Re: [for-next][PATCH 12/16] kprobes: Initialize kprobes at postcore_initcall

From: Masami Hiramatsu
Date: Tue Jul 09 2019 - 08:30:56 EST


Hi Steve,

On Wed, 3 Jul 2019 10:02:05 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Tue, 2 Jul 2019 17:50:09 +0100
> Mark Rutland <mark.rutland@xxxxxxx> wrote:
>
> > On Sun, May 26, 2019 at 03:18:40PM -0400, Steven Rostedt wrote:
> > > From: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> > >
> > > Initialize kprobes at postcore_initcall level instead of module_init
> > > since kprobes is not a module, and it depends on only subsystems
> > > initialized in core_initcall.
> > > This will allow ftrace kprobe event to add new events when it is
> > > initializing because ftrace kprobe event is initialized at
> > > later initcall level.
> > >
> > > Link: http://lkml.kernel.org/r/155851394736.15728.13626739508905120098.stgit@devnote2
> > >
> > > Signed-off-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> > > Signed-off-by: Steven Rostedt (VMware) <rostedt@xxxxxxxxxxx>
> > > ---
> > > kernel/kprobes.c | 3 +--
> > > 1 file changed, 1 insertion(+), 2 deletions(-)
> > >
> > > diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> > > index b1ea30a5540e..54aaaad00a47 100644
> > > --- a/kernel/kprobes.c
> > > +++ b/kernel/kprobes.c
> > > @@ -2289,6 +2289,7 @@ static int __init init_kprobes(void)
> > > init_test_probes();
> > > return err;
> > > }
> > > +postcore_initcall(init_kprobes);
> >
> > As a heads-up, this is causing boot-time failures on arm64.
>
> Thanks for the report.
>
> >
> > On arm64 kprobes depends on the BRK handler we register in
> > debug_traps_init(), which is an arch_initcall.
> >
> > As of this change, init_krprobes() calls init_test_probes() before
> > that's registered, so we end up hitting a BRK before we can handle it.
> >
>
> Would something like this help?
>
> -- Steve
>
> diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> index 5471efbeb937..0ca6f53c8505 100644
> --- a/kernel/kprobes.c
> +++ b/kernel/kprobes.c
> @@ -2235,6 +2235,8 @@ static struct notifier_block kprobe_module_nb = {
> extern unsigned long __start_kprobe_blacklist[];
> extern unsigned long __stop_kprobe_blacklist[];
>
> +static bool run_kprobe_tests __initdata;
> +
> static int __init init_kprobes(void)
> {
> int i, err = 0;
> @@ -2286,11 +2288,18 @@ static int __init init_kprobes(void)
> kprobes_initialized = (err == 0);
>
> if (!err)
> - init_test_probes();
> + run_kprobe_tests = true;
> return err;
> }
> subsys_initcall(init_kprobes);


Just out of curious, if arm64's handler code initialized in arch_initcall,
why this subsys_initcall() function causes a problem?

This is actually related to my boot-time tracing series, so I would like
fix this issue without this patch.

Thank you,

>
> +static int __init run_init_test_probes(void)
> +{
> + if (run_kprobe_tests)
> + init_test_probes();
> +}
> +module_init(run_init_test_probes);
> +
> #ifdef CONFIG_DEBUG_FS
> static void report_probe(struct seq_file *pi, struct kprobe *p,
> const char *sym, int offset, char *modname, struct kprobe *pp)
>


--
Masami Hiramatsu <mhiramat@xxxxxxxxxx>