On Wed, Aug 18, 2021 at 4:09 AM Linus Walleij <linus.walleij@xxxxxxxxxx> wrote: > > On Tue, Aug 17, 2021 at 9:51 AM zhiyong.tao <zhiyong.tao@xxxxxxxxxxxx> wrote: > > > In one chip, If GPIO is different, the MTXXXX_PULL_UP_RSEL_001 may > > means different actual bias resistance setting. > > > > For example, > > > > KPROW IO > > Paramters Descriptions Min Typ Max > > UNIT > > Rpd Input pull-down resistance 40 75 190 Kohm > > Rpu Input pull-up resistance 40 75 190 Kohm > > Rpd Input pull-down resistance 0.8 1.6 2 Kohm > > Rpu Input pull-up resistance 0.8 1.6 2 Kohm > > This is exactly why we should try to use SI units in the device tree. > I assume that the software can eventually configure which resistance > it gets? > > The electronics people will say make sure it is pulled down by around > 80 kOhm, they can put that on the device tree and your code can > say, "hm 40 < 80 < 190 this is OK" and let the value pass. > > We do not define these exact semantics, it is up to the driver code > to decide what to do with the Ohm value 80000 in this case, but > it makes perfect sent for me to let it pass and fail if someone > for example requests 20 kOhm, or at least print a helpful warning: > > dev_warn(dev, "the requested resistance %d is out of range, supported > range %d to %d kOhm\n", > val, low, high); > > This is what makes the SI units really helpful for people writing device > trees: solve real integration tasks and make it easy to do the right thing. I think this makes a lot of sense. The driver could select the closest setting. And from what Zhiyong mentioned offline, the resistor values aren't exact as specified in the datasheet. I suppose this is expected with any electronics. So the hardware integration will say to pull up or down by some value, and the driver will do its best to fulfill that request. That precludes DT schema checking for the values used, but I think that is a good compromise. Zhiyong also mentioned that some of their downstream integrators might not be able to deal with actual values, and would prefer symbols tied to specific RSEL values. I think that would be doable together using some _magic_ values, but I would prefer not to if it were avoidable. Regards ChenYu
