help with mtd-test failures?

[Steve deRosier]

Thanks Josh,


On Sun, Jan 4, 2015 at 12:15 AM, Josh Wu <josh.wu at atmel.com> wrote:
> Hi, Steve
>
> On 1/3/2015 10:34 AM, Steve deRosier wrote:
>>
>> Hi all,
>>
>> In doing my due dillegence on my system, I decided to run the
>> mtd-tests. I'm getting a few failures that I'd like your help in
>> understanding.
>>
>> The one that worries me comes from mtd_oobtest:
>>
>>      # modprobe mtd_oobtest dev=7
>>      mtd_oobtest: error: verify failed at 0x880000
>>      mtd_oobtest: error: verify failed at 0xe80000
>>
>> The most worrysome thing about this is the fact that the result isn't
>> consistant. sometimes I get no failures, and sometimes they're at
>> different spots.
>
> I think this should be related with nand flash's bitflips. As atmel nand
> driver just use standard oob write/read functions from nand_base.c.
> That means there is no ecc protection for mtd_oobtest.
> So if there has any nand bitflip happens in the oob area, then mtd-oobtest
> will be failed.
>
> How often does the failures comes?
> I tested on my at91sam9g25ek board (nand is: 29F2G08AAD) about 20 times, no
> error happened.
>

On our board, it happens more or less every other run.  0x880000 is
most frequent, followed by a few others at random times.

Your other email and this one made me worry about bitflips in the OOB
area, so I decided to test those.  The results are on the other
thread, but the important one here is the checking what happens to a
bit flip on the ECC data.  I did this manually and the ECC code was
able to successfully correct for this problem. I'm calling that good!

Thanks,
- Steve