2 Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5 Redistribution and use in source and binary forms, with or without
6 modification, are permitted provided that the following conditions
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
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.
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.
31 .. _multi_process_app:
33 Multi-process Sample Application
34 ================================
36 This chapter describes the example applications for multi-processing that are included in the DPDK.
41 Building the Sample Applications
42 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
43 The multi-process example applications are built in the same way as other sample applications,
44 and as documented in the *DPDK Getting Started Guide*.
47 To compile the sample application see :doc:`compiling`.
49 The applications are located in the ``multi_process`` sub-directory.
53 If just a specific multi-process application needs to be built,
54 the final make command can be run just in that application's directory,
55 rather than at the top-level multi-process directory.
57 Basic Multi-process Example
58 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
60 The examples/simple_mp folder in the DPDK release contains a basic example application to demonstrate how
61 two DPDK processes can work together using queues and memory pools to share information.
63 Running the Application
64 ^^^^^^^^^^^^^^^^^^^^^^^
66 To run the application, start one copy of the simple_mp binary in one terminal,
67 passing at least two cores in the coremask/corelist, as follows:
69 .. code-block:: console
71 ./build/simple_mp -l 0-1 -n 4 --proc-type=primary
73 For the first DPDK process run, the proc-type flag can be omitted or set to auto,
74 since all DPDK processes will default to being a primary instance,
75 meaning they have control over the hugepage shared memory regions.
76 The process should start successfully and display a command prompt as follows:
78 .. code-block:: console
80 $ ./build/simple_mp -l 0-1 -n 4 --proc-type=primary
81 EAL: coremask set to 3
82 EAL: Detected lcore 0 on socket 0
83 EAL: Detected lcore 1 on socket 0
84 EAL: Detected lcore 2 on socket 0
85 EAL: Detected lcore 3 on socket 0
88 EAL: Requesting 2 pages of size 1073741824
89 EAL: Requesting 768 pages of size 2097152
90 EAL: Ask a virtual area of 0x40000000 bytes
91 EAL: Virtual area found at 0x7ff200000000 (size = 0x40000000)
94 EAL: check igb_uio module
95 EAL: check module finished
96 EAL: Master core 0 is ready (tid=54e41820)
97 EAL: Core 1 is ready (tid=53b32700)
103 To run the secondary process to communicate with the primary process,
104 again run the same binary setting at least two cores in the coremask/corelist:
106 .. code-block:: console
108 ./build/simple_mp -l 2-3 -n 4 --proc-type=secondary
110 When running a secondary process such as that shown above, the proc-type parameter can again be specified as auto.
111 However, omitting the parameter altogether will cause the process to try and start as a primary rather than secondary process.
113 Once the process type is specified correctly,
114 the process starts up, displaying largely similar status messages to the primary instance as it initializes.
115 Once again, you will be presented with a command prompt.
117 Once both processes are running, messages can be sent between them using the send command.
118 At any stage, either process can be terminated using the quit command.
120 .. code-block:: console
122 EAL: Master core 10 is ready (tid=b5f89820) EAL: Master core 8 is ready (tid=864a3820)
123 EAL: Core 11 is ready (tid=84ffe700) EAL: Core 9 is ready (tid=85995700)
124 Starting core 11 Starting core 9
125 simple_mp > send hello_secondary simple_mp > core 9: Received 'hello_secondary'
126 simple_mp > core 11: Received 'hello_primary' simple_mp > send hello_primary
127 simple_mp > quit simple_mp > quit
131 If the primary instance is terminated, the secondary instance must also be shut-down and restarted after the primary.
132 This is necessary because the primary instance will clear and reset the shared memory regions on startup,
133 invalidating the secondary process's pointers.
134 The secondary process can be stopped and restarted without affecting the primary process.
136 How the Application Works
137 ^^^^^^^^^^^^^^^^^^^^^^^^^
139 The core of this example application is based on using two queues and a single memory pool in shared memory.
140 These three objects are created at startup by the primary process,
141 since the secondary process cannot create objects in memory as it cannot reserve memory zones,
142 and the secondary process then uses lookup functions to attach to these objects as it starts up.
