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 XBT_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 XBT_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 XBT_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 XBT_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 Performs error checking and cleanup
237 * \param comm The communication
239 static XBT_INLINE void SIMIX_communication_cleanup(smx_comm_t comm)
241 DEBUG1("Checking errors and cleaning communication %p", comm);
243 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
244 SIMIX_sem_release_forever(comm->sem);
246 /* Check for errors other than timeouts */
247 if (!SIMIX_host_get_state(SIMIX_host_self())){
249 SIMIX_rdv_remove(comm->rdv, comm);
250 SIMIX_communication_destroy(comm);
251 THROW0(host_error, 0, "Host failed");
252 } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
253 SIMIX_communication_destroy(comm);
254 THROW0(network_error, 0, "Link failure");
255 } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc)) ||
256 !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc))) {
257 /* We test both src&dst because we dunno who we are today, and we already tested myself above.
258 * So, at the end, we test the remote peer only
259 * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed.
260 * In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem)
261 * at the beginning of this function. */
262 SIMIX_communication_destroy(comm);
263 THROW0(network_error, 0, "Remote peer failed");
266 /* Copy network data */
267 SIMIX_network_copy_data(comm);
269 SIMIX_communication_destroy(comm);
273 * \brief Waits for communication completion
274 * \param comm The communication
275 * \param timeout The max amount of time to wait for the communication to finish
278 * - host_error if local peer failed
279 * - timeout_error if communication reached the timeout specified (either because of local peer or remote peer)
280 * - network_error if network failed or remote peer failed
282 static XBT_INLINE void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
284 smx_action_t act_sleep = NULL;
288 DEBUG1("Waiting for the completion of communication %p", comm);
291 act_sleep = SIMIX_action_sleep(SIMIX_host_self(), timeout);
292 if(SIMIX_process_self()==comm->src_proc)
293 comm->src_timeout = act_sleep;
295 comm->dst_timeout = act_sleep;
296 SIMIX_action_set_name(act_sleep,bprintf("Timeout for comm %p and wait on semaphore %p (max_duration:%f)", comm, comm->sem,timeout));
297 SIMIX_register_action_to_semaphore(act_sleep, comm->sem);
298 SIMIX_process_self()->waiting_action = act_sleep;
299 SIMIX_sem_block_onto(comm->sem);
300 SIMIX_process_self()->waiting_action = NULL;
301 SIMIX_unregister_action_to_semaphore(act_sleep, comm->sem);
303 SIMIX_sem_acquire(comm->sem);
306 /* Check for timeouts */
307 if ((src_timeout = ((comm->src_timeout) && (SIMIX_action_get_state(comm->src_timeout) == SURF_ACTION_DONE))) ||
308 (dst_timeout = ((comm->dst_timeout) && (SIMIX_action_get_state(comm->dst_timeout) == SURF_ACTION_DONE))) ) {
309 /* Somebody did a timeout! */
310 if (src_timeout) DEBUG1("Communication timeout from the src! %p", comm);
311 if (dst_timeout) DEBUG1("Communication timeout from the dst! %p", comm);
313 if(comm->act && SIMIX_action_get_state(comm->act) == SURF_ACTION_RUNNING)
314 SIMIX_communication_cancel(comm);
316 SIMIX_rdv_remove(comm->rdv, comm);
318 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
319 SIMIX_sem_release_forever(comm->sem);
320 SIMIX_communication_destroy(comm);
322 THROW1(timeout_error, 0, "Communication timeouted because of %s",src_timeout?"the source":"the destination");
325 DEBUG1("Communication %p complete!", comm);
326 SIMIX_communication_cleanup(comm);
330 * \brief Cancels a communication
331 * \brief comm The communication to cancel
333 XBT_INLINE void SIMIX_communication_cancel(smx_comm_t comm)
336 SIMIX_action_cancel(comm->act);
340 * \brief get the amount remaining from the communication
341 * \param comm The communication
343 XBT_INLINE double SIMIX_communication_get_remains(smx_comm_t comm)
345 return SIMIX_action_get_remains(comm->act);
348 /******************************************************************************/
349 /* SIMIX_network_copy_data callbacks */
350 /******************************************************************************/
351 static void (*SIMIX_network_copy_data_callback)(smx_comm_t, size_t) = &SIMIX_network_copy_pointer_callback;
353 void SIMIX_network_set_copy_data_callback(void (*callback)(smx_comm_t, size_t)) {
354 SIMIX_network_copy_data_callback = callback;
