if_muge: move chip type detection into muge_attach

[emaste]

Summary

Chip type detection was previously done in muge_attach_post, which happens after attach. We want the driver to instead fail to attach for unsupported devices.

(I'm not sure this is the best place for this code, but ue_attach_post isn't able to return an error and ue_attach_post_sub seems too late.)

Also replace call to muge_detach on failure with reverse teardown of only those setup tasks that have been performed.

[emaste]

alternatively ue_attach_post could be made to return an error

[• hselasky]

sys/dev/usb/net/if_muge.c

2120 ↗

(On Diff #42892)

Shouldn't this block be before uether_ifattach() else the post-attach will race with the reading of the revision register.

[emaste]

sys/dev/usb/net/if_muge.c

2120 ↗

(On Diff #42892)

That panics in lan78xx_read_reg because it relies on something set up by uether_ifattach.

[• hselasky]

I see. I then suggest:

1. add a flag to store the status from xxx_chip_init() into the softc.
2. After err = uether_ifattach(ue); and the if(err) you add:

usb_proc_drain(&ue->ue_tq);
if (sc->someflag)
goto xxx_error;

[• hselasky]

Sorry, usb_proc_drain() won't work - it will kill the taskqueue. You'll need to use usb_proc_mwait().

[emaste]

Thanks for rS334158 and the suggestion on the approach taken here.