diff --git a/src/pkg/runtime/vlop_386.s b/src/pkg/runtime/vlop_386.s
index edc659b848..9783fdc936 100644
--- a/src/pkg/runtime/vlop_386.s
+++ b/src/pkg/runtime/vlop_386.s
@@ -29,6 +29,8 @@
  * C runtime for 64-bit divide.
  */
 
+// _mul64x32(r *uint64, a uint64, b uint32)
+// sets *r = low 64 bits of 96-bit product a*b; returns high 32 bits.
 TEXT _mul64by32(SB), NOSPLIT, $0
 	MOVL	r+0(FP), CX
 	MOVL	a+4(FP), AX
@@ -38,7 +40,9 @@ TEXT _mul64by32(SB), NOSPLIT, $0
 	MOVL	a+8(FP), AX
 	MULL	b+12(FP)
 	ADDL	AX, BX
+	ADCL	$0, DX
 	MOVL	BX, 4(CX)
+	MOVL	DX, AX
 	RET
 
 TEXT _div64by32(SB), NOSPLIT, $0
diff --git a/src/pkg/runtime/vlrt_386.c b/src/pkg/runtime/vlrt_386.c
index d8bc94bd94..8d965c086e 100644
--- a/src/pkg/runtime/vlrt_386.c
+++ b/src/pkg/runtime/vlrt_386.c
@@ -147,7 +147,7 @@ _v2f(Vlong x)
 }
 
 ulong	_div64by32(Vlong, ulong, ulong*);
-void	_mul64by32(Vlong*, Vlong, ulong);
+int	_mul64by32(Vlong*, Vlong, ulong);
 
 static void
 slowdodiv(Vlong num, Vlong den, Vlong *q, Vlong *r)
@@ -232,8 +232,7 @@ dodiv(Vlong num, Vlong den, Vlong *qp, Vlong *rp)
 	if(den.hi != 0){
 		q.hi = 0;
 		n = num.hi/den.hi;
-		_mul64by32(&x, den, n);
-		if(x.hi > num.hi || (x.hi == num.hi && x.lo > num.lo))
+		if(_mul64by32(&x, den, n) || x.hi > num.hi || (x.hi == num.hi && x.lo > num.lo))
 			slowdodiv(num, den, &q, &r);
 		else {
 			q.lo = n;