cmd/internal/obj/arm64: reject misaligned stack frames, except empty frames

The layout code has to date insisted on stack frames that are 16-aligned
including the saved LR, and it ensured this by growing the frame itself.
This breaks code that refers to values near the top of the frame by positive
offset from SP, and in general it's too magical: if you see TEXT xxx, $N,
you expect that the frame size is actually N, not sometimes N and sometimes N+8.

This led to a serious bug in the compiler where ambiguously live values
were not being zeroed correctly, which in turn triggered an assertion
in the GC about finding only valid pointers. The compiler has been
fixed to always emit aligned frames, and the hand-written assembly
has also been fixed.

Now that everything is aligned, make unaligned an error instead of
something to "fix" silently.

For #9880.

Change-Id: I05f01a9df174d64b37fa19b36a6b6c5f18d5ba2d
Reviewed-on: https://go-review.googlesource.com/12848Reviewed-by: Austin Clements <austin@google.com>

src/cmd/internal/obj/arm64/obj7.go

@@ -553,7 +553,6 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym) {
 	var o int
 	var q2 *obj.Prog
 	var retjmp *obj.LSym
-	var stkadj int64
 	for p := cursym.Text; p != nil; p = p.Link {
 		o = int(p.As)
 		switch o {
@@ -567,9 +566,20 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym) {
 			if (cursym.Text.Mark&LEAF != 0) && ctxt.Autosize <= 8 {
 				ctxt.Autosize = 0
 			} else if ctxt.Autosize&(16-1) != 0 {
-				stkadj = 16 - (int64(ctxt.Autosize) & (16 - 1))
-				ctxt.Autosize += int32(stkadj)
-				cursym.Locals += int32(stkadj)
+				// The frame includes an LR.
+				// If the frame size is 8, it's only an LR,
+				// so there's no potential for breaking references to
+				// local variables by growing the frame size,
+				// because there are no local variables.
+				// But otherwise, if there is a non-empty locals section,
+				// the author of the code is responsible for making sure
+				// that the frame size is 8 mod 16.
+				if ctxt.Autosize == 8 {
+					ctxt.Autosize += 8
+					ctxt.Locals += 8
+				} else {
+					ctxt.Diag("%v: unaligned frame size %d - must be 8 mod 16 (or 0)", p, ctxt.Autosize-8)
+				}
 			}
 			p.To.Offset = int64(ctxt.Autosize) - 8
 			if ctxt.Autosize == 0 && !(cursym.Text.Mark&LEAF != 0) {

test/nosplit.go

@@ -295,7 +295,7 @@ TestCases:
 					}
 				}
 
-				if size%ptrSize == 4 {
+				if size%ptrSize == 4 || goarch == "arm64" && size != 0 && (size+8)%16 != 0 {
 					continue TestCases
 				}
 				nosplit := m[3]