IOpassthrough

IOpassthrough, often written as I/O passthrough or PCI passthrough, is a virtualization technique that assigns a physical I/O device directly to a virtual machine, bypassing the host operating system’s I/O stack. The guest gains direct control of the device, including its DMA capabilities, without mediated access through the host drivers.

The mechanism relies on an input/output memory management unit (IOMMU) such as Intel VT-d or AMD-Vi to map the device’s DMA operations into the guest’s memory space, providing isolation between devices and preventing the guest from interfering with the host or other devices. The host must typically expose the device to the guest by detaching it from the host and attaching it to the virtual machine. IOMMU groups determine which devices can be passthrough together, as devices in the same group may require sharing resources.

Common use cases include high-performance networking with direct PCIe NIC access, GPUs for compute or graphics workloads in virtual desktops or HPC environments, and storage controllers for reduced I/O latency. IOpassthrough is frequently used in servers running KVM/QEMU, VMware ESXi, or Microsoft Hyper-V, but setup details vary by hypervisor and hardware.

Requirements and setup typically involve enabling IOMMU support in hardware and firmware, verifying IOMMU groupings, and configuring the hypervisor to detach the device from the host and assign it to the target VM. Inside the guest, appropriate device drivers must be installed to utilize the passthrough device. Some devices may not support clean resets or live migration, and complex configurations can affect portability and stability.

Variants include PCIe passthrough of individual devices and SR-IOV, which presents multiple virtual functions from a single physical NIC; SR-IOV can offer similar performance with different isolation characteristics.