mtd_nandbitserr and seed using pl35x NAND controller

[Miquel Raynal]

Hello Andrea,

On 25/11/2025 at 10:12:47 GMT, Andrea Scian <andrea.scian at dave.eu> wrote:

> Dear all,
>
> I'm facing a weird behavior on mtd_nandbitserr mtd_test while working on a Zynk7k based platform, running kernel 6.6.0
>
> As many of you suggest me, this test is part of our validation for the whole lowlevel support (NAND chip, hardware design, NAND controller).
>
> I've found that this test, in my testbed, fails when using seed=1 but it seems to end successfully when using default parameters (so seed=0)
>
> Here it is the test output in the two cases:
>
> root at sw0005-devel:~# modprobe mtd_nandbiterrs seed=1

...

> --- a/drivers/mtd/nand/raw/pl35x-nand-controller.c
> +++ b/drivers/mtd/nand/raw/pl35x-nand-controller.c
> @@ -1064,7 +1064,7 @@ static int pl35x_nand_chip_init(struct pl35x_nandc *nfc,
>       plnand->cs = cs;
>  
>       chip = &plnand->chip;
> -     chip->options = NAND_BUSWIDTH_AUTO | NAND_USES_DMA | NAND_NO_SUBPAGE_WRITE;
> +     chip->options = NAND_BUSWIDTH_AUTO | NAND_USES_DMA;

I no longer remember why this flag was set, tbh. If you enable subpage
writes, you should make sure that page and subpage tests really pass,
because it means I have not tested this possibility.

Nevertheless, I find this causal effect very unsatisfying. I do not
understand how this might be related to your issue.

Are you sure of the ECC configuration (can you show the DT snippet)?

Is the issue consistent?

> * is it correct that changing seed in mtd_nandbiterrs make such a
> difference? IIUC this is just a "random" seed so I'll expect to fail
> in both cases or none

Yes, exactly. I can't make sense out of this observation :-)

> * is it correct that subpage write has been disabled in pl35x NAND
> controller? AFAIK on older kernel subpages were enable but maybe I'm
> missing something. Unfortunately git history doesn't help in this

This flag has always been there, so what does that mean? Aren't you
talking about a Xilinx downstream kernel here?

> I didn't have the chance to try a newer kernel but, at least on these
> two files, I don't see any significant changes

Did you observe what seed=0/1 meant in practice? I clearly do not
understand the link it might have with subpages.

Can you try with the mtd-utils userspace tool instead? Kernel modules
are no longer actively maintained, so maybe there is a bug (?) inside.

Also, can you try manually flipping bits with nandflipbits and target
various area in the page (outside of the first subpage, unlike the
nandbiterrs test).

Thanks,
Miquèl