usbnet - MAC and sequential interfaces

Werner Almesberger werner at
Tue Feb 3 12:12:11 CET 2009

Andy Green wrote:
> This opens a can of worms for Qi in terms of supporting GTA02 in the
> field already with random dynparts to give access to the magic ext2
> partition in NAND.

One more reason to repartition when switching to Qi ;-)

The factory partition is just another item that isn't right if one
ignores the possible differences introduced by dynamic partitions.
They are:

- going dynamic -> static for the first time:

  - may fail to load kernel from NAND (3*p)
  - may fail to mount the factory partition (72*p)

- going static -> dynamic:

  - may fail to load kernel from NAND if placed there while
    using static partitions (3*p)
  - file may disappear from the NAND rootfs (< 74*p)

- going dynamic -> static again:

  - previously deleted files in the NAND rootfs may come back from
    the dead (< 74*p)

p is the probability that a NAND eraseblock is bad. NAND rootfs
corruption has some of the highest probabilities (depending on
file system usage patterns) but it's also the hardest to notice
and even harder to attribute.

I would recommend the following policy:

- if switching to Qi on a trial basis, avoid changing things on
  the NAND rootfs. Best to avoid mounting NAND rootfs at all.

- if switching to Qi permanently, either

  - avoid touching NAND entirely henceforth,
  - erase the entire NAND and re-polulate from scratch, or
  - repartition

- while using Qi, use NOR u-boot only to DFU to the "u-boot"
  partition, no other partitions

- in case of switching back from Qi to u-boot, erase the whole
  NAND (backup your factory partition first) and re-populate
  from scratch

That should avoid all these nasty surprises without causing
too much pain elsewhere.

- Werner

More information about the openmoko-kernel mailing list