Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
make simix more resilient to PEBKAC
[simgrid.git] / src / simix / smx_global.c
index 0b2b78c..3491894 100644 (file)
@@ -4,6 +4,8 @@
 /* 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 <stdlib.h>
+
 #include "smx_private.h"
 #include "xbt/heap.h"
 #include "xbt/sysdep.h"
 #include "xbt/str.h"
 #include "xbt/ex.h"             /* ex_backtrace_display */
 #include "mc/mc.h"
+#include "mc/mc_replay.h"
 #include "simgrid/sg_config.h"
 
 #ifdef HAVE_MC
 #include "mc/mc_private.h"
+#include "mc/mc_protocol.h"
+#include "mc/mc_client.h"
 #endif
+#include "mc/mc_record.h"
 
 #ifdef HAVE_SMPI
 #include "smpi/private.h"
@@ -28,9 +34,9 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_kernel, simix,
 smx_global_t simix_global = NULL;
 static xbt_heap_t simix_timers = NULL;
 
-static void* SIMIX_action_mallocator_new_f(void);
-static void SIMIX_action_mallocator_free_f(void* action);
-static void SIMIX_action_mallocator_reset_f(void* action);
+static void* SIMIX_synchro_mallocator_new_f(void);
+static void SIMIX_synchro_mallocator_free_f(void* synchro);
+static void SIMIX_synchro_mallocator_reset_f(void* synchro);
 
 /* FIXME: Yeah, I'll do it in a portable maner one day [Mt] */
 #include <signal.h>
@@ -71,14 +77,14 @@ static void _XBT_CALL segvhandler(int signum, siginfo_t *siginfo, void *context)
 #ifdef HAVE_SMPI
     if (smpi_enabled() && !smpi_privatize_global_variables) {
       fprintf(stderr,
-        "Try to enable SMPI variable privatization with --cfg:smpi/privatize_global_variable:yes.\n");
+        "Try to enable SMPI variable privatization with --cfg=smpi/privatize_global_variables:yes.\n");
     }
 #endif
   }
 #ifdef HAVE_MC
   if (MC_is_active()) {
-    if (mc_stack_safety) {
-      MC_dump_stack_safety(mc_stack_safety);
+    if (mc_stack) {
+      MC_dump_stack_safety(mc_stack);
     }
     MC_print_statistics(mc_stats);
   }
@@ -168,10 +174,11 @@ void SIMIX_global_init(int *argc, char **argv)
     simix_global->create_process_function = SIMIX_process_create;
     simix_global->kill_process_function = SIMIX_process_kill;
     simix_global->cleanup_process_function = SIMIX_process_cleanup;
-    simix_global->action_mallocator = xbt_mallocator_new(65536,
-        SIMIX_action_mallocator_new_f, SIMIX_action_mallocator_free_f,
-        SIMIX_action_mallocator_reset_f);
+    simix_global->synchro_mallocator = xbt_mallocator_new(65536,
+        SIMIX_synchro_mallocator_new_f, SIMIX_synchro_mallocator_free_f,
+        SIMIX_synchro_mallocator_reset_f);
     simix_global->autorestart = SIMIX_host_restart_processes;
+    simix_global->mutex = xbt_os_mutex_init();
 
     surf_init(argc, argv);      /* Initialize SURF structures */
     SIMIX_context_mod_init();
@@ -199,12 +206,24 @@ void SIMIX_global_init(int *argc, char **argv)
     simix_timers = xbt_heap_new(8, &free);
   }
 
-  SIMIX_HOST_LEVEL = xbt_lib_add_level(host_lib,SIMIX_host_destroy);
   SIMIX_STORAGE_LEVEL = xbt_lib_add_level(storage_lib, SIMIX_storage_destroy);
 
   if (sg_cfg_get_boolean("clean_atexit"))
     atexit(SIMIX_clean);
 
+#ifdef HAVE_MC
+  // The communication initialization is done ASAP.
+  // We need to communicate  initialization of the different layers to the model-checker.
+  if (mc_mode == MC_MODE_NONE) {
+    if (getenv(MC_ENV_SOCKET_FD)) {
+      mc_mode = MC_MODE_CLIENT;
+      MC_client_init();
+      MC_client_hello();
+      MC_client_handle_messages();
+    }
+  }
+#endif
+
   if (_sg_cfg_exit_asap)
     exit(0);
 }
@@ -223,7 +242,14 @@ void SIMIX_clean(void)
 #endif
   if (cleaned) return; // to avoid double cleaning by java and C
   cleaned = 1;
-  /* Kill everyone (except maestro) */
+  XBT_DEBUG("SIMIX_clean called. Simulation's over.");
+  if (!xbt_dynar_is_empty(simix_global->process_to_run) && SIMIX_get_clock() == 0.0) {
+         XBT_CRITICAL("   ");
+         XBT_CRITICAL("The time is still 0, and you still have processes ready to run.");
+         XBT_CRITICAL("It seems that you forgot to run the simulation that you setup.");
+         xbt_die("Bailing out to avoid that stop-before-start madness. Please fix your code.");
+  }
+  /* Kill all processes (but maestro) */
   SIMIX_process_killall(simix_global->maestro_process, 1);
 
   /* Exit the SIMIX network module */
@@ -240,6 +266,9 @@ void SIMIX_clean(void)
   simix_global->process_to_destroy = NULL;
   xbt_dict_free(&(simix_global->registered_functions));
 
+  xbt_os_mutex_destroy(simix_global->mutex);
+  simix_global->mutex = NULL;
+
   /* Let's free maestro now */
   SIMIX_context_free(simix_global->maestro_process->context);
   xbt_free(simix_global->maestro_process->running_ctx);
@@ -264,7 +293,7 @@ void SIMIX_clean(void)
   xbt_os_timer_free(simix_global->timer_par);
 #endif
 
-  xbt_mallocator_free(simix_global->action_mallocator);
+  xbt_mallocator_free(simix_global->synchro_mallocator);
   xbt_free(simix_global);
   simix_global = NULL;
 
@@ -280,7 +309,7 @@ void SIMIX_clean(void)
  */
 XBT_INLINE double SIMIX_get_clock(void)
 {
-  if(MC_is_active()){
+  if(MC_is_active() || MC_record_replay_is_active()){
     return MC_process_clock_get(SIMIX_process_self());
   }else{
     return surf_get_clock();
@@ -306,6 +335,12 @@ static int process_syscall_color(void *p)
  */
 void SIMIX_run(void)
 {
+  if(MC_record_path) {
+    MC_record_replay_init();
+    MC_record_replay_from_string(MC_record_path);
+    return;
+  }
+
   double time = 0;
   smx_process_t process;
   surf_action_t action;
@@ -334,7 +369,7 @@ void SIMIX_run(void)
       xbt_os_cputimer_resume(simix_global->timer_seq);
 #endif
 
-      /* Move all killing processes to the end of the list, because killing a process that have an ongoing simcall is a bad idea */
+      /* Move all killer processes to the end of the list, because killing a process that have an ongoing simcall is a bad idea */
       xbt_dynar_three_way_partition(simix_global->process_that_ran, process_syscall_color);
 
       /* answer sequentially and in a fixed arbitrary order all the simcalls that were issued during that sub-round */
@@ -367,16 +402,16 @@ void SIMIX_run(void)
        *            - because the communication failed or were canceled after startup. In this case, it's called from the function
        *              we are in, by the chunk:
        *                       set = model->states.failed_action_set;
-       *                       while ((action = xbt_swag_extract(set)))
-       *                          SIMIX_simcall_post((smx_action_t) action->data);
+       *                       while ((synchro = xbt_swag_extract(set)))
+       *                          SIMIX_simcall_post((smx_synchro_t) synchro->data);
        *              This order is also fixed because it depends of the order in which the surf actions were
        *              added to the system, and only maestro can add stuff this way, through simcalls.
-       *              We thus use the inductive hypothesis once again to conclude that the order in which actions are
+       *              We thus use the inductive hypothesis once again to conclude that the order in which synchros are
        *              poped out of the swag does not depend on the user code's execution order.
-       *            - because the communication terminated. In this case, actions are served in the order given by
+       *            - because the communication terminated. In this case, synchros are served in the order given by
        *                       set = model->states.done_action_set;
-       *                       while ((action = xbt_swag_extract(set)))
-       *                          SIMIX_simcall_post((smx_action_t) action->data);
+       *                       while ((synchro = xbt_swag_extract(set)))
+       *                          SIMIX_simcall_post((smx_synchro_t) synchro->data);
        *              and the argument is very similar to the previous one.
        *            So, in any case, the orders of calls to SIMIX_comm_finish() do not depend on the order in which user processes are executed.
        *          So, in any cases, the orders of processes within process_to_run do not depend on the order in which user processes were executed previously.
@@ -394,15 +429,33 @@ void SIMIX_run(void)
 
