[PATCH] Bluetooth: Assure BREDR HCI device first in the list

[PATCH] Bluetooth: Assure BREDR HCI device first in the list

[Emeltchenko Andrei]

From: Andrei Emeltchenko <andrei.emeltchenko@intel.com>

Using different list_add to make sure that BR/EDR HCI device is the
first device in the hci_dev_list. This is needed for e.g. A2MP Discover
Command which requires that "entry for Controller ID 0x00 (Primary
BR/EDR controller) shall always be sent and shall be the first entry
in the Controller List.
Also output from hciconfig looks nicer :-)

Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
---
 net/bluetooth/hci_core.c |    6 +++++-
 1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 02a6f15..bf24218 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -1501,7 +1501,11 @@ int hci_register_dev(struct hci_dev *hdev)
 
 	sprintf(hdev->name, "hci%d", id);
 	hdev->id = id;
-	list_add(&hdev->list, head);
+
+	if (hdev->dev_type == HCI_BREDR)
+		list_add(&hdev->list, head);
+	else
+		list_add_tail(&hdev->list, head);
 
 	atomic_set(&hdev->refcnt, 1);
 	spin_lock_init(&hdev->lock);
-- 
1.7.4.1

Re: [PATCH] Bluetooth: Assure BREDR HCI device first in the list

[Mat Martineau]

Hi Andrei -

On Tue, 15 Nov 2011, Emeltchenko Andrei wrote:

> From: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
>
> Using different list_add to make sure that BR/EDR HCI device is the
> first device in the hci_dev_list. This is needed for e.g. A2MP Discover
> Command which requires that "entry for Controller ID 0x00 (Primary
> BR/EDR controller) shall always be sent and shall be the first entry
> in the Controller List.
> Also output from hciconfig looks nicer :-)
>
> Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> ---
> net/bluetooth/hci_core.c |    6 +++++-
> 1 files changed, 5 insertions(+), 1 deletions(-)
>
> diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
> index 02a6f15..bf24218 100644
> --- a/net/bluetooth/hci_core.c
> +++ b/net/bluetooth/hci_core.c
> @@ -1501,7 +1501,11 @@ int hci_register_dev(struct hci_dev *hdev)
>
> 	sprintf(hdev->name, "hci%d", id);
> 	hdev->id = id;
> -	list_add(&hdev->list, head);
> +
> +	if (hdev->dev_type == HCI_BREDR)
> +		list_add(&hdev->list, head);
> +	else
> +		list_add_tail(&hdev->list, head);
>
> 	atomic_set(&hdev->refcnt, 1);
> 	spin_lock_init(&hdev->lock);
> -- 
> 1.7.4.1

I'm don't think it works to depend on the ordering in hdev->list for 
A2MP discover, since there might be multiple BR/EDR controllers.

Controller 0x00 is the BR/EDR controller that received the A2MP 
discover request.  The rest of the list should be only AMP 
controllers, any other BR/EDR controllers should not be listed.  In 
practice, this means that the 0x00 entry in the list is hard-coded and 
hdev->list is only used to look for AMP controllers.

--
Mat Martineau
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum

Re: [PATCH] Bluetooth: Assure BREDR HCI device first in the list

[Emeltchenko Andrei]

Hi Mat,

On Tue, Nov 15, 2011 at 08:45:47AM -0800, Mat Martineau wrote:
> >Using different list_add to make sure that BR/EDR HCI device is the
> >first device in the hci_dev_list. This is needed for e.g. A2MP Discover
> >Command which requires that "entry for Controller ID 0x00 (Primary
> >BR/EDR controller) shall always be sent and shall be the first entry
> >in the Controller List.
> >Also output from hciconfig looks nicer :-)
> >
> >Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> >---
> >net/bluetooth/hci_core.c |    6 +++++-
> >1 files changed, 5 insertions(+), 1 deletions(-)
> >
> >diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
> >index 02a6f15..bf24218 100644
> >--- a/net/bluetooth/hci_core.c
> >+++ b/net/bluetooth/hci_core.c
> >@@ -1501,7 +1501,11 @@ int hci_register_dev(struct hci_dev *hdev)
> >
> >	sprintf(hdev->name, "hci%d", id);
> >	hdev->id = id;
> >-	list_add(&hdev->list, head);
> >+
> >+	if (hdev->dev_type == HCI_BREDR)
> >+		list_add(&hdev->list, head);
> >+	else
> >+		list_add_tail(&hdev->list, head);
> >
> >	atomic_set(&hdev->refcnt, 1);
> >	spin_lock_init(&hdev->lock);
> >-- 
> >1.7.4.1
> 
> I'm don't think it works to depend on the ordering in hdev->list for
> A2MP discover, since there might be multiple BR/EDR controllers.

