RDMA/cxgb3: Mark RDMA device with CXIO_ERROR_FATAL when removing
authorSteve Wise <swise@opengridcomputing.com>
Mon, 22 Feb 2010 22:07:22 +0000 (22:07 +0000)
committerRoland Dreier <rolandd@cisco.com>
Wed, 24 Feb 2010 18:40:30 +0000 (10:40 -0800)
If cxgb3 calls the iw_cxgb3 t3cclient remove function due to a device
removal event, then the iwch device must be marked with CXIO_ERROR_FATAL
since the device below us is going away.  Otherwise, we can get stuck in
a deadlock as RDMA ULPs try and deallocate objects (like MRs, QPs, etc).
So always mark the device with CXIO_ERROR_FATAL when removing.

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

index d992543890ee120f4c61b5237245e7086eefd722..ee1d8b4d45410cbf0f358e1a152f58edfb26a5dc 100644 (file)
@@ -188,6 +188,7 @@ static void close_rnic_dev(struct t3cdev *tdev)
        mutex_lock(&dev_mutex);
        list_for_each_entry_safe(dev, tmp, &dev_list, entry) {
                if (dev->rdev.t3cdev_p == tdev) {
+                       dev->rdev.flags = CXIO_ERROR_FATAL;
                        cancel_delayed_work_sync(&dev->db_drop_task);
                        list_del(&dev->entry);
                        iwch_unregister_device(dev);