• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2011 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #include "heap.h"
18 
19 #include <limits>
20 #include "android-base/thread_annotations.h"
21 #if defined(__BIONIC__) || defined(__GLIBC__)
22 #include <malloc.h>  // For mallinfo()
23 #endif
24 #include <memory>
25 #include <random>
26 #include <unistd.h>
27 #include <sys/types.h>
28 #include <vector>
29 
30 #include "android-base/stringprintf.h"
31 
32 #include "allocation_listener.h"
33 #include "art_field-inl.h"
34 #include "backtrace_helper.h"
35 #include "base/allocator.h"
36 #include "base/arena_allocator.h"
37 #include "base/dumpable.h"
38 #include "base/file_utils.h"
39 #include "base/histogram-inl.h"
40 #include "base/logging.h"  // For VLOG.
41 #include "base/memory_tool.h"
42 #include "base/mutex.h"
43 #include "base/os.h"
44 #include "base/stl_util.h"
45 #include "base/systrace.h"
46 #include "base/time_utils.h"
47 #include "base/utils.h"
48 #include "class_root-inl.h"
49 #include "common_throws.h"
50 #include "debugger.h"
51 #include "dex/dex_file-inl.h"
52 #include "entrypoints/quick/quick_alloc_entrypoints.h"
53 #include "gc/accounting/card_table-inl.h"
54 #include "gc/accounting/heap_bitmap-inl.h"
55 #include "gc/accounting/mod_union_table-inl.h"
56 #include "gc/accounting/read_barrier_table.h"
57 #include "gc/accounting/remembered_set.h"
58 #include "gc/accounting/space_bitmap-inl.h"
59 #include "gc/collector/concurrent_copying.h"
60 #include "gc/collector/mark_compact.h"
61 #include "gc/collector/mark_sweep.h"
62 #include "gc/collector/partial_mark_sweep.h"
63 #include "gc/collector/semi_space.h"
64 #include "gc/collector/sticky_mark_sweep.h"
65 #include "gc/racing_check.h"
66 #include "gc/reference_processor.h"
67 #include "gc/scoped_gc_critical_section.h"
68 #include "gc/space/bump_pointer_space.h"
69 #include "gc/space/dlmalloc_space-inl.h"
70 #include "gc/space/image_space.h"
71 #include "gc/space/large_object_space.h"
72 #include "gc/space/region_space.h"
73 #include "gc/space/rosalloc_space-inl.h"
74 #include "gc/space/space-inl.h"
75 #include "gc/space/zygote_space.h"
76 #include "gc/task_processor.h"
77 #include "gc/verification.h"
78 #include "gc_pause_listener.h"
79 #include "gc_root.h"
80 #include "handle_scope-inl.h"
81 #include "heap-inl.h"
82 #include "heap-visit-objects-inl.h"
83 #include "image.h"
84 #include "intern_table.h"
85 #include "jit/jit.h"
86 #include "jit/jit_code_cache.h"
87 #include "jni/java_vm_ext.h"
88 #include "mirror/class-inl.h"
89 #include "mirror/executable-inl.h"
90 #include "mirror/field.h"
91 #include "mirror/method_handle_impl.h"
92 #include "mirror/object-inl.h"
93 #include "mirror/object-refvisitor-inl.h"
94 #include "mirror/object_array-inl.h"
95 #include "mirror/reference-inl.h"
96 #include "mirror/var_handle.h"
97 #include "nativehelper/scoped_local_ref.h"
98 #include "obj_ptr-inl.h"
99 #ifdef ART_TARGET_ANDROID
100 #include "perfetto/heap_profile.h"
101 #endif
102 #include "reflection.h"
103 #include "runtime.h"
104 #include "javaheapprof/javaheapsampler.h"
105 #include "scoped_thread_state_change-inl.h"
106 #include "thread-inl.h"
107 #include "thread_list.h"
108 #include "verify_object-inl.h"
109 #include "well_known_classes.h"
110 
111 namespace art {
112 
113 #ifdef ART_TARGET_ANDROID
114 namespace {
115 
116 // Enable the heap sampler Callback function used by Perfetto.
EnableHeapSamplerCallback(void * enable_ptr,const AHeapProfileEnableCallbackInfo * enable_info_ptr)117 void EnableHeapSamplerCallback(void* enable_ptr,
118                                const AHeapProfileEnableCallbackInfo* enable_info_ptr) {
119   HeapSampler* sampler_self = reinterpret_cast<HeapSampler*>(enable_ptr);
120   // Set the ART profiler sampling interval to the value from Perfetto.
121   uint64_t interval = AHeapProfileEnableCallbackInfo_getSamplingInterval(enable_info_ptr);
122   if (interval > 0) {
123     sampler_self->SetSamplingInterval(interval);
124   }
125   // Else default is 4K sampling interval. However, default case shouldn't happen for Perfetto API.
126   // AHeapProfileEnableCallbackInfo_getSamplingInterval should always give the requested
127   // (non-negative) sampling interval. It is a uint64_t and gets checked for != 0
128   // Do not call heap as a temp here, it will build but test run will silently fail.
129   // Heap is not fully constructed yet in some cases.
130   sampler_self->EnableHeapSampler();
131 }
132 
133 // Disable the heap sampler Callback function used by Perfetto.
DisableHeapSamplerCallback(void * disable_ptr,const AHeapProfileDisableCallbackInfo * info_ptr ATTRIBUTE_UNUSED)134 void DisableHeapSamplerCallback(void* disable_ptr,
135                                 const AHeapProfileDisableCallbackInfo* info_ptr ATTRIBUTE_UNUSED) {
136   HeapSampler* sampler_self = reinterpret_cast<HeapSampler*>(disable_ptr);
137   sampler_self->DisableHeapSampler();
138 }
139 
140 }  // namespace
141 #endif
142 
143 namespace gc {
144 
145 DEFINE_RUNTIME_DEBUG_FLAG(Heap, kStressCollectorTransition);
146 
147 // Minimum amount of remaining bytes before a concurrent GC is triggered.
148 static constexpr size_t kMinConcurrentRemainingBytes = 128 * KB;
149 static constexpr size_t kMaxConcurrentRemainingBytes = 512 * KB;
150 // Sticky GC throughput adjustment, divided by 4. Increasing this causes sticky GC to occur more
151 // relative to partial/full GC. This may be desirable since sticky GCs interfere less with mutator
152 // threads (lower pauses, use less memory bandwidth).
GetStickyGcThroughputAdjustment(bool use_generational_cc)153 static double GetStickyGcThroughputAdjustment(bool use_generational_cc) {
154   return use_generational_cc ? 0.5 : 1.0;
155 }
156 // Whether or not we compact the zygote in PreZygoteFork.
157 static constexpr bool kCompactZygote = kMovingCollector;
158 // How many reserve entries are at the end of the allocation stack, these are only needed if the
159 // allocation stack overflows.
160 static constexpr size_t kAllocationStackReserveSize = 1024;
161 // Default mark stack size in bytes.
162 static const size_t kDefaultMarkStackSize = 64 * KB;
163 // Define space name.
164 static const char* kDlMallocSpaceName[2] = {"main dlmalloc space", "main dlmalloc space 1"};
165 static const char* kRosAllocSpaceName[2] = {"main rosalloc space", "main rosalloc space 1"};
166 static const char* kMemMapSpaceName[2] = {"main space", "main space 1"};
167 static const char* kNonMovingSpaceName = "non moving space";
168 static const char* kZygoteSpaceName = "zygote space";
169 static constexpr bool kGCALotMode = false;
170 // GC alot mode uses a small allocation stack to stress test a lot of GC.
171 static constexpr size_t kGcAlotAllocationStackSize = 4 * KB /
172     sizeof(mirror::HeapReference<mirror::Object>);
173 // Verify objet has a small allocation stack size since searching the allocation stack is slow.
174 static constexpr size_t kVerifyObjectAllocationStackSize = 16 * KB /
175     sizeof(mirror::HeapReference<mirror::Object>);
176 static constexpr size_t kDefaultAllocationStackSize = 8 * MB /
177     sizeof(mirror::HeapReference<mirror::Object>);
178 
179 // After a GC (due to allocation failure) we should retrieve at least this
180 // fraction of the current max heap size. Otherwise throw OOME.
181 static constexpr double kMinFreeHeapAfterGcForAlloc = 0.01;
182 
183 // For deterministic compilation, we need the heap to be at a well-known address.
184 static constexpr uint32_t kAllocSpaceBeginForDeterministicAoT = 0x40000000;
185 // Dump the rosalloc stats on SIGQUIT.
186 static constexpr bool kDumpRosAllocStatsOnSigQuit = false;
187 
188 static const char* kRegionSpaceName = "main space (region space)";
189 
190 // If true, we log all GCs in the both the foreground and background. Used for debugging.
191 static constexpr bool kLogAllGCs = false;
192 
193 // Use Max heap for 2 seconds, this is smaller than the usual 5s window since we don't want to leave
194 // allocate with relaxed ergonomics for that long.
195 static constexpr size_t kPostForkMaxHeapDurationMS = 2000;
196 
197 #if defined(__LP64__) || !defined(ADDRESS_SANITIZER)
198 // 300 MB (0x12c00000) - (default non-moving space capacity).
199 uint8_t* const Heap::kPreferredAllocSpaceBegin =
200     reinterpret_cast<uint8_t*>(300 * MB - kDefaultNonMovingSpaceCapacity);
201 #else
202 #ifdef __ANDROID__
203 // For 32-bit Android, use 0x20000000 because asan reserves 0x04000000 - 0x20000000.
204 uint8_t* const Heap::kPreferredAllocSpaceBegin = reinterpret_cast<uint8_t*>(0x20000000);
205 #else
206 // For 32-bit host, use 0x40000000 because asan uses most of the space below this.
207 uint8_t* const Heap::kPreferredAllocSpaceBegin = reinterpret_cast<uint8_t*>(0x40000000);
208 #endif
209 #endif
210 
211 // Log GC on regular (but fairly large) intervals during GC stress mode.
212 // It is expected that the other runtime options will be used to reduce the usual logging.
213 // This allows us to make the logging much less verbose while still reporting some
214 // progress (biased towards expensive GCs), and while still reporting pathological cases.
215 static constexpr int64_t kGcStressModeGcLogSampleFrequencyNs = MsToNs(10000);
216 
CareAboutPauseTimes()217 static inline bool CareAboutPauseTimes() {
218   return Runtime::Current()->InJankPerceptibleProcessState();
219 }
220 
VerifyBootImagesContiguity(const std::vector<gc::space::ImageSpace * > & image_spaces)221 static void VerifyBootImagesContiguity(const std::vector<gc::space::ImageSpace*>& image_spaces) {
222   uint32_t boot_image_size = 0u;
223   for (size_t i = 0u, num_spaces = image_spaces.size(); i != num_spaces; ) {
224     const ImageHeader& image_header = image_spaces[i]->GetImageHeader();
225     uint32_t reservation_size = image_header.GetImageReservationSize();
226     uint32_t image_count = image_header.GetImageSpaceCount();
227 
228     CHECK_NE(image_count, 0u);
229     CHECK_LE(image_count, num_spaces - i);
230     CHECK_NE(reservation_size, 0u);
231     for (size_t j = 1u; j != image_count; ++j) {
232       CHECK_EQ(image_spaces[i + j]->GetImageHeader().GetComponentCount(), 0u);
233       CHECK_EQ(image_spaces[i + j]->GetImageHeader().GetImageReservationSize(), 0u);
234     }
235 
236     // Check the start of the heap.
237     CHECK_EQ(image_spaces[0]->Begin() + boot_image_size, image_spaces[i]->Begin());
238     // Check contiguous layout of images and oat files.
239     const uint8_t* current_heap = image_spaces[i]->Begin();
240     const uint8_t* current_oat = image_spaces[i]->GetImageHeader().GetOatFileBegin();
241     for (size_t j = 0u; j != image_count; ++j) {
242       const ImageHeader& current_header = image_spaces[i + j]->GetImageHeader();
243       CHECK_EQ(current_heap, image_spaces[i + j]->Begin());
244       CHECK_EQ(current_oat, current_header.GetOatFileBegin());
245       current_heap += RoundUp(current_header.GetImageSize(), kPageSize);
246       CHECK_GT(current_header.GetOatFileEnd(), current_header.GetOatFileBegin());
247       current_oat = current_header.GetOatFileEnd();
248     }
249     // Check that oat files start at the end of images.
250     CHECK_EQ(current_heap, image_spaces[i]->GetImageHeader().GetOatFileBegin());
251     // Check that the reservation size equals the size of images and oat files.
252     CHECK_EQ(reservation_size, static_cast<size_t>(current_oat - image_spaces[i]->Begin()));
253 
254     boot_image_size += reservation_size;
255     i += image_count;
256   }
257 }
258 
Heap(size_t initial_size,size_t growth_limit,size_t min_free,size_t max_free,double target_utilization,double foreground_heap_growth_multiplier,size_t stop_for_native_allocs,size_t capacity,size_t non_moving_space_capacity,const std::vector<std::string> & boot_class_path,const std::vector<std::string> & boot_class_path_locations,const std::vector<int> & boot_class_path_fds,const std::vector<int> & boot_class_path_image_fds,const std::vector<int> & boot_class_path_vdex_fds,const std::vector<int> & boot_class_path_oat_fds,const std::vector<std::string> & image_file_names,const InstructionSet image_instruction_set,CollectorType foreground_collector_type,CollectorType background_collector_type,space::LargeObjectSpaceType large_object_space_type,size_t large_object_threshold,size_t parallel_gc_threads,size_t conc_gc_threads,bool low_memory_mode,size_t long_pause_log_threshold,size_t long_gc_log_threshold,bool ignore_target_footprint,bool always_log_explicit_gcs,bool use_tlab,bool verify_pre_gc_heap,bool verify_pre_sweeping_heap,bool verify_post_gc_heap,bool verify_pre_gc_rosalloc,bool verify_pre_sweeping_rosalloc,bool verify_post_gc_rosalloc,bool gc_stress_mode,bool measure_gc_performance,bool use_homogeneous_space_compaction_for_oom,bool use_generational_cc,uint64_t min_interval_homogeneous_space_compaction_by_oom,bool dump_region_info_before_gc,bool dump_region_info_after_gc)259 Heap::Heap(size_t initial_size,
260            size_t growth_limit,
261            size_t min_free,
262            size_t max_free,
263            double target_utilization,
264            double foreground_heap_growth_multiplier,
265            size_t stop_for_native_allocs,
266            size_t capacity,
267            size_t non_moving_space_capacity,
268            const std::vector<std::string>& boot_class_path,
269            const std::vector<std::string>& boot_class_path_locations,
270            const std::vector<int>& boot_class_path_fds,
271            const std::vector<int>& boot_class_path_image_fds,
272            const std::vector<int>& boot_class_path_vdex_fds,
273            const std::vector<int>& boot_class_path_oat_fds,
274            const std::vector<std::string>& image_file_names,
275            const InstructionSet image_instruction_set,
276            CollectorType foreground_collector_type,
277            CollectorType background_collector_type,
278            space::LargeObjectSpaceType large_object_space_type,
279            size_t large_object_threshold,
280            size_t parallel_gc_threads,
281            size_t conc_gc_threads,
282            bool low_memory_mode,
283            size_t long_pause_log_threshold,
284            size_t long_gc_log_threshold,
285            bool ignore_target_footprint,
286            bool always_log_explicit_gcs,
287            bool use_tlab,
288            bool verify_pre_gc_heap,
289            bool verify_pre_sweeping_heap,
290            bool verify_post_gc_heap,
291            bool verify_pre_gc_rosalloc,
292            bool verify_pre_sweeping_rosalloc,
293            bool verify_post_gc_rosalloc,
294            bool gc_stress_mode,
295            bool measure_gc_performance,
296            bool use_homogeneous_space_compaction_for_oom,
297            bool use_generational_cc,
298            uint64_t min_interval_homogeneous_space_compaction_by_oom,
299            bool dump_region_info_before_gc,
300            bool dump_region_info_after_gc)
301     : non_moving_space_(nullptr),
302       rosalloc_space_(nullptr),
303       dlmalloc_space_(nullptr),
304       main_space_(nullptr),
305       collector_type_(kCollectorTypeNone),
306       foreground_collector_type_(foreground_collector_type),
307       background_collector_type_(background_collector_type),
308       desired_collector_type_(foreground_collector_type_),
309       pending_task_lock_(nullptr),
310       parallel_gc_threads_(parallel_gc_threads),
311       conc_gc_threads_(conc_gc_threads),
312       low_memory_mode_(low_memory_mode),
313       long_pause_log_threshold_(long_pause_log_threshold),
314       long_gc_log_threshold_(long_gc_log_threshold),
315       process_cpu_start_time_ns_(ProcessCpuNanoTime()),
316       pre_gc_last_process_cpu_time_ns_(process_cpu_start_time_ns_),
317       post_gc_last_process_cpu_time_ns_(process_cpu_start_time_ns_),
318       pre_gc_weighted_allocated_bytes_(0.0),
319       post_gc_weighted_allocated_bytes_(0.0),
320       ignore_target_footprint_(ignore_target_footprint),
321       always_log_explicit_gcs_(always_log_explicit_gcs),
322       zygote_creation_lock_("zygote creation lock", kZygoteCreationLock),
323       zygote_space_(nullptr),
324       large_object_threshold_(large_object_threshold),
325       disable_thread_flip_count_(0),
326       thread_flip_running_(false),
327       collector_type_running_(kCollectorTypeNone),
328       last_gc_cause_(kGcCauseNone),
329       thread_running_gc_(nullptr),
330       last_gc_type_(collector::kGcTypeNone),
331       next_gc_type_(collector::kGcTypePartial),
332       capacity_(capacity),
333       growth_limit_(growth_limit),
334       initial_heap_size_(initial_size),
335       target_footprint_(initial_size),
336       // Using kPostMonitorLock as a lock at kDefaultMutexLevel is acquired after
337       // this one.
338       process_state_update_lock_("process state update lock", kPostMonitorLock),
339       min_foreground_target_footprint_(0),
340       min_foreground_concurrent_start_bytes_(0),
341       concurrent_start_bytes_(std::numeric_limits<size_t>::max()),
342       total_bytes_freed_ever_(0),
343       total_objects_freed_ever_(0),
344       num_bytes_allocated_(0),
345       native_bytes_registered_(0),
346       old_native_bytes_allocated_(0),
347       native_objects_notified_(0),
348       num_bytes_freed_revoke_(0),
349       num_bytes_alive_after_gc_(0),
350       verify_missing_card_marks_(false),
351       verify_system_weaks_(false),
352       verify_pre_gc_heap_(verify_pre_gc_heap),
353       verify_pre_sweeping_heap_(verify_pre_sweeping_heap),
354       verify_post_gc_heap_(verify_post_gc_heap),
355       verify_mod_union_table_(false),
356       verify_pre_gc_rosalloc_(verify_pre_gc_rosalloc),
357       verify_pre_sweeping_rosalloc_(verify_pre_sweeping_rosalloc),
358       verify_post_gc_rosalloc_(verify_post_gc_rosalloc),
359       gc_stress_mode_(gc_stress_mode),
360       /* For GC a lot mode, we limit the allocation stacks to be kGcAlotInterval allocations. This
361        * causes a lot of GC since we do a GC for alloc whenever the stack is full. When heap
362        * verification is enabled, we limit the size of allocation stacks to speed up their
363        * searching.
364        */
365       max_allocation_stack_size_(kGCALotMode ? kGcAlotAllocationStackSize
366           : (kVerifyObjectSupport > kVerifyObjectModeFast) ? kVerifyObjectAllocationStackSize :
367           kDefaultAllocationStackSize),
368       current_allocator_(kAllocatorTypeDlMalloc),
369       current_non_moving_allocator_(kAllocatorTypeNonMoving),
370       bump_pointer_space_(nullptr),
371       temp_space_(nullptr),
372       region_space_(nullptr),
373       min_free_(min_free),
374       max_free_(max_free),
375       target_utilization_(target_utilization),
376       foreground_heap_growth_multiplier_(foreground_heap_growth_multiplier),
377       stop_for_native_allocs_(stop_for_native_allocs),
378       total_wait_time_(0),
379       verify_object_mode_(kVerifyObjectModeDisabled),
380       disable_moving_gc_count_(0),
381       semi_space_collector_(nullptr),
382       active_concurrent_copying_collector_(nullptr),
383       young_concurrent_copying_collector_(nullptr),
384       concurrent_copying_collector_(nullptr),
385       is_running_on_memory_tool_(Runtime::Current()->IsRunningOnMemoryTool()),
386       use_tlab_(use_tlab),
387       main_space_backup_(nullptr),
388       min_interval_homogeneous_space_compaction_by_oom_(
389           min_interval_homogeneous_space_compaction_by_oom),
390       last_time_homogeneous_space_compaction_by_oom_(NanoTime()),
391       gcs_completed_(0u),
392       max_gc_requested_(0u),
393       pending_collector_transition_(nullptr),
394       pending_heap_trim_(nullptr),
395       use_homogeneous_space_compaction_for_oom_(use_homogeneous_space_compaction_for_oom),
396       use_generational_cc_(use_generational_cc),
397       running_collection_is_blocking_(false),
398       blocking_gc_count_(0U),
399       blocking_gc_time_(0U),
400       last_update_time_gc_count_rate_histograms_(  // Round down by the window duration.
401           (NanoTime() / kGcCountRateHistogramWindowDuration) * kGcCountRateHistogramWindowDuration),
402       gc_count_last_window_(0U),
403       blocking_gc_count_last_window_(0U),
404       gc_count_rate_histogram_("gc count rate histogram", 1U, kGcCountRateMaxBucketCount),
405       blocking_gc_count_rate_histogram_("blocking gc count rate histogram", 1U,
406                                         kGcCountRateMaxBucketCount),
407       alloc_tracking_enabled_(false),
408       alloc_record_depth_(AllocRecordObjectMap::kDefaultAllocStackDepth),
409       backtrace_lock_(nullptr),
410       seen_backtrace_count_(0u),
411       unique_backtrace_count_(0u),
412       gc_disabled_for_shutdown_(false),
413       dump_region_info_before_gc_(dump_region_info_before_gc),
414       dump_region_info_after_gc_(dump_region_info_after_gc),
415       boot_image_spaces_(),
416       boot_images_start_address_(0u),
417       boot_images_size_(0u),
418       pre_oome_gc_count_(0u) {
419   if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
420     LOG(INFO) << "Heap() entering";
421   }
422 
423   LOG(INFO) << "Using " << foreground_collector_type_ << " GC.";
424   if (!gUseUserfaultfd) {
425     // This ensures that userfaultfd syscall is done before any seccomp filter is installed.
426     // TODO(b/266731037): Remove this when we no longer need to collect metric on userfaultfd
427     // support.
428     auto [uffd_supported, minor_fault_supported] = collector::MarkCompact::GetUffdAndMinorFault();
429     // The check is just to ensure that compiler doesn't eliminate the function call above.
430     // Userfaultfd support is certain to be there if its minor-fault feature is supported.
431     CHECK_IMPLIES(minor_fault_supported, uffd_supported);
432   }
433 
434   if (gUseReadBarrier) {
435     CHECK_EQ(foreground_collector_type_, kCollectorTypeCC);
436     CHECK_EQ(background_collector_type_, kCollectorTypeCCBackground);
437   } else if (background_collector_type_ != gc::kCollectorTypeHomogeneousSpaceCompact) {
438     CHECK_EQ(IsMovingGc(foreground_collector_type_), IsMovingGc(background_collector_type_))
439         << "Changing from " << foreground_collector_type_ << " to "
440         << background_collector_type_ << " (or visa versa) is not supported.";
441   }
442   verification_.reset(new Verification(this));
443   CHECK_GE(large_object_threshold, kMinLargeObjectThreshold);
444   ScopedTrace trace(__FUNCTION__);
445   Runtime* const runtime = Runtime::Current();
446   // If we aren't the zygote, switch to the default non zygote allocator. This may update the
447   // entrypoints.
448   const bool is_zygote = runtime->IsZygote();
449   if (!is_zygote) {
450     // Background compaction is currently not supported for command line runs.
451     if (background_collector_type_ != foreground_collector_type_) {
452       VLOG(heap) << "Disabling background compaction for non zygote";
453       background_collector_type_ = foreground_collector_type_;
454     }
455   }
456   ChangeCollector(desired_collector_type_);
457   live_bitmap_.reset(new accounting::HeapBitmap(this));
458   mark_bitmap_.reset(new accounting::HeapBitmap(this));
459 
460   // We don't have hspace compaction enabled with CC.
461   if (foreground_collector_type_ == kCollectorTypeCC
462       || foreground_collector_type_ == kCollectorTypeCMC) {
463     use_homogeneous_space_compaction_for_oom_ = false;
464   }
465   bool support_homogeneous_space_compaction =
466       background_collector_type_ == gc::kCollectorTypeHomogeneousSpaceCompact ||
467       use_homogeneous_space_compaction_for_oom_;
468   // We may use the same space the main space for the non moving space if we don't need to compact
469   // from the main space.
470   // This is not the case if we support homogeneous compaction or have a moving background
471   // collector type.
472   bool separate_non_moving_space = is_zygote ||
473       support_homogeneous_space_compaction || IsMovingGc(foreground_collector_type_) ||
474       IsMovingGc(background_collector_type_);
475 
476   // Requested begin for the alloc space, to follow the mapped image and oat files
477   uint8_t* request_begin = nullptr;
478   // Calculate the extra space required after the boot image, see allocations below.
479   size_t heap_reservation_size = 0u;
480   if (separate_non_moving_space) {
481     heap_reservation_size = non_moving_space_capacity;
482   } else if (foreground_collector_type_ != kCollectorTypeCC && is_zygote) {
483     heap_reservation_size = capacity_;
484   }
485   heap_reservation_size = RoundUp(heap_reservation_size, kPageSize);
486   // Load image space(s).
487   std::vector<std::unique_ptr<space::ImageSpace>> boot_image_spaces;
488   MemMap heap_reservation;
489   if (space::ImageSpace::LoadBootImage(boot_class_path,
490                                        boot_class_path_locations,
491                                        boot_class_path_fds,
492                                        boot_class_path_image_fds,
493                                        boot_class_path_vdex_fds,
494                                        boot_class_path_oat_fds,
495                                        image_file_names,
496                                        image_instruction_set,
497                                        runtime->ShouldRelocate(),
498                                        /*executable=*/ !runtime->IsAotCompiler(),
499                                        heap_reservation_size,
500                                        runtime->AllowInMemoryCompilation(),
501                                        &boot_image_spaces,
502                                        &heap_reservation)) {
503     DCHECK_EQ(heap_reservation_size, heap_reservation.IsValid() ? heap_reservation.Size() : 0u);
504     DCHECK(!boot_image_spaces.empty());
505     request_begin = boot_image_spaces.back()->GetImageHeader().GetOatFileEnd();
506     DCHECK_IMPLIES(heap_reservation.IsValid(), request_begin == heap_reservation.Begin())
507         << "request_begin=" << static_cast<const void*>(request_begin)
508         << " heap_reservation.Begin()=" << static_cast<const void*>(heap_reservation.Begin());
509     for (std::unique_ptr<space::ImageSpace>& space : boot_image_spaces) {
510       boot_image_spaces_.push_back(space.get());
511       AddSpace(space.release());
512     }
513     boot_images_start_address_ = PointerToLowMemUInt32(boot_image_spaces_.front()->Begin());
514     uint32_t boot_images_end =
515         PointerToLowMemUInt32(boot_image_spaces_.back()->GetImageHeader().GetOatFileEnd());
516     boot_images_size_ = boot_images_end - boot_images_start_address_;
517     if (kIsDebugBuild) {
518       VerifyBootImagesContiguity(boot_image_spaces_);
519     }
520   } else {
521     if (foreground_collector_type_ == kCollectorTypeCC) {
522       // Need to use a low address so that we can allocate a contiguous 2 * Xmx space
523       // when there's no image (dex2oat for target).
524       request_begin = kPreferredAllocSpaceBegin;
525     }
526     // Gross hack to make dex2oat deterministic.
527     if (foreground_collector_type_ == kCollectorTypeMS && Runtime::Current()->IsAotCompiler()) {
528       // Currently only enabled for MS collector since that is what the deterministic dex2oat uses.
529       // b/26849108
530       request_begin = reinterpret_cast<uint8_t*>(kAllocSpaceBeginForDeterministicAoT);
531     }
532   }
533 
534   /*
535   requested_alloc_space_begin ->     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
536                                      +-  nonmoving space (non_moving_space_capacity)+-
537                                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
538                                      +-????????????????????????????????????????????+-
539                                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
540                                      +-main alloc space / bump space 1 (capacity_) +-
541                                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
542                                      +-????????????????????????????????????????????+-
543                                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
544                                      +-main alloc space2 / bump space 2 (capacity_)+-
545                                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
546   */
547 
548   MemMap main_mem_map_1;
549   MemMap main_mem_map_2;
550 
551   std::string error_str;
552   MemMap non_moving_space_mem_map;
553   if (separate_non_moving_space) {
554     ScopedTrace trace2("Create separate non moving space");
555     // If we are the zygote, the non moving space becomes the zygote space when we run
556     // PreZygoteFork the first time. In this case, call the map "zygote space" since we can't
557     // rename the mem map later.
558     const char* space_name = is_zygote ? kZygoteSpaceName : kNonMovingSpaceName;
559     // Reserve the non moving mem map before the other two since it needs to be at a specific
560     // address.
561     DCHECK_EQ(heap_reservation.IsValid(), !boot_image_spaces_.empty());
562     if (heap_reservation.IsValid()) {
563       non_moving_space_mem_map = heap_reservation.RemapAtEnd(
564           heap_reservation.Begin(), space_name, PROT_READ | PROT_WRITE, &error_str);
565     } else {
566       non_moving_space_mem_map = MapAnonymousPreferredAddress(
567           space_name, request_begin, non_moving_space_capacity, &error_str);
568     }
569     CHECK(non_moving_space_mem_map.IsValid()) << error_str;
570     DCHECK(!heap_reservation.IsValid());
571     // Try to reserve virtual memory at a lower address if we have a separate non moving space.
572     request_begin = kPreferredAllocSpaceBegin + non_moving_space_capacity;
573   }
574   // Attempt to create 2 mem maps at or after the requested begin.
575   if (foreground_collector_type_ != kCollectorTypeCC) {
576     ScopedTrace trace2("Create main mem map");
577     if (separate_non_moving_space || !is_zygote) {
578       main_mem_map_1 = MapAnonymousPreferredAddress(
579           kMemMapSpaceName[0], request_begin, capacity_, &error_str);
580     } else {
581       // If no separate non-moving space and we are the zygote, the main space must come right after
582       // the image space to avoid a gap. This is required since we want the zygote space to be
583       // adjacent to the image space.
584       DCHECK_EQ(heap_reservation.IsValid(), !boot_image_spaces_.empty());
585       main_mem_map_1 = MemMap::MapAnonymous(
586           kMemMapSpaceName[0],
587           request_begin,
588           capacity_,
589           PROT_READ | PROT_WRITE,
590           /* low_4gb= */ true,
591           /* reuse= */ false,
592           heap_reservation.IsValid() ? &heap_reservation : nullptr,
593           &error_str);
594     }
595     CHECK(main_mem_map_1.IsValid()) << error_str;
596     DCHECK(!heap_reservation.IsValid());
597   }
598   if (support_homogeneous_space_compaction ||
599       background_collector_type_ == kCollectorTypeSS ||
600       foreground_collector_type_ == kCollectorTypeSS) {
601     ScopedTrace trace2("Create main mem map 2");
602     main_mem_map_2 = MapAnonymousPreferredAddress(
603         kMemMapSpaceName[1], main_mem_map_1.End(), capacity_, &error_str);
604     CHECK(main_mem_map_2.IsValid()) << error_str;
605   }
606 
607   // Create the non moving space first so that bitmaps don't take up the address range.
608   if (separate_non_moving_space) {
609     ScopedTrace trace2("Add non moving space");
610     // Non moving space is always dlmalloc since we currently don't have support for multiple
611     // active rosalloc spaces.
612     const size_t size = non_moving_space_mem_map.Size();
613     const void* non_moving_space_mem_map_begin = non_moving_space_mem_map.Begin();
614     non_moving_space_ = space::DlMallocSpace::CreateFromMemMap(std::move(non_moving_space_mem_map),
615                                                                "zygote / non moving space",
616                                                                kDefaultStartingSize,
617                                                                initial_size,
618                                                                size,
619                                                                size,
620                                                                /* can_move_objects= */ false);
621     CHECK(non_moving_space_ != nullptr) << "Failed creating non moving space "
622         << non_moving_space_mem_map_begin;
623     non_moving_space_->SetFootprintLimit(non_moving_space_->Capacity());
624     AddSpace(non_moving_space_);
625   }
626   // Create other spaces based on whether or not we have a moving GC.
627   if (foreground_collector_type_ == kCollectorTypeCC) {
628     CHECK(separate_non_moving_space);
629     // Reserve twice the capacity, to allow evacuating every region for explicit GCs.
630     MemMap region_space_mem_map =
631         space::RegionSpace::CreateMemMap(kRegionSpaceName, capacity_ * 2, request_begin);
632     CHECK(region_space_mem_map.IsValid()) << "No region space mem map";
633     region_space_ = space::RegionSpace::Create(
634         kRegionSpaceName, std::move(region_space_mem_map), use_generational_cc_);
635     AddSpace(region_space_);
636   } else if (IsMovingGc(foreground_collector_type_)) {
637     // Create bump pointer spaces.
638     // We only to create the bump pointer if the foreground collector is a compacting GC.
639     // TODO: Place bump-pointer spaces somewhere to minimize size of card table.
640     bump_pointer_space_ = space::BumpPointerSpace::CreateFromMemMap("Bump pointer space 1",
641                                                                     std::move(main_mem_map_1));
642     CHECK(bump_pointer_space_ != nullptr) << "Failed to create bump pointer space";
643     AddSpace(bump_pointer_space_);
644     // For Concurrent Mark-compact GC we don't need the temp space to be in
645     // lower 4GB. So its temp space will be created by the GC itself.
646     if (foreground_collector_type_ != kCollectorTypeCMC) {
647       temp_space_ = space::BumpPointerSpace::CreateFromMemMap("Bump pointer space 2",
648                                                               std::move(main_mem_map_2));
649       CHECK(temp_space_ != nullptr) << "Failed to create bump pointer space";
650       AddSpace(temp_space_);
651     }
652     CHECK(separate_non_moving_space);
653   } else {
654     CreateMainMallocSpace(std::move(main_mem_map_1), initial_size, growth_limit_, capacity_);
655     CHECK(main_space_ != nullptr);
656     AddSpace(main_space_);
657     if (!separate_non_moving_space) {
658       non_moving_space_ = main_space_;
659       CHECK(!non_moving_space_->CanMoveObjects());
660     }
661     if (main_mem_map_2.IsValid()) {
662       const char* name = kUseRosAlloc ? kRosAllocSpaceName[1] : kDlMallocSpaceName[1];
663       main_space_backup_.reset(CreateMallocSpaceFromMemMap(std::move(main_mem_map_2),
664                                                            initial_size,
665                                                            growth_limit_,
666                                                            capacity_,
667                                                            name,
668                                                            /* can_move_objects= */ true));
669       CHECK(main_space_backup_.get() != nullptr);
670       // Add the space so its accounted for in the heap_begin and heap_end.
671       AddSpace(main_space_backup_.get());
672     }
673   }
674   CHECK(non_moving_space_ != nullptr);
675   CHECK(!non_moving_space_->CanMoveObjects());
676   // Allocate the large object space.
677   if (large_object_space_type == space::LargeObjectSpaceType::kFreeList) {
678     large_object_space_ = space::FreeListSpace::Create("free list large object space", capacity_);
679     CHECK(large_object_space_ != nullptr) << "Failed to create large object space";
680   } else if (large_object_space_type == space::LargeObjectSpaceType::kMap) {
681     large_object_space_ = space::LargeObjectMapSpace::Create("mem map large object space");
682     CHECK(large_object_space_ != nullptr) << "Failed to create large object space";
683   } else {
684     // Disable the large object space by making the cutoff excessively large.
685     large_object_threshold_ = std::numeric_limits<size_t>::max();
686     large_object_space_ = nullptr;
687   }
688   if (large_object_space_ != nullptr) {
689     AddSpace(large_object_space_);
690   }
691   // Compute heap capacity. Continuous spaces are sorted in order of Begin().
692   CHECK(!continuous_spaces_.empty());
693   // Relies on the spaces being sorted.
694   uint8_t* heap_begin = continuous_spaces_.front()->Begin();
695   uint8_t* heap_end = continuous_spaces_.back()->Limit();
696   size_t heap_capacity = heap_end - heap_begin;
697   // Remove the main backup space since it slows down the GC to have unused extra spaces.
698   // TODO: Avoid needing to do this.
699   if (main_space_backup_.get() != nullptr) {
700     RemoveSpace(main_space_backup_.get());
701   }
702   // Allocate the card table.
703   // We currently don't support dynamically resizing the card table.
704   // Since we don't know where in the low_4gb the app image will be located, make the card table
705   // cover the whole low_4gb. TODO: Extend the card table in AddSpace.
706   UNUSED(heap_capacity);
707   // Start at 4 KB, we can be sure there are no spaces mapped this low since the address range is
708   // reserved by the kernel.
709   static constexpr size_t kMinHeapAddress = 4 * KB;
710   card_table_.reset(accounting::CardTable::Create(reinterpret_cast<uint8_t*>(kMinHeapAddress),
711                                                   4 * GB - kMinHeapAddress));
712   CHECK(card_table_.get() != nullptr) << "Failed to create card table";
713   if (foreground_collector_type_ == kCollectorTypeCC && kUseTableLookupReadBarrier) {
714     rb_table_.reset(new accounting::ReadBarrierTable());
715     DCHECK(rb_table_->IsAllCleared());
716   }
717   if (HasBootImageSpace()) {
718     // Don't add the image mod union table if we are running without an image, this can crash if
719     // we use the CardCache implementation.
720     for (space::ImageSpace* image_space : GetBootImageSpaces()) {
721       accounting::ModUnionTable* mod_union_table = new accounting::ModUnionTableToZygoteAllocspace(
722           "Image mod-union table", this, image_space);
723       CHECK(mod_union_table != nullptr) << "Failed to create image mod-union table";
724       AddModUnionTable(mod_union_table);
725     }
726   }
727   if (collector::SemiSpace::kUseRememberedSet && non_moving_space_ != main_space_) {
728     accounting::RememberedSet* non_moving_space_rem_set =
729         new accounting::RememberedSet("Non-moving space remembered set", this, non_moving_space_);
730     CHECK(non_moving_space_rem_set != nullptr) << "Failed to create non-moving space remembered set";
731     AddRememberedSet(non_moving_space_rem_set);
732   }
733   // TODO: Count objects in the image space here?
734   num_bytes_allocated_.store(0, std::memory_order_relaxed);
735   mark_stack_.reset(accounting::ObjectStack::Create("mark stack", kDefaultMarkStackSize,
736                                                     kDefaultMarkStackSize));
737   const size_t alloc_stack_capacity = max_allocation_stack_size_ + kAllocationStackReserveSize;
738   allocation_stack_.reset(accounting::ObjectStack::Create(
739       "allocation stack", max_allocation_stack_size_, alloc_stack_capacity));
740   live_stack_.reset(accounting::ObjectStack::Create(
741       "live stack", max_allocation_stack_size_, alloc_stack_capacity));
742   // It's still too early to take a lock because there are no threads yet, but we can create locks
743   // now. We don't create it earlier to make it clear that you can't use locks during heap
744   // initialization.
745   gc_complete_lock_ = new Mutex("GC complete lock");
746   gc_complete_cond_.reset(new ConditionVariable("GC complete condition variable",
747                                                 *gc_complete_lock_));
748 
749   thread_flip_lock_ = new Mutex("GC thread flip lock");
750   thread_flip_cond_.reset(new ConditionVariable("GC thread flip condition variable",
751                                                 *thread_flip_lock_));
752   task_processor_.reset(new TaskProcessor());
753   reference_processor_.reset(new ReferenceProcessor());
754   pending_task_lock_ = new Mutex("Pending task lock");
755   if (ignore_target_footprint_) {
756     SetIdealFootprint(std::numeric_limits<size_t>::max());
757     concurrent_start_bytes_ = std::numeric_limits<size_t>::max();
758   }
759   CHECK_NE(target_footprint_.load(std::memory_order_relaxed), 0U);
760   // Create our garbage collectors.
761   for (size_t i = 0; i < 2; ++i) {
762     const bool concurrent = i != 0;
763     if ((MayUseCollector(kCollectorTypeCMS) && concurrent) ||
764         (MayUseCollector(kCollectorTypeMS) && !concurrent)) {
765       garbage_collectors_.push_back(new collector::MarkSweep(this, concurrent));
766       garbage_collectors_.push_back(new collector::PartialMarkSweep(this, concurrent));
767       garbage_collectors_.push_back(new collector::StickyMarkSweep(this, concurrent));
768     }
769   }
770   if (kMovingCollector) {
771     if (MayUseCollector(kCollectorTypeSS) ||
772         MayUseCollector(kCollectorTypeHomogeneousSpaceCompact) ||
773         use_homogeneous_space_compaction_for_oom_) {
774       semi_space_collector_ = new collector::SemiSpace(this);
775       garbage_collectors_.push_back(semi_space_collector_);
776     }
777     if (MayUseCollector(kCollectorTypeCMC)) {
778       mark_compact_ = new collector::MarkCompact(this);
779       garbage_collectors_.push_back(mark_compact_);
780     }
781     if (MayUseCollector(kCollectorTypeCC)) {
782       concurrent_copying_collector_ = new collector::ConcurrentCopying(this,
783                                                                        /*young_gen=*/false,
784                                                                        use_generational_cc_,
785                                                                        "",
786                                                                        measure_gc_performance);
787       if (use_generational_cc_) {
788         young_concurrent_copying_collector_ = new collector::ConcurrentCopying(
789             this,
790             /*young_gen=*/true,
791             use_generational_cc_,
792             "young",
793             measure_gc_performance);
794       }
795       active_concurrent_copying_collector_.store(concurrent_copying_collector_,
796                                                  std::memory_order_relaxed);
797       DCHECK(region_space_ != nullptr);
798       concurrent_copying_collector_->SetRegionSpace(region_space_);
799       if (use_generational_cc_) {
800         young_concurrent_copying_collector_->SetRegionSpace(region_space_);
801         // At this point, non-moving space should be created.
802         DCHECK(non_moving_space_ != nullptr);
803         concurrent_copying_collector_->CreateInterRegionRefBitmaps();
804       }
805       garbage_collectors_.push_back(concurrent_copying_collector_);
806       if (use_generational_cc_) {
807         garbage_collectors_.push_back(young_concurrent_copying_collector_);
808       }
809     }
810   }
811   if (!GetBootImageSpaces().empty() && non_moving_space_ != nullptr &&
812       (is_zygote || separate_non_moving_space)) {
813     // Check that there's no gap between the image space and the non moving space so that the
814     // immune region won't break (eg. due to a large object allocated in the gap). This is only
815     // required when we're the zygote.
816     // Space with smallest Begin().
817     space::ImageSpace* first_space = nullptr;
818     for (space::ImageSpace* space : boot_image_spaces_) {
819       if (first_space == nullptr || space->Begin() < first_space->Begin()) {
820         first_space = space;
821       }
822     }
823     bool no_gap = MemMap::CheckNoGaps(*first_space->GetMemMap(), *non_moving_space_->GetMemMap());
824     if (!no_gap) {
825       PrintFileToLog("/proc/self/maps", LogSeverity::ERROR);
826       MemMap::DumpMaps(LOG_STREAM(ERROR), /* terse= */ true);
827       LOG(FATAL) << "There's a gap between the image space and the non-moving space";
828     }
829   }
830   // Perfetto Java Heap Profiler Support.
831   if (runtime->IsPerfettoJavaHeapStackProfEnabled()) {
832     // Perfetto Plugin is loaded and enabled, initialize the Java Heap Profiler.
833     InitPerfettoJavaHeapProf();
834   } else {
835     // Disable the Java Heap Profiler.
836     GetHeapSampler().DisableHeapSampler();
837   }
838 
839   instrumentation::Instrumentation* const instrumentation = runtime->GetInstrumentation();
840   if (gc_stress_mode_) {
841     backtrace_lock_ = new Mutex("GC complete lock");
842   }
843   if (is_running_on_memory_tool_ || gc_stress_mode_) {
844     instrumentation->InstrumentQuickAllocEntryPoints();
845   }
846   if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
847     LOG(INFO) << "Heap() exiting";
848   }
849 }
850 
MapAnonymousPreferredAddress(const char * name,uint8_t * request_begin,size_t capacity,std::string * out_error_str)851 MemMap Heap::MapAnonymousPreferredAddress(const char* name,
852                                           uint8_t* request_begin,
853                                           size_t capacity,
854                                           std::string* out_error_str) {
855   while (true) {
856     MemMap map = MemMap::MapAnonymous(name,
857                                       request_begin,
858                                       capacity,
859                                       PROT_READ | PROT_WRITE,
860                                       /*low_4gb=*/ true,
861                                       /*reuse=*/ false,
862                                       /*reservation=*/ nullptr,
863                                       out_error_str);
864     if (map.IsValid() || request_begin == nullptr) {
865       return map;
866     }
867     // Retry a  second time with no specified request begin.
868     request_begin = nullptr;
869   }
870 }
871 
MayUseCollector(CollectorType type) const872 bool Heap::MayUseCollector(CollectorType type) const {
873   return foreground_collector_type_ == type || background_collector_type_ == type;
874 }
875 
CreateMallocSpaceFromMemMap(MemMap && mem_map,size_t initial_size,size_t growth_limit,size_t capacity,const char * name,bool can_move_objects)876 space::MallocSpace* Heap::CreateMallocSpaceFromMemMap(MemMap&& mem_map,
877                                                       size_t initial_size,
878                                                       size_t growth_limit,
879                                                       size_t capacity,
880                                                       const char* name,
881                                                       bool can_move_objects) {
882   space::MallocSpace* malloc_space = nullptr;
883   if (kUseRosAlloc) {
884     // Create rosalloc space.
885     malloc_space = space::RosAllocSpace::CreateFromMemMap(std::move(mem_map),
886                                                           name,
887                                                           kDefaultStartingSize,
888                                                           initial_size,
889                                                           growth_limit,
890                                                           capacity,
891                                                           low_memory_mode_,
892                                                           can_move_objects);
893   } else {
894     malloc_space = space::DlMallocSpace::CreateFromMemMap(std::move(mem_map),
895                                                           name,
896                                                           kDefaultStartingSize,
897                                                           initial_size,
898                                                           growth_limit,
899                                                           capacity,
900                                                           can_move_objects);
901   }
902   if (collector::SemiSpace::kUseRememberedSet) {
903     accounting::RememberedSet* rem_set  =
904         new accounting::RememberedSet(std::string(name) + " remembered set", this, malloc_space);
905     CHECK(rem_set != nullptr) << "Failed to create main space remembered set";
906     AddRememberedSet(rem_set);
907   }
908   CHECK(malloc_space != nullptr) << "Failed to create " << name;
909   malloc_space->SetFootprintLimit(malloc_space->Capacity());
910   return malloc_space;
911 }
912 
CreateMainMallocSpace(MemMap && mem_map,size_t initial_size,size_t growth_limit,size_t capacity)913 void Heap::CreateMainMallocSpace(MemMap&& mem_map,
914                                  size_t initial_size,
915                                  size_t growth_limit,
916                                  size_t capacity) {
917   // Is background compaction is enabled?
918   bool can_move_objects = IsMovingGc(background_collector_type_) !=
919       IsMovingGc(foreground_collector_type_) || use_homogeneous_space_compaction_for_oom_;
920   // If we are the zygote and don't yet have a zygote space, it means that the zygote fork will
921   // happen in the future. If this happens and we have kCompactZygote enabled we wish to compact
922   // from the main space to the zygote space. If background compaction is enabled, always pass in
923   // that we can move objets.
924   if (kCompactZygote && Runtime::Current()->IsZygote() && !can_move_objects) {
925     // After the zygote we want this to be false if we don't have background compaction enabled so
926     // that getting primitive array elements is faster.
927     can_move_objects = !HasZygoteSpace();
928   }
929   if (collector::SemiSpace::kUseRememberedSet && main_space_ != nullptr) {
930     RemoveRememberedSet(main_space_);
931   }
932   const char* name = kUseRosAlloc ? kRosAllocSpaceName[0] : kDlMallocSpaceName[0];
933   main_space_ = CreateMallocSpaceFromMemMap(std::move(mem_map),
934                                             initial_size,
935                                             growth_limit,
936                                             capacity, name,
937                                             can_move_objects);
938   SetSpaceAsDefault(main_space_);
939   VLOG(heap) << "Created main space " << main_space_;
940 }
941 
ChangeAllocator(AllocatorType allocator)942 void Heap::ChangeAllocator(AllocatorType allocator) {
943   if (current_allocator_ != allocator) {
944     // These two allocators are only used internally and don't have any entrypoints.
945     CHECK_NE(allocator, kAllocatorTypeLOS);
946     CHECK_NE(allocator, kAllocatorTypeNonMoving);
947     current_allocator_ = allocator;
948     MutexLock mu(nullptr, *Locks::runtime_shutdown_lock_);
949     SetQuickAllocEntryPointsAllocator(current_allocator_);
950     Runtime::Current()->GetInstrumentation()->ResetQuickAllocEntryPoints();
951   }
952 }
953 
IsCompilingBoot() const954 bool Heap::IsCompilingBoot() const {
955   if (!Runtime::Current()->IsAotCompiler()) {
956     return false;
957   }
958   ScopedObjectAccess soa(Thread::Current());
959   for (const auto& space : continuous_spaces_) {
960     if (space->IsImageSpace() || space->IsZygoteSpace()) {
961       return false;
962     }
963   }
964   return true;
965 }
966 
IncrementDisableMovingGC(Thread * self)967 void Heap::IncrementDisableMovingGC(Thread* self) {
968   // Need to do this holding the lock to prevent races where the GC is about to run / running when
969   // we attempt to disable it.
970   ScopedThreadStateChange tsc(self, ThreadState::kWaitingForGcToComplete);
971   MutexLock mu(self, *gc_complete_lock_);
972   ++disable_moving_gc_count_;
973   if (IsMovingGc(collector_type_running_)) {
974     WaitForGcToCompleteLocked(kGcCauseDisableMovingGc, self);
975   }
976 }
977 
DecrementDisableMovingGC(Thread * self)978 void Heap::DecrementDisableMovingGC(Thread* self) {
979   MutexLock mu(self, *gc_complete_lock_);
980   CHECK_GT(disable_moving_gc_count_, 0U);
981   --disable_moving_gc_count_;
982 }
983 
IncrementDisableThreadFlip(Thread * self)984 void Heap::IncrementDisableThreadFlip(Thread* self) {
985   // Supposed to be called by mutators. If thread_flip_running_ is true, block. Otherwise, go ahead.
986   bool is_nested = self->GetDisableThreadFlipCount() > 0;
987   self->IncrementDisableThreadFlipCount();
988   if (is_nested) {
989     // If this is a nested JNI critical section enter, we don't need to wait or increment the global
990     // counter. The global counter is incremented only once for a thread for the outermost enter.
991     return;
992   }
993   ScopedThreadStateChange tsc(self, ThreadState::kWaitingForGcThreadFlip);
994   MutexLock mu(self, *thread_flip_lock_);
995   thread_flip_cond_->CheckSafeToWait(self);
996   bool has_waited = false;
997   uint64_t wait_start = 0;
998   if (thread_flip_running_) {
999     wait_start = NanoTime();
1000     ScopedTrace trace("IncrementDisableThreadFlip");
1001     while (thread_flip_running_) {
1002       has_waited = true;
1003       thread_flip_cond_->Wait(self);
1004     }
1005   }
1006   ++disable_thread_flip_count_;
1007   if (has_waited) {
1008     uint64_t wait_time = NanoTime() - wait_start;
1009     total_wait_time_ += wait_time;
1010     if (wait_time > long_pause_log_threshold_) {
1011       LOG(INFO) << __FUNCTION__ << " blocked for " << PrettyDuration(wait_time);
1012     }
1013   }
1014 }
1015 
EnsureObjectUserfaulted(ObjPtr<mirror::Object> obj)1016 void Heap::EnsureObjectUserfaulted(ObjPtr<mirror::Object> obj) {
1017   if (gUseUserfaultfd) {
1018     // Use volatile to ensure that compiler loads from memory to trigger userfaults, if required.
1019     const uint8_t* start = reinterpret_cast<uint8_t*>(obj.Ptr());
1020     const uint8_t* end = AlignUp(start + obj->SizeOf(), kPageSize);
1021     // The first page is already touched by SizeOf().
1022     start += kPageSize;
1023     while (start < end) {
1024       ForceRead(start);
1025       start += kPageSize;
1026     }
1027   }
1028 }
1029 
DecrementDisableThreadFlip(Thread * self)1030 void Heap::DecrementDisableThreadFlip(Thread* self) {
1031   // Supposed to be called by mutators. Decrement disable_thread_flip_count_ and potentially wake up
1032   // the GC waiting before doing a thread flip.
1033   self->DecrementDisableThreadFlipCount();
1034   bool is_outermost = self->GetDisableThreadFlipCount() == 0;
1035   if (!is_outermost) {
1036     // If this is not an outermost JNI critical exit, we don't need to decrement the global counter.
1037     // The global counter is decremented only once for a thread for the outermost exit.
1038     return;
1039   }
1040   MutexLock mu(self, *thread_flip_lock_);
1041   CHECK_GT(disable_thread_flip_count_, 0U);
1042   --disable_thread_flip_count_;
1043   if (disable_thread_flip_count_ == 0) {
1044     // Potentially notify the GC thread blocking to begin a thread flip.
1045     thread_flip_cond_->Broadcast(self);
1046   }
1047 }
1048 
ThreadFlipBegin(Thread * self)1049 void Heap::ThreadFlipBegin(Thread* self) {
1050   // Supposed to be called by GC. Set thread_flip_running_ to be true. If disable_thread_flip_count_
1051   // > 0, block. Otherwise, go ahead.
1052   ScopedThreadStateChange tsc(self, ThreadState::kWaitingForGcThreadFlip);
1053   MutexLock mu(self, *thread_flip_lock_);
1054   thread_flip_cond_->CheckSafeToWait(self);
1055   bool has_waited = false;
1056   uint64_t wait_start = NanoTime();
1057   CHECK(!thread_flip_running_);
1058   // Set this to true before waiting so that frequent JNI critical enter/exits won't starve
1059   // GC. This like a writer preference of a reader-writer lock.
1060   thread_flip_running_ = true;
1061   while (disable_thread_flip_count_ > 0) {
1062     has_waited = true;
1063     thread_flip_cond_->Wait(self);
1064   }
1065   if (has_waited) {
1066     uint64_t wait_time = NanoTime() - wait_start;
1067     total_wait_time_ += wait_time;
1068     if (wait_time > long_pause_log_threshold_) {
1069       LOG(INFO) << __FUNCTION__ << " blocked for " << PrettyDuration(wait_time);
1070     }
1071   }
1072 }
1073 
ThreadFlipEnd(Thread * self)1074 void Heap::ThreadFlipEnd(Thread* self) {
1075   // Supposed to be called by GC. Set thread_flip_running_ to false and potentially wake up mutators
1076   // waiting before doing a JNI critical.
1077   MutexLock mu(self, *thread_flip_lock_);
1078   CHECK(thread_flip_running_);
1079   thread_flip_running_ = false;
1080   // Potentially notify mutator threads blocking to enter a JNI critical section.
1081   thread_flip_cond_->Broadcast(self);
1082 }
1083 
GrowHeapOnJankPerceptibleSwitch()1084 void Heap::GrowHeapOnJankPerceptibleSwitch() {
1085   MutexLock mu(Thread::Current(), process_state_update_lock_);
1086   size_t orig_target_footprint = target_footprint_.load(std::memory_order_relaxed);
1087   if (orig_target_footprint < min_foreground_target_footprint_) {
1088     target_footprint_.compare_exchange_strong(orig_target_footprint,
1089                                               min_foreground_target_footprint_,
1090                                               std::memory_order_relaxed);
1091   }
1092   if (IsGcConcurrent() && concurrent_start_bytes_ < min_foreground_concurrent_start_bytes_) {
1093     concurrent_start_bytes_ = min_foreground_concurrent_start_bytes_;
1094   }
1095 }
1096 
UpdateProcessState(ProcessState old_process_state,ProcessState new_process_state)1097 void Heap::UpdateProcessState(ProcessState old_process_state, ProcessState new_process_state) {
1098   if (old_process_state != new_process_state) {
1099     const bool jank_perceptible = new_process_state == kProcessStateJankPerceptible;
1100     if (jank_perceptible) {
1101       // Transition back to foreground right away to prevent jank.
1102       RequestCollectorTransition(foreground_collector_type_, 0);
1103       GrowHeapOnJankPerceptibleSwitch();
1104     } else {
1105       // If background_collector_type_ is kCollectorTypeHomogeneousSpaceCompact then we have
1106       // special handling which does a homogenous space compaction once but then doesn't transition
1107       // the collector. Similarly, we invoke a full compaction for kCollectorTypeCC but don't
1108       // transition the collector.
1109       RequestCollectorTransition(background_collector_type_, 0);
1110     }
1111   }
1112 }
1113 
CreateThreadPool(size_t num_threads)1114 void Heap::CreateThreadPool(size_t num_threads) {
1115   if (num_threads == 0) {
1116     num_threads = std::max(parallel_gc_threads_, conc_gc_threads_);
1117   }
1118   if (num_threads != 0) {
1119     thread_pool_.reset(new ThreadPool("Heap thread pool", num_threads));
1120   }
1121 }
1122 
WaitForWorkersToBeCreated()1123 void Heap::WaitForWorkersToBeCreated() {
1124   DCHECK(!Runtime::Current()->IsShuttingDown(Thread::Current()))
1125       << "Cannot create new threads during runtime shutdown";
1126   if (thread_pool_ != nullptr) {
1127     thread_pool_->WaitForWorkersToBeCreated();
1128   }
1129 }
1130 
MarkAllocStackAsLive(accounting::ObjectStack * stack)1131 void Heap::MarkAllocStackAsLive(accounting::ObjectStack* stack) {
1132   space::ContinuousSpace* space1 = main_space_ != nullptr ? main_space_ : non_moving_space_;
1133   space::ContinuousSpace* space2 = non_moving_space_;
1134   // TODO: Generalize this to n bitmaps?
1135   CHECK(space1 != nullptr);
1136   CHECK(space2 != nullptr);
1137   MarkAllocStack(space1->GetLiveBitmap(), space2->GetLiveBitmap(),
1138                  (large_object_space_ != nullptr ? large_object_space_->GetLiveBitmap() : nullptr),
1139                  stack);
1140 }
1141 
DeleteThreadPool()1142 void Heap::DeleteThreadPool() {
1143   thread_pool_.reset(nullptr);
1144 }
1145 
AddSpace(space::Space * space)1146 void Heap::AddSpace(space::Space* space) {
1147   CHECK(space != nullptr);
1148   WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
1149   if (space->IsContinuousSpace()) {
1150     DCHECK(!space->IsDiscontinuousSpace());
1151     space::ContinuousSpace* continuous_space = space->AsContinuousSpace();
1152     // Continuous spaces don't necessarily have bitmaps.
1153     accounting::ContinuousSpaceBitmap* live_bitmap = continuous_space->GetLiveBitmap();
1154     accounting::ContinuousSpaceBitmap* mark_bitmap = continuous_space->GetMarkBitmap();
1155     // The region space bitmap is not added since VisitObjects visits the region space objects with
1156     // special handling.
1157     if (live_bitmap != nullptr && !space->IsRegionSpace()) {
1158       CHECK(mark_bitmap != nullptr);
1159       live_bitmap_->AddContinuousSpaceBitmap(live_bitmap);
1160       mark_bitmap_->AddContinuousSpaceBitmap(mark_bitmap);
1161     }
1162     continuous_spaces_.push_back(continuous_space);
1163     // Ensure that spaces remain sorted in increasing order of start address.
1164     std::sort(continuous_spaces_.begin(), continuous_spaces_.end(),
1165               [](const space::ContinuousSpace* a, const space::ContinuousSpace* b) {
1166       return a->Begin() < b->Begin();
1167     });
1168   } else {
1169     CHECK(space->IsDiscontinuousSpace());
1170     space::DiscontinuousSpace* discontinuous_space = space->AsDiscontinuousSpace();
1171     live_bitmap_->AddLargeObjectBitmap(discontinuous_space->GetLiveBitmap());
1172     mark_bitmap_->AddLargeObjectBitmap(discontinuous_space->GetMarkBitmap());
1173     discontinuous_spaces_.push_back(discontinuous_space);
1174   }
1175   if (space->IsAllocSpace()) {
1176     alloc_spaces_.push_back(space->AsAllocSpace());
1177   }
1178 }
1179 
SetSpaceAsDefault(space::ContinuousSpace * continuous_space)1180 void Heap::SetSpaceAsDefault(space::ContinuousSpace* continuous_space) {
1181   WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
1182   if (continuous_space->IsDlMallocSpace()) {
1183     dlmalloc_space_ = continuous_space->AsDlMallocSpace();
1184   } else if (continuous_space->IsRosAllocSpace()) {
1185     rosalloc_space_ = continuous_space->AsRosAllocSpace();
1186   }
1187 }
1188 
RemoveSpace(space::Space * space)1189 void Heap::RemoveSpace(space::Space* space) {
1190   DCHECK(space != nullptr);
1191   WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
1192   if (space->IsContinuousSpace()) {
1193     DCHECK(!space->IsDiscontinuousSpace());
1194     space::ContinuousSpace* continuous_space = space->AsContinuousSpace();
1195     // Continuous spaces don't necessarily have bitmaps.
