[sanitizer] [msan] Unify aarch64 mapping

[zatrazz]

Summary

This patch unify the 39-bit and 42-bit mapping for aarch64 using only
one instrumentation algorithm (based on 39-bit mapping). This removes
compiler flag SANITIZER_AARCH64_VMA requirement for msan on aarch64.

The unifies mapping scheme now is:

{0x00000000000ULL, 0x04000000000ULL, MappingDesc::INVALID, "invalid"},
{0x04000000000ULL, 0x04100000000ULL, MappingDesc::SHADOW,  "shadow-1"},
{0x04100000000ULL, 0x04300000000ULL, MappingDesc::SHADOW,  "shadow-2"},
{0x04300000000ULL, 0x04400000000ULL, MappingDesc::ORIGIN,  "origin-1"},
{0x04400000000ULL, 0x04600000000ULL, MappingDesc::ORIGIN,  "origin-2"},
{0x04600000000ULL, 0x05500000000ULL, MappingDesc::INVALID, "invalid"},
{0x05500000000ULL, 0x05600000000ULL, MappingDesc::APP,     "app-1 (39)"},
{0x05600000000ULL, 0x07000000000ULL, MappingDesc::INVALID, "invalid"},
{0x07000000000ULL, 0x08000000000ULL, MappingDesc::APP,     "app-2 (39)"},
{0x08000000000ULL, 0x2AA00000000ULL, MappingDesc::INVALID, "invalid"},
// The segments below are only valid and mapped for 42-bit VMA kernels.
{0x2AA00000000ULL, 0x2AB00000000ULL, MappingDesc::APP,     "app-3 (42)"},
{0x2AB00000000ULL, 0x2C300000000ULL, MappingDesc::INVALID, "invalid"},
{0x2C300000000ULL, 0x2C400000000ULL, MappingDesc::SHADOW,  "shadow-3"},
{0x2C400000000ULL, 0x2C600000000ULL, MappingDesc::INVALID, "invalid"},
{0x2C600000000ULL, 0x2C700000000ULL, MappingDesc::ORIGIN,  "origin-3"},
{0x2C700000000ULL, 0x3C100000000ULL, MappingDesc::INVALID, "invalid"},
{0x3C100000000ULL, 0x3C300000000ULL, MappingDesc::SHADOW,  "shadow-4"},
{0x3C300000000ULL, 0x3C400000000ULL, MappingDesc::INVALID, "invalid"},
{0x3C400000000ULL, 0x3C600000000ULL, MappingDesc::ORIGIN,  "origin-5"},
{0x3C600000000ULL, 0x3F000000000ULL, MappingDesc::INVALID, "invalid"},
{0x3F000000000ULL, 0x40000000000ULL, MappingDesc::APP,     "app-4 (42)"},

[zatrazz]

This updates uses a different strategy, similar to the one for x86_64 new mappings. The idea is to use XOR instead and by creating multiple segments for 42-bits to increase VMA available for program use. The compiler-rt patch [1] has the information regarding the new mapping segments.

[1] http://reviews.llvm.org/D13818

[eugenis]

LGTM

lib/Transforms/Instrumentation/MemorySanitizer.cpp

 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Utils/ModuleUtils.h"
 
 using namespace llvm;
 
 #define DEBUG_TYPE "msan"
 
-// VMA size definition for architecture that support multiple sizes.
-// AArch64 has 3 VMA sizes: 39, 42 and 48.
-#ifndef SANITIZER_AARCH64_VMA
-# define SANITIZER_AARCH64_VMA 39
-#else
-# if SANITIZER_AARCH64_VMA != 39 && SANITIZER_AARCH64_VMA != 42
-#  error "invalid SANITIZER_AARCH64_VMA size"
-# endif
-#endif
-
 static const unsigned kOriginSize = 4;
 static const unsigned kMinOriginAlignment = 4;
 static const unsigned kShadowTLSAlignment = 8;
 
 // These constants must be kept in sync with the ones in msan.h.
 static const unsigned kParamTLSSize = 800;
 static const unsigned kRetvalTLSSize = 800;
 
static const MemoryMapParams Linux_PowerPC64_MemoryMapParams = {
   0x200000000000,  // AndMask
   0x100000000000,  // XorMask
   0x080000000000,  // ShadowBase
   0x1C0000000000,  // OriginBase
 };
 
 // aarch64 Linux
 static const MemoryMapParams Linux_AArch64_MemoryMapParams = {
-#if SANITIZER_AARCH64_VMA == 39
-  0x007C00000000,  // AndMask
-  0x000100000000,  // XorMask
-  0x004000000000,  // ShadowBase
-  0x004300000000,  // OriginBase
-#elif SANITIZER_AARCH64_VMA == 42
-  0x03E000000000,  // AndMask
-  0x001000000000,  // XorMask
-  0x010000000000,  // ShadowBase
-  0x012000000000,  // OriginBase
-#endif
+  0,               // AndMask (not used)
+  0x06000000000,   // XorMask
+  0,               // ShadowBase (not used)
+  0x01000000000,   // OriginBase
 };
 
 // i386 FreeBSD
 static const MemoryMapParams FreeBSD_I386_MemoryMapParams = {
   0x000180000000,  // AndMask
   0x000040000000,  // XorMask
   0x000020000000,  // ShadowBase
   0x000700000000,  // OriginBase