int main(int argc, char **argv) {
// eal args
int num_args = rte_eal_init(argc, argv);
if (num_args < 0)
rte_exit(EXIT_FAILURE, "init failed");
argc -= num_args;
argv += num_args;
// our args: [-s] port1 port2
uint8_t port1, port2;
char opt = getopt(argc, argv, "s");
bool simple_tx = opt == 's';
if (simple_tx) {
printf("Requesting simple tx path\n");
argc--;
argv++;
} else {
printf("Requesting full-featured tx path\n");
}
if (argc != 3) {
printf("usage: [-s] port1 port2\n");
return -1;
}
port1 = atoi(argv[1]);
port2 = atoi(argv[2]);
printf("Using ports %d and %d\n", port1, port2);
if (!config_port(port1, simple_tx)) return -1;
if (!config_port(port2, simple_tx)) return -1;
struct rte_mempool* pool = make_mempool();
uint64_t sent = 0;
uint64_t next_print = rte_get_tsc_hz();
uint64_t last_sent = 0;
while (true) {
sent += send_pkts(port1, pool);
sent += send_pkts(port2, pool);
uint64_t time = rte_rdtsc();
if (time >= next_print) {
double elapsed = (time - next_print + rte_get_tsc_hz()) / rte_get_tsc_hz();
uint64_t pkts = sent - last_sent;
printf("Packet rate: %.2f Mpps\n", (double) pkts / elapsed / 1000000);
next_print = time + rte_get_tsc_hz();
last_sent = sent;
}
}
return 0;
}
static void init_task_qinq_encap4(struct task_base *tbase, struct task_args *targ)
{
struct task_qinq_encap4 *task = (struct task_qinq_encap4 *)(tbase);
int socket_id = rte_lcore_to_socket_id(targ->lconf->id);
task->qinq_tag = targ->qinq_tag;
task->cpe_table = targ->cpe_table;
task->cpe_timeout = rte_get_tsc_hz()/1000*targ->cpe_table_timeout_ms;
if (!strcmp(targ->task_init->sub_mode_str, "pe")) {
PROX_PANIC(!strcmp(targ->cpe_table_name, ""), "CPE table not configured\n");
fill_table(targ, task->cpe_table);
}
#ifdef ENABLE_EXTRA_USER_STATISTICS
task->n_users = targ->n_users;
task->stats_per_user = rte_zmalloc_socket(NULL, targ->n_users * sizeof(uint32_t),
RTE_CACHE_LINE_SIZE, rte_lcore_to_socket_id(targ->lconf->id));
#endif
if (targ->runtime_flags & TASK_CLASSIFY) {
PROX_PANIC(!strcmp(targ->dscp, ""), "DSCP table not specified\n");
task->dscp = prox_sh_find_socket(socket_id, targ->dscp);
if (!task->dscp) {
int ret = lua_to_dscp(prox_lua(), GLOBAL, targ->dscp, socket_id, &task->dscp);
PROX_PANIC(ret, "Failed to create dscp table from config:\n%s\n",
get_lua_to_errors());
prox_sh_add_socket(socket_id, targ->dscp, task->dscp);
}
}
task->runtime_flags = targ->runtime_flags;
for (uint32_t i = 0; i < 64; ++i) {
task->fake_packets[i] = (struct rte_mbuf*)((uint8_t*)&task->keys[i] - sizeof (struct rte_mbuf));
}
targ->lconf->ctrl_timeout = rte_get_tsc_hz()/targ->ctrl_freq;
targ->lconf->ctrl_func_m[targ->task] = arp_msg;
/* TODO: check if it is not necessary to limit reverse mapping
for the elements that have been changing in mapping? */
for (uint32_t i =0 ; i < sizeof(targ->mapping)/sizeof(targ->mapping[0]); ++i) {
task->src_mac[targ->mapping[i]] = *(uint64_t*)&prox_port_cfg[i].eth_addr;
}
/* task->src_mac[entry->port_idx] = *(uint64_t*)&prox_port_cfg[entry->port_idx].eth_addr; */
}
static int
setup_ip_frag_tbl()
{
lcore_conf_t *qconf;
uint32_t max_flow_num = DEFAULT_FLOW_NUM;
uint32_t max_flow_ttl = DEFAULT_FLOW_TTL;
int socket;
uint64_t frag_cycles;
frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
max_flow_ttl;
for (int i = 0; i < sk.nr_lcore_ids; ++i) {
int lcore_id = sk.lcore_ids[i];
qconf = &sk.lcore_conf[lcore_id];
socket = rte_lcore_to_socket_id(lcore_id);
if (socket == SOCKET_ID_ANY) socket = 0;
if ((qconf->frag_tbl = rte_ip_frag_table_create(max_flow_num,
IP_FRAG_TBL_BUCKET_ENTRIES,
max_flow_num, frag_cycles,
socket)) == NULL)
{
RTE_LOG(ERR, "ip_frag_tbl_create(%u) on "
"lcore: %u failed\n",
max_flow_num, lcore_id);
return -1;
}
}
return 0;
}
// FIXME: support packet sizes here
static inline void main_loop_poisson(struct rte_ring* ring, uint8_t device, uint16_t queue, uint32_t target, uint32_t link_speed) {
uint64_t tsc_hz = rte_get_tsc_hz();
// control IPGs instead of IDT as IDTs < packet_time are physically impossible
std::default_random_engine rand;
uint64_t next_send = 0;
struct rte_mbuf* bufs[batch_size];
while (1) {
int rc = ring_dequeue(ring, reinterpret_cast<void**>(bufs), batch_size);
uint64_t cur = rte_get_tsc_cycles();
// nothing sent for 10 ms, restart rate control
if (((int64_t) cur - (int64_t) next_send) > (int64_t) tsc_hz / 100) {
next_send = cur;
}
if (rc == 0) {
uint32_t sent = 0;
while (sent < batch_size) {
uint64_t pkt_time = (bufs[sent]->pkt.pkt_len + 24) * 8 / (link_speed / 1000);
uint64_t avg = (uint64_t) (tsc_hz / (1000000000 / target) - pkt_time);
std::exponential_distribution<double> distribution(1.0 / avg);
while ((cur = rte_get_tsc_cycles()) < next_send);
next_send += distribution(rand) + pkt_time;
sent += rte_eth_tx_burst(device, queue, bufs + sent, 1);
}
}
}
}
static void
l2sw_main_process(struct lcore_env *env)
{
struct rte_mbuf *pkt_burst[MAX_PKT_BURST];
uint8_t n_ports = rte_eth_dev_count();
unsigned lcore_id = rte_lcore_id();
uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S
* BURST_TX_DRAIN_US;
//RTE_LOG(INFO, MARIO, "[%u] Starting main processing.\n", lcore_id);
prev_tsc = 0;
timer_tsc = 0;
while(1) {
cur_tsc = rte_rdtsc();
diff_tsc = cur_tsc - prev_tsc;
if (unlikely(diff_tsc > drain_tsc)) {
uint8_t port_id;
for(port_id = 0; port_id < n_ports; port_id++) {
if (env->tx_mbufs[port_id].len == 0)
continue;
l2sw_send_burst(env, port_id, env->tx_mbufs[port_id].len);
env->tx_mbufs[port_id].len = 0;
}
/* if timer is enabled */
if (timer_period > 0) {
/* advance the timer */
timer_tsc += diff_tsc;
/* if timer has reached its timeout */
if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
/* do this only on master core */
if (lcore_id == rte_get_master_lcore()) {
//print_stats(env);
/* reset the timer */
timer_tsc = 0;
}
}
}
prev_tsc = cur_tsc;
}
/* RX */
uint8_t port_id;
for (port_id = 0; port_id < n_ports; port_id++) {
unsigned n_rx = rte_eth_rx_burst(port_id, lcore_id,
pkt_burst, MAX_PKT_BURST);
if (n_rx != 0)
//RTE_LOG(INFO, MARIO, "[%u-%u] %u packet(s) came.\n",
// lcore_id, port_id, n_rx);
__sync_fetch_and_add(&port_statistics[port_id].rx, n_rx);
ether_in(env, pkt_burst, n_rx, port_id);
}
}
return ;
}
int tw_timer_start(tw_timer_t* timer_handle, tw_timer_cb timer_cb, uint64_t timeout) {
if (timer_handle == NULL)
return -1;
timer_handle->timer_cb = timer_cb;
timer_handle->timeout = (timeout) *( rte_get_tsc_hz() / 1000 );
return 0;
}
