3.15.0+: 32-bit multiple-of-2GB partition issues, already solved?

3.15.0+: 32-bit multiple-of-2GB partition issues, already solved?

[Michael L. Semon]

Hi!  I looked at the linux-fsdevel archives on the Web and saw the 
thread "Cannot partition 32GB disk on a 32bit machine."  So I'll 
get Alan's patch, junk my bug report, and just note some extra 
symptoms...

When a partition here is an exact multiple of 2GB...

1) lilo will fail with this message:

   part_nowrite: read:: No such device or address

   The exact-multiple partition does not need to be in /etc/lilo.conf.

2) mkfs.xfs will fail with this message:

   superblock read failed: Operation not supported
   mkfs.xfs error - read only 0 of 512 bytes

3) badblocks will report bad sectors on the read-only test.  These are 
the kind of results from `badblocks -b 512 ... /dev/2GB_multiple_dev`:

   4GB partition: 0
   8GB partition: 0, 8388608
  10GB partition: 4194304, 12582912

   badblocks comes up clean under vanilla kernel 3.15 and 3.10.41.

4) `ntpq -p` will segfault, and the strace has it ending on some kind 
   of "no such address" error as well.

`git bisect` brought me here...

root@oldsvrhw:/usr/src/kernel-git/linux# git bisect bad
37938463540b075e9166cf774c59274379f7a8ca is the first bad commit
commit 37938463540b075e9166cf774c59274379f7a8ca
Author: Al Viro <viro@zeniv.linux.org.uk>
Date:   Sat Mar 22 06:57:37 2014 -0400

    blkdev_aio_read(): switch to generic_file_read_iter(), get rid of iov_shorten()
    
    Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

:040000 040000 ad9ba87463bebb2a80d343b9c649587ead91b4a1 fc72efc46850efdf88357e7a783a1dfebbdf5132 M	fs

...though now it's looking more like the commit brought a symptom from 
elsewhere to the forefront.

Should Alan's patch not work, I'll keep working around the issue by 
not creating partitions that are multiples of 2 GB.

Thanks, and good luck!

Michael