vendor/compiler-rt/compiler-rt-trunk-r338536

17 files changed, 209 insertions, 70 deletions

diff --git a/lib/asan/asan_mapping.h b/lib/asan/asan_mapping.h
index 3f8cc004b95f5..f3696da62b30a 100644
--- a/lib/asan/asan_mapping.h
+++ b/lib/asan/asan_mapping.h
@@ -122,7 +122,7 @@
 // || `[0x400000000000, 0x47ffffffffff]` || LowShadow  ||
 // || `[0x000000000000, 0x3fffffffffff]` || LowMem     ||
 //
-// Shadow mapping on NerBSD/i386 with SHADOW_OFFSET == 0x40000000:
+// Shadow mapping on NetBSD/i386 with SHADOW_OFFSET == 0x40000000:
 // || `[0x60000000, 0xfffff000]` || HighMem    ||
 // || `[0x4c000000, 0x5fffffff]` || HighShadow ||
 // || `[0x48000000, 0x4bffffff]` || ShadowGap  ||
diff --git a/lib/asan/tests/asan_test.cc b/lib/asan/tests/asan_test.cc
index 11a3506a4853e..25d6a20834503 100644
--- a/lib/asan/tests/asan_test.cc
+++ b/lib/asan/tests/asan_test.cc
@@ -963,7 +963,7 @@ TEST(AddressSanitizer, ThreadNamesTest) {
 #if ASAN_NEEDS_SEGV
 TEST(AddressSanitizer, ShadowGapTest) {
 #if SANITIZER_WORDSIZE == 32
-  char *addr = (char*)0x22000000;
+  char *addr = (char*)0x23000000;
 #else
 # if defined(__powerpc64__)
   char *addr = (char*)0x024000800000;
diff --git a/lib/builtins/CMakeLists.txt b/lib/builtins/CMakeLists.txt
index 75ff66405649c..82332967b1047 100644
--- a/lib/builtins/CMakeLists.txt
+++ b/lib/builtins/CMakeLists.txt
@@ -567,6 +567,15 @@ else ()
         set(_arch "arm")
       endif()
 
+      # For ARM archs, exclude any VFP builtins if VFP is not supported
+      if (${arch} MATCHES "^(arm|armhf|armv7|armv7s|armv7k|armv7m|armv7em)$")
+        string(REPLACE ";" " " _TARGET_${arch}_CFLAGS "${TARGET_${arch}_CFLAGS}")
+        check_compile_definition(__VFP_FP__ "${CMAKE_C_FLAGS} ${_TARGET_${arch}_CFLAGS}" COMPILER_RT_HAS_${arch}_VFP)
+        if(NOT COMPILER_RT_HAS_${arch}_VFP)
+          list(REMOVE_ITEM ${arch}_SOURCES ${arm_Thumb1_VFPv2_SOURCES} ${arm_Thumb1_SjLj_EH_SOURCES})
+        endif()
+      endif()
+
       # Filter out generic versions of routines that are re-implemented in
       # architecture specific manner.  This prevents multiple definitions of the
       # same symbols, making the symbol selection non-deterministic.
diff --git a/lib/profile/InstrProfilingUtil.c b/lib/profile/InstrProfilingUtil.c
index d053ab160ca7b..083bf14a32604 100644
--- a/lib/profile/InstrProfilingUtil.c
+++ b/lib/profile/InstrProfilingUtil.c
@@ -35,6 +35,8 @@
 #include "InstrProfiling.h"
 #include "InstrProfilingUtil.h"
 
+COMPILER_RT_WEAK unsigned lprofDirMode = 0755;
+
 COMPILER_RT_VISIBILITY
 void __llvm_profile_recursive_mkdir(char *path) {
   int i;
@@ -47,12 +49,19 @@ void __llvm_profile_recursive_mkdir(char *path) {
 #ifdef _WIN32
     _mkdir(path);
 #else
-    mkdir(path, 0755); /* Some of these will fail, ignore it. */
+    /* Some of these will fail, ignore it. */
+    mkdir(path, __llvm_profile_get_dir_mode());
 #endif
     path[i] = save;
   }
 }
 
+COMPILER_RT_VISIBILITY
+void __llvm_profile_set_dir_mode(unsigned Mode) { lprofDirMode = Mode; }
+
+COMPILER_RT_VISIBILITY
+unsigned __llvm_profile_get_dir_mode(void) { return lprofDirMode; }
+
 #if COMPILER_RT_HAS_ATOMICS != 1
 COMPILER_RT_VISIBILITY
 uint32_t lprofBoolCmpXchg(void **Ptr, void *OldV, void *NewV) {
diff --git a/lib/profile/InstrProfilingUtil.h b/lib/profile/InstrProfilingUtil.h
index 2f2ea1b876a82..147677fc87005 100644
--- a/lib/profile/InstrProfilingUtil.h
+++ b/lib/profile/InstrProfilingUtil.h
@@ -16,6 +16,12 @@
 /*! \brief Create a directory tree. */
 void __llvm_profile_recursive_mkdir(char *Pathname);
 
+/*! Set the mode used when creating profile directories. */
+void __llvm_profile_set_dir_mode(unsigned Mode);
+
+/*! Return the directory creation mode. */
+unsigned __llvm_profile_get_dir_mode(void);
+
 int lprofLockFd(int fd);
 int lprofUnlockFd(int fd);
 
diff --git a/lib/sanitizer_common/sanitizer_mutex.h b/lib/sanitizer_common/sanitizer_mutex.h
index 4bb878ee19dc1..a93705be69184 100644
--- a/lib/sanitizer_common/sanitizer_mutex.h
+++ b/lib/sanitizer_common/sanitizer_mutex.h
@@ -73,13 +73,8 @@ class SpinMutex : public StaticSpinMutex {
 
 class BlockingMutex {
  public:
-#if SANITIZER_WINDOWS
-  // Windows does not currently support LinkerInitialized
-  explicit BlockingMutex(LinkerInitialized);
-#else
   explicit constexpr BlockingMutex(LinkerInitialized)
-      : opaque_storage_ {0, }, owner_(0) {}
-#endif
+      : opaque_storage_ {0, }, owner_ {0} {}
   BlockingMutex();
   void Lock();
   void Unlock();
diff --git a/lib/sanitizer_common/sanitizer_win.cc b/lib/sanitizer_common/sanitizer_win.cc
index b8060b2e640cb..38e567d9a7328 100644
--- a/lib/sanitizer_common/sanitizer_win.cc
+++ b/lib/sanitizer_common/sanitizer_win.cc
@@ -767,43 +767,22 @@ void *internal_start_thread(void (*func)(void *arg), void *arg) { return 0; }
 void internal_join_thread(void *th) { }
 
 // ---------------------- BlockingMutex ---------------- {{{1
-const uptr LOCK_UNINITIALIZED = 0;
-const uptr LOCK_READY = (uptr)-1;
-
-BlockingMutex::BlockingMutex(LinkerInitialized li) {
-  // FIXME: see comments in BlockingMutex::Lock() for the details.
-  CHECK(li == LINKER_INITIALIZED || owner_ == LOCK_UNINITIALIZED);
-
-  CHECK(sizeof(CRITICAL_SECTION) <= sizeof(opaque_storage_));
-  InitializeCriticalSection((LPCRITICAL_SECTION)opaque_storage_);
-  owner_ = LOCK_READY;
-}
 
 BlockingMutex::BlockingMutex() {
-  CHECK(sizeof(CRITICAL_SECTION) <= sizeof(opaque_storage_));
-  InitializeCriticalSection((LPCRITICAL_SECTION)opaque_storage_);
-  owner_ = LOCK_READY;
+  CHECK(sizeof(SRWLOCK) <= sizeof(opaque_storage_));
+  internal_memset(this, 0, sizeof(*this));
 }
 
