drivers: ethernet: xmc4xxx: Fixes the checksum of bridged packets

Due to checksum offloading, the checksum field of a packet must be
zero before sending it to packet to the ethernet controller,
otherwise the computed checksum will be incorrect.

If the packet is bridged, it already has the correct checksum in
place, which will cause the automatically computed checksum to
incorrectly become zero.

In this patch we give bridged packets the special attention they
need.

Signed-off-by: Marcelo Roberto Jimenez <marcelo.jimenez@gmail.com>

Author: mrjimenez

drivers/ethernet/eth_xmc4xxx.c

@@ -228,6 +228,123 @@ static inline void eth_xmc4xxx_trigger_dma_rx(ETH_GLOBAL_TypeDef *regs)
 	regs->RECEIVE_POLL_DEMAND = 0U;
 }
 
+static void pkt_hexdump(struct net_pkt *pkt, const char *str)
+{
+	struct net_buf *buf = pkt->buffer;
+
+	while (buf) {
+		LOG_HEXDUMP_DBG(buf->data, buf->len, str);
+		buf = buf->frags;
+	}
+}
+
+/* Clears the checksum field of an ICMPv4 packet. */
+static int eth_xmc4xxx_clear_checksum_icmpv4(struct net_pkt *pkt)
+{
+	NET_PKT_DATA_ACCESS_DEFINE(icmpv4_access, struct net_icmp_hdr);
+	struct net_icmp_hdr *icmp_hdr;
+
+	if (IS_ENABLED(CONFIG_NET_IPV4_HDR_OPTIONS)) {
+		if (net_pkt_skip(pkt, net_pkt_ipv4_opts_len(pkt))) {
+			return -ENOBUFS;
+		}
+	}
+
+	icmp_hdr = (struct net_icmp_hdr *)net_pkt_get_data(pkt, &icmpv4_access);
+	if (!icmp_hdr) {
+		return -ENOBUFS;
+	}
+
+	icmp_hdr->chksum = 0;
+	net_pkt_set_chksum_done(pkt, true);
+
+	return net_pkt_set_data(pkt, &icmpv4_access);
+}
+
+/* Clears the checksum value in packets that will go through checksum
+ * offloading. */
+static int eth_xmc4xxx_clear_checksum(struct net_pkt *pkt)
+{
+	int ret = 0;
+	uint16_t p_type;
+	uint8_t next_header = -1;
+	struct net_eth_hdr *eth_hdr = NET_ETH_HDR(pkt);
+	struct net_pkt_cursor backup;
+	bool overwrite;
+	const char *msg = NULL;
+	const char *p_type_str;
+	const char *next_header_str = "<does not apply>";
+
+	overwrite = net_pkt_is_being_overwritten(pkt);
+	net_pkt_cursor_backup(pkt, &backup);
+	net_pkt_cursor_init(pkt);
+	net_pkt_set_overwrite(pkt, true);
+	eth_hdr = NET_ETH_HDR(pkt);
+	net_pkt_skip(pkt, sizeof(struct net_eth_hdr));
+	p_type = ntohs(eth_hdr->type);
+	switch (p_type) {
+	case NET_ETH_PTYPE_IP: {
+		NET_PKT_DATA_ACCESS_DEFINE(ip_access, struct net_ipv4_hdr);
+		struct net_ipv4_hdr *ip_hdr;
+
+		p_type_str = "IPv4";
+		ip_hdr = (struct net_ipv4_hdr *)net_pkt_get_data(pkt, &ip_access);
+		next_header = ip_hdr->proto;
+		net_pkt_skip(pkt, sizeof(struct net_ipv4_hdr));
+
+		break;
+	}
+	case NET_ETH_PTYPE_IPV6: {
+		NET_PKT_DATA_ACCESS_DEFINE(ip_access, struct net_ipv6_hdr);
+		struct net_ipv6_hdr *ip_hdr;
+
+		p_type_str = "IPv6";
+		ip_hdr = (struct net_ipv6_hdr *)net_pkt_get_data(pkt, &ip_access);
+		next_header = ip_hdr->nexthdr;
+		net_pkt_skip(pkt, sizeof(struct net_ipv6_hdr));
+
+		break;
+	}
+	case NET_ETH_PTYPE_ARP:
+		p_type_str = "ARP";
+		goto end;
+	default:
+		p_type_str = "Unknown";
+		msg = "Unknown protocol type";
+		goto end;
+	}
+
+	switch (next_header) {
+	case IPPROTO_TCP:
+		next_header_str = "TCP";
+		break;
+	case IPPROTO_UDP:
+		next_header_str = "UDP";
+		break;
+	case IPPROTO_ICMP:
+		next_header_str = "ICMPv4";
+		ret = eth_xmc4xxx_clear_checksum_icmpv4(pkt);
+		break;
+	case IPPROTO_ICMPV6:
+		next_header_str = "ICMPv6";
+		break;
+	default:
+		next_header_str = "Unknown";
+		msg = "Unknown next header";
+		break;
+	}
+
+end:
+	net_pkt_cursor_restore(pkt, &backup);
+	net_pkt_set_overwrite(pkt, overwrite);
+	if (msg) {
+		LOG_DBG("p_type = 0x%04x (%s), next_header = 0x%02x (%s)\n", p_type, p_type_str,
+			next_header, next_header_str);
+		pkt_hexdump(pkt, msg);
+	}
+	return ret;
+}
+
 static int eth_xmc4xxx_send(const struct device *dev, struct net_pkt *pkt)
 {
 	struct eth_xmc4xxx_data *dev_data = dev->data;
@@ -338,6 +455,19 @@ static int eth_xmc4xxx_send(const struct device *dev, struct net_pkt *pkt)
 		return -EIO;
 	}
 
+	/* Due to checksum offloading, the checksum field must be zero before
+	 * sending the packet to the ethernet controller, otherwise the computed
+	 * checksum will be incorrect.
+	 *
+	 * If the packet is bridged, it already has the correct checksum in
+	 * place, which will cause the automatically computed checksum to
+	 * incorrectly become zero.
+	 *
+	 * Here we give bridged packets the special attention they need. */
+	if (net_pkt_is_l2_bridged(pkt)) {
+		eth_xmc4xxx_clear_checksum(pkt);
+	}
+
 	unsigned int key = irq_lock();
 
 	/* label last dma descriptor as last segment and trigger interrupt on last segment */