Re: [PATCH] usb: core: Save the config when a device is deauthorized+authorized

[Doug Anderson]

Hi,

On Fri, Dec 1, 2023 at 7:59 AM Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Thu, Nov 30, 2023 at 03:43:47PM -0800, Douglas Anderson wrote:
> > Right now, when a USB device is deauthorized (by writing 0 to the
> > "authorized" field in sysfs) and then reauthorized (by writing a 1) it
> > loses any configuration it might have had. This is because
> > usb_deauthorize_device() calls:
> >   usb_set_configuration(usb_dev, -1);
> > ...and then usb_authorize_device() calls:
> >   usb_choose_configuration(udev);
> > ...to choose the "best" configuration.
> >
> > This generally works OK and it looks like the above design was chosen
> > on purpose. In commit 93993a0a3e52 ("usb: introduce
> > usb_authorize/deauthorize()") we can see some discussion about keeping
> > the old config but it was decided not to bother since we can't save it
> > for wireless USB anyway. It can be noted that as of commit
> > 1e4c574225cc ("USB: Remove remnants of Wireless USB and UWB") wireless
> > USB is removed anyway, so there's really not a good reason not to keep
> > the old config.
> >
> > Unfortunately, throwing away the old config breaks when something has
> > decided to choose a config other than the normal "best" config.
> > Specifically, it can be noted that as of commit ec51fbd1b8a2 ("r8152:
> > add USB device driver for config selection") that the r8152 driver
> > subclasses the generic USB driver and selects a config other than the
> > one that would have been selected by usb_choose_configuration(). This
> > logic isn't re-run after a deauthorize + authorize and results in the
> > r8152 driver not being re-bound.
> >
> > Let's change things to save the old config when we deauthorize and
> > then restore it when we re-authorize. We'll disable this logic for
> > wireless USB where we re-fetch the descriptor after authorization.
>
> Would it be better to make the r8152 driver override
> usb_choose_configuration()?  This is the sort of thing that subclassing
> is intended for.

Yes, this is a nice solution. Posted.

https://lore.kernel.org/r/20231201183113.343256-1-dianders@xxxxxxxxxxxx