This is a follow-up to my previous RFC series on this topic. The goal
is to unify how KVM manages guests in the presence of non-coherent DMA
trafic and provide a way for QEMU to register VFIO groups to enable
that support. Since this changes the way KVM handles things like
WBINVD, we use the VFIO external user interface so that KVM can verify
if a user has physical hardware access. For now we assume VFIO domains
are always non-coherent, which is true if the IOMMU driver honors the
IOMMU_CACHE mapping attribute. Once this is in and we figure out what
exactly VFIO should do regarding support for coherent IOMMU domains,
we can create a new interface for KVM to call. This is noted with a
TODO in this patch. I've reworked the device interface to use
attributes and pass an address for the file descriptor since the RFC.
Alexey, I expect that a new attribute to associate an LIOBN to a group
and arch callbacks to make use of that information is all that you'll
need to add for POWER. Thanks,
Alex
---
Alex Williamson (4):
kvm: Destroy & free KVM devices on release
kvm/x86: Convert iommu_flags to iommu_noncoherent
kvm: Create non-coherent DMA registeration
kvm: Add VFIO device for handling IOMMU cache coherency
Documentation/virtual/kvm/devices/vfio.txt | 22 ++
arch/ia64/include/asm/kvm_host.h | 2
arch/powerpc/kvm/book3s_xics.c | 1
arch/x86/include/asm/kvm_host.h | 4
arch/x86/kvm/Makefile | 2
arch/x86/kvm/vmx.c | 3
arch/x86/kvm/x86.c | 21 ++
include/linux/kvm_host.h | 23 ++
include/uapi/linux/kvm.h | 4
virt/kvm/iommu.c | 22 +-
virt/kvm/kvm_main.c | 8 +
virt/kvm/vfio.c | 262 ++++++++++++++++++++++++++++
12 files changed, 355 insertions(+), 19 deletions(-)
create mode 100644 Documentation/virtual/kvm/devices/vfio.txt
create mode 100644 virt/kvm/vfio.c
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
