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;
132 xbt_fifo_foreach(rdv->comm_fifo, item, req, smx_action_t){
133 if(req->comm.type == type && (!match_fun || match_fun(data, req->comm.data))){
134 xbt_fifo_remove_item(rdv->comm_fifo, item);
135 req->comm.refcount++;
136 req->comm.rdv = NULL;
141 DEBUG0("Communication request not found");
145 /******************************************************************************/
146 /* Comunication Actions */
147 /******************************************************************************/
150 * \brief Creates a new comunicate action
151 * \param type The type of request (comm_send, comm_recv)
152 * \return The new comunicate action
154 smx_action_t SIMIX_comm_new(e_smx_comm_type_t type)
158 /* alloc structures */
159 act = xbt_new0(s_smx_action_t, 1);
160 act->type = SIMIX_ACTION_COMMUNICATE;
161 act->state = SIMIX_WAITING;
162 act->request_list = xbt_fifo_new();
164 /* set communication */
165 act->comm.type = type;
166 act->comm.refcount = 1;
169 act->category = NULL;
172 DEBUG1("Create communicate action %p", act);
178 * \brief Destroy a communicate action
179 * \param action The communicate action to be destroyed
181 void SIMIX_comm_destroy(smx_action_t action)
183 DEBUG1("Destroy action %p", action);
185 if (action->comm.refcount <= 0)
186 xbt_die(bprintf("the refcount of comm %p is already 0 before decreasing it. That's a bug!",action));
188 #ifdef HAVE_LATENCY_BOUND_TRACKING
189 //save is latency limited flag to use afterwards
190 if (action->comm.surf_comm) {
191 DEBUG2("adding key %p with latency limited value %d to the dict", action,
192 SIMIX_comm_is_latency_bounded(action));
193 xbt_dicti_set(simix_global->latency_limited_dict, (uintptr_t) action,
194 SIMIX_comm_is_latency_bounded(action));
198 action->comm.refcount--;
199 if (action->comm.refcount > 0)
201 VERB2("Really free communication %p; refcount is now %d", action,
202 action->comm.refcount);
205 TRACE_smx_action_destroy(action);
209 xbt_free(action->name);
211 xbt_fifo_free(action->request_list);
213 SIMIX_comm_destroy_internal_actions(action);
218 void SIMIX_comm_destroy_internal_actions(smx_action_t action)
220 if (action->comm.surf_comm){
221 action->comm.surf_comm->model_type->action_unref(action->comm.surf_comm);
222 action->comm.surf_comm = NULL;
225 if (action->comm.src_timeout){
226 action->comm.src_timeout->model_type->action_unref(action->comm.src_timeout);
227 action->comm.src_timeout = NULL;
230 if (action->comm.dst_timeout){
231 action->comm.dst_timeout->model_type->action_unref(action->comm.dst_timeout);
232 action->comm.dst_timeout = NULL;
236 smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv,
237 double task_size, double rate,
238 void *src_buff, size_t src_buff_size,
239 int (*match_fun)(void *, void *), void *data)
243 /* Look for communication request matching our needs.
244 If it is not found then create it and push it into the rendez-vous point */
245 action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_RECEIVE, match_fun, data);
248 action = SIMIX_comm_new(SIMIX_COMM_SEND);
249 SIMIX_rdv_push(rdv, action);
251 action->state = SIMIX_READY;
252 action->comm.type = SIMIX_COMM_READY;
255 /* Setup the communication request */
256 action->comm.src_proc = src_proc;
257 action->comm.task_size = task_size;
258 action->comm.rate = rate;
259 action->comm.src_buff = src_buff;
260 action->comm.src_buff_size = src_buff_size;
261 action->comm.data = data;
264 action->state = SIMIX_RUNNING;
268 SIMIX_comm_start(action);
272 smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv,
273 void *dst_buff, size_t *dst_buff_size,
274 int (*match_fun)(void *, void *), void *data)
278 /* Look for communication request matching our needs.
