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 action 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 action 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, action, smx_action_t){
134 if (action->comm.type == SIMIX_COMM_SEND) {
135 req_data = action->comm.src_data;
136 } else if (action->comm.type == SIMIX_COMM_RECEIVE) {
137 req_data = action->comm.dst_data;
139 if (action->comm.type == type && (!match_fun || match_fun(data, req_data))) {
140 DEBUG1("Found a matching communication action %p", action);
141 xbt_fifo_remove_item(rdv->comm_fifo, item);
142 xbt_fifo_free_item(item);
143 action->comm.refcount++;
144 action->comm.rdv = NULL;
147 DEBUG3("Sorry, communication action %p does not match our needs:"
148 " its type is %d but we are looking for a comm of type %d",
149 action, action->comm.type, type);
151 DEBUG0("No matching communication action found");
155 /******************************************************************************/
156 /* Comunication Actions */
157 /******************************************************************************/
160 * \brief Creates a new comunicate action
161 * \param type The type of request (comm_send, comm_recv)
162 * \return The new comunicate action
164 smx_action_t SIMIX_comm_new(e_smx_comm_type_t type)
168 /* alloc structures */
169 act = xbt_new0(s_smx_action_t, 1);
170 act->type = SIMIX_ACTION_COMMUNICATE;
171 act->state = SIMIX_WAITING;
172 act->request_list = xbt_fifo_new();
174 /* set communication */
175 act->comm.type = type;
176 act->comm.refcount = 1;
178 #ifdef HAVE_LATENCY_BOUND_TRACKING
179 //initialize with unknown value
180 act->latency_limited = -1;
184 act->category = NULL;
187 DEBUG1("Create communicate action %p", act);
193 * \brief Destroy a communicate action
194 * \param action The communicate action to be destroyed
196 void SIMIX_comm_destroy(smx_action_t action)
198 DEBUG2("Destroy action %p (refcount:%d)", action, action->comm.refcount);
200 if (action->comm.refcount <= 0)
201 xbt_die(bprintf("the refcount of comm %p is already 0 before decreasing it. That's a bug!",action));
203 action->comm.refcount--;
204 if (action->comm.refcount > 0)
206 DEBUG2("Really free communication %p; refcount is now %d", action,
207 action->comm.refcount);
209 #ifdef HAVE_LATENCY_BOUND_TRACKING
210 action->latency_limited = SIMIX_comm_is_latency_bounded( action ) ;
214 TRACE_smx_action_destroy(action);
218 xbt_free(action->name);
220 xbt_fifo_free(action->request_list);
222 SIMIX_comm_destroy_internal_actions(action);
227 void SIMIX_comm_destroy_internal_actions(smx_action_t action)
229 if (action->comm.surf_comm){
230 #ifdef HAVE_LATENCY_BOUND_TRACKING
231 action->latency_limited = SIMIX_comm_is_latency_bounded(action);
233 action->comm.surf_comm->model_type->action_unref(action->comm.surf_comm);
234 action->comm.surf_comm = NULL;
237 if (action->comm.src_timeout){
238 action->comm.src_timeout->model_type->action_unref(action->comm.src_timeout);
239 action->comm.src_timeout = NULL;
242 if (action->comm.dst_timeout){
243 action->comm.dst_timeout->model_type->action_unref(action->comm.dst_timeout);
244 action->comm.dst_timeout = NULL;
248 smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv,
249 double task_size, double rate,
250 void *src_buff, size_t src_buff_size,
251 int (*match_fun)(void *, void *), void *data)
255 /* Look for communication request matching our needs.
256 If it is not found then create it and push it into the rendez-vous point */
257 action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_RECEIVE, match_fun, data);
260 action = SIMIX_comm_new(SIMIX_COMM_SEND);
261 SIMIX_rdv_push(rdv, action);
263 action->state = SIMIX_READY;
264 action->comm.type = SIMIX_COMM_READY;
267 /* Setup the communication request */
268 action->comm.src_proc = src_proc;
269 action->comm.task_size = task_size;
270 action->comm.rate = rate;
271 action->comm.src_buff = src_buff;
272 action->comm.src_buff_size = src_buff_size;
273 action->comm.src_data = data;
276 action->state = SIMIX_RUNNING;
280 SIMIX_comm_start(action);
284 smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv,
285 void *dst_buff, size_t *dst_buff_size,
286 int (*match_fun)(void *, void *), void *data)
290 /* Look for communication request matching our needs.
291 * If it is not found then create it and push it into the rendez-vous point
293 action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_SEND, match_fun, data);
296 action = SIMIX_comm_new(SIMIX_COMM_RECEIVE);
297 SIMIX_rdv_push(rdv, action);
299 action->state = SIMIX_READY;
300 action->comm.type = SIMIX_COMM_READY;
303 /* Setup communication request */
304 action->comm.dst_proc = dst_proc;
305 action->comm.dst_buff = dst_buff;
306 action->comm.dst_buff_size = dst_buff_size;
307 action->comm.dst_data = data;
310 action->state = SIMIX_RUNNING;
314 SIMIX_comm_start(action);
318 void SIMIX_pre_comm_wait(smx_req_t req)
320 smx_action_t action = req->comm_wait.comm;
321 double timeout = req->comm_wait.timeout;
324 /* Associate this request to the action */
325 xbt_fifo_push(action->request_list, req);
326 req->issuer->waiting_action = action;
329 action->state = SIMIX_DONE;
330 SIMIX_comm_finish(action);
334 /* If the action has already finish perform the error handling, */
335 /* otherwise set up a waiting timeout on the right side */
336 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
337 SIMIX_comm_finish(action);
338 } 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 */
339 sleep = surf_workstation_model->extension.workstation.sleep(req->issuer->smx_host->host, timeout);
340 surf_workstation_model->action_data_set(sleep, action);
342 if (req->issuer == action->comm.src_proc)
343 action->comm.src_timeout = sleep;
345 action->comm.dst_timeout = sleep;
349 void SIMIX_pre_comm_test(smx_req_t req)
351 smx_action_t action = req->comm_test.comm;
352 req->comm_test.result = (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING);
354 if (req->comm_test.result) {
355 xbt_fifo_push(action->request_list, req);
356 SIMIX_comm_finish(action);
358 SIMIX_request_answer(req);
362 void SIMIX_pre_comm_testany(smx_req_t req, unsigned int idx)
366 xbt_dynar_t actions = req->comm_testany.comms;
367 req->comm_testany.result = -1;
371 SIMIX_request_answer(req);
373 action = xbt_dynar_get_as(actions, idx, smx_action_t);
374 xbt_fifo_push(action->request_list, req);
375 action->state = SIMIX_DONE;
376 SIMIX_comm_finish(action);
381 xbt_dynar_foreach(req->comm_testany.comms,cursor,action) {
382 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
383 req->comm_testany.result = cursor;
384 xbt_fifo_push(action->request_list, req);
385 SIMIX_comm_finish(action);
389 SIMIX_request_answer(req);
392 void SIMIX_pre_comm_waitany(smx_req_t req, unsigned int idx)
395 unsigned int cursor = 0;
396 xbt_dynar_t actions = req->comm_waitany.comms;
399 action = xbt_dynar_get_as(actions, idx, smx_action_t);
400 xbt_fifo_push(action->request_list, req);
401 req->comm_waitany.result = idx;
402 action->state = SIMIX_DONE;
403 SIMIX_comm_finish(action);
407 xbt_dynar_foreach(actions, cursor, action){
408 /* Associate this request to the action */
409 xbt_fifo_push(action->request_list, req);
410 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){
411 SIMIX_comm_finish(action);
417 void SIMIX_waitany_req_remove_from_actions(smx_req_t req)
420 unsigned int cursor = 0;
421 xbt_dynar_t actions = req->comm_waitany.comms;
423 xbt_dynar_foreach(actions, cursor, action){
424 xbt_fifo_remove(action->request_list, req);
429 * \brief Start the simulation of a communication request
430 * \param action The communication action
432 static XBT_INLINE void SIMIX_comm_start(smx_action_t action)
434 /* If both the sender and the receiver are already there, start the communication */
435 if (action->state == SIMIX_READY) {
436 smx_host_t sender = action->comm.src_proc->smx_host;
437 smx_host_t receiver = action->comm.dst_proc->smx_host;
439 DEBUG3("Starting communication %p from '%s' to '%s'", action,
440 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
442 action->comm.surf_comm = surf_workstation_model->extension.workstation.
443 communicate(sender->host, receiver->host, action->comm.task_size, action->comm.rate);
445 surf_workstation_model->action_data_set(action->comm.surf_comm, action);
447 action->state = SIMIX_RUNNING;
450 TRACE_smx_action_communicate(action, action->comm.src_proc);
453 /* If a link is failed, detect it immediately */
454 if (surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) {
455 DEBUG2("Communication from '%s' to '%s' failed to start because of a link failure",
456 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
457 action->state = SIMIX_LINK_FAILURE;
458 SIMIX_comm_destroy_internal_actions(action);
461 /* If any of the process is suspend, create the action but stop its execution,
462 it will be restarted when the sender process resume */
463 if (SIMIX_process_is_suspended(action->comm.src_proc) ||
464 SIMIX_process_is_suspended(action->comm.dst_proc)) {
465 /* FIXME: check what should happen with the action state */
466 surf_workstation_model->suspend(action->comm.surf_comm);
471 void SIMIX_comm_finish(smx_action_t action)
475 while ((req = xbt_fifo_shift(action->request_list))) {
477 /* If a waitany request is waiting for this action to finish, then remove
478 it from the other actions in the waitany list. Afterwards, get the
479 position of the actual action in the waitany request's actions dynar and
480 return it as the result of the call */
481 if (req->call == REQ_COMM_WAITANY) {
482 SIMIX_waitany_req_remove_from_actions(req);
484 req->comm_waitany.result = xbt_dynar_search(req->comm_waitany.comms, &action);
487 /* If the action is still in a rendez-vous point then remove from it */
488 if (action->comm.rdv)
489 SIMIX_rdv_remove(action->comm.rdv, action);
491 DEBUG1("SIMIX_comm_finish: action state = %d", action->state);
493 /* Check out for errors */
494 switch (action->state) {
497 DEBUG1("Communication %p complete!", action);
498 SIMIX_comm_copy_data(action);
501 case SIMIX_SRC_TIMEOUT:
503 THROW0(timeout_error, 0, "Communication timeouted because of sender");
505 CATCH(req->issuer->running_ctx->exception) {
506 req->issuer->doexception = 1;
510 case SIMIX_DST_TIMEOUT:
512 THROW0(timeout_error, 0, "Communication timeouted because of receiver");
514 CATCH(req->issuer->running_ctx->exception) {
515 req->issuer->doexception = 1;
519 case SIMIX_SRC_HOST_FAILURE:
521 if (req->issuer == action->comm.src_proc)
522 THROW0(host_error, 0, "Host failed");
524 THROW0(network_error, 0, "Remote peer failed");
526 CATCH(req->issuer->running_ctx->exception) {
527 req->issuer->doexception = 1;
531 case SIMIX_DST_HOST_FAILURE:
533 if (req->issuer == action->comm.dst_proc)
534 THROW0(host_error, 0, "Host failed");
536 THROW0(network_error, 0, "Remote peer failed");
538 CATCH(req->issuer->running_ctx->exception) {
539 req->issuer->doexception = 1;
543 case SIMIX_LINK_FAILURE:
545 DEBUG5("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)",
546 action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name,
547 req->issuer->name, req->issuer);
548 THROW0(network_error, 0, "Link failure");
550 CATCH(req->issuer->running_ctx->exception) {
551 req->issuer->doexception = 1;
559 /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
560 if (req->issuer->doexception) {
561 if (req->call == REQ_COMM_WAITANY) {
562 req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_waitany.comms, &action);
564 else if (req->call == REQ_COMM_TESTANY) {
565 req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_testany.comms, &action);
569 req->issuer->waiting_action = NULL;
570 SIMIX_request_answer(req);
574 void SIMIX_post_comm(smx_action_t action)
576 /* Update action state */
577 if (action->comm.src_timeout &&
578 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_DONE)
579 action->state = SIMIX_SRC_TIMEOUT;
580 else if (action->comm.dst_timeout &&
581 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_DONE)
582 action->state = SIMIX_DST_TIMEOUT;
583 else if (action->comm.src_timeout &&
584 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_FAILED)
585 action->state = SIMIX_SRC_HOST_FAILURE;
586 else if (action->comm.dst_timeout &&
587 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED)
588 action->state = SIMIX_DST_HOST_FAILURE;
589 else if (action->comm.surf_comm &&
590 surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED)
591 action->state = SIMIX_LINK_FAILURE;
593 action->state = SIMIX_DONE;
595 DEBUG1("SIMIX_post_comm: action state = %d", action->state);
597 /* After this point the surf actions associated with the simix communicate
598 action are no longer needed, thus we delete them. */
599 SIMIX_comm_destroy_internal_actions(action);
601 /* If there are requests associated with the action, then answer them */
602 if (xbt_fifo_size(action->request_list))
603 SIMIX_comm_finish(action);
606 void SIMIX_comm_cancel(smx_action_t action)
608 /* If the action is a waiting state means that it is still in a rdv */
609 /* so remove from it and delete it */
610 if (action->state == SIMIX_WAITING) {
611 SIMIX_rdv_remove(action->comm.rdv, action);
612 action->state = SIMIX_FAILED;
614 surf_workstation_model->action_cancel(action->comm.surf_comm);
618 void SIMIX_comm_suspend(smx_action_t action)
620 /*FIXME: shall we suspend also the timeout actions? */
621 surf_workstation_model->suspend(action->comm.surf_comm);
624 void SIMIX_comm_resume(smx_action_t action)
626 /*FIXME: check what happen with the timeouts */
627 surf_workstation_model->resume(action->comm.surf_comm);
631 /************* Action Getters **************/
634 * \brief get the amount remaining from the communication
635 * \param action The communication
637 double SIMIX_comm_get_remains(smx_action_t action)
641 switch (action->state) {
644 remains = surf_workstation_model->get_remains(action->comm.surf_comm);
649 remains = 0; /*FIXME: check what should be returned */
653 remains = 0; /*FIXME: is this correct? */
659 e_smx_state_t SIMIX_comm_get_state(smx_action_t action)
661 return action->state;
665 * \brief Return the user data associated to the sender of the communication
666 * \param action The communication
667 * \return the user data
669 void* SIMIX_comm_get_src_data(smx_action_t action)
671 return action->comm.src_data;
675 * \brief Return the user data associated to the receiver of the communication
676 * \param action The communication
677 * \return the user data
679 void* SIMIX_comm_get_dst_data(smx_action_t action)
681 return action->comm.dst_data;
684 void* SIMIX_comm_get_src_buff(smx_action_t action)
686 return action->comm.src_buff;
689 void* SIMIX_comm_get_dst_buff(smx_action_t action)
691 return action->comm.dst_buff;
694 size_t SIMIX_comm_get_src_buff_size(smx_action_t action)
696 return action->comm.src_buff_size;
699 size_t SIMIX_comm_get_dst_buff_size(smx_action_t action)
703 if (action->comm.dst_buff_size)
704 buff_size = *(action->comm.dst_buff_size);
711 smx_process_t SIMIX_comm_get_src_proc(smx_action_t action)
713 return action->comm.src_proc;
716 smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action)
718 return action->comm.dst_proc;
721 #ifdef HAVE_LATENCY_BOUND_TRACKING
723 * \brief verify if communication is latency bounded
724 * \param comm The communication
726 XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action)
728 if (action->comm.surf_comm){
729 DEBUG1("Getting latency limited for surf_action (%p)", action->comm.surf_comm);
730 action->latency_limited = surf_workstation_model->get_latency_limited(action->comm.surf_comm);
731 DEBUG1("Action limited is %d", action->latency_limited);
733 return action->latency_limited;
737 /******************************************************************************/
738 /* SIMIX_comm_copy_data callbacks */
739 /******************************************************************************/
740 static void (*SIMIX_comm_copy_data_callback) (smx_action_t, size_t) =
741 &SIMIX_comm_copy_pointer_callback;
744 SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, size_t))
746 SIMIX_comm_copy_data_callback = callback;
749 void SIMIX_comm_copy_pointer_callback(smx_action_t comm, size_t buff_size)
751 xbt_assert1((buff_size == sizeof(void *)),
752 "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
753 *(void **) (comm->comm.dst_buff) = comm->comm.src_buff;
756 void SIMIX_comm_copy_buffer_callback(smx_action_t comm, size_t buff_size)
758 memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size);
762 * \brief Copy the communication data from the sender's buffer to the receiver's one
763 * \param comm The communication
765 void SIMIX_comm_copy_data(smx_action_t comm)
767 size_t buff_size = comm->comm.src_buff_size;
768 /* If there is no data to be copy then return */
769 if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied == 1)
772 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
774 comm->comm.src_proc->smx_host->name, comm->comm.src_buff,
775 comm->comm.dst_proc->smx_host->name, comm->comm.dst_buff, buff_size);
777 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
778 if (comm->comm.dst_buff_size)
779 buff_size = MIN(buff_size, *(comm->comm.dst_buff_size));
781 /* Update the receiver's buffer size to the copied amount */
782 if (comm->comm.dst_buff_size)
783 *comm->comm.dst_buff_size = buff_size;
788 (*SIMIX_comm_copy_data_callback) (comm, buff_size);
790 /* Set the copied flag so we copy data only once */
791 /* (this function might be called from both communication ends) */
792 comm->comm.copied = 1;