 void BlockingMutex::Lock() {
-  if (owner_ == LOCK_UNINITIALIZED) {
-    // FIXME: hm, global BlockingMutex objects are not initialized?!?
-    // This might be a side effect of the clang+cl+link Frankenbuild...
-    new(this) BlockingMutex((LinkerInitialized)(LINKER_INITIALIZED + 1));
-
-    // FIXME: If it turns out the linker doesn't invoke our
-    // constructors, we should probably manually Lock/Unlock all the global
-    // locks while we're starting in one thread to avoid double-init races.
-  }
-  EnterCriticalSection((LPCRITICAL_SECTION)opaque_storage_);
-  CHECK_EQ(owner_, LOCK_READY);
+  AcquireSRWLockExclusive((PSRWLOCK)opaque_storage_);
+  CHECK_EQ(owner_, 0);
   owner_ = GetThreadSelf();
 }
 
 void BlockingMutex::Unlock() {
-  CHECK_EQ(owner_, GetThreadSelf());
-  owner_ = LOCK_READY;
-  LeaveCriticalSection((LPCRITICAL_SECTION)opaque_storage_);
+  CheckLocked();
+  owner_ = 0;
+  ReleaseSRWLockExclusive((PSRWLOCK)opaque_storage_);
 }
 
 void BlockingMutex::CheckLocked() {
diff --git a/lib/ubsan/ubsan_checks.inc b/lib/ubsan/ubsan_checks.inc
index 73c69363215af..5a7bdec2df12a 100644
--- a/lib/ubsan/ubsan_checks.inc
+++ b/lib/ubsan/ubsan_checks.inc
@@ -30,6 +30,8 @@ UBSAN_CHECK(IntegerDivideByZero, "integer-divide-by-zero",
             "integer-divide-by-zero")
 UBSAN_CHECK(FloatDivideByZero, "float-divide-by-zero", "float-divide-by-zero")
 UBSAN_CHECK(InvalidBuiltin, "invalid-builtin-use", "invalid-builtin-use")
+UBSAN_CHECK(ImplicitIntegerTruncation, "implicit-integer-truncation",
+            "implicit-integer-truncation")
 UBSAN_CHECK(InvalidShiftBase, "invalid-shift-base", "shift-base")
 UBSAN_CHECK(InvalidShiftExponent, "invalid-shift-exponent", "shift-exponent")
 UBSAN_CHECK(OutOfBoundsIndex, "out-of-bounds-index", "bounds")
diff --git a/lib/ubsan/ubsan_diag.h b/lib/ubsan/ubsan_diag.h
index d9ea410c358ca..bde749684c659 100644
--- a/lib/ubsan/ubsan_diag.h
+++ b/lib/ubsan/ubsan_diag.h
@@ -178,7 +178,7 @@ public:
   };
 
 private:
-  static const unsigned MaxArgs = 5;
+  static const unsigned MaxArgs = 8;
   static const unsigned MaxRanges = 1;
 
   /// The arguments which have been added to this diagnostic so far.
diff --git a/lib/ubsan/ubsan_handlers.cc b/lib/ubsan/ubsan_handlers.cc
index 49c6bcacbe277..e72a32cf32c29 100644
--- a/lib/ubsan/ubsan_handlers.cc
+++ b/lib/ubsan/ubsan_handlers.cc
@@ -451,6 +451,49 @@ void __ubsan::__ubsan_handle_load_invalid_value_abort(InvalidValueData *Data,
   Die();
 }
 
