GPMI NAND Regression on i.MX6S

[Sean Nyekjaer]

Hi Frieder,

On Thu, Mar 17, 2022 at 01:59:07PM +0100, Frieder Schrempf wrote:
> Am 17.03.22 um 09:06 schrieb Frieder Schrempf:
> > Hi Tim,
> > 
> > Am 16.03.22 um 17:34 schrieb Tim Harvey:
> >> On Wed, Mar 16, 2022 at 7:09 AM Fabio Estevam <festevam at gmail.com> wrote:
> >>>
> >>> Adding Han Xu's NXP email on Cc.
> >>>
> >>> On Mon, Mar 14, 2022 at 10:31 AM Frieder Schrempf
> >>> <frieder.schrempf at kontron.de> wrote:
> >>>>
> >>>> Hello everyone,
> >>>>
> >>>> sorry to dig out an old thread, but the below patch which was applied
> >>>> upstream as 616f03dabacb causes a regression for me when trying to
> >>>> attach an UBI volume with U-Boot 2022.01 on a board with i.MX6 Solo and
> >>>> AMD/Spansion parallel NAND.
> >>>>
> >>>> The failure looks like this:
> >>>>
> >>>> ubi0: attaching mtd2
> >>>> ubi0 error: ubi_io_read: error -74 (ECC error) while reading 64 bytes
> >>>> from PEB 0:0, read 64 bytes
> >>>> ubi0 error: ubi_io_read: error -74 (ECC error) while reading 2048 bytes
> >>>> from PEB 0:2048, read 2048 bytes
> >>>> ubi0 error: ubi_io_read: error -74 (ECC error) while reading 64 bytes
> >>>> from PEB 1:0, read 64 bytes
> >>>> ubi0 error: ubi_io_read: error -74 (ECC error) while reading 2048 bytes
> >>>> from PEB 1:2048, read 2048 bytes
> >>>>
> >>>> The NAND as reported by Linux is:
> >>>>
> >>>> nand: device found, Manufacturer ID: 0x01, Chip ID: 0xdc
> >>>> nand: AMD/Spansion S34ML04G1
> >>>> nand: 512 MiB, SLC, erase size: 128 KiB, page size: 2048, OOB size: 64
> >>>>
> >>>> A different revision of the same board with a different NAND from
> >>>> manufacturer ESMT doesn't show the issue:
> >>>>
> >>>> nand: device found, Manufacturer ID: 0xc8, Chip ID: 0xdc
> >>>> nand: ESMT NAND 512MiB 3,3V 8-bit
> >>>> nand: 512 MiB, SLC, erase size: 128 KiB, page size: 2048, OOB size: 64
> >>>>
> >>>> When I revert the mentioned commit (see patch here: [1]), the UBI boot
> >>>> starts working again.
> >>>>
> >>>> Does anyone know what the problem is and how to properly solve it?
> >>>>
> >>>> Thanks for any help!
> >>>> Frieder
> >>>>
> >>>> [1]
> >>>> https://eur04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fzerobin.net%2F%3F57a57a322bbdcf3c%23rZa3vHlWi%2BRxtRomoljtrngqWwiv6v4Js%2F2LNfdV10o%3D&data=04%7C01%7Cfrieder.schrempf%40kontron.de%7C978979eaf1fa48171bab08da076aea84%7C8c9d3c973fd941c8a2b1646f3942daf1%7C0%7C0%7C637830453031595997%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=NGl%2FrFnffV2oC90js0aHyTvoZjnrzW%2FD830lOnR4TbQ%3D&reserved=0
> >>>>
> >>
> >> Frieder,
> >>
> >> I see the same issue here with IMX6Q/DL GPMI NAND.
> >>
> >> If I re-flash the ubi within U-Boot (tftpboot $loadaddr rootfs.ubi &&
> >> nand erase.part rootfs && nand write $loadaddr rootfs $filesize) I
> >> find that U-Boot can attach and mount the ubi fine but Linux will have
> >> issues
> > 
> > Interesting! This sounds like U-Boot and Linux somehow diverge in how
> > they handle the ECC data in OOB. I'm pretty confident that Linux does
> > things "correctly" and U-Boot should match what Linux does in this case.
> > 
> > Does the patch (revert of 616f03dabacb) I mentioned before "solve" the
> > issue for your case, too?
> > 
> > @Han, Ye, Peng: As you signed-off the mentioned commit, do you have any
> > ideas for a fix?
> 
> So the proper fix seems to be to revert to the "legacy" BCH layout that
> is used by Linux. Sean already tried to get this fixed almost a year ago
> [1] but Han turned the change down in favor of changing the layout on
> the kernel side. But this series [2] never made it upstream and it
> doesn't look like it will anytime soon.
> 
> I will try to resurrect Sean's fix.
> 
> [1]
> https://patchwork.ozlabs.org/project/uboot/patch/20210510100043.449294-1-sean@geanix.com/
> [2]
> https://patchwork.ozlabs.org/project/linux-mtd/patch/20210522205136.19465-1-han.xu@nxp.com/
> 

Thanks for bringing this up again. I have only had time to send Han a
ping a few times.
I would still prefer that we added the new(correct) metod as a option.

/Sean