1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2017 Intel Corporation
13 #include <sys/types.h>
17 #include <sys/queue.h>
24 #include <rte_common.h>
25 #include <rte_errno.h>
26 #include <rte_byteorder.h>
28 #include <rte_debug.h>
29 #include <rte_cycles.h>
30 #include <rte_memory.h>
31 #include <rte_memcpy.h>
32 #include <rte_launch.h>
34 #include <rte_alarm.h>
35 #include <rte_per_lcore.h>
36 #include <rte_lcore.h>
37 #include <rte_atomic.h>
38 #include <rte_branch_prediction.h>
39 #include <rte_mempool.h>
40 #include <rte_malloc.h>
42 #include <rte_mbuf_pool_ops.h>
43 #include <rte_interrupts.h>
45 #include <rte_ether.h>
46 #include <rte_ethdev.h>
48 #include <rte_string_fns.h>
49 #ifdef RTE_LIBRTE_IXGBE_PMD
50 #include <rte_pmd_ixgbe.h>
52 #ifdef RTE_LIBRTE_PDUMP
53 #include <rte_pdump.h>
56 #include <rte_metrics.h>
57 #ifdef RTE_LIBRTE_BITRATESTATS
58 #include <rte_bitrate.h>
60 #ifdef RTE_LIBRTE_LATENCY_STATS
61 #include <rte_latencystats.h>
67 /* FreeBSD may not have MAP_HUGETLB (in fact, it probably doesn't) */
68 #define HUGE_FLAG (0x40000)
70 #define HUGE_FLAG MAP_HUGETLB
73 #ifndef MAP_HUGE_SHIFT
74 /* older kernels (or FreeBSD) will not have this define */
75 #define HUGE_SHIFT (26)
77 #define HUGE_SHIFT MAP_HUGE_SHIFT
80 #define EXTMEM_HEAP_NAME "extmem"
81 #define EXTBUF_ZONE_SIZE RTE_PGSIZE_2M
83 uint16_t verbose_level = 0; /**< Silent by default. */
84 int testpmd_logtype; /**< Log type for testpmd logs */
86 /* use master core for command line ? */
87 uint8_t interactive = 0;
88 uint8_t auto_start = 0;
90 char cmdline_filename[PATH_MAX] = {0};
93 * NUMA support configuration.
94 * When set, the NUMA support attempts to dispatch the allocation of the
95 * RX and TX memory rings, and of the DMA memory buffers (mbufs) for the
96 * probed ports among the CPU sockets 0 and 1.
97 * Otherwise, all memory is allocated from CPU socket 0.
99 uint8_t numa_support = 1; /**< numa enabled by default */
102 * In UMA mode,all memory is allocated from socket 0 if --socket-num is
105 uint8_t socket_num = UMA_NO_CONFIG;
108 * Select mempool allocation type:
109 * - native: use regular DPDK memory
110 * - anon: use regular DPDK memory to create mempool, but populate using
111 * anonymous memory (may not be IOVA-contiguous)
112 * - xmem: use externally allocated hugepage memory
114 uint8_t mp_alloc_type = MP_ALLOC_NATIVE;
117 * Store specified sockets on which memory pool to be used by ports
120 uint8_t port_numa[RTE_MAX_ETHPORTS];
123 * Store specified sockets on which RX ring to be used by ports
126 uint8_t rxring_numa[RTE_MAX_ETHPORTS];
129 * Store specified sockets on which TX ring to be used by ports
132 uint8_t txring_numa[RTE_MAX_ETHPORTS];
135 * Record the Ethernet address of peer target ports to which packets are
137 * Must be instantiated with the ethernet addresses of peer traffic generator
140 struct rte_ether_addr peer_eth_addrs[RTE_MAX_ETHPORTS];
141 portid_t nb_peer_eth_addrs = 0;
144 * Probed Target Environment.
146 struct rte_port *ports; /**< For all probed ethernet ports. */
147 portid_t nb_ports; /**< Number of probed ethernet ports. */
148 struct fwd_lcore **fwd_lcores; /**< For all probed logical cores. */
149 lcoreid_t nb_lcores; /**< Number of probed logical cores. */
151 portid_t ports_ids[RTE_MAX_ETHPORTS]; /**< Store all port ids. */
154 * Test Forwarding Configuration.
155 * nb_fwd_lcores <= nb_cfg_lcores <= nb_lcores
156 * nb_fwd_ports <= nb_cfg_ports <= nb_ports
158 lcoreid_t nb_cfg_lcores; /**< Number of configured logical cores. */
159 lcoreid_t nb_fwd_lcores; /**< Number of forwarding logical cores. */
160 portid_t nb_cfg_ports; /**< Number of configured ports. */
161 portid_t nb_fwd_ports; /**< Number of forwarding ports. */
163 unsigned int fwd_lcores_cpuids[RTE_MAX_LCORE]; /**< CPU ids configuration. */
164 portid_t fwd_ports_ids[RTE_MAX_ETHPORTS]; /**< Port ids configuration. */
166 struct fwd_stream **fwd_streams; /**< For each RX queue of each port. */
167 streamid_t nb_fwd_streams; /**< Is equal to (nb_ports * nb_rxq). */
170 * Forwarding engines.
172 struct fwd_engine * fwd_engines[] = {
182 &five_tuple_swap_fwd_engine,
183 #ifdef RTE_LIBRTE_IEEE1588
184 &ieee1588_fwd_engine,
189 struct rte_mempool *mempools[RTE_MAX_NUMA_NODES * MAX_SEGS_BUFFER_SPLIT];
190 uint16_t mempool_flags;
192 struct fwd_config cur_fwd_config;
193 struct fwd_engine *cur_fwd_eng = &io_fwd_engine; /**< IO mode by default. */
194 uint32_t retry_enabled;
195 uint32_t burst_tx_delay_time = BURST_TX_WAIT_US;
196 uint32_t burst_tx_retry_num = BURST_TX_RETRIES;
198 uint32_t mbuf_data_size_n = 1; /* Number of specified mbuf sizes. */
199 uint16_t mbuf_data_size[MAX_SEGS_BUFFER_SPLIT] = {
200 DEFAULT_MBUF_DATA_SIZE
201 }; /**< Mbuf data space size. */
202 uint32_t param_total_num_mbufs = 0; /**< number of mbufs in all pools - if
203 * specified on command-line. */
204 uint16_t stats_period; /**< Period to show statistics (disabled by default) */
207 * In container, it cannot terminate the process which running with 'stats-period'
208 * option. Set flag to exit stats period loop after received SIGINT/SIGTERM.
213 * Configuration of packet segments used to scatter received packets
214 * if some of split features is configured.
216 uint16_t rx_pkt_seg_lengths[MAX_SEGS_BUFFER_SPLIT];
217 uint8_t rx_pkt_nb_segs; /**< Number of segments to split */
218 uint16_t rx_pkt_seg_offsets[MAX_SEGS_BUFFER_SPLIT];
219 uint8_t rx_pkt_nb_offs; /**< Number of specified offsets */
222 * Configuration of packet segments used by the "txonly" processing engine.
224 uint16_t tx_pkt_length = TXONLY_DEF_PACKET_LEN; /**< TXONLY packet length. */
225 uint16_t tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT] = {
226 TXONLY_DEF_PACKET_LEN,
228 uint8_t tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY packets */
230 enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;
231 /**< Split policy for packets to TX. */
233 uint8_t txonly_multi_flow;
234 /**< Whether multiple flows are generated in TXONLY mode. */
236 uint32_t tx_pkt_times_inter;
237 /**< Timings for send scheduling in TXONLY mode, time between bursts. */
239 uint32_t tx_pkt_times_intra;
240 /**< Timings for send scheduling in TXONLY mode, time between packets. */
242 uint16_t nb_pkt_per_burst = DEF_PKT_BURST; /**< Number of packets per burst. */
243 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
245 /* current configuration is in DCB or not,0 means it is not in DCB mode */
246 uint8_t dcb_config = 0;
248 /* Whether the dcb is in testing status */
249 uint8_t dcb_test = 0;
252 * Configurable number of RX/TX queues.
254 queueid_t nb_hairpinq; /**< Number of hairpin queues per port. */
255 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
256 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
259 * Configurable number of RX/TX ring descriptors.
260 * Defaults are supplied by drivers via ethdev.
262 #define RTE_TEST_RX_DESC_DEFAULT 0
263 #define RTE_TEST_TX_DESC_DEFAULT 0
264 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
265 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
267 #define RTE_PMD_PARAM_UNSET -1
269 * Configurable values of RX and TX ring threshold registers.
272 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
273 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
274 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
276 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
277 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
278 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
281 * Configurable value of RX free threshold.
283 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
286 * Configurable value of RX drop enable.
288 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
291 * Configurable value of TX free threshold.
293 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
296 * Configurable value of TX RS bit threshold.
298 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
301 * Configurable value of buffered packets before sending.
303 uint16_t noisy_tx_sw_bufsz;
306 * Configurable value of packet buffer timeout.
308 uint16_t noisy_tx_sw_buf_flush_time;
311 * Configurable value for size of VNF internal memory area
312 * used for simulating noisy neighbour behaviour
314 uint64_t noisy_lkup_mem_sz;
317 * Configurable value of number of random writes done in
318 * VNF simulation memory area.
320 uint64_t noisy_lkup_num_writes;
323 * Configurable value of number of random reads done in
324 * VNF simulation memory area.
326 uint64_t noisy_lkup_num_reads;
329 * Configurable value of number of random reads/writes done in
330 * VNF simulation memory area.
332 uint64_t noisy_lkup_num_reads_writes;
335 * Receive Side Scaling (RSS) configuration.
337 uint64_t rss_hf = ETH_RSS_IP; /* RSS IP by default. */
340 * Port topology configuration
342 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
345 * Avoids to flush all the RX streams before starts forwarding.
347 uint8_t no_flush_rx = 0; /* flush by default */
350 * Flow API isolated mode.
352 uint8_t flow_isolate_all;
355 * Avoids to check link status when starting/stopping a port.
357 uint8_t no_link_check = 0; /* check by default */
360 * Don't automatically start all ports in interactive mode.
362 uint8_t no_device_start = 0;
365 * Enable link status change notification
367 uint8_t lsc_interrupt = 1; /* enabled by default */
370 * Enable device removal notification.
372 uint8_t rmv_interrupt = 1; /* enabled by default */
374 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
376 /* After attach, port setup is called on event or by iterator */
377 bool setup_on_probe_event = true;
379 /* Clear ptypes on port initialization. */
380 uint8_t clear_ptypes = true;
382 /* Hairpin ports configuration mode. */
383 uint16_t hairpin_mode;
385 /* Pretty printing of ethdev events */
386 static const char * const eth_event_desc[] = {
387 [RTE_ETH_EVENT_UNKNOWN] = "unknown",
388 [RTE_ETH_EVENT_INTR_LSC] = "link state change",
389 [RTE_ETH_EVENT_QUEUE_STATE] = "queue state",
390 [RTE_ETH_EVENT_INTR_RESET] = "reset",
391 [RTE_ETH_EVENT_VF_MBOX] = "VF mbox",
392 [RTE_ETH_EVENT_IPSEC] = "IPsec",
393 [RTE_ETH_EVENT_MACSEC] = "MACsec",
394 [RTE_ETH_EVENT_INTR_RMV] = "device removal",
395 [RTE_ETH_EVENT_NEW] = "device probed",
396 [RTE_ETH_EVENT_DESTROY] = "device released",
397 [RTE_ETH_EVENT_FLOW_AGED] = "flow aged",
398 [RTE_ETH_EVENT_MAX] = NULL,
402 * Display or mask ether events
403 * Default to all events except VF_MBOX
405 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
406 (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
407 (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
408 (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
409 (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
410 (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
411 (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV) |
412 (UINT32_C(1) << RTE_ETH_EVENT_FLOW_AGED);
414 * Decide if all memory are locked for performance.
419 * NIC bypass mode configuration options.
422 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
423 /* The NIC bypass watchdog timeout. */
424 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
428 #ifdef RTE_LIBRTE_LATENCY_STATS
431 * Set when latency stats is enabled in the commandline
433 uint8_t latencystats_enabled;
436 * Lcore ID to serive latency statistics.
438 lcoreid_t latencystats_lcore_id = -1;
443 * Ethernet device configuration.
