Hi Paolo,

On 07/07/26 4:00 pm, Paolo Abeni wrote:

On 7/1/26 1:25 PM, MD Danish Anwar wrote:

quoted

Link-local multicast addresses (01:80:c2:00:00:0x) must only be
delivered to the host port (P0) and must not be forwarded out of
the physical slave ports. icssg_fdb_add_del() was programming these
addresses with P1/P2 membership bits set, causing the firmware to
forward them out of slave ports.

Clear P1/P2 membership and set only P0 membership when
is_link_local_ether_addr() returns true.

Fixes: 487f7323f39a ("net: ti: icssg-prueth: Add helper functions to configure FDB")
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
---
 drivers/net/ethernet/ti/icssg/icssg_config.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/drivers/net/ethernet/ti/icssg/icssg_config.c b/drivers/net/ethernet/ti/icssg/icssg_config.c
index 3f8237c17d099..04a81402e3f3c 100644
--- a/drivers/net/ethernet/ti/icssg/icssg_config.c
+++ b/drivers/net/ethernet/ti/icssg/icssg_config.c

@@ -732,6 +732,16 @@ int icssg_fdb_add_del(struct prueth_emac *emac, const unsigned char *addr,
 	u8 fid = vid;
 	int ret;
 
+	/* Link-local addresses (01:80:c2:00:00:0x) must only be delivered to
+	 * the host port (P0). Clear P1/P2 membership to prevent the firmware
+	 * from forwarding them out of the physical slave ports.
+	 */
+	if (is_link_local_ether_addr(addr)) {
+		fid_c2 |= ICSSG_FDB_ENTRY_P0_MEMBERSHIP;
+		fid_c2 &= ~(ICSSG_FDB_ENTRY_P1_MEMBERSHIP |
+			    ICSSG_FDB_ENTRY_P2_MEMBERSHIP);
+	}

Sashiko gemeni mentioned it could be safer to reject entirely wrong masks:

https://sashiko.dev/#/patchset/20260701112535.4027920-1-danishanwar%40ti.com

It's not clear to me if the mentioned bad scenario is actually possibly,
please have a look.

I had a look at the Sashiko comment. This seems to be a false positive
to me.

Link-local addresses (01:80:c2:00:00:0x) are IEEE 802.1D Table 7-10
reserved addresses that bridges MUST NOT forward. They are consumed
locally by STP, LACP, LLDP, PAE, etc.

No valid protocol or user configuration would add an MDB entry for these
addresses on a slave port — doing so is a misconfiguration regardless of
hardware.

The silent enforcement in icssg_fdb_add_del() ensures the hardware
always reflects the mandatory protocol behavior. Returning -EOPNOTSUPP
would only matter if there were a legitimate caller we needed to reject
— there isn't one.

I think silently adding host port to FDB membership for Link Local
addresses is OK. This Sashiko comment can be ignored.

-- 
Thanks and Regards,
Danish