[mc] Fix NT_GNU_BUILD_ID support

[simgrid.git] / src / mc / mc_dwarf.cpp

diff --git a/src/mc/mc_dwarf.cpp b/src/mc/mc_dwarf.cpp
index 64b71b0..3d08678 100644 (file)
--- a/src/mc/mc_dwarf.cpp
+++ b/src/mc/mc_dwarf.cpp
@@ -38,7 +38,7 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_dwarf, mc, "DWARF processing");
  *
  *  The default for a given language is defined in the DWARF spec.
  *
  *
  *  The default for a given language is defined in the DWARF spec.
  *

- *  \param language consant as defined by the DWARf spec
+ *  \param language constant as defined by the DWARf spec

  */
 static uint64_t MC_dwarf_default_lower_bound(int lang);
 
  */
 static uint64_t MC_dwarf_default_lower_bound(int lang);
 


@@ -67,7 +67,7 @@ static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit);
  *  \param info the resulting object fot the library/binary file (output)
  *  \param die  the current DIE
  *  \param unit the DIE of the compile unit of the current DIE
  *  \param info the resulting object fot the library/binary file (output)
  *  \param die  the current DIE
  *  \param unit the DIE of the compile unit of the current DIE

- *  \param frame containg frame if any
+ *  \param frame containing frame if any

  */
 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
                                 Dwarf_Die * unit, simgrid::mc::Frame* frame,
  */
 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
                                 Dwarf_Die * unit, simgrid::mc::Frame* frame,


@@ -79,12 +79,12 @@ static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf
                                      Dwarf_Die * unit, simgrid::mc::Frame* frame,
                                      const char *ns);
 
                                      Dwarf_Die * unit, simgrid::mc::Frame* frame,
                                      const char *ns);
 

-/** \brief Calls MC_dwarf_handle_die on all childrend of the given die
+/** \brief Calls MC_dwarf_handle_die on all children of the given die

  *
  *  \param info the resulting object fot the library/binary file (output)
  *  \param die  the current DIE
  *  \param unit the DIE of the compile unit of the current DIE
  *
  *  \param info the resulting object fot the library/binary file (output)
  *  \param die  the current DIE
  *  \param unit the DIE of the compile unit of the current DIE

- *  \param frame containg frame if any
+ *  \param frame containing frame if any

  */
 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
                                      Dwarf_Die * unit, simgrid::mc::Frame* frame,
  */
 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
                                      Dwarf_Die * unit, simgrid::mc::Frame* frame,


@@ -95,7 +95,7 @@ static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf
  *  \param info the resulting object fot the library/binary file (output)
  *  \param die  the current DIE
  *  \param unit the DIE of the compile unit of the current DIE
  *  \param info the resulting object fot the library/binary file (output)
  *  \param die  the current DIE
  *  \param unit the DIE of the compile unit of the current DIE

- *  \param frame containg frame if any
+ *  \param frame containing frame if any

  */
 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
                                          Dwarf_Die * unit, simgrid::mc::Frame* frame,
  */
 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
                                          Dwarf_Die * unit, simgrid::mc::Frame* frame,


@@ -585,7 +585,7 @@ static void MC_dwarf_add_members(simgrid::mc::ObjectInformation* info, Dwarf_Die
       if (boost::algorithm::starts_with(member.name, "__vptr$") ||
         boost::algorithm::starts_with(member.name, "__vptr."))
         member.flags |= simgrid::mc::Member::VIRTUAL_POINTER_FLAG;
       if (boost::algorithm::starts_with(member.name, "__vptr$") ||
         boost::algorithm::starts_with(member.name, "__vptr."))
         member.flags |= simgrid::mc::Member::VIRTUAL_POINTER_FLAG;

-      // A cleaner stolution would be to check against the type:
+      // A cleaner solution would be to check against the type:

       // ---
       // tag: DW_TAG_member
       // name: "_vptr$Foo"
       // ---
       // tag: DW_TAG_member
       // name: "_vptr$Foo"


@@ -1044,6 +1044,9 @@ void read_dwarf_info(simgrid::mc::ObjectInformation* info, Dwarf* dwarf)
 static
 std::vector<char> get_build_id(Elf* elf)
 {
 static
 std::vector<char> get_build_id(Elf* elf)
 {

+  // Summary: the GNU build ID is stored in a ("GNU, NT_GNU_BUILD_ID) note
+  // found in a PT_NOTE entry in the program header table.
+

   size_t phnum;
   if (elf_getphdrnum (elf, &phnum) != 0)
     xbt_die("Could not read program headers");
   size_t phnum;
   if (elf_getphdrnum (elf, &phnum) != 0)
     xbt_die("Could not read program headers");


@@ -1059,22 +1062,21 @@ std::vector<char> get_build_id(Elf* elf)
 
     // Iterate over the notes and find the NT_GNU_BUILD_ID one:
     size_t pos = 0;
 
