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)
366 req->comm_testany.result = -1;
367 xbt_dynar_foreach(req->comm_testany.comms,cursor,action) {
368 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
369 req->comm_testany.result = cursor;
370 xbt_fifo_push(action->request_list, req);
371 SIMIX_comm_finish(action);
375 SIMIX_request_answer(req);
378 void SIMIX_pre_comm_waitany(smx_req_t req)
381 unsigned int cursor = 0;
382 xbt_dynar_t actions = req->comm_waitany.comms;
383 xbt_dynar_foreach(actions, cursor, action){
384 /* Associate this request to the action */
385 xbt_fifo_push(action->request_list, req);
386 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){
387 SIMIX_comm_finish(action);
393 void SIMIX_waitany_req_remove_from_actions(smx_req_t req)
396 unsigned int cursor = 0;
397 xbt_dynar_t actions = req->comm_waitany.comms;
399 xbt_dynar_foreach(actions, cursor, action){
400 xbt_fifo_remove(action->request_list, req);
405 * \brief Start the simulation of a communication request
406 * \param action The communication action
408 static XBT_INLINE void SIMIX_comm_start(smx_action_t action)
410 /* If both the sender and the receiver are already there, start the communication */
411 if (action->state == SIMIX_READY) {
412 smx_host_t sender = action->comm.src_proc->smx_host;
413 smx_host_t receiver = action->comm.dst_proc->smx_host;
415 DEBUG3("Starting communication %p from '%s' to '%s'", action,
416 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
418 action->comm.surf_comm = surf_workstation_model->extension.workstation.
419 communicate(sender->host, receiver->host, action->comm.task_size, action->comm.rate);
421 surf_workstation_model->action_data_set(action->comm.surf_comm, action);
423 action->state = SIMIX_RUNNING;
426 TRACE_smx_action_communicate(action, action->comm.src_proc);
429 /* If a link is failed, detect it immediately */
430 if (surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) {
431 DEBUG2("Communication from '%s' to '%s' failed to start because of a link failure",
432 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
433 action->state = SIMIX_LINK_FAILURE;
434 SIMIX_comm_destroy_internal_actions(action);
437 /* If any of the process is suspend, create the action but stop its execution,
438 it will be restarted when the sender process resume */
439 if (SIMIX_process_is_suspended(action->comm.src_proc) ||
440 SIMIX_process_is_suspended(action->comm.dst_proc)) {
441 /* FIXME: check what should happen with the action state */
442 surf_workstation_model->suspend(action->comm.surf_comm);
447 void SIMIX_comm_finish(smx_action_t action)
451 while ((req = xbt_fifo_shift(action->request_list))) {
453 /* If a waitany request is waiting for this action to finish, then remove
454 it from the other actions in the waitany list. Afterwards, get the
455 position of the actual action in the waitany request's actions dynar and
456 return it as the result of the call */
457 if (req->call == REQ_COMM_WAITANY) {
458 SIMIX_waitany_req_remove_from_actions(req);
459 req->comm_waitany.result = xbt_dynar_search(req->comm_waitany.comms, &action);
462 /* If the action is still in a rendez-vous point then remove from it */
463 if (action->comm.rdv)
464 SIMIX_rdv_remove(action->comm.rdv, action);
466 DEBUG1("SIMIX_comm_finish: action state = %d", action->state);
468 /* Check out for errors */
469 switch (action->state) {
472 DEBUG1("Communication %p complete!", action);
473 SIMIX_comm_copy_data(action);
476 case SIMIX_SRC_TIMEOUT:
478 THROW0(timeout_error, 0, "Communication timeouted because of sender");
480 CATCH(req->issuer->running_ctx->exception) {
481 req->issuer->doexception = 1;
485 case SIMIX_DST_TIMEOUT:
487 THROW0(timeout_error, 0, "Communication timeouted because of receiver");
489 CATCH(req->issuer->running_ctx->exception) {
490 req->issuer->doexception = 1;
494 case SIMIX_SRC_HOST_FAILURE:
496 if (req->issuer == action->comm.src_proc)
497 THROW0(host_error, 0, "Host failed");
499 THROW0(network_error, 0, "Remote peer failed");
501 CATCH(req->issuer->running_ctx->exception) {
502 req->issuer->doexception = 1;
506 case SIMIX_DST_HOST_FAILURE:
508 if (req->issuer == action->comm.dst_proc)
509 THROW0(host_error, 0, "Host failed");
511 THROW0(network_error, 0, "Remote peer failed");
513 CATCH(req->issuer->running_ctx->exception) {
514 req->issuer->doexception = 1;
518 case SIMIX_LINK_FAILURE:
520 DEBUG5("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)",
521 action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name,
522 req->issuer->name, req->issuer);
523 THROW0(network_error, 0, "Link failure");
525 CATCH(req->issuer->running_ctx->exception) {
526 req->issuer->doexception = 1;
534 /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
535 if (req->issuer->doexception) {
536 if (req->call == REQ_COMM_WAITANY) {
537 req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_waitany.comms, &action);
539 else if (req->call == REQ_COMM_TESTANY) {
540 req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_testany.comms, &action);
544 req->issuer->waiting_action = NULL;
545 SIMIX_request_answer(req);
549 void SIMIX_post_comm(smx_action_t action)
551 /* Update action state */
552 if (action->comm.src_timeout &&
553 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_DONE)
554 action->state = SIMIX_SRC_TIMEOUT;
555 else if (action->comm.dst_timeout &&
556 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_DONE)
557 action->state = SIMIX_DST_TIMEOUT;
558 else if (action->comm.src_timeout &&
559 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_FAILED)
560 action->state = SIMIX_SRC_HOST_FAILURE;
561 else if (action->comm.dst_timeout &&
562 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED)
563 action->state = SIMIX_DST_HOST_FAILURE;
564 else if (action->comm.surf_comm &&
565 surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED)
566 action->state = SIMIX_LINK_FAILURE;
568 action->state = SIMIX_DONE;
570 DEBUG1("SIMIX_post_comm: action state = %d", action->state);
572 /* After this point the surf actions associated with the simix communicate
573 action are no longer needed, thus we delete them. */
574 SIMIX_comm_destroy_internal_actions(action);
576 /* If there are requests associated with the action, then answer them */
577 if (xbt_fifo_size(action->request_list))
578 SIMIX_comm_finish(action);
581 void SIMIX_comm_cancel(smx_action_t action)
583 /* If the action is a waiting state means that it is still in a rdv */
584 /* so remove from it and delete it */
585 if (action->state == SIMIX_WAITING) {
586 SIMIX_rdv_remove(action->comm.rdv, action);
587 action->state = SIMIX_FAILED;
589 surf_workstation_model->action_cancel(action->comm.surf_comm);
593 void SIMIX_comm_suspend(smx_action_t action)
595 /*FIXME: shall we suspend also the timeout actions? */
596 surf_workstation_model->suspend(action->comm.surf_comm);
599 void SIMIX_comm_resume(smx_action_t action)
601 /*FIXME: check what happen with the timeouts */
602 surf_workstation_model->resume(action->comm.surf_comm);
606 /************* Action Getters **************/
609 * \brief get the amount remaining from the communication
610 * \param action The communication
612 double SIMIX_comm_get_remains(smx_action_t action)
616 switch (action->state) {
619 remains = surf_workstation_model->get_remains(action->comm.surf_comm);
624 remains = 0; /*FIXME: check what should be returned */
628 remains = 0; /*FIXME: is this correct? */
634 e_smx_state_t SIMIX_comm_get_state(smx_action_t action)
636 return action->state;
640 * \brief Return the user data associated to the sender of the communication
641 * \param action The communication
642 * \return the user data
644 void* SIMIX_comm_get_src_data(smx_action_t action)
646 return action->comm.src_data;
650 * \brief Return the user data associated to the receiver of the communication
651 * \param action The communication
652 * \return the user data
654 void* SIMIX_comm_get_dst_data(smx_action_t action)
656 return action->comm.dst_data;
659 void* SIMIX_comm_get_src_buff(smx_action_t action)
661 return action->comm.src_buff;
664 void* SIMIX_comm_get_dst_buff(smx_action_t action)
666 return action->comm.dst_buff;
669 size_t SIMIX_comm_get_src_buff_size(smx_action_t action)
671 return action->comm.src_buff_size;
674 size_t SIMIX_comm_get_dst_buff_size(smx_action_t action)
678 if (action->comm.dst_buff_size)
679 buff_size = *(action->comm.dst_buff_size);
686 smx_process_t SIMIX_comm_get_src_proc(smx_action_t action)
688 return action->comm.src_proc;
691 smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action)
693 return action->comm.dst_proc;
696 #ifdef HAVE_LATENCY_BOUND_TRACKING
698 * \brief verify if communication is latency bounded
699 * \param comm The communication
701 XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action)
703 if (action->comm.surf_comm){
704 DEBUG1("Getting latency limited for surf_action (%p)", action->comm.surf_comm);
705 action->latency_limited = surf_workstation_model->get_latency_limited(action->comm.surf_comm);
706 DEBUG1("Action limited is %d", action->latency_limited);
708 return action->latency_limited;
712 /******************************************************************************/
713 /* SIMIX_comm_copy_data callbacks */
714 /******************************************************************************/
715 static void (*SIMIX_comm_copy_data_callback) (smx_action_t, size_t) =
716 &SIMIX_comm_copy_pointer_callback;
719 SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, size_t))
721 SIMIX_comm_copy_data_callback = callback;
724 void SIMIX_comm_copy_pointer_callback(smx_action_t comm, size_t buff_size)
726 xbt_assert1((buff_size == sizeof(void *)),
727 "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
728 *(void **) (comm->comm.dst_buff) = comm->comm.src_buff;
731 void SIMIX_comm_copy_buffer_callback(smx_action_t comm, size_t buff_size)
733 memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size);
737 * \brief Copy the communication data from the sender's buffer to the receiver's one
738 * \param comm The communication
740 void SIMIX_comm_copy_data(smx_action_t comm)
742 size_t buff_size = comm->comm.src_buff_size;
743 /* If there is no data to be copy then return */
744 if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied == 1)
747 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
749 comm->comm.src_proc->smx_host->name, comm->comm.src_buff,
750 comm->comm.dst_proc->smx_host->name, comm->comm.dst_buff, buff_size);
752 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
753 if (comm->comm.dst_buff_size)
754 buff_size = MIN(buff_size, *(comm->comm.dst_buff_size));
756 /* Update the receiver's buffer size to the copied amount */
757 if (comm->comm.dst_buff_size)
758 *comm->comm.dst_buff_size = buff_size;
763 (*SIMIX_comm_copy_data_callback) (comm, buff_size);
765 /* Set the copied flag so we copy data only once */
766 /* (this function might be called from both communication ends) */
767 comm->comm.copied = 1;