[PATCH v2 2/4] regmap: Allow providing read/write callbacks through struct regmap_config

[Jean-Jacques Hiblot]

Simon,

On 24/12/2019 16:58, Simon Glass wrote:
> Hi Jean-Jacques,
>
> On Mon, 16 Dec 2019 at 03:10, Jean-Jacques Hiblot <jjhiblot at ti.com> wrote:
>> Hi Simon,
>>
>> On 10/12/2019 16:18, Simon Glass wrote:
>>> Hi Jean-Jacques,
>>>
>>> On Tue, 5 Nov 2019 at 04:47, Jean-Jacques Hiblot <jjhiblot at ti.com> wrote:
>>>> Some linux drivers provide their own read/write functions to access data
>>>> from/of the regmap. Adding support for it.
>>>>
>>>> Signed-off-by: Jean-Jacques Hiblot <jjhiblot at ti.com>
>>>>
>>>> ---
>>>>
>>>> Changes in v2:
>>>> - Only use custom accessors if {,SPL,TPL}_REGMAP_ACCESSORS is enabled
>>>>
>>>>    drivers/core/Kconfig  | 25 +++++++++++++++++++++++++
>>>>    drivers/core/regmap.c | 22 ++++++++++++++++++++--
>>>>    include/regmap.h      | 28 +++++++++++++++++++++++++---
>>>>    3 files changed, 70 insertions(+), 5 deletions(-)
>>> Coming back to the discussion on driver model....
>>>
>>> How do you specify the fields? I would expect that this would be done
>>> in the driver tree? Perhaps in a subnode of the device?
>>>
>>> Just to state what I see as the advantages of using a separate device
>>> for access:
>>>
>>> - Remove the #ifdef in the regmap struct
>>> - Easy to specify the behaviour in a device-tree node
>>> - Easy to extend as the child device can do what it likes with respect to access
>> That sure is a better abstraction. However the goal of this patch is
>> only to use the same API as linux. It allows porting the drivers as-is
>> and thus reduce the burden of maintenance.
> So how do you specify the fields? See my question above.

The fields are filled when creating the regmap.

ex:

static struct regmap_config cfg = {
	.reg_write = regmaptest_write,
	.reg_read = regmaptest_read,
};
and then
regmap = devm_regmap_init(dev, NULL, &ctx, &cfg);

You can have a look at the tests in the last patch of the series.

JJ

> It is not possible to use a similar API without importing the internal
> implementation. Linux's driver model is less homogenous.
>
> Regards,
> Simon