Merge branch 'nfp-ring-reconfig-and-xdp-support'

Jakub Kicinski says:

====================
ring reconfiguration and XDP support

This set adds support for ethtool channel API and XDP.

I kick off with ethtool get_channels() implementation.
set_channels() needs some preparations to get right.  I follow
the prepare/commit paradigm and allocate all resources before
stopping the device.  It has already been done for ndo_change_mtu
and ethtool set_ringparam(), it makes sense now to consolidate all
the required logic in one place.

XDP support requires splitting TX rings into two classes -
for the stack and for XDP.  The ring structures are identical.
The differences are in how they are connected to IRQ vector
structs and how the completion/cleanup works.  When XDP is enabled
I switch from the frag allocator to page-per-packet and map buffers
BIDIRECTIONALly.

Last but not least XDP offload is added (the patch just takes
care of the small formal differences between cls_bpf and XDP).

There is a tiny & trivial DebugFS patch in the mix, I hope it can
be taken via net-next provided we have the right Acks.

Resending with improved commit message and CCing more people on patch 10.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/netronome/nfp/nfp_bpf.h

@@ -62,6 +62,7 @@ enum nfp_bpf_action_type {
 	NN_ACT_TC_DROP,
 	NN_ACT_TC_REDIR,
 	NN_ACT_DIRECT,
+	NN_ACT_XDP,
 };
 
 /* Software register representation, hardware encoding in asm.h */

drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c

@@ -1126,7 +1126,7 @@ static int data_ind_ld4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 				     meta->insn.src_reg * 2, true, 4);
 }
 
-static int mem_ldx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+static int mem_ldx4_skb(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 {
 	if (meta->insn.off == offsetof(struct sk_buff, len))
 		emit_alu(nfp_prog, reg_both(meta->insn.dst_reg * 2),
@@ -1134,12 +1134,42 @@ static int mem_ldx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 	else
 		return -ENOTSUPP;
 
-	wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), 0);
+	return 0;
+}
+
+static int mem_ldx4_xdp(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	u32 dst = reg_both(meta->insn.dst_reg * 2);
+
+	if (meta->insn.off != offsetof(struct xdp_md, data) &&
+	    meta->insn.off != offsetof(struct xdp_md, data_end))
+		return -ENOTSUPP;
+
+	emit_alu(nfp_prog, dst, reg_none(), ALU_OP_NONE, NFP_BPF_ABI_PKT);
+
+	if (meta->insn.off == offsetof(struct xdp_md, data))
+		return 0;
+
+	emit_alu(nfp_prog, dst,	dst, ALU_OP_ADD, NFP_BPF_ABI_LEN);
 
 	return 0;
 }
 
-static int mem_stx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+static int mem_ldx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	int ret;
+
+	if (nfp_prog->act == NN_ACT_XDP)
+		ret = mem_ldx4_xdp(nfp_prog, meta);
+	else
+		ret = mem_ldx4_skb(nfp_prog, meta);
+
+	wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), 0);
+
+	return ret;
+}
+
+static int mem_stx4_skb(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 {
 	if (meta->insn.off == offsetof(struct sk_buff, mark))
 		return wrp_set_mark(nfp_prog, meta->insn.src_reg * 2);
@@ -1147,6 +1177,18 @@ static int mem_stx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 	return -ENOTSUPP;
 }
 
+static int mem_stx4_xdp(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return -ENOTSUPP;
+}
+
+static int mem_stx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	if (nfp_prog->act == NN_ACT_XDP)
+		return mem_stx4_xdp(nfp_prog, meta);
+	return mem_stx4_skb(nfp_prog, meta);
+}
+
 static int jump(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 {
 	if (meta->insn.off < 0) /* TODO */
@@ -1530,6 +1572,47 @@ static void nfp_outro_tc_da(struct nfp_prog *nfp_prog)
 	emit_ld_field(nfp_prog, reg_a(0), 0xc, reg_b(2), SHF_SC_L_SHF, 16);
 }
 
