]> git.baikalelectronics.ru Git - kernel.git/commitdiff
ixgbe: avoid bringing rings up/down as macvlans are added/removed
authorAlexander Duyck <alexander.h.duyck@intel.com>
Wed, 22 Nov 2017 18:57:29 +0000 (10:57 -0800)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Fri, 12 Jan 2018 16:20:28 +0000 (08:20 -0800)
This change makes it so that instead of bringing rings up/down for various
we just update the netdev pointer for the Rx ring and set or clear the MAC
filter for the interface. By doing it this way we can avoid a number of
races and issues in the code as things were getting messy with the macvlan
clean-up racing with the interface clean-up to bring the rings down on
shutdown.

With this change we opt to leave the rings owned by the PF interface for
both Tx and Rx and just direct the packets once they are received to the
macvlan netdev.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c
drivers/net/ethernet/intel/ixgbe/ixgbe_main.c

index dc7f3ef2957b4db17c08d770e24220ec58ab04b4..cfe5a6af04d0c2fdddcba5165e2756b44c050a07 100644 (file)
@@ -46,7 +46,7 @@ static bool ixgbe_cache_ring_dcb_sriov(struct ixgbe_adapter *adapter)
 #endif /* IXGBE_FCOE */
        struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
        int i;
-       u16 reg_idx;
+       u16 reg_idx, pool;
        u8 tcs = adapter->hw_tcs;
 
        /* verify we have DCB queueing enabled before proceeding */
@@ -58,12 +58,16 @@ static bool ixgbe_cache_ring_dcb_sriov(struct ixgbe_adapter *adapter)
                return false;
 
        /* start at VMDq register offset for SR-IOV enabled setups */
+       pool = 0;
        reg_idx = vmdq->offset * __ALIGN_MASK(1, ~vmdq->mask);
-       for (i = 0; i < adapter->num_rx_queues; i++, reg_idx++) {
+       for (i = 0, pool = 0; i < adapter->num_rx_queues; i++, reg_idx++) {
                /* If we are greater than indices move to next pool */
-               if ((reg_idx & ~vmdq->mask) >= tcs)
+               if ((reg_idx & ~vmdq->mask) >= tcs) {
+                       pool++;
                        reg_idx = __ALIGN_MASK(reg_idx, ~vmdq->mask);
+               }
                adapter->rx_ring[i]->reg_idx = reg_idx;
+               adapter->rx_ring[i]->netdev = pool ? NULL : adapter->netdev;
        }
 
        reg_idx = vmdq->offset * __ALIGN_MASK(1, ~vmdq->mask);
@@ -92,6 +96,7 @@ static bool ixgbe_cache_ring_dcb_sriov(struct ixgbe_adapter *adapter)
                for (i = fcoe->offset; i < adapter->num_rx_queues; i++) {
                        reg_idx = __ALIGN_MASK(reg_idx, ~vmdq->mask) + fcoe_tc;
                        adapter->rx_ring[i]->reg_idx = reg_idx;
+                       adapter->rx_ring[i]->netdev = adapter->netdev;
                        reg_idx++;
                }
 
@@ -182,6 +187,7 @@ static bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
                for (i = 0; i < rss_i; i++, tx_idx++, rx_idx++) {
                        adapter->tx_ring[offset + i]->reg_idx = tx_idx;
                        adapter->rx_ring[offset + i]->reg_idx = rx_idx;
+                       adapter->rx_ring[offset + i]->netdev = adapter->netdev;
                        adapter->tx_ring[offset + i]->dcb_tc = tc;
                        adapter->rx_ring[offset + i]->dcb_tc = tc;
                }
@@ -206,14 +212,15 @@ static bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
 #endif /* IXGBE_FCOE */
        struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
        struct ixgbe_ring_feature *rss = &adapter->ring_feature[RING_F_RSS];
+       u16 reg_idx, pool;
        int i;
-       u16 reg_idx;
 
        /* only proceed if VMDq is enabled */
        if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED))
                return false;
 
        /* start at VMDq register offset for SR-IOV enabled setups */
+       pool = 0;
        reg_idx = vmdq->offset * __ALIGN_MASK(1, ~vmdq->mask);
        for (i = 0; i < adapter->num_rx_queues; i++, reg_idx++) {
 #ifdef IXGBE_FCOE
@@ -222,15 +229,20 @@ static bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
                        break;
 #endif
                /* If we are greater than indices move to next pool */
-               if ((reg_idx & ~vmdq->mask) >= rss->indices)
+               if ((reg_idx & ~vmdq->mask) >= rss->indices) {
+                       pool++;
                        reg_idx = __ALIGN_MASK(reg_idx, ~vmdq->mask);
+               }
                adapter->rx_ring[i]->reg_idx = reg_idx;
+               adapter->rx_ring[i]->netdev = pool ? NULL : adapter->netdev;
        }
 
 #ifdef IXGBE_FCOE
        /* FCoE uses a linear block of queues so just assigning 1:1 */
-       for (; i < adapter->num_rx_queues; i++, reg_idx++)
+       for (; i < adapter->num_rx_queues; i++, reg_idx++) {
                adapter->rx_ring[i]->reg_idx = reg_idx;
+               adapter->rx_ring[i]->netdev = adapter->netdev;
+       }
 
 #endif
        reg_idx = vmdq->offset * __ALIGN_MASK(1, ~vmdq->mask);
@@ -267,8 +279,10 @@ static bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
 {
        int i, reg_idx;
 
-       for (i = 0; i < adapter->num_rx_queues; i++)
+       for (i = 0; i < adapter->num_rx_queues; i++) {
                adapter->rx_ring[i]->reg_idx = i;
+               adapter->rx_ring[i]->netdev = adapter->netdev;
+       }
        for (i = 0, reg_idx = 0; i < adapter->num_tx_queues; i++, reg_idx++)
                adapter->tx_ring[i]->reg_idx = reg_idx;
        for (i = 0; i < adapter->num_xdp_queues; i++, reg_idx++)
index 89f7b16c47b7d14d1c35473e98a39e8f5a498358..cdb8502ae473b49875df8f50c676bec4d32a7d01 100644 (file)
@@ -1922,10 +1922,13 @@ static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
        if (IS_ERR(skb))
                return true;
 
-       /* verify that the packet does not have any known errors */
-       if (unlikely(ixgbe_test_staterr(rx_desc,
-                                       IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
-           !(netdev->features & NETIF_F_RXALL))) {
+       /* Verify netdev is present, and that packet does not have any
+        * errors that would be unacceptable to the netdev.
+        */
+       if (!netdev ||
+           (unlikely(ixgbe_test_staterr(rx_desc,
+                                        IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
+            !(netdev->features & NETIF_F_RXALL)))) {
                dev_kfree_skb_any(skb);
                return true;
        }
@@ -5337,33 +5340,6 @@ static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
        rx_ring->next_to_use = 0;
 }
 
-static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
-                                  struct ixgbe_ring *rx_ring)
-{
-       struct ixgbe_adapter *adapter = vadapter->real_adapter;
-
-       /* shutdown specific queue receive and wait for dma to settle */
-       ixgbe_disable_rx_queue(adapter, rx_ring);
-       usleep_range(10000, 20000);
-       ixgbe_irq_disable_queues(adapter, BIT_ULL(rx_ring->queue_index));
-       ixgbe_clean_rx_ring(rx_ring);
-}
-
-static int ixgbe_fwd_ring_down(struct net_device *vdev,
-                              struct ixgbe_fwd_adapter *accel)
-{
-       struct ixgbe_adapter *adapter = accel->real_adapter;
-       unsigned int rxbase = accel->rx_base_queue;
-       int i;
-
-       for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
-               ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
-               adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
-       }
-
-       return 0;
-}
-
 static int ixgbe_fwd_ring_up(struct net_device *vdev,
                             struct ixgbe_fwd_adapter *accel)
 {
@@ -5383,25 +5359,26 @@ static int ixgbe_fwd_ring_up(struct net_device *vdev,
        accel->tx_base_queue = baseq;
 
        for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
-               ixgbe_disable_fwd_ring(accel, adapter->rx_ring[baseq + i]);
-
-       for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
                adapter->rx_ring[baseq + i]->netdev = vdev;
-               ixgbe_configure_rx_ring(adapter, adapter->rx_ring[baseq + i]);
-       }
+
+       /* Guarantee all rings are updated before we update the
+        * MAC address filter.
+        */
+       wmb();
 
        /* ixgbe_add_mac_filter will return an index if it succeeds, so we
         * need to only treat it as an error value if it is negative.
         */
        err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
                                   VMDQ_P(accel->pool));
-       if (err < 0)
-               goto fwd_queue_err;
+       if (err >= 0) {
+               ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
+               return 0;
+       }
+
+       for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
+               adapter->rx_ring[baseq + i]->netdev = NULL;
 
-       ixgbe_macvlan_set_rx_mode(vdev, VMDQ_P(accel->pool), adapter);
-       return 0;
-fwd_queue_err:
-       ixgbe_fwd_ring_down(vdev, accel);
        return err;
 }
 
@@ -9801,15 +9778,38 @@ static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
 
 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
 {
-       struct ixgbe_fwd_adapter *fwd_adapter = priv;
-       struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
-       unsigned int limit;
+       struct ixgbe_fwd_adapter *accel = priv;
+       struct ixgbe_adapter *adapter = accel->real_adapter;
+       unsigned int rxbase = accel->rx_base_queue;
+       unsigned int limit, i;
 
-       clear_bit(fwd_adapter->pool, adapter->fwd_bitmask);
+       /* delete unicast filter associated with offloaded interface */
+       ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
+                            VMDQ_P(accel->pool));
 
+       /* disable ability to receive packets for this pool */
+       IXGBE_WRITE_REG(&adapter->hw, IXGBE_VMOLR(accel->pool), 0);
+
+       /* Allow remaining Rx packets to get flushed out of the
+        * Rx FIFO before we drop the netdev for the ring.
+        */
+       usleep_range(10000, 20000);
+
+       for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
+               struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
+               struct ixgbe_q_vector *qv = ring->q_vector;
+
+               /* Make sure we aren't processing any packets and clear
+                * netdev to shut down the ring.
+                */
+               if (netif_running(adapter->netdev))
+                       napi_synchronize(&qv->napi);
+               ring->netdev = NULL;
+       }
+
+       clear_bit(accel->pool, adapter->fwd_bitmask);
        limit = find_last_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
        adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
-       ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
 
        /* go back to full RSS if we're done with our VMQs */
        if (adapter->ring_feature[RING_F_VMDQ].limit == 1) {
@@ -9823,11 +9823,11 @@ static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
 
        ixgbe_setup_tc(pdev, adapter->hw_tcs);
        netdev_dbg(pdev, "pool %i:%i queues %i:%i\n",
-                  fwd_adapter->pool, adapter->num_rx_pools,
-                  fwd_adapter->rx_base_queue,
-                  fwd_adapter->rx_base_queue +
+                  accel->pool, adapter->num_rx_pools,
+                  accel->rx_base_queue,
+                  accel->rx_base_queue +
                   adapter->num_rx_queues_per_pool);
-       kfree(fwd_adapter);
+       kfree(accel);
 }
 
 #define IXGBE_MAX_MAC_HDR_LEN          127