[PATCH treewide v3 2/4] bitfield: Add non-constant field_{prep,get}() helpers

[Geert Uytterhoeven]

Hi Jakub,

On Fri, 17 Oct 2025 at 17:19, Jakub Kicinski <kuba at kernel.org> wrote:
> On Fri, 14 Feb 2025 14:55:51 +0100 Geert Uytterhoeven wrote:
> > The existing FIELD_{GET,PREP}() macros are limited to compile-time
> > constants.  However, it is very common to prepare or extract bitfield
> > elements where the bitfield mask is not a compile-time constant.
> >
> > To avoid this limitation, the AT91 clock driver and several other
> > drivers already have their own non-const field_{prep,get}() macros.
> > Make them available for general use by consolidating them in
> > <linux/bitfield.h>, and improve them slightly:
> >   1. Avoid evaluating macro parameters more than once,
> >   2. Replace "ffs() - 1" by "__ffs()",
> >   3. Support 64-bit use on 32-bit architectures.
> >
> > This is deliberately not merged into the existing FIELD_{GET,PREP}()
> > macros, as people expressed the desire to keep stricter variants for
> > increased safety, or for performance critical paths.
>
> We already have helpers for this, please just don't know they exist :/
>
> The "const" version of the helpers are specifically defined to work
> on masks generated with BIT() and GENMASK(). If the mask is not
> constant we should expect it to have a well defined width.
>
> I strongly prefer that we do this instead and convert the users to
> the fixed-width version:
>
> ---->8----------------
>
> Subject: bitfield: open code the fixed-width non-const helpers so that people see them
>
> There is a number of useful helpers defined in bitfield.h but
> they are mostly invisible to the reader because they are all
> generated by macros. Open code the 32b versions (which are
> most commonly used) to give developers a chance to discover them.
>
> Signed-off-by: Jakub Kicinski <kuba at kernel.org>

Thanks, but this is more or less the same code which you suggested
before [1], and to which I just replied[2] after looking at the
generated assembler output on various architectures.

> @@ -188,6 +193,81 @@ static __always_inline u64 field_mask(u64 field)
>         return field / field_multiplier(field);
>  }
>  #define field_max(field)       ((typeof(field))field_mask(field))
> +
> +/**
> + * u32_encode_bits() - prepare a u32 bitfield element (non-const)
> + * @v: value to put in the field
> + * @field: shifted mask defining the field's length and position
> + *
> + * Equivalent of FIELD_PREP() for u32, field does not have to be constant.
> + *
> + * Note that the helper is available for other field widths (generated below).
> + */
> +static __always_inline __u32 u32_encode_bits(u32 v, u32 field)
> +{
> +       if (__builtin_constant_p(v) && (v & ~field_mask(field)))
> +               __field_overflow();
> +       return ((v & field_mask(field)) * field_multiplier(field));

Unfortunately gcc emits actual divisions or __*div*() calls, and
multiplications in the non-constant case.

So I don't think this is suitable as-is.

> +}

[1] https://lore.kernel.org/all/20250214073402.0129e259@kernel.org
[2] https://lore.kernel.org/all/CAMuHMdU+0HGG22FbO3wNmXtbUm9RhTopYrGghF6UrkFu-iww2A@mail.gmail.com

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds