Re: [PATCH] x86/percpu: Use __force to cast from __percpu address space

[Uros Bizjak]

On Fri, Apr 5, 2024 at 4:38 PM Nadav Amit <nadav.amit@xxxxxxxxx> wrote:
>
>
>
> > On 2 Apr 2024, at 20:50, Uros Bizjak <ubizjak@xxxxxxxxx> wrote:
> >
> > Fix sparse warning when casting from __percpu address space by using
> > __force in the cast. x86 named address spaces are not considered to
> > be subspaces of the generic (flat) address space, so explicit casts
> > are required to convert pointers between these address spaces and the
> > generic address space (the application should cast to uintptr_t and
> > apply the segment base offset). The cast to uintptr_t removes
> > __percpu address space tag and sparse reports:
> >
> >  warning: cast removes address space '__percpu' of expression
> >
> > Use __force to inform sparse that the cast is intentional.
> >
> > Reported-by: Charlemagne Lasse <charlemagnelasse@xxxxxxxxx>
> > Closes: https://lore.kernel.org/lkml/CAFGhKbzev7W4aHwhFPWwMZQEHenVgZUj7=aunFieVqZg3mt14A@xxxxxxxxxxxxxx/
> > Fixes: 9a462b9eafa6 ("x86/percpu: Use compiler segment prefix qualifier")
> > Signed-off-by: Uros Bizjak <ubizjak@xxxxxxxxx>
> > Cc: Nadav Amit <nadav.amit@xxxxxxxxx>
> > Cc: Luc Van Oostenryck <luc.vanoostenryck@xxxxxxxxx>
> > Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> > Cc: Andy Lutomirski <luto@xxxxxxxxxx>
> > Cc: Brian Gerst <brgerst@xxxxxxxxx>
> > Cc: Denys Vlasenko <dvlasenk@xxxxxxxxxx>
> > Cc: H. Peter Anvin <hpa@xxxxxxxxx>
> > Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
> > Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > Cc: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
> > ---
> > arch/x86/include/asm/percpu.h | 6 +++---
> > 1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/x86/include/asm/percpu.h b/arch/x86/include/asm/percpu.h
> > index 1f6404e0c428..20696df5d567 100644
> > --- a/arch/x86/include/asm/percpu.h
> > +++ b/arch/x86/include/asm/percpu.h
> > @@ -77,7 +77,7 @@
> > #define arch_raw_cpu_ptr(_ptr) \
> > ({ \
> > unsigned long tcp_ptr__ = __raw_my_cpu_offset; \
> > - tcp_ptr__ += (unsigned long)(_ptr); \
> > + tcp_ptr__ += (__force unsigned long)(_ptr); \
> > (typeof(*(_ptr)) __kernel __force *)tcp_ptr__; \
> > })
> > #else
> > @@ -96,8 +96,8 @@
> > #endif /* CONFIG_SMP */
> >
> > #define __my_cpu_type(var) typeof(var) __percpu_seg_override
> > -#define __my_cpu_ptr(ptr) (__my_cpu_type(*ptr) *)(uintptr_t)(ptr)
> > -#define __my_cpu_var(var) (*__my_cpu_ptr(&var))
> > +#define __my_cpu_ptr(ptr) (__my_cpu_type(*ptr)*)(__force uintptr_t)(ptr)
> > +#define __my_cpu_var(var) (*__my_cpu_ptr(&(var)))
>
> If you use "(var)” in __my_cpu_var(),, you might just as well change the first to:
>
>   #define __my_cpu_ptr(ptr) (__my_cpu_type(*(ptr))*)(__force uintptr_t)(ptr)
>
> [ not that I think of any real issue it might cause - just for consistency]

Yes, I have changed it in one of the follow-up percpu patches [1].

[1] https://lore.kernel.org/lkml/20240404094218.448963-2-ubizjak@xxxxxxxxx/

Thanks,
Uros.