Hi Marcel,
On Wed, Apr 11, 2012 at 12:48:37PM +0200, Marcel Holtmann wrote:
> > > > > > @@ -1755,16 +1756,17 @@ int hci_register_dev(struct hci_dev *hdev)
> > > > > >
> > > > > > write_lock(&hci_dev_list_lock);
> > > > > >
> > > > > > - /* Find first available device id */
> > > > > > - list_for_each(p, &hci_dev_list) {
> > > > > > - if (list_entry(p, struct hci_dev, list)->id != id)
> > > > > > - break;
> > > > > > - head = p; id++;
> > > > > > - }
> > > > > I am now a little bit confused. Is it not enough to just replace head
> > > > > with &hci_dev_list to get this fixed? Or why is this failing in the
> > > > > first place actually.
> > > >
> > > > You can see actual code above. If you have hci0 and hci1 adding third AMP
> > > > will fail since it just checks 0!=1 => break and trying to create hci1
> > > > again.
> > >
> > > I see a problem when you having only AMPs (no BR/EDR controller). Then
> > > hci0 will be skipped and keep trying to create hci1 over and over again.
> > >
> > > However in the case we have BR/EDR controller on hci0 and hci1 as AMP,
> > > then this should just work.
> > >
> > > So we need to fix the case where we have no BR/EDR controller on hci0
> > > and trying to fix something else. Or did I get confused?
> >
> > I think you got confused.
> >
> > We have hci0 (BREDR) and hci1 (AMP)
> >
> > so list_for_each would give us first entry with id=0; Adding third (AMP)
> > would compare 1 (as given by id = (hdev->dev_type == HCI_BREDR) ? 0 : 1)
> > and 0 (id for BREDR) => break from the loop and tries to use id=1 again
> > for third controller.
> >
> > Actually I do not know how the current code works at all, it is really
> > broken.
>
> so it is broken on two levels actually. The case where we have no
> BR/EDRI controller as hci0 is also broken.
>
> Can you try to come up with a different solution that is not using a bit
> mask. That one might overflow actually. Something like this:
>
> min_id = (hdev->dev_type == HCI_BREDR) ? 0 : 1)
>
> for each (...) {
> if (id >= min_id && hdev->id != id)
> break;
>
> ...
> }
I think that would not work. If we have for example hci0,hci1,hci2 and
adding another one would choose hci2 again since (2 > 1) && (hdev->id=1 !=
2)
otherwise I think this is a bit complex when compared to bitmask which is
used in netdev in dev_alloc_name but they use a page:
/* Use one page as a bit array of possible slots */
inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
BTW: what is max number of controllers we support?
Best regards
Andrei Emeltchenko
--
To unsubscribe from this list: send the line "unsubscribe linux-bluetooth" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
