[nailgun-agent] VNICs ignored due to SR-IOV filtering

[Alexandru Avadanii]

Some blades have VNICs (VFs on a PCI device), which are not SR-IOV VFs.

Random Documentation: [1]
Sample lspci output: [2]
Problematic code (filtering all VFs, not only SR-IOV): [3], introduced by [4]

Steps to reproduce:
1. Build and deploy ISO for a cluster containing nodes with VNICs (like Cavium ThunderX)
2. Bootstrap target nodes and wait for them to be detected

Expected results:
All ethernet interfaces should be detects (for the above example, 3 x VNIC + 1 x PCI NIC)

Actual results:
Only PCI NIC is detected (3 VNICs are trimmed by problematic filter [3])

Description of the environment:
 Operation system: Ubuntu Trusty 14.04 (arm64)
 Versions of components: stable/mitaka
 Related projects installed: Fuel@OPNFV

Proposed possible changes:
- change filter to check for "virtfn", "sriov_totalvfs" or "sriov_numvfs" (see [5] for related docs);
- whitelist entries based on vendor ID;

BR,
Alex

[1] http://chmod666.org/index.php/a-first-look-at-sriov-vnic-adapters/
[2] http://paste.openstack.org/show/524126/
[3] https://github.com/openstack/fuel-nailgun-agent/blob/stable/mitaka/agent#L409
[4] https://github.com/openstack/fuel-nailgun-agent/commit/01ec3c5501c4275063e81323116d4b65893a220f
[5] https://www.kernel.org/doc/Documentation/ABI/testing/sysfs-bus-pci

[Stanislaw Bogatkin (sbogatkin)]

Hi Alexandru,
I believe that this bug was fixed by commit f20415917c3110144c33b646806554ae6dd58901 [0] in 9.1 release cycle, so I closed this bug as invalid for now. If this bug is still applicable, please, reopen this bug and attach your sysfs directory files list - it would help us a lot.

[0] https://bugs.launchpad.net/fuel/+bug/1564630

[Alexandru Avadanii (alexandru-avadanii)]

Hi, Stanislaw,
I do believe this bug is still applicable - see sysfs dir list for one such interface [1].

The problem is that on certain AArch64 hardware, the interfaces are VFs, so physfn is present, but sriov_totalvfs & co are not.

However, my initial suggestion to use sriov_totalvfs is not correct, as that would filter-out interfaces that support SR-IOV, not only SR-IOV devices ...

The compromise I found to work is now submitted in Armband via [2].

Overall, I think this bug does not affect x86, so I can keep the patch in Armband, since Fuel@Openstack does not officially supporting any other arch.

Thanks,
Alex

[1] http://paste.openstack.org/show/599451/
[2] https://gerrit.opnfv.org/gerrit/#/c/28977/

[Alexandru Avadanii (alexandru-avadanii)]

Also, device/physfn dirlist - [3]
Please let me know if I should provide anything else, if not for fixing this, just for properly documenting it.

[3] http://paste.openstack.org/show/599453/