examples: fix port mask parsing failure handling
[dpdk.git] / examples / l3fwd-power / main.c
index bcd584d..afa38c1 100644 (file)
@@ -14,6 +14,7 @@
 #include <getopt.h>
 #include <unistd.h>
 #include <signal.h>
+#include <math.h>
 
 #include <rte_common.h>
 #include <rte_byteorder.h>
@@ -44,6 +45,8 @@
 #include <rte_power.h>
 #include <rte_spinlock.h>
 #include <rte_power_empty_poll.h>
+#include <rte_metrics.h>
+#include <rte_telemetry.h>
 
 #include "perf_core.h"
 #include "main.h"
 #define EMPTY_POLL_MED_THRESHOLD 350000UL
 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
 
-
+#define NUM_TELSTATS RTE_DIM(telstats_strings)
 
 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
@@ -146,17 +149,61 @@ static uint32_t enabled_port_mask = 0;
 static int promiscuous_on = 0;
 /* NUMA is enabled by default. */
 static int numa_on = 1;
-/* emptypoll is disabled by default. */
-static bool empty_poll_on;
+static bool empty_poll_stop;
 static bool empty_poll_train;
-volatile bool empty_poll_stop;
+volatile bool quit_signal;
 static struct  ep_params *ep_params;
 static struct  ep_policy policy;
 static long  ep_med_edpi, ep_hgh_edpi;
+/* timer to update telemetry every 500ms */
+static struct rte_timer telemetry_timer;
+
+/* stats index returned by metrics lib */
+int telstats_index;
+
+struct telstats_name {
+       char name[RTE_ETH_XSTATS_NAME_SIZE];
+};
+
+/* telemetry stats to be reported */
+const struct telstats_name telstats_strings[] = {
+       {"empty_poll"},
+       {"full_poll"},
+       {"busy_percent"}
+};
+
+/* core busyness in percentage */
+enum busy_rate {
+       ZERO = 0,
+       PARTIAL = 50,
+       FULL = 100
+};
+
+/* reference poll count to measure core busyness */
+#define DEFAULT_COUNT 10000
+/*
+ * reference CYCLES to be used to
+ * measure core busyness based on poll count
+ */
+#define MIN_CYCLES  1500000ULL
+#define MAX_CYCLES 22000000ULL
+
+/* (500ms) */
+#define TELEMETRY_INTERVALS_PER_SEC 2
 
 static int parse_ptype; /**< Parse packet type using rx callback, and */
                        /**< disabled by default */
 
+enum appmode {
+       APP_MODE_DEFAULT = 0,
+       APP_MODE_LEGACY,
+       APP_MODE_EMPTY_POLL,
+       APP_MODE_TELEMETRY,
+       APP_MODE_INTERRUPT
+};
+
+enum appmode app_mode;
+
 enum freq_scale_hint_t
 {
        FREQ_LOWER    =      -1,
@@ -194,13 +241,12 @@ static struct lcore_params lcore_params_array_default[] = {
 };
 
 struct lcore_params *lcore_params = lcore_params_array_default;
-uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
-                               sizeof(lcore_params_array_default[0]);
+uint16_t nb_lcore_params = RTE_DIM(lcore_params_array_default);
 
 static struct rte_eth_conf port_conf = {
        .rxmode = {
                .mq_mode        = ETH_MQ_RX_RSS,
-               .max_rx_pkt_len = ETHER_MAX_LEN,
+               .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
                .split_hdr_size = 0,
                .offloads = DEV_RX_OFFLOAD_CHECKSUM,
        },
@@ -212,10 +258,7 @@ static struct rte_eth_conf port_conf = {
        },
        .txmode = {
                .mq_mode = ETH_MQ_TX_NONE,
-       },
-       .intr_conf = {
-               .rxq = 1,
-       },
+       }
 };
 
 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
@@ -237,7 +280,7 @@ struct ipv4_5tuple {
        uint16_t port_dst;
        uint16_t port_src;
        uint8_t  proto;
-} __attribute__((__packed__));
+} __rte_packed;
 
 struct ipv6_5tuple {
        uint8_t  ip_dst[IPV6_ADDR_LEN];
@@ -245,7 +288,7 @@ struct ipv6_5tuple {
        uint16_t port_dst;
        uint16_t port_src;
        uint8_t  proto;
-} __attribute__((__packed__));
+} __rte_packed;
 
 struct ipv4_l3fwd_route {
        struct ipv4_5tuple key;
@@ -258,10 +301,10 @@ struct ipv6_l3fwd_route {
 };
 
 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
-       {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
-       {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
-       {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
-       {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
+       {{RTE_IPV4(100,10,0,1), RTE_IPV4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
+       {{RTE_IPV4(100,20,0,2), RTE_IPV4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
+       {{RTE_IPV4(100,30,0,3), RTE_IPV4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
+       {{RTE_IPV4(100,40,0,4), RTE_IPV4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 };
 
 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
@@ -282,12 +325,6 @@ static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 
 #define L3FWD_HASH_ENTRIES     1024
 
-#define IPV4_L3FWD_NUM_ROUTES \
-       (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
-
-#define IPV6_L3FWD_NUM_ROUTES \
-       (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
-
 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 #endif
@@ -300,19 +337,16 @@ struct ipv4_l3fwd_route {
 };
 
 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
-       {IPv4(1,1,1,0), 24, 0},
-       {IPv4(2,1,1,0), 24, 1},
-       {IPv4(3,1,1,0), 24, 2},
-       {IPv4(4,1,1,0), 24, 3},
-       {IPv4(5,1,1,0), 24, 4},
-       {IPv4(6,1,1,0), 24, 5},
-       {IPv4(7,1,1,0), 24, 6},
-       {IPv4(8,1,1,0), 24, 7},
+       {RTE_IPV4(1,1,1,0), 24, 0},
+       {RTE_IPV4(2,1,1,0), 24, 1},
+       {RTE_IPV4(3,1,1,0), 24, 2},
+       {RTE_IPV4(4,1,1,0), 24, 3},
+       {RTE_IPV4(5,1,1,0), 24, 4},
+       {RTE_IPV4(6,1,1,0), 24, 5},
+       {RTE_IPV4(7,1,1,0), 24, 6},
+       {RTE_IPV4(8,1,1,0), 24, 7},
 };
 
