Re: [EXT] Re: [PATCH] i2c: pca954x: Add property to skip disabling PCA954x MUX device

[Peter Rosin]

On 2019-09-30 04:43, Biwen Li wrote:
>>
>> On 2019-09-29 12:36, Biwen Li wrote:
>>> On some Layerscape boards like LS2085ARDB and LS2088ARDB, input
>>> pull-up resistors on PCA954x MUX device are missing on board, So, if
>>> MUX are disabled after powered-on, input lines will float leading to
>>> incorrect functionality.

SDA and SCL are not "inputs". They are part of a bus and are both
bidirectional signals. Speaking of input signals in an I2C context
is ambiguous.

>>
>> Hi!
>>
>> Are you saying that the parent bus of the mux is relying on some pull-ups inside
>> the mux?
> Yes, as follows:
> 													VCC
> 													-------
> 													  |-------------------
>                                                       |            |
> 													  \            \
> 													  /10K resister  / 10k resister
> 													  \            \
> 													  |            |
>                                                       |            |
>    I2C1_SCL  ------------------------                         I2C1_SCL       |       ----------------------
> --------------------|SCL           |                      ----------------------------------------|SCL          |
>    I2C1_SDA  |   PCA9547   |                        I2C1_SDA       |       |    PCA9547  |          
> --------------------|SDA           |                      ----------------------------|-----------|SDA         |
>              ------------------------                                                ----------------------
>   --wrong design(need software fix as above or hardware fix)--              --proper design--

Ok, got it (but the "picture" didn't help).

>>
>>> Hence, PCA954x MUX device should never be turned-off after power-on.
>>>
>>> Add property to skip disabling PCA954x MUX device if device tree
>>> contains "i2c-mux-never-disable"
>>> for PCA954x device node.
>>>
>>> Errata ID: E-00013 on board LS2085ARDB and LS2088ARDB (Board revision
>>> found on Rev.B, Rev.C and Rev.D)
>>
>> I think you should follow the example of the i2c-mux-gpio driver and implement
>> the idle-state property instead.
>>
>> That is a lot more consistent, assuming it solves the problem at hand?
> Got it, thanks, I will try it.

I'll wait for that patch then, instead of spending more energy on the
never-disable approach.

Cheers,
Peter