1 /* Copyright (c) 2008-2018. 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 static Dwarf_Off MC_dwarf_attr_dieoffset(Dwarf_Die * die, int attribute)
272 Dwarf_Attribute attr;
273 if (dwarf_hasattr_integrate(die, attribute) == 0)
275 dwarf_attr_integrate(die, attribute, &attr);
276 Dwarf_Die subtype_die;
277 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
278 xbt_die("Could not find DIE");
279 return dwarf_dieoffset(&subtype_die);
282 static Dwarf_Off MC_dwarf_attr_integrate_dieoffset(Dwarf_Die * die,
285 Dwarf_Attribute attr;
286 if (dwarf_hasattr_integrate(die, attribute) == 0)
288 dwarf_attr_integrate(die, DW_AT_type, &attr);
289 Dwarf_Die subtype_die;
290 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
291 xbt_die("Could not find DIE");
292 return dwarf_dieoffset(&subtype_die);
295 /** \brief Find the type/subtype (DW_AT_type) for a DIE
298 * \return DW_AT_type reference as a global offset in hexadecimal (or nullptr)
301 std::uint64_t MC_dwarf_at_type(Dwarf_Die * die)
303 return MC_dwarf_attr_integrate_dieoffset(die, DW_AT_type);
306 static uint64_t MC_dwarf_attr_integrate_addr(Dwarf_Die * die, int attribute)
308 Dwarf_Attribute attr;
309 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
312 if (dwarf_formaddr(&attr, &value) == 0)
313 return (uint64_t) value;
318 static uint64_t MC_dwarf_attr_integrate_uint(Dwarf_Die * die, int attribute,
319 uint64_t default_value)
321 Dwarf_Attribute attr;
322 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
323 return default_value;
325 return dwarf_formudata(dwarf_attr_integrate(die, attribute, &attr),
326 &value) == 0 ? (uint64_t) value : default_value;
329 static bool MC_dwarf_attr_flag(Dwarf_Die * die, int attribute, bool integrate)
331 Dwarf_Attribute attr;
332 if ((integrate ? dwarf_attr_integrate(die, attribute, &attr)
333 : dwarf_attr(die, attribute, &attr)) == 0)
337 if (dwarf_formflag(&attr, &result))
338 xbt_die("Unexpected form for attribute %s",
339 simgrid::dwarf::attrname(attribute));
343 /** @brief Find the default lower bound for a given language
345 * The default lower bound of an array (when DW_TAG_lower_bound
346 * is missing) depends on the language of the compilation unit.
348 * @param lang Language of the compilation unit (values defined in the DWARF spec)
349 * @return Default lower bound of an array in this compilation unit
351 static uint64_t MC_dwarf_default_lower_bound(int lang)
357 case DW_LANG_C_plus_plus:
361 case DW_LANG_ObjC_plus_plus:
367 case DW_LANG_Fortran77:
368 case DW_LANG_Fortran90:
369 case DW_LANG_Fortran95:
370 case DW_LANG_Modula2:
371 case DW_LANG_Pascal83:
373 case DW_LANG_Cobol74:
374 case DW_LANG_Cobol85:
377 xbt_die("No default DW_TAG_lower_bound for language %i and none given",
383 /** \brief Finds the number of elements in a DW_TAG_subrange_type or DW_TAG_enumeration_type DIE
386 * \param unit DIE of the compilation unit
387 * \return number of elements in the range
389 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die * die,
392 xbt_assert(dwarf_tag(die) == DW_TAG_enumeration_type
393 || dwarf_tag(die) == DW_TAG_subrange_type,
394 "MC_dwarf_subrange_element_count called with DIE of type %s",
395 simgrid::dwarf::tagname(die));
397 // Use DW_TAG_count if present:
398 if (dwarf_hasattr_integrate(die, DW_AT_count))
399 return MC_dwarf_attr_integrate_uint(die, DW_AT_count, 0);
400 // Otherwise compute DW_TAG_upper_bound-DW_TAG_lower_bound + 1:
402 if (not dwarf_hasattr_integrate(die, DW_AT_upper_bound))
403 // This is not really 0, but the code expects this (we do not know):
406 uint64_t upper_bound =
407 MC_dwarf_attr_integrate_uint(die, DW_AT_upper_bound, -1);
409 uint64_t lower_bound = 0;
410 if (dwarf_hasattr_integrate(die, DW_AT_lower_bound))
411 lower_bound = MC_dwarf_attr_integrate_uint(die, DW_AT_lower_bound, -1);
413 lower_bound = MC_dwarf_default_lower_bound(dwarf_srclang(unit));
414 return upper_bound - lower_bound + 1;
417 /** \brief Finds the number of elements in a array type (DW_TAG_array_type)
419 * The compilation unit might be needed because the default lower
420 * bound depends on the language of the compilation unit.
