1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2017 Intel Corporation
3 */
4
5 #include <stdint.h>
6 #include <stdio.h>
7 #include <string.h>
8 #include <sys/queue.h>
9 #include <errno.h>
10 #include <stdarg.h>
11 #include <inttypes.h>
12
13 #include <rte_common.h>
14 #include <rte_memory.h>
15 #include <rte_memzone.h>
16 #include <rte_eal.h>
17 #include <rte_byteorder.h>
18 #include <rte_atomic.h>
19 #include <rte_launch.h>
20 #include <rte_per_lcore.h>
21 #include <rte_lcore.h>
22 #include <rte_branch_prediction.h>
23 #include <rte_debug.h>
24 #include <rte_ring.h>
25 #include <rte_log.h>
26 #include <rte_mempool.h>
27 #include <rte_memcpy.h>
28 #include <rte_mbuf.h>
29 #include <rte_interrupts.h>
30 #include <rte_ether.h>
31 #include <rte_ethdev.h>
32 #include <rte_malloc.h>
33 #include <rte_string_fns.h>
34 #include <rte_cycles.h>
35 #include <rte_efd.h>
36 #include <rte_hash.h>
37
38 #include "common.h"
39 #include "args.h"
40 #include "init.h"
41
42 #define MBUFS_PER_NODE 1536
43 #define MBUFS_PER_PORT 1536
44 #define MBUF_CACHE_SIZE 512
45
46 #define RTE_MP_RX_DESC_DEFAULT 512
47 #define RTE_MP_TX_DESC_DEFAULT 512
48 #define NODE_QUEUE_RINGSIZE 128
49
50 #define NO_FLAGS 0
51
52 /* The mbuf pool for packet rx */
53 struct rte_mempool *pktmbuf_pool;
54
55 /* array of info/queues for nodes */
56 struct node *nodes;
57
58 /* EFD table */
59 struct rte_efd_table *efd_table;
60
61 /* Shared info between server and nodes */
62 struct shared_info *info;
63
64 /**
65 * Initialise the mbuf pool for packet reception for the NIC, and any other
66 * buffer pools needed by the app - currently none.
67 */
68 static int
init_mbuf_pools(void)69 init_mbuf_pools(void)
70 {
71 const unsigned int num_mbufs = (num_nodes * MBUFS_PER_NODE) +
72 (info->num_ports * MBUFS_PER_PORT);
73
74 /*
75 * Don't pass single-producer/single-consumer flags to mbuf create as it
76 * seems faster to use a cache instead
77 */
78 printf("Creating mbuf pool '%s' [%u mbufs] ...\n",
79 PKTMBUF_POOL_NAME, num_mbufs);
80 pktmbuf_pool = rte_pktmbuf_pool_create(PKTMBUF_POOL_NAME, num_mbufs,
81 MBUF_CACHE_SIZE, 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
82
83 return pktmbuf_pool == NULL; /* 0 on success */
84 }
85
86 /**
87 * Initialise an individual port:
88 * - configure number of rx and tx rings
89 * - set up each rx ring, to pull from the main mbuf pool
90 * - set up each tx ring
91 * - start the port and report its status to stdout
92 */
93 static int
init_port(uint16_t port_num)94 init_port(uint16_t port_num)
95 {
96 /* for port configuration all features are off by default */
97 struct rte_eth_conf port_conf = {
98 .rxmode = {
99 .mq_mode = ETH_MQ_RX_RSS,
100 },
101 };
102 const uint16_t rx_rings = 1, tx_rings = num_nodes;
103 uint16_t rx_ring_size = RTE_MP_RX_DESC_DEFAULT;
104 uint16_t tx_ring_size = RTE_MP_TX_DESC_DEFAULT;
105 struct rte_eth_dev_info dev_info;
106 struct rte_eth_txconf txconf;
107
108 uint16_t q;
109 int retval;
110
111 printf("Port %u init ... ", port_num);
112 fflush(stdout);
113
114 retval = rte_eth_dev_info_get(port_num, &dev_info);
115 if (retval != 0)
116 return retval;
117
118 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
119 port_conf.txmode.offloads |=
120 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
121
122 /*
123 * Standard DPDK port initialisation - config port, then set up
124 * rx and tx rings.
