Re: Linux 2.6.0-test6

[viro@parcelfarce.linux.theplanet.co.uk]

On Sun, Sep 28, 2003 at 02:19:24PM +0200, Oliver Pitzeier wrote:
> Hi folks/Linus!
> 
> Linus Torvalds wrote:
> > Ok, too long between test5 and test6 again, so the patch is 
> > pretty big. Lots of driver updates and architectures fixed, 
> > but also lots of merges from Andrew Morton. Most notably 
> > perhaps Con's scheduler changes that have been discussed 
> > extensively and made it into the -mm tree for testing.
> 
> It work's on my Intel machine, but on Alpha, I get this:
> <snip>
>   CC      init/version.o
>   LD      init/built-in.o
>   LD      .tmp_vmlinux1
> kernel/built-in.o: In function `try_to_wake_up':
> kernel/built-in.o(.text+0x438): undefined reference to `sched_clock'

Add
unsigned long long default_sched_clock(void)
{
	return (unsigned long long)jiffies * (1000000000 / HZ);
}

in kernel/sched.c and

#define sched_clock default_sched_clock

in include/asm-alpha/system.h

FWIW, the former should've been done from the very beginning and sched_clock
should've been made a weak alias for default_sched_clock.  That would avoid
the breakage of platforms original patch didn't update.

BTW, how about adding weak_alias(type, name, args, default_variant) to
compiler.h?  For most platforms it would be

#define weak_alias(type, name, args, default_variant) \
	type name args __attribute__((weak, alias(#default_variant)));

Note that we already have something similar - cond_syscall(name) would
become weak_alias(asmlinkage long, name, (void), sys_ni_syscall) and
platform-specific stuff could be taken from current definitions of this
beast.