[PATCH net-next v2 0/3] net: systemport: misc. improvements

[PATCH net-next v2 0/3] net: systemport: misc. improvements

[Florian Fainelli]

Hi David,

These patches are highly inspired by changes from Petri on bcmgenet, last patch
is a misc fix that I had pending for a while, but is not a candidate for 'net'
at this point.

Changes in v2:

- added Petri's reviewed-by tag for patches 1 and 2
- reworked patch 2 to remove a now stale comment and use an unlikely optimization

Thanks!

Florian Fainelli (3):
  net: systemport: Pre-calculate and utilize cb->bd_addr
  net: systemport: rewrite bcm_sysport_rx_refill
  net: systemport: Add a check for oversized packets

 drivers/net/ethernet/broadcom/bcmsysport.c | 113 +++++++++++++++--------------
 drivers/net/ethernet/broadcom/bcmsysport.h |   2 -
 2 files changed, 58 insertions(+), 57 deletions(-)

-- 
2.1.0

[PATCH net-next v2 1/3] net: systemport: Pre-calculate and utilize cb->bd_addr

[Florian Fainelli]

There is a 1:1 mapping between the software maintained control block in
priv->rx_cbs and the buffer address in priv->rx_bds, such that there is
no need to keep computing the buffer address when refiling a control
block.

Reviewed-by: Petri Gynther <pgynther@google.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
---
 drivers/net/ethernet/broadcom/bcmsysport.c | 18 +++++++++---------
 drivers/net/ethernet/broadcom/bcmsysport.h |  2 --
 2 files changed, 9 insertions(+), 11 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
index 084a50a555de..267330ccd595 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.c
+++ b/drivers/net/ethernet/broadcom/bcmsysport.c
@@ -549,12 +549,7 @@ static int bcm_sysport_rx_refill(struct bcm_sysport_priv *priv,
 	}
 
 	dma_unmap_addr_set(cb, dma_addr, mapping);
-	dma_desc_set_addr(priv, priv->rx_bd_assign_ptr, mapping);
-
-	priv->rx_bd_assign_index++;
-	priv->rx_bd_assign_index &= (priv->num_rx_bds - 1);
-	priv->rx_bd_assign_ptr = priv->rx_bds +
-		(priv->rx_bd_assign_index * DESC_SIZE);
+	dma_desc_set_addr(priv, cb->bd_addr, mapping);
 
 	netif_dbg(priv, rx_status, ndev, "RX refill\n");
 
@@ -568,7 +563,7 @@ static int bcm_sysport_alloc_rx_bufs(struct bcm_sysport_priv *priv)
 	unsigned int i;
 
 	for (i = 0; i < priv->num_rx_bds; i++) {
-		cb = &priv->rx_cbs[priv->rx_bd_assign_index];
+		cb = &priv->rx_cbs[i];
 		if (cb->skb)
 			continue;
 
@@ -1330,14 +1325,14 @@ static inline int tdma_enable_set(struct bcm_sysport_priv *priv,
 
 static int bcm_sysport_init_rx_ring(struct bcm_sysport_priv *priv)
 {
+	struct bcm_sysport_cb *cb;
 	u32 reg;
 	int ret;
+	int i;
 
 	/* Initialize SW view of the RX ring */
 	priv->num_rx_bds = NUM_RX_DESC;
 	priv->rx_bds = priv->base + SYS_PORT_RDMA_OFFSET;
-	priv->rx_bd_assign_ptr = priv->rx_bds;
-	priv->rx_bd_assign_index = 0;
 	priv->rx_c_index = 0;
 	priv->rx_read_ptr = 0;
 	priv->rx_cbs = kcalloc(priv->num_rx_bds, sizeof(struct bcm_sysport_cb),
@@ -1347,6 +1342,11 @@ static int bcm_sysport_init_rx_ring(struct bcm_sysport_priv *priv)
 		return -ENOMEM;
 	}
 
+	for (i = 0; i < priv->num_rx_bds; i++) {
+		cb = priv->rx_cbs + i;
+		cb->bd_addr = priv->rx_bds + i * DESC_SIZE;
+	}
+
 	ret = bcm_sysport_alloc_rx_bufs(priv);
 	if (ret) {
 		netif_err(priv, hw, priv->netdev, "SKB allocation failed\n");
diff --git a/drivers/net/ethernet/broadcom/bcmsysport.h b/drivers/net/ethernet/broadcom/bcmsysport.h
index 42a4b4a0bc14..f28bf545d7f4 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.h
+++ b/drivers/net/ethernet/broadcom/bcmsysport.h
@@ -663,8 +663,6 @@ struct bcm_sysport_priv {
 
