On 13/12/13 15:43, Wei Liu wrote:

On Thu, Dec 12, 2013 at 11:48:14PM +0000, Zoltan Kiss wrote:

quoted

Xen network protocol had implicit dependency on MAX_SKB_FRAGS. Netback has to
handle guests sending up to XEN_NETBK_LEGACY_SLOTS_MAX slots. To achieve that:
- create a new skb
- map the leftover slots to its frags (no linear buffer here!)
- chain it to the previous through skb_shinfo(skb)->frag_list
- map them
- copy the whole stuff into a brand new skb and send it to the stack
- unmap the 2 old skb's pages

Do you see performance regression with this approach?

Well, it was pretty hard to reproduce that behaviour even with NFS. I 
don't think it happens often enough that it causes a noticable 
performance regression. Anyway, it would be just as slow as the current 
grant copy with coalescing, maybe a bit slower due to the unmapping. But 
at least we use a core network function to do the coalescing.
Or, if you mean the generic performance, if this problem doesn't appear, 
then no, I don't see performance regression.

quoted

Signed-off-by: Zoltan Kiss <redacted>

---
  drivers/net/xen-netback/netback.c |   99 +++++++++++++++++++++++++++++++++++--
  1 file changed, 94 insertions(+), 5 deletions(-)

diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c
index e26cdda..f6ed1c8 100644
--- a/drivers/net/xen-netback/netback.c
+++ b/drivers/net/xen-netback/netback.c

@@ -906,11 +906,15 @@ static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif *vif,
  	u16 pending_idx = *((u16 *)skb->data);
  	int start;
  	pending_ring_idx_t index;
-	unsigned int nr_slots;
+	unsigned int nr_slots, frag_overflow = 0;

  	/* At this point shinfo->nr_frags is in fact the number of
  	 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
  	 */
+	if (shinfo->nr_frags > MAX_SKB_FRAGS) {
+		frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS;
+		shinfo->nr_frags = MAX_SKB_FRAGS;
+	}
  	nr_slots = shinfo->nr_frags;

It is also probably better to check whether shinfo->nr_frags is too
large which makes frag_overflow > MAX_SKB_FRAGS. I know skb should be
already be valid at this point but it wouldn't hurt to be more careful.

Ok, I've added this:
	/* At this point shinfo->nr_frags is in fact the number of
	 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
	 */
+	if (shinfo->nr_frags > MAX_SKB_FRAGS) {
+		if (shinfo->nr_frags > XEN_NETBK_LEGACY_SLOTS_MAX) return NULL;
+		frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS;

quoted

  	/* Skip first skb fragment if it is on same page as header fragment. */

@@ -926,6 +930,33 @@ static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif *vif,

  	BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS);

+	if (frag_overflow) {
+		struct sk_buff *nskb = alloc_skb(NET_SKB_PAD + NET_IP_ALIGN,
+				GFP_ATOMIC | __GFP_NOWARN);
+		if (unlikely(nskb == NULL)) {
+			netdev_err(vif->dev,
+				   "Can't allocate the frag_list skb.\n");
+			return NULL;
+		}
+
+		/* Packets passed to netif_rx() must have some headroom. */
+		skb_reserve(nskb, NET_SKB_PAD + NET_IP_ALIGN);
+

The code to call alloc_skb and skb_reserve is copied from other
location. I would like to have a dedicated function to allocate skb in
netback if possible.

OK