diff --git a/Kbuild b/Kbuild
index be88d8c36758..3e3125575760 100644
--- a/Kbuild
+++ b/Kbuild
@@ -3535,6 +3535,10 @@ ifdef CONFIG_BEACON_TX_OFFLOAD_MAX_VDEV
 ccflags-y += -DCFG_TGT_DEFAULT_BEACON_TX_OFFLOAD_MAX_VDEV=$(CONFIG_BEACON_TX_OFFLOAD_MAX_VDEV)
 endif
 
+ifdef CONFIG_LIMIT_IPA_TX_BUFFER
+ccflags-y += -DLIMIT_IPA_TX_BUFFER=$(CONFIG_LIMIT_IPA_TX_BUFFER)
+endif
+
 ifdef CONFIG_LOCK_STATS_ON
 ccflags-y += -DQDF_LOCK_STATS=1
 ccflags-y += -DQDF_LOCK_STATS_DESTROY_PRINT=0
diff --git a/core/dp/htt/htt_tx.c b/core/dp/htt/htt_tx.c
index 8a9a711c8a69..395f06681e04 100644
--- a/core/dp/htt/htt_tx.c
+++ b/core/dp/htt/htt_tx.c
@@ -1076,6 +1076,40 @@ void htt_tx_desc_display(void *tx_desc)
 
 #ifdef IPA_OFFLOAD
 #ifdef QCA_WIFI_3_0
+
+#ifndef LIMIT_IPA_TX_BUFFER
+#define LIMIT_IPA_TX_BUFFER 2048
+#endif
+
+/**
+ * htt_tx_ipa_get_tx_buf_count() - Update WDI TX buffers count
+ * @uc_tx_buf_cnt: TX Buffer count
+ *
+ * Return: new uc tx buffer count
+ */
+static int htt_tx_ipa_get_limit_tx_buf_count(unsigned int uc_tx_buf_cnt)
+{
+	/* In order to improve the Genoa IPA DBS KPI, need to set
+	 * IpaUcTxBufCount=2048, so tx complete ring size=2048, and
+	 * total tx buffer count = 2047.
+	 * But in fact, wlan fw just only have 5G 1100 tx desc +
+	 * 2.4G 400 desc, it can cover about 1500 packets from
+	 * IPA side.
+	 * So the remaining 2047-1500 packet are not used,
+	 * in order to save some memory, so we can use
+	 * LIMIT_IPA_TX_BUFFER to limit the max tx buffer
+	 * count, which varied from platform.
+	 * And then the tx buffer count always equal to tx complete
+	 * ring size -1 is not mandatory now.
+	 * From the trying, it has the same KPI achievement while
+	 * set LIMIT_IPA_TX_BUFFER=1500 or 2048.
+	 */
+	if (uc_tx_buf_cnt > LIMIT_IPA_TX_BUFFER)
+		return LIMIT_IPA_TX_BUFFER;
+	else
+		return uc_tx_buf_cnt;
+}
+
 /**
  * htt_tx_ipa_uc_wdi_tx_buf_alloc() - Alloc WDI TX buffers
  * @pdev: htt context
@@ -1094,11 +1128,9 @@ static int htt_tx_ipa_uc_wdi_tx_buf_alloc(struct htt_pdev_t *pdev,
 					  unsigned int uc_tx_partition_base)
 {
 	unsigned int tx_buffer_count;
-	unsigned int  tx_buffer_count_pwr2;
 	qdf_dma_addr_t buffer_paddr;
 	uint32_t *header_ptr;
 	target_paddr_t *ring_vaddr;
-	uint16_t idx;
 	qdf_mem_info_t *mem_map_table = NULL, *mem_info = NULL;
 	qdf_shared_mem_t *shared_tx_buffer;
 
@@ -1121,7 +1153,7 @@ static int htt_tx_ipa_uc_wdi_tx_buf_alloc(struct htt_pdev_t *pdev,
 		if (!shared_tx_buffer || !shared_tx_buffer->vaddr) {
 			qdf_print("IPA WDI TX buffer alloc fail %d allocated\n",
 				tx_buffer_count);
-			goto pwr2;
+			goto out;
 		}
 
 		header_ptr = shared_tx_buffer->vaddr;
@@ -1172,38 +1204,14 @@ static int htt_tx_ipa_uc_wdi_tx_buf_alloc(struct htt_pdev_t *pdev,
 		}
 	}
 
-pwr2:
-	/*
-	 * Tx complete ring buffer count should be power of 2.
-	 * So, allocated Tx buffer count should be one less than ring buffer
-	 * size.
-	 */
-	tx_buffer_count_pwr2 = qdf_rounddown_pow_of_two(tx_buffer_count + 1)
-			       - 1;
-	if (tx_buffer_count > tx_buffer_count_pwr2) {
-		QDF_TRACE(QDF_MODULE_ID_HTT, QDF_TRACE_LEVEL_INFO,
-			  "%s: Allocated Tx buffer count %d is rounded down to %d",
-			  __func__, tx_buffer_count, tx_buffer_count_pwr2);
-
-		/* Free over allocated buffers below power of 2 */
-		for (idx = tx_buffer_count_pwr2; idx < tx_buffer_count; idx++) {
-			if (pdev->ipa_uc_tx_rsc.tx_buf_pool_strg[idx]) {
-				qdf_mem_shared_mem_free(pdev->osdev,
-					pdev->ipa_uc_tx_rsc.tx_buf_pool_strg[
-								idx]);
-				 pdev->ipa_uc_tx_rsc.tx_buf_pool_strg[idx] =
-									NULL;
-			}
-		}
-	}
-
+out:
 	if (qdf_mem_smmu_s1_enabled(pdev->osdev)) {
-		cds_smmu_map_unmap(true, tx_buffer_count_pwr2,
+		cds_smmu_map_unmap(true, tx_buffer_count,
 				   mem_map_table);
 		qdf_mem_free(mem_map_table);
 	}
 
-	return tx_buffer_count_pwr2;
+	return tx_buffer_count;
 }
 
 /**
@@ -1252,6 +1260,11 @@ static void htt_tx_buf_pool_free(struct htt_pdev_t *pdev)
 	}
 }
 #else
+static int htt_tx_ipa_get_limit_tx_buf_count(unsigned int uc_tx_buf_cnt)
+{
+	return uc_tx_buf_cnt;
+}
+
 static int htt_tx_ipa_uc_wdi_tx_buf_alloc(struct htt_pdev_t *pdev,
 					  unsigned int uc_tx_buf_sz,
 					  unsigned int uc_tx_buf_cnt,
@@ -1436,6 +1449,7 @@ int htt_tx_ipa_uc_attach(struct htt_pdev_t *pdev,
 		goto free_tx_ce_idx;
 	}
 
+	uc_tx_buf_cnt = htt_tx_ipa_get_limit_tx_buf_count(uc_tx_buf_cnt);
 	/* Allocate TX BUF vAddress Storage */
 	pdev->ipa_uc_tx_rsc.tx_buf_pool_strg =
 		qdf_mem_malloc(uc_tx_buf_cnt *