Return-Path: <cyrus@holtmann.org>
Date: Wed, 08 Jul 2009 13:47:23 +0900
From: Chan-Yeol Park <chanyeol.park@samsung.com>
Subject: RE: Does anybody know conn->power_save variable in hci_conn.c?
In-reply-to: <1247024290.3384.26.camel@localhost.localdomain>
To: 'Marcel Holtmann' <marcel@holtmann.org>
Cc: linux-bluetooth@vger.kernel.org
Message-id: <004601c9ff87$2f239a20$8d6ace60$%park@samsung.com>
MIME-version: 1.0
Content-type: text/plain; charset=us-ascii
References: <002e01c9fedc$f5fcb790$e1f626b0$%park@samsung.com>
 <1246988209.3384.10.camel@localhost.localdomain>
 <004001c9ff6c$7838f7c0$68aae740$%park@samsung.com>
 <1247024290.3384.26.camel@localhost.localdomain>
Sender: linux-bluetooth-owner@vger.kernel.org
List-ID: <linux-bluetooth.vger.kernel.org>

Hi Marcel,

>>> >> /*bluetooth kernel hci_conn.c */
>>> >> 382 void hci_conn_enter_active_mode(struct hci_conn *conn)
>>> >> 390
>>> >> 391         if (conn->mode != HCI_CM_SNIFF || !conn->power_save)
>>> >> 392                 goto timer;
>>> >> 393 
>>> >> 394         if (!test_and_set_bit(HCI_CONN_MODE_CHANGE_PEND,
>>> &conn->pend)) {
>>> >> 395                 struct hci_cp_exit_sniff_mode cp;
>>> >> 396                 cp.handle = __cpu_to_le16(conn->handle);
>>> >> 397                 hci_send_cmd(hdev, OGF_LINK_POLICY,
>>> >> 398                                 OCF_EXIT_SNIFF_MODE, sizeof(cp),
>>> >> 
>>> >> If possible, could you explain why this code checks !conn->power_save
>>> var.? 
>>> >> 
>>> >> Without this , if  conn->mode==HCI_CM_SNIFF , we simply exit sniff
>>> >> mode,"OCF_EXIT_SNIFF_MODE".
>>> >> 
>>> >> As far as I understood, whenever conn->mode is HCI_CM_SNIFF
>>> conn->power_save
>>> >> is 0.
>>> >> The reason is that  hci_mode_change_evt() set conn->power_save as "0"
>>> when
>>> >> conn->mode !=HCI_CM_ACTIVE.
>>> >> 
>>> >> Consequently we don't need to check that variable.
>>> >
>>> >we do need that variable, because otherwise devices like HID which do
>>> >active sniff mode management fall over if we always try to exit sniff
>>> >mode only for a few bytes.
>>> 
>>> In case of HID, active sniff mode could be no problem 
>>> because its data rate is light compared to other profile.
>>> 
>>> But other case such as Sony Ericsson HBH-DS970,980 A2DP profile , it
could
>>> be problem.
>>> As you may already know, 
>>> they request sniff mode while HFP so their connection is too late...
>>> It made AVDTP signal so slow..
>>> 
>>> Without conn->power_save variable check procedure, I found Sony Headset
>>> works well! because they exit sniff mode 
>>> Consequently, I think conn->power_save variable procedure should be
removed
>>> except only HID case.
>>> 
>>> If you think this is totally Sony headset problem, could you explain
that?
>>
>>it is a headset problem since it is too stupid to get out of sniff mode
>>even when it put itself into it and then tries to actively transmit
>>data. However we can ensure to leave sniff mode in that case, but that
>>needs to be a per socket option.
>>
>>As I said, if you have a HID device, you have to let the HID device
>>control the sniff mode since it knows best anyway.
>>

Thanks to you
I could conclude this problem is from the stupid headset.

But we can't ignore these kind of headset-.-;, 
As you know this headset is so popular and their family model has the same
problem unfortunately.
Moreover I check another phone[CSR or commercial stack] wakes up sniff mode.

So I have a plan to patch our kernel like below,

1. Simply remove conn->power_save check procedure:
This option is only used for HID profile(as far as I know) and currently we
don't support HID.
But I worry about this, because another profiles may need this option and
this patch apply to all the ACL connection...

Could you tell me the side effect of this patch except HID profile case? 

2. Socket Option by Febien Chevalier:
I think this is similar with your recommendation.

In BlueZ-dev mailing list, socket option patch is already proposed by Fabien
Chevalier [fabchevalier@free.fr]
But I couldn't find the conclusion ,its patch and your comments.

His last patch is located on "
http://marc.info/?l=linux-bluetooth&m=122142307223008&w=2"

I think this option is more safe, because this patch is only for
user-selected connection.

On a long term view, it's recommended to accept this kind of problem and
path BlueZ kernel. 
because many user may suffer from this headset , and they don't think this
problem due to the headset.
[The reason is that another commercial stack device supports this well]


Regards
Chanyeol