Quelle StackWalk.cpp Sprache: C

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */

/* API for getting a stack trace of the C/C++ stack on the current thread */

#include "mozilla/Array.h"
#include "mozilla/ArrayUtils.h"
#include "mozilla/Atomics.h"
#include "mozilla/Attributes.h"
#include "mozilla/StackWalk.h"
#ifdef XP_WIN
#  include "mozilla/StackWalkThread.h"
#  include <io.h>
#else
#  include <unistd.h>
#endif
#include "mozilla/Sprintf.h"

#include <string.h>

#if defined(ANDROID) && defined(MOZ_LINKER)
#  include "Linker.h"
#  include <android/log.h>
#endif

using namespace mozilla;

// for _Unwind_Backtrace from libcxxrt or libunwind
// cxxabi.h from libcxxrt implicitly includes unwind.h first
#if defined(HAVE__UNWIND_BACKTRACE) && !defined(_GNU_SOURCE)
#  define _GNU_SOURCE
#endif

#if defined(HAVE_DLFCN_H) || defined(XP_DARWIN)
#  include <dlfcn.h>
#endif

#if (defined(XP_DARWIN) && \
     (defined(__i386) || defined(__ppc__) || defined(HAVE__UNWIND_BACKTRACE)))
#  define MOZ_STACKWALK_SUPPORTS_MACOSX 1
#else
#  define MOZ_STACKWALK_SUPPORTS_MACOSX 0
#endif

#if __GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 1)
#  define HAVE___LIBC_STACK_END 1
#else
#  define HAVE___LIBC_STACK_END 0
#endif

#if (defined(linux) &&                                            \
     ((defined(__GNUC__) && (defined(__i386) || defined(PPC))) || \
      defined(HAVE__UNWIND_BACKTRACE)) &&                         \
     (HAVE___LIBC_STACK_END || ANDROID))
#  define MOZ_STACKWALK_SUPPORTS_LINUX 1
#else
#  define MOZ_STACKWALK_SUPPORTS_LINUX 0
#endif

#if HAVE___LIBC_STACK_END
extern MOZ_EXPORT void* __libc_stack_end;  // from ld-linux.so
#  ifdef __aarch64__
static Atomic<uintptr_t> ldso_base;
#  endif
#endif

#ifdef ANDROID
#  include <algorithm>
#  include <unistd.h>
#  include <pthread.h>
#endif

class FrameSkipper {
public:
  constexpr FrameSkipper() : mSkipUntilAddr(0) {}
  static uintptr_t AddressFromPC(const void* aPC) {
#ifdef __arm__
    // On 32-bit ARM, mask off the thumb bit to get the instruction address.
    return uintptr_t(aPC) & ~1;
#else
    return uintptr_t(aPC);
#endif
  }
  bool ShouldSkipPC(void* aPC) {
    // Skip frames until we encounter the one we were initialized with,
    // and then never skip again.
    uintptr_t instructionAddress = AddressFromPC(aPC);
    if (mSkipUntilAddr != 0) {
      if (mSkipUntilAddr != instructionAddress) {
        return true;
      }
      mSkipUntilAddr = 0;
    }
    return false;
  }
  explicit FrameSkipper(const void* aPC) : mSkipUntilAddr(AddressFromPC(aPC)) {}

private:
  uintptr_t mSkipUntilAddr;
};

#ifdef XP_WIN

#  include <windows.h>
#  include <process.h>
#  include <stdio.h>
#  include <malloc.h>
#  include "mozilla/ArrayUtils.h"
#  include "mozilla/Atomics.h"
#  include "mozilla/StackWalk_windows.h"
#  include "mozilla/WindowsVersion.h"

#  include <imagehlp.h>
// We need a way to know if we are building for WXP (or later), as if we are, we
// need to use the newer 64-bit APIs. API_VERSION_NUMBER seems to fit the bill.
// A value of 9 indicates we want to use the new APIs.
#  if API_VERSION_NUMBER < 9
#    error Too old imagehlp.h
#  endif

#  if defined(_M_AMD64) || defined(_M_ARM64)
// We must use RtlLookupFunctionEntry to do stack walking on x86-64 and arm64,
// but internally this function does a blocking shared acquire of SRW locks
// that live in ntdll and are not exported. This is problematic when we want to
// suspend a thread and walk its stack, like we do in the profiler and the
// background hang reporter. If the suspended thread happens to hold one of the
// locks exclusively while suspended, then the stack walking thread will
// deadlock if it calls RtlLookupFunctionEntry.
//
// Note that we only care about deadlocks between the stack walking thread and
// the suspended thread. Any other deadlock scenario is considered out of
// scope, because they are unlikely to be our fault -- these other scenarios
// imply that some thread that we did not suspend is stuck holding one of the
// locks exclusively, and exclusive acquisition of these locks only happens for
// a brief time during Microsoft API calls (e.g. LdrLoadDll, LdrUnloadDll).
//
// We use one of two alternative strategies to gracefully fail to capture a
// stack instead of running into a deadlock:
//    (1) collect pointers to the ntdll internal locks at stack walk
//        initialization, then try to acquire them non-blockingly before
//        initiating any stack walk;
// or (2) mark all code paths that can potentially end up doing an exclusive
//        acquisition of the locks as stack walk suppression paths, then check
//        if any thread is currently on a stack walk suppression path before
//        initiating any stack walk;
//
// Strategy (2) can only avoid all deadlocks under the easily wronged
// assumption that we have correctly identified all existing paths that should
// be stack suppression paths. With strategy (2) we cannot collect stacks e.g.
// during the whole duration of a DLL load happening on any thread so the
// profiling results are worse.
//
// Strategy (1) guarantees no deadlock. It also gives better profiling results
// because it is more fine-grained. Therefore we always prefer strategy (1),
// and we only use strategy (2) as a fallback.

// Strategy (1): Ntdll Internal Locks
//
// The external stack walk initialization code will feed us pointers to the
// ntdll internal locks. Once we have them, we no longer need to rely on
// strategy (2).
static Atomic<bool> sStackWalkLocksInitialized;
static Array<SRWLOCK*, 2> sStackWalkLocks;

MFBT_API
void InitializeStackWalkLocks(const Array<void*, 2>& aStackWalkLocks) {
  sStackWalkLocks[0] = reinterpret_cast<SRWLOCK*>(aStackWalkLocks[0]);
  sStackWalkLocks[1] = reinterpret_cast<SRWLOCK*>(aStackWalkLocks[1]);
  sStackWalkLocksInitialized = true;
}

// Strategy (2): Stack Walk Suppressions
//
// We're using an atomic counter rather than a critical section because we
// don't require mutual exclusion with the stack walker. If the stack walker
// determines that it's safe to start unwinding the suspended thread (i.e.
// there are no suppressions when the unwind begins), then it's safe to
// continue unwinding that thread even if other threads request suppressions
// in the meantime, because we can't deadlock with those other threads.
//
// XXX: This global variable is a larger-than-necessary hammer. A more scoped
// solution would be to maintain a counter per thread, but then it would be
// more difficult for WalkStackMain64 to read the suspended thread's counter.
static Atomic<size_t> sStackWalkSuppressions;

void SuppressStackWalking() { ++sStackWalkSuppressions; }

void DesuppressStackWalking() {
  auto previousValue = sStackWalkSuppressions--;
  // We should never desuppress from 0. See bug 1687510 comment 10 for an
  // example in which this occured.
  MOZ_RELEASE_ASSERT(previousValue);
}

MFBT_API
AutoSuppressStackWalking::AutoSuppressStackWalking() { SuppressStackWalking(); }

MFBT_API
AutoSuppressStackWalking::~AutoSuppressStackWalking() {
  DesuppressStackWalking();
}

bool IsStackWalkingSafe() {
  // Use strategy (1), if initialized.
  if (sStackWalkLocksInitialized) {
    bool isSafe = false;
    if (::TryAcquireSRWLockShared(sStackWalkLocks[0])) {
      if (::TryAcquireSRWLockShared(sStackWalkLocks[1])) {
        isSafe = true;
        ::ReleaseSRWLockShared(sStackWalkLocks[1]);
      }
      ::ReleaseSRWLockShared(sStackWalkLocks[0]);
    }
    return isSafe;
  }

  // Otherwise, fall back to strategy (2).
  return sStackWalkSuppressions == 0;
}

static uint8_t* sJitCodeRegionStart;
static size_t sJitCodeRegionSize;
uint8_t* sMsMpegJitCodeRegionStart;
size_t sMsMpegJitCodeRegionSize;

MFBT_API void RegisterJitCodeRegion(uint8_t* aStart, size_t aSize) {
  // Currently we can only handle one JIT code region at a time
  MOZ_RELEASE_ASSERT(!sJitCodeRegionStart);

  sJitCodeRegionStart = aStart;
  sJitCodeRegionSize = aSize;
}

MFBT_API void UnregisterJitCodeRegion(uint8_t* aStart, size_t aSize) {
  // Currently we can only handle one JIT code region at a time
  MOZ_RELEASE_ASSERT(sJitCodeRegionStart && sJitCodeRegionStart == aStart &&
                     sJitCodeRegionSize == aSize);

  sJitCodeRegionStart = nullptr;
  sJitCodeRegionSize = 0;
}

#  endif  // _M_AMD64 || _M_ARM64

// Routine to print an error message to standard error.
static void PrintError(const char* aPrefix) {
  LPSTR lpMsgBuf;
  DWORD lastErr = GetLastError();
  FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM |
                     FORMAT_MESSAGE_IGNORE_INSERTS,
                 nullptr, lastErr,
                 MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),  // Default language
                 (LPSTR)&lpMsgBuf, 0, nullptr);
  fprintf(stderr, "### ERROR: %s: %s", aPrefix,
          lpMsgBuf ? lpMsgBuf : "(null)\n");
  fflush(stderr);
  LocalFree(lpMsgBuf);
}

class MOZ_RAII AutoCriticalSection {
public:
  explicit inline AutoCriticalSection(LPCRITICAL_SECTION aCriticalSection)
      : mCriticalSection{aCriticalSection} {
    ::EnterCriticalSection(mCriticalSection);
  }
  inline ~AutoCriticalSection() { ::LeaveCriticalSection(mCriticalSection); }

  AutoCriticalSection(AutoCriticalSection&& other) = delete;
  AutoCriticalSection operator=(AutoCriticalSection&& other) = delete;
  AutoCriticalSection(const AutoCriticalSection&) = delete;
  AutoCriticalSection operator=(const AutoCriticalSection&) = delete;

private:
  LPCRITICAL_SECTION mCriticalSection;
};

// A thread-safe safe object interface for Microsoft's DbgHelp.dll. DbgHelp
// APIs are not thread-safe and they require the use of a unique HANDLE value
// as an identifier for the current session. All this is handled internally by
// DbgHelpWrapper.
class DbgHelpWrapper {
public:
  explicit inline DbgHelpWrapper() : DbgHelpWrapper(InitFlag::BasicInit) {}
  DbgHelpWrapper(DbgHelpWrapper&& other) = delete;
  DbgHelpWrapper operator=(DbgHelpWrapper&& other) = delete;
  DbgHelpWrapper(const DbgHelpWrapper&) = delete;
  DbgHelpWrapper operator=(const DbgHelpWrapper&) = delete;

  inline bool ReadyToUse() { return mInitSuccess; }

  inline BOOL StackWalk64(
      DWORD aMachineType, HANDLE aThread, LPSTACKFRAME64 aStackFrame,
      PVOID aContextRecord, PREAD_PROCESS_MEMORY_ROUTINE64 aReadMemoryRoutine,
      PFUNCTION_TABLE_ACCESS_ROUTINE64 aFunctionTableAccessRoutine,
      PGET_MODULE_BASE_ROUTINE64 aGetModuleBaseRoutine,
      PTRANSLATE_ADDRESS_ROUTINE64 aTranslateAddress) {
    if (!ReadyToUse()) {
      return FALSE;
    }

    AutoCriticalSection guard(&sCriticalSection);
    return ::StackWalk64(aMachineType, sSessionId, aThread, aStackFrame,
                         aContextRecord, aReadMemoryRoutine,
                         aFunctionTableAccessRoutine, aGetModuleBaseRoutine,
                         aTranslateAddress);
  }

protected:
  enum class InitFlag : bool {
    BasicInit,
    WithSymbolSupport,
  };

  explicit inline DbgHelpWrapper(InitFlag initFlag) {
    mInitSuccess = Initialize(initFlag);
  }

  // DbgHelp functions are not thread-safe and should therefore be protected
  // by using this critical section through a AutoCriticalSection.
  static CRITICAL_SECTION sCriticalSection;

  // DbgHelp functions require a unique HANDLE hProcess that should be the same
  // throughout the current session. We refer to this handle as a session id.
  // Ideally the session id should be a valid HANDLE to the target process,
  // which in our case is the current process.
  //
  // However, in order to avoid conflicts with other sessions, the session id
  // should be unique and therefore not just GetCurrentProcess(), which other
  // pieces of code tend to already use (see bug 1699328).
  //
  // We therefore define sSessionId as a duplicate of the current process
  // handle, a solution that meets all the requirements listed above.
  static HANDLE sSessionId;

private:
  bool mInitSuccess;

