On Wed, Jun 01, 2005 at 06:35:44PM -0500, Eric Sandeen wrote:
> Axel Thimm wrote:
> >On Tue, May 31, 2005 at 02:28:23PM -0700, Chris Wedgwood wrote:
> >>On Tue, May 31, 2005 at 09:48:20AM +0200, Axel Thimm wrote:
> >>>is it possible to add xfs support to a 2.6.9 kernel (e.g. the RHEL4
> >>>kernel) as an external kernel module?
> >Hm, any pointers to "how"? ;)
> I've been looking at doing just this; I have a patch that lets xfs build
> as an out-of-tree 2.6 module. I'll send it to you off list, ok? The
> tricky part is getting all the xfs config options communicated to the
> xfs build. Ultimately it'd be nice to have a src.rpm like those on
> livna.org & other place that can easily be rebuilt against various
> kernels, maybe with xfs config options as --defines to the build....
Why not have an xfs.config file as part of the src.rpms instead (that
is cated to the .config file of the kernel built against)?
People seem to hate --defines. :)
And if one wants to change the config he can install the src.rpm, edit
the config and rpmbuild -ba the specfile.
> >>>Would that make sense w/ 4KSTACKS?
> >>it does but it won't give you 8K stacks, the stack size is a property
> >>of the kernel and modules will inherit this
> >Sure, that's clear. What I mean is: If you turn on xfs in RHEL4's
> >kernel is it considered safe with 4KSTACKS? If not, that would make
> >the whole point of building the kernel modules out of the tree
> >lkml and this list sometimes consider NFS & XFS a dangerous
> >combination stackwise. Urban legend or is there truth to it?
> with 4kstacks, any layering of xfs over/under other systems has the
> potential for danger. There are probably still a couple things we could
> do to reduce some of the individual large-stack functions, too.
That would be nice. The idea is to have the kmdl to plug into standard
RHEL kernels, and to not constrain the system to not use NFS, LVM etc.
> Do you have x86_64 support planned? Stack shouldn't be quite such a big
> issue there.
x86_64 is supported at ATrpms at the same level as i386 archs, but the
requests hadn't been x86_64 specific, I think all were for i386 until
now. I'd prefer a solution that works on both platforms (and even ppc
as this is the next arch to include at ATrpms :).
Axel.Thimm at ATrpms.net
Description: PGP signature