4 * Copyright(c) 2010-2012 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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.
33 * version: DPDK.L.1.2.3-3
43 #include <sys/queue.h>
46 #include <rte_common.h>
47 #include <rte_byteorder.h>
49 #include <rte_debug.h>
50 #include <rte_cycles.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
54 #include <rte_launch.h>
55 #include <rte_tailq.h>
57 #include <rte_per_lcore.h>
58 #include <rte_lcore.h>
59 #include <rte_atomic.h>
60 #include <rte_branch_prediction.h>
62 #include <rte_memory.h>
63 #include <rte_mempool.h>
65 #include <rte_memcpy.h>
66 #include <rte_interrupts.h>
68 #include <rte_ether.h>
69 #include <rte_ethdev.h>
74 #include <rte_prefetch.h>
75 #include <rte_string_fns.h>
80 #define IP_DEFTTL 64 /* from RFC 1340. */
81 #define IP_VERSION 0x40
82 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */
83 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
85 /* Pseudo Header for IPv4/UDP/TCP checksum */
87 uint32_t src_addr; /* IP address of source host. */
88 uint32_t dst_addr; /* IP address of destination host(s). */
89 uint8_t zero; /* zero. */
90 uint8_t proto; /* L4 protocol type. */
91 uint16_t len; /* L4 length. */
92 } __attribute__((__packed__));
95 /* Pseudo Header for IPv6/UDP/TCP checksum */
96 struct ipv6_psd_header {
97 uint8_t src_addr[16]; /* IP address of source host. */
98 uint8_t dst_addr[16]; /* IP address of destination host(s). */
99 uint32_t len; /* L4 length. */
100 uint8_t zero[3]; /* zero. */
101 uint8_t proto; /* L4 protocol. */
102 } __attribute__((__packed__));
105 static inline uint16_t
106 get_16b_sum(uint16_t *ptr16, uint32_t nr)
112 nr -= sizeof(uint16_t);
114 if (sum > UINT16_MAX)
118 /* If length is in odd bytes */
120 sum += *((uint8_t*)ptr16);
122 sum = ((sum & 0xffff0000) >> 16) + (sum & 0xffff);
124 return (uint16_t)sum;
127 static inline uint16_t
128 get_ipv4_cksum(struct ipv4_hdr *ipv4_hdr)
131 cksum = get_16b_sum((uint16_t*)ipv4_hdr, sizeof(struct ipv4_hdr));
132 return (uint16_t)((cksum == 0xffff)?cksum:~cksum);
137 uint16_t get_ipv4_psd_sum (struct ipv4_hdr * ip_hdr)
139 struct psd_header psd_hdr;
140 psd_hdr.src_addr = ip_hdr->src_addr;
141 psd_hdr.dst_addr = ip_hdr->dst_addr;
143 psd_hdr.proto = ip_hdr->next_proto_id;
144 psd_hdr.len = rte_cpu_to_be_16((uint16_t)(rte_be_to_cpu_16(ip_hdr->total_length)
145 - sizeof(struct ipv4_hdr)));
146 return get_16b_sum((uint16_t*)&psd_hdr, sizeof(struct psd_header));
150 uint16_t get_ipv6_psd_sum (struct ipv6_hdr * ip_hdr)
152 struct ipv6_psd_header psd_hdr;
153 rte_memcpy(psd_hdr.src_addr, ip_hdr->src_addr, sizeof(ip_hdr->src_addr)
154 + sizeof(ip_hdr->dst_addr));
159 psd_hdr.proto = ip_hdr->proto;
160 psd_hdr.len = ip_hdr->payload_len;
162 return get_16b_sum((uint16_t*)&psd_hdr, sizeof(struct ipv6_psd_header));
165 static inline uint16_t
166 get_ipv4_udptcp_checksum(struct ipv4_hdr *ipv4_hdr, uint16_t *l4_hdr)
171 l4_len = rte_be_to_cpu_16(ipv4_hdr->total_length) - sizeof(struct ipv4_hdr);
173 cksum = get_16b_sum(l4_hdr, l4_len);
174 cksum += get_ipv4_psd_sum(ipv4_hdr);
176 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
177 cksum = (~cksum) & 0xffff;
180 return (uint16_t)cksum;
184 static inline uint16_t
185 get_ipv6_udptcp_checksum(struct ipv6_hdr *ipv6_hdr, uint16_t *l4_hdr)
190 l4_len = rte_be_to_cpu_16(ipv6_hdr->payload_len);
192 cksum = get_16b_sum(l4_hdr, l4_len);
193 cksum += get_ipv6_psd_sum(ipv6_hdr);
195 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
196 cksum = (~cksum) & 0xffff;
200 return (uint16_t)cksum;
205 * Forwarding of packets. Change the checksum field with HW or SW methods
206 * The HW/SW method selection depends on the ol_flags on every packet
209 pkt_burst_checksum_forward(struct fwd_stream *fs)
211 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
212 struct rte_port *txp;
214 struct ether_hdr *eth_hdr;
215 struct ipv4_hdr *ipv4_hdr;
216 struct ipv6_hdr *ipv6_hdr;
217 struct udp_hdr *udp_hdr;
218 struct tcp_hdr *tcp_hdr;
219 struct sctp_hdr *sctp_hdr;
225 uint16_t pkt_ol_flags;
226 uint16_t tx_ol_flags;
231 uint32_t rx_bad_ip_csum;
232 uint32_t rx_bad_l4_csum;
234 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
237 uint64_t core_cycles;
240 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
241 start_tsc = rte_rdtsc();
245 * Receive a burst of packets and forward them.
247 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
249 if (unlikely(nb_rx == 0))
252 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
253 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
255 fs->rx_packets += nb_rx;
259 txp = &ports[fs->tx_port];
260 tx_ol_flags = txp->tx_ol_flags;
262 for (i = 0; i < nb_rx; i++) {
265 l2_len = sizeof(struct ether_hdr);
266 pkt_ol_flags = mb->ol_flags;
267 ol_flags = (uint16_t) (pkt_ol_flags & (~PKT_TX_L4_MASK));
269 eth_hdr = (struct ether_hdr *) mb->pkt.data;
270 if (rte_be_to_cpu_16(eth_hdr->ether_type) == ETHER_TYPE_VLAN) {
271 /* Only allow single VLAN label here */
272 l2_len += sizeof(struct vlan_hdr);
275 /* Update the L3/L4 checksum error packet count */
276 rx_bad_ip_csum += (uint16_t) ((pkt_ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
277 rx_bad_l4_csum += (uint16_t) ((pkt_ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
280 * Simplify the protocol parsing
281 * Assuming the incoming packets format as
282 * Ethernet2 + optional single VLAN
284 * + udp or tcp or sctp or others
286 if (pkt_ol_flags & PKT_RX_IPV4_HDR) {
288 /* Do not support ipv4 option field */
289 l3_len = sizeof(struct ipv4_hdr) ;
291 ipv4_hdr = (struct ipv4_hdr *) (rte_pktmbuf_mtod(mb,
292 unsigned char *) + l2_len);
294 l4_proto = ipv4_hdr->next_proto_id;
296 /* Do not delete, this is required by HW*/
297 ipv4_hdr->hdr_checksum = 0;
299 if (tx_ol_flags & 0x1) {
301 ol_flags |= PKT_TX_IP_CKSUM;
304 /* SW checksum calculation */
305 ipv4_hdr->src_addr++;
306 ipv4_hdr->hdr_checksum = get_ipv4_cksum(ipv4_hdr);
309 if (l4_proto == IPPROTO_UDP) {
310 udp_hdr = (struct udp_hdr*) (rte_pktmbuf_mtod(mb,
311 unsigned char *) + l2_len + l3_len);
312 if (tx_ol_flags & 0x2) {
314 ol_flags |= PKT_TX_UDP_CKSUM;
315 /* Pseudo header sum need be set properly */
316 udp_hdr->dgram_cksum = get_ipv4_psd_sum(ipv4_hdr);
319 /* SW Implementation, clear checksum field first */
320 udp_hdr->dgram_cksum = 0;
321 udp_hdr->dgram_cksum = get_ipv4_udptcp_checksum(ipv4_hdr,