422 * \param die the DIE of the DW_TAG_array_type
423 * \param unit the DIE of the compilation unit
424 * \return number of elements in this array type
426 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit)
428 xbt_assert(dwarf_tag(die) == DW_TAG_array_type,
429 "MC_dwarf_array_element_count called with DIE of type %s",
430 simgrid::dwarf::tagname(die));
435 for (res = dwarf_child(die, &child); res == 0;
436 res = dwarf_siblingof(&child, &child)) {
437 int child_tag = dwarf_tag(&child);
438 if (child_tag == DW_TAG_subrange_type
439 || child_tag == DW_TAG_enumeration_type)
440 result *= MC_dwarf_subrange_element_count(&child, unit);
447 /** Sort the variable by name and address.
449 * We could use boost::container::flat_set instead.
451 static bool MC_compare_variable(
452 simgrid::mc::Variable const& a, simgrid::mc::Variable const& b)
454 int cmp = strcmp(a.name.c_str(), b.name.c_str());
460 return a.address < b.address;
463 // ***** simgrid::mc::Type*
465 /** \brief Initialize the location of a member of a type
466 * (DW_AT_data_member_location of a DW_TAG_member).
468 * \param type a type (struct, class)
469 * \param member the member of the type
470 * \param child DIE of the member (DW_TAG_member)
472 static void MC_dwarf_fill_member_location(
473 simgrid::mc::Type* type, simgrid::mc::Member* member, Dwarf_Die * child)
475 if (dwarf_hasattr(child, DW_AT_data_bit_offset))
476 xbt_die("Can't groke DW_AT_data_bit_offset.");
478 if (not dwarf_hasattr_integrate(child, DW_AT_data_member_location)) {
479 if (type->type == DW_TAG_union_type)
482 ("Missing DW_AT_data_member_location field in DW_TAG_member %s of type <%"
483 PRIx64 ">%s", member->name.c_str(),
484 (uint64_t) type->id, type->name.c_str());
487 Dwarf_Attribute attr;
488 dwarf_attr_integrate(child, DW_AT_data_member_location, &attr);
489 int form = dwarf_whatform(&attr);
490 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
491 switch (form_class) {
492 case simgrid::dwarf::FormClass::ExprLoc:
493 case simgrid::dwarf::FormClass::Block:
494 // Location expression:
498 if (dwarf_getlocation(&attr, &expr, &len))
500 ("Could not read location expression DW_AT_data_member_location in DW_TAG_member %s of type <%"
501 PRIx64 ">%s", MC_dwarf_attr_integrate_string(child, DW_AT_name),
502 (uint64_t) type->id, type->name.c_str());
503 member->location_expression = simgrid::dwarf::DwarfExpression(expr, expr+len);
506 case simgrid::dwarf::FormClass::Constant:
507 // Offset from the base address of the object:
510 if (not dwarf_formudata(&attr, &offset))
511 member->offset(offset);
513 xbt_die("Cannot get %s location <%" PRIx64 ">%s",
514 MC_dwarf_attr_integrate_string(child, DW_AT_name),
515 (uint64_t) type->id, type->name.c_str());
520 // includes FormClass::LocListPtr (reference to a location list: TODO) and FormClass::Reference (it's supposed to be
521 // possible in DWARF2 but I couldn't find its semantic in the spec)
522 xbt_die("Can't handle form class (%d) / form 0x%x as DW_AT_member_location", (int)form_class, (unsigned)form);
527 /** \brief Populate the list of members of a type
529 * \param info ELF object containing the type DIE
