Re: [PATCH] init/main.c: log initcall level when initcall_debug is used

On Thursday, 3 April 2025 at 10:27:14 Geert Uytterhoeven <[email protected]> wrote:
> Hi Andrew,
> 
> On Thu, 3 Apr 2025 at 04:56, Andrew Morton <[email protected]> wrote:
> > On Sun, 16 Mar 2025 21:50:15 +0100 Francesco Valla <[email protected]> wrote:
> > > When initcall_debug is specified on the command line, the start and
> > > return point for each initcall is printed. However, no information on
> > > the initcall level is reported.
> > >
> > > Add to the initcall_debug infrastructure an additional print that
> > > informs when a new initcall level is entered. This is particularly
> > > useful when debugging dependency chains and/or working on boot time
> > > reduction.
> > >
> > > ...
> > >
> > > --- a/init/main.c
> > > +++ b/init/main.c
> > > @@ -1214,6 +1214,12 @@ trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
> > >                fn, ret, (unsigned long long)ktime_us_delta(rettime, *calltime));
> > >  }
> > >
> > > +static __init_or_module void
> > > +trace_initcall_level_cb(void *data, const char *level)
> > > +{
> > > +     printk(KERN_DEBUG "entering initcall level: %s\n", level);
> > > +}
> >
> > Please review and test this fixlet:
> >
> > --- a/init/main.c~init-mainc-log-initcall-level-when-initcall_debug-is-used-fix
> > +++ a/init/main.c
> > @@ -1217,7 +1217,7 @@ trace_initcall_finish_cb(void *data, ini
> >  static __init_or_module void
> >  trace_initcall_level_cb(void *data, const char *level)
> >  {
> > -       printk(KERN_DEBUG "entering initcall level: %s\n", level);
> > +       pr_debug("entering initcall level: %s\n", level);
> >  }
> >
> >  static ktime_t initcall_calltime;
> 
> I think the "printk(KERN_DEBUG ...)" construct is intentional.
> The message should be logged when "initcall_debug" is passed on
> the kernel command line, while pr_debug() is a no-op unless DEBUG is
> defined inside the source file.
> 
> See also the two existing users in init/main.c near
> https://elixir.bootlin.com/linux/v6.13.7/source/init/main.c#L1207.
> 

Yes, printk(KERN_DEBUG ...) here is intentional, or it would be removed,
as Geert is correctly saying.

We have another occurrence of this for the "probe" messages also related
to initcall_debug, with a nice explanatory comment:

https://elixir.bootlin.com/linux/v6.14-rc6/source/drivers/base/dd.c#L741


Thank you!

Francesco