Re: [dpdk-dev] [PATCH v2 1/1] net/mlx5: support match ICMP identifier fields

[Ori Kam <orika@nvidia.com>]

Hi Li,

PSB,

Thanks,
Ori

> -----Original Message-----
> From: Li Zhang <lizh@nvidia.com>
> Sent: Monday, September 28, 2020 6:38 AM
> Subject: [PATCH v2 1/1] net/mlx5: support match ICMP identifier fields
> 
> PRM expose fields "Icmp_header_data" in ICMP.
> Update ICMP mask parameter with ICMP identifier and sequence number
> fields.
> ICMP sequence number spec with mask, Icmp_header_data low 16 bits are set.
> ICMP identifier spec with mask, Icmp_header_data high 16 bits are set.
> 
Is it relevant only for ipv4 ICMP?

> Signed-off-by: Li Zhang <lizh@nvidia.com>
> ---
>  doc/guides/nics/mlx5.rst               |  4 ++--
>  doc/guides/rel_notes/release_20_11.rst |  2 +-
>  drivers/net/mlx5/mlx5_flow.c           |  9 +++++++--
>  drivers/net/mlx5/mlx5_flow_dv.c        | 16 +++++++++++++++-
>  4 files changed, 25 insertions(+), 6 deletions(-)
> 
> diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
> index 211c0c5a6c..576dbe5efd 100644
> --- a/doc/guides/nics/mlx5.rst
> +++ b/doc/guides/nics/mlx5.rst
> @@ -288,7 +288,7 @@ Limitations
>    - The input buffer, providing the removal size, is not validated.
>    - The buffer size must match the length of the headers to be removed.
> 
> -- ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
> +- ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching,
> IP-in-IP and MPLS flow matching are all
>    mutually exclusive features which cannot be supported together
>    (see :ref:`mlx5_firmware_config`).
> 
> @@ -1009,7 +1009,7 @@ Below are some firmware configurations listed.
> 
>      FLEX_PARSER_PROFILE_ENABLE=1
> 
> -- enable ICMP/ICMP6 code/type fields matching::
> +- enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type)
> fields matching::
> 
>      FLEX_PARSER_PROFILE_ENABLE=2
> 
> diff --git a/doc/guides/rel_notes/release_20_11.rst
> b/doc/guides/rel_notes/release_20_11.rst
> index c6642f5f94..791f133d8f 100644
> --- a/doc/guides/rel_notes/release_20_11.rst
> +++ b/doc/guides/rel_notes/release_20_11.rst
> @@ -73,7 +73,7 @@ New Features
>    * Added flag action.
>    * Added raw encap/decap actions.
>    * Added VXLAN encap/decap actions.
> -  * Added ICMP and ICMP6 matching items.
> +  * Added ICMP(code/type/identifier/sequence number) and
> ICMP6(code/type) matching items.
>    * Added option to set port mask for insertion/deletion:
>      ``--portmask=N``
>      where N represents the hexadecimal bitmask of ports used.
> diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
> index 416505f1c8..7bd5c5da94 100644
> --- a/drivers/net/mlx5/mlx5_flow.c
> +++ b/drivers/net/mlx5/mlx5_flow.c
> @@ -1303,6 +1303,7 @@ mlx5_flow_validate_item_icmp(const struct
> rte_flow_item *item,
>  			     struct rte_flow_error *error)

This function is shared between the dv and the verbs.
I think we can support this only in dv, so we need to split this function.

>  {
>  	const struct rte_flow_item_icmp *mask = item->mask;
> +	struct rte_flow_item_icmp default_mask;

I think the correct name is nic_mask.

>  	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
>  	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
>  				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
> @@ -1324,11 +1325,15 @@ mlx5_flow_validate_item_icmp(const struct
> rte_flow_item *item,
>  		return rte_flow_error_set(error, EINVAL,
>  					  RTE_FLOW_ERROR_TYPE_ITEM,
> item,
>  					  "multiple L4 layers not supported");
> +	memcpy(&default_mask, &rte_flow_item_icmp_mask,
> +	       sizeof(struct rte_flow_item_icmp));
> +	default_mask.hdr.icmp_ident = RTE_BE16(0xFFFF);
> +	default_mask.hdr.icmp_seq_nb = RTE_BE16(0xFFFF);

You don't need to mem copy just init this structure when you declare it.

>  	if (!mask)
> -		mask = &rte_flow_item_icmp_mask;
> +		mask = &default_mask;
>  	ret = mlx5_flow_item_acceptable
>  		(item, (const uint8_t *)mask,
> -		 (const uint8_t *)&rte_flow_item_icmp_mask,
> +		 (const uint8_t *)&default_mask,
>  		 sizeof(struct rte_flow_item_icmp), error);
>  	if (ret < 0)
>  		return ret;
> diff --git a/drivers/net/mlx5/mlx5_flow_dv.c
> b/drivers/net/mlx5/mlx5_flow_dv.c
> index 3819cdb266..b5d6455067 100644
> --- a/drivers/net/mlx5/mlx5_flow_dv.c
> +++ b/drivers/net/mlx5/mlx5_flow_dv.c
> @@ -7378,6 +7378,8 @@ flow_dv_translate_item_icmp(void *matcher, void
> *key,
>  {
>  	const struct rte_flow_item_icmp *icmp_m = item->mask;
>  	const struct rte_flow_item_icmp *icmp_v = item->spec;
> +	uint32_t icmp_header_data_m = 0;
> +	uint32_t icmp_header_data_v = 0;
>  	void *headers_m;
>  	void *headers_v;
>  	void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
> @@ -7396,8 +7398,14 @@ flow_dv_translate_item_icmp(void *matcher, void
> *key,
>  	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
> IPPROTO_ICMP);
>  	if (!icmp_v)
>  		return;
> -	if (!icmp_m)
> +	if (!icmp_m) {
>  		icmp_m = &rte_flow_item_icmp_mask;
> +		icmp_header_data_m = RTE_BE32(UINT32_MAX);

Why are you setting the data as default mask?

> +	} else {
> +		icmp_header_data_m = rte_cpu_to_be_16(icmp_m-
> >hdr.icmp_seq_nb);

Isn't the icmp_seq_nb already in BE?

> +		icmp_header_data_m |=
> +			rte_cpu_to_be_16(icmp_m->hdr.icmp_ident) << 16;

Again the icmp_ident is already in BE

> +	}
>  	/*
>  	 * Force flow only to match the non-fragmented IPv4 ICMP packets.
>  	 * If only the protocol is specified, no need to match the frag.
> @@ -7412,6 +7420,12 @@ flow_dv_translate_item_icmp(void *matcher, void
> *key,
>  		 icmp_m->hdr.icmp_code);
>  	MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
>  		 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
> +	icmp_header_data_v = rte_cpu_to_be_16(icmp_v->hdr.icmp_seq_nb);
> +	icmp_header_data_v |= rte_cpu_to_be_16(icmp_v->hdr.icmp_ident)
> << 16;

The BE issue again, 
> +	MLX5_SET(fte_match_set_misc3, misc3_m, icmp_header_data,
> +		 icmp_header_data_m);
> +	MLX5_SET(fte_match_set_misc3, misc3_v, icmp_header_data,
> +		 icmp_header_data_v & icmp_header_data_m);
>  }
> 
>  /**
> --
> 2.21.0