1196     accounting::ContinuousSpaceBitmap* live_bitmap = continuous_space->GetLiveBitmap();
1197     accounting::ContinuousSpaceBitmap* mark_bitmap = continuous_space->GetMarkBitmap();
1198     if (live_bitmap != nullptr && !space->IsRegionSpace()) {
1199       DCHECK(mark_bitmap != nullptr);
1200       live_bitmap_->RemoveContinuousSpaceBitmap(live_bitmap);
1201       mark_bitmap_->RemoveContinuousSpaceBitmap(mark_bitmap);
1202     }
1203     auto it = std::find(continuous_spaces_.begin(), continuous_spaces_.end(), continuous_space);
1204     DCHECK(it != continuous_spaces_.end());
1205     continuous_spaces_.erase(it);
1206   } else {
1207     DCHECK(space->IsDiscontinuousSpace());
1208     space::DiscontinuousSpace* discontinuous_space = space->AsDiscontinuousSpace();
1209     live_bitmap_->RemoveLargeObjectBitmap(discontinuous_space->GetLiveBitmap());
1210     mark_bitmap_->RemoveLargeObjectBitmap(discontinuous_space->GetMarkBitmap());
1211     auto it = std::find(discontinuous_spaces_.begin(), discontinuous_spaces_.end(),
1212                         discontinuous_space);
1213     DCHECK(it != discontinuous_spaces_.end());
1214     discontinuous_spaces_.erase(it);
1215   }
1216   if (space->IsAllocSpace()) {
1217     auto it = std::find(alloc_spaces_.begin(), alloc_spaces_.end(), space->AsAllocSpace());
1218     DCHECK(it != alloc_spaces_.end());
1219     alloc_spaces_.erase(it);
1220   }
1221 }
1222 
CalculateGcWeightedAllocatedBytes(uint64_t gc_last_process_cpu_time_ns,uint64_t current_process_cpu_time) const1223 double Heap::CalculateGcWeightedAllocatedBytes(uint64_t gc_last_process_cpu_time_ns,
1224                                                uint64_t current_process_cpu_time) const {
1225   uint64_t bytes_allocated = GetBytesAllocated();
1226   double weight = current_process_cpu_time - gc_last_process_cpu_time_ns;
1227   return weight * bytes_allocated;
1228 }
1229 
CalculatePreGcWeightedAllocatedBytes()1230 void Heap::CalculatePreGcWeightedAllocatedBytes() {
1231   uint64_t current_process_cpu_time = ProcessCpuNanoTime();
1232   pre_gc_weighted_allocated_bytes_ +=
1233     CalculateGcWeightedAllocatedBytes(pre_gc_last_process_cpu_time_ns_, current_process_cpu_time);
1234   pre_gc_last_process_cpu_time_ns_ = current_process_cpu_time;
1235 }
1236 
CalculatePostGcWeightedAllocatedBytes()1237 void Heap::CalculatePostGcWeightedAllocatedBytes() {
1238   uint64_t current_process_cpu_time = ProcessCpuNanoTime();
1239   post_gc_weighted_allocated_bytes_ +=
1240     CalculateGcWeightedAllocatedBytes(post_gc_last_process_cpu_time_ns_, current_process_cpu_time);
1241   post_gc_last_process_cpu_time_ns_ = current_process_cpu_time;
1242 }
1243 
GetTotalGcCpuTime()1244 uint64_t Heap::GetTotalGcCpuTime() {
1245   uint64_t sum = 0;
1246   for (auto* collector : garbage_collectors_) {
1247     sum += collector->GetTotalCpuTime();
1248   }
1249   return sum;
1250 }
1251 
DumpGcPerformanceInfo(std::ostream & os)1252 void Heap::DumpGcPerformanceInfo(std::ostream& os) {
1253   // Dump cumulative timings.
1254   os << "Dumping cumulative Gc timings\n";
1255   uint64_t total_duration = 0;
1256   // Dump cumulative loggers for each GC type.
1257   uint64_t total_paused_time = 0;
1258   for (auto* collector : garbage_collectors_) {
1259     total_duration += collector->GetCumulativeTimings().GetTotalNs();
1260     total_paused_time += collector->GetTotalPausedTimeNs();
1261     collector->DumpPerformanceInfo(os);
1262   }
1263   if (total_duration != 0) {
1264     const double total_seconds = total_duration / 1.0e9;
1265     const double total_cpu_seconds = GetTotalGcCpuTime() / 1.0e9;
1266     os << "Total time spent in GC: " << PrettyDuration(total_duration) << "\n";
1267     os << "Mean GC size throughput: "
1268        << PrettySize(GetBytesFreedEver() / total_seconds) << "/s"
1269        << " per cpu-time: "
1270        << PrettySize(GetBytesFreedEver() / total_cpu_seconds) << "/s\n";
1271     os << "Mean GC object throughput: "
1272        << (GetObjectsFreedEver() / total_seconds) << " objects/s\n";
1273   }
1274   uint64_t total_objects_allocated = GetObjectsAllocatedEver();
1275   os << "Total number of allocations " << total_objects_allocated << "\n";
1276   os << "Total bytes allocated " << PrettySize(GetBytesAllocatedEver()) << "\n";
1277   os << "Total bytes freed " << PrettySize(GetBytesFreedEver()) << "\n";
1278   os << "Free memory " << PrettySize(GetFreeMemory()) << "\n";
1279   os << "Free memory until GC " << PrettySize(GetFreeMemoryUntilGC()) << "\n";
1280   os << "Free memory until OOME " << PrettySize(GetFreeMemoryUntilOOME()) << "\n";
1281   os << "Total memory " << PrettySize(GetTotalMemory()) << "\n";
1282   os << "Max memory " << PrettySize(GetMaxMemory()) << "\n";
1283   if (HasZygoteSpace()) {
1284     os << "Zygote space size " << PrettySize(zygote_space_->Size()) << "\n";
1285   }
1286   os << "Total mutator paused time: " << PrettyDuration(total_paused_time) << "\n";
1287   os << "Total time waiting for GC to complete: " << PrettyDuration(total_wait_time_) << "\n";
1288   os << "Total GC count: " << GetGcCount() << "\n";
1289   os << "Total GC time: " << PrettyDuration(GetGcTime()) << "\n";
1290   os << "Total blocking GC count: " << GetBlockingGcCount() << "\n";
1291   os << "Total blocking GC time: " << PrettyDuration(GetBlockingGcTime()) << "\n";
1292   os << "Total pre-OOME GC count: " << GetPreOomeGcCount() << "\n";
1293   {
1294     MutexLock mu(Thread::Current(), *gc_complete_lock_);
1295     if (gc_count_rate_histogram_.SampleSize() > 0U) {
1296       os << "Histogram of GC count per " << NsToMs(kGcCountRateHistogramWindowDuration) << " ms: ";
1297       gc_count_rate_histogram_.DumpBins(os);
1298       os << "\n";
1299     }
1300     if (blocking_gc_count_rate_histogram_.SampleSize() > 0U) {
1301       os << "Histogram of blocking GC count per "
1302          << NsToMs(kGcCountRateHistogramWindowDuration) << " ms: ";
1303       blocking_gc_count_rate_histogram_.DumpBins(os);
1304       os << "\n";
1305     }
1306   }
1307 
1308   if (kDumpRosAllocStatsOnSigQuit && rosalloc_space_ != nullptr) {
1309     rosalloc_space_->DumpStats(os);
1310   }
1311 
1312   os << "Native bytes total: " << GetNativeBytes()
1313      << " registered: " << native_bytes_registered_.load(std::memory_order_relaxed) << "\n";
1314 
1315   os << "Total native bytes at last GC: "
1316      << old_native_bytes_allocated_.load(std::memory_order_relaxed) << "\n";
1317 
1318   BaseMutex::DumpAll(os);
1319 }
1320 
ResetGcPerformanceInfo()1321 void Heap::ResetGcPerformanceInfo() {
1322   for (auto* collector : garbage_collectors_) {
1323     collector->ResetMeasurements();
1324   }
1325 
1326   process_cpu_start_time_ns_ = ProcessCpuNanoTime();
1327 
1328   pre_gc_last_process_cpu_time_ns_ = process_cpu_start_time_ns_;
1329   pre_gc_weighted_allocated_bytes_ = 0u;
1330 
1331   post_gc_last_process_cpu_time_ns_ = process_cpu_start_time_ns_;
1332   post_gc_weighted_allocated_bytes_ = 0u;
1333 
1334   total_bytes_freed_ever_.store(0);
1335   total_objects_freed_ever_.store(0);
1336   total_wait_time_ = 0;
1337   blocking_gc_count_ = 0;
1338   blocking_gc_time_ = 0;
1339   pre_oome_gc_count_.store(0, std::memory_order_relaxed);
1340   gc_count_last_window_ = 0;
1341   blocking_gc_count_last_window_ = 0;
1342   last_update_time_gc_count_rate_histograms_ =  // Round down by the window duration.
1343       (NanoTime() / kGcCountRateHistogramWindowDuration) * kGcCountRateHistogramWindowDuration;
1344   {
1345     MutexLock mu(Thread::Current(), *gc_complete_lock_);
1346     gc_count_rate_histogram_.Reset();
1347     blocking_gc_count_rate_histogram_.Reset();
1348   }
1349 }
1350 
GetGcCount() const1351 uint64_t Heap::GetGcCount() const {
1352   uint64_t gc_count = 0U;
1353   for (auto* collector : garbage_collectors_) {
1354     gc_count += collector->GetCumulativeTimings().GetIterations();
1355   }
1356   return gc_count;
1357 }
1358 
GetGcTime() const1359 uint64_t Heap::GetGcTime() const {
1360   uint64_t gc_time = 0U;
1361   for (auto* collector : garbage_collectors_) {
1362     gc_time += collector->GetCumulativeTimings().GetTotalNs();
1363   }
1364   return gc_time;
1365 }
1366 
GetBlockingGcCount() const1367 uint64_t Heap::GetBlockingGcCount() const {
1368   return blocking_gc_count_;
1369 }
1370 
GetBlockingGcTime() const1371 uint64_t Heap::GetBlockingGcTime() const {
1372   return blocking_gc_time_;
1373 }
1374 
DumpGcCountRateHistogram(std::ostream & os) const1375 void Heap::DumpGcCountRateHistogram(std::ostream& os) const {
1376   MutexLock mu(Thread::Current(), *gc_complete_lock_);
1377   if (gc_count_rate_histogram_.SampleSize() > 0U) {
1378     gc_count_rate_histogram_.DumpBins(os);
1379   }
1380 }
1381 
DumpBlockingGcCountRateHistogram(std::ostream & os) const1382 void Heap::DumpBlockingGcCountRateHistogram(std::ostream& os) const {
1383   MutexLock mu(Thread::Current(), *gc_complete_lock_);
1384   if (blocking_gc_count_rate_histogram_.SampleSize() > 0U) {
1385     blocking_gc_count_rate_histogram_.DumpBins(os);
1386   }
1387 }
1388 
GetPreOomeGcCount() const1389 uint64_t Heap::GetPreOomeGcCount() const {
1390   return pre_oome_gc_count_.load(std::memory_order_relaxed);
1391 }
1392 
1393 ALWAYS_INLINE
GetAndOverwriteAllocationListener(Atomic<AllocationListener * > * storage,AllocationListener * new_value)1394 static inline AllocationListener* GetAndOverwriteAllocationListener(
1395     Atomic<AllocationListener*>* storage, AllocationListener* new_value) {
1396   return storage->exchange(new_value);
1397 }
1398 
~Heap()1399 Heap::~Heap() {
1400   VLOG(heap) << "Starting ~Heap()";
1401   STLDeleteElements(&garbage_collectors_);
1402   // If we don't reset then the mark stack complains in its destructor.
1403   allocation_stack_->Reset();
1404   allocation_records_.reset();
1405   live_stack_->Reset();
1406   STLDeleteValues(&mod_union_tables_);
1407   STLDeleteValues(&remembered_sets_);
1408   STLDeleteElements(&continuous_spaces_);
1409   STLDeleteElements(&discontinuous_spaces_);
1410   delete gc_complete_lock_;
1411   delete thread_flip_lock_;
1412   delete pending_task_lock_;
1413   delete backtrace_lock_;
1414   uint64_t unique_count = unique_backtrace_count_.load();
1415   uint64_t seen_count = seen_backtrace_count_.load();
1416   if (unique_count != 0 || seen_count != 0) {
1417     LOG(INFO) << "gc stress unique=" << unique_count << " total=" << (unique_count + seen_count);
1418   }
1419   VLOG(heap) << "Finished ~Heap()";
1420 }
1421 
1422 
FindContinuousSpaceFromAddress(const mirror::Object * addr) const1423 space::ContinuousSpace* Heap::FindContinuousSpaceFromAddress(const mirror::Object* addr) const {
1424   for (const auto& space : continuous_spaces_) {
1425     if (space->Contains(addr)) {
1426       return space;
1427     }
1428   }
1429   return nullptr;
1430 }
1431 
FindContinuousSpaceFromObject(ObjPtr<mirror::Object> obj,bool fail_ok) const1432 space::ContinuousSpace* Heap::FindContinuousSpaceFromObject(ObjPtr<mirror::Object> obj,
1433                                                             bool fail_ok) const {
1434   space::ContinuousSpace* space = FindContinuousSpaceFromAddress(obj.Ptr());
1435   if (space != nullptr) {
1436     return space;
1437   }
1438   if (!fail_ok) {
1439     LOG(FATAL) << "object " << obj << " not inside any spaces!";
1440   }
1441   return nullptr;
1442 }
1443 
FindDiscontinuousSpaceFromObject(ObjPtr<mirror::Object> obj,bool fail_ok) const1444 space::DiscontinuousSpace* Heap::FindDiscontinuousSpaceFromObject(ObjPtr<mirror::Object> obj,
1445                                                                   bool fail_ok) const {
1446   for (const auto& space : discontinuous_spaces_) {
1447     if (space->Contains(obj.Ptr())) {
1448       return space;
1449     }
1450   }
1451   if (!fail_ok) {
1452     LOG(FATAL) << "object " << obj << " not inside any spaces!";
1453   }
1454   return nullptr;
1455 }
1456 
FindSpaceFromObject(ObjPtr<mirror::Object> obj,bool fail_ok) const1457 space::Space* Heap::FindSpaceFromObject(ObjPtr<mirror::Object> obj, bool fail_ok) const {
1458   space::Space* result = FindContinuousSpaceFromObject(obj, true);
1459   if (result != nullptr) {
1460     return result;
1461   }
1462   return FindDiscontinuousSpaceFromObject(obj, fail_ok);
1463 }
1464 
FindSpaceFromAddress(const void * addr) const1465 space::Space* Heap::FindSpaceFromAddress(const void* addr) const {
1466   for (const auto& space : continuous_spaces_) {
1467     if (space->Contains(reinterpret_cast<const mirror::Object*>(addr))) {
1468       return space;
1469     }
1470   }
1471   for (const auto& space : discontinuous_spaces_) {
1472     if (space->Contains(reinterpret_cast<const mirror::Object*>(addr))) {
1473       return space;
1474     }
1475   }
1476   return nullptr;
1477 }
1478 
DumpSpaceNameFromAddress(const void * addr) const1479 std::string Heap::DumpSpaceNameFromAddress(const void* addr) const {
1480   space::Space* space = FindSpaceFromAddress(addr);
1481   return (space != nullptr) ? space->GetName() : "no space";
1482 }
1483 
ThrowOutOfMemoryError(Thread * self,size_t byte_count,AllocatorType allocator_type)1484 void Heap::ThrowOutOfMemoryError(Thread* self, size_t byte_count, AllocatorType allocator_type) {
1485   // If we're in a stack overflow, do not create a new exception. It would require running the
1486   // constructor, which will of course still be in a stack overflow.
1487   if (self->IsHandlingStackOverflow()) {
1488     self->SetException(
1489         Runtime::Current()->GetPreAllocatedOutOfMemoryErrorWhenHandlingStackOverflow());
1490     return;
1491   }
1492   // Allow plugins to intercept out of memory errors.
1493   Runtime::Current()->OutOfMemoryErrorHook();
1494 
1495   std::ostringstream oss;
1496   size_t total_bytes_free = GetFreeMemory();
1497   oss << "Failed to allocate a " << byte_count << " byte allocation with " << total_bytes_free
1498       << " free bytes and " << PrettySize(GetFreeMemoryUntilOOME()) << " until OOM,"
1499       << " target footprint " << target_footprint_.load(std::memory_order_relaxed)
1500       << ", growth limit "
1501       << growth_limit_;
1502   // If the allocation failed due to fragmentation, print out the largest continuous allocation.
1503   if (total_bytes_free >= byte_count) {
1504     space::AllocSpace* space = nullptr;
1505     if (allocator_type == kAllocatorTypeNonMoving) {
1506       space = non_moving_space_;
1507     } else if (allocator_type == kAllocatorTypeRosAlloc ||
1508                allocator_type == kAllocatorTypeDlMalloc) {
1509       space = main_space_;
1510     } else if (allocator_type == kAllocatorTypeBumpPointer ||
1511                allocator_type == kAllocatorTypeTLAB) {
1512       space = bump_pointer_space_;
1513     } else if (allocator_type == kAllocatorTypeRegion ||
1514                allocator_type == kAllocatorTypeRegionTLAB) {
1515       space = region_space_;
1516     }
1517 
1518     // There is no fragmentation info to log for large-object space.
1519     if (allocator_type != kAllocatorTypeLOS) {
1520       CHECK(space != nullptr) << "allocator_type:" << allocator_type
1521                               << " byte_count:" << byte_count
1522                               << " total_bytes_free:" << total_bytes_free;
1523       // LogFragmentationAllocFailure returns true if byte_count is greater than
1524       // the largest free contiguous chunk in the space. Return value false
1525       // means that we are throwing OOME because the amount of free heap after
1526       // GC is less than kMinFreeHeapAfterGcForAlloc in proportion of the heap-size.
1527       // Log an appropriate message in that case.
1528       if (!space->LogFragmentationAllocFailure(oss, byte_count)) {
1529         oss << "; giving up on allocation because <"
1530             << kMinFreeHeapAfterGcForAlloc * 100
1531             << "% of heap free after GC.";
1532       }
1533     }
1534   }
1535   self->ThrowOutOfMemoryError(oss.str().c_str());
1536 }
1537 
DoPendingCollectorTransition()1538 void Heap::DoPendingCollectorTransition() {
1539   CollectorType desired_collector_type = desired_collector_type_;
1540 
1541   if (collector_type_ == kCollectorTypeCC || collector_type_ == kCollectorTypeCMC) {
1542     // App's allocations (since last GC) more than the threshold then do TransitionGC
1543     // when the app was in background. If not then don't do TransitionGC.
1544     // num_bytes_allocated_since_gc should always be positive even if initially
1545     // num_bytes_alive_after_gc_ is coming from Zygote. This gives positive or zero value.
1546     size_t num_bytes_allocated_since_gc =
1547         UnsignedDifference(GetBytesAllocated(), num_bytes_alive_after_gc_);
1548     if (num_bytes_allocated_since_gc <
1549         (UnsignedDifference(target_footprint_.load(std::memory_order_relaxed),
1550                             num_bytes_alive_after_gc_)/4)
1551         && !kStressCollectorTransition
1552         && !IsLowMemoryMode()) {
1553       return;
1554     }
1555   }
1556 
1557   // Launch homogeneous space compaction if it is desired.
1558   if (desired_collector_type == kCollectorTypeHomogeneousSpaceCompact) {
1559     if (!CareAboutPauseTimes()) {
1560       PerformHomogeneousSpaceCompact();
1561     } else {
1562       VLOG(gc) << "Homogeneous compaction ignored due to jank perceptible process state";
1563     }
1564   } else if (desired_collector_type == kCollectorTypeCCBackground ||
1565              desired_collector_type == kCollectorTypeCMC) {
1566     if (!CareAboutPauseTimes()) {
1567       // Invoke full compaction.
1568       CollectGarbageInternal(collector::kGcTypeFull,
1569                              kGcCauseCollectorTransition,
1570                              /*clear_soft_references=*/false, GetCurrentGcNum() + 1);
1571     } else {
1572       VLOG(gc) << "background compaction ignored due to jank perceptible process state";
1573     }
1574   } else {
1575     CHECK_EQ(desired_collector_type, collector_type_) << "Unsupported collector transition";
1576   }
1577 }
1578 
Trim(Thread * self)1579 void Heap::Trim(Thread* self) {
1580   Runtime* const runtime = Runtime::Current();
1581   if (!CareAboutPauseTimes()) {
1582     // Deflate the monitors, this can cause a pause but shouldn't matter since we don't care
1583     // about pauses.
1584     ScopedTrace trace("Deflating monitors");
1585     // Avoid race conditions on the lock word for CC.
1586     ScopedGCCriticalSection gcs(self, kGcCauseTrim, kCollectorTypeHeapTrim);
1587     ScopedSuspendAll ssa(__FUNCTION__);
1588     uint64_t start_time = NanoTime();
1589     size_t count = runtime->GetMonitorList()->DeflateMonitors();
1590     VLOG(heap) << "Deflating " << count << " monitors took "
1591         << PrettyDuration(NanoTime() - start_time);
1592   }
1593   TrimIndirectReferenceTables(self);
1594   TrimSpaces(self);
1595   // Trim arenas that may have been used by JIT or verifier.
1596   runtime->GetArenaPool()->TrimMaps();
1597 }
1598 
1599 class TrimIndirectReferenceTableClosure : public Closure {
1600  public:
TrimIndirectReferenceTableClosure(Barrier * barrier)1601   explicit TrimIndirectReferenceTableClosure(Barrier* barrier) : barrier_(barrier) {
1602   }
Run(Thread * thread)1603   void Run(Thread* thread) override NO_THREAD_SAFETY_ANALYSIS {
1604     thread->GetJniEnv()->TrimLocals();
1605     // If thread is a running mutator, then act on behalf of the trim thread.
1606     // See the code in ThreadList::RunCheckpoint.
1607     barrier_->Pass(Thread::Current());
1608   }
1609 
1610  private:
1611   Barrier* const barrier_;
1612 };
1613 
TrimIndirectReferenceTables(Thread * self)1614 void Heap::TrimIndirectReferenceTables(Thread* self) {
1615   ScopedObjectAccess soa(self);
1616   ScopedTrace trace(__PRETTY_FUNCTION__);
1617   JavaVMExt* vm = soa.Vm();
1618   // Trim globals indirect reference table.
1619   vm->TrimGlobals();
1620   // Trim locals indirect reference tables.
1621   // TODO: May also want to look for entirely empty pages maintained by SmallIrtAllocator.
1622   Barrier barrier(0);
1623   TrimIndirectReferenceTableClosure closure(&barrier);
1624   ScopedThreadStateChange tsc(self, ThreadState::kWaitingForCheckPointsToRun);
1625   size_t barrier_count = Runtime::Current()->GetThreadList()->RunCheckpoint(&closure);
1626   if (barrier_count != 0) {
1627     barrier.Increment(self, barrier_count);
1628   }
1629 }
1630 
StartGC(Thread * self,GcCause cause,CollectorType collector_type)1631 void Heap::StartGC(Thread* self, GcCause cause, CollectorType collector_type) {
1632   // Need to do this before acquiring the locks since we don't want to get suspended while
1633   // holding any locks.
1634   ScopedThreadStateChange tsc(self, ThreadState::kWaitingForGcToComplete);
1635   MutexLock mu(self, *gc_complete_lock_);
1636   // Ensure there is only one GC at a time.
1637   WaitForGcToCompleteLocked(cause, self);
1638   collector_type_running_ = collector_type;
1639   last_gc_cause_ = cause;
1640   thread_running_gc_ = self;
1641 }
1642 
TrimSpaces(Thread * self)1643 void Heap::TrimSpaces(Thread* self) {
1644   // Pretend we are doing a GC to prevent background compaction from deleting the space we are
1645   // trimming.
1646   StartGC(self, kGcCauseTrim, kCollectorTypeHeapTrim);
1647   ScopedTrace trace(__PRETTY_FUNCTION__);
1648   const uint64_t start_ns = NanoTime();
1649   // Trim the managed spaces.
1650   uint64_t total_alloc_space_allocated = 0;
1651   uint64_t total_alloc_space_size = 0;
1652   uint64_t managed_reclaimed = 0;
1653   {
1654     ScopedObjectAccess soa(self);
1655     for (const auto& space : continuous_spaces_) {
1656       if (space->IsMallocSpace()) {
1657         gc::space::MallocSpace* malloc_space = space->AsMallocSpace();
1658         if (malloc_space->IsRosAllocSpace() || !CareAboutPauseTimes()) {
1659           // Don't trim dlmalloc spaces if we care about pauses since this can hold the space lock
1660           // for a long period of time.
1661           managed_reclaimed += malloc_space->Trim();
1662         }
1663         total_alloc_space_size += malloc_space->Size();
1664       }
1665     }
1666   }
1667   total_alloc_space_allocated = GetBytesAllocated();
1668   if (large_object_space_ != nullptr) {
1669     total_alloc_space_allocated -= large_object_space_->GetBytesAllocated();
1670   }
1671   if (bump_pointer_space_ != nullptr) {
1672     total_alloc_space_allocated -= bump_pointer_space_->Size();
1673   }
1674   if (region_space_ != nullptr) {
1675     total_alloc_space_allocated -= region_space_->GetBytesAllocated();
1676   }
1677   const float managed_utilization = static_cast<float>(total_alloc_space_allocated) /
1678       static_cast<float>(total_alloc_space_size);
1679   uint64_t gc_heap_end_ns = NanoTime();
1680   // We never move things in the native heap, so we can finish the GC at this point.
1681   FinishGC(self, collector::kGcTypeNone);
1682 
1683   VLOG(heap) << "Heap trim of managed (duration=" << PrettyDuration(gc_heap_end_ns - start_ns)
1684       << ", advised=" << PrettySize(managed_reclaimed) << ") heap. Managed heap utilization of "
1685       << static_cast<int>(100 * managed_utilization) << "%.";
1686 }
1687 
IsValidObjectAddress(const void * addr) const1688 bool Heap::IsValidObjectAddress(const void* addr) const {
1689   if (addr == nullptr) {
1690     return true;
1691   }
1692   return IsAligned<kObjectAlignment>(addr) && FindSpaceFromAddress(addr) != nullptr;
1693 }
1694 
IsNonDiscontinuousSpaceHeapAddress(const void * addr) const1695 bool Heap::IsNonDiscontinuousSpaceHeapAddress(const void* addr) const {
1696   return FindContinuousSpaceFromAddress(reinterpret_cast<const mirror::Object*>(addr)) != nullptr;
1697 }
1698 
IsLiveObjectLocked(ObjPtr<mirror::Object> obj,bool search_allocation_stack,bool search_live_stack,bool sorted)1699 bool Heap::IsLiveObjectLocked(ObjPtr<mirror::Object> obj,
1700                               bool search_allocation_stack,
1701                               bool search_live_stack,
1702                               bool sorted) {
1703   if (UNLIKELY(!IsAligned<kObjectAlignment>(obj.Ptr()))) {
1704     return false;
1705   }
1706   if (bump_pointer_space_ != nullptr && bump_pointer_space_->HasAddress(obj.Ptr())) {
1707     mirror::Class* klass = obj->GetClass<kVerifyNone>();
1708     if (obj == klass) {
1709       // This case happens for java.lang.Class.
1710       return true;
1711     }
1712     return VerifyClassClass(klass) && IsLiveObjectLocked(klass);
1713   } else if (temp_space_ != nullptr && temp_space_->HasAddress(obj.Ptr())) {
1714     // If we are in the allocated region of the temp space, then we are probably live (e.g. during
1715     // a GC). When a GC isn't running End() - Begin() is 0 which means no objects are contained.
1716     return temp_space_->Contains(obj.Ptr());
1717   }
1718   if (region_space_ != nullptr && region_space_->HasAddress(obj.Ptr())) {
1719     return true;
1720   }
1721   space::ContinuousSpace* c_space = FindContinuousSpaceFromObject(obj, true);
1722   space::DiscontinuousSpace* d_space = nullptr;
1723   if (c_space != nullptr) {
1724     if (c_space->GetLiveBitmap()->Test(obj.Ptr())) {
1725       return true;
1726     }
1727   } else {
1728     d_space = FindDiscontinuousSpaceFromObject(obj, true);
1729     if (d_space != nullptr) {
1730       if (d_space->GetLiveBitmap()->Test(obj.Ptr())) {
1731         return true;
1732       }
1733     }
1734   }
1735   // This is covering the allocation/live stack swapping that is done without mutators suspended.
1736   for (size_t i = 0; i < (sorted ? 1 : 5); ++i) {
1737     if (i > 0) {
1738       NanoSleep(MsToNs(10));
1739     }
1740     if (search_allocation_stack) {
1741       if (sorted) {
1742         if (allocation_stack_->ContainsSorted(obj.Ptr())) {
1743           return true;
1744         }
1745       } else if (allocation_stack_->Contains(obj.Ptr())) {
1746         return true;
1747       }
1748     }
1749 
1750     if (search_live_stack) {
1751       if (sorted) {
1752         if (live_stack_->ContainsSorted(obj.Ptr())) {
1753           return true;
1754         }
1755       } else if (live_stack_->Contains(obj.Ptr())) {
1756         return true;
1757       }
1758     }
1759   }
1760   // We need to check the bitmaps again since there is a race where we mark something as live and
1761   // then clear the stack containing it.
1762   if (c_space != nullptr) {
1763     if (c_space->GetLiveBitmap()->Test(obj.Ptr())) {
1764       return true;
1765     }
1766   } else {
1767     d_space = FindDiscontinuousSpaceFromObject(obj, true);
1768     if (d_space != nullptr && d_space->GetLiveBitmap()->Test(obj.Ptr())) {
1769       return true;
1770     }
1771   }
1772   return false;
1773 }
1774 
DumpSpaces() const1775 std::string Heap::DumpSpaces() const {
1776   std::ostringstream oss;
1777   DumpSpaces(oss);
1778   return oss.str();
1779 }
1780 
DumpSpaces(std::ostream & stream) const1781 void Heap::DumpSpaces(std::ostream& stream) const {
1782   for (const auto& space : continuous_spaces_) {
1783     accounting::ContinuousSpaceBitmap* live_bitmap = space->GetLiveBitmap();
1784     accounting::ContinuousSpaceBitmap* mark_bitmap = space->GetMarkBitmap();
1785     stream << space << " " << *space << "\n";
1786     if (live_bitmap != nullptr) {
1787       stream << live_bitmap << " " << *live_bitmap << "\n";
1788     }
1789     if (mark_bitmap != nullptr) {
1790       stream << mark_bitmap << " " << *mark_bitmap << "\n";
1791     }
1792   }
1793   for (const auto& space : discontinuous_spaces_) {
1794     stream << space << " " << *space << "\n";
1795   }
1796 }
1797 
VerifyObjectBody(ObjPtr<mirror::Object> obj)1798 void Heap::VerifyObjectBody(ObjPtr<mirror::Object> obj) {
1799   if (verify_object_mode_ == kVerifyObjectModeDisabled) {
1800     return;
1801   }
1802 
1803   // Ignore early dawn of the universe verifications.
1804   if (UNLIKELY(num_bytes_allocated_.load(std::memory_order_relaxed) < 10 * KB)) {
1805     return;
1806   }
1807   CHECK_ALIGNED(obj.Ptr(), kObjectAlignment) << "Object isn't aligned";
1808   mirror::Class* c = obj->GetFieldObject<mirror::Class, kVerifyNone>(mirror::Object::ClassOffset());
1809   CHECK(c != nullptr) << "Null class in object " << obj;
1810   CHECK_ALIGNED(c, kObjectAlignment) << "Class " << c << " not aligned in object " << obj;
1811   CHECK(VerifyClassClass(c));
1812 
1813   if (verify_object_mode_ > kVerifyObjectModeFast) {
1814     // Note: the bitmap tests below are racy since we don't hold the heap bitmap lock.
1815     CHECK(IsLiveObjectLocked(obj)) << "Object is dead " << obj << "\n" << DumpSpaces();
1816   }
1817 }
1818 
VerifyHeap()1819 void Heap::VerifyHeap() {
1820   ReaderMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
1821   auto visitor = [&](mirror::Object* obj) NO_THREAD_SAFETY_ANALYSIS {
1822     VerifyObjectBody(obj);
1823   };
1824   // Technically we need the mutator lock here to call Visit. However, VerifyObjectBody is already
1825   // NO_THREAD_SAFETY_ANALYSIS.
1826   auto no_thread_safety_analysis = [&]() NO_THREAD_SAFETY_ANALYSIS {
1827     GetLiveBitmap()->Visit(visitor);
1828   };
1829   no_thread_safety_analysis();
1830 }
1831 
RecordFree(uint64_t freed_objects,int64_t freed_bytes)1832 void Heap::RecordFree(uint64_t freed_objects, int64_t freed_bytes) {
1833   // Use signed comparison since freed bytes can be negative when background compaction foreground
1834   // transitions occurs. This is typically due to objects moving from a bump pointer space to a
1835   // free list backed space, which may increase memory footprint due to padding and binning.
1836   RACING_DCHECK_LE(freed_bytes,
1837                    static_cast<int64_t>(num_bytes_allocated_.load(std::memory_order_relaxed)));
1838   // Note: This relies on 2s complement for handling negative freed_bytes.
1839   num_bytes_allocated_.fetch_sub(static_cast<ssize_t>(freed_bytes), std::memory_order_relaxed);
1840   if (Runtime::Current()->HasStatsEnabled()) {
1841     RuntimeStats* thread_stats = Thread::Current()->GetStats();
1842     thread_stats->freed_objects += freed_objects;
1843     thread_stats->freed_bytes += freed_bytes;
1844     // TODO: Do this concurrently.
1845     RuntimeStats* global_stats = Runtime::Current()->GetStats();
1846     global_stats->freed_objects += freed_objects;
1847     global_stats->freed_bytes += freed_bytes;
1848   }
1849 }
1850 
RecordFreeRevoke()1851 void Heap::RecordFreeRevoke() {
1852   // Subtract num_bytes_freed_revoke_ from num_bytes_allocated_ to cancel out the
1853   // ahead-of-time, bulk counting of bytes allocated in rosalloc thread-local buffers.
1854   // If there's a concurrent revoke, ok to not necessarily reset num_bytes_freed_revoke_
1855   // all the way to zero exactly as the remainder will be subtracted at the next GC.
1856   size_t bytes_freed = num_bytes_freed_revoke_.load(std::memory_order_relaxed);
1857   CHECK_GE(num_bytes_freed_revoke_.fetch_sub(bytes_freed, std::memory_order_relaxed),
1858            bytes_freed) << "num_bytes_freed_revoke_ underflow";
1859   CHECK_GE(num_bytes_allocated_.fetch_sub(bytes_freed, std::memory_order_relaxed),
1860            bytes_freed) << "num_bytes_allocated_ underflow";
1861   GetCurrentGcIteration()->SetFreedRevoke(bytes_freed);
1862 }
1863 
GetRosAllocSpace(gc::allocator::RosAlloc * rosalloc) const1864 space::RosAllocSpace* Heap::GetRosAllocSpace(gc::allocator::RosAlloc* rosalloc) const {
1865   if (rosalloc_space_ != nullptr && rosalloc_space_->GetRosAlloc() == rosalloc) {
1866     return rosalloc_space_;
1867   }
1868   for (const auto& space : continuous_spaces_) {
1869     if (space->AsContinuousSpace()->IsRosAllocSpace()) {
1870       if (space->AsContinuousSpace()->AsRosAllocSpace()->GetRosAlloc() == rosalloc) {
1871         return space->AsContinuousSpace()->AsRosAllocSpace();
1872       }
1873     }
1874   }
1875   return nullptr;
1876 }
1877 
EntrypointsInstrumented()1878 static inline bool EntrypointsInstrumented() REQUIRES_SHARED(Locks::mutator_lock_) {
1879   instrumentation::Instrumentation* const instrumentation =
1880       Runtime::Current()->GetInstrumentation();
1881   return instrumentation != nullptr && instrumentation->AllocEntrypointsInstrumented();
1882 }
1883 
AllocateInternalWithGc(Thread * self,AllocatorType allocator,bool instrumented,size_t alloc_size,size_t * bytes_allocated,size_t * usable_size,size_t * bytes_tl_bulk_allocated,ObjPtr<mirror::Class> * klass)1884 mirror::Object* Heap::AllocateInternalWithGc(Thread* self,
1885                                              AllocatorType allocator,
1886                                              bool instrumented,
1887                                              size_t alloc_size,
1888                                              size_t* bytes_allocated,
1889                                              size_t* usable_size,
1890                                              size_t* bytes_tl_bulk_allocated,
1891                                              ObjPtr<mirror::Class>* klass) {
1892   bool was_default_allocator = allocator == GetCurrentAllocator();
1893   // Make sure there is no pending exception since we may need to throw an OOME.
1894   self->AssertNoPendingException();
1895   DCHECK(klass != nullptr);
1896 
1897   StackHandleScope<1> hs(self);
1898   HandleWrapperObjPtr<mirror::Class> h_klass(hs.NewHandleWrapper(klass));
1899 
1900   auto send_object_pre_alloc =
1901       [&]() REQUIRES_SHARED(Locks::mutator_lock_) REQUIRES(!Roles::uninterruptible_) {
1902         if (UNLIKELY(instrumented)) {
1903           AllocationListener* l = alloc_listener_.load(std::memory_order_seq_cst);
1904           if (UNLIKELY(l != nullptr) && UNLIKELY(l->HasPreAlloc())) {
1905             l->PreObjectAllocated(self, h_klass, &alloc_size);
1906           }
1907         }
1908       };
1909 #define PERFORM_SUSPENDING_OPERATION(op)                                          \
1910   [&]() REQUIRES(Roles::uninterruptible_) REQUIRES_SHARED(Locks::mutator_lock_) { \
1911     ScopedAllowThreadSuspension ats;                                              \
1912     auto res = (op);                                                              \
1913     send_object_pre_alloc();                                                      \
1914     return res;                                                                   \
1915   }()
1916 
1917   // The allocation failed. If the GC is running, block until it completes, and then retry the
1918   // allocation.
1919   collector::GcType last_gc =
1920       PERFORM_SUSPENDING_OPERATION(WaitForGcToComplete(kGcCauseForAlloc, self));
1921   // If we were the default allocator but the allocator changed while we were suspended,
1922   // abort the allocation.
1923   if ((was_default_allocator && allocator != GetCurrentAllocator()) ||
1924       (!instrumented && EntrypointsInstrumented())) {
1925     return nullptr;
1926   }
1927   uint32_t starting_gc_num = GetCurrentGcNum();
1928   if (last_gc != collector::kGcTypeNone) {
1929     // A GC was in progress and we blocked, retry allocation now that memory has been freed.
1930     mirror::Object* ptr = TryToAllocate<true, false>(self, allocator, alloc_size, bytes_allocated,
1931                                                      usable_size, bytes_tl_bulk_allocated);
1932     if (ptr != nullptr) {
1933       return ptr;
1934     }
1935   }
1936 
1937   auto have_reclaimed_enough = [&]() {
1938     size_t curr_bytes_allocated = GetBytesAllocated();
1939     double curr_free_heap =
1940         static_cast<double>(growth_limit_ - curr_bytes_allocated) / growth_limit_;
1941     return curr_free_heap >= kMinFreeHeapAfterGcForAlloc;
1942   };
1943   // We perform one GC as per the next_gc_type_ (chosen in GrowForUtilization),
1944   // if it's not already tried. If that doesn't succeed then go for the most
1945   // exhaustive option. Perform a full-heap collection including clearing
1946   // SoftReferences. In case of ConcurrentCopying, it will also ensure that
1947   // all regions are evacuated. If allocation doesn't succeed even after that
1948   // then there is no hope, so we throw OOME.
1949   collector::GcType tried_type = next_gc_type_;
1950   if (last_gc < tried_type) {
1951     const bool gc_ran = PERFORM_SUSPENDING_OPERATION(
1952         CollectGarbageInternal(tried_type, kGcCauseForAlloc, false, starting_gc_num + 1)
1953         != collector::kGcTypeNone);
1954 
1955     if ((was_default_allocator && allocator != GetCurrentAllocator()) ||
1956         (!instrumented && EntrypointsInstrumented())) {
1957       return nullptr;
1958     }
1959     if (gc_ran && have_reclaimed_enough()) {
1960       mirror::Object* ptr = TryToAllocate<true, false>(self, allocator,
1961                                                        alloc_size, bytes_allocated,
1962                                                        usable_size, bytes_tl_bulk_allocated);
1963       if (ptr != nullptr) {
1964         return ptr;
1965       }
1966     }
1967   }
1968   // Most allocations should have succeeded by now, so the heap is really full, really fragmented,
1969   // or the requested size is really big. Do another GC, collecting SoftReferences this time. The
1970   // VM spec requires that all SoftReferences have been collected and cleared before throwing
1971   // OOME.
1972   VLOG(gc) << "Forcing collection of SoftReferences for " << PrettySize(alloc_size)
1973            << " allocation";
1974   // TODO: Run finalization, but this may cause more allocations to occur.
1975   // We don't need a WaitForGcToComplete here either.
1976   // TODO: Should check whether another thread already just ran a GC with soft
1977   // references.
1978   DCHECK(!gc_plan_.empty());
1979   pre_oome_gc_count_.fetch_add(1, std::memory_order_relaxed);
1980   PERFORM_SUSPENDING_OPERATION(
1981       CollectGarbageInternal(gc_plan_.back(), kGcCauseForAlloc, true, GC_NUM_ANY));
1982   if ((was_default_allocator && allocator != GetCurrentAllocator()) ||
1983       (!instrumented && EntrypointsInstrumented())) {
1984     return nullptr;
1985   }
1986   mirror::Object* ptr = nullptr;
1987   if (have_reclaimed_enough()) {
1988     ptr = TryToAllocate<true, true>(self, allocator, alloc_size, bytes_allocated,
1989                                     usable_size, bytes_tl_bulk_allocated);
1990   }
1991 
1992   if (ptr == nullptr) {
1993     const uint64_t current_time = NanoTime();
1994     switch (allocator) {
1995       case kAllocatorTypeRosAlloc:
1996         // Fall-through.
1997       case kAllocatorTypeDlMalloc: {
1998         if (use_homogeneous_space_compaction_for_oom_ &&
1999             current_time - last_time_homogeneous_space_compaction_by_oom_ >
2000             min_interval_homogeneous_space_compaction_by_oom_) {
2001           last_time_homogeneous_space_compaction_by_oom_ = current_time;
2002           HomogeneousSpaceCompactResult result =
2003               PERFORM_SUSPENDING_OPERATION(PerformHomogeneousSpaceCompact());
2004           // Thread suspension could have occurred.
2005           if ((was_default_allocator && allocator != GetCurrentAllocator()) ||
2006               (!instrumented && EntrypointsInstrumented())) {
2007             return nullptr;
2008           }
2009           switch (result) {
2010             case HomogeneousSpaceCompactResult::kSuccess:
2011               // If the allocation succeeded, we delayed an oom.
2012               ptr = TryToAllocate<true, true>(self, allocator, alloc_size, bytes_allocated,
2013                                               usable_size, bytes_tl_bulk_allocated);
2014               if (ptr != nullptr) {
2015                 count_delayed_oom_++;
2016               }
2017               break;
2018             case HomogeneousSpaceCompactResult::kErrorReject:
2019               // Reject due to disabled moving GC.
2020               break;
2021             case HomogeneousSpaceCompactResult::kErrorVMShuttingDown:
2022               // Throw OOM by default.
2023               break;
2024             default: {
2025               UNIMPLEMENTED(FATAL) << "homogeneous space compaction result: "
2026                   << static_cast<size_t>(result);
2027               UNREACHABLE();
2028             }
2029           }
2030           // Always print that we ran homogeneous space compation since this can cause jank.
2031           VLOG(heap) << "Ran heap homogeneous space compaction, "
2032                     << " requested defragmentation "
2033                     << count_requested_homogeneous_space_compaction_.load()
2034                     << " performed defragmentation "
2035                     << count_performed_homogeneous_space_compaction_.load()
2036                     << " ignored homogeneous space compaction "
2037                     << count_ignored_homogeneous_space_compaction_.load()
2038                     << " delayed count = "
2039                     << count_delayed_oom_.load();
2040         }
2041         break;
2042       }
2043       default: {
2044         // Do nothing for others allocators.
2045       }
2046     }
2047   }
2048 #undef PERFORM_SUSPENDING_OPERATION
2049   // If the allocation hasn't succeeded by this point, throw an OOM error.
2050   if (ptr == nullptr) {
2051     ScopedAllowThreadSuspension ats;
2052     ThrowOutOfMemoryError(self, alloc_size, allocator);
2053   }
2054   return ptr;
2055 }
2056 
SetTargetHeapUtilization(float target)2057 void Heap::SetTargetHeapUtilization(float target) {
2058   DCHECK_GT(target, 0.1f);  // asserted in Java code
2059   DCHECK_LT(target, 1.0f);
2060   target_utilization_ = target;
2061 }
2062 
GetObjectsAllocated() const2063 size_t Heap::GetObjectsAllocated() const {
2064   Thread* const self = Thread::Current();
2065   ScopedThreadStateChange tsc(self, ThreadState::kWaitingForGetObjectsAllocated);
2066   // Prevent GC running during GetObjectsAllocated since we may get a checkpoint request that tells
2067   // us to suspend while we are doing SuspendAll. b/35232978
2068   gc::ScopedGCCriticalSection gcs(Thread::Current(),
2069                                   gc::kGcCauseGetObjectsAllocated,
2070                                   gc::kCollectorTypeGetObjectsAllocated);
2071   // Need SuspendAll here to prevent lock violation if RosAlloc does it during InspectAll.
2072   ScopedSuspendAll ssa(__FUNCTION__);
2073   ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
2074   size_t total = 0;
2075   for (space::AllocSpace* space : alloc_spaces_) {
2076     total += space->GetObjectsAllocated();
2077   }
2078   return total;
2079 }
2080 
GetObjectsAllocatedEver() const2081 uint64_t Heap::GetObjectsAllocatedEver() const {
2082   uint64_t total = GetObjectsFreedEver();
2083   // If we are detached, we can't use GetObjectsAllocated since we can't change thread states.
2084   if (Thread::Current() != nullptr) {
2085     total += GetObjectsAllocated();
2086   }
2087   return total;
2088 }
2089 
GetBytesAllocatedEver() const2090 uint64_t Heap::GetBytesAllocatedEver() const {
2091   // Force the returned value to be monotonically increasing, in the sense that if this is called
2092   // at A and B, such that A happens-before B, then the call at B returns a value no smaller than
2093   // that at A. This is not otherwise guaranteed, since num_bytes_allocated_ is decremented first,
2094   // and total_bytes_freed_ever_ is incremented later.
2095   static std::atomic<uint64_t> max_bytes_so_far(0);
2096   uint64_t so_far = max_bytes_so_far.load(std::memory_order_relaxed);
2097   uint64_t current_bytes = GetBytesFreedEver(std::memory_order_acquire);
2098   current_bytes += GetBytesAllocated();
2099   do {
2100     if (current_bytes <= so_far) {
2101       return so_far;
2102     }
2103   } while (!max_bytes_so_far.compare_exchange_weak(so_far /* updated */,
2104                                                    current_bytes, std::memory_order_relaxed));
2105   return current_bytes;
2106 }
2107 
2108 // Check whether the given object is an instance of the given class.
MatchesClass(mirror::Object * obj,Handle<mirror::Class> h_class,bool use_is_assignable_from)2109 static bool MatchesClass(mirror::Object* obj,
2110                          Handle<mirror::Class> h_class,
2111                          bool use_is_assignable_from) REQUIRES_SHARED(Locks::mutator_lock_) {
2112   mirror::Class* instance_class = obj->GetClass();
2113   CHECK(instance_class != nullptr);
2114   ObjPtr<mirror::Class> klass = h_class.Get();
2115   if (use_is_assignable_from) {
2116     return klass != nullptr && klass->IsAssignableFrom(instance_class);
2117   }
2118   return instance_class == klass;
2119 }
2120 
CountInstances(const std::vector<Handle<mirror::Class>> & classes,bool use_is_assignable_from,uint64_t * counts)2121 void Heap::CountInstances(const std::vector<Handle<mirror::Class>>& classes,
2122                           bool use_is_assignable_from,
2123                           uint64_t* counts) {
2124   auto instance_counter = [&](mirror::Object* obj) REQUIRES_SHARED(Locks::mutator_lock_) {
2125     for (size_t i = 0; i < classes.size(); ++i) {
2126       if (MatchesClass(obj, classes[i], use_is_assignable_from)) {
2127         ++counts[i];
2128       }
2129     }
2130   };
2131   VisitObjects(instance_counter);
2132 }
2133 
CollectGarbage(bool clear_soft_references,GcCause cause)2134 void Heap::CollectGarbage(bool clear_soft_references, GcCause cause) {
2135   // Even if we waited for a GC we still need to do another GC since weaks allocated during the
2136   // last GC will not have necessarily been cleared.
2137   CollectGarbageInternal(gc_plan_.back(), cause, clear_soft_references, GC_NUM_ANY);
2138 }
2139 
SupportHomogeneousSpaceCompactAndCollectorTransitions() const2140 bool Heap::SupportHomogeneousSpaceCompactAndCollectorTransitions() const {
2141   return main_space_backup_.get() != nullptr && main_space_ != nullptr &&
2142       foreground_collector_type_ == kCollectorTypeCMS;
2143 }
2144 
PerformHomogeneousSpaceCompact()2145 HomogeneousSpaceCompactResult Heap::PerformHomogeneousSpaceCompact() {
2146   Thread* self = Thread::Current();
2147   // Inc requested homogeneous space compaction.
2148   count_requested_homogeneous_space_compaction_++;
2149   // Store performed homogeneous space compaction at a new request arrival.
2150   ScopedThreadStateChange tsc(self, ThreadState::kWaitingPerformingGc);
2151   Locks::mutator_lock_->AssertNotHeld(self);
2152   {
2153     ScopedThreadStateChange tsc2(self, ThreadState::kWaitingForGcToComplete);
2154     MutexLock mu(self, *gc_complete_lock_);
2155     // Ensure there is only one GC at a time.
2156     WaitForGcToCompleteLocked(kGcCauseHomogeneousSpaceCompact, self);
2157     // Homogeneous space compaction is a copying transition, can't run it if the moving GC disable
2158     // count is non zero.
2159     // If the collector type changed to something which doesn't benefit from homogeneous space
2160     // compaction, exit.
2161     if (disable_moving_gc_count_ != 0 || IsMovingGc(collector_type_) ||
2162         !main_space_->CanMoveObjects()) {
2163       return kErrorReject;
2164     }
2165     if (!SupportHomogeneousSpaceCompactAndCollectorTransitions()) {
2166       return kErrorUnsupported;
2167     }
2168     collector_type_running_ = kCollectorTypeHomogeneousSpaceCompact;
2169   }
2170   if (Runtime::Current()->IsShuttingDown(self)) {
2171     // Don't allow heap transitions to happen if the runtime is shutting down since these can
2172     // cause objects to get finalized.
2173     FinishGC(self, collector::kGcTypeNone);
2174     return HomogeneousSpaceCompactResult::kErrorVMShuttingDown;
2175   }
2176   collector::GarbageCollector* collector;
2177   {
2178     ScopedSuspendAll ssa(__FUNCTION__);
2179     uint64_t start_time = NanoTime();
2180     // Launch compaction.
2181     space::MallocSpace* to_space = main_space_backup_.release();
2182     space::MallocSpace* from_space = main_space_;
2183     to_space->GetMemMap()->Protect(PROT_READ | PROT_WRITE);
2184     const uint64_t space_size_before_compaction = from_space->Size();
2185     AddSpace(to_space);
2186     // Make sure that we will have enough room to copy.
2187     CHECK_GE(to_space->GetFootprintLimit(), from_space->GetFootprintLimit());
2188     collector = Compact(to_space, from_space, kGcCauseHomogeneousSpaceCompact);
2189     const uint64_t space_size_after_compaction = to_space->Size();
2190     main_space_ = to_space;
2191     main_space_backup_.reset(from_space);
2192     RemoveSpace(from_space);
2193     SetSpaceAsDefault(main_space_);  // Set as default to reset the proper dlmalloc space.
2194     // Update performed homogeneous space compaction count.
2195     count_performed_homogeneous_space_compaction_++;
2196     // Print statics log and resume all threads.
2197     uint64_t duration = NanoTime() - start_time;
2198     VLOG(heap) << "Heap homogeneous space compaction took " << PrettyDuration(duration) << " size: "
2199                << PrettySize(space_size_before_compaction) << " -> "
2200                << PrettySize(space_size_after_compaction) << " compact-ratio: "
2201                << std::fixed << static_cast<double>(space_size_after_compaction) /
2202                static_cast<double>(space_size_before_compaction);
2203   }
2204   // Finish GC.
2205   // Get the references we need to enqueue.
2206   SelfDeletingTask* clear = reference_processor_->CollectClearedReferences(self);
2207   GrowForUtilization(semi_space_collector_);
2208   LogGC(kGcCauseHomogeneousSpaceCompact, collector);
2209   FinishGC(self, collector::kGcTypeFull);
2210   // Enqueue any references after losing the GC locks.
2211   clear->Run(self);
2212   clear->Finalize();
2213   {
2214     ScopedObjectAccess soa(self);
2215     soa.Vm()->UnloadNativeLibraries();
2216   }
2217   return HomogeneousSpaceCompactResult::kSuccess;
2218 }
2219 
SetDefaultConcurrentStartBytes()2220 void Heap::SetDefaultConcurrentStartBytes() {
2221   MutexLock mu(Thread::Current(), *gc_complete_lock_);
2222   if (collector_type_running_ != kCollectorTypeNone) {
2223     // If a collector is already running, just let it set concurrent_start_bytes_ .
2224     return;
2225   }
2226   SetDefaultConcurrentStartBytesLocked();
2227 }
2228 
SetDefaultConcurrentStartBytesLocked()2229 void Heap::SetDefaultConcurrentStartBytesLocked() {
2230   if (IsGcConcurrent()) {
2231     size_t target_footprint = target_footprint_.load(std::memory_order_relaxed);
2232     size_t reserve_bytes = target_footprint / 4;
2233     reserve_bytes = std::min(reserve_bytes, kMaxConcurrentRemainingBytes);
2234     reserve_bytes = std::max(reserve_bytes, kMinConcurrentRemainingBytes);
2235     concurrent_start_bytes_ = UnsignedDifference(target_footprint, reserve_bytes);
2236   } else {
2237     concurrent_start_bytes_ = std::numeric_limits<size_t>::max();
2238   }
2239 }
2240 
ChangeCollector(CollectorType collector_type)2241 void Heap::ChangeCollector(CollectorType collector_type) {
2242   // TODO: Only do this with all mutators suspended to avoid races.
2243   if (collector_type != collector_type_) {
2244     collector_type_ = collector_type;
2245     gc_plan_.clear();
2246     switch (collector_type_) {
2247       case kCollectorTypeCC: {
2248         if (use_generational_cc_) {
2249           gc_plan_.push_back(collector::kGcTypeSticky);
2250         }
2251         gc_plan_.push_back(collector::kGcTypeFull);
2252         if (use_tlab_) {
2253           ChangeAllocator(kAllocatorTypeRegionTLAB);
2254         } else {
2255           ChangeAllocator(kAllocatorTypeRegion);
2256         }
2257         break;
2258       }
2259       case kCollectorTypeCMC: {
2260         gc_plan_.push_back(collector::kGcTypeFull);
2261         if (use_tlab_) {
2262           ChangeAllocator(kAllocatorTypeTLAB);
2263         } else {
2264           ChangeAllocator(kAllocatorTypeBumpPointer);
2265         }
2266         break;
2267       }
2268       case kCollectorTypeSS: {
2269         gc_plan_.push_back(collector::kGcTypeFull);
2270         if (use_tlab_) {
2271           ChangeAllocator(kAllocatorTypeTLAB);
2272         } else {
2273           ChangeAllocator(kAllocatorTypeBumpPointer);
2274         }
2275         break;
2276       }
2277       case kCollectorTypeMS: {
2278         gc_plan_.push_back(collector::kGcTypeSticky);
2279         gc_plan_.push_back(collector::kGcTypePartial);
2280         gc_plan_.push_back(collector::kGcTypeFull);
2281         ChangeAllocator(kUseRosAlloc ? kAllocatorTypeRosAlloc : kAllocatorTypeDlMalloc);
2282         break;
2283       }
2284       case kCollectorTypeCMS: {
2285         gc_plan_.push_back(collector::kGcTypeSticky);
2286         gc_plan_.push_back(collector::kGcTypePartial);
2287         gc_plan_.push_back(collector::kGcTypeFull);
2288         ChangeAllocator(kUseRosAlloc ? kAllocatorTypeRosAlloc : kAllocatorTypeDlMalloc);
2289         break;
2290       }
2291       default: {
2292         UNIMPLEMENTED(FATAL);
2293         UNREACHABLE();
2294       }
2295     }
2296     SetDefaultConcurrentStartBytesLocked();
2297   }
2298 }
2299 
2300 // Special compacting collector which uses sub-optimal bin packing to reduce zygote space size.
2301 class ZygoteCompactingCollector final : public collector::SemiSpace {
2302  public:
ZygoteCompactingCollector(gc::Heap * heap,bool is_running_on_memory_tool)2303   ZygoteCompactingCollector(gc::Heap* heap, bool is_running_on_memory_tool)
2304       : SemiSpace(heap, "zygote collector"),
2305         bin_live_bitmap_(nullptr),
2306         bin_mark_bitmap_(nullptr),
2307         is_running_on_memory_tool_(is_running_on_memory_tool) {}
2308 
BuildBins(space::ContinuousSpace * space)2309   void BuildBins(space::ContinuousSpace* space) REQUIRES_SHARED(Locks::mutator_lock_) {
2310     bin_live_bitmap_ = space->GetLiveBitmap();
2311     bin_mark_bitmap_ = space->GetMarkBitmap();
2312     uintptr_t prev = reinterpret_cast<uintptr_t>(space->Begin());
2313     WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
2314     // Note: This requires traversing the space in increasing order of object addresses.
2315     auto visitor = [&](mirror::Object* obj) REQUIRES_SHARED(Locks::mutator_lock_) {
2316       uintptr_t object_addr = reinterpret_cast<uintptr_t>(obj);
2317       size_t bin_size = object_addr - prev;
2318       // Add the bin consisting of the end of the previous object to the start of the current object.
2319       AddBin(bin_size, prev);
2320       prev = object_addr + RoundUp(obj->SizeOf<kDefaultVerifyFlags>(), kObjectAlignment);
2321     };
2322     bin_live_bitmap_->Walk(visitor);
2323     // Add the last bin which spans after the last object to the end of the space.
2324     AddBin(reinterpret_cast<uintptr_t>(space->End()) - prev, prev);
2325   }
2326 
2327  private:
2328   // Maps from bin sizes to locations.
2329   std::multimap<size_t, uintptr_t> bins_;
2330   // Live bitmap of the space which contains the bins.
2331   accounting::ContinuousSpaceBitmap* bin_live_bitmap_;
2332   // Mark bitmap of the space which contains the bins.
2333   accounting::ContinuousSpaceBitmap* bin_mark_bitmap_;
2334   const bool is_running_on_memory_tool_;
2335 
AddBin(size_t size,uintptr_t position)2336   void AddBin(size_t size, uintptr_t position) {
2337     if (is_running_on_memory_tool_) {
2338       MEMORY_TOOL_MAKE_DEFINED(reinterpret_cast<void*>(position), size);
2339     }
2340     if (size != 0) {
2341       bins_.insert(std::make_pair(size, position));
2342     }
2343   }
2344 
ShouldSweepSpace(space::ContinuousSpace * space ATTRIBUTE_UNUSED) const2345   bool ShouldSweepSpace(space::ContinuousSpace* space ATTRIBUTE_UNUSED) const override {
2346     // Don't sweep any spaces since we probably blasted the internal accounting of the free list
2347     // allocator.
2348     return false;
2349   }
2350 
MarkNonForwardedObject(mirror::Object * obj)2351   mirror::Object* MarkNonForwardedObject(mirror::Object* obj) override
2352       REQUIRES(Locks::heap_bitmap_lock_, Locks::mutator_lock_) {
2353     size_t obj_size = obj->SizeOf<kDefaultVerifyFlags>();
2354     size_t alloc_size = RoundUp(obj_size, kObjectAlignment);
2355     mirror::Object* forward_address;
2356     // Find the smallest bin which we can move obj in.
2357     auto it = bins_.lower_bound(alloc_size);
2358     if (it == bins_.end()) {
2359       // No available space in the bins, place it in the target space instead (grows the zygote
2360       // space).
2361       size_t bytes_allocated, unused_bytes_tl_bulk_allocated;
2362       forward_address = to_space_->Alloc(
2363           self_, alloc_size, &bytes_allocated, nullptr, &unused_bytes_tl_bulk_allocated);
2364       if (to_space_live_bitmap_ != nullptr) {
2365         to_space_live_bitmap_->Set(forward_address);
2366       } else {
2367         GetHeap()->GetNonMovingSpace()->GetLiveBitmap()->Set(forward_address);
2368         GetHeap()->GetNonMovingSpace()->GetMarkBitmap()->Set(forward_address);
2369       }
2370     } else {
2371       size_t size = it->first;
2372       uintptr_t pos = it->second;
2373       bins_.erase(it);  // Erase the old bin which we replace with the new smaller bin.
2374       forward_address = reinterpret_cast<mirror::Object*>(pos);
2375       // Set the live and mark bits so that sweeping system weaks works properly.
2376       bin_live_bitmap_->Set(forward_address);
2377       bin_mark_bitmap_->Set(forward_address);
2378       DCHECK_GE(size, alloc_size);
2379       // Add a new bin with the remaining space.
2380       AddBin(size - alloc_size, pos + alloc_size);
2381     }
2382     // Copy the object over to its new location.
2383     // Historical note: We did not use `alloc_size` to avoid a Valgrind error.
2384     memcpy(reinterpret_cast<void*>(forward_address), obj, obj_size);
2385     if (kUseBakerReadBarrier) {
2386       obj->AssertReadBarrierState();
2387       forward_address->AssertReadBarrierState();
2388     }
2389     return forward_address;
2390   }
2391 };
2392 
UnBindBitmaps()2393 void Heap::UnBindBitmaps() {
2394   TimingLogger::ScopedTiming t("UnBindBitmaps", GetCurrentGcIteration()->GetTimings());
2395   for (const auto& space : GetContinuousSpaces()) {
2396     if (space->IsContinuousMemMapAllocSpace()) {
2397       space::ContinuousMemMapAllocSpace* alloc_space = space->AsContinuousMemMapAllocSpace();
2398       if (alloc_space->GetLiveBitmap() != nullptr && alloc_space->HasBoundBitmaps()) {
2399         alloc_space->UnBindBitmaps();
2400       }
2401     }
2402   }
2403 }
2404 
IncrementFreedEver()2405 void Heap::IncrementFreedEver() {
2406   // Counters are updated only by us, but may be read concurrently.
2407   // The updates should become visible after the corresponding live object info.
2408   total_objects_freed_ever_.store(total_objects_freed_ever_.load(std::memory_order_relaxed)
2409                                   + GetCurrentGcIteration()->GetFreedObjects()
2410                                   + GetCurrentGcIteration()->GetFreedLargeObjects(),
2411                                   std::memory_order_release);
2412   total_bytes_freed_ever_.store(total_bytes_freed_ever_.load(std::memory_order_relaxed)
2413                                 + GetCurrentGcIteration()->GetFreedBytes()
2414                                 + GetCurrentGcIteration()->GetFreedLargeObjectBytes(),
2415                                 std::memory_order_release);
2416 }
2417 
2418 #pragma clang diagnostic push
2419 #if !ART_USE_FUTEXES
2420 // Frame gets too large, perhaps due to Bionic pthread_mutex_lock size. We don't care.
2421 #  pragma clang diagnostic ignored "-Wframe-larger-than="
2422 #endif
2423 // This has a large frame, but shouldn't be run anywhere near the stack limit.
2424 // FIXME: BUT it did exceed... http://b/197647048
2425 #  pragma clang diagnostic ignored "-Wframe-larger-than="
PreZygoteFork()2426 void Heap::PreZygoteFork() {
2427   if (!HasZygoteSpace()) {
2428     // We still want to GC in case there is some unreachable non moving objects that could cause a
2429     // suboptimal bin packing when we compact the zygote space.
2430     CollectGarbageInternal(collector::kGcTypeFull, kGcCauseBackground, false, GC_NUM_ANY);
2431     // Trim the pages at the end of the non moving space. Trim while not holding zygote lock since
2432     // the trim process may require locking the mutator lock.
2433     non_moving_space_->Trim();
2434   }
2435   // We need to close userfaultfd fd for app/webview zygotes to avoid getattr
2436   // (stat) on the fd during fork.
2437   Thread* self = Thread::Current();
2438   MutexLock mu(self, zygote_creation_lock_);
2439   // Try to see if we have any Zygote spaces.
2440   if (HasZygoteSpace()) {
2441     return;
2442   }
2443   Runtime* runtime = Runtime::Current();
2444   runtime->GetInternTable()->AddNewTable();
2445   runtime->GetClassLinker()->MoveClassTableToPreZygote();
2446   runtime->SetupLinearAllocForPostZygoteFork(self);
2447   VLOG(heap) << "Starting PreZygoteFork";
2448   // The end of the non-moving space may be protected, unprotect it so that we can copy the zygote
2449   // there.
2450   non_moving_space_->GetMemMap()->Protect(PROT_READ | PROT_WRITE);
2451   const bool same_space = non_moving_space_ == main_space_;
2452   if (kCompactZygote) {
2453     // Temporarily disable rosalloc verification because the zygote
2454     // compaction will mess up the rosalloc internal metadata.
2455     ScopedDisableRosAllocVerification disable_rosalloc_verif(this);
2456     ZygoteCompactingCollector zygote_collector(this, is_running_on_memory_tool_);
2457     zygote_collector.BuildBins(non_moving_space_);
2458     // Create a new bump pointer space which we will compact into.
2459     space::BumpPointerSpace target_space("zygote bump space", non_moving_space_->End(),
2460                                          non_moving_space_->Limit());
2461     // Compact the bump pointer space to a new zygote bump pointer space.
2462     bool reset_main_space = false;
2463     if (IsMovingGc(collector_type_)) {
2464       if (collector_type_ == kCollectorTypeCC) {
2465         zygote_collector.SetFromSpace(region_space_);
2466       } else {
2467         zygote_collector.SetFromSpace(bump_pointer_space_);
2468       }
2469     } else {
2470       CHECK(main_space_ != nullptr);
2471       CHECK_NE(main_space_, non_moving_space_)
2472           << "Does not make sense to compact within the same space";
2473       // Copy from the main space.
2474       zygote_collector.SetFromSpace(main_space_);
2475       reset_main_space = true;
2476     }
2477     zygote_collector.SetToSpace(&target_space);
2478     zygote_collector.SetSwapSemiSpaces(false);
2479     zygote_collector.Run(kGcCauseCollectorTransition, false);
2480     if (reset_main_space) {
2481       main_space_->GetMemMap()->Protect(PROT_READ | PROT_WRITE);
2482       madvise(main_space_->Begin(), main_space_->Capacity(), MADV_DONTNEED);
2483       MemMap mem_map = main_space_->ReleaseMemMap();
2484       RemoveSpace(main_space_);
2485       space::Space* old_main_space = main_space_;
2486       CreateMainMallocSpace(std::move(mem_map),
2487                             kDefaultInitialSize,
2488                             std::min(mem_map.Size(), growth_limit_),
2489                             mem_map.Size());
2490       delete old_main_space;
2491       AddSpace(main_space_);
2492     } else {
2493       if (collector_type_ == kCollectorTypeCC) {
2494         region_space_->GetMemMap()->Protect(PROT_READ | PROT_WRITE);
2495         // Evacuated everything out of the region space, clear the mark bitmap.
2496         region_space_->GetMarkBitmap()->Clear();
2497       } else {
2498         bump_pointer_space_->GetMemMap()->Protect(PROT_READ | PROT_WRITE);
2499       }
2500     }
2501     if (temp_space_ != nullptr) {
2502       CHECK(temp_space_->IsEmpty());
2503     }
2504     IncrementFreedEver();
2505     // Update the end and write out image.
2506     non_moving_space_->SetEnd(target_space.End());
2507     non_moving_space_->SetLimit(target_space.Limit());
2508     VLOG(heap) << "Create zygote space with size=" << non_moving_space_->Size() << " bytes";
2509   }
2510   // Change the collector to the post zygote one.
2511   ChangeCollector(foreground_collector_type_);
2512   // Save the old space so that we can remove it after we complete creating the zygote space.
2513   space::MallocSpace* old_alloc_space = non_moving_space_;
2514   // Turn the current alloc space into a zygote space and obtain the new alloc space composed of
2515   // the remaining available space.
2516   // Remove the old space before creating the zygote space since creating the zygote space sets
2517   // the old alloc space's bitmaps to null.
2518   RemoveSpace(old_alloc_space);
2519   if (collector::SemiSpace::kUseRememberedSet) {
2520     // Consistency bound check.
2521     FindRememberedSetFromSpace(old_alloc_space)->AssertAllDirtyCardsAreWithinSpace();
2522     // Remove the remembered set for the now zygote space (the old
2523     // non-moving space). Note now that we have compacted objects into
2524     // the zygote space, the data in the remembered set is no longer
2525     // needed. The zygote space will instead have a mod-union table
2526     // from this point on.
2527     RemoveRememberedSet(old_alloc_space);
2528   }
2529   // Remaining space becomes the new non moving space.
2530   zygote_space_ = old_alloc_space->CreateZygoteSpace(kNonMovingSpaceName, low_memory_mode_,
2531                                                      &non_moving_space_);
2532   CHECK(!non_moving_space_->CanMoveObjects());
2533   if (same_space) {
2534     main_space_ = non_moving_space_;
2535     SetSpaceAsDefault(main_space_);
2536   }
2537   delete old_alloc_space;
2538   CHECK(HasZygoteSpace()) << "Failed creating zygote space";
2539   AddSpace(zygote_space_);
2540   non_moving_space_->SetFootprintLimit(non_moving_space_->Capacity());
2541   AddSpace(non_moving_space_);
2542   constexpr bool set_mark_bit = kUseBakerReadBarrier
2543                                 && gc::collector::ConcurrentCopying::kGrayDirtyImmuneObjects;
2544   if (set_mark_bit) {
2545     // Treat all of the objects in the zygote as marked to avoid unnecessary dirty pages. This is
2546     // safe since we mark all of the objects that may reference non immune objects as gray.
2547     zygote_space_->SetMarkBitInLiveObjects();
2548   }
2549 
2550   // Create the zygote space mod union table.
2551   accounting::ModUnionTable* mod_union_table =
2552       new accounting::ModUnionTableCardCache("zygote space mod-union table", this, zygote_space_);
2553   CHECK(mod_union_table != nullptr) << "Failed to create zygote space mod-union table";
2554 
2555   if (collector_type_ != kCollectorTypeCC && collector_type_ != kCollectorTypeCMC) {
2556     // Set all the cards in the mod-union table since we don't know which objects contain references
2557     // to large objects.
2558     mod_union_table->SetCards();
2559   } else {
2560     // Make sure to clear the zygote space cards so that we don't dirty pages in the next GC. There
2561     // may be dirty cards from the zygote compaction or reference processing. These cards are not
2562     // necessary to have marked since the zygote space may not refer to any objects not in the
2563     // zygote or image spaces at this point.
2564     mod_union_table->ProcessCards();
2565     mod_union_table->ClearTable();
2566 
2567     // For CC and CMC we never collect zygote large objects. This means we do not need to set the
2568     // cards for the zygote mod-union table and we can also clear all of the existing image
2569     // mod-union tables. The existing mod-union tables are only for image spaces and may only
2570     // reference zygote and image objects.
2571     for (auto& pair : mod_union_tables_) {
2572       CHECK(pair.first->IsImageSpace());
2573       CHECK(!pair.first->AsImageSpace()->GetImageHeader().IsAppImage());
2574       accounting::ModUnionTable* table = pair.second;
2575       table->ClearTable();
2576     }
2577   }
2578   AddModUnionTable(mod_union_table);
2579   large_object_space_->SetAllLargeObjectsAsZygoteObjects(self, set_mark_bit);
2580   if (collector::SemiSpace::kUseRememberedSet) {
2581     // Add a new remembered set for the post-zygote non-moving space.
2582     accounting::RememberedSet* post_zygote_non_moving_space_rem_set =
2583         new accounting::RememberedSet("Post-zygote non-moving space remembered set", this,
2584                                       non_moving_space_);
2585     CHECK(post_zygote_non_moving_space_rem_set != nullptr)
2586         << "Failed to create post-zygote non-moving space remembered set";
2587     AddRememberedSet(post_zygote_non_moving_space_rem_set);
2588   }
2589 }
2590 #pragma clang diagnostic pop
2591 
FlushAllocStack()2592 void Heap::FlushAllocStack() {
2593   MarkAllocStackAsLive(allocation_stack_.get());
2594   allocation_stack_->Reset();
2595 }
2596 
MarkAllocStack(accounting::ContinuousSpaceBitmap * bitmap1,accounting::ContinuousSpaceBitmap * bitmap2,accounting::LargeObjectBitmap * large_objects,accounting::ObjectStack * stack)2597 void Heap::MarkAllocStack(accounting::ContinuousSpaceBitmap* bitmap1,
2598                           accounting::ContinuousSpaceBitmap* bitmap2,
2599                           accounting::LargeObjectBitmap* large_objects,
2600                           accounting::ObjectStack* stack) {
2601   DCHECK(bitmap1 != nullptr);
2602   DCHECK(bitmap2 != nullptr);
2603   const auto* limit = stack->End();
2604   for (auto* it = stack->Begin(); it != limit; ++it) {
2605     const mirror::Object* obj = it->AsMirrorPtr();
2606     if (!kUseThreadLocalAllocationStack || obj != nullptr) {
2607       if (bitmap1->HasAddress(obj)) {
2608         bitmap1->Set(obj);
2609       } else if (bitmap2->HasAddress(obj)) {
2610         bitmap2->Set(obj);
2611       } else {
2612         DCHECK(large_objects != nullptr);
2613         large_objects->Set(obj);
2614       }
2615     }
2616   }
2617 }
2618 
SwapSemiSpaces()2619 void Heap::SwapSemiSpaces() {
2620   CHECK(bump_pointer_space_ != nullptr);
2621   CHECK(temp_space_ != nullptr);
2622   std::swap(bump_pointer_space_, temp_space_);
2623 }
2624 
Compact(space::ContinuousMemMapAllocSpace * target_space,space::ContinuousMemMapAllocSpace * source_space,GcCause gc_cause)2625 collector::GarbageCollector* Heap::Compact(space::ContinuousMemMapAllocSpace* target_space,
2626                                            space::ContinuousMemMapAllocSpace* source_space,
2627                                            GcCause gc_cause) {
2628   CHECK(kMovingCollector);
2629   if (target_space != source_space) {
2630     // Don't swap spaces since this isn't a typical semi space collection.
2631     semi_space_collector_->SetSwapSemiSpaces(false);
2632     semi_space_collector_->SetFromSpace(source_space);
2633     semi_space_collector_->SetToSpace(target_space);
2634     semi_space_collector_->Run(gc_cause, false);
2635     return semi_space_collector_;
2636   }
2637   LOG(FATAL) << "Unsupported";
2638   UNREACHABLE();
2639 }
2640 
TraceHeapSize(size_t heap_size)2641 void Heap::TraceHeapSize(size_t heap_size) {
2642   ATraceIntegerValue("Heap size (KB)", heap_size / KB);
2643 }
2644 
2645 #if defined(__GLIBC__)
2646 # define IF_GLIBC(x) x
2647 #else
2648 # define IF_GLIBC(x)
2649 #endif
2650 
GetNativeBytes()2651 size_t Heap::GetNativeBytes() {
2652   size_t malloc_bytes;
2653 #if defined(__BIONIC__) || defined(__GLIBC__)
2654   IF_GLIBC(size_t mmapped_bytes;)
2655   struct mallinfo mi = mallinfo();
2656   // In spite of the documentation, the jemalloc version of this call seems to do what we want,
2657   // and it is thread-safe.
2658   if (sizeof(size_t) > sizeof(mi.uordblks) && sizeof(size_t) > sizeof(mi.hblkhd)) {
2659     // Shouldn't happen, but glibc declares uordblks as int.
2660     // Avoiding sign extension gets us correct behavior for another 2 GB.
2661     malloc_bytes = (unsigned int)mi.uordblks;
2662     IF_GLIBC(mmapped_bytes = (unsigned int)mi.hblkhd;)
2663   } else {
2664     malloc_bytes = mi.uordblks;
2665     IF_GLIBC(mmapped_bytes = mi.hblkhd;)
2666   }
2667   // From the spec, it appeared mmapped_bytes <= malloc_bytes. Reality was sometimes
2668   // dramatically different. (b/119580449 was an early bug.) If so, we try to fudge it.
2669   // However, malloc implementations seem to interpret hblkhd differently, namely as
2670   // mapped blocks backing the entire heap (e.g. jemalloc) vs. large objects directly
2671   // allocated via mmap (e.g. glibc). Thus we now only do this for glibc, where it
2672   // previously helped, and which appears to use a reading of the spec compatible
2673   // with our adjustment.
2674 #if defined(__GLIBC__)
2675   if (mmapped_bytes > malloc_bytes) {
2676     malloc_bytes = mmapped_bytes;
2677   }
2678 #endif  // GLIBC
2679 #else  // Neither Bionic nor Glibc
2680   // We should hit this case only in contexts in which GC triggering is not critical. Effectively
2681   // disable GC triggering based on malloc().
2682   malloc_bytes = 1000;
2683 #endif
2684   return malloc_bytes + native_bytes_registered_.load(std::memory_order_relaxed);
2685   // An alternative would be to get RSS from /proc/self/statm. Empirically, that's no
2686   // more expensive, and it would allow us to count memory allocated by means other than malloc.
2687   // However it would change as pages are unmapped and remapped due to memory pressure, among
2688   // other things. It seems risky to trigger GCs as a result of such changes.
2689 }
2690 
GCNumberLt(uint32_t gc_num1,uint32_t gc_num2)2691 static inline bool GCNumberLt(uint32_t gc_num1, uint32_t gc_num2) {
2692   // unsigned comparison, assuming a non-huge difference, but dealing correctly with wrapping.
2693   uint32_t difference = gc_num2 - gc_num1;
2694   bool completed_more_than_requested = difference > 0x80000000;
2695   return difference > 0 && !completed_more_than_requested;
2696 }
2697 
2698 
CollectGarbageInternal(collector::GcType gc_type,GcCause gc_cause,bool clear_soft_references,uint32_t requested_gc_num)2699 collector::GcType Heap::CollectGarbageInternal(collector::GcType gc_type,
2700                                                GcCause gc_cause,
2701                                                bool clear_soft_references,
2702                                                uint32_t requested_gc_num) {
2703   Thread* self = Thread::Current();
2704   Runtime* runtime = Runtime::Current();
2705   // If the heap can't run the GC, silently fail and return that no GC was run.
2706   switch (gc_type) {
2707     case collector::kGcTypePartial: {
2708       if (!HasZygoteSpace()) {
2709         // Do not increment gcs_completed_ . We should retry with kGcTypeFull.
2710         return collector::kGcTypeNone;
2711       }
2712       break;
2713     }
2714     default: {
2715       // Other GC types don't have any special cases which makes them not runnable. The main case
2716       // here is full GC.
2717     }
2718   }
2719   ScopedThreadStateChange tsc(self, ThreadState::kWaitingPerformingGc);
2720   Locks::mutator_lock_->AssertNotHeld(self);
2721   if (self->IsHandlingStackOverflow()) {
2722     // If we are throwing a stack overflow error we probably don't have enough remaining stack
2723     // space to run the GC.
2724     // Count this as a GC in case someone is waiting for it to complete.
2725     gcs_completed_.fetch_add(1, std::memory_order_release);
2726     return collector::kGcTypeNone;
2727   }
2728   bool compacting_gc;
2729   {
2730     gc_complete_lock_->AssertNotHeld(self);
2731     ScopedThreadStateChange tsc2(self, ThreadState::kWaitingForGcToComplete);
2732     MutexLock mu(self, *gc_complete_lock_);
2733     // Ensure there is only one GC at a time.
2734     WaitForGcToCompleteLocked(gc_cause, self);
2735     if (requested_gc_num != GC_NUM_ANY && !GCNumberLt(GetCurrentGcNum(), requested_gc_num)) {
2736       // The appropriate GC was already triggered elsewhere.
2737       return collector::kGcTypeNone;
2738     }
2739     compacting_gc = IsMovingGc(collector_type_);
2740     // GC can be disabled if someone has a used GetPrimitiveArrayCritical.
2741     if (compacting_gc && disable_moving_gc_count_ != 0) {
2742       LOG(WARNING) << "Skipping GC due to disable moving GC count " << disable_moving_gc_count_;
2743       // Again count this as a GC.
2744       gcs_completed_.fetch_add(1, std::memory_order_release);
2745       return collector::kGcTypeNone;
2746     }
2747     if (gc_disabled_for_shutdown_) {
2748       gcs_completed_.fetch_add(1, std::memory_order_release);
2749       return collector::kGcTypeNone;
2750     }
2751     collector_type_running_ = collector_type_;
2752     last_gc_cause_ = gc_cause;
2753   }
2754   if (gc_cause == kGcCauseForAlloc && runtime->HasStatsEnabled()) {
2755     ++runtime->GetStats()->gc_for_alloc_count;
2756     ++self->GetStats()->gc_for_alloc_count;
2757   }
2758   const size_t bytes_allocated_before_gc = GetBytesAllocated();
2759 
2760   DCHECK_LT(gc_type, collector::kGcTypeMax);
2761   DCHECK_NE(gc_type, collector::kGcTypeNone);
2762 
2763   collector::GarbageCollector* collector = nullptr;
2764   // TODO: Clean this up.
2765   if (compacting_gc) {
2766     DCHECK(current_allocator_ == kAllocatorTypeBumpPointer ||
2767            current_allocator_ == kAllocatorTypeTLAB ||
2768            current_allocator_ == kAllocatorTypeRegion ||
2769            current_allocator_ == kAllocatorTypeRegionTLAB);
2770     switch (collector_type_) {
2771       case kCollectorTypeSS:
2772         semi_space_collector_->SetFromSpace(bump_pointer_space_);
2773         semi_space_collector_->SetToSpace(temp_space_);
2774         semi_space_collector_->SetSwapSemiSpaces(true);
2775         collector = semi_space_collector_;
2776         break;
2777       case kCollectorTypeCMC:
2778         collector = mark_compact_;
2779         break;
2780       case kCollectorTypeCC:
2781         collector::ConcurrentCopying* active_cc_collector;
2782         if (use_generational_cc_) {
2783           // TODO: Other threads must do the flip checkpoint before they start poking at
2784           // active_concurrent_copying_collector_. So we should not concurrency here.
2785           active_cc_collector = (gc_type == collector::kGcTypeSticky) ?
2786                   young_concurrent_copying_collector_ : concurrent_copying_collector_;
2787           active_concurrent_copying_collector_.store(active_cc_collector,
2788                                                      std::memory_order_relaxed);
2789           DCHECK(active_cc_collector->RegionSpace() == region_space_);
2790           collector = active_cc_collector;
2791         } else {
2792           collector = active_concurrent_copying_collector_.load(std::memory_order_relaxed);
2793         }
2794         break;
2795       default:
2796         LOG(FATAL) << "Invalid collector type " << static_cast<size_t>(collector_type_);
2797     }
2798     // temp_space_ will be null for kCollectorTypeCMC.
2799     if (temp_space_ != nullptr
2800         && collector != active_concurrent_copying_collector_.load(std::memory_order_relaxed)) {
2801       temp_space_->GetMemMap()->Protect(PROT_READ | PROT_WRITE);
2802       if (kIsDebugBuild) {
2803         // Try to read each page of the memory map in case mprotect didn't work properly b/19894268.
2804         temp_space_->GetMemMap()->TryReadable();
2805       }
2806       CHECK(temp_space_->IsEmpty());
2807     }
2808   } else if (current_allocator_ == kAllocatorTypeRosAlloc ||
2809       current_allocator_ == kAllocatorTypeDlMalloc) {
2810     collector = FindCollectorByGcType(gc_type);
2811   } else {
2812     LOG(FATAL) << "Invalid current allocator " << current_allocator_;
2813   }
2814 
2815   CHECK(collector != nullptr)
2816       << "Could not find garbage collector with collector_type="
2817       << static_cast<size_t>(collector_type_) << " and gc_type=" << gc_type;
2818   collector->Run(gc_cause, clear_soft_references || runtime->IsZygote());
2819   IncrementFreedEver();
2820   RequestTrim(self);
2821   // Collect cleared references.
2822   SelfDeletingTask* clear = reference_processor_->CollectClearedReferences(self);
2823   // Grow the heap so that we know when to perform the next GC.
2824   GrowForUtilization(collector, bytes_allocated_before_gc);
2825   old_native_bytes_allocated_.store(GetNativeBytes());
2826   LogGC(gc_cause, collector);
2827   FinishGC(self, gc_type);
2828   // Actually enqueue all cleared references. Do this after the GC has officially finished since
2829   // otherwise we can deadlock.
2830   clear->Run(self);
2831   clear->Finalize();
2832   // Inform DDMS that a GC completed.
2833   Dbg::GcDidFinish();
2834 
2835   // Unload native libraries for class unloading. We do this after calling FinishGC to prevent
2836   // deadlocks in case the JNI_OnUnload function does allocations.
2837   {
2838     ScopedObjectAccess soa(self);
2839     soa.Vm()->UnloadNativeLibraries();
2840   }
2841   return gc_type;
2842 }
2843 
LogGC(GcCause gc_cause,collector::GarbageCollector * collector)2844 void Heap::LogGC(GcCause gc_cause, collector::GarbageCollector* collector) {
2845   const size_t duration = GetCurrentGcIteration()->GetDurationNs();
2846   const std::vector<uint64_t>& pause_times = GetCurrentGcIteration()->GetPauseTimes();
2847   // Print the GC if it is an explicit GC (e.g. Runtime.gc()) or a slow GC
2848   // (mutator time blocked >= long_pause_log_threshold_).
2849   bool log_gc = kLogAllGCs || (gc_cause == kGcCauseExplicit && always_log_explicit_gcs_);
2850   if (!log_gc && CareAboutPauseTimes()) {
2851     // GC for alloc pauses the allocating thread, so consider it as a pause.
2852     log_gc = duration > long_gc_log_threshold_ ||
2853         (gc_cause == kGcCauseForAlloc && duration > long_pause_log_threshold_);
2854     for (uint64_t pause : pause_times) {
2855       log_gc = log_gc || pause >= long_pause_log_threshold_;
2856     }
2857   }
2858   bool is_sampled = false;
2859   if (UNLIKELY(gc_stress_mode_)) {
2860     static std::atomic_int64_t accumulated_duration_ns = 0;
2861     accumulated_duration_ns += duration;
2862     if (accumulated_duration_ns >= kGcStressModeGcLogSampleFrequencyNs) {
2863       accumulated_duration_ns -= kGcStressModeGcLogSampleFrequencyNs;
2864       log_gc = true;
2865       is_sampled = true;
2866     }
2867   }
2868   if (log_gc) {
2869     const size_t percent_free = GetPercentFree();
2870     const size_t current_heap_size = GetBytesAllocated();
2871     const size_t total_memory = GetTotalMemory();
2872     std::ostringstream pause_string;
2873     for (size_t i = 0; i < pause_times.size(); ++i) {
2874       pause_string << PrettyDuration((pause_times[i] / 1000) * 1000)
2875                    << ((i != pause_times.size() - 1) ? "," : "");
2876     }
2877     LOG(INFO) << gc_cause << " " << collector->GetName()
2878               << (is_sampled ? " (sampled)" : "")
2879               << " GC freed "  << current_gc_iteration_.GetFreedObjects() << "("
2880               << PrettySize(current_gc_iteration_.GetFreedBytes()) << ") AllocSpace objects, "
2881               << current_gc_iteration_.GetFreedLargeObjects() << "("
2882               << PrettySize(current_gc_iteration_.GetFreedLargeObjectBytes()) << ") LOS objects, "
2883               << percent_free << "% free, " << PrettySize(current_heap_size) << "/"
2884               << PrettySize(total_memory) << ", " << "paused " << pause_string.str()
2885               << " total " << PrettyDuration((duration / 1000) * 1000);
2886     VLOG(heap) << Dumpable<TimingLogger>(*current_gc_iteration_.GetTimings());
2887   }
2888 }
2889 
FinishGC(Thread * self,collector::GcType gc_type)2890 void Heap::FinishGC(Thread* self, collector::GcType gc_type) {
2891   MutexLock mu(self, *gc_complete_lock_);
2892   collector_type_running_ = kCollectorTypeNone;
2893   if (gc_type != collector::kGcTypeNone) {
2894     last_gc_type_ = gc_type;
2895 
2896     // Update stats.
2897     ++gc_count_last_window_;
2898     if (running_collection_is_blocking_) {
2899       // If the currently running collection was a blocking one,
2900       // increment the counters and reset the flag.
2901       ++blocking_gc_count_;
2902       blocking_gc_time_ += GetCurrentGcIteration()->GetDurationNs();
2903       ++blocking_gc_count_last_window_;
2904     }
2905     // Update the gc count rate histograms if due.
2906     UpdateGcCountRateHistograms();
2907   }
2908   // Reset.
2909   running_collection_is_blocking_ = false;
2910   thread_running_gc_ = nullptr;
2911   if (gc_type != collector::kGcTypeNone) {
2912     gcs_completed_.fetch_add(1, std::memory_order_release);
2913   }
2914   // Wake anyone who may have been waiting for the GC to complete.
2915   gc_complete_cond_->Broadcast(self);
2916 }
2917 
UpdateGcCountRateHistograms()2918 void Heap::UpdateGcCountRateHistograms() {
2919   // Invariant: if the time since the last update includes more than
2920   // one windows, all the GC runs (if > 0) must have happened in first
2921   // window because otherwise the update must have already taken place
2922   // at an earlier GC run. So, we report the non-first windows with
2923   // zero counts to the histograms.
2924   DCHECK_EQ(last_update_time_gc_count_rate_histograms_ % kGcCountRateHistogramWindowDuration, 0U);
2925   uint64_t now = NanoTime();
2926   DCHECK_GE(now, last_update_time_gc_count_rate_histograms_);
2927   uint64_t time_since_last_update = now - last_update_time_gc_count_rate_histograms_;
2928   uint64_t num_of_windows = time_since_last_update / kGcCountRateHistogramWindowDuration;
2929 
2930   // The computed number of windows can be incoherently high if NanoTime() is not monotonic.
2931   // Setting a limit on its maximum value reduces the impact on CPU time in such cases.
2932   if (num_of_windows > kGcCountRateHistogramMaxNumMissedWindows) {
2933     LOG(WARNING) << "Reducing the number of considered missed Gc histogram windows from "
2934                  << num_of_windows << " to " << kGcCountRateHistogramMaxNumMissedWindows;
2935     num_of_windows = kGcCountRateHistogramMaxNumMissedWindows;
2936   }
2937 
2938   if (time_since_last_update >= kGcCountRateHistogramWindowDuration) {
2939     // Record the first window.
2940     gc_count_rate_histogram_.AddValue(gc_count_last_window_ - 1);  // Exclude the current run.
2941     blocking_gc_count_rate_histogram_.AddValue(running_collection_is_blocking_ ?
2942         blocking_gc_count_last_window_ - 1 : blocking_gc_count_last_window_);
2943     // Record the other windows (with zero counts).
2944     for (uint64_t i = 0; i < num_of_windows - 1; ++i) {
2945       gc_count_rate_histogram_.AddValue(0);
2946       blocking_gc_count_rate_histogram_.AddValue(0);
2947     }
2948     // Update the last update time and reset the counters.
2949     last_update_time_gc_count_rate_histograms_ =
2950         (now / kGcCountRateHistogramWindowDuration) * kGcCountRateHistogramWindowDuration;
2951     gc_count_last_window_ = 1;  // Include the current run.
2952     blocking_gc_count_last_window_ = running_collection_is_blocking_ ? 1 : 0;
2953   }
2954   DCHECK_EQ(last_update_time_gc_count_rate_histograms_ % kGcCountRateHistogramWindowDuration, 0U);
2955 }
2956 
2957 class RootMatchesObjectVisitor : public SingleRootVisitor {
2958  public:
RootMatchesObjectVisitor(const mirror::Object * obj)2959   explicit RootMatchesObjectVisitor(const mirror::Object* obj) : obj_(obj) { }
2960 
VisitRoot(mirror::Object * root,const RootInfo & info)2961   void VisitRoot(mirror::Object* root, const RootInfo& info)
2962       override REQUIRES_SHARED(Locks::mutator_lock_) {
2963     if (root == obj_) {
2964       LOG(INFO) << "Object " << obj_ << " is a root " << info.ToString();
2965     }
2966   }
2967 
2968  private:
2969   const mirror::Object* const obj_;
2970 };
2971 
2972 
2973 class ScanVisitor {
2974  public:
operator ()(const mirror::Object * obj) const2975   void operator()(const mirror::Object* obj) const {
2976     LOG(ERROR) << "Would have rescanned object " << obj;
2977   }
2978 };
2979 
2980 // Verify a reference from an object.
2981 class VerifyReferenceVisitor : public SingleRootVisitor {
2982  public:
VerifyReferenceVisitor(Thread * self,Heap * heap,size_t * fail_count,bool verify_referent)2983   VerifyReferenceVisitor(Thread* self, Heap* heap, size_t* fail_count, bool verify_referent)
2984       REQUIRES_SHARED(Locks::mutator_lock_)
2985       : self_(self), heap_(heap), fail_count_(fail_count), verify_referent_(verify_referent) {
2986     CHECK_EQ(self_, Thread::Current());
2987   }
2988 
operator ()(ObjPtr<mirror::Class> klass ATTRIBUTE_UNUSED,ObjPtr<mirror::Reference> ref) const2989   void operator()(ObjPtr<mirror::Class> klass ATTRIBUTE_UNUSED, ObjPtr<mirror::Reference> ref) const
2990       REQUIRES_SHARED(Locks::mutator_lock_) {
2991     if (verify_referent_) {
2992       VerifyReference(ref.Ptr(), ref->GetReferent(), mirror::Reference::ReferentOffset());
2993     }
2994   }
2995 
operator ()(ObjPtr<mirror::Object> obj,MemberOffset offset,bool is_static ATTRIBUTE_UNUSED) const2996   void operator()(ObjPtr<mirror::Object> obj,
2997                   MemberOffset offset,
2998                   bool is_static ATTRIBUTE_UNUSED) const
2999       REQUIRES_SHARED(Locks::mutator_lock_) {
3000     VerifyReference(obj.Ptr(), obj->GetFieldObject<mirror::Object>(offset), offset);
3001   }
3002 
IsLive(ObjPtr<mirror::Object> obj) const3003   bool IsLive(ObjPtr<mirror::Object> obj) const NO_THREAD_SAFETY_ANALYSIS {
3004     return heap_->IsLiveObjectLocked(obj, true, false, true);
3005   }
3006 
VisitRootIfNonNull(mirror::CompressedReference<mirror::Object> * root) const3007   void VisitRootIfNonNull(mirror::CompressedReference<mirror::Object>* root) const
3008       REQUIRES_SHARED(Locks::mutator_lock_) {
3009     if (!root->IsNull()) {
3010       VisitRoot(root);
3011     }
3012   }
VisitRoot(mirror::CompressedReference<mirror::Object> * root) const3013   void VisitRoot(mirror::CompressedReference<mirror::Object>* root) const
3014       REQUIRES_SHARED(Locks::mutator_lock_) {
3015     const_cast<VerifyReferenceVisitor*>(this)->VisitRoot(
3016         root->AsMirrorPtr(), RootInfo(kRootVMInternal));
3017   }
3018 
VisitRoot(mirror::Object * root,const RootInfo & root_info)3019   void VisitRoot(mirror::Object* root, const RootInfo& root_info) override
3020       REQUIRES_SHARED(Locks::mutator_lock_) {
3021     if (root == nullptr) {
3022       LOG(ERROR) << "Root is null with info " << root_info.GetType();
3023     } else if (!VerifyReference(nullptr, root, MemberOffset(0))) {
3024       LOG(ERROR) << "Root " << root << " is dead with type " << mirror::Object::PrettyTypeOf(root)
3025           << " thread_id= " << root_info.GetThreadId() << " root_type= " << root_info.GetType();
3026     }
3027   }
3028 
3029  private:
3030   // TODO: Fix the no thread safety analysis.
3031   // Returns false on failure.
VerifyReference(mirror::Object * obj,mirror::Object * ref,MemberOffset offset) const3032   bool VerifyReference(mirror::Object* obj, mirror::Object* ref, MemberOffset offset) const
3033       NO_THREAD_SAFETY_ANALYSIS {
3034     if (ref == nullptr || IsLive(ref)) {
3035       // Verify that the reference is live.
3036       return true;
3037     }
3038     CHECK_EQ(self_, Thread::Current());  // fail_count_ is private to the calling thread.
3039     *fail_count_ += 1;
3040     if (*fail_count_ == 1) {
3041       // Only print message for the first failure to prevent spam.
3042       LOG(ERROR) << "!!!!!!!!!!!!!!Heap corruption detected!!!!!!!!!!!!!!!!!!!";
3043     }
3044     if (obj != nullptr) {
3045       // Only do this part for non roots.
3046       accounting::CardTable* card_table = heap_->GetCardTable();
3047       accounting::ObjectStack* alloc_stack = heap_->allocation_stack_.get();
3048       accounting::ObjectStack* live_stack = heap_->live_stack_.get();
3049       uint8_t* card_addr = card_table->CardFromAddr(obj);
3050       LOG(ERROR) << "Object " << obj << " references dead object " << ref << " at offset "
3051                  << offset << "\n card value = " << static_cast<int>(*card_addr);
3052       if (heap_->IsValidObjectAddress(obj->GetClass())) {
3053         LOG(ERROR) << "Obj type " << obj->PrettyTypeOf();
3054       } else {
3055         LOG(ERROR) << "Object " << obj << " class(" << obj->GetClass() << ") not a heap address";
3056       }
3057 
3058       // Attempt to find the class inside of the recently freed objects.
3059       space::ContinuousSpace* ref_space = heap_->FindContinuousSpaceFromObject(ref, true);
3060       if (ref_space != nullptr && ref_space->IsMallocSpace()) {
3061         space::MallocSpace* space = ref_space->AsMallocSpace();
3062         mirror::Class* ref_class = space->FindRecentFreedObject(ref);
3063         if (ref_class != nullptr) {
3064           LOG(ERROR) << "Reference " << ref << " found as a recently freed object with class "
3065                      << ref_class->PrettyClass();
3066         } else {
3067           LOG(ERROR) << "Reference " << ref << " not found as a recently freed object";
3068         }
3069       }
3070 
3071       if (ref->GetClass() != nullptr && heap_->IsValidObjectAddress(ref->GetClass()) &&
3072           ref->GetClass()->IsClass()) {
3073         LOG(ERROR) << "Ref type " << ref->PrettyTypeOf();
3074       } else {
3075         LOG(ERROR) << "Ref " << ref << " class(" << ref->GetClass()
3076                    << ") is not a valid heap address";
3077       }
3078 
3079       card_table->CheckAddrIsInCardTable(reinterpret_cast<const uint8_t*>(obj));
3080       void* cover_begin = card_table->AddrFromCard(card_addr);
3081       void* cover_end = reinterpret_cast<void*>(reinterpret_cast<size_t>(cover_begin) +
3082           accounting::CardTable::kCardSize);
3083       LOG(ERROR) << "Card " << reinterpret_cast<void*>(card_addr) << " covers " << cover_begin
3084           << "-" << cover_end;
3085       accounting::ContinuousSpaceBitmap* bitmap =
3086           heap_->GetLiveBitmap()->GetContinuousSpaceBitmap(obj);
3087 
3088       if (bitmap == nullptr) {
3089         LOG(ERROR) << "Object " << obj << " has no bitmap";
3090         if (!VerifyClassClass(obj->GetClass())) {
3091           LOG(ERROR) << "Object " << obj << " failed class verification!";
3092         }
3093       } else {
3094         // Print out how the object is live.
3095         if (bitmap->Test(obj)) {
3096           LOG(ERROR) << "Object " << obj << " found in live bitmap";
3097         }
3098         if (alloc_stack->Contains(const_cast<mirror::Object*>(obj))) {
3099           LOG(ERROR) << "Object " << obj << " found in allocation stack";
3100         }
3101         if (live_stack->Contains(const_cast<mirror::Object*>(obj))) {
3102           LOG(ERROR) << "Object " << obj << " found in live stack";
3103         }
3104         if (alloc_stack->Contains(const_cast<mirror::Object*>(ref))) {
3105           LOG(ERROR) << "Ref " << ref << " found in allocation stack";
3106         }
3107         if (live_stack->Contains(const_cast<mirror::Object*>(ref))) {
3108           LOG(ERROR) << "Ref " << ref << " found in live stack";
3109         }
3110         // Attempt to see if the card table missed the reference.
3111         ScanVisitor scan_visitor;
3112         uint8_t* byte_cover_begin = reinterpret_cast<uint8_t*>(card_table->AddrFromCard(card_addr));
3113         card_table->Scan<false>(bitmap, byte_cover_begin,
3114                                 byte_cover_begin + accounting::CardTable::kCardSize, scan_visitor);
3115       }
3116 
3117       // Search to see if any of the roots reference our object.
3118       RootMatchesObjectVisitor visitor1(obj);
3119       Runtime::Current()->VisitRoots(&visitor1);
3120       // Search to see if any of the roots reference our reference.
3121       RootMatchesObjectVisitor visitor2(ref);
3122       Runtime::Current()->VisitRoots(&visitor2);
3123     }
3124     return false;
3125   }
3126 
3127   Thread* const self_;
3128   Heap* const heap_;
3129   size_t* const fail_count_;
3130   const bool verify_referent_;
3131 };
3132 
3133 // Verify all references within an object, for use with HeapBitmap::Visit.
3134 class VerifyObjectVisitor {
3135  public:
VerifyObjectVisitor(Thread * self,Heap * heap,size_t * fail_count,bool verify_referent)3136   VerifyObjectVisitor(Thread* self, Heap* heap, size_t* fail_count, bool verify_referent)
3137       : self_(self), heap_(heap), fail_count_(fail_count), verify_referent_(verify_referent) {}
3138 
operator ()(mirror::Object * obj)3139   void operator()(mirror::Object* obj) REQUIRES_SHARED(Locks::mutator_lock_) {
3140     // Note: we are verifying the references in obj but not obj itself, this is because obj must
3141     // be live or else how did we find it in the live bitmap?
3142     VerifyReferenceVisitor visitor(self_, heap_, fail_count_, verify_referent_);
3143     // The class doesn't count as a reference but we should verify it anyways.
3144     obj->VisitReferences(visitor, visitor);
3145   }
3146 
VerifyRoots()3147   void VerifyRoots() REQUIRES_SHARED(Locks::mutator_lock_) REQUIRES(!Locks::heap_bitmap_lock_) {
3148     ReaderMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
3149     VerifyReferenceVisitor visitor(self_, heap_, fail_count_, verify_referent_);
3150     Runtime::Current()->VisitRoots(&visitor);
3151   }
3152 
GetFailureCount() const3153   uint32_t GetFailureCount() const REQUIRES(Locks::mutator_lock_) {
3154     CHECK_EQ(self_, Thread::Current());
3155     return *fail_count_;
3156   }
3157 
3158  private:
3159   Thread* const self_;
3160   Heap* const heap_;
3161   size_t* const fail_count_;
3162   const bool verify_referent_;
3163 };
3164 
PushOnAllocationStackWithInternalGC(Thread * self,ObjPtr<mirror::Object> * obj)3165 void Heap::PushOnAllocationStackWithInternalGC(Thread* self, ObjPtr<mirror::Object>* obj) {
3166   // Slow path, the allocation stack push back must have already failed.
3167   DCHECK(!allocation_stack_->AtomicPushBack(obj->Ptr()));
3168   do {
3169     // TODO: Add handle VerifyObject.
3170     StackHandleScope<1> hs(self);
3171     HandleWrapperObjPtr<mirror::Object> wrapper(hs.NewHandleWrapper(obj));
3172     // Push our object into the reserve region of the allocation stack. This is only required due
3173     // to heap verification requiring that roots are live (either in the live bitmap or in the
3174     // allocation stack).
3175     CHECK(allocation_stack_->AtomicPushBackIgnoreGrowthLimit(obj->Ptr()));
3176     CollectGarbageInternal(collector::kGcTypeSticky,
3177                            kGcCauseForAlloc,
3178                            false,
3179                            GetCurrentGcNum() + 1);
3180   } while (!allocation_stack_->AtomicPushBack(obj->Ptr()));
3181 }
3182 
PushOnThreadLocalAllocationStackWithInternalGC(Thread * self,ObjPtr<mirror::Object> * obj)3183 void Heap::PushOnThreadLocalAllocationStackWithInternalGC(Thread* self,
3184                                                           ObjPtr<mirror::Object>* obj) {
3185   // Slow path, the allocation stack push back must have already failed.
3186   DCHECK(!self->PushOnThreadLocalAllocationStack(obj->Ptr()));
3187   StackReference<mirror::Object>* start_address;
3188   StackReference<mirror::Object>* end_address;
3189   while (!allocation_stack_->AtomicBumpBack(kThreadLocalAllocationStackSize, &start_address,
3190                                             &end_address)) {
3191     // TODO: Add handle VerifyObject.
3192     StackHandleScope<1> hs(self);
3193     HandleWrapperObjPtr<mirror::Object> wrapper(hs.NewHandleWrapper(obj));
3194     // Push our object into the reserve region of the allocaiton stack. This is only required due
3195     // to heap verification requiring that roots are live (either in the live bitmap or in the
3196     // allocation stack).
3197     CHECK(allocation_stack_->AtomicPushBackIgnoreGrowthLimit(obj->Ptr()));
3198     // Push into the reserve allocation stack.
3199     CollectGarbageInternal(collector::kGcTypeSticky,
3200                            kGcCauseForAlloc,
3201                            false,
3202                            GetCurrentGcNum() + 1);
3203   }
3204   self->SetThreadLocalAllocationStack(start_address, end_address);
3205   // Retry on the new thread-local allocation stack.
3206   CHECK(self->PushOnThreadLocalAllocationStack(obj->Ptr()));  // Must succeed.
3207 }
3208 
3209 // Must do this with mutators suspended since we are directly accessing the allocation stacks.
VerifyHeapReferences(bool verify_referents)3210 size_t Heap::VerifyHeapReferences(bool verify_referents) {
3211   Thread* self = Thread::Current();
3212   Locks::mutator_lock_->AssertExclusiveHeld(self);
3213   // Lets sort our allocation stacks so that we can efficiently binary search them.
3214   allocation_stack_->Sort();
3215   live_stack_->Sort();
3216   // Since we sorted the allocation stack content, need to revoke all
3217   // thread-local allocation stacks.
3218   RevokeAllThreadLocalAllocationStacks(self);
3219   size_t fail_count = 0;
3220   VerifyObjectVisitor visitor(self, this, &fail_count, verify_referents);
3221   // Verify objects in the allocation stack since these will be objects which were:
3222   // 1. Allocated prior to the GC (pre GC verification).
3223   // 2. Allocated during the GC (pre sweep GC verification).
3224   // We don't want to verify the objects in the live stack since they themselves may be
3225   // pointing to dead objects if they are not reachable.
3226   VisitObjectsPaused(visitor);
3227   // Verify the roots:
3228   visitor.VerifyRoots();
3229   if (visitor.GetFailureCount() > 0) {
3230     // Dump mod-union tables.
3231     for (const auto& table_pair : mod_union_tables_) {
3232       accounting::ModUnionTable* mod_union_table = table_pair.second;
3233       mod_union_table->Dump(LOG_STREAM(ERROR) << mod_union_table->GetName() << ": ");
3234     }
3235     // Dump remembered sets.
3236     for (const auto& table_pair : remembered_sets_) {
3237       accounting::RememberedSet* remembered_set = table_pair.second;
3238       remembered_set->Dump(LOG_STREAM(ERROR) << remembered_set->GetName() << ": ");
3239     }
3240     DumpSpaces(LOG_STREAM(ERROR));
3241   }
3242   return visitor.GetFailureCount();
3243 }
3244 
3245 class VerifyReferenceCardVisitor {
3246  public:
VerifyReferenceCardVisitor(Heap * heap,bool * failed)3247   VerifyReferenceCardVisitor(Heap* heap, bool* failed)
3248       REQUIRES_SHARED(Locks::mutator_lock_,
3249                             Locks::heap_bitmap_lock_)
3250       : heap_(heap), failed_(failed) {
3251   }
3252 
3253   // There is no card marks for native roots on a class.
VisitRootIfNonNull(mirror::CompressedReference<mirror::Object> * root ATTRIBUTE_UNUSED) const3254   void VisitRootIfNonNull(mirror::CompressedReference<mirror::Object>* root ATTRIBUTE_UNUSED)
3255       const {}
VisitRoot(mirror::CompressedReference<mirror::Object> * root ATTRIBUTE_UNUSED) const3256   void VisitRoot(mirror::CompressedReference<mirror::Object>* root ATTRIBUTE_UNUSED) const {}
3257 
3258   // TODO: Fix lock analysis to not use NO_THREAD_SAFETY_ANALYSIS, requires support for
3259   // annotalysis on visitors.
operator ()(mirror::Object * obj,MemberOffset offset,bool is_static) const3260   void operator()(mirror::Object* obj, MemberOffset offset, bool is_static) const
3261       NO_THREAD_SAFETY_ANALYSIS {
3262     mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset);
3263     // Filter out class references since changing an object's class does not mark the card as dirty.
3264     // Also handles large objects, since the only reference they hold is a class reference.
3265     if (ref != nullptr && !ref->IsClass()) {
3266       accounting::CardTable* card_table = heap_->GetCardTable();
3267       // If the object is not dirty and it is referencing something in the live stack other than
3268       // class, then it must be on a dirty card.
3269       if (!card_table->AddrIsInCardTable(obj)) {
3270         LOG(ERROR) << "Object " << obj << " is not in the address range of the card table";
3271         *failed_ = true;
3272       } else if (!card_table->IsDirty(obj)) {
3273         // TODO: Check mod-union tables.
3274         // Card should be either kCardDirty if it got re-dirtied after we aged it, or
3275         // kCardDirty - 1 if it didnt get touched since we aged it.
3276         accounting::ObjectStack* live_stack = heap_->live_stack_.get();
3277         if (live_stack->ContainsSorted(ref)) {
3278           if (live_stack->ContainsSorted(obj)) {
3279             LOG(ERROR) << "Object " << obj << " found in live stack";
3280           }
3281           if (heap_->GetLiveBitmap()->Test(obj)) {
3282             LOG(ERROR) << "Object " << obj << " found in live bitmap";
3283           }
3284           LOG(ERROR) << "Object " << obj << " " << mirror::Object::PrettyTypeOf(obj)
3285                     << " references " << ref << " " << mirror::Object::PrettyTypeOf(ref)
3286                     << " in live stack";
3287 
3288           // Print which field of the object is dead.
3289           if (!obj->IsObjectArray()) {
3290             ObjPtr<mirror::Class> klass = is_static ? obj->AsClass() : obj->GetClass();
3291             CHECK(klass != nullptr);
3292             for (ArtField& field : (is_static ? klass->GetSFields() : klass->GetIFields())) {
3293               if (field.GetOffset().Int32Value() == offset.Int32Value()) {
3294                 LOG(ERROR) << (is_static ? "Static " : "") << "field in the live stack is "
3295                            << field.PrettyField();
3296                 break;
3297               }
3298             }
3299           } else {
3300             ObjPtr<mirror::ObjectArray<mirror::Object>> object_array =
3301                 obj->AsObjectArray<mirror::Object>();
3302             for (int32_t i = 0; i < object_array->GetLength(); ++i) {
3303               if (object_array->Get(i) == ref) {
3304                 LOG(ERROR) << (is_static ? "Static " : "") << "obj[" << i << "] = ref";
3305               }
3306             }
3307           }
3308 
3309           *failed_ = true;
3310         }
3311       }
3312     }
3313   }
3314 
3315  private:
3316   Heap* const heap_;
3317   bool* const failed_;
3318 };
3319 
3320 class VerifyLiveStackReferences {
3321  public:
VerifyLiveStackReferences(Heap * heap)3322   explicit VerifyLiveStackReferences(Heap* heap)
3323       : heap_(heap),
3324         failed_(false) {}
3325 
operator ()(mirror::Object * obj) const3326   void operator()(mirror::Object* obj) const
3327       REQUIRES_SHARED(Locks::mutator_lock_, Locks::heap_bitmap_lock_) {
3328     VerifyReferenceCardVisitor visitor(heap_, const_cast<bool*>(&failed_));
3329     obj->VisitReferences(visitor, VoidFunctor());
3330   }
3331 
Failed() const3332   bool Failed() const {
3333     return failed_;
3334   }
3335 
3336  private:
3337   Heap* const heap_;
3338   bool failed_;
3339 };
3340 
VerifyMissingCardMarks()3341 bool Heap::VerifyMissingCardMarks() {
3342   Thread* self = Thread::Current();
3343   Locks::mutator_lock_->AssertExclusiveHeld(self);
3344   // We need to sort the live stack since we binary search it.
3345   live_stack_->Sort();
3346   // Since we sorted the allocation stack content, need to revoke all
3347   // thread-local allocation stacks.
3348   RevokeAllThreadLocalAllocationStacks(self);
3349   VerifyLiveStackReferences visitor(this);
3350   GetLiveBitmap()->Visit(visitor);
3351   // We can verify objects in the live stack since none of these should reference dead objects.
3352   for (auto* it = live_stack_->Begin(); it != live_stack_->End(); ++it) {
3353     if (!kUseThreadLocalAllocationStack || it->AsMirrorPtr() != nullptr) {
3354       visitor(it->AsMirrorPtr());
3355     }
3356   }
3357   return !visitor.Failed();
3358 }
3359 
SwapStacks()3360 void Heap::SwapStacks() {
3361   if (kUseThreadLocalAllocationStack) {
3362     live_stack_->AssertAllZero();
3363   }
3364   allocation_stack_.swap(live_stack_);
3365 }
3366 
RevokeAllThreadLocalAllocationStacks(Thread * self)3367 void Heap::RevokeAllThreadLocalAllocationStacks(Thread* self) {
3368   // This must be called only during the pause.
3369   DCHECK(Locks::mutator_lock_->IsExclusiveHeld(self));
3370   MutexLock mu(self, *Locks::runtime_shutdown_lock_);
3371   MutexLock mu2(self, *Locks::thread_list_lock_);
3372   std::list<Thread*> thread_list = Runtime::Current()->GetThreadList()->GetList();
3373   for (Thread* t : thread_list) {
3374     t->RevokeThreadLocalAllocationStack();
3375   }
3376 }
3377 
AssertThreadLocalBuffersAreRevoked(Thread * thread)3378 void Heap::AssertThreadLocalBuffersAreRevoked(Thread* thread) {
3379   if (kIsDebugBuild) {
3380     if (rosalloc_space_ != nullptr) {
3381       rosalloc_space_->AssertThreadLocalBuffersAreRevoked(thread);
3382     }
3383     if (bump_pointer_space_ != nullptr) {
3384       bump_pointer_space_->AssertThreadLocalBuffersAreRevoked(thread);
3385     }
3386   }
3387 }
3388 
AssertAllBumpPointerSpaceThreadLocalBuffersAreRevoked()3389 void Heap::AssertAllBumpPointerSpaceThreadLocalBuffersAreRevoked() {
3390   if (kIsDebugBuild) {
3391     if (bump_pointer_space_ != nullptr) {
3392       bump_pointer_space_->AssertAllThreadLocalBuffersAreRevoked();
3393     }
3394   }
3395 }
3396 
FindModUnionTableFromSpace(space::Space * space)3397 accounting::ModUnionTable* Heap::FindModUnionTableFromSpace(space::Space* space) {
3398   auto it = mod_union_tables_.find(space);
3399   if (it == mod_union_tables_.end()) {
3400     return nullptr;
3401   }
3402   return it->second;
3403 }
3404 
FindRememberedSetFromSpace(space::Space * space)3405 accounting::RememberedSet* Heap::FindRememberedSetFromSpace(space::Space* space) {
3406   auto it = remembered_sets_.find(space);
3407   if (it == remembered_sets_.end()) {
3408     return nullptr;
3409   }
3410   return it->second;
3411 }
3412 
ProcessCards(TimingLogger * timings,bool use_rem_sets,bool process_alloc_space_cards,bool clear_alloc_space_cards)3413 void Heap::ProcessCards(TimingLogger* timings,
3414                         bool use_rem_sets,
3415                         bool process_alloc_space_cards,
3416                         bool clear_alloc_space_cards) {
3417   TimingLogger::ScopedTiming t(__FUNCTION__, timings);
3418   // Clear cards and keep track of cards cleared in the mod-union table.
3419   for (const auto& space : continuous_spaces_) {
3420     accounting::ModUnionTable* table = FindModUnionTableFromSpace(space);
3421     accounting::RememberedSet* rem_set = FindRememberedSetFromSpace(space);
3422     if (table != nullptr) {
3423       const char* name = space->IsZygoteSpace() ? "ZygoteModUnionClearCards" :
3424           "ImageModUnionClearCards";
3425       TimingLogger::ScopedTiming t2(name, timings);
3426       table->ProcessCards();
3427     } else if (use_rem_sets && rem_set != nullptr) {
3428       DCHECK(collector::SemiSpace::kUseRememberedSet) << static_cast<int>(collector_type_);
3429       TimingLogger::ScopedTiming t2("AllocSpaceRemSetClearCards", timings);
3430       rem_set->ClearCards();
3431     } else if (process_alloc_space_cards) {
3432       TimingLogger::ScopedTiming t2("AllocSpaceClearCards", timings);
3433       if (clear_alloc_space_cards) {
3434         uint8_t* end = space->End();
3435         if (space->IsImageSpace()) {
3436           // Image space end is the end of the mirror objects, it is not necessarily page or card
3437           // aligned. Align up so that the check in ClearCardRange does not fail.
3438           end = AlignUp(end, accounting::CardTable::kCardSize);
3439         }
3440         card_table_->ClearCardRange(space->Begin(), end);
3441       } else {
3442         // No mod union table for the AllocSpace. Age the cards so that the GC knows that these
3443         // cards were dirty before the GC started.
3444         // TODO: Need to use atomic for the case where aged(cleaning thread) -> dirty(other thread)
3445         // -> clean(cleaning thread).
3446         // The races are we either end up with: Aged card, unaged card. Since we have the
3447         // checkpoint roots and then we scan / update mod union tables after. We will always
3448         // scan either card. If we end up with the non aged card, we scan it it in the pause.
3449         card_table_->ModifyCardsAtomic(space->Begin(), space->End(), AgeCardVisitor(),
3450                                        VoidFunctor());
3451       }
3452     }
3453   }
3454 }
3455 
3456 struct IdentityMarkHeapReferenceVisitor : public MarkObjectVisitor {
MarkObjectart::gc::IdentityMarkHeapReferenceVisitor3457   mirror::Object* MarkObject(mirror::Object* obj) override {
3458     return obj;
3459   }
MarkHeapReferenceart::gc::IdentityMarkHeapReferenceVisitor3460   void MarkHeapReference(mirror::HeapReference<mirror::Object>*, bool) override {
3461   }
3462 };
3463 
PreGcVerificationPaused(collector::GarbageCollector * gc)3464 void Heap::PreGcVerificationPaused(collector::GarbageCollector* gc) {
3465   Thread* const self = Thread::Current();
3466   TimingLogger* const timings = current_gc_iteration_.GetTimings();
3467   TimingLogger::ScopedTiming t(__FUNCTION__, timings);
3468   if (verify_pre_gc_heap_) {
3469     TimingLogger::ScopedTiming t2("(Paused)PreGcVerifyHeapReferences", timings);
3470     size_t failures = VerifyHeapReferences();
3471     if (failures > 0) {
3472       LOG(FATAL) << "Pre " << gc->GetName() << " heap verification failed with " << failures
3473           << " failures";
3474     }
3475   }
3476   // Check that all objects which reference things in the live stack are on dirty cards.
3477   if (verify_missing_card_marks_) {
3478     TimingLogger::ScopedTiming t2("(Paused)PreGcVerifyMissingCardMarks", timings);
3479     ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
3480     SwapStacks();
3481     // Sort the live stack so that we can quickly binary search it later.
3482     CHECK(VerifyMissingCardMarks()) << "Pre " << gc->GetName()
3483                                     << " missing card mark verification failed\n" << DumpSpaces();
3484     SwapStacks();
3485   }
3486   if (verify_mod_union_table_) {
3487     TimingLogger::ScopedTiming t2("(Paused)PreGcVerifyModUnionTables", timings);
3488     ReaderMutexLock reader_lock(self, *Locks::heap_bitmap_lock_);
3489     for (const auto& table_pair : mod_union_tables_) {
3490       accounting::ModUnionTable* mod_union_table = table_pair.second;
3491       IdentityMarkHeapReferenceVisitor visitor;
3492       mod_union_table->UpdateAndMarkReferences(&visitor);
3493       mod_union_table->Verify();
3494     }
3495   }
3496 }
3497 
PreGcVerification(collector::GarbageCollector * gc)3498 void Heap::PreGcVerification(collector::GarbageCollector* gc) {
3499   if (verify_pre_gc_heap_ || verify_missing_card_marks_ || verify_mod_union_table_) {
3500     collector::GarbageCollector::ScopedPause pause(gc, false);
3501     PreGcVerificationPaused(gc);
3502   }
3503 }
3504 
PrePauseRosAllocVerification(collector::GarbageCollector * gc ATTRIBUTE_UNUSED)3505 void Heap::PrePauseRosAllocVerification(collector::GarbageCollector* gc ATTRIBUTE_UNUSED) {
3506   // TODO: Add a new runtime option for this?
3507   if (verify_pre_gc_rosalloc_) {
3508     RosAllocVerification(current_gc_iteration_.GetTimings(), "PreGcRosAllocVerification");
3509   }
3510 }
3511 
PreSweepingGcVerification(collector::GarbageCollector * gc)3512 void Heap::PreSweepingGcVerification(collector::GarbageCollector* gc) {
3513   Thread* const self = Thread::Current();
3514   TimingLogger* const timings = current_gc_iteration_.GetTimings();
3515   TimingLogger::ScopedTiming t(__FUNCTION__, timings);
3516   // Called before sweeping occurs since we want to make sure we are not going so reclaim any
3517   // reachable objects.
3518   if (verify_pre_sweeping_heap_) {
3519     TimingLogger::ScopedTiming t2("(Paused)PostSweepingVerifyHeapReferences", timings);
3520     CHECK_NE(self->GetState(), ThreadState::kRunnable);
3521     {
3522       WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
3523       // Swapping bound bitmaps does nothing.
3524       gc->SwapBitmaps();
3525     }
3526     // Pass in false since concurrent reference processing can mean that the reference referents
3527     // may point to dead objects at the point which PreSweepingGcVerification is called.
3528     size_t failures = VerifyHeapReferences(false);
3529     if (failures > 0) {
3530       LOG(FATAL) << "Pre sweeping " << gc->GetName() << " GC verification failed with " << failures
3531           << " failures";
3532     }
3533     {
3534       WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
3535       gc->SwapBitmaps();
3536     }
3537   }
3538   if (verify_pre_sweeping_rosalloc_) {
3539     RosAllocVerification(timings, "PreSweepingRosAllocVerification");
3540   }
3541 }
3542 
PostGcVerificationPaused(collector::GarbageCollector * gc)3543 void Heap::PostGcVerificationPaused(collector::GarbageCollector* gc) {
3544   // Only pause if we have to do some verification.
3545   Thread* const self = Thread::Current();
3546   TimingLogger* const timings = GetCurrentGcIteration()->GetTimings();
3547   TimingLogger::ScopedTiming t(__FUNCTION__, timings);
3548   if (verify_system_weaks_) {
3549     ReaderMutexLock mu2(self, *Locks::heap_bitmap_lock_);
3550     collector::MarkSweep* mark_sweep = down_cast<collector::MarkSweep*>(gc);
3551     mark_sweep->VerifySystemWeaks();
3552   }
3553   if (verify_post_gc_rosalloc_) {
3554     RosAllocVerification(timings, "(Paused)PostGcRosAllocVerification");
3555   }
3556   if (verify_post_gc_heap_) {
3557     TimingLogger::ScopedTiming t2("(Paused)PostGcVerifyHeapReferences", timings);
3558     size_t failures = VerifyHeapReferences();
3559     if (failures > 0) {
3560       LOG(FATAL) << "Pre " << gc->GetName() << " heap verification failed with " << failures
3561           << " failures";
3562     }
3563   }
3564 }
3565 
PostGcVerification(collector::GarbageCollector * gc)3566 void Heap::PostGcVerification(collector::GarbageCollector* gc) {
3567   if (verify_system_weaks_ || verify_post_gc_rosalloc_ || verify_post_gc_heap_) {
3568     collector::GarbageCollector::ScopedPause pause(gc, false);
3569     PostGcVerificationPaused(gc);
3570   }
3571 }
3572 
RosAllocVerification(TimingLogger * timings,const char * name)3573 void Heap::RosAllocVerification(TimingLogger* timings, const char* name) {
3574   TimingLogger::ScopedTiming t(name, timings);
3575   for (const auto& space : continuous_spaces_) {
3576     if (space->IsRosAllocSpace()) {
3577       VLOG(heap) << name << " : " << space->GetName();
3578       space->AsRosAllocSpace()->Verify();
3579     }
3580   }
3581 }
3582 
WaitForGcToComplete(GcCause cause,Thread * self)3583 collector::GcType Heap::WaitForGcToComplete(GcCause cause, Thread* self) {
3584   ScopedThreadStateChange tsc(self, ThreadState::kWaitingForGcToComplete);
3585   MutexLock mu(self, *gc_complete_lock_);
3586   return WaitForGcToCompleteLocked(cause, self);
3587 }
3588 
WaitForGcToCompleteLocked(GcCause cause,Thread * self)3589 collector::GcType Heap::WaitForGcToCompleteLocked(GcCause cause, Thread* self) {
3590   gc_complete_cond_->CheckSafeToWait(self);
3591   collector::GcType last_gc_type = collector::kGcTypeNone;
3592   GcCause last_gc_cause = kGcCauseNone;
3593   uint64_t wait_start = NanoTime();
3594   while (collector_type_running_ != kCollectorTypeNone) {
3595     if (self != task_processor_->GetRunningThread()) {
3596       // The current thread is about to wait for a currently running
3597       // collection to finish. If the waiting thread is not the heap
3598       // task daemon thread, the currently running collection is
3599       // considered as a blocking GC.
3600       running_collection_is_blocking_ = true;
3601       VLOG(gc) << "Waiting for a blocking GC " << cause;
3602     }
3603     SCOPED_TRACE << "GC: Wait For Completion " << cause;
3604     // We must wait, change thread state then sleep on gc_complete_cond_;
3605     gc_complete_cond_->Wait(self);
3606     last_gc_type = last_gc_type_;
3607     last_gc_cause = last_gc_cause_;
3608   }
3609   uint64_t wait_time = NanoTime() - wait_start;
3610   total_wait_time_ += wait_time;
3611   if (wait_time > long_pause_log_threshold_) {
3612     LOG(INFO) << "WaitForGcToComplete blocked " << cause << " on " << last_gc_cause << " for "
3613               << PrettyDuration(wait_time);
3614   }
3615   if (self != task_processor_->GetRunningThread()) {
3616     // The current thread is about to run a collection. If the thread
3617     // is not the heap task daemon thread, it's considered as a
3618     // blocking GC (i.e., blocking itself).
3619     running_collection_is_blocking_ = true;
3620     // Don't log fake "GC" types that are only used for debugger or hidden APIs. If we log these,
3621     // it results in log spam. kGcCauseExplicit is already logged in LogGC, so avoid it here too.
3622     if (cause == kGcCauseForAlloc ||
3623         cause == kGcCauseDisableMovingGc) {
3624       VLOG(gc) << "Starting a blocking GC " << cause;
3625     }
3626   }
3627   return last_gc_type;
3628 }
3629 
DumpForSigQuit(std::ostream & os)3630 void Heap::DumpForSigQuit(std::ostream& os) {
3631   os << "Heap: " << GetPercentFree() << "% free, " << PrettySize(GetBytesAllocated()) << "/"
3632      << PrettySize(GetTotalMemory()) << "; " << GetObjectsAllocated() << " objects\n";
3633   {
3634     os << "Image spaces:\n";
3635     ScopedObjectAccess soa(Thread::Current());
3636     for (const auto& space : continuous_spaces_) {
3637       if (space->IsImageSpace()) {
3638         os << space->GetName() << "\n";
3639       }
3640     }
3641   }
3642   DumpGcPerformanceInfo(os);
3643 }
3644 
GetPercentFree()3645 size_t Heap::GetPercentFree() {
3646   return static_cast<size_t>(100.0f * static_cast<float>(
3647       GetFreeMemory()) / target_footprint_.load(std::memory_order_relaxed));
3648 }
3649 
SetIdealFootprint(size_t target_footprint)3650 void Heap::SetIdealFootprint(size_t target_footprint) {
3651   if (target_footprint > GetMaxMemory()) {
3652     VLOG(gc) << "Clamp target GC heap from " << PrettySize(target_footprint) << " to "
3653              << PrettySize(GetMaxMemory());
3654     target_footprint = GetMaxMemory();
3655   }
3656   target_footprint_.store(target_footprint, std::memory_order_relaxed);
3657 }
3658 
IsMovableObject(ObjPtr<mirror::Object> obj) const3659 bool Heap::IsMovableObject(ObjPtr<mirror::Object> obj) const {
3660   if (kMovingCollector) {
3661     space::Space* space = FindContinuousSpaceFromObject(obj.Ptr(), true);
3662     if (space != nullptr) {
3663       // TODO: Check large object?
3664       return space->CanMoveObjects();
3665     }
3666   }
3667   return false;
3668 }
3669 
FindCollectorByGcType(collector::GcType gc_type)3670 collector::GarbageCollector* Heap::FindCollectorByGcType(collector::GcType gc_type) {
3671   for (auto* collector : garbage_collectors_) {
3672     if (collector->GetCollectorType() == collector_type_ &&
3673         collector->GetGcType() == gc_type) {
3674       return collector;
3675     }
3676   }
3677   return nullptr;
3678 }
3679 
HeapGrowthMultiplier() const3680 double Heap::HeapGrowthMultiplier() const {
3681   // If we don't care about pause times we are background, so return 1.0.
3682   if (!CareAboutPauseTimes()) {
3683     return 1.0;
3684   }
3685   return foreground_heap_growth_multiplier_;
3686 }
3687 
GrowForUtilization(collector::GarbageCollector * collector_ran,size_t bytes_allocated_before_gc)3688 void Heap::GrowForUtilization(collector::GarbageCollector* collector_ran,
3689                               size_t bytes_allocated_before_gc) {
3690   // We're running in the thread that set collector_type_running_ to something other than none,
3691   // thus ensuring that there is only one of us running. Thus
3692   // collector_type_running_ != kCollectorTypeNone, but that's a little tricky to turn into a
3693   // DCHECK.
3694 
3695   // We know what our utilization is at this moment.
3696   // This doesn't actually resize any memory. It just lets the heap grow more when necessary.
3697   const size_t bytes_allocated = GetBytesAllocated();
3698   // Trace the new heap size after the GC is finished.
3699   TraceHeapSize(bytes_allocated);
3700   uint64_t target_size, grow_bytes;
3701   collector::GcType gc_type = collector_ran->GetGcType();
3702   MutexLock mu(Thread::Current(), process_state_update_lock_);
3703   // Use the multiplier to grow more for foreground.
3704   const double multiplier = HeapGrowthMultiplier();
3705   if (gc_type != collector::kGcTypeSticky) {
3706     // Grow the heap for non sticky GC.
3707     uint64_t delta = bytes_allocated * (1.0 / GetTargetHeapUtilization() - 1.0);
3708     DCHECK_LE(delta, std::numeric_limits<size_t>::max()) << "bytes_allocated=" << bytes_allocated
3709         << " target_utilization_=" << target_utilization_;
3710     grow_bytes = std::min(delta, static_cast<uint64_t>(max_free_));
3711     grow_bytes = std::max(grow_bytes, static_cast<uint64_t>(min_free_));
3712     target_size = bytes_allocated + static_cast<uint64_t>(grow_bytes * multiplier);
3713     next_gc_type_ = collector::kGcTypeSticky;
3714   } else {
3715     collector::GcType non_sticky_gc_type = NonStickyGcType();
3716     // Find what the next non sticky collector will be.
3717     collector::GarbageCollector* non_sticky_collector = FindCollectorByGcType(non_sticky_gc_type);
3718     if (use_generational_cc_) {
3719       if (non_sticky_collector == nullptr) {
3720         non_sticky_collector = FindCollectorByGcType(collector::kGcTypePartial);
3721       }
3722       CHECK(non_sticky_collector != nullptr);
3723     }
3724     double sticky_gc_throughput_adjustment = GetStickyGcThroughputAdjustment(use_generational_cc_);
3725 
3726     // If the throughput of the current sticky GC >= throughput of the non sticky collector, then
3727     // do another sticky collection next.
3728     // We also check that the bytes allocated aren't over the target_footprint, or
3729     // concurrent_start_bytes in case of concurrent GCs, in order to prevent a
3730     // pathological case where dead objects which aren't reclaimed by sticky could get accumulated
3731     // if the sticky GC throughput always remained >= the full/partial throughput.
3732     size_t target_footprint = target_footprint_.load(std::memory_order_relaxed);
3733     if (current_gc_iteration_.GetEstimatedThroughput() * sticky_gc_throughput_adjustment >=
3734         non_sticky_collector->GetEstimatedMeanThroughput() &&
3735         non_sticky_collector->NumberOfIterations() > 0 &&
3736         bytes_allocated <= (IsGcConcurrent() ? concurrent_start_bytes_ : target_footprint)) {
3737       next_gc_type_ = collector::kGcTypeSticky;
3738     } else {
3739       next_gc_type_ = non_sticky_gc_type;
3740     }
3741     // If we have freed enough memory, shrink the heap back down.
3742     const size_t adjusted_max_free = static_cast<size_t>(max_free_ * multiplier);
3743     if (bytes_allocated + adjusted_max_free < target_footprint) {
3744       target_size = bytes_allocated + adjusted_max_free;
3745       grow_bytes = max_free_;
3746     } else {
3747       target_size = std::max(bytes_allocated, target_footprint);
3748       // The same whether jank perceptible or not; just avoid the adjustment.
3749       grow_bytes = 0;
3750     }
3751   }
3752   CHECK_LE(target_size, std::numeric_limits<size_t>::max());
3753   if (!ignore_target_footprint_) {
3754     SetIdealFootprint(target_size);
3755     // Store target size (computed with foreground heap growth multiplier) for updating
3756     // target_footprint_ when process state switches to foreground.
3757     // target_size = 0 ensures that target_footprint_ is not updated on
3758     // process-state switch.
3759     min_foreground_target_footprint_ =
3760         (multiplier <= 1.0 && grow_bytes > 0)
3761         ? std::min(
3762           bytes_allocated + static_cast<size_t>(grow_bytes * foreground_heap_growth_multiplier_),
3763           GetMaxMemory())
3764         : 0;
3765 
3766     if (IsGcConcurrent()) {
3767       const uint64_t freed_bytes = current_gc_iteration_.GetFreedBytes() +
3768           current_gc_iteration_.GetFreedLargeObjectBytes() +
3769           current_gc_iteration_.GetFreedRevokeBytes();
3770       // Records the number of bytes allocated at the time of GC finish,excluding the number of
3771       // bytes allocated during GC.
3772       num_bytes_alive_after_gc_ = UnsignedDifference(bytes_allocated_before_gc, freed_bytes);
3773       // Bytes allocated will shrink by freed_bytes after the GC runs, so if we want to figure out
3774       // how many bytes were allocated during the GC we need to add freed_bytes back on.
3775       // Almost always bytes_allocated + freed_bytes >= bytes_allocated_before_gc.
3776       const size_t bytes_allocated_during_gc =
3777           UnsignedDifference(bytes_allocated + freed_bytes, bytes_allocated_before_gc);
3778       // Calculate when to perform the next ConcurrentGC.
3779       // Estimate how many remaining bytes we will have when we need to start the next GC.
3780       size_t remaining_bytes = bytes_allocated_during_gc;
3781       remaining_bytes = std::min(remaining_bytes, kMaxConcurrentRemainingBytes);
3782       remaining_bytes = std::max(remaining_bytes, kMinConcurrentRemainingBytes);
3783       size_t target_footprint = target_footprint_.load(std::memory_order_relaxed);
3784       if (UNLIKELY(remaining_bytes > target_footprint)) {
3785         // A never going to happen situation that from the estimated allocation rate we will exceed
3786         // the applications entire footprint with the given estimated allocation rate. Schedule
3787         // another GC nearly straight away.
3788         remaining_bytes = std::min(kMinConcurrentRemainingBytes, target_footprint);
3789       }
3790       DCHECK_LE(target_footprint_.load(std::memory_order_relaxed), GetMaxMemory());
3791       // Start a concurrent GC when we get close to the estimated remaining bytes. When the
3792       // allocation rate is very high, remaining_bytes could tell us that we should start a GC
3793       // right away.
3794       concurrent_start_bytes_ = std::max(target_footprint - remaining_bytes, bytes_allocated);
3795       // Store concurrent_start_bytes_ (computed with foreground heap growth multiplier) for update
3796       // itself when process state switches to foreground.
3797       min_foreground_concurrent_start_bytes_ =
3798           min_foreground_target_footprint_ != 0
3799           ? std::max(min_foreground_target_footprint_ - remaining_bytes, bytes_allocated)
3800           : 0;
3801     }
3802   }
3803 }
3804 
ClampGrowthLimit()3805 void Heap::ClampGrowthLimit() {
3806   // Use heap bitmap lock to guard against races with BindLiveToMarkBitmap.
3807   ScopedObjectAccess soa(Thread::Current());
3808   WriterMutexLock mu(soa.Self(), *Locks::heap_bitmap_lock_);
3809   capacity_ = growth_limit_;
3810   for (const auto& space : continuous_spaces_) {
3811     if (space->IsMallocSpace()) {
3812       gc::space::MallocSpace* malloc_space = space->AsMallocSpace();
3813       malloc_space->ClampGrowthLimit();
3814     }
3815   }
3816   if (collector_type_ == kCollectorTypeCC) {
3817     DCHECK(region_space_ != nullptr);
3818     // Twice the capacity as CC needs extra space for evacuating objects.
3819     region_space_->ClampGrowthLimit(2 * capacity_);
3820   }
3821   // This space isn't added for performance reasons.
3822   if (main_space_backup_.get() != nullptr) {
3823     main_space_backup_->ClampGrowthLimit();
3824   }
3825 }
3826 
ClearGrowthLimit()3827 void Heap::ClearGrowthLimit() {
3828   if (target_footprint_.load(std::memory_order_relaxed) == growth_limit_
3829       && growth_limit_ < capacity_) {
3830     target_footprint_.store(capacity_, std::memory_order_relaxed);
3831     SetDefaultConcurrentStartBytes();
3832   }
3833   growth_limit_ = capacity_;
3834   ScopedObjectAccess soa(Thread::Current());
3835   for (const auto& space : continuous_spaces_) {
3836     if (space->IsMallocSpace()) {
3837       gc::space::MallocSpace* malloc_space = space->AsMallocSpace();
3838       malloc_space->ClearGrowthLimit();
3839       malloc_space->SetFootprintLimit(malloc_space->Capacity());
3840     }
3841   }
3842   // This space isn't added for performance reasons.
3843   if (main_space_backup_.get() != nullptr) {
3844     main_space_backup_->ClearGrowthLimit();
3845     main_space_backup_->SetFootprintLimit(main_space_backup_->Capacity());
3846   }
3847 }
3848 
AddFinalizerReference(Thread * self,ObjPtr<mirror::Object> * object)3849 void Heap::AddFinalizerReference(Thread* self, ObjPtr<mirror::Object>* object) {
3850   ScopedObjectAccess soa(self);
3851   StackHandleScope<1u> hs(self);
3852   // Use handle wrapper to update the `*object` if the object gets moved.
3853   HandleWrapperObjPtr<mirror::Object> h_object = hs.NewHandleWrapper(object);
3854   WellKnownClasses::java_lang_ref_FinalizerReference_add->InvokeStatic<'V', 'L'>(
3855       self, h_object.Get());
3856 }
3857 
RequestConcurrentGCAndSaveObject(Thread * self,bool force_full,uint32_t observed_gc_num,ObjPtr<mirror::Object> * obj)3858 void Heap::RequestConcurrentGCAndSaveObject(Thread* self,
3859                                             bool force_full,
3860                                             uint32_t observed_gc_num,
3861                                             ObjPtr<mirror::Object>* obj) {
3862   StackHandleScope<1> hs(self);
3863   HandleWrapperObjPtr<mirror::Object> wrapper(hs.NewHandleWrapper(obj));
3864   RequestConcurrentGC(self, kGcCauseBackground, force_full, observed_gc_num);
3865 }
3866 
3867 class Heap::ConcurrentGCTask : public HeapTask {
3868  public:
ConcurrentGCTask(uint64_t target_time,GcCause cause,bool force_full,uint32_t gc_num)3869   ConcurrentGCTask(uint64_t target_time, GcCause cause, bool force_full, uint32_t gc_num)
3870       : HeapTask(target_time), cause_(cause), force_full_(force_full), my_gc_num_(gc_num) {}
Run(Thread * self)3871   void Run(Thread* self) override {
3872     Runtime* runtime = Runtime::Current();
3873     gc::Heap* heap = runtime->GetHeap();
3874     DCHECK(GCNumberLt(my_gc_num_, heap->GetCurrentGcNum() + 2));  // <= current_gc_num + 1
3875     heap->ConcurrentGC(self, cause_, force_full_, my_gc_num_);
3876     CHECK_IMPLIES(GCNumberLt(heap->GetCurrentGcNum(), my_gc_num_), runtime->IsShuttingDown(self));
3877   }
3878 
3879  private:
3880   const GcCause cause_;
3881   const bool force_full_;  // If true, force full (or partial) collection.
3882   const uint32_t my_gc_num_;  // Sequence number of requested GC.
3883 };
3884 
CanAddHeapTask(Thread * self)3885 static bool CanAddHeapTask(Thread* self) REQUIRES(!Locks::runtime_shutdown_lock_) {
3886   Runtime* runtime = Runtime::Current();
3887   return runtime != nullptr && runtime->IsFinishedStarting() && !runtime->IsShuttingDown(self) &&
3888       !self->IsHandlingStackOverflow();
3889 }
3890 
RequestConcurrentGC(Thread * self,GcCause cause,bool force_full,uint32_t observed_gc_num)3891 bool Heap::RequestConcurrentGC(Thread* self,
3892                                GcCause cause,
3893                                bool force_full,
3894                                uint32_t observed_gc_num) {
3895   uint32_t max_gc_requested = max_gc_requested_.load(std::memory_order_relaxed);
3896   if (!GCNumberLt(observed_gc_num, max_gc_requested)) {
3897     // observed_gc_num >= max_gc_requested: Nobody beat us to requesting the next gc.
3898     if (CanAddHeapTask(self)) {
3899       // Since observed_gc_num >= max_gc_requested, this increases max_gc_requested_, if successful.
3900       if (max_gc_requested_.CompareAndSetStrongRelaxed(max_gc_requested, observed_gc_num + 1)) {
3901         task_processor_->AddTask(self, new ConcurrentGCTask(NanoTime(),  // Start straight away.
3902                                                             cause,
3903                                                             force_full,
3904                                                             observed_gc_num + 1));
3905       }
3906       DCHECK(GCNumberLt(observed_gc_num, max_gc_requested_.load(std::memory_order_relaxed)));
3907       // If we increased max_gc_requested_, then we added a task that will eventually cause
3908       // gcs_completed_ to be incremented (to at least observed_gc_num + 1).
3909       // If the CAS failed, somebody else did.
3910       return true;
3911     }
3912     return false;
3913   }
3914   return true;  // Vacuously.
3915 }
3916 
ConcurrentGC(Thread * self,GcCause cause,bool force_full,uint32_t requested_gc_num)3917 void Heap::ConcurrentGC(Thread* self, GcCause cause, bool force_full, uint32_t requested_gc_num) {
3918   if (!Runtime::Current()->IsShuttingDown(self)) {
3919     // Wait for any GCs currently running to finish. If this incremented GC number, we're done.
3920     WaitForGcToComplete(cause, self);
3921     if (GCNumberLt(GetCurrentGcNum(), requested_gc_num)) {
3922       collector::GcType next_gc_type = next_gc_type_;
3923       // If forcing full and next gc type is sticky, override with a non-sticky type.
3924       if (force_full && next_gc_type == collector::kGcTypeSticky) {
3925         next_gc_type = NonStickyGcType();
3926       }
3927       // If we can't run the GC type we wanted to run, find the next appropriate one and try
3928       // that instead. E.g. can't do partial, so do full instead.
3929       // We must ensure that we run something that ends up incrementing gcs_completed_.
3930       // In the kGcTypePartial case, the initial CollectGarbageInternal call may not have that
3931       // effect, but the subsequent KGcTypeFull call will.
3932       if (CollectGarbageInternal(next_gc_type, cause, false, requested_gc_num)
3933           == collector::kGcTypeNone) {
3934         for (collector::GcType gc_type : gc_plan_) {
3935           if (!GCNumberLt(GetCurrentGcNum(), requested_gc_num)) {
3936             // Somebody did it for us.
3937             break;
3938           }
3939           // Attempt to run the collector, if we succeed, we are done.
3940           if (gc_type > next_gc_type &&
3941               CollectGarbageInternal(gc_type, cause, false, requested_gc_num)
3942               != collector::kGcTypeNone) {
3943             break;
3944           }
3945         }
3946       }
3947     }
3948   }
3949 }
3950 
3951 class Heap::CollectorTransitionTask : public HeapTask {
3952  public:
CollectorTransitionTask(uint64_t target_time)3953   explicit CollectorTransitionTask(uint64_t target_time) : HeapTask(target_time) {}
3954 
Run(Thread * self)3955   void Run(Thread* self) override {
3956     gc::Heap* heap = Runtime::Current()->GetHeap();
3957     heap->DoPendingCollectorTransition();
3958     heap->ClearPendingCollectorTransition(self);
3959   }
3960 };
3961 
ClearPendingCollectorTransition(Thread * self)3962 void Heap::ClearPendingCollectorTransition(Thread* self) {
3963   MutexLock mu(self, *pending_task_lock_);
3964   pending_collector_transition_ = nullptr;
3965 }
3966 
RequestCollectorTransition(CollectorType desired_collector_type,uint64_t delta_time)3967 void Heap::RequestCollectorTransition(CollectorType desired_collector_type, uint64_t delta_time) {
3968   Thread* self = Thread::Current();
3969   desired_collector_type_ = desired_collector_type;
3970   if (desired_collector_type_ == collector_type_ || !CanAddHeapTask(self)) {
3971     return;
3972   }
3973   if (collector_type_ == kCollectorTypeCC) {
3974     // For CC, we invoke a full compaction when going to the background, but the collector type
3975     // doesn't change.
3976     DCHECK_EQ(desired_collector_type_, kCollectorTypeCCBackground);
3977   }
3978   DCHECK_NE(collector_type_, kCollectorTypeCCBackground);
3979   CollectorTransitionTask* added_task = nullptr;
3980   const uint64_t target_time = NanoTime() + delta_time;
3981   {
3982     MutexLock mu(self, *pending_task_lock_);
3983     // If we have an existing collector transition, update the target time to be the new target.
3984     if (pending_collector_transition_ != nullptr) {
3985       task_processor_->UpdateTargetRunTime(self, pending_collector_transition_, target_time);
3986       return;
3987     }
3988     added_task = new CollectorTransitionTask(target_time);
3989     pending_collector_transition_ = added_task;
3990   }
3991   task_processor_->AddTask(self, added_task);
3992 }
3993 
3994 class Heap::HeapTrimTask : public HeapTask {
3995  public:
HeapTrimTask(uint64_t delta_time)3996   explicit HeapTrimTask(uint64_t delta_time) : HeapTask(NanoTime() + delta_time) { }
Run(Thread * self)3997   void Run(Thread* self) override {
3998     gc::Heap* heap = Runtime::Current()->GetHeap();
3999     heap->Trim(self);
4000     heap->ClearPendingTrim(self);
4001   }
4002 };
4003 
ClearPendingTrim(Thread * self)4004 void Heap::ClearPendingTrim(Thread* self) {
4005   MutexLock mu(self, *pending_task_lock_);
4006   pending_heap_trim_ = nullptr;
4007 }
4008 
RequestTrim(Thread * self)4009 void Heap::RequestTrim(Thread* self) {
4010   if (!CanAddHeapTask(self)) {
4011     return;
4012   }
4013   // GC completed and now we must decide whether to request a heap trim (advising pages back to the
4014   // kernel) or not. Issuing a request will also cause trimming of the libc heap. As a trim scans
4015   // a space it will hold its lock and can become a cause of jank.
4016   // Note, the large object space self trims and the Zygote space was trimmed and unchanging since
4017   // forking.
4018 
4019   // We don't have a good measure of how worthwhile a trim might be. We can't use the live bitmap
4020   // because that only marks object heads, so a large array looks like lots of empty space. We
4021   // don't just call dlmalloc all the time, because the cost of an _attempted_ trim is proportional
4022   // to utilization (which is probably inversely proportional to how much benefit we can expect).
4023   // We could try mincore(2) but that's only a measure of how many pages we haven't given away,
4024   // not how much use we're making of those pages.
4025   HeapTrimTask* added_task = nullptr;
4026   {
4027     MutexLock mu(self, *pending_task_lock_);
4028     if (pending_heap_trim_ != nullptr) {
4029       // Already have a heap trim request in task processor, ignore this request.
4030       return;
4031     }
4032     added_task = new HeapTrimTask(kHeapTrimWait);
4033     pending_heap_trim_ = added_task;
4034   }
4035   task_processor_->AddTask(self, added_task);
4036 }
4037 
IncrementNumberOfBytesFreedRevoke(size_t freed_bytes_revoke)4038 void Heap::IncrementNumberOfBytesFreedRevoke(size_t freed_bytes_revoke) {
4039   size_t previous_num_bytes_freed_revoke =
4040       num_bytes_freed_revoke_.fetch_add(freed_bytes_revoke, std::memory_order_relaxed);
4041   // Check the updated value is less than the number of bytes allocated. There is a risk of
4042   // execution being suspended between the increment above and the CHECK below, leading to
4043   // the use of previous_num_bytes_freed_revoke in the comparison.
4044   CHECK_GE(num_bytes_allocated_.load(std::memory_order_relaxed),
4045            previous_num_bytes_freed_revoke + freed_bytes_revoke);
4046 }
4047 
RevokeThreadLocalBuffers(Thread * thread)4048 void Heap::RevokeThreadLocalBuffers(Thread* thread) {
4049   if (rosalloc_space_ != nullptr) {
4050     size_t freed_bytes_revoke = rosalloc_space_->RevokeThreadLocalBuffers(thread);
4051     if (freed_bytes_revoke > 0U) {
4052       IncrementNumberOfBytesFreedRevoke(freed_bytes_revoke);
4053     }
4054   }
4055   if (bump_pointer_space_ != nullptr) {
4056     CHECK_EQ(bump_pointer_space_->RevokeThreadLocalBuffers(thread), 0U);
4057   }
4058   if (region_space_ != nullptr) {
4059     CHECK_EQ(region_space_->RevokeThreadLocalBuffers(thread), 0U);
4060   }
4061 }
4062 
RevokeRosAllocThreadLocalBuffers(Thread * thread)4063 void Heap::RevokeRosAllocThreadLocalBuffers(Thread* thread) {
4064   if (rosalloc_space_ != nullptr) {
4065     size_t freed_bytes_revoke = rosalloc_space_->RevokeThreadLocalBuffers(thread);
4066     if (freed_bytes_revoke > 0U) {
4067       IncrementNumberOfBytesFreedRevoke(freed_bytes_revoke);
4068     }
4069   }
4070 }
4071 
RevokeAllThreadLocalBuffers()4072 void Heap::RevokeAllThreadLocalBuffers() {
4073   if (rosalloc_space_ != nullptr) {
4074     size_t freed_bytes_revoke = rosalloc_space_->RevokeAllThreadLocalBuffers();
4075     if (freed_bytes_revoke > 0U) {
4076       IncrementNumberOfBytesFreedRevoke(freed_bytes_revoke);
4077     }
4078   }
4079   if (bump_pointer_space_ != nullptr) {
4080     CHECK_EQ(bump_pointer_space_->RevokeAllThreadLocalBuffers(), 0U);
4081   }
4082   if (region_space_ != nullptr) {
4083     CHECK_EQ(region_space_->RevokeAllThreadLocalBuffers(), 0U);
4084   }
4085 }
4086 
4087 // For GC triggering purposes, we count old (pre-last-GC) and new native allocations as
4088 // different fractions of Java allocations.
4089 // For now, we essentially do not count old native allocations at all, so that we can preserve the
4090 // existing behavior of not limiting native heap size. If we seriously considered it, we would
4091 // have to adjust collection thresholds when we encounter large amounts of old native memory,
4092 // and handle native out-of-memory situations.
4093 
4094 static constexpr size_t kOldNativeDiscountFactor = 65536;  // Approximately infinite for now.
4095 static constexpr size_t kNewNativeDiscountFactor = 2;
4096 
4097 // If weighted java + native memory use exceeds our target by kStopForNativeFactor, and
4098 // newly allocated memory exceeds stop_for_native_allocs_, we wait for GC to complete to avoid
4099 // running out of memory.
4100 static constexpr float kStopForNativeFactor = 4.0;
4101 
4102 // Return the ratio of the weighted native + java allocated bytes to its target value.
4103 // A return value > 1.0 means we should collect. Significantly larger values mean we're falling
4104 // behind.
NativeMemoryOverTarget(size_t current_native_bytes,bool is_gc_concurrent)4105 inline float Heap::NativeMemoryOverTarget(size_t current_native_bytes, bool is_gc_concurrent) {
4106   // Collection check for native allocation. Does not enforce Java heap bounds.
4107   // With adj_start_bytes defined below, effectively checks
4108   // <java bytes allocd> + c1*<old native allocd> + c2*<new native allocd) >= adj_start_bytes,
4109   // where c3 > 1, and currently c1 and c2 are 1 divided by the values defined above.
4110   size_t old_native_bytes = old_native_bytes_allocated_.load(std::memory_order_relaxed);
4111   if (old_native_bytes > current_native_bytes) {
4112     // Net decrease; skip the check, but update old value.
4113     // It's OK to lose an update if two stores race.
4114     old_native_bytes_allocated_.store(current_native_bytes, std::memory_order_relaxed);
4115     return 0.0;
4116   } else {
4117     size_t new_native_bytes = UnsignedDifference(current_native_bytes, old_native_bytes);
4118     size_t weighted_native_bytes = new_native_bytes / kNewNativeDiscountFactor
4119         + old_native_bytes / kOldNativeDiscountFactor;
4120     size_t add_bytes_allowed = static_cast<size_t>(
4121         NativeAllocationGcWatermark() * HeapGrowthMultiplier());
4122     size_t java_gc_start_bytes = is_gc_concurrent
4123         ? concurrent_start_bytes_
4124         : target_footprint_.load(std::memory_order_relaxed);
4125     size_t adj_start_bytes = UnsignedSum(java_gc_start_bytes,
4126                                          add_bytes_allowed / kNewNativeDiscountFactor);
4127     return static_cast<float>(GetBytesAllocated() + weighted_native_bytes)
4128          / static_cast<float>(adj_start_bytes);
4129   }
4130 }
4131 
CheckGCForNative(Thread * self)4132 inline void Heap::CheckGCForNative(Thread* self) {
4133   bool is_gc_concurrent = IsGcConcurrent();
4134   uint32_t starting_gc_num = GetCurrentGcNum();
4135   size_t current_native_bytes = GetNativeBytes();
4136   float gc_urgency = NativeMemoryOverTarget(current_native_bytes, is_gc_concurrent);
4137   if (UNLIKELY(gc_urgency >= 1.0)) {
4138     if (is_gc_concurrent) {
4139       bool requested =
4140           RequestConcurrentGC(self, kGcCauseForNativeAlloc, /*force_full=*/true, starting_gc_num);
4141       if (requested && gc_urgency > kStopForNativeFactor
4142           && current_native_bytes > stop_for_native_allocs_) {
4143         // We're in danger of running out of memory due to rampant native allocation.
4144         if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
4145           LOG(INFO) << "Stopping for native allocation, urgency: " << gc_urgency;
4146         }
4147         // Count how many times we do this, so we can warn if this becomes excessive.
4148         // Stop after a while, out of excessive caution.
4149         static constexpr int kGcWaitIters = 20;
4150         for (int i = 1; i <= kGcWaitIters; ++i) {
4151           if (!GCNumberLt(GetCurrentGcNum(), max_gc_requested_.load(std::memory_order_relaxed))
4152               || WaitForGcToComplete(kGcCauseForNativeAlloc, self) != collector::kGcTypeNone) {
4153             break;
4154           }
4155           CHECK(GCNumberLt(starting_gc_num, max_gc_requested_.load(std::memory_order_relaxed)));
4156           if (i % 10 == 0) {
4157             LOG(WARNING) << "Slept " << i << " times in native allocation, waiting for GC";
4158           }
4159           static constexpr int kGcWaitSleepMicros = 2000;
4160           usleep(kGcWaitSleepMicros);  // Encourage our requested GC to start.
4161         }
4162       }
4163     } else {
4164       CollectGarbageInternal(NonStickyGcType(), kGcCauseForNativeAlloc, false, starting_gc_num + 1);
4165     }
4166   }
4167 }
4168 
4169 // About kNotifyNativeInterval allocations have occurred. Check whether we should garbage collect.
NotifyNativeAllocations(JNIEnv * env)4170 void Heap::NotifyNativeAllocations(JNIEnv* env) {
4171   native_objects_notified_.fetch_add(kNotifyNativeInterval, std::memory_order_relaxed);
4172   CheckGCForNative(Thread::ForEnv(env));
4173 }
4174 
4175 // Register a native allocation with an explicit size.
4176 // This should only be done for large allocations of non-malloc memory, which we wouldn't
4177 // otherwise see.
RegisterNativeAllocation(JNIEnv * env,size_t bytes)4178 void Heap::RegisterNativeAllocation(JNIEnv* env, size_t bytes) {
4179   // Cautiously check for a wrapped negative bytes argument.
4180   DCHECK(sizeof(size_t) < 8 || bytes < (std::numeric_limits<size_t>::max() / 2));
4181   native_bytes_registered_.fetch_add(bytes, std::memory_order_relaxed);
4182   uint32_t objects_notified =
4183       native_objects_notified_.fetch_add(1, std::memory_order_relaxed);
4184   if (objects_notified % kNotifyNativeInterval == kNotifyNativeInterval - 1
4185       || bytes > kCheckImmediatelyThreshold) {
4186     CheckGCForNative(Thread::ForEnv(env));
4187   }
4188   // Heap profiler treats this as a Java allocation with a null object.
4189   JHPCheckNonTlabSampleAllocation(Thread::Current(), nullptr, bytes);
4190 }
4191 
RegisterNativeFree(JNIEnv *,size_t bytes)4192 void Heap::RegisterNativeFree(JNIEnv*, size_t bytes) {
4193   size_t allocated;
4194   size_t new_freed_bytes;
4195   do {
4196     allocated = native_bytes_registered_.load(std::memory_order_relaxed);
4197     new_freed_bytes = std::min(allocated, bytes);
4198     // We should not be registering more free than allocated bytes.
4199     // But correctly keep going in non-debug builds.
4200     DCHECK_EQ(new_freed_bytes, bytes);
4201   } while (!native_bytes_registered_.CompareAndSetWeakRelaxed(allocated,
4202                                                               allocated - new_freed_bytes));
4203 }
4204 
GetTotalMemory() const4205 size_t Heap::GetTotalMemory() const {
4206   return std::max(target_footprint_.load(std::memory_order_relaxed), GetBytesAllocated());
4207 }
4208 
AddModUnionTable(accounting::ModUnionTable * mod_union_table)4209 void Heap::AddModUnionTable(accounting::ModUnionTable* mod_union_table) {
4210   DCHECK(mod_union_table != nullptr);
4211   mod_union_tables_.Put(mod_union_table->GetSpace(), mod_union_table);
4212 }
4213 
CheckPreconditionsForAllocObject(ObjPtr<mirror::Class> c,size_t byte_count)4214 void Heap::CheckPreconditionsForAllocObject(ObjPtr<mirror::Class> c, size_t byte_count) {
4215   // Compare rounded sizes since the allocation may have been retried after rounding the size.
4216   // See b/37885600
4217   CHECK(c == nullptr || (c->IsClassClass() && byte_count >= sizeof(mirror::Class)) ||
4218         (c->IsVariableSize() ||
4219             RoundUp(c->GetObjectSize(), kObjectAlignment) ==
4220                 RoundUp(byte_count, kObjectAlignment)))
4221       << "ClassFlags=" << c->GetClassFlags()
4222       << " IsClassClass=" << c->IsClassClass()
4223       << " byte_count=" << byte_count
4224       << " IsVariableSize=" << c->IsVariableSize()
4225       << " ObjectSize=" << c->GetObjectSize()
4226       << " sizeof(Class)=" << sizeof(mirror::Class)
4227       << " " << verification_->DumpObjectInfo(c.Ptr(), /*tag=*/ "klass");
4228   CHECK_GE(byte_count, sizeof(mirror::Object));
4229 }
4230 
AddRememberedSet(accounting::RememberedSet * remembered_set)4231 void Heap::AddRememberedSet(accounting::RememberedSet* remembered_set) {
4232   CHECK(remembered_set != nullptr);
4233   space::Space* space = remembered_set->GetSpace();
4234   CHECK(space != nullptr);
4235   CHECK(remembered_sets_.find(space) == remembered_sets_.end()) << space;
4236   remembered_sets_.Put(space, remembered_set);
4237   CHECK(remembered_sets_.find(space) != remembered_sets_.end()) << space;
4238 }
4239 
RemoveRememberedSet(space::Space * space)4240 void Heap::RemoveRememberedSet(space::Space* space) {
4241   CHECK(space != nullptr);
4242   auto it = remembered_sets_.find(space);
4243   CHECK(it != remembered_sets_.end());
4244   delete it->second;
4245   remembered_sets_.erase(it);
4246   CHECK(remembered_sets_.find(space) == remembered_sets_.end());
4247 }
4248 
ClearMarkedObjects()4249 void Heap::ClearMarkedObjects() {
4250   // Clear all of the spaces' mark bitmaps.
4251   for (const auto& space : GetContinuousSpaces()) {
4252     if (space->GetLiveBitmap() != nullptr && !space->HasBoundBitmaps()) {
4253       space->GetMarkBitmap()->Clear();
4254     }
4255   }
4256   // Clear the marked objects in the discontinous space object sets.
4257   for (const auto& space : GetDiscontinuousSpaces()) {
4258     space->GetMarkBitmap()->Clear();
4259   }
4260 }
4261 
SetAllocationRecords(AllocRecordObjectMap * records)4262 void Heap::SetAllocationRecords(AllocRecordObjectMap* records) {
4263   allocation_records_.reset(records);
4264 }
4265 
VisitAllocationRecords(RootVisitor * visitor) const4266 void Heap::VisitAllocationRecords(RootVisitor* visitor) const {
4267   if (IsAllocTrackingEnabled()) {
4268     MutexLock mu(Thread::Current(), *Locks::alloc_tracker_lock_);
4269     if (IsAllocTrackingEnabled()) {
4270       GetAllocationRecords()->VisitRoots(visitor);
4271     }
4272   }
4273 }
4274 
SweepAllocationRecords(IsMarkedVisitor * visitor) const4275 void Heap::SweepAllocationRecords(IsMarkedVisitor* visitor) const {
4276   if (IsAllocTrackingEnabled()) {
4277     MutexLock mu(Thread::Current(), *Locks::alloc_tracker_lock_);
4278     if (IsAllocTrackingEnabled()) {
4279       GetAllocationRecords()->SweepAllocationRecords(visitor);
4280     }
4281   }
4282 }
4283 
AllowNewAllocationRecords() const4284 void Heap::AllowNewAllocationRecords() const {
4285   CHECK(!gUseReadBarrier);
4286   MutexLock mu(Thread::Current(), *Locks::alloc_tracker_lock_);
4287   AllocRecordObjectMap* allocation_records = GetAllocationRecords();
4288   if (allocation_records != nullptr) {
4289     allocation_records->AllowNewAllocationRecords();
4290   }
4291 }
4292 
DisallowNewAllocationRecords() const4293 void Heap::DisallowNewAllocationRecords() const {
4294   CHECK(!gUseReadBarrier);
4295   MutexLock mu(Thread::Current(), *Locks::alloc_tracker_lock_);
4296   AllocRecordObjectMap* allocation_records = GetAllocationRecords();
4297   if (allocation_records != nullptr) {
4298     allocation_records->DisallowNewAllocationRecords();
4299   }
4300 }
4301 
BroadcastForNewAllocationRecords() const4302 void Heap::BroadcastForNewAllocationRecords() const {
4303   // Always broadcast without checking IsAllocTrackingEnabled() because IsAllocTrackingEnabled() may
4304   // be set to false while some threads are waiting for system weak access in
4305   // AllocRecordObjectMap::RecordAllocation() and we may fail to wake them up. b/27467554.
4306   MutexLock mu(Thread::Current(), *Locks::alloc_tracker_lock_);
4307   AllocRecordObjectMap* allocation_records = GetAllocationRecords();
4308   if (allocation_records != nullptr) {
4309     allocation_records->BroadcastForNewAllocationRecords();
4310   }
4311 }
4312 
4313 // Perfetto Java Heap Profiler Support.
4314 
4315 // Perfetto initialization.
InitPerfettoJavaHeapProf()4316 void Heap::InitPerfettoJavaHeapProf() {
4317   // Initialize Perfetto Heap info and Heap id.
4318   uint32_t heap_id = 1;  // Initialize to 1, to be overwritten by Perfetto heap id.
4319 #ifdef ART_TARGET_ANDROID
4320   // Register the heap and create the heapid.
4321   // Use a Perfetto heap name = "com.android.art" for the Java Heap Profiler.
4322   AHeapInfo* info = AHeapInfo_create("com.android.art");
4323   // Set the Enable Callback, there is no callback data ("nullptr").
4324   AHeapInfo_setEnabledCallback(info, &EnableHeapSamplerCallback, &heap_sampler_);
4325   // Set the Disable Callback.
4326   AHeapInfo_setDisabledCallback(info, &DisableHeapSamplerCallback, &heap_sampler_);
4327   heap_id = AHeapProfile_registerHeap(info);
4328   // Do not enable the Java Heap Profiler in this case, wait for Perfetto to enable it through
4329   // the callback function.
4330 #else
4331   // This is the host case, enable the Java Heap Profiler for host testing.
4332   // Perfetto API is currently not available on host.
4333   heap_sampler_.EnableHeapSampler();
4334 #endif
4335   heap_sampler_.SetHeapID(heap_id);
4336   VLOG(heap) << "Java Heap Profiler Initialized";
4337 }
4338 
4339 // Check if the Java Heap Profiler is enabled and initialized.
CheckPerfettoJHPEnabled()4340 int Heap::CheckPerfettoJHPEnabled() {
4341   return GetHeapSampler().IsEnabled();
4342 }
4343 
JHPCheckNonTlabSampleAllocation(Thread * self,mirror::Object * obj,size_t alloc_size)4344 void Heap::JHPCheckNonTlabSampleAllocation(Thread* self, mirror::Object* obj, size_t alloc_size) {
4345   bool take_sample = false;
4346   size_t bytes_until_sample = 0;
4347   HeapSampler& prof_heap_sampler = GetHeapSampler();
4348   if (prof_heap_sampler.IsEnabled()) {
4349     // An allocation occurred, sample it, even if non-Tlab.
4350     // In case take_sample is already set from the previous GetSampleOffset
4351     // because we tried the Tlab allocation first, we will not use this value.
4352     // A new value is generated below. Also bytes_until_sample will be updated.
4353     // Note that we are not using the return value from the GetSampleOffset in
4354     // the NonTlab case here.
4355     prof_heap_sampler.GetSampleOffset(alloc_size,
4356                                       self->GetTlabPosOffset(),
4357                                       &take_sample,
4358                                       &bytes_until_sample);
4359     prof_heap_sampler.SetBytesUntilSample(bytes_until_sample);
4360     if (take_sample) {
4361       prof_heap_sampler.ReportSample(obj, alloc_size);
4362     }
4363     VLOG(heap) << "JHP:NonTlab Non-moving or Large Allocation or RegisterNativeAllocation";
4364   }
4365 }
4366 
JHPCalculateNextTlabSize(Thread * self,size_t jhp_def_tlab_size,size_t alloc_size,bool * take_sample,size_t * bytes_until_sample)4367 size_t Heap::JHPCalculateNextTlabSize(Thread* self,
4368                                       size_t jhp_def_tlab_size,
4369                                       size_t alloc_size,
4370                                       bool* take_sample,
4371                                       size_t* bytes_until_sample) {
4372   size_t next_tlab_size = jhp_def_tlab_size;
4373   if (CheckPerfettoJHPEnabled()) {
4374     size_t next_sample_point =
4375         GetHeapSampler().GetSampleOffset(alloc_size,
4376                                          self->GetTlabPosOffset(),
4377                                          take_sample,
4378                                          bytes_until_sample);
4379     next_tlab_size = std::min(next_sample_point, jhp_def_tlab_size);
4380   }
4381   return next_tlab_size;
4382 }
4383 
AdjustSampleOffset(size_t adjustment)4384 void Heap::AdjustSampleOffset(size_t adjustment) {
4385   GetHeapSampler().AdjustSampleOffset(adjustment);
4386 }
4387 
CheckGcStressMode(Thread * self,ObjPtr<mirror::Object> * obj)4388 void Heap::CheckGcStressMode(Thread* self, ObjPtr<mirror::Object>* obj) {
4389   DCHECK(gc_stress_mode_);
4390   auto* const runtime = Runtime::Current();
4391   if (runtime->GetClassLinker()->IsInitialized() && !runtime->IsActiveTransaction()) {
4392     // Check if we should GC.
4393     bool new_backtrace = false;
4394     {
4395       static constexpr size_t kMaxFrames = 16u;
4396       MutexLock mu(self, *backtrace_lock_);
4397       FixedSizeBacktrace<kMaxFrames> backtrace;
4398       backtrace.Collect(/* skip_count= */ 2);
4399       uint64_t hash = backtrace.Hash();
4400       new_backtrace = seen_backtraces_.find(hash) == seen_backtraces_.end();
4401       if (new_backtrace) {
4402         seen_backtraces_.insert(hash);
4403       }
4404     }
4405     if (new_backtrace) {
4406       StackHandleScope<1> hs(self);
4407       auto h = hs.NewHandleWrapper(obj);
4408       CollectGarbage(/* clear_soft_references= */ false);
4409       unique_backtrace_count_.fetch_add(1);
4410     } else {
4411       seen_backtrace_count_.fetch_add(1);
4412     }
4413   }
4414 }
4415 
DisableGCForShutdown()4416 void Heap::DisableGCForShutdown() {
4417   MutexLock mu(Thread::Current(), *gc_complete_lock_);
4418   gc_disabled_for_shutdown_ = true;
4419 }
4420 
IsGCDisabledForShutdown() const4421 bool Heap::IsGCDisabledForShutdown() const {
4422   MutexLock mu(Thread::Current(), *gc_complete_lock_);
4423   return gc_disabled_for_shutdown_;
4424 }
4425 
ObjectIsInBootImageSpace(ObjPtr<mirror::Object> obj) const4426 bool Heap::ObjectIsInBootImageSpace(ObjPtr<mirror::Object> obj) const {
4427   DCHECK_EQ(IsBootImageAddress(obj.Ptr()),
4428             any_of(boot_image_spaces_.begin(),
4429                    boot_image_spaces_.end(),
4430                    [obj](gc::space::ImageSpace* space) REQUIRES_SHARED(Locks::mutator_lock_) {
4431                      return space->HasAddress(obj.Ptr());
4432                    }));
4433   return IsBootImageAddress(obj.Ptr());
4434 }
4435 
IsInBootImageOatFile(const void * p) const4436 bool Heap::IsInBootImageOatFile(const void* p) const {
4437   DCHECK_EQ(IsBootImageAddress(p),
4438             any_of(boot_image_spaces_.begin(),
4439                    boot_image_spaces_.end(),
4440                    [p](gc::space::ImageSpace* space) REQUIRES_SHARED(Locks::mutator_lock_) {
4441                      return space->GetOatFile()->Contains(p);
4442                    }));
4443   return IsBootImageAddress(p);
4444 }
4445 
SetAllocationListener(AllocationListener * l)4446 void Heap::SetAllocationListener(AllocationListener* l) {
4447   AllocationListener* old = GetAndOverwriteAllocationListener(&alloc_listener_, l);
4448 
4449   if (old == nullptr) {
4450     Runtime::Current()->GetInstrumentation()->InstrumentQuickAllocEntryPoints();
4451   }
4452 }
4453 
RemoveAllocationListener()4454 void Heap::RemoveAllocationListener() {
4455   AllocationListener* old = GetAndOverwriteAllocationListener(&alloc_listener_, nullptr);
4456 
4457   if (old != nullptr) {
4458     Runtime::Current()->GetInstrumentation()->UninstrumentQuickAllocEntryPoints();
4459   }
4460 }
4461 
SetGcPauseListener(GcPauseListener * l)4462 void Heap::SetGcPauseListener(GcPauseListener* l) {
4463   gc_pause_listener_.store(l, std::memory_order_relaxed);
4464 }
4465 
RemoveGcPauseListener()4466 void Heap::RemoveGcPauseListener() {
4467   gc_pause_listener_.store(nullptr, std::memory_order_relaxed);
4468 }
4469 
AllocWithNewTLAB(Thread * self,AllocatorType allocator_type,size_t alloc_size,bool grow,size_t * bytes_allocated,size_t * usable_size,size_t * bytes_tl_bulk_allocated)4470 mirror::Object* Heap::AllocWithNewTLAB(Thread* self,
4471                                        AllocatorType allocator_type,
4472                                        size_t alloc_size,
4473                                        bool grow,
4474                                        size_t* bytes_allocated,
4475                                        size_t* usable_size,
4476                                        size_t* bytes_tl_bulk_allocated) {
4477   mirror::Object* ret = nullptr;
4478   bool take_sample = false;
4479   size_t bytes_until_sample = 0;
4480 
4481   if (kUsePartialTlabs && alloc_size <= self->TlabRemainingCapacity()) {
4482     DCHECK_GT(alloc_size, self->TlabSize());
4483     // There is enough space if we grow the TLAB. Lets do that. This increases the
4484     // TLAB bytes.
4485     const size_t min_expand_size = alloc_size - self->TlabSize();
4486     size_t next_tlab_size = JHPCalculateNextTlabSize(self,
4487                                                      kPartialTlabSize,
4488                                                      alloc_size,
4489                                                      &take_sample,
4490                                                      &bytes_until_sample);
4491     const size_t expand_bytes = std::max(
4492         min_expand_size,
4493         std::min(self->TlabRemainingCapacity() - self->TlabSize(), next_tlab_size));
4494     if (UNLIKELY(IsOutOfMemoryOnAllocation(allocator_type, expand_bytes, grow))) {
4495       return nullptr;
4496     }
4497     *bytes_tl_bulk_allocated = expand_bytes;
4498     self->ExpandTlab(expand_bytes);
4499     DCHECK_LE(alloc_size, self->TlabSize());
4500   } else if (allocator_type == kAllocatorTypeTLAB) {
4501     DCHECK(bump_pointer_space_ != nullptr);
4502     // Try to allocate a page-aligned TLAB (not necessary though).
4503     // TODO: for large allocations, which are rare, maybe we should allocate
4504     // that object and return. There is no need to revoke the current TLAB,
4505     // particularly if it's mostly unutilized.
4506     size_t def_pr_tlab_size = RoundDown(alloc_size + kDefaultTLABSize, kPageSize) - alloc_size;
4507     size_t next_tlab_size = JHPCalculateNextTlabSize(self,
4508                                                      def_pr_tlab_size,
4509                                                      alloc_size,
4510                                                      &take_sample,
4511                                                      &bytes_until_sample);
4512     const size_t new_tlab_size = alloc_size + next_tlab_size;
4513     if (UNLIKELY(IsOutOfMemoryOnAllocation(allocator_type, new_tlab_size, grow))) {
4514       return nullptr;
4515     }
4516     // Try allocating a new thread local buffer, if the allocation fails the space must be
4517     // full so return null.
4518     if (!bump_pointer_space_->AllocNewTlab(self, new_tlab_size)) {
4519       return nullptr;
4520     }
4521     *bytes_tl_bulk_allocated = new_tlab_size;
4522     if (CheckPerfettoJHPEnabled()) {
4523       VLOG(heap) << "JHP:kAllocatorTypeTLAB, New Tlab bytes allocated= " << new_tlab_size;
4524     }
4525   } else {
4526     DCHECK(allocator_type == kAllocatorTypeRegionTLAB);
4527     DCHECK(region_space_ != nullptr);
4528     if (space::RegionSpace::kRegionSize >= alloc_size) {
4529       // Non-large. Check OOME for a tlab.
4530       if (LIKELY(!IsOutOfMemoryOnAllocation(allocator_type,
4531                                             space::RegionSpace::kRegionSize,
4532                                             grow))) {
4533         size_t def_pr_tlab_size = kUsePartialTlabs
4534                                       ? kPartialTlabSize
4535                                       : gc::space::RegionSpace::kRegionSize;
4536         size_t next_pr_tlab_size = JHPCalculateNextTlabSize(self,
4537                                                             def_pr_tlab_size,
4538                                                             alloc_size,
4539                                                             &take_sample,
4540                                                             &bytes_until_sample);
4541         const size_t new_tlab_size = kUsePartialTlabs
4542             ? std::max(alloc_size, next_pr_tlab_size)
4543             : next_pr_tlab_size;
4544         // Try to allocate a tlab.
4545         if (!region_space_->AllocNewTlab(self, new_tlab_size, bytes_tl_bulk_allocated)) {
4546           // Failed to allocate a tlab. Try non-tlab.
4547           ret = region_space_->AllocNonvirtual<false>(alloc_size,
4548                                                       bytes_allocated,
4549                                                       usable_size,
4550                                                       bytes_tl_bulk_allocated);
4551           JHPCheckNonTlabSampleAllocation(self, ret, alloc_size);
4552           return ret;
4553         }
4554         // Fall-through to using the TLAB below.
4555       } else {
4556         // Check OOME for a non-tlab allocation.
4557         if (!IsOutOfMemoryOnAllocation(allocator_type, alloc_size, grow)) {
4558           ret = region_space_->AllocNonvirtual<false>(alloc_size,
4559                                                       bytes_allocated,
4560                                                       usable_size,
4561                                                       bytes_tl_bulk_allocated);
4562           JHPCheckNonTlabSampleAllocation(self, ret, alloc_size);
4563           return ret;
4564         }
4565         // Neither tlab or non-tlab works. Give up.
4566         return nullptr;
4567       }
4568     } else {
4569       // Large. Check OOME.
4570       if (LIKELY(!IsOutOfMemoryOnAllocation(allocator_type, alloc_size, grow))) {
4571         ret = region_space_->AllocNonvirtual<false>(alloc_size,
4572                                                     bytes_allocated,
4573                                                     usable_size,
4574                                                     bytes_tl_bulk_allocated);
4575         JHPCheckNonTlabSampleAllocation(self, ret, alloc_size);
4576         return ret;
4577       }
4578       return nullptr;
4579     }
4580   }
4581   // Refilled TLAB, return.
4582   ret = self->AllocTlab(alloc_size);
4583   DCHECK(ret != nullptr);
4584   *bytes_allocated = alloc_size;
4585   *usable_size = alloc_size;
4586 
4587   // JavaHeapProfiler: Send the thread information about this allocation in case a sample is
4588   // requested.
4589   // This is the fallthrough from both the if and else if above cases => Cases that use TLAB.
4590   if (CheckPerfettoJHPEnabled()) {
4591     if (take_sample) {
4592       GetHeapSampler().ReportSample(ret, alloc_size);
4593       // Update the bytes_until_sample now that the allocation is already done.
4594       GetHeapSampler().SetBytesUntilSample(bytes_until_sample);
4595     }
4596     VLOG(heap) << "JHP:Fallthrough Tlab allocation";
4597   }
4598 
4599   return ret;
4600 }
4601 
GetVerification() const4602 const Verification* Heap::GetVerification() const {
4603   return verification_.get();
4604 }
4605 
VlogHeapGrowth(size_t old_footprint,size_t new_footprint,size_t alloc_size)4606 void Heap::VlogHeapGrowth(size_t old_footprint, size_t new_footprint, size_t alloc_size) {
4607   VLOG(heap) << "Growing heap from " << PrettySize(old_footprint) << " to "
4608              << PrettySize(new_footprint) << " for a " << PrettySize(alloc_size) << " allocation";
4609 }
4610 
4611 // Run a gc if we haven't run one since initial_gc_num. This forces processes to
4612 // reclaim memory allocated during startup, even if they don't do much
4613 // allocation post startup. If the process is actively allocating and triggering
4614 // GCs, or has moved to the background and hence forced a GC, this does nothing.
4615 class Heap::TriggerPostForkCCGcTask : public HeapTask {
4616  public:
TriggerPostForkCCGcTask(uint64_t target_time,uint32_t initial_gc_num)4617   explicit TriggerPostForkCCGcTask(uint64_t target_time, uint32_t initial_gc_num) :
4618       HeapTask(target_time), initial_gc_num_(initial_gc_num) {}
Run(Thread * self)4619   void Run(Thread* self) override {
4620     gc::Heap* heap = Runtime::Current()->GetHeap();
4621     if (heap->GetCurrentGcNum() == initial_gc_num_) {
4622       if (kLogAllGCs) {
4623         LOG(INFO) << "Forcing GC for allocation-inactive process";
4624       }
4625       heap->RequestConcurrentGC(self, kGcCauseBackground, false, initial_gc_num_);
4626     }
4627   }
4628  private:
4629   uint32_t initial_gc_num_;
4630 };
4631 
4632 // Reduce target footprint, if no GC has occurred since initial_gc_num.
4633 // If a GC already occurred, it will have done this for us.
4634 class Heap::ReduceTargetFootprintTask : public HeapTask {
4635  public:
ReduceTargetFootprintTask(uint64_t target_time,size_t new_target_sz,uint32_t initial_gc_num)4636   explicit ReduceTargetFootprintTask(uint64_t target_time, size_t new_target_sz,
4637                                      uint32_t initial_gc_num) :
4638       HeapTask(target_time), new_target_sz_(new_target_sz), initial_gc_num_(initial_gc_num) {}
Run(Thread * self)4639   void Run(Thread* self) override {
4640     gc::Heap* heap = Runtime::Current()->GetHeap();
4641     MutexLock mu(self, *(heap->gc_complete_lock_));
4642     if (heap->GetCurrentGcNum() == initial_gc_num_
4643         && heap->collector_type_running_ == kCollectorTypeNone) {
4644       size_t target_footprint = heap->target_footprint_.load(std::memory_order_relaxed);
4645       if (target_footprint > new_target_sz_) {
4646         if (heap->target_footprint_.CompareAndSetStrongRelaxed(target_footprint, new_target_sz_)) {
4647           heap->SetDefaultConcurrentStartBytesLocked();
4648         }
4649       }
4650     }
4651   }
4652  private:
4653   size_t new_target_sz_;
4654   uint32_t initial_gc_num_;
4655 };
4656 
4657 // Return a pseudo-random integer between 0 and 19999, using the uid as a seed.  We want this to
4658 // be deterministic for a given process, but to vary randomly across processes. Empirically, the
4659 // uids for processes for which this matters are distinct.
GetPseudoRandomFromUid()4660 static uint32_t GetPseudoRandomFromUid() {
4661   std::default_random_engine rng(getuid());
4662   std::uniform_int_distribution<int> dist(0, 19999);
4663   return dist(rng);
4664 }
4665 
PostForkChildAction(Thread * self)4666 void Heap::PostForkChildAction(Thread* self) {
4667   uint32_t starting_gc_num = GetCurrentGcNum();
4668   uint64_t last_adj_time = NanoTime();
4669   next_gc_type_ = NonStickyGcType();  // Always start with a full gc.
4670 
4671   LOG(INFO) << "Using " << foreground_collector_type_ << " GC.";
4672   if (gUseUserfaultfd) {
4673     DCHECK_NE(mark_compact_, nullptr);
4674     mark_compact_->CreateUserfaultfd(/*post_fork*/true);
4675   }
4676 
4677   // Temporarily increase target_footprint_ and concurrent_start_bytes_ to
4678   // max values to avoid GC during app launch.
4679   // Set target_footprint_ to the largest allowed value.
4680   SetIdealFootprint(growth_limit_);
4681   SetDefaultConcurrentStartBytes();
4682 
4683   // Shrink heap after kPostForkMaxHeapDurationMS, to force a memory hog process to GC.
4684   // This remains high enough that many processes will continue without a GC.
4685   if (initial_heap_size_ < growth_limit_) {
4686     size_t first_shrink_size = std::max(growth_limit_ / 4, initial_heap_size_);
4687     last_adj_time += MsToNs(kPostForkMaxHeapDurationMS);
4688     GetTaskProcessor()->AddTask(
4689         self, new ReduceTargetFootprintTask(last_adj_time, first_shrink_size, starting_gc_num));
4690     // Shrink to a small value after a substantial time period. This will typically force a
4691     // GC if none has occurred yet. Has no effect if there was a GC before this anyway, which
4692     // is commonly the case, e.g. because of a process transition.
4693     if (initial_heap_size_ < first_shrink_size) {
4694       last_adj_time += MsToNs(4 * kPostForkMaxHeapDurationMS);
4695       GetTaskProcessor()->AddTask(
4696           self,
4697           new ReduceTargetFootprintTask(last_adj_time, initial_heap_size_, starting_gc_num));
4698     }
4699   }
4700   // Schedule a GC after a substantial period of time. This will become a no-op if another GC is
4701   // scheduled in the interim. If not, we want to avoid holding onto start-up garbage.
4702   uint64_t post_fork_gc_time = last_adj_time
4703       + MsToNs(4 * kPostForkMaxHeapDurationMS + GetPseudoRandomFromUid());
4704   GetTaskProcessor()->AddTask(self,
4705                               new TriggerPostForkCCGcTask(post_fork_gc_time, starting_gc_num));
4706 }
4707 
VisitReflectiveTargets(ReflectiveValueVisitor * visit)4708 void Heap::VisitReflectiveTargets(ReflectiveValueVisitor *visit) {
4709   VisitObjectsPaused([&visit](mirror::Object* ref) NO_THREAD_SAFETY_ANALYSIS {
4710     art::ObjPtr<mirror::Class> klass(ref->GetClass());
4711     // All these classes are in the BootstrapClassLoader.
4712     if (!klass->IsBootStrapClassLoaded()) {
4713       return;
4714     }
4715     if (GetClassRoot<mirror::Method>()->IsAssignableFrom(klass) ||
4716         GetClassRoot<mirror::Constructor>()->IsAssignableFrom(klass)) {
4717       down_cast<mirror::Executable*>(ref)->VisitTarget(visit);
4718     } else if (art::GetClassRoot<art::mirror::Field>() == klass) {
4719       down_cast<mirror::Field*>(ref)->VisitTarget(visit);
4720     } else if (art::GetClassRoot<art::mirror::MethodHandle>()->IsAssignableFrom(klass)) {
4721       down_cast<mirror::MethodHandle*>(ref)->VisitTarget(visit);
4722     } else if (art::GetClassRoot<art::mirror::StaticFieldVarHandle>()->IsAssignableFrom(klass)) {
4723       down_cast<mirror::StaticFieldVarHandle*>(ref)->VisitTarget(visit);
4724     } else if (art::GetClassRoot<art::mirror::FieldVarHandle>()->IsAssignableFrom(klass)) {
4725       down_cast<mirror::FieldVarHandle*>(ref)->VisitTarget(visit);
4726     } else if (art::GetClassRoot<art::mirror::DexCache>()->IsAssignableFrom(klass)) {
4727       down_cast<mirror::DexCache*>(ref)->VisitReflectiveTargets(visit);
4728     }
4729   });
4730 }
4731 
AddHeapTask(gc::HeapTask * task)4732 bool Heap::AddHeapTask(gc::HeapTask* task) {
4733   Thread* const self = Thread::Current();
4734   if (!CanAddHeapTask(self)) {
4735     return false;
4736   }
4737   GetTaskProcessor()->AddTask(self, task);
4738   return true;
4739 }
4740 
4741 }  // namespace gc
4742 }  // namespace art
4743