Re: [PATCH 2/2] workqueues: export keventd_wq

[Oleg Nesterov <oleg@redhat.com>]

On 04/30, Tejun Heo wrote:
>
> On 04/30/2010 07:39 AM, Arve Hjønnevåg wrote:
> > On Thu, Apr 29, 2010 at 10:16 PM, Tejun Heo <tj@kernel.org> wrote:
> >> Hello,
> >>
> >> On 04/29/2010 09:45 PM, Oleg Nesterov wrote:
> >>> -static struct workqueue_struct *keventd_wq __read_mostly;
> >>> +struct workqueue_struct *keventd_wq __read_mostly;
> >>> +EXPORT_SYMBOL(keventd_wq);
> >>
> >> Umm... does it have to be EXPORTed?  Suspend block API can't be built
> >> as a module, right?

Right, but this allows to make schedule_xxx/flush_scheduled_work
inline and kill more EXPORT_SYMBOL's, and cmwq exports it anyway
(iirc it also renames it).

> > The suspend block api cannot be built as a module, but if
> > schedule_suspend_blocking_work will just call
> > queue_suspend_blocking_work(keventd_wq, work) it may as well be an
> > inline function which would require the export.
>
> I think it would be better to keep the thing inside the kernel, at
> least for now.

But then schedule_suspend_blocking_work() in turn needs EXPORT_SYMBOL().


OK. Let's forget this patch. We can unify schedule_suspend_blocking_work
and queue_suspend_blocking_work later, or Arve can add this export into
his patch (without EXPORT_SYMBOL) - either way I agree. This is very
minor issue, I regret I originated this almost offtopic noise ;)

Oleg.