Add linux-riscv64 nativeaot runtime build (dotnet#110688)

* Fix registers/offsets in riscv-nativeaot.

* Fix *.S files in riscv-nativeaot.

* Fix riscv64 nativeaot native build

* Managed build fixes

* Apply suggestions from code review

Co-authored-by: Tymoteusz Wenerski <[email protected]>

* Apply suggestions from code review

Co-authored-by: Tomasz Sowiński <[email protected]>

* Update other macros in AsmOffsetsVerify

* Update src/coreclr/tools/aot/ILCompiler.Compiler/Compiler/ObjectWriter/ElfObjectWriter.cs

Co-authored-by: Adeel Mujahid <[email protected]>

---------

Co-authored-by: sunlijun <[email protected]>
Co-authored-by: Tomek Sowiński <[email protected]>
Co-authored-by: Tymoteusz Wenerski <[email protected]>
Co-authored-by: Jan Kotas <[email protected]>

Author: 5 people

src/coreclr/CMakeLists.txt

@@ -151,7 +151,7 @@ add_subdirectory(tools/aot/jitinterface)
 
 if(NOT CLR_CROSS_COMPONENTS_BUILD)
   # NativeAOT only buildable for a subset of CoreCLR-supported configurations
-  if(CLR_CMAKE_HOST_ARCH_ARM64 OR CLR_CMAKE_HOST_ARCH_AMD64 OR CLR_CMAKE_HOST_ARCH_ARM OR CLR_CMAKE_HOST_ARCH_LOONGARCH64 OR (CLR_CMAKE_HOST_ARCH_I386 AND CLR_CMAKE_HOST_WIN32))
+  if(CLR_CMAKE_HOST_ARCH_ARM64 OR CLR_CMAKE_HOST_ARCH_AMD64 OR CLR_CMAKE_HOST_ARCH_ARM OR CLR_CMAKE_HOST_ARCH_LOONGARCH64 OR CLR_CMAKE_HOST_ARCH_RISCV64 OR (CLR_CMAKE_HOST_ARCH_I386 AND CLR_CMAKE_HOST_WIN32))
     add_subdirectory(nativeaot)
   endif()
 endif(NOT CLR_CROSS_COMPONENTS_BUILD)

src/coreclr/nativeaot/Common/src/Internal/Runtime/TransitionBlock.cs

@@ -35,7 +35,7 @@
 #define ENREGISTERED_RETURNTYPE_INTEGER_MAXSIZE
 #define ENREGISTERED_PARAMTYPE_MAXSIZE
 #elif TARGET_WASM
-#elif TARGET_LOONGARCH64
+#elif TARGET_LOONGARCH64 || TARGET_RISCV64
 #define CALLDESCR_ARGREGS                          // CallDescrWorker has ArgumentRegister parameter
 #define CALLDESCR_FPARGREGS                        // CallDescrWorker has FloatArgumentRegisters parameter
 #define ENREGISTERED_RETURNTYPE_MAXSIZE
@@ -360,6 +360,60 @@ internal struct ArchitectureConstants
         public const int STACK_ELEM_SIZE = 8;
         public static int StackElemSize(int size) { return (((size) + STACK_ELEM_SIZE - 1) & ~(STACK_ELEM_SIZE - 1)); }
     }
+#elif TARGET_RISCV64
+    [StructLayout(LayoutKind.Sequential)]
+    internal struct ReturnBlock
+    {
+        private IntPtr returnValue;
+        private IntPtr returnValue2;
+        private IntPtr returnValue3;
+        private IntPtr returnValue4;
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
+    internal struct ArgumentRegisters
+    {
+        private IntPtr a0;
+        private IntPtr a1;
+        private IntPtr a2;
+        private IntPtr a3;
+        private IntPtr a4;
+        private IntPtr a5;
+        private IntPtr a6;
+        private IntPtr a7;
+        public static unsafe int GetOffsetOfa7()
+        {
+            return sizeof(IntPtr) * 7;
+        }
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
+    internal struct FloatArgumentRegisters
+    {
+        private double fa0;
+        private double fa1;
+        private double fa2;
+        private double fa3;
+        private double fa4;
+        private double fa5;
+        private double fa6;
+        private double fa7;
+    }
+
+    internal struct ArchitectureConstants
+    {
+        // To avoid corner case bugs, limit maximum size of the arguments with sufficient margin
+        public const int MAX_ARG_SIZE = 0xFFFFFF;
+
+        public const int NUM_ARGUMENT_REGISTERS = 8;
+        public const int ARGUMENTREGISTERS_SIZE = NUM_ARGUMENT_REGISTERS * 8;
+        public const int ENREGISTERED_RETURNTYPE_MAXSIZE = 32;                  // bytes (four FP registers: fa0, fa1, fa2, and fa3)
+        public const int ENREGISTERED_RETURNTYPE_INTEGER_MAXSIZE = 16;          // bytes (two int registers: a0 and a1)
+        public const int ENREGISTERED_RETURNTYPE_INTEGER_MAXSIZE_PRIMITIVE = 8;
+        public const int ENREGISTERED_PARAMTYPE_MAXSIZE = 16;                   // bytes (max value type size that can be passed by value)
+        public const int STACK_ELEM_SIZE = 8;
+        public static int StackElemSize(int size) { return (((size) + STACK_ELEM_SIZE - 1) & ~(STACK_ELEM_SIZE - 1)); }
+    }
 #endif
 
     //
@@ -465,6 +519,20 @@ public static unsafe int GetOffsetOfArgumentRegisters()
             return sizeof(ReturnBlock);
         }
 
+        public IntPtr m_alignmentPad;
+#elif TARGET_RISCV64
+        public ReturnBlock m_returnBlock;
+        public static unsafe int GetOffsetOfReturnValuesBlock()
+        {
+            return 0;
+        }
+
+        public ArgumentRegisters m_argumentRegisters;
+        public static unsafe int GetOffsetOfArgumentRegisters()
+        {
+            return sizeof(ReturnBlock);
+        }
+
         public IntPtr m_alignmentPad;
 #else
 #error Portability problem

src/coreclr/nativeaot/Directory.Build.props

@@ -89,6 +89,9 @@
   <PropertyGroup Condition="'$(Platform)' == 'loongarch64'">
     <DefineConstants>TARGET_64BIT;TARGET_LOONGARCH64;$(DefineConstants)</DefineConstants>
   </PropertyGroup>
+  <PropertyGroup Condition="'$(Platform)' == 'riscv64'">
+    <DefineConstants>TARGET_64BIT;TARGET_RISCV64;$(DefineConstants)</DefineConstants>
+  </PropertyGroup>
 
   <PropertyGroup>
     <DefineConstants Condition="'$(TargetsWindows)'=='true'">TARGET_WINDOWS;$(DefineConstants)</DefineConstants>

src/coreclr/nativeaot/Runtime/AsmOffsetsVerify.cpp

@@ -28,17 +28,18 @@ class AsmOffsets
     static_assert(offsetof(Array, m_Length) == offsetof(String, m_Length), "The length field of String and Array have different offsets");
     static_assert(sizeof(((Array*)0)->m_Length) == sizeof(((String*)0)->m_Length), "The length field of String and Array have different sizes");
 
+#define TO_STRING(x) #x
+#define OFFSET_STRING(cls, member) TO_STRING(offsetof(cls, member))
+
+// Macro definition
 #define PLAT_ASM_OFFSET(offset, cls, member) \
-    static_assert((offsetof(cls, member) == 0x##offset) || (offsetof(cls, member) > 0x##offset), "Bad asm offset for '" #cls "." #member "', the actual offset is smaller than 0x" #offset "."); \
-    static_assert((offsetof(cls, member) == 0x##offset) || (offsetof(cls, member) < 0x##offset), "Bad asm offset for '" #cls "." #member "', the actual offset is larger than 0x" #offset ".");
+    static_assert(offsetof(cls, member) == 0x##offset, "Bad asm offset for '" #cls "." #member "'. Actual offset: " OFFSET_STRING(cls, member));
 
-#define PLAT_ASM_SIZEOF(size,   cls        ) \
-    static_assert((sizeof(cls) == 0x##size) || (sizeof(cls) > 0x##size), "Bad asm size for '" #cls "', the actual size is smaller than 0x" #size "."); \
-    static_assert((sizeof(cls) == 0x##size) || (sizeof(cls) < 0x##size), "Bad asm size for '" #cls "', the actual size is larger than 0x" #size ".");
+#define PLAT_ASM_SIZEOF(size, cls) \
+    static_assert(sizeof(cls) == 0x##size, "Bad asm size for '" #cls "'. Actual size: " OFFSET_STRING(cls, 0x##size));
 
 #define PLAT_ASM_CONST(constant, expr) \
-    static_assert(((expr) == 0x##constant) || ((expr) > 0x##constant), "Bad asm constant for '" #expr "', the actual value is smaller than 0x" #constant "."); \
-    static_assert(((expr) == 0x##constant) || ((expr) < 0x##constant), "Bad asm constant for '" #expr "', the actual value is larger than 0x" #constant ".");
+    static_assert((expr) == 0x##constant, "Bad asm constant for '" #expr "'. Actual value: " OFFSET_STRING(expr, 0x##constant));
 
 #include "AsmOffsets.h"
 

src/coreclr/nativeaot/Runtime/CommonMacros.h

@@ -119,7 +119,7 @@ inline bool IS_ALIGNED(T* val, uintptr_t alignment);
 #define LOG2_PTRSIZE 2
 #define POINTER_SIZE 4
 
-#elif defined(HOST_LOONGARCH64)
+#elif defined(HOST_LOONGARCH64) || defined (HOST_RISCV64)
 
 #define LOG2_PTRSIZE 3
 #define POINTER_SIZE 8

src/coreclr/nativeaot/Runtime/EHHelpers.cpp

@@ -203,6 +203,24 @@ FCIMPL3(void, RhpCopyContextFromExInfo, void * pOSContext, int32_t cbOSContext,
     pContext->Sp = pPalContext->SP;
     pContext->Ra = pPalContext->RA;
     pContext->Pc = pPalContext->IP;
+#elif defined(HOST_RISCV64)
+    pContext->A0 = pPalContext->A0;
+    pContext->A1 = pPalContext->A1;
+    pContext->S1 = pPalContext->S1;
+    pContext->S2 = pPalContext->S2;
+    pContext->S3 = pPalContext->S3;
+    pContext->S4 = pPalContext->S4;
+    pContext->S5 = pPalContext->S5;
+    pContext->S6 = pPalContext->S6;
+    pContext->S7 = pPalContext->S7;
+    pContext->S8 = pPalContext->S8;
+    pContext->S9 = pPalContext->S9;
+    pContext->S10 = pPalContext->S10;
+    pContext->S11 = pPalContext->S11;
+    pContext->Fp = pPalContext->FP;
+    pContext->Sp = pPalContext->SP;
+    pContext->Ra = pPalContext->RA;
+    pContext->Pc = pPalContext->IP;
 #elif defined(HOST_WASM)
     // No registers, no work to do yet
 #else
@@ -295,7 +313,7 @@ EXTERN_C CODE_LOCATION RhpCheckedAssignRefEBPAVLocation;
 #endif
 EXTERN_C CODE_LOCATION RhpByRefAssignRefAVLocation1;
 
-#if !defined(HOST_ARM64) && !defined(HOST_LOONGARCH64)
+#if !defined(HOST_ARM64) && !defined(HOST_LOONGARCH64) && !defined(HOST_RISCV64)
 EXTERN_C CODE_LOCATION RhpByRefAssignRefAVLocation2;
 #endif
 
@@ -328,7 +346,7 @@ static bool InWriteBarrierHelper(uintptr_t faultingIP)
         (uintptr_t)&RhpCheckedAssignRefEBPAVLocation,
 #endif
         (uintptr_t)&RhpByRefAssignRefAVLocation1,
-#if !defined(HOST_ARM64) && !defined(HOST_LOONGARCH64)
+#if !defined(HOST_ARM64) && !defined(HOST_LOONGARCH64) && !defined(HOST_RISCV64)
         (uintptr_t)&RhpByRefAssignRefAVLocation2,
 #endif
     };
@@ -410,7 +428,7 @@ static uintptr_t UnwindSimpleHelperToCaller(
     pContext->SetSp(sp+sizeof(uintptr_t)); // pop the stack
 #elif defined(HOST_ARM) || defined(HOST_ARM64)
     uintptr_t adjustedFaultingIP = pContext->GetLr();
-#elif defined(HOST_LOONGARCH64)
+#elif defined(HOST_LOONGARCH64) || defined(HOST_RISCV64)
     uintptr_t adjustedFaultingIP = pContext->GetRa();
 #else
     uintptr_t adjustedFaultingIP = 0; // initializing to make the compiler happy

