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 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
13 "Logging specific to SIMIX (network)");
15 static xbt_dict_t rdv_points = NULL;
17 static XBT_INLINE void SIMIX_comm_start(smx_action_t action);
18 static void SIMIX_comm_finish(smx_action_t action);
19 static void SIMIX_waitany_req_remove_from_actions(smx_req_t req);
20 static void SIMIX_comm_copy_data(smx_action_t comm);
21 static smx_action_t SIMIX_comm_new(e_smx_comm_type_t type);
22 static XBT_INLINE void SIMIX_comm_wait_for_completion(smx_action_t comm,
24 static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm);
25 static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm);
26 static smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type,
27 int (*match_fun)(void *, void *), void *);
28 static void SIMIX_rdv_free(void *data);
30 void SIMIX_network_init(void)
32 rdv_points = xbt_dict_new();
35 void SIMIX_network_exit(void)
37 xbt_dict_free(&rdv_points);
40 /******************************************************************************/
41 /* Rendez-Vous Points */
42 /******************************************************************************/
44 smx_rdv_t SIMIX_rdv_create(const char *name)
46 /* two processes may have pushed the same rdv_create request at the same time */
47 smx_rdv_t rdv = name ? xbt_dict_get_or_null(rdv_points, name) : NULL;
50 rdv = xbt_new0(s_smx_rvpoint_t, 1);
51 rdv->name = name ? xbt_strdup(name) : NULL;
52 rdv->comm_fifo = xbt_fifo_new();
55 xbt_dict_set(rdv_points, name, rdv, SIMIX_rdv_free);
60 void SIMIX_rdv_destroy(smx_rdv_t rdv)
63 xbt_dict_remove(rdv_points, rdv->name);
66 void SIMIX_rdv_free(void *data)
68 smx_rdv_t rdv = (smx_rdv_t) data;
71 xbt_fifo_free(rdv->comm_fifo);
75 smx_rdv_t SIMIX_rdv_get_by_name(const char *name)
77 return xbt_dict_get_or_null(rdv_points, name);
80 int SIMIX_rdv_comm_count_by_host(smx_rdv_t rdv, smx_host_t host)
82 smx_action_t comm = NULL;
83 xbt_fifo_item_t item = NULL;
86 xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_action_t) {
87 if (comm->comm.src_proc->smx_host == host)
94 smx_action_t SIMIX_rdv_get_head(smx_rdv_t rdv)
96 return xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
100 * \brief Push a communication request into a rendez-vous point
101 * \param rdv The rendez-vous point
102 * \param comm The communication request
104 static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm)
106 xbt_fifo_push(rdv->comm_fifo, comm);
107 comm->comm.rdv = rdv;
111 * \brief Remove a communication request from a rendez-vous point
112 * \param rdv The rendez-vous point
113 * \param comm The communication request
115 static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm)
117 xbt_fifo_remove(rdv->comm_fifo, comm);
118 comm->comm.rdv = NULL;
122 * \brief Checks if there is a communication request queued in a rendez-vous matching our needs
123 * \param type The type of communication we are looking for (comm_send, comm_recv)
124 * \return The communication request if found, NULL otherwise
126 smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type,
127 int (*match_fun)(void *, void *), void *data)
130 xbt_fifo_item_t item;
131 void* req_data = NULL;
133 xbt_fifo_foreach(rdv->comm_fifo, item, req, smx_action_t){
134 if(req->comm.type == SIMIX_COMM_SEND) {
135 req_data = req->comm.src_data;
136 } else if(req->comm.type == SIMIX_COMM_RECEIVE) {
137 req_data = req->comm.dst_data;
139 if (req->comm.type == type && (!match_fun || match_fun(data, req_data))) {
140 xbt_fifo_remove_item(rdv->comm_fifo, item);
141 xbt_fifo_free_item(item);
142 req->comm.refcount++;
143 req->comm.rdv = NULL;
147 DEBUG0("Communication request not found");
151 /******************************************************************************/
152 /* Comunication Actions */
153 /******************************************************************************/
156 * \brief Creates a new comunicate action
157 * \param type The type of request (comm_send, comm_recv)
158 * \return The new comunicate action
160 smx_action_t SIMIX_comm_new(e_smx_comm_type_t type)
164 /* alloc structures */
165 act = xbt_new0(s_smx_action_t, 1);
166 act->type = SIMIX_ACTION_COMMUNICATE;
167 act->state = SIMIX_WAITING;
168 act->request_list = xbt_fifo_new();
170 /* set communication */
171 act->comm.type = type;
172 act->comm.refcount = 1;
174 #ifdef HAVE_LATENCY_BOUND_TRACKING
175 //initialize with unknown value
176 act->latency_limited = -1;
180 act->category = NULL;
183 DEBUG1("Create communicate action %p", act);
189 * \brief Destroy a communicate action
190 * \param action The communicate action to be destroyed
192 void SIMIX_comm_destroy(smx_action_t action)
194 DEBUG2("Destroy action %p (refcount:%d)", action, action->comm.refcount);
196 if (action->comm.refcount <= 0)
197 xbt_die(bprintf("the refcount of comm %p is already 0 before decreasing it. That's a bug!",action));
199 action->comm.refcount--;
200 if (action->comm.refcount > 0)
202 DEBUG2("Really free communication %p; refcount is now %d", action,
203 action->comm.refcount);
205 #ifdef HAVE_LATENCY_BOUND_TRACKING
206 action->latency_limited = SIMIX_comm_is_latency_bounded( action ) ;
210 TRACE_smx_action_destroy(action);
214 xbt_free(action->name);
216 xbt_fifo_free(action->request_list);
218 SIMIX_comm_destroy_internal_actions(action);
223 void SIMIX_comm_destroy_internal_actions(smx_action_t action)
225 if (action->comm.surf_comm){
226 #ifdef HAVE_LATENCY_BOUND_TRACKING
227 action->latency_limited = SIMIX_comm_is_latency_bounded(action);
229 action->comm.surf_comm->model_type->action_unref(action->comm.surf_comm);
230 action->comm.surf_comm = NULL;
233 if (action->comm.src_timeout){
234 action->comm.src_timeout->model_type->action_unref(action->comm.src_timeout);
235 action->comm.src_timeout = NULL;
238 if (action->comm.dst_timeout){
239 action->comm.dst_timeout->model_type->action_unref(action->comm.dst_timeout);
240 action->comm.dst_timeout = NULL;
244 smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv,
245 double task_size, double rate,
246 void *src_buff, size_t src_buff_size,
247 int (*match_fun)(void *, void *), void *data)
251 /* Look for communication request matching our needs.
252 If it is not found then create it and push it into the rendez-vous point */
253 action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_RECEIVE, match_fun, data);
256 action = SIMIX_comm_new(SIMIX_COMM_SEND);
257 SIMIX_rdv_push(rdv, action);
259 action->state = SIMIX_READY;
260 action->comm.type = SIMIX_COMM_READY;
263 /* Setup the communication request */
264 action->comm.src_proc = src_proc;
265 action->comm.task_size = task_size;
266 action->comm.rate = rate;
267 action->comm.src_buff = src_buff;
268 action->comm.src_buff_size = src_buff_size;
269 action->comm.src_data = data;
272 action->state = SIMIX_RUNNING;
276 SIMIX_comm_start(action);
280 smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv,
281 void *dst_buff, size_t *dst_buff_size,
282 int (*match_fun)(void *, void *), void *data)
286 /* Look for communication request matching our needs.
287 * If it is not found then create it and push it into the rendez-vous point
289 action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_SEND, match_fun, data);
292 action = SIMIX_comm_new(SIMIX_COMM_RECEIVE);
293 SIMIX_rdv_push(rdv, action);
295 action->state = SIMIX_READY;
296 action->comm.type = SIMIX_COMM_READY;
299 /* Setup communication request */
300 action->comm.dst_proc = dst_proc;
301 action->comm.dst_buff = dst_buff;
302 action->comm.dst_buff_size = dst_buff_size;
303 action->comm.dst_data = data;
306 action->state = SIMIX_RUNNING;
310 SIMIX_comm_start(action);
314 void SIMIX_pre_comm_wait(smx_req_t req)
316 smx_action_t action = req->comm_wait.comm;
317 double timeout = req->comm_wait.timeout;
320 /* Associate this request to the action */
321 xbt_fifo_push(action->request_list, req);
322 req->issuer->waiting_action = action;
325 action->state = SIMIX_DONE;
326 SIMIX_comm_finish(action);
330 /* If the action has already finish perform the error handling, */
331 /* otherwise set up a waiting timeout on the right side */
332 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
333 SIMIX_comm_finish(action);
334 } 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 */
335 sleep = surf_workstation_model->extension.workstation.sleep(req->issuer->smx_host->host, timeout);
336 surf_workstation_model->action_data_set(sleep, action);
338 if (req->issuer == action->comm.src_proc)
339 action->comm.src_timeout = sleep;
341 action->comm.dst_timeout = sleep;
345 void SIMIX_pre_comm_test(smx_req_t req)
347 smx_action_t action = req->comm_test.comm;
348 req->comm_test.result = (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING);
350 if (req->comm_test.result) {
351 xbt_fifo_push(action->request_list, req);
352 SIMIX_comm_finish(action);
354 SIMIX_request_answer(req);
358 void SIMIX_pre_comm_testany(smx_req_t req)
362 req->comm_testany.result = -1;
363 xbt_dynar_foreach(req->comm_testany.comms,cursor,action) {
364 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
365 req->comm_testany.result = cursor;
366 xbt_fifo_push(action->request_list, req);
367 SIMIX_comm_finish(action);
371 SIMIX_request_answer(req);
374 void SIMIX_pre_comm_waitany(smx_req_t req)
377 unsigned int cursor = 0;
378 xbt_dynar_t actions = req->comm_waitany.comms;
379 xbt_dynar_foreach(actions, cursor, action){
380 /* Associate this request to the action */
381 xbt_fifo_push(action->request_list, req);
382 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){
383 SIMIX_comm_finish(action);
389 void SIMIX_waitany_req_remove_from_actions(smx_req_t req)
392 unsigned int cursor = 0;
393 xbt_dynar_t actions = req->comm_waitany.comms;
395 xbt_dynar_foreach(actions, cursor, action){
396 xbt_fifo_remove(action->request_list, req);
401 * \brief Start the simulation of a communication request
402 * \param action The communication action
404 static XBT_INLINE void SIMIX_comm_start(smx_action_t action)
406 /* If both the sender and the receiver are already there, start the communication */
407 if (action->state == SIMIX_READY) {
408 smx_host_t sender = action->comm.src_proc->smx_host;
409 smx_host_t receiver = action->comm.dst_proc->smx_host;
411 DEBUG3("Starting communication %p from '%s' to '%s'", action,
412 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
414 action->comm.surf_comm = surf_workstation_model->extension.workstation.
415 communicate(sender->host, receiver->host, action->comm.task_size, action->comm.rate);
417 surf_workstation_model->action_data_set(action->comm.surf_comm, action);
419 action->state = SIMIX_RUNNING;
422 TRACE_smx_action_communicate(action, action->comm.src_proc);
425 /* If a link is failed, detect it immediately */
426 if (surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) {
427 DEBUG2("Communication from '%s' to '%s' failed to start because of a link failure",
428 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
429 action->state = SIMIX_LINK_FAILURE;
430 SIMIX_comm_destroy_internal_actions(action);
433 /* If any of the process is suspend, create the action but stop its execution,
434 it will be restarted when the sender process resume */
435 if (SIMIX_process_is_suspended(action->comm.src_proc) ||
436 SIMIX_process_is_suspended(action->comm.dst_proc)) {
437 /* FIXME: check what should happen with the action state */
438 surf_workstation_model->suspend(action->comm.surf_comm);
443 void SIMIX_comm_finish(smx_action_t action)
447 while ((req = xbt_fifo_shift(action->request_list))) {
449 /* If a waitany request is waiting for this action to finish, then remove
450 it from the other actions in the waitany list. Afterwards, get the
451 position of the actual action in the waitany request's actions dynar and
452 return it as the result of the call */
453 if (req->call == REQ_COMM_WAITANY) {
454 SIMIX_waitany_req_remove_from_actions(req);
455 req->comm_waitany.result = xbt_dynar_search(req->comm_waitany.comms, &action);
458 /* If the action is still in a rendez-vous point then remove from it */
459 if (action->comm.rdv)
460 SIMIX_rdv_remove(action->comm.rdv, action);
462 DEBUG1("SIMIX_comm_finish: action state = %d", action->state);
464 /* Check out for errors */
465 switch (action->state) {
468 DEBUG1("Communication %p complete!", action);
469 SIMIX_comm_copy_data(action);
472 case SIMIX_SRC_TIMEOUT:
474 THROW0(timeout_error, 0, "Communication timeouted because of sender");
476 CATCH(req->issuer->running_ctx->exception) {
477 req->issuer->doexception = 1;
481 case SIMIX_DST_TIMEOUT:
483 THROW0(timeout_error, 0, "Communication timeouted because of receiver");
485 CATCH(req->issuer->running_ctx->exception) {
486 req->issuer->doexception = 1;
490 case SIMIX_SRC_HOST_FAILURE:
492 if (req->issuer == action->comm.src_proc)
493 THROW0(host_error, 0, "Host failed");
495 THROW0(network_error, 0, "Remote peer failed");
497 CATCH(req->issuer->running_ctx->exception) {
498 req->issuer->doexception = 1;
502 case SIMIX_DST_HOST_FAILURE:
504 if (req->issuer == action->comm.dst_proc)
505 THROW0(host_error, 0, "Host failed");
507 THROW0(network_error, 0, "Remote peer failed");
509 CATCH(req->issuer->running_ctx->exception) {
510 req->issuer->doexception = 1;
514 case SIMIX_LINK_FAILURE:
516 DEBUG5("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)",
517 action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name,
518 req->issuer->name, req->issuer);
519 THROW0(network_error, 0, "Link failure");
521 CATCH(req->issuer->running_ctx->exception) {
522 req->issuer->doexception = 1;
530 /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
531 if (req->issuer->doexception) {
532 if (req->call == REQ_COMM_WAITANY) {
533 req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_waitany.comms, &action);
535 else if (req->call == REQ_COMM_TESTANY) {
536 req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_testany.comms, &action);
540 req->issuer->waiting_action = NULL;
541 SIMIX_request_answer(req);
545 void SIMIX_post_comm(smx_action_t action)
547 /* Update action state */
548 if (action->comm.src_timeout &&
549 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_DONE)
550 action->state = SIMIX_SRC_TIMEOUT;
551 else if (action->comm.dst_timeout &&
552 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_DONE)
553 action->state = SIMIX_DST_TIMEOUT;
554 else if (action->comm.src_timeout &&
555 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_FAILED)
556 action->state = SIMIX_SRC_HOST_FAILURE;
557 else if (action->comm.dst_timeout &&
558 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED)
559 action->state = SIMIX_DST_HOST_FAILURE;
560 else if (action->comm.surf_comm &&
561 surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED)
562 action->state = SIMIX_LINK_FAILURE;
564 action->state = SIMIX_DONE;
566 DEBUG1("SIMIX_post_comm: action state = %d", action->state);
568 /* After this point the surf actions associated with the simix communicate
569 action are no longer needed, thus we delete them. */
570 SIMIX_comm_destroy_internal_actions(action);
572 /* If there are requests associated with the action, then answer them */
573 if (xbt_fifo_size(action->request_list))
574 SIMIX_comm_finish(action);
577 void SIMIX_comm_cancel(smx_action_t action)
579 /* If the action is a waiting state means that it is still in a rdv */
580 /* so remove from it and delete it */
581 if (action->state == SIMIX_WAITING) {
582 SIMIX_rdv_remove(action->comm.rdv, action);
583 action->state = SIMIX_FAILED;
585 surf_workstation_model->action_cancel(action->comm.surf_comm);
589 void SIMIX_comm_suspend(smx_action_t action)
591 /*FIXME: shall we suspend also the timeout actions? */
592 surf_workstation_model->suspend(action->comm.surf_comm);
595 void SIMIX_comm_resume(smx_action_t action)
597 /*FIXME: check what happen with the timeouts */
598 surf_workstation_model->resume(action->comm.surf_comm);
602 /************* Action Getters **************/
605 * \brief get the amount remaining from the communication
606 * \param action The communication
608 double SIMIX_comm_get_remains(smx_action_t action)
612 switch (action->state) {
615 remains = surf_workstation_model->get_remains(action->comm.surf_comm);
620 remains = 0; /*FIXME: check what should be returned */
624 remains = 0; /*FIXME: is this correct? */
630 e_smx_state_t SIMIX_comm_get_state(smx_action_t action)
632 return action->state;
636 * \brief Return the user data associated to the sender of the communication
637 * \param action The communication
638 * \return the user data
640 void* SIMIX_comm_get_src_data(smx_action_t action)
642 return action->comm.src_data;
646 * \brief Return the user data associated to the receiver of the communication
647 * \param action The communication
648 * \return the user data
650 void* SIMIX_comm_get_dst_data(smx_action_t action)
652 return action->comm.dst_data;
655 void* SIMIX_comm_get_src_buff(smx_action_t action)
657 return action->comm.src_buff;
660 void* SIMIX_comm_get_dst_buff(smx_action_t action)
662 return action->comm.dst_buff;
665 size_t SIMIX_comm_get_src_buff_size(smx_action_t action)
667 return action->comm.src_buff_size;
670 size_t SIMIX_comm_get_dst_buff_size(smx_action_t action)
674 if (action->comm.dst_buff_size)
675 buff_size = *(action->comm.dst_buff_size);
682 smx_process_t SIMIX_comm_get_src_proc(smx_action_t action)
684 return action->comm.src_proc;
687 smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action)
689 return action->comm.dst_proc;
692 #ifdef HAVE_LATENCY_BOUND_TRACKING
694 * \brief verify if communication is latency bounded
695 * \param comm The communication
697 XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action)
699 if (action->comm.surf_comm){
700 DEBUG1("Getting latency limited for surf_action (%p)", action->comm.surf_comm);
701 action->latency_limited = surf_workstation_model->get_latency_limited(action->comm.surf_comm);
702 DEBUG1("Action limited is %d", action->latency_limited);
704 return action->latency_limited;
708 /******************************************************************************/
709 /* SIMIX_comm_copy_data callbacks */
710 /******************************************************************************/
711 static void (*SIMIX_comm_copy_data_callback) (smx_action_t, size_t) =
712 &SIMIX_comm_copy_pointer_callback;
715 SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, size_t))
717 SIMIX_comm_copy_data_callback = callback;
720 void SIMIX_comm_copy_pointer_callback(smx_action_t comm, size_t buff_size)
722 xbt_assert1((buff_size == sizeof(void *)),
723 "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
724 *(void **) (comm->comm.dst_buff) = comm->comm.src_buff;
727 void SIMIX_comm_copy_buffer_callback(smx_action_t comm, size_t buff_size)
729 memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size);
733 * \brief Copy the communication data from the sender's buffer to the receiver's one
734 * \param comm The communication
736 void SIMIX_comm_copy_data(smx_action_t comm)
738 size_t buff_size = comm->comm.src_buff_size;
739 /* If there is no data to be copy then return */
740 if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied == 1)
743 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
745 comm->comm.src_proc->smx_host->name, comm->comm.src_buff,
746 comm->comm.dst_proc->smx_host->name, comm->comm.dst_buff, buff_size);
748 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
749 if (comm->comm.dst_buff_size)
750 buff_size = MIN(buff_size, *(comm->comm.dst_buff_size));
752 /* Update the receiver's buffer size to the copied amount */
753 if (comm->comm.dst_buff_size)
754 *comm->comm.dst_buff_size = buff_size;
759 (*SIMIX_comm_copy_data_callback) (comm, buff_size);
761 /* Set the copied flag so we copy data only once */
762 /* (this function might be called from both communication ends) */
763 comm->comm.copied = 1;