1219820Sjeff/*
2219820Sjeff * Copyright (c) 2005 Topspin Communications.  All rights reserved.
3219820Sjeff *
4219820Sjeff * This software is available to you under a choice of one of two
5219820Sjeff * licenses.  You may choose to be licensed under the terms of the GNU
6219820Sjeff * General Public License (GPL) Version 2, available from the file
7219820Sjeff * COPYING in the main directory of this source tree, or the
8219820Sjeff * OpenIB.org BSD license below:
9219820Sjeff *
10219820Sjeff *     Redistribution and use in source and binary forms, with or
11219820Sjeff *     without modification, are permitted provided that the following
12219820Sjeff *     conditions are met:
13219820Sjeff *
14219820Sjeff *      - Redistributions of source code must retain the above
15219820Sjeff *        copyright notice, this list of conditions and the following
16219820Sjeff *        disclaimer.
17219820Sjeff *
18219820Sjeff *      - Redistributions in binary form must reproduce the above
19219820Sjeff *        copyright notice, this list of conditions and the following
20219820Sjeff *        disclaimer in the documentation and/or other materials
21219820Sjeff *        provided with the distribution.
22219820Sjeff *
23219820Sjeff * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24219820Sjeff * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25219820Sjeff * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26219820Sjeff * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27219820Sjeff * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28219820Sjeff * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29219820Sjeff * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30219820Sjeff * SOFTWARE.
31219820Sjeff */
32331769Shselasky#define _GNU_SOURCE
33331769Shselasky#include <config.h>
34219820Sjeff
35219820Sjeff#include <stdio.h>
36219820Sjeff#include <stdlib.h>
37219820Sjeff#include <unistd.h>
38219820Sjeff#include <string.h>
39219820Sjeff#include <sys/types.h>
40219820Sjeff#include <sys/socket.h>
41219820Sjeff#include <sys/time.h>
42219820Sjeff#include <netdb.h>
43331769Shselasky#include <stdlib.h>
44219820Sjeff#include <getopt.h>
45219820Sjeff#include <arpa/inet.h>
46219820Sjeff#include <time.h>
47219820Sjeff
48219820Sjeff#include "pingpong.h"
49219820Sjeff
50219820Sjeffenum {
51219820Sjeff	PINGPONG_RECV_WRID = 1,
52219820Sjeff	PINGPONG_SEND_WRID = 2,
53219820Sjeff};
54219820Sjeff
55219820Sjeffstatic int page_size;
56219820Sjeff
57219820Sjeffstruct pingpong_context {
58219820Sjeff	struct ibv_context	*context;
59219820Sjeff	struct ibv_comp_channel *channel;
60219820Sjeff	struct ibv_pd		*pd;
61219820Sjeff	struct ibv_mr		*mr;
62219820Sjeff	struct ibv_cq		*cq;
63219820Sjeff	struct ibv_qp		*qp;
64219820Sjeff	void			*buf;
65219820Sjeff	int			 size;
66331769Shselasky	int			 send_flags;
67219820Sjeff	int			 rx_depth;
68219820Sjeff	int			 pending;
69331769Shselasky	struct ibv_port_attr     portinfo;
70219820Sjeff};
71219820Sjeff
72219820Sjeffstruct pingpong_dest {
73219820Sjeff	int lid;
74219820Sjeff	int qpn;
75219820Sjeff	int psn;
76219820Sjeff	union ibv_gid gid;
77219820Sjeff};
78219820Sjeff
79219820Sjeffstatic int pp_connect_ctx(struct pingpong_context *ctx, int port, int my_psn,
80219820Sjeff			  enum ibv_mtu mtu, int sl,
81219820Sjeff			  struct pingpong_dest *dest, int sgid_idx)
82219820Sjeff{
83219820Sjeff	struct ibv_qp_attr attr = {
84219820Sjeff		.qp_state		= IBV_QPS_RTR,
85219820Sjeff		.path_mtu		= mtu,
86219820Sjeff		.dest_qp_num		= dest->qpn,
87219820Sjeff		.rq_psn		= dest->psn,
88219820Sjeff		.ah_attr		= {
89219820Sjeff			.is_global	= 0,
90219820Sjeff			.dlid		= dest->lid,
91219820Sjeff			.sl		= sl,
92219820Sjeff			.src_path_bits	= 0,
93219820Sjeff			.port_num	= port
94219820Sjeff		}
95219820Sjeff	};
96219820Sjeff
97219820Sjeff	if (dest->gid.global.interface_id) {
98219820Sjeff		attr.ah_attr.is_global = 1;
99219820Sjeff		attr.ah_attr.grh.hop_limit = 1;
100219820Sjeff		attr.ah_attr.grh.dgid = dest->gid;
