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 add_region(simgrid::mc::Snapshot* snapshot, simgrid::mc::RegionType type,
116 simgrid::mc::ObjectInformation* object_info, void* start_addr, void* permanent_addr,
119 if (type == simgrid::mc::RegionType::Data)
120 xbt_assert(object_info, "Missing object info for object.");
121 else if (type == simgrid::mc::RegionType::Heap)
122 xbt_assert(not object_info, "Unexpected object info for heap region.");
124 simgrid::mc::RegionSnapshot region;
126 const bool privatization_aware = object_info && mc_model_checker->process().privatized(*object_info);
127 if (privatization_aware && MC_smpi_process_count())
128 region = simgrid::mc::privatized_region(type, start_addr, permanent_addr, size);
131 region = simgrid::mc::region(type, start_addr, permanent_addr, size);
133 region.object_info(object_info);
134 snapshot->snapshot_regions.push_back(
135 std::unique_ptr<simgrid::mc::RegionSnapshot>(new simgrid::mc::RegionSnapshot(std::move(region))));
138 static void get_memory_regions(simgrid::mc::RemoteClient* process, simgrid::mc::Snapshot* snapshot)
140 snapshot->snapshot_regions.clear();
142 for (auto const& object_info : process->object_infos)
143 add_region(snapshot, simgrid::mc::RegionType::Data, object_info.get(), object_info->start_rw, object_info->start_rw,
144 object_info->end_rw - object_info->start_rw);
146 xbt_mheap_t heap = process->get_heap();
147 void* start_heap = heap->base;
148 void* end_heap = heap->breakval;
150 add_region(snapshot, simgrid::mc::RegionType::Heap, nullptr, start_heap, start_heap,
151 (char*)end_heap - (char*)start_heap);
152 snapshot->heap_bytes_used = mmalloc_get_bytes_used_remote(heap->heaplimit, process->get_malloc_info());
155 if (mc_model_checker->process().privatized() && MC_smpi_process_count())
156 // snapshot->privatization_index = smpi_loaded_page
157 mc_model_checker->process().read_variable("smpi_loaded_page", &snapshot->privatization_index,
158 sizeof(snapshot->privatization_index));
161 snapshot->privatization_index = simgrid::mc::ProcessIndexMissing;
164 /** @brief Fills the position of the segments (executable, read-only, read/write).
166 // TODO, use the ELF segment information for more robustness
167 void find_object_address(std::vector<simgrid::xbt::VmMap> const& maps, simgrid::mc::ObjectInformation* result)
169 std::string name = simgrid::xbt::Path(result->file_name).get_base_name();
171 for (size_t i = 0; i < maps.size(); ++i) {
172 simgrid::xbt::VmMap const& reg = maps[i];
173 if (maps[i].pathname.empty())
175 std::string map_basename = simgrid::xbt::Path(maps[i].pathname).get_base_name();
176 if (map_basename != name)
179 // This is the non-GNU_RELRO-part of the data segment:
180 if (reg.prot == PROT_RW) {
181 xbt_assert(not result->start_rw, "Multiple read-write segments for %s, not supported", maps[i].pathname.c_str());
182 result->start_rw = (char*)reg.start_addr;
183 result->end_rw = (char*)reg.end_addr;
185 // The next VMA might be end of the data segment:
186 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
187 maps[i + 1].start_addr == reg.end_addr)
188 result->end_rw = (char*)maps[i + 1].end_addr;
191 // This is the text segment:
192 else if (reg.prot == PROT_RX) {
193 xbt_assert(not result->start_exec, "Multiple executable segments for %s, not supported",
194 maps[i].pathname.c_str());
195 result->start_exec = (char*)reg.start_addr;
196 result->end_exec = (char*)reg.end_addr;
198 // The next VMA might be end of the data segment:
199 if (i + 1 < maps.size() && maps[i + 1].pathname.empty() && maps[i + 1].prot == PROT_RW &&
200 maps[i + 1].start_addr == reg.end_addr) {
201 result->start_rw = (char*)maps[i + 1].start_addr;
202 result->end_rw = (char*)maps[i + 1].end_addr;
206 // This is the GNU_RELRO-part of the data segment:
207 else if (reg.prot == PROT_READ) {
208 xbt_assert(not result->start_ro, "Multiple read only segments for %s, not supported", maps[i].pathname.c_str());
209 result->start_ro = (char*)reg.start_addr;
210 result->end_ro = (char*)reg.end_addr;
214 result->start = result->start_rw;
215 if ((const void*)result->start_ro < result->start)
216 result->start = result->start_ro;
217 if ((const void*)result->start_exec < result->start)
218 result->start = result->start_exec;
220 result->end = result->end_rw;
221 if (result->end_ro && (const void*)result->end_ro > result->end)
222 result->end = result->end_ro;
223 if (result->end_exec && (const void*)result->end_exec > result->end)
224 result->end = result->end_exec;
226 xbt_assert(result->start_exec || result->start_rw || result->start_ro);
229 /************************************* Take Snapshot ************************************/
230 /****************************************************************************************/
232 /** @brief Checks whether the variable is in scope for a given IP.
