Page tree

Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.
Comment: Today's minutes


Presentations & Minutes

...

Minute taking rota

...


May 9

Upcoming meeting

Apologies

  • Kimmo, traveling 

...


May 2

Particicpants

  • Stephen
  • Kimmo
  • Gunnar
  • Harsha
  • Philippe

Apologies

  • Eugen (vacation - mentioned last week)

Minutes 

White Paper

  • Going through great new input from Harsha, texts and pictures.
  • GPU sharing..

Minutes TBD

  • . hardware assist.
  • Stephen:  It's also possible to have multiple register sets per OS, so that they are completely separated.
  • Gunnar: How is the OS ID actually managed, i.e. what actually "sets" the OS ID when VMs are switched (to avoid spoofing this must be in some privileged execution mode?, in other words some minimal Hypervisor must be involved to influence this?).
  • Stephen: Not fully sure on the details of that exactly but for some (other) aspects there is some paravirtualization involved, e.g. there are minor changes to libdrm.
  • Gunnar: I'm only missing a description of a full virtualization of GPU, where the hardware has little or no special hardware support for it?   I'm unaware if this is practically done (in embedded/automotive), outside of desktop OS virtualization like VirtualBox/VMWare?
  • Harsha: The virtio-gpu 3D chapter that follows describes the host/guest principle.
  • Gunnar:  OK, yes you are right, this is more what VirtualBox/VMWare on desktop is doing.  There is a clear host & guest role there.
  • Gunnar: But what about if the VMs are on equal standing, just running on some full virtualization / emulated GPU.  Again, do we have practical examples of this - otherwise it could be more of a minor note.
  • Stephen: I  think Intel graphics has some kind of (hardware) master that arbitrates the separation.  But that is a kind of partial hardware assist, in between full and advanced features found in other SoCs, and no hardware virt support at all.
  • Stephen: Integrity has some other implementation, but the end result includes combining graphics into the final whole.  A bit unclear on the implementation.

AMM Workshop

  • Going through AMM workshop topics.  Android↔Wayland is a topic of interest.  But who will lead this discussion?

TODO

  • Gunnar asked all participants to provide more clear topics, presentations, problem statements to discuss for the workshop.

Other TODOs


Apologies


April 25

Minutes TBD

...