Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Delete timer on removal.
[simgrid.git] / src / simix / smx_global.cpp
index bfb56d5..90deaa8 100644 (file)
@@ -1,5 +1,4 @@
-/* Copyright (c) 2007-2015. The SimGrid Team.
- * All rights reserved.                                                     */
+/* Copyright (c) 2007-2017. 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. */
@@ -7,43 +6,45 @@
 #include <functional>
 #include <memory>
 
-#include <signal.h> /* Signal handling */
-#include <stdlib.h>
 #include "src/internal_config.h"
+#include <csignal> /* Signal handling */
+#include <cstdlib>
 
+#include <xbt/algorithm.hpp>
 #include <xbt/functional.hpp>
 
-#include <simgrid/s4u/host.hpp>
+#include "simgrid/s4u/Engine.hpp"
+#include "simgrid/s4u/Host.hpp"
 
 #include "src/surf/surf_interface.hpp"
-#include "src/surf/storage_interface.hpp"
 #include "src/surf/xml/platf.hpp"
 #include "smx_private.h"
-#include "xbt/str.h"
 #include "xbt/ex.h"             /* ex_backtrace_display */
+
 #include "mc/mc.h"
-#include "src/mc/mc_replay.h"
 #include "simgrid/sg_config.h"
+#include "src/mc/mc_replay.h"
+#include "src/surf/StorageImpl.hpp"
+
+#include "src/smpi/include/smpi_process.hpp"
 
-#include "src/kernel/activity/SynchroExec.hpp"
-#include "src/kernel/activity/SynchroComm.hpp"
-#include "src/kernel/activity/SynchroSleep.hpp"
+#include "src/kernel/activity/CommImpl.hpp"
+#include "src/kernel/activity/ExecImpl.hpp"
+#include "src/kernel/activity/SleepImpl.hpp"
 #include "src/kernel/activity/SynchroIo.hpp"
 #include "src/kernel/activity/SynchroRaw.hpp"
 
-#if HAVE_MC
+#if SIMGRID_HAVE_MC
 #include "src/mc/mc_private.h"
-#include "src/mc/mc_protocol.h"
-#include "src/mc/Client.hpp"
-
-#include <stdlib.h>
-#include "src/mc/mc_protocol.h"
-#endif 
+#include "src/mc/remote/Client.hpp"
+#include "src/mc/remote/mc_protocol.h"
+#endif
 
 #include "src/mc/mc_record.h"
 
 #if HAVE_SMPI
-#include "src/smpi/private.h"
+#include "src/smpi/include/private.h"
+#include "src/smpi/include/private.hpp"
 #endif
 
 XBT_LOG_NEW_CATEGORY(simix, "All SIMIX categories");
@@ -53,11 +54,13 @@ std::unique_ptr<simgrid::simix::Global> simix_global;
 static xbt_heap_t simix_timers = nullptr;
 
 /** @brief Timer datatype */
-typedef struct s_smx_timer {
+typedef class s_smx_timer {
   double date = 0.0;
-  simgrid::xbt::Task<void()> callback;
+  s_smx_timer() = default;
 
-  s_smx_timer()=default;
+public:
+  simgrid::xbt::Task<void()> callback;
+  double getDate() { return date; }
   s_smx_timer(double date, simgrid::xbt::Task<void()> callback) : date(date), callback(std::move(callback)) {}
 } s_smx_timer_t;
 
