rawhide and cooker compared

[Hans de Goede]

Jakub Jelinek wrote:
> On Thu, Feb 21, 2008 at 12:28:03PM +0100, Hans de Goede wrote:
>> * sensors working with lm_sensors
>>
>> Note: I'm an upstream lm_sensors contributer and co-maintainer of most 
>> sensor related packages in Fedora.
>>
>> About lm_sensors not being installed by default, thats because for the 
>> average users lm_sensors is not usable as it requires manual configuration.
>>
>> About telling the user that he should run sensors-detect after installing 
>> lm_sensors, when and how do you envision this being told to the user?
> 
> Is there a subset of the probes sensors-detect does that can be done safely
> always, on all hardware (i.e. have sensors-detect --non-interactive which
> would skip the dangerous probes it asks for confirmation)?  If so, firstboot could
> run that.
> 

There are some (sortof) safe probes that can be done to determine which hwmon 
IC's are present however even fi the IC's are known we still do not have a 
properly working config.

Assuming an i2c setup there are 3 layers involved in a hwmon setup, note that 
some hwmon IC's are isa based, and thus only have the last 2 layers:

1) There is some i2c master / controller
    which is the starting point to
    communicate with for software which
    wants to talk to hwmon IC's. Usually this can be detected
    by PCI id, and the driver for this gets autoloaded by udev.

2) To the i2c (or isa) bus are attached some hwmon IC's

    A problem here is that one needs to
    know which hwmon ic's are used and at
    which i2c addresses they reside. This
    is where sensors-detect does most of its probing,
    this is usually safe, but in the past we have known
    to brick thinkpads by accidentally writing to the i2c eeprom
    holding the CMOS password.

3) Each hwmon IC has a number of input
    pins which measure (for example)
    voltage, the question here is which
    voltage is connected to which pin, and
    if a voltage divider is present
    between the voltage line and the pin.

    The pin to valtage mappings and voltage dividers
    differ from motherboard to motherboard. This is the biggest problem
    really.


We (the lm_sensors) project have long had plans to fix 2 and 3 together by 
using dmi BIOS strings to identify the motherboard (circa 80% of motherboards 
have usable id strings, some contain useless strings like "To be filled by 
OEM"), and then have a database with know good configs for tried and proven 
motherboards. We already have a small database of config files here:
http://lm-sensors.org/wiki/Configurations

And for quite a few of those I have a mailfolder with dmidecode dumps providing 
the strings. There have been several projects already to try to get a system 
like this developed (should be trivial really), but none has lifted of, the 
main problem being lack of time, we really need someone to pull the cart on 
this one. Contributers much welcome!

Thanks & Regards,

Hans