net: ethernet: mediatek: enhance the locking using the lightweight ones
authorSean Wang <sean.wang@mediatek.com>
Thu, 11 Aug 2016 09:51:00 +0000 (17:51 +0800)
committerDavid S. Miller <davem@davemloft.net>
Sat, 13 Aug 2016 03:22:56 +0000 (20:22 -0700)
Since these critical sections protected by page_lock are all entered
from the user context or bottom half context, they can be replaced
with the spin_lock() or spin_lock_bh instead of spin_lock_irqsave().

Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Acked-by: John Crispin <john@phrozen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mediatek/mtk_eth_soc.c

index b57ae3a..3a4726e 100644 (file)
@@ -353,18 +353,17 @@ static int mtk_set_mac_address(struct net_device *dev, void *p)
        int ret = eth_mac_addr(dev, p);
        struct mtk_mac *mac = netdev_priv(dev);
        const char *macaddr = dev->dev_addr;
-       unsigned long flags;
 
        if (ret)
                return ret;
 
-       spin_lock_irqsave(&mac->hw->page_lock, flags);
+       spin_lock_bh(&mac->hw->page_lock);
        mtk_w32(mac->hw, (macaddr[0] << 8) | macaddr[1],
                MTK_GDMA_MAC_ADRH(mac->id));
        mtk_w32(mac->hw, (macaddr[2] << 24) | (macaddr[3] << 16) |
                (macaddr[4] << 8) | macaddr[5],
                MTK_GDMA_MAC_ADRL(mac->id));
-       spin_unlock_irqrestore(&mac->hw->page_lock, flags);
+       spin_unlock_bh(&mac->hw->page_lock);
 
        return 0;
 }
@@ -748,7 +747,6 @@ static int mtk_start_xmit(struct sk_buff *skb, struct net_device *dev)
        struct mtk_eth *eth = mac->hw;
        struct mtk_tx_ring *ring = &eth->tx_ring;
        struct net_device_stats *stats = &dev->stats;
-       unsigned long flags;
        bool gso = false;
        int tx_num;
 
@@ -756,14 +754,14 @@ static int mtk_start_xmit(struct sk_buff *skb, struct net_device *dev)
         * however we have 2 queues running on the same ring so we need to lock
         * the ring access
         */
-       spin_lock_irqsave(&eth->page_lock, flags);
+       spin_lock(&eth->page_lock);
 
        tx_num = mtk_cal_txd_req(skb);
        if (unlikely(atomic_read(&ring->free_count) <= tx_num)) {
                mtk_stop_queue(eth);
                netif_err(eth, tx_queued, dev,
                          "Tx Ring full when queue awake!\n");
-               spin_unlock_irqrestore(&eth->page_lock, flags);
+               spin_unlock(&eth->page_lock);
                return NETDEV_TX_BUSY;
        }
 
@@ -788,12 +786,12 @@ static int mtk_start_xmit(struct sk_buff *skb, struct net_device *dev)
        if (unlikely(atomic_read(&ring->free_count) <= ring->thresh))
                mtk_stop_queue(eth);
 
-       spin_unlock_irqrestore(&eth->page_lock, flags);
+       spin_unlock(&eth->page_lock);
 
        return NETDEV_TX_OK;
 
 drop:
-       spin_unlock_irqrestore(&eth->page_lock, flags);
+       spin_unlock(&eth->page_lock);
        stats->tx_dropped++;
        dev_kfree_skb(skb);
        return NETDEV_TX_OK;
@@ -1347,16 +1345,15 @@ static int mtk_open(struct net_device *dev)
 
 static void mtk_stop_dma(struct mtk_eth *eth, u32 glo_cfg)
 {
-       unsigned long flags;
        u32 val;
        int i;
 
        /* stop the dma engine */
-       spin_lock_irqsave(&eth->page_lock, flags);
+       spin_lock_bh(&eth->page_lock);
        val = mtk_r32(eth, glo_cfg);
        mtk_w32(eth, val & ~(MTK_TX_WB_DDONE | MTK_RX_DMA_EN | MTK_TX_DMA_EN),
                glo_cfg);
-       spin_unlock_irqrestore(&eth->page_lock, flags);
+       spin_unlock_bh(&eth->page_lock);
 
        /* wait for dma stop */
        for (i = 0; i < 10; i++) {