vhost net

• document

  • KVM-Architecture-LK2010 slide p.18

  • Stefan Hajnoczi - QEMU Internals: vhost architecture

  • gaowanlong - KVM irqfd and ioeventfd

    • ioeventfd patch

    • irqfd patch

  • gaowanlong - vhost architecture

  • RoyLuo’s Notes - vhost architecture

  • virtio 后端方案 vhost

  • [PATCHv9 0/3] vhost: a kernel-level virtio server <https://www.spinics.net/lists/kvm/msg25348.html>

• vhost_net.ko

  • Host user space opens and configures kernel helper

  • virtio as guest-host interface

  • KVM interface: eventfd

    • TX trigger => ioeventfd

    • RX signal => irqfd

  • Linux interface via tap or macvtap

  • future: multi-queue vhost-net

Stefan Hajnoczi - QEMU Internals: vhost architecture

• overview

  • vhost driver provide in-kernel virtio devices for KVM. compare to QEMU, it can directly call into kernel subsystem without system call.

  • vhost-net driver emulate virtio network card in the host kernel. (driver/vhost/)

• driver model

  • /dev/vhost-net, QEMU open and init it if parameter has -netdev tap,vhost=on

  • vhost-worker-thread (vhost-<pid>): (1) handle IO event (2) perform device emulation

    • create at initialization time.

• in-kernel virtio emulation

  • vhost doesn’t emulate complete virtio PCI adapter, it restrict itself to virtqueue ops only.

  • vhost worker thread do IO:

    • wait virtqueue kick, recieve packet from tx virtqueue. transmit over tap fd.

    • do fd polling, wake up when packets come in over the tap fd. place packets to rx virtqueue

• vhost as a userspace interface

  • guest kick host 時, 需要有個方法通知 vhost worker thread.

  • virtio 不依賴 KVM module. 所以要透過 eventfd 來通知.

  • ioeventfd:

  • irqfd: allow eventfd to trigger guest interrupts