xref: /dpdk/examples/ipv4_multicast/main.c (revision 3998e2a07220844d3f3c17f76a781ced3efe0de0)
1*3998e2a0SBruce Richardson /* SPDX-License-Identifier: BSD-3-Clause
2*3998e2a0SBruce Richardson  * Copyright(c) 2010-2014 Intel Corporation
3af75078fSIntel  */
4af75078fSIntel 
5af75078fSIntel #include <stdio.h>
6af75078fSIntel #include <stdlib.h>
7af75078fSIntel #include <stdint.h>
8af75078fSIntel #include <inttypes.h>
9af75078fSIntel #include <sys/types.h>
10af75078fSIntel #include <string.h>
11af75078fSIntel #include <sys/queue.h>
12af75078fSIntel #include <stdarg.h>
13af75078fSIntel #include <errno.h>
14af75078fSIntel #include <getopt.h>
15af75078fSIntel 
16af75078fSIntel #include <rte_common.h>
17af75078fSIntel #include <rte_byteorder.h>
18af75078fSIntel #include <rte_log.h>
19af75078fSIntel #include <rte_memory.h>
20af75078fSIntel #include <rte_memcpy.h>
21af75078fSIntel #include <rte_eal.h>
22af75078fSIntel #include <rte_launch.h>
23af75078fSIntel #include <rte_atomic.h>
24af75078fSIntel #include <rte_cycles.h>
25af75078fSIntel #include <rte_prefetch.h>
26af75078fSIntel #include <rte_lcore.h>
27af75078fSIntel #include <rte_per_lcore.h>
28af75078fSIntel #include <rte_branch_prediction.h>
29af75078fSIntel #include <rte_interrupts.h>
30af75078fSIntel #include <rte_random.h>
31af75078fSIntel #include <rte_debug.h>
32af75078fSIntel #include <rte_ether.h>
33af75078fSIntel #include <rte_ethdev.h>
34af75078fSIntel #include <rte_mempool.h>
35af75078fSIntel #include <rte_mbuf.h>
36af75078fSIntel #include <rte_malloc.h>
37af75078fSIntel #include <rte_fbk_hash.h>
38af75078fSIntel #include <rte_ip.h>
39af75078fSIntel 
40af75078fSIntel #define RTE_LOGTYPE_IPv4_MULTICAST RTE_LOGTYPE_USER1
41af75078fSIntel 
42af75078fSIntel #define MAX_PORTS 16
43af75078fSIntel 
44af75078fSIntel #define	MCAST_CLONE_PORTS	2
45af75078fSIntel #define	MCAST_CLONE_SEGS	2
46af75078fSIntel 
47824cb29cSKonstantin Ananyev #define	PKT_MBUF_DATA_SIZE	RTE_MBUF_DEFAULT_BUF_SIZE
48af75078fSIntel #define	NB_PKT_MBUF	8192
49af75078fSIntel 
50ea0c20eaSOlivier Matz #define	HDR_MBUF_DATA_SIZE	(2 * RTE_PKTMBUF_HEADROOM)
51af75078fSIntel #define	NB_HDR_MBUF	(NB_PKT_MBUF * MAX_PORTS)
52af75078fSIntel 
53af75078fSIntel #define	NB_CLONE_MBUF	(NB_PKT_MBUF * MCAST_CLONE_PORTS * MCAST_CLONE_SEGS * 2)
54af75078fSIntel 
55af75078fSIntel /* allow max jumbo frame 9.5 KB */
56af75078fSIntel #define	JUMBO_FRAME_MAX_SIZE	0x2600
57af75078fSIntel 
58af75078fSIntel #define MAX_PKT_BURST 32
595c95261dSIntel #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
60af75078fSIntel 
61af75078fSIntel /* Configure how many packets ahead to prefetch, when reading packets */
62af75078fSIntel #define PREFETCH_OFFSET	3
63af75078fSIntel 
64af75078fSIntel /*
65af75078fSIntel  * Construct Ethernet multicast address from IPv4 multicast address.
66af75078fSIntel  * Citing RFC 1112, section 6.4:
67af75078fSIntel  * "An IP host group address is mapped to an Ethernet multicast address
68af75078fSIntel  * by placing the low-order 23-bits of the IP address into the low-order
69af75078fSIntel  * 23 bits of the Ethernet multicast address 01-00-5E-00-00-00 (hex)."
70af75078fSIntel  */
71af75078fSIntel #define	ETHER_ADDR_FOR_IPV4_MCAST(x)	\
72af75078fSIntel 	(rte_cpu_to_be_64(0x01005e000000ULL | ((x) & 0x7fffff)) >> 16)
73af75078fSIntel 
74af75078fSIntel /*
75af75078fSIntel  * Configurable number of RX/TX ring descriptors
76af75078fSIntel  */
77af75078fSIntel #define RTE_TEST_RX_DESC_DEFAULT 128
78af75078fSIntel #define RTE_TEST_TX_DESC_DEFAULT 512
79af75078fSIntel static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
80af75078fSIntel static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
81af75078fSIntel 
82af75078fSIntel /* ethernet addresses of ports */
83af75078fSIntel static struct ether_addr ports_eth_addr[MAX_PORTS];
84af75078fSIntel 
85af75078fSIntel /* mask of enabled ports */
86af75078fSIntel static uint32_t enabled_port_mask = 0;
87af75078fSIntel 
88f8244c63SZhiyong Yang static uint16_t nb_ports;
89af75078fSIntel 
90af75078fSIntel static int rx_queue_per_lcore = 1;
91af75078fSIntel 
92af75078fSIntel struct mbuf_table {
93af75078fSIntel 	uint16_t len;
94af75078fSIntel 	struct rte_mbuf *m_table[MAX_PKT_BURST];
95af75078fSIntel };
96af75078fSIntel 
97af75078fSIntel #define MAX_RX_QUEUE_PER_LCORE 16
98af75078fSIntel #define MAX_TX_QUEUE_PER_PORT 16
99af75078fSIntel struct lcore_queue_conf {
100af75078fSIntel 	uint64_t tx_tsc;
101af75078fSIntel 	uint16_t n_rx_queue;
102af75078fSIntel 	uint8_t rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
103af75078fSIntel 	uint16_t tx_queue_id[MAX_PORTS];
104af75078fSIntel 	struct mbuf_table tx_mbufs[MAX_PORTS];
105af75078fSIntel } __rte_cache_aligned;
106af75078fSIntel static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
107af75078fSIntel 
1085e470a66SAndriy Berestovskyy static struct rte_eth_conf port_conf = {
109af75078fSIntel 	.rxmode = {
110af75078fSIntel 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
111af75078fSIntel 		.split_hdr_size = 0,
112af75078fSIntel 		.header_split   = 0, /**< Header Split disabled */
113af75078fSIntel 		.hw_ip_checksum = 0, /**< IP checksum offload disabled */
114af75078fSIntel 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
115af75078fSIntel 		.jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
11660da774eSJeff Guo 		.hw_strip_crc   = 1, /**< CRC stripped by hardware */
117af75078fSIntel 	},
118af75078fSIntel 	.txmode = {
11932e7aa0bSIntel 		.mq_mode = ETH_MQ_TX_NONE,
120af75078fSIntel 	},
121af75078fSIntel };
122af75078fSIntel 
123af75078fSIntel static struct rte_mempool *packet_pool, *header_pool, *clone_pool;
124af75078fSIntel 
125af75078fSIntel 
126af75078fSIntel /* Multicast */
127af75078fSIntel static struct rte_fbk_hash_params mcast_hash_params = {
128af75078fSIntel 	.name = "MCAST_HASH",
129af75078fSIntel 	.entries = 1024,
130af75078fSIntel 	.entries_per_bucket = 4,
131e60f71ebSIntel 	.socket_id = 0,
132af75078fSIntel 	.hash_func = NULL,
133af75078fSIntel 	.init_val = 0,
134af75078fSIntel };
135af75078fSIntel 
136af75078fSIntel struct rte_fbk_hash_table *mcast_hash = NULL;
137af75078fSIntel 
138af75078fSIntel struct mcast_group_params {
139af75078fSIntel 	uint32_t ip;
140af75078fSIntel 	uint16_t port_mask;
141af75078fSIntel };
142af75078fSIntel 
143af75078fSIntel static struct mcast_group_params mcast_group_table[] = {
144af75078fSIntel 		{IPv4(224,0,0,101), 0x1},
145af75078fSIntel 		{IPv4(224,0,0,102), 0x2},
146af75078fSIntel 		{IPv4(224,0,0,103), 0x3},
147af75078fSIntel 		{IPv4(224,0,0,104), 0x4},
148af75078fSIntel 		{IPv4(224,0,0,105), 0x5},
149af75078fSIntel 		{IPv4(224,0,0,106), 0x6},
150af75078fSIntel 		{IPv4(224,0,0,107), 0x7},
151af75078fSIntel 		{IPv4(224,0,0,108), 0x8},
152af75078fSIntel 		{IPv4(224,0,0,109), 0x9},
153af75078fSIntel 		{IPv4(224,0,0,110), 0xA},
154af75078fSIntel 		{IPv4(224,0,0,111), 0xB},
155af75078fSIntel 		{IPv4(224,0,0,112), 0xC},
156af75078fSIntel 		{IPv4(224,0,0,113), 0xD},
157af75078fSIntel 		{IPv4(224,0,0,114), 0xE},
158af75078fSIntel 		{IPv4(224,0,0,115), 0xF},
159af75078fSIntel };
160af75078fSIntel 
161af75078fSIntel #define N_MCAST_GROUPS \
162af75078fSIntel 	(sizeof (mcast_group_table) / sizeof (mcast_group_table[0]))
163af75078fSIntel 
164af75078fSIntel 
165af75078fSIntel /* Send burst of packets on an output interface */
166af75078fSIntel static void
167f8244c63SZhiyong Yang send_burst(struct lcore_queue_conf *qconf, uint16_t port)
168af75078fSIntel {
169af75078fSIntel 	struct rte_mbuf **m_table;
170af75078fSIntel 	uint16_t n, queueid;
171af75078fSIntel 	int ret;
172af75078fSIntel 
173af75078fSIntel 	queueid = qconf->tx_queue_id[port];
174af75078fSIntel 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
175af75078fSIntel 	n = qconf->tx_mbufs[port].len;
176af75078fSIntel 
177af75078fSIntel 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
178af75078fSIntel 	while (unlikely (ret < n)) {
179af75078fSIntel 		rte_pktmbuf_free(m_table[ret]);
180af75078fSIntel 		ret++;
181af75078fSIntel 	}
182af75078fSIntel 
183af75078fSIntel 	qconf->tx_mbufs[port].len = 0;
184af75078fSIntel }
185af75078fSIntel 
186af75078fSIntel /* Get number of bits set. */
187af75078fSIntel static inline uint32_t
188af75078fSIntel bitcnt(uint32_t v)
189af75078fSIntel {
190af75078fSIntel 	uint32_t n;
191af75078fSIntel 
192af75078fSIntel 	for (n = 0; v != 0; v &= v - 1, n++)
193af75078fSIntel 		;
194af75078fSIntel 
195693f715dSHuawei Xie 	return n;
196af75078fSIntel }
197af75078fSIntel 
198af75078fSIntel /**
199af75078fSIntel  * Create the output multicast packet based on the given input packet.
200af75078fSIntel  * There are two approaches for creating outgoing packet, though both
201af75078fSIntel  * are based on data zero-copy idea, they differ in few details:
202af75078fSIntel  * First one creates a clone of the input packet, e.g - walk though all
203af75078fSIntel  * segments of the input packet, and for each of them create a new packet
204af75078fSIntel  * mbuf and attach that new mbuf to the segment (refer to rte_pktmbuf_clone()
205af75078fSIntel  * for more details). Then new mbuf is allocated for the packet header
206af75078fSIntel  * and is prepended to the 'clone' mbuf.
207af75078fSIntel  * Second approach doesn't make a clone, it just increment refcnt for all
208af75078fSIntel  * input packet segments. Then it allocates new mbuf for the packet header
209af75078fSIntel  * and prepends it to the input packet.
210af75078fSIntel  * Basically first approach reuses only input packet's data, but creates
211af75078fSIntel  * it's own copy of packet's metadata. Second approach reuses both input's
212af75078fSIntel  * packet data and metadata.
213af75078fSIntel  * The advantage of first approach - is that each outgoing packet has it's
214af75078fSIntel  * own copy of metadata, so we can safely modify data pointer of the
215af75078fSIntel  * input packet. That allows us to skip creation if the output packet for
216af75078fSIntel  * the last destination port, but instead modify input packet's header inplace,
217af75078fSIntel  * e.g: for N destination ports we need to invoke mcast_out_pkt (N-1) times.
218af75078fSIntel  * The advantage of second approach - less work for each outgoing packet,
219af75078fSIntel  * e.g: we skip "clone" operation completely. Though it comes with a price -
220af75078fSIntel  * input packet's metadata has to be intact. So for N destination ports we
221af75078fSIntel  * need to invoke mcast_out_pkt N times.
222af75078fSIntel  * So for small number of outgoing ports (and segments in the input packet)
223af75078fSIntel  * first approach will be faster.
224af75078fSIntel  * As number of outgoing ports (and/or input segments) will grow,
225af75078fSIntel  * second way will become more preferable.
226af75078fSIntel  *
227af75078fSIntel  *  @param pkt
228af75078fSIntel  *  Input packet mbuf.
229af75078fSIntel  *  @param use_clone
230af75078fSIntel  *  Control which of the two approaches described above should be used:
231af75078fSIntel  *  - 0 - use second approach:
232af75078fSIntel  *    Don't "clone" input packet.
233af75078fSIntel  *    Prepend new header directly to the input packet
234af75078fSIntel  *  - 1 - use first approach:
235af75078fSIntel  *    Make a "clone" of input packet first.
236af75078fSIntel  *    Prepend new header to the clone of the input packet
237af75078fSIntel  *  @return
238af75078fSIntel  *  - The pointer to the new outgoing packet.
239af75078fSIntel  *  - NULL if operation failed.
240af75078fSIntel  */
241af75078fSIntel static inline struct rte_mbuf *
242af75078fSIntel mcast_out_pkt(struct rte_mbuf *pkt, int use_clone)
243af75078fSIntel {
244af75078fSIntel 	struct rte_mbuf *hdr;
245af75078fSIntel 
246af75078fSIntel 	/* Create new mbuf for the header. */
247af75078fSIntel 	if (unlikely ((hdr = rte_pktmbuf_alloc(header_pool)) == NULL))
248693f715dSHuawei Xie 		return NULL;
249af75078fSIntel 
250af75078fSIntel 	/* If requested, then make a new clone packet. */
251af75078fSIntel 	if (use_clone != 0 &&
252af75078fSIntel 	    unlikely ((pkt = rte_pktmbuf_clone(pkt, clone_pool)) == NULL)) {
253af75078fSIntel 		rte_pktmbuf_free(hdr);
254693f715dSHuawei Xie 		return NULL;
255af75078fSIntel 	}
256af75078fSIntel 
257af75078fSIntel 	/* prepend new header */
258ea672a8bSOlivier Matz 	hdr->next = pkt;
259af75078fSIntel 
260af75078fSIntel 
261af75078fSIntel 	/* update header's fields */
262ea672a8bSOlivier Matz 	hdr->pkt_len = (uint16_t)(hdr->data_len + pkt->pkt_len);
2634c20622aSIlya V. Matveychikov 	hdr->nb_segs = pkt->nb_segs + 1;
264af75078fSIntel 
265af75078fSIntel 	/* copy metadata from source packet*/
266ca04aaeaSBruce Richardson 	hdr->port = pkt->port;
2677869536fSBruce Richardson 	hdr->vlan_tci = pkt->vlan_tci;
2681e685446SHelin Zhang 	hdr->vlan_tci_outer = pkt->vlan_tci_outer;
2694199fdeaSOlivier Matz 	hdr->tx_offload = pkt->tx_offload;
270ea672a8bSOlivier Matz 	hdr->hash = pkt->hash;
271af75078fSIntel 
272af75078fSIntel 	hdr->ol_flags = pkt->ol_flags;
273af75078fSIntel 
2749aaccf1aSOlivier Matz 	__rte_mbuf_sanity_check(hdr, 1);
275693f715dSHuawei Xie 	return hdr;
276af75078fSIntel }
277af75078fSIntel 
278af75078fSIntel /*
279af75078fSIntel  * Write new Ethernet header to the outgoing packet,
280af75078fSIntel  * and put it into the outgoing queue for the given port.
