From: gns <[email protected]>
From: Jingwiw <[email protected]>
Subject: [PATCH] riscv64: add initial support for riscv64

Add RISC-V 64 (rv64g) platform support with lp64d ABI to LuaJIT (not upstreamed yet).
Part1: modified files

diff --git a/bundle/LuaJIT-2.1-20220411/Makefile b/bundle/LuaJIT-2.1-20220411/Makefile
index 6fb8efa..c81a448 100644
--- a/bundle/LuaJIT-2.1-20220411/Makefile
+++ b/bundle/LuaJIT-2.1-20220411/Makefile
@@ -86,6 +86,7 @@ FILE_MAN= luajit.1
 FILE_PC= luajit.pc
 FILES_INC= lua.h lualib.h lauxlib.h luaconf.h lua.hpp luajit.h
 FILES_JITLIB= bc.lua bcsave.lua dump.lua p.lua v.lua zone.lua \
+	      dis_riscv.lua dis_riscv64.lua \
 	      dis_x86.lua dis_x64.lua dis_arm.lua dis_arm64.lua \
 	      dis_arm64be.lua dis_ppc.lua dis_mips.lua dis_mipsel.lua \
 	      dis_mips64.lua dis_mips64el.lua vmdef.lua
diff --git a/bundle/LuaJIT-2.1-20220411/src/Makefile b/bundle/LuaJIT-2.1-20220411/src/Makefile
index 68a9a7c..e1b7464 100644
--- a/bundle/LuaJIT-2.1-20220411/src/Makefile
+++ b/bundle/LuaJIT-2.1-20220411/src/Makefile
@@ -53,6 +53,7 @@ CCOPT_arm=
 CCOPT_arm64=
 CCOPT_ppc=
 CCOPT_mips=
+CCOPT_riscv64=
 #
 #CCDEBUG=
 # Uncomment the next line to generate debug information:
@@ -270,6 +271,12 @@ ifneq (,$(findstring LJ_TARGET_MIPS ,$(TARGET_TESTARCH)))
   else
     TARGET_LJARCH= mips
   endif
+else
+ifneq (,$(findstring LJ_TARGET_RISCV32 ,$(TARGET_TESTARCH)))
+    TARGET_LJARCH= riscv32
+else
+ifneq (,$(findstring LJ_TARGET_RISCV64 ,$(TARGET_TESTARCH)))
+    TARGET_LJARCH= riscv64
 else
   $(error Unsupported target architecture)
 endif
@@ -279,6 +286,8 @@ endif
 endif
 endif
 endif
+endif
+endif
 
 ifneq (,$(findstring LJ_TARGET_PS3 1,$(TARGET_TESTARCH)))
   TARGET_SYS= PS3
@@ -454,6 +463,12 @@ else
 ifneq (,$(findstring LJ_TARGET_MIPSR6 ,$(TARGET_TESTARCH)))
   DASM_AFLAGS+= -D MIPSR6
 endif
+ifneq (,$(findstring LJ_TARGET_RISCV32 ,$(TARGET_TESTARCH)))
+  DASM_AFLAGS+= -D RISCV32
+endif
+ifneq (,$(findstring LJ_TARGET_RISCV64 ,$(TARGET_TESTARCH)))
+  DASM_AFLAGS+= -D RISCV64
+endif
 ifeq (ppc,$(TARGET_LJARCH))
   ifneq (,$(findstring LJ_ARCH_SQRT 1,$(TARGET_TESTARCH)))
     DASM_AFLAGS+= -D SQRT
diff --git a/bundle/LuaJIT-2.1-20220411/src/host/buildvm.c b/bundle/LuaJIT-2.1-20220411/src/host/buildvm.c
index 4efda1b..def4a92 100644
--- a/bundle/LuaJIT-2.1-20220411/src/host/buildvm.c
+++ b/bundle/LuaJIT-2.1-20220411/src/host/buildvm.c
@@ -67,6 +67,8 @@ static int collect_reloc(BuildCtx *ctx, uint8_t *addr, int idx, int type);
 #include "../dynasm/dasm_mips.h"
 #elif LJ_TARGET_S390X
 #include "../dynasm/dasm_s390x.h"
+#elif LJ_TARGET_RISCV32 || LJ_TARGET_RISCV64
+#include "../dynasm/dasm_riscv.h"
 #else
 #error "No support for this architecture (yet)"
 #endif
diff --git a/bundle/LuaJIT-2.1-20220411/src/host/buildvm_asm.c b/bundle/LuaJIT-2.1-20220411/src/host/buildvm_asm.c
index e73f9b1..0a187d5 100644
--- a/bundle/LuaJIT-2.1-20220411/src/host/buildvm_asm.c
+++ b/bundle/LuaJIT-2.1-20220411/src/host/buildvm_asm.c
@@ -208,6 +208,15 @@ static void emit_asm_wordreloc(BuildCtx *ctx, uint8_t *p, int n,
 	  "Error: unsupported opcode %08x for %s symbol relocation.\n",
 	  ins, sym);
   exit(1);
+#elif LJ_TARGET_RISCV32 || LJ_TARGET_RISCV64
+  if ((ins & 0x7f) == 0x6fu) {
+    fprintf(ctx->fp, "\tjal %s\n", sym);
+  } else {
+    fprintf(stderr,
+  	    "Error: unsupported opcode %08x for %s symbol relocation.\n",
+  	    ins, sym);
+    exit(1);
+  }
 #else
 #error "missing relocation support for this architecture"
 #endif
@@ -327,6 +336,9 @@ void emit_asm(BuildCtx *ctx)
 #if LJ_TARGET_MIPS
   fprintf(ctx->fp, ".set nomips16\n.abicalls\n.set noreorder\n.set nomacro\n");
 #endif
+#if LJ_TARGET_RISCV64
+  fprintf(ctx->fp, ".option arch, -c\n.option norelax\n");
+#endif
 