101219820Sjeff		attr.ah_attr.grh.sgid_index = sgid_idx;
102219820Sjeff	}
103219820Sjeff	if (ibv_modify_qp(ctx->qp, &attr,
104219820Sjeff			  IBV_QP_STATE              |
105219820Sjeff			  IBV_QP_AV                 |
106219820Sjeff			  IBV_QP_PATH_MTU           |
107219820Sjeff			  IBV_QP_DEST_QPN           |
108219820Sjeff			  IBV_QP_RQ_PSN)) {
109219820Sjeff		fprintf(stderr, "Failed to modify QP to RTR\n");
110219820Sjeff		return 1;
111219820Sjeff	}
112219820Sjeff
113219820Sjeff	attr.qp_state	    = IBV_QPS_RTS;
114219820Sjeff	attr.sq_psn	    = my_psn;
115219820Sjeff	if (ibv_modify_qp(ctx->qp, &attr,
116219820Sjeff			  IBV_QP_STATE              |
117219820Sjeff			  IBV_QP_SQ_PSN)) {
118219820Sjeff		fprintf(stderr, "Failed to modify QP to RTS\n");
119219820Sjeff		return 1;
120219820Sjeff	}
121219820Sjeff
122219820Sjeff	return 0;
123219820Sjeff}
124219820Sjeff
125219820Sjeffstatic struct pingpong_dest *pp_client_exch_dest(const char *servername, int port,
126219820Sjeff						 const struct pingpong_dest *my_dest)
127219820Sjeff{
128219820Sjeff	struct addrinfo *res, *t;
129219820Sjeff	struct addrinfo hints = {
130337119Shselasky		.ai_family   = AF_UNSPEC,
131219820Sjeff		.ai_socktype = SOCK_STREAM
132219820Sjeff	};
133219820Sjeff	char *service;
134219820Sjeff	char msg[sizeof "0000:000000:000000:00000000000000000000000000000000"];
135219820Sjeff	int n;
136219820Sjeff	int sockfd = -1;
137219820Sjeff	struct pingpong_dest *rem_dest = NULL;
138219820Sjeff	char gid[33];
139219820Sjeff
140219820Sjeff	if (asprintf(&service, "%d", port) < 0)
141219820Sjeff		return NULL;
142219820Sjeff
143219820Sjeff	n = getaddrinfo(servername, service, &hints, &res);
144219820Sjeff
145219820Sjeff	if (n < 0) {
146219820Sjeff		fprintf(stderr, "%s for %s:%d\n", gai_strerror(n), servername, port);
147219820Sjeff		free(service);
148219820Sjeff		return NULL;
149219820Sjeff	}
150219820Sjeff
151219820Sjeff	for (t = res; t; t = t->ai_next) {
152219820Sjeff		sockfd = socket(t->ai_family, t->ai_socktype, t->ai_protocol);
153219820Sjeff		if (sockfd >= 0) {
154219820Sjeff			if (!connect(sockfd, t->ai_addr, t->ai_addrlen))
155219820Sjeff				break;
156219820Sjeff			close(sockfd);
157219820Sjeff			sockfd = -1;
158219820Sjeff		}
159219820Sjeff	}
160219820Sjeff
161331769Shselasky	freeaddrinfo_null(res);
162219820Sjeff	free(service);
163219820Sjeff
164219820Sjeff	if (sockfd < 0) {
165219820Sjeff		fprintf(stderr, "Couldn't connect to %s:%d\n", servername, port);
166219820Sjeff		return NULL;
167219820Sjeff	}
168219820Sjeff
169219820Sjeff	gid_to_wire_gid(&my_dest->gid, gid);
170331769Shselasky	sprintf(msg, "%04x:%06x:%06x:%s", my_dest->lid, my_dest->qpn,
171331769Shselasky							my_dest->psn, gid);
172219820Sjeff	if (write(sockfd, msg, sizeof msg) != sizeof msg) {
173219820Sjeff		fprintf(stderr, "Couldn't send local address\n");
174219820Sjeff		goto out;
175219820Sjeff	}
176219820Sjeff
177331769Shselasky	if (read(sockfd, msg, sizeof msg) != sizeof msg ||
178331769Shselasky	    write(sockfd, "done", sizeof "done") != sizeof "done") {
179331769Shselasky		perror("client read/write");
180331769Shselasky		fprintf(stderr, "Couldn't read/write remote address\n");
181219820Sjeff		goto out;
182219820Sjeff	}
183219820Sjeff
184219820Sjeff
185219820Sjeff	rem_dest = malloc(sizeof *rem_dest);
186219820Sjeff	if (!rem_dest)
187219820Sjeff		goto out;
188219820Sjeff
189331769Shselasky	sscanf(msg, "%x:%x:%x:%s", &rem_dest->lid, &rem_dest->qpn,
190331769Shselasky						&rem_dest->psn, gid);
191219820Sjeff	wire_gid_to_gid(gid, &rem_dest->gid);
192219820Sjeff
193219820Sjeffout:
194219820Sjeff	close(sockfd);
195219820Sjeff	return rem_dest;
196219820Sjeff}
197219820Sjeff
198219820Sjeffstatic struct pingpong_dest *pp_server_exch_dest(struct pingpong_context *ctx,
199219820Sjeff						 int ib_port, enum ibv_mtu mtu,
200219820Sjeff						 int port, int sl,
201219820Sjeff						 const struct pingpong_dest *my_dest,
202219820Sjeff						 int sgid_idx)
203219820Sjeff{
204219820Sjeff	struct addrinfo *res, *t;
