On 7/11/22 17:22, Andy Shevchenko wrote:
> On Mon, Jul 11, 2022 at 4:04 PM Dmitry Osipenko
> <dmitry.osipenko@xxxxxxxxxxxxx> wrote:
>> On 7/11/22 16:41, Andy Shevchenko wrote:
>>> On Mon, Jul 11, 2022 at 3:39 PM Shreeya Patel
>>> <shreeya.patel@xxxxxxxxxxxxx> wrote:
>>>> On 11/07/22 18:36, Andy Shevchenko wrote:
>>>>> On Mon, Jul 11, 2022 at 1:30 PM Shreeya Patel
>
> ...
>
>>>>>> +static const struct regmap_config ltrf216a_regmap_config = {
>>>>>> + .name = LTRF216A_DRV_NAME,
>>>>>> + .reg_bits = 8,
>>>>>> + .val_bits = 8,
>>>>>> + .max_register = LTRF216A_MAX_REG,
>>>>> Why do you use regmap locking? What for?
>>>>
>>>> Why do we want to skip the internal locking if it doesn't bring any
>>>> benefits?
>>>
>>> Can you elaborate on the "no benefits" part, please?
>>
>> Since the regmap's lock will never be contended, thus it's free to keep
>
> I'm skeptical about "free" here. My concerns are:
> 1) grosser code base;
> 2) slower code flow (even nop takes time to process).
>
>> using it. If later on we will need to change the driver's code such that
>> the lock will become needed, then we won't need to bother with
>> re-enabling it. The comment to the driver's mutex states clearly that
>> it's intended to protect the cached value.
>>
>> Hence what is point in disabling the regmap's lock? There are very few
>> drivers that disable the regmap's lock and most of them do that for the
>> good reason.
>
> Most of the drivers that have its own lock _and_ regmap lock took the
> locking scheme wrong. It is 101 when writing a driver to have a clear
> picture of what lock protects what data (or I/O).
>
> Even if lock is _almost_ free, it's still required to provide
> understanding of how each of the locks is being used in the code.
>
> That said, the main point of my review comment is to make the author
> think about it, or just elaborate if it has been thought through
> already.
Alright, thank you for clarifying yours point! I helped Shreeya a tad
with preparing the v7 and disabling the regmap's lock wasn't under
question. Since this driver isn't about performance and a nanosecond
improvement isn't worthwhile in comparison to the the I2C transfer time,
should be cleaner to keep the regmap's lock in place, IMO. I'll let
Shreeya to answer next time, sorry for jumping in.
--
Best regards,
Dmitry