   for (i = rel = 0; i < ctx->nsym; i++) {
     int32_t ofs = ctx->sym[i].ofs;
diff --git a/bundle/LuaJIT-2.1-20220411/src/jit/bcsave.lua b/bundle/LuaJIT-2.1-20220411/src/jit/bcsave.lua
index 7cb23f1..c5e22f6 100644
--- a/bundle/LuaJIT-2.1-20220411/src/jit/bcsave.lua
+++ b/bundle/LuaJIT-2.1-20220411/src/jit/bcsave.lua
@@ -85,6 +85,7 @@ local map_arch = {
   mips64el =	{ e = "le", b = 64, m = 8, f = 0x80000007, },
   mips64r6 =	{ e = "be", b = 64, m = 8, f = 0xa0000407, },
   mips64r6el =	{ e = "le", b = 64, m = 8, f = 0xa0000407, },
+  riscv64 =    { e = "le", b = 64, m = 243, f = 0x00000004, },
 }
 
 local map_os = {
diff --git a/bundle/LuaJIT-2.1-20220411/src/lib_jit.c b/bundle/LuaJIT-2.1-20220411/src/lib_jit.c
index 50c2b13..11748a4 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lib_jit.c
+++ b/bundle/LuaJIT-2.1-20220411/src/lib_jit.c
@@ -714,6 +714,75 @@ JIT_PARAMDEF(JIT_PARAMINIT)
 #include <sys/utsname.h>
 #endif
 
+#if LJ_TARGET_RISCV64
+#include <setjmp.h>
+#include <signal.h>
+static sigjmp_buf sigbuf = {0};
+static void detect_sigill(int sig)
+{
+  siglongjmp(sigbuf, 1);
+}
+
+static int riscv_compressed()
+{
+#if defined(__riscv_compressed)
+  // Don't bother checking for RVC -- would crash before getting here.
+  return 1;
+#elif defined(__GNUC__)
+  // c.nop; c.nop;
+  __asm__(".4byte 0x00010001");
+  return 1;
+#else
+  return 0;
+#endif
+}
+
+static int riscv_zba()
+{
+#if defined(__GNUC__)
+  // Don't bother verifying the result, just check if the instruction exists.
+  // add.uw zero, zero, zero
+  __asm__(".4byte 0x0800003b");
+  return 1;
+#else
+  return 0;
+#endif
+}
+
+static int riscv_zbb()
+{
+#if defined(__GNUC__)
+  register int t asm ("a0");
+  // addi a0, zero, 255; sext.b a0, a0;
+  __asm__("addi a0, zero, 255\n\t.4byte 0x60451513");
+  return t < 0;
+#else
+  return 0;
+#endif
+}
+
+static int riscv_xthead()
+{
+#if defined(__GNUC__)
+    register int t asm ("a0");
+    // C906 & C910 & C908 all have "xtheadc", XTheadBb subset "xtheadc".
+    // Therefore assume XThead* are present if XTheadBb is present.
+    // addi a0, zero, 255; th.ext a0, a0, 7, 0;
+    __asm__("addi a0, zero, 255\n\t.4byte 0x1c05250b");
+    return t == -1;		// In case of collision with other vendor extensions.
+#else
+    return 0;
+#endif
+}
+
+static uint32_t riscv_probe(int (*func)(void), uint32_t flag)
+{
+    if (sigsetjmp(sigbuf, 1) == 0) {
+        return func() ? flag : 0;
+    } else return 0;
+}
+#endif
+
 /* Arch-dependent CPU feature detection. */
 static uint32_t jit_cpudetect(void)
 {
@@ -786,6 +855,21 @@ static uint32_t jit_cpudetect(void)
 #endif
 #elif LJ_TARGET_S390X
   /* No optional CPU features to detect (for now). */
+#elif LJ_TARGET_RISCV64
+#if LJ_HASJIT
+  // SIGILL-based detection of RVC, Zba, Zbb and XThead. Welcome to the future.
+  struct sigaction old = {0}, act = {0};
+  act.sa_handler = detect_sigill;
+  sigaction(SIGILL, &act, &old);
+  flags |= riscv_probe(riscv_compressed, JIT_F_RVC);
+  flags |= riscv_probe(riscv_zba, JIT_F_RVZba);
+  flags |= riscv_probe(riscv_zbb, JIT_F_RVZbb);
+  flags |= riscv_probe(riscv_xthead, JIT_F_RVXThead);
+  sigaction(SIGILL, &old, NULL);
+
+  // Detect V/P?
+  // V have no hardware available, P not ratified yet.
+#endif
 #else
 #error "Missing CPU detection for this architecture"
 #endif
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_arch.h b/bundle/LuaJIT-2.1-20220411/src/lj_arch.h
index 9f7047d..3490af4 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_arch.h
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_arch.h
@@ -33,6 +33,10 @@
 #define LUAJIT_ARCH_mips64	7
 #define LUAJIT_ARCH_S390X	8
 #define LUAJIT_ARCH_s390x	8
+#define LUAJIT_ARCH_RISCV32	9
+#define LUAJIT_ARCH_riscv32	9
+#define LUAJIT_ARCH_RISCV64	10
+#define LUAJIT_ARCH_riscv64	10
 
