[dpdk-dev] [PATCH v2 01/62] common/cnxk: add support to lock NIX RQ contexts

[Nithin Dabilpuram <ndabilpuram@marvell.com>]

From: Satha Rao <skoteshwar@marvell.com>

This patch will consider device argument to lock rss table
in NIX.

This patch also adds few misc fixes such as disabling NIX Tx
vlan insertion conf in SMQ, enabling SSO in NIX Tx SQ
for Tx completions and TM related stats API.

Signed-off-by: Satha Rao <skoteshwar@marvell.com>
---
 drivers/common/cnxk/roc_nix.h          | 31 ++++++++++--
 drivers/common/cnxk/roc_nix_queue.c    |  2 +
 drivers/common/cnxk/roc_nix_rss.c      | 51 ++++++++++++++++++--
 drivers/common/cnxk/roc_nix_tm_utils.c | 86 +++++++++++++++++++++++++++++++++-
 drivers/common/cnxk/roc_platform.h     |  2 +
 drivers/common/cnxk/version.map        |  1 +
 6 files changed, 163 insertions(+), 10 deletions(-)

diff --git a/drivers/common/cnxk/roc_nix.h b/drivers/common/cnxk/roc_nix.h
index b39f461..6d9ac10 100644
--- a/drivers/common/cnxk/roc_nix.h
+++ b/drivers/common/cnxk/roc_nix.h
@@ -85,10 +85,11 @@ struct roc_nix_eeprom_info {
 #define ROC_NIX_LF_RX_CFG_LEN_OL3     BIT_ULL(41)
 
 /* Group 0 will be used for RSS, 1 -7 will be used for npc_flow RSS action*/
-#define ROC_NIX_RSS_GROUP_DEFAULT 0
-#define ROC_NIX_RSS_GRPS	  8
-#define ROC_NIX_RSS_RETA_MAX	  ROC_NIX_RSS_RETA_SZ_256
-#define ROC_NIX_RSS_KEY_LEN	  48 /* 352 Bits */
+#define ROC_NIX_RSS_GROUP_DEFAULT    0
+#define ROC_NIX_RSS_GRPS	     8
+#define ROC_NIX_RSS_RETA_MAX	     ROC_NIX_RSS_RETA_SZ_256
+#define ROC_NIX_RSS_KEY_LEN	     48 /* 352 Bits */
+#define ROC_NIX_RSS_MCAM_IDX_DEFAULT (-1)
 
 #define ROC_NIX_DEFAULT_HW_FRS 1514
 
@@ -184,6 +185,7 @@ struct roc_nix_sq {
 	enum roc_nix_sq_max_sqe_sz max_sqe_sz;
 	uint32_t nb_desc;
 	uint16_t qid;
+	bool sso_ena;
 	/* End of Input parameters */
 	uint16_t sqes_per_sqb_log2;
 	struct roc_nix *roc_nix;
@@ -241,6 +243,8 @@ struct roc_nix {
 	uint16_t max_sqb_count;
 	enum roc_nix_rss_reta_sz reta_sz;
 	bool enable_loop;
+	bool hw_vlan_ins;
+	uint8_t lock_rx_ctx;
 	/* End of input parameters */
 	/* LMT line base for "Per Core Tx LMT line" mode*/
 	uintptr_t lmt_base;
@@ -371,6 +375,22 @@ struct roc_nix_tm_shaper_profile {
 	void (*free_fn)(void *profile);
 };
 
+enum roc_nix_tm_node_stats_type {
+	ROC_NIX_TM_NODE_PKTS_DROPPED,
+	ROC_NIX_TM_NODE_BYTES_DROPPED,
+	ROC_NIX_TM_NODE_GREEN_PKTS,
+	ROC_NIX_TM_NODE_GREEN_BYTES,
+	ROC_NIX_TM_NODE_YELLOW_PKTS,
+	ROC_NIX_TM_NODE_YELLOW_BYTES,
+	ROC_NIX_TM_NODE_RED_PKTS,
+	ROC_NIX_TM_NODE_RED_BYTES,
+	ROC_NIX_TM_NODE_STATS_MAX,
+};
+
+struct roc_nix_tm_node_stats {
+	uint64_t stats[ROC_NIX_TM_NODE_STATS_MAX];
+};
+
 int __roc_api roc_nix_tm_node_add(struct roc_nix *roc_nix,
 				  struct roc_nix_tm_node *roc_node);
 int __roc_api roc_nix_tm_node_delete(struct roc_nix *roc_nix, uint32_t node_id,
@@ -408,6 +428,9 @@ roc_nix_tm_shaper_profile_get(struct roc_nix *roc_nix, uint32_t profile_id);
 struct roc_nix_tm_shaper_profile *__roc_api roc_nix_tm_shaper_profile_next(
 	struct roc_nix *roc_nix, struct roc_nix_tm_shaper_profile *__prev);
 
+int __roc_api roc_nix_tm_node_stats_get(struct roc_nix *roc_nix,
+					uint32_t node_id, bool clear,
+					struct roc_nix_tm_node_stats *stats);
 /*
  * TM ratelimit tree API.
  */
diff --git a/drivers/common/cnxk/roc_nix_queue.c b/drivers/common/cnxk/roc_nix_queue.c
index fbf7efa..1c62aa2 100644
--- a/drivers/common/cnxk/roc_nix_queue.c
+++ b/drivers/common/cnxk/roc_nix_queue.c
@@ -582,6 +582,7 @@ sq_cn9k_init(struct nix *nix, struct roc_nix_sq *sq, uint32_t rr_quantum,
 	aq->sq.default_chan = nix->tx_chan_base;
 	aq->sq.sqe_stype = NIX_STYPE_STF;
 	aq->sq.ena = 1;
+	aq->sq.sso_ena = !!sq->sso_ena;
 	if (aq->sq.max_sqe_size == NIX_MAXSQESZ_W8)
 		aq->sq.sqe_stype = NIX_STYPE_STP;
 	aq->sq.sqb_aura = roc_npa_aura_handle_to_aura(sq->aura_handle);
@@ -679,6 +680,7 @@ sq_init(struct nix *nix, struct roc_nix_sq *sq, uint32_t rr_quantum,
 	aq->sq.default_chan = nix->tx_chan_base;
 	aq->sq.sqe_stype = NIX_STYPE_STF;
 	aq->sq.ena = 1;
+	aq->sq.sso_ena = !!sq->sso_ena;
 	if (aq->sq.max_sqe_size == NIX_MAXSQESZ_W8)
 		aq->sq.sqe_stype = NIX_STYPE_STP;
 	aq->sq.sqb_aura = roc_npa_aura_handle_to_aura(sq->aura_handle);
diff --git a/drivers/common/cnxk/roc_nix_rss.c b/drivers/common/cnxk/roc_nix_rss.c
index 2d7b84a..7de69aa 100644
--- a/drivers/common/cnxk/roc_nix_rss.c
+++ b/drivers/common/cnxk/roc_nix_rss.c
@@ -52,7 +52,7 @@ roc_nix_rss_key_get(struct roc_nix *roc_nix, uint8_t key[ROC_NIX_RSS_KEY_LEN])
 
 static int
 nix_cn9k_rss_reta_set(struct nix *nix, uint8_t group,
-		      uint16_t reta[ROC_NIX_RSS_RETA_MAX])
+		      uint16_t reta[ROC_NIX_RSS_RETA_MAX], uint8_t lock_rx_ctx)
 {
 	struct mbox *mbox = (&nix->dev)->mbox;
 	struct nix_aq_enq_req *req;
@@ -77,6 +77,27 @@ nix_cn9k_rss_reta_set(struct nix *nix, uint8_t group,
 		req->qidx = (group * nix->reta_sz) + idx;
 		req->ctype = NIX_AQ_CTYPE_RSS;
 		req->op = NIX_AQ_INSTOP_INIT;
+
+		if (!lock_rx_ctx)
+			continue;
+
+		req = mbox_alloc_msg_nix_aq_enq(mbox);
+		if (!req) {
+			/* The shared memory buffer can be full.
+			 * Flush it and retry
+			 */
+			rc = mbox_process(mbox);
+			if (rc < 0)
+				return rc;
+			req = mbox_alloc_msg_nix_aq_enq(mbox);
+			if (!req)
+				return NIX_ERR_NO_MEM;
+		}
+		req->rss.rq = reta[idx];
+		/* Fill AQ info */
+		req->qidx = (group * nix->reta_sz) + idx;
+		req->ctype = NIX_AQ_CTYPE_RSS;
+		req->op = NIX_AQ_INSTOP_LOCK;
 	}
 
 	rc = mbox_process(mbox);
@@ -88,7 +109,7 @@ nix_cn9k_rss_reta_set(struct nix *nix, uint8_t group,
 
 static int
 nix_rss_reta_set(struct nix *nix, uint8_t group,
-		 uint16_t reta[ROC_NIX_RSS_RETA_MAX])
+		 uint16_t reta[ROC_NIX_RSS_RETA_MAX], uint8_t lock_rx_ctx)
 {
 	struct mbox *mbox = (&nix->dev)->mbox;
 	struct nix_cn10k_aq_enq_req *req;
@@ -113,6 +134,27 @@ nix_rss_reta_set(struct nix *nix, uint8_t group,
 		req->qidx = (group * nix->reta_sz) + idx;
 		req->ctype = NIX_AQ_CTYPE_RSS;
 		req->op = NIX_AQ_INSTOP_INIT;
+
+		if (!lock_rx_ctx)
+			continue;
+
+		req = mbox_alloc_msg_nix_cn10k_aq_enq(mbox);
+		if (!req) {
+			/* The shared memory buffer can be full.
+			 * Flush it and retry
+			 */
+			rc = mbox_process(mbox);
+			if (rc < 0)
+				return rc;
+			req = mbox_alloc_msg_nix_cn10k_aq_enq(mbox);
+			if (!req)
+				return NIX_ERR_NO_MEM;
+		}
+		req->rss.rq = reta[idx];
+		/* Fill AQ info */
+		req->qidx = (group * nix->reta_sz) + idx;
+		req->ctype = NIX_AQ_CTYPE_RSS;
+		req->op = NIX_AQ_INSTOP_LOCK;
 	}
 
 	rc = mbox_process(mbox);
@@ -133,9 +175,10 @@ roc_nix_rss_reta_set(struct roc_nix *roc_nix, uint8_t group,
 		return NIX_ERR_PARAM;
 
 	if (roc_model_is_cn9k())
-		rc = nix_cn9k_rss_reta_set(nix, group, reta);
+		rc = nix_cn9k_rss_reta_set(nix, group, reta,
+					   roc_nix->lock_rx_ctx);
 	else
-		rc = nix_rss_reta_set(nix, group, reta);
+		rc = nix_rss_reta_set(nix, group, reta, roc_nix->lock_rx_ctx);
 	if (rc)
 		return rc;
 
diff --git a/drivers/common/cnxk/roc_nix_tm_utils.c b/drivers/common/cnxk/roc_nix_tm_utils.c
index 1d7dd68..6b9543e 100644
--- a/drivers/common/cnxk/roc_nix_tm_utils.c
+++ b/drivers/common/cnxk/roc_nix_tm_utils.c
@@ -409,6 +409,7 @@ nix_tm_topology_reg_prep(struct nix *nix, struct nix_tm_node *node,
 			 volatile uint64_t *reg, volatile uint64_t *regval,
 			 volatile uint64_t *regval_mask)
 {
+	struct roc_nix *roc_nix = nix_priv_to_roc_nix(nix);
 	uint8_t k = 0, hw_lvl, parent_lvl;
 	uint64_t parent = 0, child = 0;
 	enum roc_nix_tm_tree tree;
@@ -454,8 +455,11 @@ nix_tm_topology_reg_prep(struct nix *nix, struct nix_tm_node *node,
 		reg[k] = NIX_AF_SMQX_CFG(schq);
 		regval[k] = (BIT_ULL(50) | NIX_MIN_HW_FRS |
 			     ((nix->mtu & 0xFFFF) << 8));
-		regval_mask[k] =
-			~(BIT_ULL(50) | GENMASK_ULL(6, 0) | GENMASK_ULL(23, 8));
+		/* Maximum Vtag insertion size as a multiple of four bytes */
+		if (roc_nix->hw_vlan_ins)
+			regval[k] |= (0x2ULL << 36);
+		regval_mask[k] = ~(BIT_ULL(50) | GENMASK_ULL(6, 0) |
+				   GENMASK_ULL(23, 8) | GENMASK_ULL(38, 36));
 		k++;
 
 		/* Parent and schedule conf */
@@ -1000,3 +1004,81 @@ nix_tm_shaper_profile_free(struct nix_tm_shaper_profile *profile)
 
 	(profile->free_fn)(profile);
 }
+
+int
+roc_nix_tm_node_stats_get(struct roc_nix *roc_nix, uint32_t node_id, bool clear,
+			  struct roc_nix_tm_node_stats *n_stats)
+{
+	struct nix *nix = roc_nix_to_nix_priv(roc_nix);
+	struct mbox *mbox = (&nix->dev)->mbox;
+	struct nix_txschq_config *req, *rsp;
+	struct nix_tm_node *node;
+	uint32_t schq;
+	int rc, i;
+
+	node = nix_tm_node_search(nix, node_id, ROC_NIX_TM_USER);
+	if (!node)
+		return NIX_ERR_TM_INVALID_NODE;
+
+	if (node->hw_lvl != NIX_TXSCH_LVL_TL1)
+		return NIX_ERR_OP_NOTSUP;
+
+	schq = node->hw_id;
+	/* Skip fetch if not requested */
+	if (!n_stats)
+		goto clear_stats;
+
+	memset(n_stats, 0, sizeof(struct roc_nix_tm_node_stats));
+	/* Check if node has HW resource */
+	if (!(node->flags & NIX_TM_NODE_HWRES))
+		return 0;
+
+	req = mbox_alloc_msg_nix_txschq_cfg(mbox);
+	req->read = 1;
+	req->lvl = NIX_TXSCH_LVL_TL1;
+
+	i = 0;
+	req->reg[i++] = NIX_AF_TL1X_DROPPED_PACKETS(schq);
+	req->reg[i++] = NIX_AF_TL1X_DROPPED_BYTES(schq);
+	req->reg[i++] = NIX_AF_TL1X_GREEN_PACKETS(schq);
+	req->reg[i++] = NIX_AF_TL1X_GREEN_BYTES(schq);
+	req->reg[i++] = NIX_AF_TL1X_YELLOW_PACKETS(schq);
+	req->reg[i++] = NIX_AF_TL1X_YELLOW_BYTES(schq);
+	req->reg[i++] = NIX_AF_TL1X_RED_PACKETS(schq);
+	req->reg[i++] = NIX_AF_TL1X_RED_BYTES(schq);
+	req->num_regs = i;
+
+	rc = mbox_process_msg(mbox, (void **)&rsp);
+	if (rc)
+		return rc;
+
+	/* Return stats */
+	n_stats->stats[ROC_NIX_TM_NODE_PKTS_DROPPED] = rsp->regval[0];
+	n_stats->stats[ROC_NIX_TM_NODE_BYTES_DROPPED] = rsp->regval[1];
+	n_stats->stats[ROC_NIX_TM_NODE_GREEN_PKTS] = rsp->regval[2];
+	n_stats->stats[ROC_NIX_TM_NODE_GREEN_BYTES] = rsp->regval[3];
+	n_stats->stats[ROC_NIX_TM_NODE_YELLOW_PKTS] = rsp->regval[4];
+	n_stats->stats[ROC_NIX_TM_NODE_YELLOW_BYTES] = rsp->regval[5];
+	n_stats->stats[ROC_NIX_TM_NODE_RED_PKTS] = rsp->regval[6];
+	n_stats->stats[ROC_NIX_TM_NODE_RED_BYTES] = rsp->regval[7];
+
+clear_stats:
+	if (!clear)
+		return 0;
+
+	/* Clear all the stats */
+	req = mbox_alloc_msg_nix_txschq_cfg(mbox);
+	req->lvl = NIX_TXSCH_LVL_TL1;
+	i = 0;
+	req->reg[i++] = NIX_AF_TL1X_DROPPED_PACKETS(schq);
+	req->reg[i++] = NIX_AF_TL1X_DROPPED_BYTES(schq);
+	req->reg[i++] = NIX_AF_TL1X_GREEN_PACKETS(schq);
+	req->reg[i++] = NIX_AF_TL1X_GREEN_BYTES(schq);
+	req->reg[i++] = NIX_AF_TL1X_YELLOW_PACKETS(schq);
+	req->reg[i++] = NIX_AF_TL1X_YELLOW_BYTES(schq);
+	req->reg[i++] = NIX_AF_TL1X_RED_PACKETS(schq);
+	req->reg[i++] = NIX_AF_TL1X_RED_BYTES(schq);
+	req->num_regs = i;
+
+	return mbox_process_msg(mbox, (void **)&rsp);
+}
diff --git a/drivers/common/cnxk/roc_platform.h b/drivers/common/cnxk/roc_platform.h
index 7864fa4..911ae15 100644
--- a/drivers/common/cnxk/roc_platform.h
+++ b/drivers/common/cnxk/roc_platform.h
@@ -127,6 +127,8 @@
 #define plt_memzone_reserve_cache_align(name, sz)                              \
 	rte_memzone_reserve_aligned(name, sz, 0, 0, RTE_CACHE_LINE_SIZE)
 #define plt_memzone_free rte_memzone_free
+#define plt_memzone_reserve_aligned(name, len, flags, align)                   \
+	rte_memzone_reserve_aligned((name), (len), 0, (flags), (align))
 
 #define plt_tsc_hz   rte_get_tsc_hz
 #define plt_delay_ms rte_delay_ms
diff --git a/drivers/common/cnxk/version.map b/drivers/common/cnxk/version.map
index 8e67c83..c39d76f 100644
--- a/drivers/common/cnxk/version.map
+++ b/drivers/common/cnxk/version.map
@@ -123,6 +123,7 @@ INTERNAL {
 	roc_nix_tm_node_parent_update;
 	roc_nix_tm_node_pkt_mode_update;
 	roc_nix_tm_node_shaper_update;
+	roc_nix_tm_node_stats_get;
 	roc_nix_tm_node_suspend_resume;
 	roc_nix_tm_prealloc_res;
 	roc_nix_tm_rlimit_sq;
-- 
2.8.4