Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
MC protocol: rename INITIAL_ADDRESSES msg to NEED_MEMINFO
[simgrid.git] / src / mc / remote / AppSide.cpp
index a32ece1..8813a4f 100644 (file)
@@ -11,7 +11,7 @@
 #include "src/kernel/actor/SimcallObserver.hpp"
 #include "src/mc/mc_base.hpp"
 #include "src/mc/mc_config.hpp"
-#include "src/mc/remote/RemoteProcess.hpp"
+#include "src/mc/sosp/RemoteProcessMemory.hpp"
 #if HAVE_SMPI
 #include "src/smpi/include/private.hpp"
 #endif
@@ -23,7 +23,6 @@
 #include <cerrno>
 #include <cstdio> // setvbuf
 #include <cstdlib>
-#include <cstring>
 #include <memory>
 #include <numeric>
 #include <sys/ptrace.h>
@@ -46,7 +45,7 @@ AppSide* AppSide::initialize()
   if (instance_)
     return instance_.get();
 
-  simgrid::mc::cfg_do_model_check = true;
+  simgrid::mc::model_checking_mode = ModelCheckingMode::APP_SIDE;
 
   setvbuf(stdout, nullptr, _IOLBF, 0);
 
@@ -65,20 +64,19 @@ AppSide* AppSide::initialize()
   instance_ = std::make_unique<simgrid::mc::AppSide>(fd);
 
   // Wait for the model-checker:
-  errno = 0;
+  if (getenv("MC_NEED_PTRACE") != nullptr) {
+    errno = 0;
 #if defined __linux__
-  ptrace(PTRACE_TRACEME, 0, nullptr, nullptr);
+    ptrace(PTRACE_TRACEME, 0, nullptr, nullptr);
 #elif defined BSD
-  ptrace(PT_TRACE_ME, 0, nullptr, 0);
+    ptrace(PT_TRACE_ME, 0, nullptr, 0);
 #else
-#error "no ptrace equivalent coded for this platform"
+    xbt_die("no ptrace equivalent coded for this platform, please don't use the liveness checker here.");
 #endif
-  xbt_assert(errno == 0 && raise(SIGSTOP) == 0, "Could not wait for the model-checker (errno = %d: %s)", errno,
-             strerror(errno));
 
-  s_mc_message_initial_addresses_t message{MessageType::INITIAL_ADDRESSES, mmalloc_get_current_heap(),
-                                           kernel::actor::ActorImpl::get_maxpid_addr()};
-  xbt_assert(instance_->channel_.send(message) == 0, "Could not send the initial message with addresses.");
+    xbt_assert(errno == 0 && raise(SIGSTOP) == 0, "Could not wait for the model-checker (errno = %d: %s)", errno,
+               strerror(errno));
+  }
 
   instance_->handle_messages();
   return instance_.get();
@@ -99,7 +97,9 @@ void AppSide::handle_deadlock_check(const s_mc_message_t*) const
     engine->display_all_actor_status();
   }
   // Send result:
-  s_mc_message_int_t answer{MessageType::DEADLOCK_CHECK_REPLY, deadlock};
+  s_mc_message_int_t answer = {};
+  answer.type  = MessageType::DEADLOCK_CHECK_REPLY;
+  answer.value = deadlock;
   xbt_assert(channel_.send(answer) == 0, "Could not send response");
 }
 void AppSide::handle_simcall_execute(const s_mc_message_simcall_execute_t* message) const
@@ -113,8 +113,7 @@ void AppSide::handle_simcall_execute(const s_mc_message_simcall_execute_t* messa
   xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send MESSAGE_WAITING to model-checker");
 
