net: thunderbolt: Add tracepoints

These are useful when debugging various performance issues.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Acked-by: Yehezkel Bernat <YehezkelShB@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/thunderbolt/Makefile

@@ -1,3 +1,6 @@
 # SPDX-License-Identifier: GPL-2.0
 obj-$(CONFIG_USB4_NET) := thunderbolt_net.o
-thunderbolt_net-objs := main.o
+thunderbolt_net-objs := main.o trace.o
+
+# Tracepoints need to know where to find trace.h
+CFLAGS_trace.o := -I$(src)

drivers/net/thunderbolt/main.c

@@ -23,6 +23,8 @@
 
 #include <net/ip6_checksum.h>
 
+#include "trace.h"
+
 /* Protocol timeouts in ms */
 #define TBNET_LOGIN_DELAY	4500
 #define TBNET_LOGIN_TIMEOUT	500
@@ -353,6 +355,8 @@ static void tbnet_free_buffers(struct tbnet_ring *ring)
 			size = TBNET_RX_PAGE_SIZE;
 		}
 
+		trace_tbnet_free_frame(i, tf->page, tf->frame.buffer_phy, dir);
+
 		if (tf->frame.buffer_phy)
 			dma_unmap_page(dma_dev, tf->frame.buffer_phy, size,
 				       dir);
@@ -526,6 +530,9 @@ static int tbnet_alloc_rx_buffers(struct tbnet *net, unsigned int nbuffers)
 		tf->frame.buffer_phy = dma_addr;
 		tf->dev = net->dev;
 
+		trace_tbnet_alloc_rx_frame(index, tf->page, dma_addr,
+					   DMA_FROM_DEVICE);
+
 		tb_ring_rx(ring->ring, &tf->frame);
 
 		ring->prod++;
@@ -602,6 +609,8 @@ static int tbnet_alloc_tx_buffers(struct tbnet *net)
 		tf->frame.callback = tbnet_tx_callback;
 		tf->frame.sof = TBIP_PDF_FRAME_START;
 		tf->frame.eof = TBIP_PDF_FRAME_END;
+
+		trace_tbnet_alloc_tx_frame(i, tf->page, dma_addr, DMA_TO_DEVICE);
 	}
 
 	ring->cons = 0;
@@ -832,12 +841,16 @@ static int tbnet_poll(struct napi_struct *napi, int budget)
 
 		hdr = page_address(page);
 		if (!tbnet_check_frame(net, tf, hdr)) {
+			trace_tbnet_invalid_rx_ip_frame(hdr->frame_size,
+				hdr->frame_id, hdr->frame_index, hdr->frame_count);
 			__free_pages(page, TBNET_RX_PAGE_ORDER);
 			dev_kfree_skb_any(net->skb);
 			net->skb = NULL;
 			continue;
 		}
 
+		trace_tbnet_rx_ip_frame(hdr->frame_size, hdr->frame_id,
+					hdr->frame_index, hdr->frame_count);
 		frame_size = le32_to_cpu(hdr->frame_size);
 
 		skb = net->skb;
@@ -871,6 +884,7 @@ static int tbnet_poll(struct napi_struct *napi, int budget)
 
 		if (last) {
 			skb->protocol = eth_type_trans(skb, net->dev);
+			trace_tbnet_rx_skb(skb);
 			napi_gro_receive(&net->napi, skb);
 			net->skb = NULL;
 		}
@@ -990,6 +1004,8 @@ static bool tbnet_xmit_csum_and_map(struct tbnet *net, struct sk_buff *skb,
 		for (i = 0; i < frame_count; i++) {
 			hdr = page_address(frames[i]->page);
 			hdr->frame_count = cpu_to_le32(frame_count);
+			trace_tbnet_tx_ip_frame(hdr->frame_size, hdr->frame_id,
+						hdr->frame_index, hdr->frame_count);
 			dma_sync_single_for_device(dma_dev,
 				frames[i]->frame.buffer_phy,
 				tbnet_frame_size(frames[i]), DMA_TO_DEVICE);
@@ -1054,6 +1070,8 @@ static bool tbnet_xmit_csum_and_map(struct tbnet *net, struct sk_buff *skb,
 		len = le32_to_cpu(hdr->frame_size) - offset;
 		wsum = csum_partial(dest, len, wsum);
 		hdr->frame_count = cpu_to_le32(frame_count);
+		trace_tbnet_tx_ip_frame(hdr->frame_size, hdr->frame_id,
+					hdr->frame_index, hdr->frame_count);
 
 		offset = 0;
 	}
@@ -1096,6 +1114,8 @@ static netdev_tx_t tbnet_start_xmit(struct sk_buff *skb,
 	bool unmap = false;
 	void *dest;
 
+	trace_tbnet_tx_skb(skb);
+
 	nframes = DIV_ROUND_UP(data_len, TBNET_MAX_PAYLOAD_SIZE);
 	if (tbnet_available_buffers(&net->tx_ring) < nframes) {
 		netif_stop_queue(net->dev);
@@ -1202,6 +1222,7 @@ static netdev_tx_t tbnet_start_xmit(struct sk_buff *skb,
 	net->stats.tx_packets++;
 	net->stats.tx_bytes += skb->len;
 
+	trace_tbnet_consume_skb(skb);
 	dev_consume_skb_any(skb);
 
 	return NETDEV_TX_OK;

drivers/net/thunderbolt/trace.c

@@ -0,0 +1,10 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Tracepoints for Thunderbolt/USB4 networking driver
+ *
+ * Copyright (C) 2023, Intel Corporation
+ * Author: Mika Westerberg <mika.westerberg@linux.intel.com>
+ */
+
+#define CREATE_TRACE_POINTS
+#include "trace.h"

drivers/net/thunderbolt/trace.h

@@ -0,0 +1,141 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Tracepoints for Thunderbolt/USB4 networking driver
+ *
+ * Copyright (C) 2023, Intel Corporation
+ * Author: Mika Westerberg <mika.westerberg@linux.intel.com>
+ */
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM thunderbolt_net
+
+#if !defined(__TRACE_THUNDERBOLT_NET_H) || defined(TRACE_HEADER_MULTI_READ)
+#define __TRACE_THUNDERBOLT_NET_H
+
+#include <linux/dma-direction.h>
+#include <linux/skbuff.h>
+#include <linux/tracepoint.h>
+
+#define DMA_DATA_DIRECTION_NAMES			\
+	{ DMA_BIDIRECTIONAL, "DMA_BIDIRECTIONAL" },	\
+	{ DMA_TO_DEVICE, "DMA_TO_DEVICE" },		\
+	{ DMA_FROM_DEVICE, "DMA_FROM_DEVICE" },		\
+	{ DMA_NONE, "DMA_NONE" }
+
+DECLARE_EVENT_CLASS(tbnet_frame,
+	TP_PROTO(unsigned int index, const void *page, dma_addr_t phys,
+		 enum dma_data_direction dir),
+	TP_ARGS(index, page, phys, dir),
+	TP_STRUCT__entry(
+		__field(unsigned int, index)
+		__field(const void *, page)
+		__field(dma_addr_t, phys)
+		__field(enum dma_data_direction, dir)
+	),
+	TP_fast_assign(
+		__entry->index = index;
+		__entry->page = page;
+		__entry->phys = phys;
+		__entry->dir = dir;
+	),
+	TP_printk("index=%u page=%p phys=%pad dir=%s",
+		  __entry->index, __entry->page, &__entry->phys,
+		__print_symbolic(__entry->dir, DMA_DATA_DIRECTION_NAMES))
+);
+
+DEFINE_EVENT(tbnet_frame, tbnet_alloc_rx_frame,
+	TP_PROTO(unsigned int index, const void *page, dma_addr_t phys,
+		 enum dma_data_direction dir),
+	TP_ARGS(index, page, phys, dir)
+);
+
+DEFINE_EVENT(tbnet_frame, tbnet_alloc_tx_frame,
+	TP_PROTO(unsigned int index, const void *page, dma_addr_t phys,
+		 enum dma_data_direction dir),
+	TP_ARGS(index, page, phys, dir)
+);
+
+DEFINE_EVENT(tbnet_frame, tbnet_free_frame,
+	TP_PROTO(unsigned int index, const void *page, dma_addr_t phys,
+		 enum dma_data_direction dir),
+	TP_ARGS(index, page, phys, dir)
+);
+
+DECLARE_EVENT_CLASS(tbnet_ip_frame,
+	TP_PROTO(__le32 size, __le16 id, __le16 index, __le32 count),
+	TP_ARGS(size, id, index, count),
+	TP_STRUCT__entry(
+		__field(u32, size)
+		__field(u16, id)
+		__field(u16, index)
+		__field(u32, count)
+	),
+	TP_fast_assign(
+		__entry->size = le32_to_cpu(size);
+		__entry->id = le16_to_cpu(id);
+		__entry->index = le16_to_cpu(index);
+		__entry->count = le32_to_cpu(count);
+	),
+	TP_printk("id=%u size=%u index=%u count=%u",
+		  __entry->id, __entry->size, __entry->index, __entry->count)
+);
+
+DEFINE_EVENT(tbnet_ip_frame, tbnet_rx_ip_frame,
+	TP_PROTO(__le32 size, __le16 id, __le16 index, __le32 count),
+	TP_ARGS(size, id, index, count)
+);
+
+DEFINE_EVENT(tbnet_ip_frame, tbnet_invalid_rx_ip_frame,
+	TP_PROTO(__le32 size, __le16 id, __le16 index, __le32 count),
+	TP_ARGS(size, id, index, count)
+);
+
+DEFINE_EVENT(tbnet_ip_frame, tbnet_tx_ip_frame,
+	TP_PROTO(__le32 size, __le16 id, __le16 index, __le32 count),
+	TP_ARGS(size, id, index, count)
+);
+
+DECLARE_EVENT_CLASS(tbnet_skb,
+	TP_PROTO(const struct sk_buff *skb),
+	TP_ARGS(skb),
+	TP_STRUCT__entry(
+		__field(const void *, addr)
+		__field(unsigned int, len)
+		__field(unsigned int, data_len)
+		__field(unsigned int, nr_frags)
+	),
+	TP_fast_assign(
+		__entry->addr = skb;
+		__entry->len = skb->len;
+		__entry->data_len = skb->data_len;
+		__entry->nr_frags = skb_shinfo(skb)->nr_frags;
+	),
+	TP_printk("skb=%p len=%u data_len=%u nr_frags=%u",
+		  __entry->addr, __entry->len, __entry->data_len,
+		  __entry->nr_frags)
+);
+
+DEFINE_EVENT(tbnet_skb, tbnet_rx_skb,
+	TP_PROTO(const struct sk_buff *skb),
+	TP_ARGS(skb)
+);
+
+DEFINE_EVENT(tbnet_skb, tbnet_tx_skb,
+	TP_PROTO(const struct sk_buff *skb),
+	TP_ARGS(skb)
+);
+
+DEFINE_EVENT(tbnet_skb, tbnet_consume_skb,
+	TP_PROTO(const struct sk_buff *skb),
+	TP_ARGS(skb)
+);
+
+#endif /* _TRACE_THUNDERBOLT_NET_H */
+
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_FILE trace
+
+#include <trace/define_trace.h>