summaryrefslogtreecommitdiffstats
path: root/runtime/gc
diff options
context:
space:
mode:
Diffstat (limited to 'runtime/gc')
-rw-r--r--runtime/gc/heap-inl.h9
-rw-r--r--runtime/gc/heap.cc95
-rw-r--r--runtime/gc/heap.h17
3 files changed, 114 insertions, 7 deletions
diff --git a/runtime/gc/heap-inl.h b/runtime/gc/heap-inl.h
index 2d54330..2ec9c86 100644
--- a/runtime/gc/heap-inl.h
+++ b/runtime/gc/heap-inl.h
@@ -174,6 +174,13 @@ inline mirror::Object* Heap::AllocObjectWithAllocator(Thread* self, mirror::Clas
} else {
DCHECK(!Dbg::IsAllocTrackingEnabled());
}
+ if (kInstrumented) {
+ if (gc_stress_mode_) {
+ CheckGcStressMode(self, &obj);
+ }
+ } else {
+ DCHECK(!gc_stress_mode_);
+ }
// IsConcurrentGc() isn't known at compile time so we can optimize by not checking it for
// the BumpPointer or TLAB allocators. This is nice since it allows the entire if statement to be
// optimized out. And for the other allocators, AllocatorMayHaveConcurrentGC is a constant since
@@ -391,7 +398,7 @@ inline bool Heap::ShouldAllocLargeObject(mirror::Class* c, size_t byte_count) co
// Zygote resulting in it being prematurely freed.
// We can only do this for primitive objects since large objects will not be within the card table
// range. This also means that we rely on SetClass not dirtying the object's card.
- return byte_count >= large_object_threshold_ && c->IsPrimitiveArray();
+ return byte_count >= large_object_threshold_ && (c->IsPrimitiveArray() || c->IsStringClass());
}
template <bool kGrow>
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index 9a70d69..57557e2 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -21,6 +21,7 @@
#include <limits>
#include <memory>
+#include <unwind.h> // For GC verification.
#include <vector>
#include "art_field-inl.h"
@@ -125,7 +126,8 @@ Heap::Heap(size_t initial_size, size_t growth_limit, size_t min_free, size_t max
bool ignore_max_footprint, 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 use_homogeneous_space_compaction_for_oom,
+ bool verify_post_gc_rosalloc, bool gc_stress_mode,
+ bool use_homogeneous_space_compaction_for_oom,
uint64_t min_interval_homogeneous_space_compaction_by_oom)
: non_moving_space_(nullptr),
rosalloc_space_(nullptr),
@@ -170,6 +172,7 @@ Heap::Heap(size_t initial_size, size_t growth_limit, size_t min_free, size_t max
verify_pre_gc_rosalloc_(verify_pre_gc_rosalloc),
verify_pre_sweeping_rosalloc_(verify_pre_sweeping_rosalloc),
verify_post_gc_rosalloc_(verify_post_gc_rosalloc),
+ gc_stress_mode_(gc_stress_mode),
/* For GC a lot mode, we limit the allocations stacks to be kGcAlotInterval allocations. This
* causes a lot of GC since we do a GC for alloc whenever the stack is full. When heap
* verification is enabled, we limit the size of allocation stacks to speed up their
@@ -209,13 +212,17 @@ Heap::Heap(size_t initial_size, size_t growth_limit, size_t min_free, size_t max
blocking_gc_count_last_window_(0U),
gc_count_rate_histogram_("gc count rate histogram", 1U, kGcCountRateMaxBucketCount),
blocking_gc_count_rate_histogram_("blocking gc count rate histogram", 1U,
- kGcCountRateMaxBucketCount) {
+ kGcCountRateMaxBucketCount),
+ backtrace_lock_(nullptr),
+ seen_backtrace_count_(0u),
+ unique_backtrace_count_(0u) {
if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
LOG(INFO) << "Heap() entering";
}
+ Runtime* const runtime = Runtime::Current();
// If we aren't the zygote, switch to the default non zygote allocator. This may update the
// entrypoints.
- const bool is_zygote = Runtime::Current()->IsZygote();
+ const bool is_zygote = runtime->IsZygote();
if (!is_zygote) {
// Background compaction is currently not supported for command line runs.
if (background_collector_type_ != foreground_collector_type_) {
@@ -507,8 +514,12 @@ Heap::Heap(size_t initial_size, size_t growth_limit, size_t min_free, size_t max
LOG(FATAL) << "There's a gap between the image space and the non-moving space";
}
}
- if (running_on_valgrind_) {
- Runtime::Current()->GetInstrumentation()->InstrumentQuickAllocEntryPoints();
+ instrumentation::Instrumentation* const instrumentation = runtime->GetInstrumentation();
+ if (gc_stress_mode_) {
+ backtrace_lock_ = new Mutex("GC complete lock");
+ }
+ if (running_on_valgrind_ || gc_stress_mode_) {
+ instrumentation->InstrumentQuickAllocEntryPoints();
}
if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
LOG(INFO) << "Heap() exiting";
@@ -1072,6 +1083,12 @@ Heap::~Heap() {
STLDeleteElements(&discontinuous_spaces_);
delete gc_complete_lock_;
delete pending_task_lock_;
+ delete backtrace_lock_;
+ if (unique_backtrace_count_.LoadRelaxed() != 0 || seen_backtrace_count_.LoadRelaxed() != 0) {
+ LOG(INFO) << "gc stress unique=" << unique_backtrace_count_.LoadRelaxed()
+ << " total=" << seen_backtrace_count_.LoadRelaxed() +
+ unique_backtrace_count_.LoadRelaxed();
+ }
VLOG(heap) << "Finished ~Heap()";
}
@@ -3675,5 +3692,73 @@ void Heap::ClearMarkedObjects() {
}
}
+// Based on debug malloc logic from libc/bionic/debug_stacktrace.cpp.
