1 /* Copyright (c) 2009, 2010. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
12 /* Pimple to get an histogram of message sizes in the simulation */
13 xbt_dict_t msg_sizes = NULL;
15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
16 "Logging specific to SIMIX (network)");
18 /******************************************************************************/
19 /* Rendez-Vous Points */
20 /******************************************************************************/
23 * \brief Creates a new rendez-vous point
24 * \param name The name of the rendez-vous point
25 * \return The created rendez-vous point
27 smx_rdv_t SIMIX_rdv_create(const char *name)
29 smx_rdv_t rdv = xbt_new0(s_smx_rvpoint_t, 1);
30 rdv->name = name ? xbt_strdup(name) : NULL;
31 rdv->read = SIMIX_mutex_init();
32 rdv->write = SIMIX_mutex_init();
33 rdv->comm_fifo = xbt_fifo_new();
39 * \brief Destroy a rendez-vous point
40 * \param name The rendez-vous point to destroy
42 void SIMIX_rdv_destroy(smx_rdv_t rdv)
46 SIMIX_mutex_destroy(rdv->read);
47 SIMIX_mutex_destroy(rdv->write);
48 xbt_fifo_free(rdv->comm_fifo);
53 * \brief Push a communication request into a rendez-vous point
54 * \param rdv The rendez-vous point
55 * \param comm The communication request
57 static inline void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm)
59 xbt_fifo_push(rdv->comm_fifo, comm);
64 * \brief Remove a communication request from a rendez-vous point
65 * \param rdv The rendez-vous point
66 * \param comm The communication request
68 static inline void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm)
70 xbt_fifo_remove(rdv->comm_fifo, comm);
75 * \brief Checks if there is a communication request queued in a rendez-vous matching our needs
76 * \param type The type of communication we are looking for (comm_send, comm_recv)
77 * \return The communication request if found, NULL otherwise.
79 smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type) {
80 smx_comm_t comm = (smx_comm_t)xbt_fifo_get_item_content(
81 xbt_fifo_get_first_item(rdv->comm_fifo));
83 if(comm && comm->type == type){
84 DEBUG0("Communication request found!");
85 xbt_fifo_shift(rdv->comm_fifo);
86 SIMIX_communication_use(comm);
91 DEBUG0("Communication request not found");
96 * \brief counts the number of communication requests of a given host pending
97 * on a rendez-vous point
98 * \param rdv The rendez-vous point
99 * \param host The host to be counted
100 * \return The number of comm request pending in the rdv
103 SIMIX_rdv_get_count_waiting_comm(smx_rdv_t rdv, smx_host_t host)
105 smx_comm_t comm = NULL;
106 xbt_fifo_item_t item = NULL;
109 xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_comm_t) {
110 if (comm->src_proc->smx_host == host)
118 * \brief returns the communication at the head of the rendez-vous
119 * \param rdv The rendez-vous point
120 * \return The communication or NULL if empty
122 XBT_INLINE smx_comm_t SIMIX_rdv_get_head(smx_rdv_t rdv)
124 return (smx_comm_t)xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
127 /** @brief adds some API-related data to the rendez-vous point */
128 XBT_INLINE void SIMIX_rdv_set_data(smx_rdv_t rdv,void *data) {
131 /** @brief gets API-related data from the rendez-vous point */
132 XBT_INLINE void *SIMIX_rdv_get_data(smx_rdv_t rdv) {
136 /******************************************************************************/
137 /* Communication Requests */
138 /******************************************************************************/
141 * \brief Creates a new communication request
142 * \param type The type of communication (comm_send, comm_recv)
143 * \return The new communication request
145 smx_comm_t SIMIX_communication_new(smx_comm_type_t type)
147 /* alloc structures */
148 smx_comm_t comm = xbt_new0(s_smx_comm_t, 1);
150 comm->sem = SIMIX_sem_init(0);
157 * \brief Destroy a communication request
158 * \param comm The request to be destroyed
160 void SIMIX_communication_destroy(smx_comm_t comm)
163 if(comm->refcount > 0)
167 SIMIX_sem_destroy(comm->sem);
172 SIMIX_action_destroy(comm->act);
176 if(comm->src_timeout){
177 SIMIX_action_destroy(comm->src_timeout);
178 comm->src_timeout = NULL;
181 if(comm->dst_timeout){
182 SIMIX_action_destroy(comm->dst_timeout);
183 comm->dst_timeout = NULL;
190 * \brief Increase the number of users of the communication.
191 * \param comm The communication request
192 * Each communication request can be used by more than one process, so it is
193 * necessary to know number of them at destroy time, to avoid freeing stuff that
194 * maybe is in use by others.
197 static inline void SIMIX_communication_use(smx_comm_t comm)
203 * \brief Start the simulation of a communication request
204 * \param comm The communication request
206 static inline void SIMIX_communication_start(smx_comm_t comm)
208 /* If both the sender and the receiver are already there, start the communication */
209 if(comm->src_proc && comm->dst_proc){
210 DEBUG1("Starting communication %p", comm);
211 comm->act = SIMIX_action_communicate(comm->src_proc->smx_host,
212 comm->dst_proc->smx_host, NULL,
213 comm->task_size, comm->rate);
215 TRACE_smx_action_communicate (comm->act, comm->src_proc);
218 /* If any of the process is suspend, create the action but stop its execution,
219 it will be restarted when the sender process resume */
220 if(SIMIX_process_is_suspended(comm->src_proc) ||
221 SIMIX_process_is_suspended(comm->dst_proc)) {
222 SIMIX_action_suspend(comm->act);
225 /* Add the communication as user data of the action */
226 comm->act->data = comm;
228 /* The semaphore will only get signaled once, but since the first unlocked guy will
229 * release_forever() the semaphore, that will unlock the second (and any other)
230 * communication partner */
231 SIMIX_register_action_to_semaphore(comm->act, comm->sem);
236 * \brief Waits for communication completion and performs error checking
237 * \param comm The communication
238 * \param timeout The max amount of time to wait for the communication to finish
241 * - host_error if local peer failed
242 * - timeout_error if communication reached the timeout specified (either because of local peer or remote peer)
243 * - network_error if network failed or remote peer failed
245 static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
247 smx_action_t act_sleep = NULL;
251 DEBUG1("Waiting for the completion of communication %p", comm);
254 act_sleep = SIMIX_action_sleep(SIMIX_host_self(), timeout);
255 if(SIMIX_process_self()==comm->src_proc)
256 comm->src_timeout = act_sleep;
258 comm->dst_timeout = act_sleep;
259 SIMIX_action_set_name(act_sleep,bprintf("Timeout for comm %p and wait on semaphore %p (max_duration:%f)", comm, comm->sem,timeout));
260 SIMIX_register_action_to_semaphore(act_sleep, comm->sem);
261 SIMIX_process_self()->waiting_action = act_sleep;
262 SIMIX_sem_block_onto(comm->sem);
263 SIMIX_process_self()->waiting_action = NULL;
264 SIMIX_unregister_action_to_semaphore(act_sleep, comm->sem);
266 SIMIX_sem_acquire(comm->sem);
269 /* Check for timeouts */
270 if ((src_timeout = ((comm->src_timeout) && (SIMIX_action_get_state(comm->src_timeout) == SURF_ACTION_DONE))) ||
271 (dst_timeout = ((comm->dst_timeout) && (SIMIX_action_get_state(comm->dst_timeout) == SURF_ACTION_DONE))) ) {
272 /* Somebody did a timeout! */
273 if (src_timeout) DEBUG1("Communication timeout from the src! %p", comm);
274 if (dst_timeout) DEBUG1("Communication timeout from the dst! %p", comm);
276 if(comm->act && SIMIX_action_get_state(comm->act) == SURF_ACTION_RUNNING)
277 SIMIX_communication_cancel(comm);
279 SIMIX_rdv_remove(comm->rdv, comm);
281 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
282 SIMIX_sem_release_forever(comm->sem);
283 SIMIX_communication_destroy(comm);
285 THROW1(timeout_error, 0, "Communication timeouted because of %s",src_timeout?"the source":"the destination");
288 DEBUG1("Communication %p complete! Let's check for errors", comm);
290 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
291 SIMIX_sem_release_forever(comm->sem);
293 /* Check for errors other than timeouts (they are catched above) */
294 if (!SIMIX_host_get_state(SIMIX_host_self())){
296 SIMIX_rdv_remove(comm->rdv, comm);
297 SIMIX_communication_destroy(comm);
298 THROW0(host_error, 0, "Host failed");
299 } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
300 SIMIX_communication_destroy(comm);
301 THROW0(network_error, 0, "Link failure");
302 } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc)) ||
303 !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc))) {
304 /* We test both src&dst because we dunno who we are today, and we already tested myself above.
305 * So, at the end, we test the remote peer only
306 * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed.
307 * In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem)
308 * at the beginning of this function. */
309 SIMIX_communication_destroy(comm);
310 THROW0(network_error, 0, "Remote peer failed");
313 SIMIX_communication_destroy(comm);
317 * \brief Cancels a communication
318 * \brief comm The communication to cancel
320 XBT_INLINE void SIMIX_communication_cancel(smx_comm_t comm)
323 SIMIX_action_cancel(comm->act);
327 * \brief get the amount remaining from the communication
328 * \param comm The communication
330 XBT_INLINE double SIMIX_communication_get_remains(smx_comm_t comm)
332 return SIMIX_action_get_remains(comm->act);
335 /******************************************************************************/
336 /* SIMIX_network_copy_data callbacks */
337 /******************************************************************************/
338 static void (*SIMIX_network_copy_data_callback)(smx_comm_t, size_t) = &SIMIX_network_copy_pointer_callback;
340 void SIMIX_network_set_copy_data_callback(void (*callback)(smx_comm_t, size_t)) {
341 SIMIX_network_copy_data_callback = callback;
344 void SIMIX_network_copy_pointer_callback(smx_comm_t comm, size_t buff_size) {
345 xbt_assert1((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)",buff_size);
346 *(void**)(comm->dst_buff) = comm->src_buff;
349 void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size) {
350 memcpy(comm->dst_buff, comm->src_buff, buff_size);
354 * \brief Copy the communication data from the sender's buffer to the receiver's one
355 * \param comm The communication
357 void SIMIX_network_copy_data(smx_comm_t comm)
359 size_t buff_size = comm->src_buff_size;
361 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
363 comm->src_proc->smx_host->name, comm->src_buff,
364 comm->dst_proc->smx_host->name, comm->dst_buff,
367 /* If there is no data to be copy then return */
368 if(!comm->src_buff || !comm->dst_buff)
371 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
372 if (comm->dst_buff_size)
373 buff_size = MIN(buff_size,*(comm->dst_buff_size));
375 /* Update the receiver's buffer size to the copied amount */
376 if (comm->dst_buff_size)
377 *comm->dst_buff_size = buff_size;
381 (*SIMIX_network_copy_data_callback)(comm, buff_size);
383 /* pimple to display the message sizes */
385 if (msg_sizes == NULL)
386 msg_sizes = xbt_dict_new();
387 uintptr_t casted_size = comm->task_size;
388 uintptr_t amount = xbt_dicti_get(msg_sizes, casted_size);
391 xbt_dicti_set(msg_sizes,casted_size, amount);
395 /* pimple to display the message sizes */
396 void SIMIX_message_sizes_output(const char *filename) {
397 FILE * out = fopen(filename,"w");
398 xbt_assert1(out,"Cannot open file %s",filename);
400 xbt_dict_cursor_t cursor;
401 xbt_dict_foreach(msg_sizes,cursor,key,data) {
402 fprintf(out,"%zu %zu\n",key,data);
408 * \brief Return the user data associated to the communication
409 * \param comm The communication
410 * \return the user data
412 XBT_INLINE void *SIMIX_communication_get_data(smx_comm_t comm)
417 /******************************************************************************/
418 /* Synchronous Communication */
419 /******************************************************************************/
421 * \brief Put a send communication request in a rendez-vous point and waits for
422 * its completion (blocking)
423 * \param rdv The rendez-vous point
424 * \param task_size The size of the communication action (for surf simulation)
425 * \param rate The rate of the communication action (for surf)
426 * \param timeout The timeout used for the waiting the completion
427 * \param src_buff The source buffer containing the message to be sent
428 * \param src_buff_size The size of the source buffer
429 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
430 * \param data User data associated to the communication object
432 * - host_error if peer failed
433 * - timeout_error if communication reached the timeout specified
434 * - network_error if network failed or peer issued a timeout
436 XBT_INLINE void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate,
437 double timeout, void *src_buff, size_t src_buff_size,
438 smx_comm_t *comm_ref, void *data)
440 *comm_ref = SIMIX_network_isend(rdv,task_size,rate,src_buff,src_buff_size,data);
441 SIMIX_network_wait(*comm_ref,timeout);
445 * \brief Put a receive communication request in a rendez-vous point and waits
446 * for its completion (blocking)
447 * \param rdv The rendez-vous point
448 * \param timeout The timeout used for the waiting the completion
449 * \param dst_buff The destination buffer to copy the received message
450 * \param src_buff_size The size of the destination buffer
451 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
453 * - host_error if peer failed
454 * - timeout_error if communication reached the timeout specified
455 * - network_error if network failed or peer issued a timeout
457 XBT_INLINE void SIMIX_network_recv(smx_rdv_t rdv, double timeout, void *dst_buff,
458 size_t *dst_buff_size, smx_comm_t *comm_ref)
460 *comm_ref = SIMIX_network_irecv(rdv,dst_buff,dst_buff_size);
461 SIMIX_network_wait(*comm_ref,timeout);
464 /******************************************************************************/
465 /* Asynchronous Communication */
466 /******************************************************************************/
467 smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate,
468 void *src_buff, size_t src_buff_size, void *data)
471 mc_transition_t trans=NULL;
473 if (_surf_do_model_check) {
474 /* Let's intercept the communication and control it from the model-checker */
475 trans = MC_create_transition(mc_isend, SIMIX_process_self(), rdv, NULL);
476 SIMIX_process_yield();
479 /* Look for communication request matching our needs.
