xen/pv-on-hvm kexec: prevent crash in xenwatch_thread() when stale watch events arrive

During repeated kexec boots xenwatch_thread() can crash because
xenbus_watch->callback is cleared by xenbus_watch_path() if a node/token
combo for a new watch happens to match an already registered watch from
an old kernel.  In this case xs_watch returns -EEXISTS, then
register_xenbus_watch() does not remove the to-be-registered watch from
the list of active watches but returns the -EEXISTS to the caller
anyway.

Because the watch is still active in xenstored it will cause an event
which will arrive in the new kernel. process_msg() will find the
encapsulated struct xenbus_watch in its list of registered watches and
puts the "empty" watch handle in the queue for xenwatch_thread().
xenwatch_thread() then calls ->callback which was cleared earlier by
xenbus_watch_path().

To prevent that crash in a guest running on an old xen toolstack remove
the special -EEXIST handling.

v2:
 - remove the EEXIST handing in register_xenbus_watch() instead of
   checking for ->callback in process_msg()

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Olaf Hering <olaf@aepfle.de>

drivers/xen/xenbus/xenbus_xs.c

@@ -638,8 +638,7 @@ int register_xenbus_watch(struct xenbus_watch *watch)
 
 	err = xs_watch(watch->node, token);
 
-	/* Ignore errors due to multiple registration. */
-	if ((err != 0) && (err != -EEXIST)) {
+	if (err) {
 		spin_lock(&watches_lock);
 		list_del(&watch->list);
 		spin_unlock(&watches_lock);