Re: [RFC PATCH] checkpatch: check for function calls with struct or union on stack

[Julia Lawall <julia.lawall@lip6.fr>]

On Fri, 27 Jul 2018, Joe Perches wrote:

> On Fri, 2018-07-27 at 10:21 +0000, David Laight wrote:
> > From: Joe Perches Sent: 27 July 2018 11:09
> > > On Fri, 2018-07-27 at 10:04 +0000, David Laight wrote:
> > > > From: Andrew Morton Sent: 26 July 2018 20:28
> > > > > On Thu, 26 Jul 2018 12:25:33 -0700 Andrew Morton <akpm@linux-foundation.org> wrote:
> > > > >
> > > > > > I'll give it a spin, see how noisy it is.
> > > > >
> > > > > Actually, I would prefer if the message, changelog and title
> > > > > used the term "passed by value".  It's a more familiar term
> > > > > and it is possible for a passed-by-value aggregate to in fact
> > > > > be passed in registers.
> > > >
> > > > You need to detect (and ignore) 'small' structures.
> > >
> > > checkpatch is stupid and basically can't do that
> > > as it has no context other than the current line.
> > >
> > > It would need a list of specific struct types to
> > > ignore.  Care to create and send that list to me?
> >
> > Does it even have the type?
>
> Yes, kinda.  But only on the line being matched.
>
> i.e.: <const> [struct or union] [type] [name]
>
> > If it has the prototype it could ignore aggregates that
> > are marked 'const'.
>
> checkpatch has no visibility of any prototype.
>
> It might make sense for this sort of check to be
> added to coccinelle or maybe as a compiler warning
> when the struct is larger than some size.
>
> Original thread for Julia:
> https://lore.kernel.org/patchwork/patch/967890/

Coccinelle doesn't directly know the size of the structure, but it can
count the number of fields.  Maybe a case with an update in the function
body or at least 3 fields is worth reporting on?

julia