[thirdparty/kernel/stable.git] / drivers / net / ethernet / broadcom / bnxt / bnxt_xdp.c

/* Broadcom NetXtreme-C/E network driver.
 *
 * Copyright (c) 2016-2017 Broadcom Limited
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation.
 */
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/pci.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/if_vlan.h>
#include <linux/bpf.h>
#include <linux/bpf_trace.h>
#include <linux/filter.h>
#include <net/page_pool/helpers.h>
#include "bnxt_hsi.h"
#include "bnxt.h"
#include "bnxt_xdp.h"

DEFINE_STATIC_KEY_FALSE(bnxt_xdp_locking_key);

struct bnxt_sw_tx_bd *bnxt_xmit_bd(struct bnxt *bp,
				   struct bnxt_tx_ring_info *txr,
				   dma_addr_t mapping, u32 len,
				   struct xdp_buff *xdp)
{
	struct skb_shared_info *sinfo;
	struct bnxt_sw_tx_bd *tx_buf;
	struct tx_bd *txbd;
	int num_frags = 0;
	u32 flags;
	u16 prod;
	int i;

	if (xdp && xdp_buff_has_frags(xdp)) {
		sinfo = xdp_get_shared_info_from_buff(xdp);
		num_frags = sinfo->nr_frags;
	}

	/* fill up the first buffer */
	prod = txr->tx_prod;
	tx_buf = &txr->tx_buf_ring[prod];
	tx_buf->nr_frags = num_frags;
	if (xdp)
		tx_buf->page = virt_to_head_page(xdp->data);

	txbd = &txr->tx_desc_ring[TX_RING(prod)][TX_IDX(prod)];
	flags = (len << TX_BD_LEN_SHIFT) |
		((num_frags + 1) << TX_BD_FLAGS_BD_CNT_SHIFT) |
		bnxt_lhint_arr[len >> 9];
	txbd->tx_bd_len_flags_type = cpu_to_le32(flags);
	txbd->tx_bd_opaque = prod;
	txbd->tx_bd_haddr = cpu_to_le64(mapping);

	/* now let us fill up the frags into the next buffers */
	for (i = 0; i < num_frags ; i++) {
		skb_frag_t *frag = &sinfo->frags[i];
		struct bnxt_sw_tx_bd *frag_tx_buf;
		dma_addr_t frag_mapping;
		int frag_len;

		prod = NEXT_TX(prod);
		WRITE_ONCE(txr->tx_prod, prod);

		/* first fill up the first buffer */
		frag_tx_buf = &txr->tx_buf_ring[prod];
		frag_tx_buf->page = skb_frag_page(frag);

		txbd = &txr->tx_desc_ring[TX_RING(prod)][TX_IDX(prod)];

		frag_len = skb_frag_size(frag);
		flags = frag_len << TX_BD_LEN_SHIFT;
		txbd->tx_bd_len_flags_type = cpu_to_le32(flags);
		frag_mapping = page_pool_get_dma_addr(skb_frag_page(frag)) +
			       skb_frag_off(frag);
		txbd->tx_bd_haddr = cpu_to_le64(frag_mapping);

		len = frag_len;
	}

	flags &= ~TX_BD_LEN;
	txbd->tx_bd_len_flags_type = cpu_to_le32(((len) << TX_BD_LEN_SHIFT) | flags |
			TX_BD_FLAGS_PACKET_END);
	/* Sync TX BD */
	wmb();
	prod = NEXT_TX(prod);
	WRITE_ONCE(txr->tx_prod, prod);

	return tx_buf;
}

static void __bnxt_xmit_xdp(struct bnxt *bp, struct bnxt_tx_ring_info *txr,
			    dma_addr_t mapping, u32 len, u16 rx_prod,
			    struct xdp_buff *xdp)
{
	struct bnxt_sw_tx_bd *tx_buf;

	tx_buf = bnxt_xmit_bd(bp, txr, mapping, len, xdp);
	tx_buf->rx_prod = rx_prod;
	tx_buf->action = XDP_TX;

}

static void __bnxt_xmit_xdp_redirect(struct bnxt *bp,
				     struct bnxt_tx_ring_info *txr,
				     dma_addr_t mapping, u32 len,
				     struct xdp_frame *xdpf)
{
	struct bnxt_sw_tx_bd *tx_buf;

	tx_buf = bnxt_xmit_bd(bp, txr, mapping, len, NULL);
	tx_buf->action = XDP_REDIRECT;
	tx_buf->xdpf = xdpf;
	dma_unmap_addr_set(tx_buf, mapping, mapping);
	dma_unmap_len_set(tx_buf, len, 0);
}

void bnxt_tx_int_xdp(struct bnxt *bp, struct bnxt_napi *bnapi, int budget)
{
	struct bnxt_tx_ring_info *txr = bnapi->tx_ring;
	struct bnxt_rx_ring_info *rxr = bnapi->rx_ring;
	bool rx_doorbell_needed = false;
	int nr_pkts = bnapi->tx_pkts;
	struct bnxt_sw_tx_bd *tx_buf;
	u16 tx_cons = txr->tx_cons;
	u16 last_tx_cons = tx_cons;
	int i, j, frags;

	if (!budget)
		return;

	for (i = 0; i < nr_pkts; i++) {
		tx_buf = &txr->tx_buf_ring[tx_cons];

		if (tx_buf->action == XDP_REDIRECT) {
			struct pci_dev *pdev = bp->pdev;

