Transitioning away from negative-errno-on-fail

Patrick Venture venture at google.com
Wed Jun 26 07:27:04 AEST 2019


Hi,

Many platforms leverage --enable-negative-errno-on-fail in
phosphor-hwmon, such that the sensor values, when unavailable are set
to -errno.  And it's up to other logic to detect that a sensor has
gone below some expected range.  This was fine with sensors that
didn't go negative, and for the most part was fine in general.

However, phosphor-hwmon is introducing a new option:
--enable-update-functional-on-fail

Each sensor reported by phosphor-hwmon will implement the Functional
decorator defined:
xyz/openbmc_project/State/Decorator/OperationalStatus.interface.yaml

There is a boolean property: Functional  when true (the default) the
value reported is valid and teh sensor is happy.  If the sensor has a
fault file, then if this reports a fault, the value will go to false.
If you enable update-functional-on-fail, then this property will also
go to false if the daemon was unable to read the sensor from sysfs.

Brandon is working on integration with phosphor-host-ipmid, such that
it'll be available there.  In this case, we'd like to deprecate
negative-errno-on-fail as an option from phosphor-hwmon.

Timeline:
    I've seen a lot of platforms coming in with that setting enabled
(including quanta-q71l).  I'd like to transition these platforms onto
update-functional-on-fail as a configuration over the next few weeks
and in the next few months, deprecate negative-errno-on-fail.  I'll
add a deprecation warning to the configuration and then eventually
it'll be dropped from support.

I'm going to attempt to reach out to each platform using this feature
once all the pieces are in place upstream and work with them on
transitioning away from it.

Patrick


More information about the openbmc mailing list