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(RegionSnapshot* 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 xbt_assert(((region->permanent_address().address()) & (xbt_pagesize - 1)) == 0, "Not at the beginning of a page");
69 xbt_assert(simgrid::mc::mmu::chunk_count(region->size()) == region->page_data().page_count());
71 for (size_t i = 0; i != region->page_data().page_count(); ++i) {
73 (void*)simgrid::mc::mmu::join(i, (std::uintptr_t)(void*)region->permanent_address().address());
74 const void* source_page = region->page_data().page(i);
75 mc_model_checker->process().write_bytes(source_page, xbt_pagesize, remote(target_page));
80 case simgrid::mc::StorageType::Privatized:
81 for (auto& p : region->privatized_data())
85 default: // includes StorageType::NoData
86 xbt_die("Storage type not supported");
92 RegionSnapshot privatized_region(RegionType region_type, void* start_addr, void* permanent_addr, std::size_t size)
94 size_t process_count = MC_smpi_process_count();
96 // Read smpi_privatization_regions from MCed:
97 smpi_privatization_region_t remote_smpi_privatization_regions;
98 mc_model_checker->process().read_variable("smpi_privatization_regions", &remote_smpi_privatization_regions,
99 sizeof(remote_smpi_privatization_regions));
100 s_smpi_privatization_region_t privatization_regions[process_count];
101 mc_model_checker->process().read_bytes(&privatization_regions, sizeof(privatization_regions),
102 remote(remote_smpi_privatization_regions));
104 std::vector<RegionSnapshot> data;
105 data.reserve(process_count);
106 for (size_t i = 0; i < process_count; i++)
107 data.push_back(region(region_type, start_addr, privatization_regions[i].address, size));
109 RegionSnapshot region = RegionSnapshot(region_type, start_addr, permanent_addr, size);
110 region.privatized_data(std::move(data));
115 static void get_memory_regions(simgrid::mc::RemoteClient* process, simgrid::mc::Snapshot* snapshot)
117 snapshot->snapshot_regions_.clear();
119 for (auto const& object_info : process->object_infos)
120 snapshot->add_region(simgrid::mc::RegionType::Data, object_info.get(), object_info->start_rw, object_info->start_rw,
121 object_info->end_rw - object_info->start_rw);
123 xbt_mheap_t heap = process->get_heap();
124 void* start_heap = heap->base;
125 void* end_heap = heap->breakval;
127 snapshot->add_region(simgrid::mc::RegionType::Heap, nullptr, start_heap, start_heap,
128 (char*)end_heap - (char*)start_heap);
129 snapshot->heap_bytes_used_ = mmalloc_get_bytes_used_remote(heap->heaplimit, process->get_malloc_info());
132 if (mc_model_checker->process().privatized() && MC_smpi_process_count())
133 // snapshot->privatization_index = smpi_loaded_page
134 mc_model_checker->process().read_variable("smpi_loaded_page", &snapshot->privatization_index_,
135 sizeof(snapshot->privatization_index_));
138 snapshot->privatization_index_ = simgrid::mc::ProcessIndexMissing;
141 /** @brief Fills the position of the segments (executable, read-only, read/write).
143 // TODO, use the ELF segment information for more robustness
144 void find_object_address(std::vector<simgrid::xbt::VmMap> const& maps, simgrid::mc::ObjectInformation* result)
146 std::string name = simgrid::xbt::Path(result->file_name).get_base_name();
148 for (size_t i = 0; i < maps.size(); ++i) {
149 simgrid::xbt::VmMap const& reg = maps[i];
150 if (maps[i].pathname.empty())
152 std::string map_basename = simgrid::xbt::Path(maps[i].pathname).get_base_name();
153 if (map_basename != name)
156 // This is the non-GNU_RELRO-part of the data segment:
157 if (reg.prot == PROT_RW) {
158 xbt_assert(not result->start_rw, "Multiple read-write segments for %s, not supported", maps[i].pathname.c_str());
159 result->start_rw = (char*)reg.start_addr;
160 result->end_rw = (char*)reg.end_addr;
162 // The next VMA might be end of the data segment:
163 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
164 maps[i + 1].start_addr == reg.end_addr)
165 result->end_rw = (char*)maps[i + 1].end_addr;
168 // This is the text segment:
169 else if (reg.prot == PROT_RX) {
170 xbt_assert(not result->start_exec, "Multiple executable segments for %s, not supported",
171 maps[i].pathname.c_str());
172 result->start_exec = (char*)reg.start_addr;
173 result->end_exec = (char*)reg.end_addr;
175 // The next VMA might be end of the data segment:
176 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
177 maps[i + 1].start_addr == reg.end_addr) {
178 result->start_rw = (char*)maps[i + 1].start_addr;
179 result->end_rw = (char*)maps[i + 1].end_addr;
183 // This is the GNU_RELRO-part of the data segment:
184 else if (reg.prot == PROT_READ) {
185 xbt_assert(not result->start_ro, "Multiple read only segments for %s, not supported", maps[i].pathname.c_str());
186 result->start_ro = (char*)reg.start_addr;
187 result->end_ro = (char*)reg.end_addr;
191 result->start = result->start_rw;
192 if ((const void*)result->start_ro < result->start)
193 result->start = result->start_ro;
194 if ((const void*)result->start_exec < result->start)
195 result->start = result->start_exec;
197 result->end = result->end_rw;
198 if (result->end_ro && (const void*)result->end_ro > result->end)
199 result->end = result->end_ro;
200 if (result->end_exec && (const void*)result->end_exec > result->end)
201 result->end = result->end_exec;
203 xbt_assert(result->start_exec || result->start_rw || result->start_ro);
206 /************************************* Take Snapshot ************************************/
207 /****************************************************************************************/
209 /** @brief Checks whether the variable is in scope for a given IP.
211 * A variable may be defined only from a given value of IP.
213 * @param var Variable description
214 * @param scope Scope description
215 * @param ip Instruction pointer
216 * @return true if the variable is valid
218 static bool valid_variable(simgrid::mc::Variable* var, simgrid::mc::Frame* scope, const void* ip)
220 // The variable is not yet valid:
221 if (scope->range.begin() + var->start_scope > (std::uint64_t)ip)
227 static void fill_local_variables_values(mc_stack_frame_t stack_frame, simgrid::mc::Frame* scope, int process_index,
228 std::vector<s_local_variable_t>& result)
230 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
232 if (not scope || not scope->range.contain(stack_frame->ip))
235 for (simgrid::mc::Variable& current_variable : scope->variables) {
237 if (not valid_variable(¤t_variable, scope, (void*)stack_frame->ip))
241 // FIXME, get rid of `region_type`
242 if ((long)stack_frame->ip > (long)process->libsimgrid_info->start_exec)
247 s_local_variable_t new_var;
248 new_var.subprogram = stack_frame->frame;
249 new_var.ip = stack_frame->ip;
250 new_var.name = current_variable.name;
251 new_var.type = current_variable.type;
252 new_var.region = region_type;
253 new_var.address = nullptr;
255 if (current_variable.address != nullptr)
256 new_var.address = current_variable.address;
257 else if (not current_variable.location_list.empty()) {
258 simgrid::dwarf::Location location = simgrid::dwarf::resolve(
259 current_variable.location_list, current_variable.object_info, &(stack_frame->unw_cursor),
260 (void*)stack_frame->frame_base, &mc_model_checker->process(), process_index);
262 if (not location.in_memory())
263 xbt_die("Cannot handle non-address variable");
264 new_var.address = location.address();
267 xbt_die("No address");
269 result.push_back(std::move(new_var));
272 // Recursive processing of nested scopes:
273 for (simgrid::mc::Frame& nested_scope : scope->scopes)
274 fill_local_variables_values(stack_frame, &nested_scope, process_index, result);
277 static std::vector<s_local_variable_t> get_local_variables_values(std::vector<s_mc_stack_frame_t>& stack_frames,
280 std::vector<s_local_variable_t> variables;
281 for (s_mc_stack_frame_t& stack_frame : stack_frames)
282 fill_local_variables_values(&stack_frame, stack_frame.frame, process_index, variables);
286 static std::vector<s_mc_stack_frame_t> unwind_stack_frames(simgrid::mc::UnwindContext* stack_context)
288 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
289 std::vector<s_mc_stack_frame_t> result;
291 unw_cursor_t c = stack_context->cursor();
293 // TODO, check condition check (unw_init_local==0 means end of frame)
297 s_mc_stack_frame_t stack_frame;
299 stack_frame.unw_cursor = c;
304 unw_get_reg(&c, UNW_REG_IP, &ip);
305 unw_get_reg(&c, UNW_REG_SP, &sp);
310 // TODO, use real addresses in frame_t instead of fixing it here
312 simgrid::mc::Frame* frame = process->find_function(remote(ip));
313 stack_frame.frame = frame;
316 stack_frame.frame_name = frame->name;
317 stack_frame.frame_base = (unw_word_t)frame->frame_base(c);
319 stack_frame.frame_base = 0;
320 stack_frame.frame_name = std::string();
323 result.push_back(std::move(stack_frame));
325 /* Stop before context switch with maestro */
326 if (frame != nullptr && frame->name == "smx_ctx_wrapper")
329 int ret = unw_step(&c);
331 xbt_die("Unexpected end of stack.");
333 xbt_die("Error while unwinding stack");
336 xbt_assert(not result.empty(), "unw_init_local failed");
341 static void take_snapshot_stacks(simgrid::mc::Snapshot* snapshot)
343 for (auto const& stack : mc_model_checker->process().stack_areas()) {
344 s_mc_snapshot_stack_t st;
346 // Read the context from remote process:
347 unw_context_t context;
348 mc_model_checker->process().read_bytes(&context, sizeof(context), remote(stack.context));
350 st.context.initialize(&mc_model_checker->process(), &context);
352 st.stack_frames = unwind_stack_frames(&st.context);
353 st.local_variables = get_local_variables_values(st.stack_frames, stack.process_index);
354 st.process_index = stack.process_index;
356 unw_word_t sp = st.stack_frames[0].sp;
358 snapshot->stacks_.push_back(std::move(st));
360 size_t stack_size = (char*)stack.address + stack.size - (char*)sp;
361 snapshot->stack_sizes_.push_back(stack_size);
365 static void snapshot_handle_ignore(simgrid::mc::Snapshot* snapshot)
367 xbt_assert(snapshot->process());
370 for (auto const& region : mc_model_checker->process().ignored_regions()) {
371 s_mc_snapshot_ignored_data_t ignored_data;
372 ignored_data.start = (void*)region.addr;
373 ignored_data.data.resize(region.size);
374 // TODO, we should do this once per privatization segment:
375 snapshot->process()->read_bytes(ignored_data.data.data(), region.size, remote(region.addr),
376 simgrid::mc::ProcessIndexDisabled);
377 snapshot->ignored_data_.push_back(std::move(ignored_data));
381 for (auto const& region : mc_model_checker->process().ignored_regions())
382 snapshot->process()->clear_bytes(remote(region.addr), region.size);
385 static void snapshot_ignore_restore(simgrid::mc::Snapshot* snapshot)
387 for (auto const& ignored_data : snapshot->ignored_data_)
388 snapshot->process()->write_bytes(ignored_data.data.data(), ignored_data.data.size(), remote(ignored_data.start));
391 std::shared_ptr<simgrid::mc::Snapshot> take_snapshot(int num_state)
393 XBT_DEBUG("Taking snapshot %i", num_state);
395 simgrid::mc::RemoteClient* mc_process = &mc_model_checker->process();
397 std::shared_ptr<simgrid::mc::Snapshot> snapshot = std::make_shared<simgrid::mc::Snapshot>(mc_process, num_state);
399 for (auto const& p : mc_model_checker->process().actors())
400 snapshot->enabled_processes_.insert(p.copy.getBuffer()->get_pid());
402 snapshot_handle_ignore(snapshot.get());
404 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
405 get_memory_regions(mc_process, snapshot.get());
407 snapshot->to_ignore_ = mc_model_checker->process().ignored_heap();
409 if (_sg_mc_max_visited_states > 0 || not _sg_mc_property_file.get().empty()) {
410 take_snapshot_stacks(snapshot.get());
412 snapshot->hash_ = simgrid::mc::hash(*snapshot);
415 snapshot_ignore_restore(snapshot.get());
419 static inline void restore_snapshot_regions(simgrid::mc::Snapshot* snapshot)
421 for (std::unique_ptr<simgrid::mc::RegionSnapshot> const& region : snapshot->snapshot_regions_) {
422 // For privatized, variables we decided it was not necessary to take the snapshot:
424 restore(region.get());
428 if (snapshot->privatization_index_ >= 0) {
429 // Fix the privatization mmap:
430 s_mc_message_restore_t message{MC_MESSAGE_RESTORE, snapshot->privatization_index_};
431 mc_model_checker->process().getChannel().send(message);
436 void restore_snapshot(std::shared_ptr<simgrid::mc::Snapshot> snapshot)
438 XBT_DEBUG("Restore snapshot %i", snapshot->num_state_);
439 restore_snapshot_regions(snapshot.get());
440 snapshot_ignore_restore(snapshot.get());
441 mc_model_checker->process().clear_cache();
445 } // namespace simgrid