VirtIO/Passthrough GPU on PVH dom0
Owner: huangrui
Time: Mon 3:45 PM 26 Jun +0200 (Europe/Prague) Final
Location: Conference Hall

  1. Some AMDGPU specific requirement on DOM0 such as hardware IOMMU instead of xen-swiotlb (large continuously buffer support for GPU requires hardware IOMMU)
  2. MSIX/MSI interrupt issue on PCIE passthrough for AMDGPU (We are using legacy GSI instead)
  3. The indirect mapping issue of IO request server on host QEMU on Xen for blob memory (“hostmem” on guest for GPU)
  4. memory-backend-memfd on qemu for xen (“-mem-path not supported with Xen”)