234 * A variable may be defined only from a given value of IP.
236 * @param var Variable description
237 * @param scope Scope description
238 * @param ip Instruction pointer
239 * @return true if the variable is valid
241 static bool valid_variable(simgrid::mc::Variable* var, simgrid::mc::Frame* scope, const void* ip)
243 // The variable is not yet valid:
244 if (scope->range.begin() + var->start_scope > (std::uint64_t)ip)
250 static void fill_local_variables_values(mc_stack_frame_t stack_frame, simgrid::mc::Frame* scope, int process_index,
251 std::vector<s_local_variable_t>& result)
253 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
255 if (not scope || not scope->range.contain(stack_frame->ip))
258 for (simgrid::mc::Variable& current_variable : scope->variables) {
260 if (not valid_variable(¤t_variable, scope, (void*)stack_frame->ip))
264 // FIXME, get rid of `region_type`
265 if ((long)stack_frame->ip > (long)process->libsimgrid_info->start_exec)
270 s_local_variable_t new_var;
271 new_var.subprogram = stack_frame->frame;
272 new_var.ip = stack_frame->ip;
273 new_var.name = current_variable.name;
274 new_var.type = current_variable.type;
275 new_var.region = region_type;
276 new_var.address = nullptr;
278 if (current_variable.address != nullptr)
279 new_var.address = current_variable.address;
280 else if (not current_variable.location_list.empty()) {
281 simgrid::dwarf::Location location = simgrid::dwarf::resolve(
282 current_variable.location_list, current_variable.object_info, &(stack_frame->unw_cursor),
283 (void*)stack_frame->frame_base, &mc_model_checker->process(), process_index);
285 if (not location.in_memory())
286 xbt_die("Cannot handle non-address variable");
287 new_var.address = location.address();
290 xbt_die("No address");
292 result.push_back(std::move(new_var));
295 // Recursive processing of nested scopes:
296 for (simgrid::mc::Frame& nested_scope : scope->scopes)
297 fill_local_variables_values(stack_frame, &nested_scope, process_index, result);
300 static std::vector<s_local_variable_t> get_local_variables_values(std::vector<s_mc_stack_frame_t>& stack_frames,
303 std::vector<s_local_variable_t> variables;
304 for (s_mc_stack_frame_t& stack_frame : stack_frames)
305 fill_local_variables_values(&stack_frame, stack_frame.frame, process_index, variables);
309 static std::vector<s_mc_stack_frame_t> unwind_stack_frames(simgrid::mc::UnwindContext* stack_context)
311 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
312 std::vector<s_mc_stack_frame_t> result;
314 unw_cursor_t c = stack_context->cursor();
316 // TODO, check condition check (unw_init_local==0 means end of frame)
320 s_mc_stack_frame_t stack_frame;
322 stack_frame.unw_cursor = c;
327 unw_get_reg(&c, UNW_REG_IP, &ip);
328 unw_get_reg(&c, UNW_REG_SP, &sp);
333 // TODO, use real addresses in frame_t instead of fixing it here
335 simgrid::mc::Frame* frame = process->find_function(remote(ip));
336 stack_frame.frame = frame;
339 stack_frame.frame_name = frame->name;
340 stack_frame.frame_base = (unw_word_t)frame->frame_base(c);
342 stack_frame.frame_base = 0;
343 stack_frame.frame_name = std::string();
346 result.push_back(std::move(stack_frame));
348 /* Stop before context switch with maestro */
349 if (frame != nullptr && frame->name == "smx_ctx_wrapper")
352 int ret = unw_step(&c);
354 xbt_die("Unexpected end of stack.");
356 xbt_die("Error while unwinding stack");
359 if (result.empty()) {
360 XBT_INFO("unw_init_local failed");
367 static std::vector<s_mc_snapshot_stack_t> take_snapshot_stacks(simgrid::mc::Snapshot* snapshot)
