This patch adjusts a couple of tests, removing spurious failures.

Coldfire has no jbsr instruction, so I changed the assembler test to use a jsr, 
which is valid on both coldfire & m68k

The charset.exp test would spuriously fail because the previous test left $pc in 
what was now malloc.  Therefore the 'break 150' line would be interpretted as a 
line in malloc's source and not the test program's source

The fileio tests were failing because we didn't quite stop where we expected in 
the source (that might be a further problem I have to investigate).  I decided 
to robustify the fileio tests by inserting calls to a 'stop' routine, and having 
a permanent breakpoint there.

ok?

nathan
-- 
Nathan Sidwell    ::   http://www.codesourcery.com   ::         CodeSourcery
nathan@codesourcery.com    ::     http://www.planetfall.pwp.blueyonder.co.uk