From mboxrd@z Thu Jan  1 00:00:00 1970
From: Graeme Russ <graeme.russ@gmail.com>
Date: Fri, 9 Mar 2012 10:26:12 +1100
Subject: [U-Boot] ns16550: tegra: Specify debugging serial port at boot.
In-Reply-To: <4F592815.9000006@wwwdotorg.org>
References: <1325890269-5613-1-git-send-email-swarren@nvidia.com>
	<4F58F5B8.6070402@wwwdotorg.org>
	<20120308183903.4575B202D69@gemini.denx.de>
	<4F590B25.8090503@wwwdotorg.org>
	<20120308212957.DE1CC202D69@gemini.denx.de>
	<4F592815.9000006@wwwdotorg.org>
Message-ID: <CALButCKini0Tjm=ycES5nBwmv_jLKy_912WTEjOpa8nXhSVAPA@mail.gmail.com>
List-Id: <u-boot.lists.denx.de>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: u-boot@lists.denx.de

Hi Stephen, Wolfgang,

On Fri, Mar 9, 2012 at 8:43 AM, Stephen Warren <swarren@wwwdotorg.org> wrote:
> On 03/08/2012 02:29 PM, Wolfgang Denk wrote:
>> Dear Stephen,
>>
>> In message <4F590B25.8090503@wwwdotorg.org> you wrote:
>>>
>>>> I don't like to see such highly architecture specific stuff in common
>>>> code, especially if it's such a dirty hack like this.
>>>
>>> Are there any hooks where we can do the same thing in SoC-specific code?
>>
>> Not without additional trickery, but I think this is actually a good
>> thing.
>>
>> The method implemented here is but a dirty hack, and should not be
>> used.

INIT_FUNC will resolve this issue long-term - Tegra can just put in:

int tegra_set_debug_port(void)
{
        /*
         * Put a 'D' in the scratchpad to let the kernel know which UART
         * for earlyprintk [D]ebugging.
         */
        serial_out('D', &com_port->spr);

        return 0;
}
INIT_FUNC(set_debug_port, tegra_set_debug_port, *serial_init)

>>> The point of this information is to enable the kernel's earlyprintk
>>> support, which runs well before the device tree, or other mechanisms,
>>> are available.
>>
>> Sorry, but I don't buy that this is the only possible way to do that.
>> Or how comes only tegra2 would need that, while all other SoCs and
>> architectures can do without it?
>
> First, OMAP does something very similar; the kernel low-level debug code
> looks at UART1's scratch pad register, and derives which UART to use
> based on the value stored there. If none of the expected values is
> found, it appears to default to UART1.
>
> On Tegra, the UART registers can't be read unless the UART is clocked
> and not in reset. So, the Tegra code looks at each UART in the system,
> and finds one that's in that state. To cater for the scenario where
> multiple UARTs are clocked-and-not-reset, the code also checks whether
> the UART scratch register contains 'D' ("D"ebug) so it's sure it picked
> the correct one.
>
> So, at leasst OMAP has set precedent here. There may be others; I didn't
> check.
>
> Some other SoCs may have only 1 UART and not need to auto-select.
>
> Some other SoCs with multiple UARTs may designate a single specific UART
> as the debug port rather than the board designer apparently randomly
> picking which one to use. In either of those cases, the kernel's
> low-level debug code can simply hard-code the UART address and do
> without the hand-shaking.

As far as I am aware, earlyprintk happens well before processing the kernel
command line so, IMHO, I don't consider setting up the hardware in order
for the kernel to get some low-level information which cannot be
provided by the command line as a hack. Reading the above, I actually think
it is quite elegant

Regards,

Graeme