Re: [PATCH] mtd: rawnand: xway: No hardcoded ECC engine for Micron Chips

[Miquel Raynal]

Hi Daniel,

Daniel Kestrel <kestrelseventyfour@xxxxxxxxx> wrote on Tue, 3 Aug 2021
16:32:56 +0200:

> Some lantiq xway devices use Micron NAND chips, which use on-die ECC.
> The hardcoded setting of NAND_ECC_ENGINE_TYPE_SOFT makes them unusable,
> because the software ECC on top of the hardware ECC produces errors for
> every read and write access, not to mention that booting does not work,
> because the boot loader uses the correct ECC when trying to load the
> kernel and stops loading on severe ECC errors.
> Removing the hardcoded settings would break a number of devices that
> work with those settings.
> Adding a DTB property was considered, but did not work, because devices
> of the same type but from different manufacture dates have different
> NAND chips and as such it is not possible to determine the NAND chip
> in advance or device specific.

I understand the problem and it is a very crappy situation.

> 
> Signed-off-by: Daniel Kestrel <kestrelseventyfour@xxxxxxxxx>
> ---
>  drivers/mtd/nand/raw/xway_nand.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/mtd/nand/raw/xway_nand.c b/drivers/mtd/nand/raw/xway_nand.c
> index 26751976e502..20cb5ce2f3b0 100644
> --- a/drivers/mtd/nand/raw/xway_nand.c
> +++ b/drivers/mtd/nand/raw/xway_nand.c
> @@ -10,6 +10,7 @@
>  #include <linux/of_platform.h>
>  
>  #include <lantiq_soc.h>
> +#include "internals.h"
>  
>  /* nand registers */
>  #define EBU_ADDSEL1		0x24
> @@ -148,7 +149,8 @@ static void xway_write_buf(struct nand_chip *chip, const u_char *buf, int len)
>  
>  static int xway_attach_chip(struct nand_chip *chip)
>  {
> -	chip->ecc.engine_type = NAND_ECC_ENGINE_TYPE_SOFT;
> +	if (chip->manufacturer.desc->id != NAND_MFR_MICRON)
> +		chip->ecc.engine_type = NAND_ECC_ENGINE_TYPE_SOFT;

Could we make this a little bit clever with something like this:
https://elixir.bootlin.com/linux/v5.13-rc7/source/drivers/mtd/nand/raw/nand_micron.c#L434

This is far from ideal, there should definitely be a change in the DT.
But given your initial comments I guess it is not possible.

Anyway I don't find a better way as, during the attach() call, we don't
yet ran the manufacturer code, hence we don't know if on-die ECC is
actually available or not.

Thanks,
Miquèl