[ceph.git] / ceph / src / seastar / dpdk / examples / multi_process / client_server_mp / mp_server / main.c

/* SPDX-License-Identifier: BSD-3-Clause
 * Copyright(c) 2010-2014 Intel Corporation
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <stdint.h>
#include <stdarg.h>
#include <inttypes.h>
#include <sys/queue.h>
#include <errno.h>
#include <netinet/ip.h>
#include <signal.h>

#include <rte_common.h>
#include <rte_memory.h>
#include <rte_eal.h>
#include <rte_launch.h>
#include <rte_per_lcore.h>
#include <rte_lcore.h>
#include <rte_branch_prediction.h>
#include <rte_atomic.h>
#include <rte_ring.h>
#include <rte_log.h>
#include <rte_debug.h>
#include <rte_mempool.h>
#include <rte_memcpy.h>
#include <rte_mbuf.h>
#include <rte_ether.h>
#include <rte_interrupts.h>
#include <rte_ethdev.h>
#include <rte_byteorder.h>
#include <rte_malloc.h>
#include <rte_string_fns.h>

#include "common.h"
#include "args.h"
#include "init.h"

/*
 * When doing reads from the NIC or the client queues,
 * use this batch size
 */
#define PACKET_READ_SIZE 32

/*
 * Local buffers to put packets in, used to send packets in bursts to the
 * clients
 */
struct client_rx_buf {
	struct rte_mbuf *buffer[PACKET_READ_SIZE];
	uint16_t count;
};

/* One buffer per client rx queue - dynamically allocate array */
static struct client_rx_buf *cl_rx_buf;

static const char *
get_printable_mac_addr(uint16_t port)
{
	static const char err_address[] = "00:00:00:00:00:00";
	static char addresses[RTE_MAX_ETHPORTS][sizeof(err_address)];

	if (unlikely(port >= RTE_MAX_ETHPORTS))
		return err_address;
	if (unlikely(addresses[port][0]=='\0')){
		struct ether_addr mac;
		rte_eth_macaddr_get(port, &mac);
		snprintf(addresses[port], sizeof(addresses[port]),
				"%02x:%02x:%02x:%02x:%02x:%02x\n",
				mac.addr_bytes[0], mac.addr_bytes[1], mac.addr_bytes[2],
				mac.addr_bytes[3], mac.addr_bytes[4], mac.addr_bytes[5]);
	}
	return addresses[port];
}

/*
 * This function displays the recorded statistics for each port
 * and for each client. It uses ANSI terminal codes to clear
 * screen when called. It is called from a single non-master
 * thread in the server process, when the process is run with more
 * than one lcore enabled.
 */
static void
do_stats_display(void)
{
	unsigned i, j;
	const char clr[] = { 27, '[', '2', 'J', '\0' };
	const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
	uint64_t port_tx[RTE_MAX_ETHPORTS], port_tx_drop[RTE_MAX_ETHPORTS];
	uint64_t client_tx[MAX_CLIENTS], client_tx_drop[MAX_CLIENTS];

	/* to get TX stats, we need to do some summing calculations */
	memset(port_tx, 0, sizeof(port_tx));
	memset(port_tx_drop, 0, sizeof(port_tx_drop));
	memset(client_tx, 0, sizeof(client_tx));
	memset(client_tx_drop, 0, sizeof(client_tx_drop));

	for (i = 0; i < num_clients; i++){
		const volatile struct tx_stats *tx = &ports->tx_stats[i];
		for (j = 0; j < ports->num_ports; j++){
			/* assign to local variables here, save re-reading volatile vars */
			const uint64_t tx_val = tx->tx[ports->id[j]];
			const uint64_t drop_val = tx->tx_drop[ports->id[j]];
			port_tx[j] += tx_val;
			port_tx_drop[j] += drop_val;
			client_tx[i] += tx_val;
			client_tx_drop[i] += drop_val;
		}
	}

