Jean-Philippe Francois wrote: > > Cirrus solution also adds a flag gPhyAutoNegoDone, so that if you hit > the return -1 code path, you still get a chance for auonegotiation when > first packet is transmitted (hard_start_xmit), and then another each > time there is a tx_timeout. > > We can : > -keep the actual code and come back to it after mainline submission. No, there's really no excuse for submitting code where we already know that it should be done better. > -take Cirrus fix. Yes, I'll try to do that, see below.[1] > -use a workqueue that do link monitoring at regular interval. > I'd like to do [1] first, then we can come back to this one. [1] I will merge the current driver with the version in linux-2.4.21-rmk1-crus-1-4-3 [3] and hope that this will fix the problem. The version resulting from this can be used as base for any other funky feature[2] or further cleanup in case that should be necessary. [2] porting it to NAPI, anyone? [3] Please tell me immediately if there is something newer available.