net: mana: Add XDP support

Add support of XDP for the MANA driver.

Supported XDP actions:
	XDP_PASS, XDP_TX, XDP_DROP, XDP_ABORTED

XDP actions not yet supported:
	XDP_REDIRECT
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/microsoft/mana/Makefile

@@ -3,4 +3,4 @@
 # Makefile for the Microsoft Azure Network Adapter driver
 
 obj-$(CONFIG_MICROSOFT_MANA) += mana.o
-mana-objs := gdma_main.o shm_channel.o hw_channel.o mana_en.o mana_ethtool.o
+mana-objs := gdma_main.o shm_channel.o hw_channel.o mana_en.o mana_ethtool.o mana_bpf.o

drivers/net/ethernet/microsoft/mana/mana.h

@@ -298,6 +298,9 @@ struct mana_rxq {
 
 	struct mana_stats stats;
 
+	struct bpf_prog __rcu *bpf_prog;
+	struct xdp_rxq_info xdp_rxq;
+
 	/* MUST BE THE LAST MEMBER:
 	 * Each receive buffer has an associated mana_recv_buf_oob.
 	 */
@@ -353,6 +356,8 @@ struct mana_port_context {
 	/* This points to an array of num_queues of RQ pointers. */
 	struct mana_rxq **rxqs;
 
+	struct bpf_prog *bpf_prog;
+
 	/* Create num_queues EQs, SQs, SQ-CQs, RQs and RQ-CQs, respectively. */
 	unsigned int max_queues;
 	unsigned int num_queues;
@@ -367,6 +372,7 @@ struct mana_port_context {
 	struct mana_ethtool_stats eth_stats;
 };
 
+int mana_start_xmit(struct sk_buff *skb, struct net_device *ndev);
 int mana_config_rss(struct mana_port_context *ac, enum TRI_STATE rx,
 		    bool update_hash, bool update_tab);
 
@@ -377,6 +383,13 @@ int mana_detach(struct net_device *ndev, bool from_close);
 int mana_probe(struct gdma_dev *gd, bool resuming);
 void mana_remove(struct gdma_dev *gd, bool suspending);
 
+void mana_xdp_tx(struct sk_buff *skb, struct net_device *ndev);
+u32 mana_run_xdp(struct net_device *ndev, struct mana_rxq *rxq,
+		 struct xdp_buff *xdp, void *buf_va, uint pkt_len);
+struct bpf_prog *mana_xdp_get(struct mana_port_context *apc);
+void mana_chn_setxdp(struct mana_port_context *apc, struct bpf_prog *prog);
+int mana_bpf(struct net_device *ndev, struct netdev_bpf *bpf);
+
 extern const struct ethtool_ops mana_ethtool_ops;
 
