[PATCH net-next 0/5] net: ethernet: ti: am65-cpsw: drop multiple functions and code cleanup

[PATCH net-next 0/5] net: ethernet: ti: am65-cpsw: drop multiple functions and code cleanup

[Roger Quadros]

We have 2 tx completion functions to handle single-port vs multi-port
variants. Merge them into one function to make maintenance easier.

We also have 2 functions to handle TX completion for SKB vs XDP.
Get rid of them too.

Also do some minor cleanups.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
---
Roger Quadros (5):
      net: ethernet: ti: am65-cpsw: remove am65_cpsw_nuss_tx_compl_packets_2g()
      net: ethernet: ti: am65_cpsw: remove cpu argument am65_cpsw_run_xdp
      net: ethernet: ti: am65-cpsw: use return instead of goto in am65_cpsw_run_xdp()
      net: ethernet: ti: am65_cpsw: move am65_cpsw_put_page() out of am65_cpsw_run_xdp()
      net: ethernet: ti am65_cpsw: Drop separate TX completion functions

 drivers/net/ethernet/ti/am65-cpsw-nuss.c | 211 +++++++++----------------------
 drivers/net/ethernet/ti/am65-cpsw-nuss.h |   8 ++
 2 files changed, 71 insertions(+), 148 deletions(-)
---
base-commit: 489b31c6cf29fff5b1f5c48625770f45cea90ada
change-id: 20250214-am65-cpsw-zc-prep-038276579e73

Best regards,
-- 
Roger Quadros <rogerq@kernel.org>

[PATCH net-next 1/5] net: ethernet: ti: am65-cpsw: remove am65_cpsw_nuss_tx_compl_packets_2g()

[Roger Quadros]

The only difference between am65_cpsw_nuss_tx_compl_packets_2g() and
am65_cpsw_nuss_tx_compl_packets() is the usage of spin_lock() and
netdev_tx_completed_queue() + am65_cpsw_nuss_tx_wake at every packet
in the latter.

Insted of having 2 separate functions for TX completion, merge them
into one. This will reduce code duplication and make maintenance easier.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
---
 drivers/net/ethernet/ti/am65-cpsw-nuss.c | 98 ++++++++------------------------
 1 file changed, 25 insertions(+), 73 deletions(-)

diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
index a2b6a30918f6..75b402132e3f 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
@@ -1502,6 +1502,7 @@ static void am65_cpsw_nuss_tx_wake(struct am65_cpsw_tx_chn *tx_chn, struct net_d
 static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 					   int chn, unsigned int budget, bool *tdown)
 {
+	bool single_port = AM65_CPSW_IS_CPSW2G(common);
 	enum am65_cpsw_tx_buf_type buf_type;
 	struct device *dev = common->dev;
 	struct am65_cpsw_tx_chn *tx_chn;
@@ -1509,6 +1510,7 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 	unsigned int total_bytes = 0;
 	struct net_device *ndev;
 	struct xdp_frame *xdpf;
+	unsigned int pkt_len;
 	struct sk_buff *skb;
 	dma_addr_t desc_dma;
 	int res, num_tx = 0;
@@ -1516,9 +1518,12 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 	tx_chn = &common->tx_chns[chn];
 
 	while (true) {
-		spin_lock(&tx_chn->lock);
+		if (!single_port)
+			spin_lock(&tx_chn->lock);
 		res = k3_udma_glue_pop_tx_chn(tx_chn->tx_chn, &desc_dma);
-		spin_unlock(&tx_chn->lock);
+		if (!single_port)
+			spin_unlock(&tx_chn->lock);
+
 		if (res == -ENODATA)
 			break;
 
@@ -1533,23 +1538,35 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 		if (buf_type == AM65_CPSW_TX_BUF_TYPE_SKB) {
 			skb = am65_cpsw_nuss_tx_compl_packet_skb(tx_chn, desc_dma);
 			ndev = skb->dev;
-			total_bytes = skb->len;
+			pkt_len = skb->len;
 			napi_consume_skb(skb, budget);
 		} else {
 			xdpf = am65_cpsw_nuss_tx_compl_packet_xdp(common, tx_chn,
 								  desc_dma, &ndev);
-			total_bytes = xdpf->len;
+			pkt_len = xdpf->len;
 			if (buf_type == AM65_CPSW_TX_BUF_TYPE_XDP_TX)
 				xdp_return_frame_rx_napi(xdpf);
 			else
 				xdp_return_frame(xdpf);
 		}
+
+		total_bytes += pkt_len;
 		num_tx++;
 
