1 /* Copyright (c) 2008-2015. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
16 #define DW_LANG_Objc DW_LANG_ObjC /* fix spelling error in older dwarf.h */
18 #include <elfutils/libdw.h>
20 #include <simgrid_config.h>
21 #include "src/simgrid/util.hpp"
23 #include <xbt/sysdep.h>
25 #include "src/mc/mc_private.h"
26 #include "src/mc/mc_dwarf.hpp"
28 #include "src/mc/Process.hpp"
29 #include "src/mc/ObjectInformation.hpp"
30 #include "src/mc/Variable.hpp"
32 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_dwarf, mc, "DWARF processing");
34 /** \brief The default DW_TAG_lower_bound for a given DW_AT_language.
36 * The default for a given language is defined in the DWARF spec.
38 * \param language consant as defined by the DWARf spec
40 static uint64_t MC_dwarf_default_lower_bound(int lang);
42 /** \brief Computes the the element_count of a DW_TAG_enumeration_type DIE
44 * This is the number of elements in a given array dimension.
46 * A reference of the compilation unit (DW_TAG_compile_unit) is
47 * needed because the default lower bound (when there is no DW_AT_lower_bound)
48 * depends of the language of the compilation unit (DW_AT_language).
50 * \param die DIE for the DW_TAG_enumeration_type or DW_TAG_subrange_type
51 * \param unit DIE of the DW_TAG_compile_unit
53 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die * die,
56 /** \brief Computes the number of elements of a given DW_TAG_array_type.
58 * \param die DIE for the DW_TAG_array_type
60 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit);
62 /** \brief Process a DIE
64 * \param info the resulting object fot the library/binary file (output)
65 * \param die the current DIE
66 * \param unit the DIE of the compile unit of the current DIE
67 * \param frame containg frame if any
69 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
70 Dwarf_Die * unit, simgrid::mc::Frame* frame,
73 /** \brief Process a type DIE
75 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
76 Dwarf_Die * unit, simgrid::mc::Frame* frame,
79 /** \brief Calls MC_dwarf_handle_die on all childrend of the given die
81 * \param info the resulting object fot the library/binary file (output)
82 * \param die the current DIE
83 * \param unit the DIE of the compile unit of the current DIE
84 * \param frame containg frame if any
86 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
87 Dwarf_Die * unit, simgrid::mc::Frame* frame,
90 /** \brief Handle a variable (DW_TAG_variable or other)
92 * \param info the resulting object fot the library/binary file (output)
93 * \param die the current DIE
94 * \param unit the DIE of the compile unit of the current DIE
95 * \param frame containg frame if any
97 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
98 Dwarf_Die * unit, simgrid::mc::Frame* frame,
101 /** \brief Get the DW_TAG_type of the DIE
104 * \return DW_TAG_type attribute as a new string (nullptr if none)
106 static std::uint64_t MC_dwarf_at_type(Dwarf_Die * die);
111 enum class TagClass {
120 /*** Class of forms defined in the DWARF standard */
121 enum class FormClass {
123 Address, // Location in the program's address space
124 Block, // Arbitrary block of bytes
127 Flag, // Boolean value
128 Reference, // Reference to another DIE
129 ExprLoc, // DWARF expression/location description
137 TagClass classify_tag(int tag)
141 case DW_TAG_array_type:
142 case DW_TAG_class_type:
143 case DW_TAG_enumeration_type:
145 case DW_TAG_pointer_type:
146 case DW_TAG_reference_type:
147 case DW_TAG_rvalue_reference_type:
148 case DW_TAG_string_type:
149 case DW_TAG_structure_type:
150 case DW_TAG_subroutine_type:
151 case DW_TAG_union_type:
152 case DW_TAG_ptr_to_member_type:
153 case DW_TAG_set_type:
154 case DW_TAG_subrange_type:
155 case DW_TAG_base_type:
156 case DW_TAG_const_type:
157 case DW_TAG_file_type:
158 case DW_TAG_packed_type:
159 case DW_TAG_volatile_type:
160 case DW_TAG_restrict_type:
161 case DW_TAG_interface_type:
162 case DW_TAG_unspecified_type:
163 case DW_TAG_shared_type:
164 return TagClass::Type;
166 case DW_TAG_subprogram:
167 return TagClass::Subprogram;
169 case DW_TAG_variable:
170 case DW_TAG_formal_parameter:
171 return TagClass::Variable;
173 case DW_TAG_lexical_block:
174 case DW_TAG_try_block:
175 case DW_TAG_catch_block:
176 case DW_TAG_inlined_subroutine:
177 case DW_TAG_with_stmt:
178 return TagClass::Scope;
180 case DW_TAG_namespace:
181 return TagClass::Namespace;
184 return TagClass::Unknown;
188 /** \brief Find the DWARF data class for a given DWARF data form
190 * This mapping is defined in the DWARF spec.
