>>>>> "Izumi" == Izumi Tsutsui <tsutsui@ceres.dti.ne.jp> writes:

    Izumi> With the attached (a bit ugly) patch against cache.c (which
    Izumi> make always wbinv use index ops) the problem seems
    Izumi> disappear, but I'm not sure what is the real problem.

    >> I've just tried this patch on my qube2: It seems to avoid the
    >> problem in pure read situation ('cmp', 'sha1' et al.), but I
    >> still can reproduce corruptions if doing 'cp file1 fileN'
    >> multiple times...

Just as a followup: Even pure read access still can produce data
corruption, but much more seldomly than without this patch.

Isn't it possible to turn the cache off just for testing purposes,
just to see if the problem is still present then?

Markus.