Got it! :)
I will improve the commit message and send v3 ASAP.
BTW, which tree should I rebase those patches onto? Is that the
for-next branch of
git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace.git. I saw
Jiri had troubles
applying those since these works are based on v6.4.0.
THX,
Ze
On Wed, May 17, 2023 at 10:54 AM Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
>
> On Wed, 17 May 2023 09:54:53 +0800
> Ze Gao <zegao2021@xxxxxxxxx> wrote:
>
> > Oops, I misunderstood your comments before.
> >
> > Yes, it's not necessary to do this reordering as regards to kprobe.
>
> Let me confirm, I meant that your current patch is correct. I just mentioned
> that kprobe_busy_{begin,end} will continue use standard version because
> kprobe itself handles that. Please update only the patch description and
> add my ack.
>
> Acked-by: Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
>
> If you add Steve's call graph for the explanation, it will help us to
> understand what will be fixed.
>
> Thank you,
>
> >
> > Thanks for your review.
> >
> > I'll rebase onto the latest tree and send v3 ASAP.
> >
> > Regards,
> > Ze
> >
> > On Wed, May 17, 2023 at 12:03 AM Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
> > >
> > > On Tue, 16 May 2023 17:47:52 +0800
> > > Ze Gao <zegao2021@xxxxxxxxx> wrote:
> > >
> > > > Precisely, these that are called within kprobe_busy_{begin, end},
> > > > which the previous patch does not resolve.
> > >
> > > Note that kprobe_busy_{begin,end} don't need to use notrace version
> > > because kprobe itself prohibits probing on preempt_count_{add,sub}.
> > >
> > > Thank you,
> > >
> > > > I will refine the commit message to make it clear.
> > > >
> > > > FYI, details can checked out here:
> > > > Link: https://lore.kernel.org/linux-trace-kernel/20230516132516.c902edcf21028874a74fb868@xxxxxxxxxx/
> > > >
> > > > Regards,
> > > > Ze
> > > >
> > > > On Tue, May 16, 2023 at 5:18 PM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > > > >
> > > > > On Tue, May 16, 2023 at 03:18:28PM +0800, Ze Gao wrote:
> > > > > > Current implementation calls kprobe related functions before doing
> > > > > > ftrace recursion check in fprobe_kprobe_handler, which opens door
> > > > > > to kernel crash due to stack recursion if preempt_count_{add, sub}
> > > > > > is traceable.
> > > > >
> > > > > Which preempt_count*() are you referring to? The ones you just made
> > > > > _notrace in the previous patch?
> > >
> > >
> > > --
> > > Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
>
>
> --
> Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
