Re: [PATCH v5 6/7] locking/lockdep: Reuse freed chain_hlocks entries

[Peter Zijlstra]

On Tue, Feb 04, 2020 at 11:45:15AM -0500, Waiman Long wrote:
> On 2/4/20 7:36 AM, Peter Zijlstra wrote:
> > --- a/kernel/locking/lockdep_proc.c
> > +++ b/kernel/locking/lockdep_proc.c
> > @@ -278,9 +278,11 @@ static int lockdep_stats_show(struct seq
> >  #ifdef CONFIG_PROVE_LOCKING
> >  	seq_printf(m, " dependency chains:             %11lu [max: %lu]\n",
> >  			lock_chain_count(), MAX_LOCKDEP_CHAINS);
> > -	seq_printf(m, " dependency chain hlocks:       %11lu [max: %lu]\n",
> > -			MAX_LOCKDEP_CHAIN_HLOCKS - nr_free_chain_hlocks,
> > +	seq_printf(m, " dependency chain hlocks used:  %11lu [max: %lu]\n",
> > +			MAX_LOCKDEP_CHAIN_HLOCKS - (nr_free_chain_hlocks - nr_lost_chain_hlocks),
> >  			MAX_LOCKDEP_CHAIN_HLOCKS);
> > +	seq_printf(m, " dependency chain hlocks free:  %11lu\n", nr_free_chain_hlocks);
> > +	seq_printf(m, " dependency chain hlocks lost:  %11lu\n", nr_lost_chain_hlocks);
> 
> I do have some comments on this. There are three buckets now - free,
> lost, used. They add up to MAX_LOCKDEP_CHAIN_HLOCKS. I don't think we
> need to list all three. We can compute the third one by subtracting max
> from the other two.
> 
> Something like:
> 
> diff --git a/kernel/locking/lockdep_proc.c b/kernel/locking/lockdep_proc.c
> index 14932ea50317..6fe6a21c58d3 100644
> --- a/kernel/locking/lockdep_proc.c
> +++ b/kernel/locking/lockdep_proc.c
> @@ -278,9 +278,12 @@ static int lockdep_stats_show(struct seq_file *m,
> void *v)
>  #ifdef CONFIG_PROVE_LOCKING
>         seq_printf(m, " dependency chains:             %11lu [max: %lu]\n",
>                         lock_chain_count(), MAX_LOCKDEP_CHAINS);
> -       seq_printf(m, " dependency chain hlocks:       %11lu [max: %lu]\n",
> -                       MAX_LOCKDEP_CHAIN_HLOCKS - nr_free_chain_hlocks,
> +       seq_printf(m, " dependency chain hlocks used:  %11lu [max: %lu]\n",
> +                       MAX_LOCKDEP_CHAIN_HLOCKS -
> +                       (nr_free_chain_hlocks + nr_lost_chain_hlocks),
>                         MAX_LOCKDEP_CHAIN_HLOCKS);
> +       seq_printf(m, " dependency chain hlocks lost:  %11lu\n",
> +                       nr_lost_chain_hlocks);
>  #endif
>  

Sure, also I tihnk the compiler is unhappy about %lu vs 'unsigned int'
for some of them.