[PATCH 01/14] mtd: nand: brcm: switch to mtd_ooblayout_ops

[Linus Walleij]

On Thu, Jan 26, 2023 at 2:02 AM William Zhang
<william.zhang at broadcom.com> wrote:

> Unfortunately the u-boot nand base code still uses nand_ecclayout
> structure because it was based on old kernel nand driver.  Your change
> cause bugcheck in the nand_scan_tail at line 4978 when mtd->oobsize is
> not one of the default size (i.e. some large nand with BCH-8 ecc
> requirement and has 224 bytes oobsize per 4K page) because ecc->layout
> is never set. Also certainly any data built based on nand_cclayout like
> mtd->oobavail will not be correct.
>
> I actually converted back to nand_ecclayout structure from mtd_ooblayout
> with this fix to solve the above issues.
> Fixes: e365de90517b ("drivers: nand: brcmnand: fix nand_chip ecc layout
> structure")

Argh yeah I see. Let's hold this series off then.
It was worth a try!

> I can see your point to get the latest from the brcmnand linux kernel
> driver but this requires updating the u-boot nand base driver to use
> mtd_ooblayout as well and all others nand controller drivers too. I am
> not sure if this is something you want to tackle right now.

No I can't do that I do not have a big enough experience with NAND
flash and no testbed for it either.

> As far as I can see, all other oob/ecc layout setting patches you back
> ported in this series are not in the original brcmstb_choose_ecc_layout
> code you replaced. So I am not worried about that we must switch to
> mtd_ooblayout_ops at this point.  If indeed there is bug in
> brcmstb_choose_ecc_layout, we can port and convert the fix to
> nand_ecclayout structure from kernel code.

OK no they seemed to be mostly improvements of the algorithm so we
can certainly live without.

> Was the bug you were hunting down in the code related to this patch?

Actually not, I think, it's one of the other patches I sent, the one enabling
BCH-1 by reading the proper ECC properties from the device tree.
That made it finally work.

The iproc NAND driver I sent should also work pretty much as-is, nothing
depends on these backports.

Yours,
Linus Walleij