	/* Clear screen and move to top left */
	printf("%s%s", clr, topLeft);

	printf("PORTS\n");
	printf("-----\n");
	for (i = 0; i < ports->num_ports; i++)
		printf("Port %u: '%s'\t", (unsigned)ports->id[i],
				get_printable_mac_addr(ports->id[i]));
	printf("\n\n");
	for (i = 0; i < ports->num_ports; i++){
		printf("Port %u - rx: %9"PRIu64"\t"
				"tx: %9"PRIu64"\n",
				(unsigned)ports->id[i], ports->rx_stats.rx[i],
				port_tx[i]);
	}

	printf("\nCLIENTS\n");
	printf("-------\n");
	for (i = 0; i < num_clients; i++){
		const unsigned long long rx = clients[i].stats.rx;
		const unsigned long long rx_drop = clients[i].stats.rx_drop;
		printf("Client %2u - rx: %9llu, rx_drop: %9llu\n"
				"            tx: %9"PRIu64", tx_drop: %9"PRIu64"\n",
				i, rx, rx_drop, client_tx[i], client_tx_drop[i]);
	}

	printf("\n");
}

/*
 * The function called from each non-master lcore used by the process.
 * The test_and_set function is used to randomly pick a single lcore on which
 * the code to display the statistics will run. Otherwise, the code just
 * repeatedly sleeps.
 */
static int
sleep_lcore(__attribute__((unused)) void *dummy)
{
	/* Used to pick a display thread - static, so zero-initialised */
	static rte_atomic32_t display_stats;

	/* Only one core should display stats */
	if (rte_atomic32_test_and_set(&display_stats)) {
		const unsigned sleeptime = 1;
		printf("Core %u displaying statistics\n", rte_lcore_id());

		/* Longer initial pause so above printf is seen */
		sleep(sleeptime * 3);

		/* Loop forever: sleep always returns 0 or <= param */
		while (sleep(sleeptime) <= sleeptime)
			do_stats_display();
	}
	return 0;
}

/*
 * Function to set all the client statistic values to zero.
 * Called at program startup.
 */
static void
clear_stats(void)
{
	unsigned i;

	for (i = 0; i < num_clients; i++)
		clients[i].stats.rx = clients[i].stats.rx_drop = 0;
}

/*
 * send a burst of traffic to a client, assuming there are packets
 * available to be sent to this client
 */
static void
flush_rx_queue(uint16_t client)
{
	uint16_t j;
	struct client *cl;

	if (cl_rx_buf[client].count == 0)
		return;

	cl = &clients[client];
	if (rte_ring_enqueue_bulk(cl->rx_q, (void **)cl_rx_buf[client].buffer,
			cl_rx_buf[client].count, NULL) == 0){
		for (j = 0; j < cl_rx_buf[client].count; j++)
			rte_pktmbuf_free(cl_rx_buf[client].buffer[j]);
		cl->stats.rx_drop += cl_rx_buf[client].count;
	}
	else
		cl->stats.rx += cl_rx_buf[client].count;

	cl_rx_buf[client].count = 0;
}

/*
 * marks a packet down to be sent to a particular client process
 */
static inline void
enqueue_rx_packet(uint8_t client, struct rte_mbuf *buf)
{
	cl_rx_buf[client].buffer[cl_rx_buf[client].count++] = buf;
}

/*
 * This function takes a group of packets and routes them
 * individually to the client process. Very simply round-robins the packets
 * without checking any of the packet contents.
 */
static void
process_packets(uint32_t port_num __rte_unused,
		struct rte_mbuf *pkts[], uint16_t rx_count)
{
	uint16_t i;
	uint8_t client = 0;

	for (i = 0; i < rx_count; i++) {
		enqueue_rx_packet(client, pkts[i]);

		if (++client == num_clients)
			client = 0;
	}

	for (i = 0; i < num_clients; i++)
		flush_rx_queue(i);
}

/*
 * Function called by the master lcore of the DPDK process.
 */
static void
do_packet_forwarding(void)
{
	unsigned port_num = 0; /* indexes the port[] array */

	for (;;) {
		struct rte_mbuf *buf[PACKET_READ_SIZE];
		uint16_t rx_count;

		/* read a port */
		rx_count = rte_eth_rx_burst(ports->id[port_num], 0, \
				buf, PACKET_READ_SIZE);
		ports->rx_stats.rx[port_num] += rx_count;

		/* Now process the NIC packets read */
		if (likely(rx_count > 0))
			process_packets(port_num, buf, rx_count);

		/* move to next port */
		if (++port_num == ports->num_ports)
			port_num = 0;
	}
}

static void
signal_handler(int signal)
{
	uint16_t port_id;

	if (signal == SIGINT)
		RTE_ETH_FOREACH_DEV(port_id) {
			rte_eth_dev_stop(port_id);
			rte_eth_dev_close(port_id);
		}
	exit(0);
}

int
main(int argc, char *argv[])
{
	signal(SIGINT, signal_handler);
	/* initialise the system */
	if (init(argc, argv) < 0 )
		return -1;
	RTE_LOG(INFO, APP, "Finished Process Init.\n");

	cl_rx_buf = calloc(num_clients, sizeof(cl_rx_buf[0]));

	/* clear statistics */
	clear_stats();

	/* put all other cores to sleep bar master */
	rte_eal_mp_remote_launch(sleep_lcore, NULL, SKIP_MASTER);

	do_packet_forwarding();
	return 0;
}
Commit	Line	Data
9f95a23c TL	1	/* SPDX-License-Identifier: BSD-3-Clause
9f95a23c TL	2	* Copyright(c) 2010-2014 Intel Corporation
7c673cae FG	3	*/
	4
	5	#include <stdio.h>
	6	#include <stdlib.h>
	7	#include <string.h>
	8	#include <unistd.h>
	9	#include <stdint.h>
	10	#include <stdarg.h>
	11	#include <inttypes.h>
7c673cae FG	12	#include <sys/queue.h>
	13	#include <errno.h>
	14	#include <netinet/ip.h>
9f95a23c	15	#include <signal.h>
7c673cae FG	16
	17	#include <rte_common.h>
	18	#include <rte_memory.h>
7c673cae	19	#include <rte_eal.h>
7c673cae FG	20	#include <rte_launch.h>
	21	#include <rte_per_lcore.h>
	22	#include <rte_lcore.h>
	23	#include <rte_branch_prediction.h>
	24	#include <rte_atomic.h>
	25	#include <rte_ring.h>
	26	#include <rte_log.h>
	27	#include <rte_debug.h>
	28	#include <rte_mempool.h>
	29	#include <rte_memcpy.h>
	30	#include <rte_mbuf.h>
	31	#include <rte_ether.h>
	32	#include <rte_interrupts.h>
7c673cae FG	33	#include <rte_ethdev.h>
	34	#include <rte_byteorder.h>
	35	#include <rte_malloc.h>
	36	#include <rte_string_fns.h>
	37
	38	#include "common.h"
	39	#include "args.h"
	40	#include "init.h"
	41
	42	/*
	43	* When doing reads from the NIC or the client queues,
	44	* use this batch size
	45	*/
	46	#define PACKET_READ_SIZE 32
	47
	48	/*
	49	* Local buffers to put packets in, used to send packets in bursts to the
	50	* clients
	51	*/
	52	struct client_rx_buf {
	53	struct rte_mbuf *buffer[PACKET_READ_SIZE];
	54	uint16_t count;
	55	};
	56
	57	/* One buffer per client rx queue - dynamically allocate array */
	58	static struct client_rx_buf *cl_rx_buf;
	59
	60	static const char *
9f95a23c	61	get_printable_mac_addr(uint16_t port)
7c673cae FG	62	{
	63	static const char err_address[] = "00:00:00:00:00:00";
	64	static char addresses[RTE_MAX_ETHPORTS][sizeof(err_address)];
	65
	66	if (unlikely(port >= RTE_MAX_ETHPORTS))
	67	return err_address;
	68	if (unlikely(addresses[port][0]=='\0')){
	69	struct ether_addr mac;
	70	rte_eth_macaddr_get(port, &mac);
	71	snprintf(addresses[port], sizeof(addresses[port]),
	72	"%02x:%02x:%02x:%02x:%02x:%02x\n",
	73	mac.addr_bytes[0], mac.addr_bytes[1], mac.addr_bytes[2],
	74	mac.addr_bytes[3], mac.addr_bytes[4], mac.addr_bytes[5]);
	75	}
	76	return addresses[port];
	77	}
	78
	79	/*
	80	* This function displays the recorded statistics for each port
	81	* and for each client. It uses ANSI terminal codes to clear
	82	* screen when called. It is called from a single non-master
	83	* thread in the server process, when the process is run with more
	84	* than one lcore enabled.
	85	*/
	86	static void
	87	do_stats_display(void)
	88	{
	89	unsigned i, j;
	90	const char clr[] = { 27, '[', '2', 'J', '\0' };
	91	const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
	92	uint64_t port_tx[RTE_MAX_ETHPORTS], port_tx_drop[RTE_MAX_ETHPORTS];
	93	uint64_t client_tx[MAX_CLIENTS], client_tx_drop[MAX_CLIENTS];
	94
	95	/* to get TX stats, we need to do some summing calculations */
	96	memset(port_tx, 0, sizeof(port_tx));
	97	memset(port_tx_drop, 0, sizeof(port_tx_drop));
	98	memset(client_tx, 0, sizeof(client_tx));
	99	memset(client_tx_drop, 0, sizeof(client_tx_drop));
	100
	101	for (i = 0; i < num_clients; i++){
	102	const volatile struct tx_stats *tx = &ports->tx_stats[i];
	103	for (j = 0; j < ports->num_ports; j++){
	104	/* assign to local variables here, save re-reading volatile vars */
	105	const uint64_t tx_val = tx->tx[ports->id[j]];
	106	const uint64_t drop_val = tx->tx_drop[ports->id[j]];
	107	port_tx[j] += tx_val;
	108	port_tx_drop[j] += drop_val;
	109	client_tx[i] += tx_val;
	110	client_tx_drop[i] += drop_val;
	111	}
	112	}
	113
	114	/* Clear screen and move to top left */
	115	printf("%s%s", clr, topLeft);
	116
	117	printf("PORTS\n");
	118	printf("-----\n");
	119	for (i = 0; i < ports->num_ports; i++)
	120	printf("Port %u: '%s'\t", (unsigned)ports->id[i],
	121	get_printable_mac_addr(ports->id[i]));
	122	printf("\n\n");
	123	for (i = 0; i < ports->num_ports; i++){
	124	printf("Port %u - rx: %9"PRIu64"\t"
	125	"tx: %9"PRIu64"\n",
126	(unsigned)ports->id[i], ports->rx_stats.rx[i],
127	port_tx[i]);
128	}
129
130	printf("\nCLIENTS\n");
131	printf("-------\n");
132	for (i = 0; i < num_clients; i++){
133	const unsigned long long rx = clients[i].stats.rx;
134	const unsigned long long rx_drop = clients[i].stats.rx_drop;
135	printf("Client %2u - rx: %9llu, rx_drop: %9llu\n"
136	" tx: %9"PRIu64", tx_drop: %9"PRIu64"\n",
137	i, rx, rx_drop, client_tx[i], client_tx_drop[i]);
138	}
139
140	printf("\n");
141	}
142
143	/*
144	* The function called from each non-master lcore used by the process.
145	* The test_and_set function is used to randomly pick a single lcore on which
146	* the code to display the statistics will run. Otherwise, the code just
147	* repeatedly sleeps.
148	*/
149	static int
150	sleep_lcore(__attribute__((unused)) void *dummy)
151	{
152	/* Used to pick a display thread - static, so zero-initialised */
153	static rte_atomic32_t display_stats;
154
155	/* Only one core should display stats */
156	if (rte_atomic32_test_and_set(&display_stats)) {
157	const unsigned sleeptime = 1;
158	printf("Core %u displaying statistics\n", rte_lcore_id());
159
160	/* Longer initial pause so above printf is seen */
161	sleep(sleeptime * 3);
162
163	/* Loop forever: sleep always returns 0 or <= param */
164	while (sleep(sleeptime) <= sleeptime)
165	do_stats_display();
166	}
167	return 0;
168	}
169
170	/*
171	* Function to set all the client statistic values to zero.
172	* Called at program startup.
173	*/
174	static void
175	clear_stats(void)
