4 * Copyright(c) 2010-2013 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.
42 #include <sys/queue.h>
45 #include <rte_common.h>
46 #include <rte_byteorder.h>
48 #include <rte_debug.h>
49 #include <rte_cycles.h>
50 #include <rte_memory.h>
51 #include <rte_memcpy.h>
52 #include <rte_memzone.h>
53 #include <rte_launch.h>
54 #include <rte_tailq.h>
56 #include <rte_per_lcore.h>
57 #include <rte_lcore.h>
58 #include <rte_atomic.h>
59 #include <rte_branch_prediction.h>
61 #include <rte_memory.h>
62 #include <rte_mempool.h>
64 #include <rte_memcpy.h>
65 #include <rte_interrupts.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
73 #include <rte_prefetch.h>
74 #include <rte_string_fns.h>
79 #define IP_DEFTTL 64 /* from RFC 1340. */
80 #define IP_VERSION 0x40
81 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */
82 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
84 static inline uint16_t
85 get_16b_sum(uint16_t *ptr16, uint32_t nr)
91 nr -= sizeof(uint16_t);
97 /* If length is in odd bytes */
99 sum += *((uint8_t*)ptr16);
101 sum = ((sum & 0xffff0000) >> 16) + (sum & 0xffff);
103 return (uint16_t)sum;
106 static inline uint16_t
107 get_ipv4_cksum(struct ipv4_hdr *ipv4_hdr)
110 cksum = get_16b_sum((uint16_t*)ipv4_hdr, sizeof(struct ipv4_hdr));
111 return (uint16_t)((cksum == 0xffff)?cksum:~cksum);
115 static inline uint16_t
116 get_ipv4_psd_sum (struct ipv4_hdr * ip_hdr)
118 /* Pseudo Header for IPv4/UDP/TCP checksum */
119 union ipv4_psd_header {
121 uint32_t src_addr; /* IP address of source host. */
122 uint32_t dst_addr; /* IP address of destination host(s). */
123 uint8_t zero; /* zero. */
124 uint8_t proto; /* L4 protocol type. */
125 uint16_t len; /* L4 length. */
126 } __attribute__((__packed__));
130 psd_hdr.src_addr = ip_hdr->src_addr;
131 psd_hdr.dst_addr = ip_hdr->dst_addr;
133 psd_hdr.proto = ip_hdr->next_proto_id;
134 psd_hdr.len = rte_cpu_to_be_16((uint16_t)(rte_be_to_cpu_16(ip_hdr->total_length)
135 - sizeof(struct ipv4_hdr)));
136 return get_16b_sum(psd_hdr.u16_arr, sizeof(psd_hdr));
139 static inline uint16_t
140 get_ipv6_psd_sum (struct ipv6_hdr * ip_hdr)
142 /* Pseudo Header for IPv6/UDP/TCP checksum */
143 union ipv6_psd_header {
145 uint8_t src_addr[16]; /* IP address of source host. */
146 uint8_t dst_addr[16]; /* IP address of destination host(s). */
147 uint32_t len; /* L4 length. */
148 uint32_t proto; /* L4 protocol - top 3 bytes must be zero */
149 } __attribute__((__packed__));
151 uint16_t u16_arr[0]; /* allow use as 16-bit values with safe aliasing */
154 rte_memcpy(&psd_hdr.src_addr, ip_hdr->src_addr,
155 sizeof(ip_hdr->src_addr) + sizeof(ip_hdr->dst_addr));
156 psd_hdr.len = ip_hdr->payload_len;
157 psd_hdr.proto = (ip_hdr->proto << 24);
159 return get_16b_sum(psd_hdr.u16_arr, sizeof(psd_hdr));
162 static inline uint16_t
163 get_ipv4_udptcp_checksum(struct ipv4_hdr *ipv4_hdr, uint16_t *l4_hdr)
168 l4_len = rte_be_to_cpu_16(ipv4_hdr->total_length) - sizeof(struct ipv4_hdr);
170 cksum = get_16b_sum(l4_hdr, l4_len);
171 cksum += get_ipv4_psd_sum(ipv4_hdr);
173 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
174 cksum = (~cksum) & 0xffff;
177 return (uint16_t)cksum;
181 static inline uint16_t
182 get_ipv6_udptcp_checksum(struct ipv6_hdr *ipv6_hdr, uint16_t *l4_hdr)
187 l4_len = rte_be_to_cpu_16(ipv6_hdr->payload_len);
189 cksum = get_16b_sum(l4_hdr, l4_len);
190 cksum += get_ipv6_psd_sum(ipv6_hdr);
192 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
193 cksum = (~cksum) & 0xffff;
197 return (uint16_t)cksum;
202 * Forwarding of packets. Change the checksum field with HW or SW methods
203 * The HW/SW method selection depends on the ol_flags on every packet
206 pkt_burst_checksum_forward(struct fwd_stream *fs)
208 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
209 struct rte_port *txp;
211 struct ether_hdr *eth_hdr;
212 struct ipv4_hdr *ipv4_hdr;
213 struct ipv6_hdr *ipv6_hdr;
214 struct udp_hdr *udp_hdr;
215 struct tcp_hdr *tcp_hdr;
216 struct sctp_hdr *sctp_hdr;
222 uint16_t pkt_ol_flags;
223 uint16_t tx_ol_flags;
229 uint32_t rx_bad_ip_csum;
230 uint32_t rx_bad_l4_csum;
232 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
235 uint64_t core_cycles;
238 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
239 start_tsc = rte_rdtsc();
243 * Receive a burst of packets and forward them.
245 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
247 if (unlikely(nb_rx == 0))
250 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
251 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
253 fs->rx_packets += nb_rx;
257 txp = &ports[fs->tx_port];
258 tx_ol_flags = txp->tx_ol_flags;
260 for (i = 0; i < nb_rx; i++) {
263 l2_len = sizeof(struct ether_hdr);
264 pkt_ol_flags = mb->ol_flags;
265 ol_flags = (uint16_t) (pkt_ol_flags & (~PKT_TX_L4_MASK));
267 eth_hdr = (struct ether_hdr *) mb->pkt.data;
268 eth_type = rte_be_to_cpu_16(eth_hdr->ether_type);
269 if (eth_type == ETHER_TYPE_VLAN) {
270 /* Only allow single VLAN label here */
271 l2_len += sizeof(struct vlan_hdr);
272 eth_type = rte_be_to_cpu_16(*(uint16_t *)
273 ((uintptr_t)ð_hdr->ether_type +
274 sizeof(struct vlan_hdr)));
277 /* Update the L3/L4 checksum error packet count */
278 rx_bad_ip_csum += (uint16_t) ((pkt_ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
279 rx_bad_l4_csum += (uint16_t) ((pkt_ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
282 * Try to figure out L3 packet type by SW.
284 if ((pkt_ol_flags & (PKT_RX_IPV4_HDR | PKT_RX_IPV4_HDR_EXT |
285 PKT_RX_IPV6_HDR | PKT_RX_IPV6_HDR_EXT)) == 0) {
286 if (eth_type == ETHER_TYPE_IPv4)
287 pkt_ol_flags |= PKT_RX_IPV4_HDR;
288 else if (eth_type == ETHER_TYPE_IPv6)
289 pkt_ol_flags |= PKT_RX_IPV6_HDR;
293 * Simplify the protocol parsing
294 * Assuming the incoming packets format as
295 * Ethernet2 + optional single VLAN
297 * + udp or tcp or sctp or others
299 if (pkt_ol_flags & PKT_RX_IPV4_HDR) {
301 /* Do not support ipv4 option field */
302 l3_len = sizeof(struct ipv4_hdr) ;
304 ipv4_hdr = (struct ipv4_hdr *) (rte_pktmbuf_mtod(mb,
305 unsigned char *) + l2_len);
307 l4_proto = ipv4_hdr->next_proto_id;
309 /* Do not delete, this is required by HW*/
310 ipv4_hdr->hdr_checksum = 0;
312 if (tx_ol_flags & 0x1) {
314 ol_flags |= PKT_TX_IP_CKSUM;
317 /* SW checksum calculation */
318 ipv4_hdr->src_addr++;
319 ipv4_hdr->hdr_checksum = get_ipv4_cksum(ipv4_hdr);
322 if (l4_proto == IPPROTO_UDP) {
323 udp_hdr = (struct udp_hdr*) (rte_pktmbuf_mtod(mb,
324 unsigned char *) + l2_len + l3_len);
325 if (tx_ol_flags & 0x2) {
327 ol_flags |= PKT_TX_UDP_CKSUM;
328 /* Pseudo header sum need be set properly */
329 udp_hdr->dgram_cksum = get_ipv4_psd_sum(ipv4_hdr);
332 /* SW Implementation, clear checksum field first */
333 udp_hdr->dgram_cksum = 0;
334 udp_hdr->dgram_cksum = get_ipv4_udptcp_checksum(ipv4_hdr,
338 else if (l4_proto == IPPROTO_TCP){
339 tcp_hdr = (struct tcp_hdr*) (rte_pktmbuf_mtod(mb,
340 unsigned char *) + l2_len + l3_len);
341 if (tx_ol_flags & 0x4) {
342 ol_flags |= PKT_TX_TCP_CKSUM;
343 tcp_hdr->cksum = get_ipv4_psd_sum(ipv4_hdr);
347 tcp_hdr->cksum = get_ipv4_udptcp_checksum(ipv4_hdr,
351 else if (l4_proto == IPPROTO_SCTP) {
352 sctp_hdr = (struct sctp_hdr*) (rte_pktmbuf_mtod(mb,
