On Sat, 19 Jun 2004 17:28:01 -0400, Jeff Garzik wrote:
> In Linux lists of model names are discouraged. It's not terribly bad in
> via-rhine, but overall these things wind up getting patches quite often,
> and become a maintenance annoyance.
>
> It's up to you as maintainer, but I would recommend removing the string
> completely. For dmesg/printk purposes, the user only needs to know they
> have a 'via-rhine' controller.
The reason I put that in is that lspci does not identify those chips
correctly (because models differ only by PCI revision, not PCI id)
and thus people get confused. But maybe I should rather file patches
against pci.ids. Okay, I think I'll remove the model names.
> >- dev = alloc_etherdev(sizeof(*rp));
> >- if (dev == NULL) {
> >+ dev = alloc_etherdev(sizeof(struct rhine_private));
> >+ if (!dev) {
> > rc = -ENOMEM;
> >- printk(KERN_ERR "init_ethernet failed for card #%d\n",
> >- card_idx);
> >+ printk(KERN_ERR "alloc_etherdev failed\n");
>
> this error message change seems like a step backwards... print out
> pci_name() or _something_ to let the user know which card failed.
It is indeed. I plan to clean up all error messages together (there
are other issues like where dev->name is defined, what information is
useful, should use a bit mask instead of debug level, etc.).
Roger
|