1 /* Copyright (c) 2009-2014. 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. */
7 #include "smx_private.h"
12 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
13 "Logging specific to SIMIX (network)");
15 static xbt_dict_t rdv_points = NULL;
16 XBT_EXPORT_NO_IMPORT(unsigned long int) smx_total_comms = 0;
18 static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall);
19 static void SIMIX_comm_copy_data(smx_action_t comm);
20 static smx_action_t SIMIX_comm_new(e_smx_comm_type_t type);
21 static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm);
22 static smx_action_t SIMIX_fifo_probe_comm(xbt_fifo_t fifo, e_smx_comm_type_t type,
23 int (*match_fun)(void *, void *,smx_action_t),
24 void *user_data, smx_action_t my_action);
25 static smx_action_t SIMIX_fifo_get_comm(xbt_fifo_t fifo, e_smx_comm_type_t type,
26 int (*match_fun)(void *, void *,smx_action_t),
27 void *user_data, smx_action_t my_action);
28 static void SIMIX_rdv_free(void *data);
29 static void SIMIX_comm_start(smx_action_t action);
31 void SIMIX_network_init(void)
33 rdv_points = xbt_dict_new_homogeneous(SIMIX_rdv_free);
36 void SIMIX_network_exit(void)
38 xbt_dict_free(&rdv_points);
41 /******************************************************************************/
42 /* Rendez-Vous Points */
43 /******************************************************************************/
45 smx_rdv_t SIMIX_pre_rdv_create(smx_simcall_t simcall, const char *name){
46 return SIMIX_rdv_create(name);
48 smx_rdv_t SIMIX_rdv_create(const char *name)
50 /* two processes may have pushed the same rdv_create simcall at the same time */
51 smx_rdv_t rdv = name ? xbt_dict_get_or_null(rdv_points, name) : NULL;
54 rdv = xbt_new0(s_smx_rvpoint_t, 1);
55 rdv->name = name ? xbt_strdup(name) : NULL;
56 rdv->comm_fifo = xbt_fifo_new();
57 rdv->done_comm_fifo = xbt_fifo_new();
58 rdv->permanent_receiver=NULL;
60 XBT_DEBUG("Creating a mailbox at %p with name %s\n", rdv, name);
63 xbt_dict_set(rdv_points, rdv->name, rdv, NULL);
68 void SIMIX_pre_rdv_destroy(smx_simcall_t simcall, smx_rdv_t rdv){
69 return SIMIX_rdv_destroy(rdv);
71 void SIMIX_rdv_destroy(smx_rdv_t rdv)
74 xbt_dict_remove(rdv_points, rdv->name);
77 void SIMIX_rdv_free(void *data)
79 XBT_DEBUG("rdv free %p", data);
80 smx_rdv_t rdv = (smx_rdv_t) data;
82 xbt_fifo_free(rdv->comm_fifo);
83 xbt_fifo_free(rdv->done_comm_fifo);
88 xbt_dict_t SIMIX_get_rdv_points()
93 smx_rdv_t SIMIX_pre_rdv_get_by_name(smx_simcall_t simcall, const char *name){
94 return SIMIX_rdv_get_by_name(name);
96 smx_rdv_t SIMIX_rdv_get_by_name(const char *name)
98 return xbt_dict_get_or_null(rdv_points, name);
101 int SIMIX_pre_rdv_comm_count_by_host(smx_simcall_t simcall, smx_rdv_t rdv, smx_host_t host){
102 return SIMIX_rdv_comm_count_by_host(rdv, host);
104 int SIMIX_rdv_comm_count_by_host(smx_rdv_t rdv, smx_host_t host)
106 smx_action_t comm = NULL;
107 xbt_fifo_item_t item = NULL;
110 xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_action_t) {
111 if (comm->comm.src_proc->smx_host == host)
118 smx_action_t SIMIX_pre_rdv_get_head(smx_simcall_t simcall, smx_rdv_t rdv){
119 return SIMIX_rdv_get_head(rdv);
121 smx_action_t SIMIX_rdv_get_head(smx_rdv_t rdv)
123 return xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
126 smx_process_t SIMIX_pre_rdv_get_receiver(smx_simcall_t simcall, smx_rdv_t rdv){
127 return SIMIX_rdv_get_receiver(rdv);
130 * \brief get the receiver (process associated to the mailbox)
131 * \param rdv The rendez-vous point
132 * \return process The receiving process (NULL if not set)
134 smx_process_t SIMIX_rdv_get_receiver(smx_rdv_t rdv)
136 return rdv->permanent_receiver;
139 void SIMIX_pre_rdv_set_receiver(smx_simcall_t simcall, smx_rdv_t rdv,
140 smx_process_t process){
141 SIMIX_rdv_set_receiver(rdv, process);
144 * \brief set the receiver of the rendez vous point to allow eager sends
145 * \param rdv The rendez-vous point
146 * \param process The receiving process
148 void SIMIX_rdv_set_receiver(smx_rdv_t rdv, smx_process_t process)
150 rdv->permanent_receiver=process;
154 * \brief Pushes a communication action into a rendez-vous point
155 * \param rdv The rendez-vous point
156 * \param comm The communication action
158 static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm)
160 xbt_fifo_push(rdv->comm_fifo, comm);
161 comm->comm.rdv = rdv;
165 * \brief Removes a communication action from a rendez-vous point
