Revert "buffer_cache: reset cached write bits after flushing invalidations"

Support `InfoType_MesosphereCurrentProcess`

.ci/scripts/windows/docker.sh

@@ -1,12 +1,27 @@
 #!/bin/bash -ex
 
+set -e
+
 cd /yuzu
 
 ccache -s
 
 mkdir build || true && cd build
-cmake .. -G Ninja -DDISPLAY_VERSION=$1 -DCMAKE_TOOLCHAIN_FILE="$(pwd)/../CMakeModules/MinGWCross.cmake" -DUSE_CCACHE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DCMAKE_BUILD_TYPE=Release -DENABLE_QT_TRANSLATION=ON
-ninja
+LDFLAGS="-fuse-ld=lld"
+# -femulated-tls required due to an incompatibility between GCC and Clang
+# TODO(lat9nq): If this is widespread, we probably need to add this to CMakeLists where appropriate
+cmake .. \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DCMAKE_CXX_FLAGS="-femulated-tls" \
+    -DCMAKE_TOOLCHAIN_FILE="$(pwd)/../CMakeModules/MinGWClangCross.cmake" \
+    -DDISPLAY_VERSION=$1 \
+    -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON \
+    -DENABLE_QT_TRANSLATION=ON \
+    -DUSE_CCACHE=ON \
+    -DYUZU_USE_BUNDLED_SDL2=OFF \
+    -DYUZU_USE_EXTERNAL_SDL2=OFF \
+    -GNinja
+ninja yuzu yuzu-cmd
 
 ccache -s
 

CMakeModules/MinGWClangCross.cmake

@@ -0,0 +1,55 @@
+set(MINGW_PREFIX                /usr/x86_64-w64-mingw32/)
+set(CMAKE_SYSTEM_NAME           Windows)
+set(CMAKE_SYSTEM_PROCESSOR      x86_64)
+
+set(CMAKE_FIND_ROOT_PATH        ${MINGW_PREFIX})
+set(SDL2_PATH                   ${MINGW_PREFIX})
+set(MINGW_TOOL_PREFIX           ${CMAKE_SYSTEM_PROCESSOR}-w64-mingw32-)
+
+# Specify the cross compiler
+set(CMAKE_C_COMPILER            ${MINGW_TOOL_PREFIX}clang)
+set(CMAKE_CXX_COMPILER          ${MINGW_TOOL_PREFIX}clang++)
+set(CMAKE_RC_COMPILER           ${MINGW_TOOL_PREFIX}windres)
+set(CMAKE_C_COMPILER_AR         ${MINGW_TOOL_PREFIX}ar)
+set(CMAKE_CXX_COMPILER_AR       ${MINGW_TOOL_PREFIX}ar)
+set(CMAKE_C_COMPILER_RANLIB     ${MINGW_TOOL_PREFIX}ranlib)
+set(CMAKE_CXX_COMPILER_RANLIB   ${MINGW_TOOL_PREFIX}ranlib)
+
+# Mingw tools
+set(STRIP                       ${MINGW_TOOL_PREFIX}strip)
+set(WINDRES                     ${MINGW_TOOL_PREFIX}windres)
+set(ENV{PKG_CONFIG}             ${MINGW_TOOL_PREFIX}pkg-config)
+
+# ccache wrapper
+option(USE_CCACHE "Use ccache for compilation" OFF)
+if(USE_CCACHE)
+    find_program(CCACHE ccache)
+    if(CCACHE)
+        message(STATUS "Using ccache found in PATH")
+        set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE ${CCACHE})
+        set_property(GLOBAL PROPERTY RULE_LAUNCH_LINK ${CCACHE})
+    else(CCACHE)
+        message(WARNING "USE_CCACHE enabled, but no ccache found")
+    endif(CCACHE)
+endif(USE_CCACHE)
+
+# Search for programs in the build host directories
+set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+
+
+# Echo modified cmake vars to screen for debugging purposes
+if(NOT DEFINED ENV{MINGW_DEBUG_INFO})
+        message("")
+        message("Custom cmake vars: (blank = system default)")
+        message("-----------------------------------------")
+        message("* CMAKE_C_COMPILER                     : ${CMAKE_C_COMPILER}")
+        message("* CMAKE_CXX_COMPILER                   : ${CMAKE_CXX_COMPILER}")
+        message("* CMAKE_RC_COMPILER                    : ${CMAKE_RC_COMPILER}")
+        message("* WINDRES                              : ${WINDRES}")
+        message("* ENV{PKG_CONFIG}                      : $ENV{PKG_CONFIG}")
+        message("* STRIP                                : ${STRIP}")
+        message("* USE_CCACHE                           : ${USE_CCACHE}")
+        message("")
+        # So that the debug info only appears once
+        set(ENV{MINGW_DEBUG_INFO} SHOWN)
+endif()

src/common/page_table.h

@@ -15,6 +15,9 @@ enum class PageType : u8 {
     Unmapped,
     /// Page is mapped to regular memory. This is the only type you can get pointers to.
     Memory,
+    /// Page is mapped to regular memory, but inaccessible from CPU fastmem and must use
+    /// the callbacks.
+    DebugMemory,
     /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and
     /// invalidation
     RasterizerCachedMemory,

src/common/param_package.cpp

@@ -76,7 +76,7 @@ std::string ParamPackage::Serialize() const {
 std::string ParamPackage::Get(const std::string& key, const std::string& default_value) const {
     auto pair = data.find(key);
     if (pair == data.end()) {
-        LOG_DEBUG(Common, "key '{}' not found", key);
+        LOG_TRACE(Common, "key '{}' not found", key);
         return default_value;
     }
 
@@ -86,7 +86,7 @@ std::string ParamPackage::Get(const std::string& key, const std::string& default
 int ParamPackage::Get(const std::string& key, int default_value) const {
     auto pair = data.find(key);
     if (pair == data.end()) {
-        LOG_DEBUG(Common, "key '{}' not found", key);
+        LOG_TRACE(Common, "key '{}' not found", key);
         return default_value;
     }
 
@@ -101,7 +101,7 @@ int ParamPackage::Get(const std::string& key, int default_value) const {
 float ParamPackage::Get(const std::string& key, float default_value) const {
     auto pair = data.find(key);
     if (pair == data.end()) {
-        LOG_DEBUG(Common, "key {} not found", key);
+        LOG_TRACE(Common, "key {} not found", key);
         return default_value;
     }
 

src/core/arm/arm_interface.cpp

@@ -95,7 +95,7 @@ void ARM_Interface::Run() {
     using Kernel::SuspendType;
 
     while (true) {
-        Kernel::KThread* current_thread{system.Kernel().CurrentScheduler()->GetCurrentThread()};
+        Kernel::KThread* current_thread{Kernel::GetCurrentThreadPointer(system.Kernel())};
         Dynarmic::HaltReason hr{};
 
         // Notify the debugger and go to sleep if a step was performed
@@ -121,8 +121,15 @@ void ARM_Interface::Run() {
 
         // Notify the debugger and go to sleep if a breakpoint was hit.
         if (Has(hr, breakpoint)) {
+            RewindBreakpointInstruction();
             system.GetDebugger().NotifyThreadStopped(current_thread);
-            current_thread->RequestSuspend(Kernel::SuspendType::Debug);
+            current_thread->RequestSuspend(SuspendType::Debug);
+            break;
+        }
+        if (Has(hr, watchpoint)) {
+            RewindBreakpointInstruction();
+            system.GetDebugger().NotifyThreadWatchpoint(current_thread, *HaltedWatchpoint());
+            current_thread->RequestSuspend(SuspendType::Debug);
             break;
         }
 
@@ -136,4 +143,36 @@ void ARM_Interface::Run() {
     }
 }
 
+void ARM_Interface::LoadWatchpointArray(const WatchpointArray& wp) {
+    watchpoints = ℘
+}
+
+const Kernel::DebugWatchpoint* ARM_Interface::MatchingWatchpoint(
+    VAddr addr, u64 size, Kernel::DebugWatchpointType access_type) const {
+    if (!watchpoints) {
+        return nullptr;
+    }
+
+    const VAddr start_address{addr};
+    const VAddr end_address{addr + size};
+
+    for (size_t i = 0; i < Core::Hardware::NUM_WATCHPOINTS; i++) {
+        const auto& watch{(*watchpoints)[i]};
+
+        if (end_address <= watch.start_address) {
+            continue;
+        }
+        if (start_address >= watch.end_address) {
+            continue;
+        }
+        if ((access_type & watch.type) == Kernel::DebugWatchpointType::None) {
+            continue;
+        }
+
+        return &watch;
+    }
+
+    return nullptr;
+}
+
 } // namespace Core

src/core/arm/arm_interface.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <array>
+#include <span>
 #include <vector>
 
 #include <dynarmic/interface/halt_reason.h>
@@ -19,13 +20,16 @@ struct PageTable;
 
 namespace Kernel {
 enum class VMAPermission : u8;
-}
+enum class DebugWatchpointType : u8;
+struct DebugWatchpoint;
+} // namespace Kernel
 
 namespace Core {
 class System;
 class CPUInterruptHandler;
 
 using CPUInterrupts = std::array<CPUInterruptHandler, Core::Hardware::NUM_CPU_CORES>;
+using WatchpointArray = std::array<Kernel::DebugWatchpoint, Core::Hardware::NUM_WATCHPOINTS>;
 
 /// Generic ARMv8 CPU interface
 class ARM_Interface {
@@ -170,6 +174,7 @@ public:
     virtual void SaveContext(ThreadContext64& ctx) = 0;
     virtual void LoadContext(const ThreadContext32& ctx) = 0;
     virtual void LoadContext(const ThreadContext64& ctx) = 0;
+    void LoadWatchpointArray(const WatchpointArray& wp);
 
