diff --git a/src/cmd/internal/obj/arm/obj5.go b/src/cmd/internal/obj/arm/obj5.go
index 752f577dc3..ce1cad9c47 100644
--- a/src/cmd/internal/obj/arm/obj5.go
+++ b/src/cmd/internal/obj/arm/obj5.go
@@ -795,38 +795,45 @@ func stacksplit(ctxt *obj.Link, p *obj.Prog, framesize int32) *obj.Prog {
 		p.Scond = C_SCOND_NE
 	}
 
-	// MOVW.LS	R14, R3
-	p = obj.Appendp(ctxt, p)
+	// BLS call-to-morestack
+	bls := obj.Appendp(ctxt, p)
+	bls.As = ABLS
+	bls.To.Type = obj.TYPE_BRANCH
 
-	p.As = AMOVW
-	p.Scond = C_SCOND_LS
-	p.From.Type = obj.TYPE_REG
-	p.From.Reg = REGLINK
-	p.To.Type = obj.TYPE_REG
-	p.To.Reg = REG_R3
-
-	// BL.LS		runtime.morestack(SB) // modifies LR, returns with LO still asserted
-	p = obj.Appendp(ctxt, p)
-
-	p.As = ABL
-	p.Scond = C_SCOND_LS
-	p.To.Type = obj.TYPE_BRANCH
-	if ctxt.Cursym.Cfunc != 0 {
-		p.To.Sym = obj.Linklookup(ctxt, "runtime.morestackc", 0)
-	} else if ctxt.Cursym.Text.From3.Offset&obj.NEEDCTXT == 0 {
-		p.To.Sym = obj.Linklookup(ctxt, "runtime.morestack_noctxt", 0)
-	} else {
-		p.To.Sym = obj.Linklookup(ctxt, "runtime.morestack", 0)
+	var last *obj.Prog
+	for last = ctxt.Cursym.Text; last.Link != nil; last = last.Link {
 	}
 
-	// BLS	start
-	p = obj.Appendp(ctxt, p)
+	// MOVW	LR, R3
+	movw := obj.Appendp(ctxt, last)
+	movw.As = AMOVW
+	movw.From.Type = obj.TYPE_REG
+	movw.From.Reg = REGLINK
+	movw.To.Type = obj.TYPE_REG
+	movw.To.Reg = REG_R3
 
-	p.As = ABLS
-	p.To.Type = obj.TYPE_BRANCH
-	p.Pcond = ctxt.Cursym.Text.Link
+	bls.Pcond = movw
 
-	return p
+	// BL runtime.morestack
+	call := obj.Appendp(ctxt, movw)
+	call.As = obj.ACALL
+	call.To.Type = obj.TYPE_BRANCH
+	morestack := "runtime.morestack"
+	switch {
+	case ctxt.Cursym.Cfunc != 0:
+		morestack = "runtime.morestackc"
+	case ctxt.Cursym.Text.From3.Offset&obj.NEEDCTXT == 0:
+		morestack = "runtime.morestack_noctxt"
+	}
+	call.To.Sym = obj.Linklookup(ctxt, morestack, 0)
+
+	// B start
+	b := obj.Appendp(ctxt, call)
+	b.As = obj.AJMP
+	b.To.Type = obj.TYPE_BRANCH
+	b.Pcond = ctxt.Cursym.Text.Link
+
+	return bls
 }
 
 func initdiv(ctxt *obj.Link) {
diff --git a/src/runtime/asm_arm.s b/src/runtime/asm_arm.s
index 291aa83cd8..a765b0c6b9 100644
--- a/src/runtime/asm_arm.s
+++ b/src/runtime/asm_arm.s
@@ -259,7 +259,6 @@ noswitch:
 
 // Called during function prolog when more stack is needed.
 // R1 frame size
-// R2 arg size
 // R3 prolog's LR
 // NB. we do not save R0 because we've forced 5c to pass all arguments
 // on the stack.
diff --git a/src/runtime/sys_arm.go b/src/runtime/sys_arm.go
index 324276e962..6e50d21098 100644
--- a/src/runtime/sys_arm.go
+++ b/src/runtime/sys_arm.go
@@ -24,7 +24,7 @@ func rewindmorestack(buf *gobuf) {
 	var inst uint32
 	if buf.pc&3 == 0 && buf.pc != 0 {
 		inst = *(*uint32)(unsafe.Pointer(buf.pc))
-		if inst>>24 == 0x9a {
+		if inst>>24 == 0x9a || inst>>24 == 0xea {
 			buf.pc += uintptr(int32(inst<<8)>>6) + 8
 			return
 		}