[EXT] [PATCH] mtd: mxs_nand: default to legacy bch and rename to modern bch option

[Frieder Schrempf]

Hi Ye,

Am 17.03.22 um 14:54 schrieb Frieder Schrempf:
> Hi Han,
> 
> Am 17.03.22 um 14:33 schrieb Han Xu:
>>
>>
>> On Thu, Mar 17, 2022 at 8:27 AM Frieder Schrempf
>> <frieder.schrempf at kontron.de <mailto:frieder.schrempf at kontron.de>> wrote:
>>
>>     Hi Stefano,
>>
>>     this old patch was delegated to you in patchwork. If you're not the
>>     correct maintainer to address, please let me know. As the NAND layer
>>     seems to be unmaintained at the moment, I'm not sure whom to ask.
>>
>>     This patch fixes a regression that was introduced by 616f03dabacb ("
>>     mtd: gpmi: change the BCH layout setting for large oob NAND") which
>>     alters the BCH layout in a way that doesn't match with the
>>     implementation in the Linux kernel.
>>
>>     This causes failures when loading an UBI image in U-Boot that was
>>     flashed by Linux or vice versa (see [1]).
>>
>>     There has been an approach to fix this through an optional devicetree
>>     property in 51cdf83eea ("mtd: gpmi: provide the option to use legacy bch
>>     geometry"), but this is not acceptable. The "legacy" BCH layout
>>     compatible with Linux should be used by default.
>>
>>     The approach to upstream the "new" layout to the kernel [2] seems to be
>>     stalled and even if it would succeed, it would break systems that use an
>>     old U-Boot and a new kernel, which is again not really acceptable in my
>>     opinion.
>>
>>
>> Hi Frieder,
>>
>> I am not in office this week. I will send another patch set to change in
>> both kernel and u-boot to fix the compatible issue.
> 
> You already claimed that months ago, but nothing happened:
> 
>> I will send patches for both kernel and u-boot to use legacy bch
>> scheme by default, and add some code to treat few MLC nand chips as
>> corner cases.
> 
> Sean's U-Boot patch is effectively reverting the default behavior to use
> the "legacy" BCH scheme. So that's in line with what you want to do and
> you can base your work on top of this fix. But we should get the basic
> fix in regardless.
> 
> Even more so because switching the layout in U-Boot by using
> fsl,legacy-bch-geometry in the devicetree requires CONFIG_DM_MTD=y,
> which causes the bootloader size to increase by around 250 KiB in my
> case which might not be an option for some boards.

Ye's reply copied over to not break the thread:

> The dt nand driver will check "fsl,legacy-bch-geometry" property to
> use legacy bch. If this can't work for you in case you don't use DM
> driver, I prefer adding a config to select the legacy bch not
> reverting the patch.

I think you miss my point. I don't really care about your preferences.
But I do care about not introducing breaking changes. IMHO the "correct"
way to introduce the new BCH layout would have been:

1. Introduce the new layout in Linux and U-Boot behind a feature-flag
(e.g. DT property)
2. Make all upstream board configurations and DTs in Linux and U-Boot
use the feature-flag.
3. Drop the flag and make the new BCH layout the default.

That way the change would be much more controlled and raise the
awareness of board maintainers that are affected.

The current situation (layout changes applied in U-Boot without any
questioning, layout changes in Linux rejected for good reasons) leaves
everyone who uses the GPMI NAND on i.MX in a situation where they
probably will see failures and need to spend some hours of debugging
until they find out the reason is that someone carelessly introduced
breaking changes.

And even with the procedure described above we break compatibility
between old and new versions of U-Boot and Linux and create a dependency
between the two which is what we should try to avoid wherever possible.

Frieder