Re: sparc64 vs. gcc 4.8

[Michael <macallan%netbsd.org@localhost>]

Hello,

some more data points:
on the sb2500, recent kernels built with gcc 4.8 will either
spontaneously reboot or deadlock, sometimes after hours under full
load, sometimes earlier. Those built with gcc 4.5 will deadlock - I
haven't seen any of those spontaneously reboot yet. Since the
spontaneous reboots happen on the U60 as well with gcc 4.8 it has to be
something else than the schizo-related spontaneous reboots we've seen
earlier on US-IIIi hardware. Also, no deadlocks on the U60.
Now to the funny part. The last stable sb2500 kernel I have found here
is a 6.99.44 from mid-June. Since there was a bunch of commits in
arch/sparc64 starting at that time I downgraded arch/sparc64 to June
1st, built a kernel and ... got a deadlock. So I suspected local
changes, checked out a clean source tree, built a kernel, got a
deadlock. Then I downgraded the whole clean source thee to July 1st and
that one is still going, almost finished a build.sh -j4 distribution
with no sign of trouble.
Looks like whatever commit triggered the deadlocks is not in
arch/sparc64. Either way it seems to be a separate issue from the
spontaneous reboots.

have fun
Michael