[PATCH 2/2] ARM: mvebu: x530: Add option for ECC

[Stefan Roese]

On 1/11/22 02:37, Chris Packham wrote:
> On Mon, Jan 10, 2022 at 9:10 PM Chris Packham <judge.packham at gmail.com> wrote:
>>
>>
>>
>> On Mon, 10 Jan 2022, 8:37 PM Stefan Roese, <sr at denx.de> wrote:
>>>
>>> On 1/6/22 04:08, Chris Packham wrote:
>>>> Some older x530 boards have layout issues that cause problems for DDR.
>>>> These are usually seen as training failures but can also cause problems
>>>> after training has completed. Add an option to enable ECC leaving the
>>>> default as N which will work with both old and new boards.
>>>>
>>>> Signed-off-by: Chris Packham <judge.packham at gmail.com>
>>>
>>> Just checking: Most likely you don't have the possibility to detect the
>>> board version at runtime?
>>
>>
>> Not in a way that can be done without working RAM. For our internal fork we maintain 2 defconfigs one with ECC one without. We can manage via the BOM which boards get which bootloader.
>>
>>>
>>> If this is not possible:
>>>
>>> Reviewed-by: Stefan Roese <sr at denx.de>
>>>
>>> Thanks,
>>> Stefan
>>>
>>>> ---
>>>>
>>>>    arch/arm/mach-mvebu/Kconfig      |  1 +
>>>>    board/alliedtelesis/x530/Kconfig | 20 ++++++++++++++++++++
>>>>    board/alliedtelesis/x530/x530.c  |  8 +++++++-
>>>>    3 files changed, 28 insertions(+), 1 deletion(-)
>>>>    create mode 100644 board/alliedtelesis/x530/Kconfig
>>>>
>>>> diff --git a/arch/arm/mach-mvebu/Kconfig b/arch/arm/mach-mvebu/Kconfig
>>>> index d23cc0c760f1..7388ade98d52 100644
>>>> --- a/arch/arm/mach-mvebu/Kconfig
>>>> +++ b/arch/arm/mach-mvebu/Kconfig
>>>> @@ -341,5 +341,6 @@ config SECURED_MODE_CSK_INDEX
>>>>
>>>>    source "board/solidrun/clearfog/Kconfig"
>>>>    source "board/kobol/helios4/Kconfig"
>>>> +source "board/alliedtelesis/x530/Kconfig"
>>>>
>>>>    endif
>>>> diff --git a/board/alliedtelesis/x530/Kconfig b/board/alliedtelesis/x530/Kconfig
>>>> new file mode 100644
>>>> index 000000000000..5c1ae36aebaa
>>>> --- /dev/null
>>>> +++ b/board/alliedtelesis/x530/Kconfig
>>>> @@ -0,0 +1,20 @@
>>>> +menu "x530 configuration"
>>>> +     depends on TARGET_X530
>>>> +
>>>> +config X530_ECC
>>>> +     bool "Enable DDR3 ECC"
>>>> +     help
>>>> +       Some of the older x530 board have layout issues which cause problems
>>>> +       for the DDR which usually exhibit as DDR training failures or
>>>> +       problems accessing DDR after training.
>>>> +
>>>> +       The known affected boards are:
>>>> +
>>>> +       * 844-001897-00 (x530-28GTXm, x530-28GPXm, GS980MX/28PSm)
>>>> +       * 844-001948-00 (GS980MX/28)
>>>> +       * 844-002008-00 (x530L-52GTX, x530L-52GPX)
>>>> +       * 844-001974-00 (x530-52GTXm, x530-52GPXm, GS980MX/52PSm)
>>>> +
>>>> +       If you have a newer board you can set Y here, otherwise say N.
>>>> +
>>>> +endmenu
>>>> diff --git a/board/alliedtelesis/x530/x530.c b/board/alliedtelesis/x530/x530.c
>>>> index 866b6e68cc16..de20684f4353 100644
>>>> --- a/board/alliedtelesis/x530/x530.c
>>>> +++ b/board/alliedtelesis/x530/x530.c
>>>> @@ -45,6 +45,12 @@ int hws_board_topology_load(struct serdes_map **serdes_map_array, u8 *count)
>>>>        return 0;
>>>>    }
>>>>
>>>> +#if CONFIG_IS_ENABLED(X530_ECC)
>>>> +     #define BUS_MASK BUS_MASK_32BIT_ECC
>>>> +#else
>>>> +     #define BUS_MASK BUS_MASK_32BIT
>>>> +#endif
>>>> +
> 
> Hmm this doesn't actually appear to be working as intended. When the
> SPL is built I end up with CONFIG_IS_ENABLED(X530_ECC) == 0 even when
> I set it to y. Do I need to define a SPL version of this Kconfig
> symbol?

AFAIK, IS_ENABLED(CONFIG_X530_ECC) would be a solution here. There is
no need for a new SPL Kconfig option this way. Please give it a try.

Thanks,
Stefan