What about following change:

-    list_add(&hdev->list, head);
+    list_add_tail(&hdev->list, head);

This way hci devices queued to the list as opposed to stacked now.
Usual use case is that we have the 1st device BR/EDR and second AMP.

> Controller 0x00 is the BR/EDR controller that received the A2MP
> discover request.  The rest of the list should be only AMP
> controllers, any other BR/EDR controllers should not be listed.  In
> practice, this means that the 0x00 entry in the list is hard-coded
> and hdev->list is only used to look for AMP controllers.

It is actually how it is done in my RFC, I wanted to make it nicer :-(

Best regards 
Andrei Emeltchenko 

Re: [PATCH] Bluetooth: Assure BREDR HCI device first in the list

[Mat Martineau]

On Wed, 16 Nov 2011, Emeltchenko Andrei wrote:

> Hi Mat,
>
> On Tue, Nov 15, 2011 at 08:45:47AM -0800, Mat Martineau wrote:
>>> Using different list_add to make sure that BR/EDR HCI device is the
>>> first device in the hci_dev_list. This is needed for e.g. A2MP Discover
>>> Command which requires that "entry for Controller ID 0x00 (Primary
>>> BR/EDR controller) shall always be sent and shall be the first entry
>>> in the Controller List.
>>> Also output from hciconfig looks nicer :-)
>>>
>>> Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
>>> ---
>>> net/bluetooth/hci_core.c |    6 +++++-
>>> 1 files changed, 5 insertions(+), 1 deletions(-)
>>>
>>> diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
>>> index 02a6f15..bf24218 100644
>>> --- a/net/bluetooth/hci_core.c
>>> +++ b/net/bluetooth/hci_core.c
>>> @@ -1501,7 +1501,11 @@ int hci_register_dev(struct hci_dev *hdev)
>>>
>>> 	sprintf(hdev->name, "hci%d", id);
>>> 	hdev->id = id;
>>> -	list_add(&hdev->list, head);
>>> +
>>> +	if (hdev->dev_type == HCI_BREDR)
>>> +		list_add(&hdev->list, head);
>>> +	else
>>> +		list_add_tail(&hdev->list, head);
>>>
>>> 	atomic_set(&hdev->refcnt, 1);
>>> 	spin_lock_init(&hdev->lock);
>>> --
>>> 1.7.4.1
>>
>> I'm don't think it works to depend on the ordering in hdev->list for
>> A2MP discover, since there might be multiple BR/EDR controllers.
>
> What about following change:
>
> -    list_add(&hdev->list, head);
> +    list_add_tail(&hdev->list, head);
>
> This way hci devices queued to the list as opposed to stacked now.
> Usual use case is that we have the 1st device BR/EDR and second AMP.

I think that's a harmless change.

You can't really count on the controllers being added in a certain 
order, though.  AMP controllers will be typically be implemented 
within WiFi drivers, and those drivers may load before or after the 
BR/EDR device is attached.

>> Controller 0x00 is the BR/EDR controller that received the A2MP
>> discover request.  The rest of the list should be only AMP
>> controllers, any other BR/EDR controllers should not be listed.  In
>> practice, this means that the 0x00 entry in the list is hard-coded
>> and hdev->list is only used to look for AMP controllers.
>
> It is actually how it is done in my RFC, I wanted to make it nicer :-(


--
Mat Martineau
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum