1 /* Copyright (c) 2008-2015. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
14 #define DW_LANG_Objc DW_LANG_ObjC /* fix spelling error in older dwarf.h */
16 #include <elfutils/libdw.h>
18 #include <simgrid_config.h>
19 #include <simgrid/util.hpp>
21 #include <xbt/sysdep.h>
23 #include <simgrid/util.hpp>
25 #include "mc_object_info.h"
26 #include "mc_private.h"
27 #include "mc_dwarf.hpp"
29 #include "mc/Process.hpp"
30 #include "mc/ObjectInformation.hpp"
31 #include "mc/Variable.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 consant 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,
57 /** \brief Computes the number of elements of a given DW_TAG_array_type.
59 * \param die DIE for the DW_TAG_array_type
61 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit);
63 /** \brief Process a DIE
65 * \param info the resulting object fot the library/binary file (output)
66 * \param die the current DIE
67 * \param unit the DIE of the compile unit of the current DIE
68 * \param frame containg frame if any
70 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
71 Dwarf_Die * unit, simgrid::mc::Frame* frame,
74 /** \brief Process a type DIE
76 static void MC_dwarf_handle_type_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
77 Dwarf_Die * unit, simgrid::mc::Frame* frame,
80 /** \brief Calls MC_dwarf_handle_die on all childrend of the given die
82 * \param info the resulting object fot the library/binary file (output)
83 * \param die the current DIE
84 * \param unit the DIE of the compile unit of the current DIE
85 * \param frame containg frame if any
87 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
88 Dwarf_Die * unit, simgrid::mc::Frame* frame,
91 /** \brief Handle a variable (DW_TAG_variable or other)
93 * \param info the resulting object fot the library/binary file (output)
94 * \param die the current DIE
95 * \param unit the DIE of the compile unit of the current DIE
96 * \param frame containg frame if any
98 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
99 Dwarf_Die * unit, simgrid::mc::Frame* frame,
102 /** \brief Get the DW_TAG_type of the DIE
105 * \return DW_TAG_type attribute as a new string (NULL if none)
107 static std::uint64_t MC_dwarf_at_type(Dwarf_Die * die);
112 enum class TagClass {
121 /*** Class of forms defined in the DWARF standard */
122 enum class FormClass {
124 Address, // Location in the program's address space
125 Block, // Arbitrary block of bytes
128 Flag, // Boolean value
129 Reference, // Reference to another DIE
130 ExprLoc, // DWARF expression/location description
138 TagClass classify_tag(int tag)
142 case DW_TAG_array_type:
143 case DW_TAG_class_type:
144 case DW_TAG_enumeration_type:
146 case DW_TAG_pointer_type:
147 case DW_TAG_reference_type:
148 case DW_TAG_rvalue_reference_type:
149 case DW_TAG_string_type:
150 case DW_TAG_structure_type:
151 case DW_TAG_subroutine_type:
152 case DW_TAG_union_type:
153 case DW_TAG_ptr_to_member_type:
154 case DW_TAG_set_type:
155 case DW_TAG_subrange_type:
156 case DW_TAG_base_type:
157 case DW_TAG_const_type:
158 case DW_TAG_file_type:
159 case DW_TAG_packed_type:
160 case DW_TAG_volatile_type:
161 case DW_TAG_restrict_type:
162 case DW_TAG_interface_type:
163 case DW_TAG_unspecified_type:
164 case DW_TAG_shared_type:
165 return TagClass::Type;
167 case DW_TAG_subprogram:
168 return TagClass::Subprogram;
170 case DW_TAG_variable:
171 case DW_TAG_formal_parameter:
172 return TagClass::Variable;
174 case DW_TAG_lexical_block:
175 case DW_TAG_try_block:
176 case DW_TAG_catch_block:
177 case DW_TAG_inlined_subroutine:
178 case DW_TAG_with_stmt:
179 return TagClass::Scope;
181 case DW_TAG_namespace:
182 return TagClass::Namespace;
185 return TagClass::Unknown;
189 /** \brief Find the DWARF data class for a given DWARF data form
191 * This mapping is defined in the DWARF spec.
193 * \param form The form (values taken from the DWARF spec)
194 * \return An internal representation for the corresponding class
197 FormClass classify_form(int form)
201 return FormClass::Address;
206 return FormClass::Block;
213 return FormClass::Constant;
216 return FormClass::String;
217 case DW_FORM_ref_addr:
222 case DW_FORM_ref_udata:
223 return FormClass::Reference;
225 case DW_FORM_flag_present:
226 return FormClass::Flag;
227 case DW_FORM_exprloc:
228 return FormClass::ExprLoc;
232 return FormClass::Unknown;
236 /** \brief Get the name of the tag of a given DIE
239 * \return name of the tag of this DIE
242 const char *tagname(Dwarf_Die * die)
244 return simgrid::dwarf::tagname(dwarf_tag(die));
252 /** \brief Get an attribute of a given DIE as a string
255 * \param attribute attribute
256 * \return value of the given attribute of the given DIE
258 static const char *MC_dwarf_attr_integrate_string(Dwarf_Die * die,
261 Dwarf_Attribute attr;
262 if (!dwarf_attr_integrate(die, attribute, &attr)) {
265 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 NULL)
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) == NULL)
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) == NULL)
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 NULL)
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) == NULL)
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) == NULL)
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 (!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 (!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 (!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 (%i) / form 0x%x as DW_AT_member_location",
543 (int) form_class, form);
548 /** \brief Populate the list of members of a type
550 * \param info ELF object containing the type DIE
551 * \param die DIE of the type
552 * \param unit DIE of the compilation unit containing the type DIE
553 * \param type the type
555 static void MC_dwarf_add_members(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
556 Dwarf_Die * unit, simgrid::mc::Type* type)
560 xbt_assert(type->members.empty());
561 for (res = dwarf_child(die, &child); res == 0;
562 res = dwarf_siblingof(&child, &child)) {
563 int tag = dwarf_tag(&child);
564 if (tag == DW_TAG_member || tag == DW_TAG_inheritance) {
566 // Skip declarations:
567 if (MC_dwarf_attr_flag(&child, DW_AT_declaration, false))
570 // Skip compile time constants:
571 if (dwarf_hasattr(&child, DW_AT_const_value))
574 // TODO, we should use another type (because is is not a type but a member)
575 simgrid::mc::Member member;
576 member.inheritance = tag == DW_TAG_inheritance;
578 const char *name = MC_dwarf_attr_integrate_string(&child, DW_AT_name);
582 MC_dwarf_attr_integrate_uint(&child, DW_AT_byte_size, 0);
583 member.type_id = MC_dwarf_at_type(&child);
585 if (dwarf_hasattr(&child, DW_AT_data_bit_offset))
586 xbt_die("Can't groke DW_AT_data_bit_offset.");
588 MC_dwarf_fill_member_location(type, &member, &child);
591 xbt_die("Missing type for member %s of <%" PRIx64 ">%s",
593 (uint64_t) type->id, type->name.c_str());
595 type->members.push_back(std::move(member));
600 /** \brief Create a MC type object from a DIE
602 * \param info current object info object
603 * \param DIE (for a given type);
604 * \param unit compilation unit of the current DIE
605 * \return MC representation of the type
607 static simgrid::mc::Type MC_dwarf_die_to_type(
608 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
609 Dwarf_Die * unit, simgrid::mc::Frame* frame,
612 simgrid::mc::Type type;
613 type.type = dwarf_tag(die);
614 type.name = std::string();
615 type.element_count = -1;
618 type.id = dwarf_dieoffset(die);
620 const char *prefix = "";
622 case DW_TAG_structure_type:
625 case DW_TAG_union_type:
628 case DW_TAG_class_type:
635 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
637 char* full_name = ns ? bprintf("%s%s::%s", prefix, ns, name) :
638 bprintf("%s%s", prefix, name);
639 type.name = std::string(full_name);
643 type.type_id = MC_dwarf_at_type(die);
645 // Some compilers do not emit DW_AT_byte_size for pointer_type,
646 // so we fill this. We currently assume that the model-checked process is in
647 // the same architecture..
