Re: [PATCH net-next 2/2] net: dsa: mt7530: add support for bridge port isolation

From: Matthias Schiffer
Date: Sun Jun 16 2024 - 04:39:58 EST


On 16/06/2024 08:52, Arınç ÜNAL wrote:
On 15/06/2024 01:21, Matthias Schiffer wrote:
Remove a pair of ports from the port matrix when both ports have the
isolated flag set.

Signed-off-by: Matthias Schiffer <mschiffer@xxxxxxxxxxxxxxxxxxxx>
---
  drivers/net/dsa/mt7530.c | 21 ++++++++++++++++++---
  drivers/net/dsa/mt7530.h |  1 +
  2 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/drivers/net/dsa/mt7530.c b/drivers/net/dsa/mt7530.c
index ecacaefdd694..44939379aba8 100644
--- a/drivers/net/dsa/mt7530.c
+++ b/drivers/net/dsa/mt7530.c
@@ -1303,7 +1303,8 @@ mt7530_stp_state_set(struct dsa_switch *ds, int port, u8 state)
  }
  static void mt7530_update_port_member(struct mt7530_priv *priv, int port,
-                      const struct net_device *bridge_dev, bool join)
+                      const struct net_device *bridge_dev,
+                      bool join)

Run git clang-format on this patch as well please.

Oops, will do.



      __must_hold(&priv->reg_mutex)
  {
      struct dsa_port *dp = dsa_to_port(priv->ds, port), *other_dp;
@@ -1311,6 +1312,7 @@ static void mt7530_update_port_member(struct mt7530_priv *priv, int port,
      struct dsa_port *cpu_dp = dp->cpu_dp;
      u32 port_bitmap = BIT(cpu_dp->index);
      int other_port;
+    bool isolated;
      dsa_switch_for_each_user_port(other_dp, priv->ds) {
          other_port = other_dp->index;
@@ -1327,7 +1329,9 @@ static void mt7530_update_port_member(struct mt7530_priv *priv, int port,
          if (!dsa_port_offloads_bridge_dev(other_dp, bridge_dev))
              continue;
-        if (join) {
+        isolated = p->isolated && other_p->isolated;
+
+        if (join && !isolated) {
              other_p->pm |= PCR_MATRIX(BIT(port));
              port_bitmap |= BIT(other_port);
          } else {

Why must other_p->isolated be true as well? If I understand correctly, when
a user port is isolated, non isolated ports can't communicate with it
whilst the CPU port can. If I were to isolate a port which is the only
isolated one at the moment, the isolated flag would not be true. Therefore,
the isolated port would not be removed from the port matrix of other user
ports. Why not only check for p->isolated?

As far as I can tell, the rules are:

- non-isolated ports can communicate with every port
- isolated ports can't communicate with other isolated ports
- communication is symmetric

You'll find that the logic works the same for non-offloaded bridge forwarding (see should_deliver() in net/bridge/br_forward.c and br_skb_isolated() in net/bridge/br_private.h).

Matthias

Attachment: OpenPGP_signature.asc
Description: OpenPGP digital signature