1 /* Copyright (c) 2008-2019. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
12 #include <boost/range/algorithm.hpp>
16 #include <elfutils/libdw.h>
18 #include <boost/algorithm/string/predicate.hpp>
20 #include "src/simgrid/util.hpp"
22 #include "xbt/string.hpp"
23 #include "xbt/sysdep.h"
24 #include <simgrid/config.h>
26 #include "src/mc/mc_dwarf.hpp"
27 #include "src/mc/mc_private.hpp"
29 #include "src/mc/ObjectInformation.hpp"
30 #include "src/mc/Variable.hpp"
31 #include "src/mc/remote/RemoteClient.hpp"
33 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_dwarf, mc, "DWARF processing");
35 /** @brief The default DW_TAG_lower_bound for a given DW_AT_language.
37 * The default for a given language is defined in the DWARF spec.
39 * @param language constant as defined by the DWARf spec
41 static uint64_t MC_dwarf_default_lower_bound(int lang);
43 /** @brief Computes the the element_count of a DW_TAG_enumeration_type DIE
45 * This is the number of elements in a given array dimension.
47 * A reference of the compilation unit (DW_TAG_compile_unit) is
48 * needed because the default lower bound (when there is no DW_AT_lower_bound)
49 * depends of the language of the compilation unit (DW_AT_language).
51 * @param die DIE for the DW_TAG_enumeration_type or DW_TAG_subrange_type
52 * @param unit DIE of the DW_TAG_compile_unit
54 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die* die, Dwarf_Die* unit);
56 /** @brief Computes the number of elements of a given DW_TAG_array_type.
58 * @param die DIE for the DW_TAG_array_type
60 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit);
62 /** @brief Process a DIE
64 * @param info the resulting object fot the library/binary file (output)
65 * @param die the current DIE
66 * @param unit the DIE of the compile unit of the current DIE
67 * @param frame containing frame if any
69 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
70 Dwarf_Die * unit, simgrid::mc::Frame* frame,
73 /** @brief Process a type DIE
75 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
76 Dwarf_Die * unit, simgrid::mc::Frame* frame,
79 /** @brief Calls MC_dwarf_handle_die on all children of the given die
81 * @param info the resulting object fot the library/binary file (output)
82 * @param die the current DIE
83 * @param unit the DIE of the compile unit of the current DIE
84 * @param frame containing frame if any
86 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
87 Dwarf_Die * unit, simgrid::mc::Frame* frame,
90 /** @brief Handle a variable (DW_TAG_variable or other)
92 * @param info the resulting object fot the library/binary file (output)
93 * @param die the current DIE
94 * @param unit the DIE of the compile unit of the current DIE
95 * @param frame containing frame if any
97 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
98 Dwarf_Die * unit, simgrid::mc::Frame* frame,
101 /** @brief Get the DW_TAG_type of the DIE
104 * @return DW_TAG_type attribute as a new string (nullptr if none)
106 static std::uint64_t MC_dwarf_at_type(Dwarf_Die * die);
111 enum class TagClass {
120 /*** Class of forms defined in the DWARF standard */
121 enum class FormClass {
123 Address, // Location in the program's address space
124 Block, // Arbitrary block of bytes
127 Flag, // Boolean value
128 Reference, // Reference to another DIE
129 ExprLoc, // DWARF expression/location description
137 TagClass classify_tag(int tag)
141 case DW_TAG_array_type:
142 case DW_TAG_class_type:
143 case DW_TAG_enumeration_type:
145 case DW_TAG_pointer_type:
146 case DW_TAG_reference_type:
147 case DW_TAG_rvalue_reference_type:
148 case DW_TAG_string_type:
149 case DW_TAG_structure_type:
150 case DW_TAG_subroutine_type:
151 case DW_TAG_union_type:
152 case DW_TAG_ptr_to_member_type:
153 case DW_TAG_set_type:
154 case DW_TAG_subrange_type:
155 case DW_TAG_base_type:
156 case DW_TAG_const_type:
157 case DW_TAG_file_type:
158 case DW_TAG_packed_type:
159 case DW_TAG_volatile_type:
160 case DW_TAG_restrict_type:
161 case DW_TAG_interface_type:
162 case DW_TAG_unspecified_type:
163 case DW_TAG_shared_type:
164 return TagClass::Type;
166 case DW_TAG_subprogram:
167 return TagClass::Subprogram;
169 case DW_TAG_variable:
170 case DW_TAG_formal_parameter:
171 return TagClass::Variable;
173 case DW_TAG_lexical_block:
174 case DW_TAG_try_block:
175 case DW_TAG_catch_block:
176 case DW_TAG_inlined_subroutine:
177 case DW_TAG_with_stmt:
178 return TagClass::Scope;
180 case DW_TAG_namespace:
181 return TagClass::Namespace;
184 return TagClass::Unknown;
188 /** @brief Find the DWARF data class for a given DWARF data form
190 * This mapping is defined in the DWARF spec.
