X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/6ee7e9c2e455536ab817ae0136acfbb53822eecd..8ba5ebec745c34478bd1082a07edcb76aab3aefb:/src/simix/smx_network.c diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c index e6a26c6a41..0ee16e8772 100644 --- a/src/simix/smx_network.c +++ b/src/simix/smx_network.c @@ -15,20 +15,17 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix, static xbt_dict_t rdv_points = NULL; unsigned long int smx_total_comms = 0; -static XBT_INLINE void SIMIX_comm_start(smx_action_t action); -static void SIMIX_comm_finish(smx_action_t action); static void SIMIX_waitany_req_remove_from_actions(smx_req_t req); static void SIMIX_comm_copy_data(smx_action_t comm); static smx_action_t SIMIX_comm_new(e_smx_comm_type_t type); static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm); -static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm); static smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type, int (*match_fun)(void *, void *), void *); static void SIMIX_rdv_free(void *data); void SIMIX_network_init(void) { - rdv_points = xbt_dict_new(); + rdv_points = xbt_dict_new_homogeneous(SIMIX_rdv_free); } void SIMIX_network_exit(void) @@ -51,7 +48,7 @@ smx_rdv_t SIMIX_rdv_create(const char *name) rdv->comm_fifo = xbt_fifo_new(); if (rdv->name) - xbt_dict_set(rdv_points, rdv->name, rdv, SIMIX_rdv_free); + xbt_dict_set(rdv_points, rdv->name, rdv, NULL); } return rdv; } @@ -65,8 +62,7 @@ void SIMIX_rdv_destroy(smx_rdv_t rdv) void SIMIX_rdv_free(void *data) { smx_rdv_t rdv = (smx_rdv_t) data; - if (rdv->name) - xbt_free(rdv->name); + xbt_free(rdv->name); xbt_fifo_free(rdv->comm_fifo); xbt_free(rdv); } @@ -111,12 +107,19 @@ static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm) * \param rdv The rendez-vous point * \param comm The communication request */ -static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm) +XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm) { xbt_fifo_remove(rdv->comm_fifo, comm); comm->comm.rdv = NULL; } +/** + * \brief Wrapper to SIMIX_rdv_get_request + */ +smx_action_t SIMIX_comm_get_send_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), void* data) { + return SIMIX_rdv_get_request(rdv, SIMIX_COMM_SEND, match_fun, data); +} + /** * \brief Checks if there is a communication action queued in a rendez-vous matching our needs * \param type The type of communication we are looking for (comm_send, comm_recv) @@ -125,11 +128,12 @@ static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm) smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type, int (*match_fun)(void *, void *), void *data) { + // FIXME rewrite this function by using SIMIX_rdv_has_send/recv_match smx_action_t action; xbt_fifo_item_t item; void* req_data = NULL; - xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){ + xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t) { if (action->comm.type == SIMIX_COMM_SEND) { req_data = action->comm.src_data; } else if (action->comm.type == SIMIX_COMM_RECEIVE) { @@ -151,6 +155,48 @@ smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type, return NULL; } +/** + * \brief Checks if there is a send communication action + * queued in a rendez-vous matching our needs. + * \return 1 if found, 0 otherwise + */ +int SIMIX_comm_has_send_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), void* data) { + + smx_action_t action; + xbt_fifo_item_t item; + + xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){ + if (action->comm.type == SIMIX_COMM_SEND + && (!match_fun || match_fun(data, action->comm.src_data))) { + XBT_DEBUG("Found a matching communication action %p", action); + return 1; + } + } + XBT_DEBUG("No matching communication action found"); + return 0; +} + +/** + * \brief Checks if there is a recv communication action + * queued in a rendez-vous matching our needs. + * \return 1 if found, 0 otherwise + */ +int SIMIX_comm_has_recv_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), void* data) { + + smx_action_t action; + xbt_fifo_item_t item; + + xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t) { + if (action->comm.type == SIMIX_COMM_RECEIVE + && (!match_fun || match_fun(data, action->comm.dst_data))) { + XBT_DEBUG("Found a matching communication action %p", action); + return 1; + } + } + XBT_DEBUG("No matching communication action found"); + return 0; +} + /******************************************************************************/ /* Comunication Actions */ /******************************************************************************/ @@ -166,6 +212,7 @@ smx_action_t SIMIX_comm_new(e_smx_comm_type_t type) /* alloc structures */ act = xbt_mallocator_get(simix_global->action_mallocator); + act->type = SIMIX_ACTION_COMMUNICATE; act->state = SIMIX_WAITING; @@ -194,11 +241,14 @@ smx_action_t SIMIX_comm_new(e_smx_comm_type_t type) */ void SIMIX_comm_destroy(smx_action_t action) { - XBT_DEBUG("Destroy action %p (refcount:%d)", action, action->comm.refcount); - - if (action->comm.refcount <= 0) - xbt_die(bprintf("the refcount of comm %p is already 0 before decreasing it. That's a bug!",action)); + XBT_DEBUG("Destroy action %p (refcount: %d), state: %d", + action, action->comm.refcount, action->state); + if (action->comm.refcount <= 0) { + xbt_backtrace_display_current(); + xbt_die("the refcount of comm %p is already 0 before decreasing it. " + "That's a bug!", action); + } action->comm.refcount--; if (action->comm.refcount > 0) return; @@ -209,17 +259,14 @@ void SIMIX_comm_destroy(smx_action_t action) action->latency_limited = SIMIX_comm_is_latency_bounded( action ) ; #endif -#ifdef HAVE_TRACING - TRACE_smx_action_destroy(action); -#endif - xbt_free(action->name); SIMIX_comm_destroy_internal_actions(action); if (action->comm.detached && action->state != SIMIX_DONE) { /* the communication has failed and was detached: * we have to free the buffer */ - ((void_f_pvoid_t) action->comm.src_data)(action->comm.src_buff); + action->comm.clean_fun(action->comm.src_buff); + action->comm.src_buff = NULL; } xbt_mallocator_release(simix_global->action_mallocator, action); @@ -249,7 +296,9 @@ void SIMIX_comm_destroy_internal_actions(smx_action_t action) smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv, double task_size, double rate, void *src_buff, size_t src_buff_size, - int (*match_fun)(void *, void *), void *data, + int (*match_fun)(void *, void *), + void (*clean_fun)(void *), // used to free the action in case of problem after a detached send + void *data, int detached) { smx_action_t action; @@ -265,12 +314,16 @@ smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv, action->state = SIMIX_READY; action->comm.type = SIMIX_COMM_READY; } + xbt_fifo_push(src_proc->comms, action); - /* If the communication action is detached then decrease the refcount - * by one, so it will be eliminated by the receivers destroy call */ + /* if the communication action is detached then decrease the refcount + * by one, so it will be eliminated by the receiver's destroy call */ if (detached) { action->comm.detached = 1; action->comm.refcount--; + action->comm.clean_fun = clean_fun; + } else { + action->comm.clean_fun = NULL; } /* Setup the communication request */ @@ -287,7 +340,7 @@ smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv, } SIMIX_comm_start(action); - return action; + return (detached ? NULL : action); } smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv, @@ -308,6 +361,7 @@ smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv, action->state = SIMIX_READY; action->comm.type = SIMIX_COMM_READY; } + xbt_fifo_push(dst_proc->comms, action); /* Setup communication request */ action->comm.dst_proc = dst_proc; @@ -326,6 +380,7 @@ smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv, void SIMIX_pre_comm_wait(smx_req_t req, smx_action_t action, double timeout, int idx) { + /* the request may be a wait, a send or a recv */ surf_action_t sleep; @@ -439,8 +494,10 @@ void SIMIX_pre_comm_waitany(smx_req_t req, int idx) } xbt_dynar_foreach(actions, cursor, action){ - /* Associate this request to the action */ + /* associate this request to the the action */ xbt_fifo_push(action->request_list, req); + + /* see if the action is already finished */ if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){ SIMIX_comm_finish(action); break; @@ -463,10 +520,12 @@ void SIMIX_waitany_req_remove_from_actions(smx_req_t req) * \brief Start the simulation of a communication request * \param action The communication action */ -static XBT_INLINE void SIMIX_comm_start(smx_action_t action) + +XBT_INLINE void SIMIX_comm_start(smx_action_t action) { /* If both the sender and the receiver are already there, start the communication */ if (action->state == SIMIX_READY) { + smx_host_t sender = action->comm.src_proc->smx_host; smx_host_t receiver = action->comm.dst_proc->smx_host; @@ -480,10 +539,6 @@ static XBT_INLINE void SIMIX_comm_start(smx_action_t action) action->state = SIMIX_RUNNING; -#ifdef HAVE_TRACING - TRACE_smx_action_communicate(action, action->comm.src_proc); -#endif - /* If a link is failed, detect it immediately */ if (surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) { XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", @@ -502,9 +557,13 @@ static XBT_INLINE void SIMIX_comm_start(smx_action_t action) } } +/** + * \brief