[PATCH] omap3_beagle: Change NAND ECC scheme back to OMAP_ECC_HAM1_CODE_HW

[Adam Ford]

On Sun, Dec 22, 2019 at 8:28 AM Tom Rini <trini at konsulko.com> wrote:
>
> On Sun, Dec 22, 2019 at 08:24:14AM -0600, Derald D. Woods wrote:
> > On Sun, Dec 22, 2019 at 09:10:50AM -0500, Tom Rini wrote:
> > > On Sun, Dec 22, 2019 at 09:46:27AM +0100, Patrik Dahlstrom wrote:
> > > > On 12/22/19 4:24 AM, Derald D. Woods wrote:
> > > > > On Sat, Dec 21, 2019 at 05:18:22PM +0100, Patrik Dahlström wrote:
> > > > >> The omap3_beagle NAND ECC scheme was changed in 4b37928d357 for unknown
> > > > >> reasons, leading to uncorrectible ecc errors. This commit changes it
> > > > >> back to what it was before.
> > > > >>
> > > > >
> > > > > Hello Patrick,
> > > > >
> > > > > Is there a setup/test that you are using for OMAP_ECC_HAM1_CODE_HW? I
> > > > > just want to give it a try. I have three OMAP3 boards, with NAND, that
> > > > > I would like to test.
> > > >
> > > > I'm using a BeagleBoard rev. C4 (yes, the one that came out in 2009)
> > > > for testing.
> > > >
> > > > >
> > > > > I also see that the SYS_NAND_ECC_BYTES should have been changed to '14'
> > > > > per the 'doc/README.nand' for OMAP_ECC_BCH8_CODE_HW_DETECTION_SW. This
> > > > > may be the issue with this particular ECC scheme.
> > > > >
> > > > I based my changes on reverting 4b37928d357, which did this:
> > > >
> > > > <snip>
> > > >  +#define CONFIG_NAND_OMAP_ECCSCHEME      OMAP_ECC_BCH8_CODE_HW_DETECTION_SW
> > > > <snip>
> > > >  -#define CONFIG_NAND_OMAP_ECCSCHEME     OMAP_ECC_HAM1_CODE_HW
> > > > <snip>
> > > >
> > > > Based on your comment, I tested this configuration:
> > > >
> > > >  #define CONFIG_SYS_NAND_ECCBYTES        14
> > > >  #define CONFIG_NAND_OMAP_ECCSCHEME      OMAP_ECC_BCH8_CODE_HW_DETECTION_SW
> > > >
> > > > It worked to boot from SD card (only had to do saveenv), but only
> > > > partly from NAND:
> > > >
> > > >  U-Boot SPL 2020.01-rc5-00006-gda26dd89c8-dirty (Dec 22 2019 - 09:26:14 +0100)
> > > >  Trying to boot from NAND
> > > >
> > > > Then it just hangs. Here's how I flashed SPL and U-Boot:
> > > >
> > > >  mmc rescan
> > > >  fatload mmc 0 80000000 MLO
> > > >  nand erase 0 80000
> > > >  nandecc hw
> > > >  cp.b 80000000 80020000 20000
> > > >  cp.b 80000000 80040000 20000
> > > >  cp.b 80000000 80060000 20000
> > > >  nand write 80000000 0 80000
> > > >  fatload mmc 0 80000000 u-boot.img
> > > >  nand erase 80000 160000
> > > >  nand write 80000000 80000 160000
> > > >
> > > > I then tried adjusting the "nandecc hw" line, but here's how that went:
> > > >
> > > >  BeagleBoard # nandecc hw bch8
> > > >  nand: error: CONFIG_NAND_OMAP_ELM required for ECC
> > > >
> > > > Unfortunately CONFIG_NAND_OMAP_ELM is not available for OMAP34XX.
> > >
> > > Yeah, ugh, I'm sorry I didn't catch this at the time (my Beagleboard is
> > > old and I worry about killing the NAND but I guess I need to move it to
> > > manual testing a few releases a year).  For the beagleboard I believe
> > > the right answer is to move back to the ECC scheme that it was before as
> > > that's what's deployed and used by anyone that's using the boards.  If
> > > memory serves there is a way to switch to doing SW and BCH8 but that's
> > > not going to be a win for these boards both for speed and for the
> > > incompatibility.
> > >
> >
> > Tom,
> >
> > Are you going to modify the remaining affected OMAP34XX boards? Or will this
> > patchset be expanded?
>
> Lets add in a few more maintainers.  From memory, there are reasons to
> move to BCH8 on omap3 platforms if for example you're moving to newer
> NAND chips that in turn require it.  If you want to keep the EVM on
> BCH8, that's fine, I don't know much about the overall user base there.
> But I do know the Beagleboard one :)

I know the Logic PD Torpedo and SOM LV boards use the BCH8 HW
detection with SW Correction because the omap34/35 had a bit with
4-bit and 1-bit wasn't sufficient.  Logic PD has some medical and
military users so having the 8-bit is preferred.  I haven't checked
lately, but to my knowledge, the Torpedo and SOM-LV boards have been
working fine with 8-bit.  I haven't changed them, so unless something
happened to the driver, I'd prefer to keep the various omap3_logic
boards as 8-bit.

I know some of the Micron flash parts have available on-chip ECC, but
I haven't tried to use them and I don't know of those drivers are
enabled in U-Boot.  That might be an option for some people who want
more than 1-bit without the overhead of the software correction on
devices without ELM.

adam
>
> --
> Tom