I have an M3 Pro MacBook with 18 GB of unified memory and would like to run Ika inside an ARM64 Linux VM using UTM. I am not asking about running Ika natively on macOS.
UTM 4.6+ supports nested virtualization for ARM64 Linux guests on M3 and newer Macs when using the Apple Virtualization backend on macOS 15+. In principle, this should expose /dev/kvm inside the Linux VM.
The proposed setup would be:
macOS on M3 Pro
└── UTM with Apple Virtualization
└── ARM64 Fedora or Debian/Ubuntu
└── Ika/Cuttlefish using nested KVM
Has anyone tested Ika in this configuration?
I am particularly interested in whether the following components are expected to work through UTM:
- Nested KVM for the inner Cuttlefish VM
- Cuttlefish networking and vsock
/dev/dri/render* access
- gfxstream, Vulkan and udmabuf
- Dynamic display resizing
guest_swiftshader as a fallback if accelerated graphics cannot work
Would the existing ARM64 RPM or Debian packages be expected to run once /dev/kvm is available, or does Ika rely on direct physical-GPU capabilities that UTM’s virtual GPU cannot provide?
If accelerated gfxstream is unlikely to work, would a headless or guest_swiftshader configuration still be a reasonable way to run a custom Cuttlefish/AOSP build?
I would be happy to test this setup and provide logs or document the process if there are particular checks or configuration changes you recommend.
I have an M3 Pro MacBook with 18 GB of unified memory and would like to run Ika inside an ARM64 Linux VM using UTM. I am not asking about running Ika natively on macOS.
UTM 4.6+ supports nested virtualization for ARM64 Linux guests on M3 and newer Macs when using the Apple Virtualization backend on macOS 15+. In principle, this should expose
/dev/kvminside the Linux VM.The proposed setup would be:
Has anyone tested Ika in this configuration?
I am particularly interested in whether the following components are expected to work through UTM:
/dev/dri/render*accessguest_swiftshaderas a fallback if accelerated graphics cannot workWould the existing ARM64 RPM or Debian packages be expected to run once
/dev/kvmis available, or does Ika rely on direct physical-GPU capabilities that UTM’s virtual GPU cannot provide?If accelerated gfxstream is unlikely to work, would a headless or
guest_swiftshaderconfiguration still be a reasonable way to run a custom Cuttlefish/AOSP build?I would be happy to test this setup and provide logs or document the process if there are particular checks or configuration changes you recommend.