Re: [PATCH net 4/6] net/mlx5: Destroy vport QoS element when no configuration remains

[Przemek Kitszel]

On 8/13/25 16:31, Tariq Toukan wrote:
> From: Carolina Jubran <cjubran@xxxxxxxxxx>
>
> If a VF has been configured and the user later clears all QoS settings,
> the vport element remains in the firmware QoS tree. This leads to
> inconsistent behavior compared to VFs that were never configured, since
> the FW assumes that unconfigured VFs are outside the QoS hierarchy.
> As a result, the bandwidth share across VFs may differ, even though
> none of them appear to have any configuration.
>
> Align the driver behavior with the FW expectation by destroying the
> vport QoS element when all configurations are removed.
>
> Fixes: c9497c98901c ("net/mlx5: Add support for setting VF min rate")
> Fixes: cf7e73770d1b ("net/mlx5: Manage TC arbiter nodes and implement full support for tc-bw")
> Signed-off-by: Carolina Jubran <cjubran@xxxxxxxxxx>
> Reviewed-by: Cosmin Ratiu <cratiu@xxxxxxxxxx>
> Signed-off-by: Tariq Toukan <tariqt@xxxxxxxxxx>
> ---
>   .../net/ethernet/mellanox/mlx5/core/esw/qos.c | 54 +++++++++++++++++--
>   1 file changed, 51 insertions(+), 3 deletions(-)


> +static bool esw_vport_qos_check_and_disable(struct mlx5_vport *vport,
> +					    struct devlink_rate *parent,
> +					    u64 tx_max, u64 tx_share,
> +					    u32 *tc_bw)
> +{
> +	struct mlx5_eswitch *esw = vport->dev->priv.eswitch;
> +
> +	if (parent || tx_max || tx_share || !esw_qos_tc_bw_disabled(tc_bw))
> +		return false;
> +
> +	esw_qos_lock(esw);
> +	mlx5_esw_qos_vport_disable_locked(vport);
> +	esw_qos_unlock(esw);
> +
> +	return true;
> +}
> +
>   int mlx5_esw_qos_init(struct mlx5_eswitch *esw)
>   {
>   	if (esw->qos.domain)
> @@ -1703,6 +1731,11 @@ int mlx5_esw_devlink_rate_leaf_tx_share_set(struct devlink_rate *rate_leaf, void
>   	if (!mlx5_esw_allowed(esw))
>   		return -EPERM;
>   
> +	if (esw_vport_qos_check_and_disable(vport, rate_leaf->parent,
> +					    rate_leaf->tx_max, tx_share,
> +					    rate_leaf->tc_bw))
> +		return 0;
> +

I would rather keep executing the code that "sets tx_share to 0 and
propagates the info", and only then prune all-0 nodes.
Same for other params (tx_max, ...)

That would be less risky and more future-proof, and also would let your
check&disable function to take less params.

Finally, the name is poor, what about:?
	esw_vport_qos_prune_empty(vport, rate_leaf);
(after applying my prev suggestion the above line will be at the
bottom of function).

Also a note, that if you apply the above, it would be also good to
keep the "esw_qos_lock() just once" (as it is now)

>   	err = esw_qos_devlink_rate_to_mbps(vport->dev, "tx_share", &tx_share, extack);
>   	if (err)
>   		return err;
> @@ -1724,6 +1757,11 @@ int mlx5_esw_devlink_rate_leaf_tx_max_set(struct devlink_rate *rate_leaf, void *
>   	if (!mlx5_esw_allowed(esw))
>   		return -EPERM;
>   
> +	if (esw_vport_qos_check_and_disable(vport, rate_leaf->parent, tx_max,
> +					    rate_leaf->tx_share,
> +					    rate_leaf->tc_bw))
> +		return 0;
> +
>   	err = esw_qos_devlink_rate_to_mbps(vport->dev, "tx_max", &tx_max, extack);
>   	if (err)
>   		return err;
> @@ -1749,6 +1787,11 @@ int mlx5_esw_devlink_rate_leaf_tc_bw_set(struct devlink_rate *rate_leaf,
>   	if (!mlx5_esw_allowed(esw))
>   		return -EPERM;
>   
> +	if (esw_vport_qos_check_and_disable(vport, rate_leaf->parent,
> +					    rate_leaf->tx_max,
> +					    rate_leaf->tx_share, tc_bw))
> +		return 0;
> +
>   	disable = esw_qos_tc_bw_disabled(tc_bw);
>   	esw_qos_lock(esw);
>   
> @@ -1930,6 +1973,11 @@ int mlx5_esw_devlink_rate_leaf_parent_set(struct devlink_rate *devlink_rate,
>   	struct mlx5_esw_sched_node *node;
>   	struct mlx5_vport *vport = priv;
>   
> +	if (esw_vport_qos_check_and_disable(vport, parent, devlink_rate->tx_max,
> +					    devlink_rate->tx_share,
> +					    devlink_rate->tc_bw))
> +		return 0;
> +
>   	if (!parent)
>   		return mlx5_esw_qos_vport_update_parent(vport, NULL, extack);
>