wrong check for physical function in pci utils

[Moshe Levi]

in pci utils the is_physical_function function check it based on existing virtfn* symbolic link. The check is incorrect because
if the PF doen't enable SR-IOV meaning sriov_numvfs is set to zero there are no virtfn* ljnks and the nova-compute recognize it as VF.

see:
root@r-ufm160:/opt/stack/logs# ls /sys/bus/pci/devices/0000\:03\:00.0/
broken_parity_status d3cold_allowed enable iommu_group modalias pools reset sriov_numvfs uevent
class device infiniband irq msi_bus power resource sriov_totalvfs vendor
commands_cache dma_mask_bits infiniband_cm local_cpulist msi_irqs real_miss resource0 subsystem vpd
config driver infiniband_mad local_cpus net remove resource0_wc subsystem_device
consistent_dma_mask_bits driver_override infiniband_verbs mlx5_num_vfs numa_node rescan sriov subsystem_vendor
root@r-ufm160:/opt/stack/logs# cat /sys/bus/pci/devices/0000\:03\:00.0/sriov_numvfs
0

root@r-ufm160:/opt/stack/logs# echo 4 > /sys/bus/pci/devices/0000\:03\:00.0/sriov_numvfs
root@r-ufm160:/opt/stack/logs# ls /sys/bus/pci/devices/0000\:03\:00.0/
broken_parity_status d3cold_allowed enable iommu_group modalias pools reset sriov_numvfs uevent virtfn3
class device infiniband irq msi_bus power resource sriov_totalvfs vendor vpd
commands_cache dma_mask_bits infiniband_cm local_cpulist msi_irqs real_miss resource0 subsystem virtfn0
config driver infiniband_mad local_cpus net remove resource0_wc subsystem_device virtfn1
consistent_dma_mask_bits driver_override infiniband_verbs mlx5_num_vfs numa_node rescan sriov subsystem_vendor virtfn2

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/227160

[Moshe Levi (moshele)]

I added the n-cpu.log and q-srv.log
the vm failed of vif bind error because nova-cmpute create port with pci_vendor 15b3:1013 which is a pf and not a vf
015-11-08 14:04:11.477 DEBUG neutron.plugins.ml2.drivers.mech_sriov.mech_driver.mech_driver [req-8dc2102e-df03-454a-a0e9-87ad58a7db56 neutron service] Unsupported pci_vendor 15b3:1013 _check_supported_pci_vendor_device /opt/stack/neutron/neutron/plugins/ml2/drivers/mech_sriov/mech_driver/mech_driver.py:171

see:
03:00.0 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4]
03:00.1 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4]
03:00.2 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4 Virtual Function]
03:00.3 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4 Virtual Function]
03:00.4 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4 Virtual Function]
03:00.5 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4 Virtual Function]

03:00.0 0200: 15b3:1013
03:00.1 0200: 15b3:1013
03:00.2 0200: 15b3:1014
03:00.3 0200: 15b3:1014
03:00.4 0200: 15b3:1014
03:00.5 0200: 15b3:1014

[Moshe Levi (moshele)]

the is the pci_passthrough_whitelist in nova
pci_passthrough_whitelist = {"address":"*:03:00.*","physical_network":"physnet1"}

[Baodong (Robert) Li (baoli)]

I think that the whtelist is not defined properly. If the address "*:03:00.*" includes some PFs, then it shouldn't use "*" in the address.

The is_physical_funciton() won't be called for the above whitelist entry since the entry contains "*".

Check https://github.com/openstack/nova/blob/master/nova/pci/manager.py#L115
https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L4806

According to the the whitelist entry, the PF will be covered by the entry, But the PF shouldn't be added by the caller from the above two references.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/227160
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=2ba4644f91aa523c2a14e32a168b853cf0b8c4e1
Submitter: Jenkins
Branch: master

commit 2ba4644f91aa523c2a14e32a168b853cf0b8c4e1
Author: Moshe Levi <email address hidden>
Date: Wed Sep 23 02:49:28 2015 +0300

    libvirt: report pci Type-PF type even when VFs are disabled

    libvirt < 1.3 reports virt_functions capability only when pf has
    VFs enabled. This workaround patch updates the is_physical_function
    function to read the sriov_totalvfs if exists and check it is
    greater than 0. The sriov_totalvfs is the number for the
    maximum possible VF for this PF. _get_pcidev_info in libvirt driver
    is updated to get the correct pci device type using this function.

    Closes-Bug: #1499204
    Change-Id: I8990c36fb1d6c66093a465930ff3f0948dd64986

[Thierry Carrez (ttx)]

This issue was fixed in the openstack/nova 13.0.0.0b2 development milestone.