-		netif_txq = netdev_get_tx_queue(ndev, chn);
+		if (!single_port) {
+			/* as packets from multi ports can be interleaved
+			 * on the same channel, we have to figure out the
+			 * port/queue at every packet and report it/wake queue.
+			 */
+			netif_txq = netdev_get_tx_queue(ndev, chn);
+			netdev_tx_completed_queue(netif_txq, 1, pkt_len);
+			am65_cpsw_nuss_tx_wake(tx_chn, ndev, netif_txq);
+		}
+	}
 
+	if (single_port) {
+		netif_txq = netdev_get_tx_queue(ndev, chn);
 		netdev_tx_completed_queue(netif_txq, num_tx, total_bytes);
-
 		am65_cpsw_nuss_tx_wake(tx_chn, ndev, netif_txq);
 	}
 
@@ -1558,66 +1575,6 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 	return num_tx;
 }
 
-static int am65_cpsw_nuss_tx_compl_packets_2g(struct am65_cpsw_common *common,
-					      int chn, unsigned int budget, bool *tdown)
-{
-	enum am65_cpsw_tx_buf_type buf_type;
-	struct device *dev = common->dev;
-	struct am65_cpsw_tx_chn *tx_chn;
-	struct netdev_queue *netif_txq;
-	unsigned int total_bytes = 0;
-	struct net_device *ndev;
-	struct xdp_frame *xdpf;
-	struct sk_buff *skb;
-	dma_addr_t desc_dma;
-	int res, num_tx = 0;
-
-	tx_chn = &common->tx_chns[chn];
-
-	while (true) {
-		res = k3_udma_glue_pop_tx_chn(tx_chn->tx_chn, &desc_dma);
-		if (res == -ENODATA)
-			break;
-
-		if (cppi5_desc_is_tdcm(desc_dma)) {
-			if (atomic_dec_and_test(&common->tdown_cnt))
-				complete(&common->tdown_complete);
-			*tdown = true;
-			break;
-		}
-
-		buf_type = am65_cpsw_nuss_buf_type(tx_chn, desc_dma);
-		if (buf_type == AM65_CPSW_TX_BUF_TYPE_SKB) {
-			skb = am65_cpsw_nuss_tx_compl_packet_skb(tx_chn, desc_dma);
-			ndev = skb->dev;
-			total_bytes += skb->len;
-			napi_consume_skb(skb, budget);
-		} else {
-			xdpf = am65_cpsw_nuss_tx_compl_packet_xdp(common, tx_chn,
-								  desc_dma, &ndev);
-			total_bytes += xdpf->len;
-			if (buf_type == AM65_CPSW_TX_BUF_TYPE_XDP_TX)
-				xdp_return_frame_rx_napi(xdpf);
-			else
-				xdp_return_frame(xdpf);
-		}
-		num_tx++;
-	}
-
-	if (!num_tx)
-		return 0;
-
-	netif_txq = netdev_get_tx_queue(ndev, chn);
-
-	netdev_tx_completed_queue(netif_txq, num_tx, total_bytes);
-
-	am65_cpsw_nuss_tx_wake(tx_chn, ndev, netif_txq);
-
-	dev_dbg(dev, "%s:%u pkt:%d\n", __func__, chn, num_tx);
-
-	return num_tx;
-}
-
 static enum hrtimer_restart am65_cpsw_nuss_tx_timer_callback(struct hrtimer *timer)
 {
 	struct am65_cpsw_tx_chn *tx_chns =
@@ -1633,13 +1590,8 @@ static int am65_cpsw_nuss_tx_poll(struct napi_struct *napi_tx, int budget)
 	bool tdown = false;
 	int num_tx;
 
-	if (AM65_CPSW_IS_CPSW2G(tx_chn->common))
-		num_tx = am65_cpsw_nuss_tx_compl_packets_2g(tx_chn->common, tx_chn->id,
-							    budget, &tdown);
-	else
-		num_tx = am65_cpsw_nuss_tx_compl_packets(tx_chn->common,
-							 tx_chn->id, budget, &tdown);
-
+	num_tx = am65_cpsw_nuss_tx_compl_packets(tx_chn->common,
+						 tx_chn->id, budget, &tdown);
 	if (num_tx >= budget)
 		return budget;
 

-- 
2.34.1

[PATCH net-next 2/5] net: ethernet: ti: am65_cpsw: remove cpu argument am65_cpsw_run_xdp

[Roger Quadros]

am65_cpsw_run_xdp() can figure out the cpu id itself.
No need to pass it around 2 functions so drop it.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
---
 drivers/net/ethernet/ti/am65-cpsw-nuss.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
index 75b402132e3f..6bd4f526cb8e 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
@@ -1167,14 +1167,14 @@ static int am65_cpsw_xdp_tx_frame(struct net_device *ndev,
 
 static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 			     struct am65_cpsw_port *port,
-			     struct xdp_buff *xdp,
-			     int cpu, int *len)
+			     struct xdp_buff *xdp, int *len)
 {
 	struct am65_cpsw_common *common = flow->common;
 	struct net_device *ndev = port->ndev;
 	int ret = AM65_CPSW_XDP_CONSUMED;
 	struct am65_cpsw_tx_chn *tx_chn;
 	struct netdev_queue *netif_txq;
+	int cpu = smp_processor_id();
 	struct xdp_frame *xdpf;
 	struct bpf_prog *prog;
 	struct page *page;
@@ -1274,7 +1274,7 @@ static void am65_cpsw_nuss_rx_csum(struct sk_buff *skb, u32 csum_info)
 }
 
 static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_rx_flow *flow,
-				     int cpu, int *xdp_state)
+				     int *xdp_state)
 {
 	struct am65_cpsw_rx_chn *rx_chn = &flow->common->rx_chns;
 	u32 buf_dma_len, pkt_len, port_id = 0, csum_info;
@@ -1334,8 +1334,7 @@ static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_rx_flow *flow,
 		xdp_init_buff(&xdp, PAGE_SIZE, &port->xdp_rxq[flow->id]);
 		xdp_prepare_buff(&xdp, page_addr, AM65_CPSW_HEADROOM,
 				 pkt_len, false);
-		*xdp_state = am65_cpsw_run_xdp(flow, port, &xdp,
-					       cpu, &pkt_len);
+		*xdp_state = am65_cpsw_run_xdp(flow, port, &xdp, &pkt_len);
 		if (*xdp_state != AM65_CPSW_XDP_PASS)
 			goto allocate;
 
@@ -1399,7 +1398,6 @@ static int am65_cpsw_nuss_rx_poll(struct napi_struct *napi_rx, int budget)
 {
 	struct am65_cpsw_rx_flow *flow = am65_cpsw_napi_to_rx_flow(napi_rx);
 	struct am65_cpsw_common *common = flow->common;
-	int cpu = smp_processor_id();
 	int xdp_state_or = 0;
 	int cur_budget, ret;
 	int xdp_state;
@@ -1408,7 +1406,7 @@ static int am65_cpsw_nuss_rx_poll(struct napi_struct *napi_rx, int budget)
 	/* process only this flow */
 	cur_budget = budget;
 	while (cur_budget--) {
-		ret = am65_cpsw_nuss_rx_packets(flow, cpu, &xdp_state);
+		ret = am65_cpsw_nuss_rx_packets(flow, &xdp_state);
 		xdp_state_or |= xdp_state;
 		if (ret)
 			break;

-- 
2.34.1

[PATCH net-next 3/5] net: ethernet: ti: am65-cpsw: use return instead of goto in am65_cpsw_run_xdp()

[Roger Quadros]

In am65_cpsw_run_xdp() instead of goto followed by return, simply return.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
---
 drivers/net/ethernet/ti/am65-cpsw-nuss.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
index 6bd4f526cb8e..468fddd0afd2 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
@@ -1193,8 +1193,7 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 
 	switch (act) {
 	case XDP_PASS:
-		ret = AM65_CPSW_XDP_PASS;
-		goto out;
+		return AM65_CPSW_XDP_PASS;
 	case XDP_TX:
 		tx_chn = &common->tx_chns[cpu % AM65_CPSW_MAX_QUEUES];
 		netif_txq = netdev_get_tx_queue(ndev, tx_chn->id);
@@ -1213,15 +1212,13 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 			goto drop;
 
 		dev_sw_netstats_rx_add(ndev, pkt_len);
-		ret = AM65_CPSW_XDP_CONSUMED;
-		goto out;
+		return AM65_CPSW_XDP_CONSUMED;
 	case XDP_REDIRECT:
 		if (unlikely(xdp_do_redirect(ndev, xdp, prog)))
 			goto drop;
 
 		dev_sw_netstats_rx_add(ndev, pkt_len);
-		ret = AM65_CPSW_XDP_REDIRECT;
-		goto out;
+		return AM65_CPSW_XDP_REDIRECT;
 	default:
 		bpf_warn_invalid_xdp_action(ndev, prog, act);
 		fallthrough;
@@ -1236,7 +1233,6 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 	page = virt_to_head_page(xdp->data);
 	am65_cpsw_put_page(flow, page, true);
 
-out:
 	return ret;
 }
 

-- 
2.34.1

[PATCH net-next 4/5] net: ethernet: ti: am65_cpsw: move am65_cpsw_put_page() out of am65_cpsw_run_xdp()

[Roger Quadros]

This allows us to re-use am65_cpsw_run_xdp() for zero copy
case. Add AM65_CPSW_XDP_TX case for successful XDP_TX so we don't
free the page while in flight.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
---
 drivers/net/ethernet/ti/am65-cpsw-nuss.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
index 468fddd0afd2..32349cc58e2e 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
@@ -164,6 +164,7 @@
 #define AM65_CPSW_CPPI_TX_PKT_TYPE 0x7
 
 /* XDP */
+#define AM65_CPSW_XDP_TX       BIT(2)
 #define AM65_CPSW_XDP_CONSUMED BIT(1)
 #define AM65_CPSW_XDP_REDIRECT BIT(0)
 #define AM65_CPSW_XDP_PASS     0
@@ -1177,7 +1178,6 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 	int cpu = smp_processor_id();
 	struct xdp_frame *xdpf;
 	struct bpf_prog *prog;
-	struct page *page;
 	int pkt_len;
 	u32 act;
 	int err;
@@ -1212,7 +1212,7 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 			goto drop;
 
 		dev_sw_netstats_rx_add(ndev, pkt_len);
-		return AM65_CPSW_XDP_CONSUMED;
+		return AM65_CPSW_XDP_TX;
 	case XDP_REDIRECT:
 		if (unlikely(xdp_do_redirect(ndev, xdp, prog)))
 			goto drop;
@@ -1230,9 +1230,6 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 		ndev->stats.rx_dropped++;
 	}
 
-	page = virt_to_head_page(xdp->data);
-	am65_cpsw_put_page(flow, page, true);
-
 	return ret;
 }
 
@@ -1331,6 +1328,12 @@ static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_rx_flow *flow,
 		xdp_prepare_buff(&xdp, page_addr, AM65_CPSW_HEADROOM,
 				 pkt_len, false);
 		*xdp_state = am65_cpsw_run_xdp(flow, port, &xdp, &pkt_len);
+		if (*xdp_state == AM65_CPSW_XDP_CONSUMED) {
+			page = virt_to_head_page(xdp.data);
+			am65_cpsw_put_page(flow, page, true);
+			goto allocate;
+		}
+
 		if (*xdp_state != AM65_CPSW_XDP_PASS)
 			goto allocate;
 

-- 
2.34.1

[PATCH net-next 5/5] net: ethernet: ti am65_cpsw: Drop separate TX completion functions

[Roger Quadros]

Drop separate TX completion functions for SKB and XDP. To do that
use the SW_DATA mechanism to store ndev and skb/xdpf for TX packets.

Use BUILD_BUG_ON_MSG() to fail build if SW_DATA size exceeds whats
available. i.e. AM65_CPSW_NAV_SW_DATA_SIZE.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
---
 drivers/net/ethernet/ti/am65-cpsw-nuss.c | 82 ++++++++++----------------------
 drivers/net/ethernet/ti/am65-cpsw-nuss.h |  8 ++++
 2 files changed, 32 insertions(+), 58 deletions(-)

diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
index 32349cc58e2e..213ec2cbe436 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
@@ -830,19 +830,19 @@ static void am65_cpsw_nuss_tx_cleanup(void *data, dma_addr_t desc_dma)
 {
 	struct am65_cpsw_tx_chn *tx_chn = data;
 	enum am65_cpsw_tx_buf_type buf_type;
+	struct am65_cpsw_tx_swdata *swdata;
 	struct cppi5_host_desc_t *desc_tx;
 	struct xdp_frame *xdpf;
 	struct sk_buff *skb;
-	void **swdata;
 
 	desc_tx = k3_cppi_desc_pool_dma2virt(tx_chn->desc_pool, desc_dma);
 	swdata = cppi5_hdesc_get_swdata(desc_tx);
 	buf_type = am65_cpsw_nuss_buf_type(tx_chn, desc_dma);
 	if (buf_type == AM65_CPSW_TX_BUF_TYPE_SKB) {
-		skb = *(swdata);
+		skb = swdata->skb;
 		dev_kfree_skb_any(skb);
 	} else {
-		xdpf = *(swdata);
+		xdpf = swdata->xdpf;
 		xdp_return_frame(xdpf);
 	}
 
@@ -1099,10 +1099,10 @@ static int am65_cpsw_xdp_tx_frame(struct net_device *ndev,
 	struct am65_cpsw_common *common = am65_ndev_to_common(ndev);
 	struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
 	struct cppi5_host_desc_t *host_desc;
+	struct am65_cpsw_tx_swdata *swdata;
 	struct netdev_queue *netif_txq;
 	dma_addr_t dma_desc, dma_buf;
 	u32 pkt_len = xdpf->len;
-	void **swdata;
 	int ret;
 
 	host_desc = k3_cppi_desc_pool_alloc(tx_chn->desc_pool);
@@ -1132,7 +1132,8 @@ static int am65_cpsw_xdp_tx_frame(struct net_device *ndev,
 	cppi5_hdesc_attach_buf(host_desc, dma_buf, pkt_len, dma_buf, pkt_len);
 
 	swdata = cppi5_hdesc_get_swdata(host_desc);
-	*(swdata) = xdpf;
+	swdata->ndev = ndev;
+	swdata->xdpf = xdpf;
 
 	/* Report BQL before sending the packet */
 	netif_txq = netdev_get_tx_queue(ndev, tx_chn->id);
@@ -1433,52 +1434,6 @@ static int am65_cpsw_nuss_rx_poll(struct napi_struct *napi_rx, int budget)
 	return num_rx;
 }
 
-static struct sk_buff *
-am65_cpsw_nuss_tx_compl_packet_skb(struct am65_cpsw_tx_chn *tx_chn,
-				   dma_addr_t desc_dma)
-{
-	struct cppi5_host_desc_t *desc_tx;
-	struct sk_buff *skb;
-	void **swdata;
-
-	desc_tx = k3_cppi_desc_pool_dma2virt(tx_chn->desc_pool,
-					     desc_dma);
-	swdata = cppi5_hdesc_get_swdata(desc_tx);
-	skb = *(swdata);
-	am65_cpsw_nuss_xmit_free(tx_chn, desc_tx);
-
-	am65_cpts_tx_timestamp(tx_chn->common->cpts, skb);
-
-	dev_sw_netstats_tx_add(skb->dev, 1, skb->len);
-
-	return skb;
-}
-
-static struct xdp_frame *
-am65_cpsw_nuss_tx_compl_packet_xdp(struct am65_cpsw_common *common,
-				   struct am65_cpsw_tx_chn *tx_chn,
-				   dma_addr_t desc_dma,
-				   struct net_device **ndev)
-{
-	struct cppi5_host_desc_t *desc_tx;
-	struct am65_cpsw_port *port;
-	struct xdp_frame *xdpf;
-	u32 port_id = 0;
-	void **swdata;
-
-	desc_tx = k3_cppi_desc_pool_dma2virt(tx_chn->desc_pool, desc_dma);
-	cppi5_desc_get_tags_ids(&desc_tx->hdr, NULL, &port_id);
-	swdata = cppi5_hdesc_get_swdata(desc_tx);
-	xdpf = *(swdata);
-	am65_cpsw_nuss_xmit_free(tx_chn, desc_tx);
-
-	port = am65_common_get_port(common, port_id);
-	dev_sw_netstats_tx_add(port->ndev, 1, xdpf->len);
-	*ndev = port->ndev;
-
-	return xdpf;
-}
-
 static void am65_cpsw_nuss_tx_wake(struct am65_cpsw_tx_chn *tx_chn, struct net_device *ndev,
 				   struct netdev_queue *netif_txq)
 {
@@ -1501,6 +1456,8 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 {
 	bool single_port = AM65_CPSW_IS_CPSW2G(common);
 	enum am65_cpsw_tx_buf_type buf_type;
+	struct am65_cpsw_tx_swdata *swdata;
+	struct cppi5_host_desc_t *desc_tx;
 	struct device *dev = common->dev;
 	struct am65_cpsw_tx_chn *tx_chn;
 	struct netdev_queue *netif_txq;
@@ -1531,15 +1488,18 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 			break;
 		}
 
+		desc_tx = k3_cppi_desc_pool_dma2virt(tx_chn->desc_pool,
+						     desc_dma);
+		swdata = cppi5_hdesc_get_swdata(desc_tx);
+		ndev = swdata->ndev;
 		buf_type = am65_cpsw_nuss_buf_type(tx_chn, desc_dma);
 		if (buf_type == AM65_CPSW_TX_BUF_TYPE_SKB) {
-			skb = am65_cpsw_nuss_tx_compl_packet_skb(tx_chn, desc_dma);
-			ndev = skb->dev;
+			skb = swdata->skb;
+			am65_cpts_tx_timestamp(tx_chn->common->cpts, skb);
 			pkt_len = skb->len;
 			napi_consume_skb(skb, budget);
 		} else {
-			xdpf = am65_cpsw_nuss_tx_compl_packet_xdp(common, tx_chn,
-								  desc_dma, &ndev);
+			xdpf = swdata->xdpf;
 			pkt_len = xdpf->len;
 			if (buf_type == AM65_CPSW_TX_BUF_TYPE_XDP_TX)
 				xdp_return_frame_rx_napi(xdpf);
@@ -1549,7 +1509,8 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 
 		total_bytes += pkt_len;
 		num_tx++;
-
+		am65_cpsw_nuss_xmit_free(tx_chn, desc_tx);
+		dev_sw_netstats_tx_add(ndev, 1, pkt_len);
 		if (!single_port) {
 			/* as packets from multi ports can be interleaved
 			 * on the same channel, we have to figure out the
@@ -1632,12 +1593,12 @@ static netdev_tx_t am65_cpsw_nuss_ndo_slave_xmit(struct sk_buff *skb,
 	struct am65_cpsw_common *common = am65_ndev_to_common(ndev);
 	struct cppi5_host_desc_t *first_desc, *next_desc, *cur_desc;
 	struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+	struct am65_cpsw_tx_swdata *swdata;
 	struct device *dev = common->dev;
 	struct am65_cpsw_tx_chn *tx_chn;
 	struct netdev_queue *netif_txq;
 	dma_addr_t desc_dma, buf_dma;
 	int ret, q_idx, i;
-	void **swdata;
 	u32 *psdata;
 	u32 pkt_len;
 
@@ -1683,7 +1644,8 @@ static netdev_tx_t am65_cpsw_nuss_ndo_slave_xmit(struct sk_buff *skb,
 	k3_udma_glue_tx_dma_to_cppi5_addr(tx_chn->tx_chn, &buf_dma);
 	cppi5_hdesc_attach_buf(first_desc, buf_dma, pkt_len, buf_dma, pkt_len);
 	swdata = cppi5_hdesc_get_swdata(first_desc);
-	*(swdata) = skb;
+	swdata->ndev = ndev;
+	swdata->skb = skb;
 	psdata = cppi5_hdesc_get_psdata(first_desc);
 
 	/* HW csum offload if enabled */
@@ -3525,6 +3487,10 @@ static int am65_cpsw_nuss_probe(struct platform_device *pdev)
 	__be64 id_temp;
 	int ret, i;
 
+	BUILD_BUG_ON_MSG(sizeof(struct am65_cpsw_tx_swdata) > AM65_CPSW_NAV_SW_DATA_SIZE,
+			 "TX SW_DATA size exceeds AM65_CPSW_NAV_SW_DATA_SIZE");
+	BUILD_BUG_ON_MSG(sizeof(struct am65_cpsw_swdata) > AM65_CPSW_NAV_SW_DATA_SIZE,
+			 "SW_DATA size exceeds AM65_CPSW_NAV_SW_DATA_SIZE");
 	common = devm_kzalloc(dev, sizeof(struct am65_cpsw_common), GFP_KERNEL);
 	if (!common)
 		return -ENOMEM;
diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.h b/drivers/net/ethernet/ti/am65-cpsw-nuss.h
index e7832a5cf3cc..917c37e4e89b 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.h
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.h
@@ -104,6 +104,14 @@ struct am65_cpsw_rx_flow {
 	char name[32];
 };
 
+struct am65_cpsw_tx_swdata {
+	struct net_device *ndev;
+	union {
+		struct sk_buff *skb;
+		struct xdp_frame *xdpf;
+	};
+};
+
 struct am65_cpsw_swdata {
 	u32 flow_id;
 	struct page *page;

-- 
2.34.1

Re: [PATCH net-next 1/5] net: ethernet: ti: am65-cpsw: remove am65_cpsw_nuss_tx_compl_packets_2g()

[Simon Horman]

On Mon, Feb 17, 2025 at 09:31:46AM +0200, Roger Quadros wrote:
> The only difference between am65_cpsw_nuss_tx_compl_packets_2g() and
> am65_cpsw_nuss_tx_compl_packets() is the usage of spin_lock() and
> netdev_tx_completed_queue() + am65_cpsw_nuss_tx_wake at every packet
> in the latter.
> 
> Insted of having 2 separate functions for TX completion, merge them

nit, in case there is a v2 for some other reason: Instead

> into one. This will reduce code duplication and make maintenance easier.
> 
> Signed-off-by: Roger Quadros <rogerq@kernel.org>

...

> diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c

...

> @@ -1533,23 +1538,35 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
>  		if (buf_type == AM65_CPSW_TX_BUF_TYPE_SKB) {
>  			skb = am65_cpsw_nuss_tx_compl_packet_skb(tx_chn, desc_dma);
>  			ndev = skb->dev;
> -			total_bytes = skb->len;
> +			pkt_len = skb->len;
>  			napi_consume_skb(skb, budget);
>  		} else {
>  			xdpf = am65_cpsw_nuss_tx_compl_packet_xdp(common, tx_chn,
>  								  desc_dma, &ndev);
> -			total_bytes = xdpf->len;
> +			pkt_len = xdpf->len;
>  			if (buf_type == AM65_CPSW_TX_BUF_TYPE_XDP_TX)
>  				xdp_return_frame_rx_napi(xdpf);
>  			else
>  				xdp_return_frame(xdpf);
>  		}
> +
> +		total_bytes += pkt_len;
>  		num_tx++;
>  
> -		netif_txq = netdev_get_tx_queue(ndev, chn);
> +		if (!single_port) {
> +			/* as packets from multi ports can be interleaved
> +			 * on the same channel, we have to figure out the
> +			 * port/queue at every packet and report it/wake queue.
> +			 */
> +			netif_txq = netdev_get_tx_queue(ndev, chn);
> +			netdev_tx_completed_queue(netif_txq, 1, pkt_len);
> +			am65_cpsw_nuss_tx_wake(tx_chn, ndev, netif_txq);
> +		}
> +	}
>  
> +	if (single_port) {
> +		netif_txq = netdev_get_tx_queue(ndev, chn);
>  		netdev_tx_completed_queue(netif_txq, num_tx, total_bytes);
> -
>  		am65_cpsw_nuss_tx_wake(tx_chn, ndev, netif_txq);
>  	}

Maybe it's not worth it, but it seems that a helper could
avoid duplication of the netif_txq handling immediately above (twice).

Regardless, this looks good to me.

Reviewed-by: Simon Horman <horms@kernel.org>

...

Re: [PATCH net-next 2/5] net: ethernet: ti: am65_cpsw: remove cpu argument am65_cpsw_run_xdp

[Simon Horman]

On Mon, Feb 17, 2025 at 09:31:47AM +0200, Roger Quadros wrote:
> am65_cpsw_run_xdp() can figure out the cpu id itself.
> No need to pass it around 2 functions so drop it.
> 
> Signed-off-by: Roger Quadros <rogerq@kernel.org>

Reviewed-by: Simon Horman <horms@kernel.org>

Re: [PATCH net-next 3/5] net: ethernet: ti: am65-cpsw: use return instead of goto in am65_cpsw_run_xdp()

[Simon Horman]

On Mon, Feb 17, 2025 at 09:31:48AM +0200, Roger Quadros wrote:
> In am65_cpsw_run_xdp() instead of goto followed by return, simply return.
> 
> Signed-off-by: Roger Quadros <rogerq@kernel.org>

Reviewed-by: Simon Horman <horms@kernel.org>

Re: [PATCH net-next 4/5] net: ethernet: ti: am65_cpsw: move am65_cpsw_put_page() out of am65_cpsw_run_xdp()

[Simon Horman]

On Mon, Feb 17, 2025 at 09:31:49AM +0200, Roger Quadros wrote:
> This allows us to re-use am65_cpsw_run_xdp() for zero copy
> case. Add AM65_CPSW_XDP_TX case for successful XDP_TX so we don't
> free the page while in flight.
> 
> Signed-off-by: Roger Quadros <rogerq@kernel.org>

Reviewed-by: Simon Horman <horms@kernel.org>

> ---
>  drivers/net/ethernet/ti/am65-cpsw-nuss.c | 13 ++++++++-----
>  1 file changed, 8 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c

...

> @@ -1230,9 +1230,6 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
>  		ndev->stats.rx_dropped++;
>  	}
>  
> -	page = virt_to_head_page(xdp->data);
> -	am65_cpsw_put_page(flow, page, true);
> -
>  	return ret;

It seems that before and after this patch ret is always initialised to
AM65_CPSW_XDP_CONSUMED and never changed. So it can be removed.

Given that with this patch the function only returns after the switch
statement, I think this would be a nice follow-up.

diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
index 20a4fc3e579f..4052c9153632 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
@@ -1172,7 +1172,6 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 {
 	struct am65_cpsw_common *common = flow->common;
 	struct net_device *ndev = port->ndev;
-	int ret = AM65_CPSW_XDP_CONSUMED;
 	struct am65_cpsw_tx_chn *tx_chn;
 	struct netdev_queue *netif_txq;
 	int cpu = smp_processor_id();
@@ -1228,9 +1227,8 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
 		fallthrough;
 	case XDP_DROP:
 		ndev->stats.rx_dropped++;
+		return AM65_CPSW_XDP_CONSUMED;
 	}
-
-	return ret;
 }
 
 /* RX psdata[2] word format - checksum information */

...

Re: [PATCH net-next 5/5] net: ethernet: ti am65_cpsw: Drop separate TX completion functions

[Simon Horman]

On Mon, Feb 17, 2025 at 09:31:50AM +0200, Roger Quadros wrote:
> Drop separate TX completion functions for SKB and XDP. To do that
> use the SW_DATA mechanism to store ndev and skb/xdpf for TX packets.
> 
> Use BUILD_BUG_ON_MSG() to fail build if SW_DATA size exceeds whats
> available. i.e. AM65_CPSW_NAV_SW_DATA_SIZE.
> 
> Signed-off-by: Roger Quadros <rogerq@kernel.org>

Reviewed-by: Simon Horman <horms@kernel.org>

Re: [PATCH net-next 0/5] net: ethernet: ti: am65-cpsw: drop multiple functions and code cleanup

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (main)
by David S. Miller <davem@davemloft.net>:

On Mon, 17 Feb 2025 09:31:45 +0200 you wrote:
> We have 2 tx completion functions to handle single-port vs multi-port
> variants. Merge them into one function to make maintenance easier.
> 
> We also have 2 functions to handle TX completion for SKB vs XDP.
> Get rid of them too.
> 
> Also do some minor cleanups.
> 
> [...]

Here is the summary with links:
  - [net-next,1/5] net: ethernet: ti: am65-cpsw: remove am65_cpsw_nuss_tx_compl_packets_2g()
    https://git.kernel.org/netdev/net-next/c/9a369ae3d143
  - [net-next,2/5] net: ethernet: ti: am65_cpsw: remove cpu argument am65_cpsw_run_xdp
    https://git.kernel.org/netdev/net-next/c/1ae26bf61517
  - [net-next,3/5] net: ethernet: ti: am65-cpsw: use return instead of goto in am65_cpsw_run_xdp()
    https://git.kernel.org/netdev/net-next/c/09057ce3774e
  - [net-next,4/5] net: ethernet: ti: am65_cpsw: move am65_cpsw_put_page() out of am65_cpsw_run_xdp()
    https://git.kernel.org/netdev/net-next/c/6d6c7933cea6
  - [net-next,5/5] net: ethernet: ti am65_cpsw: Drop separate TX completion functions
    https://git.kernel.org/netdev/net-next/c/ce643fa62a70

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

Re: [PATCH net-next 4/5] net: ethernet: ti: am65_cpsw: move am65_cpsw_put_page() out of am65_cpsw_run_xdp()

[Roger Quadros]

On 18/02/2025 20:03, Simon Horman wrote:
> On Mon, Feb 17, 2025 at 09:31:49AM +0200, Roger Quadros wrote:
>> This allows us to re-use am65_cpsw_run_xdp() for zero copy
>> case. Add AM65_CPSW_XDP_TX case for successful XDP_TX so we don't
>> free the page while in flight.
>>
>> Signed-off-by: Roger Quadros <rogerq@kernel.org>
> 
> Reviewed-by: Simon Horman <horms@kernel.org>
> 
>> ---
>>  drivers/net/ethernet/ti/am65-cpsw-nuss.c | 13 ++++++++-----
>>  1 file changed, 8 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
> 
> ...
> 
>> @@ -1230,9 +1230,6 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
>>  		ndev->stats.rx_dropped++;
>>  	}
>>  
>> -	page = virt_to_head_page(xdp->data);
>> -	am65_cpsw_put_page(flow, page, true);
>> -
>>  	return ret;
> 
> It seems that before and after this patch ret is always initialised to
> AM65_CPSW_XDP_CONSUMED and never changed. So it can be removed.
> 
> Given that with this patch the function only returns after the switch
> statement, I think this would be a nice follow-up.
> 
> diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
> index 20a4fc3e579f..4052c9153632 100644
> --- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
> +++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
> @@ -1172,7 +1172,6 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
>  {
>  	struct am65_cpsw_common *common = flow->common;
>  	struct net_device *ndev = port->ndev;
> -	int ret = AM65_CPSW_XDP_CONSUMED;
>  	struct am65_cpsw_tx_chn *tx_chn;
>  	struct netdev_queue *netif_txq;
>  	int cpu = smp_processor_id();
> @@ -1228,9 +1227,8 @@ static int am65_cpsw_run_xdp(struct am65_cpsw_rx_flow *flow,
>  		fallthrough;
>  	case XDP_DROP:
>  		ndev->stats.rx_dropped++;
> +		return AM65_CPSW_XDP_CONSUMED;
>  	}
> -
> -	return ret;
>  }
>  
>  /* RX psdata[2] word format - checksum information */
> 
> ...

Thank you for this suggestion. I will add this cleanup in my next set of patches.

-- 
cheers,
-roger