Commit fbc6daf1 authored by Amir Vadai's avatar Amir Vadai Committed by David S. Miller

net/mlx4_en: Ignore budget on TX napi polling

It is recommended that TX work not count against the quota.
The cost of TX packet liberation is a minute percentage of what it costs to
process an RX frame. Furthermore, that SKB freeing makes memory available for
other paths in the stack.

Give the TX a larger budget and be more aggressive about cleaning up the Tx
descriptors this budget could be changed using ethtool:
$ ethtool -C eth1 tx-frames-irq <budget>
Signed-off-by: default avatarAmir Vadai <amirv@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 0b030244
......@@ -417,6 +417,8 @@ static int mlx4_en_get_coalesce(struct net_device *dev,
coal->tx_coalesce_usecs = priv->tx_usecs;
coal->tx_max_coalesced_frames = priv->tx_frames;
coal->tx_max_coalesced_frames_irq = priv->tx_work_limit;
coal->rx_coalesce_usecs = priv->rx_usecs;
coal->rx_max_coalesced_frames = priv->rx_frames;
......@@ -426,6 +428,7 @@ static int mlx4_en_get_coalesce(struct net_device *dev,
coal->rx_coalesce_usecs_high = priv->rx_usecs_high;
coal->rate_sample_interval = priv->sample_interval;
coal->use_adaptive_rx_coalesce = priv->adaptive_rx_coal;
return 0;
}
......@@ -434,6 +437,9 @@ static int mlx4_en_set_coalesce(struct net_device *dev,
{
struct mlx4_en_priv *priv = netdev_priv(dev);
if (!coal->tx_max_coalesced_frames_irq)
return -EINVAL;
priv->rx_frames = (coal->rx_max_coalesced_frames ==
MLX4_EN_AUTO_CONF) ?
MLX4_EN_RX_COAL_TARGET :
......@@ -457,6 +463,7 @@ static int mlx4_en_set_coalesce(struct net_device *dev,
priv->rx_usecs_high = coal->rx_coalesce_usecs_high;
priv->sample_interval = coal->rate_sample_interval;
priv->adaptive_rx_coal = coal->use_adaptive_rx_coalesce;
priv->tx_work_limit = coal->tx_max_coalesced_frames_irq;
return mlx4_en_moderation_update(priv);
}
......
......@@ -2473,6 +2473,7 @@ int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
MLX4_WQE_CTRL_SOLICITED);
priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
priv->tx_ring_num = prof->tx_ring_num;
priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
GFP_KERNEL);
......
......@@ -351,9 +351,8 @@ int mlx4_en_free_tx_buf(struct net_device *dev, struct mlx4_en_tx_ring *ring)
return cnt;
}
static int mlx4_en_process_tx_cq(struct net_device *dev,
struct mlx4_en_cq *cq,
int budget)
static bool mlx4_en_process_tx_cq(struct net_device *dev,
struct mlx4_en_cq *cq)
{
struct mlx4_en_priv *priv = netdev_priv(dev);
struct mlx4_cq *mcq = &cq->mcq;
......@@ -372,9 +371,10 @@ static int mlx4_en_process_tx_cq(struct net_device *dev,
int factor = priv->cqe_factor;
u64 timestamp = 0;
int done = 0;
int budget = priv->tx_work_limit;
if (!priv->port_up)
return 0;
return true;
index = cons_index & size_mask;
cqe = &buf[(index << factor) + factor];
......@@ -447,7 +447,7 @@ static int mlx4_en_process_tx_cq(struct net_device *dev,
netif_tx_wake_queue(ring->tx_queue);
ring->wake_queue++;
}
return done;
return done < budget;
}
void mlx4_en_tx_irq(struct mlx4_cq *mcq)
......@@ -467,18 +467,16 @@ int mlx4_en_poll_tx_cq(struct napi_struct *napi, int budget)
struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
struct net_device *dev = cq->dev;
struct mlx4_en_priv *priv = netdev_priv(dev);
int done;
int clean_complete;
done = mlx4_en_process_tx_cq(dev, cq, budget);
clean_complete = mlx4_en_process_tx_cq(dev, cq);
if (!clean_complete)
return budget;
/* If we used up all the quota - we're probably not done yet... */
if (done < budget) {
/* Done for now */
napi_complete(napi);
mlx4_en_arm_cq(priv, cq);
return done;
}
return budget;
return 0;
}
static struct mlx4_en_tx_desc *mlx4_en_bounce_to_desc(struct mlx4_en_priv *priv,
......
......@@ -126,6 +126,8 @@ enum {
#define MAX_TX_RINGS (MLX4_EN_MAX_TX_RING_P_UP * \
MLX4_EN_NUM_UP)
#define MLX4_EN_DEFAULT_TX_WORK 256
/* Target number of packets to coalesce with interrupt moderation */
#define MLX4_EN_RX_COAL_TARGET 44
#define MLX4_EN_RX_COAL_TIME 0x10
......@@ -543,6 +545,7 @@ struct mlx4_en_priv {
__be32 ctrl_flags;
u32 flags;
u8 num_tx_rings_p_up;
u32 tx_work_limit;
u32 tx_ring_num;
u32 rx_ring_num;
u32 rx_skb_size;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment