UBIFS abnormal unaligned access at OneNAND 4KiB page

[Adrian Hunter]

Kyungmin Park wrote:
> On Mon, Aug 3, 2009 at 4:51 PM, Adrian Hunter<adrian.hunter at nokia.com> wrote:
>> Kyungmin Park wrote:
>>> On Mon, Aug 3, 2009 at 3:40 PM, Adrian Hunter<adrian.hunter at nokia.com>
>>> wrote:
>>>> Kyungmin Park wrote:
>>>>> On Mon, Aug 3, 2009 at 3:29 PM, Adrian Hunter<adrian.hunter at nokia.com>
>>>>> wrote:
>>>>>> Kyungmin Park wrote:
>>>>>>> Now I'm working with 4KiB pagesize OneNAND, I got problem with
>>>>>>> unaligned byte align again.
>>>>>>>
>>>>>>> <7>UBIFS DBG (pid 1): read_znode: LEB 78:205824, level 0, 8 branch
>>>>>>> <7>UBIFS DBG (pid 1): ubifs_read_node: LEB 78:206016, indexing node,
>>>>>>> length 48
>>>>>>> onenand_mlc_read_ops_nolock[1108] buf c41bda80 1216 48
>>>>>>> <7>UBIFS DBG (pid 1): read_znode: LEB 78:206016, level 0, 1 branch
>>>>>>> onenand_mlc_read_ops_nolock[1108] buf c40a68c0 441 13
>>>>>>> onenand_read_bufferram[674] area 0x400, buffer c40a68c0, offset 441,
>>>>>>> count
>>>>>>> 13
>>>>>>> <1>Unhandled fault: external abort on non-linefetch (0x1008) at
>>>>>>> 0xc58805ba
>>>>>>>
>>>>>>> In the previous time, we got the similar case. but the problem gone
>>>>>>> with some patch. but it happens again.
>>>>>>>
>>>>>>> Do you have any idea?
>>>>>> It is up to your driver to meet your bus requirements.  If the bus
>>>>>> requires
>>>>>> word (2 bytes) aligned accesses then you must check the alignment in
>>>>>> the
>>>>>> driver and read the ends words separately.
>>>>> It's onenand_base.c. Only change is page size is 4KiB. others are same.
>>>>>
>>>>>> UBIFS does make unaligned reads to the LPT e.g. reading 13 bytes at
>>>>>> offset
>>>>>> 441
>>>>> I think so. but now I got the these unaligned reads.
>>>>> That's mystery
>>>> Well, onenand_read_bufferram seems to check only the 'count', not the
>>>> 'offset'
>>> I added the handling of offset also
>> Show me
> 
> +       if (ONENAND_CHECK_BYTE_ACCESS(offset)) {
> +               unsigned short word;
> +
> +               printk("%s[%d] area 0x%x, buffer %p, offset %d, count %d\n", __f
> +
> +               /* Read aligned word(16-bit) size */
> +               word = this->read_word(bufferram + offset - 1);
> +               buffer[0] = (word & 0xff00) >> 8;
> +
> +               buffer++;
> +               offset++;

Is it correct to increase the buffer and the offset?


> +               count--;
> +               printk("%s[%d] area 0x%x, buffer %p, offset %d, count %d\n", __f
> +       }
>