Update the tcmalloc vendor branch to r111 (perftools version 1.8)

third_party/tcmalloc/vendor/src/tcmalloc.cc

Index: third_party/tcmalloc/vendor/src/tcmalloc.cc
===================================================================
--- third_party/tcmalloc/vendor/src/tcmalloc.cc	(revision 92996)
+++ third_party/tcmalloc/vendor/src/tcmalloc.cc	(working copy)
@@ -93,9 +93,6 @@
 #ifdef HAVE_SYS_CDEFS_H
 #include <sys/cdefs.h>                  // for __THROW
 #endif
-#ifdef HAVE_FEATURES_H
-#include <features.h>                   // for __GLIBC__
-#endif
 #if defined HAVE_STDINT_H
 #include <stdint.h>
 #elif defined HAVE_INTTYPES_H
@@ -153,6 +150,17 @@
 using STL_NAMESPACE::max;
 using STL_NAMESPACE::numeric_limits;
 using STL_NAMESPACE::vector;
+
+#include "libc_override.h"
+
+// __THROW is defined in glibc (via <sys/cdefs.h>).  It means,
+// counter-intuitively, "This function will never throw an exception."
+// It's an optional optimization tool, but we may need to use it to
+// match glibc prototypes.
+#ifndef __THROW    // I guess we're not on a glibc system
+# define __THROW   // __THROW is just an optimization, so ok to make it ""
+#endif
+
 using tcmalloc::AlignmentForSize;
 using tcmalloc::PageHeap;
 using tcmalloc::PageHeapAllocator;
@@ -162,13 +170,6 @@
 using tcmalloc::Static;
 using tcmalloc::ThreadCache;
 
-// __THROW is defined in glibc systems.  It means, counter-intuitively,
-// "This function will never throw an exception."  It's an optional
-// optimization tool, but we may need to use it to match glibc prototypes.
-#ifndef __THROW    // I guess we're not on a glibc system
-# define __THROW   // __THROW is just an optimization, so ok to make it ""
-#endif
-
 DECLARE_int64(tcmalloc_sample_parameter);
 DECLARE_double(tcmalloc_release_rate);
 
@@ -267,128 +268,6 @@
 }  // extern "C"
 #endif  // #ifndef _WIN32
 
-// Override the libc functions to prefer our own instead.  This comes
-// first so code in tcmalloc.cc can use the overridden versions.  One
-// exception: in windows, by default, we patch our code into these
-// functions (via src/windows/patch_function.cc) rather than override
-// them.  In that case, we don't want to do this overriding here.
-#if !defined(WIN32_DO_PATCHING)
-
-#if defined(__GNUC__) && !defined(__MACH__)
-  // Potentially faster variants that use the gcc alias extension.
-  // FreeBSD does support aliases, but apparently not correctly. :-(
-  // NOTE: we make many of these symbols weak, but do so in the makefile
-  //       (via objcopy -W) and not here.  That ends up being more portable.
-# define ALIAS(x) __attribute__ ((alias (x)))
-void* operator new(size_t size) throw (std::bad_alloc) ALIAS("tc_new");
-void operator delete(void* p) __THROW            ALIAS("tc_delete");
-void* operator new[](size_t size) throw (std::bad_alloc) ALIAS("tc_newarray");
-void operator delete[](void* p) __THROW          ALIAS("tc_deletearray");
-void* operator new(size_t size, const std::nothrow_t&) __THROW
-                                                 ALIAS("tc_new_nothrow");
-void* operator new[](size_t size, const std::nothrow_t&) __THROW
-                                                 ALIAS("tc_newarray_nothrow");
-void operator delete(void* size, const std::nothrow_t&) __THROW
-                                                 ALIAS("tc_delete_nothrow");
-void operator delete[](void* size, const std::nothrow_t&) __THROW
-                                                ALIAS("tc_deletearray_nothrow");
-extern "C" {
-  void* malloc(size_t size) __THROW              ALIAS("tc_malloc");
-  void  free(void* ptr) __THROW                  ALIAS("tc_free");
-  void* realloc(void* ptr, size_t size) __THROW  ALIAS("tc_realloc");
-  void* calloc(size_t n, size_t size) __THROW    ALIAS("tc_calloc");
-  void  cfree(void* ptr) __THROW                 ALIAS("tc_cfree");
-  void* memalign(size_t align, size_t s) __THROW ALIAS("tc_memalign");
-  void* valloc(size_t size) __THROW              ALIAS("tc_valloc");
-  void* pvalloc(size_t size) __THROW             ALIAS("tc_pvalloc");
-  int posix_memalign(void** r, size_t a, size_t s) __THROW
-      ALIAS("tc_posix_memalign");
-  void malloc_stats(void) __THROW                ALIAS("tc_malloc_stats");
-  int mallopt(int cmd, int value) __THROW        ALIAS("tc_mallopt");
-#ifdef HAVE_STRUCT_MALLINFO
-  struct mallinfo mallinfo(void) __THROW         ALIAS("tc_mallinfo");
-#endif
-  size_t malloc_size(void* p) __THROW            ALIAS("tc_malloc_size");
-  size_t malloc_usable_size(void* p) __THROW     ALIAS("tc_malloc_size");
-}   // extern "C"
-#else  // #if defined(__GNUC__) && !defined(__MACH__)
-// Portable wrappers
-void* operator new(size_t size)                  { return tc_new(size);       }
-void operator delete(void* p) __THROW            { tc_delete(p);              }
-void* operator new[](size_t size)                { return tc_newarray(size);  }
-void operator delete[](void* p) __THROW          { tc_deletearray(p);         }
-void* operator new(size_t size, const std::nothrow_t& nt) __THROW {
-  return tc_new_nothrow(size, nt);
-}
-void* operator new[](size_t size, const std::nothrow_t& nt) __THROW {
-  return tc_newarray_nothrow(size, nt);
-}
-void operator delete(void* ptr, const std::nothrow_t& nt) __THROW {
