1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2017 - 2018 Intel Corporation. */
4 #include <asm/barrier.h>
9 #include <linux/compiler.h>
10 #include <linux/if_link.h>
11 #include <linux/if_xdp.h>
12 #include <linux/if_ether.h>
14 #include <net/ethernet.h>
24 #include <sys/resource.h>
25 #include <sys/socket.h>
26 #include <sys/types.h>
30 #include "bpf/libbpf.h"
46 #define NUM_FRAMES (4 * 1024)
49 #define DEBUG_HEXDUMP 0
55 static unsigned long prev_time;
63 static enum benchmark_type opt_bench = BENCH_RXDROP;
64 static u32 opt_xdp_flags = XDP_FLAGS_UPDATE_IF_NOEXIST;
65 static const char *opt_if = "";
66 static int opt_ifindex;
69 static int opt_interval = 1;
70 static u32 opt_xdp_bind_flags;
73 struct xsk_umem_info {
74 struct xsk_ring_prod fq;
75 struct xsk_ring_cons cq;
76 struct xsk_umem *umem;
80 struct xsk_socket_info {
81 struct xsk_ring_cons rx;
82 struct xsk_ring_prod tx;
83 struct xsk_umem_info *umem;
84 struct xsk_socket *xsk;
85 unsigned long rx_npkts;
86 unsigned long tx_npkts;
87 unsigned long prev_rx_npkts;
88 unsigned long prev_tx_npkts;
93 struct xsk_socket_info *xsks[MAX_SOCKS];
95 static unsigned long get_nsecs(void)
99 clock_gettime(CLOCK_MONOTONIC, &ts);
100 return ts.tv_sec * 1000000000UL + ts.tv_nsec;
103 static void print_benchmark(bool running)
105 const char *bench_str = "INVALID";
107 if (opt_bench == BENCH_RXDROP)
108 bench_str = "rxdrop";
109 else if (opt_bench == BENCH_TXONLY)
110 bench_str = "txonly";
111 else if (opt_bench == BENCH_L2FWD)
114 printf("%s:%d %s ", opt_if, opt_queue, bench_str);
115 if (opt_xdp_flags & XDP_FLAGS_SKB_MODE)
117 else if (opt_xdp_flags & XDP_FLAGS_DRV_MODE)
126 printf("running...");
131 static void dump_stats(void)
133 unsigned long now = get_nsecs();
134 long dt = now - prev_time;
139 for (i = 0; i < num_socks && xsks[i]; i++) {
140 char *fmt = "%-15s %'-11.0f %'-11lu\n";
141 double rx_pps, tx_pps;
143 rx_pps = (xsks[i]->rx_npkts - xsks[i]->prev_rx_npkts) *
145 tx_pps = (xsks[i]->tx_npkts - xsks[i]->prev_tx_npkts) *
148 printf("\n sock%d@", i);
149 print_benchmark(false);
152 printf("%-15s %-11s %-11s %-11.2f\n", "", "pps", "pkts",
154 printf(fmt, "rx", rx_pps, xsks[i]->rx_npkts);
155 printf(fmt, "tx", tx_pps, xsks[i]->tx_npkts);
157 xsks[i]->prev_rx_npkts = xsks[i]->rx_npkts;
158 xsks[i]->prev_tx_npkts = xsks[i]->tx_npkts;
162 static void *poller(void *arg)
173 static void remove_xdp_program(void)
175 __u32 curr_prog_id = 0;
177 if (bpf_get_link_xdp_id(opt_ifindex, &curr_prog_id, opt_xdp_flags)) {
178 printf("bpf_get_link_xdp_id failed\n");
181 if (prog_id == curr_prog_id)
182 bpf_set_link_xdp_fd(opt_ifindex, -1, opt_xdp_flags);
183 else if (!curr_prog_id)
184 printf("couldn't find a prog id on a given interface\n");
186 printf("program on interface changed, not removing\n");
189 static void int_exit(int sig)
191 struct xsk_umem *umem = xsks[0]->umem->umem;
196 xsk_socket__delete(xsks[0]->xsk);
197 (void)xsk_umem__delete(umem);
198 remove_xdp_program();
203 static void __exit_with_error(int error, const char *file, const char *func,
206 fprintf(stderr, "%s:%s:%i: errno: %d/\"%s\"\n", file, func,
207 line, error, strerror(error));
209 remove_xdp_program();
213 #define exit_with_error(error) __exit_with_error(error, __FILE__, __func__, \
216 static const char pkt_data[] =
217 "\x3c\xfd\xfe\x9e\x7f\x71\xec\xb1\xd7\x98\x3a\xc0\x08\x00\x45\x00"
218 "\x00\x2e\x00\x00\x00\x00\x40\x11\x88\x97\x05\x08\x07\x08\xc8\x14"
219 "\x1e\x04\x10\x92\x10\x92\x00\x1a\x6d\xa3\x34\x33\x1f\x69\x40\x6b"
220 "\x54\x59\xb6\x14\x2d\x11\x44\xbf\xaf\xd9\xbe\xaa";
222 static void swap_mac_addresses(void *data)
224 struct ether_header *eth = (struct ether_header *)data;
225 struct ether_addr *src_addr = (struct ether_addr *)ð->ether_shost;
226 struct ether_addr *dst_addr = (struct ether_addr *)ð->ether_dhost;
227 struct ether_addr tmp;
230 *src_addr = *dst_addr;
234 static void hex_dump(void *pkt, size_t length, u64 addr)
236 const unsigned char *address = (unsigned char *)pkt;
237 const unsigned char *line = address;
238 size_t line_size = 32;
246 sprintf(buf, "addr=%llu", addr);
247 printf("length = %zu\n", length);
248 printf("%s | ", buf);
249 while (length-- > 0) {
250 printf("%02X ", *address++);
251 if (!(++i % line_size) || (length == 0 && i % line_size)) {
253 while (i++ % line_size)
256 printf(" | "); /* right close */
257 while (line < address) {
259 printf("%c", (c < 33 || c == 255) ? 0x2E : c);
263 printf("%s | ", buf);
269 static size_t gen_eth_frame(struct xsk_umem_info *umem, u64 addr)
271 memcpy(xsk_umem__get_data(umem->buffer, addr), pkt_data,
272 sizeof(pkt_data) - 1);
273 return sizeof(pkt_data) - 1;
276 static struct xsk_umem_info *xsk_configure_umem(void *buffer, u64 size)
278 struct xsk_umem_info *umem;
281 umem = calloc(1, sizeof(*umem));
283 exit_with_error(errno);
285 ret = xsk_umem__create(&umem->umem, buffer, size, &umem->fq, &umem->cq,
288 exit_with_error(-ret);
290 umem->buffer = buffer;
294 static struct xsk_socket_info *xsk_configure_socket(struct xsk_umem_info *umem)
296 struct xsk_socket_config cfg;
297 struct xsk_socket_info *xsk;
302 xsk = calloc(1, sizeof(*xsk));
304 exit_with_error(errno);
307 cfg.rx_size = XSK_RING_CONS__DEFAULT_NUM_DESCS;
308 cfg.tx_size = XSK_RING_PROD__DEFAULT_NUM_DESCS;
309 cfg.libbpf_flags = 0;
310 cfg.xdp_flags = opt_xdp_flags;
311 cfg.bind_flags = opt_xdp_bind_flags;
312 ret = xsk_socket__create(&xsk->xsk, opt_if, opt_queue, umem->umem,
313 &xsk->rx, &xsk->tx, &cfg);
315 exit_with_error(-ret);
317 ret = bpf_get_link_xdp_id(opt_ifindex, &prog_id, opt_xdp_flags);
319 exit_with_error(-ret);
321 ret = xsk_ring_prod__reserve(&xsk->umem->fq,
322 XSK_RING_PROD__DEFAULT_NUM_DESCS,
324 if (ret != XSK_RING_PROD__DEFAULT_NUM_DESCS)
325 exit_with_error(-ret);
327 i < XSK_RING_PROD__DEFAULT_NUM_DESCS *
328 XSK_UMEM__DEFAULT_FRAME_SIZE;
329 i += XSK_UMEM__DEFAULT_FRAME_SIZE)
330 *xsk_ring_prod__fill_addr(&xsk->umem->fq, idx++) = i;
331 xsk_ring_prod__submit(&xsk->umem->fq,
332 XSK_RING_PROD__DEFAULT_NUM_DESCS);
337 static struct option long_options[] = {
338 {"rxdrop", no_argument, 0, 'r'},
339 {"txonly", no_argument, 0, 't'},
340 {"l2fwd", no_argument, 0, 'l'},
341 {"interface", required_argument, 0, 'i'},
342 {"queue", required_argument, 0, 'q'},
343 {"poll", no_argument, 0, 'p'},
344 {"xdp-skb", no_argument, 0, 'S'},
345 {"xdp-native", no_argument, 0, 'N'},
346 {"interval", required_argument, 0, 'n'},
347 {"zero-copy", no_argument, 0, 'z'},
348 {"copy", no_argument, 0, 'c'},
352 static void usage(const char *prog)
355 " Usage: %s [OPTIONS]\n"
357 " -r, --rxdrop Discard all incoming packets (default)\n"
358 " -t, --txonly Only send packets\n"
359 " -l, --l2fwd MAC swap L2 forwarding\n"
360 " -i, --interface=n Run on interface n\n"
361 " -q, --queue=n Use queue n (default 0)\n"
362 " -p, --poll Use poll syscall\n"
363 " -S, --xdp-skb=n Use XDP skb-mod\n"
364 " -N, --xdp-native=n Enfore XDP native mode\n"
365 " -n, --interval=n Specify statistics update interval (default 1 sec).\n"
366 " -z, --zero-copy Force zero-copy mode.\n"
367 " -c, --copy Force copy mode.\n"
369 fprintf(stderr, str, prog);
373 static void parse_command_line(int argc, char **argv)
380 c = getopt_long(argc, argv, "Frtli:q:psSNn:cz", long_options,
387 opt_bench = BENCH_RXDROP;
390 opt_bench = BENCH_TXONLY;
393 opt_bench = BENCH_L2FWD;
399 opt_queue = atoi(optarg);
405 opt_xdp_flags |= XDP_FLAGS_SKB_MODE;
406 opt_xdp_bind_flags |= XDP_COPY;
409 opt_xdp_flags |= XDP_FLAGS_DRV_MODE;
412 opt_interval = atoi(optarg);
415 opt_xdp_bind_flags |= XDP_ZEROCOPY;
418 opt_xdp_bind_flags |= XDP_COPY;
421 opt_xdp_flags &= ~XDP_FLAGS_UPDATE_IF_NOEXIST;
424 usage(basename(argv[0]));
428 opt_ifindex = if_nametoindex(opt_if);
430 fprintf(stderr, "ERROR: interface \"%s\" does not exist\n",
432 usage(basename(argv[0]));
437 static void kick_tx(struct xsk_socket_info *xsk)
441 ret = sendto(xsk_socket__fd(xsk->xsk), NULL, 0, MSG_DONTWAIT, NULL, 0);
442 if (ret >= 0 || errno == ENOBUFS || errno == EAGAIN || errno == EBUSY)
444 exit_with_error(errno);
447 static inline void complete_tx_l2fwd(struct xsk_socket_info *xsk)
449 u32 idx_cq = 0, idx_fq = 0;
453 if (!xsk->outstanding_tx)
457 ndescs = (xsk->outstanding_tx > BATCH_SIZE) ? BATCH_SIZE :
460 /* re-add completed Tx buffers */
461 rcvd = xsk_ring_cons__peek(&xsk->umem->cq, ndescs, &idx_cq);
466 ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd, &idx_fq);
467 while (ret != rcvd) {
469 exit_with_error(-ret);
470 ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd,
473 for (i = 0; i < rcvd; i++)
474 *xsk_ring_prod__fill_addr(&xsk->umem->fq, idx_fq++) =
475 *xsk_ring_cons__comp_addr(&xsk->umem->cq,
478 xsk_ring_prod__submit(&xsk->umem->fq, rcvd);
479 xsk_ring_cons__release(&xsk->umem->cq, rcvd);
480 xsk->outstanding_tx -= rcvd;
481 xsk->tx_npkts += rcvd;
485 static inline void complete_tx_only(struct xsk_socket_info *xsk)
490 if (!xsk->outstanding_tx)
495 rcvd = xsk_ring_cons__peek(&xsk->umem->cq, BATCH_SIZE, &idx);
497 xsk_ring_cons__release(&xsk->umem->cq, rcvd);
498 xsk->outstanding_tx -= rcvd;
499 xsk->tx_npkts += rcvd;
503 static void rx_drop(struct xsk_socket_info *xsk)
505 unsigned int rcvd, i;
506 u32 idx_rx = 0, idx_fq = 0;
509 rcvd = xsk_ring_cons__peek(&xsk->rx, BATCH_SIZE, &idx_rx);
513 ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd, &idx_fq);
514 while (ret != rcvd) {
516 exit_with_error(-ret);
517 ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd, &idx_fq);
520 for (i = 0; i < rcvd; i++) {
521 u64 addr = xsk_ring_cons__rx_desc(&xsk->rx, idx_rx)->addr;
522 u32 len = xsk_ring_cons__rx_desc(&xsk->rx, idx_rx++)->len;
523 char *pkt = xsk_umem__get_data(xsk->umem->buffer, addr);
525 hex_dump(pkt, len, addr);
526 *xsk_ring_prod__fill_addr(&xsk->umem->fq, idx_fq++) = addr;
529 xsk_ring_prod__submit(&xsk->umem->fq, rcvd);
530 xsk_ring_cons__release(&xsk->rx, rcvd);
531 xsk->rx_npkts += rcvd;
534 static void rx_drop_all(void)
536 struct pollfd fds[MAX_SOCKS + 1];
537 int i, ret, timeout, nfds = 1;
539 memset(fds, 0, sizeof(fds));
541 for (i = 0; i < num_socks; i++) {
542 fds[i].fd = xsk_socket__fd(xsks[i]->xsk);
543 fds[i].events = POLLIN;
544 timeout = 1000; /* 1sn */
549 ret = poll(fds, nfds, timeout);
554 for (i = 0; i < num_socks; i++)
559 static void tx_only(struct xsk_socket_info *xsk)
561 int timeout, ret, nfds = 1;
562 struct pollfd fds[nfds + 1];
563 u32 idx, frame_nb = 0;
565 memset(fds, 0, sizeof(fds));
566 fds[0].fd = xsk_socket__fd(xsk->xsk);
567 fds[0].events = POLLOUT;
568 timeout = 1000; /* 1sn */
572 ret = poll(fds, nfds, timeout);
576 if (!(fds[0].revents & POLLOUT))
580 if (xsk_ring_prod__reserve(&xsk->tx, BATCH_SIZE, &idx) ==
584 for (i = 0; i < BATCH_SIZE; i++) {
585 xsk_ring_prod__tx_desc(&xsk->tx, idx + i)->addr
587 XSK_UMEM__DEFAULT_FRAME_SHIFT;
588 xsk_ring_prod__tx_desc(&xsk->tx, idx + i)->len =
589 sizeof(pkt_data) - 1;
592 xsk_ring_prod__submit(&xsk->tx, BATCH_SIZE);
593 xsk->outstanding_tx += BATCH_SIZE;
594 frame_nb += BATCH_SIZE;
595 frame_nb %= NUM_FRAMES;
598 complete_tx_only(xsk);
602 static void l2fwd(struct xsk_socket_info *xsk)
605 unsigned int rcvd, i;
606 u32 idx_rx = 0, idx_tx = 0;
610 complete_tx_l2fwd(xsk);
612 rcvd = xsk_ring_cons__peek(&xsk->rx, BATCH_SIZE,
618 ret = xsk_ring_prod__reserve(&xsk->tx, rcvd, &idx_tx);
619 while (ret != rcvd) {
621 exit_with_error(-ret);
622 ret = xsk_ring_prod__reserve(&xsk->tx, rcvd, &idx_tx);
625 for (i = 0; i < rcvd; i++) {
626 u64 addr = xsk_ring_cons__rx_desc(&xsk->rx,
628 u32 len = xsk_ring_cons__rx_desc(&xsk->rx,
630 char *pkt = xsk_umem__get_data(xsk->umem->buffer, addr);
632 swap_mac_addresses(pkt);
634 hex_dump(pkt, len, addr);
635 xsk_ring_prod__tx_desc(&xsk->tx, idx_tx)->addr = addr;
636 xsk_ring_prod__tx_desc(&xsk->tx, idx_tx++)->len = len;
639 xsk_ring_prod__submit(&xsk->tx, rcvd);
640 xsk_ring_cons__release(&xsk->rx, rcvd);
642 xsk->rx_npkts += rcvd;
643 xsk->outstanding_tx += rcvd;
647 int main(int argc, char **argv)
649 struct rlimit r = {RLIM_INFINITY, RLIM_INFINITY};
650 struct xsk_umem_info *umem;
655 parse_command_line(argc, argv);
657 if (setrlimit(RLIMIT_MEMLOCK, &r)) {
658 fprintf(stderr, "ERROR: setrlimit(RLIMIT_MEMLOCK) \"%s\"\n",
663 ret = posix_memalign(&bufs, getpagesize(), /* PAGE_SIZE aligned */
664 NUM_FRAMES * XSK_UMEM__DEFAULT_FRAME_SIZE);
666 exit_with_error(ret);
668 /* Create sockets... */
669 umem = xsk_configure_umem(bufs,
670 NUM_FRAMES * XSK_UMEM__DEFAULT_FRAME_SIZE);
671 xsks[num_socks++] = xsk_configure_socket(umem);
673 if (opt_bench == BENCH_TXONLY) {
676 for (i = 0; i < NUM_FRAMES * XSK_UMEM__DEFAULT_FRAME_SIZE;
677 i += XSK_UMEM__DEFAULT_FRAME_SIZE)
678 (void)gen_eth_frame(umem, i);
681 signal(SIGINT, int_exit);
682 signal(SIGTERM, int_exit);
683 signal(SIGABRT, int_exit);
685 setlocale(LC_ALL, "");
687 ret = pthread_create(&pt, NULL, poller, NULL);
689 exit_with_error(ret);
691 prev_time = get_nsecs();
693 if (opt_bench == BENCH_RXDROP)
695 else if (opt_bench == BENCH_TXONLY)