[PATCH RESEND u-boot-spi 0/8] Fix `mtd erase` when used with mtdpart

[Simon Glass]

Hi Marek,

On Wed, 14 Jul 2021 at 17:51, Marek Behún <marek.behun at nic.cz> wrote:
>
> Hello,
>
> I accidentally forgot to send this series to U-Boot's mailing list last
> time, meaning it did not end up in patchwork, so now I am resending it.
> Sorry for this mess.
>
> The original cover letter said:
>
> this patch series fixes the `mtd erase` command when used with mtdpart
> with a partition of non-zero offset.
>
> Currently when the `mtd erase` command is used for such a partition,
> it does not erase all blocks. Instead after a block is erased, the next
> block address not current block address + block size, but current block
> address + block size + partition offset, due to spi_nor_erase() not
> calling mtd_erase_callback():
>   => mtd erase "Rescue system"
>   Erasing 0x00000000 ... 0x006fffff (1792 eraseblock(s))
>   jedec_spi_nor spi-nor at 0: at 0x100000, len 4096
>   jedec_spi_nor spi-nor at 0: at 0x201000, len 4096
>   jedec_spi_nor spi-nor at 0: at 0x302000, len 4096
>   jedec_spi_nor spi-nor at 0: at 0x403000, len 4096
>   jedec_spi_nor spi-nor at 0: at 0x504000, len 4096
>   jedec_spi_nor spi-nor at 0: at 0x605000, len 4096
>   jedec_spi_nor spi-nor at 0: at 0x706000, len 4096
>
> This series adds some fixes to spi_nor_erase() function, then adds
> calling of mtd_erase_callback() to fix this bug.
>
> The series also contains an improvement - adding the posibility to
> interrupt spi_nor_erase() with Ctrl+C; and another one - making mtdpart's
> _erase() method more sane so that the above mentioned bug will not occur
> even if underlying driver does not call mtd_erase_callback().
>
> Moreover I would also like to start a discussion regarding the MTD
> subsystem:
> - U-Boot's MTD subsystem is based on Linux's, with many #ifdef __U_BOOT__
>   macros
> - this was done to make it easier to port Linux's patches to U-Boot
> - the problem is that it seems nobody did port Linux's MTD patches to
>   U-Boot for a long time, the code is many times different from Linux',
>   and it would be very hard to align it
> - therefore I propose to get rid of all the #ifdefs, remove the Linux
>   specific code, and continue developing the code independently from
>   Linux. This would make it impossible to apply Linux patches in some
>   kind of automatic way, but this is currently already impossible
>   anyway
> What do you guys think?

That sounds good to me, but I wonder what is the gap today?

Are you and/or Jagen going to take on this effort?

Regards,
Simon