192 * @param form The form (values taken from the DWARF spec)
193 * @return An internal representation for the corresponding class
196 FormClass classify_form(int form)
200 return FormClass::Address;
205 return FormClass::Block;
212 return FormClass::Constant;
215 return FormClass::String;
216 case DW_FORM_ref_addr:
221 case DW_FORM_ref_udata:
222 return FormClass::Reference;
224 case DW_FORM_flag_present:
225 return FormClass::Flag;
226 case DW_FORM_exprloc:
227 return FormClass::ExprLoc;
231 return FormClass::Unknown;
235 /** @brief Get the name of the tag of a given DIE
238 * @return name of the tag of this DIE
241 const char *tagname(Dwarf_Die * die)
243 return simgrid::dwarf::tagname(dwarf_tag(die));
251 /** @brief Get an attribute of a given DIE as a string
254 * @param attribute attribute
255 * @return value of the given attribute of the given DIE
257 static const char *MC_dwarf_attr_integrate_string(Dwarf_Die * die,
260 Dwarf_Attribute attr;
261 if (not dwarf_attr_integrate(die, attribute, &attr))
264 return dwarf_formstring(&attr);
267 static Dwarf_Off MC_dwarf_attr_dieoffset(Dwarf_Die * die, int attribute)
269 Dwarf_Attribute attr;
270 if (dwarf_hasattr_integrate(die, attribute) == 0)
272 dwarf_attr_integrate(die, attribute, &attr);
273 Dwarf_Die subtype_die;
274 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
275 xbt_die("Could not find DIE");
276 return dwarf_dieoffset(&subtype_die);
279 static Dwarf_Off MC_dwarf_attr_integrate_dieoffset(Dwarf_Die * die,
282 Dwarf_Attribute attr;
283 if (dwarf_hasattr_integrate(die, attribute) == 0)
285 dwarf_attr_integrate(die, DW_AT_type, &attr);
286 Dwarf_Die subtype_die;
287 if (dwarf_formref_die(&attr, &subtype_die) == nullptr)
288 xbt_die("Could not find DIE");
289 return dwarf_dieoffset(&subtype_die);
292 /** @brief Find the type/subtype (DW_AT_type) for a DIE
295 * @return DW_AT_type reference as a global offset in hexadecimal (or nullptr)
298 std::uint64_t MC_dwarf_at_type(Dwarf_Die * die)
300 return MC_dwarf_attr_integrate_dieoffset(die, DW_AT_type);
303 static uint64_t MC_dwarf_attr_integrate_addr(Dwarf_Die * die, int attribute)
305 Dwarf_Attribute attr;
306 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
309 if (dwarf_formaddr(&attr, &value) == 0)
310 return (uint64_t) value;
315 static uint64_t MC_dwarf_attr_integrate_uint(Dwarf_Die * die, int attribute,
316 uint64_t default_value)
318 Dwarf_Attribute attr;
319 if (dwarf_attr_integrate(die, attribute, &attr) == nullptr)
320 return default_value;
322 return dwarf_formudata(dwarf_attr_integrate(die, attribute, &attr),
323 &value) == 0 ? (uint64_t) value : default_value;
326 static bool MC_dwarf_attr_flag(Dwarf_Die * die, int attribute, bool integrate)
328 Dwarf_Attribute attr;
329 if ((integrate ? dwarf_attr_integrate(die, attribute, &attr)
330 : dwarf_attr(die, attribute, &attr)) == 0)
334 if (dwarf_formflag(&attr, &result))
335 xbt_die("Unexpected form for attribute %s",
336 simgrid::dwarf::attrname(attribute));
340 /** @brief Find the default lower bound for a given language
342 * The default lower bound of an array (when DW_TAG_lower_bound
343 * is missing) depends on the language of the compilation unit.
345 * @param lang Language of the compilation unit (values defined in the DWARF spec)
346 * @return Default lower bound of an array in this compilation unit
348 static uint64_t MC_dwarf_default_lower_bound(int lang)
354 case DW_LANG_C_plus_plus:
358 case DW_LANG_ObjC_plus_plus:
364 case DW_LANG_Fortran77:
365 case DW_LANG_Fortran90:
366 case DW_LANG_Fortran95:
367 case DW_LANG_Modula2:
368 case DW_LANG_Pascal83:
370 case DW_LANG_Cobol74:
371 case DW_LANG_Cobol85:
374 xbt_die("No default DW_TAG_lower_bound for language %i and none given",
380 /** @brief Finds the number of elements in a DW_TAG_subrange_type or DW_TAG_enumeration_type DIE
383 * @param unit DIE of the compilation unit
384 * @return number of elements in the range
386 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die * die,
389 xbt_assert(dwarf_tag(die) == DW_TAG_enumeration_type
390 || dwarf_tag(die) == DW_TAG_subrange_type,
391 "MC_dwarf_subrange_element_count called with DIE of type %s",
392 simgrid::dwarf::tagname(die));
394 // Use DW_TAG_count if present:
395 if (dwarf_hasattr_integrate(die, DW_AT_count))
396 return MC_dwarf_attr_integrate_uint(die, DW_AT_count, 0);
397 // Otherwise compute DW_TAG_upper_bound-DW_TAG_lower_bound + 1:
399 if (not dwarf_hasattr_integrate(die, DW_AT_upper_bound))
400 // This is not really 0, but the code expects this (we do not know):
403 uint64_t upper_bound = MC_dwarf_attr_integrate_uint(die, DW_AT_upper_bound, static_cast<uint64_t>(-1));
405 uint64_t lower_bound = 0;
406 if (dwarf_hasattr_integrate(die, DW_AT_lower_bound))
407 lower_bound = MC_dwarf_attr_integrate_uint(die, DW_AT_lower_bound, static_cast<uint64_t>(-1));
409 lower_bound = MC_dwarf_default_lower_bound(dwarf_srclang(unit));
410 return upper_bound - lower_bound + 1;
413 /** @brief Finds the number of elements in a array type (DW_TAG_array_type)
415 * The compilation unit might be needed because the default lower
416 * bound depends on the language of the compilation unit.
418 * @param die the DIE of the DW_TAG_array_type
419 * @param unit the DIE of the compilation unit
420 * @return number of elements in this array type
422 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit)
424 xbt_assert(dwarf_tag(die) == DW_TAG_array_type,
425 "MC_dwarf_array_element_count called with DIE of type %s",
426 simgrid::dwarf::tagname(die));
431 for (res = dwarf_child(die, &child); res == 0;
432 res = dwarf_siblingof(&child, &child)) {
433 int child_tag = dwarf_tag(&child);
434 if (child_tag == DW_TAG_subrange_type
435 || child_tag == DW_TAG_enumeration_type)
436 result *= MC_dwarf_subrange_element_count(&child, unit);
443 /** Sort the variable by name and address.
445 * We could use boost::container::flat_set instead.
447 static bool MC_compare_variable(
448 simgrid::mc::Variable const& a, simgrid::mc::Variable const& b)
450 int cmp = strcmp(a.name.c_str(), b.name.c_str());
456 return a.address < b.address;
459 // ***** simgrid::mc::Type*
461 /** @brief Initialize the location of a member of a type
462 * (DW_AT_data_member_location of a DW_TAG_member).
464 * @param type a type (struct, class)
465 * @param member the member of the type
466 * @param child DIE of the member (DW_TAG_member)
468 static void MC_dwarf_fill_member_location(
469 simgrid::mc::Type* type, simgrid::mc::Member* member, Dwarf_Die * child)
471 if (dwarf_hasattr(child, DW_AT_data_bit_offset))
472 xbt_die("Can't groke DW_AT_data_bit_offset.");
474 if (not dwarf_hasattr_integrate(child, DW_AT_data_member_location)) {
475 if (type->type == DW_TAG_union_type)
478 ("Missing DW_AT_data_member_location field in DW_TAG_member %s of type <%"
479 PRIx64 ">%s", member->name.c_str(),
480 (uint64_t) type->id, type->name.c_str());
483 Dwarf_Attribute attr;
484 dwarf_attr_integrate(child, DW_AT_data_member_location, &attr);
485 int form = dwarf_whatform(&attr);
486 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
487 switch (form_class) {
488 case simgrid::dwarf::FormClass::ExprLoc:
489 case simgrid::dwarf::FormClass::Block:
490 // Location expression:
494 if (dwarf_getlocation(&attr, &expr, &len))
496 ("Could not read location expression DW_AT_data_member_location in DW_TAG_member %s of type <%"
497 PRIx64 ">%s", MC_dwarf_attr_integrate_string(child, DW_AT_name),
498 (uint64_t) type->id, type->name.c_str());
499 member->location_expression = simgrid::dwarf::DwarfExpression(expr, expr+len);
502 case simgrid::dwarf::FormClass::Constant:
503 // Offset from the base address of the object:
506 if (not dwarf_formudata(&attr, &offset))
507 member->offset(offset);
509 xbt_die("Cannot get %s location <%" PRIx64 ">%s",
510 MC_dwarf_attr_integrate_string(child, DW_AT_name),
511 (uint64_t) type->id, type->name.c_str());
516 // includes FormClass::LocListPtr (reference to a location list: TODO) and FormClass::Reference (it's supposed to be
517 // possible in DWARF2 but I couldn't find its semantic in the spec)
518 xbt_die("Can't handle form class (%d) / form 0x%x as DW_AT_member_location", (int)form_class, (unsigned)form);
523 /** @brief Populate the list of members of a type
525 * @param info ELF object containing the type DIE
526 * @param die DIE of the type
527 * @param unit DIE of the compilation unit containing the type DIE
528 * @param type the type
530 static void MC_dwarf_add_members(simgrid::mc::ObjectInformation* /*info*/, Dwarf_Die* die, Dwarf_Die* /*unit*/,
531 simgrid::mc::Type* type)
535 xbt_assert(type->members.empty());
536 for (res = dwarf_child(die, &child); res == 0;
537 res = dwarf_siblingof(&child, &child)) {
538 int tag = dwarf_tag(&child);
539 if (tag == DW_TAG_member || tag == DW_TAG_inheritance) {
541 // Skip declarations:
542 if (MC_dwarf_attr_flag(&child, DW_AT_declaration, false))
545 // Skip compile time constants:
546 if (dwarf_hasattr(&child, DW_AT_const_value))
549 // TODO, we should use another type (because is is not a type but a member)
550 simgrid::mc::Member member;
551 if (tag == DW_TAG_inheritance)
552 member.flags |= simgrid::mc::Member::INHERITANCE_FLAG;
554 const char *name = MC_dwarf_attr_integrate_string(&child, DW_AT_name);
557 // Those base names are used by GCC and clang for virtual table pointers
558 // respectively ("__vptr$ClassName", "__vptr.ClassName"):
559 if (boost::algorithm::starts_with(member.name, "__vptr$") ||
560 boost::algorithm::starts_with(member.name, "__vptr."))
561 member.flags |= simgrid::mc::Member::VIRTUAL_POINTER_FLAG;
562 // A cleaner solution would be to check against the type:
564 // tag: DW_TAG_member
567 // # Type for a pointer to a vtable
568 // tag: DW_TAG_pointer_type
570 // # Type for a vtable:
571 // tag: DW_TAG_pointer_type
572 // name: "__vtbl_ptr_type"
574 // tag: DW_TAG_subroutine_type
576 // tag: DW_TAG_base_type
581 MC_dwarf_attr_integrate_uint(&child, DW_AT_byte_size, 0);
582 member.type_id = MC_dwarf_at_type(&child);
584 if (dwarf_hasattr(&child, DW_AT_data_bit_offset))
585 xbt_die("Can't groke DW_AT_data_bit_offset.");
587 MC_dwarf_fill_member_location(type, &member, &child);
589 if (not member.type_id)
590 xbt_die("Missing type for member %s of <%" PRIx64 ">%s",
592 (uint64_t) type->id, type->name.c_str());
594 type->members.push_back(std::move(member));
599 /** @brief Create a MC type object from a DIE
601 * @param info current object info object
602 * @param die DIE (for a given type)
603 * @param unit compilation unit of the current DIE
604 * @return MC representation of the type
606 static simgrid::mc::Type MC_dwarf_die_to_type(
607 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
608 Dwarf_Die * unit, simgrid::mc::Frame* frame,
611 simgrid::mc::Type type;
612 type.type = dwarf_tag(die);
613 type.name = std::string();
614 type.element_count = -1;
617 type.id = dwarf_dieoffset(die);
619 const char *prefix = "";
621 case DW_TAG_structure_type:
624 case DW_TAG_union_type:
627 case DW_TAG_class_type:
634 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
635 if (name != nullptr) {
637 type.name = simgrid::xbt::string_printf("%s%s::%s", prefix, ns, name);
639 type.name = simgrid::xbt::string_printf("%s%s", prefix, name);
642 type.type_id = MC_dwarf_at_type(die);
644 // Some compilers do not emit DW_AT_byte_size for pointer_type,
645 // so we fill this. We currently assume that the model-checked process is in
646 // the same architecture..
647 if (type.type == DW_TAG_pointer_type)
648 type.byte_size = sizeof(void*);
650 // Computation of the byte_size
651 if (dwarf_hasattr_integrate(die, DW_AT_byte_size))
652 type.byte_size = MC_dwarf_attr_integrate_uint(die, DW_AT_byte_size, 0);
653 else if (type.type == DW_TAG_array_type
654 || type.type == DW_TAG_structure_type
655 || type.type == DW_TAG_class_type) {
657 if (dwarf_aggregate_size(die, &size) == 0)
658 type.byte_size = size;
662 case DW_TAG_array_type:
663 type.element_count = MC_dwarf_array_element_count(die, unit);
664 // TODO, handle DW_byte_stride and (not) DW_bit_stride
667 case DW_TAG_pointer_type:
668 case DW_TAG_reference_type:
669 case DW_TAG_rvalue_reference_type:
672 case DW_TAG_structure_type:
673 case DW_TAG_union_type:
674 case DW_TAG_class_type:
675 MC_dwarf_add_members(info, die, unit, &type);
676 MC_dwarf_handle_children(info, die, unit, frame,
677 ns ? simgrid::xbt::string_printf("%s::%s", ns, name).c_str() : type.name.c_str());
681 XBT_DEBUG("Unhandled type: %d (%s)", type.type, simgrid::dwarf::tagname(type.type));
688 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
689 Dwarf_Die * unit, simgrid::mc::Frame* frame,
692 simgrid::mc::Type type = MC_dwarf_die_to_type(info, die, unit, frame, ns);
693 auto& t = (info->types[type.id] = std::move(type));
694 if (not t.name.empty() && type.byte_size != 0)
695 info->full_types_by_name[t.name] = &t;
698 static int mc_anonymous_variable_index = 0;
700 static std::unique_ptr<simgrid::mc::Variable> MC_die_to_variable(simgrid::mc::ObjectInformation* info, Dwarf_Die* die,
701 Dwarf_Die* /*unit*/, simgrid::mc::Frame* frame,
704 // Skip declarations:
705 if (MC_dwarf_attr_flag(die, DW_AT_declaration, false))
708 // Skip compile time constants:
709 if (dwarf_hasattr(die, DW_AT_const_value))
712 Dwarf_Attribute attr_location;
713 if (dwarf_attr(die, DW_AT_location, &attr_location) == nullptr)
714 // No location: do not add it ?
717 std::unique_ptr<simgrid::mc::Variable> variable =
718 std::unique_ptr<simgrid::mc::Variable>(new simgrid::mc::Variable());
719 variable->id = dwarf_dieoffset(die);
720 variable->global = frame == nullptr; // Can be override base on DW_AT_location
721 variable->object_info = info;
723 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
725 variable->name = name;
726 variable->type_id = MC_dwarf_at_type(die);
728 int form = dwarf_whatform(&attr_location);
729 simgrid::dwarf::FormClass form_class;
730 if (form == DW_FORM_sec_offset)
731 form_class = simgrid::dwarf::FormClass::Constant;
733 form_class = simgrid::dwarf::classify_form(form);
734 switch (form_class) {
735 case simgrid::dwarf::FormClass::ExprLoc:
736 case simgrid::dwarf::FormClass::Block:
737 // Location expression:
741 if (dwarf_getlocation(&attr_location, &expr, &len)) {
743 "Could not read location expression in DW_AT_location "
744 "of variable <%" PRIx64 ">%s",
745 (uint64_t) variable->id,
746 variable->name.c_str());
749 if (len == 1 && expr[0].atom == DW_OP_addr) {
750 variable->global = true;
751 uintptr_t offset = (uintptr_t) expr[0].number;
752 uintptr_t base = (uintptr_t) info->base_address();
753 variable->address = (void *) (base + offset);
755 variable->location_list = {
756 simgrid::dwarf::LocationListEntry(simgrid::dwarf::DwarfExpression(expr, expr + len))};
761 case simgrid::dwarf::FormClass::LocListPtr:
762 case simgrid::dwarf::FormClass::Constant:
763 // Reference to location list:
764 variable->location_list = simgrid::dwarf::location_list(
765 *info, attr_location);
769 xbt_die("Unexpected form 0x%x (%i), class 0x%x (%i) list for location in <%" PRIx64 ">%s", (unsigned)form, form,
770 (unsigned)form_class, (int)form_class, (uint64_t)variable->id, variable->name.c_str());
773 // Handle start_scope:
774 if (dwarf_hasattr(die, DW_AT_start_scope)) {
775 Dwarf_Attribute attr;
776 dwarf_attr(die, DW_AT_start_scope, &attr);
777 form = dwarf_whatform(&attr);
778 form_class = simgrid::dwarf::classify_form(form);
779 if (form_class == simgrid::dwarf::FormClass::Constant) {
781 variable->start_scope = dwarf_formudata(&attr, &value) == 0 ? (size_t)value : 0;
783 // TODO: FormClass::RangeListPtr
784 xbt_die("Unhandled form 0x%x, class 0x%X for DW_AT_start_scope of variable %s", (unsigned)form,
785 (unsigned)form_class, name == nullptr ? "?" : name);
789 if (ns && variable->global)
791 std::string(ns) + "::" + variable->name;
793 // The current code needs a variable name,
794 // generate a fake one:
795 if (variable->name.empty()) {
796 variable->name = "@anonymous#" + std::to_string(mc_anonymous_variable_index);
797 mc_anonymous_variable_index++;
802 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
803 Dwarf_Die * unit, simgrid::mc::Frame* frame,
806 std::unique_ptr<simgrid::mc::Variable> variable =
807 MC_die_to_variable(info, die, unit, frame, ns);
810 // Those arrays are sorted later:
811 if (variable->global)
812 info->global_variables.push_back(std::move(*variable));
813 else if (frame != nullptr)
814 frame->variables.push_back(std::move(*variable));
816 xbt_die("No frame for this local variable");
819 static void MC_dwarf_handle_scope_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
820 Dwarf_Die * unit, simgrid::mc::Frame* parent_frame,
823 // TODO, handle DW_TAG_type/DW_TAG_location for DW_TAG_with_stmt
824 int tag = dwarf_tag(die);
825 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
827 // (Template) Subprogram declaration:
828 if (klass == simgrid::dwarf::TagClass::Subprogram
829 && MC_dwarf_attr_flag(die, DW_AT_declaration, false))
832 if (klass == simgrid::dwarf::TagClass::Scope)
833 xbt_assert(parent_frame, "No parent scope for this scope");
835 simgrid::mc::Frame frame;
837 frame.id = dwarf_dieoffset(die);
838 frame.object_info = info;
840 if (klass == simgrid::dwarf::TagClass::Subprogram) {
841 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
843 frame.name = std::string(ns) + "::" + name;
848 frame.abstract_origin_id =
849 MC_dwarf_attr_dieoffset(die, DW_AT_abstract_origin);
851 // This is the base address for DWARF addresses.
852 // Relocated addresses are offset from this base address.
