ibm_newemac: Fix possible lockup on close

It's a bad idea to call flush_scheduled_work from within a
netdev->stop because the linkwatch will occasionally take the
rtnl lock from a workqueue context, and thus that can deadlock.

This reworks things a bit in that area to avoid the problem.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

2 files changed, 21 insertions, 11 deletions

diff --git a/drivers/net/ibm_newemac/core.c b/drivers/net/ibm_newemac/core.c
index 0de3aa2a2e44..eb0718b441bb 100644
--- a/drivers/net/ibm_newemac/core.c
+++ b/drivers/net/ibm_newemac/core.c
@@ -642,9 +642,11 @@ static void emac_reset_work(struct work_struct *work)
         DBG(dev, "reset_work" NL);
 
         mutex_lock(&dev->link_lock);
-        emac_netif_stop(dev);
-        emac_full_tx_reset(dev);
-        emac_netif_start(dev);
+        if (dev->opened) {
+                emac_netif_stop(dev);
+                emac_full_tx_reset(dev);
+                emac_netif_start(dev);
+        }
         mutex_unlock(&dev->link_lock);
 }
 
@@ -1063,10 +1065,9 @@ static int emac_open(struct net_device *ndev)
         dev->rx_sg_skb = NULL;
 
         mutex_lock(&dev->link_lock);
+        dev->opened = 1;
 
-        /* XXX Start PHY polling now. Shouldn't wr do like sungem instead and
-         * always poll the PHY even when the iface is down ? That would allow
-         * things like laptop-net to work. --BenH
+        /* Start PHY polling now.
          */
         if (dev->phy.address >= 0) {
                 int link_poll_interval;
@@ -1145,9 +1146,11 @@ static void emac_link_timer(struct work_struct *work)
         int link_poll_interval;
 
         mutex_lock(&dev->link_lock);
-
         DBG2(dev, "link timer" NL);
 
+        if (!dev->opened)
+                goto bail;
+
         if (dev->phy.def->ops->poll_link(&dev->phy)) {
                 if (!netif_carrier_ok(dev->ndev)) {
                         /* Get new link parameters */
@@ -1170,13 +1173,14 @@ static void emac_link_timer(struct work_struct *work)
                 link_poll_interval = PHY_POLL_LINK_OFF;
         }
         schedule_delayed_work(&dev->link_work, link_poll_interval);
-
+ bail:
         mutex_unlock(&dev->link_lock);
 }
 
 static void emac_force_link_update(struct emac_instance *dev)
 {
         netif_carrier_off(dev->ndev);
+        smp_rmb();
         if (dev->link_polling) {
                 cancel_rearming_delayed_work(&dev->link_work);
                 if (dev->link_polling)
@@ -1191,11 +1195,14 @@ static int emac_close(struct net_device *ndev)
 
         DBG(dev, "close" NL);
 
-        if (dev->phy.address >= 0)
+        if (dev->phy.address >= 0) {
+                dev->link_polling = 0;
                 cancel_rearming_delayed_work(&dev->link_work);
-
+        }
+        mutex_lock(&dev->link_lock);
         emac_netif_stop(dev);
-        flush_scheduled_work();
+        dev->opened = 0;
+        mutex_unlock(&dev->link_lock);
 
         emac_rx_disable(dev);
         emac_tx_disable(dev);
@@ -2756,6 +2763,8 @@ static int __devexit emac_remove(struct of_device *ofdev)
 
         unregister_netdev(dev->ndev);
 
+        flush_scheduled_work();
+
         if (emac_has_feature(dev, EMAC_FTR_HAS_TAH))
                 tah_detach(dev->tah_dev, dev->tah_port);
         if (emac_has_feature(dev, EMAC_FTR_HAS_RGMII))
diff --git a/drivers/net/ibm_newemac/core.h b/drivers/net/ibm_newemac/core.h
index 4011803117ca..a010b2463fd9 100644
--- a/drivers/net/ibm_newemac/core.h
+++ b/drivers/net/ibm_newemac/core.h
@@ -258,6 +258,7 @@ struct emac_instance {
         int                             stop_timeout;   /* in us */
         int                             no_mcast;
         int                             mcast_pending;
+        int                             opened;
         struct work_struct              reset_work;
         spinlock_t                      lock;
 };