1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2010-2015 Intel Corporation.
4 L2 Forwarding Sample Application (in Real and Virtualized Environments) with core load statistics.
5 ==================================================================================================
7 The L2 Forwarding sample application is a simple example of packet processing using
8 the Data Plane Development Kit (DPDK) which
9 also takes advantage of Single Root I/O Virtualization (SR-IOV) features in a virtualized environment.
13 This application is a variation of L2 Forwarding sample application. It demonstrate possible
14 scheme of job stats library usage therefore some parts of this document is identical with original
15 L2 forwarding application.
20 The L2 Forwarding sample application, which can operate in real and virtualized environments,
21 performs L2 forwarding for each packet that is received.
22 The destination port is the adjacent port from the enabled portmask, that is,
23 if the first four ports are enabled (portmask 0xf),
24 ports 1 and 2 forward into each other, and ports 3 and 4 forward into each other.
25 Also, the MAC addresses are affected as follows:
27 * The source MAC address is replaced by the TX port MAC address
29 * The destination MAC address is replaced by 02:00:00:00:00:TX_PORT_ID
31 This application can be used to benchmark performance using a traffic-generator, as shown in the :numref:`figure_l2_fwd_benchmark_setup_jobstats`.
33 The application can also be used in a virtualized environment as shown in :numref:`figure_l2_fwd_virtenv_benchmark_setup_jobstats`.
35 The L2 Forwarding application can also be used as a starting point for developing a new application based on the DPDK.
37 .. _figure_l2_fwd_benchmark_setup_jobstats:
39 .. figure:: img/l2_fwd_benchmark_setup.*
41 Performance Benchmark Setup (Basic Environment)
43 .. _figure_l2_fwd_virtenv_benchmark_setup_jobstats:
45 .. figure:: img/l2_fwd_virtenv_benchmark_setup.*
47 Performance Benchmark Setup (Virtualized Environment)
50 Virtual Function Setup Instructions
51 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
53 This application can use the virtual function available in the system and
54 therefore can be used in a virtual machine without passing through
55 the whole Network Device into a guest machine in a virtualized scenario.
56 The virtual functions can be enabled in the host machine or the hypervisor with the respective physical function driver.
58 For example, in a Linux* host machine, it is possible to enable a virtual function using the following command:
60 .. code-block:: console
62 modprobe ixgbe max_vfs=2,2
64 This command enables two Virtual Functions on each of Physical Function of the NIC,
65 with two physical ports in the PCI configuration space.
66 It is important to note that enabled Virtual Function 0 and 2 would belong to Physical Function 0
67 and Virtual Function 1 and 3 would belong to Physical Function 1,
68 in this case enabling a total of four Virtual Functions.
70 Compiling the Application
71 -------------------------
73 To compile the sample application see :doc:`compiling`.
75 The application is located in the ``l2fwd-jobstats`` sub-directory.
77 Running the Application
78 -----------------------
80 The application requires a number of command line options:
82 .. code-block:: console
84 ./build/l2fwd-jobstats [EAL options] -- -p PORTMASK [-q NQ] [-l]
88 * p PORTMASK: A hexadecimal bitmask of the ports to configure
90 * q NQ: A number of queues (=ports) per lcore (default is 1)
92 * l: Use locale thousands separator when formatting big numbers.
94 To run the application in linux environment with 4 lcores, 16 ports, 8 RX queues per lcore and
95 thousands separator printing, issue the command:
97 .. code-block:: console
99 $ ./build/l2fwd-jobstats -l 0-3 -n 4 -- -q 8 -p ffff -l
101 Refer to the *DPDK Getting Started Guide* for general information on running applications
102 and the Environment Abstraction Layer (EAL) options.
107 The following sections provide some explanation of the code.
109 Command Line Arguments
110 ~~~~~~~~~~~~~~~~~~~~~~
112 The L2 Forwarding sample application takes specific parameters,
113 in addition to Environment Abstraction Layer (EAL) arguments
114 (see `Running the Application`_).
115 The preferred way to parse parameters is to use the getopt() function,
116 since it is part of a well-defined and portable library.
118 The parsing of arguments is done in the l2fwd_parse_args() function.
119 The method of argument parsing is not described here.
120 Refer to the *glibc getopt(3)* man page for details.
122 EAL arguments are parsed first, then application-specific arguments.
123 This is done at the beginning of the main() function:
129 ret = rte_eal_init(argc, argv);
131 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
136 /* parse application arguments (after the EAL ones) */
138 ret = l2fwd_parse_args(argc, argv);
140 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
142 Mbuf Pool Initialization
143 ~~~~~~~~~~~~~~~~~~~~~~~~
145 Once the arguments are parsed, the mbuf pool is created.
146 The mbuf pool contains a set of mbuf objects that will be used by the driver
147 and the application to store network packet data:
151 /* create the mbuf pool */
152 l2fwd_pktmbuf_pool = rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF,
153 MEMPOOL_CACHE_SIZE, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
156 if (l2fwd_pktmbuf_pool == NULL)
157 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
159 The rte_mempool is a generic structure used to handle pools of objects.
160 In this case, it is necessary to create a pool that will be used by the driver.
161 The number of allocated pkt mbufs is NB_MBUF, with a data room size of
162 RTE_MBUF_DEFAULT_BUF_SIZE each.
163 A per-lcore cache of MEMPOOL_CACHE_SIZE mbufs is kept.
164 The memory is allocated in rte_socket_id() socket,
165 but it is possible to extend this code to allocate one mbuf pool per socket.
167 The rte_pktmbuf_pool_create() function uses the default mbuf pool and mbuf
168 initializers, respectively rte_pktmbuf_pool_init() and rte_pktmbuf_init().
169 An advanced application may want to use the mempool API to create the
170 mbuf pool with more control.
172 Driver Initialization
173 ~~~~~~~~~~~~~~~~~~~~~
175 The main part of the code in the main() function relates to the initialization of the driver.
176 To fully understand this code, it is recommended to study the chapters that related to the Poll Mode Driver
177 in the *DPDK Programmer's Guide* and the *DPDK API Reference*.
181 /* reset l2fwd_dst_ports */
183 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
184 l2fwd_dst_ports[portid] = 0;
189 * Each logical core is assigned a dedicated TX queue on each port.
191 RTE_ETH_FOREACH_DEV(portid) {
192 /* skip ports that are not enabled */
193 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
196 if (nb_ports_in_mask % 2) {
197 l2fwd_dst_ports[portid] = last_port;
198 l2fwd_dst_ports[last_port] = portid;
205 rte_eth_dev_info_get((uint8_t) portid, &dev_info);
208 The next step is to configure the RX and TX queues.
209 For each port, there is only one RX queue (only one lcore is able to poll a given port).
210 The number of TX queues depends on the number of available lcores.
211 The rte_eth_dev_configure() function is used to configure the number of queues for a port:
215 ret = rte_eth_dev_configure((uint8_t)portid, 1, 1, &port_conf);
217 rte_exit(EXIT_FAILURE, "Cannot configure device: "
221 RX Queue Initialization
222 ~~~~~~~~~~~~~~~~~~~~~~~
224 The application uses one lcore to poll one or several ports, depending on the -q option,
225 which specifies the number of queues per lcore.
227 For example, if the user specifies -q 4, the application is able to poll four ports with one lcore.
228 If there are 16 ports on the target (and if the portmask argument is -p ffff ),
229 the application will need four lcores to poll all the ports.
233 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
234 rte_eth_dev_socket_id(portid),
239 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
240 ret, (unsigned) portid);
242 The list of queues that must be polled for a given lcore is stored in a private structure called struct lcore_queue_conf.
246 struct lcore_queue_conf {
248 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
249 truct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
251 struct rte_timer rx_timers[MAX_RX_QUEUE_PER_LCORE];
252 struct rte_jobstats port_fwd_jobs[MAX_RX_QUEUE_PER_LCORE];
254 struct rte_timer flush_timer;
255 struct rte_jobstats flush_job;
256 struct rte_jobstats idle_job;
257 struct rte_jobstats_context jobs_context;
259 rte_atomic16_t stats_read_pending;
261 } __rte_cache_aligned;
263 Values of struct lcore_queue_conf:
265 * n_rx_port and rx_port_list[] are used in the main packet processing loop
266 (see Section `Receive, Process and Transmit Packets`_ later in this chapter).
268 * rx_timers and flush_timer are used to ensure forced TX on low packet rate.
270 * flush_job, idle_job and jobs_context are librte_jobstats objects used for managing l2fwd jobs.
272 * stats_read_pending and lock are used during job stats read phase.
274 TX Queue Initialization
275 ~~~~~~~~~~~~~~~~~~~~~~~
277 Each lcore should be able to transmit on any port. For every port, a single TX queue is initialized.
281 /* init one TX queue on each port */
284 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
285 rte_eth_dev_socket_id(portid),
288 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
289 ret, (unsigned) portid);
291 Jobs statistics initialization
292 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
293 There are several statistics objects available:
295 * Flush job statistics
299 rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
302 rte_timer_init(&qconf->flush_timer);
303 ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL,
304 lcore_id, &l2fwd_flush_job, NULL);
307 rte_exit(1, "Failed to reset flush job timer for lcore %u: %s",
308 lcore_id, rte_strerror(-ret));
311 * Statistics per RX port
315 rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
316 rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
318 rte_timer_init(&qconf->rx_timers[i]);
319 ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
320 l2fwd_fwd_job, (void *)(uintptr_t)i);
323 rte_exit(1, "Failed to reset lcore %u port %u job timer: %s",
324 lcore_id, qconf->rx_port_list[i], rte_strerror(-ret));
327 Following parameters are passed to rte_jobstats_init():
329 * 0 as minimal poll period
331 * drain_tsc as maximum poll period
333 * MAX_PKT_BURST as desired target value (RX burst size)
338 The forwarding path is reworked comparing to original L2 Forwarding application.
339 In the l2fwd_main_loop() function three loops are placed.
344 rte_spinlock_lock(&qconf->lock);
347 rte_jobstats_context_start(&qconf->jobs_context);
350 * - Read stats_read_pending flag
351 * - check if some real job need to be executed
353 rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
357 uint64_t now = rte_get_timer_cycles();
359 need_manage = qconf->flush_timer.expire < now;
360 /* Check if we was esked to give a stats. */
362 rte_atomic16_read(&qconf->stats_read_pending);
363 need_manage |= stats_read_pending;
365 for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
366 need_manage = qconf->rx_timers[i].expire < now;
368 } while (!need_manage);
369 rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
372 rte_jobstats_context_finish(&qconf->jobs_context);
373 } while (likely(stats_read_pending == 0));
375 rte_spinlock_unlock(&qconf->lock);
379 First infinite for loop is to minimize impact of stats reading. Lock is only locked/unlocked when asked.
381 Second inner while loop do the whole jobs management. When any job is ready, the use rte_timer_manage() is used to call the job handler.
382 In this place functions l2fwd_fwd_job() and l2fwd_flush_job() are called when needed.
383 Then rte_jobstats_context_finish() is called to mark loop end - no other jobs are ready to execute. By this time stats are ready to be read
384 and if stats_read_pending is set, loop breaks allowing stats to be read.
386 Third do-while loop is the idle job (idle stats counter). Its only purpose is monitoring if any job is ready or stats job read is pending
387 for this lcore. Statistics from this part of code is considered as the headroom available for additional processing.
389 Receive, Process and Transmit Packets
390 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
392 The main task of l2fwd_fwd_job() function is to read ingress packets from the RX queue of particular port and forward it.
393 This is done using the following code:
397 total_nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
400 for (j = 0; j < total_nb_rx; j++) {
402 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
403 l2fwd_simple_forward(m, portid);
406 Packets are read in a burst of size MAX_PKT_BURST.
407 Then, each mbuf in the table is processed by the l2fwd_simple_forward() function.
408 The processing is very simple: process the TX port from the RX port, then replace the source and destination MAC addresses.
