[Memory/CPU] UWP: Support separate code execution and write memory, FromApp functions + other Windows memory fixes

src/xenia/base/mapped_memory_win.cc

@@ -29,7 +29,7 @@ class Win32MappedMemory : public MappedMemory {
     if (data_) {
       UnmapViewOfFile(data_);
     }
-    if (mapping_handle != INVALID_HANDLE_VALUE) {
+    if (mapping_handle) {
       CloseHandle(mapping_handle);
     }
     if (file_handle != INVALID_HANDLE_VALUE) {
@@ -42,9 +42,9 @@ class Win32MappedMemory : public MappedMemory {
       UnmapViewOfFile(data_);
       data_ = nullptr;
     }
-    if (mapping_handle != INVALID_HANDLE_VALUE) {
+    if (mapping_handle) {
       CloseHandle(mapping_handle);
-      mapping_handle = INVALID_HANDLE_VALUE;
+      mapping_handle = nullptr;
     }
     if (file_handle != INVALID_HANDLE_VALUE) {
       if (truncate_size) {
@@ -65,8 +65,13 @@ class Win32MappedMemory : public MappedMemory {
     size_t aligned_length = length + (offset - aligned_offset);
 
     UnmapViewOfFile(data_);
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
     data_ = MapViewOfFile(mapping_handle, view_access_, aligned_offset >> 32,
                           aligned_offset & 0xFFFFFFFF, aligned_length);
+#else
+    data_ = MapViewOfFileFromApp(mapping_handle, ULONG(view_access_),
+                                 ULONG64(aligned_offset), aligned_length);
+#endif
     if (!data_) {
       return false;
     }
@@ -84,7 +89,7 @@ class Win32MappedMemory : public MappedMemory {
   }
 
   HANDLE file_handle = INVALID_HANDLE_VALUE;
-  HANDLE mapping_handle = INVALID_HANDLE_VALUE;
+  HANDLE mapping_handle = nullptr;
   DWORD view_access_ = 0;
 };
 
@@ -129,16 +134,28 @@ std::unique_ptr<MappedMemory> MappedMemory::Open(
     return nullptr;
   }
 
-  mm->mapping_handle = CreateFileMapping(mm->file_handle, nullptr,
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
-                                         mapping_protect, aligned_length >> 32,
+  mm->mapping_handle = CreateFileMapping(
-                                         aligned_length & 0xFFFFFFFF, nullptr);
+      mm->file_handle, nullptr, mapping_protect, DWORD(aligned_length >> 32),
-  if (mm->mapping_handle == INVALID_HANDLE_VALUE) {
+      DWORD(aligned_length), nullptr);
+#else
+  mm->mapping_handle =
+      CreateFileMappingFromApp(mm->file_handle, nullptr, ULONG(mapping_protect),
+                               ULONG64(aligned_length), nullptr);
+#endif
+  if (!mm->mapping_handle) {
     return nullptr;
   }
 
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
   mm->data_ = reinterpret_cast<uint8_t*>(MapViewOfFile(
-      mm->mapping_handle, view_access, static_cast<DWORD>(aligned_offset >> 32),
+      mm->mapping_handle, view_access, DWORD(aligned_offset >> 32),
-      static_cast<DWORD>(aligned_offset & 0xFFFFFFFF), aligned_length));
+      DWORD(aligned_offset), aligned_length));
+#else
+  mm->data_ = reinterpret_cast<uint8_t*>(
+      MapViewOfFileFromApp(mm->mapping_handle, ULONG(view_access),
+                           ULONG64(aligned_offset), aligned_length));
+#endif
   if (!mm->data_) {
     return nullptr;
   }
@@ -203,8 +220,8 @@ class Win32ChunkedMappedMemoryWriter : public ChunkedMappedMemoryWriter {
   class Chunk {
    public:
     explicit Chunk(size_t capacity)
-        : file_handle_(0),
+        : file_handle_(INVALID_HANDLE_VALUE),
-          mapping_handle_(0),
+          mapping_handle_(nullptr),
           data_(nullptr),
           offset_(0),
           capacity_(capacity),
@@ -217,7 +234,7 @@ class Win32ChunkedMappedMemoryWriter : public ChunkedMappedMemoryWriter {
       if (mapping_handle_) {
         CloseHandle(mapping_handle_);
       }
-      if (file_handle_) {
+      if (file_handle_ != INVALID_HANDLE_VALUE) {
         CloseHandle(file_handle_);
       }
     }
@@ -231,13 +248,19 @@ class Win32ChunkedMappedMemoryWriter : public ChunkedMappedMemoryWriter {
 
       file_handle_ = CreateFile(path.c_str(), file_access, file_share, nullptr,
                                 create_mode, FILE_ATTRIBUTE_NORMAL, nullptr);
-      if (!file_handle_) {
+      if (file_handle_ == INVALID_HANDLE_VALUE) {
         return false;
       }
 
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
       mapping_handle_ =
-          CreateFileMapping(file_handle_, nullptr, mapping_protect, 0,
+          CreateFileMapping(file_handle_, nullptr, mapping_protect,
-                            static_cast<DWORD>(capacity_), nullptr);
+                            DWORD(capacity_ >> 32), DWORD(capacity_), nullptr);
+#else
+      mapping_handle_ = CreateFileMappingFromApp(file_handle_, nullptr,
+                                                 ULONG(mapping_protect),
+                                                 ULONG64(capacity_), nullptr);
+#endif
       if (!mapping_handle_) {
         return false;
       }
@@ -247,10 +270,32 @@ class Win32ChunkedMappedMemoryWriter : public ChunkedMappedMemoryWriter {
       if (low_address_space) {
         bool successful = false;
         data_ = reinterpret_cast<uint8_t*>(0x10000000);
+#if !WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
+        HANDLE process = GetCurrentProcess();
+#endif
         for (int i = 0; i < 1000; ++i) {
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
           if (MapViewOfFileEx(mapping_handle_, view_access, 0, 0, capacity_,
                               data_)) {
             successful = true;
+          }
+#else
+          // VirtualAlloc2FromApp and MapViewOfFile3FromApp were added in
+          // 10.0.17134.0.
+          // https://docs.microsoft.com/en-us/uwp/win32-and-com/win32-apis
+          if (VirtualAlloc2FromApp(process, data_, capacity_,
+                                   MEM_RESERVE | MEM_RESERVE_PLACEHOLDER,
+                                   PAGE_NOACCESS, nullptr, 0)) {
+            if (MapViewOfFile3FromApp(mapping_handle_, process, data_, 0,
+                                      capacity_, MEM_REPLACE_PLACEHOLDER,
+                                      ULONG(mapping_protect), nullptr, 0)) {
+              successful = true;
+            } else {
+              VirtualFree(data_, capacity_, MEM_RELEASE);
+            }
+          }
+#endif
+          if (successful) {
             break;
           }
           data_ += capacity_;
@@ -261,8 +306,13 @@ class Win32ChunkedMappedMemoryWriter : public ChunkedMappedMemoryWriter {
           }
         }
       } else {
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
         data_ = reinterpret_cast<uint8_t*>(
             MapViewOfFile(mapping_handle_, view_access, 0, 0, capacity_));
+#else
+        data_ = reinterpret_cast<uint8_t*>(MapViewOfFileFromApp(
+            mapping_handle_, ULONG(view_access), 0, capacity_));
+#endif
       }
       if (!data_) {
         return false;

src/xenia/base/memory.cc

@@ -8,11 +8,26 @@
  */
 
 #include "xenia/base/memory.h"
+#include "xenia/base/cvar.h"
 #include "xenia/base/platform.h"
 
 #include <algorithm>
 
+DEFINE_bool(
+    writable_executable_memory, true,
+    "Allow mapping memory with both write and execute access, for simulating "
+    "behavior on platforms where that's not supported",
+    "Memory");
+
 namespace xe {
+namespace memory {
+
+bool IsWritableExecutableMemoryPreferred() {
+  return IsWritableExecutableMemorySupported() &&
+         cvars::writable_executable_memory;
+}
+
+}  // namespace memory
 
 // TODO(benvanik): fancy AVX versions.
 // https://github.com/gnuradio/volk/blob/master/kernels/volk/volk_16u_byteswap.h

src/xenia/base/memory.h

@@ -35,6 +35,7 @@ enum class PageAccess {
   kNoAccess = 0,
   kReadOnly = 1 << 0,
   kReadWrite = kReadOnly | 1 << 1,
+  kExecuteReadOnly = kReadOnly | 1 << 2,
   kExecuteReadWrite = kReadWrite | 1 << 2,
 };
 
@@ -49,6 +50,16 @@ enum class DeallocationType {
   kDecommit = 1 << 1,
 };
 
+// Whether the host allows the pages to be allocated or mapped with
+// PageAccess::kExecuteReadWrite - if not, separate mappings backed by the same
+// memory-mapped file must be used to write to executable pages.
+bool IsWritableExecutableMemorySupported();
+
+// Whether PageAccess::kExecuteReadWrite is a supported and preferred way of
+// writing executable memory, useful for simulating how Xenia would work without
+// writable executable memory on a system with it.
+bool IsWritableExecutableMemoryPreferred();
+
 // Allocates a block of memory at the given page-aligned base address.
 // Fails if the memory is not available.
 // Specify nullptr for base_address to leave it up to the system.

src/xenia/base/memory_posix.cc

@@ -39,6 +39,8 @@ uint32_t ToPosixProtectFlags(PageAccess access) {
       return PROT_READ;
     case PageAccess::kReadWrite:
       return PROT_READ | PROT_WRITE;
+    case PageAccess::kExecuteReadOnly:
+      return PROT_READ | PROT_EXEC;
     case PageAccess::kExecuteReadWrite:
       return PROT_READ | PROT_WRITE | PROT_EXEC;
     default:
@@ -47,6 +49,8 @@ uint32_t ToPosixProtectFlags(PageAccess access) {
   }
 }
 
+bool IsWritableExecutableMemorySupported() { return true; }
+
 void* AllocFixed(void* base_address, size_t length,
                  AllocationType allocation_type, PageAccess access) {
   // mmap does not support reserve / commit, so ignore allocation_type.
@@ -112,6 +116,7 @@ FileMappingHandle CreateFileMappingHandle(const std::filesystem::path& path,
       oflag = 0;
       break;
     case PageAccess::kReadOnly:
+    case PageAccess::kExecuteReadOnly:
       oflag = O_RDONLY;
       break;
     case PageAccess::kReadWrite:

src/xenia/base/memory_win.cc

@@ -42,6 +42,8 @@ DWORD ToWin32ProtectFlags(PageAccess access) {
       return PAGE_READONLY;
     case PageAccess::kReadWrite:
       return PAGE_READWRITE;
+    case PageAccess::kExecuteReadOnly:
+      return PAGE_EXECUTE_READ;
     case PageAccess::kExecuteReadWrite:
       return PAGE_EXECUTE_READWRITE;
     default:
@@ -63,6 +65,8 @@ PageAccess ToXeniaProtectFlags(DWORD access) {
       return PageAccess::kReadOnly;
     case PAGE_READWRITE:
       return PageAccess::kReadWrite;
+    case PAGE_EXECUTE_READ:
+      return PageAccess::kExecuteReadOnly;
     case PAGE_EXECUTE_READWRITE:
       return PageAccess::kExecuteReadWrite;
     default:
@@ -70,6 +74,17 @@ PageAccess ToXeniaProtectFlags(DWORD access) {
   }
 }
 
+bool IsWritableExecutableMemorySupported() {
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
+  return true;
+#else
+  // To test FromApp functions on desktop, replace
+  // WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP) with 0 in the #ifs and
+  // link to WindowsApp.lib.
+  return false;
+#endif
+}
+
 void* AllocFixed(void* base_address, size_t length,
                  AllocationType allocation_type, PageAccess access) {
   DWORD alloc_type = 0;
@@ -88,7 +103,12 @@ void* AllocFixed(void* base_address, size_t length,
       break;
   }
   DWORD protect = ToWin32ProtectFlags(access);
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
   return VirtualAlloc(base_address, length, alloc_type, protect);
+#else
+  return VirtualAllocFromApp(base_address, length, ULONG(alloc_type),
+                             ULONG(protect));
+#endif
 }
 
 bool DeallocFixed(void* base_address, size_t length,
@@ -115,13 +135,19 @@ bool Protect(void* base_address, size_t length, PageAccess access,
     *out_old_access = PageAccess::kNoAccess;
   }
   DWORD new_protect = ToWin32ProtectFlags(access);
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
   DWORD old_protect = 0;
   BOOL result = VirtualProtect(base_address, length, new_protect, &old_protect);
+#else
+  ULONG old_protect = 0;
+  BOOL result = VirtualProtectFromApp(base_address, length, ULONG(new_protect),
+                                      &old_protect);
+#endif
   if (!result) {
     return false;
   }
   if (out_old_access) {
-    *out_old_access = ToXeniaProtectFlags(old_protect);
+    *out_old_access = ToXeniaProtectFlags(DWORD(old_protect));
   }
   return true;
 }
@@ -148,9 +174,14 @@ FileMappingHandle CreateFileMappingHandle(const std::filesystem::path& path,
   DWORD protect =
       ToWin32ProtectFlags(access) | (commit ? SEC_COMMIT : SEC_RESERVE);
   auto full_path = "Local" / path;
-  return CreateFileMappingW(INVALID_HANDLE_VALUE, NULL, protect,
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
+  return CreateFileMappingW(INVALID_HANDLE_VALUE, nullptr, protect,
                             static_cast<DWORD>(length >> 32),
                             static_cast<DWORD>(length), full_path.c_str());
+#else
+  return CreateFileMappingFromApp(INVALID_HANDLE_VALUE, nullptr, ULONG(protect),
+                                  ULONG64(length), full_path.c_str());
+#endif
 }
 
 void CloseFileMappingHandle(FileMappingHandle handle,
@@ -160,6 +191,7 @@ void CloseFileMappingHandle(FileMappingHandle handle,
 
 void* MapFileView(FileMappingHandle handle, void* base_address, size_t length,
                   PageAccess access, size_t file_offset) {
+#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
   DWORD target_address_low = static_cast<DWORD>(file_offset);
   DWORD target_address_high = static_cast<DWORD>(file_offset >> 32);
   DWORD file_access = 0;
@@ -170,6 +202,9 @@ void* MapFileView(FileMappingHandle handle, void* base_address, size_t length,
     case PageAccess::kReadWrite:
       file_access = FILE_MAP_ALL_ACCESS;
       break;
+    case PageAccess::kExecuteReadOnly:
+      file_access = FILE_MAP_READ | FILE_MAP_EXECUTE;
+      break;
     case PageAccess::kExecuteReadWrite:
       file_access = FILE_MAP_ALL_ACCESS | FILE_MAP_EXECUTE;
       break;
@@ -180,6 +215,25 @@ void* MapFileView(FileMappingHandle handle, void* base_address, size_t length,
   }
   return MapViewOfFileEx(handle, file_access, target_address_high,
                          target_address_low, length, base_address);
+#else
+  // VirtualAlloc2FromApp and MapViewOfFile3FromApp were added in 10.0.17134.0.
+  // https://docs.microsoft.com/en-us/uwp/win32-and-com/win32-apis
+  HANDLE process = GetCurrentProcess();
+  void* placeholder = VirtualAlloc2FromApp(
+      process, base_address, length, MEM_RESERVE | MEM_RESERVE_PLACEHOLDER,
+      PAGE_NOACCESS, nullptr, 0);
+  if (!placeholder) {
+    return nullptr;
+  }
+  void* mapping = MapViewOfFile3FromApp(
+      handle, process, placeholder, ULONG64(file_offset), length,
+      MEM_REPLACE_PLACEHOLDER, ULONG(ToWin32ProtectFlags(access)), nullptr, 0);
+  if (!mapping) {
+    VirtualFree(placeholder, length, MEM_RELEASE);
+    return nullptr;
+  }
+  return mapping;
+#endif
 }
 
 bool UnmapFileView(FileMappingHandle handle, void* base_address,

src/xenia/cpu/backend/code_cache.h

@@ -10,6 +10,8 @@
 #ifndef XENIA_CPU_BACKEND_CODE_CACHE_H_
 #define XENIA_CPU_BACKEND_CODE_CACHE_H_
 
+#include <cstddef>
+#include <cstdint>
 #include <string>
 
 #include "xenia/cpu/function.h"
@@ -24,8 +26,8 @@ class CodeCache {
   virtual ~CodeCache() = default;
 
   virtual const std::filesystem::path& file_name() const = 0;
-  virtual uint32_t base_address() const = 0;
+  virtual uintptr_t execute_base_address() const = 0;
-  virtual uint32_t total_size() const = 0;
+  virtual size_t total_size() const = 0;
 
   // Finds a function based on the given host PC (that may be within a
   // function).

src/xenia/cpu/backend/x64/x64_code_cache.cc

@@ -41,8 +41,15 @@ X64CodeCache::~X64CodeCache() {
 
   // Unmap all views and close mapping.
   if (mapping_ != xe::memory::kFileMappingHandleInvalid) {
-    xe::memory::UnmapFileView(mapping_, generated_code_base_,
+    if (generated_code_write_base_ &&
+        generated_code_write_base_ != generated_code_execute_base_) {
+      xe::memory::UnmapFileView(mapping_, generated_code_write_base_,
                                 kGeneratedCodeSize);
+    }
+    if (generated_code_execute_base_) {
+      xe::memory::UnmapFileView(mapping_, generated_code_execute_base_,
+                                kGeneratedCodeSize);
+    }
     xe::memory::CloseFileMappingHandle(mapping_, file_name_);
     mapping_ = xe::memory::kFileMappingHandleInvalid;
   }
@@ -73,18 +80,42 @@ bool X64CodeCache::Initialize() {
   }
 
   // Map generated code region into the file. Pages are committed as required.
-  generated_code_base_ = reinterpret_cast<uint8_t*>(xe::memory::MapFileView(
+  if (xe::memory::IsWritableExecutableMemoryPreferred()) {
-      mapping_, reinterpret_cast<void*>(kGeneratedCodeBase), kGeneratedCodeSize,
+    generated_code_execute_base_ =
-      xe::memory::PageAccess::kExecuteReadWrite, 0));
+        reinterpret_cast<uint8_t*>(xe::memory::MapFileView(
-  if (!generated_code_base_) {
+            mapping_, reinterpret_cast<void*>(kGeneratedCodeExecuteBase),
+            kGeneratedCodeSize, xe::memory::PageAccess::kExecuteReadWrite, 0));
+    generated_code_write_base_ = generated_code_execute_base_;
+    if (!generated_code_execute_base_ || !generated_code_write_base_) {
       XELOGE("Unable to allocate code cache generated code storage");
       XELOGE(
           "This is likely because the {:X}-{:X} range is in use by some other "
           "system DLL",
-        static_cast<uint64_t>(kGeneratedCodeBase),
+          uint64_t(kGeneratedCodeExecuteBase),
-        kGeneratedCodeBase + kGeneratedCodeSize);
+          uint64_t(kGeneratedCodeExecuteBase + kGeneratedCodeSize));
       return false;
     }
+  } else {
+    generated_code_execute_base_ =
+        reinterpret_cast<uint8_t*>(xe::memory::MapFileView(
+            mapping_, reinterpret_cast<void*>(kGeneratedCodeExecuteBase),
+            kGeneratedCodeSize, xe::memory::PageAccess::kExecuteReadOnly, 0));
+    generated_code_write_base_ =
+        reinterpret_cast<uint8_t*>(xe::memory::MapFileView(
+            mapping_, reinterpret_cast<void*>(kGeneratedCodeWriteBase),
+            kGeneratedCodeSize, xe::memory::PageAccess::kReadWrite, 0));
+    if (!generated_code_execute_base_ || !generated_code_write_base_) {
+      XELOGE("Unable to allocate code cache generated code storage");
+      XELOGE(
+          "This is likely because the {:X}-{:X} and {:X}-{:X} ranges are in "
+          "use by some other system DLL",
+          uint64_t(kGeneratedCodeExecuteBase),
+          uint64_t(kGeneratedCodeExecuteBase + kGeneratedCodeSize),
+          uint64_t(kGeneratedCodeWriteBase),
+          uint64_t(kGeneratedCodeWriteBase + kGeneratedCodeSize));
+      return false;
+    }
+  }
 
   // Preallocate the function map to a large, reasonable size.
   generated_code_map_.reserve(kMaximumFunctionCount);
@@ -117,7 +148,7 @@ void X64CodeCache::CommitExecutableRange(uint32_t guest_low,
   xe::memory::AllocFixed(
       indirection_table_base_ + (guest_low - kIndirectionTableBase),
       guest_high - guest_low, xe::memory::AllocationType::kCommit,
-      xe::memory::PageAccess::kExecuteReadWrite);
+      xe::memory::PageAccess::kReadWrite);
 
   // Fill memory with the default value.
   uint32_t* p = reinterpret_cast<uint32_t*>(indirection_table_base_);
@@ -126,21 +157,26 @@ void X64CodeCache::CommitExecutableRange(uint32_t guest_low,
   }
 }
 
-void* X64CodeCache::PlaceHostCode(uint32_t guest_address, void* machine_code,
+void X64CodeCache::PlaceHostCode(uint32_t guest_address, void* machine_code,
-                                  const EmitFunctionInfo& func_info) {
+                                 const EmitFunctionInfo& func_info,
+                                 void*& code_execute_address_out,
+                                 void*& code_write_address_out) {
   // Same for now. We may use different pools or whatnot later on, like when
   // we only want to place guest code in a serialized cache on disk.
-  return PlaceGuestCode(guest_address, machine_code, func_info, nullptr);
+  PlaceGuestCode(guest_address, machine_code, func_info, nullptr,
+                 code_execute_address_out, code_write_address_out);
 }
 
-void* X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
+void X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
                                   const EmitFunctionInfo& func_info,
-                                   GuestFunction* function_info) {
+                                  GuestFunction* function_info,
+                                  void*& code_execute_address_out,
+                                  void*& code_write_address_out) {
   // Hold a lock while we bump the pointers up. This is important as the
   // unwind table requires entries AND code to be sorted in order.
   size_t low_mark;
   size_t high_mark;
-  uint8_t* code_address;
+  uint8_t* code_execute_address;
   UnwindReservation unwind_reservation;
   {
     auto global_lock = global_critical_region_.Acquire();
@@ -149,26 +185,33 @@ void* X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
 
     // Reserve code.
     // Always move the code to land on 16b alignment.
-    code_address = generated_code_base_ + generated_code_offset_;
+    code_execute_address =
+        generated_code_execute_base_ + generated_code_offset_;
+    code_execute_address_out = code_execute_address;
+    uint8_t* code_write_address =
+        generated_code_write_base_ + generated_code_offset_;
+    code_write_address_out = code_write_address;
     generated_code_offset_ += xe::round_up(func_info.code_size.total, 16);
 
-    auto tail_address = generated_code_base_ + generated_code_offset_;
+    auto tail_write_address =
+        generated_code_write_base_ + generated_code_offset_;
 
     // Reserve unwind info.
     // We go on the high size of the unwind info as we don't know how big we
     // need it, and a few extra bytes of padding isn't the worst thing.
-    unwind_reservation =
+    unwind_reservation = RequestUnwindReservation(generated_code_write_base_ +
-        RequestUnwindReservation(generated_code_base_ + generated_code_offset_);
+                                                  generated_code_offset_);
     generated_code_offset_ += xe::round_up(unwind_reservation.data_size, 16);
 
-    auto end_address = generated_code_base_ + generated_code_offset_;
+    auto end_write_address =
+        generated_code_write_base_ + generated_code_offset_;
 
     high_mark = generated_code_offset_;
 
     // Store in map. It is maintained in sorted order of host PC dependent on
     // us also being append-only.
     generated_code_map_.emplace_back(
-        (uint64_t(code_address - generated_code_base_) << 32) |
+        (uint64_t(code_execute_address - generated_code_execute_base_) << 32) |
             generated_code_offset_,
         function_info);
 
@@ -185,21 +228,30 @@ void* X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
       if (high_mark <= old_commit_mark) break;
 
       new_commit_mark = old_commit_mark + 16 * 1024 * 1024;
-      xe::memory::AllocFixed(generated_code_base_, new_commit_mark,
+      if (generated_code_execute_base_ == generated_code_write_base_) {
+        xe::memory::AllocFixed(generated_code_execute_base_, new_commit_mark,
                                xe::memory::AllocationType::kCommit,
                                xe::memory::PageAccess::kExecuteReadWrite);
+      } else {
+        xe::memory::AllocFixed(generated_code_execute_base_, new_commit_mark,
+                               xe::memory::AllocationType::kCommit,
+                               xe::memory::PageAccess::kExecuteReadOnly);
+        xe::memory::AllocFixed(generated_code_write_base_, new_commit_mark,
+                               xe::memory::AllocationType::kCommit,
+                               xe::memory::PageAccess::kReadWrite);
+      }
     } while (generated_code_commit_mark_.compare_exchange_weak(
         old_commit_mark, new_commit_mark));
 
     // Copy code.
-    std::memcpy(code_address, machine_code, func_info.code_size.total);
+    std::memcpy(code_write_address, machine_code, func_info.code_size.total);
 
     // Fill unused slots with 0xCC
-    std::memset(tail_address, 0xCC,
+    std::memset(tail_write_address, 0xCC,
-                static_cast<size_t>(end_address - tail_address));
+                static_cast<size_t>(end_write_address - tail_write_address));
 
     // Notify subclasses of placed code.
-    PlaceCode(guest_address, machine_code, func_info, code_address,
+    PlaceCode(guest_address, machine_code, func_info, code_execute_address,
               unwind_reservation);
   }
 
@@ -214,7 +266,7 @@ void* X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
 
     iJIT_Method_Load_V2 method = {0};
     method.method_id = iJIT_GetNewMethodID();
-    method.method_load_address = code_address;
+    method.method_load_address = code_execute_address;
     method.method_size = uint32_t(code_size);
     method.method_name = const_cast<char*>(method_name.data());
     method.module_name = function_info
@@ -230,10 +282,9 @@ void* X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
   if (guest_address && indirection_table_base_) {
     uint32_t* indirection_slot = reinterpret_cast<uint32_t*>(
         indirection_table_base_ + (guest_address - kIndirectionTableBase));
-    *indirection_slot = uint32_t(reinterpret_cast<uint64_t>(code_address));
+    *indirection_slot =
+        uint32_t(reinterpret_cast<uint64_t>(code_execute_address));
   }
-
-  return code_address;
 }
 
 uint32_t X64CodeCache::PlaceData(const void* data, size_t length) {
@@ -245,7 +296,7 @@ uint32_t X64CodeCache::PlaceData(const void* data, size_t length) {
 
     // Reserve code.
     // Always move the code to land on 16b alignment.
-    data_address = generated_code_base_ + generated_code_offset_;
+    data_address = generated_code_write_base_ + generated_code_offset_;
     generated_code_offset_ += xe::round_up(length, 16);
 
     high_mark = generated_code_offset_;
@@ -260,9 +311,18 @@ uint32_t X64CodeCache::PlaceData(const void* data, size_t length) {
     if (high_mark <= old_commit_mark) break;
 
     new_commit_mark = old_commit_mark + 16 * 1024 * 1024;
-    xe::memory::AllocFixed(generated_code_base_, new_commit_mark,
+    if (generated_code_execute_base_ == generated_code_write_base_) {
+      xe::memory::AllocFixed(generated_code_execute_base_, new_commit_mark,
                              xe::memory::AllocationType::kCommit,
                              xe::memory::PageAccess::kExecuteReadWrite);
+    } else {
+      xe::memory::AllocFixed(generated_code_execute_base_, new_commit_mark,
+                             xe::memory::AllocationType::kCommit,
+                             xe::memory::PageAccess::kExecuteReadOnly);
+      xe::memory::AllocFixed(generated_code_write_base_, new_commit_mark,
+                             xe::memory::AllocationType::kCommit,
+                             xe::memory::PageAccess::kReadWrite);
+    }
   } while (generated_code_commit_mark_.compare_exchange_weak(old_commit_mark,
                                                              new_commit_mark));
 
@@ -273,7 +333,7 @@ uint32_t X64CodeCache::PlaceData(const void* data, size_t length) {
 }
 
 GuestFunction* X64CodeCache::LookupFunction(uint64_t host_pc) {
-  uint32_t key = uint32_t(host_pc - kGeneratedCodeBase);
+  uint32_t key = uint32_t(host_pc - kGeneratedCodeExecuteBase);
   void* fn_entry = std::bsearch(
       &key, generated_code_map_.data(), generated_code_map_.size() + 1,
       sizeof(std::pair<uint32_t, Function*>),

src/xenia/cpu/backend/x64/x64_code_cache.h

@@ -11,6 +11,8 @@
 #define XENIA_CPU_BACKEND_X64_X64_CODE_CACHE_H_
 
 #include <atomic>
+#include <cstddef>
+#include <cstdint>
 #include <memory>
 #include <string>
 #include <utility>
@@ -46,8 +48,10 @@ class X64CodeCache : public CodeCache {
   virtual bool Initialize();
 
   const std::filesystem::path& file_name() const override { return file_name_; }
-  uint32_t base_address() const override { return kGeneratedCodeBase; }
+  uintptr_t execute_base_address() const override {
-  uint32_t total_size() const override { return kGeneratedCodeSize; }
+    return kGeneratedCodeExecuteBase;
+  }
+  size_t total_size() const override { return kGeneratedCodeSize; }
 
   // TODO(benvanik): ELF serialization/etc
   // TODO(benvanik): keep track of code blocks
@@ -59,11 +63,15 @@ class X64CodeCache : public CodeCache {
 
   void CommitExecutableRange(uint32_t guest_low, uint32_t guest_high);
 
-  void* PlaceHostCode(uint32_t guest_address, void* machine_code,
+  void PlaceHostCode(uint32_t guest_address, void* machine_code,
-                      const EmitFunctionInfo& func_info);
-  void* PlaceGuestCode(uint32_t guest_address, void* machine_code,
                      const EmitFunctionInfo& func_info,
-                       GuestFunction* function_info);
+                     void*& code_execute_address_out,
+                     void*& code_write_address_out);
+  void PlaceGuestCode(uint32_t guest_address, void* machine_code,
+                      const EmitFunctionInfo& func_info,
+                      GuestFunction* function_info,
+                      void*& code_execute_address_out,
+                      void*& code_write_address_out);
   uint32_t PlaceData(const void* data, size_t length);
 
   GuestFunction* LookupFunction(uint64_t host_pc) override;
@@ -71,13 +79,16 @@ class X64CodeCache : public CodeCache {
  protected:
   // All executable code falls within 0x80000000 to 0x9FFFFFFF, so we can
   // only map enough for lookups within that range.
-  static const uint64_t kIndirectionTableBase = 0x80000000;
+  static const size_t kIndirectionTableSize = 0x1FFFFFFF;
-  static const uint64_t kIndirectionTableSize = 0x1FFFFFFF;
+  static const uintptr_t kIndirectionTableBase = 0x80000000;
   // The code range is 512MB, but we know the total code games will have is
   // pretty small (dozens of mb at most) and our expansion is reasonablish
   // so 256MB should be more than enough.
-  static const uint64_t kGeneratedCodeBase = 0xA0000000;
+  static const size_t kGeneratedCodeSize = 0x0FFFFFFF;
-  static const uint64_t kGeneratedCodeSize = 0x0FFFFFFF;
+  static const uintptr_t kGeneratedCodeExecuteBase = 0xA0000000;
+  // Used for writing when PageAccess::kExecuteReadWrite is not supported.
+  static const uintptr_t kGeneratedCodeWriteBase =
+      kGeneratedCodeExecuteBase + kGeneratedCodeSize + 1;
 
   // This is picked to be high enough to cover whatever we can reasonably
   // expect. If we hit issues with this it probably means some corner case
@@ -96,7 +107,8 @@ class X64CodeCache : public CodeCache {
     return UnwindReservation();
   }
   virtual void PlaceCode(uint32_t guest_address, void* machine_code,
-                         const EmitFunctionInfo& func_info, void* code_address,
+                         const EmitFunctionInfo& func_info,
+                         void* code_execute_address,
                          UnwindReservation unwind_reservation) {}
 
   std::filesystem::path file_name_;
@@ -114,9 +126,13 @@ class X64CodeCache : public CodeCache {
   // the generated code table that correspond to the PPC functions in guest
   // space.
   uint8_t* indirection_table_base_ = nullptr;
-  // Fixed at kGeneratedCodeBase and holding all generated code, growing as
+  // Fixed at kGeneratedCodeExecuteBase and holding all generated code, growing
-  // needed.
+  // as needed.
-  uint8_t* generated_code_base_ = nullptr;
+  uint8_t* generated_code_execute_base_ = nullptr;
+  // View of the memory that backs generated_code_execute_base_ when
+  // PageAccess::kExecuteReadWrite is not supported, for writing the generated
+  // code. Equals to generated_code_execute_base_ when it's supported.
+  uint8_t* generated_code_write_base_ = nullptr;
   // Current offset to empty space in generated code.
   size_t generated_code_offset_ = 0;
   // Current high water mark of COMMITTED code.

src/xenia/cpu/backend/x64/x64_code_cache_posix.cc

@@ -27,7 +27,7 @@ class PosixX64CodeCache : public X64CodeCache {
   /*
   UnwindReservation RequestUnwindReservation(uint8_t* entry_address) override;
   void PlaceCode(uint32_t guest_address, void* machine_code, size_t code_size,
-                 size_t stack_size, void* code_address,
+                 size_t stack_size, void* code_execute_address,
                  UnwindReservation unwind_reservation) override;
 
   void InitializeUnwindEntry(uint8_t* unwind_entry_address,

src/xenia/cpu/backend/x64/x64_code_cache_win.cc

@@ -107,11 +107,12 @@ class Win32X64CodeCache : public X64CodeCache {
  private:
   UnwindReservation RequestUnwindReservation(uint8_t* entry_address) override;
   void PlaceCode(uint32_t guest_address, void* machine_code,
-                 const EmitFunctionInfo& func_info, void* code_address,
+                 const EmitFunctionInfo& func_info, void* code_execute_address,
                  UnwindReservation unwind_reservation) override;
 
   void InitializeUnwindEntry(uint8_t* unwind_entry_address,
-                             size_t unwind_table_slot, void* code_address,
+                             size_t unwind_table_slot,
+                             void* code_execute_address,
                              const EmitFunctionInfo& func_info);
 
   // Growable function table system handle.
@@ -140,9 +141,9 @@ Win32X64CodeCache::~Win32X64CodeCache() {
       delete_growable_table_(unwind_table_handle_);
     }
   } else {
-    if (generated_code_base_) {
+    if (generated_code_execute_base_) {
       RtlDeleteFunctionTable(reinterpret_cast<PRUNTIME_FUNCTION>(
-          reinterpret_cast<DWORD64>(generated_code_base_) | 0x3));
+          reinterpret_cast<DWORD64>(generated_code_execute_base_) | 0x3));
     }
   }
 }
@@ -176,10 +177,11 @@ bool Win32X64CodeCache::Initialize() {
   // Create table and register with the system. It's empty now, but we'll grow
   // it as functions are added.
   if (supports_growable_table_) {
-    if (add_growable_table_(&unwind_table_handle_, unwind_table_.data(),
+    if (add_growable_table_(
-                            unwind_table_count_, DWORD(unwind_table_.size()),
+            &unwind_table_handle_, unwind_table_.data(), unwind_table_count_,
-                            reinterpret_cast<ULONG_PTR>(generated_code_base_),
+            DWORD(unwind_table_.size()),
-                            reinterpret_cast<ULONG_PTR>(generated_code_base_ +
+            reinterpret_cast<ULONG_PTR>(generated_code_execute_base_),
+            reinterpret_cast<ULONG_PTR>(generated_code_execute_base_ +
                                         kGeneratedCodeSize))) {
       XELOGE("Unable to create unwind function table");
       return false;
@@ -188,8 +190,9 @@ bool Win32X64CodeCache::Initialize() {
     // Install a callback that the debugger will use to lookup unwind info on
     // demand.
     if (!RtlInstallFunctionTableCallback(
-            reinterpret_cast<DWORD64>(generated_code_base_) | 0x3,
+            reinterpret_cast<DWORD64>(generated_code_execute_base_) | 0x3,
-            reinterpret_cast<DWORD64>(generated_code_base_), kGeneratedCodeSize,
+            reinterpret_cast<DWORD64>(generated_code_execute_base_),
+            kGeneratedCodeSize,
             [](DWORD64 control_pc, PVOID context) {
               auto code_cache = reinterpret_cast<Win32X64CodeCache*>(context);
               return reinterpret_cast<PRUNTIME_FUNCTION>(
@@ -216,11 +219,12 @@ Win32X64CodeCache::RequestUnwindReservation(uint8_t* entry_address) {
 
 void Win32X64CodeCache::PlaceCode(uint32_t guest_address, void* machine_code,
                                   const EmitFunctionInfo& func_info,
-                                  void* code_address,
+                                  void* code_execute_address,
                                   UnwindReservation unwind_reservation) {
   // Add unwind info.
   InitializeUnwindEntry(unwind_reservation.entry_address,
-                        unwind_reservation.table_slot, code_address, func_info);
+                        unwind_reservation.table_slot, code_execute_address,
+                        func_info);
 
   if (supports_growable_table_) {
     // Notify that the unwind table has grown.
@@ -229,13 +233,15 @@ void Win32X64CodeCache::PlaceCode(uint32_t guest_address, void* machine_code,
   }
 
   // This isn't needed on x64 (probably), but is convention.
-  FlushInstructionCache(GetCurrentProcess(), code_address,
+  // On UWP, FlushInstructionCache available starting from 10.0.16299.0.
+  // https://docs.microsoft.com/en-us/uwp/win32-and-com/win32-apis
+  FlushInstructionCache(GetCurrentProcess(), code_execute_address,
                         func_info.code_size.total);
 }
 
 void Win32X64CodeCache::InitializeUnwindEntry(
-    uint8_t* unwind_entry_address, size_t unwind_table_slot, void* code_address,
+    uint8_t* unwind_entry_address, size_t unwind_table_slot,
-    const EmitFunctionInfo& func_info) {
+    void* code_execute_address, const EmitFunctionInfo& func_info) {
   auto unwind_info = reinterpret_cast<UNWIND_INFO*>(unwind_entry_address);
   UNWIND_CODE* unwind_code = nullptr;
 
@@ -299,10 +305,12 @@ void Win32X64CodeCache::InitializeUnwindEntry(
   // Add entry.
   auto& fn_entry = unwind_table_[unwind_table_slot];
   fn_entry.BeginAddress =
-      (DWORD)(reinterpret_cast<uint8_t*>(code_address) - generated_code_base_);
+      DWORD(reinterpret_cast<uint8_t*>(code_execute_address) -
+            generated_code_execute_base_);
   fn_entry.EndAddress =
-      (DWORD)(fn_entry.BeginAddress + func_info.code_size.total);
+      DWORD(fn_entry.BeginAddress + func_info.code_size.total);
-  fn_entry.UnwindData = (DWORD)(unwind_entry_address - generated_code_base_);
+  fn_entry.UnwindData =
+      DWORD(unwind_entry_address - generated_code_execute_base_);
 }
 
 void* Win32X64CodeCache::LookupUnwindInfo(uint64_t host_pc) {
@@ -310,8 +318,8 @@ void* Win32X64CodeCache::LookupUnwindInfo(uint64_t host_pc) {
       &host_pc, unwind_table_.data(), unwind_table_count_,
       sizeof(RUNTIME_FUNCTION),
       [](const void* key_ptr, const void* element_ptr) {
-        auto key =
+        auto key = *reinterpret_cast<const uintptr_t*>(key_ptr) -
-            *reinterpret_cast<const uintptr_t*>(key_ptr) - kGeneratedCodeBase;
+                   kGeneratedCodeExecuteBase;
         auto element = reinterpret_cast<const RUNTIME_FUNCTION*>(element_ptr);
         if (key < element->BeginAddress) {
           return -1;

src/xenia/cpu/backend/x64/x64_emitter.cc

@@ -125,20 +125,26 @@ void* X64Emitter::Emplace(const EmitFunctionInfo& func_info,
   // top_ points to the Xbyak buffer, and since we are in AutoGrow mode
   // it has pending relocations. We copy the top_ to our buffer, swap the
   // pointer, relocate, then return the original scratch pointer for use.
+  // top_ is used by Xbyak's ready() as both write base pointer and the absolute
+  // address base, which would not work on platforms not supporting writable
+  // executable memory, but Xenia doesn't use absolute label addresses in the
+  // generated code.
   uint8_t* old_address = top_;
-  void* new_address;
+  void* new_execute_address;
+  void* new_write_address;
   assert_true(func_info.code_size.total == size_);
   if (function) {
-    new_address = code_cache_->PlaceGuestCode(function->address(), top_,
+    code_cache_->PlaceGuestCode(function->address(), top_, func_info, function,
-                                              func_info, function);
+                                new_execute_address, new_write_address);
   } else {
-    new_address = code_cache_->PlaceHostCode(0, top_, func_info);
+    code_cache_->PlaceHostCode(0, top_, func_info, new_execute_address,
+                               new_write_address);
   }
-  top_ = reinterpret_cast<uint8_t*>(new_address);
+  top_ = reinterpret_cast<uint8_t*>(new_write_address);
   ready();
   top_ = old_address;
   reset();
-  return new_address;
+  return new_execute_address;
 }
 
 bool X64Emitter::Emit(HIRBuilder* builder, EmitFunctionInfo& func_info) {

src/xenia/cpu/ppc/testing/ppc_testing_native_main.cc

@@ -177,6 +177,9 @@ class TestRunner {
  public:
   TestRunner() {
     memory_size_ = 64 * 1024 * 1024;
+    // FIXME(Triang3l): If this is ever compiled for a platform without
+    // xe::memory::IsWritableExecutableMemorySupported, two memory mappings must
+    // be used.
     memory_ = memory::AllocFixed(nullptr, memory_size_,
                                  memory::AllocationType::kReserveCommit,
                                  memory::PageAccess::kExecuteReadWrite);

src/xenia/cpu/stack_walker_win.cc

@@ -9,6 +9,7 @@
 
 #include "xenia/cpu/stack_walker.h"
 
+#include <cstdint>
 #include <mutex>
 
 #include "xenia/base/logging.h"
@@ -120,8 +121,8 @@ class Win32StackWalker : public StackWalker {
     // They never change, so it's fine even if they are touched from multiple
     // threads.
     code_cache_ = code_cache;
-    code_cache_min_ = code_cache_->base_address();
+    code_cache_min_ = code_cache_->execute_base_address();
-    code_cache_max_ = code_cache_->base_address() + code_cache_->total_size();
+    code_cache_max_ = code_cache_min_ + code_cache_->total_size();
   }
 
   bool Initialize() {
@@ -297,13 +298,13 @@ class Win32StackWalker : public StackWalker {
   std::mutex dbghelp_mutex_;
 
   static xe::cpu::backend::CodeCache* code_cache_;
-  static uint32_t code_cache_min_;
+  static uintptr_t code_cache_min_;
-  static uint32_t code_cache_max_;
+  static uintptr_t code_cache_max_;
 };
 
 xe::cpu::backend::CodeCache* Win32StackWalker::code_cache_ = nullptr;
-uint32_t Win32StackWalker::code_cache_min_ = 0;
+uintptr_t Win32StackWalker::code_cache_min_ = 0;
-uint32_t Win32StackWalker::code_cache_max_ = 0;
+uintptr_t Win32StackWalker::code_cache_max_ = 0;
 
 std::unique_ptr<StackWalker> StackWalker::Create(
     backend::CodeCache* code_cache) {

src/xenia/emulator.cc

@@ -513,7 +513,7 @@ bool Emulator::ExceptionCallbackThunk(Exception* ex, void* data) {
 bool Emulator::ExceptionCallback(Exception* ex) {
   // Check to see if the exception occurred in guest code.
   auto code_cache = processor()->backend()->code_cache();
-  auto code_base = code_cache->base_address();
+  auto code_base = code_cache->execute_base_address();
   auto code_end = code_base + code_cache->total_size();
 
   if (!processor()->is_debugger_attached() && debugging::IsDebuggerAttached()) {

src/xenia/memory.cc

@@ -621,6 +621,10 @@ uint32_t FromPageAccess(xe::memory::PageAccess protect) {
       return kMemoryProtectRead;
     case memory::PageAccess::kReadWrite:
       return kMemoryProtectRead | kMemoryProtectWrite;
+    case memory::PageAccess::kExecuteReadOnly:
+      // Guest memory cannot be executable - this should never happen :)
+      assert_always();
+      return kMemoryProtectRead;
     case memory::PageAccess::kExecuteReadWrite:
       // Guest memory cannot be executable - this should never happen :)
       assert_always();