l2_forward_job_stats.rst revision 8b25d1ad
1..  BSD LICENSE
2    Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
3    All rights reserved.
4
5    Redistribution and use in source and binary forms, with or without
6    modification, are permitted provided that the following conditions
7    are met:
8
9    * Redistributions of source code must retain the above copyright
10    notice, this list of conditions and the following disclaimer.
11    * Redistributions in binary form must reproduce the above copyright
12    notice, this list of conditions and the following disclaimer in
13    the documentation and/or other materials provided with the
14    distribution.
15    * Neither the name of Intel Corporation nor the names of its
16    contributors may be used to endorse or promote products derived
17    from this software without specific prior written permission.
18
19    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
23    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
25    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
29    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30
31L2 Forwarding Sample Application (in Real and Virtualized Environments) with core load statistics.
32==================================================================================================
33
34The L2 Forwarding sample application is a simple example of packet processing using
35the Data Plane Development Kit (DPDK) which
36also takes advantage of Single Root I/O Virtualization (SR-IOV) features in a virtualized environment.
37
38.. note::
39
40    This application is a variation of L2 Forwarding sample application. It demonstrate possible
41    scheme of job stats library usage therefore some parts of this document is identical with original
42    L2 forwarding application.
43
44Overview
45--------
46
47The L2 Forwarding sample application, which can operate in real and virtualized environments,
48performs L2 forwarding for each packet that is received.
49The destination port is the adjacent port from the enabled portmask, that is,
50if the first four ports are enabled (portmask 0xf),
51ports 1 and 2 forward into each other, and ports 3 and 4 forward into each other.
52Also, the MAC addresses are affected as follows:
53
54*   The source MAC address is replaced by the TX port MAC address
55
56*   The destination MAC address is replaced by  02:00:00:00:00:TX_PORT_ID
57
58This application can be used to benchmark performance using a traffic-generator, as shown in the :numref:`figure_l2_fwd_benchmark_setup_jobstats`.
59
60The application can also be used in a virtualized environment as shown in :numref:`figure_l2_fwd_virtenv_benchmark_setup_jobstats`.
61
62The L2 Forwarding application can also be used as a starting point for developing a new application based on the DPDK.
63
64.. _figure_l2_fwd_benchmark_setup_jobstats:
65
66.. figure:: img/l2_fwd_benchmark_setup.*
67
68   Performance Benchmark Setup (Basic Environment)
69
70.. _figure_l2_fwd_virtenv_benchmark_setup_jobstats:
71
72.. figure:: img/l2_fwd_virtenv_benchmark_setup.*
73
74   Performance Benchmark Setup (Virtualized Environment)
75
76
77Virtual Function Setup Instructions
78~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
79
80This application can use the virtual function available in the system and
81therefore can be used in a virtual machine without passing through
82the whole Network Device into a guest machine in a virtualized scenario.
83The virtual functions can be enabled in the host machine or the hypervisor with the respective physical function driver.
84
85For example, in a Linux* host machine, it is possible to enable a virtual function using the following command:
86
87.. code-block:: console
88
89    modprobe ixgbe max_vfs=2,2
90
91This command enables two Virtual Functions on each of Physical Function of the NIC,
92with two physical ports in the PCI configuration space.
93It is important to note that enabled Virtual Function 0 and 2 would belong to Physical Function 0
94and Virtual Function 1 and 3 would belong to Physical Function 1,
95in this case enabling a total of four Virtual Functions.
96
97Compiling the Application
98-------------------------
99
100#.  Go to the example directory:
101
102    .. code-block:: console
103
104        export RTE_SDK=/path/to/rte_sdk
105        cd ${RTE_SDK}/examples/l2fwd-jobstats
106
107#.  Set the target (a default target is used if not specified). For example:
108
109    .. code-block:: console
110
111        export RTE_TARGET=x86_64-native-linuxapp-gcc
112
113    *See the DPDK Getting Started Guide* for possible RTE_TARGET values.
114
115#.  Build the application:
116
117    .. code-block:: console
118
119        make
120
121Running the Application
122-----------------------
123
124The application requires a number of command line options:
125
126.. code-block:: console
127
128    ./build/l2fwd-jobstats [EAL options] -- -p PORTMASK [-q NQ] [-l]
129
130where,
131
132*   p PORTMASK: A hexadecimal bitmask of the ports to configure
133
134*   q NQ: A number of queues (=ports) per lcore (default is 1)
135
136*   l: Use locale thousands separator when formatting big numbers.
137
138To run the application in linuxapp environment with 4 lcores, 16 ports, 8 RX queues per lcore and
139thousands  separator printing, issue the command:
140
141.. code-block:: console
142
143    $ ./build/l2fwd-jobstats -c f -n 4 -- -q 8 -p ffff -l
144
145Refer to the *DPDK Getting Started Guide* for general information on running applications
146and the Environment Abstraction Layer (EAL) options.
147
148Explanation
149-----------
150
151The following sections provide some explanation of the code.
152
153Command Line Arguments
154~~~~~~~~~~~~~~~~~~~~~~
155
156The L2 Forwarding sample application takes specific parameters,
157in addition to Environment Abstraction Layer (EAL) arguments
158(see `Running the Application`_).
159The preferred way to parse parameters is to use the getopt() function,
160since it is part of a well-defined and portable library.
161
162The parsing of arguments is done in the l2fwd_parse_args() function.
163The method of argument parsing is not described here.
164Refer to the *glibc getopt(3)* man page for details.
165
166EAL arguments are parsed first, then application-specific arguments.
167This is done at the beginning of the main() function:
168
169.. code-block:: c
170
171    /* init EAL */
172
173    ret = rte_eal_init(argc, argv);
174    if (ret < 0)
175        rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
176
177    argc -= ret;
178    argv += ret;
179
180    /* parse application arguments (after the EAL ones) */
181
182    ret = l2fwd_parse_args(argc, argv);
183    if (ret < 0)
184        rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
185
186Mbuf Pool Initialization
187~~~~~~~~~~~~~~~~~~~~~~~~
188
189Once the arguments are parsed, the mbuf pool is created.
190The mbuf pool contains a set of mbuf objects that will be used by the driver
191and the application to store network packet data:
192
193.. code-block:: c
194
195    /* create the mbuf pool */
196    l2fwd_pktmbuf_pool =
197        rte_mempool_create("mbuf_pool", NB_MBUF,
198                   MBUF_SIZE, 32,
199                   sizeof(struct rte_pktmbuf_pool_private),
200                   rte_pktmbuf_pool_init, NULL,
201                   rte_pktmbuf_init, NULL,
202                   rte_socket_id(), 0);
203
204    if (l2fwd_pktmbuf_pool == NULL)
205        rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
206
207The rte_mempool is a generic structure used to handle pools of objects.
208In this case, it is necessary to create a pool that will be used by the driver,
209which expects to have some reserved space in the mempool structure,
210sizeof(struct rte_pktmbuf_pool_private) bytes.
211The number of allocated pkt mbufs is NB_MBUF, with a size of MBUF_SIZE each.
212A per-lcore cache of 32 mbufs is kept.
213The memory is allocated in rte_socket_id() socket,
214but it is possible to extend this code to allocate one mbuf pool per socket.
215
216Two callback pointers are also given to the rte_mempool_create() function:
217
218*   The first callback pointer is to rte_pktmbuf_pool_init() and is used
219    to initialize the private data of the mempool, which is needed by the driver.
220    This function is provided by the mbuf API, but can be copied and extended by the developer.
221
222*   The second callback pointer given to rte_mempool_create() is the mbuf initializer.
223    The default is used, that is, rte_pktmbuf_init(), which is provided in the rte_mbuf library.
224    If a more complex application wants to extend the rte_pktmbuf structure for its own needs,
225    a new function derived from rte_pktmbuf_init( ) can be created.
226
227Driver Initialization
228~~~~~~~~~~~~~~~~~~~~~
229
230The main part of the code in the main() function relates to the initialization of the driver.
231To fully understand this code, it is recommended to study the chapters that related to the Poll Mode Driver
232in the *DPDK Programmer's Guide* and the *DPDK API Reference*.
233
234.. code-block:: c
235
236    nb_ports = rte_eth_dev_count();
237
238    if (nb_ports == 0)
239        rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
240
241    /* reset l2fwd_dst_ports */
242
243    for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
244        l2fwd_dst_ports[portid] = 0;
245
246    last_port = 0;
247
248    /*
249     * Each logical core is assigned a dedicated TX queue on each port.
250     */
251    for (portid = 0; portid < nb_ports; portid++) {
252        /* skip ports that are not enabled */
253        if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
254           continue;
255
256        if (nb_ports_in_mask % 2) {
257            l2fwd_dst_ports[portid] = last_port;
258            l2fwd_dst_ports[last_port] = portid;
259        }
260        else
261           last_port = portid;
262
263        nb_ports_in_mask++;
264
265        rte_eth_dev_info_get((uint8_t) portid, &dev_info);
266    }
267
268The next step is to configure the RX and TX queues.
269For each port, there is only one RX queue (only one lcore is able to poll a given port).
270The number of TX queues depends on the number of available lcores.
271The rte_eth_dev_configure() function is used to configure the number of queues for a port:
272
273.. code-block:: c
274
275    ret = rte_eth_dev_configure((uint8_t)portid, 1, 1, &port_conf);
276    if (ret < 0)
277        rte_exit(EXIT_FAILURE, "Cannot configure device: "
278            "err=%d, port=%u\n",
279            ret, portid);
280
281The global configuration is stored in a static structure:
282
283.. code-block:: c
284
285    static const struct rte_eth_conf port_conf = {
286        .rxmode = {
287            .split_hdr_size = 0,
288            .header_split = 0,   /**< Header Split disabled */
289            .hw_ip_checksum = 0, /**< IP checksum offload disabled */
290            .hw_vlan_filter = 0, /**< VLAN filtering disabled */
291            .jumbo_frame = 0,    /**< Jumbo Frame Support disabled */
292            .hw_strip_crc= 0,    /**< CRC stripped by hardware */
293        },
294
295        .txmode = {
296            .mq_mode = ETH_DCB_NONE
297        },
298    };
299
300RX Queue Initialization
301~~~~~~~~~~~~~~~~~~~~~~~
302
303The application uses one lcore to poll one or several ports, depending on the -q option,
304which specifies the number of queues per lcore.
305
306For example, if the user specifies -q 4, the application is able to poll four ports with one lcore.
307If there are 16 ports on the target (and if the portmask argument is -p ffff ),
308the application will need four lcores to poll all the ports.
309
310.. code-block:: c
311
312    ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
313                rte_eth_dev_socket_id(portid),
314                NULL,
315                l2fwd_pktmbuf_pool);
316
317    if (ret < 0)
318        rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
319                ret, (unsigned) portid);
320
321The list of queues that must be polled for a given lcore is stored in a private structure called struct lcore_queue_conf.
322
323.. code-block:: c
324
325    struct lcore_queue_conf {
326        unsigned n_rx_port;
327        unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
328        truct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
329
330        struct rte_timer rx_timers[MAX_RX_QUEUE_PER_LCORE];
331        struct rte_jobstats port_fwd_jobs[MAX_RX_QUEUE_PER_LCORE];
332
333        struct rte_timer flush_timer;
334        struct rte_jobstats flush_job;
335        struct rte_jobstats idle_job;
336        struct rte_jobstats_context jobs_context;
337
338        rte_atomic16_t stats_read_pending;
339        rte_spinlock_t lock;
340    } __rte_cache_aligned;
341
342Values of struct lcore_queue_conf:
343
344*   n_rx_port and rx_port_list[] are used in the main packet processing loop
345    (see Section `Receive, Process and Transmit Packets`_ later in this chapter).
346
347*   rx_timers and flush_timer are used to ensure forced TX on low packet rate.
348
349*   flush_job, idle_job and jobs_context are librte_jobstats objects used for managing l2fwd jobs.
350
351*   stats_read_pending and lock are used during job stats read phase.
352
353TX Queue Initialization
354~~~~~~~~~~~~~~~~~~~~~~~
355
356Each lcore should be able to transmit on any port. For every port, a single TX queue is initialized.
