multiple interface test case
Duncan Sands
baldrick at free.fr
Wed Apr 13 08:05:32 EDT 2005
Hi Roman,
> > > That is weird then: who has removed _all_ interfaces from the driver's
> > > list? On the first disconnect(), the other two interfaces must still
> > > remain on the list. Otherwise what does driver_detach() iterate over?
> >
> > I think it's this one, in device_release_driver:
> >
> > klist_del(&dev->knode_driver);
> >
> > it's called just before doing the disconnect call
> >
> > if (drv->remove)
> > drv->remove(dev);
>
> Exactly, that's what I meant, but that means that device_release_driver
> has already been called for all three interfaces, but we were talking
> about the _first_ disconnect() which is called from the _first_
> device_release_driver().
well I agree that it is unclear exactly at which point the change is
made that causes the test in usb_driver_release_interface to fire.
However one thing is clear: it has happened before the first call to
your test module's disconnect, since every call to
usb_driver_release_interface skips the call to device_release_driver.
And I know perfectly well that device_release_driver has not been
called for each interface yet, because I print debugging info for
every call to it.
> > I suppose klist_del removes the driver from the list of drivers,
> > which means that this test in usb_driver_release_interface fires:
> >
> > klist_node_attached(&dev->knode_driver)
> >
> > driver_detach iterates over devices, but this is a test on the
> > driver.
>
> klist_del() does both (similar to list_del_init()): it removes the node
> from the list, and removes the reference to the list from the node. So
> there's nothing new in -mm compared to vanilla -rc2 (BTW I've fetched
> -mm3, there are no relevant changes wrt -mm2).
>
> Frankly, I can't explain the messages above... Can you please
> double-check it?
If you stick some printk's in device_release_driver and
usb_driver_release_interface then you will see it yourself. I'm still
trying to understand exactly what's going on.
Ciao,
D.
More information about the Usbatm
mailing list