Hi,
> This locking below achieves nothing.
Ok, so I was too cautious by locking read access to a one byte
structure. I'll change that and read additional documentation on SMP ;-)
> Probably something else in
> the device struct can be reused.
Right now, I don't see which. There are other spinlocks available in the
net_device structure, but they are used by the queuing code and we
should not give up the semantic that netif_set_operstate() can be called
from everywhere. One global spinlock may be acceptable for this special
case.
> I also don't think this should be conditional, either we want
> it or we don't.
The conditional stuff is inspired from my first 2.4 version, but I'm
happy to remove it.
Btw, can you also have a look the 2.4 backport of my link state
notification feature (posting available under
http://marc.theaimsgroup.com/?l=linux-netdev&m=103722967214290&w=2, just
one typo fixed in Configure.help since then). Is this stuff acceptable
for 2.4?
Cheers, Stefan
|