cxgb3: Don't call cxgb_vlan_mode until q locks are initialized
authorRoland Dreier <roland@purestorage.com>
Mon, 30 Apr 2012 07:15:47 +0000 (07:15 +0000)
committerDavid S. Miller <davem@davemloft.net>
Tue, 1 May 2012 02:08:44 +0000 (22:08 -0400)
The driver calls cxgb_vlan_mode() from init_one().  This calls into
synchronize_rx(), which locks all the q locks, but the q locks are not
initialized until cxgb_up() -> setup_sge_qsets().  So move the call to
cxgb_vlan_mode() into cxgb_up(), after the call to setup_sge_qsets().
We also move the body of these functions up higher to avoid having to
a forward declaration.

This was found because of the lockdep warning:

    INFO: trying to register non-static key.
    the code is fine but needs lockdep annotation.
    turning off the locking correctness validator.
    Pid: 323, comm: work_for_cpu Not tainted 3.4.0-rc5 #28
    Call Trace:
     [<ffffffff8106e767>] register_lock_class+0x108/0x2d0
     [<ffffffff8106ff42>] __lock_acquire+0xd3/0xd06
     [<ffffffff81070fd0>] lock_acquire+0xbf/0xfe
     [<ffffffff813862a6>] _raw_spin_lock_irq+0x36/0x45
     [<ffffffffa01e71aa>] cxgb_vlan_mode+0x96/0xcb [cxgb3]
     [<ffffffffa01f90eb>] init_one+0x8c4/0x980 [cxgb3]
     [<ffffffff811fcbf0>] local_pci_probe+0x3f/0x70
     [<ffffffff81042206>] do_work_for_cpu+0x10/0x22
     [<ffffffff810482de>] kthread+0xa1/0xa9
     [<ffffffff8138e234>] kernel_thread_helper+0x4/0x10

Contrary to what lockdep says, the code is not fine: we are locking an
uninitialized spinlock.

Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/chelsio/cxgb3/cxgb3_main.c

index 63bfdd10bd6d857b33cb52f4702919f1843fd094..abb6ce7c1b7ed2a67ee6b064377698df2efcb9c6 100644 (file)
@@ -1149,6 +1149,48 @@ release_tpsram:
        return ret;
 }
 
+/**
+ * t3_synchronize_rx - wait for current Rx processing on a port to complete
+ * @adap: the adapter
+ * @p: the port
+ *
+ * Ensures that current Rx processing on any of the queues associated with
+ * the given port completes before returning.  We do this by acquiring and
+ * releasing the locks of the response queues associated with the port.
+ */
+static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
+{
+       int i;
+
+       for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
+               struct sge_rspq *q = &adap->sge.qs[i].rspq;
+
+               spin_lock_irq(&q->lock);
+               spin_unlock_irq(&q->lock);
+       }
+}
+
+static void cxgb_vlan_mode(struct net_device *dev, netdev_features_t features)
+{
+       struct port_info *pi = netdev_priv(dev);
+       struct adapter *adapter = pi->adapter;
+
+       if (adapter->params.rev > 0) {
+               t3_set_vlan_accel(adapter, 1 << pi->port_id,
+                                 features & NETIF_F_HW_VLAN_RX);
+       } else {
+               /* single control for all ports */
+               unsigned int i, have_vlans = features & NETIF_F_HW_VLAN_RX;
+
+               for_each_port(adapter, i)
+                       have_vlans |=
+                               adapter->port[i]->features & NETIF_F_HW_VLAN_RX;
+
+               t3_set_vlan_accel(adapter, 1, have_vlans);
+       }
+       t3_synchronize_rx(adapter, pi);
+}
+
 /**
  *     cxgb_up - enable the adapter
  *     @adapter: adapter being enabled
@@ -1161,7 +1203,7 @@ release_tpsram:
  */
 static int cxgb_up(struct adapter *adap)
 {
-       int err;
+       int i, err;
 
        if (!(adap->flags & FULL_INIT_DONE)) {
                err = t3_check_fw_version(adap);
@@ -1198,6 +1240,9 @@ static int cxgb_up(struct adapter *adap)
                if (err)
                        goto out;
 
+               for_each_port(adap, i)
+                       cxgb_vlan_mode(adap->port[i], adap->port[i]->features);
+
                setup_rss(adap);
                if (!(adap->flags & NAPI_INIT))
                        init_napi(adap);
@@ -2508,48 +2553,6 @@ static int cxgb_set_mac_addr(struct net_device *dev, void *p)
        return 0;
 }
 
-/**
- * t3_synchronize_rx - wait for current Rx processing on a port to complete
- * @adap: the adapter
- * @p: the port
- *
- * Ensures that current Rx processing on any of the queues associated with
- * the given port completes before returning.  We do this by acquiring and
- * releasing the locks of the response queues associated with the port.
- */
-static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
-{
-       int i;
-
-       for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
-               struct sge_rspq *q = &adap->sge.qs[i].rspq;
-
-               spin_lock_irq(&q->lock);
-               spin_unlock_irq(&q->lock);
-       }
-}
-
-static void cxgb_vlan_mode(struct net_device *dev, netdev_features_t features)
-{
-       struct port_info *pi = netdev_priv(dev);
-       struct adapter *adapter = pi->adapter;
-
-       if (adapter->params.rev > 0) {
-               t3_set_vlan_accel(adapter, 1 << pi->port_id,
-                                 features & NETIF_F_HW_VLAN_RX);
-       } else {
-               /* single control for all ports */
-               unsigned int i, have_vlans = features & NETIF_F_HW_VLAN_RX;
-
-               for_each_port(adapter, i)
-                       have_vlans |=
-                               adapter->port[i]->features & NETIF_F_HW_VLAN_RX;
-
-               t3_set_vlan_accel(adapter, 1, have_vlans);
-       }
-       t3_synchronize_rx(adapter, pi);
-}
-
 static netdev_features_t cxgb_fix_features(struct net_device *dev,
        netdev_features_t features)
 {
@@ -3353,9 +3356,6 @@ static int __devinit init_one(struct pci_dev *pdev,
        err = sysfs_create_group(&adapter->port[0]->dev.kobj,
                                 &cxgb3_attr_group);
 
-       for_each_port(adapter, i)
-               cxgb_vlan_mode(adapter->port[i], adapter->port[i]->features);
-
        print_port_info(adapter, ai);
        return 0;