1.. SPDX-License-Identifier: BSD-3-Clause 2 Copyright(c) 2015 Intel Corporation. 3 4RX/TX Callbacks Sample Application 5================================== 6 7The RX/TX Callbacks sample application is a packet forwarding application that 8demonstrates the use of user defined callbacks on received and transmitted 9packets. The application performs a simple latency check, using callbacks, to 10determine the time packets spend within the application. 11 12In the sample application a user defined callback is applied to all received 13packets to add a timestamp. A separate callback is applied to all packets 14prior to transmission to calculate the elapsed time, in CPU cycles. 15 16If hardware timestamping is supported by the NIC, the sample application will 17also display the average latency since the packet was timestamped in hardware, 18on top of the latency since the packet was received and processed by the RX 19callback. 20 21Compiling the Application 22------------------------- 23 24To compile the sample application see :doc:`compiling`. 25 26The application is located in the ``rxtx_callbacks`` sub-directory. 27 28 29Running the Application 30----------------------- 31 32To run the example in a ``linux`` environment: 33 34.. code-block:: console 35 36 ./<build_dir>/examples/dpdk-rxtx_callbacks -l 1 -n 4 -- [-t] 37 38Use -t to enable hardware timestamping. If not supported by the NIC, an error 39will be displayed. 40 41Refer to *DPDK Getting Started Guide* for general information on running 42applications and the Environment Abstraction Layer (EAL) options. 43 44 45 46Explanation 47----------- 48 49The ``rxtx_callbacks`` application is mainly a simple forwarding application 50based on the :doc:`skeleton`. See that section of the documentation for more 51details of the forwarding part of the application. 52 53The sections below explain the additional RX/TX callback code. 54 55 56The Main Function 57~~~~~~~~~~~~~~~~~ 58 59The ``main()`` function performs the application initialization and calls the 60execution threads for each lcore. This function is effectively identical to 61the ``main()`` function explained in :doc:`skeleton`. 62 63The ``lcore_main()`` function is also identical. 64 65The main difference is in the user defined ``port_init()`` function where the 66callbacks are added. This is explained in the next section: 67 68 69The Port Initialization Function 70~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 71 72The main functional part of the port initialization is shown below with 73comments: 74 75.. code-block:: c 76 77 static inline int 78 port_init(uint16_t port, struct rte_mempool *mbuf_pool) 79 { 80 struct rte_eth_conf port_conf = port_conf_default; 81 const uint16_t rx_rings = 1, tx_rings = 1; 82 struct rte_ether_addr addr; 83 int retval; 84 uint16_t q; 85 86 /* Configure the Ethernet device. */ 87 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf); 88 if (retval != 0) 89 return retval; 90 91 /* Allocate and set up 1 RX queue per Ethernet port. */ 92 for (q = 0; q < rx_rings; q++) { 93 retval = rte_eth_rx_queue_setup(port, q, RX_RING_SIZE, 94 rte_eth_dev_socket_id(port), NULL, mbuf_pool); 95 if (retval < 0) 96 return retval; 97 } 98 99 /* Allocate and set up 1 TX queue per Ethernet port. */ 100 for (q = 0; q < tx_rings; q++) { 101 retval = rte_eth_tx_queue_setup(port, q, TX_RING_SIZE, 102 rte_eth_dev_socket_id(port), NULL); 103 if (retval < 0) 104 return retval; 105 } 106 107 /* Start the Ethernet port. */ 108 retval = rte_eth_dev_start(port); 109 if (retval < 0) 110 return retval; 111 112 /* Enable RX in promiscuous mode for the Ethernet device. */ 113 retval = rte_eth_promiscuous_enable(port); 114 if (retval != 0) 115 return retval; 116 117 /* Add the callbacks for RX and TX.*/ 118 rte_eth_add_rx_callback(port, 0, add_timestamps, NULL); 119 rte_eth_add_tx_callback(port, 0, calc_latency, NULL); 120 121 return 0; 122 } 123 124 125The RX and TX callbacks are added to the ports/queues as function pointers: 126 127.. code-block:: c 128 129 rte_eth_add_rx_callback(port, 0, add_timestamps, NULL); 130 rte_eth_add_tx_callback(port, 0, calc_latency, NULL); 131 132More than one callback can be added and additional information can be passed 133to callback function pointers as a ``void*``. In the examples above ``NULL`` 134is used. 135 136The ``add_timestamps()`` and ``calc_latency()`` functions are explained below. 137 138 139The add_timestamps() Callback 140~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 141 142The ``add_timestamps()`` callback is added to the RX port and is applied to 143all packets received: 144 145.. code-block:: c 146 147 static uint16_t 148 add_timestamps(uint16_t port __rte_unused, uint16_t qidx __rte_unused, 149 struct rte_mbuf **pkts, uint16_t nb_pkts, void *_ __rte_unused) 150 { 151 unsigned i; 152 uint64_t now = rte_rdtsc(); 153 154 for (i = 0; i < nb_pkts; i++) 155 *tsc_field(pkts[i]) = now; 156 157 return nb_pkts; 158 } 159 160The DPDK function ``rte_rdtsc()`` is used to add a cycle count timestamp to 161each packet (see the *cycles* section of the *DPDK API Documentation* for 162details). 163 164 165The calc_latency() Callback 166~~~~~~~~~~~~~~~~~~~~~~~~~~~ 167 168The ``calc_latency()`` callback is added to the TX port and is applied to all 169packets prior to transmission: 170 171.. code-block:: c 172 173 static uint16_t 174 calc_latency(uint16_t port __rte_unused, uint16_t qidx __rte_unused, 175 struct rte_mbuf **pkts, uint16_t nb_pkts, void *_ __rte_unused) 176 { 177 uint64_t cycles = 0; 178 uint64_t now = rte_rdtsc(); 179 unsigned i; 180 181 for (i = 0; i < nb_pkts; i++) 182 cycles += now - *tsc_field(pkts[i]); 183 184 latency_numbers.total_cycles += cycles; 185 latency_numbers.total_pkts += nb_pkts; 186 187 if (latency_numbers.total_pkts > (100 * 1000 * 1000ULL)) { 188 printf("Latency = %"PRIu64" cycles\n", 189 latency_numbers.total_cycles / latency_numbers.total_pkts); 190 191 latency_numbers.total_cycles = latency_numbers.total_pkts = 0; 192 } 193 194 return nb_pkts; 195 } 196 197The ``calc_latency()`` function accumulates the total number of packets and 198the total number of cycles used. Once more than 100 million packets have been 199transmitted the average cycle count per packet is printed out and the counters 200are reset. 201