  // This function initializes sCriticalSection, sSessionId and loads DbgHelp.
  // It also calls SymInitialize if called with aInitFlag::WithSymbolSupport.
  // It is thread-safe and reentrancy-safe.
  [[nodiscard]] static bool Initialize(InitFlag aInitFlag);

  // In debug and fuzzing builds, MOZ_ASSERT and MOZ_CRASH walk the stack to
  // print it before actually crashing. This code path uses a DbgHelpWrapper
  // object, hence *any* MOZ_ASSERT or MOZ_CRASH failure reached from
  // Initialize() leads to rentrancy (see bug 1869997 for an example). Such
  // failures can occur indirectly when we load dbghelp.dll, because we
  // override various Microsoft-internal functions that are called upon DLL
  // loading. We protect against reentrancy by keeping track of the ID of the
  // thread that runs the initialization code.
  static Atomic<DWORD> sInitializationThreadId;
};

CRITICAL_SECTION DbgHelpWrapper::sCriticalSection{};
Atomic<DWORD> DbgHelpWrapper::sInitializationThreadId{0};
HANDLE DbgHelpWrapper::sSessionId{nullptr};

// Thread-safety here is ensured by the C++ standard: scoped static
// initialization is thread-safe. sInitializationThreadId is used to protect
// against reentrancy -- and only for that purpose.
[[nodiscard]] /* static */ bool DbgHelpWrapper::Initialize(
    DbgHelpWrapper::InitFlag aInitFlag) {
  // In the code below, it is only safe to reach MOZ_ASSERT or MOZ_CRASH while
  // sInitializationThreadId is set to the current thread id.
  static Atomic<DWORD> sInitializationThreadId{0};
  DWORD currentThreadId = ::GetCurrentThreadId();

  // This code relies on Windows never giving us a current thread ID of zero.
  // We make this assumption explicit, by failing if that should ever occur.
  if (!currentThreadId) {
    return false;
  }

  if (sInitializationThreadId == currentThreadId) {
    // This is a reentrant call and we must abort here.
    return false;
  }

  static const bool sHasInitializedDbgHelp = [currentThreadId]() {
    // Per the C++ standard, only one thread evers reaches this path.
    sInitializationThreadId = currentThreadId;

    ::InitializeCriticalSection(&sCriticalSection);

    if (!::DuplicateHandle(GetCurrentProcess(), GetCurrentProcess(),
                           GetCurrentProcess(), &sSessionId, 0, FALSE,
                           DUPLICATE_SAME_ACCESS)) {
      return false;
    }

    bool dbgHelpLoaded = static_cast<bool>(::LoadLibraryW(L"dbghelp.dll"));

    MOZ_ASSERT(dbgHelpLoaded);
    sInitializationThreadId = 0;
    return dbgHelpLoaded;
  }();

  // If we don't need symbol initialization, we are done. If we need it, we
  // can only proceed if DbgHelp initialization was successful.
  if (aInitFlag == InitFlag::BasicInit || !sHasInitializedDbgHelp) {
    return sHasInitializedDbgHelp;
  }

  static const bool sHasInitializedSymbols = [currentThreadId]() {
    // Per the C++ standard, only one thread evers reaches this path.
    sInitializationThreadId = currentThreadId;

    bool symbolsInitialized = false;

    {
      AutoCriticalSection guard(&sCriticalSection);
      ::SymSetOptions(SYMOPT_LOAD_LINES | SYMOPT_UNDNAME);
      symbolsInitialized =
          static_cast<bool>(::SymInitializeW(sSessionId, nullptr, TRUE));
      /* XXX At some point we need to arrange to call SymCleanup */
    }

    if (!symbolsInitialized) {
      PrintError("SymInitialize");
    }

    MOZ_ASSERT(symbolsInitialized);
    sInitializationThreadId = 0;
    return symbolsInitialized;
  }();

  return sHasInitializedSymbols;
}

// Some APIs such as SymFromAddr also require that the session id has gone
// through SymInitialize. This is handled by child class DbgHelpWrapperSym.
class DbgHelpWrapperSym : public DbgHelpWrapper {
public:
  explicit DbgHelpWrapperSym() : DbgHelpWrapper(InitFlag::WithSymbolSupport) {}

  inline BOOL SymFromAddr(DWORD64 aAddress, PDWORD64 aDisplacement,
                          PSYMBOL_INFO aSymbol) {
    if (!ReadyToUse()) {
      return FALSE;
    }

    AutoCriticalSection guard(&sCriticalSection);
    return ::SymFromAddr(sSessionId, aAddress, aDisplacement, aSymbol);
  }

  BOOL SymGetModuleInfoEspecial64(DWORD64 aAddr, PIMAGEHLP_MODULE64 aModuleInfo,
                                  PIMAGEHLP_LINE64 aLineInfo);

private:
  // Helpers for SymGetModuleInfoEspecial64
  struct CallbackEspecial64UserContext {
    HANDLE mSessionId;
    DWORD64 mAddr;
  };

  static BOOL CALLBACK CallbackEspecial64(PCSTR aModuleName,
                                          DWORD64 aModuleBase,
                                          ULONG aModuleSize,
                                          PVOID aUserContext);
};

// Wrapper around a reference to a CONTEXT, to simplify access to main
// platform-specific execution registers.
// It also avoids using CONTEXT* nullable pointers.
class CONTEXTGenericAccessors {
public:
  explicit CONTEXTGenericAccessors(CONTEXT& aCONTEXT) : mCONTEXT(aCONTEXT) {}

  CONTEXT* CONTEXTPtr() { return &mCONTEXT; }

  inline auto& PC() {
#  if defined(_M_AMD64)
    return mCONTEXT.Rip;
#  elif defined(_M_ARM64)
    return mCONTEXT.Pc;
#  elif defined(_M_IX86)
    return mCONTEXT.Eip;
#  else
#    error "unknown platform"
#  endif
  }

  inline auto& SP() {
#  if defined(_M_AMD64)
    return mCONTEXT.Rsp;
#  elif defined(_M_ARM64)
    return mCONTEXT.Sp;
#  elif defined(_M_IX86)
    return mCONTEXT.Esp;
#  else
#    error "unknown platform"
#  endif
  }

  inline auto& BP() {
#  if defined(_M_AMD64)
    return mCONTEXT.Rbp;
#  elif defined(_M_ARM64)
    return mCONTEXT.Fp;
#  elif defined(_M_IX86)
    return mCONTEXT.Ebp;
#  else
#    error "unknown platform"
#  endif
  }

private:
  CONTEXT& mCONTEXT;
};

