Re: [PATCH] mtd: rawnand: ams-delta: Drop board specific partition info

[Janusz Krzysztofik]

Hi Aaro,

Thanks for your review.

On Wednesday, March 20, 2019 2:16:30 AM CET Aaro Koskinen wrote:
> On Tue, Mar 19, 2019 at 11:37:18PM +0100, Janusz Krzysztofik wrote:
> > After recent modifications, only a hardcoded partition info makes
> > the driver device specific.  Other than that, the driver uses GPIO
> > exclusively and can be used on any hardware.
> > 
> > Drop the partition info and use MTD partition parser with default
> > list of partition types instead.
> > 
> > Amstrad Delta users should append the followig partition info to their
>                                         ^^^^^^^^
> Should be "following".
> 
> > kernel command line, possibly by embedding it in CONFIG_CMDLINE:
> > mtdparts=ams-delta-nand:3584k(Kernel),256k(u-boot),256k(u-boot_params),\
> > 256k(Amstrad_LDR),27m(File_system),768k(PBL_reserved).  For their
> > convenience, select CONFIG_MTD_CMDLINE_PARTS  symbol from that board
> > Kconfig automatically if this NAND driver is also selected.
> > 
> > Signed-off-by: Janusz Krzysztofik <jmkrzyszt@xxxxxxxxx>
> > Cc: Tony Lindgren <tony@xxxxxxxxxxx>
> 
> Could we move the fixed partition setup to the board file
> instead? Otherwise this kind of change is not really nice for the users,
> as it will likely break existing setups. The default partition layout
> should remain the same.

I'm wondering if it would be acceptable to pass partition info from a .dts 
file.  I think that would be a better, more modern approach than adding a new 
header under include/linux/platform_data.

The problem with a device tree based implementation is, I know of no u-boot 
version supporting both Amstrad Delta and FDT.  However, I've already tested 
two solutions that work for me.

One uses CONFIG_ARM_APPENDED_DTB and requires a user to manually append the 
blob to zImage and (re)generate uImage.  I'm not sure how much more user-
friendly it looks for you, compared to the command line version I proposed 
initially.

If the above is not acceptable. I can propose still another approach. The blob 
is automagically built and embedded into the kernel with some assembler glue, 
then unflattened from the board init_machine(), somehow similar to the way 
drivers/of/unittest.c does it.

Please advise which approach sounds best to you (platform_data, 
CONFIG_ARM_APPENDED_DTB or unittest like).

Thanks,
Janusz


> 
> A.
>