Re: [PATCH net-next 06/17] net: dsa: Call driver's setup callback after setting up its switchdev notifier

[Vladimir Oltean]

On 4/3/19 12:03 AM, Andrew Lunn wrote:
> On Sun, Mar 31, 2019 at 08:42:21PM +0300, Vladimir Oltean wrote:
> > This allows the driver to perform some manipulations of its own during
> > setup, using generic code.
> > One current usage scenario is for the driver to request DSA to set up
> > 802.1Q based switch tagging for its ports.
> >
> > Signed-off-by: Vladimir Oltean <olteanv@xxxxxxxxx>
> > Reviewed-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
> > ---
> >   net/dsa/dsa2.c | 8 ++++----
> >   1 file changed, 4 insertions(+), 4 deletions(-)
> >
> > diff --git a/net/dsa/dsa2.c b/net/dsa/dsa2.c
> > index c00ee464afc7..5beceb18b7e2 100644
> > --- a/net/dsa/dsa2.c
> > +++ b/net/dsa/dsa2.c
> > @@ -360,14 +360,14 @@ static int dsa_switch_setup(struct dsa_switch *ds)
> >   	if (err)
> >   		return err;
> >   
> > -	err = ds->ops->setup(ds);
> > -	if (err < 0)
> > -		return err;
> > -
> >   	err = dsa_switch_register_notifier(ds);
> >   	if (err)
> >   		return err;
>
> It seems that notifiers are the important thing here? Maybe state that
> in the commit message?
>
> I'm also wondering how safe this is in general. If we have not yet
> called the driver setup, the switch is potentially not yet ready to
> actually handle an requests that come via the notifier. If such
> notifiers can only come from the driver itself, it should be
> safe. However, if they could come from the rest of the stack, i could
> see bad things happening.
>
>      Andrew

Hi Andrew,

Correct me if I'm wrong, but the switchdev notifiers coming from the 
stack are coming through a net device, and the slave net devices are 
only created in dsa_port_setup, which is later than the code I'm 
modifying anyway? Do you have an example of potentially racy situation 
caused by this change?

Thanks,
-Vladimir