Jason Mitchell wrote:
I should post a correction here. The problem is definitely with the NetBSD hpcarm kernel. A posting to the NetBSD hpcarm mailing list that confirms it is here:http://mail-index.netbsd.org/port-hpcarm/2002/07/09/0001.html
If that's actually the problem, all you need to do is assign multiple L2 page tables to mapping the kernel (c.f. how KERNEL_PT_KERNEL_NUM is used in sys/arch/evbarm/lubbock/lubbock_machdep.c), or go the whole hog and make the L2PT allocation dynamic (c.f. ptcount_total calculations in sys/arch/evbarm/evbarm/initarm_common.c).
The first should be ~ trivial to do modulo testing it... (the second shouldn't be that difficult either, but without hardware to test on I'm not keen to try it since undoubtedly something will need debugging).
In fact, if you can test the changes (and even better if you have a working kernel source tree from which you can build test kernels) and are interested in doing so, I will try and cook the simplest set of changes to fix that when I have a few minutes and my allocated brain cycles for the day haven't all been used up.
Note that to confirm (and if the DEBUG_BEFOREMMU option works as advertised by the source), you could build a kernel with that and VERBOSE_INIT_ARM defined and make sure you at least get some output from the kernel (ie, that it really isn't hpcboot and/or WinCE conspiring against you).
--rafal