146 if (rte_eal_process_type() == RTE_PROC_PRIMARY){
147 send_ring = rte_ring_create(_PRI_2_SEC, ring_size, SOCKET0, flags);
148 recv_ring = rte_ring_create(_SEC_2_PRI, ring_size, SOCKET0, flags);
149 message_pool = rte_mempool_create(_MSG_POOL, pool_size, string_size, pool_cache, priv_data_sz, NULL, NULL, NULL, NULL, SOCKET0, flags);
151 recv_ring = rte_ring_lookup(_PRI_2_SEC);
152 send_ring = rte_ring_lookup(_SEC_2_PRI);
153 message_pool = rte_mempool_lookup(_MSG_POOL);
156 Note, however, that the named ring structure used as send_ring in the primary process is the recv_ring in the secondary process.
158 Once the rings and memory pools are all available in both the primary and secondary processes,
159 the application simply dedicates two threads to sending and receiving messages respectively.
160 The receive thread simply dequeues any messages on the receive ring, prints them,
161 and frees the buffer space used by the messages back to the memory pool.
162 The send thread makes use of the command-prompt library to interactively request user input for messages to send.
163 Once a send command is issued by the user, a buffer is allocated from the memory pool, filled in with the message contents,
164 then enqueued on the appropriate rte_ring.
166 Symmetric Multi-process Example
167 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
169 The second example of DPDK multi-process support demonstrates how a set of processes can run in parallel,
170 with each process performing the same set of packet- processing operations.
171 (Since each process is identical in functionality to the others,
172 we refer to this as symmetric multi-processing, to differentiate it from asymmetric multi- processing -
173 such as a client-server mode of operation seen in the next example,
174 where different processes perform different tasks, yet co-operate to form a packet-processing system.)
175 The following diagram shows the data-flow through the application, using two processes.
177 .. _figure_sym_multi_proc_app:
179 .. figure:: img/sym_multi_proc_app.*
181 Example Data Flow in a Symmetric Multi-process Application
184 As the diagram shows, each process reads packets from each of the network ports in use.
185 RSS is used to distribute incoming packets on each port to different hardware RX queues.
186 Each process reads a different RX queue on each port and so does not contend with any other process for that queue access.
187 Similarly, each process writes outgoing packets to a different TX queue on each port.
189 Running the Application
190 ^^^^^^^^^^^^^^^^^^^^^^^
192 As with the simple_mp example, the first instance of the symmetric_mp process must be run as the primary instance,
193 though with a number of other application- specific parameters also provided after the EAL arguments.
194 These additional parameters are:
196 * -p <portmask>, where portmask is a hexadecimal bitmask of what ports on the system are to be used.
197 For example: -p 3 to use ports 0 and 1 only.
199 * --num-procs <N>, where N is the total number of symmetric_mp instances that will be run side-by-side to perform packet processing.
200 This parameter is used to configure the appropriate number of receive queues on each network port.
202 * --proc-id <n>, where n is a numeric value in the range 0 <= n < N (number of processes, specified above).
203 This identifies which symmetric_mp instance is being run, so that each process can read a unique receive queue on each network port.
205 The secondary symmetric_mp instances must also have these parameters specified,
206 and the first two must be the same as those passed to the primary instance, or errors result.
208 For example, to run a set of four symmetric_mp instances, running on lcores 1-4,
209 all performing level-2 forwarding of packets between ports 0 and 1,
210 the following commands can be used (assuming run as root):
212 .. code-block:: console
214 # ./build/symmetric_mp -l 1 -n 4 --proc-type=auto -- -p 3 --num-procs=4 --proc-id=0
215 # ./build/symmetric_mp -l 2 -n 4 --proc-type=auto -- -p 3 --num-procs=4 --proc-id=1
216 # ./build/symmetric_mp -l 3 -n 4 --proc-type=auto -- -p 3 --num-procs=4 --proc-id=2
217 # ./build/symmetric_mp -l 4 -n 4 --proc-type=auto -- -p 3 --num-procs=4 --proc-id=3
221 In the above example, the process type can be explicitly specified as primary or secondary, rather than auto.
222 When using auto, the first process run creates all the memory structures needed for all processes -
223 irrespective of whether it has a proc-id of 0, 1, 2 or 3.
227 For the symmetric multi-process example, since all processes work in the same manner,
228 once the hugepage shared memory and the network ports are initialized,
229 it is not necessary to restart all processes if the primary instance dies.
230 Instead, that process can be restarted as a secondary,
231 by explicitly setting the proc-type to secondary on the command line.
232 (All subsequent instances launched will also need this explicitly specified,
233 as auto-detection will detect no primary processes running and therefore attempt to re-initialize shared memory.)
235 How the Application Works
236 ^^^^^^^^^^^^^^^^^^^^^^^^^
238 The initialization calls in both the primary and secondary instances are the same for the most part,
239 calling the rte_eal_init(), 1 G and 10 G driver initialization and then rte_pci_probe() functions.
240 Thereafter, the initialization done depends on whether the process is configured as a primary or secondary instance.
242 In the primary instance, a memory pool is created for the packet mbufs and the network ports to be used are initialized -
243 the number of RX and TX queues per port being determined by the num-procs parameter passed on the command-line.
244 The structures for the initialized network ports are stored in shared memory and
245 therefore will be accessible by the secondary process as it initializes.
250 rte_exit(EXIT_FAILURE, "Application must use an even number of ports\n");
252 for(i = 0; i < num_ports; i++){
253 if(proc_type == RTE_PROC_PRIMARY)
254 if (smp_port_init(ports[i], mp, (uint16_t)num_procs) < 0)
255 rte_exit(EXIT_FAILURE, "Error initializing ports\n");
258 In the secondary instance, rather than initializing the network ports, the port information exported by the primary process is used,
259 giving the secondary process access to the hardware and software rings for each network port.
260 Similarly, the memory pool of mbufs is accessed by doing a lookup for it by name:
264 mp = (proc_type == RTE_PROC_SECONDARY) ? rte_mempool_lookup(_SMP_MBUF_POOL) : rte_mempool_create(_SMP_MBUF_POOL, NB_MBUFS, MBUF_SIZE, ... )
266 Once this initialization is complete, the main loop of each process, both primary and secondary,
267 is exactly the same - each process reads from each port using the queue corresponding to its proc-id parameter,
268 and writes to the corresponding transmit queue on the output port.
270 Client-Server Multi-process Example
271 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
273 The third example multi-process application included with the DPDK shows how one can
274 use a client-server type multi-process design to do packet processing.
275 In this example, a single server process performs the packet reception from the ports being used and
276 distributes these packets using round-robin ordering among a set of client processes,
277 which perform the actual packet processing.
278 In this case, the client applications just perform level-2 forwarding of packets by sending each packet out on a different network port.
280 The following diagram shows the data-flow through the application, using two client processes.
282 .. _figure_client_svr_sym_multi_proc_app:
284 .. figure:: img/client_svr_sym_multi_proc_app.*
286 Example Data Flow in a Client-Server Symmetric Multi-process Application
289 Running the Application
290 ^^^^^^^^^^^^^^^^^^^^^^^
292 The server process must be run initially as the primary process to set up all memory structures for use by the clients.
293 In addition to the EAL parameters, the application- specific parameters are:
295 * -p <portmask >, where portmask is a hexadecimal bitmask of what ports on the system are to be used.
296 For example: -p 3 to use ports 0 and 1 only.
298 * -n <num-clients>, where the num-clients parameter is the number of client processes that will process the packets received
299 by the server application.
303 In the server process, a single thread, the master thread, that is, the lowest numbered lcore in the coremask/corelist, performs all packet I/O.
304 If a coremask/corelist is specified with more than a single lcore bit set in it,
305 an additional lcore will be used for a thread to periodically print packet count statistics.
307 Since the server application stores configuration data in shared memory, including the network ports to be used,
308 the only application parameter needed by a client process is its client instance ID.
309 Therefore, to run a server application on lcore 1 (with lcore 2 printing statistics) along with two client processes running on lcores 3 and 4,
310 the following commands could be used:
312 .. code-block:: console
314 # ./mp_server/build/mp_server -l 1-2 -n 4 -- -p 3 -n 2
315 # ./mp_client/build/mp_client -l 3 -n 4 --proc-type=auto -- -n 0
316 # ./mp_client/build/mp_client -l 4 -n 4 --proc-type=auto -- -n 1
320 If the server application dies and needs to be restarted, all client applications also need to be restarted,
321 as there is no support in the server application for it to run as a secondary process.
