xref: /linux/tools/testing/selftests/drivers/net/hw/ncdevmem.c (revision 8bf22c33e7a172fbc72464f4cc484d23a6b412ba)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * tcpdevmem netcat. Works similarly to netcat but does device memory TCP
4  * instead of regular TCP. Uses udmabuf to mock a dmabuf provider.
5  *
6  * Usage:
7  *
8  *     On server:
9  *     ncdevmem -s <server IP> [-c <client IP>] -f eth1 -l -p 5201
10  *
11  *     On client:
12  *     echo -n "hello\nworld" | \
13  *		ncdevmem -s <server IP> [-c <client IP>] -p 5201 -f eth1
14  *
15  * Note this is compatible with regular netcat. i.e. the sender or receiver can
16  * be replaced with regular netcat to test the RX or TX path in isolation.
17  *
18  * Test data validation (devmem TCP on RX only):
19  *
20  *     On server:
21  *     ncdevmem -s <server IP> [-c <client IP>] -f eth1 -l -p 5201 -v 7
22  *
23  *     On client:
24  *     yes $(echo -e \\x01\\x02\\x03\\x04\\x05\\x06) | \
25  *             head -c 1G | \
26  *             nc <server IP> 5201 -p 5201
27  *
28  * Test data validation (devmem TCP on RX and TX, validation happens on RX):
29  *
30  *	On server:
31  *	ncdevmem -s <server IP> [-c <client IP>] -l -p 5201 -v 8 -f eth1
32  *
33  *	On client:
34  *	yes $(echo -e \\x01\\x02\\x03\\x04\\x05\\x06\\x07) | \
35  *		head -c 1M | \
36  *		ncdevmem -s <server IP> [-c <client IP>] -p 5201 -f eth1
37  */
38 #define _GNU_SOURCE
39 #define __EXPORTED_HEADERS__
40 
41 #include <linux/uio.h>
42 #include <stdarg.h>
43 #include <stdio.h>
44 #include <stdlib.h>
45 #include <unistd.h>
46 #include <stdbool.h>
47 #include <string.h>
48 #include <errno.h>
49 #define __iovec_defined
50 #include <fcntl.h>
51 #include <limits.h>
52 #include <malloc.h>
53 #include <error.h>
54 #include <poll.h>
55 
56 #include <arpa/inet.h>
57 #include <sys/socket.h>
58 #include <sys/mman.h>
59 #include <sys/ioctl.h>
60 #include <sys/syscall.h>
61 #include <sys/time.h>
62 
63 #include <linux/memfd.h>
64 #include <linux/dma-buf.h>
65 #include <linux/errqueue.h>
66 #include <linux/udmabuf.h>
67 #include <linux/types.h>
68 #include <linux/netlink.h>
69 #include <linux/genetlink.h>
70 #include <linux/netdev.h>
71 #include <linux/ethtool_netlink.h>
72 #include <time.h>
73 #include <net/if.h>
74 
75 #include "netdev-user.h"
76 #include "ethtool-user.h"
77 #include <ynl.h>
78 
79 #define PAGE_SHIFT 12
80 #define TEST_PREFIX "ncdevmem"
81 #define NUM_PAGES 16000
82 
83 #ifndef MSG_SOCK_DEVMEM
84 #define MSG_SOCK_DEVMEM 0x2000000
85 #endif
86 
87 #define MAX_IOV 1024
88 
89 static size_t max_chunk;
90 static char *server_ip;
91 static char *client_ip;
92 static char *port;
93 static size_t do_validation;
94 static int start_queue = -1;
95 static int num_queues = -1;
96 static char *ifname;
97 static unsigned int ifindex;
98 static unsigned int dmabuf_id;
99 static uint32_t tx_dmabuf_id;
100 static int waittime_ms = 500;
101 static bool fail_on_linear;
102 
103 /* System state loaded by current_config_load() */
104 #define MAX_FLOWS	8
105 static int ntuple_ids[MAX_FLOWS] = { -1, -1, -1, -1, -1, -1, -1, -1, };
106 
107 struct memory_buffer {
108 	int fd;
109 	size_t size;
110 
111 	int devfd;
112 	int memfd;
113 	char *buf_mem;
114 };
115 
116 struct memory_provider {
117 	struct memory_buffer *(*alloc)(size_t size);
118 	void (*free)(struct memory_buffer *ctx);
119 	void (*memcpy_to_device)(struct memory_buffer *dst, size_t off,
120 				 void *src, int n);
121 	void (*memcpy_from_device)(void *dst, struct memory_buffer *src,
122 				   size_t off, int n);
123 };
124 
pr_err(const char * fmt,...)125 static void pr_err(const char *fmt, ...)
126 {
127 	va_list args;
128 
129 	fprintf(stderr, "%s: ", TEST_PREFIX);
130 
131 	va_start(args, fmt);
132 	vfprintf(stderr, fmt, args);
133 	va_end(args);
134 
135 	if (errno != 0)
136 		fprintf(stderr, ": %s", strerror(errno));
137 	fprintf(stderr, "\n");
138 }
139 
udmabuf_alloc(size_t size)140 static struct memory_buffer *udmabuf_alloc(size_t size)
141 {
142 	struct udmabuf_create create;
143 	struct memory_buffer *ctx;
144 	int ret;
145 
146 	ctx = malloc(sizeof(*ctx));
147 	if (!ctx)
148 		return NULL;
149 
150 	ctx->size = size;
151 
152 	ctx->devfd = open("/dev/udmabuf", O_RDWR);
153 	if (ctx->devfd < 0) {
154 		pr_err("[skip,no-udmabuf: Unable to access DMA buffer device file]");
155 		goto err_free_ctx;
156 	}
157 
158 	ctx->memfd = memfd_create("udmabuf-test", MFD_ALLOW_SEALING);
159 	if (ctx->memfd < 0) {
160 		pr_err("[skip,no-memfd]");
161 		goto err_close_dev;
162 	}
163 
164 	ret = fcntl(ctx->memfd, F_ADD_SEALS, F_SEAL_SHRINK);
165 	if (ret < 0) {
166 		pr_err("[skip,fcntl-add-seals]");
167 		goto err_close_memfd;
168 	}
169 
170 	ret = ftruncate(ctx->memfd, size);
171 	if (ret == -1) {
172 		pr_err("[FAIL,memfd-truncate]");
173 		goto err_close_memfd;
174 	}
175 
176 	memset(&create, 0, sizeof(create));
177 
178 	create.memfd = ctx->memfd;
179 	create.offset = 0;
180 	create.size = size;
181 	ctx->fd = ioctl(ctx->devfd, UDMABUF_CREATE, &create);
182 	if (ctx->fd < 0) {
183 		pr_err("[FAIL, create udmabuf]");
184 		goto err_close_fd;
185 	}
186 
187 	ctx->buf_mem = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED,
188 			    ctx->fd, 0);
189 	if (ctx->buf_mem == MAP_FAILED) {
190 		pr_err("[FAIL, map udmabuf]");
191 		goto err_close_fd;
192 	}
193 
194 	return ctx;
195 
196 err_close_fd:
197 	close(ctx->fd);
198 err_close_memfd:
199 	close(ctx->memfd);
200 err_close_dev:
201 	close(ctx->devfd);
202 err_free_ctx:
203 	free(ctx);
204 	return NULL;
205 }
206 
udmabuf_free(struct memory_buffer * ctx)207 static void udmabuf_free(struct memory_buffer *ctx)
208 {
209 	munmap(ctx->buf_mem, ctx->size);
210 	close(ctx->fd);
211 	close(ctx->memfd);
212 	close(ctx->devfd);
213 	free(ctx);
214 }
215 
udmabuf_memcpy_to_device(struct memory_buffer * dst,size_t off,void * src,int n)216 static void udmabuf_memcpy_to_device(struct memory_buffer *dst, size_t off,
217 				     void *src, int n)
218 {
219 	struct dma_buf_sync sync = {};
220 
221 	sync.flags = DMA_BUF_SYNC_START | DMA_BUF_SYNC_WRITE;
222 	ioctl(dst->fd, DMA_BUF_IOCTL_SYNC, &sync);
223 
224 	memcpy(dst->buf_mem + off, src, n);
225 
226 	sync.flags = DMA_BUF_SYNC_END | DMA_BUF_SYNC_WRITE;
227 	ioctl(dst->fd, DMA_BUF_IOCTL_SYNC, &sync);
228 }
229 
udmabuf_memcpy_from_device(void * dst,struct memory_buffer * src,size_t off,int n)230 static void udmabuf_memcpy_from_device(void *dst, struct memory_buffer *src,
231 				       size_t off, int n)
232 {
233 	struct dma_buf_sync sync = {};
234 
235 	sync.flags = DMA_BUF_SYNC_START;
236 	ioctl(src->fd, DMA_BUF_IOCTL_SYNC, &sync);
237 
238 	memcpy(dst, src->buf_mem + off, n);
239 
240 	sync.flags = DMA_BUF_SYNC_END;
241 	ioctl(src->fd, DMA_BUF_IOCTL_SYNC, &sync);
242 }
243 
244 static struct memory_provider udmabuf_memory_provider = {
245 	.alloc = udmabuf_alloc,
246 	.free = udmabuf_free,
247 	.memcpy_to_device = udmabuf_memcpy_to_device,
248 	.memcpy_from_device = udmabuf_memcpy_from_device,
249 };
250 
251 static struct memory_provider *provider = &udmabuf_memory_provider;
252 
print_nonzero_bytes(void * ptr,size_t size)253 static void print_nonzero_bytes(void *ptr, size_t size)
254 {
255 	unsigned char *p = ptr;
256 	unsigned int i;
257 
258 	for (i = 0; i < size; i++)
259 		putchar(p[i]);
260 }
261 
validate_buffer(void * line,size_t size)262 int validate_buffer(void *line, size_t size)
263 {
264 	static unsigned char seed = 1;
265 	unsigned char *ptr = line;
266 	unsigned char expected;
267 	static int errors;
268 	size_t i;
269 
270 	for (i = 0; i < size; i++) {
271 		expected = seed ? seed : '\n';
272 		if (ptr[i] != expected) {
273 			fprintf(stderr,
274 				"Failed validation: expected=%u, actual=%u, index=%lu\n",
275 				expected, ptr[i], i);
276 			errors++;
277 			if (errors > 20) {
278 				pr_err("validation failed");
279 				return -1;
280 			}
281 		}
282 		seed++;
283 		if (seed == do_validation)
284 			seed = 0;
285 	}
286 
287 	fprintf(stdout, "Validated buffer\n");
288 	return 0;
289 }
290 
291 static int
__run_command(char * out,size_t outlen,const char * cmd,va_list args)292 __run_command(char *out, size_t outlen, const char *cmd, va_list args)
293 {
294 	char command[256];
295 	FILE *fp;
296 
297 	vsnprintf(command, sizeof(command), cmd, args);
298 
299 	fprintf(stderr, "Running: %s\n", command);
300 	fp = popen(command, "r");
301 	if (!fp)
302 		return -1;
303 	if (out) {
304 		size_t len;
305 
306 		if (!fgets(out, outlen, fp))
307 			return -1;
308 
309 		/* Remove trailing newline if present */
310 		len = strlen(out);
311 		if (len && out[len - 1] == '\n')
312 			out[len - 1] = '\0';
313 	}
314 	return pclose(fp);
315 }
316 
run_command(const char * cmd,...)317 static int run_command(const char *cmd, ...)
318 {
319 	va_list args;
320 	int ret;
321 
322 	va_start(args, cmd);
323 	ret = __run_command(NULL, 0, cmd, args);
324 	va_end(args);
325 
326 	return ret;
327 }
328 
ethtool_add_flow(const char * format,...)329 static int ethtool_add_flow(const char *format, ...)
330 {
331 	char local_output[256], cmd[256];
332 	const char *id_start;
333 	int flow_idx, ret;
334 	char *endptr;
335 	long flow_id;
336 	va_list args;
337 
338 	for (flow_idx = 0; flow_idx < MAX_FLOWS; flow_idx++)
339 		if (ntuple_ids[flow_idx] == -1)
340 			break;
341 	if (flow_idx == MAX_FLOWS) {
342 		fprintf(stderr, "Error: too many flows\n");
343 		return -1;
344 	}
345 
346 	snprintf(cmd, sizeof(cmd), "ethtool -N %s %s", ifname, format);
347 
348 	va_start(args, format);
349 	ret = __run_command(local_output, sizeof(local_output), cmd, args);
350 	va_end(args);
351 
352 	if (ret != 0)
353 		return ret;
354 
355 	/* Extract the ID from the output */
356 	id_start = strstr(local_output, "Added rule with ID ");
357 	if (!id_start)
358 		return -1;
359 	id_start += strlen("Added rule with ID ");
360 
361 	flow_id = strtol(id_start, &endptr, 10);
362 	if (endptr == id_start || flow_id < 0 || flow_id > INT_MAX)
363 		return -1;
364 
365 	fprintf(stderr, "Added flow rule with ID %ld\n", flow_id);
366 	ntuple_ids[flow_idx] = flow_id;
367 	return flow_id;
368 }
369 
rxq_num(int ifindex)370 static int rxq_num(int ifindex)
371 {
372 	struct ethtool_channels_get_req *req;
373 	struct ethtool_channels_get_rsp *rsp;
374 	struct ynl_error yerr;
375 	struct ynl_sock *ys;
376 	int num = -1;
377 
378 	ys = ynl_sock_create(&ynl_ethtool_family, &yerr);
379 	if (!ys) {
380 		fprintf(stderr, "YNL: %s\n", yerr.msg);
381 		return -1;
382 	}
383 
384 	req = ethtool_channels_get_req_alloc();
385 	ethtool_channels_get_req_set_header_dev_index(req, ifindex);
386 	rsp = ethtool_channels_get(ys, req);
387 	if (rsp)
388 		num = rsp->rx_count + rsp->combined_count;
389 	ethtool_channels_get_req_free(req);
390 	ethtool_channels_get_rsp_free(rsp);
391 
392 	ynl_sock_destroy(ys);
393 
394 	return num;
395 }
396 
reset_flow_steering(void)397 static void reset_flow_steering(void)
398 {
399 	int i;
400 
401 	for (i = 0; i < MAX_FLOWS; i++) {
402 		if (ntuple_ids[i] == -1)
403 			continue;
404 		run_command("ethtool -N %s delete %d",
405 			    ifname, ntuple_ids[i]);
406 		ntuple_ids[i] = -1;
407 	}
408 }
409 
tcp_data_split_str(int val)410 static const char *tcp_data_split_str(int val)
411 {
412 	switch (val) {
413 	case 0:
414 		return "off";
415 	case 1:
416 		return "auto";
417 	case 2:
418 		return "on";
419 	default:
420 		return "?";
421 	}
422 }
423 
get_ring_config(void)424 static struct ethtool_rings_get_rsp *get_ring_config(void)
425 {
426 	struct ethtool_rings_get_req *get_req;
427 	struct ethtool_rings_get_rsp *get_rsp;
428 	struct ynl_error yerr;
429 	struct ynl_sock *ys;
430 
431 	ys = ynl_sock_create(&ynl_ethtool_family, &yerr);
432 	if (!ys) {
433 		fprintf(stderr, "YNL: %s\n", yerr.msg);
434 		return NULL;
435 	}
436 
437 	get_req = ethtool_rings_get_req_alloc();
438 	ethtool_rings_get_req_set_header_dev_index(get_req, ifindex);
439 	get_rsp = ethtool_rings_get(ys, get_req);
440 	ethtool_rings_get_req_free(get_req);
441 
442 	ynl_sock_destroy(ys);
443 
444 	return get_rsp;
445 }
446 
restore_ring_config(const struct ethtool_rings_get_rsp * config)447 static void restore_ring_config(const struct ethtool_rings_get_rsp *config)
448 {
449 	struct ethtool_rings_get_req *get_req;
450 	struct ethtool_rings_get_rsp *get_rsp;
451 	struct ethtool_rings_set_req *req;
452 	struct ynl_error yerr;
453 	struct ynl_sock *ys;
454 	int ret;
455 
456 	if (!config)
457 		return;
458 
459 	ys = ynl_sock_create(&ynl_ethtool_family, &yerr);
460 	if (!ys) {
461 		fprintf(stderr, "YNL: %s\n", yerr.msg);
462 		return;
463 	}
464 
465 	req = ethtool_rings_set_req_alloc();
466 	ethtool_rings_set_req_set_header_dev_index(req, ifindex);
467 	ethtool_rings_set_req_set_tcp_data_split(req,
468 						ETHTOOL_TCP_DATA_SPLIT_UNKNOWN);
469 	if (config->_present.hds_thresh)
470 		ethtool_rings_set_req_set_hds_thresh(req, config->hds_thresh);
471 
472 	ret = ethtool_rings_set(ys, req);
473 	if (ret < 0)
474 		fprintf(stderr, "YNL restoring HDS cfg: %s\n", ys->err.msg);
475 
476 	get_req = ethtool_rings_get_req_alloc();
477 	ethtool_rings_get_req_set_header_dev_index(get_req, ifindex);
478 	get_rsp = ethtool_rings_get(ys, get_req);
479 	ethtool_rings_get_req_free(get_req);
480 
481 	/* use explicit value if UKNOWN didn't give us the previous */
482 	if (get_rsp->tcp_data_split != config->tcp_data_split) {
483 		ethtool_rings_set_req_set_tcp_data_split(req,
484 							config->tcp_data_split);
485 		ret = ethtool_rings_set(ys, req);
486 		if (ret < 0)
487 			fprintf(stderr, "YNL restoring expl HDS cfg: %s\n",
488 				ys->err.msg);
489 	}
490 
491 	ethtool_rings_get_rsp_free(get_rsp);
492 	ethtool_rings_set_req_free(req);
493 
494 	ynl_sock_destroy(ys);
495 }
496 
497 static int
configure_headersplit(const struct ethtool_rings_get_rsp * old,bool on)498 configure_headersplit(const struct ethtool_rings_get_rsp *old, bool on)
499 {
500 	struct ethtool_rings_get_req *get_req;
501 	struct ethtool_rings_get_rsp *get_rsp;
502 	struct ethtool_rings_set_req *req;
503 	struct ynl_error yerr;
504 	struct ynl_sock *ys;
505 	int ret;
506 
507 	ys = ynl_sock_create(&ynl_ethtool_family, &yerr);
508 	if (!ys) {
509 		fprintf(stderr, "YNL: %s\n", yerr.msg);
510 		return -1;
511 	}
512 
513 	req = ethtool_rings_set_req_alloc();
514 	ethtool_rings_set_req_set_header_dev_index(req, ifindex);
515 	if (on) {
516 		ethtool_rings_set_req_set_tcp_data_split(req,
517 						ETHTOOL_TCP_DATA_SPLIT_ENABLED);
518 		if (old->_present.hds_thresh)
519 			ethtool_rings_set_req_set_hds_thresh(req, 0);
520 	} else {
521 		ethtool_rings_set_req_set_tcp_data_split(req,
522 						ETHTOOL_TCP_DATA_SPLIT_UNKNOWN);
523 	}
524 	ret = ethtool_rings_set(ys, req);
525 	if (ret < 0)
526 		fprintf(stderr, "YNL failed: %s\n", ys->err.msg);
527 	ethtool_rings_set_req_free(req);
528 
529 	if (ret == 0) {
530 		get_req = ethtool_rings_get_req_alloc();
531 		ethtool_rings_get_req_set_header_dev_index(get_req, ifindex);
532 		get_rsp = ethtool_rings_get(ys, get_req);
533 		ethtool_rings_get_req_free(get_req);
534 		if (get_rsp)
535 			fprintf(stderr, "TCP header split: %s\n",
536 				tcp_data_split_str(get_rsp->tcp_data_split));
537 		ethtool_rings_get_rsp_free(get_rsp);
538 	}
539 
540 	ynl_sock_destroy(ys);
541 
542 	return ret;
543 }
544 
configure_rss(void)545 static int configure_rss(void)
546 {
547 	return run_command("ethtool -X %s equal %d >&2", ifname, start_queue);
548 }
549 
reset_rss(void)550 static void reset_rss(void)
551 {
552 	run_command("ethtool -X %s default >&2", ifname, start_queue);
553 }
554 
check_changing_channels(unsigned int rx,unsigned int tx)555 static int check_changing_channels(unsigned int rx, unsigned int tx)
556 {
557 	struct ethtool_channels_get_req *gchan;
558 	struct ethtool_channels_set_req *schan;
559 	struct ethtool_channels_get_rsp *chan;
560 	struct ynl_error yerr;
561 	struct ynl_sock *ys;
562 	int ret;
563 
564 	fprintf(stderr, "setting channel count rx:%u tx:%u\n", rx, tx);
565 
566 	ys = ynl_sock_create(&ynl_ethtool_family, &yerr);
567 	if (!ys) {
568 		fprintf(stderr, "YNL: %s\n", yerr.msg);
569 		return -1;
570 	}
571 
572 	gchan = ethtool_channels_get_req_alloc();
573 	if (!gchan) {
574 		ret = -1;
575 		goto exit_close_sock;
576 	}
577 
578 	ethtool_channels_get_req_set_header_dev_index(gchan, ifindex);
579 	chan = ethtool_channels_get(ys, gchan);
580 	ethtool_channels_get_req_free(gchan);
581 	if (!chan) {
582 		fprintf(stderr, "YNL get channels: %s\n", ys->err.msg);
583 		ret = -1;
584 		goto exit_close_sock;
585 	}
586 
587 	schan =	ethtool_channels_set_req_alloc();
588 	if (!schan) {
589 		ret = -1;
590 		goto exit_free_chan;
591 	}
592 
593 	ethtool_channels_set_req_set_header_dev_index(schan, ifindex);
594 
595 	if (chan->_present.combined_count) {
596 		if (chan->_present.rx_count || chan->_present.tx_count) {
597 			ethtool_channels_set_req_set_rx_count(schan, 0);
598 			ethtool_channels_set_req_set_tx_count(schan, 0);
599 		}
600 
601 		if (rx == tx) {
602 			ethtool_channels_set_req_set_combined_count(schan, rx);
603 		} else if (rx > tx) {
604 			ethtool_channels_set_req_set_combined_count(schan, tx);
605 			ethtool_channels_set_req_set_rx_count(schan, rx - tx);
606 		} else {
607 			ethtool_channels_set_req_set_combined_count(schan, rx);
608 			ethtool_channels_set_req_set_tx_count(schan, tx - rx);
609 		}
610 
611 	} else if (chan->_present.rx_count) {
612 		ethtool_channels_set_req_set_rx_count(schan, rx);
613 		ethtool_channels_set_req_set_tx_count(schan, tx);
614 	} else {
615 		fprintf(stderr, "Error: device has neither combined nor rx channels\n");
616 		ret = -1;
617 		goto exit_free_schan;
618 	}
619 
620 	ret = ethtool_channels_set(ys, schan);
621 	if (ret) {
622 		fprintf(stderr, "YNL set channels: %s\n", ys->err.msg);
623 	} else {
624 		/* We were expecting a failure, go back to previous settings */
625 		ethtool_channels_set_req_set_combined_count(schan,
626 							    chan->combined_count);
627 		ethtool_channels_set_req_set_rx_count(schan, chan->rx_count);
628 		ethtool_channels_set_req_set_tx_count(schan, chan->tx_count);
629 
630 		ret = ethtool_channels_set(ys, schan);
631 		if (ret)
632 			fprintf(stderr, "YNL un-setting channels: %s\n",
633 				ys->err.msg);
634 	}
635 
636 exit_free_schan:
637 	ethtool_channels_set_req_free(schan);
638 exit_free_chan:
639 	ethtool_channels_get_rsp_free(chan);
640 exit_close_sock:
641 	ynl_sock_destroy(ys);
642 
643 	return ret;
644 }
645 
configure_flow_steering(struct sockaddr_in6 * server_sin)646 static int configure_flow_steering(struct sockaddr_in6 *server_sin)
647 {
648 	const char *type = "tcp6";
649 	const char *server_addr;
650 	char buf[40];
651 	int flow_id;
652 
653 	inet_ntop(AF_INET6, &server_sin->sin6_addr, buf, sizeof(buf));
654 	server_addr = buf;
655 
656 	if (IN6_IS_ADDR_V4MAPPED(&server_sin->sin6_addr)) {
657 		type = "tcp4";
658 		server_addr = strrchr(server_addr, ':') + 1;
659 	}
660 
661 	/* Try configure 5-tuple */
662 	flow_id = ethtool_add_flow("flow-type %s %s %s dst-ip %s %s %s dst-port %s queue %d",
663 				   type,
664 				   client_ip ? "src-ip" : "",
665 				   client_ip ?: "",
666 				   server_addr,
667 				   client_ip ? "src-port" : "",
668 				   client_ip ? port : "",
669 				   port, start_queue);
670 	if (flow_id < 0) {
671 		/* If that fails, try configure 3-tuple */
672 		flow_id = ethtool_add_flow("flow-type %s dst-ip %s dst-port %s queue %d",
673 					   type, server_addr, port, start_queue);
674 		if (flow_id < 0)
675 			/* If that fails, return error */
676 			return -1;
677 	}
678 
679 	return 0;
680 }
681 
bind_rx_queue(unsigned int ifindex,unsigned int dmabuf_fd,struct netdev_queue_id * queues,unsigned int n_queue_index,struct ynl_sock ** ys)682 static int bind_rx_queue(unsigned int ifindex, unsigned int dmabuf_fd,
683 			 struct netdev_queue_id *queues,
684 			 unsigned int n_queue_index, struct ynl_sock **ys)
685 {
686 	struct netdev_bind_rx_req *req = NULL;
687 	struct netdev_bind_rx_rsp *rsp = NULL;
688 	struct ynl_error yerr;
689 
690 	*ys = ynl_sock_create(&ynl_netdev_family, &yerr);
691 	if (!*ys) {
692 		netdev_queue_id_free(queues);
693 		fprintf(stderr, "YNL: %s\n", yerr.msg);
694 		return -1;
695 	}
696 
697 	req = netdev_bind_rx_req_alloc();
698 	netdev_bind_rx_req_set_ifindex(req, ifindex);
699 	netdev_bind_rx_req_set_fd(req, dmabuf_fd);
700 	__netdev_bind_rx_req_set_queues(req, queues, n_queue_index);
701 
702 	rsp = netdev_bind_rx(*ys, req);
703 	if (!rsp) {
704 		perror("netdev_bind_rx");
705 		goto err_close;
706 	}
707 
708 	if (!rsp->_present.id) {
709 		perror("id not present");
710 		goto err_close;
711 	}
712 
713 	fprintf(stderr, "got dmabuf id=%d\n", rsp->id);
714 	dmabuf_id = rsp->id;
715 
716 	netdev_bind_rx_req_free(req);
717 	netdev_bind_rx_rsp_free(rsp);
718 
719 	return 0;
720 
721 err_close:
722 	fprintf(stderr, "YNL failed: %s\n", (*ys)->err.msg);
723 	netdev_bind_rx_req_free(req);
724 	ynl_sock_destroy(*ys);
725 	return -1;
726 }
727 
bind_tx_queue(unsigned int ifindex,unsigned int dmabuf_fd,struct ynl_sock ** ys)728 static int bind_tx_queue(unsigned int ifindex, unsigned int dmabuf_fd,
729 			 struct ynl_sock **ys)
730 {
731 	struct netdev_bind_tx_req *req = NULL;
732 	struct netdev_bind_tx_rsp *rsp = NULL;
733 	struct ynl_error yerr;
734 
735 	*ys = ynl_sock_create(&ynl_netdev_family, &yerr);
736 	if (!*ys) {
737 		fprintf(stderr, "YNL: %s\n", yerr.msg);
738 		return -1;
739 	}
740 
741 	req = netdev_bind_tx_req_alloc();
742 	netdev_bind_tx_req_set_ifindex(req, ifindex);
743 	netdev_bind_tx_req_set_fd(req, dmabuf_fd);
744 
745 	rsp = netdev_bind_tx(*ys, req);
746 	if (!rsp) {
747 		perror("netdev_bind_tx");
748 		goto err_close;
749 	}
750 
751 	if (!rsp->_present.id) {
752 		perror("id not present");
753 		goto err_close;
754 	}
755 
756 	fprintf(stderr, "got tx dmabuf id=%d\n", rsp->id);
757 	tx_dmabuf_id = rsp->id;
758 
759 	netdev_bind_tx_req_free(req);
760 	netdev_bind_tx_rsp_free(rsp);
761 
762 	return 0;
763 
764 err_close:
765 	fprintf(stderr, "YNL failed: %s\n", (*ys)->err.msg);
766 	netdev_bind_tx_req_free(req);
767 	ynl_sock_destroy(*ys);
768 	return -1;
769 }
770 
enable_reuseaddr(int fd)771 static int enable_reuseaddr(int fd)
772 {
773 	int opt = 1;
774 	int ret;
775 
776 	ret = setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt));
777 	if (ret) {
778 		pr_err("SO_REUSEPORT failed");
779 		return -1;
780 	}
781 
782 	ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt));
783 	if (ret) {
784 		pr_err("SO_REUSEADDR failed");
785 		return -1;
786 	}
787 
788 	return 0;
789 }
790 
parse_address(const char * str,int port,struct sockaddr_in6 * sin6)791 static int parse_address(const char *str, int port, struct sockaddr_in6 *sin6)
792 {
793 	int ret;
794 
795 	sin6->sin6_family = AF_INET6;
796 	sin6->sin6_port = htons(port);
797 
798 	ret = inet_pton(sin6->sin6_family, str, &sin6->sin6_addr);
799 	if (ret != 1) {
800 		/* fallback to plain IPv4 */
801 		ret = inet_pton(AF_INET, str, &sin6->sin6_addr.s6_addr32[3]);
802 		if (ret != 1)
803 			return -1;
804 
805 		/* add ::ffff prefix */
806 		sin6->sin6_addr.s6_addr32[0] = 0;
807 		sin6->sin6_addr.s6_addr32[1] = 0;
808 		sin6->sin6_addr.s6_addr16[4] = 0;
809 		sin6->sin6_addr.s6_addr16[5] = 0xffff;
810 	}
811 
812 	return 0;
813 }
814 
create_queues(void)815 static struct netdev_queue_id *create_queues(void)
816 {
817 	struct netdev_queue_id *queues;
818 	size_t i = 0;
819 
820 	queues = netdev_queue_id_alloc(num_queues);
821 	for (i = 0; i < num_queues; i++) {
822 		netdev_queue_id_set_type(&queues[i], NETDEV_QUEUE_TYPE_RX);
823 		netdev_queue_id_set_id(&queues[i], start_queue + i);
824 	}
825 
826 	return queues;
827 }
828 
do_server(struct memory_buffer * mem)829 static int do_server(struct memory_buffer *mem)
830 {
831 	struct ethtool_rings_get_rsp *ring_config;
832 	char ctrl_data[sizeof(int) * 20000];
833 	size_t non_page_aligned_frags = 0;
834 	struct sockaddr_in6 client_addr;
835 	struct sockaddr_in6 server_sin;
836 	size_t page_aligned_frags = 0;
837 	size_t total_received = 0;
838 	socklen_t client_addr_len;
839 	bool is_devmem = false;
840 	char *tmp_mem = NULL;
841 	struct ynl_sock *ys;
842 	char iobuf[819200];
843 	int ret, err = -1;
844 	char buffer[256];
845 	int socket_fd;
846 	int client_fd;
847 
848 	ret = parse_address(server_ip, atoi(port), &server_sin);
849 	if (ret < 0) {
850 		pr_err("parse server address");
851 		return -1;
852 	}
853 
854 	ring_config = get_ring_config();
855 	if (!ring_config) {
856 		pr_err("Failed to get current ring configuration");
857 		return -1;
858 	}
859 
860 	if (configure_headersplit(ring_config, 1)) {
861 		pr_err("Failed to enable TCP header split");
862 		goto err_free_ring_config;
863 	}
864 
865 	/* Configure RSS to divert all traffic from our devmem queues */
866 	if (configure_rss()) {
867 		pr_err("Failed to configure rss");
868 		goto err_reset_headersplit;
869 	}
870 
871 	/* Flow steer our devmem flows to start_queue */
872 	if (configure_flow_steering(&server_sin)) {
873 		pr_err("Failed to configure flow steering");
874 		goto err_reset_rss;
875 	}
876 
877 	if (bind_rx_queue(ifindex, mem->fd, create_queues(), num_queues, &ys)) {
878 		pr_err("Failed to bind");
879 		goto err_reset_flow_steering;
880 	}
881 
882 	tmp_mem = malloc(mem->size);
883 	if (!tmp_mem)
884 		goto err_unbind;
885 
886 	socket_fd = socket(AF_INET6, SOCK_STREAM, 0);
887 	if (socket_fd < 0) {
888 		pr_err("Failed to create socket");
889 		goto err_free_tmp;
890 	}
891 
892 	if (enable_reuseaddr(socket_fd))
893 		goto err_close_socket;
894 
895 	fprintf(stderr, "binding to address %s:%d\n", server_ip,
896 		ntohs(server_sin.sin6_port));
897 
898 	ret = bind(socket_fd, &server_sin, sizeof(server_sin));
899 	if (ret) {
900 		pr_err("Failed to bind");
901 		goto err_close_socket;
902 	}
903 
904 	ret = listen(socket_fd, 1);
905 	if (ret) {
906 		pr_err("Failed to listen");
907 		goto err_close_socket;
908 	}
909 
910 	client_addr_len = sizeof(client_addr);
911 
912 	inet_ntop(AF_INET6, &server_sin.sin6_addr, buffer,
913 		  sizeof(buffer));
914 	fprintf(stderr, "Waiting or connection on %s:%d\n", buffer,
915 		ntohs(server_sin.sin6_port));
916 	client_fd = accept(socket_fd, &client_addr, &client_addr_len);
917 	if (client_fd < 0) {
918 		pr_err("Failed to accept");
919 		goto err_close_socket;
920 	}
921 
922 	inet_ntop(AF_INET6, &client_addr.sin6_addr, buffer,
923 		  sizeof(buffer));
924 	fprintf(stderr, "Got connection from %s:%d\n", buffer,
925 		ntohs(client_addr.sin6_port));
926 
927 	while (1) {
928 		struct iovec iov = { .iov_base = iobuf,
929 				     .iov_len = sizeof(iobuf) };
930 		struct dmabuf_cmsg *dmabuf_cmsg = NULL;
931 		struct cmsghdr *cm = NULL;
932 		struct msghdr msg = { 0 };
933 		struct dmabuf_token token;
934 		ssize_t ret;
935 
936 		is_devmem = false;
937 
938 		msg.msg_iov = &iov;
939 		msg.msg_iovlen = 1;
940 		msg.msg_control = ctrl_data;
941 		msg.msg_controllen = sizeof(ctrl_data);
942 		ret = recvmsg(client_fd, &msg, MSG_SOCK_DEVMEM);
943 		fprintf(stderr, "recvmsg ret=%ld\n", ret);
944 		if (ret < 0 && (errno == EAGAIN || errno == EWOULDBLOCK))
945 			continue;
946 		if (ret < 0) {
947 			perror("recvmsg");
948 			if (errno == EFAULT) {
949 				pr_err("received EFAULT, won't recover");
950 				goto err_close_client;
951 			}
952 			continue;
953 		}
954 		if (ret == 0) {
955 			errno = 0;
956 			pr_err("client exited");
957 			goto cleanup;
958 		}
959 
960 		for (cm = CMSG_FIRSTHDR(&msg); cm; cm = CMSG_NXTHDR(&msg, cm)) {
961 			if (cm->cmsg_level != SOL_SOCKET ||
962 			    (cm->cmsg_type != SCM_DEVMEM_DMABUF &&
963 			     cm->cmsg_type != SCM_DEVMEM_LINEAR)) {
964 				fprintf(stderr, "skipping non-devmem cmsg\n");
965 				continue;
966 			}
967 
968 			dmabuf_cmsg = (struct dmabuf_cmsg *)CMSG_DATA(cm);
969 			is_devmem = true;
970 
971 			if (cm->cmsg_type == SCM_DEVMEM_LINEAR) {
972 				/* TODO: process data copied from skb's linear
973 				 * buffer.
