Re: [PATCH v6 11/11] arm64: annotate user pointers casts detected by sparse

[Luc Van Oostenryck]

On Thu, Sep 06, 2018 at 02:16:19PM -0700, Linus Torvalds wrote:
> On Thu, Sep 6, 2018 at 2:13 PM Linus Torvalds
> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > So for example:
> >
> > >  static inline compat_uptr_t ptr_to_compat(void __user *uptr)
> > >  {
> > > -       return (u32)(unsigned long)uptr;
> > > +       return (u32)(__force unsigned long)uptr;
> > >  }
> >
> > this actually looks correct.
> 
> Side note: I do think that while the above is correct, the rest of the
> patch shows that we might be better off simply not havign the warning
> for address space changes at all for the "cast a pointer to an integer
> type" case.
> 
> When you cast to a non-pointer type, the address space issue simply
> doesn't exist at all, so the warning makes less sense.
> 
> It's really just he "pointer to one address space" being cast to
> "pointer to another address space" that should really warn, and that
> might need that "__force" thing.
> 
> Hmm? So maybe a sparse change is better for most of that patch.

Unless I'm misunderstanding something, I don't think there is
anything to change for this specific point. Sparse don't warn
(by default) on "cast from pointer with address space to integer",
as it always been the case, I think. I think it's the good choice.

It's just that recently, I've added a new flag -Wcast-from-as [1],
defaulting to 'no', specifically to *detect* these cast because of
these tagged pointers.

Note: I tend to think more and more that __force is simply too
      strong and weaker form, like __force_as and __force_bitwise
      would be more appropriate.


-- Luc Van Oostenryck

[1] d96da358c ("stricter warning for explicit cast to ulong")