[dpdk-dev] Windows community call: MoM 2021-10-27

[dpdk-dev] Windows community call: MoM 2021-10-27

[Dmitry Kozlyuk]

# About

The meeting takes place in MS Teams every two weeks on Wednesday 15:00 UTC.
Note: it is going to be rescheduled.
Ask Harini Ramakrishnan <Harini.Ramakrishnan@microsoft.com> for invitation.


# Agenda

* Patch review
* Opens


# 1. Patch review

1.1. [kmods,v2] windows/netuio: add Intel device ID (William Tu)
     http://patchwork.dpdk.org/project/dpdk/patch/20211019190102.1903-1-u9012063@gmail.com/

     Ready to be merged.

1.2. [v3] eal/windows: ensure all enabled CPUs are counted (Naty)
     http://patchwork.dpdk.org/project/dpdk/patch/1629294360-5737-1-git-send-email-navasile@linux.microsoft.com/

     Merged.

1.3. Support MLX5 crypto driver on Windows (Tal)
     http://patchwork.dpdk.org/project/dpdk/list/?series=19951

     * Limited to crypto/mlx5 PMD, doesn't require Windows maintainers review.
     * Issues cross-compiling with MinGW.

1.4. app/test: enable subset of tests on Windows (Jie)
     http://patchwork.dpdk.org/project/dpdk/list/?series=19970

     * v8 sent, needs review.
     * Thomas recommends enabling tests on library-by-library basis.

1.5. eal: Add EAL API for threading (Naty)
     http://patchwork.dpdk.org/project/dpdk/list/?series=19478

     * Failed to integrate in 21.11:

       - Comments came late and require major rework.
       - DmitryK is going to send more comments, although small ones.
       - This blocks the plan to make DPDK 21.11 static build shippable,
         because we still need pthread shim.

     * Can be integrated before the next TLS, because only introduces new API
       and a unit test for it, doesn't break ABI for non-Windows parts.

1.6. Enable the internal EAL thread API
     http://patchwork.dpdk.org/project/dpdk/list/?series=18338

     * Depends on 1.5, not integrated.
     * Cannot be merged fully until the next TLS,
       because it breaks ABI of sync primitives.
     * Needs to be revised:

       - Parts that don't break ABI can be integrated early.
       - This course of action is approved. More time for review and testing.
       - Patches need to be rearranged.

1.7. Intel patches merged.


# 2. Opens

2.1. William Tu:

There is no solution for a Windows guest running on Windows host
to get a performant paravirtual device, like NetVSC in Linux.
The only option is VF passthrough.
William wonders if QEMU on Windows allows that.
Also some customers don't want to enable HyperV role for Windows host.
Resolution: no one has relevant experience, William is going to experiment.

2.2. Dmitry Kozlyuk:

Interrupt support draft is ready, but there are fundamental issues
that may require to rework NetUIO and userspace part.
An email thread is started on the topic
explaining the issue and possible solutions
(if someone is interested but not mentioned, tell DmitryK).

Mark Cheatham (Boulder Imaging) is willing to share info about interrupt
support in their app. However, their case is quite specialized
and logic is implemented in the kernel.