Merge branch 'qed-XDP-header-adjust'

Yuval Mintz says:

====================
qede: support XDP head adjustments

Daniel has brought to my attention the fact that qede is the only driver
that currently supports XDP but still fails any program where
xdp_adjust_head is set on the bpf_prog. This series is meant to remedy
this and align qede with the rest of the driver, making it possible to
remove said field.

Patch #1 contains a minor cache-saving optimization for latter patches.

Patches #2 & #3 address existing issues with the qede implementation
[#2 should have been a part of this as it addresses something that's
affected by the additional headroom; #3 is simply here for the ride].

Patches #4 & #5 add the necessary logic in driver for ingress headroom,
the first adding the infrastrucutre needed for supporting the headroon
[as currently qede doesn't support such], and the second removing the
existing XDP limitation.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/qlogic/qede/qede.h

@@ -313,21 +313,24 @@ struct qede_rx_queue {
 	u8 data_direction;
 	u8 rxq_id;
 
+	/* Used once per each NAPI run */
+	u16 num_rx_buffers;
+
+	u16 rx_headroom;
+
 	u32 rx_buf_size;
 	u32 rx_buf_seg_size;
 
-	u64 rcv_pkts;
-
 	struct sw_rx_data *sw_rx_ring;
 	struct qed_chain rx_bd_ring;
 	struct qed_chain rx_comp_ring ____cacheline_aligned;
 
-	/* Used once per each NAPI run */
-	u16 num_rx_buffers;
-
 	/* GRO */
 	struct qede_agg_info tpa_info[ETH_TPA_MAX_AGGS_NUM];
 
+	/* Used once per each NAPI run */
+	u64 rcv_pkts;
+
 	u64 rx_hw_errors;
 	u64 rx_alloc_errors;
 	u64 rx_ip_frags;
@@ -349,6 +352,11 @@ struct sw_tx_bd {
 #define QEDE_TSO_SPLIT_BD		BIT(0)
 };
 
+struct sw_tx_xdp {
+	struct page *page;
+	dma_addr_t mapping;
+};
+
 struct qede_tx_queue {
 	u8 is_xdp;
 	bool is_legacy;
@@ -372,11 +380,11 @@ struct qede_tx_queue {
 #define QEDE_TXQ_IDX_TO_XDP(edev, idx)	((idx) + QEDE_MAX_TSS_CNT(edev))
 
 	/* Regular Tx requires skb + metadata for release purpose,
-	 * while XDP requires only the pages themselves.
+	 * while XDP requires the pages and the mapped address.
 	 */
 	union {
 		struct sw_tx_bd *skbs;
-		struct page **pages;
+		struct sw_tx_xdp *xdp;
 	} sw_tx_ring;
 
 	struct qed_chain tx_pbl;

drivers/net/ethernet/qlogic/qede/qede_filter.c

@@ -520,11 +520,6 @@ static int qede_xdp_set(struct qede_dev *edev, struct bpf_prog *prog)
 {
 	struct qede_reload_args args;
 
-	if (prog && prog->xdp_adjust_head) {
-		DP_ERR(edev, "Does not support bpf_xdp_adjust_head()\n");
-		return -EOPNOTSUPP;
-	}
-
 	/* If we're called, there was already a bpf reference increment */
 	args.func = &qede_xdp_reload_func;
 	args.u.new_prog = prog;
@@ -537,6 +532,11 @@ int qede_xdp(struct net_device *dev, struct netdev_xdp *xdp)
 {
 	struct qede_dev *edev = netdev_priv(dev);
 