			dma_unmap_single(&pdev->dev,
					 dma_unmap_addr(tx_buf, mapping),
					 dma_unmap_len(tx_buf, len),
					 DMA_TO_DEVICE);
			xdp_return_frame(tx_buf->xdpf);
			tx_buf->action = 0;
			tx_buf->xdpf = NULL;
		} else if (tx_buf->action == XDP_TX) {
			tx_buf->action = 0;
			rx_doorbell_needed = true;
			last_tx_cons = tx_cons;

			frags = tx_buf->nr_frags;
			for (j = 0; j < frags; j++) {
				tx_cons = NEXT_TX(tx_cons);
				tx_buf = &txr->tx_buf_ring[tx_cons];
				page_pool_recycle_direct(rxr->page_pool, tx_buf->page);
			}
		} else {
			bnxt_sched_reset_txr(bp, txr, i);
			return;
		}
		tx_cons = NEXT_TX(tx_cons);
	}

	bnapi->tx_pkts = 0;
	WRITE_ONCE(txr->tx_cons, tx_cons);
	if (rx_doorbell_needed) {
		tx_buf = &txr->tx_buf_ring[last_tx_cons];
		bnxt_db_write(bp, &rxr->rx_db, tx_buf->rx_prod);

	}
}

bool bnxt_xdp_attached(struct bnxt *bp, struct bnxt_rx_ring_info *rxr)
{
	struct bpf_prog *xdp_prog = READ_ONCE(rxr->xdp_prog);

	return !!xdp_prog;
}

void bnxt_xdp_buff_init(struct bnxt *bp, struct bnxt_rx_ring_info *rxr,
			u16 cons, u8 *data_ptr, unsigned int len,
			struct xdp_buff *xdp)
{
	u32 buflen = BNXT_RX_PAGE_SIZE;
	struct bnxt_sw_rx_bd *rx_buf;
	struct pci_dev *pdev;
	dma_addr_t mapping;
	u32 offset;

	pdev = bp->pdev;
	rx_buf = &rxr->rx_buf_ring[cons];
	offset = bp->rx_offset;

	mapping = rx_buf->mapping - bp->rx_dma_offset;
	dma_sync_single_for_cpu(&pdev->dev, mapping + offset, len, bp->rx_dir);

	xdp_init_buff(xdp, buflen, &rxr->xdp_rxq);
	xdp_prepare_buff(xdp, data_ptr - offset, offset, len, false);
}

void bnxt_xdp_buff_frags_free(struct bnxt_rx_ring_info *rxr,
			      struct xdp_buff *xdp)
{
	struct skb_shared_info *shinfo;
	int i;

	if (!xdp || !xdp_buff_has_frags(xdp))
		return;
	shinfo = xdp_get_shared_info_from_buff(xdp);
	for (i = 0; i < shinfo->nr_frags; i++) {
		struct page *page = skb_frag_page(&shinfo->frags[i]);

		page_pool_recycle_direct(rxr->page_pool, page);
	}
	shinfo->nr_frags = 0;
}

/* returns the following:
 * true    - packet consumed by XDP and new buffer is allocated.
 * false   - packet should be passed to the stack.
 */
bool bnxt_rx_xdp(struct bnxt *bp, struct bnxt_rx_ring_info *rxr, u16 cons,
		 struct xdp_buff xdp, struct page *page, u8 **data_ptr,
		 unsigned int *len, u8 *event)
{
	struct bpf_prog *xdp_prog = READ_ONCE(rxr->xdp_prog);
	struct bnxt_tx_ring_info *txr;
	struct bnxt_sw_rx_bd *rx_buf;
	struct pci_dev *pdev;
	dma_addr_t mapping;
	u32 tx_needed = 1;
	void *orig_data;
	u32 tx_avail;
	u32 offset;
	u32 act;

	if (!xdp_prog)
		return false;

	pdev = bp->pdev;
	offset = bp->rx_offset;

	txr = rxr->bnapi->tx_ring;
	/* BNXT_RX_PAGE_MODE(bp) when XDP enabled */
	orig_data = xdp.data;

	act = bpf_prog_run_xdp(xdp_prog, &xdp);

	tx_avail = bnxt_tx_avail(bp, txr);
	/* If the tx ring is not full, we must not update the rx producer yet
	 * because we may still be transmitting on some BDs.
	 */
	if (tx_avail != bp->tx_ring_size)
		*event &= ~BNXT_RX_EVENT;

	*len = xdp.data_end - xdp.data;
	if (orig_data != xdp.data) {
		offset = xdp.data - xdp.data_hard_start;
		*data_ptr = xdp.data_hard_start + offset;
	}

	switch (act) {
	case XDP_PASS:
		return false;

	case XDP_TX:
		rx_buf = &rxr->rx_buf_ring[cons];
		mapping = rx_buf->mapping - bp->rx_dma_offset;
		*event = 0;

		if (unlikely(xdp_buff_has_frags(&xdp))) {
			struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(&xdp);

			tx_needed += sinfo->nr_frags;
			*event = BNXT_AGG_EVENT;
		}

		if (tx_avail < tx_needed) {
			trace_xdp_exception(bp->dev, xdp_prog, act);
			bnxt_xdp_buff_frags_free(rxr, &xdp);
			bnxt_reuse_rx_data(rxr, cons, page);
			return true;
		}

		dma_sync_single_for_device(&pdev->dev, mapping + offset, *len,
					   bp->rx_dir);

		*event |= BNXT_TX_EVENT;
		__bnxt_xmit_xdp(bp, txr, mapping + offset, *len,
				NEXT_RX(rxr->rx_prod), &xdp);
		bnxt_reuse_rx_data(rxr, cons, page);
		return true;
	case XDP_REDIRECT:
		/* if we are calling this here then we know that the
		 * redirect is coming from a frame received by the
		 * bnxt_en driver.
		 */
		rx_buf = &rxr->rx_buf_ring[cons];
		mapping = rx_buf->mapping - bp->rx_dma_offset;
		dma_unmap_page_attrs(&pdev->dev, mapping,
				     BNXT_RX_PAGE_SIZE, bp->rx_dir,
				     DMA_ATTR_WEAK_ORDERING);

		/* if we are unable to allocate a new buffer, abort and reuse */
		if (bnxt_alloc_rx_data(bp, rxr, rxr->rx_prod, GFP_ATOMIC)) {
			trace_xdp_exception(bp->dev, xdp_prog, act);
			bnxt_xdp_buff_frags_free(rxr, &xdp);
			bnxt_reuse_rx_data(rxr, cons, page);
			return true;
		}

		if (xdp_do_redirect(bp->dev, &xdp, xdp_prog)) {
			trace_xdp_exception(bp->dev, xdp_prog, act);
			page_pool_recycle_direct(rxr->page_pool, page);
			return true;
		}

		*event |= BNXT_REDIRECT_EVENT;
		break;
	default:
		bpf_warn_invalid_xdp_action(bp->dev, xdp_prog, act);
		fallthrough;
	case XDP_ABORTED:
		trace_xdp_exception(bp->dev, xdp_prog, act);
		fallthrough;
	case XDP_DROP:
		bnxt_xdp_buff_frags_free(rxr, &xdp);
		bnxt_reuse_rx_data(rxr, cons, page);
		break;
	}
	return true;
}

int bnxt_xdp_xmit(struct net_device *dev, int num_frames,
		  struct xdp_frame **frames, u32 flags)
{
	struct bnxt *bp = netdev_priv(dev);
	struct bpf_prog *xdp_prog = READ_ONCE(bp->xdp_prog);
	struct pci_dev *pdev = bp->pdev;
	struct bnxt_tx_ring_info *txr;
	dma_addr_t mapping;
	int nxmit = 0;
	int ring;
	int i;

	if (!test_bit(BNXT_STATE_OPEN, &bp->state) ||
	    !bp->tx_nr_rings_xdp ||
	    !xdp_prog)
		return -EINVAL;

	ring = smp_processor_id() % bp->tx_nr_rings_xdp;
	txr = &bp->tx_ring[ring];

	if (READ_ONCE(txr->dev_state) == BNXT_DEV_STATE_CLOSING)
		return -EINVAL;

	if (static_branch_unlikely(&bnxt_xdp_locking_key))
		spin_lock(&txr->xdp_tx_lock);

	for (i = 0; i < num_frames; i++) {
		struct xdp_frame *xdp = frames[i];

		if (!bnxt_tx_avail(bp, txr))
			break;

		mapping = dma_map_single(&pdev->dev, xdp->data, xdp->len,
					 DMA_TO_DEVICE);

		if (dma_mapping_error(&pdev->dev, mapping))
			break;

		__bnxt_xmit_xdp_redirect(bp, txr, mapping, xdp->len, xdp);
		nxmit++;
	}

	if (flags & XDP_XMIT_FLUSH) {
		/* Sync BD data before updating doorbell */
		wmb();
		bnxt_db_write(bp, &txr->tx_db, txr->tx_prod);
	}

	if (static_branch_unlikely(&bnxt_xdp_locking_key))
		spin_unlock(&txr->xdp_tx_lock);

	return nxmit;
}

/* Under rtnl_lock */
static int bnxt_xdp_set(struct bnxt *bp, struct bpf_prog *prog)
{
	struct net_device *dev = bp->dev;
	int tx_xdp = 0, rc, tc;
	struct bpf_prog *old;

	if (prog && !prog->aux->xdp_has_frags &&
	    bp->dev->mtu > BNXT_MAX_PAGE_MODE_MTU) {
		netdev_warn(dev, "MTU %d larger than %d without XDP frag support.\n",
			    bp->dev->mtu, BNXT_MAX_PAGE_MODE_MTU);
		return -EOPNOTSUPP;
	}
	if (!(bp->flags & BNXT_FLAG_SHARED_RINGS)) {
		netdev_warn(dev, "ethtool rx/tx channels must be combined to support XDP.\n");
		return -EOPNOTSUPP;
	}
	if (prog)
		tx_xdp = bp->rx_nr_rings;

	tc = netdev_get_num_tc(dev);
	if (!tc)
		tc = 1;
	rc = bnxt_check_rings(bp, bp->tx_nr_rings_per_tc, bp->rx_nr_rings,
			      true, tc, tx_xdp);
	if (rc) {
		netdev_warn(dev, "Unable to reserve enough TX rings to support XDP.\n");
		return rc;
	}
	if (netif_running(dev))
		bnxt_close_nic(bp, true, false);

	old = xchg(&bp->xdp_prog, prog);
	if (old)
		bpf_prog_put(old);

	if (prog) {
		bnxt_set_rx_skb_mode(bp, true);
		xdp_features_set_redirect_target(dev, true);
	} else {
		int rx, tx;

		xdp_features_clear_redirect_target(dev);
		bnxt_set_rx_skb_mode(bp, false);
		bnxt_get_max_rings(bp, &rx, &tx, true);
		if (rx > 1) {
			bp->flags &= ~BNXT_FLAG_NO_AGG_RINGS;
			bp->dev->hw_features |= NETIF_F_LRO;
		}
	}
	bp->tx_nr_rings_xdp = tx_xdp;
	bp->tx_nr_rings = bp->tx_nr_rings_per_tc * tc + tx_xdp;
	bp->cp_nr_rings = max_t(int, bp->tx_nr_rings, bp->rx_nr_rings);
	bnxt_set_tpa_flags(bp);
	bnxt_set_ring_params(bp);

	if (netif_running(dev))
		return bnxt_open_nic(bp, true, false);

	return 0;
}

int bnxt_xdp(struct net_device *dev, struct netdev_bpf *xdp)
{
	struct bnxt *bp = netdev_priv(dev);
	int rc;

	switch (xdp->command) {
	case XDP_SETUP_PROG:
		rc = bnxt_xdp_set(bp, xdp->prog);
		break;
	default:
		rc = -EINVAL;
		break;
	}
	return rc;
}

struct sk_buff *
bnxt_xdp_build_skb(struct bnxt *bp, struct sk_buff *skb, u8 num_frags,
		   struct page_pool *pool, struct xdp_buff *xdp,
		   struct rx_cmp_ext *rxcmp1)
{
	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);

