Re: [PATCH net-next] dsa: Make offloading optional on per port basis

[Jiri Pirko <jiri@resnulli.us>]

Sun, Dec 01, 2024 at 08:42:11AM CET, andrew@andrewstrohman.com wrote:
>The author has a couple use cases for this:
>
>1) Creating a sniffer, or ethernet tap, by bridging two or more
>non-offloaded ports to the bridge, and tcpdump'ing the member
>ports. Along the same lines, it would be nice to have the ability
>to temporarily disable offloading to sniff all traffic for debugging.
>
>2) Work around bugs in the hardware switch or use features
>that are only available in software.
>
>DSA drivers can be modified to remove their port_bridge_join()
>dsa_switch_ops member to accomplish this. But, it would be better
>to make it this runtime configurable, and configurable on a per port
>basis.
>
>The key to signaling that a port is not offloading is by
>ensuring dp->bridge == NULL. With this, the VLAN and FDB
>operations that affect hardware (ie port_fdb_add, port_vlan_del, etc)
>will not run. dsa_user_fdb_event() will bail if !dp->bridge.
>dsa_user_port_obj_add() checks dsa_port_offloads_bridge_port(),
>and dsa_user_host_vlan_add() checks !dp->bridge.
>
>By being configurable on a per port basis (as opposed to switch-wide),
>we can have some subset of a switch's ports offloading and others not.
>
>While this approach is generic, and therefore will be available for all
>dsa switches, I have only tested this on a mt7530 switch. It may not be
>possible or feasible to disable offloading on other switches.
>
>A flags member was added to the dsa user port netdev private data structure
>in order to facilitate adding future dsa specific flags more easily.
>IFLA_VLAN_FLAGS was used as an example when implementing the flags member.
>
>Signed-off-by: Andy Strohman <andrew@andrewstrohman.com>

Why is this DSA specific? Plus, you say you want to disable offloading
in general (DSA_FLAG_OFFLOADING_DISABLED), but you check the flag only
when joining bridge. I mean, shouldn't this be rather something exposed
by some common UAPI?

Btw, isn't NETIF_F_HW_L2FW_DOFFLOAD what you are looking for?