+	if (IS_VF(edev)) {
+		DP_NOTICE(edev, "VFs don't support XDP\n");
+		return -EOPNOTSUPP;
+	}
+
 	switch (xdp->command) {
 	case XDP_SETUP_PROG:
 		return qede_xdp_set(edev, xdp->prog);

drivers/net/ethernet/qlogic/qede/qede_fp.c

@@ -87,7 +87,8 @@ int qede_alloc_rx_buffer(struct qede_rx_queue *rxq, bool allow_lazy)
 	rx_bd = (struct eth_rx_bd *)qed_chain_produce(&rxq->rx_bd_ring);
 	WARN_ON(!rx_bd);
 	rx_bd->addr.hi = cpu_to_le32(upper_32_bits(mapping));
-	rx_bd->addr.lo = cpu_to_le32(lower_32_bits(mapping));
+	rx_bd->addr.lo = cpu_to_le32(lower_32_bits(mapping) +
+				     rxq->rx_headroom);
 
 	rxq->sw_rx_prod++;
 	rxq->filled_buffers++;
@@ -360,7 +361,8 @@ static int qede_xdp_xmit(struct qede_dev *edev, struct qede_fastpath *fp,
 				   metadata->mapping + padding,
 				   length, PCI_DMA_TODEVICE);
 
-	txq->sw_tx_ring.pages[idx] = metadata->data;
+	txq->sw_tx_ring.xdp[idx].page = metadata->data;
+	txq->sw_tx_ring.xdp[idx].mapping = metadata->mapping;
 	txq->sw_tx_prod++;
 
 	/* Mark the fastpath for future XDP doorbell */
@@ -384,19 +386,19 @@ int qede_txq_has_work(struct qede_tx_queue *txq)
 
 static void qede_xdp_tx_int(struct qede_dev *edev, struct qede_tx_queue *txq)
 {
-	struct eth_tx_1st_bd *bd;
-	u16 hw_bd_cons;
+	u16 hw_bd_cons, idx;
 
 	hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
 	barrier();
 
 	while (hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl)) {
-		bd = (struct eth_tx_1st_bd *)qed_chain_consume(&txq->tx_pbl);
+		qed_chain_consume(&txq->tx_pbl);
+		idx = txq->sw_tx_cons & NUM_TX_BDS_MAX;
 
-		dma_unmap_single(&edev->pdev->dev, BD_UNMAP_ADDR(bd),
-				 PAGE_SIZE, DMA_BIDIRECTIONAL);
-		__free_page(txq->sw_tx_ring.pages[txq->sw_tx_cons &
-						  NUM_TX_BDS_MAX]);
+		dma_unmap_page(&edev->pdev->dev,
+			       txq->sw_tx_ring.xdp[idx].mapping,
+			       PAGE_SIZE, DMA_BIDIRECTIONAL);
+		__free_page(txq->sw_tx_ring.xdp[idx].page);
 
 		txq->sw_tx_cons++;
 		txq->xmit_pkts++;
@@ -508,7 +510,8 @@ static inline void qede_reuse_page(struct qede_rx_queue *rxq,
 	new_mapping = curr_prod->mapping + curr_prod->page_offset;
 
 	rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(new_mapping));
-	rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(new_mapping));
+	rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(new_mapping) +
+					  rxq->rx_headroom);
 
 	rxq->sw_rx_prod++;
 	curr_cons->data = NULL;
@@ -624,7 +627,6 @@ static inline void qede_skb_receive(struct qede_dev *edev,
 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
 
 	napi_gro_receive(&fp->napi, skb);
-	rxq->rcv_pkts++;
 }
 
 static void qede_set_gro_params(struct qede_dev *edev,
@@ -884,9 +886,9 @@ static inline void qede_tpa_cont(struct qede_dev *edev,
 		       "Strange - TPA cont with more than a single len_list entry\n");
 }
 
-static void qede_tpa_end(struct qede_dev *edev,
-			 struct qede_fastpath *fp,
-			 struct eth_fast_path_rx_tpa_end_cqe *cqe)
+static int qede_tpa_end(struct qede_dev *edev,
+			struct qede_fastpath *fp,
+			struct eth_fast_path_rx_tpa_end_cqe *cqe)
 {
 	struct qede_rx_queue *rxq = fp->rxq;
 	struct qede_agg_info *tpa_info;
@@ -934,11 +936,12 @@ static void qede_tpa_end(struct qede_dev *edev,
 
 	tpa_info->state = QEDE_AGG_STATE_NONE;
 
-	return;
+	return 1;
 err:
 	tpa_info->state = QEDE_AGG_STATE_NONE;
 	dev_kfree_skb_any(tpa_info->skb);
 	tpa_info->skb = NULL;
+	return 0;
 }
 
 static u8 qede_check_notunn_csum(u16 flag)
@@ -990,14 +993,15 @@ static bool qede_rx_xdp(struct qede_dev *edev,
 			struct qede_rx_queue *rxq,
 			struct bpf_prog *prog,
 			struct sw_rx_data *bd,
-			struct eth_fast_path_rx_reg_cqe *cqe)
+			struct eth_fast_path_rx_reg_cqe *cqe,
+			u16 *data_offset, u16 *len)
 {
-	u16 len = le16_to_cpu(cqe->len_on_first_bd);
 	struct xdp_buff xdp;
 	enum xdp_action act;
 
-	xdp.data = page_address(bd->data) + cqe->placement_offset;
-	xdp.data_end = xdp.data + len;
+	xdp.data_hard_start = page_address(bd->data);
+	xdp.data = xdp.data_hard_start + *data_offset;
+	xdp.data_end = xdp.data + *len;
 
 	/* Queues always have a full reset currently, so for the time
 	 * being until there's atomic program replace just mark read
@@ -1007,6 +1011,10 @@ static bool qede_rx_xdp(struct qede_dev *edev,
 	act = bpf_prog_run_xdp(prog, &xdp);
 	rcu_read_unlock();
 
+	/* Recalculate, as XDP might have changed the headers */
+	*data_offset = xdp.data - xdp.data_hard_start;
+	*len = xdp.data_end - xdp.data;
+
 	if (act == XDP_PASS)
 		return true;
 
@@ -1025,7 +1033,7 @@ static bool qede_rx_xdp(struct qede_dev *edev,
 		/* Now if there's a transmission problem, we'd still have to
 		 * throw current buffer, as replacement was already allocated.
 		 */
-		if (qede_xdp_xmit(edev, fp, bd, cqe->placement_offset, len)) {
+		if (qede_xdp_xmit(edev, fp, bd, *data_offset, *len)) {
 			dma_unmap_page(rxq->dev, bd->mapping,
 				       PAGE_SIZE, DMA_BIDIRECTIONAL);
 			__free_page(bd->data);
@@ -1052,7 +1060,7 @@ static struct sk_buff *qede_rx_allocate_skb(struct qede_dev *edev,
 					    struct sw_rx_data *bd, u16 len,
 					    u16 pad)
 {
-	unsigned int offset = bd->page_offset;
+	unsigned int offset = bd->page_offset + pad;
 	struct skb_frag_struct *frag;
 	struct page *page = bd->data;
 	unsigned int pull_len;
@@ -1069,7 +1077,7 @@ static struct sk_buff *qede_rx_allocate_skb(struct qede_dev *edev,
 	 */
 	if (len + pad <= edev->rx_copybreak) {
 		memcpy(skb_put(skb, len),
-		       page_address(page) + pad + offset, len);
+		       page_address(page) + offset, len);
 		qede_reuse_page(rxq, bd);
 		goto out;
 	}
@@ -1077,7 +1085,7 @@ static struct sk_buff *qede_rx_allocate_skb(struct qede_dev *edev,
 	frag = &skb_shinfo(skb)->frags[0];
 
 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
-			page, pad + offset, len, rxq->rx_buf_seg_size);
+			page, offset, len, rxq->rx_buf_seg_size);
 
 	va = skb_frag_address(frag);
 	pull_len = eth_get_headlen(va, QEDE_RX_HDR_SIZE);
@@ -1178,8 +1186,7 @@ static int qede_rx_process_tpa_cqe(struct qede_dev *edev,
 		qede_tpa_cont(edev, rxq, &cqe->fast_path_tpa_cont);
 		return 0;
 	case ETH_RX_CQE_TYPE_TPA_END:
-		qede_tpa_end(edev, fp, &cqe->fast_path_tpa_end);
-		return 1;
+		return qede_tpa_end(edev, fp, &cqe->fast_path_tpa_end);
 	default:
 		return 0;
 	}
@@ -1224,12 +1231,13 @@ static int qede_rx_process_cqe(struct qede_dev *edev,
 
 	fp_cqe = &cqe->fast_path_regular;
 	len = le16_to_cpu(fp_cqe->len_on_first_bd);
-	pad = fp_cqe->placement_offset;
+	pad = fp_cqe->placement_offset + rxq->rx_headroom;
 
