2 * Copyright (c) 2005 Topspin Communications. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #endif /* HAVE_CONFIG_H */
41 #include <sys/types.h>
42 #include <sys/socket.h>
47 #include <arpa/inet.h>
53 PINGPONG_RECV_WRID = 1,
54 PINGPONG_SEND_WRID = 2,
59 struct pingpong_context {
60 struct ibv_context *context;
61 struct ibv_comp_channel *channel;
70 struct ibv_port_attr portinfo;
73 struct pingpong_dest {
80 static int pp_connect_ctx(struct pingpong_context *ctx, int port, int my_psn,
81 enum ibv_mtu mtu, int sl,
82 struct pingpong_dest *dest, int sgid_idx)
84 struct ibv_qp_attr attr = {
85 .qp_state = IBV_QPS_RTR,
87 .dest_qp_num = dest->qpn,
89 .max_dest_rd_atomic = 1,
100 if (dest->gid.global.interface_id) {
101 attr.ah_attr.is_global = 1;
102 attr.ah_attr.grh.hop_limit = 1;
103 attr.ah_attr.grh.dgid = dest->gid;
104 attr.ah_attr.grh.sgid_index = sgid_idx;
106 if (ibv_modify_qp(ctx->qp, &attr,
112 IBV_QP_MAX_DEST_RD_ATOMIC |
113 IBV_QP_MIN_RNR_TIMER)) {
114 fprintf(stderr, "Failed to modify QP to RTR\n");
118 attr.qp_state = IBV_QPS_RTS;
122 attr.sq_psn = my_psn;
123 attr.max_rd_atomic = 1;
124 if (ibv_modify_qp(ctx->qp, &attr,
130 IBV_QP_MAX_QP_RD_ATOMIC)) {
131 fprintf(stderr, "Failed to modify QP to RTS\n");
138 static struct pingpong_dest *pp_client_exch_dest(const char *servername, int port,
139 const struct pingpong_dest *my_dest)
141 struct addrinfo *res, *t;
142 struct addrinfo hints = {
143 .ai_family = AF_INET,
144 .ai_socktype = SOCK_STREAM
147 char msg[sizeof "0000:000000:000000:00000000000000000000000000000000"];
150 struct pingpong_dest *rem_dest = NULL;
153 if (asprintf(&service, "%d", port) < 0)
156 n = getaddrinfo(servername, service, &hints, &res);
159 fprintf(stderr, "%s for %s:%d\n", gai_strerror(n), servername, port);
164 for (t = res; t; t = t->ai_next) {
165 sockfd = socket(t->ai_family, t->ai_socktype, t->ai_protocol);
167 if (!connect(sockfd, t->ai_addr, t->ai_addrlen))
178 fprintf(stderr, "Couldn't connect to %s:%d\n", servername, port);
182 gid_to_wire_gid(&my_dest->gid, gid);
183 sprintf(msg, "%04x:%06x:%06x:%s", my_dest->lid, my_dest->qpn, my_dest->psn, gid);
184 if (write(sockfd, msg, sizeof msg) != sizeof msg) {
185 fprintf(stderr, "Couldn't send local address\n");
189 if (read(sockfd, msg, sizeof msg) != sizeof msg) {
190 perror("client read");
191 fprintf(stderr, "Couldn't read remote address\n");
195 write(sockfd, "done", sizeof "done");
197 rem_dest = malloc(sizeof *rem_dest);
201 sscanf(msg, "%x:%x:%x:%s", &rem_dest->lid, &rem_dest->qpn, &rem_dest->psn, gid);
202 wire_gid_to_gid(gid, &rem_dest->gid);
209 static struct pingpong_dest *pp_server_exch_dest(struct pingpong_context *ctx,
210 int ib_port, enum ibv_mtu mtu,
212 const struct pingpong_dest *my_dest,
215 struct addrinfo *res, *t;
216 struct addrinfo hints = {
217 .ai_flags = AI_PASSIVE,
218 .ai_family = AF_INET,
219 .ai_socktype = SOCK_STREAM
222 char msg[sizeof "0000:000000:000000:00000000000000000000000000000000"];
224 int sockfd = -1, connfd;
225 struct pingpong_dest *rem_dest = NULL;
228 if (asprintf(&service, "%d", port) < 0)
231 n = getaddrinfo(NULL, service, &hints, &res);
234 fprintf(stderr, "%s for port %d\n", gai_strerror(n), port);
239 for (t = res; t; t = t->ai_next) {
240 sockfd = socket(t->ai_family, t->ai_socktype, t->ai_protocol);
244 setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, &n, sizeof n);
246 if (!bind(sockfd, t->ai_addr, t->ai_addrlen))
257 fprintf(stderr, "Couldn't listen to port %d\n", port);
262 connfd = accept(sockfd, NULL, 0);
265 fprintf(stderr, "accept() failed\n");
269 n = read(connfd, msg, sizeof msg);
270 if (n != sizeof msg) {
271 perror("server read");
272 fprintf(stderr, "%d/%d: Couldn't read remote address\n", n, (int) sizeof msg);
276 rem_dest = malloc(sizeof *rem_dest);
280 sscanf(msg, "%x:%x:%x:%s", &rem_dest->lid, &rem_dest->qpn, &rem_dest->psn, gid);
281 wire_gid_to_gid(gid, &rem_dest->gid);
283 if (pp_connect_ctx(ctx, ib_port, my_dest->psn, mtu, sl, rem_dest, sgid_idx)) {
284 fprintf(stderr, "Couldn't connect to remote QP\n");
291 gid_to_wire_gid(&my_dest->gid, gid);
292 sprintf(msg, "%04x:%06x:%06x:%s", my_dest->lid, my_dest->qpn, my_dest->psn, gid);
293 if (write(connfd, msg, sizeof msg) != sizeof msg) {
294 fprintf(stderr, "Couldn't send local address\n");
300 read(connfd, msg, sizeof msg);
307 #include <sys/param.h>
309 static struct pingpong_context *pp_init_ctx(struct ibv_device *ib_dev, int size,
310 int rx_depth, int port,
311 int use_event, int is_server)
313 struct pingpong_context *ctx;
315 ctx = calloc(1, sizeof *ctx);
320 ctx->rx_depth = rx_depth;
322 ctx->buf = malloc(roundup(size, page_size));
324 fprintf(stderr, "Couldn't allocate work buf.\n");
328 memset(ctx->buf, 0x7b + is_server, size);
330 ctx->context = ibv_open_device(ib_dev);
332 fprintf(stderr, "Couldn't get context for %s\n",
333 ibv_get_device_name(ib_dev));
338 ctx->channel = ibv_create_comp_channel(ctx->context);
340 fprintf(stderr, "Couldn't create completion channel\n");
346 ctx->pd = ibv_alloc_pd(ctx->context);
348 fprintf(stderr, "Couldn't allocate PD\n");
352 ctx->mr = ibv_reg_mr(ctx->pd, ctx->buf, size, IBV_ACCESS_LOCAL_WRITE);
354 fprintf(stderr, "Couldn't register MR\n");
358 ctx->cq = ibv_create_cq(ctx->context, rx_depth + 1, NULL,
361 fprintf(stderr, "Couldn't create CQ\n");
366 struct ibv_qp_init_attr attr = {
371 .max_recv_wr = rx_depth,
375 .qp_type = IBV_QPT_RC
378 ctx->qp = ibv_create_qp(ctx->pd, &attr);
380 fprintf(stderr, "Couldn't create QP\n");
386 struct ibv_qp_attr attr = {
387 .qp_state = IBV_QPS_INIT,
393 if (ibv_modify_qp(ctx->qp, &attr,
397 IBV_QP_ACCESS_FLAGS)) {
398 fprintf(stderr, "Failed to modify QP to INIT\n");
406 int pp_close_ctx(struct pingpong_context *ctx)
408 if (ibv_destroy_qp(ctx->qp)) {
409 fprintf(stderr, "Couldn't destroy QP\n");
413 if (ibv_destroy_cq(ctx->cq)) {
414 fprintf(stderr, "Couldn't destroy CQ\n");
418 if (ibv_dereg_mr(ctx->mr)) {
419 fprintf(stderr, "Couldn't deregister MR\n");
423 if (ibv_dealloc_pd(ctx->pd)) {
424 fprintf(stderr, "Couldn't deallocate PD\n");
429 if (ibv_destroy_comp_channel(ctx->channel)) {
430 fprintf(stderr, "Couldn't destroy completion channel\n");
435 if (ibv_close_device(ctx->context)) {
436 fprintf(stderr, "Couldn't release context\n");
446 static int pp_post_recv(struct pingpong_context *ctx, int n)
448 struct ibv_sge list = {
449 .addr = (uintptr_t) ctx->buf,
451 .lkey = ctx->mr->lkey
453 struct ibv_recv_wr wr = {
454 .wr_id = PINGPONG_RECV_WRID,
458 struct ibv_recv_wr *bad_wr;
461 for (i = 0; i < n; ++i)