192 * \param form The form (values taken from the DWARF spec)
193 * \return An internal representation for the corresponding class
196 FormClass classify_form(int form)
200 return FormClass::Address;
205 return FormClass::Block;
212 return FormClass::Constant;
215 return FormClass::String;
216 case DW_FORM_ref_addr:
221 case DW_FORM_ref_udata:
222 return FormClass::Reference;
224 case DW_FORM_flag_present:
225 return FormClass::Flag;
226 case DW_FORM_exprloc:
227 return FormClass::ExprLoc;
231 return FormClass::Unknown;
235 /** \brief Get the name of the tag of a given DIE
238 * \return name of the tag of this DIE
241 const char *tagname(Dwarf_Die * die)
243 return simgrid::dwarf::tagname(dwarf_tag(die));
251 /** \brief Get an attribute of a given DIE as a string
254 * \param attribute attribute
255 * \return value of the given attribute of the given DIE
257 static const char *MC_dwarf_attr_integrate_string(Dwarf_Die * die,
260 Dwarf_Attribute attr;
261 if (!dwarf_attr_integrate(die, attribute, &attr))
264 return dwarf_formstring(&attr);
267 /** \brief Get the linkage name of a DIE.
269 * Use either DW_AT_linkage_name or DW_AT_MIPS_linkage_name.
270 * DW_AT_linkage_name is standardized since DWARF 4.
271 * Before this version of DWARF, the MIPS extensions
272 * DW_AT_MIPS_linkage_name is used (at least by GCC).
275 * \return linkage name of the given DIE (or nullptr)
277 static const char *MC_dwarf_at_linkage_name(Dwarf_Die * die)
279 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_linkage_name);
281 name = MC_dwarf_attr_integrate_string(die, DW_AT_MIPS_linkage_name);
285 static Dwarf_Off MC_dwarf_attr_dieoffset(Dwarf_Die * die, int attribute)
287 Dwarf_Attribute attr;
288 if (dwarf_hasattr_integrate(die, attribute) == 0)
290 dwarf_attr_integrate(die, attribute, &attr);
291 Dwarf_Die subtype_die;
292 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
293 xbt_die("Could not find DIE");
294 return dwarf_dieoffset(&subtype_die);
297 static Dwarf_Off MC_dwarf_attr_integrate_dieoffset(Dwarf_Die * die,
300 Dwarf_Attribute attr;
301 if (dwarf_hasattr_integrate(die, attribute) == 0)
303 dwarf_attr_integrate(die, DW_AT_type, &attr);
304 Dwarf_Die subtype_die;
305 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
306 xbt_die("Could not find DIE");
307 return dwarf_dieoffset(&subtype_die);
310 /** \brief Find the type/subtype (DW_AT_type) for a DIE
313 * \return DW_AT_type reference as a global offset in hexadecimal (or nullptr)
316 std::uint64_t MC_dwarf_at_type(Dwarf_Die * die)
318 return MC_dwarf_attr_integrate_dieoffset(die, DW_AT_type);
321 static uint64_t MC_dwarf_attr_integrate_addr(Dwarf_Die * die, int attribute)
323 Dwarf_Attribute attr;
324 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
327 if (dwarf_formaddr(&attr, &value) == 0)
328 return (uint64_t) value;
333 static uint64_t MC_dwarf_attr_integrate_uint(Dwarf_Die * die, int attribute,
334 uint64_t default_value)
336 Dwarf_Attribute attr;
337 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
338 return default_value;
340 return dwarf_formudata(dwarf_attr_integrate(die, attribute, &attr),
341 &value) == 0 ? (uint64_t) value : default_value;
344 static bool MC_dwarf_attr_flag(Dwarf_Die * die, int attribute, bool integrate)
346 Dwarf_Attribute attr;
347 if ((integrate ? dwarf_attr_integrate(die, attribute, &attr)
348 : dwarf_attr(die, attribute, &attr)) == 0)
352 if (dwarf_formflag(&attr, &result))
353 xbt_die("Unexpected form for attribute %s",
354 simgrid::dwarf::attrname(attribute));
358 /** \brief Find the default lower bound for a given language
360 * The default lower bound of an array (when DW_TAG_lower_bound
361 * is missing) depends on the language of the compilation unit.
363 * \param lang Language of the compilation unit (values defined in the DWARF spec)
364 * \return Default lower bound of an array in this compilation unit
366 static uint64_t MC_dwarf_default_lower_bound(int lang)
372 case DW_LANG_C_plus_plus:
376 case DW_LANG_ObjC_plus_plus:
382 case DW_LANG_Fortran77:
383 case DW_LANG_Fortran90:
384 case DW_LANG_Fortran95:
385 case DW_LANG_Modula2:
386 case DW_LANG_Pascal83:
388 case DW_LANG_Cobol74:
389 case DW_LANG_Cobol85:
392 xbt_die("No default DW_TAG_lower_bound for language %i and none given",
398 /** \brief Finds the number of elements in a DW_TAG_subrange_type or DW_TAG_enumeration_type DIE
401 * \param unit DIE of the compilation unit
402 * \return number of elements in the range
404 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die * die,
407 xbt_assert(dwarf_tag(die) == DW_TAG_enumeration_type
408 || dwarf_tag(die) == DW_TAG_subrange_type,
409 "MC_dwarf_subrange_element_count called with DIE of type %s",
410 simgrid::dwarf::tagname(die));
412 // Use DW_TAG_count if present:
413 if (dwarf_hasattr_integrate(die, DW_AT_count))
414 return MC_dwarf_attr_integrate_uint(die, DW_AT_count, 0);
415 // Otherwise compute DW_TAG_upper_bound-DW_TAG_lower_bound + 1:
417 if (!dwarf_hasattr_integrate(die, DW_AT_upper_bound))
418 // This is not really 0, but the code expects this (we do not know):
421 uint64_t upper_bound =
422 MC_dwarf_attr_integrate_uint(die, DW_AT_upper_bound, -1);
424 uint64_t lower_bound = 0;
425 if (dwarf_hasattr_integrate(die, DW_AT_lower_bound))
426 lower_bound = MC_dwarf_attr_integrate_uint(die, DW_AT_lower_bound, -1);
428 lower_bound = MC_dwarf_default_lower_bound(dwarf_srclang(unit));
429 return upper_bound - lower_bound + 1;
432 /** \brief Finds the number of elements in a array type (DW_TAG_array_type)
434 * The compilation unit might be needed because the default lower
435 * bound depends on the language of the compilation unit.
437 * \param die the DIE of the DW_TAG_array_type
438 * \param unit the DIE of the compilation unit
439 * \return number of elements in this array type
441 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit)
443 xbt_assert(dwarf_tag(die) == DW_TAG_array_type,
444 "MC_dwarf_array_element_count called with DIE of type %s",
445 simgrid::dwarf::tagname(die));
450 for (res = dwarf_child(die, &child); res == 0;
451 res = dwarf_siblingof(&child, &child)) {
452 int child_tag = dwarf_tag(&child);
453 if (child_tag == DW_TAG_subrange_type
454 || child_tag == DW_TAG_enumeration_type)
455 result *= MC_dwarf_subrange_element_count(&child, unit);
462 /** Sort the variable by name and address.
464 * We could use boost::container::flat_set instead.
466 static bool MC_compare_variable(
467 simgrid::mc::Variable const& a, simgrid::mc::Variable const& b)
469 int cmp = strcmp(a.name.c_str(), b.name.c_str());
475 return a.address < b.address;
478 // ***** simgrid::mc::Type*
480 /** \brief Initialize the location of a member of a type
481 * (DW_AT_data_member_location of a DW_TAG_member).
483 * \param type a type (struct, class)
484 * \param member the member of the type
485 * \param child DIE of the member (DW_TAG_member)
487 static void MC_dwarf_fill_member_location(
488 simgrid::mc::Type* type, simgrid::mc::Member* member, Dwarf_Die * child)
490 if (dwarf_hasattr(child, DW_AT_data_bit_offset))
491 xbt_die("Can't groke DW_AT_data_bit_offset.");
493 if (!dwarf_hasattr_integrate(child, DW_AT_data_member_location)) {
494 if (type->type == DW_TAG_union_type)
497 ("Missing DW_AT_data_member_location field in DW_TAG_member %s of type <%"
498 PRIx64 ">%s", member->name.c_str(),
499 (uint64_t) type->id, type->name.c_str());
502 Dwarf_Attribute attr;
503 dwarf_attr_integrate(child, DW_AT_data_member_location, &attr);
504 int form = dwarf_whatform(&attr);
505 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
506 switch (form_class) {
507 case simgrid::dwarf::FormClass::ExprLoc:
508 case simgrid::dwarf::FormClass::Block:
509 // Location expression:
513 if (dwarf_getlocation(&attr, &expr, &len))
515 ("Could not read location expression DW_AT_data_member_location in DW_TAG_member %s of type <%"
516 PRIx64 ">%s", MC_dwarf_attr_integrate_string(child, DW_AT_name),
517 (uint64_t) type->id, type->name.c_str());
518 member->location_expression = simgrid::dwarf::DwarfExpression(expr, expr+len);
521 case simgrid::dwarf::FormClass::Constant:
522 // Offset from the base address of the object:
525 if (!dwarf_formudata(&attr, &offset))
526 member->offset(offset);
528 xbt_die("Cannot get %s location <%" PRIx64 ">%s",
529 MC_dwarf_attr_integrate_string(child, DW_AT_name),
530 (uint64_t) type->id, type->name.c_str());
533 case simgrid::dwarf::FormClass::LocListPtr:
534 // Reference to a location list:
536 case simgrid::dwarf::FormClass::Reference:
537 // It's supposed to be possible in DWARF2 but I couldn't find its semantic
540 xbt_die("Can't handle form class (%i) / form 0x%x as DW_AT_member_location",
541 (int) form_class, form);
546 /** \brief Populate the list of members of a type
548 * \param info ELF object containing the type DIE
549 * \param die DIE of the type
550 * \param unit DIE of the compilation unit containing the type DIE
551 * \param type the type
553 static void MC_dwarf_add_members(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
554 Dwarf_Die * unit, simgrid::mc::Type* type)
558 xbt_assert(type->members.empty());
559 for (res = dwarf_child(die, &child); res == 0;
560 res = dwarf_siblingof(&child, &child)) {
561 int tag = dwarf_tag(&child);
562 if (tag == DW_TAG_member || tag == DW_TAG_inheritance) {
564 // Skip declarations:
565 if (MC_dwarf_attr_flag(&child, DW_AT_declaration, false))
568 // Skip compile time constants:
569 if (dwarf_hasattr(&child, DW_AT_const_value))
572 // TODO, we should use another type (because is is not a type but a member)
573 simgrid::mc::Member member;
574 member.inheritance = tag == DW_TAG_inheritance;
576 const char *name = MC_dwarf_attr_integrate_string(&child, DW_AT_name);
580 MC_dwarf_attr_integrate_uint(&child, DW_AT_byte_size, 0);
581 member.type_id = MC_dwarf_at_type(&child);
583 if (dwarf_hasattr(&child, DW_AT_data_bit_offset))
584 xbt_die("Can't groke DW_AT_data_bit_offset.");
586 MC_dwarf_fill_member_location(type, &member, &child);
589 xbt_die("Missing type for member %s of <%" PRIx64 ">%s",
591 (uint64_t) type->id, type->name.c_str());
593 type->members.push_back(std::move(member));
598 /** \brief Create a MC type object from a DIE
600 * \param info current object info object
601 * \param DIE (for a given type);
602 * \param unit compilation unit of the current DIE
603 * \return MC representation of the type
605 static simgrid::mc::Type MC_dwarf_die_to_type(
606 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
607 Dwarf_Die * unit, simgrid::mc::Frame* frame,
610 simgrid::mc::Type type;
611 type.type = dwarf_tag(die);
612 type.name = std::string();
613 type.element_count = -1;
616 type.id = dwarf_dieoffset(die);
618 const char *prefix = "";
620 case DW_TAG_structure_type:
623 case DW_TAG_union_type:
626 case DW_TAG_class_type:
633 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
634 if (name != nullptr) {
635 char* full_name = ns ? bprintf("%s%s::%s", prefix, ns, name) :
636 bprintf("%s%s", prefix, name);
637 type.name = std::string(full_name);
641 type.type_id = MC_dwarf_at_type(die);
643 // Some compilers do not emit DW_AT_byte_size for pointer_type,
644 // so we fill this. We currently assume that the model-checked process is in
645 // the same architecture..
646 if (type.type == DW_TAG_pointer_type)
647 type.byte_size = sizeof(void*);
649 // Computation of the byte_size;
650 if (dwarf_hasattr_integrate(die, DW_AT_byte_size))
651 type.byte_size = MC_dwarf_attr_integrate_uint(die, DW_AT_byte_size, 0);
652 else if (type.type == DW_TAG_array_type
653 || type.type == DW_TAG_structure_type
654 || type.type == DW_TAG_class_type) {
656 if (dwarf_aggregate_size(die, &size) == 0)
657 type.byte_size = size;
661 case DW_TAG_array_type:
662 type.element_count = MC_dwarf_array_element_count(die, unit);
663 // TODO, handle DW_byte_stride and (not) DW_bit_stride
666 case DW_TAG_pointer_type:
667 case DW_TAG_reference_type:
668 case DW_TAG_rvalue_reference_type:
671 case DW_TAG_structure_type:
672 case DW_TAG_union_type:
673 case DW_TAG_class_type:
674 MC_dwarf_add_members(info, die, unit, &type);
675 char *new_ns = ns == nullptr ? xbt_strdup(type.name.c_str())
676 : bprintf("%s::%s", ns, name);
677 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
685 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
686 Dwarf_Die * unit, simgrid::mc::Frame* frame,
689 simgrid::mc::Type type = MC_dwarf_die_to_type(info, die, unit, frame, ns);
690 auto& t = (info->types[type.id] = std::move(type));
691 if (!t.name.empty() && type.byte_size != 0)
692 info->full_types_by_name[t.name] = &t;
695 static int mc_anonymous_variable_index = 0;
697 static std::unique_ptr<simgrid::mc::Variable> MC_die_to_variable(
698 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
699 Dwarf_Die * unit, simgrid::mc::Frame* frame,
702 // Skip declarations:
703 if (MC_dwarf_attr_flag(die, DW_AT_declaration, false))
706 // Skip compile time constants:
707 if (dwarf_hasattr(die, DW_AT_const_value))
710 Dwarf_Attribute attr_location;
711 if (dwarf_attr(die, DW_AT_location, &attr_location) == nullptr)
712 // No location: do not add it ?
715 std::unique_ptr<simgrid::mc::Variable> variable =
716 std::unique_ptr<simgrid::mc::Variable>(new simgrid::mc::Variable());
717 variable->id = dwarf_dieoffset(die);
718 variable->global = frame == nullptr; // Can be override base on DW_AT_location
719 variable->object_info = info;
721 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
723 variable->name = name;
724 variable->type_id = MC_dwarf_at_type(die);
726 int form = dwarf_whatform(&attr_location);
727 simgrid::dwarf::FormClass form_class;
728 if (form == DW_FORM_sec_offset)
729 form_class = simgrid::dwarf::FormClass::Constant;
731 form_class = simgrid::dwarf::classify_form(form);
732 switch (form_class) {
733 case simgrid::dwarf::FormClass::ExprLoc:
734 case simgrid::dwarf::FormClass::Block:
735 // Location expression:
739 if (dwarf_getlocation(&attr_location, &expr, &len)) {
741 "Could not read location expression in DW_AT_location "
742 "of variable <%" PRIx64 ">%s",
743 (uint64_t) variable->id,
744 variable->name.c_str());
747 if (len == 1 && expr[0].atom == DW_OP_addr) {
748 variable->global = true;
749 uintptr_t offset = (uintptr_t) expr[0].number;
750 uintptr_t base = (uintptr_t) info->base_address();
751 variable->address = (void *) (base + offset);
753 variable->location_list = {
754 simgrid::dwarf::DwarfExpression(expr, expr + len) };
759 case simgrid::dwarf::FormClass::LocListPtr:
760 case simgrid::dwarf::FormClass::Constant:
761 // Reference to location list:
762 variable->location_list = simgrid::dwarf::location_list(
763 *info, attr_location);
767 xbt_die("Unexpected form 0x%x (%i), class 0x%x (%i) list for location "
768 "in <%" PRIx64 ">%s",
769 form, form, (int) form_class, (int) form_class,
770 (uint64_t) variable->id,
771 variable->name.c_str());
774 // Handle start_scope:
775 if (dwarf_hasattr(die, DW_AT_start_scope)) {
776 Dwarf_Attribute attr;
777 dwarf_attr(die, DW_AT_start_scope, &attr);
778 int form = dwarf_whatform(&attr);
779 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
780 switch (form_class) {
781 case simgrid::dwarf::FormClass::Constant:
784 variable->start_scope =
785 dwarf_formudata(&attr, &value) == 0 ? (size_t) value : 0;
789 case simgrid::dwarf::FormClass::RangeListPtr: // TODO
792 ("Unhandled form 0x%x, class 0x%X for DW_AT_start_scope of variable %s",
793 form, (int) form_class, name == nullptr ? "?" : name);
797 if (ns && variable->global)
799 std::string(ns) + "::" + variable->name;
801 // The current code needs a variable name,
802 // generate a fake one:
803 if (variable->name.empty())
805 "@anonymous#" + std::to_string(mc_anonymous_variable_index++);
810 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
811 Dwarf_Die * unit, simgrid::mc::Frame* frame,
814 std::unique_ptr<simgrid::mc::Variable> variable =
815 MC_die_to_variable(info, die, unit, frame, ns);
818 // Those arrays are sorted later:
819 else if (variable->global)
820 info->global_variables.push_back(std::move(*variable));
821 else if (frame != nullptr)
822 frame->variables.push_back(std::move(*variable));
824 xbt_die("No frame for this local variable");
827 static void MC_dwarf_handle_scope_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
828 Dwarf_Die * unit, simgrid::mc::Frame* parent_frame,
831 // TODO, handle DW_TAG_type/DW_TAG_location for DW_TAG_with_stmt
832 int tag = dwarf_tag(die);
833 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
835 // (Template) Subprogram declaration:
836 if (klass == simgrid::dwarf::TagClass::Subprogram
837 && MC_dwarf_attr_flag(die, DW_AT_declaration, false))
840 if (klass == simgrid::dwarf::TagClass::Scope)
841 xbt_assert(parent_frame, "No parent scope for this scope");
843 simgrid::mc::Frame frame;
845 frame.id = dwarf_dieoffset(die);
846 frame.object_info = info;
848 if (klass == simgrid::dwarf::TagClass::Subprogram) {
849 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
851 frame.name = std::string(ns) + "::" + name;
856 frame.abstract_origin_id =
857 MC_dwarf_attr_dieoffset(die, DW_AT_abstract_origin);
859 // This is the base address for DWARF addresses.
860 // Relocated addresses are offset from this base address.
861 // See DWARF4 spec 7.5
862 std::uint64_t base = (std::uint64_t) info->base_address();
864 // TODO, support DW_AT_ranges
865 uint64_t low_pc = MC_dwarf_attr_integrate_addr(die, DW_AT_low_pc);
866 frame.range.begin() = low_pc ? (std::uint64_t) base + low_pc : 0;
869 Dwarf_Attribute attr;
870 if (!dwarf_attr_integrate(die, DW_AT_high_pc, &attr))
871 xbt_die("Missing DW_AT_high_pc matching with DW_AT_low_pc");
876 switch (simgrid::dwarf::classify_form(dwarf_whatform(&attr))) {
878 // DW_AT_high_pc if an offset from the low_pc:
879 case simgrid::dwarf::FormClass::Constant:
881 if (dwarf_formsdata(&attr, &offset) != 0)
882 xbt_die("Could not read constant");
883 frame.range.end() = frame.range.begin() + offset;
886 // DW_AT_high_pc is a relocatable address:
887 case simgrid::dwarf::FormClass::Address:
888 if (dwarf_formaddr(&attr, &high_pc) != 0)
889 xbt_die("Could not read address");
890 frame.range.end() = base + high_pc;
894 xbt_die("Unexpected class for DW_AT_high_pc");
899 if (klass == simgrid::dwarf::TagClass::Subprogram) {
900 Dwarf_Attribute attr_frame_base;
901 if (dwarf_attr_integrate(die, DW_AT_frame_base, &attr_frame_base))
902 frame.frame_base_location = simgrid::dwarf::location_list(*info,
907 MC_dwarf_handle_children(info, die, unit, &frame, ns);
909 // We sort them in order to have an (somewhat) efficient by name
911 std::sort(frame.variables.begin(), frame.variables.end(),
912 MC_compare_variable);
915 if (klass == simgrid::dwarf::TagClass::Subprogram)
916 info->subprograms[frame.id] = std::move(frame);
917 else if (klass == simgrid::dwarf::TagClass::Scope)
918 parent_frame->scopes.push_back(std::move(frame));
921 static void mc_dwarf_handle_namespace_die(simgrid::mc::ObjectInformation* info,
922 Dwarf_Die * die, Dwarf_Die * unit,
923 simgrid::mc::Frame* frame,
926 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
928 xbt_die("Unexpected namespace in a subprogram");
929 char *new_ns = ns == nullptr ? xbt_strdup(name)
930 : bprintf("%s::%s", ns, name);
931 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
935 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
936 Dwarf_Die * unit, simgrid::mc::Frame* frame,
939 // For each child DIE:
942 for (res = dwarf_child(die, &child); res == 0;
943 res = dwarf_siblingof(&child, &child))
944 MC_dwarf_handle_die(info, &child, unit, frame, ns);
947 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
948 Dwarf_Die * unit, simgrid::mc::Frame* frame,
951 int tag = dwarf_tag(die);
952 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
956 case simgrid::dwarf::TagClass::Type:
957 MC_dwarf_handle_type_die(info, die, unit, frame, ns);
960 // Subprogram or scope:
961 case simgrid::dwarf::TagClass::Subprogram:
962 case simgrid::dwarf::TagClass::Scope:
963 MC_dwarf_handle_scope_die(info, die, unit, frame, ns);
967 case simgrid::dwarf::TagClass::Variable:
968 MC_dwarf_handle_variable_die(info, die, unit, frame, ns);
971 case simgrid::dwarf::TagClass::Namespace:
972 mc_dwarf_handle_namespace_die(info, die, unit, frame, ns);
982 Elf64_Half get_type(Elf* elf)
984 Elf64_Ehdr* ehdr64 = elf64_getehdr(elf);
986 return ehdr64->e_type;
987 Elf32_Ehdr* ehdr32 = elf32_getehdr(elf);
989 return ehdr32->e_type;
990 xbt_die("Could not get ELF heeader");
994 void read_dwarf_info(simgrid::mc::ObjectInformation* info, Dwarf* dwarf)
996 // For each compilation unit:
997 Dwarf_Off offset = 0;
998 Dwarf_Off next_offset = 0;
1001 while (dwarf_nextcu(dwarf, offset, &next_offset, &length, nullptr, nullptr, nullptr) ==
1004 if (dwarf_offdie(dwarf, offset + length, &unit_die) != nullptr)
1005 MC_dwarf_handle_children(info, &unit_die, &unit_die, nullptr, nullptr);
1006 offset = next_offset;
1010 /** Get the build-id (NT_GNU_BUILD_ID) from the ELF file
1012 * This build-id may is used to locate an external debug (DWARF) file
1013 * for this ELF file.
1015 * @param elf libelf handle for an ELF file
1016 * @return build-id for this ELF file (or an empty vector if none is found)
1019 std::vector<char> get_build_id(Elf* elf)
1022 if (elf_getphdrnum (elf, &phnum) != 0)
1023 xbt_die("Could not read program headers");
1025 // Iterate over the program headers and find the PT_NOTE ones:
1026 for (size_t i = 0; i < phnum; ++i) {
1027 GElf_Phdr phdr_temp;
1028 GElf_Phdr *phdr = gelf_getphdr(elf, i, &phdr_temp);
1029 if (phdr->p_type != PT_NOTE)
1032 Elf_Data* data = elf_getdata_rawchunk(elf, phdr->p_offset, phdr->p_filesz, ELF_T_NHDR);
1034 // Iterate over the notes and find the NT_GNU_BUILD_ID one:
1040 pos = gelf_getnote(data, pos, &nhdr, &name_pos, &desc_pos);
1041 // A note is identified by a name "GNU" and a integer type within
1042 // the namespace defined by this name (here NT_GNU_BUILD_ID):
1043 if (nhdr.n_type == NT_GNU_BUILD_ID
1044 && nhdr.n_namesz == sizeof("GNU")
1045 && memcmp((char*) data->d_buf + name_pos, "GNU", sizeof("GNU")) == 0) {
1047 // Found the NT_GNU_BUILD_ID note:
1048 char* start = (char*) data->d_buf + desc_pos;
1049 char* end = (char*) start + nhdr.n_descsz;
1050 return std::vector<char>(start, end);
1056 return std::vector<char>();
1059 static char hexdigits[16] = {
1060 '0', '1', '2', '3', '4', '5', '6', '7', '8', '9',
1061 'a', 'b', 'c', 'd', 'e', 'f'
1064 /** Binary data to hexadecimal */
1066 std::array<char, 2> to_hex(std::uint8_t byte)
1068 // Horrid double braces!
1069 // Apparently, this is needed in C++11 (not in C++14).
1070 return { { hexdigits[byte >> 4], hexdigits[byte & 0xF] } };
1073 /** Binary data to hexadecimal */
1075 std::string to_hex(const char* data, std::size_t count)
1078 res.resize(2*count);
1079 for (std::size_t i = 0; i < count; i++) {
1080 std::array<char, 2> hex_byte = to_hex(data[i]);
1081 for (int j = 0; j < 2; ++j)
1082 res[2 * i + j] = hex_byte[j];
1087 /** Binary data to hexadecimal */
1089 std::string to_hex(std::vector<char> const& data)
1091 return to_hex(data.data(), data.size());
1094 /** Base directories for external debug files */
1096 const char* debug_paths[] = {
1098 "/usr/local/lib/debug/",
1101 /** Locate an external debug file from the NT_GNU_BUILD_ID
1103 * This is one of the mechanisms used for
1104 * [separate debug files](https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html).
1107 std::string find_by_build_id(std::vector<char> id)
1109 std::string filename;
1110 for (const char* debug_path : debug_paths) {
1111 filename = debug_path;
1112 filename += ".build-id/" + to_hex(id.data(), 1) + '/'
1113 + to_hex(id.data() + 1, id.size() - 1) + ".debug";
1114 XBT_DEBUG("Checking debug file: %s", filename.c_str());
1115 if (access(filename.c_str(), F_OK) == 0)
1118 return std::string();
1121 /** \brief Populate the debugging informations of the given ELF object
1123 * Read the DWARf information of the EFFL object and populate the
1124 * lists of types, variables, functions.
1127 void MC_dwarf_get_variables(simgrid::mc::ObjectInformation* info)
1129 if (elf_version(EV_CURRENT) == EV_NONE)
1130 xbt_die("libelf initialization error");
1132 // Open the ELF file:
1133 int fd = open(info->file_name.c_str(), O_RDONLY);
1135 xbt_die("Could not open file %s", info->file_name.c_str());
1136 Elf* elf = elf_begin(fd, ELF_C_READ, nullptr);
1138 xbt_die("Not an ELF file 1");
1139 Elf_Kind kind = elf_kind(elf);
1140 if (kind != ELF_K_ELF)
1141 xbt_die("Not an ELF file 2");
1143 // Remember if this is a `ET_EXEC` (fixed location) or `ET_DYN` (relocatable):
1144 Elf64_Half type = get_type(elf);
1145 if (type == ET_EXEC)
1146 info->flags |= simgrid::mc::ObjectInformation::Executable;
1148 // Read DWARF debug information in the file:
1149 Dwarf* dwarf = dwarf_begin_elf (elf, DWARF_C_READ, nullptr);
1150 if (dwarf != nullptr) {
1151 read_dwarf_info(info, dwarf);
1159 // If there was no DWARF in the file, try to find it in a separate file
1160 // with NT_GNU_BUILD_ID:
1161 std::vector<char> build_id = get_build_id(elf);
1162 if (!build_id.empty()) {
1166 // Find the debug file using the build id:
1167 std::string debug_file = find_by_build_id(build_id);
1168 if (debug_file.empty()) {
1169 std::string hex = to_hex(build_id);
1171 "Missing debug info for %s with build-id %s\n"
1172 "You might want to install the suitable debugging package.\n",
1173 info->file_name.c_str(), hex.c_str());
1176 // Load the DWARF info from this file:
1177 XBT_DEBUG("Load DWARF for %s from %s",
1178 info->file_name.c_str(), debug_file.c_str());
1179 fd = open(debug_file.c_str(), O_RDONLY);
1181 xbt_die("Could not open file %s", debug_file.c_str());
1182 Dwarf* dwarf = dwarf_begin(fd, DWARF_C_READ);
1183 if (dwarf == nullptr)
1184 xbt_die("No DWARF info in %s for %s",
1185 debug_file.c_str(), info->file_name.c_str());
1186 read_dwarf_info(info, dwarf);
1192 // TODO, try to find DWARF info using debug-link.
1193 // Is this method really used anywhere?
1195 xbt_die("Debugging information not found for %s\n"
1196 "Try recompiling with -g\n",
1197 info->file_name.c_str());
1200 // ***** Functions index
1202 static int MC_compare_frame_index_items(simgrid::mc::FunctionIndexEntry* a,
1203 simgrid::mc::FunctionIndexEntry* b)
1205 if (a->low_pc < b->low_pc)
1207 else if (a->low_pc == b->low_pc)
1213 static void MC_make_functions_index(simgrid::mc::ObjectInformation* info)
1215 info->functions_index.clear();
1217 for (auto& e : info->subprograms) {
1218 if (e.second.range.begin() == 0)
1220 simgrid::mc::FunctionIndexEntry entry;
1221 entry.low_pc = (void*) e.second.range.begin();
1222 entry.function = &e.second;
1223 info->functions_index.push_back(entry);
1226 info->functions_index.shrink_to_fit();
1228 // Sort the array by low_pc:
1229 std::sort(info->functions_index.begin(), info->functions_index.end(),
1230 [](simgrid::mc::FunctionIndexEntry const& a,
1231 simgrid::mc::FunctionIndexEntry const& b)
1233 return a.low_pc < b.low_pc;
1237 static void MC_post_process_variables(simgrid::mc::ObjectInformation* info)
1239 // Someone needs this to be sorted but who?
