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 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
251 /* current configuration is in DCB or not,0 means it is not in DCB mode */
252 uint8_t dcb_config = 0;
255 * Configurable number of RX/TX queues.
257 queueid_t nb_hairpinq; /**< Number of hairpin queues per port. */
258 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
259 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
262 * Configurable number of RX/TX ring descriptors.
263 * Defaults are supplied by drivers via ethdev.
265 #define RTE_TEST_RX_DESC_DEFAULT 0
266 #define RTE_TEST_TX_DESC_DEFAULT 0
267 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
268 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
270 #define RTE_PMD_PARAM_UNSET -1
272 * Configurable values of RX and TX ring threshold registers.
275 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
276 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
277 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
279 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
280 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
281 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
284 * Configurable value of RX free threshold.
286 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
289 * Configurable value of RX drop enable.
291 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
294 * Configurable value of TX free threshold.
296 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
299 * Configurable value of TX RS bit threshold.
301 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
304 * Configurable value of buffered packets before sending.
306 uint16_t noisy_tx_sw_bufsz;
309 * Configurable value of packet buffer timeout.
311 uint16_t noisy_tx_sw_buf_flush_time;
314 * Configurable value for size of VNF internal memory area
315 * used for simulating noisy neighbour behaviour
317 uint64_t noisy_lkup_mem_sz;
320 * Configurable value of number of random writes done in
321 * VNF simulation memory area.
323 uint64_t noisy_lkup_num_writes;
326 * Configurable value of number of random reads done in
327 * VNF simulation memory area.
329 uint64_t noisy_lkup_num_reads;
332 * Configurable value of number of random reads/writes done in
333 * VNF simulation memory area.
335 uint64_t noisy_lkup_num_reads_writes;
338 * Receive Side Scaling (RSS) configuration.
340 uint64_t rss_hf = ETH_RSS_IP; /* RSS IP by default. */
343 * Port topology configuration
345 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
348 * Avoids to flush all the RX streams before starts forwarding.
350 uint8_t no_flush_rx = 0; /* flush by default */
353 * Flow API isolated mode.
355 uint8_t flow_isolate_all;
358 * Avoids to check link status when starting/stopping a port.
360 uint8_t no_link_check = 0; /* check by default */
363 * Don't automatically start all ports in interactive mode.
365 uint8_t no_device_start = 0;
368 * Enable link status change notification
370 uint8_t lsc_interrupt = 1; /* enabled by default */
373 * Enable device removal notification.
375 uint8_t rmv_interrupt = 1; /* enabled by default */
377 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
379 /* After attach, port setup is called on event or by iterator */
380 bool setup_on_probe_event = true;
382 /* Clear ptypes on port initialization. */
383 uint8_t clear_ptypes = true;
385 /* Hairpin ports configuration mode. */
386 uint16_t hairpin_mode;
388 /* Pretty printing of ethdev events */
389 static const char * const eth_event_desc[] = {
390 [RTE_ETH_EVENT_UNKNOWN] = "unknown",
391 [RTE_ETH_EVENT_INTR_LSC] = "link state change",
392 [RTE_ETH_EVENT_QUEUE_STATE] = "queue state",
393 [RTE_ETH_EVENT_INTR_RESET] = "reset",
394 [RTE_ETH_EVENT_VF_MBOX] = "VF mbox",
395 [RTE_ETH_EVENT_IPSEC] = "IPsec",
396 [RTE_ETH_EVENT_MACSEC] = "MACsec",
397 [RTE_ETH_EVENT_INTR_RMV] = "device removal",
398 [RTE_ETH_EVENT_NEW] = "device probed",
399 [RTE_ETH_EVENT_DESTROY] = "device released",
400 [RTE_ETH_EVENT_FLOW_AGED] = "flow aged",
401 [RTE_ETH_EVENT_MAX] = NULL,
405 * Display or mask ether events
406 * Default to all events except VF_MBOX
408 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
409 (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
410 (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
411 (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
412 (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
413 (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
414 (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV) |
415 (UINT32_C(1) << RTE_ETH_EVENT_FLOW_AGED);
417 * Decide if all memory are locked for performance.
422 * NIC bypass mode configuration options.
425 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
426 /* The NIC bypass watchdog timeout. */
427 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
431 #ifdef RTE_LIB_LATENCYSTATS
434 * Set when latency stats is enabled in the commandline
436 uint8_t latencystats_enabled;
439 * Lcore ID to serive latency statistics.
441 lcoreid_t latencystats_lcore_id = -1;
446 * Ethernet device configuration.
448 struct rte_eth_rxmode rx_mode = {
449 /* Default maximum frame length.
450 * Zero is converted to "RTE_ETHER_MTU + PMD Ethernet overhead"
456 struct rte_eth_txmode tx_mode = {
457 .offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
460 struct rte_fdir_conf fdir_conf = {
461 .mode = RTE_FDIR_MODE_NONE,
462 .pballoc = RTE_FDIR_PBALLOC_64K,
463 .status = RTE_FDIR_REPORT_STATUS,
465 .vlan_tci_mask = 0xFFEF,
467 .src_ip = 0xFFFFFFFF,
468 .dst_ip = 0xFFFFFFFF,
471 .src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
472 .dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
474 .src_port_mask = 0xFFFF,
475 .dst_port_mask = 0xFFFF,
476 .mac_addr_byte_mask = 0xFF,
477 .tunnel_type_mask = 1,
478 .tunnel_id_mask = 0xFFFFFFFF,
483 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
486 * Display zero values by default for xstats
488 uint8_t xstats_hide_zero;
491 * Measure of CPU cycles disabled by default
493 uint8_t record_core_cycles;
496 * Display of RX and TX bursts disabled by default
498 uint8_t record_burst_stats;
500 unsigned int num_sockets = 0;
501 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
503 #ifdef RTE_LIB_BITRATESTATS
504 /* Bitrate statistics */
505 struct rte_stats_bitrates *bitrate_data;
506 lcoreid_t bitrate_lcore_id;
507 uint8_t bitrate_enabled;
510 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
511 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
514 * hexadecimal bitmask of RX mq mode can be enabled.
516 enum rte_eth_rx_mq_mode rx_mq_mode = ETH_MQ_RX_VMDQ_DCB_RSS;
519 * Used to set forced link speed
521 uint32_t eth_link_speed;
523 /* Forward function declarations */
524 static void setup_attached_port(portid_t pi);
525 static void check_all_ports_link_status(uint32_t port_mask);
526 static int eth_event_callback(portid_t port_id,
527 enum rte_eth_event_type type,
528 void *param, void *ret_param);
529 static void dev_event_callback(const char *device_name,
530 enum rte_dev_event_type type,
534 * Check if all the ports are started.
535 * If yes, return positive value. If not, return zero.
537 static int all_ports_started(void);
539 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
540 uint16_t gso_max_segment_size = RTE_ETHER_MAX_LEN - RTE_ETHER_CRC_LEN;
542 /* Holds the registered mbuf dynamic flags names. */
543 char dynf_names[64][RTE_MBUF_DYN_NAMESIZE];
546 * Helper function to check if socket is already discovered.
547 * If yes, return positive value. If not, return zero.
550 new_socket_id(unsigned int socket_id)
554 for (i = 0; i < num_sockets; i++) {
555 if (socket_ids[i] == socket_id)
562 * Setup default configuration.
565 set_default_fwd_lcores_config(void)
569 unsigned int sock_num;
572 for (i = 0; i < RTE_MAX_LCORE; i++) {
573 if (!rte_lcore_is_enabled(i))
575 sock_num = rte_lcore_to_socket_id(i);
576 if (new_socket_id(sock_num)) {
577 if (num_sockets >= RTE_MAX_NUMA_NODES) {
578 rte_exit(EXIT_FAILURE,
579 "Total sockets greater than %u\n",
582 socket_ids[num_sockets++] = sock_num;
584 if (i == rte_get_main_lcore())
586 fwd_lcores_cpuids[nb_lc++] = i;
588 nb_lcores = (lcoreid_t) nb_lc;
589 nb_cfg_lcores = nb_lcores;
594 set_def_peer_eth_addrs(void)
598 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
599 peer_eth_addrs[i].addr_bytes[0] = RTE_ETHER_LOCAL_ADMIN_ADDR;
600 peer_eth_addrs[i].addr_bytes[5] = i;
605 set_default_fwd_ports_config(void)
610 RTE_ETH_FOREACH_DEV(pt_id) {
611 fwd_ports_ids[i++] = pt_id;
613 /* Update sockets info according to the attached device */
614 int socket_id = rte_eth_dev_socket_id(pt_id);
615 if (socket_id >= 0 && new_socket_id(socket_id)) {
616 if (num_sockets >= RTE_MAX_NUMA_NODES) {
617 rte_exit(EXIT_FAILURE,
618 "Total sockets greater than %u\n",
621 socket_ids[num_sockets++] = socket_id;
625 nb_cfg_ports = nb_ports;
626 nb_fwd_ports = nb_ports;
630 set_def_fwd_config(void)
632 set_default_fwd_lcores_config();
633 set_def_peer_eth_addrs();
634 set_default_fwd_ports_config();
637 #ifndef RTE_EXEC_ENV_WINDOWS
638 /* extremely pessimistic estimation of memory required to create a mempool */
640 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
642 unsigned int n_pages, mbuf_per_pg, leftover;
643 uint64_t total_mem, mbuf_mem, obj_sz;
645 /* there is no good way to predict how much space the mempool will
646 * occupy because it will allocate chunks on the fly, and some of those
647 * will come from default DPDK memory while some will come from our
648 * external memory, so just assume 128MB will be enough for everyone.
650 uint64_t hdr_mem = 128 << 20;
652 /* account for possible non-contiguousness */
653 obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
655 TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
659 mbuf_per_pg = pgsz / obj_sz;
660 leftover = (nb_mbufs % mbuf_per_pg) > 0;
661 n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
663 mbuf_mem = n_pages * pgsz;
665 total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
667 if (total_mem > SIZE_MAX) {
668 TESTPMD_LOG(ERR, "Memory size too big\n");
671 *out = (size_t)total_mem;
677 pagesz_flags(uint64_t page_sz)
679 /* as per mmap() manpage, all page sizes are log2 of page size
680 * shifted by MAP_HUGE_SHIFT
682 int log2 = rte_log2_u64(page_sz);
684 return (log2 << HUGE_SHIFT);
688 alloc_mem(size_t memsz, size_t pgsz, bool huge)
693 /* allocate anonymous hugepages */
694 flags = MAP_ANONYMOUS | MAP_PRIVATE;
696 flags |= HUGE_FLAG | pagesz_flags(pgsz);
698 addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
699 if (addr == MAP_FAILED)
705 struct extmem_param {
709 rte_iova_t *iova_table;
710 unsigned int iova_table_len;
714 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
717 uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
718 RTE_PGSIZE_16M, RTE_PGSIZE_16G}; /* POWER */
719 unsigned int cur_page, n_pages, pgsz_idx;
720 size_t mem_sz, cur_pgsz;
721 rte_iova_t *iovas = NULL;
725 for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
726 /* skip anything that is too big */
727 if (pgsizes[pgsz_idx] > SIZE_MAX)
730 cur_pgsz = pgsizes[pgsz_idx];
732 /* if we were told not to allocate hugepages, override */
734 cur_pgsz = sysconf(_SC_PAGESIZE);
736 ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
738 TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
742 /* allocate our memory */
743 addr = alloc_mem(mem_sz, cur_pgsz, huge);
745 /* if we couldn't allocate memory with a specified page size,
746 * that doesn't mean we can't do it with other page sizes, so
752 /* store IOVA addresses for every page in this memory area */
753 n_pages = mem_sz / cur_pgsz;
755 iovas = malloc(sizeof(*iovas) * n_pages);
758 TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
761 /* lock memory if it's not huge pages */
765 /* populate IOVA addresses */
766 for (cur_page = 0; cur_page < n_pages; cur_page++) {
771 offset = cur_pgsz * cur_page;
772 cur = RTE_PTR_ADD(addr, offset);
774 /* touch the page before getting its IOVA */
775 *(volatile char *)cur = 0;
777 iova = rte_mem_virt2iova(cur);
779 iovas[cur_page] = iova;
784 /* if we couldn't allocate anything */
790 param->pgsz = cur_pgsz;
791 param->iova_table = iovas;
792 param->iova_table_len = n_pages;
799 munmap(addr, mem_sz);
805 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
807 struct extmem_param param;
810 memset(¶m, 0, sizeof(param));
812 /* check if our heap exists */
813 socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
815 /* create our heap */
816 ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
818 TESTPMD_LOG(ERR, "Cannot create heap\n");
823 ret = create_extmem(nb_mbufs, mbuf_sz, ¶m, huge);
825 TESTPMD_LOG(ERR, "Cannot create memory area\n");
829 /* we now have a valid memory area, so add it to heap */
830 ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
831 param.addr, param.len, param.iova_table,
832 param.iova_table_len, param.pgsz);
834 /* when using VFIO, memory is automatically mapped for DMA by EAL */
836 /* not needed any more */
837 free(param.iova_table);
840 TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
841 munmap(param.addr, param.len);
847 TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
853 dma_unmap_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
854 struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
859 RTE_ETH_FOREACH_DEV(pid) {
860 struct rte_eth_dev_info dev_info;
862 ret = eth_dev_info_get_print_err(pid, &dev_info);
865 "unable to get device info for port %d on addr 0x%p,"
866 "mempool unmapping will not be performed\n",
871 ret = rte_dev_dma_unmap(dev_info.device, memhdr->addr, 0, memhdr->len);
874 "unable to DMA unmap addr 0x%p "
876 memhdr->addr, dev_info.device->name);
879 ret = rte_extmem_unregister(memhdr->addr, memhdr->len);
882 "unable to un-register addr 0x%p\n", memhdr->addr);
887 dma_map_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
888 struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
891 size_t page_size = sysconf(_SC_PAGESIZE);
894 ret = rte_extmem_register(memhdr->addr, memhdr->len, NULL, 0,
898 "unable to register addr 0x%p\n", memhdr->addr);
901 RTE_ETH_FOREACH_DEV(pid) {
902 struct rte_eth_dev_info dev_info;
904 ret = eth_dev_info_get_print_err(pid, &dev_info);
907 "unable to get device info for port %d on addr 0x%p,"
908 "mempool mapping will not be performed\n",
912 ret = rte_dev_dma_map(dev_info.device, memhdr->addr, 0, memhdr->len);
915 "unable to DMA map addr 0x%p "
917 memhdr->addr, dev_info.device->name);
924 setup_extbuf(uint32_t nb_mbufs, uint16_t mbuf_sz, unsigned int socket_id,
925 char *pool_name, struct rte_pktmbuf_extmem **ext_mem)
927 struct rte_pktmbuf_extmem *xmem;
928 unsigned int ext_num, zone_num, elt_num;
931 elt_size = RTE_ALIGN_CEIL(mbuf_sz, RTE_CACHE_LINE_SIZE);
932 elt_num = EXTBUF_ZONE_SIZE / elt_size;
933 zone_num = (nb_mbufs + elt_num - 1) / elt_num;
935 xmem = malloc(sizeof(struct rte_pktmbuf_extmem) * zone_num);
937 TESTPMD_LOG(ERR, "Cannot allocate memory for "
938 "external buffer descriptors\n");
942 for (ext_num = 0; ext_num < zone_num; ext_num++) {
943 struct rte_pktmbuf_extmem *xseg = xmem + ext_num;
944 const struct rte_memzone *mz;
945 char mz_name[RTE_MEMZONE_NAMESIZE];
948 ret = snprintf(mz_name, sizeof(mz_name),
949 RTE_MEMPOOL_MZ_FORMAT "_xb_%u", pool_name, ext_num);
950 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
951 errno = ENAMETOOLONG;
955 mz = rte_memzone_reserve_aligned(mz_name, EXTBUF_ZONE_SIZE,
957 RTE_MEMZONE_IOVA_CONTIG |
959 RTE_MEMZONE_SIZE_HINT_ONLY,
963 * The caller exits on external buffer creation
964 * error, so there is no need to free memzones.
970 xseg->buf_ptr = mz->addr;
971 xseg->buf_iova = mz->iova;
972 xseg->buf_len = EXTBUF_ZONE_SIZE;
973 xseg->elt_size = elt_size;
975 if (ext_num == 0 && xmem != NULL) {
984 * Configuration initialisation done once at init time.
986 static struct rte_mempool *
987 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
988 unsigned int socket_id, uint16_t size_idx)
990 char pool_name[RTE_MEMPOOL_NAMESIZE];
991 struct rte_mempool *rte_mp = NULL;
992 #ifndef RTE_EXEC_ENV_WINDOWS
995 mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
997 mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name), size_idx);
1000 "create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
1001 pool_name, nb_mbuf, mbuf_seg_size, socket_id);
1003 switch (mp_alloc_type) {
1004 case MP_ALLOC_NATIVE:
1006 /* wrapper to rte_mempool_create() */
1007 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1008 rte_mbuf_best_mempool_ops());
1009 rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1010 mb_mempool_cache, 0, mbuf_seg_size, socket_id);
1013 #ifndef RTE_EXEC_ENV_WINDOWS
1016 rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
1017 mb_size, (unsigned int) mb_mempool_cache,
1018 sizeof(struct rte_pktmbuf_pool_private),
1019 socket_id, mempool_flags);
1023 if (rte_mempool_populate_anon(rte_mp) == 0) {
1024 rte_mempool_free(rte_mp);
1028 rte_pktmbuf_pool_init(rte_mp, NULL);
1029 rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
1030 rte_mempool_mem_iter(rte_mp, dma_map_cb, NULL);
1034 case MP_ALLOC_XMEM_HUGE:
1037 bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
1039 if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
1040 rte_exit(EXIT_FAILURE, "Could not create external memory\n");
1043 rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
1044 if (heap_socket < 0)
1045 rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
1047 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1048 rte_mbuf_best_mempool_ops());
1049 rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1050 mb_mempool_cache, 0, mbuf_seg_size,
1057 struct rte_pktmbuf_extmem *ext_mem;
1058 unsigned int ext_num;
1060 ext_num = setup_extbuf(nb_mbuf, mbuf_seg_size,
1061 socket_id, pool_name, &ext_mem);
1063 rte_exit(EXIT_FAILURE,
1064 "Can't create pinned data buffers\n");
1066 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1067 rte_mbuf_best_mempool_ops());
1068 rte_mp = rte_pktmbuf_pool_create_extbuf
1069 (pool_name, nb_mbuf, mb_mempool_cache,
1070 0, mbuf_seg_size, socket_id,
1077 rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
1081 #ifndef RTE_EXEC_ENV_WINDOWS
1084 if (rte_mp == NULL) {
1085 rte_exit(EXIT_FAILURE,
1086 "Creation of mbuf pool for socket %u failed: %s\n",
1087 socket_id, rte_strerror(rte_errno));
1088 } else if (verbose_level > 0) {
1089 rte_mempool_dump(stdout, rte_mp);
1095 * Check given socket id is valid or not with NUMA mode,
1096 * if valid, return 0, else return -1
1099 check_socket_id(const unsigned int socket_id)
1101 static int warning_once = 0;
1103 if (new_socket_id(socket_id)) {
1104 if (!warning_once && numa_support)
1106 "Warning: NUMA should be configured manually by using --port-numa-config and --ring-numa-config parameters along with --numa.\n");
1114 * Get the allowed maximum number of RX queues.
1115 * *pid return the port id which has minimal value of
1116 * max_rx_queues in all ports.
1119 get_allowed_max_nb_rxq(portid_t *pid)
1121 queueid_t allowed_max_rxq = RTE_MAX_QUEUES_PER_PORT;
1122 bool max_rxq_valid = false;
1124 struct rte_eth_dev_info dev_info;
1126 RTE_ETH_FOREACH_DEV(pi) {
1127 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1130 max_rxq_valid = true;
1131 if (dev_info.max_rx_queues < allowed_max_rxq) {
1132 allowed_max_rxq = dev_info.max_rx_queues;
1136 return max_rxq_valid ? allowed_max_rxq : 0;
1140 * Check input rxq is valid or not.
1141 * If input rxq is not greater than any of maximum number
1142 * of RX queues of all ports, it is valid.
1143 * if valid, return 0, else return -1
1146 check_nb_rxq(queueid_t rxq)
1148 queueid_t allowed_max_rxq;
1151 allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
1152 if (rxq > allowed_max_rxq) {
1154 "Fail: input rxq (%u) can't be greater than max_rx_queues (%u) of port %u\n",
1155 rxq, allowed_max_rxq, pid);
1162 * Get the allowed maximum number of TX queues.
1163 * *pid return the port id which has minimal value of
1164 * max_tx_queues in all ports.
1167 get_allowed_max_nb_txq(portid_t *pid)
1169 queueid_t allowed_max_txq = RTE_MAX_QUEUES_PER_PORT;
1170 bool max_txq_valid = false;
1172 struct rte_eth_dev_info dev_info;
1174 RTE_ETH_FOREACH_DEV(pi) {
1175 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1178 max_txq_valid = true;
1179 if (dev_info.max_tx_queues < allowed_max_txq) {
1180 allowed_max_txq = dev_info.max_tx_queues;
1184 return max_txq_valid ? allowed_max_txq : 0;
1188 * Check input txq is valid or not.
1189 * If input txq is not greater than any of maximum number
1190 * of TX queues of all ports, it is valid.
1191 * if valid, return 0, else return -1
1194 check_nb_txq(queueid_t txq)
1196 queueid_t allowed_max_txq;
1199 allowed_max_txq = get_allowed_max_nb_txq(&pid);
1200 if (txq > allowed_max_txq) {
1202 "Fail: input txq (%u) can't be greater than max_tx_queues (%u) of port %u\n",
1203 txq, allowed_max_txq, pid);
1210 * Get the allowed maximum number of RXDs of every rx queue.
1211 * *pid return the port id which has minimal value of
1212 * max_rxd in all queues of all ports.
1215 get_allowed_max_nb_rxd(portid_t *pid)
1217 uint16_t allowed_max_rxd = UINT16_MAX;
1219 struct rte_eth_dev_info dev_info;
1221 RTE_ETH_FOREACH_DEV(pi) {
1222 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1225 if (dev_info.rx_desc_lim.nb_max < allowed_max_rxd) {
1226 allowed_max_rxd = dev_info.rx_desc_lim.nb_max;
1230 return allowed_max_rxd;
1234 * Get the allowed minimal number of RXDs of every rx queue.
1235 * *pid return the port id which has minimal value of
1236 * min_rxd in all queues of all ports.
1239 get_allowed_min_nb_rxd(portid_t *pid)
1241 uint16_t allowed_min_rxd = 0;
1243 struct rte_eth_dev_info dev_info;
1245 RTE_ETH_FOREACH_DEV(pi) {
1246 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1249 if (dev_info.rx_desc_lim.nb_min > allowed_min_rxd) {
1250 allowed_min_rxd = dev_info.rx_desc_lim.nb_min;
1255 return allowed_min_rxd;
1259 * Check input rxd is valid or not.
1260 * If input rxd is not greater than any of maximum number
1261 * of RXDs of every Rx queues and is not less than any of
1262 * minimal number of RXDs of every Rx queues, it is valid.
1263 * if valid, return 0, else return -1
1266 check_nb_rxd(queueid_t rxd)
1268 uint16_t allowed_max_rxd;
1269 uint16_t allowed_min_rxd;
1272 allowed_max_rxd = get_allowed_max_nb_rxd(&pid);
1273 if (rxd > allowed_max_rxd) {
1275 "Fail: input rxd (%u) can't be greater than max_rxds (%u) of port %u\n",
1276 rxd, allowed_max_rxd, pid);
1280 allowed_min_rxd = get_allowed_min_nb_rxd(&pid);
1281 if (rxd < allowed_min_rxd) {
1283 "Fail: input rxd (%u) can't be less than min_rxds (%u) of port %u\n",
1284 rxd, allowed_min_rxd, pid);
1292 * Get the allowed maximum number of TXDs of every rx queues.
1293 * *pid return the port id which has minimal value of
1294 * max_txd in every tx queue.
1297 get_allowed_max_nb_txd(portid_t *pid)
1299 uint16_t allowed_max_txd = UINT16_MAX;
1301 struct rte_eth_dev_info dev_info;
1303 RTE_ETH_FOREACH_DEV(pi) {
1304 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1307 if (dev_info.tx_desc_lim.nb_max < allowed_max_txd) {
1308 allowed_max_txd = dev_info.tx_desc_lim.nb_max;
1312 return allowed_max_txd;
1316 * Get the allowed maximum number of TXDs of every tx queues.
1317 * *pid return the port id which has minimal value of
1318 * min_txd in every tx queue.
1321 get_allowed_min_nb_txd(portid_t *pid)
1323 uint16_t allowed_min_txd = 0;
1325 struct rte_eth_dev_info dev_info;
1327 RTE_ETH_FOREACH_DEV(pi) {
1328 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1331 if (dev_info.tx_desc_lim.nb_min > allowed_min_txd) {
1332 allowed_min_txd = dev_info.tx_desc_lim.nb_min;
1337 return allowed_min_txd;
1341 * Check input txd is valid or not.
1342 * If input txd is not greater than any of maximum number
1343 * of TXDs of every Rx queues, it is valid.
1344 * if valid, return 0, else return -1
1347 check_nb_txd(queueid_t txd)
1349 uint16_t allowed_max_txd;
1350 uint16_t allowed_min_txd;
1353 allowed_max_txd = get_allowed_max_nb_txd(&pid);
1354 if (txd > allowed_max_txd) {
1356 "Fail: input txd (%u) can't be greater than max_txds (%u) of port %u\n",
1357 txd, allowed_max_txd, pid);
1361 allowed_min_txd = get_allowed_min_nb_txd(&pid);
1362 if (txd < allowed_min_txd) {
1364 "Fail: input txd (%u) can't be less than min_txds (%u) of port %u\n",
1365 txd, allowed_min_txd, pid);
1373 * Get the allowed maximum number of hairpin queues.
1374 * *pid return the port id which has minimal value of
1375 * max_hairpin_queues in all ports.
1378 get_allowed_max_nb_hairpinq(portid_t *pid)
1380 queueid_t allowed_max_hairpinq = RTE_MAX_QUEUES_PER_PORT;
1382 struct rte_eth_hairpin_cap cap;
1384 RTE_ETH_FOREACH_DEV(pi) {
1385 if (rte_eth_dev_hairpin_capability_get(pi, &cap) != 0) {
1389 if (cap.max_nb_queues < allowed_max_hairpinq) {
1390 allowed_max_hairpinq = cap.max_nb_queues;
1394 return allowed_max_hairpinq;
1398 * Check input hairpin is valid or not.
1399 * If input hairpin is not greater than any of maximum number
1400 * of hairpin queues of all ports, it is valid.
1401 * if valid, return 0, else return -1
1404 check_nb_hairpinq(queueid_t hairpinq)
1406 queueid_t allowed_max_hairpinq;
1409 allowed_max_hairpinq = get_allowed_max_nb_hairpinq(&pid);
1410 if (hairpinq > allowed_max_hairpinq) {
1412 "Fail: input hairpin (%u) can't be greater than max_hairpin_queues (%u) of port %u\n",
1413 hairpinq, allowed_max_hairpinq, pid);
1420 init_config_port_offloads(portid_t pid, uint32_t socket_id)
1422 struct rte_port *port = &ports[pid];
1427 port->dev_conf.txmode = tx_mode;
1428 port->dev_conf.rxmode = rx_mode;
1430 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
1432 rte_exit(EXIT_FAILURE, "rte_eth_dev_info_get() failed\n");
1434 ret = update_jumbo_frame_offload(pid);
1437 "Updating jumbo frame offload failed for port %u\n",
1440 if (!(port->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1441 port->dev_conf.txmode.offloads &=
1442 ~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1444 /* Apply Rx offloads configuration */
1445 for (i = 0; i < port->dev_info.max_rx_queues; i++)
1446 port->rx_conf[i].offloads = port->dev_conf.rxmode.offloads;
1447 /* Apply Tx offloads configuration */
1448 for (i = 0; i < port->dev_info.max_tx_queues; i++)
1449 port->tx_conf[i].offloads = port->dev_conf.txmode.offloads;
1452 port->dev_conf.link_speeds = eth_link_speed;
1454 /* set flag to initialize port/queue */
1455 port->need_reconfig = 1;
1456 port->need_reconfig_queues = 1;
1457 port->socket_id = socket_id;
1458 port->tx_metadata = 0;
1461 * Check for maximum number of segments per MTU.
1462 * Accordingly update the mbuf data size.
1464 if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX &&
1465 port->dev_info.rx_desc_lim.nb_mtu_seg_max != 0) {
1466 data_size = rx_mode.max_rx_pkt_len /
1467 port->dev_info.rx_desc_lim.nb_mtu_seg_max;
1469 if ((data_size + RTE_PKTMBUF_HEADROOM) > mbuf_data_size[0]) {
1470 mbuf_data_size[0] = data_size + RTE_PKTMBUF_HEADROOM;
1471 TESTPMD_LOG(WARNING,
1472 "Configured mbuf size of the first segment %hu\n",
1482 struct rte_mempool *mbp;
1483 unsigned int nb_mbuf_per_pool;
1485 struct rte_gro_param gro_param;
1488 /* Configuration of logical cores. */
1489 fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1490 sizeof(struct fwd_lcore *) * nb_lcores,
1491 RTE_CACHE_LINE_SIZE);
1492 if (fwd_lcores == NULL) {
1493 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1494 "failed\n", nb_lcores);
1496 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1497 fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1498 sizeof(struct fwd_lcore),
1499 RTE_CACHE_LINE_SIZE);
1500 if (fwd_lcores[lc_id] == NULL) {
1501 rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1504 fwd_lcores[lc_id]->cpuid_idx = lc_id;
1507 RTE_ETH_FOREACH_DEV(pid) {
1511 socket_id = port_numa[pid];
1512 if (port_numa[pid] == NUMA_NO_CONFIG) {
1513 socket_id = rte_eth_dev_socket_id(pid);
1516 * if socket_id is invalid,
1517 * set to the first available socket.
1519 if (check_socket_id(socket_id) < 0)
1520 socket_id = socket_ids[0];
1523 socket_id = (socket_num == UMA_NO_CONFIG) ?
1526 /* Apply default TxRx configuration for all ports */
1527 init_config_port_offloads(pid, socket_id);
1530 * Create pools of mbuf.
1531 * If NUMA support is disabled, create a single pool of mbuf in
1532 * socket 0 memory by default.
1533 * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1535 * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1536 * nb_txd can be configured at run time.
1538 if (param_total_num_mbufs)
1539 nb_mbuf_per_pool = param_total_num_mbufs;
1541 nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1542 (nb_lcores * mb_mempool_cache) +
1543 RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1544 nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1550 for (i = 0; i < num_sockets; i++)
1551 for (j = 0; j < mbuf_data_size_n; j++)
1552 mempools[i * MAX_SEGS_BUFFER_SPLIT + j] =
1553 mbuf_pool_create(mbuf_data_size[j],
1559 for (i = 0; i < mbuf_data_size_n; i++)
1560 mempools[i] = mbuf_pool_create
1563 socket_num == UMA_NO_CONFIG ?
1569 gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1570 DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1572 * Records which Mbuf pool to use by each logical core, if needed.
1574 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1575 mbp = mbuf_pool_find(
1576 rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]), 0);
1579 mbp = mbuf_pool_find(0, 0);
1580 fwd_lcores[lc_id]->mbp = mbp;
1581 /* initialize GSO context */
1582 fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1583 fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1584 fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1585 fwd_lcores[lc_id]->gso_ctx.gso_size = RTE_ETHER_MAX_LEN -
1587 fwd_lcores[lc_id]->gso_ctx.flag = 0;
1592 /* create a gro context for each lcore */
1593 gro_param.gro_types = RTE_GRO_TCP_IPV4;
1594 gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1595 gro_param.max_item_per_flow = MAX_PKT_BURST;
1596 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1597 gro_param.socket_id = rte_lcore_to_socket_id(
1598 fwd_lcores_cpuids[lc_id]);
1599 fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1600 if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1601 rte_exit(EXIT_FAILURE,
1602 "rte_gro_ctx_create() failed\n");
1609 reconfig(portid_t new_port_id, unsigned socket_id)
1611 /* Reconfiguration of Ethernet ports. */
1612 init_config_port_offloads(new_port_id, socket_id);
1618 init_fwd_streams(void)
1621 struct rte_port *port;
1622 streamid_t sm_id, nb_fwd_streams_new;
1625 /* set socket id according to numa or not */
1626 RTE_ETH_FOREACH_DEV(pid) {
1628 if (nb_rxq > port->dev_info.max_rx_queues) {
1630 "Fail: nb_rxq(%d) is greater than max_rx_queues(%d)\n",
1631 nb_rxq, port->dev_info.max_rx_queues);
1634 if (nb_txq > port->dev_info.max_tx_queues) {
1636 "Fail: nb_txq(%d) is greater than max_tx_queues(%d)\n",
1637 nb_txq, port->dev_info.max_tx_queues);
1641 if (port_numa[pid] != NUMA_NO_CONFIG)
1642 port->socket_id = port_numa[pid];
1644 port->socket_id = rte_eth_dev_socket_id(pid);
1647 * if socket_id is invalid,
1648 * set to the first available socket.
1650 if (check_socket_id(port->socket_id) < 0)
1651 port->socket_id = socket_ids[0];
1655 if (socket_num == UMA_NO_CONFIG)
1656 port->socket_id = 0;
1658 port->socket_id = socket_num;
1662 q = RTE_MAX(nb_rxq, nb_txq);
1665 "Fail: Cannot allocate fwd streams as number of queues is 0\n");
1668 nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1669 if (nb_fwd_streams_new == nb_fwd_streams)
1672 if (fwd_streams != NULL) {
1673 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1674 if (fwd_streams[sm_id] == NULL)
1676 rte_free(fwd_streams[sm_id]);
1677 fwd_streams[sm_id] = NULL;
1679 rte_free(fwd_streams);
1684 nb_fwd_streams = nb_fwd_streams_new;
1685 if (nb_fwd_streams) {
1686 fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1687 sizeof(struct fwd_stream *) * nb_fwd_streams,
1688 RTE_CACHE_LINE_SIZE);
1689 if (fwd_streams == NULL)
1690 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1691 " (struct fwd_stream *)) failed\n",
1694 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1695 fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1696 " struct fwd_stream", sizeof(struct fwd_stream),
1697 RTE_CACHE_LINE_SIZE);
1698 if (fwd_streams[sm_id] == NULL)
1699 rte_exit(EXIT_FAILURE, "rte_zmalloc"
1700 "(struct fwd_stream) failed\n");
1708 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1710 uint64_t total_burst, sburst;
1712 uint64_t burst_stats[4];
1713 uint16_t pktnb_stats[4];
1715 int burst_percent[4], sburstp;
1719 * First compute the total number of packet bursts and the
1720 * two highest numbers of bursts of the same number of packets.
