[x64] Fix up unwind info for emitted functions.

- [x64] Track size of code within emitted functions (prolog, body, epilog).
- [x64] Don't use hardcoded prolog size in generated unwind info.
- [x64] Update URLs to MSDN documentation on UNWIND_INFO/UNWIND_CODE.

src/xenia/cpu/backend/x64/x64_backend.cc

@@ -407,13 +407,25 @@ HostToGuestThunk X64ThunkEmitter::EmitHostToGuestThunk() {
   // rdx = arg0 (context)
   // r8 = arg1 (guest return address)
 
+  struct _code_offsets {
+    size_t prolog;
+    size_t body;
+    size_t epilog;
+    size_t tail;
+  } code_offsets = {};
+
   const size_t stack_size = StackLayout::THUNK_STACK_SIZE;
+
+  code_offsets.prolog = getSize();
+
   // rsp + 0 = return address
   mov(qword[rsp + 8 * 3], r8);
   mov(qword[rsp + 8 * 2], rdx);
   mov(qword[rsp + 8 * 1], rcx);
   sub(rsp, stack_size);
 
+  code_offsets.body = getSize();
+
   // Save nonvolatile registers.
   EmitSaveNonvolatileRegs();
 
@@ -424,13 +436,26 @@ HostToGuestThunk X64ThunkEmitter::EmitHostToGuestThunk() {
 
   EmitLoadNonvolatileRegs();
 
+  code_offsets.epilog = getSize();
+
   add(rsp, stack_size);
   mov(rcx, qword[rsp + 8 * 1]);
   mov(rdx, qword[rsp + 8 * 2]);
   mov(r8, qword[rsp + 8 * 3]);
   ret();
 
-  void* fn = Emplace(stack_size);
+  code_offsets.tail = getSize();
+
+  assert_zero(code_offsets.prolog);
+  EmitFunctionInfo func_info = {};
+  func_info.code_size.total = getSize();
+  func_info.code_size.prolog = code_offsets.body - code_offsets.prolog;
+  func_info.code_size.body = code_offsets.epilog - code_offsets.body;
+  func_info.code_size.epilog = code_offsets.tail - code_offsets.epilog;
+  func_info.code_size.tail = getSize() - code_offsets.tail;
+  func_info.stack_size = stack_size;
+
+  void* fn = Emplace(func_info);
   return (HostToGuestThunk)fn;
 }
 
@@ -440,10 +465,22 @@ GuestToHostThunk X64ThunkEmitter::EmitGuestToHostThunk() {
   // r8  = arg1
   // r9  = arg2
 
+  struct _code_offsets {
+    size_t prolog;
+    size_t body;
+    size_t epilog;
+    size_t tail;
+  } code_offsets = {};
+
   const size_t stack_size = StackLayout::THUNK_STACK_SIZE;
+
+  code_offsets.prolog = getSize();
+
   // rsp + 0 = return address
   sub(rsp, stack_size);
 
+  code_offsets.body = getSize();
+
   // Save off volatile registers.
   EmitSaveVolatileRegs();
 
@@ -453,10 +490,23 @@ GuestToHostThunk X64ThunkEmitter::EmitGuestToHostThunk() {
 
   EmitLoadVolatileRegs();
 
+  code_offsets.epilog = getSize();
+
   add(rsp, stack_size);
   ret();
 
-  void* fn = Emplace(stack_size);
+  code_offsets.tail = getSize();
+
+  assert_zero(code_offsets.prolog);
+  EmitFunctionInfo func_info = {};
+  func_info.code_size.total = getSize();
+  func_info.code_size.prolog = code_offsets.body - code_offsets.prolog;
+  func_info.code_size.body = code_offsets.epilog - code_offsets.body;
+  func_info.code_size.epilog = code_offsets.tail - code_offsets.epilog;
+  func_info.code_size.tail = getSize() - code_offsets.tail;
+  func_info.stack_size = stack_size;
+
+  void* fn = Emplace(func_info);
   return (GuestToHostThunk)fn;
 }
 
@@ -466,11 +516,23 @@ extern "C" uint64_t ResolveFunction(void* raw_context, uint32_t target_address);
 ResolveFunctionThunk X64ThunkEmitter::EmitResolveFunctionThunk() {
   // ebx = target PPC address
   // rcx = context
+
+  struct _code_offsets {
+    size_t prolog;
+    size_t body;
+    size_t epilog;
+    size_t tail;
+  } code_offsets = {};
+
   const size_t stack_size = StackLayout::THUNK_STACK_SIZE;
 
+  code_offsets.prolog = getSize();
+
   // rsp + 0 = return address
   sub(rsp, stack_size);
 
+  code_offsets.body = getSize();
+
   // Save volatile registers
   EmitSaveVolatileRegs();
 
@@ -481,10 +543,23 @@ ResolveFunctionThunk X64ThunkEmitter::EmitResolveFunctionThunk() {
 
   EmitLoadVolatileRegs();
 
+  code_offsets.epilog = getSize();
+
   add(rsp, stack_size);
   jmp(rax);
 
-  void* fn = Emplace(stack_size);
+  code_offsets.tail = getSize();
+
+  assert_zero(code_offsets.prolog);
+  EmitFunctionInfo func_info = {};
+  func_info.code_size.total = getSize();
+  func_info.code_size.prolog = code_offsets.body - code_offsets.prolog;
+  func_info.code_size.body = code_offsets.epilog - code_offsets.body;
+  func_info.code_size.epilog = code_offsets.tail - code_offsets.epilog;
+  func_info.code_size.tail = getSize() - code_offsets.tail;
+  func_info.stack_size = stack_size;
+
+  void* fn = Emplace(func_info);
   return (ResolveFunctionThunk)fn;
 }
 

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

@@ -125,15 +125,14 @@ void X64CodeCache::CommitExecutableRange(uint32_t guest_low,
 }
 
 void* X64CodeCache::PlaceHostCode(uint32_t guest_address, void* machine_code,
-                                  size_t code_size, size_t stack_size) {
+                                  const EmitFunctionInfo& func_info) {
   // Same for now. We may use different pools or whatnot later on, like when
   // we only want to place guest code in a serialized cache on disk.
-  return PlaceGuestCode(guest_address, machine_code, code_size, stack_size,
+  return PlaceGuestCode(guest_address, machine_code, func_info, nullptr);
-                        nullptr);
 }
 
 void* X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
-                                   size_t code_size, size_t stack_size,
+                                   const EmitFunctionInfo& func_info,
                                    GuestFunction* function_info) {
   // Hold a lock while we bump the pointers up. This is important as the
   // unwind table requires entries AND code to be sorted in order.
@@ -149,7 +148,7 @@ void* X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
     // Reserve code.
     // Always move the code to land on 16b alignment.
     code_address = generated_code_base_ + generated_code_offset_;
-    generated_code_offset_ += xe::round_up(code_size, 16);
+    generated_code_offset_ += xe::round_up(func_info.code_size.total, 16);
 
     // Reserve unwind info.
     // We go on the high size of the unwind info as we don't know how big we
@@ -187,15 +186,17 @@ void* X64CodeCache::PlaceGuestCode(uint32_t guest_address, void* machine_code,
         old_commit_mark, new_commit_mark));
 
     // Copy code.
-    std::memcpy(code_address, machine_code, code_size);
+    std::memcpy(code_address, machine_code, func_info.code_size.total);
 
     // Fill unused slots with 0xCC
     std::memset(
-        code_address + code_size, 0xCC,
+        code_address + func_info.code_size.total, 0xCC,
-        xe::round_up(code_size + unwind_reservation.data_size, 16) - code_size);
+        xe::round_up(func_info.code_size.total + unwind_reservation.data_size,
+                     16) -
+            func_info.code_size.total);
 
     // Notify subclasses of placed code.
-    PlaceCode(guest_address, machine_code, code_size, stack_size, code_address,
+    PlaceCode(guest_address, machine_code, func_info, code_address,
               unwind_reservation);
   }
 

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

@@ -25,6 +25,17 @@ namespace cpu {
 namespace backend {
 namespace x64 {
 
+struct EmitFunctionInfo {
+  struct _code_size {
+    size_t prolog;
+    size_t body;
+    size_t epilog;
+    size_t tail;
+    size_t total;
+  } code_size;
+  size_t stack_size;
+};
+
 class X64CodeCache : public CodeCache {
  public:
   ~X64CodeCache() override;
@@ -48,9 +59,9 @@ class X64CodeCache : public CodeCache {
   void CommitExecutableRange(uint32_t guest_low, uint32_t guest_high);
 
   void* PlaceHostCode(uint32_t guest_address, void* machine_code,
-                      size_t code_size, size_t stack_size);
+                      const EmitFunctionInfo& func_info);
   void* PlaceGuestCode(uint32_t guest_address, void* machine_code,
-                       size_t code_size, size_t stack_size,
+                       const EmitFunctionInfo& func_info,
                        GuestFunction* function_info);
   uint32_t PlaceData(const void* data, size_t length);
 
