Re: [PATCH v4 1/4] sysctl: Add register_sysctl_init() interface

[Luis Chamberlain <mcgrof@kernel.org>]

On Fri, May 29, 2020 at 03:27:22PM +0800, Xiaoming Ni wrote:
> On 2020/5/29 15:09, Luis Chamberlain wrote:
> > On Tue, May 19, 2020 at 11:31:08AM +0800, Xiaoming Ni wrote:
> > > --- a/kernel/sysctl.c
> > > +++ b/kernel/sysctl.c
> > > @@ -3358,6 +3358,25 @@ int __init sysctl_init(void)
> > >   	kmemleak_not_leak(hdr);
> > >   	return 0;
> > >   }
> > > +
> > > +/*
> > > + * The sysctl interface is used to modify the interface value,
> > > + * but the feature interface has default values. Even if register_sysctl fails,
> > > + * the feature body function can also run. At the same time, malloc small
> > > + * fragment of memory during the system initialization phase, almost does
> > > + * not fail. Therefore, the function return is designed as void
> > > + */
> > 
> > Let's use kdoc while at it. Can you convert this to proper kdoc?
> > 
> Sorry, I do n’t know the format requirements of Kdoc, can you give me some
> tips for writing?

Sure, include/net/mac80211.h is a good example.

> > > +void __init register_sysctl_init(const char *path, struct ctl_table *table,
> > > +				 const char *table_name)
> > > +{
> > > +	struct ctl_table_header *hdr = register_sysctl(path, table);
> > > +
> > > +	if (unlikely(!hdr)) {
> > > +		pr_err("failed when register_sysctl %s to %s\n", table_name, path);
> > > +		return;
> > 
> > table_name is only used for this, however we can easily just make
> > another _register_sysctl_init() helper first, and then use a macro
> > which will concatenate this to something useful if you want to print
> > a string. I see no point in the description for this, specially since
> > the way it was used was not to be descriptive, but instead just a name
> > followed by some underscore and something else.
> > 
> Good idea, I will fix and send the patch to you as soon as possible

No rush :)

> > > +	}
> > > +	kmemleak_not_leak(hdr);
> > 
> > Is it *wrong* to run kmemleak_not_leak() when hdr was not allocated?
> > If so, can you fix the sysctl __init call itself?
> I don't understand here, do you mean that register_sysctl_init () does not
> need to call kmemleak_not_leak (hdr), or does it mean to add check hdr
> before calling kmemleak_not_leak (hdr) in sysctl_init ()?

I'm asking that the way you are adding it, you don't run
kmemleak_not_leak(hdr) if the hdr allocation filed. If that is
right then it seems that sysctl_init() might not be doing it
right.

Can that code be shared somehow?

  Luis