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>
46 #include <arpa/inet.h>
52 PINGPONG_RECV_WRID = 1,
53 PINGPONG_SEND_WRID = 2,
58 struct pingpong_context {
59 struct ibv_context *context;
60 struct ibv_comp_channel *channel;
69 struct ibv_port_attr portinfo;
72 struct pingpong_dest {
79 static int pp_connect_ctx(struct pingpong_context *ctx, int port, int my_psn,
80 enum ibv_mtu mtu, int sl,
81 struct pingpong_dest *dest, int sgid_idx)
83 struct ibv_qp_attr attr = {
84 .qp_state = IBV_QPS_RTR,
86 .dest_qp_num = dest->qpn,
97 if (dest->gid.global.interface_id) {
98 attr.ah_attr.is_global = 1;
99 attr.ah_attr.grh.hop_limit = 1;
100 attr.ah_attr.grh.dgid = dest->gid;
101 attr.ah_attr.grh.sgid_index = sgid_idx;
104 if (ibv_modify_qp(ctx->qp, &attr,
110 fprintf(stderr, "Failed to modify QP to RTR\n");
114 attr.qp_state = IBV_QPS_RTS;
115 attr.sq_psn = my_psn;
116 if (ibv_modify_qp(ctx->qp, &attr,
119 fprintf(stderr, "Failed to modify QP to RTS\n");
126 static struct pingpong_dest *pp_client_exch_dest(const char *servername, int port,
127 const struct pingpong_dest *my_dest)
129 struct addrinfo *res, *t;
130 struct addrinfo hints = {
131 .ai_family = AF_INET,
132 .ai_socktype = SOCK_STREAM
135 char msg[sizeof "0000:000000:000000:00000000000000000000000000000000"];
138 struct pingpong_dest *rem_dest = NULL;
141 if (asprintf(&service, "%d", port) < 0)
144 n = getaddrinfo(servername, service, &hints, &res);
147 fprintf(stderr, "%s for %s:%d\n", gai_strerror(n), servername, port);
152 for (t = res; t; t = t->ai_next) {
153 sockfd = socket(t->ai_family, t->ai_socktype, t->ai_protocol);
155 if (!connect(sockfd, t->ai_addr, t->ai_addrlen))
166 fprintf(stderr, "Couldn't connect to %s:%d\n", servername, port);
170 gid_to_wire_gid(&my_dest->gid, gid);
171 sprintf(msg, "%04x:%06x:%06x:%s", my_dest->lid, my_dest->qpn, my_dest->psn, gid);
172 if (write(sockfd, msg, sizeof msg) != sizeof msg) {
173 fprintf(stderr, "Couldn't send local address\n");
177 if (read(sockfd, msg, sizeof msg) != sizeof msg) {
178 perror("client read");
179 fprintf(stderr, "Couldn't read remote address\n");
183 write(sockfd, "done", sizeof "done");
185 rem_dest = malloc(sizeof *rem_dest);
189 sscanf(msg, "%x:%x:%x:%s", &rem_dest->lid, &rem_dest->qpn, &rem_dest->psn, gid);
190 wire_gid_to_gid(gid, &rem_dest->gid);
197 static struct pingpong_dest *pp_server_exch_dest(struct pingpong_context *ctx,
198 int ib_port, enum ibv_mtu mtu,
200 const struct pingpong_dest *my_dest,
203 struct addrinfo *res, *t;
204 struct addrinfo hints = {
205 .ai_flags = AI_PASSIVE,
206 .ai_family = AF_INET,
207 .ai_socktype = SOCK_STREAM
210 char msg[sizeof "0000:000000:000000:00000000000000000000000000000000"];
212 int sockfd = -1, connfd;
213 struct pingpong_dest *rem_dest = NULL;
216 if (asprintf(&service, "%d", port) < 0)
219 n = getaddrinfo(NULL, service, &hints, &res);
222 fprintf(stderr, "%s for port %d\n", gai_strerror(n), port);
227 for (t = res; t; t = t->ai_next) {
228 sockfd = socket(t->ai_family, t->ai_socktype, t->ai_protocol);
232 setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, &n, sizeof n);
234 if (!bind(sockfd, t->ai_addr, t->ai_addrlen))
