Re: [PATCH net-next 2/6] net: dsa: vsc73xx: add port_stp_state_set function

[Vladimir Oltean]

On Thu, Jun 22, 2023 at 03:01:02PM +0200, Andrew Lunn wrote:
> On Wed, Jun 21, 2023 at 10:38:22PM +0200, Paweł Dembicki wrote:
> > śr., 21 cze 2023 o 21:33 Andrew Lunn <andrew@xxxxxxx> napisał(a):
> > >
> > > > +     struct vsc73xx *vsc = ds->priv;
> > > > +     /* FIXME: STP frames isn't forwarded at this moment. BPDU frames are
> > > > +      * forwarded only from to PI/SI interface. For more info see chapter
> > > > +      * 2.7.1 (CPU Forwarding) in datasheet.
> > >
> > > Do you mean the CPU never gets to see the BPDU frames?
> > >
> > > Does the hardware have any sort of packet matching to trap frames to
> > > the CPU? Can you match on the destination MAC address
> > > 01:80:C2:00:00:00 ?
> > >
> > 
> > Analyzer in VSC73XX switches can send some kind of packages to (and
> > from) processor via registers available from SPI/Platform BUS (for
> > some external analysis).  In some cases it's possible to configure: if
> > packet will be copied or forwarded to this special CPU queue.  But
> > BPDU frames could be sent to processor via CPU queue only. So It's
> > impossible to forward bridge control data via rgmii interface.
> 
> So am i correct in saying, if you actually enable STP, and it decides
> to block a port, the BPDUs are also blocked. After a while it will
> decide the peer has gone, and unblock the port. A broadcast storm will
> then happen for a while, until a BPDU is received, at which point it
> will block the port again.
> 
>      Andrew

This is pretty much the expected behavior from a tag_8021q based
implementation with no hardware assist for control packets. tag_8021q
can provide port identification, but it cannot transform a data packet
into a control packet and it cannot force the switch to accept packets
from ports whose data plane is disabled by STP.

I am also going to review this series in the following days, but I don't
have the required amount of time right now. Perhaps during the weekend.