We've got two temperature sensors (DS18B20) on a 1-wire bus, one measuring the room temp, and the other measuring the temp inside a case, and the idea is we can change a fan pwm depending on the temp difference etc.

And it works well, for 24hrs a day, 7 days a week, until we start getting CRC errors, from which we never seem to be able to recover.

We've told w1_master_max_slave_count we have two devices and after the master finds 2 devices we stop the master searching.

We read the sensors about every 20/30 secs, and it all works well until we start getting CRC errors.

A reboot will fix the problem, BUT, as we don't power down the one-wire subsystem during a reboot, and a reboot fixes the problem, it seems to us that there must be a software solution to these CRC errors. i.e. the DS18B20 sensors are not switching internally into some sort of "silly bugger mode" from which only powering off and on again knock them back into life.

We have tried removing drivers and reloading modules, and using w1_master_remove followed by w1_master_add, but the problem persists.

Of course the main problem are these CRC errors happen only rarely, so it's difficult to try different possible solutions.

Is the a 1-wire expert in the house ?

Can anyone give some advice ?

Oh, yeah, we did try swapping out the DS18B20's for different devices, but it made no difference.

Thanks.

No one-wire expert, but I know the general idea. Bizarre method of doing anything, but anyhow…

Can we insert a cache flush between each reading? What's going on between readings? I take it one end is unpowered, so voltage on the one line needs to be cycled to keep the capacitors charged. Could this be going low?

It may be that your sensors are counterfeit. Most are, even those purchased from a reputable dealer. Try checking.

I had a bunch of DS18B20s, purchased through three different highly reputable dealers over the course of a year, and I found that every last one was fake when I checked. So I've ordered new parts from another reputable vendor but one that offers a choice up front on the order form between a fake sensor or a real Maxim DS18B20.

Thanks for the replies.

We do not run these things in parasitic mode. They are powered direct to Vdd.

We are just using the sysfs interface. Someone on the team is currently mucking about with the w1-therm driver to try and get some helpful diagnostics out of it.

I guess they could be cloned chips, but they have the Dallas mark and a proper looking batch code.

We've built up another similar system and so far that one is behaving, but it could go for days before the CRC error pops up. That is what is so annoying.

Well, if you're powered to Vdd, it's hardly 1 wire, but never mind.
Have you slapped an oscilloscope on it? I found an awful lot of errors in my time on borderline voltage levels.You have Vdd(+) & signal. How does it see 0V?

We are starting to suspect there is some sort of hardware problem. We now have built 3 identical systems, and only one (the original) is giving us problems.

But it could just be too soon.

Sounds like the time to get a hardware techie with his 'scope on the job.

Solved.

Turns out it was a solder bridge.

A bit intermittent because it wasn't a full bridge, but when stuff warmed up, contact was made.

A familiar story. It's so easily done, also.