[PATCH] ravb: fix ring memory allocation

[PATCH] ravb: fix ring memory allocation

[Sergei Shtylyov]

The driver is written as if it can adapt to a low memory situation  allocating
less RX  skbs and TX aligned buffers than the respective RX/TX ring sizes.  In
reality  though  the driver  would malfunction in this case. Stop being overly
smart and just fail in such situation -- this is achieved by moving the memory
allocation from ravb_ring_format() to ravb_ring_init().

We leave dma_map_single() calls in place but make their failure non-fatal
by marking the corresponding RX descriptors  with zero data size which should
prevent DMA to an invalid addresses.

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>

---
The patch is against Dave Miller's 'net.git' repo.

drivers/net/ethernet/renesas/ravb_main.c |   59 +++++++++++++++++--------------
 1 file changed, 34 insertions(+), 25 deletions(-)

Index: net/drivers/net/ethernet/renesas/ravb_main.c
===================================================================
--- net.orig/drivers/net/ethernet/renesas/ravb_main.c
+++ net/drivers/net/ethernet/renesas/ravb_main.c
@@ -228,9 +228,7 @@ static void ravb_ring_format(struct net_
 	struct ravb_desc *desc = NULL;
 	int rx_ring_size = sizeof(*rx_desc) * priv->num_rx_ring[q];
 	int tx_ring_size = sizeof(*tx_desc) * priv->num_tx_ring[q];
-	struct sk_buff *skb;
 	dma_addr_t dma_addr;
-	void *buffer;
 	int i;
 
 	priv->cur_rx[q] = 0;
@@ -241,41 +239,28 @@ static void ravb_ring_format(struct net_
 	memset(priv->rx_ring[q], 0, rx_ring_size);
 	/* Build RX ring buffer */
 	for (i = 0; i < priv->num_rx_ring[q]; i++) {
-		priv->rx_skb[q][i] = NULL;
-		skb = netdev_alloc_skb(ndev, PKT_BUF_SZ + RAVB_ALIGN - 1);
-		if (!skb)
-			break;
-		ravb_set_buffer_align(skb);
 		/* RX descriptor */
 		rx_desc = &priv->rx_ring[q][i];
 		/* The size of the buffer should be on 16-byte boundary. */
 		rx_desc->ds_cc = cpu_to_le16(ALIGN(PKT_BUF_SZ, 16));
-		dma_addr = dma_map_single(&ndev->dev, skb->data,
+		dma_addr = dma_map_single(&ndev->dev, priv->rx_skb[q][i]->data,
 					  ALIGN(PKT_BUF_SZ, 16),
 					  DMA_FROM_DEVICE);
-		if (dma_mapping_error(&ndev->dev, dma_addr)) {
-			dev_kfree_skb(skb);
-			break;
-		}
-		priv->rx_skb[q][i] = skb;
+		/* We just set the data size to 0 for a failed mapping which
+		 * should prevent DMA from happening...
+		 */
+		if (dma_mapping_error(&ndev->dev, dma_addr))
+			rx_desc->ds_cc = cpu_to_le16(0);
 		rx_desc->dptr = cpu_to_le32(dma_addr);
 		rx_desc->die_dt = DT_FEMPTY;
 	}
 	rx_desc = &priv->rx_ring[q][i];
 	rx_desc->dptr = cpu_to_le32((u32)priv->rx_desc_dma[q]);
 	rx_desc->die_dt = DT_LINKFIX; /* type */
-	priv->dirty_rx[q] = (u32)(i - priv->num_rx_ring[q]);
 
 	memset(priv->tx_ring[q], 0, tx_ring_size);
 	/* Build TX ring buffer */
 	for (i = 0; i < priv->num_tx_ring[q]; i++) {
-		priv->tx_skb[q][i] = NULL;
-		priv->tx_buffers[q][i] = NULL;
-		buffer = kmalloc(PKT_BUF_SZ + RAVB_ALIGN - 1, GFP_KERNEL);
-		if (!buffer)
-			break;
-		/* Aligned TX buffer */
-		priv->tx_buffers[q][i] = buffer;
 		tx_desc = &priv->tx_ring[q][i];
 		tx_desc->die_dt = DT_EEMPTY;
 	}
@@ -298,7 +283,10 @@ static void ravb_ring_format(struct net_
 static int ravb_ring_init(struct net_device *ndev, int q)
 {
 	struct ravb_private *priv = netdev_priv(ndev);
+	struct sk_buff *skb;
 	int ring_size;
+	void *buffer;
+	int i;
 
 	/* Allocate RX and TX skb rings */
 	priv->rx_skb[q] = kcalloc(priv->num_rx_ring[q],
@@ -308,12 +296,28 @@ static int ravb_ring_init(struct net_dev
 	if (!priv->rx_skb[q] || !priv->tx_skb[q])
 		goto error;
 
+	for (i = 0; i < priv->num_rx_ring[q]; i++) {
+		skb = netdev_alloc_skb(ndev, PKT_BUF_SZ + RAVB_ALIGN - 1);
+		if (!skb)
+			goto error;
+		ravb_set_buffer_align(skb);
+		priv->rx_skb[q][i] = skb;
+	}
+
 	/* Allocate rings for the aligned buffers */
 	priv->tx_buffers[q] = kcalloc(priv->num_tx_ring[q],
 				      sizeof(*priv->tx_buffers[q]), GFP_KERNEL);
 	if (!priv->tx_buffers[q])
 		goto error;
 
+	for (i = 0; i < priv->num_tx_ring[q]; i++) {
+		buffer = kmalloc(PKT_BUF_SZ + RAVB_ALIGN - 1, GFP_KERNEL);
+		if (!buffer)
+			goto error;
+		/* Aligned TX buffer */
+		priv->tx_buffers[q][i] = buffer;
+	}
+
 	/* Allocate all RX descriptors. */
 	ring_size = sizeof(struct ravb_ex_rx_desc) * (priv->num_rx_ring[q] + 1);
 	priv->rx_ring[q] = dma_alloc_coherent(NULL, ring_size,
@@ -524,6 +528,10 @@ static bool ravb_rx(struct net_device *n
 		if (--boguscnt < 0)
 			break;
 
+		/* We use 0-byte descriptors to mark the DMA mapping errors */
+		if (!pkt_len)
+			continue;
+
 		if (desc_status & MSC_MC)
 			stats->multicast++;
 
@@ -587,10 +595,11 @@ static bool ravb_rx(struct net_device *n
 						  le16_to_cpu(desc->ds_cc),
 						  DMA_FROM_DEVICE);
 			skb_checksum_none_assert(skb);
-			if (dma_mapping_error(&ndev->dev, dma_addr)) {
-				dev_kfree_skb_any(skb);
-				break;
-			}
+			/* We just set the data size to 0 for a failed mapping
+			 * which should prevent DMA  from happening...
+			 */
+			if (dma_mapping_error(&ndev->dev, dma_addr))
+				desc->ds_cc = cpu_to_le16(0);
 			desc->dptr = cpu_to_le32(dma_addr);
 			priv->rx_skb[q][entry] = skb;
 		}

Re: [PATCH] ravb: fix ring memory allocation

[David Miller]

From: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Date: Wed, 22 Jul 2015 01:31:59 +0300

> The driver is written as if it can adapt to a low memory situation  allocating
> less RX  skbs and TX aligned buffers than the respective RX/TX ring sizes.  In
> reality  though  the driver  would malfunction in this case. Stop being overly
> smart and just fail in such situation -- this is achieved by moving the memory
> allocation from ravb_ring_format() to ravb_ring_init().
> 
> We leave dma_map_single() calls in place but make their failure non-fatal
> by marking the corresponding RX descriptors  with zero data size which should
> prevent DMA to an invalid addresses.
> 
> Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>

Applied.

But the real way to handle this is to allocate all of the necessary
resources for the replacement RX SKB before unmapping and passing the
original SKB up into the stack.

That way you _NEVER_ starve the device of RX packets to receive into,
since if you fail the memory allocation or the DMA mapping, you just
put the original SKB back into the ring.