[PATCH] i2c: rk3x: adjust the LOW divison based on characteristics of SCL

[addy ke]

On 2014/9/26 10:08, Doug Anderson wrote:
> Addy,
> 
> On Thu, Sep 25, 2014 at 6:40 PM, addy ke <addy.ke at rock-chips.com> wrote:
>> Hi, Doug
>>
>> On 2014/9/26 5:52, Doug Anderson wrote:
>>> Addy,
>>>
>>> On Wed, Sep 24, 2014 at 9:36 PM, Doug Anderson <dianders at chromium.org> wrote:
>>>> Addy,
>>>>
>>>> On Wed, Sep 24, 2014 at 6:56 PM, addy ke <addy.ke at rock-chips.com> wrote:
>>>>> In my measurement,all paramter but "Data hold time" are match the characteristics of SCL bus line.
>>>>> the measured value is 0.928us("data hold time on RK3X"  ~=  "the low period / 2")
>>>>> but the maximum value described in table is 0.9us
>>>>>
>>>>> About "Data hold time", there are described in I2C specification:
>>>>> - for CBUS compatible masters for I2C-bus deivices
>>>>> - the maximum data hold time has only be met if the device does not stretch the LOW period of the SCL signal.
>>>>>
>>>>> I have tested on RK3288-Pinky board, there are no error.
>>>>> But I don't known whether this paramter will affect i2c communications.
>>>>
>>>> I'll have to spend more time tomorrow to really understand this, but
>>>> if changing the code to bias towards slightly longer "high" times
>>>> instead of "low" times helps fix it then that's fine with me.
>>>
>>> So what you're saying is that you're seeing a case where the clock
>>> goes low and the data is not valid until .928us.  Is this data that is
>>> being driven by the master or data that is being driven by the slave?
>>>
>> It is driven by the master and will be release at half of LOW period in our IC design.
> 
> Ah, I didn't know this.  Is that in the TRM somewhere?
> 
No, it was my test and confirmed by IC engineer.
You can find that there is a pulse at half of LOW period of the ninth SCL clock each bytes.

> ...so does it work to just replace:
> 
>     ideal_low_div = DIV_ROUND_UP(clk_rate * min_low_ns,
>                                  scl_rate * 8 * min_total_ns)
> 
> ...with:
> 
>     ideal_low_div = min_low_div
> 
> 
> That will assign all "extra" time to the "high" part.
> 
I Think it is more reasonable.
> -Doug
> 
> 
>