Merge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
i40e: support XDP multi-buffer

Tirthendu Sarkar says:

This patchset adds multi-buffer support for XDP. Tx side already has
support for multi-buffer. This patchset focuses on Rx side. The last
patch contains actual multi-buffer changes while the previous ones are
preparatory patches.

On receiving the first buffer of a packet, xdp_buff is built and its
subsequent buffers are added to it as frags. While 'next_to_clean' keeps
pointing to the first descriptor, the newly introduced 'next_to_process'
keeps track of every descriptor for the packet.

On receiving EOP buffer the XDP program is called and appropriate action
is taken (building skb for XDP_PASS, reusing page for XDP_DROP, adjusting
page offsets for XDP_{REDIRECT,TX}).

The patchset also streamlines page offset adjustments for buffer reuse
to make it easier to post process the rx_buffers after running XDP prog.

With this patchset there does not seem to be any performance degradation
for XDP_PASS and some improvement (~1% for XDP_TX, ~5% for XDP_DROP) when
measured using xdp_rxq_info program from samples/bpf/ for 64B packets.

v1: https://lore.kernel.org/netdev/20230306210822.3381942-1-anthony.l.nguyen@intel.com/

* '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
  i40e: add support for XDP multi-buffer Rx
  i40e: add xdp_buff to i40e_ring struct
  i40e: introduce next_to_process to i40e_ring
  i40e: use frame_sz instead of recalculating truesize for building skb
  i40e: Change size to truesize when using i40e_rx_buffer_flip()
  i40e: add pre-xdp page_count in rx_buffer
  i40e: change Rx buffer size for legacy-rx to support XDP multi-buffer
  i40e: consolidate maximum frame size calculation for vsi
====================

Link: https://lore.kernel.org/r/20230309212819.1198218-1-anthony.l.nguyen@intel.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/intel/i40e/i40e_ethtool.c

@@ -5402,6 +5402,13 @@ static int i40e_set_priv_flags(struct net_device *dev, u32 flags)
 		return -EOPNOTSUPP;
 	}
 
+	if ((changed_flags & I40E_FLAG_LEGACY_RX) &&
+	    I40E_2K_TOO_SMALL_WITH_PADDING) {
+		dev_warn(&pf->pdev->dev,
+			 "2k Rx buffer is too small to fit standard MTU and skb_shared_info\n");
+		return -EOPNOTSUPP;
+	}
+
 	if ((changed_flags & new_flags &
 	     I40E_FLAG_LINK_DOWN_ON_CLOSE_ENABLED) &&
 	    (new_flags & I40E_FLAG_MFP_ENABLED))

drivers/net/ethernet/intel/i40e/i40e_main.c

@@ -2896,15 +2896,35 @@ static void i40e_sync_filters_subtask(struct i40e_pf *pf)
 }
 
 /**
- * i40e_max_xdp_frame_size - returns the maximum allowed frame size for XDP
+ * i40e_calculate_vsi_rx_buf_len - Calculates buffer length
+ *
+ * @vsi: VSI to calculate rx_buf_len from
+ */
+static u16 i40e_calculate_vsi_rx_buf_len(struct i40e_vsi *vsi)
+{
+	if (!vsi->netdev || (vsi->back->flags & I40E_FLAG_LEGACY_RX))
+		return SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048);
+
+	return PAGE_SIZE < 8192 ? I40E_RXBUFFER_3072 : I40E_RXBUFFER_2048;
+}
+
+/**
+ * i40e_max_vsi_frame_size - returns the maximum allowed frame size for VSI
  * @vsi: the vsi
+ * @xdp_prog: XDP program
  **/
-static int i40e_max_xdp_frame_size(struct i40e_vsi *vsi)
+static int i40e_max_vsi_frame_size(struct i40e_vsi *vsi,
+				   struct bpf_prog *xdp_prog)
 {
-	if (PAGE_SIZE >= 8192 || (vsi->back->flags & I40E_FLAG_LEGACY_RX))
-		return I40E_RXBUFFER_2048;
+	u16 rx_buf_len = i40e_calculate_vsi_rx_buf_len(vsi);
+	u16 chain_len;
+
+	if (xdp_prog && !xdp_prog->aux->xdp_has_frags)
+		chain_len = 1;
 	else
-		return I40E_RXBUFFER_3072;
+		chain_len = I40E_MAX_CHAINED_RX_BUFFERS;
+
+	return min_t(u16, rx_buf_len * chain_len, I40E_MAX_RXBUFFER);
 }
 
 /**
@@ -2919,11 +2939,12 @@ static int i40e_change_mtu(struct net_device *netdev, int new_mtu)
 	struct i40e_netdev_priv *np = netdev_priv(netdev);
 	struct i40e_vsi *vsi = np->vsi;
 	struct i40e_pf *pf = vsi->back;
+	int frame_size;
 
