On Mon, Jun 12, 2006 at 03:17:35PM -0400, Alan Stern wrote:
> Greg:
>
> This patch (as714) makes the USB device (as opposed to interface) driver
> usb_generic be treated just like all other drivers, subject to matching,
> probing, and unbinding. For instance, some of the normal boilerplate
> stuff that goes along with discovery of a new USB device (registration in
> usbfs, creation of the sysfs attribute files, and selection of an initial
> configuration) is moved into the probe routine, where any normal driver
> would put it.
>
> This change isn't particularly important for now -- other than in an
> organizational sense -- but it opens the door to having alternate USB
> device drivers. The examples that spring to mind are when a host exports
> a USB device over the network or to a virtual guest operating system. In
> each case, the device shouldn't show up as a normal USB device on the host
> but rather on the client/guest. This could be accomplished by binding the
> host's device to a different driver. If this was done for a root hub, an
> entire USB device tree could automatically be exported!
Nice idea. I'm trying to remember why I didn't do this all those years
ago and can't recall why. Thanks for fixing this up.
One comment thought:
> @@ -516,14 +756,14 @@ static int usb_uevent(struct device *dev
> /* driver is often null here; dev_dbg() would oops */
> pr_debug ("usb %s: uevent\n", dev->bus_id);
>
> - /* Must check driver_data here, as on remove driver is always NULL */
> - if ((dev->driver == &usb_generic_driver) ||
> - (dev->driver_data == &usb_generic_driver_data))
> - return 0;
> -
> - intf = to_usb_interface(dev);
> - usb_dev = interface_to_usbdev (intf);
> - alt = intf->cur_altsetting;
> + if (is_usb_device(dev)) {
> + usb_dev = to_usb_device(dev);
> + alt = NULL;
> + } else {
> + intf = to_usb_interface(dev);
> + usb_dev = interface_to_usbdev (intf);
> + alt = intf->cur_altsetting;
> + }
>
> if (usb_dev->devnum < 0) {
> pr_debug ("usb %s: already deleted?\n", dev->bus_id);
> @@ -566,15 +806,22 @@ static int usb_uevent(struct device *dev
> usb_dev->descriptor.bDeviceProtocol))
> return -ENOMEM;
>
> - if (add_uevent_var(envp, num_envp, &i,
> + if (is_usb_device(dev)) {
> +
> + /* FIXME: What is an appropriate MODALIAS line for a
> + * USB device (not interface)? */
There isn't one.
In fact, you shouldn't be adding _any_ usb stuff for uevents for usb
devices, we only care about interfaces. Why did you change this?
> Index: usb-2.6/drivers/usb/core/usb.h
> ===================================================================
> --- usb-2.6.orig/drivers/usb/core/usb.h
> +++ usb-2.6/drivers/usb/core/usb.h
> @@ -31,8 +31,16 @@ extern int usb_suspend_device(struct usb
> extern int usb_resume_device(struct usb_device *dev);
>
> extern struct bus_type usb_bus_type;
> -extern struct device_driver usb_generic_driver;
> -extern int usb_generic_driver_data;
> +extern struct usb_driver usb_generic_driver;
> +
> +/* Here's how we tell apart devices and interfaces. Luckily there's
> + * no such thing as a platform USB device, so we can steal the use
> + * of the platform field. */
> +
> +static inline int is_usb_device(struct device *dev)
> +{
> + return dev->platform_data == &usb_generic_driver;
> +}
Ah, nice hack, I like it :)
> --- usb-2.6.orig/include/linux/usb.h
> +++ usb-2.6/include/linux/usb.h
> @@ -581,6 +581,8 @@ extern ssize_t usb_store_new_id(struct u
> * @driver: the driver model core driver structure.
> * @no_dynamic_id: if set to 1, the USB core will not allow dynamic ids to be
> * added to this driver by preventing the sysfs file from being created.
> + * @for_devices: indicates that the driver manages USB devices and not
> + * USB interfaces (very uncommon).
> *
> * USB drivers must provide a name, probe() and disconnect() methods,
> * and an id_table. Other driver fields are optional.
> @@ -619,6 +621,8 @@ struct usb_driver {
> struct usb_dynids dynids;
> struct device_driver driver;
> unsigned int no_dynamic_id:1;
> +
> + unsigned int for_devices:1;
Nice, this is why I didn't do this before, I didn't think of doing it
this way. It makes more sense now.
thanks,
greg k-h
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
