One thing that comes to mind is that the mapping of VID to voltage is not a simple linear function. For some of the standards it does funky things with some of the bits. So you won't be able to use the simple math operators like the other scaling functions in sensors.conf. You'd have to create a new, custom transform function that takes both the VID value and the VRM version. sensors.conf has a relatively simple equation parser. I'm not sure how easy it will be to add the VID to voltage function. If we suck that into libsensors, then you *have* to use the library to convert the VID bit pattern into a voltage or re-implement the function in the userspace language of your choice (bash, perl, python, etc.) and since these VRM standards change seemingly with every CPU rev, all those user-space applications are going to be out of date at one point or another. But the code that exists in the kernel sensor drivers today, checks the CPU, verifies which VRM standard applies to that rev of CPU and then produces the correct VID to voltage mapping automatically. If you still want to proceed, I suggest you solve the sensors.conf side of the problem first. Make sure you can scale a raw VID bit pattern into a voltage for use in setting CPU core voltage limits. Then let's discuss changing the chip drivers. :v) Grant Coady wrote: >Hi there, > >vid and vrm are two of the strangest things I met in lmsensors. > >Here we have user space telling kernel how to scale a value the >kernel does nothing with. If I produce a set of patches to remove >vid scaling from the drivers: adm1025.c, adm1026.c, asb100.c, >it87.c, lm85.c, lm87.c, pc87360.c, w83627hf.c and w83781d.c we >save nine write accessors and can remove the vrm stuff to user- >space where it belongs, with the other scaling, simplifying the >read accessors by removing the scaling requirement. > >vid sense pins don't do anything, thus the vid sense state should >be reported truthfully to user-space, it may do something with the >value. > >Please somebody tell me what I've missed here? > >Cheers, >Grant. > > >
