2 // Copyright (c) 2010-2017 Intel Corporation
4 // Licensed under the Apache License, Version 2.0 (the "License");
5 // you may not use this file except in compliance with the License.
6 // You may obtain a copy of the License at
8 // http://www.apache.org/licenses/LICENSE-2.0
10 // Unless required by applicable law or agreed to in writing, software
11 // distributed under the License is distributed on an "AS IS" BASIS,
12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 // See the License for the specific language governing permissions and
14 // limitations under the License.
22 #include <rte_cycles.h>
23 #include <rte_atomic.h>
24 #include <rte_table_hash.h>
25 #include <rte_memzone.h>
26 #include <rte_errno.h>
28 #include "prox_malloc.h"
36 #include "prox_args.h"
37 #include "prox_assert.h"
39 #include "prox_shared.h"
40 #include "prox_port_cfg.h"
42 #include "hash_utils.h"
43 #include "handle_lb_net.h"
44 #include "prox_cksum.h"
45 #include "thread_nop.h"
46 #include "thread_generic.h"
47 #include "thread_pipeline.h"
49 #include "handle_master.h"
51 #if RTE_VERSION < RTE_VERSION_NUM(1,8,0,0)
52 #define RTE_CACHE_LINE_SIZE CACHE_LINE_SIZE
55 uint8_t lb_nb_txrings = 0xff;
56 struct rte_ring *ctrl_rings[RTE_MAX_LCORE*MAX_TASKS_PER_CORE];
58 static void __attribute__((noreturn)) prox_usage(const char *prgname)
60 plog_info("\nUsage: %s [-f CONFIG_FILE] [-a|-e] [-m|-s|-i] [-w DEF] [-u] [-t]\n"
61 "\t-f CONFIG_FILE : configuration file to load, ./prox.cfg by default\n"
62 "\t-l LOG_FILE : log file name, ./prox.log by default\n"
63 "\t-p : include PID in log file name if default log file is used\n"
64 "\t-o DISPLAY: Set display to use, can be 'curses' (default), 'cli' or 'none'\n"
65 "\t-v verbosity : initial logging verbosity\n"
66 "\t-a : autostart all cores (by default)\n"
67 "\t-e : don't autostart\n"
68 "\t-n : Create NULL devices instead of using PCI devices, useful together with -i\n"
69 "\t-m : list supported task modes and exit\n"
70 "\t-s : check configuration file syntax and exit\n"
71 "\t-i : check initialization sequence and exit\n"
72 "\t-u : Listen on UDS /tmp/prox.sock\n"
73 "\t-t : Listen on TCP port 8474\n"
74 "\t-q : Pass argument to Lua interpreter, useful to define variables\n"
75 "\t-w : define variable using syntax varname=value\n"
76 "\t takes precedence over variables defined in CONFIG_FILE\n"
77 "\t-k : Log statistics to file \"stats_dump\" in current directory\n"
78 "\t-d : Run as daemon, the parent process will block until PROX is not initialized\n"
79 "\t-z : Ignore CPU topology, implies -i\n"
80 "\t-r : Change initial screen refresh rate. If set to a lower than 0.001 seconds,\n"
81 "\t screen refreshing will be disabled\n"
86 static void check_mixed_normal_pipeline(void)
88 struct lcore_cfg *lconf = NULL;
89 uint32_t lcore_id = -1;
91 while (prox_core_next(&lcore_id, 0) == 0) {
92 lconf = &lcore_cfg[lcore_id];
94 int all_thread_nop = 1;
98 for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
99 struct task_args *targ = &lconf->targs[task_id];
100 l3 = !strcmp("l3", targ->sub_mode_str);
101 all_thread_nop = all_thread_nop && !l3 &&
102 targ->task_init->thread_x == thread_nop;
104 pipeline = pipeline || targ->task_init->thread_x == thread_pipeline;
105 generic = generic || targ->task_init->thread_x == thread_generic || l3;
107 PROX_PANIC(generic && pipeline, "Can't run both pipeline and normal thread on same core\n");
110 lconf->thread_x = thread_nop;
112 lconf->thread_x = thread_generic;
117 static void check_zero_rx(void)
119 struct lcore_cfg *lconf = NULL;
120 struct task_args *targ;
122 while (core_targ_next(&lconf, &targ, 0) == 0) {
123 if (targ->nb_rxports != 0) {
124 PROX_PANIC(task_init_flag_set(targ->task_init, TASK_FEATURE_NO_RX),
125 "\tCore %u task %u: rx_ports configured while mode %s does not use it\n", lconf->id, targ->id, targ->task_init->mode_str);
130 static void check_missing_rx(void)
132 struct lcore_cfg *lconf = NULL, *rx_lconf = NULL, *tx_lconf = NULL;
133 struct task_args *targ, *rx_targ = NULL, *tx_targ = NULL;
134 struct prox_port_cfg *port;
135 uint8_t port_id, rx_port_id, ok;
137 while (core_targ_next(&lconf, &targ, 0) == 0) {
138 PROX_PANIC((targ->flags & TASK_ARG_RX_RING) && targ->rx_rings[0] == 0 && !targ->tx_opt_ring_task,
139 "Configuration Error - Core %u task %u Receiving from ring, but nobody xmitting to this ring\n", lconf->id, targ->id);
140 if (targ->nb_rxports == 0 && targ->nb_rxrings == 0) {
141 PROX_PANIC(!task_init_flag_set(targ->task_init, TASK_FEATURE_NO_RX),
142 "\tCore %u task %u: no rx_ports and no rx_rings configured while required by mode %s\n", lconf->id, targ->id, targ->task_init->mode_str);
147 while (core_targ_next(&lconf, &targ, 0) == 0) {
148 if (strcmp(targ->sub_mode_str, "l3") != 0)
151 PROX_PANIC((targ->nb_rxports == 0) && (targ->nb_txports == 0), "L3 task must have a RX or a TX port\n");
152 // If the L3 sub_mode receives from a port, check that there is at least one core/task
153 // transmitting to this port in L3 sub_mode
154 for (uint8_t i = 0; i < targ->nb_rxports; ++i) {
155 rx_port_id = targ->rx_port_queue[i].port;
158 while (core_targ_next(&tx_lconf, &tx_targ, 0) == 0) {
159 if ((port_id = tx_targ->tx_port_queue[0].port) == OUT_DISCARD)
161 if ((rx_port_id == port_id) && (tx_targ->flags & TASK_ARG_L3)){
166 PROX_PANIC(ok == 0, "RX L3 sub mode for port %d on core %d task %d, but no core/task transmitting on that port\n", rx_port_id, lconf->id, targ->id);
169 // If the L3 sub_mode transmits to a port, check that there is at least one core/task
170 // receiving from that port in L3 sub_mode.
171 if ((port_id = targ->tx_port_queue[0].port) == OUT_DISCARD)
175 plog_info("\tCore %d task %d transmitting to port %d in L3 mode\n", lconf->id, targ->id, port_id);
176 while (core_targ_next(&rx_lconf, &rx_targ, 0) == 0) {
177 for (uint8_t i = 0; i < rx_targ->nb_rxports; ++i) {
178 rx_port_id = rx_targ->rx_port_queue[i].port;
179 if ((rx_port_id == port_id) && (rx_targ->flags & TASK_ARG_L3)){
185 plog_info("\tCore %d task %d has found core %d task %d receiving from port %d\n", lconf->id, targ->id, rx_lconf->id, rx_targ->id, port_id);
189 PROX_PANIC(ok == 0, "L3 sub mode for port %d on core %d task %d, but no core/task receiving on that port\n", port_id, lconf->id, targ->id);
193 static void check_cfg_consistent(void)
197 check_mixed_normal_pipeline();
200 static void plog_all_rings(void)
202 struct lcore_cfg *lconf = NULL;
203 struct task_args *targ;
205 while (core_targ_next(&lconf, &targ, 0) == 0) {
206 for (uint8_t ring_idx = 0; ring_idx < targ->nb_rxrings; ++ring_idx) {
207 plog_info("\tCore %u, task %u, rx_ring[%u] %p\n", lconf->id, targ->id, ring_idx, targ->rx_rings[ring_idx]);
212 static int chain_flag_state(struct task_args *targ, uint64_t flag, int is_set)
214 if (task_init_flag_set(targ->task_init, flag) == is_set)
219 for (uint32_t i = 0; i < targ->n_prev_tasks; ++i) {
220 ret = chain_flag_state(targ->prev_tasks[i], flag, is_set);
227 static int chain_flag_always_set(struct task_args *targ, uint64_t flag)
229 return (!chain_flag_state(targ, flag, 0));
232 static int chain_flag_never_set(struct task_args *targ, uint64_t flag)
234 return (!chain_flag_state(targ, flag, 1));
237 static int chain_flag_sometimes_set(struct task_args *targ, uint64_t flag)
239 return (chain_flag_state(targ, flag, 1));
242 static void configure_if_tx_queues(struct task_args *targ, uint8_t socket)
246 for (uint8_t i = 0; i < targ->nb_txports; ++i) {
247 if_port = targ->tx_port_queue[i].port;
249 PROX_PANIC(if_port == OUT_DISCARD, "port misconfigured, exiting\n");
251 PROX_PANIC(!prox_port_cfg[if_port].active, "\tPort %u not used, skipping...\n", if_port);
253 int dsocket = prox_port_cfg[if_port].socket;
254 if (dsocket != -1 && dsocket != socket) {
255 plog_warn("TX core on socket %d while device on socket %d\n", socket, dsocket);
258 if (prox_port_cfg[if_port].tx_ring[0] == '\0') { // Rings-backed port can use single queue
259 targ->tx_port_queue[i].queue = prox_port_cfg[if_port].n_txq;
260 prox_port_cfg[if_port].n_txq++;
262 prox_port_cfg[if_port].n_txq = 1;
263 targ->tx_port_queue[i].queue = 0;
265 /* By default OFFLOAD is enabled, but if the whole
266 chain has NOOFFLOADS set all the way until the
267 first task that receives from a port, it will be
268 disabled for the destination port. */
269 #if RTE_VERSION < RTE_VERSION_NUM(18,8,0,1)
270 if (chain_flag_always_set(targ, TASK_FEATURE_TXQ_FLAGS_NOOFFLOADS)) {
271 prox_port_cfg[if_port].tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOOFFLOADS;
274 if (chain_flag_always_set(targ, TASK_FEATURE_TXQ_FLAGS_NOOFFLOADS)) {
275 prox_port_cfg[if_port].requested_tx_offload &= ~(DEV_TX_OFFLOAD_IPV4_CKSUM | DEV_TX_OFFLOAD_UDP_CKSUM);
281 static void configure_if_rx_queues(struct task_args *targ, uint8_t socket)
283 struct prox_port_cfg *port;
284 for (int i = 0; i < targ->nb_rxports; i++) {
285 uint8_t if_port = targ->rx_port_queue[i].port;
287 if (if_port == OUT_DISCARD) {
291 port = &prox_port_cfg[if_port];
292 PROX_PANIC(!port->active, "Port %u not used, aborting...\n", if_port);
294 if(port->rx_ring[0] != '\0') {
298 // If the mbuf size (of the rx task) is not big enough, we might receive multiple segments
299 // This is usually the case when setting a big mtu size i.e. enabling jumbo frames.
300 // If the packets get transmitted, then multi segments will have to be enabled on the TX port
301 uint16_t max_frame_size = port->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + 2 * PROX_VLAN_TAG_SIZE;
302 if (max_frame_size + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM > targ->mbuf_size) {
303 targ->task_init->flag_features |= TASK_FEATURE_TXQ_FLAGS_MULTSEGS;
305 targ->rx_port_queue[i].queue = port->n_rxq;
306 port->pool[targ->rx_port_queue[i].queue] = targ->pool;
307 port->pool_size[targ->rx_port_queue[i].queue] = targ->nb_mbuf - 1;
310 int dsocket = port->socket;
311 if (dsocket != -1 && dsocket != socket) {
312 plog_warn("RX core on socket %d while device on socket %d\n", socket, dsocket);
317 static void configure_if_queues(void)
319 struct lcore_cfg *lconf = NULL;
320 struct task_args *targ;
323 while (core_targ_next(&lconf, &targ, 0) == 0) {
324 socket = rte_lcore_to_socket_id(lconf->id);
326 configure_if_rx_queues(targ, socket);
327 configure_if_tx_queues(targ, socket);
331 static void configure_tx_queue_flags(void)
333 struct lcore_cfg *lconf = NULL;
334 struct task_args *targ;
338 while (core_targ_next(&lconf, &targ, 0) == 0) {
339 socket = rte_lcore_to_socket_id(lconf->id);
340 for (uint8_t i = 0; i < targ->nb_txports; ++i) {
341 if_port = targ->tx_port_queue[i].port;
342 #if RTE_VERSION < RTE_VERSION_NUM(18,8,0,1)
343 /* Set the ETH_TXQ_FLAGS_NOREFCOUNT flag if none of
344 the tasks up to the task transmitting to the port
346 if (chain_flag_never_set(targ, TASK_FEATURE_TXQ_FLAGS_REFCOUNT)) {
347 prox_port_cfg[if_port].tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
350 /* Set the DEV_TX_OFFLOAD_MBUF_FAST_FREE flag if none of
351 the tasks up to the task transmitting to the port
352 use refcnt and per-queue all mbufs comes from the same mempool. */
353 if (chain_flag_never_set(targ, TASK_FEATURE_TXQ_FLAGS_REFCOUNT)) {
354 if (chain_flag_never_set(targ, TASK_FEATURE_TXQ_FLAGS_MULTIPLE_MEMPOOL))
355 prox_port_cfg[if_port].requested_tx_offload |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
362 static void configure_multi_segments(void)
364 struct lcore_cfg *lconf = NULL;
365 struct task_args *targ;
368 while (core_targ_next(&lconf, &targ, 0) == 0) {
369 for (uint8_t i = 0; i < targ->nb_txports; ++i) {
370 if_port = targ->tx_port_queue[i].port;
371 // Multi segment is disabled for most tasks. It is only enabled for tasks requiring big packets.
372 #if RTE_VERSION < RTE_VERSION_NUM(18,8,0,1)
373 // We can only enable "no multi segment" if no such task exists in the chain of tasks.
374 if (chain_flag_never_set(targ, TASK_FEATURE_TXQ_FLAGS_MULTSEGS)) {
375 prox_port_cfg[if_port].tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
378 // We enable "multi segment" if at least one task requires it in the chain of tasks.