       xbt_dynar_foreach(simix_global->process_that_ran, iter, process) {
         if (process->simcall.call != SIMCALL_NONE) {
-          SIMIX_simcall_pre(&process->simcall, 0);
+          SIMIX_simcall_handle(&process->simcall, 0);
+        }
+      }
+      /* Wake up all processes waiting for a Surf action to finish */
+      xbt_dynar_foreach(model_list, iter, model) {
+        XBT_DEBUG("Handling process whose action failed");
+        while ((action = surf_model_extract_failed_action_set(model))) {
+          XBT_DEBUG("   Handling Action %p",action);
+          SIMIX_simcall_exit((smx_synchro_t) surf_action_get_data(action));
+        }
+        XBT_DEBUG("Handling process whose action terminated normally");
+        while ((action = surf_model_extract_done_action_set(model))) {
+          XBT_DEBUG("   Handling Action %p",action);
+          if (surf_action_get_data(action) == NULL)
+            XBT_DEBUG("probably vcpu's action %p, skip", action);
+          else
+            SIMIX_simcall_exit((smx_synchro_t) surf_action_get_data(action));
         }
       }
     }
 
     time = SIMIX_timer_next();
-    if (time != -1.0 || xbt_swag_size(simix_global->process_list) != 0)
+    if (time != -1.0 || xbt_swag_size(simix_global->process_list) != 0) {
+      XBT_DEBUG("Calling surf_solve");
       time = surf_solve(time);
-
+      XBT_DEBUG("Moving time ahead : %g", time);
+    }
     /* Notify all the hosts that have failed */
     /* FIXME: iterate through the list of failed host and mark each of them */
     /* as failed. On each host, signal all the running processes with host_fail */
@@ -419,14 +472,19 @@ void SIMIX_run(void)
 
     /* Wake up all processes waiting for a Surf action to finish */
     xbt_dynar_foreach(model_list, iter, model) {
-      while ((action = surf_model_extract_failed_action_set(model)))
-        SIMIX_simcall_post((smx_action_t) surf_action_get_data(action));
-
-      while ((action = surf_model_extract_done_action_set(model)))
+      XBT_DEBUG("Handling process whose action failed");
+      while ((action = surf_model_extract_failed_action_set(model))) {
+        XBT_DEBUG("   Handling Action %p",action);
+        SIMIX_simcall_exit((smx_synchro_t) surf_action_get_data(action));
+      }
+      XBT_DEBUG("Handling process whose action terminated normally");
+      while ((action = surf_model_extract_done_action_set(model))) {
+        XBT_DEBUG("   Handling Action %p",action);
         if (surf_action_get_data(action) == NULL)
           XBT_DEBUG("probably vcpu's action %p, skip", action);
         else
-          SIMIX_simcall_post((smx_action_t) surf_action_get_data(action));
+          SIMIX_simcall_exit((smx_synchro_t) surf_action_get_data(action));
+      }
     }
 
     /* Autorestart all process */
@@ -434,7 +492,7 @@ void SIMIX_run(void)
       char *hostname = NULL;
       xbt_dynar_foreach(host_that_restart,iter,hostname) {
         XBT_INFO("Restart processes on host: %s",hostname);
-        SIMIX_host_autorestart(SIMIX_host_get_by_name(hostname));
+        SIMIX_host_autorestart(sg_host_by_name(hostname));
       }
       xbt_dynar_reset(host_that_restart);
     }
@@ -444,16 +502,12 @@ void SIMIX_run(void)
 
 
     XBT_DEBUG("### time %f, empty %d", time, xbt_dynar_is_empty(simix_global->process_to_run));
-    // !(time == -1.0 && xbt_dynar_is_empty())
-
 
   } while (time != -1.0 || !xbt_dynar_is_empty(simix_global->process_to_run));
 
   if (xbt_swag_size(simix_global->process_list) != 0) {
 
-#ifdef HAVE_TRACING
-    TRACE_end();
-#endif
+       TRACE_end();
 
     XBT_CRITICAL("Oops ! Deadlock or code not perfectly clean.");
     SIMIX_display_process_status();
@@ -470,7 +524,7 @@ 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);
 
@@ -478,6 +532,16 @@ 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;
 }
 
 /**
@@ -538,82 +602,74 @@ void SIMIX_display_process_status(void)
     ("Legend of the following listing: \"Process <pid> (<name>@<host>): <status>\"");
   xbt_swag_foreach(process, simix_global->process_list) {
 
-    if (process->waiting_action) {
+    if (process->waiting_synchro) {
 
-      const char* action_description = "unknown";
-      switch (process->waiting_action->type) {
+      const char* synchro_description = "unknown";
+      switch (process->waiting_synchro->type) {
 
-      case SIMIX_ACTION_EXECUTE:
-        action_description = "execution";
+      case SIMIX_SYNC_EXECUTE:
+        synchro_description = "execution";
         break;
 
-      case SIMIX_ACTION_PARALLEL_EXECUTE:
-        action_description = "parallel execution";
+      case SIMIX_SYNC_PARALLEL_EXECUTE:
+        synchro_description = "parallel execution";
         break;
 
-      case SIMIX_ACTION_COMMUNICATE:
-        action_description = "communication";
+      case SIMIX_SYNC_COMMUNICATE:
+        synchro_description = "communication";
         break;
 
-      case SIMIX_ACTION_SLEEP:
-        action_description = "sleeping";
+      case SIMIX_SYNC_SLEEP:
+        synchro_description = "sleeping";
         break;
 
-      case SIMIX_ACTION_JOIN:
-        action_description = "joining";
+      case SIMIX_SYNC_JOIN:
+        synchro_description = "joining";
         break;
 
-      case SIMIX_ACTION_SYNCHRO:
-        action_description = "synchronization";
+      case SIMIX_SYNC_SYNCHRO:
+        synchro_description = "synchronization";
         break;
 
-      case SIMIX_ACTION_IO:
-        action_description = "I/O";
-        break;
-      /* **************************************/
-      /* TUTORIAL: New API                    */
-      case SIMIX_ACTION_NEW_API:
-        action_description = "NEW API";
-      /* **************************************/
-
+      case SIMIX_SYNC_IO:
+        synchro_description = "I/O";
         break;
       }
-      XBT_INFO("Process %lu (%s@%s): waiting for %s action %p (%s) in state %d to finish",
-          process->pid, process->name, sg_host_name(process->smx_host),
-          action_description, process->waiting_action,
-          process->waiting_action->name, (int)process->waiting_action->state);
+      XBT_INFO("Process %lu (%s@%s): waiting for %s synchro %p (%s) in state %d to finish",
+          process->pid, process->name, sg_host_name(process->host),
+          synchro_description, process->waiting_synchro,
+          process->waiting_synchro->name, (int)process->waiting_synchro->state);
     }
     else {
-      XBT_INFO("Process %lu (%s@%s)", process->pid, process->name, sg_host_name(process->smx_host));
+      XBT_INFO("Process %lu (%s@%s)", process->pid, process->name, sg_host_name(process->host));
     }
   }
 }
 
-static void* SIMIX_action_mallocator_new_f(void) {
-  smx_action_t action = xbt_new(s_smx_action_t, 1);
-  action->simcalls = xbt_fifo_new();
-  return action;
+static void* SIMIX_synchro_mallocator_new_f(void) {
+  smx_synchro_t synchro = xbt_new(s_smx_synchro_t, 1);
+  synchro->simcalls = xbt_fifo_new();
+  return synchro;
 }
 
-static void SIMIX_action_mallocator_free_f(void* action) {
-  xbt_fifo_free(((smx_action_t) action)->simcalls);
-  xbt_free(action);
+static void SIMIX_synchro_mallocator_free_f(void* synchro) {
+  xbt_fifo_free(((smx_synchro_t) synchro)->simcalls);
+  xbt_free(synchro);
 }
 
-static void SIMIX_action_mallocator_reset_f(void* action) {
+static void SIMIX_synchro_mallocator_reset_f(void* synchro) {
 
   // we also recycle the simcall list
-  xbt_fifo_t fifo = ((smx_action_t) action)->simcalls;
+  xbt_fifo_t fifo = ((smx_synchro_t) synchro)->simcalls;
   xbt_fifo_reset(fifo);
-  memset(action, 0, sizeof(s_smx_action_t));
-  ((smx_action_t) action)->simcalls = fifo;
+  memset(synchro, 0, sizeof(s_smx_synchro_t));
+  ((smx_synchro_t) synchro)->simcalls = fifo;
 }
 
-xbt_dict_t SIMIX_pre_asr_get_properties(smx_simcall_t simcall, const char *name){
+xbt_dict_t simcall_HANDLER_asr_get_properties(smx_simcall_t simcall, const char *name){
   return SIMIX_asr_get_properties(name);
 }
 xbt_dict_t SIMIX_asr_get_properties(const char *name)
 {
   return xbt_lib_get_or_null(as_router_lib, name, ROUTING_PROP_ASR_LEVEL);
 }
-