353 unsigned char *) + l2_len + l3_len);
355 if (tx_ol_flags & 0x8) {
356 ol_flags |= PKT_TX_SCTP_CKSUM;
359 /* Sanity check, only number of 4 bytes supported */
360 if ((rte_be_to_cpu_16(ipv4_hdr->total_length) % 4) != 0)
361 printf("sctp payload must be a multiple "
362 "of 4 bytes for checksum offload");
366 /* CRC32c sample code available in RFC3309 */
369 /* End of L4 Handling*/
371 else if (pkt_ol_flags & PKT_RX_IPV6_HDR) {
373 ipv6_hdr = (struct ipv6_hdr *) (rte_pktmbuf_mtod(mb,
374 unsigned char *) + l2_len);
375 l3_len = sizeof(struct ipv6_hdr) ;
376 l4_proto = ipv6_hdr->proto;
378 if (l4_proto == IPPROTO_UDP) {
379 udp_hdr = (struct udp_hdr*) (rte_pktmbuf_mtod(mb,
380 unsigned char *) + l2_len + l3_len);
381 if (tx_ol_flags & 0x2) {
383 ol_flags |= PKT_TX_UDP_CKSUM;
384 udp_hdr->dgram_cksum = get_ipv6_psd_sum(ipv6_hdr);
387 /* SW Implementation */
388 /* checksum field need be clear first */
389 udp_hdr->dgram_cksum = 0;
390 udp_hdr->dgram_cksum = get_ipv6_udptcp_checksum(ipv6_hdr,
394 else if (l4_proto == IPPROTO_TCP) {
395 tcp_hdr = (struct tcp_hdr*) (rte_pktmbuf_mtod(mb,
396 unsigned char *) + l2_len + l3_len);
397 if (tx_ol_flags & 0x4) {
398 ol_flags |= PKT_TX_TCP_CKSUM;
399 tcp_hdr->cksum = get_ipv6_psd_sum(ipv6_hdr);
403 tcp_hdr->cksum = get_ipv6_udptcp_checksum(ipv6_hdr,
407 else if (l4_proto == IPPROTO_SCTP) {
408 sctp_hdr = (struct sctp_hdr*) (rte_pktmbuf_mtod(mb,
409 unsigned char *) + l2_len + l3_len);
411 if (tx_ol_flags & 0x8) {
412 ol_flags |= PKT_TX_SCTP_CKSUM;
414 /* Sanity check, only number of 4 bytes supported by HW */
415 if ((rte_be_to_cpu_16(ipv6_hdr->payload_len) % 4) != 0)
416 printf("sctp payload must be a multiple "
417 "of 4 bytes for checksum offload");
420 /* CRC32c sample code available in RFC3309 */
424 printf("Test flow control for 1G PMD \n");
426 /* End of L6 Handling*/
430 printf("Unhandled packet type: %#hx\n", eth_type);
433 /* Combine the packet header write. VLAN is not consider here */
434 mb->pkt.vlan_macip.f.l2_len = l2_len;
435 mb->pkt.vlan_macip.f.l3_len = l3_len;
436 mb->ol_flags = ol_flags;
438 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
439 fs->tx_packets += nb_tx;
440 fs->rx_bad_ip_csum += rx_bad_ip_csum;
441 fs->rx_bad_l4_csum += rx_bad_l4_csum;
443 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
444 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
446 if (unlikely(nb_tx < nb_rx)) {
447 fs->fwd_dropped += (nb_rx - nb_tx);
449 rte_pktmbuf_free(pkts_burst[nb_tx]);
450 } while (++nb_tx < nb_rx);
452 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
453 end_tsc = rte_rdtsc();
454 core_cycles = (end_tsc - start_tsc);
455 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
460 struct fwd_engine csum_fwd_engine = {
461 .fwd_mode_name = "csum",
462 .port_fwd_begin = NULL,
463 .port_fwd_end = NULL,
464 .packet_fwd = pkt_burst_checksum_forward,