Re: [dpdk-dev] [PATCH v7 4/4] net/ice: enable protocol agnostic flow offloading in FDIR

["Zhang, Qi Z" <qi.z.zhang@intel.com>]

> -----Original Message-----
> From: Guo, Junfeng <junfeng.guo@intel.com>
> Sent: Thursday, October 28, 2021 5:14 PM
> To: Zhang, Qi Z <qi.z.zhang@intel.com>; Wu, Jingjing <jingjing.wu@intel.com>;
> Xing, Beilei <beilei.xing@intel.com>
> Cc: dev@dpdk.org; Yigit, Ferruh <ferruh.yigit@intel.com>; Xu, Ting
> <ting.xu@intel.com>; Guo, Junfeng <junfeng.guo@intel.com>
> Subject: [PATCH v7 4/4] net/ice: enable protocol agnostic flow offloading in
> FDIR
> 
> Protocol agnostic flow offloading in Flow Director is enabled by this patch
> based on the Parser Library, using existing rte_flow raw API.
> 
> Note that the raw flow requires:
> 1. byte string of raw target packet bits.
> 2. byte string of mask of target packet.
> 
> Here is an example:
> FDIR matching ipv4 dst addr with 1.2.3.4 and redirect to queue 3:
> 
> flow create 0 ingress pattern raw \
> pattern spec \
> 0000000000000000000000000800450000140000400040100000000000000
> 1020304 \ pattern mask \
> 000000000000000000000000000000000000000000000000000000000000ff
> ffffff \ / end actions queue index 3 / mark id 3 / end
> 
> Note that mask of some key bits (e.g., 0x0800 to indicate ipv4 proto) is
> optional in our cases. To avoid redundancy, we just omit the mask of 0x0800
> (with 0xFFFF) in the mask byte string example. The prefix '0x' for the spec and
> mask byte (hex) strings are also omitted here.
> 
> Signed-off-by: Junfeng Guo <junfeng.guo@intel.com>
> ---
>  doc/guides/rel_notes/release_21_11.rst |   1 +
>  drivers/net/ice/ice_ethdev.h           |  17 ++
>  drivers/net/ice/ice_fdir_filter.c      | 260 +++++++++++++++++++++++++
>  drivers/net/ice/ice_generic_flow.c     |   7 +
>  drivers/net/ice/ice_generic_flow.h     |   3 +
>  5 files changed, 288 insertions(+)
> 
> diff --git a/doc/guides/rel_notes/release_21_11.rst
> b/doc/guides/rel_notes/release_21_11.rst
> index 9c13ceed1c..cc449a4340 100644
> --- a/doc/guides/rel_notes/release_21_11.rst
> +++ b/doc/guides/rel_notes/release_21_11.rst
> @@ -167,6 +167,7 @@ New Features
> 
>  * **Updated Intel ice driver.**
> 
> +  * Added protocol agnostic flow offloading support in Flow Director.
>    * Added 1PPS out support by a devargs.
>    * Added IPv4 and L4 (TCP/UDP/SCTP) checksum hash support in RSS flow.
>    * Added DEV_RX_OFFLOAD_TIMESTAMP support.
> diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h index
> 599e0028f7..441242ee89 100644
> --- a/drivers/net/ice/ice_ethdev.h
> +++ b/drivers/net/ice/ice_ethdev.h
> @@ -318,6 +318,11 @@ struct ice_fdir_filter_conf {
>  	uint64_t input_set_o; /* used for non-tunnel or tunnel outer fields */
>  	uint64_t input_set_i; /* only for tunnel inner fields */
>  	uint32_t mark_flag;
> +
> +	struct ice_parser_profile *prof;
> +	const u8 *pkt_buf;
> +	bool parser_ena;
> +	u8 pkt_len;
>  };
> 
>  #define ICE_MAX_FDIR_FILTER_NUM		(1024 * 16)
> @@ -487,6 +492,17 @@ struct ice_devargs {
>  	uint8_t pps_out_ena;
>  };
> 
> +/**
> + * Structure to store fdir fv entry.