279 * If it is not found then create it and push it into the rendez-vous point
281 action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_SEND, match_fun, data);
284 action = SIMIX_comm_new(SIMIX_COMM_RECEIVE);
285 SIMIX_rdv_push(rdv, action);
287 action->state = SIMIX_READY;
288 action->comm.type = SIMIX_COMM_READY;
291 /* Setup communication request */
292 action->comm.dst_proc = dst_proc;
293 action->comm.dst_buff = dst_buff;
294 action->comm.dst_buff_size = dst_buff_size;
297 action->state = SIMIX_RUNNING;
301 SIMIX_comm_start(action);
305 void SIMIX_pre_comm_wait(smx_req_t req)
307 smx_action_t action = req->comm_wait.comm;
308 double timeout = req->comm_wait.timeout;
311 /* Associate this request to the action */
312 xbt_fifo_push(action->request_list, req);
313 req->issuer->waiting_action = action;
316 action->state = SIMIX_DONE;
317 SIMIX_comm_finish(action);
320 /* If the action has already finish perform the error handling, */
321 /* otherwise set up a waiting timeout on the right side */
322 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
323 SIMIX_comm_finish(action);
324 } 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 */
325 sleep = surf_workstation_model->extension.workstation.sleep(req->issuer->smx_host->host, timeout);
326 surf_workstation_model->action_data_set(sleep, action);
328 if (req->issuer == action->comm.src_proc)
329 action->comm.src_timeout = sleep;
331 action->comm.dst_timeout = sleep;
335 void SIMIX_pre_comm_test(smx_req_t req)
337 smx_action_t action = req->comm_test.comm;
338 req->comm_test.result = (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING);
340 if (req->comm_test.result) {
341 xbt_fifo_push(action->request_list, req);
342 SIMIX_comm_finish(action);
345 SIMIX_request_answer(req);
349 void SIMIX_pre_comm_testany(smx_req_t req)
353 req->comm_testany.result = -1;
354 xbt_dynar_foreach(req->comm_testany.comms,cursor,action) {
355 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
356 req->comm_testany.result = cursor;
357 xbt_fifo_push(action->request_list, req);
358 SIMIX_comm_finish(action);
362 SIMIX_request_answer(req);
365 void SIMIX_pre_comm_waitany(smx_req_t req)
368 unsigned int cursor = 0;
369 xbt_dynar_t actions = req->comm_waitany.comms;
370 xbt_dynar_foreach(actions, cursor, action){
371 /* Associate this request to the action */
372 xbt_fifo_push(action->request_list, req);
373 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){
374 SIMIX_comm_finish(action);
380 void SIMIX_waitany_req_remove_from_actions(smx_req_t req)
383 unsigned int cursor = 0;
384 xbt_dynar_t actions = req->comm_waitany.comms;
386 xbt_dynar_foreach(actions, cursor, action){
387 xbt_fifo_remove(action->request_list, req);
392 * \brief Start the simulation of a communication request
393 * \param action The communication action
395 static XBT_INLINE void SIMIX_comm_start(smx_action_t action)
397 /* If both the sender and the receiver are already there, start the communication */
398 if (action->state == SIMIX_READY) {
399 smx_host_t sender = action->comm.src_proc->smx_host;
400 smx_host_t receiver = action->comm.dst_proc->smx_host;
402 DEBUG3("Starting communication %p from '%s' to '%s'", action,
403 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
405 action->comm.surf_comm = surf_workstation_model->extension.workstation.
