X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/e1a5484101efbff9c8d982a0c4650f7953ee7e2f..ef93ada6480099151585ece90d8f4aaf0be41cd5:/src/mc/compare.cpp diff --git a/src/mc/compare.cpp b/src/mc/compare.cpp index 47d39a5254..71fac79d7b 100644 --- a/src/mc/compare.cpp +++ b/src/mc/compare.cpp @@ -1,4 +1,4 @@ -/* Copyright (c) 2008-2021. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2008-2023. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ @@ -7,16 +7,15 @@ #include "src/mc/mc_config.hpp" #include "src/mc/mc_private.hpp" +#include "src/mc/sosp/RemoteProcessMemory.hpp" #include "src/mc/sosp/Snapshot.hpp" +#include "xbt/ex.h" #include -XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_compare, xbt, "Logging specific to mc_compare in mc"); +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_compare, mc, "Logging specific to mc_compare in mc"); -using simgrid::mc::remote; - -namespace simgrid { -namespace mc { +namespace simgrid::mc { /*********************************** Heap comparison ***********************************/ /***************************************************************************************/ @@ -78,7 +77,7 @@ public: compared_pointers.clear(); } - int initHeapInformation(const s_xbt_mheap_t* heap1, const s_xbt_mheap_t* heap2, + int initHeapInformation(RemoteProcessMemory& appli, const s_xbt_mheap_t* heap1, const s_xbt_mheap_t* heap2, const std::vector& i1, const std::vector& i2); template HeapArea& equals_to_(std::size_t i, std::size_t j) @@ -127,8 +126,7 @@ public: void match_equals(const HeapLocationPairs* list); }; -} // namespace mc -} // namespace simgrid +} // namespace simgrid::mc /************************************************************************************/ @@ -140,7 +138,7 @@ static ssize_t heap_comparison_ignore_size(const std::vectorend() && pos->address == address) ? pos->size : -1; } -static bool is_stack(const simgrid::mc::RemoteProcess& process, const void* address) +static bool is_stack(const simgrid::mc::RemoteProcessMemory& process, const void* address) { auto const& stack_areas = process.stack_areas(); return std::any_of(stack_areas.begin(), stack_areas.end(), @@ -148,15 +146,14 @@ static bool is_stack(const simgrid::mc::RemoteProcess& process, const void* addr } // TODO, this should depend on the snapshot? -static bool is_block_stack(const simgrid::mc::RemoteProcess& process, int block) +static bool is_block_stack(const simgrid::mc::RemoteProcessMemory& process, int block) { auto const& stack_areas = process.stack_areas(); return std::any_of(stack_areas.begin(), stack_areas.end(), [block](auto const& stack) { return stack.block == block; }); } -namespace simgrid { -namespace mc { +namespace simgrid::mc { void StateComparator::match_equals(const HeapLocationPairs* list) { @@ -180,14 +177,14 @@ void ProcessComparisonState::initHeapInformation(const s_xbt_mheap_t* heap, cons this->types.assign(heaplimit * MAX_FRAGMENT_PER_BLOCK, nullptr); } -int StateComparator::initHeapInformation(const s_xbt_mheap_t* heap1, const s_xbt_mheap_t* heap2, - const std::vector& i1, +int StateComparator::initHeapInformation(simgrid::mc::RemoteProcessMemory& memory, const s_xbt_mheap_t* heap1, + const s_xbt_mheap_t* heap2, const std::vector& i1, const std::vector& i2) { if ((heap1->heaplimit != heap2->heaplimit) || (heap1->heapsize != heap2->heapsize)) return -1; this->heaplimit = heap1->heaplimit; - this->std_heap_copy = *mc_model_checker->get_remote_process().get_heap(); + this->std_heap_copy = *memory.get_heap(); this->processStates[0].initHeapInformation(heap1, i1); this->processStates[1].initHeapInformation(heap2, i2); return 0; @@ -202,11 +199,42 @@ static inline Region* MC_get_heap_region(const Snapshot& snapshot) xbt_die("No heap region"); } -static bool heap_area_differ(const RemoteProcess& process, StateComparator& state, const void* area1, const void* area2, - const