Re: KVM/QEMU Community Call 6th Feb Agenda Items (minutes)

["Philippe Mathieu-Daudé" <philmd@linaro.org>]

Hi,

On 2/2/24 16:55, Alex Bennée wrote:
> 
> The KVM/QEMU community call is at:
> 
>    https://meet.jit.si/kvmcallmeeting
>    @
>    6/2/2024 14:00 UTC
> 
> As I'll be away Philippe has volunteered to run the meeting.
> 
> So far we have one agenda item which is to discuss next steps from
> Markus' post about dynamic modelling:
> 
>    Message-ID: <87o7d1i7ky.fsf@pond.sub.org>
>    Date: Wed, 31 Jan 2024 21:14:21 +0100
>    Subject: Dynamic & heterogeneous machines, initial configuration: problems

I took notes on a document but am unable to share it so far,
so I'm sharing the minutes replying here.

Some people complained the call wasn't announced like previous
ones with Juan calendar invite mail, so I built this Cc list from
the old calendar invite recipient list and mixed with the list from:
https://etherpad.opendev.org/p/qemu-emulation-bof%402022-11-29

Unfortunately exporting broke the document indenting, I tried to
adapt it my best.

---< minutes >---

6 Feb 2024 | QEMU community call

Attendees:
- Philippe Mathieu-Daudé
- Zhao Liu
- Markus Armbruster
- Mark Burton
- Anton Johansson
- Cédric Le Goater
- Bernhard Beschow
- Maciej S. Szmigiero
- Manos Pitsidianakis

Notes
- Discussion around https://lore.kernel.org/all/87o7d1i7ky.fsf@pond.sub.org/
- Future of NUMA topology (current options are too limited)
     - should we add more CLI options
     - Move to QOM containers (better for DynMach)
- Ambitious projects, how to share work?
     - We can not go in all directions at once
     - start a new binary to experiment (no backwards compatibility 
problems)
     - Can we work in parallel? Are there serial steps?
     - Can we avoid working on a fork?
     - Should we avoid QMP as DSL to describe machine
- Convert QEMU code base to libraries?
- Is QDev DeviceRealize sufficient? SystemC steps:
     - Construct
     - Config start
     - Config done
     - Before end of elaboration
     - Wire / Connect
     - After end of elaboration
     - Ready to start simulation
     - Reset?
- Next steps?
     - Add new experimental target: Phil
     - Initial startup: Markus
         - Currently QMP available too late for dynamic machine
         - Define clean QEMU init state machine
     - Shared TCG frontends: Anton
     - Create QDev devices from QMP?
         - Clarify QDev states
     - Start QDev2 in-tree to avoid churn with the current QDev API?

Action items
[ ] Template for this document
[ ] Share this document out of Linaro
[ ] Setup a community calendar invite, Cc’ing interested people or 
letting them subscribe
[ ] Next meeting next week!

---< /minutes >---

In order to share tasks I put the "next steps" items on the wiki:
https://wiki.qemu.org/Dynamic_machine_and_heterogeneous_emulation_roadmap
Feel free to add any relevant steps and where you want to help.

Mark Burton asked if Peter Maydell and Paolo Bonzini could be in the
next week call (Tue 13) to discuss about next steps and whether a
in-tree prototype binary could be accepted in mainstream or not.

A such prototype binary has been proposed here:
https://lore.kernel.org/qemu-devel/20240209152945.25727-1-philmd@linaro.org/
Personally I think at this point there are no blocker so we can
discuss on the list as usual, there is no need for someone specific
presence to debate on this (yet) IMHO.

Still we'll have a call next week and discuss the roadmap. An
any other topics the community want to bring.

Regards,

Phil.