648 if (type.type == DW_TAG_pointer_type)
649 type.byte_size = sizeof(void*);
651 // Computation of the byte_size;
652 if (dwarf_hasattr_integrate(die, DW_AT_byte_size))
653 type.byte_size = MC_dwarf_attr_integrate_uint(die, DW_AT_byte_size, 0);
654 else if (type.type == DW_TAG_array_type
655 || type.type == DW_TAG_structure_type
656 || type.type == DW_TAG_class_type) {
658 if (dwarf_aggregate_size(die, &size) == 0)
659 type.byte_size = size;
663 case DW_TAG_array_type:
664 type.element_count = MC_dwarf_array_element_count(die, unit);
665 // TODO, handle DW_byte_stride and (not) DW_bit_stride
668 case DW_TAG_pointer_type:
669 case DW_TAG_reference_type:
670 case DW_TAG_rvalue_reference_type:
671 type.is_pointer_type = 1;
674 case DW_TAG_structure_type:
675 case DW_TAG_union_type:
676 case DW_TAG_class_type:
677 MC_dwarf_add_members(info, die, unit, &type);
678 char *new_ns = ns == NULL ? xbt_strdup(type.name.c_str())
679 : bprintf("%s::%s", ns, name);
680 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
685 return std::move(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 (!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(
701 simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
702 Dwarf_Die * unit, simgrid::mc::Frame* frame,
705 // Skip declarations:
706 if (MC_dwarf_attr_flag(die, DW_AT_declaration, false))
709 // Skip compile time constants:
710 if (dwarf_hasattr(die, DW_AT_const_value))
713 Dwarf_Attribute attr_location;
714 if (dwarf_attr(die, DW_AT_location, &attr_location) == NULL)
715 // No location: do not add it ?
718 std::unique_ptr<simgrid::mc::Variable> variable =
719 std::unique_ptr<simgrid::mc::Variable>(new simgrid::mc::Variable());
720 variable->dwarf_offset = dwarf_dieoffset(die);
721 variable->global = frame == NULL; // Can be override base on DW_AT_location
722 variable->object_info = info;
724 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
726 variable->name = name;
727 variable->type_id = MC_dwarf_at_type(die);
729 int form = dwarf_whatform(&attr_location);
730 simgrid::dwarf::FormClass form_class;
731 if (form == DW_FORM_sec_offset)
732 form_class = simgrid::dwarf::FormClass::Constant;
734 form_class = simgrid::dwarf::classify_form(form);
735 switch (form_class) {
736 case simgrid::dwarf::FormClass::ExprLoc:
737 case simgrid::dwarf::FormClass::Block:
738 // Location expression:
742 if (dwarf_getlocation(&attr_location, &expr, &len)) {
744 "Could not read location expression in DW_AT_location "
745 "of variable <%" PRIx64 ">%s",
746 (uint64_t) variable->dwarf_offset,
747 variable->name.c_str());
750 if (len == 1 && expr[0].atom == DW_OP_addr) {
751 variable->global = 1;
752 uintptr_t offset = (uintptr_t) expr[0].number;
753 uintptr_t base = (uintptr_t) info->base_address();
754 variable->address = (void *) (base + offset);
756 simgrid::dwarf::LocationListEntry entry;
757 entry.expression = {expr, expr + len};
758 variable->location_list = { std::move(entry) };
764 case simgrid::dwarf::FormClass::LocListPtr:
765 case simgrid::dwarf::FormClass::Constant:
766 // Reference to location list:
767 variable->location_list = simgrid::dwarf::location_list(
768 *info, attr_location);
772 xbt_die("Unexpected form 0x%x (%i), class 0x%x (%i) list for location "
773 "in <%" PRIx64 ">%s",
774 form, form, (int) form_class, (int) form_class,
775 (uint64_t) variable->dwarf_offset,
776 variable->name.c_str());
779 // Handle start_scope:
780 if (dwarf_hasattr(die, DW_AT_start_scope)) {
781 Dwarf_Attribute attr;
782 dwarf_attr(die, DW_AT_start_scope, &attr);
783 int form = dwarf_whatform(&attr);