     // Iterate over the notes and find the NT_GNU_BUILD_ID one:
     size_t pos = 0;

-    while (1) {
+    while (pos < data->d_size) {

       GElf_Nhdr nhdr;
       GElf_Nhdr nhdr;

+      // Location of the name within Elf_Data:

       size_t name_pos;
       size_t desc_pos;
       pos = gelf_getnote(data, pos, &nhdr, &name_pos, &desc_pos);
       size_t name_pos;
       size_t desc_pos;
       pos = gelf_getnote(data, pos, &nhdr, &name_pos, &desc_pos);

-      // A note is identified by a name "GNU" and a integer type within
-      // the namespace defined by this name (here NT_GNU_BUILD_ID):
+      // A build ID note is identified by the pair ("GNU", NT_GNU_BUILD_ID)
+      // (a namespace and a type within this namespace):

       if (nhdr.n_type == NT_GNU_BUILD_ID
           && nhdr.n_namesz == sizeof("GNU")
           && memcmp((char*) data->d_buf + name_pos, "GNU", sizeof("GNU")) == 0) {
       if (nhdr.n_type == NT_GNU_BUILD_ID
           && nhdr.n_namesz == sizeof("GNU")
           && memcmp((char*) data->d_buf + name_pos, "GNU", sizeof("GNU")) == 0) {

-
-        // Found the NT_GNU_BUILD_ID note:
+        XBT_DEBUG("Found GNU/NT_GNU_BUILD_ID note");

         char* start = (char*) data->d_buf + desc_pos;
         char* end = (char*) start + nhdr.n_descsz;
         return std::vector<char>(start, end);
         char* start = (char*) data->d_buf + desc_pos;
         char* end = (char*) start + nhdr.n_descsz;
         return std::vector<char>(start, end);

-

       }
     }
 
       }
     }
 


@@ -1129,18 +1131,25 @@ const char* debug_paths[] = {
  *  This is one of the mechanisms used for
  *  [separate debug files](https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html).
  */
  *  This is one of the mechanisms used for
  *  [separate debug files](https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html).
  */

+// Example:
+// /usr/lib/debug/.build-id/0b/dc77f1c29aea2b14ff5acd9a19ab3175ffdeae.debug

 static
 std::string find_by_build_id(std::vector<char> id)
 {
   std::string filename;
 static
 std::string find_by_build_id(std::vector<char> id)
 {
   std::string filename;

+  std::string hex = to_hex(id);

   for (const char* debug_path : debug_paths) {
   for (const char* debug_path : debug_paths) {

-    filename = debug_path;
-    filename += ".build-id/" + to_hex(id.data(), 1) + '/'
+    // Example:
+    filename = std::string(debug_path) + ".build-id/"
+      + to_hex(id.data(), 1) + '/'

       + to_hex(id.data() + 1, id.size() - 1) + ".debug";
     XBT_DEBUG("Checking debug file: %s", filename.c_str());
       + to_hex(id.data() + 1, id.size() - 1) + ".debug";
     XBT_DEBUG("Checking debug file: %s", filename.c_str());

-    if (access(filename.c_str(), F_OK) == 0)
+    if (access(filename.c_str(), F_OK) == 0) {
+      XBT_DEBUG("Found debug file: %s\n", hex.c_str());

       return filename;
       return filename;

+    }

   }
   }

+  XBT_DEBUG("Not debuf info found for build ID %s\n", hex.data());

   return std::string();
 }
 
   return std::string();
 }
 


@@ -1150,7 +1159,7 @@ std::string find_by_build_id(std::vector<char> id)
  *  lists of types, variables, functions.
  */
 static
  *  lists of types, variables, functions.
  */
 static

-void MC_dwarf_get_variables(simgrid::mc::ObjectInformation* info)
+void MC_load_dwarf(simgrid::mc::ObjectInformation* info)

 {
   if (elf_version(EV_CURRENT) == EV_NONE)
     xbt_die("libelf initialization error");
 {
   if (elf_version(EV_CURRENT) == EV_NONE)
     xbt_die("libelf initialization error");


@@ -1161,12 +1170,12 @@ void MC_dwarf_get_variables(simgrid::mc::ObjectInformation* info)
     xbt_die("Could not open file %s", info->file_name.c_str());
   Elf* elf = elf_begin(fd, ELF_C_READ, nullptr);
   if (elf == nullptr)
     xbt_die("Could not open file %s", info->file_name.c_str());
   Elf* elf = elf_begin(fd, ELF_C_READ, nullptr);
   if (elf == nullptr)

-    xbt_die("Not an ELF file 1");
+    xbt_die("Not an ELF file");

   Elf_Kind kind = elf_kind(elf);
   if (kind != ELF_K_ELF)
   Elf_Kind kind = elf_kind(elf);
   if (kind != ELF_K_ELF)

-    xbt_die("Not an ELF file 2");
+    xbt_die("Not an ELF file");

 
 

