Thanks for the great research and suggestions. We'll have to think of the best way to incorporate them. If we brought vrm out to /proc we could have a simple program that does the following: # fixvrm Which VID is correct: 1) 3.2V (VRM 8.2) (default) 2) 2.55V (VRM 8.5) 3) 1.675V (VRM 9.0) Enter number 1-3: 1 Your VID calculation has been changed to 1.675 V (VRM 9.0) Trent Piepho wrote: > > I've been getting lm_sensors working with my tyan tiger mp (more on that > later) and I noticed that the VID was reporting the wrong core voltage. I > tracked down the problem and figured out what was going on. > > The w83781d driver, and lots of others (lm87 I know too, because I fixed a vid > problem with that driver) use a single macro VID_FROM_REG to convert the 4 or > 5 vid lines into a core voltage, with 10 mV units. The macro follows a spec > called the VRM 8.2 DC-DC Converter Specification, you can get a pdf at intel's > site. The AthlonXP processor, and I assume all socket A chips, uses a newer > specification called VRM 9.0. There is also a VRM 8.1, 8.3, and 8.4, but they > are all mostly compatible with 8.2. VRM 9.0 is *not* backward compatible with > VRM 8.2. There is also a VRM 8.5, which is used with Pentium III Tualatin > processors, that isn't compatible with VRM 8.2. > > In order to get proper VID support, the user needs to be able to specify the > processor's VRM version. Since the code to do this conversion is the same for > any sensor chip, it makes sense not to put a separate copy in each driver. It > would be nice if it could be specified in the sensors.conf file, e.g. "set vrm > 9.0". Most drivers look at the vid setting to set the initial voltage limits > for the cpu voltage line. In order to do this correctly, the driver would > need to know the setting when it initializes the chip. > > There is also a problem with the units. The newer VRM specifications use > increments of 25 mV instead of 50 or 100 mV, so the 10mV resolution used in > the driver isn't enough. ie. magnitude for vid needs to be 3 instead of 2. > > Here is code to handle the different VRM versions. I'm not sure if the VRM > 8.5 code will work, as it's not totally clear how the VID lines will be hooked > up. > > /* VRM 8.2 - 8.4: Used for Pentium II-III chips */ > #define VID_FROM_REG(val) ((val)==0x1f?0:(val)>=0x10?510-(val)*10:\ > 205-(val)*5) > > /* VRM 8.5: Used for Pentium III-S Tualatin chips */ > #define VID_FROM_REG(val) ((val)==0x1f?0: ((((val)&16)?2.5:0) + \ > ((val)>4?205:125) - ((val)&15)*5)) > > /* VRM 9.0: Used for Pentium IV and AMD Socket A */ > #define VID_FROM_REG(val) ((val)==0x1f?0:(185-(val)*2.5))
