On 15-07-14 04:08 PM, Alexei Starovoitov wrote: > On 7/14/15 3:34 PM, David Miller wrote: >>>>> 1 get rid of burst hack for both RX and TX in pktgen (kills >>>>> >>>performance) >> #1 is a serious consideration if you don't come up with better ideas, >> since an optimization is for nothing if it knowingly breaks things. > > I've dug up the pktgen source from 2002 and see: > atomic_inc(&skb->users); > odev->hard_start_xmit(skb, odev); > so it did this trick forever. > Looks like it's a fundamental way how pktgen was working > and working still. Even when new 'burst' feature is not used, > pktgen still increments skb->users to hold skb. > At present I don't have good ideas how to redesign pktgen > and since apparently no one noticed this tc_action+pktgen > breakage for years, it's probably ok to leave everything as-is until > better ideas come. I'm not giving up yet. Just ran out of ideas. >
Right and we hit this issue when pktgen is run over any stacked device with clone_skb set. I've always put it in the don't do this category but a fix would be nice. > -- > 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 -- 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