On Wed, Dec 09, 2020 at 09:37:39AM +0100, Tobias Waldekranz wrote: > I will remove `struct dsa_lag` in v4.
Ok, thanks. It would be nice if you could also make .port_lag_leave return an int code. And I think that .port_lag_change passes more arguments than needed to the driver. > > I don't think the DSA switch tree is private to anyone. > > Well I need somewhere to store the association from LAG netdev to LAG > ID. These IDs are shared by all chips in the tree. It could be > replicated on each ds of course, but that does not feel quite right. The LAG ID does not have significance beyond the mv88e6xxx driver, does it? And even there, it's just a number. You could recalculate all IDs dynamically upon every join/leave, and they would be consistent by virtue of the fact that you use a formula which ensures consistency without storing the LAG ID anywhere. Like, say, the LAG ID is to be determined by the first struct dsa_port in the DSA switch tree that has dp->bond == lag_dev. The ID itself can be equal to (dp->ds->index * MAX_NUM_PORTS + dp->index). All switches will agree on what is the first dp in dst, since they iterate in the same way, and any LAG join/leave will notify all of them. It has to be like this anyway.
