diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/tx.c b/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
index cc6fa00d350b..e1f7a3febb50 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
@@ -2210,6 +2210,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
 	__le16 fc;
 	u8 hdr_len;
 	u16 wifi_seq;
+	bool amsdu;
 
 	txq = &trans_pcie->txq[txq_id];
 	q = &txq->q;
@@ -2301,11 +2302,18 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
 	 */
 	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
 	      hdr_len - IWL_HCMD_SCRATCHBUF_SIZE;
-	tb1_len = ALIGN(len, 4);
-
-	/* Tell NIC about any 2-byte padding after MAC header */
-	if (tb1_len != len)
-		tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
+	/* do not align A-MSDU to dword as the subframe header aligns it */
+	amsdu = ieee80211_is_data_qos(fc) &&
+		(*ieee80211_get_qos_ctl(hdr) &
+		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
+	if (trans_pcie->sw_csum_tx || !amsdu) {
+		tb1_len = ALIGN(len, 4);
+		/* Tell NIC about any 2-byte padding after MAC header */
+		if (tb1_len != len)
+			tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
+	} else {
+		tb1_len = len;
+	}
 
 	/* The first TB points to the scratchbuf data - min_copy bytes */
 	memcpy(&txq->scratchbufs[q->write_ptr], &dev_cmd->hdr,
@@ -2323,8 +2331,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
 		goto out_err;
 	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
 
-	if (ieee80211_is_data_qos(fc) &&
-	    (*ieee80211_get_qos_ctl(hdr) & IEEE80211_QOS_CTL_A_MSDU_PRESENT)) {
+	if (amsdu) {
 		if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
 						     out_meta, dev_cmd,
 						     tb1_len)))