Added support for mach_vm_remap() for dual mapping

mach_vm_remap() allows to create a dual mapping without having to use a file descriptor, which has to open a file or shm memory. The problem is that the recent macos version started displaying a popup message when such file is opened and that annoyed a lot of users. Thus, the initial code-path is no longer used, and mach_vm_remap() is used instead. This change only applies for x86 macs. Apple silicon doesn't allow dual mapping and instead uses MAP_JIT.
2025-12-17 12:34:35 +03:00 · 2024-01-25 22:23:13 +01:00
parent bfa0bf690c
commit a63d41e80b
10 changed files with 507 additions and 123 deletions
--- a/.github/workflows/build-config.json
+++ b/.github/workflows/build-config.json
@@ -12,37 +12,14 @@
  ],

  "tests": [
-    {
-      "cmd": ["asmjit_test_unit", "--quick"],
-      "optional": true
-    },
-    {
-      "cmd": ["asmjit_test_assembler"],
-      "optional": true
-    },
-    {
-      "cmd": ["asmjit_test_assembler", "--validate"],
-      "optional": true
-    },
-    {
-      "cmd": ["asmjit_test_emitters"],
-      "optional": true
-    },
-    {
-      "cmd": ["asmjit_test_compiler"],
-      "optional": true
-    },
-    {
-      "cmd": ["asmjit_test_instinfo"],
-      "optional": true
-    },
-    {
-      "cmd": ["asmjit_test_x86_sections"],
-      "optional": true
-    },
-    {
-      "cmd": ["asmjit_test_perf", "--quick"],
-      "optional": true
-    }
+    { "optional": true, "cmd": ["asmjit_test_unit", "--quick"] },
+    { "optional": true, "cmd": ["asmjit_test_assembler"] },
+    { "optional": true, "cmd": ["asmjit_test_assembler", "--validate"] },
+    { "optional": true, "cmd": ["asmjit_test_emitters"] },
+    { "optional": true, "cmd": ["asmjit_test_execute"] },
+    { "optional": true, "cmd": ["asmjit_test_compiler"] },
+    { "optional": true, "cmd": ["asmjit_test_instinfo"] },
+    { "optional": true, "cmd": ["asmjit_test_x86_sections"] },
+    { "optional": true, "cmd": ["asmjit_test_perf", "--quick"] }
  ]
 }
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -14,12 +14,12 @@ jobs:

    steps:
    - name: "Checkout"
-      uses: actions/checkout@v3
+      uses: actions/checkout@v4

    - name: "Setup node.js"
-      uses: actions/setup-node@v3
+      uses: actions/setup-node@v4
      with:
-        node-version: "16"
+        node-version: "*"

    - name: "Check Enumerations"
      run: |
@@ -141,18 +141,18 @@ jobs:

    steps:
      - name: "Checkout"
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
        with:
          path: "source"

      - name: "Checkout Build Actions"
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
        with:
          repository: build-actions/build-actions
          path: "build-actions"

      - name: "Python"
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v5
        with:
          python-version: "3.x"

--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -666,6 +666,7 @@ if (NOT ASMJIT_EMBED)
                      CFLAGS_REL ${ASMJIT_PRIVATE_CFLAGS_REL})

    foreach(_target asmjit_test_emitters
+                    asmjit_test_execute
                    asmjit_test_x86_sections)
      asmjit_add_target(${_target} TEST
                        SOURCES    test/${_target}.cpp
--- a/src/asmjit/core/globals.cpp
+++ b/src/asmjit/core/globals.cpp
@@ -87,6 +87,7 @@ ASMJIT_FAVOR_SIZE const char* DebugUtils::errorAsString(Error err) noexcept {
    "ExpressionOverflow\0"
    "FailedToOpenAnonymousMemory\0"
    "FailedToOpenFile\0"
+    "ProtectionFailure\0"
    "<Unknown>\0";

  static const uint16_t sErrorIndex[] = {
@@ -94,7 +95,7 @@ ASMJIT_FAVOR_SIZE const char* DebugUtils::errorAsString(Error err) noexcept {
    247, 264, 283, 298, 314, 333, 352, 370, 392, 410, 429, 444, 460, 474, 488,
    508, 533, 551, 573, 595, 612, 629, 645, 661, 677, 694, 709, 724, 744, 764,
    784, 817, 837, 852, 869, 888, 909, 929, 943, 964, 978, 996, 1012, 1028, 1047,
-    1073, 1088, 1104, 1119, 1134, 1164, 1188, 1207, 1235, 1252
+    1073, 1088, 1104, 1119, 1134, 1164, 1188, 1207, 1235, 1252, 1270
  };
  // @EnumStringEnd@

