NativeStackSampler implementation for Mac.

base/profiler/native_stack_sampler_mac.cc

Index: base/profiler/native_stack_sampler_mac.cc
diff --git a/base/profiler/native_stack_sampler_mac.cc b/base/profiler/native_stack_sampler_mac.cc
new file mode 100644
index 0000000000000000000000000000000000000000..b196c09436de9923d239466b222eef06909b3de0
--- /dev/null
+++ b/base/profiler/native_stack_sampler_mac.cc
@@ -0,0 +1,324 @@
+// Copyright 2015 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 <dlfcn.h>
+#include <libkern/OSByteOrder.h>
+#include <mach/mach.h>
+#include <mach/thread_status.h>
+#include <mach-o/swap.h>
+#include <stdlib.h>
+#include <uuid/uuid.h>
+
+#include <map>
+
+#include "base/profiler/native_stack_sampler.h"
+#include "base/strings/string_number_conversions.h"
+
+namespace base {
+
+namespace {
+
+// Copy of x86_64 thread context structure from x86_thread_state64_t type.
+// Copied struct since fields can have different names on different versions of
+// Darwin.
+struct ThreadContext {
+  uint64_t rax;
+  uint64_t rbx;
+  uint64_t rcx;
+  uint64_t rdx;
+  uint64_t rdi;
+  uint64_t rsi;
+  uint64_t rbp;
+  uint64_t rsp;
+  uint64_t r8;
+  uint64_t r9;
+  uint64_t r10;
+  uint64_t r11;
+  uint64_t r12;
+  uint64_t r13;
+  uint64_t r14;
+  uint64_t r15;
+  uint64_t rip;
+  uint64_t rflags;
+  uint64_t cs;
+  uint64_t fs;
+  uint64_t gs;
+};
+
+// Struct for stack walking (represents stack state on any function call that
+// pushes a frame pointer).
+struct StackFrame {
+  // Pointer to caller's frame (rbp).
+  uintptr_t base_pointer;
+  // Address in caller for callee to return to.
+  uintptr_t return_address;
+};
+
+// Helper that swaps byte order in |x| if |swap| flag is set.
+uint32_t SwapIfBig32(uint32_t x, bool swap) {
+  if (swap)
+    return OSSwapBigToHostInt32(x);
+  return x;
+}
+
+// Overwrites |dst| with StackFrame referenced in |src|. Returns true if stack
+// read was successful. Failure implies address in |src| is corrupt.
+bool SafeStackFrameRead(uintptr_t src, StackFrame* dst) {
+  vm_size_t ignored_bytes_copied = 0;
+  return vm_read_overwrite(mach_task_self(), static_cast<vm_address_t>(src),
+                           static_cast<vm_size_t>(sizeof(StackFrame)),
+                           reinterpret_cast<vm_address_t>(dst),
+                           &ignored_bytes_copied) == KERN_SUCCESS;
+}
+
+// Functions related to retrieving Mach-O Identifer
+// This walker only supports modules built for x86_64 architecture. More detail
+// can be found in the OS X ABI Mach-O File Format Reference.
+
+// Returns offset in bytes where the x86_64 header is located in binary
+// loaded at |module_addr|. Returns 0 if |module_addr| is not a valid FAT
+// Mach-O binary or has not been built for x86_64.
+off_t GetMach64HeaderOffset(const void* module_addr) {
+  const fat_header* header = reinterpret_cast<const fat_header*>(module_addr);
+  if (header->magic != FAT_MAGIC && header->magic != FAT_CIGAM)
+    return 0;
+
+  // Search all FAT architectures for x86_64.
+  const fat_arch* fat_arches = reinterpret_cast<const fat_arch*>(
+      reinterpret_cast<const uint8_t*>(module_addr) + sizeof(header));
+  for (uint32_t i = 0; i < OSSwapBigToHostInt32(header->nfat_arch); ++i) {

[robliao, 2015/09/16 01:41:45]

Cache the result of OSSwapBigToHostInt32(header->nfat_arch)

[sydli, 2015/09/18 20:29:20]

Done.

+    const fat_arch& arch = fat_arches[i];
+    if (OSSwapBigToHostInt32(arch.cputype) == CPU_TYPE_X86_64)
+      return OSSwapBigToHostInt32(arch.offset);
+  }
+  return 0;
+}
+
+// Returns true if Mach-O binary at |module_addr| was built specifically for
+// x86_64 cpu architecture.
+bool IsX64Header(const void* module_addr) {
+  const mach_header_64* header =
+      reinterpret_cast<const mach_header_64*>(module_addr);
+  if (header->magic != MH_MAGIC_64 && header->magic != MH_CIGAM_64)
+    return false;
+  bool swap = header->magic == MH_CIGAM_64;
+  return SwapIfBig32(header->cputype, swap) == CPU_TYPE_X86_64;
+}
+
+// Fills |id| with the UUID of the x86_64 Mach-O binary loaded at |module_addr|.
+// |offset| is the offset in bytes into |module_addr| where the x86_64 header
+// is located. |offset| is only relevant if binary is FAT and contains
+// multiple architecture headers.  Returns false if header is malformed or the
+// header does not specify the UUID load command.
+bool GetX64UUIDAt(const void* module_addr, unsigned char* id, off_t offset) {
+  const mach_header_64* header = reinterpret_cast<const mach_header_64*>(
+      reinterpret_cast<const uint8_t*>(module_addr) + offset);
+  if (header->magic != MH_MAGIC_64 && header->magic != MH_CIGAM_64)
+    return false;
+  bool swap = header->magic == MH_CIGAM_64;

[robliao, 2015/09/16 01:41:45]

Nit: Group statement with the next set of statements. (e.g. linebreak above and
remove linebreak below.)

