> > today tried the latest kernel from the autobuild directory and found,
> > that every 20 seconds the system hangs completely for about 1 second,
> > and the harddisk is accessed. Does anybody know what is the reason for that?
> 
> This is caused by a change in the update process in the kernel. A fix was
> introduced for a bug which means that a lot of calls to pmap_enter and
> other pmap stuff. On the SA110 this means that there is a _lot_ of cache
> flushing going on. The update process is the one which keeps the disks
> synchronised and runs at 30 second intervals. I have modified my copy of
> the sources to remove the fix based on a mail message sent to this list a
> while back. If you do not have kernel building facilities I could probably
> build one for you and e-mail it to you with this patch.

Yes, I already found that the SA is slower in some tests of the Byte
benchmark (pipe, shell scripts and so on).

Hm, might it be that there is a bit too much cache flushing done at the
Strongarm? And if it always needs to be done, perhaps it is faster to
switch off the cache for this time.

If I would get the sources from somewhere I certainly could build my own
kernel. But I have not very much experiences with programming under
Unix systems, and not too much time. I am quite happy with SA-4958.
Yes, DMA support for the Connect32 would be great, as sound support
would be, and not so much cache flushing. But I think that will come
some day, if the kerner team has some time again.


Peter Teichmann