 struct mana_obj_spec {

drivers/net/ethernet/microsoft/mana/mana_bpf.c

+// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
+/* Copyright (c) 2021, Microsoft Corporation. */
+
+#include <linux/inetdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/mm.h>
+#include <linux/bpf.h>
+#include <linux/bpf_trace.h>
+#include <net/xdp.h>
+
+#include "mana.h"
+
+void mana_xdp_tx(struct sk_buff *skb, struct net_device *ndev)
+{
+	u16 txq_idx = skb_get_queue_mapping(skb);
+	struct netdev_queue *ndevtxq;
+	int rc;
+
+	__skb_push(skb, ETH_HLEN);
+
+	ndevtxq = netdev_get_tx_queue(ndev, txq_idx);
+	__netif_tx_lock(ndevtxq, smp_processor_id());
+
+	rc = mana_start_xmit(skb, ndev);
+
+	__netif_tx_unlock(ndevtxq);
+
+	if (dev_xmit_complete(rc))
+		return;
+
+	dev_kfree_skb_any(skb);
+	ndev->stats.tx_dropped++;
+}
+
+u32 mana_run_xdp(struct net_device *ndev, struct mana_rxq *rxq,
+		 struct xdp_buff *xdp, void *buf_va, uint pkt_len)
+{
+	struct bpf_prog *prog;
+	u32 act = XDP_PASS;
+
+	rcu_read_lock();
+	prog = rcu_dereference(rxq->bpf_prog);
+
+	if (!prog)
+		goto out;
+
+	xdp_init_buff(xdp, PAGE_SIZE, &rxq->xdp_rxq);
+	xdp_prepare_buff(xdp, buf_va, XDP_PACKET_HEADROOM, pkt_len, false);
+
+	act = bpf_prog_run_xdp(prog, xdp);
+
+	switch (act) {
+	case XDP_PASS:
+	case XDP_TX:
+	case XDP_DROP:
+		break;
+
+	case XDP_ABORTED:
+		trace_xdp_exception(ndev, prog, act);
+		break;
+
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	}
+
+out:
+	rcu_read_unlock();
+
+	return act;
+}
+
+static unsigned int mana_xdp_fraglen(unsigned int len)
+{
+	return SKB_DATA_ALIGN(len) +
+	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+}
+
+struct bpf_prog *mana_xdp_get(struct mana_port_context *apc)
+{
+	ASSERT_RTNL();
+
+	return apc->bpf_prog;
+}
+
+static struct bpf_prog *mana_chn_xdp_get(struct mana_port_context *apc)
+{
+	return rtnl_dereference(apc->rxqs[0]->bpf_prog);
+}
+
+/* Set xdp program on channels */
+void mana_chn_setxdp(struct mana_port_context *apc, struct bpf_prog *prog)
+{
+	struct bpf_prog *old_prog = mana_chn_xdp_get(apc);
+	unsigned int num_queues = apc->num_queues;
+	int i;
+
+	ASSERT_RTNL();
+
+	if (old_prog == prog)
+		return;
+
+	if (prog)
+		bpf_prog_add(prog, num_queues);
+
+	for (i = 0; i < num_queues; i++)
+		rcu_assign_pointer(apc->rxqs[i]->bpf_prog, prog);
+
+	if (old_prog)
+		for (i = 0; i < num_queues; i++)
+			bpf_prog_put(old_prog);
+}
+
+static int mana_xdp_set(struct net_device *ndev, struct bpf_prog *prog,
+			struct netlink_ext_ack *extack)
+{
+	struct mana_port_context *apc = netdev_priv(ndev);
+	struct bpf_prog *old_prog;
+	int buf_max;
+
+	old_prog = mana_xdp_get(apc);
+
+	if (!old_prog && !prog)
+		return 0;
+
+	buf_max = XDP_PACKET_HEADROOM + mana_xdp_fraglen(ndev->mtu + ETH_HLEN);
+	if (prog && buf_max > PAGE_SIZE) {
+		netdev_err(ndev, "XDP: mtu:%u too large, buf_max:%u\n",
+			   ndev->mtu, buf_max);
+		NL_SET_ERR_MSG_MOD(extack, "XDP: mtu too large");
+
+		return -EOPNOTSUPP;
+	}
+
+	/* One refcnt of the prog is hold by the caller already, so
+	 * don't increase refcnt for this one.
+	 */
+	apc->bpf_prog = prog;
+
+	if (old_prog)
+		bpf_prog_put(old_prog);
+
+	if (apc->port_is_up)
+		mana_chn_setxdp(apc, prog);
+
+	return 0;
+}
+
+int mana_bpf(struct net_device *ndev, struct netdev_bpf *bpf)
+{
+	struct netlink_ext_ack *extack = bpf->extack;
+	int ret;
+
+	switch (bpf->command) {
+	case XDP_SETUP_PROG:
+		return mana_xdp_set(ndev, bpf->prog, extack);
+
+	default:
+		return -EOPNOTSUPP;
+	}
+
+	return ret;
+}

drivers/net/ethernet/microsoft/mana/mana_en.c

@@ -125,7 +125,7 @@ static int mana_map_skb(struct sk_buff *skb, struct mana_port_context *apc,
 	return -ENOMEM;
 }
 
-static int mana_start_xmit(struct sk_buff *skb, struct net_device *ndev)
+int mana_start_xmit(struct sk_buff *skb, struct net_device *ndev)
 {
 	enum mana_tx_pkt_format pkt_fmt = MANA_SHORT_PKT_FMT;
 	struct mana_port_context *apc = netdev_priv(ndev);
@@ -378,6 +378,7 @@ static const struct net_device_ops mana_devops = {
 	.ndo_start_xmit		= mana_start_xmit,
 	.ndo_validate_addr	= eth_validate_addr,
 	.ndo_get_stats64	= mana_get_stats64,
+	.ndo_bpf		= mana_bpf,
 };
 
 static void mana_cleanup_port_context(struct mana_port_context *apc)
@@ -906,6 +907,25 @@ static void mana_post_pkt_rxq(struct mana_rxq *rxq)
 	WARN_ON_ONCE(recv_buf_oob->wqe_inf.wqe_size_in_bu != 1);
 }
 
+static struct sk_buff *mana_build_skb(void *buf_va, uint pkt_len,
+				      struct xdp_buff *xdp)
+{
+	struct sk_buff *skb = build_skb(buf_va, PAGE_SIZE);
+
+	if (!skb)
+		return NULL;
+
+	if (xdp->data_hard_start) {
+		skb_reserve(skb, xdp->data - xdp->data_hard_start);
+		skb_put(skb, xdp->data_end - xdp->data);
+	} else {
+		skb_reserve(skb, XDP_PACKET_HEADROOM);
+		skb_put(skb, pkt_len);
+	}
+
+	return skb;
+}
+
 static void mana_rx_skb(void *buf_va, struct mana_rxcomp_oob *cqe,
 			struct mana_rxq *rxq)
 {
@@ -914,8 +934,10 @@ static void mana_rx_skb(void *buf_va, struct mana_rxcomp_oob *cqe,
 	uint pkt_len = cqe->ppi[0].pkt_len;
 	u16 rxq_idx = rxq->rxq_idx;
 	struct napi_struct *napi;
+	struct xdp_buff xdp = {};
 	struct sk_buff *skb;
 	u32 hash_value;
+	u32 act;
 
 	rxq->rx_cq.work_done++;
 	napi = &rxq->rx_cq.napi;
@@ -925,15 +947,16 @@ static void mana_rx_skb(void *buf_va, struct mana_rxcomp_oob *cqe,
 		return;
 	}
 
-	skb = build_skb(buf_va, PAGE_SIZE);
+	act = mana_run_xdp(ndev, rxq, &xdp, buf_va, pkt_len);
 
-	if (!skb) {
-		free_page((unsigned long)buf_va);
-		++ndev->stats.rx_dropped;
-		return;
-	}
+	if (act != XDP_PASS && act != XDP_TX)
+		goto drop;
+
+	skb = mana_build_skb(buf_va, pkt_len, &xdp);
+
+	if (!skb)
+		goto drop;
 
-	skb_put(skb, pkt_len);
 	skb->dev = napi->dev;
 
 	skb->protocol = eth_type_trans(skb, ndev);
@@ -954,12 +977,24 @@ static void mana_rx_skb(void *buf_va, struct mana_rxcomp_oob *cqe,
 			skb_set_hash(skb, hash_value, PKT_HASH_TYPE_L3);
 	}
 
+	if (act == XDP_TX) {
+		skb_set_queue_mapping(skb, rxq_idx);
+		mana_xdp_tx(skb, ndev);
+		return;
+	}
+
 	napi_gro_receive(napi, skb);
 
 	u64_stats_update_begin(&rx_stats->syncp);
 	rx_stats->packets++;
 	rx_stats->bytes += pkt_len;
 	u64_stats_update_end(&rx_stats->syncp);
+	return;
+
+drop:
+	free_page((unsigned long)buf_va);
+	++ndev->stats.rx_dropped;
+	return;
 }
 
 static void mana_process_rx_cqe(struct mana_rxq *rxq, struct mana_cq *cq,
@@ -1016,7 +1051,7 @@ static void mana_process_rx_cqe(struct mana_rxq *rxq, struct mana_cq *cq,
 	new_page = alloc_page(GFP_ATOMIC);
 
 	if (new_page) {
-		da = dma_map_page(dev, new_page, 0, rxq->datasize,
+		da = dma_map_page(dev, new_page, XDP_PACKET_HEADROOM, rxq->datasize,
 				  DMA_FROM_DEVICE);
 
 		if (dma_mapping_error(dev, da)) {
@@ -1291,6 +1326,9 @@ static void mana_destroy_rxq(struct mana_port_context *apc,
 		napi_synchronize(napi);
 
 	napi_disable(napi);
+
+	xdp_rxq_info_unreg(&rxq->xdp_rxq);
+
 	netif_napi_del(napi);
 
 	mana_destroy_wq_obj(apc, GDMA_RQ, rxq->rxobj);
@@ -1342,7 +1380,8 @@ static int mana_alloc_rx_wqe(struct mana_port_context *apc,
 		if (!page)
 			return -ENOMEM;
 
-		da = dma_map_page(dev, page, 0, rxq->datasize, DMA_FROM_DEVICE);
+		da = dma_map_page(dev, page, XDP_PACKET_HEADROOM, rxq->datasize,
+				  DMA_FROM_DEVICE);
 
 		if (dma_mapping_error(dev, da)) {
 			__free_page(page);
@@ -1485,6 +1524,12 @@ static struct mana_rxq *mana_create_rxq(struct mana_port_context *apc,
 	gc->cq_table[cq->gdma_id] = cq->gdma_cq;
 
 	netif_napi_add(ndev, &cq->napi, mana_poll, 1);
+
+	WARN_ON(xdp_rxq_info_reg(&rxq->xdp_rxq, ndev, rxq_idx,
+				 cq->napi.napi_id));
+	WARN_ON(xdp_rxq_info_reg_mem_model(&rxq->xdp_rxq,
+					   MEM_TYPE_PAGE_SHARED, NULL));
+
 	napi_enable(&cq->napi);
 
 	mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
@@ -1650,6 +1695,8 @@ int mana_alloc_queues(struct net_device *ndev)
 	if (err)
 		goto destroy_vport;
 
+	mana_chn_setxdp(apc, mana_xdp_get(apc));
+
 	return 0;
 
 destroy_vport:
@@ -1698,6 +1745,8 @@ static int mana_dealloc_queues(struct net_device *ndev)
 	if (apc->port_is_up)
 		return -EINVAL;
 
+	mana_chn_setxdp(apc, NULL);
+
 	/* No packet can be transmitted now since apc->port_is_up is false.
 	 * There is still a tiny chance that mana_poll_tx_cq() can re-enable
 	 * a txq because it may not timely see apc->port_is_up being cleared