runtime: switch to systemstack before throw in casgstatus

CL 580255 increased the frame size of entersyscall and reentersyscall,
which is causing the x/sys repository to fail to build for
windows/arm64 because of an overflow of the nosplit stack reservation.

Fix this by wrapping the other call to throw in casgstatus in a system
stack switch. This is a fatal throw anyway indicating a core runtime
invariant is broken, so this path is basically never taken. This cuts
off the nosplit frame chain and allows x/sys to build.

Change-Id: I00b16c9db3a7467413ed48953c7f8a9a750f000a
Reviewed-on: https://go-review.googlesource.com/c/go/+/580775
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
diff --git a/src/runtime/proc.go b/src/runtime/proc.go
index e469f20..56f97fa 100644
--- a/src/runtime/proc.go
+++ b/src/runtime/proc.go
@@ -1105,6 +1105,8 @@
 func casgstatus(gp *g, oldval, newval uint32) {
 	if (oldval&_Gscan != 0) || (newval&_Gscan != 0) || oldval == newval {
 		systemstack(func() {
+			// Call on the systemstack to prevent print and throw from counting
+			// against the nosplit stack reservation.
 			print("runtime: casgstatus: oldval=", hex(oldval), " newval=", hex(newval), "\n")
 			throw("casgstatus: bad incoming values")
 		})
@@ -1120,7 +1122,11 @@
 	// GC time to finish and change the state to oldval.
 	for i := 0; !gp.atomicstatus.CompareAndSwap(oldval, newval); i++ {
 		if oldval == _Gwaiting && gp.atomicstatus.Load() == _Grunnable {
-			throw("casgstatus: waiting for Gwaiting but is Grunnable")
+			systemstack(func() {
+				// Call on the systemstack to prevent throw from counting
+				// against the nosplit stack reservation.
+				throw("casgstatus: waiting for Gwaiting but is Grunnable")
+			})
 		}
 		if i == 0 {
 			nextYield = nanotime() + yieldDelay