]> git.proxmox.com Git - ceph.git/blob - ceph/src/seastar/dpdk/examples/l3fwd/l3fwd_em_hlm_sse.h
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / seastar / dpdk / examples / l3fwd / l3fwd_em_hlm_sse.h
1 /*-
2 * BSD LICENSE
3 *
4 * Copyright(c) 2016 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 #ifndef __L3FWD_EM_HLM_SSE_H__
35 #define __L3FWD_EM_HLM_SSE_H__
36
37 #include "l3fwd_sse.h"
38
39 static inline __attribute__((always_inline)) void
40 em_get_dst_port_ipv4x8(struct lcore_conf *qconf, struct rte_mbuf *m[8],
41 uint8_t portid, uint16_t dst_port[8])
42 {
43 int32_t ret[8];
44 union ipv4_5tuple_host key[8];
45 __m128i data[8];
46
47 data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
48 sizeof(struct ether_hdr) +
49 offsetof(struct ipv4_hdr, time_to_live)));
50 data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
51 sizeof(struct ether_hdr) +
52 offsetof(struct ipv4_hdr, time_to_live)));
53 data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
54 sizeof(struct ether_hdr) +
55 offsetof(struct ipv4_hdr, time_to_live)));
56 data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
57 sizeof(struct ether_hdr) +
58 offsetof(struct ipv4_hdr, time_to_live)));
59 data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
60 sizeof(struct ether_hdr) +
61 offsetof(struct ipv4_hdr, time_to_live)));
62 data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
63 sizeof(struct ether_hdr) +
64 offsetof(struct ipv4_hdr, time_to_live)));
65 data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
66 sizeof(struct ether_hdr) +
67 offsetof(struct ipv4_hdr, time_to_live)));
68 data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
69 sizeof(struct ether_hdr) +
70 offsetof(struct ipv4_hdr, time_to_live)));
71
72 key[0].xmm = _mm_and_si128(data[0], mask0.x);
73 key[1].xmm = _mm_and_si128(data[1], mask0.x);
74 key[2].xmm = _mm_and_si128(data[2], mask0.x);
75 key[3].xmm = _mm_and_si128(data[3], mask0.x);
76 key[4].xmm = _mm_and_si128(data[4], mask0.x);
77 key[5].xmm = _mm_and_si128(data[5], mask0.x);
78 key[6].xmm = _mm_and_si128(data[6], mask0.x);
79 key[7].xmm = _mm_and_si128(data[7], mask0.x);
80
81 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
82 &key[4], &key[5], &key[6], &key[7]};
83
84 rte_hash_lookup_bulk(qconf->ipv4_lookup_struct, &key_array[0], 8, ret);
85
86 dst_port[0] = (uint8_t) ((ret[0] < 0) ?
87 portid : ipv4_l3fwd_out_if[ret[0]]);
88 dst_port[1] = (uint8_t) ((ret[1] < 0) ?
89 portid : ipv4_l3fwd_out_if[ret[1]]);
90 dst_port[2] = (uint8_t) ((ret[2] < 0) ?
91 portid : ipv4_l3fwd_out_if[ret[2]]);
92 dst_port[3] = (uint8_t) ((ret[3] < 0) ?
93 portid : ipv4_l3fwd_out_if[ret[3]]);
94 dst_port[4] = (uint8_t) ((ret[4] < 0) ?
95 portid : ipv4_l3fwd_out_if[ret[4]]);
96 dst_port[5] = (uint8_t) ((ret[5] < 0) ?
97 portid : ipv4_l3fwd_out_if[ret[5]]);
98 dst_port[6] = (uint8_t) ((ret[6] < 0) ?
99 portid : ipv4_l3fwd_out_if[ret[6]]);
100 dst_port[7] = (uint8_t) ((ret[7] < 0) ?
