On Sun, Sep 23, 2007 at 05:50:10PM +0200, Joerg Sonnenberger wrote:
> On Sun, Sep 23, 2007 at 08:41:46AM -0700, Matt Thomas wrote:
> >
> > On Sep 23, 2007, at 8:21 AM, Joerg Sonnenberger wrote:
> >
> >> Hi all,
> >> I'd like to commit the attached patch to allow converting drivers on a
> >> case-by-case base to not depend on struct device being the first field.
> >> if_bge is included as sample. CFATTACH_DECL*_NEW is meant as a temporary
> >> interface until all drivers are converted and should be removed after
> >> that. This is a first and critical step to completely remove the public
> >> definition of struct device (and make it an abstract interface thereby).
> >>
> >> Joerg
> >> <softc-sep.diff>
> >
> > Shouldn't devices be allocated from their own pool rather than be
> > malloced?
> 
> Can we revisit this when the compat code is not needed any more? I guess
> the answer depends quite a bit on how large struct device is and
> therefore how big the overhead of malloc vs pool is. I also believe that
> it is certainly not performance critical, so space usage is more
> important here.

.. agreed, in which case kmem_alloc() is the appropriate choice.

Andrew