205219820Sjeff	struct addrinfo hints = {
206219820Sjeff		.ai_flags    = AI_PASSIVE,
207219820Sjeff		.ai_family   = AF_INET,
208219820Sjeff		.ai_socktype = SOCK_STREAM
209219820Sjeff	};
210219820Sjeff	char *service;
211219820Sjeff	char msg[sizeof "0000:000000:000000:00000000000000000000000000000000"];
212219820Sjeff	int n;
213219820Sjeff	int sockfd = -1, connfd;
214219820Sjeff	struct pingpong_dest *rem_dest = NULL;
215219820Sjeff	char gid[33];
216219820Sjeff
217219820Sjeff	if (asprintf(&service, "%d", port) < 0)
218219820Sjeff		return NULL;
219219820Sjeff
220219820Sjeff	n = getaddrinfo(NULL, service, &hints, &res);
221219820Sjeff
222219820Sjeff	if (n < 0) {
223219820Sjeff		fprintf(stderr, "%s for port %d\n", gai_strerror(n), port);
224219820Sjeff		free(service);
225219820Sjeff		return NULL;
226219820Sjeff	}
227219820Sjeff
228219820Sjeff	for (t = res; t; t = t->ai_next) {
229219820Sjeff		sockfd = socket(t->ai_family, t->ai_socktype, t->ai_protocol);
230219820Sjeff		if (sockfd >= 0) {
231219820Sjeff			n = 1;
232219820Sjeff
233219820Sjeff			setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, &n, sizeof n);
234219820Sjeff
235219820Sjeff			if (!bind(sockfd, t->ai_addr, t->ai_addrlen))
236219820Sjeff				break;
237219820Sjeff			close(sockfd);
238219820Sjeff			sockfd = -1;
239219820Sjeff		}
240219820Sjeff	}
241219820Sjeff
242331769Shselasky	freeaddrinfo_null(res);
243219820Sjeff	free(service);
244219820Sjeff
245219820Sjeff	if (sockfd < 0) {
246219820Sjeff		fprintf(stderr, "Couldn't listen to port %d\n", port);
247219820Sjeff		return NULL;
248219820Sjeff	}
249219820Sjeff
250219820Sjeff	listen(sockfd, 1);
251331769Shselasky	connfd = accept(sockfd, NULL, NULL);
252219820Sjeff	close(sockfd);
253219820Sjeff	if (connfd < 0) {
254219820Sjeff		fprintf(stderr, "accept() failed\n");
255219820Sjeff		return NULL;
256219820Sjeff	}
257219820Sjeff
258219820Sjeff	n = read(connfd, msg, sizeof msg);
259219820Sjeff	if (n != sizeof msg) {
260219820Sjeff		perror("server read");
261219820Sjeff		fprintf(stderr, "%d/%d: Couldn't read remote address\n", n, (int) sizeof msg);
262219820Sjeff		goto out;
263219820Sjeff	}
264219820Sjeff
265219820Sjeff	rem_dest = malloc(sizeof *rem_dest);
266219820Sjeff	if (!rem_dest)
267219820Sjeff		goto out;
268219820Sjeff
269331769Shselasky	sscanf(msg, "%x:%x:%x:%s", &rem_dest->lid, &rem_dest->qpn,
270331769Shselasky							&rem_dest->psn, gid);
271219820Sjeff	wire_gid_to_gid(gid, &rem_dest->gid);
272219820Sjeff
273331769Shselasky	if (pp_connect_ctx(ctx, ib_port, my_dest->psn, mtu, sl, rem_dest,
274331769Shselasky								sgid_idx)) {
275219820Sjeff		fprintf(stderr, "Couldn't connect to remote QP\n");
276219820Sjeff		free(rem_dest);
277219820Sjeff		rem_dest = NULL;
278219820Sjeff		goto out;
279219820Sjeff	}
280219820Sjeff
281331769Shselasky
282219820Sjeff	gid_to_wire_gid(&my_dest->gid, gid);
283331769Shselasky	sprintf(msg, "%04x:%06x:%06x:%s", my_dest->lid, my_dest->qpn,
284331769Shselasky							my_dest->psn, gid);
285331769Shselasky	if (write(connfd, msg, sizeof msg) != sizeof msg ||
286331769Shselasky	    read(connfd, msg, sizeof msg) != sizeof "done") {
287331769Shselasky		fprintf(stderr, "Couldn't send/recv local address\n");
288219820Sjeff		free(rem_dest);
289219820Sjeff		rem_dest = NULL;
290219820Sjeff		goto out;
291219820Sjeff	}
292219820Sjeff
293219820Sjeffout:
294219820Sjeff	close(connfd);
295219820Sjeff	return rem_dest;
296219820Sjeff}
297219820Sjeff
298219820Sjeffstatic struct pingpong_context *pp_init_ctx(struct ibv_device *ib_dev, int size,
299219820Sjeff					    int rx_depth, int port,
300219820Sjeff					    int use_event)
301219820Sjeff{
302219820Sjeff	struct pingpong_context *ctx;
303219820Sjeff
304219820Sjeff	ctx = calloc(1, sizeof *ctx);
305219820Sjeff	if (!ctx)
306219820Sjeff		return NULL;
307219820Sjeff
308331769Shselasky	ctx->size       = size;
309331769Shselasky	ctx->send_flags = IBV_SEND_SIGNALED;
310331769Shselasky	ctx->rx_depth   = rx_depth;
311219820Sjeff
312331769Shselasky	ctx->buf = memalign(page_size, size);
