[Intel-wired-lan] [PATCH net-next] igc: add xdp frags support to ndo_xdp_xmit

[Neftin, Sasha]

On 7/6/2022 10:54, Lorenzo Bianconi wrote:
> Add the capability to map non-linear xdp frames in XDP_TX and
> ndo_xdp_xmit callback.
> 
> Signed-off-by: Lorenzo Bianconi <lorenzo at kernel.org>
> ---
> Please note this patch is only compiled tested since I do not have
> access to a igc NIC
> ---
>   drivers/net/ethernet/intel/igc/igc_main.c | 128 ++++++++++++++--------
>   1 file changed, 83 insertions(+), 45 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/igc/igc_main.c b/drivers/net/ethernet/intel/igc/igc_main.c
> index ae17af44fe02..71657d03da03 100644
> --- a/drivers/net/ethernet/intel/igc/igc_main.c
> +++ b/drivers/net/ethernet/intel/igc/igc_main.c
> @@ -2129,65 +2129,102 @@ static bool igc_alloc_rx_buffers_zc(struct igc_ring *ring, u16 count)
>   	return ok;
>   }
>   
> -static int igc_xdp_init_tx_buffer(struct igc_tx_buffer *buffer,
> -				  struct xdp_frame *xdpf,
> -				  struct igc_ring *ring)
> -{
> -	dma_addr_t dma;
> -
> -	dma = dma_map_single(ring->dev, xdpf->data, xdpf->len, DMA_TO_DEVICE);
> -	if (dma_mapping_error(ring->dev, dma)) {
> -		netdev_err_once(ring->netdev, "Failed to map DMA for TX\n");
> -		return -ENOMEM;
> -	}
> -
> -	buffer->type = IGC_TX_BUFFER_TYPE_XDP;
> -	buffer->xdpf = xdpf;
> -	buffer->protocol = 0;
> -	buffer->bytecount = xdpf->len;
> -	buffer->gso_segs = 1;
> -	buffer->time_stamp = jiffies;
> -	dma_unmap_len_set(buffer, len, xdpf->len);
> -	dma_unmap_addr_set(buffer, dma, dma);
> -	return 0;
> -}
> -
>   /* This function requires __netif_tx_lock is held by the caller. */
>   static int igc_xdp_init_tx_descriptor(struct igc_ring *ring,
>   				      struct xdp_frame *xdpf)
>   {
> -	struct igc_tx_buffer *buffer;
> -	union igc_adv_tx_desc *desc;
> -	u32 cmd_type, olinfo_status;
> -	int err;
> +	struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
> +	u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
> +	u16 count, index = ring->next_to_use;
> +	struct igc_tx_buffer *head = &ring->tx_buffer_info[index];
> +	struct igc_tx_buffer *buffer = head;
> +	union igc_adv_tx_desc *desc = IGC_TX_DESC(ring, index);
> +	u32 olinfo_status, len = xdpf->len, cmd_type;
> +	void *data = xdpf->data;
> +	u16 i;
>   
> -	if (!igc_desc_unused(ring))
> -		return -EBUSY;
> +	count = TXD_USE_COUNT(len);
> +	for (i = 0; i < nr_frags; i++)
> +		count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
>   
> -	buffer = &ring->tx_buffer_info[ring->next_to_use];
> -	err = igc_xdp_init_tx_buffer(buffer, xdpf, ring);
> -	if (err)
> -		return err;
> +	if (igc_maybe_stop_tx(ring, count + 3)) {
> +		/* this is a hard error */
> +		return -EBUSY;
> +	}
>   
> -	cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
> -		   IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD |
> -		   buffer->bytecount;
> -	olinfo_status = buffer->bytecount << IGC_ADVTXD_PAYLEN_SHIFT;
> +	i = 0;
> +	head->bytecount = xdp_get_frame_len(xdpf);
> +	head->type = IGC_TX_BUFFER_TYPE_XDP;
> +	head->gso_segs = 1;
> +	head->xdpf = xdpf;
>   
> -	desc = IGC_TX_DESC(ring, ring->next_to_use);
> -	desc->read.cmd_type_len = cpu_to_le32(cmd_type);
> +	olinfo_status = head->bytecount << IGC_ADVTXD_PAYLEN_SHIFT;
>   	desc->read.olinfo_status = cpu_to_le32(olinfo_status);
> -	desc->read.buffer_addr = cpu_to_le64(dma_unmap_addr(buffer, dma));
>   
> -	netdev_tx_sent_queue(txring_txq(ring), buffer->bytecount);
> +	for (;;) {
> +		dma_addr_t dma;
>   
> -	buffer->next_to_watch = desc;
> +		dma = dma_map_single(ring->dev, data, len, DMA_TO_DEVICE);
> +		if (dma_mapping_error(ring->dev, dma)) {
> +			netdev_err_once(ring->netdev,
> +					"Failed to map DMA for TX\n");
> +			goto unmap;
> +		}
>   
> -	ring->next_to_use++;
> -	if (ring->next_to_use == ring->count)
> -		ring->next_to_use = 0;
> +		dma_unmap_len_set(buffer, len, len);
> +		dma_unmap_addr_set(buffer, dma, dma);
> +
> +		cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
> +			   IGC_ADVTXD_DCMD_IFCS | len;
> +
> +		desc->read.cmd_type_len = cpu_to_le32(cmd_type);
> +		desc->read.buffer_addr = cpu_to_le64(dma);
> +
> +		buffer->protocol = 0;
> +
> +		if (++index == ring->count)
> +			index = 0;
> +
> +		if (i == nr_frags)
> +			break;
> +
> +		buffer = &ring->tx_buffer_info[index];
> +		desc = IGC_TX_DESC(ring, index);
> +		desc->read.olinfo_status = 0;
> +
> +		data = skb_frag_address(&sinfo->frags[i]);
> +		len = skb_frag_size(&sinfo->frags[i]);
> +		i++;
> +	}
> +	desc->read.cmd_type_len |= cpu_to_le32(IGC_TXD_DCMD);
> +
> +	netdev_tx_sent_queue(txring_txq(ring), head->bytecount);
> +	/* set the timestamp */
> +	head->time_stamp = jiffies;
> +	/* set next_to_watch value indicating a packet is present */
> +	head->next_to_watch = desc;
> +	ring->next_to_use = index;
>   
>   	return 0;
> +
> +unmap:
> +	for (;;) {
> +		buffer = &ring->tx_buffer_info[index];
> +		if (dma_unmap_len(buffer, len))
> +			dma_unmap_page(ring->dev,
> +				       dma_unmap_addr(buffer, dma),
> +				       dma_unmap_len(buffer, len),
> +				       DMA_TO_DEVICE);
> +		dma_unmap_len_set(buffer, len, 0);
> +		if (buffer == head)
> +			break;
> +
> +		if (!index)
> +			index += ring->count;
> +		index--;
> +	}
> +
> +	return -ENOMEM;
>   }
>   
>   static struct igc_ring *igc_xdp_get_tx_ring(struct igc_adapter *adapter,
> @@ -2369,6 +2406,7 @@ static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
>   			xdp_prepare_buff(&xdp, pktbuf - igc_rx_offset(rx_ring),
>   					 igc_rx_offset(rx_ring) + pkt_offset,
>   					 size, true);
> +			xdp_buff_clear_frags_flag(&xdp);
>   
>   			skb = igc_xdp_run_prog(adapter, &xdp);
>   		}
Hello Lorenzo,
Could you provide test hints (step by step) on how to test it?
Sasha