357
358.. code-block:: c
359
360    /* init one TX queue on each port */
361
362    fflush(stdout);
363    ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
364            rte_eth_dev_socket_id(portid),
365            NULL);
366    if (ret < 0)
367        rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
368                ret, (unsigned) portid);
369
370Jobs statistics initialization
371~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
372There are several statistics objects available:
373
374*   Flush job statistics
375
376.. code-block:: c
377
378    rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
379            drain_tsc, 0);
380
381    rte_timer_init(&qconf->flush_timer);
382    ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL,
383                lcore_id, &l2fwd_flush_job, NULL);
384
385    if (ret < 0) {
386        rte_exit(1, "Failed to reset flush job timer for lcore %u: %s",
387                    lcore_id, rte_strerror(-ret));
388    }
389
390*   Statistics per RX port
391
392.. code-block:: c
393
394    rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
395    rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
396
397    rte_timer_init(&qconf->rx_timers[i]);
398    ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
399            l2fwd_fwd_job, (void *)(uintptr_t)i);
400
401    if (ret < 0) {
402        rte_exit(1, "Failed to reset lcore %u port %u job timer: %s",
403                    lcore_id, qconf->rx_port_list[i], rte_strerror(-ret));
404    }
405
406Following parameters are passed to rte_jobstats_init():
407
408*   0 as minimal poll period
409
410*   drain_tsc as maximum poll period
411
412*   MAX_PKT_BURST as desired target value (RX burst size)
413
414Main loop
415~~~~~~~~~
416
417The forwarding path is reworked comparing to original L2 Forwarding application.
418In the l2fwd_main_loop() function three loops are placed.
419
420.. code-block:: c
421
422    for (;;) {
423        rte_spinlock_lock(&qconf->lock);
424
425        do {
426            rte_jobstats_context_start(&qconf->jobs_context);
427
428            /* Do the Idle job:
429             * - Read stats_read_pending flag
430             * - check if some real job need to be executed
431             */
432            rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
433
434            do {
435                uint8_t i;
436                uint64_t now = rte_get_timer_cycles();
437
438                need_manage = qconf->flush_timer.expire < now;
439                /* Check if we was esked to give a stats. */
440                stats_read_pending =
441                        rte_atomic16_read(&qconf->stats_read_pending);
442                need_manage |= stats_read_pending;
443
444                for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
445                    need_manage = qconf->rx_timers[i].expire < now;
446
447            } while (!need_manage);
448            rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
449
450            rte_timer_manage();
451            rte_jobstats_context_finish(&qconf->jobs_context);
452        } while (likely(stats_read_pending == 0));
453
454        rte_spinlock_unlock(&qconf->lock);
455        rte_pause();
456    }
457
458First infinite for loop is to minimize impact of stats reading. Lock is only locked/unlocked when asked.
459
460Second 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.
461In this place functions l2fwd_fwd_job() and l2fwd_flush_job() are called when needed.
462Then 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
463and if stats_read_pending is set, loop breaks allowing stats to be read.
464
465Third 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
466for this lcore. Statistics from this part of code is considered as the headroom available for additional processing.
467
468Receive, Process and Transmit Packets
469~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
470
471The main task of l2fwd_fwd_job() function is to read ingress packets from the RX queue of particular port and forward it.
472This is done using the following code:
473
474.. code-block:: c
475
476    total_nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
477            MAX_PKT_BURST);
478
479    for (j = 0; j < total_nb_rx; j++) {
480        m = pkts_burst[j];
481        rte_prefetch0(rte_pktmbuf_mtod(m, void *));
482        l2fwd_simple_forward(m, portid);
483    }
484
485Packets are read in a burst of size MAX_PKT_BURST.
486Then, each mbuf in the table is processed by the l2fwd_simple_forward() function.
487The processing is very simple: process the TX port from the RX port, then replace the source and destination MAC addresses.
488
489The rte_eth_rx_burst() function writes the mbuf pointers in a local table and returns the number of available mbufs in the table.
490
491After first read second try is issued.
492
493.. code-block:: c
494
495    if (total_nb_rx == MAX_PKT_BURST) {
496        const uint16_t nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
497                MAX_PKT_BURST);
498
499        total_nb_rx += nb_rx;
500        for (j = 0; j < nb_rx; j++) {
501            m = pkts_burst[j];
502            rte_prefetch0(rte_pktmbuf_mtod(m, void *));
503            l2fwd_simple_forward(m, portid);
504        }
505    }
506
507This second read is important to give job stats library a feedback how many packets was processed.