[sydli, 2015/09/18 20:29:20]

Done.

+
+  // Search all load commands for UUID command.
+  offset += sizeof(mach_header_64);
+  for (uint32_t i = 0; i < SwapIfBig32(header->ncmds, swap); ++i) {
+    const load_command* current_cmd = reinterpret_cast<const load_command*>(
+        reinterpret_cast<const uint8_t*>(module_addr) + offset);
+
+    if (SwapIfBig32(current_cmd->cmd, swap) == LC_UUID) {
+      const uuid_command* uuid_cmd =
+          reinterpret_cast<const uuid_command*>(current_cmd);
+      memcpy(id, &uuid_cmd->uuid, sizeof(uuid_t));
+      return true;
+    }
+    offset += SwapIfBig32(current_cmd->cmdsize, swap);
+  }
+  return false;
+}
+
+// Fills |id| with Mach-O UUID retrieved from Mach-O binary loaded at
+// |module_addr|. This function returns false if the binary was not built for
+// X86_64 or if UUID cannot be found.
+bool GetUUID(const void* module_addr, unsigned char* id) {
+  off_t offset = 0;
+  // If module is not x86_64 exclusive, it could be a module that supports
+  // multiple architectures. In this case, the appropriate header will be at
+  // some non-zero offset.
+  if (!IsX64Header(module_addr) &&
+      !(offset = GetMach64HeaderOffset(module_addr))) {
+    return false;
+  }
+  return GetX64UUIDAt(module_addr, id, offset);
+}
+
+// Returns hex encoding of a 16-byte ID for binary loaded at |module_addr|.
+// Returns empty string if UUID cannot be found at |module_addr|.
+std::string GetUniqueId(const void* module_addr) {
+  unsigned char id[sizeof(uuid_t)];
+  if (!GetUUID(module_addr, id))
+    return "";
+  return HexEncode(id, sizeof(uuid_t));
+}
+
+// Functions related to grabbing a stack trace ---------------------------------
+
+// Fills |state| with |target_thread|'s context.
+bool GetThreadContext(thread_act_t target_thread, ThreadContext* state) {
+  mach_msg_type_number_t count =
+      static_cast<mach_msg_type_number_t>(MACHINE_THREAD_STATE_COUNT);
+  return thread_get_state(target_thread, x86_THREAD_STATE64,
+                          reinterpret_cast<thread_state_t>(state),
+                          &count) == KERN_SUCCESS;
+}
+
+// Walks |thread_handle|'s stack and fills |instruction_pointers|.
+// Returns number of frames in stack, unless there's a corrupt frame pointer
+// (likely if module compiled with -fno_omit_frame_pointer), in which this
+// function will return the number of frames up until the frame with a corrupt
+// frame pointer. This procedure occurs while thread is suspended, so it should
+// take as little time as possible.
+int RecordStack(mach_port_t thread_handle,
+                int max_stack_size,
+                uintptr_t instruction_pointers[]) {
+  ThreadContext state;
+  if (!GetThreadContext(thread_handle, &state))
+    return 0;
+
+  StackFrame frame;
+  frame.base_pointer = state.rbp;
+  int i = 0;
+  for (; i < max_stack_size; i++) {
+    // Three cases for end-of-stack condition:
+    //   1) A frame pointer was corrupt.
+    //   2) The next recovered rsp is not lower than the previously recovered
+    //      one, indicating the stack is not growing down.
+    //   3) Return (instruction) address is 0.
+    uintptr_t old_rsp = frame.base_pointer;
+    if (!SafeStackFrameRead(frame.base_pointer, &frame) ||
+        frame.base_pointer < old_rsp || !frame.return_address) {
+      return i;
+    }
+    instruction_pointers[i] = frame.return_address;
+  }
+  return i;
+}
+
+// NativeStackSamplerMac ------------------------------------------------------
+
+class NativeStackSamplerMac : public NativeStackSampler {
+ public:
+  explicit NativeStackSamplerMac(pid_t thread_handle);
+  ~NativeStackSamplerMac() override;
+
+  // StackSamplingProfiler::NativeStackSampler:
+  void ProfileRecordingStarting(
+      std::vector<StackSamplingProfiler::Module>* modules) override;
+  void RecordStackSample(StackSamplingProfiler::Sample* sample) override;
+  void ProfileRecordingStopped() override;
+
+ private:
+  // Gets the index for the Module containing |instruction_pointer| in
+  // |modules|, adding it if it's not already present. Returns
+  // StackSamplingProfiler::Frame::kUnknownModuleIndex if no Module can be
+  // determined for |module|.
+  size_t GetModuleIndex(const uintptr_t instruction_pointer,
+                        std::vector<StackSamplingProfiler::Module>* modules);
+
+  // Copies the stack information represented by |instruction_pointers| into
+  // |sample| and |modules|.
+  void CopyToSample(const uintptr_t instruction_pointers[],
+                    int stack_depth,
+                    StackSamplingProfiler::Sample* sample,
+                    std::vector<StackSamplingProfiler::Module>* modules);
+
+  // Weak reference: Mach handle for thread being profiled.
+  mach_port_t thread_handle_;
+
+  // Weak. Points to the modules associated with the profile being recorded
+  // between ProfileRecordingStarting() and ProfileRecordingStopped().
+  std::vector<StackSamplingProfiler::Module>* current_modules_;
+
+  // Maps a module's base address to the corresponding Module's index within
+  // current_modules_.
+  std::map<const void*, size_t> profile_module_index_;
+
+  DISALLOW_COPY_AND_ASSIGN(NativeStackSamplerMac);
+};
+
+// The PlatformThreadId of the given thread is actually a typedef of
+// mach_port_t.
+// (base/threading/platform_thread_posix.cc:128)
+NativeStackSamplerMac::NativeStackSamplerMac(pid_t thread_handle)
+    : thread_handle_(static_cast<mach_port_t>(thread_handle)),
+      current_modules_(nullptr) {}
+
+NativeStackSamplerMac::~NativeStackSamplerMac() {}
+
+void NativeStackSamplerMac::ProfileRecordingStarting(
+    std::vector<StackSamplingProfiler::Module>* modules) {
+  current_modules_ = modules;
+  profile_module_index_.clear();
+}
+
+void NativeStackSamplerMac::RecordStackSample(
+    StackSamplingProfiler::Sample* sample) {
+  DCHECK(current_modules_);
+
+  const int max_stack_size = 64;

