immunio 1.2.1 → 2.0.2

data/lua-hooks/ext/luajit/src/lj_bc.c

@@ -1,6 +1,6 @@
 /*
 ** Bytecode instruction modes.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #define lj_bc_c

data/lua-hooks/ext/luajit/src/lj_bc.h

@@ -1,6 +1,6 @@
 /*
 ** Bytecode instruction format.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #ifndef _LJ_BC_H

data/lua-hooks/ext/luajit/src/lj_bcdump.h

@@ -1,6 +1,6 @@
 /*
 ** Bytecode dump definitions.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #ifndef _LJ_BCDUMP_H

data/lua-hooks/ext/luajit/src/lj_bcread.c

@@ -1,6 +1,6 @@
 /*
 ** Bytecode reader.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #define lj_bcread_c

data/lua-hooks/ext/luajit/src/lj_bcwrite.c

@@ -1,6 +1,6 @@
 /*
 ** Bytecode writer.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #define lj_bcwrite_c

data/lua-hooks/ext/luajit/src/lj_buf.c

@@ -1,6 +1,6 @@
 /*
 ** Buffer handling.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #define lj_buf_c

data/lua-hooks/ext/luajit/src/lj_buf.h

@@ -1,6 +1,6 @@
 /*
 ** Buffer handling.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #ifndef _LJ_BUF_H

data/lua-hooks/ext/luajit/src/lj_carith.c

@@ -1,6 +1,6 @@
 /*
 ** C data arithmetic.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #include "lj_obj.h"

data/lua-hooks/ext/luajit/src/lj_carith.h

@@ -1,6 +1,6 @@
 /*
 ** C data arithmetic.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #ifndef _LJ_CARITH_H

data/lua-hooks/ext/luajit/src/lj_ccall.c

@@ -1,6 +1,6 @@
 /*
 ** FFI C call handling.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #include "lj_obj.h"
@@ -301,7 +301,7 @@
   unsigned int cl = ccall_classify_struct(cts, ctr); \
   if ((cl & 4)) { /* Combine float HFA from separate registers. */ \
     CTSize i = (cl >> 8) - 1; \
-    do { ((uint32_t *)dp)[i] = cc->fpr[i].u32; } while (i--); \
+    do { ((uint32_t *)dp)[i] = cc->fpr[i].lo; } while (i--); \
   } else { \
     if (cl > 1) sp = (uint8_t *)&cc->fpr[0]; \
     memcpy(dp, sp, ctr->size); \
@@ -331,7 +331,7 @@
 
 #define CCALL_HANDLE_COMPLEXARG \
   /* Pass complex by value in separate (!) FPRs or on stack. */ \
-  isfp = ctr->size == 2*sizeof(float) ? 2 : 1;
+  isfp = sz == 2*sizeof(float) ? 2 : 1;
 
 #define CCALL_HANDLE_REGARG \
   if (LJ_TARGET_IOS && isva) { \
@@ -359,6 +359,13 @@
     } \
   }
 
+#if LJ_BE
+#define CCALL_HANDLE_RET \
+  if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
+    sp = (uint8_t *)&cc->fpr[0].f;
+#endif
+
+
 #elif LJ_TARGET_PPC
 /* -- PPC calling conventions --------------------------------------------- */
 
@@ -407,8 +414,8 @@
   if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
     ctr = ctype_get(cts, CTID_DOUBLE);  /* FPRs always hold doubles. */
 
-#elif LJ_TARGET_MIPS
-/* -- MIPS calling conventions -------------------------------------------- */
+#elif LJ_TARGET_MIPS32
+/* -- MIPS o32 calling conventions ---------------------------------------- */
 
 #define CCALL_HANDLE_STRUCTRET \
   cc->retref = 1;  /* Return all structs by reference. */ \
@@ -483,6 +490,78 @@
     sp = (uint8_t *)&cc->fpr[0].f;
 #endif
 
