osx: JIT support

This PR;
- Enables JIT on macOS
- Fixes / improvements to xplat exception handling

Author: obastemur

CMakeLists.txt

@@ -294,8 +294,7 @@ if(CLR_CMAKE_PLATFORM_XPLAT)
    add_definitions(-DFEATURE_PAL)
 endif(CLR_CMAKE_PLATFORM_XPLAT)
 
-if(NO_JIT_SH
-   OR CLR_CMAKE_PLATFORM_DARWIN)  # TODO: JIT for OSX
+if(NO_JIT_SH)
     unset(NO_JIT_SH CACHE)      # don't cache
     unset(BuildJIT CACHE)       # also clear it just in case
     add_definitions(-DDISABLE_JIT=1)

lib/Backend/EhFrame.cpp

@@ -51,7 +51,7 @@ ubyte GetDwarfRegNum(ubyte regNum)
     return DWARF_RegNum[regNum];
 }
 
-// Enocde into ULEB128 (Unsigned Little Endian Base 128)
+// Encode into ULEB128 (Unsigned Little Endian Base 128)
 BYTE* EmitLEB128(BYTE* pc, unsigned value)
 {
     do

lib/Backend/Encoder.cpp

@@ -868,7 +868,7 @@ void Encoder::ValidateCRCOnFinalBuffer(_In_reads_bytes_(finalCodeSize) BYTE * fi
 
                     currentEndAddress = finalBufferRelocTuplePtr;
                     crcSizeToCompute = currentEndAddress - currentStartAddress;
-                    
+
                     Assert(currentEndAddress >= currentStartAddress);
 
                     finalBufferCRC = CalculateCRC(finalBufferCRC, crcSizeToCompute, currentStartAddress);
@@ -910,7 +910,7 @@ void Encoder::EnsureRelocEntryIntegrity(size_t newBufferStartAddress, size_t cod
 {
     size_t targetBrAddress = 0;
     size_t newBufferEndAddress = newBufferStartAddress + codeSize;
-    
+
     //Handle Dictionary addresses here - The target address will be in the dictionary.
     if (relocAddress < oldBufferAddress || relocAddress >= (oldBufferAddress + codeSize))
     {
@@ -939,6 +939,7 @@ void Encoder::EnsureRelocEntryIntegrity(size_t newBufferStartAddress, size_t cod
 
 uint Encoder::CalculateCRC(uint bufferCRC, size_t data)
 {
+#if defined(_WIN32) || defined(__SSE4_2__)
 #if defined(_M_IX86)
     if (AutoSystemInfo::Data.SSE4_2Available())
     {
@@ -950,6 +951,7 @@ uint Encoder::CalculateCRC(uint bufferCRC, size_t data)
         //CRC32 always returns a 32-bit result
         return (uint)_mm_crc32_u64(bufferCRC, data);
     }
+#endif
 #endif
 
     return CalculateCRC32(bufferCRC, data);
@@ -1153,7 +1155,7 @@ Encoder::ShortenBranchesAndLabelAlign(BYTE **codeStart, ptrdiff_t *codeSize, uin
     // Next, we re-write the code to shorten the BRs and adjust relocList offsets to point to new buffer.
     // We also write NOPs for aligned loops.
     BYTE* tmpBuffer = AnewArray(m_tempAlloc, BYTE, newCodeSize);
-    
+
     uint srcBufferCrc = *pShortenedBufferCRC;   //This has the intial Random CRC seed to start with.
 
     // start copying to new buffer
@@ -1257,7 +1259,7 @@ Encoder::ShortenBranchesAndLabelAlign(BYTE **codeStart, ptrdiff_t *codeSize, uin
             AssertMsg((((uint32)(label->GetPC() - buffStart)) & 0xf) == 0, "Misaligned Label");
 
             to = reloc.getLabelOrigPC() - 1;
-            
+
             CopyPartialBufferAndCalculateCRC(&dst_p, dst_size, from, to, pShortenedBufferCRC);
             srcBufferCrc = CalculateCRC(srcBufferCrc, to - from + 1, from);
 
@@ -1280,7 +1282,7 @@ Encoder::ShortenBranchesAndLabelAlign(BYTE **codeStart, ptrdiff_t *codeSize, uin
         }
     }
     // copy last chunk
-    //Exclude jumpTable content from CRC calculation. 
+    //Exclude jumpTable content from CRC calculation.
     //Though jumpTable is not part of the encoded bytes, codeSize has jumpTableSize included in it.
     CopyPartialBufferAndCalculateCRC(&dst_p, dst_size, from, buffStart + *codeSize - 1, pShortenedBufferCRC, jumpTableSize);
     srcBufferCrc = CalculateCRC(srcBufferCrc, buffStart + *codeSize - from - jumpTableSize, from);

lib/Backend/PDataManager.cpp

@@ -55,19 +55,22 @@ void PDataManager::UnregisterPdata(RUNTIME_FUNCTION* pdata)
 }
 
 #else  // !_WIN32
+
 // ----------------------------------------------------------------------------
 //  !_WIN32 x64 unwind uses .eh_frame
 // ----------------------------------------------------------------------------
 
-void PDataManager::RegisterPdata(RUNTIME_FUNCTION* pdataStart, _In_ const ULONG_PTR functionStart, _In_ const ULONG_PTR functionEnd, _Out_ PVOID* pdataTable, ULONG entryCount, ULONG maxEntryCount)
+void PDataManager::RegisterPdata(RUNTIME_FUNCTION* pdataStart,
+    _In_ const ULONG_PTR functionStart, _In_ const ULONG_PTR functionEnd,
+    _Out_ PVOID* pdataTable, ULONG entryCount, ULONG maxEntryCount)
 {
-    __register_frame(pdataStart);
+    __REGISTER_FRAME(pdataStart);
     *pdataTable = pdataStart;
 }
 
 void PDataManager::UnregisterPdata(RUNTIME_FUNCTION* pdata)
 {
-    __deregister_frame(pdata);
+    __DEREGISTER_FRAME(pdata);
 }
 
 #endif  // !_WIN32

lib/Backend/amd64/LinearScanMdA.S

@@ -7,13 +7,6 @@
 #include "unixasmmacros.inc"
 
 
-// BailOutRecord::BailOut(BailOutRecord const * bailOutRecord)
-// .extern _ZN13BailOutRecord7BailOutEPKS_
-
-// BranchBailOutRecord::BailOut(BranchBailOutRecord const * bailOutRecord, BOOL cond)
-// .extern _ZN19BranchBailOutRecord7BailOutEPKS_i
-
-
 //------------------------------------------------------------------------------
 // LinearScanMD::SaveAllRegisters(BailOutRecord *const bailOutRecord)
 
@@ -106,7 +99,7 @@ NESTED_ENTRY _ZN12LinearScanMD26SaveAllRegistersAndBailOutEP13BailOutRecord, _TE
     add rsp, 28h        // deallocate stack space
     .cfi_adjust_cfa_offset -0x28
 
-    jmp _ZN13BailOutRecord7BailOutEPKS_
+    jmp C_FUNC(_ZN13BailOutRecord7BailOutEPKS_)
 
 NESTED_END _ZN12LinearScanMD26SaveAllRegistersAndBailOutEP13BailOutRecord, _TEXT
 
@@ -132,6 +125,6 @@ NESTED_ENTRY _ZN12LinearScanMD32SaveAllRegistersAndBranchBailOutEP19BranchBailOu
     add rsp, 28h        // deallocate stack space
     .cfi_adjust_cfa_offset -0x28
 
-    jmp _ZN19BranchBailOutRecord7BailOutEPKS_i
+    jmp C_FUNC(_ZN19BranchBailOutRecord7BailOutEPKS_i)
 
 NESTED_END _ZN12LinearScanMD32SaveAllRegistersAndBranchBailOutEP19BranchBailOutRecordi, _TEXT

lib/Backend/amd64/Thunks.S

@@ -72,7 +72,7 @@ NESTED_ENTRY _ZN19NativeCodeGenerator22CheckAsmJsCodeGenThunkEPN2Js16RecyclableO
         movaps xmmword ptr [rsp + 20h], xmm2
         movaps xmmword ptr [rsp + 30h], xmm3
 
-        call _ZN19NativeCodeGenerator17CheckAsmJsCodeGenEPN2Js14ScriptFunctionE
+        call C_FUNC(_ZN19NativeCodeGenerator17CheckAsmJsCodeGenEPN2Js14ScriptFunctionE)
 
         // restore potential floating point arguments from stack
         movaps xmm0, xmmword ptr [rsp + 00h]

lib/Common/Memory/amd64/XDataAllocator.cpp

@@ -12,6 +12,10 @@ CompileAssert(false)
 #include "XDataAllocator.h"
 #include "Core/DelayLoadLibrary.h"
 
+#ifndef _WIN32
+#include "PlatformAgnostic/AssemblyCommon.h" // __REGISTER_FRAME / __DEREGISTER_FRAME
+#endif
+
 XDataAllocator::XDataAllocator(BYTE* address, uint size, HANDLE processHandle) :
     freeList(nullptr),
     start(address),
@@ -155,7 +159,7 @@ void XDataAllocator::Register(XDataAllocation * xdataInfo, ULONG_PTR functionSta
 
 #else  // !_WIN32
     Assert(ReadHead(xdataInfo->address));  // should be non-empty .eh_frame
-    __register_frame(xdataInfo->address);
+    __REGISTER_FRAME(xdataInfo->address);
 #endif
 }
 
@@ -176,6 +180,6 @@ void XDataAllocator::Unregister(XDataAllocation * xdataInfo)
 
 #else  // !_WIN32
     Assert(ReadHead(xdataInfo->address));  // should be non-empty .eh_frame
-    __deregister_frame(xdataInfo->address);
+    __DEREGISTER_FRAME(xdataInfo->address);
 #endif
 }

lib/Common/Memory/amd64/XDataAllocator.h

@@ -7,11 +7,6 @@ CompileAssert(false)
 #endif
 #pragma once
 
-#ifndef _WIN32
-extern "C" void __register_frame(const void* ehframe);
-extern "C" void __deregister_frame(const void* ehframe);
-#endif
-
 namespace Memory
 {
 #ifdef _WIN32

lib/Common/PlatformAgnostic/AssemblyCommon.h

@@ -0,0 +1,29 @@
+//-------------------------------------------------------------------------------------------------------
+// Copyright (C) Microsoft. All rights reserved.
+// Licensed under the MIT license. See LICENSE.txt file in the project root for full license information.
+//-------------------------------------------------------------------------------------------------------
+#pragma once
+
+#ifndef _WIN32
+
+extern "C" void __register_frame(const void* ehframe);
+extern "C" void __deregister_frame(const void* ehframe);
+
+#if defined(_AMD64_) && !defined(DISABLE_JIT)
+#if defined(__APPLE__) // no multi fde support
+#ifndef __IOS__
+typedef void(*mac_fde_reg_op)(const void*addr);
+void mac_fde_wrapper(const char *dataStart, mac_fde_reg_op reg_op);
+#define __REGISTER_FRAME(addr) mac_fde_wrapper((const char*)addr, __register_frame)
+#define __DEREGISTER_FRAME(addr) mac_fde_wrapper((const char*)addr, __deregister_frame)
+#else // __IOS__ // no JIT
+#define __REGISTER_FRAME(addr)
+#define __DEREGISTER_FRAME(addr)
+#endif // !__IOS__
+#else // multi FDE support
+#define __REGISTER_FRAME(addr) __register_frame(addr)
+#define __DEREGISTER_FRAME(addr) __deregister_frame(addr)
+#endif // __APPLE__
+#endif // _AMD64_ && !DISABLE_JIT
+
+#endif // !_WIN32

lib/Runtime/ByteCode/AsmJsByteCodeWriter.cpp

@@ -436,7 +436,7 @@ namespace Js
     {
         MULTISIZE_LAYOUT_WRITE(AsmReg7, op, R0, R1, R2, R3, R4, R5, R6);
     }
-    
+
     void AsmJsByteCodeWriter::AsmReg9(OpCodeAsmJs op, RegSlot R0, RegSlot R1, RegSlot R2, RegSlot R3, RegSlot R4, RegSlot R5, RegSlot R6, RegSlot R7, RegSlot R8)
     {
         MULTISIZE_LAYOUT_WRITE(AsmReg9, op, R0, R1, R2, R3, R4, R5, R6, R7, R8);
@@ -445,7 +445,7 @@ namespace Js
     {
         MULTISIZE_LAYOUT_WRITE(AsmReg10, op, R0, R1, R2, R3, R4, R5, R6, R7, R8, R9);
     }
-    
+
     void AsmJsByteCodeWriter::AsmReg11(OpCodeAsmJs op, RegSlot R0, RegSlot R1, RegSlot R2, RegSlot R3, RegSlot R4, RegSlot R5, RegSlot R6, RegSlot R7, RegSlot R8, RegSlot R9, RegSlot R10)
     {
         MULTISIZE_LAYOUT_WRITE(AsmReg11, op, R0, R1, R2, R3, R4, R5, R6, R7, R8, R9, R10);
@@ -456,7 +456,7 @@ namespace Js
     {
         MULTISIZE_LAYOUT_WRITE(AsmReg17, op, R0, R1, R2, R3, R4, R5, R6, R7, R8, R9, R10, R11, R12, R13, R14, R15, R16);
     }
-    
+
     void AsmJsByteCodeWriter::AsmReg18(OpCodeAsmJs op, RegSlot R0, RegSlot R1, RegSlot R2, RegSlot R3, RegSlot R4, RegSlot R5, RegSlot R6, RegSlot R7, RegSlot R8,
         RegSlot R9, RegSlot R10, RegSlot R11, RegSlot R12, RegSlot R13, RegSlot R14, RegSlot R15, RegSlot R16, RegSlot R17)
     {

lib/Runtime/ByteCode/ByteCodeWriter.cpp

@@ -216,7 +216,7 @@ namespace Js
             AllocateLoopHeaders();
         }
 
-        
+
 
         m_functionWrite->MarkScript(finalByteCodeBlock, finalAuxiliaryBlock, finalAuxiliaryContextBlock,
             m_byteCodeCount, m_byteCodeInLoopCount, m_byteCodeWithoutLDACount);
@@ -2397,7 +2397,7 @@ namespace Js
         //
         // Write the buffer's contents
         //
-        
+
         Assert(byteOffset < m_auxiliaryData.GetCurrentOffset());
 
