[PATCH] arm64: add support for uart earlyprintk

[catalin.marinas@arm.com (Catalin Marinas)]

On Thu, Feb 28, 2013 at 02:06:27PM +0000, Anup Patel wrote:
> On Thu, Feb 28, 2013 at 6:22 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
> > On 28/02/13 12:10, Catalin Marinas wrote:
> >> On Thu, Feb 28, 2013 at 11:34:17AM +0000, Marc Zyngier wrote:
> >>> On 28/02/13 11:01, Anup Patel wrote:
> >>>> +}
> >>>> +
> >>>> +/*
> >>>>   * PL011 single character TX.
> >>>>   */
> >>>>  static void pl011_printch(char ch)
> >>>> @@ -47,6 +62,7 @@ struct earlycon_match {
> >>>>
> >>>>  static const struct earlycon_match earlycon_match[] __initconst = {
> >>>>     { .name = "pl011", .printch = pl011_printch, },
> >>>> +   { .name = "uart", .printch = uart_printch, },
> >>>
> >>> "uart" is way too generic. pl011 is an UART too, and I suspect most of
> >>> the backends that are going to be added here over time will be UARTs.
> >>>
> >>> "uart8250" would be a possibility (and actually consistent with the rest
> >>> of the kernel, see drivers/tty/serial/8250/8250_early.c.
> >>
> >> I think it makes more sense to use the existing 8250_early.c driver. It
> >> has more features than the simple earlyprintk implementation in the
> >> 64-bit kernel (like parsing more parameters, initialising the UART). The
> >> only difference is that the early_param is called "earlycon".
> >
> > Indeed, this seems to be the best way, as it removes the need for this
> > patch altogether.
> 
> The earlycon option does not seem to work on my ARM64 kernel because it uses
> ioremap_nocache() whereas ARM64 earlyprintk uses early_io_map()

We try to get earlyprintk as early as possible and ioremap() isn't
available yet, hence the arm64-specific early_io_map(). It looks like
x86 has a CONFIG_FIX_EARLYCON_MEM to avoid the ioremap() issue but I
wouldn't add yet another option for arm64.

So we can go with your arm64 patch, once you implement the feedback from
Marc.

-- 
Catalin