+#elif LJ_TARGET_MIPS64
+/* -- MIPS n64 calling conventions ---------------------------------------- */
+
+#define CCALL_HANDLE_STRUCTRET \
+  cc->retref = !(sz <= 16); \
+  if (cc->retref) cc->gpr[ngpr++] = (GPRArg)dp;
+
+#define CCALL_HANDLE_STRUCTRET2 \
+  ccall_copy_struct(cc, ctr, dp, sp, ccall_classify_struct(cts, ctr, ct));
+
+#define CCALL_HANDLE_COMPLEXRET \
+  /* Complex values are returned in 1 or 2 FPRs. */ \
+  cc->retref = 0;
+
+#if LJ_ABI_SOFTFP	/* MIPS64 soft-float */
+
+#define CCALL_HANDLE_COMPLEXRET2 \
+  if (ctr->size == 2*sizeof(float)) {  /* Copy complex float from GPRs. */ \
+    ((intptr_t *)dp)[0] = cc->gpr[0]; \
+  } else {  /* Copy complex double from GPRs. */ \
+    ((intptr_t *)dp)[0] = cc->gpr[0]; \
+    ((intptr_t *)dp)[1] = cc->gpr[1]; \
+  }
+
+#define CCALL_HANDLE_COMPLEXARG \
+  /* Pass complex by value in 2 or 4 GPRs. */
+
+/* Position of soft-float 'float' return value depends on endianess.  */
+#define CCALL_HANDLE_RET \
+  if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
+    sp = (uint8_t *)cc->gpr + LJ_ENDIAN_SELECT(0, 4);
+
+#else			/* MIPS64 hard-float */
+
+#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 \
+  if (sz == 2*sizeof(float)) { \
+    isfp = 2; \
+    if (ngpr < maxgpr) \
+      sz *= 2; \
+  }
+
+#define CCALL_HANDLE_RET \
+  if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
+    sp = (uint8_t *)&cc->fpr[0].f;
+
+#endif
+
+#define CCALL_HANDLE_STRUCTARG \
+  /* Pass all structs by value in registers and/or on the stack. */
+
+#define CCALL_HANDLE_REGARG \
+  if (ngpr < maxgpr) { \
+    dp = &cc->gpr[ngpr]; \
+    if (ngpr + n > maxgpr) { \
+      nsp += ngpr + n - maxgpr;  /* Assumes contiguous gpr/stack fields. */ \
+      if (nsp > CCALL_MAXSTACK) goto err_nyi;  /* Too many arguments. */ \
+      ngpr = maxgpr; \
+    } else { \
+      ngpr += n; \
+    } \
+    goto done; \
+  }
+
 #else
 #error "Missing calling convention definitions for this architecture"
 #endif
@@ -722,6 +801,78 @@ noth:  /* Not a homogeneous float/double aggregate. */
 
 #endif
 
+/* -- MIPS64 ABI struct classification ---------------------------- */
+
+#if LJ_TARGET_MIPS64
+
+#define FTYPE_FLOAT	1
+#define FTYPE_DOUBLE	2
+
+/* Classify FP fields (max. 2) and their types. */
+static unsigned int ccall_classify_struct(CTState *cts, CType *ct, CType *ctf)
+{
+  int n = 0, ft = 0;
+  if ((ctf->info & CTF_VARARG) || (ct->info & CTF_UNION))
+    goto noth;
+  while (ct->sib) {
+    CType *sct;
+    ct = ctype_get(cts, ct->sib);
+    if (n == 2) {
+      goto noth;
+    } else if (ctype_isfield(ct->info)) {
+      sct = ctype_rawchild(cts, ct);
+      if (ctype_isfp(sct->info)) {
+	ft |= (sct->size == 4 ? FTYPE_FLOAT : FTYPE_DOUBLE) << 2*n;
+	n++;
+      } else {
+	goto noth;
+      }
+    } else if (ctype_isbitfield(ct->info) ||
+	       ctype_isxattrib(ct->info, CTA_SUBTYPE)) {
+      goto noth;
+    }
+  }
+  if (n <= 2)
+    return ft;
+noth:  /* Not a homogeneous float/double aggregate. */
+  return 0;  /* Struct is in GPRs. */
+}
+
+void ccall_copy_struct(CCallState *cc, CType *ctr, void *dp, void *sp, int ft)
+{
+  if (LJ_ABI_SOFTFP ? ft :
+      ((ft & 3) == FTYPE_FLOAT || (ft >> 2) == FTYPE_FLOAT)) {
+    int i, ofs = 0;
+    for (i = 0; ft != 0; i++, ft >>= 2) {
+      if ((ft & 3) == FTYPE_FLOAT) {
+#if LJ_ABI_SOFTFP
+	/* The 2nd FP struct result is in CARG1 (gpr[2]) and not CRET2. */
+	memcpy((uint8_t *)dp + ofs,
+	       (uint8_t *)&cc->gpr[2*i] + LJ_ENDIAN_SELECT(0, 4), 4);
+#else
+	*(float *)((uint8_t *)dp + ofs) = cc->fpr[i].f;
+#endif
+	ofs += 4;
+      } else {
+	ofs = (ofs + 7) & ~7;  /* 64 bit alignment. */
+#if LJ_ABI_SOFTFP
+	*(intptr_t *)((uint8_t *)dp + ofs) = cc->gpr[2*i];
+#else
+	*(double *)((uint8_t *)dp + ofs) = cc->fpr[i].d;
+#endif
+	ofs += 8;
+      }
+    }
+  } else {
+#if !LJ_ABI_SOFTFP
+    if (ft) sp = (uint8_t *)&cc->fpr[0];
+#endif
+    memcpy(dp, sp, ctr->size);
+  }
+}
+
+#endif
+
 /* -- Common C call handling ---------------------------------------------- */
 
 /* Infer the destination CTypeID for a vararg argument. */
@@ -889,6 +1040,19 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
 	*(int32_t *)dp = d->size == 1 ? (int32_t)*(int8_t *)dp :
 					(int32_t)*(int16_t *)dp;
     }
+#if LJ_TARGET_ARM64 && LJ_BE
+    if (isfp && d->size == sizeof(float))
+      ((float *)dp)[1] = ((float *)dp)[0];  /* Floats occupy high slot. */
+#endif
+#if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE)
+    if ((ctype_isinteger_or_bool(d->info) || ctype_isenum(d->info)
+#if LJ_TARGET_MIPS64
+	 || (isfp && nsp == 0)
+#endif
+	 ) && d->size <= 4) {
+      *(int64_t *)dp = (int64_t)*(int32_t *)dp;  /* Sign-extend to 64 bit. */
+    }
+#endif
 #if LJ_TARGET_X64 && LJ_ABI_WIN
     if (isva) {  /* Windows/x64 mirrors varargs in both register sets. */
       if (nfpr == ngpr)
@@ -904,7 +1068,7 @@ static int ccall_set_args(lua_State *L, CTState *cts, CType *ct,
       cc->fpr[nfpr-1].d[0] = cc->fpr[nfpr-2].d[1];  /* Split complex double. */
       cc->fpr[nfpr-2].d[1] = 0;
     }
-#elif LJ_TARGET_ARM64
+#elif LJ_TARGET_ARM64 || (LJ_TARGET_MIPS64 && !LJ_ABI_SOFTFP)
     if (isfp == 2 && (uint8_t *)dp < (uint8_t *)cc->stack) {
       /* Split float HFA or complex float into separate registers. */
       CTSize i = (sz >> 2) - 1;
@@ -951,7 +1115,8 @@ static int ccall_get_results(lua_State *L, CTState *cts, CType *ct,
     CCALL_HANDLE_COMPLEXRET2
     return 1;  /* One GC step. */
   }
-  if (LJ_BE && ctype_isinteger_or_bool(ctr->info) && ctr->size < CTSIZE_PTR)
+  if (LJ_BE && ctr->size < CTSIZE_PTR &&
+      (ctype_isinteger_or_bool(ctr->info) || ctype_isenum(ctr->info)))
     sp += (CTSIZE_PTR - ctr->size);
 #if CCALL_NUM_FPR
   if (ctype_isfp(ctr->info) || ctype_isvector(ctr->info))

