xen/netfront: improve truesize tracking
Ian Campbell [Mon, 7 Jan 2013 05:32:06 +0000 (05:32 +0000)]
Using RX_COPY_THRESHOLD is incorrect if the SKB is actually smaller
than that. We have already accounted for this in
NETFRONT_SKB_CB(skb)->pull_to so use that instead.

Fixes WARN_ON from skb_try_coalesce.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Sander Eikelenboom <linux@eikelenboom.it>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: annie li <annie.li@oracle.com>
Cc: xen-devel@lists.xen.org
Cc: netdev@vger.kernel.org
Cc: stable@kernel.org # 3.7.x only
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/xen-netfront.c

index c26e28b..7ffa43b 100644 (file)
@@ -1015,29 +1015,10 @@ err:
                i = xennet_fill_frags(np, skb, &tmpq);
 
                /*
-                * Truesize approximates the size of true data plus
-                * any supervisor overheads. Adding hypervisor
-                * overheads has been shown to significantly reduce
-                * achievable bandwidth with the default receive
-                * buffer size. It is therefore not wise to account
-                * for it here.
-                *
-                * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
-                * to RX_COPY_THRESHOLD + the supervisor
-                * overheads. Here, we add the size of the data pulled
-                * in xennet_fill_frags().
-                *
-                * We also adjust for any unused space in the main
-                * data area by subtracting (RX_COPY_THRESHOLD -
-                * len). This is especially important with drivers
-                * which split incoming packets into header and data,
-                * using only 66 bytes of the main data area (see the
-                * e1000 driver for example.)  On such systems,
-                * without this last adjustement, our achievable
-                * receive throughout using the standard receive
-                * buffer size was cut by 25%(!!!).
-                */
-               skb->truesize += skb->data_len - RX_COPY_THRESHOLD;
+                 * Truesize is the actual allocation size, even if the
+                 * allocation is only partially used.
+                 */
+               skb->truesize += PAGE_SIZE * skb_shinfo(skb)->nr_frags;
                skb->len += skb->data_len;
 
                if (rx->flags & XEN_NETRXF_csum_blank)