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 static void SIMIX_rdv_free(void *data);
29 void SIMIX_network_init(void)
31 rdv_points = xbt_dict_new();
34 void SIMIX_network_exit(void)
36 xbt_dict_free(&rdv_points);
39 /******************************************************************************/
40 /* Rendez-Vous Points */
41 /******************************************************************************/
43 smx_rdv_t SIMIX_rdv_create(const char *name)
45 /* two processes may have pushed the same rdv_create request at the same time */
46 smx_rdv_t rdv = name ? xbt_dict_get_or_null(rdv_points, name) : NULL;
49 rdv = xbt_new0(s_smx_rvpoint_t, 1);
50 rdv->name = name ? xbt_strdup(name) : NULL;
51 rdv->comm_fifo = xbt_fifo_new();
54 xbt_dict_set(rdv_points, name, rdv, SIMIX_rdv_free);
59 void SIMIX_rdv_destroy(smx_rdv_t rdv)
62 xbt_dict_remove(rdv_points, rdv->name);
65 void SIMIX_rdv_free(void *data)
67 smx_rdv_t rdv = (smx_rdv_t) data;
70 xbt_fifo_free(rdv->comm_fifo);
74 smx_rdv_t SIMIX_rdv_get_by_name(const char *name)
76 return xbt_dict_get_or_null(rdv_points, name);
79 int SIMIX_rdv_comm_count_by_host(smx_rdv_t rdv, smx_host_t host)
81 smx_action_t comm = NULL;
82 xbt_fifo_item_t item = NULL;
85 xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_action_t) {
86 if (comm->comm.src_proc->smx_host == host)
93 smx_action_t SIMIX_rdv_get_head(smx_rdv_t rdv)
95 return xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
99 * \brief Push a communication request into a rendez-vous point
100 * \param rdv The rendez-vous point
101 * \param comm The communication request
103 static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm)
105 xbt_fifo_push(rdv->comm_fifo, comm);
106 comm->comm.rdv = rdv;
110 * \brief Remove a communication request from a rendez-vous point
111 * \param rdv The rendez-vous point
112 * \param comm The communication request
114 static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm)
116 xbt_fifo_remove(rdv->comm_fifo, comm);
117 comm->comm.rdv = NULL;
121 * \brief Checks if there is a communication request queued in a rendez-vous matching our needs
122 * \param type The type of communication we are looking for (comm_send, comm_recv)
123 * \return The communication request if found, NULL otherwise
125 smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type)
127 smx_action_t comm = (smx_action_t)
128 xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
130 if (comm && comm->comm.type == type) {
131 DEBUG0("Communication request found!");
132 xbt_fifo_shift(rdv->comm_fifo);
133 comm->comm.refcount++;
134 comm->comm.rdv = NULL;
138 DEBUG0("Communication request not found");
142 /******************************************************************************/
143 /* Comunication Actions */
144 /******************************************************************************/
147 * \brief Creates a new comunicate action
148 * \param type The type of request (comm_send, comm_recv)
149 * \return The new comunicate action
151 smx_action_t SIMIX_comm_new(e_smx_comm_type_t type)
155 /* alloc structures */
156 act = xbt_new0(s_smx_action_t, 1);
157 act->type = SIMIX_ACTION_COMMUNICATE;
158 act->state = SIMIX_WAITING;
159 act->request_list = xbt_fifo_new();
161 /* set communication */
162 act->comm.type = type;
163 act->comm.refcount = 1;
166 act->category = NULL;
169 DEBUG1("Create communicate action %p", act);
175 * \brief Destroy a communicate action
176 * \param action The communicate action to be destroyed
178 void SIMIX_comm_destroy(smx_action_t action)
180 DEBUG1("Destroy action %p", action);
182 if (action->comm.refcount <= 0)
183 xbt_die(bprintf("the refcount of comm %p is already 0 before decreasing it. That's a bug!",action));
185 #ifdef HAVE_LATENCY_BOUND_TRACKING
186 //save is latency limited flag to use afterwards
187 if (action->comm.surf_comm) {
188 DEBUG2("adding key %p with latency limited value %d to the dict", action,
189 SIMIX_comm_is_latency_bounded(action));
190 xbt_dicti_set(simix_global->latency_limited_dict, (uintptr_t) action,
191 SIMIX_comm_is_latency_bounded(action));
195 action->comm.refcount--;
196 if (action->comm.refcount > 0)
198 VERB2("Really free communication %p; refcount is now %d", action,
199 action->comm.refcount);
202 TRACE_smx_action_destroy(action);
206 xbt_free(action->name);
208 xbt_fifo_free(action->request_list);
210 SIMIX_comm_destroy_internal_actions(action);
215 void SIMIX_comm_destroy_internal_actions(smx_action_t action)
217 if (action->comm.surf_comm){
218 action->comm.surf_comm->model_type->action_unref(action->comm.surf_comm);
219 action->comm.surf_comm = NULL;
222 if (action->comm.src_timeout){
223 action->comm.src_timeout->model_type->action_unref(action->comm.src_timeout);
224 action->comm.src_timeout = NULL;
227 if (action->comm.dst_timeout){
228 action->comm.dst_timeout->model_type->action_unref(action->comm.dst_timeout);
229 action->comm.dst_timeout = NULL;
233 smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv,
234 double task_size, double rate,
235 void *src_buff, size_t src_buff_size, void *data)
239 /* Look for communication request matching our needs.
240 If it is not found then create it and push it into the rendez-vous point */
241 action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_RECEIVE);
244 action = SIMIX_comm_new(SIMIX_COMM_SEND);
245 SIMIX_rdv_push(rdv, action);
247 action->state = SIMIX_READY;
248 action->comm.type = SIMIX_COMM_READY;
251 /* Setup the communication request */
252 action->comm.src_proc = src_proc;
253 action->comm.task_size = task_size;
254 action->comm.rate = rate;
255 action->comm.src_buff = src_buff;
256 action->comm.src_buff_size = src_buff_size;
257 action->comm.data = data;
260 action->state = SIMIX_RUNNING;
264 SIMIX_comm_start(action);
268 smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv,
269 void *dst_buff, size_t *dst_buff_size)
273 /* Look for communication request matching our needs.
274 * If it is not found then create it and push it into the rendez-vous point
276 action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_SEND);
279 action = SIMIX_comm_new(SIMIX_COMM_RECEIVE);
280 SIMIX_rdv_push(rdv, action);
282 action->state = SIMIX_READY;
283 action->comm.type = SIMIX_COMM_READY;
286 /* Setup communication request */
287 action->comm.dst_proc = dst_proc;
288 action->comm.dst_buff = dst_buff;
289 action->comm.dst_buff_size = dst_buff_size;
292 action->state = SIMIX_RUNNING;
296 SIMIX_comm_start(action);
300 void SIMIX_pre_comm_wait(smx_req_t req)
302 smx_action_t action = req->comm_wait.comm;
303 double timeout = req->comm_wait.timeout;
306 /* Associate this request to the action */
307 xbt_fifo_push(action->request_list, req);
308 req->issuer->waiting_action = action;
311 action->state = SIMIX_DONE;
312 SIMIX_comm_finish(action);
315 /* If the action has already finish perform the error handling, */
316 /* otherwise set up a waiting timeout on the right side */
317 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
318 SIMIX_comm_finish(action);
319 } 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 */
320 sleep = surf_workstation_model->extension.workstation.sleep(req->issuer->smx_host->host, timeout);
321 surf_workstation_model->action_data_set(sleep, action);
323 if (req->issuer == action->comm.src_proc)
324 action->comm.src_timeout = sleep;
326 action->comm.dst_timeout = sleep;
330 void SIMIX_pre_comm_test(smx_req_t req)
332 smx_action_t action = req->comm_test.comm;
333 req->comm_test.result = (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING);
335 if (req->comm_test.result) {
336 xbt_fifo_push(action->request_list, req);
337 SIMIX_comm_finish(action);
340 SIMIX_request_answer(req);
344 void SIMIX_pre_comm_waitany(smx_req_t req)
347 unsigned int cursor = 0;
348 xbt_dynar_t actions = req->comm_waitany.comms;
349 xbt_dynar_foreach(actions, cursor, action){
350 /* Associate this request to the action */
351 xbt_fifo_push(action->request_list, req);
352 if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){
353 SIMIX_comm_finish(action);
359 void SIMIX_waitany_req_remove_from_actions(smx_req_t req)
362 unsigned int cursor = 0;
363 xbt_dynar_t actions = req->comm_waitany.comms;
365 xbt_dynar_foreach(actions, cursor, action){
366 xbt_fifo_remove(action->request_list, req);
371 * \brief Start the simulation of a communication request
372 * \param action The communication action
374 static XBT_INLINE void SIMIX_comm_start(smx_action_t action)
376 /* If both the sender and the receiver are already there, start the communication */
377 if (action->state == SIMIX_READY) {
378 smx_host_t sender = action->comm.src_proc->smx_host;
379 smx_host_t receiver = action->comm.dst_proc->smx_host;
381 DEBUG3("Starting communication %p from '%s' to '%s'", action,
382 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
384 action->comm.surf_comm = surf_workstation_model->extension.workstation.
385 communicate(sender->host, receiver->host, action->comm.task_size, action->comm.rate);
387 surf_workstation_model->action_data_set(action->comm.surf_comm, action);
389 action->state = SIMIX_RUNNING;
392 TRACE_smx_action_communicate(action, action->comm.src_proc);
393 TRACE_surf_action(action->comm.surf_comm, action->category);
396 /* If a link is failed, detect it immediately */
397 if (surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) {
398 DEBUG2("Communication from '%s' to '%s' failed to start because of a link failure",
399 SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
400 action->state = SIMIX_LINK_FAILURE;
401 SIMIX_comm_destroy_internal_actions(action);
404 /* If any of the process is suspend, create the action but stop its execution,
405 it will be restarted when the sender process resume */
406 if (SIMIX_process_is_suspended(action->comm.src_proc) ||
407 SIMIX_process_is_suspended(action->comm.dst_proc)) {
408 /* FIXME: check what should happen with the action state */
409 surf_workstation_model->suspend(action->comm.surf_comm);
414 void SIMIX_comm_finish(smx_action_t action)
418 while ((req = xbt_fifo_shift(action->request_list))) {
420 /* If a waitany request is waiting for this action to finish, then remove
421 it from the other actions in the waitany list. Afterwards, get the
422 position of the actual action in the waitany request's actions dynar and
423 return it as the result of the call */
424 if (req->call == REQ_COMM_WAITANY) {
425 SIMIX_waitany_req_remove_from_actions(req);
426 req->comm_waitany.result = xbt_dynar_search(req->comm_waitany.comms, &action);
429 /* If the action is still in a rendez-vous point then remove from it */
430 if (action->comm.rdv)
431 SIMIX_rdv_remove(action->comm.rdv, action);
433 DEBUG1("SIMIX_comm_finish: action state = %d", action->state);
435 /* Check out for errors */
436 switch (action->state) {
439 DEBUG1("Communication %p complete!", action);
440 SIMIX_comm_copy_data(action);
443 case SIMIX_SRC_TIMEOUT:
445 THROW0(timeout_error, 0, "Communication timeouted because of sender");
447 CATCH(req->issuer->running_ctx->exception) {
448 req->issuer->doexception = 1;
452 case SIMIX_DST_TIMEOUT:
454 THROW0(timeout_error, 0, "Communication timeouted because of receiver");
456 CATCH(req->issuer->running_ctx->exception) {
457 req->issuer->doexception = 1;
461 case SIMIX_SRC_HOST_FAILURE:
463 if (req->issuer == action->comm.src_proc)
464 THROW0(host_error, 0, "Host failed");
466 THROW0(network_error, 0, "Remote peer failed");
468 CATCH(req->issuer->running_ctx->exception) {
469 req->issuer->doexception = 1;
473 case SIMIX_DST_HOST_FAILURE:
475 if (req->issuer == action->comm.dst_proc)
476 THROW0(host_error, 0, "Host failed");
478 THROW0(network_error, 0, "Remote peer failed");
480 CATCH(req->issuer->running_ctx->exception) {
481 req->issuer->doexception = 1;
485 case SIMIX_LINK_FAILURE:
487 DEBUG5("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)",
488 action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name,
489 req->issuer->name, req->issuer);
490 THROW0(network_error, 0, "Link failure");
492 CATCH(req->issuer->running_ctx->exception) {
493 req->issuer->doexception = 1;
500 req->issuer->waiting_action = NULL;
501 SIMIX_request_answer(req);
505 void SIMIX_post_comm(smx_action_t action)
507 /* Update action state */
508 if (action->comm.src_timeout &&
509 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_DONE)
510 action->state = SIMIX_SRC_TIMEOUT;
511 else if (action->comm.dst_timeout &&
512 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_DONE)
513 action->state = SIMIX_DST_TIMEOUT;
514 else if (action->comm.src_timeout &&
515 surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_FAILED)
516 action->state = SIMIX_SRC_HOST_FAILURE;
517 else if (action->comm.dst_timeout &&
518 surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED)
519 action->state = SIMIX_DST_HOST_FAILURE;
520 else if (action->comm.surf_comm &&
521 surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED)
522 action->state = SIMIX_LINK_FAILURE;
524 action->state = SIMIX_DONE;
526 DEBUG1("SIMIX_post_comm: action state = %d", action->state);
528 /* After this point the surf actions associated with the simix communicate
529 action are no longer needed, thus we delete them. */
530 SIMIX_comm_destroy_internal_actions(action);
532 /* If there are requests associated with the action, then answer them */
533 if (xbt_fifo_size(action->request_list))
534 SIMIX_comm_finish(action);
537 void SIMIX_comm_cancel(smx_action_t action)
539 /* If the action is a waiting state means that it is still in a rdv */
540 /* so remove from it and delete it */
541 if (action->state == SIMIX_WAITING) {
542 SIMIX_rdv_remove(action->comm.rdv, action);
543 action->state = SIMIX_FAILED;
545 surf_workstation_model->action_cancel(action->comm.surf_comm);
549 void SIMIX_comm_suspend(smx_action_t action)
551 /*FIXME: shall we suspend also the timeout actions? */
552 surf_workstation_model->suspend(action->comm.surf_comm);
555 void SIMIX_comm_resume(smx_action_t action)
557 /*FIXME: check what happen with the timeouts */
558 surf_workstation_model->resume(action->comm.surf_comm);
562 /************* Action Getters **************/
565 * \brief get the amount remaining from the communication
566 * \param action The communication
568 double SIMIX_comm_get_remains(smx_action_t action)
572 switch (action->state) {
575 remains = surf_workstation_model->get_remains(action->comm.surf_comm);
580 remains = 0; /*FIXME: check what should be returned */
584 remains = 0; /*FIXME: is this correct? */
590 e_smx_state_t SIMIX_comm_get_state(smx_action_t action)
592 return action->state;
596 * \brief Return the user data associated to the communication
597 * \param action The communication
598 * \return the user data
600 void* SIMIX_comm_get_data(smx_action_t action)
602 return action->comm.data;
605 void* SIMIX_comm_get_src_buff(smx_action_t action)
607 return action->comm.src_buff;
610 void* SIMIX_comm_get_dst_buff(smx_action_t action)
612 return action->comm.dst_buff;
615 size_t SIMIX_comm_get_src_buff_size(smx_action_t action)
617 return action->comm.src_buff_size;
620 size_t SIMIX_comm_get_dst_buff_size(smx_action_t action)
624 if (action->comm.dst_buff_size)
625 buff_size = *(action->comm.dst_buff_size);
632 smx_process_t SIMIX_comm_get_src_proc(smx_action_t action)
634 return action->comm.src_proc;
637 smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action)
639 return action->comm.dst_proc;
642 #ifdef HAVE_LATENCY_BOUND_TRACKING
644 * \brief verify if communication is latency bounded
645 * \param comm The communication
647 XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action)
649 //try to find comm on the list of finished flows
652 xbt_dict_cursor_t cursor;
653 xbt_dict_foreach(simix_global->latency_limited_dict, cursor, key, data) {
654 DEBUG2("comparing key=%p with comm=%p", (void *) key, (void *) action);
655 if ((void *) action == (void *) key) {
656 DEBUG2("key %p found, return value latency limited value %d",
657 (void *) key, (int) data);
658 xbt_dict_cursor_free(&cursor);
663 return surf_workstation_model->get_latency_limited(action->comm.surf_comm);
667 /******************************************************************************/
668 /* SIMIX_comm_copy_data callbacks */
669 /******************************************************************************/
670 static void (*SIMIX_comm_copy_data_callback) (smx_action_t, size_t) =
671 &SIMIX_comm_copy_pointer_callback;
674 SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, size_t))
676 SIMIX_comm_copy_data_callback = callback;
679 void SIMIX_comm_copy_pointer_callback(smx_action_t comm, size_t buff_size)
681 xbt_assert1((buff_size == sizeof(void *)),
682 "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
683 *(void **) (comm->comm.dst_buff) = comm->comm.src_buff;
686 void SIMIX_comm_copy_buffer_callback(smx_action_t comm, size_t buff_size)
688 memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size);
692 * \brief Copy the communication data from the sender's buffer to the receiver's one
693 * \param comm The communication
695 void SIMIX_comm_copy_data(smx_action_t comm)
697 size_t buff_size = comm->comm.src_buff_size;
698 /* If there is no data to be copy then return */
699 if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied == 1)
702 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
704 comm->comm.src_proc->smx_host->name, comm->comm.src_buff,
705 comm->comm.dst_proc->smx_host->name, comm->comm.dst_buff, buff_size);
707 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
708 if (comm->comm.dst_buff_size)
709 buff_size = MIN(buff_size, *(comm->comm.dst_buff_size));
711 /* Update the receiver's buffer size to the copied amount */
712 if (comm->comm.dst_buff_size)
713 *comm->comm.dst_buff_size = buff_size;
718 (*SIMIX_comm_copy_data_callback) (comm, buff_size);
720 /* Set the copied flag so we copy data only once */
721 /* (this function might be called from both communication ends) */
722 comm->comm.copied = 1;