[U-Boot] [PATCH 1/3] dm: serial: Do not panic if no serial ports are found

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 08-07-15 16:08, Simon Glass wrote:
> +Masahiro
>
> Hi Hans,
>
> On 8 July 2015 at 05:56, Hans de Goede <hdegoede@redhat.com> wrote:
>> Hi Simon,
>>
>>
>> On 07-07-15 09:00, Hans de Goede wrote:
>>>
>>> Hi,
>>>
>>> On 06-07-15 18:39, Simon Glass wrote:
>>>>
>>>> Hi Hans,
>>>>
>>>> On 5 July 2015 at 12:56, Hans de Goede <hdegoede@redhat.com> wrote:
>>>>>
>>>>> Some boards simply do not have any serial ports. Also no one will see
>>>>> the
>>>>> panic message as there is no where to print it if no serial port is
>>>>> found
>>>>> (and other stdout options are not yet set up at this point).
>>>>>
>>>>
>>>> It is visible (or will be when some patches land) if you have a debug
>>>> UART set up.
>>>
>>>
>>> Ok.
>>>
>>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>>> ---
>>>>>    drivers/serial/serial-uclass.c | 8 +++-----
>>>>>    1 file changed, 3 insertions(+), 5 deletions(-)
>>>>>
>>>>> diff --git a/drivers/serial/serial-uclass.c
>>>>> b/drivers/serial/serial-uclass.c
>>>>> index 815fec3..f036499 100644
>>>>> --- a/drivers/serial/serial-uclass.c
>>>>> +++ b/drivers/serial/serial-uclass.c
>>>>> @@ -27,7 +27,7 @@ static const unsigned long baudrate_table[] =
>>>>> CONFIG_SYS_BAUDRATE_TABLE;
>>>>>    #error "Serial is required before relocation - define
>>>>> CONFIG_SYS_MALLOC_F_LEN to make this work"
>>>>>    #endif
>>>>>
>>>>> -static void serial_find_console_or_panic(void)
>>>>> +static void serial_find_console(void)
>>>>>    {
>>>>>           struct udevice *dev;
>>>>>           int node;
>>>>> @@ -77,14 +77,12 @@ static void serial_find_console_or_panic(void)
>>>>>                   }
>>>>>    #undef INDEX
>>>>>           }
>>>>> -
>>>>> -       panic_str("No serial driver found");
>>>>>    }
>>>>>
>>>>>    /* Called prior to relocation */
>>>>>    int serial_init(void)
>>>>>    {
>>>>> -       serial_find_console_or_panic();
>>>>> +       serial_find_console();
>>>>>           gd->flags |= GD_FLG_SERIAL_READY;
>>>>>
>>>>>           return 0;
>>>>> @@ -93,7 +91,7 @@ int serial_init(void)
>>>>>    /* Called after relocation */
>>>>>    void serial_initialize(void)
>>>>>    {
>>>>> -       serial_find_console_or_panic();
>>>>> +       serial_find_console();
>>>>>    }
>>>>>
>>>>>    static void _serial_putc(struct udevice *dev, char ch)
>>>>
>>>>
>>>> How is this handled before driver model?
>>>
>>>
>>> It was not, the boards involved all use the A13 SoC which is the
>>> same die as the A10s but then in a different (cheaper) package
>>> with way less pins. As such uart0 is not routed to the outside
>>> on the A13. When not setting CONFIG_DM_SERIAL we are simply using
>>> uart0 as serial console, and that ends at the bonding pads at the
>>> edge of the die. doing things that way is not really useful,
>>> but there is a serial console and u-boot is happy.
>>>
>>> In the non devicetree world this sort of was a natural hack, we
>>> simply left CONSOLE_INDEX undefined for these boards, which causes
>>> it to default to 1 / uart0 and things just work. I had never given
>>> this much thought until moving to devicetree.
>>>
>>> As you can hopefully understand I do not want to do the samething
>>> with DM_SERIAL builds as that would require hacking up the dts
>>> to add a node for hardware which is effectively disabled for the
>>> A13 package of the sun5i die.
>>>
>>>   > It is possible to mark a
>>>>
>>>> device disabled in device tree - perhaps we should have a way to see
>>>> that the console is present (i.e. there is an alias) but it is
>>>> disabled?
>>>
>>>
>>> That seems rather ugly, I really do not want to add an alias to
>>> a non enabled device, that just feels wrong on all levels.
>>>
>>>> Normally a serial console is required, so I'd like to preserve this
>>>> behaviour for most boards.
>>>
>>>
>>> Ok, I actually expected as much, this was actually my second solution
>>> for the problem at hand, but as it was the simplest solution I decided
>>> to submit this version first. My first / original solution is to add a
>>> CONFIG_DM_SERIAL_NO_PANIC Kconfig option which can then be set on boards
>>> which have no serial port.
>>>
>>> Or probably better a CONFIG_DM_SERIAL_NO_SERIAL_CONSOLE option ?
>>>
>>> This option would disable the panic / disable probing all together while
>>> we would keep using DM_SERIAL on these boards, because:
>>>
>>> 1) Consistency I really want to use DM_FOO everywhere for sunxi
>>> 2) The non dm serial code cannot handle not having a serial port either,
>>>      and fixing it looks like it is going to be a bit harder (from a quick
>>>      glance).
>>
>>
>> You probably just have not gotten around to this yet, but in case it is
>> not clear, the above solution (adding a CONFIG_DM_SERIAL_NO_SERIAL_CONSOLE
>> Kconfig option) is a suggestion on how to fix. I'll happily implement this
>> if people like it, but atm I'm waiting for feedback on the suggestion.
>>
>
> I've been thinking about it. Perhaps an option to allow U-Boot to run
> without a console is best. But instead of CONFIG_DM_SERIAL_NO_PANIC,
> perhaps we could have CONFIG_REQUIRE_CONSOLE which is normally y.

That works for me, although I would prefer CONFIG_REQUIRE_SERIAL_CONSOLE,
the boards in question do have a console in the form of a cfb console.

Regards,

Hans