On 16-09-12 05:17 AM, Jesper Dangaard Brouer wrote: > On Fri, 09 Sep 2016 14:29:38 -0700 > John Fastabend <john.fastab...@gmail.com> wrote: > >> e1000 supports a single TX queue so it is being shared with the stack >> when XDP runs XDP_TX action. This requires taking the xmit lock to >> ensure we don't corrupt the tx ring. To avoid taking and dropping the >> lock per packet this patch adds a bundling implementation to submit >> a bundle of packets to the xmit routine. >> >> I tested this patch running e1000 in a VM using KVM over a tap >> device using pktgen to generate traffic along with 'ping -f -l 100'. >> >> Suggested-by: Jesper Dangaard Brouer <bro...@redhat.com> > > Thank you for actually implementing this! :-) >
Yep no problem the effects are minimal on e1000 but should be noticeable at 10/40/100gbps nics. >> Signed-off-by: John Fastabend <john.r.fastab...@intel.com> >> --- > [...] [...] >> +static void e1000_xdp_xmit_bundle(struct e1000_rx_buffer_bundle >> *buffer_info, >> + struct net_device *netdev, >> + struct e1000_adapter *adapter) >> +{ >> + struct netdev_queue *txq = netdev_get_tx_queue(netdev, 0); >> + struct e1000_tx_ring *tx_ring = adapter->tx_ring; >> + struct e1000_hw *hw = &adapter->hw; >> + int i = 0; >> + >> /* e1000 only support a single txq at the moment so the queue is being >> * shared with stack. To support this requires locking to ensure the >> * stack and XDP are not running at the same time. Devices with >> * multiple queues should allocate a separate queue space. >> + * >> + * To amortize the locking cost e1000 bundles the xmits and sends as >> + * many as possible until either running out of descriptors or failing. >> */ >> HARD_TX_LOCK(netdev, txq, smp_processor_id()); >> >> - tx_ring = adapter->tx_ring; >> - >> - if (E1000_DESC_UNUSED(tx_ring) < 2) { >> - HARD_TX_UNLOCK(netdev, txq); >> - return; >> + for (; i < E1000_XDP_XMIT_BUNDLE_MAX && buffer_info[i].buffer; i++) { > ^^^ >> + e1000_xmit_raw_frame(buffer_info[i].buffer, >> + buffer_info[i].length, >> + adapter, tx_ring); >> + buffer_info[i].buffer->rxbuf.page = NULL; >> + buffer_info[i].buffer = NULL; >> + buffer_info[i].length = 0; >> + i++; > ^^^ > Looks like "i" is incremented twice, is that correct? > >> } Yep this and a couple other issues are resolved in v3 which I'll send out in a moment. Also in v3 I kept the program in the adapter structure. Moving it into the ring structure made the code a bit uglier IMO. I agree with the logic but practically only one program can exist for e1000.