DPDK patches and discussions
 help / color / mirror / Atom feed
* [PATCH 00/16] net/hns3: some code refactor for hns3 RSS
@ 2023-03-10  9:35 Dongdong Liu
  2023-03-10  9:35 ` [PATCH 01/16] net/hns3: fix possible truncation of hash key when config Dongdong Liu
                   ` (16 more replies)
  0 siblings, 17 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

This patchset is to do some code refactor for hns3 RSS.
1. [PATCH 01/16]-[PATCH 02/16] fix some hns3 RSS bug.
2. [PATCH 03/16]-[PATCH 14/16] refactoring hns3 RSS code.
3. [PATCH 15/16]-[PATCH 16/16] reimplement hash flow function.


Huisong Li (16):
  net/hns3: fix possible truncation of hash key when config
  net/hns3: fix possible truncation of redirection table
  net/hns3: use hardware config to report hash key
  net/hns3: use hardware config to report hash types
  net/hns3: use hardware config to report redirection table
  net/hns3: separate the setting of hash algorithm
  net/hns3: separate the setting of hash key
  net/hns3: separate the setting of redirection table
  net/hns3: separate the setting of RSS types
  net/hns3: separate the setting and clearing of RSS rule
  net/hns3: use new RSS rule to configure hardware
  net/hns3: save hash algo to RSS filter list node
  net/hns3: adding queue buffer size hash rule allowed
  net/hns3: separate rte flow RSS config from hns3 rss conf
  net/hns3: reimplement hash flow function
  net/hns3: add the verification of RSS types

 drivers/net/hns3/hns3_cmd.h    |   1 +
 drivers/net/hns3/hns3_ethdev.h |   9 -
 drivers/net/hns3/hns3_flow.c   | 985 +++++++++++++++++++++++----------
 drivers/net/hns3/hns3_flow.h   |  21 +-
 drivers/net/hns3/hns3_rss.c    | 645 +++++++++++++++------
 drivers/net/hns3/hns3_rss.h    | 130 ++++-
 6 files changed, 1278 insertions(+), 513 deletions(-)

--
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 01/16] net/hns3: fix possible truncation of hash key when config
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10 19:36   ` Ferruh Yigit
  2023-03-10  9:35 ` [PATCH 02/16] net/hns3: fix possible truncation of redirection table Dongdong Liu
                   ` (15 subsequent siblings)
  16 siblings, 1 reply; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

The hash key length of hns3 driver is obtained from firmware. If the
length isn't a multiple of HNS3_RSS_HASH_KEY_NUM (16), the last part
of hash key will be truncated.

Fixes: 4a7384e3c34d ("net/hns3: refactor set RSS hash algorithm and key interface")
Fixes: c37ca66f2b27 ("net/hns3: support RSS")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_rss.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index d6e0754273..2011c18b9b 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -301,7 +301,8 @@ hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
 		req->hash_config |= (hash_algo & HNS3_RSS_HASH_ALGO_MASK);
 		req->hash_config |= (idx << HNS3_RSS_HASH_KEY_OFFSET_B);
 
-		if (idx == max_bd_num - 1)
+		if (idx == max_bd_num - 1 &&
+		    (key_len % HNS3_RSS_HASH_KEY_NUM) != 0)
 			cur_key_size = key_len % HNS3_RSS_HASH_KEY_NUM;
 		else
 			cur_key_size = HNS3_RSS_HASH_KEY_NUM;
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 02/16] net/hns3: fix possible truncation of redirection table
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
  2023-03-10  9:35 ` [PATCH 01/16] net/hns3: fix possible truncation of hash key when config Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10 19:36   ` Ferruh Yigit
  2023-03-10  9:35 ` [PATCH 03/16] net/hns3: use hardware config to report hash key Dongdong Liu
                   ` (14 subsequent siblings)
  16 siblings, 1 reply; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

The size of the redirection table is obtained from firmware. If the size
isn't a multiple of HNS3_RSS_CFG_TBL_SIZE, the redirection table from
user will be truncated.

Fixes: c37ca66f2b27 ("net/hns3: support RSS")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_rss.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index 2011c18b9b..f6d677ade8 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -329,22 +329,28 @@ int
 hns3_set_rss_indir_table(struct hns3_hw *hw, uint16_t *indir, uint16_t size)
 {
 	struct hns3_rss_indirection_table_cmd *req;
+	uint16_t max_bd_num, cfg_tbl_size;
+	uint8_t qid_msb_off, qid_msb_val;
 	struct hns3_cmd_desc desc;
-	uint8_t qid_msb_off;
-	uint8_t qid_msb_val;
 	uint16_t q_id;
 	uint16_t i, j;
 	int ret;
 
 	req = (struct hns3_rss_indirection_table_cmd *)desc.data;
-
-	for (i = 0; i < size / HNS3_RSS_CFG_TBL_SIZE; i++) {
+	max_bd_num = DIV_ROUND_UP(size, HNS3_RSS_CFG_TBL_SIZE);
+	for (i = 0; i < max_bd_num; i++) {
 		hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RSS_INDIR_TABLE,
 					  false);
 		req->start_table_index =
 				rte_cpu_to_le_16(i * HNS3_RSS_CFG_TBL_SIZE);
 		req->rss_set_bitmap = rte_cpu_to_le_16(HNS3_RSS_SET_BITMAP_MSK);
-		for (j = 0; j < HNS3_RSS_CFG_TBL_SIZE; j++) {
+
+		if (i == max_bd_num - 1 && (size % HNS3_RSS_CFG_TBL_SIZE) != 0)
+			cfg_tbl_size = size % HNS3_RSS_CFG_TBL_SIZE;
+		else
+			cfg_tbl_size = HNS3_RSS_CFG_TBL_SIZE;
+
+		for (j = 0; j < cfg_tbl_size; j++) {
 			q_id = indir[i * HNS3_RSS_CFG_TBL_SIZE + j];
 			req->rss_result_l[j] = q_id & 0xff;
 
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 03/16] net/hns3: use hardware config to report hash key
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
  2023-03-10  9:35 ` [PATCH 01/16] net/hns3: fix possible truncation of hash key when config Dongdong Liu
  2023-03-10  9:35 ` [PATCH 02/16] net/hns3: fix possible truncation of redirection table Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 04/16] net/hns3: use hardware config to report hash types Dongdong Liu
                   ` (13 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Currently, hns3_dev_rss_hash_conf_get() interface reports RSS key from
the key maintained by driver. It's better to report the key from hardware,
which is more realistic.

Fixes: c37ca66f2b27 ("net/hns3: support RSS")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_rss.c | 53 ++++++++++++++++++++++++++++++++++++-
 drivers/net/hns3/hns3_rss.h |  3 ++-
 2 files changed, 54 insertions(+), 2 deletions(-)

diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index f6d677ade8..a046cc31d1 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -321,6 +321,48 @@ hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
 	return 0;
 }
 
+int
+hns3_rss_get_algo_key(struct hns3_hw *hw,  uint8_t *hash_algo,
+		      uint8_t *key, uint8_t key_len)
+{
+	struct hns3_rss_generic_config_cmd *req;
+	struct hns3_cmd_desc desc;
+	uint16_t cur_key_size;
+	uint16_t max_bd_num;
+	uint8_t *cur_key;
+	uint16_t idx;
+	int ret;
+
+	req = (struct hns3_rss_generic_config_cmd *)desc.data;
+	max_bd_num = DIV_ROUND_UP(key_len, HNS3_RSS_HASH_KEY_NUM);
+	for (idx = 0; idx < max_bd_num; idx++) {
+		hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RSS_GENERIC_CONFIG,
+					  true);
+
+		req->hash_config |= (idx << HNS3_RSS_HASH_KEY_OFFSET_B);
+		ret = hns3_cmd_send(hw, &desc, 1);
+		if (ret) {
+			hns3_err(hw, "fail to obtain RSS algo and key from firmware, ret = %d",
+				 ret);
+			return ret;
+		}
+
+		if (idx == 0)
+			*hash_algo = req->hash_config & HNS3_RSS_HASH_ALGO_MASK;
+
+		if (idx == max_bd_num - 1 &&
+		    (key_len % HNS3_RSS_HASH_KEY_NUM) != 0)
+			cur_key_size = key_len % HNS3_RSS_HASH_KEY_NUM;
+		else
+			cur_key_size = HNS3_RSS_HASH_KEY_NUM;
+
+		cur_key = key + idx * HNS3_RSS_HASH_KEY_NUM;
+		memcpy(cur_key, req->hash_key, cur_key_size);
+	}
+
+	return 0;
+}
+
 /*
  * rss_indirection_table command function, opcode:0x0D07.
  * Used to configure the indirection table of rss.
@@ -549,13 +591,22 @@ hns3_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
 	struct hns3_adapter *hns = dev->data->dev_private;
 	struct hns3_hw *hw = &hns->hw;
 	struct hns3_rss_conf *rss_cfg = &hw->rss_info;
+	uint8_t hash_algo;
+	int ret;
 
 	rte_spinlock_lock(&hw->lock);
 	rss_conf->rss_hf = rss_cfg->conf.types;
 
 	/* Get the RSS Key required by the user */
 	if (rss_conf->rss_key && rss_conf->rss_key_len >= hw->rss_key_size) {
-		memcpy(rss_conf->rss_key, rss_cfg->key, hw->rss_key_size);
+		ret = hns3_rss_get_algo_key(hw, &hash_algo, rss_conf->rss_key,
+					    hw->rss_key_size);
+		if (ret != 0) {
+			rte_spinlock_unlock(&hw->lock);
+			hns3_err(hw, "obtain hash algo and key failed, ret = %d",
+				 ret);
+			return ret;
+		}
 		rss_conf->rss_key_len = hw->rss_key_size;
 	}
 	rte_spinlock_unlock(&hw->lock);
diff --git a/drivers/net/hns3/hns3_rss.h b/drivers/net/hns3/hns3_rss.h
index d6f81996f4..be0141f602 100644
--- a/drivers/net/hns3/hns3_rss.h
+++ b/drivers/net/hns3/hns3_rss.h
@@ -112,6 +112,7 @@ void hns3_rss_uninit(struct hns3_adapter *hns);
 int hns3_set_rss_tuple_by_rss_hf(struct hns3_hw *hw, uint64_t rss_hf);
 int hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
 			  const uint8_t *key, uint8_t key_len);
-
+int hns3_rss_get_algo_key(struct hns3_hw *hw,  uint8_t *hash_algo,
+			  uint8_t *key, uint8_t key_len);
 
 #endif /* HNS3_RSS_H */
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 04/16] net/hns3: use hardware config to report hash types
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (2 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 03/16] net/hns3: use hardware config to report hash key Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 05/16] net/hns3: use hardware config to report redirection table Dongdong Liu
                   ` (12 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Use the configuration in hardware to report hash types instead
of data maintained in software.

Fixes: c37ca66f2b27 ("net/hns3: support RSS")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_rss.c | 260 ++++++++++++++++++++++++++++--------
 drivers/net/hns3/hns3_rss.h |   1 +
 2 files changed, 208 insertions(+), 53 deletions(-)

diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index a046cc31d1..21266d64b7 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -70,6 +70,17 @@ enum hns3_tuple_field {
 	HNS3_RSS_FIELD_IPV6_FRAG_IP_S
 };
 
+#define HNS3_RSS_TUPLE_IPV4_TCP_M	GENMASK(3, 0)
+#define HNS3_RSS_TUPLE_IPV4_UDP_M	GENMASK(11, 8)
+#define HNS3_RSS_TUPLE_IPV4_SCTP_M	GENMASK(20, 16)
+#define HNS3_RSS_TUPLE_IPV4_NONF_M	GENMASK(25, 24)
+#define HNS3_RSS_TUPLE_IPV4_FLAG_M	GENMASK(27, 26)
+#define HNS3_RSS_TUPLE_IPV6_TCP_M	GENMASK(35, 32)
+#define HNS3_RSS_TUPLE_IPV6_UDP_M	GENMASK(43, 40)
+#define HNS3_RSS_TUPLE_IPV6_SCTP_M	GENMASK(52, 48)
+#define HNS3_RSS_TUPLE_IPV6_NONF_M	GENMASK(57, 56)
+#define HNS3_RSS_TUPLE_IPV6_FLAG_M	GENMASK(59, 58)
+
 enum hns3_rss_tuple_type {
 	HNS3_RSS_IP_TUPLE,
 	HNS3_RSS_IP_L4_TUPLE,
@@ -79,200 +90,249 @@ static const struct {
 	uint64_t rss_types;
 	uint16_t tuple_type;
 	uint64_t rss_field;
+	uint64_t tuple_mask;
 } hns3_set_tuple_table[] = {
 	/* IPV4-FRAG */
 	{ RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_S),
+	  HNS3_RSS_TUPLE_IPV4_FLAG_M },
 	{ RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_FLAG_M },
 	{ RTE_ETH_RSS_FRAG_IPV4,
 	  HNS3_RSS_IP_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_FLAG_M },
 
 	/* IPV4 */
 	{ RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_S),
+	  HNS3_RSS_TUPLE_IPV4_NONF_M },
 	{ RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_NONF_M },
 	{ RTE_ETH_RSS_IPV4,
 	  HNS3_RSS_IP_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_NONF_M },
 
 	/* IPV4-OTHER */
 	{ RTE_ETH_RSS_NONFRAG_IPV4_OTHER | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_S),
+	  HNS3_RSS_TUPLE_IPV4_NONF_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_OTHER | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_NONF_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
 	  HNS3_RSS_IP_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_NONF_M },
 
 	/* IPV4-TCP */
 	{ RTE_ETH_RSS_NONFRAG_IPV4_TCP | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_IP_S),
+	  HNS3_RSS_TUPLE_IPV4_TCP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_TCP | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_TCP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_TCP | RTE_ETH_RSS_L4_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_S),
+	  HNS3_RSS_TUPLE_IPV4_TCP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_TCP | RTE_ETH_RSS_L4_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_D),
+	  HNS3_RSS_TUPLE_IPV4_TCP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_TCP,
 	  HNS3_RSS_IP_L4_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_IP_S) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_IP_D) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_D),
+	  HNS3_RSS_TUPLE_IPV4_TCP_M },
 
 	/* IPV4-UDP */
 	{ RTE_ETH_RSS_NONFRAG_IPV4_UDP | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_IP_S),
+	  HNS3_RSS_TUPLE_IPV4_UDP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_UDP | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_UDP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_UDP | RTE_ETH_RSS_L4_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_S),
+	  HNS3_RSS_TUPLE_IPV4_UDP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_UDP | RTE_ETH_RSS_L4_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_D),
+	  HNS3_RSS_TUPLE_IPV4_UDP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_UDP,
 	  HNS3_RSS_IP_L4_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_IP_S) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_IP_D) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_D),
+	  HNS3_RSS_TUPLE_IPV4_UDP_M },
 
 	/* IPV4-SCTP */
 	{ RTE_ETH_RSS_NONFRAG_IPV4_SCTP | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_S),
+	  HNS3_RSS_TUPLE_IPV4_SCTP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_SCTP | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_D),
+	  HNS3_RSS_TUPLE_IPV4_SCTP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_SCTP | RTE_ETH_RSS_L4_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_S),
+	  HNS3_RSS_TUPLE_IPV4_SCTP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_SCTP | RTE_ETH_RSS_L4_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_D),
+	  HNS3_RSS_TUPLE_IPV4_SCTP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV4_SCTP,
 	  HNS3_RSS_IP_L4_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_S) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_D) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_S) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_D) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_VER) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_VER),
+	  HNS3_RSS_TUPLE_IPV4_SCTP_M },
 
 	/* IPV6-FRAG */
 	{ RTE_ETH_RSS_FRAG_IPV6 | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_FRAG_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_FRAG_IP_S),
+	  HNS3_RSS_TUPLE_IPV6_FLAG_M },
 	{ RTE_ETH_RSS_FRAG_IPV6 | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_FRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_FRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_FLAG_M },
 	{ RTE_ETH_RSS_FRAG_IPV6,
 	  HNS3_RSS_IP_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_FRAG_IP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_FRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_FRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_FLAG_M },
 
 	/* IPV6 */
 	{ RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_S),
+	  HNS3_RSS_TUPLE_IPV6_NONF_M },
 	{ RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_NONF_M },
 	{ RTE_ETH_RSS_IPV6,
 	  HNS3_RSS_IP_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_NONF_M },
 
 	/* IPV6-OTHER */
 	{ RTE_ETH_RSS_NONFRAG_IPV6_OTHER | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_S),
+	  HNS3_RSS_TUPLE_IPV6_NONF_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_OTHER | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_NONF_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
 	  HNS3_RSS_IP_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_NONF_M },
 
 	/* IPV6-TCP */
 	{ RTE_ETH_RSS_NONFRAG_IPV6_TCP | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_IP_S),
+	  HNS3_RSS_TUPLE_IPV6_TCP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_TCP | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_TCP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_TCP | RTE_ETH_RSS_L4_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_S),
+	  HNS3_RSS_TUPLE_IPV6_TCP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_TCP | RTE_ETH_RSS_L4_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_D),
+	  HNS3_RSS_TUPLE_IPV6_TCP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_TCP,
 	  HNS3_RSS_IP_L4_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_IP_S) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_IP_D) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_D),
+	  HNS3_RSS_TUPLE_IPV6_TCP_M },
 
 	/* IPV6-UDP */
 	{ RTE_ETH_RSS_NONFRAG_IPV6_UDP | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_IP_S),
+	  HNS3_RSS_TUPLE_IPV6_UDP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_UDP | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_UDP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_UDP | RTE_ETH_RSS_L4_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_S),
+	  HNS3_RSS_TUPLE_IPV6_UDP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_UDP | RTE_ETH_RSS_L4_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_D),
+	  HNS3_RSS_TUPLE_IPV6_UDP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_UDP,
 	  HNS3_RSS_IP_L4_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_IP_S) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_IP_D) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_D),
+	  HNS3_RSS_TUPLE_IPV6_UDP_M },
 
 	/* IPV6-SCTP */
 	{ RTE_ETH_RSS_NONFRAG_IPV6_SCTP | RTE_ETH_RSS_L3_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_S),
+	  HNS3_RSS_TUPLE_IPV6_SCTP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_SCTP | RTE_ETH_RSS_L3_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_D),
+	  HNS3_RSS_TUPLE_IPV6_SCTP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_SCTP | RTE_ETH_RSS_L4_SRC_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_S) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_S),
+	  HNS3_RSS_TUPLE_IPV6_SCTP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_SCTP | RTE_ETH_RSS_L4_DST_ONLY,
 	  HNS3_RSS_IP_L4_TUPLE,
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_D) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_D),
+	  HNS3_RSS_TUPLE_IPV6_SCTP_M },
 	{ RTE_ETH_RSS_NONFRAG_IPV6_SCTP,
 	  HNS3_RSS_IP_L4_TUPLE,
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_S) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_D) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_D) |
 	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_S) |
-	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_VER) },
+	  BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_VER),
+	  HNS3_RSS_TUPLE_IPV6_SCTP_M },
 };
 
 /*
@@ -575,6 +635,96 @@ hns3_dev_rss_hash_update(struct rte_eth_dev *dev,
 	return ret;
 }
 
+int
+hns3_get_rss_tuple_field(struct hns3_hw *hw, uint64_t *tuple_fields)
+{
+	struct hns3_rss_input_tuple_cmd *req;
+	struct hns3_cmd_desc desc;
+	int ret;
+
+	hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RSS_INPUT_TUPLE, true);
+	req = (struct hns3_rss_input_tuple_cmd *)desc.data;
+	ret = hns3_cmd_send(hw, &desc, 1);
+	if (ret != 0) {
+		hns3_err(hw, "fail to get RSS hash tuple fields from firmware, ret = %d",
+			 ret);
+		return ret;
+	}
+
+	*tuple_fields = rte_le_to_cpu_64(req->tuple_field);
+
+	return 0;
+}
+
+static uint64_t
+hns3_rss_tuple_fields_to_rss_hf(struct hns3_hw *hw, uint64_t tuple_fields)
+{
+	uint64_t ipv6_sctp_l4_mask =
+				BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_D) |
+				BIT_ULL(HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_S);
+	uint64_t rss_hf = 0;
+	uint64_t tuple_mask;
+	uint32_t i;
+
+	for (i = 0; i < RTE_DIM(hns3_set_tuple_table); i++) {
+		tuple_mask = hns3_set_tuple_table[i].tuple_mask;
+		/*
+		 * The RSS hash of the packet type is disabled if its tuples is
+		 * zero.
+		 */
+		if ((tuple_fields & tuple_mask) == 0)
+			continue;
+
+		/*
+		 * Some hardware don't support to use src/dst port fields to
+		 * hash for IPV6-SCTP packet.
+		 */
+		if ((hns3_set_tuple_table[i].rss_types &
+					RTE_ETH_RSS_NONFRAG_IPV6_SCTP) &&
+		    !hw->rss_info.ipv6_sctp_offload_supported)
+			tuple_mask &= ~ipv6_sctp_l4_mask;
+
+		/*
+		 * The framework (ethdev ops) or driver (rte flow API) ensure
+		 * that both L3_SRC/DST_ONLY and L4_SRC/DST_ONLY cannot be set
+		 * to driver at the same time. But if user doesn't specify
+		 * anything L3/L4_SRC/DST_ONLY, driver enables all tuple fields.
+		 * In this case, driver should not report L3/L4_SRC/DST_ONLY.
+		 */
+		if ((tuple_fields & tuple_mask) == tuple_mask) {
+			/* Skip the item enabled part tuples. */
+			if ((tuple_fields & hns3_set_tuple_table[i].rss_field) !=
+					tuple_mask)
+				continue;
+
+			rss_hf |= hns3_set_tuple_table[i].rss_types;
+			continue;
+		}
+
+		/* Match the item enabled part tuples.*/
+		if ((tuple_fields & hns3_set_tuple_table[i].rss_field) ==
+					hns3_set_tuple_table[i].rss_field)
+			rss_hf |= hns3_set_tuple_table[i].rss_types;
+	}
+
+	return rss_hf;
+}
+
+static int
+hns3_rss_hash_get_rss_hf(struct hns3_hw *hw, uint64_t *rss_hf)
+{
+	uint64_t tuple_fields;
+	int ret;
+
+	ret = hns3_get_rss_tuple_field(hw, &tuple_fields);
+	if (ret != 0)
+		return ret;
+
+	*rss_hf = hns3_rss_tuple_fields_to_rss_hf(hw, tuple_fields);
+
+	return 0;
+}
+
 /*
  * Get rss key and rss_hf types set of RSS hash configuration.
  * @param dev
@@ -590,28 +740,32 @@ hns3_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
 {
 	struct hns3_adapter *hns = dev->data->dev_private;
 	struct hns3_hw *hw = &hns->hw;
-	struct hns3_rss_conf *rss_cfg = &hw->rss_info;
 	uint8_t hash_algo;
 	int ret;
 
 	rte_spinlock_lock(&hw->lock);
-	rss_conf->rss_hf = rss_cfg->conf.types;
+	ret = hns3_rss_hash_get_rss_hf(hw, &rss_conf->rss_hf);
+	if (ret != 0) {
+		hns3_err(hw, "obtain hash tuples failed, ret = %d", ret);
+		goto out;
+	}
 
 	/* Get the RSS Key required by the user */
 	if (rss_conf->rss_key && rss_conf->rss_key_len >= hw->rss_key_size) {
 		ret = hns3_rss_get_algo_key(hw, &hash_algo, rss_conf->rss_key,
 					    hw->rss_key_size);
 		if (ret != 0) {
-			rte_spinlock_unlock(&hw->lock);
 			hns3_err(hw, "obtain hash algo and key failed, ret = %d",
 				 ret);
-			return ret;
+			goto out;
 		}
 		rss_conf->rss_key_len = hw->rss_key_size;
 	}
+
+out:
 	rte_spinlock_unlock(&hw->lock);
 
-	return 0;
+	return ret;
 }
 
 /*
diff --git a/drivers/net/hns3/hns3_rss.h b/drivers/net/hns3/hns3_rss.h
index be0141f602..17473e70e2 100644
--- a/drivers/net/hns3/hns3_rss.h
+++ b/drivers/net/hns3/hns3_rss.h
@@ -110,6 +110,7 @@ int hns3_rss_reset_indir_table(struct hns3_hw *hw);
 int hns3_config_rss(struct hns3_adapter *hns);
 void hns3_rss_uninit(struct hns3_adapter *hns);
 int hns3_set_rss_tuple_by_rss_hf(struct hns3_hw *hw, uint64_t rss_hf);
+int hns3_get_rss_tuple_field(struct hns3_hw *hw, uint64_t *tuple_fields);
 int hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
 			  const uint8_t *key, uint8_t key_len);
 int hns3_rss_get_algo_key(struct hns3_hw *hw,  uint8_t *hash_algo,
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 05/16] net/hns3: use hardware config to report redirection table
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (3 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 04/16] net/hns3: use hardware config to report hash types Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 06/16] net/hns3: separate the setting of hash algorithm Dongdong Liu
                   ` (11 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Currently, reta_query() API reports the redirection table from software.
This patch uses the one in hardware to report.

Fixes: c37ca66f2b27 ("net/hns3: support RSS")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_cmd.h |  1 +
 drivers/net/hns3/hns3_rss.c | 65 ++++++++++++++++++++++++++++++++++---
 2 files changed, 62 insertions(+), 4 deletions(-)

diff --git a/drivers/net/hns3/hns3_cmd.h b/drivers/net/hns3/hns3_cmd.h
index 994dfc48cc..eb394c9dec 100644
--- a/drivers/net/hns3/hns3_cmd.h
+++ b/drivers/net/hns3/hns3_cmd.h
@@ -606,6 +606,7 @@ struct hns3_rss_input_tuple_cmd {
 #define HNS3_RSS_CFG_TBL_SIZE_H		4
 #define HNS3_RSS_CFG_TBL_BW_H		2
 #define HNS3_RSS_CFG_TBL_BW_L		8
+#define HNS3_RSS_CFG_TBL_BW_H_M		0x3
 
 /* Configure the indirection table, opcode:0x0D07 */
 struct hns3_rss_indirection_table_cmd {
diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index 21266d64b7..fe9ad609b7 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -480,6 +480,54 @@ hns3_set_rss_indir_table(struct hns3_hw *hw, uint16_t *indir, uint16_t size)
 	return 0;
 }
 
+static int
+hns3_get_rss_indir_table(struct hns3_hw *hw, uint16_t *indir, uint16_t size)
+{
+	struct hns3_rss_indirection_table_cmd *req;
+	uint16_t max_bd_num, cfg_tbl_size;
+	uint8_t qid_msb_off, qid_msb_idx;
+	struct hns3_cmd_desc desc;
+	uint16_t q_id, q_hi, q_lo;
+	uint8_t rss_result_h;
+	uint16_t i, j;
+	int ret;
+
+	req = (struct hns3_rss_indirection_table_cmd *)desc.data;
+	max_bd_num = DIV_ROUND_UP(size, HNS3_RSS_CFG_TBL_SIZE);
+	for (i = 0; i < max_bd_num; i++) {
+		hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RSS_INDIR_TABLE,
+					  true);
+		req->start_table_index =
+				rte_cpu_to_le_16(i * HNS3_RSS_CFG_TBL_SIZE);
+		ret = hns3_cmd_send(hw, &desc, 1);
+		if (ret) {
+			hns3_err(hw, "fail to get RSS indirection table from firmware, ret = %d",
+				 ret);
+			return ret;
+		}
+
+		if (i == max_bd_num - 1 && (size % HNS3_RSS_CFG_TBL_SIZE) != 0)
+			cfg_tbl_size = size % HNS3_RSS_CFG_TBL_SIZE;
+		else
+			cfg_tbl_size = HNS3_RSS_CFG_TBL_SIZE;
+
+		for (j = 0; j < cfg_tbl_size; j++) {
+			qid_msb_idx =
+				j * HNS3_RSS_CFG_TBL_BW_H / HNS3_BITS_PER_BYTE;
+			rss_result_h = req->rss_result_h[qid_msb_idx];
+			qid_msb_off =
+				j * HNS3_RSS_CFG_TBL_BW_H % HNS3_BITS_PER_BYTE;
+			q_hi = (rss_result_h >> qid_msb_off) &
+						HNS3_RSS_CFG_TBL_BW_H_M;
+			q_lo = req->rss_result_l[j];
+			q_id = (q_hi << HNS3_RSS_CFG_TBL_BW_L) | q_lo;
+			indir[i * HNS3_RSS_CFG_TBL_SIZE + j] = q_id;
+		}
+	}
+
+	return 0;
+}
+
 int
 hns3_rss_reset_indir_table(struct hns3_hw *hw)
 {
@@ -841,10 +889,11 @@ hns3_dev_rss_reta_query(struct rte_eth_dev *dev,
 			uint16_t reta_size)
 {
 	struct hns3_adapter *hns = dev->data->dev_private;
+	uint16_t reta_table[HNS3_RSS_IND_TBL_SIZE_MAX];
 	struct hns3_hw *hw = &hns->hw;
-	struct hns3_rss_conf *rss_cfg = &hw->rss_info;
 	uint16_t idx, shift;
 	uint16_t i;
+	int ret;
 
 	if (reta_size != hw->rss_ind_tbl_size) {
 		hns3_err(hw, "The size of hash lookup table configured (%u)"
@@ -853,14 +902,22 @@ hns3_dev_rss_reta_query(struct rte_eth_dev *dev,
 		return -EINVAL;
 	}
 	rte_spinlock_lock(&hw->lock);
+	ret = hns3_get_rss_indir_table(hw, reta_table, reta_size);
+	if (ret != 0) {
+		rte_spinlock_unlock(&hw->lock);
+		hns3_err(hw, "query RSS redirection table failed, ret = %d.",
+			 ret);
+		return ret;
+	}
+	rte_spinlock_unlock(&hw->lock);
+
 	for (i = 0; i < reta_size; i++) {
 		idx = i / RTE_ETH_RETA_GROUP_SIZE;
 		shift = i % RTE_ETH_RETA_GROUP_SIZE;
 		if (reta_conf[idx].mask & (1ULL << shift))
-			reta_conf[idx].reta[shift] =
-						rss_cfg->rss_indirection_tbl[i];
+			reta_conf[idx].reta[shift] = reta_table[i];
 	}
-	rte_spinlock_unlock(&hw->lock);
+
 	return 0;
 }
 
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 06/16] net/hns3: separate the setting of hash algorithm
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (4 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 05/16] net/hns3: use hardware config to report redirection table Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10 19:36   ` Ferruh Yigit
  2023-03-10  9:35 ` [PATCH 07/16] net/hns3: separate the setting of hash key Dongdong Liu
                   ` (10 subsequent siblings)
  16 siblings, 1 reply; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Currently, the setting of hash algorithm comes from the
default configuration in driver and the rte_flow interface.
The hash algorithm that is set to hardware in both ways is
saved in hw->rss_info.conf.func.

But the 'func' in struct rte_flow_action_rss is usually used
in rte flow interface. And the ethdev ops interface may also
set hash algorithm in the future. It is not appropriate and
is a little messy for ethdev ops interface and driver default
configuration to use struct rte_flow_action_rss. So we have
to separate the RSS configuration from ethdev ops and rte
flow interface to make codes more easier to maintain.

This patch separates hash algorithm by following ways:
1) 'hash_algo' in struct hns3_rss_conf is used for ethdev ops
   interface or default configuration in driver.
2) Add a 'rte_flow_hash_algo' field in struct hns3_rss_conf
   to save algorithm from rte flow interface. The main reasons
   are as follows:
   Currently, only the last rule is used to restore the rte
   flow rule. If 'func' in RSS action is 'DEFAULT', it means
   that this rule doesn't modify algorithm and driver need to
   save current algorithm for restoring algorithm during reset
   phase.

Fixes: c37ca66f2b27 ("net/hns3: support RSS")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_flow.c | 60 +++++++++++++++++++++---------------
 drivers/net/hns3/hns3_rss.c  | 14 +--------
 drivers/net/hns3/hns3_rss.h  |  1 +
 3 files changed, 37 insertions(+), 38 deletions(-)

diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index 881d70613c..2faeb9ca52 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -1438,30 +1438,40 @@ hns3_disable_rss(struct hns3_hw *hw)
 }
 
 static int
-hns3_parse_rss_algorithm(struct hns3_hw *hw, enum rte_eth_hash_function *func,
+hns3_parse_rss_algorithm(struct hns3_hw *hw, enum rte_eth_hash_function func,
 			 uint8_t *hash_algo)
 {
-	enum rte_eth_hash_function algo_func = *func;
-	switch (algo_func) {
-	case RTE_ETH_HASH_FUNCTION_DEFAULT:
-		/* Keep *hash_algo as what it used to be */
-		algo_func = hw->rss_info.conf.func;
-		break;
-	case RTE_ETH_HASH_FUNCTION_TOEPLITZ:
-		*hash_algo = HNS3_RSS_HASH_ALGO_TOEPLITZ;
-		break;
-	case RTE_ETH_HASH_FUNCTION_SIMPLE_XOR:
-		*hash_algo = HNS3_RSS_HASH_ALGO_SIMPLE;
-		break;
-	case RTE_ETH_HASH_FUNCTION_SYMMETRIC_TOEPLITZ:
-		*hash_algo = HNS3_RSS_HASH_ALGO_SYMMETRIC_TOEP;
-		break;
-	default:
-		hns3_err(hw, "Invalid RSS algorithm configuration(%d)",
-			 algo_func);
-		return -EINVAL;
+	const uint8_t hash_func_map[] = {
+		[RTE_ETH_HASH_FUNCTION_DEFAULT] = HNS3_RSS_HASH_ALGO_TOEPLITZ,
+		[RTE_ETH_HASH_FUNCTION_TOEPLITZ] = HNS3_RSS_HASH_ALGO_TOEPLITZ,
+		[RTE_ETH_HASH_FUNCTION_SIMPLE_XOR] = HNS3_RSS_HASH_ALGO_SIMPLE,
+		[RTE_ETH_HASH_FUNCTION_SYMMETRIC_TOEPLITZ] = HNS3_RSS_HASH_ALGO_SYMMETRIC_TOEP,
+	};
+	uint8_t key[HNS3_RSS_KEY_SIZE_MAX] = {0};
+	int ret;
+
+	if (func == RTE_ETH_HASH_FUNCTION_DEFAULT) {
+		ret = hns3_rss_get_algo_key(hw, hash_algo, key,
+					    hw->rss_key_size);
+		if (ret != 0) {
+			hns3_err(hw, "fail to get current RSS hash algorithm, ret = %d",
+				 ret);
+			return ret;
+		}
+
+		/*
+		 * During the phase of reset recovery, the hash algorithm
+		 * obtained from hardware may not be the one used(saved in
+		 * rte_flow_hash_algo) when this rule is delivered.
+		 */
+		if (__atomic_load_n(&hw->reset.resetting, __ATOMIC_RELAXED) &&
+		    *hash_algo != hw->rss_info.rte_flow_hash_algo)
+			*hash_algo = hw->rss_info.rte_flow_hash_algo;
+
+		return 0;
 	}
-	*func = algo_func;
+
+	*hash_algo = hash_func_map[func];
 
 	return 0;
 }
@@ -1471,6 +1481,7 @@ hns3_hw_rss_hash_set(struct hns3_hw *hw, struct rte_flow_action_rss *rss_config)
 {
 	uint8_t rss_key[HNS3_RSS_KEY_SIZE_MAX] = {0};
 	bool use_default_key = false;
+	uint8_t hash_algo;
 	int ret;
 
 	if (rss_config->key == NULL || rss_config->key_len != hw->rss_key_size) {
@@ -1480,18 +1491,17 @@ hns3_hw_rss_hash_set(struct hns3_hw *hw, struct rte_flow_action_rss *rss_config)
 		use_default_key = true;
 	}
 
-	ret = hns3_parse_rss_algorithm(hw, &rss_config->func,
-				       &hw->rss_info.hash_algo);
+	ret = hns3_parse_rss_algorithm(hw, rss_config->func, &hash_algo);
 	if (ret)
 		return ret;
 
-	ret = hns3_rss_set_algo_key(hw, hw->rss_info.hash_algo,
+	ret = hns3_rss_set_algo_key(hw, hash_algo,
 				    use_default_key ? rss_key : rss_config->key,
 				    hw->rss_key_size);
 	if (ret)
 		return ret;
 
-	hw->rss_info.conf.func = rss_config->func;
+	hw->rss_info.rte_flow_hash_algo = hash_algo;
 
 	ret = hns3_set_rss_tuple_by_rss_hf(hw, rss_config->types);
 	if (ret)
diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index fe9ad609b7..88ba0757b5 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -1021,7 +1021,7 @@ hns3_rss_set_default_args(struct hns3_hw *hw)
 	uint16_t i;
 
 	/* Default hash algorithm */
-	rss_cfg->conf.func = RTE_ETH_HASH_FUNCTION_TOEPLITZ;
+	rss_cfg->hash_algo = HNS3_RSS_HASH_ALGO_TOEPLITZ;
 
 	memcpy(rss_cfg->key, hns3_hash_key,
 		RTE_MIN(sizeof(hns3_hash_key), hw->rss_key_size));
@@ -1045,18 +1045,6 @@ hns3_config_rss(struct hns3_adapter *hns)
 
 	enum rte_eth_rx_mq_mode mq_mode = hw->data->dev_conf.rxmode.mq_mode;
 
-	switch (hw->rss_info.conf.func) {
-	case RTE_ETH_HASH_FUNCTION_SIMPLE_XOR:
-		hw->rss_info.hash_algo = HNS3_RSS_HASH_ALGO_SIMPLE;
-		break;
-	case RTE_ETH_HASH_FUNCTION_SYMMETRIC_TOEPLITZ:
-		hw->rss_info.hash_algo = HNS3_RSS_HASH_ALGO_SYMMETRIC_TOEP;
-		break;
-	default:
-		hw->rss_info.hash_algo = HNS3_RSS_HASH_ALGO_TOEPLITZ;
-		break;
-	}
-
 	ret = hns3_rss_set_algo_key(hw, rss_cfg->hash_algo,
 				    hash_key, hw->rss_key_size);
 	if (ret)
diff --git a/drivers/net/hns3/hns3_rss.h b/drivers/net/hns3/hns3_rss.h
index 17473e70e2..6e679b709b 100644
--- a/drivers/net/hns3/hns3_rss.h
+++ b/drivers/net/hns3/hns3_rss.h
@@ -42,6 +42,7 @@ struct hns3_rss_conf {
 	/* RSS parameters :algorithm, flow_types,  key, queue */
 	struct rte_flow_action_rss conf;
 	uint8_t hash_algo; /* hash function type defined by hardware */
+	uint8_t rte_flow_hash_algo;
 	uint8_t key[HNS3_RSS_KEY_SIZE_MAX];  /* Hash key */
 	uint16_t rss_indirection_tbl[HNS3_RSS_IND_TBL_SIZE_MAX];
 	uint16_t queue[HNS3_RSS_QUEUES_BUFFER_NUM]; /* Queues indices to use */
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 07/16] net/hns3: separate the setting of hash key
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (5 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 06/16] net/hns3: separate the setting of hash algorithm Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 08/16] net/hns3: separate the setting of redirection table Dongdong Liu
                   ` (9 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

The settings of hash key comes from the ethdev ops (like, dev_configure
and rss_hash_update) and rte_flow API. For the ethdev ops, driver has
to save it to rss_info::key in hns3_hw structure so as to it can be
restored when reset is triggered. While rte_flow API no need to use
this field to save, they has a global flow_rss_list to maintain all
rules which save hash key. And hash key can be restored by this rule
information during the reset phase.

Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_rss.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index 88ba0757b5..be717b7641 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -376,8 +376,7 @@ hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
 			return ret;
 		}
 	}
-	/* Update the shadow RSS key with user specified */
-	memcpy(hw->rss_info.key, key, hw->rss_key_size);
+
 	return 0;
 }
 
@@ -671,6 +670,8 @@ hns3_dev_rss_hash_update(struct rte_eth_dev *dev,
 					    key, hw->rss_key_size);
 		if (ret)
 			goto set_algo_key_fail;
+		/* Update the shadow RSS key with user specified */
+		memcpy(hw->rss_info.key, key, hw->rss_key_size);
 	}
 	rte_spinlock_unlock(&hw->lock);
 
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 08/16] net/hns3: separate the setting of redirection table
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (6 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 07/16] net/hns3: separate the setting of hash key Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 09/16] net/hns3: separate the setting of RSS types Dongdong Liu
                   ` (8 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

The settings of redirection table comes from the ethdev ops (like,
dev_configure and rss_hash_update) and rte_flow API. For the ethdev
ops, driver has to save it to rss_info::rss_indirection_tbl in hns3_hw
structure so as to it can be restored when reset is triggered.
While rte_flow API no need to use this field to save, they has a global
RSS flow list to maintain all rules which can be used to restore the
table during the reset phase.

Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_flow.c |  2 --
 drivers/net/hns3/hns3_rss.c  | 21 +++++++++++++--------
 2 files changed, 13 insertions(+), 10 deletions(-)

diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index 2faeb9ca52..d5db09a263 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -1519,8 +1519,6 @@ hns3_update_indir_table(struct hns3_hw *hw,
 	uint32_t i;
 
 	/* Fill in redirection table */
-	memcpy(indir_tbl, hw->rss_info.rss_indirection_tbl,
-	       sizeof(hw->rss_info.rss_indirection_tbl));
 	for (i = 0, j = 0; i < hw->rss_ind_tbl_size; i++, j++) {
 		j %= num;
 		if (conf->queue[j] >= hw->alloc_rss_size) {
diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index be717b7641..037914ef04 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -472,10 +472,6 @@ hns3_set_rss_indir_table(struct hns3_hw *hw, uint16_t *indir, uint16_t size)
 		}
 	}
 
-	/* Update redirection table of hw */
-	memcpy(hw->rss_info.rss_indirection_tbl, indir,
-	       sizeof(uint16_t) * size);
-
 	return 0;
 }
 
@@ -541,8 +537,11 @@ hns3_rss_reset_indir_table(struct hns3_hw *hw)
 	}
 
 	ret = hns3_set_rss_indir_table(hw, lut, hw->rss_ind_tbl_size);
-	if (ret)
-		hns3_err(hw, "RSS uninit indir table failed: %d", ret);
+	if (ret != 0)
+		hns3_err(hw, "RSS uninit indir table failed, ret = %d.", ret);
+	else
+		memcpy(hw->rss_info.rss_indirection_tbl, lut,
+		       sizeof(uint16_t) * hw->rss_ind_tbl_size);
 	rte_free(lut);
 
 	return ret;
@@ -854,12 +853,12 @@ hns3_dev_rss_reta_update(struct rte_eth_dev *dev,
 		idx = i / RTE_ETH_RETA_GROUP_SIZE;
 		shift = i % RTE_ETH_RETA_GROUP_SIZE;
 		if (reta_conf[idx].reta[shift] >= hw->alloc_rss_size) {
-			rte_spinlock_unlock(&hw->lock);
 			hns3_err(hw, "queue id(%u) set to redirection table "
 				 "exceeds queue number(%u) allocated to a TC",
 				 reta_conf[idx].reta[shift],
 				 hw->alloc_rss_size);
-			return -EINVAL;
+			ret = -EINVAL;
+			goto out;
 		}
 
 		if (reta_conf[idx].mask & (1ULL << shift))
@@ -868,7 +867,13 @@ hns3_dev_rss_reta_update(struct rte_eth_dev *dev,
 
 	ret = hns3_set_rss_indir_table(hw, indirection_tbl,
 				       hw->rss_ind_tbl_size);
+	if (ret != 0)
+		goto out;
 
+	memcpy(rss_cfg->rss_indirection_tbl, indirection_tbl,
+	       sizeof(uint16_t) * hw->rss_ind_tbl_size);
+
+out:
 	rte_spinlock_unlock(&hw->lock);
 	return ret;
 }
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 09/16] net/hns3: separate the setting of RSS types
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (7 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 08/16] net/hns3: separate the setting of redirection table Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 10/16] net/hns3: separate the setting and clearing of RSS rule Dongdong Liu
                   ` (7 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

The settings of RSS types comes from the ethdev ops (like, dev_configure
and rss_hash_update) and rte_flow API. For the ethdev ops, driver has to
save it so as to it can be restored when reset is triggered.
While rte_flow API no need to maintain this field, it can be restored by
the saved rule.

Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_flow.c |  3 ++-
 drivers/net/hns3/hns3_rss.c  | 22 ++++++++++++++--------
 drivers/net/hns3/hns3_rss.h  |  1 +
 3 files changed, 17 insertions(+), 9 deletions(-)

diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index d5db09a263..21e00e515e 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -1433,6 +1433,7 @@ hns3_disable_rss(struct hns3_hw *hw)
 	ret = hns3_set_rss_tuple_by_rss_hf(hw, 0);
 	if (ret)
 		return ret;
+	hw->rss_info.rss_hf = 0;
 
 	return 0;
 }
@@ -1580,7 +1581,7 @@ hns3_config_rss_filter(struct hns3_hw *hw,
 	/* Filter the unsupported flow types */
 	flow_types = conf->conf.types ?
 		     rss_flow_conf.types & HNS3_ETH_RSS_SUPPORT :
-		     hw->rss_info.conf.types;
+		     hw->rss_info.rss_hf;
 	if (flow_types != rss_flow_conf.types)
 		hns3_warn(hw, "modified RSS types based on hardware support,"
 			  " requested:0x%" PRIx64 " configured:0x%" PRIx64,
diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index 037914ef04..2ef267aac6 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -627,9 +627,6 @@ hns3_set_rss_tuple_by_rss_hf(struct hns3_hw *hw, uint64_t rss_hf)
 		return ret;
 	}
 
-	/* Update supported flow types when set tuple success */
-	hw->rss_info.conf.types = rss_hf;
-
 	return 0;
 }
 
@@ -647,7 +644,7 @@ hns3_dev_rss_hash_update(struct rte_eth_dev *dev,
 			 struct rte_eth_rss_conf *rss_conf)
 {
 	struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
-	uint64_t rss_hf_bk = hw->rss_info.conf.types;
+	uint64_t rss_hf_bk = hw->rss_info.rss_hf;
 	uint8_t key_len = rss_conf->rss_key_len;
 	uint64_t rss_hf = rss_conf->rss_hf;
 	uint8_t *key = rss_conf->rss_key;
@@ -672,6 +669,7 @@ hns3_dev_rss_hash_update(struct rte_eth_dev *dev,
 		/* Update the shadow RSS key with user specified */
 		memcpy(hw->rss_info.key, key, hw->rss_key_size);
 	}
+	hw->rss_info.rss_hf = rss_hf;
 	rte_spinlock_unlock(&hw->lock);
 
 	return 0;
@@ -1029,6 +1027,7 @@ hns3_rss_set_default_args(struct hns3_hw *hw)
 	/* Default hash algorithm */
 	rss_cfg->hash_algo = HNS3_RSS_HASH_ALGO_TOEPLITZ;
 
+	hw->rss_info.rss_hf = 0;
 	memcpy(rss_cfg->key, hns3_hash_key,
 		RTE_MIN(sizeof(hns3_hash_key), hw->rss_key_size));
 
@@ -1066,15 +1065,22 @@ hns3_config_rss(struct hns3_adapter *hns)
 		return ret;
 
 	/*
-	 * When muli-queue RSS mode flag is not set or unsupported tuples are
+	 * When multi-queue RSS mode flag is not set or unsupported tuples are
 	 * set, disable all tuples.
 	 */
-	rss_hf = hw->rss_info.conf.types;
+	rss_hf = hw->rss_info.rss_hf;
 	if (!((uint32_t)mq_mode & RTE_ETH_MQ_RX_RSS_FLAG) ||
 	    !(rss_hf & HNS3_ETH_RSS_SUPPORT))
 		rss_hf = 0;
 
-	return hns3_set_rss_tuple_by_rss_hf(hw, rss_hf);
+	ret = hns3_set_rss_tuple_by_rss_hf(hw, rss_hf);
+	if (ret != 0) {
+		hns3_err(hw, "set RSS tuples failed, ret = %d.", ret);
+		return ret;
+	}
+	hw->rss_info.rss_hf = rss_hf;
+
+	return 0;
 }
 
 /*
@@ -1092,5 +1098,5 @@ hns3_rss_uninit(struct hns3_adapter *hns)
 		return;
 
 	/* Disable RSS */
-	hw->rss_info.conf.types = 0;
+	hw->rss_info.rss_hf = 0;
 }
diff --git a/drivers/net/hns3/hns3_rss.h b/drivers/net/hns3/hns3_rss.h
index 6e679b709b..21b90789d0 100644
--- a/drivers/net/hns3/hns3_rss.h
+++ b/drivers/net/hns3/hns3_rss.h
@@ -41,6 +41,7 @@
 struct hns3_rss_conf {
 	/* RSS parameters :algorithm, flow_types,  key, queue */
 	struct rte_flow_action_rss conf;
+	uint64_t rss_hf;
 	uint8_t hash_algo; /* hash function type defined by hardware */
 	uint8_t rte_flow_hash_algo;
 	uint8_t key[HNS3_RSS_KEY_SIZE_MAX];  /* Hash key */
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 10/16] net/hns3: separate the setting and clearing of RSS rule
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (8 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 09/16] net/hns3: separate the setting of RSS types Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 11/16] net/hns3: use new RSS rule to configure hardware Dongdong Liu
                   ` (6 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Separate the setting and clearing of RSS rule.

Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_flow.c | 46 +++++++++++++++++-------------------
 1 file changed, 22 insertions(+), 24 deletions(-)

diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index 21e00e515e..4da98d7adc 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -1535,8 +1535,22 @@ hns3_update_indir_table(struct hns3_hw *hw,
 }
 
 static int
-hns3_config_rss_filter(struct hns3_hw *hw,
-		       const struct hns3_rss_conf *conf, bool add)
+hns3_reset_rss_filter(struct hns3_hw *hw, const struct hns3_rss_conf *conf)
+{
+	int ret;
+
+	if (!conf->valid)
+		return 0;
+
+	ret = hns3_disable_rss(hw);
+	if (ret)
+		hns3_err(hw, "RSS disable failed(%d)", ret);
+
+	return ret;
+}
+
+static int
+hns3_config_rss_filter(struct hns3_hw *hw, const struct hns3_rss_conf *conf)
 {
 	uint64_t flow_types;
 	uint16_t num;
@@ -1553,19 +1567,6 @@ hns3_config_rss_filter(struct hns3_hw *hw,
 		.queue = conf->conf.queue,
 	};
 
-	if (!add) {
-		if (!conf->valid)
-			return 0;
-
-		ret = hns3_disable_rss(hw);
-		if (ret) {
-			hns3_err(hw, "RSS disable failed(%d)", ret);
-			return ret;
-		}
-
-		return 0;
-	}
-
 	/* Set rx queues to use */
 	num = RTE_MIN(hw->data->nb_rx_queues, rss_flow_conf.queue_num);
 	if (rss_flow_conf.queue_num > num)
@@ -1606,8 +1607,7 @@ hns3_clear_rss_filter(struct rte_eth_dev *dev)
 	rss_filter_ptr = TAILQ_FIRST(&hw->flow_rss_list);
 	while (rss_filter_ptr) {
 		TAILQ_REMOVE(&hw->flow_rss_list, rss_filter_ptr, entries);
-		ret = hns3_config_rss_filter(hw, &rss_filter_ptr->filter_info,
-					     false);
+		ret = hns3_reset_rss_filter(hw, &rss_filter_ptr->filter_info);
 		if (ret)
 			rss_rule_fail_cnt++;
 		else
@@ -1636,7 +1636,7 @@ hns3_restore_rss_filter(struct hns3_hw *hw)
 		if (!filter->filter_info.valid)
 			continue;
 
-		ret = hns3_config_rss_filter(hw, &filter->filter_info, true);
+		ret = hns3_config_rss_filter(hw, &filter->filter_info);
 		if (ret != 0) {
 			hns3_err(hw, "restore RSS filter failed, ret=%d", ret);
 			goto out;
@@ -1680,8 +1680,7 @@ hns3_rss_action_is_dup(struct hns3_hw *hw,
 }
 
 static int
-hns3_flow_parse_rss(struct rte_eth_dev *dev,
-		    const struct hns3_rss_conf *conf, bool add)
+hns3_flow_parse_rss(struct rte_eth_dev *dev, const struct hns3_rss_conf *conf)
 {
 	struct hns3_adapter *hns = dev->data->dev_private;
 	struct hns3_hw *hw = &hns->hw;
@@ -1691,7 +1690,7 @@ hns3_flow_parse_rss(struct rte_eth_dev *dev,
 		return -EINVAL;
 	}
 
-	return hns3_config_rss_filter(hw, conf, add);
+	return hns3_config_rss_filter(hw, conf);
 }
 
 static int
@@ -1778,7 +1777,7 @@ hns3_flow_create_rss_rule(struct rte_eth_dev *dev,
 		}
 	}
 
-	ret = hns3_flow_parse_rss(dev, new_conf, true);
+	ret = hns3_flow_parse_rss(dev, new_conf);
 	if (ret != 0) {
 		rte_free(rss_filter_ptr);
 		return ret;
@@ -1961,8 +1960,7 @@ hns3_flow_destroy(struct rte_eth_dev *dev, struct rte_flow *flow,
 		break;
 	case RTE_ETH_FILTER_HASH:
 		rss_filter_ptr = (struct hns3_rss_conf_ele *)flow->rule;
-		ret = hns3_config_rss_filter(hw, &rss_filter_ptr->filter_info,
-					     false);
+		ret = hns3_reset_rss_filter(hw, &rss_filter_ptr->filter_info);
 		if (ret)
 			return rte_flow_error_set(error, EIO,
 						  RTE_FLOW_ERROR_TYPE_HANDLE,
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 11/16] net/hns3: use new RSS rule to configure hardware
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (9 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 10/16] net/hns3: separate the setting and clearing of RSS rule Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 12/16] net/hns3: save hash algo to RSS filter list node Dongdong Liu
                   ` (5 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Remove redundant assignment and directly use new RSS rule to configure
hardware. Additionally, considering that the new rule configuration may
need to be modified, the 'const' of input parameter about it is removed.

Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_flow.c | 52 ++++++++++++++----------------------
 1 file changed, 20 insertions(+), 32 deletions(-)

diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index 4da98d7adc..3700d073b1 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -1482,6 +1482,7 @@ hns3_hw_rss_hash_set(struct hns3_hw *hw, struct rte_flow_action_rss *rss_config)
 {
 	uint8_t rss_key[HNS3_RSS_KEY_SIZE_MAX] = {0};
 	bool use_default_key = false;
+	uint64_t flow_types;
 	uint8_t hash_algo;
 	int ret;
 
@@ -1501,10 +1502,18 @@ hns3_hw_rss_hash_set(struct hns3_hw *hw, struct rte_flow_action_rss *rss_config)
 				    hw->rss_key_size);
 	if (ret)
 		return ret;
-
 	hw->rss_info.rte_flow_hash_algo = hash_algo;
 
-	ret = hns3_set_rss_tuple_by_rss_hf(hw, rss_config->types);
+	/* Filter the unsupported flow types */
+	flow_types = rss_config->types ?
+		     rss_config->types & HNS3_ETH_RSS_SUPPORT :
+		     hw->rss_info.rss_hf;
+	if (flow_types != rss_config->types)
+		hns3_warn(hw, "modified RSS types based on hardware support,"
+			  " requested:0x%" PRIx64 " configured:0x%" PRIx64,
+			  rss_config->types, flow_types);
+
+	ret = hns3_set_rss_tuple_by_rss_hf(hw, flow_types);
 	if (ret)
 		hns3_err(hw, "Update RSS tuples by rss hf failed %d", ret);
 
@@ -1550,48 +1559,27 @@ hns3_reset_rss_filter(struct hns3_hw *hw, const struct hns3_rss_conf *conf)
 }
 
 static int
-hns3_config_rss_filter(struct hns3_hw *hw, const struct hns3_rss_conf *conf)
+hns3_config_rss_filter(struct hns3_hw *hw, struct hns3_rss_conf *conf)
 {
-	uint64_t flow_types;
+	struct rte_flow_action_rss *rss_act;
 	uint16_t num;
 	int ret;
 
-	struct rte_flow_action_rss rss_flow_conf = {
-		.func = conf->conf.func,
-		.level = conf->conf.level,
-		.types = conf->conf.types,
-		.key_len = conf->conf.key_len,
-		.queue_num = conf->conf.queue_num,
-		.key = conf->conf.key_len ?
-		    (void *)(uintptr_t)conf->conf.key : NULL,
-		.queue = conf->conf.queue,
-	};
-
+	rss_act = &conf->conf;
 	/* Set rx queues to use */
-	num = RTE_MIN(hw->data->nb_rx_queues, rss_flow_conf.queue_num);
-	if (rss_flow_conf.queue_num > num)
+	num = RTE_MIN(hw->data->nb_rx_queues, rss_act->queue_num);
+	if (rss_act->queue_num > num)
 		hns3_warn(hw, "Config queue numbers %u are beyond the scope of truncated",
-			  rss_flow_conf.queue_num);
+			  rss_act->queue_num);
 	hns3_info(hw, "Max of contiguous %u PF queues are configured", num);
 	if (num) {
-		ret = hns3_update_indir_table(hw, &rss_flow_conf, num);
+		ret = hns3_update_indir_table(hw, rss_act, num);
 		if (ret)
 			return ret;
 	}
 
-	/* Filter the unsupported flow types */
-	flow_types = conf->conf.types ?
-		     rss_flow_conf.types & HNS3_ETH_RSS_SUPPORT :
-		     hw->rss_info.rss_hf;
-	if (flow_types != rss_flow_conf.types)
-		hns3_warn(hw, "modified RSS types based on hardware support,"
-			  " requested:0x%" PRIx64 " configured:0x%" PRIx64,
-			  rss_flow_conf.types, flow_types);
-	/* Update the useful flow types */
-	rss_flow_conf.types = flow_types;
-
 	/* Set hash algorithm and flow types by the user's config */
-	return hns3_hw_rss_hash_set(hw, &rss_flow_conf);
+	return hns3_hw_rss_hash_set(hw, rss_act);
 }
 
 static int
@@ -1680,7 +1668,7 @@ hns3_rss_action_is_dup(struct hns3_hw *hw,
 }
 
 static int
-hns3_flow_parse_rss(struct rte_eth_dev *dev, const struct hns3_rss_conf *conf)
+hns3_flow_parse_rss(struct rte_eth_dev *dev, struct hns3_rss_conf *conf)
 {
 	struct hns3_adapter *hns = dev->data->dev_private;
 	struct hns3_hw *hw = &hns->hw;
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 12/16] net/hns3: save hash algo to RSS filter list node
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (10 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 11/16] net/hns3: use new RSS rule to configure hardware Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 13/16] net/hns3: adding queue buffer size hash rule allowed Dongdong Liu
                   ` (4 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Save hash algo from rte flow RSS rule to RSS filter list node
instead of struct hns3_rss_conf.

Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_flow.c | 19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index 3700d073b1..527874df44 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -1439,7 +1439,7 @@ hns3_disable_rss(struct hns3_hw *hw)
 }
 
 static int
-hns3_parse_rss_algorithm(struct hns3_hw *hw, enum rte_eth_hash_function func,
+hns3_parse_rss_algorithm(struct hns3_hw *hw, struct hns3_rss_conf *rss_conf,
 			 uint8_t *hash_algo)
 {
 	const uint8_t hash_func_map[] = {
@@ -1451,7 +1451,7 @@ hns3_parse_rss_algorithm(struct hns3_hw *hw, enum rte_eth_hash_function func,
 	uint8_t key[HNS3_RSS_KEY_SIZE_MAX] = {0};
 	int ret;
 
-	if (func == RTE_ETH_HASH_FUNCTION_DEFAULT) {
+	if (rss_conf->conf.func == RTE_ETH_HASH_FUNCTION_DEFAULT) {
 		ret = hns3_rss_get_algo_key(hw, hash_algo, key,
 					    hw->rss_key_size);
 		if (ret != 0) {
@@ -1466,20 +1466,21 @@ hns3_parse_rss_algorithm(struct hns3_hw *hw, enum rte_eth_hash_function func,
 		 * rte_flow_hash_algo) when this rule is delivered.
 		 */
 		if (__atomic_load_n(&hw->reset.resetting, __ATOMIC_RELAXED) &&
-		    *hash_algo != hw->rss_info.rte_flow_hash_algo)
-			*hash_algo = hw->rss_info.rte_flow_hash_algo;
+		    *hash_algo != rss_conf->rte_flow_hash_algo)
+			*hash_algo = rss_conf->rte_flow_hash_algo;
 
 		return 0;
 	}
 
-	*hash_algo = hash_func_map[func];
+	*hash_algo = hash_func_map[rss_conf->conf.func];
 
 	return 0;
 }
 
 static int
-hns3_hw_rss_hash_set(struct hns3_hw *hw, struct rte_flow_action_rss *rss_config)
+hns3_hw_rss_hash_set(struct hns3_hw *hw, struct hns3_rss_conf *conf)
 {
+	struct rte_flow_action_rss *rss_config = &conf->conf;
 	uint8_t rss_key[HNS3_RSS_KEY_SIZE_MAX] = {0};
 	bool use_default_key = false;
 	uint64_t flow_types;
@@ -1493,7 +1494,7 @@ hns3_hw_rss_hash_set(struct hns3_hw *hw, struct rte_flow_action_rss *rss_config)
 		use_default_key = true;
 	}
 
-	ret = hns3_parse_rss_algorithm(hw, rss_config->func, &hash_algo);
+	ret = hns3_parse_rss_algorithm(hw, conf, &hash_algo);
 	if (ret)
 		return ret;
 
@@ -1502,7 +1503,7 @@ hns3_hw_rss_hash_set(struct hns3_hw *hw, struct rte_flow_action_rss *rss_config)
 				    hw->rss_key_size);
 	if (ret)
 		return ret;
-	hw->rss_info.rte_flow_hash_algo = hash_algo;
+	conf->rte_flow_hash_algo = hash_algo;
 
 	/* Filter the unsupported flow types */
 	flow_types = rss_config->types ?
@@ -1579,7 +1580,7 @@ hns3_config_rss_filter(struct hns3_hw *hw, struct hns3_rss_conf *conf)
 	}
 
 	/* Set hash algorithm and flow types by the user's config */
-	return hns3_hw_rss_hash_set(hw, rss_act);
+	return hns3_hw_rss_hash_set(hw, conf);
 }
 
 static int
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 13/16] net/hns3: adding queue buffer size hash rule allowed
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (11 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 12/16] net/hns3: save hash algo to RSS filter list node Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 14/16] net/hns3: separate rte flow RSS config from hns3 rss conf Dongdong Liu
                   ` (3 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

The maximum queue number from RSS flow rule allowed depends on
the maximum queue number (512) under one TC instead of the one
of Rx/Tx queue so as to eliminate restrictions on user usage.

Fixes: c37ca66f2b27 ("net/hns3: support RSS")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_rss.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/net/hns3/hns3_rss.h b/drivers/net/hns3/hns3_rss.h
index 21b90789d0..cc0bb8431d 100644
--- a/drivers/net/hns3/hns3_rss.h
+++ b/drivers/net/hns3/hns3_rss.h
@@ -37,7 +37,8 @@
 #define HNS3_RSS_HASH_ALGO_SYMMETRIC_TOEP 2
 #define HNS3_RSS_HASH_ALGO_MASK		0xf
 
-#define HNS3_RSS_QUEUES_BUFFER_NUM	64 /* Same as the Max rx/tx queue num */
+/* Same as the Max queue num under TC */
+#define HNS3_RSS_QUEUES_BUFFER_NUM	512
 struct hns3_rss_conf {
 	/* RSS parameters :algorithm, flow_types,  key, queue */
 	struct rte_flow_action_rss conf;
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 14/16] net/hns3: separate rte flow RSS config from hns3 rss conf
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (12 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 13/16] net/hns3: adding queue buffer size hash rule allowed Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 15/16] net/hns3: reimplement hash flow function Dongdong Liu
                   ` (2 subsequent siblings)
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Some RSS fields in struct hns3_rss_conf (e.g. conf, queue,
valid) are only used when create RSS flow rule, which is
unnecessary for RSS configuration information from ethdev
ops. This patch removes these fields from hns3_rss_conf
and add a new struct hns3_flow_rss_conf as rte flow
RSS filter list node element.

Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_flow.c | 23 ++++++++++++-----------
 drivers/net/hns3/hns3_flow.h | 10 +++++++++-
 drivers/net/hns3/hns3_rss.h  |  5 -----
 3 files changed, 21 insertions(+), 17 deletions(-)

diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index 527874df44..89374816aa 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -1359,7 +1359,6 @@ hns3_parse_rss_filter(struct rte_eth_dev *dev,
 {
 	struct hns3_adapter *hns = dev->data->dev_private;
 	struct hns3_hw *hw = &hns->hw;
-	struct hns3_rss_conf *rss_conf = &hw->rss_info;
 	const struct rte_flow_action_rss *rss;
 	const struct rte_flow_action *act;
 	uint32_t act_index = 0;
@@ -1374,7 +1373,7 @@ hns3_parse_rss_filter(struct rte_eth_dev *dev,
 					  act, "no valid queues");
 	}
 
-	if (rss->queue_num > RTE_DIM(rss_conf->queue))
+	if (rss->queue_num > HNS3_RSS_QUEUES_BUFFER_NUM)
 		return rte_flow_error_set(error, ENOTSUP,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, act,
 					  "queue number configured exceeds "
@@ -1439,7 +1438,7 @@ hns3_disable_rss(struct hns3_hw *hw)
 }
 
 static int
-hns3_parse_rss_algorithm(struct hns3_hw *hw, struct hns3_rss_conf *rss_conf,
+hns3_parse_rss_algorithm(struct hns3_hw *hw, struct hns3_flow_rss_conf *rss_conf,
 			 uint8_t *hash_algo)
 {
 	const uint8_t hash_func_map[] = {
@@ -1466,8 +1465,8 @@ hns3_parse_rss_algorithm(struct hns3_hw *hw, struct hns3_rss_conf *rss_conf,
 		 * rte_flow_hash_algo) when this rule is delivered.
 		 */
 		if (__atomic_load_n(&hw->reset.resetting, __ATOMIC_RELAXED) &&
-		    *hash_algo != rss_conf->rte_flow_hash_algo)
-			*hash_algo = rss_conf->rte_flow_hash_algo;
+		    *hash_algo != rss_conf->hash_algo)
+			*hash_algo = rss_conf->hash_algo;
 
 		return 0;
 	}
@@ -1478,7 +1477,7 @@ hns3_parse_rss_algorithm(struct hns3_hw *hw, struct hns3_rss_conf *rss_conf,
 }
 
 static int
-hns3_hw_rss_hash_set(struct hns3_hw *hw, struct hns3_rss_conf *conf)
+hns3_hw_rss_hash_set(struct hns3_hw *hw, struct hns3_flow_rss_conf *conf)
 {
 	struct rte_flow_action_rss *rss_config = &conf->conf;
 	uint8_t rss_key[HNS3_RSS_KEY_SIZE_MAX] = {0};
@@ -1503,7 +1502,7 @@ hns3_hw_rss_hash_set(struct hns3_hw *hw, struct hns3_rss_conf *conf)
 				    hw->rss_key_size);
 	if (ret)
 		return ret;
-	conf->rte_flow_hash_algo = hash_algo;
+	conf->hash_algo = hash_algo;
 
 	/* Filter the unsupported flow types */
 	flow_types = rss_config->types ?
@@ -1545,7 +1544,8 @@ hns3_update_indir_table(struct hns3_hw *hw,
 }
 
 static int
-hns3_reset_rss_filter(struct hns3_hw *hw, const struct hns3_rss_conf *conf)
+hns3_reset_rss_filter(struct hns3_hw *hw,
+		      const struct hns3_flow_rss_conf *conf)
 {
 	int ret;
 
@@ -1560,7 +1560,7 @@ hns3_reset_rss_filter(struct hns3_hw *hw, const struct hns3_rss_conf *conf)
 }
 
 static int
-hns3_config_rss_filter(struct hns3_hw *hw, struct hns3_rss_conf *conf)
+hns3_config_rss_filter(struct hns3_hw *hw, struct hns3_flow_rss_conf *conf)
 {
 	struct rte_flow_action_rss *rss_act;
 	uint16_t num;
@@ -1669,7 +1669,8 @@ hns3_rss_action_is_dup(struct hns3_hw *hw,
 }
 
 static int
-hns3_flow_parse_rss(struct rte_eth_dev *dev, struct hns3_rss_conf *conf)
+hns3_flow_parse_rss(struct rte_eth_dev *dev,
+		    struct hns3_flow_rss_conf *conf)
 {
 	struct hns3_adapter *hns = dev->data->dev_private;
 	struct hns3_hw *hw = &hns->hw;
@@ -1739,8 +1740,8 @@ hns3_flow_create_rss_rule(struct rte_eth_dev *dev,
 	struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 	const struct rte_flow_action_rss *rss_act;
 	struct hns3_rss_conf_ele *rss_filter_ptr;
+	struct hns3_flow_rss_conf *new_conf;
 	struct hns3_rss_conf_ele *filter_ptr;
-	struct hns3_rss_conf *new_conf;
 	int ret;
 
 	rss_filter_ptr = rte_zmalloc("hns3 rss filter",
diff --git a/drivers/net/hns3/hns3_flow.h b/drivers/net/hns3/hns3_flow.h
index e4b2fdf2e6..90126f2b6e 100644
--- a/drivers/net/hns3/hns3_flow.h
+++ b/drivers/net/hns3/hns3_flow.h
@@ -24,10 +24,18 @@ struct rte_flow {
 	uint32_t counter_id;
 };
 
+struct hns3_flow_rss_conf {
+	struct rte_flow_action_rss conf;
+	uint8_t hash_algo;
+	uint8_t key[HNS3_RSS_KEY_SIZE_MAX];  /* Hash key */
+	uint16_t queue[HNS3_RSS_QUEUES_BUFFER_NUM]; /* Queues indices to use */
+	bool valid; /* check if RSS rule is valid */
+};
+
 /* rss filter list structure */
 struct hns3_rss_conf_ele {
 	TAILQ_ENTRY(hns3_rss_conf_ele) entries;
-	struct hns3_rss_conf filter_info;
+	struct hns3_flow_rss_conf filter_info;
 };
 
 /* hns3_flow memory list structure */
diff --git a/drivers/net/hns3/hns3_rss.h b/drivers/net/hns3/hns3_rss.h
index cc0bb8431d..d19730c69c 100644
--- a/drivers/net/hns3/hns3_rss.h
+++ b/drivers/net/hns3/hns3_rss.h
@@ -40,15 +40,10 @@
 /* Same as the Max queue num under TC */
 #define HNS3_RSS_QUEUES_BUFFER_NUM	512
 struct hns3_rss_conf {
-	/* RSS parameters :algorithm, flow_types,  key, queue */
-	struct rte_flow_action_rss conf;
 	uint64_t rss_hf;
 	uint8_t hash_algo; /* hash function type defined by hardware */
-	uint8_t rte_flow_hash_algo;
 	uint8_t key[HNS3_RSS_KEY_SIZE_MAX];  /* Hash key */
 	uint16_t rss_indirection_tbl[HNS3_RSS_IND_TBL_SIZE_MAX];
-	uint16_t queue[HNS3_RSS_QUEUES_BUFFER_NUM]; /* Queues indices to use */
-	bool valid; /* check if RSS rule is valid */
 	/*
 	 * For IPv6 SCTP packets type, check whether the NIC hardware support
 	 * RSS hash using the src/dst port as the input tuple. For Kunpeng920
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 15/16] net/hns3: reimplement hash flow function
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (13 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 14/16] net/hns3: separate rte flow RSS config from hns3 rss conf Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10  9:35 ` [PATCH 16/16] net/hns3: add the verification of RSS types Dongdong Liu
  2023-03-10 20:58 ` [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Ferruh Yigit
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Currently, hns3 driver supports setting mulitiple rte flow RSS rule,
but only the last is valid. This implementation is different from
the mainstream usage of rte flow hash in the community. Please see
the discussion threads [1] and [2].

This patch sets RSS hash feature completely based on the request of
the flow rule so that multiple hash rules can take effect at the same
time. Please notice that:
1. For hns3, 'func' has only one hardware. 'key' and 'queue' have only
   one entry in hardware.
2. the overlapping part of the old rule will be overridden if the
   configuration items of a new rule overlap with those of an old rule.

The hns3_flow_validate() verifies and parses RSS or Fdir rules from
user, and saves them to a local variable at the same time. The local
variable is directly used to create RSS or Fdir rules. In this way,
we save one parsing and saving process.

[1] https://lore.kernel.org/all/DM5PR12MB46648085D7CABF1AFF2D75CDD60A9@DM5PR12MB4664.namprd12.prod.outlook.com/
[2] https://lore.kernel.org/all/f7de4db4-1b88-622f-4e03-acd3eee8a72c@oktetlabs.ru/

Fixes: c37ca66f2b27 ("net/hns3: support RSS")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_ethdev.h |   9 -
 drivers/net/hns3/hns3_flow.c   | 966 +++++++++++++++++++++++----------
 drivers/net/hns3/hns3_flow.h   |  15 +-
 drivers/net/hns3/hns3_rss.c    | 144 ++---
 drivers/net/hns3/hns3_rss.h    | 117 +++-
 5 files changed, 855 insertions(+), 396 deletions(-)

diff --git a/drivers/net/hns3/hns3_ethdev.h b/drivers/net/hns3/hns3_ethdev.h
index 2457754b3d..9acc5a3d7e 100644
--- a/drivers/net/hns3/hns3_ethdev.h
+++ b/drivers/net/hns3/hns3_ethdev.h
@@ -996,15 +996,6 @@ static inline uint32_t hns3_read_reg(void *base, uint32_t reg)
 #define hns3_read_dev(a, reg) \
 	hns3_read_reg((a)->io_base, (reg))
 
-#define NEXT_ITEM_OF_ACTION(act, actions, index)                        \
-	do {								\
-		(act) = (actions) + (index);				\
-		while ((act)->type == RTE_FLOW_ACTION_TYPE_VOID) {	\
-			(index)++;					\
-			(act) = (actions) + (index);				\
-		}							\
-	} while (0)
-
 static inline uint64_t
 hns3_atomic_test_bit(unsigned int nr, volatile uint64_t *addr)
 {
diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index 89374816aa..6ac623bea9 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -10,6 +10,125 @@
 #include "hns3_logs.h"
 #include "hns3_flow.h"
 
+#define NEXT_ITEM_OF_ACTION(act, actions, index) \
+	do { \
+		(act) = (actions) + (index); \
+		while ((act)->type == RTE_FLOW_ACTION_TYPE_VOID) { \
+			(index)++; \
+			(act) = (actions) + (index); \
+		} \
+	} while (0)
+
+#define NEXT_ITEM_OF_PATTERN(item, pattern, index) \
+	do { \
+		(item) = (pattern) + (index); \
+		while ((item)->type == RTE_FLOW_ITEM_TYPE_VOID) { \
+			(index)++; \
+			(item) = (pattern) + (index); \
+		} \
+	} while (0)
+
+#define HNS3_HASH_HDR_ETH	RTE_BIT64(0)
+#define HNS3_HASH_HDR_IPV4	RTE_BIT64(1)
+#define HNS3_HASH_HDR_IPV6	RTE_BIT64(2)
+#define HNS3_HASH_HDR_TCP	RTE_BIT64(3)
+#define HNS3_HASH_HDR_UDP	RTE_BIT64(4)
+#define HNS3_HASH_HDR_SCTP	RTE_BIT64(5)
+
+#define HNS3_HASH_VOID_NEXT_ALLOW	BIT_ULL(RTE_FLOW_ITEM_TYPE_ETH)
+
+#define HNS3_HASH_ETH_NEXT_ALLOW	(BIT_ULL(RTE_FLOW_ITEM_TYPE_IPV4) | \
+					 BIT_ULL(RTE_FLOW_ITEM_TYPE_IPV6))
+
+#define HNS3_HASH_IP_NEXT_ALLOW		(BIT_ULL(RTE_FLOW_ITEM_TYPE_TCP) | \
+					 BIT_ULL(RTE_FLOW_ITEM_TYPE_UDP) | \
+					 BIT_ULL(RTE_FLOW_ITEM_TYPE_SCTP))
+
+static const uint64_t hash_pattern_next_allow_items[] = {
+	[RTE_FLOW_ITEM_TYPE_VOID] = HNS3_HASH_VOID_NEXT_ALLOW,
+	[RTE_FLOW_ITEM_TYPE_ETH]  = HNS3_HASH_ETH_NEXT_ALLOW,
+	[RTE_FLOW_ITEM_TYPE_IPV4] = HNS3_HASH_IP_NEXT_ALLOW,
+	[RTE_FLOW_ITEM_TYPE_IPV6] = HNS3_HASH_IP_NEXT_ALLOW,
+};
+
+static const uint64_t hash_pattern_item_header[] = {
+	[RTE_FLOW_ITEM_TYPE_ETH]  = HNS3_HASH_HDR_ETH,
+	[RTE_FLOW_ITEM_TYPE_IPV4] = HNS3_HASH_HDR_IPV4,
+	[RTE_FLOW_ITEM_TYPE_IPV6] = HNS3_HASH_HDR_IPV6,
+	[RTE_FLOW_ITEM_TYPE_TCP]  = HNS3_HASH_HDR_TCP,
+	[RTE_FLOW_ITEM_TYPE_UDP]  = HNS3_HASH_HDR_UDP,
+	[RTE_FLOW_ITEM_TYPE_SCTP] = HNS3_HASH_HDR_SCTP,
+};
+
+#define HNS3_HASH_IPV4		(HNS3_HASH_HDR_ETH | HNS3_HASH_HDR_IPV4)
+#define HNS3_HASH_IPV4_TCP	(HNS3_HASH_HDR_ETH | \
+				 HNS3_HASH_HDR_IPV4 | \
+				 HNS3_HASH_HDR_TCP)
+#define HNS3_HASH_IPV4_UDP	(HNS3_HASH_HDR_ETH | \
+				 HNS3_HASH_HDR_IPV4 | \
+				 HNS3_HASH_HDR_UDP)
+#define HNS3_HASH_IPV4_SCTP	(HNS3_HASH_HDR_ETH | \
+				 HNS3_HASH_HDR_IPV4 | \
+				 HNS3_HASH_HDR_SCTP)
+#define HNS3_HASH_IPV6		(HNS3_HASH_HDR_ETH | HNS3_HASH_HDR_IPV6)
+#define HNS3_HASH_IPV6_TCP	(HNS3_HASH_HDR_ETH | \
+				 HNS3_HASH_HDR_IPV6 | \
+				 HNS3_HASH_HDR_TCP)
+#define HNS3_HASH_IPV6_UDP	(HNS3_HASH_HDR_ETH | \
+				 HNS3_HASH_HDR_IPV6 | \
+				 HNS3_HASH_HDR_UDP)
+#define HNS3_HASH_IPV6_SCTP	(HNS3_HASH_HDR_ETH | \
+				 HNS3_HASH_HDR_IPV6 | \
+				 HNS3_HASH_HDR_SCTP)
+
+static const struct hns3_hash_map_info {
+	/* flow type specified, zero means action works for all flow types. */
+	uint64_t pattern_type;
+	uint64_t rss_pctype; /* packet type with prefix RTE_ETH_RSS_xxx */
+	uint64_t l3l4_types; /* Supported L3/L4 RSS types for this packet type */
+	uint64_t hw_pctype; /* packet type in driver */
+	uint64_t tuple_mask; /* full tuples of the hw_pctype */
+} hash_map_table[] = {
+	/* IPV4 */
+	{ HNS3_HASH_IPV4,
+	  RTE_ETH_RSS_IPV4, HNS3_RSS_SUPPORT_L3_SRC_DST,
+	  HNS3_RSS_PCTYPE_IPV4_NONF, HNS3_RSS_TUPLE_IPV4_NONF_M },
+	{ HNS3_HASH_IPV4,
+	  RTE_ETH_RSS_NONFRAG_IPV4_OTHER, HNS3_RSS_SUPPORT_L3_SRC_DST,
+	  HNS3_RSS_PCTYPE_IPV4_NONF, HNS3_RSS_TUPLE_IPV4_NONF_M },
+	{ HNS3_HASH_IPV4,
+	  RTE_ETH_RSS_FRAG_IPV4, HNS3_RSS_SUPPORT_L3_SRC_DST,
+	  HNS3_RSS_PCTYPE_IPV4_FLAG, HNS3_RSS_TUPLE_IPV4_FLAG_M },
+	{ HNS3_HASH_IPV4_TCP,
+	  RTE_ETH_RSS_NONFRAG_IPV4_TCP, HNS3_RSS_SUPPORT_L3L4,
+	  HNS3_RSS_PCTYPE_IPV4_TCP, HNS3_RSS_TUPLE_IPV4_TCP_M },
+	{ HNS3_HASH_IPV4_UDP,
+	  RTE_ETH_RSS_NONFRAG_IPV4_UDP, HNS3_RSS_SUPPORT_L3L4,
+	  HNS3_RSS_PCTYPE_IPV4_UDP, HNS3_RSS_TUPLE_IPV4_UDP_M },
+	{ HNS3_HASH_IPV4_SCTP,
+	  RTE_ETH_RSS_NONFRAG_IPV4_SCTP, HNS3_RSS_SUPPORT_L3L4,
+	  HNS3_RSS_PCTYPE_IPV4_SCTP, HNS3_RSS_TUPLE_IPV4_SCTP_M },
+	/* IPV6 */
+	{ HNS3_HASH_IPV6,
+	  RTE_ETH_RSS_IPV6, HNS3_RSS_SUPPORT_L3_SRC_DST,
+	  HNS3_RSS_PCTYPE_IPV6_NONF, HNS3_RSS_TUPLE_IPV6_NONF_M },
+	{ HNS3_HASH_IPV6,
+	  RTE_ETH_RSS_NONFRAG_IPV6_OTHER, HNS3_RSS_SUPPORT_L3_SRC_DST,
+	  HNS3_RSS_PCTYPE_IPV6_NONF, HNS3_RSS_TUPLE_IPV6_NONF_M },
+	{ HNS3_HASH_IPV6,
+	  RTE_ETH_RSS_FRAG_IPV6, HNS3_RSS_SUPPORT_L3_SRC_DST,
+	  HNS3_RSS_PCTYPE_IPV6_FLAG, HNS3_RSS_TUPLE_IPV6_FLAG_M },
+	{ HNS3_HASH_IPV6_TCP,
+	  RTE_ETH_RSS_NONFRAG_IPV6_TCP, HNS3_RSS_SUPPORT_L3L4,
+	  HNS3_RSS_PCTYPE_IPV6_TCP, HNS3_RSS_TUPLE_IPV6_TCP_M },
+	{ HNS3_HASH_IPV6_UDP,
+	  RTE_ETH_RSS_NONFRAG_IPV6_UDP, HNS3_RSS_SUPPORT_L3L4,
+	  HNS3_RSS_PCTYPE_IPV6_UDP, HNS3_RSS_TUPLE_IPV6_UDP_M },
+	{ HNS3_HASH_IPV6_SCTP,
+	  RTE_ETH_RSS_NONFRAG_IPV6_SCTP, HNS3_RSS_SUPPORT_L3L4,
+	  HNS3_RSS_PCTYPE_IPV6_SCTP, HNS3_RSS_TUPLE_IPV6_SCTP_M },
+};
+
 static const uint8_t full_mask[VNI_OR_TNI_LEN] = { 0xFF, 0xFF, 0xFF };
 static const uint8_t zero_mask[VNI_OR_TNI_LEN] = { 0x00, 0x00, 0x00 };
 
@@ -79,7 +198,7 @@ net_addr_to_host(uint32_t *dst, const rte_be32_t *src, size_t len)
 }
 
 /*
- * This function is used to find rss general action.
+ * This function is used to parse filter type.
  * 1. As we know RSS is used to spread packets among several queues, the flow
  *    API provide the struct rte_flow_action_rss, user could config its field
  *    sush as: func/level/types/key/queue to control RSS function.
@@ -87,16 +206,18 @@ net_addr_to_host(uint32_t *dst, const rte_be32_t *src, size_t len)
  *    implemented by FDIR + RSS in hns3 hardware, user can create one FDIR rule
  *    which action is RSS queues region.
  * 3. When action is RSS, we use the following rule to distinguish:
- *    Case 1: pattern have ETH and action's queue_num > 0, indicate it is queue
- *            region configuration.
+ *    Case 1: pattern has ETH and all fields in RSS action except 'queues' are
+ *            zero or default, indicate it is queue region configuration.
  *    Case other: an rss general action.
  */
-static const struct rte_flow_action *
-hns3_find_rss_general_action(const struct rte_flow_item pattern[],
-			     const struct rte_flow_action actions[])
+static void
+hns3_parse_filter_type(const struct rte_flow_item pattern[],
+		       const struct rte_flow_action actions[],
+		       struct hns3_filter_info *filter_info)
 {
 	const struct rte_flow_action_rss *rss_act;
 	const struct rte_flow_action *act = NULL;
+	bool only_has_queues = false;
 	bool have_eth = false;
 
 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
@@ -105,8 +226,10 @@ hns3_find_rss_general_action(const struct rte_flow_item pattern[],
 			break;
 		}
 	}
-	if (!act)
-		return NULL;
+	if (act == NULL) {
+		filter_info->type = RTE_ETH_FILTER_FDIR;
+		return;
+	}
 
 	for (; pattern->type != RTE_FLOW_ITEM_TYPE_END; pattern++) {
 		if (pattern->type == RTE_FLOW_ITEM_TYPE_ETH) {
@@ -116,18 +239,20 @@ hns3_find_rss_general_action(const struct rte_flow_item pattern[],
 	}
 
 	rss_act = act->conf;
-	if (have_eth && rss_act->queue_num) {
+	only_has_queues = (rss_act->queue_num > 0) &&
+			  (rss_act->func == RTE_ETH_HASH_FUNCTION_DEFAULT &&
+			   rss_act->types == 0 && rss_act->key_len == 0);
+	if (have_eth && only_has_queues) {
 		/*
-		 * Pattern have ETH and action's queue_num > 0, indicate this is
-		 * queue region configuration.
-		 * Because queue region is implemented by FDIR + RSS in hns3
-		 * hardware, it needs to enter FDIR process, so here return NULL
-		 * to avoid enter RSS process.
+		 * Pattern has ETH and all fields in RSS action except 'queues'
+		 * are zero or default, which indicates this is queue region
+		 * configuration.
 		 */
-		return NULL;
+		filter_info->type = RTE_ETH_FILTER_FDIR;
+		return;
 	}
 
-	return act;
+	filter_info->type = RTE_ETH_FILTER_HASH;
 }
 
 static inline struct hns3_flow_counter *
@@ -1246,7 +1371,6 @@ hns3_filterlist_flush(struct rte_eth_dev *dev)
 {
 	struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 	struct hns3_fdir_rule_ele *fdir_rule_ptr;
-	struct hns3_rss_conf_ele *rss_filter_ptr;
 	struct hns3_flow_mem *flow_node;
 
 	fdir_rule_ptr = TAILQ_FIRST(&hw->flow_fdir_list);
@@ -1256,13 +1380,6 @@ hns3_filterlist_flush(struct rte_eth_dev *dev)
 		fdir_rule_ptr = TAILQ_FIRST(&hw->flow_fdir_list);
 	}
 
-	rss_filter_ptr = TAILQ_FIRST(&hw->flow_rss_list);
-	while (rss_filter_ptr) {
-		TAILQ_REMOVE(&hw->flow_rss_list, rss_filter_ptr, entries);
-		rte_free(rss_filter_ptr);
-		rss_filter_ptr = TAILQ_FIRST(&hw->flow_rss_list);
-	}
-
 	flow_node = TAILQ_FIRST(&hw->flow_list);
 	while (flow_node) {
 		TAILQ_REMOVE(&hw->flow_list, flow_node, entries);
@@ -1328,196 +1445,422 @@ hns3_action_rss_same(const struct rte_flow_action_rss *comp,
 }
 
 static bool
-hns3_rss_input_tuple_supported(struct hns3_hw *hw,
-			       const struct rte_flow_action_rss *rss)
+hns3_valid_ipv6_sctp_rss_types(struct hns3_hw *hw, uint64_t types)
 {
 	/*
-	 * For IP packet, it is not supported to use src/dst port fields to RSS
-	 * hash for the following packet types.
-	 * - IPV4 FRAG | IPV4 NONFRAG | IPV6 FRAG | IPV6 NONFRAG
-	 * Besides, for Kunpeng920, the NIC HW is not supported to use src/dst
-	 * port fields to RSS hash for IPV6 SCTP packet type. However, the
-	 * Kunpeng930 and future kunpeng series support to use src/dst port
-	 * fields to RSS hash for IPv6 SCTP packet type.
+	 * Some hardware don't support to use src/dst port fields to hash
+	 * for IPV6 SCTP packet type.
 	 */
-	if (rss->types & (RTE_ETH_RSS_L4_DST_ONLY | RTE_ETH_RSS_L4_SRC_ONLY) &&
-	    (rss->types & RTE_ETH_RSS_IP ||
-	    (!hw->rss_info.ipv6_sctp_offload_supported &&
-	    rss->types & RTE_ETH_RSS_NONFRAG_IPV6_SCTP)))
+	if (types & RTE_ETH_RSS_NONFRAG_IPV6_SCTP &&
+	    types & HNS3_RSS_SUPPORT_L4_SRC_DST &&
+	    !hw->rss_info.ipv6_sctp_offload_supported)
 		return false;
 
 	return true;
 }
 
-/*
- * This function is used to parse rss action validation.
- */
 static int
-hns3_parse_rss_filter(struct rte_eth_dev *dev,
-		      const struct rte_flow_action *actions,
-		      struct rte_flow_error *error)
+hns3_flow_parse_hash_func(const struct rte_flow_action_rss *rss_act,
+			  struct hns3_flow_rss_conf *rss_conf,
+			  struct rte_flow_error *error)
 {
-	struct hns3_adapter *hns = dev->data->dev_private;
-	struct hns3_hw *hw = &hns->hw;
-	const struct rte_flow_action_rss *rss;
-	const struct rte_flow_action *act;
-	uint32_t act_index = 0;
-	uint16_t n;
+	if (rss_act->func >= RTE_ETH_HASH_FUNCTION_MAX)
+		return rte_flow_error_set(error, ENOTSUP,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  NULL, "RSS hash func are not supported");
 
