Heads up: rdma-core dropped support for 32-bit arm

[Orion Poplawski]

FYI:

rdma-core 26.1-1.fc32 dropped support for %arm:

# 32-bit arm is missing required arch-specific memory barriers,
ExcludeArch: %{arm}

This broke dependecies for the arm package of openmpi 
(https://bugzilla.redhat.com/show_bug.cgi?id=1780584)

This may have affected other users of rdma-core, depending of if they 
use rdma on arm.  Using my x86_64 machine:

$ dnf repoquery --whatrequires libibverbs.so.1'()(64bit)' --source
Last metadata expiration check: 0:14:21 ago on Fri 06 Dec 2019 10:35:11 
PM MST.
ceph-14.2.4-3.fc32.src.rpm
dapl-2.1.9-10.fc31.src.rpm
fio-3.16-2.fc32.src.rpm
ga-5.6.5-6.fc31.src.rpm
glusterfs-7.0-1.fc32.src.rpm
libfabric-1.9.0-1.fc32.src.rpm
libiscsi-1.18.0-9.fc32.src.rpm
libocrdma-1.0.8-6.fc27.src.rpm
nwchem-6.8.2-1.fc32.src.rpm
openmpi-3.1.5-1.module_f32+7117+998651d7.src.rpm
orangefs-2.9.7-6.fc31.src.rpm
perftest-4.2-5.fc31.src.rpm
qemu-4.2.0-0.3.rc2.fc32.src.rpm
qperf-0.4.9-16.fc31.src.rpm
rdma-core-26.1-1.fc32.src.rpm
scsi-target-utils-1.0.70-9.fc31.src.rpm
ucx-1.6.1-1.fc32.src.rpm

This has also broken hwloc-devel on arm:

https://bugzilla.redhat.com/show_bug.cgi?id=1780813

Is this a definite hard requirement, or can we have at least a minimal 
rdma-core for arm to avoid having to propagate a bunch of arm 
conditionals down the stack?

-- 
Orion Poplawski
Manager of NWRA Technical Systems          720-772-5637
NWRA, Boulder/CoRA Office             FAX: [phone]
3380 Mitchell Lane                       orion(a)nwra.com
Boulder, CO 80301                 https://www.nwra.com/

[Doug Ledford]

The arm32 platform literally does not support the memory primitives needed to safely to
RDMA.  If we enable the support, and someone uses it, there is nothing we can do to
prevent them running the risk of memory corruption.  So we probably need to exclude arm32
from all these packages, or conditionally make the packages exclude RDMA support on
arm32.

Sent from my iPad

>  On Dec 7, 2019, at 1:04 AM, Orion Poplawski <orion(a)nwra.com&gt;
> wrote:
>  
>  FYI:
>  
>  rdma-core 26.1-1.fc32 dropped support for %arm:
>  
>  # 32-bit arm is missing required arch-specific memory barriers,
>  ExcludeArch: %{arm}
>  
>  This broke dependecies for the arm package of openmpi
> (https://bugzilla.redhat.com/show_bug.cgi?id=1780584)
>  
>  This may have affected other users of rdma-core, depending of if they use rdma on arm. 
> Using my x86_64 machine:
>  
>  $ dnf repoquery --whatrequires libibverbs.so.1'()(64bit)' --source
>  Last metadata expiration check: 0:14:21 ago on Fri 06 Dec 2019 10:35:11 PM MST.
>  ceph-14.2.4-3.fc32.src.rpm
>  dapl-2.1.9-10.fc31.src.rpm
>  fio-3.16-2.fc32.src.rpm
>  ga-5.6.5-6.fc31.src.rpm
>  glusterfs-7.0-1.fc32.src.rpm
>  libfabric-1.9.0-1.fc32.src.rpm
>  libiscsi-1.18.0-9.fc32.src.rpm
>  libocrdma-1.0.8-6.fc27.src.rpm
>  nwchem-6.8.2-1.fc32.src.rpm
>  openmpi-3.1.5-1.module_f32+7117+998651d7.src.rpm
>  orangefs-2.9.7-6.fc31.src.rpm
>  perftest-4.2-5.fc31.src.rpm
>  qemu-4.2.0-0.3.rc2.fc32.src.rpm
>  qperf-0.4.9-16.fc31.src.rpm
>  rdma-core-26.1-1.fc32.src.rpm
>  scsi-target-utils-1.0.70-9.fc31.src.rpm
>  ucx-1.6.1-1.fc32.src.rpm
>  
>  This has also broken hwloc-devel on arm:
>  
>  https://bugzilla.redhat.com/show_bug.cgi?id=1780813
>  
>  Is this a definite hard requirement, or can we have at least a minimal rdma-core for arm
> to avoid having to propagate a bunch of arm conditionals down the stack?
>  
>  -- 
>  Orion Poplawski
>  Manager of NWRA Technical Systems          720-772-5637
>  NWRA, Boulder/CoRA Office             FAX: [phone]
>  3380 Mitchell Lane                       orion(a)nwra.com
>  Boulder, CO 80301                 https://www.nwra.com/
>

