forked from chromium/chromium
-
Notifications
You must be signed in to change notification settings - Fork 0
/
sampling_heap_profiler.cc
299 lines (258 loc) · 9.95 KB
/
sampling_heap_profiler.cc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
// Copyright 2018 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "base/sampling_heap_profiler/sampling_heap_profiler.h"
#include <algorithm>
#include <cmath>
#include <utility>
#include "base/allocator/allocator_shim.h"
#include "base/allocator/buildflags.h"
#include "base/allocator/partition_allocator/partition_alloc.h"
#include "base/bind.h"
#include "base/debug/stack_trace.h"
#include "base/logging.h"
#include "base/macros.h"
#include "base/no_destructor.h"
#include "base/partition_alloc_buildflags.h"
#include "base/sampling_heap_profiler/lock_free_address_hash_set.h"
#include "base/threading/thread_local_storage.h"
#include "base/trace_event/heap_profiler_allocation_context_tracker.h" // no-presubmit-check
#include "build/build_config.h"
#if defined(OS_APPLE)
#include <pthread.h>
#endif
#if defined(OS_LINUX) || defined(OS_CHROMEOS) || defined(OS_ANDROID)
#include <sys/prctl.h>
#endif
#if defined(OS_ANDROID) && BUILDFLAG(CAN_UNWIND_WITH_CFI_TABLE) && \
defined(OFFICIAL_BUILD)
#include "base/trace_event/cfi_backtrace_android.h" // no-presubmit-check
#endif
namespace base {
constexpr uint32_t kMaxStackEntries = 256;
namespace {
// If a thread name has been set from ThreadIdNameManager, use that. Otherwise,
// gets the thread name from kernel if available or returns a string with id.
// This function intentionally leaks the allocated strings since they are used
// to tag allocations even after the thread dies.
const char* GetAndLeakThreadName() {
const char* thread_name =
base::ThreadIdNameManager::GetInstance()->GetNameForCurrentThread();
if (thread_name && *thread_name != '\0')
return thread_name;
// prctl requires 16 bytes, snprintf requires 19, pthread_getname_np requires
// 64 on macOS, see PlatformThread::SetName in platform_thread_mac.mm.
constexpr size_t kBufferLen = 64;
char name[kBufferLen];
#if defined(OS_LINUX) || defined(OS_CHROMEOS) || defined(OS_ANDROID)
// If the thread name is not set, try to get it from prctl. Thread name might
// not be set in cases where the thread started before heap profiling was
// enabled.
int err = prctl(PR_GET_NAME, name);
if (!err)
return strdup(name);
#elif defined(OS_APPLE)
int err = pthread_getname_np(pthread_self(), name, kBufferLen);
if (err == 0 && *name != '\0')
return strdup(name);
#endif // defined(OS_LINUX) || defined(OS_CHROMEOS) || defined(OS_ANDROID)
// Use tid if we don't have a thread name.
snprintf(name, sizeof(name), "Thread %lu",
static_cast<unsigned long>(base::PlatformThread::CurrentId()));
return strdup(name);
}
const char* UpdateAndGetThreadName(const char* name) {
static thread_local const char* thread_name;
if (name)
thread_name = name;
if (!thread_name)
thread_name = GetAndLeakThreadName();
return thread_name;
}
} // namespace
SamplingHeapProfiler::Sample::Sample(size_t size,
size_t total,
uint32_t ordinal)
: size(size), total(total), ordinal(ordinal) {}
SamplingHeapProfiler::Sample::Sample(const Sample&) = default;
SamplingHeapProfiler::Sample::~Sample() = default;
SamplingHeapProfiler::SamplingHeapProfiler() = default;
SamplingHeapProfiler::~SamplingHeapProfiler() {
if (record_thread_names_)
base::ThreadIdNameManager::GetInstance()->RemoveObserver(this);
}
uint32_t SamplingHeapProfiler::Start() {
#if defined(OS_ANDROID) && BUILDFLAG(CAN_UNWIND_WITH_CFI_TABLE) && \
defined(OFFICIAL_BUILD)
if (!trace_event::CFIBacktraceAndroid::GetInitializedInstance()
->can_unwind_stack_frames()) {
LOG(WARNING) << "Sampling heap profiler: Stack unwinding is not available.";
return 0;
}
#endif
AutoLock lock(start_stop_mutex_);
if (!running_sessions_++)
PoissonAllocationSampler::Get()->AddSamplesObserver(this);
return last_sample_ordinal_;
}
void SamplingHeapProfiler::Stop() {
AutoLock lock(start_stop_mutex_);
DCHECK_GT(running_sessions_, 0);
if (!--running_sessions_)
PoissonAllocationSampler::Get()->RemoveSamplesObserver(this);
}
void SamplingHeapProfiler::SetSamplingInterval(size_t sampling_interval) {
PoissonAllocationSampler::Get()->SetSamplingInterval(sampling_interval);
}
void SamplingHeapProfiler::SetRecordThreadNames(bool value) {
if (record_thread_names_ == value)
return;
record_thread_names_ = value;
if (value) {
base::ThreadIdNameManager::GetInstance()->AddObserver(this);
} else {
base::ThreadIdNameManager::GetInstance()->RemoveObserver(this);
}
}
// static
const char* SamplingHeapProfiler::CachedThreadName() {
return UpdateAndGetThreadName(nullptr);
}
// static
void** SamplingHeapProfiler::CaptureStackTrace(void** frames,
size_t max_entries,
size_t* count) {
// Skip top frames as they correspond to the profiler itself.
size_t skip_frames = 3;
#if defined(OS_ANDROID) && BUILDFLAG(CAN_UNWIND_WITH_CFI_TABLE) && \
defined(OFFICIAL_BUILD)
size_t frame_count =
base::trace_event::CFIBacktraceAndroid::GetInitializedInstance()->Unwind(
const_cast<const void**>(frames), max_entries);
#elif BUILDFLAG(CAN_UNWIND_WITH_FRAME_POINTERS)
size_t frame_count = base::debug::TraceStackFramePointers(
const_cast<const void**>(frames), max_entries, skip_frames);
skip_frames = 0;
#else
// Fall-back to capturing the stack with base::debug::CollectStackTrace,
// which is likely slower, but more reliable.
size_t frame_count =
base::debug::CollectStackTrace(const_cast<void**>(frames), max_entries);
#endif
skip_frames = std::min(skip_frames, frame_count);
*count = frame_count - skip_frames;
return frames + skip_frames;
}
void SamplingHeapProfiler::SampleAdded(
void* address,
size_t size,
size_t total,
PoissonAllocationSampler::AllocatorType type,
const char* context) {
// CaptureStack and allocation context tracking may use TLS.
// Bail out if it has been destroyed.
if (UNLIKELY(base::ThreadLocalStorage::HasBeenDestroyed()))
return;
DCHECK(PoissonAllocationSampler::ScopedMuteThreadSamples::IsMuted());
Sample sample(size, total, ++last_sample_ordinal_);
sample.allocator = type;
using CaptureMode = trace_event::AllocationContextTracker::CaptureMode;
CaptureMode capture_mode =
trace_event::AllocationContextTracker::capture_mode();
if (capture_mode == CaptureMode::PSEUDO_STACK ||
capture_mode == CaptureMode::MIXED_STACK) {
CaptureMixedStack(context, &sample);
} else {
CaptureNativeStack(context, &sample);
}
AutoLock lock(mutex_);
RecordString(sample.context);
samples_.emplace(address, std::move(sample));
}
void SamplingHeapProfiler::CaptureMixedStack(const char* context,
Sample* sample) {
auto* tracker =
trace_event::AllocationContextTracker::GetInstanceForCurrentThread();
if (!tracker)
return;
trace_event::AllocationContext allocation_context;
if (!tracker->GetContextSnapshot(&allocation_context))
return;
const base::trace_event::Backtrace& backtrace = allocation_context.backtrace;
CHECK_LE(backtrace.frame_count, kMaxStackEntries);
std::vector<void*> stack;
stack.reserve(backtrace.frame_count);
AutoLock lock(mutex_); // Needed for RecordString call.
for (int i = base::checked_cast<int>(backtrace.frame_count) - 1; i >= 0;
--i) {
const base::trace_event::StackFrame& frame = backtrace.frames[i];
if (frame.type != base::trace_event::StackFrame::Type::PROGRAM_COUNTER)
RecordString(static_cast<const char*>(frame.value));
stack.push_back(const_cast<void*>(frame.value));
}
sample->stack = std::move(stack);
if (!context)
context = allocation_context.type_name;
sample->context = context;
}
void SamplingHeapProfiler::CaptureNativeStack(const char* context,
Sample* sample) {
void* stack[kMaxStackEntries];
size_t frame_count;
// One frame is reserved for the thread name.
void** first_frame =
CaptureStackTrace(stack, kMaxStackEntries - 1, &frame_count);
DCHECK_LT(frame_count, kMaxStackEntries);
sample->stack.assign(first_frame, first_frame + frame_count);
if (record_thread_names_)
sample->thread_name = CachedThreadName();
if (!context) {
const auto* tracker =
trace_event::AllocationContextTracker::GetInstanceForCurrentThread();
if (tracker)
context = tracker->TaskContext();
}
sample->context = context;
}
const char* SamplingHeapProfiler::RecordString(const char* string) {
return string ? *strings_.insert(string).first : nullptr;
}
void SamplingHeapProfiler::SampleRemoved(void* address) {
DCHECK(base::PoissonAllocationSampler::ScopedMuteThreadSamples::IsMuted());
base::AutoLock lock(mutex_);
samples_.erase(address);
}
std::vector<SamplingHeapProfiler::Sample> SamplingHeapProfiler::GetSamples(
uint32_t profile_id) {
// Make sure the sampler does not invoke |SampleAdded| or |SampleRemoved|
// on this thread. Otherwise it could have end up with a deadlock.
// See crbug.com/882495
PoissonAllocationSampler::ScopedMuteThreadSamples no_samples_scope;
AutoLock lock(mutex_);
std::vector<Sample> samples;
samples.reserve(samples_.size());
for (auto& it : samples_) {
Sample& sample = it.second;
if (sample.ordinal > profile_id)
samples.push_back(sample);
}
return samples;
}
std::vector<const char*> SamplingHeapProfiler::GetStrings() {
PoissonAllocationSampler::ScopedMuteThreadSamples no_samples_scope;
AutoLock lock(mutex_);
return std::vector<const char*>(strings_.begin(), strings_.end());
}
// static
void SamplingHeapProfiler::Init() {
PoissonAllocationSampler::Init();
}
// static
SamplingHeapProfiler* SamplingHeapProfiler::Get() {
static NoDestructor<SamplingHeapProfiler> instance;
return instance.get();
}
void SamplingHeapProfiler::OnThreadNameChanged(const char* name) {
UpdateAndGetThreadName(name);
}
} // namespace base