Werner Almesberger wrote:
Here's a quick status update on open issues and problems I know of in
the lower reaches of the system (in no particular order):
[snip]
- JFFS2 suffers gradual corruption
The corruption manifests itself mainly in an increasing number of
complaints from JFFS2, but it may also be possible that it causes
a performance degradation or data loss/corruption (although I
haven't observed any of the latter).
I believe this is a mismatch between mkfs.jffs2 parameters and the
actual NAND geometry, but I don't have a good way to predictably
reproduce this yet (and thus couldn't tell if an attempt at solving
it is really effective or not).
Suggested resolution: I'll give this one more try later today, and
when I get back home. In the worst case, this may not get resolved
before we have to provide an image to the factory. Since we'll want
to do a complete application update for new devices anyway, that
update could take the form of a new rootfs (maybe this is the plan
anyway ?), which would then have the correct settings.
(Related to this is also the use of hardware-accelerated ECCs. I've
turned them off until the corruption is resolved, just to make sure
we don't add yet another potentially destablizing factor. HW ECCs
are probably perfectly fine, though, so we'll also have a
performance win once all this is solved.)
...
For what it's worth, I suffered a series of nasty JFFS2 corruptions that
seem to be similar (progressive) while debugging some boot and shutdown
issues on the GTA01 this past week. I was appending to a log file on
the jffs2 filesystem during the shutdown process, but I've done that
many times before both on the Neo with the 2.6.22 kernel and many many
times on other devices. I expect a truncated logfile (at worst!);
there's no reason for the jffs2 corruption to occur. (FWIW, a "sync"
after the log file was written and closed did nothing to change the
frequency with which the problem appeared.)
Unfortunately I'm off on a business trip again in a few hours and won't
be able to work on this until the weekend, but it does indicate that
perhaps there are some software issues remaining.
Mike (mwester)