508
509.. code-block:: c
510
511    /* Adjust period time in which we are running here. */
512    if (rte_jobstats_finish(job, total_nb_rx) != 0) {
513        rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
514                lcore_id, l2fwd_fwd_job, arg);
515    }
516
517To maximize performance exactly MAX_PKT_BURST is expected (the target value) to be read for each l2fwd_fwd_job() call.
518If total_nb_rx is smaller than target value job->period will be increased. If it is greater the period will be decreased.
519
520.. note::
521
522    In the following code, one line for getting the output port requires some explanation.
523
524During the initialization process, a static array of destination ports (l2fwd_dst_ports[]) is filled such that for each source port,
525a destination port is assigned that is either the next or previous enabled port from the portmask.
526Naturally, the number of ports in the portmask must be even, otherwise, the application exits.
527
528.. code-block:: c
529
530    static void
531    l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
532    {
533        struct ether_hdr *eth;
534        void *tmp;
535        unsigned dst_port;
536
537        dst_port = l2fwd_dst_ports[portid];
538
539        eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
540
541        /* 02:00:00:00:00:xx */
542
543        tmp = &eth->d_addr.addr_bytes[0];
544
545        *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t) dst_port << 40);
546
547        /* src addr */
548
549        ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], &eth->s_addr);
550
551        l2fwd_send_packet(m, (uint8_t) dst_port);
552    }
553
554Then, the packet is sent using the l2fwd_send_packet (m, dst_port) function.
555For this test application, the processing is exactly the same for all packets arriving on the same RX port.
556Therefore, it would have been possible to call the l2fwd_send_burst() function directly from the main loop
557to send all the received packets on the same TX port,
558using the burst-oriented send function, which is more efficient.
559
560However, in real-life applications (such as, L3 routing),
561packet N is not necessarily forwarded on the same port as packet N-1.
562The application is implemented to illustrate that, so the same approach can be reused in a more complex application.
563
564The l2fwd_send_packet() function stores the packet in a per-lcore and per-txport table.
565If the table is full, the whole packets table is transmitted using the l2fwd_send_burst() function:
566
567.. code-block:: c
568
569    /* Send the packet on an output interface */
570
571    static int
572    l2fwd_send_packet(struct rte_mbuf *m, uint8_t port)
573    {
574        unsigned lcore_id, len;
575        struct lcore_queue_conf *qconf;
576
577        lcore_id = rte_lcore_id();
578        qconf = &lcore_queue_conf[lcore_id];
579        len = qconf->tx_mbufs[port].len;
580        qconf->tx_mbufs[port].m_table[len] = m;
581        len++;
582
583        /* enough pkts to be sent */
584
585        if (unlikely(len == MAX_PKT_BURST)) {
586            l2fwd_send_burst(qconf, MAX_PKT_BURST, port);
587            len = 0;
588        }
589
590        qconf->tx_mbufs[port].len = len; return 0;
591    }
592
593To ensure that no packets remain in the tables, the flush job exists. The l2fwd_flush_job()
594is called periodically to for each lcore draining TX queue of each port.
595This technique introduces some latency when there are not many packets to send,
596however it improves performance:
597
598.. code-block:: c
599
600    static void
601    l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
602    {
603        uint64_t now;
604        unsigned lcore_id;
605        struct lcore_queue_conf *qconf;
606        struct mbuf_table *m_table;
607        uint8_t portid;
608
609        lcore_id = rte_lcore_id();
610        qconf = &lcore_queue_conf[lcore_id];
611
612        rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
613
614        now = rte_get_timer_cycles();
615        lcore_id = rte_lcore_id();
616        qconf = &lcore_queue_conf[lcore_id];
617        for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
618            m_table = &qconf->tx_mbufs[portid];
619            if (m_table->len == 0 || m_table->next_flush_time <= now)
620                continue;
621
622            l2fwd_send_burst(qconf, portid);
623        }
624
625
626        /* Pass target to indicate that this job is happy of time interval
627         * in which it was called. */
628        rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);
629    }
630