static void *
rte_port_ring_writer_ras_create(void *params, int socket_id, int is_ipv4)
{
struct rte_port_ring_writer_ras_params *conf =
params;
struct rte_port_ring_writer_ras *port;
uint64_t frag_cycles;
/* Check input parameters */
if (conf == NULL) {
RTE_LOG(ERR, PORT, "%s: Parameter conf is NULL\n", __func__);
return NULL;
}
if (conf->ring == NULL) {
RTE_LOG(ERR, PORT, "%s: Parameter ring is NULL\n", __func__);
return NULL;
}
if ((conf->tx_burst_sz == 0) ||
(conf->tx_burst_sz > RTE_PORT_IN_BURST_SIZE_MAX)) {
RTE_LOG(ERR, PORT, "%s: Parameter tx_burst_sz is invalid\n",
__func__);
return NULL;
}
/* Memory allocation */
port = rte_zmalloc_socket("PORT", sizeof(*port),
RTE_CACHE_LINE_SIZE, socket_id);
if (port == NULL) {
RTE_LOG(ERR, PORT, "%s: Failed to allocate socket\n", __func__);
return NULL;
}
/* Create fragmentation table */
frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S * MS_PER_S;
frag_cycles *= 100;
port->frag_tbl = rte_ip_frag_table_create(
RTE_PORT_RAS_N_BUCKETS,
RTE_PORT_RAS_N_ENTRIES_PER_BUCKET,
RTE_PORT_RAS_N_ENTRIES,
frag_cycles,
socket_id);
if (port->frag_tbl == NULL) {
RTE_LOG(ERR, PORT, "%s: rte_ip_frag_table_create failed\n",
__func__);
rte_free(port);
return NULL;
}
/* Initialization */
port->ring = conf->ring;
port->tx_burst_sz = conf->tx_burst_sz;
port->tx_buf_count = 0;
port->f_ras = (is_ipv4 == 1) ? process_ipv4 : process_ipv6;
return port;
}
static uint64_t tsc_extrapolate_backward(uint64_t tsc_from, uint64_t bytes, uint64_t tsc_minimum)
{
uint64_t tsc = tsc_from - rte_get_tsc_hz()*bytes/1250000000;
if (likely(tsc > tsc_minimum))
return tsc;
else
return tsc_minimum;
}
/**
* @brief StreamInfo copy constructor
*
* @param other Object to be copied
*
*/
DPDKAdapter::StreamInfo::StreamInfo(const StreamInfo& other)
{
devId_ = other.devId_;
delay_ = other.delay_;
lastTx_ = other.lastTx_;
numBursts_ = other.numBursts_;
txBurstSize_ = other.txBurstSize_;
burstSize_ = other.burstSize_;
sentPackets_ = other.sentPackets_;
sentBursts_ = other.sentBursts_;
curPacketId_ = other.curPacketId_;
ticksDelay_ = (uint64_t)delay_ * rte_get_tsc_hz() / SEC_TO_NSEC;
packets_ = other.packets_;
qDebug("ticksDelay_ %llu, CPU frequency %llu", ticksDelay_, rte_get_tsc_hz());
}
/**
* @brief StreamInfo constructor
*
* @param devId Port number
* @param delay Delay in nano seconds
* @param burstSize Stream burst size
* @param numBursts Number of bursts
* @param txBurstSize A minimal packets batch size put on the wire
*/
DPDKAdapter::StreamInfo::StreamInfo(unsigned char devId, unsigned int delay, unsigned int burstSize, unsigned int numBursts, uint8_t txBurstSize) :
devId_(devId),
delay_(delay),
lastTx_(0),
numBursts_(numBursts),
burstSize_(burstSize),
txBurstSize_(txBurstSize),
sentPackets_(0),
sentBursts_(0),
curPacketId_(0)
{
ticksDelay_ = (uint64_t)delay * rte_get_tsc_hz() / SEC_TO_NSEC;
qDebug("txBurstSize %u", txBurstSize);
qDebug("ticksDelay_ %llu, CPU frequency %llu", ticksDelay_, rte_get_tsc_hz());
}
static void
init_per_lcore() {
lcore_conf_t *qconf;
unsigned lcore_id = rte_lcore_id();
qconf = &sk.lcore_conf[lcore_id];
qconf->tsc_hz = rte_get_tsc_hz();
qconf->start_us = (uint64_t )ustime();
qconf->start_tsc = rte_rdtsc();
}
