Re: [PATCH] ftrace: move sysctl_ftrace_enabled to ftrace.c

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 22 Feb 2022 17:52:59 -0800
Luis Chamberlain <mcgrof@kernel.org> wrote:

> On Wed, Feb 23, 2022 at 09:23:11AM +0800, Wei Xiao wrote:
> > This moves ftrace_enabled to trace/ftrace.c.  
> 
> Hey Wei, thanks for you patch!
>                                                                                                                                                                                               
> This does not explain how this is being to help with maitenance as
> otherwise this makes kernel/sysctl.c hard to maintain and we also tend
> to get many conflicts. It also does not explain how all the filesystem
> sysctls are not gone and that this is just the next step, moving slowly
> the rest of the sysctls. Explaining this in the commit log will help
> patch review and subsystem maintainers understand the conext / logic
> behind the move.
>                                                                                                                                                                                               
> I'd be more than happy to take this if ftrace folks Ack. To avoid conflicts
> I can route this through sysctl-next which is put forward in particular
> to avoid conflicts across trees for this effort. Let me know.

I'm fine with this change going through another tree.

Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org>

for your convenience.

> 
> > diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> > index f9feb197b2da..4a5b4d6996a4 100644
> > --- a/kernel/trace/ftrace.c
> > +++ b/kernel/trace/ftrace.c
> > @@ -7846,7 +7846,8 @@ static bool is_permanent_ops_registered(void)
> >  	return false;
> >  }
> >  
> > -int
> > +#ifdef CONFIG_SYSCTL
> > +static int  
> 
> Is the ifdef really needed? It was not there before, ie, does
> ftrace not depend on sysctls? I don't see a direct relationship
> but I do wonder if its implicit.

I think because the function is now static, and that it now includes the
structure itself, that the #ifdef is added. When I first saw it, I was
a bit uncomfortable with adding more #ifdefs, but for this use case, I
believe it's OK.

> 
> >  ftrace_enable_sysctl(struct ctl_table *table, int write,
> >  		     void *buffer, size_t *lenp, loff_t *ppos)
> >  {
> > @@ -7889,3 +7890,22 @@ ftrace_enable_sysctl(struct ctl_table *table, int write,
> >  	mutex_unlock(&ftrace_lock);
> >  	return ret;
> >  }
> > +
> > +static struct ctl_table ftrace_sysctls[] = {
> > +	{
> > +		.procname       = "ftrace_enabled",
> > +		.data           = &ftrace_enabled,
> > +		.maxlen         = sizeof(int),
> > +		.mode           = 0644,
> > +		.proc_handler   = ftrace_enable_sysctl,
> > +	},
> > +	{}
> > +};
> > +
> > +static int __init ftrace_sysctl_init(void)
> > +{
> > +	register_sysctl_init("kernel", ftrace_sysctls);
> > +	return 0;
> > +}
> > +late_initcall(ftrace_sysctl_init);
> > +#endif  
> 
> There's other __init calls already on ftrace, would this be better
> placed in one of them, and then have this be a no-op iff we determine
> ftrace can be built without sysctls and then have a no-op for when
> sysctls are disabled.

I rather have the initcall here, as the other initcalls are special
needs, and not generic functions.

-- Steve