[linux-block.git] / drivers / net / ethernet / sfc / rx.c

// SPDX-License-Identifier: GPL-2.0-only
/****************************************************************************
 * Driver for Solarflare network controllers and boards
 * Copyright 2005-2006 Fen Systems Ltd.
 * Copyright 2005-2013 Solarflare Communications Inc.
 */

#include <linux/socket.h>
#include <linux/in.h>
#include <linux/slab.h>
#include <linux/ip.h>
#include <linux/ipv6.h>
#include <linux/tcp.h>
#include <linux/udp.h>
#include <linux/prefetch.h>
#include <linux/moduleparam.h>
#include <linux/iommu.h>
#include <net/ip.h>
#include <net/checksum.h>
#include <net/xdp.h>
#include <linux/bpf_trace.h>
#include "net_driver.h"
#include "efx.h"
#include "rx_common.h"
#include "filter.h"
#include "nic.h"
#include "selftest.h"
#include "workarounds.h"

/* Preferred number of descriptors to fill at once */
#define EFX_RX_PREFERRED_BATCH 8U

/* Maximum rx prefix used by any architecture. */
#define EFX_MAX_RX_PREFIX_SIZE 16

/* Size of buffer allocated for skb header area. */
#define EFX_SKB_HEADERS  128u

/* Each packet can consume up to ceil(max_frame_len / buffer_size) buffers */
#define EFX_RX_MAX_FRAGS DIV_ROUND_UP(EFX_MAX_FRAME_LEN(EFX_MAX_MTU), \
				      EFX_RX_USR_BUF_SIZE)

static void efx_rx_packet__check_len(struct efx_rx_queue *rx_queue,
				     struct efx_rx_buffer *rx_buf,
				     int len)
{
	struct efx_nic *efx = rx_queue->efx;
	unsigned max_len = rx_buf->len - efx->type->rx_buffer_padding;

	if (likely(len <= max_len))
		return;

	/* The packet must be discarded, but this is only a fatal error
	 * if the caller indicated it was
	 */
	rx_buf->flags |= EFX_RX_PKT_DISCARD;

	if (net_ratelimit())
		netif_err(efx, rx_err, efx->net_dev,
			  "RX queue %d overlength RX event (%#x > %#x)\n",
			  efx_rx_queue_index(rx_queue), len, max_len);

	efx_rx_queue_channel(rx_queue)->n_rx_overlength++;
}

/* Allocate and construct an SKB around page fragments */
static struct sk_buff *efx_rx_mk_skb(struct efx_channel *channel,
				     struct efx_rx_buffer *rx_buf,
				     unsigned int n_frags,
				     u8 *eh, int hdr_len)
{
	struct efx_nic *efx = channel->efx;
	struct sk_buff *skb;

	/* Allocate an SKB to store the headers */
	skb = netdev_alloc_skb(efx->net_dev,
			       efx->rx_ip_align + efx->rx_prefix_size +
			       hdr_len);
	if (unlikely(skb == NULL)) {
		atomic_inc(&efx->n_rx_noskb_drops);
		return NULL;
	}

	EFX_WARN_ON_ONCE_PARANOID(rx_buf->len < hdr_len);

	memcpy(skb->data + efx->rx_ip_align, eh - efx->rx_prefix_size,
	       efx->rx_prefix_size + hdr_len);
	skb_reserve(skb, efx->rx_ip_align + efx->rx_prefix_size);
	__skb_put(skb, hdr_len);

	/* Append the remaining page(s) onto the frag list */
	if (rx_buf->len > hdr_len) {
		rx_buf->page_offset += hdr_len;
		rx_buf->len -= hdr_len;

		for (;;) {
			skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
					rx_buf->page, rx_buf->page_offset,
					rx_buf->len, efx->rx_buffer_truesize);
			rx_buf->page = NULL;

			if (skb_shinfo(skb)->nr_frags == n_frags)
				break;

			rx_buf = efx_rx_buf_next(&channel->rx_queue, rx_buf);
		}
	} else {
		__free_pages(rx_buf->page, efx->rx_buffer_order);
		rx_buf->page = NULL;
		n_frags = 0;
	}

	/* Move past the ethernet header */
	skb->protocol = eth_type_trans(skb, efx->net_dev);

	skb_mark_napi_id(skb, &channel->napi_str);

	return skb;
}

void efx_rx_packet(struct efx_rx_queue *rx_queue, unsigned int index,
		   unsigned int n_frags, unsigned int len, u16 flags)
{
	struct efx_nic *efx = rx_queue->efx;
	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
	struct efx_rx_buffer *rx_buf;

	rx_queue->rx_packets++;

	rx_buf = efx_rx_buffer(rx_queue, index);
	rx_buf->flags |= flags;

	/* Validate the number of fragments and completed length */
	if (n_frags == 1) {
		if (!(flags & EFX_RX_PKT_PREFIX_LEN))
			efx_rx_packet__check_len(rx_queue, rx_buf, len);
	} else if (unlikely(n_frags > EFX_RX_MAX_FRAGS) ||
		   unlikely(len <= (n_frags - 1) * efx->rx_dma_len) ||
		   unlikely(len > n_frags * efx->rx_dma_len) ||
		   unlikely(!efx->rx_scatter)) {
		/* If this isn't an explicit discard request, either
		 * the hardware or the driver is broken.
		 */
		WARN_ON(!(len == 0 && rx_buf->flags & EFX_RX_PKT_DISCARD));
		rx_buf->flags |= EFX_RX_PKT_DISCARD;
	}

	netif_vdbg(efx, rx_status, efx->net_dev,
		   "RX queue %d received ids %x-%x len %d %s%s\n",
		   efx_rx_queue_index(rx_queue), index,
		   (index + n_frags - 1) & rx_queue->ptr_mask, len,
		   (rx_buf->flags & EFX_RX_PKT_CSUMMED) ? " [SUMMED]" : "",
		   (rx_buf->flags & EFX_RX_PKT_DISCARD) ? " [DISCARD]" : "");

	/* Discard packet, if instructed to do so.  Process the
	 * previous receive first.
	 */
	if (unlikely(rx_buf->flags & EFX_RX_PKT_DISCARD)) {
		efx_rx_flush_packet(channel);
		efx_discard_rx_packet(channel, rx_buf, n_frags);
		return;
	}

	if (n_frags == 1 && !(flags & EFX_RX_PKT_PREFIX_LEN))
		rx_buf->len = len;

	/* Release and/or sync the DMA mapping - assumes all RX buffers
	 * consumed in-order per RX queue.
	 */
	efx_sync_rx_buffer(efx, rx_buf, rx_buf->len);

	/* Prefetch nice and early so data will (hopefully) be in cache by
	 * the time we look at it.
	 */
	prefetch(efx_rx_buf_va(rx_buf));

	rx_buf->page_offset += efx->rx_prefix_size;
	rx_buf->len -= efx->rx_prefix_size;

	if (n_frags > 1) {
		/* Release/sync DMA mapping for additional fragments.
		 * Fix length for last fragment.
		 */
		unsigned int tail_frags = n_frags - 1;

		for (;;) {
			rx_buf = efx_rx_buf_next(rx_queue, rx_buf);
			if (--tail_frags == 0)
				break;
			efx_sync_rx_buffer(efx, rx_buf, efx->rx_dma_len);
		}
		rx_buf->len = len - (n_frags - 1) * efx->rx_dma_len;
		efx_sync_rx_buffer(efx, rx_buf, rx_buf->len);
	}

	/* All fragments have been DMA-synced, so recycle pages. */
	rx_buf = efx_rx_buffer(rx_queue, index);
	efx_recycle_rx_pages(channel, rx_buf, n_frags);

	/* Pipeline receives so that we give time for packet headers to be
	 * prefetched into cache.
	 */
	efx_rx_flush_packet(channel);
	channel->rx_pkt_n_frags = n_frags;
	channel->rx_pkt_index = index;
}

static void efx_rx_deliver(struct efx_channel *channel, u8 *eh,
			   struct efx_rx_buffer *rx_buf,
			   unsigned int n_frags)
{
	struct sk_buff *skb;
	u16 hdr_len = min_t(u16, rx_buf->len, EFX_SKB_HEADERS);

	skb = efx_rx_mk_skb(channel, rx_buf, n_frags, eh, hdr_len);
	if (unlikely(skb == NULL)) {
		struct efx_rx_queue *rx_queue;

		rx_queue = efx_channel_get_rx_queue(channel);
		efx_free_rx_buffers(rx_queue, rx_buf, n_frags);
		return;
	}
	skb_record_rx_queue(skb, channel->rx_queue.core_index);

	/* Set the SKB flags */
	skb_checksum_none_assert(skb);
	if (likely(rx_buf->flags & EFX_RX_PKT_CSUMMED)) {
		skb->ip_summed = CHECKSUM_UNNECESSARY;
		skb->csum_level = !!(rx_buf->flags & EFX_RX_PKT_CSUM_LEVEL);
	}

	efx_rx_skb_attach_timestamp(channel, skb);

	if (channel->type->receive_skb)
		if (channel->type->receive_skb(channel, skb))
			return;

	/* Pass the packet up */
	if (channel->rx_list != NULL)
		/* Add to list, will pass up later */
		list_add_tail(&skb->list, channel->rx_list);
	else
		/* No list, so pass it up now */
		netif_receive_skb(skb);
}

/** efx_do_xdp: perform XDP processing on a received packet
 *
 * Returns true if packet should still be delivered.
 */
static bool efx_do_xdp(struct efx_nic *efx, struct efx_channel *channel,
		       struct efx_rx_buffer *rx_buf, u8 **ehp)
{
	u8 rx_prefix[EFX_MAX_RX_PREFIX_SIZE];
	struct efx_rx_queue *rx_queue;
	struct bpf_prog *xdp_prog;
	struct xdp_frame *xdpf;
	struct xdp_buff xdp;
	u32 xdp_act;
	s16 offset;
	int err;

	xdp_prog = rcu_dereference_bh(efx->xdp_prog);
	if (!xdp_prog)
		return true;

	rx_queue = efx_channel_get_rx_queue(channel);

	if (unlikely(channel->rx_pkt_n_frags > 1)) {
		/* We can't do XDP on fragmented packets - drop. */
		efx_free_rx_buffers(rx_queue, rx_buf,
				    channel->rx_pkt_n_frags);
		if (net_ratelimit())
			netif_err(efx, rx_err, efx->net_dev,
				  "XDP is not possible with multiple receive fragments (%d)\n",
				  channel->rx_pkt_n_frags);
		channel->n_rx_xdp_bad_drops++;
		return false;
	}

	dma_sync_single_for_cpu(&efx->pci_dev->dev, rx_buf->dma_addr,
				rx_buf->len, DMA_FROM_DEVICE);

	/* Save the rx prefix. */
	EFX_WARN_ON_PARANOID(efx->rx_prefix_size > EFX_MAX_RX_PREFIX_SIZE);
	memcpy(rx_prefix, *ehp - efx->rx_prefix_size,
	       efx->rx_prefix_size);

	xdp_init_buff(&xdp, efx->rx_page_buf_step, &rx_queue->xdp_rxq_info);
	/* No support yet for XDP metadata */
	xdp_prepare_buff(&xdp, *ehp - EFX_XDP_HEADROOM, EFX_XDP_HEADROOM,
			 rx_buf->len, false);

	xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp);

	offset = (u8 *)xdp.data - *ehp;

	switch (xdp_act) {
	case XDP_PASS:
		/* Fix up rx prefix. */
		if (offset) {
			*ehp += offset;
			rx_buf->page_offset += offset;
			rx_buf->len -= offset;
			memcpy(*ehp - efx->rx_prefix_size, rx_prefix,
			       efx->rx_prefix_size);
		}
		break;

	case XDP_TX:
		/* Buffer ownership passes to tx on success. */
		xdpf = xdp_convert_buff_to_frame(&xdp);
		err = efx_xdp_tx_buffers(efx, 1, &xdpf, true);
		if (unlikely(err != 1)) {
			efx_free_rx_buffers(rx_queue, rx_buf, 1);
			if (net_ratelimit())
				netif_err(efx, rx_err, efx->net_dev,
					  "XDP TX failed (%d)\n", err);
			channel->n_rx_xdp_bad_drops++;
			trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
		} else {
			channel->n_rx_xdp_tx++;
		}
		break;

	case XDP_REDIRECT:
		err = xdp_do_redirect(efx->net_dev, &xdp, xdp_prog);
		if (unlikely(err)) {
			efx_free_rx_buffers(rx_queue, rx_buf, 1);
			if (net_ratelimit())
				netif_err(efx, rx_err, efx->net_dev,
					  "XDP redirect failed (%d)\n", err);
			channel->n_rx_xdp_bad_drops++;
			trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
		} else {
			channel->n_rx_xdp_redirect++;
		}
		break;

	default:
		bpf_warn_invalid_xdp_action(efx->net_dev, xdp_prog, xdp_act);
		efx_free_rx_buffers(rx_queue, rx_buf, 1);
		channel->n_rx_xdp_bad_drops++;
		trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
		break;

	case XDP_ABORTED:
		trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
		fallthrough;
	case XDP_DROP:
		efx_free_rx_buffers(rx_queue, rx_buf, 1);
		channel->n_rx_xdp_drops++;
		break;
	}

	return xdp_act == XDP_PASS;
}

/* Handle a received packet.  Second half: Touches packet payload. */
void __efx_rx_packet(struct efx_channel *channel)
{
	struct efx_nic *efx = channel->efx;
	struct efx_rx_buffer *rx_buf =
		efx_rx_buffer(&channel->rx_queue, channel->rx_pkt_index);
	u8 *eh = efx_rx_buf_va(rx_buf);

	/* Read length from the prefix if necessary.  This already
	 * excludes the length of the prefix itself.
	 */
	if (rx_buf->flags & EFX_RX_PKT_PREFIX_LEN)
		rx_buf->len = le16_to_cpup((__le16 *)
					   (eh + efx->rx_packet_len_offset));

	/* If we're in loopback test, then pass the packet directly to the
	 * loopback layer, and free the rx_buf here
	 */
	if (unlikely(efx->loopback_selftest)) {
		struct efx_rx_queue *rx_queue;

		efx_loopback_rx_packet(efx, eh, rx_buf->len);
		rx_queue = efx_channel_get_rx_queue(channel);
		efx_free_rx_buffers(rx_queue, rx_buf,
				    channel->rx_pkt_n_frags);
		goto out;
	}

	if (!efx_do_xdp(efx, channel, rx_buf, &eh))
		goto out;

	if (unlikely(!(efx->net_dev->features & NETIF_F_RXCSUM)))
		rx_buf->flags &= ~EFX_RX_PKT_CSUMMED;

	if ((rx_buf->flags & EFX_RX_PKT_TCP) && !channel->type->receive_skb)
		efx_rx_packet_gro(channel, rx_buf, channel->rx_pkt_n_frags, eh, 0);
	else
		efx_rx_deliver(channel, eh, rx_buf, channel->rx_pkt_n_frags);
out:
	channel->rx_pkt_n_frags = 0;
}
Commit	Line	Data
d2912cb1	1	// SPDX-License-Identifier: GPL-2.0-only
8ceee660	2	/****************************************************************************
f7a6d2c4	3	* Driver for Solarflare network controllers and boards
8ceee660	4	* Copyright 2005-2006 Fen Systems Ltd.
f7a6d2c4	5	* Copyright 2005-2013 Solarflare Communications Inc.
8ceee660 BH	6	*/
	7
	8	#include <linux/socket.h>
	9	#include <linux/in.h>
5a0e3ad6	10	#include <linux/slab.h>
8ceee660	11	#include <linux/ip.h>
c47b2d9d	12	#include <linux/ipv6.h>
8ceee660 BH	13	#include <linux/tcp.h>
8ceee660 BH	14	#include <linux/udp.h>
70c71606	15	#include <linux/prefetch.h>
6eb07caf	16	#include <linux/moduleparam.h>
2768935a	17	#include <linux/iommu.h>
8ceee660 BH	18	#include <net/ip.h>
8ceee660 BH	19	#include <net/checksum.h>
eb9a36be CM	20	#include <net/xdp.h>
eb9a36be CM	21	#include <linux/bpf_trace.h>
8ceee660	22	#include "net_driver.h"
8ceee660	23	#include "efx.h"
e1253f39	24	#include "rx_common.h"
add72477	25	#include "filter.h"
744093c9	26	#include "nic.h"
3273c2e8	27	#include "selftest.h"
8ceee660 BH	28	#include "workarounds.h"
8ceee660 BH	29
1648a23f DP	30	/* Preferred number of descriptors to fill at once */
1648a23f DP	31	#define EFX_RX_PREFERRED_BATCH 8U
8ceee660	32
eb9a36be CM	33	/* Maximum rx prefix used by any architecture. */
	34	#define EFX_MAX_RX_PREFIX_SIZE 16
	35
8ceee660	36	/* Size of buffer allocated for skb header area. */
d4ef5b6f	37	#define EFX_SKB_HEADERS 128u
8ceee660	38
85740cdf BH	39	/* Each packet can consume up to ceil(max_frame_len / buffer_size) buffers */
	40	#define EFX_RX_MAX_FRAGS DIV_ROUND_UP(EFX_MAX_FRAME_LEN(EFX_MAX_MTU), \
	41	EFX_RX_USR_BUF_SIZE)
	42
4d566063 BH	43	static void efx_rx_packet__check_len(struct efx_rx_queue *rx_queue,
4d566063 BH	44	struct efx_rx_buffer *rx_buf,
97d48a10	45	int len)
8ceee660 BH	46	{
	47	struct efx_nic *efx = rx_queue->efx;
	48	unsigned max_len = rx_buf->len - efx->type->rx_buffer_padding;
	49
	50	if (likely(len <= max_len))
	51	return;
	52
	53	/* The packet must be discarded, but this is only a fatal error
	54	* if the caller indicated it was
	55	*/
db339569	56	rx_buf->flags \|= EFX_RX_PKT_DISCARD;
8ceee660	57
