Re: [PATCH v2] sched: Make const-safe

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Dec 14, 2022 at 06:03:04PM +0000, Matthew Wilcox wrote:
> On Wed, Dec 14, 2022 at 10:11:01AM +0100, Peter Zijlstra wrote:
> > On Mon, Dec 12, 2022 at 02:49:46PM +0000, Matthew Wilcox (Oracle) wrote:
> > > With a modified container_of() that preserves constness, the compiler
> > > finds some pointers which should have been marked as const.  task_of()
> > > also needs to become const-preserving for the !FAIR_GROUP_SCHED case so
> > > that cfs_rq_of() can take a const argument.  No change to generated code.
> > 
> > More const more better I suppose.. Thanks!
> > 
> > Happen to have a sha for the container_of() commit handy?
> 
> There isn't one yet.  Obviously we can't make container_of()
> const-preserving until we've fixed all the places which would warn.
> The diff I have in my tree looks like this:
> 
> diff --git a/include/linux/container_of.h b/include/linux/container_of.h
> index 1d898f9158b4..9416e6cc8c88 100644
> --- a/include/linux/container_of.h
> +++ b/include/linux/container_of.h
> @@ -20,7 +20,10 @@
>         static_assert(__same_type(*(ptr), ((type *)0)->member) ||       \
>                       __same_type(*(ptr), void),                        \
>                       "pointer type mismatch in container_of()");       \
> -       ((type *)(__mptr - offsetof(type, member))); })
> +       __mptr -= offsetof(type, member);                               \
> +       _Generic(ptr,                                                   \
> +               const typeof(*(ptr)) *: (const type *)__mptr,           \
> +               default: ((type *)__mptr)); })
> 
>  /**
>   * container_of_const - cast a member of a structure out to the containing
> 
> I have all of fs/ and net/ compiling cleanly now.  There are a few
> places which really need the const-removing properties, and I've made
> those call a new macro called container_of_not_const(), but I don't
> like that name.

#define const_cast(T, exp)	_Generic((exp), const T : (T)(exp), default: (exp))

perhaps? Then one can write something like:

	struct task_struct *p = const_cast(struct task_struct *,
					   constainer_of(node, struct task_struct *, run_node));

The repetition is a bit naf, but at least the construct is more
generally useful.

(I really wish there was a qualifier stripping typeof() variant -- and
yes, I know about the _Atomic thing).