Answers the SIMIX requests associated to a communication action. + * \param action a finished communication action + */ void SIMIX_comm_finish(smx_action_t action) { - unsigned int destroy_count = 0; + volatile unsigned int destroy_count = 0; smx_req_t req; while ((req = xbt_fifo_shift(action->request_list))) { @@ -535,7 +594,7 @@ void SIMIX_comm_finish(smx_action_t action) case SIMIX_SRC_TIMEOUT: TRY { - THROW0(timeout_error, 0, "Communication timeouted because of sender"); + THROWF(timeout_error, 0, "Communication timeouted because of sender"); } CATCH(req->issuer->running_ctx->exception) { req->issuer->doexception = 1; @@ -544,7 +603,7 @@ void SIMIX_comm_finish(smx_action_t action) case SIMIX_DST_TIMEOUT: TRY { - THROW0(timeout_error, 0, "Communication timeouted because of receiver"); + THROWF(timeout_error, 0, "Communication timeouted because of receiver"); } CATCH(req->issuer->running_ctx->exception) { req->issuer->doexception = 1; @@ -554,9 +613,9 @@ void SIMIX_comm_finish(smx_action_t action) case SIMIX_SRC_HOST_FAILURE: TRY { if (req->issuer == action->comm.src_proc) - THROW0(host_error, 0, "Host failed"); + THROWF(host_error, 0, "Host failed"); else - THROW0(network_error, 0, "Remote peer failed"); + THROWF(network_error, 0, "Remote peer failed"); } CATCH(req->issuer->running_ctx->exception) { req->issuer->doexception = 1; @@ -566,9 +625,9 @@ void SIMIX_comm_finish(smx_action_t action) case SIMIX_DST_HOST_FAILURE: TRY { if (req->issuer == action->comm.dst_proc) - THROW0(host_error, 0, "Host failed"); + THROWF(host_error, 0, "Host failed"); else - THROW0(network_error, 0, "Remote peer failed"); + THROWF(network_error, 0, "Remote peer failed"); } CATCH(req->issuer->running_ctx->exception) { req->issuer->doexception = 1; @@ -577,18 +636,41 @@ void SIMIX_comm_finish(smx_action_t action) case SIMIX_LINK_FAILURE: TRY { - XBT_DEBUG("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)", - action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name, - req->issuer->name, req->issuer); - THROW0(network_error, 0, "Link failure"); + XBT_DEBUG("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p) detached:%d", + action, + action->comm.src_proc ? action->comm.src_proc->smx_host->name : NULL, + action->comm.dst_proc ? action->comm.dst_proc->smx_host->name : NULL, + req->issuer->name, req->issuer, action->comm.detached); + if (action->comm.src_proc == req->issuer) { + XBT_DEBUG("I'm source"); + } else if (action->comm.dst_proc == req->issuer) { + XBT_DEBUG("I'm dest"); + } else { + XBT_DEBUG("I'm neither source nor dest"); + } + THROWF(network_error, 0, "Link failure"); } CATCH(req->issuer->running_ctx->exception) { req->issuer->doexception = 1; } break; + case SIMIX_CANCELED: + TRY { + if (req->issuer == action->comm.dst_proc) { + THROWF(cancel_error, 0, "Communication canceled by the sender"); + } + else { + THROWF(cancel_error, 0, "Communication canceled by the receiver"); + } + } + CATCH(req->issuer->running_ctx->exception) { + req->issuer->doexception = 1; + } + break; + default: - THROW_IMPOSSIBLE; + xbt_die("Unexpected action state in SIMIX_comm_finish: %d", action->state); } /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */ @@ -602,6 +684,7 @@ void SIMIX_comm_finish(smx_action_t action) } req->issuer->waiting_action = NULL; + xbt_fifo_remove(req->issuer->comms, action); SIMIX_request_answer(req); destroy_count++; } @@ -610,6 +693,10 @@ void SIMIX_comm_finish(smx_action_t action) SIMIX_comm_destroy(action); } +/** + * \brief This function is called when a Surf communication action is finished. + * \param action the corresponding Simix communication + */ void SIMIX_post_comm(smx_action_t action) { /* Update action state */ @@ -626,33 +713,45 @@ void SIMIX_post_comm(smx_action_t action) surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED) action->state = SIMIX_DST_HOST_FAILURE; else if (action->comm.surf_comm && - surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) + surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) { + XBT_DEBUG("Puta madre. Surf says that the link broke"); action->state = SIMIX_LINK_FAILURE; - else + } else action->state = SIMIX_DONE; - XBT_DEBUG("SIMIX_post_comm: action state = %d", action->state); + XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d", + action, action->state, action->comm.src_proc, action->comm.dst_proc, action->comm.detached); - /* After this point the surf actions associated with the simix communicate - action are no longer needed, thus we delete them. */ + /* destroy the surf actions associated with the Simix communication */ SIMIX_comm_destroy_internal_actions(action); - /* If