[RFC,v3 5/5] mtd: spinand: skip set/get oob data bytes when interleaved case

xiangsheng.hou xiangsheng.hou at mediatek.com
Fri Nov 26 00:51:51 PST 2021


Hi Miquel,

On Mon, 2021-11-22 at 10:01 +0100, Miquel Raynal wrote:
> Hi xiangsheng,
> 
> Thanks for the figures, they are useful too understand better your
> situation.
> 
> xiangsheng.hou at mediatek.com wrote on Fri, 12 Nov 2021 16:33:34
> +0800:
> > 
> > part1 and part2 called FDM(flash disk manage data) which can be
> > used to
> > store BBM or filesystem manage data(like jffs2).
> > 
> > The FDM and FDM ECC can be configurable,
> > FDM number of bytes: 0 ~ 8 bytes
> > FDM ECC number of byte: 0 ~ FDM size 
> > 
> > Therefore, the mtk ecc driver need to handle BBM swap and OOB shift
> > operation before/after the write/read operation in
> > ecc_finish/prepare_io_req.
> 
> Not necessarily. What if you just provide a translation in
> prepare/finish which basically switches from one representation to
> the
> other?
> 

For current design which mtd_ooblayout_set_databytes in
spinand_write_to_cache_op when MTD_AUTO_OOB_MODE can not work for BBM
swap situation.

In mtk nand_ecc_prepare_io_req, there have BBM swap lead to the
oobbuf[0] have one byte main data. The expected OOB data bytes will be
OOB availabe + 1 more byte. The mtd_ooblayout_set_databytes only set
OOB available bytes (will skip the reserve byte for BBM), lead to one
byte OOB data byte lost.

+-----------------------------+-----------+
|                             |           |
|                    *        |#@@@@@     |
|                             |           |
+-----------------------------+-----------+
(#): stand for one byte main data
(*): stand for the BBM
(@): the free OOB data byte

I will send next interation in the next few days, adjust set/get OOB
data byte operation in each ECC engine and only memcpy OOB buf in
write/read cache when AUTO OOB mode, which can both solve the this
situation and the AUTO OOB issue.

> Whatever operation you try to do, we don't really care where all
> these
> sections will be located once in memory, do we?

Yes, all the divergence need be handled at each ECC engine.

Thanks
Xiangsheng Hou


More information about the Linux-mediatek mailing list