On Fri, 25 Aug 2017 00:11:29 +0200
Boris Brezillon <boris.brezil...@free-electrons.com> wrote:
> > +/*
> > + * The logical block number assigned to a physical block is stored in the
> > OOB
> > + * of the first page, in 3 16-bit copies with the following layout:
> > + *
> > + * 01234567 89abcdef
> > + * -------- --------
> > + * ECC BB xyxyxy
> > + *
> > + * When reading we check that the first two copies agree.
> > + * In case of error, matching is tried using the following pairs.
> > + * Reserved values 0xffff mean the block is kept for wear leveling.
> > + *
> > + * 01234567 89abcdef
> > + * -------- --------
> > + * ECC BB xyxy oob[8]==oob[10] && oob[9]==oob[11] -> byte0=8
> > byte1=9
> > + * ECC BB xyxy oob[10]==oob[12] && oob[11]==oob[13] -> byte0=10
> > byte1=11
> > + * ECC BB xy xy oob[12]==oob[8] && oob[13]==oob[9] -> byte0=12
> > byte1=13
>
> I know there's a depends on "MTD_NAND_SHARPSL || MTD_NAND_TMIO" in the
> Kconfig entry, but one could enable those options just to use the sharpsl
> part parser even if the OOB layout is incompatible with the sharpsl FTL.
>
> I'd recommend that you check that OOB bytes 8 to 15 are actually free
> to be used by the MTD user in sharpsl_parse_mtd_partitions().
>
> Can be done with something like that:
>
> static int sharpsl_nand_check_ooblayout(struct mtd_info *mtd)
> {
> u8 freebytes = 0;
> int section = 0;
>
> while (true) {
> struct mtd_oob_region oobfree = { };
> int ret, i;
>
> ret = mtd_ooblayout_free(mtd, section++, &oobfree);
> if (ret)
> break;
>
> if (!oobfree.length || oobfree.offset > 15 ||
> (oobfree.offset + oobfree.length) < 8)
> continue;
>
> i = oobfree.offset >= 8 ? : 8;
As you reported on IRC there's an mistake here, it should be:
i = oobfree.offset >= 8 ? oobfree.offset : 8
> for (; i < oobfree.offset + oobfree.length && i < 16; i++)
> freebytes |= BIT(i - 8);
>
> if (freebytes == 0xff)
> return 0;
> }
>
> return -ENOTSUPP;
> }
