[U-Boot] AM335x configurations - console output at early stages lost

[Vasili Galka]

Hi Simon,

On Mon, May 4, 2015 at 7:52 PM, Vasili Galka <vvv444 at gmail.com> wrote:

> Hi Simon,
>
>
> On Mon, May 4, 2015 at 7:45 PM, Simon Glass <sjg at chromium.org> wrote:
>
>> Hi Vasili,
>>
>> On 4 May 2015 at 10:21, Vasili Galka <vvv444 at gmail.com> wrote:
>> >
>> > Hi Simon,
>> >
>> > On Thu, Apr 30, 2015 at 3:38 AM, Simon Glass <sjg at chromium.org> wrote:
>> >>
>> >> Hi Vasili,
>> >>
>> >> On 29 April 2015 at 10:57, Vasili Galka <vvv444 at gmail.com> wrote:
>> >> > Hi Tom,
>> >> >
>> >> > I’m working on rebasing an old U-Boot branch for our company’s AM335x
>> >> > based board upon the current U-Boot release (v2015.04). Our branch
>> was
>> >> > initially based on the v2013.10 release (the ti/am335x/ board was
>> >> > taken as reference).
>> >> >
>> >> > After making all the code compile, I discovered there was a change in
>> >> > the U-Boot initialization sequence, introduced by commit
>> >> > a6b541b09022acb6f7c2754100ae26bd44eed1d9, that prevents our code from
>> >> > working:
>> >> >
>> >> > Prior to the above mentioned commit, the console serial port was
>> >> > initialized at a very early stage in the
>> armv7/am335x/board.c:s_init()
>> >> > function (calling preloader_console_init()). Roughly saying, it was
>> >> > the first thing done by the SPL code. This made quite much sense, as
>> >> > it enabled all the code afterwards to output status/debugging info to
>> >> > the console.
>> >> > Moving the preloader_console_init() call to spl_board_init() caused
>> it
>> >> > to be run at a very "late" stage (as part of the board_init_r()
>> >> > function). AFAIU, now all the code that runs till then has no ability
>> >> > for any console output.
>> >> > For example, in our particular case, the DDR configuration code (in
>> >> > sdram_init()) reads the DDR type from an I2C EEPROM. Obviously, if
>> >> > something goes wrong in this process, we would like to report it to
>> >> > the console.
>> >> >
>> >> > I understand there was a good reason for the mentioned commit (the
>> >> > GD), however, as you see, we have also lost some important
>> >> > functionality with it. I would appreciate your advice on how to solve
>> >> > it.
>> >>
>> >> It would certainly be very useful and it would be great if we could
>> >> printf() from as soon as we have a stack (and perhaps printch() even
>> >> earlier). I think it can be done, but I have not put in the time to
>> >> figure it out.
>> >>
>> >> You could take a look at CONFIG_DEBUG_UART which at least allows
>> >> primitive UART access before gd is set up. I wonder if we could do
>> >> something like change printf() /purchatr() to use this when gd is
>> >> NULL?
>> >>
>> >> Regards,
>> >> Simon
>> >
>> >
>> > Thanks for the suggestion! I'm attempting to make it work.
>> > Was the "Debug UART" code (added in
>> 21d004368fc8a4da07147c58dfe9a4e16d4ab761) ever tested on AM335x? It does
>> not seem to work for me, I'm currently debugging in attempt to understand
>> the cause.
>>
>> From memory I think I did test it locally but did not send patches to
>> enable it.
>>
>> You will need to enable DEBUG_UART_NS16550, and supply suitable values
>> for DEBUG_UART_BASE and DEBUG_UART_CLOCK.
>>
>> But the debug UART code (in ns16550.c) is really simple so it should
>> not be possible to debug it.
>>
>> I think it would be good to have the DEBUG_UART_... options in the
>> board's defconfig file, even though CONFIG_DEBUG_UART itself is
>> disabled. Then people can enable it easily when they want to use the
>> debug UART.
>>
>> Regards,
>> Simon
>>
>
> Yes, of course, please see below the changes I applied. I expected to see
> a character written on to the UART, but I see nothing. So I attempt to
> compare the new serial code to the old one (from 2013) that definitely
> works.
>
> Best,
> Vasili
>
> ...
>

I've altered the debug_uart_init() to mimic the behaviour of NS16550_init()
more closely and it fixed the problem. I'll send a clean patch shortly.

Many thanks!

Best regards,
Vasili