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/sysdep.h"
26 #include <simgrid_config.h>
28 #include "src/mc/mc_private.h"
29 #include "src/mc/mc_dwarf.hpp"
31 #include "src/mc/ObjectInformation.hpp"
32 #include "src/mc/Variable.hpp"
33 #include "src/mc/remote/RemoteClient.hpp"
35 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_dwarf, mc, "DWARF processing");
37 /** \brief The default DW_TAG_lower_bound for a given DW_AT_language.
39 * The default for a given language is defined in the DWARF spec.
41 * \param language constant as defined by the DWARf spec
43 static uint64_t MC_dwarf_default_lower_bound(int lang);
45 /** \brief Computes the the element_count of a DW_TAG_enumeration_type DIE
47 * This is the number of elements in a given array dimension.
49 * A reference of the compilation unit (DW_TAG_compile_unit) is
50 * needed because the default lower bound (when there is no DW_AT_lower_bound)
51 * depends of the language of the compilation unit (DW_AT_language).
53 * \param die DIE for the DW_TAG_enumeration_type or DW_TAG_subrange_type
54 * \param unit DIE of the DW_TAG_compile_unit
56 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die* die, Dwarf_Die* unit);
58 /** \brief Computes the number of elements of a given DW_TAG_array_type.
60 * \param die DIE for the DW_TAG_array_type
62 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit);
64 /** \brief Process a DIE
66 * \param info the resulting object fot the library/binary file (output)
67 * \param die the current DIE
68 * \param unit the DIE of the compile unit of the current DIE
69 * \param frame containing frame if any
71 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
72 Dwarf_Die * unit, simgrid::mc::Frame* frame,
75 /** \brief Process a type DIE
77 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
78 Dwarf_Die * unit, simgrid::mc::Frame* frame,
81 /** \brief Calls MC_dwarf_handle_die on all children of the given die
83 * \param info the resulting object fot the library/binary file (output)
84 * \param die the current DIE
85 * \param unit the DIE of the compile unit of the current DIE
86 * \param frame containing frame if any
88 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
89 Dwarf_Die * unit, simgrid::mc::Frame* frame,
92 /** \brief Handle a variable (DW_TAG_variable or other)
94 * \param info the resulting object fot the library/binary file (output)
95 * \param die the current DIE
96 * \param unit the DIE of the compile unit of the current DIE
97 * \param frame containing frame if any
99 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
100 Dwarf_Die * unit, simgrid::mc::Frame* frame,
103 /** \brief Get the DW_TAG_type of the DIE
106 * \return DW_TAG_type attribute as a new string (nullptr if none)
108 static std::uint64_t MC_dwarf_at_type(Dwarf_Die * die);
113 enum class TagClass {
122 /*** Class of forms defined in the DWARF standard */
123 enum class FormClass {
125 Address, // Location in the program's address space
126 Block, // Arbitrary block of bytes
129 Flag, // Boolean value
130 Reference, // Reference to another DIE
131 ExprLoc, // DWARF expression/location description
139 TagClass classify_tag(int tag)
143 case DW_TAG_array_type:
144 case DW_TAG_class_type:
145 case DW_TAG_enumeration_type:
147 case DW_TAG_pointer_type:
148 case DW_TAG_reference_type:
149 case DW_TAG_rvalue_reference_type:
150 case DW_TAG_string_type:
151 case DW_TAG_structure_type:
152 case DW_TAG_subroutine_type:
153 case DW_TAG_union_type:
154 case DW_TAG_ptr_to_member_type:
155 case DW_TAG_set_type:
156 case DW_TAG_subrange_type:
157 case DW_TAG_base_type:
158 case DW_TAG_const_type:
159 case DW_TAG_file_type:
160 case DW_TAG_packed_type:
161 case DW_TAG_volatile_type:
162 case DW_TAG_restrict_type:
163 case DW_TAG_interface_type:
164 case DW_TAG_unspecified_type:
165 case DW_TAG_shared_type:
166 return TagClass::Type;
168 case DW_TAG_subprogram:
169 return TagClass::Subprogram;
171 case DW_TAG_variable:
172 case DW_TAG_formal_parameter:
173 return TagClass::Variable;
175 case DW_TAG_lexical_block:
176 case DW_TAG_try_block:
177 case DW_TAG_catch_block:
178 case DW_TAG_inlined_subroutine:
179 case DW_TAG_with_stmt:
180 return TagClass::Scope;
182 case DW_TAG_namespace:
183 return TagClass::Namespace;
186 return TagClass::Unknown;
190 /** \brief Find the DWARF data class for a given DWARF data form
192 * This mapping is defined in the DWARF spec.
194 * \param form The form (values taken from the DWARF spec)
195 * \return An internal representation for the corresponding class
198 FormClass classify_form(int form)
202 return FormClass::Address;
207 return FormClass::Block;
214 return FormClass::Constant;
217 return FormClass::String;
218 case DW_FORM_ref_addr:
223 case DW_FORM_ref_udata:
224 return FormClass::Reference;
226 case DW_FORM_flag_present:
227 return FormClass::Flag;
228 case DW_FORM_exprloc:
229 return FormClass::ExprLoc;
233 return FormClass::Unknown;
237 /** \brief Get the name of the tag of a given DIE
240 * \return name of the tag of this DIE
243 const char *tagname(Dwarf_Die * die)
245 return simgrid::dwarf::tagname(dwarf_tag(die));
253 /** \brief Get an attribute of a given DIE as a string
256 * \param attribute attribute
257 * \return value of the given attribute of the given DIE
259 static const char *MC_dwarf_attr_integrate_string(Dwarf_Die * die,
262 Dwarf_Attribute attr;
263 if (not dwarf_attr_integrate(die, attribute, &attr))
266 return dwarf_formstring(&attr);
269 /** \brief Get the linkage name of a DIE.
271 * Use either DW_AT_linkage_name or DW_AT_MIPS_linkage_name.
272 * DW_AT_linkage_name is standardized since DWARF 4.
273 * Before this version of DWARF, the MIPS extensions
274 * DW_AT_MIPS_linkage_name is used (at least by GCC).
277 * \return linkage name of the given DIE (or nullptr)
279 static const char *MC_dwarf_at_linkage_name(Dwarf_Die * die)
281 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_linkage_name);
283 name = MC_dwarf_attr_integrate_string(die, DW_AT_MIPS_linkage_name);
287 static Dwarf_Off MC_dwarf_attr_dieoffset(Dwarf_Die * die, int attribute)
289 Dwarf_Attribute attr;
290 if (dwarf_hasattr_integrate(die, attribute) == 0)
292 dwarf_attr_integrate(die, attribute, &attr);
293 Dwarf_Die subtype_die;
294 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
295 xbt_die("Could not find DIE");
296 return dwarf_dieoffset(&subtype_die);
299 static Dwarf_Off MC_dwarf_attr_integrate_dieoffset(Dwarf_Die * die,
302 Dwarf_Attribute attr;
303 if (dwarf_hasattr_integrate(die, attribute) == 0)
305 dwarf_attr_integrate(die, DW_AT_type, &attr);
306 Dwarf_Die subtype_die;
307 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
308 xbt_die("Could not find DIE");
309 return dwarf_dieoffset(&subtype_die);
312 /** \brief Find the type/subtype (DW_AT_type) for a DIE
315 * \return DW_AT_type reference as a global offset in hexadecimal (or nullptr)
318 std::uint64_t MC_dwarf_at_type(Dwarf_Die * die)
320 return MC_dwarf_attr_integrate_dieoffset(die, DW_AT_type);
323 static uint64_t MC_dwarf_attr_integrate_addr(Dwarf_Die * die, int attribute)
325 Dwarf_Attribute attr;
326 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
329 if (dwarf_formaddr(&attr, &value) == 0)
330 return (uint64_t) value;
335 static uint64_t MC_dwarf_attr_integrate_uint(Dwarf_Die * die, int attribute,
336 uint64_t default_value)
338 Dwarf_Attribute attr;
339 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
340 return default_value;
342 return dwarf_formudata(dwarf_attr_integrate(die, attribute, &attr),
343 &value) == 0 ? (uint64_t) value : default_value;
346 static bool MC_dwarf_attr_flag(Dwarf_Die * die, int attribute, bool integrate)
348 Dwarf_Attribute attr;
349 if ((integrate ? dwarf_attr_integrate(die, attribute, &attr)
350 : dwarf_attr(die, attribute, &attr)) == 0)
354 if (dwarf_formflag(&attr, &result))
355 xbt_die("Unexpected form for attribute %s",
356 simgrid::dwarf::attrname(attribute));
360 /** @brief Find the default lower bound for a given language
362 * The default lower bound of an array (when DW_TAG_lower_bound
363 * is missing) depends on the language of the compilation unit.
365 * @param lang Language of the compilation unit (values defined in the DWARF spec)
366 * @return Default lower bound of an array in this compilation unit
368 static uint64_t MC_dwarf_default_lower_bound(int lang)
374 case DW_LANG_C_plus_plus:
378 case DW_LANG_ObjC_plus_plus:
384 case DW_LANG_Fortran77:
385 case DW_LANG_Fortran90:
386 case DW_LANG_Fortran95:
387 case DW_LANG_Modula2:
388 case DW_LANG_Pascal83:
390 case DW_LANG_Cobol74:
391 case DW_LANG_Cobol85:
394 xbt_die("No default DW_TAG_lower_bound for language %i and none given",
400 /** \brief Finds the number of elements in a DW_TAG_subrange_type or DW_TAG_enumeration_type DIE
403 * \param unit DIE of the compilation unit
404 * \return number of elements in the range
406 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die * die,
409 xbt_assert(dwarf_tag(die) == DW_TAG_enumeration_type
410 || dwarf_tag(die) == DW_TAG_subrange_type,
411 "MC_dwarf_subrange_element_count called with DIE of type %s",
412 simgrid::dwarf::tagname(die));
414 // Use DW_TAG_count if present:
415 if (dwarf_hasattr_integrate(die, DW_AT_count))
416 return MC_dwarf_attr_integrate_uint(die, DW_AT_count, 0);
417 // Otherwise compute DW_TAG_upper_bound-DW_TAG_lower_bound + 1:
419 if (not dwarf_hasattr_integrate(die, DW_AT_upper_bound))
420 // This is not really 0, but the code expects this (we do not know):
423 uint64_t upper_bound =
424 MC_dwarf_attr_integrate_uint(die, DW_AT_upper_bound, -1);
426 uint64_t lower_bound = 0;
427 if (dwarf_hasattr_integrate(die, DW_AT_lower_bound))
428 lower_bound = MC_dwarf_attr_integrate_uint(die, DW_AT_lower_bound, -1);
430 lower_bound = MC_dwarf_default_lower_bound(dwarf_srclang(unit));
431 return upper_bound - lower_bound + 1;
434 /** \brief Finds the number of elements in a array type (DW_TAG_array_type)
436 * The compilation unit might be needed because the default lower
437 * bound depends on the language of the compilation unit.
439 * \param die the DIE of the DW_TAG_array_type
440 * \param unit the DIE of the compilation unit
441 * \return number of elements in this array type
443 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit)
445 xbt_assert(dwarf_tag(die) == DW_TAG_array_type,
446 "MC_dwarf_array_element_count called with DIE of type %s",
447 simgrid::dwarf::tagname(die));
452 for (res = dwarf_child(die, &child); res == 0;
453 res = dwarf_siblingof(&child, &child)) {
454 int child_tag = dwarf_tag(&child);
455 if (child_tag == DW_TAG_subrange_type
456 || child_tag == DW_TAG_enumeration_type)
457 result *= MC_dwarf_subrange_element_count(&child, unit);
464 /** Sort the variable by name and address.
466 * We could use boost::container::flat_set instead.
468 static bool MC_compare_variable(
469 simgrid::mc::Variable const& a, simgrid::mc::Variable const& b)
471 int cmp = strcmp(a.name.c_str(), b.name.c_str());
477 return a.address < b.address;
480 // ***** simgrid::mc::Type*
482 /** \brief Initialize the location of a member of a type
483 * (DW_AT_data_member_location of a DW_TAG_member).
485 * \param type a type (struct, class)
486 * \param member the member of the type
487 * \param child DIE of the member (DW_TAG_member)
489 static void MC_dwarf_fill_member_location(
490 simgrid::mc::Type* type, simgrid::mc::Member* member, Dwarf_Die * child)
492 if (dwarf_hasattr(child, DW_AT_data_bit_offset))
493 xbt_die("Can't groke DW_AT_data_bit_offset.");
495 if (not dwarf_hasattr_integrate(child, DW_AT_data_member_location)) {
496 if (type->type == DW_TAG_union_type)
499 ("Missing DW_AT_data_member_location field in DW_TAG_member %s of type <%"
500 PRIx64 ">%s", member->name.c_str(),
501 (uint64_t) type->id, type->name.c_str());
504 Dwarf_Attribute attr;
505 dwarf_attr_integrate(child, DW_AT_data_member_location, &attr);
506 int form = dwarf_whatform(&attr);
507 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
508 switch (form_class) {
509 case simgrid::dwarf::FormClass::ExprLoc:
510 case simgrid::dwarf::FormClass::Block:
511 // Location expression:
515 if (dwarf_getlocation(&attr, &expr, &len))
517 ("Could not read location expression DW_AT_data_member_location in DW_TAG_member %s of type <%"
518 PRIx64 ">%s", MC_dwarf_attr_integrate_string(child, DW_AT_name),
519 (uint64_t) type->id, type->name.c_str());
520 member->location_expression = simgrid::dwarf::DwarfExpression(expr, expr+len);
523 case simgrid::dwarf::FormClass::Constant:
524 // Offset from the base address of the object:
527 if (not dwarf_formudata(&attr, &offset))
528 member->offset(offset);
530 xbt_die("Cannot get %s location <%" PRIx64 ">%s",
531 MC_dwarf_attr_integrate_string(child, DW_AT_name),
532 (uint64_t) type->id, type->name.c_str());
535 case simgrid::dwarf::FormClass::LocListPtr:
536 // Reference to a location list:
538 case simgrid::dwarf::FormClass::Reference:
539 // It's supposed to be possible in DWARF2 but I couldn't find its semantic
542 xbt_die("Can't handle form class (%d) / form 0x%x as DW_AT_member_location", (int)form_class, (unsigned)form);
547 /** \brief Populate the list of members of a type
549 * \param info ELF object containing the type DIE
550 * \param die DIE of the type
551 * \param unit DIE of the compilation unit containing the type DIE
552 * \param type the type
554 static void MC_dwarf_add_members(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
555 Dwarf_Die * unit, simgrid::mc::Type* type)
559 xbt_assert(type->members.empty());
560 for (res = dwarf_child(die, &child); res == 0;
561 res = dwarf_siblingof(&child, &child)) {
562 int tag = dwarf_tag(&child);
563 if (tag == DW_TAG_member || tag == DW_TAG_inheritance) {
565 // Skip declarations:
566 if (MC_dwarf_attr_flag(&child, DW_AT_declaration, false))
569 // Skip compile time constants:
570 if (dwarf_hasattr(&child, DW_AT_const_value))
573 // TODO, we should use another type (because is is not a type but a member)
574 simgrid::mc::Member member;
575 if (tag == DW_TAG_inheritance)
576 member.flags |= simgrid::mc::Member::INHERITANCE_FLAG;
578 const char *name = MC_dwarf_attr_integrate_string(&child, DW_AT_name);
581 // Those base names are used by GCC and clang for virtual table pointers
582 // respectively ("__vptr$ClassName", "__vptr.ClassName"):
583 if (boost::algorithm::starts_with(member.name, "__vptr$") ||
584 boost::algorithm::starts_with(member.name, "__vptr."))
585 member.flags |= simgrid::mc::Member::VIRTUAL_POINTER_FLAG;
586 // A cleaner solution would be to check against the type:
588 // tag: DW_TAG_member
591 // # Type for a pointer to a vtable
592 // tag: DW_TAG_pointer_type
594 // # Type for a vtable:
595 // tag: DW_TAG_pointer_type
596 // name: "__vtbl_ptr_type"
598 // tag: DW_TAG_subroutine_type
600 // tag: DW_TAG_base_type
605 MC_dwarf_attr_integrate_uint(&child, DW_AT_byte_size, 0);
606 member.type_id = MC_dwarf_at_type(&child);
608 if (dwarf_hasattr(&child, DW_AT_data_bit_offset))
609 xbt_die("Can't groke DW_AT_data_bit_offset.");
611 MC_dwarf_fill_member_location(type, &member, &child);
613 if (not member.type_id)
614 xbt_die("Missing type for member %s of <%" PRIx64 ">%s",
616 (uint64_t) type->id, type->name.c_str());
618 type->members.push_back(std::move(member));
623 /** \brief Create a MC type object from a DIE
625 * \param info current object info object
626 * \param DIE (for a given type);
627 * \param unit compilation unit of the current DIE
628 * \return MC representation of the type
630 static simgrid::mc::Type MC_dwarf_die_to_type(
631 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
632 Dwarf_Die * unit, simgrid::mc::Frame* frame,
635 simgrid::mc::Type type;
636 type.type = dwarf_tag(die);
637 type.name = std::string();
638 type.element_count = -1;
641 type.id = dwarf_dieoffset(die);
643 const char *prefix = "";
645 case DW_TAG_structure_type:
648 case DW_TAG_union_type:
651 case DW_TAG_class_type:
658 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
659 if (name != nullptr) {
660 char* full_name = ns ? bprintf("%s%s::%s", prefix, ns, name) :
661 bprintf("%s%s", prefix, name);
662 type.name = std::string(full_name);
666 type.type_id = MC_dwarf_at_type(die);
668 // Some compilers do not emit DW_AT_byte_size for pointer_type,
669 // so we fill this. We currently assume that the model-checked process is in
670 // the same architecture..
671 if (type.type == DW_TAG_pointer_type)
672 type.byte_size = sizeof(void*);
674 // Computation of the byte_size;
675 if (dwarf_hasattr_integrate(die, DW_AT_byte_size))
676 type.byte_size = MC_dwarf_attr_integrate_uint(die, DW_AT_byte_size, 0);
677 else if (type.type == DW_TAG_array_type
678 || type.type == DW_TAG_structure_type
679 || type.type == DW_TAG_class_type) {
681 if (dwarf_aggregate_size(die, &size) == 0)
682 type.byte_size = size;
686 case DW_TAG_array_type:
687 type.element_count = MC_dwarf_array_element_count(die, unit);
688 // TODO, handle DW_byte_stride and (not) DW_bit_stride
691 case DW_TAG_pointer_type:
692 case DW_TAG_reference_type:
693 case DW_TAG_rvalue_reference_type:
696 case DW_TAG_structure_type:
697 case DW_TAG_union_type:
698 case DW_TAG_class_type:
699 MC_dwarf_add_members(info, die, unit, &type);
700 char *new_ns = ns == nullptr ? xbt_strdup(type.name.c_str())
701 : bprintf("%s::%s", ns, name);
702 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
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::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 switch (form_class) {
803 case simgrid::dwarf::FormClass::Constant:
806 variable->start_scope =
807 dwarf_formudata(&attr, &value) == 0 ? (size_t) value : 0;
811 case simgrid::dwarf::FormClass::RangeListPtr: // TODO
813 xbt_die("Unhandled form 0x%x, class 0x%X for DW_AT_start_scope of variable %s", (unsigned)form,
814 (unsigned)form_class, name == nullptr ? "?" : name);
818 if (ns && variable->global)
820 std::string(ns) + "::" + variable->name;
822 // The current code needs a variable name,
823 // generate a fake one:
824 if (variable->name.empty())
826 "@anonymous#" + std::to_string(mc_anonymous_variable_index++);
831 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
832 Dwarf_Die * unit, simgrid::mc::Frame* frame,
835 std::unique_ptr<simgrid::mc::Variable> variable =
836 MC_die_to_variable(info, die, unit, frame, ns);
839 // Those arrays are sorted later:
840 else if (variable->global)
841 info->global_variables.push_back(std::move(*variable));
842 else if (frame != nullptr)
843 frame->variables.push_back(std::move(*variable));
845 xbt_die("No frame for this local variable");
848 static void MC_dwarf_handle_scope_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
849 Dwarf_Die * unit, simgrid::mc::Frame* parent_frame,
852 // TODO, handle DW_TAG_type/DW_TAG_location for DW_TAG_with_stmt
853 int tag = dwarf_tag(die);
854 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
856 // (Template) Subprogram declaration:
857 if (klass == simgrid::dwarf::TagClass::Subprogram
858 && MC_dwarf_attr_flag(die, DW_AT_declaration, false))
861 if (klass == simgrid::dwarf::TagClass::Scope)
862 xbt_assert(parent_frame, "No parent scope for this scope");
864 simgrid::mc::Frame frame;
866 frame.id = dwarf_dieoffset(die);
867 frame.object_info = info;
869 if (klass == simgrid::dwarf::TagClass::Subprogram) {
870 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
872 frame.name = std::string(ns) + "::" + name;
877 frame.abstract_origin_id =
878 MC_dwarf_attr_dieoffset(die, DW_AT_abstract_origin);
880 // This is the base address for DWARF addresses.
881 // Relocated addresses are offset from this base address.
882 // See DWARF4 spec 7.5
883 std::uint64_t base = (std::uint64_t) info->base_address();
885 // TODO, support DW_AT_ranges
886 uint64_t low_pc = MC_dwarf_attr_integrate_addr(die, DW_AT_low_pc);
887 frame.range.begin() = low_pc ? (std::uint64_t) base + low_pc : 0;
890 Dwarf_Attribute attr;
891 if (not dwarf_attr_integrate(die, DW_AT_high_pc, &attr))
892 xbt_die("Missing DW_AT_high_pc matching with DW_AT_low_pc");
897 switch (simgrid::dwarf::classify_form(dwarf_whatform(&attr))) {
899 // DW_AT_high_pc if an offset from the low_pc:
900 case simgrid::dwarf::FormClass::Constant:
902 if (dwarf_formsdata(&attr, &offset) != 0)
903 xbt_die("Could not read constant");
904 frame.range.end() = frame.range.begin() + offset;
907 // DW_AT_high_pc is a relocatable address:
908 case simgrid::dwarf::FormClass::Address:
909 if (dwarf_formaddr(&attr, &high_pc) != 0)
910 xbt_die("Could not read address");
911 frame.range.end() = base + high_pc;
915 xbt_die("Unexpected class for DW_AT_high_pc");
920 if (klass == simgrid::dwarf::TagClass::Subprogram) {
921 Dwarf_Attribute attr_frame_base;
922 if (dwarf_attr_integrate(die, DW_AT_frame_base, &attr_frame_base))
923 frame.frame_base_location = simgrid::dwarf::location_list(*info,
928 MC_dwarf_handle_children(info, die, unit, &frame, ns);
930 // We sort them in order to have an (somewhat) efficient by name
932 boost::range::sort(frame.variables, MC_compare_variable);
935 if (klass == simgrid::dwarf::TagClass::Subprogram)
936 info->subprograms[frame.id] = std::move(frame);
937 else if (klass == simgrid::dwarf::TagClass::Scope)
938 parent_frame->scopes.push_back(std::move(frame));
941 static void mc_dwarf_handle_namespace_die(simgrid::mc::ObjectInformation* info,
942 Dwarf_Die * die, Dwarf_Die * unit,
943 simgrid::mc::Frame* frame,
946 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
948 xbt_die("Unexpected namespace in a subprogram");
949 char *new_ns = ns == nullptr ? xbt_strdup(name)
950 : bprintf("%s::%s", ns, name);
951 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
955 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
956 Dwarf_Die * unit, simgrid::mc::Frame* frame,
959 // For each child DIE:
962 for (res = dwarf_child(die, &child); res == 0;
963 res = dwarf_siblingof(&child, &child))
964 MC_dwarf_handle_die(info, &child, unit, frame, ns);
967 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
968 Dwarf_Die * unit, simgrid::mc::Frame* frame,
971 int tag = dwarf_tag(die);
972 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
976 case simgrid::dwarf::TagClass::Type:
977 MC_dwarf_handle_type_die(info, die, unit, frame, ns);
980 // Subprogram or scope:
981 case simgrid::dwarf::TagClass::Subprogram:
982 case simgrid::dwarf::TagClass::Scope:
983 MC_dwarf_handle_scope_die(info, die, unit, frame, ns);
987 case simgrid::dwarf::TagClass::Variable:
988 MC_dwarf_handle_variable_die(info, die, unit, frame, ns);
991 case simgrid::dwarf::TagClass::Namespace:
992 mc_dwarf_handle_namespace_die(info, die, unit, frame, ns);
1002 Elf64_Half get_type(Elf* elf)
1004 Elf64_Ehdr* ehdr64 = elf64_getehdr(elf);
1006 return ehdr64->e_type;
1007 Elf32_Ehdr* ehdr32 = elf32_getehdr(elf);
1009 return ehdr32->e_type;
1010 xbt_die("Could not get ELF heeader");
1014 void read_dwarf_info(simgrid::mc::ObjectInformation* info, Dwarf* dwarf)
1016 // For each compilation unit:
1017 Dwarf_Off offset = 0;
1018 Dwarf_Off next_offset = 0;
1021 while (dwarf_nextcu(dwarf, offset, &next_offset, &length, nullptr, nullptr, nullptr) ==
1024 if (dwarf_offdie(dwarf, offset + length, &unit_die) != nullptr)
1025 MC_dwarf_handle_children(info, &unit_die, &unit_die, nullptr, nullptr);
1026 offset = next_offset;
1030 /** Get the build-id (NT_GNU_BUILD_ID) from the ELF file
1032 * This build-id may is used to locate an external debug (DWARF) file
1033 * for this ELF file.
1035 * @param elf libelf handle for an ELF file
1036 * @return build-id for this ELF file (or an empty vector if none is found)
1039 std::vector<char> get_build_id(Elf* elf)
1042 // Summary: the GNU build ID is stored in a ("GNU, NT_GNU_BUILD_ID) note
1043 // found in a PT_NOTE entry in the program header table.
1046 if (elf_getphdrnum (elf, &phnum) != 0)
1047 xbt_die("Could not read program headers");
1049 // Iterate over the program headers and find the PT_NOTE ones:
1050 for (size_t i = 0; i < phnum; ++i) {
1051 GElf_Phdr phdr_temp;
1052 GElf_Phdr *phdr = gelf_getphdr(elf, i, &phdr_temp);
1053 if (phdr->p_type != PT_NOTE)
1056 Elf_Data* data = elf_getdata_rawchunk(elf, phdr->p_offset, phdr->p_filesz, ELF_T_NHDR);
1058 // Iterate over the notes and find the NT_GNU_BUILD_ID one:
1060 while (pos < data->d_size) {
1062 // Location of the name within Elf_Data:
1065 pos = gelf_getnote(data, pos, &nhdr, &name_pos, &desc_pos);
1066 // A build ID note is identified by the pair ("GNU", NT_GNU_BUILD_ID)
1067 // (a namespace and a type within this namespace):
1068 if (nhdr.n_type == NT_GNU_BUILD_ID
1069 && nhdr.n_namesz == sizeof("GNU")
1070 && memcmp((char*) data->d_buf + name_pos, "GNU", sizeof("GNU")) == 0) {
1071 XBT_DEBUG("Found GNU/NT_GNU_BUILD_ID note");
1072 char* start = (char*) data->d_buf + desc_pos;
1073 char* end = (char*) start + nhdr.n_descsz;
1074 return std::vector<char>(start, end);
1080 return std::vector<char>();
1083 static char hexdigits[16] = {
1084 '0', '1', '2', '3', '4', '5', '6', '7', '8', '9',
1085 'a', 'b', 'c', 'd', 'e', 'f'
1088 /** Binary data to hexadecimal */
1090 std::array<char, 2> to_hex(std::uint8_t byte)
1092 // Horrid double braces!
1093 // Apparently, this is needed in C++11 (not in C++14).
1094 return { { hexdigits[byte >> 4], hexdigits[byte & 0xF] } };
1097 /** Binary data to hexadecimal */
1099 std::string to_hex(const char* data, std::size_t count)
1102 res.resize(2*count);
1103 for (std::size_t i = 0; i < count; i++) {
1104 std::array<char, 2> hex_byte = to_hex(data[i]);
1105 for (int j = 0; j < 2; ++j)
1106 res[2 * i + j] = hex_byte[j];
1111 /** Binary data to hexadecimal */
1113 std::string to_hex(std::vector<char> const& data)
1115 return to_hex(data.data(), data.size());
1118 /** Base directories for external debug files */
1120 const char* debug_paths[] = {
1122 "/usr/local/lib/debug/",
1125 /** Locate an external debug file from the NT_GNU_BUILD_ID
1127 * This is one of the mechanisms used for
1128 * [separate debug files](https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html).
1131 // /usr/lib/debug/.build-id/0b/dc77f1c29aea2b14ff5acd9a19ab3175ffdeae.debug
1133 std::string find_by_build_id(std::vector<char> id)
1135 std::string filename;
1136 std::string hex = to_hex(id);
1137 for (const char* const& debug_path : debug_paths) {
1139 filename = std::string(debug_path) + ".build-id/"
1140 + to_hex(id.data(), 1) + '/'
1141 + to_hex(id.data() + 1, id.size() - 1) + ".debug";
1142 XBT_DEBUG("Checking debug file: %s", filename.c_str());
1143 if (access(filename.c_str(), F_OK) == 0) {
1144 XBT_DEBUG("Found debug file: %s\n", hex.c_str());
1148 XBT_DEBUG("Not debuf info found for build ID %s\n", hex.data());
1149 return std::string();
1152 /** \brief Populate the debugging informations of the given ELF object
1154 * Read the DWARf information of the EFFL object and populate the
1155 * lists of types, variables, functions.
1158 void MC_load_dwarf(simgrid::mc::ObjectInformation* info)
1160 if (elf_version(EV_CURRENT) == EV_NONE)
1161 xbt_die("libelf initialization error");
1163 // Open the ELF file:
1164 int fd = open(info->file_name.c_str(), O_RDONLY);
1166 xbt_die("Could not open file %s", info->file_name.c_str());
1167 Elf* elf = elf_begin(fd, ELF_C_READ, nullptr);
1169 xbt_die("Not an ELF file");
1170 Elf_Kind kind = elf_kind(elf);
1171 if (kind != ELF_K_ELF)
1172 xbt_die("Not an ELF file");
1174 // Remember if this is a `ET_EXEC` (fixed location) or `ET_DYN`:
1175 Elf64_Half type = get_type(elf);
1176 if (type == ET_EXEC)
1177 info->flags |= simgrid::mc::ObjectInformation::Executable;
1179 // Read DWARF debug information in the file:
1180 Dwarf* dwarf = dwarf_begin_elf (elf, DWARF_C_READ, nullptr);
1181 if (dwarf != nullptr) {
1182 read_dwarf_info(info, dwarf);
1190 // If there was no DWARF in the file, try to find it in a separate file.
1191 // Different methods might be used to store the DWARF informations:
1192 // * GNU NT_GNU_BUILD_ID;
1193 // * .gnu_debuglink.
1194 // See https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html
1195 // for reference of what we are doing.
1197 // Try with NT_GNU_BUILD_ID: we find the build ID in the ELF file and then
1198 // use this ID to find the file in some known locations in the filesystem.
1199 std::vector<char> build_id = get_build_id(elf);
1200 if (not build_id.empty()) {
1204 // Find the debug file using the build id:
1205 std::string debug_file = find_by_build_id(build_id);
1206 if (debug_file.empty()) {
1207 std::string hex = to_hex(build_id);
1208 xbt_die("Missing debug info for %s with build-id %s\n"
1209 "You might want to install the suitable debugging package.\n",
1210 info->file_name.c_str(), hex.c_str());
1213 // Load the DWARF info from this file:
1214 XBT_DEBUG("Load DWARF for %s from %s",
1215 info->file_name.c_str(), debug_file.c_str());
1216 fd = open(debug_file.c_str(), O_RDONLY);
1218 xbt_die("Could not open file %s", debug_file.c_str());
1219 Dwarf* dwarf = dwarf_begin(fd, DWARF_C_READ);
1220 if (dwarf == nullptr)
1221 xbt_die("No DWARF info in %s for %s",
1222 debug_file.c_str(), info->file_name.c_str());
1223 read_dwarf_info(info, dwarf);
1229 // TODO, try to find DWARF info using .gnu_debuglink.
1233 xbt_die("Debugging information not found for %s\n"
1234 "Try recompiling with -g\n",
1235 info->file_name.c_str());
1238 // ***** Functions index
1240 static int MC_compare_frame_index_items(simgrid::mc::FunctionIndexEntry* a,
1241 simgrid::mc::FunctionIndexEntry* b)
1243 if (a->low_pc < b->low_pc)
1245 else if (a->low_pc == b->low_pc)
1251 static void MC_make_functions_index(simgrid::mc::ObjectInformation* info)
1253 info->functions_index.clear();
1255 for (auto& e : info->subprograms) {
1256 if (e.second.range.begin() == 0)
1258 simgrid::mc::FunctionIndexEntry entry;
1259 entry.low_pc = (void*) e.second.range.begin();
1260 entry.function = &e.second;
1261 info->functions_index.push_back(entry);
1264 info->functions_index.shrink_to_fit();
1266 // Sort the array by low_pc:
1267 boost::range::sort(info->functions_index,
1268 [](simgrid::mc::FunctionIndexEntry const& a,
1269 simgrid::mc::FunctionIndexEntry const& b)
1271 return a.low_pc < b.low_pc;
1275 static void MC_post_process_variables(simgrid::mc::ObjectInformation* info)
1277 // Someone needs this to be sorted but who?
1278 boost::range::sort(info->global_variables, MC_compare_variable);
1280 for (simgrid::mc::Variable& variable : info->global_variables)
1281 if (variable.type_id)
1282 variable.type = simgrid::util::find_map_ptr(
1283 info->types, variable.type_id);
1286 static void mc_post_process_scope(simgrid::mc::ObjectInformation* info, simgrid::mc::Frame* scope)
1289 if (scope->tag == DW_TAG_inlined_subroutine) {
1290 // Attach correct namespaced name in inlined subroutine:
1291 auto i = info->subprograms.find(scope->abstract_origin_id);
1292 xbt_assert(i != info->subprograms.end(),
1293 "Could not lookup abstract origin %" PRIx64,
1294 (std::uint64_t) scope->abstract_origin_id);
1295 scope->name = i->second.name;
1299 for (simgrid::mc::Variable& variable : scope->variables)
1300 if (variable.type_id)
1301 variable.type = simgrid::util::find_map_ptr(
1302 info->types, variable.type_id);
1304 // Recursive post-processing of nested-scopes:
1305 for (simgrid::mc::Frame& nested_scope : scope->scopes)
1306 mc_post_process_scope(info, &nested_scope);
1310 simgrid::mc::Type* MC_resolve_type(
1311 simgrid::mc::ObjectInformation* info, unsigned type_id)
1315 simgrid::mc::Type* type = simgrid::util::find_map_ptr(info->types, type_id);
1316 if (type == nullptr)
1319 // We already have the information on the type:
1320 if (type->byte_size != 0)
1323 // Don't have a name, we can't find a more complete version:
1324 if (type->name.empty())
1327 // Try to find a more complete description of the type:
1328 // We need to fix in order to support C++.
1329 simgrid::mc::Type** subtype = simgrid::util::find_map_ptr(
1330 info->full_types_by_name, type->name);
1336 static void MC_post_process_types(simgrid::mc::ObjectInformation* info)
1338 // Lookup "subtype" field:
1339 for (auto& i : info->types) {
1340 i.second.subtype = MC_resolve_type(info, i.second.type_id);
1341 for (simgrid::mc::Member& member : i.second.members)
1342 member.type = MC_resolve_type(info, member.type_id);
1349 /** \brief Finds informations about a given shared object/executable */
1350 std::shared_ptr<simgrid::mc::ObjectInformation> createObjectInformation(
1351 std::vector<simgrid::xbt::VmMap> const& maps, const char *name)
1353 std::shared_ptr<simgrid::mc::ObjectInformation> result =
1354 std::make_shared<simgrid::mc::ObjectInformation>();
1355 result->file_name = name;
1356 simgrid::mc::find_object_address(maps, result.get());
1357 MC_load_dwarf(result.get());
1358 MC_post_process_variables(result.get());
1359 MC_post_process_types(result.get());
1360 for (auto& entry : result.get()->subprograms)
1361 mc_post_process_scope(result.get(), &entry.second);
1362 MC_make_functions_index(result.get());
1366 /*************************************************************************/
1368 void postProcessObjectInformation(simgrid::mc::RemoteClient* process, simgrid::mc::ObjectInformation* info)
1370 for (auto& i : info->types) {
1372 simgrid::mc::Type* type = &(i.second);
1373 simgrid::mc::Type* subtype = type;
1374 while (subtype->type == DW_TAG_typedef
1375 || subtype->type == DW_TAG_volatile_type
1376 || subtype->type == DW_TAG_const_type)
1377 if (subtype->subtype)
1378 subtype = subtype->subtype;
1382 // Resolve full_type:
1383 if (not subtype->name.empty() && subtype->byte_size == 0)
1384 for (auto const& object_info : process->object_infos) {
1385 auto i = object_info->full_types_by_name.find(subtype->name);
1386 if (i != object_info->full_types_by_name.end() && not i->second->name.empty() && i->second->byte_size) {
1387 type->full_type = i->second;
1391 else type->full_type = subtype;
1402 /** Convert a DWARF register into a libunwind register
1404 * DWARF and libunwind does not use the same convention for numbering the
1405 * registers on some architectures. The function makes the necessary
1408 int dwarf_register_to_libunwind(int dwarf_register)
1410 #if defined(__x86_64__)
1411 // It seems for this arch, DWARF and libunwind agree in the numbering:
1412 return dwarf_register;
1413 #elif defined(__i386__)
1414 // Couldn't find the authoritative source of information for this.
1415 // This is inspired from http://source.winehq.org/source/dlls/dbghelp/cpu_i386.c#L517.
1416 switch (dwarf_register) {
1436 return UNW_X86_EFLAGS;
1466 xbt_die("Bad/unknown register number.");
1469 #error This architecture is not supported yet for DWARF expression evaluation.