examples/l3fwd: remove checks for SSE4
[dpdk.git] / examples / l3fwd-power / main.c
index 2f205ea..52eb835 100644 (file)
@@ -1,7 +1,7 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
  *   All rights reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
@@ -47,6 +47,7 @@
 #include <rte_common.h>
 #include <rte_byteorder.h>
 #include <rte_log.h>
+#include <rte_malloc.h>
 #include <rte_memory.h>
 #include <rte_memcpy.h>
 #include <rte_memzone.h>
@@ -65,7 +66,6 @@
 #include <rte_debug.h>
 #include <rte_ether.h>
 #include <rte_ethdev.h>
-#include <rte_ring.h>
 #include <rte_mempool.h>
 #include <rte_mbuf.h>
 #include <rte_ip.h>
 /*
  * Configurable number of RX/TX ring descriptors
  */
-#define RTE_TEST_RX_DESC_DEFAULT 128
+#define RTE_TEST_RX_DESC_DEFAULT 512
 #define RTE_TEST_TX_DESC_DEFAULT 512
 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
@@ -164,6 +164,8 @@ static uint32_t enabled_port_mask = 0;
 static int promiscuous_on = 0;
 /* NUMA is enabled by default. */
 static int numa_on = 1;
+static int parse_ptype; /**< Parse packet type using rx callback, and */
+                       /**< disabled by default */
 
 enum freq_scale_hint_t
 {
@@ -173,11 +175,6 @@ enum freq_scale_hint_t
        FREQ_HIGHEST  =       2
 };
 
-struct mbuf_table {
-       uint16_t len;
-       struct rte_mbuf *m_table[MAX_PKT_BURST];
-};
-
 struct lcore_rx_queue {
        uint8_t port_id;
        uint8_t queue_id;
@@ -226,7 +223,7 @@ static struct rte_eth_conf port_conf = {
                .hw_ip_checksum = 1, /**< IP checksum offload enabled */
                .hw_vlan_filter = 0, /**< VLAN filtering disabled */
                .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
-               .hw_strip_crc   = 0, /**< CRC stripped by hardware */
+               .hw_strip_crc   = 1, /**< CRC stripped by hardware */
        },
        .rx_adv_conf = {
                .rss_conf = {
@@ -239,9 +236,7 @@ static struct rte_eth_conf port_conf = {
        },
        .intr_conf = {
                .lsc = 1,
-#ifdef RTE_NEXT_ABI
                .rxq = 1,
-#endif
        },
 };
 
@@ -250,7 +245,7 @@ static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 
 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 
-#ifdef RTE_MACHINE_CPUFLAG_SSE4_2
+#ifdef RTE_ARCH_X86
 #include <rte_hash_crc.h>
 #define DEFAULT_HASH_FUNC       rte_hash_crc
 #else
@@ -349,8 +344,10 @@ static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 struct lcore_conf {
        uint16_t n_rx_queue;
        struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
+       uint16_t n_tx_port;
+       uint16_t tx_port_id[RTE_MAX_ETHPORTS];
        uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
-       struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
+       struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
        lookup_struct_t * ipv4_lookup_struct;
        lookup_struct_t * ipv6_lookup_struct;
 } __rte_cache_aligned;
@@ -382,6 +379,7 @@ static void
 signal_exit_now(int sigtype)
 {
        unsigned lcore_id;
+       unsigned int portid, nb_ports;
        int ret;
 
        if (sigtype == SIGINT) {
@@ -396,6 +394,15 @@ signal_exit_now(int sigtype)
                                        "library de-initialization failed on "
                                                        "core%u\n", lcore_id);
                }
+
+               nb_ports = rte_eth_dev_count();
+               for (portid = 0; portid < nb_ports; portid++) {
+                       if ((enabled_port_mask & (1 << portid)) == 0)
+                               continue;
+
+                       rte_eth_dev_stop(portid);
+                       rte_eth_dev_close(portid);
+               }
        }
 
        rte_exit(EXIT_SUCCESS, "User forced exit\n");
@@ -444,49 +451,19 @@ power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
        stats[lcore_id].sleep_time = 0;
 }
 