-	if (i40e_enabled_xdp_vsi(vsi)) {
-		int frame_size = new_mtu + I40E_PACKET_HDR_PAD;
-
-		if (frame_size > i40e_max_xdp_frame_size(vsi))
+	frame_size = i40e_max_vsi_frame_size(vsi, vsi->xdp_prog);
+	if (new_mtu > frame_size - I40E_PACKET_HDR_PAD) {
+		netdev_err(netdev, "Error changing mtu to %d, Max is %d\n",
+			   new_mtu, frame_size - I40E_PACKET_HDR_PAD);
 		return -EINVAL;
 	}
 
@@ -3595,6 +3616,8 @@ static int i40e_configure_rx_ring(struct i40e_ring *ring)
 		}
 	}
 
+	xdp_init_buff(&ring->xdp, i40e_rx_pg_size(ring) / 2, &ring->xdp_rxq);
+
 	rx_ctx.dbuff = DIV_ROUND_UP(ring->rx_buf_len,
 				    BIT_ULL(I40E_RXQ_CTX_DBUFF_SHIFT));
 
@@ -3640,10 +3663,16 @@ static int i40e_configure_rx_ring(struct i40e_ring *ring)
 	}
 
 	/* configure Rx buffer alignment */
-	if (!vsi->netdev || (vsi->back->flags & I40E_FLAG_LEGACY_RX))
+	if (!vsi->netdev || (vsi->back->flags & I40E_FLAG_LEGACY_RX)) {
+		if (I40E_2K_TOO_SMALL_WITH_PADDING) {
+			dev_info(&vsi->back->pdev->dev,
+				 "2k Rx buffer is too small to fit standard MTU and skb_shared_info\n");
+			return -EOPNOTSUPP;
+		}
 		clear_ring_build_skb_enabled(ring);
-	else
+	} else {
 		set_ring_build_skb_enabled(ring);
+	}
 
 	ring->rx_offset = i40e_rx_offset(ring);
 
@@ -3693,24 +3722,6 @@ static int i40e_vsi_configure_tx(struct i40e_vsi *vsi)
 	return err;
 }
 
-/**
- * i40e_calculate_vsi_rx_buf_len - Calculates buffer length
- *
- * @vsi: VSI to calculate rx_buf_len from
- */
-static u16 i40e_calculate_vsi_rx_buf_len(struct i40e_vsi *vsi)
-{
-	if (!vsi->netdev || (vsi->back->flags & I40E_FLAG_LEGACY_RX))
-		return I40E_RXBUFFER_2048;
-
-#if (PAGE_SIZE < 8192)
-	if (!I40E_2K_TOO_SMALL_WITH_PADDING && vsi->netdev->mtu <= ETH_DATA_LEN)
-		return I40E_RXBUFFER_1536 - NET_IP_ALIGN;
-#endif
-
-	return PAGE_SIZE < 8192 ? I40E_RXBUFFER_3072 : I40E_RXBUFFER_2048;
-}
-
 /**
  * i40e_vsi_configure_rx - Configure the VSI for Rx
  * @vsi: the VSI being configured
@@ -3722,13 +3733,15 @@ static int i40e_vsi_configure_rx(struct i40e_vsi *vsi)
 	int err = 0;
 	u16 i;
 
-	vsi->max_frame = I40E_MAX_RXBUFFER;
+	vsi->max_frame = i40e_max_vsi_frame_size(vsi, vsi->xdp_prog);
 	vsi->rx_buf_len = i40e_calculate_vsi_rx_buf_len(vsi);
 
 #if (PAGE_SIZE < 8192)
 	if (vsi->netdev && !I40E_2K_TOO_SMALL_WITH_PADDING &&
-	    vsi->netdev->mtu <= ETH_DATA_LEN)
-		vsi->max_frame = I40E_RXBUFFER_1536 - NET_IP_ALIGN;
+	    vsi->netdev->mtu <= ETH_DATA_LEN) {
+		vsi->rx_buf_len = I40E_RXBUFFER_1536 - NET_IP_ALIGN;
+		vsi->max_frame = vsi->rx_buf_len;
+	}
 #endif
 
 	/* set up individual rings */
@@ -13316,15 +13329,15 @@ static netdev_features_t i40e_features_check(struct sk_buff *skb,
 static int i40e_xdp_setup(struct i40e_vsi *vsi, struct bpf_prog *prog,
 			  struct netlink_ext_ack *extack)
 {
-	int frame_size = vsi->netdev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
+	int frame_size = i40e_max_vsi_frame_size(vsi, prog);
 	struct i40e_pf *pf = vsi->back;
 	struct bpf_prog *old_prog;
 	bool need_reset;
 	int i;
 
 	/* Don't allow frames that span over multiple buffers */
-	if (frame_size > i40e_calculate_vsi_rx_buf_len(vsi)) {
-		NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP");
+	if (vsi->netdev->mtu > frame_size - I40E_PACKET_HDR_PAD) {
+		NL_SET_ERR_MSG_MOD(extack, "MTU too large for linear frames and XDP prog does not support frags");
 		return -EINVAL;
 	}
 
@@ -13810,7 +13823,8 @@ static int i40e_config_netdev(struct i40e_vsi *vsi)
 
 		netdev->xdp_features = NETDEV_XDP_ACT_BASIC |
 				       NETDEV_XDP_ACT_REDIRECT |
-				       NETDEV_XDP_ACT_XSK_ZEROCOPY;
+				       NETDEV_XDP_ACT_XSK_ZEROCOPY |
+				       NETDEV_XDP_ACT_RX_SG;
 	} else {
 		/* Relate the VSI_VMDQ name to the VSI_MAIN name. Note that we
 		 * are still limited by IFNAMSIZ, but we're adding 'v%d\0' to

drivers/net/ethernet/intel/i40e/i40e_trace.h

@@ -162,45 +162,45 @@ DECLARE_EVENT_CLASS(
 
 	TP_PROTO(struct i40e_ring *ring,
 		 union i40e_16byte_rx_desc *desc,
-		 struct sk_buff *skb),
+		 struct xdp_buff *xdp),
 
-	TP_ARGS(ring, desc, skb),
+	TP_ARGS(ring, desc, xdp),
 
 	TP_STRUCT__entry(
 		__field(void*, ring)
 		__field(void*, desc)
-		__field(void*, skb)
+		__field(void*, xdp)
 		__string(devname, ring->netdev->name)
 	),
 
 	TP_fast_assign(
 		__entry->ring = ring;
 		__entry->desc = desc;
-		__entry->skb = skb;
+		__entry->xdp = xdp;
 		__assign_str(devname, ring->netdev->name);
 	),
 
 	TP_printk(
-		"netdev: %s ring: %p desc: %p skb %p",
+		"netdev: %s ring: %p desc: %p xdp %p",
 		__get_str(devname), __entry->ring,
-		__entry->desc, __entry->skb)
+		__entry->desc, __entry->xdp)
 );
 
 DEFINE_EVENT(
 	i40e_rx_template, i40e_clean_rx_irq,
 	TP_PROTO(struct i40e_ring *ring,
 		 union i40e_16byte_rx_desc *desc,
-		 struct sk_buff *skb),
+		 struct xdp_buff *xdp),
 
-	TP_ARGS(ring, desc, skb));
+	TP_ARGS(ring, desc, xdp));
 
 DEFINE_EVENT(
 	i40e_rx_template, i40e_clean_rx_irq_rx,
 	TP_PROTO(struct i40e_ring *ring,
 		 union i40e_16byte_rx_desc *desc,
-		 struct sk_buff *skb),
+		 struct xdp_buff *xdp),
 
-	TP_ARGS(ring, desc, skb));
+	TP_ARGS(ring, desc, xdp));
 
 DECLARE_EVENT_CLASS(
 	i40e_xmit_template,

drivers/net/ethernet/intel/i40e/i40e_txrx.c

@@ -1477,9 +1477,6 @@ void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
 	if (!rx_ring->rx_bi)
 		return;
 
-	dev_kfree_skb(rx_ring->skb);
-	rx_ring->skb = NULL;
-
 	if (rx_ring->xsk_pool) {
 		i40e_xsk_clean_rx_ring(rx_ring);
 		goto skip_free;
@@ -1524,6 +1521,7 @@ void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
 
 	rx_ring->next_to_alloc = 0;
 	rx_ring->next_to_clean = 0;
+	rx_ring->next_to_process = 0;
 	rx_ring->next_to_use = 0;
 }
 
@@ -1576,6 +1574,7 @@ int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
 
 	rx_ring->next_to_alloc = 0;
 	rx_ring->next_to_clean = 0;
+	rx_ring->next_to_process = 0;
 	rx_ring->next_to_use = 0;
 
 	/* XDP RX-queue info only needed for RX rings exposed to XDP */
@@ -1617,21 +1616,19 @@ void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
 	writel(val, rx_ring->tail);
 }
 
+#if (PAGE_SIZE >= 8192)
 static unsigned int i40e_rx_frame_truesize(struct i40e_ring *rx_ring,
 					   unsigned int size)
 {
 	unsigned int truesize;
 
-#if (PAGE_SIZE < 8192)
-	truesize = i40e_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
-#else
 	truesize = rx_ring->rx_offset ?
 		SKB_DATA_ALIGN(size + rx_ring->rx_offset) +
 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
 		SKB_DATA_ALIGN(size);
-#endif
 	return truesize;
 }
+#endif
 
 /**
  * i40e_alloc_mapped_page - recycle or make a new page
@@ -1970,7 +1967,6 @@ static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb,
  * i40e_can_reuse_rx_page - Determine if page can be reused for another Rx
  * @rx_buffer: buffer containing the page
  * @rx_stats: rx stats structure for the rx ring
- * @rx_buffer_pgcnt: buffer page refcount pre xdp_do_redirect() call
  *
  * If page is reusable, we have a green light for calling i40e_reuse_rx_page,
  * which will assign the current buffer to the buffer that next_to_alloc is
@@ -1981,8 +1977,7 @@ static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb,
  * or busy if it could not be reused.
  */
 static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer,
-				   struct i40e_rx_queue_stats *rx_stats,
-				   int rx_buffer_pgcnt)
+				   struct i40e_rx_queue_stats *rx_stats)
 {
 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
 	struct page *page = rx_buffer->page;
@@ -1995,7 +1990,7 @@ static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer,
 
 #if (PAGE_SIZE < 8192)
 	/* if we are only owner of page we can reuse it */
-	if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1)) {
+	if (unlikely((rx_buffer->page_count - pagecnt_bias) > 1)) {
 		rx_stats->page_busy_count++;
 		return false;
 	}
@@ -2021,32 +2016,13 @@ static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer,
 }
 
 /**
- * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
- * @rx_ring: rx descriptor ring to transact packets on
- * @rx_buffer: buffer containing page to add
- * @skb: sk_buff to place the data into
- * @size: packet length from rx_desc
- *
- * This function will add the data contained in rx_buffer->page to the skb.
- * It will just attach the page as a frag to the skb.
- *
- * The function will then update the page offset.
+ * i40e_rx_buffer_flip - adjusted rx_buffer to point to an unused region
+ * @rx_buffer: Rx buffer to adjust
+ * @truesize: Size of adjustment
  **/
-static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
-			     struct i40e_rx_buffer *rx_buffer,
-			     struct sk_buff *skb,
-			     unsigned int size)
+static void i40e_rx_buffer_flip(struct i40e_rx_buffer *rx_buffer,
+				unsigned int truesize)
 {
-#if (PAGE_SIZE < 8192)
-	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
-#else
-	unsigned int truesize = SKB_DATA_ALIGN(size + rx_ring->rx_offset);
-#endif
-
-	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
-			rx_buffer->page_offset, size, truesize);
-
-	/* page is being used so we must update the page offset */
 #if (PAGE_SIZE < 8192)
 	rx_buffer->page_offset ^= truesize;
 #else
@@ -2058,19 +2034,17 @@ static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
  * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
  * @rx_ring: rx descriptor ring to transact packets on
  * @size: size of buffer to add to skb
- * @rx_buffer_pgcnt: buffer page refcount
  *
  * This function will pull an Rx buffer from the ring and synchronize it
  * for use by the CPU.
  */
 static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
-						 const unsigned int size,
-						 int *rx_buffer_pgcnt)
+						 const unsigned int size)
 {
 	struct i40e_rx_buffer *rx_buffer;
 
-	rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
-	*rx_buffer_pgcnt =
+	rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_process);
+	rx_buffer->page_count =
 #if (PAGE_SIZE < 8192)
 		page_count(rx_buffer->page);
 #else
@@ -2092,25 +2066,82 @@ static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
 }
 
 /**
- * i40e_construct_skb - Allocate skb and populate it
+ * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
  * @rx_ring: rx descriptor ring to transact packets on
  * @rx_buffer: rx buffer to pull data from
+ *
+ * This function will clean up the contents of the rx_buffer.  It will
+ * either recycle the buffer or unmap it and free the associated resources.
+ */
+static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
+			       struct i40e_rx_buffer *rx_buffer)
+{
+	if (i40e_can_reuse_rx_page(rx_buffer, &rx_ring->rx_stats)) {
+		/* hand second half of page back to the ring */
+		i40e_reuse_rx_page(rx_ring, rx_buffer);
+	} else {
+		/* we are not reusing the buffer so unmap it */
+		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
+				     i40e_rx_pg_size(rx_ring),
+				     DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
+		__page_frag_cache_drain(rx_buffer->page,
+					rx_buffer->pagecnt_bias);
+		/* clear contents of buffer_info */
+		rx_buffer->page = NULL;
+	}
+}
+
+/**
+ * i40e_process_rx_buffs- Processing of buffers post XDP prog or on error
+ * @rx_ring: Rx descriptor ring to transact packets on
+ * @xdp_res: Result of the XDP program
  * @xdp: xdp_buff pointing to the data
+ **/
+static void i40e_process_rx_buffs(struct i40e_ring *rx_ring, int xdp_res,
+				  struct xdp_buff *xdp)
+{
+	u32 next = rx_ring->next_to_clean;
+	struct i40e_rx_buffer *rx_buffer;
+
+	xdp->flags = 0;
+
+	while (1) {
+		rx_buffer = i40e_rx_bi(rx_ring, next);
+		if (++next == rx_ring->count)
+			next = 0;
+
+		if (!rx_buffer->page)
+			continue;
+
+		if (xdp_res == I40E_XDP_CONSUMED)
+			rx_buffer->pagecnt_bias++;
+		else
+			i40e_rx_buffer_flip(rx_buffer, xdp->frame_sz);
+
+		/* EOP buffer will be put in i40e_clean_rx_irq() */
+		if (next == rx_ring->next_to_process)
+			return;
+
+		i40e_put_rx_buffer(rx_ring, rx_buffer);
+	}
+}
+
+/**
+ * i40e_construct_skb - Allocate skb and populate it
+ * @rx_ring: rx descriptor ring to transact packets on
+ * @xdp: xdp_buff pointing to the data
+ * @nr_frags: number of buffers for the packet
  *
  * This function allocates an skb.  It then populates it with the page
  * data from the current receive descriptor, taking care to set up the
  * skb correctly.
  */
 static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
-					  struct i40e_rx_buffer *rx_buffer,
-					  struct xdp_buff *xdp)
+					  struct xdp_buff *xdp,
+					  u32 nr_frags)
 {
 	unsigned int size = xdp->data_end - xdp->data;
-#if (PAGE_SIZE < 8192)
-	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
-#else
-	unsigned int truesize = SKB_DATA_ALIGN(size);
-#endif
+	struct i40e_rx_buffer *rx_buffer;
 	unsigned int headlen;
 	struct sk_buff *skb;
 