[Dominik 'Rathann' Mierzejewski]

On Sunday, 08 December 2019 at 04:36, Doug Ledford wrote:
>  > On Dec 7, 2019, at 1:04 AM, Orion Poplawski
> <orion(a)nwra.com&gt; wrote:
>  > 
>  > FYI:
>  > 
>  > rdma-core 26.1-1.fc32 dropped support for %arm:
>  > 
>  > # 32-bit arm is missing required arch-specific memory barriers,
>  > ExcludeArch: %{arm}
>  > 
>  > This broke dependecies for the arm package of openmpi
>  > (https://bugzilla.redhat.com/show_bug.cgi?id=1780584)
>  > 
>  > This may have affected other users of rdma-core, depending of if
>  > they use rdma on arm.  Using my x86_64 machine:
>  > 
>  > $ dnf repoquery --whatrequires libibverbs.so.1'()(64bit)' --source
>  > Last metadata expiration check: 0:14:21 ago on Fri 06 Dec 2019 10:35:11 PM MST.
>  > ceph-14.2.4-3.fc32.src.rpm
>  > dapl-2.1.9-10.fc31.src.rpm
>  > fio-3.16-2.fc32.src.rpm
>  > ga-5.6.5-6.fc31.src.rpm
>  > glusterfs-7.0-1.fc32.src.rpm
>  > libfabric-1.9.0-1.fc32.src.rpm
>  > libiscsi-1.18.0-9.fc32.src.rpm
>  > libocrdma-1.0.8-6.fc27.src.rpm
>  > nwchem-6.8.2-1.fc32.src.rpm
>  > openmpi-3.1.5-1.module_f32+7117+998651d7.src.rpm
>  > orangefs-2.9.7-6.fc31.src.rpm
>  > perftest-4.2-5.fc31.src.rpm
>  > qemu-4.2.0-0.3.rc2.fc32.src.rpm
>  > qperf-0.4.9-16.fc31.src.rpm
>  > rdma-core-26.1-1.fc32.src.rpm
>  > scsi-target-utils-1.0.70-9.fc31.src.rpm
>  > ucx-1.6.1-1.fc32.src.rpm
>  > 
>  > This has also broken hwloc-devel on arm:
>  > 
>  > https://bugzilla.redhat.com/show_bug.cgi?id=1780813
>  > 
>  > Is this a definite hard requirement, or can we have at least a
>  > minimal rdma-core for arm to avoid having to propagate a bunch of
>  > arm conditionals down the stack?
>
>  The arm32 platform literally does not support the memory primitives
>  needed to safely to RDMA.  If we enable the support, and someone uses
>  it, there is nothing we can do to prevent them running the risk of
>  memory corruption.  So we probably need to exclude arm32 from all
>  these packages, or conditionally make the packages exclude RDMA
>  support on arm32. 
The change to rdma-core.spec in commit b631ce466538bdee6e19be3286fb8cbeb5c73de6:
...
+# 32-bit arm is missing required arch-specific memory barriers,
+ExcludeArch: %{arm}
...

should have been communicated to all depdent package maintainers in
advance, as this is effectively removing the package for armv7hl from
rawhide. Cc'ing the committer.

