net: dpaa: reduce number of synchronize_net() calls

In the function dpaa_napi_del(), we execute the netif_napi_del()
for each cpu, which is actually a high overhead operation
because each call to netif_napi_del() contains a synchronize_net(),
i.e. an RCU operation. In fact, it is only necessary to call
 __netif_napi_del and use synchronize_net() once outside of the loop.
This change is similar to commit 2543a600 ("gro_cells: reduce
number of synchronize_net() calls") and commit 5198d545 (" net:
remove napi_hash_del() from driver-facing API") 5198d545.
Signed-off-by: Xi Huang <xuiagnh@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20240822072042.42750-1-xuiagnh@gmail.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/freescale/dpaa/dpaa_eth.c

@@ -3156,8 +3156,9 @@ static void dpaa_napi_del(struct net_device *net_dev)
 	for_each_possible_cpu(cpu) {
 		percpu_priv = per_cpu_ptr(priv->percpu_priv, cpu);
 
-		netif_napi_del(&percpu_priv->np.napi);
+		__netif_napi_del(&percpu_priv->np.napi);
 	}
+	synchronize_net();
 }
 
 static inline void dpaa_bp_free_pf(const struct dpaa_bp *bp,