Issues with lm_sensors on Intel SE7501cw2
mds at paradyne.com
Tue Oct 21 02:23:04 CEST 2003
one program that may be helpful for you is prog/pwm/pwmconfig.
If it's working well it should confirm the motherboard setup
you documented below, including the 'cross-connection'.
Marc Rieffel wrote:
> > > [root at gb0007 lm_sensors-2.8.1]# rmmod w83781d
> > That module isn't supposed to be loaded at this moment. There's
> > something unclear here, the rmmod command should return an error. Can
> > you test that again, checking with lsmod which modules are loaded each
> > time? Don't you have some kind of module auto-loading making trouble
> > here? Taking a look at dmesg while loading/unloading the modules could
> > help.
> I tried this again and it worked normally. With 83781d loaded, I get nothing. With w83627hf loaded, I see the sensors. Maybe I cut and pasted the wrong thing in my initial email.
> > > 2. The max6650 fan speed measurement seems to be all or nothing. If
> > > there's a (Sunon) fan connected and operational, it reports
> > 7650RPM.
> > > If there's no fan connected, it reports 0 RPM. If I physically slow
> > > the fan down, it still report 7650 RPM. (The tachs on the w83627hf
> > > sensor do show decreased rpm when I impede their fans).
> Any thoughts on this?
> > > 3. The fan speeds reported (7650 or 18,000-21,000) seem to be
> > > inconsistent with the advertised speeds. Changing the divisors
> > > doesn't seem to affect the reported values.
> Or this?
> > > 4. The speed control on the max6650 doesn't seem to work.
> > See below.
> > Strange. Could be because the driver assumes too much conditions that
> > aren't correct for your system. The MAX6650 driver was contributed by
> > John Morris. John, could you please take a look at Mark's problem?
> > > 5. The w83627hf's "pwm1" setting seems to have no effect:
> > I think that PWM needs special wiring on the motherboad in order to
> > work. Could it be that your motherboard doesn't have it?
> Well, I've figured out part of the answer to 4 & 5. Here's how my motherboard is set up.
> Fan Controlled by Monitored by
> J29/sys3 w83627hf pwm1 max6651 fan1
> J30/sys4 w83627hf pwm1 max6651 fan2
> J16/cpu1 n/a max6651 fan3
> J14/cpu2 n/a max6651 fan4
> J3/sys2 w83627hf pwm2 w83627hf fan1
> J1/sys1 w83627hf pwm2 w83627hf fan2
> J58/sys5 max6651 speed w83627hf fan3
> I can't begin to imagine why they have these chips cross-connected, with one chip controlling a fan and a different chip monitoring it. The Winbond has two pwm's, each controlling two fans, and it only operates in "open loop" mode. I read the max6651 data sheet (http://pdfserv.maxim-ic.com/en/ds/MAX6650-MAX6651.pdf), and it supports four modes, always on, always off, open loop (use pwm to achieve a desired voltage), and closed loop (use pwm to achieve a desired fan1 speed). Reading the configuration register from the driver suggests that it's in closed loop mode.
> # cat /proc/sys/dev/sensors/max6650-i2c-0-1b/xdump
> 38 42 255 0 0 0 31 2
> Bits 5 and 4 of 42 are 1 and 0
> The fact that changing the "speed" setting affects the speed of the J58 fan suggests that it's actually operating in "open loop" mode. Given that it isn't wired to monitor the fan that it's controlling, I would think there's no way that it could run in closed loop mode.
> Question: Am I interpreting the xdump values correctly?
> Request: How about making the xdump values easier to interpret, and making the configuration register writable in software, so that I can try things like setting it to always-off mode?
> > > 6. The w83627hf's "pwm2" setting is able to reduce the
> > speed of fans 1
> > > and 2, but only to about half. I can hear them change
> > speed when I do
> > > this. I can't get them to stop completely, though.
> > This may be a feature more that a bug. Fully starting a fan requires a
> > higher current. What's more, stopping a fan completely is somewhat
> > dangerous, isn't it? Same as for PWM1, I think that PWM2
> > works thank to
> > adequate wiring on your motherboard, but maybe the wiring and
> > electronics are such that the W83627HF isn't allowed to lower the fan
> > speed below a given threshold.
> The max6651 data sheet says something about how lowering the PWM and therefore the voltage on the fan may result in odd behavior, like the fan stopping completely or the tach signal not reporting at all or reporting too many times. I suppose that makes sense. That doesn't explain why setting PWM to zero doesn't stop the fan, though. If my observations are consistent with those of others, and there's no real way to stop the fans, I can live with that. I don't really need to stop the fans. I'm just trying to understand what's going on. So far everything I've tried has given me a result different from what I've expected, and that makes it hard for me to have much confidence that the system is doing what I think it is.
> > > 7. Changes to the "pwm2" setting seem to be transient. Something
> > > happens to change it back to 255.
> > >
> > > [root at oscarnode001 root]# cat
> > > /proc/sys/dev//sensors/w83627hf-isa-0290/pwm2 0 1
> > > [root at oscarnode001 root]# sleep 60
> > > [root at oscarnode001 root]# cat
> > > /proc/sys/dev//sensors/w83627hf-isa-0290/pwm2 255 1
> > Don't you by any chance have a monitoring daemon running on
> > the machine?
> I'm not aware of any, but RH9 may have installed something that I don't know about. How can I tell? What are the names of some monitoring daemons that I could check for?
> > > 8. Sometimes the "lm75" module loads and detects something. I'm not
> > > aware of there being any lm75's or compatible devices on
> > this system.
> > > (...)
> > > 9. The lm75's temperature readings are bogus. At an earlier point I
> > > got them to report reasonable values, but I can't get back to that
> > > state.
> > Could you provide the output of "i2cdetect 0"? I'd also be
> > interested in
> > the output of "i2cdump 0 0x1f" since you seem to have an unknown chip
> > there.
> ]# i2cdetect 0
> WARNING! This program can confuse your I2C bus, cause data loss and worse!
> I will probe file /dev/i2c-0
> You have five seconds to reconsider and press CTRL-C!
> 0 1 2 3 4 5 6 7 8 9 a b c d e f
> 00: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX
> 10: XX XX XX XX XX XX XX XX XX XX XX UU XX XX XX 1f
> 20: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX
> 30: 30 31 XX XX 34 35 XX XX XX XX XX XX XX XX XX XX
> 40: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX
> 50: UU UU XX XX UU UU XX XX XX XX XX XX XX XX XX XX
> 60: XX XX XX XX XX XX XX XX XX 69 XX XX XX XX XX XX
> 70: XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX XX
> > Strangely, sensors-detect didn't report any client found at these
> > addresses (0x48 and 0x49). Could you run sensors-detect again (unload
> > all client drivers first) and see if it now does? These LM75s could
> > actually be emulated by your w83627hf, although it's unusual
> > to see this
> > behavior when the w83627hf itself is on the ISA (as opposed
> > to I2C) bus.
> I tried it again. No lm75's. Output below.
> > > 10. The default value for the sensor type on the w83627hf's
> > > temperature sensor, according to sensors.conf, is
> > thermistor (3435).
> > > But for me it seems to be defaulting to "1", PII/Celeron Diode.
> > The "defaults" might be confusing, I agree. This refers to power-up
> > default for the chip. The BIOS may then reconfigure the chip
> > during the
> > boot process. I believe that this is what has happened there.
> > If you believe that we should say the things differently in the
> > configuration file, suggestions are welcome.
> It would be helpful if you included your clarification about chip vs. bios defaults.
> > Thanks for the long and detailed report. I doubt we'll be able to
> > quickly solve all the problems you encountered, since there are many,
> > still this will help us if similar problems are reported in
> > the future,
> > plus I hope I clarified some points.
> Thanks for your help.
> This whole thing has turned out to be much more complicated and difficult than I would have expected. I've been at it for over a week now, with thermometers, multimeters, oscilliscopes, duct tape and bailing wire. I finally have the temperature information I want, and I can tell if my fans are running or not, but I'm still not sure if sensors is reporting the correct fan speed.
> On the positive side, when I first tried lm_sensors a year or two ago, I got absolutely nowhere, so it's obviously come a long way since then. Keep up the good work!
More information about the lm-sensors