-/* Send burst of packets on an output interface */
-static inline int
-send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
-{
-       struct rte_mbuf **m_table;
-       int ret;
-       uint16_t queueid;
-
-       queueid = qconf->tx_queue_id[port];
-       m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
-
-       ret = rte_eth_tx_burst(port, queueid, m_table, n);
-       if (unlikely(ret < n)) {
-               do {
-                       rte_pktmbuf_free(m_table[ret]);
-               } while (++ret < n);
-       }
-
-       return 0;
-}
-
 /* Enqueue a single packet, and send burst if queue is filled */
 static inline int
 send_single_packet(struct rte_mbuf *m, uint8_t port)
 {
        uint32_t lcore_id;
-       uint16_t len;
        struct lcore_conf *qconf;
 
        lcore_id = rte_lcore_id();
-
        qconf = &lcore_conf[lcore_id];
-       len = qconf->tx_mbufs[port].len;
-       qconf->tx_mbufs[port].m_table[len] = m;
-       len++;
-
-       /* enough pkts to be sent */
-       if (unlikely(len == MAX_PKT_BURST)) {
-               send_burst(qconf, MAX_PKT_BURST, port);
-               len = 0;
-       }
 
-       qconf->tx_mbufs[port].len = len;
+       rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
+                       qconf->tx_buffer[port], m);
+
        return 0;
 }
 
@@ -633,7 +610,7 @@ static inline uint8_t
 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
                lookup_struct_t *ipv4_l3fwd_lookup_struct)
 {
-       uint8_t next_hop;
+       uint32_t next_hop;
 
        return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
                        rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
@@ -641,6 +618,48 @@ get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 }
 #endif
 
+static inline void
+parse_ptype_one(struct rte_mbuf *m)
+{
+       struct ether_hdr *eth_hdr;
+       uint32_t packet_type = RTE_PTYPE_UNKNOWN;
+       uint16_t ether_type;
+
+       eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
+       ether_type = eth_hdr->ether_type;
+       if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
+               packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
+       else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
+               packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
+
+       m->packet_type = packet_type;
+}
+
+static uint16_t
+cb_parse_ptype(uint8_t port __rte_unused, uint16_t queue __rte_unused,
+              struct rte_mbuf *pkts[], uint16_t nb_pkts,
+              uint16_t max_pkts __rte_unused,
+              void *user_param __rte_unused)
+{
+       unsigned int i;
+
+       for (i = 0; i < nb_pkts; ++i)
+               parse_ptype_one(pkts[i]);
+
+       return nb_pkts;
+}
+
+static int
+add_cb_parse_ptype(uint8_t portid, uint16_t queueid)
+{
+       printf("Port %d: softly parse packet type info\n", portid);
+       if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
+               return 0;
+
+       printf("Failed to add rx callback: port=%d\n", portid);
+       return -1;
+}
+
 static inline void
 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
                                struct lcore_conf *qconf)
@@ -652,11 +671,7 @@ l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 
        eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 
-#ifdef RTE_NEXT_ABI
        if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
-#else
-       if (m->ol_flags & PKT_RX_IPV4_HDR) {
-#endif
                /* Handle IPv4 headers.*/
                ipv4_hdr =
                        rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
@@ -691,12 +706,7 @@ l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
                ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 
                send_single_packet(m, dst_port);
-#ifdef RTE_NEXT_ABI
        } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
-#else
-       }
-       else {
-#endif
                /* Handle IPv6 headers.*/
 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
                struct ipv6_hdr *ipv6_hdr;
@@ -725,7 +735,8 @@ l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
                /* We don't currently handle IPv6 packets in LPM mode. */
                rte_pktmbuf_free(m);
 #endif
-       }
+       } else
+               rte_pktmbuf_free(m);
 
 }
 
@@ -809,6 +820,7 @@ sleep_until_rx_interrupt(int num)
                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",
@@ -915,20 +927,12 @@ main_loop(__attribute__((unused)) void *dummy)
                 */
                diff_tsc = cur_tsc - prev_tsc;
                if (unlikely(diff_tsc > drain_tsc)) {
-
-                       /*
-                        * This could be optimized (use queueid instead of
-                        * portid), but it is not called so often
-                        */
-                       for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
-                               if (qconf->tx_mbufs[portid].len == 0)
-                                       continue;
-                               send_burst(&lcore_conf[lcore_id],
-                                       qconf->tx_mbufs[portid].len,
-                                       portid);
-                               qconf->tx_mbufs[portid].len = 0;
+                       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;
                }
 
@@ -1158,7 +1162,8 @@ print_usage(const char *prgname)
                "  --config (port,queue,lcore): rx queues configuration\n"
                "  --no-numa: optional, disable numa awareness\n"
                "  --enable-jumbo: enable jumbo frame"
-               " which max packet len is PKTLEN in decimal (64-9600)\n",
+               " which max packet len is PKTLEN in decimal (64-9600)\n"
+               "  --parse-ptype: parse packet type by software\n",
                prgname);
 }
 
@@ -1252,6 +1257,8 @@ parse_config(const char *q_arg)
        return 0;
 }
 
+#define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
+
 /* Parse the argument given in the command line of the application */
 static int
 parse_args(int argc, char **argv)
@@ -1264,6 +1271,7 @@ parse_args(int argc, char **argv)
                {"config", 1, 0, 0},
                {"no-numa", 0, 0, 0},
                {"enable-jumbo", 0, 0, 0},
+               {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
                {NULL, 0, 0, 0}
        };
 
@@ -1334,6 +1342,13 @@ parse_args(int argc, char **argv)
                                (unsigned int)port_conf.rxmode.max_rx_pkt_len);
                        }
 
+                       if (!strncmp(lgopts[option_index].name,
+                                    CMD_LINE_OPT_PARSE_PTYPE,
+                                    sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
+                               printf("soft parse-ptype is enabled\n");
+                               parse_ptype = 1;
+                       }
+
                        break;
 
                default:
@@ -1346,7 +1361,7 @@ parse_args(int argc, char **argv)
                argv[optind-1] = prgname;
 
        ret = optind-1;
-       optind = 0; /* reset getopt lib */
+       optind = 1; /* reset getopt lib */
        return ret;
 }
 
@@ -1383,7 +1398,7 @@ setup_hash(int socketid)
        char s[64];
 
        /* create ipv4 hash */
-       rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
+       snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
        ipv4_l3fwd_hash_params.name = s;
        ipv4_l3fwd_hash_params.socket_id = socketid;
        ipv4_l3fwd_lookup_struct[socketid] =
@@ -1393,7 +1408,7 @@ setup_hash(int socketid)
                                "socket %d\n", socketid);
 
        /* create ipv6 hash */
-       rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
+       snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
        ipv6_l3fwd_hash_params.name = s;
        ipv6_l3fwd_hash_params.socket_id = socketid;
        ipv6_l3fwd_lookup_struct[socketid] =
@@ -1440,9 +1455,15 @@ setup_lpm(int socketid)
        char s[64];
 
        /* create the LPM table */
+       struct rte_lpm_config lpm_ipv4_config;
+
+       lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
+       lpm_ipv4_config.number_tbl8s = 256;
+       lpm_ipv4_config.flags = 0;
+
        snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
-       ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
-                               IPV4_L3FWD_LPM_MAX_RULES, 0);
+       ipv4_l3fwd_lookup_struct[socketid] =
+                       rte_lpm_create(s, socketid, &lpm_ipv4_config);
        if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
                rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
                                " on socket %d\n", socketid);
@@ -1552,7 +1573,7 @@ check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
                                continue;
                        }
                        /* clear all_ports_up flag if any link down */
-                       if (link.link_status == 0) {
+                       if (link.link_status == ETH_LINK_DOWN) {
                                all_ports_up = 0;
                                break;
                        }
@@ -1575,6 +1596,50 @@ check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
        }
 }
 
+static int check_ptype(uint8_t portid)
+{
+       int i, ret;
+       int ptype_l3_ipv4 = 0;
+#if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
+       int ptype_l3_ipv6 = 0;
+#endif
+       uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
+
+       ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
+       if (ret <= 0)
+               return 0;
+
+       uint32_t ptypes[ret];
+
+       ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
+       for (i = 0; i < ret; ++i) {
+               if (ptypes[i] & RTE_PTYPE_L3_IPV4)
+                       ptype_l3_ipv4 = 1;
+#if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
+               if (ptypes[i] & RTE_PTYPE_L3_IPV6)
+                       ptype_l3_ipv6 = 1;
+#endif
+       }
+
+       if (ptype_l3_ipv4 == 0)
+               printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
+
+#if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
+       if (ptype_l3_ipv6 == 0)
+               printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
+#endif
+
+#if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
+       if (ptype_l3_ipv4)
+#else /* APP_LOOKUP_EXACT_MATCH */
+       if (ptype_l3_ipv4 && ptype_l3_ipv6)
+#endif
+               return 1;
+
+       return 0;
+
+}
+
 int
 main(int argc, char **argv)
 {
@@ -1589,6 +1654,7 @@ main(int argc, char **argv)
        uint32_t n_tx_queue, nb_lcores;
        uint32_t dev_rxq_num, dev_txq_num;
        uint8_t portid, nb_rx_queue, queue, socketid;
+       uint16_t org_rxq_intr = port_conf.intr_conf.rxq;
 
        /* catch SIGINT and restore cpufreq governor to ondemand */
        signal(SIGINT, signal_exit_now);
@@ -1615,10 +1681,7 @@ main(int argc, char **argv)
        if (ret < 0)
                rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 
-
        nb_ports = rte_eth_dev_count();
-       if (nb_ports > RTE_MAX_ETHPORTS)
-               nb_ports = RTE_MAX_ETHPORTS;
 
        if (check_port_config(nb_ports) < 0)
                rte_exit(EXIT_FAILURE, "check_port_config failed\n");
@@ -1652,8 +1715,13 @@ main(int argc, char **argv)
                        n_tx_queue = dev_txq_num;
                printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
                        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)
+                       port_conf.intr_conf.rxq = 0;
                ret = rte_eth_dev_configure(portid, nb_rx_queue,
                                        (uint16_t)n_tx_queue, &port_conf);
+               /* Revert to original value */
+               port_conf.intr_conf.rxq = org_rxq_intr;
                if (ret < 0)
                        rte_exit(EXIT_FAILURE, "Cannot configure device: "
                                        "err=%d, port=%d\n", ret, portid);
@@ -1667,6 +1735,22 @@ main(int argc, char **argv)
                if (ret < 0)
                        rte_exit(EXIT_FAILURE, "init_mem failed\n");
 
+               for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
+                       if (rte_lcore_is_enabled(lcore_id) == 0)
+                               continue;
+
+                       /* Initialize TX buffers */
+                       qconf = &lcore_conf[lcore_id];
+                       qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
+                               RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
+                               rte_eth_dev_socket_id(portid));
+                       if (qconf->tx_buffer[portid] == NULL)
+                               rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
+                                               (unsigned) portid);
+
+                       rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
+               }
+
                /* init one TX queue per couple (lcore,port) */
                queueid = 0;
                for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
@@ -1699,6 +1783,9 @@ main(int argc, char **argv)
                        qconf = &lcore_conf[lcore_id];
                        qconf->tx_queue_id[portid] = queueid;
                        queueid++;
+
+                       qconf->tx_port_id[qconf->n_tx_port] = portid;
+                       qconf->n_tx_port++;
                }
                printf("\n");
        }
@@ -1744,6 +1831,14 @@ main(int argc, char **argv)
                                rte_exit(EXIT_FAILURE,
                                        "rte_eth_rx_queue_setup: err=%d, "
                                                "port=%d\n", ret, portid);
+
+                       if (parse_ptype) {
+                               if (add_cb_parse_ptype(portid, queueid) < 0)
+                                       rte_exit(EXIT_FAILURE,
+                                                "Fail to add ptype cb\n");
+                       } else if (!check_ptype(portid))
+                               rte_exit(EXIT_FAILURE,
+                                        "PMD can not provide needed ptypes\n");
                }
        }