Merge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
1GbE Intel Wired LAN Driver Updates 2021-05-20

This series contains updates to igc driver only.

Andre Guedes says:

This series adds AF_XDP zero-copy feature to igc driver.

The initial patches do some code refactoring, preparing the code base to
land the AF_XDP zero-copy feature, avoiding code duplications. The last
patches of the series are the ones implementing the feature.

The last patch which indeed implements AF_XDP zero-copy support was
originally way too lengthy so, for the sake of code review, I broke it
up into two patches: one adding support for the RX functionality and the
other one adding TX support.
---
v2:
Patch 8/9 - "igc: Enable RX via AF_XDP zero-copy"
 * In XDP_PASS flow, copy metadata too into the skb.
 * When HW timestamp is added by the NIC, after copying it into
   a local variable, update xdp_buff->data_meta so that
   metadata length when XDP program is called 0.
 * In igc_xdp_enable_pool(), call xsk_pool_dma_unmap() on
   failure.

Known issues:
 When an XDP application is running in Tx-Only mode with Zero-Copy
 enabled, it is not expected to add the frames to the fill-queue. I have
 noticed the following two issues in this scenario:
 - If XDP_USE_NEED_WAKEUP flag is not set by application, igc_poll()
   will go into infinite loop because the buffer allocation resulting
   in igc_clean_rx_irq_zc() indicating that all work is not done and NAPI
   should keep polling. This does not occur if XDP_USE_NEED_WAKEUP
   flag is set.
 - Since there are no buffers allocated by userspace for the fill
   queue, there is no memory allocated for the NIC to copy the data
   to. If the packet received is destined to the hardware queue where
   XDP application is running, no packets are received even on other
   queues.
 Both these issues can be mitigated by adding a few frames to the
 fill queue. The second issue can also be mitigated by making sure no
 packets are being received on the hardware queue where Rx is running.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/intel/igc/igc.h

@@ -118,6 +118,7 @@ struct igc_ring {
 	};
 
 	struct xdp_rxq_info xdp_rxq;
+	struct xsk_buff_pool *xsk_pool;
 } ____cacheline_internodealigned_in_smp;
 
 /* Board specific private data structure */
@@ -255,6 +256,11 @@ bool igc_has_link(struct igc_adapter *adapter);
 void igc_reset(struct igc_adapter *adapter);
 int igc_set_spd_dplx(struct igc_adapter *adapter, u32 spd, u8 dplx);
 void igc_update_stats(struct igc_adapter *adapter);
+void igc_disable_rx_ring(struct igc_ring *ring);
+void igc_enable_rx_ring(struct igc_ring *ring);
+void igc_disable_tx_ring(struct igc_ring *ring);
+void igc_enable_tx_ring(struct igc_ring *ring);
+int igc_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags);
 
 /* igc_dump declarations */
 void igc_rings_dump(struct igc_adapter *adapter);
@@ -390,8 +396,6 @@ enum igc_tx_flags {
 	/* olinfo flags */
 	IGC_TX_FLAGS_IPV4	= 0x10,
 	IGC_TX_FLAGS_CSUM	= 0x20,
-
-	IGC_TX_FLAGS_XDP	= 0x100,
 };
 
 enum igc_boards {
@@ -408,12 +412,19 @@ enum igc_boards {
 #define TXD_USE_COUNT(S)	DIV_ROUND_UP((S), IGC_MAX_DATA_PER_TXD)
 #define DESC_NEEDED	(MAX_SKB_FRAGS + 4)
 
+enum igc_tx_buffer_type {
+	IGC_TX_BUFFER_TYPE_SKB,
+	IGC_TX_BUFFER_TYPE_XDP,
+	IGC_TX_BUFFER_TYPE_XSK,
+};
+
 /* wrapper around a pointer to a socket buffer,
  * so a DMA handle can be stored along with the buffer
  */
 struct igc_tx_buffer {
 	union igc_adv_tx_desc *next_to_watch;
 	unsigned long time_stamp;
+	enum igc_tx_buffer_type type;
 	union {
 		struct sk_buff *skb;
 		struct xdp_frame *xdpf;
@@ -428,6 +439,8 @@ struct igc_tx_buffer {
 };
 
 struct igc_rx_buffer {
+	union {
+		struct {
 			dma_addr_t dma;
 			struct page *page;
 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
@@ -436,6 +449,9 @@ struct igc_rx_buffer {
 			__u16 page_offset;
 #endif
 			__u16 pagecnt_bias;
+		};
+		struct xdp_buff *xdp;
+	};
 };
 
 struct igc_q_vector {
@@ -521,7 +537,8 @@ enum igc_ring_flags_t {
 	IGC_RING_FLAG_RX_SCTP_CSUM,
 	IGC_RING_FLAG_RX_LB_VLAN_BSWAP,
 	IGC_RING_FLAG_TX_CTX_IDX,
-	IGC_RING_FLAG_TX_DETECT_HANG
+	IGC_RING_FLAG_TX_DETECT_HANG,
+	IGC_RING_FLAG_AF_XDP_ZC,
 };
 
 #define ring_uses_large_buffer(ring) \

drivers/net/ethernet/intel/igc/igc_base.h

@@ -78,9 +78,11 @@ union igc_adv_rx_desc {
 
 /* Additional Transmit Descriptor Control definitions */
 #define IGC_TXDCTL_QUEUE_ENABLE	0x02000000 /* Ena specific Tx Queue */
+#define IGC_TXDCTL_SWFLUSH	0x04000000 /* Transmit Software Flush */
 
 /* Additional Receive Descriptor Control definitions */
 #define IGC_RXDCTL_QUEUE_ENABLE	0x02000000 /* Ena specific Rx Queue */
+#define IGC_RXDCTL_SWFLUSH		0x04000000 /* Receive Software Flush */
 