853 // See DWARF4 spec 7.5
854 std::uint64_t base = (std::uint64_t) info->base_address();
856 // TODO, support DW_AT_ranges
857 uint64_t low_pc = MC_dwarf_attr_integrate_addr(die, DW_AT_low_pc);
858 frame.range.begin() = low_pc ? (std::uint64_t) base + low_pc : 0;
861 Dwarf_Attribute attr;
862 if (not dwarf_attr_integrate(die, DW_AT_high_pc, &attr))
863 xbt_die("Missing DW_AT_high_pc matching with DW_AT_low_pc");
868 switch (simgrid::dwarf::classify_form(dwarf_whatform(&attr))) {
870 // DW_AT_high_pc if an offset from the low_pc:
871 case simgrid::dwarf::FormClass::Constant:
873 if (dwarf_formsdata(&attr, &offset) != 0)
874 xbt_die("Could not read constant");
875 frame.range.end() = frame.range.begin() + offset;
878 // DW_AT_high_pc is a relocatable address:
879 case simgrid::dwarf::FormClass::Address:
880 if (dwarf_formaddr(&attr, &high_pc) != 0)
881 xbt_die("Could not read address");
882 frame.range.end() = base + high_pc;
886 xbt_die("Unexpected class for DW_AT_high_pc");
891 if (klass == simgrid::dwarf::TagClass::Subprogram) {
892 Dwarf_Attribute attr_frame_base;
893 if (dwarf_attr_integrate(die, DW_AT_frame_base, &attr_frame_base))
894 frame.frame_base_location = simgrid::dwarf::location_list(*info,
899 MC_dwarf_handle_children(info, die, unit, &frame, ns);
901 // We sort them in order to have an (somewhat) efficient by name
903 boost::range::sort(frame.variables, MC_compare_variable);
906 if (klass == simgrid::dwarf::TagClass::Subprogram)
907 info->subprograms[frame.id] = std::move(frame);
908 else if (klass == simgrid::dwarf::TagClass::Scope)
909 parent_frame->scopes.push_back(std::move(frame));
912 static void mc_dwarf_handle_namespace_die(simgrid::mc::ObjectInformation* info,
913 Dwarf_Die * die, Dwarf_Die * unit,
914 simgrid::mc::Frame* frame,
917 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
919 xbt_die("Unexpected namespace in a subprogram");
920 char *new_ns = ns == nullptr ? xbt_strdup(name)
921 : bprintf("%s::%s", ns, name);
922 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
926 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
927 Dwarf_Die * unit, simgrid::mc::Frame* frame,
930 // For each child DIE:
933 for (res = dwarf_child(die, &child); res == 0;
934 res = dwarf_siblingof(&child, &child))
935 MC_dwarf_handle_die(info, &child, unit, frame, ns);
938 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
939 Dwarf_Die * unit, simgrid::mc::Frame* frame,
942 int tag = dwarf_tag(die);
943 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
947 case simgrid::dwarf::TagClass::Type:
948 MC_dwarf_handle_type_die(info, die, unit, frame, ns);
951 // Subprogram or scope:
952 case simgrid::dwarf::TagClass::Subprogram:
953 case simgrid::dwarf::TagClass::Scope:
954 MC_dwarf_handle_scope_die(info, die, unit, frame, ns);
958 case simgrid::dwarf::TagClass::Variable:
959 MC_dwarf_handle_variable_die(info, die, unit, frame, ns);
962 case simgrid::dwarf::TagClass::Namespace:
963 mc_dwarf_handle_namespace_die(info, die, unit, frame, ns);
973 Elf64_Half get_type(Elf* elf)
975 Elf64_Ehdr* ehdr64 = elf64_getehdr(elf);
977 return ehdr64->e_type;
978 Elf32_Ehdr* ehdr32 = elf32_getehdr(elf);
980 return ehdr32->e_type;
981 xbt_die("Could not get ELF heeader");
985 void read_dwarf_info(simgrid::mc::ObjectInformation* info, Dwarf* dwarf)
987 // For each compilation unit:
988 Dwarf_Off offset = 0;
989 Dwarf_Off next_offset = 0;
992 while (dwarf_nextcu(dwarf, offset, &next_offset, &length, nullptr, nullptr, nullptr) ==
995 if (dwarf_offdie(dwarf, offset + length, &unit_die) != nullptr)
996 MC_dwarf_handle_children(info, &unit_die, &unit_die, nullptr, nullptr);
997 offset = next_offset;
1001 /** Get the build-id (NT_GNU_BUILD_ID) from the ELF file
1003 * This build-id may is used to locate an external debug (DWARF) file
1004 * for this ELF file.
1006 * @param elf libelf handle for an ELF file
1007 * @return build-id for this ELF file (or an empty vector if none is found)
1010 std::vector<char> get_build_id(Elf* elf)
1013 // Summary: the GNU build ID is stored in a ("GNU, NT_GNU_BUILD_ID) note
1014 // found in a PT_NOTE entry in the program header table.
1017 if (elf_getphdrnum (elf, &phnum) != 0)
1018 xbt_die("Could not read program headers");
1020 // Iterate over the program headers and find the PT_NOTE ones:
1021 for (size_t i = 0; i < phnum; ++i) {
1022 GElf_Phdr phdr_temp;
1023 GElf_Phdr *phdr = gelf_getphdr(elf, i, &phdr_temp);
1024 if (phdr->p_type != PT_NOTE)
1027 Elf_Data* data = elf_getdata_rawchunk(elf, phdr->p_offset, phdr->p_filesz, ELF_T_NHDR);
1029 // Iterate over the notes and find the NT_GNU_BUILD_ID one:
1031 while (pos < data->d_size) {
1033 // Location of the name within Elf_Data:
1036 pos = gelf_getnote(data, pos, &nhdr, &name_pos, &desc_pos);
1037 // A build ID note is identified by the pair ("GNU", NT_GNU_BUILD_ID)
1038 // (a namespace and a type within this namespace):
1039 if (nhdr.n_type == NT_GNU_BUILD_ID
1040 && nhdr.n_namesz == sizeof("GNU")
1041 && memcmp((char*) data->d_buf + name_pos, "GNU", sizeof("GNU")) == 0) {
1042 XBT_DEBUG("Found GNU/NT_GNU_BUILD_ID note");
1043 char* start = (char*) data->d_buf + desc_pos;
1044 char* end = (char*) start + nhdr.n_descsz;
1045 return std::vector<char>(start, end);
1051 return std::vector<char>();
1054 static char hexdigits[16] = {
1055 '0', '1', '2', '3', '4', '5', '6', '7', '8', '9',
1056 'a', 'b', 'c', 'd', 'e', 'f'
1059 /** Binary data to hexadecimal */
1061 std::array<char, 2> to_hex(std::uint8_t byte)
1063 // Horrid double braces!