static void
print_trace(const char *msg, struct rte_keepalive *keepcfg, int idx_core)
{
RTE_LOG(INFO, EAL, "%sLast seen %" PRId64 "ms ago.\n",
msg,
((rte_rdtsc() - keepcfg->last_alive[idx_core])*1000)
/ rte_get_tsc_hz()
);
}
void
app_ping(void)
{
unsigned i;
uint64_t timestamp, diff_tsc;
const uint64_t timeout = rte_get_tsc_hz() * APP_PING_TIMEOUT_SEC;
for (i = 0; i < RTE_MAX_LCORE; i++) {
struct app_core_params *p = &app.cores[i];
struct rte_ring *ring_req, *ring_resp;
void *msg;
struct app_msg_req *req;
int status;
if ((p->core_type != APP_CORE_FC) &&
(p->core_type != APP_CORE_FW) &&
(p->core_type != APP_CORE_RT) &&
(p->core_type != APP_CORE_RX))
continue;
ring_req = app_get_ring_req(p->core_id);
ring_resp = app_get_ring_resp(p->core_id);
/* Fill request message */
msg = (void *)rte_ctrlmbuf_alloc(app.msg_pool);
if (msg == NULL)
rte_panic("Unable to allocate new message\n");
req = (struct app_msg_req *)
rte_ctrlmbuf_data((struct rte_mbuf *)msg);
req->type = APP_MSG_REQ_PING;
/* Send request */
do {
status = rte_ring_sp_enqueue(ring_req, msg);
} while (status == -ENOBUFS);
/* Wait for response */
timestamp = rte_rdtsc();
do {
status = rte_ring_sc_dequeue(ring_resp, &msg);
diff_tsc = rte_rdtsc() - timestamp;
if (unlikely(diff_tsc > timeout))
rte_panic("Core %u of type %d does not respond "
"to requests\n", p->core_id,
p->core_type);
} while (status != 0);
/* Free message buffer */
rte_ctrlmbuf_free(msg);
}
}
/*
* Flush packets scheduled for transmit on ports
*/
static void
flush_pkts(unsigned action)
{
unsigned i = 0;
uint16_t deq_count = PKT_BURST_SIZE;
struct rte_mbuf *pkts[PKT_BURST_SIZE] = {0};
struct port_queue *pq = &port_queues[action & PORT_MASK];
struct statistics *s = &vport_stats[action];
const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
uint64_t diff_tsc = 0;
static uint64_t prev_tsc[MAX_PHYPORTS] = {0};
uint64_t cur_tsc = rte_rdtsc();
unsigned num_pkts;
diff_tsc = cur_tsc - prev_tsc[action & PORT_MASK];
if (unlikely(rte_ring_count(pq->tx_q) >= PKT_BURST_SIZE))
{
num_pkts = PKT_BURST_SIZE;
}
else
{
/* If queue idles with less than PKT_BURST packets, drain it*/
if(unlikely(diff_tsc > drain_tsc)) {
num_pkts = rte_ring_count(pq->tx_q);
}
else {
return;
}
}
if (unlikely(rte_ring_dequeue_bulk(
pq->tx_q, (void **)pkts, num_pkts) != 0))
return;
const uint16_t sent = rte_eth_tx_burst(
ports->id[action & PORT_MASK], 0, pkts, num_pkts);
prev_tsc[action & PORT_MASK] = cur_tsc;
if (unlikely(sent < num_pkts))
{
for (i = sent; i < num_pkts; i++)
rte_pktmbuf_free(pkts[i]);
s->tx_drop += (num_pkts - sent);
}
else
{
s->tx += sent;
}
}
/**
* @brief DPDKProfiler destructor
*/
DPDKProfiler::~DPDKProfiler()
{
uint64_t end = rte_get_tsc_cycles();
stats_[coreId_][name_].last_duration = (end - start_) * SEC_TO_NSEC / rte_get_tsc_hz();
stats_[coreId_][name_].total_duration += stats_[coreId_][name_].last_duration;
stats_[coreId_][name_].invoke_cnt += 1;
if (stats_[coreId_][name_].invoke_cnt == 10000000)
{
qWarning("%s on core %u : last duration %lu, medium duration %lu", name_.c_str(), coreId_, lastDurationGet(coreId_, name_), medDurationGet(coreId_, name_));
reset(coreId_, name_);
}
}
/*
* Function sends unmatched packets to vswitchd.