 /* SRRCTL bit definitions */
 #define IGC_SRRCTL_BSIZEPKT_SHIFT		10 /* Shift _right_ */

drivers/net/ethernet/intel/igc/igc_main.c

@@ -11,7 +11,7 @@
 #include <linux/pm_runtime.h>
 #include <net/pkt_sched.h>
 #include <linux/bpf_trace.h>
-
+#include <net/xdp_sock_drv.h>
 #include <net/ipv6.h>
 
 #include "igc.h"
@@ -171,6 +171,14 @@ static void igc_get_hw_control(struct igc_adapter *adapter)
 	     ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
 }
 
+static void igc_unmap_tx_buffer(struct device *dev, struct igc_tx_buffer *buf)
+{
+	dma_unmap_single(dev, dma_unmap_addr(buf, dma),
+			 dma_unmap_len(buf, len), DMA_TO_DEVICE);
+
+	dma_unmap_len_set(buf, len, 0);
+}
+
 /**
  * igc_clean_tx_ring - Free Tx Buffers
  * @tx_ring: ring to be cleaned
@@ -179,20 +187,27 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)
 {
 	u16 i = tx_ring->next_to_clean;
 	struct igc_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
+	u32 xsk_frames = 0;
 
 	while (i != tx_ring->next_to_use) {
 		union igc_adv_tx_desc *eop_desc, *tx_desc;
 
-		if (tx_buffer->tx_flags & IGC_TX_FLAGS_XDP)
+		switch (tx_buffer->type) {
+		case IGC_TX_BUFFER_TYPE_XSK:
+			xsk_frames++;
+			break;
+		case IGC_TX_BUFFER_TYPE_XDP:
 			xdp_return_frame(tx_buffer->xdpf);
-		else
+			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
+			break;
+		case IGC_TX_BUFFER_TYPE_SKB:
 			dev_kfree_skb_any(tx_buffer->skb);
-
-		/* unmap skb header data */
-		dma_unmap_single(tx_ring->dev,
-				 dma_unmap_addr(tx_buffer, dma),
-				 dma_unmap_len(tx_buffer, len),
-				 DMA_TO_DEVICE);
+			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
+			break;
+		default:
+			netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n");
+			break;
+		}
 
 		/* check for eop_desc to determine the end of the packet */
 		eop_desc = tx_buffer->next_to_watch;
@@ -211,10 +226,7 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)
 
 			/* unmap any remaining paged data */
 			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);
+				igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 		}
 
 		/* move us one more past the eop_desc for start of next pkt */
@@ -226,6 +238,9 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)
 		}
 	}
 
+	if (tx_ring->xsk_pool && xsk_frames)
+		xsk_tx_completed(tx_ring->xsk_pool, xsk_frames);
+
 	/* reset BQL for queue */
 	netdev_tx_reset_queue(txring_txq(tx_ring));
 
@@ -346,11 +361,7 @@ static int igc_setup_all_tx_resources(struct igc_adapter *adapter)
 	return err;
 }
 
-/**
- * igc_clean_rx_ring - Free Rx Buffers per Queue
- * @rx_ring: ring to free buffers from
- */
-static void igc_clean_rx_ring(struct igc_ring *rx_ring)
+static void igc_clean_rx_ring_page_shared(struct igc_ring *rx_ring)
 {
 	u16 i = rx_ring->next_to_clean;
 
@@ -383,12 +394,39 @@ static void igc_clean_rx_ring(struct igc_ring *rx_ring)
 		if (i == rx_ring->count)
 			i = 0;
 	}
+}
+
+static void igc_clean_rx_ring_xsk_pool(struct igc_ring *ring)
+{
+	struct igc_rx_buffer *bi;
+	u16 i;
+
+	for (i = 0; i < ring->count; i++) {
+		bi = &ring->rx_buffer_info[i];
+		if (!bi->xdp)
+			continue;
+
+		xsk_buff_free(bi->xdp);
+		bi->xdp = NULL;
+	}
+}
 
-	clear_ring_uses_large_buffer(rx_ring);
+/**
+ * igc_clean_rx_ring - Free Rx Buffers per Queue
+ * @ring: ring to free buffers from
+ */
+static void igc_clean_rx_ring(struct igc_ring *ring)
+{
+	if (ring->xsk_pool)
+		igc_clean_rx_ring_xsk_pool(ring);
+	else
+		igc_clean_rx_ring_page_shared(ring);
 
-	rx_ring->next_to_alloc = 0;
-	rx_ring->next_to_clean = 0;
-	rx_ring->next_to_use = 0;
+	clear_ring_uses_large_buffer(ring);
+
+	ring->next_to_alloc = 0;
+	ring->next_to_clean = 0;
+	ring->next_to_use = 0;
 }
 
 /**
@@ -414,7 +452,7 @@ void igc_free_rx_resources(struct igc_ring *rx_ring)
 {
 	igc_clean_rx_ring(rx_ring);
 
-	igc_xdp_unregister_rxq_info(rx_ring);
+	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 
 	vfree(rx_ring->rx_buffer_info);
 	rx_ring->rx_buffer_info = NULL;
@@ -453,11 +491,16 @@ int igc_setup_rx_resources(struct igc_ring *rx_ring)
 {
 	struct net_device *ndev = rx_ring->netdev;
 	struct device *dev = rx_ring->dev;
+	u8 index = rx_ring->queue_index;
 	int size, desc_len, res;
 
-	res = igc_xdp_register_rxq_info(rx_ring);
-	if (res < 0)
+	res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, ndev, index,
+			       rx_ring->q_vector->napi.napi_id);
+	if (res < 0) {
+		netdev_err(ndev, "Failed to register xdp_rxq index %u\n",
+			   index);
 		return res;
+	}
 
 	size = sizeof(struct igc_rx_buffer) * rx_ring->count;
 	rx_ring->rx_buffer_info = vzalloc(size);
@@ -483,7 +526,7 @@ int igc_setup_rx_resources(struct igc_ring *rx_ring)
 	return 0;
 
 err:
-	igc_xdp_unregister_rxq_info(rx_ring);
+	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 	vfree(rx_ring->rx_buffer_info);
 	rx_ring->rx_buffer_info = NULL;
 	netdev_err(ndev, "Unable to allocate memory for Rx descriptor ring\n");
@@ -515,9 +558,14 @@ static int igc_setup_all_rx_resources(struct igc_adapter *adapter)
 	return err;
 }
 
-static bool igc_xdp_is_enabled(struct igc_adapter *adapter)
+static struct xsk_buff_pool *igc_get_xsk_pool(struct igc_adapter *adapter,
+					      struct igc_ring *ring)
 {
-	return !!adapter->xdp_prog;
+	if (!igc_xdp_is_enabled(adapter) ||
+	    !test_bit(IGC_RING_FLAG_AF_XDP_ZC, &ring->flags))
+		return NULL;
+
+	return xsk_get_pool_from_qid(ring->netdev, ring->queue_index);
 }
 
 /**
@@ -535,6 +583,20 @@ static void igc_configure_rx_ring(struct igc_adapter *adapter,
 	int reg_idx = ring->reg_idx;
 	u32 srrctl = 0, rxdctl = 0;
 	u64 rdba = ring->dma;
+	u32 buf_size;
+
+	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
+	ring->xsk_pool = igc_get_xsk_pool(adapter, ring);
+	if (ring->xsk_pool) {
+		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
+						   MEM_TYPE_XSK_BUFF_POOL,
+						   NULL));
+		xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
+	} else {
+		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
+						   MEM_TYPE_PAGE_SHARED,
+						   NULL));
+	}
 
 	if (igc_xdp_is_enabled(adapter))
 		set_ring_uses_large_buffer(ring);
@@ -558,12 +620,15 @@ static void igc_configure_rx_ring(struct igc_adapter *adapter,
 	ring->next_to_clean = 0;
 	ring->next_to_use = 0;
 
-	/* set descriptor configuration */
-	srrctl = IGC_RX_HDR_LEN << IGC_SRRCTL_BSIZEHDRSIZE_SHIFT;
-	if (ring_uses_large_buffer(ring))
-		srrctl |= IGC_RXBUFFER_3072 >> IGC_SRRCTL_BSIZEPKT_SHIFT;
+	if (ring->xsk_pool)
+		buf_size = xsk_pool_get_rx_frame_size(ring->xsk_pool);
+	else if (ring_uses_large_buffer(ring))
+		buf_size = IGC_RXBUFFER_3072;
 	else
-		srrctl |= IGC_RXBUFFER_2048 >> IGC_SRRCTL_BSIZEPKT_SHIFT;
+		buf_size = IGC_RXBUFFER_2048;
+
+	srrctl = IGC_RX_HDR_LEN << IGC_SRRCTL_BSIZEHDRSIZE_SHIFT;
+	srrctl |= buf_size >> IGC_SRRCTL_BSIZEPKT_SHIFT;
 	srrctl |= IGC_SRRCTL_DESCTYPE_ADV_ONEBUF;
 
 	wr32(IGC_SRRCTL(reg_idx), srrctl);
@@ -618,6 +683,8 @@ static void igc_configure_tx_ring(struct igc_adapter *adapter,
 	u64 tdba = ring->dma;
 	u32 txdctl = 0;
 
+	ring->xsk_pool = igc_get_xsk_pool(adapter, ring);
+
 	/* disable the queue */
 	wr32(IGC_TXDCTL(reg_idx), 0);
 	wrfl();
@@ -1211,11 +1278,7 @@ static int igc_tx_map(struct igc_ring *tx_ring,
 	/* clear dma mappings for failed tx_buffer_info map */
 	while (tx_buffer != first) {
 		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);
+			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 
 		if (i-- == 0)
 			i += tx_ring->count;
@@ -1223,11 +1286,7 @@ static int igc_tx_map(struct igc_ring *tx_ring,
 	}
 
 	if (dma_unmap_len(tx_buffer, len))
-		dma_unmap_single(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);
+		igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 
 	dev_kfree_skb_any(tx_buffer->skb);
 	tx_buffer->skb = NULL;
@@ -1359,6 +1418,7 @@ static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
 
 	/* record the location of the first descriptor for this packet */
 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
+	first->type = IGC_TX_BUFFER_TYPE_SKB;
 	first->skb = skb;
 	first->bytecount = skb->len;
 	first->gso_segs = 1;
@@ -1930,6 +1990,63 @@ static void igc_alloc_rx_buffers(struct igc_ring *rx_ring, u16 cleaned_count)
 	}
 }
 
+static bool igc_alloc_rx_buffers_zc(struct igc_ring *ring, u16 count)
+{
+	union igc_adv_rx_desc *desc;
+	u16 i = ring->next_to_use;
+	struct igc_rx_buffer *bi;
+	dma_addr_t dma;
+	bool ok = true;
+
+	if (!count)
+		return ok;
+
+	desc = IGC_RX_DESC(ring, i);
+	bi = &ring->rx_buffer_info[i];
+	i -= ring->count;
+
+	do {
+		bi->xdp = xsk_buff_alloc(ring->xsk_pool);
+		if (!bi->xdp) {
+			ok = false;
+			break;
+		}
+
+		dma = xsk_buff_xdp_get_dma(bi->xdp);
+		desc->read.pkt_addr = cpu_to_le64(dma);
+
+		desc++;
+		bi++;
+		i++;
+		if (unlikely(!i)) {
+			desc = IGC_RX_DESC(ring, 0);
+			bi = ring->rx_buffer_info;
+			i -= ring->count;
+		}
+
+		/* Clear the length for the next_to_use descriptor. */
+		desc->wb.upper.length = 0;
+
+		count--;
+	} while (count);
+
+	i += ring->count;
+
+	if (ring->next_to_use != i) {
+		ring->next_to_use = i;
+
+		/* Force memory writes to complete before letting h/w
+		 * know there are new descriptors to fetch.  (Only
+		 * applicable for weak-ordered memory model archs,
+		 * such as IA-64).
+		 */
+		wmb();
+		writel(i, ring->tail);
+	}
+
+	return ok;
+}
+
 static int igc_xdp_init_tx_buffer(struct igc_tx_buffer *buffer,
 				  struct xdp_frame *xdpf,
 				  struct igc_ring *ring)
@@ -1942,8 +2059,8 @@ static int igc_xdp_init_tx_buffer(struct igc_tx_buffer *buffer,
 		return -ENOMEM;
 	}
 
+	buffer->type = IGC_TX_BUFFER_TYPE_XDP;
 	buffer->xdpf = xdpf;
-	buffer->tx_flags = IGC_TX_FLAGS_XDP;
 	buffer->protocol = 0;
 	buffer->bytecount = xdpf->len;
 	buffer->gso_segs = 1;
@@ -2025,38 +2142,22 @@ static int igc_xdp_xmit_back(struct igc_adapter *adapter, struct xdp_buff *xdp)
 	return res;
 }
 
-static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter,
+/* This function assumes rcu_read_lock() is held by the caller. */
+static int __igc_xdp_run_prog(struct igc_adapter *adapter,
+			      struct bpf_prog *prog,
 			      struct xdp_buff *xdp)
 {
-	struct bpf_prog *prog;
-	int res;
-	u32 act;
-
-	rcu_read_lock();
+	u32 act = bpf_prog_run_xdp(prog, xdp);
 
-	prog = READ_ONCE(adapter->xdp_prog);
-	if (!prog) {
-		res = IGC_XDP_PASS;
-		goto unlock;
-	}
-
-	act = bpf_prog_run_xdp(prog, xdp);
 	switch (act) {
 	case XDP_PASS:
-		res = IGC_XDP_PASS;
-		break;
+		return IGC_XDP_PASS;
 	case XDP_TX:
-		if (igc_xdp_xmit_back(adapter, xdp) < 0)
-			res = IGC_XDP_CONSUMED;
-		else
-			res = IGC_XDP_TX;
-		break;
+		return igc_xdp_xmit_back(adapter, xdp) < 0 ?
+			IGC_XDP_CONSUMED : IGC_XDP_TX;
 	case XDP_REDIRECT:
-		if (xdp_do_redirect(adapter->netdev, xdp, prog) < 0)
-			res = IGC_XDP_CONSUMED;
-		else
-			res = IGC_XDP_REDIRECT;
-		break;
+		return xdp_do_redirect(adapter->netdev, xdp, prog) < 0 ?
+			IGC_XDP_CONSUMED : IGC_XDP_REDIRECT;
 	default:
 		bpf_warn_invalid_xdp_action(act);
 		fallthrough;
@@ -2064,9 +2165,25 @@ static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter,
 		trace_xdp_exception(adapter->netdev, prog, act);
 		fallthrough;
 	case XDP_DROP:
-		res = IGC_XDP_CONSUMED;
-		break;
+		return IGC_XDP_CONSUMED;
 	}
+}
+
+static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter,
+					struct xdp_buff *xdp)
+{
+	struct bpf_prog *prog;
+	int res;
+
+	rcu_read_lock();
+
+	prog = READ_ONCE(adapter->xdp_prog);
+	if (!prog) {
+		res = IGC_XDP_PASS;
+		goto unlock;
+	}
+
+	res = __igc_xdp_run_prog(adapter, prog, xdp);
 
 unlock:
 	rcu_read_unlock();
@@ -2103,6 +2220,20 @@ static void igc_finalize_xdp(struct igc_adapter *adapter, int status)
 		xdp_do_flush();
 }
 
+static void igc_update_rx_stats(struct igc_q_vector *q_vector,
+				unsigned int packets, unsigned int bytes)
+{
+	struct igc_ring *ring = q_vector->rx.ring;
+
+	u64_stats_update_begin(&ring->rx_syncp);
+	ring->rx_stats.packets += packets;
+	ring->rx_stats.bytes += bytes;
+	u64_stats_update_end(&ring->rx_syncp);
+
+	q_vector->rx.total_packets += packets;
+	q_vector->rx.total_bytes += bytes;
+}
+
 static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
 {
 	unsigned int total_bytes = 0, total_packets = 0;
@@ -2223,12 +2354,7 @@ static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
 	/* place incomplete frames back on ring for completion */
 	rx_ring->skb = skb;
 
-	u64_stats_update_begin(&rx_ring->rx_syncp);
-	rx_ring->rx_stats.packets += total_packets;
-	rx_ring->rx_stats.bytes += total_bytes;
-	u64_stats_update_end(&rx_ring->rx_syncp);
-	q_vector->rx.total_packets += total_packets;
-	q_vector->rx.total_bytes += total_bytes;
+	igc_update_rx_stats(q_vector, total_packets, total_bytes);
 
 	if (cleaned_count)
 		igc_alloc_rx_buffers(rx_ring, cleaned_count);
@@ -2236,6 +2362,221 @@ static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
 	return total_packets;
 }
 
+static struct sk_buff *igc_construct_skb_zc(struct igc_ring *ring,
+					    struct xdp_buff *xdp)
+{
+	unsigned int metasize = xdp->data - xdp->data_meta;
+	unsigned int datasize = xdp->data_end - xdp->data;
+	unsigned int totalsize = metasize + datasize;
+	struct sk_buff *skb;
+
+	skb = __napi_alloc_skb(&ring->q_vector->napi,
+			       xdp->data_end - xdp->data_hard_start,
+			       GFP_ATOMIC | __GFP_NOWARN);
+	if (unlikely(!skb))
+		return NULL;
+
+	skb_reserve(skb, xdp->data_meta - xdp->data_hard_start);
+	memcpy(__skb_put(skb, totalsize), xdp->data_meta, totalsize);
+	if (metasize)
+		skb_metadata_set(skb, metasize);
+
+	return skb;
+}
+
+static void igc_dispatch_skb_zc(struct igc_q_vector *q_vector,
+				union igc_adv_rx_desc *desc,
+				struct xdp_buff *xdp,
+				ktime_t timestamp)
+{
+	struct igc_ring *ring = q_vector->rx.ring;
+	struct sk_buff *skb;
+
+	skb = igc_construct_skb_zc(ring, xdp);
+	if (!skb) {
+		ring->rx_stats.alloc_failed++;
+		return;
+	}
+
+	if (timestamp)
+		skb_hwtstamps(skb)->hwtstamp = timestamp;
+
+	if (igc_cleanup_headers(ring, desc, skb))
+		return;
+
+	igc_process_skb_fields(ring, desc, skb);
+	napi_gro_receive(&q_vector->napi, skb);
+}
+
+static int igc_clean_rx_irq_zc(struct igc_q_vector *q_vector, const int budget)
+{
+	struct igc_adapter *adapter = q_vector->adapter;
+	struct igc_ring *ring = q_vector->rx.ring;
+	u16 cleaned_count = igc_desc_unused(ring);
+	int total_bytes = 0, total_packets = 0;
+	u16 ntc = ring->next_to_clean;
+	struct bpf_prog *prog;
+	bool failure = false;
+	int xdp_status = 0;
+
+	rcu_read_lock();
+
+	prog = READ_ONCE(adapter->xdp_prog);
+
+	while (likely(total_packets < budget)) {
+		union igc_adv_rx_desc *desc;
+		struct igc_rx_buffer *bi;
+		ktime_t timestamp = 0;
+		unsigned int size;
+		int res;
+
+		desc = IGC_RX_DESC(ring, ntc);
+		size = le16_to_cpu(desc->wb.upper.length);
+		if (!size)
+			break;
+
+		/* This memory barrier is needed to keep us from reading
+		 * any other fields out of the rx_desc until we know the
+		 * descriptor has been written back
+		 */
+		dma_rmb();
+
+		bi = &ring->rx_buffer_info[ntc];
+
+		if (igc_test_staterr(desc, IGC_RXDADV_STAT_TSIP)) {
+			timestamp = igc_ptp_rx_pktstamp(q_vector->adapter,
+							bi->xdp->data);
+
+			bi->xdp->data += IGC_TS_HDR_LEN;
+
+			/* HW timestamp has been copied into local variable. Metadata
+			 * length when XDP program is called should be 0.
+			 */
+			bi->xdp->data_meta += IGC_TS_HDR_LEN;
+			size -= IGC_TS_HDR_LEN;
+		}
+
+		bi->xdp->data_end = bi->xdp->data + size;
+		xsk_buff_dma_sync_for_cpu(bi->xdp, ring->xsk_pool);
+
+		res = __igc_xdp_run_prog(adapter, prog, bi->xdp);
+		switch (res) {
+		case IGC_XDP_PASS:
+			igc_dispatch_skb_zc(q_vector, desc, bi->xdp, timestamp);
+			fallthrough;
+		case IGC_XDP_CONSUMED:
+			xsk_buff_free(bi->xdp);
+			break;
+		case IGC_XDP_TX:
+		case IGC_XDP_REDIRECT:
+			xdp_status |= res;
+			break;
+		}
+
+		bi->xdp = NULL;
+		total_bytes += size;
+		total_packets++;
+		cleaned_count++;
+		ntc++;
+		if (ntc == ring->count)
+			ntc = 0;
+	}
+
+	ring->next_to_clean = ntc;
+	rcu_read_unlock();
+
+	if (cleaned_count >= IGC_RX_BUFFER_WRITE)
+		failure = !igc_alloc_rx_buffers_zc(ring, cleaned_count);
+
+	if (xdp_status)
+		igc_finalize_xdp(adapter, xdp_status);
+
+	igc_update_rx_stats(q_vector, total_packets, total_bytes);
+
+	if (xsk_uses_need_wakeup(ring->xsk_pool)) {
+		if (failure || ring->next_to_clean == ring->next_to_use)
+			xsk_set_rx_need_wakeup(ring->xsk_pool);
+		else
+			xsk_clear_rx_need_wakeup(ring->xsk_pool);
+		return total_packets;
+	}
+
+	return failure ? budget : total_packets;
+}
+
+static void igc_update_tx_stats(struct igc_q_vector *q_vector,
+				unsigned int packets, unsigned int bytes)
+{
+	struct igc_ring *ring = q_vector->tx.ring;
+
+	u64_stats_update_begin(&ring->tx_syncp);
+	ring->tx_stats.bytes += bytes;
+	ring->tx_stats.packets += packets;
+	u64_stats_update_end(&ring->tx_syncp);
+
+	q_vector->tx.total_bytes += bytes;
+	q_vector->tx.total_packets += packets;
+}
+
+static void igc_xdp_xmit_zc(struct igc_ring *ring)
+{
+	struct xsk_buff_pool *pool = ring->xsk_pool;
+	struct netdev_queue *nq = txring_txq(ring);
+	union igc_adv_tx_desc *tx_desc = NULL;
+	int cpu = smp_processor_id();
+	u16 ntu = ring->next_to_use;
+	struct xdp_desc xdp_desc;
+	u16 budget;
+
+	if (!netif_carrier_ok(ring->netdev))
+		return;
+
+	__netif_tx_lock(nq, cpu);
+
+	budget = igc_desc_unused(ring);
+
+	while (xsk_tx_peek_desc(pool, &xdp_desc) && budget--) {
+		u32 cmd_type, olinfo_status;
+		struct igc_tx_buffer *bi;
+		dma_addr_t dma;
+
+		cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
+			   IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD |
+			   xdp_desc.len;
+		olinfo_status = xdp_desc.len << IGC_ADVTXD_PAYLEN_SHIFT;
+
+		dma = xsk_buff_raw_get_dma(pool, xdp_desc.addr);
+		xsk_buff_raw_dma_sync_for_device(pool, dma, xdp_desc.len);
+
+		tx_desc = IGC_TX_DESC(ring, ntu);
+		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
+		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
+		tx_desc->read.buffer_addr = cpu_to_le64(dma);
+
+		bi = &ring->tx_buffer_info[ntu];
+		bi->type = IGC_TX_BUFFER_TYPE_XSK;
+		bi->protocol = 0;
+		bi->bytecount = xdp_desc.len;
+		bi->gso_segs = 1;
+		bi->time_stamp = jiffies;
+		bi->next_to_watch = tx_desc;
+
+		netdev_tx_sent_queue(txring_txq(ring), xdp_desc.len);
+
+		ntu++;
+		if (ntu == ring->count)
+			ntu = 0;
+	}
+
+	ring->next_to_use = ntu;
+	if (tx_desc) {
+		igc_flush_tx_descriptors(ring);
+		xsk_tx_release(pool);
+	}
+
+	__netif_tx_unlock(nq);
+}
+
 /**
  * igc_clean_tx_irq - Reclaim resources after transmit completes
  * @q_vector: pointer to q_vector containing needed info
@@ -2252,6 +2593,7 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
 	unsigned int i = tx_ring->next_to_clean;
 	struct igc_tx_buffer *tx_buffer;
 	union igc_adv_tx_desc *tx_desc;
+	u32 xsk_frames = 0;
 