379 if (chain_flag_sometimes_set(targ, TASK_FEATURE_TXQ_FLAGS_MULTSEGS)) {
380 prox_port_cfg[if_port].requested_tx_offload |= DEV_TX_OFFLOAD_MULTI_SEGS;
387 static const char *gen_ring_name(void)
389 static char retval[] = "XX";
390 static const char* ring_names =
391 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
392 "abcdefghijklmnopqrstuvwxyz"
393 "[\\]^_`!\"#$%&'()*+,-./:;<="
399 retval[0] = ring_names[idx % strlen(ring_names)];
400 idx /= strlen(ring_names);
401 retval[1] = idx ? ring_names[(idx - 1) % strlen(ring_names)] : 0;
408 struct ring_init_stats {
409 uint32_t n_pkt_rings;
410 uint32_t n_ctrl_rings;
411 uint32_t n_opt_rings;
414 static uint32_t ring_init_stats_total(const struct ring_init_stats *ris)
416 return ris->n_pkt_rings + ris->n_ctrl_rings + ris->n_opt_rings;
419 static uint32_t count_incoming_tasks(uint32_t lcore_worker, uint32_t dest_task)
421 struct lcore_cfg *lconf = NULL;
422 struct task_args *targ;
426 while (core_targ_next(&lconf, &targ, 0) == 0) {
427 for (uint8_t idxx = 0; idxx < MAX_PROTOCOLS; ++idxx) {
428 for (uint8_t ridx = 0; ridx < targ->core_task_set[idxx].n_elems; ++ridx) {
429 ct = targ->core_task_set[idxx].core_task[ridx];
431 if (dest_task == ct.task && lcore_worker == ct.core)
439 static struct rte_ring *get_existing_ring(uint32_t lcore_id, uint32_t task_id)
441 if (!prox_core_active(lcore_id, 0))
444 struct lcore_cfg *lconf = &lcore_cfg[lcore_id];
446 if (task_id >= lconf->n_tasks_all)
449 if (lconf->targs[task_id].nb_rxrings == 0)
452 return lconf->targs[task_id].rx_rings[0];
455 static struct rte_ring *init_ring_between_tasks(struct lcore_cfg *lconf, struct task_args *starg,
456 const struct core_task ct, uint8_t ring_idx, int idx,
457 struct ring_init_stats *ris)
460 struct rte_ring *ring = NULL;
461 struct lcore_cfg *lworker;
462 struct task_args *dtarg;
464 PROX_ASSERT(prox_core_active(ct.core, 0));
465 lworker = &lcore_cfg[ct.core];
467 /* socket used is the one that the sending core resides on */
468 socket = rte_lcore_to_socket_id(lconf->id);
470 plog_info("\t\tCreating ring on socket %u with size %u\n"
471 "\t\t\tsource core, task and socket = %u, %u, %u\n"
472 "\t\t\tdestination core, task and socket = %u, %u, %u\n"
473 "\t\t\tdestination worker id = %u\n",
474 socket, starg->ring_size,
475 lconf->id, starg->id, socket,
476 ct.core, ct.task, rte_lcore_to_socket_id(ct.core),
480 struct rte_ring **dring = NULL;
482 if (ct.type == CTRL_TYPE_MSG)
483 dring = &lworker->ctrl_rings_m[ct.task];
484 else if (ct.type == CTRL_TYPE_PKT) {
485 dring = &lworker->ctrl_rings_p[ct.task];
486 starg->flags |= TASK_ARG_CTRL_RINGS_P;
490 ring = rte_ring_create(gen_ring_name(), starg->ring_size, socket, RING_F_SC_DEQ);
493 PROX_PANIC(ring == NULL, "Cannot create ring to connect I/O core %u with worker core %u\n", lconf->id, ct.core);
495 starg->tx_rings[starg->tot_n_txrings_inited] = ring;
496 starg->tot_n_txrings_inited++;
498 if (lconf->id == prox_cfg.master) {
499 ctrl_rings[ct.core*MAX_TASKS_PER_CORE + ct.task] = ring;
500 } else if (ct.core == prox_cfg.master) {
501 starg->ctrl_plane_ring = ring;
504 plog_info("\t\t\tCore %u task %u to -> core %u task %u ctrl_ring %s %p %s\n",
505 lconf->id, starg->id, ct.core, ct.task, ct.type == CTRL_TYPE_PKT?
506 "pkt" : "msg", ring, ring->name);
511 dtarg = &lworker->targs[ct.task];
512 lworker->targs[ct.task].worker_thread_id = ring_idx;
513 PROX_ASSERT(dtarg->flags & TASK_ARG_RX_RING);
514 PROX_ASSERT(ct.task < lworker->n_tasks_all);
516 /* If all the following conditions are met, the ring can be
518 if (!task_is_master(starg) && !task_is_master(dtarg) && starg->lconf->id == dtarg->lconf->id &&
519 starg->nb_txrings == 1 && idx == 0 && dtarg->task &&
520 dtarg->tot_rxrings == 1 && starg->task == dtarg->task - 1) {
521 plog_info("\t\tOptimizing away ring on core %u from task %u to task %u\n",
522 dtarg->lconf->id, starg->task, dtarg->task);
523 /* No need to set up ws_mbuf. */
524 starg->tx_opt_ring = 1;
525 /* During init of destination task, the buffer in the
526 source task will be initialized. */
527 dtarg->tx_opt_ring_task = starg;
533 int ring_created = 1;
534 /* Only create multi-producer rings if configured to do so AND
535 there is only one task sending to the task */
536 if ((prox_cfg.flags & DSF_MP_RINGS && count_incoming_tasks(ct.core, ct.task) > 1)
537 || (prox_cfg.flags & DSF_ENABLE_BYPASS)) {
538 ring = get_existing_ring(ct.core, ct.task);
