IPMI SEL refactoring comments
Tom Joseph
tomjose at linux.vnet.ibm.com
Tue Nov 27 00:26:25 AEDT 2018
Hello Jason,
I reviewed the patches for the IPMI SEL refactoring and wanted to
summarize the concerns with the proposed design. There are some current
features that are broken.
1) The sensor numbers are arbitrary in the proposed implementation and
does not account for the presence/error corresponding to inventory
objects. On IBM systems the host firmware does not rely on the SDR
repository in BMC to figure out the sensor numbers, rather the sensor
numbers are programmed into the host firmware at the build time. So the
sensor numbers cannot be arbitrary.
> I know you are working on a proposal to have a customizable map of
sensor number to path. The support is only for threshold based sensors(
targeting only /xyz/openbmc_project/sensors namespace) , it needs to
extend support for discrete sensors. The sensor types are limited to few
types(temperature, voltage,current, fan_tach, power), need to support
other sensor types in the specification. It might not be possible to
deduce the sensor type from the D-Bus object path for all sensors.
2) OpenPower systems like Witherspoon and Zaius relies on eSEL(extended
SEL format) to report debug data from the host firmware. The eSEL data
is added to the D-Bus native error log via the Add SEL entry command.
This is not handled with the refactoring.
> This is an openpower requirement and needs to be handled in an
openpower IPMI OEM library/application and not in the phosphor
implementation. A D-Bus error log containing eSEL can be created on
completion of transfer of eSEL and not depend on Add SEL entry. The
callouts associated with the eSEL can be added to the D-Bus error log by
keeping a watch on the IPMI SEL entries. This might need changes to
phosphor-logging API to support adding callout after creation of error
log. This is a proposal and looks viable.
3) The current implementation the Get SEL entry is translating the D-Bus
error log to IPMI SEL format. If there is a FRU callout associated with
the D-Bus error log(D-Bus object paths in the inventory namespace) it is
mapped to the sensor number associated with FRU and reported through SEL.
> The implementation of the Get SEL entry needs to adapt to include all
the objects paths(the patchsets target only /xyz/openbmc_project/sensors
namespace) and map the D-Bus object path to the sensor numbers(based on
the proposed customizable map).
4) In the current implementation all the D-Bus error logs have 1x1
mapping in the IPMI SEL. In the proposed solution the users will have to
rely on both D-Bus error logs and IPMI SEL to get the complete picture.
The customers cannot rely only on IPMI SEL as an event repository.
> In the current proposal D-Bus error logs reported by BMC will not be
shown by SEL commands. I am okie with Jason's proposal to set up a D-Bus
match (either by ipmid or phosphor-sel-logger) that will watch for new
logging D-Bus objects to be created and add a SEL record for them on
creation.
https://gerrit.openbmc-project.xyz/#/q/status:open++topic:%22IPMI+SEL%22
Regards,
Tom
More information about the openbmc
mailing list