On 30/06/2020 18:03, Kevin Traynor wrote:

On 30/06/2020 17:32, Stokes, Ian wrote:

quoted

quoted

-----Original Message-----
From: Kevin Traynor <redacted>
Sent: Tuesday, June 30, 2020 5:21 PM
To: Govindharajan, Hariprasad <redacted>;
stable@dpdk.org; Stokes, Ian [off-list ref]; Loftus, Ciara
[off-list ref]
Cc: dev@dpdk.org; Abhishek Marathe <redacted>;
Akhil Goyal [off-list ref]; Ali Alnubani [off-list ref];
Walker, Benjamin [off-list ref]; David Christensen
[off-list ref]; Hemant Agrawal [off-list ref];
Jerin Jacob [off-list ref]; Mcnamara, John
[off-list ref]; Ju-Hyoung Lee [off-list ref]; Luca
Boccassi [off-list ref]; Pei Zhang [off-list ref]; Yu, PingX
[off-list ref]; Xu, Qian Q [off-list ref]; Raslan Darawsheh
[off-list ref]; Thomas Monjalon [off-list ref]; Peng,
Yuan [off-list ref]; Chen, Zhaoyan [off-list ref]
Subject: Re: [dpdk-dev] 18.11.9 (LTS) patches review and test

On 30/06/2020 16:38, Govindharajan, Hariprasad wrote:

quoted

quoted

-----Original Message-----
From: dev <redacted> On Behalf Of Kevin Traynor
Sent: Friday, June 26, 2020 1:53 PM
To: stable@dpdk.org
Cc: dev@dpdk.org; Abhishek Marathe <redacted>;
Akhil Goyal [off-list ref]; Ali Alnubani
[off-list ref]; Walker, Benjamin [off-list ref];
David Christensen [off-list ref]; Hemant Agrawal
[off-list ref]; Stokes, Ian [off-list ref]; Jerin
Jacob [off-list ref]; Mcnamara, John [off-list ref];
Ju-Hyoung Lee [off-list ref]; Kevin Traynor
[off-list ref]; Luca Boccassi [off-list ref]; Pei Zhang
[off-list ref]; Yu, PingX [off-list ref]; Xu, Qian Q
[off-list ref]; Raslan Darawsheh [off-list ref];
Thomas Monjalon [off-list ref]; Peng, Yuan
[off-list ref]; Chen, Zhaoyan [off-list ref]
Subject: [dpdk-dev] 18.11.9 (LTS) patches review and test

Hi all,

Here is a list of patches targeted for LTS release 18.11.9.

The planned date for the final release is 3rd July.

Please help with testing and validation of your use cases and report
any issues/results with reply-all to this mail. For the final release
the fixes and reported validations will be added to the release notes.

A release candidate tarball can be found at:

    https://dpdk.org/browse/dpdk-stable/tag/?id=v18.11.9-rc2

These patches are located at branch 18.11 of dpdk-stable repo:
    https://dpdk.org/browse/dpdk-stable/

rc2:
Creating this to include important VF fix.
http://inbox.dpdk.org/dev/20200625035046.19820-1-
haiyue.wang@intel.com/

New patches in rc2:
145bb0e780 bus/pci: fix VF memory access
420a437247 net/tap: fix mbuf double free when writev fails

Thanks.

Kevin.

[Govindharajan, Hariprasad] Hi Kevin,

Hi G Hariprasad,

Thanks for testing.

quoted

The following performance and functional tests carried out with ixgbe,
i40e and vhost devices

DPDK 18.11.9 RC2 with OvS 2.11 and OvS 2.12

P2p throughput
P2p multi queue
PVP rss
vHost reconnect
Flow control

Zero  copy feature could not be tested.
From ovs, zero copy is used only vhost user client. So this particular patch

0309429b198800f4d17d578aad650672476ba054 is breaking the zero copy
support in ovs.

quoted

so for ovs to move to support 18.11.9, it should probably remove support for

zero-copy in the ovs code base?
The code says that zero copy is incompatible with vhost client mode.
It seems that the check for client/server was incorrect, and now it is fixed and
correctly detects this incompatible combination and returns an error to OVS.

I don't think we can call that breaking. It was previously incompatible also, it just
didn't return an error as there was a bug.

I don't see an issue with having this patch and the new behaviour.

Ian/Ciara, let me know if I'm missing something.

I think OVS only supports zero copy for vhostuserclient devices. In this case it looks like that’s no longer supported. I guess the real question becomes should the support in OVS be removed now? It is not broken as you said, just no longer supported in the OVS use case.

ah ok, it's not the commit above then, it's this one is the issue

commit 81e025d7ed6a802845909df6fb90505508dc0fbf
Author: Xuan Ding [off-list ref]
Date:   Wed Apr 29 02:59:46 2020 +0000

    vhost: prevent zero-copy with incompatible client mode

    [ upstream commit 715070ea10e6da1169deef2a3ea77ae934b4c333 ]

    In server mode, virtio-user inits under the assumption that vhost-user
    supports a list of features. However, this could be problematic when
    in_order feature is negotiated but not supported by vhost-user when
    enables dequeue_zero_copy later.

    Add handling when vhost-user enables dequeue_zero_copy as client.
--

It's unfortunate that it is both fixing a bug and removing functionality
at the same time :/

Just to wrap up this topic on this thread. We discussed further on
ovs-dev https://mail.openvswitch.org/pipermail/ovs-dev/2020-July/372320.html

The patches that were removing the functionality have now (effectively)
been reverted from 18.11. For 19.11 this will happen too for the next
release.

Ian replied that OVS testing is now passing with latest 18.11 branch
https://mail.openvswitch.org/pipermail/ovs-dev/2020-July/372535.html

quoted

I don’t think it needs to block the release of this, but it's a follow up for the OVS. I believe you'd see the same behavior for OVS master and 2.13.

Yes, these patches are already in 19.11.3.

quoted

BR
Ian

quoted

thanks,
Kevin.

quoted

Regards
G Hariprasad