445 struct rte_eth_rxmode rx_mode = {
446 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
447 /**< Default maximum frame length. */
450 struct rte_eth_txmode tx_mode = {
451 .offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
454 struct rte_fdir_conf fdir_conf = {
455 .mode = RTE_FDIR_MODE_NONE,
456 .pballoc = RTE_FDIR_PBALLOC_64K,
457 .status = RTE_FDIR_REPORT_STATUS,
459 .vlan_tci_mask = 0xFFEF,
461 .src_ip = 0xFFFFFFFF,
462 .dst_ip = 0xFFFFFFFF,
465 .src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
466 .dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
468 .src_port_mask = 0xFFFF,
469 .dst_port_mask = 0xFFFF,
470 .mac_addr_byte_mask = 0xFF,
471 .tunnel_type_mask = 1,
472 .tunnel_id_mask = 0xFFFFFFFF,
477 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
479 struct queue_stats_mappings tx_queue_stats_mappings_array[MAX_TX_QUEUE_STATS_MAPPINGS];
480 struct queue_stats_mappings rx_queue_stats_mappings_array[MAX_RX_QUEUE_STATS_MAPPINGS];
482 struct queue_stats_mappings *tx_queue_stats_mappings = tx_queue_stats_mappings_array;
483 struct queue_stats_mappings *rx_queue_stats_mappings = rx_queue_stats_mappings_array;
485 uint16_t nb_tx_queue_stats_mappings = 0;
486 uint16_t nb_rx_queue_stats_mappings = 0;
489 * Display zero values by default for xstats
491 uint8_t xstats_hide_zero;
494 * Measure of CPU cycles disabled by default
496 uint8_t record_core_cycles;
499 * Display of RX and TX bursts disabled by default
501 uint8_t record_burst_stats;
503 unsigned int num_sockets = 0;
504 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
506 #ifdef RTE_LIBRTE_BITRATESTATS
507 /* Bitrate statistics */
508 struct rte_stats_bitrates *bitrate_data;
509 lcoreid_t bitrate_lcore_id;
510 uint8_t bitrate_enabled;
513 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
514 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
517 * hexadecimal bitmask of RX mq mode can be enabled.
519 enum rte_eth_rx_mq_mode rx_mq_mode = ETH_MQ_RX_VMDQ_DCB_RSS;
521 /* Forward function declarations */
522 static void setup_attached_port(portid_t pi);
523 static void map_port_queue_stats_mapping_registers(portid_t pi,
524 struct rte_port *port);
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_master_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 /* extremely pessimistic estimation of memory required to create a mempool */
639 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
641 unsigned int n_pages, mbuf_per_pg, leftover;
642 uint64_t total_mem, mbuf_mem, obj_sz;
644 /* there is no good way to predict how much space the mempool will
645 * occupy because it will allocate chunks on the fly, and some of those
646 * will come from default DPDK memory while some will come from our
647 * external memory, so just assume 128MB will be enough for everyone.
649 uint64_t hdr_mem = 128 << 20;
651 /* account for possible non-contiguousness */
652 obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
654 TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
658 mbuf_per_pg = pgsz / obj_sz;
659 leftover = (nb_mbufs % mbuf_per_pg) > 0;
660 n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
662 mbuf_mem = n_pages * pgsz;
664 total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
666 if (total_mem > SIZE_MAX) {
667 TESTPMD_LOG(ERR, "Memory size too big\n");
670 *out = (size_t)total_mem;
676 pagesz_flags(uint64_t page_sz)
678 /* as per mmap() manpage, all page sizes are log2 of page size
679 * shifted by MAP_HUGE_SHIFT
681 int log2 = rte_log2_u64(page_sz);
683 return (log2 << HUGE_SHIFT);
687 alloc_mem(size_t memsz, size_t pgsz, bool huge)
692 /* allocate anonymous hugepages */
693 flags = MAP_ANONYMOUS | MAP_PRIVATE;
695 flags |= HUGE_FLAG | pagesz_flags(pgsz);
697 addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
698 if (addr == MAP_FAILED)
704 struct extmem_param {
708 rte_iova_t *iova_table;
709 unsigned int iova_table_len;
713 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
716 uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
717 RTE_PGSIZE_16M, RTE_PGSIZE_16G}; /* POWER */
718 unsigned int cur_page, n_pages, pgsz_idx;
719 size_t mem_sz, cur_pgsz;
720 rte_iova_t *iovas = NULL;
724 for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
725 /* skip anything that is too big */
726 if (pgsizes[pgsz_idx] > SIZE_MAX)
729 cur_pgsz = pgsizes[pgsz_idx];
731 /* if we were told not to allocate hugepages, override */
733 cur_pgsz = sysconf(_SC_PAGESIZE);
735 ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
737 TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
741 /* allocate our memory */
742 addr = alloc_mem(mem_sz, cur_pgsz, huge);
744 /* if we couldn't allocate memory with a specified page size,
745 * that doesn't mean we can't do it with other page sizes, so
751 /* store IOVA addresses for every page in this memory area */
752 n_pages = mem_sz / cur_pgsz;
754 iovas = malloc(sizeof(*iovas) * n_pages);
757 TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
760 /* lock memory if it's not huge pages */
764 /* populate IOVA addresses */
765 for (cur_page = 0; cur_page < n_pages; cur_page++) {
770 offset = cur_pgsz * cur_page;
771 cur = RTE_PTR_ADD(addr, offset);
773 /* touch the page before getting its IOVA */
774 *(volatile char *)cur = 0;
776 iova = rte_mem_virt2iova(cur);
778 iovas[cur_page] = iova;
783 /* if we couldn't allocate anything */
789 param->pgsz = cur_pgsz;
790 param->iova_table = iovas;
791 param->iova_table_len = n_pages;
798 munmap(addr, mem_sz);
804 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
806 struct extmem_param param;
809 memset(¶m, 0, sizeof(param));
811 /* check if our heap exists */
812 socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
814 /* create our heap */
815 ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
817 TESTPMD_LOG(ERR, "Cannot create heap\n");
822 ret = create_extmem(nb_mbufs, mbuf_sz, ¶m, huge);
824 TESTPMD_LOG(ERR, "Cannot create memory area\n");
828 /* we now have a valid memory area, so add it to heap */
829 ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
830 param.addr, param.len, param.iova_table,
831 param.iova_table_len, param.pgsz);
833 /* when using VFIO, memory is automatically mapped for DMA by EAL */
835 /* not needed any more */
836 free(param.iova_table);
839 TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
840 munmap(param.addr, param.len);
846 TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
852 dma_unmap_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
853 struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
858 RTE_ETH_FOREACH_DEV(pid) {
859 struct rte_eth_dev *dev =
860 &rte_eth_devices[pid];
862 ret = rte_dev_dma_unmap(dev->device, memhdr->addr, 0,
866 "unable to DMA unmap addr 0x%p "
868 memhdr->addr, dev->data->name);
871 ret = rte_extmem_unregister(memhdr->addr, memhdr->len);
874 "unable to un-register addr 0x%p\n", memhdr->addr);
879 dma_map_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
880 struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
883 size_t page_size = sysconf(_SC_PAGESIZE);
886 ret = rte_extmem_register(memhdr->addr, memhdr->len, NULL, 0,
890 "unable to register addr 0x%p\n", memhdr->addr);
893 RTE_ETH_FOREACH_DEV(pid) {
894 struct rte_eth_dev *dev =
895 &rte_eth_devices[pid];
897 ret = rte_dev_dma_map(dev->device, memhdr->addr, 0,
901 "unable to DMA map addr 0x%p "
903 memhdr->addr, dev->data->name);
909 setup_extbuf(uint32_t nb_mbufs, uint16_t mbuf_sz, unsigned int socket_id,
910 char *pool_name, struct rte_pktmbuf_extmem **ext_mem)
912 struct rte_pktmbuf_extmem *xmem;
913 unsigned int ext_num, zone_num, elt_num;
916 elt_size = RTE_ALIGN_CEIL(mbuf_sz, RTE_CACHE_LINE_SIZE);
917 elt_num = EXTBUF_ZONE_SIZE / elt_size;
918 zone_num = (nb_mbufs + elt_num - 1) / elt_num;
920 xmem = malloc(sizeof(struct rte_pktmbuf_extmem) * zone_num);
922 TESTPMD_LOG(ERR, "Cannot allocate memory for "
923 "external buffer descriptors\n");
927 for (ext_num = 0; ext_num < zone_num; ext_num++) {
928 struct rte_pktmbuf_extmem *xseg = xmem + ext_num;
929 const struct rte_memzone *mz;
930 char mz_name[RTE_MEMZONE_NAMESIZE];
933 ret = snprintf(mz_name, sizeof(mz_name),
934 RTE_MEMPOOL_MZ_FORMAT "_xb_%u", pool_name, ext_num);
935 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
936 errno = ENAMETOOLONG;
940 mz = rte_memzone_reserve_aligned(mz_name, EXTBUF_ZONE_SIZE,
942 RTE_MEMZONE_IOVA_CONTIG |
944 RTE_MEMZONE_SIZE_HINT_ONLY,
948 * The caller exits on external buffer creation
949 * error, so there is no need to free memzones.
955 xseg->buf_ptr = mz->addr;
956 xseg->buf_iova = mz->iova;
957 xseg->buf_len = EXTBUF_ZONE_SIZE;
958 xseg->elt_size = elt_size;
960 if (ext_num == 0 && xmem != NULL) {
969 * Configuration initialisation done once at init time.
971 static struct rte_mempool *
972 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
973 unsigned int socket_id, uint16_t size_idx)
975 char pool_name[RTE_MEMPOOL_NAMESIZE];
976 struct rte_mempool *rte_mp = NULL;
979 mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
980 mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name), size_idx);
983 "create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
984 pool_name, nb_mbuf, mbuf_seg_size, socket_id);
986 switch (mp_alloc_type) {
987 case MP_ALLOC_NATIVE:
989 /* wrapper to rte_mempool_create() */
990 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
991 rte_mbuf_best_mempool_ops());
992 rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
993 mb_mempool_cache, 0, mbuf_seg_size, socket_id);
998 rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
999 mb_size, (unsigned int) mb_mempool_cache,
1000 sizeof(struct rte_pktmbuf_pool_private),
1001 socket_id, mempool_flags);
1005 if (rte_mempool_populate_anon(rte_mp) == 0) {
1006 rte_mempool_free(rte_mp);
1010 rte_pktmbuf_pool_init(rte_mp, NULL);
1011 rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
1012 rte_mempool_mem_iter(rte_mp, dma_map_cb, NULL);
1016 case MP_ALLOC_XMEM_HUGE:
1019 bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
1021 if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
1022 rte_exit(EXIT_FAILURE, "Could not create external memory\n");
1025 rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
1026 if (heap_socket < 0)
1027 rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
1029 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1030 rte_mbuf_best_mempool_ops());
1031 rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1032 mb_mempool_cache, 0, mbuf_seg_size,
1038 struct rte_pktmbuf_extmem *ext_mem;
1039 unsigned int ext_num;
1041 ext_num = setup_extbuf(nb_mbuf, mbuf_seg_size,
1042 socket_id, pool_name, &ext_mem);
1044 rte_exit(EXIT_FAILURE,
1045 "Can't create pinned data buffers\n");
1047 TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1048 rte_mbuf_best_mempool_ops());
1049 rte_mp = rte_pktmbuf_pool_create_extbuf
1050 (pool_name, nb_mbuf, mb_mempool_cache,
1051 0, mbuf_seg_size, socket_id,
1058 rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
1063 if (rte_mp == NULL) {
1064 rte_exit(EXIT_FAILURE,
1065 "Creation of mbuf pool for socket %u failed: %s\n",
1066 socket_id, rte_strerror(rte_errno));
1067 } else if (verbose_level > 0) {
1068 rte_mempool_dump(stdout, rte_mp);
1074 * Check given socket id is valid or not with NUMA mode,
1075 * if valid, return 0, else return -1
1078 check_socket_id(const unsigned int socket_id)
1080 static int warning_once = 0;
1082 if (new_socket_id(socket_id)) {
1083 if (!warning_once && numa_support)
1084 printf("Warning: NUMA should be configured manually by"
1085 " using --port-numa-config and"
1086 " --ring-numa-config parameters along with"
1095 * Get the allowed maximum number of RX queues.
1096 * *pid return the port id which has minimal value of
1097 * max_rx_queues in all ports.
1100 get_allowed_max_nb_rxq(portid_t *pid)
1102 queueid_t allowed_max_rxq = RTE_MAX_QUEUES_PER_PORT;
1103 bool max_rxq_valid = false;
1105 struct rte_eth_dev_info dev_info;
1107 RTE_ETH_FOREACH_DEV(pi) {
1108 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1111 max_rxq_valid = true;
1112 if (dev_info.max_rx_queues < allowed_max_rxq) {
1113 allowed_max_rxq = dev_info.max_rx_queues;
1117 return max_rxq_valid ? allowed_max_rxq : 0;
1121 * Check input rxq is valid or not.
1122 * If input rxq is not greater than any of maximum number
1123 * of RX queues of all ports, it is valid.
1124 * if valid, return 0, else return -1
1127 check_nb_rxq(queueid_t rxq)
1129 queueid_t allowed_max_rxq;
1132 allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
1133 if (rxq > allowed_max_rxq) {
1134 printf("Fail: input rxq (%u) can't be greater "
1135 "than max_rx_queues (%u) of port %u\n",
1145 * Get the allowed maximum number of TX queues.
1146 * *pid return the port id which has minimal value of
1147 * max_tx_queues in all ports.
1150 get_allowed_max_nb_txq(portid_t *pid)
1152 queueid_t allowed_max_txq = RTE_MAX_QUEUES_PER_PORT;
1153 bool max_txq_valid = false;
1155 struct rte_eth_dev_info dev_info;
1157 RTE_ETH_FOREACH_DEV(pi) {
1158 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1161 max_txq_valid = true;
1162 if (dev_info.max_tx_queues < allowed_max_txq) {
1163 allowed_max_txq = dev_info.max_tx_queues;
1167 return max_txq_valid ? allowed_max_txq : 0;
1171 * Check input txq is valid or not.
1172 * If input txq is not greater than any of maximum number
1173 * of TX queues of all ports, it is valid.
1174 * if valid, return 0, else return -1
1177 check_nb_txq(queueid_t txq)
1179 queueid_t allowed_max_txq;
1182 allowed_max_txq = get_allowed_max_nb_txq(&pid);
1183 if (txq > allowed_max_txq) {
1184 printf("Fail: input txq (%u) can't be greater "
1185 "than max_tx_queues (%u) of port %u\n",
1195 * Get the allowed maximum number of RXDs of every rx queue.
1196 * *pid return the port id which has minimal value of
1197 * max_rxd in all queues of all ports.
1200 get_allowed_max_nb_rxd(portid_t *pid)
1202 uint16_t allowed_max_rxd = UINT16_MAX;
1204 struct rte_eth_dev_info dev_info;
1206 RTE_ETH_FOREACH_DEV(pi) {
1207 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1210 if (dev_info.rx_desc_lim.nb_max < allowed_max_rxd) {
1211 allowed_max_rxd = dev_info.rx_desc_lim.nb_max;
1215 return allowed_max_rxd;
1219 * Get the allowed minimal number of RXDs of every rx queue.
1220 * *pid return the port id which has minimal value of
1221 * min_rxd in all queues of all ports.
1224 get_allowed_min_nb_rxd(portid_t *pid)
1226 uint16_t allowed_min_rxd = 0;
1228 struct rte_eth_dev_info dev_info;
1230 RTE_ETH_FOREACH_DEV(pi) {
1231 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1234 if (dev_info.rx_desc_lim.nb_min > allowed_min_rxd) {
1235 allowed_min_rxd = dev_info.rx_desc_lim.nb_min;
1240 return allowed_min_rxd;
1244 * Check input rxd is valid or not.
1245 * If input rxd is not greater than any of maximum number
1246 * of RXDs of every Rx queues and is not less than any of
1247 * minimal number of RXDs of every Rx queues, it is valid.
1248 * if valid, return 0, else return -1
1251 check_nb_rxd(queueid_t rxd)
1253 uint16_t allowed_max_rxd;
1254 uint16_t allowed_min_rxd;
1257 allowed_max_rxd = get_allowed_max_nb_rxd(&pid);
1258 if (rxd > allowed_max_rxd) {
1259 printf("Fail: input rxd (%u) can't be greater "
1260 "than max_rxds (%u) of port %u\n",
1267 allowed_min_rxd = get_allowed_min_nb_rxd(&pid);
1268 if (rxd < allowed_min_rxd) {
1269 printf("Fail: input rxd (%u) can't be less "
1270 "than min_rxds (%u) of port %u\n",
1281 * Get the allowed maximum number of TXDs of every rx queues.
1282 * *pid return the port id which has minimal value of
1283 * max_txd in every tx queue.
1286 get_allowed_max_nb_txd(portid_t *pid)
1288 uint16_t allowed_max_txd = UINT16_MAX;
1290 struct rte_eth_dev_info dev_info;
1292 RTE_ETH_FOREACH_DEV(pi) {
1293 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1296 if (dev_info.tx_desc_lim.nb_max < allowed_max_txd) {
1297 allowed_max_txd = dev_info.tx_desc_lim.nb_max;
1301 return allowed_max_txd;
1305 * Get the allowed maximum number of TXDs of every tx queues.
1306 * *pid return the port id which has minimal value of
1307 * min_txd in every tx queue.
1310 get_allowed_min_nb_txd(portid_t *pid)
1312 uint16_t allowed_min_txd = 0;
1314 struct rte_eth_dev_info dev_info;
1316 RTE_ETH_FOREACH_DEV(pi) {
1317 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1320 if (dev_info.tx_desc_lim.nb_min > allowed_min_txd) {
1321 allowed_min_txd = dev_info.tx_desc_lim.nb_min;
1326 return allowed_min_txd;
1330 * Check input txd is valid or not.
1331 * If input txd is not greater than any of maximum number
1332 * of TXDs of every Rx queues, it is valid.
1333 * if valid, return 0, else return -1
1336 check_nb_txd(queueid_t txd)
1338 uint16_t allowed_max_txd;
1339 uint16_t allowed_min_txd;
1342 allowed_max_txd = get_allowed_max_nb_txd(&pid);
1343 if (txd > allowed_max_txd) {
1344 printf("Fail: input txd (%u) can't be greater "
1345 "than max_txds (%u) of port %u\n",
1352 allowed_min_txd = get_allowed_min_nb_txd(&pid);
1353 if (txd < allowed_min_txd) {
1354 printf("Fail: input txd (%u) can't be less "
1355 "than min_txds (%u) of port %u\n",
1366 * Get the allowed maximum number of hairpin queues.
1367 * *pid return the port id which has minimal value of
1368 * max_hairpin_queues in all ports.
1371 get_allowed_max_nb_hairpinq(portid_t *pid)
1373 queueid_t allowed_max_hairpinq = RTE_MAX_QUEUES_PER_PORT;
1375 struct rte_eth_hairpin_cap cap;
1377 RTE_ETH_FOREACH_DEV(pi) {
1378 if (rte_eth_dev_hairpin_capability_get(pi, &cap) != 0) {
1382 if (cap.max_nb_queues < allowed_max_hairpinq) {
1383 allowed_max_hairpinq = cap.max_nb_queues;
1387 return allowed_max_hairpinq;
1391 * Check input hairpin is valid or not.
1392 * If input hairpin is not greater than any of maximum number
1393 * of hairpin queues of all ports, it is valid.
1394 * if valid, return 0, else return -1
1397 check_nb_hairpinq(queueid_t hairpinq)
1399 queueid_t allowed_max_hairpinq;
1402 allowed_max_hairpinq = get_allowed_max_nb_hairpinq(&pid);
1403 if (hairpinq > allowed_max_hairpinq) {
1404 printf("Fail: input hairpin (%u) can't be greater "
1405 "than max_hairpin_queues (%u) of port %u\n",
1406 hairpinq, allowed_max_hairpinq, pid);
1416 struct rte_port *port;
1417 struct rte_mempool *mbp;
1418 unsigned int nb_mbuf_per_pool;
1420 uint8_t port_per_socket[RTE_MAX_NUMA_NODES];
1421 struct rte_gro_param gro_param;
1428 memset(port_per_socket,0,RTE_MAX_NUMA_NODES);
1430 /* Configuration of logical cores. */
1431 fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1432 sizeof(struct fwd_lcore *) * nb_lcores,
1433 RTE_CACHE_LINE_SIZE);
1434 if (fwd_lcores == NULL) {
1435 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1436 "failed\n", nb_lcores);
1438 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1439 fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1440 sizeof(struct fwd_lcore),
1441 RTE_CACHE_LINE_SIZE);
1442 if (fwd_lcores[lc_id] == NULL) {
1443 rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1446 fwd_lcores[lc_id]->cpuid_idx = lc_id;
1449 RTE_ETH_FOREACH_DEV(pid) {
1451 /* Apply default TxRx configuration for all ports */
1452 port->dev_conf.txmode = tx_mode;
1453 port->dev_conf.rxmode = rx_mode;
1455 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
1457 rte_exit(EXIT_FAILURE,
1458 "rte_eth_dev_info_get() failed\n");
1460 if (!(port->dev_info.tx_offload_capa &
1461 DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1462 port->dev_conf.txmode.offloads &=
1463 ~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1465 if (port_numa[pid] != NUMA_NO_CONFIG)
1466 port_per_socket[port_numa[pid]]++;
1468 uint32_t socket_id = rte_eth_dev_socket_id(pid);
1471 * if socket_id is invalid,
1472 * set to the first available socket.
1474 if (check_socket_id(socket_id) < 0)
1475 socket_id = socket_ids[0];
1476 port_per_socket[socket_id]++;
1480 /* Apply Rx offloads configuration */
1481 for (k = 0; k < port->dev_info.max_rx_queues; k++)
1482 port->rx_conf[k].offloads =
1483 port->dev_conf.rxmode.offloads;
1484 /* Apply Tx offloads configuration */
1485 for (k = 0; k < port->dev_info.max_tx_queues; k++)
1486 port->tx_conf[k].offloads =
1487 port->dev_conf.txmode.offloads;
1489 /* set flag to initialize port/queue */
1490 port->need_reconfig = 1;
1491 port->need_reconfig_queues = 1;
1492 port->tx_metadata = 0;
1494 /* Check for maximum number of segments per MTU. Accordingly
1495 * update the mbuf data size.
1497 if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX &&
1498 port->dev_info.rx_desc_lim.nb_mtu_seg_max != 0) {
1499 data_size = rx_mode.max_rx_pkt_len /
1500 port->dev_info.rx_desc_lim.nb_mtu_seg_max;
1502 if ((data_size + RTE_PKTMBUF_HEADROOM) >
1503 mbuf_data_size[0]) {
1504 mbuf_data_size[0] = data_size +
1505 RTE_PKTMBUF_HEADROOM;
1512 TESTPMD_LOG(WARNING,
1513 "Configured mbuf size of the first segment %hu\n",
1516 * Create pools of mbuf.
1517 * If NUMA support is disabled, create a single pool of mbuf in
1518 * socket 0 memory by default.
1519 * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1521 * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1522 * nb_txd can be configured at run time.
1524 if (param_total_num_mbufs)
1525 nb_mbuf_per_pool = param_total_num_mbufs;
1527 nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1528 (nb_lcores * mb_mempool_cache) +
1529 RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1530 nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1536 for (i = 0; i < num_sockets; i++)
1537 for (j = 0; j < mbuf_data_size_n; j++)
1538 mempools[i * MAX_SEGS_BUFFER_SPLIT + j] =
1539 mbuf_pool_create(mbuf_data_size[j],
1545 for (i = 0; i < mbuf_data_size_n; i++)
1546 mempools[i] = mbuf_pool_create
1549 socket_num == UMA_NO_CONFIG ?
1555 gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1556 DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1558 * Records which Mbuf pool to use by each logical core, if needed.