245 fprintf(stderr, "Couldn't listen to port %d\n", port);
250 connfd = accept(sockfd, NULL, 0);
253 fprintf(stderr, "accept() failed\n");
257 n = read(connfd, msg, sizeof msg);
258 if (n != sizeof msg) {
259 perror("server read");
260 fprintf(stderr, "%d/%d: Couldn't read remote address\n", n, (int) sizeof msg);
264 rem_dest = malloc(sizeof *rem_dest);
268 sscanf(msg, "%x:%x:%x:%s", &rem_dest->lid, &rem_dest->qpn, &rem_dest->psn, gid);
269 wire_gid_to_gid(gid, &rem_dest->gid);
271 if (pp_connect_ctx(ctx, ib_port, my_dest->psn, mtu, sl, rem_dest, sgid_idx)) {
272 fprintf(stderr, "Couldn't connect to remote QP\n");
278 gid_to_wire_gid(&my_dest->gid, gid);
279 sprintf(msg, "%04x:%06x:%06x:%s", my_dest->lid, my_dest->qpn, my_dest->psn, gid);
280 if (write(connfd, msg, sizeof msg) != sizeof msg) {
281 fprintf(stderr, "Couldn't send local address\n");
287 read(connfd, msg, sizeof msg);
294 static struct pingpong_context *pp_init_ctx(struct ibv_device *ib_dev, int size,
295 int rx_depth, int port,
298 struct pingpong_context *ctx;
300 ctx = calloc(1, sizeof *ctx);
305 ctx->rx_depth = rx_depth;
307 ctx->buf = malloc(roundup(size, page_size));
309 fprintf(stderr, "Couldn't allocate work buf.\n");
313 memset(ctx->buf, 0, size);
315 ctx->context = ibv_open_device(ib_dev);
317 fprintf(stderr, "Couldn't get context for %s\n",
318 ibv_get_device_name(ib_dev));
323 ctx->channel = ibv_create_comp_channel(ctx->context);
325 fprintf(stderr, "Couldn't create completion channel\n");
331 ctx->pd = ibv_alloc_pd(ctx->context);
333 fprintf(stderr, "Couldn't allocate PD\n");
337 ctx->mr = ibv_reg_mr(ctx->pd, ctx->buf, size, IBV_ACCESS_LOCAL_WRITE);
339 fprintf(stderr, "Couldn't register MR\n");
343 ctx->cq = ibv_create_cq(ctx->context, rx_depth + 1, NULL,
346 fprintf(stderr, "Couldn't create CQ\n");
351 struct ibv_qp_init_attr attr = {
356 .max_recv_wr = rx_depth,
360 .qp_type = IBV_QPT_UC
363 ctx->qp = ibv_create_qp(ctx->pd, &attr);
365 fprintf(stderr, "Couldn't create QP\n");
371 struct ibv_qp_attr attr = {
372 .qp_state = IBV_QPS_INIT,
378 if (ibv_modify_qp(ctx->qp, &attr,
382 IBV_QP_ACCESS_FLAGS)) {
383 fprintf(stderr, "Failed to modify QP to INIT\n");
391 int pp_close_ctx(struct pingpong_context *ctx)
393 if (ibv_destroy_qp(ctx->qp)) {
394 fprintf(stderr, "Couldn't destroy QP\n");
398 if (ibv_destroy_cq(ctx->cq)) {
399 fprintf(stderr, "Couldn't destroy CQ\n");
403 if (ibv_dereg_mr(ctx->mr)) {
404 fprintf(stderr, "Couldn't deregister MR\n");
408 if (ibv_dealloc_pd(ctx->pd)) {
409 fprintf(stderr, "Couldn't deallocate PD\n");
414 if (ibv_destroy_comp_channel(ctx->channel)) {
415 fprintf(stderr, "Couldn't destroy completion channel\n");
420 if (ibv_close_device(ctx->context)) {
421 fprintf(stderr, "Couldn't release context\n");
431 static int pp_post_recv(struct pingpong_context *ctx, int n)
433 struct ibv_sge list = {
434 .addr = (uintptr_t) ctx->buf,
436 .lkey = ctx->mr->lkey
438 struct ibv_recv_wr wr = {
439 .wr_id = PINGPONG_RECV_WRID,
443 struct ibv_recv_wr *bad_wr;
446 for (i = 0; i < n; ++i)
447 if (ibv_post_recv(ctx->qp, &wr, &bad_wr))
453 static int pp_post_send(struct pingpong_context *ctx)
455 struct ibv_sge list = {
456 .addr = (uintptr_t) ctx->buf,