@@ -84,8 +95,7 @@ class X64CodeCache : public CodeCache {
     return UnwindReservation();
   }
   virtual void PlaceCode(uint32_t guest_address, void* machine_code,
-                         size_t code_size, size_t stack_size,
+                         const EmitFunctionInfo& func_info, void* code_address,
-                         void* code_address,
                          UnwindReservation unwind_reservation) {}
 
   std::wstring file_name_;

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

@@ -112,13 +112,13 @@ class Win32X64CodeCache : public X64CodeCache {
 
  private:
   UnwindReservation RequestUnwindReservation(uint8_t* entry_address) override;
-  void PlaceCode(uint32_t guest_address, void* machine_code, size_t code_size,
+  void PlaceCode(uint32_t guest_address, void* machine_code,
-                 size_t stack_size, void* code_address,
+                 const EmitFunctionInfo& func_info, void* code_address,
                  UnwindReservation unwind_reservation) override;
 
   void InitializeUnwindEntry(uint8_t* unwind_entry_address,
                              size_t unwind_table_slot, void* code_address,
-                             size_t code_size, size_t stack_size);
+                             const EmitFunctionInfo& func_info);
 
   // Growable function table system handle.
   void* unwind_table_handle_ = nullptr;
@@ -222,13 +222,12 @@ Win32X64CodeCache::RequestUnwindReservation(uint8_t* entry_address) {
 }
 
 void Win32X64CodeCache::PlaceCode(uint32_t guest_address, void* machine_code,
-                                  size_t code_size, size_t stack_size,
+                                  const EmitFunctionInfo& func_info,
                                   void* code_address,
                                   UnwindReservation unwind_reservation) {
   // Add unwind info.
   InitializeUnwindEntry(unwind_reservation.entry_address,
-                        unwind_reservation.table_slot, code_address, code_size,
+                        unwind_reservation.table_slot, code_address, func_info);
-                        stack_size);
 
   if (supports_growable_table_) {
     // Notify that the unwind table has grown.
@@ -237,29 +236,29 @@ void Win32X64CodeCache::PlaceCode(uint32_t guest_address, void* machine_code,
   }
 
   // This isn't needed on x64 (probably), but is convention.
-  FlushInstructionCache(GetCurrentProcess(), code_address, code_size);
+  FlushInstructionCache(GetCurrentProcess(), code_address,
+                        func_info.code_size.total);
 }
 
-void Win32X64CodeCache::InitializeUnwindEntry(uint8_t* unwind_entry_address,
+void Win32X64CodeCache::InitializeUnwindEntry(
-                                              size_t unwind_table_slot,
+    uint8_t* unwind_entry_address, size_t unwind_table_slot, void* code_address,
-                                              void* code_address,
+    const EmitFunctionInfo& func_info) {
-                                              size_t code_size,
-                                              size_t stack_size) {
   auto unwind_info = reinterpret_cast<UNWIND_INFO*>(unwind_entry_address);
   UNWIND_CODE* unwind_code = nullptr;
 
-  if (!stack_size) {
+  assert_true(func_info.code_size.prolog < 256);  // needs to fit into a uint8_t
-    // https://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
+  auto prolog_size = static_cast<uint8_t>(func_info.code_size.prolog);
+
+  if (!func_info.stack_size) {
+    // https://docs.microsoft.com/en-us/cpp/build/exception-handling-x64#struct-unwind_info
     unwind_info->Version = 1;
     unwind_info->Flags = 0;
-    unwind_info->SizeOfProlog = 0;
+    unwind_info->SizeOfProlog = prolog_size;
     unwind_info->CountOfCodes = 0;
     unwind_info->FrameRegister = 0;
     unwind_info->FrameOffset = 0;
-  } else if (stack_size <= 128) {
+  } else if (func_info.stack_size <= 128) {
-    uint8_t prolog_size = 4;
+    // https://docs.microsoft.com/en-us/cpp/build/exception-handling-x64#struct-unwind_info
-
-    // https://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
     unwind_info->Version = 1;
     unwind_info->Flags = 0;
     unwind_info->SizeOfProlog = prolog_size;
@@ -267,17 +266,16 @@ void Win32X64CodeCache::InitializeUnwindEntry(uint8_t* unwind_entry_address,
     unwind_info->FrameRegister = 0;
     unwind_info->FrameOffset = 0;
 
-    // https://msdn.microsoft.com/en-us/library/ck9asaa9.aspx
+    // https://docs.microsoft.com/en-us/cpp/build/exception-handling-x64#struct-unwind_code
     unwind_code = &unwind_info->UnwindCode[unwind_info->CountOfCodes++];
     unwind_code->CodeOffset =
         14;  // end of instruction + 1 == offset of next instruction
     unwind_code->UnwindOp = UWOP_ALLOC_SMALL;
-    unwind_code->OpInfo = stack_size / 8 - 1;
+    unwind_code->OpInfo = func_info.stack_size / 8 - 1;
   } else {
     // TODO(benvanik): take as parameters?
-    uint8_t prolog_size = 7;
 
-    // https://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
+    // https://docs.microsoft.com/en-us/cpp/build/exception-handling-x64#struct-unwind_info
     unwind_info->Version = 1;
     unwind_info->Flags = 0;
     unwind_info->SizeOfProlog = prolog_size;
@@ -285,16 +283,16 @@ void Win32X64CodeCache::InitializeUnwindEntry(uint8_t* unwind_entry_address,
     unwind_info->FrameRegister = 0;
     unwind_info->FrameOffset = 0;
 
-    // https://msdn.microsoft.com/en-us/library/ck9asaa9.aspx
+    // https://docs.microsoft.com/en-us/cpp/build/exception-handling-x64#struct-unwind_code
     unwind_code = &unwind_info->UnwindCode[unwind_info->CountOfCodes++];
     unwind_code->CodeOffset =
         7;  // end of instruction + 1 == offset of next instruction
     unwind_code->UnwindOp = UWOP_ALLOC_LARGE;
     unwind_code->OpInfo = 0;  // One slot for size
 
-    assert_true((stack_size / 8) < 65536u);
+    assert_true((func_info.stack_size / 8) < 65536u);
     unwind_code = &unwind_info->UnwindCode[unwind_info->CountOfCodes++];
-    unwind_code->FrameOffset = (USHORT)(stack_size) / 8;
+    unwind_code->FrameOffset = (USHORT)(func_info.stack_size) / 8;
   }
 
   if (unwind_info->CountOfCodes % 1) {
@@ -307,7 +305,8 @@ void Win32X64CodeCache::InitializeUnwindEntry(uint8_t* unwind_entry_address,
   auto& fn_entry = unwind_table_[unwind_table_slot];
   fn_entry.BeginAddress =
       (DWORD)(reinterpret_cast<uint8_t*>(code_address) - generated_code_base_);
-  fn_entry.EndAddress = (DWORD)(fn_entry.BeginAddress + code_size);
+  fn_entry.EndAddress =
+      (DWORD)(fn_entry.BeginAddress + func_info.code_size.total);
   fn_entry.UnwindData = (DWORD)(unwind_entry_address - generated_code_base_);
 }
 

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

@@ -102,14 +102,14 @@ bool X64Emitter::Emit(GuestFunction* function, HIRBuilder* builder,
   source_map_arena_.Reset();
 
   // Fill the generator with code.
-  size_t stack_size = 0;
+  EmitFunctionInfo func_info = {};
-  if (!Emit(builder, &stack_size)) {
+  if (!Emit(builder, func_info)) {
     return false;
   }
 
   // Copy the final code to the cache and relocate it.
   *out_code_size = getSize();
-  *out_code_address = Emplace(stack_size, function);
+  *out_code_address = Emplace(func_info, function);
 
   // Stash source map.
   source_map_arena_.CloneContents(out_source_map);
@@ -117,18 +117,20 @@ bool X64Emitter::Emit(GuestFunction* function, HIRBuilder* builder,
   return true;
 }
 
-void* X64Emitter::Emplace(size_t stack_size, GuestFunction* function) {
+void* X64Emitter::Emplace(const EmitFunctionInfo& func_info,
+                          GuestFunction* function) {
   // To avoid changing xbyak, we do a switcharoo here.
   // top_ points to the Xbyak buffer, and since we are in AutoGrow mode
   // it has pending relocations. We copy the top_ to our buffer, swap the
   // pointer, relocate, then return the original scratch pointer for use.
   uint8_t* old_address = top_;
   void* new_address;
+  assert_true(func_info.code_size.total == size_);
   if (function) {
-    new_address = code_cache_->PlaceGuestCode(function->address(), top_, size_,
+    new_address = code_cache_->PlaceGuestCode(function->address(), top_,
-                                              stack_size, function);
+                                              func_info, function);
   } else {
-    new_address = code_cache_->PlaceHostCode(0, top_, size_, stack_size);
+    new_address = code_cache_->PlaceHostCode(0, top_, func_info);
   }
   top_ = reinterpret_cast<uint8_t*>(new_address);
   ready();
@@ -137,7 +139,7 @@ void* X64Emitter::Emplace(size_t stack_size, GuestFunction* function) {
   return new_address;
 }
 
-bool X64Emitter::Emit(HIRBuilder* builder, size_t* out_stack_size) {
+bool X64Emitter::Emit(HIRBuilder* builder, EmitFunctionInfo& func_info) {
   Xbyak::Label epilog_label;
   epilog_label_ = &epilog_label;
 
@@ -159,6 +161,15 @@ bool X64Emitter::Emit(HIRBuilder* builder, size_t* out_stack_size) {
   stack_offset -= StackLayout::GUEST_STACK_SIZE;
   stack_offset = xe::align(stack_offset, static_cast<size_t>(16));
 
+  struct _code_offsets {
+    size_t prolog;
+    size_t body;
+    size_t epilog;
+    size_t tail;
+  } code_offsets = {};
+
+  code_offsets.prolog = getSize();
+
   // Function prolog.
   // Must be 16b aligned.
   // Windows is very strict about the form of this and the epilog:
@@ -168,7 +179,7 @@ bool X64Emitter::Emit(HIRBuilder* builder, size_t* out_stack_size) {
   //     Adding or changing anything here must be matched!
   const size_t stack_size = StackLayout::GUEST_STACK_SIZE + stack_offset;
   assert_true((stack_size + 8) % 16 == 0);
-  *out_stack_size = stack_size;
+  func_info.stack_size = stack_size;
   stack_size_ = stack_size;
 
   sub(rsp, (uint32_t)stack_size);
@@ -208,6 +219,8 @@ bool X64Emitter::Emit(HIRBuilder* builder, size_t* out_stack_size) {
   mov(GetMembaseReg(),
       qword[GetContextReg() + offsetof(ppc::PPCContext, virtual_membase)]);
 
+  code_offsets.body = getSize();
+
   // Body.
   auto block = builder->first_block();
   while (block) {
@@ -236,6 +249,8 @@ bool X64Emitter::Emit(HIRBuilder* builder, size_t* out_stack_size) {
     block = block->next;
   }
 
+  code_offsets.epilog = getSize();
+
   // Function epilog.
   L(epilog_label);
   epilog_label_ = nullptr;
@@ -244,6 +259,8 @@ bool X64Emitter::Emit(HIRBuilder* builder, size_t* out_stack_size) {
   add(rsp, (uint32_t)stack_size);
   ret();
 
+  code_offsets.tail = getSize();
+
   if (cvars::emit_source_annotations) {
     nop();
     nop();
@@ -252,6 +269,13 @@ bool X64Emitter::Emit(HIRBuilder* builder, size_t* out_stack_size) {
     nop();
   }
 
+  assert_zero(code_offsets.prolog);
+  func_info.code_size.total = getSize();
+  func_info.code_size.prolog = code_offsets.body - code_offsets.prolog;
+  func_info.code_size.body = code_offsets.epilog - code_offsets.body;
+  func_info.code_size.epilog = code_offsets.tail - code_offsets.epilog;
+  func_info.code_size.tail = getSize() - code_offsets.tail;
+
   return true;
 }
 

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

@@ -39,6 +39,8 @@ namespace x64 {
 class X64Backend;
 class X64CodeCache;
 
+struct EmitFunctionInfo;
+
 enum RegisterFlags {
   REG_DEST = (1 << 0),
   REG_ABCD = (1 << 1),
@@ -222,8 +224,9 @@ class X64Emitter : public Xbyak::CodeGenerator {
   size_t stack_size() const { return stack_size_; }
 
  protected:
-  void* Emplace(size_t stack_size, GuestFunction* function = nullptr);
+  void* Emplace(const EmitFunctionInfo& func_info,
-  bool Emit(hir::HIRBuilder* builder, size_t* out_stack_size);
+                GuestFunction* function = nullptr);
+  bool Emit(hir::HIRBuilder* builder, EmitFunctionInfo& func_info);
   void EmitGetCurrentThreadId();
   void EmitTraceUserCallReturn();