[snip]

>> I hope the change in arch/um/kernel/skas/MMU.c isn't the cause of all this trouble!
> No it is not, but it is one that needs refinement. Fix coming up in 10
> minutes or so, I was just cleaning up the patches for submission.
>
> The issue is reentrancy of the core interrupt logic which the code in
> unblock_signals is supposed to prevent, but it fails (for some reason)
> to do so.

Some of it looks like a heisenbug from debugging by the way.

If you look in softirq.c you see:

#ifdef CONFIG_TRACE_IRQFLAGS
local_irq_enable <http://lxr.free-electrons.com/ident?i=local_irq_enable>()
#endif

That in UML translates to unblock_signals(). That in turn will 
re-execute hardware interrupts out of a software IRQ bottom half context.

There is still an inherent reentrancy issue, but I suspect that it is 
less with the debugging off.

Patchset coming up in next email. It throws bh warnings if you have 
CONFIG_TRACE_IRQFLAGS, but that is not something I can fix in first pass.

A.