 	/* Run eBPF program if one is attached */
 	if (xdp_prog)
-		if (!qede_rx_xdp(edev, fp, rxq, xdp_prog, bd, fp_cqe))
-			return 1;
+		if (!qede_rx_xdp(edev, fp, rxq, xdp_prog, bd, fp_cqe,
+				 &pad, &len))
+			return 0;
 
 	/* If this is an error packet then drop it */
 	flags = cqe->fast_path_regular.pars_flags.flags;
@@ -1290,8 +1298,8 @@ static int qede_rx_int(struct qede_fastpath *fp, int budget)
 {
 	struct qede_rx_queue *rxq = fp->rxq;
 	struct qede_dev *edev = fp->edev;
+	int work_done = 0, rcv_pkts = 0;
 	u16 hw_comp_cons, sw_comp_cons;
-	int work_done = 0;
 
 	hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
 	sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
@@ -1305,12 +1313,14 @@ static int qede_rx_int(struct qede_fastpath *fp, int budget)
 
 	/* Loop to complete all indicated BDs */
 	while ((sw_comp_cons != hw_comp_cons) && (work_done < budget)) {
-		qede_rx_process_cqe(edev, fp, rxq);
+		rcv_pkts += qede_rx_process_cqe(edev, fp, rxq);
 		qed_chain_recycle_consumed(&rxq->rx_comp_ring);
 		sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
 		work_done++;
 	}
 
+	rxq->rcv_pkts += rcv_pkts;
+
 	/* Allocate replacement buffers */
 	while (rxq->num_rx_buffers - rxq->filled_buffers)
 		if (qede_alloc_rx_buffer(rxq, false))

drivers/net/ethernet/qlogic/qede/qede_main.c

@@ -1187,9 +1187,11 @@ static int qede_alloc_mem_rxq(struct qede_dev *edev, struct qede_rx_queue *rxq)
 	rxq->num_rx_buffers = edev->q_num_rx_buffers;
 
 	rxq->rx_buf_size = NET_IP_ALIGN + ETH_OVERHEAD + edev->ndev->mtu;
+	rxq->rx_headroom = edev->xdp_prog ? XDP_PACKET_HEADROOM : 0;
 
-	if (rxq->rx_buf_size > PAGE_SIZE)
-		rxq->rx_buf_size = PAGE_SIZE;
+	/* Make sure that the headroom and  payload fit in a single page */
+	if (rxq->rx_buf_size + rxq->rx_headroom > PAGE_SIZE)
+		rxq->rx_buf_size = PAGE_SIZE - rxq->rx_headroom;
 
 	/* Segment size to spilt a page in multiple equal parts,
 	 * unless XDP is used in which case we'd use the entire page.
@@ -1251,7 +1253,7 @@ static void qede_free_mem_txq(struct qede_dev *edev, struct qede_tx_queue *txq)
 {
 	/* Free the parallel SW ring */
 	if (txq->is_xdp)
-		kfree(txq->sw_tx_ring.pages);
+		kfree(txq->sw_tx_ring.xdp);
 	else
 		kfree(txq->sw_tx_ring.skbs);
 
@@ -1269,9 +1271,9 @@ static int qede_alloc_mem_txq(struct qede_dev *edev, struct qede_tx_queue *txq)
 
 	/* Allocate the parallel driver ring for Tx buffers */
 	if (txq->is_xdp) {
-		size = sizeof(*txq->sw_tx_ring.pages) * TX_RING_SIZE;
-		txq->sw_tx_ring.pages = kzalloc(size, GFP_KERNEL);
-		if (!txq->sw_tx_ring.pages)
+		size = sizeof(*txq->sw_tx_ring.xdp) * TX_RING_SIZE;
+		txq->sw_tx_ring.xdp = kzalloc(size, GFP_KERNEL);
+		if (!txq->sw_tx_ring.xdp)
 			goto err;
 	} else {
 		size = sizeof(*txq->sw_tx_ring.skbs) * TX_RING_SIZE;