Re: CONFIG_RETHUNK int3 filling prevents kprobes in function body

[Peter Zijlstra]

On Tue, Sep 06, 2022 at 10:33:29AM +0900, Masami Hiramatsu wrote:
> On Mon, 5 Sep 2022 17:52:29 +0200
> Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> 
> > On Mon, Sep 05, 2022 at 05:09:16PM +0200, Peter Zijlstra wrote:
> > 
> > > > This is because kprobes decodes function body to ensure the probed address
> > > > is an instruction boundary, and if it finds the 0xcc (int3), it stops
> > > > decoding and reject probing because the int3 is usually used for a
> > > > software breakpoint and is replacing some other instruction. Without
> > > > recovering the instruction, it can not continue decoding safely.
> > > 
> > > I can't follow this logic. Decoding the single byte int3 instruction is
> > > trivial. If you want a sanity check, follow the branches you found while
> > > decoding the instruction starting at +0.
> > 
> > Specifically, kprobe is the only one scribbling random [*] instructions
> > with int3 in kernel text, so if kprobes doesn't know about the int3, it
> > must be padding.
> 
> No, kgdb is also handles int3 for its breakpoint. Of course we can
> ignore it or ask kgdb to expose the API to decode it.

I'm thinking kgdb has worse issues anyway. Much of it seems to be
wishful thinking.