[openfabrics-ewg] RHEL5 and OFED ...

[Doug Ledford]

On Sun, 2006-10-15 at 17:59 +0200, Michael S. Tsirkin wrote:
> Quoting r. Doug Ledford <dledford at redhat.com>:
> > Subject: Re: RHEL5 and OFED ...
> > 
> > On Sat, 2006-10-14 at 22:14 +0200, Michael S. Tsirkin wrote:
> > > Quoting r. Doug Ledford <dledford at redhat.com>:
> > > > Sorry.  RHEL5 Beta1 has been out for a while, but OFED 1.1 still isn't
> > > > done yet.  Obviously, I wasn't able to get something in RHEL5 that
> > > > didn't even exist prior to freeze.
> > > 
> > > Would it be possible to include patches backporting fixes in infiniband kernel
> > > components from 2.6.18/OFED 1.1 to modules that already ship with RHEL5?
> > 
> > Maybe.  It would depend on the patch.  Of course, keep in mind that
> > RHEL5 Beta1 *has* a 2.6.18 kernel.
> 
> Really? That's way cool.
> How does one generate patches for you? Patch against vanilla 2.6.18?

The best way is to install the kernel src.rpm, then go
into /usr/src/redhat/SPECS and rpmbuild --bp kernel-2.6.spec and then go
into /usr/src/redhat/BUILD/kernel-2.6.18/linux-2.6.18.noarch and patch
the appropriate files there.  Save off originals with a unique
extension, then use gendiff to create the patch.

> There are three kind of patches we might want:
> 
> 1. The first thing would be to list fixes between 2.6.18 and 2.6.19-rc1 and
>    backport these. Some of them are in OFED.

That would be helpful.  Since 2.6.19-rc looks to have integrated the
iWARP merge, the fixes are no doubt mixed in with a bunch of new code,
so I didn't pull anything from 2.6.19-rc since I was likely to break
things.  Targeted fixes that skip the iWARP changes from someone that
knows them would be helpful.

> 2. I think in cma module there are 2 main interfaces in OFED 1.1 missing in 2.6.18:
>    API to set timeouts and retry count, and rdma_establish to handle passive side

OK.

> 3. Finally, OFED has some out of kernel modules (mostly UCMA and SDP)
>    that it would be nice to add

The UCMA module is indeed missing.  The SDP module is present, but it's
the older OFED 1.0 code that could stand to be updated.

> Where does the line pass, approximately?  Backporting does take work, so I'd
> like to get a feeling on what's likely to get accepted before I commit to this.

Right now I already have a bugzilla against RHEL5 beta about the missing
UCMA, so that needs fixing anyway.  A reproducer for the SDP problems in
the OFED 1.0 stack will give me the ability to open an SDP bugzilla
against RHEL5 and fix it up.

> Now for userspace - does RHEL5 include at least libibverbs-1.0?
> This has been released a while back, and Roland makes regular bugfix releases.

It includes the OFED 1.0 libibverbs (which makes openmpi complain about
lack of out of band data support, but otherwise seems to work).

As a side note, it includes the OFED 1.0 libehca and ib_ehca kernel
module.  I didn't update the kernel code to the OFED 1.1 ehca because I
didn't know if it would introduce compatibility problems with the user
space library.  However, it might be worthwhile to update that as well
given the amount of change that has occurred and the difference in
testing between the two.

-- 
Doug Ledford <dledford at redhat.com>
              GPG KeyID: CFBFF194
              http://people.redhat.com/dledford

Infiniband specific RPMs available at
              http://people.redhat.com/dledford/Infiniband
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20061015/ac76133d/attachment.sig>