Re: [RFC PATCH v2 0/2] regmap: add APIs to bypass regmap_field and work with reg_field directly
From: Andy Shevchenko
Date: Fri Nov 13 2026 - 11:20:11 EST
On Fri, Nov 13, 2026 at 9:01 AM Victor Chernyakin
<chernyakin.victor.j@xxxxxxxxxxx> wrote:
>
> Working with the existing regmap_field_* APIs requires you to define
> a reg_field, then use one of the regmap_field_*_alloc functions to turn
> it into a dynamically allocated regmap_field. But reg_field and
> regmap_field are practically the same struct; all the *_alloc functions do
> is shuffle the data around. So this RFC proposes adding functions that
> cut out the middleman and take reg_field directly. Patch 1 adds two
> of them: regmap_reg_field_read and regmap_reg_field_write.
>
> Patch 2 ports a driver to use these new functions. The result is
> shorter and simpler, since all the code to perform the allocations
> and manage the resulting regmap_fields goes away. In theory,
> there should also be a performance benefit (we're doing fewer allocations
> and have less duplicated state), but I haven't benchmarked it, so take that
> with a grain of salt.
>
> If we like this approach, I can post a v3 that adds regmap_reg_field_*
> analogs for all the regmap_field_* functions, not just the two you see
> here.
>
>
> This is not actually a new idea! Both it and the current *_alloc approach
> were suggested as alternatives right when the regmap_field_* APIs were
> being added [1]. The *_alloc approach was chosen without much
> discussion, but I think (and Patch 2 demonstrates, I hope) that
> this approach is the cleaner one.
It may be good to Cc your attempts to the authors (and active
reviewers) of the mentioned change.
> [1] https://lore.kernel.org/20130604210137.GV31367@xxxxxxxxxxxxx
--
With Best Regards,
Andy Shevchenko