wlan-ap-Telecominfraproject/feeds/ipq95xx/mac80211/patches/qca/718-d-ath12k-remove-locks-and-optimize-tx-completion-path.patch

From c7f71aa364633e3a28a1a5506b12819c64c3424a Mon Sep 17 00:00:00 2001
From: Ramanathan Choodamani <quic_rchoodam@quicinc.com>
Date: Sat, 4 Mar 2023 20:36:53 -0800
Subject: [PATCH] ath12k: remove locks and optimize tx completion  path

Avoid taking locks while processing the tx completion. The rings
are accessed associated with specific cores for processing. There
is no need to take locks to avoid contention.

Introduce simple inline functions to update the rings directly
without too many checks in the ath12k_hif_write32 access

Move the tx completion processing to a new IRQ group to assign
specific irq smp affinity and process the tx completions
on a specific core

Signed-off-by: Balamurugan Mahalingam <quic_bmahalin@quicinc.com>
Signed-off-by: Ramanathan Choodamani <quic_rchoodam@quicinc.com>
---
 drivers/net/wireless/ath/ath12k/dp_tx.c | 11 +++--------
 drivers/net/wireless/ath/ath12k/hal.c   |  5 ++---
 drivers/net/wireless/ath/ath12k/hal.h   | 23 +++++++++++++++++++++++
 drivers/net/wireless/ath/ath12k/hw.c    |  2 +-
 4 files changed, 29 insertions(+), 12 deletions(-)

diff --git a/drivers/net/wireless/ath/ath12k/dp_tx.c b/drivers/net/wireless/ath/ath12k/dp_tx.c
index e4a67fb..9fd274f 100644
--- a/drivers/net/wireless/ath/ath12k/dp_tx.c
+++ b/drivers/net/wireless/ath/ath12k/dp_tx.c
@@ -70,11 +70,9 @@ static void ath12k_dp_tx_release_txbuf(struct ath12k_dp *dp,
 				       struct ath12k_tx_desc_info *tx_desc,
 				       u8 ring_id)
 {
-	spin_lock_bh(&dp->tx_desc_lock[ring_id]);
 	tx_desc->skb = NULL;
 	tx_desc->skb_ext_desc = NULL;
 	list_move_tail(&tx_desc->list, &dp->tx_desc_free_list[ring_id]);
-	spin_unlock_bh(&dp->tx_desc_lock[ring_id]);
 }
 
 static inline
@@ -936,14 +934,12 @@ void ath12k_dp_tx_completion_handler(struct ath12k_base *ab, int ring_id)
 	struct hal_wbm_completion_ring_tx tx_status = { 0 };
 	enum hal_wbm_rel_src_module buf_rel_source;
 
-	spin_lock_bh(&status_ring->lock);
 
-	ath12k_hal_srng_access_begin(ab, status_ring);
+	ath12k_hal_srng_access_dst_ring_begin_nolock(ab, status_ring);
 
 	valid_entries = ath12k_hal_srng_dst_num_free(ab, status_ring, false);
 	if (!valid_entries) {
-		ath12k_hal_srng_access_end(ab, status_ring);
-		spin_unlock_bh(&status_ring->lock);
+		ath12k_hal_srng_access_umac_dst_ring_end_nolock(ab->mem, status_ring);
 		return;
 	}
 
@@ -1003,9 +999,8 @@ void ath12k_dp_tx_completion_handler(struct ath12k_base *ab, int ring_id)
 					   buf_rel_source, skb_ext_desc,
 					   tx_ring->tcl_data_ring_id);
 	}
-	ath12k_hal_srng_access_end(ab, status_ring);
+	ath12k_hal_srng_access_umac_dst_ring_end_nolock(ab->mem, status_ring);
 
-	spin_unlock_bh(&status_ring->lock);
 }
 
 static int
diff --git a/drivers/net/wireless/ath/ath12k/hal.c b/drivers/net/wireless/ath/ath12k/hal.c
index 2fb931f..cea5562 100644
--- a/drivers/net/wireless/ath/ath12k/hal.c
+++ b/drivers/net/wireless/ath/ath12k/hal.c
@@ -1580,8 +1580,8 @@ u32 *ath12k_hal_srng_dst_peek(struct ath12k_base *ab, struct hal_srng *srng)
 	return NULL;
 }
 
-static void ath12k_hal_srng_prefetch_desc(struct ath12k_base *ab,
-                                          struct hal_srng *srng)
+void ath12k_hal_srng_prefetch_desc(struct ath12k_base *ab,
+				   struct hal_srng *srng)
 {
 	u32 *desc;
 
@@ -1652,7 +1652,6 @@ void ath12k_hal_srng_dst_invalidate_entry(struct ath12k_base *ab,
 	u32 *desc;
 	u32 tp, hp;
 
-	lockdep_assert_held(&srng->lock);
 
 	if (!(srng->flags & HAL_SRNG_FLAGS_CACHED) || !entries)
 	        return;
diff --git a/drivers/net/wireless/ath/ath12k/hal.h b/drivers/net/wireless/ath/ath12k/hal.h
index 542c3ac..74f5984 100644
--- a/drivers/net/wireless/ath/ath12k/hal.h
+++ b/drivers/net/wireless/ath/ath12k/hal.h
@@ -1195,6 +1195,8 @@ void ath12k_hal_reo_ring_ctrl_hash_ix3_setup(struct ath12k_base *ab,
 void ath12k_hal_reo_ring_ctrl_hash_ix2_setup(struct ath12k_base *ab,
 					     u32 ring_hash_map);
 void ath12k_hal_reo_shared_qaddr_cache_clear(struct ath12k_base *ab);
+void ath12k_hal_srng_prefetch_desc(struct ath12k_base *ab,
+				   struct hal_srng *srng);
 
 static inline
 u32 *ath12k_hal_srng_src_get_next_entry_nolock(struct ath12k_base *ab,
@@ -1236,6 +1238,16 @@ void ath12k_hal_srng_access_src_ring_begin_nolock(void *mem,
 }
 
 static inline
+void ath12k_hal_srng_access_dst_ring_begin_nolock(struct ath12k_base *ab,
+						  struct hal_srng *srng)
+{
+	srng->u.dst_ring.cached_hp = *srng->u.dst_ring.hp_addr;
+	/* Try to prefetch the next descriptor in the ring */
+	if (srng->flags & HAL_SRNG_FLAGS_CACHED)
+		ath12k_hal_srng_prefetch_desc(ab, srng);
+}
+
+static inline
 void ath12k_hal_srng_access_umac_src_ring_end_nolock(void *mem,
 						     struct hal_srng *srng)
 {
@@ -1248,4 +1260,15 @@ void ath12k_hal_srng_access_umac_src_ring_end_nolock(void *mem,
 	srng->timestamp = jiffies;
 }
 
+static inline
+void ath12k_hal_srng_access_umac_dst_ring_end_nolock(void *mem,
+						     struct hal_srng *srng)
+{
+	u32 offset = (unsigned long)srng->u.dst_ring.tp_addr - (unsigned long)mem;
+	srng->u.dst_ring.last_hp = *srng->u.dst_ring.hp_addr;
+	writel_relaxed(srng->u.dst_ring.tp, mem + HAL_DP_REG_WINDOW_OFFSET
+		       + (offset & WINDOW_RANGE_MASK));
+	srng->timestamp = jiffies;
+}
+
 #endif
diff --git a/drivers/net/wireless/ath/ath12k/hw.c b/drivers/net/wireless/ath/ath12k/hw.c
index 45f5aef..ce7c0a0 100644
--- a/drivers/net/wireless/ath/ath12k/hw.c
+++ b/drivers/net/wireless/ath/ath12k/hw.c
@@ -307,8 +307,8 @@ const struct ath12k_hw_ring_mask ath12k_hw_ring_mask_qcn92xx = {
 		ATH12K_TX_RING_MASK_0,
 		ATH12K_TX_RING_MASK_1,
 		ATH12K_TX_RING_MASK_2,
-		ATH12K_TX_RING_MASK_4,
 		0, 0, 0, 0, 0, 0, 0, 0,
+		ATH12K_TX_RING_MASK_4,
 	},
 	.rx_mon_dest = {
 		0, 0, 0, 0, 0, 0, 0, 0,
-- 
2.7.4