Re: [dpdk-stable] [PATCH] net/mlx5: fix xstats get reinitialization

[Slava Ovsiienko <viacheslavo@nvidia.com>]

Hi, Vladislav

Thank you for the fix.
In general, it looks good to me.
Please, see my comments below.

> -----Original Message-----
> From: Vladislav Grishenko <themiron@yandex-team.ru>
> Sent: Saturday, September 25, 2021 18:11
> To: Matan Azrad <matan@nvidia.com>; Slava Ovsiienko
> <viacheslavo@nvidia.com>; Shahaf Shuler <shahafs@nvidia.com>; NBU-
> Contact-Adrien Mazarguil <adrien.mazarguil@6wind.com>; Elad Persiko
> <eladpe@mellanox.com>
> Cc: dev@dpdk.org; stable@dpdk.org
> Subject: [PATCH] net/mlx5: fix xstats get reinitialization
> 
> The mlx5_xstats_get gets the device extended statistics.
> In this function the driver may reinitialize the structures that are used to read
> device counters.
> 
> In case of reinitialization, the number of counters may change, which
> wouldn't be taken into account by the get API callback and can cause a
> segmentation fault.
> 
> In case of insufficient supplied stats table size, ex. zero to query the number of
[SO] I was embarrassed a little bit with ".ex". Let's clarify?

> extended stats, reinitialization may never happen and the returned stats
> number, that is used for subsequent stats getting, will not be sufficient too.
> 
> This issue is fixed by getting and allocating the counters size after the
> reinitialization.
Yes, it would be fair.

> 
> Fixes: 1a611fdaf6ec ("net/mlx5: support missing counter in extended
> statistics")
> Fixes: a4193ae3bc4f ("net/mlx5: support extended statistics")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Vladislav Grishenko <themiron@yandex-team.ru>
> ---
>  drivers/net/mlx5/mlx5_stats.c | 35 +++++++++++++++++++++++++----------
>  1 file changed, 25 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/net/mlx5/mlx5_stats.c b/drivers/net/mlx5/mlx5_stats.c
> index ae2f5668a7..7dd7724b05 100644
> --- a/drivers/net/mlx5/mlx5_stats.c
> +++ b/drivers/net/mlx5/mlx5_stats.c
> @@ -39,23 +39,37 @@ mlx5_xstats_get(struct rte_eth_dev *dev, struct
> rte_eth_xstat *stats,
>  		unsigned int n)
>  {
>  	struct mlx5_priv *priv = dev->data->dev_private;
> -	unsigned int i;
> -	uint64_t counters[n];
>  	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
> -	uint16_t mlx5_stats_n = xstats_ctrl->mlx5_stats_n;
> +	uint16_t mlx5_stats_n;
> +	int stats_n;
> +
> +	stats_n = mlx5_os_get_stats_n(dev);
> +	if (stats_n < 0)
> +		return stats_n;
> +	if (xstats_ctrl->stats_n != stats_n)
> +		mlx5_os_stats_init(dev);
> +	mlx5_stats_n = xstats_ctrl->mlx5_stats_n;
> 
>  	if (n >= mlx5_stats_n && stats) {
> -		int stats_n;
> +		uint64_t *counters;
> +		unsigned int i;
>  		int ret;
> 
> -		stats_n = mlx5_os_get_stats_n(dev);
> -		if (stats_n < 0)
> -			return stats_n;
> -		if (xstats_ctrl->stats_n != stats_n)
> -			mlx5_os_stats_init(dev);
> +		counters = mlx5_malloc(MLX5_MEM_SYS, sizeof(*counters) *

[SO] My concern is alloc/free - can we revert back to local counter[n] on stack ?


With best regards,
Slava

> +				mlx5_stats_n, 0,
> +				SOCKET_ID_ANY);
> +		if (counters == NULL) {
> +			DRV_LOG(WARNING, "port %u unable to allocate "
> +					"memory for xstats counters",
> +					dev->data->port_id);
> +			rte_errno = ENOMEM;
> +			return -rte_errno;
> +		}
>  		ret = mlx5_os_read_dev_counters(dev, counters);
> -		if (ret)
> +		if (ret) {
> +			mlx5_free(counters);
>  			return ret;
> +		}
>  		for (i = 0; i != mlx5_stats_n; ++i) {
>  			stats[i].id = i;
>  			if (xstats_ctrl->info[i].dev) {
> @@ -76,6 +90,7 @@ mlx5_xstats_get(struct rte_eth_dev *dev, struct
> rte_eth_xstat *stats,
>  					(counters[i] - xstats_ctrl->base[i]);
>  			}
>  		}
> +		mlx5_free(counters);
>  	}
>  	mlx5_stats_n = mlx5_txpp_xstats_get(dev, stats, n, mlx5_stats_n);
>  	return mlx5_stats_n;
> --
> 2.17.1