480 If it is not found then create it and push it into the rendez-vous point */
481 comm = SIMIX_rdv_get_request(rdv, comm_recv);
484 comm = SIMIX_communication_new(comm_send);
485 SIMIX_rdv_push(rdv, comm);
488 /* Setup the communication request */
489 comm->src_proc = SIMIX_process_self();
490 comm->task_size = task_size;
492 comm->src_buff = src_buff;
493 comm->src_buff_size = src_buff_size;
496 /* Associate the simix communication to the mc transition */
497 if (_surf_do_model_check)
498 MC_transition_set_comm(trans, comm);
500 SIMIX_communication_start(comm);
504 smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_size) {
506 mc_transition_t trans=NULL;
508 if (_surf_do_model_check) {
509 /* Let's intercept the communication and control it from the model-checker */
510 trans = MC_create_transition(mc_irecv, SIMIX_process_self(), rdv, NULL);
511 SIMIX_process_yield();
514 /* Look for communication request matching our needs.
515 If it is not found then create it and push it into the rendez-vous point */
516 comm = SIMIX_rdv_get_request(rdv, comm_send);
519 comm = SIMIX_communication_new(comm_recv);
520 SIMIX_rdv_push(rdv, comm);
523 /* Setup communication request */
524 comm->dst_proc = SIMIX_process_self();
525 comm->dst_buff = dst_buff;
526 comm->dst_buff_size = dst_buff_size;
528 /* Associate the simix communication to the mc transition */
529 if (_surf_do_model_check)
530 MC_transition_set_comm(trans, comm);
532 SIMIX_communication_start(comm);
536 /** @brief blocks until the communication terminates or the timeout occurs */
537 XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout) {
538 if (_surf_do_model_check) {
539 /* Let's intercept the communication and control it from the model-checker */
540 MC_create_transition(mc_wait, SIMIX_process_self(), comm->rdv, comm);
541 SIMIX_process_yield();
543 /* Wait for communication completion */
544 SIMIX_communication_wait_for_completion(comm, timeout);
547 /** @Returns whether the (asynchronous) communication is done yet or not */
548 XBT_INLINE int SIMIX_network_test(smx_comm_t comm) {
549 if (_surf_do_model_check) {
550 /* Let's intercept the communication and control it from the model-checker */
551 MC_create_transition(mc_test, SIMIX_process_self(), comm->rdv, comm);
552 SIMIX_process_yield();
554 return comm->sem?SIMIX_sem_would_block(comm->sem):0;
557 /** @brief wait for the completion of any communication of a set
559 * @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
561 unsigned int SIMIX_network_waitany(xbt_dynar_t comms) {
562 xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t),NULL);
563 unsigned int cursor, found_comm=-1;
564 smx_comm_t comm,comm_finished=NULL;
566 if (_surf_do_model_check) {
567 /* Let's intercept the communication and control it from the model-checker */
568 MC_create_transition(mc_waitany, SIMIX_process_self(), NULL, NULL);
569 SIMIX_process_yield();
572 xbt_dynar_foreach(comms,cursor,comm)
573 xbt_dynar_push(sems,&(comm->sem));
575 DEBUG1("Waiting for the completion of communication set %p", comms);
577 found_comm = SIMIX_sem_acquire_any(sems);
578 xbt_dynar_free_container(&sems);
579 xbt_assert0(found_comm!=-1,"Cannot find which communication finished");
580 xbt_dynar_get_cpy(comms,found_comm,&comm_finished);
582 /* Check for errors and cleanup the comm */
583 SIMIX_communication_wait_for_completion(comm_finished,-1);