541 plog_info("\t\tCore %u task %u creatign MP ring %p to core %u task %u\n",
542 lconf->id, starg->id, ring, ct.core, ct.task);
546 ring = rte_ring_create(gen_ring_name(), starg->ring_size, socket, RING_F_SC_DEQ);
547 plog_info("\t\tCore %u task %u using MP ring %p from core %u task %u\n",
548 lconf->id, starg->id, ring, ct.core, ct.task);
552 ring = rte_ring_create(gen_ring_name(), starg->ring_size, socket, RING_F_SP_ENQ | RING_F_SC_DEQ);
554 PROX_PANIC(ring == NULL, "Cannot create ring to connect I/O core %u with worker core %u\n", lconf->id, ct.core);
556 starg->tx_rings[starg->tot_n_txrings_inited] = ring;
557 starg->tot_n_txrings_inited++;
560 PROX_ASSERT(dtarg->nb_rxrings < MAX_RINGS_PER_TASK);
561 dtarg->rx_rings[dtarg->nb_rxrings] = ring;
563 if (dtarg->nb_rxrings > 1)
564 dtarg->task_init->flag_features |= TASK_FEATURE_TXQ_FLAGS_MULTIPLE_MEMPOOL;
566 dtarg->nb_slave_threads = starg->core_task_set[idx].n_elems;
567 dtarg->lb_friend_core = lconf->id;
568 dtarg->lb_friend_task = starg->id;
569 plog_info("\t\tWorker thread %d has core %d, task %d as a lb friend\n", ct.core, lconf->id, starg->id);
570 plog_info("\t\tCore %u task %u tx_ring[%u] -> core %u task %u rx_ring[%u] %p %s %u WT\n",
571 lconf->id, starg->id, ring_idx, ct.core, ct.task, dtarg->nb_rxrings, ring, ring->name,
572 dtarg->nb_slave_threads);
577 static void init_rings(void)
579 struct lcore_cfg *lconf = NULL;
580 struct task_args *starg;
581 struct ring_init_stats ris = {0};
583 while (core_targ_next(&lconf, &starg, 1) == 0) {
584 plog_info("\t*** Initializing rings on core %u, task %u ***\n", lconf->id, starg->id);
585 for (uint8_t idx = 0; idx < MAX_PROTOCOLS; ++idx) {
586 for (uint8_t ring_idx = 0; ring_idx < starg->core_task_set[idx].n_elems; ++ring_idx) {
587 PROX_ASSERT(ring_idx < MAX_WT_PER_LB);
588 PROX_ASSERT(starg->tot_n_txrings_inited < MAX_RINGS_PER_TASK);
590 struct core_task ct = starg->core_task_set[idx].core_task[ring_idx];
591 init_ring_between_tasks(lconf, starg, ct, ring_idx, idx, &ris);
596 plog_info("\tInitialized %d rings:\n"
597 "\t\tNumber of packet rings: %u\n"
598 "\t\tNumber of control rings: %u\n"
599 "\t\tNumber of optimized rings: %u\n",
600 ring_init_stats_total(&ris),
606 struct prox_port_cfg *port;
607 while (core_targ_next(&lconf, &starg, 1) == 0) {
608 if ((starg->task_init) && (starg->flags & TASK_ARG_L3)) {
610 ct.core = prox_cfg.master;
612 ct.type = CTRL_TYPE_PKT;
613 struct rte_ring *rx_ring = init_ring_between_tasks(lconf, starg, ct, 0, 0, &ris);
616 ct.task = starg->id;;
617 struct rte_ring *tx_ring = init_ring_between_tasks(lcore_cfg, lcore_cfg[prox_cfg.master].targs, ct, 0, 0, &ris);
622 static void shuffle_mempool(struct rte_mempool* mempool, uint32_t nb_mbuf)
624 struct rte_mbuf** pkts = prox_zmalloc(nb_mbuf * sizeof(*pkts), rte_socket_id());
627 while ((got < nb_mbuf) && (rte_mempool_get_bulk(mempool, (void**)(pkts + got), 1) == 0))
634 idx = rand() % nb_mbuf;
635 } while (pkts[idx] == 0);
637 rte_mempool_put_bulk(mempool, (void**)&pkts[idx], 1);
644 static void set_mbuf_size(struct task_args *targ)
646 /* mbuf size can be set
647 * - from config file (highest priority, overwriting any other config) - should only be used as workaround
648 * - defaulted to MBUF_SIZE.
649 * Except if set explicitely, ensure that size is big enough for vmxnet3 driver
654 targ->mbuf_size = MBUF_SIZE;
655 struct prox_port_cfg *port;
656 uint16_t max_frame_size = 0, min_buffer_size = 0;
658 for (int i = 0; i < targ->nb_rxports; i++) {
659 uint8_t if_port = targ->rx_port_queue[i].port;
661 if (if_port == OUT_DISCARD) {
664 port = &prox_port_cfg[if_port];
665 if (max_frame_size < port->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + 2 * PROX_VLAN_TAG_SIZE)
666 max_frame_size = port->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + 2 * PROX_VLAN_TAG_SIZE;
667 if (min_buffer_size < port->min_rx_bufsize)
668 min_buffer_size = port->min_rx_bufsize;
670 // Check whether we receive from i40e. This driver have extra mbuf size requirements
671 if (strcmp(port->short_name, "i40e") == 0)
675 // i40e supports a maximum of 5 descriptors chained
676 uint16_t required_mbuf_size = RTE_ALIGN(max_frame_size / 5, 128) + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM;
677 if (required_mbuf_size > targ->mbuf_size) {
678 targ->mbuf_size = required_mbuf_size;
679 plog_info("\t\tSetting mbuf_size to %u to support frame_size %u\n", targ->mbuf_size, max_frame_size);
682 if (min_buffer_size > targ->mbuf_size) {
