examples/performance-thread: add packet type parsing
authorTomasz Kulasek <tomaszx.kulasek@intel.com>
Fri, 16 Dec 2016 15:15:16 +0000 (16:15 +0100)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Tue, 17 Jan 2017 17:40:17 +0000 (18:40 +0100)
Last changes in Niantic and Fortville NIC drivers causes that
vector Rx path is chosen by default in l3fwd-thread application.
This path doesn't support propagation of hw packet type recognition
to the packet_type field in mbuf, and packets cannot be classified
properly.

The approach to solve this problem is similar to the commit:
71a7e2424e07 ("examples/l3fwd: fix using packet type blindly").

To use sw packet analyzer, new command line option "--parse-ptype" is
introduced.

Signed-off-by: Tomasz Kulasek <tomaszx.kulasek@intel.com>
doc/guides/sample_app_ug/performance_thread.rst
examples/performance-thread/l3fwd-thread/main.c

index d7d9b08..51ef015 100644 (file)
@@ -107,6 +107,7 @@ The application has a number of command line options::
         --tx(lcore,thread)[,(lcore,thread)]
         [--enable-jumbo] [--max-pkt-len PKTLEN]]  [--no-numa]
         [--hash-entry-num] [--ipv6] [--no-lthreads] [--stat-lcore lcore]
+        [--parse-ptype]
 
 Where:
 
@@ -142,6 +143,9 @@ Where:
 * ``--stat-lcore``: optional, run CPU load stats collector on the specified
   lcore.
 
+* ``--parse-ptype:`` optional, set to use software to analyze packet type.
+  Without this option, hardware will check the packet type.
+
 The parameters of the ``--rx`` and ``--tx`` options are:
 
 * ``--rx`` parameters
index fdc90b2..53083df 100644 (file)
@@ -1,7 +1,7 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright(c) 2010-2015 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
 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
 #endif
 
+static int
+check_ptype(int portid)
+{
+       int i, ret;
+       int ipv4 = 0, ipv6 = 0;
+
+       ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
+                       0);
+       if (ret <= 0)
+               return 0;
+
+       uint32_t ptypes[ret];
+
+       ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
+                       ptypes, ret);
+       for (i = 0; i < ret; ++i) {
+               if (ptypes[i] & RTE_PTYPE_L3_IPV4)
+                       ipv4 = 1;
+               if (ptypes[i] & RTE_PTYPE_L3_IPV6)
+                       ipv6 = 1;
+       }
+
+       if (ipv4 && ipv6)
+               return 1;
+
+       return 0;
+}
+
+static inline void
+parse_ptype(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(__rte_unused uint8_t port, __rte_unused uint16_t queue,
+               struct rte_mbuf *pkts[], uint16_t nb_pkts,
+               __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
+{
+       unsigned int i;
+
+       for (i = 0; i < nb_pkts; i++)
+               parse_ptype(pkts[i]);
+
+       return nb_pkts;
+}
+
 /*
  *  When set to zero, simple forwaring path is eanbled.
  *  When set to one, optimized forwarding path is enabled.
@@ -170,8 +228,9 @@ static __m128i val_eth[RTE_MAX_ETHPORTS];
 
 /* mask of enabled ports */
 static uint32_t enabled_port_mask;
-static int promiscuous_on; /**< $et in promiscuous mode off by default. */
+static int promiscuous_on; /**< Set in promiscuous mode off by default. */
 static int numa_on = 1;    /**< NUMA is enabled by default. */
+static int parse_ptype_on;
 
 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 static int ipv6;           /**< ipv6 is false by default. */
@@ -2610,6 +2669,7 @@ print_usage(const char *prgname)
                "  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
                "  [--tx (lcore,thread)[,(lcore,thread]]"
                "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
+               "  [--parse-ptype]\n\n"
                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
                "  -P : enable promiscuous mode\n"
                "  --rx (port,queue,lcore,thread): rx queues configuration\n"
@@ -2621,7 +2681,8 @@ print_usage(const char *prgname)
                "  --enable-jumbo: enable jumbo frame"
                " which max packet len is PKTLEN in decimal (64-9600)\n"
                "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
-               "  --no-lthreads: turn off lthread model\n",
+               "  --no-lthreads: turn off lthread model\n"
+               "  --parse-ptype: set to use software to analyze packet type\n\n",
                prgname);
 }
 
@@ -2840,6 +2901,7 @@ parse_eth_dest(const char *optarg)
 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
+#define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
 
 /* Parse the argument given in the command line of the application */
 static int
@@ -2859,6 +2921,7 @@ parse_args(int argc, char **argv)
                {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
                {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
                {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
+               {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
                {NULL, 0, 0, 0}
        };
 
@@ -2935,6 +2998,12 @@ parse_args(int argc, char **argv)
                                lthreads_on = 0;
                        }
 
+                       if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
+                                       sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
+                               printf("software packet type parsing enabled\n");
+                               parse_ptype_on = 1;
+                       }
+
                        if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
                                sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
                                struct option lenopts = {"max-pkt-len", required_argument, 0,
@@ -3623,6 +3692,31 @@ main(int argc, char **argv)
                        rte_eth_promiscuous_enable(portid);
        }
 
+       for (i = 0; i < n_rx_thread; i++) {
+               lcore_id = rx_thread[i].conf.lcore_id;
+               if (rte_lcore_is_enabled(lcore_id) == 0)
+                       continue;
+
+               /* check if hw packet type is supported */
+               for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
+                       portid = rx_thread[i].rx_queue_list[queue].port_id;
+                       queueid = rx_thread[i].rx_queue_list[queue].queue_id;
+
+                       if (parse_ptype_on) {
+                               if (!rte_eth_add_rx_callback(portid, queueid,
+                                               cb_parse_ptype, NULL))
+                                       rte_exit(EXIT_FAILURE,
+                                               "Failed to add rx callback: "
+                                               "port=%d\n", portid);
+                       } else if (!check_ptype(portid))
+                               rte_exit(EXIT_FAILURE,
+                                       "Port %d cannot parse packet type.\n\n"
+                                       "Please add --parse-ptype to use sw "
+                                       "packet type analyzer.\n\n",
+                                       portid);
+               }
+       }
+
        check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
 
        if (lthreads_on) {