// Copyright 2015 The Chromium Authors. All rights reserved. // Use of this source code is governed by a BSD-style license that can be // found in the LICENSE file. #include "base/trace_event/process_memory_dump.h" #include #include "base/memory/aligned_memory.h" #include "base/process/process_metrics.h" #include "base/trace_event/memory_allocator_dump_guid.h" #include "base/trace_event/trace_event_argument.h" #include "testing/gtest/include/gtest/gtest.h" namespace base { namespace trace_event { namespace { TracedValue* GetHeapDump(const ProcessMemoryDump& pmd, const char* name) { auto it = pmd.heap_dumps().find(name); return it == pmd.heap_dumps().end() ? nullptr : it->second.get(); } } // namespace TEST(ProcessMemoryDumpTest, Clear) { scoped_ptr pmd1(new ProcessMemoryDump(nullptr)); pmd1->CreateAllocatorDump("mad1"); pmd1->CreateAllocatorDump("mad2"); ASSERT_FALSE(pmd1->allocator_dumps().empty()); pmd1->process_totals()->set_resident_set_bytes(42); pmd1->set_has_process_totals(); pmd1->process_mmaps()->AddVMRegion(ProcessMemoryMaps::VMRegion()); pmd1->set_has_process_mmaps(); pmd1->AddOwnershipEdge(MemoryAllocatorDumpGuid(42), MemoryAllocatorDumpGuid(4242)); MemoryAllocatorDumpGuid shared_mad_guid1(1); MemoryAllocatorDumpGuid shared_mad_guid2(2); pmd1->CreateSharedGlobalAllocatorDump(shared_mad_guid1); pmd1->CreateSharedGlobalAllocatorDump(shared_mad_guid2); pmd1->Clear(); ASSERT_TRUE(pmd1->allocator_dumps().empty()); ASSERT_TRUE(pmd1->allocator_dumps_edges().empty()); ASSERT_EQ(nullptr, pmd1->GetAllocatorDump("mad1")); ASSERT_EQ(nullptr, pmd1->GetAllocatorDump("mad2")); ASSERT_FALSE(pmd1->has_process_totals()); ASSERT_FALSE(pmd1->has_process_mmaps()); ASSERT_TRUE(pmd1->process_mmaps()->vm_regions().empty()); ASSERT_EQ(nullptr, pmd1->GetSharedGlobalAllocatorDump(shared_mad_guid1)); ASSERT_EQ(nullptr, pmd1->GetSharedGlobalAllocatorDump(shared_mad_guid2)); // Check that calling AsValueInto() doesn't cause a crash. scoped_ptr traced_value(new TracedValue); pmd1->AsValueInto(traced_value.get()); // Check that the pmd can be reused and behaves as expected. auto mad1 = pmd1->CreateAllocatorDump("mad1"); auto mad3 = pmd1->CreateAllocatorDump("mad3"); auto shared_mad1 = pmd1->CreateSharedGlobalAllocatorDump(shared_mad_guid1); auto shared_mad2 = pmd1->CreateWeakSharedGlobalAllocatorDump(shared_mad_guid2); ASSERT_EQ(4u, pmd1->allocator_dumps().size()); ASSERT_EQ(mad1, pmd1->GetAllocatorDump("mad1")); ASSERT_EQ(nullptr, pmd1->GetAllocatorDump("mad2")); ASSERT_EQ(mad3, pmd1->GetAllocatorDump("mad3")); ASSERT_EQ(shared_mad1, pmd1->GetSharedGlobalAllocatorDump(shared_mad_guid1)); ASSERT_EQ(MemoryAllocatorDump::Flags::DEFAULT, shared_mad1->flags()); ASSERT_EQ(shared_mad2, pmd1->GetSharedGlobalAllocatorDump(shared_mad_guid2)); ASSERT_EQ(MemoryAllocatorDump::Flags::WEAK, shared_mad2->flags()); traced_value.reset(new TracedValue); pmd1->AsValueInto(traced_value.get()); pmd1.reset(); } TEST(ProcessMemoryDumpTest, TakeAllDumpsFrom) { scoped_ptr traced_value(new TracedValue); TracedValue* heap_dumps_ptr[4]; scoped_ptr heap_dump; scoped_ptr pmd1(new ProcessMemoryDump(nullptr)); auto mad1_1 = pmd1->CreateAllocatorDump("pmd1/mad1"); auto mad1_2 = pmd1->CreateAllocatorDump("pmd1/mad2"); pmd1->AddOwnershipEdge(mad1_1->guid(), mad1_2->guid()); heap_dump.reset(new TracedValue); heap_dumps_ptr[0] = heap_dump.get(); pmd1->AddHeapDump("pmd1/heap_dump1", std::move(heap_dump)); heap_dump.reset(new TracedValue); heap_dumps_ptr[1] = heap_dump.get(); pmd1->AddHeapDump("pmd1/heap_dump2", std::move(heap_dump)); scoped_ptr pmd2(new ProcessMemoryDump(nullptr)); auto mad2_1 = pmd2->CreateAllocatorDump("pmd2/mad1"); auto mad2_2 = pmd2->CreateAllocatorDump("pmd2/mad2"); pmd2->AddOwnershipEdge(mad2_1->guid(), mad2_2->guid()); heap_dump.reset(new TracedValue); heap_dumps_ptr[2] = heap_dump.get(); pmd2->AddHeapDump("pmd2/heap_dump1", std::move(heap_dump)); heap_dump.reset(new TracedValue); heap_dumps_ptr[3] = heap_dump.get(); pmd2->AddHeapDump("pmd2/heap_dump2", std::move(heap_dump)); MemoryAllocatorDumpGuid shared_mad_guid1(1); MemoryAllocatorDumpGuid shared_mad_guid2(2); auto shared_mad1 = pmd2->CreateSharedGlobalAllocatorDump(shared_mad_guid1); auto shared_mad2 = pmd2->CreateWeakSharedGlobalAllocatorDump(shared_mad_guid2); pmd1->TakeAllDumpsFrom(pmd2.get()); // Make sure that pmd2 is empty but still usable after it has been emptied. ASSERT_TRUE(pmd2->allocator_dumps().empty()); ASSERT_TRUE(pmd2->allocator_dumps_edges().empty()); ASSERT_TRUE(pmd2->heap_dumps().empty()); pmd2->CreateAllocatorDump("pmd2/this_mad_stays_with_pmd2"); ASSERT_EQ(1u, pmd2->allocator_dumps().size()); ASSERT_EQ(1u, pmd2->allocator_dumps().count("pmd2/this_mad_stays_with_pmd2")); pmd2->AddOwnershipEdge(MemoryAllocatorDumpGuid(42), MemoryAllocatorDumpGuid(4242)); // Check that calling AsValueInto() doesn't cause a crash. pmd2->AsValueInto(traced_value.get()); // Free the |pmd2| to check that the memory ownership of the two MAD(s) // has been transferred to |pmd1|. pmd2.reset(); // Now check that |pmd1| has been effectively merged. ASSERT_EQ(6u, pmd1->allocator_dumps().size()); ASSERT_EQ(1u, pmd1->allocator_dumps().count("pmd1/mad1")); ASSERT_EQ(1u, pmd1->allocator_dumps().count("pmd1/mad2")); ASSERT_EQ(1u, pmd1->allocator_dumps().count("pmd2/mad1")); ASSERT_EQ(1u, pmd1->allocator_dumps().count("pmd1/mad2")); ASSERT_EQ(2u, pmd1->allocator_dumps_edges().size()); ASSERT_EQ(shared_mad1, pmd1->GetSharedGlobalAllocatorDump(shared_mad_guid1)); ASSERT_EQ(shared_mad2, pmd1->GetSharedGlobalAllocatorDump(shared_mad_guid2)); ASSERT_TRUE(MemoryAllocatorDump::Flags::WEAK & shared_mad2->flags()); ASSERT_EQ(4u, pmd1->heap_dumps().size()); ASSERT_EQ(heap_dumps_ptr[0], GetHeapDump(*pmd1, "pmd1/heap_dump1")); ASSERT_EQ(heap_dumps_ptr[1], GetHeapDump(*pmd1, "pmd1/heap_dump2")); ASSERT_EQ(heap_dumps_ptr[2], GetHeapDump(*pmd1, "pmd2/heap_dump1")); ASSERT_EQ(heap_dumps_ptr[3], GetHeapDump(*pmd1, "pmd2/heap_dump2")); // Check that calling AsValueInto() doesn't cause a crash. traced_value.reset(new TracedValue); pmd1->AsValueInto(traced_value.get()); pmd1.reset(); } TEST(ProcessMemoryDumpTest, Suballocations) { scoped_ptr pmd(new ProcessMemoryDump(nullptr)); const std::string allocator_dump_name = "fakealloc/allocated_objects"; pmd->CreateAllocatorDump(allocator_dump_name); // Create one allocation with an auto-assigned guid and mark it as a // suballocation of "fakealloc/allocated_objects". auto pic1_dump = pmd->CreateAllocatorDump("picturemanager/picture1"); pmd->AddSuballocation(pic1_dump->guid(), allocator_dump_name); // Same here, but this time create an allocation with an explicit guid. auto pic2_dump = pmd->CreateAllocatorDump("picturemanager/picture2", MemoryAllocatorDumpGuid(0x42)); pmd->AddSuballocation(pic2_dump->guid(), allocator_dump_name); // Now check that AddSuballocation() has created anonymous child dumps under // "fakealloc/allocated_objects". auto anon_node_1_it = pmd->allocator_dumps().find( allocator_dump_name + "/__" + pic1_dump->guid().ToString()); ASSERT_NE(pmd->allocator_dumps().end(), anon_node_1_it); auto anon_node_2_it = pmd->allocator_dumps().find(allocator_dump_name + "/__42"); ASSERT_NE(pmd->allocator_dumps().end(), anon_node_2_it); // Finally check that AddSuballocation() has created also the // edges between the pictures and the anonymous allocator child dumps. bool found_edge[2]{false, false}; for (const auto& e : pmd->allocator_dumps_edges()) { found_edge[0] |= (e.source == pic1_dump->guid() && e.target == anon_node_1_it->second->guid()); found_edge[1] |= (e.source == pic2_dump->guid() && e.target == anon_node_2_it->second->guid()); } ASSERT_TRUE(found_edge[0]); ASSERT_TRUE(found_edge[1]); // Check that calling AsValueInto() doesn't cause a crash. scoped_ptr traced_value(new TracedValue); pmd->AsValueInto(traced_value.get()); pmd.reset(); } TEST(ProcessMemoryDumpTest, GlobalAllocatorDumpTest) { scoped_ptr pmd(new ProcessMemoryDump(nullptr)); MemoryAllocatorDumpGuid shared_mad_guid(1); auto shared_mad1 = pmd->CreateWeakSharedGlobalAllocatorDump(shared_mad_guid); ASSERT_EQ(shared_mad_guid, shared_mad1->guid()); ASSERT_EQ(MemoryAllocatorDump::Flags::WEAK, shared_mad1->flags()); auto shared_mad2 = pmd->GetSharedGlobalAllocatorDump(shared_mad_guid); ASSERT_EQ(shared_mad1, shared_mad2); ASSERT_EQ(MemoryAllocatorDump::Flags::WEAK, shared_mad1->flags()); auto shared_mad3 = pmd->CreateWeakSharedGlobalAllocatorDump(shared_mad_guid); ASSERT_EQ(shared_mad1, shared_mad3); ASSERT_EQ(MemoryAllocatorDump::Flags::WEAK, shared_mad1->flags()); auto shared_mad4 = pmd->CreateSharedGlobalAllocatorDump(shared_mad_guid); ASSERT_EQ(shared_mad1, shared_mad4); ASSERT_EQ(MemoryAllocatorDump::Flags::DEFAULT, shared_mad1->flags()); auto shared_mad5 = pmd->CreateWeakSharedGlobalAllocatorDump(shared_mad_guid); ASSERT_EQ(shared_mad1, shared_mad5); ASSERT_EQ(MemoryAllocatorDump::Flags::DEFAULT, shared_mad1->flags()); } #if defined(COUNT_RESIDENT_BYTES_SUPPORTED) TEST(ProcessMemoryDumpTest, CountResidentBytes) { const size_t page_size = ProcessMemoryDump::GetSystemPageSize(); // Allocate few page of dirty memory and check if it is resident. const size_t size1 = 5 * page_size; scoped_ptr memory1( static_cast(base::AlignedAlloc(size1, page_size))); memset(memory1.get(), 0, size1); size_t res1 = ProcessMemoryDump::CountResidentBytes(memory1.get(), size1); ASSERT_EQ(res1, size1); // Allocate a large memory segment (> 8Mib). const size_t kVeryLargeMemorySize = 15 * 1024 * 1024; scoped_ptr memory2( static_cast(base::AlignedAlloc(kVeryLargeMemorySize, page_size))); memset(memory2.get(), 0, kVeryLargeMemorySize); size_t res2 = ProcessMemoryDump::CountResidentBytes(memory2.get(), kVeryLargeMemorySize); ASSERT_EQ(res2, kVeryLargeMemorySize); } #endif // defined(COUNT_RESIDENT_BYTES_SUPPORTED) } // namespace trace_event } // namespace base