OLD | NEW |
---|---|
(Empty) | |
1 // Copyright 2017 The Chromium Authors. All rights reserved. | |
2 // Use of this source code is governed by a BSD-style license that can be | |
3 // found in the LICENSE file. | |
4 | |
5 #include "base/profiler/native_stack_sampler.h" | |
6 | |
7 #include <dlfcn.h> | |
8 #include <libkern/OSByteOrder.h> | |
9 #include <libunwind.h> | |
10 #include <mach-o/swap.h> | |
11 #include <mach/kern_return.h> | |
12 #include <mach/mach.h> | |
13 #include <mach/thread_act.h> | |
14 #include <pthread.h> | |
15 #include <sys/syslimits.h> | |
16 | |
17 #include <map> | |
18 #include <memory> | |
19 | |
20 #include "base/logging.h" | |
21 #include "base/macros.h" | |
22 #include "base/memory/ptr_util.h" | |
23 #include "base/strings/string_number_conversions.h" | |
24 | |
25 namespace base { | |
26 | |
27 namespace { | |
28 | |
29 // Stack walking -------------------------------------------------------------- | |
30 | |
31 // Fills |state| with |target_thread|'s context. | |
32 // | |
33 // Note that this is called while a thread is suspended. Make very very sure | |
34 // that no shared resources (e.g. memory allocators) are used for the duration | |
35 // of this function. | |
36 bool GetThreadState(thread_act_t target_thread, x86_thread_state64_t* state) { | |
37 mach_msg_type_number_t count = | |
38 static_cast<mach_msg_type_number_t>(x86_THREAD_STATE64_COUNT); | |
39 return thread_get_state(target_thread, x86_THREAD_STATE64, | |
40 reinterpret_cast<thread_state_t>(state), | |
41 &count) == KERN_SUCCESS; | |
42 } | |
43 | |
44 // If the value at |pointer| points to the original stack, rewrites it to point | |
45 // to the corresponding location in the copied stack. | |
46 // | |
47 // Note that this is called while a thread is suspended. Make very very sure | |
48 // that no shared resources (e.g. memory allocators) are used for the duration | |
49 // of this function. | |
50 uint64_t RewritePointerIfInOriginalStack(uint64_t* original_stack_bottom, | |
51 uint64_t* original_stack_top, | |
52 uint64_t* stack_copy_bottom, | |
53 uint64_t pointer) { | |
54 uint64_t original_stack_bottom_int = | |
55 reinterpret_cast<uint64_t>(original_stack_bottom); | |
56 uint64_t original_stack_top_int = | |
57 reinterpret_cast<uint64_t>(original_stack_top); | |
58 uint64_t stack_copy_bottom_int = | |
59 reinterpret_cast<uint64_t>(stack_copy_bottom); | |
60 | |
61 if ((pointer < original_stack_bottom_int) || | |
62 (pointer >= original_stack_top_int)) { | |
63 return pointer; | |
64 } | |
65 | |
66 return stack_copy_bottom_int + (pointer - original_stack_bottom_int); | |
67 } | |
68 | |
69 // Copies the stack to a buffer while rewriting possible pointers to locations | |
70 // within the stack to point to the corresponding locations in the copy. This is | |
71 // necessary to handle stack frames with dynamic stack allocation, where a | |
72 // pointer to the beginning of the dynamic allocation area is stored on the | |
73 // stack and/or in a non-volatile register. | |
74 // | |
75 // Returns the top of the stack in the stack copy. | |
76 // | |
77 // Eager rewriting of anything that looks like a pointer to the stack, as done | |
78 // in this function, does not adversely affect the stack unwinding. The only | |
79 // other values on the stack the unwinding depends on are return addresses, | |
80 // which should not point within the stack memory. The rewriting is guaranteed | |
81 // to catch all pointers because the stacks are guaranteed by the ABI to be | |
82 // sizeof(void*) aligned. | |
83 // | |
84 // Note that this is called while a thread is suspended. Make very very sure | |
85 // that no shared resources (e.g. memory allocators) are used for the duration | |
86 // of this function. | |
87 uint64_t CopyStackAndRewritePointers(void* dest, | |
88 void* from, | |
89 void* to, | |
90 x86_thread_state64_t* thread_state) | |
91 NO_SANITIZE("address") { | |
92 uint64_t* original_stack_bottom = static_cast<uint64_t*>(from); | |
93 uint64_t* original_stack_top = static_cast<uint64_t*>(to); | |
94 uint64_t* stack_copy_bottom = static_cast<uint64_t*>(dest); | |
95 | |
96 size_t count = original_stack_top - original_stack_bottom; | |
97 for (size_t pos = 0; pos < count; ++pos) { | |
98 stack_copy_bottom[pos] = RewritePointerIfInOriginalStack( | |
99 original_stack_bottom, original_stack_top, stack_copy_bottom, | |
100 original_stack_bottom[pos]); | |
101 } | |
102 | |
103 thread_state->__rbp = | |
104 RewritePointerIfInOriginalStack(original_stack_bottom, original_stack_top, | |
105 stack_copy_bottom, thread_state->__rbp); | |
106 thread_state->__rsp = | |
107 RewritePointerIfInOriginalStack(original_stack_bottom, original_stack_top, | |
108 stack_copy_bottom, thread_state->__rsp); | |
109 | |
110 return reinterpret_cast<uint64_t>(stack_copy_bottom + count); | |
111 } | |
112 | |
113 // Walks the stack represented by |unwind_context|, calling back to the provided | |
114 // lambda for each frame. Returns false if an error occurred, otherwise returns | |
115 // true. | |
116 template <typename StackFrameCallback> | |
117 bool WalkStackFromContext(unw_context_t* unwind_context, | |
118 size_t* frame_count, | |
119 const StackFrameCallback& callback) { | |
120 unw_cursor_t unwind_cursor; | |
121 unw_init_local(&unwind_cursor, unwind_context); | |
122 | |
123 int step_result; | |
124 unw_word_t ip; | |
125 do { | |
126 ++(*frame_count); | |
127 unw_get_reg(&unwind_cursor, UNW_REG_IP, &ip); | |
128 | |
129 callback(static_cast<uintptr_t>(ip)); | |
130 | |
131 step_result = unw_step(&unwind_cursor); | |
132 } while (step_result > 0); | |
133 | |
134 if (step_result != 0) | |
135 return false; | |
136 | |
137 return true; | |
138 } | |
139 | |
140 // Walks the stack represented by |thread_state|, calling back to the provided | |
141 // lambda for each frame. | |
142 template <typename StackFrameCallback> | |
143 void WalkStack(const x86_thread_state64_t& thread_state, | |
144 uint64_t stack_top, | |
145 const StackFrameCallback& callback) { | |
146 size_t frame_count = 0; | |
147 // This uses libunwind to walk the stack. libunwind is designed to be used for | |
148 // a thread to walk its own stack. This creates two problems. | |
149 | |
150 // Problem 1: There is no official way to create a unw_context other than to | |
151 // create it from the current state of the current thread's stack. To get | |
152 // around this, forge a context. A unw_context is just a copy of the 16 main | |
153 // registers followed by the instruction pointer, nothing more. | |
154 // Coincidentally, the first 17 items of the x86_thread_state64_t type are | |
155 // exactly those registers in exactly the same order, so just bulk copy them | |
156 // over. | |
157 unw_context_t unwind_context; | |
158 memcpy(&unwind_context, &thread_state, sizeof(uint64_t) * 17); | |
159 bool result = WalkStackFromContext(&unwind_context, &frame_count, callback); | |
160 | |
161 if (!result) | |
162 return; | |
163 | |
164 if (frame_count == 1) { | |
165 // Problem 2: Because libunwind is designed to be triggered by user code on | |
166 // their own thread, if it hits a library that has no unwind info for the | |
167 // function that is being executed, it just stops. This isn't a problem in | |
168 // the normal case, but in this case, it's quite possible that the stack | |
169 // being walked is stopped in a function that bridges to the kernel and thus | |
170 // is missing the unwind info. | |
171 // | |
172 // If so, cheat by scanning the stack and trying again. Only do this once, | |
173 // and only if the first time using libunwind fails after one frame. | |
174 bool ip_in_valid_image = false; | |
175 do { | |
Mike Wittman
2017/03/02 16:35:04
How about using aliases for the register state, to
Avi (use Gerrit)
2017/03/02 19:50:44
Done.
| |
176 unwind_context.data[16] = | |
177 *reinterpret_cast<uint64_t*>(unwind_context.data[7]); // rip = *rsp | |
178 unwind_context.data[7] = unwind_context.data[7] + 8; // rsp++ | |
179 unw_cursor_t unwind_cursor; | |
180 unw_init_local(&unwind_cursor, &unwind_context); | |
181 unw_proc_info_t proc_info; | |
182 unw_get_proc_info(&unwind_cursor, &proc_info); | |
183 ip_in_valid_image = proc_info.extra != 0; | |
Mike Wittman
2017/03/02 16:35:04
Move all these calls into an IsIPInValidImage(cons
Avi (use Gerrit)
2017/03/02 19:50:44
Done.
| |
184 } while (!ip_in_valid_image && unwind_context.data[16] < stack_top); | |
Mike Wittman
2017/03/02 16:35:04
Shouldn't this be unwind_context.data[7] < stack_t
Avi (use Gerrit)
2017/03/02 19:50:44
Yes. Done.
| |
185 | |
186 if (ip_in_valid_image) | |
187 WalkStackFromContext(&unwind_context, &frame_count, callback); | |
188 } | |
189 } | |
190 | |
191 // Module identifiers --------------------------------------------------------- | |
192 | |
193 // Fills |id| with the UUID of the x86_64 Mach-O binary with the header | |
194 // |mach_header|. Returns false if the binary is malformed or does not contain | |
195 // the UUID load command. | |
196 bool GetUUID(const mach_header_64* mach_header, unsigned char* id) { | |
197 size_t offset = sizeof(mach_header_64); | |
198 size_t offset_limit = sizeof(mach_header_64) + mach_header->sizeofcmds; | |
199 for (uint32_t i = 0; (i < mach_header->ncmds) && | |
200 (offset + sizeof(load_command) < offset_limit); | |
201 ++i) { | |
202 const load_command* current_cmd = reinterpret_cast<const load_command*>( | |
203 reinterpret_cast<const uint8_t*>(mach_header) + offset); | |
204 | |
205 if (offset + current_cmd->cmdsize > offset_limit) { | |
206 // This command runs off the end of the command list. This is malformed. | |
207 return false; | |
208 } | |
209 | |
210 if (current_cmd->cmd == LC_UUID) { | |
211 if (current_cmd->cmdsize < sizeof(uuid_command)) { | |
212 // This "UUID command" is too small. This is malformed. | |
213 return false; | |
214 } | |
215 | |
216 const uuid_command* uuid_cmd = | |
217 reinterpret_cast<const uuid_command*>(current_cmd); | |
218 static_assert(sizeof(uuid_cmd->uuid) == sizeof(uuid_t), | |
219 "UUID field of UUID command should be 16 bytes."); | |
220 memcpy(id, &uuid_cmd->uuid, sizeof(uuid_t)); | |
221 return true; | |
222 } | |
223 offset += current_cmd->cmdsize; | |
224 } | |
225 return false; | |
226 } | |
227 | |
228 // Returns the hex encoding of a 16-byte ID for the binary loaded at | |
229 // |module_addr|. Returns an empty string if the UUID cannot be found at | |
230 // |module_addr|. | |
231 std::string GetUniqueId(const void* module_addr) { | |
232 const mach_header_64* mach_header = | |
233 reinterpret_cast<const mach_header_64*>(module_addr); | |
234 DCHECK_EQ(MH_MAGIC_64, mach_header->magic); | |
235 | |
236 unsigned char id[sizeof(uuid_t)]; | |
237 if (!GetUUID(mach_header, id)) | |
238 return ""; | |
239 return HexEncode(id, sizeof(uuid_t)); | |
240 } | |
241 | |
242 // Gets the index for the Module containing |instruction_pointer| in | |
243 // |modules|, adding it if it's not already present. Returns | |
244 // StackSamplingProfiler::Frame::kUnknownModuleIndex if no Module can be | |
245 // determined for |module|. | |
246 size_t GetModuleIndex(const uintptr_t instruction_pointer, | |
247 std::vector<StackSamplingProfiler::Module>* modules, | |
248 std::map<const void*, size_t>* profile_module_index) { | |
249 Dl_info inf; | |
250 if (!dladdr(reinterpret_cast<const void*>(instruction_pointer), &inf)) | |
251 return StackSamplingProfiler::Frame::kUnknownModuleIndex; | |
252 | |
253 auto module_index = profile_module_index->find(inf.dli_fbase); | |
254 if (module_index == profile_module_index->end()) { | |
255 StackSamplingProfiler::Module module( | |
256 reinterpret_cast<uintptr_t>(inf.dli_fbase), GetUniqueId(inf.dli_fbase), | |
257 base::FilePath(inf.dli_fname)); | |
258 modules->push_back(module); | |
259 module_index = | |
260 profile_module_index | |
261 ->insert(std::make_pair(inf.dli_fbase, modules->size() - 1)) | |
262 .first; | |
263 } | |
264 return module_index->second; | |
265 } | |
266 | |
267 // ScopedSuspendThread -------------------------------------------------------- | |
268 | |
269 // Suspends a thread for the lifetime of the object. | |
270 class ScopedSuspendThread { | |
271 public: | |
272 explicit ScopedSuspendThread(mach_port_t thread_port); | |
273 ~ScopedSuspendThread(); | |
274 | |
275 bool was_successful() const { return was_successful_; } | |
276 | |
277 private: | |
278 mach_port_t thread_port_; | |
279 bool was_successful_; | |
280 | |
281 DISALLOW_COPY_AND_ASSIGN(ScopedSuspendThread); | |
282 }; | |
283 | |
284 ScopedSuspendThread::ScopedSuspendThread(mach_port_t thread_port) | |
285 : thread_port_(thread_port), | |
286 was_successful_(thread_suspend(thread_port) == KERN_SUCCESS) {} | |
287 | |
288 ScopedSuspendThread::~ScopedSuspendThread() { | |
289 if (!was_successful_) | |
290 return; | |
291 | |
292 kern_return_t resume_result = thread_resume(thread_port_); | |
293 CHECK_EQ(KERN_SUCCESS, resume_result) << "thread_resume failed"; | |
294 } | |
295 | |
296 // NativeStackSamplerMac ------------------------------------------------------ | |
297 | |
298 class NativeStackSamplerMac : public NativeStackSampler { | |
299 public: | |
300 NativeStackSamplerMac(mach_port_t thread_port, | |
301 AnnotateCallback annotator, | |
302 NativeStackSamplerTestDelegate* test_delegate); | |
303 ~NativeStackSamplerMac() override; | |
304 | |
305 // StackSamplingProfiler::NativeStackSampler: | |
306 void ProfileRecordingStarting( | |
307 std::vector<StackSamplingProfiler::Module>* modules) override; | |
308 void RecordStackSample(StackSamplingProfiler::Sample* sample) override; | |
309 void ProfileRecordingStopped() override; | |
310 | |
311 private: | |
312 enum { | |
313 // Intended to hold the largest stack used by Chrome. The default macOS main | |
314 // thread stack size is 8 MB, and this allows for expansion if it occurs. | |
315 kStackCopyBufferSize = 12 * 1024 * 1024 | |
316 }; | |
317 | |
318 // Suspends the thread with |thread_port_|, copies its stack and resumes the | |
319 // thread, then records the stack frames and associated modules into |sample|. | |
320 void SuspendThreadAndRecordStack(StackSamplingProfiler::Sample* sample); | |
321 | |
322 // Weak reference: Mach port for thread being profiled. | |
323 mach_port_t thread_port_; | |
324 | |
325 const AnnotateCallback annotator_; | |
326 | |
327 NativeStackSamplerTestDelegate* const test_delegate_; | |
328 | |
329 // The stack base address corresponding to |thread_handle_|. | |
330 const void* const thread_stack_base_address_; | |
331 | |
332 // Buffer to use for copies of the stack. We use the same buffer for all the | |
333 // samples to avoid the overhead of multiple allocations and frees. | |
334 const std::unique_ptr<unsigned char[]> stack_copy_buffer_; | |
335 | |
336 // Weak. Points to the modules associated with the profile being recorded | |
337 // between ProfileRecordingStarting() and ProfileRecordingStopped(). | |
338 std::vector<StackSamplingProfiler::Module>* current_modules_ = nullptr; | |
339 | |
340 // Maps a module's base address to the corresponding Module's index within | |
341 // current_modules_. | |
342 std::map<const void*, size_t> profile_module_index_; | |
343 | |
344 DISALLOW_COPY_AND_ASSIGN(NativeStackSamplerMac); | |
345 }; | |
346 | |
347 NativeStackSamplerMac::NativeStackSamplerMac( | |
348 mach_port_t thread_port, | |
349 AnnotateCallback annotator, | |
350 NativeStackSamplerTestDelegate* test_delegate) | |
351 : thread_port_(thread_port), | |
352 annotator_(annotator), | |
353 test_delegate_(test_delegate), | |
354 thread_stack_base_address_( | |
355 pthread_get_stackaddr_np(pthread_from_mach_thread_np(thread_port))), | |
356 stack_copy_buffer_(new unsigned char[kStackCopyBufferSize]) { | |
357 DCHECK(annotator_); | |
358 | |
359 // This class suspends threads, and those threads might be suspended in dyld. | |
360 // Therefore, for all the system functions that might be linked in dynamically | |
361 // that are used while threads are suspended, make calls to them to make sure | |
362 // that they are linked up. | |
363 x86_thread_state64_t thread_state; | |
364 GetThreadState(thread_port_, &thread_state); | |
365 } | |
366 | |
367 NativeStackSamplerMac::~NativeStackSamplerMac() {} | |
368 | |
369 void NativeStackSamplerMac::ProfileRecordingStarting( | |
370 std::vector<StackSamplingProfiler::Module>* modules) { | |
371 current_modules_ = modules; | |
372 profile_module_index_.clear(); | |
373 } | |
374 | |
375 void NativeStackSamplerMac::RecordStackSample( | |
376 StackSamplingProfiler::Sample* sample) { | |
377 DCHECK(current_modules_); | |
378 | |
379 if (!stack_copy_buffer_) | |
380 return; | |
381 | |
382 SuspendThreadAndRecordStack(sample); | |
383 } | |
384 | |
385 void NativeStackSamplerMac::ProfileRecordingStopped() { | |
386 current_modules_ = nullptr; | |
387 } | |
388 | |
389 void NativeStackSamplerMac::SuspendThreadAndRecordStack( | |
390 StackSamplingProfiler::Sample* sample) { | |
391 x86_thread_state64_t thread_state; | |
392 | |
393 // Copy the stack. | |
394 | |
395 uint64_t new_stack_top; | |
Mike Wittman
2017/03/02 16:35:03
initialize to 0
Avi (use Gerrit)
2017/03/02 19:50:44
Done.
| |
396 { | |
397 // IMPORTANT NOTE: Do not do ANYTHING in this in this scope that might | |
398 // allocate memory, including indirectly via use of DCHECK/CHECK or other | |
399 // logging statements. Otherwise this code can deadlock on heap locks in the | |
400 // default heap acquired by the target thread before it was suspended. | |
401 ScopedSuspendThread suspend_thread(thread_port_); | |
402 if (!suspend_thread.was_successful()) | |
403 return; | |
404 | |
405 if (!GetThreadState(thread_port_, &thread_state)) | |
406 return; | |
407 uint64_t stack_top = reinterpret_cast<uint64_t>(thread_stack_base_address_); | |
408 uint64_t stack_bottom = thread_state.__rsp; | |
409 | |
410 if ((stack_top - stack_bottom) > kStackCopyBufferSize) | |
411 return; | |
412 | |
413 (*annotator_)(sample); | |
414 | |
415 new_stack_top = CopyStackAndRewritePointers( | |
416 stack_copy_buffer_.get(), reinterpret_cast<void*>(stack_bottom), | |
417 reinterpret_cast<void*>(stack_top), &thread_state); | |
418 } // ScopedSuspendThread | |
Mike Wittman
2017/03/02 16:35:03
nit: new_stack_top could be computed directly here
Avi (use Gerrit)
2017/03/02 19:50:44
Done.
| |
419 | |
420 if (test_delegate_) | |
421 test_delegate_->OnPreStackWalk(); | |
422 | |
423 // Walk the stack and record it. | |
424 | |
425 // Reserve enough memory for most stacks, to avoid repeated allocations. | |
426 // Approximately 99.9% of recorded stacks are 128 frames or fewer. | |
427 sample->frames.reserve(128); | |
428 | |
429 auto current_modules = current_modules_; | |
430 auto profile_module_index = &profile_module_index_; | |
431 WalkStack( | |
432 thread_state, new_stack_top, | |
433 [sample, current_modules, profile_module_index](uintptr_t frame_ip) { | |
434 sample->frames.push_back(StackSamplingProfiler::Frame( | |
435 frame_ip, | |
436 GetModuleIndex(frame_ip, current_modules, profile_module_index))); | |
437 }); | |
438 } | |
439 | |
440 } // namespace | |
441 | |
442 std::unique_ptr<NativeStackSampler> NativeStackSampler::Create( | |
443 PlatformThreadId thread_id, | |
444 AnnotateCallback annotator, | |
445 NativeStackSamplerTestDelegate* test_delegate) { | |
446 #if !defined(__x86_64__) | |
447 // No. | |
448 return nullptr; | |
449 #endif | |
450 return base::MakeUnique<NativeStackSamplerMac>(thread_id, annotator, | |
451 test_delegate); | |
452 } | |
453 | |
454 } // namespace base | |
OLD | NEW |