cmd/ld: clear unused ctxt before morestack

For non-closure functions, the context register is uninitialized
on entry and will not be used, but morestack saves it and then the
garbage collector treats it as live. This can be a source of memory
leaks if the context register points at otherwise dead memory.
Avoid this by introducing a parallel set of morestack functions
that clear the context register, and use those for the non-closure functions.

I hope this will help with some of the finalizer flakiness, but it probably won't.

Fixes issue 7244.

LGTM=dvyukov
R=khr, dvyukov
CC=golang-codereviews
https://codereview.appspot.com/71030044

3 files changed, 33 insertions, 19 deletions

diff --git a/src/liblink/obj5.c b/src/liblink/obj5.c
index ce9d4f654..91d13d8c1 100644
--- a/src/liblink/obj5.c
+++ b/src/liblink/obj5.c
@@ -194,7 +194,7 @@ prg(void)
 	return p;
 }
 
-static	Prog*	stacksplit(Link*, Prog*, int32);
+static	Prog*	stacksplit(Link*, Prog*, int32, int);
 static	void		initdiv(Link*);
 static	void	softfloat(Link*, LSym*);
 
@@ -237,9 +237,11 @@ addstacksplit(Link *ctxt, LSym *cursym)
 	
 	autosize = 0;
 
-	if(ctxt->symmorestack[0] == nil)
+	if(ctxt->symmorestack[0] == nil) {
 		ctxt->symmorestack[0] = linklookup(ctxt, "runtime.morestack", 0);
-	
+		ctxt->symmorestack[1] = linklookup(ctxt, "runtime.morestack_noctxt", 0);
+	}
+
 	q = nil;
 	
 	ctxt->cursym = cursym;
@@ -409,7 +411,7 @@ addstacksplit(Link *ctxt, LSym *cursym)
 			}
 
 			if(!(p->reg & NOSPLIT))
-				p = stacksplit(ctxt, p, autosize); // emit split check
+				p = stacksplit(ctxt, p, autosize, !(cursym->text->from.scale&NEEDCTXT)); // emit split check
 			
 			// MOVW.W		R14,$-autosize(SP)
 			p = appendp(ctxt, p);
@@ -727,7 +729,7 @@ softfloat(Link *ctxt, LSym *cursym)
 }
 
 static Prog*
-stacksplit(Link *ctxt, Prog *p, int32 framesize)
+stacksplit(Link *ctxt, Prog *p, int32 framesize, int noctxt)
 {
 	int32 arg;
 
@@ -851,7 +853,7 @@ stacksplit(Link *ctxt, Prog *p, int32 framesize)
 	p->as = ABL;
 	p->scond = C_SCOND_LS;
 	p->to.type = D_BRANCH;
-	p->to.sym = ctxt->symmorestack[0];
+	p->to.sym = ctxt->symmorestack[noctxt];
 	
 	// BLS	start
 	p = appendp(ctxt, p);
diff --git a/src/liblink/obj6.c b/src/liblink/obj6.c
index 9b99a5995..036e20c8d 100644
--- a/src/liblink/obj6.c
+++ b/src/liblink/obj6.c
@@ -342,20 +342,30 @@ static char*
 morename[] =
 {
 	"runtime.morestack00",
+	"runtime.morestack00_noctxt",
 	"runtime.morestack10",
+	"runtime.morestack10_noctxt",
 	"runtime.morestack01",
+	"runtime.morestack01_noctxt",
 	"runtime.morestack11",
+	"runtime.morestack11_noctxt",
 
 	"runtime.morestack8",
+	"runtime.morestack8_noctxt",
 	"runtime.morestack16",
+	"runtime.morestack16_noctxt",
 	"runtime.morestack24",
+	"runtime.morestack24_noctxt",
 	"runtime.morestack32",
+	"runtime.morestack32_noctxt",
 	"runtime.morestack40",
+	"runtime.morestack40_noctxt",
 	"runtime.morestack48",
+	"runtime.morestack48_noctxt",
 };
 
 static Prog*	load_g_cx(Link*, Prog*);
-static Prog*	stacksplit(Link*, Prog*, int32, int32, Prog**);
+static Prog*	stacksplit(Link*, Prog*, int32, int32, int, Prog**);
 static void	indir_cx(Link*, Addr*);
 
 static void
@@ -419,7 +429,7 @@ addstacksplit(Link *ctxt, LSym *cursym)
 		p = load_g_cx(ctxt, p); // load g into CX
 	}
 	if(!(cursym->text->from.scale & NOSPLIT))
-		p = stacksplit(ctxt, p, autoffset, textarg, &q); // emit split check
+		p = stacksplit(ctxt, p, autoffset, textarg, !(cursym->text->from.scale&NEEDCTXT), &q); // emit split check
 
