On Fri, 15 Dec 2006 21:10:19 +0100
Francois Romieu [off-list ref] wrote:

Stephen Hemminger [off-list ref] :
[...]

quoted

--- linux-2.6.20-rc1.orig/drivers/net/chelsio/sge.c
+++ linux-2.6.20-rc1/drivers/net/chelsio/sge.c

[...]

quoted

@@ -1059,37 +1062,33 @@ static void recycle_fl_buf(struct freelQ
  *	threshold and the packet is too big to copy, or (b) the packet should
  *	be copied but there is no memory for the copy.
  */
-static inline struct sk_buff *get_packet(struct pci_dev *pdev,
-					 struct freelQ *fl, unsigned int len,
-					 int dma_pad, int skb_pad,
-					 unsigned int copy_thres,
-					 unsigned int drop_thres)
+static inline struct sk_buff *get_packet(struct pci_dev *pdev, struct freelQ *fl,
+					 unsigned int len)
 {
 	struct sk_buff *skb;
-	struct freelQ_ce *ce = &fl->centries[fl->cidx];
+	const struct freelQ_ce *ce = &fl->centries[fl->cidx];
+
+	if (len < copybreak) {

If you are into cleanups, maybe add likely/unlikely (and remove the inline) ?

Why, you can't predict copybreak on off

quoted

+		skb = alloc_skb(len + 2, GFP_ATOMIC);
+		if (!skb)
+			goto use_orig_buf;
 
-	if (len < copy_thres) {
-		skb = alloc_skb(len + skb_pad, GFP_ATOMIC);
-		if (likely(skb != NULL)) {
-			skb_reserve(skb, skb_pad);
-			skb_put(skb, len);
-			pci_dma_sync_single_for_cpu(pdev,
+		skb_reserve(skb, 2);	/* align IP header */

s/2/NET_IP_ALIGN/

This is wrong, NET_IP_ALIGN is for DMA buffers only. It is defined
as 0 for those platforms where DMA alignment is important. In this
case we are copying, so we want to force 2.

Btw, since the driver supports netpoll, its blind enabling of interrupts
in t1_poll() seems old-fashoined (see thread starting at
http://lkml.org/lkml/2006/12/12/86)

-- 
Stephen Hemminger [off-list ref]