@@ -81,25 +84,26 @@ static void segvhandler(int signum, siginfo_t *siginfo, void *context)
 {
   if (siginfo->si_signo == SIGSEGV && siginfo->si_code == SEGV_ACCERR) {
     fprintf(stderr, "Access violation detected.\n"
-                    "This can result from a programming error in your code or, although less likely,\n"
-                    "from a bug in SimGrid itself.  This can also be the sign of a bug in the OS or\n"
-                    "in third-party libraries.  Failing hardware can sometimes generate such errors\n"
-                    "too.\n"
-                    "Finally, if nothing of the above applies, this can result from a stack overflow.\n"
-                    "Try to increase stack size with --cfg=contexts/stack-size (current size is %d KiB).\n",
+                    "This probably comes from a programming error in your code, or from a stack\n"
+                    "overflow. If you are certain of your code, try increasing the stack size\n"
+                    "   --cfg=contexts/stack-size=XXX (current size is %d KiB).\n"
+                    "\n"
+                    "If it does not help, this may have one of the following causes:\n"
+                    "a bug in SimGrid, a bug in the OS or a bug in a third-party libraries.\n"
+                    "Failing hardware can sometimes generate such errors too.\n"
+                    "\n"
+                    "If you think you've found a bug in SimGrid, please report it along with a\n"
+                    "Minimal Working Example (MWE) reproducing your problem and a full backtrace\n"
+                    "of the fault captured with gdb or valgrind.\n",
             smx_context_stack_size / 1024);
-    if (XBT_LOG_ISENABLED(simix_kernel, xbt_log_priority_debug)) {
-      fprintf(stderr, "siginfo = {si_signo = %d, si_errno = %d, si_code = %d, si_addr = %p}\n",
-              siginfo->si_signo, siginfo->si_errno, siginfo->si_code, siginfo->si_addr);
-    }
   } else  if (siginfo->si_signo == SIGSEGV) {
     fprintf(stderr, "Segmentation fault.\n");
 #if HAVE_SMPI
-    if (smpi_enabled() && !smpi_privatize_global_variables) {
+    if (smpi_enabled() && smpi_privatize_global_variables == SMPI_PRIVATIZE_NONE) {
 #if HAVE_PRIVATIZATION
-      fprintf(stderr, "Try to enable SMPI variable privatization with --cfg=smpi/privatize-global-variables:yes.\n");
+      fprintf(stderr, "Try to enable SMPI variable privatization with --cfg=smpi/privatization:yes.\n");
 #else
-      fprintf(stderr, "Sadly, your system does not support --cfg=smpi/privatize-global-variables:yes (yet).\n");
+      fprintf(stderr, "Sadly, your system does not support --cfg=smpi/privatization:yes (yet).\n");
 #endif /* HAVE_PRIVATIZATION */
     }
 #endif /* HAVE_SMPI */
@@ -115,7 +119,8 @@ char sigsegv_stack[SIGSTKSZ];   /* alternate stack for SIGSEGV handler */
  */
 static void install_segvhandler()
 {
-  stack_t stack, old_stack;
+  stack_t stack;
+  stack_t old_stack;
   stack.ss_sp = sigsegv_stack;
   stack.ss_size = sizeof sigsegv_stack;
   stack.ss_flags = 0;
@@ -124,13 +129,14 @@ static void install_segvhandler()
     XBT_WARN("Failed to register alternate signal stack: %s", strerror(errno));
     return;
   }
-  if (!(old_stack.ss_flags & SS_DISABLE)) {
-    XBT_DEBUG("An alternate stack was already installed (sp=%p, size=%zd, flags=%x). Restore it.",
-              old_stack.ss_sp, old_stack.ss_size, old_stack.ss_flags);
+  if (not(old_stack.ss_flags & SS_DISABLE)) {
+    XBT_DEBUG("An alternate stack was already installed (sp=%p, size=%zu, flags=%x). Restore it.", old_stack.ss_sp,
+              old_stack.ss_size, (unsigned)old_stack.ss_flags);
     sigaltstack(&old_stack, nullptr);
   }
 
-  struct sigaction action, old_action;
+  struct sigaction action;
+  struct sigaction old_action;
   action.sa_sigaction = &segvhandler;
   action.sa_flags = SA_ONSTACK | SA_RESETHAND | SA_SIGINFO;
   sigemptyset(&action.sa_mask);
@@ -177,7 +183,7 @@ XBT_PUBLIC(void) set_maestro(std::function<void()> code)
 void SIMIX_set_maestro(void (*code)(void*), void* data)
 {
 #ifdef _WIN32
-  INFO("WARNING, SIMIX_set_maestro is believed to not work on windows. Please help us investigating this issue if you need that feature");
+  XBT_INFO("WARNING, SIMIX_set_maestro is believed to not work on windows. Please help us investigating this issue if you need that feature");
 #endif
   maestro_code = std::bind(code, data);
 }
@@ -191,19 +197,16 @@ void SIMIX_set_maestro(void (*code)(void*), void* data)
  */
 void SIMIX_global_init(int *argc, char **argv)
 {
-#if HAVE_MC
+#if SIMGRID_HAVE_MC
   // The communication initialization is done ASAP.
   // We need to communicate  initialization of the different layers to the model-checker.
   simgrid::mc::Client::initialize();
 #endif
 
-  if (!simix_global) {
+  if (not simix_global) {
     simix_global = std::unique_ptr<simgrid::simix::Global>(new simgrid::simix::Global());
 
     simgrid::simix::ActorImpl proc;
-    simix_global->process_to_run = xbt_dynar_new(sizeof(smx_actor_t), nullptr);
-    simix_global->process_that_ran = xbt_dynar_new(sizeof(smx_actor_t), nullptr);
-    simix_global->process_list = xbt_swag_new(xbt_swag_offset(proc, process_hookup));
     simix_global->process_to_destroy = xbt_swag_new(xbt_swag_offset(proc, destroy_hookup));
     simix_global->maestro_process = nullptr;
     simix_global->create_process_function = &SIMIX_process_create;
@@ -226,23 +229,19 @@ void SIMIX_global_init(int *argc, char **argv)
 #endif
     /* register a function to be called by SURF after the environment creation */
     sg_platf_init();
-    simgrid::surf::on_postparse.connect(SIMIX_post_create_environment);
+    simgrid::s4u::onPlatformCreated.connect(SIMIX_post_create_environment);
     simgrid::s4u::Host::onCreation.connect([](simgrid::s4u::Host& host) {
-      host.extension_set<simgrid::simix::Host>(new simgrid::simix::Host());
+      if (host.extension<simgrid::simix::Host>() == nullptr) // another callback to the same signal may have created it
+        host.extension_set<simgrid::simix::Host>(new simgrid::simix::Host());
     });
 
-    simgrid::surf::storageCreatedCallbacks.connect([](simgrid::surf::Storage* storage) {
-      const char* name = storage->getName();
-      // TODO, create sg_storage_by_name
-      sg_storage_t s = xbt_lib_get_elm_or_null(storage_lib, name);
-      xbt_assert(s != nullptr, "Storage not found for name %s", name);
-
-      SIMIX_storage_create(name, s, nullptr);
+    simgrid::surf::storageCreatedCallbacks.connect([](simgrid::surf::StorageImpl* storage) {
+      sg_storage_t s = simgrid::s4u::Storage::byName(storage->cname());
+      xbt_assert(s != nullptr, "Storage not found for name %s", storage->cname());
     });
-
-    SIMIX_STORAGE_LEVEL = xbt_lib_add_level(storage_lib, SIMIX_storage_destroy);
   }
-  if (!simix_timers)
+
+  if (not simix_timers)
     simix_timers = xbt_heap_new(8, [](void* p) {
       delete static_cast<smx_timer_t>(p);
     });
@@ -268,7 +267,7 @@ void SIMIX_clean()
 
 #if HAVE_SMPI
   if (SIMIX_process_count()>0){
-    if(smpi_process_initialized()){
+    if(smpi_process()->initialized()){
       xbt_die("Process exited without calling MPI_Finalize - Killing simulation");
     }else{
       XBT_WARN("Process called exit when leaving - Skipping cleanups");
@@ -279,7 +278,7 @@ void SIMIX_clean()
 