 	if(autoffset) {
 		if(autoffset%ctxt->arch->regsize != 0)
@@ -674,7 +684,7 @@ load_g_cx(Link *ctxt, Prog *p)
 // On return, *jmpok is the instruction that should jump
 // to the stack frame allocation if no split is needed.
 static Prog*
-stacksplit(Link *ctxt, Prog *p, int32 framesize, int32 textarg, Prog **jmpok)
+stacksplit(Link *ctxt, Prog *p, int32 framesize, int32 textarg, int noctxt, Prog **jmpok)
 {
 	Prog *q, *q1;
 	uint32 moreconst1, moreconst2, i;
@@ -822,7 +832,7 @@ stacksplit(Link *ctxt, Prog *p, int32 framesize, int32 textarg, Prog **jmpok)
 	if(moreconst1 == 0 && moreconst2 == 0) {
 		p->as = ACALL;
 		p->to.type = D_BRANCH;
-		p->to.sym = ctxt->symmorestack[0];
+		p->to.sym = ctxt->symmorestack[0*2+noctxt];
 	} else
 	if(moreconst1 != 0 && moreconst2 == 0) {
 		p->as = AMOVL;
@@ -833,13 +843,13 @@ stacksplit(Link *ctxt, Prog *p, int32 framesize, int32 textarg, Prog **jmpok)
 		p = appendp(ctxt, p);
 		p->as = ACALL;
 		p->to.type = D_BRANCH;
-		p->to.sym = ctxt->symmorestack[1];
+		p->to.sym = ctxt->symmorestack[1*2+noctxt];
 	} else
 	if(moreconst1 == 0 && moreconst2 <= 48 && moreconst2%8 == 0) {
 		i = moreconst2/8 + 3;
 		p->as = ACALL;
 		p->to.type = D_BRANCH;
-		p->to.sym = ctxt->symmorestack[i];
+		p->to.sym = ctxt->symmorestack[i*2+noctxt];
 	} else
 	if(moreconst1 == 0 && moreconst2 != 0) {
 		p->as = AMOVL;
@@ -850,7 +860,7 @@ stacksplit(Link *ctxt, Prog *p, int32 framesize, int32 textarg, Prog **jmpok)
 		p = appendp(ctxt, p);
 		p->as = ACALL;
 		p->to.type = D_BRANCH;
-		p->to.sym = ctxt->symmorestack[2];
+		p->to.sym = ctxt->symmorestack[2*2+noctxt];
 	} else {
 		p->as = mov;
 		p->from.type = D_CONST;
@@ -861,7 +871,7 @@ stacksplit(Link *ctxt, Prog *p, int32 framesize, int32 textarg, Prog **jmpok)
 		p = appendp(ctxt, p);
 		p->as = ACALL;
 		p->to.type = D_BRANCH;
-		p->to.sym = ctxt->symmorestack[3];
+		p->to.sym = ctxt->symmorestack[3*2+noctxt];
 	}
 	
 	p = appendp(ctxt, p);
diff --git a/src/liblink/obj8.c b/src/liblink/obj8.c
index adee8c6c5..6e40d04a5 100644
--- a/src/liblink/obj8.c
+++ b/src/liblink/obj8.c
@@ -256,7 +256,7 @@ prg(void)
 }
 
 static Prog*	load_g_cx(Link*, Prog*);
-static Prog*	stacksplit(Link*, Prog*, int32, Prog**);
+static Prog*	stacksplit(Link*, Prog*, int32, int, Prog**);
 
 static void
 addstacksplit(Link *ctxt, LSym *cursym)
@@ -265,8 +265,10 @@ addstacksplit(Link *ctxt, LSym *cursym)
 	int32 autoffset, deltasp;
 	int a;
 
-	if(ctxt->symmorestack[0] == nil)
+	if(ctxt->symmorestack[0] == nil) {
 		ctxt->symmorestack[0] = linklookup(ctxt, "runtime.morestack", 0);
+		ctxt->symmorestack[1] = linklookup(ctxt, "runtime.morestack_noctxt", 0);
+	}
 
 	if(ctxt->headtype == Hplan9 && ctxt->plan9tos == nil)
 		ctxt->plan9tos = linklookup(ctxt, "_tos", 0);
@@ -291,7 +293,7 @@ addstacksplit(Link *ctxt, LSym *cursym)
 		p = load_g_cx(ctxt, p); // load g into CX
 	}
 	if(!(cursym->text->from.scale & NOSPLIT))
-		p = stacksplit(ctxt, p, autoffset, &q); // emit split check
+		p = stacksplit(ctxt, p, autoffset, !(cursym->text->from.scale&NEEDCTXT), &q); // emit split check
 
 	if(autoffset) {
 		p = appendp(ctxt, p);
@@ -499,7 +501,7 @@ load_g_cx(Link *ctxt, Prog *p)
 // On return, *jmpok is the instruction that should jump
 // to the stack frame allocation if no split is needed.
 static Prog*
-stacksplit(Link *ctxt, Prog *p, int32 framesize, Prog **jmpok)
+stacksplit(Link *ctxt, Prog *p, int32 framesize, int noctxt, Prog **jmpok)
 {
 	Prog *q, *q1;
 	int arg;
@@ -642,7 +644,7 @@ stacksplit(Link *ctxt, Prog *p, int32 framesize, Prog **jmpok)
 	p = appendp(ctxt, p);
 	p->as = ACALL;
 	p->to.type = D_BRANCH;
-	p->to.sym = ctxt->symmorestack[0];
+	p->to.sym = ctxt->symmorestack[noctxt];
 
 	p = appendp(ctxt, p);
 	p->as = AJMP;