1560 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1561 mbp = mbuf_pool_find(
1562 rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]), 0);
1565 mbp = mbuf_pool_find(0, 0);
1566 fwd_lcores[lc_id]->mbp = mbp;
1567 /* initialize GSO context */
1568 fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1569 fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1570 fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1571 fwd_lcores[lc_id]->gso_ctx.gso_size = RTE_ETHER_MAX_LEN -
1573 fwd_lcores[lc_id]->gso_ctx.flag = 0;
1576 /* Configuration of packet forwarding streams. */
1577 if (init_fwd_streams() < 0)
1578 rte_exit(EXIT_FAILURE, "FAIL from init_fwd_streams()\n");
1582 /* create a gro context for each lcore */
1583 gro_param.gro_types = RTE_GRO_TCP_IPV4;
1584 gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1585 gro_param.max_item_per_flow = MAX_PKT_BURST;
1586 for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1587 gro_param.socket_id = rte_lcore_to_socket_id(
1588 fwd_lcores_cpuids[lc_id]);
1589 fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1590 if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1591 rte_exit(EXIT_FAILURE,
1592 "rte_gro_ctx_create() failed\n");
1599 reconfig(portid_t new_port_id, unsigned socket_id)
1601 struct rte_port *port;
1604 /* Reconfiguration of Ethernet ports. */
1605 port = &ports[new_port_id];
1607 ret = eth_dev_info_get_print_err(new_port_id, &port->dev_info);
1611 /* set flag to initialize port/queue */
1612 port->need_reconfig = 1;
1613 port->need_reconfig_queues = 1;
1614 port->socket_id = socket_id;
1621 init_fwd_streams(void)
1624 struct rte_port *port;
1625 streamid_t sm_id, nb_fwd_streams_new;
1628 /* set socket id according to numa or not */
1629 RTE_ETH_FOREACH_DEV(pid) {
1631 if (nb_rxq > port->dev_info.max_rx_queues) {
1632 printf("Fail: nb_rxq(%d) is greater than "
1633 "max_rx_queues(%d)\n", nb_rxq,
1634 port->dev_info.max_rx_queues);
1637 if (nb_txq > port->dev_info.max_tx_queues) {
1638 printf("Fail: nb_txq(%d) is greater than "
1639 "max_tx_queues(%d)\n", nb_txq,
1640 port->dev_info.max_tx_queues);
1644 if (port_numa[pid] != NUMA_NO_CONFIG)
1645 port->socket_id = port_numa[pid];
1647 port->socket_id = rte_eth_dev_socket_id(pid);
1650 * if socket_id is invalid,
1651 * set to the first available socket.
1653 if (check_socket_id(port->socket_id) < 0)
1654 port->socket_id = socket_ids[0];
1658 if (socket_num == UMA_NO_CONFIG)
1659 port->socket_id = 0;
1661 port->socket_id = socket_num;
1665 q = RTE_MAX(nb_rxq, nb_txq);
1667 printf("Fail: Cannot allocate fwd streams as number of queues is 0\n");
1670 nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1671 if (nb_fwd_streams_new == nb_fwd_streams)
1674 if (fwd_streams != NULL) {
1675 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1676 if (fwd_streams[sm_id] == NULL)
1678 rte_free(fwd_streams[sm_id]);
1679 fwd_streams[sm_id] = NULL;
1681 rte_free(fwd_streams);
1686 nb_fwd_streams = nb_fwd_streams_new;
1687 if (nb_fwd_streams) {
1688 fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1689 sizeof(struct fwd_stream *) * nb_fwd_streams,
1690 RTE_CACHE_LINE_SIZE);
1691 if (fwd_streams == NULL)
1692 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1693 " (struct fwd_stream *)) failed\n",
1696 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1697 fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1698 " struct fwd_stream", sizeof(struct fwd_stream),
1699 RTE_CACHE_LINE_SIZE);
1700 if (fwd_streams[sm_id] == NULL)
1701 rte_exit(EXIT_FAILURE, "rte_zmalloc"
1702 "(struct fwd_stream) failed\n");
1710 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1712 uint64_t total_burst, sburst;
1714 uint64_t burst_stats[4];
1715 uint16_t pktnb_stats[4];
1717 int burst_percent[4], sburstp;
1721 * First compute the total number of packet bursts and the
1722 * two highest numbers of bursts of the same number of packets.
1724 memset(&burst_stats, 0x0, sizeof(burst_stats));
1725 memset(&pktnb_stats, 0x0, sizeof(pktnb_stats));
1727 /* Show stats for 0 burst size always */
1728 total_burst = pbs->pkt_burst_spread[0];
1729 burst_stats[0] = pbs->pkt_burst_spread[0];
1732 /* Find the next 2 burst sizes with highest occurrences. */
1733 for (nb_pkt = 1; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1734 nb_burst = pbs->pkt_burst_spread[nb_pkt];
1739 total_burst += nb_burst;
1741 if (nb_burst > burst_stats[1]) {
1742 burst_stats[2] = burst_stats[1];
1743 pktnb_stats[2] = pktnb_stats[1];
1744 burst_stats[1] = nb_burst;
1745 pktnb_stats[1] = nb_pkt;
1746 } else if (nb_burst > burst_stats[2]) {
1747 burst_stats[2] = nb_burst;
1748 pktnb_stats[2] = nb_pkt;
1751 if (total_burst == 0)
1754 printf(" %s-bursts : %"PRIu64" [", rx_tx, total_burst);
1755 for (i = 0, sburst = 0, sburstp = 0; i < 4; i++) {
1757 printf("%d%% of other]\n", 100 - sburstp);
1761 sburst += burst_stats[i];
1762 if (sburst == total_burst) {
1763 printf("%d%% of %d pkts]\n",
1764 100 - sburstp, (int) pktnb_stats[i]);
1769 (double)burst_stats[i] / total_burst * 100;
1770 printf("%d%% of %d pkts + ",
1771 burst_percent[i], (int) pktnb_stats[i]);
1772 sburstp += burst_percent[i];
1777 fwd_stream_stats_display(streamid_t stream_id)
1779 struct fwd_stream *fs;
1780 static const char *fwd_top_stats_border = "-------";
1782 fs = fwd_streams[stream_id];
1783 if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1784 (fs->fwd_dropped == 0))
1786 printf("\n %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1787 "TX Port=%2d/Queue=%2d %s\n",
1788 fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1789 fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1790 printf(" RX-packets: %-14"PRIu64" TX-packets: %-14"PRIu64
1791 " TX-dropped: %-14"PRIu64,
1792 fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1794 /* if checksum mode */
1795 if (cur_fwd_eng == &csum_fwd_engine) {
1796 printf(" RX- bad IP checksum: %-14"PRIu64
1797 " Rx- bad L4 checksum: %-14"PRIu64
1798 " Rx- bad outer L4 checksum: %-14"PRIu64"\n",
1799 fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1800 fs->rx_bad_outer_l4_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 } ports_stats[RTE_MAX_ETHPORTS];
1824 uint64_t total_rx_dropped = 0;
1825 uint64_t total_tx_dropped = 0;
1826 uint64_t total_rx_nombuf = 0;
1827 struct rte_eth_stats stats;
1828 uint64_t fwd_cycles = 0;
1829 uint64_t total_recv = 0;
1830 uint64_t total_xmit = 0;
1831 struct rte_port *port;
1836 memset(ports_stats, 0, sizeof(ports_stats));
1838 for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1839 struct fwd_stream *fs = fwd_streams[sm_id];
1841 if (cur_fwd_config.nb_fwd_streams >
1842 cur_fwd_config.nb_fwd_ports) {
1843 fwd_stream_stats_display(sm_id);
1845 ports_stats[fs->tx_port].tx_stream = fs;
1846 ports_stats[fs->rx_port].rx_stream = fs;
1849 ports_stats[fs->tx_port].tx_dropped += fs->fwd_dropped;
1851 ports_stats[fs->rx_port].rx_bad_ip_csum += fs->rx_bad_ip_csum;
1852 ports_stats[fs->rx_port].rx_bad_l4_csum += fs->rx_bad_l4_csum;
1853 ports_stats[fs->rx_port].rx_bad_outer_l4_csum +=
1854 fs->rx_bad_outer_l4_csum;
1856 if (record_core_cycles)
1857 fwd_cycles += fs->core_cycles;
1859 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1862 pt_id = fwd_ports_ids[i];
1863 port = &ports[pt_id];
1865 rte_eth_stats_get(pt_id, &stats);
1866 stats.ipackets -= port->stats.ipackets;
1867 stats.opackets -= port->stats.opackets;
1868 stats.ibytes -= port->stats.ibytes;
1869 stats.obytes -= port->stats.obytes;
1870 stats.imissed -= port->stats.imissed;
1871 stats.oerrors -= port->stats.oerrors;
1872 stats.rx_nombuf -= port->stats.rx_nombuf;
1874 total_recv += stats.ipackets;
1875 total_xmit += stats.opackets;
1876 total_rx_dropped += stats.imissed;
1877 total_tx_dropped += ports_stats[pt_id].tx_dropped;
1878 total_tx_dropped += stats.oerrors;
1879 total_rx_nombuf += stats.rx_nombuf;
1881 printf("\n %s Forward statistics for port %-2d %s\n",
1882 fwd_stats_border, pt_id, fwd_stats_border);
1884 if (!port->rx_queue_stats_mapping_enabled &&
1885 !port->tx_queue_stats_mapping_enabled) {
1886 printf(" RX-packets: %-14"PRIu64
1887 " RX-dropped: %-14"PRIu64
1888 "RX-total: %-"PRIu64"\n",
1889 stats.ipackets, stats.imissed,
1890 stats.ipackets + stats.imissed);
1892 if (cur_fwd_eng == &csum_fwd_engine)
1893 printf(" Bad-ipcsum: %-14"PRIu64
1894 " Bad-l4csum: %-14"PRIu64
1895 "Bad-outer-l4csum: %-14"PRIu64"\n",
1896 ports_stats[pt_id].rx_bad_ip_csum,
1897 ports_stats[pt_id].rx_bad_l4_csum,
1898 ports_stats[pt_id].rx_bad_outer_l4_csum);
1899 if (stats.ierrors + stats.rx_nombuf > 0) {
1900 printf(" RX-error: %-"PRIu64"\n",
1902 printf(" RX-nombufs: %-14"PRIu64"\n",
1906 printf(" TX-packets: %-14"PRIu64
1907 " TX-dropped: %-14"PRIu64
1908 "TX-total: %-"PRIu64"\n",
1909 stats.opackets, ports_stats[pt_id].tx_dropped,
1910 stats.opackets + ports_stats[pt_id].tx_dropped);
1912 printf(" RX-packets: %14"PRIu64
1913 " RX-dropped:%14"PRIu64
1914 " RX-total:%14"PRIu64"\n",
1915 stats.ipackets, stats.imissed,
1916 stats.ipackets + stats.imissed);
1918 if (cur_fwd_eng == &csum_fwd_engine)
1919 printf(" Bad-ipcsum:%14"PRIu64
1920 " Bad-l4csum:%14"PRIu64
1921 " Bad-outer-l4csum: %-14"PRIu64"\n",
1922 ports_stats[pt_id].rx_bad_ip_csum,
1923 ports_stats[pt_id].rx_bad_l4_csum,
1924 ports_stats[pt_id].rx_bad_outer_l4_csum);
1925 if ((stats.ierrors + stats.rx_nombuf) > 0) {
1926 printf(" RX-error:%"PRIu64"\n", stats.ierrors);
1927 printf(" RX-nombufs: %14"PRIu64"\n",
1931 printf(" TX-packets: %14"PRIu64
1932 " TX-dropped:%14"PRIu64
1933 " TX-total:%14"PRIu64"\n",
1934 stats.opackets, ports_stats[pt_id].tx_dropped,
1935 stats.opackets + ports_stats[pt_id].tx_dropped);
1938 if (record_burst_stats) {
1939 if (ports_stats[pt_id].rx_stream)
1940 pkt_burst_stats_display("RX",
1941 &ports_stats[pt_id].rx_stream->rx_burst_stats);
1942 if (ports_stats[pt_id].tx_stream)
1943 pkt_burst_stats_display("TX",
1944 &ports_stats[pt_id].tx_stream->tx_burst_stats);
1947 if (port->rx_queue_stats_mapping_enabled) {
1949 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1950 printf(" Stats reg %2d RX-packets:%14"PRIu64
1951 " RX-errors:%14"PRIu64
1952 " RX-bytes:%14"PRIu64"\n",
1953 j, stats.q_ipackets[j],
1954 stats.q_errors[j], stats.q_ibytes[j]);
1958 if (port->tx_queue_stats_mapping_enabled) {
1959 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1960 printf(" Stats reg %2d TX-packets:%14"PRIu64
1963 j, stats.q_opackets[j],
1968 printf(" %s--------------------------------%s\n",
1969 fwd_stats_border, fwd_stats_border);
1972 printf("\n %s Accumulated forward statistics for all ports"
1974 acc_stats_border, acc_stats_border);
1975 printf(" RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1977 " TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1979 total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1980 total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1981 if (total_rx_nombuf > 0)
1982 printf(" RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1983 printf(" %s++++++++++++++++++++++++++++++++++++++++++++++"
1985 acc_stats_border, acc_stats_border);
1986 if (record_core_cycles) {
1987 #define CYC_PER_MHZ 1E6
1988 if (total_recv > 0 || total_xmit > 0) {
1989 uint64_t total_pkts = 0;
1990 if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 ||
1991 strcmp(cur_fwd_eng->fwd_mode_name, "flowgen") == 0)
1992 total_pkts = total_xmit;
1994 total_pkts = total_recv;
1996 printf("\n CPU cycles/packet=%.2F (total cycles="
1997 "%"PRIu64" / total %s packets=%"PRIu64") at %"PRIu64
1999 (double) fwd_cycles / total_pkts,
2000 fwd_cycles, cur_fwd_eng->fwd_mode_name, total_pkts,
2001 (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
2007 fwd_stats_reset(void)
2013 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2014 pt_id = fwd_ports_ids[i];
2015 rte_eth_stats_get(pt_id, &ports[pt_id].stats);
2017 for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
2018 struct fwd_stream *fs = fwd_streams[sm_id];
2022 fs->fwd_dropped = 0;
2023 fs->rx_bad_ip_csum = 0;
2024 fs->rx_bad_l4_csum = 0;
2025 fs->rx_bad_outer_l4_csum = 0;
2027 memset(&fs->rx_burst_stats, 0, sizeof(fs->rx_burst_stats));
2028 memset(&fs->tx_burst_stats, 0, sizeof(fs->tx_burst_stats));
2029 fs->core_cycles = 0;
2034 flush_fwd_rx_queues(void)
2036 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2043 uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
2044 uint64_t timer_period;
2046 /* convert to number of cycles */
2047 timer_period = rte_get_timer_hz(); /* 1 second timeout */
2049 for (j = 0; j < 2; j++) {
2050 for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
2051 for (rxq = 0; rxq < nb_rxq; rxq++) {
2052 port_id = fwd_ports_ids[rxp];
2054 * testpmd can stuck in the below do while loop
2055 * if rte_eth_rx_burst() always returns nonzero
2056 * packets. So timer is added to exit this loop
2057 * after 1sec timer expiry.