--- a/src/asmjit/core/globals.h
+++ b/src/asmjit/core/globals.h
@@ -334,6 +334,10 @@ enum ErrorCode : uint32_t {
  //! \note This is a generic error that is used by internal filesystem API.
  kErrorFailedToOpenFile,

+  //! Protection failure can be returned from a virtual memory allocator or when trying to change memory access
+  //! permissions.
+  kErrorProtectionFailure,
+
  // @EnumValuesEnd@

  //! Count of AsmJit error codes.
--- a/src/asmjit/core/virtmem.cpp
+++ b/src/asmjit/core/virtmem.cpp
@@ -76,8 +76,6 @@
    #define MAP_ANONYMOUS MAP_ANON
  #endif

-  #define ASMJIT_ANONYMOUS_MEMORY_USE_FD
-
  // Android NDK doesn't provide `shm_open()` and `shm_unlink()`.
  #if !defined(__BIONIC__) && !defined(ASMJIT_NO_SHM_OPEN)
    #define ASMJIT_HAS_SHM_OPEN
@@ -89,18 +87,60 @@
    #define ASMJIT_VM_SHM_DETECT 1
  #endif

-  #if defined(__APPLE__) && TARGET_OS_OSX && ASMJIT_ARCH_ARM >= 64
+  #if defined(__APPLE__) && TARGET_OS_OSX
+    #if ASMJIT_ARCH_X86 != 0
+      #define ASMJIT_ANONYMOUS_MEMORY_USE_MACH_VM_REMAP
+    #endif
+    #if ASMJIT_ARCH_ARM >= 64
      #define ASMJIT_HAS_PTHREAD_JIT_WRITE_PROTECT_NP
    #endif
+  #endif
+
+  #if defined(__APPLE__) && ASMJIT_ARCH_X86 == 0
+    #define ASMJIT_NO_DUAL_MAPPING
+  #endif

  #if defined(__NetBSD__) && defined(MAP_REMAPDUP) && defined(PROT_MPROTECT)
-    #undef ASMJIT_ANONYMOUS_MEMORY_USE_FD
    #define ASMJIT_ANONYMOUS_MEMORY_USE_REMAPDUP
  #endif
+
+  #if !defined(ASMJIT_ANONYMOUS_MEMORY_USE_REMAPDUP) && \
+      !defined(ASMJIT_ANONYMOUS_MEMORY_USE_MACH_VM_REMAP) && \
+      !defined(ASMJIT_NO_DUAL_MAPPING)
+    #define ASMJIT_ANONYMOUS_MEMORY_USE_FD
+  #endif
 #endif

 #include <atomic>

+#if defined(ASMJIT_ANONYMOUS_MEMORY_USE_MACH_VM_REMAP)
+#include <mach/mach.h>
+#include <mach/mach_time.h>
+
+extern "C" {
+
+#ifdef mig_external
+mig_external
+#else
+extern
+#endif
+kern_return_t mach_vm_remap(
+  vm_map_t target_task,
+  mach_vm_address_t *target_address,
+  mach_vm_size_t size,
+  mach_vm_offset_t mask,
+  int flags,
+  vm_map_t src_task,
+  mach_vm_address_t src_address,
+  boolean_t copy,
+  vm_prot_t *cur_protection,
+  vm_prot_t *max_protection,
+  vm_inherit_t inheritance
+);
+
+} // {extern "C"}
+#endif
+
 ASMJIT_BEGIN_SUB_NAMESPACE(VirtMem)

 // Virtual Memory Utilities
@@ -757,8 +797,7 @@ static inline int mmMapJitFromMemoryFlags(MemoryFlags memoryFlags) noexcept {
 }

 static inline bool hasDualMappingSupport() noexcept {
-#if defined(__APPLE__) && TARGET_OS_OSX && ASMJIT_ARCH_X86 == 0
-  // Apple platforms don't allow dual-mapping on non-x86 hardware.
+#if defined(ASMJIT_NO_DUAL_MAPPING)
  return false;
 #else
  return true;
@@ -850,6 +889,7 @@ Error protect(void* p, size_t size, MemoryFlags memoryFlags) noexcept {
 // Virtual Memory [Posix] - Dual Mapping
 // =====================================

+#if !defined(ASMJIT_NO_DUAL_MAPPING)
 static Error unmapDualMapping(DualMapping* dm, size_t size) noexcept {
  Error err1 = unmapMemory(dm->rx, size);
  Error err2 = kErrorOk;
@@ -865,6 +905,7 @@ static Error unmapDualMapping(DualMapping* dm, size_t size) noexcept {
  dm->rw = nullptr;
  return kErrorOk;
 }
+#endif // !ASMJIT_NO_DUAL_MAPPING

