4 * Copyright(c) 2010-2016 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.
38 #include <sys/types.h>
40 #include <sys/queue.h>
47 #include <rte_common.h>
48 #include <rte_byteorder.h>
50 #include <rte_malloc.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
55 #include <rte_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
69 #include <rte_mempool.h>
74 #include <rte_string_fns.h>
75 #include <rte_timer.h>
76 #include <rte_power.h>
78 #include <rte_spinlock.h>
80 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
82 #define MAX_PKT_BURST 32
84 #define MIN_ZERO_POLL_COUNT 10
86 /* around 100ms at 2 Ghz */
87 #define TIMER_RESOLUTION_CYCLES 200000000ULL
89 #define TIMER_NUMBER_PER_SECOND 10
91 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
92 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
94 #define APP_LOOKUP_EXACT_MATCH 0
95 #define APP_LOOKUP_LPM 1
96 #define DO_RFC_1812_CHECKS
98 #ifndef APP_LOOKUP_METHOD
99 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
102 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
103 #include <rte_hash.h>
104 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
107 #error "APP_LOOKUP_METHOD set to incorrect value"
111 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
112 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
113 #define IPv6_BYTES(addr) \
114 addr[0], addr[1], addr[2], addr[3], \
115 addr[4], addr[5], addr[6], addr[7], \
116 addr[8], addr[9], addr[10], addr[11],\
117 addr[12], addr[13],addr[14], addr[15]
120 #define MAX_JUMBO_PKT_LEN 9600
122 #define IPV6_ADDR_LEN 16
124 #define MEMPOOL_CACHE_SIZE 256
127 * This expression is used to calculate the number of mbufs needed depending on
128 * user input, taking into account memory for rx and tx hardware rings, cache
129 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
130 * NB_MBUF never goes below a minimum value of 8192.
133 #define NB_MBUF RTE_MAX ( \
134 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
135 nb_ports*nb_lcores*MAX_PKT_BURST + \
136 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
137 nb_lcores*MEMPOOL_CACHE_SIZE), \
140 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
144 /* Configure how many packets ahead to prefetch, when reading packets */
145 #define PREFETCH_OFFSET 3
148 * Configurable number of RX/TX ring descriptors
150 #define RTE_TEST_RX_DESC_DEFAULT 512
151 #define RTE_TEST_TX_DESC_DEFAULT 512
152 static uint16_t nb_rxd
= RTE_TEST_RX_DESC_DEFAULT
;
153 static uint16_t nb_txd
= RTE_TEST_TX_DESC_DEFAULT
;
155 /* ethernet addresses of ports */
156 static struct ether_addr ports_eth_addr
[RTE_MAX_ETHPORTS
];
158 /* ethernet addresses of ports */
159 static rte_spinlock_t locks
[RTE_MAX_ETHPORTS
];
161 /* mask of enabled ports */
162 static uint32_t enabled_port_mask
= 0;
163 /* Ports set in promiscuous mode off by default. */
164 static int promiscuous_on
= 0;
165 /* NUMA is enabled by default. */
166 static int numa_on
= 1;
167 static int parse_ptype
; /**< Parse packet type using rx callback, and */
168 /**< disabled by default */
170 enum freq_scale_hint_t
178 struct lcore_rx_queue
{
181 enum freq_scale_hint_t freq_up_hint
;
182 uint32_t zero_rx_packet_count
;
184 } __rte_cache_aligned
;
186 #define MAX_RX_QUEUE_PER_LCORE 16
187 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
188 #define MAX_RX_QUEUE_PER_PORT 128
190 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
193 #define MAX_LCORE_PARAMS 1024
194 struct lcore_params
{
198 } __rte_cache_aligned
;
200 static struct lcore_params lcore_params_array
[MAX_LCORE_PARAMS
];
201 static struct lcore_params lcore_params_array_default
[] = {
213 static struct lcore_params
* lcore_params
= lcore_params_array_default
;
214 static uint16_t nb_lcore_params
= sizeof(lcore_params_array_default
) /
215 sizeof(lcore_params_array_default
[0]);
217 static struct rte_eth_conf port_conf
= {
219 .mq_mode
= ETH_MQ_RX_RSS
,
220 .max_rx_pkt_len
= ETHER_MAX_LEN
,
222 .header_split
= 0, /**< Header Split disabled */
223 .hw_ip_checksum
= 1, /**< IP checksum offload enabled */
224 .hw_vlan_filter
= 0, /**< VLAN filtering disabled */
225 .jumbo_frame
= 0, /**< Jumbo Frame Support disabled */
226 .hw_strip_crc
= 1, /**< CRC stripped by hardware */
231 .rss_hf
= ETH_RSS_UDP
,
235 .mq_mode
= ETH_MQ_TX_NONE
,
243 static struct rte_mempool
* pktmbuf_pool
[NB_SOCKETS
];
246 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
248 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
249 #include <rte_hash_crc.h>
250 #define DEFAULT_HASH_FUNC rte_hash_crc
252 #include <rte_jhash.h>
253 #define DEFAULT_HASH_FUNC rte_jhash
262 } __attribute__((__packed__
));
265 uint8_t ip_dst
[IPV6_ADDR_LEN
];
266 uint8_t ip_src
[IPV6_ADDR_LEN
];
270 } __attribute__((__packed__
));
272 struct ipv4_l3fwd_route
{
273 struct ipv4_5tuple key
;
277 struct ipv6_l3fwd_route
{
278 struct ipv6_5tuple key
;
282 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array
[] = {
283 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP
}, 0},
284 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP
}, 1},
285 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP
}, 2},
286 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP
}, 3},
289 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array
[] = {
292 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
293 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
294 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
295 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
301 typedef struct rte_hash lookup_struct_t
;
302 static lookup_struct_t
*ipv4_l3fwd_lookup_struct
[NB_SOCKETS
];
303 static lookup_struct_t
*ipv6_l3fwd_lookup_struct
[NB_SOCKETS
];
305 #define L3FWD_HASH_ENTRIES 1024
307 #define IPV4_L3FWD_NUM_ROUTES \
308 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
310 #define IPV6_L3FWD_NUM_ROUTES \
311 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
313 static uint8_t ipv4_l3fwd_out_if
[L3FWD_HASH_ENTRIES
] __rte_cache_aligned
;
314 static uint8_t ipv6_l3fwd_out_if
[L3FWD_HASH_ENTRIES
] __rte_cache_aligned
;
317 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
318 struct ipv4_l3fwd_route
{
324 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array
[] = {
325 {IPv4(1,1,1,0), 24, 0},
326 {IPv4(2,1,1,0), 24, 1},
327 {IPv4(3,1,1,0), 24, 2},
328 {IPv4(4,1,1,0), 24, 3},
329 {IPv4(5,1,1,0), 24, 4},
330 {IPv4(6,1,1,0), 24, 5},
331 {IPv4(7,1,1,0), 24, 6},
332 {IPv4(8,1,1,0), 24, 7},
335 #define IPV4_L3FWD_NUM_ROUTES \
336 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
338 #define IPV4_L3FWD_LPM_MAX_RULES 1024
340 typedef struct rte_lpm lookup_struct_t
;
341 static lookup_struct_t
*ipv4_l3fwd_lookup_struct
[NB_SOCKETS
];
346 struct lcore_rx_queue rx_queue_list
[MAX_RX_QUEUE_PER_LCORE
];
348 uint16_t tx_port_id
[RTE_MAX_ETHPORTS
];
349 uint16_t tx_queue_id
[RTE_MAX_ETHPORTS
];
350 struct rte_eth_dev_tx_buffer
*tx_buffer
[RTE_MAX_ETHPORTS
];
351 lookup_struct_t
* ipv4_lookup_struct
;
352 lookup_struct_t
* ipv6_lookup_struct
;
353 } __rte_cache_aligned
;
356 /* total sleep time in ms since last frequency scaling down */
358 /* number of long sleep recently */
359 uint32_t nb_long_sleep
;
360 /* freq. scaling up trend */
362 /* total packet processed recently */
363 uint64_t nb_rx_processed
;
364 /* total iterations looped recently */
365 uint64_t nb_iteration_looped
;
367 } __rte_cache_aligned
;
369 static struct lcore_conf lcore_conf
[RTE_MAX_LCORE
] __rte_cache_aligned
;
370 static struct lcore_stats stats
[RTE_MAX_LCORE
] __rte_cache_aligned
;
371 static struct rte_timer power_timers
[RTE_MAX_LCORE
];
373 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count
);
374 static inline enum freq_scale_hint_t
power_freq_scaleup_heuristic( \
375 unsigned lcore_id
, uint8_t port_id
, uint16_t queue_id
);
377 /* exit signal handler */
379 signal_exit_now(int sigtype
)
382 unsigned int portid
, nb_ports
;
385 if (sigtype
== SIGINT
) {
386 for (lcore_id
= 0; lcore_id
< RTE_MAX_LCORE
; lcore_id
++) {
387 if (rte_lcore_is_enabled(lcore_id
) == 0)
390 /* init power management library */
391 ret
= rte_power_exit(lcore_id
);
393 rte_exit(EXIT_FAILURE
, "Power management "
394 "library de-initialization failed on "
395 "core%u\n", lcore_id
);
398 nb_ports
= rte_eth_dev_count();
399 for (portid
= 0; portid
< nb_ports
; portid
++) {
400 if ((enabled_port_mask
& (1 << portid
)) == 0)
403 rte_eth_dev_stop(portid
);
404 rte_eth_dev_close(portid
);
408 rte_exit(EXIT_SUCCESS
, "User forced exit\n");
411 /* Freqency scale down timer callback */
413 power_timer_cb(__attribute__((unused
)) struct rte_timer
*tim
,
414 __attribute__((unused
)) void *arg
)
417 float sleep_time_ratio
;
418 unsigned lcore_id
= rte_lcore_id();
420 /* accumulate total execution time in us when callback is invoked */
421 sleep_time_ratio
= (float)(stats
[lcore_id
].sleep_time
) /
422 (float)SCALING_PERIOD
;
424 * check whether need to scale down frequency a step if it sleep a lot.
426 if (sleep_time_ratio
>= SCALING_DOWN_TIME_RATIO_THRESHOLD
) {
427 if (rte_power_freq_down
)
428 rte_power_freq_down(lcore_id
);
430 else if ( (unsigned)(stats
[lcore_id
].nb_rx_processed
/
431 stats
[lcore_id
].nb_iteration_looped
) < MAX_PKT_BURST
) {
433 * scale down a step if average packet per iteration less
436 if (rte_power_freq_down
)
437 rte_power_freq_down(lcore_id
);
441 * initialize another timer according to current frequency to ensure
442 * timer interval is relatively fixed.
444 hz
= rte_get_timer_hz();
445 rte_timer_reset(&power_timers
[lcore_id
], hz
/TIMER_NUMBER_PER_SECOND
,
446 SINGLE
, lcore_id
, power_timer_cb
, NULL
);
448 stats
[lcore_id
].nb_rx_processed
= 0;
449 stats
[lcore_id
].nb_iteration_looped
= 0;
451 stats
[lcore_id
].sleep_time
= 0;
454 /* Enqueue a single packet, and send burst if queue is filled */
456 send_single_packet(struct rte_mbuf
*m
, uint8_t port
)
459 struct lcore_conf
*qconf
;
461 lcore_id
= rte_lcore_id();
462 qconf
= &lcore_conf
[lcore_id
];
464 rte_eth_tx_buffer(port
, qconf
->tx_queue_id
[port
],
465 qconf
->tx_buffer
[port
], m
);
470 #ifdef DO_RFC_1812_CHECKS
472 is_valid_ipv4_pkt(struct ipv4_hdr
*pkt
, uint32_t link_len
)
474 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
476 * 1. The packet length reported by the Link Layer must be large
477 * enough to hold the minimum length legal IP datagram (20 bytes).
479 if (link_len
< sizeof(struct ipv4_hdr
))
482 /* 2. The IP checksum must be correct. */
483 /* this is checked in H/W */
486 * 3. The IP version number must be 4. If the version number is not 4
487 * then the packet may be another version of IP, such as IPng or
490 if (((pkt
->version_ihl
) >> 4) != 4)
493 * 4. The IP header length field must be large enough to hold the
494 * minimum length legal IP datagram (20 bytes = 5 words).
496 if ((pkt
->version_ihl
& 0xf) < 5)
500 * 5. The IP total length field must be large enough to hold the IP
501 * datagram header, whose length is specified in the IP header length
504 if (rte_cpu_to_be_16(pkt
->total_length
) < sizeof(struct ipv4_hdr
))
511 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
513 print_ipv4_key(struct ipv4_5tuple key
)
515 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
516 "proto = %d\n", (unsigned)key
.ip_dst
, (unsigned)key
.ip_src
,
517 key
.port_dst
, key
.port_src
, key
.proto
);
520 print_ipv6_key(struct ipv6_5tuple key
)
522 printf( "IP dst = " IPv6_BYTES_FMT
", IP src = " IPv6_BYTES_FMT
", "
523 "port dst = %d, port src = %d, proto = %d\n",
524 IPv6_BYTES(key
.ip_dst
), IPv6_BYTES(key
.ip_src
),
525 key
.port_dst
, key
.port_src
, key
.proto
);
528 static inline uint8_t
529 get_ipv4_dst_port(struct ipv4_hdr
*ipv4_hdr
, uint8_t portid
,
530 lookup_struct_t
* ipv4_l3fwd_lookup_struct
)
532 struct ipv4_5tuple key
;
537 key
.ip_dst
= rte_be_to_cpu_32(ipv4_hdr
->dst_addr
);
538 key
.ip_src
= rte_be_to_cpu_32(ipv4_hdr
->src_addr
);
539 key
.proto
= ipv4_hdr
->next_proto_id
;
541 switch (ipv4_hdr
->next_proto_id
) {
543 tcp
= (struct tcp_hdr
*)((unsigned char *)ipv4_hdr
+
544 sizeof(struct ipv4_hdr
));
545 key
.port_dst
= rte_be_to_cpu_16(tcp
->dst_port
);
546 key
.port_src
= rte_be_to_cpu_16(tcp
->src_port
);
550 udp
= (struct udp_hdr
*)((unsigned char *)ipv4_hdr
+
551 sizeof(struct ipv4_hdr
));
552 key
.port_dst
= rte_be_to_cpu_16(udp
->dst_port
);
553 key
.port_src
= rte_be_to_cpu_16(udp
->src_port
);
562 /* Find destination port */
563 ret
= rte_hash_lookup(ipv4_l3fwd_lookup_struct
, (const void *)&key
);
564 return (uint8_t)((ret
< 0)? portid
: ipv4_l3fwd_out_if
[ret
]);
567 static inline uint8_t
568 get_ipv6_dst_port(struct ipv6_hdr
*ipv6_hdr
, uint8_t portid
,
569 lookup_struct_t
*ipv6_l3fwd_lookup_struct
)
571 struct ipv6_5tuple key
;
576 memcpy(key
.ip_dst
, ipv6_hdr
->dst_addr
, IPV6_ADDR_LEN
);
577 memcpy(key
.ip_src
, ipv6_hdr
->src_addr
, IPV6_ADDR_LEN
);
579 key
.proto
= ipv6_hdr
->proto
;
581 switch (ipv6_hdr
->proto
) {
583 tcp
= (struct tcp_hdr
*)((unsigned char *) ipv6_hdr
+
584 sizeof(struct ipv6_hdr
));
585 key
.port_dst
= rte_be_to_cpu_16(tcp
->dst_port
);
586 key
.port_src
= rte_be_to_cpu_16(tcp
->src_port
);
590 udp
= (struct udp_hdr
*)((unsigned char *) ipv6_hdr
+
591 sizeof(struct ipv6_hdr
));
592 key
.port_dst
= rte_be_to_cpu_16(udp
->dst_port
);
593 key
.port_src
= rte_be_to_cpu_16(udp
->src_port
);
602 /* Find destination port */
603 ret
= rte_hash_lookup(ipv6_l3fwd_lookup_struct
, (const void *)&key
);
604 return (uint8_t)((ret
< 0)? portid
: ipv6_l3fwd_out_if
[ret
]);
608 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
609 static inline uint8_t
610 get_ipv4_dst_port(struct ipv4_hdr
*ipv4_hdr
, uint8_t portid
,
611 lookup_struct_t
*ipv4_l3fwd_lookup_struct
)
615 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct
,
616 rte_be_to_cpu_32(ipv4_hdr
->dst_addr
), &next_hop
) == 0)?
622 parse_ptype_one(struct rte_mbuf
*m
)
624 struct ether_hdr
*eth_hdr
;
625 uint32_t packet_type
= RTE_PTYPE_UNKNOWN
;
628 eth_hdr
= rte_pktmbuf_mtod(m
, struct ether_hdr
*);
629 ether_type
= eth_hdr
->ether_type
;
630 if (ether_type
== rte_cpu_to_be_16(ETHER_TYPE_IPv4
))
631 packet_type
|= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN
;
632 else if (ether_type
== rte_cpu_to_be_16(ETHER_TYPE_IPv6
))
633 packet_type
|= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN
;
635 m
->packet_type
= packet_type
;
639 cb_parse_ptype(uint8_t port __rte_unused
, uint16_t queue __rte_unused
,
640 struct rte_mbuf
*pkts
[], uint16_t nb_pkts
,
641 uint16_t max_pkts __rte_unused
,
642 void *user_param __rte_unused
)
646 for (i
= 0; i
< nb_pkts
; ++i
)
647 parse_ptype_one(pkts
[i
]);
653 add_cb_parse_ptype(uint8_t portid
, uint16_t queueid
)
655 printf("Port %d: softly parse packet type info\n", portid
);
656 if (rte_eth_add_rx_callback(portid
, queueid
, cb_parse_ptype
, NULL
))
659 printf("Failed to add rx callback: port=%d\n", portid
);
664 l3fwd_simple_forward(struct rte_mbuf
*m
, uint8_t portid
,
665 struct lcore_conf
*qconf
)
667 struct ether_hdr
*eth_hdr
;
668 struct ipv4_hdr
*ipv4_hdr
;
672 eth_hdr
= rte_pktmbuf_mtod(m
, struct ether_hdr
*);
674 if (RTE_ETH_IS_IPV4_HDR(m
->packet_type
)) {
675 /* Handle IPv4 headers.*/
677 rte_pktmbuf_mtod_offset(m
, struct ipv4_hdr
*,
678 sizeof(struct ether_hdr
));
680 #ifdef DO_RFC_1812_CHECKS
681 /* Check to make sure the packet is valid (RFC1812) */
682 if (is_valid_ipv4_pkt(ipv4_hdr
, m
->pkt_len
) < 0) {
688 dst_port
= get_ipv4_dst_port(ipv4_hdr
, portid
,
689 qconf
->ipv4_lookup_struct
);
690 if (dst_port
>= RTE_MAX_ETHPORTS
||
691 (enabled_port_mask
& 1 << dst_port
) == 0)
694 /* 02:00:00:00:00:xx */
695 d_addr_bytes
= ð_hdr
->d_addr
.addr_bytes
[0];
696 *((uint64_t *)d_addr_bytes
) =
697 0x000000000002 + ((uint64_t)dst_port
<< 40);
699 #ifdef DO_RFC_1812_CHECKS
700 /* Update time to live and header checksum */
701 --(ipv4_hdr
->time_to_live
);
702 ++(ipv4_hdr
->hdr_checksum
);
706 ether_addr_copy(&ports_eth_addr
[dst_port
], ð_hdr
->s_addr
);
708 send_single_packet(m
, dst_port
);
709 } else if (RTE_ETH_IS_IPV6_HDR(m
->packet_type
)) {
710 /* Handle IPv6 headers.*/
711 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
712 struct ipv6_hdr
*ipv6_hdr
;
715 rte_pktmbuf_mtod_offset(m
, struct ipv6_hdr
