runtime: make stack frames fixed size by modifying goproc/deferproc.

Calls to goproc/deferproc used to push & pop two extra arguments,
the argument size and the function to call.  Now, we allocate space
for those arguments in the outargs section so we don't have to
modify the SP.

Defers now use the stack pointer (instead of the argument pointer)
to identify which frame they are associated with.

A followon CL might simplify funcspdelta and some of the stack
walking code.

Fixes issue #8641

Change-Id: I835ec2f42f0392c5dec7cb0fe6bba6f2aed1dad8
Reviewed-on: https://go-review.googlesource.com/1601Reviewed-by: Russ Cox <rsc@golang.org>

src/cmd/5g/ggen.c

@@ -181,9 +181,14 @@ ginscall(Node *f, int proc)
 {
 	Prog *p;
 	Node n1, r, r1, con;
+	int32 extra;
 
-	if(f->type != T)
-		setmaxarg(f->type);
+	if(f->type != T) {
+		extra = 0;
+		if(proc == 1 || proc == 2)
+			extra = 2 * widthptr;
+		setmaxarg(f->type, extra);
+	}
 
 	switch(proc) {
 	default:
@@ -230,32 +235,22 @@ ginscall(Node *f, int proc)
 	case 1:	// call in new proc (go)
 	case 2:	// deferred call (defer)
 		regalloc(&r, types[tptr], N);
-		p = gins(AMOVW, N, &r);
-		p->from.type = D_OREG;
-		p->from.reg = REGSP;
-		
+		nodconst(&con, types[TINT32], argsize(f->type));
+		gins(AMOVW, &con, &r);
 		p = gins(AMOVW, &r, N);
 		p->to.type = D_OREG;
 		p->to.reg = REGSP;
-		p->to.offset = -12;
-		p->scond |= C_WBIT;
+		p->to.offset = 4;
 
 		memset(&n1, 0, sizeof n1);
 		n1.op = OADDR;
 		n1.left = f;
 		gins(AMOVW, &n1, &r);
-
 		p = gins(AMOVW, &r, N);
 		p->to.type = D_OREG;
 		p->to.reg = REGSP;
 		p->to.offset = 8;
 
-		nodconst(&con, types[TINT32], argsize(f->type));
-		gins(AMOVW, &con, &r);
-		p = gins(AMOVW, &r, N);
-		p->to.type = D_OREG;
-		p->to.reg = REGSP;
-		p->to.offset = 4;
 		regfree(&r);
 
 		if(proc == 1)
@@ -263,14 +258,6 @@ ginscall(Node *f, int proc)
 		else
 			ginscall(deferproc, 0);
 
-		nodreg(&r, types[tptr], 1);
-		p = gins(AMOVW, N, N);
-		p->from.type = D_CONST;
-		p->from.reg = REGSP;
-		p->from.offset = 12;
-		p->to.reg = REGSP;
-		p->to.type = D_REG;
-
 		if(proc == 2) {
 			nodconst(&con, types[TINT32], 0);
 			p = gins(ACMP, &con, N);
@@ -330,9 +317,11 @@ cgen_callinter(Node *n, Node *res, int proc)
 	agen(i, &nodr);		// REG = &inter
 
 	nodindreg(&nodsp, types[tptr], REGSP);
-	nodsp.xoffset = 4;
+	nodsp.xoffset = widthptr;
+	if(proc != 0)
+		nodsp.xoffset += 2 * widthptr; // leave room for size & fn
 	nodo.xoffset += widthptr;
-	cgen(&nodo, &nodsp);	// 4(SP) = 4(REG) -- i.data
+	cgen(&nodo, &nodsp);	// {4 or 12}(SP) = 4(REG) -- i.data
 
 	nodo.xoffset -= widthptr;
 	cgen(&nodo, &nodr);	// REG = 0(REG) -- i.tab

src/cmd/6g/ggen.c

@@ -176,11 +176,16 @@ void
 ginscall(Node *f, int proc)
 {
 	Prog *p;
-	Node reg, con;
+	Node reg, stk;
 	Node r1;
+	int32 extra;
 
-	if(f->type != T)
-		setmaxarg(f->type);
+	if(f->type != T) {
+		extra = 0;
+		if(proc == 1 || proc == 2)
+			extra = 2 * widthptr;
+		setmaxarg(f->type, extra);
+	}
 
 	switch(proc) {
 	default:
@@ -224,21 +229,31 @@ ginscall(Node *f, int proc)
 
 	case 1:	// call in new proc (go)
 	case 2:	// deferred call (defer)
-		nodconst(&con, types[TINT64], argsize(f->type));
-		if(widthptr == 4) {
-			nodreg(&r1, types[TINT32], D_CX);
-			gmove(f, &r1);
-			nodreg(&reg, types[TINT64], D_CX);
-			nodconst(&r1, types[TINT64], 32);
-			gins(ASHLQ, &r1, &reg);
-			gins(AORQ, &con, &reg);
-			gins(APUSHQ, &reg, N);
+		memset(&stk, 0, sizeof(stk));
+		stk.op = OINDREG;
+		stk.val.u.reg = D_SP;
+		stk.xoffset = 0;
+
+		if(widthptr == 8) {
+			// size of arguments at 0(SP)
+			ginscon(AMOVQ, argsize(f->type), &stk);
+
+			// FuncVal* at 8(SP)
+			stk.xoffset = widthptr;
+			nodreg(&reg, types[TINT64], D_AX);
+			gmove(f, &reg);
+			gins(AMOVQ, &reg, &stk);
 		} else {
-			nodreg(&reg, types[TINT64], D_CX);
+			// size of arguments at 0(SP)
+			ginscon(AMOVL, argsize(f->type), &stk);
+
+			// FuncVal* at 4(SP)
+			stk.xoffset = widthptr;
+			nodreg(&reg, types[TINT32], D_AX);
 			gmove(f, &reg);
-			gins(APUSHQ, &reg, N);
-			gins(APUSHQ, &con, N);
+			gins(AMOVL, &reg, &stk);
 		}
+
 		if(proc == 1)
 			ginscall(newproc, 0);
 		else {
@@ -246,13 +261,9 @@ ginscall(Node *f, int proc)
 				fatal("hasdefer=0 but has defer");
 			ginscall(deferproc, 0);
 		}
-		nodreg(&reg, types[TINT64], D_CX);
-		gins(APOPQ, N, &reg);
-		if(widthptr == 8)
-			gins(APOPQ, N, &reg);
 		if(proc == 2) {
-			nodreg(&reg, types[TINT64], D_AX);
-			gins(ATESTQ, &reg, &reg);
+			nodreg(&reg, types[TINT32], D_AX);
+			gins(ATESTL, &reg, &reg);
 			p = gbranch(AJEQ, T, +1);
 			cgen_ret(N);
 			patch(p, pc);
@@ -294,9 +305,12 @@ cgen_callinter(Node *n, Node *res, int proc)
 	igen(i, &nodi, res);		// REG = &inter
 
 	nodindreg(&nodsp, types[tptr], D_SP);
+        nodsp.xoffset = 0;
+	if(proc != 0)
+		nodsp.xoffset += 2 * widthptr; // leave room for size & fn
 	nodi.type = types[tptr];
 	nodi.xoffset += widthptr;
-	cgen(&nodi, &nodsp);	// 0(SP) = 8(REG) -- i.data
+	cgen(&nodi, &nodsp);	// {0, 8(nacl), or 16}(SP) = 8(REG) -- i.data
 
 	regalloc(&nodo, types[tptr], res);
 	nodi.type = types[tptr];

src/cmd/8g/ggen.c

@@ -237,10 +237,15 @@ void
 ginscall(Node *f, int proc)
 {
 	Prog *p;
-	Node reg, r1, con;
-
-	if(f->type != T)
-		setmaxarg(f->type);
+	Node reg, r1, con, stk;
+	int32 extra;
+
+	if(f->type != T) {
+		extra = 0;
+		if(proc == 1 || proc == 2)
+			extra = 2 * widthptr;
+		setmaxarg(f->type, extra);
+	}
 
 	switch(proc) {
 	default:
@@ -284,18 +289,25 @@ ginscall(Node *f, int proc)
 
 	case 1:	// call in new proc (go)
 	case 2:	// deferred call (defer)
-		nodreg(&reg, types[TINT32], D_CX);
-		gins(APUSHL, f, N);
+		memset(&stk, 0, sizeof(stk));
+		stk.op = OINDREG;
+		stk.val.u.reg = D_SP;
+		stk.xoffset = 0;
+
+		// size of arguments at 0(SP)
 		nodconst(&con, types[TINT32], argsize(f->type));
-		gins(APUSHL, &con, N);
+		gins(AMOVL, &con, &stk);
+
+		// FuncVal* at 4(SP)
+		stk.xoffset = widthptr;
+		gins(AMOVL, f, &stk);
+
 		if(proc == 1)
 			ginscall(newproc, 0);
 		else
 			ginscall(deferproc, 0);
-		gins(APOPL, N, &reg);
-		gins(APOPL, N, &reg);
 		if(proc == 2) {
-			nodreg(&reg, types[TINT64], D_AX);
+			nodreg(&reg, types[TINT32], D_AX);
 			gins(ATESTL, &reg, &reg);
 			p = gbranch(AJEQ, T, +1);
 			cgen_ret(N);
@@ -338,9 +350,12 @@ cgen_callinter(Node *n, Node *res, int proc)
 	igen(i, &nodi, res);		// REG = &inter
 
 	nodindreg(&nodsp, types[tptr], D_SP);
+	nodsp.xoffset = 0;
+	if(proc != 0)
+		nodsp.xoffset += 2 * widthptr; // leave room for size & fn
 	nodi.type = types[tptr];
 	nodi.xoffset += widthptr;
-	cgen(&nodi, &nodsp);	// 0(SP) = 4(REG) -- i.data
+	cgen(&nodi, &nodsp);	// {0 or 8}(SP) = 4(REG) -- i.data
 
 	regalloc(&nodo, types[tptr], res);
 	nodi.type = types[tptr];

src/cmd/9g/ggen.c

@@ -194,9 +194,14 @@ ginscall(Node *f, int proc)
 	Prog *p;
 	Node reg, con, reg2;
 	Node r1;
+	int32 extra;
 
-	if(f->type != T)
-		setmaxarg(f->type);
+	if(f->type != T) {
+		extra = 0;
+		if(proc == 1 || proc == 2)
+			extra = 2 * widthptr;
+		setmaxarg(f->type, extra);
+	}
 
 	switch(proc) {
 	default:
@@ -245,12 +250,6 @@ ginscall(Node *f, int proc)
 		nodreg(&reg2, types[TINT64], D_R0+4);
 		gmove(f, &reg);
 
-		p = gins(ASUB, N, N);
-		p->from.type = D_CONST;
-		p->from.offset = 3 * 8;
-		p->to.type = D_REG;
-		p->to.reg = REGSP;
-
 		gmove(&con, &reg2);
 		p = gins(AMOVW, &reg2, N);
 		p->to.type = D_OREG;
@@ -270,12 +269,6 @@ ginscall(Node *f, int proc)
 			ginscall(deferproc, 0);
 		}
 
-		p = gins(AADD, N, N);
-		p->from.type = D_CONST;
-		p->from.offset = 3 * 8;
-		p->to.type = D_REG;
-		p->to.reg = REGSP;
-
 		if(proc == 2) {
 			nodreg(&reg, types[TINT64], D_R0+3);
 			p = gins(ACMP, &reg, N);
@@ -324,9 +317,11 @@ cgen_callinter(Node *n, Node *res, int proc)
 
 	nodindreg(&nodsp, types[tptr], D_R0+REGSP);
 	nodsp.xoffset = widthptr;
+	if(proc != 0)
+		nodsp.xoffset += 2 * widthptr; // leave room for size & fn
 	nodi.type = types[tptr];
 	nodi.xoffset += widthptr;
-	cgen(&nodi, &nodsp);	// 0(SP) = 8(REG) -- i.data
+	cgen(&nodi, &nodsp);	// {8 or 24}(SP) = 8(REG) -- i.data
 
 	regalloc(&nodo, types[tptr], res);
 	nodi.type = types[tptr];

src/cmd/gc/go.h

@@ -1418,7 +1418,7 @@ Node*	cheapexpr(Node *n, NodeList **init);
 Node*	localexpr(Node *n, Type *t, NodeList **init);
 void	saveorignode(Node *n);
 int32	setlineno(Node *n);
-void	setmaxarg(Type *t);
+void	setmaxarg(Type *t, int32 extra);
 Type*	shallow(Type *t);
 int	simsimtype(Type *t);
 void	smagic(Magic *m);

src/cmd/gc/subr.c

@@ -2115,14 +2115,17 @@ localexpr(Node *n, Type *t, NodeList **init)
 }
 
 void
-setmaxarg(Type *t)
+setmaxarg(Type *t, int32 extra)
 {
 	int64 w;
 
 	dowidth(t);
 	w = t->argwid;
-	if(t->argwid >= MAXWIDTH)
+	if(w >= MAXWIDTH)
 		fatal("bad argwid %T", t);
+	w += extra;
+	if(w >= MAXWIDTH)
+		fatal("bad argwid %d + %T", extra, t);
 	if(w > maxarg)
 		maxarg = w;
 }

src/cmd/gc/walk.c

@@ -128,6 +128,35 @@ paramoutheap(Node *fn)
 	return 0;
 }
 
+// adds "adjust" to all the argument locations for the call n.
+// n must be a defer or go node that has already been walked.
+static void
+adjustargs(Node *n, int adjust)
+{
+	Node *callfunc, *arg, *lhs;
+	NodeList *args;
+
+	callfunc = n->left;
+	for(args = callfunc->list; args != 0; args = args->next) {
+		arg = args->n;
+		if(arg->op != OAS)
+			yyerror("call arg not assignment");
+		lhs = arg->left;
+		if(lhs->op == ONAME) {
+			// This is a temporary introduced by reorder1.
+			// The real store to the stack appears later in the arg list.
+			continue;
+		}
+		if(lhs->op != OINDREG) {
+			yyerror("call argument store does not use OINDREG");
+		}
+		// can't really check this in machine-indep code.
+		//if(lhs->val.u.reg != D_SP)
+		//      yyerror("call arg assign not indreg(SP)");
+		lhs->xoffset += adjust;
+	}
+}
+
 void
 walkstmt(Node **np)
 {
@@ -237,6 +266,8 @@ walkstmt(Node **np)
 			walkexpr(&n->left, &n->ninit);
 			break;
 		}
+		// make room for size & fn arguments.
+		adjustargs(n, 2 * widthptr);
 		break;
 
 	case OFOR:
@@ -270,6 +301,8 @@ walkstmt(Node **np)
 			walkexpr(&n->left, &n->ninit);
 			break;
 		}
+		// make room for size & fn arguments.
+		adjustargs(n, 2 * widthptr);
 		break;
 
 	case ORETURN:

src/runtime/heapdump.go

@@ -366,7 +366,7 @@ func dumpgoroutine(gp *g) {
 		dumpint(tagDefer)
 		dumpint(uint64(uintptr(unsafe.Pointer(d))))
 		dumpint(uint64(uintptr(unsafe.Pointer(gp))))
-		dumpint(uint64(d.argp))
+		dumpint(uint64(d.sp))
 		dumpint(uint64(d.pc))
 		dumpint(uint64(uintptr(unsafe.Pointer(d.fn))))
 		dumpint(uint64(uintptr(unsafe.Pointer(d.fn.fn))))

src/runtime/panic.go

@@ -62,13 +62,10 @@ func deferproc(siz int32, fn *funcval) { // arguments of fn follow fn
 	// the arguments of fn are in a perilous state.  The stack map
 	// for deferproc does not describe them.  So we can't let garbage
 	// collection or stack copying trigger until we've copied them out
-	// to somewhere safe.  deferproc_m does that.  Until deferproc_m,
-	// we can only call nosplit routines.
-	argp := uintptr(unsafe.Pointer(&fn))
-	argp += unsafe.Sizeof(fn)
-	if GOARCH == "arm" || GOARCH == "ppc64" || GOARCH == "ppc64le" {
-		argp += ptrSize // skip caller's saved link register
-	}
+	// to somewhere safe.  The memmove below does that.
+	// Until the copy completes, we can only call nosplit routines.
+	sp := getcallersp(unsafe.Pointer(&siz))
+	argp := uintptr(unsafe.Pointer(&fn)) + unsafe.Sizeof(fn)
 	callerpc := getcallerpc(unsafe.Pointer(&siz))
 
 	systemstack(func() {
@@ -78,7 +75,7 @@ func deferproc(siz int32, fn *funcval) { // arguments of fn follow fn
 		}
 		d.fn = fn
 		d.pc = callerpc
-		d.argp = argp
+		d.sp = sp
 		memmove(add(unsafe.Pointer(d), unsafe.Sizeof(*d)), unsafe.Pointer(argp), uintptr(siz))
 	})
 
@@ -240,8 +237,8 @@ func deferreturn(arg0 uintptr) {
 	if d == nil {
 		return
 	}
-	argp := uintptr(unsafe.Pointer(&arg0))
-	if d.argp != argp {
+	sp := getcallersp(unsafe.Pointer(&arg0))
+	if d.sp != sp {
 		return
 	}
 
@@ -250,13 +247,13 @@ func deferreturn(arg0 uintptr) {
 	// won't know the form of the arguments until the jmpdefer can
 	// flip the PC over to fn.
 	mp := acquirem()
-	memmove(unsafe.Pointer(argp), deferArgs(d), uintptr(d.siz))
+	memmove(unsafe.Pointer(&arg0), deferArgs(d), uintptr(d.siz))
 	fn := d.fn
 	d.fn = nil
 	gp._defer = d.link
 	freedefer(d)
 	releasem(mp)
-	jmpdefer(fn, argp)
+	jmpdefer(fn, uintptr(unsafe.Pointer(&arg0)))
 }
 
 // Goexit terminates the goroutine that calls it.  No other goroutine is affected.
@@ -403,7 +400,7 @@ func gopanic(e interface{}) {
 		//GC()
 
 		pc := d.pc
-		argp := unsafe.Pointer(d.argp) // must be pointer so it gets adjusted during stack copy
+		sp := unsafe.Pointer(d.sp) // must be pointer so it gets adjusted during stack copy
 		freedefer(d)
 		if p.recovered {
 			gp._panic = p.link
@@ -416,7 +413,7 @@ func gopanic(e interface{}) {
 				gp.sig = 0
 			}
 			// Pass information about recovering frame to recovery.
-			gp.sigcode0 = uintptr(argp)
+			gp.sigcode0 = uintptr(sp)
 			gp.sigcode1 = pc
 			mcall(recovery)
 			gothrow("recovery failed") // mcall should not return

src/runtime/panic1.go

@@ -4,8 +4,6 @@
 
 package runtime
 
-import "unsafe"
-
 // Code related to defer, panic and recover.
 // TODO: Merge into panic.go.
 
@@ -19,29 +17,19 @@ const hasLinkRegister = GOARCH == "arm" || GOARCH == "ppc64" || GOARCH == "ppc64
 // the caller of the deferred function returned normally.
 func recovery(gp *g) {
 	// Info about defer passed in G struct.
-	argp := (unsafe.Pointer)(gp.sigcode0)
-	pc := uintptr(gp.sigcode1)
+	sp := gp.sigcode0
+	pc := gp.sigcode1
 
 	// d's arguments need to be in the stack.
-	if argp != nil && (uintptr(argp) < gp.stack.lo || gp.stack.hi < uintptr(argp)) {
-		print("recover: ", argp, " not in [", hex(gp.stack.lo), ", ", hex(gp.stack.hi), "]\n")
+	if sp != 0 && (sp < gp.stack.lo || gp.stack.hi < sp) {
+		print("recover: ", hex(sp), " not in [", hex(gp.stack.lo), ", ", hex(gp.stack.hi), "]\n")
 		gothrow("bad recovery")
 	}
 
 	// Make the deferproc for this d return again,
 	// this time returning 1.  The calling function will
 	// jump to the standard return epilogue.
-	// The -2*sizeof(uintptr) makes up for the
-	// two extra words that are on the stack at
-	// each call to deferproc.
-	// (The pc we're returning to does pop pop
-	// before it tests the return value.)
-	// On the arm and power there are 2 saved LRs mixed in too.
-	if hasLinkRegister {
-		gp.sched.sp = uintptr(argp) - 4*ptrSize
-	} else {
-		gp.sched.sp = uintptr(argp) - 2*ptrSize
-	}
+	gp.sched.sp = sp
 	gp.sched.pc = pc
 	gp.sched.lr = 0
 	gp.sched.ret = 1

src/runtime/proc1.go

@@ -1931,10 +1931,6 @@ func malg(stacksize int32) *g {
 //go:nosplit
 func newproc(siz int32, fn *funcval) {
 	argp := add(unsafe.Pointer(&fn), ptrSize)
-	if hasLinkRegister {
-		argp = add(argp, ptrSize) // skip caller's saved LR
-	}
-
 	pc := getcallerpc(unsafe.Pointer(&siz))
 	systemstack(func() {
 		newproc1(fn, (*uint8)(argp), siz, 0, pc)

src/runtime/runtime2.go

@@ -508,7 +508,7 @@ var invalidptr int32
 type _defer struct {
 	siz     int32
 	started bool
-	argp    uintptr // where args were copied from
+	sp      uintptr // sp at time of defer
 	pc      uintptr
 	fn      *funcval
 	_panic  *_panic // panic that is running defer

src/runtime/stack1.go

@@ -499,7 +499,7 @@ func adjustdefers(gp *g, adjinfo *adjustinfo) {
 	// Defer structs themselves are never on the stack.
 	for d := gp._defer; d != nil; d = d.link {
 		adjustpointer(adjinfo, (unsafe.Pointer)(&d.fn))
-		adjustpointer(adjinfo, (unsafe.Pointer)(&d.argp))
+		adjustpointer(adjinfo, (unsafe.Pointer)(&d.sp))
 		adjustpointer(adjinfo, (unsafe.Pointer)(&d._panic))
 	}
 }

src/runtime/traceback.go

@@ -32,13 +32,11 @@ const usesLR = GOARCH != "amd64" && GOARCH != "amd64p32" && GOARCH != "386"
 
 var (
 	// initialized in tracebackinit
-	deferprocPC          uintptr
 	goexitPC             uintptr
 	jmpdeferPC           uintptr
 	mcallPC              uintptr
 	morestackPC          uintptr
 	mstartPC             uintptr
-	newprocPC            uintptr
 	rt0_goPC             uintptr
 	sigpanicPC           uintptr
 	systemstack_switchPC uintptr
@@ -51,13 +49,11 @@ func tracebackinit() {
 	// Instead of initializing the variables above in the declarations,
 	// schedinit calls this function so that the variables are
 	// initialized and available earlier in the startup sequence.
-	deferprocPC = funcPC(deferproc)
 	goexitPC = funcPC(goexit)
 	jmpdeferPC = funcPC(jmpdefer)
 	mcallPC = funcPC(mcall)
 	morestackPC = funcPC(morestack)
 	mstartPC = funcPC(mstart)
-	newprocPC = funcPC(newproc)
 	rt0_goPC = funcPC(rt0_go)
 	sigpanicPC = funcPC(sigpanic)
 	systemstack_switchPC = funcPC(systemstack_switch)
@@ -144,11 +140,10 @@ func gentraceback(pc0 uintptr, sp0 uintptr, lr0 uintptr, gp *g, skip int, pcbuf
 		frame.lr = lr0
 	}
 	waspanic := false
-	wasnewproc := false
 	printing := pcbuf == nil && callback == nil
 	_defer := gp._defer
 
-	for _defer != nil && uintptr(_defer.argp) == _NoArgs {
+	for _defer != nil && uintptr(_defer.sp) == _NoArgs {
 		_defer = _defer.link
 	}
 
@@ -251,32 +246,6 @@ func gentraceback(pc0 uintptr, sp0 uintptr, lr0 uintptr, gp *g, skip int, pcbuf
 			setArgInfo(&frame, f, callback != nil)
 		}
 
-		// Determine function SP where deferproc would find its arguments.
-		var sparg uintptr
-		if usesLR {
-			// On link register architectures, that's the standard bottom-of-stack plus 1 word
-			// for the saved LR. If the previous frame was a direct call to newproc/deferproc,
-			// however, the SP is three words lower than normal.
-			// If the function has no frame at all - perhaps it just started, or perhaps
-			// it is a leaf with no local variables - then we cannot possibly find its
-			// SP in a defer, and we might confuse its SP for its caller's SP, so
-			// leave sparg=0 in that case.
-			if frame.fp != frame.sp {
-				sparg = frame.sp + regSize
-				if wasnewproc {
-					sparg += 3 * regSize
-				}
-			}
-		} else {
-			// On x86 that's the standard bottom-of-stack, so SP exactly.
-			// If the previous frame was a direct call to newproc/deferproc, however,
-			// the SP is two words lower than normal.
-			sparg = frame.sp
-			if wasnewproc {
-				sparg += 2 * ptrSize
-			}
-		}
-
 		// Determine frame's 'continuation PC', where it can continue.
 		// Normally this is the return address on the stack, but if sigpanic
 		// is immediately below this function on the stack, then the frame
@@ -289,7 +258,7 @@ func gentraceback(pc0 uintptr, sp0 uintptr, lr0 uintptr, gp *g, skip int, pcbuf
 		// returns; everything live at earlier deferprocs is still live at that one.
 		frame.continpc = frame.pc
 		if waspanic {
-			if _defer != nil && _defer.argp == sparg {
+			if _defer != nil && _defer.sp == frame.sp {
 				frame.continpc = _defer.pc
 			} else {
 				frame.continpc = 0
@@ -297,7 +266,7 @@ func gentraceback(pc0 uintptr, sp0 uintptr, lr0 uintptr, gp *g, skip int, pcbuf
 		}
 
 		// Unwind our local defer stack past this frame.
-		for _defer != nil && (_defer.argp == sparg || _defer.argp == _NoArgs) {
+		for _defer != nil && (_defer.sp == frame.sp || _defer.sp == _NoArgs) {
 			_defer = _defer.link
 		}
 
@@ -353,7 +322,6 @@ func gentraceback(pc0 uintptr, sp0 uintptr, lr0 uintptr, gp *g, skip int, pcbuf
 
 	skipped:
 		waspanic = f.entry == sigpanicPC
-		wasnewproc = f.entry == newprocPC || f.entry == deferprocPC
 
 		// Do not unwind past the bottom of the stack.
 		if flr == nil {
@@ -438,10 +406,10 @@ func gentraceback(pc0 uintptr, sp0 uintptr, lr0 uintptr, gp *g, skip int, pcbuf
 	// incomplete information then is still better than nothing.
 	if callback != nil && n < max && _defer != nil {
 		if _defer != nil {
-			print("runtime: g", gp.goid, ": leftover defer argp=", hex(_defer.argp), " pc=", hex(_defer.pc), "\n")
+			print("runtime: g", gp.goid, ": leftover defer sp=", hex(_defer.sp), " pc=", hex(_defer.pc), "\n")
 		}
 		for _defer = gp._defer; _defer != nil; _defer = _defer.link {
-			print("\tdefer ", _defer, " argp=", hex(_defer.argp), " pc=", hex(_defer.pc), "\n")
+			print("\tdefer ", _defer, " sp=", hex(_defer.sp), " pc=", hex(_defer.pc), "\n")
 		}
 		gothrow("traceback has leftover defers")
 	}