RDMA/cxgb3: Fix regression caused by class_device -> device conversion
authorSteve Wise <swise@opengridcomputing.com>
Tue, 8 Jul 2008 21:40:05 +0000 (14:40 -0700)
committerRoland Dreier <rolandd@cisco.com>
Tue, 8 Jul 2008 21:40:05 +0000 (14:40 -0700)
The change to iwch_provider.c in commit f4e91eb4 ("IB: convert struct
class_device to struct device") undid the fix done in commit 7f049f2f
("RDMA/cxgb3: Hold rtnl_lock() around ethtool get_drvinfo call").  It
removed the calls to rtnl_lock() that serialized the iw_cxgb3 ethtool
ops calls into the cxgb3 driver.  This locking is needed to avoid
messing up the internal state of the cxgb3 driver.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
drivers/infiniband/hw/cxgb3/iwch_provider.c

index 8934178..95f82cf 100644 (file)
@@ -1096,7 +1096,9 @@ static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, ch
        struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
 
        PDBG("%s dev 0x%p\n", __func__, dev);
+       rtnl_lock();
        lldev->ethtool_ops->get_drvinfo(lldev, &info);
+       rtnl_unlock();
        return sprintf(buf, "%s\n", info.fw_version);
 }
 
@@ -1109,7 +1111,9 @@ static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
        struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
 
        PDBG("%s dev 0x%p\n", __func__, dev);
+       rtnl_lock();
        lldev->ethtool_ops->get_drvinfo(lldev, &info);
+       rtnl_unlock();
        return sprintf(buf, "%s\n", info.driver);
 }