Snapshot& snapshot1, const Snapshot& snapshot2, HeapLocationPairs* previous, - Type* type, int pointer_level); +static bool heap_area_differ(const RemoteProcessMemory& process, StateComparator& state, const void* area1, + const void* area2, const Snapshot& snapshot1, const Snapshot& snapshot2, + HeapLocationPairs* previous, Type* type, int pointer_level); -static bool mmalloc_heap_differ(const RemoteProcess& process, StateComparator& state, const Snapshot& snapshot1, +/* Compares the content of each heap fragment between the two states, at the bit level. + * + * This operation is costly (about 5 seconds per snapshots' pair to compare on a small program), + * but hard to optimize because our algorithm is too hackish. + * + * Going at bit level can trigger syntaxtic differences on states that are semantically equivalent. + * + * Padding bytes constitute the first source of such syntaxtic difference: Any malloced memory contains spaces that + * are not used to enforce the memory alignment constraints of the CPU. So, cruft of irrelevant changes could get + * added on these bits. But this case is handled properly, as any memory block is zeroed by mmalloc before being handled + * back, not only for calloc but also for malloc. So the memory interstices due to padding bytes are properly zeroed. + * + * Another source of such change comes from the order of mallocs, that may well change from one execution path to + * another. This will change the malloc fragment in which the data is stored and the pointer values (syntaxtic + * difference) while the semantic of the state remains the same. + * + * To fix this, this code relies on a hugly hack. When we see a difference during the bit-level comparison, + * we first check if it could be explained by a pointer-to-block difference. Ie, if when interpreting the memory + * area containing that difference as a pointer, I get the pointer to a valid fragment in the heap (in both snapshots). + * + * This is why we cannot pre-compute a bit-level hash of the heap content: we discover the pointers to other memory + * fragment when a difference is found during the bit-level exploration. Fixing this would require to save typing + * information about the memory fragments, which is something that could be done with https://github.com/tudasc/TypeART + * This would give us all pointers in the mallocated memory, allowing the graph traversal needed to precompute the hash. + * + * Using a hash without paying attention to malloc fragment reordering would lead to false negatives: + * semantically equivalent states would be detected as [syntaxically] different. It's of no importance for the + * state-equality reduction (we would re-explore semantically equivalent states), but it would endanger the soundness + * of the liveness model-checker, as state-equality is used to detect the loops that constitute the accepting states of + * the verified property. So we could miss counter-examples to the verified property. Not good. Not good at all. + */ +static bool mmalloc_heap_differ(const RemoteProcessMemory& process, StateComparator& state, const Snapshot& snapshot1, const Snapshot& snapshot2) { /* Check busy blocks */ @@ -239,7 +267,7 @@ static bool mmalloc_heap_differ(const RemoteProcess& process, StateComparator& s xbt_assert(heapinfo1->type >= 0, "Unknown mmalloc block type: %d", heapinfo1->type); - void* addr_block1 = ((void*)(((ADDR2UINT(i1)) - 1) * BLOCKSIZE + (char*)state.std_heap_copy.heapbase)); + void* addr_block1 = (ADDR2UINT(i1) - 1) * BLOCKSIZE + (char*)state.std_heap_copy.heapbase; if (heapinfo1->type == MMALLOC_TYPE_UNFRAGMENTED) { /* Large block */ if (is_stack(process, addr_block1)) { @@ -430,9 +458,10 @@ static bool mmalloc_heap_differ(const RemoteProcess& process, StateComparator& s * @param check_ignore * @return true when different, false otherwise (same or unknown) */ -static bool heap_area_differ_without_type(const RemoteProcess& process, StateComparator& state, const void* real_area1, - const void* real_area2, const Snapshot& snapshot1, const Snapshot& snapshot2, - HeapLocationPairs* previous, int size, int check_ignore) +static bool heap_area_differ_without_type(const RemoteProcessMemory& process, StateComparator& state, + const void* real_area1, const void* real_area2, const Snapshot& snapshot1, + const Snapshot& snapshot2, HeapLocationPairs* previous, int size, + int check_ignore) { const Region* heap_region1 = MC_get_heap_region(snapshot1); const Region* heap_region2 = MC_get_heap_region(snapshot2); @@ -493,7 +522,7 @@ static bool heap_area_differ_without_type(const RemoteProcess& process, StateCom * @param pointer_level * @return true when different, false otherwise (same or unknown) */ -static bool heap_area_differ_with_type(const simgrid::mc::RemoteProcess& process, StateComparator& state, +static bool heap_area_differ_with_type(const simgrid::mc::RemoteProcessMemory& process, StateComparator& state, const void* real_area1, const void* real_area2, const Snapshot& snapshot1, const Snapshot& snapshot2, HeapLocationPairs* previous, const Type* type, int area_size, int check_ignore, int pointer_level) @@ -716,9 +745,9 @@ static Type* get_offset_type(void* real_base_address, Type* type, int offset, in * @param pointer_level * @return true when different, false otherwise (same or unknown) */ -static bool heap_area_differ(const RemoteProcess& process, StateComparator& state, const void* area1, const void* area2, - const Snapshot& snapshot1, const Snapshot& snapshot2, HeapLocationPairs* previous, - Type* type, int pointer_level) +static bool heap_area_differ(const RemoteProcessMemory& process, StateComparator& state, const void* area1, + const void* area2, const Snapshot& snapshot1, const Snapshot& snapshot2, + HeapLocationPairs* previous, Type* type, int pointer_level) { ssize_t block1; ssize_t block2; @@ -764,9 +793,9 @@ static bool heap_area_differ(const RemoteProcess& process, StateComparator& stat // If either block is not in the expected area of memory: if (((const char*)area1 < (const char*)state.std_heap_copy.heapbase) || - (block1 > (ssize_t)state.processStates[0].heapsize) || (block1 < 1) || + (block1 > (ssize_t)state.processStates[0].heapsize) || ((const char*)area2 < (const char*)state.std_heap_copy.heapbase) || - (block2 > (ssize_t)state.processStates[1].heapsize) || (block2 < 1)) { + (block2 > (ssize_t)state.processStates[1].heapsize)) { return true; } @@ -819,8 +848,7 @@ static bool heap_area_differ(const RemoteProcess& process, StateComparator& stat } if (type_size != -1 && type_size != (ssize_t)heapinfo1->busy_block.busy_size && - type_size != (ssize_t)heapinfo2->busy_block.busy_size && - (type->name.empty() || type->name == "struct s_smx_context")) { + type_size != (ssize_t)heapinfo2->busy_block.busy_size && type->name.empty()) { if (match_pairs) state.match_equals(previous); return false; @@ -880,7 +908,7 @@ static bool heap_area_differ(const RemoteProcess& process, StateComparator& stat } // Check if the blocks are already matched together: - if (state.equals_to_<1>(block1, frag1).valid_ && state.equals_to_<2>(block2, frag2).valid_ && offset1 == offset2 && + if (state.equals_to_<1>(block1, frag1).valid_ && state.equals_to_<2>(block2, frag2).valid_ && state.fragmentsEqual(block1, frag1, block2, frag2)) { if (match_pairs) state.match_equals(previous); @@ -975,24 +1003,22 @@ static bool heap_area_differ(const RemoteProcess& process, StateComparator& stat return true; /* Start comparison */ - bool differ = type ? heap_area_differ_with_type(process, state, area1, area2, snapshot1, snapshot2, previous, type, - size, check_ignore, pointer_level) - : heap_area_differ_without_type(process, state, area1, area2, snapshot1, snapshot2, previous, size, - check_ignore); - if (differ) + if (type ? heap_area_differ_with_type(process, state, area1, area2, snapshot1, snapshot2, previous, type, size, + check_ignore, pointer_level) + : heap_area_differ_without_type(process, state, area1, area2, snapshot1, snapshot2, previous, size, + check_ignore)) return true; if (match_pairs) state.match_equals(previous); return false; } -} // namespace mc -} // namespace simgrid +} // namespace simgrid::mc /************************** Snapshot comparison *******************************/ /******************************************************************************/ -static bool areas_differ_with_type(const simgrid::mc::RemoteProcess& process, simgrid::mc::StateComparator& state, +static bool areas_differ_with_type(const simgrid::mc::RemoteProcessMemory& process, simgrid::mc::StateComparator& state, const void* real_area1, const simgrid::mc::Snapshot& snapshot1, simgrid::mc::Region* region1, const void* real_area2, const simgrid::mc::Snapshot& snapshot2, simgrid::mc::Region* region2, @@ -1119,7 +1145,8 @@ static bool areas_differ_with_type(const simgrid::mc::RemoteProcess& process, si return false; } -static bool global_variables_differ(const simgrid::mc::RemoteProcess& process, simgrid::mc::StateComparator& state, +static bool global_variables_differ(const simgrid::mc::RemoteProcessMemory& process, + simgrid::mc::StateComparator& state, const simgrid::mc::ObjectInformation* object_info, simgrid::mc::Region* r1, simgrid::mc::Region* r2, const simgrid::mc::Snapshot& snapshot1, const simgrid::mc::Snapshot& snapshot2) @@ -1146,7 +1173,7 @@ static bool global_variables_differ(const simgrid::mc::RemoteProcess& process, s return false; } -static bool local_variables_differ(const simgrid::mc::RemoteProcess& process, simgrid::mc::StateComparator& state, +static bool local_variables_differ(const simgrid::mc::RemoteProcessMemory& process, simgrid::mc::StateComparator& state, const simgrid::mc::Snapshot& snapshot1, const simgrid::mc::Snapshot& snapshot2, const_mc_snapshot_stack_t stack1, const_mc_snapshot_stack_t stack2) { @@ -1178,35 +1205,30 @@ static bool local_variables_differ(const simgrid::mc::RemoteProcess& process, si return false; } -namespace simgrid { -namespace mc { +namespace simgrid::mc { -bool snapshot_equal(const Snapshot* s1, const Snapshot* s2) +bool Snapshot::operator==(const Snapshot& other) { // TODO, make this a field of ModelChecker or something similar static StateComparator state_comparator; - const RemoteProcess& process = mc_model_checker->get_remote_process(); + RemoteProcessMemory& memory = mc_model_checker->get_remote_process_memory(); - if (s1->hash_ != s2->hash_) { - XBT_VERB("(%d - %d) Different hash: 0x%" PRIx64 "--0x%" PRIx64, s1->num_state_, s2->num_state_, s1->hash_, - s2->hash_); + if (hash_ != other.hash_) { + XBT_VERB("(%ld - %ld) Different hash: 0x%" PRIx64 "--0x%" PRIx64, this->num_state_, other.num_state_, this->hash_, + other.hash_); return false; } - XBT_VERB("(%d - %d) Same hash: 0x%" PRIx64, s1->num_state_, s2->num_state_, s1->hash_); + XBT_VERB("(%ld - %ld) Same hash: 0x%" PRIx64, this->num_state_, other.num_state_, this->hash_); - /* Compare enabled processes */ - if (s1->enabled_processes_ != s2->enabled_processes_) { - XBT_VERB("(%d - %d) Different amount of enabled processes", s1->num_state_, s2->num_state_); - return false; - } + /* TODO: re-enable the quick filter of counting enabled processes in each snapshots */ /* Compare size of stacks */ - for (unsigned long i = 0; i < s1->stacks_.size(); i++) { - size_t size_used1 = s1->stack_sizes_[i]; - size_t size_used2 = s2->stack_sizes_[i]; + for (unsigned long i = 0; i < this->stacks_.size(); i++) { + size_t size_used1 = this->stack_sizes_[i]; + size_t size_used2 = other.stack_sizes_[i]; if (size_used1 != size_used2) { - XBT_VERB("(%d - %d) Different size used in stacks: %zu - %zu", s1->num_state_, s2->num_state_, size_used1, + XBT_VERB("(%ld - %ld) Different size