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