Conclusion of thread "Deadlocks with recent SMP current"?

Sascha Schumann sascha at schumann.cx
Sat Oct 2 13:49:58 PDT 2004


> Yes, this is believed to have been resolved in 5.3.

     Good to hear.

> So, you're running 5.2 on those systems, which was documented as a
> development release and not recommended for use on production systems

     No, *I* am not running FreeBSD 5 anywhere on productive
     systems.  Unfortunaly, the ISP recommended it to this
     particular owner of two systems where I am doing some
     voluntary work in my spare time.

     One of the servers ran 5.1 with an uptime of >200 days.
     There were no indications that 5.2 contained such vast
     regressions in comparison to 5.1.  Even today, the 5.2.1
     errata page does not say anything about SMP issues.

> (and indeed contained numerous rough edges and bugs), and you're
> scared to update to 5.3 which is going to be designated as the
> beginning of the 5.3-STABLE branch (and has had a couple of months of
> extensive bugfixing and QA)?

     I would rather downgrade to 5.1 (because I know it works
     reliably on this piece of equipment), but unfortunately,
     somewhere during "make installworld", the install binary
     starts to segfault.

     truss shows that the 5.1 /usr/bin/install calls the new
     fstatfs syscall 397.  AFAICT this could only happen when the
     5.1 install binary is linked against the 5.2 libc.  I suppose
     installworld does not point LD_LIBRARY_PATH at the newly
     build libraries under /usr/obj?  If not, what is the best way
     to ensure that binaries are linked against suitable libs
     during installworld?

     Thanks
     - Sascha



More information about the freebsd-current mailing list