Re: [PATCH 3/8] arm64: kprobes: Record frame pointer with kretprobe instance

[Mark Rutland <mark.rutland@arm.com>]

On Thu, Oct 14, 2021 at 07:01:55PM +0900, Masami Hiramatsu wrote:
> On Thu, 14 Oct 2021 10:13:32 +0100
> Mark Rutland <mark.rutland@arm.com> wrote:
> 
> > On Thu, Oct 14, 2021 at 05:04:05PM +0900, Masami Hiramatsu wrote:
> > > On Wed, 13 Oct 2021 11:01:39 +0100
> > > Mark Rutland <mark.rutland@arm.com> wrote:
> > > 
> > > > On Fri, Oct 08, 2021 at 09:28:39PM +0900, Masami Hiramatsu wrote:
> > > > > Record the frame pointer instead of stack address with kretprobe
> > > > > instance as the identifier on the instance list.
> > > > > Since arm64 always enable CONFIG_FRAME_POINTER, we can use the
> > > > > actual frame pointer (x29).
> > > > 
> > > > Just to check, why do we need to use the FP rather than SP? It wasn't
> > > > clear to me if that's necessary later in the series, or if I'm missing
> > > > something here.
> > > 
> > > Actually, this is for finding correct return address from the per-task
> > > kretprobe instruction list (suppose it as a shadow stack) when it will
> > > be searched in stack-backtracing. At that point, the framepointer will
> > > be a reliable key.
> > 
> > Sure, my question was more "why isn't the SP a reliable key?", because both
> > the SP and FP should be balanced at function-entry and function-return
> > time. I'm asking because I think I'm missing a subtlety.
> 
> Ah, because SP is not used while unwinding frame. For the kretprobe,
> either FP or SP is OK. But for the stacktrace.c, I can not use SP
> and is easy to change to use FP. :)

Ah, so this is just so that stacktrace can match the address. For
clarity, would you be happy to add a sentence to the commit message like:

| This will allow the stacktrace code to find the original return
| address from the FP alone.

Thanks,
Mark.