530 * \param die DIE of the type
531 * \param unit DIE of the compilation unit containing the type DIE
532 * \param type the type
534 static void MC_dwarf_add_members(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
535 Dwarf_Die * unit, simgrid::mc::Type* type)
539 xbt_assert(type->members.empty());
540 for (res = dwarf_child(die, &child); res == 0;
541 res = dwarf_siblingof(&child, &child)) {
542 int tag = dwarf_tag(&child);
543 if (tag == DW_TAG_member || tag == DW_TAG_inheritance) {
545 // Skip declarations:
546 if (MC_dwarf_attr_flag(&child, DW_AT_declaration, false))
549 // Skip compile time constants:
550 if (dwarf_hasattr(&child, DW_AT_const_value))
553 // TODO, we should use another type (because is is not a type but a member)
554 simgrid::mc::Member member;
555 if (tag == DW_TAG_inheritance)
556 member.flags |= simgrid::mc::Member::INHERITANCE_FLAG;
558 const char *name = MC_dwarf_attr_integrate_string(&child, DW_AT_name);
561 // Those base names are used by GCC and clang for virtual table pointers
562 // respectively ("__vptr$ClassName", "__vptr.ClassName"):
563 if (boost::algorithm::starts_with(member.name, "__vptr$") ||
564 boost::algorithm::starts_with(member.name, "__vptr."))
565 member.flags |= simgrid::mc::Member::VIRTUAL_POINTER_FLAG;
566 // A cleaner solution would be to check against the type:
568 // tag: DW_TAG_member
571 // # Type for a pointer to a vtable
572 // tag: DW_TAG_pointer_type
574 // # Type for a vtable:
575 // tag: DW_TAG_pointer_type
576 // name: "__vtbl_ptr_type"
578 // tag: DW_TAG_subroutine_type
580 // tag: DW_TAG_base_type
585 MC_dwarf_attr_integrate_uint(&child, DW_AT_byte_size, 0);
586 member.type_id = MC_dwarf_at_type(&child);
588 if (dwarf_hasattr(&child, DW_AT_data_bit_offset))
589 xbt_die("Can't groke DW_AT_data_bit_offset.");
591 MC_dwarf_fill_member_location(type, &member, &child);
593 if (not member.type_id)
594 xbt_die("Missing type for member %s of <%" PRIx64 ">%s",
596 (uint64_t) type->id, type->name.c_str());
598 type->members.push_back(std::move(member));
603 /** \brief Create a MC type object from a DIE
605 * \param info current object info object
606 * \param die DIE (for a given type)
607 * \param unit compilation unit of the current DIE
608 * \return MC representation of the type
610 static simgrid::mc::Type MC_dwarf_die_to_type(
611 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
612 Dwarf_Die * unit, simgrid::mc::Frame* frame,
615 simgrid::mc::Type type;
616 type.type = dwarf_tag(die);
617 type.name = std::string();
618 type.element_count = -1;
621 type.id = dwarf_dieoffset(die);
623 const char *prefix = "";
625 case DW_TAG_structure_type:
628 case DW_TAG_union_type:
631 case DW_TAG_class_type:
638 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
639 if (name != nullptr) {
641 type.name = simgrid::xbt::string_printf("%s%s::%s", prefix, ns, name);
643 type.name = simgrid::xbt::string_printf("%s%s", prefix, name);
646 type.type_id = MC_dwarf_at_type(die);
648 // Some compilers do not emit DW_AT_byte_size for pointer_type,
649 // so we fill this. We currently assume that the model-checked process is in
650 // the same architecture..
651 if (type.type == DW_TAG_pointer_type)
652 type.byte_size = sizeof(void*);
654 // Computation of the byte_size
655 if (dwarf_hasattr_integrate(die, DW_AT_byte_size))
656 type.byte_size = MC_dwarf_attr_integrate_uint(die, DW_AT_byte_size, 0);
657 else if (type.type == DW_TAG_array_type
658 || type.type == DW_TAG_structure_type
659 || type.type == DW_TAG_class_type) {
661 if (dwarf_aggregate_size(die, &size) == 0)
662 type.byte_size = size;
666 case DW_TAG_array_type:
667 type.element_count = MC_dwarf_array_element_count(die, unit);
668 // TODO, handle DW_byte_stride and (not) DW_bit_stride
671 case DW_TAG_pointer_type:
672 case DW_TAG_reference_type:
673 case DW_TAG_rvalue_reference_type:
676 case DW_TAG_structure_type:
677 case DW_TAG_union_type:
678 case DW_TAG_class_type:
679 MC_dwarf_add_members(info, die, unit, &type);
680 MC_dwarf_handle_children(info, die, unit, frame,
681 ns ? simgrid::xbt::string_printf("%s::%s", ns, name).c_str() : type.name.c_str());
685 XBT_DEBUG("Unhandled type: %d (%s)", type.type, simgrid::dwarf::tagname(type.type));
692 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
693 Dwarf_Die * unit, simgrid::mc::Frame* frame,
696 simgrid::mc::Type type = MC_dwarf_die_to_type(info, die, unit, frame, ns);
697 auto& t = (info->types[type.id] = std::move(type));
698 if (not t.name.empty() && type.byte_size != 0)
699 info->full_types_by_name[t.name] = &t;
702 static int mc_anonymous_variable_index = 0;
704 static std::unique_ptr<simgrid::mc::Variable> MC_die_to_variable(
705 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
706 Dwarf_Die * unit, simgrid::mc::Frame* frame,
709 // Skip declarations:
710 if (MC_dwarf_attr_flag(die, DW_AT_declaration, false))
713 // Skip compile time constants:
714 if (dwarf_hasattr(die, DW_AT_const_value))
717 Dwarf_Attribute attr_location;
718 if (dwarf_attr(die, DW_AT_location, &attr_location) == nullptr)
719 // No location: do not add it ?
722 std::unique_ptr<simgrid::mc::Variable> variable =
723 std::unique_ptr<simgrid::mc::Variable>(new simgrid::mc::Variable());
724 variable->id = dwarf_dieoffset(die);
725 variable->global = frame == nullptr; // Can be override base on DW_AT_location
726 variable->object_info = info;
728 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
730 variable->name = name;
731 variable->type_id = MC_dwarf_at_type(die);
733 int form = dwarf_whatform(&attr_location);
734 simgrid::dwarf::FormClass form_class;
735 if (form == DW_FORM_sec_offset)
736 form_class = simgrid::dwarf::FormClass::Constant;
738 form_class = simgrid::dwarf::classify_form(form);
739 switch (form_class) {
740 case simgrid::dwarf::FormClass::ExprLoc:
741 case simgrid::dwarf::FormClass::Block:
742 // Location expression:
746 if (dwarf_getlocation(&attr_location, &expr, &len)) {
748 "Could not read location expression in DW_AT_location "
749 "of variable <%" PRIx64 ">%s",
750 (uint64_t) variable->id,
751 variable->name.c_str());
754 if (len == 1 && expr[0].atom == DW_OP_addr) {
755 variable->global = true;
756 uintptr_t offset = (uintptr_t) expr[0].number;
757 uintptr_t base = (uintptr_t) info->base_address();
758 variable->address = (void *) (base + offset);
760 variable->location_list = {
761 simgrid::dwarf::LocationListEntry(simgrid::dwarf::DwarfExpression(expr, expr + len))};
766 case simgrid::dwarf::FormClass::LocListPtr:
767 case simgrid::dwarf::FormClass::Constant:
768 // Reference to location list:
769 variable->location_list = simgrid::dwarf::location_list(
770 *info, attr_location);
774 xbt_die("Unexpected form 0x%x (%i), class 0x%x (%i) list for location in <%" PRIx64 ">%s", (unsigned)form, form,
775 (unsigned)form_class, (int)form_class, (uint64_t)variable->id, variable->name.c_str());
778 // Handle start_scope:
779 if (dwarf_hasattr(die, DW_AT_start_scope)) {
780 Dwarf_Attribute attr;
781 dwarf_attr(die, DW_AT_start_scope, &attr);
782 int form = dwarf_whatform(&attr);
783 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
784 if (form_class == simgrid::dwarf::FormClass::Constant) {
786 variable->start_scope = dwarf_formudata(&attr, &value) == 0 ? (size_t)value : 0;
788 // TODO: FormClass::RangeListPtr
789 xbt_die("Unhandled form 0x%x, class 0x%X for DW_AT_start_scope of variable %s", (unsigned)form,
790 (unsigned)form_class, name == nullptr ? "?" : name);
794 if (ns && variable->global)
796 std::string(ns) + "::" + variable->name;
798 // The current code needs a variable name,
799 // generate a fake one:
800 if (variable->name.empty()) {
801 variable->name = "@anonymous#" + std::to_string(mc_anonymous_variable_index);
802 mc_anonymous_variable_index++;
807 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
808 Dwarf_Die * unit, simgrid::mc::Frame* frame,
811 std::unique_ptr<simgrid::mc::Variable> variable =
812 MC_die_to_variable(info, die, unit, frame, ns);
815 // Those arrays are sorted later:
816 if (variable->global)
817 info->global_variables.push_back(std::move(*variable));
818 else if (frame != nullptr)
819 frame->variables.push_back(std::move(*variable));
821 xbt_die("No frame for this local variable");
824 static void MC_dwarf_handle_scope_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
825 Dwarf_Die * unit, simgrid::mc::Frame* parent_frame,
828 // TODO, handle DW_TAG_type/DW_TAG_location for DW_TAG_with_stmt
829 int tag = dwarf_tag(die);
830 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
832 // (Template) Subprogram declaration:
833 if (klass == simgrid::dwarf::TagClass::Subprogram
834 && MC_dwarf_attr_flag(die, DW_AT_declaration, false))
837 if (klass == simgrid::dwarf::TagClass::Scope)
838 xbt_assert(parent_frame, "No parent scope for this scope");
840 simgrid::mc::Frame frame;
842 frame.id = dwarf_dieoffset(die);
843 frame.object_info = info;
845 if (klass == simgrid::dwarf::TagClass::Subprogram) {
846 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
848 frame.name = std::string(ns) + "::" + name;
853 frame.abstract_origin_id =
854 MC_dwarf_attr_dieoffset(die, DW_AT_abstract_origin);
856 // This is the base address for DWARF addresses.
857 // Relocated addresses are offset from this base address.
858 // See DWARF4 spec 7.5
859 std::uint64_t base = (std::uint64_t) info->base_address();
861 // TODO, support DW_AT_ranges
862 uint64_t low_pc = MC_dwarf_attr_integrate_addr(die, DW_AT_low_pc);
863 frame.range.begin() = low_pc ? (std::uint64_t) base + low_pc : 0;
866 Dwarf_Attribute attr;
867 if (not dwarf_attr_integrate(die, DW_AT_high_pc, &attr))
868 xbt_die("Missing DW_AT_high_pc matching with DW_AT_low_pc");
873 switch (simgrid::dwarf::classify_form(dwarf_whatform(&attr))) {
875 // DW_AT_high_pc if an offset from the low_pc:
876 case simgrid::dwarf::FormClass::Constant:
878 if (dwarf_formsdata(&attr, &offset) != 0)
879 xbt_die("Could not read constant");
880 frame.range.end() = frame.range.begin() + offset;
883 // DW_AT_high_pc is a relocatable address:
884 case simgrid::dwarf::FormClass::Address:
885 if (dwarf_formaddr(&attr, &high_pc) != 0)
886 xbt_die("Could not read address");
887 frame.range.end() = base + high_pc;
891 xbt_die("Unexpected class for DW_AT_high_pc");
896 if (klass == simgrid::dwarf::TagClass::Subprogram) {
897 Dwarf_Attribute attr_frame_base;
898 if (dwarf_attr_integrate(die, DW_AT_frame_base, &attr_frame_base))
899 frame.frame_base_location = simgrid::dwarf::location_list(*info,
904 MC_dwarf_handle_children(info, die, unit, &frame, ns);
906 // We sort them in order to have an (somewhat) efficient by name
908 boost::range::sort(frame.variables, MC_compare_variable);
911 if (klass == simgrid::dwarf::TagClass::Subprogram)
912 info->subprograms[frame.id] = std::move(frame);
913 else if (klass == simgrid::dwarf::TagClass::Scope)
914 parent_frame->scopes.push_back(std::move(frame));
917 static void mc_dwarf_handle_namespace_die(simgrid::mc::ObjectInformation* info,
918 Dwarf_Die * die, Dwarf_Die * unit,
919 simgrid::mc::Frame* frame,
922 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
924 xbt_die("Unexpected namespace in a subprogram");
925 char *new_ns = ns == nullptr ? xbt_strdup(name)
926 : bprintf("%s::%s", ns, name);
927 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
931 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
932 Dwarf_Die * unit, simgrid::mc::Frame* frame,
935 // For each child DIE:
938 for (res = dwarf_child(die, &child); res == 0;
939 res = dwarf_siblingof(&child, &child))
940 MC_dwarf_handle_die(info, &child, unit, frame, ns);
943 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
944 Dwarf_Die * unit, simgrid::mc::Frame* frame,
947 int tag = dwarf_tag(die);
948 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
952 case simgrid::dwarf::TagClass::Type:
953 MC_dwarf_handle_type_die(info, die, unit, frame, ns);
956 // Subprogram or scope:
957 case simgrid::dwarf::TagClass::Subprogram:
958 case simgrid::dwarf::TagClass::Scope:
959 MC_dwarf_handle_scope_die(info, die, unit, frame, ns);
963 case simgrid::dwarf::TagClass::Variable:
964 MC_dwarf_handle_variable_die(info, die, unit, frame, ns);
967 case simgrid::dwarf::TagClass::Namespace:
968 mc_dwarf_handle_namespace_die(info, die, unit, frame, ns);
978 Elf64_Half get_type(Elf* elf)
980 Elf64_Ehdr* ehdr64 = elf64_getehdr(elf);
982 return ehdr64->e_type;
983 Elf32_Ehdr* ehdr32 = elf32_getehdr(elf);
985 return ehdr32->e_type;
986 xbt_die("Could not get ELF heeader");
990 void read_dwarf_info(simgrid::mc::ObjectInformation* info, Dwarf* dwarf)
992 // For each compilation unit:
993 Dwarf_Off offset = 0;
994 Dwarf_Off next_offset = 0;
997 while (dwarf_nextcu(dwarf, offset, &next_offset, &length, nullptr, nullptr, nullptr) ==
1000 if (dwarf_offdie(dwarf, offset + length, &unit_die) != nullptr)
1001 MC_dwarf_handle_children(info, &unit_die, &unit_die, nullptr, nullptr);
1002 offset = next_offset;
1006 /** Get the build-id (NT_GNU_BUILD_ID) from the ELF file
1008 * This build-id may is used to locate an external debug (DWARF) file
1009 * for this ELF file.
1011 * @param elf libelf handle for an ELF file
1012 * @return build-id for this ELF file (or an empty vector if none is found)
1015 std::vector<char> get_build_id(Elf* elf)
1018 // Summary: the GNU build ID is stored in a ("GNU, NT_GNU_BUILD_ID) note
1019 // found in a PT_NOTE entry in the program header table.
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:
1036 while (pos < data->d_size) {
1038 // Location of the name within Elf_Data:
1041 pos = gelf_getnote(data, pos, &nhdr, &name_pos, &desc_pos);
1042 // A build ID note is identified by the pair ("GNU", NT_GNU_BUILD_ID)
1043 // (a namespace and a type within this namespace):
1044 if (nhdr.n_type == NT_GNU_BUILD_ID
1045 && nhdr.n_namesz == sizeof("GNU")
1046 && memcmp((char*) data->d_buf + name_pos, "GNU", sizeof("GNU")) == 0) {
1047 XBT_DEBUG("Found GNU/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 // /usr/lib/debug/.build-id/0b/dc77f1c29aea2b14ff5acd9a19ab3175ffdeae.debug
1109 std::string find_by_build_id(std::vector<char> id)
1111 std::string filename;
1112 std::string hex = to_hex(id);
1113 for (const char* const& debug_path : debug_paths) {
1115 filename = std::string(debug_path) + ".build-id/"
1116 + to_hex(id.data(), 1) + '/'
1117 + to_hex(id.data() + 1, id.size() - 1) + ".debug";
1118 XBT_DEBUG("Checking debug file: %s", filename.c_str());
1119 if (access(filename.c_str(), F_OK) == 0) {
1120 XBT_DEBUG("Found debug file: %s\n", hex.c_str());
1124 XBT_DEBUG("Not debuf info found for build ID %s\n", hex.data());
1125 return std::string();
1128 /** \brief Populate the debugging informations of the given ELF object
1130 * Read the DWARf information of the EFFL object and populate the
1131 * lists of types, variables, functions.
