Re: [PATCH 0/4] ftrace: prevent ftrace modifications while being kprobe'd

[Ingo Molnar <mingo@elte.hu>]

* Abhishek Sagar <sagar.abhishek@gmail.com> wrote:

> Hi Steven/Ingo,
> 
> These patches address the problem of kprobe registration interfering 
> with dynamic ftrace. A kprobe registered on an mcount call-site can 
> modify that address multiple times during its duration of probing. At 
> the same time, ftrace may simultaneously modify these locations. This 
> could be fatal especially if ftrace modifies the call-site during a 
> kprobe registration on it (between arch_prepare_kprobe and 
> arch_arm_kprobe in __register_kprobe to be more precise). So as a 
> "fix", I've disabled any updates on the mcount call-site while it's 
> being kprobe'd.

applied to tip/tracing/ftrace, thanks Abhishek.

note that i had to do a number of fixups to themerge, and there was one 
chunk which did not merge cleanly and i left it out for now:

@@ -435,8 +502,12 @@ static void ftrace_replace_code(int enable)
                                continue;

                        /* ignore updates to this record's mcount site */
-                       if (get_kprobe((void *)rec->ip))
+                       if (get_kprobe((void *)rec->ip)) {
+                               freeze_record(rec);
                                continue;
+                       } else {
+                               unfreeze_record(rec);
+                       }

please send a delta patch against tip/tracing/ftrace to get that kprobes 
fix. You can pick up -tip via:

 http://people.redhat.com/mingo/tip.git/README

	Ingo