 #if defined(ASMJIT_ANONYMOUS_MEMORY_USE_REMAPDUP)
 static Error allocDualMappingUsingRemapdup(DualMapping* dmOut, size_t size, MemoryFlags memoryFlags) noexcept {
@@ -897,16 +938,105 @@ static Error allocDualMappingUsingRemapdup(DualMapping* dmOut, size_t size, Memo
 }
 #endif

-Error allocDualMapping(DualMapping* dm, size_t size, MemoryFlags memoryFlags) noexcept {
-  dm->rx = nullptr;
-  dm->rw = nullptr;
+#if defined(ASMJIT_ANONYMOUS_MEMORY_USE_MACH_VM_REMAP)
+static Error asmjitErrorFromKernResult(kern_return_t result) noexcept {
+  switch (result) {
+    case KERN_PROTECTION_FAILURE:
+      return DebugUtils::errored(kErrorProtectionFailure);
+    case KERN_NO_SPACE:
+      return DebugUtils::errored(kErrorOutOfMemory);
+    case KERN_INVALID_ARGUMENT:
+      return DebugUtils::errored(kErrorInvalidArgument);
+    default:
+      return DebugUtils::errored(kErrorInvalidState);
+  }
+}

-  if (off_t(size) <= 0)
-    return DebugUtils::errored(size == 0 ? kErrorInvalidArgument : kErrorTooLarge);
+static Error allocDualMappingUsingMachVmRemap(DualMapping* dmOut, size_t size, MemoryFlags memoryFlags) noexcept {
+  DualMapping dm {};

-#if defined(ASMJIT_ANONYMOUS_MEMORY_USE_REMAPDUP)
-  return allocDualMappingUsingRemapdup(dm, size, memoryFlags);
-#elif defined(ASMJIT_ANONYMOUS_MEMORY_USE_FD)
+  MemoryFlags mmapFlags = MemoryFlags::kAccessReadWrite | (memoryFlags & MemoryFlags::kMapShared);
+  ASMJIT_PROPAGATE(mapMemory(&dm.rx, size, mmapFlags));
+
+  vm_prot_t curProt;
+  vm_prot_t maxProt;
+
+  int rwProtectFlags = VM_PROT_READ | VM_PROT_WRITE;
+  int rxProtectFlags = VM_PROT_READ;
+
+  if (Support::test(memoryFlags, MemoryFlags::kAccessExecute))
+    rxProtectFlags |= VM_PROT_EXECUTE;
+
+  kern_return_t result {};
+  do {
+    vm_map_t task = mach_task_self();
+    mach_vm_address_t remappedAddr {};
+
+#if defined(VM_FLAGS_RANDOM_ADDR)
+    int remapFlags = VM_FLAGS_ANYWHERE | VM_FLAGS_RANDOM_ADDR;
+#else
+    int remapFlags = VM_FLAGS_ANYWHERE;
+#endif
+
+    // Try to remap the existing memory into a different address.
+    result = mach_vm_remap(
+      task,                       // target_task
+      &remappedAddr,              // target_address
+      size,                       // size
+      0,                          // mask
+      remapFlags,                 // flags
+      task,                       // src_task
+      (mach_vm_address_t)dm.rx,   // src_address
+      false,                      // copy
+      &curProt,                   // cur_protection
+      &maxProt,                   // max_protection
+      VM_INHERIT_DEFAULT);        // inheritance
+
+    if (result != KERN_SUCCESS)
+      break;
+
+    dm.rw = (void*)remappedAddr;
+
+    // Now, try to change permissions of both map regions into RW and RX. The vm_protect()
+    // API is used twice as we also want to set maximum permissions, so nobody would be
+    // allowed to change the RX region back to RW or RWX (if RWX is allowed).
+    uint32_t i;
+    for (i = 0; i < 2; i++) {
+      bool setMaximum = (i == 0);
+
+      result = vm_protect(
+        task,                       // target_task
+        (vm_address_t)dm.rx,        // address
+        size,                       // size
+        setMaximum,                 // set_maximum
+        rxProtectFlags);            // new_protection
+
+      if (result != KERN_SUCCESS)
+        break;
+
+      result = vm_protect(task,     // target_task
+        (vm_address_t)dm.rw,        // address
+        size,                       // size
+        setMaximum,                 // set_maximum
+        rwProtectFlags);            // new_protection
+
+      if (result != KERN_SUCCESS)
+        break;
+    }
+  } while (0);
+
+  if (result != KERN_SUCCESS) {
+    unmapDualMapping(&dm, size);
+    return DebugUtils::errored(asmjitErrorFromKernResult(result));
+  }
+
+  *dmOut = dm;
+  return kErrorOk;
+}
+#endif // ASMJIT_ANONYMOUS_MEMORY_USE_MACH_VM_REMAP
+
+#if defined(ASMJIT_ANONYMOUS_MEMORY_USE_FD)
+static Error allocDualMappingUsingFile(DualMapping* dm, size_t size, MemoryFlags memoryFlags) noexcept {
  bool preferTmpOverDevShm = Support::test(memoryFlags, MemoryFlags::kMappingPreferTmp);
  if (!preferTmpOverDevShm) {
    AnonymousMemoryStrategy strategy;
@@ -932,13 +1062,39 @@ Error allocDualMapping(DualMapping* dm, size_t size, MemoryFlags memoryFlags) no
  dm->rx = ptr[0];
  dm->rw = ptr[1];
  return kErrorOk;
+}
+#endif // ASMJIT_ANONYMOUS_MEMORY_USE_FD
+
+Error allocDualMapping(DualMapping* dm, size_t size, MemoryFlags memoryFlags) noexcept {
+  dm->rx = nullptr;
+  dm->rw = nullptr;
+
+#if defined(ASMJIT_NO_DUAL_MAPPING)
+  DebugUtils::unused(size, memoryFlags);
+  return DebugUtils::errored(kErrorFeatureNotEnabled);
 #else
-  #error "[asmjit] VirtMem::allocDualMapping() doesn't have implementation for the target OS and compiler"
+  if (off_t(size) <= 0)
+    return DebugUtils::errored(size == 0 ? kErrorInvalidArgument : kErrorTooLarge);
+
+#if defined(ASMJIT_ANONYMOUS_MEMORY_USE_REMAPDUP)
+  return allocDualMappingUsingRemapdup(dm, size, memoryFlags);
+#elif defined(ASMJIT_ANONYMOUS_MEMORY_USE_MACH_VM_REMAP)
+  return allocDualMappingUsingMachVmRemap(dm, size, memoryFlags);
+#elif defined(ASMJIT_ANONYMOUS_MEMORY_USE_FD)
+  return allocDualMappingUsingFile(dm, size, memoryFlags);
+#else
+  #error "[asmjit] VirtMem::allocDualMapping() doesn't have implementation for the target OS or architecture"
 #endif
+#endif // ASMJIT_NO_DUAL_MAPPING
 }

