On 05/10/2018 03:03 PM, Chris Chiu wrote:

Report from guys who can access scope. If i2c-sda-falling-time-ns=400ns
, HCNT increase to 117, the SCL high duration is 576ns as follows
https://pasteboard.co/HkwERvP.png

The original SCL high duration (HCNT = 105, 120MHz) is as follows
https://pasteboard.co/HkwFxgY.png

So the HCNT does affect but per this HCNT/LCNT value, just not 400kHz
as expected. Any suggestion?

Thanks for measurements. I was sidetracked last week so I don't have yet 
explanation why signals run faster than expected :-(

Using 120 MHz SPT I2C clocks in commit b418bbff36dd ("mfd: intel-lpss: 
Add Intel Cannonlake PCI IDs") is clearly wrong but before going to 133 
MHz (which work for you but still runs too fast) I would like to find 
explanation why it appears to be much higher.

-- 
Jarkko