[PATCH] net/ice: fix flow engines order

[PATCH] net/ice: fix flow engines order

[Vladimir Medvedkin]

Reorder flow engine evaluation for RTE flow rule installation.
Currently, the driver evaluates rules with engines in the order they are
executed in the hardware. However, in this situation, some wildcarded flows
that are also supported by the ACL engine are installed as switch engine
rules, which scales poorly with each distinct wildcard mask configuration.

Fixes: fabc9e1322e2 ("net/ice: fix flows handling")
Cc: stable@dpdk.org

Signed-off-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com>
---
 drivers/net/intel/ice/ice_generic_flow.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/net/intel/ice/ice_generic_flow.c b/drivers/net/intel/ice/ice_generic_flow.c
index 5c34e0385f..f7a44223ac 100644
--- a/drivers/net/intel/ice/ice_generic_flow.c
+++ b/drivers/net/intel/ice/ice_generic_flow.c
@@ -2239,11 +2239,11 @@ static struct ice_flow_parser *get_flow_parser(uint32_t group)
 {
 	switch (group) {
 	case 0:
-		return &ice_switch_parser;
+		return &ice_fdir_parser;
 	case 1:
 		return &ice_acl_parser;
 	case 2:
-		return &ice_fdir_parser;
+		return &ice_switch_parser;
 	default:
 		return NULL;
 	}
@@ -2298,6 +2298,13 @@ ice_flow_process_filter(struct rte_eth_dev *dev,
 	}
 
 	for (int i = 0; i < ICE_FLOW_ENGINE_NB; i++) {
+		/**
+		 * Evaluate parsers in the following order:
+		 * FDIR - for exact match rules
+		 * ACL - for some subset of wildcard matching rules
+		 * Switch - for the rest. This engine is placed after ACL
+		 * because it scales worse than ACL for different wildcard masks.
+		 **/
 		parser = get_flow_parser(i);
 		if (parser == NULL) {
 			rte_flow_error_set(error, EINVAL,
-- 
2.43.0

[PATCH v2] net/ice: fix flow engines order

[Vladimir Medvedkin]

Reorder flow engine evaluation for RTE flow rule installation.
Currently, the driver evaluates rules with engines in the order they are
executed in the hardware. However, in this situation, some wildcarded flows
that are also supported by the ACL engine are installed as switch engine
rules, which scales poorly with each distinct wildcard mask configuration.

Fixes: fabc9e1322e2 ("net/ice: fix flows handling")
Cc: stable@dpdk.org

Signed-off-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com>
---
 drivers/net/intel/ice/ice_generic_flow.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/net/intel/ice/ice_generic_flow.c b/drivers/net/intel/ice/ice_generic_flow.c
index 5c34e0385f..4049157eab 100644
--- a/drivers/net/intel/ice/ice_generic_flow.c
+++ b/drivers/net/intel/ice/ice_generic_flow.c
@@ -2239,9 +2239,9 @@ static struct ice_flow_parser *get_flow_parser(uint32_t group)
 {
 	switch (group) {
 	case 0:
-		return &ice_switch_parser;
-	case 1:
 		return &ice_acl_parser;
+	case 1:
+		return &ice_switch_parser;
 	case 2:
 		return &ice_fdir_parser;
 	default:
@@ -2298,6 +2298,13 @@ ice_flow_process_filter(struct rte_eth_dev *dev,
 	}
 
 	for (int i = 0; i < ICE_FLOW_ENGINE_NB; i++) {
+		/**
+		 * Evaluate parsers in the following order:
+		 * ACL - for some subset of wildcard matching rules
+		 * Switch - This engine is placed after ACL because
+		 * it scales worse than ACL for different wildcard masks.
+		 * FDIR - for exact match rules
+		 **/
 		parser = get_flow_parser(i);
 		if (parser == NULL) {
 			rte_flow_error_set(error, EINVAL,
-- 
2.43.0