683 plog_warn("Mbuf size might be too small. This might result in packet segmentation and memory leak\n");
688 static void setup_mempools_unique_per_socket(void)
692 struct lcore_cfg *lconf = NULL;
693 struct task_args *targ;
695 struct rte_mempool *pool[MAX_SOCKETS];
696 uint32_t mbuf_count[MAX_SOCKETS] = {0};
697 uint32_t nb_cache_mbuf[MAX_SOCKETS] = {0};
698 uint32_t mbuf_size[MAX_SOCKETS] = {0};
700 while (core_targ_next_early(&lconf, &targ, 0) == 0) {
701 PROX_PANIC(targ->task_init == NULL, "task_init = NULL, is mode specified for core %d, task %d ?\n", lconf->id, targ->id);
702 uint8_t socket = rte_lcore_to_socket_id(lconf->id);
703 PROX_ASSERT(socket < MAX_SOCKETS);
706 if (targ->rx_port_queue[0].port != OUT_DISCARD) {
707 struct prox_port_cfg* port_cfg = &prox_port_cfg[targ->rx_port_queue[0].port];
708 PROX_ASSERT(targ->nb_mbuf != 0);
709 mbuf_count[socket] += targ->nb_mbuf;
710 if (nb_cache_mbuf[socket] == 0)
711 nb_cache_mbuf[socket] = targ->nb_cache_mbuf;
713 PROX_PANIC(nb_cache_mbuf[socket] != targ->nb_cache_mbuf,
714 "all mbuf_cache must have the same size if using a unique mempool per socket\n");
716 if (mbuf_size[socket] == 0)
717 mbuf_size[socket] = targ->mbuf_size;
719 PROX_PANIC(mbuf_size[socket] != targ->mbuf_size,
720 "all mbuf_size must have the same size if using a unique mempool per socket\n");
724 for (int i = 0 ; i < MAX_SOCKETS; i++) {
725 if (mbuf_count[i] != 0) {
726 sprintf(name, "socket_%u_pool", i);
727 pool[i] = rte_mempool_create(name,
728 mbuf_count[i] - 1, mbuf_size[i],
730 sizeof(struct rte_pktmbuf_pool_private),
731 rte_pktmbuf_pool_init, NULL,
732 prox_pktmbuf_init, NULL,
734 PROX_PANIC(pool[i] == NULL, "\t\tError: cannot create mempool for socket %u\n", i);
735 plog_info("\t\tMempool %p size = %u * %u cache %u, socket %d\n", pool[i],
736 mbuf_count[i], mbuf_size[i], nb_cache_mbuf[i], i);
738 if (prox_cfg.flags & DSF_SHUFFLE) {
739 shuffle_mempool(pool[i], mbuf_count[i]);
745 while (core_targ_next_early(&lconf, &targ, 0) == 0) {
746 uint8_t socket = rte_lcore_to_socket_id(lconf->id);
748 if (targ->rx_port_queue[0].port != OUT_DISCARD) {
749 /* use this pool for the interface that the core is receiving from */
750 /* If one core receives from multiple ports, all the ports use the same mempool */
751 targ->pool = pool[socket];
752 /* Set the number of mbuf to the number of the unique mempool, so that the used and free work */
753 targ->nb_mbuf = mbuf_count[socket];
754 plog_info("\t\tMempool %p size = %u * %u cache %u, socket %d\n", targ->pool,
755 targ->nb_mbuf, mbuf_size[socket], targ->nb_cache_mbuf, socket);
760 static void setup_mempool_for_rx_task(struct lcore_cfg *lconf, struct task_args *targ)
762 const uint8_t socket = rte_lcore_to_socket_id(lconf->id);
763 struct prox_port_cfg *port_cfg = &prox_port_cfg[targ->rx_port_queue[0].port];
764 const struct rte_memzone *mz;
765 struct rte_mempool *mp = NULL;
767 char memzone_name[64];
772 /* allocate memory pool for packets */
773 PROX_ASSERT(targ->nb_mbuf != 0);
775 if (targ->pool_name[0] == '\0') {
776 sprintf(name, "core_%u_port_%u_pool", lconf->id, targ->id);
779 snprintf(memzone_name, sizeof(memzone_name)-1, "MP_%s", targ->pool_name);
780 mz = rte_memzone_lookup(memzone_name);
783 mp = (struct rte_mempool*)mz->addr;
785 targ->nb_mbuf = mp->size;
789 #ifdef RTE_LIBRTE_IVSHMEM_FALSE
790 if (mz != NULL && mp != NULL && mp->phys_addr != mz->ioremap_addr) {
791 /* Init mbufs with ioremap_addr for dma */
792 mp->phys_addr = mz->ioremap_addr;
793 mp->elt_pa[0] = mp->phys_addr + (mp->elt_va_start - (uintptr_t)mp);
795 struct prox_pktmbuf_reinit_args init_args;
797 init_args.lconf = lconf;
799 uint32_t elt_sz = mp->elt_size + mp->header_size + mp->trailer_size;
800 rte_mempool_obj_iter((void*)mp->elt_va_start, mp->size, elt_sz, 1,
801 mp->elt_pa, mp->pg_num, mp->pg_shift, prox_pktmbuf_reinit, &init_args);
805 /* Use this pool for the interface that the core is
806 receiving from if one core receives from multiple
807 ports, all the ports use the same mempool */
808 if (targ->pool == NULL) {
809 plog_info("\t\tCreating mempool with name '%s'\n", name);
810 targ->pool = rte_mempool_create(name,
811 targ->nb_mbuf - 1, targ->mbuf_size,
813 sizeof(struct rte_pktmbuf_pool_private),
814 rte_pktmbuf_pool_init, NULL,
815 prox_pktmbuf_init, lconf,
819 PROX_PANIC(targ->pool == NULL,
820 "\t\tError: cannot create mempool for core %u port %u: %s\n", lconf->id, targ->id, rte_strerror(rte_errno));
