Re: [PATCH v8 net-next 2/2] net: dsa: mv88e6xxx: mac-auth/MAB implementation

[netdev]

On 2022-11-15 16:12, Ido Schimmel wrote:
> On Tue, Nov 15, 2022 at 11:36:38AM +0100, netdev@xxxxxxxxxxxxxxxxxxxx 
> wrote:
> > On 2022-11-15 10:58, Ido Schimmel wrote:
> > > On Sat, Nov 12, 2022 at 09:37:48PM +0100, Hans J. Schultz wrote:
> > > > diff --git a/drivers/net/dsa/mv88e6xxx/global1_atu.c
> > > > b/drivers/net/dsa/mv88e6xxx/global1_atu.c
> > > > index 8a874b6fc8e1..0a57f4e7dd46 100644
> > > > --- a/drivers/net/dsa/mv88e6xxx/global1_atu.c
> > > > +++ b/drivers/net/dsa/mv88e6xxx/global1_atu.c
> > > > @@ -12,6 +12,7 @@
> > > >
> > > >  #include "chip.h"
> > > >  #include "global1.h"
> > > > +#include "switchdev.h"
> > > >
> > > >  /* Offset 0x01: ATU FID Register */
> > > >
> > > > @@ -426,6 +427,8 @@ static irqreturn_t
> > > > mv88e6xxx_g1_atu_prob_irq_thread_fn(int irq, void *dev_id)
> > > >  	if (err)
> > > >  		goto out;
> > > >
> > > > +	mv88e6xxx_reg_unlock(chip);
> > >
> > > Why? At minimum such a change needs to be explained in the commit
> > > message and probably split to a separate preparatory patch, assuming the
> > > change is actually required.
> >
> > This was a change done long time ago related to that the violation 
> > handle
> > function takes the NL lock,
> > which could lead to a double-lock deadlock afair if the chip lock is 
> > taken
> > throughout the handler.
>
> Why do you need to take RTNL lock? br_switchdev_event() which receives
> the 'SWITCHDEV_FDB_ADD_TO_BRIDGE' event has this comment:
> "/* called with RTNL or RCU */"
> And it's using br_port_get_rtnl_rcu(), so looks like RCU is enough.

As I understand, dsa_port_to_bridge_port() needs to be called with the 
NL lock taken...