loopback driver carefully uses per_cpu storage for statistics but updates
loopback_dev.last_rx
As last_rx is only used by bond driver to detect link activity, and loopback
dev wont be used as a bond slave, can we avoid this write access that slowdown
SMP platforms, because of cache ping pongs ?
If this patch is not correct, I suspect we should update trans_start in
loopback driver since the condition in the bond driver is bidirectional :
if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
Signed-off-by: Eric Dumazet <[EMAIL PROTECTED]>
--- a/drivers/net/loopback.c 2006-02-07 12:10:55.000000000 +0100
+++ b/drivers/net/loopback.c 2006-02-08 16:49:08.000000000 +0100
@@ -147,7 +147,6 @@
return 0;
}
#endif
- dev->last_rx = jiffies;
lb_stats = &per_cpu(loopback_stats, get_cpu());
lb_stats->rx_bytes += skb->len;
