Hi,

The previous patch that introduced the new interrupting
method, leaves some chances for the inferior to mess the
interrupting, since the operation is not atomic.  For
example by tweaking its own thread's priorities or if the
inferior is calling ResumeThread in its own threads while
gdbserver is stopping them.

Here is the patch that tweaks gdbserver's priorities
so to minimize those chances.

This can never be perfect, since, now we may be adding a bit
of extra entropy to other components of the system, especially
true in WinCE, where we may be inhibiting some other lower
priority thread to execute, even if just for a little.
For WinCE, I'm putting the ceiling on the highest application
priority, which is normally lower than what the time critical
drivers threads use.

This is mostly Leo's code.  I've splitted it from his
previous patch, fixed and cleaned a few things here and there,
added WinCE support, and regtested it on a local
i686-pc-cygwin gdbserver and against a remote arm-wince
gdbserver.

Leo, did I miss anything?

Cheers,
Pedro Alves