1064 // Apparently, this is needed in C++11 (not in C++14).
1065 return { { hexdigits[byte >> 4], hexdigits[byte & 0xF] } };
1068 /** Binary data to hexadecimal */
1070 std::string to_hex(const char* data, std::size_t count)
1073 res.resize(2*count);
1074 for (std::size_t i = 0; i < count; i++) {
1075 std::array<char, 2> hex_byte = to_hex(data[i]);
1076 for (int j = 0; j < 2; ++j)
1077 res[2 * i + j] = hex_byte[j];
1082 /** Binary data to hexadecimal */
1084 std::string to_hex(std::vector<char> const& data)
1086 return to_hex(data.data(), data.size());
1089 /** Base directories for external debug files */
1091 const char* debug_paths[] = {
1093 "/usr/local/lib/debug/",
1096 /** Locate an external debug file from the NT_GNU_BUILD_ID
1098 * This is one of the mechanisms used for
1099 * [separate debug files](https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html).
1102 // /usr/lib/debug/.build-id/0b/dc77f1c29aea2b14ff5acd9a19ab3175ffdeae.debug
1104 std::string find_by_build_id(std::vector<char> id)
1106 std::string filename;
1107 std::string hex = to_hex(id);
1108 for (const char* const& debug_path : debug_paths) {
1110 filename = std::string(debug_path) + ".build-id/"
1111 + to_hex(id.data(), 1) + '/'
1112 + to_hex(id.data() + 1, id.size() - 1) + ".debug";
1113 XBT_DEBUG("Checking debug file: %s", filename.c_str());
1114 if (access(filename.c_str(), F_OK) == 0) {
1115 XBT_DEBUG("Found debug file: %s\n", hex.c_str());
1119 XBT_DEBUG("Not debuf info found for build ID %s\n", hex.data());
1120 return std::string();
1123 /** @brief Populate the debugging informations of the given ELF object
1125 * Read the DWARf information of the EFFL object and populate the
1126 * lists of types, variables, functions.
1129 void MC_load_dwarf(simgrid::mc::ObjectInformation* info)
1131 if (elf_version(EV_CURRENT) == EV_NONE)
1132 xbt_die("libelf initialization error");
1134 // Open the ELF file:
1135 int fd = open(info->file_name.c_str(), O_RDONLY);
1137 xbt_die("Could not open file %s", info->file_name.c_str());
1138 Elf* elf = elf_begin(fd, ELF_C_READ, nullptr);
1140 xbt_die("Not an ELF file");
1141 Elf_Kind kind = elf_kind(elf);
1142 if (kind != ELF_K_ELF)
1143 xbt_die("Not an ELF file");
1145 // Remember if this is a `ET_EXEC` (fixed location) or `ET_DYN`:
1146 Elf64_Half type = get_type(elf);
1147 if (type == ET_EXEC)
1148 info->flags |= simgrid::mc::ObjectInformation::Executable;
1150 // Read DWARF debug information in the file:
1151 Dwarf* dwarf = dwarf_begin_elf (elf, DWARF_C_READ, nullptr);
1152 if (dwarf != nullptr) {
1153 read_dwarf_info(info, dwarf);
1161 // If there was no DWARF in the file, try to find it in a separate file.
1162 // Different methods might be used to store the DWARF informations:
1163 // * GNU NT_GNU_BUILD_ID
1165 // See https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html
1166 // for reference of what we are doing.
1168 // Try with NT_GNU_BUILD_ID: we find the build ID in the ELF file and then
1169 // use this ID to find the file in some known locations in the filesystem.
1170 std::vector<char> build_id = get_build_id(elf);
1171 if (not build_id.empty()) {
1175 // Find the debug file using the build id:
1176 std::string debug_file = find_by_build_id(build_id);
1177 if (debug_file.empty()) {
1178 std::string hex = to_hex(build_id);
1179 xbt_die("Missing debug info for %s with build-id %s\n"
1180 "You might want to install the suitable debugging package.\n",
1181 info->file_name.c_str(), hex.c_str());
1184 // Load the DWARF info from this file:
1185 XBT_DEBUG("Load DWARF for %s from %s",
1186 info->file_name.c_str(), debug_file.c_str());
1187 fd = open(debug_file.c_str(), O_RDONLY);
1189 xbt_die("Could not open file %s", debug_file.c_str());
1190 dwarf = dwarf_begin(fd, DWARF_C_READ);
1191 if (dwarf == nullptr)
1192 xbt_die("No DWARF info in %s for %s",
1193 debug_file.c_str(), info->file_name.c_str());
1194 read_dwarf_info(info, dwarf);
1200 // TODO, try to find DWARF info using .gnu_debuglink.
