[PATCH v2] net/cnxk: support Tx queue descriptor count

[skoteshwar at marvell.com]

From: Satha Rao <skoteshwar at marvell.com>

Added CNXK APIs to get used txq descriptor count.

Signed-off-by: Satha Rao <skoteshwar at marvell.com>
---

Depends-on: series-30833 ("ethdev: support Tx queue used count")

v2:
  Updated release notes and fixed API for CPT queues.

 doc/guides/nics/features/cnxk.ini      |  1 +
 doc/guides/rel_notes/release_24_03.rst |  1 +
 drivers/net/cnxk/cn10k_tx_select.c     | 22 ++++++++++++++++++++++
 drivers/net/cnxk/cn9k_tx_select.c      | 23 +++++++++++++++++++++++
 drivers/net/cnxk/cnxk_ethdev.h         | 24 ++++++++++++++++++++++++
 5 files changed, 71 insertions(+)

diff --git a/doc/guides/nics/features/cnxk.ini b/doc/guides/nics/features/cnxk.ini
index 94e7a6a..ab18f38 100644
--- a/doc/guides/nics/features/cnxk.ini
+++ b/doc/guides/nics/features/cnxk.ini
@@ -40,6 +40,7 @@ Timesync             = Y
 Timestamp offload    = Y
 Rx descriptor status = Y
 Tx descriptor status = Y
+Tx queue count       = Y
 Basic stats          = Y
 Stats per queue      = Y
 Extended stats       = Y
diff --git a/doc/guides/rel_notes/release_24_03.rst b/doc/guides/rel_notes/release_24_03.rst
index 32d0ad8..5f8fb9e 100644
--- a/doc/guides/rel_notes/release_24_03.rst
+++ b/doc/guides/rel_notes/release_24_03.rst
@@ -110,6 +110,7 @@ New Features
 
   * Added support for ``RTE_FLOW_ITEM_TYPE_PPPOES`` flow item.
   * Added support for ``RTE_FLOW_ACTION_TYPE_SAMPLE`` flow item.
+  * Added support for fast path function ``rte_eth_tx_queue_count``.
 
 * **Updated Marvell OCTEON EP driver.**
 
diff --git a/drivers/net/cnxk/cn10k_tx_select.c b/drivers/net/cnxk/cn10k_tx_select.c
index 404f5ba..aa0620e 100644
--- a/drivers/net/cnxk/cn10k_tx_select.c
+++ b/drivers/net/cnxk/cn10k_tx_select.c
@@ -20,6 +20,24 @@
 			eth_dev->tx_pkt_burst;
 }
 
+#if defined(RTE_ARCH_ARM64)
+static int
+cn10k_nix_tx_queue_count(void *tx_queue)
+{
+	struct cn10k_eth_txq *txq = (struct cn10k_eth_txq *)tx_queue;
+
+	return cnxk_nix_tx_queue_count(txq->fc_mem, txq->sqes_per_sqb_log2);
+}
+
+static int
+cn10k_nix_tx_queue_sec_count(void *tx_queue)
+{
+	struct cn10k_eth_txq *txq = (struct cn10k_eth_txq *)tx_queue;
+
+	return cnxk_nix_tx_queue_sec_count(txq->fc_mem, txq->sqes_per_sqb_log2, txq->cpt_fc);
+}
+#endif
+
 void
 cn10k_eth_set_tx_function(struct rte_eth_dev *eth_dev)
 {
@@ -63,6 +81,10 @@
 		if (dev->tx_offloads & RTE_ETH_TX_OFFLOAD_MULTI_SEGS)
 			pick_tx_func(eth_dev, nix_eth_tx_vec_burst_mseg);
 	}
+	if (dev->tx_offloads & RTE_ETH_TX_OFFLOAD_SECURITY)
+		eth_dev->tx_queue_count = cn10k_nix_tx_queue_sec_count;
+	else
+		eth_dev->tx_queue_count = cn10k_nix_tx_queue_count;
 
 	rte_mb();
 #else
diff --git a/drivers/net/cnxk/cn9k_tx_select.c b/drivers/net/cnxk/cn9k_tx_select.c
index e08883f..0e09ed6 100644
--- a/drivers/net/cnxk/cn9k_tx_select.c
+++ b/drivers/net/cnxk/cn9k_tx_select.c
@@ -20,6 +20,24 @@
 			eth_dev->tx_pkt_burst;
 }
 
+#if defined(RTE_ARCH_ARM64)
+static int
+cn9k_nix_tx_queue_count(void *tx_queue)
+{
+	struct cn9k_eth_txq *txq = (struct cn9k_eth_txq *)tx_queue;
+
+	return cnxk_nix_tx_queue_count(txq->fc_mem, txq->sqes_per_sqb_log2);
+}
+
+static int
+cn9k_nix_tx_queue_sec_count(void *tx_queue)
+{
+	struct cn9k_eth_txq *txq = (struct cn9k_eth_txq *)tx_queue;
+
+	return cnxk_nix_tx_queue_sec_count(txq->fc_mem, txq->sqes_per_sqb_log2, txq->cpt_fc);
+}
+#endif
+
 void
 cn9k_eth_set_tx_function(struct rte_eth_dev *eth_dev)
 {
@@ -60,6 +78,11 @@
 			pick_tx_func(eth_dev, nix_eth_tx_vec_burst_mseg);
 	}
 
+	if (dev->tx_offloads & RTE_ETH_TX_OFFLOAD_SECURITY)
+		eth_dev->tx_queue_count = cn9k_nix_tx_queue_sec_count;
+	else
+		eth_dev->tx_queue_count = cn9k_nix_tx_queue_count;
+
 	rte_mb();
 #else
 	RTE_SET_USED(eth_dev);
diff --git a/drivers/net/cnxk/cnxk_ethdev.h b/drivers/net/cnxk/cnxk_ethdev.h
index 37b6395..f810bb8 100644
--- a/drivers/net/cnxk/cnxk_ethdev.h
+++ b/drivers/net/cnxk/cnxk_ethdev.h
@@ -458,6 +458,30 @@ struct cnxk_eth_txq_sp {
 	return ((struct cnxk_eth_txq_sp *)__txq) - 1;
 }
 
+static inline int
+cnxk_nix_tx_queue_count(uint64_t *mem, uint16_t sqes_per_sqb_log2)
+{
+	uint64_t val;
+
+	val = rte_atomic_load_explicit(mem, rte_memory_order_relaxed);
+	val = (val << sqes_per_sqb_log2) - val;
+
+	return (val & 0xFFFF);
+}
+
+static inline int
+cnxk_nix_tx_queue_sec_count(uint64_t *mem, uint16_t sqes_per_sqb_log2, uint64_t *sec_fc)
+{
+	uint64_t sq_cnt, sec_cnt, val;
+
+	sq_cnt = rte_atomic_load_explicit(mem, rte_memory_order_relaxed);
+	sq_cnt = (sq_cnt << sqes_per_sqb_log2) - sq_cnt;
+	sec_cnt = rte_atomic_load_explicit(sec_fc, rte_memory_order_relaxed);
+	val = RTE_MAX(sq_cnt, sec_cnt);
+
+	return (val & 0xFFFF);
+}
+
 /* Common ethdev ops */
 extern struct eth_dev_ops cnxk_eth_dev_ops;
 
-- 
1.8.3.1