+class StackCrawlState {
+ public:
+ StackCrawlState(uintptr_t* frames, size_t max_depth, size_t skip_count)
+ : frames_(frames), frame_count_(0), max_depth_(max_depth), skip_count_(skip_count) {
+ }
+ size_t GetFrameCount() const {
+ return frame_count_;
+ }
+ static _Unwind_Reason_Code Callback(_Unwind_Context* context, void* arg) {
+ auto* const state = reinterpret_cast<StackCrawlState*>(arg);
+ const uintptr_t ip = _Unwind_GetIP(context);
+ // The first stack frame is get_backtrace itself. Skip it.
+ if (ip != 0 && state->skip_count_ > 0) {
+ --state->skip_count_;
+ return _URC_NO_REASON;
+ }
+ // ip may be off for ARM but it shouldn't matter since we only use it for hashing.
+ state->frames_[state->frame_count_] = ip;
+ state->frame_count_++;
+ return state->frame_count_ >= state->max_depth_ ? _URC_END_OF_STACK : _URC_NO_REASON;
+ }
+
+ private:
+ uintptr_t* const frames_;
+ size_t frame_count_;
+ const size_t max_depth_;
+ size_t skip_count_;
+};
+
+static size_t get_backtrace(uintptr_t* frames, size_t max_depth) {
+ StackCrawlState state(frames, max_depth, 0u);
+ _Unwind_Backtrace(&StackCrawlState::Callback, &state);
+ return state.GetFrameCount();
+}
+
+void Heap::CheckGcStressMode(Thread* self, mirror::Object** obj) {
+ auto* const runtime = Runtime::Current();
+ if (gc_stress_mode_ && runtime->GetClassLinker()->IsInitialized() &&
+ !runtime->IsActiveTransaction() && mirror::Class::HasJavaLangClass()) {
+ // Check if we should GC.
+ bool new_backtrace = false;
+ {
+ static constexpr size_t kMaxFrames = 16u;
+ uintptr_t backtrace[kMaxFrames];
+ const size_t frames = get_backtrace(backtrace, kMaxFrames);
+ uint64_t hash = 0;
+ for (size_t i = 0; i < frames; ++i) {
+ hash = hash * 2654435761 + backtrace[i];
+ hash += (hash >> 13) ^ (hash << 6);
+ }
+ MutexLock mu(self, *backtrace_lock_);
+ new_backtrace = seen_backtraces_.find(hash) == seen_backtraces_.end();
+ if (new_backtrace) {
+ seen_backtraces_.insert(hash);
+ }
+ }
+ if (new_backtrace) {
+ StackHandleScope<1> hs(self);
+ auto h = hs.NewHandleWrapper(obj);
+ CollectGarbage(false);
+ unique_backtrace_count_.FetchAndAddSequentiallyConsistent(1);
+ } else {
+ seen_backtrace_count_.FetchAndAddSequentiallyConsistent(1);
+ }
+ }
+}
+
} // namespace gc
} // namespace art
diff --git a/runtime/gc/heap.h b/runtime/gc/heap.h
index dac747b..81476a4 100644
--- a/runtime/gc/heap.h
+++ b/runtime/gc/heap.h
@@ -19,6 +19,7 @@
#include <iosfwd>
#include <string>
+#include <unordered_set>
#include <vector>
#include "allocator_type.h"
@@ -180,7 +181,8 @@ class Heap {
bool ignore_max_footprint, 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 use_homogeneous_space_compaction,
+ bool verify_post_gc_rosalloc, bool gc_stress_mode,
+ bool use_homogeneous_space_compaction,
uint64_t min_interval_homogeneous_space_compaction_by_oom);
~Heap();
@@ -887,6 +889,10 @@ class Heap {
void UpdateGcCountRateHistograms() EXCLUSIVE_LOCKS_REQUIRED(gc_complete_lock_);
+ // GC stress mode attempts to do one GC per unique backtrace.
+ void CheckGcStressMode(Thread* self, mirror::Object** obj)
+ SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
// All-known continuous spaces, where objects lie within fixed bounds.
std::vector<space::ContinuousSpace*> continuous_spaces_;
@@ -1042,6 +1048,7 @@ class Heap {
bool verify_pre_gc_rosalloc_;
bool verify_pre_sweeping_rosalloc_;
bool verify_post_gc_rosalloc_;
+ const bool gc_stress_mode_;
// RAII that temporarily disables the rosalloc verification during
// the zygote fork.
@@ -1192,6 +1199,14 @@ class Heap {
// The histogram of the number of blocking GC invocations per window duration.
Histogram<uint64_t> blocking_gc_count_rate_histogram_ GUARDED_BY(gc_complete_lock_);
+ // GC stress related data structures.
+ Mutex* backtrace_lock_ DEFAULT_MUTEX_ACQUIRED_AFTER;
+ // Debugging variables, seen backtraces vs unique backtraces.
+ Atomic<uint64_t> seen_backtrace_count_;
+ Atomic<uint64_t> unique_backtrace_count_;
+ // Stack trace hashes that we already saw,
+ std::unordered_set<uint64_t> seen_backtraces_ GUARDED_BY(backtrace_lock_);
+
friend class CollectorTransitionTask;
friend class collector::GarbageCollector;
friend class collector::MarkCompact;
generated by cgit v1.1 at 2025-01-24 15:00:08 +0000