Re: [Bluez-users] error on connect

[Marcel Holtmann <marcel@holtmann.org>]

Hi Marco,

> >>>>It looks like this too early packet now causes the (sensefull) 
> >>>>"hci_acldata_packet: hci1 ACL packet for unknown connection handle 6" 
> >>>>kernel error message. The dongle then waits for this acl package and thinks 
> >>>>this connection is in etablishing phase, so it is blocked.
> >>>
> >>>>From a quick look at the kernel code, it seems that an ACL packet for an
> >>>unknown connection handle may corrupt the HCI flow control.
> >>
> >>I think exactly this happens...
> > 
> > However even if I fix the HCI flow control problem, the packet is still
> > lost, because we don't know the connection handle at that time.
> 
> yes, thats what seems strainge: why is this ACL packet sent that early?
> that doesn't make sense for me (but I din't develop bluetooth)...
> In my opinion, it should be sent after the connection is etablished.

It is not only your opinion. The specification defines it this way. This
is simply a race condition inside the link manager.

> >>>>The question now is, who is responsible for holding this packed back? The 
> >>>>kernel? The dongle itself?
> >>>
> >>>The dongle is responsible for that and actually it is only allowed to
> >>>send data packets after the "Connect Complete" event. Otherwise we don't
> >>>know its handle and thus can't correctly assign it to a connection.
> >>
> >>but then it's strainge that these ACL packets are always sent before the 
> >>"connection" successfull packet. or do I right now miss something?
> > 
> > This must be a bug in the Broadcom link manager firmware. Seems like
> > BlueZ is "too fast" for these chips.
> >
> > If you wanna workaround this problem you must put potential ACL data
> > packets into a queue and then check this queue after you received the
> > connection handle.
> 
> that's pretty ugly. Maybe there's another solution for this. I think 
> that the problem isnt yet fully understood.

The problem is clear. The Broadcom link manager has a race condition
here. It should queue the package until they send the connect complete
event.

>  > You only saw that on the listing side, right?
> 
> what do you mean with this?
> on the sending side - if i remember right - the acl packet is alway sent 
> before the connection is successfull etablished. I'll check again...

Please do so, but there should be no problem. We only send any data to
the chip after we received the connect complete. All previous packages
are queued.

Regards

Marcel




-------------------------------------------------------
This SF.Net email is sponsored by: New Crystal Reports XI.
Version 11 adds new functionality designed to reduce time involved in
creating, integrating, and deploying reporting solutions. Free runtime info,
new features, or free trial, at: http://www.businessobjects.com/devxi/728
_______________________________________________
Bluez-users mailing list
Bluez-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-users