[PATCH net-next v7 0/3] enic: Use Page Pool API for receiving packets

[PATCH net-next v7 0/3] enic: Use Page Pool API for receiving packets

[John Daley]

Use the Page Pool API for RX. The Page Pool API improves bandwidth and
CPU overhead by recycling pages instead of allocating new buffers in the
driver. Also, page pool fragment allocation for smaller MTUs is used
allow multiple packets to share pages.

RX code was moved to its own file and some refactoring was done
beforehand to make the page pool changes more trasparent and to simplify
the resulting code.

Signed-off-by: John Daley <johndale@cisco.com>

---
Changes in v2:
- Fixed a valid warning found by build_clang where a variable was used
  before it was initialized. The warnings in include/linux/mm.h were not
  addressed since they do not appear to be realated to this patchset.

Changes in v3:
- Moved a function to before where is was called and removed the forward
  declaration. Reworded a commit message. No functional changes.

Changes in v4:
- Replaced call to page_pool_put_page() with page_pool_put_full_page()
  since page_pool_dev_alloc() API is used and page_pool is created with
  PP_FLAG_DMA_SYNC_DEV flag.
- Reworded final commit message one more time to try to make it clear
  that there is just one fix for the commit.

Changes in v5:
- Removed link related patches from the patchset. These were merged
  seperately.
- Removed inappropriate calls to napi_free_frags()
- Moved pp_alloc_error out of ethtool stats and accumulate into netdev
  queue stat 'alloc_error'.

Changes in v6:
- Use the page pool API for all MTUs, not just <= PAGE_SIZE. Use page
  pool 'order' field to accomodate MTUs > PAGE_SIZE. Remove the
  function pointers and functions that handled the bigger MTUs.

Changes in v7:
- Add error return code handling for page_pool_create()

John Daley (3):
  enic: Move RX functions to their own file
  enic: Simplify RX handler function
  enic: Use the Page Pool API for RX

 drivers/net/ethernet/cisco/enic/Makefile    |   2 +-
 drivers/net/ethernet/cisco/enic/enic.h      |   3 +
 drivers/net/ethernet/cisco/enic/enic_main.c | 268 +++-----------------
 drivers/net/ethernet/cisco/enic/enic_rq.c   | 242 ++++++++++++++++++
 drivers/net/ethernet/cisco/enic/enic_rq.h   |  10 +
 drivers/net/ethernet/cisco/enic/vnic_rq.h   |   2 +
 6 files changed, 287 insertions(+), 240 deletions(-)
 create mode 100644 drivers/net/ethernet/cisco/enic/enic_rq.c
 create mode 100644 drivers/net/ethernet/cisco/enic/enic_rq.h

-- 
2.35.2

[PATCH net-next v7 1/3] enic: Move RX functions to their own file

[John Daley]

Move RX handler code into it's own file in preparation for
further changes.

No functional changes.

Co-developed-by: Nelson Escobar <neescoba@cisco.com>
Signed-off-by: Nelson Escobar <neescoba@cisco.com>
Co-developed-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: John Daley <johndale@cisco.com>
---
 drivers/net/ethernet/cisco/enic/Makefile    |   2 +-
 drivers/net/ethernet/cisco/enic/enic_main.c | 238 +------------------
 drivers/net/ethernet/cisco/enic/enic_rq.c   | 242 ++++++++++++++++++++
 drivers/net/ethernet/cisco/enic/enic_rq.h   |  10 +
 4 files changed, 254 insertions(+), 238 deletions(-)
 create mode 100644 drivers/net/ethernet/cisco/enic/enic_rq.c
 create mode 100644 drivers/net/ethernet/cisco/enic/enic_rq.h

diff --git a/drivers/net/ethernet/cisco/enic/Makefile b/drivers/net/ethernet/cisco/enic/Makefile
index c3b6febfdbe4..b3b5196b2dfc 100644
--- a/drivers/net/ethernet/cisco/enic/Makefile
+++ b/drivers/net/ethernet/cisco/enic/Makefile
@@ -3,5 +3,5 @@ obj-$(CONFIG_ENIC) := enic.o
 
 enic-y := enic_main.o vnic_cq.o vnic_intr.o vnic_wq.o \
 	enic_res.o enic_dev.o enic_pp.o vnic_dev.o vnic_rq.o vnic_vic.o \
-	enic_ethtool.o enic_api.o enic_clsf.o
+	enic_ethtool.o enic_api.o enic_clsf.o enic_rq.o
 
diff --git a/drivers/net/ethernet/cisco/enic/enic_main.c b/drivers/net/ethernet/cisco/enic/enic_main.c
index 49f6cab01ed5..1d9f109346b8 100644
--- a/drivers/net/ethernet/cisco/enic/enic_main.c
+++ b/drivers/net/ethernet/cisco/enic/enic_main.c
@@ -58,6 +58,7 @@
 #include "enic_dev.h"
 #include "enic_pp.h"
 #include "enic_clsf.h"
+#include "enic_rq.h"
 
 #define ENIC_NOTIFY_TIMER_PERIOD	(2 * HZ)
 #define WQ_ENET_MAX_DESC_LEN		(1 << WQ_ENET_LEN_BITS)
@@ -1313,243 +1314,6 @@ static int enic_get_vf_port(struct net_device *netdev, int vf,
 	return -EMSGSIZE;
 }
 
-static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
-{
-	struct enic *enic = vnic_dev_priv(rq->vdev);
-
-	if (!buf->os_buf)
-		return;
-
-	dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
-			 DMA_FROM_DEVICE);
-	dev_kfree_skb_any(buf->os_buf);
-	buf->os_buf = NULL;
-}
-
-static int enic_rq_alloc_buf(struct vnic_rq *rq)
-{
-	struct enic *enic = vnic_dev_priv(rq->vdev);
-	struct net_device *netdev = enic->netdev;
-	struct sk_buff *skb;
-	unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
-	unsigned int os_buf_index = 0;
-	dma_addr_t dma_addr;
-	struct vnic_rq_buf *buf = rq->to_use;
-
-	if (buf->os_buf) {
-		enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
-				   buf->len);
-
-		return 0;
-	}
-	skb = netdev_alloc_skb_ip_align(netdev, len);
-	if (!skb) {
-		enic->rq[rq->index].stats.no_skb++;
-		return -ENOMEM;
-	}
-
-	dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len,
-				  DMA_FROM_DEVICE);
-	if (unlikely(enic_dma_map_check(enic, dma_addr))) {
-		dev_kfree_skb(skb);
-		return -ENOMEM;
-	}
-
-	enic_queue_rq_desc(rq, skb, os_buf_index,
-		dma_addr, len);
-
-	return 0;
-}
-
-static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
-				      u32 pkt_len)
-{
-	if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
-		pkt_size->large_pkt_bytes_cnt += pkt_len;
-	else
-		pkt_size->small_pkt_bytes_cnt += pkt_len;
-}
-
-static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
-			     struct vnic_rq_buf *buf, u16 len)
-{
-	struct enic *enic = netdev_priv(netdev);
-	struct sk_buff *new_skb;
-
-	if (len > enic->rx_copybreak)
-		return false;
-	new_skb = netdev_alloc_skb_ip_align(netdev, len);
-	if (!new_skb)
-		return false;
-	dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len,
-				DMA_FROM_DEVICE);
-	memcpy(new_skb->data, (*skb)->data, len);
-	*skb = new_skb;
-
-	return true;
-}
-
-static void enic_rq_indicate_buf(struct vnic_rq *rq,
-	struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
-	int skipped, void *opaque)
-{
-	struct enic *enic = vnic_dev_priv(rq->vdev);
-	struct net_device *netdev = enic->netdev;
-	struct sk_buff *skb;
-	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
-	struct enic_rq_stats *rqstats = &enic->rq[rq->index].stats;
-
-	u8 type, color, eop, sop, ingress_port, vlan_stripped;
-	u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
-	u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
-	u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
-	u8 packet_error;
-	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
-	u32 rss_hash;
-	bool outer_csum_ok = true, encap = false;
-
-	rqstats->packets++;
-	if (skipped) {
-		rqstats->desc_skip++;
-		return;
-	}
-
-	skb = buf->os_buf;
-
-	cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
-		&type, &color, &q_number, &completed_index,
-		&ingress_port, &fcoe, &eop, &sop, &rss_type,
-		&csum_not_calc, &rss_hash, &bytes_written,
-		&packet_error, &vlan_stripped, &vlan_tci, &checksum,
-		&fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
-		&fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
-		&ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
-		&fcs_ok);
-
-	if (packet_error) {
-
-		if (!fcs_ok) {
-			if (bytes_written > 0)
-				rqstats->bad_fcs++;
-			else if (bytes_written == 0)
-				rqstats->pkt_truncated++;
-		}
-
-		dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
-				 DMA_FROM_DEVICE);
-		dev_kfree_skb_any(skb);
-		buf->os_buf = NULL;
-
-		return;
-	}
-
-	if (eop && bytes_written > 0) {
-
-		/* Good receive
-		 */
-		rqstats->bytes += bytes_written;
-		if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
-			buf->os_buf = NULL;
-			dma_unmap_single(&enic->pdev->dev, buf->dma_addr,
-					 buf->len, DMA_FROM_DEVICE);
-		}
-		prefetch(skb->data - NET_IP_ALIGN);
-
-		skb_put(skb, bytes_written);
-		skb->protocol = eth_type_trans(skb, netdev);
-		skb_record_rx_queue(skb, q_number);
-		if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
-		    (type == 3)) {
-			switch (rss_type) {
-			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
-			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
-			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
-				skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
-				rqstats->l4_rss_hash++;
-				break;
-			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
-			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
-			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
-				skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
-				rqstats->l3_rss_hash++;
-				break;
-			}
-		}
-		if (enic->vxlan.vxlan_udp_port_number) {
-			switch (enic->vxlan.patch_level) {
-			case 0:
-				if (fcoe) {
-					encap = true;
-					outer_csum_ok = fcoe_fc_crc_ok;
-				}
-				break;
-			case 2:
-				if ((type == 7) &&
-				    (rss_hash & BIT(0))) {
-					encap = true;
