threads: remove low level threads on windows (#4277)

src/core/libraries/kernel/threads/pthread.cpp

@@ -197,7 +197,11 @@ int PS4_SYSV_ABI posix_pthread_detach(PthreadT pthread) {
     return 0;
 }
 
-static void RunThread(void* arg) {
+#ifdef WIN32
+static DWORD RunThread(void* arg) {
+#else
+static void* RunThread(void* arg) {
+#endif
     auto* curthread = static_cast<Pthread*>(arg);
     g_curthread = curthread;
     Common::SetCurrentThreadName(curthread->name.c_str());
@@ -214,6 +218,11 @@ static void RunThread(void* arg) {
     /* Remove thread from tracking */
     DebugState.RemoveCurrentThreadFromGuestList();
     posix_pthread_exit(ret);
+#ifdef WIN32
+    return 0;
+#else
+    return nullptr;
+#endif
 }
 
 int PS4_SYSV_ABI posix_pthread_create_name_np(PthreadT* thread, const PthreadAttrT* attr,

src/core/libraries/kernel/threads/stack.s

@@ -5,6 +5,19 @@
 _runOnAnotherStack:
   pushq %r12
   pushq %r13
+#ifdef WIN32
+  pushq %r14
+  pushq %r15
+
+  # save stack check registers
+  movq %gs:0x08, %r14 # teb->stack_bottom
+  movq %gs:0x10, %r15 # teb->stack_top
+
+  # disable stack checks
+  xorq %rcx, %rcx
+  movq %rcx, %gs:0x08 # teb->stack_bottom = 0
+  movq %rcx, %gs:0x10 # teb->stack_top = 0
+#endif
   movq %rsp, %r12
   movq %rbp, %r13
   movq %rdx, %rsp
@@ -12,6 +25,13 @@ _runOnAnotherStack:
   callq *%rsi
   movq %r13, %rbp
   movq %r12, %rsp
+#ifdef WIN32
+  # restore stack check registers
+  movq %r14, %gs:0x08 # teb->stack_bottom
+  movq %r15, %gs:0x10 # teb->stack_top
+  popq %r15
+  popq %r14
+#endif
   popq %r13
   popq %r12
   ret

src/core/thread.cpp

@@ -19,67 +19,20 @@ namespace Core {
 static constexpr u32 ORBIS_MXCSR = 0x9fc0;
 static constexpr u32 ORBIS_FPUCW = 0x037f;
 
-#ifdef _WIN64
-#define KGDT64_R3_DATA (0x28)
-#define KGDT64_R3_CODE (0x30)
-#define KGDT64_R3_CMTEB (0x50)
-#define RPL_MASK (0x03)
-#define EFLAGS_INTERRUPT_MASK (0x200)
-
-void InitializeTeb(INITIAL_TEB* teb, void* stackaddr, size_t stacksize) {
-    teb->StackBase = (void*)((u64)stackaddr + stacksize);
-    teb->StackLimit = nullptr;
-    teb->StackAllocationBase = stackaddr;
-}
-
-void InitializeContext(CONTEXT* ctx, ThreadFunc func, void* arg, void* stackaddr,
-                       size_t stacksize) {
-    /* Note: The stack has to be reversed */
-    ctx->Rsp = (u64)stackaddr + stacksize;
-    ctx->Rbp = (u64)stackaddr + stacksize;
-    ctx->Rcx = (u64)arg;
-    ctx->Rip = (u64)func;
-
-    ctx->SegGs = KGDT64_R3_DATA | RPL_MASK;
-    ctx->SegEs = KGDT64_R3_DATA | RPL_MASK;
-    ctx->SegDs = KGDT64_R3_DATA | RPL_MASK;
-    ctx->SegCs = KGDT64_R3_CODE | RPL_MASK;
-    ctx->SegSs = KGDT64_R3_DATA | RPL_MASK;
-    ctx->SegFs = KGDT64_R3_CMTEB | RPL_MASK;
-
-    ctx->EFlags = 0x3000 | EFLAGS_INTERRUPT_MASK;
-
-    ctx->ContextFlags =
-        CONTEXT_CONTROL | CONTEXT_INTEGER | CONTEXT_SEGMENTS | CONTEXT_FLOATING_POINT;
-}
-#endif
-
 NativeThread::NativeThread() : native_handle{0} {}
 
 NativeThread::~NativeThread() {}
 
 int NativeThread::Create(ThreadFunc func, void* arg) {
-    constexpr size_t stacksize = 8_KB;
-    init_stack_ptr = malloc(stacksize);
 #ifndef _WIN64
     pthread_t* pthr = reinterpret_cast<pthread_t*>(&native_handle);
-    pthread_attr_t pattr;
-    pthread_attr_init(&pattr);
-    pthread_attr_setstack(&pattr, init_stack_ptr, stacksize);
-    return pthread_create(pthr, &pattr, (PthreadFunc)func, arg);
+    return pthread_create(pthr, nullptr, func, arg);
 #else
-    CLIENT_ID clientId{};
-    INITIAL_TEB teb{};
-    CONTEXT ctx{};
-
-    clientId.UniqueProcess = GetCurrentProcess();
-    clientId.UniqueThread = GetCurrentThread();
-
-    InitializeTeb(&teb, init_stack_ptr, stacksize);
-    InitializeContext(&ctx, func, arg, init_stack_ptr, stacksize);
-
-    return NtCreateThread(&native_handle, THREAD_ALL_ACCESS, nullptr, GetCurrentProcess(),
-                          &clientId, &ctx, &teb, false);
+    native_handle = CreateThread(nullptr, 0, func, arg, 0, nullptr);
+    if (native_handle == nullptr) {
+        return GetLastError();
+    }
+    return 0;
 #endif
 }
 
@@ -91,28 +44,8 @@ void NativeThread::Exit() {
     tid = 0;
 
 #ifdef _WIN64
-    NtClose(native_handle);
     native_handle = nullptr;
-
-    /* The Windows kernel will free the stack
-       given at thread creation via INITIAL_TEB
-       (StackAllocationBase) upon thread termination.
-
-       In earlier Windows versions (NT4 to Windows Server 2003),
-       you could get around this via disabling FreeStackOnTermination
-       on the TEB. This has been removed since then.
-
-       To avoid this, we must forcefully set the TEB
-       deallocation stack pointer to NULL so ZwFreeVirtualMemory fails
-       in the kernel and our stack is not freed.
-     */
-    auto* teb = reinterpret_cast<TEB*>(NtCurrentTeb());
-    teb->DeallocationStack = nullptr;
-
-    if (init_stack_ptr != nullptr) {
-        free(init_stack_ptr);
-    }
-    NtTerminateThread(nullptr, 0);
+    ExitThread(0);
 #else
     // Disable and free the signal stack.
     constexpr stack_t sig_stack = {
@@ -124,10 +57,6 @@ void NativeThread::Exit() {
         free(sig_stack_ptr);
         sig_stack_ptr = nullptr;
     }
-
-    if (init_stack_ptr != nullptr) {
-        free(init_stack_ptr);
-    }
     pthread_exit(nullptr);
 #endif
 }

src/core/thread.h

@@ -10,9 +10,11 @@ struct PthreadAttr;
 } // namespace Libraries::Kernel
 
 namespace Core {
-
-using ThreadFunc = void (*)(void*);
-using PthreadFunc = void* (*)(void*);
+#ifdef WIN32
+using ThreadFunc = DWORD (*)(void*);
+#else
+using ThreadFunc = void* (*)(void*);
+#endif
 
 class NativeThread {
 public:
@@ -39,7 +41,6 @@ private:
     uintptr_t native_handle;
     void* sig_stack_ptr = nullptr;
 #endif
-    void* init_stack_ptr = nullptr;
     u64 tid;
 };