branches: 1.2.2; 1.2.8;
sh_frame_cache always tries to read FPSCR. Since frame.c uses gdb
exceptions now, there's an unplesant side effect that when FPSCR is
unavailable, your last display will get disabled just in case, "to
avoid infinite recursion". That happens directly in throw_exception,
so even catching that NOT_AVAILABLE_ERROR doesn't help.

Tweak the code a bit so that sh_analyze_prologue only reads FPSCR as
needed, when an FMOV instruction is encountered in the prologue.

XXX: I'm not sure if this is the right thing to do, but it seems
minimally intrusive.