@@ -212,5 +212,6 @@ void snd_hdac_bus_remove_device(struct hdac_bus *bus,
bus->caddr_tbl[codec->addr] = NULL;
clear_bit(codec->addr, &bus->codec_powered);
bus->num_codecs--;
+ flush_work(&bus->unsol_work);
}
EXPORT_SYMBOL_GPL(snd_hdac_bus_remove_device);
@@ -159,6 +159,7 @@ void snd_hdac_device_unregister(struct hdac_device *codec)
if (device_is_registered(&codec->dev)) {
hda_widget_sysfs_exit(codec);
device_del(&codec->dev);
+ snd_hdac_bus_remove_device(codec->bus, codec);
}
}
EXPORT_SYMBOL_GPL(snd_hdac_device_unregister);
When the codec device is unregistered / freed, it may release the resource while being used in an unsolicited event like the jack detection work. This leads to use-after-free. The fix here is to unregister the device at first, i.e. removing the codec from the list, then flushing the pending works to assure that all unsol events are gone. After this point, we're free from accessing the codec via unsol events, thus can release the resources gracefully. The issue was spotted originally by Intel CI, but it couldn't be reproduced reliably by its nature. So let's hope this fix really addresses the whole issues. Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=196045 Reported-by: Martin Peres <martin.peres@free.fr> Signed-off-by: Takashi Iwai <tiwai@suse.de> --- sound/hda/hdac_bus.c | 1 + sound/hda/hdac_device.c | 1 + 2 files changed, 2 insertions(+)