I got carried away agreeing, a tad too violently, with code reviewers,
but at least I'm testing diligently. The reason we cannot do persistence
on wifi yet is that wifi "set prop" ioctls use self's /dev/net/link, and
we're ensuring that properties are loaded _before_ /dev/net/link is
opened - bam, deadlock [1].
When wifi is brusselized (CR 6691666) and uses DLD control device,
things will be neater and cleaner. Meanwhile, I'm restoring 'dladm
init-linkprop -w' in net-physical and calls to dladm_init_linkprop() in
RCM/etc (these calls will have no effect on ethernet). I discussed with
Cathy and she agreed with the plan.
http://cr.opensolaris.org/~artem/pers/
Kernel code is unchanged since last time [2].
thanks,
-Artem
[1] open("/dev/net/link") invoked from dlmgmtd's property upcall
handler, trying to set persistent wifi properties, would sleep forever
in dls_devnet_prop_task_wait().
[2] http://cr.opensolaris.org/~artem/pers-0613/
_______________________________________________
networking-discuss mailing list
[email protected]
