[PATCH 21/21] Docs: add Functions parameters order section

[Jani Nikula]

On Sat, 25 Oct 2025, "Yury Norov (NVIDIA)" <yury.norov at gmail.com> wrote:
> Standardize parameters ordering in some typical cases to minimize
> confusion.
>
> Signed-off-by: Yury Norov (NVIDIA) <yury.norov at gmail.com>
> ---
>  Documentation/process/coding-style.rst | 48 ++++++++++++++++++++++++++
>  1 file changed, 48 insertions(+)
>
> diff --git a/Documentation/process/coding-style.rst b/Documentation/process/coding-style.rst
> index d1a8e5465ed9..dde24148305c 100644
> --- a/Documentation/process/coding-style.rst
> +++ b/Documentation/process/coding-style.rst
> @@ -523,6 +523,54 @@ below, compared to the **declaration** example above)::
>  	...
>   }
>  
> +6.2) Function parameters order
> +------------------------------
> +
> +The order of parameters is important both for code generation and readability.
> +Passing parameters in an unusual order is a common source of bugs. Listing
> +them in standard widely adopted order helps to avoid confusion.
> +
> +Many ABIs put first function parameter and return value in R0. If your
> +function returns one of its parameters, passing it at the very beginning
> +would lead to a better code generation. For example::
> +
> +        void *memset64(uint64_t *s, uint64_t v, size_t count);
> +        void *memcpy(void *dest, const void *src, size_t count);
> +
> +If your function doesn't propagate a parameter, but has a meaning of copying
> +and/or processing data, the best practice is following the traditional order:
> +destination, source, options, flags.
> +
> +for_each()-like iterators should take an enumerator the first. For example::
> +
> +        for_each_set_bit(bit, mask, nbits);
> +                do_something(bit);
> +
> +        list_for_each_entry(pos, head, member);
> +                do_something(pos);
> +
> +If function operates on a range or ranges of data, corresponding parameters
> +may be described as ``start - end`` or ``start - size`` pairs. In both cases,
> +the parameters should follow each other. For example::
> +
> +        int
> +        check_range(unsigned long vstart, unsigned long vend,
> +                    unsigned long kstart, unsigned long kend);
> +
> +        static inline void flush_icache_range(unsigned long start, unsigned long end);
> +
> +        static inline void flush_icache_user_page(struct vm_area_struct *vma,
> +                                            struct page *page,
> +                                            unsigned long addr, int len);
> +
> +Both ``start`` and ``end`` of the interval are inclusive.
> +
> +Describing intervals in order ``end - start`` is unfavorable. One notable
> +example is the ``GENMASK(high, low)`` macro. While such a notation is popular
> +in hardware context, particularly to describe registers structure, in context
> +of software development it looks counter intuitive and confusing. Please switch
> +to an equivalent ``BITS(low, high)`` version.
> +

GENMASK when used for defining hardware registers is completely fine,
and *much* easier to deal with when you cross check against the specs
that almost invariably define high:low.

Which other parts of coding style take on specific interfaces and tell
you to switch? Weird. I for one don't want to encourage an influx of
trivial patches doing GENMASK to BITS conversions, and then keep
rejecting them. It's just a huge collective waste of time.

Anyway, that's a lot of text on "function parameter order" to justify
BITS(), but completely skips more important principles such as "context
parameter first", or "destination first".


BR,
Jani.


>  7) Centralized exiting of functions
>  -----------------------------------

-- 
Jani Nikula, Intel