-#define IPV4_L3FWD_NUM_ROUTES \
-       (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
-
 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 
 typedef struct rte_lpm lookup_struct_t;
@@ -341,7 +375,26 @@ struct lcore_stats {
        uint64_t nb_rx_processed;
        /* total iterations looped recently */
        uint64_t nb_iteration_looped;
-       uint32_t padding[9];
+       /*
+        * Represents empty and non empty polls
+        * of rte_eth_rx_burst();
+        * ep_nep[0] holds non empty polls
+        * i.e. 0 < nb_rx <= MAX_BURST
+        * ep_nep[1] holds empty polls.
+        * i.e. nb_rx == 0
+        */
+       uint64_t ep_nep[2];
+       /*
+        * Represents full and empty+partial
+        * polls of rte_eth_rx_burst();
+        * ep_nep[0] holds empty+partial polls.
+        * i.e. 0 <= nb_rx < MAX_BURST
+        * ep_nep[1] holds full polls
+        * i.e. nb_rx == MAX_BURST
+        */
+       uint64_t fp_nfp[2];
+       enum busy_rate br;
+       rte_spinlock_t telemetry_lock;
 } __rte_cache_aligned;
 
 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
@@ -362,53 +415,23 @@ static uint8_t  freq_tlb[] = {14, 9, 1};
 
 static int is_done(void)
 {
-       return empty_poll_stop;
+       return quit_signal;
 }
 
 /* exit signal handler */
 static void
 signal_exit_now(int sigtype)
 {
-       unsigned lcore_id;
-       unsigned int portid;
-       int ret;
-
-       if (sigtype == SIGINT) {
-               if (empty_poll_on)
-                       empty_poll_stop = true;
-
 
-               for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
-                       if (rte_lcore_is_enabled(lcore_id) == 0)
-                               continue;
+       if (sigtype == SIGINT)
+               quit_signal = true;
 
-                       /* init power management library */
-                       ret = rte_power_exit(lcore_id);
-                       if (ret)
-                               rte_exit(EXIT_FAILURE, "Power management "
-                                       "library de-initialization failed on "
-                                                       "core%u\n", lcore_id);
-               }
-
-               if (!empty_poll_on) {
-                       RTE_ETH_FOREACH_DEV(portid) {
-                               if ((enabled_port_mask & (1 << portid)) == 0)
-                                       continue;
-
-                               rte_eth_dev_stop(portid);
-                               rte_eth_dev_close(portid);
-                       }
-               }
-       }
-
-       if (!empty_poll_on)
-               rte_exit(EXIT_SUCCESS, "User forced exit\n");
 }
 
 /*  Freqency scale down timer callback */
 static void
-power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
-                         __attribute__((unused)) void *arg)
+power_timer_cb(__rte_unused struct rte_timer *tim,
+                         __rte_unused void *arg)
 {
        uint64_t hz;
        float sleep_time_ratio;
@@ -466,14 +489,14 @@ send_single_packet(struct rte_mbuf *m, uint16_t port)
 
 #ifdef DO_RFC_1812_CHECKS
 static inline int
-is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
+is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 {
        /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
        /*
         * 1. The packet length reported by the Link Layer must be large
         * enough to hold the minimum length legal IP datagram (20 bytes).
         */
-       if (link_len < sizeof(struct ipv4_hdr))
+       if (link_len < sizeof(struct rte_ipv4_hdr))
                return -1;
 
        /* 2. The IP checksum must be correct. */
@@ -498,7 +521,7 @@ is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
         * datagram header, whose length is specified in the IP header length
         * field.
         */
-       if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
+       if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
                return -5;
 
        return 0;
@@ -523,12 +546,12 @@ print_ipv6_key(struct ipv6_5tuple key)
 }
 
 static inline uint16_t
-get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
+get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
                lookup_struct_t * ipv4_l3fwd_lookup_struct)
 {
        struct ipv4_5tuple key;
-       struct tcp_hdr *tcp;
-       struct udp_hdr *udp;
+       struct rte_tcp_hdr *tcp;
+       struct rte_udp_hdr *udp;
        int ret = 0;
 
        key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
@@ -537,15 +560,15 @@ get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 
        switch (ipv4_hdr->next_proto_id) {
        case IPPROTO_TCP:
-               tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
-                                       sizeof(struct ipv4_hdr));
+               tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
+                                       sizeof(struct rte_ipv4_hdr));
                key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
                key.port_src = rte_be_to_cpu_16(tcp->src_port);
                break;
 
        case IPPROTO_UDP:
-               udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
-                                       sizeof(struct ipv4_hdr));
+               udp = (struct rte_udp_hdr *)((unsigned char *)ipv4_hdr +
+                                       sizeof(struct rte_ipv4_hdr));
                key.port_dst = rte_be_to_cpu_16(udp->dst_port);
                key.port_src = rte_be_to_cpu_16(udp->src_port);
                break;
@@ -562,12 +585,12 @@ get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 }
 
 static inline uint16_t
-get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
+get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
                        lookup_struct_t *ipv6_l3fwd_lookup_struct)
 {
        struct ipv6_5tuple key;
-       struct tcp_hdr *tcp;
-       struct udp_hdr *udp;
+       struct rte_tcp_hdr *tcp;
+       struct rte_udp_hdr *udp;
        int ret = 0;
 
        memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
@@ -577,15 +600,15 @@ get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
 
        switch (ipv6_hdr->proto) {
        case IPPROTO_TCP:
-               tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
-                                       sizeof(struct ipv6_hdr));
+               tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
+                                       sizeof(struct rte_ipv6_hdr));
                key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
                key.port_src = rte_be_to_cpu_16(tcp->src_port);
                break;
 
        case IPPROTO_UDP:
-               udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
-                                       sizeof(struct ipv6_hdr));
+               udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
+                                       sizeof(struct rte_ipv6_hdr));
                key.port_dst = rte_be_to_cpu_16(udp->dst_port);
                key.port_src = rte_be_to_cpu_16(udp->src_port);
                break;
@@ -604,7 +627,7 @@ get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
 
 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 static inline uint16_t
-get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
+get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
                lookup_struct_t *ipv4_l3fwd_lookup_struct)
 {
        uint32_t next_hop;
@@ -624,9 +647,9 @@ parse_ptype_one(struct rte_mbuf *m)
 
        eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
        ether_type = eth_hdr->ether_type;
-       if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
+       if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
                packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
-       else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
+       else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
                packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 
        m->packet_type = packet_type;
@@ -662,7 +685,7 @@ l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
                                struct lcore_conf *qconf)
 {
        struct rte_ether_hdr *eth_hdr;
-       struct ipv4_hdr *ipv4_hdr;
+       struct rte_ipv4_hdr *ipv4_hdr;
        void *d_addr_bytes;
        uint16_t dst_port;
 
@@ -671,7 +694,7 @@ l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
        if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
                /* Handle IPv4 headers.*/
                ipv4_hdr =
-                       rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
+                       rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
                                                sizeof(struct rte_ether_hdr));
 
 #ifdef DO_RFC_1812_CHECKS
@@ -700,16 +723,17 @@ l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 #endif
 
                /* src addr */
-               ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
+               rte_ether_addr_copy(&ports_eth_addr[dst_port],
+                               &eth_hdr->s_addr);
 
                send_single_packet(m, dst_port);
        } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
                /* Handle IPv6 headers.*/
 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
-               struct ipv6_hdr *ipv6_hdr;
+               struct rte_ipv6_hdr *ipv6_hdr;
 
                ipv6_hdr =
-                       rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
+                       rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
                                                sizeof(struct rte_ether_hdr));
 
                dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
@@ -725,7 +749,8 @@ l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
                        0x000000000002 + ((uint64_t)dst_port << 40);
 
                /* src addr */
-               ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
+               rte_ether_addr_copy(&ports_eth_addr[dst_port],
+                               &eth_hdr->s_addr);
 
                send_single_packet(m, dst_port);
 #else
@@ -798,33 +823,40 @@ power_freq_scaleup_heuristic(unsigned lcore_id,
 static int
 sleep_until_rx_interrupt(int num)
 {
+       /*
+        * we want to track when we are woken up by traffic so that we can go
+        * back to sleep again without log spamming.
+        */
+       static bool timeout;
        struct rte_epoll_event event[num];
        int n, i;
        uint16_t port_id;
        uint8_t queue_id;
        void *data;
 
-       RTE_LOG(INFO, L3FWD_POWER,
-               "lcore %u sleeps until interrupt triggers\n",
-               rte_lcore_id());
+       if (!timeout) {
+               RTE_LOG(INFO, L3FWD_POWER,
+                               "lcore %u sleeps until interrupt triggers\n",
+                               rte_lcore_id());
+       }
 
-       n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
+       n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, 10);
        for (i = 0; i < n; i++) {
                data = event[i].epdata.data;
                port_id = ((uintptr_t)data) >> CHAR_BIT;
                queue_id = ((uintptr_t)data) &
                        RTE_LEN2MASK(CHAR_BIT, uint8_t);
-               rte_eth_dev_rx_intr_disable(port_id, queue_id);
                RTE_LOG(INFO, L3FWD_POWER,
                        "lcore %u is waked up from rx interrupt on"
                        " port %d queue %d\n",
                        rte_lcore_id(), port_id, queue_id);
        }
+       timeout = n == 0;
 
        return 0;
 }
 
-static void turn_on_intr(struct lcore_conf *qconf)
+static void turn_on_off_intr(struct lcore_conf *qconf, bool on)
 {
        int i;
        struct lcore_rx_queue *rx_queue;
@@ -837,7 +869,10 @@ static void turn_on_intr(struct lcore_conf *qconf)
                queue_id = rx_queue->queue_id;
 
                rte_spinlock_lock(&(locks[port_id]));
-               rte_eth_dev_rx_intr_enable(port_id, queue_id);
+               if (on)
+                       rte_eth_dev_rx_intr_enable(port_id, queue_id);
+               else
+                       rte_eth_dev_rx_intr_disable(port_id, queue_id);
                rte_spinlock_unlock(&(locks[port_id]));
        }
 }
@@ -867,9 +902,293 @@ static int event_register(struct lcore_conf *qconf)
 
