Remove 'Wants' on phosphor-ipmi-host

The service file for phosphor-ipmi-host already has
WantedBy=multi-user.target, which will instruct systemd to queue a start
task for phosphor-ipmi-host. Having

Remove 'Wants' on phosphor-ipmi-host

The service file for phosphor-ipmi-host already has
WantedBy=multi-user.target, which will instruct systemd to queue a start
task for phosphor-ipmi-host. Having a 'Wants' statement here is
therefore unnecessary.

This duplicate start task can become a problem if phosphor-ipmi-host
fails to start in time and is restarted by systemd. When
phosphor-ipmi-net tries to start later, it can trigger another start
task for phosphor-ipmi-host within the start burst limit, putting
phosphor-ipmi-host into a permanent timeout state.

Tested: With this change, phosphor-ipmi-host no longer has "Start
request repeated too quickly" across 1,000 powercycles.

Signed-off-by: Benjamin Fair <benjaminfair@google.com>
Change-Id: Iabcc57e2ba73522d373d91d02b12fa6e6f389979

show more ...

ipmi-net: ensure eth device present

Some OpenBMC systems may support a mix of eth0-only and eth0/eth1 ports
on their BMC device. In some cases, this support may be in a shared
flash image that suppo

ipmi-net: ensure eth device present

Some OpenBMC systems may support a mix of eth0-only and eth0/eth1 ports
on their BMC device. In some cases, this support may be in a shared
flash image that supports both systems.

For example, IBM is building a system under the "p10bmc" bitbake machine
that only has a single ethernet port, eth0, wired up on it. The system
is still utilizing an AST2600, but only wired eth0 out of the system for
cost purposes.

Since the "p10bmc" machine is a shared target there is no way to remove
services that assume eth1 is present during the build process. Instead,
rely on the fact that the device will be present in /sys/class/net/
if in fact the system supports the network device.

Tested:
- Verified that a p10bmc Rainier machine with 2 eth ports still ran
these units for both ports
- Verified that a p10bmc Bonnell machine with only eth0 ran these units
for eth0 but not eth1

Signed-off-by: Andrew Geissler <geissonator@yahoo.com>
Change-Id: I7e9db01fe26a4195802db4dcccffb412fbbe4a86

show more ...

Change Requires to Wants phosphor-ipmi-host

phosphor-ipmi-net would fail to start with missing dependencies when
phosphor-ipmi-host failed to start, and phosphor-ipmi-net would not
restart in that c

Change Requires to Wants phosphor-ipmi-host

phosphor-ipmi-net would fail to start with missing dependencies when
phosphor-ipmi-host failed to start, and phosphor-ipmi-net would not
restart in that case.

Similar change for kcsbridge in
https://gerrit.openbmc.org/c/openbmc/kcsbridge/+/53987

Change-Id: I42d179159c6e3e51db7cf5404ec5cd464083add6
Signed-off-by: Anthony <anthonyhkf@google.com>

show more ...

Ensure device to be ready before service

`phosphor-ipmi-net@.service` doesn't wait until
`sys-subsystem-net-devices-%i.device` to be ready so that it's possible
to fail to start the service.

Tested

Ensure device to be ready before service

`phosphor-ipmi-net@.service` doesn't wait until
`sys-subsystem-net-devices-%i.device` to be ready so that it's possible
to fail to start the service.

Tested:
The service now starts after the device is active and thus service now
is able to run into active(running) without any issue.

Change-Id: I28dd312bde5077d2ceb9c7d741095f297e63d178
Signed-off-by: Tom Tung <shes050117@gmail.com>

show more ...

spawn one rmcpp bridge per interface

According to the new architecture, each bridge should be a separate
process, or at the very least, be attached on a separate D-Bus
connection wit

spawn one rmcpp bridge per interface

According to the new architecture, each bridge should be a separate
process, or at the very least, be attached on a separate D-Bus
connection with a well-known name that corresponds to the channel name.

This commit brings netipmid to a state where it can be launched as:
systemctl start phosphor-ipmi-net@eth0

with a parameterized unit file and socket file that binds the socket to
the interface specified. If not launched this way, it will by default be
bound to all interfaces.

This includes the new parameterized service and socket file and the
autoconf/automake magic to install them to the correct place.

Tested-by: launch netipmid via current unit file/socket file
launch netipmid via parameterize unit file/socket file

Change-Id: Ib202015fb560da269e535f11c0ac316b17f6c262
Signed-off-by: Vernon Mauery <vernon.mauery@linux.intel.com>

show more ...