*,
716 sizeof(struct ether_hdr
));
718 dst_port
= get_ipv6_dst_port(ipv6_hdr
, portid
,
719 qconf
->ipv6_lookup_struct
);
721 if (dst_port
>= RTE_MAX_ETHPORTS
||
722 (enabled_port_mask
& 1 << dst_port
) == 0)
725 /* 02:00:00:00:00:xx */
726 d_addr_bytes
= ð_hdr
->d_addr
.addr_bytes
[0];
727 *((uint64_t *)d_addr_bytes
) =
728 0x000000000002 + ((uint64_t)dst_port
<< 40);
731 ether_addr_copy(&ports_eth_addr
[dst_port
], ð_hdr
->s_addr
);
733 send_single_packet(m
, dst_port
);
735 /* We don't currently handle IPv6 packets in LPM mode. */
743 #define MINIMUM_SLEEP_TIME 1
744 #define SUSPEND_THRESHOLD 300
746 static inline uint32_t
747 power_idle_heuristic(uint32_t zero_rx_packet_count
)
749 /* If zero count is less than 100, sleep 1us */
750 if (zero_rx_packet_count
< SUSPEND_THRESHOLD
)
751 return MINIMUM_SLEEP_TIME
;
752 /* If zero count is less than 1000, sleep 100 us which is the
753 minimum latency switching from C3/C6 to C0
756 return SUSPEND_THRESHOLD
;
761 static inline enum freq_scale_hint_t
762 power_freq_scaleup_heuristic(unsigned lcore_id
,
767 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
770 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
771 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
772 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
773 #define FREQ_UP_TREND1_ACC 1
774 #define FREQ_UP_TREND2_ACC 100
775 #define FREQ_UP_THRESHOLD 10000
777 if (likely(rte_eth_rx_descriptor_done(port_id
, queue_id
,
778 FREQ_GEAR3_RX_PACKET_THRESHOLD
) > 0)) {
779 stats
[lcore_id
].trend
= 0;
781 } else if (likely(rte_eth_rx_descriptor_done(port_id
, queue_id
,
782 FREQ_GEAR2_RX_PACKET_THRESHOLD
) > 0))
783 stats
[lcore_id
].trend
+= FREQ_UP_TREND2_ACC
;
784 else if (likely(rte_eth_rx_descriptor_done(port_id
, queue_id
,
785 FREQ_GEAR1_RX_PACKET_THRESHOLD
) > 0))
786 stats
[lcore_id
].trend
+= FREQ_UP_TREND1_ACC
;
788 if (likely(stats
[lcore_id
].trend
> FREQ_UP_THRESHOLD
)) {
789 stats
[lcore_id
].trend
= 0;
797 * force polling thread sleep until one-shot rx interrupt triggers
806 sleep_until_rx_interrupt(int num
)
808 struct rte_epoll_event event
[num
];
810 uint8_t port_id
, queue_id
;
813 RTE_LOG(INFO
, L3FWD_POWER
,
814 "lcore %u sleeps until interrupt triggers\n",
817 n
= rte_epoll_wait(RTE_EPOLL_PER_THREAD
, event
, num
, -1);
818 for (i
= 0; i
< n
; i
++) {
819 data
= event
[i
].epdata
.data
;
820 port_id
= ((uintptr_t)data
) >> CHAR_BIT
;
821 queue_id
= ((uintptr_t)data
) &
822 RTE_LEN2MASK(CHAR_BIT
, uint8_t);
823 rte_eth_dev_rx_intr_disable(port_id
, queue_id
);
824 RTE_LOG(INFO
, L3FWD_POWER
,
825 "lcore %u is waked up from rx interrupt on"
826 " port %d queue %d\n",
827 rte_lcore_id(), port_id
, queue_id
);
833 static void turn_on_intr(struct lcore_conf
*qconf
)
836 struct lcore_rx_queue
*rx_queue
;
837 uint8_t port_id
, queue_id
;
839 for (i
= 0; i
< qconf
->n_rx_queue
; ++i
) {
840 rx_queue
= &(qconf
->rx_queue_list
[i
]);
841 port_id
= rx_queue
->port_id
;
842 queue_id
= rx_queue
->queue_id
;
844 rte_spinlock_lock(&(locks
[port_id
]));
845 rte_eth_dev_rx_intr_enable(port_id
, queue_id
);
846 rte_spinlock_unlock(&(locks
[port_id
]));
850 static int event_register(struct lcore_conf
*qconf
)
852 struct lcore_rx_queue
*rx_queue
;
853 uint8_t portid
, queueid
;
858 for (i
= 0; i
< qconf
->n_rx_queue
; ++i
) {
859 rx_queue
= &(qconf
->rx_queue_list
[i
]);
860 portid
= rx_queue
->port_id
;
861 queueid
= rx_queue
->queue_id
;
862 data
= portid
<< CHAR_BIT
| queueid
;
864 ret
= rte_eth_dev_rx_intr_ctl_q(portid
, queueid
,
865 RTE_EPOLL_PER_THREAD
,
867 (void *)((uintptr_t)data
));
875 /* main processing loop */
877 main_loop(__attribute__((unused
)) void *dummy
)
879 struct rte_mbuf
*pkts_burst
[MAX_PKT_BURST
];
881 uint64_t prev_tsc
, diff_tsc
, cur_tsc
;
882 uint64_t prev_tsc_power
= 0, cur_tsc_power
, diff_tsc_power
;
884 uint8_t portid
, queueid
;
885 struct lcore_conf
*qconf
;
886 struct lcore_rx_queue
*rx_queue
;
887 enum freq_scale_hint_t lcore_scaleup_hint
;
888 uint32_t lcore_rx_idle_count
= 0;
889 uint32_t lcore_idle_hint
= 0;
892 const uint64_t drain_tsc
= (rte_get_tsc_hz() + US_PER_S
- 1) / US_PER_S
* BURST_TX_DRAIN_US
;
896 lcore_id
= rte_lcore_id();
897 qconf
= &lcore_conf
[lcore_id
];
899 if (qconf
->n_rx_queue
== 0) {
900 RTE_LOG(INFO
, L3FWD_POWER
, "lcore %u has nothing to do\n", lcore_id
);
904 RTE_LOG(INFO
, L3FWD_POWER
, "entering main loop on lcore %u\n", lcore_id
);
906 for (i
= 0; i
< qconf
->n_rx_queue
; i
++) {
907 portid
= qconf
->rx_queue_list
[i
].port_id
;
908 queueid
= qconf
->rx_queue_list
[i
].queue_id
;
909 RTE_LOG(INFO
, L3FWD_POWER
, " -- lcoreid=%u portid=%hhu "
910 "rxqueueid=%hhu\n", lcore_id
, portid
, queueid
);
913 /* add into event wait list */
914 if (event_register(qconf
) == 0)
917 RTE_LOG(INFO
, L3FWD_POWER
, "RX interrupt won't enable.\n");
920 stats
[lcore_id
].nb_iteration_looped
++;
922 cur_tsc
= rte_rdtsc();
923 cur_tsc_power
= cur_tsc
;
926 * TX burst queue drain
928 diff_tsc
= cur_tsc
- prev_tsc
;
929 if (unlikely(diff_tsc
> drain_tsc
)) {
930 for (i
= 0; i
< qconf
->n_tx_port
; ++i
) {
931 portid
= qconf
->tx_port_id
[i
];
932 rte_eth_tx_buffer_flush(portid
,
933 qconf
->tx_queue_id
[portid
],
934 qconf
->tx_buffer
[portid
]);
939 diff_tsc_power
= cur_tsc_power
- prev_tsc_power
;
940 if (diff_tsc_power
> TIMER_RESOLUTION_CYCLES
) {
942 prev_tsc_power
= cur_tsc_power
;
947 * Read packet from RX queues
949 lcore_scaleup_hint
= FREQ_CURRENT
;
950 lcore_rx_idle_count
= 0;
951 for (i
= 0; i
< qconf
->n_rx_queue
; ++i
) {
952 rx_queue
= &(qconf
->rx_queue_list
[i
]);
953 rx_queue
->idle_hint
= 0;
954 portid
= rx_queue
->port_id
;
955 queueid
= rx_queue
->queue_id
;
957 nb_rx
= rte_eth_rx_burst(portid
, queueid
, pkts_burst
,
960 stats
[lcore_id
].nb_rx_processed
+= nb_rx
;
961 if (unlikely(nb_rx
== 0)) {
963 * no packet received from rx queue, try to
964 * sleep for a while forcing CPU enter deeper
967 rx_queue
->zero_rx_packet_count
++;
969 if (rx_queue
->zero_rx_packet_count
<=
973 rx_queue
->idle_hint
= power_idle_heuristic(\
974 rx_queue
->zero_rx_packet_count
);
975 lcore_rx_idle_count
++;
977 rx_queue
->zero_rx_packet_count
= 0;
980 * do not scale up frequency immediately as
981 * user to kernel space communication is costly
982 * which might impact packet I/O for received
985 rx_queue
->freq_up_hint
=
986 power_freq_scaleup_heuristic(lcore_id
,
990 /* Prefetch first packets */
991 for (j
= 0; j
< PREFETCH_OFFSET
&& j
< nb_rx
; j
++) {
992 rte_prefetch0(rte_pktmbuf_mtod(
993 pkts_burst
[j
], void *));
996 /* Prefetch and forward already prefetched packets */
997 for (j
= 0; j
< (nb_rx
- PREFETCH_OFFSET
); j
++) {
998 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst
[
999 j
+ PREFETCH_OFFSET
], void *));
1000 l3fwd_simple_forward(pkts_burst
[j
], portid
,
1004 /* Forward remaining prefetched packets */
1005 for (; j
< nb_rx
; j
++) {
1006 l3fwd_simple_forward(pkts_burst
[j
], portid
,
1011 if (likely(lcore_rx_idle_count
!= qconf
->n_rx_queue
)) {
1012 for (i
= 1, lcore_scaleup_hint
=
1013 qconf
->rx_queue_list
[0].freq_up_hint
;
1014 i
< qconf
->n_rx_queue
; ++i
) {
1015 rx_queue
= &(qconf
->rx_queue_list
[i
]);
1016 if (rx_queue
->freq_up_hint
>
1018 lcore_scaleup_hint
=
1019 rx_queue
->freq_up_hint
;
1022 if (lcore_scaleup_hint
== FREQ_HIGHEST
) {
1023 if (rte_power_freq_max
)
1024 rte_power_freq_max(lcore_id
);
1025 } else if (lcore_scaleup_hint
== FREQ_HIGHER
) {
1026 if (rte_power_freq_up
)
1027 rte_power_freq_up(lcore_id
);
1031 * All Rx queues empty in recent consecutive polls,
1032 * sleep in a conservative manner, meaning sleep as
1035 for (i
= 1, lcore_idle_hint
=
1036 qconf
->rx_queue_list
[0].idle_hint
;
1037 i
< qconf
->n_rx_queue
; ++i
) {
1038 rx_queue
= &(qconf
->rx_queue_list
[i
]);
1039 if (rx_queue
->idle_hint
< lcore_idle_hint
)
1040 lcore_idle_hint
= rx_queue
->idle_hint
;
1043 if (lcore_idle_hint
< SUSPEND_THRESHOLD
)
1045 * execute "pause" instruction to avoid context
1046 * switch which generally take hundred of
1047 * microseconds for short sleep.
1049 rte_delay_us(lcore_idle_hint
);
1051 /* suspend until rx interrupt trigges */
1053 turn_on_intr(qconf
);
1054 sleep_until_rx_interrupt(
1057 /* start receiving packets immediately */
1060 stats
[lcore_id
].sleep_time
+= lcore_idle_hint
;
1066 check_lcore_params(void)
1068 uint8_t queue
, lcore
;
1072 for (i
= 0; i
< nb_lcore_params
; ++i
) {
1073 queue
= lcore_params
[i
].queue_id
;
1074 if (queue
>= MAX_RX_QUEUE_PER_PORT
) {
1075 printf("invalid queue number: %hhu\n", queue
);
1078 lcore
= lcore_params
[i
].lcore_id
;
1079 if (!rte_lcore_is_enabled(lcore
)) {
1080 printf("error: lcore %hhu is not enabled in lcore "
1084 if ((socketid
= rte_lcore_to_socket_id(lcore
) != 0) &&
1086 printf("warning: lcore %hhu is on socket %d with numa "
1087 "off\n", lcore
, socketid
);
1094 check_port_config(const unsigned nb_ports
)
1099 for (i
= 0; i
< nb_lcore_params
; ++i
) {
1100 portid
= lcore_params
[i
].port_id
;
1101 if ((enabled_port_mask
& (1 << portid
)) == 0) {
1102 printf("port %u is not enabled in port mask\n",
1106 if (portid
>= nb_ports
) {
1107 printf("port %u is not present on the board\n",
1116 get_port_n_rx_queues(const uint8_t port
)
1121 for (i
= 0; i
< nb_lcore_params
; ++i
) {
1122 if (lcore_params
[i
].port_id
== port
&&
1123 lcore_params
[i
].queue_id
> queue
)
1124 queue
= lcore_params
[i
].queue_id
;
1126 return (uint8_t)(++queue
);
1130 init_lcore_rx_queues(void)
1132 uint16_t i
, nb_rx_queue
;
1135 for (i
= 0; i
< nb_lcore_params
; ++i
) {
1136 lcore
= lcore_params
[i
].lcore_id
;
1137 nb_rx_queue
= lcore_conf
[lcore
].n_rx_queue
;
1138 if (nb_rx_queue
>= MAX_RX_QUEUE_PER_LCORE
) {
1139 printf("error: too many queues (%u) for lcore: %u\n",
1140 (unsigned)nb_rx_queue
+ 1, (unsigned)lcore
);
1143 lcore_conf
[lcore
].rx_queue_list
[nb_rx_queue
].port_id
=
1144 lcore_params
[i
].port_id
;
1145 lcore_conf
[lcore
].rx_queue_list
[nb_rx_queue
].queue_id
=
1146 lcore_params
[i
].queue_id
;
1147 lcore_conf
[lcore
].n_rx_queue
++;
1155 print_usage(const char *prgname
)
1157 printf ("%s [EAL options] -- -p PORTMASK -P"
1158 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1159 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1160 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1161 " -P : enable promiscuous mode\n"
1162 " --config (port,queue,lcore): rx queues configuration\n"
1163 " --no-numa: optional, disable numa awareness\n"
1164 " --enable-jumbo: enable jumbo frame"
1165 " which max packet len is PKTLEN in decimal (64-9600)\n"
1166 " --parse-ptype: parse packet type by software\n",
1170 static int parse_max_pkt_len(const char *pktlen
)
1175 /* parse decimal string */
1176 len
= strtoul(pktlen
, &end
, 10);
1177 if ((pktlen
[0] == '\0') || (end
== NULL
) || (*end
!= '\0'))
1187 parse_portmask(const char *portmask
)
1192 /* parse hexadecimal string */
1193 pm
= strtoul(portmask
, &end
, 16);
1194 if ((portmask
[0] == '\0') || (end
== NULL
) || (*end
!= '\0'))
1204 parse_config(const char *q_arg
)
1207 const char *p
, *p0
= q_arg
;
1215 unsigned long int_fld
[_NUM_FLD
];
1216 char *str_fld
[_NUM_FLD
];
1220 nb_lcore_params
= 0;
1222 while ((p
= strchr(p0
,'(')) != NULL
) {
1224 if((p0
= strchr(p
,')')) == NULL
)
1228 if(size
>= sizeof(s
))
1231 snprintf(s
, sizeof(s
), "%.*s", size
, p
);
1232 if (rte_strsplit(s
, sizeof(s
), str_fld
, _NUM_FLD
, ',') !=
1235 for (i
= 0; i
< _NUM_FLD
; i
++){
1237 int_fld
[i
] = strtoul(str_fld
[i
], &end
, 0);
1238 if (errno
!= 0 || end
== str_fld
[i
] || int_fld
[i
] >
1242 if (nb_lcore_params
>= MAX_LCORE_PARAMS
) {
1243 printf("exceeded max number of lcore params: %hu\n",
1247 lcore_params_array
[nb_lcore_params
].port_id
=
1248 (uint8_t)int_fld
[FLD_PORT
];
1249 lcore_params_array
[nb_lcore_params
].queue_id
=
1250 (uint8_t)int_fld
[FLD_QUEUE
];
1251 lcore_params_array
[nb_lcore_params
].lcore_id
=
1252 (uint8_t)int_fld
[FLD_LCORE
];
1255 lcore_params
= lcore_params_array
;
1260 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1262 /* Parse the argument given in the command line of the application */
1264 parse_args(int argc
, char **argv
)
1269 char *prgname
= argv
[0];
1270 static struct option lgopts
[] = {
1271 {"config", 1, 0, 0},
1272 {"no-numa", 0, 0, 0},
1273 {"enable-jumbo", 0, 0, 0},
1274 {CMD_LINE_OPT_PARSE_PTYPE
, 0, 0, 0},
1280 while ((opt
= getopt_long(argc
, argvopt
, "p:P",
1281 lgopts
, &option_index
)) != EOF
) {
1286 enabled_port_mask
= parse_portmask(optarg
);
1287 if (enabled_port_mask
== 0) {
1288 printf("invalid portmask\n");
1289 print_usage(prgname
);
1294 printf("Promiscuous mode selected\n");
1300 if (!strncmp(lgopts
[option_index
].name
, "config", 6)) {
1301 ret
= parse_config(optarg
);
1303 printf("invalid config\n");
1304 print_usage(prgname
);
1309 if (!strncmp(lgopts
[option_index
].name
,
1311 printf("numa is disabled \n");
1315 if (!strncmp(lgopts
[option_index
].name
,
1316 "enable-jumbo", 12)) {
1317 struct option lenopts
=
1318 {"max-pkt-len", required_argument
, \
1321 printf("jumbo frame is enabled \n");
1322 port_conf
.rxmode
.jumbo_frame
= 1;
1325 * if no max-pkt-len set, use the default value
1328 if (0 == getopt_long(argc
, argvopt
, "",
1329 &lenopts
, &option_index
)) {
1330 ret
= parse_max_pkt_len(optarg
);
1332 (ret
> MAX_JUMBO_PKT_LEN
)){
1333 printf("invalid packet "
1335 print_usage(prgname
);
1338 port_conf
.rxmode
.max_rx_pkt_len
= ret
;
1340 printf("set jumbo frame "
1341 "max packet length to %u\n",
1342 (unsigned int)port_conf
.rxmode
.max_rx_pkt_len
);
1345 if (!strncmp(lgopts
[option_index
].name
,
1346 CMD_LINE_OPT_PARSE_PTYPE
,
1347 sizeof(CMD_LINE_OPT_PARSE_PTYPE
))) {
1348 printf("soft parse-ptype is enabled\n");
1355 print_usage(prgname
);
1361 argv
[optind
-1] = prgname
;
1364 optind
= 1; /* reset getopt lib */
1369 print_ethaddr(const char *name
, const struct ether_addr
*eth_addr
)
1371 char buf
[ETHER_ADDR_FMT_SIZE
];
1372 ether_format_addr(buf
, ETHER_ADDR_FMT_SIZE
, eth_addr
);
1373 printf("%s%s", name
, buf
);
1376 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1378 setup_hash(int socketid
)
1380 struct rte_hash_parameters ipv4_l3fwd_hash_params
= {
1382 .entries
= L3FWD_HASH_ENTRIES
,
1383 .key_len
= sizeof(struct ipv4_5tuple
),
1384 .hash_func
= DEFAULT_HASH_FUNC
,
1385 .hash_func_init_val
= 0,
1388 struct rte_hash_parameters ipv6_l3fwd_hash_params
= {
1390 .entries
= L3FWD_HASH_ENTRIES
,
1391 .key_len
= sizeof(struct ipv6_5tuple
),
1392 .hash_func
= DEFAULT_HASH_FUNC
,
1393 .hash_func_init_val
= 0,
1400 /* create ipv4 hash */
1401 snprintf(s
, sizeof(s
), "ipv4_l3fwd_hash_%d", socketid
);
1402 ipv4_l3fwd_hash_params
.name
= s
;
1403 ipv4_l3fwd_hash_params
.socket_id
= socketid
;
1404 ipv4_l3fwd_lookup_struct
[socketid
] =
1405 rte_hash_create(&ipv4_l3fwd_hash_params
);
1406 if (ipv4_l3fwd_lookup_struct
[socketid
] == NULL
)
1407 rte_exit(EXIT_FAILURE
, "Unable to create the l3fwd hash on "
1408 "socket %d\n", socketid
);
1410 /* create ipv6 hash */
1411 snprintf(s
, sizeof(s
), "ipv6_l3fwd_hash_%d", socketid
);
1412 ipv6_l3fwd_hash_params
.name
= s
;
1413 ipv6_l3fwd_hash_params
.socket_id
= socketid
;
1414 ipv6_l3fwd_lookup_struct
[socketid
] =
1415 rte_hash_create(&ipv6_l3fwd_hash_params
);
1416 if (ipv6_l3fwd_lookup_struct
[socketid
] == NULL
)
1417 rte_exit(EXIT_FAILURE
, "Unable to create the l3fwd hash on "
1418 "socket %d\n", socketid
);
1421 /* populate the ipv4 hash */
1422 for (i
= 0; i
< IPV4_L3FWD_NUM_ROUTES
; i
++) {
1423 ret
= rte_hash_add_key (ipv4_l3fwd_lookup_struct
[socketid
],
1424 (void *) &ipv4_l3fwd_route_array
[i
].key
);
1426 rte_exit(EXIT_FAILURE
, "Unable to add entry %u to the"
1427 "l3fwd hash on socket %d\n", i
, socketid
);
1429 ipv4_l3fwd_out_if
[ret
] = ipv4_l3fwd_route_array
[i
].if_out
;
1430 printf("Hash: Adding key\n");
1431 print_ipv4_key(ipv4_l3fwd_route_array
[i
].key
);
1434 /* populate the ipv6 hash */
1435 for (i
= 0; i
< IPV6_L3FWD_NUM_ROUTES
; i
++) {
1436 ret
= rte_hash_add_key (ipv6_l3fwd_lookup_struct
[socketid
],
1437 (void *) &ipv6_l3fwd_route_array
[i
].key
);
1439 rte_exit(EXIT_FAILURE
, "Unable to add entry %u to the"
1440 "l3fwd hash on socket %d\n", i
, socketid
);
1442 ipv6_l3fwd_out_if
[ret
] = ipv6_l3fwd_route_array
[i
].if_out
;
1443 printf("Hash: Adding key\n");
1444 print_ipv6_key(ipv6_l3fwd_route_array
[i
].key
);
1449 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1451 setup_lpm(int socketid
)
1457 /* create the LPM table */
1458 struct rte_lpm_config lpm_ipv4_config
;
1460 lpm_ipv4_config
.max_rules
= IPV4_L3FWD_LPM_MAX_RULES
;
1461 lpm_ipv4_config
.number_tbl8s
= 256;
1462 lpm_ipv4_config
.flags
= 0;
1464 snprintf(s
, sizeof(s
), "IPV4_L3FWD_LPM_%d", socketid
);
1465 ipv4_l3fwd_lookup_struct
[socketid
] =
1466 rte_lpm_create(s
, socketid
, &lpm_ipv4_config
);
1467 if (ipv4_l3fwd_lookup_struct
[socketid
] == NULL
)
1468 rte_exit(EXIT_FAILURE
, "Unable to create the l3fwd LPM table"
1469 " on socket %d\n", socketid
);
1471 /* populate the LPM table */
1472 for (i
= 0; i
< IPV4_L3FWD_NUM_ROUTES
; i
++) {
1473 ret
= rte_lpm_add(ipv4_l3fwd_lookup_struct
[socketid
],
1474 ipv4_l3fwd_route_array
[i
].ip
,
1475 ipv4_l3fwd_route_array
[i
].depth
,
1476 ipv4_l3fwd_route_array
[i
].if_out
);
1479 rte_exit(EXIT_FAILURE
, "Unable to add entry %u to the "
1480 "l3fwd LPM table on socket %d\n",
1484 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1485 (unsigned)ipv4_l3fwd_route_array
[i
].ip
,
1486 ipv4_l3fwd_route_array
[i
].depth
,
1487 ipv4_l3fwd_route_array
[i
].if_out
);
1493 init_mem(unsigned nb_mbuf
)
1495 struct lcore_conf
*qconf
;
1500 for (lcore_id
= 0; lcore_id
< RTE_MAX_LCORE
; lcore_id
++) {
1501 if (rte_lcore_is_enabled(lcore_id
) == 0)
1505 socketid
= rte_lcore_to_socket_id(lcore_id
);
1509 if (socketid
>= NB_SOCKETS
) {
1510 rte_exit(EXIT_FAILURE
, "Socket %d of lcore %u is "
1511 "out of range %d\n", socketid
,
1512 lcore_id
, NB_SOCKETS
);
1514 if (pktmbuf_pool
[socketid
] == NULL
) {
1515 snprintf(s
, sizeof(s
), "mbuf_pool_%d", socketid
);
1516 pktmbuf_pool
[socketid
] =
1517 rte_pktmbuf_pool_create(s
, nb_mbuf
,
1518 MEMPOOL_CACHE_SIZE
, 0,
1519 RTE_MBUF_DEFAULT_BUF_SIZE
,
1521 if (pktmbuf_pool
[socketid
] == NULL
)
1522 rte_exit(EXIT_FAILURE
,
1523 "Cannot init mbuf pool on socket %d\n",
1526 printf("Allocated mbuf pool on socket %d\n",
1529 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1530 setup_lpm(socketid
);
1532 setup_hash(socketid
);
1535 qconf
= &lcore_conf
[lcore_id
];
1536 qconf
->ipv4_lookup_struct
= ipv4_l3fwd_lookup_struct
[socketid
];
1537 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1538 qconf
->ipv6_lookup_struct
= ipv6_l3fwd_lookup_struct
[socketid
];
1544 /* Check the link status of all ports in up to 9s, and print them finally */
1546 check_all_ports_link_status(uint8_t port_num
, uint32_t port_mask
)
1548 #define CHECK_INTERVAL 100 /* 100ms */
1549 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1550 uint8_t portid
, count
, all_ports_up
, print_flag
= 0;
1551 struct rte_eth_link link
;
1553 printf("\nChecking link status");
1555 for (count
= 0; count
<= MAX_CHECK_TIME
; count
++) {
1557 for (portid
= 0; portid
< port_num
; portid
++) {
1558 if ((port_mask
& (1 << portid
)) == 0)
1560 memset(&link
, 0, sizeof(link
));
1561 rte_eth_link_get_nowait(portid
, &link
);
1562 /* print link status if flag set */
1563 if (print_flag
== 1) {
1564 if (link
.link_status
)
1565 printf("Port %d Link Up - speed %u "
1566 "Mbps - %s\n", (uint8_t)portid
,
1567 (unsigned)link
.link_speed
,
1568 (link
.link_duplex
== ETH_LINK_FULL_DUPLEX
) ?
1569 ("full-duplex") : ("half-duplex\n"));
1571 printf("Port %d Link Down\n",
1575 /* clear all_ports_up flag if any link down */
1576 if (link
.link_status
== ETH_LINK_DOWN
) {
1581 /* after finally printing all link status, get out */
1582 if (print_flag
== 1)
1585 if (all_ports_up
== 0) {
1588 rte_delay_ms(CHECK_INTERVAL
);
1591 /* set the print_flag if all ports up or timeout */
1592 if (all_ports_up
== 1 || count
== (MAX_CHECK_TIME
- 1)) {
1599 static int check_ptype(uint8_t portid
)
1602 int ptype_l3_ipv4
= 0;
1603 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1604 int ptype_l3_ipv6
= 0;
1606 uint32_t ptype_mask
= RTE_PTYPE_L3_MASK
;
1608 ret
= rte_eth_dev_get_supported_ptypes(portid
, ptype_mask
, NULL
, 0);
1612 uint32_t ptypes
[ret
];
1614 ret
= rte_eth_dev_get_supported_ptypes(portid
, ptype_mask
, ptypes
, ret
);
1615 for (i
= 0; i
< ret
; ++i
) {
1616 if (ptypes
[i
] & RTE_PTYPE_L3_IPV4
)
1618 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1619 if (ptypes
[i
] & RTE_PTYPE_L3_IPV6
)
1624 if (ptype_l3_ipv4
== 0)
1625 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid
);
1627 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1628 if (ptype_l3_ipv6
== 0)
1629 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid
);
1632 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1634 #else /* APP_LOOKUP_EXACT_MATCH */
1635 if (ptype_l3_ipv4
&& ptype_l3_ipv6
)
1644 main(int argc
, char **argv
)
1646 struct lcore_conf
*qconf
;
1647 struct rte_eth_dev_info dev_info
;
1648 struct rte_eth_txconf
*txconf
;
1654 uint32_t n_tx_queue
, nb_lcores
;
1655 uint32_t dev_rxq_num
, dev_txq_num
;
1656 uint8_t portid
, nb_rx_queue
, queue
, socketid
;
1657 uint16_t org_rxq_intr
= port_conf
.intr_conf
.rxq
;
1659 /* catch SIGINT and restore cpufreq governor to ondemand */
1660 signal(SIGINT
, signal_exit_now
);
1663 ret
= rte_eal_init(argc
, argv
);
1665 rte_exit(EXIT_FAILURE
, "Invalid EAL parameters\n");
1669 /* init RTE timer library to be used late */
1670 rte_timer_subsystem_init();
1672 /* parse application arguments (after the EAL ones) */
1673 ret
= parse_args(argc
, argv
);
1675 rte_exit(EXIT_FAILURE
, "Invalid L3FWD parameters\n");
1677 if (check_lcore_params() < 0)
1678 rte_exit(EXIT_FAILURE
, "check_lcore_params failed\n");
1680 ret
= init_lcore_rx_queues();
1682 rte_exit(EXIT_FAILURE
, "init_lcore_rx_queues failed\n");
1684 nb_ports
= rte_eth_dev_count();
1686 if (check_port_config(nb_ports
) < 0)
1687 rte_exit(EXIT_FAILURE
, "check_port_config failed\n");
1689 nb_lcores
= rte_lcore_count();
1691 /* initialize all ports */
1692 for (portid
= 0; portid
< nb_ports
; portid
++) {
1693 /* skip ports that are not enabled */
1694 if ((enabled_port_mask
& (1 << portid
)) == 0) {
1695 printf("\nSkipping disabled port %d\n", portid
);
1700 printf("Initializing port %d ... ", portid
);
1703 rte_eth_dev_info_get(portid
, &dev_info
);
1704 dev_rxq_num
= dev_info
.max_rx_queues
;
1705 dev_txq_num
= dev_info
.max_tx_queues
;
1707 nb_rx_queue
= get_port_n_rx_queues(portid
);
1708 if (nb_rx_queue
> dev_rxq_num
)
1709 rte_exit(EXIT_FAILURE
,
1710 "Cannot configure not existed rxq: "
1711 "port=%d\n", portid
);
1713 n_tx_queue
= nb_lcores
;
1714 if (n_tx_queue
> dev_txq_num
)
1715 n_tx_queue
= dev_txq_num
;
1716 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1717 nb_rx_queue
, (unsigned)n_tx_queue
);
1718 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1719 if (nb_rx_queue
== 0)
1720 port_conf
.intr_conf
.rxq
= 0;
1721 ret
= rte_eth_dev_configure(portid
, nb_rx_queue
,
1722 (uint16_t)n_tx_queue
, &port_conf
);
1723 /* Revert to original value */
1724 port_conf
.intr_conf
.rxq
= org_rxq_intr
;
1726 rte_exit(EXIT_FAILURE
, "Cannot configure device: "
1727 "err=%d, port=%d\n", ret
, portid
);
1729 rte_eth_macaddr_get(portid
, &ports_eth_addr
[portid
]);
1730 print_ethaddr(" Address:", &ports_eth_addr
[portid
]);
1734 ret
= init_mem(NB_MBUF
);
1736 rte_exit(EXIT_FAILURE
, "init_mem failed\n");
1738 for (lcore_id
= 0; lcore_id
< RTE_MAX_LCORE
; lcore_id
++) {
1739 if (rte_lcore_is_enabled(lcore_id
) == 0)
1742 /* Initialize TX buffers */
1743 qconf
= &lcore_conf
[lcore_id
];
1744 qconf
->tx_buffer
[portid
] = rte_zmalloc_socket("tx_buffer",
1745 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST
), 0,
1746 rte_eth_dev_socket_id(portid
));
1747 if (qconf
->tx_buffer
[portid
] == NULL
)
1748 rte_exit(EXIT_FAILURE
, "Can't allocate tx buffer for port %u\n",
1751 rte_eth_tx_buffer_init(qconf
->tx_buffer
[portid
], MAX_PKT_BURST
);
1754 /* init one TX queue per couple (lcore,port) */
1756 for (lcore_id
= 0; lcore_id
< RTE_MAX_LCORE
; lcore_id
++) {
1757 if (rte_lcore_is_enabled(lcore_id
) == 0)
1760 if (queueid
>= dev_txq_num
)
1765 (uint8_t)rte_lcore_to_socket_id(lcore_id
);
1769 printf("txq=%u,%d,%d ", lcore_id
, queueid
, socketid
);
1772 rte_eth_dev_info_get(portid
, &dev_info
);
1773 txconf
= &dev_info
.default_txconf
;
1774 if (port_conf
.rxmode
.jumbo_frame
)
1775 txconf
->txq_flags
= 0;
1776 ret
= rte_eth_tx_queue_setup(portid
, queueid
, nb_txd
,
1779 rte_exit(EXIT_FAILURE
,
1780 "rte_eth_tx_queue_setup: err=%d, "
1781 "port=%d\n", ret
, portid
);
1783 qconf
= &lcore_conf
[lcore_id
];
1784 qconf
->tx_queue_id
[portid
] = queueid
;
1787 qconf
->tx_port_id
[qconf
->n_tx_port
] = portid
;
1793 for (lcore_id
= 0; lcore_id
< RTE_MAX_LCORE
; lcore_id
++) {
1794 if (rte_lcore_is_enabled(lcore_id
) == 0)
1797 /* init power management library */
1798 ret
= rte_power_init(lcore_id
);
1801 "Library initialization failed on core %u\n", lcore_id
);
1803 /* init timer structures for each enabled lcore */
1804 rte_timer_init(&power_timers
[lcore_id
]);
1805 hz
= rte_get_timer_hz();
1806 rte_timer_reset(&power_timers
[lcore_id
],
1807 hz
/TIMER_NUMBER_PER_SECOND
, SINGLE
, lcore_id
,
1808 power_timer_cb
, NULL
);
1810 qconf
= &lcore_conf
[lcore_id
];
1811 printf("\nInitializing rx queues on lcore %u ... ", lcore_id
);
1813 /* init RX queues */
1814 for(queue
= 0; queue
< qconf
->n_rx_queue
; ++queue
) {
1815 portid
= qconf
->rx_queue_list
[queue
].port_id
;
1816 queueid
= qconf
->rx_queue_list
[queue
].queue_id
;
1820 (uint8_t)rte_lcore_to_socket_id(lcore_id
);
1824 printf("rxq=%d,%d,%d ", portid
, queueid
, socketid
);
1827 ret
= rte_eth_rx_queue_setup(portid
, queueid
, nb_rxd
,
1829 pktmbuf_pool
[socketid
]);
1831 rte_exit(EXIT_FAILURE
,
1832 "rte_eth_rx_queue_setup: err=%d, "
1833 "port=%d\n", ret
, portid
);
1836 if (add_cb_parse_ptype(portid
, queueid
) < 0)
1837 rte_exit(EXIT_FAILURE
,
1838 "Fail to add ptype cb\n");
1839 } else if (!check_ptype(portid
))
1840 rte_exit(EXIT_FAILURE
,
1841 "PMD can not provide needed ptypes\n");
1848 for (portid
= 0; portid
< nb_ports
; portid
++) {
1849 if ((enabled_port_mask
& (1 << portid
)) == 0) {
1853 ret
= rte_eth_dev_start(portid
);
1855 rte_exit(EXIT_FAILURE
, "rte_eth_dev_start: err=%d, "
1856 "port=%d\n", ret
, portid
);
1858 * If enabled, put device in promiscuous mode.
1859 * This allows IO forwarding mode to forward packets
1860 * to itself through 2 cross-connected ports of the
1864 rte_eth_promiscuous_enable(portid
);
1865 /* initialize spinlock for each port */
1866 rte_spinlock_init(&(locks
[portid
]));
1869 check_all_ports_link_status((uint8_t)nb_ports
, enabled_port_mask
);
1871 /* launch per-lcore init on every lcore */
1872 rte_eal_mp_remote_launch(main_loop
, NULL
, CALL_MASTER
);
1873 RTE_LCORE_FOREACH_SLAVE(lcore_id
) {
1874 if (rte_eal_wait_lcore(lcore_id
) < 0)