Jean Delvare wrote: >Hi Jim, > > > >>Im looking to understand why my temps are so high, >>esp in comparison to values from my soekris as obtained from >>http://phk.freebsd.dk/soekris/env4801/ >> >> VREF = 1.214 V245 = 2.450 >> Temp 2 (status=0x81) 54 C >> >>in your code, >> >> data->in_vref = (i&0x02) ? 3025 : 2966; >> dev_dbg(&new_client->dev, "Using %s reference voltage\n", >> (i&0x02) ? "external" : "internal"); >> >>youve got 2 magical constants, where are they from ? >>I cant find mention of them in the pdf, >> >> > >3.025 is 1.235 * 2.45. Page 227 of the datasheet mentions internal Vref >= 1.235V, and at page 180, section 11.3.2 has: "Analog input voltage is >measured relative to 2.45 * Vref". > >In section 14.4.1, external Vref is given at 1.211V, and full scale is >given at 2.097, which happens to be 1.211 * 2.45. > >This is where the numbers are coming from. > thanks for the (patient) orientation. Theres a lot to miss/get in a 230 page spec. (Ill admit, I did a literal search.) >Now I always found it strange >that you could pick an external voltage with a *lower* value than the >internale voltage, and not much lower at that. Remember that I do not >have a PC87366 chip myself, so had no chance to measure the physical >values at a chip's pins to confirm that the driver does the right thing. > > >If you are using an external reference, and it is NOT 1.211 volt, then I >am not surprised if your voltages are not correct. > > Ahh. I believe Ive found a bug. the ? : assignment above chooses the larger number for the external vref, which disagrees with your explanation, and the doc refs. So attached patch fixes that, and adds 3 module_params vrefext allows changes to accommodate different boards vrefscale allows +/-50mv around nominal 2450 mv vrefint - not entirely necessary, for real tweakers. BTW, board uses external ref, and is 1.211v (per PHK's measures) I havent checked vrefext for myself. Im getting numbers in close correspondence with PHKs. I'll trim it back (the params) to suit. FWIW, this module-side tweak could be avoided by hacking sensors.conf but its ugly to work vref scaling into ALL the voltage formulas, and thermistors too. I gather one can use inputs as factors in computations too, but the man-page shows no way of adding new symbolic constants. (I added this as a support ticket so it doent get lost b4 the libsensors rewrite) other thought was to add a new sysfs node - a vref-set-point. This makes it easy to tweak the scaling/calcs done by the drivers, so is transparent to sensors.conf, keeping it less cluttered than constants. But perhaps this is too available for disciplined use; the mod-param can be set once in /etc/modprobe.d/sensors, and forgotten. >Note that the reference voltage is used to compute the voltages and the >thermistor-based temperatures, NOT the diode-based temperatures. The TMS >logical device is influenced by the reference voltage, but the value >isn't used in computations AFAIR. > > > yes - thats what I meant by saying: I think my temp answers lie elsewhere (perhaps a cast to s8 for the register value), BTW, you recall correctly; for thermistors: #define IN_FROM_REG(val,ref) (((val) * (ref) + 128) / 256) for diodes: #define TEMP_FROM_REG(val) ((val) * 1000) but that still leave me with an impossibly high value for a low-power, fanless cpu >>BTW, (since Im writing), my patchset for pc87366 sensors-dev-attrs >>(done against rc4-mm1), also applies (clean, iirc) to rc5-mm1, and >>works there. >> >> > >I'm sorry, I didn't have the time to look at it yet. I'm currently busy >with i2c core changes and other drivers (i2c-viapro and it87.) Please be >patient. > > > I can see youre busy, I hope my update wasnt an annoyance, however momentary. In any case, I gather that 13-rcX is closed for all new features, so I regard it was info only. Ill revalidate when 13-final is out, and update if needed. -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: pc87360-vref-option.patch Url: http://lists.lm-sensors.org/pipermail/lm-sensors/attachments/20050812/52220048/attachment.pl
