ARM, earlyconsole and io mapping

ARM, earlyconsole and io mapping

[Robert Jarzmik]

Hi,

I was wondering how the earlyconsole was supposed to work in order to add
support to the pxa architecture. More specifically, I don't understand how the
memory mapping is supposed to be available at the early parameters parsing.

The thing I don't understand is that :
 - in kernel initialization, parse_early_param() is called before
   machine_desc->map_io()
 - as a consequence, the architecture didn't add the vm early mapping for serial
   peripheral (which is added in pxa_map_io())
 - as a consequence, earlycon_map() will fail to map the serial MMIO
   (as in stack showed in [1])

Therefore, are there non-DT earlycon users on the list, and could they tell me
what way they get earlycon support in their arm boards please ?

Cheers.

-- 
Robert

[1] Call stack
#0  __arm_ioremap_pfn_caller (pfn=262400, offset=0, size=64, mtype=0, caller=0xc041d66c) at arch/arm/mm/ioremap.c:260
#1  0xc0015d34 in __arm_ioremap_caller (phys_addr=<value optimized out>, size=<value optimized out>, mtype=0, caller=<value optimized out>) at arch/arm/mm/ioremap.c:354
#2  0xc0015868 in __arm_ioremap (phys_addr=<value optimized out>, size=<value optimized out>, mtype=<value optimized out>) at arch/arm/mm/ioremap.c:383
#3  0xc041d66c in earlycon_map (buf=0x0, match=0x0, setup=0x40) at drivers/tty/serial/earlycon.c:48
#4  setup_earlycon (buf=0x0, match=0x0, setup=0x40) at drivers/tty/serial/earlycon.c:138
#5  0xc040a52c in do_early_param (param=0xc0424bf6 "earlycon", val=0xc0424bff "pxa,mmio,0x40100000,115200n8", unused=<value optimized out>) at init/main.c:420
#6  0xc00306bc in parse_one (doing=0xc0393668 "early options", args=0xc0424c1b "", params=<value optimized out>, num=<value optimized out>, min_level=0, max_level=0, unknown=0) at kernel/params.c:131
#7  parse_args (doing=0xc0393668 "early options", args=0xc0424c1b "", params=<value optimized out>, num=<value optimized out>, min_level=0, max_level=0, unknown=0) at kernel/params.c:216
#8  0xc040a854 in parse_early_options (cmdline=<value optimized out>) at init/main.c:430
#9  0xc040a890 in parse_early_param () at init/main.c:444
#10 0xc040d950 in setup_arch (cmdline_p=0xc07f3fd4) at arch/arm/kernel/setup.c:922
#11 0xc040a918 in start_kernel () at init/main.c:520
  => Here get_vm_area_caller() will fail in __arm_ioremap_pfn_caller().

ARM, earlyconsole and io mapping

[Thomas Petazzoni]

Dear Robert Jarzmik,

On Mon, 30 Mar 2015 14:05:33 +0200, Robert Jarzmik wrote:

> I was wondering how the earlyconsole was supposed to work in order to add
> support to the pxa architecture. More specifically, I don't understand how the
> memory mapping is supposed to be available at the early parameters parsing.
> 
> The thing I don't understand is that :
>  - in kernel initialization, parse_early_param() is called before
>    machine_desc->map_io()
>  - as a consequence, the architecture didn't add the vm early mapping for serial
>    peripheral (which is added in pxa_map_io())
>  - as a consequence, earlycon_map() will fail to map the serial MMIO
>    (as in stack showed in [1])
> 
> Therefore, are there non-DT earlycon users on the list, and could they tell me
> what way they get earlycon support in their arm boards please ?

A very early mapping of the I/O space is done in head.S, way before you
get into C code:

  http://lxr.free-electrons.com/source/arch/arm/kernel/head.S#L277

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

ARM, earlyconsole and io mapping

[Robert Jarzmik]

Thomas Petazzoni <thomas.petazzoni@free-electrons.com> writes:

> Dear Robert Jarzmik,
>
> On Mon, 30 Mar 2015 14:05:33 +0200, Robert Jarzmik wrote:
> A very early mapping of the I/O space is done in head.S, way before you
> get into C code:
>
>   http://lxr.free-electrons.com/source/arch/arm/kernel/head.S#L277
That's DEBUG_LL, right ? If that's what you mean, then I already have DEBUG_LL,
but that doesn't answer my question, as DEBUG_LL stops spitting logs after the
decompressor, and the errors I want to trap are between end of decompressor and
drivers probe.

And my question about earlycon remains, DEBUG_LL is not earlycon.

Cheers.

-- 
Robert

ARM, earlyconsole and io mapping

[Peter Hurley]

Hi Robert,

On 03/30/2015 08:05 AM, Robert Jarzmik wrote:
> Hi,
> 
> I was wondering how the earlyconsole was supposed to work in order to add
> support to the pxa architecture. More specifically, I don't understand how the
> memory mapping is supposed to be available at the early parameters parsing.
> 
> The thing I don't understand is that :
>  - in kernel initialization, parse_early_param() is called before
>    machine_desc->map_io()
>  - as a consequence, the architecture didn't add the vm early mapping for serial
>    peripheral (which is added in pxa_map_io())
>  - as a consequence, earlycon_map() will fail to map the serial MMIO
>    (as in stack showed in [1])
> 
> Therefore, are there non-DT earlycon users on the list, and could they tell me
> what way they get earlycon support in their arm boards please ?

Fixmap support has not been upstreamed yet, but if you're desperate, here's
the latest patch version:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-January/318372.html


If you're using /chosen/stdout-path, earlycon gets turned off at dummy
VT console load, so for right now, specify your earlycon and consoles on the
command line.

Regards,
Peter Hurley

ARM, earlyconsole and io mapping

[Robert Jarzmik]

Peter Hurley <peter@hurleysoftware.com> writes:

> Hi Robert,
>
> On 03/30/2015 08:05 AM, Robert Jarzmik wrote:
>> Hi,
>> 
>> I was wondering how the earlyconsole was supposed to work in order to add
>> support to the pxa architecture. More specifically, I don't understand how the
>> memory mapping is supposed to be available at the early parameters parsing.
> Fixmap support has not been upstreamed yet, but if you're desperate, here's
> the latest patch version:
> http://lists.infradead.org/pipermail/linux-arm-kernel/2015-January/318372.html

Ah I'm not desperate, that would be nice to have tool on my automation
platform. I've gone the DEBUG_LL + JTAG/__log_buf memory dump so far.

And yet I remember viewing early concole patches a few monthes back, and I
wondered how the authors could make them work ... And as my memory didn't serve
me well, I couldn't remember your name.

> If you're using /chosen/stdout-path, earlycon gets turned off at dummy
> VT console load, so for right now, specify your earlycon and consoles on the
> command line.
Ah I'm using device-tree in only 1/3rd of the PXA tests :
 - 1/3rd for legacy platform_data tests
 - 1/3rd for device-tree pxa platforms
 - 1/3rd for multiplatform kernel

Anyway, I hope you'll have Russell's comments addressed, and that your patch
will make it upstream. Thanks for the information, it is exactly what I was
asking for.

Cheers.

-- 
Robert