[U-Boot] [PATCH v4] Add board_pre_console_putc to deal with early console output

[Graeme Russ]

Hi Simon,

On Mon, Dec 5, 2011 at 3:31 PM, Simon Glass <sjg at chromium.org> wrote:
> Hi Graeme,
>
> On Sun, Dec 4, 2011 at 7:40 PM, Graeme Russ <graeme.russ at gmail.com> wrote:
>> Hi Simon,
>>
>> On Mon, Dec 5, 2011 at 2:34 PM, Simon Glass <sjg at chromium.org> wrote:
>>> Hi Stefan,
>>>
>>> On Sun, Dec 4, 2011 at 10:56 AM, Stefano Babic <sbabic at denx.de> wrote:
>>>> On 02/12/2011 19:16, Simon Glass wrote:
>>>>
>>>>> It actually started as a last-ditch panic message printer. It morphed
>>>>> into the general pre-console putc after discussions on the list (with
>>>>> Graeme).
>>>>>
>>>>> It can happen that your early board code does not know what clocks to
>>>>> use, or can't find a console, or some other critical error. It then
>>>>> calls panic() which silently dies or maybe reboots if you are lucky.
>>>>> This problem mostly comes about with device trees, where we must have
>>>>> certain info in the device tree before we can even get to relocation.
>>>>>
>>>>> The idea is that boards provide a way of outputting characters which
>>>>> tries to work on all types of boards with that SOC. For example, they
>>>>> output the characters on all UARTs with various clock options, etc.
>>>>>
>>>>> We use it on Tegra to print a friendly panic message when something is
>>>>> horribly wrong. In this case we will never make it to relocation so
>>>>> the pre-console buffer will not be displayed. So we can't rely on
>>>>> that.
>>>>
>>>> I do not know if this mechanism can be used on other SOCs, but IMHO it
>>>> does not hurt and it helps at least on Tegra, as you explained me.
>>>>
>>>> Applied to u-boot-staging, sbabic at denx.de branch
>>>
>>> Thanks. I think it can be used on any SOC with internal UARTs and a
>>> reasonably small list of permitted clock speeds. I suspect I will be
>>> trying it out on a few before long.
>>
>> Can it be used in conjuction with PRE_CONSOLE_BUFFER? i.e. panic()
>> prints the contents of the PRE_CONSOLE_BUFFER before the panic
>> message? I think that would be invaluable in determining what was
>> happening leading up to the panic
>
> It could have been when it was still a pre-console panic. But now that
> it has morphed into a general putc(), I suppose the only option is to
> somehow detect the first call and send the pre-console text first?

Couldn't the panic() function detect that the pre console buffer is
not empty and dump it before the panic message? Maybe even a generic
panic_dump_pre_console_buffer() which checks if console has been
initialised (in which case the buffer has already been dumped) and if
the buffer is not empty, use pre_console_putc() to dump the buffer.
This would be the first call in panic() followed by printing the panic
message

Regards,

Graeme