1204 xbt_die("Debugging information not found for %s\n"
1205 "Try recompiling with -g\n",
1206 info->file_name.c_str());
1209 // ***** Functions index
1211 static void MC_make_functions_index(simgrid::mc::ObjectInformation* info)
1213 info->functions_index.clear();
1215 for (auto& e : info->subprograms) {
1216 if (e.second.range.begin() == 0)
1218 simgrid::mc::FunctionIndexEntry entry;
1219 entry.low_pc = (void*) e.second.range.begin();
1220 entry.function = &e.second;
1221 info->functions_index.push_back(entry);
1224 info->functions_index.shrink_to_fit();
1226 // Sort the array by low_pc:
1227 boost::range::sort(info->functions_index,
1228 [](simgrid::mc::FunctionIndexEntry const& a,
1229 simgrid::mc::FunctionIndexEntry const& b)
1231 return a.low_pc < b.low_pc;
1235 static void MC_post_process_variables(simgrid::mc::ObjectInformation* info)
1237 // Someone needs this to be sorted but who?
1238 boost::range::sort(info->global_variables, MC_compare_variable);
1240 for (simgrid::mc::Variable& variable : info->global_variables)
1241 if (variable.type_id)
1242 variable.type = simgrid::util::find_map_ptr(
1243 info->types, variable.type_id);
1246 static void mc_post_process_scope(simgrid::mc::ObjectInformation* info, simgrid::mc::Frame* scope)
1249 if (scope->tag == DW_TAG_inlined_subroutine) {
1250 // Attach correct namespaced name in inlined subroutine:
1251 auto i = info->subprograms.find(scope->abstract_origin_id);
1252 xbt_assert(i != info->subprograms.end(),
1253 "Could not lookup abstract origin %" PRIx64,
1254 (std::uint64_t) scope->abstract_origin_id);
1255 scope->name = i->second.name;
1259 for (simgrid::mc::Variable& variable : scope->variables)
1260 if (variable.type_id)
1261 variable.type = simgrid::util::find_map_ptr(
1262 info->types, variable.type_id);
1264 // Recursive post-processing of nested-scopes:
1265 for (simgrid::mc::Frame& nested_scope : scope->scopes)
1266 mc_post_process_scope(info, &nested_scope);
1270 simgrid::mc::Type* MC_resolve_type(
1271 simgrid::mc::ObjectInformation* info, unsigned type_id)
1275 simgrid::mc::Type* type = simgrid::util::find_map_ptr(info->types, type_id);
1276 if (type == nullptr)
1279 // We already have the information on the type:
1280 if (type->byte_size != 0)
1283 // Don't have a name, we can't find a more complete version:
1284 if (type->name.empty())
1287 // Try to find a more complete description of the type:
1288 // We need to fix in order to support C++.
1289 simgrid::mc::Type** subtype = simgrid::util::find_map_ptr(
1290 info->full_types_by_name, type->name);
1296 static void MC_post_process_types(simgrid::mc::ObjectInformation* info)
1298 // Lookup "subtype" field:
1299 for (auto& i : info->types) {
1300 i.second.subtype = MC_resolve_type(info, i.second.type_id);
1301 for (simgrid::mc::Member& member : i.second.members)
1302 member.type = MC_resolve_type(info, member.type_id);
1309 /** @brief Finds informations about a given shared object/executable */
1310 std::shared_ptr<simgrid::mc::ObjectInformation> createObjectInformation(
1311 std::vector<simgrid::xbt::VmMap> const& maps, const char *name)
1313 std::shared_ptr<simgrid::mc::ObjectInformation> result =
1314 std::make_shared<simgrid::mc::ObjectInformation>();
1315 result->file_name = name;
1316 simgrid::mc::find_object_address(maps, result.get());
1317 MC_load_dwarf(result.get());
1318 MC_post_process_variables(result.get());
1319 MC_post_process_types(result.get());
1320 for (auto& entry : result.get()->subprograms)
1321 mc_post_process_scope(result.get(), &entry.second);
1322 MC_make_functions_index(result.get());
1326 /*************************************************************************/
1328 void postProcessObjectInformation(simgrid::mc::RemoteClient* process, simgrid::mc::ObjectInformation* info)
1330 for (auto& t : info->types) {
1332 simgrid::mc::Type* type = &(t.second);
1333 simgrid::mc::Type* subtype = type;
1334 while (subtype->type == DW_TAG_typedef
1335 || subtype->type == DW_TAG_volatile_type
1336 || subtype->type == DW_TAG_const_type)
1337 if (subtype->subtype)
1338 subtype = subtype->subtype;
1342 // Resolve full_type:
1343 if (not subtype->name.empty() && subtype->byte_size == 0)
1344 for (auto const& object_info : process->object_infos) {
1345 auto i = object_info->full_types_by_name.find(subtype->name);
1346 if (i != object_info->full_types_by_name.end() && not i->second->name.empty() && i->second->byte_size) {
1347 type->full_type = i->second;
1351 else type->full_type = subtype;
1362 /** Convert a DWARF register into a libunwind register
1364 * DWARF and libunwind does not use the same convention for numbering the
1365 * registers on some architectures. The function makes the necessary
1368 int dwarf_register_to_libunwind(int dwarf_register)
1370 #if defined(__x86_64__)
1371 // It seems for this arch, DWARF and libunwind agree in the numbering:
1372 return dwarf_register;
1373 #elif defined(__i386__)
1374 // Couldn't find the authoritative source of information for this.
1375 // This is inspired from http://source.winehq.org/source/dlls/dbghelp/cpu_i386.c#L517.
1376 switch (dwarf_register) {
1396 return UNW_X86_EFLAGS;
1426 xbt_die("Bad/unknown register number.");
1429 #error This architecture is not supported yet for DWARF expression evaluation.