1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
3 */
4
5 #include <stdio.h>
6 #include <stdlib.h>
7 #include <stdint.h>
8 #include <inttypes.h>
9 #include <sys/types.h>
10 #include <string.h>
11 #include <sys/queue.h>
12 #include <stdarg.h>
13 #include <errno.h>
14 #include <getopt.h>
15 #include <stdbool.h>
16 #include <netinet/in.h>
17
18 #include <rte_debug.h>
19 #include <rte_ether.h>
20 #include <rte_ethdev.h>
21 #include <rte_cycles.h>
22 #include <rte_mbuf.h>
23 #include <rte_ip.h>
24 #include <rte_tcp.h>
25 #include <rte_udp.h>
26 #include <rte_hash.h>
27
28 #include "l3fwd.h"
29 #include "l3fwd_event.h"
30 #include "em_route_parse.c"
31
32 #if defined(RTE_ARCH_X86) || defined(__ARM_FEATURE_CRC32)
33 #define EM_HASH_CRC 1
34 #endif
35
36 #ifdef EM_HASH_CRC
37 #include <rte_hash_crc.h>
38 #define DEFAULT_HASH_FUNC rte_hash_crc
39 #else
40 #include <rte_jhash.h>
41 #define DEFAULT_HASH_FUNC rte_jhash
42 #endif
43
44 #define IPV6_ADDR_LEN 16
45
46 union ipv4_5tuple_host {
47 struct {
48 uint8_t pad0;
49 uint8_t proto;
50 uint16_t pad1;
51 uint32_t ip_src;
52 uint32_t ip_dst;
53 uint16_t port_src;
54 uint16_t port_dst;
55 };
56 xmm_t xmm;
57 };
58
59 #define XMM_NUM_IN_IPV6_5TUPLE 3
60
61 union ipv6_5tuple_host {
62 struct {
63 uint16_t pad0;
64 uint8_t proto;
65 uint8_t pad1;
66 uint8_t ip_src[IPV6_ADDR_LEN];
67 uint8_t ip_dst[IPV6_ADDR_LEN];
68 uint16_t port_src;
69 uint16_t port_dst;
70 uint64_t reserve;
71 };
72 xmm_t xmm[XMM_NUM_IN_IPV6_5TUPLE];
73 };
74
75 /* 198.18.0.0/16 are set aside for RFC2544 benchmarking (RFC5735).
76 * Use RFC863 Discard Protocol.
77 */
78 const struct ipv4_l3fwd_em_route ipv4_l3fwd_em_route_array[] = {
79 {{RTE_IPV4(198, 18, 0, 0), RTE_IPV4(198, 18, 0, 1), 9, 9, IPPROTO_UDP}, 0},
80 {{RTE_IPV4(198, 18, 1, 0), RTE_IPV4(198, 18, 1, 1), 9, 9, IPPROTO_UDP}, 1},
81 {{RTE_IPV4(198, 18, 2, 0), RTE_IPV4(198, 18, 2, 1), 9, 9, IPPROTO_UDP}, 2},
82 {{RTE_IPV4(198, 18, 3, 0), RTE_IPV4(198, 18, 3, 1), 9, 9, IPPROTO_UDP}, 3},
83 {{RTE_IPV4(198, 18, 4, 0), RTE_IPV4(198, 18, 4, 1), 9, 9, IPPROTO_UDP}, 4},
84 {{RTE_IPV4(198, 18, 5, 0), RTE_IPV4(198, 18, 5, 1), 9, 9, IPPROTO_UDP}, 5},
85 {{RTE_IPV4(198, 18, 6, 0), RTE_IPV4(198, 18, 6, 1), 9, 9, IPPROTO_UDP}, 6},
86 {{RTE_IPV4(198, 18, 7, 0), RTE_IPV4(198, 18, 7, 1), 9, 9, IPPROTO_UDP}, 7},
87 {{RTE_IPV4(198, 18, 8, 0), RTE_IPV4(198, 18, 8, 1), 9, 9, IPPROTO_UDP}, 8},
88 {{RTE_IPV4(198, 18, 9, 0), RTE_IPV4(198, 18, 9, 1), 9, 9, IPPROTO_UDP}, 9},
89 {{RTE_IPV4(198, 18, 10, 0), RTE_IPV4(198, 18, 10, 1), 9, 9, IPPROTO_UDP}, 10},
90 {{RTE_IPV4(198, 18, 11, 0), RTE_IPV4(198, 18, 11, 1), 9, 9, IPPROTO_UDP}, 11},
91 {{RTE_IPV4(198, 18, 12, 0), RTE_IPV4(198, 18, 12, 1), 9, 9, IPPROTO_UDP}, 12},
92 {{RTE_IPV4(198, 18, 13, 0), RTE_IPV4(198, 18, 13, 1), 9, 9, IPPROTO_UDP}, 13},
93 {{RTE_IPV4(198, 18, 14, 0), RTE_IPV4(198, 18, 14, 1), 9, 9, IPPROTO_UDP}, 14},
94 {{RTE_IPV4(198, 18, 15, 0), RTE_IPV4(198, 18, 15, 1), 9, 9, IPPROTO_UDP}, 15},
95 };
96
97 /* 2001:0200::/48 is IANA reserved range for IPv6 benchmarking (RFC5180).
98 * Use RFC863 Discard Protocol.
