Re: [PATCH 2/3] regulator: s2mps11: Merge S2MPA01 driver

[Yadwinder Singh Brar]

On Tue, May 27, 2014 at 1:26 PM, Krzysztof Kozlowski
<k.kozlowski@xxxxxxxxxxx> wrote:
> On wto, 2014-05-27 at 12:00 +0530, Yadwinder Singh Brar wrote:
>> On Mon, May 26, 2014 at 6:50 PM, Krzysztof Kozlowski
>> <k.kozlowski@xxxxxxxxxxx> wrote:
>> > Add S2MPA01 support to the s2mps11 regulator driver. This obsoletes the
>> > s2mpa01 regulator driver.
>> >
>> > Signed-off-by: Krzysztof Kozlowski <k.kozlowski@xxxxxxxxxxx>
>>
>> > @@ -216,30 +250,20 @@ static int s2mps11_set_ramp_delay(struct regulator_dev *rdev, int ramp_delay)
>> >                         ramp_delay = s2mps11->ramp_delay16;
>> >                 break;
>> >         case S2MPX_BUCK2:
>> > -               if (!ramp_delay) {
>> > -                       ramp_enable = 0;
>> > -                       break;
>> > -               }
>> > -
>>
>> What if we want to disable ramp_delay from DT ?
>
> It will work OK because at the beginning of s2mps11_set_ramp_delay():
>         unsigned int ramp_disable = !ramp_delay;
> This 'ramp_disable' is later used if enable/disable is supported.
>>

Oh! I missed you defined a new variable "ramp_disable",
 since ramp_disable is already a label defined in same function.
It should be different, i think.

>> > -               s2mps11->ramp_delay2 = ramp_delay;
>> > +               if (s2mps11->dev_type == S2MPS11X ||
>> > +                               ramp_delay > s2mps11->ramp_delay2)
>> > +                       s2mps11->ramp_delay2 = ramp_delay;
>> > +               else /* S2MPA01 && ramp_delay <= s2mpa01->ramp_delay24 */
>> > +                       ramp_delay = s2mps11->ramp_delay2;
>>
>> Here ramp_delay = 0(ramp_disable case) is also getting over written,
>> if required to take care of it later.
>
> The same, it is already stored as 'ramp_disable' local variable.
>
>>
>> >                 break;
>> >         case S2MPX_BUCK3:
>> > -               if (!ramp_delay) {
>> > -                       ramp_enable = 0;
>> > -                       break;
>> > -               }
>>
>> [snip]
>>
>> >
>> > -       if (!ramp_enable)
>> > -               goto ramp_disable;
>> > -
>> > -       /* Ramp delay can be enabled/disabled only for buck[2346] */
>> >         if (ramp_reg->enable_supported) {
>> > +               if (ramp_disable)
>>
>> typo ?    if (!ramp_enable) / if (!ramp_delay) ?
>
> I think it is good. I changed the 'ramp_enable' into 'ramp_disable'.
>

ok, but very next statement is
             goto ramp_disable;

which seems odd and obfuscated me.

> Anyway while reviewing the code I found that I didn't updated the case
> statements with new BUCKX enum values and the register for
> enable/disable is hard-coded. I'll fix it.
>>
>> > +                       goto ramp_disable;
>> > +
>>
>>
>> Also TBH, I can't get rationale behind this merge, As i can't see
>> considerable reduction in no of C code lines in comp of added
>> complexity.
>>  Is there considerable advantage in binary stats of single driver as
>> compare to independent drivers?
>
> Overall more code is removed than added:
> 6 files changed, 454 insertions(+), 719 deletions(-)
> but you are right that the code for ramp delay is now more complex. What
> is worth noting now most of ramp delay settings are moved to an array:
>
> static const struct s2mpx_ramp_reg s2mps11_ramp_regs[] = {
>         [S2MPX_BUCK1]   = s2mps11_ramp_reg(BUCK16),
>         [S2MPX_BUCK2]   = s2mps11_buck2346_ramp_reg(BUCK2, RAMP, BUCK2),
>         [S2MPX_BUCK3]   = s2mps11_buck2346_ramp_reg(BUCK34, RAMP, BUCK3)
>
> instead of being hard-coded into the big switch statement like it was
> before.
>
> Alternative solution to complex ramp delay setting is to just use
> original functions: s2mps11_set_ramp_delay and s2mpa01_set_ramp_delay.
>
> These chips are really similar so having two drivers seems like doubling
> the effort for maintaining them.
>

I think maintaining a complex or a big file(in case we keep original
functions), itself will be an effort consuming thing and moreover
binary size of a single driver will also increase considerable as
compare to independent drivers (if its not case of multiplatform
kernel).

Anyways, i think its matter of preference of all, It will be OK, if
for others( especially maintainers, Mark ?), its OK.


Best Regards,
Yadwinder

> Thanks for comments.
>
> Best regards,
> Krzysztof
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/