822 plog_info("\t\tMempool %p size = %u * %u cache %u, socket %d\n", targ->pool,
823 targ->nb_mbuf, targ->mbuf_size, targ->nb_cache_mbuf, socket);
824 if (prox_cfg.flags & DSF_SHUFFLE) {
825 shuffle_mempool(targ->pool, targ->nb_mbuf);
829 static void setup_mempools_multiple_per_socket(void)
831 struct lcore_cfg *lconf = NULL;
832 struct task_args *targ;
834 while (core_targ_next_early(&lconf, &targ, 0) == 0) {
835 PROX_PANIC(targ->task_init == NULL, "task_init = NULL, is mode specified for core %d, task %d ?\n", lconf->id, targ->id);
836 if (targ->rx_port_queue[0].port == OUT_DISCARD)
838 setup_mempool_for_rx_task(lconf, targ);
842 static void setup_mempools(void)
844 if (prox_cfg.flags & UNIQUE_MEMPOOL_PER_SOCKET)
845 setup_mempools_unique_per_socket();
847 setup_mempools_multiple_per_socket();
850 static void set_task_lconf(void)
852 struct lcore_cfg *lconf;
853 uint32_t lcore_id = -1;
855 while(prox_core_next(&lcore_id, 1) == 0) {
856 lconf = &lcore_cfg[lcore_id];
857 for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
858 lconf->targs[task_id].lconf = lconf;
863 static void set_dest_threads(void)
865 struct lcore_cfg *lconf = NULL;
866 struct task_args *targ;
868 while (core_targ_next(&lconf, &targ, 0) == 0) {
869 for (uint8_t idx = 0; idx < MAX_PROTOCOLS; ++idx) {
870 for (uint8_t ring_idx = 0; ring_idx < targ->core_task_set[idx].n_elems; ++ring_idx) {
871 struct core_task ct = targ->core_task_set[idx].core_task[ring_idx];
873 struct task_args *dest_task = core_targ_get(ct.core, ct.task);
874 dest_task->prev_tasks[dest_task->n_prev_tasks++] = targ;
880 static void setup_all_task_structs_early_init(void)
882 struct lcore_cfg *lconf = NULL;
883 struct task_args *targ;
885 plog_info("\t*** Calling early init on all tasks ***\n");
886 while (core_targ_next(&lconf, &targ, 0) == 0) {
887 if (targ->task_init->early_init) {
888 targ->task_init->early_init(targ);
893 static void setup_all_task_structs(void)
895 struct lcore_cfg *lconf;
896 uint32_t lcore_id = -1;
897 struct task_base *tmaster = NULL;
899 while(prox_core_next(&lcore_id, 1) == 0) {
900 lconf = &lcore_cfg[lcore_id];
901 for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
902 if (task_is_master(&lconf->targs[task_id])) {
903 plog_info("\tInitializing MASTER struct for core %d task %d\n", lcore_id, task_id);
904 lconf->tasks_all[task_id] = init_task_struct(&lconf->targs[task_id]);
905 tmaster = lconf->tasks_all[task_id];
909 PROX_PANIC(tmaster == NULL, "Can't initialize master task\n");
912 while(prox_core_next(&lcore_id, 1) == 0) {
913 lconf = &lcore_cfg[lcore_id];
914 plog_info("\tInitializing struct for core %d with %d task\n", lcore_id, lconf->n_tasks_all);
915 for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
916 if (!task_is_master(&lconf->targs[task_id])) {
917 plog_info("\tInitializing struct for core %d task %d\n", lcore_id, task_id);
918 lconf->targs[task_id].tmaster = tmaster;
919 lconf->tasks_all[task_id] = init_task_struct(&lconf->targs[task_id]);
925 static void init_port_activate(void)
927 struct lcore_cfg *lconf = NULL;
928 struct task_args *targ;
931 while (core_targ_next_early(&lconf, &targ, 0) == 0) {
932 for (int i = 0; i < targ->nb_rxports; i++) {
933 port_id = targ->rx_port_queue[i].port;
934 prox_port_cfg[port_id].active = 1;
937 for (int i = 0; i < targ->nb_txports; i++) {
938 port_id = targ->tx_port_queue[i].port;
939 prox_port_cfg[port_id].active = 1;
944 /* Initialize cores and allocate mempools */
945 static void init_lcores(void)
947 struct lcore_cfg *lconf = 0;
948 uint32_t lcore_id = -1;
950 while(prox_core_next(&lcore_id, 0) == 0) {
951 uint8_t socket = rte_lcore_to_socket_id(lcore_id);
952 PROX_PANIC(socket + 1 > MAX_SOCKETS, "Can't configure core %u (on socket %u). MAX_SOCKET is set to %d\n", lcore_id, socket, MAX_SOCKETS);
955 /* need to allocate mempools as the first thing to use the lowest possible address range */
956 plog_info("=== Initializing mempools ===\n");
959 lcore_cfg_alloc_hp();
964 plog_info("=== Initializing port addresses ===\n");
967 plog_info("=== Initializing queue numbers on cores ===\n");
968 configure_if_queues();
970 plog_info("=== Initializing rings on cores ===\n");
973 configure_multi_segments();
974 configure_tx_queue_flags();
976 plog_info("=== Checking configuration consistency ===\n");
977 check_cfg_consistent();
982 static int setup_prox(int argc, char **argv)
984 if (prox_read_config_file() != 0 ||
985 prox_setup_rte(argv[0]) != 0) {