data/lua-hooks/ext/luajit/src/lj_ccall.h

@@ -1,6 +1,6 @@
 /*
 ** FFI C call handling.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #ifndef _LJ_CCALL_H
@@ -79,8 +79,8 @@ typedef union FPRArg {
 typedef intptr_t GPRArg;
 typedef union FPRArg {
   double d;
-  float f;
-  uint32_t u32;
+  struct { LJ_ENDIAN_LOHI(float f; , float g;) };
+  struct { LJ_ENDIAN_LOHI(uint32_t lo; , uint32_t hi;) };
 } FPRArg;
 
 #elif LJ_TARGET_PPC
@@ -95,11 +95,11 @@ typedef union FPRArg {
 typedef intptr_t GPRArg;
 typedef double FPRArg;
 
-#elif LJ_TARGET_MIPS
+#elif LJ_TARGET_MIPS32
 
 #define CCALL_NARG_GPR		4
 #define CCALL_NARG_FPR		(LJ_ABI_SOFTFP ? 0 : 2)
-#define CCALL_NRET_GPR		2
+#define CCALL_NRET_GPR		(LJ_ABI_SOFTFP ? 4 : 2)
 #define CCALL_NRET_FPR		(LJ_ABI_SOFTFP ? 0 : 2)
 #define CCALL_SPS_EXTRA		7
 #define CCALL_SPS_FREE		1
@@ -110,6 +110,22 @@ typedef union FPRArg {
   struct { LJ_ENDIAN_LOHI(float f; , float g;) };
 } FPRArg;
 
+#elif LJ_TARGET_MIPS64
+
+/* FP args are positional and overlay the GPR array. */
+#define CCALL_NARG_GPR		8
+#define CCALL_NARG_FPR		0
+#define CCALL_NRET_GPR		2
+#define CCALL_NRET_FPR		(LJ_ABI_SOFTFP ? 0 : 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

data/lua-hooks/ext/luajit/src/lj_ccallback.c

@@ -1,6 +1,6 @@
 /*
 ** FFI C callback handling.
-** Copyright (C) 2005-2016 Mike Pall. See Copyright Notice in luajit.h
+** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
 */
 
 #include "lj_obj.h"
@@ -63,9 +63,13 @@ static MSize CALLBACK_OFS2SLOT(MSize ofs)
 
 #define CALLBACK_MCODE_HEAD		24
 
-#elif LJ_TARGET_MIPS
+#elif LJ_TARGET_MIPS32
 
-#define CALLBACK_MCODE_HEAD		24
+#define CALLBACK_MCODE_HEAD		20
+
+#elif LJ_TARGET_MIPS64
+
+#define CALLBACK_MCODE_HEAD		52
 
 #else
 
@@ -169,16 +173,16 @@ static void callback_mcode_init(global_State *g, uint32_t *page)
   uint32_t *p = page;
   void *target = (void *)lj_vm_ffi_callback;
   MSize slot;
-  *p++ = A64I_LDRLx | A64F_D(RID_X11) | A64F_S19(4);
-  *p++ = A64I_LDRLx | A64F_D(RID_X10) | A64F_S19(5);
-  *p++ = A64I_BR | A64F_N(RID_X11);
-  *p++ = A64I_NOP;
+  *p++ = A64I_LE(A64I_LDRLx | A64F_D(RID_X11) | A64F_S19(4));
+  *p++ = A64I_LE(A64I_LDRLx | A64F_D(RID_X10) | A64F_S19(5));
+  *p++ = A64I_LE(A64I_BR | A64F_N(RID_X11));
+  *p++ = A64I_LE(A64I_NOP);
   ((void **)p)[0] = target;
   ((void **)p)[1] = g;
   p += 4;
   for (slot = 0; slot < CALLBACK_MAX_SLOT; slot++) {
-    *p++ = A64I_MOVZw | A64F_D(RID_X9) | A64F_U16(slot);
-    *p = A64I_B | A64F_S26((page-p) & 0x03ffffffu);
+    *p++ = A64I_LE(A64I_MOVZw | A64F_D(RID_X9) | A64F_U16(slot));
+    *p = A64I_LE(A64I_B | A64F_S26((page-p) & 0x03ffffffu));
     p++;
   }
   lua_assert(p - page <= CALLBACK_MCODE_SIZE);
@@ -206,14 +210,27 @@ static void callback_mcode_init(global_State *g, uint32_t *page)
 static void callback_mcode_init(global_State *g, uint32_t *page)
 {
   uint32_t *p = page;
-  void *target = (void *)lj_vm_ffi_callback;
+  uintptr_t target = (uintptr_t)(void *)lj_vm_ffi_callback;
+  uintptr_t ug = (uintptr_t)(void *)g;
   MSize slot;
-  *p++ = MIPSI_SW | MIPSF_T(RID_R1)|MIPSF_S(RID_SP) | 0;
-  *p++ = MIPSI_LUI | MIPSF_T(RID_R3) | (u32ptr(target) >> 16);
-  *p++ = MIPSI_LUI | MIPSF_T(RID_R2) | (u32ptr(g) >> 16);
-  *p++ = MIPSI_ORI | MIPSF_T(RID_R3)|MIPSF_S(RID_R3) |(u32ptr(target)&0xffff);
+#if LJ_TARGET_MIPS32
+  *p++ = MIPSI_LUI | MIPSF_T(RID_R3) | (target >> 16);
+  *p++ = MIPSI_LUI | MIPSF_T(RID_R2) | (ug >> 16);
+#else
+  *p++ = MIPSI_LUI  | MIPSF_T(RID_R3) | (target >> 48);
+  *p++ = MIPSI_LUI  | MIPSF_T(RID_R2) | (ug >> 48);
+  *p++ = MIPSI_ORI  | MIPSF_T(RID_R3)|MIPSF_S(RID_R3) | ((target >> 32) & 0xffff);
+  *p++ = MIPSI_ORI  | MIPSF_T(RID_R2)|MIPSF_S(RID_R2) | ((ug >> 32) & 0xffff);
+  *p++ = MIPSI_DSLL | MIPSF_D(RID_R3)|MIPSF_T(RID_R3) | MIPSF_A(16);
+  *p++ = MIPSI_DSLL | MIPSF_D(RID_R2)|MIPSF_T(RID_R2) | MIPSF_A(16);
+  *p++ = MIPSI_ORI  | MIPSF_T(RID_R3)|MIPSF_S(RID_R3) | ((target >> 16) & 0xffff);
+  *p++ = MIPSI_ORI  | MIPSF_T(RID_R2)|MIPSF_S(RID_R2) | ((ug >> 16) & 0xffff);
+  *p++ = MIPSI_DSLL | MIPSF_D(RID_R3)|MIPSF_T(RID_R3) | MIPSF_A(16);
+  *p++ = MIPSI_DSLL | MIPSF_D(RID_R2)|MIPSF_T(RID_R2) | MIPSF_A(16);
+#endif
+  *p++ = MIPSI_ORI  | MIPSF_T(RID_R3)|MIPSF_S(RID_R3) | (target & 0xffff);
   *p++ = MIPSI_JR | MIPSF_S(RID_R3);
-  *p++ = MIPSI_ORI | MIPSF_T(RID_R2)|MIPSF_S(RID_R2) | (u32ptr(g)&0xffff);
+  *p++ = MIPSI_ORI | MIPSF_T(RID_R2)|MIPSF_S(RID_R2) | (ug & 0xffff);
   for (slot = 0; slot < CALLBACK_MAX_SLOT; slot++) {
     *p = MIPSI_B | ((page-p-1) & 0x0000ffffu);
     p++;
@@ -425,7 +442,7 @@ void lj_ccallback_mcode_free(CTState *cts)
   if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
     *(double *)dp = *(float *)dp;  /* FPRs always hold doubles. */
 
