Android #143

renderer_vulkan: Introduce separate cmd buffer for uploads

.codespellrc

@@ -3,4 +3,4 @@
 
 [codespell]
 skip = ./.git,./build,./dist,./Doxyfile,./externals,./LICENSES,./src/android/app/src/main/res
-ignore-words-list = aci,allright,ba,canonicalizations,deques,froms,hda,inout,lod,masia,nam,nax,nd,optin,pullrequests,pullrequest,te,transfered,unstall,uscaled,vas,zink
+ignore-words-list = aci,allright,ba,canonicalizations,deques,froms,hda,inout,lod,masia,nam,nax,nce,nd,optin,pullrequests,pullrequest,te,transfered,unstall,uscaled,vas,zink

.gitmodules

@@ -4,9 +4,6 @@
 [submodule "enet"]
 	path = externals/enet
 	url = https://github.com/lsalzman/enet.git
-[submodule "inih"]
-	path = externals/inih/inih
-	url = https://github.com/benhoyt/inih.git
 [submodule "cubeb"]
 	path = externals/cubeb
 	url = https://github.com/mozilla/cubeb.git
@@ -61,3 +58,9 @@
 [submodule "breakpad"]
 	path = externals/breakpad
 	url = https://github.com/yuzu-emu/breakpad.git
+[submodule "simpleini"]
+	path = externals/simpleini
+	url = https://github.com/brofield/simpleini.git
+[submodule "oaknut"]
+	path = externals/oaknut
+	url = https://github.com/merryhime/oaknut

CMakeLists.txt

@@ -285,12 +285,12 @@ set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/bin)
 find_package(Boost 1.79.0 REQUIRED context)
 find_package(enet 1.3 MODULE)
 find_package(fmt 9 REQUIRED)
-find_package(inih 52 MODULE COMPONENTS INIReader)
 find_package(LLVM 17.0.2 MODULE COMPONENTS Demangle)
 find_package(lz4 REQUIRED)
 find_package(nlohmann_json 3.8 REQUIRED)
 find_package(Opus 1.3 MODULE)
 find_package(RenderDoc MODULE)
+find_package(SimpleIni MODULE)
 find_package(stb MODULE)
 find_package(VulkanMemoryAllocator CONFIG)
 find_package(ZLIB 1.2 REQUIRED)

CMakeModules/FindSimpleIni.cmake

@@ -0,0 +1,19 @@
+# SPDX-FileCopyrightText: 2023 Alexandre Bouvier <contact@amb.tf>
+#
+# SPDX-License-Identifier: GPL-3.0-or-later
+
+find_path(SimpleIni_INCLUDE_DIR SimpleIni.h)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(SimpleIni
+    REQUIRED_VARS SimpleIni_INCLUDE_DIR
+)
+
+if (SimpleIni_FOUND AND NOT TARGET SimpleIni::SimpleIni)
+    add_library(SimpleIni::SimpleIni INTERFACE IMPORTED)
+    set_target_properties(SimpleIni::SimpleIni PROPERTIES
+        INTERFACE_INCLUDE_DIRECTORIES "${SimpleIni_INCLUDE_DIR}"
+    )
+endif()
+
+mark_as_advanced(SimpleIni_INCLUDE_DIR)

CMakeModules/Findinih.cmake

@@ -1,27 +0,0 @@
-# SPDX-FileCopyrightText: 2022 Alexandre Bouvier <contact@amb.tf>
-#
-# SPDX-License-Identifier: GPL-3.0-or-later
-
-find_package(PkgConfig QUIET)
-pkg_search_module(INIH QUIET IMPORTED_TARGET inih)
-if (INIReader IN_LIST inih_FIND_COMPONENTS)
-    pkg_search_module(INIREADER QUIET IMPORTED_TARGET INIReader)
-    if (INIREADER_FOUND)
-        set(inih_INIReader_FOUND TRUE)
-    endif()
-endif()
-
-include(FindPackageHandleStandardArgs)
-find_package_handle_standard_args(inih
-    REQUIRED_VARS INIH_LINK_LIBRARIES
-    VERSION_VAR INIH_VERSION
-    HANDLE_COMPONENTS
-)
-
-if (inih_FOUND AND NOT TARGET inih::inih)
-    add_library(inih::inih ALIAS PkgConfig::INIH)
-endif()
-
-if (inih_FOUND AND inih_INIReader_FOUND AND NOT TARGET inih::INIReader)
-    add_library(inih::INIReader ALIAS PkgConfig::INIREADER)
-endif()

README.md

@@ -1,3 +1,12 @@
+| Pull Request | Commit | Title | Author | Merged? |
+|----|----|----|----|----|
+| [12074](https://github.com/yuzu-emu/yuzu//pull/12074) | [`15f35b865`](https://github.com/yuzu-emu/yuzu//pull/12074/files) | Implement Native Code Execution (NCE) | [GPUCode](https://github.com/GPUCode/) | Yes |
+
+
+End of merge log. You can find the original README.md below the break.
+
+-----
+
 <!--
 SPDX-FileCopyrightText: 2018 yuzu Emulator Project
 SPDX-License-Identifier: GPL-2.0-or-later

externals/CMakeLists.txt

@@ -20,6 +20,10 @@ if ((ARCHITECTURE_x86 OR ARCHITECTURE_x86_64) AND NOT TARGET xbyak::xbyak)
 endif()
 
 # Dynarmic
+if (ARCHITECTURE_arm64 AND NOT TARGET merry::oaknut)
+    add_subdirectory(oaknut)
+endif()
+
 if ((ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64) AND NOT TARGET dynarmic::dynarmic)
     set(DYNARMIC_IGNORE_ASSERTS ON)
     add_subdirectory(dynarmic)
@@ -34,11 +38,6 @@ endif()
 # Glad
 add_subdirectory(glad)
 
-# inih
-if (NOT TARGET inih::INIReader)
-    add_subdirectory(inih)
-endif()
-
 # mbedtls
 add_subdirectory(mbedtls)
 target_include_directories(mbedtls PUBLIC ./mbedtls/include)
@@ -295,3 +294,8 @@ if (YUZU_CRASH_DUMPS AND NOT TARGET libbreakpad_client)
         target_link_libraries(dump_syms PRIVATE libbreakpad_client ZLIB::ZLIB)
     endif()
 endif()
+
+# SimpleIni
+if (NOT TARGET SimpleIni::SimpleIni)
+    add_subdirectory(simpleini)
+endif()

externals/inih/CMakeLists.txt

@@ -1,13 +0,0 @@
-# SPDX-FileCopyrightText: 2014 Gui Andrade <admin@archshift.com>
-# SPDX-License-Identifier: GPL-2.0-or-later
-
-add_library(inih
-    inih/ini.c
-    inih/ini.h
-    inih/cpp/INIReader.cpp
-    inih/cpp/INIReader.h
-)
-
-create_target_directory_groups(inih)
-target_include_directories(inih INTERFACE inih/cpp)
-add_library(inih::INIReader ALIAS inih)

src/CMakeLists.txt

@@ -187,6 +187,7 @@ add_subdirectory(audio_core)
 add_subdirectory(video_core)
 add_subdirectory(network)
 add_subdirectory(input_common)
+add_subdirectory(frontend_common)
 add_subdirectory(shader_recompiler)
 
 if (YUZU_ROOM)

src/android/app/build.gradle.kts

@@ -219,7 +219,6 @@ dependencies {
     implementation("io.coil-kt:coil:2.2.2")
     implementation("androidx.core:core-splashscreen:1.0.1")
     implementation("androidx.window:window:1.2.0-beta03")
-    implementation("org.ini4j:ini4j:0.5.4")
     implementation("androidx.constraintlayout:constraintlayout:2.1.4")
     implementation("androidx.swiperefreshlayout:swiperefreshlayout:1.1.0")
     implementation("androidx.navigation:navigation-fragment-ktx:2.7.4")

src/android/app/src/main/java/org/yuzu/yuzu_emu/NativeLibrary.kt

@@ -230,8 +230,6 @@ object NativeLibrary {
      */
     external fun onTouchReleased(finger_id: Int)
 
-    external fun reloadSettings()
-
     external fun initGameIni(gameID: String?)
 
     external fun setAppDirectory(directory: String)
@@ -301,6 +299,11 @@ object NativeLibrary {
      */
     external fun getPerfStats(): DoubleArray
 
+    /**
+     * Returns the current CPU backend.
+     */
+    external fun getCpuBackend(): String
+
     /**
      * Notifies the core emulation that the orientation has changed.
      */

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/IntSetting.kt

@@ -10,6 +10,7 @@ enum class IntSetting(
     override val category: Settings.Category,
     override val androidDefault: Int? = null
 ) : AbstractIntSetting {
+    CPU_BACKEND("cpu_backend", Settings.Category.Cpu),
     CPU_ACCURACY("cpu_accuracy", Settings.Category.Cpu),
     REGION_INDEX("region_index", Settings.Category.System),
     LANGUAGE_INDEX("language_index", Settings.Category.System),

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/Settings.kt

@@ -7,7 +7,7 @@ import android.text.TextUtils
 import android.widget.Toast
 import org.yuzu.yuzu_emu.R
 import org.yuzu.yuzu_emu.YuzuApplication
-import org.yuzu.yuzu_emu.features.settings.utils.SettingsFile
+import org.yuzu.yuzu_emu.utils.NativeConfig
 
 object Settings {
     private val context get() = YuzuApplication.appContext
@@ -19,7 +19,7 @@ object Settings {
                 context.getString(R.string.ini_saved),
                 Toast.LENGTH_SHORT
             ).show()
-            SettingsFile.saveFile(SettingsFile.FILE_NAME_CONFIG)
+            NativeConfig.saveSettings()
         } else {
             // TODO: Save custom game settings
             Toast.makeText(

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt

@@ -77,6 +77,15 @@ abstract class SettingsItem(
                     "%"
                 )
             )
+            put(
+                SingleChoiceSetting(
+                    IntSetting.CPU_BACKEND,
+                    R.string.cpu_backend,
+                    0,
+                    R.array.cpuBackendArm64Names,
+                    R.array.cpuBackendArm64Values
+                )
+            )
             put(
                 SingleChoiceSetting(
                     IntSetting.CPU_ACCURACY,

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsActivity.kt

@@ -21,7 +21,6 @@ import androidx.navigation.navArgs
 import com.google.android.material.color.MaterialColors
 import kotlinx.coroutines.flow.collectLatest
 import kotlinx.coroutines.launch
-import org.yuzu.yuzu_emu.NativeLibrary
 import java.io.IOException
 import org.yuzu.yuzu_emu.R
 import org.yuzu.yuzu_emu.databinding.ActivitySettingsBinding
@@ -165,11 +164,12 @@ class SettingsActivity : AppCompatActivity() {
         settingsViewModel.shouldSave = false
 
         // Delete settings file because the user may have changed values that do not exist in the UI
+        NativeConfig.unloadConfig()
         val settingsFile = SettingsFile.getSettingsFile(SettingsFile.FILE_NAME_CONFIG)
         if (!settingsFile.delete()) {
             throw IOException("Failed to delete $settingsFile")
         }
-        NativeLibrary.reloadSettings()
+        NativeConfig.initializeConfig()
 
         Toast.makeText(
             applicationContext,

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt

@@ -269,6 +269,7 @@ class SettingsFragmentPresenter(
             add(BooleanSetting.RENDERER_DEBUG.key)
 
             add(HeaderSetting(R.string.cpu))
+            add(IntSetting.CPU_BACKEND.key)
             add(IntSetting.CPU_ACCURACY.key)
             add(BooleanSetting.CPU_DEBUG_MODE.key)
             add(SettingsItem.FASTMEM_COMBINED)

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/utils/SettingsFile.kt

@@ -3,15 +3,8 @@
 
 package org.yuzu.yuzu_emu.features.settings.utils
 
-import android.widget.Toast
 import java.io.*
-import org.ini4j.Wini
-import org.yuzu.yuzu_emu.R
-import org.yuzu.yuzu_emu.YuzuApplication
-import org.yuzu.yuzu_emu.features.settings.model.*
 import org.yuzu.yuzu_emu.utils.DirectoryInitialization
-import org.yuzu.yuzu_emu.utils.Log
-import org.yuzu.yuzu_emu.utils.NativeConfig
 
 /**
  * Contains static methods for interacting with .ini files in which settings are stored.
@@ -19,41 +12,6 @@ import org.yuzu.yuzu_emu.utils.NativeConfig
 object SettingsFile {
     const val FILE_NAME_CONFIG = "config"
 
-    /**
-     * Saves a Settings HashMap to a given .ini file on disk. If unsuccessful, outputs an error
-     * telling why it failed.
-     *
-     * @param fileName The target filename without a path or extension.
-     */
-    fun saveFile(fileName: String) {
-        val ini = getSettingsFile(fileName)
-        try {
-            val wini = Wini(ini)
-            for (specificCategory in Settings.Category.values()) {
-                val categoryHeader = NativeConfig.getConfigHeader(specificCategory.ordinal)
-                for (setting in Settings.settingsList) {
-                    if (setting.key!!.isEmpty()) continue
-
-                    val settingCategoryHeader =
-                        NativeConfig.getConfigHeader(setting.category.ordinal)
-                    val iniSetting: String? = wini.get(categoryHeader, setting.key)
-                    if (iniSetting != null || settingCategoryHeader == categoryHeader) {
-                        wini.put(settingCategoryHeader, setting.key, setting.valueAsString)
-                    }
-                }
-            }
-            wini.store()
-        } catch (e: IOException) {
-            Log.error("[SettingsFile] File not found: " + fileName + ".ini: " + e.message)
-            val context = YuzuApplication.appContext
-            Toast.makeText(
-                context,
-                context.getString(R.string.error_saving, fileName, e.message),
-                Toast.LENGTH_SHORT
-            ).show()
-        }
-    }
-
     fun getSettingsFile(fileName: String): File =
         File(DirectoryInitialization.userDirectory + "/config/" + fileName + ".ini")
 }

src/android/app/src/main/java/org/yuzu/yuzu_emu/fragments/EmulationFragment.kt

@@ -414,8 +414,10 @@ class EmulationFragment : Fragment(), SurfaceHolder.Callback {
             perfStatsUpdater = {
                 if (emulationViewModel.emulationStarted.value) {
                     val perfStats = NativeLibrary.getPerfStats()
+                    val cpuBackend = NativeLibrary.getCpuBackend()
                     if (_binding != null) {
-                        binding.showFpsText.text = String.format("FPS: %.1f", perfStats[FPS])
+                        binding.showFpsText.text =
+                            String.format("FPS: %.1f\n%s", perfStats[FPS], cpuBackend)
                     }
                     perfStatsUpdateHandler.postDelayed(perfStatsUpdater!!, 800)
                 }

src/android/app/src/main/java/org/yuzu/yuzu_emu/ui/main/MainActivity.kt

@@ -625,6 +625,7 @@ class MainActivity : AppCompatActivity(), ThemeProvider {
                 }
 
                 // Clear existing user data
+                NativeConfig.unloadConfig()
                 File(DirectoryInitialization.userDirectory!!).deleteRecursively()
 
                 // Copy archive to internal storage
@@ -643,6 +644,7 @@ class MainActivity : AppCompatActivity(), ThemeProvider {
 
                 // Reinitialize relevant data
                 NativeLibrary.initializeSystem(true)
+                NativeConfig.initializeConfig()
                 gamesViewModel.reloadGames(false)
 
                 return@newInstance getString(R.string.user_data_import_success)

src/android/app/src/main/java/org/yuzu/yuzu_emu/utils/DirectoryInitialization.kt

@@ -16,6 +16,7 @@ object DirectoryInitialization {
         if (!areDirectoriesReady) {
             initializeInternalStorage()
             NativeLibrary.initializeSystem(false)
+            NativeConfig.initializeConfig()
             areDirectoriesReady = true
         }
     }

src/android/app/src/main/java/org/yuzu/yuzu_emu/utils/InputHandler.kt

@@ -27,6 +27,8 @@ object InputHandler {
             0x054C -> getInputDS5ButtonKey(event.keyCode)
             0x057E -> getInputJoyconButtonKey(event.keyCode)
             0x1532 -> getInputRazerButtonKey(event.keyCode)
+            0x3537 -> getInputRedmagicButtonKey(event.keyCode)
+            0x358A -> getInputBackboneLabsButtonKey(event.keyCode)
             else -> getInputGenericButtonKey(event.keyCode)
         }
 
@@ -227,6 +229,42 @@ object InputHandler {
         }
     }
 
+    private fun getInputRedmagicButtonKey(key: Int): Int {
+        return when (key) {
+            KeyEvent.KEYCODE_BUTTON_A -> NativeLibrary.ButtonType.BUTTON_B
+            KeyEvent.KEYCODE_BUTTON_B -> NativeLibrary.ButtonType.BUTTON_A
+            KeyEvent.KEYCODE_BUTTON_X -> NativeLibrary.ButtonType.BUTTON_Y
+            KeyEvent.KEYCODE_BUTTON_Y -> NativeLibrary.ButtonType.BUTTON_X
+            KeyEvent.KEYCODE_BUTTON_L1 -> NativeLibrary.ButtonType.TRIGGER_L
+            KeyEvent.KEYCODE_BUTTON_R1 -> NativeLibrary.ButtonType.TRIGGER_R
+            KeyEvent.KEYCODE_BUTTON_L2 -> NativeLibrary.ButtonType.TRIGGER_ZL
+            KeyEvent.KEYCODE_BUTTON_R2 -> NativeLibrary.ButtonType.TRIGGER_ZR
+            KeyEvent.KEYCODE_BUTTON_THUMBL -> NativeLibrary.ButtonType.STICK_L
+            KeyEvent.KEYCODE_BUTTON_THUMBR -> NativeLibrary.ButtonType.STICK_R
+            KeyEvent.KEYCODE_BUTTON_START -> NativeLibrary.ButtonType.BUTTON_PLUS
+            KeyEvent.KEYCODE_BUTTON_SELECT -> NativeLibrary.ButtonType.BUTTON_MINUS
+            else -> -1
+        }
+    }
+
+    private fun getInputBackboneLabsButtonKey(key: Int): Int {
+        return when (key) {
+            KeyEvent.KEYCODE_BUTTON_A -> NativeLibrary.ButtonType.BUTTON_B
+            KeyEvent.KEYCODE_BUTTON_B -> NativeLibrary.ButtonType.BUTTON_A
+            KeyEvent.KEYCODE_BUTTON_X -> NativeLibrary.ButtonType.BUTTON_Y
+            KeyEvent.KEYCODE_BUTTON_Y -> NativeLibrary.ButtonType.BUTTON_X
+            KeyEvent.KEYCODE_BUTTON_L1 -> NativeLibrary.ButtonType.TRIGGER_L
+            KeyEvent.KEYCODE_BUTTON_R1 -> NativeLibrary.ButtonType.TRIGGER_R
+            KeyEvent.KEYCODE_BUTTON_L2 -> NativeLibrary.ButtonType.TRIGGER_ZL
+            KeyEvent.KEYCODE_BUTTON_R2 -> NativeLibrary.ButtonType.TRIGGER_ZR
+            KeyEvent.KEYCODE_BUTTON_THUMBL -> NativeLibrary.ButtonType.STICK_L
+            KeyEvent.KEYCODE_BUTTON_THUMBR -> NativeLibrary.ButtonType.STICK_R
+            KeyEvent.KEYCODE_BUTTON_START -> NativeLibrary.ButtonType.BUTTON_PLUS
+            KeyEvent.KEYCODE_BUTTON_SELECT -> NativeLibrary.ButtonType.BUTTON_MINUS
+            else -> -1
+        }
+    }
+
     private fun getInputGenericButtonKey(key: Int): Int {
         return when (key) {
             KeyEvent.KEYCODE_BUTTON_A -> NativeLibrary.ButtonType.BUTTON_A

src/android/app/src/main/java/org/yuzu/yuzu_emu/utils/NativeConfig.kt

@@ -4,6 +4,30 @@
 package org.yuzu.yuzu_emu.utils
 
 object NativeConfig {
+    /**
+     * Creates a Config object and opens the emulation config.
+     */
+    @Synchronized
+    external fun initializeConfig()
+
+    /**
+     * Destroys the stored config object. This automatically saves the existing config.
+     */
+    @Synchronized
+    external fun unloadConfig()
+
+    /**
+     * Reads values saved to the config file and saves them.
+     */
+    @Synchronized
+    external fun reloadSettings()
+
+    /**
+     * Saves settings values in memory to disk.
+     */
+    @Synchronized
+    external fun saveSettings()
+
     external fun getBoolean(key: String, getDefault: Boolean): Boolean
     external fun setBoolean(key: String, value: Boolean)
 

src/android/app/src/main/jni/CMakeLists.txt

@@ -6,9 +6,6 @@ add_library(yuzu-android SHARED
     android_common/android_common.h
     applets/software_keyboard.cpp
     applets/software_keyboard.h
-    config.cpp
-    config.h
-    default_ini.h
     emu_window/emu_window.cpp
     emu_window/emu_window.h
     id_cache.cpp
@@ -16,15 +13,17 @@ add_library(yuzu-android SHARED
     native.cpp
     native.h
     native_config.cpp
-    uisettings.cpp
+    android_settings.cpp
     game_metadata.cpp
     native_log.cpp
+    android_config.cpp
+    android_config.h
 )
 
 set_property(TARGET yuzu-android PROPERTY IMPORTED_LOCATION ${FFmpeg_LIBRARY_DIR})
 
-target_link_libraries(yuzu-android PRIVATE audio_core common core input_common)
-target_link_libraries(yuzu-android PRIVATE android camera2ndk EGL glad inih jnigraphics log)
+target_link_libraries(yuzu-android PRIVATE audio_core common core input_common frontend_common)
+target_link_libraries(yuzu-android PRIVATE android camera2ndk EGL glad jnigraphics log)
 if (ARCHITECTURE_arm64)
     target_link_libraries(yuzu-android PRIVATE adrenotools)
 endif()

src/android/app/src/main/jni/android_config.cpp

@@ -0,0 +1,70 @@
+// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "android_config.h"
+#include "android_settings.h"
+#include "common/settings_setting.h"
+
+AndroidConfig::AndroidConfig(const std::string& config_name, ConfigType config_type)
+    : Config(config_type) {
+    Initialize(config_name);
+    if (config_type != ConfigType::InputProfile) {
+        ReadAndroidValues();
+        SaveAndroidValues();
+    }
+}
+
+AndroidConfig::~AndroidConfig() {
+    if (global) {
+        AndroidConfig::SaveAllValues();
+    }
+}
+
+void AndroidConfig::ReloadAllValues() {
+    Reload();
+    ReadAndroidValues();
+    SaveAndroidValues();
+}
+
+void AndroidConfig::SaveAllValues() {
+    Save();
+    SaveAndroidValues();
+}
+
+void AndroidConfig::ReadAndroidValues() {
+    if (global) {
+        ReadAndroidUIValues();
+    }
+}
+
+void AndroidConfig::ReadAndroidUIValues() {
+    BeginGroup(Settings::TranslateCategory(Settings::Category::Android));
+
+    ReadCategory(Settings::Category::Android);
+
+    EndGroup();
+}
+
+void AndroidConfig::SaveAndroidValues() {
+    if (global) {
+        SaveAndroidUIValues();
+    }
+
+    WriteToIni();
+}
+
+void AndroidConfig::SaveAndroidUIValues() {
+    BeginGroup(Settings::TranslateCategory(Settings::Category::Android));
+
+    WriteCategory(Settings::Category::Android);
+
+    EndGroup();
+}
+
+std::vector<Settings::BasicSetting*>& AndroidConfig::FindRelevantList(Settings::Category category) {
+    auto& map = Settings::values.linkage.by_category;
+    if (map.contains(category)) {
+        return Settings::values.linkage.by_category[category];
+    }
+    return AndroidSettings::values.linkage.by_category[category];
+}

src/android/app/src/main/jni/android_config.h

@@ -0,0 +1,41 @@
+// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "frontend_common/config.h"
+
+class AndroidConfig final : public Config {
+public:
+    explicit AndroidConfig(const std::string& config_name = "config",
+                           ConfigType config_type = ConfigType::GlobalConfig);
+    ~AndroidConfig() override;
+
+    void ReloadAllValues() override;
+    void SaveAllValues() override;
+
+protected:
+    void ReadAndroidValues();
+    void ReadAndroidUIValues();
+    void ReadHidbusValues() override {}
+    void ReadDebugControlValues() override {}
+    void ReadPathValues() override {}
+    void ReadShortcutValues() override {}
+    void ReadUIValues() override {}
+    void ReadUIGamelistValues() override {}
+    void ReadUILayoutValues() override {}
+    void ReadMultiplayerValues() override {}
+
+    void SaveAndroidValues();
+    void SaveAndroidUIValues();
+    void SaveHidbusValues() override {}
+    void SaveDebugControlValues() override {}
+    void SavePathValues() override {}
+    void SaveShortcutValues() override {}
+    void SaveUIValues() override {}
+    void SaveUIGamelistValues() override {}
+    void SaveUILayoutValues() override {}
+    void SaveMultiplayerValues() override {}
+
+    std::vector<Settings::BasicSetting*>& FindRelevantList(Settings::Category category) override;
+};

src/android/app/src/main/jni/android_settings.cpp

@@ -1,7 +1,7 @@
 // SPDX-FileCopyrightText: 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include "uisettings.h"
+#include "android_settings.h"
 
 namespace AndroidSettings {
 

src/android/app/src/main/jni/config.cpp

@@ -1,330 +0,0 @@
-// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#include <memory>
-#include <optional>
-#include <sstream>
-
-#include <INIReader.h>
-#include "common/fs/file.h"
-#include "common/fs/fs.h"
-#include "common/fs/path_util.h"
-#include "common/logging/log.h"
-#include "common/settings.h"
-#include "common/settings_enums.h"
-#include "core/hle/service/acc/profile_manager.h"
-#include "input_common/main.h"
-#include "jni/config.h"
-#include "jni/default_ini.h"
-#include "uisettings.h"
-
-namespace FS = Common::FS;
-
-Config::Config(const std::string& config_name, ConfigType config_type)
-    : type(config_type), global{config_type == ConfigType::GlobalConfig} {
-    Initialize(config_name);
-}
-
-Config::~Config() = default;
-
-bool Config::LoadINI(const std::string& default_contents, bool retry) {
-    void(FS::CreateParentDir(config_loc));
-    config = std::make_unique<INIReader>(FS::PathToUTF8String(config_loc));
-    const auto config_loc_str = FS::PathToUTF8String(config_loc);
-    if (config->ParseError() < 0) {
-        if (retry) {
-            LOG_WARNING(Config, "Failed to load {}. Creating file from defaults...",
-                        config_loc_str);
-
-            void(FS::CreateParentDir(config_loc));
-            void(FS::WriteStringToFile(config_loc, FS::FileType::TextFile, default_contents));
-
-            config = std::make_unique<INIReader>(config_loc_str);
-
-            return LoadINI(default_contents, false);
-        }
-        LOG_ERROR(Config, "Failed.");
-        return false;
-    }
-    LOG_INFO(Config, "Successfully loaded {}", config_loc_str);
-    return true;
-}
-
-template <>
-void Config::ReadSetting(const std::string& group, Settings::Setting<std::string>& setting) {
-    std::string setting_value = config->Get(group, setting.GetLabel(), setting.GetDefault());
-    if (setting_value.empty()) {
-        setting_value = setting.GetDefault();
-    }
-    setting = std::move(setting_value);
-}
-
-template <>
-void Config::ReadSetting(const std::string& group, Settings::Setting<bool>& setting) {
-    setting = config->GetBoolean(group, setting.GetLabel(), setting.GetDefault());
-}
-
-template <typename Type, bool ranged>
-void Config::ReadSetting(const std::string& group, Settings::Setting<Type, ranged>& setting) {
-    setting = static_cast<Type>(
-        config->GetInteger(group, setting.GetLabel(), static_cast<long>(setting.GetDefault())));
-}
-
-void Config::ReadValues() {
-    ReadSetting("ControlsGeneral", Settings::values.mouse_enabled);
-    ReadSetting("ControlsGeneral", Settings::values.touch_device);
-    ReadSetting("ControlsGeneral", Settings::values.keyboard_enabled);
-    ReadSetting("ControlsGeneral", Settings::values.debug_pad_enabled);
-    ReadSetting("ControlsGeneral", Settings::values.vibration_enabled);
-    ReadSetting("ControlsGeneral", Settings::values.enable_accurate_vibrations);
-    ReadSetting("ControlsGeneral", Settings::values.motion_enabled);
-    Settings::values.touchscreen.enabled =
-        config->GetBoolean("ControlsGeneral", "touch_enabled", true);
-    Settings::values.touchscreen.rotation_angle =
-        config->GetInteger("ControlsGeneral", "touch_angle", 0);
-    Settings::values.touchscreen.diameter_x =
-        config->GetInteger("ControlsGeneral", "touch_diameter_x", 15);
-    Settings::values.touchscreen.diameter_y =
-        config->GetInteger("ControlsGeneral", "touch_diameter_y", 15);
-
-    int num_touch_from_button_maps =
-        config->GetInteger("ControlsGeneral", "touch_from_button_map", 0);
-    if (num_touch_from_button_maps > 0) {
-        for (int i = 0; i < num_touch_from_button_maps; ++i) {
-            Settings::TouchFromButtonMap map;
-            map.name = config->Get("ControlsGeneral",
-                                   std::string("touch_from_button_maps_") + std::to_string(i) +
-                                       std::string("_name"),
-                                   "default");
-            const int num_touch_maps = config->GetInteger(
-                "ControlsGeneral",
-                std::string("touch_from_button_maps_") + std::to_string(i) + std::string("_count"),
-                0);
-            map.buttons.reserve(num_touch_maps);
-
-            for (int j = 0; j < num_touch_maps; ++j) {
-                std::string touch_mapping =
-                    config->Get("ControlsGeneral",
-                                std::string("touch_from_button_maps_") + std::to_string(i) +
-                                    std::string("_bind_") + std::to_string(j),
-                                "");
-                map.buttons.emplace_back(std::move(touch_mapping));
-            }
-
-            Settings::values.touch_from_button_maps.emplace_back(std::move(map));
-        }
-    } else {
-        Settings::values.touch_from_button_maps.emplace_back(
-            Settings::TouchFromButtonMap{"default", {}});
-        num_touch_from_button_maps = 1;
-    }
-    Settings::values.touch_from_button_map_index = std::clamp(
-        Settings::values.touch_from_button_map_index.GetValue(), 0, num_touch_from_button_maps - 1);
-
-    ReadSetting("ControlsGeneral", Settings::values.udp_input_servers);
-
-    // Data Storage
-    ReadSetting("Data Storage", Settings::values.use_virtual_sd);
-    FS::SetYuzuPath(FS::YuzuPath::NANDDir,
-                    config->Get("Data Storage", "nand_directory",
-                                FS::GetYuzuPathString(FS::YuzuPath::NANDDir)));
-    FS::SetYuzuPath(FS::YuzuPath::SDMCDir,
-                    config->Get("Data Storage", "sdmc_directory",
-                                FS::GetYuzuPathString(FS::YuzuPath::SDMCDir)));
-    FS::SetYuzuPath(FS::YuzuPath::LoadDir,
-                    config->Get("Data Storage", "load_directory",
-                                FS::GetYuzuPathString(FS::YuzuPath::LoadDir)));
-    FS::SetYuzuPath(FS::YuzuPath::DumpDir,
-                    config->Get("Data Storage", "dump_directory",
-                                FS::GetYuzuPathString(FS::YuzuPath::DumpDir)));
-    ReadSetting("Data Storage", Settings::values.gamecard_inserted);
-    ReadSetting("Data Storage", Settings::values.gamecard_current_game);
-    ReadSetting("Data Storage", Settings::values.gamecard_path);
-
-    // System
-    ReadSetting("System", Settings::values.current_user);
-    Settings::values.current_user = std::clamp<int>(Settings::values.current_user.GetValue(), 0,
-                                                    Service::Account::MAX_USERS - 1);
-
-    // Disable docked mode by default on Android
-    Settings::values.use_docked_mode.SetValue(config->GetBoolean("System", "use_docked_mode", false)
-                                                  ? Settings::ConsoleMode::Docked
-                                                  : Settings::ConsoleMode::Handheld);
-
-    const auto rng_seed_enabled = config->GetBoolean("System", "rng_seed_enabled", false);
-    if (rng_seed_enabled) {
-        Settings::values.rng_seed.SetValue(config->GetInteger("System", "rng_seed", 0));
-    } else {
-        Settings::values.rng_seed.SetValue(0);
-    }
-    Settings::values.rng_seed_enabled.SetValue(rng_seed_enabled);
-
-    const auto custom_rtc_enabled = config->GetBoolean("System", "custom_rtc_enabled", false);
-    if (custom_rtc_enabled) {
-        Settings::values.custom_rtc = config->GetInteger("System", "custom_rtc", 0);
-    } else {
-        Settings::values.custom_rtc = 0;
-    }
-    Settings::values.custom_rtc_enabled = custom_rtc_enabled;
-
-    ReadSetting("System", Settings::values.language_index);
-    ReadSetting("System", Settings::values.region_index);
-    ReadSetting("System", Settings::values.time_zone_index);
-    ReadSetting("System", Settings::values.sound_index);
-
-    // Core
-    ReadSetting("Core", Settings::values.use_multi_core);
-    ReadSetting("Core", Settings::values.memory_layout_mode);
-
-    // Cpu
-    ReadSetting("Cpu", Settings::values.cpu_accuracy);
-    ReadSetting("Cpu", Settings::values.cpu_debug_mode);
-    ReadSetting("Cpu", Settings::values.cpuopt_page_tables);
-    ReadSetting("Cpu", Settings::values.cpuopt_block_linking);
-    ReadSetting("Cpu", Settings::values.cpuopt_return_stack_buffer);
-    ReadSetting("Cpu", Settings::values.cpuopt_fast_dispatcher);
-    ReadSetting("Cpu", Settings::values.cpuopt_context_elimination);
-    ReadSetting("Cpu", Settings::values.cpuopt_const_prop);
-    ReadSetting("Cpu", Settings::values.cpuopt_misc_ir);
-    ReadSetting("Cpu", Settings::values.cpuopt_reduce_misalign_checks);
-    ReadSetting("Cpu", Settings::values.cpuopt_fastmem);
-    ReadSetting("Cpu", Settings::values.cpuopt_fastmem_exclusives);
-    ReadSetting("Cpu", Settings::values.cpuopt_recompile_exclusives);
-    ReadSetting("Cpu", Settings::values.cpuopt_ignore_memory_aborts);
-    ReadSetting("Cpu", Settings::values.cpuopt_unsafe_unfuse_fma);
-    ReadSetting("Cpu", Settings::values.cpuopt_unsafe_reduce_fp_error);
-    ReadSetting("Cpu", Settings::values.cpuopt_unsafe_ignore_standard_fpcr);
-    ReadSetting("Cpu", Settings::values.cpuopt_unsafe_inaccurate_nan);
-    ReadSetting("Cpu", Settings::values.cpuopt_unsafe_fastmem_check);
-    ReadSetting("Cpu", Settings::values.cpuopt_unsafe_ignore_global_monitor);
-
-    // Renderer
-    ReadSetting("Renderer", Settings::values.renderer_backend);
-    ReadSetting("Renderer", Settings::values.renderer_debug);
-    ReadSetting("Renderer", Settings::values.renderer_shader_feedback);
-    ReadSetting("Renderer", Settings::values.enable_nsight_aftermath);
-    ReadSetting("Renderer", Settings::values.disable_shader_loop_safety_checks);
-    ReadSetting("Renderer", Settings::values.vulkan_device);
-
-    ReadSetting("Renderer", Settings::values.resolution_setup);
-    ReadSetting("Renderer", Settings::values.scaling_filter);
-    ReadSetting("Renderer", Settings::values.fsr_sharpening_slider);
-    ReadSetting("Renderer", Settings::values.anti_aliasing);
-    ReadSetting("Renderer", Settings::values.fullscreen_mode);
-    ReadSetting("Renderer", Settings::values.aspect_ratio);
-    ReadSetting("Renderer", Settings::values.max_anisotropy);
-    ReadSetting("Renderer", Settings::values.use_speed_limit);
-    ReadSetting("Renderer", Settings::values.speed_limit);
-    ReadSetting("Renderer", Settings::values.use_disk_shader_cache);
-    ReadSetting("Renderer", Settings::values.use_asynchronous_gpu_emulation);
-    ReadSetting("Renderer", Settings::values.vsync_mode);
-    ReadSetting("Renderer", Settings::values.shader_backend);
-    ReadSetting("Renderer", Settings::values.use_asynchronous_shaders);
-    ReadSetting("Renderer", Settings::values.nvdec_emulation);
-    ReadSetting("Renderer", Settings::values.use_fast_gpu_time);
-    ReadSetting("Renderer", Settings::values.use_vulkan_driver_pipeline_cache);
-
-    ReadSetting("Renderer", Settings::values.bg_red);
-    ReadSetting("Renderer", Settings::values.bg_green);
-    ReadSetting("Renderer", Settings::values.bg_blue);
-
-    // Use GPU accuracy normal by default on Android
-    Settings::values.gpu_accuracy = static_cast<Settings::GpuAccuracy>(config->GetInteger(
-        "Renderer", "gpu_accuracy", static_cast<u32>(Settings::GpuAccuracy::Normal)));
-
-    // Use GPU default anisotropic filtering on Android
-    Settings::values.max_anisotropy =
-        static_cast<Settings::AnisotropyMode>(config->GetInteger("Renderer", "max_anisotropy", 1));
-
-    // Disable ASTC compute by default on Android
-    Settings::values.accelerate_astc.SetValue(
-        config->GetBoolean("Renderer", "accelerate_astc", false) ? Settings::AstcDecodeMode::Gpu
-                                                                 : Settings::AstcDecodeMode::Cpu);
-
-    // Enable asynchronous presentation by default on Android
-    Settings::values.async_presentation =
-        config->GetBoolean("Renderer", "async_presentation", true);
-
-    // Disable force_max_clock by default on Android
-    Settings::values.renderer_force_max_clock =
-        config->GetBoolean("Renderer", "force_max_clock", false);
-
-    // Disable use_reactive_flushing by default on Android
-    Settings::values.use_reactive_flushing =
-        config->GetBoolean("Renderer", "use_reactive_flushing", false);
-
-    // Audio
-    ReadSetting("Audio", Settings::values.sink_id);
-    ReadSetting("Audio", Settings::values.audio_output_device_id);
-    ReadSetting("Audio", Settings::values.volume);
-
-    // Miscellaneous
-    // log_filter has a different default here than from common
-    Settings::values.log_filter = "*:Info";
-    ReadSetting("Miscellaneous", Settings::values.use_dev_keys);
-
-    // Debugging
-    Settings::values.record_frame_times =
-        config->GetBoolean("Debugging", "record_frame_times", false);
-    ReadSetting("Debugging", Settings::values.dump_exefs);
-    ReadSetting("Debugging", Settings::values.dump_nso);
-    ReadSetting("Debugging", Settings::values.enable_fs_access_log);
-    ReadSetting("Debugging", Settings::values.reporting_services);
-    ReadSetting("Debugging", Settings::values.quest_flag);
-    ReadSetting("Debugging", Settings::values.use_debug_asserts);
-    ReadSetting("Debugging", Settings::values.use_auto_stub);
-    ReadSetting("Debugging", Settings::values.disable_macro_jit);
-    ReadSetting("Debugging", Settings::values.disable_macro_hle);
-    ReadSetting("Debugging", Settings::values.use_gdbstub);
-    ReadSetting("Debugging", Settings::values.gdbstub_port);
-
-    const auto title_list = config->Get("AddOns", "title_ids", "");
-    std::stringstream ss(title_list);
-    std::string line;
-    while (std::getline(ss, line, '|')) {
-        const auto title_id = std::strtoul(line.c_str(), nullptr, 16);
-        const auto disabled_list = config->Get("AddOns", "disabled_" + line, "");
-
-        std::stringstream inner_ss(disabled_list);
-        std::string inner_line;
-        std::vector<std::string> out;
-        while (std::getline(inner_ss, inner_line, '|')) {
-            out.push_back(inner_line);
-        }
-
-        Settings::values.disabled_addons.insert_or_assign(title_id, out);
-    }
-
-    // Web Service
-    ReadSetting("WebService", Settings::values.enable_telemetry);
-    ReadSetting("WebService", Settings::values.web_api_url);
-    ReadSetting("WebService", Settings::values.yuzu_username);
-    ReadSetting("WebService", Settings::values.yuzu_token);
-
-    // Network
-    ReadSetting("Network", Settings::values.network_interface);
-
-    // Android
-    ReadSetting("Android", AndroidSettings::values.picture_in_picture);
-    ReadSetting("Android", AndroidSettings::values.screen_layout);
-}
-
-void Config::Initialize(const std::string& config_name) {
-    const auto fs_config_loc = FS::GetYuzuPath(FS::YuzuPath::ConfigDir);
-    const auto config_file = fmt::format("{}.ini", config_name);
-
-    switch (type) {
-    case ConfigType::GlobalConfig:
-        config_loc = FS::PathToUTF8String(fs_config_loc / config_file);
-        break;
-    case ConfigType::PerGameConfig:
-        config_loc = FS::PathToUTF8String(fs_config_loc / "custom" / FS::ToU8String(config_file));
-        break;
-    case ConfigType::InputProfile:
-        config_loc = FS::PathToUTF8String(fs_config_loc / "input" / config_file);
-        LoadINI(DefaultINI::android_config_file);
-        return;
-    }
-    LoadINI(DefaultINI::android_config_file);
-    ReadValues();
-}

src/android/app/src/main/jni/config.h

@@ -1,47 +0,0 @@
-// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#pragma once
-
-#include <filesystem>
-#include <memory>
-#include <optional>
-#include <string>
-
-#include "common/settings.h"
-
-class INIReader;
-
-class Config {
-    bool LoadINI(const std::string& default_contents = "", bool retry = true);
-
-public:
-    enum class ConfigType {
-        GlobalConfig,
-        PerGameConfig,
-        InputProfile,
-    };
-
-    explicit Config(const std::string& config_name = "config",
-                    ConfigType config_type = ConfigType::GlobalConfig);
-    ~Config();
-
-    void Initialize(const std::string& config_name);
-
-private:
-    /**
-     * Applies a value read from the config to a Setting.
-     *
-     * @param group The name of the INI group
-     * @param setting The yuzu setting to modify
-     */
-    template <typename Type, bool ranged>
-    void ReadSetting(const std::string& group, Settings::Setting<Type, ranged>& setting);
-
-    void ReadValues();
-
-    const ConfigType type;
-    std::unique_ptr<INIReader> config;
-    std::string config_loc;
-    const bool global;
-};

src/android/app/src/main/jni/default_ini.h

@@ -1,511 +0,0 @@
-// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#pragma once
-
-namespace DefaultINI {
-
-const char* android_config_file = R"(
-
-[ControlsP0]
-# The input devices and parameters for each Switch native input
-# The config section determines the player number where the config will be applied on. For example "ControlsP0", "ControlsP1", ...
-# It should be in the format of "engine:[engine_name],[param1]:[value1],[param2]:[value2]..."
-# Escape characters $0 (for ':'), $1 (for ',') and $2 (for '$') can be used in values
-
-# Indicates if this player should be connected at boot
-connected=
-
-# for button input, the following devices are available:
-#  - "keyboard" (default) for keyboard input. Required parameters:
-#      - "code": the code of the key to bind
-#  - "sdl" for joystick input using SDL. Required parameters:
-#      - "guid": SDL identification GUID of the joystick
-#      - "port": the index of the joystick to bind
-#      - "button"(optional): the index of the button to bind
-#      - "hat"(optional): the index of the hat to bind as direction buttons
-#      - "axis"(optional): the index of the axis to bind
-#      - "direction"(only used for hat): the direction name of the hat to bind. Can be "up", "down", "left" or "right"
-#      - "threshold"(only used for axis): a float value in (-1.0, 1.0) which the button is
-#          triggered if the axis value crosses
-#      - "direction"(only used for axis): "+" means the button is triggered when the axis value
-#          is greater than the threshold; "-" means the button is triggered when the axis value
-#          is smaller than the threshold
-button_a=
-button_b=
-button_x=
-button_y=
-button_lstick=
-button_rstick=
-button_l=
-button_r=
-button_zl=
-button_zr=
-button_plus=
-button_minus=
-button_dleft=
-button_dup=
-button_dright=
-button_ddown=
-button_lstick_left=
-button_lstick_up=
-button_lstick_right=
-button_lstick_down=
-button_sl=
-button_sr=
-button_home=
-button_screenshot=
-
-# for analog input, the following devices are available:
-#  - "analog_from_button" (default) for emulating analog input from direction buttons. Required parameters:
-#      - "up", "down", "left", "right": sub-devices for each direction.
-#          Should be in the format as a button input devices using escape characters, for example, "engine$0keyboard$1code$00"
-#      - "modifier": sub-devices as a modifier.
-#      - "modifier_scale": a float number representing the applied modifier scale to the analog input.
-#          Must be in range of 0.0-1.0. Defaults to 0.5
-#  - "sdl" for joystick input using SDL. Required parameters:
-#      - "guid": SDL identification GUID of the joystick
-#      - "port": the index of the joystick to bind
-#      - "axis_x": the index of the axis to bind as x-axis (default to 0)
-#      - "axis_y": the index of the axis to bind as y-axis (default to 1)
-lstick=
-rstick=
-
-# for motion input, the following devices are available:
-#  - "keyboard" (default) for emulating random motion input from buttons. Required parameters:
-#      - "code": the code of the key to bind
-#  - "sdl" for motion input using SDL. Required parameters:
-#      - "guid": SDL identification GUID of the joystick
-#      - "port": the index of the joystick to bind
-#      - "motion": the index of the motion sensor to bind
-#  - "cemuhookudp" for motion input using Cemu Hook protocol. Required parameters:
-#      - "guid": the IP address of the cemu hook server encoded to a hex string. for example 192.168.0.1 = "c0a80001"
-#      - "port": the port of the cemu hook server
-#      - "pad": the index of the joystick
-#      - "motion": the index of the motion sensor of the joystick to bind
-motionleft=
-motionright=
-
-[ControlsGeneral]
-# To use the debug_pad, prepend `debug_pad_` before each button setting above.
-# i.e. debug_pad_button_a=
-
-# Enable debug pad inputs to the guest
-# 0 (default): Disabled, 1: Enabled
-debug_pad_enabled =
-
-# Whether to enable or disable vibration
-# 0: Disabled, 1 (default): Enabled
-vibration_enabled=
-
-# Whether to enable or disable accurate vibrations
-# 0 (default): Disabled, 1: Enabled
-enable_accurate_vibrations=
-
-# Enables controller motion inputs
-# 0: Disabled, 1 (default): Enabled
-motion_enabled =
-
-# Defines the udp device's touch screen coordinate system for cemuhookudp devices
-#  - "min_x", "min_y", "max_x", "max_y"
-touch_device=
-
-# for mapping buttons to touch inputs.
-#touch_from_button_map=1
-#touch_from_button_maps_0_name=default
-#touch_from_button_maps_0_count=2
-#touch_from_button_maps_0_bind_0=foo
-#touch_from_button_maps_0_bind_1=bar
-# etc.
-
-# List of Cemuhook UDP servers, delimited by ','.
-# Default: 127.0.0.1:26760
-# Example: 127.0.0.1:26760,123.4.5.67:26761
-udp_input_servers =
-
-# Enable controlling an axis via a mouse input.
-# 0 (default): Off, 1: On
-mouse_panning =
-
-# Set mouse sensitivity.
-# Default: 1.0
-mouse_panning_sensitivity =
-
-# Emulate an analog control stick from keyboard inputs.
-# 0 (default): Disabled, 1: Enabled
-emulate_analog_keyboard =
-
-# Enable mouse inputs to the guest
-# 0 (default): Disabled, 1: Enabled
-mouse_enabled =
-
-# Enable keyboard inputs to the guest
-# 0 (default): Disabled, 1: Enabled
-keyboard_enabled =
-
-[Core]
-# Whether to use multi-core for CPU emulation
-# 0: Disabled, 1 (default): Enabled
-use_multi_core =
-
-# Enable unsafe extended guest system memory layout (8GB DRAM)
-# 0 (default): Disabled, 1: Enabled
-use_unsafe_extended_memory_layout =
-
-[Cpu]
-# Adjusts various optimizations.
-# Auto-select mode enables choice unsafe optimizations.
-# Accurate enables only safe optimizations.
-# Unsafe allows any unsafe optimizations.
-# 0 (default): Auto-select, 1: Accurate, 2: Enable unsafe optimizations
-cpu_accuracy =
-
-# Allow disabling safe optimizations.
-# 0 (default): Disabled, 1: Enabled
-cpu_debug_mode =
-
-# Enable inline page tables optimization (faster guest memory access)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_page_tables =
-
-# Enable block linking CPU optimization (reduce block dispatcher use during predictable jumps)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_block_linking =
-
-# Enable return stack buffer CPU optimization (reduce block dispatcher use during predictable returns)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_return_stack_buffer =
-
-# Enable fast dispatcher CPU optimization (use a two-tiered dispatcher architecture)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_fast_dispatcher =
-
-# Enable context elimination CPU Optimization (reduce host memory use for guest context)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_context_elimination =
-
-# Enable constant propagation CPU optimization (basic IR optimization)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_const_prop =
-
-# Enable miscellaneous CPU optimizations (basic IR optimization)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_misc_ir =
-
-# Enable reduction of memory misalignment checks (reduce memory fallbacks for misaligned access)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_reduce_misalign_checks =
-
-# Enable Host MMU Emulation (faster guest memory access)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_fastmem =
-
-# Enable Host MMU Emulation for exclusive memory instructions (faster guest memory access)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_fastmem_exclusives =
-
-# Enable fallback on failure of fastmem of exclusive memory instructions (faster guest memory access)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_recompile_exclusives =
-
-# Enable optimization to ignore invalid memory accesses (faster guest memory access)
-# 0: Disabled, 1 (default): Enabled
-cpuopt_ignore_memory_aborts =
-
-# Enable unfuse FMA (improve performance on CPUs without FMA)
-# Only enabled if cpu_accuracy is set to Unsafe. Automatically chosen with cpu_accuracy = Auto-select.
-# 0: Disabled, 1 (default): Enabled
-cpuopt_unsafe_unfuse_fma =
-
-# Enable faster FRSQRTE and FRECPE
-# Only enabled if cpu_accuracy is set to Unsafe.
-# 0: Disabled, 1 (default): Enabled
-cpuopt_unsafe_reduce_fp_error =
-
-# Enable faster ASIMD instructions (32 bits only)
-# Only enabled if cpu_accuracy is set to Unsafe. Automatically chosen with cpu_accuracy = Auto-select.
-# 0: Disabled, 1 (default): Enabled
-cpuopt_unsafe_ignore_standard_fpcr =
-
-# Enable inaccurate NaN handling
-# Only enabled if cpu_accuracy is set to Unsafe. Automatically chosen with cpu_accuracy = Auto-select.
-# 0: Disabled, 1 (default): Enabled
-cpuopt_unsafe_inaccurate_nan =
-
-# Disable address space checks (64 bits only)
-# Only enabled if cpu_accuracy is set to Unsafe. Automatically chosen with cpu_accuracy = Auto-select.
-# 0: Disabled, 1 (default): Enabled
-cpuopt_unsafe_fastmem_check =
-
-# Enable faster exclusive instructions
-# Only enabled if cpu_accuracy is set to Unsafe. Automatically chosen with cpu_accuracy = Auto-select.
-# 0: Disabled, 1 (default): Enabled
-cpuopt_unsafe_ignore_global_monitor =
-
-[Renderer]
-# Which backend API to use.
-# 0: OpenGL (unsupported), 1 (default): Vulkan, 2: Null
-backend =
-
-# Whether to enable asynchronous presentation (Vulkan only)
-# 0: Off, 1 (default): On
-async_presentation =
-
-# Forces the GPU to run at the maximum possible clocks (thermal constraints will still be applied).
-# 0 (default): Disabled, 1: Enabled
-force_max_clock =
-
-# Enable graphics API debugging mode.
-# 0 (default): Disabled, 1: Enabled
-debug =
-
-# Enable shader feedback.
-# 0 (default): Disabled, 1: Enabled
-renderer_shader_feedback =
-
-# Enable Nsight Aftermath crash dumps
-# 0 (default): Disabled, 1: Enabled
-nsight_aftermath =
-
-# Disable shader loop safety checks, executing the shader without loop logic changes
-# 0 (default): Disabled, 1: Enabled
-disable_shader_loop_safety_checks =
-
-# Which Vulkan physical device to use (defaults to 0)
-vulkan_device =
-
-# 0: 0.5x (360p/540p) [EXPERIMENTAL]
-# 1: 0.75x (540p/810p) [EXPERIMENTAL]
-# 2 (default): 1x (720p/1080p)
-# 3: 2x (1440p/2160p)
-# 4: 3x (2160p/3240p)
-# 5: 4x (2880p/4320p)
-# 6: 5x (3600p/5400p)
-# 7: 6x (4320p/6480p)
-resolution_setup =
-
-# Pixel filter to use when up- or down-sampling rendered frames.
-# 0: Nearest Neighbor
-# 1 (default): Bilinear
-# 2: Bicubic
-# 3: Gaussian
-# 4: ScaleForce
-# 5: AMD FidelityFX™️ Super Resolution [Vulkan Only]
-scaling_filter =
-
-# Anti-Aliasing (AA)
-# 0 (default): None, 1: FXAA
-anti_aliasing =
-
-# Whether to use fullscreen or borderless window mode
-# 0 (Windows default): Borderless window, 1 (All other default): Exclusive fullscreen
-fullscreen_mode =
-
-# Aspect ratio
-# 0: Default (16:9), 1: Force 4:3, 2: Force 21:9, 3: Force 16:10, 4: Stretch to Window
-aspect_ratio =
-
-# Anisotropic filtering
-# 0: Default, 1: 2x, 2: 4x, 3: 8x, 4: 16x
-max_anisotropy =
-
-# Whether to enable VSync or not.
-# OpenGL: Values other than 0 enable VSync
-# Vulkan: FIFO is selected if the requested mode is not supported by the driver.
-# FIFO (VSync) does not drop frames or exhibit tearing but is limited by the screen refresh rate.
-# FIFO Relaxed is similar to FIFO but allows tearing as it recovers from a slow down.
-# Mailbox can have lower latency than FIFO and does not tear but may drop frames.
-# Immediate (no synchronization) just presents whatever is available and can exhibit tearing.
-# 0: Immediate (Off), 1 (Default): Mailbox (On), 2: FIFO, 3: FIFO Relaxed
-use_vsync =
-
-# Selects the OpenGL shader backend. NV_gpu_program5 is required for GLASM. If NV_gpu_program5 is
-# not available and GLASM is selected, GLSL will be used.
-# 0: GLSL, 1 (default): GLASM, 2: SPIR-V
-shader_backend =
-
-# Whether to allow asynchronous shader building.
-# 0 (default): Off, 1: On
-use_asynchronous_shaders =
-
-# Uses reactive flushing instead of predictive flushing. Allowing a more accurate syncing of memory.
-# 0 (default): Off, 1: On
-use_reactive_flushing =
-
-# NVDEC emulation.
-# 0: Disabled, 1: CPU Decoding, 2 (default): GPU Decoding
-nvdec_emulation =
-
-# Accelerate ASTC texture decoding.
-# 0 (default): Off, 1: On
-accelerate_astc =
-
-# Turns on the speed limiter, which will limit the emulation speed to the desired speed limit value
-# 0: Off, 1: On (default)
-use_speed_limit =
-
-# Limits the speed of the game to run no faster than this value as a percentage of target speed
-# 1 - 9999: Speed limit as a percentage of target game speed. 100 (default)
-speed_limit =
-
-# Whether to use disk based shader cache
-# 0: Off, 1 (default): On
-use_disk_shader_cache =
-
-# Which gpu accuracy level to use
-# 0 (default): Normal, 1: High, 2: Extreme (Very slow)
-gpu_accuracy =
-
-# Whether to use asynchronous GPU emulation
-# 0 : Off (slow), 1 (default): On (fast)
-use_asynchronous_gpu_emulation =
-
-# Inform the guest that GPU operations completed more quickly than they did.
-# 0: Off, 1 (default): On
-use_fast_gpu_time =
-
-# Force unmodified buffers to be flushed, which can cost performance.
-# 0: Off (default), 1: On
-use_pessimistic_flushes =
-
-# Whether to use garbage collection or not for GPU caches.
-# 0 (default): Off, 1: On
-use_caches_gc =
-
-# The clear color for the renderer. What shows up on the sides of the bottom screen.
-# Must be in range of 0-255. Defaults to 0 for all.
-bg_red =
-bg_blue =
-bg_green =
-
-[Audio]
-# Which audio output engine to use.
-# auto (default): Auto-select
-# cubeb: Cubeb audio engine (if available)
-# sdl2: SDL2 audio engine (if available)
-# null: No audio output
-output_engine =
-
-# Which audio device to use.
-# auto (default): Auto-select
-output_device =
-
-# Output volume.
-# 100 (default): 100%, 0; mute
-volume =
-
-[Data Storage]
-# Whether to create a virtual SD card.
-# 1: Yes, 0 (default): No
-use_virtual_sd =
-
-# Whether or not to enable gamecard emulation
-# 1: Yes, 0 (default): No
-gamecard_inserted =
-
-# Whether or not the gamecard should be emulated as the current game
-# If 'gamecard_inserted' is 0 this setting is irrelevant
-# 1: Yes, 0 (default): No
-gamecard_current_game =
-
-# Path to an XCI file to use as the gamecard
-# If 'gamecard_inserted' is 0 this setting is irrelevant
-# If 'gamecard_current_game' is 1 this setting is irrelevant
-gamecard_path =
-
-[System]
-# Whether the system is docked
-# 1 (default): Yes, 0: No
-use_docked_mode =
-
-# Sets the seed for the RNG generator built into the switch
-# rng_seed will be ignored and randomly generated if rng_seed_enabled is false
-rng_seed_enabled =
-rng_seed =
-
-# Sets the current time (in seconds since 12:00 AM Jan 1, 1970) that will be used by the time service
-# This will auto-increment, with the time set being the time the game is started
-# This override will only occur if custom_rtc_enabled is true, otherwise the current time is used
-custom_rtc_enabled =
-custom_rtc =
-
-# Sets the systems language index
-# 0: Japanese, 1: English (default), 2: French, 3: German, 4: Italian, 5: Spanish, 6: Chinese,
-# 7: Korean, 8: Dutch, 9: Portuguese, 10: Russian, 11: Taiwanese, 12: British English, 13: Canadian French,
-# 14: Latin American Spanish, 15: Simplified Chinese, 16: Traditional Chinese, 17: Brazilian Portuguese
-language_index =
-
-# The system region that yuzu will use during emulation
-# -1: Auto-select (default), 0: Japan, 1: USA, 2: Europe, 3: Australia, 4: China, 5: Korea, 6: Taiwan
-region_index =
-
-# The system time zone that yuzu will use during emulation
-# 0: Auto-select (default), 1: Default (system archive value), Others: Index for specified time zone
-time_zone_index =
-
-# Sets the sound output mode.
-# 0: Mono, 1 (default): Stereo, 2: Surround
-sound_index =
-
-[Miscellaneous]
-# A filter which removes logs below a certain logging level.
-# Examples: *:Debug Kernel.SVC:Trace Service.*:Critical
-log_filter = *:Trace
-
-# Use developer keys
-# 0 (default): Disabled, 1: Enabled
-use_dev_keys =
-
-[Debugging]
-# Record frame time data, can be found in the log directory. Boolean value
-record_frame_times =
-# Determines whether or not yuzu will dump the ExeFS of all games it attempts to load while loading them
-dump_exefs=false
-# Determines whether or not yuzu will dump all NSOs it attempts to load while loading them
-dump_nso=false
-# Determines whether or not yuzu will save the filesystem access log.
-enable_fs_access_log=false
-# Enables verbose reporting services
-reporting_services =
-# Determines whether or not yuzu will report to the game that the emulated console is in Kiosk Mode
-# false: Retail/Normal Mode (default), true: Kiosk Mode
-quest_flag =
-# Determines whether debug asserts should be enabled, which will throw an exception on asserts.
-# false: Disabled (default), true: Enabled
-use_debug_asserts =
-# Determines whether unimplemented HLE service calls should be automatically stubbed.
-# false: Disabled (default), true: Enabled
-use_auto_stub =
-# Enables/Disables the macro JIT compiler
-disable_macro_jit=false
-# Determines whether to enable the GDB stub and wait for the debugger to attach before running.
-# false: Disabled (default), true: Enabled
-use_gdbstub=false
-# The port to use for the GDB server, if it is enabled.
-gdbstub_port=6543
-
-[WebService]
-# Whether or not to enable telemetry
-# 0: No, 1 (default): Yes
-enable_telemetry =
-# URL for Web API
-web_api_url = https://api.yuzu-emu.org
-# Username and token for yuzu Web Service
-# See https://profile.yuzu-emu.org/ for more info
-yuzu_username =
-yuzu_token =
-
-[Network]
-# Name of the network interface device to use with yuzu LAN play.
-# e.g. On *nix: 'enp7s0', 'wlp6s0u1u3u3', 'lo'
-# e.g. On Windows: 'Ethernet', 'Wi-Fi'
-network_interface =
-
-[AddOns]
-# Used to disable add-ons
-# List of title IDs of games that will have add-ons disabled (separated by '|'):
-title_ids =
-# For each title ID, have a key/value pair called `disabled_<title_id>` equal to the names of the add-ons to disable (sep. by '|')
-# e.x. disabled_0100000000010000 = Update|DLC <- disables Updates and DLC on Super Mario Odyssey
-)";
-} // namespace DefaultINI

src/android/app/src/main/jni/native.cpp

@@ -52,8 +52,8 @@
 #include "core/hle/service/am/applets/applets.h"
 #include "core/hle/service/filesystem/filesystem.h"
 #include "core/loader/loader.h"
+#include "frontend_common/config.h"
 #include "jni/android_common/android_common.h"
-#include "jni/config.h"
 #include "jni/id_cache.h"
 #include "jni/native.h"
 #include "video_core/renderer_base.h"
@@ -664,8 +664,6 @@ void Java_org_yuzu_yuzu_1emu_NativeLibrary_onTouchReleased(JNIEnv* env, jclass c
 
 void Java_org_yuzu_yuzu_1emu_NativeLibrary_initializeSystem(JNIEnv* env, jclass clazz,
                                                             jboolean reload) {
-    // Create the default config.ini.
-    Config{};
     // Initialize the emulated system.
     if (!reload) {
         EmulationSession::GetInstance().System().Initialize();
@@ -680,17 +678,6 @@ jint Java_org_yuzu_yuzu_1emu_NativeLibrary_defaultCPUCore(JNIEnv* env, jclass cl
 void Java_org_yuzu_yuzu_1emu_NativeLibrary_run__Ljava_lang_String_2Ljava_lang_String_2Z(
     JNIEnv* env, jclass clazz, jstring j_file, jstring j_savestate, jboolean j_delete_savestate) {}
 
-void Java_org_yuzu_yuzu_1emu_NativeLibrary_reloadSettings(JNIEnv* env, jclass clazz) {
-    Config{};
-}
-
-void Java_org_yuzu_yuzu_1emu_NativeLibrary_initGameIni(JNIEnv* env, jclass clazz,
-                                                       jstring j_game_id) {
-    std::string_view game_id = env->GetStringUTFChars(j_game_id, 0);
-
-    env->ReleaseStringUTFChars(j_game_id, game_id.data());
-}
-
 jdoubleArray Java_org_yuzu_yuzu_1emu_NativeLibrary_getPerfStats(JNIEnv* env, jclass clazz) {
     jdoubleArray j_stats = env->NewDoubleArray(4);
 
@@ -707,6 +694,14 @@ jdoubleArray Java_org_yuzu_yuzu_1emu_NativeLibrary_getPerfStats(JNIEnv* env, jcl
     return j_stats;
 }
 
