[Unionfs] RT patch - again

[Paul Hewlett]

On Tuesday 27 November 2007 02:26:16 Erez Zadok wrote:
> In message <200711261256.49477.paul at gccs.co.za>, Paul Hewlett writes:
> > On Sunday 25 November 2007 20:50:56 Erez Zadok wrote:
> > > In message <200711221317.39711.paul at gccs.co.za>, Paul Hewlett writes:
> > > > Please ignore previous post - here is correct patch (apologies)
> > >
> > > Paul, the problem with this patch, as I alluded to in my previous
> > > email, is that unionfs won't compile unless the RT patches are applied:
> > > the vanilla kernel doesn't have a mapping_nrpages() function.  I'd like
> > > to avoid keeping two sets of unionfs-patches, one for RT users and one
> > > for those who don't use RT.  I could have created a simple wrapper
> > > macro to replace
> >
> > Erez
> >
> > This patch should be generic ...
> >
> > Paul
> >
> >
> > diff -ur unionfs.org/dentry.c unionfs/dentry.c
> > --- unionfs.org/dentry.c        2007-11-26 11:08:22.000000000 +0200
> > +++ unionfs/dentry.c    2007-11-26 11:10:00.000000000 +0200
> > @@ -264,7 +264,11 @@
> >         /* remove all non-private mappings */
> >         unmap_mapping_range(inode->i_mapping, 0, 0, 0);
> >
> > +#ifdef CONFIG_PREEMPT_RT
> > +       if (mapping_nrpages(&inode->i_data))
> > +#else
> >         if (inode->i_data.nrpages)
> > +#endif
> >                 truncate_inode_pages(&inode->i_data, 0);
> >  }
>
> Yes, I did want a single patch, and preferably one I can keep permanently
> w/ the unionfs code.  Alas, whether you do it the way above, or with a
> wrapper macro to mapping_nrpages() which is defined all the time, you still
> get compile-time errors.  The problem is that I want whatever RT patches
> are applied to unionfs, to compile *even* if the RT patches are not
> applied, or if they are applied, then even if CONFIG_PREEMPT_RT is not
> enabled.  Try this:
>
> - apply the RT patch to your kernel
> - apply the above small patch to unionfs
> - configure a kernel *without* RT (i.e., CONFIG_PREEMPT_RT=n)
> - compile and you get this error:
>
>   CC [M]  fs/unionfs/dentry.o
> fs/unionfs/dentry.c: In function 'purge_inode_data':
> fs/unionfs/dentry.c:270: error: 'struct address_space' has no member named
> 'nrpages' make[2]: *** [fs/unionfs/dentry.o] Error 1
> make[1]: *** [fs/unionfs] Error 2
> make: *** [fs] Error 2
>
> The problem is that the RT patches *unconditionally* change the name of the
> field from .nrpages to .__nrpages.  Any solution that will work
> with/without the RT patches, and with/without CONFIG_PREEMPT_RT enabled,
> will have to change that field's name back to its original name.  The only
> solution I see at this point is to re-map the field name at compile time
> using something (ugly) like:
>
> 	CFLAGS=-D__nrpages=nrpages
>
> 
Erez

What would solve this is if the CONFIG variables in the kernel are always 
defined but to either 0 (disabled) or to 1 (enabled) and alas that is not 
going to happen.

#if CONFIG_PREEMPT_RT = 0 
...

Paul



-- 
Paul Hewlett  Technical Director 
Global Call Center Solutions Ltd, 2nd Floor, Milnerton Mall
Cnr Loxton & Koeberg Roads, 7435 Milnerton
www.gccs.co.za
Tel: +27 86 111 3433 Fax: +27 86 111 3520 Cel: +27 76 072 7906
VOIP: 087 750 7474