1 /* Copyright (c) 2008-2017. 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. */
13 #include <boost/range/algorithm.hpp>
17 #define DW_LANG_Objc DW_LANG_ObjC /* fix spelling error in older dwarf.h */
19 #include <elfutils/libdw.h>
21 #include <boost/algorithm/string/predicate.hpp>
23 #include "src/simgrid/util.hpp"
25 #include "xbt/string.hpp"
26 #include "xbt/sysdep.h"
27 #include <simgrid_config.h>
29 #include "src/mc/mc_dwarf.hpp"
30 #include "src/mc/mc_private.hpp"
32 #include "src/mc/ObjectInformation.hpp"
33 #include "src/mc/Variable.hpp"
34 #include "src/mc/remote/RemoteClient.hpp"
36 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_dwarf, mc, "DWARF processing");
38 /** \brief The default DW_TAG_lower_bound for a given DW_AT_language.
40 * The default for a given language is defined in the DWARF spec.
42 * \param language constant as defined by the DWARf spec
44 static uint64_t MC_dwarf_default_lower_bound(int lang);
46 /** \brief Computes the the element_count of a DW_TAG_enumeration_type DIE
48 * This is the number of elements in a given array dimension.
50 * A reference of the compilation unit (DW_TAG_compile_unit) is
51 * needed because the default lower bound (when there is no DW_AT_lower_bound)
52 * depends of the language of the compilation unit (DW_AT_language).
54 * \param die DIE for the DW_TAG_enumeration_type or DW_TAG_subrange_type
55 * \param unit DIE of the DW_TAG_compile_unit
57 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die* die, Dwarf_Die* unit);
59 /** \brief Computes the number of elements of a given DW_TAG_array_type.
61 * \param die DIE for the DW_TAG_array_type
63 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit);
65 /** \brief Process a DIE
67 * \param info the resulting object fot the library/binary file (output)
68 * \param die the current DIE
69 * \param unit the DIE of the compile unit of the current DIE
70 * \param frame containing frame if any
72 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
73 Dwarf_Die * unit, simgrid::mc::Frame* frame,
76 /** \brief Process a type DIE
78 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
79 Dwarf_Die * unit, simgrid::mc::Frame* frame,
82 /** \brief Calls MC_dwarf_handle_die on all children of the given die
84 * \param info the resulting object fot the library/binary file (output)
85 * \param die the current DIE
86 * \param unit the DIE of the compile unit of the current DIE
87 * \param frame containing frame if any
89 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
90 Dwarf_Die * unit, simgrid::mc::Frame* frame,
93 /** \brief Handle a variable (DW_TAG_variable or other)
95 * \param info the resulting object fot the library/binary file (output)
96 * \param die the current DIE
97 * \param unit the DIE of the compile unit of the current DIE
98 * \param frame containing frame if any
100 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
101 Dwarf_Die * unit, simgrid::mc::Frame* frame,
104 /** \brief Get the DW_TAG_type of the DIE
107 * \return DW_TAG_type attribute as a new string (nullptr if none)
109 static std::uint64_t MC_dwarf_at_type(Dwarf_Die * die);
114 enum class TagClass {
123 /*** Class of forms defined in the DWARF standard */
124 enum class FormClass {
126 Address, // Location in the program's address space
127 Block, // Arbitrary block of bytes
130 Flag, // Boolean value
131 Reference, // Reference to another DIE
132 ExprLoc, // DWARF expression/location description
140 TagClass classify_tag(int tag)
144 case DW_TAG_array_type:
145 case DW_TAG_class_type:
146 case DW_TAG_enumeration_type:
148 case DW_TAG_pointer_type:
149 case DW_TAG_reference_type:
150 case DW_TAG_rvalue_reference_type:
151 case DW_TAG_string_type:
152 case DW_TAG_structure_type:
153 case DW_TAG_subroutine_type:
154 case DW_TAG_union_type:
155 case DW_TAG_ptr_to_member_type:
156 case DW_TAG_set_type:
157 case DW_TAG_subrange_type:
158 case DW_TAG_base_type:
159 case DW_TAG_const_type:
160 case DW_TAG_file_type:
161 case DW_TAG_packed_type:
162 case DW_TAG_volatile_type:
163 case DW_TAG_restrict_type:
164 case DW_TAG_interface_type:
165 case DW_TAG_unspecified_type:
166 case DW_TAG_shared_type:
167 return TagClass::Type;
169 case DW_TAG_subprogram:
170 return TagClass::Subprogram;
172 case DW_TAG_variable:
173 case DW_TAG_formal_parameter:
174 return TagClass::Variable;
176 case DW_TAG_lexical_block:
177 case DW_TAG_try_block:
178 case DW_TAG_catch_block:
179 case DW_TAG_inlined_subroutine:
180 case DW_TAG_with_stmt:
181 return TagClass::Scope;
183 case DW_TAG_namespace:
184 return TagClass::Namespace;
187 return TagClass::Unknown;
191 /** \brief Find the DWARF data class for a given DWARF data form
193 * This mapping is defined in the DWARF spec.
