Re: [PATCH] net/hsr: Add support for the High-availability Seamless Redundancy protocol (HSRv0)

[Arvid Brodin <arvid.brodin@xdin.com>]

On 2013-06-25 02:11, Stephen Hemminger wrote:
> On Mon, 24 Jun 2013 09:44:01 -0700
> Arvid Brodin <arvid.brodin@xdin.com> wrote:
> 
>> +
>> +static void __hsr_set_operstate(struct net_device *dev, int transition)
>> +{
>> +       if (dev->operstate != transition) {
>> +               write_lock_bh(&dev_base_lock);
>> +               dev->operstate = transition;
>> +               write_unlock_bh(&dev_base_lock);
>> +               netdev_state_change(dev);
>> +       }
>> +}
> 
> Don't use dev_base_lock, you need to use rtnl_mutex() instead when
> making state changes.

This code is exactly what set_operstate() does net/core/rtnetlink.c. Are you sure I should
change it to use rtnl_lock()/rtnl_unlock()?


> Also, don't set operstate directly, it doesn't run link watch events
> which services can see. Instead use this to propogate up changes.
> 
> void netif_stacked_transfer_operstate(const struct net_device *rootdev,
> 					struct net_device *dev)

That function does not work for devices with more than one underlying physical interface.
A HSR device works (is OPER_UP) as long as at least one of its two physical interfaces is
up. Using that function to set carrier state on a HSR device would defeat the whole
purpose with HSR (redundancy). I haven't been able to find an existing interface that sets
operstate for devices that use multiple physical interfaces.

As for link watch events, doesn't the call to netdev_state_change() take care of that?


-- 
Arvid Brodin | Consultant (Linux)
T: +46-8-56254286 | M: +46-70-9714286 | arvid.brodin@xdin.com
XDIN AB | Knarrarnäsgatan 7 | SE-164 40 Kista | Sweden | xdin.com