410 The rte_eth_rx_burst() function writes the mbuf pointers in a local table and returns the number of available mbufs in the table.
412 After first read second try is issued.
416 if (total_nb_rx == MAX_PKT_BURST) {
417 const uint16_t nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
420 total_nb_rx += nb_rx;
421 for (j = 0; j < nb_rx; j++) {
423 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
424 l2fwd_simple_forward(m, portid);
428 This second read is important to give job stats library a feedback how many packets was processed.
432 /* Adjust period time in which we are running here. */
433 if (rte_jobstats_finish(job, total_nb_rx) != 0) {
434 rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
435 lcore_id, l2fwd_fwd_job, arg);
438 To maximize performance exactly MAX_PKT_BURST is expected (the target value) to be read for each l2fwd_fwd_job() call.
439 If total_nb_rx is smaller than target value job->period will be increased. If it is greater the period will be decreased.
443 In the following code, one line for getting the output port requires some explanation.
445 During the initialization process, a static array of destination ports (l2fwd_dst_ports[]) is filled such that for each source port,
446 a destination port is assigned that is either the next or previous enabled port from the portmask.
447 Naturally, the number of ports in the portmask must be even, otherwise, the application exits.
452 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
454 struct rte_ether_hdr *eth;
458 dst_port = l2fwd_dst_ports[portid];
460 eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
462 /* 02:00:00:00:00:xx */
464 tmp = ð->d_addr.addr_bytes[0];
466 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t) dst_port << 40);
470 rte_ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
472 l2fwd_send_packet(m, (uint8_t) dst_port);
475 Then, the packet is sent using the l2fwd_send_packet (m, dst_port) function.
476 For this test application, the processing is exactly the same for all packets arriving on the same RX port.
477 Therefore, it would have been possible to call the l2fwd_send_burst() function directly from the main loop
478 to send all the received packets on the same TX port,
479 using the burst-oriented send function, which is more efficient.
481 However, in real-life applications (such as, L3 routing),
482 packet N is not necessarily forwarded on the same port as packet N-1.
483 The application is implemented to illustrate that, so the same approach can be reused in a more complex application.
485 The l2fwd_send_packet() function stores the packet in a per-lcore and per-txport table.
486 If the table is full, the whole packets table is transmitted using the l2fwd_send_burst() function:
490 /* Send the packet on an output interface */
493 l2fwd_send_packet(struct rte_mbuf *m, uint16_t port)
495 unsigned lcore_id, len;
496 struct lcore_queue_conf *qconf;
498 lcore_id = rte_lcore_id();
499 qconf = &lcore_queue_conf[lcore_id];
500 len = qconf->tx_mbufs[port].len;
501 qconf->tx_mbufs[port].m_table[len] = m;
504 /* enough pkts to be sent */
506 if (unlikely(len == MAX_PKT_BURST)) {
507 l2fwd_send_burst(qconf, MAX_PKT_BURST, port);
511 qconf->tx_mbufs[port].len = len; return 0;
514 To ensure that no packets remain in the tables, the flush job exists. The l2fwd_flush_job()
515 is called periodically to for each lcore draining TX queue of each port.
516 This technique introduces some latency when there are not many packets to send,
517 however it improves performance:
522 l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
526 struct lcore_queue_conf *qconf;
527 struct mbuf_table *m_table;
530 lcore_id = rte_lcore_id();
531 qconf = &lcore_queue_conf[lcore_id];
533 rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
535 now = rte_get_timer_cycles();
536 lcore_id = rte_lcore_id();
537 qconf = &lcore_queue_conf[lcore_id];
538 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
539 m_table = &qconf->tx_mbufs[portid];
540 if (m_table->len == 0 || m_table->next_flush_time <= now)
543 l2fwd_send_burst(qconf, portid);
547 /* Pass target to indicate that this job is happy of time interval
548 * in which it was called. */
549 rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);