1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2017 Intel Corporation
12 #ifndef RTE_EXEC_ENV_WINDOWS
15 #include <sys/types.h>
19 #include <sys/queue.h>
26 #include <rte_common.h>
27 #include <rte_errno.h>
28 #include <rte_byteorder.h>
30 #include <rte_debug.h>
31 #include <rte_cycles.h>
32 #include <rte_memory.h>
33 #include <rte_memcpy.h>
34 #include <rte_launch.h>
36 #include <rte_alarm.h>
37 #include <rte_per_lcore.h>
38 #include <rte_lcore.h>
39 #include <rte_atomic.h>
40 #include <rte_branch_prediction.h>
41 #include <rte_mempool.h>
42 #include <rte_malloc.h>
44 #include <rte_mbuf_pool_ops.h>
45 #include <rte_interrupts.h>
47 #include <rte_ether.h>
48 #include <rte_ethdev.h>
50 #include <rte_string_fns.h>
52 #include <rte_pmd_ixgbe.h>
55 #include <rte_pdump.h>
58 #include <rte_metrics.h>
59 #ifdef RTE_LIB_BITRATESTATS
60 #include <rte_bitrate.h>
62 #ifdef RTE_LIB_LATENCYSTATS
63 #include <rte_latencystats.h>
65 #ifdef RTE_EXEC_ENV_WINDOWS
72 /* FreeBSD may not have MAP_HUGETLB (in fact, it probably doesn't) */
73 #define HUGE_FLAG (0x40000)
75 #define HUGE_FLAG MAP_HUGETLB
78 #ifndef MAP_HUGE_SHIFT
79 /* older kernels (or FreeBSD) will not have this define */
80 #define HUGE_SHIFT (26)
82 #define HUGE_SHIFT MAP_HUGE_SHIFT
85 #define EXTMEM_HEAP_NAME "extmem"
86 #define EXTBUF_ZONE_SIZE RTE_PGSIZE_2M
88 uint16_t verbose_level = 0; /**< Silent by default. */
89 int testpmd_logtype; /**< Log type for testpmd logs */
91 /* use main core for command line ? */
92 uint8_t interactive = 0;
93 uint8_t auto_start = 0;
95 char cmdline_filename[PATH_MAX] = {0};
98 * NUMA support configuration.
99 * When set, the NUMA support attempts to dispatch the allocation of the
100 * RX and TX memory rings, and of the DMA memory buffers (mbufs) for the
101 * probed ports among the CPU sockets 0 and 1.
102 * Otherwise, all memory is allocated from CPU socket 0.
104 uint8_t numa_support = 1; /**< numa enabled by default */
107 * In UMA mode,all memory is allocated from socket 0 if --socket-num is
110 uint8_t socket_num = UMA_NO_CONFIG;
113 * Select mempool allocation type:
114 * - native: use regular DPDK memory
115 * - anon: use regular DPDK memory to create mempool, but populate using
116 * anonymous memory (may not be IOVA-contiguous)
117 * - xmem: use externally allocated hugepage memory
119 uint8_t mp_alloc_type = MP_ALLOC_NATIVE;
122 * Store specified sockets on which memory pool to be used by ports
125 uint8_t port_numa[RTE_MAX_ETHPORTS];
128 * Store specified sockets on which RX ring to be used by ports
131 uint8_t rxring_numa[RTE_MAX_ETHPORTS];
134 * Store specified sockets on which TX ring to be used by ports
137 uint8_t txring_numa[RTE_MAX_ETHPORTS];
140 * Record the Ethernet address of peer target ports to which packets are
142 * Must be instantiated with the ethernet addresses of peer traffic generator
145 struct rte_ether_addr peer_eth_addrs[RTE_MAX_ETHPORTS];
146 portid_t nb_peer_eth_addrs = 0;
149 * Probed Target Environment.
151 struct rte_port *ports; /**< For all probed ethernet ports. */
152 portid_t nb_ports; /**< Number of probed ethernet ports. */
153 struct fwd_lcore **fwd_lcores; /**< For all probed logical cores. */
154 lcoreid_t nb_lcores; /**< Number of probed logical cores. */
156 portid_t ports_ids[RTE_MAX_ETHPORTS]; /**< Store all port ids. */
159 * Test Forwarding Configuration.
160 * nb_fwd_lcores <= nb_cfg_lcores <= nb_lcores
161 * nb_fwd_ports <= nb_cfg_ports <= nb_ports
163 lcoreid_t nb_cfg_lcores; /**< Number of configured logical cores. */
164 lcoreid_t nb_fwd_lcores; /**< Number of forwarding logical cores. */
165 portid_t nb_cfg_ports; /**< Number of configured ports. */
166 portid_t nb_fwd_ports; /**< Number of forwarding ports. */
168 unsigned int fwd_lcores_cpuids[RTE_MAX_LCORE]; /**< CPU ids configuration. */
169 portid_t fwd_ports_ids[RTE_MAX_ETHPORTS]; /**< Port ids configuration. */
171 struct fwd_stream **fwd_streams; /**< For each RX queue of each port. */
172 streamid_t nb_fwd_streams; /**< Is equal to (nb_ports * nb_rxq). */
175 * Forwarding engines.
177 struct fwd_engine * fwd_engines[] = {
187 &five_tuple_swap_fwd_engine,
188 #ifdef RTE_LIBRTE_IEEE1588
189 &ieee1588_fwd_engine,
194 struct rte_mempool *mempools[RTE_MAX_NUMA_NODES * MAX_SEGS_BUFFER_SPLIT];
195 uint16_t mempool_flags;
197 struct fwd_config cur_fwd_config;
198 struct fwd_engine *cur_fwd_eng = &io_fwd_engine; /**< IO mode by default. */
199 uint32_t retry_enabled;
200 uint32_t burst_tx_delay_time = BURST_TX_WAIT_US;
201 uint32_t burst_tx_retry_num = BURST_TX_RETRIES;
203 uint32_t mbuf_data_size_n = 1; /* Number of specified mbuf sizes. */
204 uint16_t mbuf_data_size[MAX_SEGS_BUFFER_SPLIT] = {
205 DEFAULT_MBUF_DATA_SIZE
206 }; /**< Mbuf data space size. */
207 uint32_t param_total_num_mbufs = 0; /**< number of mbufs in all pools - if
208 * specified on command-line. */
209 uint16_t stats_period; /**< Period to show statistics (disabled by default) */
212 * In container, it cannot terminate the process which running with 'stats-period'
213 * option. Set flag to exit stats period loop after received SIGINT/SIGTERM.
218 * Configuration of packet segments used to scatter received packets
219 * if some of split features is configured.
221 uint16_t rx_pkt_seg_lengths[MAX_SEGS_BUFFER_SPLIT];
222 uint8_t rx_pkt_nb_segs; /**< Number of segments to split */
223 uint16_t rx_pkt_seg_offsets[MAX_SEGS_BUFFER_SPLIT];
224 uint8_t rx_pkt_nb_offs; /**< Number of specified offsets */
227 * Configuration of packet segments used by the "txonly" processing engine.
229 uint16_t tx_pkt_length = TXONLY_DEF_PACKET_LEN; /**< TXONLY packet length. */
230 uint16_t tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT] = {
231 TXONLY_DEF_PACKET_LEN,
233 uint8_t tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY packets */
235 enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;
236 /**< Split policy for packets to TX. */
238 uint8_t txonly_multi_flow;
239 /**< Whether multiple flows are generated in TXONLY mode. */
241 uint32_t tx_pkt_times_inter;
242 /**< Timings for send scheduling in TXONLY mode, time between bursts. */
244 uint32_t tx_pkt_times_intra;
245 /**< Timings for send scheduling in TXONLY mode, time between packets. */
247 uint16_t nb_pkt_per_burst = DEF_PKT_BURST; /**< Number of packets per burst. */
248 uint16_t nb_pkt_flowgen_clones; /**< Number of Tx packet clones to send in flowgen mode. */
249 int nb_flows_flowgen = 1024; /**< Number of flows in flowgen mode. */
250 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
252 /* current configuration is in DCB or not,0 means it is not in DCB mode */
253 uint8_t dcb_config = 0;
256 * Configurable number of RX/TX queues.
258 queueid_t nb_hairpinq; /**< Number of hairpin queues per port. */
259 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
260 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
263 * Configurable number of RX/TX ring descriptors.
264 * Defaults are supplied by drivers via ethdev.
266 #define RTE_TEST_RX_DESC_DEFAULT 0
267 #define RTE_TEST_TX_DESC_DEFAULT 0
268 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
269 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
271 #define RTE_PMD_PARAM_UNSET -1
273 * Configurable values of RX and TX ring threshold registers.
276 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
277 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
278 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
280 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
281 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
282 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
285 * Configurable value of RX free threshold.
287 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
290 * Configurable value of RX drop enable.
292 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
295 * Configurable value of TX free threshold.
297 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
300 * Configurable value of TX RS bit threshold.
302 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
305 * Configurable value of buffered packets before sending.
307 uint16_t noisy_tx_sw_bufsz;
310 * Configurable value of packet buffer timeout.
312 uint16_t noisy_tx_sw_buf_flush_time;
315 * Configurable value for size of VNF internal memory area
316 * used for simulating noisy neighbour behaviour
318 uint64_t noisy_lkup_mem_sz;
321 * Configurable value of number of random writes done in
322 * VNF simulation memory area.
324 uint64_t noisy_lkup_num_writes;
327 * Configurable value of number of random reads done in
328 * VNF simulation memory area.
330 uint64_t noisy_lkup_num_reads;
333 * Configurable value of number of random reads/writes done in
334 * VNF simulation memory area.
336 uint64_t noisy_lkup_num_reads_writes;
339 * Receive Side Scaling (RSS) configuration.
341 uint64_t rss_hf = ETH_RSS_IP; /* RSS IP by default. */
344 * Port topology configuration
346 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
349 * Avoids to flush all the RX streams before starts forwarding.
351 uint8_t no_flush_rx = 0; /* flush by default */
354 * Flow API isolated mode.
356 uint8_t flow_isolate_all;
359 * Avoids to check link status when starting/stopping a port.
361 uint8_t no_link_check = 0; /* check by default */
364 * Don't automatically start all ports in interactive mode.
366 uint8_t no_device_start = 0;
369 * Enable link status change notification
371 uint8_t lsc_interrupt = 1; /* enabled by default */
374 * Enable device removal notification.
376 uint8_t rmv_interrupt = 1; /* enabled by default */
378 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
380 /* After attach, port setup is called on event or by iterator */
381 bool setup_on_probe_event = true;
383 /* Clear ptypes on port initialization. */
384 uint8_t clear_ptypes = true;
386 /* Hairpin ports configuration mode. */
387 uint16_t hairpin_mode;
389 /* Pretty printing of ethdev events */
390 static const char * const eth_event_desc[] = {
391 [RTE_ETH_EVENT_UNKNOWN] = "unknown",
392 [RTE_ETH_EVENT_INTR_LSC] = "link state change",
393 [RTE_ETH_EVENT_QUEUE_STATE] = "queue state",
394 [RTE_ETH_EVENT_INTR_RESET] = "reset",
395 [RTE_ETH_EVENT_VF_MBOX] = "VF mbox",
396 [RTE_ETH_EVENT_IPSEC] = "IPsec",
397 [RTE_ETH_EVENT_MACSEC] = "MACsec",
398 [RTE_ETH_EVENT_INTR_RMV] = "device removal",
399 [RTE_ETH_EVENT_NEW] = "device probed",
400 [RTE_ETH_EVENT_DESTROY] = "device released",
401 [RTE_ETH_EVENT_FLOW_AGED] = "flow aged",
402 [RTE_ETH_EVENT_MAX] = NULL,
406 * Display or mask ether events
407 * Default to all events except VF_MBOX
409 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
410 (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
411 (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
412 (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
413 (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
414 (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
415 (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV) |
416 (UINT32_C(1) << RTE_ETH_EVENT_FLOW_AGED);
418 * Decide if all memory are locked for performance.
423 * NIC bypass mode configuration options.
426 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
427 /* The NIC bypass watchdog timeout. */
428 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
432 #ifdef RTE_LIB_LATENCYSTATS
435 * Set when latency stats is enabled in the commandline
437 uint8_t latencystats_enabled;
440 * Lcore ID to serive latency statistics.
442 lcoreid_t latencystats_lcore_id = -1;
447 * Ethernet device configuration.
449 struct rte_eth_rxmode rx_mode = {
450 /* Default maximum frame length.
451 * Zero is converted to "RTE_ETHER_MTU + PMD Ethernet overhead"
457 struct rte_eth_txmode tx_mode = {
458 .offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
461 struct rte_fdir_conf fdir_conf = {
462 .mode = RTE_FDIR_MODE_NONE,
463 .pballoc = RTE_FDIR_PBALLOC_64K,
464 .status = RTE_FDIR_REPORT_STATUS,
466 .vlan_tci_mask = 0xFFEF,
468 .src_ip = 0xFFFFFFFF,
469 .dst_ip = 0xFFFFFFFF,
472 .src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
473 .dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
475 .src_port_mask = 0xFFFF,
476 .dst_port_mask = 0xFFFF,
477 .mac_addr_byte_mask = 0xFF,
478 .tunnel_type_mask = 1,
479 .tunnel_id_mask = 0xFFFFFFFF,
484 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
487 * Display zero values by default for xstats
489 uint8_t xstats_hide_zero;
492 * Measure of CPU cycles disabled by default
494 uint8_t record_core_cycles;
497 * Display of RX and TX bursts disabled by default
499 uint8_t record_burst_stats;
501 unsigned int num_sockets = 0;
502 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
504 #ifdef RTE_LIB_BITRATESTATS
505 /* Bitrate statistics */
506 struct rte_stats_bitrates *bitrate_data;
507 lcoreid_t bitrate_lcore_id;
508 uint8_t bitrate_enabled;
511 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
512 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
515 * hexadecimal bitmask of RX mq mode can be enabled.
517 enum rte_eth_rx_mq_mode rx_mq_mode = ETH_MQ_RX_VMDQ_DCB_RSS;
520 * Used to set forced link speed
522 uint32_t eth_link_speed;
525 * ID of the current process in multi-process, used to
526 * configure the queues to be polled.
531 * Number of processes in multi-process, used to
532 * configure the queues to be polled.
534 unsigned int num_procs = 1;
537 eth_dev_configure_mp(uint16_t port_id, uint16_t nb_rx_q, uint16_t nb_tx_q,
538 const struct rte_eth_conf *dev_conf)
540 if (is_proc_primary())
541 return rte_eth_dev_configure(port_id, nb_rx_q, nb_tx_q,
547 eth_dev_start_mp(uint16_t port_id)
549 if (is_proc_primary())
550 return rte_eth_dev_start(port_id);
556 eth_dev_stop_mp(uint16_t port_id)
558 if (is_proc_primary())
559 return rte_eth_dev_stop(port_id);
565 mempool_free_mp(struct rte_mempool *mp)
567 if (is_proc_primary())
568 rte_mempool_free(mp);
572 eth_dev_set_mtu_mp(uint16_t port_id, uint16_t mtu)
574 if (is_proc_primary())
575 return rte_eth_dev_set_mtu(port_id, mtu);
580 /* Forward function declarations */
581 static void setup_attached_port(portid_t pi);
582 static void check_all_ports_link_status(uint32_t port_mask);
583 static int eth_event_callback(portid_t port_id,
584 enum rte_eth_event_type type,
585 void *param, void *ret_param);
586 static void dev_event_callback(const char *device_name,
587 enum rte_dev_event_type type,
591 * Check if all the ports are started.
592 * If yes, return positive value. If not, return zero.
594 static int all_ports_started(void);
596 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
597 uint16_t gso_max_segment_size = RTE_ETHER_MAX_LEN - RTE_ETHER_CRC_LEN;
599 /* Holds the registered mbuf dynamic flags names. */
600 char dynf_names[64][RTE_MBUF_DYN_NAMESIZE];
603 * Helper function to check if socket is already discovered.
604 * If yes, return positive value. If not, return zero.
607 new_socket_id(unsigned int socket_id)
611 for (i = 0; i < num_sockets; i++) {
612 if (socket_ids[i] == socket_id)
619 * Setup default configuration.
622 set_default_fwd_lcores_config(void)
626 unsigned int sock_num;
629 for (i = 0; i < RTE_MAX_LCORE; i++) {
630 if (!rte_lcore_is_enabled(i))
632 sock_num = rte_lcore_to_socket_id(i);
633 if (new_socket_id(sock_num)) {
634 if (num_sockets >= RTE_MAX_NUMA_NODES) {
635 rte_exit(EXIT_FAILURE,
636 "Total sockets greater than %u\n",
639 socket_ids[num_sockets++] = sock_num;
641 if (i == rte_get_main_lcore())
643 fwd_lcores_cpuids[nb_lc++] = i;
645 nb_lcores = (lcoreid_t) nb_lc;
646 nb_cfg_lcores = nb_lcores;
651 set_def_peer_eth_addrs(void)
655 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
656 peer_eth_addrs[i].addr_bytes[0] = RTE_ETHER_LOCAL_ADMIN_ADDR;
657 peer_eth_addrs[i].addr_bytes[5] = i;
662 set_default_fwd_ports_config(void)
667 RTE_ETH_FOREACH_DEV(pt_id) {
668 fwd_ports_ids[i++] = pt_id;
670 /* Update sockets info according to the attached device */
671 int socket_id = rte_eth_dev_socket_id(pt_id);
672 if (socket_id >= 0 && new_socket_id(socket_id)) {
673 if (num_sockets >= RTE_MAX_NUMA_NODES) {
674 rte_exit(EXIT_FAILURE,
675 "Total sockets greater than %u\n",
678 socket_ids[num_sockets++] = socket_id;
682 nb_cfg_ports = nb_ports;
683 nb_fwd_ports = nb_ports;
687 set_def_fwd_config(void)
689 set_default_fwd_lcores_config();
690 set_def_peer_eth_addrs();
691 set_default_fwd_ports_config();
694 #ifndef RTE_EXEC_ENV_WINDOWS
695 /* extremely pessimistic estimation of memory required to create a mempool */
697 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
699 unsigned int n_pages, mbuf_per_pg, leftover;
700 uint64_t total_mem, mbuf_mem, obj_sz;
702 /* there is no good way to predict how much space the mempool will
703 * occupy because it will allocate chunks on the fly, and some of those
704 * will come from default DPDK memory while some will come from our
705 * external memory, so just assume 128MB will be enough for everyone.
707 uint64_t hdr_mem = 128 << 20;
709 /* account for possible non-contiguousness */
710 obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
712 TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
716 mbuf_per_pg = pgsz / obj_sz;
717 leftover = (nb_mbufs % mbuf_per_pg) > 0;
718 n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
720 mbuf_mem = n_pages * pgsz;
722 total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
724 if (total_mem > SIZE_MAX) {
725 TESTPMD_LOG(ERR, "Memory size too big\n");
728 *out = (size_t)total_mem;
734 pagesz_flags(uint64_t page_sz)
736 /* as per mmap() manpage, all page sizes are log2 of page size
737 * shifted by MAP_HUGE_SHIFT
739 int log2 = rte_log2_u64(page_sz);
741 return (log2 << HUGE_SHIFT);
745 alloc_mem(size_t memsz, size_t pgsz, bool huge)
750 /* allocate anonymous hugepages */
751 flags = MAP_ANONYMOUS | MAP_PRIVATE;
753 flags |= HUGE_FLAG | pagesz_flags(pgsz);
755 addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
756 if (addr == MAP_FAILED)
762 struct extmem_param {
766 rte_iova_t *iova_table;
767 unsigned int iova_table_len;
771 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
774 uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
775 RTE_PGSIZE_16M, RTE_PGSIZE_16G}; /* POWER */
776 unsigned int cur_page, n_pages, pgsz_idx;
777 size_t mem_sz, cur_pgsz;
778 rte_iova_t *iovas = NULL;
782 for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
783 /* skip anything that is too big */
784 if (pgsizes[pgsz_idx] > SIZE_MAX)
787 cur_pgsz = pgsizes[pgsz_idx];
789 /* if we were told not to allocate hugepages, override */
791 cur_pgsz = sysconf(_SC_PAGESIZE);
793 ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
795 TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
799 /* allocate our memory */
800 addr = alloc_mem(mem_sz, cur_pgsz, huge);
802 /* if we couldn't allocate memory with a specified page size,
803 * that doesn't mean we can't do it with other page sizes, so
809 /* store IOVA addresses for every page in this memory area */
810 n_pages = mem_sz / cur_pgsz;
812 iovas = malloc(sizeof(*iovas) * n_pages);
815 TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
818 /* lock memory if it's not huge pages */
822 /* populate IOVA addresses */
823 for (cur_page = 0; cur_page < n_pages; cur_page++) {
828 offset = cur_pgsz * cur_page;
829 cur = RTE_PTR_ADD(addr, offset);
831 /* touch the page before getting its IOVA */
832 *(volatile char *)cur = 0;
834 iova = rte_mem_virt2iova(cur);
836 iovas[cur_page] = iova;
841 /* if we couldn't allocate anything */
847 param->pgsz = cur_pgsz;
848 param->iova_table = iovas;
849 param->iova_table_len = n_pages;
856 munmap(addr, mem_sz);
862 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
864 struct extmem_param param;
867 memset(¶m, 0, sizeof(param));
869 /* check if our heap exists */
870 socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
872 /* create our heap */
873 ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
875 TESTPMD_LOG(ERR, "Cannot create heap\n");
880 ret = create_extmem(nb_mbufs, mbuf_sz, ¶m, huge);
882 TESTPMD_LOG(ERR, "Cannot create memory area\n");
886 /* we now have a valid memory area, so add it to heap */
887 ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
888 param.addr, param.len, param.iova_table,
889 param.iova_table_len, param.pgsz);
891 /* when using VFIO, memory is automatically mapped for DMA by EAL */
893 /* not needed any more */
894 free(param.iova_table);
897 TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
898 munmap(param.addr, param.len);
904 TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
910 dma_unmap_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
911 struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
916 RTE_ETH_FOREACH_DEV(pid) {
917 struct rte_eth_dev_info dev_info;
919 ret = eth_dev_info_get_print_err(pid, &dev_info);
922 "unable to get device info for port %d on addr 0x%p,"
923 "mempool unmapping will not be performed\n",
928 ret = rte_dev_dma_unmap(dev_info.device, memhdr->addr, 0, memhdr->len);
931 "unable to DMA unmap addr 0x%p "
933 memhdr->addr, dev_info.device->name);
936 ret = rte_extmem_unregister(memhdr->addr, memhdr->len);
939 "unable to un-register addr 0x%p\n", memhdr->addr);
944 dma_map_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
945 struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
948 size_t page_size = sysconf(_SC_PAGESIZE);
951 ret = rte_extmem_register(memhdr->addr, memhdr->len, NULL, 0,
955 "unable to register addr 0x%p\n", memhdr->addr);
958 RTE_ETH_FOREACH_DEV(pid) {
959 struct rte_eth_dev_info dev_info;
961 ret = eth_dev_info_get_print_err(pid, &dev_info);
964 "unable to get device info for port %d on addr 0x%p,"
965 "mempool mapping will not be performed\n",
969 ret = rte_dev_dma_map(dev_info.device, memhdr->addr, 0, memhdr->len);
972 "unable to DMA map addr 0x%p "
974 memhdr->addr, dev_info.device->name);
981 setup_extbuf(uint32_t nb_mbufs, uint16_t mbuf_sz, unsigned int socket_id,
982 char *pool_name, struct rte_pktmbuf_extmem **ext_mem)
984 struct rte_pktmbuf_extmem *xmem;
985 unsigned int ext_num, zone_num, elt_num;
988 elt_size = RTE_ALIGN_CEIL(mbuf_sz, RTE_CACHE_LINE_SIZE);
989 elt_num = EXTBUF_ZONE_SIZE / elt_size;
990 zone_num = (nb_mbufs + elt_num - 1) / elt_num;
992 xmem = malloc(sizeof(struct rte_pktmbuf_extmem) * zone_num);
994 TESTPMD_LOG(ERR, "Cannot allocate memory for "
995 "external buffer descriptors\n");
999 for (ext_num = 0; ext_num < zone_num; ext_num++) {
1000 struct rte_pktmbuf_extmem *xseg = xmem + ext_num;
1001 const struct rte_memzone *mz;
1002 char mz_name[RTE_MEMZONE_NAMESIZE];
1005 ret = snprintf(mz_name, sizeof(mz_name),
1006 RTE_MEMPOOL_MZ_FORMAT "_xb_%u", pool_name, ext_num);
1007 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
1008 errno = ENAMETOOLONG;
1012 mz = rte_memzone_reserve_aligned(mz_name, EXTBUF_ZONE_SIZE,
1014 RTE_MEMZONE_IOVA_CONTIG |
1016 RTE_MEMZONE_SIZE_HINT_ONLY,
1020 * The caller exits on external buffer creation
1021 * error, so there is no need to free memzones.
1027 xseg->buf_ptr = mz->addr;
1028 xseg->buf_iova = mz->iova;
1029 xseg->buf_len = EXTBUF_ZONE_SIZE;
1030 xseg->elt_size = elt_size;
1032 if (ext_num == 0 && xmem != NULL) {
1041 * Configuration initialisation done once at init time.
1043 static struct rte_mempool *
1044 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
1045 unsigned int socket_id, uint16_t size_idx)
1047 char pool_name[RTE_MEMPOOL_NAMESIZE];
1048 struct rte_mempool *rte_mp = NULL;
1049 #ifndef RTE_EXEC_ENV_WINDOWS
1052 mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
1054 mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name), size_idx);
1055 if (!is_proc_primary()) {
1056 rte_mp = rte_mempool_lookup(pool_name);
1058 rte_exit(EXIT_FAILURE,
1059 "Get mbuf pool for socket %u failed: %s\n",
1060 socket_id, rte_strerror(rte_errno));
1065 "create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
1066 pool_name, nb_mbuf, mbuf_seg_size, socket_id);
1068 switch (mp_alloc_type) {
1069 case MP_ALLOC_NATIVE:
1071 /* wrapper to rte_mempool_create() */
1072 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1073 rte_mbuf_best_mempool_ops());
1074 rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1075 mb_mempool_cache, 0, mbuf_seg_size, socket_id);
1078 #ifndef RTE_EXEC_ENV_WINDOWS
1081 rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
1082 mb_size, (unsigned int) mb_mempool_cache,
1083 sizeof(struct rte_pktmbuf_pool_private),
1084 socket_id, mempool_flags);
1088 if (rte_mempool_populate_anon(rte_mp) == 0) {
1089 rte_mempool_free(rte_mp);
1093 rte_pktmbuf_pool_init(rte_mp, NULL);
1094 rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
1095 rte_mempool_mem_iter(rte_mp, dma_map_cb, NULL);
1099 case MP_ALLOC_XMEM_HUGE:
1102 bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
1104 if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
1105 rte_exit(EXIT_FAILURE, "Could not create external memory\n");
1108 rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
1109 if (heap_socket < 0)
1110 rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
1112 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1113 rte_mbuf_best_mempool_ops());
1114 rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1115 mb_mempool_cache, 0, mbuf_seg_size,
1122 struct rte_pktmbuf_extmem *ext_mem;
1123 unsigned int ext_num;
1125 ext_num = setup_extbuf(nb_mbuf, mbuf_seg_size,
1126 socket_id, pool_name, &ext_mem);
1128 rte_exit(EXIT_FAILURE,
1129 "Can't create pinned data buffers\n");
1131 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1132 rte_mbuf_best_mempool_ops());
1133 rte_mp = rte_pktmbuf_pool_create_extbuf
1134 (pool_name, nb_mbuf, mb_mempool_cache,
1135 0, mbuf_seg_size, socket_id,
1142 rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
1146 #ifndef RTE_EXEC_ENV_WINDOWS
1149 if (rte_mp == NULL) {
1150 rte_exit(EXIT_FAILURE,
1151 "Creation of mbuf pool for socket %u failed: %s\n",
1152 socket_id, rte_strerror(rte_errno));
1153 } else if (verbose_level > 0) {
1154 rte_mempool_dump(stdout, rte_mp);
1160 * Check given socket id is valid or not with NUMA mode,
1161 * if valid, return 0, else return -1
1164 check_socket_id(const unsigned int socket_id)
1166 static int warning_once = 0;
1168 if (new_socket_id(socket_id)) {
1169 if (!warning_once && numa_support)
1171 "Warning: NUMA should be configured manually by using --port-numa-config and --ring-numa-config parameters along with --numa.\n");
1179 * Get the allowed maximum number of RX queues.
1180 * *pid return the port id which has minimal value of
1181 * max_rx_queues in all ports.
1184 get_allowed_max_nb_rxq(portid_t *pid)
1186 queueid_t allowed_max_rxq = RTE_MAX_QUEUES_PER_PORT;
1187 bool max_rxq_valid = false;
1189 struct rte_eth_dev_info dev_info;
1191 RTE_ETH_FOREACH_DEV(pi) {
1192 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1195 max_rxq_valid = true;
1196 if (dev_info.max_rx_queues < allowed_max_rxq) {
1197 allowed_max_rxq = dev_info.max_rx_queues;
1201 return max_rxq_valid ? allowed_max_rxq : 0;
1205 * Check input rxq is valid or not.
1206 * If input rxq is not greater than any of maximum number
1207 * of RX queues of all ports, it is valid.
1208 * if valid, return 0, else return -1
1211 check_nb_rxq(queueid_t rxq)
1213 queueid_t allowed_max_rxq;
1216 allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
1217 if (rxq > allowed_max_rxq) {
1219 "Fail: input rxq (%u) can't be greater than max_rx_queues (%u) of port %u\n",
1220 rxq, allowed_max_rxq, pid);
1227 * Get the allowed maximum number of TX queues.
1228 * *pid return the port id which has minimal value of
1229 * max_tx_queues in all ports.
1232 get_allowed_max_nb_txq(portid_t *pid)
1234 queueid_t allowed_max_txq = RTE_MAX_QUEUES_PER_PORT;
1235 bool max_txq_valid = false;
1237 struct rte_eth_dev_info dev_info;
1239 RTE_ETH_FOREACH_DEV(pi) {
1240 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1243 max_txq_valid = true;
1244 if (dev_info.max_tx_queues < allowed_max_txq) {
1245 allowed_max_txq = dev_info.max_tx_queues;
1249 return max_txq_valid ? allowed_max_txq : 0;
1253 * Check input txq is valid or not.
1254 * If input txq is not greater than any of maximum number
1255 * of TX queues of all ports, it is valid.
1256 * if valid, return 0, else return -1
1259 check_nb_txq(queueid_t txq)
1261 queueid_t allowed_max_txq;
1264 allowed_max_txq = get_allowed_max_nb_txq(&pid);
1265 if (txq > allowed_max_txq) {
1267 "Fail: input txq (%u) can't be greater than max_tx_queues (%u) of port %u\n",
1268 txq, allowed_max_txq, pid);
1275 * Get the allowed maximum number of RXDs of every rx queue.
1276 * *pid return the port id which has minimal value of
1277 * max_rxd in all queues of all ports.
1280 get_allowed_max_nb_rxd(portid_t *pid)
1282 uint16_t allowed_max_rxd = UINT16_MAX;
1284 struct rte_eth_dev_info dev_info;
1286 RTE_ETH_FOREACH_DEV(pi) {
1287 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1290 if (dev_info.rx_desc_lim.nb_max < allowed_max_rxd) {
1291 allowed_max_rxd = dev_info.rx_desc_lim.nb_max;
1295 return allowed_max_rxd;
1299 * Get the allowed minimal number of RXDs of every rx queue.
1300 * *pid return the port id which has minimal value of
1301 * min_rxd in all queues of all ports.
1304 get_allowed_min_nb_rxd(portid_t *pid)
1306 uint16_t allowed_min_rxd = 0;
1308 struct rte_eth_dev_info dev_info;
1310 RTE_ETH_FOREACH_DEV(pi) {
1311 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1314 if (dev_info.rx_desc_lim.nb_min > allowed_min_rxd) {
1315 allowed_min_rxd = dev_info.rx_desc_lim.nb_min;
1320 return allowed_min_rxd;
1324 * Check input rxd is valid or not.
1325 * If input rxd is not greater than any of maximum number
1326 * of RXDs of every Rx queues and is not less than any of
1327 * minimal number of RXDs of every Rx queues, it is valid.
1328 * if valid, return 0, else return -1
1331 check_nb_rxd(queueid_t rxd)
1333 uint16_t allowed_max_rxd;
1334 uint16_t allowed_min_rxd;
1337 allowed_max_rxd = get_allowed_max_nb_rxd(&pid);
1338 if (rxd > allowed_max_rxd) {
1340 "Fail: input rxd (%u) can't be greater than max_rxds (%u) of port %u\n",
1341 rxd, allowed_max_rxd, pid);
1345 allowed_min_rxd = get_allowed_min_nb_rxd(&pid);
1346 if (rxd < allowed_min_rxd) {
1348 "Fail: input rxd (%u) can't be less than min_rxds (%u) of port %u\n",
1349 rxd, allowed_min_rxd, pid);
1357 * Get the allowed maximum number of TXDs of every rx queues.
1358 * *pid return the port id which has minimal value of
1359 * max_txd in every tx queue.
1362 get_allowed_max_nb_txd(portid_t *pid)
1364 uint16_t allowed_max_txd = UINT16_MAX;
1366 struct rte_eth_dev_info dev_info;
1368 RTE_ETH_FOREACH_DEV(pi) {
1369 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1372 if (dev_info.tx_desc_lim.nb_max < allowed_max_txd) {
1373 allowed_max_txd = dev_info.tx_desc_lim.nb_max;
1377 return allowed_max_txd;
1381 * Get the allowed maximum number of TXDs of every tx queues.
1382 * *pid return the port id which has minimal value of
1383 * min_txd in every tx queue.
1386 get_allowed_min_nb_txd(portid_t *pid)
1388 uint16_t allowed_min_txd = 0;
1390 struct rte_eth_dev_info dev_info;
1392 RTE_ETH_FOREACH_DEV(pi) {
1393 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1396 if (dev_info.tx_desc_lim.nb_min > allowed_min_txd) {
1397 allowed_min_txd = dev_info.tx_desc_lim.nb_min;
1402 return allowed_min_txd;
1406 * Check input txd is valid or not.
1407 * If input txd is not greater than any of maximum number
1408 * of TXDs of every Rx queues, it is valid.
1409 * if valid, return 0, else return -1
1412 check_nb_txd(queueid_t txd)
1414 uint16_t allowed_max_txd;
1415 uint16_t allowed_min_txd;
1418 allowed_max_txd = get_allowed_max_nb_txd(&pid);
1419 if (txd > allowed_max_txd) {
1421 "Fail: input txd (%u) can't be greater than max_txds (%u) of port %u\n",
1422 txd, allowed_max_txd, pid);
1426 allowed_min_txd = get_allowed_min_nb_txd(&pid);
1427 if (txd < allowed_min_txd) {
1429 "Fail: input txd (%u) can't be less than min_txds (%u) of port %u\n",
1430 txd, allowed_min_txd, pid);
1438 * Get the allowed maximum number of hairpin queues.
1439 * *pid return the port id which has minimal value of
1440 * max_hairpin_queues in all ports.
1443 get_allowed_max_nb_hairpinq(portid_t *pid)
1445 queueid_t allowed_max_hairpinq = RTE_MAX_QUEUES_PER_PORT;
1447 struct rte_eth_hairpin_cap cap;
1449 RTE_ETH_FOREACH_DEV(pi) {
1450 if (rte_eth_dev_hairpin_capability_get(pi, &cap) != 0) {
1454 if (cap.max_nb_queues < allowed_max_hairpinq) {
1455 allowed_max_hairpinq = cap.max_nb_queues;
1459 return allowed_max_hairpinq;
1463 * Check input hairpin is valid or not.
1464 * If input hairpin is not greater than any of maximum number
1465 * of hairpin queues of all ports, it is valid.
1466 * if valid, return 0, else return -1
1469 check_nb_hairpinq(queueid_t hairpinq)
1471 queueid_t allowed_max_hairpinq;
1474 allowed_max_hairpinq = get_allowed_max_nb_hairpinq(&pid);
1475 if (hairpinq > allowed_max_hairpinq) {
1477 "Fail: input hairpin (%u) can't be greater than max_hairpin_queues (%u) of port %u\n",
1478 hairpinq, allowed_max_hairpinq, pid);
1485 init_config_port_offloads(portid_t pid, uint32_t socket_id)
1487 struct rte_port *port = &ports[pid];
1492 port->dev_conf.txmode = tx_mode;
1493 port->dev_conf.rxmode = rx_mode;
1495 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
1497 rte_exit(EXIT_FAILURE, "rte_eth_dev_info_get() failed\n");
1499 ret = update_jumbo_frame_offload(pid);
1502 "Updating jumbo frame offload failed for port %u\n",
1505 if (!(port->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1506 port->dev_conf.txmode.offloads &=
1507 ~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1509 /* Apply Rx offloads configuration */
1510 for (i = 0; i < port->dev_info.max_rx_queues; i++)
1511 port->rx_conf[i].offloads = port->dev_conf.rxmode.offloads;
1512 /* Apply Tx offloads configuration */
1513 for (i = 0; i < port->dev_info.max_tx_queues; i++)
1514 port->tx_conf[i].offloads = port->dev_conf.txmode.offloads;
1517 port->dev_conf.link_speeds = eth_link_speed;
1519 /* set flag to initialize port/queue */
1520 port->need_reconfig = 1;
1521 port->need_reconfig_queues = 1;
1522 port->socket_id = socket_id;
1523 port->tx_metadata = 0;
1526 * Check for maximum number of segments per MTU.
1527 * Accordingly update the mbuf data size.
