Re: [PATCH net-next v2 2/3] net: dsa: integrate with SWITCHDEV for HW bridging

[Guenter Roeck <linux@roeck-us.net>]

On Tue, Feb 24, 2015 at 01:04:48PM -0800, Florian Fainelli wrote:
> On 24/02/15 12:58, Guenter Roeck wrote:
> > On Tue, Feb 24, 2015 at 12:49:35PM -0800, Florian Fainelli wrote:
> >> On 24/02/15 12:43, Guenter Roeck wrote:
> >>> Hi Florian,
> >>>
> >>> On Tue, Feb 24, 2015 at 12:08:30PM -0800, Florian Fainelli wrote:
> >>>> In order to support bridging offloads in DSA switch drivers, select
> >>>> NET_SWITCHDEV to get access to the port_stp_update and parent_get_id
> >>>> NDOs that we are required to implement.
> >>>>
> >>>> To facilitate the integratation at the DSA driver level, we implement 3
> >>>> types of operations:
> >>>>
> >>>> - port_join_bridge
> >>>> - port_leave_bridge
> >>>> - port_stp_update
> >>>>
> >>>> DSA will resolve which switch ports that are currently bridge port
> >>>> members as some Switch hardware/drivers need to know about that to limit
> >>>> the register programming to just the relevant registers (especially for
> >>>> slow MDIO buses).
> >>>>
> >>>> We also take care of setting the correct STP state when slave network
> >>>> devices are brought up/down while being bridge members.
> >>>>
> >>>> Finally, when a port is leaving the bridge, we make sure we set in
> >>>> BR_STATE_FORWARDING state, otherwise the bridge layer would leave it
> >>>> disabled as a result of having left the bridge.
> >>>>
> >>>> Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
> >>>> ---
> >>> [ ... ]
> >>>
> >>>>  
> >>>> +/* Return a bitmask of all ports being currently bridged within a given bridge
> >>>> + * device. Note that on leave, the mask will still return the bitmask of ports
> >>>> + * currently bridged, prior to port removal, and this is exactly what we want.
> >>>> + */
> >>>> +static u32 dsa_slave_br_port_mask(struct dsa_switch *ds,
> >>>> +				  struct net_device *bridge)
> >>>> +{
> >>>> +	struct dsa_slave_priv *p;
> >>>> +	unsigned int port;
> >>>> +	u32 mask = 0;
> >>>> +
> >>>> +	for (port = 0; port < DSA_MAX_PORTS; port++) {
> >>>> +		if (!((1 << port) & ds->phys_port_mask))
> >>>> +			continue;
> >>>> +
> >>>> +		p = netdev_priv(ds->ports[port]);
> >>>> +
> >>>
> >>> ds->ports[port] can still be NULL here since the function can be called
> >>> before all ports are initialized (which is what I had actually seen).
> >>
> >> Can it now? I re-ordered the patches such that your change comes first
> >> (and I forgot to update the commit message) and by the time we get
> >> called from register_netdev(), ds->ports[port] has already been assigned.
> >>
> >> Am I missing something here?
> > 
> > Yes, because phys_port_mask is set to include _all_ ports, not just
> > the ones already registered.
> > 
> > Assume ports 0..2 have been registered, phys_port_mask is 0x1f, and
> > dsa_slave_br_port_mask is called for a state change on port 0.
> > The loop will check ports 3 and 4 which have not yet been registered.
> 
> Ok, I re-added the check.
> 
> > 
> > Strictly speaking we might want to consider adding the same check
> > into the suspend and resume functions, at least if suspend  /remove
> > can ever happen before the system is fully initialized.
> 
> Yes, that is possibly a problem, I don't think I could reproduce that
> with my current setup because everything needs to be built into the kernel.

Does suspend/resume have anything to do with building code into the kernel ?

There may be a more practical problem trying to reproduce this; you would
have to somehow manage to suspend the system in the maybe one second
or less where phys_port_mask is already initialized but not the slave ports.

I'll submit a patch to address that problem, just to be sure, following the
logic of better safe than sorry.

Would it make sense to introduce a macro such as ds_port_initialized(ds, port)
or ds_port_configured(ds, port) to check for this condition ? I see that it is
used it in the bcm_sf2 driver's suspend/remove functions as well.

Guenter