458 .lkey = ctx->mr->lkey
460 struct ibv_send_wr wr = {
461 .wr_id = PINGPONG_SEND_WRID,
464 .opcode = IBV_WR_SEND,
465 .send_flags = IBV_SEND_SIGNALED,
467 struct ibv_send_wr *bad_wr;
469 return ibv_post_send(ctx->qp, &wr, &bad_wr);
472 static void usage(const char *argv0)
475 printf(" %s start a server and wait for connection\n", argv0);
476 printf(" %s <host> connect to server at <host>\n", argv0);
478 printf("Options:\n");
479 printf(" -p, --port=<port> listen on/connect to port <port> (default 18515)\n");
480 printf(" -d, --ib-dev=<dev> use IB device <dev> (default first device found)\n");
481 printf(" -i, --ib-port=<port> use port <port> of IB device (default 1)\n");
482 printf(" -s, --size=<size> size of message to exchange (default 4096)\n");
483 printf(" -m, --mtu=<size> path MTU (default 1024)\n");
484 printf(" -r, --rx-depth=<dep> number of receives to post at a time (default 500)\n");
485 printf(" -n, --iters=<iters> number of exchanges (default 1000)\n");
486 printf(" -l, --sl=<sl> service level value\n");
487 printf(" -e, --events sleep on CQ events (default poll)\n");
488 printf(" -g, --gid-idx=<gid index> local port gid index\n");
491 int main(int argc, char *argv[])
493 struct ibv_device **dev_list;
494 struct ibv_device *ib_dev;
495 struct pingpong_context *ctx;
496 struct pingpong_dest my_dest;
497 struct pingpong_dest *rem_dest;
498 struct timeval start, end;
499 char *ib_devname = NULL;
500 char *servername = NULL;
504 enum ibv_mtu mtu = IBV_MTU_1024;
510 int num_cq_events = 0;
515 srand48(getpid() * time(NULL));
520 static struct option long_options[] = {
521 { .name = "port", .has_arg = 1, .val = 'p' },
522 { .name = "ib-dev", .has_arg = 1, .val = 'd' },
523 { .name = "ib-port", .has_arg = 1, .val = 'i' },
524 { .name = "size", .has_arg = 1, .val = 's' },
525 { .name = "mtu", .has_arg = 1, .val = 'm' },
526 { .name = "rx-depth", .has_arg = 1, .val = 'r' },
527 { .name = "iters", .has_arg = 1, .val = 'n' },
528 { .name = "sl", .has_arg = 1, .val = 'l' },
529 { .name = "events", .has_arg = 0, .val = 'e' },
530 { .name = "gid-idx", .has_arg = 1, .val = 'g' },
534 c = getopt_long(argc, argv, "p:d:i:s:m:r:n:l:eg:", long_options, NULL);
540 port = strtol(optarg, NULL, 0);
541 if (port < 0 || port > 65535) {
548 ib_devname = strdup(optarg);
552 ib_port = strtol(optarg, NULL, 0);
560 size = strtol(optarg, NULL, 0);
564 mtu = pp_mtu_to_enum(strtol(optarg, NULL, 0));
572 rx_depth = strtol(optarg, NULL, 0);
576 iters = strtol(optarg, NULL, 0);
580 sl = strtol(optarg, NULL, 0);
588 gidx = strtol(optarg, NULL, 0);
597 if (optind == argc - 1)
598 servername = strdup(argv[optind]);
599 else if (optind < argc) {
604 page_size = sysconf(_SC_PAGESIZE);
606 dev_list = ibv_get_device_list(NULL);
608 perror("Failed to get IB devices list");
615 fprintf(stderr, "No IB devices found\n");
620 for (i = 0; dev_list[i]; ++i)
621 if (!strcmp(ibv_get_device_name(dev_list[i]), ib_devname))
623 ib_dev = dev_list[i];
625 fprintf(stderr, "IB device %s not found\n", ib_devname);
630 ctx = pp_init_ctx(ib_dev, size, rx_depth, ib_port, use_event);
634 routs = pp_post_recv(ctx, ctx->rx_depth);
635 if (routs < ctx->rx_depth) {
636 fprintf(stderr, "Couldn't post receive (%d)\n", routs);
641 if (ibv_req_notify_cq(ctx->cq, 0)) {
642 fprintf(stderr, "Couldn't request CQ notification\n");
646 if (pp_get_port_info(ctx->context, ib_port, &ctx->portinfo)) {
647 fprintf(stderr, "Couldn't get port info\n");
651 my_dest.lid = ctx->portinfo.lid;
652 if (ctx->portinfo.link_layer == IBV_LINK_LAYER_INFINIBAND && !my_dest.lid) {
653 fprintf(stderr, "Couldn't get local LID\n");
658 if (ibv_query_gid(ctx->context, ib_port, gidx, &my_dest.gid)) {
659 fprintf(stderr, "Could not get local gid for gid index %d\n", gidx);
663 memset(&my_dest.gid, 0, sizeof my_dest.gid);
665 my_dest.qpn = ctx->qp->qp_num;
666 my_dest.psn = lrand48() & 0xffffff;
667 inet_ntop(AF_INET6, &my_dest.gid, gid, sizeof gid);
668 printf(" local address: LID 0x%04x, QPN 0x%06x, PSN 0x%06x, GID %s\n",
669 my_dest.lid, my_dest.qpn, my_dest.psn, gid);
672 rem_dest = pp_client_exch_dest(servername, port, &my_dest);
674 rem_dest = pp_server_exch_dest(ctx, ib_port, mtu, port, sl, &my_dest, gidx);
679 inet_ntop(AF_INET6, &rem_dest->gid, gid, sizeof gid);
680 printf(" remote address: LID 0x%04x, QPN 0x%06x, PSN 0x%06x, GID %s\n",
681 rem_dest->lid, rem_dest->qpn, rem_dest->psn, gid);
684 if (pp_connect_ctx(ctx, ib_port, my_dest.psn, mtu, sl, rem_dest, gidx))
687 ctx->pending = PINGPONG_RECV_WRID;
690 if (pp_post_send(ctx)) {
691 fprintf(stderr, "Couldn't post send\n");
694 ctx->pending |= PINGPONG_SEND_WRID;
697 if (gettimeofday(&start, NULL)) {
698 perror("gettimeofday");
703 while (rcnt < iters || scnt < iters) {
705 struct ibv_cq *ev_cq;
708 if (ibv_get_cq_event(ctx->channel, &ev_cq, &ev_ctx)) {
709 fprintf(stderr, "Failed to get cq_event\n");
715 if (ev_cq != ctx->cq) {
716 fprintf(stderr, "CQ event for unknown CQ %p\n", ev_cq);
720 if (ibv_req_notify_cq(ctx->cq, 0)) {
721 fprintf(stderr, "Couldn't request CQ notification\n");
731 ne = ibv_poll_cq(ctx->cq, 2, wc);
733 fprintf(stderr, "poll CQ failed %d\n", ne);
736 } while (!use_event && ne < 1);
738 for (i = 0; i < ne; ++i) {
739 if (wc[i].status != IBV_WC_SUCCESS) {
740 fprintf(stderr, "Failed status %s (%d) for wr_id %d\n",
741 ibv_wc_status_str(wc[i].status),
742 wc[i].status, (int) wc[i].wr_id);
746 switch ((int) wc[i].wr_id) {
747 case PINGPONG_SEND_WRID:
751 case PINGPONG_RECV_WRID:
753 routs += pp_post_recv(ctx, ctx->rx_depth - routs);
754 if (routs < ctx->rx_depth) {
756 "Couldn't post receive (%d)\n",
766 fprintf(stderr, "Completion for unknown wr_id %d\n",
771 ctx->pending &= ~(int) wc[i].wr_id;
772 if (scnt < iters && !ctx->pending) {
773 if (pp_post_send(ctx)) {
774 fprintf(stderr, "Couldn't post send\n");
777 ctx->pending = PINGPONG_RECV_WRID |
784 if (gettimeofday(&end, NULL)) {
785 perror("gettimeofday");
790 float usec = (end.tv_sec - start.tv_sec) * 1000000 +
791 (end.tv_usec - start.tv_usec);
792 long long bytes = (long long) size * iters * 2;
794 printf("%lld bytes in %.2f seconds = %.2f Mbit/sec\n",
795 bytes, usec / 1000000., bytes * 8. / usec);
796 printf("%d iters in %.2f seconds = %.2f usec/iter\n",
797 iters, usec / 1000000., usec / iters);
800 ibv_ack_cq_events(ctx->cq, num_cq_events);
802 if (pp_close_ctx(ctx))
805 ibv_free_device_list(dev_list);