989 if (prox_cfg.flags & DSF_CHECK_SYNTAX) {
990 plog_info("=== Configuration file syntax has been checked ===\n\n");
994 init_port_activate();
995 plog_info("=== Initializing rte devices ===\n");
996 if (!(prox_cfg.flags & DSF_USE_DUMMY_DEVICES))
998 init_rte_dev(prox_cfg.flags & DSF_USE_DUMMY_DEVICES);
999 plog_info("=== Calibrating TSC overhead ===\n");
1001 plog_info("\tTSC running at %"PRIu64" Hz\n", rte_get_tsc_hz());
1004 plog_info("=== Initializing ports ===\n");
1007 setup_all_task_structs_early_init();
1008 plog_info("=== Initializing tasks ===\n");
1009 setup_all_task_structs();
1011 if (prox_cfg.logbuf_size) {
1012 prox_cfg.logbuf = prox_zmalloc(prox_cfg.logbuf_size, rte_socket_id());
1013 PROX_PANIC(prox_cfg.logbuf == NULL, "Failed to allocate memory for logbuf with size = %d\n", prox_cfg.logbuf_size);
1016 if (prox_cfg.flags & DSF_CHECK_INIT) {
1017 plog_info("=== Initialization sequence completed ===\n\n");
1021 /* Current way that works to disable DPDK logging */
1022 FILE *f = fopen("/dev/null", "r");
1023 rte_openlog_stream(f);
1024 plog_info("=== PROX started ===\n");
1028 static int success = 0;
1029 static void siguser_handler(int signal)
1031 if (signal == SIGUSR1)
1037 static void sigabrt_handler(__attribute__((unused)) int signum)
1039 /* restore default disposition for SIGABRT and SIGPIPE */
1040 signal(SIGABRT, SIG_DFL);
1041 signal(SIGPIPE, SIG_DFL);
1043 /* ignore further Ctrl-C */
1044 signal(SIGINT, SIG_IGN);
1046 /* more drastic exit on tedious termination signal */
1047 plog_info("Aborting...\n");
1048 if (lcore_cfg != NULL) {
1050 pthread_t thread_id, tid0, tid = pthread_self();
1051 memset(&tid0, 0, sizeof(tid0));
1053 /* cancel all threads except current one */
1055 while (prox_core_next(&lcore_id, 1) == 0) {
1056 thread_id = lcore_cfg[lcore_id].thread_id;
1057 if (pthread_equal(thread_id, tid0))
1059 if (pthread_equal(thread_id, tid))
1061 pthread_cancel(thread_id);
1064 /* wait for cancelled threads to terminate */
1066 while (prox_core_next(&lcore_id, 1) == 0) {
1067 thread_id = lcore_cfg[lcore_id].thread_id;
1068 if (pthread_equal(thread_id, tid0))
1070 if (pthread_equal(thread_id, tid))
1072 pthread_join(thread_id, NULL);
1079 /* close ports on termination signal */
1080 close_ports_atexit();
1086 static void sigterm_handler(int signum)
1088 /* abort on second Ctrl-C */
1089 if (signum == SIGINT)
1090 signal(SIGINT, sigabrt_handler);
1092 /* gracefully quit on harmless termination signal */
1093 /* ports will subsequently get closed at resulting exit */
1097 int main(int argc, char **argv)
1099 /* set en_US locale to print big numbers with ',' */
1100 setlocale(LC_NUMERIC, "en_US.utf-8");
1102 if (prox_parse_args(argc, argv) != 0){
1103 prox_usage(argv[0]);
1106 plog_init(prox_cfg.log_name, prox_cfg.log_name_pid);
1107 plog_info("=== " PROGRAM_NAME " " VERSION_STR " ===\n");
1108 plog_info("\tUsing DPDK %s\n", rte_version() + sizeof(RTE_VER_PREFIX));
1111 if (prox_cfg.flags & DSF_LIST_TASK_MODES) {
1112 /* list supported task modes and exit */
1114 return EXIT_SUCCESS;
1117 /* close ports at normal exit */
1118 atexit(close_ports_atexit);
1119 /* gracefully quit on harmless termination signals */
1120 signal(SIGHUP, sigterm_handler);
1121 signal(SIGINT, sigterm_handler);
1122 signal(SIGQUIT, sigterm_handler);
1123 signal(SIGTERM, sigterm_handler);
1124 signal(SIGUSR1, sigterm_handler);
1125 signal(SIGUSR2, sigterm_handler);
1126 /* more drastic exit on tedious termination signals */
1127 signal(SIGABRT, sigabrt_handler);
1128 signal(SIGPIPE, sigabrt_handler);
1130 if (prox_cfg.flags & DSF_DAEMON) {
1131 signal(SIGUSR1, siguser_handler);
1132 signal(SIGUSR2, siguser_handler);
1133 plog_info("=== Running in Daemon mode ===\n");
1134 plog_info("\tForking child and waiting for setup completion\n");
1136 pid_t ppid = getpid();
1139 plog_err("Failed to fork process to run in daemon mode\n");
1140 return EXIT_FAILURE;
1148 kill(ppid, SIGUSR2);
1149 return EXIT_FAILURE;
1151 if (setup_prox(argc, argv) != 0) {
1152 kill(ppid, SIGUSR2);
1153 return EXIT_FAILURE;
1156 kill(ppid, SIGUSR1);
1157 run(prox_cfg.flags);
1158 return EXIT_SUCCESS;
1162 /* Before exiting the parent, wait until the
1163 child process has finished setting up */
1165 if (prox_cfg.logbuf) {
1166 file_print(prox_cfg.logbuf);
1168 return success? EXIT_SUCCESS : EXIT_FAILURE;
1172 if (setup_prox(argc, argv) != 0)
1173 return EXIT_FAILURE;
1174 run(prox_cfg.flags);
1175 return EXIT_SUCCESS;