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>

diff --git a/drivers/net/ibm_newemac/core.c b/drivers/net/ibm_newemac/core.c
index 0de3aa2a2e44172d99598556ee7d5b2e9c721997..eb0718b441bbb01523739a7093b0f7fcd87d769e 100644 (file)
--- 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 4011803117caae85be457cc5b6774b351c097fb1..a010b2463fd93ba3a285a40063941f03399472dc 100644 (file)
--- 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;
 };