[U-Boot] Problem with attaching UBI partition

[Marek Vasut]

On 03/03/2016 12:51 PM, Stefan Roese wrote:
> Hi Chin Liang,
> 
> On 02.03.2016 13:24, Chin Liang See wrote:
>>> On 01.03.2016 14:38, Chin Liang See wrote:
>>>> On Tue, 2016-03-01 at 08:23 +0100, Stefan Roese wrote:
>>>>> On 01.03.2016 07:53, Chin Liang See wrote:
>>>>>> On Mon, 2016-02-29 at 23:55 +0100, Marek Vasut wrote:
>>>>>>> On 02/24/2016 09:59 AM, Bakhvalov, Denis (Nokia - PL/Wroclaw)
>>>>>>> wrote:
>>>>>>>> Hello Heiko,
>>>>>>>>
>>>>>>>>> Which U-Boot version? Which board?
>>>>>>>>
>>>>>>>> This is U-Boot v2016.03-rc1. I have custom board with
>>>>>>>> socfpga
>>>>>>>> Arria5
>>>>>>>> onboard.
>>>>>>>>
>>>>>>>>> Where does this leading 0xff come from? There seems a
>>>>>>>>> problem
>>>>>>>>> with
>>>>>>>>> your spi nor flash driver?
>>>>>>>>
>>>>>>>> Yes, you're right. I have problems with the driver. As I
>>>>>>>> mentioned
>>>>>>>> in
>>>>>>>> previous mail when I read the contents from the flash some
>>>>>>>> data
>>>>>>>> is
>>>>>>>> corrupted.
>>>>>>>>
>>>>>>>> How to find out if the problem is because U-Boot has no
>>>>>>>> support
>>>>>>>> for
>>>>>>>> my flash (Spansion S25FL512S NOR flash with SPI) or I have
>>>>>>>> not
>>>>>>>> proper
>>>>>>>> configured SPI in U-Boot?
>>>>>>>
>>>>>>> I believe there is a problem with caches , which we still
>>>>>>> didn't
>>>>>>> identify. CCing Dinh and Chin, as they were the last ones
>>>>>>> looking
>>>>>>> into this problem.
>>>>>>>
>>>>>>> For now, try doing "dcache off" before using the QSPI NOR.
>>>>>>>
>>>>>>
>>>>>> We managed to get it work for socfpga. One of the issue dragged
>>>>>> me
>>>>>> long
>>>>>> is the CONFIG_SPI_FLASH_USE_4K_SECTORS. Need to ensure its not
>>>>>> defined
>>>>>> in the defconfig. Hope it helps
>>>>>
>>>>> This is not clear to me. You mean you were able to reproduce and
>>>>> solve this cache (S-bit) related issue by disabling
>>>>> CONFIG_SPI_FLASH_USE_4K_SECTORS? This option is disabled for all
>>>>> SoCFPGA board in mainline.
>>>>>
>>>>> Could you please explain in more details what exactly you did to
>>>>> solve this issue on SPI NOR?
>>>>>
>>>>
>>>> Nope, I am not replicating the issue. My board is having Micron
>>>> instead
>>>> of Spansion flash. In previously, CONFIG_SPI_FLASH_USE_4K_SECTORS
>>>> is
>>>> defined in socfpga as most serial NOR flash are supporting 4k sub
>>>> sector. With that, it would be good to ensure the 4K is undefined.
>>>>
>>>> In this case, wonder any details on the failure? Intermittent?
>>>
>>> I think we are talking about different things here. I'm referring
>>> to Marek mentioning the cache problem (S-bit related) on current
>>> SoCFPGA U-Boot mainline. With caches enabled you can experience
>>> problems on QSPI and on USB (reported from Marek - I didn't test
>>> USB yet).
>>>
>>
>> For QSPI, wonder the issue is on the read or write?
> 
> I'm not 100% sure here if this never happens on read. But it definitely
> happens on write. saveeenv to SPI NOR causes serious problems here
> for example.
>  
>>> So do you have any updated on this cache / S-bit problem?
>>
>> I am still debugging this as I notice the hub is not able to detect the
>> mass storage. This is only for certain pen drive.
>>
>>> Or can you
>>> use QSPI NOR without any issues on current mainline U-Boot with
>>> caches enabled on your platforms?
>>>
>>
>> The issue I have is on USB only. For NOR, I can ubifsmount and
>> ubifsload with Micron NOR flash without issue. This is with dcache
>> enabled
> 
> Strange. I have Micron here on the SR1500 as well:
> 
> => sf probe     
> SF: Detected N25Q128 with page size 256 Bytes, erase size 64 KiB, total 16 MiB
> 
> Thanks,
> Stefan
> 

The testcase is that you write some 8 MiB or so of data into the SPI
NOR. Upon read-back, you will notice that the last two or so bytes of
some sectors were not written completely and are corrupted.

-- 
Best regards,
Marek Vasut