         OpLayoutAuxNoReg data;
@@ -3242,7 +3242,9 @@ namespace Js
     }
 
     template <>
-    inline void ByteCodeWriter::Data::EncodeOpCode<SmallLayout>(uint16 op, ByteCodeWriter* writer)
+    void ByteCodeWriter::Data::EncodeOpCode<SmallLayout>(
+        uint16 op,
+        ByteCodeWriter* writer)
     {
         DebugOnly(const uint offset = currentOffset);
         if (op <= (uint16)Js::OpCode::MaxByteSizedOpcodes)
@@ -3256,7 +3258,8 @@ namespace Js
             Write(&byteop, sizeof(byte));
             Write(&op, sizeof(uint16));
         }
-        Assert(OpCodeUtil::EncodedSize((Js::OpCode)op, SmallLayout) == (currentOffset - offset));
+        Assert(OpCodeUtil::EncodedSize((Js::OpCode)op, SmallLayout)
+               == (currentOffset - offset));
     }
 
     template <LayoutSize layoutSize>

lib/Runtime/ByteCode/ByteCodeWriter.h

@@ -96,6 +96,7 @@ namespace Js
 
             template <LayoutSize layoutSize> void EncodeOpCode(uint16 op, ByteCodeWriter* writer);
             template <> void EncodeOpCode<SmallLayout>(uint16 op, ByteCodeWriter* writer);
+
             // EncodeT functions return the global offset where the opcode has been encoded
             template <LayoutSize layoutSize> uint EncodeT(OpCode op, ByteCodeWriter* writer);
             template <LayoutSize layoutSize> uint EncodeT(OpCode op, const void * rawData, int byteSize, ByteCodeWriter* writer);

lib/Runtime/Language/DynamicProfileInfo.cpp

@@ -889,10 +889,11 @@ namespace Js
 
     // We are overloading the value types to store whether it is a mod by power of 2.
     // TaggedInt:
-    inline void DynamicProfileInfo::RecordModulusOpType(FunctionBody* body, ProfileId profileId, bool isModByPowerOf2)
+    void DynamicProfileInfo::RecordModulusOpType(FunctionBody* body,
+                                 ProfileId profileId, bool isModByPowerOf2)
     {
         Assert(profileId < body->GetProfiledDivOrRemCount());
-        // allow one op of the modulus to be optimized - anyway
+        /* allow one op of the modulus to be optimized - anyway */
         if (divideTypeInfo[profileId].IsUninitialized())
         {
             divideTypeInfo[profileId] = ValueType::GetInt(true);
@@ -901,11 +902,13 @@ namespace Js
         {
             if (isModByPowerOf2)
             {
-                divideTypeInfo[profileId] = divideTypeInfo[profileId].Merge(ValueType::GetInt(true));
+                divideTypeInfo[profileId] = divideTypeInfo[profileId]
+                                           .Merge(ValueType::GetInt(true));
             }
             else
             {
-                divideTypeInfo[profileId] = divideTypeInfo[profileId].Merge(ValueType::Float);
+                divideTypeInfo[profileId] = divideTypeInfo[profileId]
+                                           .Merge(ValueType::Float);
             }
         }
     }

lib/Runtime/Language/DynamicProfileInfo.h

@@ -5,7 +5,7 @@
 #pragma once
 
 // DisableJit-TODO
-#if ENABLE_PROFILE_INFO 
+#if ENABLE_PROFILE_INFO
 
 #ifdef DYNAMIC_PROFILE_MUTATOR
 class DynamicProfileMutatorImpl;
@@ -375,6 +375,7 @@ namespace Js
         ValueType * GetDivideTypeInfo() const { return divideTypeInfo; }
 
         void RecordModulusOpType(FunctionBody* body, ProfileId profileId, bool isModByPowerOf2);
+
         bool IsModulusOpByPowerOf2(FunctionBody* body, ProfileId profileId) const;
 
         void RecordSwitchType(FunctionBody* body, ProfileId switchId, Var object);

lib/Runtime/Language/InterpreterStackFrame.cpp

@@ -1696,7 +1696,6 @@ namespace Js
     {
 #if DYNAMIC_INTERPRETER_THUNK
         Assert(function);
-
         Js::FunctionBody *functionBody = function->GetFunctionBody();
         JavascriptMethod entrypoint = functionBody->EnsureDynamicInterpreterThunk(function->GetFunctionEntryPointInfo());
         Assert(!IsDelayDynamicInterpreterThunk(functionBody->GetDirectEntryPoint(function->GetEntryPointInfo())));