-  // Remember if this is a `ET_EXEC` (fixed location) or `ET_DYN` (relocatable):
+  // Remember if this is a `ET_EXEC` (fixed location) or `ET_DYN`:

   Elf64_Half type = get_type(elf);
   if (type == ET_EXEC)
     info->flags |= simgrid::mc::ObjectInformation::Executable;
   Elf64_Half type = get_type(elf);
   if (type == ET_EXEC)
     info->flags |= simgrid::mc::ObjectInformation::Executable;


@@ -1182,8 +1191,15 @@ void MC_dwarf_get_variables(simgrid::mc::ObjectInformation* info)
   }
   dwarf_end(dwarf);
 
   }
   dwarf_end(dwarf);
 

-  // If there was no DWARF in the file, try to find it in a separate file
-  // with NT_GNU_BUILD_ID:
+  // If there was no DWARF in the file, try to find it in a separate file.
+  // Different methods might be used to store the DWARF informations:
+  //  * GNU NT_GNU_BUILD_ID;
+  //  * .gnu_debuglink.
+  // See https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html
+  // for reference of what we are doing.
+
+  // Try with NT_GNU_BUILD_ID: we find the build ID in the ELF file and then
+  // use this ID to find the file in some known locations in the filesystem.

   std::vector<char> build_id = get_build_id(elf);
   if (!build_id.empty()) {
     elf_end(elf);
   std::vector<char> build_id = get_build_id(elf);
   if (!build_id.empty()) {
     elf_end(elf);


@@ -1193,8 +1209,7 @@ void MC_dwarf_get_variables(simgrid::mc::ObjectInformation* info)
     std::string debug_file = find_by_build_id(build_id);
     if (debug_file.empty()) {
       std::string hex = to_hex(build_id);
     std::string debug_file = find_by_build_id(build_id);
     if (debug_file.empty()) {
       std::string hex = to_hex(build_id);

-      xbt_die(
-        "Missing debug info for %s with build-id %s\n"
+      xbt_die("Missing debug info for %s with build-id %s\n"

         "You might want to install the suitable debugging package.\n",
         info->file_name.c_str(), hex.c_str());
     }
         "You might want to install the suitable debugging package.\n",
         info->file_name.c_str(), hex.c_str());
     }


@@ -1215,9 +1230,10 @@ void MC_dwarf_get_variables(simgrid::mc::ObjectInformation* info)
     return;
   }
 
     return;
   }
 

-  // TODO, try to find DWARF info using debug-link.
-  // Is this method really used anywhere?
+  // TODO, try to find DWARF info using .gnu_debuglink.

 
 

+  elf_end(elf);
+  close(fd);

   xbt_die("Debugging information not found for %s\n"
     "Try recompiling with -g\n",
     info->file_name.c_str());
   xbt_die("Debugging information not found for %s\n"
     "Try recompiling with -g\n",
     info->file_name.c_str());


@@ -1343,7 +1359,7 @@ std::shared_ptr<simgrid::mc::ObjectInformation> createObjectInformation(
     std::make_shared<simgrid::mc::ObjectInformation>();
   result->file_name = name;
   simgrid::mc::find_object_address(maps, result.get());
     std::make_shared<simgrid::mc::ObjectInformation>();
   result->file_name = name;
   simgrid::mc::find_object_address(maps, result.get());

-  MC_dwarf_get_variables(result.get());
+  MC_load_dwarf(result.get());

   MC_post_process_variables(result.get());
   MC_post_process_types(result.get());
   for (auto& entry : result.get()->subprograms)
   MC_post_process_variables(result.get());
   MC_post_process_types(result.get());
   for (auto& entry : result.get()->subprograms)


@@ -1393,7 +1409,7 @@ namespace dwarf {
  *
  *  DWARF and libunwind does not use the same convention for numbering the
  *  registers on some architectures. The function makes the necessary
  *
  *  DWARF and libunwind does not use the same convention for numbering the
  *  registers on some architectures. The function makes the necessary

- *  convertion.
+ *  conversion.

  */
 int dwarf_register_to_libunwind(int dwarf_register)
 {
  */
 int dwarf_register_to_libunwind(int dwarf_register)
 {


@@ -1401,7 +1417,7 @@ int dwarf_register_to_libunwind(int dwarf_register)
   // It seems for this arch, DWARF and libunwind agree in the numbering:
   return dwarf_register;
 #elif defined(__i386__)
   // It seems for this arch, DWARF and libunwind agree in the numbering:
   return dwarf_register;
 #elif defined(__i386__)

-  // Could't find the authoritative source of information for this.
+  // Couldn't find the authoritative source of information for this.

   // This is inspired from http://source.winehq.org/source/dlls/dbghelp/cpu_i386.c#L517.
   switch (dwarf_register) {
   case 0:
   // This is inspired from http://source.winehq.org/source/dlls/dbghelp/cpu_i386.c#L517.
   switch (dwarf_register) {
   case 0: