[Intel-wired-lan] [PATCH 09/10] igc: Enable RX via AF_XDP zero-copy

[Maciej Fijalkowski <maciej.fijalkowski@intel.com>]

On Tue, Dec 22, 2020 at 09:43:39AM -0800, Andre Guedes wrote:
> Quoting Maciej Fijalkowski (2020-12-22 05:14:28)
> > > +static int igc_clean_rx_irq_zc(struct igc_q_vector *q_vector, const int budget)
> > > +{
> > > +     struct igc_adapter *adapter = q_vector->adapter;
> > > +     struct igc_ring *ring = q_vector->rx.ring;
> > > +     u16 cleaned_count = igc_desc_unused(ring);
> > > +     int total_bytes = 0, total_packets = 0;
> > > +     struct bpf_prog *prog;
> > > +     bool failure = false;
> > > +     int xdp_status = 0;
> > > +
> > > +     rcu_read_lock();
> > > +
> > > +     prog = READ_ONCE(adapter->xdp_prog);
> > > +
> > > +     while (likely(total_packets < budget)) {
> > > +             union igc_adv_rx_desc *desc;
> > > +             struct igc_rx_buffer *bi;
> > > +             unsigned int size;
> > > +             int res;
> > > +
> > > +             desc = IGC_RX_DESC(ring, ring->next_to_clean);
> > > +             size = le16_to_cpu(desc->wb.upper.length);
> > > +             if (!size)
> > > +                     break;
> > > +
> > > +             /* This memory barrier is needed to keep us from reading
> > > +              * any other fields out of the rx_desc until we know the
> > > +              * descriptor has been written back
> > > +              */
> > > +             dma_rmb();
> > > +
> > > +             bi = &ring->rx_buffer_info[ring->next_to_clean];
> > > +
> > > +             if (igc_test_staterr(desc, IGC_RXDADV_STAT_TSIP)) {
> > > +                     /* FIXME: For now, if packet buffer contains timestamp
> > > +                      * information, we discard it. Once XDP infrastructe
> > 
> > nit: s/infrastructe/infrastructure
> 
> Thanks, I'll fix it in the v2.
> 
> > > +int igc_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
> > > +{
> > > +     struct igc_adapter *adapter = netdev_priv(dev);
> > > +     struct igc_q_vector *q_vector;
> > > +     struct igc_ring *ring;
> > > +
> > > +     if (test_bit(__IGC_DOWN, &adapter->state))
> > > +             return -ENETDOWN;
> > > +
> > > +     if (!igc_xdp_is_enabled(adapter))
> > > +             return -ENXIO;
> > > +
> > > +     if (queue_id >= adapter->num_rx_queues)
> > > +             return -EINVAL;
> > > +
> > > +     ring = adapter->rx_ring[queue_id];
> > > +
> > > +     if (!ring->xsk_pool)
> > > +             return -ENXIO;
> > 
> > ring local variable feels a bit unnecessary.
> 
> It's just for the sake of code readability. I thought it reads better than:
> 
> 	if (!adapter->rx_ring[queue_id]->xsk_pool)
> 		return -ENXIO;

Yeah, okay then.

> 
> > > +static int igc_xdp_enable_pool(struct igc_adapter *adapter,
> > > +                            struct xsk_buff_pool *pool, u16 queue_id)
> > > +{
> > > +     struct net_device *ndev = adapter->netdev;
> > > +     struct device *dev = &adapter->pdev->dev;
> > > +     struct igc_ring *rx_ring;
> > > +     struct napi_struct *napi;
> > > +     bool needs_reset;
> > > +     u32 frame_size;
> > > +     int err;
> > > +
> > > +     if (queue_id >= adapter->num_rx_queues)
> > > +             return -EINVAL;
> > > +
> > > +     frame_size = xsk_pool_get_rx_frame_size(pool);
> > > +     if (frame_size < ETH_FRAME_LEN + VLAN_HLEN * 2) {
> > > +             /* For now, when XDP is enabled, the driver doesn't support
> > > +              * frames that span over multiple buffers. The max frame size
> > > +              * considered here is the ethernet frame size + vlan double
> > > +              * tagging.
> > > +              */
> > 
> > I don't really follow that. You check if chunk size is less than
> > ETH_FRAME_LEN + VLAN_HLEN * 2. chunk size will be at least 2k if I recall
> > correctly. How is that related to fragmented buffers?
> 
> Note that the code above checks the xsk frame size, not chunk size.
> 
> Yes, the chunk mim size is 2k indeed. However, if I'm reading the code
> correctly, the umem headroom can have an arbitrary value defined by the user
> (see xdp_umem_reg()). In this case, the xsk frame size could potentially be
> less than the ethernet frame size.

Ok I see. So you actually meant that for lower frame size given to HW we
would start to fragment the frames?

Comment was a bit unclear, hence my confusion.

Anyway, I feel like this is something generic that should be checked
earlier at xsk code? Meaning, when we actually create the umem with user
provided config at setsockopt?