[lm-sensors] Backport of lm77.c for 2.4.x

[khali@linux-fr.org (Jean Delvare)]

Hi Michael,

> I'm currently working on a backport of Andras Bali's lm77.c (as found in
> current 2.6.x kernels) to kernel series 2.4.x. The driver is basically
> working on my local test platform (PC Engines WRAP2). Now the
> fine-tuning has to be done, which brings up several questions.
> 
> 1.
> I'd like to hear some suggestions on what proc files the driver should
> create, apart from the usual "temp". This might be necessary as Jean
> pointed out in IRC yesterday, since the LM77 offers 5 values rather than
> just 3:
> 
> * Tmin, minimum temperature (if the current temperature goes below this
> value, an alarm bit is set)
> * Tmax, maximum temperature (if the current temperature goes above this
> value, an alarm bit is set)
> * Tcrit, critical temperature (if the current temperature reaches or
> goes above this limit, the LM77 will reset the system until temperature
> reaches Tcrit - Thyst
> * T, current temperature
> * Thyst, hysteresis
> 
> There is just one hysteresis value that is used wit Tmin, Tmax and
> Tcrit.
> 
> Currently I think of putting Tmin, Tmax and T into the "temp" proc file,
> and:
> a. Tcrit and Thyst into two distinct files, or
> b. Tcrit and Thyst into one file.
> 
> Suggestions?

I'm fine with this proposal, with a preference for option (b).

> 2.
> I'd like to allow the user to enable the "fault queue" feature that the
> LM77 provides. Quoting the data sheet:
> 
> "A fault queue of up to 4 faults is provided to prevent false
> tripping when the LM77 is used in noisy environments. The 4
> faults must occur consecutively to set flags as well as INT
> and T_CRIT_A outputs."
> 
> I'll add a module parameter that allows the user to enable the fault
> queue upon initialization. But I think it might be useful to also add a
> proc file that allows to enable/disable the fault queue at a later
> point, without the need of reloading the module.
> 
> Objections? Any suggestion regarding the name for the corresponding proc
> file?

My personal opinion is that this kind of tuning should be done by the
BIOS, as the system manufacturer should know whether the fault queue is
needed for the hardware combination used and the expected conditions of
the system.

Do you have a personnal need for this feature? Supporting every feature
of every chip isn't our goal. However, if you have the need and insist
on including this feature in your driver, I won't object. "fault_queue"
seems to be a sensible name for the feature.

> 3.
> The LM77 offers a shutdown mode which reduces power consumption. Is
> there any support in the lm_sensors package for suspend/resume (ACPI)? I
> think it would make sense to enable the shutdown mode upon suspend and
> disable it upon resume.

No power state support in lm_sensors CVS (for Linux 2.4) and there
won't ever be.

No support in Linux 2.6 either, but this could change if anyone wants
to work on this. I have no idea how difficult this would be. Please
realize that hardware monitoring is a risky area when it comes to
stopping chips. Any hardware monitoring chip can virtually be used to
control the cooling of the system. If the system is put to sleep, it is
questionable whether we really want to immediately stop any fan which
could be controlled by the chip. Of course, it is reasonable to assume
that a sleeping system shouldn't produce too much heat, but we still
must pay attention to every possible situation if we ever want to add
power states support to our hardware monitoring drivers.

I also wonder if the power consumption of these chips is high enough to
simply care about it.

Thanks,
-- 
Jean Delvare