK_FULL_DUPLEX +#define RTE_ETH_LINK_HALF_DUPLEX ETH_LINK_HALF_DUPLEX +#define RTE_ETH_VLAN_STRIP_OFFLOAD ETH_VLAN_STRIP_OFFLOAD +#define RTE_ETH_VLAN_FILTER_OFFLOAD ETH_VLAN_FILTER_OFFLOAD +#define RTE_ETH_VLAN_EXTEND_OFFLOAD ETH_VLAN_EXTEND_OFFLOAD +#define RTE_ETH_LINK_SPEED_200G ETH_LINK_SPEED_200G +#define RTE_ETH_LINK_SPEED_100G ETH_LINK_SPEED_100G +#define RTE_ETH_LINK_SPEED_56G ETH_LINK_SPEED_56G +#define RTE_ETH_LINK_SPEED_50G ETH_LINK_SPEED_50G +#define RTE_ETH_LINK_SPEED_40G ETH_LINK_SPEED_40G +#define RTE_ETH_LINK_SPEED_25G ETH_LINK_SPEED_25G +#define RTE_ETH_LINK_SPEED_20G ETH_LINK_SPEED_20G +#define RTE_ETH_LINK_SPEED_10G ETH_LINK_SPEED_10G +#define RTE_ETH_LINK_SPEED_5G ETH_LINK_SPEED_5G +#define RTE_ETH_LINK_SPEED_2_5G ETH_LINK_SPEED_2_5G +#define RTE_ETH_LINK_SPEED_1G ETH_LINK_SPEED_1G +#define RTE_ETH_RSS_IP ETH_RSS_IP +#define RTE_ETH_RSS_UDP ETH_RSS_UDP +#define RTE_ETH_RSS_TCP ETH_RSS_TCP +#endif + /* * fd.io coding-style-patch-verification: ON * diff --git a/src/plugins/dpdk/device/driver.c b/src/plugins/dpdk/device/driver.c new file mode 100644 index 00000000000..9c368dd9038 --- /dev/null +++ b/src/plugins/dpdk/device/driver.c @@ -0,0 +1,154 @@ +/* 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, + .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" }), + .interface_name_prefix = "VirtualFunctionEthernet", + .use_intel_phdr_cksum = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_ixgbe_vf", "Intel 82599 VF" }), + .interface_name_prefix = "VirtualFunctionEthernet", + .use_intel_phdr_cksum = 1, + .program_vlans = 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-4/5/6 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, + }, + { + .drivers = DPDK_DRIVERS ({ "net_vmxnet3", "VMware VMXNET3" }), + .interface_name_prefix = "GigabitEthernet", + .enable_rxq_int = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_virtio", "Red Hat Virtio" }), + .interface_name_prefix = "GigabitEthernet", + .n_rx_desc = 256, + .n_tx_desc = 256, + .mq_mode_none = 1, + .enable_rxq_int = 1, + }, + { + .drivers = DPDK_DRIVERS ({ "net_gve", "Google vNIC" }), + .interface_name_prefix = "VirtualFunctionEthernet", + } +}; + +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/flow.c b/src/plugins/dpdk/device/flow.c index a090ec0e930..635f6f37ebf 100644 --- a/src/plugins/dpdk/device/flow.c +++ b/src/plugins/dpdk/device/flow.c @@ -21,7 +21,7 @@ #include <vnet/ip/ip.h> #include <vnet/ethernet/ethernet.h> #include <vnet/ethernet/arp_packet.h> -#include <vnet/vxlan/vxlan.h> +#include <vxlan/vxlan.h> #include <dpdk/device/dpdk.h> #include <dpdk/device/dpdk_priv.h> #include <vppinfra/error.h> @@ -29,22 +29,30 @@ #define FLOW_IS_ETHERNET_CLASS(f) \ (f->type == VNET_FLOW_TYPE_ETHERNET) -#define FLOW_IS_IPV4_CLASS(f) \ - ((f->type == VNET_FLOW_TYPE_IP4) || \ - (f->type == VNET_FLOW_TYPE_IP4_N_TUPLE) || \ - (f->type == VNET_FLOW_TYPE_IP4_N_TUPLE_TAGGED) || \ - (f->type == VNET_FLOW_TYPE_IP4_VXLAN) || \ - (f->type == VNET_FLOW_TYPE_IP4_GTPC) || \ - (f->type == VNET_FLOW_TYPE_IP4_GTPU) || \ - (f->type == VNET_FLOW_TYPE_IP4_L2TPV3OIP) || \ - (f->type == VNET_FLOW_TYPE_IP4_IPSEC_ESP) || \ - (f->type == VNET_FLOW_TYPE_IP4_IPSEC_AH)) - -#define FLOW_IS_IPV6_CLASS(f) \ - ((f->type == VNET_FLOW_TYPE_IP6) || \ - (f->type == VNET_FLOW_TYPE_IP6_N_TUPLE) || \ - (f->type == VNET_FLOW_TYPE_IP6_N_TUPLE_TAGGED) || \ - (f->type == VNET_FLOW_TYPE_IP6_VXLAN)) +#define FLOW_IS_IPV4_CLASS(f) \ + ((f->type == VNET_FLOW_TYPE_IP4) || \ + (f->type == VNET_FLOW_TYPE_IP4_N_TUPLE) || \ + (f->type == VNET_FLOW_TYPE_IP4_N_TUPLE_TAGGED) || \ + (f->type == VNET_FLOW_TYPE_IP4_VXLAN) || \ + (f->type == VNET_FLOW_TYPE_IP4_GTPC) || \ + (f->type == VNET_FLOW_TYPE_IP4_GTPU) || \ + (f->type == VNET_FLOW_TYPE_IP4_L2TPV3OIP) || \ + (f->type == VNET_FLOW_TYPE_IP4_IPSEC_ESP) || \ + (f->type == VNET_FLOW_TYPE_IP4_IPSEC_AH) || \ + (f->type == VNET_FLOW_TYPE_IP4_IP4) || \ + (f->type == VNET_FLOW_TYPE_IP4_IP6) || \ + (f->type == VNET_FLOW_TYPE_IP4_IP4_N_TUPLE) || \ + (f->type == VNET_FLOW_TYPE_IP4_IP6_N_TUPLE)) + +#define FLOW_IS_IPV6_CLASS(f) \ + ((f->type == VNET_FLOW_TYPE_IP6) || \ + (f->type == VNET_FLOW_TYPE_IP6_N_TUPLE) || \ + (f->type == VNET_FLOW_TYPE_IP6_N_TUPLE_TAGGED) || \ + (f->type == VNET_FLOW_TYPE_IP6_VXLAN) || \ + (f->type == VNET_FLOW_TYPE_IP6_IP4) || \ + (f->type == VNET_FLOW_TYPE_IP6_IP6) || \ + (f->type == VNET_FLOW_TYPE_IP6_IP4_N_TUPLE) || \ + (f->type == VNET_FLOW_TYPE_IP6_IP6_N_TUPLE)) /* check if flow is VLAN sensitive */ #define FLOW_HAS_VLAN_TAG(f) \ @@ -70,6 +78,13 @@ (f->type == VNET_FLOW_TYPE_IP4_GTPC) || \ (f->type == VNET_FLOW_TYPE_IP4_GTPU)) +/* check if flow has a inner TCP/UDP header */ +#define FLOW_HAS_INNER_N_TUPLE(f) \ + ((f->type == VNET_FLOW_TYPE_IP4_IP4_N_TUPLE) || \ + (f->type == VNET_FLOW_TYPE_IP4_IP6_N_TUPLE) || \ + (f->type == VNET_FLOW_TYPE_IP6_IP4_N_TUPLE) || \ + (f->type == VNET_FLOW_TYPE_IP6_IP6_N_TUPLE)) + /* constant structs */ static const struct rte_flow_attr ingress = {.ingress = 1 }; @@ -103,6 +118,25 @@ dpdk_flow_convert_rss_types (u64 type, u64 * dpdk_rss_type) return; } +/** Maximum number of queue indices in struct rte_flow_action_rss. */ +#define ACTION_RSS_QUEUE_NUM 128 + +static inline void +dpdk_flow_convert_rss_queues (u32 queue_index, u32 queue_num, + struct rte_flow_action_rss *rss) +{ + u16 *queues = clib_mem_alloc (sizeof (*queues) * ACTION_RSS_QUEUE_NUM); + int i; + + for (i = 0; i < queue_num; i++) + queues[i] = queue_index++; + + rss->queue_num = queue_num; + rss->queue = queues; + + return; +} + static inline enum rte_eth_hash_function dpdk_flow_convert_rss_func (vnet_rss_function_t func) { @@ -134,14 +168,15 @@ static int dpdk_flow_add (dpdk_device_t * xd, vnet_flow_t * f, dpdk_flow_entry_t * fe) { struct rte_flow_item_eth eth[2] = { }; - struct rte_flow_item_ipv4 ip4[2] = { }; - struct rte_flow_item_ipv6 ip6[2] = { }; - struct rte_flow_item_udp udp[2] = { }; - struct rte_flow_item_tcp tcp[2] = { }; + struct rte_flow_item_ipv4 ip4[2] = {}, in_ip4[2] = {}; + struct rte_flow_item_ipv6 ip6[2] = {}, in_ip6[2] = {}; + struct rte_flow_item_udp udp[2] = {}, in_UDP[2] = {}; + struct rte_flow_item_tcp tcp[2] = {}, in_TCP[2] = {}; struct rte_flow_item_gtp gtp[2] = { }; struct rte_flow_item_l2tpv3oip l2tp[2] = { }; struct rte_flow_item_esp esp[2] = { }; struct rte_flow_item_ah ah[2] = { }; + struct rte_flow_item_raw generic[2] = {}; struct rte_flow_action_mark mark = { 0 }; struct rte_flow_action_queue queue = { 0 }; struct rte_flow_action_rss rss = { 0 }; @@ -165,6 +200,20 @@ dpdk_flow_add (dpdk_device_t * xd, vnet_flow_t * f, dpdk_flow_entry_t * fe) u8 protocol = IP_PROTOCOL_RESERVED; int rv = 0; + /* Handle generic flow first */ + if (f->type == VNET_FLOW_TYPE_GENERIC) + { + generic[0].pattern = f->generic.pattern.spec; + generic[1].pattern = f->generic.pattern.mask; + + vec_add2 (items, item, 1); + item->type = RTE_FLOW_ITEM_TYPE_RAW; + item->spec = generic; + item->mask = generic + 1; + + goto pattern_end; + } + enum { FLOW_UNKNOWN_CLASS, @@ -285,7 +334,8 @@ dpdk_flow_add (dpdk_device_t * xd, vnet_flow_t * f, dpdk_flow_entry_t * fe) if ((ip6_ptr->src_addr.mask.as_u64[0] == 0) && (ip6_ptr->src_addr.mask.as_u64[1] == 0) && - (!ip6_ptr->protocol.mask)) + (ip6_ptr->dst_addr.mask.as_u64[0] == 0) && + (ip6_ptr->dst_addr.mask.as_u64[1] == 0) && (!ip6_ptr->protocol.mask)) { item->spec = NULL; item->mask = NULL; @@ -437,13 +487,127 @@ dpdk_flow_add (dpdk_device_t * xd, vnet_flow_t * f, dpdk_flow_entry_t * fe) item->mask = raw + 1; } break; + case IP_PROTOCOL_IPV6: + item->type = RTE_FLOW_ITEM_TYPE_IPV6; +#define fill_inner_ip6_with_outer_ipv(OUTER_IP_VER) \ + if (f->type == VNET_FLOW_TYPE_IP##OUTER_IP_VER##_IP6 || \ + f->type == VNET_FLOW_TYPE_IP##OUTER_IP_VER##_IP6_N_TUPLE) \ + { \ + vnet_flow_ip##OUTER_IP_VER##_ip6_t *ptr = &f->ip##OUTER_IP_VER##_ip6; \ + if ((ptr->in_src_addr.mask.as_u64[0] == 0) && \ + (ptr->in_src_addr.mask.as_u64[1] == 0) && \ + (ptr->in_dst_addr.mask.as_u64[0] == 0) && \ + (ptr->in_dst_addr.mask.as_u64[1] == 0) && (!ptr->in_protocol.mask)) \ + { \ + item->spec = NULL; \ + item->mask = NULL; \ + } \ + else \ + { \ + clib_memcpy (in_ip6[0].hdr.src_addr, &ptr->in_src_addr.addr, \ + ARRAY_LEN (ptr->in_src_addr.addr.as_u8)); \ + clib_memcpy (in_ip6[1].hdr.src_addr, &ptr->in_src_addr.mask, \ + ARRAY_LEN (ptr->in_src_addr.mask.as_u8)); \ + clib_memcpy (in_ip6[0].hdr.dst_addr, &ptr->in_dst_addr.addr, \ + ARRAY_LEN (ptr->in_dst_addr.addr.as_u8)); \ + clib_memcpy (in_ip6[1].hdr.dst_addr, &ptr->in_dst_addr.mask, \ + ARRAY_LEN (ptr->in_dst_addr.mask.as_u8)); \ + item->spec = in_ip6; \ + item->mask = in_ip6 + 1; \ + } \ + } + fill_inner_ip6_with_outer_ipv (6) fill_inner_ip6_with_outer_ipv (4) +#undef fill_inner_ip6_with_outer_ipv + break; + case IP_PROTOCOL_IP_IN_IP: + item->type = RTE_FLOW_ITEM_TYPE_IPV4; + +#define fill_inner_ip4_with_outer_ipv(OUTER_IP_VER) \ + if (f->type == VNET_FLOW_TYPE_IP##OUTER_IP_VER##_IP4 || \ + f->type == VNET_FLOW_TYPE_IP##OUTER_IP_VER##_IP4_N_TUPLE) \ + { \ + vnet_flow_ip##OUTER_IP_VER##_ip4_t *ptr = &f->ip##OUTER_IP_VER##_ip4; \ + if ((!ptr->in_src_addr.mask.as_u32) && \ + (!ptr->in_dst_addr.mask.as_u32) && (!ptr->in_protocol.mask)) \ + { \ + item->spec = NULL; \ + item->mask = NULL; \ + } \ + else \ + { \ + in_ip4[0].hdr.src_addr = ptr->in_src_addr.addr.as_u32; \ + in_ip4[1].hdr.src_addr = ptr->in_src_addr.mask.as_u32; \ + in_ip4[0].hdr.dst_addr = ptr->in_dst_addr.addr.as_u32; \ + in_ip4[1].hdr.dst_addr = ptr->in_dst_addr.mask.as_u32; \ + item->spec = in_ip4; \ + item->mask = in_ip4 + 1; \ + } \ + } + fill_inner_ip4_with_outer_ipv (6) fill_inner_ip4_with_outer_ipv (4) +#undef fill_inner_ip4_with_outer_ipv + break; default: rv = VNET_FLOW_ERROR_NOT_SUPPORTED; goto done; } + if (FLOW_HAS_INNER_N_TUPLE (f)) + { + vec_add2 (items, item, 1); + +#define fill_inner_n_tuple_of(proto) \ + item->type = RTE_FLOW_ITEM_TYPE_##proto; \ + if ((ptr->in_src_port.mask == 0) && (ptr->in_dst_port.mask == 0)) \ + { \ + item->spec = NULL; \ + item->mask = NULL; \ + } \ + else \ + { \ + in_##proto[0].hdr.src_port = \ + clib_host_to_net_u16 (ptr->in_src_port.port); \ + in_##proto[1].hdr.src_port = \ + clib_host_to_net_u16 (ptr->in_src_port.mask); \ + in_##proto[0].hdr.dst_port = \ + clib_host_to_net_u16 (ptr->in_dst_port.port); \ + in_##proto[1].hdr.dst_port = \ + clib_host_to_net_u16 (ptr->in_dst_port.mask); \ + item->spec = in_##proto; \ + item->mask = in_##proto + 1; \ + } + +#define fill_inner_n_tuple(OUTER_IP_VER, INNER_IP_VER) \ + if (f->type == \ + VNET_FLOW_TYPE_IP##OUTER_IP_VER##_IP##INNER_IP_VER##_N_TUPLE) \ + { \ + vnet_flow_ip##OUTER_IP_VER##_ip##INNER_IP_VER##_n_tuple_t *ptr = \ + &f->ip##OUTER_IP_VER##_ip##INNER_IP_VER##_n_tuple; \ + switch (ptr->in_protocol.prot) \ + { \ + case IP_PROTOCOL_UDP: \ + fill_inner_n_tuple_of (UDP) break; \ + case IP_PROTOCOL_TCP: \ + fill_inner_n_tuple_of (TCP) break; \ + default: \ + break; \ + } \ + } + fill_inner_n_tuple (6, 4) fill_inner_n_tuple (4, 4) + fill_inner_n_tuple (6, 6) fill_inner_n_tuple (4, 6) +#undef fill_inner_n_tuple +#undef fill_inner_n_tuple_of + } + pattern_end: + if ((f->actions & VNET_FLOW_ACTION_RSS) && + (f->rss_types & (1ULL << VNET_FLOW_RSS_TYPES_ESP))) + { + + vec_add2 (items, item, 1); + item->type = RTE_FLOW_ITEM_TYPE_ESP; + } + vec_add2 (items, item, 1); item->type = RTE_FLOW_ITEM_TYPE_END; @@ -482,6 +646,10 @@ pattern_end: /* convert types to DPDK rss bitmask */ dpdk_flow_convert_rss_types (f->rss_types, &rss_type); + if (f->queue_num) + /* convert rss queues to array */ + dpdk_flow_convert_rss_queues (f->queue_index, f->queue_num, &rss); + rss.types = rss_type; if ((rss.func = dpdk_flow_convert_rss_func (f->rss_fun)) == RTE_ETH_HASH_FUNCTION_MAX) @@ -547,6 +715,7 @@ int dpdk_flow_ops_fn (vnet_main_t * vnm, vnet_flow_dev_op_t op, u32 dev_instance, u32 flow_index, uword * private_data) { + vlib_main_t *vm = vlib_get_main (); dpdk_main_t *dm = &dpdk_main; vnet_flow_t *flow = vnet_get_flow (flow_index); dpdk_device_t *xd = vec_elt_at_index (dm->devices, dev_instance); @@ -557,7 +726,7 @@ dpdk_flow_ops_fn (vnet_main_t * vnm, vnet_flow_dev_op_t op, u32 dev_instance, /* recycle old flow lookup entries only after the main loop counter increases - i.e. previously DMA'ed packets were handled */ if (vec_len (xd->parked_lookup_indexes) > 0 && - xd->parked_loop_count != dm->vlib_main->main_loop_count) + xd->parked_loop_count != vm->main_loop_count) { u32 *fl_index; @@ -580,7 +749,7 @@ dpdk_flow_ops_fn (vnet_main_t * vnm, vnet_flow_dev_op_t op, u32 dev_instance, fle = pool_elt_at_index (xd->flow_lookup_entries, fe->mark); clib_memset (fle, -1, sizeof (*fle)); vec_add1 (xd->parked_lookup_indexes, fe->mark); - xd->parked_loop_count = dm->vlib_main->main_loop_count; + xd->parked_loop_count = vm->main_loop_count; } clib_memset (fe, 0, sizeof (*fe)); @@ -644,6 +813,15 @@ dpdk_flow_ops_fn (vnet_main_t * vnm, vnet_flow_dev_op_t op, u32 dev_instance, case VNET_FLOW_TYPE_IP4_L2TPV3OIP: case VNET_FLOW_TYPE_IP4_IPSEC_ESP: case VNET_FLOW_TYPE_IP4_IPSEC_AH: + case VNET_FLOW_TYPE_IP4_IP4: + case VNET_FLOW_TYPE_IP4_IP4_N_TUPLE: + case VNET_FLOW_TYPE_IP4_IP6: + case VNET_FLOW_TYPE_IP4_IP6_N_TUPLE: + case VNET_FLOW_TYPE_IP6_IP4: + case VNET_FLOW_TYPE_IP6_IP4_N_TUPLE: + case VNET_FLOW_TYPE_IP6_IP6: + case VNET_FLOW_TYPE_IP6_IP6_N_TUPLE: + case VNET_FLOW_TYPE_GENERIC: if ((rv = dpdk_flow_add (xd, flow, fe))) goto done; break; diff --git a/src/plugins/dpdk/device/format.c b/src/plugins/dpdk/device/format.c index 24994aa9426..c4170c20329 100644 --- a/src/plugins/dpdk/device/format.c +++ b/src/plugins/dpdk/device/format.c @@ -17,9 +17,6 @@ #include <vppinfra/format.h> #include <assert.h> -#define __USE_GNU -#include <dlfcn.h> - #include <vnet/ethernet/ethernet.h> #include <vnet/ethernet/sfp.h> #include <dpdk/device/dpdk.h> @@ -49,18 +46,28 @@ #endif #define foreach_dpdk_pkt_rx_offload_flag \ - _ (PKT_RX_VLAN, "RX packet is a 802.1q VLAN packet") \ - _ (PKT_RX_RSS_HASH, "RX packet with RSS hash result") \ - _ (PKT_RX_FDIR, "RX packet with FDIR infos") \ - _ (PKT_RX_L4_CKSUM_BAD, "L4 cksum of RX pkt. is not OK") \ - _ (PKT_RX_IP_CKSUM_BAD, "IP cksum of RX pkt. is not OK") \ - _ (PKT_RX_OUTER_IP_CKSUM_BAD, "External IP header checksum error") \ - _ (PKT_RX_VLAN_STRIPPED, "RX packet VLAN tag stripped") \ - _ (PKT_RX_IP_CKSUM_GOOD, "IP cksum of RX pkt. is valid") \ - _ (PKT_RX_L4_CKSUM_GOOD, "L4 cksum of RX pkt. is valid") \ - _ (PKT_RX_IEEE1588_PTP, "RX IEEE1588 L2 Ethernet PT Packet") \ - _ (PKT_RX_IEEE1588_TMST, "RX IEEE1588 L2/L4 timestamped packet") \ - _ (PKT_RX_QINQ_STRIPPED, "RX packet QinQ tags stripped") + _ (RX_FDIR, "RX packet with FDIR infos") \ + _ (RX_FDIR_FLX, "RX packet with FDIR_FLX info") \ + _ (RX_FDIR_ID, "RX packet with FDIR_ID info") \ + _ (RX_IEEE1588_PTP, "RX IEEE1588 L2 Ethernet PT Packet") \ + _ (RX_IEEE1588_TMST, "RX IEEE1588 L2/L4 timestamped packet") \ + _ (RX_IP_CKSUM_BAD, "IP cksum of RX pkt. is not OK") \ + _ (RX_IP_CKSUM_GOOD, "IP cksum of RX pkt. is valid") \ + _ (RX_IP_CKSUM_NONE, "no IP cksum of RX pkt.") \ + _ (RX_L4_CKSUM_BAD, "L4 cksum of RX pkt. is not OK") \ + _ (RX_L4_CKSUM_GOOD, "L4 cksum of RX pkt. is valid") \ + _ (RX_L4_CKSUM_NONE, "no L4 cksum of RX pkt.") \ + _ (RX_LRO, "LRO packet") \ + _ (RX_OUTER_IP_CKSUM_BAD, "External IP header checksum error") \ + _ (RX_OUTER_L4_CKSUM_BAD, "External L4 header checksum error") \ + _ (RX_OUTER_L4_CKSUM_GOOD, "External L4 header checksum OK") \ + _ (RX_QINQ, "RX packet with QinQ tags") \ + _ (RX_QINQ_STRIPPED, "RX packet QinQ tags stripped") \ + _ (RX_RSS_HASH, "RX packet with RSS hash result") \ + _ (RX_SEC_OFFLOAD, "RX packet with security offload") \ + _ (RX_SEC_OFFLOAD_FAILED, "RX packet with security offload failed") \ + _ (RX_VLAN, "RX packet is a 802.1q VLAN packet") \ + _ (RX_VLAN_STRIPPED, "RX packet VLAN tag stripped") #define foreach_dpdk_pkt_type \ _ (L2, ETHER, "Ethernet packet") \ @@ -103,14 +110,32 @@ _ (INNER_L4, NONFRAG, "Inner non-fragmented IP packet") #define foreach_dpdk_pkt_tx_offload_flag \ - _ (PKT_TX_VLAN_PKT, "TX packet is a 802.1q VLAN packet") \ - _ (PKT_TX_TUNNEL_VXLAN, "TX packet is a VXLAN packet") \ - _ (PKT_TX_IP_CKSUM, "IP cksum of TX pkt. computed by NIC") \ - _ (PKT_TX_TCP_CKSUM, "TCP cksum of TX pkt. computed by NIC") \ - _ (PKT_TX_SCTP_CKSUM, "SCTP cksum of TX pkt. computed by NIC") \ - _ (PKT_TX_OUTER_IP_CKSUM, "Outer IP cksum of Tx pkt. computed by NIC") \ - _ (PKT_TX_TCP_SEG, "TSO of TX pkt. done by NIC") \ - _ (PKT_TX_IEEE1588_TMST, "TX IEEE1588 packet to timestamp") + _ (TX_IEEE1588_TMST, "TX IEEE1588 packet to timestamp") \ + _ (TX_IPV4, "TX IPV4") \ + _ (TX_IPV6, "TX IPV6") \ + _ (TX_IP_CKSUM, "IP cksum of TX pkt. computed by NIC") \ + _ (TX_MACSEC, "TX MACSEC") \ + _ (TX_OUTER_IPV4, "TX outer IPV4") \ + _ (TX_OUTER_IPV6, "TX outer IPV6") \ + _ (TX_OUTER_IP_CKSUM, "Outer IP cksum of Tx pkt. computed by NIC") \ + _ (TX_OUTER_UDP_CKSUM, "TX outer UDP cksum") \ + _ (TX_QINQ, "TX QINQ") \ + _ (TX_SCTP_CKSUM, "SCTP cksum of TX pkt. computed by NIC") \ + _ (TX_SEC_OFFLOAD, "TX SEC OFFLOAD") \ + _ (TX_TCP_CKSUM, "TCP cksum of TX pkt. computed by NIC") \ + _ (TX_TCP_SEG, "TSO of TX pkt. done by NIC") \ + _ (TX_TUNNEL_GENEVE, "TX tunnel GENEVE") \ + _ (TX_TUNNEL_GRE, "TX tunnel GRE") \ + _ (TX_TUNNEL_GTP, "TX tunnel GTP") \ + _ (TX_TUNNEL_IP, "TX tunnel IP") \ + _ (TX_TUNNEL_IPIP, "TX tunnel IPIP") \ + _ (TX_TUNNEL_MPLSINUDP, "TX tunnel MPLSinUDP") \ + _ (TX_TUNNEL_UDP, "TX tunnel UDP") \ + _ (TX_TUNNEL_VXLAN, "TX packet is a VXLAN packet") \ + _ (TX_TUNNEL_VXLAN_GPE, "TX tunnel VXLAN GPE") \ + _ (TX_UDP_CKSUM, "TX UDP cksum") \ + _ (TX_UDP_SEG, "TX UDP SEG") \ + _ (TX_VLAN, "TX packet is a 802.1q VLAN packet") #define foreach_dpdk_pkt_offload_flag \ foreach_dpdk_pkt_rx_offload_flag \ @@ -123,105 +148,10 @@ u8 * format_dpdk_device_name (u8 * s, va_list * args) { dpdk_main_t *dm = &dpdk_main; - char *devname_format; - char *device_name; 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"; - - switch (xd->port_type) - { - case VNET_DPDK_PORT_TYPE_ETH_1G: - device_name = "GigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_2_5G: - device_name = "Two_FiveGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_5G: - device_name = "FiveGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_10G: - device_name = "TenGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_20G: - device_name = "TwentyGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_25G: - device_name = "TwentyFiveGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_40G: - device_name = "FortyGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_50G: - device_name = "FiftyGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_56G: - device_name = "FiftySixGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_100G: - device_name = "HundredGigabitEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_SWITCH: - device_name = "EthernetSwitch"; - break; - - case VNET_DPDK_PORT_TYPE_ETH_VF: - device_name = "VirtualFunctionEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_AF_PACKET: - return format (s, "af_packet%d", xd->af_packet_instance_num); - - case VNET_DPDK_PORT_TYPE_VIRTIO_USER: - device_name = "VirtioUser"; - break; - - case VNET_DPDK_PORT_TYPE_VHOST_ETHER: - device_name = "VhostEthernet"; - break; - - case VNET_DPDK_PORT_TYPE_FAILSAFE: - device_name = "FailsafeEthernet"; - break; - - default: - case VNET_DPDK_PORT_TYPE_UNKNOWN: - device_name = "UnknownEthernet"; - break; - } - - 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 * @@ -243,126 +173,12 @@ 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 * @@ -378,10 +194,11 @@ format_dpdk_link_status (u8 * s, va_list * args) { u32 promisc = rte_eth_promiscuous_get (xd->port_id); - s = format (s, "%s duplex ", (l->link_duplex == ETH_LINK_FULL_DUPLEX) ? - "full" : "half"); - s = format (s, "mtu %d %s\n", hi->max_packet_bytes, promisc ? - " promisc" : ""); + s = format (s, "%s duplex ", + (l->link_duplex == RTE_ETH_LINK_FULL_DUPLEX) ? "full" : + "half"); + s = format (s, "max-frame-size %d %s\n", hi->max_frame_size, + promisc ? " promisc" : ""); } else s = format (s, "\n"); @@ -419,8 +236,6 @@ format_offload (u8 * s, va_list * va) uword i, l; l = ~0; - if (clib_mem_is_vec (id)) - l = vec_len (id); if (id) for (i = 0; id[i] != 0 && i < l; i++) @@ -523,15 +338,29 @@ format_dpdk_device_module_info (u8 * s, va_list * args) return s; } -static const char * -ptr2sname (void *p) +u8 * +format_dpdk_burst_fn (u8 *s, va_list *args) { - Dl_info info = { 0 }; + dpdk_device_t *xd = va_arg (*args, dpdk_device_t *); + vlib_rx_or_tx_t dir = va_arg (*args, vlib_rx_or_tx_t); + void *p; + clib_elf_symbol_t sym; - if (dladdr (p, &info) == 0) - return 0; +#if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0) +#define rte_eth_fp_ops rte_eth_devices +#endif + + p = (dir == VLIB_TX) ? rte_eth_fp_ops[xd->port_id].tx_pkt_burst : + rte_eth_fp_ops[xd->port_id].rx_pkt_burst; - return info.dli_sname; + if (clib_elf_symbol_by_address (pointer_to_uword (p), &sym)) + { + return format (s, "%s", clib_elf_symbol_name (&sym)); + } + else + { + return format (s, "(not available)"); + } } static u8 * @@ -549,16 +378,51 @@ format_switch_info (u8 * s, va_list * args) } u8 * +format_dpdk_rte_device (u8 *s, va_list *args) +{ + struct rte_device *d = va_arg (*args, struct rte_device *); + + if (!d) + return format (s, "not available"); + +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + s = + format (s, "name: %s, numa: %d", rte_dev_name (d), rte_dev_numa_node (d)); + + if (rte_dev_driver (d)) + s = format (s, ", driver: %s", rte_driver_name (rte_dev_driver (d))); + + if (rte_dev_bus (d)) + s = format (s, ", bus: %s", rte_bus_name (rte_dev_bus (d))); +#else + s = format (s, "name: %s, numa: %d", d->name, d->numa_node); + + if (d->driver) + s = format (s, ", driver: %s", d->driver->name); + + if (d->bus) + s = format (s, ", bus: %s", d->bus->name); +#endif + + return s; +} + +u8 * format_dpdk_device (u8 * s, va_list * args) { u32 dev_instance = va_arg (*args, u32); int verbose = va_arg (*args, int); dpdk_main_t *dm = &dpdk_main; + vlib_main_t *vm = vlib_get_main (); dpdk_device_t *xd = vec_elt_at_index (dm->devices, dev_instance); u32 indent = format_get_indent (s); - f64 now = vlib_time_now (dm->vlib_main); + f64 now = vlib_time_now (vm); struct rte_eth_dev_info di; struct rte_eth_burst_mode mode; + struct rte_pci_device *pci; + struct rte_eth_rss_conf rss_conf; + int vlan_off; + int retval; dpdk_update_counters (xd, now); dpdk_update_link_state (xd, now); @@ -569,126 +433,114 @@ format_dpdk_device (u8 * s, va_list * args) format_white_space, indent + 2, format_dpdk_link_status, xd); s = format (s, "%Uflags: %U\n", format_white_space, indent + 2, format_dpdk_device_flags, xd); +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + if (rte_dev_devargs (di.device) && rte_dev_devargs (di.device)->args) + s = format (s, "%UDevargs: %s\n", format_white_space, indent + 2, + rte_dev_devargs (di.device)->args); +#else if (di.device->devargs && di.device->devargs->args) s = format (s, "%UDevargs: %s\n", format_white_space, indent + 2, di.device->devargs->args); - s = format (s, "%Urx: queues %d (max %d), desc %d " +#endif + s = format (s, + "%Urx: queues %d (max %d), desc %d " "(min %d max %d align %d)\n", - format_white_space, indent + 2, xd->rx_q_used, di.max_rx_queues, - xd->nb_rx_desc, di.rx_desc_lim.nb_min, di.rx_desc_lim.nb_max, - di.rx_desc_lim.nb_align); - s = format (s, "%Utx: queues %d (max %d), desc %d " + format_white_space, indent + 2, xd->conf.n_rx_queues, + di.max_rx_queues, xd->conf.n_rx_desc, di.rx_desc_lim.nb_min, + di.rx_desc_lim.nb_max, di.rx_desc_lim.nb_align); + s = format (s, + "%Utx: queues %d (max %d), desc %d " "(min %d max %d align %d)\n", - format_white_space, indent + 2, xd->tx_q_used, di.max_tx_queues, - xd->nb_tx_desc, di.tx_desc_lim.nb_min, di.tx_desc_lim.nb_max, - di.tx_desc_lim.nb_align); + format_white_space, indent + 2, xd->conf.n_tx_queues, + di.max_tx_queues, xd->conf.n_tx_desc, di.tx_desc_lim.nb_min, + di.tx_desc_lim.nb_max, di.tx_desc_lim.nb_align); - if (xd->flags & DPDK_DEVICE_FLAG_PMD) - { - struct rte_pci_device *pci; - struct rte_eth_rss_conf rss_conf; - int vlan_off; - int retval; + rss_conf.rss_key = 0; + rss_conf.rss_hf = 0; + retval = rte_eth_dev_rss_hash_conf_get (xd->port_id, &rss_conf); + if (retval < 0) + clib_warning ("rte_eth_dev_rss_hash_conf_get returned %d", retval); - rss_conf.rss_key = 0; - rss_conf.rss_hf = 0; - retval = rte_eth_dev_rss_hash_conf_get (xd->port_id, &rss_conf); - if (retval < 0) - clib_warning ("rte_eth_dev_rss_hash_conf_get returned %d", retval); + pci = dpdk_get_pci_device (&di); - pci = dpdk_get_pci_device (&di); + if (pci) + { + u8 *s2; + if (xd->cpu_socket > -1) + s2 = format (0, "%d", xd->cpu_socket); + else + s2 = format (0, "unknown"); + s = format (s, + "%Upci: device %04x:%04x subsystem %04x:%04x " + "address %04x:%02x:%02x.