Re: [dpdk-dev] [dpdk-dev v3 2/4] net/iavf: add RSS configuration for VFs

[Jeff Guo <jia.guo@intel.com>]

hi, qi


On 4/13/2020 10:02 AM, Zhang, Qi Z wrote:
> Hi Jeff:
>
>> -----Original Message-----
>> From: Guo, Jia <jia.guo@intel.com>
>> Sent: Saturday, April 11, 2020 8:10 AM
>> To: Ye, Xiaolong <xiaolong.ye@intel.com>; Zhang, Qi Z <qi.z.zhang@intel.com>
>> Cc: dev@dpdk.org; Wu, Jingjing <jingjing.wu@intel.com>; Cao, Yahui
>> <yahui.cao@intel.com>; Su, Simei <simei.su@intel.com>; Guo, Jia
>> <jia.guo@intel.com>
>> Subject: [dpdk-dev v3 2/4] net/iavf: add RSS configuration for VFs
>>
>> The VF must be capable of configuring RSS. Add a virtchnl handler to parse a
>> specific RSS configuration, and process the configuration for VFs, such as add
>> or delete a RSS rule.
>>
>> Signed-off-by: Jeff Guo <jia.guo@intel.com>
>> ---
>> v3->v2:
>> 1.add doc in release note
>> 2.refine some naming base on virtchnl definition.
>> ---
>>   doc/guides/rel_notes/release_20_05.rst |    2 +
>>   drivers/net/iavf/Makefile              |    1 +
>>   drivers/net/iavf/iavf.h                |    2 +
>>   drivers/net/iavf/iavf_hash.c           | 1108
>> ++++++++++++++++++++++++
>>   drivers/net/iavf/iavf_vchnl.c          |   33 +-
>>   drivers/net/iavf/meson.build           |    1 +
>>   6 files changed, 1142 insertions(+), 5 deletions(-)  create mode 100644
>> drivers/net/iavf/iavf_hash.c
>>
>> diff --git a/doc/guides/rel_notes/release_20_05.rst
>> b/doc/guides/rel_notes/release_20_05.rst
>> index 4b81893ff..b5962d8e4 100644
>> --- a/doc/guides/rel_notes/release_20_05.rst
>> +++ b/doc/guides/rel_notes/release_20_05.rst
>> @@ -98,6 +98,8 @@ New Features
>>     Update the Intel iavf driver with new features and improvements,
>> including:
>>
>>     * Added generic filter support.
>> +  * Added advanced RSS configuration for VFs.
>> +
>>
>>   Removed Items
>>   -------------
>> diff --git a/drivers/net/iavf/Makefile b/drivers/net/iavf/Makefile index
>> 1bf0f26b5..7b0093a3e 100644
>> --- a/drivers/net/iavf/Makefile
>> +++ b/drivers/net/iavf/Makefile
>> @@ -24,6 +24,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_IAVF_PMD) +=