/**
* Walk the stack, translating PC's found into strings and recording the
* chain in aBuffer. For this to work properly, the DLLs must be rebased
* so that the address in the file agrees with the address in memory.
* Otherwise StackWalk will return FALSE when it hits a frame in a DLL
* whose in memory address doesn't match its in-file address.
*/

static void DoMozStackWalkThread(MozWalkStackCallback aCallback,
                                 const void* aFirstFramePC, uint32_t aMaxFrames,
                                 void* aClosure, HANDLE aThread,
                                 CONTEXT* aContext) {
#  if defined(_M_IX86)
  DbgHelpWrapper dbgHelp;
  if (!dbgHelp.ReadyToUse()) {
    return;
  }
#  endif

  HANDLE targetThread = aThread;
  bool walkCallingThread;
  if (!targetThread) {
    targetThread = ::GetCurrentThread();
    walkCallingThread = true;
  } else {
    DWORD targetThreadId = ::GetThreadId(targetThread);
    DWORD currentThreadId = ::GetCurrentThreadId();
    walkCallingThread = (targetThreadId == currentThreadId);
  }

  // If not already provided, get a context for the specified thread.
  CONTEXT context_buf;
  if (!aContext) {
    memset(&context_buf, 0, sizeof(CONTEXT));
    context_buf.ContextFlags = CONTEXT_FULL;
    if (walkCallingThread) {
      ::RtlCaptureContext(&context_buf);
    } else if (!GetThreadContext(targetThread, &context_buf)) {
      return;
    }
  }
  CONTEXTGenericAccessors context{aContext ? *aContext : context_buf};

#  if defined(_M_IX86)
  // Setup initial stack frame to walk from.
  STACKFRAME64 frame64;
  memset(&frame64, 0, sizeof(frame64));
  frame64.AddrPC.Offset = context.PC();
  frame64.AddrStack.Offset = context.SP();
  frame64.AddrFrame.Offset = context.BP();
  frame64.AddrPC.Mode = AddrModeFlat;
  frame64.AddrStack.Mode = AddrModeFlat;
  frame64.AddrFrame.Mode = AddrModeFlat;
  frame64.AddrReturn.Mode = AddrModeFlat;
#  endif

#  if defined(_M_AMD64) || defined(_M_ARM64)
  // If at least one thread (we don't know which) may be holding a lock that
  // can deadlock RtlLookupFunctionEntry, we can't proceed because that thread
  // may be the one that we're trying to walk the stack of.
  //
  // But if there is no such thread by this point, then our target thread can't
  // be holding a lock, so it's safe to proceed. By virtue of being suspended,
  // the target thread can't acquire any new locks during our stack walking, so
  // we only need to do this check once. Other threads may temporarily acquire
  // the locks while we're walking the stack, but that's mostly fine -- calling
  // RtlLookupFunctionEntry will make us wait for them to release the locks,
  // but at least we won't deadlock.
  if (!IsStackWalkingSafe()) {
    return;
  }

  bool firstFrame = true;
#  endif

  FrameSkipper skipper(aFirstFramePC);

  uint32_t frames = 0;

  // Now walk the stack.
  while (true) {
    DWORD64 addr;
    DWORD64 spaddr;

#  if defined(_M_IX86)
    // 32-bit frame unwinding.
    BOOL ok = dbgHelp.StackWalk64(
        IMAGE_FILE_MACHINE_I386, targetThread, &frame64, context.CONTEXTPtr(),
        nullptr,
        ::SymFunctionTableAccess64,  // function table access routine
        ::SymGetModuleBase64,        // module base routine
        0);

    if (ok) {
      addr = frame64.AddrPC.Offset;
      spaddr = frame64.AddrStack.Offset;
    } else {
      addr = 0;
      spaddr = 0;
      if (walkCallingThread) {
        PrintError("WalkStack64");
      }
    }

    if (!ok) {
      break;
    }

#  elif defined(_M_AMD64) || defined(_M_ARM64)

    auto currentInstr = context.PC();

    // If we reach a frame in JIT code, we don't have enough information to
    // unwind, so we have to give up.
    if (sJitCodeRegionStart && (uint8_t*)currentInstr >= sJitCodeRegionStart &&
        (uint8_t*)currentInstr < sJitCodeRegionStart + sJitCodeRegionSize) {
      break;
    }

    // We must also avoid msmpeg2vdec.dll's JIT region: they don't generate
    // unwind data, so their JIT unwind callback just throws up its hands and
    // terminates the process.
    if (sMsMpegJitCodeRegionStart &&
        (uint8_t*)currentInstr >= sMsMpegJitCodeRegionStart &&
        (uint8_t*)currentInstr <
            sMsMpegJitCodeRegionStart + sMsMpegJitCodeRegionSize) {
      break;
    }

    // 64-bit frame unwinding.
    // Try to look up unwind metadata for the current function.
    ULONG64 imageBase;
    PRUNTIME_FUNCTION runtimeFunction =
        RtlLookupFunctionEntry(currentInstr, &imageBase, NULL);

    if (runtimeFunction) {
      PVOID dummyHandlerData;
      ULONG64 dummyEstablisherFrame;
      RtlVirtualUnwind(UNW_FLAG_NHANDLER, imageBase, currentInstr,
                       runtimeFunction, context.CONTEXTPtr(), &dummyHandlerData,
                       &dummyEstablisherFrame, nullptr);
    } else if (firstFrame) {
      // Leaf functions can be unwound by hand.
      context.PC() = *reinterpret_cast<DWORD64*>(context.SP());
      context.SP() += sizeof(void*);
    } else {
      // Something went wrong.
      break;
    }

    addr = context.PC();
    spaddr = context.SP();
    firstFrame = false;
#  else
#    error "unknown platform"
#  endif

    if (addr == 0) {
      break;
    }

    if (skipper.ShouldSkipPC((void*)addr)) {
      continue;
    }

    aCallback(++frames, (void*)addr, (void*)spaddr, aClosure);

    if (aMaxFrames != 0 && frames == aMaxFrames) {
      break;
    }

#  if defined(_M_IX86)
    if (frame64.AddrReturn.Offset == 0) {
      break;
    }
#  endif
  }
}

MFBT_API void MozStackWalkThread(MozWalkStackCallback aCallback,
                                 uint32_t aMaxFrames, void* aClosure,
                                 HANDLE aThread, CONTEXT* aContext) {
  // We don't pass a aFirstFramePC because we walk the stack for another
  // thread.
  DoMozStackWalkThread(aCallback, nullptr, aMaxFrames, aClosure, aThread,
                       aContext);
}

