[U-Boot] [PATCH] serial: 16550: Add JZ47xx support

[Simon Glass]

Hi Marek,

On 26 May 2016 at 10:47, Marek Vasut <marex at denx.de> wrote:
> On 05/26/2016 06:44 PM, Simon Glass wrote:
>> Hi Marek,
>>
>> On 26 May 2016 at 10:34, Marek Vasut <marex at denx.de> wrote:
>>> On 05/26/2016 03:29 PM, Simon Glass wrote:
>>>> Hi Marek,
>>>>
>>>> On 25 May 2016 at 16:35, Marek Vasut <marex at denx.de> wrote:
>>>>> On 05/26/2016 12:31 AM, Daniel Schwierzeck wrote:
>>>>>>
>>>>>>
>>>>>> Am 26.05.2016 um 00:21 schrieb Marek Vasut:
>>>>>>> On 05/26/2016 12:17 AM, Daniel Schwierzeck wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> Am 25.05.2016 um 02:19 schrieb Marek Vasut:
>>>>>>>>> The Ingenic JZ47xx requires special bit (UART_EN) set in FCR register
>>>>>>>>> in order to work at all. Add this special case handling into the driver.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Marek Vasut <marex at denx.de>
>>>>>>>>> Cc: Tom Rini <trini at konsulko.com>
>>>>>>>>> Cc: Simon Glass <sjg at chromium.org>
>>>>>>>>> Cc: Daniel Schwierzeck <daniel.schwierzeck at gmail.com>
>>>>>>>>> Cc: Paul Burton <paul.burton at imgtec.com>
>>>>>>>>> ---
>>>>>>>>>  drivers/serial/ns16550.c | 8 ++++++++
>>>>>>>>>  1 file changed, 8 insertions(+)
>>>>>>>>>
>>>>>>>>> diff --git a/drivers/serial/ns16550.c b/drivers/serial/ns16550.c
>>>>>>>>> index 30ba0aa..1323881 100644
>>>>>>>>> --- a/drivers/serial/ns16550.c
>>>>>>>>> +++ b/drivers/serial/ns16550.c
>>>>>>>>> @@ -50,6 +50,14 @@ DECLARE_GLOBAL_DATA_PTR;
>>>>>>>>>  #endif
>>>>>>>>>  #endif
>>>>>>>>>
>>>>>>>>> +#ifdef CONFIG_ARCH_JZ47XX
>>>>>>>>> +#undef UART_FCRVAL
>>>>>>>>> +/* Ingenic JZ47xx SoCs require that a 'UART Module Enable' bit be set */
>>>>>>>>> +#define UART_FCR_UME              0x10
>>>>>>>>> +#define UART_FCRVAL (UART_FCR_FIFO_EN | UART_FCR_RXSR |   \
>>>>>>>>> +               UART_FCR_TXSR | UART_FCR_UME)
>>>>>>>>> +#endif
>>>>>>>>
>>>>>>>> I think this could be added as DT property
>>>>>>>
>>>>>>> Not for SPL, which has 14 kiB size limit and it is itching to overflow.
>>>>>>> I am literally counting bytes in the SPL and removing slop from
>>>>>>> structures to make it fit, just barely. With the USB loader, I can
>>>>>>> brutalize the SPL into having extremely rudimentary UART support now
>>>>>>> (like printch() being the most advanced output mechanism, but you can
>>>>>>> only use it three times, otherwise the code won't fit and the board is
>>>>>>> eaten by demons) and this is where this patch comes into play.
>>>>>>>
>>>>>>> So yes, for full u-boot, this _should_ be part of DT. For SPL, please apply.
>>>>>>>
>>>>>>
>>>>>> ok, but wouldn't it be better to introduce an option like
>>>>>> CONFIG_SYS_NS16550_UME instead of using the SoC-specific
>>>>>> CONFIG_ARCH_JZ47XX. This driver is messed up enough ;)
>>>>>
>>>>> I was undecided between this (like the IER) and adding new ifdef (like
>>>>> SOC_KEYSTONE). Whichever way is fine with me. Yeah, the driver is
>>>>> repugnant for sure.
>>>>>
>>>>>>>>> +
>>>>>>>>>  #ifndef CONFIG_SYS_NS16550_IER
>>>>>>>>>  #define CONFIG_SYS_NS16550_IER  0x00
>>>>>>>>>  #endif /* CONFIG_SYS_NS16550_IER */
>>>>>>>>>
>>>>
>>>> That way seems better to me. You should be able to add your UME flag
>>>> as a Kconfig for this driver, in drivers/serial/Kconfig, which
>>>> defaults to 0. It would be good to keep out board-specific stuff from
>>>> this file, as you did with OMAP1510.
>>>
>>> I'm not really sure I want to expose the CONFIG_SYS_NS16550_FCR override
>>> via Kconfig. The extra bits should be set via DT props
>>> unless there is some really good reason why that cannot be done
>>> (like size limitation in SPL or DT not available).
>>>
>>> I tried these approaches:
>>> 1) Add Kconfig option for CONFIG_SYS_NS16550_FCR , where you
>>>    set specific byte value. This makes it hard to figure out
>>>    which bits are set in the FCR just by looking at the value.
>>> 2) Add Kconfig option selected if ARCH_JZ47XX is selected and
>>>    use that in the driver to add the extra UME bit.
>>> 3) Define CONFIG_SYS_NS16550_FCR in include/configs/board.h
>>>    using the macros from ns16550.h . This makes it obvious
>>>    which bits are set.
>>>
>>> I am undecided between 2 and 3, but inclined to go with 3.
>>> What do you think ?
>>
>> How about making your feature an option, like CONFIG_SYS_NS16550_UEN,
>> which defaults to 0, but in your case is 0x10. You can OR it into the
>> value.
>
> That's quite close to 1) , but then what if someone comes and wants to
> remove some bits from the FCR instead of adding some?

Seems like speculative complexity to me - let's worry about it when it happens!

- Simon