2017-12-19 19:11 GMT+08:00 Cornelia Huck <cohuck@xxxxxxxxxx>: > On Sun, 17 Dec 2017 21:48:05 +0800 > weiping zhang <zwp10758@xxxxxxxxx> wrote: > >> As mentioned at drivers/base/core.c: >> /* >> * NOTE: _Never_ directly free @dev after calling this function, even >> * if it returned an error! Always use put_device() to give up the >> * reference initialized in this function instead. >> */ >> virtio_register_device may fail before/after call device_register, the >> caller should do a proper cleanup. Caller cann't use kfree directly, >> if virtio_register_device has already called device_register. Caller >> cann't use put_device directly, if virtio_register_device has not yet >> call device_register, because kobject_put may give a warning cause >> dev->kobj has not been initialized. > > This comment makes me inclined to think that we should also rethink > register_virtio_device(). On failure, we cannot do kfree() due to > driver core interaction; but we cannot do a put_device() either, since > the refcount may not yet have been initialized -- unless we check the > device status, which triggers I/O (at least on s390). > > We really want to do the same cleanup on error in every case. What > about splitting device_register() into device_initialize() and > device_add()? If we move device_initialize() before getting an index, > we should be fine with doing put_device() on error in every case. > Good idea (^_^), I''ll apply it at V4. _______________________________________________ Virtualization mailing list Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linuxfoundation.org/mailman/listinfo/virtualization