*/
void
send_packet_to_vswitchd(struct rte_mbuf *mbuf, struct dpdk_upcall *info)
{
int rslt = 0;
void *mbuf_ptr = NULL;
const uint64_t dpif_send_tsc =
(rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * DPIF_SEND_US;
uint64_t cur_tsc = 0;
uint64_t diff_tsc = 0;
static uint64_t prev_tsc = 0;
/* send one packet, delete information about segments */
rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf);
/* allocate space before the packet for the upcall info */
mbuf_ptr = rte_pktmbuf_prepend(mbuf, sizeof(*info));
if (mbuf_ptr == NULL) {
printf("Cannot prepend upcall info\n");
rte_pktmbuf_free(mbuf);
stats_vswitch_tx_drop_increment(INC_BY_1);
stats_vport_tx_drop_increment(VSWITCHD, INC_BY_1);
return;
}
rte_memcpy(mbuf_ptr, info, sizeof(*info));
/* send the packet and the upcall info to the daemon */
rslt = rte_ring_mp_enqueue(vswitchd_packet_ring, mbuf);
if (rslt < 0) {
if (rslt == -ENOBUFS) {
rte_pktmbuf_free(mbuf);
stats_vswitch_tx_drop_increment(INC_BY_1);
stats_vport_tx_drop_increment(VSWITCHD, INC_BY_1);
return;
} else {
stats_vport_overrun_increment(VSWITCHD, INC_BY_1);
}
}
stats_vport_tx_increment(VSWITCHD, INC_BY_1);
cur_tsc = rte_rdtsc();
diff_tsc = cur_tsc - prev_tsc;
prev_tsc = cur_tsc;
/* Only signal the daemon after 100 milliseconds */
if (unlikely(diff_tsc > dpif_send_tsc))
send_signal_to_dpif();
}
// FIXME: link speed is hardcoded to 10gbit (but not really relevant for this use case where you should have only one packet anyways)
// this is only optimized for latency measurements/timestamping, not packet capture
// packet capturing would benefit from running the whole rx thread in C to avoid gc/jit pauses
uint16_t receive_with_timestamps_software(uint8_t port_id, uint16_t queue_id, struct rte_mbuf* rx_pkts[], uint16_t nb_pkts, uint64_t timestamps[]) {
uint32_t cycles_per_byte = rte_get_tsc_hz() / 10000000.0 / 0.8;
while (is_running()) {
uint64_t tsc = read_rdtsc();
uint16_t rx = rte_eth_rx_burst(port_id, queue_id, rx_pkts, nb_pkts);
uint16_t prev_pkt_size = 0;
for (int i = 0; i < rx; i++) {
timestamps[i] = tsc + prev_pkt_size * cycles_per_byte;
prev_pkt_size = rx_pkts[i]->pkt_len + 24;
}
if (rx > 0) {
return rx;
}
}
return 0;
}
static int tsc_diff_to_tv(uint64_t beg, uint64_t end, struct timeval *tv)
{
if (end < beg) {
return -1;
}
uint64_t diff = end - beg;
uint64_t sec_tsc = rte_get_tsc_hz();
uint64_t sec = diff/sec_tsc;
tv->tv_sec = sec;
diff -= sec*sec_tsc;
tv->tv_usec = diff*1000000/sec_tsc;
return 0;
}
static void
app_init_pipelines(struct app_params *app)
{
uint32_t p_id;
for (p_id = 0; p_id < app->n_pipelines; p_id++) {
struct app_pipeline_params *params =
&app->pipeline_params[p_id];
struct app_pipeline_data *data = &app->pipeline_data[p_id];
struct pipeline_type *ptype;
struct pipeline_params pp;
APP_LOG(app, HIGH, "Initializing %s ...", params->name);
ptype = app_pipeline_type_find(app, params->type);
if (ptype == NULL)
rte_panic("Init error: Unknown pipeline type \"%s\"\n",
params->type);
app_pipeline_params_get(app, params, &pp);
/* Back-end */
data->be = NULL;
if (ptype->be_ops->f_init) {
data->be = ptype->be_ops->f_init(&pp, (void *) app);
if (data->be == NULL)
rte_panic("Pipeline instance \"%s\" back-end "
"init error\n", params->name);
}
/* Front-end */
data->fe = NULL;
if (ptype->fe_ops->f_init) {
data->fe = ptype->fe_ops->f_init(&pp, (void *) app);
if (data->fe == NULL)
rte_panic("Pipeline instance \"%s\" front-end "
"init error\n", params->name);
}
data->ptype = ptype;
data->timer_period = (rte_get_tsc_hz() *
params->timer_period) / 100;
}
}
/**
* functional test for rte_meter_trtcm_color_blind_check
*/
static inline int
tm_test_trtcm_color_blind_check(void)
{
#define TRTCM_BLIND_CHECK_MSG "trtcm_blind_check"
uint64_t time;
struct rte_meter_trtcm tm;
uint64_t hz = rte_get_tsc_hz();
/* Test green */
if(rte_meter_trtcm_config(&tm, &tparams) != 0)
melog(TRTCM_BLIND_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_trtcm_color_blind_check(
&tm, time, TM_TEST_TRTCM_CBS_DF - 1)
!= e_RTE_METER_GREEN)
melog(TRTCM_BLIND_CHECK_MSG" GREEN");
/* Test yellow */
if(rte_meter_trtcm_config(&tm, &tparams) != 0)
melog(TRTCM_BLIND_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_trtcm_color_blind_check(
&tm, time, TM_TEST_TRTCM_CBS_DF + 1)
!= e_RTE_METER_YELLOW)
melog(TRTCM_BLIND_CHECK_MSG" YELLOW");
if(rte_meter_trtcm_config(&tm, &tparams) != 0)
melog(TRTCM_BLIND_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_trtcm_color_blind_check(
&tm, time, TM_TEST_TRTCM_PBS_DF - 1)
!= e_RTE_METER_YELLOW)
melog(TRTCM_BLIND_CHECK_MSG" YELLOW");
/* Test red */
if(rte_meter_trtcm_config(&tm, &tparams) != 0)
melog(TRTCM_BLIND_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_trtcm_color_blind_check(
&tm, time, TM_TEST_TRTCM_PBS_DF + 1)
!= e_RTE_METER_RED)
melog(TRTCM_BLIND_CHECK_MSG" RED");
return 0;
}
struct rte_keepalive *
rte_keepalive_create(rte_keepalive_failure_callback_t callback,
void *data)
{
struct rte_keepalive *keepcfg;
keepcfg = rte_zmalloc("RTE_EAL_KEEPALIVE",
sizeof(struct rte_keepalive),
RTE_CACHE_LINE_SIZE);
if (keepcfg != NULL) {
keepcfg->callback = callback;
keepcfg->callback_data = data;
keepcfg->tsc_initial = rte_rdtsc();
keepcfg->tsc_mhz = rte_get_tsc_hz() / 1000;
}
return keepcfg;
}
/*
* This function must be called by application to initialize.
* the rate of polling for driver, timer, readable & writable socket lists
* Paramters: drv_poll_interval,timer_poll_interval,tx_ready_sockets_poll_interval,
* rx_ready_sockets_poll_interval - all in micros
* Returns: None
*
*/
void app_glue_init_poll_intervals(int drv_poll_interval,
int timer_poll_interval,
int tx_ready_sockets_poll_interval,
int rx_ready_sockets_poll_interval)
{
syslog(LOG_INFO,"%s %d %d %d %d %d\n",__func__,__LINE__,
drv_poll_interval,timer_poll_interval,tx_ready_sockets_poll_interval,
rx_ready_sockets_poll_interval);
float cycles_in_micro = rte_get_tsc_hz()/1000000;
app_glue_drv_poll_interval = cycles_in_micro*(float)drv_poll_interval;
app_glue_timer_poll_interval = cycles_in_micro*(float)timer_poll_interval;
app_glue_tx_ready_sockets_poll_interval = cycles_in_micro*(float)tx_ready_sockets_poll_interval;
app_glue_rx_ready_sockets_poll_interval = cycles_in_micro*(float)rx_ready_sockets_poll_interval;
syslog(LOG_INFO,"%s %d %"PRIu64" %"PRIu64" %"PRIu64" %"PRIu64"\n",__func__,__LINE__,
app_glue_drv_poll_interval,app_glue_timer_poll_interval,
app_glue_tx_ready_sockets_poll_interval,app_glue_rx_ready_sockets_poll_interval);
}
/* Send burst of outgoing packet, if timeout expires. */
static inline void
send_timeout_burst(void)
{
uint64_t cur_tsc;
uint8_t port;
const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
cur_tsc = rte_rdtsc();
if (likely (cur_tsc < port_tx_conf.tx_tsc + drain_tsc))
return;
for (port = 0; port < MAX_PORTS; port++) {
if (port_tx_conf.tx_mbufs[port].len != 0)
send_burst(port);
}
port_tx_conf.tx_tsc = cur_tsc;
}
/**
* functional test for rte_meter_srtcm_color_blind_check
*/
static inline int
tm_test_srtcm_color_blind_check(void)
{
#define SRTCM_BLIND_CHECK_MSG "srtcm_blind_check"
struct rte_meter_srtcm sm;
uint64_t time;
uint64_t hz = rte_get_tsc_hz();
/* Test green */
if(rte_meter_srtcm_config(&sm, &sparams) != 0)
melog(SRTCM_BLIND_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_srtcm_color_blind_check(
&sm, time, TM_TEST_SRTCM_CBS_DF - 1)
!= e_RTE_METER_GREEN)
melog(SRTCM_BLIND_CHECK_MSG" GREEN");
/* Test yellow */
if(rte_meter_srtcm_config(&sm, &sparams) != 0)
melog(SRTCM_BLIND_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_srtcm_color_blind_check(
&sm, time, TM_TEST_SRTCM_CBS_DF + 1)
!= e_RTE_METER_YELLOW)
melog(SRTCM_BLIND_CHECK_MSG" YELLOW");
if(rte_meter_srtcm_config(&sm, &sparams) != 0)
melog(SRTCM_BLIND_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_srtcm_color_blind_check(
&sm, time, (uint32_t)sm.ebs - 1) != e_RTE_METER_YELLOW)
melog(SRTCM_BLIND_CHECK_MSG" YELLOW");
/* Test red */
if(rte_meter_srtcm_config(&sm, &sparams) != 0)
melog(SRTCM_BLIND_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_srtcm_color_blind_check(
&sm, time, TM_TEST_SRTCM_EBS_DF + 1)
!= e_RTE_METER_RED)
melog(SRTCM_BLIND_CHECK_MSG" RED");
return 0;
}
static inline int wait_command_handled(struct lcore_cfg *lconf)
{
uint64_t t1 = rte_rdtsc(), t2;
while (lconf_is_req(lconf)) {
t2 = rte_rdtsc();
if (t2 - t1 > 5 * rte_get_tsc_hz()) {
// Failed to handle command ...
for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
struct task_args *targs = &lconf->targs[task_id];
if (!(targs->flags & TASK_ARG_DROP)) {
plogx_err("Failed to handle command - task is in NO_DROP and might be stuck...\n");
return - 1;
}
}
plogx_err("Failed to handle command\n");
return -1;
}
}
return 0;
}
static int setup_prox(int argc, char **argv)
{
if (prox_read_config_file() != 0 ||
prox_setup_rte(argv[0]) != 0) {
return -1;
}
if (prox_cfg.flags & DSF_CHECK_SYNTAX) {
plog_info("=== Configuration file syntax has been checked ===\n\n");
exit(EXIT_SUCCESS);
}
init_port_activate();
plog_info("=== Initializing rte devices ===\n");
if (!(prox_cfg.flags & DSF_USE_DUMMY_DEVICES))
init_rte_ring_dev();
init_rte_dev(prox_cfg.flags & DSF_USE_DUMMY_DEVICES);
plog_info("=== Calibrating TSC overhead ===\n");
clock_init();
plog_info("\tTSC running at %"PRIu64" Hz\n", rte_get_tsc_hz());
init_lcores();
plog_info("=== Initializing ports ===\n");
init_port_all();
if (prox_cfg.logbuf_size) {
prox_cfg.logbuf = prox_zmalloc(prox_cfg.logbuf_size, rte_socket_id());
PROX_PANIC(prox_cfg.logbuf == NULL, "Failed to allocate memory for logbuf with size = %d\n", prox_cfg.logbuf_size);
}
if (prox_cfg.flags & DSF_CHECK_INIT) {
plog_info("=== Initialization sequence completed ===\n\n");
exit(EXIT_SUCCESS);
}
/* Current way that works to disable DPDK logging */
FILE *f = fopen("/dev/null", "r");
rte_openlog_stream(f);
plog_info("=== PROX started ===\n");
return 0;
}
int sflow_timer_callback() {
uint64_t expired_ticks = rte_rdtsc() - (rte_get_tsc_hz() * L4_SFLOW_EXPIRED_SECONDS);
int expired_sockets = 0;
sflow_socket_t* socket;
rte_rwlock_write_lock(&sflow_hash_lock);
for (int i = 0; i < L4_SFLOW_HASH_SIZE; ++i) {
socket = sflow_sockets[i];
if (!socket) continue;
if (socket->rx_ticks < expired_ticks) {
rte_spinlock_recursive_lock(&socket->lock);
sflow_socket_delete(&socket->key, 1);
rte_spinlock_recursive_unlock(&socket->lock);
++expired_sockets;
}
}
rte_rwlock_write_unlock(&sflow_hash_lock);
RTE_LOG(NOTICE, L3L4, "deleted %d expired sflow sockets\n", expired_sockets);
return 0;
}
/**
* @in[4] : the flags packets carries.
* @in[4] : the flags function expect to return.
* It will do blind check at the time of 1 second from beginning.
