Re: [PATCH 2/5] USB: Make it possible to "subclass" usb_device_driver

[Bastien Nocera <hadess@hadess.net>]

More replies inline (which I always miss)

On Wed, 2019-10-09 at 10:34 -0400, Alan Stern wrote:
> On Wed, 9 Oct 2019, Bastien Nocera wrote:
> 
> > The kernel currenly has only 2 usb_device_drivers, one generic one,
> one
> > that completely replaces the generic one to make USB devices usable
> over
> > a network.
> 
> Presumably your first driver is in generic.c.  Where is the second
> one?
> 
> > Use the newly exported generic driver functions when a driver
> declares
> > to want them run, in addition to its own code. This makes it
> possible to
> > write drivers that extend the generic USB driver.
> > 
> > Signed-off-by: Bastien Nocera <hadess@hadess.net>
> 
> This has a few problems.  The biggest one is that the device core
> does 
> not guarantee any order of driver probing.  If generic.c is probed 
> first, the subclass driver will never get probed -- which is a
> pretty 
> fatal flaw.
> 
> > ---
> >  drivers/usb/core/driver.c | 36 ++++++++++++++++++++++++++++++-----
> -
> >  include/linux/usb.h       |  1 +
> >  2 files changed, 31 insertions(+), 6 deletions(-)
> > 
> > diff --git a/drivers/usb/core/driver.c b/drivers/usb/core/driver.c
> > index 2b27d232d7a7..863e380a272b 100644
> > --- a/drivers/usb/core/driver.c
> > +++ b/drivers/usb/core/driver.c
> > @@ -261,10 +261,17 @@ static int usb_probe_device(struct device
> *dev)
> >        */
> >       if (!udriver->supports_autosuspend)
> >               error = usb_autoresume_device(udev);
> > +     if (error)
> > +             return error;
> >  
> > -     if (!error)
> > -             error = udriver->probe(udev);
> > -     return error;
> > +     if (udriver->generic_init)
> > +             error = usb_generic_driver_probe(udev);
> > +     if (error)
> > +             return error;
> > +
> > +     if (udriver->probe)
> > +             return udriver->probe(udev);
> > +     return 0;
> >  }
> >  
> >  /* called from driver core with dev locked */
> > @@ -273,7 +280,10 @@ static int usb_unbind_device(struct device
> *dev)
> >       struct usb_device *udev = to_usb_device(dev);
> >       struct usb_device_driver *udriver = to_usb_device_driver(dev-
> >driver);
> >  
> > -     udriver->disconnect(udev);
> > +     if (udriver->generic_init)
> > +             usb_generic_driver_disconnect(udev);
> > +     if (udriver->disconnect)
> > +             udriver->disconnect(udev);
> 
> The order is wrong.  The disconnects should always be done in
> reverse 
> order of probing.  This is true whenever you have a destructor for a 
> subclass; the subclasses destructor runs before the superclass's 
> destructor.

Fixed. Fixed in the suspend function as well.

> >       if (!udriver->supports_autosuspend)
> >               usb_autosuspend_device(udev);
> >       return 0;
> > @@ -886,6 +896,14 @@ int usb_register_device_driver(struct
> usb_device_driver *new_udriver,
> >       if (usb_disabled())
> >               return -ENODEV;
> >  
> > +     if (new_udriver->probe == NULL &&
> > +         !new_udriver->generic_init) {
> 
> There's no point adding this extra test.  Even subclass drivers
> should 
> have a probe function.

Removed.

> > +             printk(KERN_ERR "%s: error %d registering device "
> > +                    "        driver %s, no probe() function\n",
> 
> Don't split character strings.  They are an exception to the 80-
> column 
> limit.

I was using the error message just below in the function as an example.
A bad one apparently. This is gone in any case.

> > +                    usbcore_name, retval, new_udriver->name);
> > +             return -EINVAL;
> > +     }
> > +
> >       new_udriver->drvwrap.for_devices = 1;
> >       new_udriver->drvwrap.driver.name = new_udriver->name;
> >       new_udriver->drvwrap.driver.bus = &usb_bus_type;
> > @@ -1149,7 +1167,10 @@ static int usb_suspend_device(struct
> usb_device *udev, pm_message_t msg)
> >               udev->do_remote_wakeup = 0;
> >               udriver = &usb_generic_driver;
> >       }
> > -     status = udriver->suspend(udev, msg);
> > +     if (udriver->generic_init)
> > +             status = usb_generic_driver_suspend (udev, msg);
> > +     if (status == 0 && udriver->suspend)
> > +             status = udriver->suspend(udev, msg);
> 
> Again, the order is wrong.  Suspend the subclass driver first.

Done, as mentioned above.

> >   done:
> >       dev_vdbg(&udev->dev, "%s: status %d\n", __func__, status);
> > @@ -1181,7 +1202,10 @@ static int usb_resume_device(struct
> usb_device *udev, pm_message_t msg)
> >               udev->reset_resume = 1;
> >  
> >       udriver = to_usb_device_driver(udev->dev.driver);
> > -     status = udriver->resume(udev, msg);
> > +     if (udriver->generic_init)
> > +             status = usb_generic_driver_resume (udev, msg);
> > +     if (status == 0 && udriver->resume)
> > +             status = udriver->resume(udev, msg);
> >  
> >   done:
> >       dev_vdbg(&udev->dev, "%s: status %d\n", __func__, status);
> > diff --git a/include/linux/usb.h b/include/linux/usb.h
> > index e656e7b4b1e4..fb9ad3511e55 100644
> > --- a/include/linux/usb.h
> > +++ b/include/linux/usb.h
> > @@ -1242,6 +1242,7 @@ struct usb_device_driver {
> >       const struct attribute_group **dev_groups;
> >       struct usbdrv_wrap drvwrap;
> >       unsigned int supports_autosuspend:1;
> > +     unsigned int generic_init:1;
> 
> How about using a name that actually says something about the
> driver?  
> Such as generic_subclass?  Or subclass_of_generic?
> 
> "init" has nothing to do with anything.

generic_subclass it will be. I've also documented it in the header.

> >  };
> >  #define      to_usb_device_driver(d) container_of(d, struct
> usb_device_driver, \
> >               drvwrap.driver)
> 
> Alan Stern
>