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>
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;
71 static int opt_xsk_frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE;
74 struct xsk_umem_info {
75 struct xsk_ring_prod fq;
76 struct xsk_ring_cons cq;
77 struct xsk_umem *umem;
81 struct xsk_socket_info {
82 struct xsk_ring_cons rx;
83 struct xsk_ring_prod tx;
84 struct xsk_umem_info *umem;
85 struct xsk_socket *xsk;
86 unsigned long rx_npkts;
87 unsigned long tx_npkts;
88 unsigned long prev_rx_npkts;
89 unsigned long prev_tx_npkts;
94 struct xsk_socket_info *xsks[MAX_SOCKS];
96 static unsigned long get_nsecs(void)
100 clock_gettime(CLOCK_MONOTONIC, &ts);
101 return ts.tv_sec * 1000000000UL + ts.tv_nsec;
104 static void print_benchmark(bool running)
106 const char *bench_str = "INVALID";
108 if (opt_bench == BENCH_RXDROP)
109 bench_str = "rxdrop";
110 else if (opt_bench == BENCH_TXONLY)
111 bench_str = "txonly";
112 else if (opt_bench == BENCH_L2FWD)
115 printf("%s:%d %s ", opt_if, opt_queue, bench_str);
116 if (opt_xdp_flags & XDP_FLAGS_SKB_MODE)
118 else if (opt_xdp_flags & XDP_FLAGS_DRV_MODE)
127 printf("running...");
132 static void dump_stats(void)
134 unsigned long now = get_nsecs();
135 long dt = now - prev_time;
140 for (i = 0; i < num_socks && xsks[i]; i++) {
141 char *fmt = "%-15s %'-11.0f %'-11lu\n";
142 double rx_pps, tx_pps;
144 rx_pps = (xsks[i]->rx_npkts - xsks[i]->prev_rx_npkts) *
146 tx_pps = (xsks[i]->tx_npkts - xsks[i]->prev_tx_npkts) *
149 printf("\n sock%d@", i);
150 print_benchmark(false);
153 printf("%-15s %-11s %-11s %-11.2f\n", "", "pps", "pkts",
155 printf(fmt, "rx", rx_pps, xsks[i]->rx_npkts);
156 printf(fmt, "tx", tx_pps, xsks[i]->tx_npkts);
158 xsks[i]->prev_rx_npkts = xsks[i]->rx_npkts;
159 xsks[i]->prev_tx_npkts = xsks[i]->tx_npkts;
163 static void *poller(void *arg)
174 static void remove_xdp_program(void)
176 __u32 curr_prog_id = 0;
178 if (bpf_get_link_xdp_id(opt_ifindex, &curr_prog_id, opt_xdp_flags)) {
179 printf("bpf_get_link_xdp_id failed\n");
182 if (prog_id == curr_prog_id)
183 bpf_set_link_xdp_fd(opt_ifindex, -1, opt_xdp_flags);
184 else if (!curr_prog_id)
185 printf("couldn't find a prog id on a given interface\n");
187 printf("program on interface changed, not removing\n");
190 static void int_exit(int sig)
192 struct xsk_umem *umem = xsks[0]->umem->umem;
197 xsk_socket__delete(xsks[0]->xsk);
198 (void)xsk_umem__delete(umem);
199 remove_xdp_program();
204 static void __exit_with_error(int error, const char *file, const char *func,
207 fprintf(stderr, "%s:%s:%i: errno: %d/\"%s\"\n", file, func,
208 line, error, strerror(error));
210 remove_xdp_program();
214 #define exit_with_error(error) __exit_with_error(error, __FILE__, __func__, \
217 static const char pkt_data[] =
218 "\x3c\xfd\xfe\x9e\x7f\x71\xec\xb1\xd7\x98\x3a\xc0\x08\x00\x45\x00"
219 "\x00\x2e\x00\x00\x00\x00\x40\x11\x88\x97\x05\x08\x07\x08\xc8\x14"
220 "\x1e\x04\x10\x92\x10\x92\x00\x1a\x6d\xa3\x34\x33\x1f\x69\x40\x6b"
221 "\x54\x59\xb6\x14\x2d\x11\x44\xbf\xaf\xd9\xbe\xaa";
223 static void swap_mac_addresses(void *data)
225 struct ether_header *eth = (struct ether_header *)data;
226 struct ether_addr *src_addr = (struct ether_addr *)ð->ether_shost;
227 struct ether_addr *dst_addr = (struct ether_addr *)ð->ether_dhost;
228 struct ether_addr tmp;
231 *src_addr = *dst_addr;
235 static void hex_dump(void *pkt, size_t length, u64 addr)
237 const unsigned char *address = (unsigned char *)pkt;
238 const unsigned char *line = address;
239 size_t line_size = 32;
247 sprintf(buf, "addr=%llu", addr);
248 printf("length = %zu\n", length);
249 printf("%s | ", buf);
250 while (length-- > 0) {
251 printf("%02X ", *address++);
252 if (!(++i % line_size) || (length == 0 && i % line_size)) {
254 while (i++ % line_size)
257 printf(" | "); /* right close */
258 while (line < address) {
260 printf("%c", (c < 33 || c == 255) ? 0x2E : c);
264 printf("%s | ", buf);
270 static size_t gen_eth_frame(struct xsk_umem_info *umem, u64 addr)
272 memcpy(xsk_umem__get_data(umem->buffer, addr), pkt_data,
273 sizeof(pkt_data) - 1);
274 return sizeof(pkt_data) - 1;
277 static struct xsk_umem_info *xsk_configure_umem(void *buffer, u64 size)
279 struct xsk_umem_info *umem;
280 struct xsk_umem_config cfg = {
281 .fill_size = XSK_RING_PROD__DEFAULT_NUM_DESCS,
282 .comp_size = XSK_RING_CONS__DEFAULT_NUM_DESCS,
283 .frame_size = opt_xsk_frame_size,
284 .frame_headroom = XSK_UMEM__DEFAULT_FRAME_HEADROOM,
288 umem = calloc(1, sizeof(*umem));
290 exit_with_error(errno);
292 ret = xsk_umem__create(&umem->umem, buffer, size, &umem->fq, &umem->cq,
295 exit_with_error(-ret);
297 umem->buffer = buffer;
301 static struct xsk_socket_info *xsk_configure_socket(struct xsk_umem_info *umem)
303 struct xsk_socket_config cfg;
304 struct xsk_socket_info *xsk;
309 xsk = calloc(1, sizeof(*xsk));
311 exit_with_error(errno);
314 cfg.rx_size = XSK_RING_CONS__DEFAULT_NUM_DESCS;
315 cfg.tx_size = XSK_RING_PROD__DEFAULT_NUM_DESCS;
316 cfg.libbpf_flags = 0;
317 cfg.xdp_flags = opt_xdp_flags;
318 cfg.bind_flags = opt_xdp_bind_flags;
319 ret = xsk_socket__create(&xsk->xsk, opt_if, opt_queue, umem->umem,
320 &xsk->rx, &xsk->tx, &cfg);
322 exit_with_error(-ret);
324 ret = bpf_get_link_xdp_id(opt_ifindex, &prog_id, opt_xdp_flags);
326 exit_with_error(-ret);
328 ret = xsk_ring_prod__reserve(&xsk->umem->fq,
329 XSK_RING_PROD__DEFAULT_NUM_DESCS,
331 if (ret != XSK_RING_PROD__DEFAULT_NUM_DESCS)
332 exit_with_error(-ret);
333 for (i = 0; i < XSK_RING_PROD__DEFAULT_NUM_DESCS; i++)
334 *xsk_ring_prod__fill_addr(&xsk->umem->fq, idx++) =
335 i * opt_xsk_frame_size;
336 xsk_ring_prod__submit(&xsk->umem->fq,
337 XSK_RING_PROD__DEFAULT_NUM_DESCS);
342 static struct option long_options[] = {
343 {"rxdrop", no_argument, 0, 'r'},
344 {"txonly", no_argument, 0, 't'},
345 {"l2fwd", no_argument, 0, 'l'},
346 {"interface", required_argument, 0, 'i'},
347 {"queue", required_argument, 0, 'q'},
348 {"poll", no_argument, 0, 'p'},
349 {"xdp-skb", no_argument, 0, 'S'},
350 {"xdp-native", no_argument, 0, 'N'},
351 {"interval", required_argument, 0, 'n'},
352 {"zero-copy", no_argument, 0, 'z'},
353 {"copy", no_argument, 0, 'c'},
354 {"frame-size", required_argument, 0, 'f'},
358 static void usage(const char *prog)
361 " Usage: %s [OPTIONS]\n"
363 " -r, --rxdrop Discard all incoming packets (default)\n"
364 " -t, --txonly Only send packets\n"
365 " -l, --l2fwd MAC swap L2 forwarding\n"
366 " -i, --interface=n Run on interface n\n"
367 " -q, --queue=n Use queue n (default 0)\n"
368 " -p, --poll Use poll syscall\n"
369 " -S, --xdp-skb=n Use XDP skb-mod\n"
370 " -N, --xdp-native=n Enfore XDP native mode\n"
371 " -n, --interval=n Specify statistics update interval (default 1 sec).\n"
372 " -z, --zero-copy Force zero-copy mode.\n"
373 " -c, --copy Force copy mode.\n"
374 " -f, --frame-size=n Set the frame size (must be a power of two, default is %d).\n"
376 fprintf(stderr, str, prog, XSK_UMEM__DEFAULT_FRAME_SIZE);
380 static void parse_command_line(int argc, char **argv)
387 c = getopt_long(argc, argv, "Frtli:q:psSNn:czf:", long_options,
394 opt_bench = BENCH_RXDROP;
397 opt_bench = BENCH_TXONLY;
400 opt_bench = BENCH_L2FWD;
406 opt_queue = atoi(optarg);
412 opt_xdp_flags |= XDP_FLAGS_SKB_MODE;
413 opt_xdp_bind_flags |= XDP_COPY;
416 opt_xdp_flags |= XDP_FLAGS_DRV_MODE;
419 opt_interval = atoi(optarg);
422 opt_xdp_bind_flags |= XDP_ZEROCOPY;
425 opt_xdp_bind_flags |= XDP_COPY;
428 opt_xdp_flags &= ~XDP_FLAGS_UPDATE_IF_NOEXIST;
431 opt_xsk_frame_size = atoi(optarg);
434 usage(basename(argv[0]));
438 opt_ifindex = if_nametoindex(opt_if);
440 fprintf(stderr, "ERROR: interface \"%s\" does not exist\n",
442 usage(basename(argv[0]));
445 if (opt_xsk_frame_size & (opt_xsk_frame_size - 1)) {
446 fprintf(stderr, "--frame-size=%d is not a power of two\n",
448 usage(basename(argv[0]));
452 static void kick_tx(struct xsk_socket_info *xsk)
456 ret = sendto(xsk_socket__fd(xsk->xsk), NULL, 0, MSG_DONTWAIT, NULL, 0);
457 if (ret >= 0 || errno == ENOBUFS || errno == EAGAIN || errno == EBUSY)
459 exit_with_error(errno);
462 static inline void complete_tx_l2fwd(struct xsk_socket_info *xsk)
464 u32 idx_cq = 0, idx_fq = 0;
468 if (!xsk->outstanding_tx)
472 ndescs = (xsk->outstanding_tx > BATCH_SIZE) ? BATCH_SIZE :
475 /* re-add completed Tx buffers */
476 rcvd = xsk_ring_cons__peek(&xsk->umem->cq, ndescs, &idx_cq);
481 ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd, &idx_fq);
482 while (ret != rcvd) {
484 exit_with_error(-ret);
485 ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd,
488 for (i = 0; i < rcvd; i++)
489 *xsk_ring_prod__fill_addr(&xsk->umem->fq, idx_fq++) =
490 *xsk_ring_cons__comp_addr(&xsk->umem->cq,
493 xsk_ring_prod__submit(&xsk->umem->fq, rcvd);
494 xsk_ring_cons__release(&xsk->umem->cq, rcvd);
495 xsk->outstanding_tx -= rcvd;
496 xsk->tx_npkts += rcvd;
500 static inline void complete_tx_only(struct xsk_socket_info *xsk)
505 if (!xsk->outstanding_tx)
510 rcvd = xsk_ring_cons__peek(&xsk->umem->cq, BATCH_SIZE, &idx);
512 xsk_ring_cons__release(&xsk->umem->cq, rcvd);
513 xsk->outstanding_tx -= rcvd;
514 xsk->tx_npkts += rcvd;
518 static void rx_drop(struct xsk_socket_info *xsk)
520 unsigned int rcvd, i;
521 u32 idx_rx = 0, idx_fq = 0;
524 rcvd = xsk_ring_cons__peek(&xsk->rx, BATCH_SIZE, &idx_rx);
528 ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd, &idx_fq);
529 while (ret != rcvd) {
531 exit_with_error(-ret);
532 ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd, &idx_fq);
535 for (i = 0; i < rcvd; i++) {
536 u64 addr = xsk_ring_cons__rx_desc(&xsk->rx, idx_rx)->addr;
537 u32 len = xsk_ring_cons__rx_desc(&xsk->rx, idx_rx++)->len;
538 char *pkt = xsk_umem__get_data(xsk->umem->buffer, addr);
540 hex_dump(pkt, len, addr);
541 *xsk_ring_prod__fill_addr(&xsk->umem->fq, idx_fq++) = addr;
544 xsk_ring_prod__submit(&xsk->umem->fq, rcvd);
545 xsk_ring_cons__release(&xsk->rx, rcvd);
546 xsk->rx_npkts += rcvd;
549 static void rx_drop_all(void)
551 struct pollfd fds[MAX_SOCKS + 1];
552 int i, ret, timeout, nfds = 1;
554 memset(fds, 0, sizeof(fds));
556 for (i = 0; i < num_socks; i++) {
557 fds[i].fd = xsk_socket__fd(xsks[i]->xsk);
558 fds[i].events = POLLIN;
559 timeout = 1000; /* 1sn */
564 ret = poll(fds, nfds, timeout);
569 for (i = 0; i < num_socks; i++)
574 static void tx_only(struct xsk_socket_info *xsk)
576 int timeout, ret, nfds = 1;
577 struct pollfd fds[nfds + 1];
578 u32 idx, frame_nb = 0;
580 memset(fds, 0, sizeof(fds));
581 fds[0].fd = xsk_socket__fd(xsk->xsk);
582 fds[0].events = POLLOUT;
583 timeout = 1000; /* 1sn */
587 ret = poll(fds, nfds, timeout);
591 if (!(fds[0].revents & POLLOUT))
595 if (xsk_ring_prod__reserve(&xsk->tx, BATCH_SIZE, &idx) ==
599 for (i = 0; i < BATCH_SIZE; i++) {
600 xsk_ring_prod__tx_desc(&xsk->tx, idx + i)->addr
601 = (frame_nb + i) * opt_xsk_frame_size;
602 xsk_ring_prod__tx_desc(&xsk->tx, idx + i)->len =
603 sizeof(pkt_data) - 1;
606 xsk_ring_prod__submit(&xsk->tx, BATCH_SIZE);
607 xsk->outstanding_tx += BATCH_SIZE;
608 frame_nb += BATCH_SIZE;
609 frame_nb %= NUM_FRAMES;
612 complete_tx_only(xsk);
616 static void l2fwd(struct xsk_socket_info *xsk)
619 unsigned int rcvd, i;
620 u32 idx_rx = 0, idx_tx = 0;
624 complete_tx_l2fwd(xsk);
626 rcvd = xsk_ring_cons__peek(&xsk->rx, BATCH_SIZE,
632 ret = xsk_ring_prod__reserve(&xsk->tx, rcvd, &idx_tx);
633 while (ret != rcvd) {
635 exit_with_error(-ret);
636 ret = xsk_ring_prod__reserve(&xsk->tx, rcvd, &idx_tx);
639 for (i = 0; i < rcvd; i++) {
640 u64 addr = xsk_ring_cons__rx_desc(&xsk->rx,
642 u32 len = xsk_ring_cons__rx_desc(&xsk->rx,
644 char *pkt = xsk_umem__get_data(xsk->umem->buffer, addr);
646 swap_mac_addresses(pkt);
648 hex_dump(pkt, len, addr);
649 xsk_ring_prod__tx_desc(&xsk->tx, idx_tx)->addr = addr;
650 xsk_ring_prod__tx_desc(&xsk->tx, idx_tx++)->len = len;
653 xsk_ring_prod__submit(&xsk->tx, rcvd);
654 xsk_ring_cons__release(&xsk->rx, rcvd);
656 xsk->rx_npkts += rcvd;
657 xsk->outstanding_tx += rcvd;
661 int main(int argc, char **argv)
663 struct rlimit r = {RLIM_INFINITY, RLIM_INFINITY};
664 struct xsk_umem_info *umem;
669 parse_command_line(argc, argv);
671 if (setrlimit(RLIMIT_MEMLOCK, &r)) {
672 fprintf(stderr, "ERROR: setrlimit(RLIMIT_MEMLOCK) \"%s\"\n",
677 ret = posix_memalign(&bufs, getpagesize(), /* PAGE_SIZE aligned */
678 NUM_FRAMES * opt_xsk_frame_size);
680 exit_with_error(ret);
682 /* Create sockets... */
683 umem = xsk_configure_umem(bufs, NUM_FRAMES * opt_xsk_frame_size);
684 xsks[num_socks++] = xsk_configure_socket(umem);
686 if (opt_bench == BENCH_TXONLY) {
689 for (i = 0; i < NUM_FRAMES; i++)
690 (void)gen_eth_frame(umem, i * opt_xsk_frame_size);
693 signal(SIGINT, int_exit);
694 signal(SIGTERM, int_exit);
695 signal(SIGABRT, int_exit);
697 setlocale(LC_ALL, "");
699 ret = pthread_create(&pt, NULL, poller, NULL);
701 exit_with_error(ret);
703 prev_time = get_nsecs();
705 if (opt_bench == BENCH_RXDROP)
707 else if (opt_bench == BENCH_TXONLY)