net: stmmac: adding multiple napi mechanism
authorJoao Pinto <Joao.Pinto@synopsys.com>
Thu, 6 Apr 2017 08:49:11 +0000 (09:49 +0100)
committerDavid S. Miller <davem@davemloft.net>
Fri, 7 Apr 2017 14:18:27 +0000 (07:18 -0700)
This patch adds the napi variable to the stmmac_rx_queue
structure and forces that operations like netif_queue_stopped,
netif_wake_queue, netif_stop_queue, netdev_reset_queue and
netdev_sent_queue be made by queue.

Signed-off-by: Joao Pinto <jpinto@synopsys.com>
Tested-by: Niklas Cassel <niklas.cassel@axis.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/stmicro/stmmac/stmmac.h
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c

index 359f8fd..33efe70 100644 (file)
@@ -72,6 +72,7 @@ struct stmmac_rx_queue {
        u32 rx_zeroc_thresh;
        dma_addr_t dma_rx_phy;
        u32 rx_tail_addr;
+       struct napi_struct napi ____cacheline_aligned_in_smp;
 };
 
 struct stmmac_priv {
@@ -91,8 +92,6 @@ struct stmmac_priv {
        u32 rx_riwt;
        int hwts_rx_en;
 
-       struct napi_struct napi ____cacheline_aligned_in_smp;
-
        void __iomem *ioaddr;
        struct net_device *dev;
        struct device *device;
index 56a081f..a89f76b 100644 (file)
@@ -138,6 +138,64 @@ static void stmmac_verify_args(void)
                eee_timer = STMMAC_DEFAULT_LPI_TIMER;
 }
 
+/**
+ * stmmac_disable_all_queues - Disable all queues
+ * @priv: driver private structure
+ */
+static void stmmac_disable_all_queues(struct stmmac_priv *priv)
+{
+       u32 rx_queues_cnt = priv->plat->rx_queues_to_use;
+       u32 queue;
+
+       for (queue = 0; queue < rx_queues_cnt; queue++) {
+               struct stmmac_rx_queue *rx_q = &priv->rx_queue[queue];
+
+               napi_disable(&rx_q->napi);
+       }
+}
+
+/**
+ * stmmac_enable_all_queues - Enable all queues
+ * @priv: driver private structure
+ */
+static void stmmac_enable_all_queues(struct stmmac_priv *priv)
+{
+       u32 rx_queues_cnt = priv->plat->rx_queues_to_use;
+       u32 queue;
+
+       for (queue = 0; queue < rx_queues_cnt; queue++) {
+               struct stmmac_rx_queue *rx_q = &priv->rx_queue[queue];
+
+               napi_enable(&rx_q->napi);
+       }
+}
+
+/**
+ * stmmac_stop_all_queues - Stop all queues
+ * @priv: driver private structure
+ */
+static void stmmac_stop_all_queues(struct stmmac_priv *priv)
+{
+       u32 tx_queues_cnt = priv->plat->tx_queues_to_use;
+       u32 queue;
+
+       for (queue = 0; queue < tx_queues_cnt; queue++)
+               netif_tx_stop_queue(netdev_get_tx_queue(priv->dev, queue));
+}
+
+/**
+ * stmmac_start_all_queues - Start all queues
+ * @priv: driver private structure
+ */
+static void stmmac_start_all_queues(struct stmmac_priv *priv)
+{
+       u32 tx_queues_cnt = priv->plat->tx_queues_to_use;
+       u32 queue;
+
+       for (queue = 0; queue < tx_queues_cnt; queue++)
+               netif_tx_start_queue(netdev_get_tx_queue(priv->dev, queue));
+}
+
 /**
  * stmmac_clk_csr_set - dynamically set the MDC clock
  * @priv: driver private structure
@@ -1262,7 +1320,6 @@ static int init_dma_tx_desc_rings(struct net_device *dev)
 
                for (i = 0; i < DMA_TX_SIZE; i++) {
                        struct dma_desc *p;
-
                        if (priv->extend_desc)
                                p = &((tx_q->dma_etx + i)->basic);
                        else
@@ -1286,9 +1343,9 @@ static int init_dma_tx_desc_rings(struct net_device *dev)
 
                tx_q->dirty_tx = 0;
                tx_q->cur_tx = 0;
-       }
 
-       netdev_reset_queue(priv->dev);
+               netdev_tx_reset_queue(netdev_get_tx_queue(priv->dev, queue));
+       }
 
        return 0;
 }
@@ -1805,13 +1862,16 @@ static void stmmac_tx_clean(struct stmmac_priv *priv, u32 queue)
        }
        tx_q->dirty_tx = entry;
 
-       netdev_completed_queue(priv->dev, pkts_compl, bytes_compl);
+       netdev_tx_completed_queue(netdev_get_tx_queue(priv->dev, queue),
+                                 pkts_compl, bytes_compl);
+
+       if (unlikely(netif_tx_queue_stopped(netdev_get_tx_queue(priv->dev,
+                                                               queue))) &&
+           stmmac_tx_avail(priv, queue) > STMMAC_TX_THRESH) {
 
-       if (unlikely(netif_queue_stopped(priv->dev) &&
-           stmmac_tx_avail(priv, queue) > STMMAC_TX_THRESH)) {
                netif_dbg(priv, tx_done, priv->dev,
                          "%s: restart transmit\n", __func__);
-               netif_wake_queue(priv->dev);
+               netif_tx_wake_queue(netdev_get_tx_queue(priv->dev, queue));
        }
 
        if ((priv->eee_enabled) && (!priv->tx_path_in_lpi_mode)) {
@@ -1843,7 +1903,7 @@ static void stmmac_tx_err(struct stmmac_priv *priv, u32 chan)
        struct stmmac_tx_queue *tx_q = &priv->tx_queue[chan];
        int i;
 
-       netif_stop_queue(priv->dev);
+       netif_tx_stop_queue(netdev_get_tx_queue(priv->dev, chan));
 
        stmmac_stop_tx_dma(priv, chan);
        dma_free_tx_skbufs(priv, chan);
@@ -1858,11 +1918,11 @@ static void stmmac_tx_err(struct stmmac_priv *priv, u32 chan)
                                                     (i == DMA_TX_SIZE - 1));
        tx_q->dirty_tx = 0;
        tx_q->cur_tx = 0;
-       netdev_reset_queue(priv->dev);
+       netdev_tx_reset_queue(netdev_get_tx_queue(priv->dev, chan));
        stmmac_start_tx_dma(priv, chan);
 
        priv->dev->stats.tx_errors++;
-       netif_wake_queue(priv->dev);
+       netif_tx_wake_queue(netdev_get_tx_queue(priv->dev, chan));
 }
 
 /**
@@ -1907,12 +1967,14 @@ static void stmmac_dma_interrupt(struct stmmac_priv *priv)
        u32 chan;
 
        for (chan = 0; chan < tx_channel_count; chan++) {
+               struct stmmac_rx_queue *rx_q = &priv->rx_queue[chan];
+
                status = priv->hw->dma->dma_interrupt(priv->ioaddr,
                                                      &priv->xstats, chan);
                if (likely((status & handle_rx)) || (status & handle_tx)) {
-                       if (likely(napi_schedule_prep(&priv->napi))) {
+                       if (likely(napi_schedule_prep(&rx_q->napi))) {
                                stmmac_disable_dma_irq(priv, chan);
-                               __napi_schedule(&priv->napi);
+                               __napi_schedule(&rx_q->napi);
                        }
                }
 
@@ -2554,8 +2616,8 @@ static int stmmac_open(struct net_device *dev)
                }
        }
 
-       napi_enable(&priv->napi);
-       netif_start_queue(dev);
+       stmmac_enable_all_queues(priv);
+       stmmac_start_all_queues(priv);
 
        return 0;
 
@@ -2598,9 +2660,9 @@ static int stmmac_release(struct net_device *dev)
                phy_disconnect(dev->phydev);
        }
 
-       netif_stop_queue(dev);
+       stmmac_stop_all_queues(priv);
 
-       napi_disable(&priv->napi);
+       stmmac_disable_all_queues(priv);
 
        del_timer_sync(&priv->txtimer);
 
@@ -2717,8 +2779,9 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev)
        /* Desc availability based on threshold should be enough safe */
        if (unlikely(stmmac_tx_avail(priv, queue) <
                (((skb->len - proto_hdr_len) / TSO_MAX_BUFF_SIZE + 1)))) {
-               if (!netif_queue_stopped(dev)) {
-                       netif_stop_queue(dev);
+               if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, queue))) {
+                       netif_tx_stop_queue(netdev_get_tx_queue(priv->dev,
+                                                               queue));
                        /* This is a hard error, log it. */
                        netdev_err(priv->dev,
                                   "%s: Tx Ring full when queue awake\n",
@@ -2798,7 +2861,7 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev)
        if (unlikely(stmmac_tx_avail(priv, queue) <= (MAX_SKB_FRAGS + 1))) {
                netif_dbg(priv, hw, priv->dev, "%s: stop transmitted packets\n",
                          __func__);
-               netif_stop_queue(dev);
+               netif_tx_stop_queue(netdev_get_tx_queue(priv->dev, queue));
        }
 
        dev->stats.tx_bytes += skb->len;
@@ -2855,7 +2918,7 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev)
                print_pkt(skb->data, skb_headlen(skb));
        }
 
-       netdev_sent_queue(dev, skb->len);
+       netdev_tx_sent_queue(netdev_get_tx_queue(dev, queue), skb->len);
 
        priv->hw->dma->set_tx_tail_ptr(priv->ioaddr, tx_q->tx_tail_addr,
                                       queue);
@@ -2899,8 +2962,9 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
        }
 
        if (unlikely(stmmac_tx_avail(priv, queue) < nfrags + 1)) {
-               if (!netif_queue_stopped(dev)) {
-                       netif_stop_queue(dev);
+               if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, queue))) {
+                       netif_tx_stop_queue(netdev_get_tx_queue(priv->dev,
+                                                               queue));
                        /* This is a hard error, log it. */
                        netdev_err(priv->dev,
                                   "%s: Tx Ring full when queue awake\n",
@@ -2998,7 +3062,7 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
        if (unlikely(stmmac_tx_avail(priv, queue) <= (MAX_SKB_FRAGS + 1))) {
                netif_dbg(priv, hw, priv->dev, "%s: stop transmitted packets\n",
                          __func__);
-               netif_stop_queue(dev);
+               netif_tx_stop_queue(netdev_get_tx_queue(priv->dev, queue));
        }
 
        dev->stats.tx_bytes += skb->len;
@@ -3061,7 +3125,7 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
                dma_wmb();
        }
 
-       netdev_sent_queue(dev, skb->len);
+       netdev_tx_sent_queue(netdev_get_tx_queue(dev, queue), skb->len);
 
        if (priv->synopsys_id < DWMAC_CORE_4_00)
                priv->hw->dma->enable_dma_transmission(priv->ioaddr);
@@ -3361,7 +3425,7 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit, u32 queue)
                        else
                                skb->ip_summed = CHECKSUM_UNNECESSARY;
 
-                       napi_gro_receive(&priv->napi, skb);
+                       napi_gro_receive(&rx_q->napi, skb);
 
                        priv->dev->stats.rx_packets++;
                        priv->dev->stats.rx_bytes += frame_len;
@@ -3386,11 +3450,13 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit, u32 queue)
  */
 static int stmmac_poll(struct napi_struct *napi, int budget)
 {
-       struct stmmac_priv *priv = container_of(napi, struct stmmac_priv, napi);
+       struct stmmac_rx_queue *rx_q =
+               container_of(napi, struct stmmac_rx_queue, napi);
+       struct stmmac_priv *priv = rx_q->priv_data;
        u32 tx_count = priv->plat->tx_queues_to_use;
-       u32 chan = STMMAC_CHAN0;
+       u32 chan = rx_q->queue_index;
        int work_done = 0;
-       u32 queue = chan;
+       u32 queue;
 
        priv->xstats.napi_poll++;
 