> + */
> +struct ice_fdir_prof_info {
> +	struct LIST_ENTRY_TYPE l_entry;
> +
> +	struct ice_parser_profile prof;
> +	u16 ptg;
> +	u64 fdir_actived_cnt;
> +};
> +
>  /**
>   * Structure to store private data for each PF/VF instance.
>   */
> @@ -509,6 +525,7 @@ struct ice_adapter {
>  	struct rte_timecounter rx_tstamp_tc;
>  	struct rte_timecounter tx_tstamp_tc;
>  	bool ptp_ena;
> +	struct LIST_HEAD_TYPE fdir_prof_list;
>  #ifdef RTE_ARCH_X86
>  	bool rx_use_avx2;
>  	bool rx_use_avx512;
> diff --git a/drivers/net/ice/ice_fdir_filter.c b/drivers/net/ice/ice_fdir_filter.c
> index bd627e3aa8..bcf105d1bd 100644
> --- a/drivers/net/ice/ice_fdir_filter.c
> +++ b/drivers/net/ice/ice_fdir_filter.c
> @@ -107,6 +107,7 @@
>  	ICE_INSET_NAT_T_ESP_SPI)
> 
>  static struct ice_pattern_match_item ice_fdir_pattern_list[] = {
> +	{pattern_raw,					ICE_INSET_NONE,
> 	ICE_INSET_NONE,			ICE_INSET_NONE},
>  	{pattern_ethertype,				ICE_FDIR_INSET_ETH,
> 	ICE_INSET_NONE,			ICE_INSET_NONE},
>  	{pattern_eth_ipv4,				ICE_FDIR_INSET_ETH_IPV4,
> 	ICE_INSET_NONE,			ICE_INSET_NONE},
>  	{pattern_eth_ipv4_udp,				ICE_FDIR_INSET_ETH_IPV4_UDP,
> 	ICE_INSET_NONE,			ICE_INSET_NONE},
> @@ -1158,6 +1159,8 @@ ice_fdir_init(struct ice_adapter *ad)
>  	if (ret)
>  		return ret;
> 
> +	INIT_LIST_HEAD(&ad->fdir_prof_list);
> +
>  	parser = &ice_fdir_parser;
> 
>  	return ice_register_parser(parser, ad); @@ -1188,6 +1191,24 @@
> ice_fdir_is_tunnel_profile(enum ice_fdir_tunnel_type tunnel_type)
>  		return 0;
>  }
> 
> +static int
> +ice_fdir_add_del_raw(struct ice_pf *pf,
> +		     struct ice_fdir_filter_conf *filter,
> +		     bool add)
> +{
> +	struct ice_hw *hw = ICE_PF_TO_HW(pf);
> +
> +	unsigned char *pkt = (unsigned char *)pf->fdir.prg_pkt;
> +	rte_memcpy(pkt, filter->pkt_buf, filter->pkt_len);
> +
> +	struct ice_fltr_desc desc;
> +	memset(&desc, 0, sizeof(desc));
> +	filter->input.comp_report = ICE_FXD_FLTR_QW0_COMP_REPORT_SW;
> +	ice_fdir_get_prgm_desc(hw, &filter->input, &desc, add);
> +
> +	return ice_fdir_programming(pf, &desc); }
> +
>  static int
>  ice_fdir_add_del_filter(struct ice_pf *pf,
>  			struct ice_fdir_filter_conf *filter, @@ -1303,6 +1324,100 @@
> ice_fdir_create_filter(struct ice_adapter *ad,
>  	struct ice_fdir_fltr_pattern key;
>  	bool is_tun;
>  	int ret;
> +	int i;
> +
> +	if (filter->parser_ena) {
> +		struct ice_hw *hw = ICE_PF_TO_HW(pf);
> +
> +		int id = ice_find_first_bit(filter->prof->ptypes, UINT16_MAX);
> +		u16 ctrl_vsi = pf->fdir.fdir_vsi->idx;
> +		u16 main_vsi = pf->main_vsi->idx;
> +		bool fv_found = false;
> +		u16 vsi_num;
> +
> +		struct ice_fdir_prof_info *pi;
> +		LIST_FOR_EACH_ENTRY(pi, &ad->fdir_prof_list,
> +				    ice_fdir_prof_info, l_entry) {
> +			if (pi->ptg != hw->blk[ICE_BLK_FD].xlt1.t[id])
> +				continue;
> +			if (!pi->fdir_actived_cnt) {
> +				vsi_num = ice_get_hw_vsi_num(hw, ctrl_vsi);
> +				ret = ice_rem_prof_id_flow(hw, ICE_BLK_FD,
> +							   vsi_num, id);

Looks like you delay the "profile remove" until next call ice_fdir_add_del_filter, 
Why not just do this at the exact moment that "fdir_actived_cnt = 0"?



> +				if (ret)
> +					return -rte_errno;
> +
> +				vsi_num = ice_get_hw_vsi_num(hw, main_vsi);
> +				ret = ice_rem_prof_id_flow(hw, ICE_BLK_FD,
> +							   vsi_num, id);
> +				if (ret)
> +					return -rte_errno;
> +
> +				LIST_DEL(&pi->l_entry);
> +				continue;
> +			}
> +			for (i = 0; i < ICE_MAX_FV_WORDS; i++)
> +				if (pi->prof.fv[i].proto_id !=
> +				    filter->prof->fv[i].proto_id ||
> +				    pi->prof.fv[i].offset !=
> +				    filter->prof->fv[i].offset)
> +					break;
> +			if (i == ICE_MAX_FV_WORDS) {
> +				fv_found = true;
> +				pi->fdir_actived_cnt++;
> +				break;
> +			}
> +		}
> +
> +		if (!fv_found) {
> +			ret = ice_flow_set_hw_prof(hw, main_vsi, ctrl_vsi,
> +						   filter->prof, ICE_BLK_FD);
> +			if (ret)
> +				return -rte_errno;
> +		}
> +
> +		ret = ice_fdir_add_del_raw(pf, filter, true);
> +		if (ret)
> +			return -rte_errno;
> +
> +		if (!fv_found) {
> +			pi = (struct ice_fdir_prof_info *)
> +				ice_malloc(hw, sizeof(*pi));
> +			if (!pi)
> +				return ICE_ERR_NO_MEMORY;
> +
> +			memset(&pi->prof, 0, sizeof(pi->prof));
> +
> +			for (i = 0; i < filter->prof->fv_num; i++) {
> +				pi->prof.fv[i].proto_id =
> +					filter->prof->fv[i].proto_id;
> +				pi->prof.fv[i].offset =
> +					filter->prof->fv[i].offset;
> +				pi->prof.fv[i].spec = filter->prof->fv[i].spec;
> +				pi->prof.fv[i].msk = filter->prof->fv[i].msk;
> +			}
> +			pi->ptg = hw->blk[ICE_BLK_FD].xlt1.t[id];
> +			pi->fdir_actived_cnt = 1;
> +
> +			LIST_ADD(&pi->l_entry, &ad->fdir_prof_list);
> +		}
> +
> +		if (filter->mark_flag == 1)
> +			ice_fdir_rx_parsing_enable(ad, 1);
> +
> +		entry = rte_zmalloc("fdir_entry", sizeof(*entry), 0);
> +		if (!entry)
> +			return -rte_errno;
> +
> +		rte_memcpy(entry, filter, sizeof(*filter));
> +
> +		filter->prof = NULL;
> +		filter->pkt_buf = NULL;
> +
> +		flow->rule = entry;
> +
> +		return 0;
> +	}
> 
>  	ice_fdir_extract_fltr_key(&key, filter);
>  	node = ice_fdir_entry_lookup(fdir_info, &key); @@ -1397,6 +1512,44
> @@ ice_fdir_destroy_filter(struct ice_adapter *ad,
> 
>  	filter = (struct ice_fdir_filter_conf *)flow->rule;
> 
> +	if (filter->parser_ena) {
> +		struct ice_hw *hw = ICE_PF_TO_HW(pf);
> +
> +		ret = ice_fdir_add_del_raw(pf, filter, false);
> +		if (ret)
> +			return -rte_errno;
> +
> +		int id = ice_find_first_bit(filter->prof->ptypes, UINT16_MAX);
> +		int i;
> +		struct ice_fdir_prof_info *pi;
