Le mardi 27 avril 2010 à 15:08 -0700, David Miller a écrit :
quoted text
> From: Eric Dumazet <eric.dumazet@gmail.com>
> Date: Fri, 23 Apr 2010 12:26:06 +0200
> 
> > Le vendredi 23 avril 2010 à 16:12 +0800, Changli Gao a écrit :
> >> batch skb dequeueing from softnet input_pkt_queue.
> >> 
> >> batch skb dequeueing from softnet input_pkt_queue to reduce potential lock
> >> contention when RPS is enabled.
> >> 
> >> Note: in the worst case, the number of packets in a softnet_data may be double
> >> of netdev_max_backlog.
> >> 
> >> Signed-off-by: Changli Gao <xiaosuo@gmail.com>
> >> ----
> > 
> > Oops, reading it again, I found process_backlog() was still taking the
> > lock twice, if only one packet is waiting in input_pkt_queue.
> > 
> > Possible fix, on top of your patch :
> 
> I've applied Changli's patch with this fixup added to it.
> 
> If there are any follow-on changes necessary after further analysis,
> please send patches on top of this work.
> 

Thanks David, I was about to resubmit the cumulative patch ;)

On my 'old' dev machine (two quad core), RPS is able to get a 300%
increase on udpsink test on 20 flows.

I yet have to make routing/firewalling tests as well.

I also noticed bnx2x driver has some strange prefetch() calls.

[PATCH net-next-2.6] bnx2x: Remove two prefetch()

1) Even on 64bit arches, sizeof(struct sk_buff) < 256
2) No need to prefetch same pointer twice.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
---

diff --git a/drivers/net/bnx2x_main.c b/drivers/net/bnx2x_main.c
index 613f727..f706ed1 100644
--- a/drivers/net/bnx2x_main.c
+++ b/drivers/net/bnx2x_main.c
@@ -1617,7 +1617,6 @@ static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
 			rx_buf = &fp->rx_buf_ring[bd_cons];
 			skb = rx_buf->skb;
 			prefetch(skb);
-			prefetch((u8 *)skb + 256);
 			len = le16_to_cpu(cqe->fast_path_cqe.pkt_len);
 			pad = cqe->fast_path_cqe.placement_offset;
 
@@ -1668,7 +1667,6 @@ static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
 					dma_unmap_addr(rx_buf, mapping),
 						   pad + RX_COPY_THRESH,
 						   DMA_FROM_DEVICE);
-			prefetch(skb);
 			prefetch(((char *)(skb)) + 128);
 
 			/* is this an error packet? */



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html