runtime: remove type-punning for Type.gc[0], gc[1]

Depending on flags&KindGCProg,
gc[0] and gc[1] are either pointers or inlined bitmap bits.
That's not compatible with a precise garbage collector:
it needs to be always pointers or never pointers.

Change the inlined bitmap case to store a pointer to an
out-of-line bitmap in gc[0]. The out-of-line bitmaps are
dedup'ed, so that for example all pointer types share the
same out-of-line bitmap.

Fixes #8864.

LGTM=r
R=golang-codereviews, dvyukov, r
CC=golang-codereviews, iant, khr, rlh
https://golang.org/cl/155820043

5 files changed, 30 insertions, 8 deletions

diff --git a/src/cmd/gc/reflect.c b/src/cmd/gc/reflect.c
index 4892ab7570..d0ebf6b481 100644
--- a/src/cmd/gc/reflect.c
+++ b/src/cmd/gc/reflect.c
@@ -716,9 +716,10 @@ static int
 dcommontype(Sym *s, int ot, Type *t)
 {
 	int i, alg, sizeofAlg, gcprog;
-	Sym *sptr, *algsym, *zero, *gcprog0, *gcprog1;
+	Sym *sptr, *algsym, *zero, *gcprog0, *gcprog1, *sbits;
 	uint8 gcmask[16];
 	static Sym *algarray;
+	uint64 x1, x2;
 	char *p;
 	
 	if(ot != 0)
@@ -804,8 +805,26 @@ dcommontype(Sym *s, int ot, Type *t)
 		ot = dsymptr(s, ot, gcprog1, 0);
 	} else {
 		gengcmask(t, gcmask);
-		for(i = 0; i < 2*widthptr; i++)
-			ot = duint8(s, ot, gcmask[i]);
+		x1 = 0;
+		for(i=0; i<8; i++)
+			x1 = x1<<8 | gcmask[i];
+		if(widthptr == 4) {
+			p = smprint("gcbits.%#016x", x1);
+		} else {
+			x2 = 0;
+			for(i=0; i<8; i++)
+				x2 = x2<<8 | gcmask[i+8];
+			p = smprint("gcbits.%#016llux%016llux", x1, x2);
+		}
+		sbits = pkglookup(p, runtimepkg);
+		if((sbits->flags & SymUniq) == 0) {
+			sbits->flags |= SymUniq;
+			for(i = 0; i < 2*widthptr; i++)
+				duint8(sbits, i, gcmask[i]);
+			ggloblsym(sbits, 2*widthptr, DUPOK|RODATA);
+		}
+		ot = dsymptr(s, ot, sbits, 0);
+		ot = duintptr(s, ot, 0);
 	}
 	p = smprint("%-uT", t);
 	//print("dcommontype: %s\n", p);
diff --git a/src/cmd/ld/decodesym.c b/src/cmd/ld/decodesym.c
index c530669425..037263dce7 100644
--- a/src/cmd/ld/decodesym.c
+++ b/src/cmd/ld/decodesym.c
@@ -111,7 +111,10 @@ decodetype_gcprog(LSym *s)
 uint8*
 decodetype_gcmask(LSym *s)
 {
-	return (uint8*)(s->p + 1*PtrSize + 8 + 1*PtrSize);
+	LSym *mask;
+	
+	mask = decode_reloc_sym(s, 1*PtrSize + 8 + 1*PtrSize);
+	return mask->p;
 }
 
 // Type.ArrayType.elem and Type.SliceType.Elem
diff --git a/src/reflect/type.go b/src/reflect/type.go
index f099546d27..a36c0ba604 100644
--- a/src/reflect/type.go
+++ b/src/reflect/type.go
@@ -1523,8 +1523,8 @@ func (gc *gcProg) appendProg(t *rtype) {
 		// The program is stored in t.gc[0], skip unroll flag.
 		prog = (*[1 << 30]byte)(unsafe.Pointer(t.gc[0]))[1:]
 	} else {
-		// The mask is embed directly in t.gc.
-		prog = (*[1 << 30]byte)(unsafe.Pointer(&t.gc[0]))[:]
+		// The mask is linked directly in t.gc.
+		prog = (*[2 * ptrSize]byte)(unsafe.Pointer(t.gc[0]))[:]
 	}
 	for i := uintptr(0); i < nptr; i++ {
 		gc.appendWord(extractGCWord(prog, i))
diff --git a/src/runtime/malloc.go b/src/runtime/malloc.go
index 99d14e3145..9b4264f2b3 100644
--- a/src/runtime/malloc.go
+++ b/src/runtime/malloc.go
@@ -270,7 +270,7 @@ func mallocgc(size uintptr, typ *_type, flags int) unsafe.Pointer {
 			}
 			ptrmask = (*uint8)(add(unsafe.Pointer(ptrmask), 1)) // skip the unroll flag byte
 		} else {
-			ptrmask = (*uint8)(unsafe.Pointer(&typ.gc[0])) // embed mask
+			ptrmask = (*uint8)(unsafe.Pointer(typ.gc[0])) // pointer to unrolled mask
 		}
 		if size == 2*ptrSize {
 			*xbits = *ptrmask | bitBoundary
diff --git a/src/runtime/type.h b/src/runtime/type.h
index de82e886f2..f5b4f9d13f 100644
--- a/src/runtime/type.h
+++ b/src/runtime/type.h
@@ -23,7 +23,7 @@ struct Type
 	uint8 kind;
 	void* alg;
 	// gc stores type info required for garbage collector.
-	// If (kind&KindGCProg)==0, then gc directly contains sparse GC bitmap
+	// If (kind&KindGCProg)==0, then gc[0] points at sparse GC bitmap
 	// (no indirection), 4 bits per word.
 	// If (kind&KindGCProg)!=0, then gc[1] points to a compiler-generated
 	// read-only GC program; and gc[0] points to BSS space for sparse GC bitmap.