166 * \param rdv The rendez-vous point
167 * \param comm The communication action
169 XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm)
171 xbt_fifo_remove(rdv->comm_fifo, comm);
172 comm->comm.rdv = NULL;
176 * \brief Checks if there is a communication action queued in a fifo matching our needs
177 * \param type The type of communication we are looking for (comm_send, comm_recv)
178 * \return The communication action if found, NULL otherwise
180 smx_action_t SIMIX_fifo_get_comm(xbt_fifo_t fifo, e_smx_comm_type_t type,
181 int (*match_fun)(void *, void *,smx_action_t),
182 void *this_user_data, smx_action_t my_action)
185 xbt_fifo_item_t item;
186 void* other_user_data = NULL;
188 xbt_fifo_foreach(fifo, item, action, smx_action_t) {
189 if (action->comm.type == SIMIX_COMM_SEND) {
190 other_user_data = action->comm.src_data;
191 } else if (action->comm.type == SIMIX_COMM_RECEIVE) {
192 other_user_data = action->comm.dst_data;
194 if (action->comm.type == type &&
195 (!match_fun || match_fun(this_user_data, other_user_data, action)) &&
196 (!action->comm.match_fun || action->comm.match_fun(other_user_data, this_user_data, my_action))) {
197 XBT_DEBUG("Found a matching communication action %p", action);
198 xbt_fifo_remove_item(fifo, item);
199 xbt_fifo_free_item(item);
200 action->comm.refcount++;
202 action->comm.rdv_cpy = action->comm.rdv;
204 action->comm.rdv = NULL;
207 XBT_DEBUG("Sorry, communication action %p does not match our needs:"
208 " its type is %d but we are looking for a comm of type %d (or maybe the filtering didn't match)",
209 action, (int)action->comm.type, (int)type);
211 XBT_DEBUG("No matching communication action found");
217 * \brief Checks if there is a communication action queued in a fifo matching our needs, but leave it there
218 * \param type The type of communication we are looking for (comm_send, comm_recv)
219 * \return The communication action if found, NULL otherwise
221 smx_action_t SIMIX_fifo_probe_comm(xbt_fifo_t fifo, e_smx_comm_type_t type,
222 int (*match_fun)(void *, void *,smx_action_t),
223 void *this_user_data, smx_action_t my_action)
226 xbt_fifo_item_t item;
227 void* other_user_data = NULL;
229 xbt_fifo_foreach(fifo, item, action, smx_action_t) {
230 if (action->comm.type == SIMIX_COMM_SEND) {
231 other_user_data = action->comm.src_data;
232 } else if (action->comm.type == SIMIX_COMM_RECEIVE) {
233 other_user_data = action->comm.dst_data;
235 if (action->comm.type == type &&
236 (!match_fun || match_fun(this_user_data, other_user_data, action)) &&
237 (!action->comm.match_fun || action->comm.match_fun(other_user_data, this_user_data, my_action))) {
238 XBT_DEBUG("Found a matching communication action %p", action);
239 action->comm.refcount++;
243 XBT_DEBUG("Sorry, communication action %p does not match our needs:"
244 " its type is %d but we are looking for a comm of type %d (or maybe the filtering didn't match)",
245 action, (int)action->comm.type, (int)type);
247 XBT_DEBUG("No matching communication action found");
250 /******************************************************************************/
251 /* Communication Actions */
252 /******************************************************************************/
255 * \brief Creates a new communicate action
256 * \param type The direction of communication (comm_send, comm_recv)
257 * \return The new communicate action
259 smx_action_t SIMIX_comm_new(e_smx_comm_type_t type)
263 /* alloc structures */
264 act = xbt_mallocator_get(simix_global->action_mallocator);
266 act->type = SIMIX_ACTION_COMMUNICATE;
267 act->state = SIMIX_WAITING;
269 /* set communication */
270 act->comm.type = type;
271 act->comm.refcount = 1;
272 act->comm.src_data=NULL;
273 act->comm.dst_data=NULL;
276 #ifdef HAVE_LATENCY_BOUND_TRACKING
277 //initialize with unknown value
278 act->latency_limited = -1;
282 act->category = NULL;
285 XBT_DEBUG("Create communicate action %p", act);
292 * \brief Destroy a communicate action
293 * \param action The communicate action to be destroyed
295 void SIMIX_comm_destroy(smx_action_t action)
297 XBT_DEBUG("Destroy action %p (refcount: %d), state: %d",
298 action, action->comm.refcount, (int)action->state);
300 if (action->comm.refcount <= 0) {
301 xbt_backtrace_display_current();
302 xbt_die("The refcount of comm %p is already 0 before decreasing it. "
303 "That's a bug! If you didn't test and/or wait the same communication twice in your code, then the bug is SimGrid's...", action);
305 action->comm.refcount--;
306 if (action->comm.refcount > 0)
308 XBT_DEBUG("Really free communication %p; refcount is now %d", action,
309 action->comm.refcount);
311 #ifdef HAVE_LATENCY_BOUND_TRACKING
312 action->latency_limited = SIMIX_comm_is_latency_bounded( action ) ;
315 xbt_free(action->name);
316 SIMIX_comm_destroy_internal_actions(action);
318 if (action->comm.detached && action->state != SIMIX_DONE) {
319 /* the communication has failed and was detached:
320 * we have to free the buffer */
321 if (action->comm.clean_fun) {
322 action->comm.clean_fun(action->comm.src_buff);
324 action->comm.src_buff = NULL;
328 SIMIX_rdv_remove(action->comm.rdv, action);
330 xbt_mallocator_release(simix_global->action_mallocator, action);
333 void SIMIX_comm_destroy_internal_actions(smx_action_t action)
335 if (action->comm.surf_comm){
336 #ifdef HAVE_LATENCY_BOUND_TRACKING
337 action->latency_limited = SIMIX_comm_is_latency_bounded(action);
339 surf_action_unref(action->comm.surf_comm);
340 action->comm.surf_comm = NULL;
343 if (action->comm.src_timeout){
344 surf_action_unref(action->comm.src_timeout);
345 action->comm.src_timeout = NULL;
348 if (action->comm.dst_timeout){
349 surf_action_unref(action->comm.dst_timeout);
350 action->comm.dst_timeout = NULL;
354 void SIMIX_pre_comm_send(smx_simcall_t simcall, smx_rdv_t rdv,
355 double task_size, double rate,
356 void *src_buff, size_t src_buff_size,
357 int (*match_fun)(void *, void *,smx_action_t),
358 void *data, double timeout){
359 smx_action_t comm = SIMIX_comm_isend(simcall->issuer, rdv, task_size, rate,
360 src_buff, src_buff_size, match_fun, NULL,
362 SIMCALL_SET_MC_VALUE(simcall, 0);
363 SIMIX_pre_comm_wait(simcall, comm, timeout);
365 smx_action_t SIMIX_pre_comm_isend(smx_simcall_t simcall, smx_rdv_t rdv,
366 double task_size, double rate,
367 void *src_buff, size_t src_buff_size,
368 int (*match_fun)(void *, void *,smx_action_t),
369 void (*clean_fun)(void *),
370 void *data, int detached){
371 return SIMIX_comm_isend(simcall->issuer, rdv, task_size, rate, src_buff,
372 src_buff_size, match_fun, clean_fun, data, detached);
375 smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv,
376 double task_size, double rate,
377 void *src_buff, size_t src_buff_size,
378 int (*match_fun)(void *, void *,smx_action_t),
379 void (*clean_fun)(void *), // used to free the action in case of problem after a detached send
383 XBT_DEBUG("send from %p\n", rdv);
385 /* Prepare an action describing us, so that it gets passed to the user-provided filter of other side */
386 smx_action_t this_action = SIMIX_comm_new(SIMIX_COMM_SEND);
388 /* Look for communication action matching our needs. We also provide a description of
389 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
391 * If it is not found then push our communication into the rendez-vous point */
392 smx_action_t other_action = SIMIX_fifo_get_comm(rdv->comm_fifo, SIMIX_COMM_RECEIVE, match_fun, data, this_action);
395 other_action = this_action;
397 if (rdv->permanent_receiver!=NULL){
398 //this mailbox is for small messages, which have to be sent right now
399 other_action->state = SIMIX_READY;
400 other_action->comm.dst_proc=rdv->permanent_receiver;
401 other_action->comm.refcount++;
402 xbt_fifo_push(rdv->done_comm_fifo,other_action);
403 other_action->comm.rdv=rdv;
404 XBT_DEBUG("pushing a message into the permanent receive fifo %p, comm %p \n", rdv, &(other_action->comm));
407 SIMIX_rdv_push(rdv, this_action);
410 XBT_DEBUG("Receive already pushed\n");
412 SIMIX_comm_destroy(this_action);
413 --smx_total_comms; // this creation was a pure waste
415 other_action->state = SIMIX_READY;
416 other_action->comm.type = SIMIX_COMM_READY;
419 xbt_fifo_push(src_proc->comms, other_action);
421 /* if the communication action is detached then decrease the refcount
422 * by one, so it will be eliminated by the receiver's destroy call */
424 other_action->comm.detached = 1;
425 other_action->comm.refcount--;
426 other_action->comm.clean_fun = clean_fun;
428 other_action->comm.clean_fun = NULL;
431 /* Setup the communication action */
432 other_action->comm.src_proc = src_proc;
433 other_action->comm.task_size = task_size;
434 other_action->comm.rate = rate;
435 other_action->comm.src_buff = src_buff;
436 other_action->comm.src_buff_size = src_buff_size;
437 other_action->comm.src_data = data;
439 other_action->comm.match_fun = match_fun;
