[PATCH net-next v2 0/5] r8152: RX improve

[PATCH net-next v2 0/5] r8152: RX improve

[Hayes Wang]

v2:
For patch #2, replace list_for_each_safe with list_for_each_entry_safe.
Remove unlikely in WARN_ON. Adjust the coding style.

For patch #4, replace list_for_each_safe with list_for_each_entry_safe.
Remove "else" after "continue".

For patch #5. replace sysfs with ethtool to modify rx_copybreak and
rx_pending.

v1:
The different chips use different rx buffer size.

Use skb_add_rx_frag() to reduce memory copy for RX.

Hayes Wang (5):
  r8152: separate the rx buffer size
  r8152: replace array with linking list for rx information
  r8152: use alloc_pages for rx buffer
  r8152: support skb_add_rx_frag
  r8152: change rx_copybreak and rx_pending through ethtool

 drivers/net/usb/r8152.c | 374 ++++++++++++++++++++++++++++++++--------
 1 file changed, 304 insertions(+), 70 deletions(-)

-- 
2.21.0

[PATCH net-next v2 1/5] r8152: separate the rx buffer size

[Hayes Wang]

The different chips may accept different rx buffer sizes. The RTL8152
supports 16K bytes, and RTL8153 support 32K bytes.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
---
 drivers/net/usb/r8152.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/net/usb/r8152.c b/drivers/net/usb/r8152.c
index 0cc03a9ff545..94da79028a65 100644
--- a/drivers/net/usb/r8152.c
+++ b/drivers/net/usb/r8152.c
@@ -749,6 +749,7 @@ struct r8152 {
 	u32 msg_enable;
 	u32 tx_qlen;
 	u32 coalesce;
+	u32 rx_buf_sz;
 	u16 ocp_base;
 	u16 speed;
 	u8 *intr_buff;
@@ -1516,13 +1517,13 @@ static int alloc_all_mem(struct r8152 *tp)
 	skb_queue_head_init(&tp->rx_queue);
 
 	for (i = 0; i < RTL8152_MAX_RX; i++) {
-		buf = kmalloc_node(agg_buf_sz, GFP_KERNEL, node);
+		buf = kmalloc_node(tp->rx_buf_sz, GFP_KERNEL, node);
 		if (!buf)
 			goto err1;
 
 		if (buf != rx_agg_align(buf)) {
 			kfree(buf);
-			buf = kmalloc_node(agg_buf_sz + RX_ALIGN, GFP_KERNEL,
+			buf = kmalloc_node(tp->rx_buf_sz + RX_ALIGN, GFP_KERNEL,
 					   node);
 			if (!buf)
 				goto err1;
@@ -2113,7 +2114,7 @@ int r8152_submit_rx(struct r8152 *tp, struct rx_agg *agg, gfp_t mem_flags)
 		return 0;
 
 	usb_fill_bulk_urb(agg->urb, tp->udev, usb_rcvbulkpipe(tp->udev, 1),
-			  agg->head, agg_buf_sz,
+			  agg->head, tp->rx_buf_sz,
 			  (usb_complete_t)read_bulk_callback, agg);
 
 	ret = usb_submit_urb(agg->urb, mem_flags);
@@ -2447,7 +2448,7 @@ static void r8153_set_rx_early_timeout(struct r8152 *tp)
 
 static void r8153_set_rx_early_size(struct r8152 *tp)
 {
-	u32 ocp_data = agg_buf_sz - rx_reserved_size(tp->netdev->mtu);
+	u32 ocp_data = tp->rx_buf_sz - rx_reserved_size(tp->netdev->mtu);
 
 	switch (tp->version) {
 	case RTL_VER_03:
@@ -5115,6 +5116,7 @@ static int rtl_ops_init(struct r8152 *tp)
 		ops->in_nway		= rtl8152_in_nway;
 		ops->hw_phy_cfg		= r8152b_hw_phy_cfg;
 		ops->autosuspend_en	= rtl_runtime_suspend_enable;
+		tp->rx_buf_sz		= 16 * 1024;
 		break;
 
 	case RTL_VER_03:
@@ -5132,6 +5134,7 @@ static int rtl_ops_init(struct r8152 *tp)
 		ops->in_nway		= rtl8153_in_nway;
 		ops->hw_phy_cfg		= r8153_hw_phy_cfg;
 		ops->autosuspend_en	= rtl8153_runtime_enable;
+		tp->rx_buf_sz		= 32 * 1024;
 		break;
 
 	case RTL_VER_08:
@@ -5147,6 +5150,7 @@ static int rtl_ops_init(struct r8152 *tp)
 		ops->in_nway		= rtl8153_in_nway;
 		ops->hw_phy_cfg		= r8153b_hw_phy_cfg;
 		ops->autosuspend_en	= rtl8153b_runtime_enable;
+		tp->rx_buf_sz		= 32 * 1024;
 		break;
 
 	default:
-- 
2.21.0

[PATCH net-next v2 2/5] r8152: replace array with linking list for rx information

[Hayes Wang]