src/coreclr/nativeaot/Runtime/MiscHelpers.cpp

@@ -335,6 +335,7 @@ FCIMPL1(uint8_t *, RhGetCodeTarget, uint8_t * pCodeOrg)
         int64_t distToTarget = ((int64_t)pCode[0] << 38) >> 36;
         return (uint8_t *)pCode + distToTarget;
     }
+
 #elif TARGET_LOONGARCH64
     uint32_t * pCode = (uint32_t *)pCodeOrg;
     // is this "addi.d $a0, $a0, 8"?
@@ -370,6 +371,45 @@ FCIMPL1(uint8_t *, RhGetCodeTarget, uint8_t * pCodeOrg)
         distToTarget += ((((int64_t)pCode[1] & ~0x3ff) << 38) >> 46);
         return (uint8_t *)((int64_t)pCode + distToTarget);
     }
+
+#elif TARGET_RISCV64
+    uint32_t * pCode = (uint32_t *)pCodeOrg;
+    if (pCode[0] == 0x00850513)  // Encoding for `addi a0, a0, 8` in 32-bit instruction format
+    {
+        // unboxing sequence
+        unboxingStub = true;
+        pCode++;
+    }
+    // is this an indirect jump?
+    // lui t0, imm; jalr t0, t0, imm12
+    if ((pCode[0] & 0x7f) == 0x17 &&                 // auipc
+        (pCode[1] & 0x707f) == 0x3003 &&             // ld with funct3=011
+        (pCode[2] & 0x707f) == 0x0067)               // jr (jalr with x0 as rd and funct3=000)
+    {
+        // Compute the distance to the IAT cell
+        int64_t distToIatCell = (((int32_t)pCode[0]) >> 12) << 12;  // Extract imm20 from auipc
+        distToIatCell += ((int32_t)pCode[1]) >> 20;                    // Add imm12 from ld
+
+        uint8_t ** pIatCell = (uint8_t **)(((int64_t)pCode & ~0xfff) + distToIatCell);
+        return *pIatCell;
+    }
+
+    // Is this an unboxing stub followed by a relative jump?
+    // auipc t0, imm20; jalr ra, imm12(t0)
+    else if (unboxingStub &&
+            (pCode[0] & 0x7f) == 0x17 &&                 // auipc opcode
+            (pCode[1] & 0x707f) == 0x0067)              // jalr opcode with funct3=000
+    {
+        // Extract imm20 from auipc
+        int64_t distToTarget = (((int32_t)pCode[0]) >> 12) << 12;  // Extract imm20 (bits 31:12)
+
+        // Extract imm12 from jalr
+        distToTarget += ((int32_t)pCode[1]) >> 20;  // Extract imm12 (bits 31:20)
+
+        // Calculate the final target address relative to PC
+        return (uint8_t *)((int64_t)pCode + distToTarget);
+    }
+
 #else
     UNREFERENCED_PARAMETER(unboxingStub);
     PORTABILITY_ASSERT("RhGetCodeTarget");

src/coreclr/nativeaot/Runtime/PalRedhawk.h

@@ -531,6 +531,83 @@ typedef struct DECLSPEC_ALIGN(16) _CONTEXT {
     }
 } CONTEXT, *PCONTEXT;
 