784 simgrid::dwarf::FormClass form_class = simgrid::dwarf::classify_form(form);
785 switch (form_class) {
786 case simgrid::dwarf::FormClass::Constant:
789 variable->start_scope =
790 dwarf_formudata(&attr, &value) == 0 ? (size_t) value : 0;
794 case simgrid::dwarf::FormClass::RangeListPtr: // TODO
797 ("Unhandled form 0x%x, class 0x%X for DW_AT_start_scope of variable %s",
798 form, (int) form_class, name == NULL ? "?" : name);
802 if (ns && variable->global)
804 std::string(ns) + "::" + variable->name;
806 // The current code needs a variable name,
807 // generate a fake one:
808 if (variable->name.empty())
810 "@anonymous#" + std::to_string(mc_anonymous_variable_index++);
812 return std::move(variable);
815 static void MC_dwarf_handle_variable_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
816 Dwarf_Die * unit, simgrid::mc::Frame* frame,
819 std::unique_ptr<simgrid::mc::Variable> variable =
820 MC_die_to_variable(info, die, unit, frame, ns);
823 // Those arrays are sorted later:
824 else if (variable->global)
825 info->global_variables.push_back(std::move(*variable));
826 else if (frame != nullptr)
827 frame->variables.push_back(std::move(*variable));
829 xbt_die("No frame for this local variable");
832 static void MC_dwarf_handle_scope_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
833 Dwarf_Die * unit, simgrid::mc::Frame* parent_frame,
836 // TODO, handle DW_TAG_type/DW_TAG_location for DW_TAG_with_stmt
837 int tag = dwarf_tag(die);
838 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
840 // (Template) Subprogram declaration:
841 if (klass == simgrid::dwarf::TagClass::Subprogram
842 && MC_dwarf_attr_flag(die, DW_AT_declaration, false))
845 if (klass == simgrid::dwarf::TagClass::Scope)
846 xbt_assert(parent_frame, "No parent scope for this scope");
848 simgrid::mc::Frame frame;
851 frame.id = dwarf_dieoffset(die);
852 frame.object_info = info;
854 if (klass == simgrid::dwarf::TagClass::Subprogram) {
855 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
857 frame.name = std::string(ns) + "::" + name;
864 frame.abstract_origin_id =
865 MC_dwarf_attr_dieoffset(die, DW_AT_abstract_origin);
867 // This is the base address for DWARF addresses.
868 // Relocated addresses are offset from this base address.
869 // See DWARF4 spec 7.5
870 void *base = info->base_address();
872 // TODO, support DW_AT_ranges
873 uint64_t low_pc = MC_dwarf_attr_integrate_addr(die, DW_AT_low_pc);
874 frame.low_pc = low_pc ? ((char *) base) + low_pc : 0;
877 Dwarf_Attribute attr;
878 if (!dwarf_attr_integrate(die, DW_AT_high_pc, &attr)) {
879 xbt_die("Missing DW_AT_high_pc matching with DW_AT_low_pc");
885 switch (simgrid::dwarf::classify_form(dwarf_whatform(&attr))) {
887 // DW_AT_high_pc if an offset from the low_pc:
888 case simgrid::dwarf::FormClass::Constant:
890 if (dwarf_formsdata(&attr, &offset) != 0)
891 xbt_die("Could not read constant");
892 frame.high_pc = (void *) ((char *) frame.low_pc + offset);
895 // DW_AT_high_pc is a relocatable address:
896 case simgrid::dwarf::FormClass::Address:
897 if (dwarf_formaddr(&attr, &high_pc) != 0)
898 xbt_die("Could not read address");
899 frame.high_pc = ((char *) base) + high_pc;
903 xbt_die("Unexpected class for DW_AT_high_pc");
908 if (klass == simgrid::dwarf::TagClass::Subprogram) {
909 Dwarf_Attribute attr_frame_base;
910 if (dwarf_attr_integrate(die, DW_AT_frame_base, &attr_frame_base))
911 frame.frame_base_location = simgrid::dwarf::location_list(*info,
916 MC_dwarf_handle_children(info, die, unit, &frame, ns);
918 // Someone needs this to be sorted but who?
