> : I'm still working on COMPAT_IRIX. It seems that during signal delivery,
> : the kernel has to build a signal frame with a field containing errno
> : address in userspace. It is used by the signal trampoline to set errno
> : on sigreturn exit.
> I have a suspicion that this value might be passed to the kernel from the
> libc sigreturn call in an extra register.  It's rather unusual for the
> kernel to have to grovel the ELF symbol tables to find such a beast.

On IRIX, the kernel already has to handle psection relocation and load (see
/sys/compat/irix/irix_syssgi.c:irix_mapelf()), I wouldn't be that suprised if
it also has to dig for errno.

I ran a test program on IRIX and checked registers using gdb at signal handler
invocation. errno address seems to be set there:

(gdb) x/32w $sp
0x7fff2ab0:     0x00000000      0x00000000      0x00000000      0x00000000
0x7fff2ac0:     0x00000000      0x00000000      0x00000000      0x00000000
0x7fff2ad0:     0x00000000      0x00000000      0x00000000      0x00000000
0x7fff2ae0:     0x00000000      0x00000000      0x00000000      0x00000000
0x7fff2af0:     0x00000000      0x00000000      0x00000000      0x0fb4f9b0
0x7fff2b00:     0x00000000      0x80000001      0x00000000      0x7fff2b10
0x7fff2b10:     0x0000000f      0x00000000      0x00000000      0x00000000
0x7fff2b20:     0x00000000      0x00000000      0x7fff0000      0x00006b58
(gdb) x/1w 0x0fb4f9b0
0xfb4f9b0 <errno>:      0x00000000

If it's done at signal handler invocation, then 
1) the kernel knows it
2) the user process always has a register pointing there. 
3) another solution?

Dumping the registers during normal program operation shows no errno address,
hence solution 2) seems wrong.
(gdb) info reg
          zero       at       v0       v1       a0       a1       a2       a3
 R0   00000000 00000000 00000000 0000bded 0000bec8 00000001 00000026 00000000
            t0       t1       t2       t3       t4       t5       t6       t7
 R8   00000042 00000000 0fb4e931 00000042 00000040 00001000 0fb54fa0 00001000
            s0       s1       s2       s3       s4       s5       s6       s7
 R16  0000bec8 7fff2f74 7fff2f7c 7fff2fc8 00000000 00000000 00000000 00000000
            t8       t9       k0       k1       gp       sp       fp       ra
 R24  0fb502e4 0fa4541c 00000000 8000003f 1000c0c0 7fff2ef8 00000000 00400e14
            pc    cause      bad       hi       lo      fsr      fir
      0fa455a4 00000008 00000000 00000050 00000073 00000000 00000000 

-- 
Emmanuel Dreyfus
UNIX *is* user friendly. It is just a bit selective about his friends
manu@netbsd.org