462 if (ibv_post_recv(ctx->qp, &wr, &bad_wr))
468 static int pp_post_send(struct pingpong_context *ctx)
470 struct ibv_sge list = {
471 .addr = (uintptr_t) ctx->buf,
473 .lkey = ctx->mr->lkey
475 struct ibv_send_wr wr = {
476 .wr_id = PINGPONG_SEND_WRID,
479 .opcode = IBV_WR_SEND,
480 .send_flags = IBV_SEND_SIGNALED,
482 struct ibv_send_wr *bad_wr;
484 return ibv_post_send(ctx->qp, &wr, &bad_wr);
487 static void usage(const char *argv0)
490 printf(" %s start a server and wait for connection\n", argv0);
491 printf(" %s <host> connect to server at <host>\n", argv0);
493 printf("Options:\n");
494 printf(" -p, --port=<port> listen on/connect to port <port> (default 18515)\n");
495 printf(" -d, --ib-dev=<dev> use IB device <dev> (default first device found)\n");
496 printf(" -i, --ib-port=<port> use port <port> of IB device (default 1)\n");
497 printf(" -s, --size=<size> size of message to exchange (default 4096)\n");
498 printf(" -m, --mtu=<size> path MTU (default 1024)\n");
499 printf(" -r, --rx-depth=<dep> number of receives to post at a time (default 500)\n");
500 printf(" -n, --iters=<iters> number of exchanges (default 1000)\n");
501 printf(" -l, --sl=<sl> service level value\n");
502 printf(" -e, --events sleep on CQ events (default poll)\n");
503 printf(" -g, --gid-idx=<gid index> local port gid index\n");
506 int main(int argc, char *argv[])
508 struct ibv_device **dev_list;
509 struct ibv_device *ib_dev;
510 struct pingpong_context *ctx;
511 struct pingpong_dest my_dest;
512 struct pingpong_dest *rem_dest;
513 struct timeval start, end;
514 char *ib_devname = NULL;
515 char *servername = NULL;
519 enum ibv_mtu mtu = IBV_MTU_1024;
525 int num_cq_events = 0;
530 srand48(getpid() * time(NULL));
535 static struct option long_options[] = {
536 { .name = "port", .has_arg = 1, .val = 'p' },
537 { .name = "ib-dev", .has_arg = 1, .val = 'd' },
538 { .name = "ib-port", .has_arg = 1, .val = 'i' },
539 { .name = "size", .has_arg = 1, .val = 's' },
540 { .name = "mtu", .has_arg = 1, .val = 'm' },
541 { .name = "rx-depth", .has_arg = 1, .val = 'r' },
542 { .name = "iters", .has_arg = 1, .val = 'n' },
543 { .name = "sl", .has_arg = 1, .val = 'l' },
544 { .name = "events", .has_arg = 0, .val = 'e' },
545 { .name = "gid-idx", .has_arg = 1, .val = 'g' },
549 c = getopt_long(argc, argv, "p:d:i:s:m:r:n:l:eg:", long_options, NULL);
555 port = strtol(optarg, NULL, 0);
556 if (port < 0 || port > 65535) {
563 ib_devname = strdup(optarg);
567 ib_port = strtol(optarg, NULL, 0);
575 size = strtol(optarg, NULL, 0);
579 mtu = pp_mtu_to_enum(strtol(optarg, NULL, 0));
587 rx_depth = strtol(optarg, NULL, 0);
591 iters = strtol(optarg, NULL, 0);
595 sl = strtol(optarg, NULL, 0);
603 gidx = strtol(optarg, NULL, 0);
612 if (optind == argc - 1)
613 servername = strdup(argv[optind]);
614 else if (optind < argc) {
619 page_size = sysconf(_SC_PAGESIZE);
621 dev_list = ibv_get_device_list(NULL);
623 perror("Failed to get IB devices list");
630 fprintf(stderr, "No IB devices found\n");
635 for (i = 0; dev_list[i]; ++i)
636 if (!strcmp(ibv_get_device_name(dev_list[i]), ib_devname))
638 ib_dev = dev_list[i];
640 fprintf(stderr, "IB device %s not found\n", ib_devname);
645 ctx = pp_init_ctx(ib_dev, size, rx_depth, ib_port, use_event, !servername);