1529 if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX &&
1530 port->dev_info.rx_desc_lim.nb_mtu_seg_max != 0) {
1531 data_size = rx_mode.max_rx_pkt_len /
1532 port->dev_info.rx_desc_lim.nb_mtu_seg_max;
1534 if ((data_size + RTE_PKTMBUF_HEADROOM) > mbuf_data_size[0]) {
1535 mbuf_data_size[0] = data_size + RTE_PKTMBUF_HEADROOM;
1536 TESTPMD_LOG(WARNING,
1537 "Configured mbuf size of the first segment %hu\n",
1547 struct rte_mempool *mbp;
1548 unsigned int nb_mbuf_per_pool;
1550 struct rte_gro_param gro_param;
1553 /* Configuration of logical cores. */
1554 fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1555 sizeof(struct fwd_lcore *) * nb_lcores,
1556 RTE_CACHE_LINE_SIZE);
1557 if (fwd_lcores == NULL) {
1558 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1559 "failed\n", nb_lcores);
1561 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1562 fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1563 sizeof(struct fwd_lcore),
1564 RTE_CACHE_LINE_SIZE);
1565 if (fwd_lcores[lc_id] == NULL) {
1566 rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1569 fwd_lcores[lc_id]->cpuid_idx = lc_id;
1572 RTE_ETH_FOREACH_DEV(pid) {
1576 socket_id = port_numa[pid];
1577 if (port_numa[pid] == NUMA_NO_CONFIG) {
1578 socket_id = rte_eth_dev_socket_id(pid);
1581 * if socket_id is invalid,
1582 * set to the first available socket.
1584 if (check_socket_id(socket_id) < 0)
1585 socket_id = socket_ids[0];
1588 socket_id = (socket_num == UMA_NO_CONFIG) ?
1591 /* Apply default TxRx configuration for all ports */
1592 init_config_port_offloads(pid, socket_id);
1595 * Create pools of mbuf.
1596 * If NUMA support is disabled, create a single pool of mbuf in
1597 * socket 0 memory by default.
1598 * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1600 * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1601 * nb_txd can be configured at run time.
1603 if (param_total_num_mbufs)
1604 nb_mbuf_per_pool = param_total_num_mbufs;
1606 nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1607 (nb_lcores * mb_mempool_cache) +
1608 RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1609 nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1615 for (i = 0; i < num_sockets; i++)
1616 for (j = 0; j < mbuf_data_size_n; j++)
1617 mempools[i * MAX_SEGS_BUFFER_SPLIT + j] =
1618 mbuf_pool_create(mbuf_data_size[j],
1624 for (i = 0; i < mbuf_data_size_n; i++)
1625 mempools[i] = mbuf_pool_create
1628 socket_num == UMA_NO_CONFIG ?
1634 gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1635 DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1637 * Records which Mbuf pool to use by each logical core, if needed.
1639 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1640 mbp = mbuf_pool_find(
1641 rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]), 0);
1644 mbp = mbuf_pool_find(0, 0);
1645 fwd_lcores[lc_id]->mbp = mbp;
1646 /* initialize GSO context */
1647 fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1648 fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1649 fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1650 fwd_lcores[lc_id]->gso_ctx.gso_size = RTE_ETHER_MAX_LEN -
1652 fwd_lcores[lc_id]->gso_ctx.flag = 0;
1657 /* create a gro context for each lcore */
1658 gro_param.gro_types = RTE_GRO_TCP_IPV4;
1659 gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1660 gro_param.max_item_per_flow = MAX_PKT_BURST;
1661 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1662 gro_param.socket_id = rte_lcore_to_socket_id(
1663 fwd_lcores_cpuids[lc_id]);
1664 fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1665 if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1666 rte_exit(EXIT_FAILURE,
1667 "rte_gro_ctx_create() failed\n");
1674 reconfig(portid_t new_port_id, unsigned socket_id)
1676 /* Reconfiguration of Ethernet ports. */
1677 init_config_port_offloads(new_port_id, socket_id);
1683 init_fwd_streams(void)
1686 struct rte_port *port;
1687 streamid_t sm_id, nb_fwd_streams_new;
1690 /* set socket id according to numa or not */
1691 RTE_ETH_FOREACH_DEV(pid) {
1693 if (nb_rxq > port->dev_info.max_rx_queues) {
1695 "Fail: nb_rxq(%d) is greater than max_rx_queues(%d)\n",
1696 nb_rxq, port->dev_info.max_rx_queues);
1699 if (nb_txq > port->dev_info.max_tx_queues) {
1701 "Fail: nb_txq(%d) is greater than max_tx_queues(%d)\n",
1702 nb_txq, port->dev_info.max_tx_queues);
1706 if (port_numa[pid] != NUMA_NO_CONFIG)
1707 port->socket_id = port_numa[pid];
1709 port->socket_id = rte_eth_dev_socket_id(pid);
1712 * if socket_id is invalid,
1713 * set to the first available socket.
1715 if (check_socket_id(port->socket_id) < 0)
1716 port->socket_id = socket_ids[0];
1720 if (socket_num == UMA_NO_CONFIG)
1721 port->socket_id = 0;
1723 port->socket_id = socket_num;
1727 q = RTE_MAX(nb_rxq, nb_txq);
1730 "Fail: Cannot allocate fwd streams as number of queues is 0\n");
1733 nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1734 if (nb_fwd_streams_new == nb_fwd_streams)
1737 if (fwd_streams != NULL) {
1738 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1739 if (fwd_streams[sm_id] == NULL)
1741 rte_free(fwd_streams[sm_id]);
1742 fwd_streams[sm_id] = NULL;
1744 rte_free(fwd_streams);
1749 nb_fwd_streams = nb_fwd_streams_new;
1750 if (nb_fwd_streams) {
1751 fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1752 sizeof(struct fwd_stream *) * nb_fwd_streams,
1753 RTE_CACHE_LINE_SIZE);
1754 if (fwd_streams == NULL)
1755 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1756 " (struct fwd_stream *)) failed\n",
1759 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1760 fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1761 " struct fwd_stream", sizeof(struct fwd_stream),
1762 RTE_CACHE_LINE_SIZE);
1763 if (fwd_streams[sm_id] == NULL)
1764 rte_exit(EXIT_FAILURE, "rte_zmalloc"
1765 "(struct fwd_stream) failed\n");
1773 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1775 uint64_t total_burst, sburst;
1777 uint64_t burst_stats[4];
1778 uint16_t pktnb_stats[4];
1780 int burst_percent[4], sburstp;
1784 * First compute the total number of packet bursts and the
1785 * two highest numbers of bursts of the same number of packets.
1787 memset(&burst_stats, 0x0, sizeof(burst_stats));
1788 memset(&pktnb_stats, 0x0, sizeof(pktnb_stats));
1790 /* Show stats for 0 burst size always */
1791 total_burst = pbs->pkt_burst_spread[0];
1792 burst_stats[0] = pbs->pkt_burst_spread[0];
1795 /* Find the next 2 burst sizes with highest occurrences. */
1796 for (nb_pkt = 1; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1797 nb_burst = pbs->pkt_burst_spread[nb_pkt];
1802 total_burst += nb_burst;
1804 if (nb_burst > burst_stats[1]) {
1805 burst_stats[2] = burst_stats[1];
1806 pktnb_stats[2] = pktnb_stats[1];
1807 burst_stats[1] = nb_burst;
1808 pktnb_stats[1] = nb_pkt;
1809 } else if (nb_burst > burst_stats[2]) {
1810 burst_stats[2] = nb_burst;
1811 pktnb_stats[2] = nb_pkt;
1814 if (total_burst == 0)
1817 printf(" %s-bursts : %"PRIu64" [", rx_tx, total_burst);
1818 for (i = 0, sburst = 0, sburstp = 0; i < 4; i++) {
1820 printf("%d%% of other]\n", 100 - sburstp);
1824 sburst += burst_stats[i];
1825 if (sburst == total_burst) {
1826 printf("%d%% of %d pkts]\n",
1827 100 - sburstp, (int) pktnb_stats[i]);
1832 (double)burst_stats[i] / total_burst * 100;
1833 printf("%d%% of %d pkts + ",
1834 burst_percent[i], (int) pktnb_stats[i]);
1835 sburstp += burst_percent[i];
1840 fwd_stream_stats_display(streamid_t stream_id)
1842 struct fwd_stream *fs;
1843 static const char *fwd_top_stats_border = "-------";
1845 fs = fwd_streams[stream_id];
1846 if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1847 (fs->fwd_dropped == 0))
1849 printf("\n %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1850 "TX Port=%2d/Queue=%2d %s\n",
1851 fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1852 fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1853 printf(" RX-packets: %-14"PRIu64" TX-packets: %-14"PRIu64
1854 " TX-dropped: %-14"PRIu64,
1855 fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1857 /* if checksum mode */
1858 if (cur_fwd_eng == &csum_fwd_engine) {
1859 printf(" RX- bad IP checksum: %-14"PRIu64
1860 " Rx- bad L4 checksum: %-14"PRIu64
1861 " Rx- bad outer L4 checksum: %-14"PRIu64"\n",
1862 fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1863 fs->rx_bad_outer_l4_csum);
1864 printf(" RX- bad outer IP checksum: %-14"PRIu64"\n",
1865 fs->rx_bad_outer_ip_csum);
1870 if (record_burst_stats) {
1871 pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1872 pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1877 fwd_stats_display(void)
1879 static const char *fwd_stats_border = "----------------------";
1880 static const char *acc_stats_border = "+++++++++++++++";
1882 struct fwd_stream *rx_stream;
1883 struct fwd_stream *tx_stream;
1884 uint64_t tx_dropped;
1885 uint64_t rx_bad_ip_csum;
1886 uint64_t rx_bad_l4_csum;
1887 uint64_t rx_bad_outer_l4_csum;
1888 uint64_t rx_bad_outer_ip_csum;
1889 } ports_stats[RTE_MAX_ETHPORTS];
1890 uint64_t total_rx_dropped = 0;
1891 uint64_t total_tx_dropped = 0;
1892 uint64_t total_rx_nombuf = 0;
1893 struct rte_eth_stats stats;
1894 uint64_t fwd_cycles = 0;
1895 uint64_t total_recv = 0;
1896 uint64_t total_xmit = 0;
1897 struct rte_port *port;
1902 memset(ports_stats, 0, sizeof(ports_stats));
1904 for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1905 struct fwd_stream *fs = fwd_streams[sm_id];
1907 if (cur_fwd_config.nb_fwd_streams >
1908 cur_fwd_config.nb_fwd_ports) {
1909 fwd_stream_stats_display(sm_id);
1911 ports_stats[fs->tx_port].tx_stream = fs;
1912 ports_stats[fs->rx_port].rx_stream = fs;
1915 ports_stats[fs->tx_port].tx_dropped += fs->fwd_dropped;
1917 ports_stats[fs->rx_port].rx_bad_ip_csum += fs->rx_bad_ip_csum;
1918 ports_stats[fs->rx_port].rx_bad_l4_csum += fs->rx_bad_l4_csum;
1919 ports_stats[fs->rx_port].rx_bad_outer_l4_csum +=
1920 fs->rx_bad_outer_l4_csum;
1921 ports_stats[fs->rx_port].rx_bad_outer_ip_csum +=
1922 fs->rx_bad_outer_ip_csum;
1924 if (record_core_cycles)
1925 fwd_cycles += fs->core_cycles;
1927 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1928 pt_id = fwd_ports_ids[i];
1929 port = &ports[pt_id];
1931 rte_eth_stats_get(pt_id, &stats);
1932 stats.ipackets -= port->stats.ipackets;
1933 stats.opackets -= port->stats.opackets;
1934 stats.ibytes -= port->stats.ibytes;
1935 stats.obytes -= port->stats.obytes;
1936 stats.imissed -= port->stats.imissed;
1937 stats.oerrors -= port->stats.oerrors;
1938 stats.rx_nombuf -= port->stats.rx_nombuf;
1940 total_recv += stats.ipackets;
1941 total_xmit += stats.opackets;
1942 total_rx_dropped += stats.imissed;
1943 total_tx_dropped += ports_stats[pt_id].tx_dropped;
1944 total_tx_dropped += stats.oerrors;
1945 total_rx_nombuf += stats.rx_nombuf;
1947 printf("\n %s Forward statistics for port %-2d %s\n",
1948 fwd_stats_border, pt_id, fwd_stats_border);
1950 printf(" RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64
1951 "RX-total: %-"PRIu64"\n", stats.ipackets, stats.imissed,
1952 stats.ipackets + stats.imissed);
1954 if (cur_fwd_eng == &csum_fwd_engine) {
1955 printf(" Bad-ipcsum: %-14"PRIu64
1956 " Bad-l4csum: %-14"PRIu64
1957 "Bad-outer-l4csum: %-14"PRIu64"\n",
1958 ports_stats[pt_id].rx_bad_ip_csum,
1959 ports_stats[pt_id].rx_bad_l4_csum,
1960 ports_stats[pt_id].rx_bad_outer_l4_csum);
1961 printf(" Bad-outer-ipcsum: %-14"PRIu64"\n",
1962 ports_stats[pt_id].rx_bad_outer_ip_csum);
1964 if (stats.ierrors + stats.rx_nombuf > 0) {
1965 printf(" RX-error: %-"PRIu64"\n", stats.ierrors);
1966 printf(" RX-nombufs: %-14"PRIu64"\n", stats.rx_nombuf);
1969 printf(" TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64
1970 "TX-total: %-"PRIu64"\n",
1971 stats.opackets, ports_stats[pt_id].tx_dropped,
1972 stats.opackets + ports_stats[pt_id].tx_dropped);
1974 if (record_burst_stats) {
1975 if (ports_stats[pt_id].rx_stream)
1976 pkt_burst_stats_display("RX",
1977 &ports_stats[pt_id].rx_stream->rx_burst_stats);
1978 if (ports_stats[pt_id].tx_stream)
1979 pkt_burst_stats_display("TX",
1980 &ports_stats[pt_id].tx_stream->tx_burst_stats);
1983 printf(" %s--------------------------------%s\n",
1984 fwd_stats_border, fwd_stats_border);
1987 printf("\n %s Accumulated forward statistics for all ports"
1989 acc_stats_border, acc_stats_border);
1990 printf(" RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1992 " TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1994 total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1995 total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1996 if (total_rx_nombuf > 0)
1997 printf(" RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1998 printf(" %s++++++++++++++++++++++++++++++++++++++++++++++"
2000 acc_stats_border, acc_stats_border);
2001 if (record_core_cycles) {
2002 #define CYC_PER_MHZ 1E6
2003 if (total_recv > 0 || total_xmit > 0) {
2004 uint64_t total_pkts = 0;
2005 if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 ||
2006 strcmp(cur_fwd_eng->fwd_mode_name, "flowgen") == 0)
2007 total_pkts = total_xmit;
2009 total_pkts = total_recv;
2011 printf("\n CPU cycles/packet=%.2F (total cycles="
2012 "%"PRIu64" / total %s packets=%"PRIu64") at %"PRIu64
2014 (double) fwd_cycles / total_pkts,
2015 fwd_cycles, cur_fwd_eng->fwd_mode_name, total_pkts,
2016 (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
2022 fwd_stats_reset(void)
2028 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2029 pt_id = fwd_ports_ids[i];
2030 rte_eth_stats_get(pt_id, &ports[pt_id].stats);
2032 for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
2033 struct fwd_stream *fs = fwd_streams[sm_id];
2037 fs->fwd_dropped = 0;
2038 fs->rx_bad_ip_csum = 0;
2039 fs->rx_bad_l4_csum = 0;
2040 fs->rx_bad_outer_l4_csum = 0;
2041 fs->rx_bad_outer_ip_csum = 0;
2043 memset(&fs->rx_burst_stats, 0, sizeof(fs->rx_burst_stats));
2044 memset(&fs->tx_burst_stats, 0, sizeof(fs->tx_burst_stats));
2045 fs->core_cycles = 0;
2050 flush_fwd_rx_queues(void)
2052 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2059 uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
2060 uint64_t timer_period;
2062 if (num_procs > 1) {
2063 printf("multi-process not support for flushing fwd Rx queues, skip the below lines and return.\n");
2067 /* convert to number of cycles */
2068 timer_period = rte_get_timer_hz(); /* 1 second timeout */
2070 for (j = 0; j < 2; j++) {
2071 for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
2072 for (rxq = 0; rxq < nb_rxq; rxq++) {
2073 port_id = fwd_ports_ids[rxp];
2075 * testpmd can stuck in the below do while loop
2076 * if rte_eth_rx_burst() always returns nonzero
2077 * packets. So timer is added to exit this loop
2078 * after 1sec timer expiry.