322 Any client processes that need restarting can be restarted without affecting the server process.
324 How the Application Works
325 ^^^^^^^^^^^^^^^^^^^^^^^^^
327 The server process performs the network port and data structure initialization much as the symmetric multi-process application does when run as primary.
328 One additional enhancement in this sample application is that the server process stores its port configuration data in a memory zone in hugepage shared memory.
329 This eliminates the need for the client processes to have the portmask parameter passed into them on the command line,
330 as is done for the symmetric multi-process application, and therefore eliminates mismatched parameters as a potential source of errors.
332 In the same way that the server process is designed to be run as a primary process instance only,
333 the client processes are designed to be run as secondary instances only.
334 They have no code to attempt to create shared memory objects.
335 Instead, handles to all needed rings and memory pools are obtained via calls to rte_ring_lookup() and rte_mempool_lookup().
336 The network ports for use by the processes are obtained by loading the network port drivers and probing the PCI bus,
337 which will, as in the symmetric multi-process example,
338 automatically get access to the network ports using the settings already configured by the primary/server process.
340 Once all applications are initialized, the server operates by reading packets from each network port in turn and
341 distributing those packets to the client queues (software rings, one for each client process) in round-robin order.
342 On the client side, the packets are read from the rings in as big of bursts as possible, then routed out to a different network port.
343 The routing used is very simple. All packets received on the first NIC port are transmitted back out on the second port and vice versa.
344 Similarly, packets are routed between the 3rd and 4th network ports and so on.
345 The sending of packets is done by writing the packets directly to the network ports; they are not transferred back via the server process.
347 In both the server and the client processes, outgoing packets are buffered before being sent,
348 so as to allow the sending of multiple packets in a single burst to improve efficiency.
349 For example, the client process will buffer packets to send,
350 until either the buffer is full or until we receive no further packets from the server.
352 Master-slave Multi-process Example
353 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
355 The fourth example of DPDK multi-process support demonstrates a master-slave model that
356 provide the capability of application recovery if a slave process crashes or meets unexpected conditions.
357 In addition, it also demonstrates the floating process,
358 which can run among different cores in contrast to the traditional way of binding a process/thread to a specific CPU core,
359 using the local cache mechanism of mempool structures.
361 This application performs the same functionality as the L2 Forwarding sample application,
362 therefore this chapter does not cover that part but describes functionality that is introduced in this multi-process example only.
363 Please refer to :doc:`l2_forward_real_virtual` for more information.
365 Unlike previous examples where all processes are started from the command line with input arguments, in this example,
366 only one process is spawned from the command line and that process creates other processes.
367 The following section describes this in more detail.
369 Master-slave Process Models
370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
372 The process spawned from the command line is called the *master process* in this document.
373 A process created by the master is called a *slave process*.
374 The application has only one master process, but could have multiple slave processes.
376 Once the master process begins to run, it tries to initialize all the resources such as
377 memory, CPU cores, driver, ports, and so on, as the other examples do.
378 Thereafter, it creates slave processes, as shown in the following figure.
380 .. _figure_master_slave_proc:
382 .. figure:: img/master_slave_proc.*
384 Master-slave Process Workflow
387 The master process calls the rte_eal_mp_remote_launch() EAL function to launch an application function for each pinned thread through the pipe.
388 Then, it waits to check if any slave processes have exited.
389 If so, the process tries to re-initialize the resources that belong to that slave and launch them in the pinned thread entry again.
390 The following section describes the recovery procedures in more detail.
392 For each pinned thread in EAL, after reading any data from the pipe, it tries to call the function that the application specified.
393 In this master specified function, a fork() call creates a slave process that performs the L2 forwarding task.
394 Then, the function waits until the slave exits, is killed or crashes. Thereafter, it notifies the master of this event and returns.
395 Finally, the EAL pinned thread waits until the new function is launched.
397 After discussing the master-slave model, it is necessary to mention another issue, global and static variables.
399 For multiple-thread cases, all global and static variables have only one copy and they can be accessed by any thread if applicable.
