X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/15eb59567704ce6613cefcc628878db5fd8f9bbf..b30602ffd8e090c68e99a37ec5e4e7403debd7ce:/src/simix/smx_network.c

diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c
index 941dcce71b..dd7b49a4f3 100644
--- a/src/simix/smx_network.c
+++ b/src/simix/smx_network.c
@@ -1,13 +1,12 @@
-/* 	$Id$	 */
-
-/* Copyright (c) 2009 Cristian Rosa.
-   All rights reserved.                                          */
+/* Copyright (c) 2009, 2010. The SimGrid Team.
+ * All rights reserved.                                                     */
 
 /* This program is free software; you can redistribute it and/or modify it
  * under the terms of the license (GNU LGPL) which comes with this package. */
 
 #include "private.h"
 #include "xbt/log.h"
+#include "mc/mc.h"
 #include "xbt/dict.h"
 
 /* Pimple to get an histogram of message sizes in the simulation */
@@ -85,7 +84,7 @@ smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type) {
     DEBUG0("Communication request found!");
     xbt_fifo_shift(rdv->comm_fifo);
     SIMIX_communication_use(comm);
-    comm->rdv = NULL;    
+    comm->rdv = NULL;
     return comm;
   }
 
@@ -202,7 +201,7 @@ static inline void SIMIX_communication_use(smx_comm_t comm)
 
 /**
  *  \brief Start the simulation of a communication request
- *  \param comm The   comm->rdv = NULL;communication request
+ *  \param comm The communication request
  */
 static inline void SIMIX_communication_start(smx_comm_t comm)
 {
@@ -239,9 +238,9 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
  *  \param timeout The max amount of time to wait for the communication to finish
  *
  *  Throws:
- *   - host_error if peer failed
- *   - timeout_error if communication reached the timeout specified
- *   - network_error if network failed or peer issued a timeout
+ *   - host_error if local peer failed
+ *   - timeout_error if communication reached the timeout specified (either because of local peer or remote peer)
+ *   - network_error if network failed or remote peer failed
  */
 static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
 {
@@ -292,7 +291,7 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub
   SIMIX_sem_release_forever(comm->sem);
   
   /* Check for errors other than timeouts (they are catched above) */
-  if(!SIMIX_host_get_state(SIMIX_host_self())){
+  if (!SIMIX_host_get_state(SIMIX_host_self())){
     if(comm->rdv)
       SIMIX_rdv_remove(comm->rdv, comm);
     SIMIX_communication_destroy(comm);
@@ -300,7 +299,20 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub
   } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
     SIMIX_communication_destroy(comm);
     THROW0(network_error, 0, "Link failure");
+  } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc)) ||
+      !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc))) {
+    /* We test both src&dst because we dunno who we are today, and we already tested myself above.
+     *    So, at the end, we test the remote peer only
+     * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed.
+     *    In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem)
+     *    at the beginning of this function. */
+    SIMIX_communication_destroy(comm);
+    THROW0(network_error, 0, "Remote peer failed");
+
   }
+  /* Copy network data */
+  SIMIX_network_copy_data(comm);  
+
   SIMIX_communication_destroy(comm);
 }
 
@@ -349,16 +361,16 @@ void SIMIX_network_copy_data(smx_comm_t comm)
 {
   size_t buff_size = comm->src_buff_size;
 
+  /* If there is no data to be copy then return */
+  if(!comm->src_buff || !comm->dst_buff || comm->copied == 1)
+    return;
+
   DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
       comm,
       comm->src_proc->smx_host->name, comm->src_buff,
       comm->dst_proc->smx_host->name, comm->dst_buff,
       buff_size);
 
-  /* If there is no data to be copy then return */
-  if(!comm->src_buff || !comm->dst_buff)
-    return;
-  
   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
   if (comm->dst_buff_size)
     buff_size = MIN(buff_size,*(comm->dst_buff_size));
@@ -371,6 +383,10 @@ void SIMIX_network_copy_data(smx_comm_t comm)
     return;
   (*SIMIX_network_copy_data_callback)(comm, buff_size);
 
+  /* Set the copied flag so we copy data only once */
+  /* (this function might be called from both communication ends)*/
+  comm->copied = 1;
+  
   /* pimple to display the message sizes */
   {
     if (msg_sizes == NULL)
@@ -460,6 +476,10 @@ smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate,
 {
   smx_comm_t comm;
 