2080 prev_tsc = rte_rdtsc();
2082 nb_rx = rte_eth_rx_burst(port_id, rxq,
2083 pkts_burst, MAX_PKT_BURST);
2084 for (i = 0; i < nb_rx; i++)
2085 rte_pktmbuf_free(pkts_burst[i]);
2087 cur_tsc = rte_rdtsc();
2088 diff_tsc = cur_tsc - prev_tsc;
2089 timer_tsc += diff_tsc;
2090 } while ((nb_rx > 0) &&
2091 (timer_tsc < timer_period));
2095 rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
2100 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
2102 struct fwd_stream **fsm;
2105 #ifdef RTE_LIB_BITRATESTATS
2106 uint64_t tics_per_1sec;
2107 uint64_t tics_datum;
2108 uint64_t tics_current;
2109 uint16_t i, cnt_ports;
2111 cnt_ports = nb_ports;
2112 tics_datum = rte_rdtsc();
2113 tics_per_1sec = rte_get_timer_hz();
2115 fsm = &fwd_streams[fc->stream_idx];
2116 nb_fs = fc->stream_nb;
2118 for (sm_id = 0; sm_id < nb_fs; sm_id++)
2119 (*pkt_fwd)(fsm[sm_id]);
2120 #ifdef RTE_LIB_BITRATESTATS
2121 if (bitrate_enabled != 0 &&
2122 bitrate_lcore_id == rte_lcore_id()) {
2123 tics_current = rte_rdtsc();
2124 if (tics_current - tics_datum >= tics_per_1sec) {
2125 /* Periodic bitrate calculation */
2126 for (i = 0; i < cnt_ports; i++)
2127 rte_stats_bitrate_calc(bitrate_data,
2129 tics_datum = tics_current;
2133 #ifdef RTE_LIB_LATENCYSTATS
2134 if (latencystats_enabled != 0 &&
2135 latencystats_lcore_id == rte_lcore_id())
2136 rte_latencystats_update();
2139 } while (! fc->stopped);
2143 start_pkt_forward_on_core(void *fwd_arg)
2145 run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
2146 cur_fwd_config.fwd_eng->packet_fwd);
2151 * Run the TXONLY packet forwarding engine to send a single burst of packets.
2152 * Used to start communication flows in network loopback test configurations.
2155 run_one_txonly_burst_on_core(void *fwd_arg)
2157 struct fwd_lcore *fwd_lc;
2158 struct fwd_lcore tmp_lcore;
2160 fwd_lc = (struct fwd_lcore *) fwd_arg;
2161 tmp_lcore = *fwd_lc;
2162 tmp_lcore.stopped = 1;
2163 run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
2168 * Launch packet forwarding:
2169 * - Setup per-port forwarding context.
2170 * - launch logical cores with their forwarding configuration.
2173 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
2175 port_fwd_begin_t port_fwd_begin;
2180 port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
2181 if (port_fwd_begin != NULL) {
2182 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2183 (*port_fwd_begin)(fwd_ports_ids[i]);
2185 for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
2186 lc_id = fwd_lcores_cpuids[i];
2187 if ((interactive == 0) || (lc_id != rte_lcore_id())) {
2188 fwd_lcores[i]->stopped = 0;
2189 diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
2190 fwd_lcores[i], lc_id);
2193 "launch lcore %u failed - diag=%d\n",
2200 * Launch packet forwarding configuration.
2203 start_packet_forwarding(int with_tx_first)
2205 port_fwd_begin_t port_fwd_begin;
2206 port_fwd_end_t port_fwd_end;
2209 if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
2210 rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
2212 if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
2213 rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
2215 if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
2216 strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
2217 (!nb_rxq || !nb_txq))
2218 rte_exit(EXIT_FAILURE,
2219 "Either rxq or txq are 0, cannot use %s fwd mode\n",
2220 cur_fwd_eng->fwd_mode_name);
2222 if (all_ports_started() == 0) {
2223 fprintf(stderr, "Not all ports were started\n");
2226 if (test_done == 0) {
2227 fprintf(stderr, "Packet forwarding already started\n");
2235 flush_fwd_rx_queues();
2237 pkt_fwd_config_display(&cur_fwd_config);
2238 rxtx_config_display();
2241 if (with_tx_first) {
2242 port_fwd_begin = tx_only_engine.port_fwd_begin;
2243 if (port_fwd_begin != NULL) {
2244 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2245 (*port_fwd_begin)(fwd_ports_ids[i]);
2247 while (with_tx_first--) {
2248 launch_packet_forwarding(
2249 run_one_txonly_burst_on_core);
2250 rte_eal_mp_wait_lcore();
2252 port_fwd_end = tx_only_engine.port_fwd_end;
2253 if (port_fwd_end != NULL) {
2254 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2255 (*port_fwd_end)(fwd_ports_ids[i]);
2258 launch_packet_forwarding(start_pkt_forward_on_core);
2262 stop_packet_forwarding(void)
2264 port_fwd_end_t port_fwd_end;
2270 fprintf(stderr, "Packet forwarding not started\n");
2273 printf("Telling cores to stop...");
2274 for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
2275 fwd_lcores[lc_id]->stopped = 1;
2276 printf("\nWaiting for lcores to finish...\n");
2277 rte_eal_mp_wait_lcore();
2278 port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
2279 if (port_fwd_end != NULL) {
2280 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2281 pt_id = fwd_ports_ids[i];
2282 (*port_fwd_end)(pt_id);
2286 fwd_stats_display();
2288 printf("\nDone.\n");
2293 dev_set_link_up(portid_t pid)
2295 if (rte_eth_dev_set_link_up(pid) < 0)
2296 fprintf(stderr, "\nSet link up fail.\n");
2300 dev_set_link_down(portid_t pid)
2302 if (rte_eth_dev_set_link_down(pid) < 0)
2303 fprintf(stderr, "\nSet link down fail.\n");
2307 all_ports_started(void)
2310 struct rte_port *port;
2312 RTE_ETH_FOREACH_DEV(pi) {
2314 /* Check if there is a port which is not started */
2315 if ((port->port_status != RTE_PORT_STARTED) &&
2316 (port->slave_flag == 0))
2320 /* No port is not started */
2325 port_is_stopped(portid_t port_id)
2327 struct rte_port *port = &ports[port_id];
2329 if ((port->port_status != RTE_PORT_STOPPED) &&
2330 (port->slave_flag == 0))
2336 all_ports_stopped(void)
2340 RTE_ETH_FOREACH_DEV(pi) {
2341 if (!port_is_stopped(pi))
2349 port_is_started(portid_t port_id)
2351 if (port_id_is_invalid(port_id, ENABLED_WARN))
2354 if (ports[port_id].port_status != RTE_PORT_STARTED)
2360 /* Configure the Rx and Tx hairpin queues for the selected port. */
2362 setup_hairpin_queues(portid_t pi, portid_t p_pi, uint16_t cnt_pi)
2365 struct rte_eth_hairpin_conf hairpin_conf = {
2370 struct rte_port *port = &ports[pi];
2371 uint16_t peer_rx_port = pi;
2372 uint16_t peer_tx_port = pi;
2373 uint32_t manual = 1;
2374 uint32_t tx_exp = hairpin_mode & 0x10;
2376 if (!(hairpin_mode & 0xf)) {
2380 } else if (hairpin_mode & 0x1) {
2381 peer_tx_port = rte_eth_find_next_owned_by(pi + 1,
2382 RTE_ETH_DEV_NO_OWNER);
2383 if (peer_tx_port >= RTE_MAX_ETHPORTS)
2384 peer_tx_port = rte_eth_find_next_owned_by(0,
2385 RTE_ETH_DEV_NO_OWNER);
2386 if (p_pi != RTE_MAX_ETHPORTS) {
2387 peer_rx_port = p_pi;
2391 /* Last port will be the peer RX port of the first. */
2392 RTE_ETH_FOREACH_DEV(next_pi)
2393 peer_rx_port = next_pi;
2396 } else if (hairpin_mode & 0x2) {
2398 peer_rx_port = p_pi;
2400 peer_rx_port = rte_eth_find_next_owned_by(pi + 1,
2401 RTE_ETH_DEV_NO_OWNER);
2402 if (peer_rx_port >= RTE_MAX_ETHPORTS)
2405 peer_tx_port = peer_rx_port;
2409 for (qi = nb_txq, i = 0; qi < nb_hairpinq + nb_txq; qi++) {
2410 hairpin_conf.peers[0].port = peer_rx_port;
2411 hairpin_conf.peers[0].queue = i + nb_rxq;
2412 hairpin_conf.manual_bind = !!manual;
2413 hairpin_conf.tx_explicit = !!tx_exp;
2414 diag = rte_eth_tx_hairpin_queue_setup
2415 (pi, qi, nb_txd, &hairpin_conf);
2420 /* Fail to setup rx queue, return */
2421 if (rte_atomic16_cmpset(&(port->port_status),
2423 RTE_PORT_STOPPED) == 0)
2425 "Port %d can not be set back to stopped\n", pi);
2426 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2428 /* try to reconfigure queues next time */
2429 port->need_reconfig_queues = 1;
2432 for (qi = nb_rxq, i = 0; qi < nb_hairpinq + nb_rxq; qi++) {
2433 hairpin_conf.peers[0].port = peer_tx_port;
2434 hairpin_conf.peers[0].queue = i + nb_txq;
2435 hairpin_conf.manual_bind = !!manual;
2436 hairpin_conf.tx_explicit = !!tx_exp;
2437 diag = rte_eth_rx_hairpin_queue_setup
2438 (pi, qi, nb_rxd, &hairpin_conf);
2443 /* Fail to setup rx queue, return */
2444 if (rte_atomic16_cmpset(&(port->port_status),
2446 RTE_PORT_STOPPED) == 0)
2448 "Port %d can not be set back to stopped\n", pi);
2449 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2451 /* try to reconfigure queues next time */
2452 port->need_reconfig_queues = 1;
2458 /* Configure the Rx with optional split. */
2460 rx_queue_setup(uint16_t port_id, uint16_t rx_queue_id,
2461 uint16_t nb_rx_desc, unsigned int socket_id,
2462 struct rte_eth_rxconf *rx_conf, struct rte_mempool *mp)
2464 union rte_eth_rxseg rx_useg[MAX_SEGS_BUFFER_SPLIT] = {};
2465 unsigned int i, mp_n;
2468 if (rx_pkt_nb_segs <= 1 ||
2469 (rx_conf->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) == 0) {
2470 rx_conf->rx_seg = NULL;
2471 rx_conf->rx_nseg = 0;
2472 ret = rte_eth_rx_queue_setup(port_id, rx_queue_id,
2473 nb_rx_desc, socket_id,
2477 for (i = 0; i < rx_pkt_nb_segs; i++) {
2478 struct rte_eth_rxseg_split *rx_seg = &rx_useg[i].split;
2479 struct rte_mempool *mpx;
2481 * Use last valid pool for the segments with number
2482 * exceeding the pool index.
2484 mp_n = (i > mbuf_data_size_n) ? mbuf_data_size_n - 1 : i;
2485 mpx = mbuf_pool_find(socket_id, mp_n);
2486 /* Handle zero as mbuf data buffer size. */
2487 rx_seg->length = rx_pkt_seg_lengths[i] ?
2488 rx_pkt_seg_lengths[i] :
2489 mbuf_data_size[mp_n];
2490 rx_seg->offset = i < rx_pkt_nb_offs ?
2491 rx_pkt_seg_offsets[i] : 0;
2492 rx_seg->mp = mpx ? mpx : mp;
2494 rx_conf->rx_nseg = rx_pkt_nb_segs;
2495 rx_conf->rx_seg = rx_useg;
2496 ret = rte_eth_rx_queue_setup(port_id, rx_queue_id, nb_rx_desc,
2497 socket_id, rx_conf, NULL);
2498 rx_conf->rx_seg = NULL;
2499 rx_conf->rx_nseg = 0;
2504 start_port(portid_t pid)
2506 int diag, need_check_link_status = -1;
2508 portid_t p_pi = RTE_MAX_ETHPORTS;
2509 portid_t pl[RTE_MAX_ETHPORTS];
2510 portid_t peer_pl[RTE_MAX_ETHPORTS];
2511 uint16_t cnt_pi = 0;
2512 uint16_t cfg_pi = 0;
2515 struct rte_port *port;
2516 struct rte_eth_hairpin_cap cap;
2518 if (port_id_is_invalid(pid, ENABLED_WARN))
2521 RTE_ETH_FOREACH_DEV(pi) {
2522 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2525 need_check_link_status = 0;
2527 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
2528 RTE_PORT_HANDLING) == 0) {
2529 fprintf(stderr, "Port %d is now not stopped\n", pi);
2533 if (port->need_reconfig > 0) {
2534 port->need_reconfig = 0;
2536 if (flow_isolate_all) {
2537 int ret = port_flow_isolate(pi, 1);
2540 "Failed to apply isolated mode on port %d\n",
2545 configure_rxtx_dump_callbacks(0);
2546 printf("Configuring Port %d (socket %u)\n", pi,
2548 if (nb_hairpinq > 0 &&
2549 rte_eth_dev_hairpin_capability_get(pi, &cap)) {
2551 "Port %d doesn't support hairpin queues\n",
2555 /* configure port */
2556 diag = eth_dev_configure_mp(pi, nb_rxq + nb_hairpinq,
2557 nb_txq + nb_hairpinq,
2560 if (rte_atomic16_cmpset(&(port->port_status),
2561 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2563 "Port %d can not be set back to stopped\n",
2565 fprintf(stderr, "Fail to configure port %d\n",
2567 /* try to reconfigure port next time */
2568 port->need_reconfig = 1;
2572 if (port->need_reconfig_queues > 0 && is_proc_primary()) {
2573 port->need_reconfig_queues = 0;
2574 /* setup tx queues */
2575 for (qi = 0; qi < nb_txq; qi++) {
2576 if ((numa_support) &&
2577 (txring_numa[pi] != NUMA_NO_CONFIG))
2578 diag = rte_eth_tx_queue_setup(pi, qi,
2579 port->nb_tx_desc[qi],
2581 &(port->tx_conf[qi]));
2583 diag = rte_eth_tx_queue_setup(pi, qi,
2584 port->nb_tx_desc[qi],
2586 &(port->tx_conf[qi]));
2591 /* Fail to setup tx queue, return */
2592 if (rte_atomic16_cmpset(&(port->port_status),
2594 RTE_PORT_STOPPED) == 0)
2596 "Port %d can not be set back to stopped\n",
2599 "Fail to configure port %d tx queues\n",
2601 /* try to reconfigure queues next time */
2602 port->need_reconfig_queues = 1;
2605 for (qi = 0; qi < nb_rxq; qi++) {
2606 /* setup rx queues */
2607 if ((numa_support) &&
2608 (rxring_numa[pi] != NUMA_NO_CONFIG)) {
2609 struct rte_mempool * mp =
2611 (rxring_numa[pi], 0);
2614 "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2619 diag = rx_queue_setup(pi, qi,
2620 port->nb_rx_desc[qi],
2622 &(port->rx_conf[qi]),
2625 struct rte_mempool *mp =
2627 (port->socket_id, 0);
2630 "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2634 diag = rx_queue_setup(pi, qi,
2635 port->nb_rx_desc[qi],
2637 &(port->rx_conf[qi]),
2643 /* Fail to setup rx queue, return */
2644 if (rte_atomic16_cmpset(&(port->port_status),
2646 RTE_PORT_STOPPED) == 0)
2648 "Port %d can not be set back to stopped\n",
2651 "Fail to configure port %d rx queues\n",
2653 /* try to reconfigure queues next time */
2654 port->need_reconfig_queues = 1;
2657 /* setup hairpin queues */
2658 if (setup_hairpin_queues(pi, p_pi, cnt_pi) != 0)
2661 configure_rxtx_dump_callbacks(verbose_level);
2663 diag = rte_eth_dev_set_ptypes(pi, RTE_PTYPE_UNKNOWN,
2667 "Port %d: Failed to disable Ptype parsing\n",
2675 diag = eth_dev_start_mp(pi);
2677 fprintf(stderr, "Fail to start port %d: %s\n",
2678 pi, rte_strerror(-diag));
2680 /* Fail to setup rx queue, return */
2681 if (rte_atomic16_cmpset(&(port->port_status),
2682 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2684 "Port %d can not be set back to stopped\n",
2689 if (rte_atomic16_cmpset(&(port->port_status),
2690 RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2691 fprintf(stderr, "Port %d can not be set into started\n",
2694 if (eth_macaddr_get_print_err(pi, &port->eth_addr) == 0)
2695 printf("Port %d: " RTE_ETHER_ADDR_PRT_FMT "\n", pi,
2696 RTE_ETHER_ADDR_BYTES(&port->eth_addr));
2698 /* at least one port started, need checking link status */
2699 need_check_link_status = 1;
2704 if (need_check_link_status == 1 && !no_link_check)
2705 check_all_ports_link_status(RTE_PORT_ALL);
2706 else if (need_check_link_status == 0)
2707 fprintf(stderr, "Please stop the ports first\n");
2709 if (hairpin_mode & 0xf) {
2713 /* bind all started hairpin ports */
2714 for (i = 0; i < cfg_pi; i++) {
2716 /* bind current Tx to all peer Rx */
2717 peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2718 RTE_MAX_ETHPORTS, 1);
2721 for (j = 0; j < peer_pi; j++) {
2722 if (!port_is_started(peer_pl[j]))
2724 diag = rte_eth_hairpin_bind(pi, peer_pl[j]);
2727 "Error during binding hairpin Tx port %u to %u: %s\n",
2729 rte_strerror(-diag));
2733 /* bind all peer Tx to current Rx */
2734 peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2735 RTE_MAX_ETHPORTS, 0);
2738 for (j = 0; j < peer_pi; j++) {
2739 if (!port_is_started(peer_pl[j]))
2741 diag = rte_eth_hairpin_bind(peer_pl[j], pi);
2744 "Error during binding hairpin Tx port %u to %u: %s\n",
2746 rte_strerror(-diag));
2758 stop_port(portid_t pid)
2761 struct rte_port *port;
2762 int need_check_link_status = 0;
2763 portid_t peer_pl[RTE_MAX_ETHPORTS];
2766 if (port_id_is_invalid(pid, ENABLED_WARN))
2769 printf("Stopping ports...\n");
2771 RTE_ETH_FOREACH_DEV(pi) {
2772 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2775 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2777 "Please remove port %d from forwarding configuration.\n",
2782 if (port_is_bonding_slave(pi)) {
2784 "Please remove port %d from bonded device.\n",
2790 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2791 RTE_PORT_HANDLING) == 0)
2794 if (hairpin_mode & 0xf) {
2797 rte_eth_hairpin_unbind(pi, RTE_MAX_ETHPORTS);
2798 /* unbind all peer Tx from current Rx */
2799 peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2800 RTE_MAX_ETHPORTS, 0);
2803 for (j = 0; j < peer_pi; j++) {
2804 if (!port_is_started(peer_pl[j]))
2806 rte_eth_hairpin_unbind(peer_pl[j], pi);
2810 if (port->flow_list)
2811 port_flow_flush(pi);
2813 if (eth_dev_stop_mp(pi) != 0)
2814 RTE_LOG(ERR, EAL, "rte_eth_dev_stop failed for port %u\n",
2817 if (rte_atomic16_cmpset(&(port->port_status),
2818 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2819 fprintf(stderr, "Port %d can not be set into stopped\n",
2821 need_check_link_status = 1;
2823 if (need_check_link_status && !no_link_check)
2824 check_all_ports_link_status(RTE_PORT_ALL);
2830 remove_invalid_ports_in(portid_t *array, portid_t *total)
2833 portid_t new_total = 0;
2835 for (i = 0; i < *total; i++)
2836 if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2837 array[new_total] = array[i];
2844 remove_invalid_ports(void)
2846 remove_invalid_ports_in(ports_ids, &nb_ports);
2847 remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2848 nb_cfg_ports = nb_fwd_ports;
2852 close_port(portid_t pid)
2855 struct rte_port *port;
2857 if (port_id_is_invalid(pid, ENABLED_WARN))
2860 printf("Closing ports...\n");
2862 RTE_ETH_FOREACH_DEV(pi) {
2863 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2866 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2868 "Please remove port %d from forwarding configuration.\n",
2873 if (port_is_bonding_slave(pi)) {
2875 "Please remove port %d from bonded device.\n",
2881 if (rte_atomic16_cmpset(&(port->port_status),
2882 RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2883 fprintf(stderr, "Port %d is already closed\n", pi);
2887 if (is_proc_primary()) {
2888 port_flow_flush(pi);
2889 rte_eth_dev_close(pi);
2893 remove_invalid_ports();
2898 reset_port(portid_t pid)
2902 struct rte_port *port;
2904 if (port_id_is_invalid(pid, ENABLED_WARN))
2907 if ((pid == (portid_t)RTE_PORT_ALL && !all_ports_stopped()) ||
2908 (pid != (portid_t)RTE_PORT_ALL && !port_is_stopped(pid))) {
2910 "Can not reset port(s), please stop port(s) first.\n");
2914 printf("Resetting ports...\n");
2916 RTE_ETH_FOREACH_DEV(pi) {
2917 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2920 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2922 "Please remove port %d from forwarding configuration.\n",
2927 if (port_is_bonding_slave(pi)) {
2929 "Please remove port %d from bonded device.\n",
2934 diag = rte_eth_dev_reset(pi);
2937 port->need_reconfig = 1;
2938 port->need_reconfig_queues = 1;
2940 fprintf(stderr, "Failed to reset port %d. diag=%d\n",
2949 attach_port(char *identifier)
2952 struct rte_dev_iterator iterator;
2954 printf("Attaching a new port...\n");
2956 if (identifier == NULL) {
2957 fprintf(stderr, "Invalid parameters are specified\n");
2961 if (rte_dev_probe(identifier) < 0) {
2962 TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2966 /* first attach mode: event */
2967 if (setup_on_probe_event) {
2968 /* new ports are detected on RTE_ETH_EVENT_NEW event */
2969 for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2970 if (ports[pi].port_status == RTE_PORT_HANDLING &&
2971 ports[pi].need_setup != 0)
2972 setup_attached_port(pi);
2976 /* second attach mode: iterator */
2977 RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2978 /* setup ports matching the devargs used for probing */
2979 if (port_is_forwarding(pi))
2980 continue; /* port was already attached before */
2981 setup_attached_port(pi);
2986 setup_attached_port(portid_t pi)
2988 unsigned int socket_id;
2991 socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2992 /* if socket_id is invalid, set to the first available socket. */
2993 if (check_socket_id(socket_id) < 0)
2994 socket_id = socket_ids[0];
2995 reconfig(pi, socket_id);
2996 ret = rte_eth_promiscuous_enable(pi);
2999 "Error during enabling promiscuous mode for port %u: %s - ignore\n",
3000 pi, rte_strerror(-ret));
3002 ports_ids[nb_ports++] = pi;
3003 fwd_ports_ids[nb_fwd_ports++] = pi;
3004 nb_cfg_ports = nb_fwd_ports;
3005 ports[pi].need_setup = 0;
3006 ports[pi].port_status = RTE_PORT_STOPPED;
3008 printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
3013 detach_device(struct rte_device *dev)
3018 fprintf(stderr, "Device already removed\n");
3022 printf("Removing a device...\n");
3024 RTE_ETH_FOREACH_DEV_OF(sibling, dev) {
3025 if (ports[sibling].port_status != RTE_PORT_CLOSED) {
3026 if (ports[sibling].port_status != RTE_PORT_STOPPED) {
3027 fprintf(stderr, "Port %u not stopped\n",
3031 port_flow_flush(sibling);
3035 if (rte_dev_remove(dev) < 0) {
3036 TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
3039 remove_invalid_ports();
3041 printf("Device is detached\n");
3042 printf("Now total ports is %d\n", nb_ports);
3048 detach_port_device(portid_t port_id)
3051 struct rte_eth_dev_info dev_info;
3053 if (port_id_is_invalid(port_id, ENABLED_WARN))
3056 if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3057 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3058 fprintf(stderr, "Port not stopped\n");
3061 fprintf(stderr, "Port was not closed\n");
3064 ret = eth_dev_info_get_print_err(port_id, &dev_info);
3067 "Failed to get device info for port %d, not detaching\n",
3071 detach_device(dev_info.device);
3075 detach_devargs(char *identifier)
3077 struct rte_dev_iterator iterator;
3078 struct rte_devargs da;
3081 printf("Removing a device...\n");
3083 memset(&da, 0, sizeof(da));
3084 if (rte_devargs_parsef(&da, "%s", identifier)) {
3085 fprintf(stderr, "cannot parse identifier\n");
3089 RTE_ETH_FOREACH_MATCHING_DEV(port_id, identifier, &iterator) {
