net: mana: Handle SKB if TX SGEs exceed hardware limit

The MANA hardware supports a maximum of 30 scatter-gather entries (SGEs)
per TX WQE. Exceeding this limit can cause TX failures.
Add ndo_features_check() callback to validate SKB layout before
transmission. For GSO SKBs that would exceed the hardware SGE limit, clear
NETIF_F_GSO_MASK to enforce software segmentation in the stack.
Add a fallback in mana_start_xmit() to linearize non-GSO SKBs that still
exceed the SGE limit.

Also, Add ethtool counter for SKBs linearized

Co-developed-by: Dipayaan Roy <dipayanroy@linux.microsoft.com>
Signed-off-by: Dipayaan Roy <dipayanroy@linux.microsoft.com>
Signed-off-by: Aditya Garg <gargaditya@linux.microsoft.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Link: https://patch.msgid.link/1763464269-10431-2-git-send-email-gargaditya@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 38 insertions, 2 deletions

diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c
index 13f47be7aca6..7b49ab005e2d 100644
--- a/drivers/net/ethernet/microsoft/mana/mana_en.c
+++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
@@ -11,6 +11,7 @@
 #include <linux/mm.h>
 #include <linux/pci.h>
 #include <linux/export.h>
+#include <linux/skbuff.h>
 
 #include <net/checksum.h>
 #include <net/ip6_checksum.h>
@@ -329,6 +330,21 @@ netdev_tx_t mana_start_xmit(struct sk_buff *skb, struct net_device *ndev)
 	cq = &apc->tx_qp[txq_idx].tx_cq;
 	tx_stats = &txq->stats;
 
+	BUILD_BUG_ON(MAX_TX_WQE_SGL_ENTRIES != MANA_MAX_TX_WQE_SGL_ENTRIES);
+	if (MAX_SKB_FRAGS + 2 > MAX_TX_WQE_SGL_ENTRIES &&
+	    skb_shinfo(skb)->nr_frags + 2 > MAX_TX_WQE_SGL_ENTRIES) {
+		/* GSO skb with Hardware SGE limit exceeded is not expected here
+		 * as they are handled in mana_features_check() callback
+		 */
+		if (skb_linearize(skb)) {
+			netdev_warn_once(ndev, "Failed to linearize skb with nr_frags=%d and is_gso=%d\n",
+					 skb_shinfo(skb)->nr_frags,
+					 skb_is_gso(skb));
+			goto tx_drop_count;
+		}
+		apc->eth_stats.tx_linear_pkt_cnt++;
+	}
+
 	pkg.tx_oob.s_oob.vcq_num = cq->gdma_id;
 	pkg.tx_oob.s_oob.vsq_frame = txq->vsq_frame;
 
@@ -442,8 +458,6 @@ netdev_tx_t mana_start_xmit(struct sk_buff *skb, struct net_device *ndev)
 		}
 	}
 
-	WARN_ON_ONCE(pkg.wqe_req.num_sge > MAX_TX_WQE_SGL_ENTRIES);
-
 	if (pkg.wqe_req.num_sge <= ARRAY_SIZE(pkg.sgl_array)) {
 		pkg.wqe_req.sgl = pkg.sgl_array;
 	} else {
@@ -518,6 +532,25 @@ tx_drop:
 	return NETDEV_TX_OK;
 }
 
+#if (MAX_SKB_FRAGS + 2 > MANA_MAX_TX_WQE_SGL_ENTRIES)
+static netdev_features_t mana_features_check(struct sk_buff *skb,
+					     struct net_device *ndev,
+					     netdev_features_t features)
+{
+	if (skb_shinfo(skb)->nr_frags + 2 > MAX_TX_WQE_SGL_ENTRIES) {
+		/* Exceeds HW SGE limit.
+		 * GSO case:
+		 *   Disable GSO so the stack will software-segment the skb
+		 *   into smaller skbs that fit the SGE budget.
+		 * Non-GSO case:
+		 *   The xmit path will attempt skb_linearize() as a fallback.
+		 */
+		features &= ~NETIF_F_GSO_MASK;
+	}
+	return features;
+}
+#endif
+
 static void mana_get_stats64(struct net_device *ndev,
 			     struct rtnl_link_stats64 *st)
 {
@@ -883,6 +916,9 @@ static const struct net_device_ops mana_devops = {
 	.ndo_open		= mana_open,
 	.ndo_stop		= mana_close,
 	.ndo_select_queue	= mana_select_queue,
+#if (MAX_SKB_FRAGS + 2 > MANA_MAX_TX_WQE_SGL_ENTRIES)
+	.ndo_features_check	= mana_features_check,
+#endif
 	.ndo_start_xmit		= mana_start_xmit,
 	.ndo_validate_addr	= eth_validate_addr,
 	.ndo_get_stats64	= mana_get_stats64,