Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit fa37a958 authored by Yevgeny Petrilin's avatar Yevgeny Petrilin Committed by David S. Miller
Browse files

mlx4_en: Moving to work with GRO

parent 0533943c
Loading
Loading
Loading
Loading
+2 −54
Original line number Original line Diff line number Diff line
@@ -39,21 +39,6 @@
#include "en_port.h"
#include "en_port.h"




static void mlx4_en_update_lro_stats(struct mlx4_en_priv *priv)
{
	int i;

	priv->port_stats.lro_aggregated = 0;
	priv->port_stats.lro_flushed = 0;
	priv->port_stats.lro_no_desc = 0;

	for (i = 0; i < priv->rx_ring_num; i++) {
		priv->port_stats.lro_aggregated += priv->rx_ring[i].lro.stats.aggregated;
		priv->port_stats.lro_flushed += priv->rx_ring[i].lro.stats.flushed;
		priv->port_stats.lro_no_desc += priv->rx_ring[i].lro.stats.no_desc;
	}
}

static void
static void
mlx4_en_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *drvinfo)
mlx4_en_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *drvinfo)
{
{
@@ -112,7 +97,7 @@ static const char main_strings[][ETH_GSTRING_LEN] = {
	"tx_heartbeat_errors", "tx_window_errors",
	"tx_heartbeat_errors", "tx_window_errors",


	/* port statistics */
	/* port statistics */
	"lro_aggregated", "lro_flushed", "lro_no_desc", "tso_packets",
	"tso_packets",
	"queue_stopped", "wake_queue", "tx_timeout", "rx_alloc_failed",
	"queue_stopped", "wake_queue", "tx_timeout", "rx_alloc_failed",
	"rx_csum_good", "rx_csum_none", "tx_chksum_offload",
	"rx_csum_good", "rx_csum_none", "tx_chksum_offload",


@@ -122,7 +107,7 @@ static const char main_strings[][ETH_GSTRING_LEN] = {
	"tx_prio_1", "tx_prio_2", "tx_prio_3", "tx_prio_4", "tx_prio_5",
	"tx_prio_1", "tx_prio_2", "tx_prio_3", "tx_prio_4", "tx_prio_5",
	"tx_prio_6", "tx_prio_7",
	"tx_prio_6", "tx_prio_7",
};
};
#define NUM_MAIN_STATS	21
#define NUM_MAIN_STATS	18
#define NUM_ALL_STATS	(NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PKT_STATS + NUM_PERF_STATS)
#define NUM_ALL_STATS	(NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PKT_STATS + NUM_PERF_STATS)


static const char mlx4_en_test_names[][ETH_GSTRING_LEN]= {
static const char mlx4_en_test_names[][ETH_GSTRING_LEN]= {
@@ -174,8 +159,6 @@ static void mlx4_en_get_ethtool_stats(struct net_device *dev,


	spin_lock_bh(&priv->stats_lock);
	spin_lock_bh(&priv->stats_lock);


	mlx4_en_update_lro_stats(priv);

	for (i = 0; i < NUM_MAIN_STATS; i++)
	for (i = 0; i < NUM_MAIN_STATS; i++)
		data[index++] = ((unsigned long *) &priv->stats)[i];
		data[index++] = ((unsigned long *) &priv->stats)[i];
	for (i = 0; i < NUM_PORT_STATS; i++)
	for (i = 0; i < NUM_PORT_STATS; i++)
@@ -439,40 +422,6 @@ static void mlx4_en_get_ringparam(struct net_device *dev,
	param->tx_pending = priv->tx_ring[0].size;
	param->tx_pending = priv->tx_ring[0].size;
}
}


static int mlx4_ethtool_op_set_flags(struct net_device *dev, u32 data)
{
	struct mlx4_en_priv *priv = netdev_priv(dev);
	struct mlx4_en_dev *mdev = priv->mdev;
	int rc = 0;
	int changed = 0;