[robliao, 2015/09/16 01:41:45]

Optional: It would be nice to unify this value somehow.

+  uintptr_t instruction_pointers[max_stack_size] = {0};
+
+  thread_suspend(thread_handle_);

[Robert Sesek, 2015/09/17 22:06:56]

In the email thread, Mark mentioned that this could deadlock if the thread about
to be suspended is holding a lock in dyld. He gave some ideas for handling that
you may want to consider.

[sydli, 2015/09/18 20:29:20]

Now forcibly binds library used during stack walk. Not sure about the best way
to find the kernel library pathname, though; assuming it's always in
/usr/lib/system feels very wrong. Is there a better way?

+  int stack_depth =
+      RecordStack(thread_handle_, max_stack_size, instruction_pointers);
+  thread_resume(thread_handle_);
+  CopyToSample(instruction_pointers, stack_depth, sample, current_modules_);
+}
+
+void NativeStackSamplerMac::ProfileRecordingStopped() {
+  current_modules_ = nullptr;
+}
+
+size_t NativeStackSamplerMac::GetModuleIndex(
+    const uintptr_t instruction_pointer,
+    std::vector<StackSamplingProfiler::Module>* modules) {
+  Dl_info inf;
+  if (!dladdr(reinterpret_cast<const void*>(instruction_pointer), &inf))
+    return StackSamplingProfiler::Frame::kUnknownModuleIndex;
+
+  auto module_index = profile_module_index_.find(inf.dli_fbase);
+  if (module_index == profile_module_index_.end()) {
+    StackSamplingProfiler::Module module(inf.dli_fbase,
+                                         GetUniqueId(inf.dli_fbase),
+                                         base::FilePath(inf.dli_fname));
+    modules->push_back(module);
+    module_index = profile_module_index_.insert(
+        std::make_pair(inf.dli_fbase, modules->size() - 1)).first;
+  }
+  return module_index->second;
+}
+
+void NativeStackSamplerMac::CopyToSample(
+    const uintptr_t instruction_pointers[],
+    int stack_depth,
+    StackSamplingProfiler::Sample* sample,
+    std::vector<StackSamplingProfiler::Module>* modules) {
+  sample->clear();
+  sample->reserve(stack_depth);
+
+  for (int i = 0; i < stack_depth; i++) {
+    sample->push_back(StackSamplingProfiler::Frame(
+        reinterpret_cast<const void*>(instruction_pointers[i]),
+        GetModuleIndex(instruction_pointers[i], modules)));
+  }
+}
+
+}  // namespace
+
+scoped_ptr<NativeStackSampler> NativeStackSampler::Create(
+    PlatformThreadId thread_id) {
+#if defined(__i386__)
+  return nullptr;
+#endif
+  return scoped_ptr<NativeStackSampler>(new NativeStackSamplerMac(thread_id));
+}
+
+}  // namespace base