Add --sanity option and fill in expectaions for Intel X710 #2
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Do not fail test if actual number of Tx descriptor is more than requested, but effective number exactly matches requested. Use verdicts to highlight the behaviour specifics. Signed-off-by: Andrew Rybchenko <[email protected]>
Make it clear in verdicts that expected hash is zero, but obtained value is not zero. Signed-off-by: Andrew Rybchenko <[email protected]>
Helper API to create Rx mempool has some logic which we don't need here. We want to fully control data room size. Signed-off-by: Andrew Rybchenko <[email protected]>
Mapping to URL is site-specific and should live in ts-rigs if present. Signed-off-by: Andrew Rybchenko <[email protected]>
Use new unified format for references in key values. Signed-off-by: Andrew Rybchenko <[email protected]>
DPDK-Bugzilla-ID: 1097 Signed-off-by: Andrew Rybchenko <[email protected]>
DPDK-Bugzilla-ID: 1097 Signed-off-by: Andrew Rybchenko <[email protected]>
Doing so will help to keep RSS configuration scheme consistent across the relevant tests. Signed-off-by: Ivan Malov <[email protected]> Reviewed-by: Andrew Rybchenko <[email protected]>
In the case of net/i40e, a deferred start Tx queue accepts packets instead of rejecting them. The packets get stuck. When the user starts the queue, the packets get sent. Such behaviour might be a bug. Make the test classify it. Signed-off-by: Ivan Malov <[email protected]> Reviewed-by: Andrew Rybchenko <[email protected]>
In the case of net/i40e, RSS hash offload is enabled unconditionally whenever MQ mode RSS is configured. Signed-off-by: Ivan Malov <[email protected]> Reviewed-by: Andrew Rybchenko <[email protected]>
Fixes: f407cf4 ("DPDK TS going public") Signed-off-by: Ivan Malov <[email protected]> Reviewed-by: Andrew Rybchenko <[email protected]>
Doing so will help to keep RSS configuration scheme consistent across the relevant tests. Signed-off-by: Ivan Malov <[email protected]> Reviewed-by: Andrew Rybchenko <[email protected]>
Doing so will help to keep RSS configuration scheme consistent across the relevant tests. Signed-off-by: Ivan Malov <[email protected]> Reviewed-by: Andrew Rybchenko <[email protected]>
net/ice does not support rte_flow_isolate(). Signed-off-by: Andrew Rybchenko <[email protected]>
net/ice uses helper API to prepare packet for checksum offload. The APi requires contiguous packet headers. Signed-off-by: Andrew Rybchenko <[email protected]>
The driver provides RSS hash even if the offload is disabled. Signed-off-by: Andrew Rybchenko <[email protected]>
net/ice does not provide driver callback for rte_eth_dev_flow_ctrl_get(). Signed-off-by: Andrew Rybchenko <[email protected]>
Make it easier to run all tests which use RTE flow rules with transfer attribute set. Signed-off-by: Andrew Rybchenko <[email protected]>
net/ice does not support flow transfer rules. Signed-off-by: Andrew Rybchenko <[email protected]>
net/ice does not support UDP port configuration for GENEVE tunnels. Signed-off-by: Andrew Rybchenko <[email protected]>
Add missing waiting for link up on Tester side before traffic checks. Signed-off-by: Andrew Rybchenko <[email protected]>
Reword verdicts and expect behaviour displayed by upstream DPDK. Signed-off-by: Ivan Malov <[email protected]>
The current code starts CSAP for the 1st time with NULL pattern, then generates mbufs and the real pattern. Reorder the steps so that the real pattern is used. Fixes: f407cf4 ("DPDK TS going public") Signed-off-by: Ivan Malov <[email protected]>
In the case of net/i40e, going for runtime Tx queue setup causes failures on device start. Provide TRC expectations. When the bug has been fixed, the test will fail in a different way because of a bug in deferred Tx queue start. Assuming the partial fix has been applied for that bug too, provide updates for the test which help to classify remaining regressions. Signed-off-by: Ivan Malov <[email protected]>
Doing so will allow to update TRC for more IUT variants. Signed-off-by: Ivan Malov <[email protected]>
Follow up changes in TE [1].
[1] 7feba7708191 ("lib/rpc_dpdk: decouple RSS types from flows")
Signed-off-by: Andrew Rybchenko <[email protected]>
PMD provides RSS hash information even when the offload is not enabled. Signed-off-by: Andrew Rybchenko <[email protected]>
DPDK v22.11-rc1. NIC is MCX516A-CCAT. Signed-off-by: Andrew Rybchenko <[email protected]>
DPDK v22.11-rc1. NIC is MCX516A-CCAT. Signed-off-by: Andrew Rybchenko <[email protected]>
mlx5 changes RSS redirecation table when device is started. Signed-off-by: Andrew Rybchenko <[email protected]>
Not all drivers support MTU change when device is started. So, it is better to avoid and only dedicated tests should do it. Signed-off-by: Andrew Rybchenko <[email protected]>
i40e does not provide tx_pkts and tx_octets extended statistics used by the test. Everything else looks good. Signed-off-by: Andrew Rybchenko <[email protected]>
When packet buffers mempool is populated data room size should include RTE_PKTMBUF_HEADROOM since it is always subtracted when Rx buffer size is calculated. Signed-off-by: Andrew Rybchenko <[email protected]>
Update the test to take min_rx_bufsize into account and expect Rx queue setup failure if requested Rx buffer size is smaller than the minimum. Signed-off-by: Andrew Rybchenko <[email protected]>
When device is not started yet, it is incorrect and not actually required to wait for link up. Fixes: f407cf4 ("DPDK TS going public") Signed-off-by: Andrew Rybchenko <[email protected]>
i40e does not allow to change MTU when device is started. Signed-off-by: Andrew Rybchenko <[email protected]>
i40e fails to start device if an Rx queue is not setup yet. DPDK-Bugzilla-ID: 1110 Signed-off-by: Andrew Rybchenko <[email protected]>
There is a FW LLDP daemon on Intel X710 which generate LLDP packets from time ot time. It breaks ransom tests due to unexpected packet received. Disable FW LLDP daemon to avoid unexpected packets. Signed-off-by: Andrew Rybchenko <[email protected]>
i40e does not provide rx_pkts and rx_octets extended stats. Signed-off-by: Andrew Rybchenko <[email protected]>
Nvidia ConnectX-5 uses bifurcated driver and network interface is available on IUT side as well. However, it does not have disable-fw-lldp private flag and corresponding error was incorrectly returned to the caller. Fixes: 05cfdcc ("prologue: fix random failures on some tests on Intel X710") Signed-off-by: Andrew Rybchenko <[email protected]>
Enable Tx offloads explicitly to let PMD know which offloads are required. Signed-off-by: Andrew Rybchenko <[email protected]>
If IUT port is restarted, wait for link up on Tester side once again to avoid random packet drops because of link down. Fix randon test failures on Intel X710. Signed-off-by: Andrew Rybchenko <[email protected]>
i40e does not fill in TCI's in mbuf if stripping is disabled. Signed-off-by: Andrew Rybchenko <[email protected]>
i40e does not fill in TCI's in mbuf if stripping is disabled. Signed-off-by: Andrew Rybchenko <[email protected]>
|
QEMU virtio-net testing logs: |
Fixes: f407cf4 ("DPDK TS going public") Signed-off-by: Andrew Rybchenko <[email protected]>
Nvidia ConnectX-5 uses bifurcated driver mlx5_core and network interface is present and grabbed by Test Agent. Changing MTU via DPDK changes network interface MTU as well. Chaning MTU to a value smaller than IPv6 minimum MTU make IPv6 on the interface unavailable and related IPv6 instances disappear. Restore using history does not help since test does not change the interface MTU via CS. Configuration restore using backup fails since it tries to restore these IPv6 instances first. The problem is fixed in TE making /agent/sys/net/ipv6 dependent from /agent/interface/mtu. However, it is better to fix it here as well since there is no point to decrease MTU smaller than default. OL-Redmine-Id: 12587 Signed-off-by: Andrew Rybchenko <[email protected]>
Increase default MSS when its value is not important for the test purposes to be bigger than minimum i40e MSS which is 256. Try to make corresponding test iterations useful for i40e. Signed-off-by: Andrew Rybchenko <[email protected]>
Fix many header segments iterations to avoid usage of artificially small MSS which results in many packets as well. Also too small MSS makes the test useless for some PMDs (e.g. i40e) which have bigger minimum MSS. Use so-called default MSS 257 which is bigger than minimum MSS in i40e (256). Use default payload size 1024 to avoid potential issues with dummy TSO which is not the goal of these iterations. Signed-off-by: Andrew Rybchenko <[email protected]>
Both i40e and ice use SW checksum helper which requires contiguous headers. Signed-off-by: Andrew Rybchenko <[email protected]>
Prepare to reuse the helper in another test. Signed-off-by: Andrew Rybchenko <[email protected]>
The function to release AF_PACKET ethdev pretends to handle -ENODEV case, but it lack required waiting for RPC failure. Fixes: f407cf4 ("DPDK TS going public") Signed-off-by: Andrew Rybchenko <[email protected]>
Take into account limits on a number of segment in mbuf to be sent. If constructed mbuf violates these limits, expect Tx preapre to reject it. Signed-off-by: Andrew Rybchenko <[email protected]>
i40e rejects MSS smaller than 256. Signed-off-by: Andrew Rybchenko <[email protected]>
i40e does not support outer UDP checksum offload, but reports the offload as supported. Fill in expectations with references to corresponding DPDK bug. DPDK-Bugzilla-ID: 1125 Signed-off-by: Andrew Rybchenko <[email protected]>
|
I don't really understand what the result should be after adding these two commits: If I understand correctly, this functionality was added to the TE in I've added to my key2html the following lines: It had no effect on the html log. I see that your logs look the same. However, the substitution works in trc-brief.html in my case. However, in your trc-brief.html, substituting links does not work. Could you please tell me what is expected behavior? |
|
It is a TRC key2html. So, it has never worked for HTML logs. It is good that substitution works for you in trc-brief.html. It is the expected behaviour. It is expected that my logs have no substitution, unfortunately. I have no corresponding config locally. I'll fix it. |
|
I have one more PR waiting in a queue. I just don't want to submit it since it overlaps with this one and spoil your efforts here. Please, move forward. |
|
Thanks Andrew! I've pushed your commits with proper review trailers. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Update tests to produce sensible verdicts and take more limitations into account.
Example log:
https://ts-factory.io/bublik/v2/runs/221577