On 6/26/20, Jesper Dangaard Brouer <bro...@redhat.com> wrote: > On Fri, 26 Jun 2020 14:40:38 +0300 > Denis Kirjanov <k...@linux-powerpc.org> wrote: > >> diff --git a/drivers/net/xen-netfront.c b/drivers/net/xen-netfront.c >> index 482c6c8..91a3b53 100644 >> --- a/drivers/net/xen-netfront.c >> +++ b/drivers/net/xen-netfront.c > [...] >> @@ -560,6 +572,67 @@ static u16 xennet_select_queue(struct net_device >> *dev, struct sk_buff *skb, >> return queue_idx; >> } >> >> +static int xennet_xdp_xmit_one(struct net_device *dev, >> + struct netfront_queue *queue, >> + struct xdp_frame *xdpf) >> +{ >> + struct netfront_info *np = netdev_priv(dev); >> + struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); >> + struct xen_netif_tx_request *tx; >> + int notify; >> + >> + tx = xennet_make_first_txreq(queue, NULL, >> + virt_to_page(xdpf->data), >> + offset_in_page(xdpf->data), >> + xdpf->len); >> + >> + RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); >> + if (notify) >> + notify_remote_via_irq(queue->tx_irq); > > Is this an expensive operation?
Hi Jesper, actually not. > > Do you think this can be moved outside the loop? > So that it is called once per bulk. I've tested both variants and it turned out that there is no difference between both: with xdp_redirect I see the similar picture: PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 1074 root 20 0 0 0 0 R 100.0 0.0 1:17.60 cpumap/3/m+ 9 root 20 0 0 0 0 R 93.7 0.0 1:37.03 ksoftirqd/0 with the similar pps rate reported by xdp_redirect. So I'll keep the code as is in the next submission. Thanks! > > >> + u64_stats_update_begin(&tx_stats->syncp); >> + tx_stats->bytes += xdpf->len; >> + tx_stats->packets++; >> + u64_stats_update_end(&tx_stats->syncp); >> + >> + xennet_tx_buf_gc(queue); >> + >> + return 0; >> +} >> + >> +static int xennet_xdp_xmit(struct net_device *dev, int n, >> + struct xdp_frame **frames, u32 flags) >> +{ >> + unsigned int num_queues = dev->real_num_tx_queues; >> + struct netfront_info *np = netdev_priv(dev); >> + struct netfront_queue *queue = NULL; >> + unsigned long irq_flags; >> + int drops = 0; >> + int i, err; >> + >> + if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) >> + return -EINVAL; >> + >> + queue = &np->queues[smp_processor_id() % num_queues]; >> + >> + spin_lock_irqsave(&queue->tx_lock, irq_flags); >> + for (i = 0; i < n; i++) { >> + struct xdp_frame *xdpf = frames[i]; >> + >> + if (!xdpf) >> + continue; >> + err = xennet_xdp_xmit_one(dev, queue, xdpf); >> + if (err) { >> + xdp_return_frame_rx_napi(xdpf); >> + drops++; >> + } >> + } >> + spin_unlock_irqrestore(&queue->tx_lock, irq_flags); >> + >> + return n - drops; >> +} > > > -- > Best regards, > Jesper Dangaard Brouer > MSc.CS, Principal Kernel Engineer at Red Hat > LinkedIn: http://www.linkedin.com/in/brouer > >