-					outer_csum_ok = (rss_hash & BIT(1)) &&
-							(rss_hash & BIT(2));
-				}
-				break;
-			}
-		}
-
-		/* Hardware does not provide whole packet checksum. It only
-		 * provides pseudo checksum. Since hw validates the packet
-		 * checksum but not provide us the checksum value. use
-		 * CHECSUM_UNNECESSARY.
-		 *
-		 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
-		 * inner csum_ok. outer_csum_ok is set by hw when outer udp
-		 * csum is correct or is zero.
-		 */
-		if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
-		    tcp_udp_csum_ok && outer_csum_ok &&
-		    (ipv4_csum_ok || ipv6)) {
-			skb->ip_summed = CHECKSUM_UNNECESSARY;
-			skb->csum_level = encap;
-			if (encap)
-				rqstats->csum_unnecessary_encap++;
-			else
-				rqstats->csum_unnecessary++;
-		}
-
-		if (vlan_stripped) {
-			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
-			rqstats->vlan_stripped++;
-		}
-		skb_mark_napi_id(skb, &enic->napi[rq->index]);
-		if (!(netdev->features & NETIF_F_GRO))
-			netif_receive_skb(skb);
-		else
-			napi_gro_receive(&enic->napi[q_number], skb);
-		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
-			enic_intr_update_pkt_size(&cq->pkt_size_counter,
-						  bytes_written);
-	} else {
-
-		/* Buffer overflow
-		 */
-		rqstats->pkt_truncated++;
-		dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
-				 DMA_FROM_DEVICE);
-		dev_kfree_skb_any(skb);
-		buf->os_buf = NULL;
-	}
-}
-
-static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
-	u8 type, u16 q_number, u16 completed_index, void *opaque)
-{
-	struct enic *enic = vnic_dev_priv(vdev);
-
-	vnic_rq_service(&enic->rq[q_number].vrq, cq_desc,
-		completed_index, VNIC_RQ_RETURN_DESC,
-		enic_rq_indicate_buf, opaque);
-
-	return 0;
-}
-
 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
 {
 	unsigned int intr = enic_msix_rq_intr(enic, rq->index);
diff --git a/drivers/net/ethernet/cisco/enic/enic_rq.c b/drivers/net/ethernet/cisco/enic/enic_rq.c
new file mode 100644
index 000000000000..e5b2f581c055
--- /dev/null
+++ b/drivers/net/ethernet/cisco/enic/enic_rq.c
@@ -0,0 +1,242 @@
+// SPDX-License-Identifier: GPL-2.0-only
+// Copyright 2024 Cisco Systems, Inc.  All rights reserved.
+
+#include <linux/skbuff.h>
+#include <linux/if_vlan.h>
+#include <net/busy_poll.h>
+#include "enic.h"
+#include "enic_res.h"
+#include "enic_rq.h"
+#include "vnic_rq.h"
+#include "cq_enet_desc.h"
+
+#define ENIC_LARGE_PKT_THRESHOLD                1000
+
+static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
+				      u32 pkt_len)
+{
+	if (pkt_len > ENIC_LARGE_PKT_THRESHOLD)
+		pkt_size->large_pkt_bytes_cnt += pkt_len;
+	else
+		pkt_size->small_pkt_bytes_cnt += pkt_len;
+}
+
+static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
+			     struct vnic_rq_buf *buf, u16 len)
+{
+	struct enic *enic = netdev_priv(netdev);
+	struct sk_buff *new_skb;
+
+	if (len > enic->rx_copybreak)
+		return false;
+	new_skb = netdev_alloc_skb_ip_align(netdev, len);
+	if (!new_skb)
+		return false;
+	dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len,
+				DMA_FROM_DEVICE);
+	memcpy(new_skb->data, (*skb)->data, len);
+	*skb = new_skb;
+
+	return true;
+}
+
+int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, u8 type,
+		    u16 q_number, u16 completed_index, void *opaque)
+{
+	struct enic *enic = vnic_dev_priv(vdev);
+
+	vnic_rq_service(&enic->rq[q_number].vrq, cq_desc, completed_index,
+			VNIC_RQ_RETURN_DESC, enic_rq_indicate_buf, opaque);
+	return 0;
+}
+
+int enic_rq_alloc_buf(struct vnic_rq *rq)
+{
+	struct enic *enic = vnic_dev_priv(rq->vdev);
+	struct net_device *netdev = enic->netdev;
+	struct sk_buff *skb;
+	unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
+	unsigned int os_buf_index = 0;
+	dma_addr_t dma_addr;
+	struct vnic_rq_buf *buf = rq->to_use;
+
+	if (buf->os_buf) {
+		enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
+				   buf->len);
+
+		return 0;
+	}
+	skb = netdev_alloc_skb_ip_align(netdev, len);
+	if (!skb) {
+		enic->rq[rq->index].stats.no_skb++;
+		return -ENOMEM;
+	}
+
+	dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len,
+				  DMA_FROM_DEVICE);
+	if (unlikely(enic_dma_map_check(enic, dma_addr))) {
+		dev_kfree_skb(skb);
+		return -ENOMEM;
+	}
+
+	enic_queue_rq_desc(rq, skb, os_buf_index, dma_addr, len);
+
+	return 0;
+}
+
+void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
+{
+	struct enic *enic = vnic_dev_priv(rq->vdev);
+
+	if (!buf->os_buf)
+		return;
+
+	dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
+			 DMA_FROM_DEVICE);
+	dev_kfree_skb_any(buf->os_buf);
+	buf->os_buf = NULL;
+}
+
+void enic_rq_indicate_buf(struct vnic_rq *rq, struct cq_desc *cq_desc,
+			  struct vnic_rq_buf *buf, int skipped, void *opaque)
+{
+	struct enic *enic = vnic_dev_priv(rq->vdev);
+	struct net_device *netdev = enic->netdev;
+	struct sk_buff *skb;
+	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
+	struct enic_rq_stats *rqstats = &enic->rq[rq->index].stats;
+
+	u8 type, color, eop, sop, ingress_port, vlan_stripped;
+	u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
+	u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
+	u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
+	u8 packet_error;
+	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
+	u32 rss_hash;
+	bool outer_csum_ok = true, encap = false;
+
+	rqstats->packets++;
+	if (skipped) {
+		rqstats->desc_skip++;
+		return;
+	}
+
+	skb = buf->os_buf;
+
+	cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc, &type, &color,
+			    &q_number, &completed_index, &ingress_port, &fcoe,
+			    &eop, &sop, &rss_type, &csum_not_calc, &rss_hash,
+			    &bytes_written, &packet_error, &vlan_stripped,
+			    &vlan_tci, &checksum, &fcoe_sof, &fcoe_fc_crc_ok,
+			    &fcoe_enc_error, &fcoe_eof, &tcp_udp_csum_ok, &udp,
+			    &tcp, &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
+			    &fcs_ok);
+
+	if (packet_error) {
+		if (!fcs_ok) {
+			if (bytes_written > 0)
+				rqstats->bad_fcs++;
+			else if (bytes_written == 0)
+				rqstats->pkt_truncated++;
+		}
+
+		dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
+				 DMA_FROM_DEVICE);
+		dev_kfree_skb_any(skb);
+		buf->os_buf = NULL;
+
+		return;
+	}
+
+	if (eop && bytes_written > 0) {
+		/* Good receive
+		 */
+		rqstats->bytes += bytes_written;
+		if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
+			buf->os_buf = NULL;
+			dma_unmap_single(&enic->pdev->dev, buf->dma_addr,
+					 buf->len, DMA_FROM_DEVICE);
+		}
+		prefetch(skb->data - NET_IP_ALIGN);
+
+		skb_put(skb, bytes_written);
+		skb->protocol = eth_type_trans(skb, netdev);
+		skb_record_rx_queue(skb, q_number);
+		if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
+		    type == 3) {
+			switch (rss_type) {
+			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
+			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
+			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
+				skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
+				rqstats->l4_rss_hash++;
+				break;
+			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
+			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
+			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
+				skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
+				rqstats->l3_rss_hash++;
+				break;
+			}
+		}
+		if (enic->vxlan.vxlan_udp_port_number) {
+			switch (enic->vxlan.patch_level) {
+			case 0:
+				if (fcoe) {
+					encap = true;
+					outer_csum_ok = fcoe_fc_crc_ok;
+				}
+				break;
+			case 2:
+				if (type == 7 &&
+				    (rss_hash & BIT(0))) {
+					encap = true;
+					outer_csum_ok = (rss_hash & BIT(1)) &&
+							(rss_hash & BIT(2));
+				}
+				break;
+			}
+		}
+
+		/* Hardware does not provide whole packet checksum. It only
+		 * provides pseudo checksum. Since hw validates the packet
+		 * checksum but not provide us the checksum value. use
+		 * CHECSUM_UNNECESSARY.
+		 *
+		 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
+		 * inner csum_ok. outer_csum_ok is set by hw when outer udp
+		 * csum is correct or is zero.
+		 */
+		if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
+		    tcp_udp_csum_ok && outer_csum_ok &&
+		    (ipv4_csum_ok || ipv6)) {
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
+			skb->csum_level = encap;
+			if (encap)
+				rqstats->csum_unnecessary_encap++;
+			else
+				rqstats->csum_unnecessary++;
+		}
+
+		if (vlan_stripped) {
+			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
+			rqstats->vlan_stripped++;
+		}
+		skb_mark_napi_id(skb, &enic->napi[rq->index]);
+		if (!(netdev->features & NETIF_F_GRO))
+			netif_receive_skb(skb);
+		else
+			napi_gro_receive(&enic->napi[q_number], skb);
+		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
+			enic_intr_update_pkt_size(&cq->pkt_size_counter,
+						  bytes_written);
+	} else {
+		/* Buffer overflow
+		 */
+		rqstats->pkt_truncated++;
+		dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
+				 DMA_FROM_DEVICE);
+		dev_kfree_skb_any(skb);
+		buf->os_buf = NULL;
+	}
+}
diff --git a/drivers/net/ethernet/cisco/enic/enic_rq.h b/drivers/net/ethernet/cisco/enic/enic_rq.h
new file mode 100644
index 000000000000..a75d07562686
--- /dev/null
+++ b/drivers/net/ethernet/cisco/enic/enic_rq.h
@@ -0,0 +1,10 @@
+/* SPDX-License-Identifier: GPL-2.0-only
+ * Copyright 2024 Cisco Systems, Inc.  All rights reserved.
+ */
+
+int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, u8 type,
+		    u16 q_number, u16 completed_index, void *opaque);
+void enic_rq_indicate_buf(struct vnic_rq *rq, struct cq_desc *cq_desc,
+			  struct vnic_rq_buf *buf, int skipped, void *opaque);
+int enic_rq_alloc_buf(struct vnic_rq *rq);
+void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf);
-- 
2.39.3

[PATCH net-next v7 2/3] enic: Simplify RX handler function

[John Daley]

Split up RX handler function in preparation for moving
to a page pool based implementation.

No functional changes.

Co-developed-by: Nelson Escobar <neescoba@cisco.com>
Signed-off-by: Nelson Escobar <neescoba@cisco.com>
Co-developed-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: John Daley <johndale@cisco.com>
---
 drivers/net/ethernet/cisco/enic/enic_rq.c | 162 +++++++++++++---------
 1 file changed, 93 insertions(+), 69 deletions(-)

diff --git a/drivers/net/ethernet/cisco/enic/enic_rq.c b/drivers/net/ethernet/cisco/enic/enic_rq.c
index e5b2f581c055..48aa385aa831 100644
--- a/drivers/net/ethernet/cisco/enic/enic_rq.c
+++ b/drivers/net/ethernet/cisco/enic/enic_rq.c
@@ -50,6 +50,94 @@ int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, u8 type,
 	return 0;
 }
 
+static void enic_rq_set_skb_flags(struct vnic_rq *vrq, u8 type, u32 rss_hash,
+				  u8 rss_type, u8 fcoe, u8 fcoe_fc_crc_ok,
+				  u8 vlan_stripped, u8 csum_not_calc,
+				  u8 tcp_udp_csum_ok, u8 ipv6, u8 ipv4_csum_ok,
+				  u16 vlan_tci, struct sk_buff *skb)
+{
+	struct enic *enic = vnic_dev_priv(vrq->vdev);
+	struct net_device *netdev = enic->netdev;
+	struct enic_rq_stats *rqstats =  &enic->rq[vrq->index].stats;
+	bool outer_csum_ok = true, encap = false;
+
+	if ((netdev->features & NETIF_F_RXHASH) && rss_hash && type == 3) {
+		switch (rss_type) {
+		case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
+		case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
+		case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
+			skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
+			rqstats->l4_rss_hash++;
+			break;
+		case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
+		case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
+		case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
+			skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
+			rqstats->l3_rss_hash++;
+			break;
+		}
+	}
+	if (enic->vxlan.vxlan_udp_port_number) {
+		switch (enic->vxlan.patch_level) {
+		case 0:
+			if (fcoe) {
+				encap = true;
+				outer_csum_ok = fcoe_fc_crc_ok;
+			}
+			break;
+		case 2:
+			if (type == 7 && (rss_hash & BIT(0))) {
+				encap = true;
+				outer_csum_ok = (rss_hash & BIT(1)) &&
+						(rss_hash & BIT(2));
+			}
+			break;
+		}
+	}
+
+	/* Hardware does not provide whole packet checksum. It only
+	 * provides pseudo checksum. Since hw validates the packet
+	 * checksum but not provide us the checksum value. use
+	 * CHECSUM_UNNECESSARY.
+	 *
+	 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
+	 * inner csum_ok. outer_csum_ok is set by hw when outer udp
+	 * csum is correct or is zero.
+	 */
+	if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
+	    tcp_udp_csum_ok && outer_csum_ok && (ipv4_csum_ok || ipv6)) {
+		skb->ip_summed = CHECKSUM_UNNECESSARY;
+		skb->csum_level = encap;
+		if (encap)
+			rqstats->csum_unnecessary_encap++;
+		else
+			rqstats->csum_unnecessary++;
+	}
+
+	if (vlan_stripped) {
+		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
+		rqstats->vlan_stripped++;
+	}
+}
+
+static bool enic_rq_pkt_error(struct vnic_rq *vrq, u8 packet_error, u8 fcs_ok,
+			      u16 bytes_written)
+{
+	struct enic *enic = vnic_dev_priv(vrq->vdev);
+	struct enic_rq_stats *rqstats = &enic->rq[vrq->index].stats;
+
+	if (packet_error) {
+		if (!fcs_ok) {
+			if (bytes_written > 0)
+				rqstats->bad_fcs++;
+			else if (bytes_written == 0)
+				rqstats->pkt_truncated++;
+		}
+		return true;
+	}
+	return false;
+}
+
 int enic_rq_alloc_buf(struct vnic_rq *rq)
 {
 	struct enic *enic = vnic_dev_priv(rq->vdev);
@@ -113,7 +201,6 @@ void enic_rq_indicate_buf(struct vnic_rq *rq, struct cq_desc *cq_desc,
 	u8 packet_error;
 	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
 	u32 rss_hash;
-	bool outer_csum_ok = true, encap = false;
 
 	rqstats->packets++;
 	if (skipped) {
@@ -132,14 +219,7 @@ void enic_rq_indicate_buf(struct vnic_rq *rq, struct cq_desc *cq_desc,
 			    &tcp, &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
 			    &fcs_ok);
 
-	if (packet_error) {
-		if (!fcs_ok) {
-			if (bytes_written > 0)
-				rqstats->bad_fcs++;
-			else if (bytes_written == 0)
-				rqstats->pkt_truncated++;
-		}
-
+	if (enic_rq_pkt_error(rq, packet_error, fcs_ok, bytes_written)) {
 		dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
 				 DMA_FROM_DEVICE);
 		dev_kfree_skb_any(skb);
@@ -162,66 +242,10 @@ void enic_rq_indicate_buf(struct vnic_rq *rq, struct cq_desc *cq_desc,
 		skb_put(skb, bytes_written);
 		skb->protocol = eth_type_trans(skb, netdev);
 		skb_record_rx_queue(skb, q_number);
-		if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
-		    type == 3) {
-			switch (rss_type) {
-			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
-			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
-			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
-				skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
-				rqstats->l4_rss_hash++;
-				break;
-			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
-			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
-			case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
-				skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
-				rqstats->l3_rss_hash++;
-				break;
-			}
-		}
-		if (enic->vxlan.vxlan_udp_port_number) {
-			switch (enic->vxlan.patch_level) {
-			case 0:
-				if (fcoe) {
-					encap = true;
-					outer_csum_ok = fcoe_fc_crc_ok;
-				}
-				break;
-			case 2:
-				if (type == 7 &&
-				    (rss_hash & BIT(0))) {
-					encap = true;
-					outer_csum_ok = (rss_hash & BIT(1)) &&
-							(rss_hash & BIT(2));
-				}
-				break;
-			}
-		}
-
-		/* Hardware does not provide whole packet checksum. It only
-		 * provides pseudo checksum. Since hw validates the packet
-		 * checksum but not provide us the checksum value. use
-		 * CHECSUM_UNNECESSARY.
-		 *
-		 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
-		 * inner csum_ok. outer_csum_ok is set by hw when outer udp
-		 * csum is correct or is zero.
-		 */
-		if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
-		    tcp_udp_csum_ok && outer_csum_ok &&
-		    (ipv4_csum_ok || ipv6)) {
-			skb->ip_summed = CHECKSUM_UNNECESSARY;
-			skb->csum_level = encap;
-			if (encap)
-				rqstats->csum_unnecessary_encap++;
-			else
-				rqstats->csum_unnecessary++;
-		}
-
-		if (vlan_stripped) {
-			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
-			rqstats->vlan_stripped++;
-		}
+		enic_rq_set_skb_flags(rq, type, rss_hash, rss_type, fcoe,
+				      fcoe_fc_crc_ok, vlan_stripped,
+				      csum_not_calc, tcp_udp_csum_ok, ipv6,
+				      ipv4_csum_ok, vlan_tci, skb);
 		skb_mark_napi_id(skb, &enic->napi[rq->index]);
 		if (!(netdev->features & NETIF_F_GRO))
 			netif_receive_skb(skb);
-- 
2.39.3

[PATCH net-next v7 3/3] enic: Use the Page Pool API for RX

[John Daley]

The Page Pool API improves bandwidth and CPU overhead by recycling pages
instead of allocating new buffers in the driver. Make use of page pool
fragment allocation for smaller MTUs so that multiple packets can share
a page.

The RQ descriptor field 'os_buf' is repurposed to hold page pointers
allocated from page_pool instead of SKBs. When packets arrive, SKBs are
allocated and the page pointers attached instead of preallocating SKBs.

'alloc_fail' netdev statistic is incremented when page_pool_dev_alloc()
fails.

Co-developed-by: Nelson Escobar <neescoba@cisco.com>
Signed-off-by: Nelson Escobar <neescoba@cisco.com>
Co-developed-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: John Daley <johndale@cisco.com>
---
 drivers/net/ethernet/cisco/enic/enic.h      |  3 +
 drivers/net/ethernet/cisco/enic/enic_main.c | 33 +++++++-
 drivers/net/ethernet/cisco/enic/enic_rq.c   | 94 ++++++++-------------
 drivers/net/ethernet/cisco/enic/vnic_rq.h   |  2 +
 4 files changed, 71 insertions(+), 61 deletions(-)

diff --git a/drivers/net/ethernet/cisco/enic/enic.h b/drivers/net/ethernet/cisco/enic/enic.h
index 10b7e02ba4d0..2ccf2d2a77db 100644
--- a/drivers/net/ethernet/cisco/enic/enic.h
+++ b/drivers/net/ethernet/cisco/enic/enic.h
@@ -17,6 +17,7 @@
 #include "vnic_nic.h"
 #include "vnic_rss.h"
 #include <linux/irq.h>
+#include <net/page_pool/helpers.h>
 
 #define DRV_NAME		"enic"
 #define DRV_DESCRIPTION		"Cisco VIC Ethernet NIC Driver"
@@ -158,6 +159,7 @@ struct enic_rq_stats {
 	u64 pkt_truncated;		/* truncated pkts */
 	u64 no_skb;			/* out of skbs */
 	u64 desc_skip;			/* Rx pkt went into later buffer */
+	u64 pp_alloc_fail;		/* page pool alloc failure */
 };
 
 struct enic_wq {
@@ -169,6 +171,7 @@ struct enic_wq {
 struct enic_rq {
 	struct vnic_rq vrq;
 	struct enic_rq_stats stats;
+	struct page_pool *pool;
 } ____cacheline_aligned;
 
 /* Per-instance private data structure */
diff --git a/drivers/net/ethernet/cisco/enic/enic_main.c b/drivers/net/ethernet/cisco/enic/enic_main.c
index 1d9f109346b8..447c54dcd89b 100644
--- a/drivers/net/ethernet/cisco/enic/enic_main.c
+++ b/drivers/net/ethernet/cisco/enic/enic_main.c
@@ -1736,6 +1736,17 @@ static int enic_open(struct net_device *netdev)
 	struct enic *enic = netdev_priv(netdev);
 	unsigned int i;
 	int err, ret;
+	unsigned int max_pkt_len = netdev->mtu + VLAN_ETH_HLEN;
+	struct page_pool_params pp_params = {
+		.order = get_order(max_pkt_len),
+		.pool_size = enic->config.rq_desc_count,
+		.nid = dev_to_node(&enic->pdev->dev),
+		.dev = &enic->pdev->dev,
+		.dma_dir = DMA_FROM_DEVICE,
+		.max_len = (max_pkt_len > PAGE_SIZE) ? max_pkt_len : PAGE_SIZE,
+		.netdev = netdev,
+		.flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV,
+	};
 
 	err = enic_request_intr(enic);
 	if (err) {
@@ -1753,6 +1764,16 @@ static int enic_open(struct net_device *netdev)
 	}
 
 	for (i = 0; i < enic->rq_count; i++) {
+		/* create a page pool for each RQ */
+		pp_params.napi = &enic->napi[i];
+		pp_params.queue_idx = i;
+		enic->rq[i].pool = page_pool_create(&pp_params);
+		if (IS_ERR(enic->rq[i].pool)) {
+			err = PTR_ERR(enic->rq[i].pool);
+			enic->rq[i].pool = NULL;
+			goto err_out_free_rq;
+		}
+
 		/* enable rq before updating rq desc */
 		vnic_rq_enable(&enic->rq[i].vrq);
 		vnic_rq_fill(&enic->rq[i].vrq, enic_rq_alloc_buf);
@@ -1793,8 +1814,11 @@ static int enic_open(struct net_device *netdev)
 err_out_free_rq:
 	for (i = 0; i < enic->rq_count; i++) {
 		ret = vnic_rq_disable(&enic->rq[i].vrq);
-		if (!ret)
+		if (!ret) {
 			vnic_rq_clean(&enic->rq[i].vrq, enic_free_rq_buf);
+			page_pool_destroy(enic->rq[i].pool);
+			enic->rq[i].pool = NULL;
+		}
 	}
 	enic_dev_notify_unset(enic);
 err_out_free_intr:
@@ -1852,8 +1876,11 @@ static int enic_stop(struct net_device *netdev)
 
 	for (i = 0; i < enic->wq_count; i++)
 		vnic_wq_clean(&enic->wq[i].vwq, enic_free_wq_buf);
-	for (i = 0; i < enic->rq_count; i++)
+	for (i = 0; i < enic->rq_count; i++) {
 		vnic_rq_clean(&enic->rq[i].vrq, enic_free_rq_buf);
+		page_pool_destroy(enic->rq[i].pool);
+		enic->rq[i].pool = NULL;
+	}
 	for (i = 0; i < enic->cq_count; i++)
 		vnic_cq_clean(&enic->cq[i]);
 	for (i = 0; i < enic->intr_count; i++)
@@ -2363,6 +2390,7 @@ static void enic_get_queue_stats_rx(struct net_device *dev, int idx,
 	rxs->hw_drop_overruns = rqstats->pkt_truncated;
 	rxs->csum_unnecessary = rqstats->csum_unnecessary +
 				rqstats->csum_unnecessary_encap;
+	rxs->alloc_fail = rqstats->pp_alloc_fail;
 }
 
 static void enic_get_queue_stats_tx(struct net_device *dev, int idx,
@@ -2390,6 +2418,7 @@ static void enic_get_base_stats(struct net_device *dev,
 	rxs->hw_drops = 0;
 	rxs->hw_drop_overruns = 0;
 	rxs->csum_unnecessary = 0;
+	rxs->alloc_fail = 0;
 	txs->bytes = 0;
 	txs->packets = 0;
 	txs->csum_none = 0;
diff --git a/drivers/net/ethernet/cisco/enic/enic_rq.c b/drivers/net/ethernet/cisco/enic/enic_rq.c
index 48aa385aa831..e3228ef7988a 100644
--- a/drivers/net/ethernet/cisco/enic/enic_rq.c
+++ b/drivers/net/ethernet/cisco/enic/enic_rq.c
@@ -21,25 +21,6 @@ static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
 		pkt_size->small_pkt_bytes_cnt += pkt_len;
 }
 
-static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
-			     struct vnic_rq_buf *buf, u16 len)
-{
-	struct enic *enic = netdev_priv(netdev);
-	struct sk_buff *new_skb;
-
-	if (len > enic->rx_copybreak)
-		return false;
-	new_skb = netdev_alloc_skb_ip_align(netdev, len);
-	if (!new_skb)
-		return false;
-	dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len,
-				DMA_FROM_DEVICE);
-	memcpy(new_skb->data, (*skb)->data, len);
-	*skb = new_skb;
-
-	return true;
-}
-
 int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, u8 type,
 		    u16 q_number, u16 completed_index, void *opaque)
 {
@@ -142,11 +123,15 @@ int enic_rq_alloc_buf(struct vnic_rq *rq)
 {
 	struct enic *enic = vnic_dev_priv(rq->vdev);
 	struct net_device *netdev = enic->netdev;
-	struct sk_buff *skb;
+	struct enic_rq *erq = &enic->rq[rq->index];
+	struct enic_rq_stats *rqstats = &erq->stats;
+	unsigned int offset = 0;
 	unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
 	unsigned int os_buf_index = 0;
 	dma_addr_t dma_addr;
 	struct vnic_rq_buf *buf = rq->to_use;
+	struct page *page;
+	unsigned int truesize = len;
 
 	if (buf->os_buf) {
 		enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
@@ -154,20 +139,16 @@ int enic_rq_alloc_buf(struct vnic_rq *rq)
 
 		return 0;
 	}
-	skb = netdev_alloc_skb_ip_align(netdev, len);
-	if (!skb) {
-		enic->rq[rq->index].stats.no_skb++;
-		return -ENOMEM;
-	}
 
-	dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len,
-				  DMA_FROM_DEVICE);
-	if (unlikely(enic_dma_map_check(enic, dma_addr))) {
-		dev_kfree_skb(skb);
+	page = page_pool_dev_alloc(erq->pool, &offset, &truesize);
+	if (unlikely(!page)) {
+		rqstats->pp_alloc_fail++;
 		return -ENOMEM;
 	}
-
-	enic_queue_rq_desc(rq, skb, os_buf_index, dma_addr, len);
+	buf->offset = offset;
+	buf->truesize = truesize;
+	dma_addr = page_pool_get_dma_addr(page) + offset;
+	enic_queue_rq_desc(rq, (void *)page, os_buf_index, dma_addr, len);
 
 	return 0;
 }
@@ -175,13 +156,12 @@ int enic_rq_alloc_buf(struct vnic_rq *rq)
 void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
 {
 	struct enic *enic = vnic_dev_priv(rq->vdev);
+	struct enic_rq *erq = &enic->rq[rq->index];
 
 	if (!buf->os_buf)
 		return;
 
-	dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
-			 DMA_FROM_DEVICE);
-	dev_kfree_skb_any(buf->os_buf);
+	page_pool_put_full_page(erq->pool, (struct page *)buf->os_buf, true);
 	buf->os_buf = NULL;
 }
 
@@ -189,10 +169,10 @@ void enic_rq_indicate_buf(struct vnic_rq *rq, struct cq_desc *cq_desc,
 			  struct vnic_rq_buf *buf, int skipped, void *opaque)
 {
 	struct enic *enic = vnic_dev_priv(rq->vdev);
-	struct net_device *netdev = enic->netdev;
 	struct sk_buff *skb;
 	struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
 	struct enic_rq_stats *rqstats = &enic->rq[rq->index].stats;
+	struct napi_struct *napi;
 
 	u8 type, color, eop, sop, ingress_port, vlan_stripped;
 	u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
@@ -208,8 +188,6 @@ void enic_rq_indicate_buf(struct vnic_rq *rq, struct cq_desc *cq_desc,
 		return;
 	}
 
-	skb = buf->os_buf;
-
 	cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc, &type, &color,
 			    &q_number, &completed_index, &ingress_port, &fcoe,
 			    &eop, &sop, &rss_type, &csum_not_calc, &rss_hash,
@@ -219,48 +197,46 @@ void enic_rq_indicate_buf(struct vnic_rq *rq, struct cq_desc *cq_desc,
 			    &tcp, &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
 			    &fcs_ok);
 
-	if (enic_rq_pkt_error(rq, packet_error, fcs_ok, bytes_written)) {
-		dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
-				 DMA_FROM_DEVICE);
-		dev_kfree_skb_any(skb);
-		buf->os_buf = NULL;
-
+	if (enic_rq_pkt_error(rq, packet_error, fcs_ok, bytes_written))
 		return;
-	}
 
 	if (eop && bytes_written > 0) {
 		/* Good receive
 		 */
 		rqstats->bytes += bytes_written;
-		if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
-			buf->os_buf = NULL;
-			dma_unmap_single(&enic->pdev->dev, buf->dma_addr,
-					 buf->len, DMA_FROM_DEVICE);
+		napi = &enic->napi[rq->index];
+		skb = napi_get_frags(napi);
+		if (unlikely(!skb)) {
+			net_warn_ratelimited("%s: skb alloc error rq[%d], desc[%d]\n",
+					     enic->netdev->name, rq->index,
+					     completed_index);
+			rqstats->no_skb++;
+			return;
 		}
+
 		prefetch(skb->data - NET_IP_ALIGN);
 
-		skb_put(skb, bytes_written);
-		skb->protocol = eth_type_trans(skb, netdev);
+		dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr,
+					bytes_written, DMA_FROM_DEVICE);
+		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
+				(struct page *)buf->os_buf, buf->offset,
+				bytes_written, buf->truesize);
 		skb_record_rx_queue(skb, q_number);
 		enic_rq_set_skb_flags(rq, type, rss_hash, rss_type, fcoe,
 				      fcoe_fc_crc_ok, vlan_stripped,
 				      csum_not_calc, tcp_udp_csum_ok, ipv6,
 				      ipv4_csum_ok, vlan_tci, skb);
-		skb_mark_napi_id(skb, &enic->napi[rq->index]);
-		if (!(netdev->features & NETIF_F_GRO))
-			netif_receive_skb(skb);
-		else
-			napi_gro_receive(&enic->napi[q_number], skb);
+		skb_mark_for_recycle(skb);
+		napi_gro_frags(napi);
 		if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
 			enic_intr_update_pkt_size(&cq->pkt_size_counter,
 						  bytes_written);
+		buf->os_buf = NULL;
+		buf->dma_addr = 0;
+		buf = buf->next;
 	} else {
 		/* Buffer overflow
 		 */
 		rqstats->pkt_truncated++;
-		dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
-				 DMA_FROM_DEVICE);
-		dev_kfree_skb_any(skb);
-		buf->os_buf = NULL;
 	}
 }
diff --git a/drivers/net/ethernet/cisco/enic/vnic_rq.h b/drivers/net/ethernet/cisco/enic/vnic_rq.h
index 0bc595abc03b..2ee4be2b9a34 100644
--- a/drivers/net/ethernet/cisco/enic/vnic_rq.h
+++ b/drivers/net/ethernet/cisco/enic/vnic_rq.h
@@ -61,6 +61,8 @@ struct vnic_rq_buf {
 	unsigned int index;
 	void *desc;
 	uint64_t wr_id;
+	unsigned int offset;
+	unsigned int truesize;
 };
 
 enum enic_poll_state {
-- 
2.39.3

Re: [PATCH net-next v7 0/3] enic: Use Page Pool API for receiving packets

[Jakub Kicinski]

On Sun, 19 Jan 2025 12:00:15 -0800 John Daley wrote:
> Use the Page Pool API for RX. The Page Pool API improves bandwidth and
> CPU overhead by recycling pages instead of allocating new buffers in the
> driver. Also, page pool fragment allocation for smaller MTUs is used
> allow multiple packets to share pages.
> 
> RX code was moved to its own file and some refactoring was done
> beforehand to make the page pool changes more trasparent and to simplify
> the resulting code.

Sorry to say, we already closed net-next for the 6.14 merge window:
https://lore.kernel.org/r/20250117182059.7ce1196f@kernel.org/

You'll have to repost in 2 weeks.