Re: r8169 tx problem (1s pause with ping)

[David Gundersen <gundy@iinet.net.au>]

> diff --git a/drivers/net/r8169.c b/drivers/net/r8169.c
> index 8f3e0da..8c0851f 100644
> --- a/drivers/net/r8169.c
> +++ b/drivers/net/r8169.c
> @@ -2682,6 +2688,8 @@ static void rtl8169_tx_interrupt(struct net_device *dev,
>  		    (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)) {
>  			netif_wake_queue(dev);
>  		}
> +		if (tp->dirty_tx != tp->cur_rx)
> +			RTL_W8(TxPoll, NPQ);
>  	}
>  }
>  

Hi Francois,

A few things:

- Why are you checking dirty_tx against cur_rx (shouldn't it be cur_tx?)?

- Is there a possibility that the driver could be triggering the card to 
send invalid packets with that code?

I'm thinking in _start_xmit, the cur_tx pointer (assuming that's what 
you meant to include above) gets incremented when the packet is sent to 
the card (the RTL_W8(TxPoll,NPQ)) to indicate that the card _should_ be 
able to process packets up to that point in the queue.

The interrupt routine comes along later to clean up the buffers between 
dirty_tx (the last packet that the driver knows the card has sent) and 
cur_tx (the point that the card could potentially be up to).   What 
seems could happen with the code above is that the card gets told that a 
packet is ready to be sent when really it's not.  I'm not sure how it 
would deal with such a situation.

Regards,
Dave.