Re: sysmon_envsys - Setting sensor on-chip thresholds from userland

[Christoph Egger <Christoph_Egger%gmx.de@localhost>]

Paul Goyette wrote:
> I'm working on a driver for On-DIMM temperature sensors for memory 
> modules, similar to OpenBSD's sdtemp but with sysmon_envsys(9) support.

Great!

> These sensors have on-chip limit registers for a Lower, Uppper, and 
> Critical boundary setting, and have on-chip comparison and flag-bit 
> setting when the actual temp transitions from one zone to another.
> (The sensor chips also have an event output pin, which might be able to 
> generate an I2C interrupt, but that's a question for another day!)
>
> I've been muddling through the sysmon_envsys stuff, and I'm having 
> difficulty understanding how a userland program might provide some 
> threshold values to the driver.  Would it be correct (or at least 
> acceptable) for the driver's sensor refresh routine to grovel through 
> the sensor's event queue to find the values associated with certain 
> monitor types (ie, ENVSYS_FMONWARNUNDER, WARNOVER, and CRITOVER)?

Yes, that's right.

> I'm also a bit more confused about all this, since the example in the 
> sysmon_envsys(9) man page seems to use a hard-wired threshold for 
> setting the sensor state, rather than using a limit from userland.
>
> Any guidance in this area would be greatly appreciated.

When I wrote the amdtemp(4) driver, I used sys/dev/acpi/aiboost.c
as example code.
You in turn may use amdtemp(4) and aiboost(4) as example code :)

In your softc you need these two members:

        struct sysmon_envsys *sc_sme;
        envsys_data_t *sc_sensor;

In the attach routine, run the envsys setup.

The refresh routine is of type
void (*refresh)(struct sysmon_envsys *sme, envsys_data_t *edata);

In the refresh routine read out the sensor data fill out edata.

If all is done right, envstat will show your driver and
the sensor data your driver provides.


Christoph