-#elif LJ_TARGET_MIPS
+#elif LJ_TARGET_MIPS32
 
 #define CALLBACK_HANDLE_GPR \
   if (n > 1) ngpr = (ngpr + 1u) & ~1u;  /* Align to regpair. */ \
@@ -451,6 +468,29 @@ void lj_ccallback_mcode_free(CTState *cts)
   UNUSED(isfp);
 #endif
 
+#define CALLBACK_HANDLE_RET \
+  if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
+    ((float *)dp)[1] = *(float *)dp;
+
+#elif LJ_TARGET_MIPS64
+
+#if !LJ_ABI_SOFTFP	/* MIPS64 hard-float */
+#define CALLBACK_HANDLE_REGARG \
+  if (ngpr + n <= maxgpr) { \
+    sp = isfp ? (void*) &cts->cb.fpr[ngpr] : (void*) &cts->cb.gpr[ngpr]; \
+    ngpr += n; \
+    goto done; \
+  }
+#else			/* MIPS64 soft-float */
+#define CALLBACK_HANDLE_REGARG \
+  if (ngpr + n <= maxgpr) { \
+    UNUSED(isfp); \
+    sp = (void*) &cts->cb.gpr[ngpr]; \
+    ngpr += n; \
+    goto done; \
+  }
+#endif
+
 #define CALLBACK_HANDLE_RET \
   if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
     ((float *)dp)[1] = *(float *)dp;
@@ -542,7 +582,11 @@ static void callback_conv_args(CTState *cts, lua_State *L)
       nsp += n;
 
     done:
-      if (LJ_BE && cta->size < CTSIZE_PTR)
+      if (LJ_BE && cta->size < CTSIZE_PTR
+#if LJ_TARGET_MIPS64
+	  && !(isfp && nsp)
+#endif
+	 )
 	sp = (void *)((uint8_t *)sp + CTSIZE_PTR-cta->size);
       gcsteps += lj_cconv_tv_ct(cts, cta, 0, o++, sp);
     }
@@ -579,6 +623,10 @@ static void callback_conv_result(CTState *cts, lua_State *L, TValue *o)
 #if CCALL_NUM_FPR
     if (ctype_isfp(ctr->info))
       dp = (uint8_t *)&cts->cb.fpr[0];
+#endif
+#if LJ_TARGET_ARM64 && LJ_BE
+    if (ctype_isfp(ctr->info) && ctr->size == sizeof(float))
+      dp = (uint8_t *)&cts->cb.fpr[0].f[1];
 #endif
     lj_cconv_ct_tv(cts, ctr, dp, o, 0);
 #ifdef CALLBACK_HANDLE_RET
@@ -593,6 +641,12 @@ static void callback_conv_result(CTState *cts, lua_State *L, TValue *o)
 	*(int32_t *)dp = ctr->size == 1 ? (int32_t)*(int8_t *)dp :
 					  (int32_t)*(int16_t *)dp;
     }
+#if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE)
+    /* 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)))
+      *(int64_t *)dp = (int64_t)*(int32_t *)dp;
+#endif
 #if LJ_TARGET_X86
     if (ctype_isfp(ctr->info))
       cts->cb.gpr[2] = ctr->size == sizeof(float) ? 1 : 2;