[dpdk-dev] [Bug 578] i40evf_check_api_version(): PF/VF API version mismatch:(0.0)-(1.1) with multiple DPDK instances using different VFs with the same PF

[bugzilla at dpdk.org]

https://bugs.dpdk.org/show_bug.cgi?id=578

            Bug ID: 578
           Summary: i40evf_check_api_version(): PF/VF API version
                    mismatch:(0.0)-(1.1) with multiple DPDK instances
                    using different VFs with the same PF
           Product: DPDK
           Version: 20.08
          Hardware: All
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: Normal
         Component: ethdev
          Assignee: dev at dpdk.org
          Reporter: juraj.linkes at pantheon.tech
  Target Milestone: ---

ISSUE DESCRIPTION:
VPP uses DPDK during startup with these init args:
EAL init args: -c 200000000 -n 4 --in-memory --log-level debug --file-prefix
vpp -w 0000:91:03.0 -w 0000:91:03.1 --master-lcore 33

When there's only one VPP starting up, the initialization is successful. When
we add multiple VPPs the initialization sometimes fails:
2020/11/16 08:58:49:660 notice     dpdk           EAL: Detected 64 lcore(s)
2020/11/16 08:58:49:660 notice     dpdk           EAL: Detected 2 NUMA nodes
2020/11/16 08:58:49:660 notice     dpdk           EAL: Selected IOVA mode 'VA'
2020/11/16 08:58:49:660 notice     dpdk           EAL: Probing VFIO support...
2020/11/16 08:58:49:660 notice     dpdk           EAL: VFIO support initialized
2020/11/16 08:58:49:660 notice     dpdk           EAL:   using IOMMU type 1
(Type 1)
2020/11/16 08:58:49:660 notice     dpdk           EAL: Probe PCI driver:
net_i40e_vf (8086:154c) device: 0000:91:03.0 (socket 1)
2020/11/16 08:58:49:660 notice     dpdk           i40evf_check_api_version():
PF/VF API version mismatch:(0.0)-(1.1)
2020/11/16 08:58:49:660 notice     dpdk           i40evf_init_vf(): check_api
version failed
2020/11/16 08:58:49:660 notice     dpdk           i40evf_dev_init(): Init vf
failed
2020/11/16 08:58:49:660 notice     dpdk           EAL: Releasing pci mapped
resource for 0000:91:03.0
2020/11/16 08:58:49:660 notice     dpdk           EAL: Calling
pci_unmap_resource for 0000:91:03.0 at 0x2101000000
2020/11/16 08:58:49:660 notice     dpdk           EAL: Calling
pci_unmap_resource for 0000:91:03.0 at 0x2101010000
2020/11/16 08:58:49:660 notice     dpdk           EAL: Requested device
0000:91:03.0 cannot be used
2020/11/16 08:58:49:660 notice     dpdk           EAL:   using IOMMU type 1
(Type 1)


EXPANDED SETUP/FAILURE EXPLANATION:
There is one testsuite running tests serially in one Jenkins job. Each test
starts VPP anew, so the DPDK initialization happens multiple times during a
testsuite run. When there is only one Jenkins job running, no failure is
observed. With multiple Jenkins jobs (and multiple testsuites running in
parallel), there is a small likelihood that a small number (usually just 1) of
initialization failures will occur, suggesting that when multiple DPDK
instances are retrieving PF/VF version there's a race condition resulting in PF
version 0.0 being retrieved.
Each Jenkins job uses two different VFs from the same PF.

An illustration of what likely happens:
VPP1 in job1 starts and DPDK initializes VF1 and VF2 from a PF.
VPP2 in job2 starts and DPDK initializes VF3 and VF4 from the same PF.
One or more of VF1, VF2, VF3 or VF4 don't get properly initialized because of:
i40evf_check_api_version(): PF/VF API version mismatch:(0.0)-(1.1)

NIC:
Intel Ethernet Converged Network Adapter XL710-Q2
cat /sys/class/net/enp5s0f0/device/device 
0x1583
cat /sys/class/net/enp5s0f0/device/vendor 
0x8086
cat /sys/class/net/enp5s0f0/device/subsystem_device 
0x0001
cat /sys/class/net/enp5s0f0/device/subsystem_vendor 
0x8086

DRIVER/FIRMWARE VERSION:
driver: i40e
version: 2.1.14-k
firmware-version: 6.01 0x800035da 1.1747.0
expansion-rom-version: 
bus-info: 0000:05:00.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: yes

-- 
You are receiving this mail because:
You are the assignee for the bug.