[lkml]   [2020]   [Mar]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Patch in this message
Subject[PATCH v10 11/11] vfio: Document nested stage control
The VFIO API was enhanced to support nested stage control: a bunch of
new iotcls, one DMA FAULT region and an associated specific IRQ.

Let's document the process to follow to set up nested mode.

Signed-off-by: Eric Auger <>


v8 -> v9:
- new layout for the DMA FAULT memory region and specific IRQ

v2 -> v3:
- document the new fault API

v1 -> v2:
- use the new ioctl names
- add doc related to fault handling
Documentation/driver-api/vfio.rst | 77 +++++++++++++++++++++++++++++++
1 file changed, 77 insertions(+)

diff --git a/Documentation/driver-api/vfio.rst b/Documentation/driver-api/vfio.rst
index f1a4d3c3ba0b..563ebcec9224 100644
--- a/Documentation/driver-api/vfio.rst
+++ b/Documentation/driver-api/vfio.rst
@@ -239,6 +239,83 @@ group and can access them as follows::
/* Gratuitous device reset and go... */
ioctl(device, VFIO_DEVICE_RESET);

+IOMMU Dual Stage Control
+Some IOMMUs support 2 stages/levels of translation. "Stage" corresponds to
+the ARM terminology while "level" corresponds to Intel's VTD terminology. In
+the following text we use either without distinction.
+This is useful when the guest is exposed with a virtual IOMMU and some
+devices are assigned to the guest through VFIO. Then the guest OS can use
+stage 1 (IOVA -> GPA), while the hypervisor uses stage 2 for VM isolation
+(GPA -> HPA).
+The guest gets ownership of the stage 1 page tables and also owns stage 1
+configuration structures. The hypervisor owns the root configuration structure
+(for security reason), including stage 2 configuration. This works as long
+configuration structures and page table format are compatible between the
+virtual IOMMU and the physical IOMMU.
+Assuming the HW supports it, this nested mode is selected by choosing the
+This forces the hypervisor to use the stage 2, leaving stage 1 available for
+guest usage.
+Once groups are attached to the container, the guest stage 1 translation
+configuration data can be passed to VFIO by using
+ioctl(container, VFIO_IOMMU_SET_PASID_TABLE, &pasid_table_info);
+This allows to combine the guest stage 1 configuration structure along with
+the hypervisor stage 2 configuration structure. Stage 1 configuration
+structures are dependent on the IOMMU type.
+As the stage 1 translation is fully delegated to the HW, translation faults
+encountered during the translation process need to be propagated up to
+the virtualizer and re-injected into the guest.
+The userspace must be prepared to receive faults. The VFIO-PCI device
+exposes one dedicated DMA FAULT region: it contains a ring buffer and
+its header that allows to manage the head/tail indices. The region is
+identified by the following index/subindex:
+region capability that allows the userspace to retrieve the ABI version
+of the fault records filled by the host.
+On top of that region, the userspace can be notified whenever a fault
+occurs at the physical level. It can use the VFIO_IRQ_TYPE_NESTED/
+VFIO_IRQ_SUBTYPE_DMA_FAULT specific IRQ to attach the eventfd to be
+The ring buffer containing the fault records can be mmapped. When
+the userspace consumes a fault in the queue, it should increment
+the consumer index to allow new fault records to replace the used ones.
+The queue size and the entry size can be retrieved in the header.
+The tail index should never overshoot the producer index as in any
+other circular buffer scheme. Also it must be less than the queue size
+otherwise the change fails.
+When the guest invalidates stage 1 related caches, invalidations must be
+forwarded to the host through
+ioctl(container, VFIO_IOMMU_CACHE_INVALIDATE, &inv_data);
+Those invalidations can happen at various granularity levels, page, context, ...
+The ARM SMMU specification introduces another challenge: MSIs are translated by
+both the virtual SMMU and the physical SMMU. To build a nested mapping for the
+IOVA programmed into the assigned device, the guest needs to pass its IOVA/MSI
+doorbell GPA binding to the host. Then the hypervisor can build a nested stage 2
+binding eventually translating into the physical MSI doorbell.
+This is achieved by calling
+ioctl(container, VFIO_IOMMU_SET_MSI_BINDING, &guest_binding);

 \ /
  Last update: 2020-03-20 17:21    [W:0.164 / U:3.624 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site