Re: [lm-sensors] [PATCH v3 1/2] hwmon: (coretemp) Don't use

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

Hi Jean,

On Tue, Sep 27, 2011 at 09:05:07AM -0400, Jean Delvare wrote:
> Hi Guenter,
> 
> Sorry for the late reply.
> 
> On Wed, 21 Sep 2011 14:28:18 -0700, Guenter Roeck wrote:
> > On Wed, Sep 21, 2011 at 05:01:58PM -0400, Jean Delvare wrote:
> > > On Wed, 21 Sep 2011 13:35:28 -0700, Guenter Roeck wrote:
> > > > Not sure about discarding the value either. If we drop the attribute if
> > > > ttarget is 0, people will wonder where tempX_max disappeared to. Worst
> > > > case tempX_max will show up at the same temperature as tempX_crit, so we
> > > > would know (or learn) about it and users would still have the attribute
> > > > available.
> > > 
> > > This is a read-only value, there's no point in making it "available" if
> > > the value is known to be wrong.
> >
> > But how do we know that or if a reading of 0 would be wrong ? Or a reading
> > larger than X ?
> 
> I'm not quite sure what your concern is. I never talked about filtering
> large register values (which would mean low ttarget temperatures -
> remember that ttarget = tjmax - register). All my proposal is about is
> filtering register value of 0, because it means ttarget = tjmax, which
> in turn means that the supposed effect of t > ttarget (all fans forced
> to full speed) as no chance to happen because the CPU dies or stops at
> tjmax. The only reason why I propose this is because the bits in
> question are undocumented, so we can't rule out that there are models
> out there which do support the MSR in general but don't implement the
> specific bits.
> 
A value of 1 doesn't make much sense either, since t > ttarget means that the
temperature already reached tjmax, and the CPU is just as dead. And a value of,
say, 100 doesn't make sense either. So declaring that "0" is invalid but all
other values are valid is pretty much arbitrary, as would any other limits.
So why bother trying to declare what is valid and what isn't.

> Anyway, I'll propose a patch, maybe it will make my intents clearer.
> 
I am actually ok with adding it in. I just would not bother myself.

> > > Here again, I'm fine if your patch doesn't include that change, after
> > > all you're mainly reverting to pre 3.0 behavior so it makes sense to
> > > stick to what we had before for now. I can submit patches for the
> > > proposed changes later, and we can discuss them again then.
> >
> > Let's do it that way. I really want to limit the changes to 3.1
> > at this point - we are a bit too close to release for my liking.
> 
> Agreed. But anyway, with kernel.org being MIA for so long, this release
> cycle is nothing I like to start with.
> 
Good point ... at the time my assumption was that Linus would release 3.1 
either this week or next week, which as it looks like is not going to happen.

Thanks,
Guenter

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