5a6681e2 EC	58	if (net_ratelimit())
	59	netif_err(efx, rx_err, efx->net_dev,
	60	"RX queue %d overlength RX event (%#x > %#x)\n",
	61	efx_rx_queue_index(rx_queue), len, max_len);
8ceee660	62
ba1e8a35	63	efx_rx_queue_channel(rx_queue)->n_rx_overlength++;
8ceee660 BH	64	}
8ceee660 BH	65
85740cdf	66	/* Allocate and construct an SKB around page fragments */
97d48a10 AR	67	static struct sk_buff efx_rx_mk_skb(struct efx_channel channel,
97d48a10 AR	68	struct efx_rx_buffer *rx_buf,
85740cdf	69	unsigned int n_frags,
97d48a10 AR	70	u8 *eh, int hdr_len)
	71	{
	72	struct efx_nic *efx = channel->efx;
	73	struct sk_buff *skb;
18e1d2be	74
97d48a10	75	/* Allocate an SKB to store the headers */
2ccd0b19 BH	76	skb = netdev_alloc_skb(efx->net_dev,
	77	efx->rx_ip_align + efx->rx_prefix_size +
	78	hdr_len);
e4d112e4 EC	79	if (unlikely(skb == NULL)) {
e4d112e4 EC	80	atomic_inc(&efx->n_rx_noskb_drops);
97d48a10	81	return NULL;
e4d112e4	82	}
97d48a10	83
e01b16a7	84	EFX_WARN_ON_ONCE_PARANOID(rx_buf->len < hdr_len);
97d48a10	85
2ccd0b19 BH	86	memcpy(skb->data + efx->rx_ip_align, eh - efx->rx_prefix_size,
	87	efx->rx_prefix_size + hdr_len);
	88	skb_reserve(skb, efx->rx_ip_align + efx->rx_prefix_size);
	89	__skb_put(skb, hdr_len);
97d48a10	90
85740cdf	91	/* Append the remaining page(s) onto the frag list */
97d48a10	92	if (rx_buf->len > hdr_len) {
85740cdf BH	93	rx_buf->page_offset += hdr_len;
	94	rx_buf->len -= hdr_len;
	95
	96	for (;;) {
c438a801 CJ	97	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
	98	rx_buf->page, rx_buf->page_offset,
	99	rx_buf->len, efx->rx_buffer_truesize);
85740cdf	100	rx_buf->page = NULL;
c438a801	101
85740cdf BH	102	if (skb_shinfo(skb)->nr_frags == n_frags)
	103	break;
	104
	105	rx_buf = efx_rx_buf_next(&channel->rx_queue, rx_buf);
	106	}
97d48a10 AR	107	} else {
97d48a10 AR	108	__free_pages(rx_buf->page, efx->rx_buffer_order);
85740cdf BH	109	rx_buf->page = NULL;
85740cdf BH	110	n_frags = 0;
18e1d2be	111	}
97d48a10	112
97d48a10 AR	113	/* Move past the ethernet header */
	114	skb->protocol = eth_type_trans(skb, efx->net_dev);
	115
36763266 AR	116	skb_mark_napi_id(skb, &channel->napi_str);
36763266 AR	117
97d48a10	118	return skb;
8ceee660 BH	119	}
8ceee660 BH	120
8ceee660	121	void efx_rx_packet(struct efx_rx_queue *rx_queue, unsigned int index,
85740cdf	122	unsigned int n_frags, unsigned int len, u16 flags)
8ceee660 BH	123	{
8ceee660 BH	124	struct efx_nic *efx = rx_queue->efx;
ba1e8a35	125	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
8ceee660	126	struct efx_rx_buffer *rx_buf;
8ceee660	127
8ccf3800 AR	128	rx_queue->rx_packets++;
8ccf3800 AR	129
8ceee660	130	rx_buf = efx_rx_buffer(rx_queue, index);
179ea7f0	131	rx_buf->flags \|= flags;
8ceee660	132
85740cdf BH	133	/* Validate the number of fragments and completed length */
85740cdf BH	134	if (n_frags == 1) {
3dced740 BH	135	if (!(flags & EFX_RX_PKT_PREFIX_LEN))
3dced740 BH	136	efx_rx_packet__check_len(rx_queue, rx_buf, len);
85740cdf	137	} else if (unlikely(n_frags > EFX_RX_MAX_FRAGS) \|\|
e8c68c0a JC	138	unlikely(len <= (n_frags - 1) * efx->rx_dma_len) \|\|
e8c68c0a JC	139	unlikely(len > n_frags * efx->rx_dma_len) \|\|
85740cdf BH	140	unlikely(!efx->rx_scatter)) {
	141	/* If this isn't an explicit discard request, either
	142	* the hardware or the driver is broken.
	143	*/
	144	WARN_ON(!(len == 0 && rx_buf->flags & EFX_RX_PKT_DISCARD));
	145	rx_buf->flags \|= EFX_RX_PKT_DISCARD;
	146	}
8ceee660	147
62776d03	148	netif_vdbg(efx, rx_status, efx->net_dev,
85740cdf	149	"RX queue %d received ids %x-%x len %d %s%s\n",
ba1e8a35	150	efx_rx_queue_index(rx_queue), index,
85740cdf	151	(index + n_frags - 1) & rx_queue->ptr_mask, len,
db339569 BH	152	(rx_buf->flags & EFX_RX_PKT_CSUMMED) ? " [SUMMED]" : "",
db339569 BH	153	(rx_buf->flags & EFX_RX_PKT_DISCARD) ? " [DISCARD]" : "");
8ceee660	154
85740cdf BH	155	/* Discard packet, if instructed to do so. Process the
	156	* previous receive first.
	157	*/
db339569	158	if (unlikely(rx_buf->flags & EFX_RX_PKT_DISCARD)) {
85740cdf	159	efx_rx_flush_packet(channel);
734d4e15	160	efx_discard_rx_packet(channel, rx_buf, n_frags);
85740cdf	161	return;
8ceee660 BH	162	}
8ceee660 BH	163
3dced740	164	if (n_frags == 1 && !(flags & EFX_RX_PKT_PREFIX_LEN))
85740cdf BH	165	rx_buf->len = len;
85740cdf BH	166
2768935a DP	167	/* Release and/or sync the DMA mapping - assumes all RX buffers
2768935a DP	168	* consumed in-order per RX queue.
8ceee660	169	*/
2768935a	170	efx_sync_rx_buffer(efx, rx_buf, rx_buf->len);
8ceee660 BH	171
	172	/* Prefetch nice and early so data will (hopefully) be in cache by
	173	* the time we look at it.
	174	*/
5036b7c7	175	prefetch(efx_rx_buf_va(rx_buf));
8ceee660	176
43a3739d JC	177	rx_buf->page_offset += efx->rx_prefix_size;
43a3739d JC	178	rx_buf->len -= efx->rx_prefix_size;
85740cdf BH	179
	180	if (n_frags > 1) {
	181	/* Release/sync DMA mapping for additional fragments.
	182	* Fix length for last fragment.
	183	*/
	184	unsigned int tail_frags = n_frags - 1;
	185
	186	for (;;) {
	187	rx_buf = efx_rx_buf_next(rx_queue, rx_buf);
	188	if (--tail_frags == 0)
	189	break;
e8c68c0a	190	efx_sync_rx_buffer(efx, rx_buf, efx->rx_dma_len);
85740cdf	191	}
e8c68c0a	192	rx_buf->len = len - (n_frags - 1) * efx->rx_dma_len;
2768935a	193	efx_sync_rx_buffer(efx, rx_buf, rx_buf->len);
85740cdf	194	}
b74e3e8c	195
734d4e15	196	/* All fragments have been DMA-synced, so recycle pages. */
2768935a	197	rx_buf = efx_rx_buffer(rx_queue, index);
734d4e15	198	efx_recycle_rx_pages(channel, rx_buf, n_frags);
2768935a	199
8ceee660 BH	200	/* Pipeline receives so that we give time for packet headers to be
	201	* prefetched into cache.
	202	*/
ff734ef4	203	efx_rx_flush_packet(channel);
85740cdf BH	204	channel->rx_pkt_n_frags = n_frags;
85740cdf BH	205	channel->rx_pkt_index = index;
8ceee660 BH	206	}
8ceee660 BH	207
97d48a10	208	static void efx_rx_deliver(struct efx_channel channel, u8 eh,
85740cdf BH	209	struct efx_rx_buffer *rx_buf,
85740cdf BH	210	unsigned int n_frags)
1ddceb4c BH	211	{
1ddceb4c BH	212	struct sk_buff *skb;
97d48a10	213	u16 hdr_len = min_t(u16, rx_buf->len, EFX_SKB_HEADERS);
1ddceb4c	214
85740cdf	215	skb = efx_rx_mk_skb(channel, rx_buf, n_frags, eh, hdr_len);
97d48a10	216	if (unlikely(skb == NULL)) {
9eb0a5d1 DP	217	struct efx_rx_queue *rx_queue;
	218
	219	rx_queue = efx_channel_get_rx_queue(channel);
	220	efx_free_rx_buffers(rx_queue, rx_buf, n_frags);
97d48a10 AR	221	return;
	222	}
	223	skb_record_rx_queue(skb, channel->rx_queue.core_index);
1ddceb4c BH	224
	225	/* Set the SKB flags */
	226	skb_checksum_none_assert(skb);
da50ae2e	227	if (likely(rx_buf->flags & EFX_RX_PKT_CSUMMED)) {
c99dffc4	228	skb->ip_summed = CHECKSUM_UNNECESSARY;
da50ae2e JC	229	skb->csum_level = !!(rx_buf->flags & EFX_RX_PKT_CSUM_LEVEL);
da50ae2e JC	230	}
1ddceb4c	231
bd9a265d JC	232	efx_rx_skb_attach_timestamp(channel, skb);
bd9a265d JC	233
c31e5f9f	234	if (channel->type->receive_skb)
4a74dc65	235	if (channel->type->receive_skb(channel, skb))
97d48a10	236	return;
4a74dc65 BH	237
4a74dc65 BH	238	/* Pass the packet up */
e090bfb9 EC	239	if (channel->rx_list != NULL)
	240	/* Add to list, will pass up later */
	241	list_add_tail(&skb->list, channel->rx_list);
	242	else
	243	/* No list, so pass it up now */
	244	netif_receive_skb(skb);
1ddceb4c BH	245	}
1ddceb4c BH	246
eb9a36be CM	247	/** efx_do_xdp: perform XDP processing on a received packet
	248	*
	249	* Returns true if packet should still be delivered.
	250	*/
	251	static bool efx_do_xdp(struct efx_nic efx, struct efx_channel channel,
	252	struct efx_rx_buffer rx_buf, u8 *ehp)
	253	{
	254	u8 rx_prefix[EFX_MAX_RX_PREFIX_SIZE];
	255	struct efx_rx_queue *rx_queue;
	256	struct bpf_prog *xdp_prog;
dfe44c1f	257	struct xdp_frame *xdpf;
eb9a36be CM	258	struct xdp_buff xdp;
	259	u32 xdp_act;
	260	s16 offset;
	261	int err;
	262
4eb14e3f THJ	263	xdp_prog = rcu_dereference_bh(efx->xdp_prog);
4eb14e3f THJ	264	if (!xdp_prog)
eb9a36be	265	return true;
eb9a36be CM	266
	267	rx_queue = efx_channel_get_rx_queue(channel);
	268
	269	if (unlikely(channel->rx_pkt_n_frags > 1)) {
	270	/* We can't do XDP on fragmented packets - drop. */
eb9a36be CM	271	efx_free_rx_buffers(rx_queue, rx_buf,
	272	channel->rx_pkt_n_frags);
	273	if (net_ratelimit())
	274	netif_err(efx, rx_err, efx->net_dev,
	275	"XDP is not possible with multiple receive fragments (%d)\n",
	276	channel->rx_pkt_n_frags);
cd846bef	277	channel->n_rx_xdp_bad_drops++;
eb9a36be CM	278	return false;
	279	}
	280
	281	dma_sync_single_for_cpu(&efx->pci_dev->dev, rx_buf->dma_addr,
	282	rx_buf->len, DMA_FROM_DEVICE);
	283
	284	/* Save the rx prefix. */
	285	EFX_WARN_ON_PARANOID(efx->rx_prefix_size > EFX_MAX_RX_PREFIX_SIZE);
	286	memcpy(rx_prefix, *ehp - efx->rx_prefix_size,
	287	efx->rx_prefix_size);
	288
43b5169d	289	xdp_init_buff(&xdp, efx->rx_page_buf_step, &rx_queue->xdp_rxq_info);
eb9a36be	290	/* No support yet for XDP metadata */
be9df4af LB	291	xdp_prepare_buff(&xdp, *ehp - EFX_XDP_HEADROOM, EFX_XDP_HEADROOM,
be9df4af LB	292	rx_buf->len, false);
eb9a36be CM	293
eb9a36be CM	294	xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp);
eb9a36be CM	295
	296	offset = (u8 )xdp.data - ehp;
	297
	298	switch (xdp_act) {
	299	case XDP_PASS:
	300	/* Fix up rx prefix. */
	301	if (offset) {
	302	*ehp += offset;
	303	rx_buf->page_offset += offset;
	304	rx_buf->len -= offset;
	305	memcpy(*ehp - efx->rx_prefix_size, rx_prefix,
	306	efx->rx_prefix_size);
	307	}
	308	break;
	309
	310	case XDP_TX:
dfe44c1f	311	/* Buffer ownership passes to tx on success. */
1b698fa5	312	xdpf = xdp_convert_buff_to_frame(&xdp);
dfe44c1f CM	313	err = efx_xdp_tx_buffers(efx, 1, &xdpf, true);
	314	if (unlikely(err != 1)) {
	315	efx_free_rx_buffers(rx_queue, rx_buf, 1);
	316	if (net_ratelimit())
	317	netif_err(efx, rx_err, efx->net_dev,
	318	"XDP TX failed (%d)\n", err);
cd846bef	319	channel->n_rx_xdp_bad_drops++;
9440a875	320	trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
cd846bef CM	321	} else {
cd846bef CM	322	channel->n_rx_xdp_tx++;
dfe44c1f CM	323	}
dfe44c1f CM	324	break;
eb9a36be CM	325
	326	case XDP_REDIRECT:
	327	err = xdp_do_redirect(efx->net_dev, &xdp, xdp_prog);
	328	if (unlikely(err)) {
	329	efx_free_rx_buffers(rx_queue, rx_buf, 1);
	330	if (net_ratelimit())
	331	netif_err(efx, rx_err, efx->net_dev,
	332	"XDP redirect failed (%d)\n", err);
cd846bef	333	channel->n_rx_xdp_bad_drops++;
9440a875	334	trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
cd846bef CM	335	} else {
cd846bef CM	336	channel->n_rx_xdp_redirect++;
eb9a36be CM	337	}
	338	break;
	339
	340	default:
c8064e5b	341	bpf_warn_invalid_xdp_action(efx->net_dev, xdp_prog, xdp_act);
eb9a36be	342	efx_free_rx_buffers(rx_queue, rx_buf, 1);
cd846bef	343	channel->n_rx_xdp_bad_drops++;
9440a875	344	trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
eb9a36be CM	345	break;
	346
	347	case XDP_ABORTED:
	348	trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
df561f66	349	fallthrough;
eb9a36be CM	350	case XDP_DROP:
eb9a36be CM	351	efx_free_rx_buffers(rx_queue, rx_buf, 1);
cd846bef	352	channel->n_rx_xdp_drops++;
eb9a36be CM	353	break;
	354	}
	355
	356	return xdp_act == XDP_PASS;
	357	}
	358
8ceee660	359	/* Handle a received packet. Second half: Touches packet payload. */
85740cdf	360	void __efx_rx_packet(struct efx_channel *channel)
8ceee660 BH	361	{
8ceee660 BH	362	struct efx_nic *efx = channel->efx;
85740cdf BH	363	struct efx_rx_buffer *rx_buf =
85740cdf BH	364	efx_rx_buffer(&channel->rx_queue, channel->rx_pkt_index);
b74e3e8c	365	u8 *eh = efx_rx_buf_va(rx_buf);
604f6049	366
3dced740 BH	367	/* Read length from the prefix if necessary. This already
	368	* excludes the length of the prefix itself.
	369	*/
	370	if (rx_buf->flags & EFX_RX_PKT_PREFIX_LEN)
	371	rx_buf->len = le16_to_cpup((__le16 *)
	372	(eh + efx->rx_packet_len_offset));
	373
3273c2e8 BH	374	/* If we're in loopback test, then pass the packet directly to the
	375	* loopback layer, and free the rx_buf here
	376	*/
	377	if (unlikely(efx->loopback_selftest)) {
9eb0a5d1 DP	378	struct efx_rx_queue *rx_queue;
9eb0a5d1 DP	379
a526f140	380	efx_loopback_rx_packet(efx, eh, rx_buf->len);
9eb0a5d1 DP	381	rx_queue = efx_channel_get_rx_queue(channel);
	382	efx_free_rx_buffers(rx_queue, rx_buf,
	383	channel->rx_pkt_n_frags);
85740cdf	384	goto out;
3273c2e8 BH	385	}
3273c2e8 BH	386
eb9a36be CM	387	if (!efx_do_xdp(efx, channel, rx_buf, &eh))
	388	goto out;
	389
abfe9039	390	if (unlikely(!(efx->net_dev->features & NETIF_F_RXCSUM)))
db339569	391	rx_buf->flags &= ~EFX_RX_PKT_CSUMMED;
ab3cf6d0	392
e7fe9491	393	if ((rx_buf->flags & EFX_RX_PKT_TCP) && !channel->type->receive_skb)
4d9c0a2d	394	efx_rx_packet_gro(channel, rx_buf, channel->rx_pkt_n_frags, eh, 0);
1ddceb4c	395	else
85740cdf BH	396	efx_rx_deliver(channel, eh, rx_buf, channel->rx_pkt_n_frags);
	397	out:
	398	channel->rx_pkt_n_frags = 0;
8ceee660	399	}