 /* Target OS. */
 #define LUAJIT_OS_OTHER		0
@@ -69,6 +73,10 @@
 #define LUAJIT_TARGET	LUAJIT_ARCH_MIPS64
 #elif defined(__mips__) || defined(__mips) || defined(__MIPS__) || defined(__MIPS)
 #define LUAJIT_TARGET	LUAJIT_ARCH_MIPS32
+#elif defined(__riscv) && __riscv_xlen == 32
+#define LUAJIT_TARGET LUAJIT_ARCH_RISCV32
+#elif defined(__riscv) && __riscv_xlen == 64
+#define LUAJIT_TARGET LUAJIT_ARCH_RISCV64
 #else
 #error "No support for this architecture (yet)"
 #endif
@@ -451,6 +459,32 @@
 #define LJ_TARGET_GC64		1
 #define LJ_ARCH_NOJIT		1	/* NYI */
 
+#elif LUAJIT_TARGET == LUAJIT_ARCH_RISCV32
+#error "No support for RISC-V 32"
+
+#elif LUAJIT_TARGET == LUAJIT_ARCH_RISCV64
+#if defined(__riscv_float_abi_double)
+
+#define LJ_ARCH_NAME		"riscv64"
+#define LJ_ARCH_BITS		64
+#define LJ_ARCH_ENDIAN		LUAJIT_LE	/* Forget about BE for now */
+#define LJ_TARGET_RISCV64	1
+#define LJ_TARGET_GC64		1
+#define LJ_TARGET_EHRETREG	0 // TODO
+#define LJ_TARGET_EHRAREG	1
+#define LJ_TARGET_JUMPRANGE	30	/* JAL +-2^20 = +-1MB,\
+        AUIPC+JALR +-2^31 = +-2GB, leave 1 bit to avoid AUIPC corner case */
+#define LJ_TARGET_MASKSHIFT	1
+#define LJ_TARGET_MASKROT	1
+#define LJ_TARGET_UNIFYROT	2	/* Want only IR_BROR, no ROLI */
+#define LJ_ARCH_NUMMODE		LJ_NUMMODE_DUAL
+// for now
+#define LUAJIT_NO_UNWIND	1
+
+#else
+#error "No support for RISC-V 64 Soft-float/Single-float"
+#endif
+
 #else
 #error "No target architecture defined"
 #endif
@@ -525,6 +559,13 @@
 /* MIPS32ON64 aka n32 ABI support might be desirable, but difficult. */
 #error "Only n64 ABI supported for MIPS64"
 #endif
+#elif LJ_TARGET_RISCV
+#if !defined(__riscv_float_abi_double)
+#error "Only RISC-V 64 double float supported for now"
+#endif
+#if defined(__riscv_compressed)
+#error "Compressed instructions not supported for now"
+#endif
 #endif
 #endif
 
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_asm.c b/bundle/LuaJIT-2.1-20220411/src/lj_asm.c
index d811808..4deca17 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_asm.c
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_asm.c
@@ -185,6 +185,8 @@ IRFLDEF(FLOFS)
 #include "lj_emit_ppc.h"
 #elif LJ_TARGET_MIPS
 #include "lj_emit_mips.h"
+#elif LJ_TARGET_RISCV64
+#include "lj_emit_riscv.h"
 #else
 #error "Missing instruction emitter for target CPU"
 #endif
@@ -1664,6 +1666,8 @@ static void asm_loop(ASMState *as)
 #include "lj_asm_mips.h"
 #elif LJ_TARGET_S390X
 #include "lj_asm_s390x.h"
+#elif LJ_TARGET_RISCV64
+#include "lj_asm_riscv64.h"
 #else
 #error "Missing assembler for target CPU"
 #endif
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_ccall.c b/bundle/LuaJIT-2.1-20220411/src/lj_ccall.c
index 8162b95..96bc3b8 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_ccall.c
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_ccall.c
@@ -686,6 +686,97 @@
     if (ngpr < maxgpr) { dp = &cc->gpr[ngpr++]; goto done; } \
   }
 
+#elif LJ_TARGET_RISCV64
+/* -- RISC-V lp64d calling conventions ------------------------------------ */
+
+#define CCALL_HANDLE_STRUCTRET \
+  /* Return structs of size > 16 by reference. */ \
+  cc->retref = !(sz <= 16); \
+  if (cc->retref) cc->gpr[ngpr++] = (GPRArg)dp;
+
+#define CCALL_HANDLE_STRUCTRET2 \
+  unsigned int cl = ccall_classify_struct(cts, ctr); \
+  if ((cl & 4) && (cl >> 8) <= 2) { \
+    CTSize i = (cl >> 8) - 1; \
+    do { ((float *)dp)[i] = cc->fpr[i].f; } while (i--); \
+  } else { \
+    if (cl > 1) { \
+      sp = (uint8_t *)&cc->fpr[0]; \
+      if ((cl >> 8) > 2) \
+        sp = (uint8_t *)&cc->gpr[0]; \
+    } \
+      memcpy(dp, sp, ctr->size); \
+  } \
+
+#define CCALL_HANDLE_COMPLEXRET \
+  /* Complex values are returned in 1 or 2 FPRs. */ \
+  cc->retref = 0;
+
+#define CCALL_HANDLE_COMPLEXRET2 \
+  if (ctr->size == 2*sizeof(float)) {  /* Copy complex float from FPRs. */ \
+    ((float *)dp)[0] = cc->fpr[0].f; \
+    ((float *)dp)[1] = cc->fpr[1].f; \
+  } else {  /* Copy complex double from FPRs. */ \
+    ((double *)dp)[0] = cc->fpr[0].d; \
+    ((double *)dp)[1] = cc->fpr[1].d; \
+  }
+
+#define CCALL_HANDLE_COMPLEXARG \
+  /* Pass long double complex by reference. */ \
+  if (sz == 2*sizeof(long double)) { \
+    rp = cdataptr(lj_cdata_new(cts, did, sz)); \
+    sz = CTSIZE_PTR; \
+  } \
+  /* Pass complex in two FPRs or on stack. */ \
+  else if (sz == 2*sizeof(float)) { \
+    isfp = 2; \
+    sz = 2*CTSIZE_PTR; \
+  } else { \
+    isfp = 1; \
+    sz = 2*CTSIZE_PTR; \
+  }
+
+#define CCALL_HANDLE_RET \
+  if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
+    sp = (uint8_t *)&cc->fpr[0].f;
+
+#define CCALL_HANDLE_STRUCTARG \
+  /* Pass structs of size >16 by reference. */ \
+  unsigned int cl = ccall_classify_struct(cts, d); \
+  nff = cl >> 8; \
+  if (sz > 16) { \
+    rp = cdataptr(lj_cdata_new(cts, did, sz)); \
+    sz = CTSIZE_PTR; \
+  } \
+  /* Pass struct in FPRs. */ \
+  if (cl > 1) { \
+    isfp = (cl & 4) ? 2 : 1; \
+  }
+
+
+#define CCALL_HANDLE_REGARG \
+  if (isfp && (!isva)) {  /* Try to pass argument in FPRs. */ \
+    int n2 = ctype_isvector(d->info) ? 1 : \
+            isfp == 1 ? n : 2; \
+    if (nfpr + n2 <= CCALL_NARG_FPR && nff <= 2) { \
+      dp = &cc->fpr[nfpr]; \
+      nfpr += n2; \
+      goto done; \
+    } else { \
+      if (ngpr + n2 <= maxgpr) { \
+       dp = &cc->gpr[ngpr]; \
+       ngpr += n2; \
+       goto done; \
+      } \
+    } \
+  } else {  /* Try to pass argument in GPRs. */ \
+      if (ngpr + n <= maxgpr) { \
+        dp = &cc->gpr[ngpr]; \
+        ngpr += n; \
+        goto done; \
+    } \
+  }
+
 #else
 #error "Missing calling convention definitions for this architecture"
 #endif
@@ -1045,6 +1136,51 @@ static void ccall_copy_struct(CCallState *cc, CType *ctr, void *dp, void *sp,
 
