On Mon, Oct 23, 2017 at 09:54:26AM +0200, Romain Porte wrote: > On 21/10/2017 18:20, Guenter Roeck wrote: > >This should be part of device programming in manufacturing, just like, > >say, current calibration. It should not be user programmable, even less so > >runtime programmable. On top of that, we definitely don't want to make > >STORE_USER_ALL available to user space. Both can too easily result in > >a bricked device (bad enough that the register values are writable using > >i2cset). > My point is that I am exactly using Linux's pmbus implementation in order to > perform a manufacturing calibration which is indeed a one-time operation. > > I agree that exposing these sysfs entries to end users can be dangerous. It > can be useful if you actually want to perform a component calibration using > Linux. Actually I am running an userspace application that does the > calibration by writing to these sysfs entries. This driver is useful for > manufacturing calibration and I think can be useful for other Linux users > who wants to perform this kind of operation too. > > If this driver is dangerous for the end-user, how can we keep the features > of this driver for manufacturing calibration using Linux? Maybe keep the > generic driver for normal use and propose this specific driver as 'Advanced > TPS544C20 driver [DANGER]'? >
You can use i2cset for your purpose and still use Linux. There is no need to have kernel support for it. With that, you can program all registers, not just a single one. Guenter