Re: s390(64) per_cpu in modules (ipv6)

[Rusty Russell <rusty@rustcorp.com.au>]

On Fri, 2004-10-15 at 19:15, Martin Schwidefsky wrote:
> 
> 
> Rusty Russell <rusty@rustcorp.com.au> wrote on 15/10/2004 03:41:40 AM:

> > The worse problem is that a (static) per-cpu var declared *inside* a
> > function gets renamed by gcc; IIRC some generic code used to do this.
> 
> __thread in the kernel would be a real innovation, but I fear it isn't easy.
> The problem with the per_cpu__x variables in modules is solved for s390x
> by the way.

Sure, but it doesn't solve this case, AFAICT:

void func(void)
{
	static DEFINE_PER_CPU(x, int);

	__get_per_cpu(x)++;
}

The compiler will create a variable called "per_cpu__x.0" and your asm
reference to "per_cpu__x" will cause a link failure, no?  Obviously, you
would have noticed this, so I'm wondering what I'm missing.

I hit this in mm/page-writeback.c:balance_dirty_pages_ratelimited().

Confused,
Rusty.
-- 
Anyone who quotes me in their signature is an idiot -- Rusty Russell