+static void handleImplicitConversion(ImplicitConversionData *Data,
+                                     ReportOptions Opts, ValueHandle Src,
+                                     ValueHandle Dst) {
+  SourceLocation Loc = Data->Loc.acquire();
+  ErrorType ET = ErrorType::GenericUB;
+
+  switch (Data->Kind) {
+  case ICCK_IntegerTruncation:
+    ET = ErrorType::ImplicitIntegerTruncation;
+    break;
+  }
+
+  if (ignoreReport(Loc, Opts, ET))
+    return;
+
+  const TypeDescriptor &SrcTy = Data->FromType;
+  const TypeDescriptor &DstTy = Data->ToType;
+
+  ScopedReport R(Opts, Loc, ET);
+
+  // FIXME: is it possible to dump the values as hex with fixed width?
+
+  Diag(Loc, DL_Error, ET,
+       "implicit conversion from type %0 of value %1 (%2-bit, %3signed) to "
+       "type %4 changed the value to %5 (%6-bit, %7signed)")
+      << SrcTy << Value(SrcTy, Src) << SrcTy.getIntegerBitWidth()
+      << (SrcTy.isSignedIntegerTy() ? "" : "un") << DstTy << Value(DstTy, Dst)
+      << DstTy.getIntegerBitWidth() << (DstTy.isSignedIntegerTy() ? "" : "un");
+}
+
+void __ubsan::__ubsan_handle_implicit_conversion(ImplicitConversionData *Data,
+                                                 ValueHandle Src,
+                                                 ValueHandle Dst) {
+  GET_REPORT_OPTIONS(false);
+  handleImplicitConversion(Data, Opts, Src, Dst);
+}
+void __ubsan::__ubsan_handle_implicit_conversion_abort(
+    ImplicitConversionData *Data, ValueHandle Src, ValueHandle Dst) {
+  GET_REPORT_OPTIONS(true);
+  handleImplicitConversion(Data, Opts, Src, Dst);
+  Die();
+}
+
 static void handleInvalidBuiltin(InvalidBuiltinData *Data, ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
   ErrorType ET = ErrorType::InvalidBuiltin;
diff --git a/lib/ubsan/ubsan_handlers.h b/lib/ubsan/ubsan_handlers.h
index ed3c8f0b1b0ef..07644b7ea15d6 100644
--- a/lib/ubsan/ubsan_handlers.h
+++ b/lib/ubsan/ubsan_handlers.h
@@ -122,6 +122,23 @@ struct InvalidValueData {
 /// \brief Handle a load of an invalid value for the type.
 RECOVERABLE(load_invalid_value, InvalidValueData *Data, ValueHandle Val)
 
+/// Known implicit conversion check kinds.
+/// Keep in sync with the enum of the same name in CGExprScalar.cpp
+enum ImplicitConversionCheckKind : unsigned char {
+  ICCK_IntegerTruncation = 0,
+};
+
+struct ImplicitConversionData {
+  SourceLocation Loc;
+  const TypeDescriptor &FromType;
+  const TypeDescriptor &ToType;
+  /* ImplicitConversionCheckKind */ unsigned char Kind;
+};
+
+/// \brief Implict conversion that changed the value.
+RECOVERABLE(implicit_conversion, ImplicitConversionData *Data, ValueHandle Src,
+            ValueHandle Dst)
+
 /// Known builtin check kinds.
 /// Keep in sync with the enum of the same name in CodeGenFunction.h
 enum BuiltinCheckKind : unsigned char {
diff --git a/lib/ubsan/ubsan_interface.inc b/lib/ubsan/ubsan_interface.inc
index 782c621a2b47c..0be6010ad27bb 100644
--- a/lib/ubsan/ubsan_interface.inc
+++ b/lib/ubsan/ubsan_interface.inc
@@ -22,6 +22,8 @@ INTERFACE_FUNCTION(__ubsan_handle_float_cast_overflow)
 INTERFACE_FUNCTION(__ubsan_handle_float_cast_overflow_abort)
 INTERFACE_FUNCTION(__ubsan_handle_function_type_mismatch)
 INTERFACE_FUNCTION(__ubsan_handle_function_type_mismatch_abort)
+INTERFACE_FUNCTION(__ubsan_handle_implicit_conversion)
+INTERFACE_FUNCTION(__ubsan_handle_implicit_conversion_abort)
 INTERFACE_FUNCTION(__ubsan_handle_invalid_builtin)
 INTERFACE_FUNCTION(__ubsan_handle_invalid_builtin_abort)
 INTERFACE_FUNCTION(__ubsan_handle_load_invalid_value)
diff --git a/lib/ubsan_minimal/ubsan_minimal_handlers.cc b/lib/ubsan_minimal/ubsan_minimal_handlers.cc
index f25a75a863c4e..e8fc3a8499b52 100644
--- a/lib/ubsan_minimal/ubsan_minimal_handlers.cc
+++ b/lib/ubsan_minimal/ubsan_minimal_handlers.cc
@@ -109,6 +109,7 @@ HANDLER(float_cast_overflow, "float-cast-overflow")
 HANDLER(load_invalid_value, "load-invalid-value")
 HANDLER(invalid_builtin, "invalid-builtin")
 HANDLER(function_type_mismatch, "function-type-mismatch")
+HANDLER(implicit_conversion, "implicit-conversion")
 HANDLER(nonnull_arg, "nonnull-arg")
 HANDLER(nonnull_return, "nonnull-return")
 HANDLER(nullability_arg, "nullability-arg")
diff --git a/lib/xray/tests/unit/profile_collector_test.cc b/lib/xray/tests/unit/profile_collector_test.cc
index b7dbe567312a0..67049af2cd5f9 100644
--- a/lib/xray/tests/unit/profile_collector_test.cc
+++ b/lib/xray/tests/unit/profile_collector_test.cc
@@ -15,6 +15,8 @@
 #include "xray_profile_collector.h"
 #include "xray_profiling_flags.h"
 #include <cstdint>
+#include <cstring>
+#include <memory>
 #include <thread>
 #include <utility>
 #include <vector>
@@ -24,6 +26,29 @@ namespace {
 
 static constexpr auto kHeaderSize = 16u;
 
+constexpr uptr ExpectedProfilingVersion = 0x20180424;
+
+struct ExpectedProfilingFileHeader {
+  const u64 MagicBytes = 0x7872617970726f66; // Identifier for XRay profiling
+                                             // files 'xrayprof' in hex.
+  const u64 Version = ExpectedProfilingVersion;
+  u64 Timestamp = 0;
+  u64 PID = 0;
+};
+
+void ValidateFileHeaderBlock(XRayBuffer B) {
+  ASSERT_NE(static_cast<const void *>(B.Data), nullptr);
+  ASSERT_EQ(B.Size, sizeof(ExpectedProfilingFileHeader));
+  typename std::aligned_storage<sizeof(ExpectedProfilingFileHeader)>::type
+      FileHeaderStorage;
+  ExpectedProfilingFileHeader ExpectedHeader;
+  std::memcpy(&FileHeaderStorage, B.Data, B.Size);
+  auto &FileHeader =
+      *reinterpret_cast<ExpectedProfilingFileHeader *>(&FileHeaderStorage);
+  ASSERT_EQ(ExpectedHeader.MagicBytes, FileHeader.MagicBytes);
+  ASSERT_EQ(ExpectedHeader.Version, FileHeader.Version);
+}
+
 void ValidateBlock(XRayBuffer B) {
   profilingFlags()->setDefaults();
   ASSERT_NE(static_cast<const void *>(B.Data), nullptr);
@@ -107,9 +132,13 @@ TEST(profileCollectorServiceTest, PostSerializeCollect) {
   // Then we serialize the data.
   profileCollectorService::serialize();
 
-  // Then we go through a single buffer to see whether we're getting the data we
-  // expect.
+  // Then we go through two buffers to see whether we're getting the data we
+  // expect. The first block must always be as large as a file header, which
+  // will have a fixed size.
   auto B = profileCollectorService::nextBuffer({nullptr, 0});
+  ValidateFileHeaderBlock(B);
+
+  B = profileCollectorService::nextBuffer(B);
   ValidateBlock(B);
   u32 BlockSize;
   u32 BlockNum;
@@ -169,6 +198,9 @@ TEST(profileCollectorServiceTest, PostSerializeCollectMultipleThread) {
 
   // Ensure that we see two buffers.
   auto B = profileCollectorService::nextBuffer({nullptr, 0});
+  ValidateFileHeaderBlock(B);
+
+  B = profileCollectorService::nextBuffer(B);
   ValidateBlock(B);
 
   B = profileCollectorService::nextBuffer(B);
diff --git a/lib/xray/xray_buffer_queue.cc b/lib/xray/xray_buffer_queue.cc
index 8dfcc23540b12..3ce728900787b 100644
--- a/lib/xray/xray_buffer_queue.cc
+++ b/lib/xray/xray_buffer_queue.cc
@@ -13,17 +13,50 @@
 //
 //===----------------------------------------------------------------------===//
 #include "xray_buffer_queue.h"
-#include "sanitizer_common/sanitizer_allocator_internal.h"
 #include "sanitizer_common/sanitizer_common.h"
 #include "sanitizer_common/sanitizer_libc.h"
+#include "sanitizer_common/sanitizer_posix.h"
 #include <memory>
+#include <sys/mman.h>
+
+#ifndef MAP_NORESERVE
+// no-op on NetBSD (at least), unsupported flag on FreeBSD
+#define MAP_NORESERVE 0
+#endif
 
 using namespace __xray;
 using namespace __sanitizer;
 
