Re: [PATCH v2 2/5] gpio: pca953x: add register definitions for pcal6524 and fix address calculation

From: H. Nikolaus Schaller
Date: Tue Apr 10 2018 - 11:51:33 EST


Hi,

> Am 10.04.2018 um 16:41 schrieb Andy Shevchenko <andy.shevchenko@xxxxxxxxx>:
>
> On Wed, Apr 4, 2018 at 10:00 PM, H. Nikolaus Schaller <hns@xxxxxxxxxxxxx> wrote:
>> PCAL chips ("L" seems to stand for "latched") have additional
>> registers starting at address 0x40 to control the latches,
>> interrupt mask, pull-up and pull down etc.
>>
>> The constants are so far defined in a way that they fit for
>> the pcal9555a when shifted by the number of banks, i.e. multiplied
>> by 2.
>>
>> Now the pcal6524 has 3 banks which means the relative offset
>> must be multiplied by 4 which gives a wrong result if not done
>> carefully, since the base offset is already included in the offset.
>>
>> For the basic registers shared with all pca93xx/tca64xx chips
>> there is no such offset.
>>
>> Therefore, we add code to adjust the register number for exended
>> registers in this case.
>>
>> And we add additional register offset constants (not yet used by
>> the driver code) which are specific to the pcal6524.
>
>> -#define PCAL953X_IN_LATCH 34
>> -#define PCAL953X_INT_MASK 37
>> -#define PCAL953X_INT_STAT 38
>
> I prefer either to change first all the rest to be 2 digit hex values
> first, or leave decimal in this patch.

Ok, let's change them all to hex (like in the data sheets).

>
>> +#define PCAL953X_OUT_STRENGTH 0x20
>> +#define PCAL953X_IN_LATCH 0x22
>> +#define PCAL953X_PULL_EN 0x23
>> +#define PCAL953X_PULL_SEL 0x24
>> +#define PCAL953X_INT_MASK 0x25
>> +#define PCAL953X_INT_STAT 0x26
>> +#define PCAL953X_OUT_CONF 0x27
>> +
>> +#define PCAL6524_INT_EDGE 0x28
>> +#define PCAL6524_INT_CLR 0x2a
>> +#define PCAL6524_IN_STATUS 0x2b
>> +#define PCAL6524_OUT_INDCONF 0x2c
>> +#define PCAL6524_DEBOUNCE 0x2d
>
>> + /* adjust register address for pcal6524 */
>> + if (reg >= PCAL953X_OUT_STRENGTH)
>> + reg -= PCAL953X_OUT_STRENGTH >> 1;
>
> I don't like this condition. Can we avoid relying on the register
> offset in regard to some flag / compatible string / etc?

I don't think so.

for TCA6416 and PCAL9555 the real registers are e.g.:
PCA953X_INPUT 0x00 / 0x01
PCA953X_OUTPUT 0x02 / 0x03

i.e. every address constant is shifted left 1 bit in the 16 bit
accessors.

for PCAL9555 some extended registers:

PCAL953X_OUT_STRENGTH 0x40 / 0x41 / 0x42 / 0x43
PCAL953X_IN_LATCH 0x44 / 0x45

i.e. PCAL953X_IN_LATCH is also simply shifted left by 1 bit
because the extended constants are defined that they fit.

for TCA6424 and PCAL6524:
PCA953X_INPUT 0x00 / 0x01 / 0x02
PCA953X_OUTPUT 0x04 / 0x05 / 0x06

every address is shifted left 2 bits in the 24 bit accessors.

for PCAL6524:
PCAL953X_OUT_STRENGTH 0x40 / 0x41 / 0x42 / 0x43 / 0x44 / 0x45
PCAL953X_IN_LATCH 0x48 / 0x49 / 0x4a

simply shifting left by 2 bits gives wrong result which is this
special case. It is not a general chip property but a mix of chip
bit-width and depends on accessing the first or second bank.

So the full condition to apply this address fix is:

if (24 bit access && second bank access)
do correction.

and it can be simplified to what I suggest, because the 8/16/24 bit
accessors are already separate functions.

I don't think this can't be hard-coded into flags / compatible string.

>
>> + /* adjust register address for pcal6524 */
>> + if (reg >= PCAL953X_OUT_STRENGTH)
>> + reg -= PCAL953X_OUT_STRENGTH >> 1;
>
> Ditto.
>
> --
> With Best Regards,
> Andy Shevchenko

BR and thanks,
Nikolaus