On Thu, Apr 20, 2017 at 2:32 AM, Martin KaFai Lau <ka...@fb.com> wrote:
> We have observed a sudden spike in rx/tx_packets and rx/tx_bytes
> reported under /proc/net/dev. There is a race in mlx5e_update_stats()
> and some of the get-stats functions (the one that we hit is the
> mlx5e_get_stats() which is called by ndo_get_stats64()).
>
> In particular, the very first thing mlx5e_update_sw_counters()
> does is 'memset(s, 0, sizeof(*s))'. For example, if mlx5e_get_stats()
> is unlucky at one point, rx_bytes and rx_packets could be 0. One second
> later, a normal (and much bigger than 0) value will be reported.
>
> This patch is to use a 'struct mlx5e_sw_stats temp' to avoid
> a direct memset zero on priv->stats.sw.
>
> mlx5e_update_vport_counters() has a similar race. Hence, addressed
> together.
>
> I am lucky enough to catch this 0-reset in rx multicast:
> eth0: 41457665 76804 70 0 0 70 0 47085 15586634
> 87502 3 0 0 0 3 0
> eth0: 41459860 76815 70 0 0 70 0 47094 15588376
> 87516 3 0 0 0 3 0
> eth0: 41460577 76822 70 0 0 70 0 0 15589083
> 87521 3 0 0 0 3 0
> eth0: 41463293 76838 70 0 0 70 0 47108 15595872
> 87538 3 0 0 0 3 0
> eth0: 41463379 76839 70 0 0 70 0 47116 15596138
> 87539 3 0 0 0 3 0
>
> Cc: Saeed Mahameed <sae...@mellanox.com>
> Suggested-by: Eric Dumazet <eric.duma...@gmail.com>
> Signed-off-by: Martin KaFai Lau <ka...@fb.com>
> ---
> drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
> b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
> index 66c133757a5e..246786bb861b 100644
> --- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
> @@ -174,7 +174,7 @@ static void mlx5e_tx_timeout_work(struct work_struct
> *work)
>
> static void mlx5e_update_sw_counters(struct mlx5e_priv *priv)
> {
> - struct mlx5e_sw_stats *s = &priv->stats.sw;
> + struct mlx5e_sw_stats temp, *s = &temp;
> struct mlx5e_rq_stats *rq_stats;
> struct mlx5e_sq_stats *sq_stats;
> u64 tx_offload_none = 0;
> @@ -229,12 +229,14 @@ static void mlx5e_update_sw_counters(struct mlx5e_priv
> *priv)
> s->link_down_events_phy = MLX5_GET(ppcnt_reg,
> priv->stats.pport.phy_counters,
>
> counter_set.phys_layer_cntrs.link_down_events);
> + memcpy(&priv->stats.sw, s, sizeof(*s));
> }
>
> static void mlx5e_update_vport_counters(struct mlx5e_priv *priv)
> {
> + struct mlx5e_vport_stats temp;
> int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
> - u32 *out = (u32 *)priv->stats.vport.query_vport_out;
> + u32 *out = (u32 *)temp.query_vport_out;
> u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)] = {0};
> struct mlx5_core_dev *mdev = priv->mdev;
>
> @@ -245,6 +247,7 @@ static void mlx5e_update_vport_counters(struct mlx5e_priv
> *priv)
>
> memset(out, 0, outlen);
Actually you don't need any temp here, it is safe to just remove this
redundant memset
and mlx5_cmd_exec will do the copy for you.
> mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen);
> + memcpy(priv->stats.vport.query_vport_out, out, outlen);
> }
Anyway we still need a spin lock here, and also for all the counters
under priv->stats which are affected by this race as well.
If you want I can accept this as a temporary fix for net and Gal will
work on a spin lock based mechanism to fix the memcpy race for all the
counters.
>
> static void mlx5e_update_pport_counters(struct mlx5e_priv *priv)
> --
> 2.9.3
>
