On Wed, 17 Feb 2010 16:17:16 +0100
Eric Dumazet [off-list ref] wrote:

Le mercredi 17 f=C3=A9vrier 2010 =C3=A0 15:55 +0100, Anatolij Gustschin a=

 =C3=A9crit :
...

quoted

+static struct sk_buff *tx_skb_align_workaround(struct net_device *dev,
+					       struct sk_buff *skb)
+{
+	struct sk_buff *new_skb;
+	struct fs_enet_private *fep =3D netdev_priv(dev);
+
+	/* Alloc new skb */
+	new_skb =3D dev_alloc_skb(ENET_RX_FRSIZE + 4);

=20
=20
ENET_RX_FRSIZE looks strange in TX path
=20
Why not using skb->len + 4 instead of ENET_RX_FRSIZE + 4 ?

I will fix it.

quoted

+	if (!new_skb) {
+		dev_warn(fep->dev, "Memory squeeze, dropping tx packet.\n");

=20
I am just wondering if this is ratelimited ?

Right, it should be ratelimited, will fix it, too.

...

quoted

 static int fs_enet_start_xmit(struct sk_buff *skb, struct net_device *=

dev)

quoted

 {
 	struct fs_enet_private *fep =3D netdev_priv(dev);

@@ -588,6 +619,19 @@ static int fs_enet_start_xmit(struct sk_buff *skb,=

 struct net_device *dev)

quoted

 	u16 sc;
 	unsigned long flags;
=20
+#ifdef CONFIG_FS_ENET_MPC5121_FEC
+	if (((unsigned long)skb->data) & 0x3) {
+		skb =3D tx_skb_align_workaround(dev, skb);
+		if (!skb) {
+			/*
+			 * We have lost packet due to memory allocation error
+			 * in tx_skb_align_workaround(). Hopefully original
+			 * skb is still valid, so try transmit it later.
+			 */

=20
Could you define 'try to transmit later' ?
Who is responsible to trigger this event ?

The function returns NETDEV_TX_BUSY here, skb is not
consumed and will be requeued by sch_direct_xmit(), so
it is scheduled for resending later.

Thanks,
Anatolij