kprobes: Disabling optimized kprobes for entry text section

You can crash the kernel (with root/admin privileges) using kprobe tracer by running:

 echo "p system_call_after_swapgs" > ./kprobe_events
 echo 1 > ./events/kprobes/enable

The reason is that at the system_call_after_swapgs label, the
kernel stack is not set up. If optimized kprobes are enabled,
the user space stack is being used in this case (see optimized
kprobe template) and this might result in a crash.

There are several places like this over the entry code
(entry_$BIT). As it seems there's no any reasonable/maintainable
way to disable only those places where the stack is not ready, I
switched off the whole entry code from kprobe optimizing.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: acme@redhat.com
Cc: fweisbec@gmail.com
Cc: ananth@in.ibm.com
Cc: davem@davemloft.net
Cc: a.p.zijlstra@chello.nl
Cc: eric.dumazet@gmail.com
Cc: 2nddept-manager@sdl.hitachi.co.jp
LKML-Reference: <1298298313-5980-3-git-send-email-jolsa@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

arch/x86/kernel/kprobes.c

@@ -1276,6 +1276,14 @@ static int __kprobes can_optimize(unsigned long paddr)
 	if (!kallsyms_lookup_size_offset(paddr, &size, &offset))
 		return 0;
 
+	/*
+	 * Do not optimize in the entry code due to the unstable
+	 * stack handling.
+	 */
+	if ((paddr >= (unsigned long )__entry_text_start) &&
+	    (paddr <  (unsigned long )__entry_text_end))
+		return 0;
+
 	/* Check there is enough space for a relative jump. */
 	if (size - offset < RELATIVEJUMP_SIZE)
 		return 0;