Re: Re: [PATCH v3 -tip] [BUGFIX] x86/kprobes: Fix to recover instructions on optimized path

[Ingo Molnar <mingo at elte dot hu>]

* Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> wrote:

> (2012/02/27 18:34), Ingo Molnar wrote:
> > 
> > * Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> wrote:
> > 
> >> +
> >> +#ifdef CONFIG_OPTPROBES
> >> +static unsigned long __recover_optprobed_insn(struct kprobe *kp,
> >> +					      kprobe_opcode_t *buf,
> >> +					      unsigned long addr)
> >> +{
> >> +	long offs = addr - (unsigned long)kp->addr - 1;
> >> +	struct optimized_kprobe *op = container_of(kp, struct optimized_kprobe, kp);
> >> +
> >> +	/*
> >> +	 * If the kprobe can be optimized, original bytes which can be
> >> +	 * overwritten by jump destination address. In this case, original
> >> +	 * bytes must be recovered from op->optinsn.copied_insn buffer.
> >> +	 */
> >> +	memcpy(buf, (void *)addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
> >> +	if (addr == (unsigned long)kp->addr) {
> >> +		buf[0] = kp->opcode;
> >> +		memcpy(buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
> >> +	} else
> >> +		memcpy(buf, op->optinsn.copied_insn + offs, RELATIVE_ADDR_SIZE - offs);
> >> +
> >> +	return (unsigned long)buf;
> >> +}
> >> +#endif
> > 
> > Why not stick this into a new kprobes-opt.c file?
> 
> Would you mean that I should split all optprobe stuffs into 
> new file?

Yeah, that would be sensible I think - and it might help avoid 
similar complications in the future.

Could (and probably should) be done in a separate patch - to 
keep the bits that you already fixed and tested intact.

> > This should be a separate, kprobes_recover_opt() method and 
> > be inside kprobes-opt.c as well.
> 
> OK, I'll do that. But I think it should be separated work. 
> Just for the bugfix, I think this should go into this style, 
> because this should be pushed into stable tree too.

I don't think we can push such a large and complex looking patch 
into v3.3 (let alone into -stable) - it's v3.4 material, and 
that's why I asked for the cleaner split-out as well. This 
optprobes code is really non-obvious at the moment and a 
split-out might improve that and might make future fixes easier 
to merge.

Thanks,

	Ingo