On Oct 18,  9:43pm, kilbi@rad.rwth-aachen.de (Markus W Kilbinger) wrote:
-- Subject: Re: Missing some floating point function?

| Zoulas> Looks like a NULL pointer (NULL+0x4) dereference bug.
| 
| (What does it mean?)
| 
| I've extracted perl's Configure C source fragment and recompiled it
| explicitly: It's really and only the cast:
| 
|   int i32;
|   double f, g;
|   [...]
|   i32  = (int) g;
| 
| depending (I didn't write that last time) on the mips specific
| compilation optimization: Binaries compiled with anything like
| '-mips2' or higher seem to produce the panic:
| 
|   gcc -O -s -mips2 -o test-casti32 test-casti32.c
| 
| while plain (or '-mips1') binaries seem to run fine.
| 
| Is this a (known) limitation of cobalt/mips port/libs?
| A bug in gcc?
| 
| Anyway, a kernel panic for this kind of user 'stupidity' is not very
| kind... ;-)
| 

No, this is a kernel bug. The kernel traps the floating point operation
and then has a bug handling it. The null pointer bug is in the kernel.
With mips1, probably the compiler does not generate floating point code
and does the equivalent of soft-float so that the buggy code does not get
invoked.

christos