357 void SIMIX_network_copy_pointer_callback(smx_comm_t comm, size_t buff_size) {
358 xbt_assert1((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)",buff_size);
359 *(void**)(comm->dst_buff) = comm->src_buff;
362 void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size) {
363 memcpy(comm->dst_buff, comm->src_buff, buff_size);
367 * \brief Copy the communication data from the sender's buffer to the receiver's one
368 * \param comm The communication
370 void SIMIX_network_copy_data(smx_comm_t comm)
372 size_t buff_size = comm->src_buff_size;
373 uintptr_t casted_size = 0;
374 uintptr_t amount = 0;
375 /* If there is no data to be copy then return */
376 if(!comm->src_buff || !comm->dst_buff || comm->copied == 1)
379 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
381 comm->src_proc->smx_host->name, comm->src_buff,
382 comm->dst_proc->smx_host->name, comm->dst_buff,
385 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
386 if (comm->dst_buff_size)
387 buff_size = MIN(buff_size,*(comm->dst_buff_size));
389 /* Update the receiver's buffer size to the copied amount */
390 if (comm->dst_buff_size)
391 *comm->dst_buff_size = buff_size;
395 (*SIMIX_network_copy_data_callback)(comm, buff_size);
397 /* Set the copied flag so we copy data only once */
398 /* (this function might be called from both communication ends)*/
401 /* pimple to display the message sizes */
403 if (msg_sizes == NULL)
404 msg_sizes = xbt_dict_new();
405 casted_size = comm->task_size;
406 amount = xbt_dicti_get(msg_sizes, casted_size);
409 xbt_dicti_set(msg_sizes,casted_size, amount);
413 /* pimple to display the message sizes */
414 void SIMIX_message_sizes_output(const char *filename) {
417 xbt_dict_cursor_t cursor;
419 out = fopen(filename,"w");
420 xbt_assert1(out,"Cannot open file %s",filename);
422 xbt_dict_foreach(msg_sizes,cursor,key,data) {
423 fprintf(out,"%zu %zu\n",key,data);
429 * \brief Return the user data associated to the communication
430 * \param comm The communication
431 * \return the user data
433 XBT_INLINE void *SIMIX_communication_get_data(smx_comm_t comm)
438 /******************************************************************************/
439 /* Synchronous Communication */
440 /******************************************************************************/
442 * \brief Put a send communication request in a rendez-vous point and waits for
443 * its completion (blocking)
444 * \param rdv The rendez-vous point
445 * \param task_size The size of the communication action (for surf simulation)
446 * \param rate The rate of the communication action (for surf)
447 * \param timeout The timeout used for the waiting the completion
448 * \param src_buff The source buffer containing the message to be sent
449 * \param src_buff_size The size of the source buffer
450 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
451 * \param data User data associated to the communication object
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_send(smx_rdv_t rdv, double task_size, double rate,
458 double timeout, void *src_buff, size_t src_buff_size,
459 smx_comm_t *comm_ref, void *data)
461 *comm_ref = SIMIX_network_isend(rdv,task_size,rate,src_buff,src_buff_size,data);
462 SIMIX_network_wait(*comm_ref,timeout);
466 * \brief Put a receive communication request in a rendez-vous point and waits
467 * for its completion (blocking)
468 * \param rdv The rendez-vous point
469 * \param timeout The timeout used for the waiting the completion
470 * \param dst_buff The destination buffer to copy the received message
471 * \param src_buff_size The size of the destination buffer
472 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
474 * - host_error if peer failed
475 * - timeout_error if communication reached the timeout specified
476 * - network_error if network failed or peer issued a timeout
478 XBT_INLINE void SIMIX_network_recv(smx_rdv_t rdv, double timeout, void *dst_buff,
479 size_t *dst_buff_size, smx_comm_t *comm_ref)
481 *comm_ref = SIMIX_network_irecv(rdv,dst_buff,dst_buff_size);
482 SIMIX_network_wait(*comm_ref,timeout);
485 /******************************************************************************/
486 /* Asynchronous Communication */
487 /******************************************************************************/
488 smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate,
489 void *src_buff, size_t src_buff_size, void *data)
493 /*If running in model-checking mode then intercept the communication action*/
495 if (_surf_do_model_check)
496 MC_trans_intercept_isend(rdv);
498 /* Look for communication request matching our needs.
