From: Martin Quinson <martin.quinson@loria.fr>
Date: Sun, 26 Apr 2015 21:39:35 +0000 (+0200)
Subject: Remove the kill_time timer when a process stops (fix #18888)
X-Git-Tag: v3_12~732^2~50^2~2
X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/commitdiff_plain/9bd89d253a72967d4bb3c041c5fb5df8b210a556

Remove the kill_time timer when a process stops (fix #18888)
---

diff --git a/ChangeLog b/ChangeLog
index af38a7ee48..455983dd86 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -32,6 +32,8 @@ SimGrid (3.12) NOT RELEASED; urgency=low
    For example, a communication is a sort of synchronization involving
    the communicating processes (that may block until the exchange) and
    the platform. The same can be said from computations, etc.
+ * Bug fixes:
+   - #18888: segfault when a process ends before its kill_time
    
  SMPI: 
  * New functions
diff --git a/include/simgrid/simix.h b/include/simgrid/simix.h
index a79adf3726..ba20c93bfc 100644
--- a/include/simgrid/simix.h
+++ b/include/simgrid/simix.h
@@ -52,8 +52,6 @@ typedef enum {
 } e_smx_state_t;
 /** @} */
 
-typedef struct s_smx_timer* smx_timer_t;
-
 /* ******************************** Synchro ************************************ */
 /**
  * \ingroup simix_synchro_management
@@ -235,8 +233,12 @@ XBT_PUBLIC(void) SIMIX_run(void);
 XBT_PUBLIC(double) SIMIX_get_clock(void);
 
 /* Timer functions FIXME: should these be public? */
-XBT_PUBLIC(void) SIMIX_timer_set(double date, void *function, void *arg);
+typedef struct s_smx_timer* smx_timer_t;
+
+XBT_PUBLIC(smx_timer_t) SIMIX_timer_set(double date, void *function, void *arg);
+XBT_PUBLIC(void) SIMIX_timer_remove(smx_timer_t timer);
 XBT_PUBLIC(double) SIMIX_timer_next(void);
+XBT_PUBLIC(double) SIMIX_timer_get_date(smx_timer_t timer);
 
 XBT_PUBLIC(void) SIMIX_display_process_status(void);
 
diff --git a/src/simix/libsmx.c b/src/simix/libsmx.c
index 35b3c8ec62..f604cec2fd 100644
--- a/src/simix/libsmx.c
+++ b/src/simix/libsmx.c
@@ -737,7 +737,7 @@ void simcall_process_set_kill_time(smx_process_t process, double kill_time)
     if (simix_global->kill_process_function) {
       XBT_DEBUG("Set kill time %f for process %s(%s)",kill_time, process->name,
           sg_host_name(process->smx_host));
-      SIMIX_timer_set(kill_time, simix_global->kill_process_function, process);
+      process->kill_timer = SIMIX_timer_set(kill_time, simix_global->kill_process_function, process);
     }
   }
 }
diff --git a/src/simix/smx_global.c b/src/simix/smx_global.c
index 0011f977f2..2689131134 100644
--- a/src/simix/smx_global.c
+++ b/src/simix/smx_global.c
@@ -516,7 +516,7 @@ void SIMIX_run(void)
  *   \param arg Parameters of the function
  *
  */
-XBT_INLINE void SIMIX_timer_set(double date, void *function, void *arg)
+XBT_INLINE smx_timer_t SIMIX_timer_set(double date, void *function, void *arg)
 {
   smx_timer_t timer = xbt_new0(s_smx_timer_t, 1);
 
@@ -524,6 +524,16 @@ XBT_INLINE void SIMIX_timer_set(double date, void *function, void *arg)
   timer->func = function;
   timer->args = arg;
   xbt_heap_push(simix_timers, timer, date);
+  return timer;
+}
+/** @brief cancels a timer that was added earlier */
+XBT_INLINE void SIMIX_timer_remove(smx_timer_t timer) {
+	xbt_heap_rm_elm(simix_timers, timer, timer->date);
+}
+
+/** @brief Returns the date at which the timer will trigger (or 0 if NULL timer) */
+XBT_INLINE double SIMIX_timer_get_date(smx_timer_t timer) {
+	return timer?timer->date:0;
 }
 
 /**
diff --git a/src/simix/smx_process.c b/src/simix/smx_process.c
index 8fa3c6b8da..d466c2c9d6 100644
--- a/src/simix/smx_process.c
+++ b/src/simix/smx_process.c
@@ -49,6 +49,10 @@ void SIMIX_process_cleanup(smx_process_t process)
 
   SIMIX_process_on_exit_runall(process);
 
+  /* Unregister from the kill timer if any */
+  if (process->kill_timer != NULL)
+	  SIMIX_timer_remove(process->kill_timer);
+
   /* cancel non-blocking communications */
   smx_synchro_t synchro;
   while ((synchro = xbt_fifo_pop(process->comms))) {
@@ -157,7 +161,7 @@ void SIMIX_process_stop(smx_process_t arg) {
   if (arg->auto_restart && !SIMIX_host_get_state(arg->smx_host)) {
     SIMIX_host_add_auto_restart_process(arg->smx_host,arg->name,arg->code, arg->data,
                                         sg_host_name(arg->smx_host),
-                                        arg->kill_time,
+                                        SIMIX_timer_get_date(arg->kill_timer),
                                         arg->argc,arg->argv,arg->properties,
                                         arg->auto_restart);
   }
@@ -260,7 +264,6 @@ void SIMIX_process_create(smx_process_t *process,
     (*process)->code = code;
     (*process)->argc = argc;
     (*process)->argv = argv;
-    (*process)->kill_time = kill_time;
 
 
     XBT_VERB("Create context %s", (*process)->name);
@@ -290,7 +293,7 @@ void SIMIX_process_create(smx_process_t *process,
     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
       XBT_DEBUG("Process %s(%s) will be kill at time %f", (*process)->name,
           sg_host_name((*process)->smx_host), kill_time);
-      SIMIX_timer_set(kill_time, simix_global->kill_process_function, *process);
+      (*process)->kill_timer = SIMIX_timer_set(kill_time, simix_global->kill_process_function, *process);
     }
   }
 }
@@ -952,18 +955,15 @@ void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
   return SIMIX_process_restart(process, simcall->issuer);
 }
-/**
- * \brief Restart a process.
- * Restart a process, starting it again from the beginning.
- */
+/** @brief Restart a process, starting it again from the beginning. */
 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_name(process->smx_host));
   //retrieve the arguments of the old process
-  //FIXME: Factorise this with SIMIX_host_add_auto_restart_process ?
+  //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
   s_smx_process_arg_t arg;
   arg.code = process->code;
   arg.hostname = sg_host_name(process->smx_host);
-  arg.kill_time = process->kill_time;
+  arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
   arg.argc = process->argc;
   arg.data = process->data;
   int i;
diff --git a/src/simix/smx_process_private.h b/src/simix/smx_process_private.h
index 8b4f8ed9ba..b3cd4bcede 100644
--- a/src/simix/smx_process_private.h
+++ b/src/simix/smx_process_private.h
@@ -58,8 +58,7 @@ typedef struct s_smx_process {
   xbt_main_func_t code;
   int argc;
   char **argv;
-  double kill_time;
-
+  smx_timer_t kill_timer;
 } s_smx_process_t;