Re: NetBSD/java threading issues

[Francois Tigeot <ftigeot%wolfpond.org@localhost>]

On Sat, Dec 11, 2010 at 08:34:54PM +0000, David Brownlee wrote:
> On 11 December 2010 20:23, Francois Tigeot <ftigeot%wolfpond.org@localhost> 
> wrote:
> > On Sat, Dec 11, 2010 at 07:50:07PM +0000, David Brownlee wrote:
> >
> >> If so could you try booting with only
> >> one CPU enabled and see if the threading issues exhibit?
> >> Also,
> >> possibly try running the netbsd-4 binary under netbsd-5 (probably with
> >> 'sysctl -w kern.no_sa_support=0' first).
> >
> > I don't think this will change anything: Martin S. Weber had the same thread
> > always waiting problem with Linux binaries...
> 
> That will ensure its using the old 'sa' threading library, which ships
> with netbsd-5 for compat...

Okay, I understand the logic.

I just finished a complete set of tests with NetBSD-5.1. In the end,
disabling one of the two CPUs on boot changed absolutely nothing, so
we have only four cases left for boostrapping wip/jdk15:

1. NetBSD 4.0.1 kaffe binary running under NetBSD 5.1

   Eternal sleep, kaffe uses 0% CPU

2. 4.0.1 kaffe binary running under 5.1 + kern.no_sa_support=0

   Eternal sleep, kaffe uses 0% CPU

3. NetBSD 5.1 kernel, NetBSD 4.0.1 kaffe binary, NetBSD 4.0.1 userland,
   kern.no_sa_support=1

   [...]
   cd ../generated;   /usr/pkgsrc/wip/jdk15/work/bin/java -classpath .
   MakeDeps diffs UnixPlatform platform.current includeDB.current
   /usr/pkgsrc/wip/jdk15/work/hotspot/build/bsd/platform_amd64 includeDB
   kaffe-bin: libpthread: sa_register failed

3. NetBSD 5.1 kernel, NetBSD 4.0.1 kaffe binary, NetBSD 4.0.1 userland,
   kern.no_sa_support=0

   The jdk build runs fine


At a first glance, we've learned two things:
- the new NetBSD 5.1 kernel itself is fine; if there's a bug
somewhere, it has to be in userland.
- the compat environment for NetBSD 4 binaries is not the same
as a real NetBSD 4 userland.

-- 
Francois Tigeot