Re: [PATCH] UBIFS: add tolerance to use variable writesize

[Artem Bityutskiy <dedekind1@gmail.com>]

On Wed, 2013-05-15 at 18:21 +0200, Dirk Behme wrote:
> From: Achim Dahlhoff <Achim.Dahlhoff@de.bosch.com>
> 
> UBIfs adapts it‘s structures according to the minimum writeSize reported
> by the MTD device. For NOR flash devices, this is normally 1. For NOR
> devices with internal hardware-ECC it might be more, such as the S-Die
> flash chips of Micron which can use an internal ECC if the chip is
> accessed in 32 byte chunks.
> 
> The UBIfs mount process checks and compares the writeSize set in the
> image and the writeSize reported from the /dev/mtd . UBIfs will fail
> to mount if the values differ. It should, though, not be a problem to
> mount an image which was created with a writeSize larger than that of
> the MTD, if it is larger by an integer factor.
> 
> This commit changes the check in a way so it will allow the image
> writeSize to be larger than that of the MTD by an integer factor.
> It will allow to create images and deploy them on different devices
> using different writeSizes. writeSize found using values of 1, 8 and 32.
> 
> Signed-off-by: Achim Dahlhoff <Achim.Dahlhoff@de.bosch.com>

Hi Dirk,

first of all, sorry for long delay.

In UBIFS we have 2 flash parameters which are relevant to this
discussion:

1. min_io_size - minimum size of the I/O buffer, comes from
mtd->writesize
2. max_write_size - optimzl size of the I/O buffer, comes from
mtd->writebufsize

Take a look at io.c.

So it looks like what you actually want to do is to make sure you have
the right max_write_size, in which case UBIFS will do most of its writes
in max_wirte_size units. E.g., make it to be 32.

This can be done by teaching your flash driver to export correct
'mtd->writebufsize'. My guess is that you do not do this, and it
defaults to 'mtd->writesize'.

-- 
Best Regards,
Artem Bityutskiy