/*-
* BSD LICENSE
*
- * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
- * All rights reserved.
+ * Copyright(c) 2010-2017 Intel Corporation. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
#include <rte_cycles.h>
#include <rte_malloc.h>
#include <rte_debug.h>
+#include <rte_prefetch.h>
#include <rte_distributor.h>
#define RX_RING_SIZE 256
#define TX_RING_SIZE 512
#define NUM_MBUFS ((64*1024)-1)
-#define MBUF_DATA_SIZE (2048 + RTE_PKTMBUF_HEADROOM)
#define MBUF_CACHE_SIZE 250
#define BURST_SIZE 32
#define RTE_RING_SZ 1024
-/* uncommnet below line to enable debug logs */
-/* #define DEBUG */
-
-#ifdef DEBUG
-#define LOG_LEVEL RTE_LOG_DEBUG
-#define LOG_DEBUG(log_type, fmt, args...) do { \
- RTE_LOG(DEBUG, log_type, fmt, ##args) \
-} while (0)
-#else
-#define LOG_LEVEL RTE_LOG_INFO
-#define LOG_DEBUG(log_type, fmt, args...) do {} while (0)
-#endif
-
#define RTE_LOGTYPE_DISTRAPP RTE_LOGTYPE_USER1
+#define ANSI_COLOR_RED "\x1b[31m"
+#define ANSI_COLOR_RESET "\x1b[0m"
+
/* mask of enabled ports */
static uint32_t enabled_port_mask;
volatile uint8_t quit_signal;
volatile uint8_t quit_signal_rx;
+volatile uint8_t quit_signal_dist;
+volatile uint8_t quit_signal_work;
static volatile struct app_stats {
struct {
uint64_t rx_pkts;
uint64_t returned_pkts;
uint64_t enqueued_pkts;
+ uint64_t enqdrop_pkts;
} rx __rte_cache_aligned;
+ int pad1 __rte_cache_aligned;
+
+ struct {
+ uint64_t in_pkts;
+ uint64_t ret_pkts;
+ uint64_t sent_pkts;
+ uint64_t enqdrop_pkts;
+ } dist __rte_cache_aligned;
+ int pad2 __rte_cache_aligned;
struct {
uint64_t dequeue_pkts;
uint64_t tx_pkts;
+ uint64_t enqdrop_pkts;
} tx __rte_cache_aligned;
+ int pad3 __rte_cache_aligned;
+
+ uint64_t worker_pkts[64] __rte_cache_aligned;
+
+ int pad4 __rte_cache_aligned;
+
+ uint64_t worker_bursts[64][8] __rte_cache_aligned;
+
+ int pad5 __rte_cache_aligned;
+
+ uint64_t port_rx_pkts[64] __rte_cache_aligned;
+ uint64_t port_tx_pkts[64] __rte_cache_aligned;
} app_stats;
+struct app_stats prev_app_stats;
+
static const struct rte_eth_conf port_conf_default = {
.rxmode = {
.mq_mode = ETH_MQ_RX_RSS,
struct rte_mbuf *mbufs[BURST_SIZE];
};
+static void print_stats(void);
+
/*
* Initialises a given port using global settings and with the rx buffers
* coming from the mbuf_pool passed as parameter
struct rte_eth_link link;
rte_eth_link_get_nowait(port, &link);
- if (!link.link_status) {
+ while (!link.link_status) {
+ printf("Waiting for Link up on port %"PRIu8"\n", port);
sleep(1);
rte_eth_link_get_nowait(port, &link);
}
struct rte_mempool *mem_pool;
};
-static void
+static int
quit_workers(struct rte_distributor *d, struct rte_mempool *p)
{
const unsigned num_workers = rte_lcore_count() - 2;
unsigned i;
struct rte_mbuf *bufs[num_workers];
- rte_mempool_get_bulk(p, (void *)bufs, num_workers);
+
+ if (rte_mempool_get_bulk(p, (void *)bufs, num_workers) != 0) {
+ printf("line %d: Error getting mbufs from pool\n", __LINE__);
+ return -1;
+ }
for (i = 0; i < num_workers; i++)
bufs[i]->hash.rss = i << 1;
rte_distributor_process(d, bufs, num_workers);
rte_mempool_put_bulk(p, (void *)bufs, num_workers);
+
+ return 0;
}
static int
struct rte_mbuf *bufs[BURST_SIZE*2];
const uint16_t nb_rx = rte_eth_rx_burst(port, 0, bufs,
BURST_SIZE);
+ if (unlikely(nb_rx == 0)) {
+ if (++port == nb_ports)
+ port = 0;
+ continue;
+ }
app_stats.rx.rx_pkts += nb_rx;
rte_distributor_process(d, bufs, nb_rx);
const uint16_t nb_ret = rte_distributor_returned_pkts(d,
bufs, BURST_SIZE*2);
app_stats.rx.returned_pkts += nb_ret;
- if (unlikely(nb_ret == 0))
+ if (unlikely(nb_ret == 0)) {
+ if (++port == nb_ports)
+ port = 0;
continue;
+ }
uint16_t sent = rte_ring_enqueue_burst(r, (void *)bufs, nb_ret);
app_stats.rx.enqueued_pkts += sent;
if (unlikely(sent < nb_ret)) {
- LOG_DEBUG(DISTRAPP, "%s:Packet loss due to full ring\n", __func__);
+ RTE_LOG_DP(DEBUG, DISTRAPP,
+ "%s:Packet loss due to full ring\n", __func__);
while (sent < nb_ret)
rte_pktmbuf_free(bufs[sent++]);
}
* get packets till quit_signal is actually been
* received and they gracefully shutdown
*/
- quit_workers(d, mem_pool);
+ if (quit_workers(d, mem_pool) != 0)
+ return -1;
/* rx thread should quit at last */
return 0;
}
app_stats.tx.tx_pkts += nb_tx;
if (unlikely(nb_tx < outbuf->count)) {
- LOG_DEBUG(DISTRAPP, "%s:Packet loss with tx_burst\n", __func__);
+ RTE_LOG_DP(DEBUG, DISTRAPP,
+ "%s:Packet loss with tx_burst\n", __func__);
do {
rte_pktmbuf_free(outbuf->mbufs[nb_tx]);
} while (++nb_tx < outbuf->count);
/* for traffic we receive, queue it up for transmit */
uint16_t i;
- _mm_prefetch(bufs[0], 0);
- _mm_prefetch(bufs[1], 0);
- _mm_prefetch(bufs[2], 0);
+ rte_prefetch_non_temporal((void *)bufs[0]);
+ rte_prefetch_non_temporal((void *)bufs[1]);
+ rte_prefetch_non_temporal((void *)bufs[2]);
for (i = 0; i < nb_rx; i++) {
struct output_buffer *outbuf;
uint8_t outp;
- _mm_prefetch(bufs[i + 3], 0);
+ rte_prefetch_non_temporal((void *)bufs[i + 3]);
/*
* workers should update in_port to hold the
* output port value
print_stats(void)
{
struct rte_eth_stats eth_stats;
- unsigned i;
-
- printf("\nRX thread stats:\n");
- printf(" - Received: %"PRIu64"\n", app_stats.rx.rx_pkts);
- printf(" - Processed: %"PRIu64"\n", app_stats.rx.returned_pkts);
- printf(" - Enqueued: %"PRIu64"\n", app_stats.rx.enqueued_pkts);
-
- printf("\nTX thread stats:\n");
- printf(" - Dequeued: %"PRIu64"\n", app_stats.tx.dequeue_pkts);
- printf(" - Transmitted: %"PRIu64"\n", app_stats.tx.tx_pkts);
+ unsigned int i, j;
+ const unsigned int num_workers = rte_lcore_count() - 4;
for (i = 0; i < rte_eth_dev_count(); i++) {
rte_eth_stats_get(i, ð_stats);
- printf("\nPort %u stats:\n", i);
- printf(" - Pkts in: %"PRIu64"\n", eth_stats.ipackets);
- printf(" - Pkts out: %"PRIu64"\n", eth_stats.opackets);
- printf(" - In Errs: %"PRIu64"\n", eth_stats.ierrors);
- printf(" - Out Errs: %"PRIu64"\n", eth_stats.oerrors);
- printf(" - Mbuf Errs: %"PRIu64"\n", eth_stats.rx_nombuf);
+ app_stats.port_rx_pkts[i] = eth_stats.ipackets;
+ app_stats.port_tx_pkts[i] = eth_stats.opackets;
+ }
+
+ printf("\n\nRX Thread:\n");
+ for (i = 0; i < rte_eth_dev_count(); i++) {
+ printf("Port %u Pktsin : %5.2f\n", i,
+ (app_stats.port_rx_pkts[i] -
+ prev_app_stats.port_rx_pkts[i])/1000000.0);
+ prev_app_stats.port_rx_pkts[i] = app_stats.port_rx_pkts[i];
+ }
+ printf(" - Received: %5.2f\n",
+ (app_stats.rx.rx_pkts -
+ prev_app_stats.rx.rx_pkts)/1000000.0);
+ printf(" - Returned: %5.2f\n",
+ (app_stats.rx.returned_pkts -
+ prev_app_stats.rx.returned_pkts)/1000000.0);
+ printf(" - Enqueued: %5.2f\n",
+ (app_stats.rx.enqueued_pkts -
+ prev_app_stats.rx.enqueued_pkts)/1000000.0);
+ printf(" - Dropped: %s%5.2f%s\n", ANSI_COLOR_RED,
+ (app_stats.rx.enqdrop_pkts -
+ prev_app_stats.rx.enqdrop_pkts)/1000000.0,
+ ANSI_COLOR_RESET);
+
+ printf("Distributor thread:\n");
+ printf(" - In: %5.2f\n",
+ (app_stats.dist.in_pkts -
+ prev_app_stats.dist.in_pkts)/1000000.0);
+ printf(" - Returned: %5.2f\n",
+ (app_stats.dist.ret_pkts -
+ prev_app_stats.dist.ret_pkts)/1000000.0);
+ printf(" - Sent: %5.2f\n",
+ (app_stats.dist.sent_pkts -
+ prev_app_stats.dist.sent_pkts)/1000000.0);
+ printf(" - Dropped %s%5.2f%s\n", ANSI_COLOR_RED,
+ (app_stats.dist.enqdrop_pkts -
+ prev_app_stats.dist.enqdrop_pkts)/1000000.0,
+ ANSI_COLOR_RESET);
+
+ printf("TX thread:\n");
+ printf(" - Dequeued: %5.2f\n",
+ (app_stats.tx.dequeue_pkts -
+ prev_app_stats.tx.dequeue_pkts)/1000000.0);
+ for (i = 0; i < rte_eth_dev_count(); i++) {
+ printf("Port %u Pktsout: %5.2f\n",
+ i, (app_stats.port_tx_pkts[i] -
+ prev_app_stats.port_tx_pkts[i])/1000000.0);
+ prev_app_stats.port_tx_pkts[i] = app_stats.port_tx_pkts[i];
+ }
+ printf(" - Transmitted: %5.2f\n",
+ (app_stats.tx.tx_pkts -
+ prev_app_stats.tx.tx_pkts)/1000000.0);
+ printf(" - Dropped: %s%5.2f%s\n", ANSI_COLOR_RED,
+ (app_stats.tx.enqdrop_pkts -
+ prev_app_stats.tx.enqdrop_pkts)/1000000.0,
+ ANSI_COLOR_RESET);
+
+ prev_app_stats.rx.rx_pkts = app_stats.rx.rx_pkts;
+ prev_app_stats.rx.returned_pkts = app_stats.rx.returned_pkts;
+ prev_app_stats.rx.enqueued_pkts = app_stats.rx.enqueued_pkts;
+ prev_app_stats.rx.enqdrop_pkts = app_stats.rx.enqdrop_pkts;
+ prev_app_stats.dist.in_pkts = app_stats.dist.in_pkts;
+ prev_app_stats.dist.ret_pkts = app_stats.dist.ret_pkts;
+ prev_app_stats.dist.sent_pkts = app_stats.dist.sent_pkts;
+ prev_app_stats.dist.enqdrop_pkts = app_stats.dist.enqdrop_pkts;
+ prev_app_stats.tx.dequeue_pkts = app_stats.tx.dequeue_pkts;
+ prev_app_stats.tx.tx_pkts = app_stats.tx.tx_pkts;
+ prev_app_stats.tx.enqdrop_pkts = app_stats.tx.enqdrop_pkts;
+
+ for (i = 0; i < num_workers; i++) {
+ printf("Worker %02u Pkts: %5.2f. Bursts(1-8): ", i,
+ (app_stats.worker_pkts[i] -
+ prev_app_stats.worker_pkts[i])/1000000.0);
+ for (j = 0; j < 8; j++) {
+ printf("%"PRIu64" ", app_stats.worker_bursts[i][j]);
+ app_stats.worker_bursts[i][j] = 0;
+ }
+ printf("\n");
+ prev_app_stats.worker_pkts[i] = app_stats.worker_pkts[i];
}
}
{
struct rte_distributor *d = p->d;
const unsigned id = p->worker_id;
+ unsigned int num = 0;
+ unsigned int i;
+
/*
* for single port, xor_val will be zero so we won't modify the output
* port, otherwise we send traffic from 0 to 1, 2 to 3, and vice versa
*/
const unsigned xor_val = (rte_eth_dev_count() > 1);
- struct rte_mbuf *buf = NULL;
+ struct rte_mbuf *buf[8] __rte_cache_aligned;
+
+ for (i = 0; i < 8; i++)
+ buf[i] = NULL;
printf("\nCore %u acting as worker core.\n", rte_lcore_id());
while (!quit_signal) {
- buf = rte_distributor_get_pkt(d, id, buf);
- buf->port ^= xor_val;
+ num = rte_distributor_get_pkt(d, id, buf, buf, num);
+ /* Do a little bit of work for each packet */
+ for (i = 0; i < num; i++) {
+ uint64_t t = rte_rdtsc()+100;
+
+ while (rte_rdtsc() < t)
+ rte_pause();
+ buf[i]->port ^= xor_val;
+ }
}
return 0;
}
argv[optind-1] = prgname;
- optind = 0; /* reset getopt lib */
+ optind = 1; /* reset getopt lib */
return 0;
}
unsigned nb_ports;
uint8_t portid;
uint8_t nb_ports_available;
+ uint64_t t, freq;
/* catch ctrl-c so we can print on exit */
signal(SIGINT, int_handler);
"when using a single port\n");
mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL",
- NUM_MBUFS * nb_ports, MBUF_CACHE_SIZE, 0, MBUF_DATA_SIZE,
- rte_socket_id());
+ NUM_MBUFS * nb_ports, MBUF_CACHE_SIZE, 0,
+ RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
if (mbuf_pool == NULL)
rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
nb_ports_available = nb_ports;
}
d = rte_distributor_create("PKT_DIST", rte_socket_id(),
- rte_lcore_count() - 2);
+ rte_lcore_count() - 2,
+ RTE_DIST_ALG_BURST);
if (d == NULL)
rte_exit(EXIT_FAILURE, "Cannot create distributor\n");
}
/* call lcore_main on master core only */
struct lcore_params p = { 0, d, output_ring, mbuf_pool};
- lcore_rx(&p);
+
+ if (lcore_rx(&p) != 0)
+ return -1;
+
+ freq = rte_get_timer_hz();
+ t = rte_rdtsc() + freq;
+ while (!quit_signal_dist) {
+ if (t < rte_rdtsc()) {
+ print_stats();
+ t = rte_rdtsc() + freq;
+ }
+ usleep(1000);
+ }
RTE_LCORE_FOREACH_SLAVE(lcore_id) {
if (rte_eal_wait_lcore(lcore_id) < 0)