919 std::sort(frame.variables.begin(), frame.variables.end(),
920 MC_compare_variable);
923 if (klass == simgrid::dwarf::TagClass::Subprogram)
924 info->subprograms[frame.id] = frame;
925 else if (klass == simgrid::dwarf::TagClass::Scope)
926 parent_frame->scopes.push_back(std::move(frame));
929 static void mc_dwarf_handle_namespace_die(simgrid::mc::ObjectInformation* info,
930 Dwarf_Die * die, Dwarf_Die * unit,
931 simgrid::mc::Frame* frame,
934 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
936 xbt_die("Unexpected namespace in a subprogram");
937 char *new_ns = ns == NULL ? xbt_strdup(name)
938 : bprintf("%s::%s", ns, name);
939 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
943 static void MC_dwarf_handle_children(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
944 Dwarf_Die * unit, simgrid::mc::Frame* frame,
947 // For each child DIE:
950 for (res = dwarf_child(die, &child); res == 0;
951 res = dwarf_siblingof(&child, &child)) {
952 MC_dwarf_handle_die(info, &child, unit, frame, ns);
956 static void MC_dwarf_handle_die(simgrid::mc::ObjectInformation* info, Dwarf_Die * die,
957 Dwarf_Die * unit, simgrid::mc::Frame* frame,
960 int tag = dwarf_tag(die);
961 simgrid::dwarf::TagClass klass = simgrid::dwarf::classify_tag(tag);
965 case simgrid::dwarf::TagClass::Type:
966 MC_dwarf_handle_type_die(info, die, unit, frame, ns);
969 // Subprogram or scope:
970 case simgrid::dwarf::TagClass::Subprogram:
971 case simgrid::dwarf::TagClass::Scope:
972 MC_dwarf_handle_scope_die(info, die, unit, frame, ns);
976 case simgrid::dwarf::TagClass::Variable:
977 MC_dwarf_handle_variable_die(info, die, unit, frame, ns);
980 case simgrid::dwarf::TagClass::Namespace:
981 mc_dwarf_handle_namespace_die(info, die, unit, frame, ns);
991 Elf64_Half MC_dwarf_elf_type(Dwarf* dwarf)
993 Elf* elf = dwarf_getelf(dwarf);
994 Elf64_Ehdr* ehdr64 = elf64_getehdr(elf);
996 return ehdr64->e_type;
997 Elf32_Ehdr* ehdr32 = elf32_getehdr(elf);
999 return ehdr32->e_type;
1000 xbt_die("Could not get ELF heeader");
1003 /** \brief Populate the debugging informations of the given ELF object
1005 * Read the DWARf information of the EFFL object and populate the
1006 * lists of types, variables, functions.
1009 void MC_dwarf_get_variables(simgrid::mc::ObjectInformation* info)
1011 int fd = open(info->file_name.c_str(), O_RDONLY);
1013 xbt_die("Could not open file %s", info->file_name.c_str());
1014 Dwarf *dwarf = dwarf_begin(fd, DWARF_C_READ);
1016 xbt_die("Missing debugging information in %s\n"
1017 "Your program and its dependencies must have debugging information.\n"
1018 "You might want to recompile with -g or install the suitable debugging package.\n",
1019 info->file_name.c_str());
1021 Elf64_Half elf_type = MC_dwarf_elf_type(dwarf);
1022 if (elf_type == ET_EXEC)
1023 info->flags |= simgrid::mc::ObjectInformation::Executable;
1025 // For each compilation unit:
1026 Dwarf_Off offset = 0;
1027 Dwarf_Off next_offset = 0;
1030 while (dwarf_nextcu(dwarf, offset, &next_offset, &length, NULL, NULL, NULL) ==
1033 if (dwarf_offdie(dwarf, offset + length, &unit_die) != NULL)
1034 MC_dwarf_handle_children(info, &unit_die, &unit_die, NULL, NULL);
1035 offset = next_offset;
1042 // ***** Functions index
1044 static int MC_compare_frame_index_items(simgrid::mc::FunctionIndexEntry* a,
1045 simgrid::mc::FunctionIndexEntry* b)
1047 if (a->low_pc < b->low_pc)
1049 else if (a->low_pc == b->low_pc)
1055 static void MC_make_functions_index(simgrid::mc::ObjectInformation* info)
1057 info->functions_index.clear();
1059 for (auto& e : info->subprograms) {
1060 if (e.second.low_pc == nullptr)
1062 simgrid::mc::FunctionIndexEntry entry;
1063 entry.low_pc = e.second.low_pc;
1064 entry.function = &e.second;
1065 info->functions_index.push_back(entry);
1068 info->functions_index.shrink_to_fit();
1070 // Sort the array by low_pc:
1071 std::sort(info->functions_index.begin(), info->functions_index.end(),
1072 [](simgrid::mc::FunctionIndexEntry const& a,
1073 simgrid::mc::FunctionIndexEntry const& b)
1075 return a.low_pc < b.low_pc;
1079 static void MC_post_process_variables(simgrid::mc::ObjectInformation* info)
1081 // Someone needs this to be sorted but who?
