diff --git a/src/runtime/386/asm.s b/src/runtime/386/asm.s
index b7138105f7..cb6af7a155 100644
--- a/src/runtime/386/asm.s
+++ b/src/runtime/386/asm.s
@@ -157,15 +157,17 @@ TEXT cas(SB), 7, $0
 	MOVL	$1, AX
 	RET
 
-// void jmpdefer(byte*);
+// void jmpdefer(fn, sp);
+// called from deferreturn.
 // 1. pop the caller
 // 2. sub 5 bytes from the callers return
 // 3. jmp to the argument
 TEXT jmpdefer(SB), 7, $0
-	MOVL	4(SP), AX	// function
-	ADDL	$(4+56), SP	// pop saved PC and callers frame
-	SUBL	$5, (SP)	// reposition his return address
-	JMP	AX		// and goto function
+	MOVL	4(SP), AX	// fn
+	MOVL	8(SP), BX	// caller sp
+	LEAL	-4(BX), SP	// caller sp after CALL
+	SUBL	$5, (SP)	// return to CALL again
+	JMP	AX	// but first run the deferred function
 
 TEXT	sys·memclr(SB),7,$0
 	MOVL	4(SP), DI		// arg 1 addr
diff --git a/src/runtime/amd64/asm.s b/src/runtime/amd64/asm.s
index b69259e314..6fc01bbc98 100644
--- a/src/runtime/amd64/asm.s
+++ b/src/runtime/amd64/asm.s
@@ -172,7 +172,7 @@ TEXT setspgoto(SB), 7, $0
 	MOVQ	AX, SP
 	PUSHQ	CX
 	JMP	BX
-	POPQ	AX
+	POPQ	AX	// not reached
 	RET
 
 // bool cas(int32 *val, int32 old, int32 new)
@@ -194,12 +194,14 @@ TEXT cas(SB), 7, $0
 	MOVL	$1, AX
 	RET
 
-// void jmpdefer(byte*);
+// void jmpdefer(fn, sp);
+// called from deferreturn.
 // 1. pop the caller
 // 2. sub 5 bytes from the callers return
 // 3. jmp to the argument
 TEXT jmpdefer(SB), 7, $0
-	MOVQ	8(SP), AX	// function
-	ADDQ	$(8+56), SP	// pop saved PC and callers frame
-	SUBQ	$5, (SP)	// reposition his return address
-	JMP	AX		// and goto function
+	MOVQ	8(SP), AX	// fn
+	MOVQ	16(SP), BX	// caller sp
+	LEAQ	-8(BX), SP	// caller sp after CALL
+	SUBQ	$5, (SP)	// return to CALL again
+	JMP	AX	// but first run the deferred function
diff --git a/src/runtime/proc.c b/src/runtime/proc.c
index d51a6c013e..1d065e6d2a 100644
--- a/src/runtime/proc.c
+++ b/src/runtime/proc.c
@@ -607,7 +607,7 @@ oldstack(void)
 	Stktop *top;
 	uint32 args;
 	byte *sp;
-	uint64 oldsp, oldpc, oldbase, oldguard;
+	uintptr oldsp, oldpc, oldbase, oldguard;
 
 // printf("oldstack m->cret=%p\n", m->cret);
 
@@ -622,10 +622,10 @@ oldstack(void)
 		mcpy(top->oldsp+2*sizeof(uintptr), sp, args);
 	}
 
-	oldsp = (uint64)top->oldsp + 8;
-	oldpc = *(uint64*)(top->oldsp + 8);
-	oldbase = (uint64)top->oldbase;
-	oldguard = (uint64)top->oldguard;
+	oldsp = (uintptr)top->oldsp + sizeof(uintptr);
+	oldpc = *(uintptr*)oldsp;
+	oldbase = (uintptr)top->oldbase;
+	oldguard = (uintptr)top->oldguard;
 
 	stackfree((byte*)m->curg->stackguard - StackGuard);
 
@@ -645,6 +645,7 @@ oldstack(void)
 	gogoret(&m->morestack, m->cret);
 }
 
+#pragma textflag 7
 void
 lessstack(void)
 {
@@ -818,13 +819,11 @@ sys·deferproc(int32 siz, byte* fn, byte* arg0)
 
 #pragma textflag 7
 void
-sys·deferreturn(int32 arg0)
+sys·deferreturn(uintptr arg0)
 {
-	// warning: jmpdefer knows the frame size
-	// of this routine. dont change anything
-	// that might change the frame size
 	Defer *d;
-	byte *sp;
+	byte *sp, *fn;
+	uintptr *caller;
 
 	d = g->defer;
 	if(d == nil)
@@ -834,10 +833,10 @@ sys·deferreturn(int32 arg0)
 		return;
 	mcpy(d->sp, d->args, d->siz);
 	g->defer = d->link;
-	sp = d->fn;
+	fn = d->fn;
 	free(d);
-	jmpdefer(sp);
-}
+	jmpdefer(fn, sp);
+  }
 
 void
 runtime·Breakpoint(void)
diff --git a/src/runtime/runtime.h b/src/runtime/runtime.h
index f2926037aa..8b92c446c8 100644
--- a/src/runtime/runtime.h
+++ b/src/runtime/runtime.h
@@ -314,7 +314,7 @@ int32	write(int32, void*, int32);
 void	close(int32);
 int32	fstat(int32, void*);
 bool	cas(uint32*, uint32, uint32);
-void	jmpdefer(byte*);
+void	jmpdefer(byte*, void*);
 void	exit1(int32);
 void	ready(G*);
 byte*	getenv(int8*);