On Tue, Apr 26, 2005 at 11:57:34AM -0400, Alan Stern wrote:
> On Tue, 26 Apr 2005, Roman Kagan wrote:
>
> > On Mon, Apr 25, 2005 at 04:01:07PM -0400, Alan Stern wrote:
> > > I think usb_driver_claim_interface is correct as it stands. It was a
> > > mistake to leave out from usb_driver_release_interface originally the
> > > line
> > > setting iface->condition to USB_INTERFACE_UNBINDING.
> >
> > But it would be asymmetric then.
>
> In what sense? You mean because we don't also protect against a driver
> calling usb_driver_claim_interface from within that interface's probe?
> We don't need to check for that, do we?
Well, I'm not sure, but we do it now (from 2.6.12-rc2):
int usb_driver_claim_interface(struct usb_driver *driver,
struct usb_interface *iface, void* priv)
{
struct device *dev = &iface->dev;
if (dev->driver)
return -EBUSY;
dev->driver = &driver->driver;
usb_set_intfdata(iface, priv);
iface->condition = USB_INTERFACE_BOUND;
/* if interface was already added, bind now; else let
* the future device_add() bind it, bypassing probe()
*/
if (!list_empty (&dev->bus_list))
device_bind_driver(dev);
return 0;
}
>
> > FWIW my personal preference would be
> > to get rid of usb_interface.condition altogether, and always use the
> > driver base facilities to deduce the current interface bound state.
> > Otherwise maintaining a private field in sync with what the driver base
> > thinks is always error prone.
>
> There are other reasons for keeping usb_interface.condition. For one
> thing, the driver-model data doesn't tell when a bind or unbind is in
> progress. That information is used elsewhere in usbcore.
AFAICT this "elsewhere" is only usb_lock_device_for_reset(). And it
looks like all its users (scsi reset in storage/scsiglue.c and
image/microtek.c) don't really care if it is transitioning or already in
an established bound state.
> Unfortunately there are several spots where a device can be bound to a
> driver, and the code generally doesn't check for errors. This makes it
> difficult to check klist_node_attached at each spot. The best solution
> would be always to have device_release_driver wait for the klist_node's
> completion before releasing the semaphore. But there's no simple way to
> do that, given how driver_detach works.
Agreed.
Cheers,
Roman.
-------------------------------------------------------
This SF.Net email is sponsored by: NEC IT Guy Games.
Get your fingers limbered up and give it your best shot. 4 great events, 4
opportunities to win big! Highest score wins.NEC IT Guy Games. Play to
win an NEC 61 plasma display. Visit http://www.necitguy.com/?r=20
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
