Re: [lm-sensors] [PATCH] hwmon: (adt7475) Fix temperature fault

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 11/15/2009 05:54 PM, Jean Delvare wrote:
> On Mon, 9 Nov 2009 13:26:21 +0100, Jean Delvare wrote:
>> Hi Hans,
>>
>> On Mon, 09 Nov 2009 11:51:54 +0100, Hans de Goede wrote:
>>> I've a adt7475 on my home development machine in the Netherlands
>>> (I'm currently in Brno).
>>>
>>> I can (and would like to) verify this fix when I'm back home (friday).
>>> The adt7475 on my machine is used to control an additional fan, and has
>>> no external temperature sensors connected (AFAIK).
>>>
>>> I've tested the original adt7475 driver on this machine, but I may
>>> have simply out "temp1 ignore" (and temp3) in my sensors.conf missing
>>> this.
>>
>> Thanks for the feedback. I have tested on 3 dumps I have here, one
>> ADT7475 and two ADT7473. The ADT7475 reports -64°C for temp1 and temp3
>> and has both fault bits set to 1. The two ADT7473 reports reasonable
>> values for both temp1 and temp3 and have the fault bits set to 0.
>> Assuming that the ADT7473 and ADT7475 are almost the same chip and
>> should thus behave the same, my change would be correct.
>>
>> I have also noticed that Jordan's original code did not have the
>> inversion:
>>
>> http://lists.lm-sensors.org/pipermail/lm-sensors/2008-January/022338.html
>>
>> 	case FAULT:
>> 		/* Note - only for remote1 and remote2 */
>>
>> 		ret = sprintf(buf, "%d\n",
>> 			(data->alarms&  (sattr->index ? 0x4000 : 0x8000)));
>> 		break;
>>
>> but it lacked normalization (output value 0x4000 or 0x8000 instead of
>> 1.) Maybe when normalizing the logic was inverted by accident?
>>
>> The few ADT7475 data points I have found on the web suggest that on
>> many boards temp1 and temp3 are not connected. This could explain why
>> the bug went unnoticed so far. Why vendors use these chips, which are
>> specifically designed for automatic temperature-based fan speed
>> control, in designs where they don't bother wiring the thermal sensors,
>> is beyond me.
>>
>> Anyway, I will wait for you to return home and test on your system,
>> before sending the patch to Linus.
>
> Hans, did you have the time to test this patch?
>

I did not get around to it sofar, but I had already planned to do it this
evening, so I just have. Your patch seems to be correct when run on real
hardware (and the current code in the kernel is thus wrong).

Regards,

Hans


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