I think the aging settings are per-bridge, not per-bridge-port, so the
policy control you're talking about wouldn't end up here on
/sbin/bridge link.

Agree.

However, I would argue even with hardware aging capability, we still
should use Linux for aging since all the controls are already there
and it just works.  It keeps the swdev model simple and the swdev
driver simple.  Do you have a counter-argument for why enabling
hardware aging would be better?

When traffic is being forwarded in hardware and if the entries are to
be software aged, the aging logic needs to periodically(once every
configured age interval) retrieve the FDB entry's hit bit in the
hardware. If this bit is not set it deletes the entry. If the hit bit
is set, it clears it. This could be CPU intensive if the FDB size runs
into few thousand entries.

If hardware aging is enabled, the above job is offloaded to hardware.

The current aging logic does not  differentiate between software
forwarded and hardware forwarded entries. When the traffic is being
forwarded in hardware, the aging logic is deleting the entries after
age interval.

So the solution could be

1)When hardware aging is enabled, only the notifications(probably via
br_fdb_external_learn_delete())  delete the fdb entries.
2)When software aging is enabled, aging logic needs to differentiate
between software forwarded and hardware forwarded entries. For entries
forwarded in hardware, the above aging logic needs to be implemented.



On Fri, Jan 2, 2015 at 11:27 PM, Scott Feldman [off-list ref] wrote:

On Fri, Jan 2, 2015 at 9:14 AM, Siva Mannem [off-list ref] wrote:

quoted

Hi,

quoted

+.BR "learning_sync on " or " learning_sync off "
+Controls whether a given port will sync MAC addresses learned on device port to
+bridge FDB.
+

For the FDB entries synced from device port to bridge FDB, can the
device port also mention that it will take care of aging the synced
entries? I am thinking of a use case where the port supports hardware
learning and hardware aging?

I think the aging settings are per-bridge, not per-bridge-port, so the
policy control you're talking about wouldn't end up here on
/sbin/bridge link.

However, I would argue even with hardware aging capability, we still
should use Linux for aging since all the controls are already there
and it just works.  It keeps the swdev model simple and the swdev
driver simple.  Do you have a counter-argument for why enabling
hardware aging would be better?

-scott