Re: [PATCH 1/2] mm/slub: Fix cmp_loc_by_count() to return 0 when counts are equal

[Kuan-Wei Chiu <visitorckw@gmail.com>]

Hi Joshua,

On Mon, Aug 25, 2025 at 07:48:36AM -0700, Joshua Hahn wrote:
> On Mon, 25 Aug 2025 09:34:18 +0800 Kuan-Wei Chiu <visitorckw@gmail.com> wrote:
> 
> > The comparison function cmp_loc_by_count() used for sorting stack trace
> > locations in debugfs currently returns -1 if a->count > b->count and 1
> > otherwise. This breaks the antisymmetry property required by sort(),
> > because when two counts are equal, both cmp(a, b) and cmp(b, a) return
> > 1.
> > 
> > This can lead to undefined or incorrect ordering results. Fix it by
> > explicitly returning 0 when the counts are equal, ensuring that the
> > comparison function follows the expected mathematical properties.
> > 
> > Fixes: 553c0369b3e1 ("mm/slub: sort debugfs output by frequency of stack traces")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
> > ---
> >  mm/slub.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/mm/slub.c b/mm/slub.c
> > index 30003763d224..c91b3744adbc 100644
> > --- a/mm/slub.c
> > +++ b/mm/slub.c
> > @@ -7718,8 +7718,9 @@ static int cmp_loc_by_count(const void *a, const void *b, const void *data)
> >  
> >  	if (loc1->count > loc2->count)
> >  		return -1;
> > -	else
> > +	if (loc1->count < loc2->count)
> >  		return 1;
> > +	return 0;
> >  }
> 
> Hello Kuan-Wei,
> 
> This is a great catch! I was thinking that in addition to separating out the
> == case, we can also simplify the behavior by just opting to use the
> cmp_int macro, which is defined in the <linux/sort.h> header, which is
> already included in mm/slub.c. For the description, we have:
> 
>  * Return: 1 if the left argument is greater than the right one; 0 if the
>  * arguments are equal; -1 if the left argument is less than the right one.
> 
> So in this case, we can replace the entire code block above with:
> 
> return cmp_int(loc2->count, loc1->count);
> 
> or
> 
> return -1 * cmp_int(loc1->count, loc2->count);
> 
> if you prefer to keep the position of loc1 and loc2. I guess we do lose
> some interpretability of what -1 and 1 would refer to here, but I think
> a comment should be able to take care of that.
> 
> Please let me know what you think. I hope you have a great day!
> Joshua

Thanks for the suggestion!
If we're going with the cmp_int() macro, I personally prefer
return cmp_int(loc2->count, loc1->count);
this avoids the need to explain the extra * (-1), and I think cmp_int()
is simple enough to be easily understood by readers.
That said, both options work fine for me.

Regards,
Kuan-Wei