> +		LIST_FOR_EACH_ENTRY(pi, &ad->fdir_prof_list,
> +				    ice_fdir_prof_info, l_entry) {
> +			if (pi->ptg != hw->blk[ICE_BLK_FD].xlt1.t[id])
> +				continue;
> +			for (i = 0; i < ICE_MAX_FV_WORDS; i++)
> +				if (pi->prof.fv[i].proto_id !=
> +				    filter->prof->fv[i].proto_id ||
> +				    pi->prof.fv[i].offset !=
> +				    filter->prof->fv[i].offset)
> +					break;
> +			if (i == ICE_MAX_FV_WORDS) {
> +				pi->fdir_actived_cnt--;

Here you can remove the profile.

> +				break;
> +			}
> +		}

can all above part be moved into ice_fdir_add_del_filter, this make the add /delete implementation symmetric .

> +
> +		if (filter->mark_flag == 1)
> +			ice_fdir_rx_parsing_enable(ad, 0);
> +
> +		flow->rule = NULL;
> +		filter->prof = NULL;
> +		filter->pkt_buf = NULL;
> +
> +		rte_free(filter);
> +
> +		return 0;
> +	}
> +
>  	is_tun = ice_fdir_is_tunnel_profile(filter->tunnel_type);
> 
>  	if (filter->counter) {
> @@ -1675,6 +1828,7 @@ ice_fdir_parse_pattern(__rte_unused struct
> ice_adapter *ad,
>  	enum rte_flow_item_type l3 = RTE_FLOW_ITEM_TYPE_END;
>  	enum rte_flow_item_type l4 = RTE_FLOW_ITEM_TYPE_END;
>  	enum ice_fdir_tunnel_type tunnel_type =
> ICE_FDIR_TUNNEL_TYPE_NONE;
> +	const struct rte_flow_item_raw *raw_spec, *raw_mask;
>  	const struct rte_flow_item_eth *eth_spec, *eth_mask;
>  	const struct rte_flow_item_ipv4 *ipv4_spec, *ipv4_last, *ipv4_mask;
>  	const struct rte_flow_item_ipv6 *ipv6_spec, *ipv6_mask; @@ -1702,6
> +1856,9 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
>  	struct ice_fdir_extra *p_ext_data;
>  	struct ice_fdir_v4 *p_v4 = NULL;
>  	struct ice_fdir_v6 *p_v6 = NULL;
> +	struct ice_parser_result rslt;
> +	struct ice_parser *psr;
> +	uint8_t item_num = 0;
> 
>  	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
>  		if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN) @@ -1713,6
> +1870,7 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
>  		    item->type == RTE_FLOW_ITEM_TYPE_GTP_PSC) {
>  			is_outer = false;
>  		}
> +		item_num++;
>  	}
> 
>  	/* This loop parse flow pattern and distinguish Non-tunnel and tunnel
> @@ -1733,6 +1891,101 @@ ice_fdir_parse_pattern(__rte_unused struct
> ice_adapter *ad,
>  			    &input_set_i : &input_set_o;
> 
>  		switch (item_type) {
> +		case RTE_FLOW_ITEM_TYPE_RAW:
> +			raw_spec = item->spec;
> +			raw_mask = item->mask;
> +
> +			if (item_num != 1)
> +				break;
> +
> +			/* convert raw spec & mask from byte string to int */
