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.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;