mlx4: fix DMA mapping leak when allocation fails
authorThadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Mon, 6 Feb 2012 08:39:50 +0000 (08:39 +0000)
committerDavid S. Miller <davem@davemloft.net>
Mon, 6 Feb 2012 19:42:28 +0000 (14:42 -0500)
mlx4_en_prepare_rx_desc does not correctly clean up after it finds an
allocation failure. It should unmap a page before calling put_page, but
it only calls the later.

This bug would prevent a device removal using hotplug after setting the
device MTU to 9000 and opening the network interface. After the fix, we
still see the allocation failure with MTU 9000, but we are able to
remove the device.

Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlx4/en_rx.c

index d1c631e8293a929f2c0b7a00c0b1b30695b2167f..d4ad8c226b5115430066a4f41221f7ee41384e58 100644 (file)
@@ -168,8 +168,12 @@ static int mlx4_en_prepare_rx_desc(struct mlx4_en_priv *priv,
        return 0;
 
 err:
-       while (i--)
+       while (i--) {
+               dma_addr_t dma = be64_to_cpu(rx_desc->data[i].addr);
+               pci_unmap_single(priv->mdev->pdev, dma, skb_frags[i].size,
+                                PCI_DMA_FROMDEVICE);
                put_page(skb_frags[i].page);
+       }
        return -ENOMEM;
 }