>> iavf_ethdev.c
>>   SRCS-$(CONFIG_RTE_LIBRTE_IAVF_PMD) += iavf_vchnl.c
>>   SRCS-$(CONFIG_RTE_LIBRTE_IAVF_PMD) += iavf_rxtx.c
>>   SRCS-$(CONFIG_RTE_LIBRTE_IAVF_PMD) += iavf_generic_flow.c
>> +SRCS-$(CONFIG_RTE_LIBRTE_IAVF_PMD) += iavf_hash.c
>>   ifeq ($(CONFIG_RTE_ARCH_X86), y)
>>   SRCS-$(CONFIG_RTE_LIBRTE_IAVF_PMD) += iavf_rxtx_vec_sse.c  endif diff
>> --git a/drivers/net/iavf/iavf.h b/drivers/net/iavf/iavf.h index
>> 78bdaff20..d813296d3 100644
>> --- a/drivers/net/iavf/iavf.h
>> +++ b/drivers/net/iavf/iavf.h
>> @@ -252,4 +252,6 @@ int iavf_config_promisc(struct iavf_adapter
>> *adapter, bool enable_unicast,  int iavf_add_del_eth_addr(struct
>> iavf_adapter *adapter,
>>   			 struct rte_ether_addr *addr, bool add);  int
>> iavf_add_del_vlan(struct iavf_adapter *adapter, uint16_t vlanid, bool add);
>> +int iavf_add_del_rss_cfg(struct iavf_adapter *adapter,
>> +			 struct virtchnl_rss_cfg *rss_cfg, bool add);
>>   #endif /* _IAVF_ETHDEV_H_ */
>> diff --git a/drivers/net/iavf/iavf_hash.c b/drivers/net/iavf/iavf_hash.c new
>> file mode 100644 index 000000000..f831738cc
>> --- /dev/null
>> +++ b/drivers/net/iavf/iavf_hash.c
>> @@ -0,0 +1,1108 @@
>> +/* SPDX-License-Identifier: BSD-3-Clause
>> + * Copyright(c) 2020 Intel Corporation
>> + */
>> +
>> +
>> +static uint64_t
>> +iavf_rss_hf_refine(uint64_t rss_hf, const struct rte_flow_item pattern[],
>> +		   const struct rte_flow_action *action,
>> +		   struct rte_flow_error *error)
>> +{
>> +	const struct rte_flow_item *item;
>> +
>> +	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++)
>> {
>> +		if (item->type == RTE_FLOW_ITEM_TYPE_GTP_PSC) {
>> +			const struct rte_flow_action_rss *rss = action->conf;
>> +			const struct rte_flow_item_gtp_psc *psc = item->spec;
>> +
>> +			if (psc && ((psc->pdu_type == GTP_EH_PDU_LINK_UP &&
>> +				     (rss->types & ETH_RSS_L3_SRC_ONLY)) ||
> Why we need to check SRC_ONLY here, the purpose of the function is to refine the rss_hf,
> Should we'd better input set check to the stage " Find matched proto hdrs according to hash type" later