+template <class T> static T *allocRaw(size_t N) {
+  // TODO: Report errors?
+  // We use MAP_NORESERVE on platforms where it's supported to ensure that the
+  // pages we're allocating for XRay never end up in pages that can be swapped
+  // in/out. We're doing this because for FDR mode, we want to ensure that
+  // writes to the buffers stay resident in memory to prevent XRay itself from
+  // causing swapping/thrashing.
+  //
+  // In the case when XRay pages cannot be swapped in/out or there's not enough
+  // RAM to back these pages, we're willing to cause a segmentation fault
+  // instead of introducing latency in the measurement. We assume here that
+  // there are enough pages that are swappable in/out outside of the buffers
+  // being used by FDR mode (which are bounded and configurable anyway) to allow
+  // us to keep using always-resident memory.
+  //
+  // TODO: Make this configurable?
+  void *A = reinterpret_cast<void *>(
+      internal_mmap(NULL, N * sizeof(T), PROT_WRITE | PROT_READ,
+                    MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE, -1, 0));
+  return (A == MAP_FAILED) ? nullptr : reinterpret_cast<T *>(A);
+}
+
+template <class T> static void deallocRaw(T *ptr, size_t N) {
+  // TODO: Report errors?
+  if (ptr != nullptr)
+    internal_munmap(ptr, N);
+}
+
 template <class T> static T *initArray(size_t N) {
-  auto A = reinterpret_cast<T *>(
-      InternalAlloc(N * sizeof(T), nullptr, kCacheLineSize));
+  auto A = allocRaw<T>(N);
   if (A != nullptr)
     while (N > 0)
       new (A + (--N)) T();
@@ -42,19 +75,19 @@ BufferQueue::BufferQueue(size_t B, size_t N, bool &Success)
     // Clean up the buffers we've already allocated.
     for (auto B = Buffers, E = Buffers + BufferCount; B != E; ++B)
       B->~BufferRep();
-    InternalFree(Buffers);
+    deallocRaw(Buffers, N);
     Success = false;
     return;
   };
 
   for (size_t i = 0; i < N; ++i) {
     auto &T = Buffers[i];
-    void *Tmp = InternalAlloc(BufferSize, nullptr, 64);
+    void *Tmp = allocRaw<char>(BufferSize);
     if (Tmp == nullptr) {
       Success = false;
       return;
     }
-    void *Extents = InternalAlloc(sizeof(BufferExtents), nullptr, 64);
+    auto *Extents = allocRaw<BufferExtents>(1);
     if (Extents == nullptr) {
       Success = false;
       return;
@@ -62,7 +95,7 @@ BufferQueue::BufferQueue(size_t B, size_t N, bool &Success)
     auto &Buf = T.Buff;
     Buf.Data = Tmp;
     Buf.Size = B;
-    Buf.Extents = reinterpret_cast<BufferExtents *>(Extents);
+    Buf.Extents = Extents;
     OwnedBuffers[i] = Tmp;
   }
   Success = true;
@@ -128,11 +161,11 @@ BufferQueue::~BufferQueue() {
   for (auto I = Buffers, E = Buffers + BufferCount; I != E; ++I) {
     auto &T = *I;
     auto &Buf = T.Buff;
-    InternalFree(Buf.Data);
-    InternalFree(Buf.Extents);
+    deallocRaw(Buf.Data, Buf.Size);
+    deallocRaw(Buf.Extents, 1);
   }
   for (auto B = Buffers, E = Buffers + BufferCount; B != E; ++B)
     B->~BufferRep();
-  InternalFree(Buffers);
-  InternalFree(OwnedBuffers);
+  deallocRaw(Buffers, BufferCount);
+  deallocRaw(OwnedBuffers, BufferCount);
 }
diff --git a/lib/xray/xray_profile_collector.cc b/lib/xray/xray_profile_collector.cc
index a43744d9a0cb2..17a611eeacb86 100644
--- a/lib/xray/xray_profile_collector.cc
+++ b/lib/xray/xray_profile_collector.cc
@@ -37,6 +37,19 @@ struct ProfileBuffer {
   size_t Size;
 };
 
+// Current version of the profile format.
+constexpr u64 XRayProfilingVersion = 0x20180424;
+
+// Identifier for XRay profiling files 'xrayprof' in hex.
+constexpr u64 XRayMagicBytes = 0x7872617970726f66;
+
+struct XRayProfilingFileHeader {
+  const u64 MagicBytes = XRayMagicBytes;
+  const u64 Version = XRayProfilingVersion;
+  u64 Timestamp = 0; // System time in nanoseconds.
+  u64 PID = 0;       // Process ID.
+};
+
 struct BlockHeader {
   u32 BlockSize;
   u32 BlockNum;
@@ -302,7 +315,22 @@ XRayBuffer nextBuffer(XRayBuffer B) {
   if (ProfileBuffers == nullptr || ProfileBuffers->Size() == 0)
     return {nullptr, 0};
 
-  if (B.Data == nullptr)
+  static pthread_once_t Once = PTHREAD_ONCE_INIT;
+  static typename std::aligned_storage<sizeof(XRayProfilingFileHeader)>::type
+      FileHeaderStorage;
+  pthread_once(&Once,
+               +[] { new (&FileHeaderStorage) XRayProfilingFileHeader{}; });
+
+  if (UNLIKELY(B.Data == nullptr)) {
+    // The first buffer should always contain the file header information.
+    auto &FileHeader =
+        *reinterpret_cast<XRayProfilingFileHeader *>(&FileHeaderStorage);
+    FileHeader.Timestamp = NanoTime();
+    FileHeader.PID = internal_getpid();
+    return {&FileHeaderStorage, sizeof(XRayProfilingFileHeader)};
+  }
+
+  if (UNLIKELY(B.Data == &FileHeaderStorage))
     return {(*ProfileBuffers)[0].Data, (*ProfileBuffers)[0].Size};
 
   BlockHeader Header;
diff --git a/lib/xray/xray_profiling.cc b/lib/xray/xray_profiling.cc
index 786084c772268..d4b4345d764af 100644
--- a/lib/xray/xray_profiling.cc
+++ b/lib/xray/xray_profiling.cc
@@ -32,16 +32,6 @@ namespace __xray {
 
 namespace {
 
-constexpr uptr XRayProfilingVersion = 0x20180424;
-
-struct XRayProfilingFileHeader {
-  const u64 MagicBytes = 0x7872617970726f66; // Identifier for XRay profiling
-                                             // files 'xrayprof' in hex.
-  const uptr Version = XRayProfilingVersion;
-  uptr Timestamp = 0; // System time in nanoseconds.
-  uptr PID = 0;       // Process ID.
-};
-
 atomic_sint32_t ProfilerLogFlushStatus = {
     XRayLogFlushStatus::XRAY_LOG_NOT_FLUSHING};
 
@@ -144,14 +134,7 @@ XRayLogFlushStatus profilingFlush() XRAY_NEVER_INSTRUMENT {
         if (Verbosity())
           Report("profiling: Failed to flush to file, dropping data.\n");
       } else {
-        XRayProfilingFileHeader Header;
-        Header.Timestamp = NanoTime();
-        Header.PID = internal_getpid();
-        retryingWriteAll(Fd, reinterpret_cast<const char *>(&Header),
-                         reinterpret_cast<const char *>(&Header) +
-                             sizeof(Header));
-
-        // Now for each of the threads, write out the profile data as we would
+        // Now for each of the buffers, write out the profile data as we would
         // see it in memory, verbatim.
         while (B.Data != nullptr && B.Size != 0) {
           retryingWriteAll(Fd, reinterpret_cast<const char *>(B.Data),