Re: lockdep tracing and using of printk return value ?

[Joe Perches <joe@perches.com>]

On Tue, 2020-05-19 at 16:45 +0200, Peter Zijlstra wrote:
> On Tue, May 19, 2020 at 05:41:47AM -0700, Joe Perches wrote:
> > Except for some ancient code in drivers/scsi, this code
> > may be the only kernel use of the printk return value.
> 
> Is using the printk() return value a problem?

Maybe.  The printk code isn't obviously correct.

https://lore.kernel.org/lkml/20200518204415.d1a3adaba597ce5b232b4b2a@linux-foundation.org/

> > Code that uses the printk return value in
> > kernel/locking/lockdep.c is odd because the printk
> > return length includes both the length of a KERN_<LEVEL>
> > prefix and the newline.  depth also seems double counted.
> 
> Yeah, it seems dodgy. OTOH printk() really ought to discard the
> KERN_<level> crud from the return size.

Or change it to the actual output of '<' level '>'
instead of the internal KERN_SOH level.

> > Perhaps there's a better way to calculate this?
> > 
> > Maybe:
> > ---
> >  kernel/locking/lockdep.c | 8 +++-----
> >  1 file changed, 3 insertions(+), 5 deletions(-)
> > 
> > diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
> > index 2fadc2635946..265227edc550 100644
> > --- a/kernel/locking/lockdep.c
> > +++ b/kernel/locking/lockdep.c
> > @@ -1960,11 +1960,9 @@ static void print_lock_class_header(struct lock_class *class, int depth)
> >  
> >  	for (bit = 0; bit < LOCK_USAGE_STATES; bit++) {
> >  		if (class->usage_mask & (1 << bit)) {
> > -			int len = depth;
> > -
> > -			len += printk("%*s   %s", depth, "", usage_str[bit]);
> > -			len += printk(KERN_CONT " at:\n");
> > -			print_lock_trace(class->usage_traces[bit], len);
> > +			printk("%*s   %s at:\n", depth, "", usage_str[bit]);
> > +			print_lock_trace(class->usage_traces[bit],
> > +					 depth + 3 + strlen(usage_str[bit]);
> >  		}
> >  	}
> >  	printk("%*s }\n", depth, "");
> 
> Doesn't seem crazy...