101 portid : ipv4_l3fwd_out_if[ret[7]]);
102
103 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
104 (enabled_port_mask & 1 << dst_port[0]) == 0)
105 dst_port[0] = portid;
106
107 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
108 (enabled_port_mask & 1 << dst_port[1]) == 0)
109 dst_port[1] = portid;
110
111 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
112 (enabled_port_mask & 1 << dst_port[2]) == 0)
113 dst_port[2] = portid;
114
115 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
116 (enabled_port_mask & 1 << dst_port[3]) == 0)
117 dst_port[3] = portid;
118
119 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
120 (enabled_port_mask & 1 << dst_port[4]) == 0)
121 dst_port[4] = portid;
122
123 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
124 (enabled_port_mask & 1 << dst_port[5]) == 0)
125 dst_port[5] = portid;
126
127 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
128 (enabled_port_mask & 1 << dst_port[6]) == 0)
129 dst_port[6] = portid;
130
131 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
132 (enabled_port_mask & 1 << dst_port[7]) == 0)
133 dst_port[7] = portid;
134
135 }
136
137 static inline void
138 get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
139 __m128i mask1, union ipv6_5tuple_host *key)
140 {
141 __m128i tmpdata0 = _mm_loadu_si128(
142 rte_pktmbuf_mtod_offset(m0, __m128i *,
143 sizeof(struct ether_hdr) +
144 offsetof(struct ipv6_hdr, payload_len)));
145
146 __m128i tmpdata1 = _mm_loadu_si128(
147 rte_pktmbuf_mtod_offset(m0, __m128i *,
148 sizeof(struct ether_hdr) +
149 offsetof(struct ipv6_hdr, payload_len) +
150 sizeof(__m128i)));
151
152 __m128i tmpdata2 = _mm_loadu_si128(
153 rte_pktmbuf_mtod_offset(m0, __m128i *,
154 sizeof(struct ether_hdr) +
155 offsetof(struct ipv6_hdr, payload_len) +
156 sizeof(__m128i) + sizeof(__m128i)));
157
158 key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
159 key->xmm[1] = tmpdata1;
160 key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
161 }
162
163 static inline __attribute__((always_inline)) void
164 em_get_dst_port_ipv6x8(struct lcore_conf *qconf, struct rte_mbuf *m[8],
165 uint8_t portid, uint16_t dst_port[8])
166 {
167 int32_t ret[8];
168 union ipv6_5tuple_host key[8];
169
170 get_ipv6_5tuple(m[0], mask1.x, mask2.x, &key[0]);
171 get_ipv6_5tuple(m[1], mask1.x, mask2.x, &key[1]);
172 get_ipv6_5tuple(m[2], mask1.x, mask2.x, &key[2]);
173 get_ipv6_5tuple(m[3], mask1.x, mask2.x, &key[3]);
174 get_ipv6_5tuple(m[4], mask1.x, mask2.x, &key[4]);
175 get_ipv6_5tuple(m[5], mask1.x, mask2.x, &key[5]);
176 get_ipv6_5tuple(m[6], mask1.x, mask2.x, &key[6]);
177 get_ipv6_5tuple(m[7], mask1.x, mask2.x, &key[7]);
178
179 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
180 &key[4], &key[5], &key[6], &key[7]};
181
182 rte_hash_lookup_bulk(qconf->ipv6_lookup_struct, &key_array[0], 8, ret);
183
184 dst_port[0] = (uint8_t) ((ret[0] < 0) ?
185 portid : ipv6_l3fwd_out_if[ret[0]]);
186 dst_port[1] = (uint8_t) ((ret[1] < 0) ?
187 portid : ipv6_l3fwd_out_if[ret[1]]);
188 dst_port[2] = (uint8_t) ((ret[2] < 0) ?
189 portid : ipv6_l3fwd_out_if[ret[2]]);
190 dst_port[3] = (uint8_t) ((ret[3] < 0) ?
191 portid : ipv6_l3fwd_out_if[ret[3]]);
192 dst_port[4] = (uint8_t) ((ret[4] < 0) ?
193 portid : ipv6_l3fwd_out_if[ret[4]]);
194 dst_port[5] = (uint8_t) ((ret[5] < 0) ?
195 portid : ipv6_l3fwd_out_if[ret[5]]);
196 dst_port[6] = (uint8_t) ((ret[6] < 0) ?
197 portid : ipv6_l3fwd_out_if[ret[6]]);
198 dst_port[7] = (uint8_t) ((ret[7] < 0) ?
