Buscar

Este contenido no está disponible en el idioma seleccionado.

E.2. A Deep-dive into IOMMU Groups

download PDF
An IOMMU group is defined as the smallest set of devices that can be considered isolated from the IOMMU’s perspective. The first step to achieve isolation is granularity. If the IOMMU cannot differentiate devices into separate IOVA spaces, they are not isolated. For example, if multiple devices attempt to alias to the same IOVA space, the IOMMU is not able to distinguish between them. This is the reason why a typical x86 PC will group all conventional-PCI devices together, with all of them aliased to the same requester ID, the PCIe-to-PCI bridge. Legacy KVM device assignment allows a user to assign these conventional-PCI devices separately, but the configuration fails because the IOMMU cannot distinguish between the devices. As VFIO is governed by IOMMU groups, it prevents any configuration that violates this most basic requirement of IOMMU granularity.
The next step is to determine whether the transactions from the device actually reach the IOMMU. The PCIe specification allows for transactions to be re-routed within the interconnect fabric. A PCIe downstream port can re-route a transaction from one downstream device to another. The downstream ports of a PCIe switch may be interconnected to allow re-routing from one port to another. Even within a multifunction endpoint device, a transaction from one function may be delivered directly to another function. These transactions from one device to another are called peer-to-peer transactions and can destroy the isolation of devices operating in separate IOVA spaces. Imagine for instance, if the network interface card assigned to a guest virtual machine, attempts a DMA write operation to a virtual address within its own IOVA space. However in the physical space, that same address belongs to a peer disk controller owned by the host. As the IOVA to physical translation for the device is only performed at the IOMMU, any interconnect attempting to optimize the data path of that transaction could mistakenly redirect the DMA write operation to the disk controller before it gets to the IOMMU for translation.
To solve this problem, the PCI Express specification includes support for PCIe Access Control Services (ACS), which provides visibility and control of these redirects. This is an essential component for isolating devices from one another, which is often missing in interconnects and multifunction endpoints. Without ACS support at every level from the device to the IOMMU, it must be assumed that redirection is possible. This will, therefore, break the isolation of all devices below the point lacking ACS support in the PCI topology. IOMMU groups in a PCI environment take this isolation into account, grouping together devices which are capable of untranslated peer-to-peer DMA.
In summary, the IOMMU group represents the smallest set of devices for which the IOMMU has visibility and which is isolated from other groups. VFIO uses this information to enforce safe ownership of devices for user space. With the exception of bridges, root ports, and switches (all examples of interconnect fabric), all devices within an IOMMU group must be bound to a VFIO device driver or known safe stub driver. For PCI, these drivers are vfio-pci and pci-stub. pci-stub is allowed simply because it is known that the host does not interact with devices via this driver[2]. If an error occurs indicating the group is not viable when using VFIO, it means that all of the devices in the group need to be bound to an appropriate host driver. Using virsh nodedev-dumpxml to explore the composition of an IOMMU group and virsh nodedev-detach to bind devices to VFIO compatible drivers, will help resolve such problems.


[2] The exception is legacy KVM device assignment, which often interacts with the device while bound to the pci-stub driver. Red Hat Enterprise Linux 7 does not include legacy KVM device assignment, avoiding this interaction and potential conflict. Therefore, mixing the use of VFIO and legacy KVM device assignment within the same IOMMU group is not recommended.
Red Hat logoGithubRedditYoutubeTwitter

Aprender

Pruebe, compre y venda

Comunidades

Acerca de la documentación de Red Hat

Ayudamos a los usuarios de Red Hat a innovar y alcanzar sus objetivos con nuestros productos y servicios con contenido en el que pueden confiar.

Hacer que el código abierto sea más inclusivo

Red Hat se compromete a reemplazar el lenguaje problemático en nuestro código, documentación y propiedades web. Para más detalles, consulte el Blog de Red Hat.

Acerca de Red Hat

Ofrecemos soluciones reforzadas que facilitan a las empresas trabajar en plataformas y entornos, desde el centro de datos central hasta el perímetro de la red.

© 2024 Red Hat, Inc.