Linux mdadm --grow RAID6: Something wrong - reshape aborted

I have a RAID60 that I want to expand.

The current is: 2 axles each having 9 disks + 2 spares.

The future is: 4 axles each having 10 disks + 1 spare.

So I need to do some --grow to reshape the drives.

I thought this would be enough:

mdadm -v --grow /dev/md1 --raid-devices=10 --backup-file=/root/back-md1
mdadm -v --grow /dev/md2 --raid-devices=10 --backup-file=/root/back-md2
mdadm -v --grow /dev/md0 --raid-devices=4 --add /dev/md3 /dev/md4

The last command works, but the 2 first commands fail with:

mdadm: Need to backup 7168K of critical section..
mdadm: /dev/md2: Something wrong - reshape aborted

How can I --grow a RAID6 to use more devices?

Info about the system:

$ mdadm --version
mdadm - v3.2.5 - 18th May 2012
$ uname -r
3.5.0-17-generic

Makefile to reproduce the problem:

all:        install_parallel createmd0 createmd3 createmd4

install_parallel:
    parallel --version || wget -O - pi.dk/3 | bash

loop:
    # make enough loop-devices
    seq 0 50 | parallel mknod -m 660 /dev/loop{} b 7 {} || true
    seq 0 50 | parallel dd if=/dev/zero of=l{} bs=1M count=1
    seq 0 50 | parallel losetup -f l{}

createmd1: loop
    mdadm --create /dev/md1 -c 128 --level=6 --raid-devices=9 -x 2 /dev/loop1? /dev/loop1

createmd2: loop
    mdadm --create /dev/md2 -c 128 --level=6 --raid-devices=9 -x 2 /dev/loop2? /dev/loop2

createmd3: loop
    mdadm --create /dev/md3 -c 128 --level=6 --raid-devices=10 -x 1 /dev/loop3? /dev/loop3

createmd4: loop
    mdadm --create /dev/md4 -c 128 --level=6 --raid-devices=10 -x 1 /dev/loop4? /dev/loop4

createmd0: loop createmd1 createmd2
    mdadm --create /dev/md0 -c 512 --level=0 --raid-devices=2 /dev/md1 /dev/md2

grow:
    mdadm -v --grow /dev/md1 --raid-devices=10 --backup-file=/root/back-md1
    mdadm -v --grow /dev/md2 --raid-devices=10 --backup-file=/root/back-md2
    mdadm -v --grow /dev/md0 --raid-devices=4 --add /dev/md3 /dev/md4

info:
    mdadm --version
    uname -r

clean:
    mdadm --stop /dev/md0 /dev/md? || true
    seq 0 50 | parallel losetup -d /dev/loop{} || true
    seq 0 50 | parallel rm l{}

NeilBrown (neilb (o) suse.de) answered this by email.

The problem is the loopback-devices are too small. 1 MB is too little. If size is changed to 30MB it works:

seq 0 50 | parallel dd if=/dev/zero of=l{} bs=30M count=1