Re: [RFC PATCH] always probe UART HW when options are not specified

[Sebastian Frias]

Hi Peter,

On 01/13/2016 05:34 PM, Peter Hurley wrote:
> On 01/13/2016 03:14 AM, Sebastian Frias wrote:
> > Hi Peter,
> >
> > On 01/12/2016 08:47 PM, Peter Hurley wrote:
> > > On 01/12/2016 06:22 AM, Sebastian Frias wrote:
> > > > For the record, I'm using a SoC emulator, and thus do not have a bootloader per se and there are a bunch of other things that I cannot count on.
> > > > The emulator has the UART pre-setup, so I just need Linux to take over without changing the parameters.
> > > > Ideally, I would like to have the same image of Linux+DT to start in any instance of the emulator or real chips, regardless of the clock ratios, that's why I sort of need Linux to not change the UART speed, which is quite tricky because there are no clock generators in the emulator.
> > >
> > > Got it, thanks for the info.
> > > Please test the series I just cc'd you on plus the patch I sent
> > > you yesterday.
> > >
> > > That should get you an earlycon up and running on that simulator;
> > > let me know if it doesn't and we'll go from there.
> >
> > Ok, thanks.
> > I will try as soon as we finish rebasing our changes on top of
> > "mainline" (or HEAD, or is it "-next"? I don't know how you guys call
> > the most recent code base for Linux)
>
> The basic tree organization is
>
>     Linus's tree  <---- linux-next[date]  <----- maintainers' trees
>     (mainline)          (next)                   (eg., Greg's tty tree)
>
> The patches I sent you (along with any required modifications during the
> review cycle) are based on the tty-next branch in Greg's tty tree here
> git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty.git
>
> As of this email, Greg's tree is based on Linus's 4.4-rc6 which should
> be stable enough for you to test these patches on.

Thanks for your explanation :-)

> > Actually, I tried yesterday on a 4.1.13 but
> > "[drivers/tty/serial/earlycon.c:62] earlycon_map()" was still the
> > last message I got, just as with the OF_EARLYCON_DECLARE hack I had
> > previously talk about, and so I would like to test on the same
> > conditions than you, mainline.
>
> So just to confirm, you applied "8250: Add Au1x00/RT288x earlycon support"
> to 4.1.13 and the earlycon didn't come up when you used the kernel
> command line option like so:
>
> 	earlycon=rt288x,mmio32,0x10700
>
> Is that correct?

Sort of, I applied your patch "8250: Add Au1x00/RT288x earlycon 
support", and then tried with something like:

	aliases {
		serial0 = &uart;
	};

	chosen {
		bootargs = "earlycon console mem=256M earlyprintk debug ignore_loglevel";
		stdout-path = "serial0:115200n8";
	};

	uart: serial@10700 {
		compatible = "ralink,rt2880-uart";
		reg = <0x10700 0x30>;
		interrupts = <1 IRQ_TYPE_LEVEL_HIGH>;
		clock-frequency = <7372800>;
		reg-shift = <2>;
	};

I can see that the code is correctly picking up the rt288x from the DT 
(I put logs):

...
[drivers/of/fdt.c:834] early_init_dt_scan_chosen_serial(): 
'ralink,rt2880-uart'
[drivers/tty/serial/earlycon.c:239] of_setup_earlycon(): addr=00010700 
setup@c024c14c
[drivers/tty/serial/earlycon.c:62] earlycon_map(): paddr 0x00010700 size 
4096

(last log visible on UART)

where 0xc024c14c resolves to 'early_rt288x_setup' from your patch (in 
the disassembly)

This happens when parsing the first token of the bootargs, "earlycon".

I will let you know how it goes the tests on 4.4.

Best regards,

Sebastian