Re: [PATCH net-next 2/4] atl1c: improve performance by avoiding unnecessary pcie writes on xmit

[Gatis Peisenieks]

On 2021-05-12 00:39, Eric Dumazet wrote:
> > @@ -2270,8 +2272,11 @@ static netdev_tx_t atl1c_xmit_frame(struct 
> > sk_buff *skb,
> >  		atl1c_tx_rollback(adapter, tpd, type);
> >  		dev_kfree_skb_any(skb);
> >  	} else {
> > -		netdev_sent_queue(adapter->netdev, skb->len);
> > -		atl1c_tx_queue(adapter, skb, tpd, type);
> > +		bool more = netdev_xmit_more();
> > +
> > +		__netdev_sent_queue(adapter->netdev, skb->len, more);
>
>
> This is probably buggy.
>
> You must check and use the return code of this function,
> as in :
>
> 	bool door_bell = __netdev_sent_queue(adapter->netdev, skb->len,
> netdev_xmit_more());
>
> 	if (door_bell)
> 		atl1c_tx_queue(adapter, type);

Eric, thank you for taking your time to look at this!

You are correct, tx queue can get stopped in __netdev_sent_queue
and if there were more packets coming the submit to HW would be
missed / unnecessarily delayed.


> > +		if (!more)
> > +			atl1c_tx_queue(adapter, type);
> >  	}
> >
> >  	return NETDEV_TX_OK;