Re: [PATCH 2/6] x86: allow to call text_poke_bp during boot

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Sat, Oct 19, 2013 at 03:19:19PM -0400, Steven Rostedt wrote:
> 
> [ Added Paul because he'll understand this ]
> 
> On Sat, 19 Oct 2013 15:16:58 -0400
> Steven Rostedt <rostedt@goodmis.org> wrote:
> 
> > On Sun, 20 Oct 2013 00:02:32 +0900
> > Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> wrote:
> > 
> > > (2013/10/18 23:27), Petr Mladek wrote:
> > > > We would like to use text_poke_bp in ftrace. It might be called also during
> > > > boot when the interupts are disabled. We need to enable them for syncing
> > > > the cores on each CPU. Otherwise, there might be a deadlock, see the
> > > > warning in "smp_call_function_many", kernel/smp.c:371.
> > > 
> > > Steven, is this really needed?
> > > I think if this is the special use(e.g. boottime test),
> > > we'd better to run it after boot...
> > > 
> > 
> > It's used to convert the calls to mcount to nops. But maybe a better
> > thing to do is to check if we only have a single CPU:
> > 
> > static void run_sync(void)
> > {
> > 	if (num_online_cpus() != 1)
> 
> Hmm, to be more robust to handle our future "ideal" machines, perhaps
> this should be:
> 
> 	/* Ideally we would like to run on zero CPUS! */
> 	if (num_online_cpus() < 2)

To be really safe, shouldn't you use complex numbers?  Just in case
you end up running on a system with 5i-3 CPUs or something.  ;-)

							Thanx, Paul

> -- Steve
> 
> 
> > 		on_each_cpu(do_sync_core, NULL, 1);
> > }
> > 
> > 
> > I believe that the only time we call this function with interrupts
> > disabled is before SMP is set up. Thus, the above change would handle
> > that case.
> > 
> > -- Steve
>