	if (!skb)
		return NULL;
	skb_checksum_none_assert(skb);
	if (RX_CMP_L4_CS_OK(rxcmp1)) {
		if (bp->dev->features & NETIF_F_RXCSUM) {
			skb->ip_summed = CHECKSUM_UNNECESSARY;
			skb->csum_level = RX_CMP_ENCAP(rxcmp1);
		}
	}
	xdp_update_skb_shared_info(skb, num_frags,
				   sinfo->xdp_frags_size,
				   BNXT_RX_PAGE_SIZE * sinfo->nr_frags,
				   xdp_buff_is_frag_pfmemalloc(xdp));
	return skb;
}
Commit	Line	Data
c6d30e83 MC	1	/* Broadcom NetXtreme-C/E network driver.
	2	*
	3	* Copyright (c) 2016-2017 Broadcom Limited
	4	*
	5	* This program is free software; you can redistribute it and/or modify
	6	* it under the terms of the GNU General Public License as published by
	7	* the Free Software Foundation.
	8	*/
	9	#include <linux/kernel.h>
	10	#include <linux/errno.h>
	11	#include <linux/pci.h>
	12	#include <linux/netdevice.h>
	13	#include <linux/etherdevice.h>
	14	#include <linux/if_vlan.h>
	15	#include <linux/bpf.h>
	16	#include <linux/bpf_trace.h>
	17	#include <linux/filter.h>
a9ca9f9c	18	#include <net/page_pool/helpers.h>
c6d30e83 MC	19	#include "bnxt_hsi.h"
	20	#include "bnxt.h"
	21	#include "bnxt_xdp.h"
	22
4f81def2 PC	23	DEFINE_STATIC_KEY_FALSE(bnxt_xdp_locking_key);
4f81def2 PC	24
c1ba92a8 MC	25	struct bnxt_sw_tx_bd bnxt_xmit_bd(struct bnxt bp,
c1ba92a8 MC	26	struct bnxt_tx_ring_info *txr,
a7559bc8 AG	27	dma_addr_t mapping, u32 len,
a7559bc8 AG	28	struct xdp_buff *xdp)
38413406	29	{
a7559bc8	30	struct skb_shared_info *sinfo;
53f8c2d3	31	struct bnxt_sw_tx_bd *tx_buf;
38413406	32	struct tx_bd *txbd;
a7559bc8	33	int num_frags = 0;
38413406 MC	34	u32 flags;
38413406 MC	35	u16 prod;
a7559bc8 AG	36	int i;
	37
	38	if (xdp && xdp_buff_has_frags(xdp)) {
	39	sinfo = xdp_get_shared_info_from_buff(xdp);
	40	num_frags = sinfo->nr_frags;
	41	}
38413406	42
a7559bc8	43	/* fill up the first buffer */
38413406 MC	44	prod = txr->tx_prod;
38413406 MC	45	tx_buf = &txr->tx_buf_ring[prod];
a7559bc8 AG	46	tx_buf->nr_frags = num_frags;
	47	if (xdp)
	48	tx_buf->page = virt_to_head_page(xdp->data);
38413406 MC	49
38413406 MC	50	txbd = &txr->tx_desc_ring[TX_RING(prod)][TX_IDX(prod)];
53f8c2d3 MC	51	flags = (len << TX_BD_LEN_SHIFT) \|
	52	((num_frags + 1) << TX_BD_FLAGS_BD_CNT_SHIFT) \|
	53	bnxt_lhint_arr[len >> 9];
38413406 MC	54	txbd->tx_bd_len_flags_type = cpu_to_le32(flags);
	55	txbd->tx_bd_opaque = prod;
	56	txbd->tx_bd_haddr = cpu_to_le64(mapping);
	57
a7559bc8 AG	58	/* now let us fill up the frags into the next buffers */
	59	for (i = 0; i < num_frags ; i++) {
	60	skb_frag_t *frag = &sinfo->frags[i];
	61	struct bnxt_sw_tx_bd *frag_tx_buf;
a7559bc8 AG	62	dma_addr_t frag_mapping;
	63	int frag_len;
	64
	65	prod = NEXT_TX(prod);
36647b20	66	WRITE_ONCE(txr->tx_prod, prod);
a7559bc8 AG	67
	68	/* first fill up the first buffer */
	69	frag_tx_buf = &txr->tx_buf_ring[prod];
	70	frag_tx_buf->page = skb_frag_page(frag);
	71
	72	txbd = &txr->tx_desc_ring[TX_RING(prod)][TX_IDX(prod)];
	73
	74	frag_len = skb_frag_size(frag);
a7559bc8 AG	75	flags = frag_len << TX_BD_LEN_SHIFT;
a7559bc8 AG	76	txbd->tx_bd_len_flags_type = cpu_to_le32(flags);
23c93c3b AG	77	frag_mapping = page_pool_get_dma_addr(skb_frag_page(frag)) +
23c93c3b AG	78	skb_frag_off(frag);
a7559bc8 AG	79	txbd->tx_bd_haddr = cpu_to_le64(frag_mapping);
	80
	81	len = frag_len;
	82	}
	83
	84	flags &= ~TX_BD_LEN;
	85	txbd->tx_bd_len_flags_type = cpu_to_le32(((len) << TX_BD_LEN_SHIFT) \| flags \|
	86	TX_BD_FLAGS_PACKET_END);
	87	/* Sync TX BD */
	88	wmb();
38413406	89	prod = NEXT_TX(prod);
36647b20	90	WRITE_ONCE(txr->tx_prod, prod);
a7559bc8	91
53f8c2d3	92	return tx_buf;
c1ba92a8 MC	93	}
	94
	95	static void __bnxt_xmit_xdp(struct bnxt bp, struct bnxt_tx_ring_info txr,
a7559bc8 AG	96	dma_addr_t mapping, u32 len, u16 rx_prod,
a7559bc8 AG	97	struct xdp_buff *xdp)
c1ba92a8 MC	98	{
	99	struct bnxt_sw_tx_bd *tx_buf;
	100
a7559bc8	101	tx_buf = bnxt_xmit_bd(bp, txr, mapping, len, xdp);
c1ba92a8 MC	102	tx_buf->rx_prod = rx_prod;
c1ba92a8 MC	103	tx_buf->action = XDP_TX;
a7559bc8	104
38413406 MC	105	}
38413406 MC	106
f18c2b77 AG	107	static void __bnxt_xmit_xdp_redirect(struct bnxt *bp,
	108	struct bnxt_tx_ring_info *txr,
	109	dma_addr_t mapping, u32 len,
	110	struct xdp_frame *xdpf)
	111	{
	112	struct bnxt_sw_tx_bd *tx_buf;
	113
a7559bc8	114	tx_buf = bnxt_xmit_bd(bp, txr, mapping, len, NULL);
f18c2b77 AG	115	tx_buf->action = XDP_REDIRECT;
	116	tx_buf->xdpf = xdpf;
	117	dma_unmap_addr_set(tx_buf, mapping, mapping);
	118	dma_unmap_len_set(tx_buf, len, 0);
	119	}
	120
37b61cda	121	void bnxt_tx_int_xdp(struct bnxt bp, struct bnxt_napi bnapi, int budget)
38413406 MC	122	{
	123	struct bnxt_tx_ring_info *txr = bnapi->tx_ring;
	124	struct bnxt_rx_ring_info *rxr = bnapi->rx_ring;
c1ba92a8	125	bool rx_doorbell_needed = false;
37b61cda	126	int nr_pkts = bnapi->tx_pkts;
38413406 MC	127	struct bnxt_sw_tx_bd *tx_buf;
	128	u16 tx_cons = txr->tx_cons;
	129	u16 last_tx_cons = tx_cons;
a7559bc8	130	int i, j, frags;
38413406	131
37b61cda JK	132	if (!budget)
	133	return;
	134
38413406	135	for (i = 0; i < nr_pkts; i++) {
c1ba92a8 MC	136	tx_buf = &txr->tx_buf_ring[tx_cons];
c1ba92a8 MC	137
f18c2b77 AG	138	if (tx_buf->action == XDP_REDIRECT) {
	139	struct pci_dev *pdev = bp->pdev;
	140
	141	dma_unmap_single(&pdev->dev,
	142	dma_unmap_addr(tx_buf, mapping),
	143	dma_unmap_len(tx_buf, len),
df70303d	144	DMA_TO_DEVICE);
f18c2b77 AG	145	xdp_return_frame(tx_buf->xdpf);
	146	tx_buf->action = 0;
	147	tx_buf->xdpf = NULL;
	148	} else if (tx_buf->action == XDP_TX) {
2b56b3d9	149	tx_buf->action = 0;
c1ba92a8 MC	150	rx_doorbell_needed = true;
c1ba92a8 MC	151	last_tx_cons = tx_cons;
a7559bc8 AG	152
	153	frags = tx_buf->nr_frags;
	154	for (j = 0; j < frags; j++) {
	155	tx_cons = NEXT_TX(tx_cons);
	156	tx_buf = &txr->tx_buf_ring[tx_cons];
	157	page_pool_recycle_direct(rxr->page_pool, tx_buf->page);
	158	}
2b56b3d9 JK	159	} else {
	160	bnxt_sched_reset_txr(bp, txr, i);
	161	return;
c1ba92a8	162	}
38413406	163	tx_cons = NEXT_TX(tx_cons);
38413406	164	}
37b61cda JK	165
37b61cda JK	166	bnapi->tx_pkts = 0;
36647b20	167	WRITE_ONCE(txr->tx_cons, tx_cons);
c1ba92a8	168	if (rx_doorbell_needed) {
38413406	169	tx_buf = &txr->tx_buf_ring[last_tx_cons];
c1ba92a8	170	bnxt_db_write(bp, &rxr->rx_db, tx_buf->rx_prod);
a7559bc8	171
38413406	172	}
38413406 MC	173	}
38413406 MC	174
b231c3f3 AG	175	bool bnxt_xdp_attached(struct bnxt bp, struct bnxt_rx_ring_info rxr)
	176	{
	177	struct bpf_prog *xdp_prog = READ_ONCE(rxr->xdp_prog);
	178
	179	return !!xdp_prog;
	180	}
	181
	182	void bnxt_xdp_buff_init(struct bnxt bp, struct bnxt_rx_ring_info rxr,
bbfc17e5	183	u16 cons, u8 *data_ptr, unsigned int len,
b231c3f3 AG	184	struct xdp_buff *xdp)
b231c3f3 AG	185	{
f6974b4c	186	u32 buflen = BNXT_RX_PAGE_SIZE;
b231c3f3 AG	187	struct bnxt_sw_rx_bd *rx_buf;
	188	struct pci_dev *pdev;
	189	dma_addr_t mapping;
	190	u32 offset;
	191
	192	pdev = bp->pdev;
	193	rx_buf = &rxr->rx_buf_ring[cons];
	194	offset = bp->rx_offset;
	195
	196	mapping = rx_buf->mapping - bp->rx_dma_offset;
bbfc17e5	197	dma_sync_single_for_cpu(&pdev->dev, mapping + offset, len, bp->rx_dir);
b231c3f3	198
7dd3de7c	199	xdp_init_buff(xdp, buflen, &rxr->xdp_rxq);
bbfc17e5	200	xdp_prepare_buff(xdp, data_ptr - offset, offset, len, false);
b231c3f3 AG	201	}
b231c3f3 AG	202
a7559bc8 AG	203	void bnxt_xdp_buff_frags_free(struct bnxt_rx_ring_info *rxr,
	204	struct xdp_buff *xdp)
	205	{
	206	struct skb_shared_info *shinfo;
	207	int i;
	208
	209	if (!xdp \|\| !xdp_buff_has_frags(xdp))
	210	return;
	211	shinfo = xdp_get_shared_info_from_buff(xdp);
	212	for (i = 0; i < shinfo->nr_frags; i++) {
	213	struct page *page = skb_frag_page(&shinfo->frags[i]);
	214
	215	page_pool_recycle_direct(rxr->page_pool, page);
	216	}
	217	shinfo->nr_frags = 0;
	218	}
	219
c6d30e83 MC	220	/* returns the following:
	221	* true - packet consumed by XDP and new buffer is allocated.
	222	* false - packet should be passed to the stack.
	223	*/
	224	bool bnxt_rx_xdp(struct bnxt bp, struct bnxt_rx_ring_info rxr, u16 cons,
9b3e6078 MC	225	struct xdp_buff xdp, struct page page, u8 *data_ptr,
9b3e6078 MC	226	unsigned int len, u8 event)
c6d30e83 MC	227	{
c6d30e83 MC	228	struct bpf_prog *xdp_prog = READ_ONCE(rxr->xdp_prog);
38413406	229	struct bnxt_tx_ring_info *txr;
