Hello.  If you do a /sbin/sysctl -a |grep fpu_present on your broken
machine, what do you get back?  If the value is 1, that's probbably your
problem.  The loaded math libraries are  expecting to have some kernel
support for math emulation.  However, if it's 0, then there's definitely a
problem.
-Brian
On Feb 25,  9:09pm, Pierre Dubuc wrote:
} Subject: Re: Status of math_emulate
} -----BEGIN PGP SIGNED MESSAGE-----
} Hash: SHA1
} 
} Replying to message <6e9741c60702251724p611b5329p374ab28f9ea65ed1@mail.gmai...
} 
} >
} > You may wish to try only defining one kind of CPU type (I386) instead
} > of having I486 there; the fact that the kernel is detecting a 486DX
} > and GENERIC has I486 defined might be tripping you up.
} >
} 
} I rebuilt a kernel with just I386 (and MATH_EMULATE) defined. The probe 
} yields this interesting sequence now...
} 
} NetBSD 3.1 (I386) #0: Sun Feb 25 20:53:06 EST 2007
}          pldubuc@localhost.localhost:/usr/src/sys/arch/i386/compile/I386
} total memory = 15616 KB
} avail memory = 13164 KB
} mainbus0 (root)
} cpu0 at mainbus0: (uniprocessor)
} cpu0: Intel 486DX (486-class)
} NOTICE: this kernel does not support i486 CPU class
} NOTICE: lowering CPU class to i386
} ...
} 
} The situation remains the same, though. Without an npx, the MATH_EMULATE 
} option is required, but the implementation of sqrt is flawed. Or rather, 
} by browsing math_emulate.c, it is altogether absent.
} 
} Does anyone think a PR is justified? I'm pretty sure the developers have 
} better things to do than supporting 15 year old hardware, but I'm new to 
} NetBSD, so I don't know how those things go.
} 
} - -- 
} Pierre Dubuc
} pldubuc@yahoo.ca
} -----BEGIN PGP SIGNATURE-----
} Version: GnuPG v1.4.6 (NetBSD)
} 
} iEYEARECAAYFAkXiQUoACgkQaMA1iJ2qYmJV9gCeOYVXr4Ma6iZMMdWo4j77lUBz
} MJUAnRY30SLopPM4k+72wIAYMlsfLrqe
} =dc3F
} -----END PGP SIGNATURE-----
} 
>-- End of excerpt from Pierre Dubuc