[dpdk-stable] [PATCH] net/mlx5: fix inconsistent link status

[Yuanhan Liu]

On Tue, Feb 21, 2017 at 09:23:29AM +0200, shahafs wrote:
> From: Shahaf Shuler <shahafs at mellanox.com>
> 
> [ backported from upstream commit a9f2fbc42f0c58df614896a407fd694899315233 ]

Applied to dpdk-stable/16.11.

Thanks for the backport.

	--yliu

> 
> Querying the link status can end up being in an inconsistent state,
> like the port is reporting speed although it is down.
> 
> For this case another query is scheduled.
> 
> A race condition can occur between the scheduled query and link
> status interrupt handlers.
> 
> When the scheduled query by-pass interrupt handlers, the link status
> will be stuck in an inconsistent state.
> 
> This patch addresses the race condition by not blocking link status
> queries in case delayed query is used.
> 
> Fixes: 198a3c339a8f ("mlx5: handle link status interrupts")
> Cc: stable at dpdk.org
> 
> Signed-off-by: Shahaf Shuler <shahafs at mellanox.com>
> Acked-by: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com>
> ---
>  drivers/net/mlx5/mlx5_ethdev.c | 26 +++++++++++---------------
>  1 file changed, 11 insertions(+), 15 deletions(-)
> 
> diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
> index a709a3a..06cfd01 100644
> --- a/drivers/net/mlx5/mlx5_ethdev.c
> +++ b/drivers/net/mlx5/mlx5_ethdev.c
> @@ -1183,7 +1183,7 @@ struct priv *
>  priv_dev_link_status_handler(struct priv *priv, struct rte_eth_dev *dev)
>  {
>  	struct ibv_async_event event;
> -	int port_change = 0;
> +	struct rte_eth_link *link = &dev->data->dev_link;
>  	int ret = 0;
>  
>  	/* Read all message and acknowledge them. */
> @@ -1191,29 +1191,24 @@ struct priv *
>  		if (ibv_get_async_event(priv->ctx, &event))
>  			break;
>  
> -		if (event.event_type == IBV_EVENT_PORT_ACTIVE ||
> -		    event.event_type == IBV_EVENT_PORT_ERR)
> -			port_change = 1;
> -		else
> +		if (event.event_type != IBV_EVENT_PORT_ACTIVE &&
> +		    event.event_type != IBV_EVENT_PORT_ERR)
>  			DEBUG("event type %d on port %d not handled",
>  			      event.event_type, event.element.port_num);
>  		ibv_ack_async_event(&event);
>  	}
> -
> -	if (port_change ^ priv->pending_alarm) {
> -		struct rte_eth_link *link = &dev->data->dev_link;
> -
> -		priv->pending_alarm = 0;
> -		mlx5_link_update_unlocked(dev, 0);
> -		if (((link->link_speed == 0) && link->link_status) ||
> -		    ((link->link_speed != 0) && !link->link_status)) {
> +	mlx5_link_update(dev, 0);
> +	if (((link->link_speed == 0) && link->link_status) ||
> +	    ((link->link_speed != 0) && !link->link_status)) {
> +		if (!priv->pending_alarm) {
>  			/* Inconsistent status, check again later. */
>  			priv->pending_alarm = 1;
>  			rte_eal_alarm_set(MLX5_ALARM_TIMEOUT_US,
>  					  mlx5_dev_link_status_handler,
>  					  dev);
> -		} else
> -			ret = 1;
> +		}
> +	} else {
> +		ret = 1;
>  	}
>  	return ret;
>  }
> @@ -1233,6 +1228,7 @@ struct priv *
>  
>  	priv_lock(priv);
>  	assert(priv->pending_alarm == 1);
> +	priv->pending_alarm = 0;
>  	ret = priv_dev_link_status_handler(priv, dev);
>  	priv_unlock(priv);
>  	if (ret)
> -- 
> 1.8.3.1