[net-next v0 13/14] net/mlx5e: Allocate per-channel stats dynamically at first usage
flat view
From: Saeed Mahameed <saeed@kernel.org>
Date: 2021-12-22 03:16:30
Subsystem:
mellanox ethernet driver (mlx5e), mellanox mlx5 core vpi driver, networking drivers, the rest · Maintainers:
Saeed Mahameed, Tariq Toukan, Mark Bloch, Leon Romanovsky, Andrew Lunn, "David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Linus Torvalds
From: Lama Kayal <redacted> Make stats allocation per-channel dynamic on demand, at channel open operation. Previously the stats array was pre-allocated for the maximum possible number of channels. Here we defer the per-channel stats instance allocation upon its first usage, so that it's allocated only if really needed. Allocating stats on demand helps maintain a more memory-efficient code, as we're saving memory when the used number of channels is smaller than the maximum. The stats memory instances are still freed in mlx5e_priv_arrays_free(), so that they are persistent to channels' closure. Memory size allocated for struct mlx5e_channel_stats is 3648 bytes. If maximum number of channel stands for 64, the total memory space allocated for stats is 3648x64 = 228K bytes. In scenarios where the number of channels in use is significantly smaller than maximum number, the memory saved can be remarkable. Signed-off-by: Lama Kayal <redacted> Reviewed-by: Tariq Toukan <tariqt@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com> --- .../net/ethernet/mellanox/mlx5/core/en_main.c | 41 +++++++++++++------ 1 file changed, 28 insertions(+), 13 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 504844097d20..1a47108805fd 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c@@ -2176,6 +2176,30 @@ static u8 mlx5e_enumerate_lag_port(struct mlx5_core_dev *mdev, int ix) return (ix + port_aff_bias) % mlx5e_get_num_lag_ports(mdev); } +static int mlx5e_channel_stats_alloc(struct mlx5e_priv *priv, int ix, int cpu) +{ + if (ix > priv->stats_nch) { + netdev_warn(priv->netdev, "Unexpected channel stats index %d > %d\n", ix, + priv->stats_nch); + return -EINVAL; + } + + if (priv->channel_stats[ix]) + return 0; + + /* Asymmetric dynamic memory allocation. + * Freed in mlx5e_priv_arrays_free, not on channel closure. + */ + mlx5e_dbg(DRV, priv, "Creating channel stats %d\n", ix); + priv->channel_stats[ix] = kvzalloc_node(sizeof(**priv->channel_stats), + GFP_KERNEL, cpu_to_node(cpu)); + if (!priv->channel_stats[ix]) + return -ENOMEM; + priv->stats_nch++; + + return 0; +} + static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix, struct mlx5e_params *params, struct mlx5e_channel_param *cparam,
@@ -2193,6 +2217,10 @@ static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix, if (err) return err; + err = mlx5e_channel_stats_alloc(priv, ix, cpu); + if (err) + return err; + c = kvzalloc_node(sizeof(*c), GFP_KERNEL, cpu_to_node(cpu)); if (!c) return -ENOMEM;
@@ -5153,7 +5181,6 @@ int mlx5e_priv_init(struct mlx5e_priv *priv, priv->netdev = netdev; priv->msglevel = MLX5E_MSG_LEVEL; priv->max_nch = nch; - priv->stats_nch = nch; priv->max_opened_tc = 1; if (!alloc_cpumask_var(&priv->scratchpad.cpumask, GFP_KERNEL))
@@ -5196,20 +5223,8 @@ int mlx5e_priv_init(struct mlx5e_priv *priv, if (!priv->channel_stats) goto err_free_channel_tc2realtxq; - for (i = 0; i < priv->stats_nch; i++) { - priv->channel_stats[i] = kvzalloc_node(sizeof(**priv->channel_stats), - GFP_KERNEL, node); - if (!priv->channel_stats[i]) - goto err_free_channel_stats; - } - return 0; -err_free_channel_stats: - while (--i >= 0) - kvfree(priv->channel_stats[i]); - kfree(priv->channel_stats); - i = nch; err_free_channel_tc2realtxq: while (--i >= 0) kfree(priv->channel_tc2realtxq[i]);
--
2.33.1