MFBT_API void MozStackWalk(MozWalkStackCallback aCallback,
                           const void* aFirstFramePC, uint32_t aMaxFrames,
                           void* aClosure) {
  DoMozStackWalkThread(aCallback, aFirstFramePC ? aFirstFramePC : CallerPC(),
                       aMaxFrames, aClosure, nullptr, nullptr);
}

/* static */ BOOL CALLBACK
DbgHelpWrapperSym::CallbackEspecial64(PCSTR aModuleName, DWORD64 aModuleBase,
                                      ULONG aModuleSize, PVOID aUserContext) {
  BOOL retval = TRUE;
  auto context = reinterpret_cast<CallbackEspecial64UserContext*>(aUserContext);
  DWORD64 addr = context->mAddr;

  /*
   * You'll want to control this if we are running on an
   *  architecture where the addresses go the other direction.
   * Not sure this is even a realistic consideration.
   */
  const BOOL addressIncreases = TRUE;

  /*
   * If it falls in side the known range, load the symbols.
   */
  if (addressIncreases
          ? (addr >= aModuleBase && addr <= (aModuleBase + aModuleSize))
          : (addr <= aModuleBase && addr >= (aModuleBase - aModuleSize))) {
    retval = !!::SymLoadModule64(context->mSessionId, nullptr, aModuleName,
                                 nullptr, aModuleBase, aModuleSize);
    if (!retval) {
      PrintError("SymLoadModule64");
    }
  }

  return retval;
}

/*
* SymGetModuleInfoEspecial
*
* Attempt to determine the module information.
* Bug 112196 says this DLL may not have been loaded at the time
*  SymInitialize was called, and thus the module information
*  and symbol information is not available.
* This code rectifies that problem.
*/

// New members were added to IMAGEHLP_MODULE64 (that show up in the
// Platform SDK that ships with VC8, but not the Platform SDK that ships
// with VC7.1, i.e., between DbgHelp 6.0 and 6.1), but we don't need to
// use them, and it's useful to be able to function correctly with the
// older library.  (Stock Windows XP SP2 seems to ship with dbghelp.dll
// version 5.1.)  Since Platform SDK version need not correspond to
// compiler version, and the version number in debughlp.h was NOT bumped
// when these changes were made, ifdef based on a constant that was
// added between these versions.
#  ifdef SSRVOPT_SETCONTEXT
#    define NS_IMAGEHLP_MODULE64_SIZE                                        \
      (((offsetof(IMAGEHLP_MODULE64, LoadedPdbName) + sizeof(DWORD64) - 1) / \
        sizeof(DWORD64)) *                                                   \
       sizeof(DWORD64))
#  else
#    define NS_IMAGEHLP_MODULE64_SIZE sizeof(IMAGEHLP_MODULE64)
#  endif

BOOL DbgHelpWrapperSym::SymGetModuleInfoEspecial64(
    DWORD64 aAddr, PIMAGEHLP_MODULE64 aModuleInfo, PIMAGEHLP_LINE64 aLineInfo) {
  if (!ReadyToUse()) {
    return FALSE;
  }

  AutoCriticalSection guard(&sCriticalSection);
  BOOL retval = FALSE;

  /*
   * Init the vars if we have em.
   */
  aModuleInfo->SizeOfStruct = NS_IMAGEHLP_MODULE64_SIZE;
  if (aLineInfo) {
    aLineInfo->SizeOfStruct = sizeof(IMAGEHLP_LINE64);
  }

  /*
   * Give it a go.
   * It may already be loaded.
   */
  retval = ::SymGetModuleInfo64(sSessionId, aAddr, aModuleInfo);
  if (retval == FALSE) {
    /*
     * Not loaded, here's the magic.
     * Go through all the modules.
     */
    CallbackEspecial64UserContext context{
        .mSessionId = sSessionId,
        .mAddr = aAddr,
    };
    BOOL enumRes = ::EnumerateLoadedModules64(
        sSessionId, CallbackEspecial64, reinterpret_cast<PVOID>(&context));
    if (enumRes != FALSE) {
      /*
       * One final go.
       * If it fails, then well, we have other problems.
       */
      retval = ::SymGetModuleInfo64(sSessionId, aAddr, aModuleInfo);
    }
  }

  /*
   * If we got module info, we may attempt line info as well.
   * We will not report failure if this does not work.
   */
  if (retval != FALSE && aLineInfo) {
    DWORD displacement = 0;
    BOOL lineRes = FALSE;
    lineRes =
        ::SymGetLineFromAddr64(sSessionId, aAddr, &displacement, aLineInfo);
    if (!lineRes) {
      // Clear out aLineInfo to indicate that it's not valid
      memset(aLineInfo, 0, sizeof(*aLineInfo));
    }
  }

  return retval;
}

MFBT_API bool MozDescribeCodeAddress(void* aPC,
                                     MozCodeAddressDetails* aDetails) {
  aDetails->library[0] = '\0';
  aDetails->loffset = 0;
  aDetails->filename[0] = '\0';
  aDetails->lineno = 0;
  aDetails->function[0] = '\0';
  aDetails->foffset = 0;

  DbgHelpWrapperSym dbgHelp;
  if (!dbgHelp.ReadyToUse()) {
    return false;
  }

  // Attempt to load module info before we attempt to resolve the symbol.
  // This just makes sure we get good info if available.
  DWORD64 addr = (DWORD64)aPC;
  IMAGEHLP_MODULE64 modInfo;
  IMAGEHLP_LINE64 lineInfo;
  BOOL modInfoRes;
  modInfoRes = dbgHelp.SymGetModuleInfoEspecial64(addr, &modInfo, &lineInfo);

  if (modInfoRes) {
    strncpy(aDetails->library, modInfo.LoadedImageName,
            sizeof(aDetails->library));
    aDetails->library[std::size(aDetails->library) - 1] = '\0';
    aDetails->loffset = (char*)aPC - (char*)modInfo.BaseOfImage;

    if (lineInfo.FileName) {
      strncpy(aDetails->filename, lineInfo.FileName,
              sizeof(aDetails->filename));
      aDetails->filename[std::size(aDetails->filename) - 1] = '\0';
      aDetails->lineno = lineInfo.LineNumber;
    }
  }

  ULONG64 buffer[(sizeof(SYMBOL_INFO) + MAX_SYM_NAME * sizeof(TCHAR) +
                  sizeof(ULONG64) - 1) /
                 sizeof(ULONG64)];
  PSYMBOL_INFO pSymbol = (PSYMBOL_INFO)buffer;
  pSymbol->SizeOfStruct = sizeof(SYMBOL_INFO);
  pSymbol->MaxNameLen = MAX_SYM_NAME;

