svn commit: r253255 - head/sys/netgraph/bluetooth/drivers/ubt

[Marko Zec]

On Saturday 13 July 2013 09:31:19 Mikolaj Golub wrote:
> On Fri, Jul 12, 2013 at 10:40:08PM -0700, Adrian Chadd wrote:
> > Here's what I have running on my stable/9 VIMAGE laptop.
> >
> > http://people.freebsd.org/~adrian/ath/20130712-vimage-default-attach-de
> >tach.diff
> >
> > How's this look to put into -HEAD?
>
> Your patch looks similar to Marko's:
>
> http://lists.freebsd.org/pipermail/freebsd-hackers/attachments/20121115/0
>824ca85/attachment.diff
>
> Marko took your idea of setting vnet0 context in
> device_probe_and_attach(), although there is no is_default_vnet check,
> and I also think it is not necessary here -- just unconditionally set
> vnet0 is ok.
>
> Also, as I understand, after handling vnet context in
> device_probe_and_attach(), there is no need in curvnet setting in
> kldload/kldunload, and Marko removed it.
>
> In net/if.c, similarly to you, Marko sets vnet context in if_detach(),
> but also in if_free(), so this would be good to commit too (I suppose
> Marko had good reasons to do this).
>
> I can't comment the netgraph related parts of your patch as I have no
> idea what is going on here. Though I think this could be committed
> separately if it is needed.

I fear both Adrian's and my patch fall short of addressing the issue of 
generic USB device attach handling in VIMAGE kernels, i.e. the patches 
discussed in this thread might be OK for preventing panics with ng_ubt, but 
what about USB ethernet or wifi devices?  If I recall correctly the USB 
stack has a separate thread which handles device attach events, and IMO the 
curvnet context should be set there to cover a broader set of different 
devices.

I might be able to find an USB ethernet device for testing next week, but 
ATM I don't have any piece of USB hardware I could experiment with...

Marko