+static void nfp_outro_xdp(struct nfp_prog *nfp_prog)
+{
+	/* XDP return codes:
+	 *   0 aborted  0x82 -> drop,  count as stat3
+	 *   1    drop  0x22 -> drop,  count as stat1
+	 *   2    pass  0x11 -> pass,  count as stat0
+	 *   3      tx  0x44 -> redir, count as stat2
+	 *   * unknown  0x82 -> drop,  count as stat3
+	 */
+	/* Target for aborts */
+	nfp_prog->tgt_abort = nfp_prog_current_offset(nfp_prog);
+
+	emit_br_def(nfp_prog, nfp_prog->tgt_done, 2);
+
+	emit_alu(nfp_prog, reg_a(0),
+		 reg_none(), ALU_OP_NONE, NFP_BPF_ABI_FLAGS);
+	emit_ld_field(nfp_prog, reg_a(0), 0xc, reg_imm(0x82), SHF_SC_L_SHF, 16);
+
+	/* Target for normal exits */
+	nfp_prog->tgt_out = nfp_prog_current_offset(nfp_prog);
+
+	/* if R0 > 3 jump to abort */
+	emit_alu(nfp_prog, reg_none(), reg_imm(3), ALU_OP_SUB, reg_b(0));
+	emit_br(nfp_prog, BR_BLO, nfp_prog->tgt_abort, 0);
+
+	wrp_immed(nfp_prog, reg_b(2), 0x44112282);
+
+	emit_shf(nfp_prog, reg_a(1),
+		 reg_none(), SHF_OP_NONE, reg_b(0), SHF_SC_L_SHF, 3);
+
+	emit_alu(nfp_prog, reg_none(), reg_a(1), ALU_OP_OR, reg_imm(0));
+	emit_shf(nfp_prog, reg_b(2),
+		 reg_imm(0xff), SHF_OP_AND, reg_b(2), SHF_SC_R_SHF, 0);
+
+	emit_br_def(nfp_prog, nfp_prog->tgt_done, 2);
+
+	emit_alu(nfp_prog, reg_a(0),
+		 reg_none(), ALU_OP_NONE, NFP_BPF_ABI_FLAGS);
+	emit_ld_field(nfp_prog, reg_a(0), 0xc, reg_b(2), SHF_SC_L_SHF, 16);
+}
+
 static void nfp_outro(struct nfp_prog *nfp_prog)
 {
 	switch (nfp_prog->act) {
@@ -1540,6 +1623,9 @@ static void nfp_outro(struct nfp_prog *nfp_prog)
 	case NN_ACT_TC_REDIR:
 		nfp_outro_tc_legacy(nfp_prog);
 		break;
+	case NN_ACT_XDP:
+		nfp_outro_xdp(nfp_prog);
+		break;
 	}
 }
 

drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c

@@ -80,6 +80,9 @@ nfp_bpf_check_exit(struct nfp_prog *nfp_prog,
 {
 	const struct bpf_reg_state *reg0 = &env->cur_state.regs[0];
 
+	if (nfp_prog->act == NN_ACT_XDP)
+		return 0;
+
 	if (reg0->type != CONST_IMM) {
 		pr_info("unsupported exit state: %d, imm: %llx\n",
 			reg0->type, reg0->imm);

drivers/net/ethernet/netronome/nfp/nfp_net.h

@@ -171,7 +171,10 @@ struct nfp_net_tx_desc {
  *		on the head's buffer). Equal to skb->len for non-TSO packets.
  */
 struct nfp_net_tx_buf {
+	union {
 		struct sk_buff *skb;
+		void *frag;
+	};
 	dma_addr_t dma_addr;
 	short int fidx;
 	u16 pkt_cnt;
@@ -341,6 +344,7 @@ struct nfp_net_rx_ring {
  * @napi:           NAPI structure for this ring vec
  * @tx_ring:        Pointer to TX ring
  * @rx_ring:        Pointer to RX ring
+ * @xdp_ring:	    Pointer to an extra TX ring for XDP
  * @irq_idx:        Index into MSI-X table
  * @rx_sync:	    Seqlock for atomic updates of RX stats
  * @rx_pkts:        Number of received packets
@@ -384,6 +388,8 @@ struct nfp_net_r_vector {
 	u64 hw_csum_rx_inner_ok;
 	u64 hw_csum_rx_error;
 
+	struct nfp_net_tx_ring *xdp_ring;
+
 	struct u64_stats_sync tx_sync;
 	u64 tx_pkts;
 	u64 tx_bytes;
@@ -429,9 +435,11 @@ struct nfp_stat_pair {
  * @is_vf:              Is the driver attached to a VF?
  * @fw_loaded:          Is the firmware loaded?
  * @bpf_offload_skip_sw:  Offloaded BPF program will not be rerun by cls_bpf
+ * @bpf_offload_xdp:	Offloaded BPF program is XDP
  * @ctrl:               Local copy of the control register/word.
  * @fl_bufsz:           Currently configured size of the freelist buffers
  * @rx_offset:		Offset in the RX buffers where packet data starts
+ * @xdp_prog:		Installed XDP program
  * @cpp:                Pointer to the CPP handle
  * @nfp_dev_cpp:        Pointer to the NFP Device handle
  * @ctrl_area:          Pointer to the CPP area for the control BAR
@@ -451,6 +459,7 @@ struct nfp_stat_pair {
  * @max_tx_rings:       Maximum number of TX rings supported by the Firmware
  * @max_rx_rings:       Maximum number of RX rings supported by the Firmware
  * @num_tx_rings:       Currently configured number of TX rings
+ * @num_stack_tx_rings:	Number of TX rings used by the stack (not XDP)
  * @num_rx_rings:       Currently configured number of RX rings
  * @txd_cnt:            Size of the TX ring in number of descriptors
  * @rxd_cnt:            Size of the RX ring in number of descriptors
@@ -494,12 +503,15 @@ struct nfp_net {
 	unsigned is_vf:1;
 	unsigned fw_loaded:1;
 	unsigned bpf_offload_skip_sw:1;
+	unsigned bpf_offload_xdp:1;
 
 	u32 ctrl;
 	u32 fl_bufsz;
 
 	u32 rx_offset;
 
+	struct bpf_prog *xdp_prog;
+
 	struct nfp_net_tx_ring *tx_rings;
 	struct nfp_net_rx_ring *rx_rings;
 
@@ -532,6 +544,7 @@ struct nfp_net {
 	unsigned int max_rx_rings;
 
 	unsigned int num_tx_rings;
+	unsigned int num_stack_tx_rings;
 	unsigned int num_rx_rings;
 
 	int stride_tx;
@@ -583,6 +596,13 @@ struct nfp_net {
 	struct dentry *debugfs_dir;
 };
 
+struct nfp_net_ring_set {
+	unsigned int n_rings;
+	unsigned int mtu;
+	unsigned int dcnt;
+	void *rings;
+};
+
 /* Functions to read/write from/to a BAR
  * Performs any endian conversion necessary.
  */
@@ -771,7 +791,9 @@ void nfp_net_rss_write_key(struct nfp_net *nn);
 void nfp_net_coalesce_write_cfg(struct nfp_net *nn);
 int nfp_net_irqs_alloc(struct nfp_net *nn);
 void nfp_net_irqs_disable(struct nfp_net *nn);
-int nfp_net_set_ring_size(struct nfp_net *nn, u32 rxd_cnt, u32 txd_cnt);
+int
+nfp_net_ring_reconfig(struct nfp_net *nn, struct bpf_prog **xdp_prog,
+		      struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx);
 
 #ifdef CONFIG_NFP_NET_DEBUG
 void nfp_net_debugfs_create(void);
@@ -797,8 +819,6 @@ static inline void nfp_net_debugfs_adapter_del(struct nfp_net *nn)
 #endif /* CONFIG_NFP_NET_DEBUG */
 
 void nfp_net_filter_stats_timer(unsigned long data);
-int
-nfp_net_bpf_offload(struct nfp_net *nn, u32 handle, __be16 proto,
-		    struct tc_cls_bpf_offload *cls_bpf);
+int nfp_net_bpf_offload(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf);
 
 #endif /* _NFP_NET_H_ */

drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c

@@ -114,6 +114,16 @@ static const struct file_operations nfp_rx_q_fops = {
 	.llseek = seq_lseek
 };
 
+static int nfp_net_debugfs_tx_q_open(struct inode *inode, struct file *f);
+
+static const struct file_operations nfp_tx_q_fops = {
+	.owner = THIS_MODULE,
+	.open = nfp_net_debugfs_tx_q_open,
+	.release = single_release,
+	.read = seq_read,
+	.llseek = seq_lseek
+};
+
 static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data)
 {
 	struct nfp_net_r_vector *r_vec = file->private;
@@ -126,10 +136,13 @@ static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data)
 
 	rtnl_lock();
 
-	if (!r_vec->nfp_net || !r_vec->tx_ring)
+	if (debugfs_real_fops(file->file) == &nfp_tx_q_fops)
+		tx_ring = r_vec->tx_ring;
+	else
+		tx_ring = r_vec->xdp_ring;
+	if (!r_vec->nfp_net || !tx_ring)
 		goto out;
 	nn = r_vec->nfp_net;
-	tx_ring = r_vec->tx_ring;
 	if (!netif_running(nn->netdev))
 		goto out;
 