   // Finish the RPC from the server: return a serialized observer, to build a Transition on Checker side
-  s_mc_message_simcall_execute_answer_t answer;
-  memset(&answer, 0, sizeof(answer));
+  s_mc_message_simcall_execute_answer_t answer = {};
   answer.type = MessageType::SIMCALL_EXECUTE_ANSWER;
   std::stringstream stream;
   if (actor->simcall_.observer_ != nullptr) {
@@ -151,28 +150,27 @@ void AppSide::handle_finalize(const s_mc_message_int_t* msg) const
   if (terminate_asap)
     ::_Exit(0);
 }
+void AppSide::handle_need_meminfo()
+{
+  this->need_memory_info_                  = true;
+  s_mc_message_need_meminfo_reply_t answer = {};
+  answer.type                              = MessageType::NEED_MEMINFO_REPLY;
+  answer.mmalloc_default_mdp               = mmalloc_get_current_heap();
+  xbt_assert(channel_.send(answer) == 0, "Could not send response to the request for meminfo.");
+}
 void AppSide::handle_actors_status() const
 {
   auto const& actor_list = kernel::EngineImpl::get_instance()->get_actor_list();
-  const int num_actors   = actor_list.size();
-  XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %d actors to go.", num_actors);
-
-  std::vector<s_mc_message_actors_status_one_t> status(num_actors);
-  int i                 = 0;
-  int total_transitions = 0;
-
-  for (auto const& [aid, actor] : actor_list) {
-    status[i].aid            = aid;
-    status[i].enabled        = mc::actor_is_enabled(actor);
-    status[i].max_considered = actor->simcall_.observer_->get_max_consider();
-    status[i].n_transitions  = mc::actor_is_enabled(actor) ? status[i].max_considered : 0;
-    total_transitions += status[i].n_transitions;
-    i++;
-  }
+  XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %zu actors to go.", actor_list.size());
+
+  std::vector<s_mc_message_actors_status_one_t> status;
+  for (auto const& [aid, actor] : actor_list)
+    status.emplace_back(s_mc_message_actors_status_one_t{aid, mc::actor_is_enabled(actor),
+                                                         actor->simcall_.observer_->get_max_consider()});
 
-  struct s_mc_message_actors_status_answer_t answer {
-    MessageType::ACTORS_STATUS_REPLY, num_actors, total_transitions
-  };
+  struct s_mc_message_actors_status_answer_t answer = {};
+  answer.type             = MessageType::ACTORS_STATUS_REPLY;
+  answer.count            = static_cast<int>(status.size());
 
   xbt_assert(channel_.send(answer) == 0, "Could not send ACTORS_STATUS_REPLY msg");
   if (answer.count > 0) {
@@ -181,52 +179,53 @@ void AppSide::handle_actors_status() const
   }
 
   // Serialize each transition to describe what each actor is doing
-  if (total_transitions > 0) {
-    std::vector<s_mc_message_simcall_probe_one_t> probes(total_transitions);
-    auto probes_iter = probes.begin();
-
-    for (const auto& actor_status : status) {
-      if (not actor_status.enabled)
-        continue;
-
-      const auto& actor        = actor_list.at(actor_status.aid);
-      const int max_considered = actor_status.max_considered;
-
-      for (int times_considered = 0; times_considered < max_considered; times_considered++, probes_iter++) {
-        std::stringstream stream;
-        s_mc_message_simcall_probe_one_t& probe = *probes_iter;
-
-        if (actor->simcall_.observer_ != nullptr) {
-          actor->simcall_.observer_->prepare(times_considered);
-          actor->simcall_.observer_->serialize(stream);
-        } else {
-          stream << (short)mc::Transition::Type::UNKNOWN;
-        }
-
-        std::string str = stream.str();
-        xbt_assert(str.size() + 1 <= probe.buffer.size(),
-                   "The serialized transition is too large for the buffer. Please fix the code.");
-        strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
-        probe.buffer.back() = '\0';
+  XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
+  for (const auto& actor_status : status) {
+    if (not actor_status.enabled)
+      continue;
+
+    const auto& actor        = actor_list.at(actor_status.aid);
+    const int max_considered = actor_status.max_considered;
+
+    for (int times_considered = 0; times_considered < max_considered; times_considered++) {
+      std::stringstream stream;
+      s_mc_message_simcall_probe_one_t probe;
+
+      if (actor->simcall_.observer_ != nullptr) {
+        actor->simcall_.observer_->prepare(times_considered);
+        actor->simcall_.observer_->serialize(stream);
+      } else {
+        stream << (short)mc::Transition::Type::UNKNOWN;
       }
-      // NOTE: We do NOT need to reset `times_considered` for each actor's
-      // simcall observer here to the "original" value (i.e. the value BEFORE
-      // multiple prepare() calls were made for serialization purposes) since
-      // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
-      // the transition before execution.
-    }
-    XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
 
-    for (const auto& probe : probes)
+      std::string str = stream.str();
+      xbt_assert(str.size() + 1 <= probe.buffer.size(),
+                 "The serialized transition is too large for the buffer. Please fix the code.");
+      strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
+      probe.buffer.back() = '\0';
+
       xbt_assert(channel_.send(probe) == 0, "Could not send ACTOR_TRANSITION_PROBE payload");
+    }
+    // NOTE: We do NOT need to reset `times_considered` for each actor's
+    // simcall observer here to the "original" value (i.e. the value BEFORE
+    // multiple prepare() calls were made for serialization purposes) since
+    // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
+    // the transition before execution.
   }
 }
+void AppSide::handle_actors_maxpid() const
+{
+  s_mc_message_int_t answer = {};
+  answer.type               = MessageType::ACTORS_MAXPID_REPLY;
+  answer.value              = kernel::actor::ActorImpl::get_maxpid();
+  xbt_assert(channel_.send(answer) == 0, "Could not send response");
+}
 
 #define assert_msg_size(_name_, _type_)                                                                                \
   xbt_assert(received_size == sizeof(_type_), "Unexpected size for " _name_ " (%zd != %zu)", received_size,            \
              sizeof(_type_))
 
-void AppSide::handle_messages() const
+void AppSide::handle_messages()
 {
   while (true) { // Until we get a CONTINUE message
     XBT_DEBUG("Waiting messages from model-checker");
@@ -234,6 +233,10 @@ void AppSide::handle_messages() const
     std::array<char, MC_MESSAGE_LENGTH> message_buffer;
     ssize_t received_size = channel_.receive(message_buffer.data(), message_buffer.size());
 
+    if (received_size == 0) {
+      XBT_DEBUG("Socket closed on the Checker side, bailing out.");
+      ::_Exit(0); // Nobody's listening to that process anymore => exit as quickly as possible.
+    }
     xbt_assert(received_size >= 0, "Could not receive commands from the model-checker");
     xbt_assert(static_cast<size_t>(received_size) >= sizeof(s_mc_message_t), "Cannot handle short message (size=%zd)",
                received_size);
@@ -259,11 +262,21 @@ void AppSide::handle_messages() const
         handle_finalize((s_mc_message_int_t*)message_buffer.data());
         break;
 
+      case MessageType::NEED_MEMINFO:
+        assert_msg_size("NEED_MEMINFO", s_mc_message_t);
+        handle_need_meminfo();
+        break;
+
       case MessageType::ACTORS_STATUS:
         assert_msg_size("ACTORS_STATUS", s_mc_message_t);
         handle_actors_status();
         break;
 
+      case MessageType::ACTORS_MAXPID:
+        assert_msg_size("ACTORS_MAXPID", s_mc_message_t);
+        handle_actors_maxpid();
+        break;
+
       default:
         xbt_die("Received unexpected message %s (%i)", to_c_str(message->type), static_cast<int>(message->type));
         break;
@@ -271,7 +284,7 @@ void AppSide::handle_messages() const
   }
 }
 
-void AppSide::main_loop() const
+void AppSide::main_loop()
 {
   simgrid::mc::processes_time.resize(simgrid::kernel::actor::ActorImpl::get_maxpid());
   MC_ignore_heap(simgrid::mc::processes_time.data(),
@@ -287,7 +300,7 @@ void AppSide::main_loop() const
   }
 }
 
-void AppSide::report_assertion_failure() const
+void AppSide::report_assertion_failure()
 {
   xbt_assert(channel_.send(MessageType::ASSERTION_FAILED) == 0, "Could not send assertion to model-checker");
   this->handle_messages();
@@ -295,10 +308,10 @@ void AppSide::report_assertion_failure() const
 
 void AppSide::ignore_memory(void* addr, std::size_t size) const
 {
-  if (not MC_is_active())
+  if (not MC_is_active() || not need_memory_info_)
     return;
 
-  s_mc_message_ignore_memory_t message;
+  s_mc_message_ignore_memory_t message = {};
   message.type = MessageType::IGNORE_MEMORY;
   message.addr = (std::uintptr_t)addr;
   message.size = size;
@@ -307,12 +320,12 @@ void AppSide::ignore_memory(void* addr, std::size_t size) const
 
 void AppSide::ignore_heap(void* address, std::size_t size) const
 {
-  if (not MC_is_active())
+  if (not MC_is_active() || not need_memory_info_)
     return;
 
   const s_xbt_mheap_t* heap = mmalloc_get_current_heap();
 
-  s_mc_message_ignore_heap_t message;
+  s_mc_message_ignore_heap_t message = {};
   message.type    = MessageType::IGNORE_HEAP;
   message.address = address;
   message.size    = size;
@@ -330,10 +343,10 @@ void AppSide::ignore_heap(void* address, std::size_t size) const
 
 void AppSide::unignore_heap(void* address, std::size_t size) const
 {
-  if (not MC_is_active())
+  if (not MC_is_active() || not need_memory_info_)
     return;
 
-  s_mc_message_ignore_memory_t message;
+  s_mc_message_ignore_memory_t message = {};
   message.type = MessageType::UNIGNORE_HEAP;
   message.addr = (std::uintptr_t)address;
   message.size = size;
@@ -342,11 +355,12 @@ void AppSide::unignore_heap(void* address, std::size_t size) const
 
 void AppSide::declare_symbol(const char* name, int* value) const
 {
-  if (not MC_is_active())
+  if (not MC_is_active() || not need_memory_info_) {
+    XBT_CRITICAL("Ignore AppSide::declare_symbol(%s)", name);
     return;
+  }
 
-  s_mc_message_register_symbol_t message;
-  memset(&message, 0, sizeof(message));
+  s_mc_message_register_symbol_t message = {};
   message.type = MessageType::REGISTER_SYMBOL;
   xbt_assert(strlen(name) + 1 <= message.name.size(), "Symbol is too long");
   strncpy(message.name.data(), name, message.name.size() - 1);
@@ -363,19 +377,18 @@ void AppSide::declare_symbol(const char* name, int* value) const
  */
 void AppSide::declare_stack(void* stack, size_t size, ucontext_t* context) const
 {
-  if (not MC_is_active())
+  if (not MC_is_active() || not need_memory_info_)
     return;
 
   const s_xbt_mheap_t* heap = mmalloc_get_current_heap();
 
-  s_stack_region_t region;
-  memset(&region, 0, sizeof(region));
+  s_stack_region_t region = {};
   region.address = stack;
   region.context = context;
   region.size    = size;
   region.block   = ((char*)stack - (char*)heap->heapbase) / BLOCKSIZE + 1;
 
-  s_mc_message_stack_region_t message;
+  s_mc_message_stack_region_t message = {};
   message.type         = MessageType::STACK_REGION;
   message.stack_region = region;
   xbt_assert(channel_.send(message) == 0, "Could not send STACK_REGION to model-checker");