there are requests associated with the action, then answer them */ - if (xbt_fifo_size(action->request_list)) + /* remove the communication action from the list of pending communications + * of both processes (if they still exist) */ + if (action->comm.src_proc) { + xbt_fifo_remove(action->comm.src_proc->comms, action); + } + if (action->comm.dst_proc) { + xbt_fifo_remove(action->comm.dst_proc->comms, action); + } + + /* if there are requests associated with the action, then answer them */ + if (xbt_fifo_size(action->request_list)) { SIMIX_comm_finish(action); + } } void SIMIX_comm_cancel(smx_action_t action) { - /* If the action is a waiting state means that it is still in a rdv */ + /* if the action is a waiting state means that it is still in a rdv */ /* so remove from it and delete it */ if (action->state == SIMIX_WAITING) { SIMIX_rdv_remove(action->comm.rdv, action); - action->state = SIMIX_FAILED; - } else { - /* When running the MC there are no surf actions */ - if(!MC_IS_ENABLED) - surf_workstation_model->action_cancel(action->comm.surf_comm); + action->state = SIMIX_CANCELED; + } + else if (!MC_IS_ENABLED /* when running the MC there are no surf actions */ + && (action->state == SIMIX_READY || action->state == SIMIX_RUNNING)) { + + surf_workstation_model->action_cancel(action->comm.surf_comm); } } @@ -679,6 +778,10 @@ double SIMIX_comm_get_remains(smx_action_t action) { double remains; + if(!action){ + return 0; + } + switch (action->state) { case SIMIX_RUNNING: @@ -739,6 +842,9 @@ smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action) */ XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action) { + if(!action){ + return 0; + } if (action->comm.surf_comm){ XBT_DEBUG("Getting latency limited for surf_action (%p)", action->comm.surf_comm); action->latency_limited = surf_workstation_model->get_latency_limited(action->comm.surf_comm); @@ -751,25 +857,36 @@ XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action) /******************************************************************************/ /* SIMIX_comm_copy_data callbacks */ /******************************************************************************/ -static void (*SIMIX_comm_copy_data_callback) (smx_action_t, size_t) = +static void (*SIMIX_comm_copy_data_callback) (smx_action_t, void*, size_t) = &SIMIX_comm_copy_pointer_callback; void -SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, size_t)) +SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, void*, size_t)) { SIMIX_comm_copy_data_callback = callback; } -void SIMIX_comm_copy_pointer_callback(smx_action_t comm, size_t buff_size) +void SIMIX_comm_copy_pointer_callback(smx_action_t comm, void* buff, size_t buff_size) { - xbt_assert1((buff_size == sizeof(void *)), + xbt_assert((buff_size == sizeof(void *)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size); - *(void **) (comm->comm.dst_buff) = comm->comm.src_buff; + *(void **) (comm->comm.dst_buff) = buff; } -void SIMIX_comm_copy_buffer_callback(smx_action_t comm, size_t buff_size) +void SIMIX_comm_copy_buffer_callback(smx_action_t comm, void* buff, size_t buff_size) { - memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size); + XBT_DEBUG("Copy the data over"); + memcpy(comm->comm.dst_buff, buff, buff_size); +} + +void smpi_comm_copy_data_callback(smx_action_t comm, void* buff, size_t buff_size) +{ + XBT_DEBUG("Copy the data over"); + memcpy(comm->comm.dst_buff, buff, buff_size); + if (comm->comm.detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the original buffer available to the application ASAP + xbt_free(buff); + comm->comm.src_buff = NULL; + } } /** @@ -785,8 +902,10 @@ void SIMIX_comm_copy_data(smx_action_t comm) XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", comm, - comm->comm.src_proc->smx_host->name, comm->comm.src_buff, - comm->comm.dst_proc->smx_host->name, comm->comm.dst_buff, buff_size); + comm->comm.src_proc ? comm->comm.src_proc->smx_host->name : "a finished process", + comm->comm.src_buff, + comm->comm.dst_proc ? comm->comm.dst_proc->smx_host->name : "a finished process", + comm->comm.dst_buff, buff_size); /* Copy at most dst_buff_size bytes of the message to receiver's buffer */ if (comm->comm.dst_buff_size) @@ -796,10 +915,8 @@ void SIMIX_comm_copy_data(smx_action_t comm) if (comm->comm.dst_buff_size) *comm->comm.dst_buff_size = buff_size; - if (buff_size == 0) - return; - - (*SIMIX_comm_copy_data_callback) (comm, buff_size); + if (buff_size > 0) + SIMIX_comm_copy_data_callback (comm, comm->comm.src_buff, buff_size); /* Set the copied flag so we copy data only once */ /* (this function might be called from both communication ends) */