[U-Boot] [PATCH v4 0/6] Support for the Turris Omnia router

[Andreas Färber]

Am 20.01.2018 um 02:40 schrieb Andreas Färber:
> Hi,
> 
> Am 18.01.2018 um 18:20 schrieb Stefan Roese:
>> On 17.01.2018 16:52, Andreas Färber wrote:
>>> Am 09.06.2017 um 19:28 schrieb Marek Behún:
>>>> This is the fourth version of patches for adding support for the
>>>> Turris Omnia board, a router developed by the CZ.NIC.
>>>
>>> I'm still facing trouble testing turris_omnia on latest v2018.01.
>>>
>>> First, that made me notice there's no README for how to test and deploy.
>>> I'm aware of temporary:
>>> sendbeacon /dev/ttyUSBx
>>
>> I have to admit, that don't know anything about this "sendbeacon"
>> tool.
>>
>>> ./tools/kwboot -t -B 115200 /dev/ttyUSBx -b u-boot-spl.kwb -p
>>
>> This is what I have used, when I tested / debugged images for
>> Armada XP / 38x. Please note that the init sequence is somewhat
>> "fragile" - so I added the -q and -s parameters, to optionally
>> finetune the startup timings:
>>
>> # kwboot
>> ...
>>   -q <req-delay>:  use specific request-delay
>>   -s <resp-timeo>: use specific response-timeout
>>
>> You might what to play a bit with these parameters as well.
> 
> I saw them but had no idea what to pass as values. ;)
> I did try -a, but it worked with and without.
> 
>> BTW: I don't have access to the Omnia router, so I can't
>> test anything on this specific platform.
>>
>> BTW2: Kosta from Marvell just recently added a new tool / script,
>> to help debug / boot Marvell MVEBU boards:
>>
>> tools/mrvl_uart.sh
>>
>> He told me that its better to use than the "old" kwboot tool.
>> I never found the time to use it up until now, so I have no
>> personal experience. But I'm pretty sure that Kosta did a
>> great job here. So please give it a try.
> 
> I did not get it to work ... or was not patient enough.
> 
>>> # or without -p when s/BOOT_FROM spi/BOOT_FROM uart/
>>> and permanent:
>>> tftpboot 0x1000000 u-boot-spl.kwb
>>> sf probe
>>> sf update 0x1000000 0 $filesize
>>>
>>> I used to have the original factory CZ.NIC U-Boot in SPI and booted test
>>> versions only via sendbeacon+kwboot.
>>>
>>> With mainline that appears to be broken - the CONFIG_ARMADA_38X code in
>>> arch/arm/mach-mvebu/spl.c seems to run into !boot_device and instead of
>>> UART tries to boot from SPI - nothing happens then and kwboot complains.
>>> I can force it to continue booting from UART by commenting out the if.
>>> So Stefan, it looks like your auto-detection is not working here and the
>>> Kconfig option to force it was dropped prematurely.
>>
>> Hmmm. Then some patch must have broken this UART boot-ability. Could
>> you by any chance git-bisect, to check which patch broke this
>> functionality? Perhaps some of the newer patches from Sean Nyekjaer?
> 
> I've so far found that v2017.11 had UART boot working okay.

git-bisect pointed to this commit:

e83e2b390038c9075642cb243a6292241beb8d73 is the first bad commit
commit e83e2b390038c9075642cb243a6292241beb8d73
Author: Sean Nyekjaer <sean.nyekjaer at prevas.dk>
Date:   Fri Nov 24 14:01:28 2017 +0100

    arm: mvebu: fix boot from UART when in fallback mode

    It's the first 8 bits of the bootrom error register that
    contain the boot error/fallback error code. Let's check that
    and continue to boot from UART.

    Signed-off-by: Sean Nyekjaer <sean.nyekjaer at prevas.dk>
    Signed-off-by: Stefan Roese <sr at denx.de>

:040000 040000 c4c5cb4287ae8c3ced749a9734213a5844ddf1d9
772ec1e6401cbb2616b1337ff8757b72240458b3 M	arch

Regards,
Andreas

>>> When forcing UART, as soon as the progress surpasses 99%, it reboots
>>> into SPI SPL without any error message.
>>> Using the old U-Boot fork I've tried flashing u-boot-spl.kwb - similarly
>>> it gets stuck in the SPL trying to boot on from SPI. After a while it
>>> reboots, and so on in a loop.
>>>
>>> So it seems that non-SPL U-Boot 2018.01 is broken, on my 2 GB model.
>>
>> Hmmm. I can only
> 
> I have reconfirmed (with our GCC 7.2.1) that no version works for me:
> 
> v2017.07 did not have turris_omnia defconfig yet
> b6ee860b87d8accedfb7a1fd5414f7c483603234 bad (adds turris_omnia)
> v2017.09 bad
> v2017.11 bad
> v2018.01 bad
> master bad (c4cb6e64bf068eaa1f7c96cb37da7ae6d40bbbff / arc merge)
> 
> So I have nothing to bisect and wonder how Marek tested this...
> 
>>> I also ran into a couple DDR3 training failures when booting via UART.
>>> No such training problems observed booting from SPI.
>>
>> Using the same U-Boot version, meaning same DDR init code?
> 
> Yes. Only difference being BOOT_FROM and the spl.c modification.
> 
>> Thats
>> strange. I have not really worked with Armada 38x for quite some
>> time, but I don't remember any such problems that you describe
>> here. I might need to dig up an Armada 38x board and give it a
>> try myself...
> 
> Regards,
> Andreas
> 


-- 
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)