wlan-ap-Telecominfraproject/feeds/ipq95xx/mac80211/patches/qca/718-c-ath12k-optimize-datapath-tx-completion-routine.patch

From 5ef626578252e8ee8dfdbd08f9cdbca1ab557f27 Mon Sep 17 00:00:00 2001
From: Ramanathan Choodamani <quic_rchoodam@quicinc.com>
Date: Sat, 4 Mar 2023 20:31:42 -0800
Subject: [PATCH] ath12k: optimize datapath tx completion routine

Avoid copying all descriptors to a cached memory list and then
processing it again. Avoid two loops to process it.
Use a single loop to process the tx completion

Signed-off-by: Balamurugan Mahalingam <quic_bmahalin@quicinc.com>

Signed-off-by: Ramanathan Choodamani <quic_rchoodam@quicinc.com>
---
 drivers/net/wireless/ath/ath12k/dp.c    | 19 +----------------
 drivers/net/wireless/ath/ath12k/dp.h    |  3 ---
 drivers/net/wireless/ath/ath12k/dp_tx.c | 36 +++++++++++++--------------------
 3 files changed, 15 insertions(+), 43 deletions(-)

diff --git a/drivers/net/wireless/ath/ath12k/dp.c b/drivers/net/wireless/ath/ath12k/dp.c
index 0af711b..5aa0147 100644
--- a/drivers/net/wireless/ath/ath12k/dp.c
+++ b/drivers/net/wireless/ath/ath12k/dp.c
@@ -247,9 +247,9 @@ int ath12k_dp_srng_setup(struct ath12k_base *ab, struct dp_srng *ring,
 		/* Allocate the reo dst and tx completion rings from cacheable memory */
 		switch (type) {
 		case HAL_REO_DST:
-		case HAL_WBM2SW_RELEASE:
 			cached = true;
 			break;
+		case HAL_WBM2SW_RELEASE:
 		default:
 			cached = false;
 		}
@@ -1295,7 +1295,6 @@ static void ath12k_dp_reoq_lut_cleanup(struct ath12k_base *ab)
 void ath12k_dp_free(struct ath12k_base *ab)
 {
 	struct ath12k_dp *dp = &ab->dp;
-	int i;
 
 	ath12k_dp_link_desc_cleanup(ab, dp->link_desc_banks,
 				    HAL_WBM_IDLE_LINK, &dp->wbm_idle_ring);
@@ -1307,9 +1306,6 @@ void ath12k_dp_free(struct ath12k_base *ab)
 
 	ath12k_dp_rx_reo_cmd_list_cleanup(ab);
 
-	for (i = 0; i < ab->hw_params->max_tx_ring; i++)
-		kfree(dp->tx_ring[i].tx_status);
-
 	ath12k_dp_rx_free(ab);
 	/* Deinit any SOC level resource */
 }
@@ -1723,17 +1719,6 @@ int ath12k_dp_alloc(struct ath12k_base *ab)
 
 	for (i = 0; i < ab->hw_params->max_tx_ring; i++) {
 		dp->tx_ring[i].tcl_data_ring_id = i;
-
-		dp->tx_ring[i].tx_status_head = 0;
-		dp->tx_ring[i].tx_status_tail = DP_TX_COMP_RING_SIZE - 1;
-		dp->tx_ring[i].tx_status = kmalloc(size, GFP_KERNEL);
-		if (!dp->tx_ring[i].tx_status) {
-			ret = -ENOMEM;
-			/* FIXME The allocated tx status is not freed
-			 * properly here
-			 */
-			goto fail_cmn_reoq_cleanup;
-		}
 	}
 
 	for (i = 0; i < HAL_DSCP_TID_MAP_TBL_NUM_ENTRIES_MAX; i++)
@@ -1749,8 +1734,6 @@ int ath12k_dp_alloc(struct ath12k_base *ab)
 
 fail_dp_rx_free:
 	ath12k_dp_rx_free(ab);
-
-fail_cmn_reoq_cleanup:
 	ath12k_dp_reoq_lut_cleanup(ab);
 
 fail_cmn_srng_cleanup:
diff --git a/drivers/net/wireless/ath/ath12k/dp.h b/drivers/net/wireless/ath/ath12k/dp.h
index 3726c61..d994767 100644
--- a/drivers/net/wireless/ath/ath12k/dp.h
+++ b/drivers/net/wireless/ath/ath12k/dp.h
@@ -44,9 +44,6 @@ struct dp_tx_ring {
 	u8 tcl_data_ring_id;
 	struct dp_srng tcl_data_ring;
 	struct dp_srng tcl_comp_ring;
-	struct hal_wbm_completion_ring_tx *tx_status;
-	int tx_status_head;
-	int tx_status_tail;
 };
 
 struct ath12k_pdev_mon_stats {
diff --git a/drivers/net/wireless/ath/ath12k/dp_tx.c b/drivers/net/wireless/ath/ath12k/dp_tx.c
index a461c7d..e4a67fb 100644
--- a/drivers/net/wireless/ath/ath12k/dp_tx.c
+++ b/drivers/net/wireless/ath/ath12k/dp_tx.c
@@ -932,8 +932,8 @@ void ath12k_dp_tx_completion_handler(struct ath12k_base *ab, int ring_id)
 	u32 *desc, desc_id;
 	u8 mac_id;
 	u64 desc_va;
-	int valid_entries, count = 0, i = 0;
-	struct hal_wbm_completion_ring_tx *tx_status;
+	int valid_entries;
+	struct hal_wbm_completion_ring_tx tx_status = { 0 };
 	enum hal_wbm_rel_src_module buf_rel_source;
 
 	spin_lock_bh(&status_ring->lock);
@@ -953,29 +953,18 @@ void ath12k_dp_tx_completion_handler(struct ath12k_base *ab, int ring_id)
 		if (!ath12k_dp_tx_completion_valid(desc))
 			continue;
 
-		memcpy(&tx_ring->tx_status[count],
-		       desc, sizeof(struct hal_wbm_release_ring));
-		count++;
+		memcpy(&tx_status, desc,
+				sizeof(struct hal_wbm_completion_ring_tx));
 
-		if (count == DP_TX_COMP_RING_SIZE)
-			break;
-	}
-
-	ath12k_hal_srng_access_end(ab, status_ring);
-
-	spin_unlock_bh(&status_ring->lock);
 
-	while (count--) {
-		tx_status = &tx_ring->tx_status[i++];
-
-		if (likely(u32_get_bits(tx_status->info0, HAL_WBM_COMPL_TX_INFO0_CC_DONE))) {
+		if (likely(FIELD_GET(HAL_WBM_COMPL_TX_INFO0_CC_DONE, tx_status.info0))) {
 			/* HW done cookie conversion */
-			desc_va = ((u64)tx_status->buf_va_hi << 32 |
-					tx_status->buf_va_lo);
+			desc_va = ((u64)tx_status.buf_va_hi << 32 |
+					tx_status.buf_va_lo);
 			tx_desc = (struct ath12k_tx_desc_info *)((unsigned long)desc_va);
 		} else {
 			/* SW does cookie conversion to VA */
-			desc_id = u32_get_bits(tx_status->buf_va_hi,
+			desc_id = u32_get_bits(tx_status.buf_va_hi,
 					       BUFFER_ADDR_INFO1_SW_COOKIE);
 
 			tx_desc = ath12k_dp_get_tx_desc(ab, desc_id);
@@ -993,11 +982,11 @@ void ath12k_dp_tx_completion_handler(struct ath12k_base *ab, int ring_id)
 		 */
 		ath12k_dp_tx_release_txbuf(dp, tx_desc, tx_desc->pool_id);
 		buf_rel_source = FIELD_GET(HAL_WBM_RELEASE_INFO0_REL_SRC_MODULE,
-					   tx_status->info0);
+					   tx_status.info0);
 		ab->soc_stats.wbm_rel_source[buf_rel_source]++;
 		if (unlikely(buf_rel_source == HAL_WBM_REL_SRC_MODULE_FW)) {
 			ath12k_dp_tx_process_htt_tx_complete(ab,
-							     (void *)tx_status,
+							     (void *)&tx_status,
 							     mac_id, msdu,
 							     tx_ring,
 							     skb_ext_desc,
@@ -1010,10 +999,13 @@ void ath12k_dp_tx_completion_handler(struct ath12k_base *ab, int ring_id)
 		if (atomic_dec_and_test(&ar->dp.num_tx_pending))
 			wake_up(&ar->dp.tx_empty_waitq);
 
-		ath12k_dp_tx_complete_msdu(ar, msdu, tx_status,
+		ath12k_dp_tx_complete_msdu(ar, msdu, &tx_status,
 					   buf_rel_source, skb_ext_desc,
 					   tx_ring->tcl_data_ring_id);
 	}
+	ath12k_hal_srng_access_end(ab, status_ring);
+
+	spin_unlock_bh(&status_ring->lock);
 }
 
 static int
-- 
2.7.4