Re: Modernize NetBSD/powerpc support in GDB

[Mark Kettenis <mark.kettenis@xs4all.nl>]

> From: "Nathan J. Williams" <nathanw@wasabisystems.com>
> Date: 11 May 2006 21:55:00 -0400
> 
> This seems good, though less dramatic than you report. On my macppc
> system, it took the testsuite from 788 to 753 failures (though with a
> notable number of PASS->FAIL changes as well)

I had some PASS->FAIL changes as well; I looked at a few of them but
they all seemed to be related to GDB missing a breakpoint.  Presumably
these breakpoints are now missed because an earlier test now PASSes
and the instruction cache is in a somewhat different state now.

> ; on a AMCC 440 system, it went from 179 to 170 failures, with only
> one PASS->FAIL regression (which is a mysterious additional ^M in
> the output of "info all-registers".

Thanks for testing; I'm committing the diff then ;-).  I've seen
spurious ^M's in the "info all-registers" output on OpenBSD/sparc and
OpenBSD/sparc64 as well.

> Summary diffs attached, and I'm trying to figure out
> why macppc is so much worse than the 440.

My bet is still that it is a cache flushing issue, especially since
the 440, which uses different code to flush the cache, is so much
better.