Re: killing stalled ACL connection

[Luiz Augusto von Dentz <luiz.dentz@gmail.com>]

Hi,

On Fri, Apr 16, 2010 at 5:28 PM, Pavan Savoy <pavan_savoy@yahoo.co.in> wrote:
>
> --- On Fri, 16/4/10, Luiz Augusto von Dentz <luiz.dentz@gmail.com> wrote:
>
>> From: Luiz Augusto von Dentz <luiz.dentz@gmail.com>
>> Subject: Re: killing stalled ACL connection
>> To: "Pavan Savoy" <pavan_savoy@yahoo.co.in>
>> Cc: linux-bluetooth@vger.kernel.org
>> Date: Friday, 16 April, 2010, 1:32 PM
>> Hi,
>>
>> On Thu, Apr 15, 2010 at 10:25 PM, Pavan Savoy <pavan_savoy@yahoo.co.in>
>> wrote:
>> > I am working on a platform which has a very aggressive
>> power management features.
>> > So consider a situation, where an a2dp connection
>> exists with a headset, and i pause the media over AVRCP, so
>> my phone now goes into suspend state (shutting off UART
>> clocks),
>> >
>> > and when I wake-up (resume) using the AVRCP
>> connection, via the play/pause button, I see the platform
>> waking up and media player receiving the input, However I
>> also see this,
>> >
>> > hci0  ACL tx timeout.
>> > killing stalled ACL connection.
>> >
>> > and media is not able to play on the headset anymore.
>> > Although the media player thinks it's being played.
>> >
>> > Now where should I look into, as to why a perfectly
>> nice ACL (l2cap, a2dp) connection was dropped during the
>> suspend ?
>>
>> I guess you mean the system is suspended, to be honest I
>> thought that
>> would mean shutting down the bluetooth chip too which
>> obviously will
>> drop all the connections, but that doesn't seems to be the
>> case here,
>> does it? Anyway if you shutdown the bus which you
>> communicate with the
>> chip it could eventually timeout, I guess in such situation
>> it is
>> better to enter sniff mode, use some kind of bus that can
>> wakeup/powerup automatically if there is data to transfer
>> or a
>> combination of both.
>
> UART does this kind of automatically, I mean there is no s/w calls that are made it go into suspend state. (there is a hardware module which does that - we just have to configure it)
>
> And yes, bluetooth chip is actually alive and active after suspend/resume.
> However it's only the ACL connections that are getting dropped.
>
> Is there some sort of timeout involved ? with these ACL connections ?
> because I guess even the LMP's time to leave messages would not be going to headset (if there is such a thing called TTL messages).
>
> I need some hint as to how I can debug this ...

You can start here (net/bluetooth/hci_core.c:1408):

	if (!test_bit(HCI_RAW, &hdev->flags)) {
		/* ACL tx timeout must be longer than maximum
		 * link supervision timeout (40.9 seconds) */
		if (!hdev->acl_cnt && time_after(jiffies, hdev->acl_last_tx + HZ * 45))
			hci_acl_tx_to(hdev);
	}

-- 
Luiz Augusto von Dentz
Computer Engineer