Dear Mr. Woodhouse,

I've found a bug in the functions serving the BLKRRPART 
ioctl() in the 2.4 branch of MTD blkdevs. See the attached 
patch, I believe it contains a fix.

My DiskOnChip was detected just fine, but when I flushed the 
original partition and issued a BLKRRPART, the device would 
report half its original size. If I did it again, it would
report a quarter size... etc.

Seems like the driver would re-init device size to a size 
denominated in 1k blocks, rather than sectors (512B).

Attached are two simple proggies that I'm using in my 
partitioning scripts.


Otherwise the driver seems to work just fine.
Actually I also had to insert 

#ifndef CONFIG_MTD_CFI_GEOMETRY
#define CONFIG_MTD_MAP_BANK_WIDTH_1
#endif

at the beginning of $KERNEL/include/linux/mtd/map.h

to make it compile. Seems like a dependency mismatch between 
drivers/mtd/chips/Config.in and the headers/code. I don't have 
a CFI MTD, but I do need drivers/mtd/chips/chipreg.o even for 
my DiskOnChip (now a generic NFTL device).


Thanks for the great job you're doing on MTD :-)

Frank Rysanek