Re: [RFC/RFT 2/2] RISC-V: kprobes/kretprobe support

[Zong Li <zong.li@sifive.com>]

On Thu, Dec 19, 2019 at 5:28 PM Zong Li <zong.li@sifive.com> wrote:
>
> On Wed, Dec 18, 2019 at 9:10 PM Patrick Stählin <me@packi.ch> wrote:
> >
> > Hi all
> >
> > On 18.12.19 10:14, Zong Li wrote:
> > > On Wed, Dec 18, 2019 at 5:09 PM Zong Li <zong.li@sifive.com> wrote:
> > >>
> > >> From: mhiramat@kernel.org (Masami Hiramatsu)
> > >>
> > >> On Wed, 14 Nov 2018 21:52:57 +0100
> > >> Patrick Staehlin <me@packi.ch> wrote:
> > >>
> > >>> Yeah, I think it's simpler.
> > >>>
> > >>> And I found that the kprobe_breakpoint_handler() was called without
> > >>> checking !user_mode(regs). In that case, you should add the check in
> > >>> front of kprobe_breakpoint_handler() call.
> > >>>
> > >>> Thank you,
> > >>
> > >> Hi all,
> > >>
> > >> Is there any update? I was wondering if this patch are keep going? If
> > >> not, I think I could pick it up to go head
> >
> > I am still working on it, albeit slowly, holiday season coming up here
> > in Switzerland may accelerate that a bit. All the feedback I got from
> > Masami has been implemented.
>
> As Masami's suggestion, I don't see stop_machine in your implementation [1],
> are there some concerns to use it on SMP?

Hi all,
I introduced the patching code framework. Maybe it is helpful for
self-modify code here. (Please see:
https://lkml.org/lkml/2020/3/9/679)
This patch series is reviewing, maybe we could apply the
implementation after it was merged.


>
> > Additionally I added instruction simulation for everything except memory
> > accesses. I am currently working on getting compressed instructions
> > decoded properly into regular instructions but that is very tedious work.
> > I guess I am two or three full days of work away from getting a RFC/RFT
> > v2 series ready I guess that will happen in early January.
>
> Thanks for the efforts.
>
> >
> > What I currently have is at [1], that is mostly untested as of now
> > (beware I will rebase/squash that branch regularly). What I could use
> > help with in the future (as in for v2), is somebody testing it on real
> > hardware, as I've never gotten any testing feedback on my original
> > patch-series (that would not have worked properly because of the lacking
> > cache-flush).
>
> I give some quick tests by using kprobe sanity test and simple LKM on
> Hifive Unleashed board,
> it seems to work normally. I could help to test your next version patch as well.
>
> >
> > [1] https://github.com/packi/linux/tree/kprobes-riscv
> >
> > Patrick