Hi,

> And here's another device with programmable clock stretching[1]. Lots
> of discussions in searching about masters not supporting clock
> stretching more than I found of slaves which are configurable. At
> least in that case, we should be able to derive it from master
> compatible strings. But for the cases where both ends can support
> stretching or not, seems like we need an i2c property.

I haven't read all this, but regarding clock stretching, this is the
current state:

* clock streching is defined in the I2C specs and thus masters are
  assumed to handle it properly
* if they can't, this is a quirk and we have a flag for it:
  I2C_AQ_NO_CLK_STRETCH
* client drivers can check for quirks and act accordingly:
  i2c_check_quirks(struct i2c_adapter *adap, u64 quirks)

Note: there is currently no user of this feature because mainlining the
client got stuck for some reason. So, I'd be happy if all this is useful
to you.

Regards,

   Wolfram