On 25 Aug 2023, at 13:34, David Marchand wrote:
> On Fri, Aug 25, 2023 at 12:57 PM Eelco Chaudron <echau...@redhat.com> wrote:
>>>
>>> +m4_define([OVS_DPDK_STOP_VSWITCHD],
>>> + [OVS_VSWITCHD_STOP([dnl
>>> +$1";/does not exist. The Open vSwitch kernel module is probably not
>>> loaded./d
>>> +/does not support MTU configuration,/d
>>> +/EAL: No \(available\|free\) .*hugepages reported/d
>>> +/Failed to enable flow control/d
>>> +/Rx checksum offload is not supported on/d
>>> +/TELEMETRY: No legacy callbacks, legacy socket not created/d"])
>>
>> On my VMs (with all patches applied) all the existing dpdk tests and new
>> ones are failing due to the following error messages:
>
> As you mention, the issue predates this current patch and it points at
> a different problem than the logs matching.
I was wrong, it are only the new system tests that are failing, the existing
ones have the --no-pci option.
So I guess you should add the same for the system traffic one, i.e. the
following seems to work:
--- a/tests/system-dpdk-macros.at
+++ b/tests/system-dpdk-macros.at
@@ -147,7 +147,7 @@ m4_define([OVS_TRAFFIC_VSWITCHD_START],
[
OVS_DPDK_PRE_CHECK()
OVS_WAIT_WHILE([ip link show ovs-netdev])
- OVS_DPDK_START([], [--disable-system --no-pci], [$3])
+ OVS_DPDK_START([--no-pci], [--disable-system], [$3])
dnl Add bridges, ports, etc.
OVS_WAIT_WHILE([ip link show br0])
AT_CHECK([ovs-vsctl -- _ADD_BR([br0]) -- $1 m4_if([$2], [], [], [|
uuidfilt])], [0], [$2])
>>
>> +2023-08-25T10:09:42.252Z|00016|dpdk|ERR|eth_virtio_pci_init(): Failed to
>> init PCI device
>> +2023-08-25T10:09:42.252Z|00017|dpdk|ERR|EAL: Requested device 0000:00:05.0
>> cannot be used
>> +2023-08-25T10:09:42Z|00016|dpdk|ERR|eth_virtio_pci_init(): Failed to init
>> PCI device
>> +2023-08-25T10:09:42Z|00017|dpdk|ERR|EAL: Requested device 0000:00:05.0
>> cannot be used
>>
>> Maybe we could exclude them here (this is what I did) or add some --no-pci
>> option (and make sure the ones that do need them have it).
>>
>> @@ -93,6 +93,8 @@ $1";/does not exist. The Open vSwitch kernel module is
>> probably not loaded./d
>> /eth_dev_tap_create(): .*: failed to create multiq qdisc./d
>> /eth_dev_tap_create(): Disabling rte flow support: No such file or
>> directory/d
>> /tap_mp_req_on_rxtx(): Failed to send start req to secondary/d
>> +/eth_virtio_pci_init(): Failed to init PCI device/d
>> +c
>> /does not exist. The Open vSwitch kernel module is probably not loaded./d"])
>> ])
>>
>
> Those warnings are due to the fact that OVS lets DPDK try to grab all
> devices by default.
> And here, we are affected by a port that you don't seem to care about.
>
>
> Besides, about log matching, in the context of the phy port tests, we
> do want to catch errors like "EAL: Requested device .* cannot be
> used".
> You could also imagine someone passing the PCI device of a virtio-net
> for those tests to run so I would not filter the virtio errors.
> So I don't think we should ignore those specific logs in the common macro.
>
>
> Passing --no-pci in the tests that wants to probe one pci device is not
> doable.
>
>
> What I think instead is that having DPDK probes all devices by default
> is not desirable, though it has been working like this since OVS
> integrated DPDK.
> I have a patch in store for a long time, I just never found an
> explicit need for it and it had some drawback in one usecase (multiple
> ports for a single pci device), maybe I could send a rfc.
See above, as it’s a false alarm and we do not need to make this work ;)
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
