2014-03-25 10:05 GMT-07:00 Arnd Bergmann <arnd@xxxxxxxx>:
> On Tuesday 25 March 2014 10:00:30 Florian Fainelli wrote:
>> 2014-03-25 1:12 GMT-07:00 Arnd Bergmann <arnd@xxxxxxxx>:
>> > On Tuesday 25 March 2014 12:06:31 Zhangfei Gao wrote:
>> >> Dear Arnd
>> >>
>> >> On Mon, Mar 24, 2014 at 11:18 PM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
>> >> > On Monday 24 March 2014 22:14:56 Zhangfei Gao wrote:
>> >> >
>> >> >> +
>> >> >> +static void hip04_tx_reclaim(struct net_device *ndev, bool force)
>> >> >> +{
>> >> >> + struct hip04_priv *priv = netdev_priv(ndev);
>> >> >> + unsigned tx_head = priv->tx_head;
>> >> >> + unsigned tx_tail = priv->tx_tail;
>> >> >> + struct tx_desc *desc = &priv->tx_desc[priv->tx_tail];
>> >> >> +
>> >> >> + while (tx_tail != tx_head) {
>> >> >> + if (desc->send_addr != 0) {
>> >> >> + if (force)
>> >> >> + desc->send_addr = 0;
>> >> >> + else
>> >> >> + break;
>> >> >> + }
>> >> >> + if (priv->tx_phys[tx_tail]) {
>> >> >> + dma_unmap_single(&ndev->dev, priv->tx_phys[tx_tail],
>> >> >> + priv->tx_skb[tx_tail]->len, DMA_TO_DEVICE);
>> >> >> + priv->tx_phys[tx_tail] = 0;
>> >> >> + }
>> >> >> + dev_kfree_skb_irq(priv->tx_skb[tx_tail]);
>> >> >> + priv->tx_skb[tx_tail] = NULL;
>> >> >> + tx_tail = TX_NEXT(tx_tail);
>> >> >> + priv->tx_count--;
>> >> >> + }
>> >> >> + priv->tx_tail = tx_tail;
>> >> >> +}
>> >> >
>> >> > I think you still need to find a solution to ensure that the tx reclaim is
>> >> > called eventually through a method other than start_xmit.
>> >>
>> >> In the iperf stress test, if move reclaim to poll, there is some
>> >> error, sometimes sending zero packets.
>> >> While keep reclaim in the xmit to reclaim transmitted packets looks
>> >> stable in the test,
>> >> There TX_DESC_NUM desc can be used.
>> >
>> > What I meant is that you need a correct implementation, presumably
>> > you added a bug when you moved the function to poll(), and also you
>> > forgot to add a timer.
>>
>> Using a timer to ensure completion of TX packets is a trick that
>> worked in the past, but now that the networking stack got smarter,
>> this might artificially increase the processing time of packets in the
>> transmit path, and this will defeat features like TCP small queues
>> etc.. as could be seen with the mvneta driver [1]. The best way really
>> is to rely on TX completion interrupts when those exist as they cannot
>> lie about the hardware status (in theory) and they should provide the
>> fastest way to complete TX packets.
>
> By as Zhangfei Gao pointed out, this hardware does not have a working
> TX completion interrupt. Using timers to do this has always just been
> a workaround for broken hardware IMHO.
Ok, well that's really unfortunate, to achieve the best of everything,
the workaround should probably look like:
- keep reclaiming TX buffers in ndo_start_xmit() in case you push more
packets to the NICs than your timer can free
- reclaim TX buffers in NAPI poll() context for "symetrical" workloads
where e.g: TCP ACKs received allow you to complete TX buffers
- have a timer like you suggest which should help with transmit only
workloads at a slow rate
--
Florian
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