+  /*If running in model-checking mode then intercept the communication action*/
+  if (_surf_do_model_check)
+    MC_trans_intercept_isend(rdv);
+  
   /* Look for communication request matching our needs.
      If it is not found then create it and push it into the rendez-vous point */
   comm = SIMIX_rdv_get_request(rdv, comm_recv);
@@ -476,16 +496,22 @@ smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate,
   comm->src_buff = src_buff;
   comm->src_buff_size = src_buff_size;
   comm->data = data;
-
+  
   SIMIX_communication_start(comm);
   return comm;
 }
 
-smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_size) {
+smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_size)
+{
   smx_comm_t comm;
 
+  /*If running in model-checking mode then intercept the communication action*/
+  if (_surf_do_model_check)
+    MC_trans_intercept_irecv(rdv);
+  
   /* Look for communication request matching our needs.
-     If it is not found then create it and push it into the rendez-vous point */
+   * If it is not found then create it and push it into the rendez-vous point
+   */
   comm = SIMIX_rdv_get_request(rdv, comm_send);
 
   if(!comm){
@@ -497,58 +523,63 @@ smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_s
   comm->dst_proc = SIMIX_process_self();
   comm->dst_buff = dst_buff;
   comm->dst_buff_size = dst_buff_size;
-
+ 
   SIMIX_communication_start(comm);
   return comm;
 }
 
 /** @brief blocks until the communication terminates or the timeout occurs */
-XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout) {
+XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout)
+{
+  /*If running in model-checking mode then intercept the communication action*/
+  if (_surf_do_model_check)
+    MC_trans_intercept_wait(comm);
+  
   /* Wait for communication completion */
   SIMIX_communication_wait_for_completion(comm, timeout);
 }
 
 /** @Returns whether the (asynchronous) communication is done yet or not */
-XBT_INLINE int SIMIX_network_test(smx_comm_t comm) {
-  return comm->sem?SIMIX_sem_would_block(comm->sem):0;
+XBT_INLINE int SIMIX_network_test(smx_comm_t comm)
+{
+  /*If running in model-checking mode then intercept the communication action*/
+  if (_surf_do_model_check)
+    MC_trans_intercept_test(comm);
+
+  /* Copy data if the communication is done */
+  if(comm->sem && !SIMIX_sem_would_block(comm->sem)){
+    /* Copy network data */
+    SIMIX_network_copy_data(comm);
+    return TRUE;
+  }
+  return FALSE;
 }
 
 /** @brief wait for the completion of any communication of a set
  *
  *  @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
  */
-unsigned int SIMIX_network_waitany(xbt_dynar_t comms) {
+unsigned int SIMIX_network_waitany(xbt_dynar_t comms)
+{
   xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t),NULL);
   unsigned int cursor, found_comm=-1;
   smx_comm_t comm,comm_finished=NULL;
 
-  xbt_dynar_foreach(comms,cursor,comm){
+  /*If running in model-checking mode then intercept the communication action*/
+  if (_surf_do_model_check)
+    MC_trans_intercept_waitany(comms);
+    
+  xbt_dynar_foreach(comms,cursor,comm)
     xbt_dynar_push(sems,&(comm->sem));
-  }
 
   DEBUG1("Waiting for the completion of communication set %p", comms);
 
   found_comm = SIMIX_sem_acquire_any(sems);
+  xbt_dynar_free_container(&sems);
   xbt_assert0(found_comm!=-1,"Cannot find which communication finished");
   xbt_dynar_get_cpy(comms,found_comm,&comm_finished);
 
-  DEBUG1("Communication %p complete! Let's check for errors", comm_finished);
-
-  /* Make sure that everyone sleeping on that semaphore is awake,
-   * and that nobody will ever block on it */
-  SIMIX_sem_release_forever(comm_finished->sem);
-
-  /* Check for errors */
-  if(!SIMIX_host_get_state(SIMIX_host_self())){
-    if(comm_finished->rdv)
-      SIMIX_rdv_remove(comm_finished->rdv, comm_finished);
-    SIMIX_communication_destroy(comm_finished);
-    THROW0(host_error, 0, "Host failed");
-  } else if (SIMIX_action_get_state(comm_finished->act) == SURF_ACTION_FAILED){
-    SIMIX_communication_destroy(comm_finished);
-    THROW0(network_error, 0, "Link failure");
-  }
-  SIMIX_communication_destroy(comm_finished);
-
+  /* Check for errors and cleanup the comm */
+  SIMIX_communication_wait_for_completion(comm_finished,-1);
   return found_comm;
-}
+}
\ No newline at end of file