The ethernet driver tx queue is stopped when the queue length exceeds what is allowed. It should only be started again when the queue length is back within bounds.
The logic here was just reenabling the queue when any buffers had been freed. the queue was stopped whenever the length exceeded 1000 (MAX_OUT_QUEUE_DEPTH), but then was essentially immediately started again. On a congested link, the queue length would just keep increasing up to around 8000 (for average size packets), at which point the hardware would start refusing the packets and they would begin to be dropped. This prevented the qdisc layer from effectively managing and prioritising packets, as essentially all packets were being allowed into the driver queue and then were being dropped by the hardware. This change only restarts the queue if the length is less than 1000 (MAX_OUT_QUEUE_DEPTH). Reviewed-by: Kyeong Yoo <kyeong....@alliedtelesis.co.nz> Reviewed-by: Chris Packham <chris.pack...@alliedtelesis.co.nz> Reviewed-by: Richard Laing <richard.la...@alliedtelesis.co.nz> --- drivers/staging/octeon/ethernet-tx.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/staging/octeon/ethernet-tx.c b/drivers/staging/octeon/ethernet-tx.c index c053c4a..31292fe 100644 --- a/drivers/staging/octeon/ethernet-tx.c +++ b/drivers/staging/octeon/ethernet-tx.c @@ -126,7 +126,7 @@ static void cvm_oct_free_tx_skbs(struct net_device *dev) } total_remaining += skb_queue_len(&priv->tx_free_list[qos]); } - if (total_freed >= 0 && netif_queue_stopped(dev)) + if (total_remaining < MAX_OUT_QUEUE_DEPTH && netif_queue_stopped(dev)) netif_wake_queue(dev); if (total_remaining) cvm_oct_kick_tx_poll_watchdog(); -- 2.6.4 -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html