Merge branch 'XDP-redirect-memory-return-API'

Jesper Dangaard Brouer says:

====================
XDP redirect memory return API

Submitted against net-next, as it contains NIC driver changes.

This patchset works towards supporting different XDP RX-ring memory
allocators.  As this will be needed by the AF_XDP zero-copy mode.

The patchset uses mlx5 as the sample driver, which gets implemented
XDP_REDIRECT RX-mode, but not ndo_xdp_xmit (as this API is subject to
change thought the patchset).

A new struct xdp_frame is introduced (modeled after cpumap xdp_pkt).
And both ndo_xdp_xmit and the new xdp_return_frame end-up using this.

Support for a driver supplied allocator is implemented, and a
refurbished version of page_pool is the first return allocator type
introduced.  This will be a integration point for AF_XDP zero-copy.

The mlx5 driver evolve into using the page_pool, and see a performance
increase (with ndo_xdp_xmit out ixgbe driver) from 6Mpps to 12Mpps.

The patchset stop at 16 patches (one over limit), but more API changes
are planned.  Specifically extending ndo_xdp_xmit and xdp_return_frame
APIs to support bulking.  As this will address some known limits.

V2: Updated according to Tariq's feedback
V3: Updated based on feedback from Jason Wang and Alex Duyck
V4: Updated based on feedback from Tariq and Jason
V5: Fix SPDX license, add Tariq's reviews, improve patch desc for perf test
V6: Updated based on feedback from Eric Dumazet and Alex Duyck
V7: Adapt to i40e that got XDP_REDIRECT support in-between
V8:
 Updated based on feedback kbuild test robot, and adjust for mlx5 changes
 page_pool only compiled into kernel when drivers Kconfig 'select' feature
V9:
 Remove some inline statements, let compiler decide what to inline
 Fix return value in virtio_net driver
 Adjust for mlx5 changes in-between submissions
V10:
 Minor adjust for mlx5 requested by Tariq
 Resubmit against net-next
V11: avoid leaking info stored in frame data on page reuse
====================

Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

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

@@ -638,7 +638,7 @@ static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
 		if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
 			kfree(tx_buffer->raw_buf);
 		else if (ring_is_xdp(ring))
-			page_frag_free(tx_buffer->raw_buf);
+			xdp_return_frame(tx_buffer->xdpf);
 		else
 			dev_kfree_skb_any(tx_buffer->skb);
 		if (dma_unmap_len(tx_buffer, len))
@@ -841,7 +841,7 @@ static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
 
 		/* free the skb/XDP data */
 		if (ring_is_xdp(tx_ring))
-			page_frag_free(tx_buf->raw_buf);
+			xdp_return_frame(tx_buf->xdpf);
 		else
 			napi_consume_skb(tx_buf->skb, napi_budget);
 
@@ -2203,9 +2203,20 @@ static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
 #define I40E_XDP_CONSUMED 1
 #define I40E_XDP_TX 2
 
-static int i40e_xmit_xdp_ring(struct xdp_buff *xdp,
+static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
 			      struct i40e_ring *xdp_ring);
 
+static int i40e_xmit_xdp_tx_ring(struct xdp_buff *xdp,
+				 struct i40e_ring *xdp_ring)
+{
+	struct xdp_frame *xdpf = convert_to_xdp_frame(xdp);
+
+	if (unlikely(!xdpf))
+		return I40E_XDP_CONSUMED;
+
+	return i40e_xmit_xdp_ring(xdpf, xdp_ring);
+}
+
 /**
  * i40e_run_xdp - run an XDP program
  * @rx_ring: Rx ring being processed
@@ -2225,13 +2236,15 @@ static struct sk_buff *i40e_run_xdp(struct i40e_ring *rx_ring,
 	if (!xdp_prog)
 		goto xdp_out;
 
+	prefetchw(xdp->data_hard_start); /* xdp_frame write */
+
 	act = bpf_prog_run_xdp(xdp_prog, xdp);
 	switch (act) {
 	case XDP_PASS:
 		break;
 	case XDP_TX:
 		xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->queue_index];
-		result = i40e_xmit_xdp_ring(xdp, xdp_ring);
+		result = i40e_xmit_xdp_tx_ring(xdp, xdp_ring);
 		break;
 	case XDP_REDIRECT:
 		err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
@@ -3478,28 +3491,28 @@ static inline int i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
  * @xdp: data to transmit
  * @xdp_ring: XDP Tx ring
  **/
-static int i40e_xmit_xdp_ring(struct xdp_buff *xdp,
+static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
 			      struct i40e_ring *xdp_ring)
 {
-	u32 size = xdp->data_end - xdp->data;
 	u16 i = xdp_ring->next_to_use;
 	struct i40e_tx_buffer *tx_bi;
 	struct i40e_tx_desc *tx_desc;
+	u32 size = xdpf->len;
 	dma_addr_t dma;
 
 	if (!unlikely(I40E_DESC_UNUSED(xdp_ring))) {
 		xdp_ring->tx_stats.tx_busy++;
 		return I40E_XDP_CONSUMED;
 	}
 
-	dma = dma_map_single(xdp_ring->dev, xdp->data, size, DMA_TO_DEVICE);
+	dma = dma_map_single(xdp_ring->dev, xdpf->data, size, DMA_TO_DEVICE);
 	if (dma_mapping_error(xdp_ring->dev, dma))
 		return I40E_XDP_CONSUMED;
 
 	tx_bi = &xdp_ring->tx_bi[i];
 	tx_bi->bytecount = size;
 	tx_bi->gso_segs = 1;
-	tx_bi->raw_buf = xdp->data;
+	tx_bi->xdpf = xdpf;
 
 	/* record length, and DMA address */
 	dma_unmap_len_set(tx_bi, len, size);
@@ -3675,7 +3688,7 @@ netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
  *
  * Returns Zero if sent, else an error code
  **/
-int i40e_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
+int i40e_xdp_xmit(struct net_device *dev, struct xdp_frame *xdpf)
 {
 	struct i40e_netdev_priv *np = netdev_priv(dev);
 	unsigned int queue_index = smp_processor_id();
@@ -3688,7 +3701,7 @@ int i40e_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
 	if (!i40e_enabled_xdp_vsi(vsi) || queue_index >= vsi->num_queue_pairs)
 		return -ENXIO;
 
-	err = i40e_xmit_xdp_ring(xdp, vsi->xdp_rings[queue_index]);
+	err = i40e_xmit_xdp_ring(xdpf, vsi->xdp_rings[queue_index]);
 	if (err != I40E_XDP_TX)
 		return -ENOSPC;
 

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

@@ -306,6 +306,7 @@ static inline unsigned int i40e_txd_use_count(unsigned int size)
 struct i40e_tx_buffer {
 	struct i40e_tx_desc *next_to_watch;
 	union {
+		struct xdp_frame *xdpf;
 		struct sk_buff *skb;
 		void *raw_buf;
 	};
@@ -510,7 +511,7 @@ u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw);
 void i40e_detect_recover_hung(struct i40e_vsi *vsi);
 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
 bool __i40e_chk_linearize(struct sk_buff *skb);
-int i40e_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp);
+int i40e_xdp_xmit(struct net_device *dev, struct xdp_frame *xdpf);
 void i40e_xdp_flush(struct net_device *dev);
 
 /**

drivers/net/ethernet/intel/ixgbe/ixgbe.h

@@ -241,8 +241,7 @@ struct ixgbe_tx_buffer {
 	unsigned long time_stamp;
 	union {
 		struct sk_buff *skb;
-		/* XDP uses address ptr on irq_clean */
-		void *data;
+		struct xdp_frame *xdpf;
 	};
 	unsigned int bytecount;
 	unsigned short gso_segs;

drivers/net/ethernet/intel/ixgbe/ixgbe_main.c

@@ -1216,7 +1216,7 @@ static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
 
 		/* free the skb */
 		if (ring_is_xdp(tx_ring))
-			page_frag_free(tx_buffer->data);
+			xdp_return_frame(tx_buffer->xdpf);
 		else
 			napi_consume_skb(tx_buffer->skb, napi_budget);
 
@@ -2262,14 +2262,15 @@ static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
 #define IXGBE_XDP_TX 2
 
 static int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
-			       struct xdp_buff *xdp);
+			       struct xdp_frame *xdpf);
 
 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
 				     struct ixgbe_ring *rx_ring,
 				     struct xdp_buff *xdp)
 {
 	int err, result = IXGBE_XDP_PASS;
 	struct bpf_prog *xdp_prog;
+	struct xdp_frame *xdpf;
 	u32 act;
 
 	rcu_read_lock();
@@ -2278,12 +2279,19 @@ static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
 	if (!xdp_prog)
 		goto xdp_out;
 
+	prefetchw(xdp->data_hard_start); /* xdp_frame write */
+
 	act = bpf_prog_run_xdp(xdp_prog, xdp);
 	switch (act) {
 	case XDP_PASS:
 		break;
 	case XDP_TX:
-		result = ixgbe_xmit_xdp_ring(adapter, xdp);
+		xdpf = convert_to_xdp_frame(xdp);
+		if (unlikely(!xdpf)) {
+			result = IXGBE_XDP_CONSUMED;
+			break;
+		}
+		result = ixgbe_xmit_xdp_ring(adapter, xdpf);
 		break;
 	case XDP_REDIRECT:
 		err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
@@ -5797,7 +5805,7 @@ static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
 
 		/* Free all the Tx ring sk_buffs */
 		if (ring_is_xdp(tx_ring))
-			page_frag_free(tx_buffer->data);
+			xdp_return_frame(tx_buffer->xdpf);
 		else
 			dev_kfree_skb_any(tx_buffer->skb);
 
@@ -6370,7 +6378,7 @@ int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
 	struct device *dev = rx_ring->dev;
 	int orig_node = dev_to_node(dev);
 	int ring_node = -1;
-	int size;
+	int size, err;
 
 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
 
@@ -6407,6 +6415,13 @@ int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
 			     rx_ring->queue_index) < 0)
 		goto err;
 
+	err = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq,
+					 MEM_TYPE_PAGE_SHARED, NULL);
+	if (err) {
+		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
+		goto err;
+	}
+
 	rx_ring->xdp_prog = adapter->xdp_prog;
 
 	return 0;
@@ -8336,7 +8351,7 @@ static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
 }
 
 static int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
-			       struct xdp_buff *xdp)
+			       struct xdp_frame *xdpf)
 {
 	struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
 	struct ixgbe_tx_buffer *tx_buffer;
@@ -8345,12 +8360,12 @@ static int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
 	dma_addr_t dma;
 	u16 i;
 
-	len = xdp->data_end - xdp->data;
+	len = xdpf->len;
 
 	if (unlikely(!ixgbe_desc_unused(ring)))
 		return IXGBE_XDP_CONSUMED;
 
-	dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
+	dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
 	if (dma_mapping_error(ring->dev, dma))
 		return IXGBE_XDP_CONSUMED;
 
@@ -8365,7 +8380,8 @@ static int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
 
 	dma_unmap_len_set(tx_buffer, len, len);
 	dma_unmap_addr_set(tx_buffer, dma, dma);
-	tx_buffer->data = xdp->data;
+	tx_buffer->xdpf = xdpf;
+
 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
 
 	/* put descriptor type bits */
@@ -9996,7 +10012,7 @@ static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
 	}
 }
 
-static int ixgbe_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
+static int ixgbe_xdp_xmit(struct net_device *dev, struct xdp_frame *xdpf)
 {
 	struct ixgbe_adapter *adapter = netdev_priv(dev);
 	struct ixgbe_ring *ring;
@@ -10012,7 +10028,7 @@ static int ixgbe_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
 	if (unlikely(!ring))
 		return -ENXIO;
 
-	err = ixgbe_xmit_xdp_ring(adapter, xdp);
+	err = ixgbe_xmit_xdp_ring(adapter, xdpf);
 	if (err != IXGBE_XDP_TX)
 		return -ENOSPC;
 

drivers/net/ethernet/mellanox/mlx5/core/Kconfig

@@ -30,6 +30,7 @@ config MLX5_CORE_EN
 	bool "Mellanox Technologies ConnectX-4 Ethernet support"
 	depends on NETDEVICES && ETHERNET && INET && PCI && MLX5_CORE
 	depends on IPV6=y || IPV6=n || MLX5_CORE=m
+	select PAGE_POOL
 	default n
 	---help---
 	  Ethernet support in Mellanox Technologies ConnectX-4 NIC.

drivers/net/ethernet/mellanox/mlx5/core/en.h

@@ -53,6 +53,8 @@
 #include "mlx5_core.h"
 #include "en_stats.h"
 