400 So, they can be used to sync or share data among threads.
402 In the previous examples, each process has separate global and static variables in memory and are independent of each other.
403 If it is necessary to share the knowledge, some communication mechanism should be deployed, such as, memzone, ring, shared memory, and so on.
404 The global or static variables are not a valid approach to share data among processes.
405 For variables in this example, on the one hand, the slave process inherits all the knowledge of these variables after being created by the master.
406 On the other hand, other processes cannot know if one or more processes modifies them after slave creation since that
407 is the nature of a multiple process address space.
408 But this does not mean that these variables cannot be used to share or sync data; it depends on the use case.
409 The following are the possible use cases:
411 #. The master process starts and initializes a variable and it will never be changed after slave processes created. This case is OK.
413 #. After the slave processes are created, the master or slave cores need to change a variable, but other processes do not need to know the change.
414 This case is also OK.
416 #. After the slave processes are created, the master or a slave needs to change a variable.
417 In the meantime, one or more other process needs to be aware of the change.
418 In this case, global and static variables cannot be used to share knowledge. Another communication mechanism is needed.
419 A simple approach without lock protection can be a heap buffer allocated by rte_malloc or mem zone.
421 Slave Process Recovery Mechanism
422 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
424 Before talking about the recovery mechanism, it is necessary to know what is needed before a new slave instance can run if a previous one exited.
426 When a slave process exits, the system returns all the resources allocated for this process automatically.
427 However, this does not include the resources that were allocated by the DPDK. All the hardware resources are shared among the processes,
428 which include memzone, mempool, ring, a heap buffer allocated by the rte_malloc library, and so on.
429 If the new instance runs and the allocated resource is not returned, either resource allocation failed or the hardware resource is lost forever.
431 When a slave process runs, it may have dependencies on other processes.
432 They could have execution sequence orders; they could share the ring to communicate; they could share the same port for reception and forwarding;
433 they could use lock structures to do exclusive access in some critical path.
434 What happens to the dependent process(es) if the peer leaves?
435 The consequence are varied since the dependency cases are complex.
436 It depends on what the processed had shared.
437 However, it is necessary to notify the peer(s) if one slave exited.
438 Then, the peer(s) will be aware of that and wait until the new instance begins to run.
440 Therefore, to provide the capability to resume the new slave instance if the previous one exited, it is necessary to provide several mechanisms:
442 #. Keep a resource list for each slave process.
443 Before a slave process run, the master should prepare a resource list.
444 After it exits, the master could either delete the allocated resources and create new ones,
445 or re-initialize those for use by the new instance.
447 #. Set up a notification mechanism for slave process exit cases. After the specific slave leaves,
448 the master should be notified and then help to create a new instance.
449 This mechanism is provided in Section `Master-slave Process Models`_.
451 #. Use a synchronization mechanism among dependent processes.
452 The master should have the capability to stop or kill slave processes that have a dependency on the one that has exited.
453 Then, after the new instance of exited slave process begins to run, the dependency ones could resume or run from the start.
454 The example sends a STOP command to slave processes dependent on the exited one, then they will exit.
455 Thereafter, the master creates new instances for the exited slave processes.
457 The following diagram describes slave process recovery.
459 .. _figure_slave_proc_recov:
461 .. figure:: img/slave_proc_recov.*
463 Slave Process Recovery Process Flow
466 Floating Process Support
467 ^^^^^^^^^^^^^^^^^^^^^^^^
469 When the DPDK application runs, there is always a -c option passed in to indicate the cores that are enabled.
470 Then, the DPDK creates a thread for each enabled core.
471 By doing so, it creates a 1:1 mapping between the enabled core and each thread.
472 The enabled core always has an ID, therefore, each thread has a unique core ID in the DPDK execution environment.
473 With the ID, each thread can easily access the structures or resources exclusively belonging to it without using function parameter passing.
474 It can easily use the rte_lcore_id() function to get the value in every function that is called.
476 For threads/processes not created in that way, either pinned to a core or not, they will not own a unique ID and the
477 rte_lcore_id() function will not work in the correct way.
478 However, sometimes these threads/processes still need the unique ID mechanism to do easy access on structures or resources.
