On Tue, Dec 09, 2014 at 05:01:35PM +0100, Sebastian Andrzej Siewior wrote: > On 12/09/2014 04:24 PM, 'Greg Kroah-Hartman' wrote: > > On Mon, Dec 08, 2014 at 09:44:05AM +0000, David Laight wrote: > >> From: Greg Kroah-Hartman > >>> On Fri, Dec 05, 2014 at 09:03:57PM +0100, Sebastian Andrzej Siewior wrote: > >>>> Consider the following scenario: > >>>> - plugin a webcam > >>>> - play the stream via gst-launch-0.10 v4l2src device=/dev/video0 > >>>> - remove the USB-HCD during playback via "rmmod $HCD" > >>>> > >>>> and now wait for the crash > >>> > >>> Which you deserve, why did you ever remove a kernel module? That's racy > >>> and _never_ recommended, which is why it never happens automatically and > >>> only root can do it. > >> > >> Really drivers and subsystems should have the required locking (etc) to > >> ensure that kernel modules can either be unloaded, or that the unload > >> request itself fails if the device is busy. > >> > >> It shouldn't be considered a 'shoot self in foot' operation. > >> OTOH there are likely to be bugs. > > > > This is not always the case, sorry, removing a kernel module is a known > > racy condition, and sometimes adding all of the locking required to try > > to make it "safe" just isn't worth it overall, as this is something that > > _only_ a developer does. > > I wasn't are of that. rmmod does not mention this. Kconfig does not > mention this and suggest y as default (for MODULE_UNLOAD) . rmmod -f > likely causes problems but this is not the case here. If you want to > avoid rmmod why not mark a driver that it is not safe to remove it? And > why not make it work? Because sometimes fixes to make rmmod work "properly" entail slowing down the whole "normal" path. There is inherit problems with the core of the module unload code for all modules that are known, and are not going to be fixed because this isn't something that really matters. > You can unbind the HCD driver from the PCI-device via sysfs and this is > not something not only a developer does. This "unbind" calls the remove > function of the driver and the only difference between unbind and rmmod > is that the module remains inserted (but this is no news for you). If unbind causes a problem, it's the same problem that could happen if the hardware is hot-unplugged (like on a PCI card.) Stuff like that _does_ need to be fixed, and if your test shows this is a problem, I am all for fixing that. thanks, greg k-h -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