* At the time, it will use packets length of cbs -1, cbs + 1,
* ebs -1 and ebs +1 with flag in[0], in[1], in[2] and in[3] to do
* aware check, expect flag out[0], out[1], out[2] and out[3]
*/
static inline int
tm_test_trtcm_aware_check
(enum rte_meter_color in[4], enum rte_meter_color out[4])
{
#define TRTCM_AWARE_CHECK_MSG "trtcm_aware_check"
struct rte_meter_trtcm tm;
uint64_t time;
uint64_t hz = rte_get_tsc_hz();
if(rte_meter_trtcm_config(&tm, &tparams) != 0)
melog(TRTCM_AWARE_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_trtcm_color_aware_check(
&tm, time, TM_TEST_TRTCM_CBS_DF - 1, in[0]) != out[0])
melog(TRTCM_AWARE_CHECK_MSG" %u:%u", in[0], out[0]);
if(rte_meter_trtcm_config(&tm, &tparams) != 0)
melog(TRTCM_AWARE_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_trtcm_color_aware_check(
&tm, time, TM_TEST_TRTCM_CBS_DF + 1, in[1]) != out[1])
melog(TRTCM_AWARE_CHECK_MSG" %u:%u", in[1], out[1]);
if(rte_meter_trtcm_config(&tm, &tparams) != 0)
melog(TRTCM_AWARE_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_trtcm_color_aware_check(
&tm, time, TM_TEST_TRTCM_PBS_DF - 1, in[2]) != out[2])
melog(TRTCM_AWARE_CHECK_MSG" %u:%u", in[2], out[2]);
if(rte_meter_trtcm_config(&tm, &tparams) != 0)
melog(TRTCM_AWARE_CHECK_MSG);
time = rte_get_tsc_cycles() + hz;
if(rte_meter_trtcm_color_aware_check(
&tm, time, TM_TEST_TRTCM_PBS_DF + 1, in[3]) != out[3])
melog(TRTCM_AWARE_CHECK_MSG" %u:%u", in[3], out[3]);
return 0;
}
static inline void main_loop_cbr(struct rte_ring* ring, uint8_t device, uint16_t queue, uint32_t target) {
uint64_t tsc_hz = rte_get_tsc_hz();
uint64_t id_cycles = (uint64_t) (target / (1000000000.0 / ((double) tsc_hz)));
uint64_t next_send = 0;
struct rte_mbuf* bufs[batch_size];
while (1) {
int rc = ring_dequeue(ring, reinterpret_cast<void**>(bufs), batch_size);
uint64_t cur = rte_get_tsc_cycles();
// nothing sent for 10 ms, restart rate control
if (((int64_t) cur - (int64_t) next_send) > (int64_t) tsc_hz / 100) {
next_send = cur;
}
if (rc == 0) {
uint32_t sent = 0;
while (sent < batch_size) {
while ((cur = rte_get_tsc_cycles()) < next_send);
next_send += id_cycles;
sent += rte_eth_tx_burst(device, queue, bufs + sent, 1);
}
}
}
}
/**
* @brief Save mbuf burst to the capture buffer
*
* @param devId Port number
* @param burstBuf mbuf burst
* @param pktCount Number of packets in the burst
*
* @return true on success
*/
void DPDKAdapter::saveToBuf(uint8_t devId, MBuf_t** burstBuf, uint8_t pktCount)
{
MBuf_t* m = NULL;
DeviceInfo& devInfo = devices[devId];
uint64_t rxTicksEnd = rte_get_tsc_cycles();
uint64_t ticksDiff = rxTicksEnd - devInfo.rxTicksStart;
uint64_t timestamp = (SEC_TO_NSEC * ticksDiff) / rte_get_tsc_hz();
struct pcap_pkthdr hdr;
memset(&hdr, 0, sizeof(pcap_pkthdr));
uint32_t sec = timestamp / SEC_TO_NSEC;
hdr.ts.tv_sec = sec;
uint32_t usec = (timestamp - hdr.ts.tv_sec * SEC_TO_NSEC) / MSEC_TO_NSEC;
hdr.ts.tv_usec = usec;
for(uint8_t pkt = 0; pkt < pktCount; pkt++)
{
m = burstBuf[pkt];
hdr.caplen = m->pkt.data_len;
hdr.len = hdr.caplen;
if(devInfo.captureDataSize + sizeof(hdr) + m->pkt.data_len > devInfo.captureDataLength)
{
qDebug("Capture buffer is full with %u bytes", devInfo.captureDataSize);
devInfo.captureDataSize = 0;
}
memcpy(devInfo.captureData + devInfo.captureDataSize, &hdr, sizeof(hdr));
devInfo.captureDataSize += sizeof(hdr);
memcpy(devInfo.captureData + devInfo.captureDataSize, m->pkt.data, m->pkt.data_len);
devInfo.captureDataSize += m->pkt.data_len;
}
}
