Re: [PATCH v2 (repost)] locking/lockdep: add debug_show_all_lock_holders()

[Waiman Long]

On 1/14/23 04:53, Tetsuo Handa wrote:
> On 2023/01/14 18:36, Ingo Molnar wrote:
> > Yeah, so note how you introduce a function with a parameter:
> >
> > 	void __debug_show_all_locks(bool show_stack)
> >
> > ... only to then *hide* the new parameter via helper functions:
> >
> > 	static inline void debug_show_all_locks(void)
> > 	{
> > 		__debug_show_all_locks(false);
> > 	}
> >
> > 	static inline void debug_show_all_lock_holders(void)
> > 	{
> > 		__debug_show_all_locks(true);
> > 	}
> >
> > ... which is a *strong* hint by our universe that the new parameter was
> > probably a bad idea to begin with.
> >
> > Given how small debug_show_all_locks() is to begin with, I'd suggest simply
> > duplicating the loop into debug_show_all_lock_holders() or so.
> Initial version at https://lkml.kernel.org/r/82af40cc-bf85-2b53-b8f9-dfc12e66a781@xxxxxxxxxxxxxxxxxxx
> was duplicating the loop.
>
> Waiman Long suggested me not to duplicate the loop at
> https://lkml.kernel.org/r/3e027453-fda4-3891-3ec3-5623f1525e56@xxxxxxxxxx .
>
> Please talk with Waiman. I'm fine with either approach.

My original concern was that two functions are very similar with some 
minor difference. My suggestion was to use a common helper to reduce the 
code redundancy and future maintenance.

I do have some nits about the patch. The show_stack parameter isn't 
informative. Maybe you can use show_tasks as the parameter name since 
the major difference is the calling of sched_show_task().

Define a new helper like debug_show_all_locks_tasks(bool show_tasks), 
use it directly in check_hung_uninterruptible_tasks() and make 
debug_show_all_lock() call debug_show_all_locks_tasks().

Ingo, will that OK with you?

Cheers,
Longman