 	/* Receive queue */
 	void __iomem		*rx_bds;
-	void __iomem		*rx_bd_assign_ptr;
-	unsigned int		rx_bd_assign_index;
 	struct bcm_sysport_cb	*rx_cbs;
 	unsigned int		num_rx_bds;
 	unsigned int		rx_read_ptr;
-- 
2.1.0

[PATCH net-next v2 2/3] net: systemport: rewrite bcm_sysport_rx_refill

[Florian Fainelli]

Currently, bcm_sysport_desc_rx() calls bcm_sysport_rx_refill() at the end of Rx
packet processing loop, after the current Rx packet has already been passed to
napi_gro_receive(). However, bcm_sysport_rx_refill() might fail to allocate a new
Rx skb, thus leaving a hole on the Rx queue where no valid Rx buffer exists.

To eliminate this situation:

1. Rewrite bcm_sysport_rx_refill() to retain the current Rx skb on the
Rx queue if a new replacement Rx skb can't be allocated and DMA-mapped.
In this case, the data on the current Rx skb is effectively dropped.

2. Modify bcm_sysport_desc_rx() to call bcm_sysport_rx_refill() at the
top of Rx packet processing loop, so that the new replacement Rx skb is
already in place before the current Rx skb is processed.

This is loosely inspired from d6707bec5986 ("net: bcmgenet: rewrite
bcmgenet_rx_refill()")

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
---
 drivers/net/ethernet/broadcom/bcmsysport.c | 87 ++++++++++++++----------------
 1 file changed, 41 insertions(+), 46 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
index 267330ccd595..62ea403e15b8 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.c
+++ b/drivers/net/ethernet/broadcom/bcmsysport.c
@@ -524,62 +524,70 @@ static void bcm_sysport_free_cb(struct bcm_sysport_cb *cb)
 	dma_unmap_addr_set(cb, dma_addr, 0);
 }
 
-static int bcm_sysport_rx_refill(struct bcm_sysport_priv *priv,
-				 struct bcm_sysport_cb *cb)
+static struct sk_buff *bcm_sysport_rx_refill(struct bcm_sysport_priv *priv,
+					     struct bcm_sysport_cb *cb)
 {
 	struct device *kdev = &priv->pdev->dev;
 	struct net_device *ndev = priv->netdev;
+	struct sk_buff *skb, *rx_skb;
 	dma_addr_t mapping;
-	int ret;
 
-	cb->skb = netdev_alloc_skb(priv->netdev, RX_BUF_LENGTH);
-	if (!cb->skb) {
+	/* Allocate a new SKB for a new packet */
+	skb = netdev_alloc_skb(priv->netdev, RX_BUF_LENGTH);
+	if (!skb) {
+		priv->mib.alloc_rx_buff_failed++;
 		netif_err(priv, rx_err, ndev, "SKB alloc failed\n");
-		return -ENOMEM;
+		return NULL;
 	}
 
-	mapping = dma_map_single(kdev, cb->skb->data,
+	mapping = dma_map_single(kdev, skb->data,
 				 RX_BUF_LENGTH, DMA_FROM_DEVICE);
-	ret = dma_mapping_error(kdev, mapping);
-	if (ret) {
+	if (dma_mapping_error(kdev, mapping)) {
 		priv->mib.rx_dma_failed++;
-		bcm_sysport_free_cb(cb);
+		dev_kfree_skb_any(skb);
 		netif_err(priv, rx_err, ndev, "DMA mapping failure\n");
-		return ret;
+		return NULL;
 	}
 
+	/* Grab the current SKB on the ring */
+	rx_skb = cb->skb;
+	if (likely(rx_skb))
+		dma_unmap_single(kdev, dma_unmap_addr(cb, dma_addr),
+				 RX_BUF_LENGTH, DMA_FROM_DEVICE);
+
+	/* Put the new SKB on the ring */
+	cb->skb = skb;
 	dma_unmap_addr_set(cb, dma_addr, mapping);
 	dma_desc_set_addr(priv, cb->bd_addr, mapping);
 
 	netif_dbg(priv, rx_status, ndev, "RX refill\n");
 
