10/40GbE per port runs only in NO-FEC and Force mode (ignores user configuration)
10/40
-
-
+
Copper
Up to 3
30
10/40GbE per port runs only in NO-FEC and Force mode (ignores user configuration)
25
-
+
-
Optic
Up to 100
40
25GbE per port runs only in RS-FEC and Force mode (ignores user configuration). LR modules are supported only when the device supports RS-FEC with these modules
25
-
+
-
Copper
Up to 5
40
25GbE per port runs only in RS-FEC and Force mode (ignores user configuration). LR modules are supported only when the device supports RS-FEC with these modules
50
-
+
-
Optic
Up to 100
40
50GbE per port runs only in RS-FEC and Force mode (ignores user configuration)
50
-
+
-
Copper
Up to 5
40
50GbE per port runs only in RS-FEC and Force mode (ignores user configuration)
100
-
+
-
Optic
Up to 100
40
100GbE per port runs only in RS FEC and Force mode (ignores user configuration). LR4/BiDi/SWDM4 modules are supported only when the device supports RS-FEC with these modules
100
-
+
-
Copper
Up to 100
40
100GbE per port runs only in RS FEC and Force mode (ignores user configuration). LR4/BiDi/SWDM4 modules are supported only when the device supports RS-FEC with these modules
200
-
+
-
Optic
Up to 30
60
Mellanox to Mellanox only
200
-
+
-
Copper
Up to 2
60
Mellanox to Mellanox only
Known Issues
Category
Description
Network configuration
Order of operations: Network configuration should be done bottom-up, teardown should be done top-down
ACL
ACL key size: Rules that match on L2 fields and the complete IPv6 header cannot be specified
FW automatic upgrade
Spectrum-2 in-driver firmware upgrade is not supported. See manual instructions Firmware upgrade
Multicast snooping
MDB records mapped to the same multicast MAC will share the same MDB record in the device. Deleting such records from the bridge driver will result in the following error being printed to the kernel log: "Unable to remove port from MC DB"
FRR
FRR 7.3+ installs routes using nexthop objects which are not supported by mlxsw, causing it to reject the routes. This can be fixed by the command no zebra nexthop kernel enable. See this ticket for more details
Routing
When deleting an IP neighbour from user space, the neighbour transitions to FAILED state instead of being deleted completely. The neighbour is correctly deleted from the hardware data path and not marked with offload keyword. It can be deleted completely by issuing the deletion command again