281af75078fSIntel  */
282af75078fSIntel static inline void
283af75078fSIntel mcast_send_pkt(struct rte_mbuf *pkt, struct ether_addr *dest_addr,
284f8244c63SZhiyong Yang 		struct lcore_queue_conf *qconf, uint16_t port)
285af75078fSIntel {
286af75078fSIntel 	struct ether_hdr *ethdr;
287af75078fSIntel 	uint16_t len;
288af75078fSIntel 
289af75078fSIntel 	/* Construct Ethernet header. */
290af75078fSIntel 	ethdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, (uint16_t)sizeof(*ethdr));
29150705e8eSThomas Monjalon 	RTE_ASSERT(ethdr != NULL);
292af75078fSIntel 
293af75078fSIntel 	ether_addr_copy(dest_addr, &ethdr->d_addr);
294af75078fSIntel 	ether_addr_copy(&ports_eth_addr[port], &ethdr->s_addr);
295af75078fSIntel 	ethdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
296af75078fSIntel 
297af75078fSIntel 	/* Put new packet into the output queue */
298af75078fSIntel 	len = qconf->tx_mbufs[port].len;
299af75078fSIntel 	qconf->tx_mbufs[port].m_table[len] = pkt;
300af75078fSIntel 	qconf->tx_mbufs[port].len = ++len;
301af75078fSIntel 
302af75078fSIntel 	/* Transmit packets */
303af75078fSIntel 	if (unlikely(MAX_PKT_BURST == len))
304af75078fSIntel 		send_burst(qconf, port);
305af75078fSIntel }
306af75078fSIntel 
307af75078fSIntel /* Multicast forward of the input packet */
308af75078fSIntel static inline void
309af75078fSIntel mcast_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf)
310af75078fSIntel {
311af75078fSIntel 	struct rte_mbuf *mc;
312af75078fSIntel 	struct ipv4_hdr *iphdr;
313af75078fSIntel 	uint32_t dest_addr, port_mask, port_num, use_clone;
314af75078fSIntel 	int32_t hash;
315f8244c63SZhiyong Yang 	uint16_t port;
316af75078fSIntel 	union {
317af75078fSIntel 		uint64_t as_int;
318af75078fSIntel 		struct ether_addr as_addr;
319af75078fSIntel 	} dst_eth_addr;
320af75078fSIntel 
321af75078fSIntel 	/* Remove the Ethernet header from the input packet */
322af75078fSIntel 	iphdr = (struct ipv4_hdr *)rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
32350705e8eSThomas Monjalon 	RTE_ASSERT(iphdr != NULL);
324af75078fSIntel 
325af75078fSIntel 	dest_addr = rte_be_to_cpu_32(iphdr->dst_addr);
326af75078fSIntel 
327af75078fSIntel 	/*
328af75078fSIntel 	 * Check that it is a valid multicast address and
329af75078fSIntel 	 * we have some active ports assigned to it.
330af75078fSIntel 	 */
331af75078fSIntel 	if(!IS_IPV4_MCAST(dest_addr) ||
332af75078fSIntel 	    (hash = rte_fbk_hash_lookup(mcast_hash, dest_addr)) <= 0 ||
333af75078fSIntel 	    (port_mask = hash & enabled_port_mask) == 0) {
334af75078fSIntel 		rte_pktmbuf_free(m);
335af75078fSIntel 		return;
336af75078fSIntel 	}
337af75078fSIntel 
338af75078fSIntel 	/* Calculate number of destination ports. */
339af75078fSIntel 	port_num = bitcnt(port_mask);
340af75078fSIntel 
341af75078fSIntel 	/* Should we use rte_pktmbuf_clone() or not. */
342af75078fSIntel 	use_clone = (port_num <= MCAST_CLONE_PORTS &&
343ea672a8bSOlivier Matz 	    m->nb_segs <= MCAST_CLONE_SEGS);
344af75078fSIntel 
345af75078fSIntel 	/* Mark all packet's segments as referenced port_num times */
346af75078fSIntel 	if (use_clone == 0)
347af75078fSIntel 		rte_pktmbuf_refcnt_update(m, (uint16_t)port_num);
348af75078fSIntel 
349af75078fSIntel 	/* construct destination ethernet address */
350af75078fSIntel 	dst_eth_addr.as_int = ETHER_ADDR_FOR_IPV4_MCAST(dest_addr);
351af75078fSIntel 
352af75078fSIntel 	for (port = 0; use_clone != port_mask; port_mask >>= 1, port++) {
353af75078fSIntel 
354af75078fSIntel 		/* Prepare output packet and send it out. */
355af75078fSIntel 		if ((port_mask & 1) != 0) {
356af75078fSIntel 			if (likely ((mc = mcast_out_pkt(m, use_clone)) != NULL))
357af75078fSIntel 				mcast_send_pkt(mc, &dst_eth_addr.as_addr,
358af75078fSIntel 						qconf, port);
359af75078fSIntel 			else if (use_clone == 0)
360af75078fSIntel 				rte_pktmbuf_free(m);
361af75078fSIntel 		}
362af75078fSIntel 	}
363af75078fSIntel 
364af75078fSIntel 	/*
365af75078fSIntel 	 * If we making clone packets, then, for the last destination port,
366af75078fSIntel 	 * we can overwrite input packet's metadata.
367af75078fSIntel 	 */
368af75078fSIntel 	if (use_clone != 0)
369af75078fSIntel 		mcast_send_pkt(m, &dst_eth_addr.as_addr, qconf, port);
370af75078fSIntel 	else
371af75078fSIntel 		rte_pktmbuf_free(m);
372af75078fSIntel }
373af75078fSIntel 
374af75078fSIntel /* Send burst of outgoing packet, if timeout expires. */
375af75078fSIntel static inline void
376af75078fSIntel send_timeout_burst(struct lcore_queue_conf *qconf)
377af75078fSIntel {
378af75078fSIntel 	uint64_t cur_tsc;
379f8244c63SZhiyong Yang 	uint16_t portid;
3805c95261dSIntel 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
381af75078fSIntel 
382af75078fSIntel 	cur_tsc = rte_rdtsc();
3835c95261dSIntel 	if (likely (cur_tsc < qconf->tx_tsc + drain_tsc))
384af75078fSIntel 		return;
385af75078fSIntel 
386af75078fSIntel 	for (portid = 0; portid < MAX_PORTS; portid++) {
387af75078fSIntel 		if (qconf->tx_mbufs[portid].len != 0)
388af75078fSIntel 			send_burst(qconf, portid);
389af75078fSIntel 	}
390af75078fSIntel 	qconf->tx_tsc = cur_tsc;
391af75078fSIntel }
392af75078fSIntel 
393af75078fSIntel /* main processing loop */
394cdfd5dbbSIntel static int
395af75078fSIntel main_loop(__rte_unused void *dummy)
396af75078fSIntel {
397af75078fSIntel 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
3986441b9f6SIntel 	unsigned lcore_id;
399af75078fSIntel 	int i, j, nb_rx;
400f8244c63SZhiyong Yang 	uint16_t portid;
401af75078fSIntel 	struct lcore_queue_conf *qconf;
402af75078fSIntel 
403af75078fSIntel 	lcore_id = rte_lcore_id();
404af75078fSIntel 	qconf = &lcore_queue_conf[lcore_id];
405af75078fSIntel 
406af75078fSIntel 
407af75078fSIntel 	if (qconf->n_rx_queue == 0) {
408af75078fSIntel 		RTE_LOG(INFO, IPv4_MULTICAST, "lcore %u has nothing to do\n",
409af75078fSIntel 		    lcore_id);
410cdfd5dbbSIntel 		return 0;
411af75078fSIntel 	}
412af75078fSIntel 
413af75078fSIntel 	RTE_LOG(INFO, IPv4_MULTICAST, "entering main loop on lcore %u\n",
414af75078fSIntel 	    lcore_id);
415af75078fSIntel 
416af75078fSIntel 	for (i = 0; i < qconf->n_rx_queue; i++) {
417af75078fSIntel 
418af75078fSIntel 		portid = qconf->rx_queue_list[i];
419af75078fSIntel 		RTE_LOG(INFO, IPv4_MULTICAST, " -- lcoreid=%u portid=%d\n",
420f8244c63SZhiyong Yang 		    lcore_id, portid);
421af75078fSIntel 	}
422af75078fSIntel 
423af75078fSIntel 	while (1) {
424af75078fSIntel 
425af75078fSIntel 		/*
426af75078fSIntel 		 * Read packet from RX queues
427af75078fSIntel 		 */
428af75078fSIntel 		for (i = 0; i < qconf->n_rx_queue; i++) {
429af75078fSIntel 
430af75078fSIntel 			portid = qconf->rx_queue_list[i];
431af75078fSIntel 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
432af75078fSIntel 						 MAX_PKT_BURST);
433af75078fSIntel 
434af75078fSIntel 			/* Prefetch first packets */
435af75078fSIntel 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
436af75078fSIntel 				rte_prefetch0(rte_pktmbuf_mtod(
437af75078fSIntel 						pkts_burst[j], void *));
438af75078fSIntel 			}
439af75078fSIntel 
440af75078fSIntel 			/* Prefetch and forward already prefetched packets */
441af75078fSIntel 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
442af75078fSIntel 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
443af75078fSIntel 						j + PREFETCH_OFFSET], void *));
444af75078fSIntel 				mcast_forward(pkts_burst[j], qconf);
445af75078fSIntel 			}
446af75078fSIntel 
447af75078fSIntel 			/* Forward remaining prefetched packets */
448af75078fSIntel 			for (; j < nb_rx; j++) {
449af75078fSIntel 				mcast_forward(pkts_burst[j], qconf);
450af75078fSIntel 			}
451af75078fSIntel 		}
452af75078fSIntel 
453af75078fSIntel 		/* Send out packets from TX queues */
454af75078fSIntel 		send_timeout_burst(qconf);
455af75078fSIntel 	}
456af75078fSIntel }
457af75078fSIntel 
458af75078fSIntel /* display usage */
459af75078fSIntel static void
460af75078fSIntel print_usage(const char *prgname)
461af75078fSIntel {
462af75078fSIntel 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
463af75078fSIntel 	    "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
464af75078fSIntel 	    "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
465af75078fSIntel 	    prgname);
466af75078fSIntel }
467af75078fSIntel 
468af75078fSIntel static uint32_t
469af75078fSIntel parse_portmask(const char *portmask)
470af75078fSIntel {
471af75078fSIntel 	char *end = NULL;
472af75078fSIntel 	unsigned long pm;
473af75078fSIntel 
474af75078fSIntel 	/* parse hexadecimal string */
475af75078fSIntel 	pm = strtoul(portmask, &end, 16);
476af75078fSIntel 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
477af75078fSIntel 		return 0;
478af75078fSIntel 
479693f715dSHuawei Xie 	return (uint32_t)pm;
480af75078fSIntel }
481af75078fSIntel 
482af75078fSIntel static int
483af75078fSIntel parse_nqueue(const char *q_arg)
484af75078fSIntel {
485af75078fSIntel 	char *end = NULL;
486af75078fSIntel 	unsigned long n;
487af75078fSIntel 
488af75078fSIntel 	/* parse numerical string */
489af75078fSIntel 	errno = 0;
490af75078fSIntel 	n = strtoul(q_arg, &end, 0);
491af75078fSIntel 	if (errno != 0 || end == NULL || *end != '\0' ||
492af75078fSIntel 			n == 0 || n >= MAX_RX_QUEUE_PER_LCORE)
493693f715dSHuawei Xie 		return -1;
494af75078fSIntel 
495693f715dSHuawei Xie 	return n;
496af75078fSIntel }
497af75078fSIntel 
498af75078fSIntel /* Parse the argument given in the command line of the application */
499af75078fSIntel static int
500af75078fSIntel parse_args(int argc, char **argv)
501af75078fSIntel {
502af75078fSIntel 	int opt, ret;
503af75078fSIntel 	char **argvopt;
504af75078fSIntel 	int option_index;
505af75078fSIntel 	char *prgname = argv[0];
506af75078fSIntel 	static struct option lgopts[] = {
507af75078fSIntel 		{NULL, 0, 0, 0}
508af75078fSIntel 	};
509af75078fSIntel 
510af75078fSIntel 	argvopt = argv;
511af75078fSIntel 
512af75078fSIntel 	while ((opt = getopt_long(argc, argvopt, "p:q:",
513af75078fSIntel 				  lgopts, &option_index)) != EOF) {
514af75078fSIntel 
515af75078fSIntel 		switch (opt) {
516af75078fSIntel 		/* portmask */
517af75078fSIntel 		case 'p':
518af75078fSIntel 			enabled_port_mask = parse_portmask(optarg);
519af75078fSIntel 			if (enabled_port_mask == 0) {
520af75078fSIntel 				printf("invalid portmask\n");
521af75078fSIntel 				print_usage(prgname);
522af75078fSIntel 				return -1;
523af75078fSIntel 			}
524af75078fSIntel 			break;
525af75078fSIntel 
526af75078fSIntel 		/* nqueue */
527af75078fSIntel 		case 'q':
528af75078fSIntel 			rx_queue_per_lcore = parse_nqueue(optarg);
529af75078fSIntel 			if (rx_queue_per_lcore < 0) {
530af75078fSIntel 				printf("invalid queue number\n");
531af75078fSIntel 				print_usage(prgname);
532af75078fSIntel 				return -1;
533af75078fSIntel 			}
534af75078fSIntel 			break;
535af75078fSIntel 
536af75078fSIntel 		default:
537af75078fSIntel 			print_usage(prgname);
538af75078fSIntel 			return -1;
539af75078fSIntel 		}
540af75078fSIntel 	}
541af75078fSIntel 
542af75078fSIntel 	if (optind >= 0)
543af75078fSIntel 		argv[optind-1] = prgname;
544af75078fSIntel 
545af75078fSIntel 	ret = optind-1;
5469d5ca532SKeith Wiles 	optind = 1; /* reset getopt lib */
547af75078fSIntel 	return ret;
548af75078fSIntel }
549af75078fSIntel 
550af75078fSIntel static void
551af75078fSIntel print_ethaddr(const char *name, struct ether_addr *eth_addr)
552af75078fSIntel {
553ec3d82dbSCunming Liang 	char buf[ETHER_ADDR_FMT_SIZE];
554ec3d82dbSCunming Liang 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
555ec3d82dbSCunming Liang 	printf("%s%s", name, buf);
556af75078fSIntel }
557af75078fSIntel 
558af75078fSIntel static int
559af75078fSIntel init_mcast_hash(void)
560af75078fSIntel {
561af75078fSIntel 	uint32_t i;
562af75078fSIntel 
563e60f71ebSIntel 	mcast_hash_params.socket_id = rte_socket_id();
564af75078fSIntel 	mcast_hash = rte_fbk_hash_create(&mcast_hash_params);
565af75078fSIntel 	if (mcast_hash == NULL){
566af75078fSIntel 		return -1;
567af75078fSIntel 	}
568af75078fSIntel 
569af75078fSIntel 	for (i = 0; i < N_MCAST_GROUPS; i ++){
570af75078fSIntel 		if (rte_fbk_hash_add_key(mcast_hash,
571af75078fSIntel 			mcast_group_table[i].ip,
572af75078fSIntel 			mcast_group_table[i].port_mask) < 0) {
573af75078fSIntel 			return -1;
574af75078fSIntel 		}
575af75078fSIntel 	}
576af75078fSIntel 
577af75078fSIntel 	return 0;
578af75078fSIntel }
579af75078fSIntel 
580d3641ae8SIntel /* Check the link status of all ports in up to 9s, and print them finally */
581d3641ae8SIntel static void
582f8244c63SZhiyong Yang check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
583d3641ae8SIntel {
584d3641ae8SIntel #define CHECK_INTERVAL 100 /* 100ms */
585d3641ae8SIntel #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
586f8244c63SZhiyong Yang 	uint16_t portid;
587f8244c63SZhiyong Yang 	uint8_t count, all_ports_up, print_flag = 0;
588d3641ae8SIntel 	struct rte_eth_link link;
589d3641ae8SIntel 
590d3641ae8SIntel 	printf("\nChecking link status");
591d3641ae8SIntel 	fflush(stdout);
592d3641ae8SIntel 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
593d3641ae8SIntel 		all_ports_up = 1;
594d3641ae8SIntel 		for (portid = 0; portid < port_num; portid++) {
595d3641ae8SIntel 			if ((port_mask & (1 << portid)) == 0)
596d3641ae8SIntel 				continue;
597d3641ae8SIntel 			memset(&link, 0, sizeof(link));
598d3641ae8SIntel 			rte_eth_link_get_nowait(portid, &link);
599d3641ae8SIntel 			/* print link status if flag set */
600d3641ae8SIntel 			if (print_flag == 1) {
601d3641ae8SIntel 				if (link.link_status)
602f8244c63SZhiyong Yang 					printf(
603f8244c63SZhiyong Yang 					"Port%d Link Up. Speed %u Mbps - %s\n",
604f8244c63SZhiyong Yang 					portid, link.link_speed,
605d3641ae8SIntel 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
606d3641ae8SIntel 					("full-duplex") : ("half-duplex\n"));
607d3641ae8SIntel 				else
608f8244c63SZhiyong Yang 					printf("Port %d Link Down\n", portid);
609d3641ae8SIntel 				continue;
610d3641ae8SIntel 			}
611d3641ae8SIntel 			/* clear all_ports_up flag if any link down */
61209419f23SThomas Monjalon 			if (link.link_status == ETH_LINK_DOWN) {
613d3641ae8SIntel 				all_ports_up = 0;
614d3641ae8SIntel 				break;
615d3641ae8SIntel 			}
616d3641ae8SIntel 		}
617d3641ae8SIntel 		/* after finally printing all link status, get out */
618d3641ae8SIntel 		if (print_flag == 1)
619d3641ae8SIntel 			break;
620d3641ae8SIntel 
621d3641ae8SIntel 		if (all_ports_up == 0) {
622d3641ae8SIntel 			printf(".");
623d3641ae8SIntel 			fflush(stdout);
624d3641ae8SIntel 			rte_delay_ms(CHECK_INTERVAL);
625d3641ae8SIntel 		}
626d3641ae8SIntel 
627d3641ae8SIntel 		/* set the print_flag if all ports up or timeout */
628d3641ae8SIntel 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
629d3641ae8SIntel 			print_flag = 1;
630d3641ae8SIntel 			printf("done\n");
631d3641ae8SIntel 		}
632d3641ae8SIntel 	}
633d3641ae8SIntel }
634d3641ae8SIntel 
635af75078fSIntel int
63698a16481SDavid Marchand main(int argc, char **argv)
637af75078fSIntel {
638af75078fSIntel 	struct lcore_queue_conf *qconf;
63981f7ecd9SPablo de Lara 	struct rte_eth_dev_info dev_info;
64081f7ecd9SPablo de Lara 	struct rte_eth_txconf *txconf;
641af75078fSIntel 	int ret;
642af75078fSIntel 	uint16_t queueid;
6439787d22fSIntel 	unsigned lcore_id = 0, rx_lcore_id = 0;
644af75078fSIntel 	uint32_t n_tx_queue, nb_lcores;
645f8244c63SZhiyong Yang 	uint16_t portid;
646af75078fSIntel 
647af75078fSIntel 	/* init EAL */
648af75078fSIntel 	ret = rte_eal_init(argc, argv);
649af75078fSIntel 	if (ret < 0)
650af75078fSIntel 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
651af75078fSIntel 	argc -= ret;
652af75078fSIntel 	argv += ret;
653af75078fSIntel 
654af75078fSIntel 	/* parse application arguments (after the EAL ones) */
655af75078fSIntel 	ret = parse_args(argc, argv);
656af75078fSIntel 	if (ret < 0)
657af75078fSIntel 		rte_exit(EXIT_FAILURE, "Invalid IPV4_MULTICAST parameters\n");
658af75078fSIntel 
659af75078fSIntel 	/* create the mbuf pools */
660ea0c20eaSOlivier Matz 	packet_pool = rte_pktmbuf_pool_create("packet_pool", NB_PKT_MBUF, 32,
661ea0c20eaSOlivier Matz 		0, PKT_MBUF_DATA_SIZE, rte_socket_id());
662af75078fSIntel 
663af75078fSIntel 	if (packet_pool == NULL)
664af75078fSIntel 		rte_exit(EXIT_FAILURE, "Cannot init packet mbuf pool\n");
665af75078fSIntel 
666ea0c20eaSOlivier Matz 	header_pool = rte_pktmbuf_pool_create("header_pool", NB_HDR_MBUF, 32,
667ea0c20eaSOlivier Matz 		0, HDR_MBUF_DATA_SIZE, rte_socket_id());
668af75078fSIntel 
669af75078fSIntel 	if (header_pool == NULL)
670af75078fSIntel 		rte_exit(EXIT_FAILURE, "Cannot init header mbuf pool\n");
671af75078fSIntel 
672ea0c20eaSOlivier Matz 	clone_pool = rte_pktmbuf_pool_create("clone_pool", NB_CLONE_MBUF, 32,
673ea0c20eaSOlivier Matz 		0, 0, rte_socket_id());
674af75078fSIntel 
675af75078fSIntel 	if (clone_pool == NULL)
676af75078fSIntel 		rte_exit(EXIT_FAILURE, "Cannot init clone mbuf pool\n");
677af75078fSIntel 
678af75078fSIntel 	nb_ports = rte_eth_dev_count();
679af75078fSIntel 	if (nb_ports == 0)
680af75078fSIntel 		rte_exit(EXIT_FAILURE, "No physical ports!\n");
681af75078fSIntel 	if (nb_ports > MAX_PORTS)
682af75078fSIntel 		nb_ports = MAX_PORTS;
683af75078fSIntel 
684af75078fSIntel 	nb_lcores = rte_lcore_count();
685af75078fSIntel 
686af75078fSIntel 	/* initialize all ports */
687af75078fSIntel 	for (portid = 0; portid < nb_ports; portid++) {
688af75078fSIntel 		/* skip ports that are not enabled */
689af75078fSIntel 		if ((enabled_port_mask & (1 << portid)) == 0) {
690af75078fSIntel 			printf("Skipping disabled port %d\n", portid);
691af75078fSIntel 			continue;
692af75078fSIntel 		}
693af75078fSIntel 
694af75078fSIntel 		qconf = &lcore_queue_conf[rx_lcore_id];
695af75078fSIntel 
6965e470a66SAndriy Berestovskyy 		/* limit the frame size to the maximum supported by NIC */
6975e470a66SAndriy Berestovskyy 		rte_eth_dev_info_get(portid, &dev_info);
6985e470a66SAndriy Berestovskyy 		port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
6995e470a66SAndriy Berestovskyy 		    dev_info.max_rx_pktlen, port_conf.rxmode.max_rx_pkt_len);
7005e470a66SAndriy Berestovskyy 
701af75078fSIntel 		/* get the lcore_id for this port */
702af75078fSIntel 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
703af75078fSIntel 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
704af75078fSIntel 
705af75078fSIntel 			rx_lcore_id ++;
706af75078fSIntel 			qconf = &lcore_queue_conf[rx_lcore_id];
707af75078fSIntel 
708af75078fSIntel 			if (rx_lcore_id >= RTE_MAX_LCORE)
709af75078fSIntel 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
710af75078fSIntel 		}
711af75078fSIntel 		qconf->rx_queue_list[qconf->n_rx_queue] = portid;
712af75078fSIntel 		qconf->n_rx_queue++;
713af75078fSIntel 
714af75078fSIntel 		/* init port */
715af75078fSIntel 		printf("Initializing port %d on lcore %u... ", portid,
716af75078fSIntel 		       rx_lcore_id);
717af75078fSIntel 		fflush(stdout);
718af75078fSIntel 
719af75078fSIntel 		n_tx_queue = nb_lcores;
720af75078fSIntel 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
721af75078fSIntel 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
722af75078fSIntel 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
723af75078fSIntel 					    &port_conf);
724af75078fSIntel 		if (ret < 0)
725af75078fSIntel 			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
726af75078fSIntel 				  ret, portid);
727af75078fSIntel 
72860efb44fSRoman Zhukov 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
72960efb44fSRoman Zhukov 						       &nb_txd);
73060efb44fSRoman Zhukov 		if (ret < 0)
73160efb44fSRoman Zhukov 			rte_exit(EXIT_FAILURE,
73260efb44fSRoman Zhukov 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
73360efb44fSRoman Zhukov 				 ret, portid);
73460efb44fSRoman Zhukov 
735af75078fSIntel 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
736af75078fSIntel 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
737af75078fSIntel 		printf(", ");
738af75078fSIntel 
739af75078fSIntel 		/* init one RX queue */
740af75078fSIntel 		queueid = 0;
741af75078fSIntel 		printf("rxq=%hu ", queueid);
742af75078fSIntel 		fflush(stdout);
743af75078fSIntel 		ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
74481f7ecd9SPablo de Lara 					     rte_eth_dev_socket_id(portid),
74581f7ecd9SPablo de Lara 					     NULL,
746af75078fSIntel 					     packet_pool);
747af75078fSIntel 		if (ret < 0)
748af75078fSIntel 			rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, port=%d\n",
749af75078fSIntel 				  ret, portid);
750af75078fSIntel 
751af75078fSIntel 		/* init one TX queue per couple (lcore,port) */
752af75078fSIntel 		queueid = 0;
753af75078fSIntel 
754af75078fSIntel 		RTE_LCORE_FOREACH(lcore_id) {
755af75078fSIntel 			if (rte_lcore_is_enabled(lcore_id) == 0)
756af75078fSIntel 				continue;
757af75078fSIntel 			printf("txq=%u,%hu ", lcore_id, queueid);
758af75078fSIntel 			fflush(stdout);
75981f7ecd9SPablo de Lara 
76081f7ecd9SPablo de Lara 			txconf = &dev_info.default_txconf;
76181f7ecd9SPablo de Lara 			txconf->txq_flags = 0;
762af75078fSIntel 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
76381f7ecd9SPablo de Lara 						     rte_lcore_to_socket_id(lcore_id), txconf);
764af75078fSIntel 			if (ret < 0)
765af75078fSIntel 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
766af75078fSIntel 					  "port=%d\n", ret, portid);
767af75078fSIntel 
768af75078fSIntel 			qconf = &lcore_queue_conf[lcore_id];
769af75078fSIntel 			qconf->tx_queue_id[portid] = queueid;
770af75078fSIntel 			queueid++;
771af75078fSIntel 		}
772af75078fSIntel 
773af75078fSIntel 		/* Start device */
774af75078fSIntel 		ret = rte_eth_dev_start(portid);
775af75078fSIntel 		if (ret < 0)
776af75078fSIntel 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
777af75078fSIntel 				  ret, portid);
778af75078fSIntel 
779d3641ae8SIntel 		printf("done:\n");
780af75078fSIntel 	}
781af75078fSIntel 
782d3641ae8SIntel 	check_all_ports_link_status(nb_ports, enabled_port_mask);
783af75078fSIntel 
784af75078fSIntel 	/* initialize the multicast hash */
785af75078fSIntel 	int retval = init_mcast_hash();
786af75078fSIntel 	if (retval != 0)
787af75078fSIntel 		rte_exit(EXIT_FAILURE, "Cannot build the multicast hash\n");
788af75078fSIntel 
789af75078fSIntel 	/* launch per-lcore init on every lcore */
790af75078fSIntel 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
791af75078fSIntel 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
792af75078fSIntel 		if (rte_eal_wait_lcore(lcore_id) < 0)
793af75078fSIntel 			return -1;
794af75078fSIntel 	}
795af75078fSIntel 
796af75078fSIntel 	return 0;
797af75078fSIntel }
798