-  return tc_delete_nothrow(ptr, nt);
-}
-void operator delete[](void* ptr, const std::nothrow_t& nt) __THROW {
-  return tc_deletearray_nothrow(ptr, nt);
-}
-extern "C" {
-  void* malloc(size_t s) __THROW                 { return tc_malloc(s);       }
-  void  free(void* p) __THROW                    { tc_free(p);                }
-  void* realloc(void* p, size_t s) __THROW       { return tc_realloc(p, s);   }
-  void* calloc(size_t n, size_t s) __THROW       { return tc_calloc(n, s);    }
-  void  cfree(void* p) __THROW                   { tc_cfree(p);               }
-  void* memalign(size_t a, size_t s) __THROW     { return tc_memalign(a, s);  }
-  void* valloc(size_t s) __THROW                 { return tc_valloc(s);       }
-  void* pvalloc(size_t s) __THROW                { return tc_pvalloc(s);      }
-  int posix_memalign(void** r, size_t a, size_t s) __THROW {
-    return tc_posix_memalign(r, a, s);
-  }
-  void malloc_stats(void) __THROW                { tc_malloc_stats();         }
-  int mallopt(int cmd, int v) __THROW            { return tc_mallopt(cmd, v); }
-#ifdef HAVE_STRUCT_MALLINFO
-  struct mallinfo mallinfo(void) __THROW         { return tc_mallinfo();      }
-#endif
-  size_t malloc_size(void* p) __THROW            { return tc_malloc_size(p); }
-  size_t malloc_usable_size(void* p) __THROW     { return tc_malloc_size(p); }
-}  // extern "C"
-#endif  // #if defined(__GNUC__)
-
-// Some library routines on RedHat 9 allocate memory using malloc()
-// and free it using __libc_free() (or vice-versa).  Since we provide
-// our own implementations of malloc/free, we need to make sure that
-// the __libc_XXX variants (defined as part of glibc) also point to
-// the same implementations.
-#ifdef __GLIBC__       // only glibc defines __libc_*
-extern "C" {
-#ifdef ALIAS
-  void* __libc_malloc(size_t size)               ALIAS("tc_malloc");
-  void  __libc_free(void* ptr)                   ALIAS("tc_free");
-  void* __libc_realloc(void* ptr, size_t size)   ALIAS("tc_realloc");
-  void* __libc_calloc(size_t n, size_t size)     ALIAS("tc_calloc");
-  void  __libc_cfree(void* ptr)                  ALIAS("tc_cfree");
-  void* __libc_memalign(size_t align, size_t s)  ALIAS("tc_memalign");
-  void* __libc_valloc(size_t size)               ALIAS("tc_valloc");
-  void* __libc_pvalloc(size_t size)              ALIAS("tc_pvalloc");
-  int __posix_memalign(void** r, size_t a, size_t s) ALIAS("tc_posix_memalign");
-#else  // #ifdef ALIAS
-  void* __libc_malloc(size_t size)               { return malloc(size);       }
-  void  __libc_free(void* ptr)                   { free(ptr);                 }
-  void* __libc_realloc(void* ptr, size_t size)   { return realloc(ptr, size); }
-  void* __libc_calloc(size_t n, size_t size)     { return calloc(n, size);    }
-  void  __libc_cfree(void* ptr)                  { cfree(ptr);                }
-  void* __libc_memalign(size_t align, size_t s)  { return memalign(align, s); }
-  void* __libc_valloc(size_t size)               { return valloc(size);       }
-  void* __libc_pvalloc(size_t size)              { return pvalloc(size);      }
-  int __posix_memalign(void** r, size_t a, size_t s) {
-    return posix_memalign(r, a, s);
-  }
-#endif  // #ifdef ALIAS
-}   // extern "C"
-#endif  // ifdef __GLIBC__
-
-#undef ALIAS
-
-#endif  // #ifndef(WIN32_DO_PATCHING)
-
-
 // ----------------------- IMPLEMENTATION -------------------------------
 
 static int tc_new_mode = 0;  // See tc_set_new_mode().
@@ -430,9 +309,10 @@
   for (int cl = 0; cl < kNumClasses; ++cl) {
     const int length = Static::central_cache()[cl].length();
     const int tc_length = Static::central_cache()[cl].tc_length();
+    const size_t cache_overhead = Static::central_cache()[cl].OverheadBytes();
     const size_t size = static_cast<uint64_t>(
         Static::sizemap()->ByteSizeForClass(cl));
-    r->central_bytes += (size * length);
+    r->central_bytes += (size * length) + cache_overhead;
     r->transfer_bytes += (size * tc_length);
     if (class_count) class_count[cl] = length + tc_length;
   }
@@ -811,6 +691,25 @@
   // file.
   virtual size_t GetAllocatedSize(void* ptr);
 
+  // This duplicates some of the logic in GetSizeWithCallback, but is
+  // faster.  This is important on OS X, where this function is called
+  // on every allocation operation.
+  virtual Ownership GetOwnership(const void* ptr) {
+    const PageID p = reinterpret_cast<uintptr_t>(ptr) >> kPageShift;
+    // The rest of tcmalloc assumes that all allocated pointers use at
+    // most kAddressBits bits.  If ptr doesn't, then it definitely
+    // wasn't alloacted by tcmalloc.
+    if ((p >> (kAddressBits - kPageShift)) > 0) {
+      return kNotOwned;
+    }
+    size_t cl = Static::pageheap()->GetSizeClassIfCached(p);
+    if (cl != 0) {
+      return kOwned;
+    }
+    const Span *span = Static::pageheap()->GetDescriptor(p);
+    return span ? kOwned : kNotOwned;
+  }
+
   virtual void GetFreeListSizes(vector<MallocExtension::FreeListInfo>* v) {
     static const char* kCentralCacheType = "tcmalloc.central";
     static const char* kTransferCacheType = "tcmalloc.transfer";
@@ -924,9 +823,12 @@
     // Check whether the kernel also supports TLS (needs to happen at runtime)
     tcmalloc::CheckIfKernelSupportsTLS();
 #endif
-#ifdef WIN32_DO_PATCHING
-    // patch the windows VirtualAlloc, etc.
-    PatchWindowsFunctions();    // defined in windows/patch_functions.cc
+    ReplaceSystemAlloc();    // defined in libc_override_*.h
+#if defined(__APPLE__)
+    // To break the recursive call of malloc, as malloc -> TCMALLOC_MESSAGE
+    // -> snprintf -> localeconv_l -> malloc, on MacOS.
+    char buf[32];
+    snprintf(buf, sizeof(buf), "%d", tcmallocguard_refcount);
 #endif
     tc_free(tc_malloc(1));
     ThreadCache::InitTSD();
@@ -1023,8 +925,8 @@
   static const int N = 1000;
   char buffer[N];
   TCMalloc_Printer printer(buffer, N);
-  printer.printf("tcmalloc: large alloc %llu bytes == %p @ ",
-                 static_cast<unsigned long long>(num_pages) << kPageShift,
+  printer.printf("tcmalloc: large alloc %"PRIu64" bytes == %p @ ",
+                 static_cast<uint64>(num_pages) << kPageShift,
                  result);
   for (int i = 0; i < stack.depth; i++) {
     printer.printf(" %p", stack.stack[i]);
@@ -1184,6 +1086,8 @@
   return do_free_with_callback(ptr, &InvalidFree);
 }
 
+// NOTE: some logic here is duplicated in GetOwnership (above), for
+// speed.  If you change this function, look at that one too.
 inline size_t GetSizeWithCallback(void* ptr,
                                   size_t (*invalid_getsize_fn)(void*)) {
   if (ptr == NULL)
@@ -1193,7 +1097,7 @@
   if (cl != 0) {
     return Static::sizemap()->ByteSizeForClass(cl);
   } else {
-    Span *span = Static::pageheap()->GetDescriptor(p);
+    const Span *span = Static::pageheap()->GetDescriptor(p);
     if (span == NULL) {  // means we do not own this memory
       return (*invalid_getsize_fn)(ptr);
     } else if (span->sizeclass != 0) {
@@ -1476,6 +1380,8 @@
 
 // As promised, the definition of this function, declared above.
 size_t TCMallocImplementation::GetAllocatedSize(void* ptr) {
+  ASSERT(TCMallocImplementation::GetOwnership(ptr)
+         != TCMallocImplementation::kNotOwned);
   return GetSizeWithCallback(ptr, &InvalidGetAllocatedSize);
 }
 
@@ -1673,26 +1579,7 @@
 #endif
 
 extern "C" PERFTOOLS_DLL_DECL size_t tc_malloc_size(void* ptr) __THROW {
-  return GetSizeWithCallback(ptr, &InvalidGetAllocatedSize);
+  return MallocExtension::instance()->GetAllocatedSize(ptr);
 }
 
-
-// Override __libc_memalign in libc on linux boxes specially.
-// They have a bug in libc that causes them to (very rarely) allocate
-// with __libc_memalign() yet deallocate with free() and the
-// definitions above don't catch it.
-// This function is an exception to the rule of calling MallocHook method
-// from the stack frame of the allocation function;
-// heap-checker handles this special case explicitly.
-static void *MemalignOverride(size_t align, size_t size, const void *caller)
-    __THROW ATTRIBUTE_SECTION(google_malloc);
-
-static void *MemalignOverride(size_t align, size_t size, const void *caller)
-    __THROW {
-  void* result = do_memalign_or_cpp_memalign(align, size);
-  MallocHook::InvokeNewHook(result, size);
-  return result;
-}
-void *(*__memalign_hook)(size_t, size_t, const void *) = MemalignOverride;
-
 #endif  // TCMALLOC_USING_DEBUGALLOCATION