1722 memset(&burst_stats, 0x0, sizeof(burst_stats));
1723 memset(&pktnb_stats, 0x0, sizeof(pktnb_stats));
1725 /* Show stats for 0 burst size always */
1726 total_burst = pbs->pkt_burst_spread[0];
1727 burst_stats[0] = pbs->pkt_burst_spread[0];
1730 /* Find the next 2 burst sizes with highest occurrences. */
1731 for (nb_pkt = 1; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1732 nb_burst = pbs->pkt_burst_spread[nb_pkt];
1737 total_burst += nb_burst;
1739 if (nb_burst > burst_stats[1]) {
1740 burst_stats[2] = burst_stats[1];
1741 pktnb_stats[2] = pktnb_stats[1];
1742 burst_stats[1] = nb_burst;
1743 pktnb_stats[1] = nb_pkt;
1744 } else if (nb_burst > burst_stats[2]) {
1745 burst_stats[2] = nb_burst;
1746 pktnb_stats[2] = nb_pkt;
1749 if (total_burst == 0)
1752 printf(" %s-bursts : %"PRIu64" [", rx_tx, total_burst);
1753 for (i = 0, sburst = 0, sburstp = 0; i < 4; i++) {
1755 printf("%d%% of other]\n", 100 - sburstp);
1759 sburst += burst_stats[i];
1760 if (sburst == total_burst) {
1761 printf("%d%% of %d pkts]\n",
1762 100 - sburstp, (int) pktnb_stats[i]);
1767 (double)burst_stats[i] / total_burst * 100;
1768 printf("%d%% of %d pkts + ",
1769 burst_percent[i], (int) pktnb_stats[i]);
1770 sburstp += burst_percent[i];
1775 fwd_stream_stats_display(streamid_t stream_id)
1777 struct fwd_stream *fs;
1778 static const char *fwd_top_stats_border = "-------";
1780 fs = fwd_streams[stream_id];
1781 if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1782 (fs->fwd_dropped == 0))
1784 printf("\n %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1785 "TX Port=%2d/Queue=%2d %s\n",
1786 fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1787 fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1788 printf(" RX-packets: %-14"PRIu64" TX-packets: %-14"PRIu64
1789 " TX-dropped: %-14"PRIu64,
1790 fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1792 /* if checksum mode */
1793 if (cur_fwd_eng == &csum_fwd_engine) {
1794 printf(" RX- bad IP checksum: %-14"PRIu64
1795 " Rx- bad L4 checksum: %-14"PRIu64
1796 " Rx- bad outer L4 checksum: %-14"PRIu64"\n",
1797 fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1798 fs->rx_bad_outer_l4_csum);
1799 printf(" RX- bad outer IP checksum: %-14"PRIu64"\n",
1800 fs->rx_bad_outer_ip_csum);
1805 if (record_burst_stats) {
1806 pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1807 pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1812 fwd_stats_display(void)
1814 static const char *fwd_stats_border = "----------------------";
1815 static const char *acc_stats_border = "+++++++++++++++";
1817 struct fwd_stream *rx_stream;
1818 struct fwd_stream *tx_stream;
1819 uint64_t tx_dropped;
1820 uint64_t rx_bad_ip_csum;
1821 uint64_t rx_bad_l4_csum;
1822 uint64_t rx_bad_outer_l4_csum;
1823 uint64_t rx_bad_outer_ip_csum;
1824 } ports_stats[RTE_MAX_ETHPORTS];
1825 uint64_t total_rx_dropped = 0;
1826 uint64_t total_tx_dropped = 0;
1827 uint64_t total_rx_nombuf = 0;
1828 struct rte_eth_stats stats;
1829 uint64_t fwd_cycles = 0;
1830 uint64_t total_recv = 0;
1831 uint64_t total_xmit = 0;
1832 struct rte_port *port;
1837 memset(ports_stats, 0, sizeof(ports_stats));
1839 for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1840 struct fwd_stream *fs = fwd_streams[sm_id];
1842 if (cur_fwd_config.nb_fwd_streams >
1843 cur_fwd_config.nb_fwd_ports) {
1844 fwd_stream_stats_display(sm_id);
1846 ports_stats[fs->tx_port].tx_stream = fs;
1847 ports_stats[fs->rx_port].rx_stream = fs;
1850 ports_stats[fs->tx_port].tx_dropped += fs->fwd_dropped;
1852 ports_stats[fs->rx_port].rx_bad_ip_csum += fs->rx_bad_ip_csum;
1853 ports_stats[fs->rx_port].rx_bad_l4_csum += fs->rx_bad_l4_csum;
1854 ports_stats[fs->rx_port].rx_bad_outer_l4_csum +=
1855 fs->rx_bad_outer_l4_csum;
1856 ports_stats[fs->rx_port].rx_bad_outer_ip_csum +=
1857 fs->rx_bad_outer_ip_csum;
1859 if (record_core_cycles)
1860 fwd_cycles += fs->core_cycles;
1862 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1863 pt_id = fwd_ports_ids[i];
1864 port = &ports[pt_id];
1866 rte_eth_stats_get(pt_id, &stats);
1867 stats.ipackets -= port->stats.ipackets;
1868 stats.opackets -= port->stats.opackets;
1869 stats.ibytes -= port->stats.ibytes;
1870 stats.obytes -= port->stats.obytes;
1871 stats.imissed -= port->stats.imissed;
1872 stats.oerrors -= port->stats.oerrors;
1873 stats.rx_nombuf -= port->stats.rx_nombuf;
1875 total_recv += stats.ipackets;
1876 total_xmit += stats.opackets;
1877 total_rx_dropped += stats.imissed;
1878 total_tx_dropped += ports_stats[pt_id].tx_dropped;
1879 total_tx_dropped += stats.oerrors;
1880 total_rx_nombuf += stats.rx_nombuf;
1882 printf("\n %s Forward statistics for port %-2d %s\n",
1883 fwd_stats_border, pt_id, fwd_stats_border);
1885 printf(" RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64
1886 "RX-total: %-"PRIu64"\n", stats.ipackets, stats.imissed,
1887 stats.ipackets + stats.imissed);
1889 if (cur_fwd_eng == &csum_fwd_engine) {
1890 printf(" Bad-ipcsum: %-14"PRIu64
1891 " Bad-l4csum: %-14"PRIu64
1892 "Bad-outer-l4csum: %-14"PRIu64"\n",
1893 ports_stats[pt_id].rx_bad_ip_csum,
1894 ports_stats[pt_id].rx_bad_l4_csum,
1895 ports_stats[pt_id].rx_bad_outer_l4_csum);
1896 printf(" Bad-outer-ipcsum: %-14"PRIu64"\n",
1897 ports_stats[pt_id].rx_bad_outer_ip_csum);
1899 if (stats.ierrors + stats.rx_nombuf > 0) {
1900 printf(" RX-error: %-"PRIu64"\n", stats.ierrors);
1901 printf(" RX-nombufs: %-14"PRIu64"\n", stats.rx_nombuf);
1904 printf(" TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64
1905 "TX-total: %-"PRIu64"\n",
1906 stats.opackets, ports_stats[pt_id].tx_dropped,
1907 stats.opackets + ports_stats[pt_id].tx_dropped);
1909 if (record_burst_stats) {
1910 if (ports_stats[pt_id].rx_stream)
1911 pkt_burst_stats_display("RX",
1912 &ports_stats[pt_id].rx_stream->rx_burst_stats);
1913 if (ports_stats[pt_id].tx_stream)
1914 pkt_burst_stats_display("TX",
1915 &ports_stats[pt_id].tx_stream->tx_burst_stats);
1918 printf(" %s--------------------------------%s\n",
1919 fwd_stats_border, fwd_stats_border);
1922 printf("\n %s Accumulated forward statistics for all ports"
1924 acc_stats_border, acc_stats_border);
1925 printf(" RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1927 " TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1929 total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1930 total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1931 if (total_rx_nombuf > 0)
1932 printf(" RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1933 printf(" %s++++++++++++++++++++++++++++++++++++++++++++++"
1935 acc_stats_border, acc_stats_border);
1936 if (record_core_cycles) {
1937 #define CYC_PER_MHZ 1E6
1938 if (total_recv > 0 || total_xmit > 0) {
1939 uint64_t total_pkts = 0;
1940 if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 ||
1941 strcmp(cur_fwd_eng->fwd_mode_name, "flowgen") == 0)
1942 total_pkts = total_xmit;
1944 total_pkts = total_recv;
1946 printf("\n CPU cycles/packet=%.2F (total cycles="
1947 "%"PRIu64" / total %s packets=%"PRIu64") at %"PRIu64
1949 (double) fwd_cycles / total_pkts,
1950 fwd_cycles, cur_fwd_eng->fwd_mode_name, total_pkts,
1951 (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
1957 fwd_stats_reset(void)
1963 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1964 pt_id = fwd_ports_ids[i];
1965 rte_eth_stats_get(pt_id, &ports[pt_id].stats);
1967 for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1968 struct fwd_stream *fs = fwd_streams[sm_id];
1972 fs->fwd_dropped = 0;
1973 fs->rx_bad_ip_csum = 0;
1974 fs->rx_bad_l4_csum = 0;
1975 fs->rx_bad_outer_l4_csum = 0;
1976 fs->rx_bad_outer_ip_csum = 0;
1978 memset(&fs->rx_burst_stats, 0, sizeof(fs->rx_burst_stats));
1979 memset(&fs->tx_burst_stats, 0, sizeof(fs->tx_burst_stats));
1980 fs->core_cycles = 0;
1985 flush_fwd_rx_queues(void)
1987 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1994 uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
1995 uint64_t timer_period;
1997 /* convert to number of cycles */
1998 timer_period = rte_get_timer_hz(); /* 1 second timeout */
2000 for (j = 0; j < 2; j++) {
2001 for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
2002 for (rxq = 0; rxq < nb_rxq; rxq++) {
2003 port_id = fwd_ports_ids[rxp];
2005 * testpmd can stuck in the below do while loop
2006 * if rte_eth_rx_burst() always returns nonzero
2007 * packets. So timer is added to exit this loop
2008 * after 1sec timer expiry.
2010 prev_tsc = rte_rdtsc();
2012 nb_rx = rte_eth_rx_burst(port_id, rxq,
2013 pkts_burst, MAX_PKT_BURST);
2014 for (i = 0; i < nb_rx; i++)
2015 rte_pktmbuf_free(pkts_burst[i]);
2017 cur_tsc = rte_rdtsc();
2018 diff_tsc = cur_tsc - prev_tsc;
2019 timer_tsc += diff_tsc;
2020 } while ((nb_rx > 0) &&
2021 (timer_tsc < timer_period));
2025 rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
2030 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
2032 struct fwd_stream **fsm;
2035 #ifdef RTE_LIB_BITRATESTATS
2036 uint64_t tics_per_1sec;
2037 uint64_t tics_datum;
2038 uint64_t tics_current;
2039 uint16_t i, cnt_ports;
2041 cnt_ports = nb_ports;
2042 tics_datum = rte_rdtsc();
2043 tics_per_1sec = rte_get_timer_hz();
2045 fsm = &fwd_streams[fc->stream_idx];
2046 nb_fs = fc->stream_nb;
2048 for (sm_id = 0; sm_id < nb_fs; sm_id++)
2049 (*pkt_fwd)(fsm[sm_id]);
2050 #ifdef RTE_LIB_BITRATESTATS
2051 if (bitrate_enabled != 0 &&
2052 bitrate_lcore_id == rte_lcore_id()) {
2053 tics_current = rte_rdtsc();
2054 if (tics_current - tics_datum >= tics_per_1sec) {
2055 /* Periodic bitrate calculation */
2056 for (i = 0; i < cnt_ports; i++)
2057 rte_stats_bitrate_calc(bitrate_data,
2059 tics_datum = tics_current;
2063 #ifdef RTE_LIB_LATENCYSTATS
2064 if (latencystats_enabled != 0 &&
2065 latencystats_lcore_id == rte_lcore_id())
2066 rte_latencystats_update();
2069 } while (! fc->stopped);
2073 start_pkt_forward_on_core(void *fwd_arg)
2075 run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
2076 cur_fwd_config.fwd_eng->packet_fwd);
2081 * Run the TXONLY packet forwarding engine to send a single burst of packets.
