On Fri, Jul 27, 2007 at 05:36:14AM +0100, Al Viro wrote:
> On Fri, Jul 27, 2007 at 06:27:44AM +0200, Sam Ravnborg wrote:
> > On Fri, Jul 27, 2007 at 02:18:13AM +0100, Al Viro wrote:
> > > On Thu, Jul 26, 2007 at 11:01:41PM +0200, Sam Ravnborg wrote:
> > >
> > > > +static void *__init_refok alloc_rte(unsigned long size)
> > > > +{
> > > > + return alloc_bootmem(size);
> > > > +}
> > >
> > > That makes no sense at all. If we ever call that after freeing initmem,
> > > we are screwed, period. Sounds like __init fodder.
> >
> > The call site has logic to prevent this from being called after init.
> > And the call site cannot be made __init and to limit the scope of
> > the __init_refok a small function is used.
> >
> > So unless I mis-understood something the above should be OK.
>
> Then the call site must be __init_refok, AFAICS. The point is,
> the callers of alloc_rte() are where the analysis belongs; use of
> __init_refok means that you assert that all calls of __init *from*
> *it* are safe. Otherwise it just becomes "make modpost STFU and
> let's hope that code is really OK".
The above function is suposed to be used solely from
iosapic_alloc_rte() and not for general use.
And instead of declaring all of iosapic_alloc_rte() as __init_refok
then the single line doing the call to an __init function
was factored out and marked __init.
alloc_rte() can be used from everywhere but the naming suggest
that this is for rte only.
To make it more clear what the function was used for I could have made
it like this:
/*
* Call site shall make sure this isused only during early init.
* Use __init_refok to avoid warnings from modpost.
*/
static void * __init_refok alloc_rte(unsigned long size)
{
return alloc_bootmen(sizeof(struct iosapic_rte_info) * size);
}
Would that make the intention more clear?
As for marking all of iosapic_alloc_rte() __init_refok this was not done
because that would then not detect any new __init usage within the function.
That could be done too obviously - it is anyway only 30 loc.
Sam
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
