Update detours library.

detours/creatwth.cpp

@@ -7,34 +7,15 @@
 //  Copyright (c) Microsoft Corporation.  All rights reserved.
 //
 
-#if _MSC_VER >= 1900
-#pragma warning(push)
-#pragma warning(disable:4091) // empty typedef
-#endif
-#define _CRT_STDIO_ARBITRARY_WIDE_SPECIFIERS 1
-#define _ARM_WINAPI_PARTITION_DESKTOP_SDK_AVAILABLE 1
-#include <windows.h>
-#include <stddef.h>
-#pragma warning(push)
-#if _MSC_VER > 1400
-#pragma warning(disable:6102 6103) // /analyze warnings
-#endif
-#include <strsafe.h>
-#pragma warning(pop)
-
 // #define DETOUR_DEBUG 1
 #define DETOURS_INTERNAL
-
 #include "detours.h"
+#include <stddef.h>
 
 #if DETOURS_VERSION != 0x4c0c1   // 0xMAJORcMINORcPATCH
 #error detours.h version mismatch
 #endif
 
-#if _MSC_VER >= 1900
-#pragma warning(pop)
-#endif
-
 #define IMPORT_DIRECTORY OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_IMPORT]
 #define BOUND_DIRECTORY OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_BOUND_IMPORT]
 #define CLR_DIRECTORY OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_COM_DESCRIPTOR]

detours/detours.cpp

@@ -7,33 +7,15 @@
 //  Copyright (c) Microsoft Corporation.  All rights reserved.
 //
 
-#pragma warning(disable:4068) // unknown pragma (suppress)
-
-#if _MSC_VER >= 1900
-#pragma warning(push)
-#pragma warning(disable:4091) // empty typedef
-#endif
-
-#define _ARM_WINAPI_PARTITION_DESKTOP_SDK_AVAILABLE 1
-#include <windows.h>
-
-#if (_MSC_VER < 1299)
-#pragma warning(disable: 4710)
-#endif
 
 //#define DETOUR_DEBUG 1
 #define DETOURS_INTERNAL
-
 #include "detours.h"
 
 #if DETOURS_VERSION != 0x4c0c1   // 0xMAJORcMINORcPATCH
 #error detours.h version mismatch
 #endif
 
-#if _MSC_VER >= 1900
-#pragma warning(pop)
-#endif
-
 #define NOTHROW
 
 //////////////////////////////////////////////////////////////////////////////
@@ -880,7 +862,8 @@ struct _DETOUR_TRAMPOLINE
 {
     // An ARM64 instruction is 4 bytes long.
     //
-    // The overwrite is always 2 instructions plus a literal, so 16 bytes, 4 instructions.
+    // The overwrite is always composed of 3 instructions (12 bytes) which perform an indirect jump
+	// using _DETOUR_TRAMPOLINE::pbDetour as the address holding the target location.
     //
     // Copied instructions can expand.
     //
@@ -915,7 +898,7 @@ struct _DETOUR_TRAMPOLINE
 C_ASSERT(sizeof(_DETOUR_TRAMPOLINE) == 184);
 
 enum {
-    SIZE_OF_JMP = 16
+    SIZE_OF_JMP = 12
 };
 
 inline ULONG fetch_opcode(PBYTE pbCode)
@@ -929,6 +912,79 @@ inline void write_opcode(PBYTE &pbCode, ULONG Opcode)
     pbCode += 4;
 }
 