> +			unsigned char *tmp_spec =
> +				(uint8_t *)(uintptr_t)raw_spec->pattern;
> +			unsigned char *tmp_mask =
> +				(uint8_t *)(uintptr_t)raw_mask->pattern;
> +			uint16_t udp_port = 0;
> +			uint16_t tmp_val = 0;
> +			uint8_t pkt_len = 0;
> +			uint8_t tmp = 0;
> +			int i, j;
> +
> +			pkt_len = strlen((char *)(uintptr_t)raw_spec->pattern);
> +			if (strlen((char *)(uintptr_t)raw_mask->pattern) !=
> +				pkt_len)
> +				return -rte_errno;
> +
> +			for (i = 0, j = 0; i < pkt_len; i += 2, j++) {
> +				tmp = tmp_spec[i];
> +				if (tmp >= 'a' && tmp <= 'f')
> +					tmp_val = tmp - 'a' + 10;
> +				if (tmp >= 'A' && tmp <= 'F')
> +					tmp_val = tmp - 'A' + 10;
> +				if (tmp >= '0' && tmp <= '9')
> +					tmp_val = tmp - '0';
> +
> +				tmp_val *= 16;
> +				tmp = tmp_spec[i + 1];
> +				if (tmp >= 'a' && tmp <= 'f')
> +					tmp_spec[j] = tmp_val + tmp - 'a' + 10;
> +				if (tmp >= 'A' && tmp <= 'F')
> +					tmp_spec[j] = tmp_val + tmp - 'A' + 10;
> +				if (tmp >= '0' && tmp <= '9')
> +					tmp_spec[j] = tmp_val + tmp - '0';
> +
> +				tmp = tmp_mask[i];
> +				if (tmp >= 'a' && tmp <= 'f')
> +					tmp_val = tmp - 'a' + 10;
> +				if (tmp >= 'A' && tmp <= 'F')
> +					tmp_val = tmp - 'A' + 10;
> +				if (tmp >= '0' && tmp <= '9')
> +					tmp_val = tmp - '0';
> +
> +				tmp_val *= 16;
> +				tmp = tmp_mask[i + 1];
> +				if (tmp >= 'a' && tmp <= 'f')
> +					tmp_mask[j] = tmp_val + tmp - 'a' + 10;
> +				if (tmp >= 'A' && tmp <= 'F')
> +					tmp_mask[j] = tmp_val + tmp - 'A' + 10;
> +				if (tmp >= '0' && tmp <= '9')
> +					tmp_mask[j] = tmp_val + tmp - '0';
> +			}
> +
> +			pkt_len /= 2;
> +
> +			if (ice_parser_create(&ad->hw, &psr))
> +				return -rte_errno;
> +			if (ice_get_open_tunnel_port(&ad->hw, TNL_VXLAN,
> +						     &udp_port))
> +				ice_parser_vxlan_tunnel_set(psr, udp_port,
> +							    true);
> +			if (ice_parser_run(psr, tmp_spec, pkt_len, &rslt))
> +				return -rte_errno;
> +			ice_parser_destroy(psr);
> +
> +			if (!tmp_mask)
> +				return -rte_errno;
> +
> +			filter->prof = (struct ice_parser_profile *)
> +				ice_malloc(&ad->hw, sizeof(*filter->prof));
> +			if (!filter->prof)
> +				return -ENOMEM;
> +
> +			if (ice_parser_profile_init(&rslt, tmp_spec, tmp_mask,
> +				pkt_len, ICE_BLK_FD, true, filter->prof))
> +				return -rte_errno;
> +
> +			u8 *pkt_buf = (u8 *)ice_malloc(&ad->hw, pkt_len + 1);
> +			if (!pkt_buf)
> +				return -ENOMEM;
> +			rte_memcpy(pkt_buf, tmp_spec, pkt_len);
> +			filter->pkt_buf = pkt_buf;
> +
> +			filter->pkt_len = pkt_len;
> +
> +			filter->parser_ena = true;
> +
> +			break;
> +
>  		case RTE_FLOW_ITEM_TYPE_ETH:
>  			flow_type = ICE_FLTR_PTYPE_NON_IP_L2;
>  			eth_spec = item->spec;
> @@ -2198,6 +2451,7 @@ ice_fdir_parse(struct ice_adapter *ad,
>  	struct ice_fdir_filter_conf *filter = &pf->fdir.conf;
>  	struct ice_pattern_match_item *item = NULL;
>  	uint64_t input_set;
> +	bool raw = false;
>  	int ret;
> 
>  	memset(filter, 0, sizeof(*filter));
> @@ -2213,7 +2467,13 @@ ice_fdir_parse(struct ice_adapter *ad,
>  	ret = ice_fdir_parse_pattern(ad, pattern, error, filter);
>  	if (ret)
>  		goto error;
> +
> +	if (item->pattern_list[0] == RTE_FLOW_ITEM_TYPE_RAW)
> +		raw = true;
> +
>  	input_set = filter->input_set_o | filter->input_set_i;
> +	input_set = raw ? ~input_set : input_set;
> +
>  	if (!input_set || filter->input_set_o &
>  	    ~(item->input_set_mask_o | ICE_INSET_ETHERTYPE) ||
>  	    filter->input_set_i & ~item->input_set_mask_i) { diff --git
> a/drivers/net/ice/ice_generic_flow.c b/drivers/net/ice/ice_generic_flow.c
> index 02f854666a..d3391c86c0 100644
> --- a/drivers/net/ice/ice_generic_flow.c
> +++ b/drivers/net/ice/ice_generic_flow.c
> @@ -65,6 +65,12 @@ enum rte_flow_item_type pattern_empty[] = {
>  	RTE_FLOW_ITEM_TYPE_END,
>  };
> 
> +/* raw */
> +enum rte_flow_item_type pattern_raw[] = {
> +	RTE_FLOW_ITEM_TYPE_RAW,
> +	RTE_FLOW_ITEM_TYPE_END,
> +};
> +
>  /* L2 */
>  enum rte_flow_item_type pattern_ethertype[] = {
>  	RTE_FLOW_ITEM_TYPE_ETH,
> @@ -2081,6 +2087,7 @@ struct ice_ptype_match {  };
> 
>  static struct ice_ptype_match ice_ptype_map[] = {
> +	{pattern_raw,					ICE_PTYPE_IPV4_PAY},
>  	{pattern_eth_ipv4,				ICE_PTYPE_IPV4_PAY},
>  	{pattern_eth_ipv4_udp,				ICE_PTYPE_IPV4_UDP_PAY},
>  	{pattern_eth_ipv4_tcp,				ICE_PTYPE_IPV4_TCP_PAY},
> diff --git a/drivers/net/ice/ice_generic_flow.h
> b/drivers/net/ice/ice_generic_flow.h
> index 8845a3e156..1b030c0466 100644
> --- a/drivers/net/ice/ice_generic_flow.h
> +++ b/drivers/net/ice/ice_generic_flow.h
> @@ -124,6 +124,9 @@
>  /* empty pattern */
>  extern enum rte_flow_item_type pattern_empty[];
> 
> +/* raw pattern */
> +extern enum rte_flow_item_type pattern_raw[];
> +
>  /* L2 */
>  extern enum rte_flow_item_type pattern_ethertype[];  extern enum
> rte_flow_item_type pattern_ethertype_vlan[];
> --
> 2.25.1