tree ab6b5dfa461ac4b8a6f3ff9a6c74bbdd168dd6cf
parent 4686bb5598712b6c7f5541c9ac5e7ac75c8e07df
author Andrew Geissler <geissonator@yahoo.com> 1547830848 -0600
committer Brad Bishop <bradleyb@fuzziesquirrel.com> 1550095456 -0500

Ensure ipmid doesn't start until on dbus

Other services rely on phosphor-ipmi-host providing needed dbus
interfaces once its been started. Due to other services providing
identical dbus object paths, the normal mapper wait services are not an
option here.

Ensure systemd does not consider this service active until it has
registered the host control dbus object.

I'm working on a long term fix to this in
https://gerrit.openbmc-project.xyz/#/c/openbmc/phosphor-objmgr/+/17990/
(a refactor with unit tests) and
https://gerrit.openbmc-project.xyz/#/c/openbmc/phosphor-objmgr/+/17837/
(close race condition due to introspection) but I think this change here
makes things better overall so I'd like to get it merged

Testing:
Verified systemd now waits for ipmid to really get up and running and
ran through HW CI

Jan 18 19:56:00 witherspoon systemd[1]: Starting Phosphor Inband IPMI...
Jan 18 19:56:03 witherspoon ipmid[1336]: JSON file not found
Jan 18 19:56:09 witherspoon systemd[1]: Started Phosphor Inband IPMI.

I had our internal test team try out this patch on a Romulus. They were
seeing this issue this is trying to fix on a very regular basis. With
this change, they completed 265 successful boot tests. The ipmid process
does host 2 well-know names, openbmc_project.Ipmi.Host and
xyz.openbmc_project.Control.Host. I verified via a bus capture that
xyz.openbmc_project.Control.Host is the last one to be put on the bus.

Resolves openbmc/phosphor-state-manager#2

(From meta-phosphor rev: bea22141ca95b38f30adcec248a853d3d268af79)

Change-Id: I532b31647951045adc8cbd0141e8630cdc350e27
Signed-off-by: Andrew Geissler <geissonator@yahoo.com>
Signed-off-by: Brad Bishop <bradleyb@fuzziesquirrel.com>