cmd/gc, runtime: fix race, nacl for writebarrier changes

The racewalk code was not updated for the new write barriers.
Make it more future-proof.

The new write barrier code assumed that +1 pointer would
be aligned properly for any type that might follow, but that's
not true on 32-bit systems where some types are 64-bit aligned.
The only system like that today is nacl/amd64p32.
Insert a dummy pointer so that the ambiguously typed
value is at +2 pointers, which is always max-aligned.

LGTM=r
R=r
CC=golang-codereviews, iant, khr
https://golang.org/cl/158890046

src/cmd/gc/builtin.c

@@ -87,9 +87,9 @@ char *runtimeimport =
 	"func @\"\".writebarrierstring (@\"\".dst·1 *any, @\"\".src·2 any)\n"
 	"func @\"\".writebarrierslice (@\"\".dst·1 *any, @\"\".src·2 any)\n"
 	"func @\"\".writebarrieriface (@\"\".dst·1 *any, @\"\".src·2 any)\n"
-	"func @\"\".writebarrierfat2 (@\"\".dst·1 *any, @\"\".src·2 any)\n"
-	"func @\"\".writebarrierfat3 (@\"\".dst·1 *any, @\"\".src·2 any)\n"
-	"func @\"\".writebarrierfat4 (@\"\".dst·1 *any, @\"\".src·2 any)\n"
+	"func @\"\".writebarrierfat2 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat3 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat4 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
 	"func @\"\".writebarrierfat (@\"\".typ·1 *byte, @\"\".dst·2 *any, @\"\".src·3 *any)\n"
 	"func @\"\".selectnbsend (@\"\".chanType·2 *byte, @\"\".hchan·3 chan<- any, @\"\".elem·4 *any) (? bool)\n"
 	"func @\"\".selectnbrecv (@\"\".chanType·2 *byte, @\"\".elem·3 *any, @\"\".hchan·4 <-chan any) (? bool)\n"

src/cmd/gc/racewalk.c

@@ -210,12 +210,7 @@ racewalknode(Node **np, NodeList **init, int wr, int skip)
 	case OCALLFUNC:
 		// Instrument dst argument of runtime.writebarrier* calls
 		// as we do not instrument runtime code.
-		if(n->left->sym != S && n->left->sym->pkg == runtimepkg &&
-			(strcmp(n->left->sym->name, "writebarrierptr") == 0 ||
-			strcmp(n->left->sym->name, "writebarrierstring") == 0 ||
-			strcmp(n->left->sym->name, "writebarrierslice") == 0 ||
-			strcmp(n->left->sym->name, "writebarrieriface") == 0 ||
-			strcmp(n->left->sym->name, "writebarrierfat") == 0)) {
+		if(n->left->sym != S && n->left->sym->pkg == runtimepkg && strncmp(n->left->sym->name, "writebarrier", 12) == 0) {
 			// Find the dst argument.
 			// The list can be reordered, so it's not necessary just the first or the second element.
 			for(l = n->list; l; l = l->next) {

src/cmd/gc/runtime.go

@@ -112,9 +112,13 @@ func writebarrierptr(dst *any, src any)
 func writebarrierstring(dst *any, src any)
 func writebarrierslice(dst *any, src any)
 func writebarrieriface(dst *any, src any)
-func writebarrierfat2(dst *any, src any)
-func writebarrierfat3(dst *any, src any)
-func writebarrierfat4(dst *any, src any)
+
+// The unused *byte argument makes sure that src is 2-pointer-aligned,
+// which is the maximum alignment on NaCl amd64p32
+// (and possibly on 32-bit systems if we start 64-bit aligning uint64s).
+func writebarrierfat2(dst *any, _ *byte, src any)
+func writebarrierfat3(dst *any, _ *byte, src any)
+func writebarrierfat4(dst *any, _ *byte, src any)
 func writebarrierfat(typ *byte, dst *any, src *any)
 
 func selectnbsend(chanType *byte, hchan chan<- any, elem *any) bool

src/cmd/gc/walk.c

@@ -2060,13 +2060,13 @@ applywritebarrier(Node *n, NodeList **init)
 				l, n->right);
 		} else if(t->width == 2*widthptr) {
 			n = mkcall1(writebarrierfn("writebarrierfat2", t, n->right->type), T, init,
-				l, n->right);
+				l, nodnil(), n->right);
 		} else if(t->width == 3*widthptr) {
 			n = mkcall1(writebarrierfn("writebarrierfat3", t, n->right->type), T, init,
-				l, n->right);
+				l, nodnil(), n->right);
 		} else if(t->width == 4*widthptr) {
 			n = mkcall1(writebarrierfn("writebarrierfat4", t, n->right->type), T, init,
-				l, n->right);
+				l, nodnil(), n->right);
 		} else {
 			r = n->right;
 			while(r->op == OCONVNOP)

src/runtime/mgc0.go

@@ -110,20 +110,20 @@ func writebarrieriface(dst *[2]uintptr, src [2]uintptr) {
 }
 
 //go:nosplit
-func writebarrierfat2(dst *[2]uintptr, src [2]uintptr) {
+func writebarrierfat2(dst *[2]uintptr, _ *byte, src [2]uintptr) {
 	dst[0] = src[0]
 	dst[1] = src[1]
 }
 
 //go:nosplit
-func writebarrierfat3(dst *[3]uintptr, src [3]uintptr) {
+func writebarrierfat3(dst *[3]uintptr, _ *byte, src [3]uintptr) {
 	dst[0] = src[0]
 	dst[1] = src[1]
 	dst[2] = src[2]
 }
 
 //go:nosplit
-func writebarrierfat4(dst *[4]uintptr, src [4]uintptr) {
+func writebarrierfat4(dst *[4]uintptr, _ *byte, src [4]uintptr) {
 	dst[0] = src[0]
 	dst[1] = src[1]
 	dst[2] = src[2]