Re: Building of extended nbd fails on ppc

[Michael Schmitz <schmitz@opal.biophys.uni-duesseldorf.de>]

> The build process stops at the following line:
> (By the way, I'm using kernel-2.4.20-ben10 on an iBook v2.2)
>
> gcc  -o enbd-server enbd-server.o socket.o logging.o alarm.o file.o\
> select.o fileserver.o db.o stream.o ioctl.o ipaddrfile.o pidfile.o\
> interface.o lockfile.o md5.o lock.o shmem.o hash.o
>
> lock.o(.text+0x18): In function `lock_try':
> : undefined reference to `atomic_dec_and_test'
>                           ^^^^^^^^^^^^^^^^^^^
> lock.o(.text+0x18): In function `lock_try':
> : relocation truncated to fit: R_PPC_REL24 atomic_dec_and_test
> lock.o(.text+0x30): In function `lock_try':
> : undefined reference to `atomic_inc'
> [...]

atomic functions are only accessible to kernel code...

static __inline__ int atomic_dec_return(atomic_t *v)
{
        int t;

        __asm__ __volatile__(
"1:     lwarx   %0,0,%1         # atomic_dec_return\n\
        addic   %0,%0,-1\n\
        stwcx.  %0,0,%1\n\
        bne-    1b"
        SMP_ISYNC
        : "=&r" (t)
        : "r" (&v->counter)
        : "cc", "memory");

        return t;
}

The SMP_ISYNC is just a plain isync on SMP systems. On non-SMP systems it
should be safe to use these functions. On SMP systems, the isync might
prevent this (privileged instruction, perhaps -- BenH??).
Maybe only the SMP_ISYNC def should be protected by #ifdef __KERNEL__?

Just copy the code ...

	Michael