        return 0;
 }
+
+/* main processing loop */
+static int main_intr_loop(__rte_unused void *dummy)
+{
+       struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
+       unsigned int lcore_id;
+       uint64_t prev_tsc, diff_tsc, cur_tsc;
+       int i, j, nb_rx;
+       uint8_t queueid;
+       uint16_t portid;
+       struct lcore_conf *qconf;
+       struct lcore_rx_queue *rx_queue;
+       uint32_t lcore_rx_idle_count = 0;
+       uint32_t lcore_idle_hint = 0;
+       int intr_en = 0;
+
+       const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
+                                  US_PER_S * BURST_TX_DRAIN_US;
+
+       prev_tsc = 0;
+
+       lcore_id = rte_lcore_id();
+       qconf = &lcore_conf[lcore_id];
+
+       if (qconf->n_rx_queue == 0) {
+               RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
+                               lcore_id);
+               return 0;
+       }
+
+       RTE_LOG(INFO, L3FWD_POWER, "entering main interrupt loop on lcore %u\n",
+                       lcore_id);
+
+       for (i = 0; i < qconf->n_rx_queue; i++) {
+               portid = qconf->rx_queue_list[i].port_id;
+               queueid = qconf->rx_queue_list[i].queue_id;
+               RTE_LOG(INFO, L3FWD_POWER,
+                               " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
+                               lcore_id, portid, queueid);
+       }
+
+       /* add into event wait list */
+       if (event_register(qconf) == 0)
+               intr_en = 1;
+       else
+               RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
+
+       while (!is_done()) {
+               stats[lcore_id].nb_iteration_looped++;
+
+               cur_tsc = rte_rdtsc();
+
+               /*
+                * TX burst queue drain
+                */
+               diff_tsc = cur_tsc - prev_tsc;
+               if (unlikely(diff_tsc > drain_tsc)) {
+                       for (i = 0; i < qconf->n_tx_port; ++i) {
+                               portid = qconf->tx_port_id[i];
+                               rte_eth_tx_buffer_flush(portid,
+                                               qconf->tx_queue_id[portid],
+                                               qconf->tx_buffer[portid]);
+                       }
+                       prev_tsc = cur_tsc;
+               }
+
+start_rx:
+               /*
+                * Read packet from RX queues
+                */
+               lcore_rx_idle_count = 0;
+               for (i = 0; i < qconf->n_rx_queue; ++i) {
+                       rx_queue = &(qconf->rx_queue_list[i]);
+                       rx_queue->idle_hint = 0;
+                       portid = rx_queue->port_id;
+                       queueid = rx_queue->queue_id;
+
+                       nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
+                                       MAX_PKT_BURST);
+
+                       stats[lcore_id].nb_rx_processed += nb_rx;
+                       if (unlikely(nb_rx == 0)) {
+                               /**
+                                * no packet received from rx queue, try to
+                                * sleep for a while forcing CPU enter deeper
+                                * C states.
+                                */
+                               rx_queue->zero_rx_packet_count++;
+
+                               if (rx_queue->zero_rx_packet_count <=
+                                               MIN_ZERO_POLL_COUNT)
+                                       continue;
+
+                               rx_queue->idle_hint = power_idle_heuristic(
+                                               rx_queue->zero_rx_packet_count);
+                               lcore_rx_idle_count++;
+                       } else {
+                               rx_queue->zero_rx_packet_count = 0;
+                       }
+
+                       /* Prefetch first packets */
+                       for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
+                               rte_prefetch0(rte_pktmbuf_mtod(
+                                               pkts_burst[j], void *));
+                       }
+
+                       /* Prefetch and forward already prefetched packets */
+                       for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
+                               rte_prefetch0(rte_pktmbuf_mtod(
+                                               pkts_burst[j + PREFETCH_OFFSET],
+                                               void *));
+                               l3fwd_simple_forward(
+                                               pkts_burst[j], portid, qconf);
+                       }
+
+                       /* Forward remaining prefetched packets */
+                       for (; j < nb_rx; j++) {
+                               l3fwd_simple_forward(
+                                               pkts_burst[j], portid, qconf);
+                       }
+               }
+
+               if (unlikely(lcore_rx_idle_count == qconf->n_rx_queue)) {
+                       /**
+                        * All Rx queues empty in recent consecutive polls,
+                        * sleep in a conservative manner, meaning sleep as
+                        * less as possible.
+                        */
+                       for (i = 1,
+                           lcore_idle_hint = qconf->rx_queue_list[0].idle_hint;
+                                       i < qconf->n_rx_queue; ++i) {
+                               rx_queue = &(qconf->rx_queue_list[i]);
+                               if (rx_queue->idle_hint < lcore_idle_hint)
+                                       lcore_idle_hint = rx_queue->idle_hint;
+                       }
+
+                       if (lcore_idle_hint < SUSPEND_THRESHOLD)
+                               /**
+                                * execute "pause" instruction to avoid context
+                                * switch which generally take hundred of
+                                * microseconds for short sleep.
+                                */
+                               rte_delay_us(lcore_idle_hint);
+                       else {
+                               /* suspend until rx interrupt triggers */
+                               if (intr_en) {
+                                       turn_on_off_intr(qconf, 1);
+                                       sleep_until_rx_interrupt(
+                                                       qconf->n_rx_queue);
+                                       turn_on_off_intr(qconf, 0);
+                                       /**
+                                        * start receiving packets immediately
+                                        */
+                                       if (likely(!is_done()))
+                                               goto start_rx;
+                               }
+                       }
+                       stats[lcore_id].sleep_time += lcore_idle_hint;
+               }
+       }
+
+       return 0;
+}
+
 /* main processing loop */
 static int
-main_empty_poll_loop(__attribute__((unused)) void *dummy)
+main_telemetry_loop(__rte_unused void *dummy)
+{
+       struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
+       unsigned int lcore_id;
+       uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
+       int i, j, nb_rx;
+       uint8_t queueid;
+       uint16_t portid;
+       struct lcore_conf *qconf;
+       struct lcore_rx_queue *rx_queue;
+       uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
+       uint64_t poll_count;
+       enum busy_rate br;
+
+       const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
+                                       US_PER_S * BURST_TX_DRAIN_US;
+
+       poll_count = 0;
+       prev_tsc = 0;
+       prev_tel_tsc = 0;
+
+       lcore_id = rte_lcore_id();
+       qconf = &lcore_conf[lcore_id];
+
+       if (qconf->n_rx_queue == 0) {
+               RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
+                       lcore_id);
+               return 0;
+       }
+
+       RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
+               lcore_id);
+
+       for (i = 0; i < qconf->n_rx_queue; i++) {
+               portid = qconf->rx_queue_list[i].port_id;
+               queueid = qconf->rx_queue_list[i].queue_id;
+               RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
+                       "rxqueueid=%hhu\n", lcore_id, portid, queueid);
+       }
+
+       while (!is_done()) {
+
+               cur_tsc = rte_rdtsc();
+               /*
+                * TX burst queue drain
+                */
+               diff_tsc = cur_tsc - prev_tsc;
+               if (unlikely(diff_tsc > drain_tsc)) {
+                       for (i = 0; i < qconf->n_tx_port; ++i) {
+                               portid = qconf->tx_port_id[i];
+                               rte_eth_tx_buffer_flush(portid,
+                                               qconf->tx_queue_id[portid],
+                                               qconf->tx_buffer[portid]);
+                       }
+                       prev_tsc = cur_tsc;
+               }
+
+               /*
+                * Read packet from RX queues
+                */
+               for (i = 0; i < qconf->n_rx_queue; ++i) {
+                       rx_queue = &(qconf->rx_queue_list[i]);
+                       portid = rx_queue->port_id;
+                       queueid = rx_queue->queue_id;
+
+                       nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
+                                                               MAX_PKT_BURST);
+                       ep_nep[nb_rx == 0]++;
+                       fp_nfp[nb_rx == MAX_PKT_BURST]++;
+                       poll_count++;
+                       if (unlikely(nb_rx == 0))
+                               continue;
+
+                       /* Prefetch first packets */
+                       for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
+                               rte_prefetch0(rte_pktmbuf_mtod(
+                                               pkts_burst[j], void *));
+                       }
+
+                       /* Prefetch and forward already prefetched packets */
+                       for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
+                               rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
+                                               j + PREFETCH_OFFSET], void *));
+                               l3fwd_simple_forward(pkts_burst[j], portid,
+                                                               qconf);
+                       }
+
+                       /* Forward remaining prefetched packets */
+                       for (; j < nb_rx; j++) {
+                               l3fwd_simple_forward(pkts_burst[j], portid,
+                                                               qconf);
+                       }
+               }
+               if (unlikely(poll_count >= DEFAULT_COUNT)) {
+                       diff_tsc = cur_tsc - prev_tel_tsc;
+                       if (diff_tsc >= MAX_CYCLES) {
+                               br = FULL;
+                       } else if (diff_tsc > MIN_CYCLES &&
+                                       diff_tsc < MAX_CYCLES) {
+                               br = (diff_tsc * 100) / MAX_CYCLES;
+                       } else {
+                               br = ZERO;
+                       }
+                       poll_count = 0;
+                       prev_tel_tsc = cur_tsc;
+                       /* update stats for telemetry */
+                       rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
+                       stats[lcore_id].ep_nep[0] = ep_nep[0];
+                       stats[lcore_id].ep_nep[1] = ep_nep[1];
+                       stats[lcore_id].fp_nfp[0] = fp_nfp[0];
+                       stats[lcore_id].fp_nfp[1] = fp_nfp[1];
+                       stats[lcore_id].br = br;
+                       rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
+               }
+       }
+
+       return 0;
+}
+/* main processing loop */
+static int
+main_empty_poll_loop(__rte_unused void *dummy)
 {
        struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
        unsigned int lcore_id;
@@ -973,7 +1292,7 @@ main_empty_poll_loop(__attribute__((unused)) void *dummy)
 }
 /* main processing loop */
 static int
-main_loop(__attribute__((unused)) void *dummy)
+main_legacy_loop(__rte_unused void *dummy)
 {
        struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
        unsigned lcore_id;
@@ -1018,7 +1337,7 @@ main_loop(__attribute__((unused)) void *dummy)
        else
                RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 
-       while (1) {
+       while (!is_done()) {
                stats[lcore_id].nb_iteration_looped++;
 
                cur_tsc = rte_rdtsc();
@@ -1150,20 +1469,24 @@ start_rx:
                                 */
                                rte_delay_us(lcore_idle_hint);
                        else {
-                               /* suspend until rx interrupt trigges */
+                               /* suspend until rx interrupt triggers */
                                if (intr_en) {
-                                       turn_on_intr(qconf);
+                                       turn_on_off_intr(qconf, 1);
                                        sleep_until_rx_interrupt(
                                                qconf->n_rx_queue);
+                                       turn_on_off_intr(qconf, 0);
                                        /**
                                         * start receiving packets immediately
                                         */
-                                       goto start_rx;
+                                       if (likely(!is_done()))
+                                               goto start_rx;
                                }
                        }
                        stats[lcore_id].sleep_time += lcore_idle_hint;
                }
        }
+
+       return 0;
 }
 
 static int
@@ -1190,6 +1513,11 @@ check_lcore_params(void)
                        printf("warning: lcore %hhu is on socket %d with numa "
                                                "off\n", lcore, socketid);
                }
+               if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
+                       printf("cannot enable master core %d in config for telemetry mode\n",
+                               rte_lcore_id());
+                       return -1;
+               }
        }
        return 0;
 }
@@ -1273,8 +1601,12 @@ print_usage(const char *prgname)
                "  --enable-jumbo: enable jumbo frame"
                " which max packet len is PKTLEN in decimal (64-9600)\n"
                "  --parse-ptype: parse packet type by software\n"
+               "  --legacy: use legacy interrupt-based scaling\n"
                "  --empty-poll: enable empty poll detection"
-               " follow (training_flag, high_threshold, med_threshold)\n",
+               " follow (training_flag, high_threshold, med_threshold)\n"
+               " --telemetry: enable telemetry mode, to update"
+               " empty polls, full polls, and core busyness to telemetry\n"
+               " --interrupt-only: enable interrupt-only mode\n",
                prgname);
 }
 
@@ -1303,10 +1635,7 @@ parse_portmask(const char *portmask)
        /* parse hexadecimal string */
        pm = strtoul(portmask, &end, 16);
        if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
-               return -1;
-
-       if (pm == 0)
-               return -1;
+               return 0;
 
        return pm;
 }
@@ -1417,6 +1746,10 @@ parse_ep_config(const char *q_arg)
 
 }
 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
+#define CMD_LINE_OPT_LEGACY "legacy"
+#define CMD_LINE_OPT_EMPTY_POLL "empty-poll"
+#define CMD_LINE_OPT_INTERRUPT_ONLY "interrupt-only"
+#define CMD_LINE_OPT_TELEMETRY "telemetry"
 
 /* Parse the argument given in the command line of the application */
 static int
@@ -1433,8 +1766,11 @@ parse_args(int argc, char **argv)
                {"high-perf-cores", 1, 0, 0},
                {"no-numa", 0, 0, 0},
                {"enable-jumbo", 0, 0, 0},
-               {"empty-poll", 1, 0, 0},
+               {CMD_LINE_OPT_EMPTY_POLL, 1, 0, 0},
                {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
+               {CMD_LINE_OPT_LEGACY, 0, 0, 0},
+               {CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
+               {CMD_LINE_OPT_INTERRUPT_ONLY, 0, 0, 0},
                {NULL, 0, 0, 0}
        };
 
@@ -1507,9 +1843,23 @@ parse_args(int argc, char **argv)
                        }
 
                        if (!strncmp(lgopts[option_index].name,
-                                               "empty-poll", 10)) {
-                               printf("empty-poll is enabled\n");
-                               empty_poll_on = true;
+                                       CMD_LINE_OPT_LEGACY,
+                                       sizeof(CMD_LINE_OPT_LEGACY))) {
+                               if (app_mode != APP_MODE_DEFAULT) {
+                                       printf(" legacy mode is mutually exclusive with other modes\n");
+                                       return -1;
+                               }
+                               app_mode = APP_MODE_LEGACY;
+                               printf("legacy mode is enabled\n");
+                       }
+
+                       if (!strncmp(lgopts[option_index].name,
+                                       CMD_LINE_OPT_EMPTY_POLL, 10)) {
+                               if (app_mode != APP_MODE_DEFAULT) {
+                                       printf(" empty-poll mode is mutually exclusive with other modes\n");
+                                       return -1;
+                               }
+                               app_mode = APP_MODE_EMPTY_POLL;
                                ret = parse_ep_config(optarg);
 
                                if (ret) {
@@ -1517,7 +1867,29 @@ parse_args(int argc, char **argv)
                                        print_usage(prgname);
                                        return -1;
                                }
+                               printf("empty-poll is enabled\n");
+                       }
+
+                       if (!strncmp(lgopts[option_index].name,
+                                       CMD_LINE_OPT_TELEMETRY,
+                                       sizeof(CMD_LINE_OPT_TELEMETRY))) {
+                               if (app_mode != APP_MODE_DEFAULT) {
+                                       printf(" telemetry mode is mutually exclusive with other modes\n");
+                                       return -1;
+                               }
+                               app_mode = APP_MODE_TELEMETRY;
+                               printf("telemetry mode is enabled\n");
+                       }
 
+                       if (!strncmp(lgopts[option_index].name,
+                                       CMD_LINE_OPT_INTERRUPT_ONLY,
+                                       sizeof(CMD_LINE_OPT_INTERRUPT_ONLY))) {
+                               if (app_mode != APP_MODE_DEFAULT) {
+                                       printf(" interrupt-only mode is mutually exclusive with other modes\n");
+                                       return -1;
+                               }
+                               app_mode = APP_MODE_INTERRUPT;
+                               printf("interrupt-only mode is enabled\n");
                        }
 
                        if (!strncmp(lgopts[option_index].name,
@@ -1534,7 +1906,7 @@ parse_args(int argc, char **argv)
 
                                /**
                                 * if no max-pkt-len set, use the default value
-                                * ETHER_MAX_LEN
+                                * RTE_ETHER_MAX_LEN
                                 */
                                if (0 == getopt_long(argc, argvopt, "",
                                                &lenopts, &option_index)) {
@@ -1579,8 +1951,8 @@ parse_args(int argc, char **argv)
 static void
 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
 {
-       char buf[ETHER_ADDR_FMT_SIZE];
-       ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
+       char buf[RTE_ETHER_ADDR_FMT_SIZE];
+       rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
        printf("%s%s", name, buf);
 }
 
@@ -1630,7 +2002,7 @@ setup_hash(int socketid)
 
 
        /* populate the ipv4 hash */
-       for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
+       for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
                ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
                                (void *) &ipv4_l3fwd_route_array[i].key);
                if (ret < 0) {
@@ -1643,7 +2015,7 @@ setup_hash(int socketid)
        }
 
        /* populate the ipv6 hash */
-       for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
+       for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
                ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
                                (void *) &ipv6_l3fwd_route_array[i].key);
                if (ret < 0) {
@@ -1680,7 +2052,7 @@ setup_lpm(int socketid)
                                " on socket %d\n", socketid);
 
        /* populate the LPM table */
-       for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
+       for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
                ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
                        ipv4_l3fwd_route_array[i].ip,
                        ipv4_l3fwd_route_array[i].depth,
@@ -1761,6 +2133,7 @@ check_all_ports_link_status(uint32_t port_mask)
        uint8_t count, all_ports_up, print_flag = 0;
        uint16_t portid;
        struct rte_eth_link link;
+       int ret;
 
        printf("\nChecking link status");
        fflush(stdout);
@@ -1770,7 +2143,14 @@ check_all_ports_link_status(uint32_t port_mask)
                        if ((port_mask & (1 << portid)) == 0)
                                continue;
                        memset(&link, 0, sizeof(link));
-                       rte_eth_link_get_nowait(portid, &link);
+                       ret = rte_eth_link_get_nowait(portid, &link);
+                       if (ret < 0) {
+                               all_ports_up = 0;
+                               if (print_flag == 1)
+                                       printf("Port %u link get failed: %s\n",
+                                               portid, rte_strerror(-ret));
+                               continue;
+                       }
                        /* print link status if flag set */
                        if (print_flag == 1) {
                                if (link.link_status)
@@ -1778,7 +2158,7 @@ check_all_ports_link_status(uint32_t port_mask)
                                                "Mbps - %s\n", (uint8_t)portid,
                                                (unsigned)link.link_speed,
                                (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
-                                       ("full-duplex") : ("half-duplex\n"));
+                                       ("full-duplex") : ("half-duplex"));
                                else
                                        printf("Port %d Link Down\n",
                                                (uint8_t)portid);
@@ -1855,19 +2235,124 @@ static int check_ptype(uint16_t portid)
 static int
 init_power_library(void)
 {
-       int ret = 0, lcore_id;
-       for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
-               if (rte_lcore_is_enabled(lcore_id)) {
-                       /* init power management library */
-                       ret = rte_power_init(lcore_id);
-                       if (ret)
-                               RTE_LOG(ERR, POWER,
+       enum power_management_env env;
+       unsigned int lcore_id;
+       int ret = 0;
+
+       RTE_LCORE_FOREACH(lcore_id) {
+               /* init power management library */
+               ret = rte_power_init(lcore_id);
+               if (ret) {
+                       RTE_LOG(ERR, POWER,
                                "Library initialization failed on core %u\n",
                                lcore_id);
+                       return ret;
+               }
+               /* we're not supporting the VM channel mode */
+               env = rte_power_get_env();
+               if (env != PM_ENV_ACPI_CPUFREQ &&
+                               env != PM_ENV_PSTATE_CPUFREQ) {
+                       RTE_LOG(ERR, POWER,
+                               "Only ACPI and PSTATE mode are supported\n");
+                       return -1;
                }
        }
        return ret;
 }
+
+static int
+deinit_power_library(void)
+{
+       unsigned int lcore_id;
+       int ret = 0;
+
+       RTE_LCORE_FOREACH(lcore_id) {
+               /* deinit power management library */
+               ret = rte_power_exit(lcore_id);
+               if (ret) {
+                       RTE_LOG(ERR, POWER,
+                               "Library deinitialization failed on core %u\n",
+                               lcore_id);
+                       return ret;
+               }
+       }
+       return ret;
+}
+
+static void
+get_current_stat_values(uint64_t *values)
+{
+       unsigned int lcore_id = rte_lcore_id();
+       struct lcore_conf *qconf;
+       uint64_t app_eps = 0, app_fps = 0, app_br = 0;
+       uint64_t count = 0;
+
+       RTE_LCORE_FOREACH_SLAVE(lcore_id) {
+               qconf = &lcore_conf[lcore_id];
+               if (qconf->n_rx_queue == 0)
+                       continue;
+               count++;
+               rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
+               app_eps += stats[lcore_id].ep_nep[1];
+               app_fps += stats[lcore_id].fp_nfp[1];
+               app_br += stats[lcore_id].br;
+               rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
+       }
+
+       if (count > 0) {
+               values[0] = app_eps/count;
+               values[1] = app_fps/count;
+               values[2] = app_br/count;
+       } else
+               memset(values, 0, sizeof(uint64_t) * NUM_TELSTATS);
+
+}
+
+static void
+update_telemetry(__rte_unused struct rte_timer *tim,
+               __rte_unused void *arg)
+{
+       int ret;
+       uint64_t values[NUM_TELSTATS] = {0};
+
+       get_current_stat_values(values);
+       ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
+                                       values, RTE_DIM(values));
+       if (ret < 0)
+               RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
+}
+
+static int
+handle_app_stats(const char *cmd __rte_unused,
+               const char *params __rte_unused,
+               struct rte_tel_data *d)
+{
+       uint64_t values[NUM_TELSTATS] = {0};
+       uint32_t i;
+
+       rte_tel_data_start_dict(d);
+       get_current_stat_values(values);
+       for (i = 0; i < NUM_TELSTATS; i++)
+               rte_tel_data_add_dict_u64(d, telstats_strings[i].name,
+                               values[i]);
+       return 0;
+}
+
+static void
+telemetry_setup_timer(void)
+{
+       int lcore_id = rte_lcore_id();
+       uint64_t hz = rte_get_timer_hz();
+       uint64_t ticks;
+
+       ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
+       rte_timer_reset_sync(&telemetry_timer,
+                       ticks,
+                       PERIODICAL,
+                       lcore_id,
+                       update_telemetry,
+                       NULL);
+}
 static void
 empty_poll_setup_timer(void)
 {
@@ -1902,7 +2387,10 @@ launch_timer(unsigned int lcore_id)
 