   smx_cleaned = 1;
   XBT_DEBUG("SIMIX_clean called. Simulation's over.");
-  if (!xbt_dynar_is_empty(simix_global->process_to_run) && SIMIX_get_clock() <= 0.0) {
+  if (not simix_global->process_to_run.empty() && 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.");
@@ -294,15 +293,17 @@ void SIMIX_clean()
   xbt_heap_free(simix_timers);
   simix_timers = nullptr;
   /* Free the remaining data structures */
-  xbt_dynar_free(&simix_global->process_to_run);
-  xbt_dynar_free(&simix_global->process_that_ran);
+  simix_global->process_to_run.clear();
+  simix_global->process_that_ran.clear();
   xbt_swag_free(simix_global->process_to_destroy);
-  xbt_swag_free(simix_global->process_list);
-  simix_global->process_list = nullptr;
+  simix_global->process_list.clear();
   simix_global->process_to_destroy = nullptr;
 
   xbt_os_mutex_destroy(simix_global->mutex);
   simix_global->mutex = nullptr;
+#if SIMGRID_HAVE_MC
+  xbt_dynar_free(&simix_global->actors_vector);
+#endif
 
   /* Let's free maestro now */
   delete simix_global->maestro_process->context;
@@ -316,7 +317,6 @@ void SIMIX_clean()
   surf_exit();
 
   simix_global = nullptr;
-  return;
 }
 
 
@@ -335,29 +335,16 @@ double SIMIX_get_clock()
   }
 }
 
-static int process_syscall_color(void *p)
-{
-  switch ((*(smx_actor_t *)p)->simcall.call) {
-  case SIMCALL_NONE:
-  case SIMCALL_PROCESS_KILL:
-    return 2;
-  case SIMCALL_PROCESS_RESUME:
-    return 1;
-  default:
-    return 0;
-  }
-}
-
 /** Wake up all processes waiting for a Surf action to finish */
 static void SIMIX_wake_processes()
 {
   surf_action_t action;
 
-  for(auto model : *all_existing_models) {
+  for (auto const& model : *all_existing_models) {
     XBT_DEBUG("Handling the processes whose action failed (if any)");
     while ((action = surf_model_extract_failed_action_set(model))) {
       XBT_DEBUG("   Handling Action %p",action);
-      SIMIX_simcall_exit((smx_activity_t) action->getData());
+      SIMIX_simcall_exit(static_cast<simgrid::kernel::activity::ActivityImpl*>(action->getData()));
     }
     XBT_DEBUG("Handling the processes whose action terminated normally (if any)");
     while ((action = surf_model_extract_done_action_set(model))) {
@@ -365,7 +352,7 @@ static void SIMIX_wake_processes()
       if (action->getData() == nullptr)
         XBT_DEBUG("probably vcpu's action %p, skip", action);
       else
-        SIMIX_simcall_exit((smx_activity_t) action->getData());
+        SIMIX_simcall_exit(static_cast<simgrid::kernel::activity::ActivityImpl*>(action->getData()));
     }
   }
 }
@@ -411,7 +398,7 @@ static bool SIMIX_execute_tasks()
       task();
 
     simix_global->tasksTemp.clear();
-  } while (!simix_global->tasks.empty());
+  } while (not simix_global->tasks.empty());
 
   return true;
 }
@@ -428,22 +415,18 @@ void SIMIX_run()
   }
 
   double time = 0;
-  smx_actor_t process;
 
   do {
-    XBT_DEBUG("New Schedule Round; size(queue)=%lu", xbt_dynar_length(simix_global->process_to_run));
+    XBT_DEBUG("New Schedule Round; size(queue)=%zu", simix_global->process_to_run.size());
 
     SIMIX_execute_tasks();
 
-    while (!xbt_dynar_is_empty(simix_global->process_to_run)) {
-      XBT_DEBUG("New Sub-Schedule Round; size(queue)=%lu", xbt_dynar_length(simix_global->process_to_run));
+    while (not simix_global->process_to_run.empty()) {
+      XBT_DEBUG("New Sub-Schedule Round; size(queue)=%zu", simix_global->process_to_run.size());
 
       /* Run all processes that are ready to run, possibly in parallel */
       SIMIX_process_runall();
 
-      /* 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 */
 
       /* WARNING, the order *must* be fixed or you'll jeopardize the simulation reproducibility (see RR-7653) */
@@ -499,8 +482,7 @@ void SIMIX_run()
        *   That would thus be a pure waste of time.
        */
 
-      unsigned int iter;
-      xbt_dynar_foreach(simix_global->process_that_ran, iter, process) {
+      for (smx_actor_t const& process : simix_global->process_that_ran) {
         if (process->simcall.call != SIMCALL_NONE) {
           SIMIX_simcall_handle(&process->simcall, 0);
         }
@@ -511,10 +493,16 @@ void SIMIX_run()
         SIMIX_wake_processes();
       } while (SIMIX_execute_tasks());
 
+      /* If only daemon processes remain, cancel their actions, mark them to die and reschedule them */
+      if (simix_global->process_list.size() == simix_global->daemons.size())
+        for (auto const& dmon : simix_global->daemons) {
+          XBT_DEBUG("Kill %s", dmon->cname());
+          SIMIX_process_kill(dmon, simix_global->maestro_process);
+        }
     }
 
     time = SIMIX_timer_next();
-    if (time > -1.0 || xbt_swag_size(simix_global->process_list) != 0) {
+    if (time > -1.0 || simix_global->process_list.empty() == false) {
       XBT_DEBUG("Calling surf_solve");
       time = surf_solve(time);
       XBT_DEBUG("Moving time ahead : %g", time);
@@ -534,8 +522,8 @@ void SIMIX_run()
     } while (again);
 
     /* Autorestart all process */
-    for (auto host: host_that_restart) {
-      XBT_INFO("Restart processes on host: %s", sg_host_get_name(host));
+    for (auto const& host : host_that_restart) {
+      XBT_INFO("Restart processes on host %s", host->getCname());
       SIMIX_host_autorestart(host);
     }
     host_that_restart.clear();
@@ -543,15 +531,15 @@ void SIMIX_run()
     /* Clean processes to destroy */
     SIMIX_process_empty_trash();
 
-    XBT_DEBUG("### time %f, empty %d", time, xbt_dynar_is_empty(simix_global->process_to_run));
+    XBT_DEBUG("### time %f, #processes %zu, #to_run %zu", time, simix_global->process_list.size(),
+              simix_global->process_to_run.size());
 
-    if (xbt_dynar_is_empty(simix_global->process_to_run) &&
-        xbt_swag_size(simix_global->process_list) != 0)
-    simgrid::simix::onDeadlock();
+    if (simix_global->process_to_run.empty() && not simix_global->process_list.empty())
+      simgrid::simix::onDeadlock();
 
-  } while (time > -1.0 || !xbt_dynar_is_empty(simix_global->process_to_run));
+  } while (time > -1.0 || not simix_global->process_to_run.empty());
 
-  if (xbt_swag_size(simix_global->process_list) != 0) {
+  if (simix_global->process_list.size() != 0) {
 
     TRACE_end();
 
@@ -559,6 +547,7 @@ void SIMIX_run()
     SIMIX_display_process_status();
     xbt_abort();
   }
+  simgrid::s4u::onSimulationEnd();
 }
 
 /**
@@ -566,13 +555,13 @@ void SIMIX_run()
  *
  * Set the date to execute the function on the surf.
  *   \param date Date to execute function
- *   \param function Function to be executed
+ *   \param callback Function to be executed
  *   \param arg Parameters of the function
  *
  */
 smx_timer_t SIMIX_timer_set(double date, void (*callback)(void*), void *arg)
 {
-  smx_timer_t timer = new s_smx_timer_t(date, [=](){ callback(arg); });
+  smx_timer_t timer = new s_smx_timer_t(date, [callback, arg]() { callback(arg); });
   xbt_heap_push(simix_timers, timer, date);
   return timer;
 }
@@ -586,12 +575,12 @@ smx_timer_t SIMIX_timer_set(double date, simgrid::xbt::Task<void()> callback)
 
 /** @brief cancels a timer that was added earlier */
 void SIMIX_timer_remove(smx_timer_t timer) {
-  xbt_heap_rm_elm(simix_timers, timer, timer->date);
+  delete static_cast<smx_timer_t>(xbt_heap_rm_elm(simix_timers, timer, timer->getDate()));
 }
 
