Jean,
Taking on board your comments and looking more closely at the output of
the sensors command I can see that the values returned to userland from
the kernel should indeed be x1000 (magnitude 3) and not x100 (mag 2)!
>
> Your fix is probably incomplete: I guess that the high limits should be
> ~80 and ~70 degrees C, and the hyst limits (which appear to really be
> low limits, not hysteresis) presumably -128 degrees C. So the limits
> magnitudes are wrong too.
>
>
>> [root at noakes lm_sensors-2.10.3]# ipmitool -I open sdr elist
>> Temp | 01h | ok | 3.2 | 38 degrees C
>> Temp | 02h | ok | 3.3 | 42 degrees C
>> Planar Temp | 04h | ok | 7.1 | 32 degrees C
>> VRD 0 Temp | 05h | ok | 7.1 | 26 degrees C
>> VRD 1 Temp | 06h | ok | 7.1 | 27 degrees C
>> CMOS Battery | 10h | ok | 7.1 | 3.08 Volts
>> VCORE | 11h | ok | 3.2 | State Deasserted
>> VCORE | 12h | ok | 3.3 | State Deasserted
>> PROC VTT | 13h | ok | 7.1 | State Deasserted
>> 1.5V PG | 14h | ok | 7.1 | State Deasserted
>> 1.8V PG | 15h | ok | 7.1 | State Deasserted
>> Presence | 20h | ok | 3.2 | Present
>> Presence | 21h | ok | 3.3 | Present
>> Fan 1 | 30h | ok | 7.1 | 3525 RPM
>> Fan 2A | 31h | ok | 7.1 | 7650 RPM
>> Fan 2B | 32h | ok | 7.1 | 5400 RPM
>> Fan 3A | 33h | ok | 7.1 | 7725 RPM
>> Fan 3B | 34h | ok | 7.1 | 5550 RPM
>> Fan 4A | 35h | ok | 7.1 | 7425 RPM
>> Fan 4B | 36h | ok | 7.1 | 5475 RPM
>> Fan 5A | 37h | ok | 7.1 | 7500 RPM
>> Fan 5B | 38h | ok | 7.1 | 5475 RPM
>> Status | 40h | ok | 3.2 | Presence detected
>> Status | 41h | ok | 3.3 | Presence detected
>> VRM | 44h | ok | 3.2 | Presence detected
>> VRM | 45h | ok | 3.3 | Presence detected
>> OS Watchdog | 50h | ok | 7.1 |
>> SEL | 51h | ns | 7.1 | Disabled
>> Intrusion | 52h | ok | 7.1 |
>> Fan Redundancy | 54h | ok | 7.1 | Fully Redundant
>> ECC Corr Err | 01h | ok | 34.6 | Presence Detected, Configuration Error
>> ECC Uncorr Err | 02h | ok | 34.6 | Presence Detected, Configuration Error
>> I/O Channel Chk | 03h | ns | 34.6 | Disabled
>> PCI Parity Err | 04h | ok | 34.6 | EISA failsafe timeout, Bus
>> Correctable error
>> PCI System Err | 05h | ok | 34.6 | EISA failsafe timeout, Bus
>> Correctable error
>> SBE Log Disabled | 06h | ok | 34.6 |
>> Logging Disabled | 07h | ns | 34.6 | Disabled
>> Unknown | 08h | ns | 34.6 | Disabled
>> PROC Protocol | 0Ah | ns | 34.6 | Disabled
>> PROC Bus PERR | 0Bh | ns | 34.6 | Disabled
>> PROC Init Err | 0Ch | ns | 34.6 | Disabled
>> PROC Machine Chk | 0Dh | ns | 34.6 | Disabled
>> Memory Spared | 11h | ok | 34.6 | Fully Redundant
>> Memory Mirrored | 12h | ok | 34.6 | Fully Redundant
>> Memory RAID | 13h | ok | 34.6 | Fully Redundant
>> Memory Added | 14h | ok | 34.6 | Correctable ECC
>> Memory Removed | 15h | ok | 34.6 | Correctable ECC
>> PCIE Fatal Err | 18h | ns | 34.6 | Disabled
>> Chipset Err | 19h | ns | 34.6 | Disabled
>> Err Reg Pointer | 1Ah | ns | 34.6 | No Reading
>>
>> The attached patch files fix this problem for the FC7 version of
>> lm_sensors 2.10.3 !
