fix off by 4 bug in morestack (lr again). remove storing of r0

now that all arguments are passed on the stack.

go/test: passes 89% (310/345)

R=rsc
APPROVED=rsc
DELTA=33  (13 added, 14 deleted, 6 changed)
OCL=36009
CL=36022

3 files changed, 19 insertions, 20 deletions

diff --git a/src/cmd/5l/noop.c b/src/cmd/5l/noop.c
index 61cd36550..e932750b5 100644
--- a/src/cmd/5l/noop.c
+++ b/src/cmd/5l/noop.c
@@ -128,7 +128,7 @@ noops(void)
 	Bflush(&bso);
 
 	pmorestack = P;
-	symmorestack = lookup("runtime·morestackx", 0);
+	symmorestack = lookup("runtime·morestack", 0);
 
 	if(symmorestack->type == STEXT)
 	for(p = firstp; p != P; p = p->link) {
@@ -358,9 +358,10 @@ noops(void)
 				// split stack check for small functions
 				// MOVW			g_stackguard(g), R1
 				// CMP			R1, $-autosize(SP)
+				// MOVW.LO		$autosize, R1
 				// MOVW.LO		$args, R2
-				// MOVW.W.LO	R14, R3
-				// BL.LO		runtime·morestackx(SB) // modifies LR
+				// MOVW.W.LO		R14, R3
+				// BL.LO			runtime·morestack(SB) // modifies LR
 				// MOVW.W		R14,$-autosize(SP)
 
 				// TODO(kaib): add more trampolines
@@ -383,12 +384,22 @@ noops(void)
 				p->from.offset = -autosize;
 				p->reg = REGSP;
 
-				// MOVW.LO		$args, R2
+				// MOVW.LO		$autosize, R1
+				p = appendp(p);
+				p->as = AMOVW;
+				p->scond = C_SCOND_LO;
+				p->from.type = D_CONST;
+				p->from.offset = 0;
+				p->to.type = D_REG;
+				p->to.reg = 1;
+
+				// MOVW.LO		$args +4, R2
+				// also need to store the extra 4 bytes.
 				p = appendp(p);
 				p->as = AMOVW;
 				p->scond = C_SCOND_LO;
 				p->from.type = D_CONST;
-				p->from.offset = curtext->to.offset2 & ~7;
+				p->from.offset = (curtext->to.offset2 & ~7) + 4;
 				p->to.type = D_REG;
 				p->to.reg = 2;
 
@@ -401,7 +412,7 @@ noops(void)
 				p->to.type = D_REG;
 				p->to.reg = 3;
 
-				// BL.LO		runtime·morestackx(SB) // modifies LR
+				// BL.LO		runtime·morestack(SB) // modifies LR
 				p = appendp(p);
 				p->as = ABL;
 				p->scond = C_SCOND_LO;
diff --git a/src/pkg/runtime/arm/asm.s b/src/pkg/runtime/arm/asm.s
index aa21ab850..98c9e06bf 100644
--- a/src/pkg/runtime/arm/asm.s
+++ b/src/pkg/runtime/arm/asm.s
@@ -120,7 +120,6 @@ TEXT gogocall(SB), 7, $-4
 	MOVW	0(g), R2		// make sure g != nil
 	MOVW	gobuf_sp(R0), SP	// restore SP
 	MOVW	gobuf_pc(R0), LR
-	MOVW	gobuf_r0(R0), R0
 	MOVW	R1, PC
 
 /*
@@ -131,6 +130,8 @@ TEXT gogocall(SB), 7, $-4
 // R1 frame size
 // R2 arg size
 // R3 prolog's LR
+// NB. we do not save R0 because the we've forced 5c to pass all arguments
+// on the stack.
 // using frame size $-4 means do not save LR on stack.
 TEXT runtime·morestack(SB),7,$-4
 	// Cannot grow scheduler stack (m->g0).
@@ -149,7 +150,6 @@ TEXT runtime·morestack(SB),7,$-4
 	MOVW	SP, (m_morebuf+gobuf_sp)(m) // f's caller's SP
 	MOVW	SP, m_morefp(m) // f's caller's SP
 	MOVW	g, (m_morebuf+gobuf_g)(m)
-	MOVW	R0, (m_morebuf+gobuf_r0)(m)
 
 	// Set m->morepc to f's PC.
 	MOVW	LR, m_morepc(m)
@@ -169,7 +169,6 @@ TEXT reflect·call(SB), 7, $-4
 	// restore when returning from f.
 	MOVW	LR, (m_morebuf+gobuf_pc)(m)	// our caller's PC
 	MOVW	SP, (m_morebuf+gobuf_sp)(m)	// our caller's SP
-	MOVW	R0, (m_morebuf+gobuf_r0)(m)
 	MOVW	g,  (m_morebuf+gobuf_g)(m)
 
 	// Set up morestack arguments to call f on a new stack.
@@ -204,16 +203,6 @@ TEXT runtime·lessstack(SB), 7, $-4
 	MOVW	(m_sched+gobuf_sp)(m), SP
 	B	oldstack(SB)
 
-// Optimization to make inline stack splitting code smaller
-// R0 is original first argument
-// R2 is argsize
-// R3 is LR for f (f's caller's PC)
-// using frame size $-4 means do not save LR on stack.
-TEXT runtime·morestackx(SB), 7, $-4
-	MOVW	$0, R1		// set frame size
-	B	runtime·morestack(SB)
-
-
 // void jmpdefer(fn, sp);
 // called from deferreturn.
 // 1. grab stored LR for caller
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index fac0008ee..068e2bea1 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -152,7 +152,6 @@ struct	Gobuf
 	byte*	sp;
 	byte*	pc;
 	G*	g;
-	uintptr	r0;		// used on arm
 };
 struct	G
 {