Changes wrt. v1:
- added some cleanup patches
- simplify handling of 'quirks' flags as suggested by Russell King.
- remove DIV_ROUND_UP() from byte swapping loop as suggested by
Eric Dumazet
Changes wrt. v2:
- rebased against next-20141028
- added some more cleanups in fec.h
- removed unused return value from swap_buffer()
- fixed messed swab32s() call in swap_buffer2()
- fixed messed up setup of fep->quirks
Changes wrt. v3:
- removed the cleanup patches; will send later to net-next,
when the bugfix patch has been applied
commit 1b7bde6d659d ("net: fec: implement rx_copybreak to improve rx performance")
introduced a regression for i.MX28. The swap_buffer() function doing
the endian conversion of the received data on i.MX28 may access memory
beyond the actual packet size in the DMA buffer. fec_enet_copybreak()
does not copy those bytes, so that the last bytes of a packet may be
filled with invalid data after swapping.
This will likely lead to checksum errors on received packets.
E.g. when trying to mount an NFS rootfs:
UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
Do the byte swapping and copying to the new skb in one go if
necessary.
Signed-off-by: Lothar Waßmann <[email protected]>
---
drivers/net/ethernet/freescale/fec_main.c | 23 +++++++++++++++++++----
1 file changed, 19 insertions(+), 4 deletions(-)
diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c
index c27128d..3dca494 100644
--- a/drivers/net/ethernet/freescale/fec_main.c
+++ b/drivers/net/ethernet/freescale/fec_main.c
@@ -298,6 +298,16 @@ static void *swap_buffer(void *bufaddr, int len)
return bufaddr;
}
+static void swap_buffer2(void *dst_buf, void *src_buf, int len)
+{
+ int i;
+ unsigned int *src = src_buf;
+ unsigned int *dst = dst_buf;
+
+ for (i = 0; i < len; i += 4, src++, dst++)
+ *dst = swab32p(src);
+}
+
static void fec_dump(struct net_device *ndev)
{
struct fec_enet_private *fep = netdev_priv(ndev);
@@ -1307,7 +1317,7 @@ fec_enet_new_rxbdp(struct net_device *ndev, struct bufdesc *bdp, struct sk_buff
}
static bool fec_enet_copybreak(struct net_device *ndev, struct sk_buff **skb,
- struct bufdesc *bdp, u32 length)
+ struct bufdesc *bdp, u32 length, bool swap)
{
struct fec_enet_private *fep = netdev_priv(ndev);
struct sk_buff *new_skb;
@@ -1322,7 +1332,10 @@ static bool fec_enet_copybreak(struct net_device *ndev, struct sk_buff **skb,
dma_sync_single_for_cpu(&fep->pdev->dev, bdp->cbd_bufaddr,
FEC_ENET_RX_FRSIZE - fep->rx_align,
DMA_FROM_DEVICE);
- memcpy(new_skb->data, (*skb)->data, length);
+ if (!swap)
+ memcpy(new_skb->data, (*skb)->data, length);
+ else
+ swap_buffer2(new_skb->data, (*skb)->data, length);
*skb = new_skb;
return true;
@@ -1352,6 +1365,7 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
u16 vlan_tag;
int index = 0;
bool is_copybreak;
+ bool need_swap = id_entry->driver_data & FEC_QUIRK_SWAP_FRAME;
#ifdef CONFIG_M532x
flush_cache_all();
@@ -1415,7 +1429,8 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
* include that when passing upstream as it messes up
* bridging applications.
*/
- is_copybreak = fec_enet_copybreak(ndev, &skb, bdp, pkt_len - 4);
+ is_copybreak = fec_enet_copybreak(ndev, &skb, bdp, pkt_len - 4,
+ need_swap);
if (!is_copybreak) {
skb_new = netdev_alloc_skb(ndev, FEC_ENET_RX_FRSIZE);
if (unlikely(!skb_new)) {
@@ -1430,7 +1445,7 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
prefetch(skb->data - NET_IP_ALIGN);
skb_put(skb, pkt_len - 4);
data = skb->data;
- if (id_entry->driver_data & FEC_QUIRK_SWAP_FRAME)
+ if (!is_copybreak && need_swap)
swap_buffer(data, pkt_len);
/* Extract the enhanced buffer descriptor */
--
1.7.10.4
On Fri, Nov 7, 2014 at 7:02 AM, Lothar Waßmann <[email protected]> wrote:
> commit 1b7bde6d659d ("net: fec: implement rx_copybreak to improve rx performance")
> introduced a regression for i.MX28. The swap_buffer() function doing
> the endian conversion of the received data on i.MX28 may access memory
> beyond the actual packet size in the DMA buffer. fec_enet_copybreak()
> does not copy those bytes, so that the last bytes of a packet may be
> filled with invalid data after swapping.
> This will likely lead to checksum errors on received packets.
> E.g. when trying to mount an NFS rootfs:
> UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
>
> Do the byte swapping and copying to the new skb in one go if
> necessary.
>
> Signed-off-by: Lothar Waßmann <[email protected]>
With this patch I am able to NFS mount on mx28 again. Thanks, Lothar!
Tested-by: Fabio Estevam <[email protected]>
From: Fabio Estevam <[email protected]>
Date: Fri, 7 Nov 2014 10:26:15 -0200
> On Fri, Nov 7, 2014 at 7:02 AM, Lothar Wa?mann <[email protected]> wrote:
>> commit 1b7bde6d659d ("net: fec: implement rx_copybreak to improve rx performance")
>> introduced a regression for i.MX28. The swap_buffer() function doing
>> the endian conversion of the received data on i.MX28 may access memory
>> beyond the actual packet size in the DMA buffer. fec_enet_copybreak()
>> does not copy those bytes, so that the last bytes of a packet may be
>> filled with invalid data after swapping.
>> This will likely lead to checksum errors on received packets.
>> E.g. when trying to mount an NFS rootfs:
>> UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
>>
>> Do the byte swapping and copying to the new skb in one go if
>> necessary.
>>
>> Signed-off-by: Lothar Wa?mann <[email protected]>
>
> With this patch I am able to NFS mount on mx28 again. Thanks, Lothar!
>
> Tested-by: Fabio Estevam <[email protected]>
Applied, thanks everyone.