1082 std::sort(info->global_variables.begin(), info->global_variables.end(),
1083 MC_compare_variable);
1085 for(simgrid::mc::Variable& variable : info->global_variables)
1086 if (variable.type_id)
1087 variable.type = simgrid::util::find_map_ptr(
1088 info->types, variable.type_id);
1091 static void mc_post_process_scope(simgrid::mc::ObjectInformation* info, simgrid::mc::Frame* scope)
1094 if (scope->tag == DW_TAG_inlined_subroutine) {
1095 // Attach correct namespaced name in inlined subroutine:
1096 auto i = info->subprograms.find(scope->abstract_origin_id);
1097 xbt_assert(i != info->subprograms.end(),
1098 "Could not lookup abstract origin %" PRIx64,
1099 (std::uint64_t) scope->abstract_origin_id);
1100 scope->name = i->second.name;
1104 for (simgrid::mc::Variable& variable : scope->variables)
1105 if (variable.type_id)
1106 variable.type = simgrid::util::find_map_ptr(
1107 info->types, variable.type_id);
1109 // Recursive post-processing of nested-scopes:
1110 for (simgrid::mc::Frame& nested_scope : scope->scopes)
1111 mc_post_process_scope(info, &nested_scope);
1116 simgrid::mc::Type* MC_resolve_type(
1117 simgrid::mc::ObjectInformation* info, unsigned type_id)
1121 simgrid::mc::Type* type = simgrid::util::find_map_ptr(info->types, type_id);
1122 if (type == nullptr)
1125 // We already have the information on the type:
1126 if (type->byte_size != 0)
1129 // Don't have a name, we can't find a more complete version:
1130 if (type->name.empty())
1133 // Try to find a more complete description of the type:
1134 // We need to fix in order to support C++.
1135 simgrid::mc::Type** subtype = simgrid::util::find_map_ptr(
1136 info->full_types_by_name, type->name);
1142 static void MC_post_process_types(simgrid::mc::ObjectInformation* info)
1144 // Lookup "subtype" field:
1145 for(auto& i : info->types) {
1146 i.second.subtype = MC_resolve_type(info, i.second.type_id);
1147 for (simgrid::mc::Member& member : i.second.members)
1148 member.type = MC_resolve_type(info, member.type_id);
1152 /** \brief Finds informations about a given shared object/executable */
1153 std::shared_ptr<simgrid::mc::ObjectInformation> MC_find_object_info(
1154 std::vector<simgrid::mc::VmMap> const& maps, const char *name)
1156 std::shared_ptr<simgrid::mc::ObjectInformation> result =
1157 std::make_shared<simgrid::mc::ObjectInformation>();
1158 result->file_name = name;
1159 MC_find_object_address(maps, result.get());
1160 MC_dwarf_get_variables(result.get());
1161 MC_post_process_variables(result.get());
1162 MC_post_process_types(result.get());
1163 for (auto& entry : result.get()->subprograms)
1164 mc_post_process_scope(result.get(), &entry.second);
1165 MC_make_functions_index(result.get());
1166 return std::move(result);
1169 /*************************************************************************/
1171 void MC_post_process_object_info(simgrid::mc::Process* process, simgrid::mc::ObjectInformation* info)
1173 for (auto& i : info->types) {
1175 simgrid::mc::Type* type = &(i.second);
1176 simgrid::mc::Type* subtype = type;
1177 while (subtype->type == DW_TAG_typedef
1178 || subtype->type == DW_TAG_volatile_type
1179 || subtype->type == DW_TAG_const_type)
1180 if (subtype->subtype)
1181 subtype = subtype->subtype;
1185 // Resolve full_type:
1186 if (!subtype->name.empty() && subtype->byte_size == 0) {
1187 for (auto const& object_info : process->object_infos) {
1188 auto i = object_info->full_types_by_name.find(subtype->name);
1189 if (i != object_info->full_types_by_name.end()
1190 && !i->second->name.empty() && i->second->byte_size) {
1191 type->full_type = i->second;
1195 } else type->full_type = subtype;
1203 /** Convert a DWARF register into a libunwind register
1205 * DWARF and libunwind does not use the same convention for numbering the
1206 * registers on some architectures. The function makes the necessary
1209 int dwarf_register_to_libunwind(int dwarf_register)
1211 #if defined(__x86_64__)
1212 // It seems for this arch, DWARF and libunwind agree in the numbering:
1213 return dwarf_register;
1214 #elif defined(__i386__)
1215 // Could't find the authoritative source of information for this.
1216 // This is inspired from http://source.winehq.org/source/dlls/dbghelp/cpu_i386.c#L517.
1217 switch (dwarf_register) {
1237 return UNW_X86_EFLAGS;
1267 xbt_die("Bad/unknown register number.");
1270 #error This architecture is not supported yet for DWARF expression evaluation.