RDMA/cxgb3: Wait at least one schedule cycle during device removal
authorSteve Wise <swise@opengridcomputing.com>
Wed, 3 Mar 2010 15:06:34 +0000 (15:06 +0000)
committerRoland Dreier <rolandd@cisco.com>
Thu, 11 Mar 2010 22:00:35 +0000 (14:00 -0800)
During a hot-plug LLD removal event or an EEH error event, iw_cxgb3
must ensure that any/all threads that might be in a cxgb3 exported
function must return from the function before iw_cxgb3 returns from
its event processing.  Do this by calling synchronize_net().

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

index ee1d8b4..63f975f 100644 (file)
@@ -189,6 +189,7 @@ static void close_rnic_dev(struct t3cdev *tdev)
        list_for_each_entry_safe(dev, tmp, &dev_list, entry) {
                if (dev->rdev.t3cdev_p == tdev) {
                        dev->rdev.flags = CXIO_ERROR_FATAL;
+                       synchronize_net();
                        cancel_delayed_work_sync(&dev->db_drop_task);
                        list_del(&dev->entry);
                        iwch_unregister_device(dev);
@@ -217,6 +218,7 @@ static void iwch_event_handler(struct t3cdev *tdev, u32 evt, u32 port_id)
        switch (evt) {
        case OFFLOAD_STATUS_DOWN: {
                rdev->flags = CXIO_ERROR_FATAL;
+               synchronize_net();
                event.event  = IB_EVENT_DEVICE_FATAL;
                dispatch = 1;
                break;