Re: [PATCH -tip v14 03/12] kprobes: checks probe address is instruction boudary on x86

From: Frederic Weisbecker <fweisbec@gmail.com>
To: Masami Hiramatsu <mhiramat@redhat.com>
Cc: "Ingo Molnar" <mingo@elte.hu>,
	"Steven Rostedt" <rostedt@goodmis.org>,
	lkml <linux-kernel@vger.kernel.org>,
	systemtap <systemtap@sources.redhat.com>,
	kvm <kvm@vger.kernel.org>,
	DLE <dle-develop@lists.sourceforge.net>,
	"Ananth N Mavinakayanahalli" <ananth@in.ibm.com>,
	"Avi Kivity" <avi@redhat.com>, "Andi Kleen" <ak@linux.intel.com>,
	"Christoph Hellwig" <hch@infradead.org>,
	"Frank Ch. Eigler" <fche@redhat.com>,
	"H. Peter Anvin" <hpa@zytor.com>,
	"Jason Baron" <jbaron@redhat.com>,
	"Jim Keniston" <jkenisto@us.ibm.com>,
	"K.Prasad" <prasad@linux.vnet.ibm.com>,
	"Lai Jiangshan" <laijs@cn.fujitsu.com>,
	"Li Zefan" <lizf@cn.fujitsu.com>,
	PrzemysławPawełczyk <przemyslaw@pawelczyk.it>,
	"Roland McGrath" <roland@redhat.com>,
	"Sam Ravnborg" <sam@ravnborg.org>,
	"Srikar Dronamraju" <srikar@linux.vnet.ibm.com>,
	"Tom Zanussi" <tzanussi@gmail.com>,
	"Vegard Nossum" <vegard.nossum@gmail.com>
Subject: Re: [PATCH -tip v14 03/12] kprobes: checks probe address is instruction boudary on x86
Date: Wed, 19 Aug 2009 02:46:29 +0200	[thread overview]
Message-ID: <20090819004627.GH5231@nowhere> (raw)
In-Reply-To: <4A8B4515.3090508@redhat.com>

On Tue, Aug 18, 2009 at 08:19:33PM -0400, Masami Hiramatsu wrote:
> Frederic Weisbecker wrote:
> > On Tue, Aug 18, 2009 at 07:17:39PM -0400, Masami Hiramatsu wrote:
> >> Frederic Weisbecker wrote:
> >>>> +	while (addr < paddr) {
> >>>> +		kernel_insn_init(&insn, (void *)addr);
> >>>> +		insn_get_opcode(&insn);
> >>>> +
> >>>> +		/* Check if the instruction has been modified. */
> >>>> +		if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION) {
> >>>> +			ret = recover_probed_instruction(buf, addr);
> >>>
> >>>
> >>>
> >>> I'm confused about the reason of this recovering. Is it to remove
> >>> kprobes behind the current setting one in the current function?
> >>
> >> No, it recovers just an instruction which is probed by a kprobe,
> >> because we need to know the first byte of this instruction for
> >> decoding it.
> 
> Ah, sorry, it was not accurate. the function recovers an instruction
> on the buffer(buf), not on the real kernel text. :)

Ah ok. I'll just add a small comment about that then, and apply
it.

> >>
> >> Perhaps we'd better to have more generic interface (text_peek?)
> >> for it because another subsystem (e.g. kgdb) may want to insert int3...
> >>
> >> Thank you,
> > 
> > 
> > Aah, I see now, it's to keep a sane check of the instructions
> > boundaries without int 3 artifacts in the middle.
> > 
> > But in that case, you should re-arm the breakpoint after your
> > check, right?
> > 
> > Or may be you could do the check without repatching?
> 
> Yes, it doesn't modify kernel text, just recover an original
> instruction from kernel text and backup byte on a buffer.

Ok.

> > May be by doing a copy of insn.opcode.bytes and replacing bytes[0]
> > with what a random kprobe has stolen?
> 
> Hm, no, this function is protected from other kprobes by kprobe_mutex.
> 
> Thank you,

Right, thanks!

> -- 
> Masami Hiramatsu
> 
> Software Engineer
> Hitachi Computer Products (America), Inc.
> Software Solutions Division
> 
> e-mail: mhiramat@redhat.com
> 