2059 prev_tsc = rte_rdtsc();
2061 nb_rx = rte_eth_rx_burst(port_id, rxq,
2062 pkts_burst, MAX_PKT_BURST);
2063 for (i = 0; i < nb_rx; i++)
2064 rte_pktmbuf_free(pkts_burst[i]);
2066 cur_tsc = rte_rdtsc();
2067 diff_tsc = cur_tsc - prev_tsc;
2068 timer_tsc += diff_tsc;
2069 } while ((nb_rx > 0) &&
2070 (timer_tsc < timer_period));
2074 rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
2079 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
2081 struct fwd_stream **fsm;
2084 #ifdef RTE_LIBRTE_BITRATESTATS
2085 uint64_t tics_per_1sec;
2086 uint64_t tics_datum;
2087 uint64_t tics_current;
2088 uint16_t i, cnt_ports;
2090 cnt_ports = nb_ports;
2091 tics_datum = rte_rdtsc();
2092 tics_per_1sec = rte_get_timer_hz();
2094 fsm = &fwd_streams[fc->stream_idx];
2095 nb_fs = fc->stream_nb;
2097 for (sm_id = 0; sm_id < nb_fs; sm_id++)
2098 (*pkt_fwd)(fsm[sm_id]);
2099 #ifdef RTE_LIBRTE_BITRATESTATS
2100 if (bitrate_enabled != 0 &&
2101 bitrate_lcore_id == rte_lcore_id()) {
2102 tics_current = rte_rdtsc();
2103 if (tics_current - tics_datum >= tics_per_1sec) {
2104 /* Periodic bitrate calculation */
2105 for (i = 0; i < cnt_ports; i++)
2106 rte_stats_bitrate_calc(bitrate_data,
2108 tics_datum = tics_current;
2112 #ifdef RTE_LIBRTE_LATENCY_STATS
2113 if (latencystats_enabled != 0 &&
2114 latencystats_lcore_id == rte_lcore_id())
2115 rte_latencystats_update();
2118 } while (! fc->stopped);
2122 start_pkt_forward_on_core(void *fwd_arg)
2124 run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
2125 cur_fwd_config.fwd_eng->packet_fwd);
2130 * Run the TXONLY packet forwarding engine to send a single burst of packets.
2131 * Used to start communication flows in network loopback test configurations.
2134 run_one_txonly_burst_on_core(void *fwd_arg)
2136 struct fwd_lcore *fwd_lc;
2137 struct fwd_lcore tmp_lcore;
2139 fwd_lc = (struct fwd_lcore *) fwd_arg;
2140 tmp_lcore = *fwd_lc;
2141 tmp_lcore.stopped = 1;
2142 run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
2147 * Launch packet forwarding:
2148 * - Setup per-port forwarding context.
2149 * - launch logical cores with their forwarding configuration.
2152 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
2154 port_fwd_begin_t port_fwd_begin;
2159 port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
2160 if (port_fwd_begin != NULL) {
2161 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2162 (*port_fwd_begin)(fwd_ports_ids[i]);
2164 for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
2165 lc_id = fwd_lcores_cpuids[i];
2166 if ((interactive == 0) || (lc_id != rte_lcore_id())) {
2167 fwd_lcores[i]->stopped = 0;
2168 diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
2169 fwd_lcores[i], lc_id);
2171 printf("launch lcore %u failed - diag=%d\n",
2178 * Launch packet forwarding configuration.
2181 start_packet_forwarding(int with_tx_first)
2183 port_fwd_begin_t port_fwd_begin;
2184 port_fwd_end_t port_fwd_end;
2185 struct rte_port *port;
2189 if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
2190 rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
2192 if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
2193 rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
2195 if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
2196 strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
2197 (!nb_rxq || !nb_txq))
2198 rte_exit(EXIT_FAILURE,
2199 "Either rxq or txq are 0, cannot use %s fwd mode\n",
2200 cur_fwd_eng->fwd_mode_name);
2202 if (all_ports_started() == 0) {
2203 printf("Not all ports were started\n");
2206 if (test_done == 0) {
2207 printf("Packet forwarding already started\n");
2213 for (i = 0; i < nb_fwd_ports; i++) {
2214 pt_id = fwd_ports_ids[i];
2215 port = &ports[pt_id];
2216 if (!port->dcb_flag) {
2217 printf("In DCB mode, all forwarding ports must "
2218 "be configured in this mode.\n");
2222 if (nb_fwd_lcores == 1) {
2223 printf("In DCB mode,the nb forwarding cores "
2224 "should be larger than 1.\n");
2233 flush_fwd_rx_queues();
2235 pkt_fwd_config_display(&cur_fwd_config);
2236 rxtx_config_display();
2239 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2240 pt_id = fwd_ports_ids[i];
2241 port = &ports[pt_id];
2242 map_port_queue_stats_mapping_registers(pt_id, port);
2244 if (with_tx_first) {
2245 port_fwd_begin = tx_only_engine.port_fwd_begin;
2246 if (port_fwd_begin != NULL) {
2247 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2248 (*port_fwd_begin)(fwd_ports_ids[i]);
2250 while (with_tx_first--) {
2251 launch_packet_forwarding(
2252 run_one_txonly_burst_on_core);
2253 rte_eal_mp_wait_lcore();
2255 port_fwd_end = tx_only_engine.port_fwd_end;
2256 if (port_fwd_end != NULL) {
2257 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2258 (*port_fwd_end)(fwd_ports_ids[i]);
2261 launch_packet_forwarding(start_pkt_forward_on_core);
2265 stop_packet_forwarding(void)
2267 port_fwd_end_t port_fwd_end;
2273 printf("Packet forwarding not started\n");
2276 printf("Telling cores to stop...");
2277 for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
2278 fwd_lcores[lc_id]->stopped = 1;
2279 printf("\nWaiting for lcores to finish...\n");
2280 rte_eal_mp_wait_lcore();
2281 port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
2282 if (port_fwd_end != NULL) {
2283 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2284 pt_id = fwd_ports_ids[i];
2285 (*port_fwd_end)(pt_id);
2289 fwd_stats_display();
2291 printf("\nDone.\n");
2296 dev_set_link_up(portid_t pid)
2298 if (rte_eth_dev_set_link_up(pid) < 0)
2299 printf("\nSet link up fail.\n");
2303 dev_set_link_down(portid_t pid)
2305 if (rte_eth_dev_set_link_down(pid) < 0)
2306 printf("\nSet link down fail.\n");
2310 all_ports_started(void)
2313 struct rte_port *port;
2315 RTE_ETH_FOREACH_DEV(pi) {
2317 /* Check if there is a port which is not started */
2318 if ((port->port_status != RTE_PORT_STARTED) &&
2319 (port->slave_flag == 0))
2323 /* No port is not started */
2328 port_is_stopped(portid_t port_id)
2330 struct rte_port *port = &ports[port_id];
2332 if ((port->port_status != RTE_PORT_STOPPED) &&
2333 (port->slave_flag == 0))
2339 all_ports_stopped(void)
2343 RTE_ETH_FOREACH_DEV(pi) {
2344 if (!port_is_stopped(pi))
2352 port_is_started(portid_t port_id)
2354 if (port_id_is_invalid(port_id, ENABLED_WARN))
2357 if (ports[port_id].port_status != RTE_PORT_STARTED)
2363 /* Configure the Rx and Tx hairpin queues for the selected port. */
2365 setup_hairpin_queues(portid_t pi, portid_t p_pi, uint16_t cnt_pi)
2368 struct rte_eth_hairpin_conf hairpin_conf = {
2373 struct rte_port *port = &ports[pi];
2374 uint16_t peer_rx_port = pi;
2375 uint16_t peer_tx_port = pi;
2376 uint32_t manual = 1;
2377 uint32_t tx_exp = hairpin_mode & 0x10;
2379 if (!(hairpin_mode & 0xf)) {
2383 } else if (hairpin_mode & 0x1) {
2384 peer_tx_port = rte_eth_find_next_owned_by(pi + 1,
2385 RTE_ETH_DEV_NO_OWNER);
2386 if (peer_tx_port >= RTE_MAX_ETHPORTS)
2387 peer_tx_port = rte_eth_find_next_owned_by(0,
2388 RTE_ETH_DEV_NO_OWNER);
2389 if (p_pi != RTE_MAX_ETHPORTS) {
2390 peer_rx_port = p_pi;
2394 /* Last port will be the peer RX port of the first. */
2395 RTE_ETH_FOREACH_DEV(next_pi)
2396 peer_rx_port = next_pi;
2399 } else if (hairpin_mode & 0x2) {
2401 peer_rx_port = p_pi;
2403 peer_rx_port = rte_eth_find_next_owned_by(pi + 1,
2404 RTE_ETH_DEV_NO_OWNER);
2405 if (peer_rx_port >= RTE_MAX_ETHPORTS)
2408 peer_tx_port = peer_rx_port;
2412 for (qi = nb_txq, i = 0; qi < nb_hairpinq + nb_txq; qi++) {
2413 hairpin_conf.peers[0].port = peer_rx_port;
2414 hairpin_conf.peers[0].queue = i + nb_rxq;
2415 hairpin_conf.manual_bind = !!manual;
2416 hairpin_conf.tx_explicit = !!tx_exp;
2417 diag = rte_eth_tx_hairpin_queue_setup
2418 (pi, qi, nb_txd, &hairpin_conf);
2423 /* Fail to setup rx queue, return */
2424 if (rte_atomic16_cmpset(&(port->port_status),
2426 RTE_PORT_STOPPED) == 0)
2427 printf("Port %d can not be set back "
2428 "to stopped\n", pi);
2429 printf("Fail to configure port %d hairpin "
2431 /* try to reconfigure queues next time */
2432 port->need_reconfig_queues = 1;
2435 for (qi = nb_rxq, i = 0; qi < nb_hairpinq + nb_rxq; qi++) {
2436 hairpin_conf.peers[0].port = peer_tx_port;
2437 hairpin_conf.peers[0].queue = i + nb_txq;
2438 hairpin_conf.manual_bind = !!manual;
2439 hairpin_conf.tx_explicit = !!tx_exp;
2440 diag = rte_eth_rx_hairpin_queue_setup
2441 (pi, qi, nb_rxd, &hairpin_conf);
2446 /* Fail to setup rx queue, return */
2447 if (rte_atomic16_cmpset(&(port->port_status),
2449 RTE_PORT_STOPPED) == 0)
2450 printf("Port %d can not be set back "
2451 "to stopped\n", pi);
2452 printf("Fail to configure port %d hairpin "
2454 /* try to reconfigure queues next time */
2455 port->need_reconfig_queues = 1;
2462 start_port(portid_t pid)
2464 int diag, need_check_link_status = -1;
2466 portid_t p_pi = RTE_MAX_ETHPORTS;
2467 portid_t pl[RTE_MAX_ETHPORTS];
2468 portid_t peer_pl[RTE_MAX_ETHPORTS];
2469 uint16_t cnt_pi = 0;
2470 uint16_t cfg_pi = 0;
2473 struct rte_port *port;
2474 struct rte_ether_addr mac_addr;
2475 struct rte_eth_hairpin_cap cap;
2477 if (port_id_is_invalid(pid, ENABLED_WARN))
2482 RTE_ETH_FOREACH_DEV(pi) {
2483 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2486 need_check_link_status = 0;
2488 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
2489 RTE_PORT_HANDLING) == 0) {
2490 printf("Port %d is now not stopped\n", pi);
2494 if (port->need_reconfig > 0) {
2495 port->need_reconfig = 0;
2497 if (flow_isolate_all) {
2498 int ret = port_flow_isolate(pi, 1);
2500 printf("Failed to apply isolated"
2501 " mode on port %d\n", pi);
2505 configure_rxtx_dump_callbacks(0);
2506 printf("Configuring Port %d (socket %u)\n", pi,
2508 if (nb_hairpinq > 0 &&
2509 rte_eth_dev_hairpin_capability_get(pi, &cap)) {
2510 printf("Port %d doesn't support hairpin "
2514 /* configure port */
2515 diag = rte_eth_dev_configure(pi, nb_rxq + nb_hairpinq,
2516 nb_txq + nb_hairpinq,
2519 if (rte_atomic16_cmpset(&(port->port_status),
2520 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2521 printf("Port %d can not be set back "
2522 "to stopped\n", pi);
2523 printf("Fail to configure port %d\n", pi);
2524 /* try to reconfigure port next time */
2525 port->need_reconfig = 1;
2529 if (port->need_reconfig_queues > 0) {
2530 port->need_reconfig_queues = 0;
2531 /* setup tx queues */
2532 for (qi = 0; qi < nb_txq; qi++) {
2533 if ((numa_support) &&
2534 (txring_numa[pi] != NUMA_NO_CONFIG))
2535 diag = rte_eth_tx_queue_setup(pi, qi,
2536 port->nb_tx_desc[qi],
2538 &(port->tx_conf[qi]));
2540 diag = rte_eth_tx_queue_setup(pi, qi,
2541 port->nb_tx_desc[qi],
2543 &(port->tx_conf[qi]));
2548 /* Fail to setup tx queue, return */
2549 if (rte_atomic16_cmpset(&(port->port_status),
2551 RTE_PORT_STOPPED) == 0)
2552 printf("Port %d can not be set back "
2553 "to stopped\n", pi);
2554 printf("Fail to configure port %d tx queues\n",
2556 /* try to reconfigure queues next time */
2557 port->need_reconfig_queues = 1;
2560 for (qi = 0; qi < nb_rxq; qi++) {
2561 /* setup rx queues */
2562 if ((numa_support) &&
2563 (rxring_numa[pi] != NUMA_NO_CONFIG)) {
2564 struct rte_mempool * mp =
2566 (rxring_numa[pi], 0);
2568 printf("Failed to setup RX queue:"
2569 "No mempool allocation"
2570 " on the socket %d\n",
2575 diag = rte_eth_rx_queue_setup(pi, qi,
2576 port->nb_rx_desc[qi],
2578 &(port->rx_conf[qi]),
2581 struct rte_mempool *mp =
2583 (port->socket_id, 0);
2585 printf("Failed to setup RX queue:"
2586 "No mempool allocation"
2587 " on the socket %d\n",
2591 diag = rte_eth_rx_queue_setup(pi, qi,
2592 port->nb_rx_desc[qi],
2594 &(port->rx_conf[qi]),
2600 /* Fail to setup rx queue, return */
2601 if (rte_atomic16_cmpset(&(port->port_status),
2603 RTE_PORT_STOPPED) == 0)
2604 printf("Port %d can not be set back "
2605 "to stopped\n", pi);
2606 printf("Fail to configure port %d rx queues\n",
2608 /* try to reconfigure queues next time */
2609 port->need_reconfig_queues = 1;
2612 /* setup hairpin queues */
2613 if (setup_hairpin_queues(pi, p_pi, cnt_pi) != 0)
2616 configure_rxtx_dump_callbacks(verbose_level);
2618 diag = rte_eth_dev_set_ptypes(pi, RTE_PTYPE_UNKNOWN,
2622 "Port %d: Failed to disable Ptype parsing\n",
2630 if (rte_eth_dev_start(pi) < 0) {
2631 printf("Fail to start port %d\n", pi);
2633 /* Fail to setup rx queue, return */
2634 if (rte_atomic16_cmpset(&(port->port_status),
2635 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2636 printf("Port %d can not be set back to "
2641 if (rte_atomic16_cmpset(&(port->port_status),
2642 RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2643 printf("Port %d can not be set into started\n", pi);
2645 if (eth_macaddr_get_print_err(pi, &mac_addr) == 0)
2646 printf("Port %d: %02X:%02X:%02X:%02X:%02X:%02X\n", pi,
2647 mac_addr.addr_bytes[0], mac_addr.addr_bytes[1],
2648 mac_addr.addr_bytes[2], mac_addr.addr_bytes[3],
2649 mac_addr.addr_bytes[4], mac_addr.addr_bytes[5]);
2651 /* at least one port started, need checking link status */
2652 need_check_link_status = 1;
2657 if (need_check_link_status == 1 && !no_link_check)
2658 check_all_ports_link_status(RTE_PORT_ALL);
2659 else if (need_check_link_status == 0)
2660 printf("Please stop the ports first\n");
2662 if (hairpin_mode & 0xf) {
2666 /* bind all started hairpin ports */
2667 for (i = 0; i < cfg_pi; i++) {
2669 /* bind current Tx to all peer Rx */
2670 peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2671 RTE_MAX_ETHPORTS, 1);
2674 for (j = 0; j < peer_pi; j++) {
2675 if (!port_is_started(peer_pl[j]))
2677 diag = rte_eth_hairpin_bind(pi, peer_pl[j]);
2679 printf("Error during binding hairpin"
2680 " Tx port %u to %u: %s\n",
2682 rte_strerror(-diag));
2686 /* bind all peer Tx to current Rx */
2687 peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2688 RTE_MAX_ETHPORTS, 0);
2691 for (j = 0; j < peer_pi; j++) {
2692 if (!port_is_started(peer_pl[j]))
2694 diag = rte_eth_hairpin_bind(peer_pl[j], pi);
2696 printf("Error during binding hairpin"
2697 " Tx port %u to %u: %s\n",
2699 rte_strerror(-diag));
2711 stop_port(portid_t pid)
2714 struct rte_port *port;
2715 int need_check_link_status = 0;
2716 portid_t peer_pl[RTE_MAX_ETHPORTS];
2724 if (port_id_is_invalid(pid, ENABLED_WARN))
2727 printf("Stopping ports...\n");
2729 RTE_ETH_FOREACH_DEV(pi) {
2730 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2733 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2734 printf("Please remove port %d from forwarding configuration.\n", pi);
2738 if (port_is_bonding_slave(pi)) {
2739 printf("Please remove port %d from bonded device.\n", pi);
2744 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2745 RTE_PORT_HANDLING) == 0)
2748 if (hairpin_mode & 0xf) {
2751 rte_eth_hairpin_unbind(pi, RTE_MAX_ETHPORTS);
2752 /* unbind all peer Tx from current Rx */
2753 peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2754 RTE_MAX_ETHPORTS, 0);
2757 for (j = 0; j < peer_pi; j++) {
2758 if (!port_is_started(peer_pl[j]))
2760 rte_eth_hairpin_unbind(peer_pl[j], pi);
2764 rte_eth_dev_stop(pi);
2766 if (rte_atomic16_cmpset(&(port->port_status),
2767 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2768 printf("Port %d can not be set into stopped\n", pi);
2769 need_check_link_status = 1;
2771 if (need_check_link_status && !no_link_check)
2772 check_all_ports_link_status(RTE_PORT_ALL);
2778 remove_invalid_ports_in(portid_t *array, portid_t *total)
2781 portid_t new_total = 0;
2783 for (i = 0; i < *total; i++)
2784 if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2785 array[new_total] = array[i];
2792 remove_invalid_ports(void)
2794 remove_invalid_ports_in(ports_ids, &nb_ports);
2795 remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2796 nb_cfg_ports = nb_fwd_ports;
2800 close_port(portid_t pid)
2803 struct rte_port *port;
2805 if (port_id_is_invalid(pid, ENABLED_WARN))
2808 printf("Closing ports...\n");
2810 RTE_ETH_FOREACH_DEV(pi) {
2811 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2814 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2815 printf("Please remove port %d from forwarding configuration.\n", pi);
2819 if (port_is_bonding_slave(pi)) {
2820 printf("Please remove port %d from bonded device.\n", pi);
2825 if (rte_atomic16_cmpset(&(port->port_status),
2826 RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2827 printf("Port %d is already closed\n", pi);
2831 port_flow_flush(pi);
2832 rte_eth_dev_close(pi);
2835 remove_invalid_ports();
2840 reset_port(portid_t pid)
2844 struct rte_port *port;
2846 if (port_id_is_invalid(pid, ENABLED_WARN))
2849 if ((pid == (portid_t)RTE_PORT_ALL && !all_ports_stopped()) ||
2850 (pid != (portid_t)RTE_PORT_ALL && !port_is_stopped(pid))) {
2851 printf("Can not reset port(s), please stop port(s) first.\n");
2855 printf("Resetting ports...\n");
2857 RTE_ETH_FOREACH_DEV(pi) {
2858 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2861 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2862 printf("Please remove port %d from forwarding "
2863 "configuration.\n", pi);
2867 if (port_is_bonding_slave(pi)) {
2868 printf("Please remove port %d from bonded device.\n",
2873 diag = rte_eth_dev_reset(pi);
2876 port->need_reconfig = 1;
2877 port->need_reconfig_queues = 1;
2879 printf("Failed to reset port %d. diag=%d\n", pi, diag);
2887 attach_port(char *identifier)
2890 struct rte_dev_iterator iterator;
2892 printf("Attaching a new port...\n");
2894 if (identifier == NULL) {
2895 printf("Invalid parameters are specified\n");
2899 if (rte_dev_probe(identifier) < 0) {
2900 TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2904 /* first attach mode: event */
2905 if (setup_on_probe_event) {
2906 /* new ports are detected on RTE_ETH_EVENT_NEW event */
2907 for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2908 if (ports[pi].port_status == RTE_PORT_HANDLING &&
2909 ports[pi].need_setup != 0)
2910 setup_attached_port(pi);
2914 /* second attach mode: iterator */
2915 RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2916 /* setup ports matching the devargs used for probing */
2917 if (port_is_forwarding(pi))
2918 continue; /* port was already attached before */
2919 setup_attached_port(pi);
2924 setup_attached_port(portid_t pi)
2926 unsigned int socket_id;
2929 socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2930 /* if socket_id is invalid, set to the first available socket. */
2931 if (check_socket_id(socket_id) < 0)
2932 socket_id = socket_ids[0];
2933 reconfig(pi, socket_id);
2934 ret = rte_eth_promiscuous_enable(pi);
2936 printf("Error during enabling promiscuous mode for port %u: %s - ignore\n",
2937 pi, rte_strerror(-ret));
2939 ports_ids[nb_ports++] = pi;
2940 fwd_ports_ids[nb_fwd_ports++] = pi;
2941 nb_cfg_ports = nb_fwd_ports;
2942 ports[pi].need_setup = 0;
2943 ports[pi].port_status = RTE_PORT_STOPPED;
2945 printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
2950 detach_device(struct rte_device *dev)
2955 printf("Device already removed\n");
2959 printf("Removing a device...\n");
2961 RTE_ETH_FOREACH_DEV_OF(sibling, dev) {
2962 if (ports[sibling].port_status != RTE_PORT_CLOSED) {
2963 if (ports[sibling].port_status != RTE_PORT_STOPPED) {
2964 printf("Port %u not stopped\n", sibling);
2967 port_flow_flush(sibling);
2971 if (rte_dev_remove(dev) < 0) {
2972 TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
2975 remove_invalid_ports();
2977 printf("Device is detached\n");
2978 printf("Now total ports is %d\n", nb_ports);
2984 detach_port_device(portid_t port_id)
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 printf("Port not stopped\n");
2994 printf("Port was not closed\n");
2997 detach_device(rte_eth_devices[port_id].device);
3001 detach_devargs(char *identifier)
3003 struct rte_dev_iterator iterator;
3004 struct rte_devargs da;
3007 printf("Removing a device...\n");
3009 memset(&da, 0, sizeof(da));
3010 if (rte_devargs_parsef(&da, "%s", identifier)) {
3011 printf("cannot parse identifier\n");
3017 RTE_ETH_FOREACH_MATCHING_DEV(port_id, identifier, &iterator) {
3018 if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3019 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3020 printf("Port %u not stopped\n", port_id);
3021 rte_eth_iterator_cleanup(&iterator);
3024 port_flow_flush(port_id);
3028 if (rte_eal_hotplug_remove(da.bus->name, da.name) != 0) {
3029 TESTPMD_LOG(ERR, "Failed to detach device %s(%s)\n",
3030 da.name, da.bus->name);
3034 remove_invalid_ports();
3036 printf("Device %s is detached\n", identifier);
3037 printf("Now total ports is %d\n", nb_ports);
3049 stop_packet_forwarding();
3051 for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3053 if (mp_alloc_type == MP_ALLOC_ANON)
3054 rte_mempool_mem_iter(mempools[i], dma_unmap_cb,
3058 if (ports != NULL) {
3060 RTE_ETH_FOREACH_DEV(pt_id) {
3061 printf("\nStopping port %d...\n", pt_id);
3065 RTE_ETH_FOREACH_DEV(pt_id) {
3066 printf("\nShutting down port %d...\n", pt_id);
3073 ret = rte_dev_event_monitor_stop();
3076 "fail to stop device event monitor.");
3080 ret = rte_dev_event_callback_unregister(NULL,
3081 dev_event_callback, NULL);
3084 "fail to unregister device event callback.\n");
3088 ret = rte_dev_hotplug_handle_disable();
3091 "fail to disable hotplug handling.\n");
3095 for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3097 rte_mempool_free(mempools[i]);
3100 printf("\nBye...\n");
3103 typedef void (*cmd_func_t)(void);
3104 struct pmd_test_command {
3105 const char *cmd_name;
3106 cmd_func_t cmd_func;
3109 /* Check the link status of all ports in up to 9s, and print them finally */
3111 check_all_ports_link_status(uint32_t port_mask)
3113 #define CHECK_INTERVAL 100 /* 100ms */
3114 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3116 uint8_t count, all_ports_up, print_flag = 0;
3117 struct rte_eth_link link;
3119 char link_status[RTE_ETH_LINK_MAX_STR_LEN];
3121 printf("Checking link statuses...\n");
3123 for (count = 0; count <= MAX_CHECK_TIME; count++) {
3125 RTE_ETH_FOREACH_DEV(portid) {
3126 if ((port_mask & (1 << portid)) == 0)
3128 memset(&link, 0, sizeof(link));
3129 ret = rte_eth_link_get_nowait(portid, &link);
3132 if (print_flag == 1)
3133 printf("Port %u link get failed: %s\n",
3134 portid, rte_strerror(-ret));
3137 /* print link status if flag set */
3138 if (print_flag == 1) {
3139 rte_eth_link_to_str(link_status,
3140 sizeof(link_status), &link);
3141 printf("Port %d %s\n", portid, link_status);
3144 /* clear all_ports_up flag if any link down */
3145 if (link.link_status == ETH_LINK_DOWN) {
3150 /* after finally printing all link status, get out */
3151 if (print_flag == 1)
3154 if (all_ports_up == 0) {
3156 rte_delay_ms(CHECK_INTERVAL);
3159 /* set the print_flag if all ports up or timeout */
3160 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3170 rmv_port_callback(void *arg)
3172 int need_to_start = 0;
3173 int org_no_link_check = no_link_check;
3174 portid_t port_id = (intptr_t)arg;
3175 struct rte_device *dev;
3177 RTE_ETH_VALID_PORTID_OR_RET(port_id);
3179 if (!test_done && port_is_forwarding(port_id)) {
3181 stop_packet_forwarding();
3185 no_link_check = org_no_link_check;
3187 /* Save rte_device pointer before closing ethdev port */
3188 dev = rte_eth_devices[port_id].device;
3189 close_port(port_id);
3190 detach_device(dev); /* might be already removed or have more ports */
3193 start_packet_forwarding(0);
3196 /* This function is used by the interrupt thread */
3198 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
3201 RTE_SET_USED(param);
3202 RTE_SET_USED(ret_param);
3204 if (type >= RTE_ETH_EVENT_MAX) {
3205 fprintf(stderr, "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
3206 port_id, __func__, type);
3208 } else if (event_print_mask & (UINT32_C(1) << type)) {
3209 printf("\nPort %" PRIu16 ": %s event\n", port_id,
3210 eth_event_desc[type]);
3215 case RTE_ETH_EVENT_NEW:
3216 ports[port_id].need_setup = 1;
3217 ports[port_id].port_status = RTE_PORT_HANDLING;
3219 case RTE_ETH_EVENT_INTR_RMV:
3220 if (port_id_is_invalid(port_id, DISABLED_WARN))
3222 if (rte_eal_alarm_set(100000,
3223 rmv_port_callback, (void *)(intptr_t)port_id))
3224 fprintf(stderr, "Could not set up deferred device removal\n");
3226 case RTE_ETH_EVENT_DESTROY:
3227 ports[port_id].port_status = RTE_PORT_CLOSED;
3228 printf("Port %u is closed\n", port_id);
3237 register_eth_event_callback(void)
3240 enum rte_eth_event_type event;
3242 for (event = RTE_ETH_EVENT_UNKNOWN;
3243 event < RTE_ETH_EVENT_MAX; event++) {
3244 ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
3249 TESTPMD_LOG(ERR, "Failed to register callback for "
3250 "%s event\n", eth_event_desc[event]);
3258 /* This function is used by the interrupt thread */
3260 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
3261 __rte_unused void *arg)
3266 if (type >= RTE_DEV_EVENT_MAX) {
3267 fprintf(stderr, "%s called upon invalid event %d\n",
3273 case RTE_DEV_EVENT_REMOVE:
3274 RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
3276 ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
3278 RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
3283 * Because the user's callback is invoked in eal interrupt
3284 * callback, the interrupt callback need to be finished before
3285 * it can be unregistered when detaching device. So finish
3286 * callback soon and use a deferred removal to detach device
3287 * is need. It is a workaround, once the device detaching be
3288 * moved into the eal in the future, the deferred removal could
3291 if (rte_eal_alarm_set(100000,
3292 rmv_port_callback, (void *)(intptr_t)port_id))
3294 "Could not set up deferred device removal\n");
3296 case RTE_DEV_EVENT_ADD:
3297 RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
3299 /* TODO: After finish kernel driver binding,
3300 * begin to attach port.
3309 set_tx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
3313 uint8_t mapping_found = 0;
3315 for (i = 0; i < nb_tx_queue_stats_mappings; i++) {
3316 if ((tx_queue_stats_mappings[i].port_id == port_id) &&
3317 (tx_queue_stats_mappings[i].queue_id < nb_txq )) {
3318 diag = rte_eth_dev_set_tx_queue_stats_mapping(port_id,
3319 tx_queue_stats_mappings[i].queue_id,
3320 tx_queue_stats_mappings[i].stats_counter_id);
3327 port->tx_queue_stats_mapping_enabled = 1;
3332 set_rx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
3336 uint8_t mapping_found = 0;
3338 for (i = 0; i < nb_rx_queue_stats_mappings; i++) {
3339 if ((rx_queue_stats_mappings[i].port_id == port_id) &&
3340 (rx_queue_stats_mappings[i].queue_id < nb_rxq )) {
3341 diag = rte_eth_dev_set_rx_queue_stats_mapping(port_id,
3342 rx_queue_stats_mappings[i].queue_id,
3343 rx_queue_stats_mappings[i].stats_counter_id);
3350 port->rx_queue_stats_mapping_enabled = 1;
3355 map_port_queue_stats_mapping_registers(portid_t pi, struct rte_port *port)
3359 diag = set_tx_queue_stats_mapping_registers(pi, port);
3361 if (diag == -ENOTSUP) {
3362 port->tx_queue_stats_mapping_enabled = 0;
3363 printf("TX queue stats mapping not supported port id=%d\n", pi);
3366 rte_exit(EXIT_FAILURE,
3367 "set_tx_queue_stats_mapping_registers "
3368 "failed for port id=%d diag=%d\n",
3372 diag = set_rx_queue_stats_mapping_registers(pi, port);
3374 if (diag == -ENOTSUP) {
3375 port->rx_queue_stats_mapping_enabled = 0;
3376 printf("RX queue stats mapping not supported port id=%d\n", pi);
3379 rte_exit(EXIT_FAILURE,
3380 "set_rx_queue_stats_mapping_registers "
3381 "failed for port id=%d diag=%d\n",
3387 rxtx_port_config(struct rte_port *port)
3392 for (qid = 0; qid < nb_rxq; qid++) {
3393 offloads = port->rx_conf[qid].offloads;
3394 port->rx_conf[qid] = port->dev_info.default_rxconf;
3396 port->rx_conf[qid].offloads = offloads;
3398 /* Check if any Rx parameters have been passed */
3399 if (rx_pthresh != RTE_PMD_PARAM_UNSET)
3400 port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
3402 if (rx_hthresh != RTE_PMD_PARAM_UNSET)
3403 port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
3405 if (rx_wthresh != RTE_PMD_PARAM_UNSET)
3406 port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
3408 if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
3409 port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
3411 if (rx_drop_en != RTE_PMD_PARAM_UNSET)
3412 port->rx_conf[qid].rx_drop_en = rx_drop_en;
3414 port->nb_rx_desc[qid] = nb_rxd;
3417 for (qid = 0; qid < nb_txq; qid++) {
3418 offloads = port->tx_conf[qid].offloads;
3419 port->tx_conf[qid] = port->dev_info.default_txconf;
3421 port->tx_conf[qid].offloads = offloads;
3423 /* Check if any Tx parameters have been passed */
3424 if (tx_pthresh != RTE_PMD_PARAM_UNSET)
3425 port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
3427 if (tx_hthresh != RTE_PMD_PARAM_UNSET)
3428 port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
3430 if (tx_wthresh != RTE_PMD_PARAM_UNSET)
3431 port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
3433 if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
3434 port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
3436 if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
3437 port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
3439 port->nb_tx_desc[qid] = nb_txd;
3444 init_port_config(void)
3447 struct rte_port *port;
3450 RTE_ETH_FOREACH_DEV(pid) {
3452 port->dev_conf.fdir_conf = fdir_conf;
3454 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
3459 port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3460 port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
3461 rss_hf & port->dev_info.flow_type_rss_offloads;
3463 port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3464 port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
3467 if (port->dcb_flag == 0) {
3468 if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
3469 port->dev_conf.rxmode.mq_mode =
3470 (enum rte_eth_rx_mq_mode)
3471 (rx_mq_mode & ETH_MQ_RX_RSS);
3473 port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
3476 rxtx_port_config(port);
3478 ret = eth_macaddr_get_print_err(pid, &port->eth_addr);
3482 map_port_queue_stats_mapping_registers(pid, port);
3483 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
3484 rte_pmd_ixgbe_bypass_init(pid);
3487 if (lsc_interrupt &&
3488 (rte_eth_devices[pid].data->dev_flags &
3489 RTE_ETH_DEV_INTR_LSC))
3490 port->dev_conf.intr_conf.lsc = 1;
3491 if (rmv_interrupt &&
3492 (rte_eth_devices[pid].data->dev_flags &
3493 RTE_ETH_DEV_INTR_RMV))
3494 port->dev_conf.intr_conf.rmv = 1;
3498 void set_port_slave_flag(portid_t slave_pid)
3500 struct rte_port *port;
3502 port = &ports[slave_pid];
3503 port->slave_flag = 1;
3506 void clear_port_slave_flag(portid_t slave_pid)
3508 struct rte_port *port;
3510 port = &ports[slave_pid];
3511 port->slave_flag = 0;
3514 uint8_t port_is_bonding_slave(portid_t slave_pid)
3516 struct rte_port *port;
3518 port = &ports[slave_pid];
3519 if ((rte_eth_devices[slave_pid].data->dev_flags &
3520 RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
3525 const uint16_t vlan_tags[] = {
3526 0, 1, 2, 3, 4, 5, 6, 7,
3527 8, 9, 10, 11, 12, 13, 14, 15,
3528 16, 17, 18, 19, 20, 21, 22, 23,
3529 24, 25, 26, 27, 28, 29, 30, 31
3533 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
3534 enum dcb_mode_enable dcb_mode,
3535 enum rte_eth_nb_tcs num_tcs,
3540 struct rte_eth_rss_conf rss_conf;
3543 * Builds up the correct configuration for dcb+vt based on the vlan tags array
3544 * given above, and the number of traffic classes available for use.
3546 if (dcb_mode == DCB_VT_ENABLED) {
3547 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3548 ð_conf->rx_adv_conf.vmdq_dcb_conf;
3549 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3550 ð_conf->tx_adv_conf.vmdq_dcb_tx_conf;
3552 /* VMDQ+DCB RX and TX configurations */
3553 vmdq_rx_conf->enable_default_pool = 0;
3554 vmdq_rx_conf->default_pool = 0;
3555 vmdq_rx_conf->nb_queue_pools =
3556 (num_tcs == ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3557 vmdq_tx_conf->nb_queue_pools =
3558 (num_tcs == ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3560 vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
3561 for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
3562 vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
3563 vmdq_rx_conf->pool_map[i].pools =
3564 1 << (i % vmdq_rx_conf->nb_queue_pools);
3566 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3567 vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
3568 vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
3571 /* set DCB mode of RX and TX of multiple queues */
3572 eth_conf->rxmode.mq_mode =
3573 (enum rte_eth_rx_mq_mode)
3574 (rx_mq_mode & ETH_MQ_RX_VMDQ_DCB);
3575 eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
3577 struct rte_eth_dcb_rx_conf *rx_conf =
3578 ð_conf->rx_adv_conf.dcb_rx_conf;
3579 struct rte_eth_dcb_tx_conf *tx_conf =
3580 ð_conf->tx_adv_conf.dcb_tx_conf;
3582 memset(&rss_conf, 0, sizeof(struct rte_eth_rss_conf));
3584 rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
3588 rx_conf->nb_tcs = num_tcs;
3589 tx_conf->nb_tcs = num_tcs;
3591 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3592 rx_conf->dcb_tc[i] = i % num_tcs;
3593 tx_conf->dcb_tc[i] = i % num_tcs;
3596 eth_conf->rxmode.mq_mode =
3597 (enum rte_eth_rx_mq_mode)
3598 (rx_mq_mode & ETH_MQ_RX_DCB_RSS);
3599 eth_conf->rx_adv_conf.rss_conf = rss_conf;
3600 eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
3604 eth_conf->dcb_capability_en =
3605 ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
3607 eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
3613 init_port_dcb_config(portid_t pid,
3614 enum dcb_mode_enable dcb_mode,
3615 enum rte_eth_nb_tcs num_tcs,
3618 struct rte_eth_conf port_conf;
3619 struct rte_port *rte_port;
3623 rte_port = &ports[pid];
3625 memset(&port_conf, 0, sizeof(struct rte_eth_conf));
3626 /* Enter DCB configuration status */
3629 port_conf.rxmode = rte_port->dev_conf.rxmode;
3630 port_conf.txmode = rte_port->dev_conf.txmode;
3632 /*set configuration of DCB in vt mode and DCB in non-vt mode*/
3633 retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
3636 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3638 /* re-configure the device . */
3639 retval = rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
3643 retval = eth_dev_info_get_print_err(pid, &rte_port->dev_info);
3647 /* If dev_info.vmdq_pool_base is greater than 0,
3648 * the queue id of vmdq pools is started after pf queues.
3650 if (dcb_mode == DCB_VT_ENABLED &&
3651 rte_port->dev_info.vmdq_pool_base > 0) {
3652 printf("VMDQ_DCB multi-queue mode is nonsensical"
3653 " for port %d.", pid);
3657 /* Assume the ports in testpmd have the same dcb capability
3658 * and has the same number of rxq and txq in dcb mode
3660 if (dcb_mode == DCB_VT_ENABLED) {
3661 if (rte_port->dev_info.max_vfs > 0) {
3662 nb_rxq = rte_port->dev_info.nb_rx_queues;
3663 nb_txq = rte_port->dev_info.nb_tx_queues;
3665 nb_rxq = rte_port->dev_info.max_rx_queues;
3666 nb_txq = rte_port->dev_info.max_tx_queues;
3669 /*if vt is disabled, use all pf queues */
3670 if (rte_port->dev_info.vmdq_pool_base == 0) {
3671 nb_rxq = rte_port->dev_info.max_rx_queues;
3672 nb_txq = rte_port->dev_info.max_tx_queues;
3674 nb_rxq = (queueid_t)num_tcs;
3675 nb_txq = (queueid_t)num_tcs;
3679 rx_free_thresh = 64;
3681 memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
3683 rxtx_port_config(rte_port);
3685 rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3686 for (i = 0; i < RTE_DIM(vlan_tags); i++)
3687 rx_vft_set(pid, vlan_tags[i], 1);
3689 retval = eth_macaddr_get_print_err(pid, &rte_port->eth_addr);
3693 map_port_queue_stats_mapping_registers(pid, rte_port);
3695 rte_port->dcb_flag = 1;
3705 /* Configuration of Ethernet ports. */
3706 ports = rte_zmalloc("testpmd: ports",
3707 sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3708 RTE_CACHE_LINE_SIZE);
3709 if (ports == NULL) {
3710 rte_exit(EXIT_FAILURE,
3711 "rte_zmalloc(%d struct rte_port) failed\n",
3714 for (i = 0; i < RTE_MAX_ETHPORTS; i++)
3715 LIST_INIT(&ports[i].flow_tunnel_list);
3716 /* Initialize ports NUMA structures */
3717 memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3718 memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3719 memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3733 const char clr[] = { 27, '[', '2', 'J', '\0' };
3734 const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3736 /* Clear screen and move to top left */
3737 printf("%s%s", clr, top_left);
3739 printf("\nPort statistics ====================================");
3740 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3741 nic_stats_display(fwd_ports_ids[i]);
3747 signal_handler(int signum)
3749 if (signum == SIGINT || signum == SIGTERM) {
3750 printf("\nSignal %d received, preparing to exit...\n",
3752 #ifdef RTE_LIBRTE_PDUMP
3753 /* uninitialize packet capture framework */
3756 #ifdef RTE_LIBRTE_LATENCY_STATS
3757 if (latencystats_enabled != 0)
3758 rte_latencystats_uninit();
3761 /* Set flag to indicate the force termination. */
3763 /* exit with the expected status */
3764 signal(signum, SIG_DFL);
3765 kill(getpid(), signum);
3770 main(int argc, char** argv)
3777 signal(SIGINT, signal_handler);
3778 signal(SIGTERM, signal_handler);
3780 testpmd_logtype = rte_log_register("testpmd");
3781 if (testpmd_logtype < 0)
3782 rte_exit(EXIT_FAILURE, "Cannot register log type");
3783 rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3785 diag = rte_eal_init(argc, argv);
3787 rte_exit(EXIT_FAILURE, "Cannot init EAL: %s\n",
3788 rte_strerror(rte_errno));
3790 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
3791 rte_exit(EXIT_FAILURE,
3792 "Secondary process type not supported.\n");
3794 ret = register_eth_event_callback();
3796 rte_exit(EXIT_FAILURE, "Cannot register for ethdev events");
3798 #ifdef RTE_LIBRTE_PDUMP
3799 /* initialize packet capture framework */
3804 RTE_ETH_FOREACH_DEV(port_id) {
3805 ports_ids[count] = port_id;
3808 nb_ports = (portid_t) count;
3810 TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3812 /* allocate port structures, and init them */
3815 set_def_fwd_config();
3817 rte_exit(EXIT_FAILURE, "No cores defined for forwarding\n"
3818 "Check the core mask argument\n");
3820 /* Bitrate/latency stats disabled by default */
3821 #ifdef RTE_LIBRTE_BITRATESTATS
3822 bitrate_enabled = 0;
3824 #ifdef RTE_LIBRTE_LATENCY_STATS
3825 latencystats_enabled = 0;
3828 /* on FreeBSD, mlockall() is disabled by default */
3829 #ifdef RTE_EXEC_ENV_FREEBSD
3838 launch_args_parse(argc, argv);
3840 if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3841 TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3845 if (tx_first && interactive)
3846 rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3847 "interactive mode.\n");
3849 if (tx_first && lsc_interrupt) {
3850 printf("Warning: lsc_interrupt needs to be off when "
3851 " using tx_first. Disabling.\n");
3855 if (!nb_rxq && !nb_txq)
3856 printf("Warning: Either rx or tx queues should be non-zero\n");
3858 if (nb_rxq > 1 && nb_rxq > nb_txq)
3859 printf("Warning: nb_rxq=%d enables RSS configuration, "
3860 "but nb_txq=%d will prevent to fully test it.\n",
3866 ret = rte_dev_hotplug_handle_enable();
3869 "fail to enable hotplug handling.");
3873 ret = rte_dev_event_monitor_start();
3876 "fail to start device event monitoring.");
3880 ret = rte_dev_event_callback_register(NULL,
3881 dev_event_callback, NULL);
3884 "fail to register device event callback\n");
3889 if (!no_device_start && start_port(RTE_PORT_ALL) != 0)
3890 rte_exit(EXIT_FAILURE, "Start ports failed\n");
3892 /* set all ports to promiscuous mode by default */
3893 RTE_ETH_FOREACH_DEV(port_id) {
3894 ret = rte_eth_promiscuous_enable(port_id);
3896 printf("Error during enabling promiscuous mode for port %u: %s - ignore\n",
3897 port_id, rte_strerror(-ret));
3900 /* Init metrics library */
3901 rte_metrics_init(rte_socket_id());
3903 #ifdef RTE_LIBRTE_LATENCY_STATS
3904 if (latencystats_enabled != 0) {
3905 int ret = rte_latencystats_init(1, NULL);
3907 printf("Warning: latencystats init()"
3908 " returned error %d\n", ret);
3909 printf("Latencystats running on lcore %d\n",
3910 latencystats_lcore_id);
3914 /* Setup bitrate stats */
3915 #ifdef RTE_LIBRTE_BITRATESTATS
3916 if (bitrate_enabled != 0) {
3917 bitrate_data = rte_stats_bitrate_create();
3918 if (bitrate_data == NULL)
3919 rte_exit(EXIT_FAILURE,
3920 "Could not allocate bitrate data.\n");
3921 rte_stats_bitrate_reg(bitrate_data);
3925 #ifdef RTE_LIBRTE_CMDLINE
3926 if (strlen(cmdline_filename) != 0)
3927 cmdline_read_from_file(cmdline_filename);
3929 if (interactive == 1) {
3931 printf("Start automatic packet forwarding\n");
3932 start_packet_forwarding(0);
3944 printf("No commandline core given, start packet forwarding\n");
3945 start_packet_forwarding(tx_first);
3946 if (stats_period != 0) {
3947 uint64_t prev_time = 0, cur_time, diff_time = 0;
3948 uint64_t timer_period;
3950 /* Convert to number of cycles */
3951 timer_period = stats_period * rte_get_timer_hz();
3953 while (f_quit == 0) {
3954 cur_time = rte_get_timer_cycles();
3955 diff_time += cur_time - prev_time;
3957 if (diff_time >= timer_period) {
3959 /* Reset the timer */
3962 /* Sleep to avoid unnecessary checks */
3963 prev_time = cur_time;
3968 printf("Press enter to exit\n");
3969 rc = read(0, &c, 1);
3975 ret = rte_eal_cleanup();
3977 rte_exit(EXIT_FAILURE,
3978 "EAL cleanup failed: %s\n", strerror(-ret));
3980 return EXIT_SUCCESS;