+struct ARM64_INDIRECT_JMP {
+    struct {
+        ULONG Rd : 5;
+        ULONG immhi : 19;
+        ULONG iop : 5;
+        ULONG immlo : 2;
+        ULONG op : 1;
+    } ardp;
+
+    struct {
+        ULONG Rt : 5;
+        ULONG Rn : 5;
+        ULONG imm : 12;
+        ULONG opc : 2;
+        ULONG iop1 : 2;
+        ULONG V : 1;
+        ULONG iop2 : 3;
+        ULONG size : 2;
+    } ldr;
+
+    ULONG br;
+};
+
+#pragma warning(push)
+#pragma warning(disable:4201)
+
+union ARM64_INDIRECT_IMM {
+    struct {
+        ULONG64 pad : 12;
+        ULONG64 adrp_immlo : 2;
+        ULONG64 adrp_immhi : 19;
+    };
+
+    LONG64 value;
+};
+
+#pragma warning(pop)
+
+PBYTE detour_gen_jmp_indirect(BYTE *pbCode, ULONG64 *pbJmpVal)
+{
+    // adrp x17, [jmpval]
+    // ldr x17, [x17, jmpval]
+    // br x17
+
+    struct ARM64_INDIRECT_JMP *pIndJmp;
+    union ARM64_INDIRECT_IMM jmpIndAddr;
+
+    jmpIndAddr.value = (((LONG64)pbJmpVal) & 0xFFFFFFFFFFFFF000) - 
+                       (((LONG64)pbCode) & 0xFFFFFFFFFFFFF000);
+
+    pIndJmp = (struct ARM64_INDIRECT_JMP *)pbCode;
+    pbCode = (BYTE *)(pIndJmp + 1);
+
+    pIndJmp->ardp.Rd = 17;
+    pIndJmp->ardp.immhi = jmpIndAddr.adrp_immhi;
+    pIndJmp->ardp.iop = 0x10;
+    pIndJmp->ardp.immlo = jmpIndAddr.adrp_immlo;
+    pIndJmp->ardp.op = 1;
+
+    pIndJmp->ldr.Rt = 17;
+    pIndJmp->ldr.Rn = 17;
+    pIndJmp->ldr.imm = (((ULONG64)pbJmpVal) & 0xFFF) / 8;
+    pIndJmp->ldr.opc = 1;
+    pIndJmp->ldr.iop1 = 1;
+    pIndJmp->ldr.V = 0;
+    pIndJmp->ldr.iop2 = 7;
+    pIndJmp->ldr.size = 3;
+
+    pIndJmp->br = 0xD61F0220;
+
+    return pbCode;
+}
+
 PBYTE detour_gen_jmp_immediate(PBYTE pbCode, PBYTE *ppPool, PBYTE pbJmpVal)
 {
     PBYTE pbLiteral;
@@ -995,7 +1051,7 @@ inline PBYTE detour_skip_jmp(PBYTE pbCode, PVOID *ppGlobals)
     the bottom 12 bits cleared to zero, and then writes the result to a general-purpose register. This permits the
     calculation of the address at a 4KB aligned memory region. In conjunction with an ADD (immediate) instruction, or
     a Load/Store instruction with a 12-bit immediate offset, this allows for the calculation of, or access to, any address
-    within ±4GB of the current PC.
+    within +/- 4GB of the current PC.
 
 PC-rel. addressing
     This section describes the encoding of the PC-rel. addressing instruction class. The encodings in this section are
@@ -1062,7 +1118,10 @@ inline void detour_find_jmp_bounds(PBYTE pbCode,
                                    PDETOUR_TRAMPOLINE *ppLower,
                                    PDETOUR_TRAMPOLINE *ppUpper)
 {
-    // We have to place trampolines within +/- 2GB of code.
+    // The encoding used by detour_gen_jmp_indirect actually enables a
+    // displacement of +/- 4GiB. In the future, this could be changed to
+    // reflect that. For now, just reuse the x86 logic which is plenty.
+
     ULONG_PTR lo = detour_2gb_below((ULONG_PTR)pbCode);
     ULONG_PTR hi = detour_2gb_above((ULONG_PTR)pbCode);
     DETOUR_TRACE(("[%p..%p..%p]\n", lo, pbCode, hi));
@@ -1250,6 +1309,65 @@ static PVOID detour_alloc_region_from_hi(PBYTE pbLo, PBYTE pbHi)
     return NULL;
 }
 
