Re: [Freeipmi-devel] Sensors incorrect assumption about discrete sensors

[Anand Babu]

Original code was correct.

"Generic - discrete sensor" and "Sensor Specific - discrete sensors"
are different.

Original code classified event-reading based on 36.1.

When event/reading type code is between 0x01 to 0x0C, you have to 
sub switch-case using table 36.2.

It was confusing because of the MACRO names.
We should rename them as
IPMI_SENSOR_CLASS_DIGITAL_DISCRETE =>
    IPMI_SENSOR_CLASS_GENERIC_DISCRETE
IPMI_SENSOR_CLASS_DISCRETE =>
    IPMI_SENSOR_CLASS_SENSOR_SPECIFIC_DISCRETE.

Happy Hacking,
-ab

,----[ Albert Chu <address@hidden> ]
| It seems my "fix" of ipmi_sensor_classify was only half a fix.  Fish's
| sensors code incorrectly assumes that a "discrete sensor" has an
| event/reading type code of 0x6Fh.  Thus, it always interprets states
| based on the the sensor specific data (table 36-3 of the IPMI spec).
| Instead it should check the event/reading type code first, to make
| sure it is 0x6Fh.  If it isn't 0x6F, then it should be using the
| generic sensor data (table 36-2).
| 
| I think this only affects 1 sensor, Power Unit Redund, on Tiger4.  So
| I'm not too hung up delaying Alpha5-Qa1 for this bug.  But I think its
| something that should be fixed soon.
| 
| Al
`----

----- Original Message -----
From: Albert Chu <address@hidden>
Date: Friday, March 26, 2004 2:53 pm
Subject: [Freeipmi-devel] Couple of major changes ...

> Made a few changes that are pretty significant that I thought I should
> mention.
> 
> unassemble_ipmi_kcs_pkt:  similar to ipmi_lan_pkt, there is no 
> guaranteethat the packet returned from ipmi_kcs_read will be 
> atleast the size
> of tmpl_hdr_kcs + tmpl_cmd.  In particular, if comp_code != 
> success, the
> package may be much smaller.  So we cannot just error out if the 
> packetis smaller than we expect.
> 
> tmpl_get_sensor_threshold_reading_rs: Removed the "reserved3" 
> field. 
> This field is optionally returned from the BMC.  On tiger4, it is not
> returned at all.  On those machines that it is returned,
> unassemble_ipmi_kcs_pkt will ensure it isn't copied at all to the
> obj_cmd buffer.
> 
> ipmi_sensor_classify: This function returned incorrect classes on some
> event type codes, leading to some incorrect output in sensors.  As far
> as I can tell, this did not break anything, although there was a 
> chanceit could have.
> 
> Al


-- 
 _.|_ 
(_||_)
Free as in Freedom <www.gnu.org>