406 communicate(sender->host, receiver->host, action->comm.task_size, action->comm.rate);
408 surf_workstation_model->action_data_set(action->comm.surf_comm, action);
410 action->state = SIMIX_RUNNING;
413 TRACE_smx_action_communicate(action, action->comm.src_proc);
416 /* If a link is failed, detect it immediately */
417 if (surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) {
418 DEBUG2("Communication from '%s' to '%s' failed to start because of a link failure",
419 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
420 action->state = SIMIX_LINK_FAILURE;
421 SIMIX_comm_destroy_internal_actions(action);
424 /* If any of the process is suspend, create the action but stop its execution,
425 it will be restarted when the sender process resume */
426 if (SIMIX_process_is_suspended(action->comm.src_proc) ||
427 SIMIX_process_is_suspended(action->comm.dst_proc)) {
428 /* FIXME: check what should happen with the action state */
429 surf_workstation_model->suspend(action->comm.surf_comm);
434 void SIMIX_comm_finish(smx_action_t action)
438 while ((req = xbt_fifo_shift(action->request_list))) {
440 /* If a waitany request is waiting for this action to finish, then remove
441 it from the other actions in the waitany list. Afterwards, get the
442 position of the actual action in the waitany request's actions dynar and
443 return it as the result of the call */
444 if (req->call == REQ_COMM_WAITANY) {
445 SIMIX_waitany_req_remove_from_actions(req);
446 req->comm_waitany.result = xbt_dynar_search(req->comm_waitany.comms, &action);
449 /* If the action is still in a rendez-vous point then remove from it */
450 if (action->comm.rdv)
451 SIMIX_rdv_remove(action->comm.rdv, action);
453 DEBUG1("SIMIX_comm_finish: action state = %d", action->state);
455 /* Check out for errors */
456 switch (action->state) {
459 DEBUG1("Communication %p complete!", action);
460 SIMIX_comm_copy_data(action);
463 case SIMIX_SRC_TIMEOUT:
465 THROW0(timeout_error, 0, "Communication timeouted because of sender");
467 CATCH(req->issuer->running_ctx->exception) {
468 req->issuer->doexception = 1;
472 case SIMIX_DST_TIMEOUT:
474 THROW0(timeout_error, 0, "Communication timeouted because of receiver");
476 CATCH(req->issuer->running_ctx->exception) {
477 req->issuer->doexception = 1;
481 case SIMIX_SRC_HOST_FAILURE:
483 if (req->issuer == action->comm.src_proc)
484 THROW0(host_error, 0, "Host failed");
486 THROW0(network_error, 0, "Remote peer failed");
488 CATCH(req->issuer->running_ctx->exception) {
489 req->issuer->doexception = 1;
493 case SIMIX_DST_HOST_FAILURE:
495 if (req->issuer == action->comm.dst_proc)
496 THROW0(host_error, 0, "Host failed");
498 THROW0(network_error, 0, "Remote peer failed");
500 CATCH(req->issuer->running_ctx->exception) {
501 req->issuer->doexception = 1;
505 case SIMIX_LINK_FAILURE:
507 DEBUG5("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)",
508 action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name,
509 req->issuer->name, req->issuer);
510 THROW0(network_error, 0, "Link failure");
512 CATCH(req->issuer->running_ctx->exception) {
513 req->issuer->doexception = 1;
520 req->issuer->waiting_action = NULL;
521 SIMIX_request_answer(req);
525 void SIMIX_post_comm(smx_action_t action)
527 /* Update action state */
528 if (action->comm.src_timeout &&
529 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_DONE)
530 action->state = SIMIX_SRC_TIMEOUT;
531 else if (action->comm.dst_timeout &&
532 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_DONE)
533 action->state = SIMIX_DST_TIMEOUT;
534 else if (action->comm.src_timeout &&
535 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_FAILED)
536 action->state = SIMIX_SRC_HOST_FAILURE;
537 else if (action->comm.dst_timeout &&
538 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED)
539 action->state = SIMIX_DST_HOST_FAILURE;
540 else if (action->comm.surf_comm &&
541 surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED)
542 action->state = SIMIX_LINK_FAILURE;
544 action->state = SIMIX_DONE;
546 DEBUG1("SIMIX_post_comm: action state = %d", action->state);
548 /* After this point the surf actions associated with the simix communicate
549 action are no longer needed, thus we delete them. */
550 SIMIX_comm_destroy_internal_actions(action);
552 /* If there are requests associated with the action, then answer them */
553 if (xbt_fifo_size(action->request_list))
554 SIMIX_comm_finish(action);
557 void SIMIX_comm_cancel(smx_action_t action)
