3 /* Copyright (c) 2009 Cristian Rosa.
4 All rights reserved. */
6 /* This program is free software; you can redistribute it and/or modify it
7 * under the terms of the license (GNU LGPL) which comes with this package. */
12 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
13 "Logging specific to SIMIX (network)");
15 /******************************************************************************/
16 /* Rendez-Vous Points */
17 /******************************************************************************/
20 * \brief Creates a new rendez-vous point
21 * \param name The name of the rendez-vous point
22 * \return The created rendez-vous point
24 smx_rdv_t SIMIX_rdv_create(const char *name)
26 smx_rdv_t rdv = xbt_new0(s_smx_rvpoint_t, 1);
27 rdv->name = name ? xbt_strdup(name) : NULL;
28 rdv->read = SIMIX_mutex_init();
29 rdv->write = SIMIX_mutex_init();
30 rdv->comm_fifo = xbt_fifo_new();
36 * \brief Destroy a rendez-vous point
37 * \param name The rendez-vous point to destroy
39 void SIMIX_rdv_destroy(smx_rdv_t rdv)
43 SIMIX_mutex_destroy(rdv->read);
44 SIMIX_mutex_destroy(rdv->write);
45 xbt_fifo_free(rdv->comm_fifo);
50 * \brief Push a communication request into a rendez-vous point
51 * \param rdv The rendez-vous point
52 * \param comm The communication request
54 static inline void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm)
56 xbt_fifo_push(rdv->comm_fifo, comm);
61 * \brief Remove a communication request from a rendez-vous point
62 * \param rdv The rendez-vous point
63 * \param comm The communication request
65 static inline void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm)
67 xbt_fifo_remove(rdv->comm_fifo, comm);
72 * \brief Checks if there is a communication request queued in a rendez-vous matching our needs
73 * \param type The type of communication we are looking for (comm_send, comm_recv)
74 * \return The communication request if found, NULL otherwise.
76 smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type)
78 smx_comm_t comm = (smx_comm_t)xbt_fifo_get_item_content(
79 xbt_fifo_get_first_item(rdv->comm_fifo));
81 if(comm && comm->type == type){
82 DEBUG0("Communication request found!");
83 xbt_fifo_shift(rdv->comm_fifo);
84 SIMIX_communication_use(comm);
89 /* no relevant request found. Return NULL */
90 DEBUG0("Communication request not found. I assume that other side will arrive later on.");
95 * \brief counts the number of communication requests of a given host pending
96 * on a rendez-vous point
97 * \param rdv The rendez-vous point
98 * \param host The host to be counted
99 * \return The number of comm request pending in the rdv
102 SIMIX_rdv_get_count_waiting_comm(smx_rdv_t rdv, smx_host_t host)
104 smx_comm_t comm = NULL;
105 xbt_fifo_item_t item = NULL;
108 xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_comm_t) {
109 if (comm->src_proc->smx_host == host)
117 * \brief returns the communication at the head of the rendez-vous
118 * \param rdv The rendez-vous point
119 * \return The communication or NULL if empty
121 smx_comm_t SIMIX_rdv_get_head(smx_rdv_t rdv)
123 return (smx_comm_t)xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
126 /** @brief adds some API-related data to the rendez-vous point */
127 void SIMIX_rdv_set_data(smx_rdv_t rdv,void *data) {
130 /** @brief gets API-related data from the rendez-vous point */
131 void *SIMIX_rdv_get_data(smx_rdv_t rdv) {
135 /******************************************************************************/
136 /* Communication Requests */
137 /******************************************************************************/
140 * \brief Creates a new communication request
141 * \param type The type of communication (comm_send, comm_recv)
142 * \return The new communication request
144 smx_comm_t SIMIX_communication_new(smx_comm_type_t type)
146 /* alloc structures */
147 smx_comm_t comm = xbt_new0(s_smx_comm_t, 1);
149 comm->sem = SIMIX_sem_init(0);
156 * \brief Destroy a communication request
157 * \param comm The request to be destroyed
159 void SIMIX_communication_destroy(smx_comm_t comm)
162 if(comm->refcount > 0)
166 SIMIX_sem_destroy(comm->sem);
171 SIMIX_action_destroy(comm->act);
179 * \brief Increase the number of users of the communication.
180 * \param comm The communication request
181 * Each communication request can be used by more than one process, so it is
182 * necessary to know number of them at destroy time, to avoid freeing stuff that
183 * maybe is in use by others.
186 static inline void SIMIX_communication_use(smx_comm_t comm)
192 * \brief Start the simulation of a communication request
193 * \param comm The comm->rdv = NULL;communication request
195 static inline void SIMIX_communication_start(smx_comm_t comm)
197 /* If both the sender and the receiver are already there, start the communication */
198 if(comm->src_proc && comm->dst_proc){
199 DEBUG1("Starting communication %p", comm);
200 comm->act = SIMIX_action_communicate(comm->src_proc->smx_host,
201 comm->dst_proc->smx_host, NULL,
202 comm->task_size, comm->rate);
204 /* If any of the process is suspend, create the action but stop its execution,
205 it will be restarted when the sender process resume */
206 if(SIMIX_process_is_suspended(comm->src_proc) ||
207 SIMIX_process_is_suspended(comm->dst_proc)) {
208 SIMIX_action_set_priority(comm->act, 0);
211 /* Add the communication as user data of the action */
212 comm->act->data = comm;
214 /* The semaphore will only get signaled once, but since the first unlocked guy will
215 * release_forever() the semaphore, that will unlock the second (and any other)
216 * communication partner */
217 SIMIX_register_action_to_semaphore(comm->act, comm->sem);
222 * \brief Waits for communication completion and performs error checking
223 * \param comm The communication
224 * \param timeout The max amount of time to wait for the communication to finish
227 * - host_error if peer failed
228 * - timeout_error if communication reached the timeout specified
229 * - network_error if network failed or peer issued a timeout
231 static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
235 DEBUG1("Waiting for the completion of communication %p", comm);
239 SIMIX_sem_acquire_timeout(comm->sem, timeout);
242 /* If there is a timeout then cancel the communication if it is running or
243 remove it from the rendez-vous otherwise. Then signal the other peer,
244 destroy the communication and retrow the exception. */
245 if(e.category == timeout_error){
246 DEBUG1("Communication timeout! %p", comm);
247 if(comm->act && SIMIX_action_get_state(comm->act) == SURF_ACTION_RUNNING)
248 SIMIX_communication_cancel(comm);
250 SIMIX_rdv_remove(comm->rdv, comm);
252 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
253 SIMIX_sem_release_forever(comm->sem);
254 SIMIX_communication_destroy(comm);
259 SIMIX_sem_acquire(comm->sem);
262 DEBUG1("Communication %p complete! Let's check for errors", comm);
264 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
265 SIMIX_sem_release_forever(comm->sem);
267 /* Check for errors other than timeouts (they are catched above) */
268 if(!SIMIX_host_get_state(SIMIX_host_self())){
270 SIMIX_rdv_remove(comm->rdv, comm);
271 SIMIX_communication_destroy(comm);
272 THROW0(host_error, 0, "Host failed");
273 } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
274 SIMIX_communication_destroy(comm);
275 THROW0(network_error, 0, "Link failure");
280 * \brief Cancels a communication
281 * \brief comm The communication to cancel
283 void SIMIX_communication_cancel(smx_comm_t comm)
285 SIMIX_action_cancel(comm->act);
289 * \brief get the amount remaining from the communication
290 * \param comm The communication
292 double SIMIX_communication_get_remains(smx_comm_t comm)
294 return SIMIX_action_get_remains(comm->act);
298 * \brief Copy the communication data from the sender's buffer to the receiver's one
299 * \param comm The communication
301 void SIMIX_network_copy_data(smx_comm_t comm)
303 size_t src_buff_size = comm->src_buff_size;
304 size_t dst_buff_size = *comm->dst_buff_size;
306 xbt_assert(src_buff_size == dst_buff_size);
308 /* If there is no data to copy then return */
309 if(!comm->src_buff || !comm->dst_buff || dst_buff_size == 0)
312 memcpy(comm->dst_buff, comm->src_buff, dst_buff_size);
314 DEBUG6("Copying comm %p data from %s -> %s (%zu bytes %p->%p)",
315 comm, comm->src_proc->smx_host->name, comm->dst_proc->smx_host->name,
316 dst_buff_size,comm->src_buff,comm->dst_buff);
320 * \brief Return the user data associated to the communication
322 * In MSG and GRAS, that data is the exchanged task/msg itself, since
323 * (i) In MSG, the receiver still wants to read the task although the communication didn't complete.
324 * (ii) In GRAS, we need to retrieve that gras_msg_t during the select
325 * \param comm The communication
326 * \return the user data
328 void *SIMIX_communication_get_sentdata(smx_comm_t comm)
333 /******************************************************************************/
334 /* Synchronous Communication */
335 /******************************************************************************/
337 * \brief Put a send communication request in a rendez-vous point and waits for
338 * its completion (blocking)
339 * \param rdv The rendez-vous point
340 * \param task_size The size of the communication action (for surf simulation)
341 * \param rate The rate of the communication action (for surf)
342 * \param timeout The timeout used for the waiting the completion
343 * \param src_buff The source buffer containing the message to be sent
344 * \param src_buff_size The size of the source buffer
345 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
346 * \param data User data associated to the communication object
348 * - host_error if peer failed
349 * - timeout_error if communication reached the timeout specified
350 * - network_error if network failed or peer issued a timeout
352 void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate,
353 double timeout, void *src_buff, size_t src_buff_size,
354 smx_comm_t *comm_ref, void *data)
356 *comm_ref = SIMIX_network_isend(rdv,task_size,rate,src_buff,src_buff_size,data);
357 SIMIX_network_wait(*comm_ref,timeout);
361 * \brief Put a receive communication request in a rendez-vous point and waits
362 * for its completion (blocking)
363 * \param rdv The rendez-vous point
364 * \param timeout The timeout used for the waiting the completion
365 * \param dst_buff The destination buffer to copy the received message
366 * \param src_buff_size The size of the destination buffer
367 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
369 * - host_error if peer failed
370 * - timeout_error if communication reached the timeout specified
371 * - network_error if network failed or peer issued a timeout
373 void SIMIX_network_recv(smx_rdv_t rdv, double timeout, void *dst_buff,
374 size_t *dst_buff_size, smx_comm_t *comm_ref)
376 *comm_ref = SIMIX_network_irecv(rdv,dst_buff,dst_buff_size);
377 SIMIX_network_wait(*comm_ref,timeout);
380 /******************************************************************************/
381 /* Asynchronous Communication */
382 /******************************************************************************/
383 smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate,
384 void *src_buff, size_t src_buff_size, void *data)
388 /* Look for communication request matching our needs.
389 If it is not found then create it and push it into the rendez-vous point */
390 comm = SIMIX_rdv_get_request(rdv, comm_recv);
393 comm = SIMIX_communication_new(comm_send);
394 SIMIX_rdv_push(rdv, comm);
397 /* Setup the communication request */
398 comm->src_proc = SIMIX_process_self();
399 comm->task_size = task_size;
401 comm->src_buff = src_buff;
402 comm->src_buff_size = src_buff_size;
405 SIMIX_communication_start(comm);
409 smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_size) {
412 /* Look for communication request matching our needs.
413 If it is not found then create it and push it into the rendez-vous point */
414 comm = SIMIX_rdv_get_request(rdv, comm_send);
417 comm = SIMIX_communication_new(comm_recv);
418 SIMIX_rdv_push(rdv, comm);
421 /* Setup communication request */
422 comm->dst_proc = SIMIX_process_self();
423 comm->dst_buff = dst_buff;
424 comm->dst_buff_size = dst_buff_size;
426 DEBUG2("Receive data for %p into %p",comm,comm->dst_buff);
427 SIMIX_communication_start(comm);
431 /** @brief blocks until the communication terminates or the timeout occurs */
432 void SIMIX_network_wait(smx_comm_t comm, double timeout) {
433 /* Wait for communication completion */
434 SIMIX_communication_wait_for_completion(comm, timeout);
436 SIMIX_communication_destroy(comm);
439 /** @Returns whether the (asynchronous) communication is done yet or not */
440 int SIMIX_network_test(smx_comm_t comm) {
441 return comm->sem?SIMIX_sem_would_block(comm->sem):0;
444 /** @brief wait for the completion of any communication of a set
446 * @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
448 unsigned int SIMIX_network_waitany(xbt_dynar_t comms) {
449 xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t),NULL);
450 unsigned int cursor, found_comm=-1;
451 smx_comm_t comm,comm_finished=NULL;
453 xbt_dynar_foreach(comms,cursor,comm){
454 xbt_dynar_push(sems,&(comm->sem));
457 DEBUG1("Waiting for the completion of communication set %p", comms);
459 found_comm = SIMIX_sem_acquire_any(sems);
460 xbt_assert0(found_comm!=-1,"Cannot find which communication finished");
461 xbt_dynar_get_cpy(comms,found_comm,&comm_finished);
463 DEBUG1("Communication %p complete! Let's check for errors", comm_finished);
465 /* Make sure that everyone sleeping on that semaphore is awake,
466 * and that nobody will ever block on it */
467 SIMIX_sem_release_forever(comm_finished->sem);
469 /* Check for errors */
470 if(!SIMIX_host_get_state(SIMIX_host_self())){
471 if(comm_finished->rdv)
472 SIMIX_rdv_remove(comm_finished->rdv, comm_finished);
473 SIMIX_communication_destroy(comm_finished);
474 THROW0(host_error, 0, "Host failed");
475 } else if (SIMIX_action_get_state(comm_finished->act) == SURF_ACTION_FAILED){
476 SIMIX_communication_destroy(comm_finished);
477 THROW0(network_error, 0, "Link failure");