Increase timeout for host to respond to attentions

OpenBMC is dealing with some performance issues right after
a reboot. Systemd does almost too good a job at parallelizing
all of the services which causes a 6X hit on the processor
during the initial boot once mapper is ready.

In situations where the BMC is rebooted when the host is up,
this performance issue coupled with errors queued up by the
host to send down to the BMC, can cause OpenBMC to exceed
its current 3 second timeout for attentions.

This initial message from OpenBMC to the host is done to
check if the host is running. If the message times out then
it's assumed the host is not running which causes all sorts
of issues for the system.

Testing has shown it can take up to 10 seconds for the
path to complete. Given the uncertainty of future performance,
the unknowns on how many errors the host may have queued up,
and the minimal impact a larger timeout has, this commit is
increasing the timeout to 30 seconds.

Resolves openbmc/openbmc#2832

Change-Id: I0d1baf601b6441a149e786e1f5de607862d86a52
Signed-off-by: Andrew Geissler <geissonator@yahoo.com>
1 file changed