1 /* Copyright (c) 2014-2022. 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. */
8 #include <sys/mman.h> // PROT_READ and friends
11 #include "src/mc/inspect/Frame.hpp"
12 #include "src/mc/inspect/ObjectInformation.hpp"
13 #include "src/mc/inspect/Variable.hpp"
14 #include "src/mc/mc_private.hpp"
15 #include "xbt/file.hpp"
20 /* For an executable object, addresses are virtual address (there is no offset) i.e.
21 * \f$\text{virtual address} = \{dwarf address}\f$
23 * For a shared object, the addresses are offset from the beginning of the shared object (the base address of the
24 * mapped shared object must be used as offset
25 * i.e. \f$\text{virtual address} = \text{shared object base address}
26 * + \text{dwarf address}\f$.
28 void* ObjectInformation::base_address() const
30 // For an executable (more precisely for an ET_EXEC) the base it 0:
31 if (this->executable())
34 // For an a shared-object (ET_DYN, including position-independent executables) the base address is its lowest address:
35 void* result = this->start_exec;
36 if (this->start_rw != nullptr && result > (void*)this->start_rw)
37 result = this->start_rw;
38 if (this->start_ro != nullptr && result > (void*)this->start_ro)
39 result = this->start_ro;
43 Frame* ObjectInformation::find_function(const void* ip)
45 ensure_dwarf_loaded();
47 /* This is implemented by binary search on a sorted array.
49 * We do quite a lot of those so we want this to be cache efficient.
50 * We pack the only information we need in the index entries in order
51 * to successfully do the binary search. We do not need the high_pc
52 * during the binary search (only at the end) so it is not included
53 * in the index entry. We could use parallel arrays as well.
55 * Note the usage of reverse iterators to match the correct interval.
57 auto pos = std::lower_bound(this->functions_index.rbegin(), this->functions_index.rend(), ip,
58 [](auto const& func, auto const* addr) { return func.low_pc > addr; });
60 /* At this point, the search is over.
61 * Either we have found the correct function or we do not know
62 * any function corresponding to this instruction address.
63 * Only at the point do we dereference the function pointer. */
64 return (pos != this->functions_index.rend() && reinterpret_cast<std::uint64_t>(ip) < pos->function->range.end())
69 const Variable* ObjectInformation::find_variable(const char* var_name)
71 ensure_dwarf_loaded();
73 auto pos = std::lower_bound(this->global_variables.begin(), this->global_variables.end(), var_name,
74 [](auto const& var, const char* name) { return var.name < name; });
75 return (pos != this->global_variables.end() && pos->name == var_name) ? &(*pos) : nullptr;
78 void ObjectInformation::remove_global_variable(const char* var_name)
81 auto pos1 = std::lower_bound(this->global_variables.begin(), this->global_variables.end(), var_name,
82 [](auto const& var, const char* name) { return var.name < name; });
83 // Find the whole range:
84 auto pos2 = std::upper_bound(pos1, this->global_variables.end(), var_name,
85 [](const char* name, auto const& var) { return name < var.name; });
86 // Remove the whole range:
87 this->global_variables.erase(pos1, pos2);
90 /** Ignore a local variable in a scope
92 * Ignore all instances of variables with a given name in any (possibly inlined) subprogram with a given namespaced
95 * @param var_name Name of the local variable to ignore
96 * @param subprogram_name Name of the subprogram to ignore (nullptr for any)
97 * @param subprogram (possibly inlined) Subprogram of the scope current scope
98 * @param scope Current scope
100 static void remove_local_variable(Frame& scope, const char* var_name, const char* subprogram_name,
101 Frame const& subprogram)
103 // If the current subprogram matches the given name:
104 if (subprogram_name == nullptr || (not subprogram.name.empty() && subprogram.name == subprogram_name)) {
105 // Try to find the variable and remove it:
108 auto pos = std::lower_bound(scope.variables.begin(), scope.variables.end(), var_name,
109 [](auto const& var, const char* name) { return var.name < name; });
110 if (pos != scope.variables.end() && pos->name == var_name) {
111 // Variable found, remove it:
112 scope.variables.erase(pos);
116 // And recursive processing in nested scopes:
117 for (Frame& nested_scope : scope.scopes) {
118 // The new scope may be an inlined subroutine, in this case we want to use its
119 // namespaced name in recursive calls:
120 Frame const& nested_subprogram = nested_scope.tag == DW_TAG_inlined_subroutine ? nested_scope : subprogram;
121 remove_local_variable(nested_scope, var_name, subprogram_name, nested_subprogram);
125 void ObjectInformation::remove_local_variable(const char* var_name, const char* subprogram_name)
127 for (auto& entry : this->subprograms)
128 mc::remove_local_variable(entry.second, var_name, subprogram_name, entry.second);
131 /** @brief Fills the position of the segments (executable, read-only, read/write) */
132 // TODO, use the ELF segment information for more robustness
133 void find_object_address(std::vector<xbt::VmMap> const& maps, ObjectInformation* result)
135 const int PROT_RW = PROT_READ | PROT_WRITE;
136 const int PROT_RX = PROT_READ | PROT_EXEC;
138 std::string name = xbt::Path(result->file_name).get_base_name();
140 for (size_t i = 0; i < maps.size(); ++i) {
141 simgrid::xbt::VmMap const& reg = maps[i];
142 if (maps[i].pathname.empty())
144 std::string map_basename = simgrid::xbt::Path(maps[i].pathname).get_base_name();
145 if (map_basename != name)
148 // This is the non-GNU_RELRO-part of the data segment:
149 if (reg.prot == PROT_RW) {
150 xbt_assert(not result->start_rw, "Multiple read-write segments for %s, not supported", maps[i].pathname.c_str());
151 result->start_rw = (char*)reg.start_addr;
152 result->end_rw = (char*)reg.end_addr;
154 // The next VMA might be end of the data segment:
155 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
156 maps[i + 1].start_addr == reg.end_addr)
157 result->end_rw = (char*)maps[i + 1].end_addr;
160 // This is the text segment:
161 else if (reg.prot == PROT_RX) {
162 xbt_assert(not result->start_exec, "Multiple executable segments for %s, not supported",
163 maps[i].pathname.c_str());
164 result->start_exec = (char*)reg.start_addr;
165 result->end_exec = (char*)reg.end_addr;
167 // The next VMA might be end of the data segment:
168 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
169 maps[i + 1].start_addr == reg.end_addr) {
170 result->start_rw = (char*)maps[i + 1].start_addr;
171 result->end_rw = (char*)maps[i + 1].end_addr;
175 // This is the GNU_RELRO-part of the data segment:
176 else if (reg.prot == PROT_READ) {
177 xbt_assert(not result->start_ro,
178 "Multiple read-only segments for %s, not supported. Compiling with the following may help: "
179 "-g -Wl,-znorelro -Wl,-znoseparate-code",
180 maps[i].pathname.c_str());
181 result->start_ro = (char*)reg.start_addr;
182 result->end_ro = (char*)reg.end_addr;
186 result->start = result->start_rw;
187 if ((const void*)result->start_ro < result->start)
188 result->start = result->start_ro;
189 if ((const void*)result->start_exec < result->start)
190 result->start = result->start_exec;
192 result->end = result->end_rw;
193 if (result->end_ro && (const void*)result->end_ro > result->end)
194 result->end = result->end_ro;
195 if (result->end_exec && (const void*)result->end_exec > result->end)
196 result->end = result->end_exec;
198 xbt_assert(result->start_exec || result->start_rw || result->start_ro);
202 } // namespace simgrid