+struct page_pool;
+
 #define MLX5_SET_CFG(p, f, v) MLX5_SET(create_flow_group_in, p, f, v)
 
 #define MLX5E_ETH_HARD_MTU (ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN)
@@ -392,6 +394,7 @@ struct mlx5e_xdpsq {
 	struct {
 		struct mlx5e_dma_info     *di;
 		bool                       doorbell;
+		bool                       redirect_flush;
 	} db;
 
 	/* read only */
@@ -533,6 +536,7 @@ struct mlx5e_rq {
 	unsigned int           hw_mtu;
 	struct mlx5e_xdpsq     xdpsq;
 	DECLARE_BITMAP(flags, 8);
+	struct page_pool      *page_pool;
 
 	/* control */
 	struct mlx5_wq_ctrl    wq_ctrl;

drivers/net/ethernet/mellanox/mlx5/core/en_main.c

@@ -35,6 +35,7 @@
 #include <linux/mlx5/fs.h>
 #include <net/vxlan.h>
 #include <linux/bpf.h>
+#include <net/page_pool.h>
 #include "eswitch.h"
 #include "en.h"
 #include "en_tc.h"
@@ -389,10 +390,11 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 			  struct mlx5e_rq_param *rqp,
 			  struct mlx5e_rq *rq)
 {
+	struct page_pool_params pp_params = { 0 };
 	struct mlx5_core_dev *mdev = c->mdev;
 	void *rqc = rqp->rqc;
 	void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
-	u32 byte_count;
+	u32 byte_count, pool_size;
 	int npages;
 	int wq_sz;
 	int err;
@@ -432,9 +434,12 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 	rq->buff.map_dir = rq->xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
 	rq->buff.headroom = mlx5e_get_rq_headroom(mdev, params);
+	pool_size = 1 << params->log_rq_mtu_frames;
 
 	switch (rq->wq_type) {
 	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+
+		pool_size = MLX5_MPWRQ_PAGES_PER_WQE << mlx5e_mpwqe_get_log_rq_size(params);
 		rq->post_wqes = mlx5e_post_rx_mpwqes;
 		rq->dealloc_wqe = mlx5e_dealloc_rx_mpwqe;
 
@@ -512,6 +517,32 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 		rq->mkey_be = c->mkey_be;
 	}
 
+	/* Create a page_pool and register it with rxq */
+	pp_params.order     = rq->buff.page_order;
+	pp_params.flags     = 0; /* No-internal DMA mapping in page_pool */
+	pp_params.pool_size = pool_size;
+	pp_params.nid       = cpu_to_node(c->cpu);
+	pp_params.dev       = c->pdev;
+	pp_params.dma_dir   = rq->buff.map_dir;
+
+	/* page_pool can be used even when there is no rq->xdp_prog,
+	 * given page_pool does not handle DMA mapping there is no
+	 * required state to clear. And page_pool gracefully handle
+	 * elevated refcnt.
+	 */
+	rq->page_pool = page_pool_create(&pp_params);
+	if (IS_ERR(rq->page_pool)) {
+		if (rq->wq_type != MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ)
+			kfree(rq->wqe.frag_info);
+		err = PTR_ERR(rq->page_pool);
+		rq->page_pool = NULL;
+		goto err_rq_wq_destroy;
+	}
+	err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
+					 MEM_TYPE_PAGE_POOL, rq->page_pool);
+	if (err)
+		goto err_rq_wq_destroy;
+
 	for (i = 0; i < wq_sz; i++) {
 		struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
 
@@ -548,6 +579,8 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 	if (rq->xdp_prog)
 		bpf_prog_put(rq->xdp_prog);
 	xdp_rxq_info_unreg(&rq->xdp_rxq);
+	if (rq->page_pool)
+		page_pool_destroy(rq->page_pool);
 	mlx5_wq_destroy(&rq->wq_ctrl);
 
 	return err;
@@ -561,6 +594,8 @@ static void mlx5e_free_rq(struct mlx5e_rq *rq)
 		bpf_prog_put(rq->xdp_prog);
 
 	xdp_rxq_info_unreg(&rq->xdp_rxq);
+	if (rq->page_pool)
+		page_pool_destroy(rq->page_pool);
 
 	switch (rq->wq_type) {
 	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ: