cmd/5l: support -Z (zero stack frame at function entry)

also added appropriate docs to cmd/ld/doc.go
(largely copied from Russ's CL 6938073).

R=rsc
CC=golang-dev
https://golang.org/cl/7004049

3 files changed, 75 insertions, 2 deletions

diff --git a/src/cmd/5l/l.h b/src/cmd/5l/l.h
index 5b05932fdd..0ceafdd112 100644
--- a/src/cmd/5l/l.h
+++ b/src/cmd/5l/l.h
@@ -420,8 +420,9 @@ int32	immaddr(int32);
 int32	opbra(int, int);
 int	brextra(Prog*);
 int	isbranch(Prog*);
-void fnptrs(void);
+void	fnptrs(void);
 void	doelf(void);
+void	dozerostk(void); // used by -Z
 
 vlong		addaddr(Sym *s, Sym *t);
 vlong		addsize(Sym *s, Sym *t);
diff --git a/src/cmd/5l/obj.c b/src/cmd/5l/obj.c
index a99f67d949..6a95a5e572 100644
--- a/src/cmd/5l/obj.c
+++ b/src/cmd/5l/obj.c
@@ -255,7 +255,12 @@ main(int argc, char *argv[])
 	doelf();
 	follow();
 	softfloat();
-	noops();
+	// 5l -Z means zero the stack frame on entry.
+	// This slows down function calls but can help avoid
+	// false positives in garbage collection.
+	if(debug['Z'])
+		dozerostk();
+	noops(); // generate stack split prolog, handle div/mod, etc.
 	dostkcheck();
 	span();
 	addexport();
diff --git a/src/cmd/5l/pass.c b/src/cmd/5l/pass.c
index 50593ced97..c22b860858 100644
--- a/src/cmd/5l/pass.c
+++ b/src/cmd/5l/pass.c
@@ -333,3 +333,70 @@ rnd(int32 v, int32 r)
 	v -= c;
 	return v;
 }
+
+void
+dozerostk(void)
+{
+	Prog *p, *pl;
+	int32 autoffset;
+
+	for(cursym = textp; cursym != nil; cursym = cursym->next) {
+		if(cursym->text == nil || cursym->text->link == nil)
+			continue;				
+		p = cursym->text;
+		autoffset = p->to.offset;
+		if(autoffset < 0)
+			autoffset = 0;
+		if(autoffset && !(p->reg&NOSPLIT)) {
+			// MOVW $4(R13), R1
+			p = appendp(p);
+			p->as = AMOVW;
+			p->from.type = D_CONST;
+			p->from.reg = 13;
+			p->from.offset = 4;
+			p->to.type = D_REG;
+			p->to.reg = 1;
+
+			// MOVW $n(R13), R2
+			p = appendp(p);
+			p->as = AMOVW;
+			p->from.type = D_CONST;
+			p->from.reg = 13;
+			p->from.offset = 4 + autoffset;
+			p->to.type = D_REG;
+			p->to.reg = 2;
+
+			// MOVW $0, R3
+			p = appendp(p);
+			p->as = AMOVW;
+			p->from.type = D_CONST;
+			p->from.offset = 0;
+			p->to.type = D_REG;
+			p->to.reg = 3;
+
+			// L:
+			//	MOVW.P R3, 0(R1) +4
+			//	CMP R1, R2
+			//	BNE L
+			p = pl = appendp(p);
+			p->as = AMOVW;
+			p->from.type = D_REG;
+			p->from.reg = 3;
+			p->to.type = D_OREG;
+			p->to.reg = 1;
+			p->to.offset = 4;
+			p->scond |= C_PBIT;
+
+			p = appendp(p);
+			p->as = ACMP;
+			p->from.type = D_REG;
+			p->from.reg = 1;
+			p->reg = 2;
+
+			p = appendp(p);
+			p->as = ABNE;
+			p->to.type = D_BRANCH;
+			p->cond = pl;
+		}
+	}
+}