  DWORD64 displacement;
  BOOL ok = dbgHelp.SymFromAddr(addr, &displacement, pSymbol);

  if (ok) {
    strncpy(aDetails->function, pSymbol->Name, sizeof(aDetails->function));
    aDetails->function[std::size(aDetails->function) - 1] = '\0';
    aDetails->foffset = static_cast<ptrdiff_t>(displacement);
  }

  return true;
}

// i386 or PPC Linux stackwalking code
//
// Changes to to OS/Architecture support here should be reflected in
// build/moz.configure/memory.configure
#elif HAVE_DLADDR &&                                           \
    (HAVE__UNWIND_BACKTRACE || MOZ_STACKWALK_SUPPORTS_LINUX || \
     MOZ_STACKWALK_SUPPORTS_MACOSX)

#  include <stdlib.h>
#  include <stdio.h>

// On glibc 2.1, the Dl_info api defined in <dlfcn.h> is only exposed
// if __USE_GNU is defined.  I suppose its some kind of standards
// adherence thing.
//
#  if (__GLIBC_MINOR__ >= 1) && !defined(__USE_GNU)
#    define __USE_GNU
#  endif

// This thing is exported by libstdc++
// Yes, this is a gcc only hack
#  if defined(MOZ_DEMANGLE_SYMBOLS)
#    include <cxxabi.h>
#  endif  // MOZ_DEMANGLE_SYMBOLS

namespace mozilla {

void DemangleSymbol(const char* aSymbol, char* aBuffer, int aBufLen) {
  aBuffer[0] = '\0';

#  if defined(MOZ_DEMANGLE_SYMBOLS)
  /* See demangle.h in the gcc source for the voodoo */
  char* demangled = abi::__cxa_demangle(aSymbol, 0, 0, 0);

  if (demangled) {
    strncpy(aBuffer, demangled, aBufLen);
    aBuffer[aBufLen - 1] = '\0';
    free(demangled);
  }
#  endif  // MOZ_DEMANGLE_SYMBOLS
}

}  // namespace mozilla

// {x86, ppc} x {Linux, Mac} stackwalking code.
//
// Changes to to OS/Architecture support here should be reflected in
// build/moz.configure/memory.configure
#  if ((defined(__i386) || defined(PPC) || defined(__ppc__)) && \
       (MOZ_STACKWALK_SUPPORTS_MACOSX || MOZ_STACKWALK_SUPPORTS_LINUX))

static void DoFramePointerStackWalk(MozWalkStackCallback aCallback,
                                    const void* aFirstFramePC,
                                    uint32_t aMaxFrames, void* aClosure,
                                    void** aBp, void* aStackEnd);

MFBT_API void MozStackWalk(MozWalkStackCallback aCallback,
                           const void* aFirstFramePC, uint32_t aMaxFrames,
                           void* aClosure) {
  // Get the frame pointer
  void** bp = (void**)__builtin_frame_address(0);

  void* stackEnd;
#    if HAVE___LIBC_STACK_END
  stackEnd = __libc_stack_end;
#    elif defined(XP_DARWIN)
  stackEnd = pthread_get_stackaddr_np(pthread_self());
#    elif defined(ANDROID)
  pthread_attr_t sattr;
  pthread_attr_init(&sattr);
  pthread_getattr_np(pthread_self(), &sattr);
  void* stackBase = stackEnd = nullptr;
  size_t stackSize = 0;
  if (gettid() != getpid()) {
    // bionic's pthread_attr_getstack doesn't tell the truth for the main
    // thread (see bug 846670). So don't use it for the main thread.
    if (!pthread_attr_getstack(&sattr, &stackBase, &stackSize)) {
      stackEnd = static_cast<char*>(stackBase) + stackSize;
    } else {
      stackEnd = nullptr;
    }
  }
  if (!stackEnd) {
    // So consider the current frame pointer + an arbitrary size of 8MB
    // (modulo overflow ; not really arbitrary as it's the default stack
    // size for the main thread) if pthread_attr_getstack failed for
    // some reason (or was skipped).
    static const uintptr_t kMaxStackSize = 8 * 1024 * 1024;
    uintptr_t maxStackStart = uintptr_t(-1) - kMaxStackSize;
    uintptr_t stackStart = std::max(maxStackStart, uintptr_t(bp));
    stackEnd = reinterpret_cast<void*>(stackStart + kMaxStackSize);
  }
#    else
#      error Unsupported configuration
#    endif
  DoFramePointerStackWalk(aCallback, aFirstFramePC, aMaxFrames, aClosure, bp,
                          stackEnd);
}

#  elif defined(HAVE__UNWIND_BACKTRACE)

// libgcc_s.so symbols _Unwind_Backtrace@@GCC_3.3 and _Unwind_GetIP@@GCC_3.0
#    include <unwind.h>

struct unwind_info {
  MozWalkStackCallback callback;
  FrameSkipper skipper;
  int maxFrames;
  int numFrames;
  void* closure;
};

static _Unwind_Reason_Code unwind_callback(struct _Unwind_Context* context,
                                           void* closure) {
  _Unwind_Reason_Code ret = _URC_NO_REASON;
  unwind_info* info = static_cast<unwind_info*>(closure);
  void* pc = reinterpret_cast<void*>(_Unwind_GetIP(context));
#    if HAVE___LIBC_STACK_END && defined(__aarch64__)
  // Work around https://sourceware.org/bugzilla/show_bug.cgi?id=32612
  // The _dl_tlsdesc_dynamic function can't be unwound through with
  // _Unwind_Backtrace when glibc is built with aarch64 PAC (that leads
  // to a crash).
  // Unfortunately, we can't get the address of that specific function, so
  // we just disallow all of ld-linux-aarch64.so.1: when we hit an address
  // in there, we make _Unwind_Backtrace stop.
  // In the case of _dl_tlsdesc_dynamic, this would stop the stackwalk at
  // tls_get_addr_tail, which is enough information to know the stack comes
  // from ld.so, and we even get inlining info giving us malloc,
  // allocate_dtv_entry and allocate_and_init, which is plenty enough and
  // better than nothing^Hcrashing.
  // To figure out whether the frame falls into ld-linux-aarch64.so.1, we
  // use __libc_stack_end (which lives there and is .data) as upper bound
  // (assuming .data comes after .text), and get the base address of the
  // library via dladdr.
  if (!ldso_base) {
    Dl_info info;
    dladdr(&__libc_stack_end, &info);
    ldso_base = (uintptr_t)info.dli_fbase;
  }
  if (ldso_base && ((uintptr_t)pc > ldso_base) &&
      (uintptr_t)pc < (uintptr_t)&__libc_stack_end) {
    // Any error code will do, we just want to stop the walk even when
    // we haven't reached the limit.
    ret = _URC_FOREIGN_EXCEPTION_CAUGHT;
  }
#    endif
  // TODO Use something like '_Unwind_GetGR()' to get the stack pointer.
  if (!info->skipper.ShouldSkipPC(pc)) {
    info->numFrames++;
    (*info->callback)(info->numFrames, pc, nullptr, info->closure);
    if (info->maxFrames != 0 && info->numFrames == info->maxFrames) {
      // Again, any error code that stops the walk will do.
      return _URC_FOREIGN_EXCEPTION_CAUGHT;
    }
  }
  return ret;
}