 Error releaseDualMapping(DualMapping* dm, size_t size) noexcept {
+#if defined(ASMJIT_NO_DUAL_MAPPING)
+  DebugUtils::unused(dm, size);
+  return DebugUtils::errored(kErrorFeatureNotEnabled);
+#else
  return unmapDualMapping(dm, size);
+#endif // ASMJIT_NO_DUAL_MAPPING
 }
 #endif

@@ -1041,6 +1197,7 @@ UNIT(virt_mem) {
  INFO("  flags:");
  INFO("    kEnabled: %s"    , Support::test(hardenedFlags, VirtMem::HardenedRuntimeFlags::kEnabled    ) ? "true" : "false");
  INFO("    kMapJit: %s"     , Support::test(hardenedFlags, VirtMem::HardenedRuntimeFlags::kMapJit     ) ? "true" : "false");
+  INFO("    kDualMapping: %s", Support::test(hardenedFlags, VirtMem::HardenedRuntimeFlags::kDualMapping) ? "true" : "false");
 }

 ASMJIT_END_NAMESPACE
--- a/src/asmjit/core/virtmem.h
+++ b/src/asmjit/core/virtmem.h
@@ -10,6 +10,7 @@
 #ifndef ASMJIT_NO_JIT

 #include "../core/globals.h"
+#include "../core/support.h"

 ASMJIT_BEGIN_NAMESPACE

@@ -218,15 +219,28 @@ enum class HardenedRuntimeFlags : uint32_t {
  //! Read+Write+Execute can only be allocated with MAP_JIT flag (Apple specific, only available on Apple platforms).
  kMapJit = 0x00000002u,

-  //! Read+Write+Executa can be allocated with dual mapping approach (one region with RW and the other with RX).
+  //! Read+Write+Execute can be allocated with dual mapping approach (one region with RW and the other with RX).
  kDualMapping = 0x00000004u
 };
 ASMJIT_DEFINE_ENUM_FLAGS(HardenedRuntimeFlags)

 //! Hardened runtime information.
 struct HardenedRuntimeInfo {
+  //! \name Members
+  //! \{
+
  //! Hardened runtime flags.
  HardenedRuntimeFlags flags;
+
+  //! \}
+
+  //! \name Accessors
+  //! \{
+
+  //! Tests whether the hardened runtime `flag` is set.
+  ASMJIT_INLINE_NODEBUG bool hasFlag(HardenedRuntimeFlags flag) const noexcept { return Support::test(flags, flag); }
+
+  //! \}
 };

 //! Returns runtime features provided by the OS.
--- a/src/asmjit/x86/x86emitter.h
+++ b/src/asmjit/x86/x86emitter.h
@@ -2827,7 +2827,7 @@ public:
  ASMJIT_INST_3x(vpand, Vpand, Vec, Vec, Mem)                          // AVX+
  ASMJIT_INST_3x(vpandd, Vpandd, Vec, Vec, Vec)                        //      AVX512_F{kz|b32}
  ASMJIT_INST_3x(vpandd, Vpandd, Vec, Vec, Mem)                        //      AVX512_F{kz|b32}
-  ASMJIT_INST_3x(vpandn, Vpandn, Vec, Vec, Vec)                        // AV+
+  ASMJIT_INST_3x(vpandn, Vpandn, Vec, Vec, Vec)                        // AVX+
  ASMJIT_INST_3x(vpandn, Vpandn, Vec, Vec, Mem)                        // AVX+
  ASMJIT_INST_3x(vpandnd, Vpandnd, Vec, Vec, Vec)                      //      AVX512_F{kz|b32}
  ASMJIT_INST_3x(vpandnd, Vpandnd, Vec, Vec, Mem)                      //      AVX512_F{kz|b32}
@@ -3186,7 +3186,7 @@ public:
  ASMJIT_INST_2x(vpopcntq, Vpopcntq, Vec, Mem)                         //      AVX512_VPOPCNTDQ{kz|b64}
  ASMJIT_INST_2x(vpopcntw, Vpopcntw, Vec, Vec)                         //      AVX512_BITALG{kz|b32}
  ASMJIT_INST_2x(vpopcntw, Vpopcntw, Vec, Mem)                         //      AVX512_BITALG{kz|b32}
-  ASMJIT_INST_3x(vpor, Vpor, Vec, Vec, Vec)                            // AV+
+  ASMJIT_INST_3x(vpor, Vpor, Vec, Vec, Vec)                            // AVX+
  ASMJIT_INST_3x(vpor, Vpor, Vec, Vec, Mem)                            // AVX+
  ASMJIT_INST_3x(vpord, Vpord, Vec, Vec, Vec)                          //      AVX512_F{kz|b32}
  ASMJIT_INST_3x(vpord, Vpord, Vec, Vec, Mem)                          //      AVX512_F{kz|b32}
--- a/test/asmjit_test_emitters.cpp
+++ b/test/asmjit_test_emitters.cpp
@@ -5,8 +5,24 @@

 #include <asmjit/core.h>

-#if ASMJIT_ARCH_X86  && !defined(ASMJIT_NO_X86) && !defined(ASMJIT_NO_JIT)
+static void printInfo() noexcept {
+  printf("AsmJit Emitters Test-Suite v%u.%u.%u\n",
+    unsigned((ASMJIT_LIBRARY_VERSION >> 16)       ),
+    unsigned((ASMJIT_LIBRARY_VERSION >>  8) & 0xFF),
+    unsigned((ASMJIT_LIBRARY_VERSION      ) & 0xFF));
+}
+
+#if !defined(ASMJIT_NO_JIT) && ( \
+    (ASMJIT_ARCH_X86 != 0  && !defined(ASMJIT_NO_X86    )) || \
+    (ASMJIT_ARCH_ARM == 64 && !defined(ASMJIT_NO_AARCH64)) )
+
+#if ASMJIT_ARCH_X86 != 0
 #include <asmjit/x86.h>
+#endif
+
+#if ASMJIT_ARCH_ARM == 64
+#include <asmjit/a64.h>
+#endif

 #include <stdio.h>
 #include <stdlib.h>
@@ -17,9 +33,13 @@ using namespace asmjit;
 // Signature of the generated function.
 typedef void (*SumIntsFunc)(int* dst, const int* a, const int* b);

+// X86 Backend
+// -----------
+
+#if ASMJIT_ARCH_X86 != 0
 // This function works with both x86::Assembler and x86::Builder. It shows how
 // `x86::Emitter` can be used to make your code more generic.
-static void makeRawFunc(x86::Emitter* emitter) noexcept {
+static void generateFuncWithEmitter(x86::Emitter* emitter) noexcept {
  // Decide which registers will be mapped to function arguments. Try changing
  // registers of `dst`, `src_a`, and `src_b` and see what happens in function's
  // prolog and epilog.
@@ -39,8 +59,8 @@ static void makeRawFunc(x86::Emitter* emitter) noexcept {
  FuncFrame frame;
  frame.init(func);

-  // Make XMM0 and XMM1 dirty. VEC group includes XMM|YMM|ZMM registers.
-  frame.addDirtyRegs(x86::xmm0, x86::xmm1);
+  // Make or registers dirty.
+  frame.addDirtyRegs(vec0, vec1);

  FuncArgsAssignment args(&func);         // Create arguments assignment context.
  args.assignAll(dst, src_a, src_b);      // Assign our registers to arguments.
@@ -63,7 +83,7 @@ static void makeRawFunc(x86::Emitter* emitter) noexcept {

 #ifndef ASMJIT_NO_COMPILER
 // This function works with x86::Compiler, provided for comparison.
-static void makeCompiledFunc(x86::Compiler* cc) noexcept {
+static void generateFuncWithCompiler(x86::Compiler* cc) noexcept {
  x86::Gp dst = cc->newIntPtr("dst");
  x86::Gp src_a = cc->newIntPtr("src_a");
  x86::Gp src_b = cc->newIntPtr("src_b");
@@ -83,6 +103,154 @@ static void makeCompiledFunc(x86::Compiler* cc) noexcept {
 }
 #endif

+static Error generateFunc(CodeHolder& code, EmitterType emitterType) noexcept {
+  switch (emitterType) {
+    case EmitterType::kAssembler: {
+      printf("Using x86::Assembler:\n");
+      x86::Assembler a(&code);
+      generateFuncWithEmitter(a.as<x86::Emitter>());
+      return kErrorOk;
+    }
+
+#ifndef ASMJIT_NO_BUILDER
+    case EmitterType::kBuilder: {
+      printf("Using x86::Builder:\n");
+      x86::Builder cb(&code);
+      generateFuncWithEmitter(cb.as<x86::Emitter>());
+
+      return cb.finalize();
+    }
+#endif
+
+#ifndef ASMJIT_NO_COMPILER
+    case EmitterType::kCompiler: {
+      printf("Using x86::Compiler:\n");
+      x86::Compiler cc(&code);
+      generateFuncWithCompiler(&cc);
+
+      return cc.finalize();
+    }
+#endif
+
+    default: {
+      printf("** FAILURE: No emitter to use **\n");
+      exit(1);
+    }
+  }
+}
+#endif
+
+// AArch64 Backend
+// ---------------
+
+#if ASMJIT_ARCH_ARM == 64
+// This function works with both a64::Assembler and a64::Builder. It shows how
+// `a64::Emitter` can be used to make your code more generic.
+static void generateFuncWithEmitter(a64::Emitter* emitter) noexcept {
+  // Decide which registers will be mapped to function arguments. Try changing
+  // registers of `dst`, `src_a`, and `src_b` and see what happens in function's
+  // prolog and epilog.
+  a64::Gp dst   = a64::x0;
+  a64::Gp src_a = a64::x1;
+  a64::Gp src_b = a64::x2;
+
+  // Decide which vector registers to use. We use these to keep the code generic,
+  // you can switch to any other registers when needed.
+  a64::Vec vec0 = a64::v0;
+  a64::Vec vec1 = a64::v1;
+  a64::Vec vec2 = a64::v2;
+
+  // Create and initialize `FuncDetail` and `FuncFrame`.
+  FuncDetail func;
+  func.init(FuncSignature::build<void, int*, const int*, const int*>(), emitter->environment());
+
+  FuncFrame frame;
+  frame.init(func);
+
+  // Make XMM0 and XMM1 dirty. VEC group includes XMM|YMM|ZMM registers.
+  frame.addDirtyRegs(vec0, vec1, vec2);
+
+  FuncArgsAssignment args(&func);                // Create arguments assignment context.
+  args.assignAll(dst, src_a, src_b);             // Assign our registers to arguments.
+  args.updateFuncFrame(frame);                   // Reflect our args in FuncFrame.
+  frame.finalize();
+
+  // Emit prolog and allocate arguments to registers.
+  emitter->emitProlog(frame);
+  emitter->emitArgsAssignment(frame, args);
+
+  emitter->ld1(vec0.b16(), a64::ptr(src_a));     // Load 4 ints from [src_a] to vec0.
+  emitter->ld1(vec1.b16(), a64::ptr(src_b));     // Load 4 ints from [src_b] to vec1.
+  emitter->add(vec2.s4(), vec0.s4(), vec1.s4()); // Add 4 ints of vec0 and vec1 and store to vec2.
+  emitter->st1(vec2.b16(), a64::ptr(dst));       // Store the result (vec2) to [dst].
+
+  // Emit epilog and return.
+  emitter->emitEpilog(frame);
+}
+
+#ifndef ASMJIT_NO_COMPILER
+// This function works with x86::Compiler, provided for comparison.
+static void generateFuncWithCompiler(a64::Compiler* cc) noexcept {
+  a64::Gp dst = cc->newIntPtr("dst");
+  a64::Gp src_a = cc->newIntPtr("src_a");
+  a64::Gp src_b = cc->newIntPtr("src_b");
+  a64::Vec vec0 = cc->newVecQ("vec0");
+  a64::Vec vec1 = cc->newVecQ("vec1");
+  a64::Vec vec2 = cc->newVecQ("vec2");
+
+  FuncNode* funcNode = cc->addFunc(FuncSignature::build<void, int*, const int*, const int*>());
+  funcNode->setArg(0, dst);
+  funcNode->setArg(1, src_a);
+  funcNode->setArg(2, src_b);
+
+  cc->ld1(vec0.b16(), a64::ptr(src_a));          // Load 4 ints from [src_a] to vec0.
+  cc->ld1(vec1.b16(), a64::ptr(src_b));          // Load 4 ints from [src_b] to vec1.
+  cc->add(vec2.s4(), vec0.s4(), vec1.s4());      // Add 4 ints of vec0 and vec1 and store to vec2.
+  cc->st1(vec2.b16(), a64::ptr(dst));            // Store the result (vec2) to [dst].
+  cc->endFunc();
+}
+#endif
+
+static Error generateFunc(CodeHolder& code, EmitterType emitterType) noexcept {
+  switch (emitterType) {
+    case EmitterType::kAssembler: {
+      printf("Using a64::Assembler:\n");
+      a64::Assembler a(&code);
+      generateFuncWithEmitter(a.as<a64::Emitter>());
+      return kErrorOk;
+    }
+
+#ifndef ASMJIT_NO_BUILDER
+    case EmitterType::kBuilder: {
+      printf("Using a64::Builder:\n");
+      a64::Builder cb(&code);
+      generateFuncWithEmitter(cb.as<a64::Emitter>());
+
+      return cb.finalize();
+    }
+#endif
+
+#ifndef ASMJIT_NO_COMPILER
+    case EmitterType::kCompiler: {
+      printf("Using a64::Compiler:\n");
+      a64::Compiler cc(&code);
+      generateFuncWithCompiler(&cc);
+
+      return cc.finalize();
+    }
+#endif
+
+    default: {
+      printf("** FAILURE: No emitter to use **\n");
+      exit(1);
+    }
+  }
+}
+#endif
+
+// Testing
+// -------
+
 static uint32_t testFunc(JitRuntime& rt, EmitterType emitterType) noexcept {
 #ifndef ASMJIT_NO_LOGGING
  FileLogger logger(stdout);
@@ -96,57 +264,18 @@ static uint32_t testFunc(JitRuntime& rt, EmitterType emitterType) noexcept {
  code.setLogger(&logger);
 #endif

-  Error err = kErrorOk;
-  switch (emitterType) {
-    case EmitterType::kAssembler: {
-      printf("Using x86::Assembler:\n");
-      x86::Assembler a(&code);
-      makeRawFunc(a.as<x86::Emitter>());
-      break;
-    }
-
-#ifndef ASMJIT_NO_BUILDER
-    case EmitterType::kBuilder: {
-      printf("Using x86::Builder:\n");
-      x86::Builder cb(&code);
-      makeRawFunc(cb.as<x86::Emitter>());
-
-      err = cb.finalize();
+  Error err = generateFunc(code, emitterType);
  if (err) {
-        printf("** FAILURE: x86::Builder::finalize() failed (%s) **\n", DebugUtils::errorAsString(err));
+    printf("** FAILURE: Failed to generate a function: %s **\n", DebugUtils::errorAsString(err));
    return 1;
  }
-      break;
-    }
-#endif
-
-#ifndef ASMJIT_NO_COMPILER
-    case EmitterType::kCompiler: {
-      printf("Using x86::Compiler:\n");
-      x86::Compiler cc(&code);
-      makeCompiledFunc(&cc);
-
-      err = cc.finalize();
-      if (err) {
-        printf("** FAILURE: x86::Compiler::finalize() failed (%s) **\n", DebugUtils::errorAsString(err));
-        return 1;
-      }
-      break;
-    }
-#endif
-
-    default: {
-      printf("** FAILURE: No emitter to use **\n");
-      return 1;
-    }
-  }

  // Add the code generated to the runtime.
  SumIntsFunc fn;
  err = rt.add(&fn, &code);

  if (err) {
-    printf("** FAILURE: JitRuntime::add() failed (%s) **\n", DebugUtils::errorAsString(err));
+    printf("** FAILURE: JitRuntime::add() failed: %s **\n", DebugUtils::errorAsString(err));
    return 1;
  }

@@ -160,27 +289,24 @@ static uint32_t testFunc(JitRuntime& rt, EmitterType emitterType) noexcept {
  printf("Result = { %d %d %d %d }\n\n", out[0], out[1], out[2], out[3]);

  rt.release(fn);
-  return !(out[0] == 5 && out[1] == 8 && out[2] == 4 && out[3] == 9);
+  return out[0] == 5 && out[1] == 8 && out[2] == 4 && out[3] == 9;
 }

 int main() {
-  printf("AsmJit Emitters Test-Suite v%u.%u.%u\n",
-    unsigned((ASMJIT_LIBRARY_VERSION >> 16)       ),
-    unsigned((ASMJIT_LIBRARY_VERSION >>  8) & 0xFF),
-    unsigned((ASMJIT_LIBRARY_VERSION      ) & 0xFF));
+  printInfo();
  printf("\n");

