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/remote/mc_protocol.h"
34 #include "src/mc/sosp/mc_snapshot.hpp"
36 #include "src/mc/sosp/RegionSnapshot.hpp"
38 using simgrid::mc::remote;
40 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc, "Logging specific to mc_checkpoint");
42 #define PROT_RWX (PROT_READ | PROT_WRITE | PROT_EXEC)
43 #define PROT_RW (PROT_READ | PROT_WRITE)
44 #define PROT_RX (PROT_READ | PROT_EXEC)
49 /************************************ Free functions **************************************/
50 /*****************************************************************************************/
52 /** @brief Restore a region from a snapshot
54 * @param region Target region
56 static void restore(RegionSnapshot* region)
58 switch (region->storage_type()) {
59 case simgrid::mc::StorageType::Flat:
60 mc_model_checker->process().write_bytes(region->flat_data().get(), region->size(), region->permanent_address());
63 case simgrid::mc::StorageType::Chunked:
64 xbt_assert(((region->permanent_address().address()) & (xbt_pagesize - 1)) == 0, "Not at the beginning of a page");
65 xbt_assert(simgrid::mc::mmu::chunk_count(region->size()) == region->page_data().page_count());
67 for (size_t i = 0; i != region->page_data().page_count(); ++i) {
69 (void*)simgrid::mc::mmu::join(i, (std::uintptr_t)(void*)region->permanent_address().address());
70 const void* source_page = region->page_data().page(i);
71 mc_model_checker->process().write_bytes(source_page, xbt_pagesize, remote(target_page));
76 case simgrid::mc::StorageType::Privatized:
77 for (auto& p : region->privatized_data())
81 default: // includes StorageType::NoData
82 xbt_die("Storage type not supported");
88 RegionSnapshot privatized_region(RegionType region_type, void* start_addr, void* permanent_addr, std::size_t size)
90 size_t process_count = MC_smpi_process_count();
92 // Read smpi_privatization_regions from MCed:
93 smpi_privatization_region_t remote_smpi_privatization_regions;
94 mc_model_checker->process().read_variable("smpi_privatization_regions", &remote_smpi_privatization_regions,
95 sizeof(remote_smpi_privatization_regions));
96 s_smpi_privatization_region_t privatization_regions[process_count];
97 mc_model_checker->process().read_bytes(&privatization_regions, sizeof(privatization_regions),
98 remote(remote_smpi_privatization_regions));
100 std::vector<RegionSnapshot> data;
101 data.reserve(process_count);
102 for (size_t i = 0; i < process_count; i++)
103 data.push_back(region(region_type, start_addr, privatization_regions[i].address, size));
105 RegionSnapshot region = RegionSnapshot(region_type, start_addr, permanent_addr, size);
106 region.privatized_data(std::move(data));
111 static void get_memory_regions(simgrid::mc::RemoteClient* process, simgrid::mc::Snapshot* snapshot)
113 snapshot->snapshot_regions_.clear();
115 for (auto const& object_info : process->object_infos)
116 snapshot->add_region(simgrid::mc::RegionType::Data, object_info.get(), object_info->start_rw, object_info->start_rw,
117 object_info->end_rw - object_info->start_rw);
119 xbt_mheap_t heap = process->get_heap();
120 void* start_heap = heap->base;
121 void* end_heap = heap->breakval;
123 snapshot->add_region(simgrid::mc::RegionType::Heap, nullptr, start_heap, start_heap,
124 (char*)end_heap - (char*)start_heap);
125 snapshot->heap_bytes_used_ = mmalloc_get_bytes_used_remote(heap->heaplimit, process->get_malloc_info());
128 if (mc_model_checker->process().privatized() && MC_smpi_process_count())
129 // snapshot->privatization_index = smpi_loaded_page
130 mc_model_checker->process().read_variable("smpi_loaded_page", &snapshot->privatization_index_,
131 sizeof(snapshot->privatization_index_));
134 snapshot->privatization_index_ = simgrid::mc::ProcessIndexMissing;
137 /** @brief Fills the position of the segments (executable, read-only, read/write).
