xref: /dpdk/examples/server_node_efd/server/main.c (revision 4c00cfdc)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2016-2017 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <string.h>
37 #include <unistd.h>
38 #include <stdint.h>
39 #include <stdarg.h>
40 #include <inttypes.h>
41 #include <sys/queue.h>
42 #include <errno.h>
43 #include <netinet/ip.h>
44 
45 #include <rte_common.h>
46 #include <rte_memory.h>
47 #include <rte_eal.h>
48 #include <rte_launch.h>
49 #include <rte_per_lcore.h>
50 #include <rte_lcore.h>
51 #include <rte_branch_prediction.h>
52 #include <rte_atomic.h>
53 #include <rte_ring.h>
54 #include <rte_log.h>
55 #include <rte_debug.h>
56 #include <rte_mempool.h>
57 #include <rte_memcpy.h>
58 #include <rte_mbuf.h>
59 #include <rte_ether.h>
60 #include <rte_interrupts.h>
61 #include <rte_pci.h>
62 #include <rte_ethdev.h>
63 #include <rte_byteorder.h>
64 #include <rte_malloc.h>
65 #include <rte_string_fns.h>
66 #include <rte_efd.h>
67 #include <rte_ip.h>
68 
69 #include "common.h"
70 #include "args.h"
71 #include "init.h"
72 
73 /*
74  * When doing reads from the NIC or the node queues,
75  * use this batch size
76  */
77 #define PACKET_READ_SIZE 32
78 
79 /*
80  * Local buffers to put packets in, used to send packets in bursts to the
81  * nodes
82  */
83 struct node_rx_buf {
84 	struct rte_mbuf *buffer[PACKET_READ_SIZE];
85 	uint16_t count;
86 };
87 
88 struct efd_stats {
89 	uint64_t distributed;
90 	uint64_t drop;
91 } flow_dist_stats;
92 
93 /* One buffer per node rx queue - dynamically allocate array */
94 static struct node_rx_buf *cl_rx_buf;
95 
96 static const char *
97 get_printable_mac_addr(uint16_t port)
98 {
99 	static const char err_address[] = "00:00:00:00:00:00";
100 	static char addresses[RTE_MAX_ETHPORTS][sizeof(err_address)];
101 	struct ether_addr mac;
102 
103 	if (unlikely(port >= RTE_MAX_ETHPORTS))
104 		return err_address;
105 	if (unlikely(addresses[port][0] == '\0')) {
106 		rte_eth_macaddr_get(port, &mac);
107 		snprintf(addresses[port], sizeof(addresses[port]),
108 				"%02x:%02x:%02x:%02x:%02x:%02x\n",
109 				mac.addr_bytes[0], mac.addr_bytes[1],
110 				mac.addr_bytes[2], mac.addr_bytes[3],
111 				mac.addr_bytes[4], mac.addr_bytes[5]);
112 	}
113 	return addresses[port];
114 }
115 
116 /*
117  * This function displays the recorded statistics for each port
118  * and for each node. It uses ANSI terminal codes to clear
119  * screen when called. It is called from a single non-master
120  * thread in the server process, when the process is run with more
121  * than one lcore enabled.
122  */
123 static void
124 do_stats_display(void)
125 {
126 	unsigned int i, j;
127 	const char clr[] = {27, '[', '2', 'J', '\0'};
128 	const char topLeft[] = {27, '[', '1', ';', '1', 'H', '\0'};
129 	uint64_t port_tx[RTE_MAX_ETHPORTS], port_tx_drop[RTE_MAX_ETHPORTS];
130 	uint64_t node_tx[MAX_NODES], node_tx_drop[MAX_NODES];
131 
132 	/* to get TX stats, we need to do some summing calculations */
133 	memset(port_tx, 0, sizeof(port_tx));
134 	memset(port_tx_drop, 0, sizeof(port_tx_drop));
135 	memset(node_tx, 0, sizeof(node_tx));
136 	memset(node_tx_drop, 0, sizeof(node_tx_drop));
137 
138 	for (i = 0; i < num_nodes; i++) {
139 		const struct tx_stats *tx = &info->tx_stats[i];
140 
141 		for (j = 0; j < info->num_ports; j++) {
142 			const uint64_t tx_val = tx->tx[info->id[j]];
143 			const uint64_t drop_val = tx->tx_drop[info->id[j]];
144 
145 			port_tx[j] += tx_val;
146 			port_tx_drop[j] += drop_val;
147 			node_tx[i] += tx_val;
148 			node_tx_drop[i] += drop_val;
149 		}
150 	}
151 
152 	/* Clear screen and move to top left */
153 	printf("%s%s", clr, topLeft);
154 
155 	printf("PORTS\n");
156 	printf("-----\n");
157 	for (i = 0; i < info->num_ports; i++)
158 		printf("Port %u: '%s'\t", (unsigned int)info->id[i],
159 				get_printable_mac_addr(info->id[i]));
160 	printf("\n\n");
161 	for (i = 0; i < info->num_ports; i++) {
162 		printf("Port %u - rx: %9"PRIu64"\t"
163 				"tx: %9"PRIu64"\n",
164 				(unsigned int)info->id[i], info->rx_stats.rx[i],
165 				port_tx[i]);
166 	}
167 
168 	printf("\nSERVER\n");
169 	printf("-----\n");
170 	printf("distributed: %9"PRIu64", drop: %9"PRIu64"\n",
171 			flow_dist_stats.distributed, flow_dist_stats.drop);
172 
173 	printf("\nNODES\n");
174 	printf("-------\n");
175 	for (i = 0; i < num_nodes; i++) {
176 		const unsigned long long rx = nodes[i].stats.rx;
177 		const unsigned long long rx_drop = nodes[i].stats.rx_drop;
178 		const struct filter_stats *filter = &info->filter_stats[i];
179 
180 		printf("Node %2u - rx: %9llu, rx_drop: %9llu\n"
181 				"            tx: %9"PRIu64", tx_drop: %9"PRIu64"\n"
182 				"            filter_passed: %9"PRIu64", "
183 				"filter_drop: %9"PRIu64"\n",
184 				i, rx, rx_drop, node_tx[i], node_tx_drop[i],
185 				filter->passed, filter->drop);
186 	}
187 
188 	printf("\n");
189 }
190 
191 /*
192  * The function called from each non-master lcore used by the process.
193  * The test_and_set function is used to randomly pick a single lcore on which
194  * the code to display the statistics will run. Otherwise, the code just
195  * repeatedly sleeps.
196  */
197 static int
198 sleep_lcore(__attribute__((unused)) void *dummy)
199 {
200 	/* Used to pick a display thread - static, so zero-initialised */
201 	static rte_atomic32_t display_stats;
202 
203 	/* Only one core should display stats */
204 	if (rte_atomic32_test_and_set(&display_stats)) {
205 		const unsigned int sleeptime = 1;
206 
207 		printf("Core %u displaying statistics\n", rte_lcore_id());
208 
209 		/* Longer initial pause so above printf is seen */
210 		sleep(sleeptime * 3);
211 
212 		/* Loop forever: sleep always returns 0 or <= param */
213 		while (sleep(sleeptime) <= sleeptime)
214 			do_stats_display();
215 	}
216 	return 0;
217 }
218 
219 /*
220  * Function to set all the node statistic values to zero.
221  * Called at program startup.
222  */
223 static void
224 clear_stats(void)
225 {
226 	unsigned int i;
227 
228 	for (i = 0; i < num_nodes; i++)
229 		nodes[i].stats.rx = nodes[i].stats.rx_drop = 0;
230 }
231 
232 /*
233  * send a burst of traffic to a node, assuming there are packets
234  * available to be sent to this node
235  */
236 static void
237 flush_rx_queue(uint16_t node)
238 {
239 	uint16_t j;
240 	struct node *cl;
241 
242 	if (cl_rx_buf[node].count == 0)
243 		return;
244 
245 	cl = &nodes[node];
246 	if (rte_ring_enqueue_bulk(cl->rx_q, (void **)cl_rx_buf[node].buffer,
247 			cl_rx_buf[node].count, NULL) != cl_rx_buf[node].count){
248 		for (j = 0; j < cl_rx_buf[node].count; j++)
249 			rte_pktmbuf_free(cl_rx_buf[node].buffer[j]);
250 		cl->stats.rx_drop += cl_rx_buf[node].count;
251 	} else
252 		cl->stats.rx += cl_rx_buf[node].count;
253 
254 	cl_rx_buf[node].count = 0;
255 }
256 
257 /*
258  * marks a packet down to be sent to a particular node process
259  */
260 static inline void
261 enqueue_rx_packet(uint8_t node, struct rte_mbuf *buf)
262 {
263 	cl_rx_buf[node].buffer[cl_rx_buf[node].count++] = buf;
264 }
265 
266 /*
267  * This function takes a group of packets and routes them
268  * individually to the node process. Very simply round-robins the packets
269  * without checking any of the packet contents.
270  */
271 static void
272 process_packets(uint32_t port_num __rte_unused, struct rte_mbuf *pkts[],
273 		uint16_t rx_count, unsigned int socket_id)
274 {
275 	uint16_t i;
276 	uint8_t node;
277 	efd_value_t data[RTE_EFD_BURST_MAX];
278 	const void *key_ptrs[RTE_EFD_BURST_MAX];
279 
280 	struct ipv4_hdr *ipv4_hdr;
281 	uint32_t ipv4_dst_ip[RTE_EFD_BURST_MAX];
282 
283 	for (i = 0; i < rx_count; i++) {
284 		/* Handle IPv4 header.*/
285 		ipv4_hdr = rte_pktmbuf_mtod_offset(pkts[i], struct ipv4_hdr *,
286 				sizeof(struct ether_hdr));
287 		ipv4_dst_ip[i] = ipv4_hdr->dst_addr;
288 		key_ptrs[i] = (void *)&ipv4_dst_ip[i];
289 	}
290 
291 	rte_efd_lookup_bulk(efd_table, socket_id, rx_count,
292 				(const void **) key_ptrs, data);
293 	for (i = 0; i < rx_count; i++) {
294 		node = (uint8_t) ((uintptr_t)data[i]);
295 
296 		if (node >= num_nodes) {
297 			/*
298 			 * Node is out of range, which means that
299 			 * flow has not been inserted
300 			 */
301 			flow_dist_stats.drop++;
302 			rte_pktmbuf_free(pkts[i]);
303 		} else {
304 			flow_dist_stats.distributed++;
305 			enqueue_rx_packet(node, pkts[i]);
306 		}
307 	}
308 
309 	for (i = 0; i < num_nodes; i++)
310 		flush_rx_queue(i);
311 }
312 
313 /*
314  * Function called by the master lcore of the DPDK process.
315  */
316 static void
317 do_packet_forwarding(void)
318 {
319 	unsigned int port_num = 0; /* indexes the port[] array */
320 	unsigned int socket_id = rte_socket_id();
321 
322 	for (;;) {
323 		struct rte_mbuf *buf[PACKET_READ_SIZE];
324 		uint16_t rx_count;
325 
326 		/* read a port */
327 		rx_count = rte_eth_rx_burst(info->id[port_num], 0,
328 				buf, PACKET_READ_SIZE);
329 		info->rx_stats.rx[port_num] += rx_count;
330 
331 		/* Now process the NIC packets read */
332 		if (likely(rx_count > 0))
333 			process_packets(port_num, buf, rx_count, socket_id);
334 
335 		/* move to next port */
336 		if (++port_num == info->num_ports)
337 			port_num = 0;
338 	}
339 }
340 
341 int
342 main(int argc, char *argv[])
343 {
344 	/* initialise the system */
345 	if (init(argc, argv) < 0)
346 		return -1;
347 	RTE_LOG(INFO, APP, "Finished Process Init.\n");
348 
349 	cl_rx_buf = calloc(num_nodes, sizeof(cl_rx_buf[0]));
350 
351 	/* clear statistics */
352 	clear_stats();
353 
354 	/* put all other cores to sleep bar master */
355 	rte_eal_mp_remote_launch(sleep_lcore, NULL, SKIP_MASTER);
356 
357 	do_packet_forwarding();
358 	return 0;
359 }
360