On Fri, May 29, 2020 at 04:31:26PM +0800, Hillf Danton wrote: > Hold another grab to dev to prevent it from going home before work gets > done with it. > > +++ b/drivers/infiniband/core/uverbs_main.c > @@ -1152,6 +1152,8 @@ static int ib_uverbs_add_one(struct ib_d > device->ops.mmap ? &uverbs_mmap_fops : &uverbs_fops); > uverbs_dev->cdev.owner = THIS_MODULE; > > + /* pair with put_device() in ib_uverbs_remove_one() */ > + get_device(&uverbs_dev->dev); > ret = cdev_device_add(&uverbs_dev->cdev, &uverbs_dev->dev); > if (ret) > goto err_uapi; Doesn't look right, the put_device() in uverbs_remove_one() pairs with the device_initialize() in this function. The only thing I can think of is we called remove_once twice somehow or had an extra put on some error path. But I couldn't find any flow that would do either of those things Jason
