Re: [PATCH v10 04/26] mtd: rawnand: sunxi: reject pages exceeding the ECC step limit
From: Miquel Raynal
Date: Thu Nov 19 2026 - 06:58:35 EST
On 05/11/2026 at 11:55:20 -07, James Hilliard <james.hilliard1@xxxxxxxxx> wrote:
> The controller has 16 ECC status slots on A10/A23 and 32 on H6/H616.
> Hardware-ECC setup checks the strength and available OOB space, but does
> not check that the page fits those slots. NAND identification can supply
> a larger page size, and the NAND core does not enforce this
> controller-specific limit.
>
> If enough OOB space is available, an unsupported geometry can pass attach
> and reach DMA code that indexes status and user-data registers by ECC
> step. H6/H616 also represent enabled data blocks with a 32-bit mask. The
> number of steps must fit the controller even when their user data fits
> the register bank.
>
> Reject excessive step counts before allocating user-data lengths or
> installing the hardware-ECC callbacks. Check the final step size, after
> strength maximization and the conversion from 512-byte to 1-KiB steps,
> so valid geometries retain their existing layout. This also keeps PIO
> fallback from accepting a page that the controller cannot handle.
>
> Fixes: 1fef62c1423b ("mtd: nand: add sunxi NAND flash controller support")
> Reported-by: Sashiko <sashiko-bot@xxxxxxxxxx>
> Closes: https://lore.kernel.org/r/sashiko-outbox-173724@xxxxxxxxxx
> Signed-off-by: James Hilliard <james.hilliard1@xxxxxxxxx>
> ---
> drivers/mtd/nand/raw/sunxi_nand.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/drivers/mtd/nand/raw/sunxi_nand.c b/drivers/mtd/nand/raw/sunxi_nand.c
> index 662d96fc2535..1a61f608ad91 100644
> --- a/drivers/mtd/nand/raw/sunxi_nand.c
> +++ b/drivers/mtd/nand/raw/sunxi_nand.c
> @@ -2223,6 +2223,11 @@ static int sunxi_nand_hw_ecc_ctrl_init(struct nand_chip *nand,
> ecc->bytes = ALIGN(ecc->bytes, 2);
>
> nsectors = mtd->writesize / ecc->size;
> + if (nsectors > nfc->caps->max_ecc_steps) {
> + dev_err(nfc->dev, "unsupported number of ECC steps: %d (max %u)\n",
> + nsectors, nfc->caps->max_ecc_steps);
> + return -EINVAL;
> + }
I'm fine with this patch, but if your manual says that the controller
support chips with pages up to 32k and you wire a chip -that does not
exist on earth yet- with a larger page size, screw you? This is again
one of those over defensive approaches...