199 portid : ipv6_l3fwd_out_if[ret[7]]);
200
201 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
202 (enabled_port_mask & 1 << dst_port[0]) == 0)
203 dst_port[0] = portid;
204
205 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
206 (enabled_port_mask & 1 << dst_port[1]) == 0)
207 dst_port[1] = portid;
208
209 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
210 (enabled_port_mask & 1 << dst_port[2]) == 0)
211 dst_port[2] = portid;
212
213 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
214 (enabled_port_mask & 1 << dst_port[3]) == 0)
215 dst_port[3] = portid;
216
217 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
218 (enabled_port_mask & 1 << dst_port[4]) == 0)
219 dst_port[4] = portid;
220
221 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
222 (enabled_port_mask & 1 << dst_port[5]) == 0)
223 dst_port[5] = portid;
224
225 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
226 (enabled_port_mask & 1 << dst_port[6]) == 0)
227 dst_port[6] = portid;
228
229 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
230 (enabled_port_mask & 1 << dst_port[7]) == 0)
231 dst_port[7] = portid;
232
233 }
234
235 static inline __attribute__((always_inline)) uint16_t
236 em_get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
237 uint8_t portid)
238 {
239 uint8_t next_hop;
240 struct ipv4_hdr *ipv4_hdr;
241 struct ipv6_hdr *ipv6_hdr;
242 uint32_t tcp_or_udp;
243 uint32_t l3_ptypes;
244
245 tcp_or_udp = pkt->packet_type & (RTE_PTYPE_L4_TCP | RTE_PTYPE_L4_UDP);
246 l3_ptypes = pkt->packet_type & RTE_PTYPE_L3_MASK;
247
248 if (tcp_or_udp && (l3_ptypes == RTE_PTYPE_L3_IPV4)) {
249
250 /* Handle IPv4 headers.*/
251 ipv4_hdr = rte_pktmbuf_mtod_offset(pkt, struct ipv4_hdr *,
252 sizeof(struct ether_hdr));
253
254 next_hop = em_get_ipv4_dst_port(ipv4_hdr, portid,
255 qconf->ipv4_lookup_struct);
256
257 if (next_hop >= RTE_MAX_ETHPORTS ||
258 (enabled_port_mask & 1 << next_hop) == 0)
259 next_hop = portid;
260
261 return next_hop;
262
263 } else if (tcp_or_udp && (l3_ptypes == RTE_PTYPE_L3_IPV6)) {
264
265 /* Handle IPv6 headers.*/
266 ipv6_hdr = rte_pktmbuf_mtod_offset(pkt, struct ipv6_hdr *,
267 sizeof(struct ether_hdr));
268
269 next_hop = em_get_ipv6_dst_port(ipv6_hdr, portid,
270 qconf->ipv6_lookup_struct);
271
272 if (next_hop >= RTE_MAX_ETHPORTS ||
273 (enabled_port_mask & 1 << next_hop) == 0)
274 next_hop = portid;
275
276 return next_hop;
277
278 }
279
280 return portid;
281 }
282
283 /*
284 * Buffer optimized handling of packets, invoked
285 * from main_loop.
286 */
287 static inline void
288 l3fwd_em_send_packets(int nb_rx, struct rte_mbuf **pkts_burst,
289 uint8_t portid, struct lcore_conf *qconf)
290 {
291 int32_t j;
292 uint16_t dst_port[MAX_PKT_BURST];
293
294 /*
295 * Send nb_rx - nb_rx%8 packets
296 * in groups of 8.
297 */
298 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
299
300 for (j = 0; j < n; j += 8) {
301
302 uint32_t pkt_type =
303 pkts_burst[j]->packet_type &
304 pkts_burst[j+1]->packet_type &
305 pkts_burst[j+2]->packet_type &
306 pkts_burst[j+3]->packet_type &
307 pkts_burst[j+4]->packet_type &
308 pkts_burst[j+5]->packet_type &
309 pkts_burst[j+6]->packet_type &
310 pkts_burst[j+7]->packet_type;
311
312 uint32_t l3_type = pkt_type & RTE_PTYPE_L3_MASK;
313 uint32_t tcp_or_udp = pkt_type &
314 (RTE_PTYPE_L4_TCP | RTE_PTYPE_L4_UDP);
315
316 if (tcp_or_udp && (l3_type == RTE_PTYPE_L3_IPV4)) {
317
318 em_get_dst_port_ipv4x8(qconf, &pkts_burst[j], portid, &dst_port[j]);
319
320 } else if (tcp_or_udp && (l3_type == RTE_PTYPE_L3_IPV6)) {
321
322 em_get_dst_port_ipv6x8(qconf, &pkts_burst[j], portid, &dst_port[j]);
323
324 } else {
325 dst_port[j] = em_get_dst_port(qconf, pkts_burst[j], portid);
326 dst_port[j+1] = em_get_dst_port(qconf, pkts_burst[j+1], portid);
327 dst_port[j+2] = em_get_dst_port(qconf, pkts_burst[j+2], portid);
328 dst_port[j+3] = em_get_dst_port(qconf, pkts_burst[j+3], portid);
329 dst_port[j+4] = em_get_dst_port(qconf, pkts_burst[j+4], portid);
330 dst_port[j+5] = em_get_dst_port(qconf, pkts_burst[j+5], portid);
331 dst_port[j+6] = em_get_dst_port(qconf, pkts_burst[j+6], portid);
332 dst_port[j+7] = em_get_dst_port(qconf, pkts_burst[j+7], portid);
333 }
334 }
335
336 for (; j < nb_rx; j++)
337 dst_port[j] = em_get_dst_port(qconf, pkts_burst[j], portid);
338
339 send_packets_multi(qconf, pkts_burst, dst_port, nb_rx);
340
341 }
342 #endif /* __L3FWD_EM_SSE_HLM_H__ */