Most recent kernel where this bug did not occur:
Distribution: Fedora Core 7
Hardware Environment: AMD Athlon 64
Software Environment:
Problem Description:

Negative timestamps (i.e. before 1970) are incorrectly handled due to some obvious 64bit issues.  They show up normally as long as the cache has not been purged.

Steps to reproduce:

Touch a file in an ext3-file system with a negative timestamp, e.g.:

  # touch -t 196901010000 /opt/foo

Unless you have been writing a lot of data to this partition after the last command, the next one should display the following:

  #  ls -l /opt/foo
  -rw-r--r-- 1 root root 0 1969-01-01 00:00 /opt/foo

This is still correct.  But now we purge the file system cache by unmounting and mounting the partition again:

  # umount /opt/foo
  # mount /opt/foo

Now the timestamp will be corrupted:

  # ls -l /opt/foo 
  -rw-r--r-- 1 root root 0 2105-02-07 06:28 /opt/foo

It seems obvious that the upper 32 bits of the 64 bits representing the time stamp get lost, which explains the above observation.

On 32bit architectures there is no such problem.  I haven't managed to reproduce this problem with other file systems (e.g. XFS).