-	return 0;
+	/* Return the current SKB to the caller */
+	return rx_skb;
 }
 
 static int bcm_sysport_alloc_rx_bufs(struct bcm_sysport_priv *priv)
 {
 	struct bcm_sysport_cb *cb;
-	int ret = 0;
+	struct sk_buff *skb;
 	unsigned int i;
 
 	for (i = 0; i < priv->num_rx_bds; i++) {
 		cb = &priv->rx_cbs[i];
-		if (cb->skb)
-			continue;
-
-		ret = bcm_sysport_rx_refill(priv, cb);
-		if (ret)
-			break;
+		skb = bcm_sysport_rx_refill(priv, cb);
+		if (skb)
+			dev_kfree_skb(skb);
+		if (!cb->skb)
+			return -ENOMEM;
 	}
 
-	return ret;
+	return 0;
 }
 
 /* Poll the hardware for up to budget packets to process */
 static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
 					unsigned int budget)
 {
-	struct device *kdev = &priv->pdev->dev;
 	struct net_device *ndev = priv->netdev;
 	unsigned int processed = 0, to_process;
 	struct bcm_sysport_cb *cb;
@@ -587,7 +595,6 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
 	unsigned int p_index;
 	u16 len, status;
 	struct bcm_rsb *rsb;
-	int ret;
 
 	/* Determine how much we should process since last call */
 	p_index = rdma_readl(priv, RDMA_PROD_INDEX);
@@ -605,29 +612,15 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
 
 	while ((processed < to_process) && (processed < budget)) {
 		cb = &priv->rx_cbs[priv->rx_read_ptr];
-		skb = cb->skb;
-
-		processed++;
-		priv->rx_read_ptr++;
-
-		if (priv->rx_read_ptr == priv->num_rx_bds)
-			priv->rx_read_ptr = 0;
+		skb = bcm_sysport_rx_refill(priv, cb);
 
-		/* We do not have a backing SKB, so we do not a corresponding
-		 * DMA mapping for this incoming packet since
-		 * bcm_sysport_rx_refill always either has both skb and mapping
-		 * or none.
-		 */
 		if (unlikely(!skb)) {
 			netif_err(priv, rx_err, ndev, "out of memory!\n");
 			ndev->stats.rx_dropped++;
 			ndev->stats.rx_errors++;
-			goto refill;
+			goto next;
 		}
 
-		dma_unmap_single(kdev, dma_unmap_addr(cb, dma_addr),
-				 RX_BUF_LENGTH, DMA_FROM_DEVICE);
-
 		/* Extract the Receive Status Block prepended */
 		rsb = (struct bcm_rsb *)skb->data;
 		len = (rsb->rx_status_len >> DESC_LEN_SHIFT) & DESC_LEN_MASK;
@@ -643,8 +636,8 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
 			netif_err(priv, rx_status, ndev, "fragmented packet!\n");
 			ndev->stats.rx_dropped++;
 			ndev->stats.rx_errors++;
-			bcm_sysport_free_cb(cb);
-			goto refill;
+			dev_kfree_skb_any(skb);
+			goto next;
 		}
 
 		if (unlikely(status & (RX_STATUS_ERR | RX_STATUS_OVFLOW))) {
@@ -653,8 +646,8 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
 				ndev->stats.rx_over_errors++;
 			ndev->stats.rx_dropped++;
 			ndev->stats.rx_errors++;
-			bcm_sysport_free_cb(cb);
-			goto refill;
+			dev_kfree_skb_any(skb);
+			goto next;
 		}
 
 		skb_put(skb, len);
@@ -681,10 +674,12 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
 		ndev->stats.rx_bytes += len;
 
 		napi_gro_receive(&priv->napi, skb);
-refill:
-		ret = bcm_sysport_rx_refill(priv, cb);
-		if (ret)
-			priv->mib.alloc_rx_buff_failed++;
+next:
+		processed++;
+		priv->rx_read_ptr++;
+
+		if (unlikely(priv->rx_read_ptr == priv->num_rx_bds))
+			priv->rx_read_ptr = 0;
 	}
 
 	return processed;
-- 
2.1.0

[PATCH net-next v2 3/3] net: systemport: Add a check for oversized packets

[Florian Fainelli]

Occasionnaly we may get oversized packets from the hardware which exceed
the nomimal 2KiB buffer size we allocate SKBs with. Add an early check
which drops the packet to avoid invoking skb_over_panic() and move on to
processing the next packet.

Reviewed-by: Petri Gynther <pgynther@google.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
---
 drivers/net/ethernet/broadcom/bcmsysport.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
index 62ea403e15b8..bbd8676a9675 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.c
+++ b/drivers/net/ethernet/broadcom/bcmsysport.c
@@ -632,6 +632,14 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
 			  p_index, priv->rx_c_index, priv->rx_read_ptr,
 			  len, status);
 
+		if (unlikely(len > RX_BUF_LENGTH)) {
+			netif_err(priv, rx_status, ndev, "oversized packet\n");
+			ndev->stats.rx_length_errors++;
+			ndev->stats.rx_errors++;
+			dev_kfree_skb_any(skb);
+			goto next;
+		}
+
 		if (unlikely(!(status & DESC_EOP) || !(status & DESC_SOP))) {
 			netif_err(priv, rx_status, ndev, "fragmented packet!\n");
 			ndev->stats.rx_dropped++;
-- 
2.1.0

Re: [PATCH net-next v2 2/3] net: systemport: rewrite bcm_sysport_rx_refill

[Petri Gynther]

On Fri, May 29, 2015 at 9:42 AM, Florian Fainelli <f.fainelli@gmail.com> wrote:
> Currently, bcm_sysport_desc_rx() calls bcm_sysport_rx_refill() at the end of Rx
> packet processing loop, after the current Rx packet has already been passed to
> napi_gro_receive(). However, bcm_sysport_rx_refill() might fail to allocate a new
> Rx skb, thus leaving a hole on the Rx queue where no valid Rx buffer exists.
>
> To eliminate this situation:
>
> 1. Rewrite bcm_sysport_rx_refill() to retain the current Rx skb on the
> Rx queue if a new replacement Rx skb can't be allocated and DMA-mapped.
> In this case, the data on the current Rx skb is effectively dropped.
>
> 2. Modify bcm_sysport_desc_rx() to call bcm_sysport_rx_refill() at the
> top of Rx packet processing loop, so that the new replacement Rx skb is
> already in place before the current Rx skb is processed.
>
> This is loosely inspired from d6707bec5986 ("net: bcmgenet: rewrite
> bcmgenet_rx_refill()")
>
> Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>