649 routs = pp_post_recv(ctx, ctx->rx_depth);
650 if (routs < ctx->rx_depth) {
651 fprintf(stderr, "Couldn't post receive (%d)\n", routs);
656 if (ibv_req_notify_cq(ctx->cq, 0)) {
657 fprintf(stderr, "Couldn't request CQ notification\n");
662 if (pp_get_port_info(ctx->context, ib_port, &ctx->portinfo)) {
663 fprintf(stderr, "Couldn't get port info\n");
667 my_dest.lid = ctx->portinfo.lid;
668 if (ctx->portinfo.link_layer == IBV_LINK_LAYER_INFINIBAND && !my_dest.lid) {
669 fprintf(stderr, "Couldn't get local LID\n");
674 if (ibv_query_gid(ctx->context, ib_port, gidx, &my_dest.gid)) {
675 fprintf(stderr, "Could not get local gid for gid index %d\n", gidx);
679 memset(&my_dest.gid, 0, sizeof my_dest.gid);
681 my_dest.qpn = ctx->qp->qp_num;
682 my_dest.psn = lrand48() & 0xffffff;
683 inet_ntop(AF_INET6, &my_dest.gid, gid, sizeof gid);
684 printf(" local address: LID 0x%04x, QPN 0x%06x, PSN 0x%06x, GID %s\n",
685 my_dest.lid, my_dest.qpn, my_dest.psn, gid);
689 rem_dest = pp_client_exch_dest(servername, port, &my_dest);
691 rem_dest = pp_server_exch_dest(ctx, ib_port, mtu, port, sl, &my_dest, gidx);
696 inet_ntop(AF_INET6, &rem_dest->gid, gid, sizeof gid);
697 printf(" remote address: LID 0x%04x, QPN 0x%06x, PSN 0x%06x, GID %s\n",
698 rem_dest->lid, rem_dest->qpn, rem_dest->psn, gid);
701 if (pp_connect_ctx(ctx, ib_port, my_dest.psn, mtu, sl, rem_dest, gidx))
704 ctx->pending = PINGPONG_RECV_WRID;
707 if (pp_post_send(ctx)) {
708 fprintf(stderr, "Couldn't post send\n");
711 ctx->pending |= PINGPONG_SEND_WRID;
714 if (gettimeofday(&start, NULL)) {
715 perror("gettimeofday");
720 while (rcnt < iters || scnt < iters) {
722 struct ibv_cq *ev_cq;
725 if (ibv_get_cq_event(ctx->channel, &ev_cq, &ev_ctx)) {
726 fprintf(stderr, "Failed to get cq_event\n");
732 if (ev_cq != ctx->cq) {
733 fprintf(stderr, "CQ event for unknown CQ %p\n", ev_cq);
737 if (ibv_req_notify_cq(ctx->cq, 0)) {
738 fprintf(stderr, "Couldn't request CQ notification\n");
748 ne = ibv_poll_cq(ctx->cq, 2, wc);
750 fprintf(stderr, "poll CQ failed %d\n", ne);
754 } while (!use_event && ne < 1);
756 for (i = 0; i < ne; ++i) {
757 if (wc[i].status != IBV_WC_SUCCESS) {
758 fprintf(stderr, "Failed status %s (%d) for wr_id %d\n",
759 ibv_wc_status_str(wc[i].status),
760 wc[i].status, (int) wc[i].wr_id);
764 switch ((int) wc[i].wr_id) {
765 case PINGPONG_SEND_WRID:
769 case PINGPONG_RECV_WRID:
771 routs += pp_post_recv(ctx, ctx->rx_depth - routs);
772 if (routs < ctx->rx_depth) {
774 "Couldn't post receive (%d)\n",
784 fprintf(stderr, "Completion for unknown wr_id %d\n",
789 ctx->pending &= ~(int) wc[i].wr_id;
790 if (scnt < iters && !ctx->pending) {
791 if (pp_post_send(ctx)) {
792 fprintf(stderr, "Couldn't post send\n");
795 ctx->pending = PINGPONG_RECV_WRID |
802 if (gettimeofday(&end, NULL)) {
803 perror("gettimeofday");
808 float usec = (end.tv_sec - start.tv_sec) * 1000000 +
809 (end.tv_usec - start.tv_usec);
810 long long bytes = (long long) size * iters * 2;
812 printf("%lld bytes in %.2f seconds = %.2f Mbit/sec\n",
813 bytes, usec / 1000000., bytes * 8. / usec);
814 printf("%d iters in %.2f seconds = %.2f usec/iter\n",
815 iters, usec / 1000000., usec / iters);
818 ibv_ack_cq_events(ctx->cq, num_cq_events);
820 if (pp_close_ctx(ctx))
823 ibv_free_device_list(dev_list);