vhost virtio host-side backends (net, SCSI, vsock, vDPA)
Kernel-side accelerators for virtio devices used by virtual machines: instead of the hypervisor (typically QEMU/KVM) handling every guest I/O request in userspace, vhost moves the data path for networking, SCSI storage, vsock guest-host sockets, and vDPA hardware offload directly into the host kernel for much higher throughput and lower latency.
recommendation
It should stay because vhost is a foundational piece of modern Linux virtualization, used by essentially every KVM/QEMU deployment for fast virtio networking, storage, and vsock, and it is the kernel side of the vDPA hardware-offload story. Upstream activity is healthy, with core patches and bug fixes landing as recently as 2026, and current kernel and QEMU documentation still treat vhost as a first-class path for live migration, virtio-net failover, and even AWS Nitro Enclave-style workflows. There is no in-kernel replacement that covers the same role.
repository signals
sources
- lore.kernel.org
Recent upstream maintenance is ongoing: a vhost core patch ('vhost: Bump max_mem_regions default to 256') touched drivers/vhost/vhost.c on 2026-03-19.
- lore.kernel.org
Bug-fix traffic is still active in 2026: 'vhost: fix vhost_get_avail_idx for a non empty ring' touched drivers/vhost/vhost.c on 2026-03-02.
- kernel.org
Current kernel documentation still uses vhost in supported virtio-net standby/live-migration setups, with libvirt examples showing '<driver name='vhost' queues='4'/>'.
- qemu.org
Current QEMU documentation says VirtIO device implementation can be provided by QEMU together with the kernel ('known as vhost'), indicating vhost remains a first-class deployment path.
- qemu.org
Current QEMU Nitro Enclaves documentation references vhost-vsock/vhost-user-vsock for enclave communication, showing continued modern virtualization use.
codex reasoning notes (technical)
`drivers/vhost` is real driver code (Kconfig exposes VHOST_NET/VHOST_SCSI/VHOST_VSOCK/VHOST_VDPA modules; local shell read of drivers/vhost/Kconfig). lore_file_timeline on drivers/vhost/vhost.c showed heavy 2021-2026 activity and fresh 2026 patches via lore-http. Web open of current kernel docs and QEMU docs showed vhost still documented in active virtio, failover/live-migration, and enclave workflows. This is virtualization infrastructure rather than sellable hardware, so hardware fields are null/false; there is no natural in-kernel replacement covering the same host-backend role.