Re: Implementation of POSIX shared memory objects

[Antti Kantee <pooka%cs.hut.fi@localhost>]

On Thu Feb 12 2009 at 00:37:57 +0900, YAMAMOTO Takashi wrote:
> > ad%netbsd.org@localhost wrote:
> > > On Wed, Feb 11, 2009 at 10:13:17AM +0000, Mindaugas Rasiukevicius wrote:
> > > 
> > > > Any reads/writes? Apart from greater file-system overhead (as opposed to
> > > > aobj), I think it can involve disks. :)
> > > 
> > > tmpfs == swap backed uao
> > 
> > Right, but this is dependency on tmpfs, which one might not necessary want
> > to have. I think it is not that friendly for embedded devices too.
> 
> while it can be, it doesn't need to be tmpfs.  ie. same as /tmp.

I just did a special-case implementation of this in userspace.  Comparing
the complexity of it to the suggested kernel patch, I think implementing
this in the kernel is not the right way to go.

.. well, and everyone knows how i feel about pushing unnecessary stuff
into the kernel.