used in stacks: %zu - %zu", num_state_, other.num_state_, size_used1, size_used2); return false; } @@ -1214,32 +1236,33 @@ bool snapshot_equal(const Snapshot* s1, const Snapshot* s2) /* Init heap information used in heap comparison algorithm */ const s_xbt_mheap_t* heap1 = static_cast( - s1->read_bytes(alloca(sizeof(s_xbt_mheap_t)), sizeof(s_xbt_mheap_t), process.heap_address, ReadOptions::lazy())); + this->read_bytes(alloca(sizeof(s_xbt_mheap_t)), sizeof(s_xbt_mheap_t), memory.heap_address, ReadOptions::lazy())); const s_xbt_mheap_t* heap2 = static_cast( - s2->read_bytes(alloca(sizeof(s_xbt_mheap_t)), sizeof(s_xbt_mheap_t), process.heap_address, ReadOptions::lazy())); - if (state_comparator.initHeapInformation(heap1, heap2, s1->to_ignore_, s2->to_ignore_) == -1) { - XBT_VERB("(%d - %d) Different heap information", s1->num_state_, s2->num_state_); + other.read_bytes(alloca(sizeof(s_xbt_mheap_t)), sizeof(s_xbt_mheap_t), memory.heap_address, ReadOptions::lazy())); + if (state_comparator.initHeapInformation(memory, heap1, heap2, this->to_ignore_, other.to_ignore_) == -1) { + XBT_VERB("(%ld - %ld) Different heap information", this->num_state_, other.num_state_); return false; } /* Stacks comparison */ - for (unsigned int cursor = 0; cursor < s1->stacks_.size(); cursor++) { - const_mc_snapshot_stack_t stack1 = &s1->stacks_[cursor]; - const_mc_snapshot_stack_t stack2 = &s2->stacks_[cursor]; + for (unsigned int cursor = 0; cursor < this->stacks_.size(); cursor++) { + const_mc_snapshot_stack_t stack1 = &this->stacks_[cursor]; + const_mc_snapshot_stack_t stack2 = &other.stacks_[cursor]; - if (local_variables_differ(process, state_comparator, *s1, *s2, stack1, stack2)) { - XBT_VERB("(%d - %d) Different local variables between stacks %u", s1->num_state_, s2->num_state_, cursor + 1); + if (local_variables_differ(memory, state_comparator, *this, other, stack1, stack2)) { + XBT_VERB("(%ld - %ld) Different local variables between stacks %u", this->num_state_, other.num_state_, + cursor + 1); return false; } } - size_t regions_count = s1->snapshot_regions_.size(); - if (regions_count != s2->snapshot_regions_.size()) + size_t regions_count = this->snapshot_regions_.size(); + if (regions_count != other.snapshot_regions_.size()) return false; for (size_t k = 0; k != regions_count; ++k) { - Region* region1 = s1->snapshot_regions_[k].get(); - Region* region2 = s2->snapshot_regions_[k].get(); + Region* region1 = this->snapshot_regions_[k].get(); + Region* region2 = other.snapshot_regions_[k].get(); // Preconditions: if (region1->region_type() != RegionType::Data) @@ -1250,22 +1273,22 @@ bool snapshot_equal(const Snapshot* s1, const Snapshot* s2) xbt_assert(region1->object_info()); /* Compare global variables */ - if (global_variables_differ(process, state_comparator, region1->object_info(), region1, region2, *s1, *s2)) { + if (global_variables_differ(memory, state_comparator, region1->object_info(), region1, region2, *this, other)) { std::string const& name = region1->object_info()->file_name; - XBT_VERB("(%d - %d) Different global variables in %s", s1->num_state_, s2->num_state_, name.c_str()); + XBT_VERB("(%ld - %ld) Different global variables in %s", this->num_state_, other.num_state_, name.c_str()); return false; } } + XBT_VERB(" Compare heap..."); /* Compare heap */ - if (mmalloc_heap_differ(process, state_comparator, *s1, *s2)) { - XBT_VERB("(%d - %d) Different heap (mmalloc_compare)", s1->num_state_, s2->num_state_); + if (mmalloc_heap_differ(memory, state_comparator, *this, other)) { + XBT_VERB("(%ld - %ld) Different heap (mmalloc_heap_differ)", this->num_state_, other.num_state_); return false; } - XBT_VERB("(%d - %d) No difference found", s1->num_state_, s2->num_state_); + XBT_VERB("(%ld - %ld) No difference found", this->num_state_, other.num_state_); return true; } -} // namespace mc -} // namespace simgrid +} // namespace simgrid::mc