sr-iov · shipping · also SR-IOV, Single Root I/O Virtualization, Single-Root I/O Virtualization

Single Root I/O Virtualization (SR-IOV)

SR-IOV is a PCI Express capability that lets a Physical Function expose dependent Virtual Functions as separately addressable PCI functions.

Related Observatory article: Single Root I/O Virtualization (SR-IOV) on hesela.com

A Virtual Function (VF) has its own PCI configuration-space address and register mapping, not a separate physical drive. Linux v6.12 distinguishes enabling VFs through sriov_numvfs from host-driver probing; neither operation alone assigns a VF to a guest. Set sriov_drivers_autoprobe before enabling VFs; changing it does not affect VFs that have already been probed. [1]

In NVMe Base 2.0e, support for both SR-IOV and Virtualization Enhancements maps Physical Function (PF) controllers to primary controllers and their VFs to secondary controllers. Flexible queue and interrupt resources assigned to a secondary controller are configured while that controller is Offline. A namespace is separate: Figure 19 permits both private and shared namespaces, not one namespace per VF. [2]

The same NVMe edition takes associated secondary controllers Offline when the primary is disabled or undergoes a Controller Level Reset. Function separation is not reset independence. Disabling a VF register mapping is not itself a controller reset: Base 2.0e specifies that clearing SR-IOV Control.VF MSE hides the VF controller properties without resetting their values. For Linux device assignment, DMA isolation also depends on the IOMMU group and platform topology. [2] [3]

Relations

related: PCIe storage attachment, NVMe, NVMe namespace, NVM subsystem

Limitations

Enabling a VF does not establish guest assignment, controller readiness or namespace attachment. [1, 2]
Separately addressable functions do not by themselves establish independent DMA isolation; Linux VFIO retains IOMMU-group ownership constraints. [3]

Citations

Linux v6.12: PCI Express I/O Virtualization Howto, sections 3.1.1, 3.2.1 and 3.3.1 (accessed 2026-11-02)
Linux v6.12: VFIO, Groups, Devices, and IOMMUs; Device cdev ownership (accessed 2026-10-31)

Sources & caveats

Primary-source explanation using pinned editions, not a claim about the latest specification or support in every NVMe device.
No VF was enabled or assigned in this work; no isolation, reset, performance or live-migration experiment was performed.

JSON: /data/terms/sr-iov.json