[PATCH net-next 00/12] cleanup & optimizations & bugfixes for HNS3 driver

[PATCH net-next 00/12] cleanup & optimizations & bugfixes for HNS3 driver

[Huazhong Tan]

This patchset contains some cleanup related to hns3_enet_ring
struct and tx bd filling process, optimizations related
to rx page reusing, barrier using and tso handling process,
bugfixes related to tunnel type handling and error handling for
desc filling.

Yunsheng Lin (12):
  net: hns3: unify maybe_stop_tx for TSO and non-TSO case
  net: hns3: use napi_schedule_irqoff in hard interrupts handlers
  net: hns3: add counter for times RX pages gets allocated
  net: hns3: add linearizing checking for TSO case
  net: hns3: fix for tunnel type handling in hns3_rx_checksum
  net: hns3: refactor BD filling for l2l3l4 info
  net: hns3: combine len and checksum handling for inner and outer
    header.
  net: hns3: fix error handling for desc filling
  net: hns3: optimize the barrier using when cleaning TX BD
  net: hns3: unify the page reusing for page size 4K and 64K
  net: hns3: some cleanup for struct hns3_enet_ring
  net: hns3: use devm_kcalloc when allocating desc_cb

 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c    | 499 ++++++++++-----------
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.h    |  17 +-
 drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c |   2 +
 3 files changed, 230 insertions(+), 288 deletions(-)

-- 
2.7.4

[PATCH net-next 01/12] net: hns3: unify maybe_stop_tx for TSO and non-TSO case

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

Currently, maybe_stop_tx ops for TSO and non-TSO case share some BD
calculation code, so this patch unifies the maybe_stop_tx by removing
the maybe_stop_tx ops. skb_is_gso() can be used to differentiate the
case between TSO and non-TSO case if there is need to handle special
case for TSO case.

This patch also add tx_copy field in "ethtool --statistics" to help
better debug the performance issue caused by calling skb_copy.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c    | 128 ++++++++-------------
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.h    |   7 +-
 drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c |   1 +
 3 files changed, 50 insertions(+), 86 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 96272e6..06dda77 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -1154,64 +1154,48 @@ static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
 	return 0;
 }
 
-static int hns3_nic_maybe_stop_tso(struct sk_buff **out_skb, int *bnum,
-				   struct hns3_enet_ring *ring)
+static int hns3_nic_bd_num(struct sk_buff *skb)
 {
-	struct sk_buff *skb = *out_skb;
-	struct sk_buff *new_skb = NULL;
-	struct skb_frag_struct *frag;
-	int bdnum_for_frag;
-	int frag_num;
-	int buf_num;
-	int size;
-	int i;
+	int size = skb_headlen(skb);
+	int i, bd_num;
 
-	size = skb_headlen(skb);
-	buf_num = hns3_tx_bd_count(size);
+	/* if the total len is within the max bd limit */
+	if (likely(skb->len <= HNS3_MAX_BD_SIZE))
+		return skb_shinfo(skb)->nr_frags + 1;
 
-	frag_num = skb_shinfo(skb)->nr_frags;
-	for (i = 0; i < frag_num; i++) {
-		frag = &skb_shinfo(skb)->frags[i];
-		size = skb_frag_size(frag);
-		bdnum_for_frag = hns3_tx_bd_count(size);
-		if (unlikely(bdnum_for_frag > HNS3_MAX_BD_PER_FRAG))
-			return -ENOMEM;
+	bd_num = hns3_tx_bd_count(size);
 
-		buf_num += bdnum_for_frag;
-	}
+	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
+		int frag_bd_num;
 
-	if (unlikely(buf_num > HNS3_MAX_BD_PER_FRAG)) {
-		buf_num = hns3_tx_bd_count(skb->len);
-		if (ring_space(ring) < buf_num)
-			return -EBUSY;
-		/* manual split the send packet */
-		new_skb = skb_copy(skb, GFP_ATOMIC);
-		if (!new_skb)
+		size = skb_frag_size(frag);
+		frag_bd_num = hns3_tx_bd_count(size);
+
+		if (unlikely(frag_bd_num > HNS3_MAX_BD_PER_FRAG))
 			return -ENOMEM;
-		dev_kfree_skb_any(skb);
-		*out_skb = new_skb;
-	}
 
-	if (unlikely(ring_space(ring) < buf_num))
-		return -EBUSY;
+		bd_num += frag_bd_num;
+	}
 
-	*bnum = buf_num;
-	return 0;
+	return bd_num;
 }
 
-static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum,
-				  struct hns3_enet_ring *ring)
+static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
+				  struct sk_buff **out_skb)
 {
 	struct sk_buff *skb = *out_skb;
-	struct sk_buff *new_skb = NULL;
-	int buf_num;
+	int bd_num;
 
-	/* No. of segments (plus a header) */
-	buf_num = skb_shinfo(skb)->nr_frags + 1;
+	bd_num = hns3_nic_bd_num(skb);
+	if (bd_num < 0)
+		return bd_num;
+
+	if (unlikely(bd_num > HNS3_MAX_BD_PER_FRAG)) {
+		struct sk_buff *new_skb;
 
-	if (unlikely(buf_num > HNS3_MAX_BD_PER_FRAG)) {
-		buf_num = hns3_tx_bd_count(skb->len);
-		if (ring_space(ring) < buf_num)
+		bd_num = hns3_tx_bd_count(skb->len);
+		if (unlikely(ring_space(ring) < bd_num))
 			return -EBUSY;
 		/* manual split the send packet */
 		new_skb = skb_copy(skb, GFP_ATOMIC);
@@ -1219,14 +1203,16 @@ static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum,
 			return -ENOMEM;
 		dev_kfree_skb_any(skb);
 		*out_skb = new_skb;
+
+		u64_stats_update_begin(&ring->syncp);
+		ring->stats.tx_copy++;
+		u64_stats_update_end(&ring->syncp);
 	}
 
-	if (unlikely(ring_space(ring) < buf_num))
+	if (unlikely(ring_space(ring) < bd_num))
 		return -EBUSY;
 
-	*bnum = buf_num;
-
-	return 0;
+	return bd_num;
 }
 
 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
@@ -1277,22 +1263,23 @@ netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
 	/* Prefetch the data used later */
 	prefetch(skb->data);
 
-	switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
-	case -EBUSY:
-		u64_stats_update_begin(&ring->syncp);
-		ring->stats.tx_busy++;
-		u64_stats_update_end(&ring->syncp);
+	buf_num = hns3_nic_maybe_stop_tx(ring, &skb);
+	if (unlikely(buf_num <= 0)) {
+		if (buf_num == -EBUSY) {
+			u64_stats_update_begin(&ring->syncp);
+			ring->stats.tx_busy++;
+			u64_stats_update_end(&ring->syncp);
+			goto out_net_tx_busy;
+		} else if (buf_num == -ENOMEM) {
+			u64_stats_update_begin(&ring->syncp);
+			ring->stats.sw_err_cnt++;
+			u64_stats_update_end(&ring->syncp);
+		}
 
-		goto out_net_tx_busy;
-	case -ENOMEM:
-		u64_stats_update_begin(&ring->syncp);
-		ring->stats.sw_err_cnt++;
-		u64_stats_update_end(&ring->syncp);
-		netdev_err(netdev, "no memory to xmit!\n");
+		if (net_ratelimit())
+			netdev_err(netdev, "xmit error: %d!\n", buf_num);
 
 		goto out_err_tx_ok;
-	default:
-		break;
 	}
 
 	/* No. of segments (plus a header) */
@@ -1397,13 +1384,6 @@ static int hns3_nic_set_features(struct net_device *netdev,
 	bool enable;
 	int ret;
 
-	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
-		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
-			priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
-		else
-			priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
-	}
-
 	if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
 		enable = !!(features & NETIF_F_GRO_HW);
 		ret = h->ae_algo->ops->set_gro_en(h, enable);
@@ -3733,17 +3713,6 @@ static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
 		h->ae_algo->ops->del_all_fd_entries(h, clear_list);
 }
 
-static void hns3_nic_set_priv_ops(struct net_device *netdev)
-{
-	struct hns3_nic_priv *priv = netdev_priv(netdev);
-
-	if ((netdev->features & NETIF_F_TSO) ||
-	    (netdev->features & NETIF_F_TSO6))
-		priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
-	else
-		priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
-}
-
 static int hns3_client_start(struct hnae3_handle *handle)
 {
 	if (!handle->ae_algo->ops->client_start)
@@ -3810,7 +3779,6 @@ static int hns3_client_init(struct hnae3_handle *handle)
 	netdev->netdev_ops = &hns3_nic_netdev_ops;
 	SET_NETDEV_DEV(netdev, &pdev->dev);
 	hns3_ethtool_set_ops(netdev);
-	hns3_nic_set_priv_ops(netdev);
 
 	/* Carrier off reporting is important to ethtool even BEFORE open */
 	netif_carrier_off(netdev);
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
index 2b4f5ea..f669412 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
@@ -376,6 +376,7 @@ struct ring_stats {
 			u64 tx_err_cnt;
 			u64 restart_queue;
 			u64 tx_busy;
+			u64 tx_copy;
 		};
 		struct {
 			u64 rx_pkts;
@@ -444,11 +445,6 @@ struct hns3_nic_ring_data {
 	void (*fini_process)(struct hns3_nic_ring_data *);
 };
 
-struct hns3_nic_ops {
-	int (*maybe_stop_tx)(struct sk_buff **out_skb,
-			     int *bnum, struct hns3_enet_ring *ring);
-};
-
 enum hns3_flow_level_range {
 	HNS3_FLOW_LOW = 0,
 	HNS3_FLOW_MID = 1,
@@ -538,7 +534,6 @@ struct hns3_nic_priv {
 	u32 port_id;
 	struct net_device *netdev;
 	struct device *dev;
-	struct hns3_nic_ops ops;
 
 	/**
 	 * the cb for nic to manage the ring buffer, the first half of the
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
index 1746943..7256ed4 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
@@ -29,6 +29,7 @@ static const struct hns3_stats hns3_txq_stats[] = {
 	HNS3_TQP_STAT("errors", tx_err_cnt),
 	HNS3_TQP_STAT("wake", restart_queue),
 	HNS3_TQP_STAT("busy", tx_busy),
+	HNS3_TQP_STAT("copy", tx_copy),
 };
 
 #define HNS3_TXQ_STATS_COUNT ARRAY_SIZE(hns3_txq_stats)
-- 
2.7.4

[PATCH net-next 02/12] net: hns3: use napi_schedule_irqoff in hard interrupts handlers

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

napi_schedule_irqoff is introduced to be used from hard interrupts
handlers or when irqs are already masked, see:

https://lists.openwall.net/netdev/2014/10/29/2

So this patch replaces napi_schedule with napi_schedule_irqoff.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 06dda77..ff3c9f1 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -74,7 +74,7 @@ static irqreturn_t hns3_irq_handle(int irq, void *vector)
 {
 	struct hns3_enet_tqp_vector *tqp_vector = vector;
 
-	napi_schedule(&tqp_vector->napi);
+	napi_schedule_irqoff(&tqp_vector->napi);
 
 	return IRQ_HANDLED;
 }
-- 
2.7.4

[PATCH net-next 03/12] net: hns3: add counter for times RX pages gets allocated

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

Currently, using "ethtool --statistics" can show how many time RX
page have been reused, but there is no counter for RX page not
being reused.

This patch adds non_reuse_pg counter to better debug the performance
issue, because it is hard to determine when the RX page is reused
or not if there is no such counter.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c    | 4 ++++
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.h    | 1 +
 drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c | 1 +
 3 files changed, 6 insertions(+)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index ff3c9f1..453fdf4 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -2315,6 +2315,10 @@ hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring, int cleand_count)
 				break;
 			}
 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
+
+			u64_stats_update_begin(&ring->syncp);
+			ring->stats.non_reuse_pg++;
+			u64_stats_update_end(&ring->syncp);
 		}
 
 		ring_ptr_move_fw(ring, next_to_use);
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
index f669412..9680a68 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
@@ -389,6 +389,7 @@ struct ring_stats {
 			u64 l2_err;
 			u64 l3l4_csum_err;
 			u64 rx_multicast;
+			u64 non_reuse_pg;
 		};
 	};
 };
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
index 7256ed4..d1588ea 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_ethtool.c
@@ -49,6 +49,7 @@ static const struct hns3_stats hns3_rxq_stats[] = {
 	HNS3_TQP_STAT("l2_err", l2_err),
 	HNS3_TQP_STAT("l3l4_csum_err", l3l4_csum_err),
 	HNS3_TQP_STAT("multicast", rx_multicast),
+	HNS3_TQP_STAT("non_reuse_pg", non_reuse_pg),
 };
 
 #define HNS3_RXQ_STATS_COUNT ARRAY_SIZE(hns3_rxq_stats)
-- 
2.7.4

[PATCH net-next 04/12] net: hns3: add linearizing checking for TSO case

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

HW requires every continuous 8 buffer data to be larger than MSS,
we simplify it by ensuring skb_headlen + the first continuous
7 frags to to be larger than GSO header len + mss, and the
remaining continuous 7 frags to be larger than MSS except the
last 7 frags.

This patch adds hns3_skb_need_linearized to handle it for TSO
case.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 45 +++++++++++++++++++++++++
 1 file changed, 45 insertions(+)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 453fdf4..944e0ae 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -1181,6 +1181,47 @@ static int hns3_nic_bd_num(struct sk_buff *skb)
 	return bd_num;
 }
 
+static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
+{
+	if (!skb->encapsulation)
+		return skb_transport_offset(skb) + tcp_hdrlen(skb);
+
+	return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
+}
+
+/* HW need every continuous 8 buffer data to be larger than MSS,
+ * we simplify it by ensuring skb_headlen + the first continuous
+ * 7 frags to to be larger than gso header len + mss, and the remaining
+ * continuous 7 frags to be larger than MSS except the last 7 frags.
+ */
+static bool hns3_skb_need_linearized(struct sk_buff *skb)
+{
+	int bd_limit = HNS3_MAX_BD_PER_FRAG - 1;
+	unsigned int tot_len = 0;
+	int i;
+
+	for (i = 0; i < bd_limit; i++)
+		tot_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
+
+	/* ensure headlen + the first 7 frags is greater than mss + header
+	 * and the first 7 frags is greater than mss.
+	 */
+	if (((tot_len + skb_headlen(skb)) < (skb_shinfo(skb)->gso_size +
+	    hns3_gso_hdr_len(skb))) || (tot_len < skb_shinfo(skb)->gso_size))
+		return true;
+
+	/* ensure the remaining continuous 7 buffer is greater than mss */
+	for (i = 0; i < (skb_shinfo(skb)->nr_frags - bd_limit - 1); i++) {
+		tot_len -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
+		tot_len += skb_frag_size(&skb_shinfo(skb)->frags[i + bd_limit]);
+
+		if (tot_len < skb_shinfo(skb)->gso_size)
+			return true;
+	}
+
+	return false;
+}
+
 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
 				  struct sk_buff **out_skb)
 {
@@ -1194,6 +1235,9 @@ static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
 	if (unlikely(bd_num > HNS3_MAX_BD_PER_FRAG)) {
 		struct sk_buff *new_skb;
 
+		if (skb_is_gso(skb) && !hns3_skb_need_linearized(skb))
+			goto out;
+
 		bd_num = hns3_tx_bd_count(skb->len);
 		if (unlikely(ring_space(ring) < bd_num))
 			return -EBUSY;
@@ -1209,6 +1253,7 @@ static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
 		u64_stats_update_end(&ring->syncp);
 	}
 
+out:
 	if (unlikely(ring_space(ring) < bd_num))
 		return -EBUSY;
 
-- 
2.7.4

[PATCH net-next 05/12] net: hns3: fix for tunnel type handling in hns3_rx_checksum

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

According to hardware user manual, the tunnel packet type is
available in the rx.ol_info field of struct hns3_desc. Currently
the tunnel packet type is decided by the rx.l234_info, which may
cause RX checksum handling error.

This patch fixes it by using the correct field in struct hns3_desc
to decide the tunnel packet type.

Fixes: 76ad4f0ee747 ("net: hns3: Add support of HNS3 Ethernet Driver for hip08 SoC")
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 944e0ae..14312ff 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -2463,7 +2463,7 @@ static int hns3_gro_complete(struct sk_buff *skb)
 }
 
 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
-			     u32 l234info, u32 bd_base_info)
+			     u32 l234info, u32 bd_base_info, u32 ol_info)
 {
 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
 	int l3_type, l4_type;
@@ -2490,7 +2490,7 @@ static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
 		return;
 	}
 
-	ol4_type = hnae3_get_field(l234info, HNS3_RXD_OL4ID_M,
+	ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
 				   HNS3_RXD_OL4ID_S);
 	switch (ol4_type) {
 	case HNS3_OL4_TYPE_MAC_IN_UDP:
@@ -2694,7 +2694,7 @@ static int hns3_add_frag(struct hns3_enet_ring *ring, struct hns3_desc *desc,
 
 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
 				     struct sk_buff *skb, u32 l234info,
-				     u32 bd_base_info)
+				     u32 bd_base_info, u32 ol_info)
 {
 	u16 gro_count;
 	u32 l3_type;
@@ -2703,7 +2703,7 @@ static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
 				    HNS3_RXD_GRO_COUNT_S);
 	/* if there is no HW GRO, do not set gro params */
 	if (!gro_count) {
-		hns3_rx_checksum(ring, skb, l234info, bd_base_info);
+		hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info);
 		return 0;
 	}
 
@@ -2743,7 +2743,7 @@ static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
 {
 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
 	enum hns3_pkt_l2t_type l2_frame_type;
-	u32 bd_base_info, l234info;
+	u32 bd_base_info, l234info, ol_info;
 	struct hns3_desc *desc;
 	unsigned int len;
 	int pre_ntc, ret;
@@ -2757,6 +2757,7 @@ static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
 	desc = &ring->desc[pre_ntc];
 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
 	l234info = le32_to_cpu(desc->rx.l234_info);
+	ol_info = le32_to_cpu(desc->rx.ol_info);
 
 	/* Based on hw strategy, the tag offloaded will be stored at
 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
@@ -2796,7 +2797,8 @@ static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
 	skb->protocol = eth_type_trans(skb, netdev);
 
 	/* This is needed in order to enable forwarding support */
-	ret = hns3_set_gro_and_checksum(ring, skb, l234info, bd_base_info);
+	ret = hns3_set_gro_and_checksum(ring, skb, l234info,
+					bd_base_info, ol_info);
 	if (unlikely(ret)) {
 		u64_stats_update_begin(&ring->syncp);
 		ring->stats.rx_err_cnt++;
-- 
2.7.4

[PATCH net-next 06/12] net: hns3: refactor BD filling for l2l3l4 info

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

This patch separates the inner and outer l2l3l4 len handling in
hns3_set_l2l3l4_len, this is a preparation to combine the l2l3l4
len and checksum handling for inner and outer header.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 62 +++++++++----------------
 1 file changed, 23 insertions(+), 39 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 14312ff..9170743 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -741,65 +741,49 @@ static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto,
 				u8 il4_proto, u32 *type_cs_vlan_tso,
 				u32 *ol_type_vlan_len_msec)
 {
+	unsigned char *l2_hdr = skb->data;
+	u8 l4_proto = ol4_proto;
 	union l3_hdr_info l3;
 	union l4_hdr_info l4;
-	unsigned char *l2_hdr;
-	u8 l4_proto = ol4_proto;
-	u32 ol2_len;
-	u32 ol3_len;
-	u32 ol4_len;
 	u32 l2_len;
 	u32 l3_len;
+	u32 l4_len;
 
 	l3.hdr = skb_network_header(skb);
 	l4.hdr = skb_transport_header(skb);
 
-	/* compute L2 header size for normal packet, defined in 2 Bytes */
-	l2_len = l3.hdr - skb->data;
-	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
-
-	/* tunnel packet*/
+	/* tunnel packet */
 	if (skb->encapsulation) {
+		/* not MAC in UDP, MAC in GRE (0x6558) */
+		if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE))
+			return;
+
 		/* compute OL2 header size, defined in 2 Bytes */
-		ol2_len = l2_len;
+		l2_len = l3.hdr - skb->data;
 		hns3_set_field(*ol_type_vlan_len_msec,
-			       HNS3_TXD_L2LEN_S, ol2_len >> 1);
+			       HNS3_TXD_L2LEN_S, l2_len >> 1);
 
 		/* compute OL3 header size, defined in 4 Bytes */
-		ol3_len = l4.hdr - l3.hdr;
+		l3_len = l4.hdr - l3.hdr;
 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S,
-			       ol3_len >> 2);
-
-		/* MAC in UDP, MAC in GRE (0x6558)*/
-		if ((ol4_proto == IPPROTO_UDP) || (ol4_proto == IPPROTO_GRE)) {
-			/* switch MAC header ptr from outer to inner header.*/
-			l2_hdr = skb_inner_mac_header(skb);
-
-			/* compute OL4 header size, defined in 4 Bytes. */
-			ol4_len = l2_hdr - l4.hdr;
-			hns3_set_field(*ol_type_vlan_len_msec,
-				       HNS3_TXD_L4LEN_S, ol4_len >> 2);
+			       l3_len >> 2);
 
-			/* switch IP header ptr from outer to inner header */
-			l3.hdr = skb_inner_network_header(skb);
+		l2_hdr = skb_inner_mac_header(skb);
+		/* compute OL4 header size, defined in 4 Bytes. */
+		l4_len = l2_hdr - l4.hdr;
+		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S,
+			       l4_len >> 2);
 
-			/* compute inner l2 header size, defined in 2 Bytes. */
-			l2_len = l3.hdr - l2_hdr;
-			hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S,
-				       l2_len >> 1);
-		} else {
-			/* skb packet types not supported by hardware,
-			 * txbd len fild doesn't be filled.
-			 */
-			return;
-		}
-
-		/* switch L4 header pointer from outer to inner */
+		/* switch to inner header */
+		l2_hdr = skb_inner_mac_header(skb);
+		l3.hdr = skb_inner_network_header(skb);
 		l4.hdr = skb_inner_transport_header(skb);
-
 		l4_proto = il4_proto;
 	}
 
+	l2_len = l3.hdr - l2_hdr;
+	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
+
 	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
 	l3_len = l4.hdr - l3.hdr;
 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
-- 
2.7.4

[PATCH net-next 07/12] net: hns3: combine len and checksum handling for inner and outer header.

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

When filling len and checksum info to description, there is some
similar checking or calculation.

So this patch adds hns3_set_l2l3l4 to fill the inner(/normal)
header's len and checksum info. If it is a encapsulation skb, it
calls hns3_set_outer_l2l3l4 to handle the outer header's len and
checksum info, in order to avoid some similar checking or
calculation.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 192 ++++++++++--------------
 1 file changed, 81 insertions(+), 111 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 9170743..7224b38 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -737,79 +737,6 @@ static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
 	return 0;
 }
 
-static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto,
-				u8 il4_proto, u32 *type_cs_vlan_tso,
-				u32 *ol_type_vlan_len_msec)
-{
-	unsigned char *l2_hdr = skb->data;
-	u8 l4_proto = ol4_proto;
-	union l3_hdr_info l3;
-	union l4_hdr_info l4;
-	u32 l2_len;
-	u32 l3_len;
-	u32 l4_len;
-
-	l3.hdr = skb_network_header(skb);
-	l4.hdr = skb_transport_header(skb);
-
-	/* tunnel packet */
-	if (skb->encapsulation) {
-		/* not MAC in UDP, MAC in GRE (0x6558) */
-		if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE))
-			return;
-
-		/* compute OL2 header size, defined in 2 Bytes */
-		l2_len = l3.hdr - skb->data;
-		hns3_set_field(*ol_type_vlan_len_msec,
-			       HNS3_TXD_L2LEN_S, l2_len >> 1);
-
-		/* compute OL3 header size, defined in 4 Bytes */
-		l3_len = l4.hdr - l3.hdr;
-		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S,
-			       l3_len >> 2);
-
-		l2_hdr = skb_inner_mac_header(skb);
-		/* compute OL4 header size, defined in 4 Bytes. */
-		l4_len = l2_hdr - l4.hdr;
-		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S,
-			       l4_len >> 2);
-
-		/* switch to inner header */
-		l2_hdr = skb_inner_mac_header(skb);
-		l3.hdr = skb_inner_network_header(skb);
-		l4.hdr = skb_inner_transport_header(skb);
-		l4_proto = il4_proto;
-	}
-
-	l2_len = l3.hdr - l2_hdr;
-	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
-
-	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
-	l3_len = l4.hdr - l3.hdr;
-	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
-
-	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
-	switch (l4_proto) {
-	case IPPROTO_TCP:
-		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
-			       l4.tcp->doff);
-		break;
-	case IPPROTO_SCTP:
-		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
-			       (sizeof(struct sctphdr) >> 2));
-		break;
-	case IPPROTO_UDP:
-		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
-			       (sizeof(struct udphdr) >> 2));
-		break;
-	default:
-		/* skb packet types not supported by hardware,
-		 * txbd len fild doesn't be filled.
-		 */
-		return;
-	}
-}
-
 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
  * and it is udp packet, which has a dest port as the IANA assigned.
  * the hardware is expected to do the checksum offload, but the
@@ -831,46 +758,71 @@ static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
 	return true;
 }
 
-static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
-				   u8 il4_proto, u32 *type_cs_vlan_tso,
-				   u32 *ol_type_vlan_len_msec)
+static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
+				  u32 *ol_type_vlan_len_msec)
 {
+	u32 l2_len, l3_len, l4_len;
+	unsigned char *il2_hdr;
 	union l3_hdr_info l3;
-	u32 l4_proto = ol4_proto;
+	union l4_hdr_info l4;
 
 	l3.hdr = skb_network_header(skb);
+	l4.hdr = skb_transport_header(skb);
 
-	/* define OL3 type and tunnel type(OL4).*/
-	if (skb->encapsulation) {
-		/* define outer network header type.*/
-		if (skb->protocol == htons(ETH_P_IP)) {
-			if (skb_is_gso(skb))
-				hns3_set_field(*ol_type_vlan_len_msec,
-					       HNS3_TXD_OL3T_S,
-					       HNS3_OL3T_IPV4_CSUM);
-			else
-				hns3_set_field(*ol_type_vlan_len_msec,
-					       HNS3_TXD_OL3T_S,
-					       HNS3_OL3T_IPV4_NO_CSUM);
-
-		} else if (skb->protocol == htons(ETH_P_IPV6)) {
-			hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
-				       HNS3_OL3T_IPV6);
-		}
+	/* compute OL2 header size, defined in 2 Bytes */
+	l2_len = l3.hdr - skb->data;
+	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
+
+	/* compute OL3 header size, defined in 4 Bytes */
+	l3_len = l4.hdr - l3.hdr;
+	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
 