+jstring Java_org_yuzu_yuzu_1emu_NativeLibrary_getCpuBackend(JNIEnv* env, jclass clazz) {
+    if (Settings::IsNceEnabled()) {
+        return ToJString(env, "NCE");
+    }
+
+    return ToJString(env, "JIT");
+}
+
 void Java_org_yuzu_yuzu_1emu_utils_DirectoryInitialization_setSysDirectory(JNIEnv* env,
                                                                            jclass clazz,
                                                                            jstring j_path) {}

src/android/app/src/main/jni/native_config.cpp

@@ -5,11 +5,14 @@
 
 #include <jni.h>
 
+#include "android_config.h"
+#include "android_settings.h"
 #include "common/logging/log.h"
 #include "common/settings.h"
+#include "frontend_common/config.h"
 #include "jni/android_common/android_common.h"
-#include "jni/config.h"
-#include "uisettings.h"
+
+std::unique_ptr<AndroidConfig> config;
 
 template <typename T>
 Settings::Setting<T>* getSetting(JNIEnv* env, jstring jkey) {
@@ -28,6 +31,22 @@ Settings::Setting<T>* getSetting(JNIEnv* env, jstring jkey) {
 
 extern "C" {
 
+void Java_org_yuzu_yuzu_1emu_utils_NativeConfig_initializeConfig(JNIEnv* env, jobject obj) {
+    config = std::make_unique<AndroidConfig>();
+}
+
+void Java_org_yuzu_yuzu_1emu_utils_NativeConfig_unloadConfig(JNIEnv* env, jobject obj) {
+    config.reset();
+}
+
+void Java_org_yuzu_yuzu_1emu_utils_NativeConfig_reloadSettings(JNIEnv* env, jobject obj) {
+    config->AndroidConfig::ReloadAllValues();
+}
+
+void Java_org_yuzu_yuzu_1emu_utils_NativeConfig_saveSettings(JNIEnv* env, jobject obj) {
+    config->AndroidConfig::SaveAllValues();
+}
+
 jboolean Java_org_yuzu_yuzu_1emu_utils_NativeConfig_getBoolean(JNIEnv* env, jobject obj,
                                                                jstring jkey, jboolean getDefault) {
     auto setting = getSetting<bool>(env, jkey);

src/android/app/src/main/res/values/arrays.xml

@@ -175,6 +175,24 @@
         <item>2</item>
     </integer-array>
 
+    <string-array name="cpuBackendArm64Names">
+        <item>@string/cpu_backend_dynarmic</item>
+        <item>@string/cpu_backend_nce</item>
+    </string-array>
+
+    <integer-array name="cpuBackendArm64Values">
+        <item>0</item>
+        <item>1</item>
+    </integer-array>
+
+    <string-array name="cpuBackendX86Names">
+        <item>@string/cpu_backend_dynarmic</item>
+    </string-array>
+
+    <integer-array name="cpuBackendX86Values">
+        <item>0</item>
+    </integer-array>
+
     <string-array name="cpuAccuracyNames">
         <item>@string/auto</item>
         <item>@string/cpu_accuracy_accurate</item>

src/android/app/src/main/res/values/strings.xml

@@ -185,6 +185,7 @@
     <string name="frame_limit_enable_description">Limits emulation speed to a specified percentage of normal speed.</string>
     <string name="frame_limit_slider">Limit speed percent</string>
     <string name="frame_limit_slider_description">Specifies the percentage to limit emulation speed. 100% is the normal speed. Values higher or lower will increase or decrease the speed limit.</string>
+    <string name="cpu_backend">CPU backend</string>
     <string name="cpu_accuracy">CPU accuracy</string>
     <string name="value_with_units">%1$s%2$s</string>
 
@@ -416,6 +417,10 @@
     <string name="ratio_force_sixteen_ten">Force 16:10</string>
     <string name="ratio_stretch">Stretch to window</string>
 
+    <!-- CPU Backend -->
+    <string name="cpu_backend_dynarmic">Dynarmic (Slow)</string>
+    <string name="cpu_backend_nce">Native code execution (NCE)</string>
+
     <!-- CPU Accuracy -->
     <string name="cpu_accuracy_accurate">Accurate</string>
     <string name="cpu_accuracy_unsafe">Unsafe</string>

src/common/CMakeLists.txt

@@ -52,6 +52,7 @@ add_library(common STATIC
     fiber.cpp
     fiber.h
     fixed_point.h
+    free_region_manager.h
     fs/file.cpp
     fs/file.h
     fs/fs.cpp
@@ -166,6 +167,13 @@ if (WIN32)
   target_link_libraries(common PRIVATE ntdll)
 endif()
 
+if (NOT WIN32)
+  target_sources(common PRIVATE
+    signal_chain.cpp
+    signal_chain.h
+  )
+endif()
+
 if(ANDROID)
     target_sources(common
         PRIVATE

src/common/free_region_manager.h

@@ -0,0 +1,55 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <mutex>
+#include <boost/icl/interval_set.hpp>
+
+namespace Common {
+
+class FreeRegionManager {
+public:
+    explicit FreeRegionManager() = default;
+    ~FreeRegionManager() = default;
+
+    void SetAddressSpace(void* start, size_t size) {
+        this->FreeBlock(start, size);
+    }
+
+    std::pair<void*, size_t> FreeBlock(void* block_ptr, size_t size) {
+        std::scoped_lock lk(m_mutex);
+
+        // Check to see if we are adjacent to any regions.
+        auto start_address = reinterpret_cast<uintptr_t>(block_ptr);
+        auto end_address = start_address + size;
+        auto it = m_free_regions.find({start_address - 1, end_address + 1});
+
+        // If we are, join with them, ensuring we stay in bounds.
+        if (it != m_free_regions.end()) {
+            start_address = std::min(start_address, it->lower());
+            end_address = std::max(end_address, it->upper());
+        }
+
+        // Free the relevant region.
+        m_free_regions.insert({start_address, end_address});
+
+        // Return the adjusted pointers.
+        block_ptr = reinterpret_cast<void*>(start_address);
+        size = end_address - start_address;
+        return {block_ptr, size};
+    }
+
+    void AllocateBlock(void* block_ptr, size_t size) {
+        std::scoped_lock lk(m_mutex);
+
+        auto address = reinterpret_cast<uintptr_t>(block_ptr);
+        m_free_regions.subtract({address, address + size});
+    }
+
+private:
+    std::mutex m_mutex;
+    boost::icl::interval_set<uintptr_t> m_free_regions;
+};
+
+} // namespace Common

src/common/host_memory.cpp

@@ -21,15 +21,18 @@
 #include <boost/icl/interval_set.hpp>
 #include <fcntl.h>
 #include <sys/mman.h>
+#include <sys/random.h>
 #include <unistd.h>
 #include "common/scope_exit.h"
 
 #endif // ^^^ Linux ^^^
 
 #include <mutex>
+#include <random>
 
 #include "common/alignment.h"
 #include "common/assert.h"
+#include "common/free_region_manager.h"
 #include "common/host_memory.h"
 #include "common/logging/log.h"
 
@@ -141,7 +144,7 @@ public:
         Release();
     }
 
-    void Map(size_t virtual_offset, size_t host_offset, size_t length) {
+    void Map(size_t virtual_offset, size_t host_offset, size_t length, MemoryPermission perms) {
         std::unique_lock lock{placeholder_mutex};
         if (!IsNiechePlaceholder(virtual_offset, length)) {
             Split(virtual_offset, length);
@@ -160,7 +163,7 @@ public:
         }
     }
 
-    void Protect(size_t virtual_offset, size_t length, bool read, bool write) {
+    void Protect(size_t virtual_offset, size_t length, bool read, bool write, bool execute) {
         DWORD new_flags{};
         if (read && write) {
             new_flags = PAGE_READWRITE;
@@ -186,6 +189,11 @@ public:
         }
     }
 
+    void EnableDirectMappedAddress() {
+        // TODO
+        UNREACHABLE();
+    }
+
     const size_t backing_size; ///< Size of the backing memory in bytes
     const size_t virtual_size; ///< Size of the virtual address placeholder in bytes
 
@@ -353,6 +361,55 @@ private:
 
 #elif defined(__linux__) || defined(__FreeBSD__) // ^^^ Windows ^^^ vvv Linux vvv
 
+#ifdef ARCHITECTURE_arm64
+
+static void* ChooseVirtualBase(size_t virtual_size) {
+    constexpr uintptr_t Map39BitSize = (1ULL << 39);
+    constexpr uintptr_t Map36BitSize = (1ULL << 36);
+
+    // This is not a cryptographic application, we just want something random.
+    std::mt19937_64 rng;
+
+    // We want to ensure we are allocating at an address aligned to the L2 block size.
+    // For Qualcomm devices, we must also allocate memory above 36 bits.
+    const size_t lower = Map36BitSize / HugePageSize;
+    const size_t upper = (Map39BitSize - virtual_size) / HugePageSize;
+    const size_t range = upper - lower;
+
+    // Try up to 64 times to allocate memory at random addresses in the range.
+    for (int i = 0; i < 64; i++) {
+        // Calculate a possible location.
+        uintptr_t hint_address = ((rng() % range) + lower) * HugePageSize;
+
+        // Try to map.
+        // Note: we may be able to take advantage of MAP_FIXED_NOREPLACE here.
+        void* map_pointer =
+            mmap(reinterpret_cast<void*>(hint_address), virtual_size, PROT_READ | PROT_WRITE,
+                 MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0);
+
+        // If we successfully mapped, we're done.
+        if (reinterpret_cast<uintptr_t>(map_pointer) == hint_address) {
+            return map_pointer;
+        }
+
+        // Unmap if necessary, and try again.
+        if (map_pointer != MAP_FAILED) {
+            munmap(map_pointer, virtual_size);
+        }
+    }
+
+    return MAP_FAILED;
+}
+
+#else
+
+static void* ChooseVirtualBase(size_t virtual_size) {
+    return mmap(nullptr, virtual_size, PROT_READ | PROT_WRITE,
+                MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0);
+}
+
+#endif
+
 class HostMemory::Impl {
 public:
     explicit Impl(size_t backing_size_, size_t virtual_size_)
@@ -415,8 +472,7 @@ public:
             }
         }
 #else
-        virtual_base = static_cast<u8*>(mmap(nullptr, virtual_size, PROT_NONE,
-                                             MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0));
+        virtual_base = virtual_map_base = static_cast<u8*>(ChooseVirtualBase(virtual_size));
         if (virtual_base == MAP_FAILED) {
             LOG_CRITICAL(HW_Memory, "mmap failed: {}", strerror(errno));
             throw std::bad_alloc{};
@@ -424,7 +480,7 @@ public:
         madvise(virtual_base, virtual_size, MADV_HUGEPAGE);
 #endif
 
-        placeholders.add({0, virtual_size});
+        free_manager.SetAddressSpace(virtual_base, virtual_size);
         good = true;
     }
 
@@ -432,14 +488,29 @@ public:
         Release();
     }
 
-    void Map(size_t virtual_offset, size_t host_offset, size_t length) {
-        {
-            std::scoped_lock lock{placeholder_mutex};
-            placeholders.subtract({virtual_offset, virtual_offset + length});
-        }
+    void Map(size_t virtual_offset, size_t host_offset, size_t length, MemoryPermission perms) {
+        // Intersect the range with our address space.
+        AdjustMap(&virtual_offset, &length);
 
-        void* ret = mmap(virtual_base + virtual_offset, length, PROT_READ | PROT_WRITE,
-                         MAP_SHARED | MAP_FIXED, fd, host_offset);
+        // We are removing a placeholder.
+        free_manager.AllocateBlock(virtual_base + virtual_offset, length);
+
+        // Deduce mapping protection flags.
+        int flags = PROT_NONE;
+        if (True(perms & MemoryPermission::Read)) {
+            flags |= PROT_READ;
+        }
+        if (True(perms & MemoryPermission::Write)) {
+            flags |= PROT_WRITE;
+        }
+#ifdef ARCHITECTURE_arm64
+        if (True(perms & MemoryPermission::Execute)) {
+            flags |= PROT_EXEC;
+        }
+#endif
+
+        void* ret = mmap(virtual_base + virtual_offset, length, flags, MAP_SHARED | MAP_FIXED, fd,
+                         host_offset);
         ASSERT_MSG(ret != MAP_FAILED, "mmap failed: {}", strerror(errno));
     }
 
@@ -447,47 +518,54 @@ public:
         // The method name is wrong. We're still talking about the virtual range.
         // We don't want to unmap, we want to reserve this memory.
 
-        {
-            std::scoped_lock lock{placeholder_mutex};
-            auto it = placeholders.find({virtual_offset - 1, virtual_offset + length + 1});
+        // Intersect the range with our address space.
+        AdjustMap(&virtual_offset, &length);
 
-            if (it != placeholders.end()) {
-                size_t prev_upper = virtual_offset + length;
-                virtual_offset = std::min(virtual_offset, it->lower());
-                length = std::max(it->upper(), prev_upper) - virtual_offset;
-            }
+        // Merge with any adjacent placeholder mappings.
+        auto [merged_pointer, merged_size] =
+            free_manager.FreeBlock(virtual_base + virtual_offset, length);
 
-            placeholders.add({virtual_offset, virtual_offset + length});
-        }
-
-        void* ret = mmap(virtual_base + virtual_offset, length, PROT_NONE,
+        void* ret = mmap(merged_pointer, merged_size, PROT_NONE,
                          MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
         ASSERT_MSG(ret != MAP_FAILED, "mmap failed: {}", strerror(errno));
     }
 
-    void Protect(size_t virtual_offset, size_t length, bool read, bool write) {
-        int flags = 0;
+    void Protect(size_t virtual_offset, size_t length, bool read, bool write, bool execute) {
+        // Intersect the range with our address space.
+        AdjustMap(&virtual_offset, &length);
+
+        int flags = PROT_NONE;
         if (read) {
             flags |= PROT_READ;
         }
         if (write) {
             flags |= PROT_WRITE;
         }
+#ifdef ARCHITECTURE_arm64
+        if (execute) {
+            flags |= PROT_EXEC;
+        }
+#endif
         int ret = mprotect(virtual_base + virtual_offset, length, flags);
         ASSERT_MSG(ret == 0, "mprotect failed: {}", strerror(errno));
     }
 
+    void EnableDirectMappedAddress() {
+        virtual_base = nullptr;
+    }
+
     const size_t backing_size; ///< Size of the backing memory in bytes
     const size_t virtual_size; ///< Size of the virtual address placeholder in bytes
 
     u8* backing_base{reinterpret_cast<u8*>(MAP_FAILED)};
     u8* virtual_base{reinterpret_cast<u8*>(MAP_FAILED)};
+    u8* virtual_map_base{reinterpret_cast<u8*>(MAP_FAILED)};
 
 private:
     /// Release all resources in the object
     void Release() {
-        if (virtual_base != MAP_FAILED) {
-            int ret = munmap(virtual_base, virtual_size);
+        if (virtual_map_base != MAP_FAILED) {
+            int ret = munmap(virtual_map_base, virtual_size);
             ASSERT_MSG(ret == 0, "munmap failed: {}", strerror(errno));
         }
 
@@ -502,10 +580,29 @@ private:
         }
     }
 
-    int fd{-1}; // memfd file descriptor, -1 is the error value of memfd_create
+    void AdjustMap(size_t* virtual_offset, size_t* length) {
+        if (virtual_base != nullptr) {
+            return;
+        }
 
-    boost::icl::interval_set<size_t> placeholders; ///< Mapped placeholders
-    std::mutex placeholder_mutex;                  ///< Mutex for placeholders
+        // If we are direct mapped, we want to make sure we are operating on a region
+        // that is in range of our virtual mapping.
+        size_t intended_start = *virtual_offset;
+        size_t intended_end = intended_start + *length;
+        size_t address_space_start = reinterpret_cast<size_t>(virtual_map_base);
+        size_t address_space_end = address_space_start + virtual_size;
+
+        if (address_space_start > intended_end || intended_start > address_space_end) {
+            *virtual_offset = 0;
+            *length = 0;
+        } else {
+            *virtual_offset = std::max(intended_start, address_space_start);
+            *length = std::min(intended_end, address_space_end) - *virtual_offset;
+        }
+    }
+
+    int fd{-1}; // memfd file descriptor, -1 is the error value of memfd_create
+    FreeRegionManager free_manager{};
 };
 
 #else // ^^^ Linux ^^^ vvv Generic vvv
@@ -518,11 +615,13 @@ public:
         throw std::bad_alloc{};
     }
 
-    void Map(size_t virtual_offset, size_t host_offset, size_t length) {}
+    void Map(size_t virtual_offset, size_t host_offset, size_t length, MemoryPermission perm) {}
 
     void Unmap(size_t virtual_offset, size_t length) {}
 
-    void Protect(size_t virtual_offset, size_t length, bool read, bool write) {}
+    void Protect(size_t virtual_offset, size_t length, bool read, bool write, bool execute) {}
+
+    void EnableDirectMappedAddress() {}
 
     u8* backing_base{nullptr};
     u8* virtual_base{nullptr};
@@ -535,15 +634,16 @@ HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_)
     try {
         // Try to allocate a fastmem arena.
         // The implementation will fail with std::bad_alloc on errors.
-        impl = std::make_unique<HostMemory::Impl>(AlignUp(backing_size, PageAlignment),
-                                                  AlignUp(virtual_size, PageAlignment) +
-                                                      3 * HugePageSize);
+        impl =
+            std::make_unique<HostMemory::Impl>(AlignUp(backing_size, PageAlignment),
+                                               AlignUp(virtual_size, PageAlignment) + HugePageSize);
         backing_base = impl->backing_base;
         virtual_base = impl->virtual_base;
 
         if (virtual_base) {
-            virtual_base += 2 * HugePageSize - 1;
-            virtual_base -= reinterpret_cast<size_t>(virtual_base) & (HugePageSize - 1);
+            // Ensure the virtual base is aligned to the L2 block size.
+            virtual_base = reinterpret_cast<u8*>(
+                Common::AlignUp(reinterpret_cast<uintptr_t>(virtual_base), HugePageSize));
             virtual_base_offset = virtual_base - impl->virtual_base;
         }
 
@@ -562,7 +662,8 @@ HostMemory::HostMemory(HostMemory&&) noexcept = default;
 
 HostMemory& HostMemory::operator=(HostMemory&&) noexcept = default;
 
-void HostMemory::Map(size_t virtual_offset, size_t host_offset, size_t length) {
+void HostMemory::Map(size_t virtual_offset, size_t host_offset, size_t length,
+                     MemoryPermission perms) {
     ASSERT(virtual_offset % PageAlignment == 0);
     ASSERT(host_offset % PageAlignment == 0);
     ASSERT(length % PageAlignment == 0);
@@ -571,7 +672,7 @@ void HostMemory::Map(size_t virtual_offset, size_t host_offset, size_t length) {
     if (length == 0 || !virtual_base || !impl) {
         return;
     }
-    impl->Map(virtual_offset + virtual_base_offset, host_offset, length);
+    impl->Map(virtual_offset + virtual_base_offset, host_offset, length, perms);
 }
 
 void HostMemory::Unmap(size_t virtual_offset, size_t length) {
@@ -584,14 +685,22 @@ void HostMemory::Unmap(size_t virtual_offset, size_t length) {
     impl->Unmap(virtual_offset + virtual_base_offset, length);
 }
 
-void HostMemory::Protect(size_t virtual_offset, size_t length, bool read, bool write) {
+void HostMemory::Protect(size_t virtual_offset, size_t length, bool read, bool write,
+                         bool execute) {
     ASSERT(virtual_offset % PageAlignment == 0);
     ASSERT(length % PageAlignment == 0);
     ASSERT(virtual_offset + length <= virtual_size);
     if (length == 0 || !virtual_base || !impl) {
         return;
     }
-    impl->Protect(virtual_offset + virtual_base_offset, length, read, write);
+    impl->Protect(virtual_offset + virtual_base_offset, length, read, write, execute);
+}
+
+void HostMemory::EnableDirectMappedAddress() {
+    if (impl) {
+        impl->EnableDirectMappedAddress();
+        virtual_size += reinterpret_cast<uintptr_t>(virtual_base);
+    }
 }
 
 } // namespace Common

src/common/host_memory.h

@@ -4,11 +4,20 @@
 #pragma once
 
 #include <memory>
+#include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/virtual_buffer.h"
 
 namespace Common {
 
+enum class MemoryPermission : u32 {
+    Read = 1 << 0,
+    Write = 1 << 1,
+    ReadWrite = Read | Write,
+    Execute = 1 << 2,
+};
+DECLARE_ENUM_FLAG_OPERATORS(MemoryPermission)
+
 /**
  * A low level linear memory buffer, which supports multiple mappings
  * Its purpose is to rebuild a given sparse memory layout, including mirrors.
@@ -31,11 +40,13 @@ public:
     HostMemory(HostMemory&& other) noexcept;
     HostMemory& operator=(HostMemory&& other) noexcept;
 
-    void Map(size_t virtual_offset, size_t host_offset, size_t length);
+    void Map(size_t virtual_offset, size_t host_offset, size_t length, MemoryPermission perms);
 
     void Unmap(size_t virtual_offset, size_t length);
 
-    void Protect(size_t virtual_offset, size_t length, bool read, bool write);
+    void Protect(size_t virtual_offset, size_t length, bool read, bool write, bool execute = false);
+
+    void EnableDirectMappedAddress();
 
     [[nodiscard]] u8* BackingBasePointer() noexcept {
         return backing_base;

src/common/settings.cpp

@@ -41,6 +41,7 @@ SWITCHABLE(AspectRatio, true);
 SWITCHABLE(AstcDecodeMode, true);
 SWITCHABLE(AstcRecompression, true);
 SWITCHABLE(AudioMode, true);
+SWITCHABLE(CpuBackend, true);
 SWITCHABLE(CpuAccuracy, true);
 SWITCHABLE(FullscreenMode, true);
 SWITCHABLE(GpuAccuracy, true);
@@ -155,6 +156,22 @@ bool IsFastmemEnabled() {
     return true;
 }
 
+static bool is_nce_enabled = false;
+
+void SetNceEnabled(bool is_39bit) {
+    const bool is_nce_selected = values.cpu_backend.GetValue() == CpuBackend::Nce;
+    is_nce_enabled = IsFastmemEnabled() && is_nce_selected && is_39bit;
+    if (is_nce_selected && !is_nce_enabled) {
+        LOG_WARNING(
+            Common,
+            "Program does not utilize 39-bit address space, unable to natively execute code");
+    }
+}
+
+bool IsNceEnabled() {
+    return is_nce_enabled;
+}
+
 bool IsDockedMode() {
     return values.use_docked_mode.GetValue() == Settings::ConsoleMode::Docked;
 }
@@ -206,9 +223,9 @@ const char* TranslateCategory(Category category) {
     case Category::UiAudio:
         return "UiAudio";
     case Category::UiLayout:
-        return "UiLayout";
+        return "UILayout";
     case Category::UiGameList:
-        return "UiGameList";
+        return "UIGameList";
     case Category::Screenshots:
         return "Screenshots";
     case Category::Shortcuts:

src/common/settings.h

@@ -63,6 +63,7 @@ SWITCHABLE(AspectRatio, true);
 SWITCHABLE(AstcDecodeMode, true);
 SWITCHABLE(AstcRecompression, true);
 SWITCHABLE(AudioMode, true);
+SWITCHABLE(CpuBackend, true);
 SWITCHABLE(CpuAccuracy, true);
 SWITCHABLE(FullscreenMode, true);
 SWITCHABLE(GpuAccuracy, true);
@@ -179,6 +180,14 @@ struct Values {
                                              &use_speed_limit};
 
     // Cpu
+    SwitchableSetting<CpuBackend, true> cpu_backend{
+        linkage,         CpuBackend::Dynarmic, CpuBackend::Dynarmic,
+#ifdef ARCHITECTURE_arm64
+        CpuBackend::Nce,
+#else
+                                                    CpuBackend::Dynarmic,
+#endif
+        "cpu_backend",   Category::Cpu};
     SwitchableSetting<CpuAccuracy, true> cpu_accuracy{linkage,           CpuAccuracy::Auto,
                                                       CpuAccuracy::Auto, CpuAccuracy::Paranoid,
                                                       "cpu_accuracy",    Category::Cpu};
@@ -232,7 +241,11 @@ struct Values {
     SwitchableSetting<bool> use_asynchronous_gpu_emulation{
         linkage, true, "use_asynchronous_gpu_emulation", Category::Renderer};
     SwitchableSetting<AstcDecodeMode, true> accelerate_astc{linkage,
+#ifdef ANDROID
+                                                            AstcDecodeMode::Cpu,
+#else
                                                             AstcDecodeMode::Gpu,
+#endif
                                                             AstcDecodeMode::Cpu,
                                                             AstcDecodeMode::CpuAsynchronous,
                                                             "accelerate_astc",
@@ -304,7 +317,11 @@ struct Values {
         linkage, 0, "bg_blue", Category::Renderer, Specialization::Default, true, true};
 
     SwitchableSetting<GpuAccuracy, true> gpu_accuracy{linkage,
+#ifdef ANDROID
+                                                      GpuAccuracy::Normal,
+#else
                                                       GpuAccuracy::High,
+#endif
                                                       GpuAccuracy::Normal,
                                                       GpuAccuracy::Extreme,
                                                       "gpu_accuracy",
@@ -313,20 +330,38 @@ struct Values {
                                                       true,
                                                       true};
     GpuAccuracy current_gpu_accuracy{GpuAccuracy::High};
-    SwitchableSetting<AnisotropyMode, true> max_anisotropy{
-        linkage,          AnisotropyMode::Automatic, AnisotropyMode::Automatic, AnisotropyMode::X16,
-        "max_anisotropy", Category::RendererAdvanced};
+    SwitchableSetting<AnisotropyMode, true> max_anisotropy{linkage,
+#ifdef ANDROID
+                                                           AnisotropyMode::Default,
+#else
+                                                           AnisotropyMode::Automatic,
+#endif
+                                                           AnisotropyMode::Automatic,
+                                                           AnisotropyMode::X16,
+                                                           "max_anisotropy",
+                                                           Category::RendererAdvanced};
     SwitchableSetting<AstcRecompression, true> astc_recompression{linkage,
                                                                   AstcRecompression::Uncompressed,
                                                                   AstcRecompression::Uncompressed,
                                                                   AstcRecompression::Bc3,
                                                                   "astc_recompression",
                                                                   Category::RendererAdvanced};
-    SwitchableSetting<bool> async_presentation{linkage, false, "async_presentation",
-                                               Category::RendererAdvanced};
+    SwitchableSetting<bool> async_presentation{linkage,
+#ifdef ANDROID
+                                               true,
+#else
+                                               false,
+#endif
+                                               "async_presentation", Category::RendererAdvanced};
     SwitchableSetting<bool> renderer_force_max_clock{linkage, false, "force_max_clock",
                                                      Category::RendererAdvanced};
-    SwitchableSetting<bool> use_reactive_flushing{linkage, true, "use_reactive_flushing",
+    SwitchableSetting<bool> use_reactive_flushing{linkage,
+#ifdef ANDROID
+                                                  false,
+#else
+                                                  true,
+#endif
+                                                  "use_reactive_flushing",
                                                   Category::RendererAdvanced};
     SwitchableSetting<bool> use_asynchronous_shaders{linkage, false, "use_asynchronous_shaders",
                                                      Category::RendererAdvanced};
@@ -358,6 +393,8 @@ struct Values {
                                           Category::RendererDebug};
     // TODO: remove this once AMDVLK supports VK_EXT_depth_bias_control
     bool renderer_amdvlk_depth_bias_workaround{};
+    Setting<bool> disable_buffer_reorder{linkage, false, "disable_buffer_reorder",
+                                         Category::RendererDebug};
 
     // System
     SwitchableSetting<Language, true> language_index{linkage,
@@ -390,7 +427,11 @@ struct Values {
     Setting<s32> current_user{linkage, 0, "current_user", Category::System};
 
     SwitchableSetting<ConsoleMode> use_docked_mode{linkage,
+#ifdef ANDROID
+                                                   ConsoleMode::Handheld,
+#else
                                                    ConsoleMode::Docked,
+#endif
                                                    "use_docked_mode",
                                                    Category::System,
                                                    Specialization::Radio,
@@ -534,6 +575,8 @@ bool IsGPULevelExtreme();
 bool IsGPULevelHigh();
 
 bool IsFastmemEnabled();
+void SetNceEnabled(bool is_64bit);
+bool IsNceEnabled();
 
 bool IsDockedMode();
 

src/common/settings_enums.h

@@ -129,6 +129,8 @@ ENUM(ShaderBackend, Glsl, Glasm, SpirV);
 
 ENUM(GpuAccuracy, Normal, High, Extreme);
 
+ENUM(CpuBackend, Dynarmic, Nce);
+
 ENUM(CpuAccuracy, Auto, Accurate, Unsafe, Paranoid);
 
 ENUM(MemoryLayout, Memory_4Gb, Memory_6Gb, Memory_8Gb);

src/common/signal_chain.cpp

@@ -0,0 +1,42 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <dlfcn.h>
+
+#include "common/assert.h"
+#include "common/dynamic_library.h"
+#include "common/scope_exit.h"
+#include "common/signal_chain.h"
+
+namespace Common {
+
+template <typename T>
+T* LookupLibcSymbol(const char* name) {
+#if defined(__BIONIC__)
+    Common::DynamicLibrary provider("libc.so");
+    if (!provider.IsOpen()) {
+        UNREACHABLE_MSG("Failed to open libc!");
+    }
+#else
+    // For other operating environments, we assume the symbol is not overridden.
+    const char* base = nullptr;
+    Common::DynamicLibrary provider(base);
+#endif
+
+    void* sym = provider.GetSymbolAddress(name);
+    if (sym == nullptr) {
+        sym = dlsym(RTLD_DEFAULT, name);
+    }
+    if (sym == nullptr) {
+        UNREACHABLE_MSG("Unable to find symbol {}!", name);
+    }
+
+    return reinterpret_cast<T*>(sym);
+}
+
+int SigAction(int signum, const struct sigaction* act, struct sigaction* oldact) {
+    static auto libc_sigaction = LookupLibcSymbol<decltype(sigaction)>("sigaction");
+    return libc_sigaction(signum, act, oldact);
+}
+
+} // namespace Common

src/common/signal_chain.h

@@ -0,0 +1,19 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#ifndef _WIN32
+
+#include <signal.h>
+
+namespace Common {
+
+// Android's ART overrides sigaction with its own wrapper. This is problematic for SIGSEGV
+// in particular, because ART's handler accesses tpidr_el0, which conflicts with NCE.
+// This extracts the libc symbol and calls it directly.
+int SigAction(int signum, const struct sigaction* act, struct sigaction* oldact);
+
+} // namespace Common
+
+#endif

src/core/CMakeLists.txt

@@ -926,6 +926,23 @@ if (ENABLE_WEB_SERVICE)
     target_link_libraries(core PRIVATE web_service)
 endif()
 
+if (ARCHITECTURE_arm64 AND (ANDROID OR ${CMAKE_SYSTEM_NAME} STREQUAL "Linux"))
+    target_compile_definitions(core PRIVATE -DHAS_NCE)
+    enable_language(C ASM)
+    set(CMAKE_ASM_FLAGS "${CFLAGS} -x assembler-with-cpp")
+
+    target_sources(core PRIVATE
+        arm/nce/arm_nce.cpp
+        arm/nce/arm_nce.h
+        arm/nce/arm_nce.s
+        arm/nce/guest_context.h
+        arm/nce/patch.cpp
+        arm/nce/patch.h
+        arm/nce/instructions.h
+    )
+    target_link_libraries(core PRIVATE merry::oaknut)
+endif()
+
 if (ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64)
     target_sources(core PRIVATE
         arm/dynarmic/arm_dynarmic.h

src/core/arm/arm_interface.cpp

@@ -201,6 +201,8 @@ void ARM_Interface::Run() {
         if (True(hr & HaltReason::DataAbort)) {
             if (system.DebuggerEnabled()) {
                 system.GetDebugger().NotifyThreadWatchpoint(current_thread, *HaltedWatchpoint());
+            } else {
+                LogBacktrace();
             }
             current_thread->RequestSuspend(SuspendType::Debug);
             break;

src/core/arm/arm_interface.h

@@ -81,6 +81,9 @@ public:
     // thread context to be 800 bytes in size.
     static_assert(sizeof(ThreadContext64) == 0x320);
 
