follow-up re system software

Kristoffer Ericson kristoffer.ericson at
Sat Nov 21 05:34:03 EST 2009

On Sat, 21 Nov 2009 04:37:30 +0100
Lars-Peter Clausen <lars at> wrote:

> Hash: SHA1
> Wolfgang Spraul wrote:
> > Ron,
> >
> >> Is that issue considered a show stopper? until one understands if
> >> the root cause is hardware or software, one could argue it as
> >> being STOP SHIP severity.
> >
> > No, not a show stopper. Almost the same hardware design as ours has
> > shipped over 60,000 units already. If there would be a serious
> > microSD problem it would have been found. What that means is that
> > there is at least a workaround to the problem. Also, I vaguely
> > recall that we have not seen this issue when we started to work
> > with the original kernel, but only after we cleaned up the
> >  sources and up-leveled to 2.6.31. All hardware ships with bugs.
> > Many times you can even argue what is a hardware bug and what a
> > software bug. The lowest levels of software routinely have to
> > 'cover up' / 'workaround' hardware bugs, making the hardware look
> > 'stable' for higher levels of software.
> >
> > For this particular problem, I am confident enough that we will
> > find a software workaround in the end to not make it a STOP SHIP
> > severity. Otherwise we would never ship anything, and you would
> > happily use lots of other consumer electronics with bugs inside you
> > just don't know about.
> >
> > However, I am clearly aware of this bug, and hopefully we get to
> > the bottom of it soon... The beauty of open - everything is known
> > :-) Wolfgang
> Hi
> Just wanted to let you know that I identified the problem and fixed it :)
> Turned out to be, as I suspected, a caching issue. The MIPS
> architecture has two independent caches: One for instructions and one
> for data.
> If data is read from the sd card it goes into the data cache. If it
> hasn't been used in while it's probably going to be replaced by other
> data and the pages holding the sd card data are going to be flushed
> which means their contents is written to the main memory.
> Now if we want to boot from a sd card we are going to execute the data
> read from the card. But as said earlier MIPS has two different caches.
> So in our case the instruction cache is going to be filled from the
> main memory. But the data read from the sd card has not been flushed
> from the data cached yet and so we end up executing random leftover
> data. The solution is to flush the the data cache after reading from
> the sd.


> - - Lars
> Version: GnuPG v1.4.9 (GNU/Linux)
> Comment: Using GnuPG with Mozilla -
> =5Zck
> _______________________________________________
> Qi Developer Mailing List
> Mail to list (members only): developer at
> Subscribe or Unsubscribe:

Kristoffer Ericson <kristoffer.ericson at>

More information about the discussion mailing list