-		/* define tunnel type(OL4).*/
-		switch (l4_proto) {
-		case IPPROTO_UDP:
+	il2_hdr = skb_inner_mac_header(skb);
+	/* compute OL4 header size, defined in 4 Bytes. */
+	l4_len = il2_hdr - l4.hdr;
+	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
+
+	/* define outer network header type */
+	if (skb->protocol == htons(ETH_P_IP)) {
+		if (skb_is_gso(skb))
 			hns3_set_field(*ol_type_vlan_len_msec,
-				       HNS3_TXD_TUNTYPE_S,
-				       HNS3_TUN_MAC_IN_UDP);
-			break;
-		case IPPROTO_GRE:
+				       HNS3_TXD_OL3T_S,
+				       HNS3_OL3T_IPV4_CSUM);
+		else
 			hns3_set_field(*ol_type_vlan_len_msec,
-				       HNS3_TXD_TUNTYPE_S,
-				       HNS3_TUN_NVGRE);
-			break;
-		default:
+				       HNS3_TXD_OL3T_S,
+				       HNS3_OL3T_IPV4_NO_CSUM);
+
+	} else if (skb->protocol == htons(ETH_P_IPV6)) {
+		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
+			       HNS3_OL3T_IPV6);
+	}
+
+	if (ol4_proto == IPPROTO_UDP)
+		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
+			       HNS3_TUN_MAC_IN_UDP);
+	else if (ol4_proto == IPPROTO_GRE)
+		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
+			       HNS3_TUN_NVGRE);
+}
+
+static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
+			   u8 il4_proto, u32 *type_cs_vlan_tso,
+			   u32 *ol_type_vlan_len_msec)
+{
+	unsigned char *l2_hdr = l2_hdr = skb->data;
+	u32 l4_proto = ol4_proto;
+	union l4_hdr_info l4;
+	union l3_hdr_info l3;
+	u32 l2_len, l3_len;
+
+	l4.hdr = skb_transport_header(skb);
+	l3.hdr = skb_network_header(skb);
+
+	/* handle encapsulation skb */
+	if (skb->encapsulation) {
+		/* If this is a not UDP/GRE encapsulation skb */
+		if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
 			/* drop the skb tunnel packet if hardware don't support,
 			 * because hardware can't calculate csum when TSO.
 			 */
@@ -884,7 +836,12 @@ static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
 			return 0;
 		}
 
+		hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
+
+		/* switch to inner header */
+		l2_hdr = skb_inner_mac_header(skb);
 		l3.hdr = skb_inner_network_header(skb);
+		l4.hdr = skb_inner_transport_header(skb);
 		l4_proto = il4_proto;
 	}
 
@@ -902,11 +859,22 @@ static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
 			       HNS3_L3T_IPV6);
 	}
 
+	/* compute inner(/normal) L2 header size, defined in 2 Bytes */
+	l2_len = l3.hdr - l2_hdr;
+	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
+
+	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
+	l3_len = l4.hdr - l3.hdr;
+	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
+
+	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
 	switch (l4_proto) {
 	case IPPROTO_TCP:
 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
 			       HNS3_L4T_TCP);
+		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
+			       l4.tcp->doff);
 		break;
 	case IPPROTO_UDP:
 		if (hns3_tunnel_csum_bug(skb))
@@ -915,11 +883,15 @@ static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
 			       HNS3_L4T_UDP);
+		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
+			       (sizeof(struct udphdr) >> 2));
 		break;
 	case IPPROTO_SCTP:
 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
 			       HNS3_L4T_SCTP);
+		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
+			       (sizeof(struct sctphdr) >> 2));
 		break;
 	default:
 		/* drop the skb tunnel packet if hardware don't support,
@@ -1050,12 +1022,10 @@ static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
 			ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
 			if (unlikely(ret))
 				return ret;
-			hns3_set_l2l3l4_len(skb, ol4_proto, il4_proto,
-					    &type_cs_vlan_tso,
-					    &ol_type_vlan_len_msec);
-			ret = hns3_set_l3l4_type_csum(skb, ol4_proto, il4_proto,
-						      &type_cs_vlan_tso,
-						      &ol_type_vlan_len_msec);
+
+			ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
+					      &type_cs_vlan_tso,
+					      &ol_type_vlan_len_msec);
 			if (unlikely(ret))
 				return ret;
 
-- 
2.7.4

[PATCH net-next 08/12] net: hns3: fix error handling for desc filling

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

When desc filling fails in hns3_nic_net_xmit, it will call
hns3_clear_desc to unmap the dma mapping. But currently the
ring->next_to_use points to the desc where the desc filling
or dma mapping return error, which means the desc that
ring->next_to_use points to has not done the dma mapping,
the desc that need unmapping is before the ring->next_to_use.

This patch fixes it by calling ring_ptr_move_bw(next_to_use)
before doing unmapping operation, and set desc_cb->dma to
zero to avoid freeing it again when unloading.

Also, when filling skb head or frag fails, both need to unmap
all the way back to next_to_use_head, so remove one desc filling
error handling.

Fixes: 76ad4f0ee747 ("net: hns3: Add support of HNS3 Ethernet Driver for hip08 SoC")
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 18 +++++++-----------
 1 file changed, 7 insertions(+), 11 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 7224b38..21eac68 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -1224,6 +1224,9 @@ static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
 		if (ring->next_to_use == next_to_use_orig)
 			break;
 
+		/* rollback one */
+		ring_ptr_move_bw(ring, next_to_use);
+
 		/* unmap the descriptor dma address */
 		if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
 			dma_unmap_single(dev,
@@ -1237,9 +1240,7 @@ static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
 				       DMA_TO_DEVICE);
 
 		ring->desc_cb[ring->next_to_use].length = 0;
-
-		/* rollback one */
-		ring_ptr_move_bw(ring, next_to_use);
+		ring->desc_cb[ring->next_to_use].dma = 0;
 	}
 }
 
@@ -1252,7 +1253,6 @@ netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
 	struct netdev_queue *dev_queue;
 	struct skb_frag_struct *frag;
 	int next_to_use_head;
-	int next_to_use_frag;
 	int buf_num;
 	int seg_num;
 	int size;
@@ -1291,9 +1291,8 @@ netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
 	ret = hns3_fill_desc(ring, skb, size, seg_num == 1 ? 1 : 0,
 			     DESC_TYPE_SKB);
 	if (unlikely(ret))
-		goto head_fill_err;
+		goto fill_err;
 
-	next_to_use_frag = ring->next_to_use;
 	/* Fill the fragments */
 	for (i = 1; i < seg_num; i++) {
 		frag = &skb_shinfo(skb)->frags[i - 1];
@@ -1304,7 +1303,7 @@ netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
 				     DESC_TYPE_PAGE);
 
 		if (unlikely(ret))
-			goto frag_fill_err;
+			goto fill_err;
 	}
 
 	/* Complete translate all packets */
@@ -1317,10 +1316,7 @@ netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
 
 	return NETDEV_TX_OK;
 
-frag_fill_err:
-	hns3_clear_desc(ring, next_to_use_frag);
-
-head_fill_err:
+fill_err:
 	hns3_clear_desc(ring, next_to_use_head);
 
 out_err_tx_ok:
-- 
2.7.4

[PATCH net-next 09/12] net: hns3: optimize the barrier using when cleaning TX BD

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

Currently, a barrier is used when cleaning each TX BD, which may
cause performance degradation.

This patch optimizes it to use one barrier when cleaning TX BD
each round.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 29 +++++++++++++------------
 1 file changed, 15 insertions(+), 14 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 21eac68..25d607c 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -2186,20 +2186,25 @@ static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
 	ring->desc[i].rx.bd_base_info = 0;
 }
 
-static void hns3_nic_reclaim_one_desc(struct hns3_enet_ring *ring, int *bytes,
-				      int *pkts)
+static void hns3_nic_reclaim_desc(struct hns3_enet_ring *ring, int head,
+				  int *bytes, int *pkts)
 {
 	int ntc = ring->next_to_clean;
 	struct hns3_desc_cb *desc_cb;
 
-	desc_cb = &ring->desc_cb[ntc];
-	(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
-	(*bytes) += desc_cb->length;
-	/* desc_cb will be cleaned, after hnae3_free_buffer_detach*/
-	hns3_free_buffer_detach(ring, ntc);
+	while (head != ntc) {
+		desc_cb = &ring->desc_cb[ntc];
+		(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
+		(*bytes) += desc_cb->length;
+		/* desc_cb will be cleaned, after hnae3_free_buffer_detach */
+		hns3_free_buffer_detach(ring, ntc);
 
-	if (++ntc == ring->desc_num)
-		ntc = 0;
+		if (++ntc == ring->desc_num)
+			ntc = 0;
+
+		/* Issue prefetch for next Tx descriptor */
+		prefetch(&ring->desc_cb[ntc]);
+	}
 
 	/* This smp_store_release() pairs with smp_load_acquire() in
 	 * ring_space called by hns3_nic_net_xmit.
@@ -2244,11 +2249,7 @@ void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
 
 	bytes = 0;
 	pkts = 0;
-	while (head != ring->next_to_clean) {
-		hns3_nic_reclaim_one_desc(ring, &bytes, &pkts);
-		/* Issue prefetch for next Tx descriptor */
-		prefetch(&ring->desc_cb[ring->next_to_clean]);
-	}
+	hns3_nic_reclaim_desc(ring, head, &bytes, &pkts);
 
 	ring->tqp_vector->tx_group.total_bytes += bytes;
 	ring->tqp_vector->tx_group.total_packets += pkts;
-- 
2.7.4

[PATCH net-next 10/12] net: hns3: unify the page reusing for page size 4K and 64K

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

When page size is 64K, RX buffer is currently not reused when the
page_offset is moved to last buffer. This patch adds checking to
decide whether the buffer page can be reused when last_offset is
moved beyond last offset.

If the driver is the only user of page when page_offset is moved
to beyond last offset, then buffer can be reused and page_offset
is set to zero.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 45 +++++++------------------
 1 file changed, 13 insertions(+), 32 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 25d607c..b6e73fe 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -2328,50 +2328,31 @@ static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
 				struct hns3_enet_ring *ring, int pull_len,
 				struct hns3_desc_cb *desc_cb)
 {
-	struct hns3_desc *desc;
-	u32 truesize;
-	int size;
-	int last_offset;
-	bool twobufs;
-
-	twobufs = ((PAGE_SIZE < 8192) &&
-		hnae3_buf_size(ring) == HNS3_BUFFER_SIZE_2048);
-
-	desc = &ring->desc[ring->next_to_clean];
-	size = le16_to_cpu(desc->rx.size);
-
-	truesize = hnae3_buf_size(ring);
-
-	if (!twobufs)
-		last_offset = hnae3_page_size(ring) - hnae3_buf_size(ring);
+	struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
+	int size = le16_to_cpu(desc->rx.size);
+	u32 truesize = hnae3_buf_size(ring);
 
 	skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
 			size - pull_len, truesize);
 
-	 /* Avoid re-using remote pages,flag default unreuse */
-	if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
-		return;
-
-	if (twobufs) {
-		/* If we are only owner of page we can reuse it */
-		if (likely(page_count(desc_cb->priv) == 1)) {
-			/* Flip page offset to other buffer */
-			desc_cb->page_offset ^= truesize;
-
-			desc_cb->reuse_flag = 1;
-			/* bump ref count on page before it is given*/
-			get_page(desc_cb->priv);
-		}
+	/* Avoid re-using remote pages, or the stack is still using the page
+	 * when page_offset rollback to zero, flag default unreuse
+	 */
+	if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()) ||
+	    (!desc_cb->page_offset && page_count(desc_cb->priv) > 1))
 		return;
-	}
 
 	/* Move offset up to the next cache line */
 	desc_cb->page_offset += truesize;
 
-	if (desc_cb->page_offset <= last_offset) {
+	if (desc_cb->page_offset + truesize <= hnae3_page_size(ring)) {
 		desc_cb->reuse_flag = 1;
 		/* Bump ref count on page before it is given*/
 		get_page(desc_cb->priv);
+	} else if (page_count(desc_cb->priv) == 1) {
+		desc_cb->reuse_flag = 1;
+		desc_cb->page_offset = 0;
+		get_page(desc_cb->priv);
 	}
 }
 
-- 
2.7.4

[PATCH net-next 11/12] net: hns3: some cleanup for struct hns3_enet_ring

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

This patch removes some unused field in struct hns3_enet_ring,
use ring->dev for ring_to_dev macro, and use dev consistently
in hns3_fill_desc.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 2 +-
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.h | 9 +--------
 2 files changed, 2 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index b6e73fe..65fb421 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -1051,7 +1051,7 @@ static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
 	}
 
-	if (unlikely(dma_mapping_error(ring->dev, dma))) {
+	if (unlikely(dma_mapping_error(dev, dma))) {
 		ring->stats.sw_err_cnt++;
 		return -ENOMEM;
 	}
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
index 9680a68..c14480f 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.h
@@ -401,7 +401,6 @@ struct hns3_enet_ring {
 	struct hns3_enet_ring *next;
 	struct hns3_enet_tqp_vector *tqp_vector;
 	struct hnae3_queue *tqp;
-	char ring_name[HNS3_RING_NAME_LEN];
 	struct device *dev; /* will be used for DMA mapping of descriptors */
 
 	/* statistic */
@@ -411,9 +410,6 @@ struct hns3_enet_ring {
 	dma_addr_t desc_dma_addr;
 	u32 buf_size;       /* size for hnae_desc->addr, preset by AE */
 	u16 desc_num;       /* total number of desc */
-	u16 max_desc_num_per_pkt;
-	u16 max_raw_data_sz_per_desc;
-	u16 max_pkt_size;
 	int next_to_use;    /* idx of next spare desc */
 
 	/* idx of lastest sent desc, the ring is empty when equal to
@@ -427,9 +423,6 @@ struct hns3_enet_ring {
 
 	u32 flag;          /* ring attribute */
 
-	int numa_node;
-	cpumask_t affinity_mask;
-
 	int pending_buf;
 	struct sk_buff *skb;
 	struct sk_buff *tail_skb;
@@ -629,7 +622,7 @@ static inline bool hns3_nic_resetting(struct net_device *netdev)
 #define hnae3_queue_xmit(tqp, buf_num) writel_relaxed(buf_num, \
 		(tqp)->io_base + HNS3_RING_TX_RING_TAIL_REG)
 
-#define ring_to_dev(ring) (&(ring)->tqp->handle->pdev->dev)
+#define ring_to_dev(ring) ((ring)->dev)
 
 #define ring_to_dma_dir(ring) (HNAE3_IS_TX_RING(ring) ? \
 	DMA_TO_DEVICE : DMA_FROM_DEVICE)
-- 
2.7.4

[PATCH net-next 12/12] net: hns3: use devm_kcalloc when allocating desc_cb

[Huazhong Tan]

From: Yunsheng Lin <linyunsheng@huawei.com>

This patch uses devm_kcalloc instead of kcalloc when allocating
ring->desc_cb, because devm_kcalloc not only ensure to free the
memory when the dev is deallocted, but also allocate the memory
from it's device memory node.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
---
 drivers/net/ethernet/hisilicon/hns3/hns3_enet.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
index 65fb421..18711e0 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c
@@ -3478,8 +3478,8 @@ static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
 		return -EINVAL;
 
-	ring->desc_cb = kcalloc(ring->desc_num, sizeof(ring->desc_cb[0]),
-				GFP_KERNEL);
+	ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
+				     sizeof(ring->desc_cb[0]), GFP_KERNEL);
 	if (!ring->desc_cb) {
 		ret = -ENOMEM;
 		goto out;
@@ -3500,7 +3500,7 @@ static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
 out_with_desc:
 	hns3_free_desc(ring);
 out_with_desc_cb:
-	kfree(ring->desc_cb);
+	devm_kfree(ring_to_dev(ring), ring->desc_cb);
 	ring->desc_cb = NULL;
 out:
 	return ret;
@@ -3509,7 +3509,7 @@ static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
 static void hns3_fini_ring(struct hns3_enet_ring *ring)
 {
 	hns3_free_desc(ring);
-	kfree(ring->desc_cb);
+	devm_kfree(ring_to_dev(ring), ring->desc_cb);
 	ring->desc_cb = NULL;
 	ring->next_to_clean = 0;
 	ring->next_to_use = 0;
-- 
2.7.4

Re: [PATCH net-next 00/12] cleanup & optimizations & bugfixes for HNS3 driver

[David Miller]

From: Huazhong Tan <tanhuazhong@huawei.com>
Date: Mon, 6 May 2019 10:48:40 +0800

> This patchset contains some cleanup related to hns3_enet_ring
> struct and tx bd filling process, optimizations related
> to rx page reusing, barrier using and tso handling process,
> bugfixes related to tunnel type handling and error handling for
> desc filling.

Series applied, thanks.