+#elif defined(TARGET_RISCV64)
+
+#define CONTEXT_RISCV64   0x01000000L
+
+#define CONTEXT_CONTROL (CONTEXT_RISCV64 | 0x1L)
+#define CONTEXT_INTEGER (CONTEXT_RISCV64 | 0x2L)
+
+#define RISCV64_MAX_BREAKPOINTS     8
+#define RISCV64_MAX_WATCHPOINTS     2
+
+typedef struct DECLSPEC_ALIGN(16) _CONTEXT {
+    //
+    // Control flags.
+    //
+    uint32_t ContextFlags;
+
+    //
+    // Integer registers
+    //
+    uint64_t X0;
+    uint64_t Ra;
+    uint64_t Sp;
+    uint64_t Gp;
+    uint64_t Tp;
+    uint64_t T0;
+    uint64_t T1;
+    uint64_t T2;
+    uint64_t Fp;
+    uint64_t S1;
+    uint64_t A0;
+    uint64_t A1;
+    uint64_t A2;
+    uint64_t A3;
+    uint64_t A4;
+    uint64_t A5;
+    uint64_t A6;
+    uint64_t A7;
+    uint64_t S2;
+    uint64_t S3;
+    uint64_t S4;
+    uint64_t S5;
+    uint64_t S6;
+    uint64_t S7;
+    uint64_t S8;
+    uint64_t S9;
+    uint64_t S10;
+    uint64_t S11;
+    uint64_t T3;
+    uint64_t T4;
+    uint64_t T5;
+    uint64_t T6;
+    uint64_t Pc;
+
+    //
+    // Floating Point Registers
+    //
+    uint64_t F[32];
+    uint32_t Fcsr;
+
+    void SetIp(uintptr_t ip) { Pc = ip; }
+    void SetArg0Reg(uintptr_t val) { A0 = val; }
+    void SetArg1Reg(uintptr_t val) { A1 = val; }
+    uintptr_t GetIp() { return Pc; }
+    uintptr_t GetRa() { return Ra; }
+    uintptr_t GetSp() { return Sp; }
+
+    template <typename F>
+    void ForEachPossibleObjectRef(F lambda)
+    {
+        for (uint64_t* pReg = &X0; pReg <= &T6; pReg++)
+            lambda((size_t*)pReg);
+
+        // RA can be used as a scratch register
+        lambda((size_t*)&Ra);
+    }
+} CONTEXT, *PCONTEXT;
+
 #elif defined(HOST_WASM)
 
 typedef struct DECLSPEC_ALIGN(8) _CONTEXT {

src/coreclr/nativeaot/Runtime/PalRedhawkCommon.h

@@ -61,6 +61,7 @@ struct PAL_LIMITED_CONTEXT
     uintptr_t GetLr() const { return LR; }
     void SetIp(uintptr_t ip) { IP = ip; }
     void SetSp(uintptr_t sp) { SP = sp; }
+
 #elif defined(TARGET_ARM64)
     uintptr_t  FP;
     uintptr_t  LR;
@@ -91,6 +92,7 @@ struct PAL_LIMITED_CONTEXT
     uintptr_t GetLr() const { return LR; }
     void SetIp(uintptr_t ip) { IP = ip; }
     void SetSp(uintptr_t sp) { SP = sp; }
+
 #elif defined(TARGET_LOONGARCH64)
     uintptr_t  FP;
     uintptr_t  RA;
@@ -121,6 +123,37 @@ struct PAL_LIMITED_CONTEXT
     void SetIp(uintptr_t ip) { IP = ip; }
     void SetSp(uintptr_t sp) { SP = sp; }
 
+#elif defined(TARGET_RISCV64)
+
+    uintptr_t  FP;
+    uintptr_t  RA;
+
+    uintptr_t  A0;
+    uintptr_t  A1;
+    uintptr_t  S1;
+    uintptr_t  S2;
+    uintptr_t  S3;
+    uintptr_t  S4;
+    uintptr_t  S5;
+    uintptr_t  S6;
+    uintptr_t  S7;
+    uintptr_t  S8;
+    uintptr_t  S9;
+    uintptr_t  S10;
+    uintptr_t  S11;
+
+    uintptr_t  SP;
+    uintptr_t  IP;
+
+    uint64_t  F[12];
+
+    uintptr_t GetIp() const { return IP; }
+    uintptr_t GetSp() const { return SP; }
+    uintptr_t GetFp() const { return FP; }
+    uintptr_t GetRa() const { return RA; }
+    void SetIp(uintptr_t ip) { IP = ip; }
+    void SetSp(uintptr_t sp) { SP = sp; }
+
 #elif defined(UNIX_AMD64_ABI)
     // Param regs: rdi, rsi, rdx, rcx, r8, r9, scratch: rax, rdx (both return val), preserved: rbp, rbx, r12-r15
     uintptr_t  IP;