This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: [PATCH -tip v4 6/6] [RFC] kprobes/x86: Call exception handlers directly from do_int3/do_debug
- From: Masami Hiramatsu <masami dot hiramatsu dot pt at hitachi dot com>
- To: Jiri Kosina <jkosina at suse dot cz>
- Cc: Steven Rostedt <rostedt at goodmis dot org>, Ingo Molnar <mingo at kernel dot org>, Andi Kleen <ak at linux dot intel dot com>, Ananth N Mavinakayanahalli <ananth at in dot ibm dot com>, Sandeepa Prabhu <sandeepa dot prabhu at linaro dot org>, Frederic Weisbecker <fweisbec at gmail dot com>, x86 at kernel dot org, lkml <linux-kernel at vger dot kernel dot org>, Andrew Morton <akpm at linux-foundation dot org>, Ingo Molnar <mingo at redhat dot com>, systemtap at sourceware dot org, "H. Peter Anvin" <hpa at zytor dot com>, Sasha Levin <sasha dot levin at oracle dot com>, Thomas Gleixner <tglx at linutronix dot de>, Seiji Aguchi <seiji dot aguchi at hds dot com>, "David S. Miller" <davem at davemloft dot net>
- Date: Thu, 12 Dec 2013 13:40:26 +0900
- Subject: Re: [PATCH -tip v4 6/6] [RFC] kprobes/x86: Call exception handlers directly from do_int3/do_debug
- Authentication-results: sourceware.org; auth=none
- References: <20131204012841 dot 22118 dot 82992 dot stgit at kbuild-fedora dot novalocal> <20131204012856 dot 22118 dot 91079 dot stgit at kbuild-fedora dot novalocal> <20131203213904 dot 5ff1794c at gandalf dot local dot home> <alpine dot LRH dot 2 dot 00 dot 1312111418550 dot 6318 at twin dot jikos dot cz>
(2013/12/11 22:31), Jiri Kosina wrote:
> On Tue, 3 Dec 2013, Steven Rostedt wrote:
>
>>> To avoid a kernel crash by probing on lockdep code, call
>>> kprobe_int3_handler and kprobe_debug_handler directly
>>> from do_int3 and do_debug. Since there is a locking code
>>> in notify_die, lockdep code can be invoked. And because
>>> the lockdep involves printk() related things, theoretically,
>>> we need to prohibit probing on much more code...
>>>
>>> Anyway, most of the int3 handlers in the kernel are already
>>> called from do_int3 directly, e.g. ftrace_int3_handler,
>>> poke_int3_handler, kgdb_ll_trap. Actually only
>>> kprobe_exceptions_notify is on the notifier_call_chain.
>>>
>>> So I think this is not a crazy thing.
>>
>> What? Oh, yeah. No, using notifiers in int3 handler is the crazy
>> thing ;-)
>
> Yeah, it's broken. Obviously, if you happen to trigger int3 before the
> notifier has been registered, it'd cause int3 exception to be unhandled.
> See
>
> commit 17f41571bb2c4a398785452ac2718a6c5d77180e
> Author: Jiri Kosina <jkosina@suse.cz>
> Date: Tue Jul 23 10:09:28 2013 +0200
>
> kprobes/x86: Call out into INT3 handler directly instead of using notifier
>
> for one such issue that happened with jump labels.
>
>> Hmm, if there's no users of the int3 notifier, should we just remove it?
>
> Hmm, there are still uprobes, right?
Right, uprobes still use it, however, since it only handles user-space
breakpoint, there is no problem.
Thank you!
--
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com