1240 std::sort(info->global_variables.begin(), info->global_variables.end(),
1241 MC_compare_variable);
1243 for(simgrid::mc::Variable& variable : info->global_variables)
1244 if (variable.type_id)
1245 variable.type = simgrid::util::find_map_ptr(
1246 info->types, variable.type_id);
1249 static void mc_post_process_scope(simgrid::mc::ObjectInformation* info, simgrid::mc::Frame* scope)
1252 if (scope->tag == DW_TAG_inlined_subroutine) {
1253 // Attach correct namespaced name in inlined subroutine:
1254 auto i = info->subprograms.find(scope->abstract_origin_id);
1255 xbt_assert(i != info->subprograms.end(),
1256 "Could not lookup abstract origin %" PRIx64,
1257 (std::uint64_t) scope->abstract_origin_id);
1258 scope->name = i->second.name;
1262 for (simgrid::mc::Variable& variable : scope->variables)
1263 if (variable.type_id)
1264 variable.type = simgrid::util::find_map_ptr(
1265 info->types, variable.type_id);
1267 // Recursive post-processing of nested-scopes:
1268 for (simgrid::mc::Frame& nested_scope : scope->scopes)
1269 mc_post_process_scope(info, &nested_scope);
1274 simgrid::mc::Type* MC_resolve_type(
1275 simgrid::mc::ObjectInformation* info, unsigned type_id)
1279 simgrid::mc::Type* type = simgrid::util::find_map_ptr(info->types, type_id);
1280 if (type == nullptr)
1283 // We already have the information on the type:
1284 if (type->byte_size != 0)
1287 // Don't have a name, we can't find a more complete version:
1288 if (type->name.empty())
1291 // Try to find a more complete description of the type:
1292 // We need to fix in order to support C++.
1293 simgrid::mc::Type** subtype = simgrid::util::find_map_ptr(
1294 info->full_types_by_name, type->name);
1300 static void MC_post_process_types(simgrid::mc::ObjectInformation* info)
1302 // Lookup "subtype" field:
1303 for(auto& i : info->types) {
1304 i.second.subtype = MC_resolve_type(info, i.second.type_id);
1305 for (simgrid::mc::Member& member : i.second.members)
1306 member.type = MC_resolve_type(info, member.type_id);
1313 /** \brief Finds informations about a given shared object/executable */
1314 std::shared_ptr<simgrid::mc::ObjectInformation> createObjectInformation(
1315 std::vector<simgrid::xbt::VmMap> const& maps, const char *name)
1317 std::shared_ptr<simgrid::mc::ObjectInformation> result =
1318 std::make_shared<simgrid::mc::ObjectInformation>();
1319 result->file_name = name;
1320 simgrid::mc::find_object_address(maps, result.get());
1321 MC_dwarf_get_variables(result.get());
1322 MC_post_process_variables(result.get());
1323 MC_post_process_types(result.get());
1324 for (auto& entry : result.get()->subprograms)
1325 mc_post_process_scope(result.get(), &entry.second);
1326 MC_make_functions_index(result.get());
1330 /*************************************************************************/
1332 void postProcessObjectInformation(simgrid::mc::Process* process, simgrid::mc::ObjectInformation* info)
1334 for (auto& i : info->types) {
1336 simgrid::mc::Type* type = &(i.second);
1337 simgrid::mc::Type* subtype = type;
1338 while (subtype->type == DW_TAG_typedef
1339 || subtype->type == DW_TAG_volatile_type
1340 || subtype->type == DW_TAG_const_type)
1341 if (subtype->subtype)
1342 subtype = subtype->subtype;
1346 // Resolve full_type:
1347 if (!subtype->name.empty() && subtype->byte_size == 0)
1348 for (auto const& object_info : process->object_infos) {
1349 auto i = object_info->full_types_by_name.find(subtype->name);
1350 if (i != object_info->full_types_by_name.end()
1351 && !i->second->name.empty() && i->second->byte_size) {
1352 type->full_type = i->second;
1356 else type->full_type = subtype;
1367 /** Convert a DWARF register into a libunwind register
1369 * DWARF and libunwind does not use the same convention for numbering the
1370 * registers on some architectures. The function makes the necessary
1373 int dwarf_register_to_libunwind(int dwarf_register)
1375 #if defined(__x86_64__)
1376 // It seems for this arch, DWARF and libunwind agree in the numbering:
1377 return dwarf_register;
1378 #elif defined(__i386__)
1379 // Could't find the authoritative source of information for this.
1380 // This is inspired from http://source.winehq.org/source/dlls/dbghelp/cpu_i386.c#L517.
1381 switch (dwarf_register) {
1401 return UNW_X86_EFLAGS;
1431 xbt_die("Bad/unknown register number.");
1434 #error This architecture is not supported yet for DWARF expression evaluation.