@@ -148,9 +161,14 @@ static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data)
 			   txd->vals[2], txd->vals[3]);
 
 		skb = READ_ONCE(tx_ring->txbufs[i].skb);
-		if (skb)
+		if (skb) {
+			if (tx_ring == r_vec->tx_ring)
 				seq_printf(file, " skb->head=%p skb->data=%p",
 					   skb->head, skb->data);
+			else
+				seq_printf(file, " frag=%p", skb);
+		}
+
 		if (tx_ring->txbufs[i].dma_addr)
 			seq_printf(file, " dma_addr=%pad",
 				   &tx_ring->txbufs[i].dma_addr);
@@ -176,7 +194,7 @@ static int nfp_net_debugfs_tx_q_open(struct inode *inode, struct file *f)
 	return single_open(f, nfp_net_debugfs_tx_q_read, inode->i_private);
 }
 
-static const struct file_operations nfp_tx_q_fops = {
+static const struct file_operations nfp_xdp_q_fops = {
 	.owner = THIS_MODULE,
 	.open = nfp_net_debugfs_tx_q_open,
 	.release = single_release,
@@ -186,7 +204,7 @@ static const struct file_operations nfp_tx_q_fops = {
 
 void nfp_net_debugfs_adapter_add(struct nfp_net *nn)
 {
-	struct dentry *queues, *tx, *rx;
+	struct dentry *queues, *tx, *rx, *xdp;
 	char int_name[16];
 	int i;
 
@@ -204,16 +222,19 @@ void nfp_net_debugfs_adapter_add(struct nfp_net *nn)
 
 	rx = debugfs_create_dir("rx", queues);
 	tx = debugfs_create_dir("tx", queues);
-	if (IS_ERR_OR_NULL(rx) || IS_ERR_OR_NULL(tx))
+	xdp = debugfs_create_dir("xdp", queues);
+	if (IS_ERR_OR_NULL(rx) || IS_ERR_OR_NULL(tx) || IS_ERR_OR_NULL(xdp))
 		return;
 
-	for (i = 0; i < nn->num_rx_rings; i++) {
+	for (i = 0; i < min(nn->max_rx_rings, nn->max_r_vecs); i++) {
 		sprintf(int_name, "%d", i);
 		debugfs_create_file(int_name, S_IRUSR, rx,
 				    &nn->r_vecs[i], &nfp_rx_q_fops);
+		debugfs_create_file(int_name, S_IRUSR, xdp,
+				    &nn->r_vecs[i], &nfp_xdp_q_fops);
 	}
 
-	for (i = 0; i < nn->num_tx_rings; i++) {
+	for (i = 0; i < min(nn->max_tx_rings, nn->max_r_vecs); i++) {
 		sprintf(int_name, "%d", i);
 		debugfs_create_file(int_name, S_IRUSR, tx,
 				    &nn->r_vecs[i], &nfp_tx_q_fops);

drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c

@@ -158,6 +158,28 @@ static void nfp_net_get_ringparam(struct net_device *netdev,
 	ring->tx_pending = nn->txd_cnt;
 }
 
+static int nfp_net_set_ring_size(struct nfp_net *nn, u32 rxd_cnt, u32 txd_cnt)
+{
+	struct nfp_net_ring_set *reconfig_rx = NULL, *reconfig_tx = NULL;
+	struct nfp_net_ring_set rx = {
+		.n_rings = nn->num_rx_rings,
+		.mtu = nn->netdev->mtu,
+		.dcnt = rxd_cnt,
+	};
+	struct nfp_net_ring_set tx = {
+		.n_rings = nn->num_tx_rings,
+		.dcnt = txd_cnt,
+	};
+
+	if (nn->rxd_cnt != rxd_cnt)
+		reconfig_rx = ℞
+	if (nn->txd_cnt != txd_cnt)
+		reconfig_tx = &tx;
+
+	return nfp_net_ring_reconfig(nn, &nn->xdp_prog,
+				     reconfig_rx, reconfig_tx);
+}
+
 static int nfp_net_set_ringparam(struct net_device *netdev,
 				 struct ethtool_ringparam *ring)
 {
@@ -614,6 +636,76 @@ static int nfp_net_set_coalesce(struct net_device *netdev,
 	return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD);
 }
 
+static void nfp_net_get_channels(struct net_device *netdev,
+				 struct ethtool_channels *channel)
+{
+	struct nfp_net *nn = netdev_priv(netdev);
+	unsigned int num_tx_rings;
+
+	num_tx_rings = nn->num_tx_rings;
+	if (nn->xdp_prog)
+		num_tx_rings -= nn->num_rx_rings;
+
+	channel->max_rx = min(nn->max_rx_rings, nn->max_r_vecs);
+	channel->max_tx = min(nn->max_tx_rings, nn->max_r_vecs);
+	channel->max_combined = min(channel->max_rx, channel->max_tx);
+	channel->max_other = NFP_NET_NON_Q_VECTORS;
+	channel->combined_count = min(nn->num_rx_rings, num_tx_rings);
+	channel->rx_count = nn->num_rx_rings - channel->combined_count;
+	channel->tx_count = num_tx_rings - channel->combined_count;
+	channel->other_count = NFP_NET_NON_Q_VECTORS;
+}
+
+static int nfp_net_set_num_rings(struct nfp_net *nn, unsigned int total_rx,
+				 unsigned int total_tx)
+{
+	struct nfp_net_ring_set *reconfig_rx = NULL, *reconfig_tx = NULL;
+	struct nfp_net_ring_set rx = {
+		.n_rings = total_rx,
+		.mtu = nn->netdev->mtu,
+		.dcnt = nn->rxd_cnt,
+	};
+	struct nfp_net_ring_set tx = {
+		.n_rings = total_tx,
+		.dcnt = nn->txd_cnt,
+	};
+
+	if (nn->num_rx_rings != total_rx)
+		reconfig_rx = ℞
+	if (nn->num_stack_tx_rings != total_tx ||
+	    (nn->xdp_prog && reconfig_rx))
+		reconfig_tx = &tx;
+
+	/* nfp_net_check_config() will catch tx.n_rings > nn->max_tx_rings */
+	if (nn->xdp_prog)
+		tx.n_rings += total_rx;
+
+	return nfp_net_ring_reconfig(nn, &nn->xdp_prog,
+				     reconfig_rx, reconfig_tx);
+}
+
+static int nfp_net_set_channels(struct net_device *netdev,
+				struct ethtool_channels *channel)
+{
+	struct nfp_net *nn = netdev_priv(netdev);
+	unsigned int total_rx, total_tx;
+
+	/* Reject unsupported */
+	if (!channel->combined_count ||
+	    channel->other_count != NFP_NET_NON_Q_VECTORS ||
+	    (channel->rx_count && channel->tx_count))
+		return -EINVAL;
+
+	total_rx = channel->combined_count + channel->rx_count;
+	total_tx = channel->combined_count + channel->tx_count;
+
+	if (total_rx > min(nn->max_rx_rings, nn->max_r_vecs) ||
+	    total_tx > min(nn->max_tx_rings, nn->max_r_vecs))
+		return -EINVAL;
+
+	return nfp_net_set_num_rings(nn, total_rx, total_tx);
+}
+
 static const struct ethtool_ops nfp_net_ethtool_ops = {
 	.get_drvinfo		= nfp_net_get_drvinfo,
 	.get_link		= ethtool_op_get_link,
@@ -632,6 +724,8 @@ static const struct ethtool_ops nfp_net_ethtool_ops = {
 	.get_regs		= nfp_net_get_regs,
 	.get_coalesce           = nfp_net_get_coalesce,
 	.set_coalesce           = nfp_net_set_coalesce,
+	.get_channels		= nfp_net_get_channels,
+	.set_channels		= nfp_net_set_channels,
 };
 
 void nfp_net_set_ethtool_ops(struct net_device *netdev)

drivers/net/ethernet/netronome/nfp/nfp_net_offload.c

@@ -111,6 +111,9 @@ nfp_net_bpf_get_act(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf)
 	const struct tc_action *a;
 	LIST_HEAD(actions);
 
+	if (!cls_bpf->exts)
+		return NN_ACT_XDP;
+
 	/* TC direct action */
 	if (cls_bpf->exts_integrated) {
 		if (tc_no_actions(cls_bpf->exts))
@@ -233,9 +236,7 @@ static int nfp_net_bpf_stop(struct nfp_net *nn)
 	return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
 }
 
-int
-nfp_net_bpf_offload(struct nfp_net *nn, u32 handle, __be16 proto,
-		    struct tc_cls_bpf_offload *cls_bpf)
+int nfp_net_bpf_offload(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf)
 {
 	struct nfp_bpf_result res;
 	dma_addr_t dma_addr;

include/linux/debugfs.h

@@ -52,7 +52,8 @@ extern struct srcu_struct debugfs_srcu;
  * Must only be called under the protection established by
  * debugfs_use_file_start().
  */
-static inline const struct file_operations *debugfs_real_fops(struct file *filp)
+static inline const struct file_operations *
+debugfs_real_fops(const struct file *filp)
 	__must_hold(&debugfs_srcu)
 {
 	/*