Re: MODULAR option and advertised semaphore support

[Nicolas Joly <njoly%pasteur.fr@localhost>]

On Fri, Nov 05, 2010 at 04:46:41PM +0200, Jukka Ruohonen wrote:
> On Fri, Nov 05, 2010 at 03:14:44PM +0100, Nicolas Joly wrote:
> > While testing semaphore support, i noticed that advertised support is
> > currently hardcoded ... which does not deals nicely with the currently
> > available module framework.
> > 
> > By example, you'll loose if you have a kernel without
> > P1003_1B_SEMAPHORE option, but the module cannot be loaded for
> > whatever reason.
> > 
> > The attached patch aims at making it fully dynamic, at the cost of
> > loading the ksem module.
> 
> If I understood correctly, why not just make the SYSCTL_SETUP disappear?

Well, maybe; but you then have to fix it somewhere else, actually for
sysconf(_SC_SEMAPHORES) which currently rely on sysctl.

> It can not be used with modules anyways, and probably should be removed
> entirely at some point. See also aio(3) and mq(3) for similar changes.

Moving the whole sysctl tree into the module won't work unless the
module is actually loaded; sysconf(_SC_ASYNCHRONOUS_IO) will claim
that's unsupported which is not true, using a syscall will make the
module autoload and it will work just fine.

-- 
Nicolas Joly

Biological Software and Databanks.
Institut Pasteur, Paris.