SCMI taking longer than expected, tustzone work. Running this in its own VM, Do we want to include this under Stratos ? Work has been done in KWG. VG: Have you started on SCMI in guest VM : AZ yes, would like to accelerate this work. VG: SCMI project is working in this area. There is some overlap. I think we can extend this to no secure VM. VG : we have a regular SCMI call, Satyaki has been in some of the calls, took op-tee server and trying to put it into a VM - what does it take to use the same server as a backend in host linux? AZ can have back end in Linux as well, VG: need to think. BM: a diagram would help, AZ presented to every guest.
AZ: need to make this more visible in Linaro
Vincent shared slides ***** link here ****** : manifest questions, watchdogs
Follow up Azzedine and Vincent to talk non-secure SCMI server. Azzedine to generate a couple of slides
Planning to support RB5, will have 1 or 2 people in QC supporting effort.
Has proprietary hypervisor, there are some issue but think we can open up what is needed
Need to ensure these hypervisors needs are covered
hope to have updates next week
AJB : what is upstream support : T - Linux 5.10 has 150 patches accepted, still have about 150 to go. Landing team has the remaining patches hosted.
AB : open firmware ? : T it is closed from thundercomm, has QC TrustZone, you cannot run own EL2 yet. Hope to provide firmware that will support some use cases, crosvm etc. no nested virtualization, can do plain guest with kernel. not N virtual machines at this point. Internal discussion opening to accept patches - good direction to talks. Type one, not planning to promote adoption
Arnd Bergmann How much do we need to be involved in supporting direct HW implementations of a virtio interface
AB pci-virtio from intel being removed, QC have sent patches, there is a discussion about making an endpoint discussion. GIven HW link between two linuxes, you can talk over PCI. Axsis have this working on non PCI device running different system images or perhaps instruction sets.
AZ: this is PCIE specific ? : AB it is intel specific now, but the infrastructure could be used elsewhere.
BM rpmesg - you can put smart in HW, issues around a write that instantaneously changes a CPU register. Can work with synchronous, you may want to break that assumption of immediate feedback
SS did work and use case was not able to support blocking
AB : understood this would be a fourth bus type
AB perhaps not re merge but adopt rpmsg
NXP planning to re present this work. TI has work touches related areas to talk between Linux over PCI etc Mathieu : put it on ice.