Hi, beilei
> -----Original Message-----
> From: Xing, Beilei <beilei.x...@intel.com>
> Sent: Tuesday, December 29, 2020 2:18 PM
> To: Guo, Jia <jia....@intel.com>; dev@dpdk.org
> Cc: Xing, Beilei <beilei.x...@intel.com>; sta...@dpdk.org; Sun, Chenmin
> <chenmin....@intel.com>
> Subject: [PATCH v2] net/i40e: fix flex payload rule conflict issue
>
> From: Beilei Xing <beilei.x...@intel.com>
>
> With the following commands, the second flow can't be created successfully.
>
> 1. flow create 0 ingress pattern eth / ipv4 / udp /
> raw relative is 1 pattern is 0102030405 / end
> actions drop / end
> 2. flow destroy 0 rule 0
> 3. flow create 0 ingress pattern eth / ipv4 / udp /
> raw relative is 1 pattern is 010203040506 / end
> actions drop / end
>
> The root cause is that a flag for flex pit isn't reset.
>
> Fixes: 6ced3dd72f5f ("net/i40e: support flexible payload parsing for FDIR")
> Cc: sta...@dpdk.org
>
> Reported-by: Chenmin Sun<chenmin....@intel.com>
> Signed-off-by: Beilei Xing <beilei.x...@intel.com>
> ---
>
> v2 changes:
> - Add fix line.
> - Refine comments.
>
> drivers/net/i40e/i40e_flow.c | 10 +++++++++-
> 1 file changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/net/i40e/i40e_flow.c b/drivers/net/i40e/i40e_flow.c
> index b09ff6590d..65e0b69356 100644
> --- a/drivers/net/i40e/i40e_flow.c
> +++ b/drivers/net/i40e/i40e_flow.c
> @@ -5284,6 +5284,7 @@ i40e_flow_destroy(struct rte_eth_dev *dev,
> enum rte_filter_type filter_type = flow->filter_type;
> struct i40e_fdir_info *fdir_info = &pf->fdir;
> int ret = 0;
> + int i;
>
> switch (filter_type) {
> case RTE_ETH_FILTER_ETHERTYPE:
> @@ -5299,9 +5300,13 @@ i40e_flow_destroy(struct rte_eth_dev *dev,
> &((struct i40e_fdir_filter *)flow->rule)->fdir,
> 0);
>
> - /* If the last flow is destroyed, disable fdir. */
> + /* When the last flow is destroyed. */
> if (!ret && TAILQ_EMPTY(&pf->fdir.fdir_list)) {
> + /* Disable FDIR processing. */
> i40e_fdir_rx_proc_enable(dev, 0);
> + /* Reset the flex_pit_flag. */
> + for (i = 0; i < I40E_MAX_FLXPLD_LAYER; i++)
> + pf->fdir.flex_pit_flag[i] = 0;
Why reset all flex_pit_flag when destroy the last flow, if destroy other flow,
is it no need to reset corresponding flex_pit_flag which set before when the
flow added?
And reset flag should be ahead of the FDIR disabling I think.
> }
> break;
> case RTE_ETH_FILTER_HASH:
> @@ -5515,6 +5520,9 @@ i40e_flow_flush_fdir_filter(struct i40e_pf *pf)
> pf->fdir.flex_mask_flag[pctype] = 0;
> }
>
> + for (i = 0; i < I40E_MAX_FLXPLD_LAYER; i++)
> + pf->fdir.flex_pit_flag[i] = 0;
> +
> /* Disable FDIR processing as all FDIR rules are now flushed
> */
> i40e_fdir_rx_proc_enable(dev, 0);
> }
> --
> 2.26.2