The original method uses an array to store the rx information. The
new one uses a list to link each rx structure. Then, it is possible
to increase/decrease the number of rx structure dynamically.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
---
 drivers/net/usb/r8152.c | 182 +++++++++++++++++++++++++++-------------
 1 file changed, 125 insertions(+), 57 deletions(-)

diff --git a/drivers/net/usb/r8152.c b/drivers/net/usb/r8152.c
index 94da79028a65..d063c9b358e5 100644
--- a/drivers/net/usb/r8152.c
+++ b/drivers/net/usb/r8152.c
@@ -22,6 +22,7 @@
 #include <linux/mdio.h>
 #include <linux/usb/cdc.h>
 #include <linux/suspend.h>
+#include <linux/atomic.h>
 #include <linux/acpi.h>
 
 /* Information for net-next */
@@ -694,7 +695,7 @@ struct tx_desc {
 struct r8152;
 
 struct rx_agg {
-	struct list_head list;
+	struct list_head list, info_list;
 	struct urb *urb;
 	struct r8152 *context;
 	void *buffer;
@@ -719,7 +720,7 @@ struct r8152 {
 	struct net_device *netdev;
 	struct urb *intr_urb;
 	struct tx_agg tx_info[RTL8152_MAX_TX];
-	struct rx_agg rx_info[RTL8152_MAX_RX];
+	struct list_head rx_info;
 	struct list_head rx_done, tx_free;
 	struct sk_buff_head tx_queue, rx_queue;
 	spinlock_t rx_lock, tx_lock;
@@ -744,6 +745,8 @@ struct r8152 {
 		void (*autosuspend_en)(struct r8152 *tp, bool enable);
 	} rtl_ops;
 
+	atomic_t rx_count;
+
 	int intr_interval;
 	u32 saved_wolopts;
 	u32 msg_enable;
@@ -1468,18 +1471,81 @@ static inline void *tx_agg_align(void *data)
 	return (void *)ALIGN((uintptr_t)data, TX_ALIGN);
 }
 
+static void free_rx_agg(struct r8152 *tp, struct rx_agg *agg)
+{
+	list_del(&agg->info_list);
+
+	usb_free_urb(agg->urb);
+	kfree(agg->buffer);
+	kfree(agg);
+
+	atomic_dec(&tp->rx_count);
+}
+
+static struct rx_agg *alloc_rx_agg(struct r8152 *tp, gfp_t mflags)
+{
+	struct net_device *netdev = tp->netdev;
+	int node = netdev->dev.parent ? dev_to_node(netdev->dev.parent) : -1;
+	struct rx_agg *rx_agg;
+	unsigned long flags;
+	u8 *buf;
+
+	rx_agg = kmalloc_node(sizeof(*rx_agg), mflags, node);
+	if (!rx_agg)
+		return NULL;
+
+	buf = kmalloc_node(tp->rx_buf_sz, mflags, node);
+	if (!buf)
+		goto free_rx;
+
+	if (buf != rx_agg_align(buf)) {
+		kfree(buf);
+		buf = kmalloc_node(tp->rx_buf_sz + RX_ALIGN, mflags,
+				   node);
+		if (!buf)
+			goto free_rx;
+	}
+
+	rx_agg->buffer = buf;
+	rx_agg->head = rx_agg_align(buf);
+
+	rx_agg->urb = usb_alloc_urb(0, mflags);
+	if (!rx_agg->urb)
+		goto free_buf;
+
+	rx_agg->context = tp;
+
+	INIT_LIST_HEAD(&rx_agg->list);
+	INIT_LIST_HEAD(&rx_agg->info_list);
+	spin_lock_irqsave(&tp->rx_lock, flags);
+	list_add_tail(&rx_agg->info_list, &tp->rx_info);
+	spin_unlock_irqrestore(&tp->rx_lock, flags);
+
+	atomic_inc(&tp->rx_count);
+
+	return rx_agg;
+
+free_buf:
+	kfree(rx_agg->buffer);
+free_rx:
+	kfree(rx_agg);
+	return NULL;
+}
+
 static void free_all_mem(struct r8152 *tp)
 {
+	struct rx_agg *agg, *agg_next;
+	unsigned long flags;
 	int i;
 
-	for (i = 0; i < RTL8152_MAX_RX; i++) {
-		usb_free_urb(tp->rx_info[i].urb);
-		tp->rx_info[i].urb = NULL;
+	spin_lock_irqsave(&tp->rx_lock, flags);
 
-		kfree(tp->rx_info[i].buffer);
-		tp->rx_info[i].buffer = NULL;
-		tp->rx_info[i].head = NULL;
-	}
+	list_for_each_entry_safe(agg, agg_next, &tp->rx_info, info_list)
+		free_rx_agg(tp, agg);
+
+	spin_unlock_irqrestore(&tp->rx_lock, flags);
+
+	WARN_ON(atomic_read(&tp->rx_count));
 
 	for (i = 0; i < RTL8152_MAX_TX; i++) {
 		usb_free_urb(tp->tx_info[i].urb);
@@ -1503,46 +1569,28 @@ static int alloc_all_mem(struct r8152 *tp)
 	struct usb_interface *intf = tp->intf;
 	struct usb_host_interface *alt = intf->cur_altsetting;
 	struct usb_host_endpoint *ep_intr = alt->endpoint + 2;
-	struct urb *urb;
 	int node, i;
-	u8 *buf;
 
 	node = netdev->dev.parent ? dev_to_node(netdev->dev.parent) : -1;
 
 	spin_lock_init(&tp->rx_lock);
 	spin_lock_init(&tp->tx_lock);
+	INIT_LIST_HEAD(&tp->rx_info);
 	INIT_LIST_HEAD(&tp->tx_free);
 	INIT_LIST_HEAD(&tp->rx_done);
 	skb_queue_head_init(&tp->tx_queue);
 	skb_queue_head_init(&tp->rx_queue);
+	atomic_set(&tp->rx_count, 0);
 
 	for (i = 0; i < RTL8152_MAX_RX; i++) {
-		buf = kmalloc_node(tp->rx_buf_sz, GFP_KERNEL, node);
-		if (!buf)
+		if (!alloc_rx_agg(tp, GFP_KERNEL))
 			goto err1;
-
-		if (buf != rx_agg_align(buf)) {
-			kfree(buf);
-			buf = kmalloc_node(tp->rx_buf_sz + RX_ALIGN, GFP_KERNEL,
-					   node);
-			if (!buf)
-				goto err1;
-		}
-
-		urb = usb_alloc_urb(0, GFP_KERNEL);
-		if (!urb) {
-			kfree(buf);
-			goto err1;
-		}
-
-		INIT_LIST_HEAD(&tp->rx_info[i].list);
-		tp->rx_info[i].context = tp;
-		tp->rx_info[i].urb = urb;
-		tp->rx_info[i].buffer = buf;
-		tp->rx_info[i].head = rx_agg_align(buf);
 	}
 
 	for (i = 0; i < RTL8152_MAX_TX; i++) {
+		struct urb *urb;
+		u8 *buf;
+
 		buf = kmalloc_node(agg_buf_sz, GFP_KERNEL, node);
 		if (!buf)
 			goto err1;
@@ -2331,44 +2379,64 @@ static void rxdy_gated_en(struct r8152 *tp, bool enable)
 
 static int rtl_start_rx(struct r8152 *tp)
 {
-	int i, ret = 0;
+	struct rx_agg *agg, *agg_next;
+	struct list_head tmp_list;
+	unsigned long flags;
+	int ret = 0;
 
-	INIT_LIST_HEAD(&tp->rx_done);
-	for (i = 0; i < RTL8152_MAX_RX; i++) {
-		INIT_LIST_HEAD(&tp->rx_info[i].list);
-		ret = r8152_submit_rx(tp, &tp->rx_info[i], GFP_KERNEL);
-		if (ret)
-			break;
-	}
+	INIT_LIST_HEAD(&tmp_list);
 
-	if (ret && ++i < RTL8152_MAX_RX) {
-		struct list_head rx_queue;
-		unsigned long flags;
+	spin_lock_irqsave(&tp->rx_lock, flags);
 
-		INIT_LIST_HEAD(&rx_queue);
+	INIT_LIST_HEAD(&tp->rx_done);
 
-		do {
-			struct rx_agg *agg = &tp->rx_info[i++];
-			struct urb *urb = agg->urb;
+	list_splice_init(&tp->rx_info, &tmp_list);
 
-			urb->actual_length = 0;
-			list_add_tail(&agg->list, &rx_queue);
-		} while (i < RTL8152_MAX_RX);
+	spin_unlock_irqrestore(&tp->rx_lock, flags);
 
-		spin_lock_irqsave(&tp->rx_lock, flags);
-		list_splice_tail(&rx_queue, &tp->rx_done);
-		spin_unlock_irqrestore(&tp->rx_lock, flags);
+	list_for_each_entry_safe(agg, agg_next, &tmp_list, info_list) {
+		INIT_LIST_HEAD(&agg->list);
+
+		if (ret < 0)
+			list_add_tail(&agg->list, &tp->rx_done);
+		else
+			ret = r8152_submit_rx(tp, agg, GFP_KERNEL);
 	}
 
+	spin_lock_irqsave(&tp->rx_lock, flags);
+	WARN_ON(!list_empty(&tp->rx_info));
+	list_splice(&tmp_list, &tp->rx_info);
+	spin_unlock_irqrestore(&tp->rx_lock, flags);
+
 	return ret;
 }
 
 static int rtl_stop_rx(struct r8152 *tp)
 {
-	int i;
+	struct rx_agg *agg, *agg_next;
+	struct list_head tmp_list;
+	unsigned long flags;
+
+	INIT_LIST_HEAD(&tmp_list);
+
+	/* The usb_kill_urb() couldn't be used in atomic.
+	 * Therefore, move the list of rx_info to a tmp one.
+	 * Then, list_for_each_entry_safe could be used without
+	 * spin lock.
+	 */
+
+	spin_lock_irqsave(&tp->rx_lock, flags);
+	list_splice_init(&tp->rx_info, &tmp_list);
+	spin_unlock_irqrestore(&tp->rx_lock, flags);
+
+	list_for_each_entry_safe(agg, agg_next, &tmp_list, info_list)
+		usb_kill_urb(agg->urb);
 
-	for (i = 0; i < RTL8152_MAX_RX; i++)
-		usb_kill_urb(tp->rx_info[i].urb);
+	/* Move back the list of temp to the rx_info */
+	spin_lock_irqsave(&tp->rx_lock, flags);
+	WARN_ON(!list_empty(&tp->rx_info));
+	list_splice(&tmp_list, &tp->rx_info);
+	spin_unlock_irqrestore(&tp->rx_lock, flags);
 
 	while (!skb_queue_empty(&tp->rx_queue))
 		dev_kfree_skb(__skb_dequeue(&tp->rx_queue));
-- 
2.21.0

[PATCH net-next v2 3/5] r8152: use alloc_pages for rx buffer

[Hayes Wang]

Replace kmalloc_node() with alloc_pages() for rx buffer.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
---
 drivers/net/usb/r8152.c | 31 +++++++++++--------------------
 1 file changed, 11 insertions(+), 20 deletions(-)

diff --git a/drivers/net/usb/r8152.c b/drivers/net/usb/r8152.c
index d063c9b358e5..f41cb728e999 100644
--- a/drivers/net/usb/r8152.c
+++ b/drivers/net/usb/r8152.c
@@ -698,8 +698,8 @@ struct rx_agg {
 	struct list_head list, info_list;
 	struct urb *urb;
 	struct r8152 *context;
+	struct page *page;
 	void *buffer;
-	void *head;
 };
 
 struct tx_agg {
@@ -1476,7 +1476,7 @@ static void free_rx_agg(struct r8152 *tp, struct rx_agg *agg)
 	list_del(&agg->info_list);
 
 	usb_free_urb(agg->urb);
-	kfree(agg->buffer);
+	__free_pages(agg->page, get_order(tp->rx_buf_sz));
 	kfree(agg);
 
 	atomic_dec(&tp->rx_count);
@@ -1486,28 +1486,19 @@ static struct rx_agg *alloc_rx_agg(struct r8152 *tp, gfp_t mflags)
 {
 	struct net_device *netdev = tp->netdev;
 	int node = netdev->dev.parent ? dev_to_node(netdev->dev.parent) : -1;
+	unsigned int order = get_order(tp->rx_buf_sz);
 	struct rx_agg *rx_agg;
 	unsigned long flags;
-	u8 *buf;
 
 	rx_agg = kmalloc_node(sizeof(*rx_agg), mflags, node);
 	if (!rx_agg)
 		return NULL;
 
-	buf = kmalloc_node(tp->rx_buf_sz, mflags, node);
-	if (!buf)
+	rx_agg->page = alloc_pages(mflags, order);
+	if (!rx_agg->page)
 		goto free_rx;
 
-	if (buf != rx_agg_align(buf)) {
-		kfree(buf);
-		buf = kmalloc_node(tp->rx_buf_sz + RX_ALIGN, mflags,
-				   node);
-		if (!buf)
-			goto free_rx;
-	}
-
-	rx_agg->buffer = buf;
-	rx_agg->head = rx_agg_align(buf);
+	rx_agg->buffer = page_address(rx_agg->page);
 
 	rx_agg->urb = usb_alloc_urb(0, mflags);
 	if (!rx_agg->urb)
@@ -1526,7 +1517,7 @@ static struct rx_agg *alloc_rx_agg(struct r8152 *tp, gfp_t mflags)
 	return rx_agg;
 
 free_buf:
-	kfree(rx_agg->buffer);
+	__free_pages(rx_agg->page, order);
 free_rx:
 	kfree(rx_agg);
 	return NULL;
@@ -2003,8 +1994,8 @@ static int rx_bottom(struct r8152 *tp, int budget)
 		if (urb->actual_length < ETH_ZLEN)
 			goto submit;
 
-		rx_desc = agg->head;
-		rx_data = agg->head;
+		rx_desc = agg->buffer;
+		rx_data = agg->buffer;
 		len_used += sizeof(struct rx_desc);
 
 		while (urb->actual_length > len_used) {
@@ -2051,7 +2042,7 @@ static int rx_bottom(struct r8152 *tp, int budget)
 find_next_rx:
 			rx_data = rx_agg_align(rx_data + pkt_len + ETH_FCS_LEN);
 			rx_desc = (struct rx_desc *)rx_data;
-			len_used = (int)(rx_data - (u8 *)agg->head);
+			len_used = (int)(rx_data - (u8 *)agg->buffer);
 			len_used += sizeof(struct rx_desc);
 		}
 
@@ -2162,7 +2153,7 @@ int r8152_submit_rx(struct r8152 *tp, struct rx_agg *agg, gfp_t mem_flags)
 		return 0;
 
 	usb_fill_bulk_urb(agg->urb, tp->udev, usb_rcvbulkpipe(tp->udev, 1),
-			  agg->head, tp->rx_buf_sz,
+			  agg->buffer, tp->rx_buf_sz,
 			  (usb_complete_t)read_bulk_callback, agg);
 
 	ret = usb_submit_urb(agg->urb, mem_flags);
-- 
2.21.0

[PATCH net-next v2 4/5] r8152: support skb_add_rx_frag

[Hayes Wang]

Use skb_add_rx_frag() to reduce the memory copy for rx data.

Use a new list of rx_used to store the rx buffer which couldn't be
reused yet.

Besides, the total number of rx buffer may be increased or decreased
dynamically. And it is limited by RTL8152_MAX_RX_AGG.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
---
 drivers/net/usb/r8152.c | 120 +++++++++++++++++++++++++++++++++++-----
 1 file changed, 105 insertions(+), 15 deletions(-)

diff --git a/drivers/net/usb/r8152.c b/drivers/net/usb/r8152.c
index f41cb728e999..2ae04522cd5a 100644
--- a/drivers/net/usb/r8152.c
+++ b/drivers/net/usb/r8152.c
@@ -584,6 +584,9 @@ enum rtl_register_content {
 #define TX_ALIGN		4
 #define RX_ALIGN		8
 
+#define RTL8152_MAX_RX_AGG	(10 * RTL8152_MAX_RX)
+#define RTL8152_RXFG_HEADSZ	256
+
 #define INTR_LINK		0x0004
 
 #define RTL8152_REQT_READ	0xc0
@@ -720,7 +723,7 @@ struct r8152 {
 	struct net_device *netdev;
 	struct urb *intr_urb;
 	struct tx_agg tx_info[RTL8152_MAX_TX];
-	struct list_head rx_info;
+	struct list_head rx_info, rx_used;
 	struct list_head rx_done, tx_free;
 	struct sk_buff_head tx_queue, rx_queue;
 	spinlock_t rx_lock, tx_lock;
@@ -1476,7 +1479,7 @@ static void free_rx_agg(struct r8152 *tp, struct rx_agg *agg)
 	list_del(&agg->info_list);
 
 	usb_free_urb(agg->urb);
-	__free_pages(agg->page, get_order(tp->rx_buf_sz));
+	put_page(agg->page);
 	kfree(agg);
 
 	atomic_dec(&tp->rx_count);
@@ -1494,7 +1497,7 @@ static struct rx_agg *alloc_rx_agg(struct r8152 *tp, gfp_t mflags)
 	if (!rx_agg)
 		return NULL;
 
-	rx_agg->page = alloc_pages(mflags, order);
+	rx_agg->page = alloc_pages(mflags | __GFP_COMP, order);
 	if (!rx_agg->page)
 		goto free_rx;
 
@@ -1947,6 +1950,46 @@ static u8 r8152_rx_csum(struct r8152 *tp, struct rx_desc *rx_desc)
 	return checksum;
 }
 
+static inline bool rx_count_exceed(struct r8152 *tp)
+{
+	return atomic_read(&tp->rx_count) > RTL8152_MAX_RX;
+}
+
+static inline int agg_offset(struct rx_agg *agg, void *addr)
+{
+	return (int)(addr - agg->buffer);
+}
+
+static struct rx_agg *rtl_get_free_rx(struct r8152 *tp, gfp_t mflags)
+{
+	struct rx_agg *agg, *agg_next, *agg_free = NULL;
+	unsigned long flags;
+
+	spin_lock_irqsave(&tp->rx_lock, flags);
+
+	list_for_each_entry_safe(agg, agg_next, &tp->rx_used, list) {
+		if (page_count(agg->page) == 1) {
+			if (!agg_free) {
+				list_del_init(&agg->list);
+				agg_free = agg;
+				continue;
+			}
+			if (rx_count_exceed(tp)) {
+				list_del_init(&agg->list);
+				free_rx_agg(tp, agg);
+			}
+			break;
+		}
+	}
+
+	spin_unlock_irqrestore(&tp->rx_lock, flags);
+
+	if (!agg_free && atomic_read(&tp->rx_count) < RTL8152_MAX_RX_AGG)
+		agg_free = alloc_rx_agg(tp, mflags);
+
+	return agg_free;
+}
+
 static int rx_bottom(struct r8152 *tp, int budget)
 {
 	unsigned long flags;
@@ -1982,7 +2025,7 @@ static int rx_bottom(struct r8152 *tp, int budget)
 
 	list_for_each_safe(cursor, next, &rx_queue) {
 		struct rx_desc *rx_desc;
-		struct rx_agg *agg;
+		struct rx_agg *agg, *agg_free;
 		int len_used = 0;
 		struct urb *urb;
 		u8 *rx_data;
@@ -1994,6 +2037,8 @@ static int rx_bottom(struct r8152 *tp, int budget)
 		if (urb->actual_length < ETH_ZLEN)
 			goto submit;
 
+		agg_free = rtl_get_free_rx(tp, GFP_ATOMIC);
+
 		rx_desc = agg->buffer;
 		rx_data = agg->buffer;
 		len_used += sizeof(struct rx_desc);
@@ -2001,7 +2046,7 @@ static int rx_bottom(struct r8152 *tp, int budget)
 		while (urb->actual_length > len_used) {
 			struct net_device *netdev = tp->netdev;
 			struct net_device_stats *stats = &netdev->stats;
-			unsigned int pkt_len;
+			unsigned int pkt_len, rx_frag_head_sz;
 			struct sk_buff *skb;
 
 			/* limite the skb numbers for rx_queue */
@@ -2019,22 +2064,37 @@ static int rx_bottom(struct r8152 *tp, int budget)
 			pkt_len -= ETH_FCS_LEN;
 			rx_data += sizeof(struct rx_desc);
 
-			skb = napi_alloc_skb(napi, pkt_len);
+			if (!agg_free || RTL8152_RXFG_HEADSZ > pkt_len)
+				rx_frag_head_sz = pkt_len;
+			else
+				rx_frag_head_sz = RTL8152_RXFG_HEADSZ;
+
+			skb = napi_alloc_skb(napi, rx_frag_head_sz);
 			if (!skb) {
 				stats->rx_dropped++;
 				goto find_next_rx;
 			}
 
 			skb->ip_summed = r8152_rx_csum(tp, rx_desc);
-			memcpy(skb->data, rx_data, pkt_len);
-			skb_put(skb, pkt_len);
+			memcpy(skb->data, rx_data, rx_frag_head_sz);
+			skb_put(skb, rx_frag_head_sz);
+			pkt_len -= rx_frag_head_sz;
+			rx_data += rx_frag_head_sz;
+			if (pkt_len) {
+				skb_add_rx_frag(skb, 0, agg->page,
+						agg_offset(agg, rx_data),
+						pkt_len,
+						SKB_DATA_ALIGN(pkt_len));
+				get_page(agg->page);
+			}
+
 			skb->protocol = eth_type_trans(skb, netdev);
 			rtl_rx_vlan_tag(rx_desc, skb);
 			if (work_done < budget) {
 				napi_gro_receive(napi, skb);
 				work_done++;
 				stats->rx_packets++;
-				stats->rx_bytes += pkt_len;
+				stats->rx_bytes += skb->len;
 			} else {
 				__skb_queue_tail(&tp->rx_queue, skb);
 			}
@@ -2042,10 +2102,24 @@ static int rx_bottom(struct r8152 *tp, int budget)
 find_next_rx:
 			rx_data = rx_agg_align(rx_data + pkt_len + ETH_FCS_LEN);
 			rx_desc = (struct rx_desc *)rx_data;
-			len_used = (int)(rx_data - (u8 *)agg->buffer);
+			len_used = agg_offset(agg, rx_data);
 			len_used += sizeof(struct rx_desc);
 		}
 
+		WARN_ON(!agg_free && page_count(agg->page) > 1);
+
+		if (agg_free) {
+			spin_lock_irqsave(&tp->rx_lock, flags);
+			if (page_count(agg->page) == 1) {
+				list_add(&agg_free->list, &tp->rx_used);
+			} else {
+				list_add_tail(&agg->list, &tp->rx_used);
+				agg = agg_free;
+				urb = agg->urb;
+			}
+			spin_unlock_irqrestore(&tp->rx_lock, flags);
+		}
+
 submit:
 		if (!ret) {
 			ret = r8152_submit_rx(tp, agg, GFP_ATOMIC);
@@ -2373,13 +2447,14 @@ static int rtl_start_rx(struct r8152 *tp)
 	struct rx_agg *agg, *agg_next;
 	struct list_head tmp_list;
 	unsigned long flags;
-	int ret = 0;
+	int ret = 0, i = 0;
 
 	INIT_LIST_HEAD(&tmp_list);
 
 	spin_lock_irqsave(&tp->rx_lock, flags);
 
 	INIT_LIST_HEAD(&tp->rx_done);
+	INIT_LIST_HEAD(&tp->rx_used);
 
 	list_splice_init(&tp->rx_info, &tmp_list);
 
@@ -2388,10 +2463,18 @@ static int rtl_start_rx(struct r8152 *tp)
 	list_for_each_entry_safe(agg, agg_next, &tmp_list, info_list) {
 		INIT_LIST_HEAD(&agg->list);
 
-		if (ret < 0)
+		/* Only RTL8152_MAX_RX rx_agg need to be submitted. */
+		if (++i > RTL8152_MAX_RX) {
+			spin_lock_irqsave(&tp->rx_lock, flags);
+			list_add_tail(&agg->list, &tp->rx_used);
+			spin_unlock_irqrestore(&tp->rx_lock, flags);
+		} else if (unlikely(ret < 0)) {
+			spin_lock_irqsave(&tp->rx_lock, flags);
 			list_add_tail(&agg->list, &tp->rx_done);
-		else
+			spin_unlock_irqrestore(&tp->rx_lock, flags);
+		} else {
 			ret = r8152_submit_rx(tp, agg, GFP_KERNEL);
+		}
 	}
 
 	spin_lock_irqsave(&tp->rx_lock, flags);
@@ -2420,8 +2503,15 @@ static int rtl_stop_rx(struct r8152 *tp)
 	list_splice_init(&tp->rx_info, &tmp_list);
 	spin_unlock_irqrestore(&tp->rx_lock, flags);
 
-	list_for_each_entry_safe(agg, agg_next, &tmp_list, info_list)
-		usb_kill_urb(agg->urb);
+	list_for_each_entry_safe(agg, agg_next, &tmp_list, info_list) {
+		/* At least RTL8152_MAX_RX rx_agg have the page_count being
+		 * equal to 1, so the other ones could be freed safely.
+		 */
+		if (page_count(agg->page) > 1)
+			free_rx_agg(tp, agg);
+		else
+			usb_kill_urb(agg->urb);
+	}
 
 	/* Move back the list of temp to the rx_info */
 	spin_lock_irqsave(&tp->rx_lock, flags);
-- 
2.21.0

[PATCH net-next v2 5/5] r8152: change rx_copybreak and rx_pending through ethtool

[Hayes Wang]

Let the rx_copybreak and rx_pending could be modified by
ethtool.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
---
 drivers/net/usb/r8152.c | 91 ++++++++++++++++++++++++++++++++++++++---
 1 file changed, 86 insertions(+), 5 deletions(-)

diff --git a/drivers/net/usb/r8152.c b/drivers/net/usb/r8152.c
index 2ae04522cd5a..40d18e866269 100644
--- a/drivers/net/usb/r8152.c
+++ b/drivers/net/usb/r8152.c
@@ -26,7 +26,7 @@
 #include <linux/acpi.h>
 
 /* Information for net-next */
-#define NETNEXT_VERSION		"09"
+#define NETNEXT_VERSION		"10"
 
 /* Information for net */
 #define NET_VERSION		"10"
@@ -584,7 +584,7 @@ enum rtl_register_content {
 #define TX_ALIGN		4
 #define RX_ALIGN		8
 
-#define RTL8152_MAX_RX_AGG	(10 * RTL8152_MAX_RX)
+#define RTL8152_RX_MAX_PENDING	4096
 #define RTL8152_RXFG_HEADSZ	256
 
 #define INTR_LINK		0x0004
@@ -756,6 +756,9 @@ struct r8152 {
 	u32 tx_qlen;
 	u32 coalesce;
 	u32 rx_buf_sz;
+	u32 rx_copybreak;
+	u32 rx_pending;
+
 	u16 ocp_base;
 	u16 speed;
 	u8 *intr_buff;
@@ -1984,7 +1987,7 @@ static struct rx_agg *rtl_get_free_rx(struct r8152 *tp, gfp_t mflags)
 
 	spin_unlock_irqrestore(&tp->rx_lock, flags);
 
-	if (!agg_free && atomic_read(&tp->rx_count) < RTL8152_MAX_RX_AGG)
+	if (!agg_free && atomic_read(&tp->rx_count) < tp->rx_pending)
 		agg_free = alloc_rx_agg(tp, mflags);
 
 	return agg_free;
@@ -2064,10 +2067,10 @@ static int rx_bottom(struct r8152 *tp, int budget)
 			pkt_len -= ETH_FCS_LEN;
 			rx_data += sizeof(struct rx_desc);
 
-			if (!agg_free || RTL8152_RXFG_HEADSZ > pkt_len)
+			if (!agg_free || tp->rx_copybreak > pkt_len)
 				rx_frag_head_sz = pkt_len;
 			else
-				rx_frag_head_sz = RTL8152_RXFG_HEADSZ;
+				rx_frag_head_sz = tp->rx_copybreak;
 
 			skb = napi_alloc_skb(napi, rx_frag_head_sz);
 			if (!skb) {
@@ -5104,6 +5107,77 @@ static int rtl8152_set_coalesce(struct net_device *netdev,
 	return ret;
 }
 
+static int rtl8152_get_tunable(struct net_device *netdev,
+			       const struct ethtool_tunable *tunable, void *d)
+{
+	struct r8152 *tp = netdev_priv(netdev);
+
+	switch (tunable->id) {
+	case ETHTOOL_RX_COPYBREAK:
+		*(u32 *)d = tp->rx_copybreak;
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+
+	return 0;
+}
+
+static int rtl8152_set_tunable(struct net_device *netdev,
+			       const struct ethtool_tunable *tunable,
+			       const void *d)
+{
+	struct r8152 *tp = netdev_priv(netdev);
+	u32 val;
+
+	switch (tunable->id) {
+	case ETHTOOL_RX_COPYBREAK:
+		val = *(u32 *)d;
+		if (val < ETH_ZLEN) {
+			netif_err(tp, rx_err, netdev,
+				  "Invalid rx copy break value\n");
+			return -EINVAL;
+		}
+
+		if (tp->rx_copybreak != val) {
+			napi_disable(&tp->napi);
+			tp->rx_copybreak = val;
+			napi_enable(&tp->napi);
+		}
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+
+	return 0;
+}
+
+static void rtl8152_get_ringparam(struct net_device *netdev,
+				  struct ethtool_ringparam *ring)
+{
+	struct r8152 *tp = netdev_priv(netdev);
+
+	ring->rx_max_pending = RTL8152_RX_MAX_PENDING;
+	ring->rx_pending = tp->rx_pending;
+}
+
+static int rtl8152_set_ringparam(struct net_device *netdev,
+				 struct ethtool_ringparam *ring)
+{
+	struct r8152 *tp = netdev_priv(netdev);
+
+	if (ring->rx_pending < (RTL8152_MAX_RX * 2))
+		return -EINVAL;
+
+	if (tp->rx_pending != ring->rx_pending) {
+		napi_disable(&tp->napi);
+		tp->rx_pending = ring->rx_pending;
+		napi_enable(&tp->napi);
+	}
+
+	return 0;
+}
+
 static const struct ethtool_ops ops = {
 	.get_drvinfo = rtl8152_get_drvinfo,
 	.get_link = ethtool_op_get_link,
@@ -5121,6 +5195,10 @@ static const struct ethtool_ops ops = {
 	.set_eee = rtl_ethtool_set_eee,
 	.get_link_ksettings = rtl8152_get_link_ksettings,
 	.set_link_ksettings = rtl8152_set_link_ksettings,
+	.get_tunable = rtl8152_get_tunable,
+	.set_tunable = rtl8152_set_tunable,
+	.get_ringparam = rtl8152_get_ringparam,
+	.set_ringparam = rtl8152_set_ringparam,
 };
 
 static int rtl8152_ioctl(struct net_device *netdev, struct ifreq *rq, int cmd)
@@ -5474,6 +5552,9 @@ static int rtl8152_probe(struct usb_interface *intf,
 	tp->speed = tp->mii.supports_gmii ? SPEED_1000 : SPEED_100;
 	tp->duplex = DUPLEX_FULL;
 
+	tp->rx_copybreak = RTL8152_RXFG_HEADSZ;
+	tp->rx_pending = 10 * RTL8152_MAX_RX;
+
 	intf->needs_remote_wakeup = 1;
 
 	tp->rtl_ops.init(tp);
-- 
2.21.0

Re: [PATCH net-next v2 0/5] r8152: RX improve

[Jakub Kicinski]

On Tue, 13 Aug 2019 11:42:04 +0800, Hayes Wang wrote:
> v2:
> For patch #2, replace list_for_each_safe with list_for_each_entry_safe.
> Remove unlikely in WARN_ON. Adjust the coding style.
> 
> For patch #4, replace list_for_each_safe with list_for_each_entry_safe.
> Remove "else" after "continue".
> 
> For patch #5. replace sysfs with ethtool to modify rx_copybreak and
> rx_pending.
> 
> v1:
> The different chips use different rx buffer size.
> 
> Use skb_add_rx_frag() to reduce memory copy for RX.

Applied, thank you.

Re: [PATCH net-next v2 4/5] r8152: support skb_add_rx_frag

[Eric Dumazet]

On 8/13/19 5:42 AM, Hayes Wang wrote:
> Use skb_add_rx_frag() to reduce the memory copy for rx data.
> 
> Use a new list of rx_used to store the rx buffer which couldn't be
> reused yet.
> 
> Besides, the total number of rx buffer may be increased or decreased
> dynamically. And it is limited by RTL8152_MAX_RX_AGG.
> 
> Signed-off-by: Hayes Wang <hayeswang@realtek.com>
>

...

>  			skb->protocol = eth_type_trans(skb, netdev);
>  			rtl_rx_vlan_tag(rx_desc, skb);
>  			if (work_done < budget) {
>  				napi_gro_receive(napi, skb);
>  				work_done++;
>  				stats->rx_packets++;
> -				stats->rx_bytes += pkt_len;
> +				stats->rx_bytes += skb->len;

use-after-free. skb is no longer in your hands after napi_gro_receive()

[PATCH net-next] r8152: fix accessing skb after napi_gro_receive

[Hayes Wang]

Fix accessing skb after napi_gro_receive which is caused by
commit 47922fcde536 ("r8152: support skb_add_rx_frag").

Fixes: 47922fcde536 ("r8152: support skb_add_rx_frag")
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
---
 drivers/net/usb/r8152.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/usb/r8152.c b/drivers/net/usb/r8152.c
index 40d18e866269..b1db6df6f4ab 100644
--- a/drivers/net/usb/r8152.c
+++ b/drivers/net/usb/r8152.c
@@ -2094,10 +2094,10 @@ static int rx_bottom(struct r8152 *tp, int budget)
 			skb->protocol = eth_type_trans(skb, netdev);
 			rtl_rx_vlan_tag(rx_desc, skb);
 			if (work_done < budget) {
-				napi_gro_receive(napi, skb);
 				work_done++;
 				stats->rx_packets++;
 				stats->rx_bytes += skb->len;
+				napi_gro_receive(napi, skb);
 			} else {
 				__skb_queue_tail(&tp->rx_queue, skb);
 			}
-- 
2.21.0

Re: [PATCH net-next] r8152: fix accessing skb after napi_gro_receive

[David Miller]

From: Hayes Wang <hayeswang@realtek.com>
Date: Mon, 19 Aug 2019 11:15:19 +0800

> Fix accessing skb after napi_gro_receive which is caused by
> commit 47922fcde536 ("r8152: support skb_add_rx_frag").
> 
> Fixes: 47922fcde536 ("r8152: support skb_add_rx_frag")
> Signed-off-by: Hayes Wang <hayeswang@realtek.com>

Applied, thanks.

Re: [PATCH net-next] r8152: fix accessing skb after napi_gro_receive

[Eric Dumazet]

On 8/19/19 5:15 AM, Hayes Wang wrote:
> Fix accessing skb after napi_gro_receive which is caused by
> commit 47922fcde536 ("r8152: support skb_add_rx_frag").
> 
> Fixes: 47922fcde536 ("r8152: support skb_add_rx_frag")
> Signed-off-by: Hayes Wang <hayeswang@realtek.com>
> ---

It is customary to add a tag to credit the reporter...

Something like :

Reported-by: ....

Thanks.

RE: [PATCH net-next] r8152: fix accessing skb after napi_gro_receive

[Hayes Wang]

Eric Dumazet [mailto:eric.dumazet@gmail.com]
> Sent: Friday, August 30, 2019 12:32 AM
> To: Hayes Wang; netdev@vger.kernel.org
> Cc: nic_swsd; linux-kernel@vger.kernel.org
> Subject: Re: [PATCH net-next] r8152: fix accessing skb after napi_gro_receive
> 
> On 8/19/19 5:15 AM, Hayes Wang wrote:
> > Fix accessing skb after napi_gro_receive which is caused by
> > commit 47922fcde536 ("r8152: support skb_add_rx_frag").
> >
> > Fixes: 47922fcde536 ("r8152: support skb_add_rx_frag")
> > Signed-off-by: Hayes Wang <hayeswang@realtek.com>
> > ---
> 
> It is customary to add a tag to credit the reporter...
> 
> Something like :
> 
> Reported-by: ....
> 
> Thanks.

Sorry. It's my mistake.
I would note that next time.

Best Regards,
Hayes