Re: [PATCH 2/5] slab: Add __alloc_size attributes for better bounds checking

[Joe Perches <joe@perches.com>]

On Thu, 2021-08-19 at 01:27 +0100, Matthew Wilcox wrote:
> On Tue, Aug 17, 2021 at 10:31:32PM -0700, Joe Perches wrote:
> > Lastly __alloc_size should probably be added to checkpatch
> > 
> > Maybe:
> > ---
> >  scripts/checkpatch.pl | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
[]
> > @@ -489,7 +489,8 @@ our $Attribute	= qr{
> >  			____cacheline_aligned|
> >  			____cacheline_aligned_in_smp|
> >  			____cacheline_internodealigned_in_smp|
> > -			__weak
> > +			__weak|
> > +			__alloc_size\s*\(\s*\d+\s*(?:,\s*d+\s*){0,5}\)
> 
> Should probably be added to kernel-doc as well.  Any other awful regexes
> that need to be changed to understand it?  And can we commonise the
> regexes that do exist into a perl helper library?

probably, but there would need to be some library work done and
changes made to both utilities so they could use the same $helpers.

And there are several nominally incomplete regexes already in
kernel-doc and I'm not at all familiar with kernel-doc.

e.g.: kernel-doc has:

my $attribute = qr{__attribute__\s*\(\([a-z0-9,_\*\s\(\)]*\)\)}i;

but __attribute__ can have quotes like:

__attribute__((section("foo")))

and spaces around and and I believe between (( and )) like:

__attribute__ ((packed))

so those wouldn't match.

The use of parentheses internal to attributes like __align__(8) may
not work particularly well either given greedy matching.