Return-Path: <cyrus@holtmann.org>
Subject: RE: Does anybody know conn->power_save variable in hci_conn.c?
From: Marcel Holtmann <marcel@holtmann.org>
To: Chan-Yeol Park <chanyeol.park@samsung.com>
Cc: linux-bluetooth@vger.kernel.org
In-Reply-To: <004001c9ff6c$7838f7c0$68aae740$%park@samsung.com>
References: <002e01c9fedc$f5fcb790$e1f626b0$%park@samsung.com>
	 <1246988209.3384.10.camel@localhost.localdomain>
	 <004001c9ff6c$7838f7c0$68aae740$%park@samsung.com>
Content-Type: text/plain
Date: Tue, 07 Jul 2009 20:38:10 -0700
Message-Id: <1247024290.3384.26.camel@localhost.localdomain>
Mime-Version: 1.0
Sender: linux-bluetooth-owner@vger.kernel.org
List-ID: <linux-bluetooth.vger.kernel.org>

Hi Chan-Yeol,

> >> /*bluetooth kernel hci_conn.c */
> >> 382 void hci_conn_enter_active_mode(struct hci_conn *conn)
> >> 383 {
> >> 384         struct hci_dev *hdev = conn->hdev;
> >> 385
> >> 386         BT_DBG("conn %p mode %d", conn, conn->mode);
> >> 387
> >> 388         if (test_bit(HCI_RAW, &hdev->flags))
> >> 389                 return;
> >> 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),
> &cp);
> >> 399         }
> >> 400
> >> 401 timer:
> >> 402         if (hdev->idle_timeout > 0)
> >> 403                 mod_timer(&conn->idle_timer,
> >> 404                         jiffies +
> msecs_to_jiffies(hdev->idle_timeout));
> >> 405 }
> >> 406 
> >> 
> >> 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.

Regards

Marcel