Re: [lm-sensors] [PATCH] lm-sensors: Add support for new hwmon

[Guenter Roeck <guenter.roeck@ericsson.com>]

On Wed, Jun 30, 2010 at 11:22:48AM -0400, Jean Delvare wrote:
> Hi Guenter,
> 
> On Wed, 30 Jun 2010 07:20:21 -0700, Guenter Roeck wrote:
> > On Wed, Jun 30, 2010 at 09:50:51AM -0400, Jean Delvare wrote:
> > > Hi Guenter,
> > > 
> > > On Tue, 29 Jun 2010 23:06:18 -0700, Guenter Roeck wrote:
> > > > This patch adds support for the following new attributes to libsensors and
> > > > to the sensors command.
> > > >
> > > > inX_lcrit
> > > > inX_crit
> > > > inX_fault
> > > > temp_lcrit
> > > > powerX_cap
> > > > powerX_max
> > > > powerX_crit
> > > > powerX_alarm
> > > > powerX_fault
> > > > currX_lcrit
> > > > currX_crit
> > > > currX_fault
> > > 
> > > Fair enough, assuming these matches the additions to
> > > Documentation/hwmon/sysfs-interface. I'm a little curious about
> > > inX_fault, powerX_fault and currX_fault though. _fault files are for
> > > hardware failures. While I can imagine a thermal diode failing, how
> > > could a voltage or current sensor be broken?
> >
> > I thought about that myself, ie if it would be better to introduce
> > _lcrit_alarm and _crit_alarm instead. My ultimate conclusion was that the chips
> > I am working with right now are voltage controllers, and thus exceeding the limits
> > points to a fault rather than a critical limit alarm. Action taken by the chip 
> > is typically also drastic; raising a fault alarm is only the least drastic action
> > possible. More likely the chip will be configured to reset the board or shut down
> > power completely.
> > 
> > This applies to chips which have both "alarm" and "critical" limits.
> > 
> > So even if the event does not exactly match the "fault" description in the API,
> > I concluded that "fault" is a better match to what happens than "crit_alarm" or
> > "lcrit_alarm". I don't feel too strongly about that, though, so if you disagree 
> > I'll be happy to change the code and API to "lcrit_alarm" and "crit_alarm"
> > for those sensors.
> 
> Yes, I disagree, and yes, please change. "alarms" in the hwmon
> framework are flags raised because of out-of-limits measurements, and
> we should stick to this to avoid any confusion. And "faults" are
> something else. Let's stick to this.
> 
Ok, I'll do that. I'll also re-submit the driver for SMM665 to reflect the changes.

> in[0-*]_fault should probably be removed from
> Documentation/hwmon/sysfs-interface altogether, to clear the confusion.

Guess so.

Guenter

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors