i40evf: fix panic during MTU change
authorMitch Williams <mitch.a.williams@intel.com>
Fri, 19 Jun 2015 15:56:30 +0000 (08:56 -0700)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Fri, 26 Jun 2015 09:51:31 +0000 (02:51 -0700)
Down was requesting queue disables, but then exited immediately
without waiting for the queues to actually disable.  This could
allow any function called after i40evf_down to run immediately,
including i40evf_up, and causes a memory leak.

Removing the whole reinit_locked function is the best way
to go about this, and allows for the driver to handle the
state changes by requesting reset from the periodic timer.

Also, add a couple WARN_ONs in slow path to help us recognize
if we re-introduce this issue or missed any cases.

Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ethernet/intel/i40evf/i40e_txrx.c
drivers/net/ethernet/intel/i40evf/i40evf.h
drivers/net/ethernet/intel/i40evf/i40evf_ethtool.c
drivers/net/ethernet/intel/i40evf/i40evf_main.c

index f54996f..395f32f 100644 (file)
@@ -484,6 +484,8 @@ int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
        if (!dev)
                return -ENOMEM;
 
+       /* warn if we are about to overwrite the pointer */
+       WARN_ON(tx_ring->tx_bi);
        bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
        tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
        if (!tx_ring->tx_bi)
@@ -644,6 +646,8 @@ int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
        struct device *dev = rx_ring->dev;
        int bi_size;
 
+       /* warn if we are about to overwrite the pointer */
+       WARN_ON(rx_ring->rx_bi);
        bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
        rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
        if (!rx_ring->rx_bi)
index 1b98c25..fea3b75 100644 (file)
@@ -264,7 +264,6 @@ extern const char i40evf_driver_version[];
 
 int i40evf_up(struct i40evf_adapter *adapter);
 void i40evf_down(struct i40evf_adapter *adapter);
-void i40evf_reinit_locked(struct i40evf_adapter *adapter);
 void i40evf_reset(struct i40evf_adapter *adapter);
 void i40evf_set_ethtool_ops(struct net_device *netdev);
 void i40evf_update_stats(struct i40evf_adapter *adapter);
index f4e7766..2b53c87 100644 (file)
@@ -267,8 +267,10 @@ static int i40evf_set_ringparam(struct net_device *netdev,
        adapter->tx_desc_count = new_tx_count;
        adapter->rx_desc_count = new_rx_count;
 
-       if (netif_running(netdev))
-               i40evf_reinit_locked(adapter);
+       if (netif_running(netdev)) {
+               adapter->flags |= I40EVF_FLAG_RESET_NEEDED;
+               schedule_work(&adapter->reset_task);
+       }
 
        return 0;
 }
index 7c53aca..5c73374 100644 (file)
@@ -170,7 +170,8 @@ static void i40evf_tx_timeout(struct net_device *netdev)
        struct i40evf_adapter *adapter = netdev_priv(netdev);
 
        adapter->tx_timeout_count++;
-       if (!(adapter->flags & I40EVF_FLAG_RESET_PENDING)) {
+       if (!(adapter->flags & (I40EVF_FLAG_RESET_PENDING |
+                               I40EVF_FLAG_RESET_NEEDED))) {
                adapter->flags |= I40EVF_FLAG_RESET_NEEDED;
                schedule_work(&adapter->reset_task);
        }
@@ -1470,8 +1471,8 @@ static void i40evf_configure_rss(struct i40evf_adapter *adapter)
        i40e_flush(hw);
 }
 
-#define I40EVF_RESET_WAIT_MS 100
-#define I40EVF_RESET_WAIT_COUNT 200
+#define I40EVF_RESET_WAIT_MS 10
+#define I40EVF_RESET_WAIT_COUNT 500
 /**
  * i40evf_reset_task - Call-back task to handle hardware reset
  * @work: pointer to work_struct
@@ -1495,10 +1496,17 @@ static void i40evf_reset_task(struct work_struct *work)
                                &adapter->crit_section))
                usleep_range(500, 1000);
 
+       i40evf_misc_irq_disable(adapter);
        if (adapter->flags & I40EVF_FLAG_RESET_NEEDED) {
-               dev_info(&adapter->pdev->dev, "Requesting reset from PF\n");
+               adapter->flags &= ~I40EVF_FLAG_RESET_NEEDED;
+               /* Restart the AQ here. If we have been reset but didn't
+                * detect it, or if the PF had to reinit, our AQ will be hosed.
+                */
+               i40evf_shutdown_adminq(hw);
+               i40evf_init_adminq(hw);
                i40evf_request_reset(adapter);
        }
+       adapter->flags |= I40EVF_FLAG_RESET_PENDING;
 
        /* poll until we see the reset actually happen */
        for (i = 0; i < I40EVF_RESET_WAIT_COUNT; i++) {
@@ -1507,10 +1515,10 @@ static void i40evf_reset_task(struct work_struct *work)
                if ((rstat_val != I40E_VFR_VFACTIVE) &&
                    (rstat_val != I40E_VFR_COMPLETED))
                        break;
-               msleep(I40EVF_RESET_WAIT_MS);
+               usleep_range(500, 1000);
        }
        if (i == I40EVF_RESET_WAIT_COUNT) {
-               adapter->flags &= ~I40EVF_FLAG_RESET_PENDING;
+               dev_info(&adapter->pdev->dev, "Never saw reset\n");
                goto continue_reset; /* act like the reset happened */
        }
 
@@ -1518,11 +1526,12 @@ static void i40evf_reset_task(struct work_struct *work)
        for (i = 0; i < I40EVF_RESET_WAIT_COUNT; i++) {
                rstat_val = rd32(hw, I40E_VFGEN_RSTAT) &
                            I40E_VFGEN_RSTAT_VFR_STATE_MASK;
-               if ((rstat_val == I40E_VFR_VFACTIVE) ||
-                   (rstat_val == I40E_VFR_COMPLETED))
+               if (rstat_val == I40E_VFR_VFACTIVE)
                        break;
                msleep(I40EVF_RESET_WAIT_MS);
        }
+       /* extra wait to make sure minimum wait is met */
+       msleep(I40EVF_RESET_WAIT_MS);
        if (i == I40EVF_RESET_WAIT_COUNT) {
                struct i40evf_mac_filter *f, *ftmp;
                struct i40evf_vlan_filter *fv, *fvtmp;
@@ -1534,11 +1543,10 @@ static void i40evf_reset_task(struct work_struct *work)
 
                if (netif_running(adapter->netdev)) {
                        set_bit(__I40E_DOWN, &adapter->vsi.state);
-                       i40evf_irq_disable(adapter);
-                       i40evf_napi_disable_all(adapter);
-                       netif_tx_disable(netdev);
-                       netif_tx_stop_all_queues(netdev);
                        netif_carrier_off(netdev);
+                       netif_tx_disable(netdev);
+                       i40evf_napi_disable_all(adapter);
+                       i40evf_irq_disable(adapter);
                        i40evf_free_traffic_irqs(adapter);
                        i40evf_free_all_tx_resources(adapter);
                        i40evf_free_all_rx_resources(adapter);
@@ -1550,6 +1558,7 @@ static void i40evf_reset_task(struct work_struct *work)
                        list_del(&f->list);
                        kfree(f);
                }
+
                list_for_each_entry_safe(fv, fvtmp, &adapter->vlan_filter_list,
                                         list) {
                        list_del(&fv->list);
@@ -1564,22 +1573,27 @@ static void i40evf_reset_task(struct work_struct *work)
                i40evf_shutdown_adminq(hw);
                adapter->netdev->flags &= ~IFF_UP;
                clear_bit(__I40EVF_IN_CRITICAL_TASK, &adapter->crit_section);
+               adapter->flags &= ~I40EVF_FLAG_RESET_PENDING;
+               dev_info(&adapter->pdev->dev, "Reset task did not complete, VF disabled\n");
                return; /* Do not attempt to reinit. It's dead, Jim. */
        }
 
 continue_reset:
-       adapter->flags &= ~I40EVF_FLAG_RESET_PENDING;
-
-       i40evf_irq_disable(adapter);
-
        if (netif_running(adapter->netdev)) {
-               i40evf_napi_disable_all(adapter);
-               netif_tx_disable(netdev);
-               netif_tx_stop_all_queues(netdev);
                netif_carrier_off(netdev);
+               netif_tx_stop_all_queues(netdev);
+               i40evf_napi_disable_all(adapter);
        }
+       i40evf_irq_disable(adapter);
 
        adapter->state = __I40EVF_RESETTING;
+       adapter->flags &= ~I40EVF_FLAG_RESET_PENDING;
+
+       /* free the Tx/Rx rings and descriptors, might be better to just
+        * re-use them sometime in the future
+        */
+       i40evf_free_all_rx_resources(adapter);
+       i40evf_free_all_tx_resources(adapter);
 
        /* kill and reinit the admin queue */
        if (i40evf_shutdown_adminq(hw))
@@ -1603,6 +1617,7 @@ continue_reset:
        adapter->aq_required = I40EVF_FLAG_AQ_ADD_MAC_FILTER;
        adapter->aq_required |= I40EVF_FLAG_AQ_ADD_VLAN_FILTER;
        clear_bit(__I40EVF_IN_CRITICAL_TASK, &adapter->crit_section);
+       i40evf_misc_irq_enable(adapter);
 
        mod_timer(&adapter->watchdog_timer, jiffies + 2);
 
@@ -1624,7 +1639,10 @@ continue_reset:
                        goto reset_err;
 
                i40evf_irq_enable(adapter, true);
+       } else {
+               adapter->state = __I40EVF_DOWN;
        }
+
        return;
 reset_err:
        dev_err(&adapter->pdev->dev, "failed to allocate resources during reinit\n");
@@ -1667,6 +1685,11 @@ static void i40evf_adminq_task(struct work_struct *work)
                        memset(event.msg_buf, 0, I40EVF_MAX_AQ_BUF_SIZE);
        } while (pending);
 
+       if ((adapter->flags &
+            (I40EVF_FLAG_RESET_PENDING | I40EVF_FLAG_RESET_NEEDED)) ||
+           adapter->state == __I40EVF_RESETTING)
+               goto freedom;
+
        /* check for error indications */
        val = rd32(hw, hw->aq.arq.len);
        oldval = val;
@@ -1702,6 +1725,7 @@ static void i40evf_adminq_task(struct work_struct *work)
        if (oldval != val)
                wr32(hw, hw->aq.asq.len, val);
 
+freedom:
        kfree(event.msg_buf);
 out:
        /* re-enable Admin queue interrupt cause */
@@ -1897,47 +1921,6 @@ static struct net_device_stats *i40evf_get_stats(struct net_device *netdev)
 }
 
 /**
- * i40evf_reinit_locked - Software reinit
- * @adapter: board private structure
- *
- * Reinititalizes the ring structures in response to a software configuration
- * change. Roughly the same as close followed by open, but skips releasing
- * and reallocating the interrupts.
- **/
-void i40evf_reinit_locked(struct i40evf_adapter *adapter)
-{
-       struct net_device *netdev = adapter->netdev;
-       int err;
-
-       WARN_ON(in_interrupt());
-
-       i40evf_down(adapter);
-
-       /* allocate transmit descriptors */
-       err = i40evf_setup_all_tx_resources(adapter);
-       if (err)
-               goto err_reinit;
-
-       /* allocate receive descriptors */
-       err = i40evf_setup_all_rx_resources(adapter);
-       if (err)
-               goto err_reinit;
-
-       i40evf_configure(adapter);
-
-       err = i40evf_up_complete(adapter);
-       if (err)
-               goto err_reinit;
-
-       i40evf_irq_enable(adapter, true);
-       return;
-
-err_reinit:
-       dev_err(&adapter->pdev->dev, "failed to allocate resources during reinit\n");
-       i40evf_close(netdev);
-}
-
-/**
  * i40evf_change_mtu - Change the Maximum Transfer Unit
  * @netdev: network interface device structure
  * @new_mtu: new value for maximum frame size
@@ -1952,9 +1935,10 @@ static int i40evf_change_mtu(struct net_device *netdev, int new_mtu)
        if ((new_mtu < 68) || (max_frame > I40E_MAX_RXBUFFER))
                return -EINVAL;
 
-       /* must set new MTU before calling down or up */
        netdev->mtu = new_mtu;
-       i40evf_reinit_locked(adapter);
+       adapter->flags |= I40EVF_FLAG_RESET_NEEDED;
+       schedule_work(&adapter->reset_task);
+
        return 0;
 }