G'day,
  I had a similar observation when using the 386/387 assembly. I had wanted
a mathlib badly so I made try to compile the code from DJ's GPP (for MS-DOS).
The maths routine are written mostly in assembly language. I compiled the
C and assembly code and made libm.a, ranlib the things. All without problem.
  I then proceeded to write a small C program with a sin() and tan() call.
That is when I had

device not available: 0000
EIP:    000f:00000060
EFLAGS: 00010246
ESP:    0017:03FFFED0
fs: 0010
base: 10000000. limit: 04000000
Stack: 00000044  66666666 40026666 03fffe48
Pid: 2123, process nr: 4
dd 44 24 04 d9 fe df e0 9e 7b

  A wild guess, does it have anything to do with the fact that on the 486, the
maths coprocessor is already incorporated onto the chip it-self?

  BTW, I am trying to compile the bsd-reno's maths library temporary. Anybody
with warning, advice, question please e-mail to me.

nicholas@cs.uwa.oz.au