 /** @brief Returns the date at which the timer will trigger (or 0 if nullptr timer) */
 double SIMIX_timer_get_date(smx_timer_t timer) {
-  return timer?timer->date:0;
+  return timer ? timer->getDate() : 0;
 }
 
 /**
@@ -635,69 +624,45 @@ void SIMIX_function_register_process_cleanup(void_pfn_smxprocess_t function)
 
 void SIMIX_display_process_status()
 {
-  if (simix_global->process_list == nullptr) {
-    return;
-  }
-
-  smx_actor_t process = nullptr;
-  int nbprocess = xbt_swag_size(simix_global->process_list);
+  int nbprocess = simix_global->process_list.size();
 
   XBT_INFO("%d processes are still running, waiting for something.", nbprocess);
   /*  List the process and their state */
   XBT_INFO("Legend of the following listing: \"Process <pid> (<name>@<host>): <status>\"");
-  xbt_swag_foreach(process, simix_global->process_list) {
+  for (auto const& kv : simix_global->process_list) {
+    smx_actor_t process = kv.second;
 
     if (process->waiting_synchro) {
 
       const char* synchro_description = "unknown";
 
-      if (dynamic_cast<simgrid::kernel::activity::Exec*>(process->waiting_synchro) != nullptr)
+      if (boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(process->waiting_synchro) != nullptr)
         synchro_description = "execution";
 
-      if (dynamic_cast<simgrid::kernel::activity::Comm*>(process->waiting_synchro) != nullptr)
+      if (boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(process->waiting_synchro) != nullptr)
         synchro_description = "communication";
 
-      if (dynamic_cast<simgrid::kernel::activity::Sleep*>(process->waiting_synchro) != nullptr)
+      if (boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(process->waiting_synchro) != nullptr)
         synchro_description = "sleeping";
 
-      if (dynamic_cast<simgrid::kernel::activity::Raw*>(process->waiting_synchro) != nullptr)
+      if (boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(process->waiting_synchro) != nullptr)
         synchro_description = "synchronization";
 
-      if (dynamic_cast<simgrid::kernel::activity::Io*>(process->waiting_synchro) != nullptr)
+      if (boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(process->waiting_synchro) != nullptr)
         synchro_description = "I/O";
 
-
-      /*
-        switch (process->waiting_synchro->type) {
-      case SIMIX_SYNC_PARALLEL_EXECUTE:
-        synchro_description = "parallel execution";
-        break;
-
-      case SIMIX_SYNC_JOIN:
-        synchro_description = "joining";
-        break;
-*/
-
-      XBT_INFO("Process %lu (%s@%s): waiting for %s synchro %p (%s) in state %d to finish",
-          process->pid, process->name.c_str(), sg_host_get_name(process->host),
-          synchro_description, process->waiting_synchro,
-          process->waiting_synchro->name.c_str(), (int)process->waiting_synchro->state);
+      XBT_INFO("Process %lu (%s@%s): waiting for %s synchro %p (%s) in state %d to finish", process->pid,
+               process->cname(), process->host->getCname(), synchro_description, process->waiting_synchro.get(),
+               process->waiting_synchro->name.c_str(), (int)process->waiting_synchro->state);
     }
     else {
-      XBT_INFO("Process %lu (%s@%s)", process->pid, process->name.c_str(), sg_host_get_name(process->host));
+      XBT_INFO("Process %lu (%s@%s)", process->pid, process->cname(), process->host->getCname());
     }
   }
 }
 
-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 static_cast<xbt_dict_t>(xbt_lib_get_or_null(as_router_lib, name, ROUTING_PROP_ASR_LEVEL));
-}
-
 int SIMIX_is_maestro()
 {
-  return simix_global==nullptr /*SimDag*/|| SIMIX_process_self() == simix_global->maestro_process;
+  smx_actor_t self = SIMIX_process_self();
+  return simix_global == nullptr /*SimDag*/ || self == nullptr || self == simix_global->maestro_process;
 }