3090 if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3091 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3092 fprintf(stderr, "Port %u not stopped\n",
3094 rte_eth_iterator_cleanup(&iterator);
3095 rte_devargs_reset(&da);
3098 port_flow_flush(port_id);
3102 if (rte_eal_hotplug_remove(da.bus->name, da.name) != 0) {
3103 TESTPMD_LOG(ERR, "Failed to detach device %s(%s)\n",
3104 da.name, da.bus->name);
3105 rte_devargs_reset(&da);
3109 remove_invalid_ports();
3111 printf("Device %s is detached\n", identifier);
3112 printf("Now total ports is %d\n", nb_ports);
3114 rte_devargs_reset(&da);
3125 stop_packet_forwarding();
3127 #ifndef RTE_EXEC_ENV_WINDOWS
3128 for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3130 if (mp_alloc_type == MP_ALLOC_ANON)
3131 rte_mempool_mem_iter(mempools[i], dma_unmap_cb,
3136 if (ports != NULL) {
3138 RTE_ETH_FOREACH_DEV(pt_id) {
3139 printf("\nStopping port %d...\n", pt_id);
3143 RTE_ETH_FOREACH_DEV(pt_id) {
3144 printf("\nShutting down port %d...\n", pt_id);
3151 ret = rte_dev_event_monitor_stop();
3154 "fail to stop device event monitor.");
3158 ret = rte_dev_event_callback_unregister(NULL,
3159 dev_event_callback, NULL);
3162 "fail to unregister device event callback.\n");
3166 ret = rte_dev_hotplug_handle_disable();
3169 "fail to disable hotplug handling.\n");
3173 for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3175 mempool_free_mp(mempools[i]);
3178 printf("\nBye...\n");
3181 typedef void (*cmd_func_t)(void);
3182 struct pmd_test_command {
3183 const char *cmd_name;
3184 cmd_func_t cmd_func;
3187 /* Check the link status of all ports in up to 9s, and print them finally */
3189 check_all_ports_link_status(uint32_t port_mask)
3191 #define CHECK_INTERVAL 100 /* 100ms */
3192 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3194 uint8_t count, all_ports_up, print_flag = 0;
3195 struct rte_eth_link link;
3197 char link_status[RTE_ETH_LINK_MAX_STR_LEN];
3199 printf("Checking link statuses...\n");
3201 for (count = 0; count <= MAX_CHECK_TIME; count++) {
3203 RTE_ETH_FOREACH_DEV(portid) {
3204 if ((port_mask & (1 << portid)) == 0)
3206 memset(&link, 0, sizeof(link));
3207 ret = rte_eth_link_get_nowait(portid, &link);
3210 if (print_flag == 1)
3212 "Port %u link get failed: %s\n",
3213 portid, rte_strerror(-ret));
3216 /* print link status if flag set */
3217 if (print_flag == 1) {
3218 rte_eth_link_to_str(link_status,
3219 sizeof(link_status), &link);
3220 printf("Port %d %s\n", portid, link_status);
3223 /* clear all_ports_up flag if any link down */
3224 if (link.link_status == ETH_LINK_DOWN) {
3229 /* after finally printing all link status, get out */
3230 if (print_flag == 1)
3233 if (all_ports_up == 0) {
3235 rte_delay_ms(CHECK_INTERVAL);
3238 /* set the print_flag if all ports up or timeout */
3239 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3249 rmv_port_callback(void *arg)
3251 int need_to_start = 0;
3252 int org_no_link_check = no_link_check;
3253 portid_t port_id = (intptr_t)arg;
3254 struct rte_eth_dev_info dev_info;
3257 RTE_ETH_VALID_PORTID_OR_RET(port_id);
3259 if (!test_done && port_is_forwarding(port_id)) {
3261 stop_packet_forwarding();
3265 no_link_check = org_no_link_check;
3267 ret = eth_dev_info_get_print_err(port_id, &dev_info);
3270 "Failed to get device info for port %d, not detaching\n",
3273 struct rte_device *device = dev_info.device;
3274 close_port(port_id);
3275 detach_device(device); /* might be already removed or have more ports */
3278 start_packet_forwarding(0);
3281 /* This function is used by the interrupt thread */
3283 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
3286 RTE_SET_USED(param);
3287 RTE_SET_USED(ret_param);
3289 if (type >= RTE_ETH_EVENT_MAX) {
3291 "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
3292 port_id, __func__, type);
3294 } else if (event_print_mask & (UINT32_C(1) << type)) {
3295 printf("\nPort %" PRIu16 ": %s event\n", port_id,
3296 eth_event_desc[type]);
3301 case RTE_ETH_EVENT_NEW:
3302 ports[port_id].need_setup = 1;
3303 ports[port_id].port_status = RTE_PORT_HANDLING;
3305 case RTE_ETH_EVENT_INTR_RMV:
3306 if (port_id_is_invalid(port_id, DISABLED_WARN))
3308 if (rte_eal_alarm_set(100000,
3309 rmv_port_callback, (void *)(intptr_t)port_id))
3311 "Could not set up deferred device removal\n");
3313 case RTE_ETH_EVENT_DESTROY:
3314 ports[port_id].port_status = RTE_PORT_CLOSED;
3315 printf("Port %u is closed\n", port_id);
3324 register_eth_event_callback(void)
3327 enum rte_eth_event_type event;
3329 for (event = RTE_ETH_EVENT_UNKNOWN;
3330 event < RTE_ETH_EVENT_MAX; event++) {
3331 ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
3336 TESTPMD_LOG(ERR, "Failed to register callback for "
3337 "%s event\n", eth_event_desc[event]);
3345 /* This function is used by the interrupt thread */
3347 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
3348 __rte_unused void *arg)
3353 if (type >= RTE_DEV_EVENT_MAX) {
3354 fprintf(stderr, "%s called upon invalid event %d\n",
3360 case RTE_DEV_EVENT_REMOVE:
3361 RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
3363 ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
3365 RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
3370 * Because the user's callback is invoked in eal interrupt
3371 * callback, the interrupt callback need to be finished before
3372 * it can be unregistered when detaching device. So finish
3373 * callback soon and use a deferred removal to detach device
3374 * is need. It is a workaround, once the device detaching be
3375 * moved into the eal in the future, the deferred removal could
3378 if (rte_eal_alarm_set(100000,
3379 rmv_port_callback, (void *)(intptr_t)port_id))
3381 "Could not set up deferred device removal\n");
3383 case RTE_DEV_EVENT_ADD:
3384 RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
3386 /* TODO: After finish kernel driver binding,
3387 * begin to attach port.
3396 rxtx_port_config(struct rte_port *port)
3401 for (qid = 0; qid < nb_rxq; qid++) {
3402 offloads = port->rx_conf[qid].offloads;
3403 port->rx_conf[qid] = port->dev_info.default_rxconf;
3405 port->rx_conf[qid].offloads = offloads;
3407 /* Check if any Rx parameters have been passed */
3408 if (rx_pthresh != RTE_PMD_PARAM_UNSET)
3409 port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
3411 if (rx_hthresh != RTE_PMD_PARAM_UNSET)
3412 port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
3414 if (rx_wthresh != RTE_PMD_PARAM_UNSET)
3415 port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
3417 if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
3418 port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
3420 if (rx_drop_en != RTE_PMD_PARAM_UNSET)
3421 port->rx_conf[qid].rx_drop_en = rx_drop_en;
3423 port->nb_rx_desc[qid] = nb_rxd;
3426 for (qid = 0; qid < nb_txq; qid++) {
3427 offloads = port->tx_conf[qid].offloads;
3428 port->tx_conf[qid] = port->dev_info.default_txconf;
3430 port->tx_conf[qid].offloads = offloads;
3432 /* Check if any Tx parameters have been passed */
3433 if (tx_pthresh != RTE_PMD_PARAM_UNSET)
3434 port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
3436 if (tx_hthresh != RTE_PMD_PARAM_UNSET)
3437 port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
3439 if (tx_wthresh != RTE_PMD_PARAM_UNSET)
3440 port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
3442 if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
3443 port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
3445 if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
3446 port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
3448 port->nb_tx_desc[qid] = nb_txd;
3453 * Helper function to arrange max_rx_pktlen value and JUMBO_FRAME offload,
3454 * MTU is also aligned if JUMBO_FRAME offload is not set.
3456 * port->dev_info should be set before calling this function.
3458 * return 0 on success, negative on error
3461 update_jumbo_frame_offload(portid_t portid)
3463 struct rte_port *port = &ports[portid];
3464 uint32_t eth_overhead;
3465 uint64_t rx_offloads;
3469 /* Update the max_rx_pkt_len to have MTU as RTE_ETHER_MTU */
3470 if (port->dev_info.max_mtu != UINT16_MAX &&
3471 port->dev_info.max_rx_pktlen > port->dev_info.max_mtu)
3472 eth_overhead = port->dev_info.max_rx_pktlen -
3473 port->dev_info.max_mtu;
3475 eth_overhead = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
3477 rx_offloads = port->dev_conf.rxmode.offloads;
3479 /* Default config value is 0 to use PMD specific overhead */
3480 if (port->dev_conf.rxmode.max_rx_pkt_len == 0)
3481 port->dev_conf.rxmode.max_rx_pkt_len = RTE_ETHER_MTU + eth_overhead;
3483 if (port->dev_conf.rxmode.max_rx_pkt_len <= RTE_ETHER_MTU + eth_overhead) {
3484 rx_offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3487 if ((port->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_JUMBO_FRAME) == 0) {
3489 "Frame size (%u) is not supported by port %u\n",
3490 port->dev_conf.rxmode.max_rx_pkt_len,
3494 rx_offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
3498 if (rx_offloads != port->dev_conf.rxmode.offloads) {
3501 port->dev_conf.rxmode.offloads = rx_offloads;
3503 /* Apply JUMBO_FRAME offload configuration to Rx queue(s) */
3504 for (qid = 0; qid < port->dev_info.nb_rx_queues; qid++) {
3506 port->rx_conf[qid].offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
3508 port->rx_conf[qid].offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3512 /* If JUMBO_FRAME is set MTU conversion done by ethdev layer,
3513 * if unset do it here
3515 if ((rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) == 0) {
3516 ret = eth_dev_set_mtu_mp(portid,
3517 port->dev_conf.rxmode.max_rx_pkt_len - eth_overhead);
3520 "Failed to set MTU to %u for port %u\n",
3521 port->dev_conf.rxmode.max_rx_pkt_len - eth_overhead,
3529 init_port_config(void)
3532 struct rte_port *port;
3535 RTE_ETH_FOREACH_DEV(pid) {
3537 port->dev_conf.fdir_conf = fdir_conf;
3539 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
3544 port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3545 port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
3546 rss_hf & port->dev_info.flow_type_rss_offloads;
3548 port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3549 port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
3552 if (port->dcb_flag == 0) {
3553 if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
3554 port->dev_conf.rxmode.mq_mode =
3555 (enum rte_eth_rx_mq_mode)
3556 (rx_mq_mode & ETH_MQ_RX_RSS);
3558 port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
3561 rxtx_port_config(port);
3563 ret = eth_macaddr_get_print_err(pid, &port->eth_addr);
3567 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
3568 rte_pmd_ixgbe_bypass_init(pid);
3571 if (lsc_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_LSC))
3572 port->dev_conf.intr_conf.lsc = 1;
3573 if (rmv_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_RMV))
3574 port->dev_conf.intr_conf.rmv = 1;
3578 void set_port_slave_flag(portid_t slave_pid)
3580 struct rte_port *port;
3582 port = &ports[slave_pid];
3583 port->slave_flag = 1;
3586 void clear_port_slave_flag(portid_t slave_pid)
3588 struct rte_port *port;
3590 port = &ports[slave_pid];
3591 port->slave_flag = 0;
3594 uint8_t port_is_bonding_slave(portid_t slave_pid)
3596 struct rte_port *port;
3597 struct rte_eth_dev_info dev_info;
3600 port = &ports[slave_pid];
3601 ret = eth_dev_info_get_print_err(slave_pid, &dev_info);
3604 "Failed to get device info for port id %d,"
3605 "cannot determine if the port is a bonded slave",
3609 if ((*dev_info.dev_flags & RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
3614 const uint16_t vlan_tags[] = {
3615 0, 1, 2, 3, 4, 5, 6, 7,
3616 8, 9, 10, 11, 12, 13, 14, 15,
3617 16, 17, 18, 19, 20, 21, 22, 23,
3618 24, 25, 26, 27, 28, 29, 30, 31
3622 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
3623 enum dcb_mode_enable dcb_mode,
3624 enum rte_eth_nb_tcs num_tcs,
3629 struct rte_eth_rss_conf rss_conf;
3632 * Builds up the correct configuration for dcb+vt based on the vlan tags array
3633 * given above, and the number of traffic classes available for use.
3635 if (dcb_mode == DCB_VT_ENABLED) {
3636 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3637 ð_conf->rx_adv_conf.vmdq_dcb_conf;
3638 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3639 ð_conf->tx_adv_conf.vmdq_dcb_tx_conf;
3641 /* VMDQ+DCB RX and TX configurations */
3642 vmdq_rx_conf->enable_default_pool = 0;
3643 vmdq_rx_conf->default_pool = 0;
3644 vmdq_rx_conf->nb_queue_pools =
3645 (num_tcs == ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3646 vmdq_tx_conf->nb_queue_pools =
3647 (num_tcs == ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3649 vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
3650 for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
3651 vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
3652 vmdq_rx_conf->pool_map[i].pools =
3653 1 << (i % vmdq_rx_conf->nb_queue_pools);
3655 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3656 vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
3657 vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
3660 /* set DCB mode of RX and TX of multiple queues */
3661 eth_conf->rxmode.mq_mode =
3662 (enum rte_eth_rx_mq_mode)
3663 (rx_mq_mode & ETH_MQ_RX_VMDQ_DCB);
3664 eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
3666 struct rte_eth_dcb_rx_conf *rx_conf =
3667 ð_conf->rx_adv_conf.dcb_rx_conf;
3668 struct rte_eth_dcb_tx_conf *tx_conf =
3669 ð_conf->tx_adv_conf.dcb_tx_conf;
3671 memset(&rss_conf, 0, sizeof(struct rte_eth_rss_conf));
3673 rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
3677 rx_conf->nb_tcs = num_tcs;
3678 tx_conf->nb_tcs = num_tcs;
3680 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3681 rx_conf->dcb_tc[i] = i % num_tcs;
3682 tx_conf->dcb_tc[i] = i % num_tcs;
3685 eth_conf->rxmode.mq_mode =
3686 (enum rte_eth_rx_mq_mode)
3687 (rx_mq_mode & ETH_MQ_RX_DCB_RSS);
3688 eth_conf->rx_adv_conf.rss_conf = rss_conf;
3689 eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
3693 eth_conf->dcb_capability_en =
3694 ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
3696 eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
3702 init_port_dcb_config(portid_t pid,
3703 enum dcb_mode_enable dcb_mode,
3704 enum rte_eth_nb_tcs num_tcs,
3707 struct rte_eth_conf port_conf;
3708 struct rte_port *rte_port;
3712 if (num_procs > 1) {
3713 printf("The multi-process feature doesn't support dcb.\n");
3716 rte_port = &ports[pid];
3718 memset(&port_conf, 0, sizeof(struct rte_eth_conf));
3720 port_conf.rxmode = rte_port->dev_conf.rxmode;
3721 port_conf.txmode = rte_port->dev_conf.txmode;
3723 /*set configuration of DCB in vt mode and DCB in non-vt mode*/
3724 retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
3727 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3729 /* re-configure the device . */
3730 retval = rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
3734 retval = eth_dev_info_get_print_err(pid, &rte_port->dev_info);
3738 /* If dev_info.vmdq_pool_base is greater than 0,
3739 * the queue id of vmdq pools is started after pf queues.
3741 if (dcb_mode == DCB_VT_ENABLED &&
3742 rte_port->dev_info.vmdq_pool_base > 0) {
3744 "VMDQ_DCB multi-queue mode is nonsensical for port %d.\n",
3749 /* Assume the ports in testpmd have the same dcb capability
3750 * and has the same number of rxq and txq in dcb mode
3752 if (dcb_mode == DCB_VT_ENABLED) {
3753 if (rte_port->dev_info.max_vfs > 0) {
3754 nb_rxq = rte_port->dev_info.nb_rx_queues;
3755 nb_txq = rte_port->dev_info.nb_tx_queues;
3757 nb_rxq = rte_port->dev_info.max_rx_queues;
3758 nb_txq = rte_port->dev_info.max_tx_queues;
3761 /*if vt is disabled, use all pf queues */
3762 if (rte_port->dev_info.vmdq_pool_base == 0) {
3763 nb_rxq = rte_port->dev_info.max_rx_queues;
3764 nb_txq = rte_port->dev_info.max_tx_queues;
3766 nb_rxq = (queueid_t)num_tcs;
3767 nb_txq = (queueid_t)num_tcs;
3771 rx_free_thresh = 64;
3773 memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
3775 rxtx_port_config(rte_port);
3777 rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3778 for (i = 0; i < RTE_DIM(vlan_tags); i++)
3779 rx_vft_set(pid, vlan_tags[i], 1);
3781 retval = eth_macaddr_get_print_err(pid, &rte_port->eth_addr);
3785 rte_port->dcb_flag = 1;
3787 /* Enter DCB configuration status */
3798 /* Configuration of Ethernet ports. */
3799 ports = rte_zmalloc("testpmd: ports",
3800 sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3801 RTE_CACHE_LINE_SIZE);
3802 if (ports == NULL) {
3803 rte_exit(EXIT_FAILURE,
3804 "rte_zmalloc(%d struct rte_port) failed\n",
3807 for (i = 0; i < RTE_MAX_ETHPORTS; i++)
3808 LIST_INIT(&ports[i].flow_tunnel_list);
3809 /* Initialize ports NUMA structures */
3810 memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3811 memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3812 memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3826 const char clr[] = { 27, '[', '2', 'J', '\0' };
3827 const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3829 /* Clear screen and move to top left */
3830 printf("%s%s", clr, top_left);
3832 printf("\nPort statistics ====================================");
3833 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3834 nic_stats_display(fwd_ports_ids[i]);
3840 signal_handler(int signum)
3842 if (signum == SIGINT || signum == SIGTERM) {
3843 fprintf(stderr, "\nSignal %d received, preparing to exit...\n",
3845 #ifdef RTE_LIB_PDUMP
3846 /* uninitialize packet capture framework */
3849 #ifdef RTE_LIB_LATENCYSTATS
3850 if (latencystats_enabled != 0)
3851 rte_latencystats_uninit();
3854 /* Set flag to indicate the force termination. */
3856 /* exit with the expected status */
3857 #ifndef RTE_EXEC_ENV_WINDOWS
3858 signal(signum, SIG_DFL);
3859 kill(getpid(), signum);
3865 main(int argc, char** argv)
3872 signal(SIGINT, signal_handler);
3873 signal(SIGTERM, signal_handler);
3875 testpmd_logtype = rte_log_register("testpmd");
3876 if (testpmd_logtype < 0)
3877 rte_exit(EXIT_FAILURE, "Cannot register log type");
3878 rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3880 diag = rte_eal_init(argc, argv);
3882 rte_exit(EXIT_FAILURE, "Cannot init EAL: %s\n",
3883 rte_strerror(rte_errno));
3885 ret = register_eth_event_callback();
3887 rte_exit(EXIT_FAILURE, "Cannot register for ethdev events");
3889 #ifdef RTE_LIB_PDUMP
3890 /* initialize packet capture framework */
3895 RTE_ETH_FOREACH_DEV(port_id) {
3896 ports_ids[count] = port_id;
3899 nb_ports = (portid_t) count;
3901 TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3903 /* allocate port structures, and init them */
3906 set_def_fwd_config();
3908 rte_exit(EXIT_FAILURE, "No cores defined for forwarding\n"
3909 "Check the core mask argument\n");
3911 /* Bitrate/latency stats disabled by default */
3912 #ifdef RTE_LIB_BITRATESTATS
3913 bitrate_enabled = 0;
3915 #ifdef RTE_LIB_LATENCYSTATS
3916 latencystats_enabled = 0;
3919 /* on FreeBSD, mlockall() is disabled by default */
3920 #ifdef RTE_EXEC_ENV_FREEBSD
3929 launch_args_parse(argc, argv);
3931 #ifndef RTE_EXEC_ENV_WINDOWS
3932 if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3933 TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3938 if (tx_first && interactive)
3939 rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3940 "interactive mode.\n");
3942 if (tx_first && lsc_interrupt) {
3944 "Warning: lsc_interrupt needs to be off when using tx_first. Disabling.\n");
3948 if (!nb_rxq && !nb_txq)
3950 "Warning: Either rx or tx queues should be non-zero\n");
3952 if (nb_rxq > 1 && nb_rxq > nb_txq)
3954 "Warning: nb_rxq=%d enables RSS configuration, but nb_txq=%d will prevent to fully test it.\n",
3960 ret = rte_dev_hotplug_handle_enable();
3963 "fail to enable hotplug handling.");
3967 ret = rte_dev_event_monitor_start();
3970 "fail to start device event monitoring.");
3974 ret = rte_dev_event_callback_register(NULL,
3975 dev_event_callback, NULL);
3978 "fail to register device event callback\n");
3983 if (!no_device_start && start_port(RTE_PORT_ALL) != 0)
3984 rte_exit(EXIT_FAILURE, "Start ports failed\n");
3986 /* set all ports to promiscuous mode by default */
3987 RTE_ETH_FOREACH_DEV(port_id) {
3988 ret = rte_eth_promiscuous_enable(port_id);
3991 "Error during enabling promiscuous mode for port %u: %s - ignore\n",
3992 port_id, rte_strerror(-ret));
3995 /* Init metrics library */
3996 rte_metrics_init(rte_socket_id());
3998 #ifdef RTE_LIB_LATENCYSTATS
3999 if (latencystats_enabled != 0) {
4000 int ret = rte_latencystats_init(1, NULL);
4003 "Warning: latencystats init() returned error %d\n",
4005 fprintf(stderr, "Latencystats running on lcore %d\n",
4006 latencystats_lcore_id);
4010 /* Setup bitrate stats */
4011 #ifdef RTE_LIB_BITRATESTATS
4012 if (bitrate_enabled != 0) {
4013 bitrate_data = rte_stats_bitrate_create();
4014 if (bitrate_data == NULL)
4015 rte_exit(EXIT_FAILURE,
4016 "Could not allocate bitrate data.\n");
4017 rte_stats_bitrate_reg(bitrate_data);
4021 #ifdef RTE_LIB_CMDLINE
4022 if (strlen(cmdline_filename) != 0)
4023 cmdline_read_from_file(cmdline_filename);
4025 if (interactive == 1) {
4027 printf("Start automatic packet forwarding\n");
4028 start_packet_forwarding(0);
4040 printf("No commandline core given, start packet forwarding\n");
4041 start_packet_forwarding(tx_first);
4042 if (stats_period != 0) {
4043 uint64_t prev_time = 0, cur_time, diff_time = 0;
4044 uint64_t timer_period;
4046 /* Convert to number of cycles */
4047 timer_period = stats_period * rte_get_timer_hz();
4049 while (f_quit == 0) {
4050 cur_time = rte_get_timer_cycles();
4051 diff_time += cur_time - prev_time;
4053 if (diff_time >= timer_period) {
4055 /* Reset the timer */
4058 /* Sleep to avoid unnecessary checks */
4059 prev_time = cur_time;
4060 rte_delay_us_sleep(US_PER_S);
4064 printf("Press enter to exit\n");
4065 rc = read(0, &c, 1);
4071 ret = rte_eal_cleanup();
4073 rte_exit(EXIT_FAILURE,
4074 "EAL cleanup failed: %s\n", strerror(-ret));
4076 return EXIT_SUCCESS;