agree with you the point that we don't need to check the SRC/DST here, 
because we could exactly check it later.


>> +				    (!psc->pdu_type &&
> Better use GTP_EH_PDU_LINK_DOWN


make sense.


>> +				     (rss->types & ETH_RSS_L3_DST_ONLY)))) {
>> +				rss_hf |= ETH_RSS_GTPU;
>> +			} else {
>> +				rte_flow_error_set(error, EINVAL,
>> +						   RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
>> +						   pattern,
>> +						   "Invalid input set");
>> +				return -rte_errno;
>> +			}
>> +		}
>> +	}
>> +
>> +	return rss_hf;
>> +}
>> +
>> +static int
>> +iavf_hash_parse_action(struct iavf_pattern_match_item
>> *pattern_match_item,
>> +		       const struct rte_flow_item pattern[],
>> +		       const struct rte_flow_action actions[],
>> +		       void **meta, struct rte_flow_error *error) {
>> +	struct iavf_rss_meta *rss_meta = (struct iavf_rss_meta *)*meta;
>> +	struct rss_type_match_hdr *m = (struct rss_type_match_hdr *)
>> +				(pattern_match_item->meta);
>> +	uint32_t type_list_len = RTE_DIM(iavf_hash_type_list);
>> +	struct iavf_hash_match_type *type_match_item;
>> +	enum rte_flow_action_type action_type;
>> +	const struct rte_flow_action_rss *rss;
>> +	const struct rte_flow_action *action;
>> +	uint64_t rss_hf;
>> +	uint16_t i;
>> +	bool item_found = false;
>> +
>> +	/* Supported action is RSS. */
>> +	for (action = actions; action->type !=
>> +		RTE_FLOW_ACTION_TYPE_END; action++) {
>> +		action_type = action->type;
>> +		switch (action_type) {
>> +		case RTE_FLOW_ACTION_TYPE_RSS:
>> +			rss = action->conf;
>> +			rss_hf = rss->types;
>> +
>> +			/**
>> +			 * Check simultaneous use of SRC_ONLY and DST_ONLY
>> +			 * of the same level.
>> +			 */
>> +			rss_hf = rte_eth_rss_hf_refine(rss_hf);
>> +
>> +			/**
>> +			 * Check the item spec with the rss action and
>> +			 * refine rss hash field.
>> +			 */
>> +			rss_hf = iavf_rss_hf_refine(rss_hf, pattern, action,
>> +						    error);
>> +
>> +			/* Check if pattern is empty. */
>> +			if (pattern_match_item->pattern_list !=
>> +				iavf_pattern_empty && rss->func ==
>> +				RTE_ETH_HASH_FUNCTION_SIMPLE_XOR)
>> +				return rte_flow_error_set(error, ENOTSUP,
>> +					RTE_FLOW_ERROR_TYPE_ACTION, action,
>> +					"Not supported flow");
>> +
>> +			/* Check if rss types match pattern. */
>> +			if (rss->func != RTE_ETH_HASH_FUNCTION_SIMPLE_XOR) {
>> +				if (((rss_hf & ETH_RSS_IPV4) !=
>> +					m->eth_rss_hint) &&
>> +				    ((rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) !=
>> +					m->eth_rss_hint) &&
>> +				    ((rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) !=
>> +					m->eth_rss_hint) &&
>> +				    ((rss_hf & ETH_RSS_NONFRAG_IPV4_SCTP) !=
>> +					m->eth_rss_hint) &&
>> +				    ((rss_hf & ETH_RSS_IPV6) !=
>> +					m->eth_rss_hint) &&
>> +				    ((rss_hf & ETH_RSS_NONFRAG_IPV6_UDP) !=
>> +					m->eth_rss_hint) &&
>> +				    ((rss_hf & ETH_RSS_NONFRAG_IPV6_TCP) !=
>> +					m->eth_rss_hint) &&
>> +				    ((rss_hf & ETH_RSS_NONFRAG_IPV6_SCTP) !=
>> +					m->eth_rss_hint))
>> +					return rte_flow_error_set(error,
>> +					ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
>> +					action, "Not supported RSS types");
>> +			}
> Why we need above check ? what happens if the input set does not include any IP layer for example session id of PFCP?
>
> ......


PFCP is also include IP later, but yes, it should support for much case 
case, such as only ether.


>
>> +static int
>> +iavf_hash_create(__rte_unused struct iavf_adapter *ad,
>> +		 __rte_unused struct rte_flow *flow, void *meta,
>> +		 __rte_unused struct rte_flow_error *error) {
>> +	struct iavf_rss_meta *rss_meta = (struct iavf_rss_meta *)meta;
>> +	struct virtchnl_rss_cfg *rss_cfg;
>> +	int ret = 0;
>> +
>> +	rss_cfg = rte_zmalloc("iavf rss rule",
>> +			      sizeof(struct virtchnl_rss_cfg), 0);
>> +	if (!rss_cfg) {
>> +		rte_flow_error_set(error, EINVAL,
>> +				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
>> +				   "No memory for rss rule");
>> +		return -ENOMEM;
>> +	}
>> +
>> +	rss_cfg->proto_hdrs = *rss_meta->proto_hdrs;
>> +	rss_cfg->rss_algorithm = rss_meta->hash_function;
>> +
>> +	ret = iavf_add_del_rss_cfg(ad, rss_cfg, true);
>> +	if (!ret) {
>> +		flow->rule = rss_cfg;
>> +	} else {
>> +		PMD_DRV_LOG(ERR, "fail to add RSS configure");
>> +		rte_free(rss_cfg);
>> +	}
> The "meta" be created by iavf_hash_parse_pattern_action is assumed to be freed here,
>

sure.