99 */
100 const struct ipv6_l3fwd_em_route ipv6_l3fwd_em_route_array[] = {
101 {{{32, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
102 {32, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 0},
103 {{{32, 1, 2, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0},
104 {32, 1, 2, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 1},
105 {{{32, 1, 2, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0},
106 {32, 1, 2, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 2},
107 {{{32, 1, 2, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 0},
108 {32, 1, 2, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 3},
109 {{{32, 1, 2, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 0},
110 {32, 1, 2, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 4},
111 {{{32, 1, 2, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0, 0, 0},
112 {32, 1, 2, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 5},
113 {{{32, 1, 2, 0, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0, 0, 0},
114 {32, 1, 2, 0, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 6},
115 {{{32, 1, 2, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 0},
116 {32, 1, 2, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 7},
117 {{{32, 1, 2, 0, 0, 0, 0, 8, 0, 0, 0, 0, 0, 0, 0, 0},
118 {32, 1, 2, 0, 0, 0, 0, 8, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 8},
119 {{{32, 1, 2, 0, 0, 0, 0, 9, 0, 0, 0, 0, 0, 0, 0, 0},
120 {32, 1, 2, 0, 0, 0, 0, 9, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 9},
121 {{{32, 1, 2, 0, 0, 0, 0, 10, 0, 0, 0, 0, 0, 0, 0, 0},
122 {32, 1, 2, 0, 0, 0, 0, 10, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 10},
123 {{{32, 1, 2, 0, 0, 0, 0, 11, 0, 0, 0, 0, 0, 0, 0, 0},
124 {32, 1, 2, 0, 0, 0, 0, 11, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 11},
125 {{{32, 1, 2, 0, 0, 0, 0, 12, 0, 0, 0, 0, 0, 0, 0, 0},
126 {32, 1, 2, 0, 0, 0, 0, 12, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 12},
127 {{{32, 1, 2, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0, 0, 0, 0},
128 {32, 1, 2, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 13},
129 {{{32, 1, 2, 0, 0, 0, 0, 14, 0, 0, 0, 0, 0, 0, 0, 0},
130 {32, 1, 2, 0, 0, 0, 0, 14, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 14},
131 {{{32, 1, 2, 0, 0, 0, 0, 15, 0, 0, 0, 0, 0, 0, 0, 0},
132 {32, 1, 2, 0, 0, 0, 0, 15, 0, 0, 0, 0, 0, 0, 0, 1}, 9, 9, IPPROTO_UDP}, 15},
133 };
134
135 struct rte_hash *ipv4_l3fwd_em_lookup_struct[NB_SOCKETS];
136 struct rte_hash *ipv6_l3fwd_em_lookup_struct[NB_SOCKETS];
137
138 static inline uint32_t
ipv4_hash_crc(const void * data,__rte_unused uint32_t data_len,uint32_t init_val)139 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
140 uint32_t init_val)
141 {
142 const union ipv4_5tuple_host *k;
143 uint32_t t;
144 const uint32_t *p;
145
146 k = data;
147 t = k->proto;
148 p = (const uint32_t *)&k->port_src;
149
150 #ifdef EM_HASH_CRC
151 init_val = rte_hash_crc_4byte(t, init_val);
152 init_val = rte_hash_crc_4byte(k->ip_src, init_val);
153 init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
154 init_val = rte_hash_crc_4byte(*p, init_val);
155 #else
156 init_val = rte_jhash_1word(t, init_val);
157 init_val = rte_jhash_1word(k->ip_src, init_val);
158 init_val = rte_jhash_1word(k->ip_dst, init_val);
159 init_val = rte_jhash_1word(*p, init_val);
160 #endif
161
162 return init_val;
163 }
164
165 static inline uint32_t
ipv6_hash_crc(const void * data,__rte_unused uint32_t data_len,uint32_t init_val)166 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
167 uint32_t init_val)
168 {
169 const union ipv6_5tuple_host *k;
170 uint32_t t;
171 const uint32_t *p;
172 #ifdef EM_HASH_CRC
173 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
174 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
175 #endif
176
177 k = data;
178 t = k->proto;
179 p = (const uint32_t *)&k->port_src;
180
181 #ifdef EM_HASH_CRC
182 ip_src0 = (const uint32_t *) k->ip_src;
183 ip_src1 = (const uint32_t *)(k->ip_src+4);
184 ip_src2 = (const uint32_t *)(k->ip_src+8);
185 ip_src3 = (const uint32_t *)(k->ip_src+12);
186 ip_dst0 = (const uint32_t *) k->ip_dst;
187 ip_dst1 = (const uint32_t *)(k->ip_dst+4);
188 ip_dst2 = (const uint32_t *)(k->ip_dst+8);
189 ip_dst3 = (const uint32_t *)(k->ip_dst+12);
190 init_val = rte_hash_crc_4byte(t, init_val);
191 init_val = rte_hash_crc_4byte(*ip_src0, init_val);
192 init_val = rte_hash_crc_4byte(*ip_src1, init_val);
193 init_val = rte_hash_crc_4byte(*ip_src2, init_val);
194 init_val = rte_hash_crc_4byte(*ip_src3, init_val);
195 init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
196 init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
197 init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
198 init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
199 init_val = rte_hash_crc_4byte(*p, init_val);
200 #else
201 init_val = rte_jhash_1word(t, init_val);
202 init_val = rte_jhash(k->ip_src,
203 sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
204 init_val = rte_jhash(k->ip_dst,
205 sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
206 init_val = rte_jhash_1word(*p, init_val);
207 #endif
208 return init_val;
209 }
210
211 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
212 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
213
214 static rte_xmm_t mask0;
215 static rte_xmm_t mask1;
216 static rte_xmm_t mask2;
217
218 #if defined(__SSE2__)
219 static inline xmm_t
em_mask_key(void * key,xmm_t mask)220 em_mask_key(void *key, xmm_t mask)
221 {
222 __m128i data = _mm_loadu_si128((__m128i *)(key));
223
224 return _mm_and_si128(data, mask);
225 }
226 #elif defined(__ARM_NEON)
227 static inline xmm_t
em_mask_key(void * key,xmm_t mask)228 em_mask_key(void *key, xmm_t mask)
229 {
230 int32x4_t data = vld1q_s32((int32_t *)key);
231
232 return vandq_s32(data, mask);
233 }
234 #elif defined(__ALTIVEC__)
235 static inline xmm_t
em_mask_key(void * key,xmm_t mask)236 em_mask_key(void *key, xmm_t mask)
237 {
238 xmm_t data = vec_ld(0, (xmm_t *)(key));
239
240 return vec_and(data, mask);
241 }
242 #else
243 #error No vector engine (SSE, NEON, ALTIVEC) available, check your toolchain
244 #endif
245
246 /* Performing hash-based lookups. 8< */
247 static inline uint16_t
em_get_ipv4_dst_port(void * ipv4_hdr,uint16_t portid,void * lookup_struct)248 em_get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid, void *lookup_struct)
249 {
250 int ret = 0;
251 union ipv4_5tuple_host key;
252 struct rte_hash *ipv4_l3fwd_lookup_struct =
253 (struct rte_hash *)lookup_struct;
254
255 ipv4_hdr = (uint8_t *)ipv4_hdr +
256 offsetof(struct rte_ipv4_hdr, time_to_live);
257
258 /*
259 * Get 5 tuple: dst port, src port, dst IP address,
260 * src IP address and protocol.
261 */
262 key.xmm = em_mask_key(ipv4_hdr, mask0.x);
263
264 /* Find destination port */
265 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
266 return (ret < 0) ? portid : ipv4_l3fwd_out_if[ret];
267 }
268 /* >8 End of performing hash-based lookups. */
269
270 static inline uint16_t
em_get_ipv6_dst_port(void * ipv6_hdr,uint16_t portid,void * lookup_struct)271 em_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)
272 {
273 int ret = 0;
274 union ipv6_5tuple_host key;
275 struct rte_hash *ipv6_l3fwd_lookup_struct =
276 (struct rte_hash *)lookup_struct;
277
278 ipv6_hdr = (uint8_t *)ipv6_hdr +
279 offsetof(struct rte_ipv6_hdr, payload_len);
280 void *data0 = ipv6_hdr;
281 void *data1 = ((uint8_t *)ipv6_hdr) + sizeof(xmm_t);
282 void *data2 = ((uint8_t *)ipv6_hdr) + sizeof(xmm_t) + sizeof(xmm_t);
283
284 /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
285 key.xmm[0] = em_mask_key(data0, mask1.x);
286
287 /*
288 * Get part of 5 tuple: dst IP address lower 96 bits
289 * and src IP address higher 32 bits.
290 */
291 #if defined RTE_ARCH_X86
292 key.xmm[1] = _mm_loadu_si128(data1);
293 #else
294 key.xmm[1] = *(xmm_t *)data1;
295 #endif
296
297 /*
298 * Get part of 5 tuple: dst port and src port
299 * and dst IP address higher 32 bits.
300 */
301 key.xmm[2] = em_mask_key(data2, mask2.x);
302
303 /* Find destination port */
304 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
305 return (ret < 0) ? portid : ipv6_l3fwd_out_if[ret];
306 }
307
308 #if defined RTE_ARCH_X86 || defined __ARM_NEON
309 #if defined(NO_HASH_MULTI_LOOKUP)
310 #include "l3fwd_em_sequential.h"
311 #else
312 #include "l3fwd_em_hlm.h"
313 #endif
314 #else
315 #include "l3fwd_em.h"
316 #endif
317
318 static void
convert_ipv4_5tuple(struct ipv4_5tuple * key1,union ipv4_5tuple_host * key2)319 convert_ipv4_5tuple(struct ipv4_5tuple *key1,
320 union ipv4_5tuple_host *key2)
321 {
322 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
323 key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
324 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
325 key2->port_src = rte_cpu_to_be_16(key1->port_src);
326 key2->proto = key1->proto;
327 key2->pad0 = 0;
328 key2->pad1 = 0;
329 }
330
331 static void
convert_ipv6_5tuple(struct ipv6_5tuple * key1,union ipv6_5tuple_host * key2)332 convert_ipv6_5tuple(struct ipv6_5tuple *key1,
333 union ipv6_5tuple_host *key2)
334 {
335 uint32_t i;
336
337 for (i = 0; i < 16; i++) {
338 key2->ip_dst[i] = key1->ip_dst[i];
339 key2->ip_src[i] = key1->ip_src[i];
340 }
341 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
342 key2->port_src = rte_cpu_to_be_16(key1->port_src);
343 key2->proto = key1->proto;
344 key2->pad0 = 0;
345 key2->pad1 = 0;
346 key2->reserve = 0;
347 }
348
349 #define BYTE_VALUE_MAX 256
350 #define ALL_32_BITS 0xffffffff
351 #define BIT_8_TO_15 0x0000ff00
352
353 static inline void
populate_ipv4_flow_into_table(const struct rte_hash * h)354 populate_ipv4_flow_into_table(const struct rte_hash *h)
355 {
356 int i;
357 int32_t ret;
358 struct rte_eth_dev_info dev_info;
359 char srcbuf[INET6_ADDRSTRLEN];
360 char dstbuf[INET6_ADDRSTRLEN];
361
362 mask0 = (rte_xmm_t){.u32 = {BIT_8_TO_15, ALL_32_BITS,
363 ALL_32_BITS, ALL_32_BITS} };
364
365 for (i = 0; i < route_num_v4; i++) {
366 struct em_rule *entry;
367 union ipv4_5tuple_host newkey;
368 struct in_addr src;
369 struct in_addr dst;
370
371 if ((1 << em_route_base_v4[i].if_out &
372 enabled_port_mask) == 0)
373 continue;
374
375 entry = &em_route_base_v4[i];
376 convert_ipv4_5tuple(&(entry->v4_key), &newkey);
377 ret = rte_hash_add_key(h, (void *) &newkey);
378 if (ret < 0) {
379 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
380 " to the l3fwd hash.\n", i);
381 }
382 ipv4_l3fwd_out_if[ret] = entry->if_out;
383 ret = rte_eth_dev_info_get(em_route_base_v4[i].if_out,
384 &dev_info);
385 if (ret != 0)
386 rte_exit(EXIT_FAILURE,
387 "Error during getting device (port %u) info: %s\n",
388 em_route_base_v4[i].if_out, strerror(-ret));
389
390 src.s_addr = htonl(em_route_base_v4[i].v4_key.ip_src);
391 dst.s_addr = htonl(em_route_base_v4[i].v4_key.ip_dst);
392 printf("EM: Adding route %s, %s, %d, %d, %d (%d) [%s]\n",
393 inet_ntop(AF_INET, &dst, dstbuf, sizeof(dstbuf)),
394 inet_ntop(AF_INET, &src, srcbuf, sizeof(srcbuf)),
395 em_route_base_v4[i].v4_key.port_dst,
396 em_route_base_v4[i].v4_key.port_src,
397 em_route_base_v4[i].v4_key.proto,
398 em_route_base_v4[i].if_out, dev_info.device->name);
399 }
400 printf("Hash: Adding 0x%" PRIx64 " keys\n",
401 (uint64_t)route_num_v4);
402 }
403
404 #define BIT_16_TO_23 0x00ff0000
405 static inline void
populate_ipv6_flow_into_table(const struct rte_hash * h)406 populate_ipv6_flow_into_table(const struct rte_hash *h)
407 {
408 int i;
409 int32_t ret;
410 struct rte_eth_dev_info dev_info;
411 char srcbuf[INET6_ADDRSTRLEN];
412 char dstbuf[INET6_ADDRSTRLEN];
413
414 mask1 = (rte_xmm_t){.u32 = {BIT_16_TO_23, ALL_32_BITS,
415 ALL_32_BITS, ALL_32_BITS} };
416
417 mask2 = (rte_xmm_t){.u32 = {ALL_32_BITS, ALL_32_BITS, 0, 0} };
418
419 for (i = 0; i < route_num_v6; i++) {
420 struct em_rule *entry;
421 union ipv6_5tuple_host newkey;
422
423 if ((1 << em_route_base_v6[i].if_out &
424 enabled_port_mask) == 0)
425 continue;
426
427 entry = &em_route_base_v6[i];
428 convert_ipv6_5tuple(&(entry->v6_key), &newkey);
429 ret = rte_hash_add_key(h, (void *) &newkey);
430 if (ret < 0) {
431 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
432 " to the l3fwd hash.\n", i);
433 }
434 ipv6_l3fwd_out_if[ret] = entry->if_out;
435 ret = rte_eth_dev_info_get(em_route_base_v6[i].if_out,
436 &dev_info);
437 if (ret != 0)
438 rte_exit(EXIT_FAILURE,
439 "Error during getting device (port %u) info: %s\n",
440 em_route_base_v6[i].if_out, strerror(-ret));
441
442 printf("EM: Adding route %s, %s, %d, %d, %d (%d) [%s]\n",
443 inet_ntop(AF_INET6, em_route_base_v6[i].v6_key.ip_dst,
444 dstbuf, sizeof(dstbuf)),
445 inet_ntop(AF_INET6, em_route_base_v6[i].v6_key.ip_src,
446 srcbuf, sizeof(srcbuf)),
447 em_route_base_v6[i].v6_key.port_dst,
448 em_route_base_v6[i].v6_key.port_src,
449 em_route_base_v6[i].v6_key.proto,
450 em_route_base_v6[i].if_out, dev_info.device->name);
451 }
452 printf("Hash: Adding 0x%" PRIx64 "keys\n",
453 (uint64_t)route_num_v6);
454 }
455
456 /* Requirements:
457 * 1. IP packets without extension;
458 * 2. L4 payload should be either TCP or UDP.
459 */
460 int
em_check_ptype(int portid)461 em_check_ptype(int portid)
462 {
463 int i, ret;
464 int ptype_l3_ipv4_ext = 0;
465 int ptype_l3_ipv6_ext = 0;
466 int ptype_l4_tcp = 0;
467 int ptype_l4_udp = 0;
468 uint32_t ptype_mask = RTE_PTYPE_L3_MASK | RTE_PTYPE_L4_MASK;
469
470 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
471 if (ret <= 0)
472 return 0;
473
474 uint32_t ptypes[ret];
475
476 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
477 for (i = 0; i < ret; ++i) {
478 switch (ptypes[i]) {
479 case RTE_PTYPE_L3_IPV4_EXT:
480 ptype_l3_ipv4_ext = 1;
481 break;
482 case RTE_PTYPE_L3_IPV6_EXT:
483 ptype_l3_ipv6_ext = 1;
484 break;
485 case RTE_PTYPE_L4_TCP:
486 ptype_l4_tcp = 1;
487 break;
488 case RTE_PTYPE_L4_UDP:
489 ptype_l4_udp = 1;
490 break;
491 }
492 }
493
494 if (ptype_l3_ipv4_ext == 0)
495 printf("port %d cannot parse RTE_PTYPE_L3_IPV4_EXT\n", portid);
496 if (ptype_l3_ipv6_ext == 0)
497 printf("port %d cannot parse RTE_PTYPE_L3_IPV6_EXT\n", portid);
498 if (!ptype_l3_ipv4_ext || !ptype_l3_ipv6_ext)
499 return 0;
500
501 if (ptype_l4_tcp == 0)
502 printf("port %d cannot parse RTE_PTYPE_L4_TCP\n", portid);
503 if (ptype_l4_udp == 0)
504 printf("port %d cannot parse RTE_PTYPE_L4_UDP\n", portid);
505 if (ptype_l4_tcp && ptype_l4_udp)
506 return 1;
507
508 return 0;
509 }
510
511 static inline void
em_parse_ptype(struct rte_mbuf * m)512 em_parse_ptype(struct rte_mbuf *m)
513 {
514 struct rte_ether_hdr *eth_hdr;
515 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
516 uint16_t ether_type;
517 void *l3;
518 int hdr_len;
519 struct rte_ipv4_hdr *ipv4_hdr;
520 struct rte_ipv6_hdr *ipv6_hdr;
521
522 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
523 ether_type = eth_hdr->ether_type;
524 l3 = (uint8_t *)eth_hdr + sizeof(struct rte_ether_hdr);
525 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
526 ipv4_hdr = (struct rte_ipv4_hdr *)l3;
527 hdr_len = rte_ipv4_hdr_len(ipv4_hdr);
528 if (hdr_len == sizeof(struct rte_ipv4_hdr)) {
529 packet_type |= RTE_PTYPE_L3_IPV4;
530 if (ipv4_hdr->next_proto_id == IPPROTO_TCP)
531 packet_type |= RTE_PTYPE_L4_TCP;
532 else if (ipv4_hdr->next_proto_id == IPPROTO_UDP)
533 packet_type |= RTE_PTYPE_L4_UDP;
534 } else
535 packet_type |= RTE_PTYPE_L3_IPV4_EXT;
536 } else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
537 ipv6_hdr = (struct rte_ipv6_hdr *)l3;
538 if (ipv6_hdr->proto == IPPROTO_TCP)
539 packet_type |= RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP;
540 else if (ipv6_hdr->proto == IPPROTO_UDP)
541 packet_type |= RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP;
542 else
543 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
544 }
545
546 m->packet_type = packet_type;
547 }
548
549 uint16_t
em_cb_parse_ptype(uint16_t port __rte_unused,uint16_t queue __rte_unused,struct rte_mbuf * pkts[],uint16_t nb_pkts,uint16_t max_pkts __rte_unused,void * user_param __rte_unused)550 em_cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
551 struct rte_mbuf *pkts[], uint16_t nb_pkts,
552 uint16_t max_pkts __rte_unused,
553 void *user_param __rte_unused)
554 {
555 unsigned i;
556
557 for (i = 0; i < nb_pkts; ++i)
558 em_parse_ptype(pkts[i]);
559
560 return nb_pkts;
561 }
562
563 /* main processing loop */
564 int
em_main_loop(__rte_unused void * dummy)565 em_main_loop(__rte_unused void *dummy)
566 {
567 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
568 unsigned lcore_id;
569 uint64_t prev_tsc, diff_tsc, cur_tsc;
570 int i, nb_rx;
571 uint8_t queueid;
572 uint16_t portid;
573 struct lcore_conf *qconf;
574 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
575 US_PER_S * BURST_TX_DRAIN_US;
576
577 lcore_id = rte_lcore_id();
578 qconf = &lcore_conf[lcore_id];
579
580 const uint16_t n_rx_q = qconf->n_rx_queue;
581 const uint16_t n_tx_p = qconf->n_tx_port;
582 if (n_rx_q == 0) {
583 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
584 return 0;
585 }
586
587 RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
588
589 for (i = 0; i < n_rx_q; i++) {
590
591 portid = qconf->rx_queue_list[i].port_id;
592 queueid = qconf->rx_queue_list[i].queue_id;
593 RTE_LOG(INFO, L3FWD,
594 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
595 lcore_id, portid, queueid);
596 }
597
598 cur_tsc = rte_rdtsc();
599 prev_tsc = cur_tsc;
600
601 while (!force_quit) {
602
603 /*
604 * TX burst queue drain
605 */
606 diff_tsc = cur_tsc - prev_tsc;
607 if (unlikely(diff_tsc > drain_tsc)) {
608
609 for (i = 0; i < n_tx_p; ++i) {
610 portid = qconf->tx_port_id[i];
611 if (qconf->tx_mbufs[portid].len == 0)
612 continue;
613 send_burst(qconf,
614 qconf->tx_mbufs[portid].len,
615 portid);
616 qconf->tx_mbufs[portid].len = 0;
617 }
618
619 prev_tsc = cur_tsc;
620 }
621
622 /*
623 * Read packet from RX queues
624 */
625 for (i = 0; i < n_rx_q; ++i) {
626 portid = qconf->rx_queue_list[i].port_id;
627 queueid = qconf->rx_queue_list[i].queue_id;
628 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
629 MAX_PKT_BURST);
630 if (nb_rx == 0)
631 continue;
632
633 #if defined RTE_ARCH_X86 || defined __ARM_NEON
634 l3fwd_em_send_packets(nb_rx, pkts_burst,
635 portid, qconf);
636 #else
637 l3fwd_em_no_opt_send_packets(nb_rx, pkts_burst,
638 portid, qconf);
639 #endif
640 }
641
642 cur_tsc = rte_rdtsc();
643 }
644
645 return 0;
646 }
647
648 static __rte_always_inline void
em_event_loop_single(struct l3fwd_event_resources * evt_rsrc,const uint8_t flags)649 em_event_loop_single(struct l3fwd_event_resources *evt_rsrc,
650 const uint8_t flags)
651 {
652 const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
653 const uint8_t tx_q_id = evt_rsrc->evq.event_q_id[
654 evt_rsrc->evq.nb_queues - 1];
655 const uint8_t event_d_id = evt_rsrc->event_d_id;
656 uint8_t deq = 0, enq = 0;
657 struct lcore_conf *lconf;
658 unsigned int lcore_id;
659 struct rte_event ev;
660
661 if (event_p_id < 0)
662 return;
663
664 lcore_id = rte_lcore_id();
665 lconf = &lcore_conf[lcore_id];
666
667 RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
668 while (!force_quit) {
669 deq = rte_event_dequeue_burst(event_d_id, event_p_id, &ev, 1,
670 0);
671 if (!deq)
672 continue;
673
674 struct rte_mbuf *mbuf = ev.mbuf;
675
676 #if defined RTE_ARCH_X86 || defined __ARM_NEON
677 mbuf->port = em_get_dst_port(lconf, mbuf, mbuf->port);
678 process_packet(mbuf, &mbuf->port);
679 #else
680 l3fwd_em_simple_process(mbuf, lconf);
681 #endif
682 if (mbuf->port == BAD_PORT) {
683 rte_pktmbuf_free(mbuf);
684 continue;
685 }
686
687 if (flags & L3FWD_EVENT_TX_ENQ) {
688 ev.queue_id = tx_q_id;
689 ev.op = RTE_EVENT_OP_FORWARD;
690 do {
691 enq = rte_event_enqueue_burst(
692 event_d_id, event_p_id, &ev, 1);
693 } while (!enq && !force_quit);
694 }
695
696 if (flags & L3FWD_EVENT_TX_DIRECT) {
697 rte_event_eth_tx_adapter_txq_set(mbuf, 0);
698 do {
699 enq = rte_event_eth_tx_adapter_enqueue(
700 event_d_id, event_p_id, &ev, 1, 0);
701 } while (!enq && !force_quit);
702 }
703 }
704
705 l3fwd_event_worker_cleanup(event_d_id, event_p_id, &ev, enq, deq, 0);
706 }
707
708 static __rte_always_inline void
em_event_loop_burst(struct l3fwd_event_resources * evt_rsrc,const uint8_t flags)709 em_event_loop_burst(struct l3fwd_event_resources *evt_rsrc,
710 const uint8_t flags)
711 {
712 const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
713 const uint8_t tx_q_id = evt_rsrc->evq.event_q_id[
714 evt_rsrc->evq.nb_queues - 1];
715 const uint8_t event_d_id = evt_rsrc->event_d_id;
716 const uint16_t deq_len = evt_rsrc->deq_depth;
717 struct rte_event events[MAX_PKT_BURST];
718 int i, nb_enq = 0, nb_deq = 0;
719 struct lcore_conf *lconf;
720 unsigned int lcore_id;
721
722 if (event_p_id < 0)
723 return;
724
725 lcore_id = rte_lcore_id();
726
727 lconf = &lcore_conf[lcore_id];
728
729 RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
730
731 while (!force_quit) {
732 /* Read events from RX queues */
733 nb_deq = rte_event_dequeue_burst(event_d_id, event_p_id,
734 events, deq_len, 0);
735 if (nb_deq == 0) {
736 rte_pause();
737 continue;
738 }
739
740 #if defined RTE_ARCH_X86 || defined __ARM_NEON
741 l3fwd_em_process_events(nb_deq, (struct rte_event **)&events,
742 lconf);
743 #else
744 l3fwd_em_no_opt_process_events(nb_deq,
745 (struct rte_event **)&events,
746 lconf);
747 #endif
748 for (i = 0; i < nb_deq; i++) {
749 if (flags & L3FWD_EVENT_TX_ENQ) {
750 events[i].queue_id = tx_q_id;
751 events[i].op = RTE_EVENT_OP_FORWARD;
752 }
753
754 if (flags & L3FWD_EVENT_TX_DIRECT)
755 rte_event_eth_tx_adapter_txq_set(events[i].mbuf,
756 0);
757 }
758
759 if (flags & L3FWD_EVENT_TX_ENQ) {
760 nb_enq = rte_event_enqueue_burst(event_d_id, event_p_id,
761 events, nb_deq);
762 while (nb_enq < nb_deq && !force_quit)
763 nb_enq += rte_event_enqueue_burst(event_d_id,
764 event_p_id, events + nb_enq,
765 nb_deq - nb_enq);
766 }
767
768 if (flags & L3FWD_EVENT_TX_DIRECT) {
769 nb_enq = rte_event_eth_tx_adapter_enqueue(event_d_id,
770 event_p_id, events, nb_deq, 0);
771 while (nb_enq < nb_deq && !force_quit)
772 nb_enq += rte_event_eth_tx_adapter_enqueue(
773 event_d_id, event_p_id,
774 events + nb_enq,
775 nb_deq - nb_enq, 0);
776 }
777 }
778
779 l3fwd_event_worker_cleanup(event_d_id, event_p_id, events, nb_enq,
780 nb_deq, 0);
781 }
782
783 static __rte_always_inline void
em_event_loop(struct l3fwd_event_resources * evt_rsrc,const uint8_t flags)784 em_event_loop(struct l3fwd_event_resources *evt_rsrc,
785 const uint8_t flags)
786 {
787 if (flags & L3FWD_EVENT_SINGLE)
788 em_event_loop_single(evt_rsrc, flags);
789 if (flags & L3FWD_EVENT_BURST)
790 em_event_loop_burst(evt_rsrc, flags);
791 }
792
793 int __rte_noinline
em_event_main_loop_tx_d(__rte_unused void * dummy)794 em_event_main_loop_tx_d(__rte_unused void *dummy)
795 {
796 struct l3fwd_event_resources *evt_rsrc =
797 l3fwd_get_eventdev_rsrc();
798
799 em_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT | L3FWD_EVENT_SINGLE);
800 return 0;
801 }
802
803 int __rte_noinline
em_event_main_loop_tx_d_burst(__rte_unused void * dummy)804 em_event_main_loop_tx_d_burst(__rte_unused void *dummy)
805 {
806 struct l3fwd_event_resources *evt_rsrc =
807 l3fwd_get_eventdev_rsrc();
808
809 em_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT | L3FWD_EVENT_BURST);
810 return 0;
811 }
812
813 int __rte_noinline
em_event_main_loop_tx_q(__rte_unused void * dummy)814 em_event_main_loop_tx_q(__rte_unused void *dummy)
815 {
816 struct l3fwd_event_resources *evt_rsrc =
817 l3fwd_get_eventdev_rsrc();
818
819 em_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ | L3FWD_EVENT_SINGLE);
820 return 0;
821 }
822
823 int __rte_noinline
em_event_main_loop_tx_q_burst(__rte_unused void * dummy)824 em_event_main_loop_tx_q_burst(__rte_unused void *dummy)
825 {
826 struct l3fwd_event_resources *evt_rsrc =
827 l3fwd_get_eventdev_rsrc();
828
829 em_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ | L3FWD_EVENT_BURST);
830 return 0;
831 }
832
833 /* Same eventdev loop for single and burst of vector */
834 static __rte_always_inline void
em_event_loop_vector(struct l3fwd_event_resources * evt_rsrc,const uint8_t flags)835 em_event_loop_vector(struct l3fwd_event_resources *evt_rsrc,
836 const uint8_t flags)
837 {
838 const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
839 const uint8_t tx_q_id =
840 evt_rsrc->evq.event_q_id[evt_rsrc->evq.nb_queues - 1];
841 const uint8_t event_d_id = evt_rsrc->event_d_id;
842 const uint16_t deq_len = evt_rsrc->deq_depth;
843 struct rte_event events[MAX_PKT_BURST];
844 int i, nb_enq = 0, nb_deq = 0;
845 struct lcore_conf *lconf;
846 unsigned int lcore_id;
847
848 if (event_p_id < 0)
849 return;
850
851 lcore_id = rte_lcore_id();
852 lconf = &lcore_conf[lcore_id];
853
854 RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
855
856 while (!force_quit) {
857 /* Read events from RX queues */
858 nb_deq = rte_event_dequeue_burst(event_d_id, event_p_id, events,
859 deq_len, 0);
860 if (nb_deq == 0) {
861 rte_pause();
862 continue;
863 }
864
865 for (i = 0; i < nb_deq; i++) {
866 if (flags & L3FWD_EVENT_TX_ENQ) {
867 events[i].queue_id = tx_q_id;
868 events[i].op = RTE_EVENT_OP_FORWARD;
869 }
870
871 #if defined RTE_ARCH_X86 || defined __ARM_NEON
872 l3fwd_em_process_event_vector(events[i].vec, lconf);
873 #else
874 l3fwd_em_no_opt_process_event_vector(events[i].vec,
875 lconf);
876 #endif
877 if (flags & L3FWD_EVENT_TX_DIRECT)
878 event_vector_txq_set(events[i].vec, 0);
879 }
880
881 if (flags & L3FWD_EVENT_TX_ENQ) {
882 nb_enq = rte_event_enqueue_burst(event_d_id, event_p_id,
883 events, nb_deq);
884 while (nb_enq < nb_deq && !force_quit)
885 nb_enq += rte_event_enqueue_burst(
886 event_d_id, event_p_id, events + nb_enq,
887 nb_deq - nb_enq);
888 }
889
890 if (flags & L3FWD_EVENT_TX_DIRECT) {
891 nb_enq = rte_event_eth_tx_adapter_enqueue(
892 event_d_id, event_p_id, events, nb_deq, 0);
893 while (nb_enq < nb_deq && !force_quit)
894 nb_enq += rte_event_eth_tx_adapter_enqueue(
895 event_d_id, event_p_id, events + nb_enq,
896 nb_deq - nb_enq, 0);
897 }
898 }
899
900 l3fwd_event_worker_cleanup(event_d_id, event_p_id, events, nb_enq,
901 nb_deq, 1);
902 }
903
904 int __rte_noinline
em_event_main_loop_tx_d_vector(__rte_unused void * dummy)905 em_event_main_loop_tx_d_vector(__rte_unused void *dummy)
906 {
907 struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
908
909 em_event_loop_vector(evt_rsrc, L3FWD_EVENT_TX_DIRECT);
910 return 0;
911 }
912
913 int __rte_noinline
em_event_main_loop_tx_d_burst_vector(__rte_unused void * dummy)914 em_event_main_loop_tx_d_burst_vector(__rte_unused void *dummy)
915 {
916 struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
917
918 em_event_loop_vector(evt_rsrc, L3FWD_EVENT_TX_DIRECT);
919 return 0;
920 }
921
922 int __rte_noinline
em_event_main_loop_tx_q_vector(__rte_unused void * dummy)923 em_event_main_loop_tx_q_vector(__rte_unused void *dummy)
924 {
925 struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
926
927 em_event_loop_vector(evt_rsrc, L3FWD_EVENT_TX_ENQ);
928 return 0;
929 }
930
931 int __rte_noinline
em_event_main_loop_tx_q_burst_vector(__rte_unused void * dummy)932 em_event_main_loop_tx_q_burst_vector(__rte_unused void *dummy)
933 {
934 struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
935
936 em_event_loop_vector(evt_rsrc, L3FWD_EVENT_TX_ENQ);
937 return 0;
938 }
939
940 /* Initialize exact match (hash) parameters. 8< */
941 void
setup_hash(const int socketid)942 setup_hash(const int socketid)
943 {
944 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
945 .name = NULL,
946 .entries = L3FWD_HASH_ENTRIES,
947 .key_len = sizeof(union ipv4_5tuple_host),
948 .hash_func = ipv4_hash_crc,
949 .hash_func_init_val = 0,
950 };
951
952 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
953 .name = NULL,
954 .entries = L3FWD_HASH_ENTRIES,
955 .key_len = sizeof(union ipv6_5tuple_host),
956 .hash_func = ipv6_hash_crc,
957 .hash_func_init_val = 0,
958 };
959
960 char s[64];
961
962 /* create ipv4 hash */
963 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
964 ipv4_l3fwd_hash_params.name = s;
965 ipv4_l3fwd_hash_params.socket_id = socketid;
966 ipv4_l3fwd_em_lookup_struct[socketid] =
967 rte_hash_create(&ipv4_l3fwd_hash_params);
968 if (ipv4_l3fwd_em_lookup_struct[socketid] == NULL)
969 rte_exit(EXIT_FAILURE,
970 "Unable to create the l3fwd hash on socket %d\n",
971 socketid);
972
973 /* create ipv6 hash */
974 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
975 ipv6_l3fwd_hash_params.name = s;
976 ipv6_l3fwd_hash_params.socket_id = socketid;
977 ipv6_l3fwd_em_lookup_struct[socketid] =
978 rte_hash_create(&ipv6_l3fwd_hash_params);
979 if (ipv6_l3fwd_em_lookup_struct[socketid] == NULL)
980 rte_exit(EXIT_FAILURE,
981 "Unable to create the l3fwd hash on socket %d\n",
982 socketid);
983
984 /*
985 * Use data from ipv4/ipv6 l3fwd config file
986 * directly to initialize the hash table.
987 */
988 if (ipv6 == 0) {
989 /* populate the ipv4 hash */
990 populate_ipv4_flow_into_table(
991 ipv4_l3fwd_em_lookup_struct[socketid]);
992 } else {
993 /* populate the ipv6 hash */
994 populate_ipv6_flow_into_table(
995 ipv6_l3fwd_em_lookup_struct[socketid]);
996 }
997 }
998 /* >8 End of initialization of hash parameters. */
999
1000 /* Return ipv4/ipv6 em fwd lookup struct. */
1001 void *
em_get_ipv4_l3fwd_lookup_struct(const int socketid)1002 em_get_ipv4_l3fwd_lookup_struct(const int socketid)
1003 {
1004 return ipv4_l3fwd_em_lookup_struct[socketid];
1005 }
1006
1007 void *
em_get_ipv6_l3fwd_lookup_struct(const int socketid)1008 em_get_ipv6_l3fwd_lookup_struct(const int socketid)
1009 {
1010 return ipv6_l3fwd_em_lookup_struct[socketid];
1011 }
1012