     /// Clears the exclusive monitor's state.
     virtual void ClearExclusiveState() = 0;
@@ -198,18 +203,24 @@ public:
     static constexpr Dynarmic::HaltReason break_loop = Dynarmic::HaltReason::UserDefined2;
     static constexpr Dynarmic::HaltReason svc_call = Dynarmic::HaltReason::UserDefined3;
     static constexpr Dynarmic::HaltReason breakpoint = Dynarmic::HaltReason::UserDefined4;
+    static constexpr Dynarmic::HaltReason watchpoint = Dynarmic::HaltReason::UserDefined5;
 
 protected:
     /// System context that this ARM interface is running under.
     System& system;
     CPUInterrupts& interrupt_handlers;
+    const WatchpointArray* watchpoints;
     bool uses_wall_clock;
 
     static void SymbolicateBacktrace(Core::System& system, std::vector<BacktraceEntry>& out);
+    const Kernel::DebugWatchpoint* MatchingWatchpoint(
+        VAddr addr, u64 size, Kernel::DebugWatchpointType access_type) const;
 
     virtual Dynarmic::HaltReason RunJit() = 0;
     virtual Dynarmic::HaltReason StepJit() = 0;
     virtual u32 GetSvcNumber() const = 0;
+    virtual const Kernel::DebugWatchpoint* HaltedWatchpoint() const = 0;
+    virtual void RewindBreakpointInstruction() = 0;
 };
 
 } // namespace Core

src/core/arm/dynarmic/arm_dynarmic_32.cpp

@@ -29,45 +29,62 @@ using namespace Common::Literals;
 class DynarmicCallbacks32 : public Dynarmic::A32::UserCallbacks {
 public:
     explicit DynarmicCallbacks32(ARM_Dynarmic_32& parent_)
-        : parent{parent_}, memory(parent.system.Memory()) {}
+        : parent{parent_},
+          memory(parent.system.Memory()), debugger_enabled{parent.system.DebuggerEnabled()} {}
 
     u8 MemoryRead8(u32 vaddr) override {
+        CheckMemoryAccess(vaddr, 1, Kernel::DebugWatchpointType::Read);
         return memory.Read8(vaddr);
     }
     u16 MemoryRead16(u32 vaddr) override {
+        CheckMemoryAccess(vaddr, 2, Kernel::DebugWatchpointType::Read);
         return memory.Read16(vaddr);
     }
     u32 MemoryRead32(u32 vaddr) override {
+        CheckMemoryAccess(vaddr, 4, Kernel::DebugWatchpointType::Read);
         return memory.Read32(vaddr);
     }
     u64 MemoryRead64(u32 vaddr) override {
+        CheckMemoryAccess(vaddr, 8, Kernel::DebugWatchpointType::Read);
         return memory.Read64(vaddr);
     }
 
     void MemoryWrite8(u32 vaddr, u8 value) override {
-        memory.Write8(vaddr, value);
+        if (CheckMemoryAccess(vaddr, 1, Kernel::DebugWatchpointType::Write)) {
+            memory.Write8(vaddr, value);
+        }
     }
     void MemoryWrite16(u32 vaddr, u16 value) override {
-        memory.Write16(vaddr, value);
+        if (CheckMemoryAccess(vaddr, 2, Kernel::DebugWatchpointType::Write)) {
+            memory.Write16(vaddr, value);
+        }
     }
     void MemoryWrite32(u32 vaddr, u32 value) override {
-        memory.Write32(vaddr, value);
+        if (CheckMemoryAccess(vaddr, 4, Kernel::DebugWatchpointType::Write)) {
+            memory.Write32(vaddr, value);
+        }
     }
     void MemoryWrite64(u32 vaddr, u64 value) override {
-        memory.Write64(vaddr, value);
+        if (CheckMemoryAccess(vaddr, 8, Kernel::DebugWatchpointType::Write)) {
+            memory.Write64(vaddr, value);
+        }
     }
 
     bool MemoryWriteExclusive8(u32 vaddr, u8 value, u8 expected) override {
-        return memory.WriteExclusive8(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 1, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive8(vaddr, value, expected);
     }
     bool MemoryWriteExclusive16(u32 vaddr, u16 value, u16 expected) override {
-        return memory.WriteExclusive16(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 2, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive16(vaddr, value, expected);
     }
     bool MemoryWriteExclusive32(u32 vaddr, u32 value, u32 expected) override {
-        return memory.WriteExclusive32(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 4, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive32(vaddr, value, expected);
     }
     bool MemoryWriteExclusive64(u32 vaddr, u64 value, u64 expected) override {
-        return memory.WriteExclusive64(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 8, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive64(vaddr, value, expected);
     }
 
     void InterpreterFallback(u32 pc, std::size_t num_instructions) override {
@@ -77,8 +94,8 @@ public:
     }
 
     void ExceptionRaised(u32 pc, Dynarmic::A32::Exception exception) override {
-        if (parent.system.DebuggerEnabled()) {
-            parent.jit.load()->Regs()[15] = pc;
+        if (debugger_enabled) {
+            parent.SaveContext(parent.breakpoint_context);
             parent.jit.load()->HaltExecution(ARM_Interface::breakpoint);
             return;
         }
@@ -117,9 +134,26 @@ public:
         return std::max<s64>(parent.system.CoreTiming().GetDowncount(), 0);
     }
 
+    bool CheckMemoryAccess(VAddr addr, u64 size, Kernel::DebugWatchpointType type) {
+        if (!debugger_enabled) {
+            return true;
+        }
+
+        const auto match{parent.MatchingWatchpoint(addr, size, type)};
+        if (match) {
+            parent.SaveContext(parent.breakpoint_context);
+            parent.jit.load()->HaltExecution(ARM_Interface::watchpoint);
+            parent.halted_watchpoint = match;
+            return false;
+        }
+
+        return true;
+    }
+
     ARM_Dynarmic_32& parent;
     Core::Memory::Memory& memory;
     std::size_t num_interpreted_instructions{};
+    bool debugger_enabled{};
     static constexpr u64 minimum_run_cycles = 1000U;
 };
 
@@ -154,6 +188,11 @@ std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable*
     config.code_cache_size = 512_MiB;
     config.far_code_offset = 400_MiB;
 
+    // Allow memory fault handling to work
+    if (system.DebuggerEnabled()) {
+        config.check_halt_on_memory_access = true;
+    }
+
     // null_jit
     if (!page_table) {
         // Don't waste too much memory on null_jit
@@ -248,6 +287,14 @@ u32 ARM_Dynarmic_32::GetSvcNumber() const {
     return svc_swi;
 }
 
+const Kernel::DebugWatchpoint* ARM_Dynarmic_32::HaltedWatchpoint() const {
+    return halted_watchpoint;
+}
+
+void ARM_Dynarmic_32::RewindBreakpointInstruction() {
+    LoadContext(breakpoint_context);
+}
+
 ARM_Dynarmic_32::ARM_Dynarmic_32(System& system_, CPUInterrupts& interrupt_handlers_,
                                  bool uses_wall_clock_, ExclusiveMonitor& exclusive_monitor_,
                                  std::size_t core_index_)

src/core/arm/dynarmic/arm_dynarmic_32.h

@@ -72,6 +72,8 @@ protected:
     Dynarmic::HaltReason RunJit() override;
     Dynarmic::HaltReason StepJit() override;
     u32 GetSvcNumber() const override;
+    const Kernel::DebugWatchpoint* HaltedWatchpoint() const override;
+    void RewindBreakpointInstruction() override;
 
 private:
     std::shared_ptr<Dynarmic::A32::Jit> MakeJit(Common::PageTable* page_table) const;
@@ -98,6 +100,10 @@ private:
 
     // SVC callback
     u32 svc_swi{};
+
+    // Watchpoint info
+    const Kernel::DebugWatchpoint* halted_watchpoint;
+    ThreadContext32 breakpoint_context;
 };
 
 } // namespace Core

src/core/arm/dynarmic/arm_dynarmic_64.cpp

@@ -29,55 +29,76 @@ using namespace Common::Literals;
 class DynarmicCallbacks64 : public Dynarmic::A64::UserCallbacks {
 public:
     explicit DynarmicCallbacks64(ARM_Dynarmic_64& parent_)
-        : parent{parent_}, memory(parent.system.Memory()) {}
+        : parent{parent_},
+          memory(parent.system.Memory()), debugger_enabled{parent.system.DebuggerEnabled()} {}
 
     u8 MemoryRead8(u64 vaddr) override {
+        CheckMemoryAccess(vaddr, 1, Kernel::DebugWatchpointType::Read);
         return memory.Read8(vaddr);
     }
     u16 MemoryRead16(u64 vaddr) override {
+        CheckMemoryAccess(vaddr, 2, Kernel::DebugWatchpointType::Read);
         return memory.Read16(vaddr);
     }
     u32 MemoryRead32(u64 vaddr) override {
+        CheckMemoryAccess(vaddr, 4, Kernel::DebugWatchpointType::Read);
         return memory.Read32(vaddr);
     }
     u64 MemoryRead64(u64 vaddr) override {
+        CheckMemoryAccess(vaddr, 8, Kernel::DebugWatchpointType::Read);
         return memory.Read64(vaddr);
     }
     Vector MemoryRead128(u64 vaddr) override {
+        CheckMemoryAccess(vaddr, 16, Kernel::DebugWatchpointType::Read);
         return {memory.Read64(vaddr), memory.Read64(vaddr + 8)};
     }
 
     void MemoryWrite8(u64 vaddr, u8 value) override {
-        memory.Write8(vaddr, value);
+        if (CheckMemoryAccess(vaddr, 1, Kernel::DebugWatchpointType::Write)) {
+            memory.Write8(vaddr, value);
+        }
     }
     void MemoryWrite16(u64 vaddr, u16 value) override {
-        memory.Write16(vaddr, value);
+        if (CheckMemoryAccess(vaddr, 2, Kernel::DebugWatchpointType::Write)) {
+            memory.Write16(vaddr, value);
+        }
     }
     void MemoryWrite32(u64 vaddr, u32 value) override {
-        memory.Write32(vaddr, value);
+        if (CheckMemoryAccess(vaddr, 4, Kernel::DebugWatchpointType::Write)) {
+            memory.Write32(vaddr, value);
+        }
     }
     void MemoryWrite64(u64 vaddr, u64 value) override {
-        memory.Write64(vaddr, value);
+        if (CheckMemoryAccess(vaddr, 8, Kernel::DebugWatchpointType::Write)) {
+            memory.Write64(vaddr, value);
+        }
     }
     void MemoryWrite128(u64 vaddr, Vector value) override {
-        memory.Write64(vaddr, value[0]);
-        memory.Write64(vaddr + 8, value[1]);
+        if (CheckMemoryAccess(vaddr, 16, Kernel::DebugWatchpointType::Write)) {
+            memory.Write64(vaddr, value[0]);
+            memory.Write64(vaddr + 8, value[1]);
+        }
     }
 
     bool MemoryWriteExclusive8(u64 vaddr, std::uint8_t value, std::uint8_t expected) override {
-        return memory.WriteExclusive8(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 1, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive8(vaddr, value, expected);
     }
     bool MemoryWriteExclusive16(u64 vaddr, std::uint16_t value, std::uint16_t expected) override {
-        return memory.WriteExclusive16(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 2, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive16(vaddr, value, expected);
     }
     bool MemoryWriteExclusive32(u64 vaddr, std::uint32_t value, std::uint32_t expected) override {
-        return memory.WriteExclusive32(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 4, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive32(vaddr, value, expected);
     }
     bool MemoryWriteExclusive64(u64 vaddr, std::uint64_t value, std::uint64_t expected) override {
-        return memory.WriteExclusive64(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 8, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive64(vaddr, value, expected);
     }
     bool MemoryWriteExclusive128(u64 vaddr, Vector value, Vector expected) override {
-        return memory.WriteExclusive128(vaddr, value, expected);
+        return CheckMemoryAccess(vaddr, 16, Kernel::DebugWatchpointType::Write) &&
+               memory.WriteExclusive128(vaddr, value, expected);
     }
 
     void InterpreterFallback(u64 pc, std::size_t num_instructions) override {
@@ -118,8 +139,8 @@ public:
         case Dynarmic::A64::Exception::Yield:
             return;
         default:
-            if (parent.system.DebuggerEnabled()) {
-                parent.jit.load()->SetPC(pc);
+            if (debugger_enabled) {
+                parent.SaveContext(parent.breakpoint_context);
                 parent.jit.load()->HaltExecution(ARM_Interface::breakpoint);
                 return;
             }
@@ -160,10 +181,27 @@ public:
         return parent.system.CoreTiming().GetClockTicks();
     }
 
+    bool CheckMemoryAccess(VAddr addr, u64 size, Kernel::DebugWatchpointType type) {
+        if (!debugger_enabled) {
+            return true;
+        }
+
+        const auto match{parent.MatchingWatchpoint(addr, size, type)};
+        if (match) {
+            parent.SaveContext(parent.breakpoint_context);
+            parent.jit.load()->HaltExecution(ARM_Interface::watchpoint);
+            parent.halted_watchpoint = match;
+            return false;
+        }
+
+        return true;
+    }
+
     ARM_Dynarmic_64& parent;
     Core::Memory::Memory& memory;
     u64 tpidrro_el0 = 0;
     u64 tpidr_el0 = 0;
+    bool debugger_enabled{};
     static constexpr u64 minimum_run_cycles = 1000U;
 };
 
@@ -214,6 +252,11 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable*
     config.code_cache_size = 512_MiB;
     config.far_code_offset = 400_MiB;
 
+    // Allow memory fault handling to work
+    if (system.DebuggerEnabled()) {
+        config.check_halt_on_memory_access = true;
+    }
+
     // null_jit
     if (!page_table) {
         // Don't waste too much memory on null_jit
@@ -308,6 +351,14 @@ u32 ARM_Dynarmic_64::GetSvcNumber() const {
     return svc_swi;
 }
 
+const Kernel::DebugWatchpoint* ARM_Dynarmic_64::HaltedWatchpoint() const {
+    return halted_watchpoint;
+}
+
+void ARM_Dynarmic_64::RewindBreakpointInstruction() {
+    LoadContext(breakpoint_context);
+}
+
 ARM_Dynarmic_64::ARM_Dynarmic_64(System& system_, CPUInterrupts& interrupt_handlers_,
                                  bool uses_wall_clock_, ExclusiveMonitor& exclusive_monitor_,
                                  std::size_t core_index_)

src/core/arm/dynarmic/arm_dynarmic_64.h

@@ -66,6 +66,8 @@ protected:
     Dynarmic::HaltReason RunJit() override;
     Dynarmic::HaltReason StepJit() override;
     u32 GetSvcNumber() const override;
+    const Kernel::DebugWatchpoint* HaltedWatchpoint() const override;
+    void RewindBreakpointInstruction() override;
 
 private:
     std::shared_ptr<Dynarmic::A64::Jit> MakeJit(Common::PageTable* page_table,
@@ -91,6 +93,10 @@ private:
 
     // SVC callback
     u32 svc_swi{};
+
+    // Breakpoint info
+    const Kernel::DebugWatchpoint* halted_watchpoint;
+    ThreadContext64 breakpoint_context;
 };
 
 } // namespace Core

src/core/cpu_manager.cpp

@@ -26,6 +26,7 @@ void CpuManager::ThreadStart(std::stop_token stop_token, CpuManager& cpu_manager
 
 void CpuManager::Initialize() {
     num_cores = is_multicore ? Core::Hardware::NUM_CPU_CORES : 1;
+    gpu_barrier = std::make_unique<Common::Barrier>(num_cores + 1);
 
     for (std::size_t core = 0; core < num_cores; core++) {
         core_data[core].host_thread = std::jthread(ThreadStart, std::ref(*this), core);
@@ -94,7 +95,7 @@ void* CpuManager::GetStartFuncParameter() {
 void CpuManager::MultiCoreRunGuestThread() {
     auto& kernel = system.Kernel();
     kernel.CurrentScheduler()->OnThreadStart();
-    auto* thread = kernel.CurrentScheduler()->GetCurrentThread();
+    auto* thread = kernel.CurrentScheduler()->GetSchedulerCurrentThread();
     auto& host_context = thread->GetHostContext();
     host_context->SetRewindPoint(GuestRewindFunction, this);
     MultiCoreRunGuestLoop();
@@ -131,7 +132,7 @@ void CpuManager::MultiCoreRunIdleThread() {
 void CpuManager::SingleCoreRunGuestThread() {
     auto& kernel = system.Kernel();
     kernel.CurrentScheduler()->OnThreadStart();
-    auto* thread = kernel.CurrentScheduler()->GetCurrentThread();
+    auto* thread = kernel.CurrentScheduler()->GetSchedulerCurrentThread();
     auto& host_context = thread->GetHostContext();
     host_context->SetRewindPoint(GuestRewindFunction, this);
     SingleCoreRunGuestLoop();
@@ -171,7 +172,7 @@ void CpuManager::PreemptSingleCore(bool from_running_enviroment) {
     {
         auto& kernel = system.Kernel();
         auto& scheduler = kernel.Scheduler(current_core);
-        Kernel::KThread* current_thread = scheduler.GetCurrentThread();
+        Kernel::KThread* current_thread = scheduler.GetSchedulerCurrentThread();
         if (idle_count >= 4 || from_running_enviroment) {
             if (!from_running_enviroment) {
                 system.CoreTiming().Idle();
@@ -183,7 +184,7 @@ void CpuManager::PreemptSingleCore(bool from_running_enviroment) {
         }
         current_core.store((current_core + 1) % Core::Hardware::NUM_CPU_CORES);
         system.CoreTiming().ResetTicks();
-        scheduler.Unload(scheduler.GetCurrentThread());
+        scheduler.Unload(scheduler.GetSchedulerCurrentThread());
 
         auto& next_scheduler = kernel.Scheduler(current_core);
         Common::Fiber::YieldTo(current_thread->GetHostContext(), *next_scheduler.ControlContext());
@@ -192,10 +193,8 @@ void CpuManager::PreemptSingleCore(bool from_running_enviroment) {
     // May have changed scheduler
     {
         auto& scheduler = system.Kernel().Scheduler(current_core);
-        scheduler.Reload(scheduler.GetCurrentThread());
-        if (!scheduler.IsIdle()) {
-            idle_count = 0;
-        }
+        scheduler.Reload(scheduler.GetSchedulerCurrentThread());
+        idle_count = 0;
     }
 }
 
@@ -230,11 +229,14 @@ void CpuManager::RunThread(std::size_t core) {
     });
 
     // Running
+    gpu_barrier->Sync();
+
     if (!is_async_gpu && !is_multicore) {
         system.GPU().ObtainContext();
     }
 
-    auto current_thread = system.Kernel().CurrentScheduler()->GetCurrentThread();
+    auto* current_thread = system.Kernel().CurrentScheduler()->GetIdleThread();
+    Kernel::SetCurrentThread(system.Kernel(), current_thread);
     Common::Fiber::YieldTo(data.host_context, *current_thread->GetHostContext());
 }
 

src/core/cpu_manager.h

@@ -43,6 +43,10 @@ public:
         is_async_gpu = is_async;
     }
 
+    void OnGpuReady() {
+        gpu_barrier->Sync();
+    }
+
     void Initialize();
     void Shutdown();
 
@@ -81,6 +85,7 @@ private:
         std::jthread host_thread;
     };
 
+    std::unique_ptr<Common::Barrier> gpu_barrier{};
     std::array<CoreData, Core::Hardware::NUM_CPU_CORES> core_data{};
 
     bool is_async_gpu{};

src/core/debugger/debugger.cpp

@@ -44,12 +44,14 @@ static std::span<const u8> ReceiveInto(Readable& r, Buffer& buffer) {
 
 enum class SignalType {
     Stopped,
+    Watchpoint,
     ShuttingDown,
 };
 
 struct SignalInfo {
     SignalType type;
     Kernel::KThread* thread;
+    const Kernel::DebugWatchpoint* watchpoint;
 };
 
 namespace Core {
@@ -157,13 +159,19 @@ private:
     void PipeData(std::span<const u8> data) {
         switch (info.type) {
         case SignalType::Stopped:
+        case SignalType::Watchpoint:
             // Stop emulation.
             PauseEmulation();
 
             // Notify the client.
             active_thread = info.thread;
             UpdateActiveThread();
-            frontend->Stopped(active_thread);
+
+            if (info.type == SignalType::Watchpoint) {
+                frontend->Watchpoint(active_thread, *info.watchpoint);
+            } else {
+                frontend->Stopped(active_thread);
+            }
 
             break;
         case SignalType::ShuttingDown:
@@ -290,12 +298,17 @@ Debugger::Debugger(Core::System& system, u16 port) {
 Debugger::~Debugger() = default;
 
 bool Debugger::NotifyThreadStopped(Kernel::KThread* thread) {
-    return impl && impl->SignalDebugger(SignalInfo{SignalType::Stopped, thread});
+    return impl && impl->SignalDebugger(SignalInfo{SignalType::Stopped, thread, nullptr});
+}
+
+bool Debugger::NotifyThreadWatchpoint(Kernel::KThread* thread,
+                                      const Kernel::DebugWatchpoint& watch) {
+    return impl && impl->SignalDebugger(SignalInfo{SignalType::Watchpoint, thread, &watch});
 }
 
 void Debugger::NotifyShutdown() {
     if (impl) {
-        impl->SignalDebugger(SignalInfo{SignalType::ShuttingDown, nullptr});
+        impl->SignalDebugger(SignalInfo{SignalType::ShuttingDown, nullptr, nullptr});
     }
 }
 

src/core/debugger/debugger.h

@@ -9,7 +9,8 @@
 
 namespace Kernel {
 class KThread;
-}
+struct DebugWatchpoint;
+} // namespace Kernel
 
 namespace Core {
 class System;
@@ -40,6 +41,11 @@ public:
      */
     void NotifyShutdown();
 
+    /*
+     * Notify the debugger that the given thread has stopped due to hitting a watchpoint.
+     */
+    bool NotifyThreadWatchpoint(Kernel::KThread* thread, const Kernel::DebugWatchpoint& watch);
+
 private:
     std::unique_ptr<DebuggerImpl> impl;
 };

src/core/debugger/debugger_interface.h

@@ -11,7 +11,8 @@
 
 namespace Kernel {
 class KThread;
-}
+struct DebugWatchpoint;
+} // namespace Kernel
 
 namespace Core {
 
@@ -71,6 +72,11 @@ public:
      */
     virtual void ShuttingDown() = 0;
 
+    /*
+     * Called when emulation has stopped on a watchpoint.
+     */
+    virtual void Watchpoint(Kernel::KThread* thread, const Kernel::DebugWatchpoint& watch) = 0;
+
     /**
      * Called when new data is asynchronously received on the client socket.
      * A list of actions to perform is returned.

src/core/debugger/gdbstub.cpp

@@ -112,6 +112,23 @@ void GDBStub::Stopped(Kernel::KThread* thread) {
     SendReply(arch->ThreadStatus(thread, GDB_STUB_SIGTRAP));
 }
 
+void GDBStub::Watchpoint(Kernel::KThread* thread, const Kernel::DebugWatchpoint& watch) {
+    const auto status{arch->ThreadStatus(thread, GDB_STUB_SIGTRAP)};
+
+    switch (watch.type) {
+    case Kernel::DebugWatchpointType::Read:
+        SendReply(fmt::format("{}rwatch:{:x};", status, watch.start_address));
+        break;
+    case Kernel::DebugWatchpointType::Write:
+        SendReply(fmt::format("{}watch:{:x};", status, watch.start_address));
+        break;
+    case Kernel::DebugWatchpointType::ReadOrWrite:
+    default:
+        SendReply(fmt::format("{}awatch:{:x};", status, watch.start_address));
+        break;
+    }
+}
+
 std::vector<DebuggerAction> GDBStub::ClientData(std::span<const u8> data) {
     std::vector<DebuggerAction> actions;
     current_command.insert(current_command.end(), data.begin(), data.end());
@@ -235,6 +252,7 @@ void GDBStub::ExecuteCommand(std::string_view packet, std::vector<DebuggerAction
         const auto sep{std::find(command.begin(), command.end(), '=') - command.begin() + 1};
         const size_t reg{static_cast<size_t>(strtoll(command.data(), nullptr, 16))};
         arch->RegWrite(backend.GetActiveThread(), reg, std::string_view(command).substr(sep));
+        SendReply(GDB_STUB_REPLY_OK);
         break;
     }
     case 'm': {
@@ -278,44 +296,124 @@ void GDBStub::ExecuteCommand(std::string_view packet, std::vector<DebuggerAction
     case 'c':
         actions.push_back(DebuggerAction::Continue);
         break;
-    case 'Z': {
-        const auto addr_sep{std::find(command.begin(), command.end(), ',') - command.begin() + 1};
-        const size_t addr{static_cast<size_t>(strtoll(command.data() + addr_sep, nullptr, 16))};
-
-        if (system.Memory().IsValidVirtualAddress(addr)) {
-            replaced_instructions[addr] = system.Memory().Read32(addr);
-            system.Memory().Write32(addr, arch->BreakpointInstruction());
-            system.InvalidateCpuInstructionCacheRange(addr, sizeof(u32));
-
-            SendReply(GDB_STUB_REPLY_OK);
-        } else {
-            SendReply(GDB_STUB_REPLY_ERR);
-        }
+    case 'Z':
+        HandleBreakpointInsert(command);
         break;
-    }
-    case 'z': {
-        const auto addr_sep{std::find(command.begin(), command.end(), ',') - command.begin() + 1};
-        const size_t addr{static_cast<size_t>(strtoll(command.data() + addr_sep, nullptr, 16))};
-
-        const auto orig_insn{replaced_instructions.find(addr)};
-        if (system.Memory().IsValidVirtualAddress(addr) &&
-            orig_insn != replaced_instructions.end()) {
-            system.Memory().Write32(addr, orig_insn->second);
-            system.InvalidateCpuInstructionCacheRange(addr, sizeof(u32));
-            replaced_instructions.erase(addr);
-
-            SendReply(GDB_STUB_REPLY_OK);
-        } else {
-            SendReply(GDB_STUB_REPLY_ERR);
-        }
+    case 'z':
+        HandleBreakpointRemove(command);
         break;
-    }
     default:
         SendReply(GDB_STUB_REPLY_EMPTY);
         break;
     }
 }
 
+enum class BreakpointType {
+    Software = 0,
+    Hardware = 1,
+    WriteWatch = 2,
+    ReadWatch = 3,
+    AccessWatch = 4,
+};
+
+void GDBStub::HandleBreakpointInsert(std::string_view command) {
+    const auto type{static_cast<BreakpointType>(strtoll(command.data(), nullptr, 16))};
+    const auto addr_sep{std::find(command.begin(), command.end(), ',') - command.begin() + 1};
+    const auto size_sep{std::find(command.begin() + addr_sep, command.end(), ',') -
+                        command.begin() + 1};
+    const size_t addr{static_cast<size_t>(strtoll(command.data() + addr_sep, nullptr, 16))};
+    const size_t size{static_cast<size_t>(strtoll(command.data() + size_sep, nullptr, 16))};
+
+    if (!system.Memory().IsValidVirtualAddressRange(addr, size)) {
+        SendReply(GDB_STUB_REPLY_ERR);
+        return;
+    }
+
+    bool success{};
+
+    switch (type) {
+    case BreakpointType::Software:
+        replaced_instructions[addr] = system.Memory().Read32(addr);
+        system.Memory().Write32(addr, arch->BreakpointInstruction());
+        system.InvalidateCpuInstructionCacheRange(addr, sizeof(u32));
+        success = true;
+        break;
+    case BreakpointType::WriteWatch:
+        success = system.CurrentProcess()->InsertWatchpoint(system, addr, size,
+                                                            Kernel::DebugWatchpointType::Write);
+        break;
+    case BreakpointType::ReadWatch:
+        success = system.CurrentProcess()->InsertWatchpoint(system, addr, size,
+                                                            Kernel::DebugWatchpointType::Read);
+        break;
+    case BreakpointType::AccessWatch:
+        success = system.CurrentProcess()->InsertWatchpoint(
+            system, addr, size, Kernel::DebugWatchpointType::ReadOrWrite);
+        break;
+    case BreakpointType::Hardware:
+    default:
+        SendReply(GDB_STUB_REPLY_EMPTY);
+        return;
+    }
+
+    if (success) {
+        SendReply(GDB_STUB_REPLY_OK);
+    } else {
+        SendReply(GDB_STUB_REPLY_ERR);
+    }
+}
+
+void GDBStub::HandleBreakpointRemove(std::string_view command) {
+    const auto type{static_cast<BreakpointType>(strtoll(command.data(), nullptr, 16))};
+    const auto addr_sep{std::find(command.begin(), command.end(), ',') - command.begin() + 1};
+    const auto size_sep{std::find(command.begin() + addr_sep, command.end(), ',') -
+                        command.begin() + 1};
+    const size_t addr{static_cast<size_t>(strtoll(command.data() + addr_sep, nullptr, 16))};
+    const size_t size{static_cast<size_t>(strtoll(command.data() + size_sep, nullptr, 16))};
+
+    if (!system.Memory().IsValidVirtualAddressRange(addr, size)) {
+        SendReply(GDB_STUB_REPLY_ERR);
+        return;
+    }
+
+    bool success{};
+
+    switch (type) {
+    case BreakpointType::Software: {
+        const auto orig_insn{replaced_instructions.find(addr)};
+        if (orig_insn != replaced_instructions.end()) {
+            system.Memory().Write32(addr, orig_insn->second);
+            system.InvalidateCpuInstructionCacheRange(addr, sizeof(u32));
+            replaced_instructions.erase(addr);
+            success = true;
+        }
+        break;
+    }
+    case BreakpointType::WriteWatch:
+        success = system.CurrentProcess()->RemoveWatchpoint(system, addr, size,
+                                                            Kernel::DebugWatchpointType::Write);
+        break;
+    case BreakpointType::ReadWatch:
+        success = system.CurrentProcess()->RemoveWatchpoint(system, addr, size,
+                                                            Kernel::DebugWatchpointType::Read);
+        break;
+    case BreakpointType::AccessWatch:
+        success = system.CurrentProcess()->RemoveWatchpoint(
+            system, addr, size, Kernel::DebugWatchpointType::ReadOrWrite);
+        break;
+    case BreakpointType::Hardware:
+    default:
+        SendReply(GDB_STUB_REPLY_EMPTY);
+        return;
+    }
+
+    if (success) {
+        SendReply(GDB_STUB_REPLY_OK);
+    } else {
+        SendReply(GDB_STUB_REPLY_ERR);
+    }
+}
+
 // Structure offsets are from Atmosphere
 // See osdbg_thread_local_region.os.horizon.hpp and osdbg_thread_type.os.horizon.hpp
 

src/core/debugger/gdbstub.h

@@ -24,6 +24,7 @@ public:
     void Connected() override;
     void Stopped(Kernel::KThread* thread) override;
     void ShuttingDown() override;
+    void Watchpoint(Kernel::KThread* thread, const Kernel::DebugWatchpoint& watch) override;
     std::vector<DebuggerAction> ClientData(std::span<const u8> data) override;
 
 private:
@@ -31,6 +32,8 @@ private:
     void ExecuteCommand(std::string_view packet, std::vector<DebuggerAction>& actions);
     void HandleVCont(std::string_view command, std::vector<DebuggerAction>& actions);
     void HandleQuery(std::string_view command);
+    void HandleBreakpointInsert(std::string_view command);
+    void HandleBreakpointRemove(std::string_view command);
     std::vector<char>::const_iterator CommandEnd() const;
     std::optional<std::string> DetachCommand();
     Kernel::KThread* GetThreadByID(u64 thread_id);

src/core/hardware_properties.h

@@ -25,6 +25,9 @@ constexpr std::array<s32, Common::BitSize<u64>()> VirtualToPhysicalCoreMap{
     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3,
 };
 
+// Cortex-A57 supports 4 memory watchpoints
+constexpr u64 NUM_WATCHPOINTS = 4;
+
 } // namespace Hardware
 
 } // namespace Core

src/core/hle/kernel/k_address_arbiter.cpp

@@ -234,7 +234,7 @@ ResultCode KAddressArbiter::SignalAndModifyByWaitingCountIfEqual(VAddr addr, s32
 
 ResultCode KAddressArbiter::WaitIfLessThan(VAddr addr, s32 value, bool decrement, s64 timeout) {
     // Prepare to wait.
-    KThread* cur_thread = kernel.CurrentScheduler()->GetCurrentThread();
+    KThread* cur_thread = GetCurrentThreadPointer(kernel);
     ThreadQueueImplForKAddressArbiter wait_queue(kernel, std::addressof(thread_tree));
 
     {
@@ -287,7 +287,7 @@ ResultCode KAddressArbiter::WaitIfLessThan(VAddr addr, s32 value, bool decrement
 
 ResultCode KAddressArbiter::WaitIfEqual(VAddr addr, s32 value, s64 timeout) {
     // Prepare to wait.
-    KThread* cur_thread = kernel.CurrentScheduler()->GetCurrentThread();
+    KThread* cur_thread = GetCurrentThreadPointer(kernel);
     ThreadQueueImplForKAddressArbiter wait_queue(kernel, std::addressof(thread_tree));
 
     {

src/core/hle/kernel/k_condition_variable.cpp

@@ -106,7 +106,7 @@ KConditionVariable::KConditionVariable(Core::System& system_)
 KConditionVariable::~KConditionVariable() = default;
 
 ResultCode KConditionVariable::SignalToAddress(VAddr addr) {
-    KThread* owner_thread = kernel.CurrentScheduler()->GetCurrentThread();
+    KThread* owner_thread = GetCurrentThreadPointer(kernel);
 
     // Signal the address.
     {
@@ -147,7 +147,7 @@ ResultCode KConditionVariable::SignalToAddress(VAddr addr) {
 }
 
 ResultCode KConditionVariable::WaitForAddress(Handle handle, VAddr addr, u32 value) {
-    KThread* cur_thread = kernel.CurrentScheduler()->GetCurrentThread();
+    KThread* cur_thread = GetCurrentThreadPointer(kernel);
     ThreadQueueImplForKConditionVariableWaitForAddress wait_queue(kernel);
 
     // Wait for the address.

src/core/hle/kernel/k_interrupt_manager.cpp

@@ -15,8 +15,7 @@ void HandleInterrupt(KernelCore& kernel, s32 core_id) {
         return;
     }
 
-    auto& scheduler = kernel.Scheduler(core_id);
-    auto& current_thread = *scheduler.GetCurrentThread();
+    auto& current_thread = GetCurrentThread(kernel);
 
     // If the user disable count is set, we may need to pin the current thread.
     if (current_thread.GetUserDisableCount() && !process->GetPinnedThread(core_id)) {
@@ -26,7 +25,7 @@ void HandleInterrupt(KernelCore& kernel, s32 core_id) {
         process->PinCurrentThread(core_id);
 
         // Set the interrupt flag for the thread.
-        scheduler.GetCurrentThread()->SetInterruptFlag();
+        GetCurrentThread(kernel).SetInterruptFlag();
     }
 }
 

src/core/hle/kernel/k_page_table.cpp

@@ -65,7 +65,6 @@ ResultCode KPageTable::InitializeForProcess(FileSys::ProgramAddressSpaceType as_
     std::size_t alias_region_size{GetSpaceSize(KAddressSpaceInfo::Type::Alias)};
     std::size_t heap_region_size{GetSpaceSize(KAddressSpaceInfo::Type::Heap)};
 
-    ASSERT(start <= code_addr);
     ASSERT(code_addr < code_addr + code_size);
     ASSERT(code_addr + code_size - 1 <= end - 1);
 

src/core/hle/kernel/k_process.cpp

@@ -57,18 +57,13 @@ void SetupMainThread(Core::System& system, KProcess& owner_process, u32 priority
     thread->GetContext64().cpu_registers[0] = 0;
     thread->GetContext32().cpu_registers[1] = thread_handle;
     thread->GetContext64().cpu_registers[1] = thread_handle;
-    thread->DisableDispatch();
 
-    auto& kernel = system.Kernel();
-    // Threads by default are dormant, wake up the main thread so it runs when the scheduler fires
-    {
-        KScopedSchedulerLock lock{kernel};
-        thread->SetState(ThreadState::Runnable);
-
-        if (system.DebuggerEnabled()) {
-            thread->RequestSuspend(SuspendType::Debug);
-        }
+    if (system.DebuggerEnabled()) {
+        thread->RequestSuspend(SuspendType::Debug);
     }
+
+    // Run our thread.
+    void(thread->Run());
 }
 } // Anonymous namespace
 
@@ -181,7 +176,8 @@ void KProcess::PinCurrentThread(s32 core_id) {
     ASSERT(kernel.GlobalSchedulerContext().IsLocked());
 
     // Get the current thread.
-    KThread* cur_thread = kernel.Scheduler(static_cast<std::size_t>(core_id)).GetCurrentThread();
+    KThread* cur_thread =
+        kernel.Scheduler(static_cast<std::size_t>(core_id)).GetSchedulerCurrentThread();
 
     // If the thread isn't terminated, pin it.
     if (!cur_thread->IsTerminationRequested()) {
@@ -198,7 +194,8 @@ void KProcess::UnpinCurrentThread(s32 core_id) {
     ASSERT(kernel.GlobalSchedulerContext().IsLocked());
 
     // Get the current thread.
-    KThread* cur_thread = kernel.Scheduler(static_cast<std::size_t>(core_id)).GetCurrentThread();
+    KThread* cur_thread =
+        kernel.Scheduler(static_cast<std::size_t>(core_id)).GetSchedulerCurrentThread();
 
     // Unpin it.
     cur_thread->Unpin();
@@ -425,11 +422,11 @@ void KProcess::PrepareForTermination() {
     ChangeStatus(ProcessStatus::Exiting);
 
     const auto stop_threads = [this](const std::vector<KThread*>& in_thread_list) {
-        for (auto& thread : in_thread_list) {
+        for (auto* thread : in_thread_list) {
             if (thread->GetOwnerProcess() != this)
                 continue;
 
-            if (thread == kernel.CurrentScheduler()->GetCurrentThread())
+            if (thread == GetCurrentThreadPointer(kernel))
                 continue;
 
             // TODO(Subv): When are the other running/ready threads terminated?
@@ -584,6 +581,52 @@ ResultCode KProcess::DeleteThreadLocalRegion(VAddr addr) {
     return ResultSuccess;
 }
 
+bool KProcess::InsertWatchpoint(Core::System& system, VAddr addr, u64 size,
+                                DebugWatchpointType type) {
+    const auto watch{std::find_if(watchpoints.begin(), watchpoints.end(), [&](const auto& wp) {
+        return wp.type == DebugWatchpointType::None;
+    })};
+
+    if (watch == watchpoints.end()) {
+        return false;
+    }
+
+    watch->start_address = addr;
+    watch->end_address = addr + size;
+    watch->type = type;
+
+    for (VAddr page = Common::AlignDown(addr, PageSize); page < addr + size; page += PageSize) {
+        debug_page_refcounts[page]++;
+        system.Memory().MarkRegionDebug(page, PageSize, true);
+    }
+
+    return true;
+}
+
+bool KProcess::RemoveWatchpoint(Core::System& system, VAddr addr, u64 size,
+                                DebugWatchpointType type) {
+    const auto watch{std::find_if(watchpoints.begin(), watchpoints.end(), [&](const auto& wp) {
+        return wp.start_address == addr && wp.end_address == addr + size && wp.type == type;
+    })};
+
+    if (watch == watchpoints.end()) {
+        return false;
+    }
+
+    watch->start_address = 0;
+    watch->end_address = 0;
+    watch->type = DebugWatchpointType::None;
+
+    for (VAddr page = Common::AlignDown(addr, PageSize); page < addr + size; page += PageSize) {
+        debug_page_refcounts[page]--;
+        if (!debug_page_refcounts[page]) {
+            system.Memory().MarkRegionDebug(page, PageSize, false);
+        }
+    }
+
+    return true;
+}
+
 void KProcess::LoadModule(CodeSet code_set, VAddr base_addr) {
     const auto ReprotectSegment = [&](const CodeSet::Segment& segment,
                                       Svc::MemoryPermission permission) {

src/core/hle/kernel/k_process.h

@@ -7,6 +7,7 @@
 #include <array>
 #include <cstddef>
 #include <list>
+#include <map>
 #include <string>
 #include "common/common_types.h"
 #include "core/hle/kernel/k_address_arbiter.h"
@@ -68,6 +69,20 @@ enum class ProcessActivity : u32 {
     Paused,
 };
 
+enum class DebugWatchpointType : u8 {
+    None = 0,
+    Read = 1 << 0,
+    Write = 1 << 1,
+    ReadOrWrite = Read | Write,
+};
+DECLARE_ENUM_FLAG_OPERATORS(DebugWatchpointType);
+
+struct DebugWatchpoint {
+    VAddr start_address;
+    VAddr end_address;
+    DebugWatchpointType type;
+};
+
 class KProcess final : public KAutoObjectWithSlabHeapAndContainer<KProcess, KWorkerTask> {
     KERNEL_AUTOOBJECT_TRAITS(KProcess, KSynchronizationObject);
 
@@ -374,6 +389,19 @@ public:
     // Frees a used TLS slot identified by the given address
     ResultCode DeleteThreadLocalRegion(VAddr addr);
 
+    ///////////////////////////////////////////////////////////////////////////////////////////////
+    // Debug watchpoint management
+
+    // Attempts to insert a watchpoint into a free slot. Returns false if none are available.
+    bool InsertWatchpoint(Core::System& system, VAddr addr, u64 size, DebugWatchpointType type);
+
+    // Attempts to remove the watchpoint specified by the given parameters.
+    bool RemoveWatchpoint(Core::System& system, VAddr addr, u64 size, DebugWatchpointType type);
+
+    const std::array<DebugWatchpoint, Core::Hardware::NUM_WATCHPOINTS>& GetWatchpoints() const {
+        return watchpoints;
+    }
+
 private:
     void PinThread(s32 core_id, KThread* thread) {
         ASSERT(0 <= core_id && core_id < static_cast<s32>(Core::Hardware::NUM_CPU_CORES));
@@ -478,6 +506,8 @@ private:
     std::array<KThread*, Core::Hardware::NUM_CPU_CORES> running_threads{};
     std::array<u64, Core::Hardware::NUM_CPU_CORES> running_thread_idle_counts{};
     std::array<KThread*, Core::Hardware::NUM_CPU_CORES> pinned_threads{};
+    std::array<DebugWatchpoint, Core::Hardware::NUM_WATCHPOINTS> watchpoints{};
+    std::map<VAddr, u64> debug_page_refcounts;
 
     KThread* exception_thread{};
 

src/core/hle/kernel/k_scheduler.cpp

@@ -317,7 +317,7 @@ void KScheduler::RotateScheduledQueue(s32 cpu_core_id, s32 priority) {
 
     {
         KThread* best_thread = priority_queue.GetScheduledFront(cpu_core_id);
-        if (best_thread == GetCurrentThread()) {
+        if (best_thread == GetCurrentThreadPointer(kernel)) {
             best_thread = priority_queue.GetScheduledNext(cpu_core_id, best_thread);
         }
 
@@ -424,7 +424,7 @@ void KScheduler::YieldWithoutCoreMigration(KernelCore& kernel) {
     ASSERT(kernel.CurrentProcess() != nullptr);
 
     // Get the current thread and process.
-    KThread& cur_thread = Kernel::GetCurrentThread(kernel);
+    KThread& cur_thread = GetCurrentThread(kernel);
     KProcess& cur_process = *kernel.CurrentProcess();
 
     // If the thread's yield count matches, there's nothing for us to do.
@@ -463,7 +463,7 @@ void KScheduler::YieldWithCoreMigration(KernelCore& kernel) {
     ASSERT(kernel.CurrentProcess() != nullptr);
 
     // Get the current thread and process.
-    KThread& cur_thread = Kernel::GetCurrentThread(kernel);
+    KThread& cur_thread = GetCurrentThread(kernel);
     KProcess& cur_process = *kernel.CurrentProcess();
 
     // If the thread's yield count matches, there's nothing for us to do.
@@ -551,7 +551,7 @@ void KScheduler::YieldToAnyThread(KernelCore& kernel) {
     ASSERT(kernel.CurrentProcess() != nullptr);
 
     // Get the current thread and process.
-    KThread& cur_thread = Kernel::GetCurrentThread(kernel);
+    KThread& cur_thread = GetCurrentThread(kernel);
     KProcess& cur_process = *kernel.CurrentProcess();
 
     // If the thread's yield count matches, there's nothing for us to do.
@@ -642,7 +642,7 @@ KScheduler::~KScheduler() {
     ASSERT(!idle_thread);
 }
 
-KThread* KScheduler::GetCurrentThread() const {
+KThread* KScheduler::GetSchedulerCurrentThread() const {
     if (auto result = current_thread.load(); result) {
         return result;
     }
@@ -654,7 +654,7 @@ u64 KScheduler::GetLastContextSwitchTicks() const {
 }
 
 void KScheduler::RescheduleCurrentCore() {
-    ASSERT(GetCurrentThread()->GetDisableDispatchCount() == 1);
+    ASSERT(GetCurrentThread(system.Kernel()).GetDisableDispatchCount() == 1);
 
     auto& phys_core = system.Kernel().PhysicalCore(core_id);
     if (phys_core.IsInterrupted()) {
@@ -665,7 +665,7 @@ void KScheduler::RescheduleCurrentCore() {
     if (state.needs_scheduling.load()) {
         Schedule();
     } else {
-        GetCurrentThread()->EnableDispatch();
+        GetCurrentThread(system.Kernel()).EnableDispatch();
         guard.Unlock();
     }
 }
@@ -710,6 +710,7 @@ void KScheduler::Reload(KThread* thread) {
     Core::ARM_Interface& cpu_core = system.ArmInterface(core_id);
     cpu_core.LoadContext(thread->GetContext32());
     cpu_core.LoadContext(thread->GetContext64());
+    cpu_core.LoadWatchpointArray(thread->GetOwnerProcess()->GetWatchpoints());
     cpu_core.SetTlsAddress(thread->GetTLSAddress());
     cpu_core.SetTPIDR_EL0(thread->GetTPIDR_EL0());
     cpu_core.ClearExclusiveState();
@@ -717,13 +718,18 @@ void KScheduler::Reload(KThread* thread) {
 
 void KScheduler::SwitchContextStep2() {
     // Load context of new thread
-    Reload(GetCurrentThread());
+    Reload(GetCurrentThreadPointer(system.Kernel()));
 
     RescheduleCurrentCore();
 }
 
+void KScheduler::Schedule() {
+    ASSERT(GetCurrentThread(system.Kernel()).GetDisableDispatchCount() == 1);
+    this->ScheduleImpl();
+}
+
 void KScheduler::ScheduleImpl() {
-    KThread* previous_thread = GetCurrentThread();
+    KThread* previous_thread = GetCurrentThreadPointer(system.Kernel());
     KThread* next_thread = state.highest_priority_thread;
 
     state.needs_scheduling.store(false);
@@ -761,6 +767,7 @@ void KScheduler::ScheduleImpl() {
     old_context = &previous_thread->GetHostContext();
 
     // Set the new thread.
+    SetCurrentThread(system.Kernel(), next_thread);
     current_thread.store(next_thread);
 
     guard.Unlock();
@@ -804,6 +811,7 @@ void KScheduler::SwitchToCurrent() {
                 }
             }
             auto thread = next_thread ? next_thread : idle_thread;
+            SetCurrentThread(system.Kernel(), thread);
             Common::Fiber::YieldTo(switch_fiber, *thread->GetHostContext());
         } while (!is_switch_pending());
     }
@@ -829,6 +837,7 @@ void KScheduler::Initialize() {
     idle_thread = KThread::Create(system.Kernel());
     ASSERT(KThread::InitializeIdleThread(system, idle_thread, core_id).IsSuccess());
     idle_thread->SetName(fmt::format("IdleThread:{}", core_id));
+    idle_thread->EnableDispatch();
 }
 
 KScopedSchedulerLock::KScopedSchedulerLock(KernelCore& kernel)

src/core/hle/kernel/k_scheduler.h

@@ -48,18 +48,13 @@ public:
     void Reload(KThread* thread);
 
     /// Gets the current running thread
-    [[nodiscard]] KThread* GetCurrentThread() const;
+    [[nodiscard]] KThread* GetSchedulerCurrentThread() const;
 
     /// Gets the idle thread
     [[nodiscard]] KThread* GetIdleThread() const {
         return idle_thread;
     }
 
-    /// Returns true if the scheduler is idle
-    [[nodiscard]] bool IsIdle() const {
-        return GetCurrentThread() == idle_thread;
-    }
-
     /// Gets the timestamp for the last context switch in ticks.
     [[nodiscard]] u64 GetLastContextSwitchTicks() const;
 
@@ -149,10 +144,7 @@ private:
 
     void RotateScheduledQueue(s32 cpu_core_id, s32 priority);
 
-    void Schedule() {
-        ASSERT(GetCurrentThread()->GetDisableDispatchCount() == 1);
-        this->ScheduleImpl();
-    }
+    void Schedule();
 
     /// Switches the CPU's active thread context to that of the specified thread
     void ScheduleImpl();

src/core/hle/kernel/k_thread.cpp

@@ -225,7 +225,7 @@ ResultCode KThread::Initialize(KThreadFunction func, uintptr_t arg, VAddr user_s
     // Setup the stack parameters.
     StackParameters& sp = GetStackParameters();
     sp.cur_thread = this;
-    sp.disable_count = 0;
+    sp.disable_count = 1;
     SetInExceptionHandler();
 
     // Set thread ID.
@@ -382,7 +382,7 @@ void KThread::FinishTermination() {
         for (std::size_t i = 0; i < static_cast<std::size_t>(Core::Hardware::NUM_CPU_CORES); ++i) {
             KThread* core_thread{};
             do {
-                core_thread = kernel.Scheduler(i).GetCurrentThread();
+                core_thread = kernel.Scheduler(i).GetSchedulerCurrentThread();
             } while (core_thread == this);
         }
     }
@@ -631,7 +631,7 @@ ResultCode KThread::SetCoreMask(s32 core_id_, u64 v_affinity_mask) {
             s32 thread_core;
             for (thread_core = 0; thread_core < static_cast<s32>(Core::Hardware::NUM_CPU_CORES);
                  ++thread_core) {
-                if (kernel.Scheduler(thread_core).GetCurrentThread() == this) {
+                if (kernel.Scheduler(thread_core).GetSchedulerCurrentThread() == this) {
                     thread_is_current = true;
                     break;
                 }
@@ -748,6 +748,19 @@ void KThread::Continue() {
     KScheduler::OnThreadStateChanged(kernel, this, old_state);
 }
 
+void KThread::WaitUntilSuspended() {
+    // Make sure we have a suspend requested.
+    ASSERT(IsSuspendRequested());
+
+    // Loop until the thread is not executing on any core.
+    for (std::size_t i = 0; i < static_cast<std::size_t>(Core::Hardware::NUM_CPU_CORES); ++i) {
+        KThread* core_thread{};
+        do {
+            core_thread = kernel.Scheduler(i).GetSchedulerCurrentThread();
+        } while (core_thread == this);
+    }
+}
+
 ResultCode KThread::SetActivity(Svc::ThreadActivity activity) {
     // Lock ourselves.
     KScopedLightLock lk(activity_pause_lock);
@@ -809,7 +822,7 @@ ResultCode KThread::SetActivity(Svc::ThreadActivity activity) {
                 // Check if the thread is currently running.
                 // If it is, we'll need to retry.
                 for (auto i = 0; i < static_cast<s32>(Core::Hardware::NUM_CPU_CORES); ++i) {
-                    if (kernel.Scheduler(i).GetCurrentThread() == this) {
+                    if (kernel.Scheduler(i).GetSchedulerCurrentThread() == this) {
                         thread_is_current = true;
                         break;
                     }
@@ -1014,8 +1027,6 @@ ResultCode KThread::Run() {
         // Set our state and finish.
         SetState(ThreadState::Runnable);
 
-        DisableDispatch();
-
         return ResultSuccess;
     }
 }
@@ -1164,6 +1175,10 @@ std::shared_ptr<Common::Fiber>& KThread::GetHostContext() {
     return host_context;
 }
 
+void SetCurrentThread(KernelCore& kernel, KThread* thread) {
+    kernel.SetCurrentEmuThread(thread);
+}
+
 KThread* GetCurrentThreadPointer(KernelCore& kernel) {
     return kernel.GetCurrentEmuThread();
 }

src/core/hle/kernel/k_thread.h

@@ -106,6 +106,7 @@ enum class StepState : u32 {
     StepPerformed, ///< Thread has stepped, waiting to be scheduled again
 };
 
+void SetCurrentThread(KernelCore& kernel, KThread* thread);
 [[nodiscard]] KThread* GetCurrentThreadPointer(KernelCore& kernel);
 [[nodiscard]] KThread& GetCurrentThread(KernelCore& kernel);
 [[nodiscard]] s32 GetCurrentCoreId(KernelCore& kernel);
@@ -207,6 +208,8 @@ public:
 
     void Continue();
 
+    void WaitUntilSuspended();
+
     constexpr void SetSyncedIndex(s32 index) {
         synced_index = index;
     }

src/core/hle/kernel/kernel.cpp

@@ -254,7 +254,6 @@ struct KernelCore::Impl {
                                                          core_id)
                        .IsSuccess());
             shutdown_threads[core_id]->SetName(fmt::format("SuspendThread:{}", core_id));
-            shutdown_threads[core_id]->DisableDispatch();
         }
     }
 
@@ -332,6 +331,8 @@ struct KernelCore::Impl {
         return is_shutting_down.load(std::memory_order_relaxed);
     }
 
+    static inline thread_local KThread* current_thread{nullptr};
+
     KThread* GetCurrentEmuThread() {
         // If we are shutting down the kernel, none of this is relevant anymore.
         if (IsShuttingDown()) {
@@ -342,7 +343,12 @@ struct KernelCore::Impl {
         if (thread_id >= Core::Hardware::NUM_CPU_CORES) {
             return GetHostDummyThread();
         }
-        return schedulers[thread_id]->GetCurrentThread();
+
+        return current_thread;
+    }
+
+    void SetCurrentEmuThread(KThread* thread) {
+        current_thread = thread;
     }
 
     void DeriveInitialMemoryLayout() {
@@ -1025,6 +1031,10 @@ KThread* KernelCore::GetCurrentEmuThread() const {
     return impl->GetCurrentEmuThread();
 }
 
+void KernelCore::SetCurrentEmuThread(KThread* thread) {
+    impl->SetCurrentEmuThread(thread);
+}
+
 KMemoryManager& KernelCore::MemoryManager() {
     return *impl->memory_manager;
 }
@@ -1079,6 +1089,13 @@ void KernelCore::Suspend(bool suspended) {
 
     for (auto* process : GetProcessList()) {
         process->SetActivity(activity);
+
+        if (should_suspend) {
+            // Wait for execution to stop
+            for (auto* thread : process->GetThreadList()) {
+                thread->WaitUntilSuspended();
+            }
+        }
     }
 }
 

src/core/hle/kernel/kernel.h

@@ -226,6 +226,9 @@ public:
     /// Gets the current host_thread/guest_thread pointer.
     KThread* GetCurrentEmuThread() const;
 
+    /// Sets the current guest_thread pointer.
+    void SetCurrentEmuThread(KThread* thread);
+
     /// Gets the current host_thread handle.
     u32 GetCurrentHostThreadID() const;
 

src/core/hle/kernel/svc.cpp

@@ -327,7 +327,6 @@ static ResultCode SendSyncRequest(Core::System& system, Handle handle) {
 
     LOG_TRACE(Kernel_SVC, "called handle=0x{:08X}({})", handle, session->GetName());
 
-    auto thread = kernel.CurrentScheduler()->GetCurrentThread();
     {
         KScopedSchedulerLock lock(kernel);
 
@@ -337,7 +336,7 @@ static ResultCode SendSyncRequest(Core::System& system, Handle handle) {
         session->SendSyncRequest(&GetCurrentThread(kernel), system.Memory(), system.CoreTiming());
     }
 
-    return thread->GetWaitResult();
+    return GetCurrentThread(kernel).GetWaitResult();
 }
 
 static ResultCode SendSyncRequest32(Core::System& system, Handle handle) {
@@ -624,7 +623,7 @@ static void Break(Core::System& system, u32 reason, u64 info1, u64 info2) {
 
         handle_debug_buffer(info1, info2);
 
-        auto* const current_thread = system.Kernel().CurrentScheduler()->GetCurrentThread();
+        auto* const current_thread = GetCurrentThreadPointer(system.Kernel());
         const auto thread_processor_id = current_thread->GetActiveCore();
         system.ArmInterface(static_cast<std::size_t>(thread_processor_id)).LogBacktrace();
     }
@@ -692,6 +691,9 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, Handle
         // 6.0.0+
         TotalPhysicalMemoryAvailableWithoutSystemResource = 21,
         TotalPhysicalMemoryUsedWithoutSystemResource = 22,
+
+        // Homebrew only
+        MesosphereCurrentProcess = 65001,
     };
 
     const auto info_id_type = static_cast<GetInfoType>(info_id);
@@ -884,7 +886,7 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, Handle
 
         const auto& core_timing = system.CoreTiming();
         const auto& scheduler = *system.Kernel().CurrentScheduler();
-        const auto* const current_thread = scheduler.GetCurrentThread();
+        const auto* const current_thread = GetCurrentThreadPointer(system.Kernel());
         const bool same_thread = current_thread == thread.GetPointerUnsafe();
 
         const u64 prev_ctx_ticks = scheduler.GetLastContextSwitchTicks();
@@ -914,6 +916,27 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, Handle
         *result = system.Kernel().CurrentScheduler()->GetIdleThread()->GetCpuTime();
         return ResultSuccess;
     }
+    case GetInfoType::MesosphereCurrentProcess: {
+        // Verify the input handle is invalid.
+        R_UNLESS(handle == InvalidHandle, ResultInvalidHandle);
+
+        // Verify the sub-type is valid.
+        R_UNLESS(info_sub_id == 0, ResultInvalidCombination);
+
+        // Get the handle table.
+        KProcess* current_process = system.Kernel().CurrentProcess();
+        KHandleTable& handle_table = current_process->GetHandleTable();
+
+        // Get a new handle for the current process.
+        Handle tmp;
+        R_TRY(handle_table.Add(&tmp, current_process));
+
+        // Set the output.
+        *result = tmp;
+
+        // We succeeded.
+        return ResultSuccess;
+    }
     default:
         LOG_ERROR(Kernel_SVC, "Unimplemented svcGetInfo id=0x{:016X}", info_id);
         return ResultInvalidEnumValue;
@@ -1103,7 +1126,7 @@ static ResultCode GetThreadContext(Core::System& system, VAddr out_context, Hand
         if (thread->GetRawState() != ThreadState::Runnable) {
             bool current = false;
             for (auto i = 0; i < static_cast<s32>(Core::Hardware::NUM_CPU_CORES); ++i) {
-                if (thread.GetPointerUnsafe() == kernel.Scheduler(i).GetCurrentThread()) {
+                if (thread.GetPointerUnsafe() == kernel.Scheduler(i).GetSchedulerCurrentThread()) {
                     current = true;
                     break;
                 }
@@ -1851,7 +1874,7 @@ static ResultCode StartThread32(Core::System& system, Handle thread_handle) {
 static void ExitThread(Core::System& system) {
     LOG_DEBUG(Kernel_SVC, "called, pc=0x{:08X}", system.CurrentArmInterface().GetPC());
 
-    auto* const current_thread = system.Kernel().CurrentScheduler()->GetCurrentThread();
+    auto* const current_thread = GetCurrentThreadPointer(system.Kernel());
     system.GlobalSchedulerContext().RemoveThread(current_thread);
     current_thread->Exit();
     system.Kernel().UnregisterInUseObject(current_thread);
@@ -2993,7 +3016,7 @@ void Call(Core::System& system, u32 immediate) {
     auto& kernel = system.Kernel();
     kernel.EnterSVCProfile();
 
-    auto* thread = kernel.CurrentScheduler()->GetCurrentThread();
+    auto* thread = GetCurrentThreadPointer(kernel);
     thread->SetIsCallingSvc();
 
     const FunctionDef* info = system.CurrentProcess()->Is64BitProcess() ? GetSVCInfo64(immediate)

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

@@ -686,7 +686,7 @@ ICommonStateGetter::ICommonStateGetter(Core::System& system_,
         {66, &ICommonStateGetter::SetCpuBoostMode, "SetCpuBoostMode"},
         {67, nullptr, "CancelCpuBoostMode"},
         {68, nullptr, "GetBuiltInDisplayType"},
-        {80, nullptr, "PerformSystemButtonPressingIfInFocus"},
+        {80, &ICommonStateGetter::PerformSystemButtonPressingIfInFocus, "PerformSystemButtonPressingIfInFocus"},
         {90, nullptr, "SetPerformanceConfigurationChangedNotification"},
         {91, nullptr, "GetCurrentPerformanceConfiguration"},
         {100, nullptr, "SetHandlingHomeButtonShortPressedEnabled"},
@@ -826,6 +826,16 @@ void ICommonStateGetter::SetCpuBoostMode(Kernel::HLERequestContext& ctx) {
     apm_sys->SetCpuBoostMode(ctx);
 }
 
+void ICommonStateGetter::PerformSystemButtonPressingIfInFocus(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const auto system_button{rp.PopEnum<SystemButtonType>()};
+
+    LOG_WARNING(Service_AM, "(STUBBED) called, system_button={}", system_button);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(ResultSuccess);
+}
+
 void ICommonStateGetter::SetRequestExitToLibraryAppletAtExecuteNextProgramEnabled(
     Kernel::HLERequestContext& ctx) {
     LOG_WARNING(Service_AM, "(STUBBED) called");

src/core/hle/service/am/am.h

@@ -220,6 +220,18 @@ private:
         Docked = 1,
     };
 
+    // This is nn::am::service::SystemButtonType
+    enum class SystemButtonType {
+        None,
+        HomeButtonShortPressing,
+        HomeButtonLongPressing,
+        PowerButtonShortPressing,
+        PowerButtonLongPressing,
+        ShutdownSystem,
+        CaptureButtonShortPressing,
+        CaptureButtonLongPressing,
+    };
+
     void GetEventHandle(Kernel::HLERequestContext& ctx);
     void ReceiveMessage(Kernel::HLERequestContext& ctx);
     void GetCurrentFocusState(Kernel::HLERequestContext& ctx);
@@ -234,6 +246,7 @@ private:
     void EndVrModeEx(Kernel::HLERequestContext& ctx);
     void GetDefaultDisplayResolution(Kernel::HLERequestContext& ctx);
     void SetCpuBoostMode(Kernel::HLERequestContext& ctx);
+    void PerformSystemButtonPressingIfInFocus(Kernel::HLERequestContext& ctx);
     void SetRequestExitToLibraryAppletAtExecuteNextProgramEnabled(Kernel::HLERequestContext& ctx);
 
     std::shared_ptr<AppletMessageQueue> msg_queue;

src/core/memory.cpp

@@ -67,6 +67,16 @@ struct Memory::Impl {
         return system.DeviceMemory().GetPointer(paddr) + vaddr;
     }
 
+    [[nodiscard]] u8* GetPointerFromDebugMemory(VAddr vaddr) const {
+        const PAddr paddr{current_page_table->backing_addr[vaddr >> PAGE_BITS]};
+
+        if (paddr == 0) {
+            return {};
+        }
+
+        return system.DeviceMemory().GetPointer(paddr) + vaddr;
+    }
+
     u8 Read8(const VAddr addr) {
         return Read<u8>(addr);
     }
@@ -187,6 +197,12 @@ struct Memory::Impl {
                 on_memory(copy_amount, mem_ptr);
                 break;
             }
+            case Common::PageType::DebugMemory: {
+                DEBUG_ASSERT(pointer);
+                u8* const mem_ptr{GetPointerFromDebugMemory(current_vaddr)};
+                on_memory(copy_amount, mem_ptr);
+                break;
+            }
             case Common::PageType::RasterizerCachedMemory: {
                 u8* const host_ptr{GetPointerFromRasterizerCachedMemory(current_vaddr)};
                 on_rasterizer(current_vaddr, copy_amount, host_ptr);
@@ -316,6 +332,58 @@ struct Memory::Impl {
             });
     }
 
+    void MarkRegionDebug(VAddr vaddr, u64 size, bool debug) {
+        if (vaddr == 0) {
+            return;
+        }
+
+        // Iterate over a contiguous CPU address space, marking/unmarking the region.
+        // The region is at a granularity of CPU pages.
+
+        const u64 num_pages = ((vaddr + size - 1) >> PAGE_BITS) - (vaddr >> PAGE_BITS) + 1;
+        for (u64 i = 0; i < num_pages; ++i, vaddr += PAGE_SIZE) {
+            const Common::PageType page_type{
+                current_page_table->pointers[vaddr >> PAGE_BITS].Type()};
+            if (debug) {
+                // Switch page type to debug if now debug
+                switch (page_type) {
+                case Common::PageType::Unmapped:
+                    ASSERT_MSG(false, "Attempted to mark unmapped pages as debug");
+                    break;
+                case Common::PageType::RasterizerCachedMemory:
+                case Common::PageType::DebugMemory:
+                    // Page is already marked.
+                    break;
+                case Common::PageType::Memory:
+                    current_page_table->pointers[vaddr >> PAGE_BITS].Store(
+                        nullptr, Common::PageType::DebugMemory);
+                    break;
+                default:
+                    UNREACHABLE();
+                }
+            } else {
+                // Switch page type to non-debug if now non-debug
+                switch (page_type) {
+                case Common::PageType::Unmapped:
+                    ASSERT_MSG(false, "Attempted to mark unmapped pages as non-debug");
+                    break;
+                case Common::PageType::RasterizerCachedMemory:
+                case Common::PageType::Memory:
+                    // Don't mess with already non-debug or rasterizer memory.
+                    break;
+                case Common::PageType::DebugMemory: {
+                    u8* const pointer{GetPointerFromDebugMemory(vaddr & ~PAGE_MASK)};
+                    current_page_table->pointers[vaddr >> PAGE_BITS].Store(
+                        pointer - (vaddr & ~PAGE_MASK), Common::PageType::Memory);
+                    break;
+                }
+                default:
+                    UNREACHABLE();
+                }
+            }
+        }
+    }
+
     void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
         if (vaddr == 0) {
             return;
@@ -342,6 +410,7 @@ struct Memory::Impl {
                     // It is not necessary for a process to have this region mapped into its address
                     // space, for example, a system module need not have a VRAM mapping.
                     break;
+                case Common::PageType::DebugMemory:
                 case Common::PageType::Memory:
                     current_page_table->pointers[vaddr >> PAGE_BITS].Store(
                         nullptr, Common::PageType::RasterizerCachedMemory);
@@ -360,6 +429,7 @@ struct Memory::Impl {
                     // It is not necessary for a process to have this region mapped into its address
                     // space, for example, a system module need not have a VRAM mapping.
                     break;
+                case Common::PageType::DebugMemory:
                 case Common::PageType::Memory:
                     // There can be more than one GPU region mapped per CPU region, so it's common
                     // that this area is already unmarked as cached.
@@ -460,6 +530,8 @@ struct Memory::Impl {
         case Common::PageType::Memory:
             ASSERT_MSG(false, "Mapped memory page without a pointer @ 0x{:016X}", vaddr);
             return nullptr;
+        case Common::PageType::DebugMemory:
+            return GetPointerFromDebugMemory(vaddr);
         case Common::PageType::RasterizerCachedMemory: {
             u8* const host_ptr{GetPointerFromRasterizerCachedMemory(vaddr)};
             on_rasterizer();
@@ -591,7 +663,8 @@ bool Memory::IsValidVirtualAddress(const VAddr vaddr) const {
         return false;
     }
     const auto [pointer, type] = page_table.pointers[page].PointerType();
-    return pointer != nullptr || type == Common::PageType::RasterizerCachedMemory;
+    return pointer != nullptr || type == Common::PageType::RasterizerCachedMemory ||
+           type == Common::PageType::DebugMemory;
 }
 
 bool Memory::IsValidVirtualAddressRange(VAddr base, u64 size) const {
@@ -707,4 +780,8 @@ void Memory::RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
     impl->RasterizerMarkRegionCached(vaddr, size, cached);
 }
 
+void Memory::MarkRegionDebug(VAddr vaddr, u64 size, bool debug) {
+    impl->MarkRegionDebug(vaddr, size, debug);
+}
+
 } // namespace Core::Memory

src/core/memory.h

@@ -446,6 +446,17 @@ public:
      */
     void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached);
 
+    /**
+     * Marks each page within the specified address range as debug or non-debug.
+     * Debug addresses are not accessible from fastmem pointers.
+     *
+     * @param vaddr The virtual address indicating the start of the address range.
+     * @param size  The size of the address range in bytes.
+     * @param debug Whether or not any pages within the address range should be
+     *              marked as debug or non-debug.
+     */
+    void MarkRegionDebug(VAddr vaddr, u64 size, bool debug);
+
 private:
     Core::System& system;
 

src/video_core/buffer_cache/buffer_base.h

@@ -211,7 +211,7 @@ public:
     void FlushCachedWrites() noexcept {
         flags &= ~BufferFlagBits::CachedWrites;
         const u64 num_words = NumWords();
-        u64* const cached_words = Array<Type::CachedCPU>();
+        const u64* const cached_words = Array<Type::CachedCPU>();
         u64* const untracked_words = Array<Type::Untracked>();
         u64* const cpu_words = Array<Type::CPU>();
         for (u64 word_index = 0; word_index < num_words; ++word_index) {
@@ -219,7 +219,6 @@ public:
             NotifyRasterizer<false>(word_index, untracked_words[word_index], cached_bits);
             untracked_words[word_index] |= cached_bits;
             cpu_words[word_index] |= cached_bits;
-            cached_words[word_index] = 0;
         }
     }
 

src/yuzu/bootmanager.cpp

@@ -29,6 +29,7 @@
 #include "common/scm_rev.h"
 #include "common/settings.h"
 #include "core/core.h"
+#include "core/cpu_manager.h"
 #include "core/frontend/framebuffer_layout.h"
 #include "input_common/drivers/keyboard.h"
 #include "input_common/drivers/mouse.h"
@@ -73,6 +74,8 @@ void EmuThread::run() {
 
     gpu.ReleaseContext();
 
+    system.GetCpuManager().OnGpuReady();
+
     // Holds whether the cpu was running during the last iteration,
     // so that the DebugModeLeft signal can be emitted before the
     // next execution step

src/yuzu_cmd/yuzu.cpp

@@ -21,6 +21,7 @@
 #include "common/string_util.h"
 #include "common/telemetry.h"
 #include "core/core.h"
+#include "core/cpu_manager.h"
 #include "core/crypto/key_manager.h"
 #include "core/file_sys/registered_cache.h"
 #include "core/file_sys/vfs_real.h"
@@ -216,6 +217,7 @@ int main(int argc, char** argv) {
 
     // Core is loaded, start the GPU (makes the GPU contexts current to this thread)
     system.GPU().Start();
+    system.GetCpuManager().OnGpuReady();
 
     if (Settings::values.use_disk_shader_cache.GetValue()) {
         system.Renderer().ReadRasterizer()->LoadDiskResources(