139 // TODO, use the ELF segment information for more robustness
140 void find_object_address(std::vector<simgrid::xbt::VmMap> const& maps, simgrid::mc::ObjectInformation* result)
142 std::string name = simgrid::xbt::Path(result->file_name).get_base_name();
144 for (size_t i = 0; i < maps.size(); ++i) {
145 simgrid::xbt::VmMap const& reg = maps[i];
146 if (maps[i].pathname.empty())
148 std::string map_basename = simgrid::xbt::Path(maps[i].pathname).get_base_name();
149 if (map_basename != name)
152 // This is the non-GNU_RELRO-part of the data segment:
153 if (reg.prot == PROT_RW) {
154 xbt_assert(not result->start_rw, "Multiple read-write segments for %s, not supported", maps[i].pathname.c_str());
155 result->start_rw = (char*)reg.start_addr;
156 result->end_rw = (char*)reg.end_addr;
158 // The next VMA might be end of the data segment:
159 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
160 maps[i + 1].start_addr == reg.end_addr)
161 result->end_rw = (char*)maps[i + 1].end_addr;
164 // This is the text segment:
165 else if (reg.prot == PROT_RX) {
166 xbt_assert(not result->start_exec, "Multiple executable segments for %s, not supported",
167 maps[i].pathname.c_str());
168 result->start_exec = (char*)reg.start_addr;
169 result->end_exec = (char*)reg.end_addr;
171 // The next VMA might be end of the data segment:
172 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
173 maps[i + 1].start_addr == reg.end_addr) {
174 result->start_rw = (char*)maps[i + 1].start_addr;
175 result->end_rw = (char*)maps[i + 1].end_addr;
179 // This is the GNU_RELRO-part of the data segment:
180 else if (reg.prot == PROT_READ) {
181 xbt_assert(not result->start_ro, "Multiple read only segments for %s, not supported", maps[i].pathname.c_str());
182 result->start_ro = (char*)reg.start_addr;
183 result->end_ro = (char*)reg.end_addr;
187 result->start = result->start_rw;
188 if ((const void*)result->start_ro < result->start)
189 result->start = result->start_ro;
190 if ((const void*)result->start_exec < result->start)
191 result->start = result->start_exec;
193 result->end = result->end_rw;
194 if (result->end_ro && (const void*)result->end_ro > result->end)
195 result->end = result->end_ro;
196 if (result->end_exec && (const void*)result->end_exec > result->end)
197 result->end = result->end_exec;
199 xbt_assert(result->start_exec || result->start_rw || result->start_ro);
202 /************************************* Take Snapshot ************************************/
203 /****************************************************************************************/
205 /** @brief Checks whether the variable is in scope for a given IP.
207 * A variable may be defined only from a given value of IP.
209 * @param var Variable description
210 * @param scope Scope description
211 * @param ip Instruction pointer
212 * @return true if the variable is valid
214 static bool valid_variable(simgrid::mc::Variable* var, simgrid::mc::Frame* scope, const void* ip)
216 // The variable is not yet valid:
217 if (scope->range.begin() + var->start_scope > (std::uint64_t)ip)
223 static void fill_local_variables_values(mc_stack_frame_t stack_frame, simgrid::mc::Frame* scope, int process_index,
224 std::vector<s_local_variable_t>& result)
226 if (not scope || not scope->range.contain(stack_frame->ip))
229 for (simgrid::mc::Variable& current_variable : scope->variables) {
231 if (not valid_variable(¤t_variable, scope, (void*)stack_frame->ip))
234 s_local_variable_t new_var;
235 new_var.subprogram = stack_frame->frame;
236 new_var.ip = stack_frame->ip;
237 new_var.name = current_variable.name;
238 new_var.type = current_variable.type;
239 new_var.address = nullptr;
241 if (current_variable.address != nullptr)
242 new_var.address = current_variable.address;
243 else if (not current_variable.location_list.empty()) {
244 simgrid::dwarf::Location location = simgrid::dwarf::resolve(
245 current_variable.location_list, current_variable.object_info, &(stack_frame->unw_cursor),
246 (void*)stack_frame->frame_base, &mc_model_checker->process(), process_index);
248 if (not location.in_memory())
249 xbt_die("Cannot handle non-address variable");
250 new_var.address = location.address();
253 xbt_die("No address");
255 result.push_back(std::move(new_var));
258 // Recursive processing of nested scopes:
259 for (simgrid::mc::Frame& nested_scope : scope->scopes)
260 fill_local_variables_values(stack_frame, &nested_scope, process_index, result);
263 static std::vector<s_local_variable_t> get_local_variables_values(std::vector<s_mc_stack_frame_t>& stack_frames,
266 std::vector<s_local_variable_t> variables;
267 for (s_mc_stack_frame_t& stack_frame : stack_frames)
268 fill_local_variables_values(&stack_frame, stack_frame.frame, process_index, variables);
272 static std::vector<s_mc_stack_frame_t> unwind_stack_frames(simgrid::mc::UnwindContext* stack_context)
274 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
275 std::vector<s_mc_stack_frame_t> result;
277 unw_cursor_t c = stack_context->cursor();
279 // TODO, check condition check (unw_init_local==0 means end of frame)
283 s_mc_stack_frame_t stack_frame;
285 stack_frame.unw_cursor = c;
290 unw_get_reg(&c, UNW_REG_IP, &ip);
291 unw_get_reg(&c, UNW_REG_SP, &sp);
296 // TODO, use real addresses in frame_t instead of fixing it here
298 simgrid::mc::Frame* frame = process->find_function(remote(ip));
299 stack_frame.frame = frame;
302 stack_frame.frame_name = frame->name;
303 stack_frame.frame_base = (unw_word_t)frame->frame_base(c);
305 stack_frame.frame_base = 0;
306 stack_frame.frame_name = std::string();
309 result.push_back(std::move(stack_frame));
311 /* Stop before context switch with maestro */
312 if (frame != nullptr && frame->name == "smx_ctx_wrapper")
315 int ret = unw_step(&c);
317 xbt_die("Unexpected end of stack.");
319 xbt_die("Error while unwinding stack");
322 xbt_assert(not result.empty(), "unw_init_local failed");
327 static void take_snapshot_stacks(simgrid::mc::Snapshot* snapshot)
329 for (auto const& stack : mc_model_checker->process().stack_areas()) {
330 s_mc_snapshot_stack_t st;
332 // Read the context from remote process:
333 unw_context_t context;
334 mc_model_checker->process().read_bytes(&context, sizeof(context), remote(stack.context));
336 st.context.initialize(&mc_model_checker->process(), &context);
338 st.stack_frames = unwind_stack_frames(&st.context);
339 st.local_variables = get_local_variables_values(st.stack_frames, stack.process_index);
340 st.process_index = stack.process_index;
342 unw_word_t sp = st.stack_frames[0].sp;
344 snapshot->stacks_.push_back(std::move(st));
346 size_t stack_size = (char*)stack.address + stack.size - (char*)sp;
347 snapshot->stack_sizes_.push_back(stack_size);
351 static void snapshot_handle_ignore(simgrid::mc::Snapshot* snapshot)
353 xbt_assert(snapshot->process());
356 for (auto const& region : mc_model_checker->process().ignored_regions()) {
357 s_mc_snapshot_ignored_data_t ignored_data;
358 ignored_data.start = (void*)region.addr;
359 ignored_data.data.resize(region.size);
360 // TODO, we should do this once per privatization segment:
361 snapshot->process()->read_bytes(ignored_data.data.data(), region.size, remote(region.addr),
362 simgrid::mc::ProcessIndexDisabled);
363 snapshot->ignored_data_.push_back(std::move(ignored_data));
367 for (auto const& region : mc_model_checker->process().ignored_regions())
368 snapshot->process()->clear_bytes(remote(region.addr), region.size);
371 static void snapshot_ignore_restore(simgrid::mc::Snapshot* snapshot)
373 for (auto const& ignored_data : snapshot->ignored_data_)
374 snapshot->process()->write_bytes(ignored_data.data.data(), ignored_data.data.size(), remote(ignored_data.start));
377 std::shared_ptr<simgrid::mc::Snapshot> take_snapshot(int num_state)
379 XBT_DEBUG("Taking snapshot %i", num_state);
381 simgrid::mc::RemoteClient* mc_process = &mc_model_checker->process();
383 std::shared_ptr<simgrid::mc::Snapshot> snapshot = std::make_shared<simgrid::mc::Snapshot>(mc_process, num_state);
385 for (auto const& p : mc_model_checker->process().actors())
386 snapshot->enabled_processes_.insert(p.copy.getBuffer()->get_pid());
388 snapshot_handle_ignore(snapshot.get());
390 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
391 get_memory_regions(mc_process, snapshot.get());
393 snapshot->to_ignore_ = mc_model_checker->process().ignored_heap();
395 if (_sg_mc_max_visited_states > 0 || not _sg_mc_property_file.get().empty()) {
396 take_snapshot_stacks(snapshot.get());
398 snapshot->hash_ = simgrid::mc::hash(*snapshot);
401 snapshot_ignore_restore(snapshot.get());
405 static inline void restore_snapshot_regions(simgrid::mc::Snapshot* snapshot)
407 for (std::unique_ptr<simgrid::mc::RegionSnapshot> const& region : snapshot->snapshot_regions_) {
408 // For privatized, variables we decided it was not necessary to take the snapshot:
410 restore(region.get());
414 if (snapshot->privatization_index_ >= 0) {
415 // Fix the privatization mmap:
416 s_mc_message_restore_t message{MC_MESSAGE_RESTORE, snapshot->privatization_index_};
417 mc_model_checker->process().getChannel().send(message);
422 void restore_snapshot(std::shared_ptr<simgrid::mc::Snapshot> snapshot)
424 XBT_DEBUG("Restore snapshot %i", snapshot->num_state_);
425 restore_snapshot_regions(snapshot.get());
426 snapshot_ignore_restore(snapshot.get());
427 mc_model_checker->process().clear_cache();
431 } // namespace simgrid