qcacmn: Rx optimizations to improve small pkt perf

basically tried reading peer_id from peer structure
to avoid 1 cache-line miss per pkt and access mac address
from frame data for WDS only if WDS condition is valid.
The other optimization is to avoid accessing queue_mapping
field of skb for skb received which is at 140 offset
This again helps to avoid i cache miss.
Another place is to do a memcpy only when we intend to
add an ast entry.

Change-Id: I7e328060c505bc21419d045e77329c2cda9e3644
CRs-fixed: 2388857

dp/wifi3.0/dp_peer.c

@@ -626,6 +626,10 @@ int dp_peer_add_ast(struct dp_soc *soc,
 		ast_entry = dp_peer_ast_hash_find_by_pdevid(soc, mac_addr,
 							    pdev->pdev_id);
 		if (ast_entry) {
+			if ((type == CDP_TXRX_AST_TYPE_MEC) &&
+			    (ast_entry->type == CDP_TXRX_AST_TYPE_MEC))
+				ast_entry->is_active = TRUE;
+
 			qdf_spin_unlock_bh(&soc->ast_lock);
 			return 0;
 		}
@@ -834,6 +838,13 @@ void dp_peer_del_ast(struct dp_soc *soc, struct dp_ast_entry *ast_entry)
 
 	dp_peer_ast_send_wds_del(soc, ast_entry);
 
+	/*
+	 * release the reference only if it is mapped
+	 * to ast_table
+	 */
+	if (ast_entry->is_mapped)
+		soc->ast_table[ast_entry->ast_idx] = NULL;
+
 	/*
 	 * if peer map v2 is enabled we are not freeing ast entry
 	 * here and it is supposed to be freed in unmap event (after
@@ -853,13 +864,6 @@ void dp_peer_del_ast(struct dp_soc *soc, struct dp_ast_entry *ast_entry)
 			return;
 	}
 
-	/*
-	 * release the reference only if it is mapped
-	 * to ast_table
-	 */
-	if (ast_entry->is_mapped)
-		soc->ast_table[ast_entry->ast_idx] = NULL;
-
 	/* SELF and STATIC entries are removed in teardown itself */
 	if (ast_entry->next_hop)
 		TAILQ_REMOVE(&peer->ast_entry_list, ast_entry, ase_list_elem);

dp/wifi3.0/dp_rx.c

@@ -1386,15 +1386,11 @@ static void dp_rx_msdu_stats_update(struct dp_soc *soc,
 				    struct cdp_tid_rx_stats *tid_stats)
 {
 	bool is_ampdu, is_not_amsdu;
-	uint16_t peer_id;
 	uint32_t sgi, mcs, tid, nss, bw, reception_type, pkt_type;
 	struct dp_vdev *vdev = peer->vdev;
 	qdf_ether_header_t *eh;
 	uint16_t msdu_len = QDF_NBUF_CB_RX_PKT_LEN(nbuf);
 
-	peer_id = DP_PEER_METADATA_PEER_ID_GET(
-			       hal_rx_mpdu_peer_meta_data_get(rx_tlv_hdr));
-
 	is_not_amsdu = qdf_nbuf_is_rx_chfrag_start(nbuf) &
 			qdf_nbuf_is_rx_chfrag_end(nbuf);
 
@@ -1481,7 +1477,7 @@ static void dp_rx_msdu_stats_update(struct dp_soc *soc,
 			return;
 
 		dp_wdi_event_handler(WDI_EVENT_UPDATE_DP_STATS, vdev->pdev->soc,
-				     &peer->stats, peer_id,
+				     &peer->stats, peer->peer_ids[0],
 				     UPDATE_PEER_STATS,
 				     vdev->pdev->pdev_id);
 #endif
@@ -2105,8 +2101,6 @@ done:
 		if (soc->process_rx_status)
 			dp_rx_cksum_offload(vdev->pdev, nbuf, rx_tlv_hdr);
 
-		dp_set_rx_queue(nbuf, ring_id);
-
 		/* Update the protocol tag in SKB based on CCE metadata */
 		dp_rx_update_protocol_tag(soc, vdev, nbuf, rx_tlv_hdr,
 					  reo_ring_num, false, true);

dp/wifi3.0/dp_rx.h

@@ -439,7 +439,7 @@ void dp_rx_add_to_free_desc_list(union dp_rx_desc_list_elem_t **head,
 #ifdef FEATURE_WDS
 static inline void
 dp_rx_wds_add_or_update_ast(struct dp_soc *soc, struct dp_peer *ta_peer,
-			    uint8_t *wds_src_mac, uint8_t is_ad4_valid,
+			    qdf_nbuf_t nbuf, uint8_t is_ad4_valid,
 			    uint8_t is_sa_valid, uint8_t is_chfrag_start,
 			    uint16_t sa_idx, uint16_t sa_sw_peer_id)
 {
@@ -449,6 +449,7 @@ dp_rx_wds_add_or_update_ast(struct dp_soc *soc, struct dp_peer *ta_peer,
 	uint32_t ret = 0;
 	struct dp_neighbour_peer *neighbour_peer = NULL;
 	struct dp_pdev *pdev = ta_peer->vdev->pdev;
+	uint8_t wds_src_mac[QDF_MAC_ADDR_SIZE];
 
 	/* For AP mode : Do wds source port learning only if it is a
 	 * 4-address mpdu
@@ -470,6 +471,10 @@ dp_rx_wds_add_or_update_ast(struct dp_soc *soc, struct dp_peer *ta_peer,
 	}
 
 	if (qdf_unlikely(!is_sa_valid)) {
+		qdf_mem_copy(wds_src_mac,
+			     (qdf_nbuf_data(nbuf) + QDF_MAC_ADDR_SIZE),
+			     QDF_MAC_ADDR_SIZE);
+
 		ret = dp_peer_add_ast(soc,
 					ta_peer,
 					wds_src_mac,
@@ -497,6 +502,10 @@ dp_rx_wds_add_or_update_ast(struct dp_soc *soc, struct dp_peer *ta_peer,
 		 * cached.
 		 */
 		if (!soc->ast_override_support) {
+			qdf_mem_copy(wds_src_mac,
+				     (qdf_nbuf_data(nbuf) + QDF_MAC_ADDR_SIZE),
+				     QDF_MAC_ADDR_SIZE);
+
 			ret = dp_peer_add_ast(soc,
 					      ta_peer,
 					      wds_src_mac,
@@ -512,6 +521,11 @@ dp_rx_wds_add_or_update_ast(struct dp_soc *soc, struct dp_peer *ta_peer,
 			 * to FW.
 			 */
 			if (pdev->neighbour_peers_added) {
+				qdf_mem_copy(wds_src_mac,
+					     (qdf_nbuf_data(nbuf) +
+					      QDF_MAC_ADDR_SIZE),
+					     QDF_MAC_ADDR_SIZE);
+
 				qdf_spin_lock_bh(&pdev->neighbour_peer_mutex);
 				TAILQ_FOREACH(neighbour_peer,
 					      &pdev->neighbour_peers_list,
@@ -624,7 +638,6 @@ dp_rx_wds_srcport_learn(struct dp_soc *soc,
 {
 	uint16_t sa_sw_peer_id = hal_rx_msdu_end_sa_sw_peer_id_get(rx_tlv_hdr);
 	uint8_t sa_is_valid = hal_rx_msdu_end_sa_is_valid_get(rx_tlv_hdr);
-	uint8_t wds_src_mac[QDF_MAC_ADDR_SIZE];
 	uint16_t sa_idx;
 	uint8_t is_chfrag_start = 0;
 	uint8_t is_ad4_valid = 0;
@@ -636,15 +649,12 @@ dp_rx_wds_srcport_learn(struct dp_soc *soc,
 	if (is_chfrag_start)
 		is_ad4_valid = hal_rx_get_mpdu_mac_ad4_valid(rx_tlv_hdr);
 
-	memcpy(wds_src_mac, (qdf_nbuf_data(nbuf) + QDF_MAC_ADDR_SIZE),
-	       QDF_MAC_ADDR_SIZE);
-
 	/*
 	 * Get the AST entry from HW SA index and mark it as active
 	 */
 	sa_idx = hal_rx_msdu_end_sa_idx_get(rx_tlv_hdr);
 
-	dp_rx_wds_add_or_update_ast(soc, ta_peer, wds_src_mac, is_ad4_valid,
+	dp_rx_wds_add_or_update_ast(soc, ta_peer, nbuf, is_ad4_valid,
 				    sa_is_valid, is_chfrag_start,
 				    sa_idx, sa_sw_peer_id);