479 For example, the DPDK mempool library provides a local cache mechanism
480 (refer to :ref:`mempool_local_cache`)
481 for fast element allocation and freeing.
482 If using a non-unique ID or a fake one,
483 a race condition occurs if two or more threads/ processes with the same core ID try to use the local cache.
485 Therefore, unused core IDs from the passing of parameters with the -c option are used to organize the core ID allocation array.
486 Once the floating process is spawned, it tries to allocate a unique core ID from the array and release it on exit.
488 A natural way to spawn a floating process is to use the fork() function and allocate a unique core ID from the unused core ID array.
489 However, it is necessary to write new code to provide a notification mechanism for slave exit
490 and make sure the process recovery mechanism can work with it.
492 To avoid producing redundant code, the Master-Slave process model is still used to spawn floating processes,
493 then cancel the affinity to specific cores.
494 Besides that, clear the core ID assigned to the DPDK spawning a thread that has a 1:1 mapping with the core mask.
495 Thereafter, get a new core ID from the unused core ID allocation array.
500 This example has a command line similar to the L2 Forwarding sample application with a few differences.
502 To run the application, start one copy of the l2fwd_fork binary in one terminal.
503 Unlike the L2 Forwarding example,
504 this example requires at least three cores since the master process will wait and be accountable for slave process recovery.
505 The command is as follows:
507 .. code-block:: console
509 #./build/l2fwd_fork -l 2-4 -n 4 -- -p 3 -f
511 This example provides another -f option to specify the use of floating process.
512 If not specified, the example will use a pinned process to perform the L2 forwarding task.
514 To verify the recovery mechanism, proceed as follows: First, check the PID of the slave processes:
516 .. code-block:: console
518 #ps -fe | grep l2fwd_fork
519 root 5136 4843 29 11:11 pts/1 00:00:05 ./build/l2fwd_fork
520 root 5145 5136 98 11:11 pts/1 00:00:11 ./build/l2fwd_fork
521 root 5146 5136 98 11:11 pts/1 00:00:11 ./build/l2fwd_fork
523 Then, kill one of the slaves:
525 .. code-block:: console
529 After 1 or 2 seconds, check whether the slave has resumed:
531 .. code-block:: console
533 #ps -fe | grep l2fwd_fork
534 root 5136 4843 3 11:11 pts/1 00:00:06 ./build/l2fwd_fork
535 root 5247 5136 99 11:14 pts/1 00:00:01 ./build/l2fwd_fork
536 root 5248 5136 99 11:14 pts/1 00:00:01 ./build/l2fwd_fork
538 It can also monitor the traffic generator statics to see whether slave processes have resumed.
543 As described in previous sections,
544 not all global and static variables need to change to be accessible in multiple processes;
545 it depends on how they are used.
547 the statics info on packets dropped/forwarded/received count needs to be updated by the slave process,
548 and the master needs to see the update and print them out.
549 So, it needs to allocate a heap buffer using rte_zmalloc.
550 In addition, if the -f option is specified,
551 an array is needed to store the allocated core ID for the floating process so that the master can return it
552 after a slave has exited accidentally.
557 l2fwd_malloc_shared_struct(void)
559 port_statistics = rte_zmalloc("port_stat", sizeof(struct l2fwd_port_statistics) * RTE_MAX_ETHPORTS, 0);
561 if (port_statistics == NULL)
564 /* allocate mapping_id array */
569 mapping_id = rte_malloc("mapping_id", sizeof(unsigned) * RTE_MAX_LCORE, 0);
570 if (mapping_id == NULL)
573 for (i = 0 ;i < RTE_MAX_LCORE; i++)
574 mapping_id[i] = INVALID_MAPPING_ID;
580 For each slave process, packets are received from one port and forwarded to another port that another slave is operating on.
581 If the other slave exits accidentally, the port it is operating on may not work normally,
582 so the first slave cannot forward packets to that port.
583 There is a dependency on the port in this case. So, the master should recognize the dependency.
584 The following is the code to detect this dependency:
588 for (portid = 0; portid < nb_ports; portid++) {
589 /* skip ports that are not enabled */
591 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
594 /* Find pair ports' lcores */
596 find_lcore = find_pair_lcore = 0;
597 pair_port = l2fwd_dst_ports[portid];
599 for (i = 0; i < RTE_MAX_LCORE; i++) {
600 if (!rte_lcore_is_enabled(i))
603 for (j = 0; j < lcore_queue_conf[i].n_rx_port;j++) {
604 if (lcore_queue_conf[i].rx_port_list[j] == portid) {
610 if (lcore_queue_conf[i].rx_port_list[j] == pair_port) {
617 if (find_lcore && find_pair_lcore)
621 if (!find_lcore || !find_pair_lcore)
622 rte_exit(EXIT_FAILURE, "Not find port=%d pair\\n", portid);
624 printf("lcore %u and %u paired\\n", lcore, pair_lcore);
626 lcore_resource[lcore].pair_id = pair_lcore;
627 lcore_resource[pair_lcore].pair_id = lcore;
630 Before launching the slave process,
631 it is necessary to set up the communication channel between the master and slave so that
632 the master can notify the slave if its peer process with the dependency exited.
633 In addition, the master needs to register a callback function in the case where a specific slave exited.
637 for (i = 0; i < RTE_MAX_LCORE; i++) {
638 if (lcore_resource[i].enabled) {
639 /* Create ring for master and slave communication */
641 ret = create_ms_ring(i);
643 rte_exit(EXIT_FAILURE, "Create ring for lcore=%u failed",i);
645 if (flib_register_slave_exit_notify(i,slave_exit_cb) != 0)
646 rte_exit(EXIT_FAILURE, "Register master_trace_slave_exit failed");
650 After launching the slave process, the master waits and prints out the port statics periodically.
651 If an event indicating that a slave process exited is detected,
652 it sends the STOP command to the peer and waits until it has also exited.
653 Then, it tries to clean up the execution environment and prepare new resources.
654 Finally, the new slave instance is launched.
660 cur_tsc = rte_rdtsc();
661 diff_tsc = cur_tsc - prev_tsc;
663 /* if timer is enabled */
665 if (timer_period > 0) {
666 /* advance the timer */
667 timer_tsc += diff_tsc;
669 /* if timer has reached its timeout */
670 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
673 /* reset the timer */
680 /* Check any slave need restart or recreate */
682 rte_spinlock_lock(&res_lock);
684 for (i = 0; i < RTE_MAX_LCORE; i++) {
685 struct lcore_resource_struct *res = &lcore_resource[i];
686 struct lcore_resource_struct *pair = &lcore_resource[res->pair_id];
688 /* If find slave exited, try to reset pair */
690 if (res->enabled && res->flags && pair->enabled) {
692 master_sendcmd_with_ack(pair->lcore_id, CMD_STOP);
693 rte_spinlock_unlock(&res_lock);
695 rte_spinlock_lock(&res_lock);
700 if (reset_pair(res->lcore_id, pair->lcore_id) != 0)
701 rte_exit(EXIT_FAILURE, "failed to reset slave");
707 rte_spinlock_unlock(&res_lock);
710 When the slave process is spawned and starts to run, it checks whether the floating process option is applied.
711 If so, it clears the affinity to a specific core and also sets the unique core ID to 0.
712 Then, it tries to allocate a new core ID.
713 Since the core ID has changed, the resource allocated by the master cannot work,
714 so it remaps the resource to the new core ID slot.
719 l2fwd_launch_one_lcore( attribute ((unused)) void *dummy)
721 unsigned lcore_id = rte_lcore_id();
726 /* Change it to floating process, also change it's lcore_id */
728 clear_cpu_affinity();
730 RTE_PER_LCORE(_lcore_id) = 0;
734 if (flib_assign_lcore_id() < 0 ) {
735 printf("flib_assign_lcore_id failed\n");
739 flcore_id = rte_lcore_id();
741 /* Set mapping id, so master can return it after slave exited */
743 mapping_id[lcore_id] = flcore_id;
744 printf("Org lcore_id = %u, cur lcore_id = %u\n",lcore_id, flcore_id);
745 remapping_slave_resource(lcore_id, flcore_id);
750 /* return lcore_id before return */
752 flib_free_lcore_id(rte_lcore_id());
753 mapping_id[lcore_id] = INVALID_MAPPING_ID;