[Qi] bad magic when booting from NAND
dmitry.kurochkin at gmail.com
Fri Apr 10 20:57:39 CEST 2009
On Fri, Apr 10, 2009 at 10:06 PM, Michael Trimarchi
<trimarchi at gandalf.sssup.it> wrote:
> Dmitry Kurochkin wrote:
>> I have found the problem. The error was because we check if the
>> current block is bad or the next one is bad. And skip the current
>> block even if only the next one is bad. This way we skip good block
>> before a bad one.
>> Attached is a patch to fix it. Besides I have removed 4 bad block
>> count limit. Instead I check if we read till the end of nand. Or at
>> least this is what this code supposed to do:
>> if (start_block512 >> 2 > BAD_BLOCK_OFFSET)
>> /* end of NAND */
>> return -1;
>> I am not sure this is the best and/or correct way to do the stop check.
>> A similar check for bad block ahead the current one was added to
>> u-boot in commit 4ec60973cf2e028de905f6ea1f81ef99ca60d834 by Michael:
>> Change the skip bad block. Avoid false positives by only checking
>> the beginning of a block and check the second page
> It's mine, but I don't know why Andy don't patch qi too.
U-Boot code is similar to qi but has some differences. In U-boot
nand_read_ll works with bytes while in Qi it works with 512-bytes
block. In U-Boot we check if a NAND block (64*2048) is bad. While in
qi the bad check is done for each page (2048).
So checking one page ahead the current one work in U-Boot since the
check itself is done only in the beginning of a block. But in qi
checking one page ahead leads to a false bad for page before bad. My
guess is that Qi check was wrongly copied from U-Boot.
Now I am pretty confident in my fix except for the stop condition
part. And tests confirm that. It would be nice if someone review the
patch and commit it.
>> Qi code does the same thing. But it seems to be wrong at least in my case.
More information about the openmoko-kernel