Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
argument why parallel execution don't jeopardize reproducibility even if we never...
[simgrid.git] / src / simix / smx_process.c
index 366cc77..aa1378a 100644 (file)
@@ -4,11 +4,10 @@
 /* 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 "smx_private.h"
 #include "xbt/sysdep.h"
 #include "xbt/log.h"
 #include "xbt/dict.h"
-#include "msg/mailbox.h"
 #include "mc/mc.h"
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
@@ -31,7 +30,16 @@ XBT_INLINE smx_process_t SIMIX_process_self(void)
 }
 
 /**
- * \brief Move a process to the list of processes to destroy.
+ * \brief Returns whether a process has pending asynchronous communications.
+ * \return true if there are asynchronous communications in this process
+ */
+int SIMIX_process_has_pending_comms(smx_process_t process) {
+
+  return xbt_fifo_size(process->comms) > 0;
+}
+
+/**
+ * \brief Moves a process to the list of processes to destroy.
  */
 void SIMIX_process_cleanup(smx_process_t process)
 {
@@ -42,38 +50,53 @@ void SIMIX_process_cleanup(smx_process_t process)
   smx_action_t action;
   while ((action = xbt_fifo_pop(process->comms))) {
 
-    /* make sure no one will finish the comm after this process is destroyed */
+    /* make sure no one will finish the comm after this process is destroyed,
+     * because src_proc or dst_proc would be an invalid pointer */
     SIMIX_comm_cancel(action);
 
     if (action->comm.src_proc == process) {
-      XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d",
-          action, action->comm.detached, action->state);
+      XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
+          action, action->comm.detached, (int)action->state, action->comm.src_proc, action->comm.dst_proc);
       action->comm.src_proc = NULL;
 
       if (action->comm.detached) {
-        /* the receiver was supposed to destroy the comm after completion,
-         * but the comm will actually never finish */
-        action->comm.refcount++;
+         if (action->comm.refcount == 0) {
+           XBT_DEBUG("Increase the refcount before destroying it since it's detached");
+           /* I'm not supposed to destroy a detached comm from the sender side,
+            * unless there is no receiver matching the rdv */
+           action->comm.refcount++;
+           SIMIX_comm_destroy(action);
+         }
+         else {
+           XBT_DEBUG("Don't destroy it since its refcount is %d", action->comm.refcount);
+         }
+      } else {
+        SIMIX_comm_destroy(action);
       }
     }
     else if (action->comm.dst_proc == process){
-      XBT_DEBUG("Found an unfinished recv comm %p, state %d", action, action->state);
+      XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
+          action, (int)action->state, action->comm.src_proc, action->comm.dst_proc);
       action->comm.dst_proc = NULL;
+
+      if (action->comm.detached && action->comm.refcount == 1
+          && action->comm.src_proc != NULL) {
+        /* the comm will be freed right now, remove it from the sender */
+        xbt_fifo_remove(action->comm.src_proc->comms, action);
+      }
+      SIMIX_comm_destroy(action);
     }
     else {
-      XBT_DEBUG("Strange, I'm not in comm %p, state = %d, src = %p, dst = %p", action,
-          action->state, action->comm.src_proc, action->comm.dst_proc);
-      THROW_IMPOSSIBLE;
+      xbt_die("Communication action %p is in my list but I'm not the sender "
+          "or the receiver", action);
     }
-
-    /* FIXME uncommenting this instruction crashes complex simulations
-    SIMIX_comm_destroy(action); */
   }
 
   /*xbt_swag_remove(process, simix_global->process_to_run);*/
   xbt_swag_remove(process, simix_global->process_list);
   xbt_swag_remove(process, process->smx_host->process_list);
   xbt_swag_insert(process, simix_global->process_to_destroy);
+  process->context->iwannadie = 0;
 }
 
 /** 
@@ -90,15 +113,12 @@ void SIMIX_process_empty_trash(void)
     SIMIX_context_free(process->context);
 
     /* Free the exception allocated at creation time */
-    if (process->running_ctx)
-      free(process->running_ctx);
-    if (process->properties)
-      xbt_dict_free(&process->properties);
+    free(process->running_ctx);
+    xbt_dict_free(&process->properties);
 
     xbt_fifo_free(process->comms);
 
     free(process->name);
-    process->name = NULL;
     free(process);
   }
 }
@@ -117,7 +137,7 @@ void SIMIX_create_maestro_process()
   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
-  maestro->request.issuer = maestro;
+  maestro->simcall.issuer = maestro;
 
   simix_global->maestro_process = maestro;
   return;
@@ -136,6 +156,7 @@ smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
       args->code,
       args->data,
       args->hostname,
+      args->kill_time,
       args->argc,
       args->argv,
       args->properties);
@@ -147,8 +168,8 @@ smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
  * \brief Internal function to create a process.
  *
  * This function actually creates the process.
- * It may be called when a REQ_PROCESS_CREATE request occurs,
- * or directly for SIMIX internal purposes.
+ * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
+ * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
  *
  * \return the process created
  */
@@ -157,6 +178,7 @@ void SIMIX_process_create(smx_process_t *process,
                           xbt_main_func_t code,
                           void *data,
                           const char *hostname,
+                          double kill_time,
                           int argc, char **argv,
                           xbt_dict_t properties) {
 
@@ -180,6 +202,7 @@ void SIMIX_process_create(smx_process_t *process,
     (*process)->smx_host = host;
     (*process)->data = data;
     (*process)->comms = xbt_fifo_new();
+    (*process)->simcall.issuer = *process;
 
     XBT_VERB("Create context %s", (*process)->name);
     (*process)->context = SIMIX_context_new(code, argc, argv,
@@ -201,6 +224,14 @@ void SIMIX_process_create(smx_process_t *process,
     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, host->name);
     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
   }
+
+  if (kill_time > SIMIX_get_clock()) {
+    if (simix_global->kill_process_function) {
+      XBT_DEBUG("Process %s(%s) will be kill at time %f", (*process)->name,
+          (*process)->smx_host->name, kill_time);
+      SIMIX_timer_set(kill_time, simix_global->kill_process_function, *process);
+    }
+  }
 }
 
 /**
@@ -225,7 +256,7 @@ void SIMIX_process_runall(void)
 /**
  * \brief Internal function to kill a SIMIX process.
  *
- * This function may be called when a REQ_PROCESS_KILL request occurs,
+ * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
  * or directly for SIMIX internal purposes.
  *
  * \param process poor victim
@@ -250,6 +281,7 @@ void SIMIX_process_kill(smx_process_t process) {
         break;
 
       case SIMIX_ACTION_COMMUNICATE:
+        xbt_fifo_remove(process->comms, process->waiting_action);
         SIMIX_comm_destroy(process->waiting_action);
         break;
 
@@ -258,17 +290,17 @@ void SIMIX_process_kill(smx_process_t process) {
        break;
 
       case SIMIX_ACTION_SYNCHRO:
-       SIMIX_synchro_stop_waiting(process, &process->request);
+       SIMIX_synchro_stop_waiting(process, &process->simcall);
        SIMIX_synchro_destroy(process->waiting_action);
        break;
 
       case SIMIX_ACTION_IO:
-       THROW_UNIMPLEMENTED;
-       break;
+        SIMIX_io_destroy(process->waiting_action);
+        break;
     }
   }
-
-  xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
+  if(!xbt_dynar_member(simix_global->process_to_run, &(process)))
+    xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 }
 
 /**
@@ -285,7 +317,7 @@ void SIMIX_process_killall(smx_process_t issuer)
     }
   }
 
-  SIMIX_context_runall(simix_global->process_to_run);
+  SIMIX_context_runall();
 
   SIMIX_process_empty_trash();
 }
@@ -304,30 +336,35 @@ void SIMIX_pre_process_change_host(smx_process_t process, smx_host_t dest)
   process->new_host = dest;
 }
 
-void SIMIX_pre_process_suspend(smx_req_t req)
+void SIMIX_pre_process_suspend(smx_simcall_t simcall)
 {
-  smx_process_t process = req->process_suspend.process;
-  SIMIX_process_suspend(process, req->issuer);
-
-  if (process != req->issuer) {
-    SIMIX_request_answer(req);
+  smx_process_t process = simcall->process_suspend.process;
+  smx_action_t action_suspend =
+      SIMIX_process_suspend(process, simcall->issuer);
+
+  if (process != simcall->issuer) {
+    SIMIX_simcall_answer(simcall);
+  } else {
+    xbt_fifo_push(action_suspend->simcalls, simcall);
+    process->waiting_action = action_suspend;
+    SIMIX_host_execution_suspend(process->waiting_action);
   }
-  /* If we are suspending ourselves, then just do not replay the request. */
+  /* If we are suspending ourselves, then just do not finish the simcall now */
 }
 
-void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
+smx_action_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 {
   xbt_assert((process != NULL), "Invalid parameters");
 
   if (process->suspended) {
     XBT_DEBUG("Process '%s' is already suspended", process->name);
-    return;
+    return NULL;
   }
 
   process->suspended = 1;
 
   /* If we are suspending another process, and it is waiting on an action,
-     suspend it's action. */
+     suspend its action. */
   if (process != issuer) {
 
     if (process->waiting_action) {
@@ -347,10 +384,21 @@ void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
           SIMIX_process_sleep_suspend(process->waiting_action);
           break;
 
+        case SIMIX_ACTION_SYNCHRO:
+          /* Suspension is delayed to when the process is rescheduled. */
+          break;
+
         default:
-          THROW_IMPOSSIBLE;
+          xbt_die("Internal error in SIMIX_process_suspend: unexpected action type %d",
+              (int)process->waiting_action->type);
       }
+      return NULL;
+    } else {
+      /* Suspension is delayed to when the process is rescheduled. */
+      return NULL;
     }
+  } else {
+    return SIMIX_host_execute("suspend", process->smx_host, 0.0, 1.0);
   }
 }
 
@@ -358,11 +406,14 @@ void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 {
   xbt_assert((process != NULL), "Invalid parameters");
 
-  if (!process->suspended) {
-    XBT_DEBUG("Process '%s' is not suspended", process->name);
+  XBT_IN("process = %p, issuer = %p", process, issuer);
+
+  if(process->context->iwannadie) {
+    XBT_VERB("Ignoring request to suspend a process that is currently dying.");
     return;
   }
 
+  if(!process->suspended) return;
   process->suspended = 0;
 
   /* If we are resuming another process, resume the action it was waiting for
@@ -386,14 +437,19 @@ void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
           SIMIX_process_sleep_resume(process->waiting_action);
           break;
 
+        case SIMIX_ACTION_SYNCHRO:
+          /* I cannot resume it now. This is delayed to when the process is rescheduled at
+           * the end of the synchro. */
+          break;
+
         default:
-          THROW_IMPOSSIBLE;
+          xbt_die("Internal error in SIMIX_process_resume: unexpected action type %d",
+              (int)process->waiting_action->type);
       }
     }
-    else {
-      xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
-    }
-  }
+  } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
+
+  XBT_OUT();
 }
 
 int SIMIX_process_get_maxpid(void) {
@@ -405,18 +461,21 @@ int SIMIX_process_count(void)
   return xbt_swag_size(simix_global->process_list);
 }
 
-void* SIMIX_process_self_get_data(void)
+void* SIMIX_process_self_get_data(smx_process_t self)
 {
-  smx_process_t me = SIMIX_process_self();
-  if (!me) {
+  xbt_assert(self == SIMIX_process_self(), "This is not the current process");
+
+  if (!self) {
     return NULL;
   }
-  return SIMIX_process_get_data(me);
+  return SIMIX_process_get_data(self);
 }
 
-void SIMIX_process_self_set_data(void *data)
+void SIMIX_process_self_set_data(smx_process_t self, void *data)
 {
-  SIMIX_process_set_data(SIMIX_process_self(), data);
+  xbt_assert(self == SIMIX_process_self(), "This is not the current process");
+
+  SIMIX_process_set_data(self, data);
 }
 
 void* SIMIX_process_get_data(smx_process_t process)
@@ -434,7 +493,7 @@ smx_host_t SIMIX_process_get_host(smx_process_t process)
   return process->smx_host;
 }
 
-/* needs to be public and without request because it is called
+/* needs to be public and without simcall because it is called
    by exceptions and logging events */
 const char* SIMIX_process_self_get_name(void) {
 
@@ -472,17 +531,17 @@ xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
   return process->properties;
 }
 
-void SIMIX_pre_process_sleep(smx_req_t req)
+void SIMIX_pre_process_sleep(smx_simcall_t simcall)
 {
   if (MC_IS_ENABLED) {
-    MC_process_clock_add(req->issuer, req->process_sleep.duration);
-    req->process_sleep.result = SIMIX_DONE;
-    SIMIX_request_answer(req);
+    MC_process_clock_add(simcall->issuer, simcall->process_sleep.duration);
+    simcall->process_sleep.result = SIMIX_DONE;
+    SIMIX_simcall_answer(simcall);
     return;
   }
-  smx_action_t action = SIMIX_process_sleep(req->issuer, req->process_sleep.duration);
-  xbt_fifo_push(action->request_list, req);
-  req->issuer->waiting_action = action;
+  smx_action_t action = SIMIX_process_sleep(simcall->issuer, simcall->process_sleep.duration);
+  xbt_fifo_push(action->simcalls, simcall);
+  simcall->issuer->waiting_action = action;
 }
 
 smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
@@ -516,10 +575,10 @@ smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
 
 void SIMIX_post_process_sleep(smx_action_t action)
 {
-  smx_req_t req;
+  smx_simcall_t simcall;
   e_smx_state_t state;
 
-  while ((req = xbt_fifo_shift(action->request_list))) {
+  while ((simcall = xbt_fifo_shift(action->simcalls))) {
 
     switch(surf_workstation_model->action_state_get(action->sleep.surf_sleep)){
       case SURF_ACTION_FAILED:
@@ -534,9 +593,9 @@ void SIMIX_post_process_sleep(smx_action_t action)
         THROW_IMPOSSIBLE;
         break;
     }
-    req->process_sleep.result = state;
-    req->issuer->waiting_action = NULL;
-    SIMIX_request_answer(req);
+    simcall->process_sleep.result = state;
+    simcall->issuer->waiting_action = NULL;
+    SIMIX_simcall_answer(simcall);
   }
   SIMIX_process_sleep_destroy(action);
 }
@@ -560,13 +619,15 @@ void SIMIX_process_sleep_resume(smx_action_t action)
 }
 
 /** 
- * Calling this function makes the process to yield.
- * Only the processes can call this function, giving back the control to maestro
+ * \brief Calling this function makes the process to yield.
+ *
+ * Only the current process can call this function, giving back the control to
+ * maestro.
+ *
+ * \param self the current process
  */
-void SIMIX_process_yield(void)
+void SIMIX_process_yield(smx_process_t self)
 {
-  smx_process_t self = SIMIX_process_self();
-
   XBT_DEBUG("Yield process '%s'", self->name);
 
   /* Go into sleep and return control to maestro */
@@ -580,10 +641,16 @@ void SIMIX_process_yield(void)
     SIMIX_context_stop(self->context);
   }
 
+  if(self->suspended) {
+    xbt_assert(!self->doexception, "Gloups! This exception may be lost by subsequent calls.");
+    self->suspended = 0;
+    SIMIX_process_suspend(self,self);
+  }
+
   if (self->doexception) {
     XBT_DEBUG("Wait, maestro left me an exception");
     self->doexception = 0;
-    RETHROW;
+    SMX_THROW();
   }
   
   if (self->new_host) {
@@ -612,3 +679,35 @@ smx_context_t SIMIX_process_get_context(smx_process_t p) {
 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
   p->context = c;
 }
+
+/**
+ * \brief Returns the list of processes to run.
+ */
+xbt_dynar_t SIMIX_process_get_runnable(void)
+{
+  return simix_global->process_to_run;
+}
+
+/**
+ * \brief Returns the process from PID.
+ */
+smx_process_t SIMIX_process_from_PID(int PID)
+{
+       smx_process_t proc;
+       xbt_swag_foreach(proc, simix_global->process_list)
+       {
+        if(proc->pid == PID)
+        return proc;
+       }
+       return NULL;
+}
+
+/** @brief returns a dynar containg all currently existing processes */
+xbt_dynar_t SIMIX_processes_as_dynar(void) {
+  smx_process_t proc;
+  xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
+  xbt_swag_foreach(proc, simix_global->process_list) {
+    xbt_dynar_push(res,&proc);
+  }
+  return res;
+}