559 /* If the action is a waiting state means that it is still in a rdv */
560 /* so remove from it and delete it */
561 if (action->state == SIMIX_WAITING) {
562 SIMIX_rdv_remove(action->comm.rdv, action);
563 action->state = SIMIX_FAILED;
565 surf_workstation_model->action_cancel(action->comm.surf_comm);
569 void SIMIX_comm_suspend(smx_action_t action)
571 /*FIXME: shall we suspend also the timeout actions? */
572 surf_workstation_model->suspend(action->comm.surf_comm);
575 void SIMIX_comm_resume(smx_action_t action)
577 /*FIXME: check what happen with the timeouts */
578 surf_workstation_model->resume(action->comm.surf_comm);
582 /************* Action Getters **************/
585 * \brief get the amount remaining from the communication
586 * \param action The communication
588 double SIMIX_comm_get_remains(smx_action_t action)
592 switch (action->state) {
595 remains = surf_workstation_model->get_remains(action->comm.surf_comm);
600 remains = 0; /*FIXME: check what should be returned */
604 remains = 0; /*FIXME: is this correct? */
610 e_smx_state_t SIMIX_comm_get_state(smx_action_t action)
612 return action->state;
616 * \brief Return the user data associated to the communication
617 * \param action The communication
618 * \return the user data
620 void* SIMIX_comm_get_data(smx_action_t action)
622 return action->comm.data;
625 void* SIMIX_comm_get_src_buff(smx_action_t action)
627 return action->comm.src_buff;
630 void* SIMIX_comm_get_dst_buff(smx_action_t action)
632 return action->comm.dst_buff;
635 size_t SIMIX_comm_get_src_buff_size(smx_action_t action)
637 return action->comm.src_buff_size;
640 size_t SIMIX_comm_get_dst_buff_size(smx_action_t action)
644 if (action->comm.dst_buff_size)
645 buff_size = *(action->comm.dst_buff_size);
652 smx_process_t SIMIX_comm_get_src_proc(smx_action_t action)
654 return action->comm.src_proc;
657 smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action)
659 return action->comm.dst_proc;
662 #ifdef HAVE_LATENCY_BOUND_TRACKING
664 * \brief verify if communication is latency bounded
665 * \param comm The communication
667 XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action)
669 //try to find comm on the list of finished flows
672 xbt_dict_cursor_t cursor;
673 xbt_dict_foreach(simix_global->latency_limited_dict, cursor, key, data) {
674 DEBUG2("comparing key=%p with comm=%p", (void *) key, (void *) action);
675 if ((void *) action == (void *) key) {
676 DEBUG2("key %p found, return value latency limited value %d",
677 (void *) key, (int) data);
678 xbt_dict_cursor_free(&cursor);
683 return surf_workstation_model->get_latency_limited(action->comm.surf_comm);
687 /******************************************************************************/
688 /* SIMIX_comm_copy_data callbacks */
689 /******************************************************************************/
690 static void (*SIMIX_comm_copy_data_callback) (smx_action_t, size_t) =
691 &SIMIX_comm_copy_pointer_callback;
694 SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, size_t))
696 SIMIX_comm_copy_data_callback = callback;
699 void SIMIX_comm_copy_pointer_callback(smx_action_t comm, size_t buff_size)
701 xbt_assert1((buff_size == sizeof(void *)),
702 "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
703 *(void **) (comm->comm.dst_buff) = comm->comm.src_buff;
706 void SIMIX_comm_copy_buffer_callback(smx_action_t comm, size_t buff_size)
708 memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size);
712 * \brief Copy the communication data from the sender's buffer to the receiver's one
713 * \param comm The communication
715 void SIMIX_comm_copy_data(smx_action_t comm)
717 size_t buff_size = comm->comm.src_buff_size;
718 /* If there is no data to be copy then return */
719 if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied == 1)
722 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
724 comm->comm.src_proc->smx_host->name, comm->comm.src_buff,
725 comm->comm.dst_proc->smx_host->name, comm->comm.dst_buff, buff_size);
727 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
728 if (comm->comm.dst_buff_size)
729 buff_size = MIN(buff_size, *(comm->comm.dst_buff_size));
731 /* Update the receiver's buffer size to the copied amount */
732 if (comm->comm.dst_buff_size)
733 *comm->comm.dst_buff_size = buff_size;
738 (*SIMIX_comm_copy_data_callback) (comm, buff_size);
740 /* Set the copied flag so we copy data only once */
741 /* (this function might be called from both communication ends) */
742 comm->comm.copied = 1;