igc: Remove no need declaration of the igc_configure
authorSasha Neftin <sasha.neftin@intel.com>
Tue, 19 Nov 2019 11:44:01 +0000 (13:44 +0200)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Sun, 5 Jan 2020 06:47:15 +0000 (22:47 -0800)
We want to avoid forward-declarations of function if possible.
Rearrange the igc_configure function implementation.

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ethernet/intel/igc/igc_main.c

index 95a62f8386682923bf8f1a895518a905439e34a2..1023c9226a8e32f1d60695e4abdee92b72121194 100644 (file)
@@ -53,7 +53,6 @@ MODULE_DEVICE_TABLE(pci, igc_pci_tbl);
 
 /* forward declaration */
 static int igc_sw_init(struct igc_adapter *);
-static void igc_configure(struct igc_adapter *adapter);
 static void igc_set_rx_mode(struct net_device *netdev);
 static void igc_write_itr(struct igc_q_vector *q_vector);
 static void igc_assign_vector(struct igc_q_vector *q_vector, int msix_vector);
@@ -1985,6 +1984,51 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
        return !!budget;
 }
 
+static void igc_nfc_filter_restore(struct igc_adapter *adapter)
+{
+       struct igc_nfc_filter *rule;
+
+       spin_lock(&adapter->nfc_lock);
+
+       hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
+               igc_add_filter(adapter, rule);
+
+       spin_unlock(&adapter->nfc_lock);
+}
+
+/**
+ * igc_configure - configure the hardware for RX and TX
+ * @adapter: private board structure
+ */
+static void igc_configure(struct igc_adapter *adapter)
+{
+       struct net_device *netdev = adapter->netdev;
+       int i = 0;
+
+       igc_get_hw_control(adapter);
+       igc_set_rx_mode(netdev);
+
+       igc_setup_tctl(adapter);
+       igc_setup_mrqc(adapter);
+       igc_setup_rctl(adapter);
+
+       igc_nfc_filter_restore(adapter);
+       igc_configure_tx(adapter);
+       igc_configure_rx(adapter);
+
+       igc_rx_fifo_flush_base(&adapter->hw);
+
+       /* call igc_desc_unused which always leaves
+        * at least 1 descriptor unused to make sure
+        * next_to_use != next_to_clean
+        */
+       for (i = 0; i < adapter->num_rx_queues; i++) {
+               struct igc_ring *ring = adapter->rx_ring[i];
+
+               igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
+       }
+}
+
 /**
  * igc_up - Open the interface and prepare it to handle traffic
  * @adapter: board private structure
@@ -2206,18 +2250,6 @@ static void igc_nfc_filter_exit(struct igc_adapter *adapter)
        spin_unlock(&adapter->nfc_lock);
 }
 
-static void igc_nfc_filter_restore(struct igc_adapter *adapter)
-{
-       struct igc_nfc_filter *rule;
-
-       spin_lock(&adapter->nfc_lock);
-
-       hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
-               igc_add_filter(adapter, rule);
-
-       spin_unlock(&adapter->nfc_lock);
-}
-
 /**
  * igc_down - Close the interface
  * @adapter: board private structure
@@ -2441,39 +2473,6 @@ igc_features_check(struct sk_buff *skb, struct net_device *dev,
        return features;
 }
 
-/**
- * igc_configure - configure the hardware for RX and TX
- * @adapter: private board structure
- */
-static void igc_configure(struct igc_adapter *adapter)
-{
-       struct net_device *netdev = adapter->netdev;
-       int i = 0;
-
-       igc_get_hw_control(adapter);
-       igc_set_rx_mode(netdev);
-
-       igc_setup_tctl(adapter);
-       igc_setup_mrqc(adapter);
-       igc_setup_rctl(adapter);
-
-       igc_nfc_filter_restore(adapter);
-       igc_configure_tx(adapter);
-       igc_configure_rx(adapter);
-
-       igc_rx_fifo_flush_base(&adapter->hw);
-
-       /* call igc_desc_unused which always leaves
-        * at least 1 descriptor unused to make sure
-        * next_to_use != next_to_clean
-        */
-       for (i = 0; i < adapter->num_rx_queues; i++) {
-               struct igc_ring *ring = adapter->rx_ring[i];
-
-               igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
-       }
-}
-
 /* If the filter to be added and an already existing filter express
  * the same address and address type, it should be possible to only
  * override the other configurations, for example the queue to steer