%02x numa %v\n", + format_white_space, indent + 2, pci->id.vendor_id, + pci->id.device_id, pci->id.subsystem_vendor_id, + pci->id.subsystem_device_id, pci->addr.domain, pci->addr.bus, + pci->addr.devid, pci->addr.function, s2); + vec_free (s2); + } - if (pci) - { - u8 *s2; - if (xd->cpu_socket > -1) - s2 = format (0, "%d", xd->cpu_socket); - else - s2 = format (0, "unknown"); - s = format (s, "%Upci: device %04x:%04x subsystem %04x:%04x " - "address %04x:%02x:%02x.%02x numa %v\n", - format_white_space, indent + 2, pci->id.vendor_id, - pci->id.device_id, pci->id.subsystem_vendor_id, - pci->id.subsystem_device_id, pci->addr.domain, - pci->addr.bus, pci->addr.devid, pci->addr.function, s2); - vec_free (s2); - } + if (di.switch_info.domain_id != RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) + { + s = format (s, "%Uswitch info: %U\n", format_white_space, indent + 2, + format_switch_info, &di.switch_info); + } - if (di.switch_info.domain_id != RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) - { - s = - format (s, "%Uswitch info: %U\n", format_white_space, indent + 2, - format_switch_info, &di.switch_info); - } + if (1 < verbose) + { + s = format (s, "%Umodule: %U\n", format_white_space, indent + 2, + format_dpdk_device_module_info, xd); + } - if (1 < verbose) - { - s = format (s, "%Umodule: %U\n", format_white_space, indent + 2, - format_dpdk_device_module_info, xd); - } + s = format (s, "%Umax rx packet len: %d\n", format_white_space, indent + 2, + di.max_rx_pktlen); + s = format (s, "%Upromiscuous: unicast %s all-multicast %s\n", + format_white_space, indent + 2, + rte_eth_promiscuous_get (xd->port_id) ? "on" : "off", + rte_eth_allmulticast_get (xd->port_id) ? "on" : "off"); + vlan_off = rte_eth_dev_get_vlan_offload (xd->port_id); + s = format (s, "%Uvlan offload: strip %s filter %s qinq %s\n", + format_white_space, indent + 2, + vlan_off & RTE_ETH_VLAN_STRIP_OFFLOAD ? "on" : "off", + vlan_off & RTE_ETH_VLAN_FILTER_OFFLOAD ? "on" : "off", + vlan_off & RTE_ETH_VLAN_EXTEND_OFFLOAD ? "on" : "off"); + s = format (s, "%Urx offload avail: %U\n", format_white_space, indent + 2, + format_dpdk_rx_offload_caps, di.rx_offload_capa); + s = format (s, "%Urx offload active: %U\n", format_white_space, indent + 2, + format_dpdk_rx_offload_caps, xd->enabled_rx_off); + s = format (s, "%Utx offload avail: %U\n", format_white_space, indent + 2, + format_dpdk_tx_offload_caps, di.tx_offload_capa); + s = format (s, "%Utx offload active: %U\n", format_white_space, indent + 2, + format_dpdk_tx_offload_caps, xd->enabled_tx_off); + s = format (s, + "%Urss avail: %U\n" + "%Urss active: %U\n", + format_white_space, indent + 2, format_dpdk_rss_hf_name, + di.flow_type_rss_offloads, format_white_space, indent + 2, + format_dpdk_rss_hf_name, rss_conf.rss_hf); + + if (rte_eth_tx_burst_mode_get (xd->port_id, 0, &mode) == 0) + { + s = format (s, "%Utx burst mode: %s%s\n", format_white_space, indent + 2, + mode.info, + mode.flags & RTE_ETH_BURST_FLAG_PER_QUEUE ? " (per queue)" : + ""); + } - s = format (s, "%Umax rx packet len: %d\n", format_white_space, - indent + 2, di.max_rx_pktlen); - s = format (s, "%Upromiscuous: unicast %s all-multicast %s\n", - format_white_space, indent + 2, - rte_eth_promiscuous_get (xd->port_id) ? "on" : "off", - rte_eth_allmulticast_get (xd->port_id) ? "on" : "off"); - vlan_off = rte_eth_dev_get_vlan_offload (xd->port_id); - s = format (s, "%Uvlan offload: strip %s filter %s qinq %s\n", - format_white_space, indent + 2, - vlan_off & ETH_VLAN_STRIP_OFFLOAD ? "on" : "off", - vlan_off & ETH_VLAN_FILTER_OFFLOAD ? "on" : "off", - vlan_off & ETH_VLAN_EXTEND_OFFLOAD ? "on" : "off"); - s = format (s, "%Urx offload avail: %U\n", - format_white_space, indent + 2, - format_dpdk_rx_offload_caps, di.rx_offload_capa); - s = format (s, "%Urx offload active: %U\n", - format_white_space, indent + 2, - format_dpdk_rx_offload_caps, xd->port_conf.rxmode.offloads); - s = format (s, "%Utx offload avail: %U\n", - format_white_space, indent + 2, - format_dpdk_tx_offload_caps, di.tx_offload_capa); - s = format (s, "%Utx offload active: %U\n", - format_white_space, indent + 2, - format_dpdk_tx_offload_caps, xd->port_conf.txmode.offloads); - s = format (s, "%Urss avail: %U\n" - "%Urss active: %U\n", - format_white_space, indent + 2, - format_dpdk_rss_hf_name, di.flow_type_rss_offloads, - format_white_space, indent + 2, - format_dpdk_rss_hf_name, rss_conf.rss_hf); - - if (rte_eth_tx_burst_mode_get (xd->port_id, 0, &mode) == 0) - { - s = format (s, "%Utx burst mode: %s%s\n", - format_white_space, indent + 2, - mode.info, - mode.flags & RTE_ETH_BURST_FLAG_PER_QUEUE ? - " (per queue)" : ""); - } - else - { - s = format (s, "%Utx burst function: %s\n", - format_white_space, indent + 2, - ptr2sname (rte_eth_devices[xd->port_id].tx_pkt_burst)); - } + s = format (s, "%Utx burst function: %U\n", format_white_space, indent + 2, + format_dpdk_burst_fn, xd, VLIB_TX); - if (rte_eth_rx_burst_mode_get (xd->port_id, 0, &mode) == 0) - { - s = format (s, "%Urx burst mode: %s%s\n", - format_white_space, indent + 2, - mode.info, - mode.flags & RTE_ETH_BURST_FLAG_PER_QUEUE ? - " (per queue)" : ""); - } - else - { - s = format (s, "%Urx burst function: %s\n", - format_white_space, indent + 2, - ptr2sname (rte_eth_devices[xd->port_id].rx_pkt_burst)); - } + if (rte_eth_rx_burst_mode_get (xd->port_id, 0, &mode) == 0) + { + s = format (s, "%Urx burst mode: %s%s\n", format_white_space, indent + 2, + mode.info, + mode.flags & RTE_ETH_BURST_FLAG_PER_QUEUE ? " (per queue)" : + ""); } + s = format (s, "%Urx burst function: %U\n", format_white_space, indent + 2, + format_dpdk_burst_fn, xd, VLIB_RX); + /* $$$ MIB counters */ { #define _(N, V) \ @@ -713,7 +565,6 @@ format_dpdk_device (u8 * s, va_list * args) if (ret >= 0 && ret <= len) { - /* *INDENT-OFF* */ vec_foreach_index(i, xd->xstats) { xstat = vec_elt_at_index(xd->xstats, i); @@ -725,7 +576,6 @@ format_dpdk_device (u8 * s, va_list * args) xstat->value); } } - /* *INDENT-ON* */ vec_free (xstat_names); } @@ -756,14 +606,12 @@ format_dpdk_tx_trace (u8 * s, va_list * va) dpdk_main_t *dm = &dpdk_main; dpdk_device_t *xd = vec_elt_at_index (dm->devices, t->device_index); u32 indent = format_get_indent (s); - vnet_sw_interface_t *sw = vnet_get_sw_interface (vnm, xd->sw_if_index); - s = format (s, "%U tx queue %d", - format_vnet_sw_interface_name, vnm, sw, t->queue_index); + s = format (s, "%U tx queue %d", format_vnet_sw_if_index_name, vnm, + xd->sw_if_index, t->queue_index); - s = format (s, "\n%Ubuffer 0x%x: %U", - format_white_space, indent, - t->buffer_index, format_vnet_buffer, &t->buffer); + s = format (s, "\n%Ubuffer 0x%x: %U", format_white_space, indent, + t->buffer_index, format_vnet_buffer_no_chain, &t->buffer); s = format (s, "\n%U%U", format_white_space, indent, @@ -787,14 +635,12 @@ format_dpdk_rx_trace (u8 * s, va_list * va) dpdk_device_t *xd = vec_elt_at_index (dm->devices, t->device_index); format_function_t *f; u32 indent = format_get_indent (s); - vnet_sw_interface_t *sw = vnet_get_sw_interface (vnm, xd->sw_if_index); - s = format (s, "%U rx queue %d", - format_vnet_sw_interface_name, vnm, sw, t->queue_index); + s = format (s, "%U rx queue %d", format_vnet_sw_if_index_name, vnm, + xd->sw_if_index, t->queue_index); - s = format (s, "\n%Ubuffer 0x%x: %U", - format_white_space, indent, - t->buffer_index, format_vnet_buffer, &t->buffer); + s = format (s, "\n%Ubuffer 0x%x: %U", format_white_space, indent, + t->buffer_index, format_vnet_buffer_no_chain, &t->buffer); s = format (s, "\n%U%U", format_white_space, indent, @@ -855,11 +701,11 @@ format_dpdk_pkt_offload_flags (u8 * s, va_list * va) s = format (s, "Packet Offload Flags"); -#define _(F, S) \ - if (*ol_flags & F) \ - { \ - s = format (s, "\n%U%s (0x%04x) %s", \ - format_white_space, indent, #F, F, S); \ +#define _(F, S) \ + if ((*ol_flags & RTE_MBUF_F_##F) == RTE_MBUF_F_##F) \ + { \ + s = format (s, "\n%U%s (0x%04x) %s", format_white_space, indent, \ + "PKT_" #F, RTE_MBUF_F_##F, S); \ } foreach_dpdk_pkt_offload_flag @@ -887,7 +733,7 @@ u8 * format_dpdk_rte_mbuf_tso (u8 *s, va_list *va) { struct rte_mbuf *mb = va_arg (*va, struct rte_mbuf *); - if (mb->ol_flags & PKT_TX_TCP_SEG) + if (mb->ol_flags & RTE_MBUF_F_TX_TCP_SEG) { s = format (s, "l4_len %u tso_segsz %u", mb->l4_len, mb->tso_segsz); } @@ -940,8 +786,9 @@ format_dpdk_rte_mbuf (u8 * s, va_list * va) s = format (s, "\n%U%U", format_white_space, indent, format_dpdk_pkt_offload_flags, &mb->ol_flags); - if ((mb->ol_flags & PKT_RX_VLAN) && - ((mb->ol_flags & (PKT_RX_VLAN_STRIPPED | PKT_RX_QINQ_STRIPPED)) == 0)) + if ((mb->ol_flags & RTE_MBUF_F_RX_VLAN) && + ((mb->ol_flags & + (RTE_MBUF_F_RX_VLAN_STRIPPED | RTE_MBUF_F_RX_QINQ_STRIPPED)) == 0)) { ethernet_vlan_header_tv_t *vlan_hdr = ((ethernet_vlan_header_tv_t *) & (eth_hdr->type)); diff --git a/src/plugins/dpdk/device/init.c b/src/plugins/dpdk/device/init.c index aebbb64dd76..827f8801ca6 100644 --- a/src/plugins/dpdk/device/init.c +++ b/src/plugins/dpdk/device/init.c @@ -21,14 +21,16 @@ #include <vlib/unix/unix.h> #include <vlib/log.h> +#include <vnet/vnet.h> #include <vnet/ethernet/ethernet.h> #include <vnet/interface/rx_queue_funcs.h> +#include <vnet/interface/tx_queue_funcs.h> #include <dpdk/buffer.h> #include <dpdk/device/dpdk.h> #include <dpdk/cryptodev/cryptodev.h> #include <vlib/pci/pci.h> #include <vlib/vmbus/vmbus.h> - +#include <vlib/stats/stats.h> #include <rte_ring.h> #include <rte_vect.h> @@ -43,71 +45,69 @@ #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 */ +/* dev_info.speed_capa -> interface name mapppings */ +const struct +{ + u32 link_speed; + const char *pfx; +} if_name_prefixes[] = { + /* sorted, higher speed first */ + { RTE_ETH_LINK_SPEED_200G, "TwoHundredGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_100G, "HundredGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_56G, "FiftySixGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_50G, "FiftyGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_40G, "FortyGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_25G, "TwentyFiveGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_20G, "TwentyGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_10G, "TenGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_5G, "FiveGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_2_5G, "TwoDotFiveGigabitEthernet" }, + { RTE_ETH_LINK_SPEED_1G, "GigabitEthernet" }, +}; -static dpdk_port_type_t -port_type_from_speed_capa (struct rte_eth_dev_info *dev_info) +static clib_error_t * +dpdk_set_max_frame_size (vnet_main_t *vnm, vnet_hw_interface_t *hi, + u32 frame_size) { + dpdk_main_t *dm = &dpdk_main; + dpdk_device_t *xd = vec_elt_at_index (dm->devices, hi->dev_instance); + int rv; + u32 mtu; - 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; -} + mtu = frame_size - xd->driver_frame_overhead; -static dpdk_port_type_t -port_type_from_link_speed (u32 link_speed) -{ - switch (link_speed) + rv = rte_eth_dev_set_mtu (xd->port_id, mtu); + + if (rv < 0) { - case ETH_SPEED_NUM_1G: - return VNET_DPDK_PORT_TYPE_ETH_1G; - case ETH_SPEED_NUM_2_5G: - return VNET_DPDK_PORT_TYPE_ETH_2_5G; - case ETH_SPEED_NUM_5G: - return VNET_DPDK_PORT_TYPE_ETH_5G; - case ETH_SPEED_NUM_10G: - return VNET_DPDK_PORT_TYPE_ETH_10G; - case ETH_SPEED_NUM_20G: - return VNET_DPDK_PORT_TYPE_ETH_20G; - case ETH_SPEED_NUM_25G: - return VNET_DPDK_PORT_TYPE_ETH_25G; - case ETH_SPEED_NUM_40G: - return VNET_DPDK_PORT_TYPE_ETH_40G; - case ETH_SPEED_NUM_50G: - return VNET_DPDK_PORT_TYPE_ETH_50G; - case ETH_SPEED_NUM_56G: - return VNET_DPDK_PORT_TYPE_ETH_56G; - case ETH_SPEED_NUM_100G: - return VNET_DPDK_PORT_TYPE_ETH_100G; - default: - return VNET_DPDK_PORT_TYPE_UNKNOWN; + dpdk_log_err ("[%u] rte_eth_dev_set_mtu failed (mtu %u, rv %d)", + xd->port_id, mtu, rv); + switch (rv) + { + case -ENOTSUP: + return vnet_error (VNET_ERR_UNSUPPORTED, + "dpdk driver doesn't support MTU change"); + case -EBUSY: + return vnet_error (VNET_ERR_BUSY, "port is running"); + case -EINVAL: + return vnet_error (VNET_ERR_INVALID_VALUE, "invalid MTU"); + default: + return vnet_error (VNET_ERR_BUG, + "unexpected return value %d returned from " + "rte_eth_dev_set_mtu(...)", + rv); + } } + else + dpdk_log_debug ("[%u] max_frame_size set to %u by setting MTU to %u", + xd->port_id, frame_size, mtu); + + return 0; } static u32 @@ -121,15 +121,11 @@ dpdk_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi, u32 flags) { case ETHERNET_INTERFACE_FLAG_DEFAULT_L3: /* set to L3/non-promisc mode */ - xd->flags &= ~DPDK_DEVICE_FLAG_PROMISC; + dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_PROMISC, 0); break; case ETHERNET_INTERFACE_FLAG_ACCEPT_ALL: - xd->flags |= DPDK_DEVICE_FLAG_PROMISC; + dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_PROMISC, 1); break; - case ETHERNET_INTERFACE_FLAG_MTU: - xd->port_conf.rxmode.max_rx_pkt_len = hi->max_packet_bytes; - dpdk_device_setup (xd); - return 0; default: return ~0; } @@ -145,12 +141,6 @@ dpdk_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi, u32 flags) return old; } -static int -dpdk_port_crc_strip_enabled (dpdk_device_t * xd) -{ - return !(xd->port_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC); -} - /* The function check_l3cache helps check if Level 3 cache exists or not on current CPUs return value 1: exist. return value 0: not exist. @@ -192,701 +182,433 @@ check_l3cache () return 0; } -static void -dpdk_enable_l4_csum_offload (dpdk_device_t * xd) +static dpdk_device_config_t * +dpdk_find_startup_config (struct rte_eth_dev_info *di) { - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM; - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM; - xd->flags |= DPDK_DEVICE_FLAG_TX_OFFLOAD | - DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM; + dpdk_main_t *dm = &dpdk_main; + struct rte_pci_device *pci_dev; + vlib_pci_addr_t pci_addr; +#ifdef __linux__ + struct rte_vmbus_device *vmbus_dev; + vlib_vmbus_addr_t vmbus_addr; +#endif /* __linux__ */ + 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); + } + +#ifdef __linux__ + if ((vmbus_dev = dpdk_get_vmbus_device (di))) + { + unformat_input_t input_vmbus; +#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) + const char *dev_name = rte_dev_name (di->device); +#else + const char *dev_name = di->device->name; +#endif + unformat_init_string (&input_vmbus, dev_name, strlen (dev_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); + } +#endif /* __linux__ */ + + if (p) + return pool_elt_at_index (dm->conf->dev_confs, p[0]); + return &dm->conf->default_devconf; +} + +/* + * Initialise or refresh the xstats counters for a device + */ +void +dpdk_counters_xstats_init (dpdk_device_t *xd) +{ + int len, ret, i; + struct rte_eth_xstat_name *xstats_names = 0; + char *name; + dpdk_driver_t *dr = xd->driver; + + /* Only support xstats for supported drivers */ + if (!dr) + return; + + len = rte_eth_xstats_get_names (xd->port_id, 0, 0); + if (len < 0) + { + dpdk_log_err ("[%u] rte_eth_xstats_get_names failed: %d", xd->port_id, + len); + return; + } + /* Counters for this driver is already initialised */ + if (vec_len (dr->xstats_counters) == len) + { + vec_foreach_index (i, dr->xstats_counters) + { + vlib_validate_simple_counter (&dr->xstats_counters[i], + xd->sw_if_index); + vlib_zero_simple_counter (&dr->xstats_counters[i], xd->sw_if_index); + } + return; + } + + /* Same driver, different interface, different length of counter array. */ + ASSERT (vec_len (dr->xstats_counters) == 0); + + vec_validate (xstats_names, len - 1); + + ret = rte_eth_xstats_get_names (xd->port_id, xstats_names, len); + if (ret >= 0 && ret <= len) + { + vec_validate (dr->xstats_counters, len - 1); + vec_foreach_index (i, xstats_names) + { + name = (char *) format (0, "/if/%s/%s%c", dr->drivers->name, + xstats_names[i].name, 0); + + /* There is a bug in the ENA driver where the xstats names are not + * unique. */ + if (vlib_stats_find_entry_index (name) != STAT_SEGMENT_INDEX_INVALID) + { + vec_free (name); + name = (char *) format (0, "/if/%s/%s_%d%c", dr->drivers->name, + xstats_names[i].name, i, 0); + } + + dr->xstats_counters[i].name = name; + dr->xstats_counters[i].stat_segment_name = name; + dr->xstats_counters[i].counters = 0; + vlib_validate_simple_counter (&dr->xstats_counters[i], + xd->sw_if_index); + vlib_zero_simple_counter (&dr->xstats_counters[i], xd->sw_if_index); + vec_free (name); + } + } + vec_free (xstats_names); } static clib_error_t * dpdk_lib_init (dpdk_main_t * dm) { - u32 nports; - u32 mtu, max_rx_frame; - int i; - clib_error_t *error; - vlib_main_t *vm = vlib_get_main (); + vnet_main_t *vnm = vnet_get_main (); + u16 port_id; 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; - u8 af_packet_instance_num = 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); - - if (dm->conf->enable_tcp_udp_checksum) - dm->buffer_flags_template &= ~(VNET_BUFFER_F_L4_CHECKSUM_CORRECT - | VNET_BUFFER_F_L4_CHECKSUM_COMPUTED); + char *if_num_fmt; /* vlib_buffer_t template */ vec_validate_aligned (dm->per_thread_data, tm->n_vlib_mains - 1, CLIB_CACHE_LINE_BYTES); - for (i = 0; i < tm->n_vlib_mains; i++) + for (int i = 0; i < tm->n_vlib_mains; i++) { dpdk_per_thread_data_t *ptd = vec_elt_at_index (dm->per_thread_data, i); clib_memset (&ptd->buffer_template, 0, sizeof (vlib_buffer_t)); - ptd->buffer_template.flags = dm->buffer_flags_template; vnet_buffer (&ptd->buffer_template)->sw_if_index[VLIB_TX] = (u32) ~ 0; } - /* *INDENT-OFF* */ - RTE_ETH_FOREACH_DEV(i) + 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; + dm->default_port_conf.n_rx_queues = 1; + dm->default_port_conf.n_tx_queues = tm->n_vlib_mains; + dm->default_port_conf.rss_hf = + RTE_ETH_RSS_IP | RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP; + dm->default_port_conf.max_lro_pkt_size = DPDK_MAX_LRO_SIZE_DEFAULT; + + if ((clib_mem_get_default_hugepage_size () == 2 << 20) && + check_l3cache () == 0) + dm->default_port_conf.n_rx_desc = dm->default_port_conf.n_tx_desc = 512; + + RTE_ETH_FOREACH_DEV (port_id) { u8 addr[6]; - int vlan_off; - struct rte_eth_dev_info dev_info; - struct rte_pci_device *pci_dev; - struct rte_vmbus_device *vmbus_dev; - dpdk_portid_t next_port_id; + int rv, q; + struct rte_eth_dev_info di; dpdk_device_config_t *devconf = 0; - vlib_pci_addr_t pci_addr; - vlib_vmbus_addr_t vmbus_addr; - uword *p = 0; + vnet_eth_interface_registration_t eir = {}; + dpdk_driver_t *dr; + i8 numa_node; - if (!rte_eth_dev_is_valid_port(i)) + if (!rte_eth_dev_is_valid_port (port_id)) continue; - rte_eth_dev_info_get (i, &dev_info); - - if (dev_info.device == 0) + if ((rv = rte_eth_dev_info_get (port_id, &di)) != 0) { - dpdk_log_notice ("DPDK bug: missing device info. Skipping %s device", - dev_info.driver_name); + dpdk_log_warn ("[%u] failed to get device info. skipping device.", + port_id); continue; } - pci_dev = dpdk_get_pci_device (&dev_info); - - 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 (&dev_info); + 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_vector (&input_vmbus, (u8 *) dev_info.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); - } + dpdk_log_notice ("[%d] Device %s blacklisted. Skipping...", port_id, + di.driver_name); + continue; } - if (p) + vec_add2_aligned (dm->devices, xd, 1, CLIB_CACHE_LINE_BYTES); + 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)); + + /* find driver datea for this PMD */ + if ((dr = dpdk_driver_find (di.driver_name, &xd->if_desc))) { - devconf = pool_elt_at_index (dm->conf->dev_confs, p[0]); - /* If device is blacklisted, we should skip it */ - if (devconf->is_blacklisted) - { - continue; - } + 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; + xd->conf.enable_rxq_int = dr->enable_rxq_int; + 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 - devconf = &dm->conf->default_devconf; + dpdk_log_warn ("[%u] unknown driver '%s'", port_id, di.driver_name); - /* Create vnet interface */ - vec_add2_aligned (dm->devices, xd, 1, CLIB_CACHE_LINE_BYTES); - xd->nb_rx_desc = DPDK_NB_RX_DESC_DEFAULT; - xd->nb_tx_desc = DPDK_NB_TX_DESC_DEFAULT; - xd->cpu_socket = (i8) rte_eth_dev_socket_id (i); - if (p) + if (devconf->name) { xd->name = devconf->name; } - - /* Handle representor devices that share the same PCI ID */ - if (dev_info.switch_info.domain_id != RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) - { - if (dev_info.switch_info.port_id != (uint16_t)-1) - xd->interface_name_suffix = format (0, "%d", dev_info.switch_info.port_id); - } - /* Handle interface naming for devices with multiple ports sharing same PCI ID */ - else if (pci_dev && - ((next_port_id = rte_eth_find_next (i + 1)) != RTE_MAX_ETHPORTS)) + else { - struct rte_eth_dev_info di = { 0 }; - struct rte_pci_device *next_pci_dev; - rte_eth_dev_info_get (next_port_id, &di); - next_pci_dev = di.device ? RTE_DEV_TO_PCI (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 = i; - } - 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", i - 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; - - clib_memcpy (&xd->tx_conf, &dev_info.default_txconf, - sizeof (struct rte_eth_txconf)); - if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_IPV4_CKSUM) - { - xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_IPV4_CKSUM; - xd->flags |= DPDK_DEVICE_FLAG_RX_IP4_CKSUM; - } + 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); - if (dm->conf->enable_tcp_udp_checksum) - { - if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_UDP_CKSUM) - xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_UDP_CKSUM; - if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_CKSUM) - xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_TCP_CKSUM; + /* 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); } - if (dm->conf->no_multi_seg) - { - xd->port_conf.txmode.offloads &= ~DEV_TX_OFFLOAD_MULTI_SEGS; - xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME; - xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_SCATTER; - } - else + /* number of RX and TX queues */ + if (devconf->num_tx_queues > 0) { - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS; - xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME; - xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER; - xd->flags |= DPDK_DEVICE_FLAG_MAYBE_MULTISEG; + if (di.max_tx_queues < devconf->num_tx_queues) + dpdk_log_warn ("[%u] Configured number of TX queues (%u) is " + "bigger than maximum supported (%u)", + port_id, devconf->num_tx_queues, di.max_tx_queues); + xd->conf.n_tx_queues = devconf->num_tx_queues; } - xd->tx_q_used = clib_min (dev_info.max_tx_queues, tm->n_vlib_mains); - - if (devconf->num_tx_queues > 0 - && devconf->num_tx_queues < xd->tx_q_used) - xd->tx_q_used = clib_min (xd->tx_q_used, devconf->num_tx_queues); + xd->conf.n_tx_queues = clib_min (di.max_tx_queues, xd->conf.n_tx_queues); - if (devconf->num_rx_queues > 1 - && dev_info.max_rx_queues >= devconf->num_rx_queues) + if (devconf->num_rx_queues > 1 && + di.max_rx_queues >= devconf->num_rx_queues) { - xd->rx_q_used = devconf->num_rx_queues; - xd->port_conf.rxmode.mq_mode = ETH_MQ_RX_RSS; - if (devconf->rss_fn == 0) - xd->port_conf.rx_adv_conf.rss_conf.rss_hf = - ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP; - else + xd->conf.n_rx_queues = devconf->num_rx_queues; + if (devconf->rss_fn) { u64 unsupported_bits; - xd->port_conf.rx_adv_conf.rss_conf.rss_hf = devconf->rss_fn; - unsupported_bits = xd->port_conf.rx_adv_conf.rss_conf.rss_hf; - unsupported_bits &= ~dev_info.flow_type_rss_offloads; + xd->conf.rss_hf = devconf->rss_fn; + unsupported_bits = xd->conf.rss_hf; + unsupported_bits &= ~di.flow_type_rss_offloads; if (unsupported_bits) dpdk_log_warn ("Unsupported RSS hash functions: %U", format_dpdk_rss_hf_name, unsupported_bits); } - xd->port_conf.rx_adv_conf.rss_conf.rss_hf &= - dev_info.flow_type_rss_offloads; + xd->conf.rss_hf &= di.flow_type_rss_offloads; + dpdk_log_debug ("[%u] rss_hf: %U", port_id, format_dpdk_rss_hf_name, + xd->conf.rss_hf); } - else - xd->rx_q_used = 1; - vec_validate_aligned (xd->rx_queues, xd->rx_q_used - 1, +#ifndef RTE_VLAN_HLEN +#define RTE_VLAN_HLEN 4 +#endif + xd->driver_frame_overhead = + RTE_ETHER_HDR_LEN + 2 * RTE_VLAN_HLEN + RTE_ETHER_CRC_LEN; +#if RTE_VERSION >= RTE_VERSION_NUM(21, 11, 0, 0) + q = di.max_rx_pktlen - di.max_mtu; + + /* attempt to protect from bogus value provided by pmd */ + if (q < (2 * xd->driver_frame_overhead) && q > 0 && + di.max_mtu != UINT16_MAX) + xd->driver_frame_overhead = q; + dpdk_log_debug ("[%u] min_mtu: %u, max_mtu: %u, min_rx_bufsize: %u, " + "max_rx_pktlen: %u, max_lro_pkt_size: %u", + xd->port_id, di.min_mtu, di.max_mtu, di.min_rx_bufsize, + di.max_rx_pktlen, di.max_lro_pkt_size); +#endif + dpdk_log_debug ("[%u] driver frame overhead is %u", port_id, + xd->driver_frame_overhead); + + /* 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; + + dpdk_log_debug ( + "[%u] n_rx_queues: %u n_tx_queues: %u n_rx_desc: %u n_tx_desc: %u", + port_id, xd->conf.n_rx_queues, xd->conf.n_tx_queues, + xd->conf.n_rx_desc, xd->conf.n_tx_desc); + + vec_validate_aligned (xd->rx_queues, xd->conf.n_rx_queues - 1, + CLIB_CACHE_LINE_BYTES); + vec_validate_aligned (xd->tx_queues, xd->conf.n_tx_queues - 1, CLIB_CACHE_LINE_BYTES); - xd->flags |= DPDK_DEVICE_FLAG_PMD; - - /* workaround for drivers not setting driver_name */ - if ((!dev_info.driver_name) && (pci_dev)) - dev_info.driver_name = pci_dev->driver->driver.name; - - ASSERT (dev_info.driver_name); - - if (!xd->pmd) - { - - -#define _(s,f) else if (dev_info.driver_name && \ - !strcmp(dev_info.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; - xd->nb_rx_desc = DPDK_NB_RX_DESC_DEFAULT; - xd->nb_tx_desc = DPDK_NB_TX_DESC_DEFAULT; - - switch (xd->pmd) - { - /* Drivers with valid speed_capa set */ - case VNET_DPDK_PMD_I40E: - xd->flags |= DPDK_DEVICE_FLAG_INT_UNMASKABLE; - 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->port_type = port_type_from_speed_capa (&dev_info); - 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; - - if (dm->conf->no_tx_checksum_offload == 0) - { - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM; - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM; - xd->flags |= - DPDK_DEVICE_FLAG_TX_OFFLOAD | - DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM; - } - - xd->port_conf.intr_conf.rxq = 1; - break; - case VNET_DPDK_PMD_CXGBE: - case VNET_DPDK_PMD_MLX4: - case VNET_DPDK_PMD_MLX5: - case VNET_DPDK_PMD_QEDE: - case VNET_DPDK_PMD_BNXT: - xd->port_type = port_type_from_speed_capa (&dev_info); - break; - - /* SR-IOV VFs */ - case VNET_DPDK_PMD_I40EVF: - xd->flags |= DPDK_DEVICE_FLAG_INT_UNMASKABLE; - case VNET_DPDK_PMD_IGBVF: - case VNET_DPDK_PMD_IXGBEVF: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - if (dm->conf->no_tx_checksum_offload == 0) - { - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM; - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM; - xd->flags |= - DPDK_DEVICE_FLAG_TX_OFFLOAD | - DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM; - } - /* DPDK bug in multiqueue... */ - /* xd->port_conf.intr_conf.rxq = 1; */ - break; - - /* iAVF */ - case VNET_DPDK_PMD_IAVF: - xd->flags |= DPDK_DEVICE_FLAG_INT_UNMASKABLE; - 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; - - if (dm->conf->no_tx_checksum_offload == 0) - { - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM; - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM; - xd->flags |= - DPDK_DEVICE_FLAG_TX_OFFLOAD | - DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM; - } - /* 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; - - if (dm->conf->no_tx_checksum_offload == 0) - { - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM; - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM; - xd->flags |= DPDK_DEVICE_FLAG_TX_OFFLOAD; - } - break; - - case VNET_DPDK_PMD_ENA: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_SCATTER; - xd->port_conf.intr_conf.rxq = 1; - break; - - case VNET_DPDK_PMD_DPAA2: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_10G; - break; - - /* Cisco VIC */ - case VNET_DPDK_PMD_ENIC: - { - struct rte_eth_link l; - rte_eth_link_get_nowait (i, &l); - xd->port_type = port_type_from_link_speed (l.link_speed); - if (dm->conf->enable_tcp_udp_checksum) - dpdk_enable_l4_csum_offload (xd); - } - 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->port_conf.rxmode.mq_mode = ETH_MQ_RX_NONE; - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_1G; - xd->nb_rx_desc = DPDK_NB_RX_DESC_VIRTIO; - xd->nb_tx_desc = DPDK_NB_TX_DESC_VIRTIO; - /* - * Enable use of RX interrupts if supported. - * - * There is no device flag or capability for this, so - * use the same check that the virtio driver does. - */ - if (pci_dev && rte_intr_cap_multiple (&pci_dev->intr_handle)) - xd->port_conf.intr_conf.rxq = 1; - break; - - /* vmxnet3 */ - case VNET_DPDK_PMD_VMXNET3: - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_1G; - xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS; - break; - - case VNET_DPDK_PMD_AF_PACKET: - xd->port_type = VNET_DPDK_PORT_TYPE_AF_PACKET; - xd->af_packet_instance_num = af_packet_instance_num++; - 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->port_conf.intr_conf.lsc = 1; - break; - - case VNET_DPDK_PMD_NETVSC: - { - struct rte_eth_link l; - rte_eth_link_get_nowait (i, &l); - xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF; - } - break; - - default: - xd->port_type = VNET_DPDK_PORT_TYPE_UNKNOWN; - } - - if (devconf->num_rx_desc) - xd->nb_rx_desc = devconf->num_rx_desc; - else { - - /* If num_rx_desc is not specified by VPP user, the current CPU is working - with 2M page and has no L3 cache, default num_rx_desc is changed to 512 - from original 1024 to help reduce TLB misses. - */ - if ((clib_mem_get_default_hugepage_size () == 2 << 20) - && check_l3cache() == 0) - xd->nb_rx_desc = 512; - } - - if (devconf->num_tx_desc) - xd->nb_tx_desc = devconf->num_tx_desc; - else { - - /* If num_tx_desc is not specified by VPP user, the current CPU is working - with 2M page and has no L3 cache, default num_tx_desc is changed to 512 - from original 1024 to help reduce TLB misses. - */ - if ((clib_mem_get_default_hugepage_size () == 2 << 20) - && check_l3cache() == 0) - xd->nb_tx_desc = 512; - } - } - - 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 (i, (void *) addr); - - xd->port_id = i; - xd->device_index = xd - dm->devices; - xd->per_interface_next_index = ~0; - - /* assign interface to input thread */ - int q; - - error = ethernet_register_interface - (dm->vnet_main, dpdk_device_class.index, xd->device_index, - /* ethernet address */ addr, - &xd->hw_if_index, dpdk_flag_change); - if (error) - return error; - - /* - * Ensure default mtu is not > the mtu read from the hardware. - * Otherwise rte_eth_dev_configure() will fail and the port will - * not be available. - * Calculate max_frame_size and mtu supported by NIC - */ - if (ETHERNET_MAX_PACKET_BYTES > dev_info.max_rx_pktlen) - { - /* - * This device does not support the platforms's max frame - * size. Use it's advertised mru instead. - */ - max_rx_frame = dev_info.max_rx_pktlen; - mtu = dev_info.max_rx_pktlen - sizeof (ethernet_header_t); - } + 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; + eir.cb.set_max_frame_size = dpdk_set_max_frame_size; + xd->hw_if_index = vnet_eth_register_interface (vnm, &eir); + hi = vnet_get_hw_interface (vnm, xd->hw_if_index); + numa_node = (i8) rte_eth_dev_socket_id (port_id); + if (numa_node == SOCKET_ID_ANY) + /* numa_node is not set, default to 0 */ + hi->numa_node = xd->cpu_socket = 0; else - { - /* VPP treats MTU and max_rx_pktlen both equal to - * ETHERNET_MAX_PACKET_BYTES, if dev_info.max_rx_pktlen >= - * ETHERNET_MAX_PACKET_BYTES + sizeof(ethernet_header_t) - */ - if (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES + - sizeof (ethernet_header_t))) - { - mtu = ETHERNET_MAX_PACKET_BYTES; - max_rx_frame = ETHERNET_MAX_PACKET_BYTES; - - /* - * Some platforms do not account for Ethernet FCS (4 bytes) in - * MTU calculations. To interop with them increase mru but only - * if the device's settings can support it. - */ - if (dpdk_port_crc_strip_enabled (xd) && - (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES + - sizeof (ethernet_header_t) + - 4))) - { - max_rx_frame += 4; - } - } - else - { - max_rx_frame = ETHERNET_MAX_PACKET_BYTES; - mtu = ETHERNET_MAX_PACKET_BYTES - sizeof (ethernet_header_t); - - if (dpdk_port_crc_strip_enabled (xd) && - (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES + 4))) - { - max_rx_frame += 4; - } - } - } - - if (xd->pmd == VNET_DPDK_PMD_FAILSAFE) - { - /* failsafe device numerables are reported with active device only, - * need to query the mtu for current device setup to overwrite - * reported value. - */ - uint16_t dev_mtu; - if (!rte_eth_dev_get_mtu (i, &dev_mtu)) - { - mtu = dev_mtu; - max_rx_frame = mtu + sizeof (ethernet_header_t); - - if (dpdk_port_crc_strip_enabled (xd)) - { - max_rx_frame += 4; - } - } - } + hi->numa_node = xd->cpu_socket = numa_node; + sw = vnet_get_hw_sw_interface (vnm, xd->hw_if_index); + xd->sw_if_index = sw->sw_if_index; + dpdk_log_debug ("[%u] interface %v created", port_id, hi->name); - /*Set port rxmode config */ - xd->port_conf.rxmode.max_rx_pkt_len = max_rx_frame; + if (devconf->tag) + vnet_set_sw_interface_tag (vnm, devconf->tag, sw->sw_if_index); - sw = vnet_get_hw_sw_interface (dm->vnet_main, xd->hw_if_index); - xd->sw_if_index = sw->sw_if_index; - vnet_hw_if_set_input_node (dm->vnet_main, xd->hw_if_index, - dpdk_input_node.index); + 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 i; + int j; q = 0; - clib_bitmap_foreach (i, devconf->workers) { + clib_bitmap_foreach (j, devconf->workers) + { dpdk_rx_queue_t *rxq = vec_elt_at_index (xd->rx_queues, q); rxq->queue_index = vnet_hw_if_register_rx_queue ( - dm->vnet_main, xd->hw_if_index, q++, - vdm->first_worker_thread_index + i); - } + vnm, xd->hw_if_index, q++, vdm->first_worker_thread_index + j); + } } else - for (q = 0; q < xd->rx_q_used; q++) + for (q = 0; q < xd->conf.n_rx_queues; q++) { dpdk_rx_queue_t *rxq = vec_elt_at_index (xd->rx_queues, q); rxq->queue_index = vnet_hw_if_register_rx_queue ( - dm->vnet_main, xd->hw_if_index, q, VNET_HW_IF_RXQ_THREAD_ANY); + vnm, xd->hw_if_index, q, VNET_HW_IF_RXQ_THREAD_ANY); } - vnet_hw_if_update_runtime_data (dm->vnet_main, xd->hw_if_index); - - /*Get vnet hardware interface */ - hi = vnet_get_hw_interface (dm->vnet_main, xd->hw_if_index); + for (q = 0; q < xd->conf.n_tx_queues; q++) + { + dpdk_tx_queue_t *txq = vec_elt_at_index (xd->tx_queues, q); + txq->queue_index = + vnet_hw_if_register_tx_queue (vnm, xd->hw_if_index, q); + } - /*Override default max_packet_bytes and max_supported_bytes set in - * ethernet_register_interface() above*/ - if (hi) + for (q = 0; q < tm->n_vlib_mains; q++) { - hi->max_packet_bytes = mtu; - hi->max_supported_packet_bytes = max_rx_frame; - hi->numa_node = xd->cpu_socket; - - /* Indicate ability to support L3 DMAC filtering and - * initialize interface to L3 non-promisc mode */ - hi->caps |= VNET_HW_INTERFACE_CAP_SUPPORTS_MAC_FILTER; - ethernet_set_flags (dm->vnet_main, xd->hw_if_index, - ETHERNET_INTERFACE_FLAG_DEFAULT_L3); + u32 qi = xd->tx_queues[q % xd->conf.n_tx_queues].queue_index; + vnet_hw_if_tx_queue_assign_thread (vnm, qi, q); } - if (dm->conf->no_tx_checksum_offload == 0) - if (xd->flags & DPDK_DEVICE_FLAG_TX_OFFLOAD && hi != NULL) - { - hi->caps |= VNET_HW_INTERFACE_CAP_SUPPORTS_TX_IP4_CKSUM | - VNET_HW_INTERFACE_CAP_SUPPORTS_TX_TCP_CKSUM | - VNET_HW_INTERFACE_CAP_SUPPORTS_TX_UDP_CKSUM; - } - if (devconf->tso == DPDK_DEVICE_TSO_ON && hi != NULL) + if (devconf->tso == DPDK_DEVICE_TSO_ON) { /*tcp_udp checksum must be enabled*/ - if ((dm->conf->enable_tcp_udp_checksum) && - (hi->caps & VNET_HW_INTERFACE_CAP_SUPPORTS_TX_CKSUM)) - { - hi->caps |= VNET_HW_INTERFACE_CAP_SUPPORTS_TCP_GSO | - VNET_HW_INTERFACE_CAP_SUPPORTS_UDP_GSO; - xd->port_conf.txmode.offloads |= - DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_UDP_TSO; - } + if (xd->conf.enable_tcp_udp_checksum == 0) + dpdk_log_warn ("[%u] TCP/UDP checksum offload must be enabled", + xd->port_id); + else if ((di.tx_offload_capa & RTE_ETH_TX_OFFLOAD_TCP_TSO) == 0) + dpdk_log_warn ("[%u] TSO not supported by device", xd->port_id); else - clib_warning ("%s: TCP/UDP checksum offload must be enabled", - hi->name); + xd->conf.enable_tso = 1; } + if (devconf->max_lro_pkt_size) + xd->conf.max_lro_pkt_size = devconf->max_lro_pkt_size; + + if (devconf->disable_rxq_int) + xd->conf.enable_rxq_int = 0; + dpdk_device_setup (xd); /* rss queues should be configured after dpdk_device_setup() */ - if ((hi != NULL) && (devconf->rss_queues != NULL)) - { - if (vnet_hw_interface_set_rss_queues - (vnet_get_main (), hi, devconf->rss_queues)) - { - clib_warning ("%s: Failed to set rss queues", hi->name); - } - } + if (devconf->rss_queues) + { + if (vnet_hw_interface_set_rss_queues (vnet_get_main (), hi, + devconf->rss_queues)) + 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, i, - format_dpdk_device_errors, xd); - - /* - * A note on Cisco VIC (PMD_ENIC) and VLAN: - * - * With Cisco VIC vNIC, every ingress packet is tagged. On a - * trunk vNIC (C series "standalone" server), packets on no VLAN - * are tagged with vlan 0. On an access vNIC (standalone or B - * series "blade" server), packets on the default/native VLAN - * are tagged with that vNIC's VLAN. VPP expects these packets - * to be untagged, and previously enabled VLAN strip on VIC by - * default. But it also broke vlan sub-interfaces. - * - * The VIC adapter has "untag default vlan" ingress VLAN rewrite - * mode, which removes tags from these packets. VPP now includes - * a local patch for the enic driver to use this untag mode, so - * enabling vlan stripping is no longer needed. In future, the - * driver + dpdk will have an API to set the mode after - * rte_eal_init. Then, this note and local patch will be - * removed. - */ - - /* - * VLAN stripping: default to VLAN strip disabled, unless specified - * otherwise in the startup config. - */ - - vlan_off = rte_eth_dev_get_vlan_offload (xd->port_id); - if (devconf->vlan_strip_offload == DPDK_DEVICE_VLAN_STRIP_ON) - { - vlan_off |= ETH_VLAN_STRIP_OFFLOAD; - if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) >= 0) - dpdk_log_info ("VLAN strip enabled for interface\n"); - else - dpdk_log_warn ("VLAN strip cannot be supported by interface\n"); - xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_STRIP; - } - else - { - if (vlan_off & ETH_VLAN_STRIP_OFFLOAD) - { - vlan_off &= ~ETH_VLAN_STRIP_OFFLOAD; - if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) >= 0) - dpdk_log_warn ("set VLAN offload failed\n"); - } - xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP; - } - - if (hi) - hi->max_packet_bytes = xd->port_conf.rxmode.max_rx_pkt_len - - sizeof (ethernet_header_t); - else - dpdk_log_warn ("hi NULL"); - - if (dm->conf->no_multi_seg) - mtu = mtu > ETHER_MAX_LEN ? ETHER_MAX_LEN : mtu; - - rte_eth_dev_set_mtu (xd->port_id, mtu); -} + dpdk_log_err ("[%u] setup failed Errors:\n %U", port_id, + format_dpdk_device_errors, xd); + dpdk_counters_xstats_init (xd); + } - /* *INDENT-ON* */ + for (int i = 0; i < vec_len (dm->devices); i++) + vnet_hw_if_update_runtime_data (vnm, dm->devices[i].hw_if_index); return 0; } @@ -903,7 +625,6 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf) int i; addrs = vlib_pci_get_all_dev_addrs (); - /* *INDENT-OFF* */ vec_foreach (addr, addrs) { dpdk_device_config_t * devconf = 0; @@ -922,8 +643,18 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf) continue; } +#ifdef __FreeBSD__ + /* + * The defines for the PCI_CLASS_* types are platform specific and differ + * on FreeBSD. + */ + if (d->device_class != PCI_CLASS_NETWORK && + d->device_class != PCI_CLASS_PROCESSOR_CO) + continue; +#else if (d->device_class != PCI_CLASS_NETWORK_ETHERNET && d->device_class != PCI_CLASS_PROCESSOR_CO) continue; +#endif /* __FreeBSD__ */ if (num_whitelisted) { @@ -991,9 +722,13 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf) else if (d->vendor_id == 0x8086 && d->device_class == PCI_CLASS_NETWORK_ETHERNET) ; /* all Intel QAT devices VFs */ - else if (d->vendor_id == 0x8086 && d->device_class == PCI_CLASS_PROCESSOR_CO && - (d->device_id == 0x0443 || d->device_id == 0x18a1 || d->device_id == 0x19e3 || - d->device_id == 0x37c9 || d->device_id == 0x6f55)) + else if (d->vendor_id == 0x8086 && + d->device_class == PCI_CLASS_PROCESSOR_CO && + (d->device_id == 0x0443 || d->device_id == 0x18a1 || + d->device_id == 0x19e3 || d->device_id == 0x37c9 || + d->device_id == 0x6f55 || d->device_id == 0x18ef || + d->device_id == 0x4941 || d->device_id == 0x4943 || + d->device_id == 0x4945)) ; /* Cisco VIC */ else if (d->vendor_id == 0x1137 && @@ -1021,10 +756,28 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf) { continue; } - /* Mellanox CX6, CX6VF, CX6DX, CX6DXVF */ - else if (d->vendor_id == 0x15b3 && d->device_id >= 0x101b && d->device_id <= 0x101e) + /* Mellanox CX6, CX6VF, CX6DX, CX6DXVF, CX6LX */ + else if (d->vendor_id == 0x15b3 && + (d->device_id >= 0x101b && d->device_id <= 0x101f)) { - continue; + continue; + } + /* Mellanox CX7 */ + else if (d->vendor_id == 0x15b3 && d->device_id == 0x1021) + { + continue; + } + /* Mellanox BF, BFVF */ + else if (d->vendor_id == 0x15b3 && + (d->device_id >= 0xa2d2 && d->device_id <= 0Xa2d3)) + { + continue; + } + /* Mellanox BF2, BF3 */ + else if (d->vendor_id == 0x15b3 && + (d->device_id == 0xa2d6 || d->device_id == 0xa2dc)) + { + continue; } /* Broadcom NetXtreme S, and E series only */ else if (d->vendor_id == 0x14e4 && @@ -1039,6 +792,9 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf) d->device_id == 0x1614 || d->device_id == 0x1606 || d->device_id == 0x1609 || d->device_id == 0x1614))) ; + /* Google vNIC */ + else if (d->vendor_id == 0x1ae0 && d->device_id == 0x0042) + ; else { dpdk_log_warn ("Unsupported PCI device 0x%04x:0x%04x found " @@ -1047,7 +803,8 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf) continue; } - error = vlib_pci_bind_to_uio (vm, addr, (char *) conf->uio_driver_name); + error = vlib_pci_bind_to_uio (vm, addr, (char *) conf->uio_driver_name, + conf->uio_bind_force); if (error) { @@ -1063,7 +820,6 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf) clib_error_report (error); } } - /* *INDENT-ON* */ vec_free (pci_addr); vlib_pci_free_device_info (d); } @@ -1078,7 +834,6 @@ dpdk_bind_vmbus_devices_to_uio (dpdk_config_main_t * conf) addrs = vlib_vmbus_get_all_dev_addrs (); - /* *INDENT-OFF* */ vec_foreach (addr, addrs) { dpdk_device_config_t *devconf = 0; @@ -1143,7 +898,6 @@ dpdk_bind_vmbus_devices_to_uio (dpdk_config_main_t * conf) clib_error_report (error); } } - /* *INDENT-ON* */ } uword @@ -1240,7 +994,9 @@ 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, "tag %s", &devconf->tag)) ; else if (unformat (input, "workers %U", unformat_bitmap_list, &devconf->workers)) @@ -1253,10 +1009,10 @@ dpdk_device_config (dpdk_config_main_t *conf, void *addr, if (error) break; } - else if (unformat (input, "vlan-strip-offload off")) - devconf->vlan_strip_offload = DPDK_DEVICE_VLAN_STRIP_OFF; - else if (unformat (input, "vlan-strip-offload on")) - devconf->vlan_strip_offload = DPDK_DEVICE_VLAN_STRIP_ON; + else if (unformat (input, "no-rx-interrupts")) + { + devconf->disable_rxq_int = 1; + } else if (unformat (input, "tso on")) { devconf->tso = DPDK_DEVICE_TSO_ON; @@ -1270,6 +1026,9 @@ dpdk_device_config (dpdk_config_main_t *conf, void *addr, else if (unformat (input, "rss-queues %U", unformat_bitmap_list, &devconf->rss_queues)) ; + else if (unformat (input, "max-lro-pkt-size %u", + &devconf->max_lro_pkt_size)) + ; else { error = clib_error_return (0, "unknown input `%U'", @@ -1310,14 +1069,26 @@ dpdk_log_read_ready (clib_file_t * uf) n = read (uf->file_descriptor, s + len, n_try); if (n < 0 && errno != EAGAIN) return clib_error_return_unix (0, "read"); - _vec_len (s) = len + (n < 0 ? 0 : n); + vec_set_len (s, len + (n < 0 ? 0 : n)); } unformat_init_vector (&input, s); while (unformat_user (&input, unformat_line, &line)) { - dpdk_log_notice ("%v", line); + int skip = 0; + vec_add1 (line, 0); + + /* unfortunatelly DPDK polutes log with this error messages + * even when we pass --in-memory which means no secondary process */ + if (strstr ((char *) line, "WARNING! Base virtual address hint")) + skip = 1; + else if (strstr ((char *) line, "This may cause issues with mapping " + "memory into secondary processes")) + skip = 1; + vec_pop (line); + if (!skip) + dpdk_log_notice ("%v", line); vec_free (line); } @@ -1326,29 +1097,51 @@ dpdk_log_read_ready (clib_file_t * uf) } static clib_error_t * +dpdk_set_stat_poll_interval (f64 interval) +{ + if (interval < DPDK_MIN_STATS_POLL_INTERVAL) + return clib_error_return (0, "wrong stats-poll-interval value"); + + dpdk_main.stat_poll_interval = interval; + return 0; +} + +static clib_error_t * +dpdk_set_link_state_poll_interval (f64 interval) +{ + if (interval < DPDK_MIN_LINK_POLL_INTERVAL) + return clib_error_return (0, "wrong link-state-poll-interval value"); + + dpdk_main.link_state_poll_interval = interval; + return 0; +} + +static clib_error_t * dpdk_config (vlib_main_t * vm, unformat_input_t * input) { + dpdk_main_t *dm = &dpdk_main; clib_error_t *error = 0; dpdk_config_main_t *conf = &dpdk_config_main; - vlib_thread_main_t *tm = vlib_get_thread_main (); dpdk_device_config_t *devconf; vlib_pci_addr_t pci_addr = { 0 }; vlib_vmbus_addr_t vmbus_addr = { 0 }; unformat_input_t sub_input; +#ifdef __linux + vlib_thread_main_t *tm = vlib_get_thread_main (); uword default_hugepage_sz, x; + u8 file_prefix = 0; +#endif /* __linux__ */ u8 *s, *tmp = 0; int ret, i; int num_whitelisted = 0; int eal_no_hugetlb = 0; u8 no_pci = 0; u8 no_vmbus = 0; - u8 file_prefix = 0; u8 *socket_mem = 0; - u8 *huge_dir_path = 0; u32 vendor, device, domain, bus, func; - - huge_dir_path = - format (0, "%s/hugepages%c", vlib_unix_get_runtime_dir (), 0); + void *fmt_func; + void *fmt_addr; + f64 poll_interval; conf->device_config_index_by_pci_addr = hash_create (0, sizeof (uword)); mhash_init (&conf->device_config_index_by_vmbus_addr, sizeof (uword), @@ -1366,19 +1159,36 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) conf->enable_telemetry = 1; else if (unformat (input, "enable-tcp-udp-checksum")) - conf->enable_tcp_udp_checksum = 1; - + { + dm->default_port_conf.enable_tcp_udp_checksum = 1; + if (unformat (input, "enable-outer-checksum-offload")) + dm->default_port_conf.enable_outer_checksum_offload = 1; + } else if (unformat (input, "no-tx-checksum-offload")) - conf->no_tx_checksum_offload = 1; + dm->default_port_conf.disable_tx_checksum_offload = 1; else if (unformat (input, "decimal-interface-names")) conf->interface_name_format_decimal = 1; else if (unformat (input, "no-multi-seg")) - conf->no_multi_seg = 1; + dm->default_port_conf.disable_multi_seg = 1; + else if (unformat (input, "enable-lro")) + dm->default_port_conf.enable_lro = 1; else if (unformat (input, "max-simd-bitwidth %U", unformat_max_simd_bitwidth, &conf->max_simd_bitwidth)) ; + else if (unformat (input, "link-state-poll-interval %f", &poll_interval)) + { + error = dpdk_set_link_state_poll_interval (poll_interval); + if (error != 0) + return error; + } + else if (unformat (input, "stats-poll-interval %f", &poll_interval)) + { + error = dpdk_set_stat_poll_interval (poll_interval); + if (error != 0) + return error; + } else if (unformat (input, "dev default %U", unformat_vlib_cli_sub_input, &sub_input)) { @@ -1433,13 +1243,10 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) num_whitelisted++; } - else if (unformat (input, "num-mem-channels %d", &conf->nchannels)) - conf->nchannels_set_manually = 0; - else if (unformat (input, "num-crypto-mbufs %d", - &conf->num_crypto_mbufs)) - ; else if (unformat (input, "uio-driver %s", &conf->uio_driver_name)) ; + else if (unformat (input, "uio-bind-force")) + conf->uio_bind_force = 1; else if (unformat (input, "socket-mem %s", &socket_mem)) ; else if (unformat (input, "no-pci")) @@ -1489,6 +1296,7 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) } foreach_eal_double_hyphen_predicate_arg #undef _ +#ifdef __linux__ #define _(a) \ else if (unformat(input, #a " %s", &s)) \ { \ @@ -1504,6 +1312,7 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) } foreach_eal_double_hyphen_arg #undef _ +#endif /* __linux__ */ #define _(a,b) \ else if (unformat(input, #a " %s", &s)) \ { \ @@ -1514,28 +1323,13 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) } foreach_eal_single_hyphen_arg #undef _ -#define _(a,b) \ - else if (unformat(input, #a " %s", &s)) \ - { \ - tmp = format (0, "-%s%c", #b, 0); \ - vec_add1 (conf->eal_init_args, tmp); \ - vec_add1 (s, 0); \ - vec_add1 (conf->eal_init_args, s); \ - conf->a##_set_manually = 1; \ - } - foreach_eal_single_hyphen_mandatory_arg -#undef _ else if (unformat (input, "default")) ; else if (unformat_skip_white_space (input)) ; - else - { - error = clib_error_return (0, "unknown input `%U'", + else return clib_error_return (0, "unknown input `%U'", format_unformat_error, input); - goto done; - } } if (!conf->uio_driver_name) @@ -1545,9 +1339,13 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) { vec_add1 (conf->eal_init_args, (u8 *) "--in-memory"); +#ifdef __linux__ + /* + * FreeBSD performs huge page prealloc through a dedicated kernel mode + * this process is only required on Linux. + */ default_hugepage_sz = clib_mem_get_default_hugepage_size (); - /* *INDENT-OFF* */ clib_bitmap_foreach (x, tm->cpu_socket_bitmap) { clib_error_t *e; @@ -1560,7 +1358,7 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) if ((e = clib_sysfs_prealloc_hugepages(x, 0, n_pages))) clib_error_report (e); } - /* *INDENT-ON* */ +#endif /* __linux__ */ } /* on/off dpdk's telemetry thread */ @@ -1569,6 +1367,7 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) vec_add1 (conf->eal_init_args, (u8 *) "--no-telemetry"); } +#ifdef __linux__ if (!file_prefix) { tmp = format (0, "--file-prefix%c", 0); @@ -1576,41 +1375,7 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) tmp = format (0, "vpp%c", 0); vec_add1 (conf->eal_init_args, tmp); } - - if (error) - return error; - - /* I'll bet that -c and -n must be the first and second args... */ - if (!conf->coremask_set_manually) - { - vlib_thread_registration_t *tr; - uword *coremask = 0; - int i; - - /* main thread core */ - coremask = clib_bitmap_set (coremask, tm->main_lcore, 1); - - for (i = 0; i < vec_len (tm->registrations); i++) - { - tr = tm->registrations[i]; - coremask = clib_bitmap_or (coremask, tr->coremask); - } - - vec_insert (conf->eal_init_args, 2, 1); - conf->eal_init_args[1] = (u8 *) "-c"; - tmp = format (0, "%U%c", format_bitmap_hex, coremask, 0); - conf->eal_init_args[2] = tmp; - clib_bitmap_free (coremask); - } - - if (!conf->nchannels_set_manually) - { - vec_insert (conf->eal_init_args, 2, 3); - conf->eal_init_args[3] = (u8 *) "-n"; - tmp = format (0, "%d", conf->nchannels); - vec_terminate_c_string (tmp); - conf->eal_init_args[4] = tmp; - } +#endif if (no_pci == 0 && geteuid () == 0) dpdk_bind_devices_to_uio (conf); @@ -1622,15 +1387,11 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) if (devconf->x == 0 && conf->default_devconf.x > 0) \ devconf->x = conf->default_devconf.x ; - /* *INDENT-OFF* */ pool_foreach (devconf, conf->dev_confs) { /* default per-device config items */ foreach_dpdk_device_config_item - /* copy vlan_strip config from default device */ - _ (vlan_strip_offload) - /* copy tso config from default device */ _ (tso) @@ -1640,56 +1401,57 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) /* copy rss_queues config from default device */ _ (rss_queues) - /* add DPDK EAL whitelist/blacklist entry */ - if (num_whitelisted > 0 && devconf->is_blacklisted == 0 && - devconf->dev_addr_type == VNET_DEV_ADDR_PCI) - { - tmp = format (0, "-a%c", 0); - vec_add1 (conf->eal_init_args, tmp); - if (devconf->devargs) + /* assume that default is PCI */ + fmt_func = format_vlib_pci_addr; + fmt_addr = &devconf->pci_addr; + + if (devconf->dev_addr_type == VNET_DEV_ADDR_VMBUS) + { + fmt_func = format_vlib_vmbus_addr; + fmt_addr = &devconf->vmbus_addr; + } + + /* add DPDK EAL whitelist/blacklist entry */ + if (num_whitelisted > 0 && devconf->is_blacklisted == 0) + { + tmp = format (0, "-a%c", 0); + vec_add1 (conf->eal_init_args, tmp); + if (devconf->devargs) { - tmp = format (0, "%U,%s%c", format_vlib_pci_addr, - &devconf->pci_addr, devconf->devargs, 0); + tmp = + format (0, "%U,%s%c", fmt_func, fmt_addr, devconf->devargs, 0); } else { - tmp = format (0, "%U%c", format_vlib_pci_addr, &devconf->pci_addr, 0); + tmp = format (0, "%U%c", fmt_func, fmt_addr, 0); } vec_add1 (conf->eal_init_args, tmp); - } - else if (num_whitelisted == 0 && devconf->is_blacklisted != 0 && - devconf->dev_addr_type == VNET_DEV_ADDR_PCI) - { - tmp = format (0, "-b%c", 0); - vec_add1 (conf->eal_init_args, tmp); - tmp = format (0, "%U%c", format_vlib_pci_addr, &devconf->pci_addr, 0); - vec_add1 (conf->eal_init_args, tmp); - } + } + else if (num_whitelisted == 0 && devconf->is_blacklisted != 0) + { + tmp = format (0, "-b%c", 0); + vec_add1 (conf->eal_init_args, tmp); + tmp = format (0, "%U%c", fmt_func, fmt_addr, 0); + vec_add1 (conf->eal_init_args, tmp); + } } - /* *INDENT-ON* */ #undef _ - /* set master-lcore */ - tmp = format (0, "--main-lcore%c", 0); - vec_add1 (conf->eal_init_args, tmp); - tmp = format (0, "%u%c", tm->main_lcore, 0); - vec_add1 (conf->eal_init_args, tmp); - - if (socket_mem) clib_warning ("socket-mem argument is deprecated"); /* NULL terminate the "argv" vector, in case of stupidity */ vec_add1 (conf->eal_init_args, 0); - _vec_len (conf->eal_init_args) -= 1; + vec_dec_len (conf->eal_init_args, 1); /* Set up DPDK eal and packet mbuf pool early. */ int log_fds[2] = { 0 }; if (pipe (log_fds) == 0) { - if (fcntl (log_fds[1], F_SETFL, O_NONBLOCK) == 0) + if (fcntl (log_fds[0], F_SETFL, O_NONBLOCK) == 0 && + fcntl (log_fds[1], F_SETFL, O_NONBLOCK) == 0) { FILE *f = fdopen (log_fds[1], "a"); if (f && rte_openlog_stream (f) == 0) @@ -1720,6 +1482,8 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) dpdk_log_notice ("EAL init args: %s", conf->eal_init_args_str); ret = rte_eal_init (vec_len (conf->eal_init_args), (char **) conf->eal_init_args); + if (ret < 0) + return clib_error_return (0, "rte_eal_init returned %d", ret); /* enable the AVX-512 vPMDs in DPDK */ if (clib_cpu_supports_avx512_bitalg () && @@ -1731,20 +1495,11 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input) RTE_VECT_SIMD_256 : RTE_VECT_SIMD_512); - /* lazy umount hugepages */ - umount2 ((char *) huge_dir_path, MNT_DETACH); - rmdir ((char *) huge_dir_path); - vec_free (huge_dir_path); - - if (ret < 0) - return clib_error_return (0, "rte_eal_init returned %d", ret); - /* main thread 1st */ if ((error = dpdk_buffer_pools_create (vm))) return error; -done: - return error; + return 0; } VLIB_CONFIG_FUNCTION (dpdk_config, "dpdk"); @@ -1757,10 +1512,6 @@ dpdk_update_link_state (dpdk_device_t * xd, f64 now) u32 hw_flags = 0; u8 hw_flags_chg = 0; - /* only update link state for PMD interfaces */ - if ((xd->flags & DPDK_DEVICE_FLAG_PMD) == 0) - return; - xd->time_last_link_update = now ? now : xd->time_last_link_update; clib_memset (&xd->link, 0, sizeof (xd->link)); rte_eth_link_get_nowait (xd->port_id, &xd->link); @@ -1788,35 +1539,32 @@ dpdk_update_link_state (dpdk_device_t * xd, f64 now) ed->new_link_state = (u8) xd->link.link_status; } - if ((xd->link.link_duplex != prev_link.link_duplex)) + hw_flags_chg = ((xd->link.link_duplex != prev_link.link_duplex) || + (xd->link.link_status != prev_link.link_status)); + + if (xd->link.link_speed != prev_link.link_speed) + vnet_hw_interface_set_link_speed (vnm, xd->hw_if_index, + (xd->link.link_speed == UINT32_MAX) ? + UINT32_MAX : + xd->link.link_speed * 1000); + + if (hw_flags_chg) { - hw_flags_chg = 1; + if (xd->link.link_status) + hw_flags |= VNET_HW_INTERFACE_FLAG_LINK_UP; + switch (xd->link.link_duplex) { - case ETH_LINK_HALF_DUPLEX: + case RTE_ETH_LINK_HALF_DUPLEX: hw_flags |= VNET_HW_INTERFACE_FLAG_HALF_DUPLEX; break; - case ETH_LINK_FULL_DUPLEX: + case RTE_ETH_LINK_FULL_DUPLEX: hw_flags |= VNET_HW_INTERFACE_FLAG_FULL_DUPLEX; break; default: break; } - } - if (xd->link.link_speed != prev_link.link_speed) - vnet_hw_interface_set_link_speed (vnm, xd->hw_if_index, - xd->link.link_speed * 1000); - if (xd->link.link_status != prev_link.link_status) - { - hw_flags_chg = 1; - - if (xd->link.link_status) - hw_flags |= VNET_HW_INTERFACE_FLAG_LINK_UP; - } - - if (hw_flags_chg) - { if (LINK_STATE_ELOGS) { ELOG_TYPE_DECLARE (e) = @@ -1846,6 +1594,7 @@ dpdk_process (vlib_main_t * vm, vlib_node_runtime_t * rt, vlib_frame_t * f) dpdk_device_t *xd; vlib_thread_main_t *tm = vlib_get_thread_main (); + vlib_worker_thread_barrier_sync (vm); error = dpdk_lib_init (dm); if (error) @@ -1862,6 +1611,7 @@ dpdk_process (vlib_main_t * vm, vlib_node_runtime_t * rt, vlib_frame_t * f) } } + vlib_worker_thread_barrier_release (vm); tm->worker_thread_release = 1; f64 now = vlib_time_now (vm); @@ -1870,16 +1620,17 @@ dpdk_process (vlib_main_t * vm, vlib_node_runtime_t * rt, vlib_frame_t * f) dpdk_update_link_state (xd, now); } + f64 timeout = + clib_min (dm->link_state_poll_interval, dm->stat_poll_interval); + while (1) { - /* - * check each time through the loop in case intervals are changed - */ - f64 min_wait = dm->link_state_poll_interval < dm->stat_poll_interval ? - dm->link_state_poll_interval : dm->stat_poll_interval; - + f64 min_wait = clib_max (timeout, DPDK_MIN_POLL_INTERVAL); vlib_process_wait_for_event_or_clock (vm, min_wait); + timeout = + clib_min (dm->link_state_poll_interval, dm->stat_poll_interval); + if (dm->admin_up_down_in_progress) /* skip the poll if an admin up down is in progress (on any interface) */ continue; @@ -1893,19 +1644,25 @@ dpdk_process (vlib_main_t * vm, vlib_node_runtime_t * rt, vlib_frame_t * f) dpdk_update_link_state (xd, now); } - } + now = vlib_time_now (vm); + vec_foreach (xd, dm->devices) + { + timeout = clib_min (timeout, xd->time_last_stats_update + + dm->stat_poll_interval - now); + timeout = clib_min (timeout, xd->time_last_link_update + + dm->link_state_poll_interval - now); + } + } return 0; } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (dpdk_process_node,static) = { .function = dpdk_process, .type = VLIB_NODE_TYPE_PROCESS, .name = "dpdk-process", .process_log2_n_stack_bytes = 17, }; -/* *INDENT-ON* */ static clib_error_t * dpdk_init (vlib_main_t * vm) @@ -1921,40 +1678,31 @@ dpdk_init (vlib_main_t * vm) "Data in cache line 0 is bigger than cache line size"); STATIC_ASSERT (offsetof (frame_queue_trace_t, cacheline0) == 0, "Cache line marker must be 1st element in frame_queue_trace_t"); - STATIC_ASSERT (RTE_CACHE_LINE_SIZE == 1 << CLIB_LOG2_CACHE_LINE_BYTES, - "DPDK RTE CACHE LINE SIZE does not match with 1<<CLIB_LOG2_CACHE_LINE_BYTES"); dpdk_cli_reference (); - dm->vlib_main = vm; - dm->vnet_main = vnet_get_main (); dm->conf = &dpdk_config_main; - dm->conf->nchannels = 4; vec_add1 (dm->conf->eal_init_args, (u8 *) "vnet"); - /* Default vlib_buffer_t flags, DISABLES tcp/udp checksumming... */ - dm->buffer_flags_template = (VLIB_BUFFER_TOTAL_LENGTH_VALID | - VLIB_BUFFER_EXT_HDR_VALID | - VNET_BUFFER_F_L4_CHECKSUM_COMPUTED | - VNET_BUFFER_F_L4_CHECKSUM_CORRECT); - dm->stat_poll_interval = DPDK_STATS_POLL_INTERVAL; dm->link_state_poll_interval = DPDK_LINK_POLL_INTERVAL; dm->log_default = vlib_log_register_class ("dpdk", 0); dm->log_cryptodev = vlib_log_register_class ("dpdk", "cryptodev"); - dm->log_ipsec = vlib_log_register_class ("dpdk", "ipsec"); return error; } VLIB_INIT_FUNCTION (dpdk_init); -/* - * fd.io coding-style-patch-verification: ON - * - * Local Variables: - * eval: (c-set-style "gnu") - * End: - */ +static clib_error_t * +dpdk_worker_thread_init (vlib_main_t *vm) +{ + if (rte_thread_register () < 0) + clib_panic ("dpdk: cannot register thread %u - %s", vm->thread_index, + rte_strerror (rte_errno)); + return 0; +} + +VLIB_WORKER_INIT_FUNCTION (dpdk_worker_thread_init); diff --git a/src/plugins/dpdk/device/node.c b/src/plugins/dpdk/device/node.c index 25222856912..ca1690b708f 100644 --- a/src/plugins/dpdk/device/node.c +++ b/src/plugins/dpdk/device/node.c @@ -23,10 +23,10 @@ #include <dpdk/device/dpdk.h> #include <vnet/classify/vnet_classify.h> #include <vnet/mpls/packet.h> -#include <vnet/handoff.h> #include <vnet/devices/devices.h> #include <vnet/interface/rx_queue_funcs.h> #include <vnet/feature/feature.h> +#include <vnet/tcp/tcp_packet.h> #include <dpdk/device/dpdk_priv.h> @@ -36,9 +36,13 @@ static char *dpdk_error_strings[] = { #undef _ }; -/* make sure all flags we need are stored in lower 8 bits */ -STATIC_ASSERT ((PKT_RX_IP_CKSUM_BAD | PKT_RX_FDIR) < - 256, "dpdk flags not un lower byte, fix needed"); +/* make sure all flags we need are stored in lower 32 bits */ +STATIC_ASSERT ((u64) (RTE_MBUF_F_RX_IP_CKSUM_BAD | RTE_MBUF_F_RX_L4_CKSUM_BAD | + RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_LRO) < (1ULL << 32), + "dpdk flags not in lower word, fix needed"); + +STATIC_ASSERT (RTE_MBUF_F_RX_L4_CKSUM_BAD == (1ULL << 3), + "bit number of RTE_MBUF_F_RX_L4_CKSUM_BAD is no longer 3!"); static_always_inline uword dpdk_process_subseq_segs (vlib_main_t * vm, vlib_buffer_t * b, @@ -97,13 +101,13 @@ dpdk_prefetch_buffer_x4 (struct rte_mbuf *mb[]) { vlib_buffer_t *b; b = vlib_buffer_from_rte_mbuf (mb[0]); - clib_prefetch_load (b); + clib_prefetch_store (b); b = vlib_buffer_from_rte_mbuf (mb[1]); - clib_prefetch_load (b); + clib_prefetch_store (b); b = vlib_buffer_from_rte_mbuf (mb[2]); - clib_prefetch_load (b); + clib_prefetch_store (b); b = vlib_buffer_from_rte_mbuf (mb[3]); - clib_prefetch_load (b); + clib_prefetch_store (b); } /** \brief Main DPDK input node @@ -125,18 +129,18 @@ dpdk_prefetch_buffer_x4 (struct rte_mbuf *mb[]) @em Uses: - <code>struct rte_mbuf mb->ol_flags</code> - - PKT_RX_IP_CKSUM_BAD + - RTE_MBUF_F_RX_IP_CKSUM_BAD @em Sets: - <code>b->error</code> if the packet is to be dropped immediately - <code>b->current_data, b->current_length</code> - - adjusted as needed to skip the L2 header in direct-dispatch cases + - adjusted as needed to skip the L2 header in direct-dispatch cases - <code>vnet_buffer(b)->sw_if_index[VLIB_RX]</code> - - rx interface sw_if_index + - rx interface sw_if_index - <code>vnet_buffer(b)->sw_if_index[VLIB_TX] = ~0</code> - - required by ipX-lookup + - required by ipX-lookup - <code>b->flags</code> - - to indicate multi-segment pkts (VLIB_BUFFER_NEXT_PRESENT), etc. + - to indicate multi-segment pkts (VLIB_BUFFER_NEXT_PRESENT), etc. <em>Next Nodes:</em> - Static arcs to: error-drop, ethernet-input, @@ -145,31 +149,30 @@ dpdk_prefetch_buffer_x4 (struct rte_mbuf *mb[]) <code>xd->per_interface_next_index</code> */ -static_always_inline u16 -dpdk_ol_flags_extract (struct rte_mbuf **mb, u16 * flags, int count) +static_always_inline u32 +dpdk_ol_flags_extract (struct rte_mbuf **mb, u32 *flags, int count) { - u16 rv = 0; + u32 rv = 0; int i; for (i = 0; i < count; i++) { /* all flags we are interested in are in lower 8 bits but that might change */ - flags[i] = (u16) mb[i]->ol_flags; + flags[i] = (u32) mb[i]->ol_flags; rv |= flags[i]; } return rv; } static_always_inline uword -dpdk_process_rx_burst (vlib_main_t * vm, dpdk_per_thread_data_t * ptd, - uword n_rx_packets, int maybe_multiseg, - u16 * or_flagsp) +dpdk_process_rx_burst (vlib_main_t *vm, dpdk_per_thread_data_t *ptd, + uword n_rx_packets, int maybe_multiseg, u32 *or_flagsp) { u32 n_left = n_rx_packets; vlib_buffer_t *b[4]; struct rte_mbuf **mb = ptd->mbufs; uword n_bytes = 0; - u16 *flags, or_flags = 0; + u32 *flags, or_flags = 0; vlib_buffer_t bt; mb = ptd->mbufs; @@ -254,7 +257,7 @@ dpdk_process_flow_offload (dpdk_device_t * xd, dpdk_per_thread_data_t * ptd, /* TODO prefetch and quad-loop */ for (n = 0; n < n_rx_packets; n++) { - if ((ptd->flags[n] & PKT_RX_FDIR_ID) == 0) + if ((ptd->flags[n] & RTE_MBUF_F_RX_FDIR_ID) == 0) continue; fle = pool_elt_at_index (xd->flow_lookup_entries, @@ -277,6 +280,65 @@ dpdk_process_flow_offload (dpdk_device_t * xd, dpdk_per_thread_data_t * ptd, } } +static_always_inline u16 +dpdk_lro_find_l4_hdr_sz (vlib_buffer_t *b) +{ + u16 l4_hdr_sz = 0; + u16 current_offset = 0; + ethernet_header_t *e; + tcp_header_t *tcp; + u8 *data = vlib_buffer_get_current (b); + u16 ethertype; + e = (void *) data; + current_offset += sizeof (e[0]); + ethertype = clib_net_to_host_u16 (e->type); + if (ethernet_frame_is_tagged (ethertype)) + { + ethernet_vlan_header_t *vlan = (ethernet_vlan_header_t *) (e + 1); + ethertype = clib_net_to_host_u16 (vlan->type); + current_offset += sizeof (*vlan); + if (ethertype == ETHERNET_TYPE_VLAN) + { + vlan++; + current_offset += sizeof (*vlan); + ethertype = clib_net_to_host_u16 (vlan->type); + } + } + data += current_offset; + if (ethertype == ETHERNET_TYPE_IP4) + { + data += sizeof (ip4_header_t); + tcp = (void *) data; + l4_hdr_sz = tcp_header_bytes (tcp); + } + else + { + /* FIXME: extension headers...*/ + data += sizeof (ip6_header_t); + tcp = (void *) data; + l4_hdr_sz = tcp_header_bytes (tcp); + } + return l4_hdr_sz; +} + +static_always_inline void +dpdk_process_lro_offload (dpdk_device_t *xd, dpdk_per_thread_data_t *ptd, + uword n_rx_packets) +{ + uword n; + vlib_buffer_t *b0; + for (n = 0; n < n_rx_packets; n++) + { + b0 = vlib_buffer_from_rte_mbuf (ptd->mbufs[n]); + if (ptd->flags[n] & RTE_MBUF_F_RX_LRO) + { + b0->flags |= VNET_BUFFER_F_GSO; + vnet_buffer2 (b0)->gso_size = ptd->mbufs[n]->tso_segsz; + vnet_buffer2 (b0)->gso_l4_hdr_sz = dpdk_lro_find_l4_hdr_sz (b0); + } + } +} + static_always_inline u32 dpdk_device_input (vlib_main_t * vm, dpdk_main_t * dm, dpdk_device_t * xd, vlib_node_runtime_t * node, u32 thread_index, u16 queue_id) @@ -289,7 +351,7 @@ dpdk_device_input (vlib_main_t * vm, dpdk_main_t * dm, dpdk_device_t * xd, struct rte_mbuf **mb; vlib_buffer_t *b0; u16 *next; - u16 or_flags; + u32 or_flags; u32 n; int single_next = 0; @@ -303,12 +365,13 @@ dpdk_device_input (vlib_main_t * vm, dpdk_main_t * dm, dpdk_device_t * xd, /* get up to DPDK_RX_BURST_SZ buffers from PMD */ while (n_rx_packets < DPDK_RX_BURST_SZ) { - n = rte_eth_rx_burst (xd->port_id, queue_id, - ptd->mbufs + n_rx_packets, - DPDK_RX_BURST_SZ - n_rx_packets); + u32 n_to_rx = clib_min (DPDK_RX_BURST_SZ - n_rx_packets, 32); + + n = rte_eth_rx_burst (xd->port_id, queue_id, ptd->mbufs + n_rx_packets, + n_to_rx); n_rx_packets += n; - if (n < 32) + if (n < n_to_rx) break; } @@ -318,6 +381,7 @@ dpdk_device_input (vlib_main_t * vm, dpdk_main_t * dm, dpdk_device_t * xd, /* Update buffer template */ vnet_buffer (bt)->sw_if_index[VLIB_RX] = xd->sw_if_index; bt->error = node->errors[DPDK_ERROR_NONE]; + bt->flags = xd->buffer_flags; /* as DPDK is allocating empty buffers from mempool provided before interface start for each queue, it is safe to store this in the template */ bt->buffer_pool_index = rxq->buffer_pool_index; @@ -332,14 +396,34 @@ dpdk_device_input (vlib_main_t * vm, dpdk_main_t * dm, dpdk_device_t * xd, /* as all packets belong to the same interface feature arc lookup can be don once and result stored in the buffer template */ if (PREDICT_FALSE (vnet_device_input_have_features (xd->sw_if_index))) - vnet_feature_start_device_input_x1 (xd->sw_if_index, &next_index, bt); + vnet_feature_start_device_input (xd->sw_if_index, &next_index, bt); if (xd->flags & DPDK_DEVICE_FLAG_MAYBE_MULTISEG) n_rx_bytes = dpdk_process_rx_burst (vm, ptd, n_rx_packets, 1, &or_flags); else n_rx_bytes = dpdk_process_rx_burst (vm, ptd, n_rx_packets, 0, &or_flags); - if (PREDICT_FALSE (or_flags & PKT_RX_FDIR)) + if (PREDICT_FALSE ((or_flags & RTE_MBUF_F_RX_LRO))) + dpdk_process_lro_offload (xd, ptd, n_rx_packets); + + if (PREDICT_FALSE ((or_flags & RTE_MBUF_F_RX_L4_CKSUM_BAD) && + (xd->buffer_flags & VNET_BUFFER_F_L4_CHECKSUM_CORRECT))) + { + for (n = 0; n < n_rx_packets; n++) + { + /* Check and reset VNET_BUFFER_F_L4_CHECKSUM_CORRECT flag + if RTE_MBUF_F_RX_L4_CKSUM_BAD is set. + The magic num 3 is the bit number of RTE_MBUF_F_RX_L4_CKSUM_BAD + which is defined in DPDK. + Have made a STATIC_ASSERT in this file to ensure this. + */ + b0 = vlib_buffer_from_rte_mbuf (ptd->mbufs[n]); + b0->flags ^= (ptd->flags[n] & RTE_MBUF_F_RX_L4_CKSUM_BAD) + << (VNET_BUFFER_F_LOG2_L4_CHECKSUM_CORRECT - 3); + } + } + + if (PREDICT_FALSE (or_flags & RTE_MBUF_F_RX_FDIR)) { /* some packets will need to go to different next nodes */ for (n = 0; n < n_rx_packets; n++) @@ -348,7 +432,7 @@ dpdk_device_input (vlib_main_t * vm, dpdk_main_t * dm, dpdk_device_t * xd, /* flow offload - process if rx flow offload enabled and at least one packet is marked */ if (PREDICT_FALSE ((xd->flags & DPDK_DEVICE_FLAG_RX_FLOW_OFFLOAD) && - (or_flags & PKT_RX_FDIR))) + (or_flags & RTE_MBUF_F_RX_FDIR))) dpdk_process_flow_offload (xd, ptd, n_rx_packets); /* enqueue buffers to the next node */ @@ -385,7 +469,7 @@ dpdk_device_input (vlib_main_t * vm, dpdk_main_t * dm, dpdk_device_t * xd, marked as ip4 checksum bad we can notify ethernet input so it can send pacets to ip4-input-no-checksum node */ if (xd->flags & DPDK_DEVICE_FLAG_RX_IP4_CKSUM && - (or_flags & PKT_RX_IP_CKSUM_BAD) == 0) + (or_flags & RTE_MBUF_F_RX_IP_CKSUM_BAD) == 0) f->flags |= ETH_INPUT_FRAME_F_IP4_CKSUM_OK; vlib_frame_no_append (f); } @@ -459,7 +543,7 @@ VLIB_NODE_FN (dpdk_input_node) (vlib_main_t * vm, vlib_node_runtime_t * node, dpdk_device_t *xd; uword n_rx_packets = 0; vnet_hw_if_rxq_poll_vector_t *pv; - u32 thread_index = node->thread_index; + u32 thread_index = vm->thread_index; /* * Poll all devices on this cpu for input/interrupts. @@ -476,7 +560,6 @@ VLIB_NODE_FN (dpdk_input_node) (vlib_main_t * vm, vlib_node_runtime_t * node, return n_rx_packets; } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (dpdk_input_node) = { .type = VLIB_NODE_TYPE_INPUT, .name = "dpdk-input", @@ -492,7 +575,6 @@ VLIB_REGISTER_NODE (dpdk_input_node) = { .n_errors = DPDK_N_ERROR, .error_strings = dpdk_error_strings, }; -/* *INDENT-ON* */ /* * fd.io coding-style-patch-verification: ON diff --git a/src/plugins/dpdk/main.c b/src/plugins/dpdk/main.c index 413a0f0df9d..9781d0ed7f0 100644 --- a/src/plugins/dpdk/main.c +++ b/src/plugins/dpdk/main.c @@ -13,13 +13,6 @@ * limitations under the License. */ -#include <unistd.h> -#include <sys/types.h> -#include <sys/stat.h> -#include <fcntl.h> -#include <linux/vfio.h> -#include <sys/ioctl.h> - #include <vnet/vnet.h> #include <vnet/plugin/plugin.h> #include <dpdk/device/dpdk.h> @@ -79,19 +72,14 @@ rte_delay_us_override_cb (unsigned us) static clib_error_t * dpdk_main_init (vlib_main_t * vm) { - dpdk_main_t * dm = &dpdk_main; clib_error_t * error = 0; - dm->vlib_main = vm; - dm->vnet_main = vnet_get_main (); - /* register custom delay function */ rte_delay_us_callback_register (rte_delay_us_override_cb); return error; } -/* *INDENT-OFF* */ VLIB_INIT_FUNCTION (dpdk_main_init) = { .runs_after = VLIB_INITS("dpdk_init"), @@ -101,4 +89,3 @@ VLIB_PLUGIN_REGISTER () = { .version = VPP_BUILD_VER, .description = "Data Plane Development Kit (DPDK)", }; -/* *INDENT-ON* */ diff --git a/src/plugins/dpdk/thread.c b/src/plugins/dpdk/thread.c deleted file mode 100644 index 3a3fcc6cea6..00000000000 --- a/src/plugins/dpdk/thread.c +++ /dev/null @@ -1,85 +0,0 @@ -/* - * Copyright (c) 2017 Cisco and/or its affiliates. - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at: - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -#include <rte_config.h> - -#include <rte_common.h> -#include <rte_log.h> -#include <rte_memory.h> -#include <rte_memzone.h> -#include <rte_tailq.h> -#include <rte_eal.h> -#include <rte_per_lcore.h> -#include <rte_launch.h> -#include <rte_atomic.h> -#include <rte_cycles.h> -#include <rte_prefetch.h> -#include <rte_lcore.h> -#include <rte_per_lcore.h> -#include <rte_branch_prediction.h> -#include <rte_interrupts.h> -#include <rte_pci.h> -#include <rte_random.h> -#include <rte_debug.h> -#include <rte_ether.h> -#include <rte_ethdev.h> -#include <rte_ring.h> -#include <rte_mempool.h> -#include <rte_mbuf.h> -#include <rte_version.h> - -#include <vlib/vlib.h> -#include <vnet/vnet.h> -#include <dpdk/device/dpdk.h> -#include <dpdk/device/dpdk_priv.h> - -static clib_error_t * -dpdk_launch_thread (void *fp, vlib_worker_thread_t * w, unsigned lcore_id) -{ - int r; - r = rte_eal_remote_launch (fp, (void *) w, lcore_id); - if (r) - return clib_error_return (0, "Failed to launch thread %u", lcore_id); - return 0; -} - -static clib_error_t * -dpdk_thread_set_lcore (u32 thread, u16 lcore) -{ - return 0; -} - -static vlib_thread_callbacks_t callbacks = { - .vlib_launch_thread_cb = &dpdk_launch_thread, - .vlib_thread_set_lcore_cb = &dpdk_thread_set_lcore, -}; - -static clib_error_t * -dpdk_thread_init (vlib_main_t * vm) -{ - vlib_thread_cb_register (vm, &callbacks); - return 0; -} - -VLIB_INIT_FUNCTION (dpdk_thread_init); - -/** @endcond */ -/* - * fd.io coding-style-patch-verification: ON - * - * Local Variables: - * eval: (c-set-style "gnu") - * End: - */ |