974 				 */
975 				fprintf(stderr,
976 					"SCM_DEVMEM_LINEAR. dmabuf_cmsg->frag_size=%u\n",
977 					dmabuf_cmsg->frag_size);
978 
979 				if (fail_on_linear) {
980 					pr_err("received SCM_DEVMEM_LINEAR but --fail-on-linear (-L) set");
981 					goto err_close_client;
982 				}
983 
984 				continue;
985 			}
986 
987 			token.token_start = dmabuf_cmsg->frag_token;
988 			token.token_count = 1;
989 
990 			total_received += dmabuf_cmsg->frag_size;
991 			fprintf(stderr,
992 				"received frag_page=%llu, in_page_offset=%llu, frag_offset=%llu, frag_size=%u, token=%u, total_received=%lu, dmabuf_id=%u\n",
993 				dmabuf_cmsg->frag_offset >> PAGE_SHIFT,
994 				dmabuf_cmsg->frag_offset % getpagesize(),
995 				dmabuf_cmsg->frag_offset,
996 				dmabuf_cmsg->frag_size, dmabuf_cmsg->frag_token,
997 				total_received, dmabuf_cmsg->dmabuf_id);
998 
999 			if (dmabuf_cmsg->dmabuf_id != dmabuf_id) {
1000 				pr_err("received on wrong dmabuf_id: flow steering error");
1001 				goto err_close_client;
1002 			}
1003 
1004 			if (dmabuf_cmsg->frag_size % getpagesize())
1005 				non_page_aligned_frags++;
1006 			else
1007 				page_aligned_frags++;
1008 
1009 			provider->memcpy_from_device(tmp_mem, mem,
1010 						     dmabuf_cmsg->frag_offset,
1011 						     dmabuf_cmsg->frag_size);
1012 
1013 			if (do_validation) {
1014 				if (validate_buffer(tmp_mem,
1015 						    dmabuf_cmsg->frag_size))
1016 					goto err_close_client;
1017 			} else {
1018 				print_nonzero_bytes(tmp_mem,
1019 						    dmabuf_cmsg->frag_size);
1020 			}
1021 
1022 			ret = setsockopt(client_fd, SOL_SOCKET,
1023 					 SO_DEVMEM_DONTNEED, &token,
1024 					 sizeof(token));
1025 			if (ret != 1) {
1026 				pr_err("SO_DEVMEM_DONTNEED not enough tokens");
1027 				goto err_close_client;
1028 			}
1029 		}
1030 		if (!is_devmem) {
1031 			pr_err("flow steering error");
1032 			goto err_close_client;
1033 		}
1034 
1035 		fprintf(stderr, "total_received=%lu\n", total_received);
1036 	}
1037 
1038 	fprintf(stderr, "%s: ok\n", TEST_PREFIX);
1039 
1040 	fprintf(stderr, "page_aligned_frags=%lu, non_page_aligned_frags=%lu\n",
1041 		page_aligned_frags, non_page_aligned_frags);
1042 
1043 cleanup:
1044 	err = 0;
1045 
1046 err_close_client:
1047 	close(client_fd);
1048 err_close_socket:
1049 	close(socket_fd);
1050 err_free_tmp:
1051 	free(tmp_mem);
1052 err_unbind:
1053 	ynl_sock_destroy(ys);
1054 err_reset_flow_steering:
1055 	reset_flow_steering();
1056 err_reset_rss:
1057 	reset_rss();
1058 err_reset_headersplit:
1059 	restore_ring_config(ring_config);
1060 err_free_ring_config:
1061 	ethtool_rings_get_rsp_free(ring_config);
1062 	return err;
1063 }
1064 
run_devmem_tests(void)1065 int run_devmem_tests(void)
1066 {
1067 	struct ethtool_rings_get_rsp *ring_config;
1068 	struct netdev_queue_id *queues;
1069 	struct memory_buffer *mem;
1070 	struct ynl_sock *ys;
1071 	int err = -1;
1072 
1073 	mem = provider->alloc(getpagesize() * NUM_PAGES);
1074 	if (!mem) {
1075 		pr_err("Failed to allocate memory buffer");
1076 		return -1;
1077 	}
1078 
1079 	ring_config = get_ring_config();
1080 	if (!ring_config) {
1081 		pr_err("Failed to get current ring configuration");
1082 		goto err_free_mem;
1083 	}
1084 
1085 	/* Configure RSS to divert all traffic from our devmem queues */
1086 	if (configure_rss()) {
1087 		pr_err("rss error");
1088 		goto err_free_ring_config;
1089 	}
1090 
1091 	if (configure_headersplit(ring_config, 1)) {
1092 		pr_err("Failed to configure header split");
1093 		goto err_reset_rss;
1094 	}
1095 
1096 	queues = netdev_queue_id_alloc(num_queues);
1097 	if (!queues) {
1098 		pr_err("Failed to allocate empty queues array");
1099 		goto err_reset_headersplit;
1100 	}
1101 
1102 	if (!bind_rx_queue(ifindex, mem->fd, queues, num_queues, &ys)) {
1103 		pr_err("Binding empty queues array should have failed");
1104 		goto err_unbind;
1105 	}
1106 
1107 	if (configure_headersplit(ring_config, 0)) {
1108 		pr_err("Failed to configure header split");
1109 		goto err_reset_headersplit;
1110 	}
1111 
1112 	queues = create_queues();
1113 	if (!queues) {
1114 		pr_err("Failed to create queues");
1115 		goto err_reset_headersplit;
1116 	}
1117 
1118 	if (!bind_rx_queue(ifindex, mem->fd, queues, num_queues, &ys)) {
1119 		pr_err("Configure dmabuf with header split off should have failed");
1120 		goto err_unbind;
1121 	}
1122 
1123 	if (configure_headersplit(ring_config, 1)) {
1124 		pr_err("Failed to configure header split");
1125 		goto err_reset_headersplit;
1126 	}
1127 
1128 	queues = create_queues();
1129 	if (!queues) {
1130 		pr_err("Failed to create queues");
1131 		goto err_reset_headersplit;
1132 	}
1133 
1134 	if (bind_rx_queue(ifindex, mem->fd, queues, num_queues, &ys)) {
1135 		pr_err("Failed to bind");
1136 		goto err_reset_headersplit;
1137 	}
1138 
1139 	/* Deactivating a bound queue should not be legal */
1140 	if (!check_changing_channels(num_queues, num_queues)) {
1141 		pr_err("Deactivating a bound queue should be illegal");
1142 		goto err_unbind;
1143 	}
1144 
1145 	err = 0;
1146 	goto err_unbind;
1147 
1148 err_unbind:
1149 	ynl_sock_destroy(ys);
1150 err_reset_headersplit:
1151 	restore_ring_config(ring_config);
1152 err_reset_rss:
1153 	reset_rss();
1154 err_free_ring_config:
1155 	ethtool_rings_get_rsp_free(ring_config);
1156 err_free_mem:
1157 	provider->free(mem);
1158 	return err;
1159 }
1160 
gettimeofday_ms(void)1161 static uint64_t gettimeofday_ms(void)
1162 {
1163 	struct timeval tv;
1164 
1165 	gettimeofday(&tv, NULL);
1166 	return (tv.tv_sec * 1000ULL) + (tv.tv_usec / 1000ULL);
1167 }
1168 
do_poll(int fd)1169 static int do_poll(int fd)
1170 {
1171 	struct pollfd pfd;
1172 	int ret;
1173 
1174 	pfd.revents = 0;
1175 	pfd.fd = fd;
1176 
1177 	ret = poll(&pfd, 1, waittime_ms);
1178 	if (ret == -1) {
1179 		pr_err("poll");
1180 		return -1;
1181 	}
1182 
1183 	return ret && (pfd.revents & POLLERR);
1184 }
1185 
wait_compl(int fd)1186 static int wait_compl(int fd)
1187 {
1188 	int64_t tstop = gettimeofday_ms() + waittime_ms;
1189 	char control[CMSG_SPACE(100)] = {};
1190 	struct sock_extended_err *serr;
1191 	struct msghdr msg = {};
1192 	struct cmsghdr *cm;
1193 	__u32 hi, lo;
1194 	int ret;
1195 
1196 	msg.msg_control = control;
1197 	msg.msg_controllen = sizeof(control);
1198 
1199 	while (gettimeofday_ms() < tstop) {
1200 		ret = do_poll(fd);
1201 		if (ret < 0)
1202 			return ret;
1203 		if (!ret)
1204 			continue;
1205 
1206 		ret = recvmsg(fd, &msg, MSG_ERRQUEUE);
1207 		if (ret < 0) {
1208 			if (errno == EAGAIN)
1209 				continue;
1210 			pr_err("recvmsg(MSG_ERRQUEUE)");
1211 			return -1;
1212 		}
1213 		if (msg.msg_flags & MSG_CTRUNC) {
1214 			pr_err("MSG_CTRUNC");
1215 			return -1;
1216 		}
1217 
1218 		for (cm = CMSG_FIRSTHDR(&msg); cm; cm = CMSG_NXTHDR(&msg, cm)) {
1219 			if (cm->cmsg_level != SOL_IP &&
1220 			    cm->cmsg_level != SOL_IPV6)
1221 				continue;
1222 			if (cm->cmsg_level == SOL_IP &&
1223 			    cm->cmsg_type != IP_RECVERR)
1224 				continue;
1225 			if (cm->cmsg_level == SOL_IPV6 &&
1226 			    cm->cmsg_type != IPV6_RECVERR)
1227 				continue;
1228 
1229 			serr = (void *)CMSG_DATA(cm);
1230 			if (serr->ee_origin != SO_EE_ORIGIN_ZEROCOPY) {
1231 				pr_err("wrong origin %u", serr->ee_origin);
1232 				return -1;
1233 			}
1234 			if (serr->ee_errno != 0) {
1235 				pr_err("wrong errno %d", serr->ee_errno);
1236 				return -1;
1237 			}
1238 
1239 			hi = serr->ee_data;
1240 			lo = serr->ee_info;
1241 
1242 			fprintf(stderr, "tx complete [%d,%d]\n", lo, hi);
1243 			return 0;
1244 		}
1245 	}
1246 
1247 	pr_err("did not receive tx completion");
1248 	return -1;
1249 }
1250 
do_client(struct memory_buffer * mem)1251 static int do_client(struct memory_buffer *mem)
1252 {
1253 	char ctrl_data[CMSG_SPACE(sizeof(__u32))];
1254 	struct sockaddr_in6 server_sin;
1255 	struct sockaddr_in6 client_sin;
1256 	struct ynl_sock *ys = NULL;
1257 	struct iovec iov[MAX_IOV];
1258 	struct msghdr msg = {};
1259 	ssize_t line_size = 0;
1260 	struct cmsghdr *cmsg;
1261 	char *line = NULL;
1262 	int ret, err = -1;
1263 	size_t len = 0;
1264 	int socket_fd;
1265 	__u32 ddmabuf;
1266 	int opt = 1;
1267 
1268 	ret = parse_address(server_ip, atoi(port), &server_sin);
1269 	if (ret < 0) {
1270 		pr_err("parse server address");
1271 		return -1;
1272 	}
1273 
1274 	if (client_ip) {
1275 		ret = parse_address(client_ip, atoi(port), &client_sin);
1276 		if (ret < 0) {
1277 			pr_err("parse client address");
1278 			return ret;
1279 		}
1280 	}
1281 
1282 	socket_fd = socket(AF_INET6, SOCK_STREAM, 0);
1283 	if (socket_fd < 0) {
1284 		pr_err("create socket");
1285 		return -1;
1286 	}
1287 
1288 	if (enable_reuseaddr(socket_fd))
1289 		goto err_close_socket;
1290 
1291 	ret = setsockopt(socket_fd, SOL_SOCKET, SO_BINDTODEVICE, ifname,
1292 			 strlen(ifname) + 1);
1293 	if (ret) {
1294 		pr_err("bindtodevice");
1295 		goto err_close_socket;
1296 	}
1297 
1298 	if (bind_tx_queue(ifindex, mem->fd, &ys)) {
1299 		pr_err("Failed to bind");
1300 		goto err_close_socket;
1301 	}
1302 
1303 	if (client_ip) {
1304 		ret = bind(socket_fd, &client_sin, sizeof(client_sin));
1305 		if (ret) {
1306 			pr_err("bind");
1307 			goto err_unbind;
1308 		}
1309 	}
1310 
1311 	ret = setsockopt(socket_fd, SOL_SOCKET, SO_ZEROCOPY, &opt, sizeof(opt));
1312 	if (ret) {
1313 		pr_err("set sock opt");
1314 		goto err_unbind;
1315 	}
1316 
1317 	fprintf(stderr, "Connect to %s %d (via %s)\n", server_ip,
1318 		ntohs(server_sin.sin6_port), ifname);
1319 
1320 	ret = connect(socket_fd, &server_sin, sizeof(server_sin));
1321 	if (ret) {
1322 		pr_err("connect");
1323 		goto err_unbind;
1324 	}
1325 
1326 	while (1) {
1327 		free(line);
1328 		line = NULL;
1329 		line_size = getline(&line, &len, stdin);
1330 
1331 		if (line_size < 0)
1332 			break;
1333 
1334 		if (max_chunk) {
1335 			msg.msg_iovlen =
