Re: [RFC PATCH net-next 01/11] net: dsa: microchip: lan937x: add cascade tailtag

[Vladimir Oltean <olteanv@gmail.com>]

On Thu, Feb 02, 2023 at 06:29:20PM +0530, Rakesh Sankaranarayanan wrote:
> cascade tailtag contains 3 bytes of information, it includes
> additional bytes for accomodating port number in second switch.
> Destination port bitmap on first switch is at bit position 7:0 and
> of second switch is at bit position 15:8, add new tailtag xmit and
> rcv functions for cascade with proper formatting. Add new tag protocol
> for cascading and link with new xmit and rcv functions.
> 
> Signed-off-by: Rakesh Sankaranarayanan <rakesh.sankaranarayanan@microchip.com>
> ---
>  include/net/dsa.h |  2 ++
>  net/dsa/tag_ksz.c | 80 ++++++++++++++++++++++++++++++++++++++++++++---
>  2 files changed, 77 insertions(+), 5 deletions(-)
> 
> diff --git a/include/net/dsa.h b/include/net/dsa.h
> index a15f17a38eca..55651ad29193 100644
> --- a/include/net/dsa.h
> +++ b/include/net/dsa.h
> @@ -390,6 +399,7 @@ MODULE_ALIAS_DSA_TAG_DRIVER(DSA_TAG_PROTO_KSZ9893, KSZ9893_NAME);
>   *	  (eg, 0x00=port1, 0x02=port3, 0x07=port8)
>   */
>  #define LAN937X_EGRESS_TAG_LEN		2
> +#define LAN937X_CASCADE_TAG_LEN		3
>  
>  #define LAN937X_TAIL_TAG_BLOCKING_OVERRIDE	BIT(11)
>  #define LAN937X_TAIL_TAG_LOOKUP			BIT(12)
> @@ -442,11 +452,71 @@ static const struct dsa_device_ops lan937x_netdev_ops = {
>  DSA_TAG_DRIVER(lan937x_netdev_ops);
>  MODULE_ALIAS_DSA_TAG_DRIVER(DSA_TAG_PROTO_LAN937X, LAN937X_NAME);
>  
> +/* For xmit, 3/7 bytes are added before FCS.
> + * ---------------------------------------------------------------------------
> + * DA(6bytes)|SA(6bytes)|....|Data(nbytes)|ts(4bytes)|tag0(1byte)|tag1(1byte)|
> + * tag2(1byte)|FCS(4bytes)
> + * ---------------------------------------------------------------------------
> + * ts   : time stamp (Present only if PTP is enabled in the Hardware)
> + * tag0 : represents tag override, lookup and valid
> + * tag1 : each bit represents destination port map through switch 2
> + *	  (eg, 0x01=port1, 0x02=port2, 0x80=port8)
> + * tag2 : each bit represents destination port map through switch 1
> + *	  (eg, 0x01=port1, 0x02=port2, 0x80=port8)
> + *
> + * For rcv, 1/5 bytes is added before FCS.

Plural is only used for more than 5 bytes?
"5 bytes is added" vs "7 bytes are added"

> + * ---------------------------------------------------------------------------
> + * DA(6bytes)|SA(6bytes)|....|Data(nbytes)|ts(4bytes)|tag0(1byte)|FCS(4bytes)
> + * ---------------------------------------------------------------------------
> + * ts   : time stamp (Present only if bit 7 of tag0 is set)
> + * tag0 : zero-based value represents port
> + *	  (eg, 0x00=port1, 0x02=port3, 0x07=port8)
> + */
> +static struct sk_buff *lan937x_cascade_xmit(struct sk_buff *skb,
> +					    struct net_device *dev)
> +{
> +	struct dsa_port *dp = dsa_slave_to_port(dev);
> +	const struct ethhdr *hdr = eth_hdr(skb);
> +	__be32 *tag;
> +	u32 val;
> +
> +	if (skb->ip_summed == CHECKSUM_PARTIAL && skb_checksum_help(skb))
> +		return NULL;
> +
> +	tag = skb_put(skb, LAN937X_CASCADE_TAG_LEN);
> +
> +	val |= BIT((dp->index + (8 * dp->ds->index)));

This only works because cascade port is port 0, right?

> +
> +	if (is_link_local_ether_addr(hdr->h_dest))
> +		val |= (LAN937X_TAIL_TAG_BLOCKING_OVERRIDE << 8);
> +
> +	val |= (LAN937X_TAIL_TAG_VALID << 8);
> +
> +	put_unaligned_be24(val, tag);
> +
> +	return skb;
> +}
> +
> +static const struct dsa_device_ops lan937x_cascade_netdev_ops = {
> +	.name   = LAN937X_CASCADE_NAME,
> +	.proto  = DSA_TAG_PROTO_LAN937X_CASCADE,
> +	.xmit   = lan937x_cascade_xmit,
> +	.rcv    = ksz9477_rcv,
> +	.connect = ksz_connect,
> +	.disconnect = ksz_disconnect,
> +	.needed_tailroom = LAN937X_CASCADE_TAG_LEN,
> +};

Nope, no new lan937x_cascade tagging protocol.

If you read Documentation/networking/dsa/dsa.rst, it says:

| From the perspective of the network stack, all switches within the same DSA
| switch tree use the same tagging protocol.

Unless you are prepared to remove this assumption from the DSA framework,
you need to fold the cascade tag handling into the regular lan937x
tagging protocol (and declare the larger needed_tailroom to cover the
tail tag case).

You can look at dp->ds->index when figuring out the length of the tail
tag that should be inserted.

There's a lot of consolidation that could (and should) be done first
between lan937x_xmit() and ksz9477_xmit(), prior to adding the support
for cascade tagging.

> +
> +DSA_TAG_DRIVER(lan937x_cascade_netdev_ops);
> +MODULE_ALIAS_DSA_TAG_DRIVER(DSA_TAG_PROTO_LAN937x_CASCADE,
> +			    LAN937X_CASCADE_NAME);
> +
>  static struct dsa_tag_driver *dsa_tag_driver_array[] = {
>  	&DSA_TAG_DRIVER_NAME(ksz8795_netdev_ops),
>  	&DSA_TAG_DRIVER_NAME(ksz9477_netdev_ops),
>  	&DSA_TAG_DRIVER_NAME(ksz9893_netdev_ops),
>  	&DSA_TAG_DRIVER_NAME(lan937x_netdev_ops),
> +	&DSA_TAG_DRIVER_NAME(lan937x_cascade_netdev_ops),
>  };
>  
>  module_dsa_tag_drivers(dsa_tag_driver_array);
> -- 
> 2.34.1
>