Re: [PATCH] kprobes: Fix to init kprobes in subsys_initcall

[Masami Hiramatsu]

On Tue, 4 Jun 2019 05:06:29 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Mon,  3 Jun 2019 22:04:42 +0900
> Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
> 
> > Since arm64 kernel initializes breakpoint trap vector in arch_initcall(),
> > initializing kprobe (and run smoke test) in postcore_initcall() causes
> > a kernel panic.
> > 
> > To fix this issue, move the kprobe initialization in subsys_initcall()
> > (which is called right afer the arch_initcall).
> > 
> > In-kernel kprobe users (ftrace and bpf) are using fs_initcall() which is
> > called after subsys_initcall(), so this shouldn't cause more problem.
> > 
> > Reported-by: Anders Roxell <anders.roxell@xxxxxxxxxx>
> > Fixes: b5f8b32c93b2 ("kprobes: Initialize kprobes at postcore_initcall")
> 
> Hi Masami,
> 
> Is this really a fix, or is this just needed to add kprobes to the command line?

Hi Steve,

Yes, this is a real fix, not related to the kprobes command line feature.

First of all, the problem was that kprobes(module_init) was initialized
later than trace_kprobe(fs_initcall). the commit b5f8b32c93b2 is to solve
it. But I missed that the break-point hander initialization can be done 
in arch_initcall on some archs. This is fixing that bug.

Thank you,


> 
> -- Steve
> 
> 
> > Signed-off-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> > ---
> >  kernel/kprobes.c |    2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> > index 54aaaad00a47..5471efbeb937 100644
> > --- a/kernel/kprobes.c
> > +++ b/kernel/kprobes.c
> > @@ -2289,7 +2289,7 @@ static int __init init_kprobes(void)
> >  		init_test_probes();
> >  	return err;
> >  }
> > -postcore_initcall(init_kprobes);
> > +subsys_initcall(init_kprobes);
> >  
> >  #ifdef CONFIG_DEBUG_FS
> >  static void report_probe(struct seq_file *pi, struct kprobe *p,
> 


-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>