369 std::vector<s_mc_snapshot_stack_t> res;
371 for (auto const& stack : mc_model_checker->process().stack_areas()) {
372 s_mc_snapshot_stack_t st;
374 // Read the context from remote process:
375 unw_context_t context;
376 mc_model_checker->process().read_bytes(&context, sizeof(context), remote(stack.context));
378 st.context.initialize(&mc_model_checker->process(), &context);
380 st.stack_frames = unwind_stack_frames(&st.context);
381 st.local_variables = get_local_variables_values(st.stack_frames, stack.process_index);
382 st.process_index = stack.process_index;
384 unw_word_t sp = st.stack_frames[0].sp;
386 res.push_back(std::move(st));
388 size_t stack_size = (char*)stack.address + stack.size - (char*)sp;
389 snapshot->stack_sizes.push_back(stack_size);
395 static void snapshot_handle_ignore(simgrid::mc::Snapshot* snapshot)
397 xbt_assert(snapshot->process());
400 for (auto const& region : mc_model_checker->process().ignored_regions()) {
401 s_mc_snapshot_ignored_data_t ignored_data;
402 ignored_data.start = (void*)region.addr;
403 ignored_data.data.resize(region.size);
404 // TODO, we should do this once per privatization segment:
405 snapshot->process()->read_bytes(ignored_data.data.data(), region.size, remote(region.addr),
406 simgrid::mc::ProcessIndexDisabled);
407 snapshot->ignored_data.push_back(std::move(ignored_data));
411 for (auto const& region : mc_model_checker->process().ignored_regions())
412 snapshot->process()->clear_bytes(remote(region.addr), region.size);
415 static void snapshot_ignore_restore(simgrid::mc::Snapshot* snapshot)
417 for (auto const& ignored_data : snapshot->ignored_data)
418 snapshot->process()->write_bytes(ignored_data.data.data(), ignored_data.data.size(), remote(ignored_data.start));
421 std::shared_ptr<simgrid::mc::Snapshot> take_snapshot(int num_state)
423 XBT_DEBUG("Taking snapshot %i", num_state);
425 simgrid::mc::RemoteClient* mc_process = &mc_model_checker->process();
427 std::shared_ptr<simgrid::mc::Snapshot> snapshot = std::make_shared<simgrid::mc::Snapshot>(mc_process, num_state);
429 for (auto const& p : mc_model_checker->process().actors())
430 snapshot->enabled_processes.insert(p.copy.getBuffer()->get_pid());
432 snapshot_handle_ignore(snapshot.get());
434 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
435 get_memory_regions(mc_process, snapshot.get());
437 snapshot->to_ignore = mc_model_checker->process().ignored_heap();
439 if (_sg_mc_max_visited_states > 0 || not _sg_mc_property_file.get().empty()) {
440 snapshot->stacks = take_snapshot_stacks(snapshot.get());
442 snapshot->hash = simgrid::mc::hash(*snapshot);
445 snapshot_ignore_restore(snapshot.get());
449 static inline void restore_snapshot_regions(simgrid::mc::Snapshot* snapshot)
451 for (std::unique_ptr<simgrid::mc::RegionSnapshot> const& region : snapshot->snapshot_regions) {
452 // For privatized, variables we decided it was not necessary to take the snapshot:
454 restore(region.get());
458 if (snapshot->privatization_index >= 0) {
459 // Fix the privatization mmap:
460 s_mc_message_restore_t message{MC_MESSAGE_RESTORE, snapshot->privatization_index};
461 mc_model_checker->process().getChannel().send(message);
466 void restore_snapshot(std::shared_ptr<simgrid::mc::Snapshot> snapshot)
468 XBT_DEBUG("Restore snapshot %i", snapshot->num_state);
469 restore_snapshot_regions(snapshot.get());
470 snapshot_ignore_restore(snapshot.get());
471 mc_model_checker->process().clear_cache();
475 } // namespace simgrid