	if (data & ~ETH_FLAG_LRO)
		return -EOPNOTSUPP;

	if (data & ETH_FLAG_LRO) {
		if (!(dev->features & NETIF_F_LRO))
			changed = 1;
	} else if (dev->features & NETIF_F_LRO) {
		changed = 1;
	}

	if (changed) {
		if (netif_running(dev)) {
			mutex_lock(&mdev->state_lock);
			mlx4_en_stop_port(dev);
		}
		dev->features ^= NETIF_F_LRO;
		if (netif_running(dev)) {
			rc = mlx4_en_start_port(dev);
			if (rc)
				en_err(priv, "Failed to restart port\n");
			mutex_unlock(&mdev->state_lock);
		}
	}

	return rc;
}

const struct ethtool_ops mlx4_en_ethtool_ops = {
const struct ethtool_ops mlx4_en_ethtool_ops = {
	.get_drvinfo = mlx4_en_get_drvinfo,
	.get_drvinfo = mlx4_en_get_drvinfo,
	.get_settings = mlx4_en_get_settings,
	.get_settings = mlx4_en_get_settings,
@@ -502,7 +451,6 @@ const struct ethtool_ops mlx4_en_ethtool_ops = {
	.get_ringparam = mlx4_en_get_ringparam,
	.get_ringparam = mlx4_en_get_ringparam,
	.set_ringparam = mlx4_en_set_ringparam,
	.set_ringparam = mlx4_en_set_ringparam,
	.get_flags = ethtool_op_get_flags,
	.get_flags = ethtool_op_get_flags,
	.set_flags = mlx4_ethtool_op_set_flags,
};
};




+1 −1
Original line number Original line Diff line number Diff line
@@ -1047,7 +1047,7 @@ int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
	dev->features |= NETIF_F_HW_VLAN_TX |
	dev->features |= NETIF_F_HW_VLAN_TX |
			 NETIF_F_HW_VLAN_RX |
			 NETIF_F_HW_VLAN_RX |
			 NETIF_F_HW_VLAN_FILTER;
			 NETIF_F_HW_VLAN_FILTER;
	dev->features |= NETIF_F_LRO;
	dev->features |= NETIF_F_GRO;
	if (mdev->LSO_support) {
	if (mdev->LSO_support) {
		dev->features |= NETIF_F_TSO;
		dev->features |= NETIF_F_TSO;
		dev->features |= NETIF_F_TSO6;
		dev->features |= NETIF_F_TSO6;
+13 −53
Original line number Original line Diff line number Diff line
@@ -42,18 +42,6 @@
#include "mlx4_en.h"
#include "mlx4_en.h"




static int mlx4_en_get_frag_header(struct skb_frag_struct *frags, void **mac_hdr,
				   void **ip_hdr, void **tcpudp_hdr,
				   u64 *hdr_flags, void *priv)
{
	*mac_hdr = page_address(frags->page) + frags->page_offset;
	*ip_hdr = *mac_hdr + ETH_HLEN;
	*tcpudp_hdr = (struct tcphdr *)(*ip_hdr + sizeof(struct iphdr));
	*hdr_flags = LRO_IPV4 | LRO_TCP;

	return 0;
}

static int mlx4_en_alloc_frag(struct mlx4_en_priv *priv,
static int mlx4_en_alloc_frag(struct mlx4_en_priv *priv,
			      struct mlx4_en_rx_desc *rx_desc,
			      struct mlx4_en_rx_desc *rx_desc,
			      struct skb_frag_struct *skb_frags,
			      struct skb_frag_struct *skb_frags,
@@ -312,24 +300,6 @@ int mlx4_en_create_rx_ring(struct mlx4_en_priv *priv,
	}
	}
	ring->buf = ring->wqres.buf.direct.buf;
	ring->buf = ring->wqres.buf.direct.buf;


	/* Configure lro mngr */
	memset(&ring->lro, 0, sizeof(struct net_lro_mgr));
	ring->lro.dev = priv->dev;
	ring->lro.features = LRO_F_NAPI;
	ring->lro.frag_align_pad = NET_IP_ALIGN;
	ring->lro.ip_summed = CHECKSUM_UNNECESSARY;
	ring->lro.ip_summed_aggr = CHECKSUM_UNNECESSARY;
	ring->lro.max_desc = MLX4_EN_MAX_LRO_DESCRIPTORS;
	ring->lro.max_aggr = MAX_SKB_FRAGS;
	ring->lro.lro_arr = kcalloc(MLX4_EN_MAX_LRO_DESCRIPTORS,
				    sizeof(struct net_lro_desc),
				    GFP_KERNEL);
	if (!ring->lro.lro_arr) {
		en_err(priv, "Failed to allocate lro array\n");
		goto err_map;
	}
	ring->lro.get_frag_header = mlx4_en_get_frag_header;

	return 0;
	return 0;


err_map:
err_map:
@@ -412,7 +382,6 @@ void mlx4_en_destroy_rx_ring(struct mlx4_en_priv *priv,
{
{
	struct mlx4_en_dev *mdev = priv->mdev;
	struct mlx4_en_dev *mdev = priv->mdev;


	kfree(ring->lro.lro_arr);
	mlx4_en_unmap_buffer(&ring->wqres.buf);
	mlx4_en_unmap_buffer(&ring->wqres.buf);
	mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size + TXBB_SIZE);
	mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size + TXBB_SIZE);
	vfree(ring->rx_info);
	vfree(ring->rx_info);
@@ -563,7 +532,6 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
	struct mlx4_cqe *cqe;
	struct mlx4_cqe *cqe;
	struct mlx4_en_rx_ring *ring = &priv->rx_ring[cq->ring];
	struct mlx4_en_rx_ring *ring = &priv->rx_ring[cq->ring];
	struct skb_frag_struct *skb_frags;
	struct skb_frag_struct *skb_frags;
	struct skb_frag_struct lro_frags[MLX4_EN_MAX_RX_FRAGS];
	struct mlx4_en_rx_desc *rx_desc;
	struct mlx4_en_rx_desc *rx_desc;
	struct sk_buff *skb;
	struct sk_buff *skb;
	int index;
	int index;
@@ -623,37 +591,33 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
				 * - TCP/IP (v4)
				 * - TCP/IP (v4)
				 * - without IP options
				 * - without IP options
				 * - not an IP fragment */
				 * - not an IP fragment */
				if (mlx4_en_can_lro(cqe->status) &&
				if (dev->features & NETIF_F_GRO) {
				    dev->features & NETIF_F_LRO) {
					struct sk_buff *gro_skb = napi_get_frags(&cq->napi);


					nr = mlx4_en_complete_rx_desc(
					nr = mlx4_en_complete_rx_desc(
						priv, rx_desc,
						priv, rx_desc,
						skb_frags, lro_frags,
						skb_frags, skb_shinfo(gro_skb)->frags,
						ring->page_alloc, length);
						ring->page_alloc, length);
					if (!nr)
					if (!nr)
						goto next;
						goto next;


					skb_shinfo(gro_skb)->nr_frags = nr;
					gro_skb->len = length;
					gro_skb->data_len = length;
					gro_skb->truesize += length;
					gro_skb->ip_summed = CHECKSUM_UNNECESSARY;

					if (priv->vlgrp && (cqe->vlan_my_qpn &
					if (priv->vlgrp && (cqe->vlan_my_qpn &
							    cpu_to_be32(MLX4_CQE_VLAN_PRESENT_MASK))) {
							    cpu_to_be32(MLX4_CQE_VLAN_PRESENT_MASK)))
						lro_vlan_hwaccel_receive_frags(
						vlan_gro_frags(&cq->napi, priv->vlgrp, be16_to_cpu(cqe->sl_vid));
						       &ring->lro, lro_frags,
					else
						       length, length,
						napi_gro_frags(&cq->napi);
						       priv->vlgrp,
						       be16_to_cpu(cqe->sl_vid),
						       NULL, 0);
					} else
						lro_receive_frags(&ring->lro,
								  lro_frags,
								  length,
								  length,
								  NULL, 0);


					goto next;
					goto next;
				}
				}


				/* LRO not possible, complete processing here */
				/* LRO not possible, complete processing here */
				ip_summed = CHECKSUM_UNNECESSARY;
				ip_summed = CHECKSUM_UNNECESSARY;
				INC_PERF_COUNTER(priv->pstats.lro_misses);
			} else {
			} else {
				ip_summed = CHECKSUM_NONE;
				ip_summed = CHECKSUM_NONE;
				priv->port_stats.rx_chksum_none++;
				priv->port_stats.rx_chksum_none++;
@@ -694,14 +658,10 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
		if (++polled == budget) {
		if (++polled == budget) {
			/* We are here because we reached the NAPI budget -
			/* We are here because we reached the NAPI budget -
			 * flush only pending LRO sessions */
			 * flush only pending LRO sessions */
			lro_flush_all(&ring->lro);
			goto out;
			goto out;
		}
		}
	}
	}


	/* If CQ is empty flush all LRO sessions unconditionally */
	lro_flush_all(&ring->lro);

out:
out:
	AVG_PERF_COUNTER(priv->pstats.rx_coal_avg, polled);
	AVG_PERF_COUNTER(priv->pstats.rx_coal_avg, polled);
	mlx4_cq_set_ci(&cq->mcq);
	mlx4_cq_set_ci(&cq->mcq);
+1 −0
Original line number Original line Diff line number Diff line
@@ -38,6 +38,7 @@
#include <linux/skbuff.h>
#include <linux/skbuff.h>
#include <linux/if_vlan.h>
#include <linux/if_vlan.h>
#include <linux/vmalloc.h>
#include <linux/vmalloc.h>
#include <linux/tcp.h>


#include "mlx4_en.h"
#include "mlx4_en.h"


+0 −5
Original line number Original line Diff line number Diff line
@@ -38,7 +38,6 @@
#include <linux/list.h>
#include <linux/list.h>
#include <linux/mutex.h>
#include <linux/mutex.h>
#include <linux/netdevice.h>
#include <linux/netdevice.h>
#include <linux/inet_lro.h>


#include <linux/mlx4/device.h>
#include <linux/mlx4/device.h>
#include <linux/mlx4/qp.h>
#include <linux/mlx4/qp.h>
@@ -254,7 +253,6 @@ struct mlx4_en_rx_desc {
struct mlx4_en_rx_ring {
struct mlx4_en_rx_ring {
	struct mlx4_hwq_resources wqres;
	struct mlx4_hwq_resources wqres;
	struct mlx4_en_rx_alloc page_alloc[MLX4_EN_MAX_RX_FRAGS];
	struct mlx4_en_rx_alloc page_alloc[MLX4_EN_MAX_RX_FRAGS];
	struct net_lro_mgr lro;
	u32 size ;	/* number of Rx descs*/
	u32 size ;	/* number of Rx descs*/
	u32 actual_size;
	u32 actual_size;
	u32 size_mask;
	u32 size_mask;
@@ -379,9 +377,6 @@ struct mlx4_en_pkt_stats {
};
};


struct mlx4_en_port_stats {
struct mlx4_en_port_stats {
	unsigned long lro_aggregated;
	unsigned long lro_flushed;
	unsigned long lro_no_desc;
	unsigned long tso_packets;
	unsigned long tso_packets;
	unsigned long queue_stopped;
	unsigned long queue_stopped;
	unsigned long wake_queue;
	unsigned long wake_queue;