>>
>
>
>> --- ./lib/chips.c.orig 2007-08-06 10:25:49.000000000 +1000
>> +++ ./lib/chips.c 2007-08-06 10:25:01.000000000 +1000
>> @@ -5551,7 +5551,7 @@ static sensors_chip_feature bmc_features
>> SENSORS_BMC_FAN1+9, RW },
>> BMC_SYSCTL_FAN1+9, VALUE(1), 0 },
>> { { SENSORS_BMC_TEMP1, "temp1", NOMAP, NOMAP, R },
>> - BMC_SYSCTL_TEMP1, VALUE(3), 2 , "temp1_input", 3 },
>> + BMC_SYSCTL_TEMP1, VALUE(3), 2 , "temp1_input", 2 },
>> { { SENSORS_BMC_TEMP1_MIN, "temp1_min", SENSORS_BMC_TEMP1, SENSORS_BMC_TEMP1,
>> RW },
>> BMC_SYSCTL_TEMP1, VALUE(2), 2 },
>>
>
> Not correct, sorry. The temperature magnitude is standardized to 3 for
> all drivers in Linux 2.6 (see Documentation/hwmon/sysfs-interface). If
> you had to change it to 2 here, it means that the bmcsensors driver
> itself does not respect the standard. The bmcsensors driver needs to be
> fixed, not libsensors.
>
I'm using ipmisensors, not the older (and ?obsolete?) bmcsensors but the
same comment applies in either case.
> Or rather, libsensors needs to be fixed for all the other temperature
> channels, to use the standard magnitude of 3 instead of 2.
>
> Yani, looking at the patch which introduced this bug:
> http://www.lm-sensors.org/changeset/2888
> I have too comments:
>
> * in2 and in3 both map to in2_input, and after that all the voltage
> channels are shifted by one. I assume this is a bug? Or were you trying
> to compensate for voltages channels starting at 0 instead of 1?
>
This shifting bug is caused by the asynchronous nature of the IPMI
message passing i/f. The ipmidriver as originally coded simply fired
off multiple GET requests without waiting for a reply. It saved the
pointer to the last SDR object for which a request was issued in the bmc
object, overwriting any previous value. In my case, the 2nd GET was
issued before the response to the 1st GET arrived back resulting in an
incorrect value being set for the 2nd sensor and everything being
shifted by one. Only the last sensor was correct. I've developed a
patch to fix this behaviour.
I've also adjusted the SDR list creation to do a list_add_tail() rather
than list_add(). This ensures that the SDR list is in ascending order
of SDR number!
The "disconnect" between the kernel driver exporting the sensors data
via /sys/... and the usermode code needs to be eliminated. The 'spec'
does indeed state that TEMP and VOLT values are in milli<xxx> units so
the change to ipmisensors in decplaces() that returns 2 for TEMP
readings is incorrect and needs to be reverted to it's previous value of 3.
I've attached a (partial) patch which shows the changes I made to fix
the usage of the current_sdr pointer. It also reverts the change in
decplaces() that returned 2 for TEMP sensors instead of 3.
Of course along with these changes, lib/chips.c needs to be changed to
make ALL of the bmc/ipmi temp sensor values magnitude 3.
> * Everything in this patch (except the bugs) only duplicates the
> default mappings libsensors uses to convert Linux 2.4 symbols into
> Linux 2.6 sysfs file names. So I believe that the patch was simply not
> needed. As a matter of fact, the min and max limit symbols were left
> untouched, and work fine. Can you please comment on this? I plan to
> revert this changeset.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.lm-sensors.org/pipermail/lm-sensors/attachments/20070813/edfa1dab/attachment.html
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: ipmisensors-current_sdr-fix.patch
Url: http://lists.lm-sensors.org/pipermail/lm-sensors/attachments/20070813/edfa1dab/attachment.pl
