I'm still having an occasional problem with corruption when I use this
patch, but it's very different from all previous problems and it's not
clear to me that it's a software problem. Briefly, when I copy a
large directory tree using tar and diff the tree, I get a number of
single bit errors. These errors are not always in the same bit
(although they're usually in the low 4 bits or so), and they aren't at
the same location each time modulo a page or even modulo 8 bytes. The
fact that there's no apparent pattern, combined with the fact that
other people have performed the same test at my request and not had
any problems, leads me to believe that this might be a hardware
problem of some sort (such as a flaky SIMM). However, I occasionally
get single bit errors in consecutive bytes. Since I typically get
5-20 errors while copying 130 MB of data, I would not expect to get
consecutive errors very frequently.
Nonetheless, I would like everyone using my version of kernel memcpy()
to perform this test and email me their results, along with their
hardware setup. I'm most interested in the motherboard, disks, and
controller, as I suspect that that's where the problem lies if it's in
fact hardware. If you do have a problem, I'd like to know if it's the
same thing -- scattered single bit errors.
BTW, if I do a make -j of the kernel, either with or without my patch,
I have problems (segv's, mostly). I have ample swap space (130 MB
with 16 MB of DRAM).
-- Robert Krawitz <rlk@tiac.net> http://www.tiac.net/users/rlk/Member of the League for Programming Freedom -- mail lpf@uunet.uu.net Tall Clubs International -- tci-request@aptinc.com or 1-800-521-2512