325 else if (l4_proto == IPPROTO_TCP){
326 tcp_hdr = (struct tcp_hdr*) (rte_pktmbuf_mtod(mb,
327 unsigned char *) + l2_len + l3_len);
328 if (tx_ol_flags & 0x4) {
329 ol_flags |= PKT_TX_TCP_CKSUM;
330 tcp_hdr->cksum = get_ipv4_psd_sum(ipv4_hdr);
334 tcp_hdr->cksum = get_ipv4_udptcp_checksum(ipv4_hdr,
338 else if (l4_proto == IPPROTO_SCTP) {
339 sctp_hdr = (struct sctp_hdr*) (rte_pktmbuf_mtod(mb,
340 unsigned char *) + l2_len + l3_len);
342 if (tx_ol_flags & 0x8) {
343 ol_flags |= PKT_TX_SCTP_CKSUM;
346 /* Sanity check, only number of 4 bytes supported */
347 if ((rte_be_to_cpu_16(ipv4_hdr->total_length) % 4) != 0)
348 printf("sctp payload must be a multiple "
349 "of 4 bytes for checksum offload");
353 /* CRC32c sample code available in RFC3309 */
356 /* End of L4 Handling*/
360 ipv6_hdr = (struct ipv6_hdr *) (rte_pktmbuf_mtod(mb,
361 unsigned char *) + l2_len);
362 l3_len = sizeof(struct ipv6_hdr) ;
363 l4_proto = ipv6_hdr->proto;
365 if (l4_proto == IPPROTO_UDP) {
366 udp_hdr = (struct udp_hdr*) (rte_pktmbuf_mtod(mb,
367 unsigned char *) + l2_len + l3_len);
368 if (tx_ol_flags & 0x2) {
370 ol_flags |= PKT_TX_UDP_CKSUM;
371 udp_hdr->dgram_cksum = get_ipv6_psd_sum(ipv6_hdr);
374 /* SW Implementation */
375 /* checksum field need be clear first */
376 udp_hdr->dgram_cksum = 0;
377 udp_hdr->dgram_cksum = get_ipv6_udptcp_checksum(ipv6_hdr,
381 else if (l4_proto == IPPROTO_TCP) {
382 tcp_hdr = (struct tcp_hdr*) (rte_pktmbuf_mtod(mb,
383 unsigned char *) + l2_len + l3_len);
384 if (tx_ol_flags & 0x4) {
385 ol_flags |= PKT_TX_TCP_CKSUM;
386 tcp_hdr->cksum = get_ipv6_psd_sum(ipv6_hdr);
390 tcp_hdr->cksum = get_ipv6_udptcp_checksum(ipv6_hdr,
394 else if (l4_proto == IPPROTO_SCTP) {
395 sctp_hdr = (struct sctp_hdr*) (rte_pktmbuf_mtod(mb,
396 unsigned char *) + l2_len + l3_len);
398 if (tx_ol_flags & 0x8) {
399 ol_flags |= PKT_TX_SCTP_CKSUM;
401 /* Sanity check, only number of 4 bytes supported by HW */
402 if ((rte_be_to_cpu_16(ipv6_hdr->payload_len) % 4) != 0)
403 printf("sctp payload must be a multiple "
404 "of 4 bytes for checksum offload");
407 /* CRC32c sample code available in RFC3309 */
411 printf("Test flow control for 1G PMD \n");
413 /* End of L4 Handling*/
416 /* Combine the packet header write. VLAN is not consider here */
417 mb->pkt.l2_len = l2_len;
418 mb->pkt.l3_len = l3_len;
419 mb->ol_flags = ol_flags;
421 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
422 fs->tx_packets += nb_tx;
423 fs->rx_bad_ip_csum += rx_bad_ip_csum;
424 fs->rx_bad_l4_csum += rx_bad_l4_csum;
426 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
427 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
429 if (unlikely(nb_tx < nb_rx)) {
430 fs->fwd_dropped += (nb_rx - nb_tx);
432 rte_pktmbuf_free(pkts_burst[nb_tx]);
433 } while (++nb_tx < nb_rx);
435 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
436 end_tsc = rte_rdtsc();
437 core_cycles = (end_tsc - start_tsc);
438 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
443 struct fwd_engine csum_fwd_engine = {
444 .fwd_mode_name = "csum",
445 .port_fwd_begin = NULL,
446 .port_fwd_end = NULL,
447 .packet_fwd = pkt_burst_checksum_forward,