+static PVOID detour_alloc_trampoline_allocate_new(PBYTE pbTarget,
+                                                  PDETOUR_TRAMPOLINE pLo,
+                                                  PDETOUR_TRAMPOLINE pHi)
+{
+    PVOID pbTry = NULL;
+
+    // NB: We must always also start the search at an offset from pbTarget
+    //     in order to maintain ASLR entropy.
+
+#if defined(DETOURS_64BIT)
+    // Try looking 1GB below or lower.
+    if (pbTry == NULL && pbTarget > (PBYTE)0x40000000) {
+        pbTry = detour_alloc_region_from_hi((PBYTE)pLo, pbTarget - 0x40000000);
+    }
+    // Try looking 1GB above or higher.
+    if (pbTry == NULL && pbTarget < (PBYTE)0xffffffff40000000) {
+        pbTry = detour_alloc_region_from_lo(pbTarget + 0x40000000, (PBYTE)pHi);
+    }
+    // Try looking 1GB below or higher.
+    if (pbTry == NULL && pbTarget > (PBYTE)0x40000000) {
+        pbTry = detour_alloc_region_from_lo(pbTarget - 0x40000000, pbTarget);
+    }
+    // Try looking 1GB above or lower.
+    if (pbTry == NULL && pbTarget < (PBYTE)0xffffffff40000000) {
+        pbTry = detour_alloc_region_from_hi(pbTarget, pbTarget + 0x40000000);
+    }
+#endif
+
+    // Try anything below.
+    if (pbTry == NULL) {
+        pbTry = detour_alloc_region_from_hi((PBYTE)pLo, pbTarget);
+    }
+    // try anything above.
+    if (pbTry == NULL) {
+        pbTry = detour_alloc_region_from_lo(pbTarget, (PBYTE)pHi);
+    }
+
+    return pbTry;
+}
+
+PVOID WINAPI DetourAllocateRegionWithinJumpBounds(_In_ LPCVOID pbTarget,
+                                                  _Out_ PDWORD pcbAllocatedSize)
+{
+    PDETOUR_TRAMPOLINE pLo;
+    PDETOUR_TRAMPOLINE pHi;
+    detour_find_jmp_bounds((PBYTE)pbTarget, &pLo, &pHi);
+
+    PVOID pbNewlyAllocated =
+        detour_alloc_trampoline_allocate_new((PBYTE)pbTarget, pLo, pHi);
+    if (pbNewlyAllocated == NULL) {
+        DETOUR_TRACE(("Couldn't find available memory region!\n"));
+        *pcbAllocatedSize = 0;
+        return NULL;
+    }
+
+    *pcbAllocatedSize = DETOUR_REGION_SIZE;
+    return pbNewlyAllocated;
+}
+
 static PDETOUR_TRAMPOLINE detour_alloc_trampoline(PBYTE pbTarget)
 {
     // We have to place trampolines within +/- 2GB of target.
@@ -1294,41 +1412,10 @@ static PDETOUR_TRAMPOLINE detour_alloc_trampoline(PBYTE pbTarget)
     // Round pbTarget down to 64KB block.
     pbTarget = pbTarget - (PtrToUlong(pbTarget) & 0xffff);
 
-    PVOID pbTry = NULL;
+    PVOID pbNewlyAllocated =
-
+        detour_alloc_trampoline_allocate_new(pbTarget, pLo, pHi);
-    // NB: We must always also start the search at an offset from pbTarget
+    if (pbNewlyAllocated != NULL) {
-    //     in order to maintain ASLR entropy.
+        s_pRegion = (DETOUR_REGION*)pbNewlyAllocated;
-
-#if defined(DETOURS_64BIT)
-    // Try looking 1GB below or lower.
-    if (pbTry == NULL && pbTarget > (PBYTE)0x40000000) {
-        pbTry = detour_alloc_region_from_hi((PBYTE)pLo, pbTarget - 0x40000000);
-    }
-    // Try looking 1GB above or higher.
-    if (pbTry == NULL && pbTarget < (PBYTE)0xffffffff40000000) {
-        pbTry = detour_alloc_region_from_lo(pbTarget + 0x40000000, (PBYTE)pHi);
-    }
-    // Try looking 1GB below or higher.
-    if (pbTry == NULL && pbTarget > (PBYTE)0x40000000) {
-        pbTry = detour_alloc_region_from_lo(pbTarget - 0x40000000, pbTarget);
-    }
-    // Try looking 1GB above or lower.
-    if (pbTry == NULL && pbTarget < (PBYTE)0xffffffff40000000) {
-        pbTry = detour_alloc_region_from_hi(pbTarget, pbTarget + 0x40000000);
-    }
-#endif
-
-    // Try anything below.
-    if (pbTry == NULL) {
-        pbTry = detour_alloc_region_from_hi((PBYTE)pLo, pbTarget);
-    }
-    // try anything above.
-    if (pbTry == NULL) {
-        pbTry = detour_alloc_region_from_lo(pbTarget, (PBYTE)pHi);
-    }
-
-    if (pbTry != NULL) {
-        s_pRegion = (DETOUR_REGION*)pbTry;
         s_pRegion->dwSignature = DETOUR_REGION_SIGNATURE;
         s_pRegion->pFree = NULL;
         s_pRegion->pNext = s_pRegions;
@@ -1655,7 +1742,7 @@ LONG WINAPI DetourTransactionCommitEx(_Out_opt_ PVOID **pppFailedPointer)
 #endif // DETOURS_ARM
 
 #ifdef DETOURS_ARM64
-            PBYTE pbCode = detour_gen_jmp_immediate(o->pbTarget, NULL, o->pTrampoline->pbDetour);
+            PBYTE pbCode = detour_gen_jmp_indirect(o->pbTarget, (ULONG64*)&(o->pTrampoline->pbDetour));
             pbCode = detour_gen_brk(pbCode, o->pTrampoline->pbRemain);
             *o->ppbPointer = o->pTrampoline->rbCode;
             UNREFERENCED_PARAMETER(pbCode);

detours/detours.h

@@ -16,6 +16,34 @@
 //////////////////////////////////////////////////////////////////////////////
 //
 
+#ifdef DETOURS_INTERNAL
+
+#define _CRT_STDIO_ARBITRARY_WIDE_SPECIFIERS 1
+#define _ARM_WINAPI_PARTITION_DESKTOP_SDK_AVAILABLE 1
+
+#pragma warning(disable:4068) // unknown pragma (suppress)
+
+#if _MSC_VER >= 1900
+#pragma warning(push)
+#pragma warning(disable:4091) // empty typedef
+#endif
+
+#include <windows.h>
+#if (_MSC_VER < 1310)
+#else
+#pragma warning(push)
+#if _MSC_VER > 1400
+#pragma warning(disable:6102 6103) // /analyze warnings
+#endif
+#include <strsafe.h>
+#pragma warning(pop)
+#endif
+
+#endif // DETOURS_INTERNAL
+
+//////////////////////////////////////////////////////////////////////////////
+//
+
 #undef DETOURS_X64
 #undef DETOURS_X86
 #undef DETOURS_IA64
@@ -61,7 +89,12 @@
 //#define DETOURS_OPTION_BITS 32
 #endif
 
-#define VER_DETOURS_BITS    DETOUR_STRINGIFY(DETOURS_BITS)
+/////////////////////////////////////////////////////////////// Helper Macros.
+//
+#define DETOURS_STRINGIFY_(x)    #x
+#define DETOURS_STRINGIFY(x)    DETOURS_STRINGIFY_(x)
+
+#define VER_DETOURS_BITS    DETOURS_STRINGIFY(DETOURS_BITS)
 
 //////////////////////////////////////////////////////////////////////////////
 //
@@ -387,7 +420,6 @@ typedef struct _DETOUR_EXE_RESTORE
 #ifdef IMAGE_NT_OPTIONAL_HDR64_MAGIC    // some environments do not have this
         BYTE                raw[sizeof(IMAGE_NT_HEADERS64) +
                                 sizeof(IMAGE_SECTION_HEADER) * 32];
-        C_ASSERT(sizeof(IMAGE_NT_HEADERS64) == 0x108);
 #else
         BYTE                raw[0x108 + sizeof(IMAGE_SECTION_HEADER) * 32];
 #endif
@@ -396,6 +428,10 @@ typedef struct _DETOUR_EXE_RESTORE
 
 } DETOUR_EXE_RESTORE, *PDETOUR_EXE_RESTORE;
 
+#ifdef IMAGE_NT_OPTIONAL_HDR64_MAGIC
+C_ASSERT(sizeof(IMAGE_NT_HEADERS64) == 0x108);
+#endif
+
 // The size can change, but assert for clarity due to the muddying #ifdefs.
 #ifdef _WIN64
 C_ASSERT(sizeof(DETOUR_EXE_RESTORE) == 0x688);
@@ -431,11 +467,6 @@ typedef struct _DETOUR_EXE_HELPER
       0,\
 }
 
-/////////////////////////////////////////////////////////////// Helper Macros.
-//
-#define DETOURS_STRINGIFY(x)    DETOURS_STRINGIFY_(x)
-#define DETOURS_STRINGIFY_(x)    #x
-
 ///////////////////////////////////////////////////////////// Binary Typedefs.
 //
 typedef BOOL (CALLBACK *PF_DETOUR_BINARY_BYWAY_CALLBACK)(
@@ -523,6 +554,8 @@ PVOID WINAPI DetourCopyInstruction(_In_opt_ PVOID pDst,
                                    _Out_opt_ LONG *plExtra);
 BOOL WINAPI DetourSetCodeModule(_In_ HMODULE hModule,
                                 _In_ BOOL fLimitReferencesToModule);
+PVOID WINAPI DetourAllocateRegionWithinJumpBounds(_In_ LPCVOID pbTarget,
+                                                  _Out_ PDWORD pcbAllocatedSize);
 
 ///////////////////////////////////////////////////// Loaded Binary Functions.
 //

detours/detver.h

@@ -14,8 +14,8 @@
 #include <detours.h>
 #else
 #ifndef DETOURS_STRINGIFY
-#define DETOURS_STRINGIFY(x)    DETOURS_STRINGIFY_(x)
 #define DETOURS_STRINGIFY_(x)    #x
+#define DETOURS_STRINGIFY(x)    DETOURS_STRINGIFY_(x)
 #endif
 