125 */
126 retval = rte_eth_dev_configure(port_num, rx_rings, tx_rings, &port_conf);
127 if (retval != 0)
128 return retval;
129
130 retval = rte_eth_dev_adjust_nb_rx_tx_desc(port_num, &rx_ring_size,
131 &tx_ring_size);
132 if (retval != 0)
133 return retval;
134
135 for (q = 0; q < rx_rings; q++) {
136 retval = rte_eth_rx_queue_setup(port_num, q, rx_ring_size,
137 rte_eth_dev_socket_id(port_num),
138 NULL, pktmbuf_pool);
139 if (retval < 0)
140 return retval;
141 }
142
143 txconf = dev_info.default_txconf;
144 txconf.offloads = port_conf.txmode.offloads;
145 for (q = 0; q < tx_rings; q++) {
146 retval = rte_eth_tx_queue_setup(port_num, q, tx_ring_size,
147 rte_eth_dev_socket_id(port_num),
148 &txconf);
149 if (retval < 0)
150 return retval;
151 }
152
153 retval = rte_eth_promiscuous_enable(port_num);
154 if (retval != 0)
155 return retval;
156
157 retval = rte_eth_dev_start(port_num);
158 if (retval < 0)
159 return retval;
160
161 printf("done:\n");
162
163 return 0;
164 }
165
166 /**
167 * Set up the DPDK rings which will be used to pass packets, via
168 * pointers, between the multi-process server and node processes.
169 * Each node needs one RX queue.
170 */
171 static int
init_shm_rings(void)172 init_shm_rings(void)
173 {
174 unsigned int i;
175 unsigned int socket_id;
176 const char *q_name;
177 const unsigned int ringsize = NODE_QUEUE_RINGSIZE;
178
179 nodes = rte_malloc("node details",
180 sizeof(*nodes) * num_nodes, 0);
181 if (nodes == NULL)
182 rte_exit(EXIT_FAILURE, "Cannot allocate memory for "
183 "node program details\n");
184
185 for (i = 0; i < num_nodes; i++) {
186 /* Create an RX queue for each node */
187 socket_id = rte_socket_id();
188 q_name = get_rx_queue_name(i);
189 nodes[i].rx_q = rte_ring_create(q_name,
190 ringsize, socket_id,
191 RING_F_SP_ENQ | RING_F_SC_DEQ);
192 if (nodes[i].rx_q == NULL)
193 rte_exit(EXIT_FAILURE, "Cannot create rx ring queue "
194 "for node %u\n", i);
195 }
196 return 0;
197 }
198
199 /*
200 * Create EFD table which will contain all the flows
201 * that will be distributed among the nodes
202 */
203 static void
create_efd_table(void)204 create_efd_table(void)
205 {
206 uint8_t socket_id = rte_socket_id();
207
208 /* create table */
209 efd_table = rte_efd_create("flow table", num_flows * 2, sizeof(uint32_t),
210 1 << socket_id, socket_id);
211
212 if (efd_table == NULL)
213 rte_exit(EXIT_FAILURE, "Problem creating the flow table\n");
214 }
215
216 static void
populate_efd_table(void)217 populate_efd_table(void)
218 {
219 unsigned int i;
220 int32_t ret;
221 uint32_t ip_dst;
222 uint8_t socket_id = rte_socket_id();
223 uint64_t node_id;
224
225 /* Add flows in table */
226 for (i = 0; i < num_flows; i++) {
227 node_id = i % num_nodes;
228
229 ip_dst = rte_cpu_to_be_32(i);
230 ret = rte_efd_update(efd_table, socket_id,
231 (void *)&ip_dst, (efd_value_t)node_id);
232 if (ret < 0)
233 rte_exit(EXIT_FAILURE, "Unable to add entry %u in "
234 "EFD table\n", i);
235 }
236
237 printf("EFD table: Adding 0x%x keys\n", num_flows);
238 }
239
240 /* Check the link status of all ports in up to 9s, and print them finally */
241 static void
check_all_ports_link_status(uint16_t port_num,uint32_t port_mask)242 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
243 {
244 #define CHECK_INTERVAL 100 /* 100ms */
245 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
246 uint8_t count, all_ports_up, print_flag = 0;
247 uint16_t portid;
248 struct rte_eth_link link;
249 int ret;
250 char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
251
252 printf("\nChecking link status");
253 fflush(stdout);
254 for (count = 0; count <= MAX_CHECK_TIME; count++) {
255 all_ports_up = 1;
256 for (portid = 0; portid < port_num; portid++) {
257 if ((port_mask & (1 << info->id[portid])) == 0)
258 continue;
259 memset(&link, 0, sizeof(link));
260 ret = rte_eth_link_get_nowait(info->id[portid], &link);
261 if (ret < 0) {
262 all_ports_up = 0;
263 if (print_flag == 1)
264 printf("Port %u link get failed: %s\n",
265 portid, rte_strerror(-ret));
266 continue;
267 }
268 /* print link status if flag set */
269 if (print_flag == 1) {
270 rte_eth_link_to_str(link_status_text,
271 sizeof(link_status_text), &link);
272 printf("Port %d %s\n", info->id[portid],
273 link_status_text);
274 continue;
275 }
276 /* clear all_ports_up flag if any link down */
277 if (link.link_status == ETH_LINK_DOWN) {
278 all_ports_up = 0;
279 break;
280 }
281 }
282 /* after finally printing all link status, get out */
283 if (print_flag == 1)
284 break;
285
286 if (all_ports_up == 0) {
287 printf(".");
288 fflush(stdout);
289 rte_delay_ms(CHECK_INTERVAL);
290 }
291
292 /* set the print_flag if all ports up or timeout */
293 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
294 print_flag = 1;
295 printf("done\n");
296 }
297 }
298 }
299
300 /**
301 * Main init function for the multi-process server app,
302 * calls subfunctions to do each stage of the initialisation.
303 */
304 int
init(int argc,char * argv[])305 init(int argc, char *argv[])
306 {
307 int retval;
308 const struct rte_memzone *mz;
309 uint8_t i, total_ports;
310
311 /* init EAL, parsing EAL args */
312 retval = rte_eal_init(argc, argv);
313 if (retval < 0)
314 return -1;
315 argc -= retval;
316 argv += retval;
317
318 /* get total number of ports */
319 total_ports = rte_eth_dev_count_avail();
320
321 /* set up array for port data */
322 mz = rte_memzone_reserve(MZ_SHARED_INFO, sizeof(*info),
323 rte_socket_id(), NO_FLAGS);
324 if (mz == NULL)
325 rte_exit(EXIT_FAILURE, "Cannot reserve memory zone "
326 "for port information\n");
327 memset(mz->addr, 0, sizeof(*info));
328 info = mz->addr;
329
330 /* parse additional, application arguments */
331 retval = parse_app_args(total_ports, argc, argv);
332 if (retval != 0)
333 return -1;
334
335 /* initialise mbuf pools */
336 retval = init_mbuf_pools();
337 if (retval != 0)
338 rte_exit(EXIT_FAILURE, "Cannot create needed mbuf pools\n");
339
340 /* now initialise the ports we will use */
341 for (i = 0; i < info->num_ports; i++) {
342 retval = init_port(info->id[i]);
343 if (retval != 0)
344 rte_exit(EXIT_FAILURE, "Cannot initialise port %u\n",
345 (unsigned int) i);
346 }
347
348 check_all_ports_link_status(info->num_ports, (~0x0));
349
350 /* initialise the node queues/rings for inter-eu comms */
351 init_shm_rings();
352
353 /* Create the EFD table */
354 create_efd_table();
355
356 /* Populate the EFD table */
357 populate_efd_table();
358
359 /* Share the total number of nodes */
360 info->num_nodes = num_nodes;
361
362 /* Share the total number of flows */
363 info->num_flows = num_flows;
364 return 0;
365 }
366