313219820Sjeff	if (!ctx->buf) {
314219820Sjeff		fprintf(stderr, "Couldn't allocate work buf.\n");
315331769Shselasky		goto clean_ctx;
316219820Sjeff	}
317219820Sjeff
318331769Shselasky	/* FIXME memset(ctx->buf, 0, size); */
319331769Shselasky	memset(ctx->buf, 0x7b, size);
320219820Sjeff
321219820Sjeff	ctx->context = ibv_open_device(ib_dev);
322219820Sjeff	if (!ctx->context) {
323219820Sjeff		fprintf(stderr, "Couldn't get context for %s\n",
324219820Sjeff			ibv_get_device_name(ib_dev));
325331769Shselasky		goto clean_buffer;
326219820Sjeff	}
327219820Sjeff
328219820Sjeff	if (use_event) {
329219820Sjeff		ctx->channel = ibv_create_comp_channel(ctx->context);
330219820Sjeff		if (!ctx->channel) {
331219820Sjeff			fprintf(stderr, "Couldn't create completion channel\n");
332331769Shselasky			goto clean_device;
333219820Sjeff		}
334219820Sjeff	} else
335219820Sjeff		ctx->channel = NULL;
336219820Sjeff
337219820Sjeff	ctx->pd = ibv_alloc_pd(ctx->context);
338219820Sjeff	if (!ctx->pd) {
339219820Sjeff		fprintf(stderr, "Couldn't allocate PD\n");
340331769Shselasky		goto clean_comp_channel;
341219820Sjeff	}
342219820Sjeff
343219820Sjeff	ctx->mr = ibv_reg_mr(ctx->pd, ctx->buf, size, IBV_ACCESS_LOCAL_WRITE);
344219820Sjeff	if (!ctx->mr) {
345219820Sjeff		fprintf(stderr, "Couldn't register MR\n");
346331769Shselasky		goto clean_pd;
347219820Sjeff	}
348219820Sjeff
349219820Sjeff	ctx->cq = ibv_create_cq(ctx->context, rx_depth + 1, NULL,
350219820Sjeff				ctx->channel, 0);
351219820Sjeff	if (!ctx->cq) {
352219820Sjeff		fprintf(stderr, "Couldn't create CQ\n");
353331769Shselasky		goto clean_mr;
354219820Sjeff	}
355219820Sjeff
356219820Sjeff	{
357331769Shselasky		struct ibv_qp_attr attr;
358331769Shselasky		struct ibv_qp_init_attr init_attr = {
359219820Sjeff			.send_cq = ctx->cq,
360219820Sjeff			.recv_cq = ctx->cq,
361219820Sjeff			.cap     = {
362219820Sjeff				.max_send_wr  = 1,
363219820Sjeff				.max_recv_wr  = rx_depth,
364219820Sjeff				.max_send_sge = 1,
365219820Sjeff				.max_recv_sge = 1
366219820Sjeff			},
367219820Sjeff			.qp_type = IBV_QPT_UC
368219820Sjeff		};
369219820Sjeff
370331769Shselasky		ctx->qp = ibv_create_qp(ctx->pd, &init_attr);
371219820Sjeff		if (!ctx->qp)  {
372219820Sjeff			fprintf(stderr, "Couldn't create QP\n");
373331769Shselasky			goto clean_cq;
374219820Sjeff		}
375331769Shselasky		ibv_query_qp(ctx->qp, &attr, IBV_QP_CAP, &init_attr);
376331769Shselasky		if (init_attr.cap.max_inline_data >= size) {
377331769Shselasky			ctx->send_flags |= IBV_SEND_INLINE;
378331769Shselasky		}
379219820Sjeff	}
380219820Sjeff
381219820Sjeff	{
382219820Sjeff		struct ibv_qp_attr attr = {
383219820Sjeff			.qp_state        = IBV_QPS_INIT,
384219820Sjeff			.pkey_index      = 0,
385219820Sjeff			.port_num        = port,
386219820Sjeff			.qp_access_flags = 0
387219820Sjeff		};
388219820Sjeff
389219820Sjeff		if (ibv_modify_qp(ctx->qp, &attr,
390219820Sjeff				  IBV_QP_STATE              |
391219820Sjeff				  IBV_QP_PKEY_INDEX         |
392219820Sjeff				  IBV_QP_PORT               |
393219820Sjeff				  IBV_QP_ACCESS_FLAGS)) {
394219820Sjeff			fprintf(stderr, "Failed to modify QP to INIT\n");
395331769Shselasky			goto clean_qp;
396219820Sjeff		}
397219820Sjeff	}
398219820Sjeff
399219820Sjeff	return ctx;
400331769Shselasky
401331769Shselaskyclean_qp:
402331769Shselasky	ibv_destroy_qp(ctx->qp);
403331769Shselasky
404331769Shselaskyclean_cq:
405331769Shselasky	ibv_destroy_cq(ctx->cq);
406331769Shselasky
407331769Shselaskyclean_mr:
408331769Shselasky	ibv_dereg_mr(ctx->mr);
409331769Shselasky
410331769Shselaskyclean_pd:
411331769Shselasky	ibv_dealloc_pd(ctx->pd);
412331769Shselasky
413331769Shselaskyclean_comp_channel:
414331769Shselasky	if (ctx->channel)
415331769Shselasky		ibv_destroy_comp_channel(ctx->channel);
416331769Shselasky
417331769Shselaskyclean_device:
418331769Shselasky	ibv_close_device(ctx->context);
419331769Shselasky
420331769Shselaskyclean_buffer:
421331769Shselasky	free(ctx->buf);
422331769Shselasky
423331769Shselaskyclean_ctx:
424331769Shselasky	free(ctx);
425331769Shselasky
426331769Shselasky	return NULL;
427219820Sjeff}
428219820Sjeff
429331769Shselaskystatic int pp_close_ctx(struct pingpong_context *ctx)
430219820Sjeff{
431219820Sjeff	if (ibv_destroy_qp(ctx->qp)) {
432219820Sjeff		fprintf(stderr, "Couldn't destroy QP\n");
433219820Sjeff		return 1;
434219820Sjeff	}
435219820Sjeff
436219820Sjeff	if (ibv_destroy_cq(ctx->cq)) {
437219820Sjeff		fprintf(stderr, "Couldn't destroy CQ\n");
438219820Sjeff		return 1;
439219820Sjeff	}
440219820Sjeff
441219820Sjeff	if (ibv_dereg_mr(ctx->mr)) {
442219820Sjeff		fprintf(stderr, "Couldn't deregister MR\n");
443219820Sjeff		return 1;
444219820Sjeff	}
445219820Sjeff
446219820Sjeff	if (ibv_dealloc_pd(ctx->pd)) {
447219820Sjeff		fprintf(stderr, "Couldn't deallocate PD\n");
448219820Sjeff		return 1;
449219820Sjeff	}
450219820Sjeff
451219820Sjeff	if (ctx->channel) {
452219820Sjeff		if (ibv_destroy_comp_channel(ctx->channel)) {
453219820Sjeff			fprintf(stderr, "Couldn't destroy completion channel\n");
454219820Sjeff			return 1;
455219820Sjeff		}
456219820Sjeff	}
457219820Sjeff
458219820Sjeff	if (ibv_close_device(ctx->context)) {
459219820Sjeff		fprintf(stderr, "Couldn't release context\n");
460219820Sjeff		return 1;
461219820Sjeff	}
462219820Sjeff
463219820Sjeff	free(ctx->buf);
464219820Sjeff	free(ctx);
465219820Sjeff
466219820Sjeff	return 0;
467219820Sjeff}
468219820Sjeff
469219820Sjeffstatic int pp_post_recv(struct pingpong_context *ctx, int n)
470219820Sjeff{
471219820Sjeff	struct ibv_sge list = {
472219820Sjeff		.addr	= (uintptr_t) ctx->buf,
473219820Sjeff		.length = ctx->size,
474219820Sjeff		.lkey	= ctx->mr->lkey
475219820Sjeff	};
476219820Sjeff	struct ibv_recv_wr wr = {
477219820Sjeff		.wr_id	    = PINGPONG_RECV_WRID,
478219820Sjeff		.sg_list    = &list,
479219820Sjeff		.num_sge    = 1,
480219820Sjeff	};
481219820Sjeff	struct ibv_recv_wr *bad_wr;
482219820Sjeff	int i;
483219820Sjeff
484219820Sjeff	for (i = 0; i < n; ++i)
485219820Sjeff		if (ibv_post_recv(ctx->qp, &wr, &bad_wr))
486219820Sjeff			break;
487219820Sjeff
488219820Sjeff	return i;
489219820Sjeff}
490219820Sjeff
491219820Sjeffstatic int pp_post_send(struct pingpong_context *ctx)
492219820Sjeff{
493219820Sjeff	struct ibv_sge list = {
494219820Sjeff		.addr	= (uintptr_t) ctx->buf,
495219820Sjeff		.length = ctx->size,
496219820Sjeff		.lkey	= ctx->mr->lkey
497219820Sjeff	};
498219820Sjeff	struct ibv_send_wr wr = {
499219820Sjeff		.wr_id	    = PINGPONG_SEND_WRID,
500219820Sjeff		.sg_list    = &list,
501219820Sjeff		.num_sge    = 1,
502219820Sjeff		.opcode     = IBV_WR_SEND,
503331769Shselasky		.send_flags = ctx->send_flags,
504219820Sjeff	};
505219820Sjeff	struct ibv_send_wr *bad_wr;
506219820Sjeff
507219820Sjeff	return ibv_post_send(ctx->qp, &wr, &bad_wr);
508219820Sjeff}
509219820Sjeff
510219820Sjeffstatic void usage(const char *argv0)
511219820Sjeff{
512219820Sjeff	printf("Usage:\n");
513219820Sjeff	printf("  %s            start a server and wait for connection\n", argv0);
514219820Sjeff	printf("  %s <host>     connect to server at <host>\n", argv0);
515219820Sjeff	printf("\n");
516219820Sjeff	printf("Options:\n");
517219820Sjeff	printf("  -p, --port=<port>      listen on/connect to port <port> (default 18515)\n");
518219820Sjeff	printf("  -d, --ib-dev=<dev>     use IB device <dev> (default first device found)\n");
519219820Sjeff	printf("  -i, --ib-port=<port>   use port <port> of IB device (default 1)\n");
520219820Sjeff	printf("  -s, --size=<size>      size of message to exchange (default 4096)\n");
521219820Sjeff	printf("  -m, --mtu=<size>       path MTU (default 1024)\n");
522219820Sjeff	printf("  -r, --rx-depth=<dep>   number of receives to post at a time (default 500)\n");
523219820Sjeff	printf("  -n, --iters=<iters>    number of exchanges (default 1000)\n");
524219820Sjeff	printf("  -l, --sl=<sl>          service level value\n");
525219820Sjeff	printf("  -e, --events           sleep on CQ events (default poll)\n");
526219820Sjeff	printf("  -g, --gid-idx=<gid index> local port gid index\n");
527219820Sjeff}
528219820Sjeff
529219820Sjeffint main(int argc, char *argv[])
530219820Sjeff{
531219820Sjeff	struct ibv_device      **dev_list;
532219820Sjeff	struct ibv_device	*ib_dev;
533219820Sjeff	struct pingpong_context *ctx;
534219820Sjeff	struct pingpong_dest     my_dest;
535219820Sjeff	struct pingpong_dest    *rem_dest;
536219820Sjeff	struct timeval           start, end;
537219820Sjeff	char                    *ib_devname = NULL;
538219820Sjeff	char                    *servername = NULL;
539331769Shselasky	unsigned int             port = 18515;
540219820Sjeff	int                      ib_port = 1;
541331769Shselasky	unsigned int             size = 4096;
542219820Sjeff	enum ibv_mtu		 mtu = IBV_MTU_1024;
543331769Shselasky	unsigned int             rx_depth = 500;
544331769Shselasky	unsigned int             iters = 1000;
545219820Sjeff	int                      use_event = 0;
546219820Sjeff	int                      routs;
547219820Sjeff	int                      rcnt, scnt;
548219820Sjeff	int                      num_cq_events = 0;
549219820Sjeff	int                      sl = 0;
550219820Sjeff	int			 gidx = -1;
551219820Sjeff	char			 gid[33];
552219820Sjeff
553219820Sjeff	srand48(getpid() * time(NULL));
554219820Sjeff
555219820Sjeff	while (1) {
556219820Sjeff		int c;
557219820Sjeff
558219820Sjeff		static struct option long_options[] = {
559219820Sjeff			{ .name = "port",     .has_arg = 1, .val = 'p' },
560219820Sjeff			{ .name = "ib-dev",   .has_arg = 1, .val = 'd' },
561219820Sjeff			{ .name = "ib-port",  .has_arg = 1, .val = 'i' },
562219820Sjeff			{ .name = "size",     .has_arg = 1, .val = 's' },
563219820Sjeff			{ .name = "mtu",      .has_arg = 1, .val = 'm' },
564219820Sjeff			{ .name = "rx-depth", .has_arg = 1, .val = 'r' },
565219820Sjeff			{ .name = "iters",    .has_arg = 1, .val = 'n' },
566219820Sjeff			{ .name = "sl",       .has_arg = 1, .val = 'l' },
567219820Sjeff			{ .name = "events",   .has_arg = 0, .val = 'e' },
568219820Sjeff			{ .name = "gid-idx",  .has_arg = 1, .val = 'g' },
569331769Shselasky			{}
570219820Sjeff		};
571219820Sjeff
572331769Shselasky		c = getopt_long(argc, argv, "p:d:i:s:m:r:n:l:eg:",
573331769Shselasky							long_options, NULL);
574219820Sjeff		if (c == -1)
575219820Sjeff			break;
576219820Sjeff
577219820Sjeff		switch (c) {
578219820Sjeff		case 'p':
579331769Shselasky			port = strtoul(optarg, NULL, 0);
580331769Shselasky			if (port > 65535) {
581219820Sjeff				usage(argv[0]);
582219820Sjeff				return 1;
583219820Sjeff			}
584219820Sjeff			break;
585219820Sjeff
586219820Sjeff		case 'd':
587331769Shselasky			ib_devname = strdupa(optarg);
588219820Sjeff			break;
589219820Sjeff
590219820Sjeff		case 'i':
591219820Sjeff			ib_port = strtol(optarg, NULL, 0);
592331769Shselasky			if (ib_port < 1) {
593219820Sjeff				usage(argv[0]);
594219820Sjeff				return 1;
595219820Sjeff			}
596219820Sjeff			break;
597219820Sjeff
598219820Sjeff		case 's':
599331769Shselasky			size = strtoul(optarg, NULL, 0);
600219820Sjeff			break;
601219820Sjeff
602219820Sjeff		case 'm':
603219820Sjeff			mtu = pp_mtu_to_enum(strtol(optarg, NULL, 0));
604331769Shselasky			if (mtu == 0) {
605219820Sjeff				usage(argv[0]);
606219820Sjeff				return 1;
607219820Sjeff			}
608219820Sjeff			break;
609219820Sjeff
610219820Sjeff		case 'r':
611331769Shselasky			rx_depth = strtoul(optarg, NULL, 0);
612219820Sjeff			break;
613219820Sjeff
614219820Sjeff		case 'n':
615331769Shselasky			iters = strtoul(optarg, NULL, 0);
616219820Sjeff			break;
617219820Sjeff
618219820Sjeff		case 'l':
619219820Sjeff			sl = strtol(optarg, NULL, 0);
620219820Sjeff			break;
621219820Sjeff
622219820Sjeff		case 'e':
623219820Sjeff			++use_event;
624219820Sjeff			break;
625219820Sjeff
626219820Sjeff		case 'g':
627219820Sjeff			gidx = strtol(optarg, NULL, 0);
628219820Sjeff			break;
629219820Sjeff
630219820Sjeff		default:
631219820Sjeff			usage(argv[0]);
632219820Sjeff			return 1;
633219820Sjeff		}
634219820Sjeff	}
635219820Sjeff
636219820Sjeff	if (optind == argc - 1)
637331769Shselasky		servername = strdupa(argv[optind]);
638219820Sjeff	else if (optind < argc) {
639219820Sjeff		usage(argv[0]);
640219820Sjeff		return 1;
641219820Sjeff	}
642219820Sjeff
643219820Sjeff	page_size = sysconf(_SC_PAGESIZE);
644219820Sjeff
645219820Sjeff	dev_list = ibv_get_device_list(NULL);
646219820Sjeff	if (!dev_list) {
647219820Sjeff		perror("Failed to get IB devices list");
648219820Sjeff		return 1;
649219820Sjeff	}
650219820Sjeff
651219820Sjeff	if (!ib_devname) {
652219820Sjeff		ib_dev = *dev_list;
653219820Sjeff		if (!ib_dev) {
654219820Sjeff			fprintf(stderr, "No IB devices found\n");
655219820Sjeff			return 1;
656219820Sjeff		}
657219820Sjeff	} else {
658219820Sjeff		int i;
659219820Sjeff		for (i = 0; dev_list[i]; ++i)
660219820Sjeff			if (!strcmp(ibv_get_device_name(dev_list[i]), ib_devname))
661219820Sjeff				break;
662219820Sjeff		ib_dev = dev_list[i];
663219820Sjeff		if (!ib_dev) {
664219820Sjeff			fprintf(stderr, "IB device %s not found\n", ib_devname);
665219820Sjeff			return 1;
666219820Sjeff		}
667219820Sjeff	}
668219820Sjeff
669219820Sjeff	ctx = pp_init_ctx(ib_dev, size, rx_depth, ib_port, use_event);
670219820Sjeff	if (!ctx)
671219820Sjeff		return 1;
672219820Sjeff
673219820Sjeff	routs = pp_post_recv(ctx, ctx->rx_depth);
674219820Sjeff	if (routs < ctx->rx_depth) {
675219820Sjeff		fprintf(stderr, "Couldn't post receive (%d)\n", routs);
676219820Sjeff		return 1;
677219820Sjeff	}
678219820Sjeff
679219820Sjeff	if (use_event)
680219820Sjeff		if (ibv_req_notify_cq(ctx->cq, 0)) {
681219820Sjeff			fprintf(stderr, "Couldn't request CQ notification\n");
682219820Sjeff			return 1;
683219820Sjeff		}
684219820Sjeff
685331769Shselasky
686219820Sjeff	if (pp_get_port_info(ctx->context, ib_port, &ctx->portinfo)) {
687219820Sjeff		fprintf(stderr, "Couldn't get port info\n");
688219820Sjeff		return 1;
689219820Sjeff	}
690219820Sjeff
691219820Sjeff	my_dest.lid = ctx->portinfo.lid;
692331769Shselasky	if (ctx->portinfo.link_layer != IBV_LINK_LAYER_ETHERNET &&
693331769Shselasky							!my_dest.lid) {
694219820Sjeff		fprintf(stderr, "Couldn't get local LID\n");
695219820Sjeff		return 1;
696219820Sjeff	}
697219820Sjeff
698219820Sjeff	if (gidx >= 0) {
699219820Sjeff		if (ibv_query_gid(ctx->context, ib_port, gidx, &my_dest.gid)) {
700331769Shselasky			fprintf(stderr, "can't read sgid of index %d\n", gidx);
701219820Sjeff			return 1;
702219820Sjeff		}
703219820Sjeff	} else
704219820Sjeff		memset(&my_dest.gid, 0, sizeof my_dest.gid);
705219820Sjeff
706219820Sjeff	my_dest.qpn = ctx->qp->qp_num;
707219820Sjeff	my_dest.psn = lrand48() & 0xffffff;
708219820Sjeff	inet_ntop(AF_INET6, &my_dest.gid, gid, sizeof gid);
709219820Sjeff	printf("  local address:  LID 0x%04x, QPN 0x%06x, PSN 0x%06x, GID %s\n",
710219820Sjeff	       my_dest.lid, my_dest.qpn, my_dest.psn, gid);
711219820Sjeff
712331769Shselasky
713219820Sjeff	if (servername)
714219820Sjeff		rem_dest = pp_client_exch_dest(servername, port, &my_dest);
715219820Sjeff	else
716331769Shselasky		rem_dest = pp_server_exch_dest(ctx, ib_port, mtu, port, sl,
717331769Shselasky								&my_dest, gidx);
718219820Sjeff
719219820Sjeff	if (!rem_dest)
720219820Sjeff		return 1;
721219820Sjeff
722219820Sjeff	inet_ntop(AF_INET6, &rem_dest->gid, gid, sizeof gid);
723219820Sjeff	printf("  remote address: LID 0x%04x, QPN 0x%06x, PSN 0x%06x, GID %s\n",
724219820Sjeff	       rem_dest->lid, rem_dest->qpn, rem_dest->psn, gid);
725219820Sjeff
726219820Sjeff	if (servername)
727331769Shselasky		if (pp_connect_ctx(ctx, ib_port, my_dest.psn, mtu, sl, rem_dest,
728331769Shselasky					gidx))
729219820Sjeff			return 1;
730219820Sjeff
731219820Sjeff	ctx->pending = PINGPONG_RECV_WRID;
732219820Sjeff
733219820Sjeff	if (servername) {
734219820Sjeff		if (pp_post_send(ctx)) {
735219820Sjeff			fprintf(stderr, "Couldn't post send\n");
736219820Sjeff			return 1;
737219820Sjeff		}
738219820Sjeff		ctx->pending |= PINGPONG_SEND_WRID;
739219820Sjeff	}
740219820Sjeff
741219820Sjeff	if (gettimeofday(&start, NULL)) {
742219820Sjeff		perror("gettimeofday");
743219820Sjeff		return 1;
744219820Sjeff	}
745219820Sjeff
746219820Sjeff	rcnt = scnt = 0;
747219820Sjeff	while (rcnt < iters || scnt < iters) {
748219820Sjeff		if (use_event) {
749219820Sjeff			struct ibv_cq *ev_cq;
750219820Sjeff			void          *ev_ctx;
751219820Sjeff
752219820Sjeff			if (ibv_get_cq_event(ctx->channel, &ev_cq, &ev_ctx)) {
753219820Sjeff				fprintf(stderr, "Failed to get cq_event\n");
754219820Sjeff				return 1;
755219820Sjeff			}
756219820Sjeff
757219820Sjeff			++num_cq_events;
758219820Sjeff
759219820Sjeff			if (ev_cq != ctx->cq) {
760219820Sjeff				fprintf(stderr, "CQ event for unknown CQ %p\n", ev_cq);
761219820Sjeff				return 1;
762219820Sjeff			}
763219820Sjeff
764219820Sjeff			if (ibv_req_notify_cq(ctx->cq, 0)) {
765219820Sjeff				fprintf(stderr, "Couldn't request CQ notification\n");
766219820Sjeff				return 1;
767219820Sjeff			}
768219820Sjeff		}
769219820Sjeff
770219820Sjeff		{
771219820Sjeff			struct ibv_wc wc[2];
772219820Sjeff			int ne, i;
773219820Sjeff
774219820Sjeff			do {
775219820Sjeff				ne = ibv_poll_cq(ctx->cq, 2, wc);
776219820Sjeff				if (ne < 0) {
777219820Sjeff					fprintf(stderr, "poll CQ failed %d\n", ne);
778219820Sjeff					return 1;
779219820Sjeff				}
780331769Shselasky
781219820Sjeff			} while (!use_event && ne < 1);
782219820Sjeff
783219820Sjeff			for (i = 0; i < ne; ++i) {
784219820Sjeff				if (wc[i].status != IBV_WC_SUCCESS) {
785219820Sjeff					fprintf(stderr, "Failed status %s (%d) for wr_id %d\n",
786219820Sjeff						ibv_wc_status_str(wc[i].status),
787219820Sjeff						wc[i].status, (int) wc[i].wr_id);
788219820Sjeff					return 1;
789219820Sjeff				}
790219820Sjeff
791219820Sjeff				switch ((int) wc[i].wr_id) {
792219820Sjeff				case PINGPONG_SEND_WRID:
793219820Sjeff					++scnt;
794219820Sjeff					break;
795219820Sjeff
796219820Sjeff				case PINGPONG_RECV_WRID:
797219820Sjeff					if (--routs <= 1) {
798219820Sjeff						routs += pp_post_recv(ctx, ctx->rx_depth - routs);
799219820Sjeff						if (routs < ctx->rx_depth) {
800219820Sjeff							fprintf(stderr,
801219820Sjeff								"Couldn't post receive (%d)\n",
802219820Sjeff								routs);
803219820Sjeff							return 1;
804219820Sjeff						}
805219820Sjeff					}
806219820Sjeff
807219820Sjeff					++rcnt;
808219820Sjeff					break;
809219820Sjeff
810219820Sjeff				default:
811219820Sjeff					fprintf(stderr, "Completion for unknown wr_id %d\n",
812219820Sjeff						(int) wc[i].wr_id);
813219820Sjeff					return 1;
814219820Sjeff				}
815219820Sjeff
816219820Sjeff				ctx->pending &= ~(int) wc[i].wr_id;
817219820Sjeff				if (scnt < iters && !ctx->pending) {
818219820Sjeff					if (pp_post_send(ctx)) {
819219820Sjeff						fprintf(stderr, "Couldn't post send\n");
820219820Sjeff						return 1;
821219820Sjeff					}
822219820Sjeff					ctx->pending = PINGPONG_RECV_WRID |
823219820Sjeff						       PINGPONG_SEND_WRID;
824219820Sjeff				}
825219820Sjeff			}
826219820Sjeff		}
827219820Sjeff	}
828219820Sjeff
829219820Sjeff	if (gettimeofday(&end, NULL)) {
830219820Sjeff		perror("gettimeofday");
831219820Sjeff		return 1;
832219820Sjeff	}
833219820Sjeff
834219820Sjeff	{
835219820Sjeff		float usec = (end.tv_sec - start.tv_sec) * 1000000 +
836219820Sjeff			(end.tv_usec - start.tv_usec);
837219820Sjeff		long long bytes = (long long) size * iters * 2;
838219820Sjeff
839219820Sjeff		printf("%lld bytes in %.2f seconds = %.2f Mbit/sec\n",
840219820Sjeff		       bytes, usec / 1000000., bytes * 8. / usec);
841219820Sjeff		printf("%d iters in %.2f seconds = %.2f usec/iter\n",
842219820Sjeff		       iters, usec / 1000000., usec / iters);
843219820Sjeff	}
844219820Sjeff
845219820Sjeff	ibv_ack_cq_events(ctx->cq, num_cq_events);
846219820Sjeff
847219820Sjeff	if (pp_close_ctx(ctx))
848219820Sjeff		return 1;
849219820Sjeff
850219820Sjeff	ibv_free_device_list(dev_list);
851219820Sjeff	free(rem_dest);
852219820Sjeff
853219820Sjeff	return 0;
854219820Sjeff}
855