 #define VER_FILEFLAGSMASK   0x3fL
@@ -24,4 +24,4 @@
 #define VER_FILETYPE        0x00000002L
 #define VER_FILESUBTYPE     0x00000000L
 #endif
-#define VER_DETOURS_BITS    DETOUR_STRINGIFY(DETOURS_BITS)
+#define VER_DETOURS_BITS    DETOURS_STRINGIFY(DETOURS_BITS)

detours/disasm.cpp

@@ -7,28 +7,15 @@
 //  Copyright (c) Microsoft Corporation.  All rights reserved.
 //
 
-#if _MSC_VER >= 1900
-#pragma warning(push)
-#pragma warning(disable:4091) // empty typedef
-#endif
-
-#define _ARM_WINAPI_PARTITION_DESKTOP_SDK_AVAILABLE 1
-#include <windows.h>
-#include <limits.h>
-
 // #define DETOUR_DEBUG 1
 #define DETOURS_INTERNAL
-
 #include "detours.h"
+#include <limits.h>
 
 #if DETOURS_VERSION != 0x4c0c1   // 0xMAJORcMINORcPATCH
 #error detours.h version mismatch
 #endif
 
-#if _MSC_VER >= 1900
-#pragma warning(pop)
-#endif
-
 #undef ASSERT
 #define ASSERT(x)
 
@@ -260,6 +247,11 @@ class CDetourDis
 #define ENTRY_CopyFF                ENTRY_DataIgnored &CDetourDis::CopyFF
 #define ENTRY_CopyVex2              ENTRY_DataIgnored &CDetourDis::CopyVex2
 #define ENTRY_CopyVex3              ENTRY_DataIgnored &CDetourDis::CopyVex3
+#define ENTRY_CopyEvex              ENTRY_DataIgnored &CDetourDis::CopyEvex // 62, 3 byte payload, then normal with implied prefixes like vex
+#define ENTRY_CopyXop               ENTRY_DataIgnored &CDetourDis::CopyXop   // 0x8F ... POP /0 or AMD XOP
+#define ENTRY_CopyBytesXop          5, 5, 4, 0, 0, &CDetourDis::CopyBytes // 0x8F xop1 xop2 opcode modrm
+#define ENTRY_CopyBytesXop1         6, 6, 4, 0, 0, &CDetourDis::CopyBytes // 0x8F xop1 xop2 opcode modrm ... imm8
+#define ENTRY_CopyBytesXop4         9, 9, 4, 0, 0, &CDetourDis::CopyBytes // 0x8F xop1 xop2 opcode modrm ... imm32
 #define ENTRY_Invalid               ENTRY_DataIgnored &CDetourDis::Invalid
 #define ENTRY_End                   ENTRY_DataIgnored NULL
 
@@ -289,6 +281,9 @@ class CDetourDis
     PBYTE CopyVex2(REFCOPYENTRY pEntry, PBYTE pbDst, PBYTE pbSrc);
     PBYTE CopyVex3(REFCOPYENTRY pEntry, PBYTE pbDst, PBYTE pbSrc);
     PBYTE CopyVexCommon(BYTE m, PBYTE pbDst, PBYTE pbSrc);
+    PBYTE CopyVexEvexCommon(BYTE m, PBYTE pbDst, PBYTE pbSrc, BYTE p);
+    PBYTE CopyEvex(REFCOPYENTRY pEntry, PBYTE pbDst, PBYTE pbSrc);
+    PBYTE CopyXop(REFCOPYENTRY pEntry, PBYTE pbDst, PBYTE pbSrc);
 
   protected:
     static const COPYENTRY  s_rceCopyTable[257];
@@ -303,6 +298,7 @@ class CDetourDis
     BOOL                m_bAddressOverride;
     BOOL                m_bRaxOverride; // AMD64 only
     BOOL                m_bVex;
+    BOOL                m_bEvex;
     BOOL                m_bF2;
     BOOL                m_bF3; // x86 only
     BYTE                m_nSegmentOverride;
@@ -337,6 +333,7 @@ CDetourDis::CDetourDis(_Out_opt_ PBYTE *ppbTarget, _Out_opt_ LONG *plExtra)
     m_bF2 = FALSE;
     m_bF3 = FALSE;
     m_bVex = FALSE;
+    m_bEvex = FALSE;
 
     m_ppbTarget = ppbTarget ? ppbTarget : &m_pbScratchTarget;
     m_plExtra = plExtra ? plExtra : &m_lScratchExtra;
@@ -368,8 +365,11 @@ PBYTE CDetourDis::CopyBytes(REFCOPYENTRY pEntry, PBYTE pbDst, PBYTE pbSrc)
 {
     UINT nBytesFixed;
 
-    ASSERT(!m_bVex || pEntry->nFlagBits == 0);
+    if (m_bVex || m_bEvex)
-    ASSERT(!m_bVex || pEntry->nFixedSize == pEntry->nFixedSize16);
+    {
+        ASSERT(pEntry->nFlagBits == 0);
+        ASSERT(pEntry->nFixedSize == pEntry->nFixedSize16);
+    }
 
     UINT const nModOffset = pEntry->nModOffset;
     UINT const nFlagBits = pEntry->nFlagBits;
@@ -748,32 +748,41 @@ PBYTE CDetourDis::CopyFF(REFCOPYENTRY pEntry, PBYTE pbDst, PBYTE pbSrc)
     return pbOut;
 }
 
-PBYTE CDetourDis::CopyVexCommon(BYTE m, PBYTE pbDst, PBYTE pbSrc)
+PBYTE CDetourDis::CopyVexEvexCommon(BYTE m, PBYTE pbDst, PBYTE pbSrc, BYTE p)
 // m is first instead of last in the hopes of pbDst/pbSrc being
 // passed along efficiently in the registers they were already in.
 {
     static const COPYENTRY ceF38 = { 0x38, ENTRY_CopyBytes2Mod };
     static const COPYENTRY ceF3A = { 0x3A, ENTRY_CopyBytes2Mod1 };
-    static const COPYENTRY Invalid = { 0xC4, ENTRY_Invalid };
+    static const COPYENTRY ceInvalid = { 0xC4, ENTRY_Invalid };
 
-    m_bVex = TRUE;
+    switch (p & 3) {
-    REFCOPYENTRY pEntry;
-    switch (m) {
-    default: pEntry = &Invalid; break;
-    case 1:  pEntry = &s_rceCopyTable0F[pbSrc[0]]; break;
-    case 2:  pEntry = &ceF38; break;
-    case 3:  pEntry = &ceF3A; break;
-    }
-
-    switch (pbSrc[-1] & 3) { // p in last byte
     case 0: break;
     case 1: m_bOperandOverride = TRUE; break;
     case 2: m_bF3 = TRUE; break;
     case 3: m_bF2 = TRUE; break;
     }
 
+    REFCOPYENTRY pEntry;
+
+    switch (m) {
+    default: return Invalid(&ceInvalid, pbDst, pbSrc);
+    case 1:  pEntry = &s_rceCopyTable0F[pbSrc[0]];
              return (this->*pEntry->pfCopy)(pEntry, pbDst, pbSrc);
+    case 2:  return CopyBytes(&ceF38, pbDst, pbSrc);
+    case 3:  return CopyBytes(&ceF3A, pbDst, pbSrc);
     }
+}
+
+PBYTE CDetourDis::CopyVexCommon(BYTE m, PBYTE pbDst, PBYTE pbSrc)
+// m is first instead of last in the hopes of pbDst/pbSrc being
+// passed along efficiently in the registers they were already in.
+{
+    m_bVex = TRUE;
+    BYTE const p = (BYTE)(pbSrc[-1] & 3); // p in last byte
+    return CopyVexEvexCommon(m, pbDst, pbSrc, p);
+}
+
 
 PBYTE CDetourDis::CopyVex3(REFCOPYENTRY, PBYTE pbDst, PBYTE pbSrc)
 // 3 byte VEX prefix 0xC4
@@ -835,6 +844,78 @@ PBYTE CDetourDis::CopyVex2(REFCOPYENTRY, PBYTE pbDst, PBYTE pbSrc)
     return CopyVexCommon(1, pbDst + 2, pbSrc + 2);
 }
 
+PBYTE CDetourDis::CopyEvex(REFCOPYENTRY, PBYTE pbDst, PBYTE pbSrc)
+// 62, 3 byte payload, x86 with implied prefixes like Vex
+// for 32bit, mode 0xC0 else fallback to bound /r
+{
+    // NOTE: Intel and Wikipedia number these differently.
+    // Intel says 0-2, Wikipedia says 1-3.
+
+    BYTE const p0 = pbSrc[1];
+
+#ifdef DETOURS_X86
+    const static COPYENTRY ceBound = { 0x62, ENTRY_CopyBytes2Mod };
+    if ((p0 & 0xC0) != 0xC0) {
+        return CopyBytes(&ceBound, pbDst, pbSrc);
+    }
+#endif
+
+    static const COPYENTRY ceInvalid = { 0x62, ENTRY_Invalid };
+
+    if ((p0 & 0x0C) != 0)
+        return Invalid(&ceInvalid, pbDst, pbSrc);
+
+    BYTE const p1 = pbSrc[2];
+
+    if ((p1 & 0x04) != 0x04)
+        return Invalid(&ceInvalid, pbDst, pbSrc);
+
+    // Copy 4 byte prefix.
+    *(UNALIGNED ULONG *)pbDst = *(UNALIGNED ULONG*)pbSrc;
+
+    m_bEvex = TRUE;
+
+#ifdef DETOURS_X64
+    m_bRaxOverride |= !!(p1 & 0x80); // w
+#endif
+
+    return CopyVexEvexCommon(p0 & 3u, pbDst + 4, pbSrc + 4, p1 & 3u);
+}
+
+PBYTE CDetourDis::CopyXop(REFCOPYENTRY, PBYTE pbDst, PBYTE pbSrc)
+/* 3 byte AMD XOP prefix 0x8F
+byte0: 0x8F
+byte1: RXBmmmmm
+byte2: WvvvvLpp
+byte3: opcode
+mmmmm >= 8, else pop
+mmmmm only otherwise defined for 8, 9, A.
+pp is like VEX but only instructions with 0 are defined
+*/
+{
+    const static COPYENTRY cePop = { 0x8F, ENTRY_CopyBytes2Mod };
+    const static COPYENTRY ceXop = { 0x8F, ENTRY_CopyBytesXop };
+    const static COPYENTRY ceXop1 = { 0x8F, ENTRY_CopyBytesXop1 };
+    const static COPYENTRY ceXop4 = { 0x8F, ENTRY_CopyBytesXop4 };
+
+    BYTE const m = (BYTE)(pbSrc[1] & 0x1F);
+    ASSERT(m <= 10);
+    switch (m)
+    {
+    default:
+        return CopyBytes(&cePop, pbDst, pbSrc);
+
+    case 8: // modrm with 8bit immediate
+        return CopyBytes(&ceXop1, pbDst, pbSrc);
+
+    case 9: // modrm with no immediate
+        return CopyBytes(&ceXop, pbDst, pbSrc);
+
+    case 10: // modrm with 32bit immediate
+        return CopyBytes(&ceXop4, pbDst, pbSrc);
+    }
+}
+
 //////////////////////////////////////////////////////////////////////////////
 //
 PBYTE CDetourDis::s_pbModuleBeg = NULL;
@@ -1030,11 +1111,11 @@ const CDetourDis::COPYENTRY CDetourDis::s_rceCopyTable[257] =
 #ifdef DETOURS_X64
     { 0x60, ENTRY_Invalid },                            // Invalid
     { 0x61, ENTRY_Invalid },                            // Invalid
-    { 0x62, ENTRY_Invalid },                            // Invalid (not yet implemented Intel EVEX support)
+    { 0x62, ENTRY_CopyEvex },                           // EVEX / AVX512
 #else
     { 0x60, ENTRY_CopyBytes1 },                         // PUSHAD
     { 0x61, ENTRY_CopyBytes1 },                         // POPAD
-    { 0x62, ENTRY_CopyBytes2Mod },                      // BOUND /r
+    { 0x62, ENTRY_CopyEvex },                           // BOUND /r and EVEX / AVX512
 #endif
     { 0x63, ENTRY_CopyBytes2Mod },                      // 32bit ARPL /r, 64bit MOVSXD
     { 0x64, ENTRY_CopyBytesSegment },                   // FS prefix
@@ -1084,7 +1165,7 @@ const CDetourDis::COPYENTRY CDetourDis::s_rceCopyTable[257] =
     { 0x8C, ENTRY_CopyBytes2Mod },                      // MOV /r
     { 0x8D, ENTRY_CopyBytes2Mod },                      // LEA /r
     { 0x8E, ENTRY_CopyBytes2Mod },                      // MOV /r
-    { 0x8F, ENTRY_CopyBytes2Mod },                      // POP /0
+    { 0x8F, ENTRY_CopyXop },                            // POP /0 or AMD XOP
     { 0x90, ENTRY_CopyBytes1 },                         // NOP
     { 0x91, ENTRY_CopyBytes1 },                         // XCHG
     { 0x92, ENTRY_CopyBytes1 },                         // XCHG

detours/image.cpp

@@ -9,39 +9,18 @@
 //  Used for for payloads, byways, and imports.
 //
 
-#if _MSC_VER >= 1900
+#if _MSC_VER < 1299
-#pragma warning(push)
-#pragma warning(disable:4091) // empty typedef
-#endif
-#define _CRT_STDIO_ARBITRARY_WIDE_SPECIFIERS 1
-#define _ARM_WINAPI_PARTITION_DESKTOP_SDK_AVAILABLE 1
-#include <windows.h>
-#if _MSC_VER >= 1310
-#pragma warning(push)
-#if _MSC_VER > 1400
-#pragma warning(disable:6102 6103) // /analyze warnings
-#endif
-#include <strsafe.h>
-#pragma warning(pop)
-#endif
-
-#if (_MSC_VER < 1299)
 #pragma warning(disable: 4710)
 #endif
 