Why was this built on arm32 before if RDMA is not support arm32?

Regards,
Dominik
-- 
Fedora   https://getfedora.org  |  RPM Fusion  http://rpmfusion.org
There should be a science of discontent. People need hard times and
oppression to develop psychic muscles.
        -- from "Collected Sayings of Muad'Dib" by the Princess Irulan

[Honggang LI]

On Mon, Dec 09, 2019 at 12:17:43PM +0100, Dominik 'Rathann' Mierzejewski wrote:
>  On Sunday, 08 December 2019 at 04:36, Doug Ledford wrote:
>  > > On Dec 7, 2019, at 1:04 AM, Orion Poplawski <orion(a)nwra.com&gt; wrote:
>  > > 
>  > > FYI:
>  > > 
>  > > rdma-core 26.1-1.fc32 dropped support for %arm:
>  > > 
>  > > # 32-bit arm is missing required arch-specific memory barriers,
>  > > ExcludeArch: %{arm}
>  > > 
>  > > This broke dependecies for the arm package of openmpi
>  > > (https://bugzilla.redhat.com/show_bug.cgi?id=1780584)
>  > > 
>  > > This may have affected other users of rdma-core, depending of if
>  > > they use rdma on arm.  Using my x86_64 machine:
>  > > 
>  > > $ dnf repoquery --whatrequires libibverbs.so.1'()(64bit)' --source
>  > > Last metadata expiration check: 0:14:21 ago on Fri 06 Dec 2019 10:35:11 PM
> MST.
>  > > ceph-14.2.4-3.fc32.src.rpm
>  > > dapl-2.1.9-10.fc31.src.rpm
>  > > fio-3.16-2.fc32.src.rpm
>  > > ga-5.6.5-6.fc31.src.rpm
>  > > glusterfs-7.0-1.fc32.src.rpm
>  > > libfabric-1.9.0-1.fc32.src.rpm
>  > > libiscsi-1.18.0-9.fc32.src.rpm
>  > > libocrdma-1.0.8-6.fc27.src.rpm
>  > > nwchem-6.8.2-1.fc32.src.rpm
>  > > openmpi-3.1.5-1.module_f32+7117+998651d7.src.rpm
>  > > orangefs-2.9.7-6.fc31.src.rpm
>  > > perftest-4.2-5.fc31.src.rpm
>  > > qemu-4.2.0-0.3.rc2.fc32.src.rpm
>  > > qperf-0.4.9-16.fc31.src.rpm
>  > > rdma-core-26.1-1.fc32.src.rpm
>  > > scsi-target-utils-1.0.70-9.fc31.src.rpm
>  > > ucx-1.6.1-1.fc32.src.rpm
>  > > 
>  > > This has also broken hwloc-devel on arm:
>  > > 
>  > > https://bugzilla.redhat.com/show_bug.cgi?id=1780813
>  > > 
>  > > Is this a definite hard requirement, or can we have at least a
>  > > minimal rdma-core for arm to avoid having to propagate a bunch of
>  > > arm conditionals down the stack?
>  >
>  > The arm32 platform literally does not support the memory primitives
>  > needed to safely to RDMA.  If we enable the support, and someone uses
>  > it, there is nothing we can do to prevent them running the risk of
>  > memory corruption.  So we probably need to exclude arm32 from all
>  > these packages, or conditionally make the packages exclude RDMA
>  > support on arm32.
>  
>  The change to rdma-core.spec in commit b631ce466538bdee6e19be3286fb8cbeb5c73de6:
>  ...
>  +# 32-bit arm is missing required arch-specific memory barriers,
>  +ExcludeArch: %{arm}
>  ...
>  
>  should have been communicated to all depdent package maintainers in
>  advance, as this is effectively removing the package for armv7hl from
>  rawhide. Cc'ing the committer.
>  
>  Why was this built on arm32 before if RDMA is not support arm32? 
We did not intentionally built rdma stack package for arm32.

https://bugzilla.redhat.com/show_bug.cgi?id=1484155

Please see this bug for details. The first version of rdma-core-12
did not built on arm32. Unfortunately, rdma-core-16.2-1.fc28 had been
built on arm32. It seems enabled by the fedora package building system
by default. I do not mean to blame fedora package building system. It is
our fault did not fix this time, when rdma-core-16.2-1.fc28 built on
arm32.


>  Regards,
>  Dominik
>  -- 
>  Fedora   https://getfedora.org  |  RPM Fusion  http://rpmfusion.org
>  There should be a science of discontent. People need hard times and
>  oppression to develop psychic muscles.
>          -- from "Collected Sayings of Muad'Dib" by the Princess Irulan
>

[Dominik 'Rathann' Mierzejewski]

On Monday, 09 December 2019 at 14:15, Honggang LI wrote:
>  On Mon, Dec 09, 2019 at 12:17:43PM +0100, Dominik 'Rathann'
> Mierzejewski wrote: 
[...]
>  > The change to rdma-core.spec in commit
> b631ce466538bdee6e19be3286fb8cbeb5c73de6:
>  > ...
>  > +# 32-bit arm is missing required arch-specific memory barriers,
>  > +ExcludeArch: %{arm}
>  > ...
>  > 
>  > should have been communicated to all depdent package maintainers in
>  > advance, as this is effectively removing the package for armv7hl from
>  > rawhide. Cc'ing the committer.
>  > 
>  > Why was this built on arm32 before if RDMA is not supported on arm32?
>  
>  We did not intentionally built rdma stack package for arm32. 
You did not intentially disable building on arm32. All Fedora packages
are built on all arches unless explicitly disabled:
https://docs.fedoraproject.org/en-US/packaging-guidelines/#_architecture_...

>  https://bugzilla.redhat.com/show_bug.cgi?id=1484155
>  
>  Please see this bug for details. The first version of rdma-core-12
>  did not built on arm32. Unfortunately, rdma-core-16.2-1.fc28 had been
>  built on arm32. It seems enabled by the fedora package building system
>  by default. I do not mean to blame fedora package building system. It is
>  our fault did not fix this time, when rdma-core-16.2-1.fc28 built on
>  arm32. 
Thanks for the background. I'm not questioning your decision to stop
building rdma for armv7hl, but it needs to be coordinated with the
dependent packages. Please re-enable it, work with the respective
maintainers to disable rdma support on armv7hl and then put ExcludeArch
for armv7hl back in.

Regards,
Dominik
-- 
Fedora   https://getfedora.org  |  RPM Fusion  http://rpmfusion.org
There should be a science of discontent. People need hard times and
oppression to develop psychic muscles.
        -- from "Collected Sayings of Muad'Dib" by the Princess Irulan

[Daniel P. Berrangé]

On Fri, Dec 06, 2019 at 11:04:53PM -0700, Orion Poplawski wrote:
>  FYI:
>  
>  rdma-core 26.1-1.fc32 dropped support for %arm:
>  
>  # 32-bit arm is missing required arch-specific memory barriers,
>  ExcludeArch: %{arm}
>  
>  This broke dependecies for the arm package of openmpi
>  (https://bugzilla.redhat.com/show_bug.cgi?id=1780584)
>  
>  This may have affected other users of rdma-core, depending of if they use
>  rdma on arm.  Using my x86_64 machine:
>  
>  $ dnf repoquery --whatrequires libibverbs.so.1'()(64bit)' --source 
>  qemu-4.2.0-0.3.rc2.fc32.src.rpm 
QEMU was fixed a week ago when this dropped arm32 support
first hit the repos (rhbz#1778517)


Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

[Honggang LI]

On Fri, Dec 06, 2019 at 11:04:53PM -0700, Orion Poplawski wrote:
>  FYI:
>  
>  rdma-core 26.1-1.fc32 dropped support for %arm:
>  
>  # 32-bit arm is missing required arch-specific memory barriers,
>  ExcludeArch: %{arm}
>  
>  This broke dependecies for the arm package of openmpi
>  (https://bugzilla.redhat.com/show_bug.cgi?id=1780584)
>  
>  This may have affected other users of rdma-core, depending of if they use
>  rdma on arm.  Using my x86_64 machine:
>  
>  $ dnf repoquery --whatrequires libibverbs.so.1'()(64bit)' --source
>  Last metadata expiration check: 0:14:21 ago on Fri 06 Dec 2019 10:35:11 PM
>  MST.
>  ceph-14.2.4-3.fc32.src.rpm
>  dapl-2.1.9-10.fc31.src.rpm 
dapl is obsolete in upstream. It should be retired.

>  fio-3.16-2.fc32.src.rpm
>  ga-5.6.5-6.fc31.src.rpm
>  glusterfs-7.0-1.fc32.src.rpm
>  libfabric-1.9.0-1.fc32.src.rpm 
I fix this dependency issue for libfabric.

>  libiscsi-1.18.0-9.fc32.src.rpm
>  libocrdma-1.0.8-6.fc27.src.rpm 
libocrdma is obsolete. It should be retired.

>  nwchem-6.8.2-1.fc32.src.rpm
>  openmpi-3.1.5-1.module_f32+7117+998651d7.src.rpm
>  orangefs-2.9.7-6.fc31.src.rpm
>  perftest-4.2-5.fc31.src.rpm 
I will fix this dependency issue for perftest.

>  qemu-4.2.0-0.3.rc2.fc32.src.rpm
>  qperf-0.4.9-16.fc31.src.rpm 
I will fix this dependency issue for qperf.

>  rdma-core-26.1-1.fc32.src.rpm 
No, it does not impact itself.

>  scsi-target-utils-1.0.70-9.fc31.src.rpm
>  ucx-1.6.1-1.fc32.src.rpm 
I will work with Mellanox to fix this dependency issue.

>  
>  This has also broken hwloc-devel on arm:
>  
>  https://bugzilla.redhat.com/show_bug.cgi?id=1780813
>  
>  Is this a definite hard requirement, or can we have at least a minimal
>  rdma-core for arm to avoid having to propagate a bunch of arm conditionals
>  down the stack?
>  
>  -- 
>  Orion Poplawski
>  Manager of NWRA Technical Systems          720-772-5637
>  NWRA, Boulder/CoRA Office             FAX: [phone]
>  3380 Mitchell Lane                       orion(a)nwra.com
>  Boulder, CO 80301                 https://www.nwra.com/
>

[Dominik 'Rathann' Mierzejewski]

On Tuesday, 10 December 2019 at 10:47, Honggang LI wrote:
[...]
>  I will work with Mellanox to fix this dependency issue. 
Thank you!

Regards,
Dominik
-- 
Fedora   https://getfedora.org  |  RPM Fusion  http://rpmfusion.org
There should be a science of discontent. People need hard times and
oppression to develop psychic muscles.
        -- from "Collected Sayings of Muad'Dib" by the Princess Irulan

[Honggang LI]

On Mon, Dec 09, 2019 at 02:40:29PM +0100, Dominik 'Rathann' Mierzejewski wrote:
>  On Monday, 09 December 2019 at 14:15, Honggang LI wrote:
>  > On Mon, Dec 09, 2019 at 12:17:43PM +0100, Dominik 'Rathann' Mierzejewski
> wrote:
>  [...]
>  > > The change to rdma-core.spec in commit
> b631ce466538bdee6e19be3286fb8cbeb5c73de6:
>  > > ...
>  > > +# 32-bit arm is missing required arch-specific memory barriers,
>  > > +ExcludeArch: %{arm}
>  > > ...
>  > > 
>  > > should have been communicated to all depdent package maintainers in
>  > > advance, as this is effectively removing the package for armv7hl from
>  > > rawhide. Cc'ing the committer.
>  > > 
>  > > Why was this built on arm32 before if RDMA is not supported on arm32?
>  > 
>  > We did not intentionally built rdma stack package for arm32.
>  
>  You did not intentially disable building on arm32. All Fedora packages
>  are built on all arches unless explicitly disabled:
>  https://docs.fedoraproject.org/en-US/packaging-guidelines/#_architecture_...
>  
>  > https://bugzilla.redhat.com/show_bug.cgi?id=1484155
>  > 
>  > Please see this bug for details. The first version of rdma-core-12
>  > did not built on arm32. Unfortunately, rdma-core-16.2-1.fc28 had been
>  > built on arm32. It seems enabled by the fedora package building system
>  > by default. I do not mean to blame fedora package building system. It is
>  > our fault did not fix this time, when rdma-core-16.2-1.fc28 built on
>  > arm32.
>  
>  Thanks for the background. I'm not questioning your decision to stop
>  building rdma for armv7hl, but it needs to be coordinated with the
>  dependent packages. Please re-enable it, work with the respective 
It seems not a good choice for me to re-enable and disable it again.

I'm trying to get an arm32 machine in our lab. And then I will narrow
down all packages impacted by this. Will open bug for all impacted
packages.

>  maintainers to disable rdma support on armv7hl and then put
> ExcludeArch
>  for armv7hl back in.
>  
>  Regards,
>  Dominik
>  -- 
>  Fedora   https://getfedora.org  |  RPM Fusion  http://rpmfusion.org
>  There should be a science of discontent. People need hard times and
>  oppression to develop psychic muscles.
>          -- from "Collected Sayings of Muad'Dib" by the Princess Irulan
>  _______________________________________________
>  devel mailing list -- devel(a)lists.fedoraproject.org
>  To unsubscribe send an email to devel-leave(a)lists.fedoraproject.org
>  Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/
>  List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
>  List Archives:
> https://lists.fedoraproject.org/archives/list/[e-mail]

[Dominik 'Rathann' Mierzejewski]

On Tuesday, 10 December 2019 at 11:14, Honggang LI wrote:
>  On Mon, Dec 09, 2019 at 02:40:29PM +0100, Dominik 'Rathann'
>  Mierzejewski wrote: 
[...]
>  > Thanks for the background. I'm not questioning your decision
> to stop
>  > building rdma for armv7hl, but it needs to be coordinated with the
>  > dependent packages. Please re-enable it, work with the respective
>  
>  It seems not a good choice for me to re-enable and disable it again. 
I hope you can fix the dependency issues quickly. Otherwise you'll be
leaving a number of packages in a state where they're not installable
and the packages depending on them are not rebuildable for an extended
period of time.

>  I'm trying to get an arm32 machine in our lab. And then I will
> narrow
>  down all packages impacted by this. Will open bug for all impacted
>  packages. 
You can use one of the ARM VMs available to Fedora packagers:
https://fedoraproject.org/wiki/Test_Machine_Resources_For_Package_Maintai...

Thanks for trying to do the right thing after all, I appreciate it.

Regards,
Dominik
-- 
Fedora   https://getfedora.org  |  RPM Fusion  http://rpmfusion.org
There should be a science of discontent. People need hard times and
oppression to develop psychic muscles.
        -- from "Collected Sayings of Muad'Dib" by the Princess Irulan

[Daniel P. Berrangé]

On Tue, Dec 10, 2019 at 11:37:06AM +0100, Dominik 'Rathann' Mierzejewski wrote:
>  On Tuesday, 10 December 2019 at 11:14, Honggang LI wrote:
>  > On Mon, Dec 09, 2019 at 02:40:29PM +0100, Dominik 'Rathann'
>  > Mierzejewski wrote:
>  [...]
>  > > Thanks for the background. I'm not questioning your decision to stop
>  > > building rdma for armv7hl, but it needs to be coordinated with the
>  > > dependent packages. Please re-enable it, work with the respective
>  > 
>  > It seems not a good choice for me to re-enable and disable it again.
>  
>  I hope you can fix the dependency issues quickly. Otherwise you'll be
>  leaving a number of packages in a state where they're not installable
>  and the packages depending on them are not rebuildable for an extended
>  period of time. 
I'd expect it to be a trivial fix for any package maintainer who is
affected by this. I doubt any of them have a hard dependnacy on
rdma, so should largely be a case of wrapping the rdma BuildRequires
in an arch conditional to exclude arm7. This change has been in
rawhide for 12 days now, so I'd hope the package maintainers have
already done this themselves.

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

[Honggang LI]

On Fri, Dec 06, 2019 at 11:04:53PM -0700, Orion Poplawski wrote:
>  FYI:
>  
>  rdma-core 26.1-1.fc32 dropped support for %arm:
>  
>  # 32-bit arm is missing required arch-specific memory barriers,
>  ExcludeArch: %{arm}
>  
>  This broke dependecies for the arm package of openmpi
>  (https://bugzilla.redhat.com/show_bug.cgi?id=1780584)
>  
>  This may have affected other users of rdma-core, depending of if they use
>  rdma on arm.  Using my x86_64 machine:
>  
Confirmed only openmpi impacted by this issue. Please see inline comments
for details.

Opened https://bugzilla.redhat.com/show_bug.cgi?id=1781765 for openmpi.

>  $ dnf repoquery --whatrequires libibverbs.so.1'()(64bit)'
> --source
>  Last metadata expiration check: 0:14:21 ago on Fri 06 Dec 2019 10:35:11 PM
>  MST.
>  ceph-14.2.4-3.fc32.src.rpm 
ExcludeArch:	i686 armv7hl

>  dapl-2.1.9-10.fc31.src.rpm 
dapl should be retired.

>  fio-3.16-2.fc32.src.rpm 
%ifnarch %{arm}
BuildRequires:  numactl-devel
BuildRequires:  librdmacm-devel
%endif

>  ga-5.6.5-6.fc31.src.rpm 
ExclusiveArch: %{ix86} x86_64

>  glusterfs-7.0-1.fc32.src.rpm 
# No RDMA Support on armv7hl
%ifarch armv7hl
%global _without_rdma --disable-ibverbs
%endif

>  libfabric-1.9.0-1.fc32.src.rpm 
# RDMA not available on 32-bit
ARM: #1484155
%ifnarch %{arm}
BuildRequires:  libibverbs-devel
BuildRequires:  librdmacm-devel
%endif

>  libiscsi-1.18.0-9.fc32.src.rpm 
%ifnarch %{arm}
BuildRequires: rdma-core-devel
%endif

>  libocrdma-1.0.8-6.fc27.src.rpm 
libocrdma should be retired.

>  nwchem-6.8.2-1.fc32.src.rpm 
ExclusiveArch: x86_64 %{ix86}

>  openmpi-3.1.5-1.module_f32+7117+998651d7.src.rpm
https://bugzilla.redhat.com/show_bug.cgi?id=1781765

>  orangefs-2.9.7-6.fc31.src.rpm 
%ifnarch armv7hl
BuildRequires: libibverbs-devel
%endif

>  perftest-4.2-5.fc31.src.rpm 
ExcludeArch:    s390 s390x %{arm}

>  qemu-4.2.0-0.3.rc2.fc32.src.rpm 
qemu maintainer had fixed this
issue.

>  qperf-0.4.9-16.fc31.src.rpm 
ExcludeArch:    %{arm}

>  rdma-core-26.1-1.fc32.src.rpm 
It does not impact itself.

>  scsi-target-utils-1.0.70-9.fc31.src.rpm 
%ifnarch s390 s390x
%{arm}
%global with_rdma 1
%endif

>  ucx-1.6.1-1.fc32.src.rpm 
# UCX currently supports only the
following architectures
ExclusiveArch: aarch64 ppc64le x86_64

>  
>  This has also broken hwloc-devel on arm:
>  
>  https://bugzilla.redhat.com/show_bug.cgi?id=1780813
>  
>  Is this a definite hard requirement, or can we have at least a minimal
>  rdma-core for arm to avoid having to propagate a bunch of arm conditionals
>  down the stack?
>  
>  -- 
>  Orion Poplawski
>  Manager of NWRA Technical Systems          720-772-5637
>  NWRA, Boulder/CoRA Office             FAX: [phone]
>  3380 Mitchell Lane                       orion(a)nwra.com
>  Boulder, CO 80301                 https://www.nwra.com/
>