        RTE_LOG(INFO, POWER, "Bring up the Timer\n");
 
-       empty_poll_setup_timer();
+       if (app_mode == APP_MODE_EMPTY_POLL)
+               empty_poll_setup_timer();
+       else
+               telemetry_setup_timer();
 
        cycles_10ms = rte_get_timer_hz() / 100;
 
@@ -1921,6 +2409,42 @@ launch_timer(unsigned int lcore_id)
        return 0;
 }
 
+static int
+autodetect_mode(void)
+{
+       RTE_LOG(NOTICE, L3FWD_POWER, "Operating mode not specified, probing frequency scaling support...\n");
+
+       /*
+        * Empty poll and telemetry modes have to be specifically requested to
+        * be enabled, but we can auto-detect between interrupt mode with or
+        * without frequency scaling. Both ACPI and pstate can be used.
+        */
+       if (rte_power_check_env_supported(PM_ENV_ACPI_CPUFREQ))
+               return APP_MODE_LEGACY;
+       if (rte_power_check_env_supported(PM_ENV_PSTATE_CPUFREQ))
+               return APP_MODE_LEGACY;
+
+       RTE_LOG(NOTICE, L3FWD_POWER, "Frequency scaling not supported, selecting interrupt-only mode\n");
+
+       return APP_MODE_INTERRUPT;
+}
+
+static const char *
+mode_to_str(enum appmode mode)
+{
+       switch (mode) {
+       case APP_MODE_LEGACY:
+               return "legacy";
+       case APP_MODE_EMPTY_POLL:
+               return "empty poll";
+       case APP_MODE_TELEMETRY:
+               return "telemetry";
+       case APP_MODE_INTERRUPT:
+               return "interrupt-only";
+       default:
+               return "invalid";
+       }
+}
 
 int
 main(int argc, char **argv)
@@ -1937,6 +2461,7 @@ main(int argc, char **argv)
        uint32_t dev_rxq_num, dev_txq_num;
        uint8_t nb_rx_queue, queue, socketid;
        uint16_t portid;
+       const char *ptr_strings[NUM_TELSTATS];
 
        /* catch SIGINT and restore cpufreq governor to ondemand */
        signal(SIGINT, signal_exit_now);
@@ -1956,8 +2481,16 @@ main(int argc, char **argv)
        if (ret < 0)
                rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
 
-       if (init_power_library())
-               RTE_LOG(ERR, L3FWD_POWER, "init_power_library failed\n");
+       if (app_mode == APP_MODE_DEFAULT)
+               app_mode = autodetect_mode();
+
+       RTE_LOG(INFO, L3FWD_POWER, "Selected operation mode: %s\n",
+                       mode_to_str(app_mode));
+
+       /* only legacy and empty poll mode rely on power library */
+       if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
+                       init_power_library())
+               rte_exit(EXIT_FAILURE, "init_power_library failed\n");
 
        if (update_lcore_params() < 0)
                rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
@@ -1979,6 +2512,9 @@ main(int argc, char **argv)
        /* initialize all ports */
        RTE_ETH_FOREACH_DEV(portid) {
                struct rte_eth_conf local_port_conf = port_conf;
+               /* not all app modes need interrupts */
+               bool need_intr = app_mode == APP_MODE_LEGACY ||
+                               app_mode == APP_MODE_INTERRUPT;
 
                /* skip ports that are not enabled */
                if ((enabled_port_mask & (1 << portid)) == 0) {
@@ -1990,7 +2526,12 @@ main(int argc, char **argv)
                printf("Initializing port %d ... ", portid );
                fflush(stdout);
 
-               rte_eth_dev_info_get(portid, &dev_info);
+               ret = rte_eth_dev_info_get(portid, &dev_info);
+               if (ret != 0)
+                       rte_exit(EXIT_FAILURE,
+                               "Error during getting device (port %u) info: %s\n",
+                               portid, strerror(-ret));
+
                dev_rxq_num = dev_info.max_rx_queues;
                dev_txq_num = dev_info.max_tx_queues;
 
@@ -2007,8 +2548,17 @@ main(int argc, char **argv)
                        nb_rx_queue, (unsigned)n_tx_queue );
                /* If number of Rx queue is 0, no need to enable Rx interrupt */
                if (nb_rx_queue == 0)
-                       local_port_conf.intr_conf.rxq = 0;
-               rte_eth_dev_info_get(portid, &dev_info);
+                       need_intr = false;
+
+               if (need_intr)
+                       local_port_conf.intr_conf.rxq = 1;
+
+               ret = rte_eth_dev_info_get(portid, &dev_info);
+               if (ret != 0)
+                       rte_exit(EXIT_FAILURE,
+                               "Error during getting device (port %u) info: %s\n",
+                               portid, strerror(-ret));
+
                if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
                        local_port_conf.txmode.offloads |=
                                DEV_TX_OFFLOAD_MBUF_FAST_FREE;
@@ -2037,7 +2587,12 @@ main(int argc, char **argv)
                                 "Cannot adjust number of descriptors: err=%d, port=%d\n",
                                 ret, portid);
 
-               rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
+               ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
+               if (ret < 0)
+                       rte_exit(EXIT_FAILURE,
+                                "Cannot get MAC address: err=%d, port=%d\n",
+                                ret, portid);
+
                print_ethaddr(" Address:", &ports_eth_addr[portid]);
                printf(", ");
 
@@ -2103,7 +2658,7 @@ main(int argc, char **argv)
                if (rte_lcore_is_enabled(lcore_id) == 0)
                        continue;
 
-               if (empty_poll_on == false) {
+               if (app_mode == APP_MODE_LEGACY) {
                        /* init timer structures for each enabled lcore */
                        rte_timer_init(&power_timers[lcore_id]);
                        hz = rte_get_timer_hz();
@@ -2118,13 +2673,9 @@ main(int argc, char **argv)
                /* init RX queues */
                for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
                        struct rte_eth_rxconf rxq_conf;
-                       struct rte_eth_dev *dev;
-                       struct rte_eth_conf *conf;
 
                        portid = qconf->rx_queue_list[queue].port_id;
                        queueid = qconf->rx_queue_list[queue].queue_id;
-                       dev = &rte_eth_devices[portid];
-                       conf = &dev->data->dev_conf;
 
                        if (numa_on)
                                socketid = \
@@ -2135,9 +2686,14 @@ main(int argc, char **argv)
                        printf("rxq=%d,%d,%d ", portid, queueid, socketid);
                        fflush(stdout);
 
-                       rte_eth_dev_info_get(portid, &dev_info);
+                       ret = rte_eth_dev_info_get(portid, &dev_info);
+                       if (ret != 0)
+                               rte_exit(EXIT_FAILURE,
+                                       "Error during getting device (port %u) info: %s\n",
+                                       portid, strerror(-ret));
+
                        rxq_conf = dev_info.default_rxconf;
-                       rxq_conf.offloads = conf->rxmode.offloads;
+                       rxq_conf.offloads = port_conf.rxmode.offloads;
                        ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
                                socketid, &rxq_conf,
                                pktmbuf_pool[socketid]);
@@ -2174,15 +2730,20 @@ main(int argc, char **argv)
                 * to itself through 2 cross-connected  ports of the
                 * target machine.
                 */
-               if (promiscuous_on)
-                       rte_eth_promiscuous_enable(portid);
+               if (promiscuous_on) {
+                       ret = rte_eth_promiscuous_enable(portid);
+                       if (ret != 0)
+                               rte_exit(EXIT_FAILURE,
+                                       "rte_eth_promiscuous_enable: err=%s, port=%u\n",
+                                       rte_strerror(-ret), portid);
+               }
                /* initialize spinlock for each port */
                rte_spinlock_init(&(locks[portid]));
        }
 
        check_all_ports_link_status(enabled_port_mask);
 
-       if (empty_poll_on == true) {
+       if (app_mode == APP_MODE_EMPTY_POLL) {
 
                if (empty_poll_train) {
                        policy.state = TRAINING;
@@ -2201,15 +2762,41 @@ main(int argc, char **argv)
 
 
        /* launch per-lcore init on every lcore */
-       if (empty_poll_on == false) {
-               rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
-       } else {
+       if (app_mode == APP_MODE_LEGACY) {
+               rte_eal_mp_remote_launch(main_legacy_loop, NULL, CALL_MASTER);
+       } else if (app_mode == APP_MODE_EMPTY_POLL) {
                empty_poll_stop = false;
                rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
                                SKIP_MASTER);
+       } else if (app_mode == APP_MODE_TELEMETRY) {
+               unsigned int i;
+
+               /* Init metrics library */
+               rte_metrics_init(rte_socket_id());
+               /** Register stats with metrics library */
+               for (i = 0; i < NUM_TELSTATS; i++)
+                       ptr_strings[i] = telstats_strings[i].name;
+
+               ret = rte_metrics_reg_names(ptr_strings, NUM_TELSTATS);
+               if (ret >= 0)
+                       telstats_index = ret;
+               else
+                       rte_exit(EXIT_FAILURE, "failed to register metrics names");
+
+               RTE_LCORE_FOREACH_SLAVE(lcore_id) {
+                       rte_spinlock_init(&stats[lcore_id].telemetry_lock);
+               }
+               rte_timer_init(&telemetry_timer);
+               rte_telemetry_register_cmd("/l3fwd-power/stats",
+                               handle_app_stats,
+                               "Returns global power stats. Parameters: None");
+               rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
+                                               SKIP_MASTER);
+       } else if (app_mode == APP_MODE_INTERRUPT) {
+               rte_eal_mp_remote_launch(main_intr_loop, NULL, CALL_MASTER);
        }
 
-       if (empty_poll_on == true)
+       if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
                launch_timer(rte_lcore_id());
 
        RTE_LCORE_FOREACH_SLAVE(lcore_id) {
@@ -2217,8 +2804,24 @@ main(int argc, char **argv)
                        return -1;
        }
 
-       if (empty_poll_on)
+       RTE_ETH_FOREACH_DEV(portid)
+       {
+               if ((enabled_port_mask & (1 << portid)) == 0)
+                       continue;
+
+               rte_eth_dev_stop(portid);
+               rte_eth_dev_close(portid);
+       }
+
+       if (app_mode == APP_MODE_EMPTY_POLL)
                rte_power_empty_poll_stat_free();
 
+       if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
+                       deinit_power_library())
+               rte_exit(EXIT_FAILURE, "deinit_power_library failed\n");
+
+       if (rte_eal_cleanup() < 0)
+               RTE_LOG(ERR, L3FWD_POWER, "EAL cleanup failed\n");
+
        return 0;
 }