@@ -2150,48 +2181,60 @@ static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
 	memcpy(__skb_put(skb, headlen), xdp->data,
 	       ALIGN(headlen, sizeof(long)));
 
+	rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
 	/* update all of the pointers */
 	size -= headlen;
 	if (size) {
+		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
+			dev_kfree_skb(skb);
+			return NULL;
+		}
 		skb_add_rx_frag(skb, 0, rx_buffer->page,
 				rx_buffer->page_offset + headlen,
-				size, truesize);
-
+				size, xdp->frame_sz);
 		/* buffer is used by skb, update page_offset */
-#if (PAGE_SIZE < 8192)
-		rx_buffer->page_offset ^= truesize;
-#else
-		rx_buffer->page_offset += truesize;
-#endif
+		i40e_rx_buffer_flip(rx_buffer, xdp->frame_sz);
 	} else {
 		/* buffer is unused, reset bias back to rx_buffer */
 		rx_buffer->pagecnt_bias++;
 	}
 
+	if (unlikely(xdp_buff_has_frags(xdp))) {
+		struct skb_shared_info *sinfo, *skinfo = skb_shinfo(skb);
+
+		sinfo = xdp_get_shared_info_from_buff(xdp);
+		memcpy(&skinfo->frags[skinfo->nr_frags], &sinfo->frags[0],
+		       sizeof(skb_frag_t) * nr_frags);
+
+		xdp_update_skb_shared_info(skb, skinfo->nr_frags + nr_frags,
+					   sinfo->xdp_frags_size,
+					   nr_frags * xdp->frame_sz,
+					   xdp_buff_is_frag_pfmemalloc(xdp));
+
+		/* First buffer has already been processed, so bump ntc */
+		if (++rx_ring->next_to_clean == rx_ring->count)
+			rx_ring->next_to_clean = 0;
+
+		i40e_process_rx_buffs(rx_ring, I40E_XDP_PASS, xdp);
+	}
+
 	return skb;
 }
 
 /**
  * i40e_build_skb - Build skb around an existing buffer
  * @rx_ring: Rx descriptor ring to transact packets on
- * @rx_buffer: Rx buffer to pull data from
  * @xdp: xdp_buff pointing to the data
+ * @nr_frags: number of buffers for the packet
  *
  * This function builds an skb around an existing Rx buffer, taking care
  * to set up the skb correctly and avoid any memcpy overhead.
  */
 static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
-				      struct i40e_rx_buffer *rx_buffer,
-				      struct xdp_buff *xdp)
+				      struct xdp_buff *xdp,
+				      u32 nr_frags)
 {
 	unsigned int metasize = xdp->data - xdp->data_meta;
-#if (PAGE_SIZE < 8192)
-	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
-#else
-	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
-				SKB_DATA_ALIGN(xdp->data_end -
-					       xdp->data_hard_start);
-#endif
 	struct sk_buff *skb;
 
 	/* Prefetch first cache line of first page. If xdp->data_meta
@@ -2202,7 +2245,7 @@ static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
 	net_prefetch(xdp->data_meta);
 
 	/* build an skb around the page buffer */
-	skb = napi_build_skb(xdp->data_hard_start, truesize);
+	skb = napi_build_skb(xdp->data_hard_start, xdp->frame_sz);
 	if (unlikely(!skb))
 		return NULL;
 
@@ -2212,42 +2255,25 @@ static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
 	if (metasize)
 		skb_metadata_set(skb, metasize);
 
-	/* buffer is used by skb, update page_offset */
-#if (PAGE_SIZE < 8192)
-	rx_buffer->page_offset ^= truesize;
-#else
-	rx_buffer->page_offset += truesize;
-#endif
+	if (unlikely(xdp_buff_has_frags(xdp))) {
+		struct skb_shared_info *sinfo;
 
-	return skb;
-}
+		sinfo = xdp_get_shared_info_from_buff(xdp);
+		xdp_update_skb_shared_info(skb, nr_frags,
+					   sinfo->xdp_frags_size,
+					   nr_frags * xdp->frame_sz,
+					   xdp_buff_is_frag_pfmemalloc(xdp));
 
-/**
- * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
- * @rx_ring: rx descriptor ring to transact packets on
- * @rx_buffer: rx buffer to pull data from
- * @rx_buffer_pgcnt: rx buffer page refcount pre xdp_do_redirect() call
- *
- * This function will clean up the contents of the rx_buffer.  It will
- * either recycle the buffer or unmap it and free the associated resources.
- */
-static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
-			       struct i40e_rx_buffer *rx_buffer,
-			       int rx_buffer_pgcnt)
-{
-	if (i40e_can_reuse_rx_page(rx_buffer, &rx_ring->rx_stats, rx_buffer_pgcnt)) {
-		/* hand second half of page back to the ring */
-		i40e_reuse_rx_page(rx_ring, rx_buffer);
+		i40e_process_rx_buffs(rx_ring, I40E_XDP_PASS, xdp);
 	} else {
-		/* we are not reusing the buffer so unmap it */
-		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
-				     i40e_rx_pg_size(rx_ring),
-				     DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
-		__page_frag_cache_drain(rx_buffer->page,
-					rx_buffer->pagecnt_bias);
-		/* clear contents of buffer_info */
-		rx_buffer->page = NULL;
+		struct i40e_rx_buffer *rx_buffer;
+
+		rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
+		/* buffer is used by skb, update page_offset */
+		i40e_rx_buffer_flip(rx_buffer, xdp->frame_sz);
 	}
+
+	return skb;
 }
 
 /**
@@ -2332,25 +2358,6 @@ static int i40e_run_xdp(struct i40e_ring *rx_ring, struct xdp_buff *xdp, struct
 	return result;
 }
 
-/**
- * i40e_rx_buffer_flip - adjusted rx_buffer to point to an unused region
- * @rx_ring: Rx ring
- * @rx_buffer: Rx buffer to adjust
- * @size: Size of adjustment
- **/
-static void i40e_rx_buffer_flip(struct i40e_ring *rx_ring,
-				struct i40e_rx_buffer *rx_buffer,
-				unsigned int size)
-{
-	unsigned int truesize = i40e_rx_frame_truesize(rx_ring, size);
-
-#if (PAGE_SIZE < 8192)
-	rx_buffer->page_offset ^= truesize;
-#else
-	rx_buffer->page_offset += truesize;
-#endif
-}
-
 /**
  * i40e_xdp_ring_update_tail - Updates the XDP Tx ring tail register
  * @xdp_ring: XDP Tx ring
@@ -2409,16 +2416,65 @@ void i40e_finalize_xdp_rx(struct i40e_ring *rx_ring, unsigned int xdp_res)
 }
 
 /**
- * i40e_inc_ntc: Advance the next_to_clean index
+ * i40e_inc_ntp: Advance the next_to_process index
  * @rx_ring: Rx ring
  **/
-static void i40e_inc_ntc(struct i40e_ring *rx_ring)
+static void i40e_inc_ntp(struct i40e_ring *rx_ring)
+{
+	u32 ntp = rx_ring->next_to_process + 1;
+
+	ntp = (ntp < rx_ring->count) ? ntp : 0;
+	rx_ring->next_to_process = ntp;
+	prefetch(I40E_RX_DESC(rx_ring, ntp));
+}
+
+/**
+ * i40e_add_xdp_frag: Add a frag to xdp_buff
+ * @xdp: xdp_buff pointing to the data
+ * @nr_frags: return number of buffers for the packet
+ * @rx_buffer: rx_buffer holding data of the current frag
+ * @size: size of data of current frag
+ */
+static int i40e_add_xdp_frag(struct xdp_buff *xdp, u32 *nr_frags,
+			     struct i40e_rx_buffer *rx_buffer, u32 size)
 {
-	u32 ntc = rx_ring->next_to_clean + 1;
+	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
+
+	if (!xdp_buff_has_frags(xdp)) {
+		sinfo->nr_frags = 0;
+		sinfo->xdp_frags_size = 0;
+		xdp_buff_set_frags_flag(xdp);
+	} else if (unlikely(sinfo->nr_frags >= MAX_SKB_FRAGS)) {
+		/* Overflowing packet: All frags need to be dropped */
+		return -ENOMEM;
+	}
+
+	__skb_fill_page_desc_noacc(sinfo, sinfo->nr_frags++, rx_buffer->page,
+				   rx_buffer->page_offset, size);
 
-	ntc = (ntc < rx_ring->count) ? ntc : 0;
-	rx_ring->next_to_clean = ntc;
-	prefetch(I40E_RX_DESC(rx_ring, ntc));
+	sinfo->xdp_frags_size += size;
+
+	if (page_is_pfmemalloc(rx_buffer->page))
+		xdp_buff_set_frag_pfmemalloc(xdp);
+	*nr_frags = sinfo->nr_frags;
+
+	return 0;
+}
+
+/**
+ * i40e_consume_xdp_buff - Consume all the buffers of the packet and update ntc
+ * @rx_ring: rx descriptor ring to transact packets on
+ * @xdp: xdp_buff pointing to the data
+ * @rx_buffer: rx_buffer of eop desc
+ */
+static void i40e_consume_xdp_buff(struct i40e_ring *rx_ring,
+				  struct xdp_buff *xdp,
+				  struct i40e_rx_buffer *rx_buffer)
+{
+	i40e_process_rx_buffs(rx_ring, I40E_XDP_CONSUMED, xdp);
+	i40e_put_rx_buffer(rx_ring, rx_buffer);
+	rx_ring->next_to_clean = rx_ring->next_to_process;
+	xdp->data = NULL;
 }
 
 /**
@@ -2437,38 +2493,36 @@ static void i40e_inc_ntc(struct i40e_ring *rx_ring)
 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget,
 			     unsigned int *rx_cleaned)
 {
-	unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
+	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
+	u16 clean_threshold = rx_ring->count / 2;
 	unsigned int offset = rx_ring->rx_offset;
-	struct sk_buff *skb = rx_ring->skb;
+	struct xdp_buff *xdp = &rx_ring->xdp;
 	unsigned int xdp_xmit = 0;
 	struct bpf_prog *xdp_prog;
 	bool failure = false;
-	struct xdp_buff xdp;
 	int xdp_res = 0;
 
-#if (PAGE_SIZE < 8192)
-	frame_sz = i40e_rx_frame_truesize(rx_ring, 0);
-#endif
-	xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
-
 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
 
 	while (likely(total_rx_packets < (unsigned int)budget)) {
+		u16 ntp = rx_ring->next_to_process;
 		struct i40e_rx_buffer *rx_buffer;
 		union i40e_rx_desc *rx_desc;
-		int rx_buffer_pgcnt;
+		struct sk_buff *skb;
 		unsigned int size;
+		u32 nfrags = 0;
+		bool neop;
 		u64 qword;
 
 		/* return some buffers to hardware, one at a time is too slow */
-		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
+		if (cleaned_count >= clean_threshold) {
 			failure = failure ||
 				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
 			cleaned_count = 0;
 		}
 
-		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
+		rx_desc = I40E_RX_DESC(rx_ring, ntp);
 
 		/* status_error_len will always be zero for unused descriptors
 		 * because it's cleared in cleanup, and overlaps with hdr_addr
@@ -2487,8 +2541,8 @@ static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget,
 			i40e_clean_programming_status(rx_ring,
 						      rx_desc->raw.qword[0],
 						      qword);
-			rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
-			i40e_inc_ntc(rx_ring);
+			rx_buffer = i40e_rx_bi(rx_ring, ntp);
+			i40e_inc_ntp(rx_ring);
 			i40e_reuse_rx_page(rx_ring, rx_buffer);
 			cleaned_count++;
 			continue;
@@ -2499,59 +2553,62 @@ static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget,
 		if (!size)
 			break;
 
-		i40e_trace(clean_rx_irq, rx_ring, rx_desc, skb);
-		rx_buffer = i40e_get_rx_buffer(rx_ring, size, &rx_buffer_pgcnt);
-
+		i40e_trace(clean_rx_irq, rx_ring, rx_desc, xdp);
 		/* retrieve a buffer from the ring */
-		if (!skb) {
+		rx_buffer = i40e_get_rx_buffer(rx_ring, size);
+
+		neop = i40e_is_non_eop(rx_ring, rx_desc);
+		i40e_inc_ntp(rx_ring);
+
+		if (!xdp->data) {
 			unsigned char *hard_start;
 
 			hard_start = page_address(rx_buffer->page) +
 				     rx_buffer->page_offset - offset;
-			xdp_prepare_buff(&xdp, hard_start, offset, size, true);
-			xdp_buff_clear_frags_flag(&xdp);
+			xdp_prepare_buff(xdp, hard_start, offset, size, true);
 #if (PAGE_SIZE > 4096)
 			/* At larger PAGE_SIZE, frame_sz depend on len size */
-			xdp.frame_sz = i40e_rx_frame_truesize(rx_ring, size);
+			xdp->frame_sz = i40e_rx_frame_truesize(rx_ring, size);
 #endif
-			xdp_res = i40e_run_xdp(rx_ring, &xdp, xdp_prog);
+		} else if (i40e_add_xdp_frag(xdp, &nfrags, rx_buffer, size) &&
+			   !neop) {
+			/* Overflowing packet: Drop all frags on EOP */
+			i40e_consume_xdp_buff(rx_ring, xdp, rx_buffer);
+			break;
 		}
 