 // #define DETOUR_DEBUG 1
 #define DETOURS_INTERNAL
-
 #include "detours.h"
 
 #if DETOURS_VERSION != 0x4c0c1   // 0xMAJORcMINORcPATCH
 #error detours.h version mismatch
 #endif
 
-#if _MSC_VER >= 1900
-#pragma warning(pop)
-#endif
-
 namespace Detour
 {
 //////////////////////////////////////////////////////////////////////////////
@@ -1714,17 +1693,13 @@ BOOL CImage::Write(HANDLE hFile)
         m_nNextFileAddr = Max(m_SectionHeaders[n].PointerToRawData +
                               m_SectionHeaders[n].SizeOfRawData,
                               m_nNextFileAddr);
-#if 0
+        // Old images have VirtualSize == 0 as a matter of course, e.g. NT 3.1.
-        m_nNextVirtAddr = Max(m_SectionHeaders[n].VirtualAddress +
+        // In which case, use SizeOfRawData instead.
-                              m_SectionHeaders[n].Misc.VirtualSize,
-                              m_nNextVirtAddr);
-#else
         m_nNextVirtAddr = Max(m_SectionHeaders[n].VirtualAddress +
                               (m_SectionHeaders[n].Misc.VirtualSize
                                ? m_SectionHeaders[n].Misc.VirtualSize
                                : SectionAlign(m_SectionHeaders[n].SizeOfRawData)),
                               m_nNextVirtAddr);
-#endif
 
         m_nExtraOffset = Max(m_nNextFileAddr, m_nExtraOffset);
 
@@ -1857,7 +1832,7 @@ BOOL CImage::Write(HANDLE hFile)
         for (CImageImportFile *pImportFile = m_pImportFiles;
              pImportFile != NULL; pImportFile = pImportFile->m_pNextFile) {
 
-            ZeroMemory(piidDst, sizeof(piidDst));
+            ZeroMemory(piidDst, sizeof(*piidDst));
             nameTable.Allocate(pImportFile->m_pszName, (DWORD *)&piidDst->Name);
             piidDst->TimeDateStamp = 0;
             piidDst->ForwarderChain = pImportFile->m_nForwarderChain;
@@ -1899,7 +1874,7 @@ BOOL CImage::Write(HANDLE hFile)
             }
             piidDst++;
         }
-        ZeroMemory(piidDst, sizeof(piidDst));
+        ZeroMemory(piidDst, sizeof(*piidDst));
 
         //////////////////////////////////////////////////////////////////////////
         //

detours/modules.cpp

@@ -9,27 +9,6 @@
 //  Module enumeration functions.
 //
 
-#define _CRT_STDIO_ARBITRARY_WIDE_SPECIFIERS 1
-
-#pragma warning(disable:4068) // unknown pragma (suppress)
-
-#if _MSC_VER >= 1900
-#pragma warning(push)
-#pragma warning(disable:4091) // empty typedef
-#endif
-
-#define _ARM_WINAPI_PARTITION_DESKTOP_SDK_AVAILABLE 1
-#include <windows.h>
-#if (_MSC_VER < 1310)
-#else
-#pragma warning(push)
-#if _MSC_VER > 1400
-#pragma warning(disable:6102 6103) // /analyze warnings
-#endif
-#include <strsafe.h>
-#pragma warning(pop)
-#endif
-
 // #define DETOUR_DEBUG 1
 #define DETOURS_INTERNAL
 #include "detours.h"
@@ -38,10 +17,6 @@
 #error detours.h version mismatch
 #endif
 
-#if _MSC_VER >= 1900
-#pragma warning(pop)
-#endif
-
 #define CLR_DIRECTORY OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_COM_DESCRIPTOR]
 #define IAT_DIRECTORY OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_IAT]
 
@@ -164,8 +139,8 @@ PDETOUR_SYM_INFO DetourLoadImageHlp(VOID)
     return pSymInfo;
 }
 
-PVOID WINAPI DetourFindFunction(_In_ PCSTR pszModule,
+PVOID WINAPI DetourFindFunction(_In_ LPCSTR pszModule,
-                                _In_ PCSTR pszFunction)
+                                _In_ LPCSTR pszFunction)
 {
     /////////////////////////////////////////////// First, try GetProcAddress.
     //