On Tue, 2011-12-20 at 10:42 -0800, Mike Dunn wrote:
> +		/*
> +		 * For unpartitioned devices, create a single "partition" that
> +		 * spans the entire device, so that driver methods go through
> +		 * partition wrappers in all cases.
> +		 */
> +		struct mtd_partition single_part = {
> +			.name = (char *)mtd->name,
> +			.offset = 0,
> +			.size = mtd->size,
> +			.mask_flags = 0,
> +			.ecclayout = mtd->ecclayout,
> +		};
> +		err = add_mtd_partitions(mtd, &single_part, 1);
>  	}

There is a problem with this approach :-(

Look at the 'mtd_blkpg_ioctl()' function which is used to re-partition
MTD device from user-space. It is a bit strange, but we use block device
for re-partition in order to not invent new ioctls.

This function has this code:

                /* Only master mtd device must be used to add partitions */
                if (mtd_is_partition(mtd))
                        return -EINVAL;

and your patch  will brake re-partitioning.

So I guess we should go back to your first patch where you just change
the ->read() / ->read_oob() interface.

Or:

1. Rework whole MTD interface from ->read()/->write()/etc (pointers to
functions) to mtd_read(mtd, ...)/mtd_write(mtd, ...)/etc.

2. In the new functions you may do the stuff you do at patch 2.

For me the second approach is more plausible.

-- 
Best Regards,
Artem Bityutskiy