-	NEXT_ITEM_OF_ACTION(act, actions, act_index);
-	rss = act->conf;
+	rss_conf->conf.func = rss_act->func;
+	return 0;
+}
 
-	if (rss == NULL) {
+static int
+hns3_flow_parse_hash_key(struct hns3_hw *hw,
+			 const struct rte_flow_action_rss *rss_act,
+			 struct hns3_flow_rss_conf *rss_conf,
+			 struct rte_flow_error *error)
+{
+	if (rss_act->key_len != hw->rss_key_size)
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  act, "no valid queues");
-	}
+					  NULL, "invalid RSS key length");
+
+	if (rss_act->key != NULL)
+		memcpy(rss_conf->key, rss_act->key, rss_act->key_len);
+	else
+		memcpy(rss_conf->key, hns3_hash_key,
+			RTE_MIN(sizeof(hns3_hash_key), rss_act->key_len));
+	/* Need to record if user sets hash key. */
+	rss_conf->conf.key = rss_act->key;
+	rss_conf->conf.key_len = rss_act->key_len;
 
-	if (rss->queue_num > HNS3_RSS_QUEUES_BUFFER_NUM)
+	return 0;
+}
+
+static int
+hns3_flow_parse_queues(struct hns3_hw *hw,
+		       const struct rte_flow_action_rss *rss_act,
+		       struct hns3_flow_rss_conf *rss_conf,
+		       struct rte_flow_error *error)
+{
+	uint16_t i;
+
+	if (rss_act->queue_num > hw->rss_ind_tbl_size)
 		return rte_flow_error_set(error, ENOTSUP,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, act,
-					  "queue number configured exceeds "
-					  "queue buffer size driver supported");
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  NULL,
+					  "queue number can not exceed RSS indirection table.");
 
-	for (n = 0; n < rss->queue_num; n++) {
-		if (rss->queue[n] < hw->alloc_rss_size)
-			continue;
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, act,
-					  "queue id must be less than queue number allocated to a TC");
+	if (rss_act->queue_num > HNS3_RSS_QUEUES_BUFFER_NUM)
+		return rte_flow_error_set(error, ENOTSUP,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  NULL,
+					  "queue number configured exceeds queue buffer size driver supported");
+
+	for (i = 0; i < rss_act->queue_num; i++) {
+		if (rss_act->queue[i] >= hw->alloc_rss_size)
+			return rte_flow_error_set(error, EINVAL,
+						RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+						NULL,
+						"queue id must be less than queue number allocated to a TC");
 	}
 
-	if (!(rss->types & HNS3_ETH_RSS_SUPPORT) && rss->types)
+	memcpy(rss_conf->queue, rss_act->queue,
+	       rss_act->queue_num * sizeof(rss_conf->queue[0]));
+	rss_conf->conf.queue = rss_conf->queue;
+	rss_conf->conf.queue_num = rss_act->queue_num;
+
+	return 0;
+}
+
+static int
+hns3_flow_get_hw_pctype(struct hns3_hw *hw,
+			const struct rte_flow_action_rss *rss_act,
+			const struct hns3_hash_map_info *map,
+			struct hns3_flow_rss_conf *rss_conf,
+			struct rte_flow_error *error)
+{
+	uint64_t l3l4_src_dst, l3l4_refine, left_types;
+
+	if (rss_act->types == 0) {
+		/* Disable RSS hash of this packet type if types is zero. */
+		rss_conf->hw_pctypes |= map->hw_pctype;
+		return 0;
+	}
+
+	/*
+	 * Can not have extra types except rss_pctype and l3l4_type in this map.
+	 */
+	left_types = ~map->rss_pctype & rss_act->types;
+	if (left_types & ~map->l3l4_types)
 		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  act,
-					  "Flow types is unsupported by "
-					  "hns3's RSS");
-	if (rss->func >= RTE_ETH_HASH_FUNCTION_MAX)
-		return rte_flow_error_set(error, ENOTSUP,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, act,
-					  "RSS hash func are not supported");
-	if (rss->level)
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+					  "cannot set extra types.");
+
+	l3l4_src_dst = left_types;
+	/* L3/L4 SRC and DST shouldn't be specified at the same time. */
+	l3l4_refine = rte_eth_rss_hf_refine(l3l4_src_dst);
+	if (l3l4_refine != l3l4_src_dst)
 		return rte_flow_error_set(error, ENOTSUP,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, act,
-					  "a nonzero RSS encapsulation level is not supported");
-	if (rss->key_len && rss->key_len != hw->rss_key_size)
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+					  "cannot specify L3_SRC/DST_ONLY or L4_SRC/DST_ONLY at the same.");
+
+	if (!hns3_valid_ipv6_sctp_rss_types(hw, rss_act->types))
 		return rte_flow_error_set(error, ENOTSUP,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, act,
-					  "invalid RSS key length");
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+					  "hardware doesn't support to use L4 src/dst to hash for IPV6-SCTP.");
 
-	if (!hns3_rss_input_tuple_supported(hw, rss))
-		return rte_flow_error_set(error, EINVAL,
+	rss_conf->hw_pctypes |= map->hw_pctype;
+
+	return 0;
+}
+
+static int
+hns3_flow_parse_rss_types_by_ptype(struct hns3_hw *hw,
+				   const struct rte_flow_action_rss *rss_act,
+				   uint64_t pattern_type,
+				   struct hns3_flow_rss_conf *rss_conf,
+				   struct rte_flow_error *error)
+{
+	const struct hns3_hash_map_info *map;
+	bool matched = false;
+	uint16_t i;
+	int ret;
+
+	for (i = 0; i < RTE_DIM(hash_map_table); i++) {
+		map = &hash_map_table[i];
+		if (map->pattern_type != pattern_type) {
+			/*
+			 * If the target pattern type is already matched with
+			 * the one before this pattern in the hash map table,
+			 * no need to continue walk.
+			 */
+			if (matched)
+				break;
+			continue;
+		}
+		matched = true;
+
+		/*
+		 * If pattern type is matched and the 'types' is zero, all packet flow
+		 * types related to this pattern type disable RSS hash.
+		 * Otherwise, RSS types must match the pattern type and cannot have no
+		 * extra or unsupported types.
+		 */
+		if (rss_act->types != 0 && !(map->rss_pctype & rss_act->types))
+			continue;
+
+		ret = hns3_flow_get_hw_pctype(hw, rss_act, map, rss_conf, error);
+		if (ret != 0)
+			return ret;
+	}
+
+	if (rss_conf->hw_pctypes != 0)
+		return 0;
+
+	if (matched)
+		return rte_flow_error_set(error, ENOTSUP,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  &rss->types,
-					  "input RSS types are not supported");
+					  NULL, "RSS types are unsupported");
 
-	act_index++;
+	return rte_flow_error_set(error, ENOTSUP,
+				  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+				  NULL, "Pattern specified is unsupported");
+}
 
-	/* Check if the next not void action is END */
-	NEXT_ITEM_OF_ACTION(act, actions, act_index);
-	if (act->type != RTE_FLOW_ACTION_TYPE_END)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION,
-					  act, "Not supported action.");
+static uint64_t
+hns3_flow_get_all_hw_pctypes(uint64_t types)
+{
+	uint64_t hw_pctypes = 0;
+	uint16_t i;
 
-	return 0;
+	for (i = 0; i < RTE_DIM(hash_map_table); i++) {
+		if (types & hash_map_table[i].rss_pctype)
+			hw_pctypes |= hash_map_table[i].hw_pctype;
+	}
+
+	return hw_pctypes;
 }
 
 static int
-hns3_disable_rss(struct hns3_hw *hw)
+hns3_flow_parse_rss_types(struct hns3_hw *hw,
+			  const struct rte_flow_action_rss *rss_act,
+			  uint64_t pattern_type,
+			  struct hns3_flow_rss_conf *rss_conf,
+			  struct rte_flow_error *error)
+{
+	rss_conf->conf.types = rss_act->types;
+
+	/* no pattern specified to set global RSS types. */
+	if (pattern_type == 0) {
+		if (rss_act->types & ~HNS3_ETH_RSS_SUPPORT)
+			hns3_warn(hw, "some types in the requested RSS types (0x%" PRIx64 ") aren't supported, they are ignored.",
+				  rss_act->types);
+		rss_conf->hw_pctypes =
+				hns3_flow_get_all_hw_pctypes(rss_act->types);
+		return 0;
+	}
+
+	return hns3_flow_parse_rss_types_by_ptype(hw, rss_act, pattern_type,
+						  rss_conf, error);
+}
+
+static int
+hns3_flow_parse_hash_global_conf(struct rte_eth_dev *dev,
+				 const struct rte_flow_action_rss *rss_act,
+				 struct hns3_flow_rss_conf *rss_conf,
+				 struct rte_flow_error *error)
 {
+	struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 	int ret;
 
-	ret = hns3_set_rss_tuple_by_rss_hf(hw, 0);
-	if (ret)
+	ret = hns3_flow_parse_hash_func(rss_act, rss_conf, error);
+	if (ret != 0)
 		return ret;
-	hw->rss_info.rss_hf = 0;
 
-	return 0;
+	if (rss_act->queue_num > 0) {
+		ret = hns3_flow_parse_queues(hw, rss_act, rss_conf, error);
+		if (ret != 0)
+			return ret;
+	}
+
+	if (rss_act->key_len > 0) {
+		ret = hns3_flow_parse_hash_key(hw, rss_act, rss_conf, error);
+		if (ret != 0)
+			return ret;
+	}
+
+	return hns3_flow_parse_rss_types(hw, rss_act, rss_conf->pattern_type,
+					 rss_conf, error);
 }
 
 static int
-hns3_parse_rss_algorithm(struct hns3_hw *hw, struct hns3_flow_rss_conf *rss_conf,
-			 uint8_t *hash_algo)
-{
-	const uint8_t hash_func_map[] = {
-		[RTE_ETH_HASH_FUNCTION_DEFAULT] = HNS3_RSS_HASH_ALGO_TOEPLITZ,
-		[RTE_ETH_HASH_FUNCTION_TOEPLITZ] = HNS3_RSS_HASH_ALGO_TOEPLITZ,
-		[RTE_ETH_HASH_FUNCTION_SIMPLE_XOR] = HNS3_RSS_HASH_ALGO_SIMPLE,
-		[RTE_ETH_HASH_FUNCTION_SYMMETRIC_TOEPLITZ] = HNS3_RSS_HASH_ALGO_SYMMETRIC_TOEP,
-	};
-	uint8_t key[HNS3_RSS_KEY_SIZE_MAX] = {0};
-	int ret;
+hns3_flow_parse_pattern_type(const struct rte_flow_item pattern[],
+			     uint64_t *ptype, struct rte_flow_error *error)
+{
+	enum rte_flow_item_type pre_type = RTE_FLOW_ITEM_TYPE_VOID;
+	const char *message = "Pattern specified isn't supported";
+	uint64_t item_hdr, pattern_hdrs = 0;
+	enum rte_flow_item_type cur_type;
 
-	if (rss_conf->conf.func == RTE_ETH_HASH_FUNCTION_DEFAULT) {
-		ret = hns3_rss_get_algo_key(hw, hash_algo, key,
-					    hw->rss_key_size);
-		if (ret != 0) {
-			hns3_err(hw, "fail to get current RSS hash algorithm, ret = %d",
-				 ret);
-			return ret;
+	for (; pattern->type != RTE_FLOW_ITEM_TYPE_END; pattern++) {
+		if (pattern->type == RTE_FLOW_ITEM_TYPE_VOID)
+			continue;
+		if (pattern->mask || pattern->spec || pattern->last) {
+			message = "Header info shouldn't be specified";
+			goto unsup;
 		}
 
-		/*
-		 * During the phase of reset recovery, the hash algorithm
-		 * obtained from hardware may not be the one used(saved in
-		 * rte_flow_hash_algo) when this rule is delivered.
-		 */
-		if (__atomic_load_n(&hw->reset.resetting, __ATOMIC_RELAXED) &&
-		    *hash_algo != rss_conf->hash_algo)
-			*hash_algo = rss_conf->hash_algo;
+		/* Check the sub-item allowed by the previous item . */
+		if (pre_type >= RTE_DIM(hash_pattern_next_allow_items) ||
+		    !(hash_pattern_next_allow_items[pre_type] &
+				BIT_ULL(pattern->type)))
+			goto unsup;
+
+		cur_type = pattern->type;
+		/* Unsupported for current type being greater than array size. */
+		if (cur_type >= RTE_DIM(hash_pattern_item_header))
+			goto unsup;
+
+		/* The value is zero, which means unsupported current header. */
+		item_hdr = hash_pattern_item_header[cur_type];
+		if (item_hdr == 0)
+			goto unsup;
+
+		/* Have duplicate pattern header. */
+		if (item_hdr & pattern_hdrs)
+			goto unsup;
+		pre_type = cur_type;
+		pattern_hdrs |= item_hdr;
+	}
 
+	if (pattern_hdrs != 0) {
+		*ptype = pattern_hdrs;
 		return 0;
 	}
 
-	*hash_algo = hash_func_map[rss_conf->conf.func];
+unsup:
+	return rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
+				  pattern, message);
+}
+
+static int
+hns3_flow_parse_pattern_act(struct rte_eth_dev *dev,
+			    const struct rte_flow_item pattern[],
+			    const struct rte_flow_action_rss *rss_act,
+			    struct hns3_flow_rss_conf *rss_conf,
+			    struct rte_flow_error *error)
+{
+	struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+	int ret;
+
+	ret = hns3_flow_parse_hash_func(rss_act, rss_conf, error);
+	if (ret != 0)
+		return ret;
+
+	if (rss_act->key_len > 0) {
+		ret = hns3_flow_parse_hash_key(hw, rss_act, rss_conf, error);
+		if (ret != 0)
+			return ret;
+	}
+
+	if (rss_act->queue_num > 0) {
+		ret = hns3_flow_parse_queues(hw, rss_act, rss_conf, error);
+		if (ret != 0)
+			return ret;
+	}
+
+	ret = hns3_flow_parse_pattern_type(pattern, &rss_conf->pattern_type,
+					   error);
+	if (ret != 0)
+		return ret;
+
+	ret = hns3_flow_parse_rss_types(hw, rss_act, rss_conf->pattern_type,
+					rss_conf, error);
+	if (ret != 0)
+		return ret;
+
+	if (rss_act->func != RTE_ETH_HASH_FUNCTION_DEFAULT ||
+	    rss_act->key_len > 0 || rss_act->queue_num > 0)
+		hns3_warn(hw, "hash func, key and queues are global config, which work for all flow types. "
+			  "Recommend: don't set them together with pattern.");
 
 	return 0;
 }
 
+static bool
+hns3_rss_action_is_dup(struct hns3_hw *hw,
+		       const struct hns3_flow_rss_conf *conf)
+{
+	struct hns3_rss_conf_ele *filter;
+
+	TAILQ_FOREACH(filter, &hw->flow_rss_list, entries) {
+		if (conf->pattern_type != filter->filter_info.pattern_type)
+			continue;
+
+		if (hns3_action_rss_same(&filter->filter_info.conf, &conf->conf))
+			return true;
+	}
+
+	return false;
+}
+
+/*
+ * This function is used to parse rss action validation.
+ */
 static int
-hns3_hw_rss_hash_set(struct hns3_hw *hw, struct hns3_flow_rss_conf *conf)
+hns3_parse_rss_filter(struct rte_eth_dev *dev,
+		      const struct rte_flow_item pattern[],
+		      const struct rte_flow_action *actions,
+		      struct hns3_flow_rss_conf *rss_conf,
+		      struct rte_flow_error *error)
 {
-	struct rte_flow_action_rss *rss_config = &conf->conf;
-	uint8_t rss_key[HNS3_RSS_KEY_SIZE_MAX] = {0};
-	bool use_default_key = false;
-	uint64_t flow_types;
-	uint8_t hash_algo;
+	struct hns3_adapter *hns = dev->data->dev_private;
+	const struct rte_flow_action_rss *rss_act;
+	const struct rte_flow_action *act;
+	const struct rte_flow_item *pat;
+	struct hns3_hw *hw = &hns->hw;
+	uint32_t index = 0;
 	int ret;
 
-	if (rss_config->key == NULL || rss_config->key_len != hw->rss_key_size) {
-		hns3_warn(hw, "Default RSS hash key to be set");
-		memcpy(rss_key, hns3_hash_key,
-			RTE_MIN(sizeof(hns3_hash_key), hw->rss_key_size));
-		use_default_key = true;
+	NEXT_ITEM_OF_ACTION(act, actions, index);
+	if (actions[1].type != RTE_FLOW_ACTION_TYPE_END)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION,
+					  &actions[1],
+					  "Only support one action for RSS.");
+
+	rss_act = (const struct rte_flow_action_rss *)act->conf;
+	if (rss_act == NULL) {
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  act, "lost RSS action configuration");
 	}
 
-	ret = hns3_parse_rss_algorithm(hw, conf, &hash_algo);
-	if (ret)
+	if (rss_act->level != 0)
+		return rte_flow_error_set(error, ENOTSUP,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  act,
+					  "RSS level is not supported");
+
+	index = 0;
+	NEXT_ITEM_OF_PATTERN(pat, pattern, index);
+	if (pat[0].type == RTE_FLOW_ITEM_TYPE_END) {
+		rss_conf->pattern_type = 0;
+		ret = hns3_flow_parse_hash_global_conf(dev, rss_act,
+						       rss_conf, error);
+	} else {
+		ret = hns3_flow_parse_pattern_act(dev, pat, rss_act,
+						  rss_conf, error);
+	}
+	if (ret != 0)
 		return ret;
 
-	ret = hns3_rss_set_algo_key(hw, hash_algo,
-				    use_default_key ? rss_key : rss_config->key,
-				    hw->rss_key_size);
-	if (ret)
-		return ret;
-	conf->hash_algo = hash_algo;
-
-	/* Filter the unsupported flow types */
-	flow_types = rss_config->types ?
-		     rss_config->types & HNS3_ETH_RSS_SUPPORT :
-		     hw->rss_info.rss_hf;
-	if (flow_types != rss_config->types)
-		hns3_warn(hw, "modified RSS types based on hardware support,"
-			  " requested:0x%" PRIx64 " configured:0x%" PRIx64,
-			  rss_config->types, flow_types);
-
-	ret = hns3_set_rss_tuple_by_rss_hf(hw, flow_types);
-	if (ret)
-		hns3_err(hw, "Update RSS tuples by rss hf failed %d", ret);
+	if (hns3_rss_action_is_dup(hw, rss_conf))
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  act, "duplicate RSS rule");
 
-	return ret;
+	return 0;
 }
 
 static int
@@ -1543,44 +1886,106 @@ hns3_update_indir_table(struct hns3_hw *hw,
 	return hns3_set_rss_indir_table(hw, indir_tbl, hw->rss_ind_tbl_size);
 }
 
+static uint64_t
+hns3_flow_get_pctype_tuple_mask(uint64_t hw_pctype)
+{
+	uint64_t tuple_mask = 0;
+	uint16_t i;
+
+	for (i = 0; i < RTE_DIM(hash_map_table); i++) {
+		if (hw_pctype == hash_map_table[i].hw_pctype) {
+			tuple_mask = hash_map_table[i].tuple_mask;
+			break;
+		}
+	}
+
+	return tuple_mask;
+}
+
 static int
-hns3_reset_rss_filter(struct hns3_hw *hw,
-		      const struct hns3_flow_rss_conf *conf)
+hns3_flow_set_rss_ptype_tuple(struct hns3_hw *hw,
+			      struct hns3_flow_rss_conf *rss_conf)
 {
+	uint64_t old_tuple_fields, new_tuple_fields;
+	uint64_t hw_pctypes, tuples, tuple_mask = 0;
+	bool cfg_global_tuple;
 	int ret;
 
-	if (!conf->valid)
-		return 0;
+	cfg_global_tuple = (rss_conf->pattern_type == 0);
+	if (!cfg_global_tuple) {
+		/*
+		 * To ensure that different packets do not affect each other,
+		 * we have to first read all tuple fields, and then only modify
+		 * the tuples for the specified packet type.
+		 */
+		ret = hns3_get_rss_tuple_field(hw, &old_tuple_fields);
+		if (ret != 0)
+			return ret;
 
-	ret = hns3_disable_rss(hw);
-	if (ret)
-		hns3_err(hw, "RSS disable failed(%d)", ret);
+		new_tuple_fields = old_tuple_fields;
+		hw_pctypes = rss_conf->hw_pctypes;
+		while (hw_pctypes > 0) {
+			uint32_t idx = rte_bsf64(hw_pctypes);
+			uint64_t pctype = BIT_ULL(idx);
+
+			tuple_mask = hns3_flow_get_pctype_tuple_mask(pctype);
+			tuples = hns3_rss_calc_tuple_filed(hw,
+							rss_conf->conf.types);
+			new_tuple_fields &= ~tuple_mask;
+			new_tuple_fields |= tuples;
+			hw_pctypes &= ~pctype;
+		}
+	} else {
+		new_tuple_fields =
+			hns3_rss_calc_tuple_filed(hw, rss_conf->conf.types);
+	}
 
-	return ret;
+	ret = hns3_set_rss_tuple_field(hw, new_tuple_fields);
+	if (ret != 0)
+		return ret;
+
+	hns3_info(hw, "RSS tuple fields changed from 0x%" PRIx64 " to 0x%" PRIx64,
+		  old_tuple_fields, new_tuple_fields);
+
+	return 0;
 }
 
 static int
-hns3_config_rss_filter(struct hns3_hw *hw, struct hns3_flow_rss_conf *conf)
+hns3_config_rss_filter(struct hns3_hw *hw,
+		       struct hns3_flow_rss_conf *rss_conf)
 {
 	struct rte_flow_action_rss *rss_act;
-	uint16_t num;
 	int ret;
 
-	rss_act = &conf->conf;
-	/* Set rx queues to use */
-	num = RTE_MIN(hw->data->nb_rx_queues, rss_act->queue_num);
-	if (rss_act->queue_num > num)
-		hns3_warn(hw, "Config queue numbers %u are beyond the scope of truncated",
-			  rss_act->queue_num);
-	hns3_info(hw, "Max of contiguous %u PF queues are configured", num);
-	if (num) {
-		ret = hns3_update_indir_table(hw, rss_act, num);
-		if (ret)
+	rss_act = &rss_conf->conf;
+	if (rss_act->queue_num > 0) {
+		ret = hns3_update_indir_table(hw, rss_act, rss_act->queue_num);
+		if (ret) {
+			hns3_err(hw, "set queues action failed, ret = %d", ret);
+			return ret;
+		}
+	}
+
+	if (rss_act->key_len > 0 ||
+	    rss_act->func != RTE_ETH_HASH_FUNCTION_DEFAULT) {
+		ret = hns3_update_rss_algo_key(hw, rss_act->func, rss_conf->key,
+					       rss_act->key_len);
+		if (ret != 0) {
+			hns3_err(hw, "set func or hash key action failed, ret = %d",
+				 ret);
 			return ret;
+		}
+	}
+
+	if (rss_conf->hw_pctypes > 0) {
+		ret = hns3_flow_set_rss_ptype_tuple(hw, rss_conf);
+		if (ret != 0) {
+			hns3_err(hw, "set types action failed, ret = %d", ret);
+			return ret;
+		}
 	}
 
-	/* Set hash algorithm and flow types by the user's config */
-	return hns3_hw_rss_hash_set(hw, conf);
+	return 0;
 }
 
 static int
@@ -1589,50 +1994,44 @@ hns3_clear_rss_filter(struct rte_eth_dev *dev)
 	struct hns3_adapter *hns = dev->data->dev_private;
 	struct hns3_rss_conf_ele *rss_filter_ptr;
 	struct hns3_hw *hw = &hns->hw;
-	int rss_rule_succ_cnt = 0; /* count for success of clearing RSS rules */
-	int rss_rule_fail_cnt = 0; /* count for failure of clearing RSS rules */
-	int ret = 0;
 
 	rss_filter_ptr = TAILQ_FIRST(&hw->flow_rss_list);
 	while (rss_filter_ptr) {
 		TAILQ_REMOVE(&hw->flow_rss_list, rss_filter_ptr, entries);
-		ret = hns3_reset_rss_filter(hw, &rss_filter_ptr->filter_info);
-		if (ret)
-			rss_rule_fail_cnt++;
-		else
-			rss_rule_succ_cnt++;
 		rte_free(rss_filter_ptr);
 		rss_filter_ptr = TAILQ_FIRST(&hw->flow_rss_list);
 	}
 
-	if (rss_rule_fail_cnt) {
-		hns3_err(hw, "fail to delete all RSS filters, success num = %d fail num = %d",
-			 rss_rule_succ_cnt, rss_rule_fail_cnt);
-		ret = -EIO;
-	}
-
-	return ret;
+	return hns3_config_rss(hns);
 }
 
 static int
-hns3_restore_rss_filter(struct hns3_hw *hw)
+hns3_reconfig_all_rss_filter(struct hns3_hw *hw)
 {
 	struct hns3_rss_conf_ele *filter;
-	int ret = 0;
+	uint32_t rule_no = 0;
+	int ret;
 
-	pthread_mutex_lock(&hw->flows_lock);
 	TAILQ_FOREACH(filter, &hw->flow_rss_list, entries) {
-		if (!filter->filter_info.valid)
-			continue;
-
 		ret = hns3_config_rss_filter(hw, &filter->filter_info);
 		if (ret != 0) {
-			hns3_err(hw, "restore RSS filter failed, ret=%d", ret);
-			goto out;
+			hns3_err(hw, "config %uth RSS filter failed, ret = %d",
+				 rule_no, ret);
+			return ret;
 		}
+		rule_no++;
 	}
 
-out:
+	return 0;
+}
+
+static int
+hns3_restore_rss_filter(struct hns3_hw *hw)
+{
+	int ret;
+
+	pthread_mutex_lock(&hw->flows_lock);
+	ret = hns3_reconfig_all_rss_filter(hw);
 	pthread_mutex_unlock(&hw->flows_lock);
 
 	return ret;
@@ -1651,38 +2050,6 @@ hns3_restore_filter(struct hns3_adapter *hns)
 	return hns3_restore_rss_filter(hw);
 }
 
-static bool
-hns3_rss_action_is_dup(struct hns3_hw *hw,
-		       const struct rte_flow_action_rss *act)
-{
-	struct hns3_rss_conf_ele *filter;
-
-	TAILQ_FOREACH(filter, &hw->flow_rss_list, entries) {
-		if (!filter->filter_info.valid)
-			continue;
-
-		if (hns3_action_rss_same(&filter->filter_info.conf, act))
-			return true;
-	}
-
-	return false;
-}
-
-static int
-hns3_flow_parse_rss(struct rte_eth_dev *dev,
-		    struct hns3_flow_rss_conf *conf)
-{
-	struct hns3_adapter *hns = dev->data->dev_private;
-	struct hns3_hw *hw = &hns->hw;
-
-	if (hns3_rss_action_is_dup(hw, &conf->conf)) {
-		hns3_err(hw, "duplicate RSS configuration");
-		return -EINVAL;
-	}
-
-	return hns3_config_rss_filter(hw, conf);
-}
-
 static int
 hns3_flow_args_check(const struct rte_flow_attr *attr,
 		     const struct rte_flow_item pattern[],
@@ -1716,32 +2083,55 @@ static int
 hns3_flow_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
 		   const struct rte_flow_item pattern[],
 		   const struct rte_flow_action actions[],
-		   struct rte_flow_error *error)
+		   struct rte_flow_error *error,
+		   struct hns3_filter_info *filter_info)
 {
-	struct hns3_fdir_rule fdir_rule;
+	union hns3_filter_conf *conf;
 	int ret;
 
 	ret = hns3_flow_args_check(attr, pattern, actions, error);
 	if (ret)
 		return ret;
 
-	if (hns3_find_rss_general_action(pattern, actions))
-		return hns3_parse_rss_filter(dev, actions, error);
+	hns3_parse_filter_type(pattern, actions, filter_info);
+	conf = &filter_info->conf;
+	if (filter_info->type == RTE_ETH_FILTER_HASH)
+		return hns3_parse_rss_filter(dev, pattern, actions,
+					     &conf->rss_conf, error);
 
-	memset(&fdir_rule, 0, sizeof(struct hns3_fdir_rule));
-	return hns3_parse_fdir_filter(dev, pattern, actions, &fdir_rule, error);
+	return hns3_parse_fdir_filter(dev, pattern, actions,
+				      &conf->fdir_conf, error);
+}
+
+static int
+hns3_flow_rebuild_all_rss_filter(struct hns3_adapter *hns)
+{
+	struct hns3_hw *hw = &hns->hw;
+	int ret;
+
+	ret = hns3_config_rss(hns);
+	if (ret != 0) {
+		hns3_err(hw, "restore original RSS configuration failed, ret = %d.",
+			 ret);
+		return ret;
+	}
+	ret = hns3_reconfig_all_rss_filter(hw);
+	if (ret != 0)
+		hns3_err(hw, "rebuild all RSS filter failed, ret = %d.", ret);
+
+	return ret;
 }
 
 static int
 hns3_flow_create_rss_rule(struct rte_eth_dev *dev,
-			  const struct rte_flow_action *act,
+			  struct hns3_flow_rss_conf *rss_conf,
 			  struct rte_flow *flow)
 {
 	struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
-	const struct rte_flow_action_rss *rss_act;
+	struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
 	struct hns3_rss_conf_ele *rss_filter_ptr;
 	struct hns3_flow_rss_conf *new_conf;
-	struct hns3_rss_conf_ele *filter_ptr;
+	struct rte_flow_action_rss *rss_act;
 	int ret;
 
 	rss_filter_ptr = rte_zmalloc("hns3 rss filter",
@@ -1751,35 +2141,28 @@ hns3_flow_create_rss_rule(struct rte_eth_dev *dev,
 		return -ENOMEM;
 	}
 
-	rss_act = (const struct rte_flow_action_rss *)act->conf;
 	new_conf = &rss_filter_ptr->filter_info;
-	memcpy(&new_conf->conf, rss_act, sizeof(*rss_act));
-	if (rss_act->queue_num > 0) {
-		memcpy(new_conf->queue, rss_act->queue,
-		       rss_act->queue_num * sizeof(new_conf->queue[0]));
+	memcpy(new_conf, rss_conf, sizeof(*new_conf));
+	rss_act = &new_conf->conf;
+	if (rss_act->queue_num > 0)
 		new_conf->conf.queue = new_conf->queue;
-	}
-	if (rss_act->key_len > 0) {
-		if (rss_act->key != NULL) {
-			memcpy(new_conf->key, rss_act->key,
-			       rss_act->key_len * sizeof(new_conf->key[0]));
-			new_conf->conf.key = new_conf->key;
-		}
-	}
+	/*
+	 * There are two ways to deliver hash key action:
+	 * 1> 'key_len' is greater than zero and 'key' isn't NULL.
+	 * 2> 'key_len' is greater than zero, but 'key' is NULL.
+	 * For case 2, we need to keep 'key' of the new_conf is NULL so as to
+	 * inherit the configuration from user in case of failing to verify
+	 * duplicate rule later.
+	 */
+	if (rss_act->key_len > 0 && rss_act->key != NULL)
+		new_conf->conf.key = new_conf->key;
 
-	ret = hns3_flow_parse_rss(dev, new_conf);
+	ret = hns3_config_rss_filter(hw, new_conf);
 	if (ret != 0) {
 		rte_free(rss_filter_ptr);
+		(void)hns3_flow_rebuild_all_rss_filter(hns);
 		return ret;
 	}
-	rss_filter_ptr->filter_info.valid = true;
-
-	/*
-	 * When create a new RSS rule, the old rule will be overlaid and set
-	 * invalid.
-	 */
-	TAILQ_FOREACH(filter_ptr, &hw->flow_rss_list, entries)
-		filter_ptr->filter_info.valid = false;
 
 	TAILQ_INSERT_TAIL(&hw->flow_rss_list, rss_filter_ptr, entries);
 	flow->rule = rss_filter_ptr;
@@ -1790,31 +2173,24 @@ hns3_flow_create_rss_rule(struct rte_eth_dev *dev,
 
 static int
 hns3_flow_create_fdir_rule(struct rte_eth_dev *dev,
-			   const struct rte_flow_item pattern[],
-			   const struct rte_flow_action actions[],
+			   struct hns3_fdir_rule *fdir_rule,
 			   struct rte_flow_error *error,
 			   struct rte_flow *flow)
 {
 	struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 	struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
 	struct hns3_fdir_rule_ele *fdir_rule_ptr;
-	struct hns3_fdir_rule fdir_rule;
 	bool indir;
 	int ret;
 
-	memset(&fdir_rule, 0, sizeof(struct hns3_fdir_rule));
-	ret = hns3_parse_fdir_filter(dev, pattern, actions, &fdir_rule, error);
-	if (ret != 0)
-		return ret;
-
-	indir = !!(fdir_rule.flags & HNS3_RULE_FLAG_COUNTER_INDIR);
-	if (fdir_rule.flags & HNS3_RULE_FLAG_COUNTER) {
-		ret = hns3_counter_new(dev, indir, fdir_rule.act_cnt.id,
+	indir = !!(fdir_rule->flags & HNS3_RULE_FLAG_COUNTER_INDIR);
+	if (fdir_rule->flags & HNS3_RULE_FLAG_COUNTER) {
+		ret = hns3_counter_new(dev, indir, fdir_rule->act_cnt.id,
 				       error);
 		if (ret != 0)
 			return ret;
 
-		flow->counter_id = fdir_rule.act_cnt.id;
+		flow->counter_id = fdir_rule->act_cnt.id;
 	}
 
 	fdir_rule_ptr = rte_zmalloc("hns3 fdir rule",
@@ -1830,11 +2206,11 @@ hns3_flow_create_fdir_rule(struct rte_eth_dev *dev,
 	 * rules to the hardware to simplify the rollback of rules in the
 	 * hardware.
 	 */
-	ret = hns3_fdir_filter_program(hns, &fdir_rule, false);
+	ret = hns3_fdir_filter_program(hns, fdir_rule, false);
 	if (ret != 0)
 		goto err_fdir_filter;
 
-	memcpy(&fdir_rule_ptr->fdir_conf, &fdir_rule,
+	memcpy(&fdir_rule_ptr->fdir_conf, fdir_rule,
 		sizeof(struct hns3_fdir_rule));
 	TAILQ_INSERT_TAIL(&hw->flow_fdir_list, fdir_rule_ptr, entries);
 	flow->rule = fdir_rule_ptr;
@@ -1845,8 +2221,8 @@ hns3_flow_create_fdir_rule(struct rte_eth_dev *dev,
 err_fdir_filter:
 	rte_free(fdir_rule_ptr);
 err_malloc:
-	if (fdir_rule.flags & HNS3_RULE_FLAG_COUNTER)
-		hns3_counter_release(dev, fdir_rule.act_cnt.id);
+	if (fdir_rule->flags & HNS3_RULE_FLAG_COUNTER)
+		hns3_counter_release(dev, fdir_rule->act_cnt.id);
 
 	return ret;
 }
@@ -1864,13 +2240,15 @@ hns3_flow_create(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
 		 struct rte_flow_error *error)
 {
 	struct hns3_adapter *hns = dev->data->dev_private;
-	struct hns3_hw *hw = &hns->hw;
+	struct hns3_filter_info filter_info = {0};
 	struct hns3_flow_mem *flow_node;
-	const struct rte_flow_action *act;
+	struct hns3_hw *hw = &hns->hw;
+	union hns3_filter_conf *conf;
 	struct rte_flow *flow;
 	int ret;
 
-	ret = hns3_flow_validate(dev, attr, pattern, actions, error);
+	ret = hns3_flow_validate(dev, attr, pattern, actions, error,
+				 &filter_info);
 	if (ret)
 		return NULL;
 
@@ -1890,13 +2268,12 @@ hns3_flow_create(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
 	}
 
 	flow_node->flow = flow;
+	conf = &filter_info.conf;
 	TAILQ_INSERT_TAIL(&hw->flow_list, flow_node, entries);
-
-	act = hns3_find_rss_general_action(pattern, actions);
-	if (act)
-		ret = hns3_flow_create_rss_rule(dev, act, flow);
+	if (filter_info.type == RTE_ETH_FILTER_HASH)
+		ret = hns3_flow_create_rss_rule(dev, &conf->rss_conf, flow);
 	else
-		ret = hns3_flow_create_fdir_rule(dev, pattern, actions,
+		ret = hns3_flow_create_fdir_rule(dev, &conf->fdir_conf,
 						 error, flow);
 	if (ret == 0)
 		return flow;
@@ -1950,15 +2327,10 @@ hns3_flow_destroy(struct rte_eth_dev *dev, struct rte_flow *flow,
 		break;
 	case RTE_ETH_FILTER_HASH:
 		rss_filter_ptr = (struct hns3_rss_conf_ele *)flow->rule;
-		ret = hns3_reset_rss_filter(hw, &rss_filter_ptr->filter_info);
-		if (ret)
-			return rte_flow_error_set(error, EIO,
-						  RTE_FLOW_ERROR_TYPE_HANDLE,
-						  flow,
-						  "Destroy RSS fail.Try again");
 		TAILQ_REMOVE(&hw->flow_rss_list, rss_filter_ptr, entries);
 		rte_free(rss_filter_ptr);
 		rss_filter_ptr = NULL;
+		(void)hns3_flow_rebuild_all_rss_filter(hns);
 		break;
 	default:
 		return rte_flow_error_set(error, EINVAL,
@@ -2064,10 +2436,12 @@ hns3_flow_validate_wrap(struct rte_eth_dev *dev,
 			struct rte_flow_error *error)
 {
 	struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+	struct hns3_filter_info filter_info = {0};
 	int ret;
 
 	pthread_mutex_lock(&hw->flows_lock);
-	ret = hns3_flow_validate(dev, attr, pattern, actions, error);
+	ret = hns3_flow_validate(dev, attr, pattern, actions, error,
+				 &filter_info);
 	pthread_mutex_unlock(&hw->flows_lock);
 
 	return ret;
diff --git a/drivers/net/hns3/hns3_flow.h b/drivers/net/hns3/hns3_flow.h
index 90126f2b6e..1b49673f11 100644
--- a/drivers/net/hns3/hns3_flow.h
+++ b/drivers/net/hns3/hns3_flow.h
@@ -9,6 +9,7 @@
 #include <ethdev_driver.h>
 
 #include "hns3_rss.h"
+#include "hns3_fdir.h"
 
 struct hns3_flow_counter {
 	LIST_ENTRY(hns3_flow_counter) next; /* Pointer to the next counter. */
@@ -26,10 +27,10 @@ struct rte_flow {
 
 struct hns3_flow_rss_conf {
 	struct rte_flow_action_rss conf;
-	uint8_t hash_algo;
 	uint8_t key[HNS3_RSS_KEY_SIZE_MAX];  /* Hash key */
 	uint16_t queue[HNS3_RSS_QUEUES_BUFFER_NUM]; /* Queues indices to use */
-	bool valid; /* check if RSS rule is valid */
+	uint64_t pattern_type;
+	uint64_t hw_pctypes; /* packet types in driver */
 };
 
 /* rss filter list structure */
@@ -53,6 +54,16 @@ struct rte_flow_action_handle {
 	uint32_t counter_id;
 };
 
+union hns3_filter_conf {
+	struct hns3_fdir_rule fdir_conf;
+	struct hns3_flow_rss_conf rss_conf;
+};
+
+struct hns3_filter_info {
+	enum rte_filter_type type;
+	union hns3_filter_conf conf;
+};
+
 TAILQ_HEAD(hns3_rss_filter_list, hns3_rss_conf_ele);
 TAILQ_HEAD(hns3_flow_mem_list, hns3_flow_mem);
 
diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index 2ef267aac6..370565a841 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -18,69 +18,13 @@ const uint8_t hns3_hash_key[HNS3_RSS_KEY_SIZE] = {
 	0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA
 };
 
-enum hns3_tuple_field {
-	/* IPV4_TCP ENABLE FIELD */
-	HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_D = 0,
-	HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_S,
-	HNS3_RSS_FIELD_IPV4_TCP_EN_IP_D,
-	HNS3_RSS_FIELD_IPV4_TCP_EN_IP_S,
-
-	/* IPV4_UDP ENABLE FIELD */
-	HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_D = 8,
-	HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_S,
-	HNS3_RSS_FIELD_IPV4_UDP_EN_IP_D,
-	HNS3_RSS_FIELD_IPV4_UDP_EN_IP_S,
-
-	/* IPV4_SCTP ENABLE FIELD */
-	HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_D = 16,
-	HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_S,
-	HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_D,
-	HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_S,
-	HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_VER,
-
-	/* IPV4 ENABLE FIELD */
-	HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D = 24,
-	HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_S,
-	HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_D,
-	HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_S,
-
-	/* IPV6_TCP ENABLE FIELD */
-	HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_D = 32,
-	HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_S,
-	HNS3_RSS_FIELD_IPV6_TCP_EN_IP_D,
-	HNS3_RSS_FIELD_IPV6_TCP_EN_IP_S,
-
-	/* IPV6_UDP ENABLE FIELD */
-	HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_D = 40,
-	HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_S,
-	HNS3_RSS_FIELD_IPV6_UDP_EN_IP_D,
-	HNS3_RSS_FIELD_IPV6_UDP_EN_IP_S,
-
-	/* IPV6_SCTP ENABLE FIELD */
-	HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_D = 48,
-	HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_S,
-	HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_D,
-	HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_S,
-	HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_VER,
-
-	/* IPV6 ENABLE FIELD */
-	HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D = 56,
-	HNS3_RSS_FIELD_IPV6_NONFRAG_IP_S,
-	HNS3_RSS_FIELD_IPV6_FRAG_IP_D,
-	HNS3_RSS_FIELD_IPV6_FRAG_IP_S
+const uint8_t hns3_hash_func_map[] = {
+	[RTE_ETH_HASH_FUNCTION_DEFAULT] = HNS3_RSS_HASH_ALGO_TOEPLITZ,
+	[RTE_ETH_HASH_FUNCTION_TOEPLITZ] = HNS3_RSS_HASH_ALGO_TOEPLITZ,
+	[RTE_ETH_HASH_FUNCTION_SIMPLE_XOR] = HNS3_RSS_HASH_ALGO_SIMPLE,
+	[RTE_ETH_HASH_FUNCTION_SYMMETRIC_TOEPLITZ] = HNS3_RSS_HASH_ALGO_SYMMETRIC_TOEP,
 };
 
-#define HNS3_RSS_TUPLE_IPV4_TCP_M	GENMASK(3, 0)
-#define HNS3_RSS_TUPLE_IPV4_UDP_M	GENMASK(11, 8)
-#define HNS3_RSS_TUPLE_IPV4_SCTP_M	GENMASK(20, 16)
-#define HNS3_RSS_TUPLE_IPV4_NONF_M	GENMASK(25, 24)
-#define HNS3_RSS_TUPLE_IPV4_FLAG_M	GENMASK(27, 26)
-#define HNS3_RSS_TUPLE_IPV6_TCP_M	GENMASK(35, 32)
-#define HNS3_RSS_TUPLE_IPV6_UDP_M	GENMASK(43, 40)
-#define HNS3_RSS_TUPLE_IPV6_SCTP_M	GENMASK(52, 48)
-#define HNS3_RSS_TUPLE_IPV6_NONF_M	GENMASK(57, 56)
-#define HNS3_RSS_TUPLE_IPV6_FLAG_M	GENMASK(59, 58)
-
 enum hns3_rss_tuple_type {
 	HNS3_RSS_IP_TUPLE,
 	HNS3_RSS_IP_L4_TUPLE,
@@ -573,7 +517,7 @@ hns3_rss_check_l3l4_types(struct hns3_hw *hw, uint64_t rss_hf)
 		hns3_warn(hw, "packet type isn't specified, L4_SRC/DST_ONLY is ignored.");
 }
 
-static uint64_t
+uint64_t
 hns3_rss_calc_tuple_filed(struct hns3_hw *hw, uint64_t rss_hf)
 {
 	uint64_t l3_only_mask = RTE_ETH_RSS_L3_SRC_ONLY |
@@ -609,25 +553,35 @@ hns3_rss_calc_tuple_filed(struct hns3_hw *hw, uint64_t rss_hf)
 }
 
 int
-hns3_set_rss_tuple_by_rss_hf(struct hns3_hw *hw, uint64_t rss_hf)
+hns3_set_rss_tuple_field(struct hns3_hw *hw, uint64_t tuple_fields)
 {
 	struct hns3_rss_input_tuple_cmd *req;
 	struct hns3_cmd_desc desc;
-	uint64_t tuple_field;
 	int ret;
 
 	hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RSS_INPUT_TUPLE, false);
 	req = (struct hns3_rss_input_tuple_cmd *)desc.data;
-
-	tuple_field = hns3_rss_calc_tuple_filed(hw, rss_hf);
-	req->tuple_field = rte_cpu_to_le_64(tuple_field);
+	req->tuple_field = rte_cpu_to_le_64(tuple_fields);
 	ret = hns3_cmd_send(hw, &desc, 1);
-	if (ret) {
-		hns3_err(hw, "Update RSS flow types tuples failed %d", ret);
-		return ret;
-	}
+	if (ret != 0)
+		hns3_err(hw, "set RSS hash tuple fields failed ret = %d", ret);
 
-	return 0;
+	return ret;
+}
+
+int
+hns3_set_rss_tuple_by_rss_hf(struct hns3_hw *hw, uint64_t rss_hf)
+{
+	uint64_t tuple_fields;
+	int ret;
+
+	tuple_fields = hns3_rss_calc_tuple_filed(hw, rss_hf);
+	ret = hns3_set_rss_tuple_field(hw, tuple_fields);
+	if (ret != 0)
+		hns3_err(hw, "Update RSS flow types tuples failed, ret = %d",
+			 ret);
+
+	return ret;
 }
 
 /*
@@ -999,6 +953,52 @@ hns3_set_rss_tc_mode(struct hns3_hw *hw)
 	return ret;
 }
 
+/*
+ * Note: the 'hash_algo' is defined by enum rte_eth_hash_function.
+ */
+int
+hns3_update_rss_algo_key(struct hns3_hw *hw, uint8_t hash_func,
+			 uint8_t *key, uint8_t key_len)
+{
+	uint8_t rss_key[HNS3_RSS_KEY_SIZE_MAX] = {0};
+	bool modify_key, modify_algo;
+	uint8_t hash_algo;
+	int ret;
+
+	modify_key = (key != NULL && key_len > 0);
+	modify_algo = hash_func != RTE_ETH_HASH_FUNCTION_DEFAULT;
+	if (!modify_key && !modify_algo)
+		return 0;
+
+	if (modify_algo && hash_func >= RTE_DIM(hns3_hash_func_map)) {
+		hns3_err(hw, "hash func (%u) is unsupported.", hash_func);
+		return -ENOTSUP;
+	}
+	if (modify_key && key_len != hw->rss_key_size) {
+		hns3_err(hw, "hash key length (%u) is invalid.", key_len);
+		return -EINVAL;
+	}
+
+	ret = hns3_rss_get_algo_key(hw, &hash_algo, rss_key, hw->rss_key_size);
+	if (ret != 0) {
+		hns3_err(hw, "fail to get RSS hash algorithm and key, ret = %d",
+			 ret);
+		return ret;
+	}
+
+	if (modify_algo)
+		hash_algo = hns3_hash_func_map[hash_func];
+	if (modify_key)
+		memcpy(rss_key, key, key_len);
+
+	ret = hns3_rss_set_algo_key(hw, hash_algo, rss_key, hw->rss_key_size);
+	if (ret != 0)
+		hns3_err(hw, "fail to set RSS hash algorithm and key, ret = %d",
+			 ret);
+
+	return ret;
+}
+
 static void
 hns3_rss_tuple_uninit(struct hns3_hw *hw)
 {
diff --git a/drivers/net/hns3/hns3_rss.h b/drivers/net/hns3/hns3_rss.h
index d19730c69c..d672481a14 100644
--- a/drivers/net/hns3/hns3_rss.h
+++ b/drivers/net/hns3/hns3_rss.h
@@ -8,23 +8,102 @@
 #include <rte_ethdev.h>
 #include <rte_flow.h>
 
-#define HNS3_ETH_RSS_SUPPORT ( \
-	RTE_ETH_RSS_IPV4 | \
-	RTE_ETH_RSS_FRAG_IPV4 | \
-	RTE_ETH_RSS_NONFRAG_IPV4_TCP | \
-	RTE_ETH_RSS_NONFRAG_IPV4_UDP | \
-	RTE_ETH_RSS_NONFRAG_IPV4_SCTP | \
-	RTE_ETH_RSS_NONFRAG_IPV4_OTHER | \
-	RTE_ETH_RSS_IPV6 | \
-	RTE_ETH_RSS_FRAG_IPV6 | \
-	RTE_ETH_RSS_NONFRAG_IPV6_TCP | \
-	RTE_ETH_RSS_NONFRAG_IPV6_UDP | \
-	RTE_ETH_RSS_NONFRAG_IPV6_SCTP | \
-	RTE_ETH_RSS_NONFRAG_IPV6_OTHER | \
-	RTE_ETH_RSS_L3_SRC_ONLY | \
-	RTE_ETH_RSS_L3_DST_ONLY | \
-	RTE_ETH_RSS_L4_SRC_ONLY | \
-	RTE_ETH_RSS_L4_DST_ONLY)
+#define HNS3_RSS_SUPPORT_L3_SRC_DST	(RTE_ETH_RSS_L3_SRC_ONLY | \
+					 RTE_ETH_RSS_L3_DST_ONLY)
+#define HNS3_RSS_SUPPORT_L4_SRC_DST	(RTE_ETH_RSS_L4_SRC_ONLY | \
+					 RTE_ETH_RSS_L4_DST_ONLY)
+#define HNS3_RSS_SUPPORT_L3L4		(HNS3_RSS_SUPPORT_L3_SRC_DST | \
+					 HNS3_RSS_SUPPORT_L4_SRC_DST)
+
+#define HNS3_RSS_SUPPORT_FLOW_TYPE	(RTE_ETH_RSS_IPV4 | \
+					 RTE_ETH_RSS_FRAG_IPV4 | \
+					 RTE_ETH_RSS_NONFRAG_IPV4_TCP | \
+					 RTE_ETH_RSS_NONFRAG_IPV4_UDP | \
+					 RTE_ETH_RSS_NONFRAG_IPV4_SCTP | \
+					 RTE_ETH_RSS_NONFRAG_IPV4_OTHER | \
+					 RTE_ETH_RSS_IPV6 | \
+					 RTE_ETH_RSS_FRAG_IPV6 | \
+					 RTE_ETH_RSS_NONFRAG_IPV6_TCP | \
+					 RTE_ETH_RSS_NONFRAG_IPV6_UDP | \
+					 RTE_ETH_RSS_NONFRAG_IPV6_SCTP | \
+					 RTE_ETH_RSS_NONFRAG_IPV6_OTHER)
+
+#define HNS3_ETH_RSS_SUPPORT		(HNS3_RSS_SUPPORT_FLOW_TYPE | \
+					 HNS3_RSS_SUPPORT_L3L4)
+
+enum hns3_tuple_field {
+	/* IPV4_TCP ENABLE FIELD */
+	HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_D = 0,
+	HNS3_RSS_FIELD_IPV4_TCP_EN_TCP_S,
+	HNS3_RSS_FIELD_IPV4_TCP_EN_IP_D,
+	HNS3_RSS_FIELD_IPV4_TCP_EN_IP_S,
+
+	/* IPV4_UDP ENABLE FIELD */
+	HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_D = 8,
+	HNS3_RSS_FIELD_IPV4_UDP_EN_UDP_S,
+	HNS3_RSS_FIELD_IPV4_UDP_EN_IP_D,
+	HNS3_RSS_FIELD_IPV4_UDP_EN_IP_S,
+
+	/* IPV4_SCTP ENABLE FIELD */
+	HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_D = 16,
+	HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_S,
+	HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_D,
+	HNS3_RSS_FIELD_IPV4_SCTP_EN_IP_S,
+	HNS3_RSS_FIELD_IPV4_SCTP_EN_SCTP_VER,
+
+	/* IPV4 ENABLE FIELD */
+	HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_D = 24,
+	HNS3_RSS_FIELD_IPV4_EN_NONFRAG_IP_S,
+	HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_D,
+	HNS3_RSS_FIELD_IPV4_EN_FRAG_IP_S,
+
+	/* IPV6_TCP ENABLE FIELD */
+	HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_D = 32,
+	HNS3_RSS_FIELD_IPV6_TCP_EN_TCP_S,
+	HNS3_RSS_FIELD_IPV6_TCP_EN_IP_D,
+	HNS3_RSS_FIELD_IPV6_TCP_EN_IP_S,
+
+	/* IPV6_UDP ENABLE FIELD */
+	HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_D = 40,
+	HNS3_RSS_FIELD_IPV6_UDP_EN_UDP_S,
+	HNS3_RSS_FIELD_IPV6_UDP_EN_IP_D,
+	HNS3_RSS_FIELD_IPV6_UDP_EN_IP_S,
+
+	/* IPV6_SCTP ENABLE FIELD */
+	HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_D = 48,
+	HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_S,
+	HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_D,
+	HNS3_RSS_FIELD_IPV6_SCTP_EN_IP_S,
+	HNS3_RSS_FIELD_IPV6_SCTP_EN_SCTP_VER,
+
+	/* IPV6 ENABLE FIELD */
+	HNS3_RSS_FIELD_IPV6_NONFRAG_IP_D = 56,
+	HNS3_RSS_FIELD_IPV6_NONFRAG_IP_S,
+	HNS3_RSS_FIELD_IPV6_FRAG_IP_D,
+	HNS3_RSS_FIELD_IPV6_FRAG_IP_S
+};
+
+#define HNS3_RSS_PCTYPE_IPV4_TCP	BIT_ULL(0)
+#define HNS3_RSS_PCTYPE_IPV4_UDP	BIT_ULL(8)
+#define HNS3_RSS_PCTYPE_IPV4_SCTP	BIT_ULL(16)
+#define HNS3_RSS_PCTYPE_IPV4_NONF	BIT_ULL(24)
+#define HNS3_RSS_PCTYPE_IPV4_FLAG	BIT_ULL(26)
+#define HNS3_RSS_PCTYPE_IPV6_TCP	BIT_ULL(32)
+#define HNS3_RSS_PCTYPE_IPV6_UDP	BIT_ULL(40)
+#define HNS3_RSS_PCTYPE_IPV6_SCTP	BIT_ULL(48)
+#define HNS3_RSS_PCTYPE_IPV6_NONF	BIT_ULL(56)
+#define HNS3_RSS_PCTYPE_IPV6_FLAG	BIT_ULL(58)
+
+#define HNS3_RSS_TUPLE_IPV4_TCP_M	GENMASK(3, 0)
+#define HNS3_RSS_TUPLE_IPV4_UDP_M	GENMASK(11, 8)
+#define HNS3_RSS_TUPLE_IPV4_SCTP_M	GENMASK(20, 16)
+#define HNS3_RSS_TUPLE_IPV4_NONF_M	GENMASK(25, 24)
+#define HNS3_RSS_TUPLE_IPV4_FLAG_M	GENMASK(27, 26)
+#define HNS3_RSS_TUPLE_IPV6_TCP_M	GENMASK(35, 32)
+#define HNS3_RSS_TUPLE_IPV6_UDP_M	GENMASK(43, 40)
+#define HNS3_RSS_TUPLE_IPV6_SCTP_M	GENMASK(52, 48)
+#define HNS3_RSS_TUPLE_IPV6_NONF_M	GENMASK(57, 56)
+#define HNS3_RSS_TUPLE_IPV6_FLAG_M	GENMASK(59, 58)
 
 #define HNS3_RSS_IND_TBL_SIZE	512 /* The size of hash lookup table */
 #define HNS3_RSS_IND_TBL_SIZE_MAX 2048
@@ -108,10 +187,14 @@ int hns3_rss_reset_indir_table(struct hns3_hw *hw);
 int hns3_config_rss(struct hns3_adapter *hns);
 void hns3_rss_uninit(struct hns3_adapter *hns);
 int hns3_set_rss_tuple_by_rss_hf(struct hns3_hw *hw, uint64_t rss_hf);
+int hns3_set_rss_tuple_field(struct hns3_hw *hw, uint64_t tuple_fields);
 int hns3_get_rss_tuple_field(struct hns3_hw *hw, uint64_t *tuple_fields);
 int hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
 			  const uint8_t *key, uint8_t key_len);
 int hns3_rss_get_algo_key(struct hns3_hw *hw,  uint8_t *hash_algo,
 			  uint8_t *key, uint8_t key_len);
+uint64_t hns3_rss_calc_tuple_filed(struct hns3_hw *hw, uint64_t rss_hf);
+int hns3_update_rss_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
+			     uint8_t *key, uint8_t key_len);
 
 #endif /* HNS3_RSS_H */
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* [PATCH 16/16] net/hns3: add the verification of RSS types
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (14 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 15/16] net/hns3: reimplement hash flow function Dongdong Liu
@ 2023-03-10  9:35 ` Dongdong Liu
  2023-03-10 20:58 ` [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Ferruh Yigit
  16 siblings, 0 replies; 24+ messages in thread
From: Dongdong Liu @ 2023-03-10  9:35 UTC (permalink / raw)
  To: dev, ferruh.yigit, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, liudongdong3, lihuisong

From: Huisong Li <lihuisong@huawei.com>

Add the verification of RSS types from ethdev ops and rte flow without
pattern attribute. The following cases are invalid:
1. types contains unsupported RSS type but hasn't type driver support.
2. types has L3 src/dst but hasn't supported packet type.
3. types has L4 src/dst but hasn't supported packet type and hasn't IP
   packet type.

Fixes: 13c3993240c8 ("net/hns3: add L3 and L4 RSS types")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
 drivers/net/hns3/hns3_flow.c | 12 +++---
 drivers/net/hns3/hns3_rss.c  | 74 +++++++++++++++++++++++++-----------
 drivers/net/hns3/hns3_rss.h  |  3 +-
 3 files changed, 60 insertions(+), 29 deletions(-)

diff --git a/drivers/net/hns3/hns3_flow.c b/drivers/net/hns3/hns3_flow.c
index 6ac623bea9..e132d88fa1 100644
--- a/drivers/net/hns3/hns3_flow.c
+++ b/drivers/net/hns3/hns3_flow.c
@@ -1652,9 +1652,10 @@ hns3_flow_parse_rss_types(struct hns3_hw *hw,
 
 	/* no pattern specified to set global RSS types. */
 	if (pattern_type == 0) {
-		if (rss_act->types & ~HNS3_ETH_RSS_SUPPORT)
-			hns3_warn(hw, "some types in the requested RSS types (0x%" PRIx64 ") aren't supported, they are ignored.",
-				  rss_act->types);
+		if (!hns3_check_rss_types_valid(hw, rss_act->types))
+			return rte_flow_error_set(error, EINVAL,
+					RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					NULL, "RSS types is invalid.");
 		rss_conf->hw_pctypes =
 				hns3_flow_get_all_hw_pctypes(rss_act->types);
 		return 0;
@@ -1929,15 +1930,14 @@ hns3_flow_set_rss_ptype_tuple(struct hns3_hw *hw,
 			uint64_t pctype = BIT_ULL(idx);
 
 			tuple_mask = hns3_flow_get_pctype_tuple_mask(pctype);
-			tuples = hns3_rss_calc_tuple_filed(hw,
-							rss_conf->conf.types);
+			tuples = hns3_rss_calc_tuple_filed(rss_conf->conf.types);
 			new_tuple_fields &= ~tuple_mask;
 			new_tuple_fields |= tuples;
 			hw_pctypes &= ~pctype;
 		}
 	} else {
 		new_tuple_fields =
-			hns3_rss_calc_tuple_filed(hw, rss_conf->conf.types);
+			hns3_rss_calc_tuple_filed(rss_conf->conf.types);
 	}
 
 	ret = hns3_set_rss_tuple_field(hw, new_tuple_fields);
diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
index 370565a841..036d4a8650 100644
--- a/drivers/net/hns3/hns3_rss.c
+++ b/drivers/net/hns3/hns3_rss.c
@@ -491,34 +491,62 @@ hns3_rss_reset_indir_table(struct hns3_hw *hw)
 	return ret;
 }
 
-static void
-hns3_rss_check_l3l4_types(struct hns3_hw *hw, uint64_t rss_hf)
+bool
+hns3_check_rss_types_valid(struct hns3_hw *hw, uint64_t types)
 {
 	uint64_t ip_mask = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
 			   RTE_ETH_RSS_NONFRAG_IPV4_OTHER |
 			   RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
 			   RTE_ETH_RSS_NONFRAG_IPV6_OTHER;
-	uint64_t l4_mask = RTE_ETH_RSS_NONFRAG_IPV4_TCP |
-			   RTE_ETH_RSS_NONFRAG_IPV4_UDP |
-			   RTE_ETH_RSS_NONFRAG_IPV4_SCTP |
-			   RTE_ETH_RSS_NONFRAG_IPV6_TCP |
-			   RTE_ETH_RSS_NONFRAG_IPV6_UDP |
-			   RTE_ETH_RSS_NONFRAG_IPV6_SCTP;
-	uint64_t l3_src_dst_mask = RTE_ETH_RSS_L3_SRC_ONLY |
-				   RTE_ETH_RSS_L3_DST_ONLY;
-	uint64_t l4_src_dst_mask = RTE_ETH_RSS_L4_SRC_ONLY |
-				   RTE_ETH_RSS_L4_DST_ONLY;
-
-	if (rss_hf & l3_src_dst_mask &&
-	    !(rss_hf & ip_mask || rss_hf & l4_mask))
-		hns3_warn(hw, "packet type isn't specified, L3_SRC/DST_ONLY is ignored.");
-
-	if (rss_hf & l4_src_dst_mask && !(rss_hf & l4_mask))
-		hns3_warn(hw, "packet type isn't specified, L4_SRC/DST_ONLY is ignored.");
+	uint64_t ip_l4_mask = RTE_ETH_RSS_NONFRAG_IPV4_TCP |
+			      RTE_ETH_RSS_NONFRAG_IPV4_UDP |
+			      RTE_ETH_RSS_NONFRAG_IPV4_SCTP |
+			      RTE_ETH_RSS_NONFRAG_IPV6_TCP |
+			      RTE_ETH_RSS_NONFRAG_IPV6_UDP |
+			      RTE_ETH_RSS_NONFRAG_IPV6_SCTP;
+	bool has_l4_src_dst = !!(types & HNS3_RSS_SUPPORT_L4_SRC_DST);
+	bool has_ip_pkt = !!(types & ip_mask);
+	uint64_t final_types;
+
+	if (types == 0)
+		return true;
+
+	if ((types & HNS3_ETH_RSS_SUPPORT) == 0) {
+		hns3_err(hw, "specified types(0x%" PRIx64 ") are unsupported.",
+			 types);
+		return false;
+	}
+
+	if ((types & HNS3_RSS_SUPPORT_L3_SRC_DST) != 0 &&
+	    (types & HNS3_RSS_SUPPORT_FLOW_TYPE) == 0) {
+		hns3_err(hw, "IP or IP-TCP/UDP/SCTP packet type isn't specified, L3_SRC/DST_ONLY cannot be set.");
+		return false;
+	}
+
+	if (has_l4_src_dst && (types & ip_l4_mask) == 0) {
+		if (!has_ip_pkt) {
+			hns3_err(hw, "IP-TCP/UDP/SCTP packet type isn't specified, L4_SRC/DST_ONLY cannot be set.");
+			return false;
+		}
+		/*
+		 * For the case that the types has L4_SRC/DST_ONLY but hasn't
+		 * IP-TCP/UDP/SCTP packet type, this types is considered valid
+		 * if it also has IP packet type.
+		 */
+		hns3_warn(hw, "L4_SRC/DST_ONLY is ignored because of no including L4 packet.");
+	}
+
+	if ((types & ~HNS3_ETH_RSS_SUPPORT) != 0) {
+		final_types = types & HNS3_ETH_RSS_SUPPORT;
+		hns3_warn(hw, "set RSS types based on hardware support, requested:0x%" PRIx64 " configured:0x%" PRIx64 "",
+			  types, final_types);
+	}
+
+	return true;
 }
 
 uint64_t
-hns3_rss_calc_tuple_filed(struct hns3_hw *hw, uint64_t rss_hf)
+hns3_rss_calc_tuple_filed(uint64_t rss_hf)
 {
 	uint64_t l3_only_mask = RTE_ETH_RSS_L3_SRC_ONLY |
 				RTE_ETH_RSS_L3_DST_ONLY;
@@ -547,7 +575,6 @@ hns3_rss_calc_tuple_filed(struct hns3_hw *hw, uint64_t rss_hf)
 		    !has_l3_l4_only)
 			tuple |= hns3_set_tuple_table[i].rss_field;
 	}
-	hns3_rss_check_l3l4_types(hw, rss_hf);
 
 	return tuple;
 }
@@ -575,7 +602,7 @@ hns3_set_rss_tuple_by_rss_hf(struct hns3_hw *hw, uint64_t rss_hf)
 	uint64_t tuple_fields;
 	int ret;
 
-	tuple_fields = hns3_rss_calc_tuple_filed(hw, rss_hf);
+	tuple_fields = hns3_rss_calc_tuple_filed(rss_hf);
 	ret = hns3_set_rss_tuple_field(hw, tuple_fields);
 	if (ret != 0)
 		hns3_err(hw, "Update RSS flow types tuples failed, ret = %d",
@@ -610,6 +637,9 @@ hns3_dev_rss_hash_update(struct rte_eth_dev *dev,
 		return -EINVAL;
 	}
 
+	if (!hns3_check_rss_types_valid(hw, rss_hf))
+		return -EINVAL;
+
 	rte_spinlock_lock(&hw->lock);
 	ret = hns3_set_rss_tuple_by_rss_hf(hw, rss_hf);
 	if (ret)
diff --git a/drivers/net/hns3/hns3_rss.h b/drivers/net/hns3/hns3_rss.h
index d672481a14..415430a399 100644
--- a/drivers/net/hns3/hns3_rss.h
+++ b/drivers/net/hns3/hns3_rss.h
@@ -186,6 +186,7 @@ int hns3_set_rss_indir_table(struct hns3_hw *hw, uint16_t *indir,
 int hns3_rss_reset_indir_table(struct hns3_hw *hw);
 int hns3_config_rss(struct hns3_adapter *hns);
 void hns3_rss_uninit(struct hns3_adapter *hns);
+bool hns3_check_rss_types_valid(struct hns3_hw *hw, uint64_t types);
 int hns3_set_rss_tuple_by_rss_hf(struct hns3_hw *hw, uint64_t rss_hf);
 int hns3_set_rss_tuple_field(struct hns3_hw *hw, uint64_t tuple_fields);
 int hns3_get_rss_tuple_field(struct hns3_hw *hw, uint64_t *tuple_fields);
@@ -193,7 +194,7 @@ int hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
 			  const uint8_t *key, uint8_t key_len);
 int hns3_rss_get_algo_key(struct hns3_hw *hw,  uint8_t *hash_algo,
 			  uint8_t *key, uint8_t key_len);
-uint64_t hns3_rss_calc_tuple_filed(struct hns3_hw *hw, uint64_t rss_hf);
+uint64_t hns3_rss_calc_tuple_filed(uint64_t rss_hf);
 int hns3_update_rss_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
 			     uint8_t *key, uint8_t key_len);
 
-- 
2.22.0


^ permalink raw reply	[flat|nested] 24+ messages in thread

* Re: [PATCH 01/16] net/hns3: fix possible truncation of hash key when config
  2023-03-10  9:35 ` [PATCH 01/16] net/hns3: fix possible truncation of hash key when config Dongdong Liu
@ 2023-03-10 19:36   ` Ferruh Yigit
  2023-03-29  1:56     ` lihuisong (C)
  0 siblings, 1 reply; 24+ messages in thread
From: Ferruh Yigit @ 2023-03-10 19:36 UTC (permalink / raw)
  To: Dongdong Liu, dev, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, lihuisong

On 3/10/2023 9:35 AM, Dongdong Liu wrote:
> From: Huisong Li <lihuisong@huawei.com>
> 
> The hash key length of hns3 driver is obtained from firmware. If the
> length isn't a multiple of HNS3_RSS_HASH_KEY_NUM (16), the last part
> of hash key will be truncated.
> 

I am not sure if the explanation is correct, according below code last
part of the key is truncated if key_len *is* multiple of
HNS3_RSS_HASH_KEY_NUM.

Because code assumes "key_len % HNS3_RSS_HASH_KEY_NUM" will give the
remaining part of the key, but when key_len is multiple of
HNS3_RSS_HASH_KEY_NUM it gives 0, causing last HNS3_RSS_HASH_KEY_NUM
chunk truncated.

If above understanding correct, I can fix commit log while merging.

> Fixes: 4a7384e3c34d ("net/hns3: refactor set RSS hash algorithm and key interface")
> Fixes: c37ca66f2b27 ("net/hns3: support RSS")
> Cc: stable@dpdk.org
> 

I am not sure if `c37ca66f2b27 ("net/hns3: support RSS")` is needed
here, issue seems because of commit 4a7384e3c34d, so this should be:

Fixes: 4a7384e3c34d ("net/hns3: refactor set RSS hash algorithm and key
interface")
Cc: stable@dpdk.org


> Signed-off-by: Huisong Li <lihuisong@huawei.com>
> Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
> ---
>  drivers/net/hns3/hns3_rss.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
> index d6e0754273..2011c18b9b 100644
> --- a/drivers/net/hns3/hns3_rss.c
> +++ b/drivers/net/hns3/hns3_rss.c
> @@ -301,7 +301,8 @@ hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
>  		req->hash_config |= (hash_algo & HNS3_RSS_HASH_ALGO_MASK);
>  		req->hash_config |= (idx << HNS3_RSS_HASH_KEY_OFFSET_B);
>  
> -		if (idx == max_bd_num - 1)
> +		if (idx == max_bd_num - 1 &&
> +		    (key_len % HNS3_RSS_HASH_KEY_NUM) != 0)
>  			cur_key_size = key_len % HNS3_RSS_HASH_KEY_NUM;
>  		else
>  			cur_key_size = HNS3_RSS_HASH_KEY_NUM;


^ permalink raw reply	[flat|nested] 24+ messages in thread

* Re: [PATCH 02/16] net/hns3: fix possible truncation of redirection table
  2023-03-10  9:35 ` [PATCH 02/16] net/hns3: fix possible truncation of redirection table Dongdong Liu
@ 2023-03-10 19:36   ` Ferruh Yigit
  0 siblings, 0 replies; 24+ messages in thread
From: Ferruh Yigit @ 2023-03-10 19:36 UTC (permalink / raw)
  To: Dongdong Liu, dev, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, lihuisong

On 3/10/2023 9:35 AM, Dongdong Liu wrote:
> From: Huisong Li <lihuisong@huawei.com>
> 
> The size of the redirection table is obtained from firmware. If the size
> isn't a multiple of HNS3_RSS_CFG_TBL_SIZE, the redirection table from
> user will be truncated.
> 
> Fixes: c37ca66f2b27 ("net/hns3: support RSS")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Huisong Li <lihuisong@huawei.com>
> Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
> ---
>  drivers/net/hns3/hns3_rss.c | 16 +++++++++++-----
>  1 file changed, 11 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
> index 2011c18b9b..f6d677ade8 100644
> --- a/drivers/net/hns3/hns3_rss.c
> +++ b/drivers/net/hns3/hns3_rss.c
> @@ -329,22 +329,28 @@ int
>  hns3_set_rss_indir_table(struct hns3_hw *hw, uint16_t *indir, uint16_t size)
>  {
>  	struct hns3_rss_indirection_table_cmd *req;
> +	uint16_t max_bd_num, cfg_tbl_size;
> +	uint8_t qid_msb_off, qid_msb_val;
>  	struct hns3_cmd_desc desc;
> -	uint8_t qid_msb_off;
> -	uint8_t qid_msb_val;

This change seems unrelated, I can drop while merging if there is no
other change request.

>  	uint16_t q_id;
>  	uint16_t i, j;
>  	int ret;
>  
>  	req = (struct hns3_rss_indirection_table_cmd *)desc.data;
> -
> -	for (i = 0; i < size / HNS3_RSS_CFG_TBL_SIZE; i++) {
> +	max_bd_num = DIV_ROUND_UP(size, HNS3_RSS_CFG_TBL_SIZE);
> +	for (i = 0; i < max_bd_num; i++) {
>  		hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RSS_INDIR_TABLE,
>  					  false);
>  		req->start_table_index =
>  				rte_cpu_to_le_16(i * HNS3_RSS_CFG_TBL_SIZE);
>  		req->rss_set_bitmap = rte_cpu_to_le_16(HNS3_RSS_SET_BITMAP_MSK);
> -		for (j = 0; j < HNS3_RSS_CFG_TBL_SIZE; j++) {
> +
> +		if (i == max_bd_num - 1 && (size % HNS3_RSS_CFG_TBL_SIZE) != 0)
> +			cfg_tbl_size = size % HNS3_RSS_CFG_TBL_SIZE;
> +		else
> +			cfg_tbl_size = HNS3_RSS_CFG_TBL_SIZE;
> +
> +		for (j = 0; j < cfg_tbl_size; j++) {
>  			q_id = indir[i * HNS3_RSS_CFG_TBL_SIZE + j];
>  			req->rss_result_l[j] = q_id & 0xff;
>  


^ permalink raw reply	[flat|nested] 24+ messages in thread

* Re: [PATCH 06/16] net/hns3: separate the setting of hash algorithm
  2023-03-10  9:35 ` [PATCH 06/16] net/hns3: separate the setting of hash algorithm Dongdong Liu
@ 2023-03-10 19:36   ` Ferruh Yigit
  2023-03-29  1:58     ` lihuisong (C)
  0 siblings, 1 reply; 24+ messages in thread
From: Ferruh Yigit @ 2023-03-10 19:36 UTC (permalink / raw)
  To: Dongdong Liu, dev, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, lihuisong, Ori Kam

On 3/10/2023 9:35 AM, Dongdong Liu wrote:
> From: Huisong Li <lihuisong@huawei.com>
> 
> Currently, the setting of hash algorithm comes from the
> default configuration in driver and the rte_flow interface.
> The hash algorithm that is set to hardware in both ways is
> saved in hw->rss_info.conf.func.
> 
> But the 'func' in struct rte_flow_action_rss is usually used
> in rte flow interface. And the ethdev ops interface may also
> set hash algorithm in the future. It is not appropriate and
> is a little messy for ethdev ops interface and driver default
> configuration to use struct rte_flow_action_rss. So we have
> to separate the RSS configuration from ethdev ops and rte
> flow interface to make codes more easier to maintain.
> 

Agree that it is not ideal to have two different ways for same/similar
control path functionality in ethdev.

+Ori to discuss if this is a common problem and can be resolved in
ethdev layer.


I can see some of remaining patches are related to this ethdev / flow
API RSS separation. I will continue with this set, but I believe it is
better if this issue addressed in higher level.

> This patch separates hash algorithm by following ways:
> 1) 'hash_algo' in struct hns3_rss_conf is used for ethdev ops
>    interface or default configuration in driver.
> 2) Add a 'rte_flow_hash_algo' field in struct hns3_rss_conf
>    to save algorithm from rte flow interface. The main reasons
>    are as follows:
>    Currently, only the last rule is used to restore the rte
>    flow rule. If 'func' in RSS action is 'DEFAULT', it means
>    that this rule doesn't modify algorithm and driver need to
>    save current algorithm for restoring algorithm during reset
>    phase.
> 
> Fixes: c37ca66f2b27 ("net/hns3: support RSS")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Huisong Li <lihuisong@huawei.com>
> Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>

<...>


^ permalink raw reply	[flat|nested] 24+ messages in thread

* Re: [PATCH 00/16] net/hns3: some code refactor for hns3 RSS
  2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
                   ` (15 preceding siblings ...)
  2023-03-10  9:35 ` [PATCH 16/16] net/hns3: add the verification of RSS types Dongdong Liu
@ 2023-03-10 20:58 ` Ferruh Yigit
  16 siblings, 0 replies; 24+ messages in thread
From: Ferruh Yigit @ 2023-03-10 20:58 UTC (permalink / raw)
  To: Dongdong Liu, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, lihuisong, dev

On 3/10/2023 9:35 AM, Dongdong Liu wrote:
> This patchset is to do some code refactor for hns3 RSS.
> 1. [PATCH 01/16]-[PATCH 02/16] fix some hns3 RSS bug.
> 2. [PATCH 03/16]-[PATCH 14/16] refactoring hns3 RSS code.
> 3. [PATCH 15/16]-[PATCH 16/16] reimplement hash flow function.
> 
> 
> Huisong Li (16):
>   net/hns3: fix possible truncation of hash key when config
>   net/hns3: fix possible truncation of redirection table
>   net/hns3: use hardware config to report hash key
>   net/hns3: use hardware config to report hash types
>   net/hns3: use hardware config to report redirection table
>   net/hns3: separate the setting of hash algorithm
>   net/hns3: separate the setting of hash key
>   net/hns3: separate the setting of redirection table
>   net/hns3: separate the setting of RSS types
>   net/hns3: separate the setting and clearing of RSS rule
>   net/hns3: use new RSS rule to configure hardware
>   net/hns3: save hash algo to RSS filter list node
>   net/hns3: adding queue buffer size hash rule allowed
>   net/hns3: separate rte flow RSS config from hns3 rss conf
>   net/hns3: reimplement hash flow function
>   net/hns3: add the verification of RSS types


with mentioned minor modifications in 1/16 & 2/16

Series applied to dpdk-next-net/main, thanks.

^ permalink raw reply	[flat|nested] 24+ messages in thread

* Re: [PATCH 01/16] net/hns3: fix possible truncation of hash key when config
  2023-03-10 19:36   ` Ferruh Yigit
@ 2023-03-29  1:56     ` lihuisong (C)
  0 siblings, 0 replies; 24+ messages in thread
From: lihuisong (C) @ 2023-03-29  1:56 UTC (permalink / raw)
  To: Ferruh Yigit, dev
  Cc: stable, yisen.zhuang, Dongdong Liu, andrew.rybchenko, thomas


在 2023/3/11 3:36, Ferruh Yigit 写道:
> On 3/10/2023 9:35 AM, Dongdong Liu wrote:
>> From: Huisong Li <lihuisong@huawei.com>
>>
>> The hash key length of hns3 driver is obtained from firmware. If the
>> length isn't a multiple of HNS3_RSS_HASH_KEY_NUM (16), the last part
>> of hash key will be truncated.
>>
> I am not sure if the explanation is correct, according below code last
> part of the key is truncated if key_len *is* multiple of
> HNS3_RSS_HASH_KEY_NUM.
>
> Because code assumes "key_len % HNS3_RSS_HASH_KEY_NUM" will give the
> remaining part of the key, but when key_len is multiple of
> HNS3_RSS_HASH_KEY_NUM it gives 0, causing last HNS3_RSS_HASH_KEY_NUM
> chunk truncated.
>
> If above understanding correct, I can fix commit log while merging.
Sorry for my late reply. Your understanding correct. Thanks.
>
>> Fixes: 4a7384e3c34d ("net/hns3: refactor set RSS hash algorithm and key interface")
>> Fixes: c37ca66f2b27 ("net/hns3: support RSS")
>> Cc: stable@dpdk.org
>>
> I am not sure if `c37ca66f2b27 ("net/hns3: support RSS")` is needed
> here, issue seems because of commit 4a7384e3c34d, so this should be:
>
> Fixes: 4a7384e3c34d ("net/hns3: refactor set RSS hash algorithm and key
> interface")
> Cc: stable@dpdk.org
>
>
>> Signed-off-by: Huisong Li <lihuisong@huawei.com>
>> Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
>> ---
>>   drivers/net/hns3/hns3_rss.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/hns3/hns3_rss.c b/drivers/net/hns3/hns3_rss.c
>> index d6e0754273..2011c18b9b 100644
>> --- a/drivers/net/hns3/hns3_rss.c
>> +++ b/drivers/net/hns3/hns3_rss.c
>> @@ -301,7 +301,8 @@ hns3_rss_set_algo_key(struct hns3_hw *hw, uint8_t hash_algo,
>>   		req->hash_config |= (hash_algo & HNS3_RSS_HASH_ALGO_MASK);
>>   		req->hash_config |= (idx << HNS3_RSS_HASH_KEY_OFFSET_B);
>>   
>> -		if (idx == max_bd_num - 1)
>> +		if (idx == max_bd_num - 1 &&
>> +		    (key_len % HNS3_RSS_HASH_KEY_NUM) != 0)
>>   			cur_key_size = key_len % HNS3_RSS_HASH_KEY_NUM;
>>   		else
>>   			cur_key_size = HNS3_RSS_HASH_KEY_NUM;
> .

^ permalink raw reply	[flat|nested] 24+ messages in thread

* Re: [PATCH 06/16] net/hns3: separate the setting of hash algorithm
  2023-03-10 19:36   ` Ferruh Yigit
@ 2023-03-29  1:58     ` lihuisong (C)
  2023-03-29  8:13       ` Ferruh Yigit
  0 siblings, 1 reply; 24+ messages in thread
From: lihuisong (C) @ 2023-03-29  1:58 UTC (permalink / raw)
  To: Ferruh Yigit, Dongdong Liu, dev, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, Ori Kam


在 2023/3/11 3:36, Ferruh Yigit 写道:
> On 3/10/2023 9:35 AM, Dongdong Liu wrote:
>> From: Huisong Li <lihuisong@huawei.com>
>>
>> Currently, the setting of hash algorithm comes from the
>> default configuration in driver and the rte_flow interface.
>> The hash algorithm that is set to hardware in both ways is
>> saved in hw->rss_info.conf.func.
>>
>> But the 'func' in struct rte_flow_action_rss is usually used
>> in rte flow interface. And the ethdev ops interface may also
>> set hash algorithm in the future. It is not appropriate and
>> is a little messy for ethdev ops interface and driver default
>> configuration to use struct rte_flow_action_rss. So we have
>> to separate the RSS configuration from ethdev ops and rte
>> flow interface to make codes more easier to maintain.
>>
> Agree that it is not ideal to have two different ways for same/similar
> control path functionality in ethdev.
>
> +Ori to discuss if this is a common problem and can be resolved in
> ethdev layer.
This patchset is aimed to decouple the configuration API and the 
structure used by
ethdev ops and rte flow API in driver. I think this can be ignored.
>
> I can see some of remaining patches are related to this ethdev / flow
> API RSS separation. I will continue with this set, but I believe it is
> better if this issue addressed in higher level.
As far as I know, the priority of rte flow hash rule is higher than 
ethdev ops by default,
and there are many other rules.
But the implementation guide documentation about it is not clear.
It may be better if we can clarify the documentation to guide driver coding.
>> This patch separates hash algorithm by following ways:
>> 1) 'hash_algo' in struct hns3_rss_conf is used for ethdev ops
>>     interface or default configuration in driver.
>> 2) Add a 'rte_flow_hash_algo' field in struct hns3_rss_conf
>>     to save algorithm from rte flow interface. The main reasons
>>     are as follows:
>>     Currently, only the last rule is used to restore the rte
>>     flow rule. If 'func' in RSS action is 'DEFAULT', it means
>>     that this rule doesn't modify algorithm and driver need to
>>     save current algorithm for restoring algorithm during reset
>>     phase.
>>
>> Fixes: c37ca66f2b27 ("net/hns3: support RSS")
>> Cc: stable@dpdk.org
>>
>> Signed-off-by: Huisong Li <lihuisong@huawei.com>
>> Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
> <...>
>
> .

^ permalink raw reply	[flat|nested] 24+ messages in thread

* Re: [PATCH 06/16] net/hns3: separate the setting of hash algorithm
  2023-03-29  1:58     ` lihuisong (C)
@ 2023-03-29  8:13       ` Ferruh Yigit
  0 siblings, 0 replies; 24+ messages in thread
From: Ferruh Yigit @ 2023-03-29  8:13 UTC (permalink / raw)
  To: lihuisong (C), Dongdong Liu, dev, thomas, andrew.rybchenko
  Cc: stable, yisen.zhuang, Ori Kam

On 3/29/2023 2:58 AM, lihuisong (C) wrote:
> 
> 在 2023/3/11 3:36, Ferruh Yigit 写道:
>> On 3/10/2023 9:35 AM, Dongdong Liu wrote:
>>> From: Huisong Li <lihuisong@huawei.com>
>>>
>>> Currently, the setting of hash algorithm comes from the
>>> default configuration in driver and the rte_flow interface.
>>> The hash algorithm that is set to hardware in both ways is
>>> saved in hw->rss_info.conf.func.
>>>
>>> But the 'func' in struct rte_flow_action_rss is usually used
>>> in rte flow interface. And the ethdev ops interface may also
>>> set hash algorithm in the future. It is not appropriate and
>>> is a little messy for ethdev ops interface and driver default
>>> configuration to use struct rte_flow_action_rss. So we have
>>> to separate the RSS configuration from ethdev ops and rte
>>> flow interface to make codes more easier to maintain.
>>>
>> Agree that it is not ideal to have two different ways for same/similar
>> control path functionality in ethdev.
>>
>> +Ori to discuss if this is a common problem and can be resolved in
>> ethdev layer.
> This patchset is aimed to decouple the configuration API and the
> structure used by
> ethdev ops and rte flow API in driver. I think this can be ignored.
>>
>> I can see some of remaining patches are related to this ethdev / flow
>> API RSS separation. I will continue with this set, but I believe it is
>> better if this issue addressed in higher level.
> As far as I know, the priority of rte flow hash rule is higher than
> ethdev ops by default,
> and there are many other rules.
> But the implementation guide documentation about it is not clear.
> It may be better if we can clarify the documentation to guide driver
> coding.

I am not aware of any priority between flow API hash rule and ethdev
APIs, Ori & Andrew may comment better. And +1 to document this.


>>> This patch separates hash algorithm by following ways:
>>> 1) 'hash_algo' in struct hns3_rss_conf is used for ethdev ops
>>>     interface or default configuration in driver.
>>> 2) Add a 'rte_flow_hash_algo' field in struct hns3_rss_conf
>>>     to save algorithm from rte flow interface. The main reasons
>>>     are as follows:
>>>     Currently, only the last rule is used to restore the rte
>>>     flow rule. If 'func' in RSS action is 'DEFAULT', it means
>>>     that this rule doesn't modify algorithm and driver need to
>>>     save current algorithm for restoring algorithm during reset
>>>     phase.
>>>
>>> Fixes: c37ca66f2b27 ("net/hns3: support RSS")
>>> Cc: stable@dpdk.org
>>>
>>> Signed-off-by: Huisong Li <lihuisong@huawei.com>
>>> Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
>> <...>
>>
>> .


^ permalink raw reply	[flat|nested] 24+ messages in thread

end of thread, other threads:[~2023-03-29  8:13 UTC | newest]

Thread overview: 24+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-03-10  9:35 [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Dongdong Liu
2023-03-10  9:35 ` [PATCH 01/16] net/hns3: fix possible truncation of hash key when config Dongdong Liu
2023-03-10 19:36   ` Ferruh Yigit
2023-03-29  1:56     ` lihuisong (C)
2023-03-10  9:35 ` [PATCH 02/16] net/hns3: fix possible truncation of redirection table Dongdong Liu
2023-03-10 19:36   ` Ferruh Yigit
2023-03-10  9:35 ` [PATCH 03/16] net/hns3: use hardware config to report hash key Dongdong Liu
2023-03-10  9:35 ` [PATCH 04/16] net/hns3: use hardware config to report hash types Dongdong Liu
2023-03-10  9:35 ` [PATCH 05/16] net/hns3: use hardware config to report redirection table Dongdong Liu
2023-03-10  9:35 ` [PATCH 06/16] net/hns3: separate the setting of hash algorithm Dongdong Liu
2023-03-10 19:36   ` Ferruh Yigit
2023-03-29  1:58     ` lihuisong (C)
2023-03-29  8:13       ` Ferruh Yigit
2023-03-10  9:35 ` [PATCH 07/16] net/hns3: separate the setting of hash key Dongdong Liu
2023-03-10  9:35 ` [PATCH 08/16] net/hns3: separate the setting of redirection table Dongdong Liu
2023-03-10  9:35 ` [PATCH 09/16] net/hns3: separate the setting of RSS types Dongdong Liu
2023-03-10  9:35 ` [PATCH 10/16] net/hns3: separate the setting and clearing of RSS rule Dongdong Liu
2023-03-10  9:35 ` [PATCH 11/16] net/hns3: use new RSS rule to configure hardware Dongdong Liu
2023-03-10  9:35 ` [PATCH 12/16] net/hns3: save hash algo to RSS filter list node Dongdong Liu
2023-03-10  9:35 ` [PATCH 13/16] net/hns3: adding queue buffer size hash rule allowed Dongdong Liu
2023-03-10  9:35 ` [PATCH 14/16] net/hns3: separate rte flow RSS config from hns3 rss conf Dongdong Liu
2023-03-10  9:35 ` [PATCH 15/16] net/hns3: reimplement hash flow function Dongdong Liu
2023-03-10  9:35 ` [PATCH 16/16] net/hns3: add the verification of RSS types Dongdong Liu
2023-03-10 20:58 ` [PATCH 00/16] net/hns3: some code refactor for hns3 RSS Ferruh Yigit

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).