2082 * Used to start communication flows in network loopback test configurations.
2085 run_one_txonly_burst_on_core(void *fwd_arg)
2087 struct fwd_lcore *fwd_lc;
2088 struct fwd_lcore tmp_lcore;
2090 fwd_lc = (struct fwd_lcore *) fwd_arg;
2091 tmp_lcore = *fwd_lc;
2092 tmp_lcore.stopped = 1;
2093 run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
2098 * Launch packet forwarding:
2099 * - Setup per-port forwarding context.
2100 * - launch logical cores with their forwarding configuration.
2103 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
2105 port_fwd_begin_t port_fwd_begin;
2110 port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
2111 if (port_fwd_begin != NULL) {
2112 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2113 (*port_fwd_begin)(fwd_ports_ids[i]);
2115 for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
2116 lc_id = fwd_lcores_cpuids[i];
2117 if ((interactive == 0) || (lc_id != rte_lcore_id())) {
2118 fwd_lcores[i]->stopped = 0;
2119 diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
2120 fwd_lcores[i], lc_id);
2123 "launch lcore %u failed - diag=%d\n",
2130 * Launch packet forwarding configuration.
2133 start_packet_forwarding(int with_tx_first)
2135 port_fwd_begin_t port_fwd_begin;
2136 port_fwd_end_t port_fwd_end;
2139 if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
2140 rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
2142 if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
2143 rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
2145 if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
2146 strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
2147 (!nb_rxq || !nb_txq))
2148 rte_exit(EXIT_FAILURE,
2149 "Either rxq or txq are 0, cannot use %s fwd mode\n",
2150 cur_fwd_eng->fwd_mode_name);
2152 if (all_ports_started() == 0) {
2153 fprintf(stderr, "Not all ports were started\n");
2156 if (test_done == 0) {
2157 fprintf(stderr, "Packet forwarding already started\n");
2165 flush_fwd_rx_queues();
2167 pkt_fwd_config_display(&cur_fwd_config);
2168 rxtx_config_display();
2171 if (with_tx_first) {
2172 port_fwd_begin = tx_only_engine.port_fwd_begin;
2173 if (port_fwd_begin != NULL) {
2174 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2175 (*port_fwd_begin)(fwd_ports_ids[i]);
2177 while (with_tx_first--) {
2178 launch_packet_forwarding(
2179 run_one_txonly_burst_on_core);
2180 rte_eal_mp_wait_lcore();
2182 port_fwd_end = tx_only_engine.port_fwd_end;
2183 if (port_fwd_end != NULL) {
2184 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2185 (*port_fwd_end)(fwd_ports_ids[i]);
2188 launch_packet_forwarding(start_pkt_forward_on_core);
2192 stop_packet_forwarding(void)
2194 port_fwd_end_t port_fwd_end;
2200 fprintf(stderr, "Packet forwarding not started\n");
2203 printf("Telling cores to stop...");
2204 for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
2205 fwd_lcores[lc_id]->stopped = 1;
2206 printf("\nWaiting for lcores to finish...\n");
2207 rte_eal_mp_wait_lcore();
2208 port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
2209 if (port_fwd_end != NULL) {
2210 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2211 pt_id = fwd_ports_ids[i];
2212 (*port_fwd_end)(pt_id);
2216 fwd_stats_display();
2218 printf("\nDone.\n");
2223 dev_set_link_up(portid_t pid)
2225 if (rte_eth_dev_set_link_up(pid) < 0)
2226 fprintf(stderr, "\nSet link up fail.\n");
2230 dev_set_link_down(portid_t pid)
2232 if (rte_eth_dev_set_link_down(pid) < 0)
2233 fprintf(stderr, "\nSet link down fail.\n");
2237 all_ports_started(void)
2240 struct rte_port *port;
2242 RTE_ETH_FOREACH_DEV(pi) {
2244 /* Check if there is a port which is not started */
2245 if ((port->port_status != RTE_PORT_STARTED) &&
2246 (port->slave_flag == 0))
2250 /* No port is not started */
2255 port_is_stopped(portid_t port_id)
2257 struct rte_port *port = &ports[port_id];
2259 if ((port->port_status != RTE_PORT_STOPPED) &&
2260 (port->slave_flag == 0))
2266 all_ports_stopped(void)
2270 RTE_ETH_FOREACH_DEV(pi) {
2271 if (!port_is_stopped(pi))
2279 port_is_started(portid_t port_id)
2281 if (port_id_is_invalid(port_id, ENABLED_WARN))
2284 if (ports[port_id].port_status != RTE_PORT_STARTED)
2290 /* Configure the Rx and Tx hairpin queues for the selected port. */
2292 setup_hairpin_queues(portid_t pi, portid_t p_pi, uint16_t cnt_pi)
2295 struct rte_eth_hairpin_conf hairpin_conf = {
2300 struct rte_port *port = &ports[pi];
2301 uint16_t peer_rx_port = pi;
2302 uint16_t peer_tx_port = pi;
2303 uint32_t manual = 1;
2304 uint32_t tx_exp = hairpin_mode & 0x10;
2306 if (!(hairpin_mode & 0xf)) {
2310 } else if (hairpin_mode & 0x1) {
2311 peer_tx_port = rte_eth_find_next_owned_by(pi + 1,
2312 RTE_ETH_DEV_NO_OWNER);
2313 if (peer_tx_port >= RTE_MAX_ETHPORTS)
2314 peer_tx_port = rte_eth_find_next_owned_by(0,
2315 RTE_ETH_DEV_NO_OWNER);
2316 if (p_pi != RTE_MAX_ETHPORTS) {
2317 peer_rx_port = p_pi;
2321 /* Last port will be the peer RX port of the first. */
2322 RTE_ETH_FOREACH_DEV(next_pi)
2323 peer_rx_port = next_pi;
2326 } else if (hairpin_mode & 0x2) {
2328 peer_rx_port = p_pi;
2330 peer_rx_port = rte_eth_find_next_owned_by(pi + 1,
2331 RTE_ETH_DEV_NO_OWNER);
2332 if (peer_rx_port >= RTE_MAX_ETHPORTS)
2335 peer_tx_port = peer_rx_port;
2339 for (qi = nb_txq, i = 0; qi < nb_hairpinq + nb_txq; qi++) {
2340 hairpin_conf.peers[0].port = peer_rx_port;
2341 hairpin_conf.peers[0].queue = i + nb_rxq;
2342 hairpin_conf.manual_bind = !!manual;
2343 hairpin_conf.tx_explicit = !!tx_exp;
2344 diag = rte_eth_tx_hairpin_queue_setup
2345 (pi, qi, nb_txd, &hairpin_conf);
2350 /* Fail to setup rx queue, return */
2351 if (rte_atomic16_cmpset(&(port->port_status),
2353 RTE_PORT_STOPPED) == 0)
2355 "Port %d can not be set back to stopped\n", pi);
2356 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2358 /* try to reconfigure queues next time */
2359 port->need_reconfig_queues = 1;
2362 for (qi = nb_rxq, i = 0; qi < nb_hairpinq + nb_rxq; qi++) {
2363 hairpin_conf.peers[0].port = peer_tx_port;
2364 hairpin_conf.peers[0].queue = i + nb_txq;
2365 hairpin_conf.manual_bind = !!manual;
2366 hairpin_conf.tx_explicit = !!tx_exp;
2367 diag = rte_eth_rx_hairpin_queue_setup
2368 (pi, qi, nb_rxd, &hairpin_conf);
2373 /* Fail to setup rx queue, return */
2374 if (rte_atomic16_cmpset(&(port->port_status),
2376 RTE_PORT_STOPPED) == 0)
2378 "Port %d can not be set back to stopped\n", pi);
2379 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2381 /* try to reconfigure queues next time */
2382 port->need_reconfig_queues = 1;
2388 /* Configure the Rx with optional split. */
2390 rx_queue_setup(uint16_t port_id, uint16_t rx_queue_id,
2391 uint16_t nb_rx_desc, unsigned int socket_id,
2392 struct rte_eth_rxconf *rx_conf, struct rte_mempool *mp)
2394 union rte_eth_rxseg rx_useg[MAX_SEGS_BUFFER_SPLIT] = {};
2395 unsigned int i, mp_n;
2398 if (rx_pkt_nb_segs <= 1 ||
2399 (rx_conf->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) == 0) {
2400 rx_conf->rx_seg = NULL;
2401 rx_conf->rx_nseg = 0;
2402 ret = rte_eth_rx_queue_setup(port_id, rx_queue_id,
2403 nb_rx_desc, socket_id,
2407 for (i = 0; i < rx_pkt_nb_segs; i++) {
2408 struct rte_eth_rxseg_split *rx_seg = &rx_useg[i].split;
2409 struct rte_mempool *mpx;
2411 * Use last valid pool for the segments with number
2412 * exceeding the pool index.
2414 mp_n = (i > mbuf_data_size_n) ? mbuf_data_size_n - 1 : i;
2415 mpx = mbuf_pool_find(socket_id, mp_n);
2416 /* Handle zero as mbuf data buffer size. */
2417 rx_seg->length = rx_pkt_seg_lengths[i] ?
2418 rx_pkt_seg_lengths[i] :
2419 mbuf_data_size[mp_n];
2420 rx_seg->offset = i < rx_pkt_nb_offs ?