Reviewed-by: Petri Gynther <pgynther@google.com>

> ---
>  drivers/net/ethernet/broadcom/bcmsysport.c | 87 ++++++++++++++----------------
>  1 file changed, 41 insertions(+), 46 deletions(-)
>
> diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
> index 267330ccd595..62ea403e15b8 100644
> --- a/drivers/net/ethernet/broadcom/bcmsysport.c
> +++ b/drivers/net/ethernet/broadcom/bcmsysport.c
> @@ -524,62 +524,70 @@ static void bcm_sysport_free_cb(struct bcm_sysport_cb *cb)
>         dma_unmap_addr_set(cb, dma_addr, 0);
>  }
>
> -static int bcm_sysport_rx_refill(struct bcm_sysport_priv *priv,
> -                                struct bcm_sysport_cb *cb)
> +static struct sk_buff *bcm_sysport_rx_refill(struct bcm_sysport_priv *priv,
> +                                            struct bcm_sysport_cb *cb)
>  {
>         struct device *kdev = &priv->pdev->dev;
>         struct net_device *ndev = priv->netdev;
> +       struct sk_buff *skb, *rx_skb;
>         dma_addr_t mapping;
> -       int ret;
>
> -       cb->skb = netdev_alloc_skb(priv->netdev, RX_BUF_LENGTH);
> -       if (!cb->skb) {
> +       /* Allocate a new SKB for a new packet */
> +       skb = netdev_alloc_skb(priv->netdev, RX_BUF_LENGTH);
> +       if (!skb) {
> +               priv->mib.alloc_rx_buff_failed++;
>                 netif_err(priv, rx_err, ndev, "SKB alloc failed\n");
> -               return -ENOMEM;
> +               return NULL;
>         }
>
> -       mapping = dma_map_single(kdev, cb->skb->data,
> +       mapping = dma_map_single(kdev, skb->data,
>                                  RX_BUF_LENGTH, DMA_FROM_DEVICE);
> -       ret = dma_mapping_error(kdev, mapping);
> -       if (ret) {
> +       if (dma_mapping_error(kdev, mapping)) {
>                 priv->mib.rx_dma_failed++;
> -               bcm_sysport_free_cb(cb);
> +               dev_kfree_skb_any(skb);
>                 netif_err(priv, rx_err, ndev, "DMA mapping failure\n");
> -               return ret;
> +               return NULL;
>         }
>
> +       /* Grab the current SKB on the ring */
> +       rx_skb = cb->skb;
> +       if (likely(rx_skb))
> +               dma_unmap_single(kdev, dma_unmap_addr(cb, dma_addr),
> +                                RX_BUF_LENGTH, DMA_FROM_DEVICE);
> +
> +       /* Put the new SKB on the ring */
> +       cb->skb = skb;
>         dma_unmap_addr_set(cb, dma_addr, mapping);
>         dma_desc_set_addr(priv, cb->bd_addr, mapping);
>
>         netif_dbg(priv, rx_status, ndev, "RX refill\n");
>
> -       return 0;
> +       /* Return the current SKB to the caller */
> +       return rx_skb;
>  }
>
>  static int bcm_sysport_alloc_rx_bufs(struct bcm_sysport_priv *priv)
>  {
>         struct bcm_sysport_cb *cb;
> -       int ret = 0;
> +       struct sk_buff *skb;
>         unsigned int i;
>
>         for (i = 0; i < priv->num_rx_bds; i++) {
>                 cb = &priv->rx_cbs[i];
> -               if (cb->skb)
> -                       continue;
> -
> -               ret = bcm_sysport_rx_refill(priv, cb);
> -               if (ret)
> -                       break;
> +               skb = bcm_sysport_rx_refill(priv, cb);
> +               if (skb)
> +                       dev_kfree_skb(skb);
> +               if (!cb->skb)
> +                       return -ENOMEM;
>         }
>
> -       return ret;
> +       return 0;
>  }
>
>  /* Poll the hardware for up to budget packets to process */
>  static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
>                                         unsigned int budget)
>  {
> -       struct device *kdev = &priv->pdev->dev;
>         struct net_device *ndev = priv->netdev;
>         unsigned int processed = 0, to_process;
>         struct bcm_sysport_cb *cb;
> @@ -587,7 +595,6 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
>         unsigned int p_index;
>         u16 len, status;
>         struct bcm_rsb *rsb;
> -       int ret;
>
>         /* Determine how much we should process since last call */
>         p_index = rdma_readl(priv, RDMA_PROD_INDEX);
> @@ -605,29 +612,15 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
>
>         while ((processed < to_process) && (processed < budget)) {
>                 cb = &priv->rx_cbs[priv->rx_read_ptr];
> -               skb = cb->skb;
> -
> -               processed++;
> -               priv->rx_read_ptr++;
> -
> -               if (priv->rx_read_ptr == priv->num_rx_bds)
> -                       priv->rx_read_ptr = 0;
> +               skb = bcm_sysport_rx_refill(priv, cb);
>
> -               /* We do not have a backing SKB, so we do not a corresponding
> -                * DMA mapping for this incoming packet since
> -                * bcm_sysport_rx_refill always either has both skb and mapping
> -                * or none.
> -                */
>                 if (unlikely(!skb)) {
>                         netif_err(priv, rx_err, ndev, "out of memory!\n");
>                         ndev->stats.rx_dropped++;
>                         ndev->stats.rx_errors++;
> -                       goto refill;
> +                       goto next;
>                 }
>
> -               dma_unmap_single(kdev, dma_unmap_addr(cb, dma_addr),
> -                                RX_BUF_LENGTH, DMA_FROM_DEVICE);
> -
>                 /* Extract the Receive Status Block prepended */
>                 rsb = (struct bcm_rsb *)skb->data;
>                 len = (rsb->rx_status_len >> DESC_LEN_SHIFT) & DESC_LEN_MASK;
> @@ -643,8 +636,8 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
>                         netif_err(priv, rx_status, ndev, "fragmented packet!\n");
>                         ndev->stats.rx_dropped++;
>                         ndev->stats.rx_errors++;
> -                       bcm_sysport_free_cb(cb);
> -                       goto refill;
> +                       dev_kfree_skb_any(skb);
> +                       goto next;
>                 }
>
>                 if (unlikely(status & (RX_STATUS_ERR | RX_STATUS_OVFLOW))) {
> @@ -653,8 +646,8 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
>                                 ndev->stats.rx_over_errors++;
>                         ndev->stats.rx_dropped++;
>                         ndev->stats.rx_errors++;
> -                       bcm_sysport_free_cb(cb);
> -                       goto refill;
> +                       dev_kfree_skb_any(skb);
> +                       goto next;
>                 }
>
>                 skb_put(skb, len);
> @@ -681,10 +674,12 @@ static unsigned int bcm_sysport_desc_rx(struct bcm_sysport_priv *priv,
>                 ndev->stats.rx_bytes += len;
>
>                 napi_gro_receive(&priv->napi, skb);
> -refill:
> -               ret = bcm_sysport_rx_refill(priv, cb);
> -               if (ret)
> -                       priv->mib.alloc_rx_buff_failed++;
> +next:
> +               processed++;
> +               priv->rx_read_ptr++;
> +
> +               if (unlikely(priv->rx_read_ptr == priv->num_rx_bds))
> +                       priv->rx_read_ptr = 0;
>         }
>
>         return processed;
> --
> 2.1.0
>