441 if (MC_is_active()) {
442 other_action->state = SIMIX_RUNNING;
443 return (detached ? NULL : other_action);
446 SIMIX_comm_start(other_action);
447 return (detached ? NULL : other_action);
450 void SIMIX_pre_comm_recv(smx_simcall_t simcall, smx_rdv_t rdv,
451 void *dst_buff, size_t *dst_buff_size,
452 int (*match_fun)(void *, void *, smx_action_t),
453 void *data, double timeout, double rate)
455 smx_action_t comm = SIMIX_comm_irecv(simcall->issuer, rdv, dst_buff,
456 dst_buff_size, match_fun, data, rate);
457 SIMCALL_SET_MC_VALUE(simcall, 0);
458 SIMIX_pre_comm_wait(simcall, comm, timeout);
461 smx_action_t SIMIX_pre_comm_irecv(smx_simcall_t simcall, smx_rdv_t rdv,
462 void *dst_buff, size_t *dst_buff_size,
463 int (*match_fun)(void *, void *, smx_action_t),
464 void *data, double rate)
466 return SIMIX_comm_irecv(simcall->issuer, rdv, dst_buff, dst_buff_size,
467 match_fun, data, rate);
470 smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv,
471 void *dst_buff, size_t *dst_buff_size,
472 int (*match_fun)(void *, void *, smx_action_t),
473 void *data, double rate)
475 XBT_DEBUG("recv from %p %p\n", rdv, rdv->comm_fifo);
476 smx_action_t this_action = SIMIX_comm_new(SIMIX_COMM_RECEIVE);
478 smx_action_t other_action;
479 //communication already done, get it inside the fifo of completed comms
480 //permanent receive v1
481 //int already_received=0;
482 if(rdv->permanent_receiver && xbt_fifo_size(rdv->done_comm_fifo)!=0){
484 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication\n");
485 //find a match in the already received fifo
486 other_action = SIMIX_fifo_get_comm(rdv->done_comm_fifo, SIMIX_COMM_SEND, match_fun, data, this_action);
487 //if not found, assume the receiver came first, register it to the mailbox in the classical way
489 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request into fifo\n");
490 other_action = this_action;
491 SIMIX_rdv_push(rdv, this_action);
493 if(other_action->comm.surf_comm && SIMIX_comm_get_remains(other_action)==0.0)
495 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it\n",&(other_action->comm));
496 other_action->state = SIMIX_DONE;
497 other_action->comm.type = SIMIX_COMM_DONE;
498 other_action->comm.rdv = NULL;
499 //SIMIX_comm_destroy(this_action);
500 //--smx_total_comms; // this creation was a pure waste
501 //already_received=1;
502 //other_action->comm.refcount--;
504 XBT_DEBUG("Not yet finished, we have to wait %d\n", xbt_fifo_size(rdv->comm_fifo));
506 other_action->comm.refcount--;
507 SIMIX_comm_destroy(this_action);
508 --smx_total_comms; // this creation was a pure waste
511 /* Prepare an action describing us, so that it gets passed to the user-provided filter of other side */
513 /* Look for communication action matching our needs. We also provide a description of
514 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
516 * If it is not found then push our communication into the rendez-vous point */
517 other_action = SIMIX_fifo_get_comm(rdv->comm_fifo, SIMIX_COMM_SEND, match_fun, data, this_action);
520 XBT_DEBUG("Receive pushed first %d\n", xbt_fifo_size(rdv->comm_fifo));
521 other_action = this_action;
522 SIMIX_rdv_push(rdv, this_action);
524 SIMIX_comm_destroy(this_action);
525 --smx_total_comms; // this creation was a pure waste
526 other_action->state = SIMIX_READY;
527 other_action->comm.type = SIMIX_COMM_READY;
528 //other_action->comm.refcount--;
530 xbt_fifo_push(dst_proc->comms, other_action);
533 /* Setup communication action */
534 other_action->comm.dst_proc = dst_proc;
535 other_action->comm.dst_buff = dst_buff;
536 other_action->comm.dst_buff_size = dst_buff_size;
537 other_action->comm.dst_data = data;
540 (other_action->comm.rate == -1.0 || rate < other_action->comm.rate))
541 other_action->comm.rate = rate;
543 other_action->comm.match_fun = match_fun;
546 /*if(already_received)//do the actual copy, because the first one after the comm didn't have all the info
547 SIMIX_comm_copy_data(other_action);*/
550 if (MC_is_active()) {
551 other_action->state = SIMIX_RUNNING;
555 SIMIX_comm_start(other_action);
560 smx_action_t SIMIX_pre_comm_iprobe(smx_simcall_t simcall, smx_rdv_t rdv,
562 int (*match_fun)(void *, void *, smx_action_t),
564 return SIMIX_comm_iprobe(simcall->issuer, rdv, src, tag, match_fun, data);
567 smx_action_t SIMIX_comm_iprobe(smx_process_t dst_proc, smx_rdv_t rdv, int src,
568 int tag, int (*match_fun)(void *, void *, smx_action_t), void *data)
570 XBT_DEBUG("iprobe from %p %p\n", rdv, rdv->comm_fifo);
571 smx_action_t this_action = SIMIX_comm_new(SIMIX_COMM_RECEIVE);
573 smx_action_t other_action=NULL;
574 if(rdv->permanent_receiver && xbt_fifo_size(rdv->done_comm_fifo)!=0){
575 //find a match in the already received fifo
576 XBT_DEBUG("first try in the perm recv mailbox \n");
578 other_action = SIMIX_fifo_probe_comm(rdv->done_comm_fifo, SIMIX_COMM_SEND, match_fun, data, this_action);
582 XBT_DEBUG("second try in the other mailbox");
583 other_action = SIMIX_fifo_probe_comm(rdv->comm_fifo, SIMIX_COMM_SEND, match_fun, data, this_action);
586 if(other_action)other_action->comm.refcount--;
588 SIMIX_comm_destroy(this_action);
593 void SIMIX_pre_comm_wait(smx_simcall_t simcall, smx_action_t action, double timeout)
595 /* the simcall may be a wait, a send or a recv */
598 /* Associate this simcall to the wait action */
599 XBT_DEBUG("SIMIX_pre_comm_wait, %p", action);
601 xbt_fifo_push(action->simcalls, simcall);
602 simcall->issuer->waiting_action = action;
604 if (MC_is_active()) {
605 int idx = SIMCALL_GET_MC_VALUE(simcall);
607 action->state = SIMIX_DONE;
609 /* If we reached this point, the wait simcall must have a timeout */
610 /* Otherwise it shouldn't be enabled and executed by the MC */
614 if (action->comm.src_proc == simcall->issuer)
615 action->state = SIMIX_SRC_TIMEOUT;
617 action->state = SIMIX_DST_TIMEOUT;
620 SIMIX_comm_finish(action);
624 /* If the action has already finish perform the error handling, */
625 /* otherwise set up a waiting timeout on the right side */
626 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
627 SIMIX_comm_finish(action);
628 } else { /* if (timeout >= 0) { we need a surf sleep action even when there is no timeout, otherwise surf won't tell us when the host fails */
629 sleep = surf_workstation_sleep(simcall->issuer->smx_host, timeout);
630 surf_action_set_data(sleep, action);
632 if (simcall->issuer == action->comm.src_proc)
633 action->comm.src_timeout = sleep;
635 action->comm.dst_timeout = sleep;
639 void SIMIX_pre_comm_test(smx_simcall_t simcall, smx_action_t action)
642 simcall_comm_test__set__result(simcall, action->comm.src_proc && action->comm.dst_proc);
643 if(simcall_comm_test__get__result(simcall)){
644 action->state = SIMIX_DONE;
645 xbt_fifo_push(action->simcalls, simcall);
646 SIMIX_comm_finish(action);
648 SIMIX_simcall_answer(simcall);
653 simcall_comm_test__set__result(simcall, (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING));
654 if (simcall_comm_test__get__result(simcall)) {
655 xbt_fifo_push(action->simcalls, simcall);
656 SIMIX_comm_finish(action);
658 SIMIX_simcall_answer(simcall);
662 void SIMIX_pre_comm_testany(smx_simcall_t simcall, xbt_dynar_t actions)
666 simcall_comm_testany__set__result(simcall, -1);
669 int idx = SIMCALL_GET_MC_VALUE(simcall);
671 SIMIX_simcall_answer(simcall);
673 action = xbt_dynar_get_as(actions, idx, smx_action_t);
674 simcall_comm_testany__set__result(simcall, idx);
675 xbt_fifo_push(action->simcalls, simcall);
676 action->state = SIMIX_DONE;
677 SIMIX_comm_finish(action);
682 xbt_dynar_foreach(simcall_comm_testany__get__comms(simcall), cursor,action) {
683 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
684 simcall_comm_testany__set__result(simcall, cursor);
685 xbt_fifo_push(action->simcalls, simcall);
686 SIMIX_comm_finish(action);
690 SIMIX_simcall_answer(simcall);
693 void SIMIX_pre_comm_waitany(smx_simcall_t simcall, xbt_dynar_t actions)
696 unsigned int cursor = 0;
699 int idx = SIMCALL_GET_MC_VALUE(simcall);
700 action = xbt_dynar_get_as(actions, idx, smx_action_t);
701 xbt_fifo_push(action->simcalls, simcall);
702 simcall_comm_waitany__set__result(simcall, idx);
703 action->state = SIMIX_DONE;
704 SIMIX_comm_finish(action);
708 xbt_dynar_foreach(actions, cursor, action){
709 /* associate this simcall to the the action */
710 xbt_fifo_push(action->simcalls, simcall);
712 /* see if the action is already finished */
713 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){
714 SIMIX_comm_finish(action);
720 void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall)
723 unsigned int cursor = 0;
724 xbt_dynar_t actions = simcall_comm_waitany__get__comms(simcall);
726 xbt_dynar_foreach(actions, cursor, action) {
727 xbt_fifo_remove(action->simcalls, simcall);
732 * \brief Starts the simulation of a communication action.
733 * \param action the communication action
735 static XBT_INLINE void SIMIX_comm_start(smx_action_t action)
737 /* If both the sender and the receiver are already there, start the communication */
738 if (action->state == SIMIX_READY) {
740 smx_host_t sender = action->comm.src_proc->smx_host;
741 smx_host_t receiver = action->comm.dst_proc->smx_host;
743 XBT_DEBUG("Starting communication %p from '%s' to '%s'", action,
744 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
746 action->comm.surf_comm = surf_workstation_model_communicate(surf_workstation_model,
748 action->comm.task_size, action->comm.rate);
750 surf_action_set_data(action->comm.surf_comm, action);
752 action->state = SIMIX_RUNNING;
754 /* If a link is failed, detect it immediately */
755 if (surf_action_get_state(action->comm.surf_comm) == SURF_ACTION_FAILED) {
756 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure",
757 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
758 action->state = SIMIX_LINK_FAILURE;
759 SIMIX_comm_destroy_internal_actions(action);
762 /* If any of the process is suspend, create the action but stop its execution,
763 it will be restarted when the sender process resume */
764 if (SIMIX_process_is_suspended(action->comm.src_proc) ||
765 SIMIX_process_is_suspended(action->comm.dst_proc)) {
766 /* FIXME: check what should happen with the action state */
768 if (SIMIX_process_is_suspended(action->comm.src_proc))
769 XBT_DEBUG("The communication is suspended on startup because src (%s:%s) were suspended since it initiated the communication",
770 SIMIX_host_get_name(action->comm.src_proc->smx_host), action->comm.src_proc->name);
772 XBT_DEBUG("The communication is suspended on startup because dst (%s:%s) were suspended since it initiated the communication",
773 SIMIX_host_get_name(action->comm.dst_proc->smx_host), action->comm.dst_proc->name);
775 surf_action_suspend(action->comm.surf_comm);
782 * \brief Answers the SIMIX simcalls associated to a communication action.
783 * \param action a finished communication action
785 void SIMIX_comm_finish(smx_action_t action)
787 unsigned int destroy_count = 0;
788 smx_simcall_t simcall;
791 while ((simcall = xbt_fifo_shift(action->simcalls))) {
793 /* If a waitany simcall is waiting for this action to finish, then remove
794 it from the other actions in the waitany list. Afterwards, get the
795 position of the actual action in the waitany dynar and
796 return it as the result of the simcall */
797 if (simcall->call == SIMCALL_COMM_WAITANY) {
798 SIMIX_waitany_remove_simcall_from_actions(simcall);
800 simcall_comm_waitany__set__result(simcall, xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &action));
803 /* If the action is still in a rendez-vous point then remove from it */
804 if (action->comm.rdv)
805 SIMIX_rdv_remove(action->comm.rdv, action);
807 XBT_DEBUG("SIMIX_comm_finish: action state = %d", (int)action->state);
809 /* Check out for errors */
810 switch (action->state) {
813 XBT_DEBUG("Communication %p complete!", action);
814 SIMIX_comm_copy_data(action);
817 case SIMIX_SRC_TIMEOUT:
818 SMX_EXCEPTION(simcall->issuer, timeout_error, 0,
819 "Communication timeouted because of sender");
822 case SIMIX_DST_TIMEOUT:
823 SMX_EXCEPTION(simcall->issuer, timeout_error, 0,
824 "Communication timeouted because of receiver");
827 case SIMIX_SRC_HOST_FAILURE:
828 if (simcall->issuer == action->comm.src_proc)
829 simcall->issuer->context->iwannadie = 1;
830 // SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
832 SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed");
835 case SIMIX_DST_HOST_FAILURE:
836 if (simcall->issuer == action->comm.dst_proc)
837 simcall->issuer->context->iwannadie = 1;
838 // SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
840 SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed");
843 case SIMIX_LINK_FAILURE:
844 XBT_DEBUG("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p) detached:%d",
846 action->comm.src_proc ? sg_host_name(action->comm.src_proc->smx_host) : NULL,
847 action->comm.dst_proc ? sg_host_name(action->comm.dst_proc->smx_host) : NULL,
848 simcall->issuer->name, simcall->issuer, action->comm.detached);
849 if (action->comm.src_proc == simcall->issuer) {
850 XBT_DEBUG("I'm source");
851 } else if (action->comm.dst_proc == simcall->issuer) {
852 XBT_DEBUG("I'm dest");
854 XBT_DEBUG("I'm neither source nor dest");
856 SMX_EXCEPTION(simcall->issuer, network_error, 0, "Link failure");
860 if (simcall->issuer == action->comm.dst_proc)
861 SMX_EXCEPTION(simcall->issuer, cancel_error, 0,
862 "Communication canceled by the sender");
864 SMX_EXCEPTION(simcall->issuer, cancel_error, 0,
865 "Communication canceled by the receiver");
869 xbt_die("Unexpected action state in SIMIX_comm_finish: %d", (int)action->state);
872 /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
873 if (simcall->issuer->doexception) {
874 if (simcall->call == SIMCALL_COMM_WAITANY) {
875 simcall->issuer->running_ctx->exception.value = xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &action);
877 else if (simcall->call == SIMCALL_COMM_TESTANY) {
878 simcall->issuer->running_ctx->exception.value = xbt_dynar_search(simcall_comm_testany__get__comms(simcall), &action);
882 if (surf_resource_get_state(surf_workstation_resource_priv(simcall->issuer->smx_host)) != SURF_RESOURCE_ON) {
883 simcall->issuer->context->iwannadie = 1;
886 simcall->issuer->waiting_action = NULL;
887 xbt_fifo_remove(simcall->issuer->comms, action);
888 if(action->comm.detached){
892 if(simcall->issuer == action->comm.src_proc){
893 if(action->comm.dst_proc){
894 xbt_swag_foreach(proc, simix_global->process_list)
896 if(proc==action->comm.dst_proc){
902 if(still_alive) xbt_fifo_remove(action->comm.dst_proc->comms, action);
904 if(simcall->issuer == action->comm.dst_proc){
905 if(action->comm.src_proc)
906 if(action->comm.dst_proc){
907 xbt_swag_foreach(proc, simix_global->process_list)
909 if(proc==action->comm.src_proc){
915 if(still_alive) xbt_fifo_remove(action->comm.src_proc->comms, action);
918 SIMIX_simcall_answer(simcall);
922 while (destroy_count-- > 0)
923 SIMIX_comm_destroy(action);
927 * \brief This function is called when a Surf communication action is finished.
928 * \param action the corresponding Simix communication
930 void SIMIX_post_comm(smx_action_t action)
932 /* Update action state */
933 if (action->comm.src_timeout &&
934 surf_action_get_state(action->comm.src_timeout) == SURF_ACTION_DONE)
935 action->state = SIMIX_SRC_TIMEOUT;
936 else if (action->comm.dst_timeout &&
937 surf_action_get_state(action->comm.dst_timeout) == SURF_ACTION_DONE)
938 action->state = SIMIX_DST_TIMEOUT;
939 else if (action->comm.src_timeout &&
940 surf_action_get_state(action->comm.src_timeout) == SURF_ACTION_FAILED)
941 action->state = SIMIX_SRC_HOST_FAILURE;
942 else if (action->comm.dst_timeout &&
943 surf_action_get_state(action->comm.dst_timeout) == SURF_ACTION_FAILED)
944 action->state = SIMIX_DST_HOST_FAILURE;
945 else if (action->comm.surf_comm &&
946 surf_action_get_state(action->comm.surf_comm) == SURF_ACTION_FAILED) {
947 XBT_DEBUG("Puta madre. Surf says that the link broke");
948 action->state = SIMIX_LINK_FAILURE;
950 action->state = SIMIX_DONE;
952 XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d",
953 action, (int)action->state, action->comm.src_proc, action->comm.dst_proc, action->comm.detached);
955 /* destroy the surf actions associated with the Simix communication */
956 SIMIX_comm_destroy_internal_actions(action);
958 /* remove the communication action from the list of pending communications
959 * of both processes (if they still exist) */
960 if (action->comm.src_proc) {
961 xbt_fifo_remove(action->comm.src_proc->comms, action);
963 if (action->comm.dst_proc) {
964 xbt_fifo_remove(action->comm.dst_proc->comms, action);
967 /* if there are simcalls associated with the action, then answer them */
968 if (xbt_fifo_size(action->simcalls)) {
969 SIMIX_comm_finish(action);
973 void SIMIX_pre_comm_cancel(smx_simcall_t simcall, smx_action_t action){
974 SIMIX_comm_cancel(action);
976 void SIMIX_comm_cancel(smx_action_t action)
978 /* if the action is a waiting state means that it is still in a rdv */
979 /* so remove from it and delete it */
980 if (action->state == SIMIX_WAITING) {
981 SIMIX_rdv_remove(action->comm.rdv, action);
982 action->state = SIMIX_CANCELED;
984 else if (!MC_is_active() /* when running the MC there are no surf actions */
985 && (action->state == SIMIX_READY || action->state == SIMIX_RUNNING)) {
987 surf_action_cancel(action->comm.surf_comm);
991 void SIMIX_comm_suspend(smx_action_t action)
993 /*FIXME: shall we suspend also the timeout actions? */
994 if (action->comm.surf_comm)
995 surf_action_suspend(action->comm.surf_comm);
996 /* in the other case, the action will be suspended on creation, in SIMIX_comm_start() */
999 void SIMIX_comm_resume(smx_action_t action)
1001 /*FIXME: check what happen with the timeouts */
1002 if (action->comm.surf_comm)
1003 surf_action_resume(action->comm.surf_comm);
1004 /* in the other case, the action were not really suspended yet, see SIMIX_comm_suspend() and SIMIX_comm_start() */
1008 /************* Action Getters **************/
1010 double SIMIX_pre_comm_get_remains(smx_simcall_t simcall, smx_action_t action){
1011 return SIMIX_comm_get_remains(action);
1014 * \brief get the amount remaining from the communication
1015 * \param action The communication
1017 double SIMIX_comm_get_remains(smx_action_t action)
1025 switch (action->state) {
1028 remains = surf_action_get_remains(action->comm.surf_comm);
1033 remains = 0; /*FIXME: check what should be returned */
1037 remains = 0; /*FIXME: is this correct? */
1043 e_smx_state_t SIMIX_pre_comm_get_state(smx_simcall_t simcall, smx_action_t action){
1044 return SIMIX_comm_get_state(action);
1046 e_smx_state_t SIMIX_comm_get_state(smx_action_t action)
1048 return action->state;
1051 void* SIMIX_pre_comm_get_src_data(smx_simcall_t simcall, smx_action_t action){
1052 return SIMIX_comm_get_src_data(action);
1055 * \brief Return the user data associated to the sender of the communication
1056 * \param action The communication
1057 * \return the user data
1059 void* SIMIX_comm_get_src_data(smx_action_t action)
1061 return action->comm.src_data;
1064 void* SIMIX_pre_comm_get_dst_data(smx_simcall_t simcall, smx_action_t action){
1065 return SIMIX_comm_get_dst_data(action);
1068 * \brief Return the user data associated to the receiver of the communication
1069 * \param action The communication
1070 * \return the user data
1072 void* SIMIX_comm_get_dst_data(smx_action_t action)
1074 return action->comm.dst_data;
1077 smx_process_t SIMIX_pre_comm_get_src_proc(smx_simcall_t simcall, smx_action_t action){
1078 return SIMIX_comm_get_src_proc(action);
1080 smx_process_t SIMIX_comm_get_src_proc(smx_action_t action)
1082 return action->comm.src_proc;
1085 smx_process_t SIMIX_pre_comm_get_dst_proc(smx_simcall_t simcall, smx_action_t action){
1086 return SIMIX_comm_get_dst_proc(action);
1088 smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action)
1090 return action->comm.dst_proc;
1093 #ifdef HAVE_LATENCY_BOUND_TRACKING
1094 int SIMIX_pre_comm_is_latency_bounded(smx_simcall_t simcall, smx_action_t action)
1096 return SIMIX_comm_is_latency_bounded(action);
1100 * \brief verify if communication is latency bounded
1101 * \param comm The communication
1103 int SIMIX_comm_is_latency_bounded(smx_action_t action)
1108 if (action->comm.surf_comm){
1109 XBT_DEBUG("Getting latency limited for surf_action (%p)", action->comm.surf_comm);
1110 action->latency_limited = surf_network_action_get_latency_limited(action->comm.surf_comm);
1111 XBT_DEBUG("Action limited is %d", action->latency_limited);
1113 return action->latency_limited;
1117 /******************************************************************************/
1118 /* SIMIX_comm_copy_data callbacks */
1119 /******************************************************************************/
1120 static void (*SIMIX_comm_copy_data_callback) (smx_action_t, void*, size_t) =
1121 &SIMIX_comm_copy_pointer_callback;
1124 SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, void*, size_t))
1126 SIMIX_comm_copy_data_callback = callback;
1129 void SIMIX_comm_copy_pointer_callback(smx_action_t comm, void* buff, size_t buff_size)
1131 xbt_assert((buff_size == sizeof(void *)),
1132 "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
1133 *(void **) (comm->comm.dst_buff) = buff;
1136 void SIMIX_comm_copy_buffer_callback(smx_action_t comm, void* buff, size_t buff_size)
1138 XBT_DEBUG("Copy the data over");
1139 memcpy(comm->comm.dst_buff, buff, buff_size);
1140 if (comm->comm.detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the original buffer available to the application ASAP
1142 comm->comm.src_buff = NULL;
1148 * \brief Copy the communication data from the sender's buffer to the receiver's one
1149 * \param comm The communication
1151 void SIMIX_comm_copy_data(smx_action_t comm)
1153 size_t buff_size = comm->comm.src_buff_size;
1154 /* If there is no data to be copy then return */
1155 if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied)
1158 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
1160 comm->comm.src_proc ? sg_host_name(comm->comm.src_proc->smx_host) : "a finished process",
1161 comm->comm.src_buff,
1162 comm->comm.dst_proc ? sg_host_name(comm->comm.dst_proc->smx_host) : "a finished process",
1163 comm->comm.dst_buff, buff_size);
1165 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
1166 if (comm->comm.dst_buff_size)
1167 buff_size = MIN(buff_size, *(comm->comm.dst_buff_size));
1169 /* Update the receiver's buffer size to the copied amount */
1170 if (comm->comm.dst_buff_size)
1171 *comm->comm.dst_buff_size = buff_size;
1174 SIMIX_comm_copy_data_callback (comm, comm->comm.src_buff, buff_size);
1176 /* Set the copied flag so we copy data only once */
1177 /* (this function might be called from both communication ends) */
1178 comm->comm.copied = 1;