2421 rx_pkt_seg_offsets[i] : 0;
2422 rx_seg->mp = mpx ? mpx : mp;
2424 rx_conf->rx_nseg = rx_pkt_nb_segs;
2425 rx_conf->rx_seg = rx_useg;
2426 ret = rte_eth_rx_queue_setup(port_id, rx_queue_id, nb_rx_desc,
2427 socket_id, rx_conf, NULL);
2428 rx_conf->rx_seg = NULL;
2429 rx_conf->rx_nseg = 0;
2434 start_port(portid_t pid)
2436 int diag, need_check_link_status = -1;
2438 portid_t p_pi = RTE_MAX_ETHPORTS;
2439 portid_t pl[RTE_MAX_ETHPORTS];
2440 portid_t peer_pl[RTE_MAX_ETHPORTS];
2441 uint16_t cnt_pi = 0;
2442 uint16_t cfg_pi = 0;
2445 struct rte_port *port;
2446 struct rte_eth_hairpin_cap cap;
2448 if (port_id_is_invalid(pid, ENABLED_WARN))
2451 RTE_ETH_FOREACH_DEV(pi) {
2452 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2455 need_check_link_status = 0;
2457 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
2458 RTE_PORT_HANDLING) == 0) {
2459 fprintf(stderr, "Port %d is now not stopped\n", pi);
2463 if (port->need_reconfig > 0) {
2464 port->need_reconfig = 0;
2466 if (flow_isolate_all) {
2467 int ret = port_flow_isolate(pi, 1);
2470 "Failed to apply isolated mode on port %d\n",
2475 configure_rxtx_dump_callbacks(0);
2476 printf("Configuring Port %d (socket %u)\n", pi,
2478 if (nb_hairpinq > 0 &&
2479 rte_eth_dev_hairpin_capability_get(pi, &cap)) {
2481 "Port %d doesn't support hairpin queues\n",
2485 /* configure port */
2486 diag = rte_eth_dev_configure(pi, nb_rxq + nb_hairpinq,
2487 nb_txq + nb_hairpinq,
2490 if (rte_atomic16_cmpset(&(port->port_status),
2491 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2493 "Port %d can not be set back to stopped\n",
2495 fprintf(stderr, "Fail to configure port %d\n",
2497 /* try to reconfigure port next time */
2498 port->need_reconfig = 1;
2502 if (port->need_reconfig_queues > 0) {
2503 port->need_reconfig_queues = 0;
2504 /* setup tx queues */
2505 for (qi = 0; qi < nb_txq; qi++) {
2506 if ((numa_support) &&
2507 (txring_numa[pi] != NUMA_NO_CONFIG))
2508 diag = rte_eth_tx_queue_setup(pi, qi,
2509 port->nb_tx_desc[qi],
2511 &(port->tx_conf[qi]));
2513 diag = rte_eth_tx_queue_setup(pi, qi,
2514 port->nb_tx_desc[qi],
2516 &(port->tx_conf[qi]));
2521 /* Fail to setup tx queue, return */
2522 if (rte_atomic16_cmpset(&(port->port_status),
2524 RTE_PORT_STOPPED) == 0)
2526 "Port %d can not be set back to stopped\n",
2529 "Fail to configure port %d tx queues\n",
2531 /* try to reconfigure queues next time */
2532 port->need_reconfig_queues = 1;
2535 for (qi = 0; qi < nb_rxq; qi++) {
2536 /* setup rx queues */
2537 if ((numa_support) &&
2538 (rxring_numa[pi] != NUMA_NO_CONFIG)) {
2539 struct rte_mempool * mp =
2541 (rxring_numa[pi], 0);
2544 "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2549 diag = rx_queue_setup(pi, qi,
2550 port->nb_rx_desc[qi],
2552 &(port->rx_conf[qi]),
2555 struct rte_mempool *mp =
2557 (port->socket_id, 0);
2560 "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2564 diag = rx_queue_setup(pi, qi,
2565 port->nb_rx_desc[qi],
2567 &(port->rx_conf[qi]),
2573 /* Fail to setup rx queue, return */
2574 if (rte_atomic16_cmpset(&(port->port_status),
2576 RTE_PORT_STOPPED) == 0)
2578 "Port %d can not be set back to stopped\n",
2581 "Fail to configure port %d rx queues\n",
2583 /* try to reconfigure queues next time */
2584 port->need_reconfig_queues = 1;
2587 /* setup hairpin queues */
2588 if (setup_hairpin_queues(pi, p_pi, cnt_pi) != 0)
2591 configure_rxtx_dump_callbacks(verbose_level);
2593 diag = rte_eth_dev_set_ptypes(pi, RTE_PTYPE_UNKNOWN,
2597 "Port %d: Failed to disable Ptype parsing\n",
2605 diag = rte_eth_dev_start(pi);
2607 fprintf(stderr, "Fail to start port %d: %s\n",
2608 pi, rte_strerror(-diag));
2610 /* Fail to setup rx queue, return */
2611 if (rte_atomic16_cmpset(&(port->port_status),
2612 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2614 "Port %d can not be set back to stopped\n",
2619 if (rte_atomic16_cmpset(&(port->port_status),
2620 RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2621 fprintf(stderr, "Port %d can not be set into started\n",
2624 if (eth_macaddr_get_print_err(pi, &port->eth_addr) == 0)
2625 printf("Port %d: %02X:%02X:%02X:%02X:%02X:%02X\n", pi,
2626 port->eth_addr.addr_bytes[0],
2627 port->eth_addr.addr_bytes[1],
2628 port->eth_addr.addr_bytes[2],
2629 port->eth_addr.addr_bytes[3],
2630 port->eth_addr.addr_bytes[4],
2631 port->eth_addr.addr_bytes[5]);
2633 /* at least one port started, need checking link status */
2634 need_check_link_status = 1;
2639 if (need_check_link_status == 1 && !no_link_check)
2640 check_all_ports_link_status(RTE_PORT_ALL);
2641 else if (need_check_link_status == 0)
2642 fprintf(stderr, "Please stop the ports first\n");
2644 if (hairpin_mode & 0xf) {
2648 /* bind all started hairpin ports */
2649 for (i = 0; i < cfg_pi; i++) {
2651 /* bind current Tx to all peer Rx */
2652 peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2653 RTE_MAX_ETHPORTS, 1);
2656 for (j = 0; j < peer_pi; j++) {
2657 if (!port_is_started(peer_pl[j]))
2659 diag = rte_eth_hairpin_bind(pi, peer_pl[j]);
2662 "Error during binding hairpin Tx port %u to %u: %s\n",
2664 rte_strerror(-diag));
2668 /* bind all peer Tx to current Rx */
2669 peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2670 RTE_MAX_ETHPORTS, 0);
2673 for (j = 0; j < peer_pi; j++) {
2674 if (!port_is_started(peer_pl[j]))
2676 diag = rte_eth_hairpin_bind(peer_pl[j], pi);
2679 "Error during binding hairpin Tx port %u to %u: %s\n",
2681 rte_strerror(-diag));
2693 stop_port(portid_t pid)
2696 struct rte_port *port;
2697 int need_check_link_status = 0;
2698 portid_t peer_pl[RTE_MAX_ETHPORTS];
2701 if (port_id_is_invalid(pid, ENABLED_WARN))
2704 printf("Stopping ports...\n");
2706 RTE_ETH_FOREACH_DEV(pi) {
2707 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2710 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2712 "Please remove port %d from forwarding configuration.\n",
2717 if (port_is_bonding_slave(pi)) {
2719 "Please remove port %d from bonded device.\n",
2725 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2726 RTE_PORT_HANDLING) == 0)
2729 if (hairpin_mode & 0xf) {
2732 rte_eth_hairpin_unbind(pi, RTE_MAX_ETHPORTS);
2733 /* unbind all peer Tx from 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 rte_eth_hairpin_unbind(peer_pl[j], pi);
2745 if (port->flow_list)
2746 port_flow_flush(pi);
2748 if (rte_eth_dev_stop(pi) != 0)
2749 RTE_LOG(ERR, EAL, "rte_eth_dev_stop failed for port %u\n",
2752 if (rte_atomic16_cmpset(&(port->port_status),
2753 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2754 fprintf(stderr, "Port %d can not be set into stopped\n",
2756 need_check_link_status = 1;
2758 if (need_check_link_status && !no_link_check)
2759 check_all_ports_link_status(RTE_PORT_ALL);
2765 remove_invalid_ports_in(portid_t *array, portid_t *total)
2768 portid_t new_total = 0;
2770 for (i = 0; i < *total; i++)
2771 if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2772 array[new_total] = array[i];
2779 remove_invalid_ports(void)
2781 remove_invalid_ports_in(ports_ids, &nb_ports);
2782 remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2783 nb_cfg_ports = nb_fwd_ports;
2787 close_port(portid_t pid)
2790 struct rte_port *port;
2792 if (port_id_is_invalid(pid, ENABLED_WARN))
2795 printf("Closing ports...\n");
2797 RTE_ETH_FOREACH_DEV(pi) {
2798 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2801 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2803 "Please remove port %d from forwarding configuration.\n",
2808 if (port_is_bonding_slave(pi)) {
2810 "Please remove port %d from bonded device.\n",
2816 if (rte_atomic16_cmpset(&(port->port_status),
2817 RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2818 fprintf(stderr, "Port %d is already closed\n", pi);
2822 port_flow_flush(pi);
2823 rte_eth_dev_close(pi);
2826 remove_invalid_ports();
2831 reset_port(portid_t pid)
2835 struct rte_port *port;
2837 if (port_id_is_invalid(pid, ENABLED_WARN))
2840 if ((pid == (portid_t)RTE_PORT_ALL && !all_ports_stopped()) ||
2841 (pid != (portid_t)RTE_PORT_ALL && !port_is_stopped(pid))) {
2843 "Can not reset port(s), please stop port(s) first.\n");
2847 printf("Resetting ports...\n");
2849 RTE_ETH_FOREACH_DEV(pi) {
2850 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2853 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2855 "Please remove port %d from forwarding configuration.\n",
2860 if (port_is_bonding_slave(pi)) {
2862 "Please remove port %d from bonded device.\n",
2867 diag = rte_eth_dev_reset(pi);
2870 port->need_reconfig = 1;
2871 port->need_reconfig_queues = 1;
2873 fprintf(stderr, "Failed to reset port %d. diag=%d\n",
2882 attach_port(char *identifier)
2885 struct rte_dev_iterator iterator;
2887 printf("Attaching a new port...\n");
2889 if (identifier == NULL) {
2890 fprintf(stderr, "Invalid parameters are specified\n");
2894 if (rte_dev_probe(identifier) < 0) {
2895 TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2899 /* first attach mode: event */
2900 if (setup_on_probe_event) {
2901 /* new ports are detected on RTE_ETH_EVENT_NEW event */
2902 for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2903 if (ports[pi].port_status == RTE_PORT_HANDLING &&
2904 ports[pi].need_setup != 0)
2905 setup_attached_port(pi);
2909 /* second attach mode: iterator */
2910 RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2911 /* setup ports matching the devargs used for probing */
2912 if (port_is_forwarding(pi))
2913 continue; /* port was already attached before */
2914 setup_attached_port(pi);
2919 setup_attached_port(portid_t pi)
2921 unsigned int socket_id;
2924 socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2925 /* if socket_id is invalid, set to the first available socket. */
2926 if (check_socket_id(socket_id) < 0)
2927 socket_id = socket_ids[0];
2928 reconfig(pi, socket_id);
2929 ret = rte_eth_promiscuous_enable(pi);
2932 "Error during enabling promiscuous mode for port %u: %s - ignore\n",
2933 pi, rte_strerror(-ret));
2935 ports_ids[nb_ports++] = pi;
2936 fwd_ports_ids[nb_fwd_ports++] = pi;
2937 nb_cfg_ports = nb_fwd_ports;
2938 ports[pi].need_setup = 0;
2939 ports[pi].port_status = RTE_PORT_STOPPED;
2941 printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
2946 detach_device(struct rte_device *dev)
2951 fprintf(stderr, "Device already removed\n");
2955 printf("Removing a device...\n");
2957 RTE_ETH_FOREACH_DEV_OF(sibling, dev) {
2958 if (ports[sibling].port_status != RTE_PORT_CLOSED) {
2959 if (ports[sibling].port_status != RTE_PORT_STOPPED) {
2960 fprintf(stderr, "Port %u not stopped\n",
2964 port_flow_flush(sibling);
2968 if (rte_dev_remove(dev) < 0) {
2969 TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
2972 remove_invalid_ports();
2974 printf("Device is detached\n");
2975 printf("Now total ports is %d\n", nb_ports);
2981 detach_port_device(portid_t port_id)
2984 struct rte_eth_dev_info dev_info;
2986 if (port_id_is_invalid(port_id, ENABLED_WARN))
2989 if (ports[port_id].port_status != RTE_PORT_CLOSED) {
2990 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
2991 fprintf(stderr, "Port not stopped\n");
2994 fprintf(stderr, "Port was not closed\n");
2997 ret = eth_dev_info_get_print_err(port_id, &dev_info);
3000 "Failed to get device info for port %d, not detaching\n",
3004 detach_device(dev_info.device);
3008 detach_devargs(char *identifier)
3010 struct rte_dev_iterator iterator;
3011 struct rte_devargs da;
3014 printf("Removing a device...\n");
3016 memset(&da, 0, sizeof(da));
3017 if (rte_devargs_parsef(&da, "%s", identifier)) {
3018 fprintf(stderr, "cannot parse identifier\n");
3022 RTE_ETH_FOREACH_MATCHING_DEV(port_id, identifier, &iterator) {
3023 if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3024 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3025 fprintf(stderr, "Port %u not stopped\n",
3027 rte_eth_iterator_cleanup(&iterator);
3028 rte_devargs_reset(&da);
3031 port_flow_flush(port_id);
3035 if (rte_eal_hotplug_remove(da.bus->name, da.name) != 0) {
3036 TESTPMD_LOG(ERR, "Failed to detach device %s(%s)\n",
3037 da.name, da.bus->name);
3038 rte_devargs_reset(&da);
3042 remove_invalid_ports();
3044 printf("Device %s is detached\n", identifier);
3045 printf("Now total ports is %d\n", nb_ports);
3047 rte_devargs_reset(&da);
3058 stop_packet_forwarding();
3060 #ifndef RTE_EXEC_ENV_WINDOWS
3061 for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3063 if (mp_alloc_type == MP_ALLOC_ANON)
3064 rte_mempool_mem_iter(mempools[i], dma_unmap_cb,
3069 if (ports != NULL) {
3071 RTE_ETH_FOREACH_DEV(pt_id) {
3072 printf("\nStopping port %d...\n", pt_id);
3076 RTE_ETH_FOREACH_DEV(pt_id) {
3077 printf("\nShutting down port %d...\n", pt_id);
3084 ret = rte_dev_event_monitor_stop();
3087 "fail to stop device event monitor.");
3091 ret = rte_dev_event_callback_unregister(NULL,
3092 dev_event_callback, NULL);
3095 "fail to unregister device event callback.\n");
3099 ret = rte_dev_hotplug_handle_disable();
3102 "fail to disable hotplug handling.\n");
3106 for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3108 rte_mempool_free(mempools[i]);
3111 printf("\nBye...\n");
3114 typedef void (*cmd_func_t)(void);
3115 struct pmd_test_command {
3116 const char *cmd_name;
3117 cmd_func_t cmd_func;
3120 /* Check the link status of all ports in up to 9s, and print them finally */
3122 check_all_ports_link_status(uint32_t port_mask)
3124 #define CHECK_INTERVAL 100 /* 100ms */
3125 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3127 uint8_t count, all_ports_up, print_flag = 0;
3128 struct rte_eth_link link;
3130 char link_status[RTE_ETH_LINK_MAX_STR_LEN];
3132 printf("Checking link statuses...\n");
3134 for (count = 0; count <= MAX_CHECK_TIME; count++) {
3136 RTE_ETH_FOREACH_DEV(portid) {
3137 if ((port_mask & (1 << portid)) == 0)
3139 memset(&link, 0, sizeof(link));
3140 ret = rte_eth_link_get_nowait(portid, &link);
3143 if (print_flag == 1)
3145 "Port %u link get failed: %s\n",
3146 portid, rte_strerror(-ret));
3149 /* print link status if flag set */
3150 if (print_flag == 1) {
3151 rte_eth_link_to_str(link_status,
3152 sizeof(link_status), &link);
3153 printf("Port %d %s\n", portid, link_status);
3156 /* clear all_ports_up flag if any link down */
3157 if (link.link_status == ETH_LINK_DOWN) {
3162 /* after finally printing all link status, get out */
3163 if (print_flag == 1)
3166 if (all_ports_up == 0) {
3168 rte_delay_ms(CHECK_INTERVAL);
3171 /* set the print_flag if all ports up or timeout */
3172 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3182 rmv_port_callback(void *arg)
3184 int need_to_start = 0;
3185 int org_no_link_check = no_link_check;
3186 portid_t port_id = (intptr_t)arg;
3187 struct rte_eth_dev_info dev_info;
3190 RTE_ETH_VALID_PORTID_OR_RET(port_id);
3192 if (!test_done && port_is_forwarding(port_id)) {
3194 stop_packet_forwarding();
3198 no_link_check = org_no_link_check;
3200 close_port(port_id);
3201 ret = eth_dev_info_get_print_err(port_id, &dev_info);
3204 "Failed to get device info for port %d, not detaching\n",
3207 detach_device(dev_info.device); /* might be already removed or have more ports */
3209 start_packet_forwarding(0);
3212 /* This function is used by the interrupt thread */
3214 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
3217 RTE_SET_USED(param);
3218 RTE_SET_USED(ret_param);
3220 if (type >= RTE_ETH_EVENT_MAX) {
3222 "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
3223 port_id, __func__, type);
3225 } else if (event_print_mask & (UINT32_C(1) << type)) {
3226 printf("\nPort %" PRIu16 ": %s event\n", port_id,
3227 eth_event_desc[type]);
3232 case RTE_ETH_EVENT_NEW:
3233 ports[port_id].need_setup = 1;
3234 ports[port_id].port_status = RTE_PORT_HANDLING;
3236 case RTE_ETH_EVENT_INTR_RMV:
3237 if (port_id_is_invalid(port_id, DISABLED_WARN))
3239 if (rte_eal_alarm_set(100000,
3240 rmv_port_callback, (void *)(intptr_t)port_id))
3242 "Could not set up deferred device removal\n");
3244 case RTE_ETH_EVENT_DESTROY:
3245 ports[port_id].port_status = RTE_PORT_CLOSED;
3246 printf("Port %u is closed\n", port_id);
3255 register_eth_event_callback(void)
3258 enum rte_eth_event_type event;
3260 for (event = RTE_ETH_EVENT_UNKNOWN;
3261 event < RTE_ETH_EVENT_MAX; event++) {
3262 ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
3267 TESTPMD_LOG(ERR, "Failed to register callback for "
3268 "%s event\n", eth_event_desc[event]);
3276 /* This function is used by the interrupt thread */
3278 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
3279 __rte_unused void *arg)
3284 if (type >= RTE_DEV_EVENT_MAX) {
3285 fprintf(stderr, "%s called upon invalid event %d\n",
3291 case RTE_DEV_EVENT_REMOVE:
3292 RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
3294 ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
3296 RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
3301 * Because the user's callback is invoked in eal interrupt
3302 * callback, the interrupt callback need to be finished before
3303 * it can be unregistered when detaching device. So finish
3304 * callback soon and use a deferred removal to detach device
3305 * is need. It is a workaround, once the device detaching be
3306 * moved into the eal in the future, the deferred removal could
3309 if (rte_eal_alarm_set(100000,
3310 rmv_port_callback, (void *)(intptr_t)port_id))
3312 "Could not set up deferred device removal\n");
3314 case RTE_DEV_EVENT_ADD:
3315 RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
3317 /* TODO: After finish kernel driver binding,
3318 * begin to attach port.
3327 rxtx_port_config(struct rte_port *port)
3332 for (qid = 0; qid < nb_rxq; qid++) {
3333 offloads = port->rx_conf[qid].offloads;
3334 port->rx_conf[qid] = port->dev_info.default_rxconf;
3336 port->rx_conf[qid].offloads = offloads;
3338 /* Check if any Rx parameters have been passed */
3339 if (rx_pthresh != RTE_PMD_PARAM_UNSET)
3340 port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
3342 if (rx_hthresh != RTE_PMD_PARAM_UNSET)
3343 port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
3345 if (rx_wthresh != RTE_PMD_PARAM_UNSET)
3346 port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
3348 if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
3349 port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
3351 if (rx_drop_en != RTE_PMD_PARAM_UNSET)
3352 port->rx_conf[qid].rx_drop_en = rx_drop_en;
3354 port->nb_rx_desc[qid] = nb_rxd;
3357 for (qid = 0; qid < nb_txq; qid++) {
3358 offloads = port->tx_conf[qid].offloads;
3359 port->tx_conf[qid] = port->dev_info.default_txconf;
3361 port->tx_conf[qid].offloads = offloads;
3363 /* Check if any Tx parameters have been passed */
3364 if (tx_pthresh != RTE_PMD_PARAM_UNSET)
3365 port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
3367 if (tx_hthresh != RTE_PMD_PARAM_UNSET)
3368 port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
3370 if (tx_wthresh != RTE_PMD_PARAM_UNSET)
3371 port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
3373 if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
3374 port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
3376 if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
3377 port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
3379 port->nb_tx_desc[qid] = nb_txd;
3384 * Helper function to arrange max_rx_pktlen value and JUMBO_FRAME offload,
3385 * MTU is also aligned if JUMBO_FRAME offload is not set.
3387 * port->dev_info should be set before calling this function.
3389 * return 0 on success, negative on error
3392 update_jumbo_frame_offload(portid_t portid)
3394 struct rte_port *port = &ports[portid];
3395 uint32_t eth_overhead;
3396 uint64_t rx_offloads;
3400 /* Update the max_rx_pkt_len to have MTU as RTE_ETHER_MTU */
3401 if (port->dev_info.max_mtu != UINT16_MAX &&
3402 port->dev_info.max_rx_pktlen > port->dev_info.max_mtu)
3403 eth_overhead = port->dev_info.max_rx_pktlen -
3404 port->dev_info.max_mtu;
3406 eth_overhead = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
3408 rx_offloads = port->dev_conf.rxmode.offloads;
3410 /* Default config value is 0 to use PMD specific overhead */
3411 if (port->dev_conf.rxmode.max_rx_pkt_len == 0)
3412 port->dev_conf.rxmode.max_rx_pkt_len = RTE_ETHER_MTU + eth_overhead;
3414 if (port->dev_conf.rxmode.max_rx_pkt_len <= RTE_ETHER_MTU + eth_overhead) {
3415 rx_offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3418 if ((port->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_JUMBO_FRAME) == 0) {
3420 "Frame size (%u) is not supported by port %u\n",
3421 port->dev_conf.rxmode.max_rx_pkt_len,
3425 rx_offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
3429 if (rx_offloads != port->dev_conf.rxmode.offloads) {
3432 port->dev_conf.rxmode.offloads = rx_offloads;
3434 /* Apply JUMBO_FRAME offload configuration to Rx queue(s) */
3435 for (qid = 0; qid < port->dev_info.nb_rx_queues; qid++) {
3437 port->rx_conf[qid].offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
3439 port->rx_conf[qid].offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3443 /* If JUMBO_FRAME is set MTU conversion done by ethdev layer,
3444 * if unset do it here
3446 if ((rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) == 0) {
3447 ret = rte_eth_dev_set_mtu(portid,
3448 port->dev_conf.rxmode.max_rx_pkt_len - eth_overhead);
3451 "Failed to set MTU to %u for port %u\n",
3452 port->dev_conf.rxmode.max_rx_pkt_len - eth_overhead,
3460 init_port_config(void)
3463 struct rte_port *port;
3466 RTE_ETH_FOREACH_DEV(pid) {
3468 port->dev_conf.fdir_conf = fdir_conf;
3470 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
3475 port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3476 port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
3477 rss_hf & port->dev_info.flow_type_rss_offloads;
3479 port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3480 port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
3483 if (port->dcb_flag == 0) {
3484 if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
3485 port->dev_conf.rxmode.mq_mode =
3486 (enum rte_eth_rx_mq_mode)
3487 (rx_mq_mode & ETH_MQ_RX_RSS);
3489 port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
3492 rxtx_port_config(port);
3494 ret = eth_macaddr_get_print_err(pid, &port->eth_addr);
3498 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
3499 rte_pmd_ixgbe_bypass_init(pid);
3502 if (lsc_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_LSC))
3503 port->dev_conf.intr_conf.lsc = 1;
3504 if (rmv_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_RMV))
3505 port->dev_conf.intr_conf.rmv = 1;
3509 void set_port_slave_flag(portid_t slave_pid)
3511 struct rte_port *port;
3513 port = &ports[slave_pid];
3514 port->slave_flag = 1;
3517 void clear_port_slave_flag(portid_t slave_pid)
3519 struct rte_port *port;
3521 port = &ports[slave_pid];
3522 port->slave_flag = 0;
3525 uint8_t port_is_bonding_slave(portid_t slave_pid)
3527 struct rte_port *port;
3528 struct rte_eth_dev_info dev_info;
3531 port = &ports[slave_pid];
3532 ret = eth_dev_info_get_print_err(slave_pid, &dev_info);
3535 "Failed to get device info for port id %d,"
3536 "cannot determine if the port is a bonded slave",
3540 if ((*dev_info.dev_flags & RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
3545 const uint16_t vlan_tags[] = {
3546 0, 1, 2, 3, 4, 5, 6, 7,
3547 8, 9, 10, 11, 12, 13, 14, 15,
3548 16, 17, 18, 19, 20, 21, 22, 23,
3549 24, 25, 26, 27, 28, 29, 30, 31
3553 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
3554 enum dcb_mode_enable dcb_mode,
3555 enum rte_eth_nb_tcs num_tcs,
3560 struct rte_eth_rss_conf rss_conf;
3563 * Builds up the correct configuration for dcb+vt based on the vlan tags array
3564 * given above, and the number of traffic classes available for use.
3566 if (dcb_mode == DCB_VT_ENABLED) {
3567 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3568 ð_conf->rx_adv_conf.vmdq_dcb_conf;
3569 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3570 ð_conf->tx_adv_conf.vmdq_dcb_tx_conf;
3572 /* VMDQ+DCB RX and TX configurations */
3573 vmdq_rx_conf->enable_default_pool = 0;
3574 vmdq_rx_conf->default_pool = 0;
3575 vmdq_rx_conf->nb_queue_pools =
3576 (num_tcs == ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3577 vmdq_tx_conf->nb_queue_pools =
3578 (num_tcs == ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3580 vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
3581 for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
3582 vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
3583 vmdq_rx_conf->pool_map[i].pools =
3584 1 << (i % vmdq_rx_conf->nb_queue_pools);
3586 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3587 vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
3588 vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
3591 /* set DCB mode of RX and TX of multiple queues */
3592 eth_conf->rxmode.mq_mode =
3593 (enum rte_eth_rx_mq_mode)
3594 (rx_mq_mode & ETH_MQ_RX_VMDQ_DCB);
3595 eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
3597 struct rte_eth_dcb_rx_conf *rx_conf =
3598 ð_conf->rx_adv_conf.dcb_rx_conf;
3599 struct rte_eth_dcb_tx_conf *tx_conf =
3600 ð_conf->tx_adv_conf.dcb_tx_conf;
3602 memset(&rss_conf, 0, sizeof(struct rte_eth_rss_conf));
3604 rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
3608 rx_conf->nb_tcs = num_tcs;
3609 tx_conf->nb_tcs = num_tcs;
3611 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3612 rx_conf->dcb_tc[i] = i % num_tcs;
3613 tx_conf->dcb_tc[i] = i % num_tcs;
3616 eth_conf->rxmode.mq_mode =
3617 (enum rte_eth_rx_mq_mode)
3618 (rx_mq_mode & ETH_MQ_RX_DCB_RSS);
3619 eth_conf->rx_adv_conf.rss_conf = rss_conf;
3620 eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
3624 eth_conf->dcb_capability_en =
3625 ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
3627 eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
3633 init_port_dcb_config(portid_t pid,
3634 enum dcb_mode_enable dcb_mode,
3635 enum rte_eth_nb_tcs num_tcs,
3638 struct rte_eth_conf port_conf;
3639 struct rte_port *rte_port;
3643 rte_port = &ports[pid];
3645 memset(&port_conf, 0, sizeof(struct rte_eth_conf));
3647 port_conf.rxmode = rte_port->dev_conf.rxmode;
3648 port_conf.txmode = rte_port->dev_conf.txmode;
3650 /*set configuration of DCB in vt mode and DCB in non-vt mode*/
3651 retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
3654 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3656 /* re-configure the device . */
3657 retval = rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
3661 retval = eth_dev_info_get_print_err(pid, &rte_port->dev_info);
3665 /* If dev_info.vmdq_pool_base is greater than 0,
3666 * the queue id of vmdq pools is started after pf queues.
3668 if (dcb_mode == DCB_VT_ENABLED &&
3669 rte_port->dev_info.vmdq_pool_base > 0) {
3671 "VMDQ_DCB multi-queue mode is nonsensical for port %d.\n",
3676 /* Assume the ports in testpmd have the same dcb capability
3677 * and has the same number of rxq and txq in dcb mode
3679 if (dcb_mode == DCB_VT_ENABLED) {
3680 if (rte_port->dev_info.max_vfs > 0) {
3681 nb_rxq = rte_port->dev_info.nb_rx_queues;
3682 nb_txq = rte_port->dev_info.nb_tx_queues;
3684 nb_rxq = rte_port->dev_info.max_rx_queues;
3685 nb_txq = rte_port->dev_info.max_tx_queues;
3688 /*if vt is disabled, use all pf queues */
3689 if (rte_port->dev_info.vmdq_pool_base == 0) {
3690 nb_rxq = rte_port->dev_info.max_rx_queues;
3691 nb_txq = rte_port->dev_info.max_tx_queues;
3693 nb_rxq = (queueid_t)num_tcs;
3694 nb_txq = (queueid_t)num_tcs;
3698 rx_free_thresh = 64;
3700 memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
3702 rxtx_port_config(rte_port);
3704 rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3705 for (i = 0; i < RTE_DIM(vlan_tags); i++)
3706 rx_vft_set(pid, vlan_tags[i], 1);
3708 retval = eth_macaddr_get_print_err(pid, &rte_port->eth_addr);
3712 rte_port->dcb_flag = 1;
3714 /* Enter DCB configuration status */
3725 /* Configuration of Ethernet ports. */
3726 ports = rte_zmalloc("testpmd: ports",
3727 sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3728 RTE_CACHE_LINE_SIZE);
3729 if (ports == NULL) {
3730 rte_exit(EXIT_FAILURE,
3731 "rte_zmalloc(%d struct rte_port) failed\n",
3734 for (i = 0; i < RTE_MAX_ETHPORTS; i++)
3735 LIST_INIT(&ports[i].flow_tunnel_list);
3736 /* Initialize ports NUMA structures */
3737 memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3738 memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3739 memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3753 const char clr[] = { 27, '[', '2', 'J', '\0' };
3754 const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3756 /* Clear screen and move to top left */
3757 printf("%s%s", clr, top_left);
3759 printf("\nPort statistics ====================================");
3760 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3761 nic_stats_display(fwd_ports_ids[i]);
3767 signal_handler(int signum)
3769 if (signum == SIGINT || signum == SIGTERM) {
3770 fprintf(stderr, "\nSignal %d received, preparing to exit...\n",
3772 #ifdef RTE_LIB_PDUMP
3773 /* uninitialize packet capture framework */
3776 #ifdef RTE_LIB_LATENCYSTATS
3777 if (latencystats_enabled != 0)
3778 rte_latencystats_uninit();
3781 /* Set flag to indicate the force termination. */
3783 /* exit with the expected status */
3784 #ifndef RTE_EXEC_ENV_WINDOWS
3785 signal(signum, SIG_DFL);
3786 kill(getpid(), signum);
3792 main(int argc, char** argv)
3799 signal(SIGINT, signal_handler);
3800 signal(SIGTERM, signal_handler);
3802 testpmd_logtype = rte_log_register("testpmd");
3803 if (testpmd_logtype < 0)
3804 rte_exit(EXIT_FAILURE, "Cannot register log type");
3805 rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3807 diag = rte_eal_init(argc, argv);
3809 rte_exit(EXIT_FAILURE, "Cannot init EAL: %s\n",
3810 rte_strerror(rte_errno));
3812 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
3813 rte_exit(EXIT_FAILURE,
3814 "Secondary process type not supported.\n");
3816 ret = register_eth_event_callback();
3818 rte_exit(EXIT_FAILURE, "Cannot register for ethdev events");
3820 #ifdef RTE_LIB_PDUMP
3821 /* initialize packet capture framework */
3826 RTE_ETH_FOREACH_DEV(port_id) {
3827 ports_ids[count] = port_id;
3830 nb_ports = (portid_t) count;
3832 TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3834 /* allocate port structures, and init them */
3837 set_def_fwd_config();
3839 rte_exit(EXIT_FAILURE, "No cores defined for forwarding\n"
3840 "Check the core mask argument\n");
3842 /* Bitrate/latency stats disabled by default */
3843 #ifdef RTE_LIB_BITRATESTATS
3844 bitrate_enabled = 0;
3846 #ifdef RTE_LIB_LATENCYSTATS
3847 latencystats_enabled = 0;
3850 /* on FreeBSD, mlockall() is disabled by default */
3851 #ifdef RTE_EXEC_ENV_FREEBSD
3860 launch_args_parse(argc, argv);
3862 #ifndef RTE_EXEC_ENV_WINDOWS
3863 if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3864 TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3869 if (tx_first && interactive)
3870 rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3871 "interactive mode.\n");
3873 if (tx_first && lsc_interrupt) {
3875 "Warning: lsc_interrupt needs to be off when using tx_first. Disabling.\n");
3879 if (!nb_rxq && !nb_txq)
3881 "Warning: Either rx or tx queues should be non-zero\n");
3883 if (nb_rxq > 1 && nb_rxq > nb_txq)
3885 "Warning: nb_rxq=%d enables RSS configuration, but nb_txq=%d will prevent to fully test it.\n",
3891 ret = rte_dev_hotplug_handle_enable();
3894 "fail to enable hotplug handling.");
3898 ret = rte_dev_event_monitor_start();
3901 "fail to start device event monitoring.");
3905 ret = rte_dev_event_callback_register(NULL,
3906 dev_event_callback, NULL);
3909 "fail to register device event callback\n");
3914 if (!no_device_start && start_port(RTE_PORT_ALL) != 0)
3915 rte_exit(EXIT_FAILURE, "Start ports failed\n");
3917 /* set all ports to promiscuous mode by default */
3918 RTE_ETH_FOREACH_DEV(port_id) {
3919 ret = rte_eth_promiscuous_enable(port_id);
3922 "Error during enabling promiscuous mode for port %u: %s - ignore\n",
3923 port_id, rte_strerror(-ret));
3926 /* Init metrics library */
3927 rte_metrics_init(rte_socket_id());
3929 #ifdef RTE_LIB_LATENCYSTATS
3930 if (latencystats_enabled != 0) {
3931 int ret = rte_latencystats_init(1, NULL);
3934 "Warning: latencystats init() returned error %d\n",
3936 fprintf(stderr, "Latencystats running on lcore %d\n",
3937 latencystats_lcore_id);
3941 /* Setup bitrate stats */
3942 #ifdef RTE_LIB_BITRATESTATS
3943 if (bitrate_enabled != 0) {
3944 bitrate_data = rte_stats_bitrate_create();
3945 if (bitrate_data == NULL)
3946 rte_exit(EXIT_FAILURE,
3947 "Could not allocate bitrate data.\n");
3948 rte_stats_bitrate_reg(bitrate_data);
3952 #ifdef RTE_LIB_CMDLINE
3953 if (strlen(cmdline_filename) != 0)
3954 cmdline_read_from_file(cmdline_filename);
3956 if (interactive == 1) {
3958 printf("Start automatic packet forwarding\n");
3959 start_packet_forwarding(0);
3971 printf("No commandline core given, start packet forwarding\n");
3972 start_packet_forwarding(tx_first);
3973 if (stats_period != 0) {
3974 uint64_t prev_time = 0, cur_time, diff_time = 0;
3975 uint64_t timer_period;
3977 /* Convert to number of cycles */
3978 timer_period = stats_period * rte_get_timer_hz();
3980 while (f_quit == 0) {
3981 cur_time = rte_get_timer_cycles();
3982 diff_time += cur_time - prev_time;
3984 if (diff_time >= timer_period) {
3986 /* Reset the timer */
3989 /* Sleep to avoid unnecessary checks */
3990 prev_time = cur_time;
3991 rte_delay_us_sleep(US_PER_S);
3995 printf("Press enter to exit\n");
3996 rc = read(0, &c, 1);
4002 ret = rte_eal_cleanup();
4004 rte_exit(EXIT_FAILURE,
4005 "EAL cleanup failed: %s\n", strerror(-ret));
4007 return EXIT_SUCCESS;