195 * \param form The form (values taken from the DWARF spec)
196 * \return An internal representation for the corresponding class
199 FormClass classify_form(int form)
203 return FormClass::Address;
208 return FormClass::Block;
215 return FormClass::Constant;
218 return FormClass::String;
219 case DW_FORM_ref_addr:
224 case DW_FORM_ref_udata:
225 return FormClass::Reference;
227 case DW_FORM_flag_present:
228 return FormClass::Flag;
229 case DW_FORM_exprloc:
230 return FormClass::ExprLoc;
234 return FormClass::Unknown;
238 /** \brief Get the name of the tag of a given DIE
241 * \return name of the tag of this DIE
244 const char *tagname(Dwarf_Die * die)
246 return simgrid::dwarf::tagname(dwarf_tag(die));
254 /** \brief Get an attribute of a given DIE as a string
257 * \param attribute attribute
258 * \return value of the given attribute of the given DIE
260 static const char *MC_dwarf_attr_integrate_string(Dwarf_Die * die,
263 Dwarf_Attribute attr;
264 if (not dwarf_attr_integrate(die, attribute, &attr))
267 return dwarf_formstring(&attr);
270 /** \brief Get the linkage name of a DIE.
272 * Use either DW_AT_linkage_name or DW_AT_MIPS_linkage_name.
273 * DW_AT_linkage_name is standardized since DWARF 4.
274 * Before this version of DWARF, the MIPS extensions
275 * DW_AT_MIPS_linkage_name is used (at least by GCC).
278 * \return linkage name of the given DIE (or nullptr)
280 static const char *MC_dwarf_at_linkage_name(Dwarf_Die * die)
282 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_linkage_name);
284 name = MC_dwarf_attr_integrate_string(die, DW_AT_MIPS_linkage_name);
288 static Dwarf_Off MC_dwarf_attr_dieoffset(Dwarf_Die * die, int attribute)
290 Dwarf_Attribute attr;
291 if (dwarf_hasattr_integrate(die, attribute) == 0)
293 dwarf_attr_integrate(die, attribute, &attr);
294 Dwarf_Die subtype_die;
295 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
296 xbt_die("Could not find DIE");
297 return dwarf_dieoffset(&subtype_die);
300 static Dwarf_Off MC_dwarf_attr_integrate_dieoffset(Dwarf_Die * die,
303 Dwarf_Attribute attr;
304 if (dwarf_hasattr_integrate(die, attribute) == 0)
306 dwarf_attr_integrate(die, DW_AT_type, &attr);
307 Dwarf_Die subtype_die;
308 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
309 xbt_die("Could not find DIE");
310 return dwarf_dieoffset(&subtype_die);
313 /** \brief Find the type/subtype (DW_AT_type) for a DIE
316 * \return DW_AT_type reference as a global offset in hexadecimal (or nullptr)
319 std::uint64_t MC_dwarf_at_type(Dwarf_Die * die)
321 return MC_dwarf_attr_integrate_dieoffset(die, DW_AT_type);
324 static uint64_t MC_dwarf_attr_integrate_addr(Dwarf_Die * die, int attribute)
326 Dwarf_Attribute attr;
327 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
330 if (dwarf_formaddr(&attr, &value) == 0)
331 return (uint64_t) value;
336 static uint64_t MC_dwarf_attr_integrate_uint(Dwarf_Die * die, int attribute,
337 uint64_t default_value)
339 Dwarf_Attribute attr;
340 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
341 return default_value;
343 return dwarf_formudata(dwarf_attr_integrate(die, attribute, &attr),
344 &value) == 0 ? (uint64_t) value : default_value;
347 static bool MC_dwarf_attr_flag(Dwarf_Die * die, int attribute, bool integrate)
349 Dwarf_Attribute attr;
350 if ((integrate ? dwarf_attr_integrate(die, attribute, &attr)
351 : dwarf_attr(die, attribute, &attr)) == 0)
355 if (dwarf_formflag(&attr, &result))
356 xbt_die("Unexpected form for attribute %s",
357 simgrid::dwarf::attrname(attribute));
361 /** @brief Find the default lower bound for a given language
363 * The default lower bound of an array (when DW_TAG_lower_bound
364 * is missing) depends on the language of the compilation unit.
366 * @param lang Language of the compilation unit (values defined in the DWARF spec)
367 * @return Default lower bound of an array in this compilation unit
369 static uint64_t MC_dwarf_default_lower_bound(int lang)
375 case DW_LANG_C_plus_plus:
379 case DW_LANG_ObjC_plus_plus:
385 case DW_LANG_Fortran77:
386 case DW_LANG_Fortran90:
387 case DW_LANG_Fortran95:
388 case DW_LANG_Modula2:
389 case DW_LANG_Pascal83:
391 case DW_LANG_Cobol74:
392 case DW_LANG_Cobol85:
395 xbt_die("No default DW_TAG_lower_bound for language %i and none given",
401 /** \brief Finds the number of elements in a DW_TAG_subrange_type or DW_TAG_enumeration_type DIE
404 * \param unit DIE of the compilation unit
405 * \return number of elements in the range
407 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die * die,
410 xbt_assert(dwarf_tag(die) == DW_TAG_enumeration_type
411 || dwarf_tag(die) == DW_TAG_subrange_type,
412 "MC_dwarf_subrange_element_count called with DIE of type %s",
413 simgrid::dwarf::tagname(die));
415 // Use DW_TAG_count if present:
416 if (dwarf_hasattr_integrate(die, DW_AT_count))
417 return MC_dwarf_attr_integrate_uint(die, DW_AT_count, 0);
418 // Otherwise compute DW_TAG_upper_bound-DW_TAG_lower_bound + 1:
420 if (not dwarf_hasattr_integrate(die, DW_AT_upper_bound))
421 // This is not really 0, but the code expects this (we do not know):
424 uint64_t upper_bound =
425 MC_dwarf_attr_integrate_uint(die, DW_AT_upper_bound, -1);
427 uint64_t lower_bound = 0;
428 if (dwarf_hasattr_integrate(die, DW_AT_lower_bound))
429 lower_bound = MC_dwarf_attr_integrate_uint(die, DW_AT_lower_bound, -1);
431 lower_bound = MC_dwarf_default_lower_bound(dwarf_srclang(unit));
432 return upper_bound - lower_bound + 1;
435 /** \brief Finds the number of elements in a array type (DW_TAG_array_type)
437 * The compilation unit might be needed because the default lower
438 * bound depends on the language of the compilation unit.