+		if (neop)
+			continue;
+
+		xdp_res = i40e_run_xdp(rx_ring, xdp, xdp_prog);
+
 		if (xdp_res) {
-			if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
-				xdp_xmit |= xdp_res;
-				i40e_rx_buffer_flip(rx_ring, rx_buffer, size);
+			xdp_xmit |= xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR);
+
+			if (unlikely(xdp_buff_has_frags(xdp))) {
+				i40e_process_rx_buffs(rx_ring, xdp_res, xdp);
+				size = xdp_get_buff_len(xdp);
+			} else if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
+				i40e_rx_buffer_flip(rx_buffer, xdp->frame_sz);
 			} else {
 				rx_buffer->pagecnt_bias++;
 			}
 			total_rx_bytes += size;
-			total_rx_packets++;
-		} else if (skb) {
-			i40e_add_rx_frag(rx_ring, rx_buffer, skb, size);
-		} else if (ring_uses_build_skb(rx_ring)) {
-			skb = i40e_build_skb(rx_ring, rx_buffer, &xdp);
 		} else {
-			skb = i40e_construct_skb(rx_ring, rx_buffer, &xdp);
-		}
+			if (ring_uses_build_skb(rx_ring))
+				skb = i40e_build_skb(rx_ring, xdp, nfrags);
+			else
+				skb = i40e_construct_skb(rx_ring, xdp, nfrags);
 
-		/* exit if we failed to retrieve a buffer */
-		if (!xdp_res && !skb) {
+			/* drop if we failed to retrieve a buffer */
+			if (!skb) {
 				rx_ring->rx_stats.alloc_buff_failed++;
-			rx_buffer->pagecnt_bias++;
+				i40e_consume_xdp_buff(rx_ring, xdp, rx_buffer);
 				break;
 			}
 
-		i40e_put_rx_buffer(rx_ring, rx_buffer, rx_buffer_pgcnt);
-		cleaned_count++;
-
-		i40e_inc_ntc(rx_ring);
-		if (i40e_is_non_eop(rx_ring, rx_desc))
-			continue;
-
-		if (xdp_res || i40e_cleanup_headers(rx_ring, skb, rx_desc)) {
-			skb = NULL;
-			continue;
-		}
+			if (i40e_cleanup_headers(rx_ring, skb, rx_desc))
+				goto process_next;
 
 			/* probably a little skewed due to removing CRC */
 			total_rx_bytes += skb->len;
@@ -2559,16 +2616,21 @@ static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget,
 			/* populate checksum, VLAN, and protocol */
 			i40e_process_skb_fields(rx_ring, rx_desc, skb);
 
-		i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, skb);
+			i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, xdp);
 			napi_gro_receive(&rx_ring->q_vector->napi, skb);
-		skb = NULL;
+		}
 
 		/* update budget accounting */
 		total_rx_packets++;
+process_next:
+		cleaned_count += nfrags + 1;
+		i40e_put_rx_buffer(rx_ring, rx_buffer);
+		rx_ring->next_to_clean = rx_ring->next_to_process;
+
+		xdp->data = NULL;
 	}
 
 	i40e_finalize_xdp_rx(rx_ring, xdp_xmit);
-	rx_ring->skb = skb;
 
 	i40e_update_rx_stats(rx_ring, total_rx_bytes, total_rx_packets);
 

drivers/net/ethernet/intel/i40e/i40e_txrx.h

@@ -277,6 +277,7 @@ struct i40e_rx_buffer {
 	struct page *page;
 	__u32 page_offset;
 	__u16 pagecnt_bias;
+	__u32 page_count;
 };
 
 struct i40e_queue_stats {
@@ -336,6 +337,17 @@ struct i40e_ring {
 	u8 dcb_tc;			/* Traffic class of ring */
 	u8 __iomem *tail;
 
+	/* Storing xdp_buff on ring helps in saving the state of partially built
+	 * packet when i40e_clean_rx_ring_irq() must return before it sees EOP
+	 * and to resume packet building for this ring in the next call to
+	 * i40e_clean_rx_ring_irq().
+	 */
+	struct xdp_buff xdp;
+
+	/* Next descriptor to be processed; next_to_clean is updated only on
+	 * processing EOP descriptor
+	 */
+	u16 next_to_process;
 	/* high bit set means dynamic, use accessor routines to read/write.
 	 * hardware only supports 2us resolution for the ITR registers.
 	 * these values always store the USER setting, and must be converted
@@ -380,14 +392,6 @@ struct i40e_ring {
 
 	struct rcu_head rcu;		/* to avoid race on free */
 	u16 next_to_alloc;
-	struct sk_buff *skb;		/* When i40e_clean_rx_ring_irq() must
-					 * return before it sees the EOP for
-					 * the current packet, we save that skb
-					 * here and resume receiving this
-					 * packet the next time
-					 * i40e_clean_rx_ring_irq() is called
-					 * for this ring.
-					 */
 
 	struct i40e_channel *ch;
 	u16 rx_offset;