MFBT_API void MozStackWalk(MozWalkStackCallback aCallback,
                           const void* aFirstFramePC, uint32_t aMaxFrames,
                           void* aClosure) {
  unwind_info info;
  info.callback = aCallback;
  info.skipper = FrameSkipper(aFirstFramePC ? aFirstFramePC : CallerPC());
  info.maxFrames = aMaxFrames;
  info.numFrames = 0;
  info.closure = aClosure;

  // We ignore the return value from _Unwind_Backtrace. There are three main
  // reasons for this.
  // - On ARM/Android bionic's _Unwind_Backtrace usually (always?) returns
  //   _URC_FAILURE.  See
  //   https://bugzilla.mozilla.org/show_bug.cgi?id=717853#c110.
  // - If aMaxFrames != 0, we want to stop early, and the only way to do that
  //   is to make unwind_callback return something other than _URC_NO_REASON,
  //   which causes _Unwind_Backtrace to return a non-success code.
  // - MozStackWalk doesn't have a return value anyway.
  (void)_Unwind_Backtrace(unwind_callback, &info);
}

#  endif

bool MFBT_API MozDescribeCodeAddress(void* aPC,
                                     MozCodeAddressDetails* aDetails) {
  aDetails->library[0] = '\0';
  aDetails->loffset = 0;
  aDetails->filename[0] = '\0';
  aDetails->lineno = 0;
  aDetails->function[0] = '\0';
  aDetails->foffset = 0;

  Dl_info info;

#  if defined(ANDROID) && defined(MOZ_LINKER)
  int ok = __wrap_dladdr(aPC, &info);
#  else
  int ok = dladdr(aPC, &info);
#  endif

  if (!ok) {
    return true;
  }

  strncpy(aDetails->library, info.dli_fname, sizeof(aDetails->library));
  aDetails->library[std::size(aDetails->library) - 1] = '\0';
  aDetails->loffset = (char*)aPC - (char*)info.dli_fbase;

#  if !defined(XP_FREEBSD)
  // On FreeBSD, dli_sname is unusably bad, it often returns things like
  // 'gtk_xtbin_new' or 'XRE_GetBootstrap' instead of long C++ symbols. Just let
  // GetFunction do the lookup directly in the ELF image.

  const char* symbol = info.dli_sname;
  if (!symbol || symbol[0] == '\0') {
    return true;
  }

  DemangleSymbol(symbol, aDetails->function, sizeof(aDetails->function));

  if (aDetails->function[0] == '\0') {
    // Just use the mangled symbol if demangling failed.
    strncpy(aDetails->function, symbol, sizeof(aDetails->function));
    aDetails->function[std::size(aDetails->function) - 1] = '\0';
  }

  aDetails->foffset = (char*)aPC - (char*)info.dli_saddr;
#  endif

  return true;
}

#else  // unsupported platform.

MFBT_API void MozStackWalk(MozWalkStackCallback aCallback,
                           const void* aFirstFramePC, uint32_t aMaxFrames,
                           void* aClosure) {}

MFBT_API bool MozDescribeCodeAddress(void* aPC,
                                     MozCodeAddressDetails* aDetails) {
  aDetails->library[0] = '\0';
  aDetails->loffset = 0;
  aDetails->filename[0] = '\0';
  aDetails->lineno = 0;
  aDetails->function[0] = '\0';
  aDetails->foffset = 0;
  return false;
}

#endif

#if defined(XP_WIN) || defined(XP_MACOSX) || defined(XP_LINUX)

#  if defined(XP_MACOSX) && defined(__aarch64__)
// On macOS arm64, system libraries are arm64e binaries, and arm64e can do
// pointer authentication: The low bits of the pointer are the actual pointer
// value, and the high bits are an encrypted hash. During stackwalking, we need
// to strip off this hash. In theory, ptrauth_strip would be the right function
// to call for this. However, that function is a no-op unless it's called from
// code which also builds as arm64e - which we do not. So we cannot use it. So
// for now, we hardcode a mask that seems to work today: 40 bits for the pointer
// and 24 bits for the hash seems to do the trick. We can worry about
// dynamically computing the correct mask if this ever stops working.
const uintptr_t kPointerMask =
    (uintptr_t(1) << 40) - 1;  // 40 bits pointer, 24 bit PAC
#  else
const uintptr_t kPointerMask = ~uintptr_t(0);
#  endif

MOZ_ASAN_IGNORE
static void DoFramePointerStackWalk(MozWalkStackCallback aCallback,
                                    const void* aFirstFramePC,
                                    uint32_t aMaxFrames, void* aClosure,
                                    void** aBp, void* aStackEnd) {
  // Stack walking code courtesy Kipp's "leaky".

  FrameSkipper skipper(aFirstFramePC);
  uint32_t numFrames = 0;

  // Sanitize the given aBp. Assume that something reasonably close to
  // but before the stack end is going be a valid frame pointer. Also
  // check that it is an aligned address. This increases the chances
  // that if the pointer is not valid (which might happen if the caller
  // called __builtin_frame_address(1) and its frame is busted for some
  // reason), we won't read it, leading to a crash. Because the calling
  // code is not using frame pointers when returning, it might actually
  // recover just fine.
  static const uintptr_t kMaxStackSize = 8 * 1024 * 1024;
  if (uintptr_t(aBp) < uintptr_t(aStackEnd) -
                           std::min(kMaxStackSize, uintptr_t(aStackEnd)) ||
      aBp >= aStackEnd || (uintptr_t(aBp) & 3)) {
    return;
  }

  while (aBp) {
    void** next = (void**)*aBp;
    // aBp may not be a frame pointer on i386 if code was compiled with
    // -fomit-frame-pointer, so do some sanity checks.
    // (aBp should be a frame pointer on ppc(64) but checking anyway may help
    // a little if the stack has been corrupted.)
    // We don't need to check against the begining of the stack because
    // we can assume that aBp > sp
    if (next <= aBp || next >= aStackEnd || (uintptr_t(next) & 3)) {
      break;
    }
#  if (defined(__ppc__) && defined(XP_MACOSX)) || defined(__powerpc64__)
    // ppc mac or powerpc64 linux
    void* pc = *(aBp + 2);
    aBp += 3;
#  else  // i386 or powerpc32 linux
    void* pc = *(aBp + 1);
    aBp += 2;
#  endif

    // Strip off pointer authentication hash, if present. For now, it looks
    // like only return addresses require stripping, and stack pointers do
    // not. This might change in the future.
    pc = (void*)((uintptr_t)pc & kPointerMask);

    if (!skipper.ShouldSkipPC(pc)) {
      // Assume that the SP points to the BP of the function
      // it called. We can't know the exact location of the SP
      // but this should be sufficient for our use the SP
      // to order elements on the stack.
      numFrames++;
      (*aCallback)(numFrames, pc, aBp, aClosure);
      if (aMaxFrames != 0 && numFrames == aMaxFrames) {
        break;
      }
    }
    aBp = next;
  }
}

namespace mozilla {

MFBT_API void FramePointerStackWalk(MozWalkStackCallback aCallback,
                                    uint32_t aMaxFrames, void* aClosure,
                                    void** aBp, void* aStackEnd) {
  // We don't pass a aFirstFramePC because we start walking the stack from the
  // frame at aBp.
  DoFramePointerStackWalk(aCallback, nullptr, aMaxFrames, aClosure, aBp,
                          aStackEnd);
}

}  // namespace mozilla

#else

namespace mozilla {
MFBT_API void FramePointerStackWalk(MozWalkStackCallback aCallback,
                                    uint32_t aMaxFrames, void* aClosure,
                                    void** aBp, void* aStackEnd) {}
}  // namespace mozilla

#endif

MFBT_API int MozFormatCodeAddressDetails(
    char* aBuffer, uint32_t aBufferSize, uint32_t aFrameNumber, void* aPC,
    const MozCodeAddressDetails* aDetails) {
  return MozFormatCodeAddress(aBuffer, aBufferSize, aFrameNumber, aPC,
                              aDetails->function, aDetails->library,
                              aDetails->loffset, aDetails->filename,
                              aDetails->lineno);
}

MFBT_API int MozFormatCodeAddress(char* aBuffer, uint32_t aBufferSize,
                                  uint32_t aFrameNumber, const void* aPC,
                                  const char* aFunction, const char* aLibrary,
                                  ptrdiff_t aLOffset, const char* aFileName,
                                  uint32_t aLineNo) {
  const char* function = aFunction && aFunction[0] ? aFunction : "???";
  if (aFileName && aFileName[0]) {
    // We have a filename and (presumably) a line number. Use them.
    return SprintfBuf(aBuffer, aBufferSize, "#%02u: %s (%s:%u)", aFrameNumber,
                      function, aFileName, aLineNo);
  } else if (aLibrary && aLibrary[0]) {
    // We have no filename, but we do have a library name. Use it and the
    // library offset, and print them in a way that `fix_stacks.py` can
    // post-process.
    return SprintfBuf(aBuffer, aBufferSize, "#%02u: %s[%s +0x%" PRIxPTR "]",
                      aFrameNumber, function, aLibrary,
                      static_cast<uintptr_t>(aLOffset));
  } else {
    // We have nothing useful to go on. (The format string is split because
    // '??)' is a trigraph and causes a warning, sigh.)
    return SprintfBuf(aBuffer, aBufferSize,
                      "#%02u: ??? (???:???"
                      ")",
                      aFrameNumber);
  }
}

static void EnsureWrite(FILE* aStream, const char* aBuf, size_t aLen) {
#ifdef XP_WIN
  int fd = _fileno(aStream);
#else
  int fd = fileno(aStream);
#endif
  while (aLen > 0) {
#ifdef XP_WIN
    auto written = _write(fd, aBuf, aLen);
#else
    auto written = write(fd, aBuf, aLen);
#endif
    if (written <= 0 || size_t(written) > aLen) {
      break;
    }
    aBuf += written;
    aLen -= written;
  }
}

template <int N>
static int PrintStackFrameBuf(char (&aBuf)[N], uint32_t aFrameNumber, void* aPC,
                              void* aSP) {
  MozCodeAddressDetails details;
  MozDescribeCodeAddress(aPC, &details);
  int len =
      MozFormatCodeAddressDetails(aBuf, N - 1, aFrameNumber, aPC, &details);
  len = std::min(len, N - 2);
  aBuf[len++] = '\n';
  aBuf[len] = '\0';
  return len;
}

static void PrintStackFrame(uint32_t aFrameNumber, void* aPC, void* aSP,
                            void* aClosure) {
  FILE* stream = (FILE*)aClosure;
  char buf[1025];  // 1024 + 1 for trailing '\n'
  int len = PrintStackFrameBuf(buf, aFrameNumber, aPC, aSP);
  fflush(stream);
  EnsureWrite(stream, buf, len);
}

static bool WalkTheStackEnabled() {
  static bool result = [] {
    char* value = getenv("MOZ_DISABLE_WALKTHESTACK");
    return !(value && value[0]);
  }();
  return result;
}

MFBT_API void MozWalkTheStack(FILE* aStream, const void* aFirstFramePC,
                              uint32_t aMaxFrames) {
  if (WalkTheStackEnabled()) {
    MozStackWalk(PrintStackFrame, aFirstFramePC ? aFirstFramePC : CallerPC(),
                 aMaxFrames, aStream);
  }
}

static void WriteStackFrame(uint32_t aFrameNumber, void* aPC, void* aSP,
                            void* aClosure) {
  auto writer = (void (*)(const char*))aClosure;
  char buf[1024];
  PrintStackFrameBuf(buf, aFrameNumber, aPC, aSP);
  writer(buf);
}

MFBT_API void MozWalkTheStackWithWriter(void (*aWriter)(const char*),
                                        const void* aFirstFramePC,
                                        uint32_t aMaxFrames) {
  if (WalkTheStackEnabled()) {
    MozStackWalk(WriteStackFrame, aFirstFramePC ? aFirstFramePC : CallerPC(),
                 aMaxFrames, (void*)aWriter);
  }
}

quality99%

¤ Dauer der Verarbeitung: 0.24 Sekunden (vorverarbeitet) ¤

Wurzel

Suchen

Beweissystem der NASA

Beweissystem Isabelle

NIST Cobol Testsuite

Cephes Mathematical Library

Wiener Entwicklungsmethode

Haftungshinweis

Die Informationen auf dieser Webseite wurden nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit, noch Qualität der bereit gestellten Informationen zugesichert.

Bemerkung:

Die farbliche Syntaxdarstellung ist noch experimentell.