499 If it is not found then create it and push it into the rendez-vous point */
500 comm = SIMIX_rdv_get_request(rdv, comm_recv);
503 comm = SIMIX_communication_new(comm_send);
504 SIMIX_rdv_push(rdv, comm);
507 /* Setup the communication request */
508 comm->src_proc = SIMIX_process_self();
509 comm->task_size = task_size;
511 comm->src_buff = src_buff;
512 comm->src_buff_size = src_buff_size;
515 SIMIX_communication_start(comm);
519 smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_size)
523 /*If running in model-checking mode then intercept the communication action*/
525 if (_surf_do_model_check)
526 MC_trans_intercept_irecv(rdv);
528 /* Look for communication request matching our needs.
529 * If it is not found then create it and push it into the rendez-vous point
531 comm = SIMIX_rdv_get_request(rdv, comm_send);
534 comm = SIMIX_communication_new(comm_recv);
535 SIMIX_rdv_push(rdv, comm);
538 /* Setup communication request */
539 comm->dst_proc = SIMIX_process_self();
540 comm->dst_buff = dst_buff;
541 comm->dst_buff_size = dst_buff_size;
543 SIMIX_communication_start(comm);
547 /** @brief blocks until the communication terminates or the timeout occurs */
548 XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout)
550 /*If running in model-checking mode then intercept the communication action*/
552 if (_surf_do_model_check)
553 MC_trans_intercept_wait(comm);
555 /* Wait for communication completion */
556 SIMIX_communication_wait_for_completion(comm, timeout);
559 /** @Returns whether the (asynchronous) communication is done yet or not */
560 XBT_INLINE int SIMIX_network_test(smx_comm_t comm)
562 /*If running in model-checking mode then intercept the communication action*/
564 if (_surf_do_model_check)
565 MC_trans_intercept_test(comm);
568 /* Copy data if the communication is done */
569 if(comm->sem && !SIMIX_sem_would_block(comm->sem)){
570 /* Copy network data */
571 SIMIX_network_copy_data(comm);
577 /** @brief wait for the completion of any communication of a set
579 * @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
581 unsigned int SIMIX_network_waitany(xbt_dynar_t comms)
583 xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t),NULL);
584 unsigned int cursor, found_comm=-1;
585 smx_comm_t comm,comm_finished=NULL;
587 /*If running in model-checking mode then intercept the communication action*/
589 if (_surf_do_model_check)
590 MC_trans_intercept_waitany(comms);
592 xbt_dynar_foreach(comms,cursor,comm)
593 xbt_dynar_push(sems,&(comm->sem));
595 DEBUG1("Waiting for the completion of communication set %p", comms);
597 found_comm = SIMIX_sem_acquire_any(sems);
598 xbt_dynar_free_container(&sems);
599 xbt_assert0(found_comm!=-1,"Cannot find which communication finished");
600 xbt_dynar_get_cpy(comms,found_comm,&comm_finished);
602 /* Check for errors and cleanup the comm */
603 SIMIX_communication_cleanup(comm_finished);