Re: [RFC v2 8/9] Bluetooth: drop HCI_UART_INIT_PENDING support

[Martin Blumenstingl <martin.blumenstingl-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org>]

Hi Loic,

On Wed, Jan 3, 2018 at 6:14 PM, Loic Poulain <loic.poulain-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
> Hi Martin,
>
> On 2 January 2018 at 22:06, Martin Blumenstingl
> <martin.blumenstingl-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org> wrote:
>> Hi Marcel,
>>
>> thank you for looking into this latest version!
>>
>> On Tue, Jan 2, 2018 at 12:04 PM, Marcel Holtmann <marcel-kz+m5ild9QBg9hUCZPvPmw@public.gmane.org> wrote:
>>> Hi Martin,
>>>
>>>> The three-wire (H5) protocol is the only protocol which uses
>>>> HCI_UART_INIT_PENDING.
>>>> Unfortunately the benefits of using this flag are currently unknown. It
>>>> was added in commit 9f2aee848fe6 ("Bluetooth: Add delayed init sequence
>>>> support for UART controllers"). In my experiments (with the
>>>> "rtk_hciattach" tool - a customized version of hciattach for Realtek
>>>> chipsets) I started the tool before and after this patch while the
>>>> Bluetooth chipset was disabled (by pulling it's enable GPIO LOW). In
>>>> both cases hci0 was not created - thus HCI_UART_INIT_PENDING is not
>>>> required in that case.
>>>>
>>>> Removing this code also has another benefit: hci_serdev.c does not
>>>> support the delayed initialization / registration. Thus the protocol
>>>> implementation (hci_h5) never receives any data with this check still
>>>> in place. For the H5 protocol this means that the initialization never
>>>> completes (because the sync response never arrives). Even if the
>>>> initialization would succeed later on the drivers would call
>>>> hci_uart_init_ready() which schedules the registration which is
>>>> currently not implemented by hci_serdev.c.
>>>>
>>>> Removing the HCI_UART_INIT_PENDING check makes the code easier to read
>>>> and also fixes the initalization of devices (implemented with the serdev
>>>> library) which use the H5 protocol.
>>>>
>>>> Signed-off-by: Martin Blumenstingl <martin.blumenstingl-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org>
>
> I think the original goal is to perform H5 init peacefully.
> The H5 protocol needs to be open in order to send/receive H5 link
> packets during the H5 initialization/synchronization step.
> During this stage, driver prevents upper stack to send any HCI packet
> by delaying the HCI device registration.
thank you for this explanation
do you have a test-case (manual list of steps, a scripts, etc.) which
I can use to compare the old and new behavior?

based on your explanation this would mean that we should add
HCI_UART_INIT_PENDING support to the serdev code rather than removing
it


Regards
Martin