net: mvneta: improve suspend/resume
authorJisheng Zhang <Jisheng.Zhang@synaptics.com>
Mon, 2 Apr 2018 03:24:59 +0000 (11:24 +0800)
committerDavid S. Miller <davem@davemloft.net>
Mon, 2 Apr 2018 15:14:03 +0000 (11:14 -0400)
Current suspend/resume implementation reuses the mvneta_open() and
mvneta_close(), but it could be optimized to take only necessary
actions during suspend/resume.

One obvious problem of current implementation is: after hundreds of
system suspend/resume cycles, the resume of mvneta could fail due to
fragmented dma coherent memory. After this patch, the non-necessary
memory alloc/free is optimized out.

Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/marvell/mvneta.c

index d4fbad2..17a904c 100644 (file)
@@ -4571,16 +4571,45 @@ static int mvneta_remove(struct platform_device *pdev)
 #ifdef CONFIG_PM_SLEEP
 static int mvneta_suspend(struct device *device)
 {
+       int queue;
        struct net_device *dev = dev_get_drvdata(device);
        struct mvneta_port *pp = netdev_priv(dev);
 
+       if (!netif_running(dev))
+               goto clean_exit;
+
+       if (!pp->neta_armada3700) {
+               spin_lock(&pp->lock);
+               pp->is_stopped = true;
+               spin_unlock(&pp->lock);
+
+               cpuhp_state_remove_instance_nocalls(online_hpstate,
+                                                   &pp->node_online);
+               cpuhp_state_remove_instance_nocalls(CPUHP_NET_MVNETA_DEAD,
+                                                   &pp->node_dead);
+       }
+
        rtnl_lock();
-       if (netif_running(dev))
-               mvneta_stop(dev);
+       mvneta_stop_dev(pp);
        rtnl_unlock();
+
+       for (queue = 0; queue < rxq_number; queue++) {
+               struct mvneta_rx_queue *rxq = &pp->rxqs[queue];
+
+               mvneta_rxq_drop_pkts(pp, rxq);
+       }
+
+       for (queue = 0; queue < txq_number; queue++) {
+               struct mvneta_tx_queue *txq = &pp->txqs[queue];
+
+               mvneta_txq_hw_deinit(pp, txq);
+       }
+
+clean_exit:
        netif_device_detach(dev);
        clk_disable_unprepare(pp->clk_bus);
        clk_disable_unprepare(pp->clk);
+
        return 0;
 }
 
@@ -4589,7 +4618,7 @@ static int mvneta_resume(struct device *device)
        struct platform_device *pdev = to_platform_device(device);
        struct net_device *dev = dev_get_drvdata(device);
        struct mvneta_port *pp = netdev_priv(dev);
-       int err;
+       int err, queue;
 
        clk_prepare_enable(pp->clk);
        if (!IS_ERR(pp->clk_bus))
@@ -4611,12 +4640,38 @@ static int mvneta_resume(struct device *device)
        }
 
        netif_device_attach(dev);
-       rtnl_lock();
-       if (netif_running(dev)) {
-               mvneta_open(dev);
-               mvneta_set_rx_mode(dev);
+
+       if (!netif_running(dev))
+               return 0;
+
+       for (queue = 0; queue < rxq_number; queue++) {
+               struct mvneta_rx_queue *rxq = &pp->rxqs[queue];
+
+               rxq->next_desc_to_proc = 0;
+               mvneta_rxq_hw_init(pp, rxq);
+       }
+
+       for (queue = 0; queue < txq_number; queue++) {
+               struct mvneta_tx_queue *txq = &pp->txqs[queue];
+
+               txq->next_desc_to_proc = 0;
+               mvneta_txq_hw_init(pp, txq);
        }
+
+       if (!pp->neta_armada3700) {
+               spin_lock(&pp->lock);
+               pp->is_stopped = false;
+               spin_unlock(&pp->lock);
+               cpuhp_state_add_instance_nocalls(online_hpstate,
+                                                &pp->node_online);
+               cpuhp_state_add_instance_nocalls(CPUHP_NET_MVNETA_DEAD,
+                                                &pp->node_dead);
+       }
+
+       rtnl_lock();
+       mvneta_start_dev(pp);
        rtnl_unlock();
+       mvneta_set_rx_mode(dev);
 
        return 0;
 }