 #endif
 
+/* -- RISC-V ABI struct classification ---------------------------- */
+
+#if LJ_TARGET_RISCV64
+
+static unsigned int ccall_classify_struct(CTState *cts, CType *ct)
+{
+  CTSize sz = ct->size;
+  unsigned int r = 0, n = 0, isu = (ct->info & CTF_UNION);
+  while (ct->sib) {
+    CType *sct;
+    ct = ctype_get(cts, ct->sib);
+    if (ctype_isfield(ct->info)) {
+      sct = ctype_rawchild(cts, ct);
+      if (ctype_isfp(sct->info)) {
+	r |= sct->size;
+	if (!isu) n++; else if (n == 0) n = 1;
+      } else if (ctype_iscomplex(sct->info)) {
+	r |= (sct->size >> 1);
+	if (!isu) n += 2; else if (n < 2) n = 2;
+      } else if (ctype_isstruct(sct->info)) {
+	goto substruct;
+      } else {
+	goto noth;
+      }
+    } else if (ctype_isbitfield(ct->info)) {
+      goto noth;
+    } else if (ctype_isxattrib(ct->info, CTA_SUBTYPE)) {
+      sct = ctype_rawchild(cts, ct);
+    substruct:
+      if (sct->size > 0) {
+	unsigned int s = ccall_classify_struct(cts, sct);
+	if (s <= 1) goto noth;
+	r |= (s & 255);
+	if (!isu) n += (s >> 8); else if (n < (s >>8)) n = (s >> 8);
+      }
+    }
+  }
+  if ((r == 4 || r == 8) && n <= 4)
+    return r + (n << 8);
+noth:  /* Not a homogeneous float/double aggregate. */
+  return (sz <= 16);  /* Return structs of size <= 16 in GPRs. */
+}
+
+#endif
+
 /* -- Common C call handling ---------------------------------------------- */
 
 /* Infer the destination CTypeID for a vararg argument. */
@@ -1091,6 +1227,10 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
 #endif
 #endif
 
+#if LJ_TARGET_RISCV64
+  int nff = 0;
+#endif
+
   /* Clear unused regs to get some determinism in case of misdeclaration. */
   memset(cc->gpr, 0, sizeof(cc->gpr));
 #if CCALL_NUM_FPR
@@ -1265,7 +1405,11 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
         *(int64_t *)dp = (int64_t)*(int32_t *)dp;
     }
 #endif
-#if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE)
+#if LJ_TARGET_RISCV64
+    if (isfp && d->size == sizeof(float))
+      ((uint32_t *)dp)[1] = 0xffffffffu;  /* Float NaN boxing */
+#endif
+#if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE) || LJ_TARGET_RISCV64
     if ((ctype_isinteger_or_bool(d->info) || ctype_isenum(d->info)
 #if LJ_TARGET_MIPS64
 	 || (isfp && nsp == 0)
@@ -1305,13 +1449,21 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
       CTSize i = (sz >> 2) - 1;
       do { ((uint64_t *)dp)[i] = ((uint32_t *)dp)[i]; } while (i--);
     }
+#elif LJ_TARGET_RISCV64
+    if (isfp == 2 && nff <= 2) {
+      /* Split complex float into separate registers. */
+      CTSize i = (sz >> 2) - 1;
+      do {
+        ((uint64_t *)dp)[i] = 0xffffffff00000000ul | ((uint32_t *)dp)[i];
+      } while (i--);
+    }
 #else
     UNUSED(isfp);
 #endif
   }
   if (fid) lj_err_caller(L, LJ_ERR_FFI_NUMARG);  /* Too few arguments. */
 
-#if LJ_TARGET_X64 || (LJ_TARGET_PPC && !LJ_ABI_SOFTFP)
+#if LJ_TARGET_X64 || (LJ_TARGET_PPC && !LJ_ABI_SOFTFP) || LJ_TARGET_RISCV64
   cc->nfpr = nfpr;  /* Required for vararg functions. */
 #endif
   cc->nsp = nsp;
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_ccall.h b/bundle/LuaJIT-2.1-20220411/src/lj_ccall.h
index 5245553..29985ff 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_ccall.h
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_ccall.h
@@ -154,6 +154,21 @@ typedef union FPRArg {
   float f;
 } FPRArg;
 
+#elif LJ_TARGET_RISCV64
+
+#define CCALL_NARG_GPR		8
+#define CCALL_NARG_FPR		8
+#define CCALL_NRET_GPR		2
+#define CCALL_NRET_FPR		2
+#define CCALL_SPS_EXTRA		3
+#define CCALL_SPS_FREE		1
+
+typedef intptr_t GPRArg;
+typedef union FPRArg {
+  double d;
+  struct { LJ_ENDIAN_LOHI(float f; , float g;) };
+} FPRArg;
+
 #else
 #error "Missing calling convention definitions for this architecture"
 #endif
@@ -196,7 +211,7 @@ typedef LJ_ALIGN(CCALL_ALIGN_CALLSTATE) struct CCallState {
   uint8_t resx87;		/* Result on x87 stack: 1:float, 2:double. */
 #elif LJ_TARGET_ARM64
   void *retp;			/* Aggregate return pointer in x8. */
-#elif LJ_TARGET_PPC
+#elif LJ_TARGET_PPC || LJ_TARGET_RISCV64
   uint8_t nfpr;			/* Number of arguments in FPRs. */
 #endif
 #if LJ_32
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_ccallback.c b/bundle/LuaJIT-2.1-20220411/src/lj_ccallback.c
index c1e67ab..0c3d80c 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_ccallback.c
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_ccallback.c
@@ -91,6 +91,10 @@ static MSize CALLBACK_OFS2SLOT(MSize ofs)
 
 #define CALLBACK_MCODE_HEAD		52
 
+#elif LJ_TARGET_RISCV64
+
+#define CALLBACK_MCODE_HEAD		68
+
 #else
 
 /* Missing support for this architecture. */
@@ -293,6 +297,39 @@ static void *callback_mcode_init(global_State *g, uint32_t *page)
   }
   return p;
 }
+#elif LJ_TARGET_RISCV64
+static void *callback_mcode_init(global_State *g, uint32_t *page)
+{
+  uint32_t *p = page;
+  uintptr_t target = (uintptr_t)(void *)lj_vm_ffi_callback;
+  uintptr_t ug = (uintptr_t)(void *)g;
+  uintptr_t target_hi = (target >> 32), target_lo = target & 0xffffffffULL;
+  uintptr_t ug_hi = (ug >> 32), ug_lo = ug & 0xffffffffULL;
+  MSize slot;
+  *p++ = RISCVI_LUI  | RISCVF_D(RID_X6) | RISCVF_IMMU(RISCVF_HI(target_hi));
+  *p++ = RISCVI_LUI  | RISCVF_D(RID_X7) | RISCVF_IMMU(RISCVF_HI(ug_hi));
+  *p++ = RISCVI_ADDI | RISCVF_D(RID_X6) | RISCVF_S1(RID_X6) | RISCVF_IMMI(RISCVF_LO(target_hi));
+  *p++ = RISCVI_ADDI | RISCVF_D(RID_X7) | RISCVF_S1(RID_X7) | RISCVF_IMMI(RISCVF_LO(ug_hi));
+  *p++ = RISCVI_SLLI | RISCVF_D(RID_X6) | RISCVF_S1(RID_X6) | RISCVF_SHAMT(11);
+  *p++ = RISCVI_SLLI | RISCVF_D(RID_X7) | RISCVF_S1(RID_X7) | RISCVF_SHAMT(11);
+  *p++ = RISCVI_ADDI | RISCVF_D(RID_X6) | RISCVF_S1(RID_X6) | RISCVF_IMMI(target_lo >> 21);
+  *p++ = RISCVI_ADDI | RISCVF_D(RID_X7) | RISCVF_S1(RID_X7) | RISCVF_IMMI(ug_lo >> 21);
+  *p++ = RISCVI_SLLI | RISCVF_D(RID_X6) | RISCVF_S1(RID_X6) | RISCVF_SHAMT(11);
+  *p++ = RISCVI_SLLI | RISCVF_D(RID_X7) | RISCVF_S1(RID_X7) | RISCVF_SHAMT(11);
+  *p++ = RISCVI_ADDI | RISCVF_D(RID_X6) | RISCVF_S1(RID_X6) | RISCVF_IMMI((target_lo >> 10) & 0x7ff);
+  *p++ = RISCVI_ADDI | RISCVF_D(RID_X7) | RISCVF_S1(RID_X7) | RISCVF_IMMI((ug_lo >> 10) & 0x7ff);
+  *p++ = RISCVI_SLLI | RISCVF_D(RID_X6) | RISCVF_S1(RID_X6) | RISCVF_SHAMT(10);
+  *p++ = RISCVI_SLLI | RISCVF_D(RID_X7) | RISCVF_S1(RID_X7) | RISCVF_SHAMT(10);
+  *p++ = RISCVI_ADDI | RISCVF_D(RID_X6) | RISCVF_S1(RID_X6) | RISCVF_IMMI(target_lo & 0x3ff);
+  *p++ = RISCVI_ADDI | RISCVF_D(RID_X7) | RISCVF_S1(RID_X7) | RISCVF_IMMI(ug_lo & 0x3ff);
+  *p++ = RISCVI_JALR | RISCVF_D(RID_X0) | RISCVF_S1(RID_X6) | RISCVF_IMMJ(0);
+  for (slot = 0; slot < CALLBACK_MAX_SLOT; slot++) {
+    *p++ = RISCVI_LUI | RISCVF_D(RID_X5) | RISCVF_IMMU(slot);
+    *p = RISCVI_JAL | RISCVF_IMMJ(((char *)page-(char *)p));
+    p++;
+  }
+  return p;
+}
 #else
 /* Missing support for this architecture. */
 #define callback_mcode_init(g, p)	(p)
@@ -567,6 +604,31 @@ void lj_ccallback_mcode_free(CTState *cts)
   }
 #endif
 
+#define CALLBACK_HANDLE_RET \
+  if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
+    ((float *)dp)[1] = *(float *)dp;
+
+#elif LJ_TARGET_RISCV64
+
+#define CALLBACK_HANDLE_REGARG \
+  if (isfp) { \
+    if (nfpr + n <= CCALL_NARG_FPR) { \
+      sp = &cts->cb.fpr[nfpr]; \
+      nfpr += n; \
+      goto done; \
+    } else if (ngpr + n <= maxgpr) { \
+      sp = &cts->cb.gpr[ngpr]; \
+      ngpr += n; \
+      goto done; \
+    } \
+  } else { \
+    if (ngpr + n <= maxgpr) { \
+      sp = &cts->cb.gpr[ngpr]; \
+      ngpr += n; \
+      goto done; \
+    } \
+  }
+
 #define CALLBACK_HANDLE_RET \
   if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
     ((float *)dp)[1] = *(float *)dp;
@@ -735,7 +797,7 @@ static void callback_conv_result(CTState *cts, lua_State *L, TValue *o)
         *(int64_t *)dp = (int64_t)*(int32_t *)dp;
     }
 #endif
-#if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE)
+#if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE) || LJ_TARGET_RISCV64
     /* Always sign-extend results to 64 bits. Even a soft-fp 'float'. */
     if (ctr->size <= 4 &&
 	(LJ_ABI_SOFTFP || ctype_isinteger_or_bool(ctr->info)))
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_dispatch.c b/bundle/LuaJIT-2.1-20220411/src/lj_dispatch.c
index ded382a..32688a0 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_dispatch.c
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_dispatch.c
@@ -56,6 +56,15 @@ static const ASMFunction dispatch_got[] = {
 #undef GOTFUNC
 #endif
 
+#if LJ_TARGET_RISCV64
+#include <math.h>
+#define GOTFUNC(name)	(ASMFunction)name,
+static const ASMFunction dispatch_got[] = {
+  GOTDEF(GOTFUNC)
+};
+#undef GOTFUNC
+#endif
+
 /* Initialize instruction dispatch table and hot counters. */
 void lj_dispatch_init(GG_State *GG)
 {
@@ -76,7 +85,7 @@ void lj_dispatch_init(GG_State *GG)
   GG->g.bc_cfunc_ext = GG->g.bc_cfunc_int = BCINS_AD(BC_FUNCC, LUA_MINSTACK, 0);
   for (i = 0; i < GG_NUM_ASMFF; i++)
     GG->bcff[i] = BCINS_AD(BC__MAX+i, 0, 0);
-#if LJ_TARGET_MIPS
+#if LJ_TARGET_MIPS || LJ_TARGET_RISCV64
   memcpy(GG->got, dispatch_got, LJ_GOT__MAX*sizeof(ASMFunction *));
 #endif
 }
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_dispatch.h b/bundle/LuaJIT-2.1-20220411/src/lj_dispatch.h
index 0594af5..37407f4 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_dispatch.h
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_dispatch.h
@@ -66,6 +66,35 @@ GOTDEF(GOTENUM)
 };
 #endif
 
+#if LJ_TARGET_RISCV64
+/* Need our own global offset table to wrap RISC-V PIC intern / extern calls */
+
+#if LJ_HASJIT
+#define JITGOTDEF(_)	_(lj_err_trace) _(lj_trace_exit) _(lj_trace_hot)
+#else
+#define JITGOTDEF(_)
+#endif
+#if LJ_HASFFI
+#define FFIGOTDEF(_) \
+  _(lj_meta_equal_cd) _(lj_ccallback_enter) _(lj_ccallback_leave)
+#else
+#define FFIGOTDEF(_)
+#endif
+
+#define GOTDEF(_) \
+  _(floor) _(ceil) _(trunc) _(log) _(log10) _(exp) _(sin) _(cos) _(tan) \
+  _(asin) _(acos) _(atan) _(sinh) _(cosh) _(tanh) _(frexp) _(modf) _(atan2) \
+  _(pow) _(fmod) _(ldexp) \
+  JITGOTDEF(_) FFIGOTDEF(_)
+
+enum {
+#define GOTENUM(name) LJ_GOT_##name,
+GOTDEF(GOTENUM)
+#undef GOTENUM
+  LJ_GOT__MAX
+};
+#endif
+
 /* Type of hot counter. Must match the code in the assembler VM. */
 /* 16 bits are sufficient. Only 0.0015% overhead with maximum slot penalty. */
 typedef uint16_t HotCount;
@@ -93,7 +122,7 @@ typedef struct GG_State {
   /* Make g reachable via K12 encoded DISPATCH-relative addressing. */
   uint8_t align1[(16-sizeof(global_State))&15];
 #endif
-#if LJ_TARGET_MIPS
+#if LJ_TARGET_MIPS || LJ_TARGET_RISCV64
   ASMFunction got[LJ_GOT__MAX];		/* Global offset table. */
 #endif
 #if LJ_HASJIT
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_frame.h b/bundle/LuaJIT-2.1-20220411/src/lj_frame.h
index 4058311..26b8081 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_frame.h
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_frame.h
@@ -287,6 +287,15 @@ enum { LJ_CONT_TAILCALL, LJ_CONT_FFI_CALLBACK };  /* Special continuations. */
 ** need to change to 3.
 */
 #define CFRAME_SHIFT_MULTRES	0
+#elif LJ_TARGET_RISCV64
+#define CFRAME_OFS_ERRF		252
+#define CFRAME_OFS_NRES		248
+#define CFRAME_OFS_PREV		240
+#define CFRAME_OFS_L		232
+#define CFRAME_OFS_PC		224
+#define CFRAME_OFS_MULTRES	0
+#define CFRAME_SIZE		256
+#define CFRAME_SHIFT_MULTRES	3
 #else
 #error "Missing CFRAME_* definitions for this architecture"
 #endif
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_jit.h b/bundle/LuaJIT-2.1-20220411/src/lj_jit.h
index 74b40fd..23d027b 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_jit.h
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_jit.h
@@ -67,6 +67,15 @@
 #endif
 #endif
 
+#elif LJ_TARGET_RISCV64
+
+#define JIT_F_RVC		(JIT_F_CPU << 0)
+#define JIT_F_RVZba		(JIT_F_CPU << 1)
+#define JIT_F_RVZbb		(JIT_F_CPU << 2)
+#define JIT_F_RVXThead		(JIT_F_CPU << 3)
+
+#define JIT_F_CPUSTRING		"\003RVC\003Zba\003Zbb\006XThead"
+
 #else
 
 #define JIT_F_CPUSTRING		""
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_target.h b/bundle/LuaJIT-2.1-20220411/src/lj_target.h
index 3831cb6..4534567 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_target.h
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_target.h
@@ -55,7 +55,7 @@ typedef uint32_t RegSP;
 /* Bitset for registers. 32 registers suffice for most architectures.
 ** Note that one set holds bits for both GPRs and FPRs.
 */
-#if LJ_TARGET_PPC || LJ_TARGET_MIPS || LJ_TARGET_ARM64
+#if LJ_TARGET_PPC || LJ_TARGET_MIPS || LJ_TARGET_ARM64 || LJ_TARGET_RISCV64
 typedef uint64_t RegSet;
 #else
 typedef uint32_t RegSet;
@@ -69,7 +69,7 @@ typedef uint32_t RegSet;
 #define rset_set(rs, r)		(rs |= RID2RSET(r))
 #define rset_clear(rs, r)	(rs &= ~RID2RSET(r))
 #define rset_exclude(rs, r)	(rs & ~RID2RSET(r))
-#if LJ_TARGET_PPC || LJ_TARGET_MIPS || LJ_TARGET_ARM64
+#if LJ_TARGET_PPC || LJ_TARGET_MIPS || LJ_TARGET_ARM64 || LJ_TARGET_RISCV64
 #define rset_picktop(rs)	((Reg)(__builtin_clzll(rs)^63))
 #define rset_pickbot(rs)	((Reg)__builtin_ctzll(rs))
 #else
@@ -146,6 +146,8 @@ typedef uint32_t RegCost;
 #include "lj_target_mips.h"
 #elif LJ_TARGET_S390X
 #include "lj_target_s390x.h"
+#elif LJ_TARGET_RISCV64
+#include "lj_target_riscv.h"
 #else
 #error "Missing include for target CPU"
 #endif
diff --git a/bundle/LuaJIT-2.1-20220411/src/lj_vmmath.c b/bundle/LuaJIT-2.1-20220411/src/lj_vmmath.c
index b6cc60b..2186621 100644
--- a/bundle/LuaJIT-2.1-20220411/src/lj_vmmath.c
+++ b/bundle/LuaJIT-2.1-20220411/src/lj_vmmath.c
@@ -58,7 +58,8 @@ double lj_vm_foldarith(double x, double y, int op)
 
 /* -- Helper functions for generated machine code ------------------------- */
 
-#if (LJ_HASJIT && !(LJ_TARGET_ARM || LJ_TARGET_ARM64 || LJ_TARGET_PPC)) || LJ_TARGET_MIPS
+#if (LJ_HASJIT && !(LJ_TARGET_ARM || LJ_TARGET_ARM64 || LJ_TARGET_PPC)) || LJ_TARGET_MIPS \
+ || LJ_TARGET_RISCV64
 int32_t LJ_FASTCALL lj_vm_modi(int32_t a, int32_t b)
 {
   uint32_t y, ua, ub;