[U-Boot] IMX6 NAND boot regression

[Tim Harvey]

On Sat, Feb 2, 2019 at 5:30 AM Adam Ford <aford173 at gmail.com> wrote:
>
> On Sat, Feb 2, 2019 at 12:29 AM Jagan Teki <jagan at amarulasolutions.com> wrote:
> >
> > +Adam, Shyam
> >
> > On Sat, 2 Feb, 2019, 8:49 AM Stefan Agner <stefan.agner at toradex.com wrote:
> >
> > > Hi Tim,
> > >
> > > On 02.02.19 03:32, Tim Harvey wrote:
> > > > Stefan,
> > > >
> > > > I'm trying to track down an IMX6 SPL NAND boot regression that started
> > > > in v2018.07 with your patch series to mxs_nand.
> > >
> > > I am sorry about that. Unfortunately I did not had a design at hand where
> > > I was able to test the NAND driver in SPL...
> > >
> > > >
> > > > I bisected it back to '5346c31e305a37d39f535cc0d5ae87d8b7e81230 mtd:
> > > > nand: mxs_nand: use self init'. With this particular patch nand bbt
> > > > scanning would crash the board because of nand_chip.scan_btt not being
> > > > assigned. This was later fixed in
> > > > '96d0be07e7498e7174daa6f3b56fc807b9feb71d MTD: nand: mxs_nand_spl: Fix
> > > > empty function pointer for BBT' but cherry-picking that on top of
> > > > 5346c31 fixes the immediate crash while scanning but then I find that
> > > > mxs_read_page_ecc() hangs on the 4th page of reading u-boot.img from
> > > > the NAND. This gets worse 2 patches later where in
> > > > '28897e8d21f8e197e259a91c693de09cd81f2d5a: mtd: nand: mxs_nand: use
> > > > structure for BCH geometry' I find that the first byte of every page
> > > > read is wrong because mxs_nand_swap_block_mark() is getting called on
> > > > the page which swaps the first bytes with oob.
> > > >
> > > > There are several IMX6 boards out there using both NAND and SPL I
> > > > believe that I would assume were broken by this series. Any ideas on
> > > > the proper resolution?
> > >
> >
> > Look like 2017.03 can be stable boot from nand as for as my test is concern.
> >
> > We are also trying hard using git bisect, but seems like multiple breakings.
> >
> > Will keep posted if something move further.
>
>
> From a different thread, someone was able to test these patches and
> found they fixed their booting issues:
>
>  There was a broken function pointer here that was fixed and applied
>  the imx-master, but pending merge with master
>  http://patchwork.ozlabs.org/patch/1019440/

right, this one was the 96d0be07 merged upstream already. I'm not
clear how any NAND SPL would have worked after Sefan's series back in
June without this one.

>
>  Configure ECC from SPL here:
>  http://patchwork.ozlabs.org/patch/1020160/

For my boards with Micron  MT29F2G08AB 2Gbit NAND (oob=64) this was
enough but with Micron MT29F16G08AD 16Gbit (oob=224) and Cypress
S34ML16G202BH 16GBit (oob=128) NAND flash your other patch is needed.

>
>  Remove hard-coded ECC parameters since the patch above can autoset them.
>  http://patchwork.ozlabs.org/patch/1026638/
>

Thanks Adam, this one resolved the boot issue with Micron MT29F16G08AD
16Gbit (oob=224) and Cypress S34ML16G202BH 16Gbit (oob=128) NAND.

I will ack those last 2 patches that are still pending and take note
to avoid anything between v2018.07 and the next release v2019.03 for
IMX NAND SPL.

It kind of sucks that patches that affected IMX NAND SPL were accepted
with no testing on a NAND SPL board or cc's sent to maintainers of
boards using IMX NAND SPL but I understand the difficulty in even
identifying those maintainers and realize it is everyones
responsibility to test RC's on their own boards so I accept
responsibility for that. I'll make a note to remember that U-Boot is
broken for IMX NAND SPL between 2017.07 and 2019.03.

Thanks,

Tim