1336 				(line_size + max_chunk - 1) / max_chunk;
1337 			if (msg.msg_iovlen > MAX_IOV) {
1338 				pr_err("can't partition %zd bytes into maximum of %d chunks",
1339 				       line_size, MAX_IOV);
1340 				goto err_free_line;
1341 			}
1342 
1343 			for (int i = 0; i < msg.msg_iovlen; i++) {
1344 				iov[i].iov_base = (void *)(i * max_chunk);
1345 				iov[i].iov_len = max_chunk;
1346 			}
1347 
1348 			iov[msg.msg_iovlen - 1].iov_len =
1349 				line_size - (msg.msg_iovlen - 1) * max_chunk;
1350 		} else {
1351 			iov[0].iov_base = 0;
1352 			iov[0].iov_len = line_size;
1353 			msg.msg_iovlen = 1;
1354 		}
1355 
1356 		msg.msg_iov = iov;
1357 		provider->memcpy_to_device(mem, 0, line, line_size);
1358 
1359 		msg.msg_control = ctrl_data;
1360 		msg.msg_controllen = sizeof(ctrl_data);
1361 
1362 		cmsg = CMSG_FIRSTHDR(&msg);
1363 		cmsg->cmsg_level = SOL_SOCKET;
1364 		cmsg->cmsg_type = SCM_DEVMEM_DMABUF;
1365 		cmsg->cmsg_len = CMSG_LEN(sizeof(__u32));
1366 
1367 		ddmabuf = tx_dmabuf_id;
1368 
1369 		*((__u32 *)CMSG_DATA(cmsg)) = ddmabuf;
1370 
1371 		ret = sendmsg(socket_fd, &msg, MSG_ZEROCOPY);
1372 		if (ret < 0) {
1373 			pr_err("Failed sendmsg");
1374 			goto err_free_line;
1375 		}
1376 
1377 		fprintf(stderr, "sendmsg_ret=%d\n", ret);
1378 
1379 		if (ret != line_size) {
1380 			pr_err("Did not send all bytes %d vs %zd", ret, line_size);
1381 			goto err_free_line;
1382 		}
1383 
1384 		if (wait_compl(socket_fd))
1385 			goto err_free_line;
1386 	}
1387 
1388 	fprintf(stderr, "%s: tx ok\n", TEST_PREFIX);
1389 
1390 	err = 0;
1391 
1392 err_free_line:
1393 	free(line);
1394 err_unbind:
1395 	ynl_sock_destroy(ys);
1396 err_close_socket:
1397 	close(socket_fd);
1398 	return err;
1399 }
1400 
main(int argc,char * argv[])1401 int main(int argc, char *argv[])
1402 {
1403 	struct memory_buffer *mem;
1404 	int is_server = 0, opt;
1405 	int ret, err = 1;
1406 
1407 	while ((opt = getopt(argc, argv, "Lls:c:p:v:q:t:f:z:")) != -1) {
1408 		switch (opt) {
1409 		case 'L':
1410 			fail_on_linear = true;
1411 			break;
1412 		case 'l':
1413 			is_server = 1;
1414 			break;
1415 		case 's':
1416 			server_ip = optarg;
1417 			break;
1418 		case 'c':
1419 			client_ip = optarg;
1420 			break;
1421 		case 'p':
1422 			port = optarg;
1423 			break;
1424 		case 'v':
1425 			do_validation = atoll(optarg);
1426 			break;
1427 		case 'q':
1428 			num_queues = atoi(optarg);
1429 			break;
1430 		case 't':
1431 			start_queue = atoi(optarg);
1432 			break;
1433 		case 'f':
1434 			ifname = optarg;
1435 			break;
1436 		case 'z':
1437 			max_chunk = atoi(optarg);
1438 			break;
1439 		case '?':
1440 			fprintf(stderr, "unknown option: %c\n", optopt);
1441 			break;
1442 		}
1443 	}
1444 
1445 	if (!ifname) {
1446 		pr_err("Missing -f argument");
1447 		return 1;
1448 	}
1449 
1450 	ifindex = if_nametoindex(ifname);
1451 
1452 	fprintf(stderr, "using ifindex=%u\n", ifindex);
1453 
1454 	if (!server_ip && !client_ip) {
1455 		if (start_queue < 0 && num_queues < 0) {
1456 			num_queues = rxq_num(ifindex);
1457 			if (num_queues < 0) {
1458 				pr_err("couldn't detect number of queues");
1459 				return 1;
1460 			}
1461 			if (num_queues < 2) {
1462 				pr_err("number of device queues is too low");
1463 				return 1;
1464 			}
1465 			/* make sure can bind to multiple queues */
1466 			start_queue = num_queues / 2;
1467 			num_queues /= 2;
1468 		}
1469 
1470 		if (start_queue < 0 || num_queues < 0) {
1471 			pr_err("Both -t and -q are required");
1472 			return 1;
1473 		}
1474 
1475 		return run_devmem_tests();
1476 	}
1477 
1478 	if (start_queue < 0 && num_queues < 0) {
1479 		num_queues = rxq_num(ifindex);
1480 		if (num_queues < 2) {
1481 			pr_err("number of device queues is too low");
1482 			return 1;
1483 		}
1484 
1485 		num_queues = 1;
1486 		start_queue = rxq_num(ifindex) - num_queues;
1487 
1488 		if (start_queue < 0) {
1489 			pr_err("couldn't detect number of queues");
1490 			return 1;
1491 		}
1492 
1493 		fprintf(stderr, "using queues %d..%d\n", start_queue, start_queue + num_queues);
1494 	}
1495 
1496 	for (; optind < argc; optind++)
1497 		fprintf(stderr, "extra arguments: %s\n", argv[optind]);
1498 
1499 	if (start_queue < 0) {
1500 		pr_err("Missing -t argument");
1501 		return 1;
1502 	}
1503 
1504 	if (num_queues < 0) {
1505 		pr_err("Missing -q argument");
1506 		return 1;
1507 	}
1508 
1509 	if (!server_ip) {
1510 		pr_err("Missing -s argument");
1511 		return 1;
1512 	}
1513 
1514 	if (!port) {
1515 		pr_err("Missing -p argument");
1516 		return 1;
1517 	}
1518 
1519 	mem = provider->alloc(getpagesize() * NUM_PAGES);
1520 	if (!mem) {
1521 		pr_err("Failed to allocate memory buffer");
1522 		return 1;
1523 	}
1524 
1525 	ret = is_server ? do_server(mem) : do_client(mem);
1526 	if (ret)
1527 		goto err_free_mem;
1528 
1529 	err = 0;
1530 
1531 err_free_mem:
1532 	provider->free(mem);
1533 	return err;
1534 }
1535