1134 void MC_load_dwarf(simgrid::mc::ObjectInformation* info)
1136 if (elf_version(EV_CURRENT) == EV_NONE)
1137 xbt_die("libelf initialization error");
1139 // Open the ELF file:
1140 int fd = open(info->file_name.c_str(), O_RDONLY);
1142 xbt_die("Could not open file %s", info->file_name.c_str());
1143 Elf* elf = elf_begin(fd, ELF_C_READ, nullptr);
1145 xbt_die("Not an ELF file");
1146 Elf_Kind kind = elf_kind(elf);
1147 if (kind != ELF_K_ELF)
1148 xbt_die("Not an ELF file");
1150 // Remember if this is a `ET_EXEC` (fixed location) or `ET_DYN`:
1151 Elf64_Half type = get_type(elf);
1152 if (type == ET_EXEC)
1153 info->flags |= simgrid::mc::ObjectInformation::Executable;
1155 // Read DWARF debug information in the file:
1156 Dwarf* dwarf = dwarf_begin_elf (elf, DWARF_C_READ, nullptr);
1157 if (dwarf != nullptr) {
1158 read_dwarf_info(info, dwarf);
1166 // If there was no DWARF in the file, try to find it in a separate file.
1167 // Different methods might be used to store the DWARF informations:
1168 // * GNU NT_GNU_BUILD_ID
1170 // See https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html
1171 // for reference of what we are doing.
1173 // Try with NT_GNU_BUILD_ID: we find the build ID in the ELF file and then
1174 // use this ID to find the file in some known locations in the filesystem.
1175 std::vector<char> build_id = get_build_id(elf);
1176 if (not build_id.empty()) {
1180 // Find the debug file using the build id:
1181 std::string debug_file = find_by_build_id(build_id);
1182 if (debug_file.empty()) {
1183 std::string hex = to_hex(build_id);
1184 xbt_die("Missing debug info for %s with build-id %s\n"
1185 "You might want to install the suitable debugging package.\n",
1186 info->file_name.c_str(), hex.c_str());
1189 // Load the DWARF info from this file:
1190 XBT_DEBUG("Load DWARF for %s from %s",
1191 info->file_name.c_str(), debug_file.c_str());
1192 fd = open(debug_file.c_str(), O_RDONLY);
1194 xbt_die("Could not open file %s", debug_file.c_str());
1195 Dwarf* dwarf = dwarf_begin(fd, DWARF_C_READ);
1196 if (dwarf == nullptr)
1197 xbt_die("No DWARF info in %s for %s",
1198 debug_file.c_str(), info->file_name.c_str());
1199 read_dwarf_info(info, dwarf);
1205 // TODO, try to find DWARF info using .gnu_debuglink.
1209 xbt_die("Debugging information not found for %s\n"
1210 "Try recompiling with -g\n",
1211 info->file_name.c_str());
1214 // ***** Functions index
1216 static void MC_make_functions_index(simgrid::mc::ObjectInformation* info)
1218 info->functions_index.clear();
1220 for (auto& e : info->subprograms) {
1221 if (e.second.range.begin() == 0)
1223 simgrid::mc::FunctionIndexEntry entry;
1224 entry.low_pc = (void*) e.second.range.begin();
1225 entry.function = &e.second;
1226 info->functions_index.push_back(entry);
1229 info->functions_index.shrink_to_fit();
1231 // Sort the array by low_pc:
1232 boost::range::sort(info->functions_index,
1233 [](simgrid::mc::FunctionIndexEntry const& a,
1234 simgrid::mc::FunctionIndexEntry const& b)
1236 return a.low_pc < b.low_pc;
1240 static void MC_post_process_variables(simgrid::mc::ObjectInformation* info)
1242 // Someone needs this to be sorted but who?
1243 boost::range::sort(info->global_variables, MC_compare_variable);
1245 for (simgrid::mc::Variable& variable : info->global_variables)
1246 if (variable.type_id)
1247 variable.type = simgrid::util::find_map_ptr(
1248 info->types, variable.type_id);
1251 static void mc_post_process_scope(simgrid::mc::ObjectInformation* info, simgrid::mc::Frame* scope)
1254 if (scope->tag == DW_TAG_inlined_subroutine) {
1255 // Attach correct namespaced name in inlined subroutine:
1256 auto i = info->subprograms.find(scope->abstract_origin_id);
1257 xbt_assert(i != info->subprograms.end(),
1258 "Could not lookup abstract origin %" PRIx64,
1259 (std::uint64_t) scope->abstract_origin_id);
1260 scope->name = i->second.name;
1264 for (simgrid::mc::Variable& variable : scope->variables)
1265 if (variable.type_id)
1266 variable.type = simgrid::util::find_map_ptr(
1267 info->types, variable.type_id);
1269 // Recursive post-processing of nested-scopes:
1270 for (simgrid::mc::Frame& nested_scope : scope->scopes)
1271 mc_post_process_scope(info, &nested_scope);
1275 simgrid::mc::Type* MC_resolve_type(
1276 simgrid::mc::ObjectInformation* info, unsigned type_id)
1280 simgrid::mc::Type* type = simgrid::util::find_map_ptr(info->types, type_id);
1281 if (type == nullptr)
1284 // We already have the information on the type:
1285 if (type->byte_size != 0)
1288 // Don't have a name, we can't find a more complete version:
1289 if (type->name.empty())
1292 // Try to find a more complete description of the type:
1293 // We need to fix in order to support C++.
1294 simgrid::mc::Type** subtype = simgrid::util::find_map_ptr(
1295 info->full_types_by_name, type->name);
1301 static void MC_post_process_types(simgrid::mc::ObjectInformation* info)
1303 // Lookup "subtype" field:
1304 for (auto& i : info->types) {
1305 i.second.subtype = MC_resolve_type(info, i.second.type_id);
1306 for (simgrid::mc::Member& member : i.second.members)
1307 member.type = MC_resolve_type(info, member.type_id);
1314 /** \brief Finds informations about a given shared object/executable */
1315 std::shared_ptr<simgrid::mc::ObjectInformation> createObjectInformation(
1316 std::vector<simgrid::xbt::VmMap> const& maps, const char *name)
1318 std::shared_ptr<simgrid::mc::ObjectInformation> result =
1319 std::make_shared<simgrid::mc::ObjectInformation>();
1320 result->file_name = name;
1321 simgrid::mc::find_object_address(maps, result.get());
1322 MC_load_dwarf(result.get());
1323 MC_post_process_variables(result.get());
1324 MC_post_process_types(result.get());
1325 for (auto& entry : result.get()->subprograms)
1326 mc_post_process_scope(result.get(), &entry.second);
1327 MC_make_functions_index(result.get());
1331 /*************************************************************************/
1333 void postProcessObjectInformation(simgrid::mc::RemoteClient* process, simgrid::mc::ObjectInformation* info)
1335 for (auto& i : info->types) {
1337 simgrid::mc::Type* type = &(i.second);
1338 simgrid::mc::Type* subtype = type;
1339 while (subtype->type == DW_TAG_typedef
1340 || subtype->type == DW_TAG_volatile_type
1341 || subtype->type == DW_TAG_const_type)
1342 if (subtype->subtype)
1343 subtype = subtype->subtype;
1347 // Resolve full_type:
1348 if (not subtype->name.empty() && subtype->byte_size == 0)
1349 for (auto const& object_info : process->object_infos) {
1350 auto i = object_info->full_types_by_name.find(subtype->name);
1351 if (i != object_info->full_types_by_name.end() && not 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 // Couldn'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.