Re: [PATCH] mtd: rawnand: stm32_fmc2: fix broken ECC

[Miquel Raynal <miquel.raynal@bootlin.com>]

Hi Christophe,

Christophe Kerello <christophe.kerello@st.com> wrote on Thu, 29 Oct
2020 17:38:12 +0100:

> Since commit d7157ff49a5b ("mtd: rawnand: Use the ECC framework user
> input parsing bits"), ECC are broken in FMC2 driver in case of
> nand-ecc-step-size and nand-ecc-strength are not set in the device tree.
> The default user configuration set in FMC2 driver is lost when
> rawnand_dt_init function is called. To avoid to lose the default user
> configuration, it is needed to move it in the new user_conf structure.
> 
> Signed-off-by: Christophe Kerello <christophe.kerello@st.com>
> Fixes: d7157ff49a5b ("mtd: rawnand: Use the ECC framework user input parsing bits")
> ---
>  drivers/mtd/nand/raw/stm32_fmc2_nand.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/mtd/nand/raw/stm32_fmc2_nand.c b/drivers/mtd/nand/raw/stm32_fmc2_nand.c
> index b31a581..dc86ac9 100644
> --- a/drivers/mtd/nand/raw/stm32_fmc2_nand.c
> +++ b/drivers/mtd/nand/raw/stm32_fmc2_nand.c
> @@ -1846,6 +1846,7 @@ static int stm32_fmc2_nfc_probe(struct platform_device *pdev)
>  	struct resource *res;
>  	struct mtd_info *mtd;
>  	struct nand_chip *chip;
> +	struct nand_device *nanddev;
>  	struct resource cres;
>  	int chip_cs, mem_region, ret, irq;
>  	int start_region = 0;
> @@ -1952,10 +1953,11 @@ static int stm32_fmc2_nfc_probe(struct platform_device *pdev)
>  	chip->options |= NAND_BUSWIDTH_AUTO | NAND_NO_SUBPAGE_WRITE |
>  			 NAND_USES_DMA;
>  
> -	/* Default ECC settings */
> +	/* Default ECC user settings */
>  	chip->ecc.engine_type = NAND_ECC_ENGINE_TYPE_ON_HOST;
> -	chip->ecc.size = FMC2_ECC_STEP_SIZE;
> -	chip->ecc.strength = FMC2_ECC_BCH8;
> +	nanddev = mtd_to_nanddev(mtd);
> +	nanddev->ecc.user_conf.step_size = FMC2_ECC_STEP_SIZE;
> +	nanddev->ecc.user_conf.strength = FMC2_ECC_BCH8;
>  
>  	/* Scan to find existence of the device */
>  	ret = nand_scan(chip, nand->ncs);

Sorry for breaking the driver with this change, but now I think we
should have all ECC related bits in ->attach() instead of ->probe().
The ->attach() hook is called during the nand_scan() operation and at
this point the chip's requirements/layout are known (not before). I
know that certain controllers don't really care about that, here your
simply hardcode these two fields and you don't need to know anything
about the chip's properties. But as a bid to harmonize all drivers with
the target of a generic ECC engine in mind, I think it's now time to
move these three lines (chip->ecc.* = ...) at the top of ->attach().
Also, these fields should have been populated by the core so perhaps
the best approach is to check if the user requirements are synced with
the controller's capabilities and error out otherwise?

We plan to send a fixes PR for -rc2, if the v2 arrives today I'll
integrate it.

Thanks,
Miquèl

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/