+    /// Perform any backend-specific initialization.
+    virtual void Initialize() {}
+
     /// Runs the CPU until an event happens
     void Run();
 

src/core/arm/nce/arm_nce.cpp

@@ -0,0 +1,400 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <cinttypes>
+#include <memory>
+
+#include "common/signal_chain.h"
+#include "core/arm/nce/arm_nce.h"
+#include "core/arm/nce/patch.h"
+#include "core/core.h"
+#include "core/memory.h"
+
+#include "core/hle/kernel/k_process.h"
+
+#include <signal.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+
+namespace Core {
+
+namespace {
+
+struct sigaction g_orig_action;
+
+// Verify assembly offsets.
+using NativeExecutionParameters = Kernel::KThread::NativeExecutionParameters;
+static_assert(offsetof(NativeExecutionParameters, native_context) == TpidrEl0NativeContext);
+static_assert(offsetof(NativeExecutionParameters, lock) == TpidrEl0Lock);
+static_assert(offsetof(NativeExecutionParameters, magic) == TpidrEl0TlsMagic);
+
+fpsimd_context* GetFloatingPointState(mcontext_t& host_ctx) {
+    _aarch64_ctx* header = reinterpret_cast<_aarch64_ctx*>(&host_ctx.__reserved);
+    while (header->magic != FPSIMD_MAGIC) {
+        header = reinterpret_cast<_aarch64_ctx*>(reinterpret_cast<char*>(header) + header->size);
+    }
+    return reinterpret_cast<fpsimd_context*>(header);
+}
+
+} // namespace
+
+void* ARM_NCE::RestoreGuestContext(void* raw_context) {
+    // Retrieve the host context.
+    auto& host_ctx = static_cast<ucontext_t*>(raw_context)->uc_mcontext;
+
+    // Thread-local parameters will be located in x9.
+    auto* tpidr = reinterpret_cast<NativeExecutionParameters*>(host_ctx.regs[9]);
+    auto* guest_ctx = static_cast<GuestContext*>(tpidr->native_context);
+
+    // Retrieve the host floating point state.
+    auto* fpctx = GetFloatingPointState(host_ctx);
+
+    // Save host callee-saved registers.
+    std::memcpy(guest_ctx->host_ctx.host_saved_vregs.data(), &fpctx->vregs[8],
+                sizeof(guest_ctx->host_ctx.host_saved_vregs));
+    std::memcpy(guest_ctx->host_ctx.host_saved_regs.data(), &host_ctx.regs[19],
+                sizeof(guest_ctx->host_ctx.host_saved_regs));
+
+    // Save stack pointer.
+    guest_ctx->host_ctx.host_sp = host_ctx.sp;
+
+    // Restore all guest state except tpidr_el0.
+    host_ctx.sp = guest_ctx->sp;
+    host_ctx.pc = guest_ctx->pc;
+    host_ctx.pstate = guest_ctx->pstate;
+    fpctx->fpcr = guest_ctx->fpcr;
+    fpctx->fpsr = guest_ctx->fpsr;
+    std::memcpy(host_ctx.regs, guest_ctx->cpu_registers.data(), sizeof(host_ctx.regs));
+    std::memcpy(fpctx->vregs, guest_ctx->vector_registers.data(), sizeof(fpctx->vregs));
+
+    // Return the new thread-local storage pointer.
+    return tpidr;
+}
+
+void ARM_NCE::SaveGuestContext(GuestContext* guest_ctx, void* raw_context) {
+    // Retrieve the host context.
+    auto& host_ctx = static_cast<ucontext_t*>(raw_context)->uc_mcontext;
+
+    // Retrieve the host floating point state.
+    auto* fpctx = GetFloatingPointState(host_ctx);
+
+    // Save all guest registers except tpidr_el0.
+    std::memcpy(guest_ctx->cpu_registers.data(), host_ctx.regs, sizeof(host_ctx.regs));
+    std::memcpy(guest_ctx->vector_registers.data(), fpctx->vregs, sizeof(fpctx->vregs));
+    guest_ctx->fpsr = fpctx->fpsr;
+    guest_ctx->fpcr = fpctx->fpcr;
+    guest_ctx->pstate = static_cast<u32>(host_ctx.pstate);
+    guest_ctx->pc = host_ctx.pc;
+    guest_ctx->sp = host_ctx.sp;
+
+    // Restore stack pointer.
+    host_ctx.sp = guest_ctx->host_ctx.host_sp;
+
+    // Restore host callee-saved registers.
+    std::memcpy(&host_ctx.regs[19], guest_ctx->host_ctx.host_saved_regs.data(),
+                sizeof(guest_ctx->host_ctx.host_saved_regs));
+    std::memcpy(&fpctx->vregs[8], guest_ctx->host_ctx.host_saved_vregs.data(),
+                sizeof(guest_ctx->host_ctx.host_saved_vregs));
+
+    // Return from the call on exit by setting pc to x30.
+    host_ctx.pc = guest_ctx->host_ctx.host_saved_regs[11];
+
+    // Clear esr_el1 and return it.
+    host_ctx.regs[0] = guest_ctx->esr_el1.exchange(0);
+}
+
+bool ARM_NCE::HandleGuestFault(GuestContext* guest_ctx, void* raw_info, void* raw_context) {
+    auto& host_ctx = static_cast<ucontext_t*>(raw_context)->uc_mcontext;
+    auto* info = static_cast<siginfo_t*>(raw_info);
+
+    // Try to handle an invalid access.
+    // TODO: handle accesses which split a page?
+    const Common::ProcessAddress addr =
+        (reinterpret_cast<u64>(info->si_addr) & ~Memory::YUZU_PAGEMASK);
+    if (guest_ctx->system->ApplicationMemory().InvalidateNCE(addr, Memory::YUZU_PAGESIZE)) {
+        // We handled the access successfully and are returning to guest code.
+        return true;
+    }
+
+    // We can't handle the access, so determine why we crashed.
+    const bool is_prefetch_abort = host_ctx.pc == reinterpret_cast<u64>(info->si_addr);
+
+    // For data aborts, skip the instruction and return to guest code.
+    // This will allow games to continue in many scenarios where they would otherwise crash.
+    if (!is_prefetch_abort) {
+        host_ctx.pc += 4;
+        return true;
+    }
+
+    // This is a prefetch abort.
+    guest_ctx->esr_el1.fetch_or(static_cast<u64>(HaltReason::PrefetchAbort));
+
+    // Forcibly mark the context as locked. We are still running.
+    // We may race with SignalInterrupt here:
+    // - If we lose the race, then SignalInterrupt will send us a signal we are masking,
+    //   and it will do nothing when it is unmasked, as we have already left guest code.
+    // - If we win the race, then SignalInterrupt will wait for us to unlock first.
+    auto& thread_params = guest_ctx->parent->running_thread->GetNativeExecutionParameters();
+    thread_params.lock.store(SpinLockLocked);
+
+    // Return to host.
+    SaveGuestContext(guest_ctx, raw_context);
+    return false;
+}
+
+void ARM_NCE::HandleHostFault(int sig, void* raw_info, void* raw_context) {
+    return g_orig_action.sa_sigaction(sig, static_cast<siginfo_t*>(raw_info), raw_context);
+}
+
+HaltReason ARM_NCE::RunJit() {
+    // Get the thread parameters.
+    // TODO: pass the current thread down from ::Run
+    auto* thread = Kernel::GetCurrentThreadPointer(system.Kernel());
+    auto* thread_params = &thread->GetNativeExecutionParameters();
+
+    {
+        // Lock our core context.
+        std::scoped_lock lk{lock};
+
+        // We should not be running.
+        ASSERT(running_thread == nullptr);
+
+        // Check if we need to run. If we have already been halted, we are done.
+        u64 halt = guest_ctx.esr_el1.exchange(0);
+        if (halt != 0) {
+            return static_cast<HaltReason>(halt);
+        }
+
+        // Mark that we are running.
+        running_thread = thread;
+
+        // Acquire the lock on the thread parameters.
+        // This allows us to force synchronization with SignalInterrupt.
+        LockThreadParameters(thread_params);
+    }
+
+    // Assign current members.
+    guest_ctx.parent = this;
+    thread_params->native_context = &guest_ctx;
+    thread_params->tpidr_el0 = guest_ctx.tpidr_el0;
+    thread_params->tpidrro_el0 = guest_ctx.tpidrro_el0;
+    thread_params->is_running = true;
+
+    HaltReason halt{};
+
+    // TODO: finding and creating the post handler needs to be locked
+    // to deal with dynamic loading of NROs.
+    const auto& post_handlers = system.ApplicationProcess()->GetPostHandlers();
+    if (auto it = post_handlers.find(guest_ctx.pc); it != post_handlers.end()) {
+        halt = ReturnToRunCodeByTrampoline(thread_params, &guest_ctx, it->second);
+    } else {
+        halt = ReturnToRunCodeByExceptionLevelChange(thread_id, thread_params);
+    }
+
+    // Unload members.
+    // The thread does not change, so we can persist the old reference.
+    guest_ctx.tpidr_el0 = thread_params->tpidr_el0;
+    thread_params->native_context = nullptr;
+    thread_params->is_running = false;
+
+    // Unlock the thread parameters.
+    UnlockThreadParameters(thread_params);
+
+    {
+        // Lock the core context.
+        std::scoped_lock lk{lock};
+
+        // On exit, we no longer have an active thread.
+        running_thread = nullptr;
+    }
+
+    // Return the halt reason.
+    return halt;
+}
+
+HaltReason ARM_NCE::StepJit() {
+    return HaltReason::StepThread;
+}
+
+u32 ARM_NCE::GetSvcNumber() const {
+    return guest_ctx.svc_swi;
+}
+
+ARM_NCE::ARM_NCE(System& system_, bool uses_wall_clock_, std::size_t core_index_)
+    : ARM_Interface{system_, uses_wall_clock_}, core_index{core_index_} {
+    guest_ctx.system = &system_;
+}
+
+ARM_NCE::~ARM_NCE() = default;
+
+void ARM_NCE::Initialize() {
+    thread_id = gettid();
+
+    // Setup our signals
+    static std::once_flag flag;
+    std::call_once(flag, [] {
+        using HandlerType = decltype(sigaction::sa_sigaction);
+
+        sigset_t signal_mask;
+        sigemptyset(&signal_mask);
+        sigaddset(&signal_mask, ReturnToRunCodeByExceptionLevelChangeSignal);
+        sigaddset(&signal_mask, BreakFromRunCodeSignal);
+        sigaddset(&signal_mask, GuestFaultSignal);
+
+        struct sigaction return_to_run_code_action {};
+        return_to_run_code_action.sa_flags = SA_SIGINFO | SA_ONSTACK;
+        return_to_run_code_action.sa_sigaction = reinterpret_cast<HandlerType>(
+            &ARM_NCE::ReturnToRunCodeByExceptionLevelChangeSignalHandler);
+        return_to_run_code_action.sa_mask = signal_mask;
+        Common::SigAction(ReturnToRunCodeByExceptionLevelChangeSignal, &return_to_run_code_action,
+                          nullptr);
+
+        struct sigaction break_from_run_code_action {};
+        break_from_run_code_action.sa_flags = SA_SIGINFO | SA_ONSTACK;
+        break_from_run_code_action.sa_sigaction =
+            reinterpret_cast<HandlerType>(&ARM_NCE::BreakFromRunCodeSignalHandler);
+        break_from_run_code_action.sa_mask = signal_mask;
+        Common::SigAction(BreakFromRunCodeSignal, &break_from_run_code_action, nullptr);
+
+        struct sigaction fault_action {};
+        fault_action.sa_flags = SA_SIGINFO | SA_ONSTACK | SA_RESTART;
+        fault_action.sa_sigaction =
+            reinterpret_cast<HandlerType>(&ARM_NCE::GuestFaultSignalHandler);
+        fault_action.sa_mask = signal_mask;
+        Common::SigAction(GuestFaultSignal, &fault_action, &g_orig_action);
+
+        // Simplify call for g_orig_action.
+        // These fields occupy the same space in memory, so this should be a no-op in practice.
+        if (!(g_orig_action.sa_flags & SA_SIGINFO)) {
+            g_orig_action.sa_sigaction =
+                reinterpret_cast<decltype(g_orig_action.sa_sigaction)>(g_orig_action.sa_handler);
+        }
+    });
+}
+
+void ARM_NCE::SetPC(u64 pc) {
+    guest_ctx.pc = pc;
+}
+
+u64 ARM_NCE::GetPC() const {
+    return guest_ctx.pc;
+}
+
+u64 ARM_NCE::GetSP() const {
+    return guest_ctx.sp;
+}
+
+u64 ARM_NCE::GetReg(int index) const {
+    return guest_ctx.cpu_registers[index];
+}
+
+void ARM_NCE::SetReg(int index, u64 value) {
+    guest_ctx.cpu_registers[index] = value;
+}
+
+u128 ARM_NCE::GetVectorReg(int index) const {
+    return guest_ctx.vector_registers[index];
+}
+
+void ARM_NCE::SetVectorReg(int index, u128 value) {
+    guest_ctx.vector_registers[index] = value;
+}
+
+u32 ARM_NCE::GetPSTATE() const {
+    return guest_ctx.pstate;
+}
+
+void ARM_NCE::SetPSTATE(u32 pstate) {
+    guest_ctx.pstate = pstate;
+}
+
+u64 ARM_NCE::GetTlsAddress() const {
+    return guest_ctx.tpidrro_el0;
+}
+
+void ARM_NCE::SetTlsAddress(u64 address) {
+    guest_ctx.tpidrro_el0 = address;
+}
+
+u64 ARM_NCE::GetTPIDR_EL0() const {
+    return guest_ctx.tpidr_el0;
+}
+
+void ARM_NCE::SetTPIDR_EL0(u64 value) {
+    guest_ctx.tpidr_el0 = value;
+}
+
+void ARM_NCE::SaveContext(ThreadContext64& ctx) const {
+    ctx.cpu_registers = guest_ctx.cpu_registers;
+    ctx.sp = guest_ctx.sp;
+    ctx.pc = guest_ctx.pc;
+    ctx.pstate = guest_ctx.pstate;
+    ctx.vector_registers = guest_ctx.vector_registers;
+    ctx.fpcr = guest_ctx.fpcr;
+    ctx.fpsr = guest_ctx.fpsr;
+    ctx.tpidr = guest_ctx.tpidr_el0;
+}
+
+void ARM_NCE::LoadContext(const ThreadContext64& ctx) {
+    guest_ctx.cpu_registers = ctx.cpu_registers;
+    guest_ctx.sp = ctx.sp;
+    guest_ctx.pc = ctx.pc;
+    guest_ctx.pstate = ctx.pstate;
+    guest_ctx.vector_registers = ctx.vector_registers;
+    guest_ctx.fpcr = ctx.fpcr;
+    guest_ctx.fpsr = ctx.fpsr;
+    guest_ctx.tpidr_el0 = ctx.tpidr;
+}
+
+void ARM_NCE::SignalInterrupt() {
+    // Lock core context.
+    std::scoped_lock lk{lock};
+
+    // Add break loop condition.
+    guest_ctx.esr_el1.fetch_or(static_cast<u64>(HaltReason::BreakLoop));
+
+    // If there is no thread running, we are done.
+    if (running_thread == nullptr) {
+        return;
+    }
+
+    // Lock the thread context.
+    auto* params = &running_thread->GetNativeExecutionParameters();
+    LockThreadParameters(params);
+
+    if (params->is_running) {
+        // We should signal to the running thread.
+        // The running thread will unlock the thread context.
+        syscall(SYS_tkill, thread_id, BreakFromRunCodeSignal);
+    } else {
+        // If the thread is no longer running, we have nothing to do.
+        UnlockThreadParameters(params);
+    }
+}
+
+void ARM_NCE::ClearInterrupt() {
+    guest_ctx.esr_el1 = {};
+}
+
+void ARM_NCE::ClearInstructionCache() {
+    // TODO: This is not possible to implement correctly on Linux because
+    // we do not have any access to ic iallu.
+
+    // Require accesses to complete.
+    std::atomic_thread_fence(std::memory_order_seq_cst);
+}
+
+void ARM_NCE::InvalidateCacheRange(u64 addr, std::size_t size) {
+    this->ClearInstructionCache();
+}
+
+void ARM_NCE::ClearExclusiveState() {
+    // No-op.
+}
+
+void ARM_NCE::PageTableChanged(Common::PageTable& page_table,
+                               std::size_t new_address_space_size_in_bits) {
+    // No-op. Page table is never used.
+}
+
+} // namespace Core

src/core/arm/nce/arm_nce.h

@@ -0,0 +1,108 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <atomic>
+#include <memory>
+#include <span>
+#include <unordered_map>
+#include <vector>
+
+#include "core/arm/arm_interface.h"
+#include "core/arm/nce/guest_context.h"
+
+namespace Core::Memory {
+class Memory;
+}
+
+namespace Core {
+
+class System;
+
+class ARM_NCE final : public ARM_Interface {
+public:
+    ARM_NCE(System& system_, bool uses_wall_clock_, std::size_t core_index_);
+
+    ~ARM_NCE() override;
+
+    void Initialize() override;
+    void SetPC(u64 pc) override;
+    u64 GetPC() const override;
+    u64 GetSP() const override;
+    u64 GetReg(int index) const override;
+    void SetReg(int index, u64 value) override;
+    u128 GetVectorReg(int index) const override;
+    void SetVectorReg(int index, u128 value) override;
+
+    u32 GetPSTATE() const override;
+    void SetPSTATE(u32 pstate) override;
+    u64 GetTlsAddress() const override;
+    void SetTlsAddress(u64 address) override;
+    void SetTPIDR_EL0(u64 value) override;
+    u64 GetTPIDR_EL0() const override;
+
+    Architecture GetArchitecture() const override {
+        return Architecture::Aarch64;
+    }
+
+    void SaveContext(ThreadContext32& ctx) const override {}
+    void SaveContext(ThreadContext64& ctx) const override;
+    void LoadContext(const ThreadContext32& ctx) override {}
+    void LoadContext(const ThreadContext64& ctx) override;
+
+    void SignalInterrupt() override;
+    void ClearInterrupt() override;
+    void ClearExclusiveState() override;
+    void ClearInstructionCache() override;
+    void InvalidateCacheRange(u64 addr, std::size_t size) override;
+    void PageTableChanged(Common::PageTable& new_page_table,
+                          std::size_t new_address_space_size_in_bits) override;
+
+protected:
+    HaltReason RunJit() override;
+    HaltReason StepJit() override;
+
+    u32 GetSvcNumber() const override;
+
+    const Kernel::DebugWatchpoint* HaltedWatchpoint() const override {
+        return nullptr;
+    }
+
+    void RewindBreakpointInstruction() override {}
+
+private:
+    // Assembly definitions.
+    static HaltReason ReturnToRunCodeByTrampoline(void* tpidr, GuestContext* ctx,
+                                                  u64 trampoline_addr);
+    static HaltReason ReturnToRunCodeByExceptionLevelChange(int tid, void* tpidr);
+
+    static void ReturnToRunCodeByExceptionLevelChangeSignalHandler(int sig, void* info,
+                                                                   void* raw_context);
+    static void BreakFromRunCodeSignalHandler(int sig, void* info, void* raw_context);
+    static void GuestFaultSignalHandler(int sig, void* info, void* raw_context);
+
+    static void LockThreadParameters(void* tpidr);
+    static void UnlockThreadParameters(void* tpidr);
+
+private:
+    // C++ implementation functions for assembly definitions.
+    static void* RestoreGuestContext(void* raw_context);
+    static void SaveGuestContext(GuestContext* ctx, void* raw_context);
+    static bool HandleGuestFault(GuestContext* ctx, void* info, void* raw_context);
+    static void HandleHostFault(int sig, void* info, void* raw_context);
+
+public:
+    // Members set on initialization.
+    std::size_t core_index{};
+    pid_t thread_id{-1};
+
+    // Core context.
+    GuestContext guest_ctx;
+
+    // Thread and invalidation info.
+    std::mutex lock;
+    Kernel::KThread* running_thread{};
+};
+
+} // namespace Core

src/core/arm/nce/arm_nce.s

@@ -0,0 +1,222 @@
+/* SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project */
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#include "core/arm/nce/arm_nce_asm_definitions.h"
+
+#define LOAD_IMMEDIATE_32(reg, val)                     \
+    mov     reg, #(((val) >> 0x00) & 0xFFFF);           \
+    movk    reg, #(((val) >> 0x10) & 0xFFFF), lsl #16
+
+
+/* static HaltReason Core::ARM_NCE::ReturnToRunCodeByTrampoline(void* tpidr, Core::GuestContext* ctx, u64 trampoline_addr) */
+.section    .text._ZN4Core7ARM_NCE27ReturnToRunCodeByTrampolineEPvPNS_12GuestContextEm, "ax", %progbits
+.global     _ZN4Core7ARM_NCE27ReturnToRunCodeByTrampolineEPvPNS_12GuestContextEm
+.type       _ZN4Core7ARM_NCE27ReturnToRunCodeByTrampolineEPvPNS_12GuestContextEm, %function
+_ZN4Core7ARM_NCE27ReturnToRunCodeByTrampolineEPvPNS_12GuestContextEm:
+    /* Back up host sp to x3. */
+    /* Back up host tpidr_el0 to x4. */
+    mov     x3, sp
+    mrs     x4, tpidr_el0
+
+    /* Load guest sp. x5 is used as a scratch register. */
+    ldr     x5, [x1, #(GuestContextSp)]
+    mov     sp, x5
+
+    /* Offset GuestContext pointer to the host member. */
+    add     x5, x1, #(GuestContextHostContext)
+
+    /* Save original host sp and tpidr_el0 (x3, x4) to host context. */
+    stp     x3, x4, [x5, #(HostContextSpTpidrEl0)]
+
+    /* Save all callee-saved host GPRs. */
+    stp     x19, x20, [x5, #(HostContextRegs+0x0)]
+    stp     x21, x22, [x5, #(HostContextRegs+0x10)]
+    stp     x23, x24, [x5, #(HostContextRegs+0x20)]
+    stp     x25, x26, [x5, #(HostContextRegs+0x30)]
+    stp     x27, x28, [x5, #(HostContextRegs+0x40)]
+    stp     x29, x30, [x5, #(HostContextRegs+0x50)]
+
+    /* Save all callee-saved host FPRs. */
+    stp     q8, q9,   [x5, #(HostContextVregs+0x0)]
+    stp     q10, q11, [x5, #(HostContextVregs+0x20)]
+    stp     q12, q13, [x5, #(HostContextVregs+0x40)]
+    stp     q14, q15, [x5, #(HostContextVregs+0x60)]
+
+    /* Load guest tpidr_el0 from argument. */
+    msr     tpidr_el0, x0
+
+    /* Tail call the trampoline to restore guest state. */
+    br      x2
+
+
+/* static HaltReason Core::ARM_NCE::ReturnToRunCodeByExceptionLevelChange(int tid, void* tpidr) */
+.section    .text._ZN4Core7ARM_NCE37ReturnToRunCodeByExceptionLevelChangeEiPv, "ax", %progbits
+.global     _ZN4Core7ARM_NCE37ReturnToRunCodeByExceptionLevelChangeEiPv
+.type       _ZN4Core7ARM_NCE37ReturnToRunCodeByExceptionLevelChangeEiPv, %function
+_ZN4Core7ARM_NCE37ReturnToRunCodeByExceptionLevelChangeEiPv:
+    /* This jumps to the signal handler, which will restore the entire context. */
+    /* On entry, x0 = thread id, which is already in the right place. */
+
+    /* Move tpidr to x9 so it is not trampled. */
+    mov     x9, x1
+
+    /* Set up arguments. */
+    mov     x8, #(__NR_tkill)
+    mov     x1, #(ReturnToRunCodeByExceptionLevelChangeSignal)
+
+    /* Tail call the signal handler. */
+    svc     #0
+
+    /* Block execution from flowing here. */
+    brk     #1000
+
+
+/* static void Core::ARM_NCE::ReturnToRunCodeByExceptionLevelChangeSignalHandler(int sig, void* info, void* raw_context) */
+.section    .text._ZN4Core7ARM_NCE50ReturnToRunCodeByExceptionLevelChangeSignalHandlerEiPvS1_, "ax", %progbits
+.global     _ZN4Core7ARM_NCE50ReturnToRunCodeByExceptionLevelChangeSignalHandlerEiPvS1_
+.type       _ZN4Core7ARM_NCE50ReturnToRunCodeByExceptionLevelChangeSignalHandlerEiPvS1_, %function
+_ZN4Core7ARM_NCE50ReturnToRunCodeByExceptionLevelChangeSignalHandlerEiPvS1_:
+    stp     x29, x30, [sp, #-0x10]!
+    mov     x29, sp
+
+    /* Call the context restorer with the raw context. */
+    mov     x0, x2
+    bl      _ZN4Core7ARM_NCE19RestoreGuestContextEPv
+
+    /* Save the old value of tpidr_el0. */
+    mrs     x8, tpidr_el0
+    ldr     x9, [x0, #(TpidrEl0NativeContext)]
+    str     x8, [x9, #(GuestContextHostContext + HostContextTpidrEl0)]
+
+    /* Set our new tpidr_el0. */
+    msr     tpidr_el0, x0
+
+    /* Unlock the context. */
+    bl      _ZN4Core7ARM_NCE22UnlockThreadParametersEPv
+
+    /* Returning from here will enter the guest. */
+    ldp     x29, x30, [sp], #0x10
+    ret
+
+
+/* static void Core::ARM_NCE::BreakFromRunCodeSignalHandler(int sig, void* info, void* raw_context) */
+.section    .text._ZN4Core7ARM_NCE29BreakFromRunCodeSignalHandlerEiPvS1_, "ax", %progbits
+.global     _ZN4Core7ARM_NCE29BreakFromRunCodeSignalHandlerEiPvS1_
+.type       _ZN4Core7ARM_NCE29BreakFromRunCodeSignalHandlerEiPvS1_, %function
+_ZN4Core7ARM_NCE29BreakFromRunCodeSignalHandlerEiPvS1_:
+    /* Check to see if we have the correct TLS magic. */
+    mrs     x8, tpidr_el0
+    ldr     w9, [x8, #(TpidrEl0TlsMagic)]
+
+    LOAD_IMMEDIATE_32(w10, TlsMagic)
+
+    cmp     w9, w10
+    b.ne    1f
+
+    /* Correct TLS magic, so this is a guest interrupt. */
+    /* Restore host tpidr_el0. */
+    ldr     x0, [x8, #(TpidrEl0NativeContext)]
+    ldr     x3, [x0, #(GuestContextHostContext + HostContextTpidrEl0)]
+    msr     tpidr_el0, x3
+
+    /* Tail call the restorer. */
+    mov     x1, x2
+    b       _ZN4Core7ARM_NCE16SaveGuestContextEPNS_12GuestContextEPv
+
+    /* Returning from here will enter host code. */
+
+1:
+    /* Incorrect TLS magic, so this is a spurious signal. */
+    ret
+
+
+/* static void Core::ARM_NCE::GuestFaultSignalHandler(int sig, void* info, void* raw_context) */
+.section    .text._ZN4Core7ARM_NCE23GuestFaultSignalHandlerEiPvS1_, "ax", %progbits
+.global     _ZN4Core7ARM_NCE23GuestFaultSignalHandlerEiPvS1_
+.type       _ZN4Core7ARM_NCE23GuestFaultSignalHandlerEiPvS1_, %function
+_ZN4Core7ARM_NCE23GuestFaultSignalHandlerEiPvS1_:
+    /* Check to see if we have the correct TLS magic. */
+    mrs     x8, tpidr_el0
+    ldr     w9, [x8, #(TpidrEl0TlsMagic)]
+
+    LOAD_IMMEDIATE_32(w10, TlsMagic)
+
+    cmp     w9, w10
+    b.eq    1f
+
+    /* Incorrect TLS magic, so this is a host fault. */
+    /* Tail call the handler. */
+    b       _ZN4Core7ARM_NCE15HandleHostFaultEiPvS1_
+
+1:
+    /* Correct TLS magic, so this is a guest fault. */
+    stp     x29, x30, [sp, #-0x20]!
+    str     x19, [sp, #0x10]
+    mov     x29, sp
+
+    /* Save the old tpidr_el0. */
+    mov     x19, x8
+
+    /* Restore host tpidr_el0. */
+    ldr     x0, [x8, #(TpidrEl0NativeContext)]
+    ldr     x3, [x0, #(GuestContextHostContext + HostContextTpidrEl0)]
+    msr     tpidr_el0, x3
+
+    /* Call the handler. */
+    bl       _ZN4Core7ARM_NCE16HandleGuestFaultEPNS_12GuestContextEPvS3_
+
+    /* If the handler returned false, we want to preserve the host tpidr_el0. */
+    cbz     x0, 2f
+
+    /* Otherwise, restore guest tpidr_el0. */
+    msr     tpidr_el0, x19
+
+2:
+    ldr     x19, [sp, #0x10]
+    ldp     x29, x30, [sp], #0x20
+    ret
+
+
+/* static void Core::ARM_NCE::LockThreadParameters(void* tpidr) */
+.section    .text._ZN4Core7ARM_NCE20LockThreadParametersEPv, "ax", %progbits
+.global     _ZN4Core7ARM_NCE20LockThreadParametersEPv
+.type       _ZN4Core7ARM_NCE20LockThreadParametersEPv, %function
+_ZN4Core7ARM_NCE20LockThreadParametersEPv:
+    /* Offset to lock member. */
+    add     x0, x0, #(TpidrEl0Lock)
+
+1:
+    /* Clear the monitor. */
+    clrex
+
+2:
+    /* Load-linked with acquire ordering. */
+    ldaxr   w1, [x0]
+
+    /* If the value was SpinLockLocked, clear monitor and retry. */
+    cbz     w1, 1b
+
+    /* Store-conditional SpinLockLocked with relaxed ordering. */
+    stxr    w1, wzr, [x0]
+
+    /* If we failed to store, retry. */
+    cbnz    w1, 2b
+
+    ret
+
+
+/* static void Core::ARM_NCE::UnlockThreadParameters(void* tpidr) */
+.section    .text._ZN4Core7ARM_NCE22UnlockThreadParametersEPv, "ax", %progbits
+.global     _ZN4Core7ARM_NCE22UnlockThreadParametersEPv
+.type       _ZN4Core7ARM_NCE22UnlockThreadParametersEPv, %function
+_ZN4Core7ARM_NCE22UnlockThreadParametersEPv:
+    /* Offset to lock member. */
+    add     x0, x0, #(TpidrEl0Lock)
+
+    /* Load SpinLockUnlocked. */
+    mov     w1, #(SpinLockUnlocked)
+
+    /* Store value with release ordering. */
+    stlr    w1, [x0]
+
+    ret

src/core/arm/nce/arm_nce_asm_definitions.h

@@ -0,0 +1,29 @@
+/* SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project */
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#pragma once
+
+#define __ASSEMBLY__
+
+#include <asm-generic/signal.h>
+#include <asm-generic/unistd.h>
+
+#define ReturnToRunCodeByExceptionLevelChangeSignal SIGUSR2
+#define BreakFromRunCodeSignal SIGURG
+#define GuestFaultSignal SIGSEGV
+
+#define GuestContextSp 0xF8
+#define GuestContextHostContext 0x320
+
+#define HostContextSpTpidrEl0 0xE0
+#define HostContextTpidrEl0 0xE8
+#define HostContextRegs 0x0
+#define HostContextVregs 0x60
+
+#define TpidrEl0NativeContext 0x10
+#define TpidrEl0Lock 0x18
+#define TpidrEl0TlsMagic 0x20
+#define TlsMagic 0x555a5559
+
+#define SpinLockLocked 0
+#define SpinLockUnlocked 1

src/core/arm/nce/guest_context.h

@@ -0,0 +1,50 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "core/arm/arm_interface.h"
+#include "core/arm/nce/arm_nce_asm_definitions.h"
+
+namespace Core {
+
+class ARM_NCE;
+class System;
+
+struct HostContext {
+    alignas(16) std::array<u64, 12> host_saved_regs{};
+    alignas(16) std::array<u128, 8> host_saved_vregs{};
+    u64 host_sp{};
+    void* host_tpidr_el0{};
+};
+
+struct GuestContext {
+    std::array<u64, 31> cpu_registers{};
+    u64 sp{};
+    u64 pc{};
+    u32 fpcr{};
+    u32 fpsr{};
+    std::array<u128, 32> vector_registers{};
+    u32 pstate{};
+    alignas(16) HostContext host_ctx{};
+    u64 tpidrro_el0{};
+    u64 tpidr_el0{};
+    std::atomic<u64> esr_el1{};
+    u32 nzcv{};
+    u32 svc_swi{};
+    System* system{};
+    ARM_NCE* parent{};
+};
+
+// Verify assembly offsets.
+static_assert(offsetof(GuestContext, sp) == GuestContextSp);
+static_assert(offsetof(GuestContext, host_ctx) == GuestContextHostContext);
+static_assert(offsetof(HostContext, host_sp) == HostContextSpTpidrEl0);
+static_assert(offsetof(HostContext, host_tpidr_el0) - 8 == HostContextSpTpidrEl0);
+static_assert(offsetof(HostContext, host_tpidr_el0) == HostContextTpidrEl0);
+static_assert(offsetof(HostContext, host_saved_regs) == HostContextRegs);
+static_assert(offsetof(HostContext, host_saved_vregs) == HostContextVregs);
+
+} // namespace Core

src/core/arm/nce/instructions.h

@@ -0,0 +1,147 @@
+// SPDX-FileCopyrightText: Copyright © 2020 Skyline Team and Contributors
+// SPDX-License-Identifier: MPL-2.0
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+
+namespace Core::NCE {
+
+enum SystemRegister : u32 {
+    TpidrEl0 = 0x5E82,
+    TpidrroEl0 = 0x5E83,
+    CntfrqEl0 = 0x5F00,
+    CntpctEl0 = 0x5F01,
+};
+
+// https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/SVC--Supervisor-Call-
+union SVC {
+    constexpr explicit SVC(u32 raw_) : raw{raw_} {}
+
+    constexpr bool Verify() {
+        return (this->GetSig0() == 0x1 && this->GetSig1() == 0x6A0);
+    }
+
+    constexpr u32 GetSig0() {
+        return decltype(sig0)::ExtractValue(raw);
+    }
+
+    constexpr u32 GetValue() {
+        return decltype(value)::ExtractValue(raw);
+    }
+
+    constexpr u32 GetSig1() {
+        return decltype(sig1)::ExtractValue(raw);
+    }
+
+    u32 raw;
+
+private:
+    BitField<0, 5, u32> sig0;   // 0x1
+    BitField<5, 16, u32> value; // 16-bit immediate
+    BitField<21, 11, u32> sig1; // 0x6A0
+};
+static_assert(sizeof(SVC) == sizeof(u32));
+static_assert(SVC(0xD40000C1).Verify());
+static_assert(SVC(0xD40000C1).GetValue() == 0x6);
+
+// https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/MRS--Move-System-Register-
+union MRS {
+    constexpr explicit MRS(u32 raw_) : raw{raw_} {}
+
+    constexpr bool Verify() {
+        return (this->GetSig() == 0xD53);
+    }
+
+    constexpr u32 GetRt() {
+        return decltype(rt)::ExtractValue(raw);
+    }
+
+    constexpr u32 GetSystemReg() {
+        return decltype(system_reg)::ExtractValue(raw);
+    }
+
+    constexpr u32 GetSig() {
+        return decltype(sig)::ExtractValue(raw);
+    }
+
+    u32 raw;
+
+private:
+    BitField<0, 5, u32> rt;          // destination register
+    BitField<5, 15, u32> system_reg; // source system register
+    BitField<20, 12, u32> sig;       // 0xD53
+};
+static_assert(sizeof(MRS) == sizeof(u32));
+static_assert(MRS(0xD53BE020).Verify());
+static_assert(MRS(0xD53BE020).GetSystemReg() == CntpctEl0);
+static_assert(MRS(0xD53BE020).GetRt() == 0x0);
+
+// https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/MSR--register---Move-general-purpose-register-to-System-Register-
+union MSR {
+    constexpr explicit MSR(u32 raw_) : raw{raw_} {}
+
+    constexpr bool Verify() {
+        return this->GetSig() == 0xD51;
+    }
+
+    constexpr u32 GetRt() {
+        return decltype(rt)::ExtractValue(raw);
+    }
+
+    constexpr u32 GetSystemReg() {
+        return decltype(system_reg)::ExtractValue(raw);
+    }
+
+    constexpr u32 GetSig() {
+        return decltype(sig)::ExtractValue(raw);
+    }
+
+    u32 raw;
+
+private:
+    BitField<0, 5, u32> rt;          // source register
+    BitField<5, 15, u32> system_reg; // destination system register
+    BitField<20, 12, u32> sig;       // 0xD51
+};
+static_assert(sizeof(MSR) == sizeof(u32));
+static_assert(MSR(0xD51BD040).Verify());
+static_assert(MSR(0xD51BD040).GetSystemReg() == TpidrEl0);
+static_assert(MSR(0xD51BD040).GetRt() == 0x0);
+
+// https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/LDXR--Load-Exclusive-Register-
+// https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/LDXP--Load-Exclusive-Pair-of-Registers-
+// https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/STXR--Store-Exclusive-Register-
+// https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/STXP--Store-Exclusive-Pair-of-registers-
+union Exclusive {
+    constexpr explicit Exclusive(u32 raw_) : raw{raw_} {}
+
+    constexpr bool Verify() {
+        return this->GetSig() == 0x10;
+    }
+
+    constexpr u32 GetSig() {
+        return decltype(sig)::ExtractValue(raw);
+    }
+
+    constexpr u32 AsOrdered() {
+        return raw | decltype(o0)::FormatValue(1);
+    }
+
+    u32 raw;
+
+private:
+    BitField<0, 5, u32> rt;    // memory operand
+    BitField<5, 5, u32> rn;    // register operand 1
+    BitField<10, 5, u32> rt2;  // register operand 2
+    BitField<15, 1, u32> o0;   // ordered
+    BitField<16, 5, u32> rs;   // status register
+    BitField<21, 2, u32> l;    // operation type
+    BitField<23, 7, u32> sig;  // 0x10
+    BitField<30, 2, u32> size; // size
+};
+static_assert(Exclusive(0xC85FFC00).Verify());
+static_assert(Exclusive(0xC85FFC00).AsOrdered() == 0xC85FFC00);
+static_assert(Exclusive(0xC85F7C00).AsOrdered() == 0xC85FFC00);
+static_assert(Exclusive(0xC8200440).AsOrdered() == 0xC8208440);
+
+} // namespace Core::NCE

src/core/arm/nce/patch.cpp

@@ -0,0 +1,474 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "common/arm64/native_clock.h"
+#include "common/bit_cast.h"
+#include "common/literals.h"
+#include "core/arm/nce/arm_nce.h"
+#include "core/arm/nce/guest_context.h"
+#include "core/arm/nce/instructions.h"
+#include "core/arm/nce/patch.h"
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/hle/kernel/svc.h"
+
+namespace Core::NCE {
+
+using namespace Common::Literals;
+using namespace oaknut::util;
+
+using NativeExecutionParameters = Kernel::KThread::NativeExecutionParameters;
+
+constexpr size_t MaxRelativeBranch = 128_MiB;
+constexpr u32 ModuleCodeIndex = 0x24 / sizeof(u32);
+
+Patcher::Patcher() : c(m_patch_instructions) {}
+
+Patcher::~Patcher() = default;
+
+void Patcher::PatchText(const Kernel::PhysicalMemory& program_image,
+                        const Kernel::CodeSet::Segment& code) {
+
+    // Write save context helper function.
+    c.l(m_save_context);
+    WriteSaveContext();
+
+    // Write load context helper function.
+    c.l(m_load_context);
+    WriteLoadContext();
+
+    // Retrieve text segment data.
+    const auto text = std::span{program_image}.subspan(code.offset, code.size);
+    const auto text_words =
+        std::span<const u32>{reinterpret_cast<const u32*>(text.data()), text.size() / sizeof(u32)};
+
+    // Loop through instructions, patching as needed.
+    for (u32 i = ModuleCodeIndex; i < static_cast<u32>(text_words.size()); i++) {
+        const u32 inst = text_words[i];
+
+        const auto AddRelocations = [&] {
+            const uintptr_t this_offset = i * sizeof(u32);
+            const uintptr_t next_offset = this_offset + sizeof(u32);
+
+            // Relocate from here to patch.
+            this->BranchToPatch(this_offset);
+
+            // Relocate from patch to next instruction.
+            return next_offset;
+        };
+
+        // SVC
+        if (auto svc = SVC{inst}; svc.Verify()) {
+            WriteSvcTrampoline(AddRelocations(), svc.GetValue());
+            continue;
+        }
+
+        // MRS Xn, TPIDR_EL0
+        // MRS Xn, TPIDRRO_EL0
+        if (auto mrs = MRS{inst};
+            mrs.Verify() && (mrs.GetSystemReg() == TpidrroEl0 || mrs.GetSystemReg() == TpidrEl0)) {
+            const auto src_reg = mrs.GetSystemReg() == TpidrroEl0 ? oaknut::SystemReg::TPIDRRO_EL0
+                                                                  : oaknut::SystemReg::TPIDR_EL0;
+            const auto dest_reg = oaknut::XReg{static_cast<int>(mrs.GetRt())};
+            WriteMrsHandler(AddRelocations(), dest_reg, src_reg);
+            continue;
+        }
+
+        // MRS Xn, CNTPCT_EL0
+        if (auto mrs = MRS{inst}; mrs.Verify() && mrs.GetSystemReg() == CntpctEl0) {
+            WriteCntpctHandler(AddRelocations(), oaknut::XReg{static_cast<int>(mrs.GetRt())});
+            continue;
+        }
+
+        // MRS Xn, CNTFRQ_EL0
+        if (auto mrs = MRS{inst}; mrs.Verify() && mrs.GetSystemReg() == CntfrqEl0) {
+            UNREACHABLE();
+        }
+
+        // MSR TPIDR_EL0, Xn
+        if (auto msr = MSR{inst}; msr.Verify() && msr.GetSystemReg() == TpidrEl0) {
+            WriteMsrHandler(AddRelocations(), oaknut::XReg{static_cast<int>(msr.GetRt())});
+            continue;
+        }
+
+        if (auto exclusive = Exclusive{inst}; exclusive.Verify()) {
+            m_exclusives.push_back(i);
+        }
+    }
+
+    // Determine patching mode for the final relocation step
+    const size_t image_size = program_image.size();
+    this->mode = image_size > MaxRelativeBranch ? PatchMode::PreText : PatchMode::PostData;
+}
+
+void Patcher::RelocateAndCopy(Common::ProcessAddress load_base,
+                              const Kernel::CodeSet::Segment& code,
+                              Kernel::PhysicalMemory& program_image,
+                              EntryTrampolines* out_trampolines) {
+    const size_t patch_size = GetSectionSize();
+    const size_t image_size = program_image.size();
+
+    // Retrieve text segment data.
+    const auto text = std::span{program_image}.subspan(code.offset, code.size);
+    const auto text_words =
+        std::span<u32>{reinterpret_cast<u32*>(text.data()), text.size() / sizeof(u32)};
+
+    const auto ApplyBranchToPatchRelocation = [&](u32* target, const Relocation& rel) {
+        oaknut::CodeGenerator rc{target};
+        if (mode == PatchMode::PreText) {
+            rc.B(rel.patch_offset - patch_size - rel.module_offset);
+        } else {
+            rc.B(image_size - rel.module_offset + rel.patch_offset);
+        }
+    };
+
+    const auto ApplyBranchToModuleRelocation = [&](u32* target, const Relocation& rel) {
+        oaknut::CodeGenerator rc{target};
+        if (mode == PatchMode::PreText) {
+            rc.B(patch_size - rel.patch_offset + rel.module_offset);
+        } else {
+            rc.B(rel.module_offset - image_size - rel.patch_offset);
+        }
+    };
+
+    const auto RebasePatch = [&](ptrdiff_t patch_offset) {
+        if (mode == PatchMode::PreText) {
+            return GetInteger(load_base) + patch_offset;
+        } else {
+            return GetInteger(load_base) + image_size + patch_offset;
+        }
+    };
+
+    const auto RebasePc = [&](uintptr_t module_offset) {
+        if (mode == PatchMode::PreText) {
+            return GetInteger(load_base) + patch_size + module_offset;
+        } else {
+            return GetInteger(load_base) + module_offset;
+        }
+    };
+
+    // We are now ready to relocate!
+    for (const Relocation& rel : m_branch_to_patch_relocations) {
+        ApplyBranchToPatchRelocation(text_words.data() + rel.module_offset / sizeof(u32), rel);
+    }
+    for (const Relocation& rel : m_branch_to_module_relocations) {
+        ApplyBranchToModuleRelocation(m_patch_instructions.data() + rel.patch_offset / sizeof(u32),
+                                      rel);
+    }
+
+    // Rewrite PC constants and record post trampolines
+    for (const Relocation& rel : m_write_module_pc_relocations) {
+        oaknut::CodeGenerator rc{m_patch_instructions.data() + rel.patch_offset / sizeof(u32)};
+        rc.dx(RebasePc(rel.module_offset));
+    }
+    for (const Trampoline& rel : m_trampolines) {
+        out_trampolines->insert({RebasePc(rel.module_offset), RebasePatch(rel.patch_offset)});
+    }
+
+    // Cortex-A57 seems to treat all exclusives as ordered, but newer processors do not.
+    // Convert to ordered to preserve this assumption.
+    for (const ModuleTextAddress i : m_exclusives) {
+        auto exclusive = Exclusive{text_words[i]};
+        text_words[i] = exclusive.AsOrdered();
+    }
+
+    // Copy to program image
+    if (this->mode == PatchMode::PreText) {
+        std::memcpy(program_image.data(), m_patch_instructions.data(),
+                    m_patch_instructions.size() * sizeof(u32));
+    } else {
+        program_image.resize(image_size + patch_size);
+        std::memcpy(program_image.data() + image_size, m_patch_instructions.data(),
+                    m_patch_instructions.size() * sizeof(u32));
+    }
+}
+
+size_t Patcher::GetSectionSize() const noexcept {
+    return Common::AlignUp(m_patch_instructions.size() * sizeof(u32), Core::Memory::YUZU_PAGESIZE);
+}
+
+void Patcher::WriteLoadContext() {
+    // This function was called, which modifies X30, so use that as a scratch register.
+    // SP contains the guest X30, so save our return X30 to SP + 8, since we have allocated 16 bytes
+    // of stack.
+    c.STR(X30, SP, 8);
+    c.MRS(X30, oaknut::SystemReg::TPIDR_EL0);
+    c.LDR(X30, X30, offsetof(NativeExecutionParameters, native_context));
+
+    // Load system registers.
+    c.LDR(W0, X30, offsetof(GuestContext, fpsr));
+    c.MSR(oaknut::SystemReg::FPSR, X0);
+    c.LDR(W0, X30, offsetof(GuestContext, fpcr));
+    c.MSR(oaknut::SystemReg::FPCR, X0);
+    c.LDR(W0, X30, offsetof(GuestContext, nzcv));
+    c.MSR(oaknut::SystemReg::NZCV, X0);
+
+    // Load all vector registers.
+    static constexpr size_t VEC_OFF = offsetof(GuestContext, vector_registers);
+    for (int i = 0; i <= 30; i += 2) {
+        c.LDP(oaknut::QReg{i}, oaknut::QReg{i + 1}, X30, VEC_OFF + 16 * i);
+    }
+
+    // Load all general-purpose registers except X30.
+    for (int i = 0; i <= 28; i += 2) {
+        c.LDP(oaknut::XReg{i}, oaknut::XReg{i + 1}, X30, 8 * i);
+    }
+
+    // Reload our return X30 from the stack and return.
+    // The patch code will reload the guest X30 for us.
+    c.LDR(X30, SP, 8);
+    c.RET();
+}
+
+void Patcher::WriteSaveContext() {
+    // This function was called, which modifies X30, so use that as a scratch register.
+    // SP contains the guest X30, so save our X30 to SP + 8, since we have allocated 16 bytes of
+    // stack.
+    c.STR(X30, SP, 8);
+    c.MRS(X30, oaknut::SystemReg::TPIDR_EL0);
+    c.LDR(X30, X30, offsetof(NativeExecutionParameters, native_context));
+
+    // Store all general-purpose registers except X30.
+    for (int i = 0; i <= 28; i += 2) {
+        c.STP(oaknut::XReg{i}, oaknut::XReg{i + 1}, X30, 8 * i);
+    }
+
+    // Store all vector registers.
+    static constexpr size_t VEC_OFF = offsetof(GuestContext, vector_registers);
+    for (int i = 0; i <= 30; i += 2) {
+        c.STP(oaknut::QReg{i}, oaknut::QReg{i + 1}, X30, VEC_OFF + 16 * i);
+    }
+
+    // Store guest system registers, X30 and SP, using X0 as a scratch register.
+    c.STR(X0, SP, PRE_INDEXED, -16);
+    c.LDR(X0, SP, 16);
+    c.STR(X0, X30, 8 * 30);
+    c.ADD(X0, SP, 32);
+    c.STR(X0, X30, offsetof(GuestContext, sp));
+    c.MRS(X0, oaknut::SystemReg::FPSR);
+    c.STR(W0, X30, offsetof(GuestContext, fpsr));
+    c.MRS(X0, oaknut::SystemReg::FPCR);
+    c.STR(W0, X30, offsetof(GuestContext, fpcr));
+    c.MRS(X0, oaknut::SystemReg::NZCV);
+    c.STR(W0, X30, offsetof(GuestContext, nzcv));
+    c.LDR(X0, SP, POST_INDEXED, 16);
+
+    // Reload our return X30 from the stack, and return.
+    c.LDR(X30, SP, 8);
+    c.RET();
+}
+
+void Patcher::WriteSvcTrampoline(ModuleDestLabel module_dest, u32 svc_id) {
+    // We are about to start saving state, so we need to lock the context.
+    this->LockContext();
+
+    // Store guest X30 to the stack. Then, save the context and restore the stack.
+    // This will save all registers except PC, but we know PC at patch time.
+    c.STR(X30, SP, PRE_INDEXED, -16);
+    c.BL(m_save_context);
+    c.LDR(X30, SP, POST_INDEXED, 16);
+
+    // Now that we've saved all registers, we can use any registers as scratch.
+    // Store PC + 4 to arm interface, since we know the instruction offset from the entry point.
+    oaknut::Label pc_after_svc;
+    c.MRS(X1, oaknut::SystemReg::TPIDR_EL0);
+    c.LDR(X1, X1, offsetof(NativeExecutionParameters, native_context));
+    c.LDR(X2, pc_after_svc);
+    c.STR(X2, X1, offsetof(GuestContext, pc));
+
+    // Store SVC number to execute when we return
+    c.MOV(X2, svc_id);
+    c.STR(W2, X1, offsetof(GuestContext, svc_swi));
+
+    // We are calling a SVC. Clear esr_el1 and return it.
+    static_assert(std::is_same_v<std::underlying_type_t<HaltReason>, u64>);
+    oaknut::Label retry;
+    c.ADD(X2, X1, offsetof(GuestContext, esr_el1));
+    c.l(retry);
+    c.LDAXR(X0, X2);
+    c.STLXR(W3, XZR, X2);
+    c.CBNZ(W3, retry);
+
+    // Add "calling SVC" flag. Since this is X0, this is now our return value.
+    c.ORR(X0, X0, static_cast<u64>(HaltReason::SupervisorCall));
+
+    // Offset the GuestContext pointer to the HostContext member.
+    // STP has limited range of [-512, 504] which we can't reach otherwise
+    // NB: Due to this all offsets below are from the start of HostContext.
+    c.ADD(X1, X1, offsetof(GuestContext, host_ctx));
+
+    // Reload host TPIDR_EL0 and SP.
+    static_assert(offsetof(HostContext, host_sp) + 8 == offsetof(HostContext, host_tpidr_el0));
+    c.LDP(X2, X3, X1, offsetof(HostContext, host_sp));
+    c.MOV(SP, X2);
+    c.MSR(oaknut::SystemReg::TPIDR_EL0, X3);
+
+    // Load callee-saved host registers and return to host.
+    static constexpr size_t HOST_REGS_OFF = offsetof(HostContext, host_saved_regs);
+    static constexpr size_t HOST_VREGS_OFF = offsetof(HostContext, host_saved_vregs);
+    c.LDP(X19, X20, X1, HOST_REGS_OFF);
+    c.LDP(X21, X22, X1, HOST_REGS_OFF + 2 * sizeof(u64));
+    c.LDP(X23, X24, X1, HOST_REGS_OFF + 4 * sizeof(u64));
+    c.LDP(X25, X26, X1, HOST_REGS_OFF + 6 * sizeof(u64));
+    c.LDP(X27, X28, X1, HOST_REGS_OFF + 8 * sizeof(u64));
+    c.LDP(X29, X30, X1, HOST_REGS_OFF + 10 * sizeof(u64));
+    c.LDP(Q8, Q9, X1, HOST_VREGS_OFF);
+    c.LDP(Q10, Q11, X1, HOST_VREGS_OFF + 2 * sizeof(u128));
+    c.LDP(Q12, Q13, X1, HOST_VREGS_OFF + 4 * sizeof(u128));
+    c.LDP(Q14, Q15, X1, HOST_VREGS_OFF + 6 * sizeof(u128));
+    c.RET();
+
+    // Write the post-SVC trampoline address, which will jump back to the guest after restoring its
+    // state.
+    m_trampolines.push_back({c.offset(), module_dest});
+
+    // Host called this location. Save the return address so we can
+    // unwind the stack properly when jumping back.
+    c.MRS(X2, oaknut::SystemReg::TPIDR_EL0);
+    c.LDR(X2, X2, offsetof(NativeExecutionParameters, native_context));
+    c.ADD(X0, X2, offsetof(GuestContext, host_ctx));
+    c.STR(X30, X0, offsetof(HostContext, host_saved_regs) + 11 * sizeof(u64));
+
+    // Reload all guest registers except X30 and PC.
+    // The function also expects 16 bytes of stack already allocated.
+    c.STR(X30, SP, PRE_INDEXED, -16);
+    c.BL(m_load_context);
+    c.LDR(X30, SP, POST_INDEXED, 16);
+
+    // Use X1 as a scratch register to restore X30.
+    c.STR(X1, SP, PRE_INDEXED, -16);
+    c.MRS(X1, oaknut::SystemReg::TPIDR_EL0);
+    c.LDR(X1, X1, offsetof(NativeExecutionParameters, native_context));
+    c.LDR(X30, X1, offsetof(GuestContext, cpu_registers) + sizeof(u64) * 30);
+    c.LDR(X1, SP, POST_INDEXED, 16);
+
+    // Unlock the context.
+    this->UnlockContext();
+
+    // Jump back to the instruction after the emulated SVC.
+    this->BranchToModule(module_dest);
+
+    // Store PC after call.
+    c.l(pc_after_svc);
+    this->WriteModulePc(module_dest);
+}
+
+void Patcher::WriteMrsHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg,
+                              oaknut::SystemReg src_reg) {
+    // Retrieve emulated TLS register from GuestContext.
+    c.MRS(dest_reg, oaknut::SystemReg::TPIDR_EL0);
+    if (src_reg == oaknut::SystemReg::TPIDRRO_EL0) {
+        c.LDR(dest_reg, dest_reg, offsetof(NativeExecutionParameters, tpidrro_el0));
+    } else {
+        c.LDR(dest_reg, dest_reg, offsetof(NativeExecutionParameters, tpidr_el0));
+    }
+
+    // Jump back to the instruction after the emulated MRS.
+    this->BranchToModule(module_dest);
+}
+
+void Patcher::WriteMsrHandler(ModuleDestLabel module_dest, oaknut::XReg src_reg) {
+    const auto scratch_reg = src_reg.index() == 0 ? X1 : X0;
+    c.STR(scratch_reg, SP, PRE_INDEXED, -16);
+
+    // Save guest value to NativeExecutionParameters::tpidr_el0.
+    c.MRS(scratch_reg, oaknut::SystemReg::TPIDR_EL0);
+    c.STR(src_reg, scratch_reg, offsetof(NativeExecutionParameters, tpidr_el0));
+
+    // Restore scratch register.
+    c.LDR(scratch_reg, SP, POST_INDEXED, 16);
+
+    // Jump back to the instruction after the emulated MSR.
+    this->BranchToModule(module_dest);
+}
+
+void Patcher::WriteCntpctHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg) {
+    static Common::Arm64::NativeClock clock{};
+    const auto factor = clock.GetGuestCNTFRQFactor();
+    const auto raw_factor = Common::BitCast<std::array<u64, 2>>(factor);
+
+    const auto use_x2_x3 = dest_reg.index() == 0 || dest_reg.index() == 1;
+    oaknut::XReg scratch0 = use_x2_x3 ? X2 : X0;
+    oaknut::XReg scratch1 = use_x2_x3 ? X3 : X1;
+
+    oaknut::Label factorlo;
+    oaknut::Label factorhi;
+
+    // Save scratches.
+    c.STP(scratch0, scratch1, SP, PRE_INDEXED, -16);
+
+    // Load counter value.
+    c.MRS(dest_reg, oaknut::SystemReg::CNTVCT_EL0);
+
+    // Load scaling factor.
+    c.LDR(scratch0, factorlo);
+    c.LDR(scratch1, factorhi);
+
+    // Multiply low bits and get result.
+    c.UMULH(scratch0, dest_reg, scratch0);
+
+    // Multiply high bits and add low bit result.
+    c.MADD(dest_reg, dest_reg, scratch1, scratch0);
+
+    // Reload scratches.
+    c.LDP(scratch0, scratch1, SP, POST_INDEXED, 16);
+
+    // Jump back to the instruction after the emulated MRS.
+    this->BranchToModule(module_dest);
+
+    // Scaling factor constant values.
+    c.l(factorlo);
+    c.dx(raw_factor[0]);
+    c.l(factorhi);
+    c.dx(raw_factor[1]);
+}
+
+void Patcher::LockContext() {
+    oaknut::Label retry;
+
+    // Save scratches.
+    c.STP(X0, X1, SP, PRE_INDEXED, -16);
+
+    // Reload lock pointer.
+    c.l(retry);
+    c.CLREX();
+    c.MRS(X0, oaknut::SystemReg::TPIDR_EL0);
+    c.ADD(X0, X0, offsetof(NativeExecutionParameters, lock));
+
+    static_assert(SpinLockLocked == 0);
+
+    // Load-linked with acquire ordering.
+    c.LDAXR(W1, X0);
+
+    // If the value was SpinLockLocked, clear monitor and retry.
+    c.CBZ(W1, retry);
+
+    // Store-conditional SpinLockLocked with relaxed ordering.
+    c.STXR(W1, WZR, X0);
+
+    // If we failed to store, retry.
+    c.CBNZ(W1, retry);
+
+    // We succeeded! Reload scratches.
+    c.LDP(X0, X1, SP, POST_INDEXED, 16);
+}
+
+void Patcher::UnlockContext() {
+    // Save scratches.
+    c.STP(X0, X1, SP, PRE_INDEXED, -16);
+
+    // Load lock pointer.
+    c.MRS(X0, oaknut::SystemReg::TPIDR_EL0);
+    c.ADD(X0, X0, offsetof(NativeExecutionParameters, lock));
+
+    // Load SpinLockUnlocked.
+    c.MOV(W1, SpinLockUnlocked);
+
+    // Store value with release ordering.
+    c.STLR(W1, X0);
+
+    // Load scratches.
+    c.LDP(X0, X1, SP, POST_INDEXED, 16);
+}
+
+} // namespace Core::NCE

src/core/arm/nce/patch.h

@@ -0,0 +1,98 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <span>
+#include <unordered_map>
+#include <vector>
+#include <oaknut/code_block.hpp>
+#include <oaknut/oaknut.hpp>
+
+#include "common/common_types.h"
+#include "core/hle/kernel/code_set.h"
+#include "core/hle/kernel/k_typed_address.h"
+#include "core/hle/kernel/physical_memory.h"
+
+namespace Core::NCE {
+
+enum class PatchMode : u32 {
+    None,
+    PreText,  ///< Patch section is inserted before .text
+    PostData, ///< Patch section is inserted after .data
+};
+
+using ModuleTextAddress = u64;
+using PatchTextAddress = u64;
+using EntryTrampolines = std::unordered_map<ModuleTextAddress, PatchTextAddress>;
+
+class Patcher {
+public:
+    explicit Patcher();
+    ~Patcher();
+
+    void PatchText(const Kernel::PhysicalMemory& program_image,
+                   const Kernel::CodeSet::Segment& code);
+    void RelocateAndCopy(Common::ProcessAddress load_base, const Kernel::CodeSet::Segment& code,
+                         Kernel::PhysicalMemory& program_image, EntryTrampolines* out_trampolines);
+    size_t GetSectionSize() const noexcept;
+
+    [[nodiscard]] PatchMode GetPatchMode() const noexcept {
+        return mode;
+    }
+
+private:
+    using ModuleDestLabel = uintptr_t;
+
+    struct Trampoline {
+        ptrdiff_t patch_offset;
+        uintptr_t module_offset;
+    };
+
+    void WriteLoadContext();
+    void WriteSaveContext();
+    void LockContext();
+    void UnlockContext();
+    void WriteSvcTrampoline(ModuleDestLabel module_dest, u32 svc_id);
+    void WriteMrsHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg,
+                         oaknut::SystemReg src_reg);
+    void WriteMsrHandler(ModuleDestLabel module_dest, oaknut::XReg src_reg);
+    void WriteCntpctHandler(ModuleDestLabel module_dest, oaknut::XReg dest_reg);
+
+private:
+    void BranchToPatch(uintptr_t module_dest) {
+        m_branch_to_patch_relocations.push_back({c.offset(), module_dest});
+    }
+
+    void BranchToModule(uintptr_t module_dest) {
+        m_branch_to_module_relocations.push_back({c.offset(), module_dest});
+        c.dw(0);
+    }
+
+    void WriteModulePc(uintptr_t module_dest) {
+        m_write_module_pc_relocations.push_back({c.offset(), module_dest});
+        c.dx(0);
+    }
+
+private:
+    // List of patch instructions we have generated.
+    std::vector<u32> m_patch_instructions{};
+
+    // Relocation type for relative branch from module to patch.
+    struct Relocation {
+        ptrdiff_t patch_offset;  ///< Offset in bytes from the start of the patch section.
+        uintptr_t module_offset; ///< Offset in bytes from the start of the text section.
+    };
+
+    oaknut::VectorCodeGenerator c;
+    std::vector<Trampoline> m_trampolines;
+    std::vector<Relocation> m_branch_to_patch_relocations{};
+    std::vector<Relocation> m_branch_to_module_relocations{};
+    std::vector<Relocation> m_write_module_pc_relocations{};
+    std::vector<ModuleTextAddress> m_exclusives{};
+    oaknut::Label m_save_context{};
+    oaknut::Label m_load_context{};
+    PatchMode mode{PatchMode::None};
+};
+
+} // namespace Core::NCE

src/core/cpu_manager.cpp

@@ -211,6 +211,8 @@ void CpuManager::RunThread(std::stop_token token, std::size_t core) {
         system.GPU().ObtainContext();
     }
 
+    system.ArmInterface(core).Initialize();
+
     auto& kernel = system.Kernel();
     auto& scheduler = *kernel.CurrentScheduler();
     auto* thread = scheduler.GetSchedulerCurrentThread();

src/core/device_memory.cpp

@@ -6,7 +6,7 @@
 
 namespace Core {
 
-#ifdef ANDROID
+#ifdef HAS_NCE
 constexpr size_t VirtualReserveSize = 1ULL << 38;
 #else
 constexpr size_t VirtualReserveSize = 1ULL << 39;
@@ -15,6 +15,7 @@ constexpr size_t VirtualReserveSize = 1ULL << 39;
 DeviceMemory::DeviceMemory()
     : buffer{Kernel::Board::Nintendo::Nx::KSystemControl::Init::GetIntendedMemorySize(),
              VirtualReserveSize} {}
+
 DeviceMemory::~DeviceMemory() = default;
 
 } // namespace Core

src/core/hid/emulated_controller.cpp

@@ -509,9 +509,11 @@ void EmulatedController::ReloadInput() {
         });
     }
     turbo_button_state = 0;
+    is_initalized = true;
 }
 
 void EmulatedController::UnloadInput() {
+    is_initalized = false;
     for (auto& button : button_devices) {
         button.reset();
     }
@@ -1207,6 +1209,9 @@ void EmulatedController::SetNfc(const Common::Input::CallbackStatus& callback) {
 }
 
 bool EmulatedController::SetVibration(std::size_t device_index, VibrationValue vibration) {
+    if (!is_initalized) {
+        return false;
+    }
     if (device_index >= output_devices.size()) {
         return false;
     }
@@ -1242,6 +1247,10 @@ bool EmulatedController::IsVibrationEnabled(std::size_t device_index) {
     const auto player_index = Service::HID::NpadIdTypeToIndex(npad_id_type);
     const auto& player = Settings::values.players.GetValue()[player_index];
 
+    if (!is_initalized) {
+        return false;
+    }
+
     if (!player.vibration_enabled) {
         return false;
     }
@@ -1261,6 +1270,10 @@ Common::Input::DriverResult EmulatedController::SetPollingMode(
     EmulatedDeviceIndex device_index, Common::Input::PollingMode polling_mode) {
     LOG_INFO(Service_HID, "Set polling mode {}, device_index={}", polling_mode, device_index);
 
+    if (!is_initalized) {
+        return Common::Input::DriverResult::InvalidHandle;
+    }
+
     auto& left_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Left)];
     auto& right_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
     auto& nfc_output_device = output_devices[3];
@@ -1306,6 +1319,10 @@ bool EmulatedController::SetCameraFormat(
     Core::IrSensor::ImageTransferProcessorFormat camera_format) {
     LOG_INFO(Service_HID, "Set camera format {}", camera_format);
 
+    if (!is_initalized) {
+        return false;
+    }
+
     auto& right_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
     auto& camera_output_device = output_devices[2];
 
@@ -1329,6 +1346,11 @@ void EmulatedController::SetRingParam(Common::ParamPackage param) {
 }
 
 bool EmulatedController::HasNfc() const {
+
+    if (!is_initalized) {
+        return false;
+    }
+
     const auto& nfc_output_device = output_devices[3];
 
     switch (npad_type) {
@@ -1366,6 +1388,10 @@ bool EmulatedController::RemoveNfcHandle() {
 }
 
 bool EmulatedController::StartNfcPolling() {
+    if (!is_initalized) {
+        return false;
+    }
+
     auto& nfc_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
     auto& nfc_virtual_output_device = output_devices[3];
 
@@ -1377,6 +1403,10 @@ bool EmulatedController::StartNfcPolling() {
 }
 
 bool EmulatedController::StopNfcPolling() {
+    if (!is_initalized) {
+        return false;
+    }
+
     auto& nfc_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
     auto& nfc_virtual_output_device = output_devices[3];
 
@@ -1388,6 +1418,10 @@ bool EmulatedController::StopNfcPolling() {
 }
 
 bool EmulatedController::ReadAmiiboData(std::vector<u8>& data) {
+    if (!is_initalized) {
+        return false;
+    }
+
     auto& nfc_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
     auto& nfc_virtual_output_device = output_devices[3];
 
@@ -1400,6 +1434,10 @@ bool EmulatedController::ReadAmiiboData(std::vector<u8>& data) {
 
 bool EmulatedController::ReadMifareData(const Common::Input::MifareRequest& request,
                                         Common::Input::MifareRequest& out_data) {
+    if (!is_initalized) {
+        return false;
+    }
+
     auto& nfc_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
     auto& nfc_virtual_output_device = output_devices[3];
 
@@ -1412,6 +1450,10 @@ bool EmulatedController::ReadMifareData(const Common::Input::MifareRequest& requ
 }
 
 bool EmulatedController::WriteMifareData(const Common::Input::MifareRequest& request) {
+    if (!is_initalized) {
+        return false;
+    }
+
     auto& nfc_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
     auto& nfc_virtual_output_device = output_devices[3];
 
@@ -1423,6 +1465,10 @@ bool EmulatedController::WriteMifareData(const Common::Input::MifareRequest& req
 }
 
 bool EmulatedController::WriteNfc(const std::vector<u8>& data) {
+    if (!is_initalized) {
+        return false;
+    }
+
     auto& nfc_output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)];
     auto& nfc_virtual_output_device = output_devices[3];
 
@@ -1434,6 +1480,10 @@ bool EmulatedController::WriteNfc(const std::vector<u8>& data) {
 }
 
 void EmulatedController::SetLedPattern() {
+    if (!is_initalized) {
+        return;
+    }
+
     for (auto& device : output_devices) {
         if (!device) {
             continue;

src/core/hid/emulated_controller.h

@@ -559,6 +559,7 @@ private:
     NpadStyleTag supported_style_tag{NpadStyleSet::All};
     bool is_connected{false};
     bool is_configuring{false};
+    bool is_initalized{false};
     bool system_buttons_enabled{true};
     f32 motion_sensitivity{Core::HID::MotionInput::IsAtRestStandard};
     u32 turbo_button_state{0};

src/core/hle/kernel/code_set.h

@@ -75,12 +75,26 @@ struct CodeSet final {
         return segments[2];
     }
 
+#ifdef HAS_NCE
+    Segment& PatchSegment() {
+        return patch_segment;
+    }
+
+    const Segment& PatchSegment() const {
+        return patch_segment;
+    }
+#endif
+
     /// The overall data that backs this code set.
     Kernel::PhysicalMemory memory;
 
     /// The segments that comprise this code set.
     std::array<Segment, 3> segments;
 
+#ifdef HAS_NCE
+    Segment patch_segment;
+#endif
+
     /// The entry point address for this code set.
     KProcessAddress entrypoint = 0;
 };

src/core/hle/kernel/k_address_space_info.cpp

@@ -25,8 +25,8 @@ constexpr std::array<KAddressSpaceInfo, 13> AddressSpaceInfos{{
    { .bit_width = 36, .address = 2_GiB       , .size = 64_GiB  - 2_GiB  , .type = KAddressSpaceInfo::Type::MapLarge, },
    { .bit_width = 36, .address = Size_Invalid, .size = 8_GiB            , .type = KAddressSpaceInfo::Type::Heap,     },
    { .bit_width = 36, .address = Size_Invalid, .size = 6_GiB            , .type = KAddressSpaceInfo::Type::Alias,    },
-#ifdef ANDROID
-   // With Android, we use a 38-bit address space due to memory limitations. This should (safely) truncate ASLR region.
+#ifdef HAS_NCE
+   // With NCE, we use a 38-bit address space due to memory limitations. This should (safely) truncate ASLR region.
    { .bit_width = 39, .address = 128_MiB     , .size = 256_GiB - 128_MiB, .type = KAddressSpaceInfo::Type::Map39Bit, },
 #else
    { .bit_width = 39, .address = 128_MiB     , .size = 512_GiB - 128_MiB, .type = KAddressSpaceInfo::Type::Map39Bit, },

src/core/hle/kernel/k_page_table_base.cpp

@@ -88,6 +88,22 @@ Result FlushDataCache(AddressType addr, u64 size) {
     R_SUCCEED();
 }
 
+constexpr Common::MemoryPermission ConvertToMemoryPermission(KMemoryPermission perm) {
+    Common::MemoryPermission perms{};
+    if (True(perm & KMemoryPermission::UserRead)) {
+        perms |= Common::MemoryPermission::Read;
+    }
+    if (True(perm & KMemoryPermission::UserWrite)) {
+        perms |= Common::MemoryPermission::Write;
+    }
+#ifdef HAS_NCE
+    if (True(perm & KMemoryPermission::UserExecute)) {
+        perms |= Common::MemoryPermission::Execute;
+    }
+#endif
+    return perms;
+}
+
 } // namespace
 
 void KPageTableBase::MemoryRange::Open() {
@@ -170,7 +186,8 @@ Result KPageTableBase::InitializeForProcess(Svc::CreateProcessFlag as_type, bool
                                             KMemoryManager::Pool pool, KProcessAddress code_address,
                                             size_t code_size, KSystemResource* system_resource,
                                             KResourceLimit* resource_limit,
-                                            Core::Memory::Memory& memory) {
+                                            Core::Memory::Memory& memory,
+                                            KProcessAddress aslr_space_start) {
     // Calculate region extents.
     const size_t as_width = GetAddressSpaceWidth(as_type);
     const KProcessAddress start = 0;
@@ -211,7 +228,8 @@ Result KPageTableBase::InitializeForProcess(Svc::CreateProcessFlag as_type, bool
         heap_region_size = GetSpaceSize(KAddressSpaceInfo::Type::Heap);
         stack_region_size = GetSpaceSize(KAddressSpaceInfo::Type::Stack);
         kernel_map_region_size = GetSpaceSize(KAddressSpaceInfo::Type::MapSmall);
-        m_code_region_start = GetSpaceStart(KAddressSpaceInfo::Type::Map39Bit);
+        m_code_region_start = m_address_space_start + aslr_space_start +
+                              GetSpaceStart(KAddressSpaceInfo::Type::Map39Bit);
         m_code_region_end = m_code_region_start + GetSpaceSize(KAddressSpaceInfo::Type::Map39Bit);
         m_alias_code_region_start = m_code_region_start;
         m_alias_code_region_end = m_code_region_end;
@@ -5643,7 +5661,8 @@ Result KPageTableBase::Operate(PageLinkedList* page_list, KProcessAddress virt_a
     case OperationType::Map: {
         ASSERT(virt_addr != 0);
         ASSERT(Common::IsAligned(GetInteger(virt_addr), PageSize));
-        m_memory->MapMemoryRegion(*m_impl, virt_addr, num_pages * PageSize, phys_addr);
+        m_memory->MapMemoryRegion(*m_impl, virt_addr, num_pages * PageSize, phys_addr,
+                                  ConvertToMemoryPermission(properties.perm));
 
         // Open references to pages, if we should.
         if (this->IsHeapPhysicalAddress(phys_addr)) {
@@ -5658,8 +5677,11 @@ Result KPageTableBase::Operate(PageLinkedList* page_list, KProcessAddress virt_a
     }
     case OperationType::ChangePermissions:
     case OperationType::ChangePermissionsAndRefresh:
-    case OperationType::ChangePermissionsAndRefreshAndFlush:
+    case OperationType::ChangePermissionsAndRefreshAndFlush: {
+        m_memory->ProtectRegion(*m_impl, virt_addr, num_pages * PageSize,
+                                ConvertToMemoryPermission(properties.perm));
         R_SUCCEED();
+    }
     default:
         UNREACHABLE();
     }
@@ -5687,7 +5709,8 @@ Result KPageTableBase::Operate(PageLinkedList* page_list, KProcessAddress virt_a
             const size_t size{node.GetNumPages() * PageSize};
 
             // Map the pages.
-            m_memory->MapMemoryRegion(*m_impl, virt_addr, size, node.GetAddress());
+            m_memory->MapMemoryRegion(*m_impl, virt_addr, size, node.GetAddress(),
+                                      ConvertToMemoryPermission(properties.perm));
 
             virt_addr += size;
         }

src/core/hle/kernel/k_page_table_base.h

@@ -235,7 +235,8 @@ public:
                                 bool enable_device_address_space_merge, bool from_back,
                                 KMemoryManager::Pool pool, KProcessAddress code_address,
                                 size_t code_size, KSystemResource* system_resource,
-                                KResourceLimit* resource_limit, Core::Memory::Memory& memory);
+                                KResourceLimit* resource_limit, Core::Memory::Memory& memory,
+                                KProcessAddress aslr_space_start);
 
     void Finalize();
 

src/core/hle/kernel/k_process.cpp

@@ -300,7 +300,7 @@ Result KProcess::Initialize(const Svc::CreateProcessParameter& params, const KPa
             False(params.flags & Svc::CreateProcessFlag::DisableDeviceAddressSpaceMerge);
         R_TRY(m_page_table.Initialize(as_type, enable_aslr, enable_das_merge, !enable_aslr, pool,
                                       params.code_address, params.code_num_pages * PageSize,
-                                      m_system_resource, res_limit, this->GetMemory()));
+                                      m_system_resource, res_limit, this->GetMemory(), 0));
     }
     ON_RESULT_FAILURE_2 {
         m_page_table.Finalize();
@@ -332,7 +332,7 @@ Result KProcess::Initialize(const Svc::CreateProcessParameter& params, const KPa
 
 Result KProcess::Initialize(const Svc::CreateProcessParameter& params,
                             std::span<const u32> user_caps, KResourceLimit* res_limit,
-                            KMemoryManager::Pool pool) {
+                            KMemoryManager::Pool pool, KProcessAddress aslr_space_start) {
     ASSERT(res_limit != nullptr);
 
     // Set members.
@@ -393,7 +393,7 @@ Result KProcess::Initialize(const Svc::CreateProcessParameter& params,
             False(params.flags & Svc::CreateProcessFlag::DisableDeviceAddressSpaceMerge);
         R_TRY(m_page_table.Initialize(as_type, enable_aslr, enable_das_merge, !enable_aslr, pool,
                                       params.code_address, code_size, m_system_resource, res_limit,
-                                      this->GetMemory()));
+                                      this->GetMemory(), aslr_space_start));
     }
     ON_RESULT_FAILURE_2 {
         m_page_table.Finalize();
@@ -1128,7 +1128,7 @@ KProcess::KProcess(KernelCore& kernel)
 KProcess::~KProcess() = default;
 
 Result KProcess::LoadFromMetadata(const FileSys::ProgramMetadata& metadata, std::size_t code_size,
-                                  bool is_hbl) {
+                                  KProcessAddress aslr_space_start, bool is_hbl) {
     // Create a resource limit for the process.
     const auto physical_memory_size =
         m_kernel.MemoryManager().GetSize(Kernel::KMemoryManager::Pool::Application);
@@ -1179,7 +1179,7 @@ Result KProcess::LoadFromMetadata(const FileSys::ProgramMetadata& metadata, std:
         .name = {},
         .version = {},
         .program_id = metadata.GetTitleID(),
-        .code_address = code_address,
+        .code_address = code_address + GetInteger(aslr_space_start),
         .code_num_pages = static_cast<s32>(code_size / PageSize),
         .flags = flag,
         .reslimit = Svc::InvalidHandle,
@@ -1193,7 +1193,7 @@ Result KProcess::LoadFromMetadata(const FileSys::ProgramMetadata& metadata, std:
 
     // Initialize for application process.
     R_TRY(this->Initialize(params, metadata.GetKernelCapabilities(), res_limit,
-                           KMemoryManager::Pool::Application));
+                           KMemoryManager::Pool::Application, aslr_space_start));
 
     // Assign remaining properties.
     m_is_hbl = is_hbl;
@@ -1214,6 +1214,17 @@ void KProcess::LoadModule(CodeSet code_set, KProcessAddress base_addr) {
     ReprotectSegment(code_set.CodeSegment(), Svc::MemoryPermission::ReadExecute);
     ReprotectSegment(code_set.RODataSegment(), Svc::MemoryPermission::Read);
     ReprotectSegment(code_set.DataSegment(), Svc::MemoryPermission::ReadWrite);
+
+#ifdef HAS_NCE
+    if (Settings::IsNceEnabled()) {
+        auto& buffer = m_kernel.System().DeviceMemory().buffer;
+        const auto& code = code_set.CodeSegment();
+        const auto& patch = code_set.PatchSegment();
+        buffer.Protect(GetInteger(base_addr + code.addr), code.size, true, true, true);
+        buffer.Protect(GetInteger(base_addr + patch.addr), patch.size, true, true, true);
+        ReprotectSegment(code_set.PatchSegment(), Svc::MemoryPermission::None);
+    }
+#endif
 }
 
 bool KProcess::InsertWatchpoint(KProcessAddress addr, u64 size, DebugWatchpointType type) {

src/core/hle/kernel/k_process.h

@@ -120,6 +120,9 @@ private:
     std::atomic<s64> m_num_ipc_messages{};
     std::atomic<s64> m_num_ipc_replies{};
     std::atomic<s64> m_num_ipc_receives{};
+#ifdef HAS_NCE
+    std::unordered_map<u64, u64> m_post_handlers{};
+#endif
 
 private:
     Result StartTermination();
@@ -150,7 +153,8 @@ public:
                       std::span<const u32> caps, KResourceLimit* res_limit,
                       KMemoryManager::Pool pool, bool immortal);
     Result Initialize(const Svc::CreateProcessParameter& params, std::span<const u32> user_caps,
-                      KResourceLimit* res_limit, KMemoryManager::Pool pool);
+                      KResourceLimit* res_limit, KMemoryManager::Pool pool,
+                      KProcessAddress aslr_space_start);
     void Exit();
 
     const char* GetName() const {
@@ -466,6 +470,12 @@ public:
 
     static void Switch(KProcess* cur_process, KProcess* next_process);
 
+#ifdef HAS_NCE
+    std::unordered_map<u64, u64>& GetPostHandlers() noexcept {
+        return m_post_handlers;
+    }
+#endif
+
 public:
     // Attempts to insert a watchpoint into a free slot. Returns false if none are available.
     bool InsertWatchpoint(KProcessAddress addr, u64 size, DebugWatchpointType type);
@@ -479,7 +489,7 @@ public:
 
 public:
     Result LoadFromMetadata(const FileSys::ProgramMetadata& metadata, std::size_t code_size,
-                            bool is_hbl);
+                            KProcessAddress aslr_space_start, bool is_hbl);
 
     void LoadModule(CodeSet code_set, KProcessAddress base_addr);
 

src/core/hle/kernel/k_process_page_table.h

@@ -23,10 +23,11 @@ public:
     Result Initialize(Svc::CreateProcessFlag as_type, bool enable_aslr, bool enable_das_merge,
                       bool from_back, KMemoryManager::Pool pool, KProcessAddress code_address,
                       size_t code_size, KSystemResource* system_resource,
-                      KResourceLimit* resource_limit, Core::Memory::Memory& memory) {
-        R_RETURN(m_page_table.InitializeForProcess(as_type, enable_aslr, enable_das_merge,
-                                                   from_back, pool, code_address, code_size,
-                                                   system_resource, resource_limit, memory));
+                      KResourceLimit* resource_limit, Core::Memory::Memory& memory,
+                      KProcessAddress aslr_space_start) {
+        R_RETURN(m_page_table.InitializeForProcess(
+            as_type, enable_aslr, enable_das_merge, from_back, pool, code_address, code_size,
+            system_resource, resource_limit, memory, aslr_space_start));
     }
 
     void Finalize() {

src/core/hle/kernel/k_thread.h

@@ -655,6 +655,21 @@ public:
         return m_stack_top;
     }
 
+public:
+    // TODO: This shouldn't be defined in kernel namespace
+    struct NativeExecutionParameters {
+        u64 tpidr_el0{};
+        u64 tpidrro_el0{};
+        void* native_context{};
+        std::atomic<u32> lock{1};
+        bool is_running{};
+        u32 magic{Common::MakeMagic('Y', 'U', 'Z', 'U')};
+    };
+
+    NativeExecutionParameters& GetNativeExecutionParameters() {
+        return m_native_execution_parameters;
+    }
+
 private:
     KThread* RemoveWaiterByKey(bool* out_has_waiters, KProcessAddress key,
                                bool is_kernel_address_key);
@@ -914,6 +929,7 @@ private:
     ThreadWaitReasonForDebugging m_wait_reason_for_debugging{};
     uintptr_t m_argument{};
     KProcessAddress m_stack_top{};
+    NativeExecutionParameters m_native_execution_parameters{};
 
 public:
     using ConditionVariableThreadTreeType = ConditionVariableThreadTree;

src/core/hle/kernel/physical_core.cpp

@@ -1,8 +1,12 @@
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include "common/settings.h"
 #include "core/arm/dynarmic/arm_dynarmic_32.h"
 #include "core/arm/dynarmic/arm_dynarmic_64.h"
+#ifdef HAS_NCE
+#include "core/arm/nce/arm_nce.h"
+#endif
 #include "core/core.h"
 #include "core/hle/kernel/k_scheduler.h"
 #include "core/hle/kernel/kernel.h"
@@ -14,7 +18,8 @@ PhysicalCore::PhysicalCore(std::size_t core_index, Core::System& system, KSchedu
     : m_core_index{core_index}, m_system{system}, m_scheduler{scheduler} {
 #if defined(ARCHITECTURE_x86_64) || defined(ARCHITECTURE_arm64)
     // TODO(bunnei): Initialization relies on a core being available. We may later replace this with
-    // a 32-bit instance of Dynarmic. This should be abstracted out to a CPU manager.
+    // an NCE interface or a 32-bit instance of Dynarmic. This should be abstracted out to a CPU
+    // manager.
     auto& kernel = system.Kernel();
     m_arm_interface = std::make_unique<Core::ARM_Dynarmic_64>(
         system, kernel.IsMulticore(),
@@ -28,6 +33,13 @@ PhysicalCore::PhysicalCore(std::size_t core_index, Core::System& system, KSchedu
 PhysicalCore::~PhysicalCore() = default;
 
 void PhysicalCore::Initialize(bool is_64_bit) {
+#if defined(HAS_NCE)
+    if (Settings::IsNceEnabled()) {
+        m_arm_interface = std::make_unique<Core::ARM_NCE>(m_system, m_system.Kernel().IsMulticore(),
+                                                          m_core_index);
+        return;
+    }
+#endif
 #if defined(ARCHITECTURE_x86_64) || defined(ARCHITECTURE_arm64)
     auto& kernel = m_system.Kernel();
     if (!is_64_bit) {

src/core/hle/service/am/applets/applet_cabinet.cpp

@@ -122,7 +122,8 @@ void Cabinet::DisplayCompleted(bool apply_changes, std::string_view amiibo_name)
         Service::NFP::RegisterInfoPrivate register_info{};
         std::memcpy(register_info.amiibo_name.data(), amiibo_name.data(),
                     std::min(amiibo_name.size(), register_info.amiibo_name.size() - 1));
-
+        register_info.mii_store_data.BuildRandom(Mii::Age::All, Mii::Gender::All, Mii::Race::All);
+        register_info.mii_store_data.SetNickname({u'y', u'u', u'z', u'u'});
         nfp_device->SetRegisterInfoPrivate(register_info);
         break;
     }

src/core/hle/service/set/set_sys.cpp

@@ -19,19 +19,8 @@
 
 namespace Service::Set {
 
-namespace {
-constexpr u64 SYSTEM_VERSION_FILE_MINOR_REVISION_OFFSET = 0x05;
-
-enum class GetFirmwareVersionType {
-    Version1,
-    Version2,
-};
-
-void GetFirmwareVersionImpl(Core::System& system, HLERequestContext& ctx,
-                            GetFirmwareVersionType type) {
-    ASSERT_MSG(ctx.GetWriteBufferSize() == 0x100,
-               "FirmwareVersion output buffer must be 0x100 bytes in size!");
-
+Result GetFirmwareVersionImpl(FirmwareVersionFormat& out_firmware, Core::System& system,
+                              GetFirmwareVersionType type) {
     constexpr u64 FirmwareVersionSystemDataId = 0x0100000000000809;
     auto& fsc = system.GetFileSystemController();
 
@@ -52,39 +41,34 @@ void GetFirmwareVersionImpl(Core::System& system, HLERequestContext& ctx,
             FileSys::SystemArchive::SynthesizeSystemArchive(FirmwareVersionSystemDataId));
     }
 
-    const auto early_exit_failure = [&ctx](std::string_view desc, Result code) {
+    const auto early_exit_failure = [](std::string_view desc, Result code) {
         LOG_ERROR(Service_SET, "General failure while attempting to resolve firmware version ({}).",
                   desc);
-        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(code);
+        return code;
     };
 
     const auto ver_file = romfs->GetFile("file");
     if (ver_file == nullptr) {
-        early_exit_failure("The system version archive didn't contain the file 'file'.",
-                           FileSys::ERROR_INVALID_ARGUMENT);
-        return;
+        return early_exit_failure("The system version archive didn't contain the file 'file'.",
+                                  FileSys::ERROR_INVALID_ARGUMENT);
     }
 
     auto data = ver_file->ReadAllBytes();
-    if (data.size() != 0x100) {
-        early_exit_failure("The system version file 'file' was not the correct size.",
-                           FileSys::ERROR_OUT_OF_BOUNDS);
-        return;
+    if (data.size() != sizeof(FirmwareVersionFormat)) {
+        return early_exit_failure("The system version file 'file' was not the correct size.",
+                                  FileSys::ERROR_OUT_OF_BOUNDS);
     }
 
+    std::memcpy(&out_firmware, data.data(), sizeof(FirmwareVersionFormat));
+
     // If the command is GetFirmwareVersion (as opposed to GetFirmwareVersion2), hardware will
     // zero out the REVISION_MINOR field.
     if (type == GetFirmwareVersionType::Version1) {
-        data[SYSTEM_VERSION_FILE_MINOR_REVISION_OFFSET] = 0;
+        out_firmware.revision_minor = 0;
     }
 
-    ctx.WriteBuffer(data);
-
-    IPC::ResponseBuilder rb{ctx, 2};
-    rb.Push(ResultSuccess);
+    return ResultSuccess;
 }
-} // Anonymous namespace
 
 void SET_SYS::SetLanguageCode(HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
@@ -98,12 +82,32 @@ void SET_SYS::SetLanguageCode(HLERequestContext& ctx) {
 
 void SET_SYS::GetFirmwareVersion(HLERequestContext& ctx) {
     LOG_DEBUG(Service_SET, "called");
-    GetFirmwareVersionImpl(system, ctx, GetFirmwareVersionType::Version1);
+
+    FirmwareVersionFormat firmware_data{};
+    const auto result =
+        GetFirmwareVersionImpl(firmware_data, system, GetFirmwareVersionType::Version1);
+
+    if (result.IsSuccess()) {
+        ctx.WriteBuffer(firmware_data);
+    }
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(result);
 }
 
 void SET_SYS::GetFirmwareVersion2(HLERequestContext& ctx) {
     LOG_DEBUG(Service_SET, "called");
-    GetFirmwareVersionImpl(system, ctx, GetFirmwareVersionType::Version2);
+
+    FirmwareVersionFormat firmware_data{};
+    const auto result =
+        GetFirmwareVersionImpl(firmware_data, system, GetFirmwareVersionType::Version2);
+
+    if (result.IsSuccess()) {
+        ctx.WriteBuffer(firmware_data);
+    }
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(result);
 }
 
 void SET_SYS::GetAccountSettings(HLERequestContext& ctx) {

src/core/hle/service/set/set_sys.h

@@ -4,6 +4,7 @@
 #pragma once
 
 #include "common/uuid.h"
+#include "core/hle/result.h"
 #include "core/hle/service/service.h"
 #include "core/hle/service/time/clock_types.h"
 
@@ -12,6 +13,29 @@ class System;
 }
 
 namespace Service::Set {
+enum class LanguageCode : u64;
+enum class GetFirmwareVersionType {
+    Version1,
+    Version2,
+};
+
+struct FirmwareVersionFormat {
+    u8 major;
+    u8 minor;
+    u8 micro;
+    INSERT_PADDING_BYTES(1);
+    u8 revision_major;
+    u8 revision_minor;
+    INSERT_PADDING_BYTES(2);
+    std::array<char, 0x20> platform;
+    std::array<u8, 0x40> version_hash;
+    std::array<char, 0x18> display_version;
+    std::array<char, 0x80> display_title;
+};
+static_assert(sizeof(FirmwareVersionFormat) == 0x100, "FirmwareVersionFormat is an invalid size");
+
+Result GetFirmwareVersionImpl(FirmwareVersionFormat& out_firmware, Core::System& system,
+                              GetFirmwareVersionType type);
 
 class SET_SYS final : public ServiceFramework<SET_SYS> {
 public:

src/core/hle/service/time/clock_types.h

@@ -11,6 +11,11 @@
 #include "core/hle/service/time/errors.h"
 #include "core/hle/service/time/time_zone_types.h"
 
+// Defined by WinBase.h on Windows
+#ifdef GetCurrentTime
+#undef GetCurrentTime
+#endif
+
 namespace Service::Time::Clock {
 
 enum class TimeType : u8 {

src/core/loader/deconstructed_rom_directory.cpp

@@ -3,6 +3,7 @@
 
 #include <cstring>
 #include "common/logging/log.h"
+#include "common/settings.h"
 #include "core/core.h"
 #include "core/file_sys/content_archive.h"
 #include "core/file_sys/control_metadata.h"
@@ -14,6 +15,10 @@
 #include "core/loader/deconstructed_rom_directory.h"
 #include "core/loader/nso.h"
 
+#ifdef HAS_NCE
+#include "core/arm/nce/patch.h"
+#endif
+
 namespace Loader {
 
 AppLoader_DeconstructedRomDirectory::AppLoader_DeconstructedRomDirectory(FileSys::VirtualFile file_,
@@ -124,21 +129,43 @@ AppLoader_DeconstructedRomDirectory::LoadResult AppLoader_DeconstructedRomDirect
     }
     metadata.Print();
 
-    const auto static_modules = {"rtld",    "main",    "subsdk0", "subsdk1", "subsdk2",
-                                 "subsdk3", "subsdk4", "subsdk5", "subsdk6", "subsdk7",
-                                 "subsdk8", "subsdk9", "sdk"};
+    // Enable NCE only for programs with 39-bit address space.
+    const bool is_39bit =
+        metadata.GetAddressSpaceType() == FileSys::ProgramAddressSpaceType::Is39Bit;
+    Settings::SetNceEnabled(is_39bit);
+
+    const std::array static_modules = {"rtld",    "main",    "subsdk0", "subsdk1", "subsdk2",
+                                       "subsdk3", "subsdk4", "subsdk5", "subsdk6", "subsdk7",
+                                       "subsdk8", "subsdk9", "sdk"};
+
+    std::size_t code_size{};
+
+    // Define an nce patch context for each potential module.
+#ifdef HAS_NCE
+    std::array<Core::NCE::Patcher, 13> module_patchers;
+#endif
+
+    const auto GetPatcher = [&](size_t i) -> Core::NCE::Patcher* {
+#ifdef HAS_NCE
+        if (Settings::IsNceEnabled()) {
+            return &module_patchers[i];
+        }
+#endif
+        return nullptr;
+    };
 
     // Use the NSO module loader to figure out the code layout
-    std::size_t code_size{};
-    for (const auto& module : static_modules) {
+    for (size_t i = 0; i < static_modules.size(); i++) {
+        const auto& module = static_modules[i];
         const FileSys::VirtualFile module_file{dir->GetFile(module)};
         if (!module_file) {
             continue;
         }
 
         const bool should_pass_arguments = std::strcmp(module, "rtld") == 0;
-        const auto tentative_next_load_addr = AppLoader_NSO::LoadModule(
-            process, system, *module_file, code_size, should_pass_arguments, false);
+        const auto tentative_next_load_addr =
+            AppLoader_NSO::LoadModule(process, system, *module_file, code_size,
+                                      should_pass_arguments, false, {}, GetPatcher(i));
         if (!tentative_next_load_addr) {
             return {ResultStatus::ErrorLoadingNSO, {}};
         }
@@ -146,8 +173,18 @@ AppLoader_DeconstructedRomDirectory::LoadResult AppLoader_DeconstructedRomDirect
         code_size = *tentative_next_load_addr;
     }
 
+    // Enable direct memory mapping in case of NCE.
+    const u64 fastmem_base = [&]() -> size_t {
+        if (Settings::IsNceEnabled()) {
+            auto& buffer = system.DeviceMemory().buffer;
+            buffer.EnableDirectMappedAddress();
+            return reinterpret_cast<u64>(buffer.VirtualBasePointer());
+        }
+        return 0;
+    }();
+
     // Setup the process code layout
-    if (process.LoadFromMetadata(metadata, code_size, is_hbl).IsError()) {
+    if (process.LoadFromMetadata(metadata, code_size, fastmem_base, is_hbl).IsError()) {
         return {ResultStatus::ErrorUnableToParseKernelMetadata, {}};
     }
 
@@ -157,7 +194,8 @@ AppLoader_DeconstructedRomDirectory::LoadResult AppLoader_DeconstructedRomDirect
     VAddr next_load_addr{base_address};
     const FileSys::PatchManager pm{metadata.GetTitleID(), system.GetFileSystemController(),
                                    system.GetContentProvider()};
-    for (const auto& module : static_modules) {
+    for (size_t i = 0; i < static_modules.size(); i++) {
+        const auto& module = static_modules[i];
         const FileSys::VirtualFile module_file{dir->GetFile(module)};
         if (!module_file) {
             continue;
@@ -165,15 +203,16 @@ AppLoader_DeconstructedRomDirectory::LoadResult AppLoader_DeconstructedRomDirect
 
         const VAddr load_addr{next_load_addr};
         const bool should_pass_arguments = std::strcmp(module, "rtld") == 0;
-        const auto tentative_next_load_addr = AppLoader_NSO::LoadModule(
-            process, system, *module_file, load_addr, should_pass_arguments, true, pm);
+        const auto tentative_next_load_addr =
+            AppLoader_NSO::LoadModule(process, system, *module_file, load_addr,
+                                      should_pass_arguments, true, pm, GetPatcher(i));
         if (!tentative_next_load_addr) {
             return {ResultStatus::ErrorLoadingNSO, {}};
         }
 
         next_load_addr = *tentative_next_load_addr;
         modules.insert_or_assign(load_addr, module);
-        LOG_DEBUG(Loader, "loaded module {} @ 0x{:X}", module, load_addr);
+        LOG_DEBUG(Loader, "loaded module {} @ {:#X}", module, load_addr);
     }
 
     // Find the RomFS by searching for a ".romfs" file in this directory

src/core/loader/kip.cpp

@@ -91,7 +91,8 @@ AppLoader::LoadResult AppLoader_KIP::Load(Kernel::KProcess& process,
 
     // Setup the process code layout
     if (process
-            .LoadFromMetadata(FileSys::ProgramMetadata::GetDefault(), program_image.size(), false)
+            .LoadFromMetadata(FileSys::ProgramMetadata::GetDefault(), program_image.size(), 0,
+                              false)
             .IsError()) {
         return {ResultStatus::ErrorNotInitialized, {}};
     }

src/core/loader/nro.cpp

@@ -22,6 +22,10 @@
 #include "core/loader/nso.h"
 #include "core/memory.h"
 
+#ifdef HAS_NCE
+#include "core/arm/nce/patch.h"
+#endif
+
 namespace Loader {
 
 struct NroSegmentHeader {
@@ -139,7 +143,8 @@ static constexpr u32 PageAlignSize(u32 size) {
     return static_cast<u32>((size + Core::Memory::YUZU_PAGEMASK) & ~Core::Memory::YUZU_PAGEMASK);
 }
 
-static bool LoadNroImpl(Kernel::KProcess& process, const std::vector<u8>& data) {
+static bool LoadNroImpl(Core::System& system, Kernel::KProcess& process,
+                        const std::vector<u8>& data) {
     if (data.size() < sizeof(NroHeader)) {
         return {};
     }
@@ -194,14 +199,61 @@ static bool LoadNroImpl(Kernel::KProcess& process, const std::vector<u8>& data)
 
     codeset.DataSegment().size += bss_size;
     program_image.resize(static_cast<u32>(program_image.size()) + bss_size);
+    size_t image_size = program_image.size();
+
+#ifdef HAS_NCE
+    const auto& code = codeset.CodeSegment();
+
+    // NROs always have a 39-bit address space.
+    Settings::SetNceEnabled(true);
+
+    // Create NCE patcher
+    Core::NCE::Patcher patch{};
+
+    if (Settings::IsNceEnabled()) {
+        // Patch SVCs and MRS calls in the guest code
+        patch.PatchText(program_image, code);
+
+        // We only support PostData patching for NROs.
+        ASSERT(patch.GetPatchMode() == Core::NCE::PatchMode::PostData);
+
+        // Update patch section.
+        auto& patch_segment = codeset.PatchSegment();
+        patch_segment.addr = image_size;
+        patch_segment.size = static_cast<u32>(patch.GetSectionSize());
+
+        // Add patch section size to the module size.
+        image_size += patch_segment.size;
+    }
+#endif
+
+    // Enable direct memory mapping in case of NCE.
+    const u64 fastmem_base = [&]() -> size_t {
+        if (Settings::IsNceEnabled()) {
+            auto& buffer = system.DeviceMemory().buffer;
+            buffer.EnableDirectMappedAddress();
+            return reinterpret_cast<u64>(buffer.VirtualBasePointer());
+        }
+        return 0;
+    }();
 
     // Setup the process code layout
     if (process
-            .LoadFromMetadata(FileSys::ProgramMetadata::GetDefault(), program_image.size(), false)
+            .LoadFromMetadata(FileSys::ProgramMetadata::GetDefault(), image_size, fastmem_base,
+                              false)
             .IsError()) {
         return false;
     }
 
+    // Relocate code patch and copy to the program_image if running under NCE.
+    // This needs to be after LoadFromMetadata so we can use the process entry point.
+#ifdef HAS_NCE
+    if (Settings::IsNceEnabled()) {
+        patch.RelocateAndCopy(process.GetEntryPoint(), code, program_image,
+                              &process.GetPostHandlers());
+    }
+#endif
+
     // Load codeset for current process
     codeset.memory = std::move(program_image);
     process.LoadModule(std::move(codeset), process.GetEntryPoint());
@@ -209,8 +261,9 @@ static bool LoadNroImpl(Kernel::KProcess& process, const std::vector<u8>& data)
     return true;
 }
 
-bool AppLoader_NRO::LoadNro(Kernel::KProcess& process, const FileSys::VfsFile& nro_file) {
-    return LoadNroImpl(process, nro_file.ReadAllBytes());
+bool AppLoader_NRO::LoadNro(Core::System& system, Kernel::KProcess& process,
+                            const FileSys::VfsFile& nro_file) {
+    return LoadNroImpl(system, process, nro_file.ReadAllBytes());
 }
 
 AppLoader_NRO::LoadResult AppLoader_NRO::Load(Kernel::KProcess& process, Core::System& system) {
@@ -218,7 +271,7 @@ AppLoader_NRO::LoadResult AppLoader_NRO::Load(Kernel::KProcess& process, Core::S
         return {ResultStatus::ErrorAlreadyLoaded, {}};
     }
 
-    if (!LoadNro(process, *file)) {
+    if (!LoadNro(system, process, *file)) {
         return {ResultStatus::ErrorLoadingNRO, {}};
     }
 

src/core/loader/nro.h

@@ -54,7 +54,7 @@ public:
     bool IsRomFSUpdatable() const override;
 
 private:
-    bool LoadNro(Kernel::KProcess& process, const FileSys::VfsFile& nro_file);
+    bool LoadNro(Core::System& system, Kernel::KProcess& process, const FileSys::VfsFile& nro_file);
 
     std::vector<u8> icon_data;
     std::unique_ptr<FileSys::NACP> nacp;

src/core/loader/nso.cpp

@@ -20,6 +20,10 @@
 #include "core/loader/nso.h"
 #include "core/memory.h"
 
+#ifdef HAS_NCE
+#include "core/arm/nce/patch.h"
+#endif
+
 namespace Loader {
 namespace {
 struct MODHeader {
@@ -72,7 +76,8 @@ FileType AppLoader_NSO::IdentifyType(const FileSys::VirtualFile& in_file) {
 std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::KProcess& process, Core::System& system,
                                                const FileSys::VfsFile& nso_file, VAddr load_base,
                                                bool should_pass_arguments, bool load_into_process,
-                                               std::optional<FileSys::PatchManager> pm) {
+                                               std::optional<FileSys::PatchManager> pm,
+                                               Core::NCE::Patcher* patch) {
     if (nso_file.GetSize() < sizeof(NSOHeader)) {
         return std::nullopt;
     }
@@ -86,6 +91,16 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::KProcess& process, Core::
         return std::nullopt;
     }
 
+    // Allocate some space at the beginning if we are patching in PreText mode.
+    const size_t module_start = [&]() -> size_t {
+#ifdef HAS_NCE
+        if (patch && patch->GetPatchMode() == Core::NCE::PatchMode::PreText) {
+            return patch->GetSectionSize();
+        }
+#endif
+        return 0;
+    }();
+
     // Build program image
     Kernel::CodeSet codeset;
     Kernel::PhysicalMemory program_image;
@@ -95,11 +110,12 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::KProcess& process, Core::
         if (nso_header.IsSegmentCompressed(i)) {
             data = DecompressSegment(data, nso_header.segments[i]);
         }
-        program_image.resize(nso_header.segments[i].location + static_cast<u32>(data.size()));
-        std::memcpy(program_image.data() + nso_header.segments[i].location, data.data(),
-                    data.size());
-        codeset.segments[i].addr = nso_header.segments[i].location;
-        codeset.segments[i].offset = nso_header.segments[i].location;
+        program_image.resize(module_start + nso_header.segments[i].location +
+                             static_cast<u32>(data.size()));
+        std::memcpy(program_image.data() + module_start + nso_header.segments[i].location,
+                    data.data(), data.size());
+        codeset.segments[i].addr = module_start + nso_header.segments[i].location;
+        codeset.segments[i].offset = module_start + nso_header.segments[i].location;
         codeset.segments[i].size = nso_header.segments[i].size;
     }
 
@@ -118,7 +134,7 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::KProcess& process, Core::
     }
 
     codeset.DataSegment().size += nso_header.segments[2].bss_size;
-    const u32 image_size{
+    u32 image_size{
         PageAlignSize(static_cast<u32>(program_image.size()) + nso_header.segments[2].bss_size)};
     program_image.resize(image_size);
 
@@ -129,16 +145,45 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::KProcess& process, Core::
     // Apply patches if necessary
     const auto name = nso_file.GetName();
     if (pm && (pm->HasNSOPatch(nso_header.build_id, name) || Settings::values.dump_nso)) {
-        std::vector<u8> pi_header(sizeof(NSOHeader) + program_image.size());
+        std::span<u8> patchable_section(program_image.data() + module_start,
+                                        program_image.size() - module_start);
+        std::vector<u8> pi_header(sizeof(NSOHeader) + patchable_section.size());
         std::memcpy(pi_header.data(), &nso_header, sizeof(NSOHeader));
-        std::memcpy(pi_header.data() + sizeof(NSOHeader), program_image.data(),
-                    program_image.size());
+        std::memcpy(pi_header.data() + sizeof(NSOHeader), patchable_section.data(),
+                    patchable_section.size());
 
         pi_header = pm->PatchNSO(pi_header, name);
 
-        std::copy(pi_header.begin() + sizeof(NSOHeader), pi_header.end(), program_image.data());
+        std::copy(pi_header.begin() + sizeof(NSOHeader), pi_header.end(), patchable_section.data());
     }
 
+#ifdef HAS_NCE
+    // If we are computing the process code layout and using nce backend, patch.
+    const auto& code = codeset.CodeSegment();
+    if (patch && patch->GetPatchMode() == Core::NCE::PatchMode::None) {
+        // Patch SVCs and MRS calls in the guest code
+        patch->PatchText(program_image, code);
+
+        // Add patch section size to the module size.
+        image_size += static_cast<u32>(patch->GetSectionSize());
+    } else if (patch) {
+        // Relocate code patch and copy to the program_image.
+        patch->RelocateAndCopy(load_base, code, program_image, &process.GetPostHandlers());
+
+        // Update patch section.
+        auto& patch_segment = codeset.PatchSegment();
+        patch_segment.addr =
+            patch->GetPatchMode() == Core::NCE::PatchMode::PreText ? 0 : image_size;
+        patch_segment.size = static_cast<u32>(patch->GetSectionSize());
+
+        // Add patch section size to the module size. In PreText mode image_size
+        // already contains the patch segment as part of module_start.
+        if (patch->GetPatchMode() == Core::NCE::PatchMode::PostData) {
+            image_size += patch_segment.size;
+        }
+    }
+#endif
+
     // If we aren't actually loading (i.e. just computing the process code layout), we are done
     if (!load_into_process) {
         return load_base + image_size;

src/core/loader/nso.h

@@ -15,6 +15,10 @@ namespace Core {
 class System;
 }
 
+namespace Core::NCE {
+class Patcher;
+}
+
 namespace Kernel {
 class KProcess;
 }
@@ -88,7 +92,8 @@ public:
     static std::optional<VAddr> LoadModule(Kernel::KProcess& process, Core::System& system,
                                            const FileSys::VfsFile& nso_file, VAddr load_base,
                                            bool should_pass_arguments, bool load_into_process,
-                                           std::optional<FileSys::PatchManager> pm = {});
+                                           std::optional<FileSys::PatchManager> pm = {},
+                                           Core::NCE::Patcher* patch = nullptr);
 
     LoadResult Load(Kernel::KProcess& process, Core::System& system) override;
 

src/core/memory.cpp

@@ -53,7 +53,7 @@ struct Memory::Impl {
     }
 
     void MapMemoryRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size,
-                         Common::PhysicalAddress target) {
+                         Common::PhysicalAddress target, Common::MemoryPermission perms) {
         ASSERT_MSG((size & YUZU_PAGEMASK) == 0, "non-page aligned size: {:016X}", size);
         ASSERT_MSG((base & YUZU_PAGEMASK) == 0, "non-page aligned base: {:016X}", GetInteger(base));
         ASSERT_MSG(target >= DramMemoryMap::Base, "Out of bounds target: {:016X}",
@@ -63,7 +63,7 @@ struct Memory::Impl {
 
         if (Settings::IsFastmemEnabled()) {
             system.DeviceMemory().buffer.Map(GetInteger(base),
-                                             GetInteger(target) - DramMemoryMap::Base, size);
+                                             GetInteger(target) - DramMemoryMap::Base, size, perms);
         }
     }
 
@@ -78,6 +78,51 @@ struct Memory::Impl {
         }
     }
 
+    void ProtectRegion(Common::PageTable& page_table, VAddr vaddr, u64 size,
+                       Common::MemoryPermission perms) {
+        ASSERT_MSG((size & YUZU_PAGEMASK) == 0, "non-page aligned size: {:016X}", size);
+        ASSERT_MSG((vaddr & YUZU_PAGEMASK) == 0, "non-page aligned base: {:016X}", vaddr);
+
+        if (!Settings::IsFastmemEnabled()) {
+            return;
+        }
+
+        const bool is_r = True(perms & Common::MemoryPermission::Read);
+        const bool is_w = True(perms & Common::MemoryPermission::Write);
+        const bool is_x =
+            True(perms & Common::MemoryPermission::Execute) && Settings::IsNceEnabled();
+
+        if (!current_page_table) {
+            system.DeviceMemory().buffer.Protect(vaddr, size, is_r, is_w, is_x);
+            return;
+        }
+
+        u64 protect_bytes{};
+        u64 protect_begin{};
+        for (u64 addr = vaddr; addr < vaddr + size; addr += YUZU_PAGESIZE) {
+            const Common::PageType page_type{
+                current_page_table->pointers[addr >> YUZU_PAGEBITS].Type()};
+            switch (page_type) {
+            case Common::PageType::RasterizerCachedMemory:
+                if (protect_bytes > 0) {
+                    system.DeviceMemory().buffer.Protect(protect_begin, protect_bytes, is_r, is_w,
+                                                         is_x);
+                    protect_bytes = 0;
+                }
+                break;
+            default:
+                if (protect_bytes == 0) {
+                    protect_begin = addr;
+                }
+                protect_bytes += YUZU_PAGESIZE;
+            }
+        }
+
+        if (protect_bytes > 0) {
+            system.DeviceMemory().buffer.Protect(protect_begin, protect_bytes, is_r, is_w, is_x);
+        }
+    }
+
     [[nodiscard]] u8* GetPointerFromRasterizerCachedMemory(u64 vaddr) const {
         const Common::PhysicalAddress paddr{
             current_page_table->backing_addr[vaddr >> YUZU_PAGEBITS]};
@@ -831,14 +876,19 @@ void Memory::SetCurrentPageTable(Kernel::KProcess& process, u32 core_id) {
 }
 
 void Memory::MapMemoryRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size,
-                             Common::PhysicalAddress target) {
-    impl->MapMemoryRegion(page_table, base, size, target);
+                             Common::PhysicalAddress target, Common::MemoryPermission perms) {
+    impl->MapMemoryRegion(page_table, base, size, target, perms);
 }
 
 void Memory::UnmapRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size) {
     impl->UnmapRegion(page_table, base, size);
 }
 
+void Memory::ProtectRegion(Common::PageTable& page_table, Common::ProcessAddress vaddr, u64 size,
+                           Common::MemoryPermission perms) {
+    impl->ProtectRegion(page_table, GetInteger(vaddr), size, perms);
+}
+
 bool Memory::IsValidVirtualAddress(const Common::ProcessAddress vaddr) const {
     const Kernel::KProcess& process = *system.ApplicationProcess();
     const auto& page_table = process.GetPageTable().GetImpl();
@@ -1001,4 +1051,17 @@ void Memory::FlushRegion(Common::ProcessAddress dest_addr, size_t size) {
     impl->FlushRegion(dest_addr, size);
 }
 
+bool Memory::InvalidateNCE(Common::ProcessAddress vaddr, size_t size) {
+    bool mapped = true;
+    u8* const ptr = impl->GetPointerImpl(
+        GetInteger(vaddr),
+        [&] {
+            LOG_ERROR(HW_Memory, "Unmapped InvalidateNCE for {} bytes @ {:#x}", size,
+                      GetInteger(vaddr));
+            mapped = false;
+        },
+        [&] { impl->system.GPU().InvalidateRegion(GetInteger(vaddr), size); });
+    return mapped && ptr != nullptr;
+}
+
 } // namespace Core::Memory

src/core/memory.h

@@ -15,8 +15,9 @@
 #include "core/hle/result.h"
 
 namespace Common {
+enum class MemoryPermission : u32;
 struct PageTable;
-}
+} // namespace Common
 
 namespace Core {
 class System;
@@ -82,9 +83,10 @@ public:
      * @param size       The amount of bytes to map. Must be page-aligned.
      * @param target     Buffer with the memory backing the mapping. Must be of length at least
      *                   `size`.
+     * @param perms      The permissions to map the memory with.
      */
     void MapMemoryRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size,
-                         Common::PhysicalAddress target);
+                         Common::PhysicalAddress target, Common::MemoryPermission perms);
 
     /**
      * Unmaps a region of the emulated process address space.
@@ -95,6 +97,17 @@ public:
      */
     void UnmapRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size);
 
+    /**
+     * Protects a region of the emulated process address space with the new permissions.
+     *
+     * @param page_table The page table of the emulated process.
+     * @param base       The start address to re-protect. Must be page-aligned.
+     * @param size       The amount of bytes to protect. Must be page-aligned.
+     * @param perms      The permissions the address range is mapped.
+     */
+    void ProtectRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size,
+                       Common::MemoryPermission perms);
+
     /**
      * Checks whether or not the supplied address is a valid virtual
      * address for the current process.
@@ -472,6 +485,7 @@ public:
 
     void SetGPUDirtyManagers(std::span<Core::GPUDirtyMemoryManager> managers);
     void InvalidateRegion(Common::ProcessAddress dest_addr, size_t size);
+    bool InvalidateNCE(Common::ProcessAddress vaddr, size_t size);
     void FlushRegion(Common::ProcessAddress dest_addr, size_t size);
 
 private:

src/frontend_common/CMakeLists.txt

@@ -0,0 +1,10 @@
+# SPDX-FileCopyrightText: 2023 yuzu Emulator Project
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+add_library(frontend_common STATIC
+    config.cpp
+    config.h
+)
+
+create_target_directory_groups(frontend_common)
+target_link_libraries(frontend_common PUBLIC core SimpleIni::SimpleIni PRIVATE common Boost::headers)

src/frontend_common/config.h

@@ -0,0 +1,211 @@
+// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <memory>
+#include <string>
+#include "common/settings.h"
+
+#define SI_NO_CONVERSION
+#include <SimpleIni.h>
+#include <boost/algorithm/string/replace.hpp>
+
+// Workaround for conflicting definition in libloaderapi.h caused by SimpleIni
+#undef LoadString
+#undef CreateFile
+#undef DeleteFile
+#undef CopyFile
+#undef CreateDirectory
+#undef MoveFile
+
+namespace Core {
+class System;
+}
+
+class Config {
+public:
+    enum class ConfigType {
+        GlobalConfig,
+        PerGameConfig,
+        InputProfile,
+    };
+
+    virtual ~Config() = default;
+
+    void ClearControlPlayerValues() const;
+
+    [[nodiscard]] const std::string& GetConfigFilePath() const;
+
+    [[nodiscard]] bool Exists(const std::string& section, const std::string& key) const;
+
+protected:
+    explicit Config(ConfigType config_type = ConfigType::GlobalConfig);
+
+    void Initialize(const std::string& config_name = "config");
+    void Initialize(std::optional<std::string> config_path);
+
+    void WriteToIni() const;
+
+    void SetUpIni();
+    [[nodiscard]] bool IsCustomConfig() const;
+
+    void Reload();
+    void Save();
+
+    /**
+     * Derived config classes must implement this so they can reload all platform-specific
+     * values and global ones.
+     */
+    virtual void ReloadAllValues() = 0;
+
+    /**
+     * Derived config classes must implement this so they can save all platform-specific
+     * and global values.
+     */
+    virtual void SaveAllValues() = 0;
+
+    void ReadValues();
+    void ReadPlayerValues(std::size_t player_index);
+
+    void ReadTouchscreenValues();
+    void ReadMotionTouchValues();
+
+    // Read functions bases off the respective config section names.
+    void ReadAudioValues();
+    void ReadControlValues();
+    void ReadCoreValues();
+    void ReadDataStorageValues();
+    void ReadDebuggingValues();
+    void ReadServiceValues();
+    void ReadDisabledAddOnValues();
+    void ReadMiscellaneousValues();
+    void ReadCpuValues();
+    void ReadRendererValues();
+    void ReadScreenshotValues();
+    void ReadSystemValues();
+    void ReadWebServiceValues();
+    void ReadNetworkValues();
+
+    // Read platform specific sections
+    virtual void ReadHidbusValues() = 0;
+    virtual void ReadDebugControlValues() = 0;
+    virtual void ReadPathValues() = 0;
+    virtual void ReadShortcutValues() = 0;
+    virtual void ReadUIValues() = 0;
+    virtual void ReadUIGamelistValues() = 0;
+    virtual void ReadUILayoutValues() = 0;
+    virtual void ReadMultiplayerValues() = 0;
+
+    void SaveValues();
+    void SavePlayerValues(std::size_t player_index);
+    void SaveTouchscreenValues();
+    void SaveMotionTouchValues();
+
+    // Save functions based off the respective config section names.
+    void SaveAudioValues();
+    void SaveControlValues();
+    void SaveCoreValues();
+    void SaveDataStorageValues();
+    void SaveDebuggingValues();
+    void SaveNetworkValues();
+    void SaveDisabledAddOnValues();
+    void SaveMiscellaneousValues();
+    void SaveCpuValues();
+    void SaveRendererValues();
+    void SaveScreenshotValues();
+    void SaveSystemValues();
+    void SaveWebServiceValues();
+
+    // Save platform specific sections
+    virtual void SaveHidbusValues() = 0;
+    virtual void SaveDebugControlValues() = 0;
+    virtual void SavePathValues() = 0;
+    virtual void SaveShortcutValues() = 0;
+    virtual void SaveUIValues() = 0;
+    virtual void SaveUIGamelistValues() = 0;
+    virtual void SaveUILayoutValues() = 0;
+    virtual void SaveMultiplayerValues() = 0;
+
+    virtual std::vector<Settings::BasicSetting*>& FindRelevantList(Settings::Category category) = 0;
+
+    /**
+     * Reads a setting from the qt_config.
+     *
+     * @param key The setting's identifier
+     * @param default_value The value to use when the setting is not already present in the config
+     */
+    bool ReadBooleanSetting(const std::string& key,
+                            std::optional<bool> default_value = std::nullopt);
+    s64 ReadIntegerSetting(const std::string& key, std::optional<s64> default_value = std::nullopt);
+    u64 ReadUnsignedIntegerSetting(const std::string& key,
+                                   std::optional<u64> default_value = std::nullopt);
+    double ReadDoubleSetting(const std::string& key,
+                             std::optional<double> default_value = std::nullopt);
+    std::string ReadStringSetting(const std::string& key,
+                                  std::optional<std::string> default_value = std::nullopt);
+
+    /**
+     * Writes a setting to the qt_config.
+     *
+     * @param key The setting's idetentifier
+     * @param value Value of the setting
+     * @param default_value Default of the setting if not present in config
+     * @param use_global Specifies if the custom or global config should be in use, for custom
+     * configs
+     */
+    template <typename Type = int>
+    void WriteSetting(const std::string& key, const Type& value,
+                      const std::optional<Type>& default_value = std::nullopt,
+                      const std::optional<bool>& use_global = std::nullopt);
+    void WriteSettingInternal(const std::string& key, const std::string& value);
+
+    void ReadCategory(Settings::Category category);
+    void WriteCategory(Settings::Category category);
+    void ReadSettingGeneric(Settings::BasicSetting* setting);
+    void WriteSettingGeneric(const Settings::BasicSetting* setting);
+
+    template <typename T>
+    [[nodiscard]] std::string ToString(const T& value_) {
+        if constexpr (std::is_same_v<T, std::string>) {
+            return value_;
+        } else if constexpr (std::is_same_v<T, std::optional<u32>>) {
+            return value_.has_value() ? std::to_string(*value_) : "none";
+        } else if constexpr (std::is_same_v<T, bool>) {
+            return value_ ? "true" : "false";
+        } else if constexpr (std::is_same_v<T, u64>) {
+            return std::to_string(static_cast<u64>(value_));
+        } else {
+            return std::to_string(static_cast<s64>(value_));
+        }
+    }
+
+    void BeginGroup(const std::string& group);
+    void EndGroup();
+    std::string GetSection();
+    [[nodiscard]] std::string GetGroup() const;
+    static std::string AdjustKey(const std::string& key);
+    static std::string AdjustOutputString(const std::string& string);
+    std::string GetFullKey(const std::string& key, bool skipArrayIndex);
+    int BeginArray(const std::string& array);
+    void EndArray();
+    void SetArrayIndex(int index);
+
+    const ConfigType type;
+    std::unique_ptr<CSimpleIniA> config;
+    std::string config_loc;
+    const bool global;
+
+private:
+    inline static std::array<char, 19> special_characters = {'!', '#', '$',  '%',  '^', '&', '*',
+                                                             '|', ';', '\'', '\"', ',', '<', '.',
+                                                             '>', '?', '`',  '~',  '='};
+
+    struct ConfigArray {
+        std::string name;
+        int size;
+        int index;
+    };
+    std::vector<ConfigArray> array_stack;
+    std::vector<std::string> key_stack;
+};

src/tests/common/host_memory.cpp

@@ -11,6 +11,7 @@ using namespace Common::Literals;
 
 static constexpr size_t VIRTUAL_SIZE = 1ULL << 39;
 static constexpr size_t BACKING_SIZE = 4_GiB;
+static constexpr auto PERMS = Common::MemoryPermission::ReadWrite;
 
 TEST_CASE("HostMemory: Initialize and deinitialize", "[common]") {
     { HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE); }
@@ -19,7 +20,7 @@ TEST_CASE("HostMemory: Initialize and deinitialize", "[common]") {
 
 TEST_CASE("HostMemory: Simple map", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x5000, 0x8000, 0x1000);
+    mem.Map(0x5000, 0x8000, 0x1000, PERMS);
 
     volatile u8* const data = mem.VirtualBasePointer() + 0x5000;
     data[0] = 50;
@@ -28,8 +29,8 @@ TEST_CASE("HostMemory: Simple map", "[common]") {
 
 TEST_CASE("HostMemory: Simple mirror map", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x5000, 0x3000, 0x2000);
-    mem.Map(0x8000, 0x4000, 0x1000);
+    mem.Map(0x5000, 0x3000, 0x2000, PERMS);
+    mem.Map(0x8000, 0x4000, 0x1000, PERMS);
 
     volatile u8* const mirror_a = mem.VirtualBasePointer() + 0x5000;
     volatile u8* const mirror_b = mem.VirtualBasePointer() + 0x8000;
@@ -39,7 +40,7 @@ TEST_CASE("HostMemory: Simple mirror map", "[common]") {
 
 TEST_CASE("HostMemory: Simple unmap", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x5000, 0x3000, 0x2000);
+    mem.Map(0x5000, 0x3000, 0x2000, PERMS);
 
     volatile u8* const data = mem.VirtualBasePointer() + 0x5000;
     data[75] = 50;
@@ -50,7 +51,7 @@ TEST_CASE("HostMemory: Simple unmap", "[common]") {
 
 TEST_CASE("HostMemory: Simple unmap and remap", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x5000, 0x3000, 0x2000);
+    mem.Map(0x5000, 0x3000, 0x2000, PERMS);
 
     volatile u8* const data = mem.VirtualBasePointer() + 0x5000;
     data[0] = 50;
@@ -58,79 +59,79 @@ TEST_CASE("HostMemory: Simple unmap and remap", "[common]") {
 
     mem.Unmap(0x5000, 0x2000);
 
-    mem.Map(0x5000, 0x3000, 0x2000);
+    mem.Map(0x5000, 0x3000, 0x2000, PERMS);
     REQUIRE(data[0] == 50);
 
-    mem.Map(0x7000, 0x2000, 0x5000);
+    mem.Map(0x7000, 0x2000, 0x5000, PERMS);
     REQUIRE(data[0x3000] == 50);
 }
 
 TEST_CASE("HostMemory: Nieche allocation", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x0000, 0, 0x20000);
+    mem.Map(0x0000, 0, 0x20000, PERMS);
     mem.Unmap(0x0000, 0x4000);
-    mem.Map(0x1000, 0, 0x2000);
-    mem.Map(0x3000, 0, 0x1000);
-    mem.Map(0, 0, 0x1000);
+    mem.Map(0x1000, 0, 0x2000, PERMS);
+    mem.Map(0x3000, 0, 0x1000, PERMS);
+    mem.Map(0, 0, 0x1000, PERMS);
 }
 
 TEST_CASE("HostMemory: Full unmap", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x8000, 0, 0x4000);
+    mem.Map(0x8000, 0, 0x4000, PERMS);
     mem.Unmap(0x8000, 0x4000);
-    mem.Map(0x6000, 0, 0x16000);
+    mem.Map(0x6000, 0, 0x16000, PERMS);
 }
 
 TEST_CASE("HostMemory: Right out of bounds unmap", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x0000, 0, 0x4000);
+    mem.Map(0x0000, 0, 0x4000, PERMS);
     mem.Unmap(0x2000, 0x4000);
-    mem.Map(0x2000, 0x80000, 0x4000);
+    mem.Map(0x2000, 0x80000, 0x4000, PERMS);
 }
 
 TEST_CASE("HostMemory: Left out of bounds unmap", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x8000, 0, 0x4000);
+    mem.Map(0x8000, 0, 0x4000, PERMS);
     mem.Unmap(0x6000, 0x4000);
-    mem.Map(0x8000, 0, 0x2000);
+    mem.Map(0x8000, 0, 0x2000, PERMS);
 }
 
 TEST_CASE("HostMemory: Multiple placeholder unmap", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x0000, 0, 0x4000);
-    mem.Map(0x4000, 0, 0x1b000);
+    mem.Map(0x0000, 0, 0x4000, PERMS);
+    mem.Map(0x4000, 0, 0x1b000, PERMS);
     mem.Unmap(0x3000, 0x1c000);
-    mem.Map(0x3000, 0, 0x20000);
+    mem.Map(0x3000, 0, 0x20000, PERMS);
 }
 
 TEST_CASE("HostMemory: Unmap between placeholders", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x0000, 0, 0x4000);
-    mem.Map(0x4000, 0, 0x4000);
+    mem.Map(0x0000, 0, 0x4000, PERMS);
+    mem.Map(0x4000, 0, 0x4000, PERMS);
     mem.Unmap(0x2000, 0x4000);
-    mem.Map(0x2000, 0, 0x4000);
+    mem.Map(0x2000, 0, 0x4000, PERMS);
 }
 
 TEST_CASE("HostMemory: Unmap to origin", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x4000, 0, 0x4000);
-    mem.Map(0x8000, 0, 0x4000);
+    mem.Map(0x4000, 0, 0x4000, PERMS);
+    mem.Map(0x8000, 0, 0x4000, PERMS);
     mem.Unmap(0x4000, 0x4000);
-    mem.Map(0, 0, 0x4000);
-    mem.Map(0x4000, 0, 0x4000);
+    mem.Map(0, 0, 0x4000, PERMS);
+    mem.Map(0x4000, 0, 0x4000, PERMS);
 }
 
 TEST_CASE("HostMemory: Unmap to right", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x4000, 0, 0x4000);
-    mem.Map(0x8000, 0, 0x4000);
+    mem.Map(0x4000, 0, 0x4000, PERMS);
+    mem.Map(0x8000, 0, 0x4000, PERMS);
     mem.Unmap(0x8000, 0x4000);
-    mem.Map(0x8000, 0, 0x4000);
+    mem.Map(0x8000, 0, 0x4000, PERMS);
 }
 
 TEST_CASE("HostMemory: Partial right unmap check bindings", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x4000, 0x10000, 0x4000);
+    mem.Map(0x4000, 0x10000, 0x4000, PERMS);
 
     volatile u8* const ptr = mem.VirtualBasePointer() + 0x4000;
     ptr[0x1000] = 17;
@@ -142,7 +143,7 @@ TEST_CASE("HostMemory: Partial right unmap check bindings", "[common]") {
 
 TEST_CASE("HostMemory: Partial left unmap check bindings", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x4000, 0x10000, 0x4000);
+    mem.Map(0x4000, 0x10000, 0x4000, PERMS);
 
     volatile u8* const ptr = mem.VirtualBasePointer() + 0x4000;
     ptr[0x3000] = 19;
@@ -156,7 +157,7 @@ TEST_CASE("HostMemory: Partial left unmap check bindings", "[common]") {
 
 TEST_CASE("HostMemory: Partial middle unmap check bindings", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x4000, 0x10000, 0x4000);
+    mem.Map(0x4000, 0x10000, 0x4000, PERMS);
 
     volatile u8* const ptr = mem.VirtualBasePointer() + 0x4000;
     ptr[0x0000] = 19;
@@ -170,8 +171,8 @@ TEST_CASE("HostMemory: Partial middle unmap check bindings", "[common]") {
 
 TEST_CASE("HostMemory: Partial sparse middle unmap and check bindings", "[common]") {
     HostMemory mem(BACKING_SIZE, VIRTUAL_SIZE);
-    mem.Map(0x4000, 0x10000, 0x2000);
-    mem.Map(0x6000, 0x20000, 0x2000);
+    mem.Map(0x4000, 0x10000, 0x2000, PERMS);
+    mem.Map(0x6000, 0x20000, 0x2000, PERMS);
 
     volatile u8* const ptr = mem.VirtualBasePointer() + 0x4000;
     ptr[0x0000] = 19;

src/video_core/CMakeLists.txt

@@ -15,6 +15,7 @@ add_library(video_core STATIC
     buffer_cache/buffer_cache.cpp
     buffer_cache/buffer_cache.h
     buffer_cache/memory_tracker_base.h
+    buffer_cache/usage_tracker.h
     buffer_cache/word_manager.h
     cache_types.h
     cdma_pusher.cpp

src/video_core/buffer_cache/buffer_cache.h

@@ -67,6 +67,7 @@ void BufferCache<P>::TickFrame() {
     if (!channel_state) {
         return;
     }
+    runtime.TickFrame(slot_buffers);
 
     // Calculate hits and shots and move hit bits to the right
     const u32 hits = std::reduce(channel_state->uniform_cache_hits.begin(),
@@ -230,7 +231,10 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
     for (const IntervalType& add_interval : tmp_intervals) {
         common_ranges.add(add_interval);
     }
-    runtime.CopyBuffer(dest_buffer, src_buffer, copies);
+    const auto& copy = copies[0];
+    src_buffer.MarkUsage(copy.src_offset, copy.size);
+    dest_buffer.MarkUsage(copy.dst_offset, copy.size);
+    runtime.CopyBuffer(dest_buffer, src_buffer, copies, true);
     if (has_new_downloads) {
         memory_tracker.MarkRegionAsGpuModified(*cpu_dest_address, amount);
     }
@@ -258,9 +262,10 @@ bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) {
     common_ranges.subtract(subtract_interval);
 
     const BufferId buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size));
-    auto& dest_buffer = slot_buffers[buffer];
+    Buffer& dest_buffer = slot_buffers[buffer];
     const u32 offset = dest_buffer.Offset(*cpu_dst_address);
     runtime.ClearBuffer(dest_buffer, offset, size, value);
+    dest_buffer.MarkUsage(offset, size);
     return true;
 }
 
@@ -603,6 +608,7 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
             VAddr orig_cpu_addr = static_cast<VAddr>(second_copy.src_offset);
             const IntervalType base_interval{orig_cpu_addr, orig_cpu_addr + copy.size};
             async_downloads += std::make_pair(base_interval, 1);
+            buffer.MarkUsage(copy.src_offset, copy.size);
             runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
             normalized_copies.push_back(second_copy);
         }
@@ -621,8 +627,9 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
                     // Have in mind the staging buffer offset for the copy
                     copy.dst_offset += download_staging.offset;
                     const std::array copies{copy};
-                    runtime.CopyBuffer(download_staging.buffer, slot_buffers[buffer_id], copies,
-                                       false);
+                    Buffer& buffer = slot_buffers[buffer_id];
+                    buffer.MarkUsage(copy.src_offset, copy.size);
+                    runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
                 }
                 runtime.PostCopyBarrier();
                 runtime.Finish();
@@ -742,7 +749,7 @@ void BufferCache<P>::BindHostIndexBuffer() {
                 {BufferCopy{.src_offset = upload_staging.offset, .dst_offset = 0, .size = size}}};
             std::memcpy(upload_staging.mapped_span.data(),
                         draw_state.inline_index_draw_indexes.data(), size);
-            runtime.CopyBuffer(buffer, upload_staging.buffer, copies);
+            runtime.CopyBuffer(buffer, upload_staging.buffer, copies, true);
         } else {
             buffer.ImmediateUpload(0, draw_state.inline_index_draw_indexes);
         }
@@ -754,6 +761,7 @@ void BufferCache<P>::BindHostIndexBuffer() {
             offset + draw_state.index_buffer.first * draw_state.index_buffer.FormatSizeInBytes();
         runtime.BindIndexBuffer(buffer, new_offset, size);
     } else {
+        buffer.MarkUsage(offset, size);
         runtime.BindIndexBuffer(draw_state.topology, draw_state.index_buffer.format,
                                 draw_state.index_buffer.first, draw_state.index_buffer.count,
                                 buffer, offset, size);
@@ -790,6 +798,7 @@ void BufferCache<P>::BindHostVertexBuffers() {
 
             const u32 stride = maxwell3d->regs.vertex_streams[index].stride;
             const u32 offset = buffer.Offset(binding.cpu_addr);
+            buffer.MarkUsage(offset, binding.size);
 
             host_bindings.buffers.push_back(&buffer);
             host_bindings.offsets.push_back(offset);
@@ -895,6 +904,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
     if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
         channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size;
     }
+    buffer.MarkUsage(offset, size);
     if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
         runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size);
     } else {
@@ -913,6 +923,7 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
+        buffer.MarkUsage(offset, size);
         const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
 
         if (is_written) {
@@ -943,6 +954,7 @@ void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) {
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
         const PixelFormat format = binding.format;
+        buffer.MarkUsage(offset, size);
         if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
             if (((channel_state->image_texture_buffers[stage] >> index) & 1) != 0) {
                 runtime.BindImageBuffer(buffer, offset, size, format);
@@ -975,9 +987,10 @@ void BufferCache<P>::BindHostTransformFeedbackBuffers() {
         MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, size);
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
+        buffer.MarkUsage(offset, size);
         host_bindings.buffers.push_back(&buffer);
         host_bindings.offsets.push_back(offset);
-        host_bindings.sizes.push_back(binding.size);
+        host_bindings.sizes.push_back(size);
     }
     if (host_bindings.buffers.size() > 0) {
         runtime.BindTransformFeedbackBuffers(host_bindings);
@@ -1001,6 +1014,7 @@ void BufferCache<P>::BindHostComputeUniformBuffers() {
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
+        buffer.MarkUsage(offset, size);
         if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
             runtime.BindComputeUniformBuffer(binding_index, buffer, offset, size);
             ++binding_index;
@@ -1021,6 +1035,7 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
         SynchronizeBuffer(buffer, binding.cpu_addr, size);
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
+        buffer.MarkUsage(offset, size);
         const bool is_written =
             ((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
 
@@ -1053,6 +1068,7 @@ void BufferCache<P>::BindHostComputeTextureBuffers() {
 
         const u32 offset = buffer.Offset(binding.cpu_addr);
         const PixelFormat format = binding.format;
+        buffer.MarkUsage(offset, size);
         if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
             if (((channel_state->image_compute_texture_buffers >> index) & 1) != 0) {
                 runtime.BindImageBuffer(buffer, offset, size, format);
@@ -1172,10 +1188,11 @@ void BufferCache<P>::UpdateVertexBuffer(u32 index) {
     if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end)) {
         size = static_cast<u32>(gpu_memory->MaxContinuousRange(gpu_addr_begin, size));
     }
+    const BufferId buffer_id = FindBuffer(*cpu_addr, size);
     channel_state->vertex_buffers[index] = Binding{
         .cpu_addr = *cpu_addr,
         .size = size,
-        .buffer_id = FindBuffer(*cpu_addr, size),
+        .buffer_id = buffer_id,
     };
 }
 
@@ -1401,7 +1418,8 @@ void BufferCache<P>::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id,
         .dst_offset = dst_base_offset,
         .size = overlap.SizeBytes(),
     });
-    runtime.CopyBuffer(new_buffer, overlap, copies);
+    new_buffer.MarkUsage(copies[0].dst_offset, copies[0].size);
+    runtime.CopyBuffer(new_buffer, overlap, copies, true);
     DeleteBuffer(overlap_id, true);
 }
 
@@ -1414,7 +1432,9 @@ BufferId BufferCache<P>::CreateBuffer(VAddr cpu_addr, u32 wanted_size) {
     const u32 size = static_cast<u32>(overlap.end - overlap.begin);
     const BufferId new_buffer_id = slot_buffers.insert(runtime, rasterizer, overlap.begin, size);
     auto& new_buffer = slot_buffers[new_buffer_id];
-    runtime.ClearBuffer(new_buffer, 0, new_buffer.SizeBytes(), 0);
+    const size_t size_bytes = new_buffer.SizeBytes();
+    runtime.ClearBuffer(new_buffer, 0, size_bytes, 0);
+    new_buffer.MarkUsage(0, size_bytes);
     for (const BufferId overlap_id : overlap.ids) {
         JoinOverlap(new_buffer_id, overlap_id, !overlap.has_stream_leap);
     }
@@ -1467,11 +1487,6 @@ void BufferCache<P>::TouchBuffer(Buffer& buffer, BufferId buffer_id) noexcept {
 
 template <class P>
 bool BufferCache<P>::SynchronizeBuffer(Buffer& buffer, VAddr cpu_addr, u32 size) {
-    return SynchronizeBufferImpl(buffer, cpu_addr, size);
-}
-
-template <class P>
-bool BufferCache<P>::SynchronizeBufferImpl(Buffer& buffer, VAddr cpu_addr, u32 size) {
     boost::container::small_vector<BufferCopy, 4> copies;
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
@@ -1493,51 +1508,6 @@ bool BufferCache<P>::SynchronizeBufferImpl(Buffer& buffer, VAddr cpu_addr, u32 s
     return false;
 }
 
-template <class P>
-bool BufferCache<P>::SynchronizeBufferNoModified(Buffer& buffer, VAddr cpu_addr, u32 size) {
-    boost::container::small_vector<BufferCopy, 4> copies;
-    u64 total_size_bytes = 0;
-    u64 largest_copy = 0;
-    IntervalSet found_sets{};
-    auto make_copies = [&] {
-        for (auto& interval : found_sets) {
-            const std::size_t sub_size = interval.upper() - interval.lower();
-            const VAddr cpu_addr_ = interval.lower();
-            copies.push_back(BufferCopy{
-                .src_offset = total_size_bytes,
-                .dst_offset = cpu_addr_ - buffer.CpuAddr(),
-                .size = sub_size,
-            });
-            total_size_bytes += sub_size;
-            largest_copy = std::max<u64>(largest_copy, sub_size);
-        }
-        const std::span<BufferCopy> copies_span(copies.data(), copies.size());
-        UploadMemory(buffer, total_size_bytes, largest_copy, copies_span);
-    };
-    memory_tracker.ForEachUploadRange(cpu_addr, size, [&](u64 cpu_addr_out, u64 range_size) {
-        const VAddr base_adr = cpu_addr_out;
-        const VAddr end_adr = base_adr + range_size;
-        const IntervalType add_interval{base_adr, end_adr};
-        found_sets.add(add_interval);
-    });
-    if (found_sets.empty()) {
-        return true;
-    }
-    const IntervalType search_interval{cpu_addr, cpu_addr + size};
-    auto it = common_ranges.lower_bound(search_interval);
-    auto it_end = common_ranges.upper_bound(search_interval);
-    if (it == common_ranges.end()) {
-        make_copies();
-        return false;
-    }
-    while (it != it_end) {
-        found_sets.subtract(*it);
-        it++;
-    }
-    make_copies();
-    return false;
-}
-
 template <class P>
 void BufferCache<P>::UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy,
                                   std::span<BufferCopy> copies) {
@@ -1586,7 +1556,8 @@ void BufferCache<P>::MappedUploadMemory([[maybe_unused]] Buffer& buffer,
             // Apply the staging offset
             copy.src_offset += upload_staging.offset;
         }
-        runtime.CopyBuffer(buffer, upload_staging.buffer, copies);
+        const bool can_reorder = runtime.CanReorderUpload(buffer, copies);
+        runtime.CopyBuffer(buffer, upload_staging.buffer, copies, true, can_reorder);
     }
 }
 
@@ -1628,7 +1599,8 @@ void BufferCache<P>::InlineMemoryImplementation(VAddr dest_address, size_t copy_
         }};
         u8* const src_pointer = upload_staging.mapped_span.data();
         std::memcpy(src_pointer, inlined_buffer.data(), copy_size);
-        runtime.CopyBuffer(buffer, upload_staging.buffer, copies);
+        const bool can_reorder = runtime.CanReorderUpload(buffer, copies);
+        runtime.CopyBuffer(buffer, upload_staging.buffer, copies, true, can_reorder);
     } else {
         buffer.ImmediateUpload(buffer.Offset(dest_address), inlined_buffer.first(copy_size));
     }
@@ -1681,8 +1653,9 @@ void BufferCache<P>::DownloadBufferMemory(Buffer& buffer, VAddr cpu_addr, u64 si
         for (BufferCopy& copy : copies) {
             // Modify copies to have the staging offset in mind
             copy.dst_offset += download_staging.offset;
+            buffer.MarkUsage(copy.src_offset, copy.size);
         }
-        runtime.CopyBuffer(download_staging.buffer, buffer, copies_span);
+        runtime.CopyBuffer(download_staging.buffer, buffer, copies_span, true);
         runtime.Finish();
         for (const BufferCopy& copy : copies) {
             const VAddr copy_cpu_addr = buffer.CpuAddr() + copy.src_offset;

src/video_core/buffer_cache/buffer_cache_base.h

@@ -529,10 +529,6 @@ private:
 
     bool SynchronizeBuffer(Buffer& buffer, VAddr cpu_addr, u32 size);
 
-    bool SynchronizeBufferImpl(Buffer& buffer, VAddr cpu_addr, u32 size);
-
-    bool SynchronizeBufferNoModified(Buffer& buffer, VAddr cpu_addr, u32 size);
-
     void UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy,
                       std::span<BufferCopy> copies);
 

src/video_core/buffer_cache/usage_tracker.h

@@ -0,0 +1,79 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include "common/alignment.h"
+#include "common/common_types.h"
+
+namespace VideoCommon {
+
+class UsageTracker {
+    static constexpr size_t BYTES_PER_BIT_SHIFT = 6;
+    static constexpr size_t PAGE_SHIFT = 6 + BYTES_PER_BIT_SHIFT;
+    static constexpr size_t PAGE_BYTES = 1 << PAGE_SHIFT;
+
+public:
+    explicit UsageTracker(size_t size) {
+        const size_t num_pages = (size >> PAGE_SHIFT) + 1;
+        pages.resize(num_pages, 0ULL);
+    }
+
+    void Reset() noexcept {
+        std::ranges::fill(pages, 0ULL);
+    }
+
+    void Track(u64 offset, u64 size) noexcept {
+        const size_t page = offset >> PAGE_SHIFT;
+        const size_t page_end = (offset + size) >> PAGE_SHIFT;
+        TrackPage(page, offset, size);
+        if (page == page_end) {
+            return;
+        }
+        for (size_t i = page + 1; i < page_end; i++) {
+            pages[i] = ~u64{0};
+        }
+        const size_t offset_end = offset + size;
+        const size_t offset_end_page_aligned = Common::AlignDown(offset_end, PAGE_BYTES);
+        TrackPage(page_end, offset_end_page_aligned, offset_end - offset_end_page_aligned);
+    }
+
+    [[nodiscard]] bool IsUsed(u64 offset, u64 size) const noexcept {
+        const size_t page = offset >> PAGE_SHIFT;
+        const size_t page_end = (offset + size) >> PAGE_SHIFT;
+        if (IsPageUsed(page, offset, size)) {
+            return true;
+        }
+        for (size_t i = page + 1; i < page_end; i++) {
+            if (pages[i] != 0) {
+                return true;
+            }
+        }
+        const size_t offset_end = offset + size;
+        const size_t offset_end_page_aligned = Common::AlignDown(offset_end, PAGE_BYTES);
+        return IsPageUsed(page_end, offset_end_page_aligned, offset_end - offset_end_page_aligned);
+    }
+
+private:
+    void TrackPage(u64 page, u64 offset, u64 size) noexcept {
+        const size_t offset_in_page = offset % PAGE_BYTES;
+        const size_t first_bit = offset_in_page >> BYTES_PER_BIT_SHIFT;
+        const size_t num_bits = std::min(size, PAGE_BYTES) >> BYTES_PER_BIT_SHIFT;
+        const size_t mask = ~u64{0} >> (64 - num_bits);
+        pages[page] |= (~u64{0} & mask) << first_bit;
+    }
+
+    bool IsPageUsed(u64 page, u64 offset, u64 size) const noexcept {
+        const size_t offset_in_page = offset % PAGE_BYTES;
+        const size_t first_bit = offset_in_page >> BYTES_PER_BIT_SHIFT;
+        const size_t num_bits = std::min(size, PAGE_BYTES) >> BYTES_PER_BIT_SHIFT;
+        const size_t mask = ~u64{0} >> (64 - num_bits);
+        const size_t mask2 = (~u64{0} & mask) << first_bit;
+        return (pages[page] & mask2) != 0;
+    }
+
+private:
+    std::vector<u64> pages;
+};
+
+} // namespace VideoCommon

src/video_core/renderer_opengl/gl_buffer_cache.cpp

@@ -178,13 +178,14 @@ void BufferCacheRuntime::CopyBuffer(GLuint dst_buffer, Buffer& src_buffer,
 }
 
 void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, GLuint src_buffer,
-                                    std::span<const VideoCommon::BufferCopy> copies, bool barrier) {
+                                    std::span<const VideoCommon::BufferCopy> copies, bool barrier,
+                                    bool) {
     CopyBuffer(dst_buffer.Handle(), src_buffer, copies, barrier);
 }
 
 void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer,
-                                    std::span<const VideoCommon::BufferCopy> copies) {
-    CopyBuffer(dst_buffer.Handle(), src_buffer.Handle(), copies);
+                                    std::span<const VideoCommon::BufferCopy> copies, bool) {
+    CopyBuffer(dst_buffer.Handle(), src_buffer.Handle(), copies, true);
 }
 
 void BufferCacheRuntime::PreCopyBarrier() {

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -30,6 +30,8 @@ public:
 
     void MakeResident(GLenum access) noexcept;
 
+    void MarkUsage(u64 offset, u64 size) {}
+
     [[nodiscard]] GLuint View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format);
 
     [[nodiscard]] GLuint64EXT HostGpuAddr() const noexcept {
@@ -66,22 +68,29 @@ public:
 
     [[nodiscard]] StagingBufferMap DownloadStagingBuffer(size_t size);
 
+    bool CanReorderUpload(const Buffer&, std::span<const VideoCommon::BufferCopy>) {
+        return false;
+    }
+
     void CopyBuffer(GLuint dst_buffer, GLuint src_buffer,
-                    std::span<const VideoCommon::BufferCopy> copies, bool barrier = true);
+                    std::span<const VideoCommon::BufferCopy> copies, bool barrier);
 
     void CopyBuffer(GLuint dst_buffer, Buffer& src_buffer,
-                    std::span<const VideoCommon::BufferCopy> copies, bool barrier = true);
+                    std::span<const VideoCommon::BufferCopy> copies, bool barrier);
 
     void CopyBuffer(Buffer& dst_buffer, GLuint src_buffer,
-                    std::span<const VideoCommon::BufferCopy> copies, bool barrier = true);
+                    std::span<const VideoCommon::BufferCopy> copies, bool barrier,
+                    bool can_reorder_upload = false);
 
     void CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer,
-                    std::span<const VideoCommon::BufferCopy> copies);
+                    std::span<const VideoCommon::BufferCopy> copies, bool);
 
     void PreCopyBarrier();
     void PostCopyBarrier();
     void Finish();
 
+    void TickFrame(VideoCommon::SlotVector<Buffer>&) noexcept {}
+
     void ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value);
 
     void BindIndexBuffer(Buffer& buffer, u32 offset, u32 size);

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -79,13 +79,13 @@ vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allo
 } // Anonymous namespace
 
 Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
-    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {}
+    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params), tracker{4096} {}
 
 Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
                VAddr cpu_addr_, u64 size_bytes_)
     : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_),
-      device{&runtime.device}, buffer{
-                                   CreateBuffer(*device, runtime.memory_allocator, SizeBytes())} {
+      device{&runtime.device}, buffer{CreateBuffer(*device, runtime.memory_allocator, SizeBytes())},
+      tracker{SizeBytes()} {
     if (runtime.device.HasDebuggingToolAttached()) {
         buffer.SetObjectNameEXT(fmt::format("Buffer 0x{:x}", CpuAddr()).c_str());
     }
@@ -355,12 +355,31 @@ bool BufferCacheRuntime::CanReportMemoryUsage() const {
     return device.CanReportMemoryUsage();
 }
 
+void BufferCacheRuntime::TickFrame(VideoCommon::SlotVector<Buffer>& slot_buffers) noexcept {
+    for (auto it = slot_buffers.begin(); it != slot_buffers.end(); it++) {
+        it->ResetUsageTracking();
+    }
+}
+
 void BufferCacheRuntime::Finish() {
     scheduler.Finish();
 }
 
+bool BufferCacheRuntime::CanReorderUpload(const Buffer& buffer,
+                                          std::span<const VideoCommon::BufferCopy> copies) {
+    if (Settings::values.disable_buffer_reorder) {
+        return false;
+    }
+    const bool can_use_upload_cmdbuf =
+        std::ranges::all_of(copies, [&](const VideoCommon::BufferCopy& copy) {
+            return !buffer.IsRegionUsed(copy.dst_offset, copy.size);
+        });
+    return can_use_upload_cmdbuf;
+}
+
 void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
-                                    std::span<const VideoCommon::BufferCopy> copies, bool barrier) {
+                                    std::span<const VideoCommon::BufferCopy> copies, bool barrier,
+                                    bool can_reorder_upload) {
     if (dst_buffer == VK_NULL_HANDLE || src_buffer == VK_NULL_HANDLE) {
         return;
     }
@@ -376,9 +395,18 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
         .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
         .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
     };
+
     // Measuring a popular game, this number never exceeds the specified size once data is warmed up
     boost::container::small_vector<VkBufferCopy, 8> vk_copies(copies.size());
     std::ranges::transform(copies, vk_copies.begin(), MakeBufferCopy);
+    if (src_buffer == staging_pool.StreamBuf() && can_reorder_upload) {
+        scheduler.RecordWithUploadBuffer([src_buffer, dst_buffer, vk_copies](
+                                             vk::CommandBuffer, vk::CommandBuffer upload_cmdbuf) {
+            upload_cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies);
+        });
+        return;
+    }
+
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([src_buffer, dst_buffer, vk_copies, barrier](vk::CommandBuffer cmdbuf) {
         if (barrier) {

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -5,6 +5,7 @@
 
 #include "video_core/buffer_cache/buffer_cache_base.h"
 #include "video_core/buffer_cache/memory_tracker_base.h"
+#include "video_core/buffer_cache/usage_tracker.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
@@ -34,6 +35,18 @@ public:
         return *buffer;
     }
 
+    [[nodiscard]] bool IsRegionUsed(u64 offset, u64 size) const noexcept {
+        return tracker.IsUsed(offset, size);
+    }
+
+    void MarkUsage(u64 offset, u64 size) noexcept {
+        tracker.Track(offset, size);
+    }
+
+    void ResetUsageTracking() noexcept {
+        tracker.Reset();
+    }
+
     operator VkBuffer() const noexcept {
         return *buffer;
     }
@@ -49,6 +62,7 @@ private:
     const Device* device{};
     vk::Buffer buffer;
     std::vector<BufferView> views;
+    VideoCommon::UsageTracker tracker;
 };
 
 class QuadArrayIndexBuffer;
@@ -67,6 +81,8 @@ public:
                                 ComputePassDescriptorQueue& compute_pass_descriptor_queue,
                                 DescriptorPool& descriptor_pool);
 
+    void TickFrame(VideoCommon::SlotVector<Buffer>& slot_buffers) noexcept;
+
     void Finish();
 
     u64 GetDeviceLocalMemory() const;
@@ -79,12 +95,15 @@ public:
 
     [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size, bool deferred = false);
 
+    bool CanReorderUpload(const Buffer& buffer, std::span<const VideoCommon::BufferCopy> copies);
+
     void FreeDeferredStagingBuffer(StagingBufferRef& ref);
 
     void PreCopyBarrier();
 
     void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer,
-                    std::span<const VideoCommon::BufferCopy> copies, bool barrier = true);
+                    std::span<const VideoCommon::BufferCopy> copies, bool barrier,
+                    bool can_reorder_upload = false);
 
     void PostCopyBarrier();
 

src/video_core/renderer_vulkan/vk_master_semaphore.cpp

@@ -100,12 +100,14 @@ void MasterSemaphore::Wait(u64 tick) {
     Refresh();
 }
 
-VkResult MasterSemaphore::SubmitQueue(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                                      VkSemaphore wait_semaphore, u64 host_tick) {
+VkResult MasterSemaphore::SubmitQueue(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf,
+                                      VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
+                                      u64 host_tick) {
     if (semaphore) {
-        return SubmitQueueTimeline(cmdbuf, signal_semaphore, wait_semaphore, host_tick);
+        return SubmitQueueTimeline(cmdbuf, upload_cmdbuf, signal_semaphore, wait_semaphore,
+                                   host_tick);
     } else {
-        return SubmitQueueFence(cmdbuf, signal_semaphore, wait_semaphore, host_tick);
+        return SubmitQueueFence(cmdbuf, upload_cmdbuf, signal_semaphore, wait_semaphore, host_tick);
     }
 }
 
@@ -115,6 +117,7 @@ static constexpr std::array<VkPipelineStageFlags, 2> wait_stage_masks{
 };
 
 VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
+                                              vk::CommandBuffer& upload_cmdbuf,
                                               VkSemaphore signal_semaphore,
                                               VkSemaphore wait_semaphore, u64 host_tick) {
     const VkSemaphore timeline_semaphore = *semaphore;
@@ -123,6 +126,8 @@ VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
     const std::array signal_values{host_tick, u64(0)};
     const std::array signal_semaphores{timeline_semaphore, signal_semaphore};
 
+    const std::array cmdbuffers{*upload_cmdbuf, *cmdbuf};
+
     const u32 num_wait_semaphores = wait_semaphore ? 1 : 0;
     const VkTimelineSemaphoreSubmitInfo timeline_si{
         .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO,
@@ -138,8 +143,8 @@ VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
         .waitSemaphoreCount = num_wait_semaphores,
         .pWaitSemaphores = &wait_semaphore,
         .pWaitDstStageMask = wait_stage_masks.data(),
-        .commandBufferCount = 1,
-        .pCommandBuffers = cmdbuf.address(),
+        .commandBufferCount = static_cast<u32>(cmdbuffers.size()),
+        .pCommandBuffers = cmdbuffers.data(),
         .signalSemaphoreCount = num_signal_semaphores,
         .pSignalSemaphores = signal_semaphores.data(),
     };
@@ -147,19 +152,23 @@ VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
     return device.GetGraphicsQueue().Submit(submit_info);
 }
 
-VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                                           VkSemaphore wait_semaphore, u64 host_tick) {
+VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf,
+                                           vk::CommandBuffer& upload_cmdbuf,
+                                           VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
+                                           u64 host_tick) {
     const u32 num_signal_semaphores = signal_semaphore ? 1 : 0;
     const u32 num_wait_semaphores = wait_semaphore ? 1 : 0;
 
+    const std::array cmdbuffers{*upload_cmdbuf, *cmdbuf};
+
     const VkSubmitInfo submit_info{
         .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
         .pNext = nullptr,
         .waitSemaphoreCount = num_wait_semaphores,
         .pWaitSemaphores = &wait_semaphore,
         .pWaitDstStageMask = wait_stage_masks.data(),
-        .commandBufferCount = 1,
-        .pCommandBuffers = cmdbuf.address(),
+        .commandBufferCount = static_cast<u32>(cmdbuffers.size()),
+        .pCommandBuffers = cmdbuffers.data(),
         .signalSemaphoreCount = num_signal_semaphores,
         .pSignalSemaphores = &signal_semaphore,
     };

src/video_core/renderer_vulkan/vk_master_semaphore.h

@@ -52,14 +52,16 @@ public:
     void Wait(u64 tick);
 
     /// Submits the device graphics queue, updating the tick as necessary
-    VkResult SubmitQueue(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                         VkSemaphore wait_semaphore, u64 host_tick);
+    VkResult SubmitQueue(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf,
+                         VkSemaphore signal_semaphore, VkSemaphore wait_semaphore, u64 host_tick);
 
 private:
-    VkResult SubmitQueueTimeline(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                                 VkSemaphore wait_semaphore, u64 host_tick);
-    VkResult SubmitQueueFence(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                              VkSemaphore wait_semaphore, u64 host_tick);
+    VkResult SubmitQueueTimeline(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf,
+                                 VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
+                                 u64 host_tick);
+    VkResult SubmitQueueFence(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf,
+                              VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
+                              u64 host_tick);
 
     void WaitThread(std::stop_token token);
 

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -22,11 +22,12 @@ namespace Vulkan {
 
 MICROPROFILE_DECLARE(Vulkan_WaitForWorker);
 
-void Scheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) {
+void Scheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf,
+                                         vk::CommandBuffer upload_cmdbuf) {
     auto command = first;
     while (command != nullptr) {
         auto next = command->GetNext();
-        command->Execute(cmdbuf);
+        command->Execute(cmdbuf, upload_cmdbuf);
         command->~Command();
         command = next;
     }
@@ -180,7 +181,7 @@ void Scheduler::WorkerThread(std::stop_token stop_token) {
             // Perform the work, tracking whether the chunk was a submission
             // before executing.
             const bool has_submit = work->HasSubmit();
-            work->ExecuteAll(current_cmdbuf);
+            work->ExecuteAll(current_cmdbuf, current_upload_cmdbuf);
 
             // If the chunk was a submission, reallocate the command buffer.
             if (has_submit) {
@@ -205,6 +206,13 @@ void Scheduler::AllocateWorkerCommandBuffer() {
         .flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
         .pInheritanceInfo = nullptr,
     });
+    current_upload_cmdbuf = vk::CommandBuffer(command_pool->Commit(), device.GetDispatchLoader());
+    current_upload_cmdbuf.Begin({
+        .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
+        .pNext = nullptr,
+        .flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
+        .pInheritanceInfo = nullptr,
+    });
 }
 
 u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) {
@@ -212,7 +220,17 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se
     InvalidateState();
 
     const u64 signal_value = master_semaphore->NextTick();
-    Record([signal_semaphore, wait_semaphore, signal_value, this](vk::CommandBuffer cmdbuf) {
+    RecordWithUploadBuffer([signal_semaphore, wait_semaphore, signal_value,
+                            this](vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) {
+        static constexpr VkMemoryBarrier WRITE_BARRIER{
+            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+        };
+        upload_cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
+                                      VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER);
+        upload_cmdbuf.End();
         cmdbuf.End();
 
         if (on_submit) {
@@ -221,7 +239,7 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se
 
         std::scoped_lock lock{submit_mutex};
         switch (const VkResult result = master_semaphore->SubmitQueue(
-                    cmdbuf, signal_semaphore, wait_semaphore, signal_value)) {
+                    cmdbuf, upload_cmdbuf, signal_semaphore, wait_semaphore, signal_value)) {
         case VK_SUCCESS:
             break;
         case VK_ERROR_DEVICE_LOST:

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -80,7 +80,8 @@ public:
 
     /// Send work to a separate thread.
     template <typename T>
-    void Record(T&& command) {
+        requires std::is_invocable_v<T, vk::CommandBuffer, vk::CommandBuffer>
+    void RecordWithUploadBuffer(T&& command) {
         if (chunk->Record(command)) {
             return;
         }
@@ -88,6 +89,15 @@ public:
         (void)chunk->Record(command);
     }
 
+    template <typename T>
+        requires std::is_invocable_v<T, vk::CommandBuffer>
+    void Record(T&& c) {
+        this->RecordWithUploadBuffer(
+            [command = std::move(c)](vk::CommandBuffer cmdbuf, vk::CommandBuffer) {
+                command(cmdbuf);
+            });
+    }
+
     /// Returns the current command buffer tick.
     [[nodiscard]] u64 CurrentTick() const noexcept {
         return master_semaphore->CurrentTick();
@@ -119,7 +129,7 @@ private:
     public:
         virtual ~Command() = default;
 
-        virtual void Execute(vk::CommandBuffer cmdbuf) const = 0;
+        virtual void Execute(vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) const = 0;
 
         Command* GetNext() const {
             return next;
@@ -142,8 +152,8 @@ private:
         TypedCommand(TypedCommand&&) = delete;
         TypedCommand& operator=(TypedCommand&&) = delete;
 
-        void Execute(vk::CommandBuffer cmdbuf) const override {
-            command(cmdbuf);
+        void Execute(vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) const override {
+            command(cmdbuf, upload_cmdbuf);
         }
 
     private:
@@ -152,7 +162,7 @@ private:
 
     class CommandChunk final {
     public:
-        void ExecuteAll(vk::CommandBuffer cmdbuf);
+        void ExecuteAll(vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf);
 
         template <typename T>
         bool Record(T& command) {
@@ -228,6 +238,7 @@ private:
     VideoCommon::QueryCacheBase<QueryCacheParams>* query_cache = nullptr;
 
     vk::CommandBuffer current_cmdbuf;
+    vk::CommandBuffer current_upload_cmdbuf;
 
     std::unique_ptr<CommandChunk> chunk;
     std::function<void()> on_submit;

src/video_core/renderer_vulkan/vk_smaa.cpp

@@ -672,7 +672,7 @@ void SMAA::UploadImages(Scheduler& scheduler) {
     UploadImage(m_device, m_allocator, scheduler, m_static_images[Search], search_extent,
                 VK_FORMAT_R8_UNORM, ARRAY_TO_SPAN(searchTexBytes));
 
-    scheduler.Record([&](vk::CommandBuffer& cmdbuf) {
+    scheduler.Record([&](vk::CommandBuffer cmdbuf) {
         for (auto& images : m_dynamic_images) {
             for (size_t i = 0; i < MaxDynamicImage; i++) {
                 ClearColorImage(cmdbuf, *images.images[i]);
@@ -707,7 +707,7 @@ VkImageView SMAA::Draw(Scheduler& scheduler, size_t image_index, VkImage source_
     UpdateDescriptorSets(source_image_view, image_index);
 
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([=, this](vk::CommandBuffer& cmdbuf) {
+    scheduler.Record([=, this](vk::CommandBuffer cmdbuf) {
         TransitionImageLayout(cmdbuf, source_image, VK_IMAGE_LAYOUT_GENERAL);
         TransitionImageLayout(cmdbuf, edges_image, VK_IMAGE_LAYOUT_GENERAL);
         BeginRenderPass(cmdbuf, m_renderpasses[EdgeDetection], edge_detection_framebuffer,

src/video_core/renderer_vulkan/vk_staging_buffer_pool.h

@@ -36,6 +36,10 @@ public:
     StagingBufferRef Request(size_t size, MemoryUsage usage, bool deferred = false);
     void FreeDeferred(StagingBufferRef& ref);
 
+    [[nodiscard]] VkBuffer StreamBuf() const noexcept {
+        return *stream_buffer;
+    }
+
     void TickFrame();
 
 private: