tmakin/ack-tegra
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ANDROID: KVM: arm64: Add documentation for pvIOMMU UAPI

Browse Source 
Document pvIOMMU UAPI under VFIO.

Bug: 357781595
Bug: 348382247
Bug: 236685427
Change-Id: I9efd8bb09e4850d905172c7617b85fe843ee607b
Signed-off-by: Mostafa Saleh <smostafa@google.com>
This commit is contained in:
Mostafa Saleh
2024-07-22 11:34:51 +00:00
committed by Carlos Llamas
parent 53e073efe1
commit 8561efaa1b
2 changed files with 70 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Documentation/driver-api/vfio.rst
+23
View File
@@ -670,6 +670,29 @@ This implementation has some specifics:
-------------------------------------------------------------------------------
pvIOMMU implementation note
-------------------------------
1) pKVM provides mutual distrust between host kernel and protected VMs(pVM)
One solution to provide DMA isolation in this model, is to move the IOMMU
control to the hypervisor and para-virtualize the IOMMU interface for
the host and guest kernels. (none of them have direct access to IOMMU
programming interface).
2) In the case of device assignment, the host can't map memory for the
guest kernel in the IOMMU (as it is not trusted).
3) To statify these requirements a new VFIO IOMMU container type is added
VFIO_PKVM_IOMMU which attaches the device to a blocking domain, and
denies any MAP_DMA/UNMAP_DMA IOCTLs.
4) The guest IOMMU is configured by the host (virtual toplogy) but the mappings
are controlled by the guest, you can find more about this in:
- Documentation/virt/kvm/devices/vfio.rst
- Documentation/virt/kvm/arm/pviommu.rst
So, a new container type is added: `VFIO_PKVM_IOMMU`
-------------------------------------------------------------------------------
.. [1] VFIO was originally an acronym for "Virtual Function I/O" in its
initial implementation by Tom Lyon while as Cisco. We've since
outgrown the acronym, but it's catchy.
Documentation/virt/kvm/devices/vfio.rst
+47
View File
@@ -18,6 +18,7 @@ KVM, a reference to the VFIO file is held by KVM.
Groups:
KVM_DEV_VFIO_FILE
alias: KVM_DEV_VFIO_GROUP
KVM_DEV_VFIO_PVIOMMU
KVM_DEV_VFIO_FILE attributes:
KVM_DEV_VFIO_FILE_ADD: Add a VFIO file (group/device) to VFIO-KVM device
@@ -59,3 +60,49 @@ callback. It is the same for device file descriptor via character device
open which gets device access via VFIO_DEVICE_BIND_IOMMUFD. For such file
descriptors, FILE_ADD should be invoked before VFIO_DEVICE_BIND_IOMMUFD
to support the drivers mentioned in prior sentence as well.
KVM_DEV_VFIO_PVIOMMU attributes:
KVM_DEV_VFIO_PVIOMMU_ATTACH: Create and attach a pvIOMMU instance to the
KVM VM associated with this device, returns a file descriptor "pviommufd"
for this IOMMU which supports some IOCTLs.
KVM_DEV_VFIO_PVIOMMU_GET_INFO: Retrieve information about IOMMUs for a VFIO
devicefd, using the following struct:
struct kvm_vfio_iommu_info {
__u32 size;
__s32 device_fd;
__u32 out_nr_sids;
__u32 __reserved;
};
Where kvm_vfio_iommu_info.device_fd the input VFIO device ID, and
kvm_vfio_iommu_info.out_nr_sids is the number stream IDs(endpoint) for
this device, this similar to VFIO_DEVICE_GET_IRQ_INFO which returns the
number of irqs for a VFIO device.
size is added to allow the struct to be extended, while __reserved
must be zero.
The rest of the IOCTLs are used to configure the pvIOMMU are part of
the pviommufd returned from the attach operation:
KVM_PVIOMMU_SET_CONFIG: Configure pvIOMMU for an endpoint for a device,
where the input struct:
struct kvm_vfio_iommu_config {
__u32 size;
__s32 device_fd;
__u32 sid_idx;
__u32 vsid;
__u32 __reserved;
};
kvm_vfio_iommu_config.device_fd is the VFIO devicefd,
kvm_vfio_iommu_config.sid_idx is a valid index based on number of sids
from KVM_DEV_VFIO_PVIOMMU_GET_INFO.
And kvm_vfio_iommu_config.vsid is the virtual sid seen by the guest,
It is the VMM responsibility to describe this to the guest, the pvIOMMU
would translate an IOMMU request with this vsid to the physical SID of
the device for the index specified.
This is similar to VFIO IOCTL VFIO_DEVICE_SET_IRQS but for the IOMMU.
size is added to allow the struct to be extended, while __reserved
must be zero.