  JitRuntime rt;
  unsigned nFailed = 0;

-  nFailed += testFunc(rt, EmitterType::kAssembler);
+  nFailed += !testFunc(rt, EmitterType::kAssembler);

 #ifndef ASMJIT_NO_BUILDER
-  nFailed += testFunc(rt, EmitterType::kBuilder);
+  nFailed += !testFunc(rt, EmitterType::kBuilder);
 #endif

 #ifndef ASMJIT_NO_COMPILER
-  nFailed += testFunc(rt, EmitterType::kCompiler);
+  nFailed += !testFunc(rt, EmitterType::kCompiler);
 #endif

  if (!nFailed)
@@ -192,7 +318,8 @@ int main() {
 }
 #else
 int main() {
-  printf("AsmJit X86 Emitter Test is disabled on non-x86 hosts or when compiled with ASMJIT_NO_JIT option\n\n");
+  printInfo();
+  printf("\nThis test is currently disabled - no JIT or no support for the target architecture\n");
  return 0;
 }
 #endif // ASMJIT_ARCH_X86 && !ASMJIT_NO_X86 && !ASMJIT_NO_JIT
--- a/test/asmjit_test_execute.cpp
+++ b/test/asmjit_test_execute.cpp
@@ -0,0 +1,103 @@
+// This file is part of AsmJit project <https://asmjit.com>
+//
+// See asmjit.h or LICENSE.md for license and copyright information
+// SPDX-License-Identifier: Zlib
+
+#include <asmjit/core.h>
+
+static void printInfo() noexcept {
+  printf("AsmJit Execute Test-Suite v%u.%u.%u\n",
+    unsigned((ASMJIT_LIBRARY_VERSION >> 16)       ),
+    unsigned((ASMJIT_LIBRARY_VERSION >>  8) & 0xFF),
+    unsigned((ASMJIT_LIBRARY_VERSION      ) & 0xFF));
+}
+
+#if !defined(ASMJIT_NO_JIT) && ( \
+    (ASMJIT_ARCH_X86 != 0  && !defined(ASMJIT_NO_X86    )) || \
+    (ASMJIT_ARCH_ARM == 64 && !defined(ASMJIT_NO_AARCH64)) )
+
+#if ASMJIT_ARCH_X86 != 0
+#include <asmjit/x86.h>
+#endif
+
+#if ASMJIT_ARCH_ARM == 64
+#include <asmjit/a64.h>
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+using namespace asmjit;
+
+// Signature of the generated function.
+typedef void (*EmptyFunc)(void);
+
+// Generate Empty Function
+// -----------------------
+
+#if ASMJIT_ARCH_X86 != 0
+static void generateEmptyFunc(CodeHolder& code) noexcept {
+  x86::Assembler a(&code);
+  a.ret();
+}
+#endif
+
+#if ASMJIT_ARCH_ARM == 64
+static void generateEmptyFunc(CodeHolder& code) noexcept {
+  a64::Assembler a(&code);
+  a.ret(a64::x30);
+}
+#endif
+
+// Testing
+// -------
+
+static void executeEmptyFunc(JitRuntime& rt) noexcept {
+  CodeHolder code;
+  code.init(rt.environment(), rt.cpuFeatures());
+
+  EmptyFunc fn;
+
+  generateEmptyFunc(code);
+  Error err = rt.add(&fn, &code);
+
+  if (err) {
+    printf("** FAILURE: JitRuntime::add() failed: %s **\n", DebugUtils::errorAsString(err));
+    exit(1);
+  }
+
+  fn();
+
+  rt.release(&fn);
+}
+
+int main() {
+  printInfo();
+  printf("\n");
+
+  {
+    printf("Trying to execute empty function with JitRuntime (default settings)\n");
+    JitRuntime rt;
+    executeEmptyFunc(rt);
+  }
+
+  if (VirtMem::hardenedRuntimeInfo().hasFlag(VirtMem::HardenedRuntimeFlags::kDualMapping)) {
+    printf("Trying to execute empty function with JitRuntime (dual-mapped)\n");
+    JitAllocator::CreateParams params {};
+    params.options |= JitAllocatorOptions::kUseDualMapping;
+    JitRuntime rt;
+    executeEmptyFunc(rt);
+  }
+
+  // If we are here we were successful, otherwise the process would crash.
+  printf("** SUCCESS **\n");
+  return 0;
+}
+#else
+int main() {
+  printInfo();
+  printf("\nThis test is currently disabled - no JIT or no support for the target architecture\n");
+  return 0;
+}
+#endif // ASMJIT_ARCH_X86 && !ASMJIT_NO_X86 && !ASMJIT_NO_JIT