Re: [net-next PATCH v13 13/14] net: dsa: Add Airoha AN8855 5-Port Gigabit DSA Switch driver

[Vladimir Oltean]

On Sat, Mar 15, 2025 at 04:43:53PM +0100, Christian Marangi wrote:
> +/* Similar to MT7530 also trap link local frame and special frame to CPU */
> +static int an8855_trap_special_frames(struct an8855_priv *priv)
> +{
> +	int ret;
> +
> +	/* Trap BPDUs to the CPU port(s) and egress them
> +	 * VLAN-untagged.
> +	 */
> +	ret = regmap_update_bits(priv->regmap, AN8855_BPC,
> +				 AN8855_BPDU_BPDU_FR | AN8855_BPDU_EG_TAG |
> +				 AN8855_BPDU_PORT_FW,
> +				 AN8855_BPDU_BPDU_FR |
> +				 FIELD_PREP(AN8855_BPDU_EG_TAG, AN8855_VLAN_EG_UNTAGGED) |
> +				 FIELD_PREP(AN8855_BPDU_PORT_FW, AN8855_BPDU_CPU_ONLY));
> +	if (ret)
> +		return ret;
> +
> +	/* Trap 802.1X PAE frames to the CPU port(s) and egress them
> +	 * VLAN-untagged.
> +	 */
> +	ret = regmap_update_bits(priv->regmap, AN8855_PAC,
> +				 AN8855_PAE_BPDU_FR | AN8855_PAE_EG_TAG |
> +				 AN8855_PAE_PORT_FW,
> +				 AN8855_PAE_BPDU_FR |
> +				 FIELD_PREP(AN8855_PAE_EG_TAG, AN8855_VLAN_EG_UNTAGGED) |
> +				 FIELD_PREP(AN8855_PAE_PORT_FW, AN8855_BPDU_CPU_ONLY));
> +	if (ret)
> +		return ret;
> +
> +	/* Trap frames with :01 MAC DAs to the CPU port(s) and egress
> +	 * them VLAN-untagged.
> +	 */
> +	ret = regmap_update_bits(priv->regmap, AN8855_RGAC1,
> +				 AN8855_R01_BPDU_FR | AN8855_R01_EG_TAG |
> +				 AN8855_R01_PORT_FW,
> +				 AN8855_R01_BPDU_FR |
> +				 FIELD_PREP(AN8855_R01_EG_TAG, AN8855_VLAN_EG_UNTAGGED) |
> +				 FIELD_PREP(AN8855_R01_PORT_FW, AN8855_BPDU_CPU_ONLY));
> +	if (ret)
> +		return ret;
> +
> +	/* Trap frames with :02 MAC DAs to the CPU port(s) and egress
> +	 * them VLAN-untagged.
> +	 */
> +	ret = regmap_update_bits(priv->regmap, AN8855_RGAC1,
> +				 AN8855_R02_BPDU_FR | AN8855_R02_EG_TAG |
> +				 AN8855_R02_PORT_FW,
> +				 AN8855_R02_BPDU_FR |
> +				 FIELD_PREP(AN8855_R02_EG_TAG, AN8855_VLAN_EG_UNTAGGED) |
> +				 FIELD_PREP(AN8855_R02_PORT_FW, AN8855_BPDU_CPU_ONLY));
> +	if (ret)
> +		return ret;
> +
> +	/* Trap frames with :03 MAC DAs to the CPU port(s) and egress
> +	 * them VLAN-untagged.
> +	 */
> +	ret = regmap_update_bits(priv->regmap, AN8855_RGAC1,
> +				 AN8855_R03_BPDU_FR | AN8855_R03_EG_TAG |
> +				 AN8855_R03_PORT_FW,
> +				 AN8855_R03_BPDU_FR |
> +				 FIELD_PREP(AN8855_R03_EG_TAG, AN8855_VLAN_EG_UNTAGGED) |
> +				 FIELD_PREP(AN8855_R03_PORT_FW, AN8855_BPDU_CPU_ONLY));
> +	if (ret)
> +		return ret;
> +
> +	/* Trap frames with :0E MAC DAs to the CPU port(s) and egress
> +	 * them VLAN-untagged.
> +	 */
> +	return regmap_update_bits(priv->regmap, AN8855_RGAC1,
> +				  AN8855_R0E_BPDU_FR | AN8855_R0E_EG_TAG |
> +				  AN8855_R0E_PORT_FW,
> +				  AN8855_R0E_BPDU_FR |
> +				  FIELD_PREP(AN8855_R0E_EG_TAG, AN8855_VLAN_EG_UNTAGGED) |
> +				  FIELD_PREP(AN8855_R0E_PORT_FW, AN8855_BPDU_CPU_ONLY));
> +}

Is there a way in which you could group the registers a bit more?
The function occupies 2 screens :-/

There are 4 read-modify-write operations in succession to the RGAC1
register. Maybe you can converge them into a single regmap_update_bits()
call.

Also, for packets which reach the CPU via a trap, we shouldn't set
skb->offload_fwd_mark = 1. In other words, if the bridge layer wants to
forward them in software (including to other an8855 ports), let it do so.
The common example given in commit 515853ccecc6 ("bridge: allow
forwarding some link local frames") is 802.1X PAE (01-80-C2-00-00-03).

I notice mtk_tag_rcv() calls dsa_default_offload_fwd_mark() with no
pre-condition. Do you know whether there exists any bit in the RX tag
which signifies whether the packet was received because of a trap
(or if it was autonomously forwarded by the switch to the other bridge
ports as well)? The offload_fwd_mark bit should be set based on
something like that.