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 "src/internal_config.h"
13 #include "src/mc/mc_private.hpp"
14 #include "src/smpi/include/private.hpp"
15 #include "xbt/file.hpp"
16 #include "xbt/mmalloc.h"
17 #include "xbt/module.h"
19 #include "src/xbt/mmalloc/mmprivate.h"
21 #include "src/simix/smx_private.hpp"
24 #include <libunwind.h>
26 #include "src/mc/mc_private.hpp"
29 #include "src/mc/mc_config.hpp"
30 #include "src/mc/mc_hash.hpp"
31 #include "src/mc/mc_mmu.hpp"
32 #include "src/mc/mc_smx.hpp"
33 #include "src/mc/mc_unw.hpp"
34 #include "src/mc/remote/mc_protocol.h"
35 #include "src/mc/sosp/mc_snapshot.hpp"
37 #include "src/mc/Frame.hpp"
38 #include "src/mc/ObjectInformation.hpp"
39 #include "src/mc/Variable.hpp"
40 #include "src/mc/sosp/RegionSnapshot.hpp"
42 using simgrid::mc::remote;
44 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc, "Logging specific to mc_checkpoint");
46 #define PROT_RWX (PROT_READ | PROT_WRITE | PROT_EXEC)
47 #define PROT_RW (PROT_READ | PROT_WRITE)
48 #define PROT_RX (PROT_READ | PROT_EXEC)
53 /************************************ Free functions **************************************/
54 /*****************************************************************************************/
56 /** @brief Restore a region from a snapshot
58 * @param region Target region
60 static void restore(mc_mem_region_t region)
62 switch (region->storage_type()) {
63 case simgrid::mc::StorageType::Flat:
64 mc_model_checker->process().write_bytes(region->flat_data().get(), region->size(), region->permanent_address());
67 case simgrid::mc::StorageType::Chunked:
68 mc_region_restore_sparse(&mc_model_checker->process(), region);
71 case simgrid::mc::StorageType::Privatized:
72 for (auto& p : region->privatized_data())
76 default: // includes StorageType::NoData
77 xbt_die("Storage type not supported");
83 RegionSnapshot privatized_region(RegionType region_type, void* start_addr, void* permanent_addr, std::size_t size)
85 size_t process_count = MC_smpi_process_count();
87 // Read smpi_privatization_regions from MCed:
88 smpi_privatization_region_t remote_smpi_privatization_regions;
89 mc_model_checker->process().read_variable("smpi_privatization_regions", &remote_smpi_privatization_regions,
90 sizeof(remote_smpi_privatization_regions));
91 s_smpi_privatization_region_t privatization_regions[process_count];
92 mc_model_checker->process().read_bytes(&privatization_regions, sizeof(privatization_regions),
93 remote(remote_smpi_privatization_regions));
95 std::vector<simgrid::mc::RegionSnapshot> data;
96 data.reserve(process_count);
97 for (size_t i = 0; i < process_count; i++)
98 data.push_back(simgrid::mc::region(region_type, start_addr, privatization_regions[i].address, size));
100 simgrid::mc::RegionSnapshot region = simgrid::mc::RegionSnapshot(region_type, start_addr, permanent_addr, size);
101 region.privatized_data(std::move(data));
106 static void add_region(int index, simgrid::mc::Snapshot* snapshot, simgrid::mc::RegionType type,
107 simgrid::mc::ObjectInformation* object_info, void* start_addr, void* permanent_addr,
110 if (type == simgrid::mc::RegionType::Data)
111 xbt_assert(object_info, "Missing object info for object.");
112 else if (type == simgrid::mc::RegionType::Heap)
113 xbt_assert(not object_info, "Unexpected object info for heap region.");
115 simgrid::mc::RegionSnapshot region;
117 const bool privatization_aware = object_info && mc_model_checker->process().privatized(*object_info);
118 if (privatization_aware && MC_smpi_process_count())
119 region = simgrid::mc::privatized_region(type, start_addr, permanent_addr, size);
122 region = simgrid::mc::region(type, start_addr, permanent_addr, size);
124 region.object_info(object_info);
125 snapshot->snapshot_regions[index] =
126 std::unique_ptr<simgrid::mc::RegionSnapshot>(new simgrid::mc::RegionSnapshot(std::move(region)));
129 static void get_memory_regions(simgrid::mc::RemoteClient* process, simgrid::mc::Snapshot* snapshot)
131 const size_t n = process->object_infos.size();
132 snapshot->snapshot_regions.resize(n + 1);
134 for (auto const& object_info : process->object_infos) {
135 add_region(i, snapshot, simgrid::mc::RegionType::Data, object_info.get(), object_info->start_rw,
136 object_info->start_rw, object_info->end_rw - object_info->start_rw);
140 xbt_mheap_t heap = process->get_heap();
141 void* start_heap = heap->base;
142 void* end_heap = heap->breakval;
144 add_region(n, snapshot, simgrid::mc::RegionType::Heap, nullptr, start_heap, start_heap,
145 (char*)end_heap - (char*)start_heap);
146 snapshot->heap_bytes_used = mmalloc_get_bytes_used_remote(heap->heaplimit, process->get_malloc_info());
149 if (mc_model_checker->process().privatized() && MC_smpi_process_count())
150 // snapshot->privatization_index = smpi_loaded_page
151 mc_model_checker->process().read_variable("smpi_loaded_page", &snapshot->privatization_index,
152 sizeof(snapshot->privatization_index));
155 snapshot->privatization_index = simgrid::mc::ProcessIndexMissing;
158 /** @brief Fills the position of the segments (executable, read-only, read/write).
160 // TODO, use the ELF segment information for more robustness
161 void find_object_address(std::vector<simgrid::xbt::VmMap> const& maps, simgrid::mc::ObjectInformation* result)
163 std::string name = simgrid::xbt::Path(result->file_name).get_base_name();
165 for (size_t i = 0; i < maps.size(); ++i) {
166 simgrid::xbt::VmMap const& reg = maps[i];
167 if (maps[i].pathname.empty())
169 std::string map_basename = simgrid::xbt::Path(maps[i].pathname).get_base_name();
170 if (map_basename != name)
173 // This is the non-GNU_RELRO-part of the data segment:
174 if (reg.prot == PROT_RW) {
175 xbt_assert(not result->start_rw, "Multiple read-write segments for %s, not supported", maps[i].pathname.c_str());
176 result->start_rw = (char*)reg.start_addr;
177 result->end_rw = (char*)reg.end_addr;
179 // The next VMA might be end of the data segment:
180 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
181 maps[i + 1].start_addr == reg.end_addr)
182 result->end_rw = (char*)maps[i + 1].end_addr;
185 // This is the text segment:
186 else if (reg.prot == PROT_RX) {
187 xbt_assert(not result->start_exec, "Multiple executable segments for %s, not supported",
188 maps[i].pathname.c_str());
189 result->start_exec = (char*)reg.start_addr;
190 result->end_exec = (char*)reg.end_addr;
192 // The next VMA might be end of the data segment:
193 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
194 maps[i + 1].start_addr == reg.end_addr) {
195 result->start_rw = (char*)maps[i + 1].start_addr;
196 result->end_rw = (char*)maps[i + 1].end_addr;
200 // This is the GNU_RELRO-part of the data segment:
201 else if (reg.prot == PROT_READ) {
202 xbt_assert(not result->start_ro, "Multiple read only segments for %s, not supported", maps[i].pathname.c_str());
203 result->start_ro = (char*)reg.start_addr;
204 result->end_ro = (char*)reg.end_addr;
208 result->start = result->start_rw;
209 if ((const void*)result->start_ro < result->start)
210 result->start = result->start_ro;
211 if ((const void*)result->start_exec < result->start)
212 result->start = result->start_exec;
214 result->end = result->end_rw;
215 if (result->end_ro && (const void*)result->end_ro > result->end)
216 result->end = result->end_ro;
217 if (result->end_exec && (const void*)result->end_exec > result->end)
218 result->end = result->end_exec;
220 xbt_assert(result->start_exec || result->start_rw || result->start_ro);
223 /************************************* Take Snapshot ************************************/
224 /****************************************************************************************/
226 /** @brief Checks whether the variable is in scope for a given IP.
228 * A variable may be defined only from a given value of IP.
230 * @param var Variable description
231 * @param scope Scope description
232 * @param ip Instruction pointer
233 * @return true if the variable is valid
235 static bool valid_variable(simgrid::mc::Variable* var, simgrid::mc::Frame* scope, const void* ip)
237 // The variable is not yet valid:
238 if (scope->range.begin() + var->start_scope > (std::uint64_t)ip)
244 static void fill_local_variables_values(mc_stack_frame_t stack_frame, simgrid::mc::Frame* scope, int process_index,
245 std::vector<s_local_variable_t>& result)
247 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
249 if (not scope || not scope->range.contain(stack_frame->ip))
252 for (simgrid::mc::Variable& current_variable : scope->variables) {
254 if (not valid_variable(¤t_variable, scope, (void*)stack_frame->ip))
258 // FIXME, get rid of `region_type`
259 if ((long)stack_frame->ip > (long)process->libsimgrid_info->start_exec)
264 s_local_variable_t new_var;
265 new_var.subprogram = stack_frame->frame;
266 new_var.ip = stack_frame->ip;
267 new_var.name = current_variable.name;
268 new_var.type = current_variable.type;
269 new_var.region = region_type;
270 new_var.address = nullptr;
272 if (current_variable.address != nullptr)
273 new_var.address = current_variable.address;
274 else if (not current_variable.location_list.empty()) {
275 simgrid::dwarf::Location location = simgrid::dwarf::resolve(
276 current_variable.location_list, current_variable.object_info, &(stack_frame->unw_cursor),
277 (void*)stack_frame->frame_base, &mc_model_checker->process(), process_index);
279 if (not location.in_memory())
280 xbt_die("Cannot handle non-address variable");
281 new_var.address = location.address();
284 xbt_die("No address");
286 result.push_back(std::move(new_var));
289 // Recursive processing of nested scopes:
290 for (simgrid::mc::Frame& nested_scope : scope->scopes)
291 fill_local_variables_values(stack_frame, &nested_scope, process_index, result);
294 static std::vector<s_local_variable_t> get_local_variables_values(std::vector<s_mc_stack_frame_t>& stack_frames,
297 std::vector<s_local_variable_t> variables;
298 for (s_mc_stack_frame_t& stack_frame : stack_frames)
299 fill_local_variables_values(&stack_frame, stack_frame.frame, process_index, variables);
303 static std::vector<s_mc_stack_frame_t> unwind_stack_frames(simgrid::mc::UnwindContext* stack_context)
305 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
306 std::vector<s_mc_stack_frame_t> result;
308 unw_cursor_t c = stack_context->cursor();
310 // TODO, check condition check (unw_init_local==0 means end of frame)
314 s_mc_stack_frame_t stack_frame;
316 stack_frame.unw_cursor = c;
321 unw_get_reg(&c, UNW_REG_IP, &ip);
322 unw_get_reg(&c, UNW_REG_SP, &sp);
327 // TODO, use real addresses in frame_t instead of fixing it here
329 simgrid::mc::Frame* frame = process->find_function(remote(ip));
330 stack_frame.frame = frame;
333 stack_frame.frame_name = frame->name;
334 stack_frame.frame_base = (unw_word_t)frame->frame_base(c);
336 stack_frame.frame_base = 0;
337 stack_frame.frame_name = std::string();
340 result.push_back(std::move(stack_frame));
342 /* Stop before context switch with maestro */
343 if (frame != nullptr && frame->name == "smx_ctx_wrapper")
346 int ret = unw_step(&c);
348 xbt_die("Unexpected end of stack.");
350 xbt_die("Error while unwinding stack");
353 if (result.empty()) {
354 XBT_INFO("unw_init_local failed");
361 static std::vector<s_mc_snapshot_stack_t> take_snapshot_stacks(simgrid::mc::Snapshot* snapshot)
363 std::vector<s_mc_snapshot_stack_t> res;
365 for (auto const& stack : mc_model_checker->process().stack_areas()) {
366 s_mc_snapshot_stack_t st;
368 // Read the context from remote process:
369 unw_context_t context;
370 mc_model_checker->process().read_bytes(&context, sizeof(context), remote(stack.context));
372 st.context.initialize(&mc_model_checker->process(), &context);
374 st.stack_frames = unwind_stack_frames(&st.context);
375 st.local_variables = get_local_variables_values(st.stack_frames, stack.process_index);
376 st.process_index = stack.process_index;
378 unw_word_t sp = st.stack_frames[0].sp;
380 res.push_back(std::move(st));
382 size_t stack_size = (char*)stack.address + stack.size - (char*)sp;
383 snapshot->stack_sizes.push_back(stack_size);
389 static void snapshot_handle_ignore(simgrid::mc::Snapshot* snapshot)
391 xbt_assert(snapshot->process());
394 for (auto const& region : mc_model_checker->process().ignored_regions()) {
395 s_mc_snapshot_ignored_data_t ignored_data;
396 ignored_data.start = (void*)region.addr;
397 ignored_data.data.resize(region.size);
398 // TODO, we should do this once per privatization segment:
399 snapshot->process()->read_bytes(ignored_data.data.data(), region.size, remote(region.addr),
400 simgrid::mc::ProcessIndexDisabled);
401 snapshot->ignored_data.push_back(std::move(ignored_data));
405 for (auto const& region : mc_model_checker->process().ignored_regions())
406 snapshot->process()->clear_bytes(remote(region.addr), region.size);
409 static void snapshot_ignore_restore(simgrid::mc::Snapshot* snapshot)
411 for (auto const& ignored_data : snapshot->ignored_data)
412 snapshot->process()->write_bytes(ignored_data.data.data(), ignored_data.data.size(), remote(ignored_data.start));
415 std::shared_ptr<simgrid::mc::Snapshot> take_snapshot(int num_state)
417 XBT_DEBUG("Taking snapshot %i", num_state);
419 simgrid::mc::RemoteClient* mc_process = &mc_model_checker->process();
421 std::shared_ptr<simgrid::mc::Snapshot> snapshot = std::make_shared<simgrid::mc::Snapshot>(mc_process, num_state);
423 for (auto const& p : mc_model_checker->process().actors())
424 snapshot->enabled_processes.insert(p.copy.getBuffer()->get_pid());
426 snapshot_handle_ignore(snapshot.get());
428 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
429 get_memory_regions(mc_process, snapshot.get());
431 snapshot->to_ignore = mc_model_checker->process().ignored_heap();
433 if (_sg_mc_max_visited_states > 0 || not _sg_mc_property_file.get().empty()) {
434 snapshot->stacks = take_snapshot_stacks(snapshot.get());
436 snapshot->hash = simgrid::mc::hash(*snapshot);
442 snapshot_ignore_restore(snapshot.get());
446 static inline void restore_snapshot_regions(simgrid::mc::Snapshot* snapshot)
448 for (std::unique_ptr<s_mc_mem_region_t> const& region : snapshot->snapshot_regions) {
449 // For privatized, variables we decided it was not necessary to take the snapshot:
451 restore(region.get());
455 if (snapshot->privatization_index >= 0) {
456 // Fix the privatization mmap:
457 s_mc_message_restore_t message{MC_MESSAGE_RESTORE, snapshot->privatization_index};
458 mc_model_checker->process().getChannel().send(message);
463 void restore_snapshot(std::shared_ptr<simgrid::mc::Snapshot> snapshot)
465 XBT_DEBUG("Restore snapshot %i", snapshot->num_state);
466 restore_snapshot_regions(snapshot.get());
467 snapshot_ignore_restore(snapshot.get());
468 mc_model_checker->process().clear_cache();
472 } // namespace simgrid