176	{
177	unsigned i;
178
179	for (i = 0; i < num_clients; i++)
180	clients[i].stats.rx = clients[i].stats.rx_drop = 0;
181	}
182
183	/*
184	* send a burst of traffic to a client, assuming there are packets
185	* available to be sent to this client
186	*/
187	static void
188	flush_rx_queue(uint16_t client)
189	{
190	uint16_t j;
191	struct client *cl;
192
193	if (cl_rx_buf[client].count == 0)
194	return;
195
196	cl = &clients[client];
197	if (rte_ring_enqueue_bulk(cl->rx_q, (void **)cl_rx_buf[client].buffer,
11fdf7f2	198	cl_rx_buf[client].count, NULL) == 0){
7c673cae FG	199	for (j = 0; j < cl_rx_buf[client].count; j++)
	200	rte_pktmbuf_free(cl_rx_buf[client].buffer[j]);
	201	cl->stats.rx_drop += cl_rx_buf[client].count;
	202	}
	203	else
	204	cl->stats.rx += cl_rx_buf[client].count;
	205
	206	cl_rx_buf[client].count = 0;
	207	}
	208
	209	/*
	210	* marks a packet down to be sent to a particular client process
	211	*/
	212	static inline void
	213	enqueue_rx_packet(uint8_t client, struct rte_mbuf *buf)
	214	{
	215	cl_rx_buf[client].buffer[cl_rx_buf[client].count++] = buf;
	216	}
	217
	218	/*
	219	* This function takes a group of packets and routes them
	220	* individually to the client process. Very simply round-robins the packets
	221	* without checking any of the packet contents.
	222	*/
	223	static void
	224	process_packets(uint32_t port_num __rte_unused,
	225	struct rte_mbuf *pkts[], uint16_t rx_count)
	226	{
	227	uint16_t i;
	228	uint8_t client = 0;
	229
	230	for (i = 0; i < rx_count; i++) {
	231	enqueue_rx_packet(client, pkts[i]);
	232
	233	if (++client == num_clients)
	234	client = 0;
	235	}
	236
	237	for (i = 0; i < num_clients; i++)
	238	flush_rx_queue(i);
	239	}
	240
	241	/*
	242	* Function called by the master lcore of the DPDK process.
	243	*/
	244	static void
	245	do_packet_forwarding(void)
	246	{
	247	unsigned port_num = 0; /* indexes the port[] array */
	248
	249	for (;;) {
	250	struct rte_mbuf *buf[PACKET_READ_SIZE];
	251	uint16_t rx_count;
	252
	253	/* read a port */
	254	rx_count = rte_eth_rx_burst(ports->id[port_num], 0, \
	255	buf, PACKET_READ_SIZE);
	256	ports->rx_stats.rx[port_num] += rx_count;
	257
	258	/* Now process the NIC packets read */
	259	if (likely(rx_count > 0))
	260	process_packets(port_num, buf, rx_count);
	261
	262	/* move to next port */
263	if (++port_num == ports->num_ports)
264	port_num = 0;
265	}
266	}
267
9f95a23c TL	268	static void
	269	signal_handler(int signal)
	270	{
	271	uint16_t port_id;
	272
	273	if (signal == SIGINT)
	274	RTE_ETH_FOREACH_DEV(port_id) {
	275	rte_eth_dev_stop(port_id);
	276	rte_eth_dev_close(port_id);
	277	}
	278	exit(0);
	279	}
	280
7c673cae FG	281	int
	282	main(int argc, char *argv[])
	283	{
9f95a23c	284	signal(SIGINT, signal_handler);
7c673cae FG	285	/* initialise the system */
	286	if (init(argc, argv) < 0 )
	287	return -1;
	288	RTE_LOG(INFO, APP, "Finished Process Init.\n");
	289
	290	cl_rx_buf = calloc(num_clients, sizeof(cl_rx_buf[0]));
	291
	292	/* clear statistics */
	293	clear_stats();
	294
	295	/* put all other cores to sleep bar master */
	296	rte_eal_mp_remote_launch(sleep_lcore, NULL, SKIP_MASTER);
	297
	298	do_packet_forwarding();
	299	return 0;
	300	}