Barebox / Kernel Omap ECC inconsistency?

[Miquel Raynal]

Hi Colin,

rogerq at kernel.org wrote on Wed, 2 Nov 2022 09:12:27 +0200:

> Hi Colin,
> 
> On 01/11/2022 21:09, Colin Foster wrote:
> > Hello,
> > 
> > I'm trying to revive a product that runs on a Phytec OMAP 4460 SOM. I
> > submitted a .dts RFC a month or so ago, and plan to perform the
> > suggestions and resubmit, but I'm up against one main hurdle that seems
> > to be related to flash OOB/ECC. (get_maintainers on
> > drivers/mtd/nand/raw/omap2.c is how I got this email list)
> > 
> > Barebox has "native" support for the Phytec SOM:
> > https://git.pengutronix.de/cgit/barebox/tree/arch/arm/boards/phytec-phycore-omap4460
> > 
> > It seems like Barebox is writing and expecting ECC bits to start at an
> > offset of 12 bytes, while the kernel (and Barebox comments suggest) the
> > ECC bytes should start at 2. I'm seeing this with
> > `nanddump -n -o -l 0x41000 -f mtdxnanddump /dev/mtdx`
> > 
> > Barebox created partition with UBI (mtd3)
> > ...
> > 00000800  ff ff ff ff ff ff ff ff  ff ff ff ff 56 49 fd 17
> > 00000810  b2 25 60 1a 42 1d eb 56  5d ff ff ff ff ff ff ff
> > ...
> > 
> > Kernel created partition with UBI (mtd4)
> > ...
> > 00000800  ff ff 07 73 04 ac 57 6b  9b 1f 92 49 ab e0 b9 ff
> > 00000810  ff ff ff ff ff ff ff ff  ff ff ff ff ff ff ff ff
> > ...
> > 
> > 
> > My question: 
> > 
> > Am I right to assume this is an issue in Barebox? Perhaps this is just a  
> 
> I'm guessing so. Both u-boot and Linux for OMAP put the ECC bytes right
> after the Bad block marker which is 2 bytes.

Yep. I checked, this has been like that since at least 2014, I don't
think we changed the layout in U-Boot/Linux "recently"... (I haven't
checked earlier, by laziness).

> https://elixir.bootlin.com/linux/latest/source/drivers/mtd/nand/raw/omap2.c#L1729
> https://elixir.bootlin.com/linux/latest/source/drivers/mtd/nand/raw/omap2.c#L134
> 
> > bug that has been fairly dormant for 15 years. If that is the case, I
> > assume there's probably no hope in getting this mainlined, and "native"
> > barebox support is just a ruse.
> > 
> > If that isn't the case, is there a hidden "shift OOB by 10" config
> > option that I'm missing? Or am I interpreting this data incorrectly?
> > 
> > 
> > Any suggestions would be greatly appreciated.  
> 
> You should fix the OMAP NAND driver/config in Barebox to match that
> with Linux OMAP NAND driver if you want them to run on the same system.

Agreed.

Thanks,
Miquèl