440 * \param die the DIE of the DW_TAG_array_type
441 * \param unit the DIE of the compilation unit
442 * \return number of elements in this array type
444 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit)
446 xbt_assert(dwarf_tag(die) == DW_TAG_array_type,
447 "MC_dwarf_array_element_count called with DIE of type %s",
448 simgrid::dwarf::tagname(die));
453 for (res = dwarf_child(die, &child); res == 0;
454 res = dwarf_siblingof(&child, &child)) {
455 int child_tag = dwarf_tag(&child);
456 if (child_tag == DW_TAG_subrange_type
457 || child_tag == DW_TAG_enumeration_type)
458 result *= MC_dwarf_subrange_element_count(&child, unit);
465 /** Sort the variable by name and address.
467 * We could use boost::container::flat_set instead.
469 static bool MC_compare_variable(
470 simgrid::mc::Variable const& a, simgrid::mc::Variable const& b)
472 int cmp = strcmp(a.name.c_str(), b.name.c_str());
478 return a.address < b.address;
481 // ***** simgrid::mc::Type*
483 /** \brief Initialize the location of a member of a type
484 * (DW_AT_data_member_location of a DW_TAG_member).
486 * \param type a type (struct, class)
487 * \param member the member of the type
488 * \param child DIE of the member (DW_TAG_member)
490 static void MC_dwarf_fill_member_location(
491 simgrid::mc::Type* type, simgrid::mc::Member* member, Dwarf_Die * child)
493 if (dwarf_hasattr(child, DW_AT_data_bit_offset))
494 xbt_die("Can't groke DW_AT_data_bit_offset.");
496 if (not dwarf_hasattr_integrate(child, DW_AT_data_member_location)) {
497 if (type->type == DW_TAG_union_type)
500 ("Missing DW_AT_data_member_location field in DW_TAG_member %s of type <%"
501 PRIx64 ">%s", member->name.c_str(),
502 (uint64_t) type->id, type->name.c_str());
505 Dwarf_Attribute attr;
506 dwarf_attr_integrate(child, DW_AT_data_member_location, &attr);
507 int form = dwarf_whatform(&attr);
508 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
509 switch (form_class) {
510 case simgrid::dwarf::FormClass::ExprLoc:
511 case simgrid::dwarf::FormClass::Block:
512 // Location expression:
516 if (dwarf_getlocation(&attr, &expr, &len))
518 ("Could not read location expression DW_AT_data_member_location in DW_TAG_member %s of type <%"
519 PRIx64 ">%s", MC_dwarf_attr_integrate_string(child, DW_AT_name),
520 (uint64_t) type->id, type->name.c_str());
521 member->location_expression = simgrid::dwarf::DwarfExpression(expr, expr+len);
524 case simgrid::dwarf::FormClass::Constant:
525 // Offset from the base address of the object:
528 if (not dwarf_formudata(&attr, &offset))
529 member->offset(offset);
531 xbt_die("Cannot get %s location <%" PRIx64 ">%s",
532 MC_dwarf_attr_integrate_string(child, DW_AT_name),
533 (uint64_t) type->id, type->name.c_str());
538 // includes FormClass::LocListPtr (reference to a location list: TODO) and FormClass::Reference (it's supposed to be
539 // possible in DWARF2 but I couldn't find its semantic in the spec)
540 xbt_die("Can't handle form class (%d) / form 0x%x as DW_AT_member_location", (int)form_class, (unsigned)form);
545 /** \brief Populate the list of members of a type
547 * \param info ELF object containing the type DIE
548 * \param die DIE of the type
549 * \param unit DIE of the compilation unit containing the type DIE
550 * \param type the type
552 static void MC_dwarf_add_members(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
553 Dwarf_Die * unit, simgrid::mc::Type* type)
557 xbt_assert(type->members.empty());
558 for (res = dwarf_child(die, &child); res == 0;
559 res = dwarf_siblingof(&child, &child)) {
560 int tag = dwarf_tag(&child);
561 if (tag == DW_TAG_member || tag == DW_TAG_inheritance) {
563 // Skip declarations:
564 if (MC_dwarf_attr_flag(&child, DW_AT_declaration, false))
567 // Skip compile time constants:
568 if (dwarf_hasattr(&child, DW_AT_const_value))
571 // TODO, we should use another type (because is is not a type but a member)
572 simgrid::mc::Member member;
573 if (tag == DW_TAG_inheritance)
574 member.flags |= simgrid::mc::Member::INHERITANCE_FLAG;
576 const char *name = MC_dwarf_attr_integrate_string(&child, DW_AT_name);
579 // Those base names are used by GCC and clang for virtual table pointers
580 // respectively ("__vptr$ClassName", "__vptr.ClassName"):
581 if (boost::algorithm::starts_with(member.name, "__vptr$") ||
582 boost::algorithm::starts_with(member.name, "__vptr."))
583 member.flags |= simgrid::mc::Member::VIRTUAL_POINTER_FLAG;
584 // A cleaner solution would be to check against the type:
586 // tag: DW_TAG_member
589 // # Type for a pointer to a vtable
590 // tag: DW_TAG_pointer_type
592 // # Type for a vtable:
593 // tag: DW_TAG_pointer_type
594 // name: "__vtbl_ptr_type"
596 // tag: DW_TAG_subroutine_type
598 // tag: DW_TAG_base_type
603 MC_dwarf_attr_integrate_uint(&child, DW_AT_byte_size, 0);
604 member.type_id = MC_dwarf_at_type(&child);
606 if (dwarf_hasattr(&child, DW_AT_data_bit_offset))
607 xbt_die("Can't groke DW_AT_data_bit_offset.");
609 MC_dwarf_fill_member_location(type, &member, &child);
611 if (not member.type_id)
612 xbt_die("Missing type for member %s of <%" PRIx64 ">%s",
614 (uint64_t) type->id, type->name.c_str());
616 type->members.push_back(std::move(member));
621 /** \brief Create a MC type object from a DIE
623 * \param info current object info object
624 * \param die DIE (for a given type)
625 * \param unit compilation unit of the current DIE
626 * \return MC representation of the type
628 static simgrid::mc::Type MC_dwarf_die_to_type(
629 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
630 Dwarf_Die * unit, simgrid::mc::Frame* frame,
633 simgrid::mc::Type type;
634 type.type = dwarf_tag(die);
635 type.name = std::string();
636 type.element_count = -1;
639 type.id = dwarf_dieoffset(die);
641 const char *prefix = "";
643 case DW_TAG_structure_type:
646 case DW_TAG_union_type:
649 case DW_TAG_class_type:
656 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
657 if (name != nullptr) {
659 type.name = simgrid::xbt::string_printf("%s%s::%s", prefix, ns, name);
661 type.name = simgrid::xbt::string_printf("%s%s", prefix, name);
664 type.type_id = MC_dwarf_at_type(die);
666 // Some compilers do not emit DW_AT_byte_size for pointer_type,
667 // so we fill this. We currently assume that the model-checked process is in
668 // the same architecture..
669 if (type.type == DW_TAG_pointer_type)
670 type.byte_size = sizeof(void*);
672 // Computation of the byte_size
673 if (dwarf_hasattr_integrate(die, DW_AT_byte_size))
674 type.byte_size = MC_dwarf_attr_integrate_uint(die, DW_AT_byte_size, 0);
675 else if (type.type == DW_TAG_array_type
676 || type.type == DW_TAG_structure_type
677 || type.type == DW_TAG_class_type) {
679 if (dwarf_aggregate_size(die, &size) == 0)
680 type.byte_size = size;
684 case DW_TAG_array_type:
685 type.element_count = MC_dwarf_array_element_count(die, unit);
686 // TODO, handle DW_byte_stride and (not) DW_bit_stride
689 case DW_TAG_pointer_type:
690 case DW_TAG_reference_type:
691 case DW_TAG_rvalue_reference_type:
694 case DW_TAG_structure_type:
695 case DW_TAG_union_type:
696 case DW_TAG_class_type:
697 MC_dwarf_add_members(info, die, unit, &type);
698 MC_dwarf_handle_children(info, die, unit, frame,
699 ns ? simgrid::xbt::string_printf("%s::%s", ns, name).c_str() : type.name.c_str());
703 XBT_DEBUG("Unhandled type: %d (%s)", type.type, simgrid::dwarf::tagname(type.type));
710 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
711 Dwarf_Die * unit, simgrid::mc::Frame* frame,
714 simgrid::mc::Type type = MC_dwarf_die_to_type(info, die, unit, frame, ns);
715 auto& t = (info->types[type.id] = std::move(type));
716 if (not t.name.empty() && type.byte_size != 0)
717 info->full_types_by_name[t.name] = &t;
720 static int mc_anonymous_variable_index = 0;
722 static std::unique_ptr<simgrid::mc::Variable> MC_die_to_variable(
723 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
724 Dwarf_Die * unit, simgrid::mc::Frame* frame,
727 // Skip declarations:
728 if (MC_dwarf_attr_flag(die, DW_AT_declaration, false))
731 // Skip compile time constants:
732 if (dwarf_hasattr(die, DW_AT_const_value))
735 Dwarf_Attribute attr_location;
736 if (dwarf_attr(die, DW_AT_location, &attr_location) == nullptr)
737 // No location: do not add it ?
740 std::unique_ptr<simgrid::mc::Variable> variable =
741 std::unique_ptr<simgrid::mc::Variable>(new simgrid::mc::Variable());
742 variable->id = dwarf_dieoffset(die);
743 variable->global = frame == nullptr; // Can be override base on DW_AT_location
744 variable->object_info = info;
746 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
748 variable->name = name;
749 variable->type_id = MC_dwarf_at_type(die);
751 int form = dwarf_whatform(&attr_location);
752 simgrid::dwarf::FormClass form_class;
753 if (form == DW_FORM_sec_offset)
754 form_class = simgrid::dwarf::FormClass::Constant;
756 form_class = simgrid::dwarf::classify_form(form);
757 switch (form_class) {
758 case simgrid::dwarf::FormClass::ExprLoc:
759 case simgrid::dwarf::FormClass::Block:
760 // Location expression:
764 if (dwarf_getlocation(&attr_location, &expr, &len)) {
766 "Could not read location expression in DW_AT_location "
767 "of variable <%" PRIx64 ">%s",
768 (uint64_t) variable->id,
769 variable->name.c_str());
772 if (len == 1 && expr[0].atom == DW_OP_addr) {
773 variable->global = true;
774 uintptr_t offset = (uintptr_t) expr[0].number;
775 uintptr_t base = (uintptr_t) info->base_address();
776 variable->address = (void *) (base + offset);
778 variable->location_list = {
779 simgrid::dwarf::LocationListEntry(simgrid::dwarf::DwarfExpression(expr, expr + len))};
784 case simgrid::dwarf::FormClass::LocListPtr:
785 case simgrid::dwarf::FormClass::Constant:
786 // Reference to location list:
787 variable->location_list = simgrid::dwarf::location_list(
788 *info, attr_location);
792 xbt_die("Unexpected form 0x%x (%i), class 0x%x (%i) list for location in <%" PRIx64 ">%s", (unsigned)form, form,
793 (unsigned)form_class, (int)form_class, (uint64_t)variable->id, variable->name.c_str());
796 // Handle start_scope:
797 if (dwarf_hasattr(die, DW_AT_start_scope)) {
798 Dwarf_Attribute attr;
799 dwarf_attr(die, DW_AT_start_scope, &attr);
800 int form = dwarf_whatform(&attr);
801 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
802 if (form_class == simgrid::dwarf::FormClass::Constant) {
804 variable->start_scope = dwarf_formudata(&attr, &value) == 0 ? (size_t)value : 0;
806 // TODO: FormClass::RangeListPtr
807 xbt_die("Unhandled form 0x%x, class 0x%X for DW_AT_start_scope of variable %s", (unsigned)form,
808 (unsigned)form_class, name == nullptr ? "?" : name);
812 if (ns && variable->global)
814 std::string(ns) + "::" + variable->name;
816 // The current code needs a variable name,
817 // generate a fake one:
818 if (variable->name.empty()) {
819 variable->name = "@anonymous#" + std::to_string(mc_anonymous_variable_index);
820 mc_anonymous_variable_index++;
825 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
826 Dwarf_Die * unit, simgrid::mc::Frame* frame,
829 std::unique_ptr<simgrid::mc::Variable> variable =
830 MC_die_to_variable(info, die, unit, frame, ns);
833 // Those arrays are sorted later:
834 if (variable->global)
835 info->global_variables.push_back(std::move(*variable));
836 else if (frame != nullptr)
837 frame->variables.push_back(std::move(*variable));
839 xbt_die("No frame for this local variable");
842 static void MC_dwarf_handle_scope_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
843 Dwarf_Die * unit, simgrid::mc::Frame* parent_frame,
846 // TODO, handle DW_TAG_type/DW_TAG_location for DW_TAG_with_stmt
847 int tag = dwarf_tag(die);
848 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
850 // (Template) Subprogram declaration:
851 if (klass == simgrid::dwarf::TagClass::Subprogram
852 && MC_dwarf_attr_flag(die, DW_AT_declaration, false))
855 if (klass == simgrid::dwarf::TagClass::Scope)
856 xbt_assert(parent_frame, "No parent scope for this scope");
858 simgrid::mc::Frame frame;
860 frame.id = dwarf_dieoffset(die);
861 frame.object_info = info;
863 if (klass == simgrid::dwarf::TagClass::Subprogram) {
864 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
866 frame.name = std::string(ns) + "::" + name;
871 frame.abstract_origin_id =
872 MC_dwarf_attr_dieoffset(die, DW_AT_abstract_origin);
874 // This is the base address for DWARF addresses.
875 // Relocated addresses are offset from this base address.
876 // See DWARF4 spec 7.5
877 std::uint64_t base = (std::uint64_t) info->base_address();
879 // TODO, support DW_AT_ranges
880 uint64_t low_pc = MC_dwarf_attr_integrate_addr(die, DW_AT_low_pc);
881 frame.range.begin() = low_pc ? (std::uint64_t) base + low_pc : 0;
884 Dwarf_Attribute attr;
885 if (not dwarf_attr_integrate(die, DW_AT_high_pc, &attr))
886 xbt_die("Missing DW_AT_high_pc matching with DW_AT_low_pc");
891 switch (simgrid::dwarf::classify_form(dwarf_whatform(&attr))) {
893 // DW_AT_high_pc if an offset from the low_pc:
894 case simgrid::dwarf::FormClass::Constant:
896 if (dwarf_formsdata(&attr, &offset) != 0)
897 xbt_die("Could not read constant");
898 frame.range.end() = frame.range.begin() + offset;
901 // DW_AT_high_pc is a relocatable address:
902 case simgrid::dwarf::FormClass::Address:
903 if (dwarf_formaddr(&attr, &high_pc) != 0)
904 xbt_die("Could not read address");
905 frame.range.end() = base + high_pc;
909 xbt_die("Unexpected class for DW_AT_high_pc");
914 if (klass == simgrid::dwarf::TagClass::Subprogram) {
915 Dwarf_Attribute attr_frame_base;
916 if (dwarf_attr_integrate(die, DW_AT_frame_base, &attr_frame_base))
917 frame.frame_base_location = simgrid::dwarf::location_list(*info,
922 MC_dwarf_handle_children(info, die, unit, &frame, ns);
924 // We sort them in order to have an (somewhat) efficient by name
926 boost::range::sort(frame.variables, MC_compare_variable);
929 if (klass == simgrid::dwarf::TagClass::Subprogram)
930 info->subprograms[frame.id] = std::move(frame);
931 else if (klass == simgrid::dwarf::TagClass::Scope)
932 parent_frame->scopes.push_back(std::move(frame));
935 static void mc_dwarf_handle_namespace_die(simgrid::mc::ObjectInformation* info,
936 Dwarf_Die * die, Dwarf_Die * unit,
937 simgrid::mc::Frame* frame,
940 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
942 xbt_die("Unexpected namespace in a subprogram");
943 char *new_ns = ns == nullptr ? xbt_strdup(name)
944 : bprintf("%s::%s", ns, name);
945 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
949 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
950 Dwarf_Die * unit, simgrid::mc::Frame* frame,
953 // For each child DIE:
956 for (res = dwarf_child(die, &child); res == 0;
957 res = dwarf_siblingof(&child, &child))
958 MC_dwarf_handle_die(info, &child, unit, frame, ns);
961 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
962 Dwarf_Die * unit, simgrid::mc::Frame* frame,
965 int tag = dwarf_tag(die);
966 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
970 case simgrid::dwarf::TagClass::Type:
971 MC_dwarf_handle_type_die(info, die, unit, frame, ns);
974 // Subprogram or scope:
975 case simgrid::dwarf::TagClass::Subprogram:
976 case simgrid::dwarf::TagClass::Scope:
977 MC_dwarf_handle_scope_die(info, die, unit, frame, ns);
981 case simgrid::dwarf::TagClass::Variable:
982 MC_dwarf_handle_variable_die(info, die, unit, frame, ns);
985 case simgrid::dwarf::TagClass::Namespace:
986 mc_dwarf_handle_namespace_die(info, die, unit, frame, ns);
996 Elf64_Half get_type(Elf* elf)
998 Elf64_Ehdr* ehdr64 = elf64_getehdr(elf);
1000 return ehdr64->e_type;
1001 Elf32_Ehdr* ehdr32 = elf32_getehdr(elf);
1003 return ehdr32->e_type;
1004 xbt_die("Could not get ELF heeader");
1008 void read_dwarf_info(simgrid::mc::ObjectInformation* info, Dwarf* dwarf)
1010 // For each compilation unit:
1011 Dwarf_Off offset = 0;
1012 Dwarf_Off next_offset = 0;
1015 while (dwarf_nextcu(dwarf, offset, &next_offset, &length, nullptr, nullptr, nullptr) ==
1018 if (dwarf_offdie(dwarf, offset + length, &unit_die) != nullptr)
1019 MC_dwarf_handle_children(info, &unit_die, &unit_die, nullptr, nullptr);
1020 offset = next_offset;
1024 /** Get the build-id (NT_GNU_BUILD_ID) from the ELF file
1026 * This build-id may is used to locate an external debug (DWARF) file
1027 * for this ELF file.
1029 * @param elf libelf handle for an ELF file
1030 * @return build-id for this ELF file (or an empty vector if none is found)
1033 std::vector<char> get_build_id(Elf* elf)
1036 // Summary: the GNU build ID is stored in a ("GNU, NT_GNU_BUILD_ID) note
1037 // found in a PT_NOTE entry in the program header table.
1040 if (elf_getphdrnum (elf, &phnum) != 0)
1041 xbt_die("Could not read program headers");
1043 // Iterate over the program headers and find the PT_NOTE ones:
1044 for (size_t i = 0; i < phnum; ++i) {
1045 GElf_Phdr phdr_temp;
1046 GElf_Phdr *phdr = gelf_getphdr(elf, i, &phdr_temp);
1047 if (phdr->p_type != PT_NOTE)
1050 Elf_Data* data = elf_getdata_rawchunk(elf, phdr->p_offset, phdr->p_filesz, ELF_T_NHDR);
1052 // Iterate over the notes and find the NT_GNU_BUILD_ID one:
1054 while (pos < data->d_size) {
1056 // Location of the name within Elf_Data:
1059 pos = gelf_getnote(data, pos, &nhdr, &name_pos, &desc_pos);
1060 // A build ID note is identified by the pair ("GNU", NT_GNU_BUILD_ID)
1061 // (a namespace and a type within this namespace):
1062 if (nhdr.n_type == NT_GNU_BUILD_ID
1063 && nhdr.n_namesz == sizeof("GNU")
1064 && memcmp((char*) data->d_buf + name_pos, "GNU", sizeof("GNU")) == 0) {
1065 XBT_DEBUG("Found GNU/NT_GNU_BUILD_ID note");
1066 char* start = (char*) data->d_buf + desc_pos;
1067 char* end = (char*) start + nhdr.n_descsz;
1068 return std::vector<char>(start, end);
1074 return std::vector<char>();
1077 static char hexdigits[16] = {
1078 '0', '1', '2', '3', '4', '5', '6', '7', '8', '9',
1079 'a', 'b', 'c', 'd', 'e', 'f'
1082 /** Binary data to hexadecimal */
1084 std::array<char, 2> to_hex(std::uint8_t byte)
1086 // Horrid double braces!
1087 // Apparently, this is needed in C++11 (not in C++14).
1088 return { { hexdigits[byte >> 4], hexdigits[byte & 0xF] } };
1091 /** Binary data to hexadecimal */
1093 std::string to_hex(const char* data, std::size_t count)
1096 res.resize(2*count);
1097 for (std::size_t i = 0; i < count; i++) {
1098 std::array<char, 2> hex_byte = to_hex(data[i]);
1099 for (int j = 0; j < 2; ++j)
1100 res[2 * i + j] = hex_byte[j];
1105 /** Binary data to hexadecimal */
1107 std::string to_hex(std::vector<char> const& data)
1109 return to_hex(data.data(), data.size());
1112 /** Base directories for external debug files */
1114 const char* debug_paths[] = {
1116 "/usr/local/lib/debug/",
1119 /** Locate an external debug file from the NT_GNU_BUILD_ID
1121 * This is one of the mechanisms used for
1122 * [separate debug files](https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html).
1125 // /usr/lib/debug/.build-id/0b/dc77f1c29aea2b14ff5acd9a19ab3175ffdeae.debug
1127 std::string find_by_build_id(std::vector<char> id)
1129 std::string filename;
1130 std::string hex = to_hex(id);
1131 for (const char* const& debug_path : debug_paths) {
1133 filename = std::string(debug_path) + ".build-id/"
1134 + to_hex(id.data(), 1) + '/'
1135 + to_hex(id.data() + 1, id.size() - 1) + ".debug";
1136 XBT_DEBUG("Checking debug file: %s", filename.c_str());
1137 if (access(filename.c_str(), F_OK) == 0) {
1138 XBT_DEBUG("Found debug file: %s\n", hex.c_str());
1142 XBT_DEBUG("Not debuf info found for build ID %s\n", hex.data());
1143 return std::string();
1146 /** \brief Populate the debugging informations of the given ELF object
1148 * Read the DWARf information of the EFFL object and populate the
1149 * lists of types, variables, functions.
1152 void MC_load_dwarf(simgrid::mc::ObjectInformation* info)
1154 if (elf_version(EV_CURRENT) == EV_NONE)
1155 xbt_die("libelf initialization error");
1157 // Open the ELF file:
1158 int fd = open(info->file_name.c_str(), O_RDONLY);
1160 xbt_die("Could not open file %s", info->file_name.c_str());
1161 Elf* elf = elf_begin(fd, ELF_C_READ, nullptr);
1163 xbt_die("Not an ELF file");
1164 Elf_Kind kind = elf_kind(elf);
1165 if (kind != ELF_K_ELF)
1166 xbt_die("Not an ELF file");
1168 // Remember if this is a `ET_EXEC` (fixed location) or `ET_DYN`:
1169 Elf64_Half type = get_type(elf);
1170 if (type == ET_EXEC)
1171 info->flags |= simgrid::mc::ObjectInformation::Executable;
1173 // Read DWARF debug information in the file:
1174 Dwarf* dwarf = dwarf_begin_elf (elf, DWARF_C_READ, nullptr);
1175 if (dwarf != nullptr) {
1176 read_dwarf_info(info, dwarf);
1184 // If there was no DWARF in the file, try to find it in a separate file.
1185 // Different methods might be used to store the DWARF informations:
1186 // * GNU NT_GNU_BUILD_ID
1188 // See https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html
1189 // for reference of what we are doing.
1191 // Try with NT_GNU_BUILD_ID: we find the build ID in the ELF file and then
1192 // use this ID to find the file in some known locations in the filesystem.
1193 std::vector<char> build_id = get_build_id(elf);
1194 if (not build_id.empty()) {
1198 // Find the debug file using the build id:
1199 std::string debug_file = find_by_build_id(build_id);
1200 if (debug_file.empty()) {
1201 std::string hex = to_hex(build_id);
1202 xbt_die("Missing debug info for %s with build-id %s\n"
1203 "You might want to install the suitable debugging package.\n",
1204 info->file_name.c_str(), hex.c_str());
1207 // Load the DWARF info from this file:
1208 XBT_DEBUG("Load DWARF for %s from %s",
1209 info->file_name.c_str(), debug_file.c_str());
1210 fd = open(debug_file.c_str(), O_RDONLY);
1212 xbt_die("Could not open file %s", debug_file.c_str());
1213 Dwarf* dwarf = dwarf_begin(fd, DWARF_C_READ);
1214 if (dwarf == nullptr)
1215 xbt_die("No DWARF info in %s for %s",
1216 debug_file.c_str(), info->file_name.c_str());
1217 read_dwarf_info(info, dwarf);
1223 // TODO, try to find DWARF info using .gnu_debuglink.
1227 xbt_die("Debugging information not found for %s\n"
1228 "Try recompiling with -g\n",
1229 info->file_name.c_str());
1232 // ***** Functions index
1234 static int MC_compare_frame_index_items(simgrid::mc::FunctionIndexEntry* a,
1235 simgrid::mc::FunctionIndexEntry* b)
1237 if (a->low_pc < b->low_pc)
1239 else if (a->low_pc == b->low_pc)
1245 static void MC_make_functions_index(simgrid::mc::ObjectInformation* info)
1247 info->functions_index.clear();
1249 for (auto& e : info->subprograms) {
1250 if (e.second.range.begin() == 0)
1252 simgrid::mc::FunctionIndexEntry entry;
1253 entry.low_pc = (void*) e.second.range.begin();
1254 entry.function = &e.second;
1255 info->functions_index.push_back(entry);
1258 info->functions_index.shrink_to_fit();
1260 // Sort the array by low_pc:
1261 boost::range::sort(info->functions_index,
1262 [](simgrid::mc::FunctionIndexEntry const& a,
1263 simgrid::mc::FunctionIndexEntry const& b)
1265 return a.low_pc < b.low_pc;
1269 static void MC_post_process_variables(simgrid::mc::ObjectInformation* info)
1271 // Someone needs this to be sorted but who?
1272 boost::range::sort(info->global_variables, MC_compare_variable);
1274 for (simgrid::mc::Variable& variable : info->global_variables)
1275 if (variable.type_id)
1276 variable.type = simgrid::util::find_map_ptr(
1277 info->types, variable.type_id);
1280 static void mc_post_process_scope(simgrid::mc::ObjectInformation* info, simgrid::mc::Frame* scope)
1283 if (scope->tag == DW_TAG_inlined_subroutine) {
1284 // Attach correct namespaced name in inlined subroutine:
1285 auto i = info->subprograms.find(scope->abstract_origin_id);
1286 xbt_assert(i != info->subprograms.end(),
1287 "Could not lookup abstract origin %" PRIx64,
1288 (std::uint64_t) scope->abstract_origin_id);
1289 scope->name = i->second.name;
1293 for (simgrid::mc::Variable& variable : scope->variables)
1294 if (variable.type_id)
1295 variable.type = simgrid::util::find_map_ptr(
1296 info->types, variable.type_id);
1298 // Recursive post-processing of nested-scopes:
1299 for (simgrid::mc::Frame& nested_scope : scope->scopes)
1300 mc_post_process_scope(info, &nested_scope);
1304 simgrid::mc::Type* MC_resolve_type(
1305 simgrid::mc::ObjectInformation* info, unsigned type_id)
1309 simgrid::mc::Type* type = simgrid::util::find_map_ptr(info->types, type_id);
1310 if (type == nullptr)
1313 // We already have the information on the type:
1314 if (type->byte_size != 0)
1317 // Don't have a name, we can't find a more complete version:
1318 if (type->name.empty())
1321 // Try to find a more complete description of the type:
1322 // We need to fix in order to support C++.
1323 simgrid::mc::Type** subtype = simgrid::util::find_map_ptr(
1324 info->full_types_by_name, type->name);
1330 static void MC_post_process_types(simgrid::mc::ObjectInformation* info)
1332 // Lookup "subtype" field:
1333 for (auto& i : info->types) {
1334 i.second.subtype = MC_resolve_type(info, i.second.type_id);
1335 for (simgrid::mc::Member& member : i.second.members)
1336 member.type = MC_resolve_type(info, member.type_id);
1343 /** \brief Finds informations about a given shared object/executable */
1344 std::shared_ptr<simgrid::mc::ObjectInformation> createObjectInformation(
1345 std::vector<simgrid::xbt::VmMap> const& maps, const char *name)
1347 std::shared_ptr<simgrid::mc::ObjectInformation> result =
1348 std::make_shared<simgrid::mc::ObjectInformation>();
1349 result->file_name = name;
1350 simgrid::mc::find_object_address(maps, result.get());
1351 MC_load_dwarf(result.get());
1352 MC_post_process_variables(result.get());
1353 MC_post_process_types(result.get());
1354 for (auto& entry : result.get()->subprograms)
1355 mc_post_process_scope(result.get(), &entry.second);
1356 MC_make_functions_index(result.get());
1360 /*************************************************************************/
1362 void postProcessObjectInformation(simgrid::mc::RemoteClient* process, simgrid::mc::ObjectInformation* info)
1364 for (auto& i : info->types) {
1366 simgrid::mc::Type* type = &(i.second);
1367 simgrid::mc::Type* subtype = type;
1368 while (subtype->type == DW_TAG_typedef
1369 || subtype->type == DW_TAG_volatile_type
1370 || subtype->type == DW_TAG_const_type)
1371 if (subtype->subtype)
1372 subtype = subtype->subtype;
1376 // Resolve full_type:
1377 if (not subtype->name.empty() && subtype->byte_size == 0)
1378 for (auto const& object_info : process->object_infos) {
1379 auto i = object_info->full_types_by_name.find(subtype->name);
1380 if (i != object_info->full_types_by_name.end() && not i->second->name.empty() && i->second->byte_size) {
1381 type->full_type = i->second;
1385 else type->full_type = subtype;
1396 /** Convert a DWARF register into a libunwind register
1398 * DWARF and libunwind does not use the same convention for numbering the
1399 * registers on some architectures. The function makes the necessary
1402 int dwarf_register_to_libunwind(int dwarf_register)
1404 #if defined(__x86_64__)
1405 // It seems for this arch, DWARF and libunwind agree in the numbering:
1406 return dwarf_register;
1407 #elif defined(__i386__)
1408 // Couldn't find the authoritative source of information for this.
1409 // This is inspired from http://source.winehq.org/source/dlls/dbghelp/cpu_i386.c#L517.
1410 switch (dwarf_register) {
1430 return UNW_X86_EFLAGS;
1460 xbt_die("Bad/unknown register number.");
1463 #error This architecture is not supported yet for DWARF expression evaluation.