@@ -3398,9 +3464,7 @@ static int stmmac_poll(struct napi_struct *napi, int budget)
        for (queue = 0; queue < tx_count; queue++)
                stmmac_tx_clean(priv, queue);
 
-       queue = chan;
-
-       work_done = stmmac_rx(priv, budget, queue);
+       work_done = stmmac_rx(priv, budget, rx_q->queue_index);
        if (work_done < budget) {
                napi_complete_done(napi, work_done);
                stmmac_enable_dma_irq(priv, chan);
@@ -3989,11 +4053,14 @@ int stmmac_dvr_probe(struct device *device,
                     struct plat_stmmacenet_data *plat_dat,
                     struct stmmac_resources *res)
 {
-       int ret = 0;
        struct net_device *ndev = NULL;
        struct stmmac_priv *priv;
+       int ret = 0;
+       u32 queue;
 
-       ndev = alloc_etherdev(sizeof(struct stmmac_priv));
+       ndev = alloc_etherdev_mqs(sizeof(struct stmmac_priv),
+                                 MTL_MAX_TX_QUEUES,
+                                 MTL_MAX_RX_QUEUES);
        if (!ndev)
                return -ENOMEM;
 
@@ -4035,6 +4102,10 @@ int stmmac_dvr_probe(struct device *device,
        if (ret)
                goto error_hw_init;
 
+       /* Configure real RX and TX queues */
+       ndev->real_num_rx_queues = priv->plat->rx_queues_to_use;
+       ndev->real_num_tx_queues = priv->plat->tx_queues_to_use;
+
        ndev->netdev_ops = &stmmac_netdev_ops;
 
        ndev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
@@ -4084,7 +4155,12 @@ int stmmac_dvr_probe(struct device *device,
                         "Enable RX Mitigation via HW Watchdog Timer\n");
        }
 
-       netif_napi_add(ndev, &priv->napi, stmmac_poll, 64);
+       for (queue = 0; queue < priv->plat->rx_queues_to_use; queue++) {
+               struct stmmac_rx_queue *rx_q = &priv->rx_queue[queue];
+
+               netif_napi_add(ndev, &rx_q->napi, stmmac_poll,
+                              (8 * priv->plat->rx_queues_to_use));
+       }
 
        spin_lock_init(&priv->lock);
 
@@ -4129,7 +4205,11 @@ error_netdev_register:
            priv->hw->pcs != STMMAC_PCS_RTBI)
                stmmac_mdio_unregister(ndev);
 error_mdio_register:
-       netif_napi_del(&priv->napi);
+       for (queue = 0; queue < priv->plat->rx_queues_to_use; queue++) {
+               struct stmmac_rx_queue *rx_q = &priv->rx_queue[queue];
+
+               netif_napi_del(&rx_q->napi);
+       }
 error_hw_init:
        free_netdev(ndev);
 
@@ -4191,9 +4271,9 @@ int stmmac_suspend(struct device *dev)
        spin_lock_irqsave(&priv->lock, flags);
 
        netif_device_detach(ndev);
-       netif_stop_queue(ndev);
+       stmmac_stop_all_queues(priv);
 
-       napi_disable(&priv->napi);
+       stmmac_disable_all_queues(priv);
 
        /* Stop TX/RX DMA */
        stmmac_stop_all_dma(priv);
@@ -4296,9 +4376,9 @@ int stmmac_resume(struct device *dev)
        stmmac_init_tx_coalesce(priv);
        stmmac_set_rx_mode(ndev);
 
-       napi_enable(&priv->napi);
+       stmmac_enable_all_queues(priv);
 
-       netif_start_queue(ndev);
+       stmmac_start_all_queues(priv);
 
        spin_unlock_irqrestore(&priv->lock, flags);