c6d30e83 MC	230	struct bnxt_sw_rx_bd *rx_buf;
c6d30e83 MC	231	struct pci_dev *pdev;
c6d30e83	232	dma_addr_t mapping;
a7559bc8	233	u32 tx_needed = 1;
c6d30e83	234	void *orig_data;
38413406	235	u32 tx_avail;
c6d30e83 MC	236	u32 offset;
	237	u32 act;
	238
	239	if (!xdp_prog)
	240	return false;
	241
	242	pdev = bp->pdev;
c6d30e83 MC	243	offset = bp->rx_offset;
c6d30e83 MC	244
f18c2b77	245	txr = rxr->bnapi->tx_ring;
43b5169d	246	/* BNXT_RX_PAGE_MODE(bp) when XDP enabled */
c6d30e83	247	orig_data = xdp.data;
c6d30e83	248
c6d30e83	249	act = bpf_prog_run_xdp(xdp_prog, &xdp);
c6d30e83	250
38413406 MC	251	tx_avail = bnxt_tx_avail(bp, txr);
	252	/* If the tx ring is not full, we must not update the rx producer yet
	253	* because we may still be transmitting on some BDs.
	254	*/
	255	if (tx_avail != bp->tx_ring_size)
	256	*event &= ~BNXT_RX_EVENT;
	257
b968e735	258	*len = xdp.data_end - xdp.data;
9b3e6078	259	if (orig_data != xdp.data) {
c6d30e83	260	offset = xdp.data - xdp.data_hard_start;
9b3e6078 MC	261	*data_ptr = xdp.data_hard_start + offset;
9b3e6078 MC	262	}
b231c3f3	263
c6d30e83 MC	264	switch (act) {
	265	case XDP_PASS:
	266	return false;
	267
38413406	268	case XDP_TX:
b231c3f3 AG	269	rx_buf = &rxr->rx_buf_ring[cons];
b231c3f3 AG	270	mapping = rx_buf->mapping - bp->rx_dma_offset;
a7559bc8 AG	271	*event = 0;
	272
	273	if (unlikely(xdp_buff_has_frags(&xdp))) {
	274	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(&xdp);
b231c3f3	275
a7559bc8 AG	276	tx_needed += sinfo->nr_frags;
	277	*event = BNXT_AGG_EVENT;
	278	}
	279
	280	if (tx_avail < tx_needed) {
38413406	281	trace_xdp_exception(bp->dev, xdp_prog, act);
a7559bc8	282	bnxt_xdp_buff_frags_free(rxr, &xdp);
38413406 MC	283	bnxt_reuse_rx_data(rxr, cons, page);
	284	return true;
	285	}
	286
38413406 MC	287	dma_sync_single_for_device(&pdev->dev, mapping + offset, *len,
38413406 MC	288	bp->rx_dir);
a7559bc8 AG	289
a7559bc8 AG	290	*event \|= BNXT_TX_EVENT;
52c06092	291	__bnxt_xmit_xdp(bp, txr, mapping + offset, *len,
a7559bc8	292	NEXT_RX(rxr->rx_prod), &xdp);
38413406 MC	293	bnxt_reuse_rx_data(rxr, cons, page);
38413406 MC	294	return true;
f18c2b77 AG	295	case XDP_REDIRECT:
	296	/* if we are calling this here then we know that the
	297	* redirect is coming from a frame received by the
	298	* bnxt_en driver.
	299	*/
b231c3f3 AG	300	rx_buf = &rxr->rx_buf_ring[cons];
b231c3f3 AG	301	mapping = rx_buf->mapping - bp->rx_dma_offset;
f18c2b77	302	dma_unmap_page_attrs(&pdev->dev, mapping,
f6974b4c	303	BNXT_RX_PAGE_SIZE, bp->rx_dir,
f18c2b77 AG	304	DMA_ATTR_WEAK_ORDERING);
	305
	306	/* if we are unable to allocate a new buffer, abort and reuse */
	307	if (bnxt_alloc_rx_data(bp, rxr, rxr->rx_prod, GFP_ATOMIC)) {
	308	trace_xdp_exception(bp->dev, xdp_prog, act);
a7559bc8	309	bnxt_xdp_buff_frags_free(rxr, &xdp);
f18c2b77 AG	310	bnxt_reuse_rx_data(rxr, cons, page);
	311	return true;
	312	}
	313
	314	if (xdp_do_redirect(bp->dev, &xdp, xdp_prog)) {
	315	trace_xdp_exception(bp->dev, xdp_prog, act);
322b87ca	316	page_pool_recycle_direct(rxr->page_pool, page);
f18c2b77 AG	317	return true;
	318	}
	319
	320	*event \|= BNXT_REDIRECT_EVENT;
	321	break;
c6d30e83	322	default:
c8064e5b	323	bpf_warn_invalid_xdp_action(bp->dev, xdp_prog, act);
df561f66	324	fallthrough;
c6d30e83 MC	325	case XDP_ABORTED:
c6d30e83 MC	326	trace_xdp_exception(bp->dev, xdp_prog, act);
df561f66	327	fallthrough;
c6d30e83	328	case XDP_DROP:
a7559bc8	329	bnxt_xdp_buff_frags_free(rxr, &xdp);
c6d30e83 MC	330	bnxt_reuse_rx_data(rxr, cons, page);
	331	break;
	332	}
	333	return true;
	334	}
	335
f18c2b77 AG	336	int bnxt_xdp_xmit(struct net_device *dev, int num_frames,
	337	struct xdp_frame **frames, u32 flags)
	338	{
	339	struct bnxt *bp = netdev_priv(dev);
	340	struct bpf_prog *xdp_prog = READ_ONCE(bp->xdp_prog);
	341	struct pci_dev *pdev = bp->pdev;
	342	struct bnxt_tx_ring_info *txr;
	343	dma_addr_t mapping;
fdc13979	344	int nxmit = 0;
f18c2b77 AG	345	int ring;
	346	int i;
	347
	348	if (!test_bit(BNXT_STATE_OPEN, &bp->state) \|\|
	349	!bp->tx_nr_rings_xdp \|\|
	350	!xdp_prog)
	351	return -EINVAL;
	352
	353	ring = smp_processor_id() % bp->tx_nr_rings_xdp;
	354	txr = &bp->tx_ring[ring];
	355
27d4073f RJ	356	if (READ_ONCE(txr->dev_state) == BNXT_DEV_STATE_CLOSING)
	357	return -EINVAL;
	358
4f81def2 PC	359	if (static_branch_unlikely(&bnxt_xdp_locking_key))
	360	spin_lock(&txr->xdp_tx_lock);
	361
f18c2b77 AG	362	for (i = 0; i < num_frames; i++) {
	363	struct xdp_frame *xdp = frames[i];
	364
27d4073f	365	if (!bnxt_tx_avail(bp, txr))
fdc13979	366	break;
f18c2b77 AG	367
	368	mapping = dma_map_single(&pdev->dev, xdp->data, xdp->len,
	369	DMA_TO_DEVICE);
	370
fdc13979 LB	371	if (dma_mapping_error(&pdev->dev, mapping))
	372	break;
	373
f18c2b77	374	__bnxt_xmit_xdp_redirect(bp, txr, mapping, xdp->len, xdp);
fdc13979	375	nxmit++;
f18c2b77 AG	376	}
	377
	378	if (flags & XDP_XMIT_FLUSH) {
	379	/* Sync BD data before updating doorbell */
	380	wmb();
	381	bnxt_db_write(bp, &txr->tx_db, txr->tx_prod);
	382	}
	383
4f81def2 PC	384	if (static_branch_unlikely(&bnxt_xdp_locking_key))
	385	spin_unlock(&txr->xdp_tx_lock);
	386
fdc13979	387	return nxmit;
f18c2b77 AG	388	}
f18c2b77 AG	389
c6d30e83 MC	390	/* Under rtnl_lock */
	391	static int bnxt_xdp_set(struct bnxt bp, struct bpf_prog prog)
	392	{
	393	struct net_device *dev = bp->dev;
	394	int tx_xdp = 0, rc, tc;
	395	struct bpf_prog *old;
	396
9f4b2830 AG	397	if (prog && !prog->aux->xdp_has_frags &&
	398	bp->dev->mtu > BNXT_MAX_PAGE_MODE_MTU) {
	399	netdev_warn(dev, "MTU %d larger than %d without XDP frag support.\n",
c6d30e83 MC	400	bp->dev->mtu, BNXT_MAX_PAGE_MODE_MTU);
	401	return -EOPNOTSUPP;
	402	}
	403	if (!(bp->flags & BNXT_FLAG_SHARED_RINGS)) {
	404	netdev_warn(dev, "ethtool rx/tx channels must be combined to support XDP.\n");
	405	return -EOPNOTSUPP;
	406	}
1abeacc1	407	if (prog)
c6d30e83 MC	408	tx_xdp = bp->rx_nr_rings;
	409
	410	tc = netdev_get_num_tc(dev);
	411	if (!tc)
	412	tc = 1;
98fdbe73 MC	413	rc = bnxt_check_rings(bp, bp->tx_nr_rings_per_tc, bp->rx_nr_rings,
98fdbe73 MC	414	true, tc, tx_xdp);
c6d30e83 MC	415	if (rc) {
	416	netdev_warn(dev, "Unable to reserve enough TX rings to support XDP.\n");
	417	return rc;
	418	}
	419	if (netif_running(dev))
	420	bnxt_close_nic(bp, true, false);
	421
	422	old = xchg(&bp->xdp_prog, prog);
	423	if (old)
	424	bpf_prog_put(old);
	425
	426	if (prog) {
	427	bnxt_set_rx_skb_mode(bp, true);
66c0e13a	428	xdp_features_set_redirect_target(dev, true);
c6d30e83 MC	429	} else {
	430	int rx, tx;
	431
66c0e13a	432	xdp_features_clear_redirect_target(dev);
c6d30e83 MC	433	bnxt_set_rx_skb_mode(bp, false);
	434	bnxt_get_max_rings(bp, &rx, &tx, true);
	435	if (rx > 1) {
	436	bp->flags &= ~BNXT_FLAG_NO_AGG_RINGS;
	437	bp->dev->hw_features \|= NETIF_F_LRO;
	438	}
	439	}
	440	bp->tx_nr_rings_xdp = tx_xdp;
	441	bp->tx_nr_rings = bp->tx_nr_rings_per_tc * tc + tx_xdp;
	442	bp->cp_nr_rings = max_t(int, bp->tx_nr_rings, bp->rx_nr_rings);
c6d30e83 MC	443	bnxt_set_tpa_flags(bp);
	444	bnxt_set_ring_params(bp);
	445
	446	if (netif_running(dev))
	447	return bnxt_open_nic(bp, true, false);
	448
	449	return 0;
	450	}
	451
f4e63525	452	int bnxt_xdp(struct net_device dev, struct netdev_bpf xdp)
c6d30e83 MC	453	{
	454	struct bnxt *bp = netdev_priv(dev);
	455	int rc;
	456
	457	switch (xdp->command) {
	458	case XDP_SETUP_PROG:
	459	rc = bnxt_xdp_set(bp, xdp->prog);
	460	break;
c6d30e83 MC	461	default:
	462	rc = -EINVAL;
	463	break;
	464	}
	465	return rc;
	466	}
1dc4c557 AG	467
	468	struct sk_buff *
	469	bnxt_xdp_build_skb(struct bnxt bp, struct sk_buff skb, u8 num_frags,
	470	struct page_pool pool, struct xdp_buff xdp,
	471	struct rx_cmp_ext *rxcmp1)
	472	{
	473	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
	474
	475	if (!skb)
	476	return NULL;
	477	skb_checksum_none_assert(skb);
	478	if (RX_CMP_L4_CS_OK(rxcmp1)) {
	479	if (bp->dev->features & NETIF_F_RXCSUM) {
	480	skb->ip_summed = CHECKSUM_UNNECESSARY;
	481	skb->csum_level = RX_CMP_ENCAP(rxcmp1);
	482	}
	483	}
	484	xdp_update_skb_shared_info(skb, num_frags,
	485	sinfo->xdp_frags_size,
f6974b4c	486	BNXT_RX_PAGE_SIZE * sinfo->nr_frags,
1dc4c557 AG	487	xdp_buff_is_frag_pfmemalloc(xdp));
	488	return skb;
	489	}