Re: OpenSlug: ipkg -> segmentation faults ?
- --- In firstname.lastname@example.org, "lew1193" <lew001@g...> wrote:
> I think I have a very sick slug... (file size OK). Will LinksysI think this is a true H/W error.
> want a reason for return? How do I convince them there's really
> a H/W problem? Can I activate FTP on R63 with no disk, and ftp
> to/from the slug's tmpfs (to show corruption -- surely it also has
> a tmpfs)?
In the tmpfs (RAM) of the slug (under OpenSlug), I created (with
repeated cat) a ~ 1MB file containing a repetition of the same
line, eg. "0123456789abcdef" (plus LF gives 17 bytes, a prime
Then I copied (using cp(1)) this file to toto1, toto2, and so on
(still in tmpfs), checking md5sums each time. At toto8, the first
invalid md5sum appeared. Continued for a few more copies. Then I
did following on all files that report invalid md5sums:
grep -n -v 0123456789abcdef toto8 ...
This returns a few bad lines. Notice that the errored bytes are
the original ones with the top bit set to 1.
Other tests with different patterns show similar results.
It's a single bit here and there so this suggests that it's
unlikely to be memory config confusion. Other than the fact that
it's always the top bit at an odd byte position in the file that
gets corrupted, I haven't been able to see a clear pattern.
If confirmed, this will be the single most expensive bit of my
career :-). I have spent hours trying to figure out why the
device had data corruption problems...
I wonder if there are boot-time options to do a more thorough
memory check and exclude bad areas?