Re: [PATCH v3] mtd: rawnand: brcmnand: correctly verify erased pages

[Ãlvaro FernÃndez Rojas]

Hi MiquÃl,


El mar., 12 may. 2020 a las 9:34, Miquel Raynal
(<miquel.raynal@xxxxxxxxxxx>) escribiÃ:
>
> Hi Ãlvaro,
>
> Ãlvaro FernÃndez Rojas <noltari@xxxxxxxxx> wrote on Tue, 12 May 2020
> 09:24:32 +0200:
>
> > Hi MiquÃl
> >
> > > El 12 may 2020, a las 9:16, Miquel Raynal <miquel.raynal@xxxxxxxxxxx> escribiÃ:
> > >
> > > Hi Ãlvaro,
> > >
> > > Ãlvaro FernÃndez Rojas <noltari@xxxxxxxxx> wrote on Tue, 12 May 2020
> > > 08:51:11 +0200:
> > >
> > >> The current code checks that the whole OOB area is erased.
> > >> This is a problem when JFFS2 cleanmarkers are added to the OOB, since it will
> > >> fail due to the usable OOB bytes not being 0xff.
> > >> Correct this by only checking that data and ECC bytes aren't 0xff.
> > >>
> > >> Fixes: 02b88eea9f9c ("mtd: brcmnand: Add check for erased page bitflips")
> > >> Signed-off-by: Ãlvaro FernÃndez Rojas <noltari@xxxxxxxxx>
> > >> ---
> > >> v3: Fix commit log and merge nand_check_erased_ecc_chunk calls.
> > >> v2: Add Fixes tag
> > >>
> > >> drivers/mtd/nand/raw/brcmnand/brcmnand.c | 19 ++++++++++++++-----
> > >> 1 file changed, 14 insertions(+), 5 deletions(-)
> > >>
> > >> diff --git a/drivers/mtd/nand/raw/brcmnand/brcmnand.c b/drivers/mtd/nand/raw/brcmnand/brcmnand.c
> > >> index e4e3ceeac38f..80fe01f03516 100644
> > >> --- a/drivers/mtd/nand/raw/brcmnand/brcmnand.c
> > >> +++ b/drivers/mtd/nand/raw/brcmnand/brcmnand.c
> > >> @@ -2018,8 +2018,9 @@ static int brcmnand_read_by_pio(struct mtd_info *mtd, struct nand_chip *chip,
> > >> static int brcmstb_nand_verify_erased_page(struct mtd_info *mtd,
> > >>              struct nand_chip *chip, void *buf, u64 addr)
> > >> {
> > >> +  struct mtd_oob_region oobecc;
> > >>    int i, sas;
> > >> -  void *oob = chip->oob_poi;
> > >> +  void *oob;
> > >>    int bitflips = 0;
> > >>    int page = addr >> chip->page_shift;
> > >>    int ret;
> > >> @@ -2035,11 +2036,19 @@ static int brcmstb_nand_verify_erased_page(struct mtd_info *mtd,
> > >>    if (ret)
> > >>            return ret;
> > >>
> > >> -  for (i = 0; i < chip->ecc.steps; i++, oob += sas) {
> > >> +  for (i = 0; i < chip->ecc.steps; i++) {
> > >>            ecc_chunk = buf + chip->ecc.size * i;
> > >> -          ret = nand_check_erased_ecc_chunk(ecc_chunk,
> > >> -                                            chip->ecc.size,
> > >> -                                            oob, sas, NULL, 0,
> > >> +
> > >> +          if (mtd->ooblayout->ecc(mtd, i, &oobecc)) {
> > >
> > > Please use the mtdcore.c's helpers
> > > (mtd_ooblayout_set/get_data/free/ecc/bytes).

Ok, I will use mtd_ooblayout_ecc function.

> > >
> > > Also, what are you trying to discriminate with the return code of the
> > > function? Shouldn't this function "always" work?
> >
> > Just making sure it doesnât return an ERANGE in case chip->ecc.size doesnât match the sections from mtd->ooblayout->ecc, which shouldnât happen, so I think we can remove that...
>
> The style we prefer for error checking is:
>
>         ret = function();
>         if (ret)
>                 do someting;
>
> instead of:
>
>         if (function())
>
> Anyway, I really don't know if it can happen or not. I suppose it does.
> What I don't understand is your "oob = chip->oob_poi + oobecc.offset".
> If you expect an error, then you should not update this pointer, right?

After switching to mtd_ooblayout_ecc, error checking isn't needed anymore.

>
> Don't you need to use 2 * i instead of i here? Following your other
> contribution, sections are distributed like "data/ecc/data/ecc/etc".

No, we're checking ECC bytes in the OOB, not about usable bytes in the
OOB area, which is what my other patch changes.

>
> >
> > >
> > >> +                  oob = NULL;
> > >> +                  oobecc.length = 0;
> > >> +          } else {
> > >> +                  oob = chip->oob_poi + oobecc.offset;
> > >> +          }
> > >> +
> > >> +          ret = nand_check_erased_ecc_chunk(ecc_chunk, chip->ecc.size,
> > >> +                                            oob, oobecc.length,
> > >> +                                            NULL, 0,
> > >>                                              chip->ecc.strength);
> > >
> > > As I told you, this helper takes "maid data" then "spare area" then
> > > "ecc bytes". The names are pretty important here as you want to avoid
> > > checking the spare OOB bytes on purpose, so maybe you could have more
> > > meaningful names and call "ecc" instead of "oob" the ecc region?
> >
> > Actually I thought you meant the commit log, not the code itself...
>
> No problem ;) I meant both actually, And I think you should name the
> oob pointer ecc_bytes.
>
> >
> > >
> > >>            if (ret < 0)
> > >>                    return ret;
> > >
> > >
> > > Thanks,
> > > MiquÃl
> >
> > Regards,
> > Ãlvaro.
> >
>
>
>
>
> Thanks,
> MiquÃl

Regards,
Ãlvaro.