[dpdk-dev] [PATCH v2] net/i40e: Fail rte_flow MARK requests if RX func was vectorized

[Zhang, Qi Z]

> -----Original Message-----
> From: Ergin, Mesut A
> Sent: Saturday, May 25, 2019 6:50 AM
> To: Xing, Beilei <beilei.xing at intel.com>; Zhang, Qi Z <qi.z.zhang at intel.com>
> Cc: dev at dpdk.org; Ergin, Mesut A <mesut.a.ergin at intel.com>
> Subject: [PATCH v2] net/i40e: Fail rte_flow MARK requests if RX func was
> vectorized
> 
> Runtime requests to install an rte_flow with MARK action should fail if a vector
> RX function was already chosen for the device during configuration time.
> 
> Currently, i40e rte_flow driver would successfully install the flow with MARK
> action, even when vector RX functions are in use. However, those vector RX
> functions will fail to retrieve the MARK data from the device descriptor into
> the mbuf. The original app installing the flow would never know what went
> wrong. The change introduced in this patch must be reverted if/when vector
> RX functions start supporting correct FDIR processing for MARK actions.
> 
> Signed-off-by: Mesut Ali Ergin <mesut.a.ergin at intel.com>

You can add below fix line with cc stable
Fixes: 2e67a7fbf3ff ("net/i40e: config flow director automatically")
Cc: stable at dpdk.org

> 
> v2:
>  - Check added to fail MARK request only if the device was started
> ---
>  drivers/net/i40e/i40e_flow.c | 22 ++++++++++++++++++++++
>  1 file changed, 22 insertions(+)
> 
> diff --git a/drivers/net/i40e/i40e_flow.c b/drivers/net/i40e/i40e_flow.c index
> 3afd779..d1233ed 100644
> --- a/drivers/net/i40e/i40e_flow.c
> +++ b/drivers/net/i40e/i40e_flow.c
> @@ -22,12 +22,20 @@
>  #include "base/i40e_type.h"
>  #include "base/i40e_prototype.h"
>  #include "i40e_ethdev.h"
> +#include "i40e_rxtx.h"
> 
>  #define I40E_IPV6_TC_MASK	(0xFF << I40E_FDIR_IPv6_TC_OFFSET)
>  #define I40E_IPV6_FRAG_HEADER	44
>  #define I40E_TENANT_ARRAY_NUM	3
>  #define I40E_TCI_MASK		0xFFFF
> 
> +/* Check whether device RX function is vectorized  */
> +#define I40E_RX_IS_VECTOR(dev)	\
> +	(dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec || \
> +	 dev->rx_pkt_burst == i40e_recv_pkts_vec || \
> +	 dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec_avx2 || \
> +	 dev->rx_pkt_burst == i40e_recv_pkts_vec_avx2)

Some code style warning in patchwork need to be fixed.
Maybe we can just wrap the check algorithm into one function (ex. i40e_check_mark_support.)
And in future when the mark offload is ready, we can keep the function name and just replace the inside logic? What do you think?

Regards
Qi

> +
>  static int i40e_flow_validate(struct rte_eth_dev *dev,
>  			      const struct rte_flow_attr *attr,
>  			      const struct rte_flow_item pattern[], @@ -3079,6
> +3087,13 @@ i40e_flow_parse_fdir_action(struct rte_eth_dev *dev,
>  		filter->action.behavior = I40E_FDIR_PASSTHRU;
>  		break;
>  	case RTE_FLOW_ACTION_TYPE_MARK:
> +		if (I40E_RX_IS_VECTOR(dev) && dev->data->dev_started) {
> +			/* MARK not supported w/ Vector RX on a started device*/
> +			rte_flow_error_set(error, EINVAL,
> +					   RTE_FLOW_ERROR_TYPE_ACTION, act,
> +					   "Action not supported: Vector RX has been
> active.");
> +			return -rte_errno;
> +		}
>  		filter->action.behavior = I40E_FDIR_PASSTHRU;
>  		mark_spec = act->conf;
>  		filter->action.report_status = I40E_FDIR_REPORT_ID; @@ -3103,6
> +3118,13 @@ i40e_flow_parse_fdir_action(struct rte_eth_dev *dev,
>  			   "Invalid action.");
>  			return -rte_errno;
>  		}
> +		if (I40E_RX_IS_VECTOR(dev) && dev->data->dev_started) {
> +			/* MARK not supported w/ Vector RX on a started device*/
> +			rte_flow_error_set(error, EINVAL,
> +					   RTE_FLOW_ERROR_TYPE_ACTION, act,
> +					   "Action not supported: Vector RX has been
> active.");
> +			return -rte_errno;
> +		}
>  		mark_spec = act->conf;
>  		filter->action.report_status = I40E_FDIR_REPORT_ID;
>  		filter->soft_id = mark_spec->id;
> --
> 2.7.4