 	if (test_bit(__IGC_DOWN, &adapter->state))
 		return true;
@@ -2281,19 +2623,22 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
 		total_bytes += tx_buffer->bytecount;
 		total_packets += tx_buffer->gso_segs;
 
-		if (tx_buffer->tx_flags & IGC_TX_FLAGS_XDP)
+		switch (tx_buffer->type) {
+		case IGC_TX_BUFFER_TYPE_XSK:
+			xsk_frames++;
+			break;
+		case IGC_TX_BUFFER_TYPE_XDP:
 			xdp_return_frame(tx_buffer->xdpf);
-		else
+			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
+			break;
+		case IGC_TX_BUFFER_TYPE_SKB:
 			napi_consume_skb(tx_buffer->skb, napi_budget);
-
-		/* unmap skb header data */
-		dma_unmap_single(tx_ring->dev,
-				 dma_unmap_addr(tx_buffer, dma),
-				 dma_unmap_len(tx_buffer, len),
-				 DMA_TO_DEVICE);
-
-		/* clear tx_buffer data */
-		dma_unmap_len_set(tx_buffer, len, 0);
+			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
+			break;
+		default:
+			netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n");
+			break;
+		}
 
 		/* clear last DMA location and unmap remaining buffers */
 		while (tx_desc != eop_desc) {
@@ -2307,13 +2652,8 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
 			}
 
 			/* unmap any remaining paged data */
-			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 (dma_unmap_len(tx_buffer, len))
+				igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 		}
 
 		/* move us one more past the eop_desc for start of next pkt */
@@ -2338,12 +2678,16 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
 
 	i += tx_ring->count;
 	tx_ring->next_to_clean = i;
-	u64_stats_update_begin(&tx_ring->tx_syncp);
-	tx_ring->tx_stats.bytes += total_bytes;
-	tx_ring->tx_stats.packets += total_packets;
-	u64_stats_update_end(&tx_ring->tx_syncp);
-	q_vector->tx.total_bytes += total_bytes;
-	q_vector->tx.total_packets += total_packets;
+
+	igc_update_tx_stats(q_vector, total_packets, total_bytes);
+
+	if (tx_ring->xsk_pool) {
+		if (xsk_frames)
+			xsk_tx_completed(tx_ring->xsk_pool, xsk_frames);
+		if (xsk_uses_need_wakeup(tx_ring->xsk_pool))
+			xsk_set_tx_need_wakeup(tx_ring->xsk_pool);
+		igc_xdp_xmit_zc(tx_ring);
+	}
 
 	if (test_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
 		struct igc_hw *hw = &adapter->hw;
@@ -2923,6 +3267,9 @@ static void igc_configure(struct igc_adapter *adapter)
 	for (i = 0; i < adapter->num_rx_queues; i++) {
 		struct igc_ring *ring = adapter->rx_ring[i];
 
