igb: add xdp frags support to ndo_xdp_xmit

Add the capability to map non-linear xdp frames in XDP_TX and
ndo_xdp_xmit callback.

Signed-off-by: Lorenzo Bianconi <[e-mail]>
Tested-by: Chandan Kumar Rout <[e-mail]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Jakub Kicinski <[e-mail]>

drivers/net/ethernet/intel/igb/igb_main.c

···
 		      struct igb_ring *tx_ring,
 		      struct xdp_frame *xdpf)
 {
-	union e1000_adv_tx_desc *tx_desc;
-	u32 len, cmd_type, olinfo_status;
-	struct igb_tx_buffer *tx_buffer;
-	dma_addr_t dma;
-	u16 i;
+	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, i, index = tx_ring->next_to_use;
+	struct igb_tx_buffer *tx_head = &tx_ring->tx_buffer_info[index];
+	struct igb_tx_buffer *tx_buffer = tx_head;
+	union e1000_adv_tx_desc *tx_desc = IGB_TX_DESC(tx_ring, index);
+	u32 len = xdpf->len, cmd_type, olinfo_status;
+	void *data = xdpf->data;
 
-	len = xdpf->len;
+	count = TXD_USE_COUNT(len);
+	for (i = 0; i < nr_frags; i++)
+		count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
 
-	if (unlikely(!igb_desc_unused(tx_ring)))
+	if (igb_maybe_stop_tx(tx_ring, count + 3))
 		return IGB_XDP_CONSUMED;
 
-	dma = dma_map_single(tx_ring->dev, xdpf->data, len, DMA_TO_DEVICE);
-	if (dma_mapping_error(tx_ring->dev, dma))
-		return IGB_XDP_CONSUMED;
-
+	i = 0;
 	/* record the location of the first descriptor for this packet */
-	tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
-	tx_buffer->bytecount = len;
-	tx_buffer->gso_segs = 1;
-	tx_buffer->protocol = 0;
+	tx_head->bytecount = xdp_get_frame_len(xdpf);
+	tx_head->type = IGB_TYPE_XDP;
+	tx_head->gso_segs = 1;
+	tx_head->xdpf = xdpf;
 
-	i = tx_ring->next_to_use;
-	tx_desc = IGB_TX_DESC(tx_ring, i);
-
-	dma_unmap_len_set(tx_buffer, len, len);
-	dma_unmap_addr_set(tx_buffer, dma, dma);
-	tx_buffer->type = IGB_TYPE_XDP;
-	tx_buffer->xdpf = xdpf;
-
-	tx_desc->read.buffer_addr = cpu_to_le64(dma);
-
-	/* put descriptor type bits */
-	cmd_type = E1000_ADVTXD_DTYP_DATA |
-		   E1000_ADVTXD_DCMD_DEXT |
-		   E1000_ADVTXD_DCMD_IFCS;
-	cmd_type |= len | IGB_TXD_DCMD;
-	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
-
-	olinfo_status = len << E1000_ADVTXD_PAYLEN_SHIFT;
+	olinfo_status = tx_head->bytecount << E1000_ADVTXD_PAYLEN_SHIFT;
 	/* 82575 requires a unique index per ring */
 	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
 		olinfo_status |= tx_ring->reg_idx << 4;
-
 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
 
-	netdev_tx_sent_queue(txring_txq(tx_ring), tx_buffer->bytecount);
+	for (;;) {
+		dma_addr_t dma;
 
+		dma = dma_map_single(tx_ring->dev, data, len, DMA_TO_DEVICE);
+		if (dma_mapping_error(tx_ring->dev, dma))
+			goto unmap;
+
+		/* record length, and DMA address */
+		dma_unmap_len_set(tx_buffer, len, len);
+		dma_unmap_addr_set(tx_buffer, dma, dma);
+
+		/* put descriptor type bits */
+		cmd_type = E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_DEXT |
+			   E1000_ADVTXD_DCMD_IFCS | len;
+
+		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
+		tx_desc->read.buffer_addr = cpu_to_le64(dma);
+
+		tx_buffer->protocol = 0;
+
+		if (++index == tx_ring->count)
+			index = 0;
+
+		if (i == nr_frags)
+			break;
+
+		tx_buffer = &tx_ring->tx_buffer_info[index];
+		tx_desc = IGB_TX_DESC(tx_ring, index);
+		tx_desc->read.olinfo_status = 0;
+
+		data = skb_frag_address(&sinfo->frags[i]);
+		len = skb_frag_size(&sinfo->frags[i]);
+		i++;
+	}
+	tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_TXD_DCMD);
+
+	netdev_tx_sent_queue(txring_txq(tx_ring), tx_head->bytecount);
 	/* set the timestamp */
-	tx_buffer->time_stamp = jiffies;
+	tx_head->time_stamp = jiffies;
 
 	/* Avoid any potential race with xdp_xmit and cleanup */
 	smp_wmb();
 
 	/* set next_to_watch value indicating a packet is present */
-	i++;
-	if (i == tx_ring->count)
-		i = 0;
-
-	tx_buffer->next_to_watch = tx_desc;
-	tx_ring->next_to_use = i;
+	tx_head->next_to_watch = tx_desc;
+	tx_ring->next_to_use = index;
 
 	/* Make sure there is space in the ring for the next send. */
 	igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
 
 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more())
-		writel(i, tx_ring->tail);
+		writel(index, tx_ring->tail);
 
 	return IGB_XDP_TX;
+
+unmap:
+	for (;;) {
+		tx_buffer = &tx_ring->tx_buffer_info[index];
+		if (dma_unmap_len(tx_buffer, len))
+			dma_unmap_page(tx_ring->dev,
+				       dma_unmap_addr(tx_buffer, dma),
+				       dma_unmap_len(tx_buffer, len),
+				       DMA_TO_DEVICE);
+		dma_unmap_len_set(tx_buffer, len, 0);
+		if (tx_buffer == tx_head)
+			break;
+
+		if (!index)
+			index += tx_ring->count;
+		index--;
+	}
+
+	return IGB_XDP_CONSUMED;
 }
 
 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
···
 			unsigned int offset = pkt_offset + igb_rx_offset(rx_ring);
 
 			xdp_prepare_buff(&xdp, hard_start, offset, size, true);
+			xdp_buff_clear_frags_flag(&xdp);
 #if (PAGE_SIZE > 4096)
 			/* At larger PAGE_SIZE, frame_sz depend on len size */
 			xdp.frame_sz = igb_rx_frame_truesize(rx_ring, size);