+		if (ring->xsk_pool)
+			igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring));
+		else
 			igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
 	}
 }
@@ -3538,14 +3885,17 @@ static int igc_poll(struct napi_struct *napi, int budget)
 	struct igc_q_vector *q_vector = container_of(napi,
 						     struct igc_q_vector,
 						     napi);
+	struct igc_ring *rx_ring = q_vector->rx.ring;
 	bool clean_complete = true;
 	int work_done = 0;
 
 	if (q_vector->tx.ring)
 		clean_complete = igc_clean_tx_irq(q_vector, budget);
 
-	if (q_vector->rx.ring) {
-		int cleaned = igc_clean_rx_irq(q_vector, budget);
+	if (rx_ring) {
+		int cleaned = rx_ring->xsk_pool ?
+			      igc_clean_rx_irq_zc(q_vector, budget) :
+			      igc_clean_rx_irq(q_vector, budget);
 
 		work_done += cleaned;
 		if (cleaned >= budget)
@@ -5183,6 +5533,9 @@ static int igc_bpf(struct net_device *dev, struct netdev_bpf *bpf)
 	switch (bpf->command) {
 	case XDP_SETUP_PROG:
 		return igc_xdp_set_prog(adapter, bpf->prog, bpf->extack);
+	case XDP_SETUP_XSK_POOL:
+		return igc_xdp_setup_pool(adapter, bpf->xsk.pool,
+					  bpf->xsk.queue_id);
 	default:
 		return -EOPNOTSUPP;
 	}
@@ -5228,6 +5581,43 @@ static int igc_xdp_xmit(struct net_device *dev, int num_frames,
 	return num_frames - drops;
 }
 
+static void igc_trigger_rxtxq_interrupt(struct igc_adapter *adapter,
+					struct igc_q_vector *q_vector)
+{
+	struct igc_hw *hw = &adapter->hw;
+	u32 eics = 0;
+
+	eics |= q_vector->eims_value;
+	wr32(IGC_EICS, eics);
+}
+
+int igc_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
+{
+	struct igc_adapter *adapter = netdev_priv(dev);
+	struct igc_q_vector *q_vector;
+	struct igc_ring *ring;
+
+	if (test_bit(__IGC_DOWN, &adapter->state))
+		return -ENETDOWN;
+
+	if (!igc_xdp_is_enabled(adapter))
+		return -ENXIO;
+
+	if (queue_id >= adapter->num_rx_queues)
+		return -EINVAL;
+
+	ring = adapter->rx_ring[queue_id];
+
+	if (!ring->xsk_pool)
+		return -ENXIO;
+
+	q_vector = adapter->q_vector[queue_id];
+	if (!napi_if_scheduled_mark_missed(&q_vector->napi))
+		igc_trigger_rxtxq_interrupt(adapter, q_vector);
+
+	return 0;
+}
+
 static const struct net_device_ops igc_netdev_ops = {
 	.ndo_open		= igc_open,
 	.ndo_stop		= igc_close,
@@ -5243,6 +5633,7 @@ static const struct net_device_ops igc_netdev_ops = {
 	.ndo_setup_tc		= igc_setup_tc,
 	.ndo_bpf		= igc_bpf,
 	.ndo_xdp_xmit		= igc_xdp_xmit,
+	.ndo_xsk_wakeup		= igc_xsk_wakeup,
 };
 
 /* PCIe configuration access */
@@ -5995,6 +6386,61 @@ struct net_device *igc_get_hw_dev(struct igc_hw *hw)
 	return adapter->netdev;
 }
 
+static void igc_disable_rx_ring_hw(struct igc_ring *ring)
+{
+	struct igc_hw *hw = &ring->q_vector->adapter->hw;
+	u8 idx = ring->reg_idx;
+	u32 rxdctl;
+
+	rxdctl = rd32(IGC_RXDCTL(idx));
+	rxdctl &= ~IGC_RXDCTL_QUEUE_ENABLE;
+	rxdctl |= IGC_RXDCTL_SWFLUSH;
+	wr32(IGC_RXDCTL(idx), rxdctl);
+}
+
+void igc_disable_rx_ring(struct igc_ring *ring)
+{
+	igc_disable_rx_ring_hw(ring);
+	igc_clean_rx_ring(ring);
+}
+
+void igc_enable_rx_ring(struct igc_ring *ring)
+{
+	struct igc_adapter *adapter = ring->q_vector->adapter;
+
+	igc_configure_rx_ring(adapter, ring);
+
+	if (ring->xsk_pool)
+		igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring));
+	else
+		igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
+}
+
+static void igc_disable_tx_ring_hw(struct igc_ring *ring)
+{
+	struct igc_hw *hw = &ring->q_vector->adapter->hw;
+	u8 idx = ring->reg_idx;
+	u32 txdctl;
+
+	txdctl = rd32(IGC_TXDCTL(idx));
+	txdctl &= ~IGC_TXDCTL_QUEUE_ENABLE;
+	txdctl |= IGC_TXDCTL_SWFLUSH;
+	wr32(IGC_TXDCTL(idx), txdctl);
+}
+
+void igc_disable_tx_ring(struct igc_ring *ring)
+{
+	igc_disable_tx_ring_hw(ring);
+	igc_clean_tx_ring(ring);
+}
+
+void igc_enable_tx_ring(struct igc_ring *ring)
+{
+	struct igc_adapter *adapter = ring->q_vector->adapter;
+
+	igc_configure_tx_ring(adapter, ring);
+}
+
 /**
  * igc_init_module - Driver Registration Routine
  *

drivers/net/ethernet/intel/igc/igc_xdp.c

 // SPDX-License-Identifier: GPL-2.0
 /* Copyright (c) 2020, Intel Corporation. */
 
+#include <net/xdp_sock_drv.h>
+
 #include "igc.h"
 #include "igc_xdp.h"
 
@@ -32,29 +34,112 @@ int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog,
 	return 0;
 }
 
-int igc_xdp_register_rxq_info(struct igc_ring *ring)
+static int igc_xdp_enable_pool(struct igc_adapter *adapter,
+			       struct xsk_buff_pool *pool, u16 queue_id)
 {
-	struct net_device *dev = ring->netdev;
+	struct net_device *ndev = adapter->netdev;
+	struct device *dev = &adapter->pdev->dev;
+	struct igc_ring *rx_ring, *tx_ring;
+	struct napi_struct *napi;
+	bool needs_reset;
+	u32 frame_size;
 	int err;
 
-	err = xdp_rxq_info_reg(&ring->xdp_rxq, dev, ring->queue_index, 0);
+	if (queue_id >= adapter->num_rx_queues ||
+	    queue_id >= adapter->num_tx_queues)
+		return -EINVAL;
+
+	frame_size = xsk_pool_get_rx_frame_size(pool);
+	if (frame_size < ETH_FRAME_LEN + VLAN_HLEN * 2) {
+		/* When XDP is enabled, the driver doesn't support frames that
+		 * span over multiple buffers. To avoid that, we check if xsk
+		 * frame size is big enough to fit the max ethernet frame size
+		 * + vlan double tagging.
+		 */
+		return -EOPNOTSUPP;
+	}
+
+	err = xsk_pool_dma_map(pool, dev, IGC_RX_DMA_ATTR);
 	if (err) {
-		netdev_err(dev, "Failed to register xdp rxq info\n");
+		netdev_err(ndev, "Failed to map xsk pool\n");
 		return err;
 	}
 
-	err = xdp_rxq_info_reg_mem_model(&ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
-					 NULL);
+	needs_reset = netif_running(adapter->netdev) && igc_xdp_is_enabled(adapter);
+
+	rx_ring = adapter->rx_ring[queue_id];
+	tx_ring = adapter->tx_ring[queue_id];
+	/* Rx and Tx rings share the same napi context. */
+	napi = &rx_ring->q_vector->napi;
+
+	if (needs_reset) {
+		igc_disable_rx_ring(rx_ring);
+		igc_disable_tx_ring(tx_ring);
+		napi_disable(napi);
+	}
+
+	set_bit(IGC_RING_FLAG_AF_XDP_ZC, &rx_ring->flags);
+	set_bit(IGC_RING_FLAG_AF_XDP_ZC, &tx_ring->flags);
+
+	if (needs_reset) {
+		napi_enable(napi);
+		igc_enable_rx_ring(rx_ring);
+		igc_enable_tx_ring(tx_ring);
+
+		err = igc_xsk_wakeup(ndev, queue_id, XDP_WAKEUP_RX);
 		if (err) {
-		netdev_err(dev, "Failed to register xdp rxq mem model\n");
-		xdp_rxq_info_unreg(&ring->xdp_rxq);
+			xsk_pool_dma_unmap(pool, IGC_RX_DMA_ATTR);
 			return err;
 		}
+	}
+
+	return 0;
+}
+
+static int igc_xdp_disable_pool(struct igc_adapter *adapter, u16 queue_id)
+{
+	struct igc_ring *rx_ring, *tx_ring;
+	struct xsk_buff_pool *pool;
+	struct napi_struct *napi;
+	bool needs_reset;
+
+	if (queue_id >= adapter->num_rx_queues ||
+	    queue_id >= adapter->num_tx_queues)
+		return -EINVAL;
+
+	pool = xsk_get_pool_from_qid(adapter->netdev, queue_id);
+	if (!pool)
+		return -EINVAL;
+
+	needs_reset = netif_running(adapter->netdev) && igc_xdp_is_enabled(adapter);
+
+	rx_ring = adapter->rx_ring[queue_id];
+	tx_ring = adapter->tx_ring[queue_id];
+	/* Rx and Tx rings share the same napi context. */
+	napi = &rx_ring->q_vector->napi;
+
+	if (needs_reset) {
+		igc_disable_rx_ring(rx_ring);
+		igc_disable_tx_ring(tx_ring);
+		napi_disable(napi);
+	}
+
+	xsk_pool_dma_unmap(pool, IGC_RX_DMA_ATTR);
+	clear_bit(IGC_RING_FLAG_AF_XDP_ZC, &rx_ring->flags);
+	clear_bit(IGC_RING_FLAG_AF_XDP_ZC, &tx_ring->flags);
+
+	if (needs_reset) {
+		napi_enable(napi);
+		igc_enable_rx_ring(rx_ring);
+		igc_enable_tx_ring(tx_ring);
+	}
 
 	return 0;
 }
 
-void igc_xdp_unregister_rxq_info(struct igc_ring *ring)
+int igc_xdp_setup_pool(struct igc_adapter *adapter, struct xsk_buff_pool *pool,
+		       u16 queue_id)
 {
-	xdp_rxq_info_unreg(&ring->xdp_rxq);
+	return pool ? igc_xdp_enable_pool(adapter, pool, queue_id) :
+		      igc_xdp_disable_pool(adapter, queue_id);
 }

drivers/net/ethernet/intel/igc/igc_xdp.h

@@ -6,8 +6,12 @@
 
 int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog,
 		     struct netlink_ext_ack *extack);
+int igc_xdp_setup_pool(struct igc_adapter *adapter, struct xsk_buff_pool *pool,
+		       u16 queue_id);
 
-int igc_xdp_register_rxq_info(struct igc_ring *ring);
-void igc_xdp_unregister_rxq_info(struct igc_ring *ring);
+static inline bool igc_xdp_is_enabled(struct igc_adapter *adapter)
+{
+	return !!adapter->xdp_prog;
+}
 
 #endif /* _IGC_XDP_H_ */