This doc introduces the steps to try the vhost-pci implementation.
-
Compilation configuration 1.1 In addition to the configuration options that are required for virtio-net, the following options are required for vhost-pci-net in the guest kernel .config: "CONFIG_HOTPLUG_PCI_ACPI=y", "CONFIG_VHOST_PCI_NET=y"
-
Booting configuration 2.1 Slave guest 2.1.1 Kernel booting add "vhost_pci_net.napi_weight=0", to let the vhost-pci-net driver use polling to receive packets. 2.1.2 QEMU booting add "-chardev socket,id=slave1,server,wait=off,path=/opt/vhost-pci-slave1
-vhost-pci-slave socket,chardev=slave1", to create a vhost-pci-slave 2.2 Master guest 2.2.1 Kernel booting add "virtio_net.napi_weight=0", to let the virtio_net driver use polling to receive packets. 2.2.2 QEMU booting 1) "-m 512M -mem-prealloc -realtime mlock=on
"-object memory-backend-file,id=mem,size=512M,mem-path=/mnt/hugepages,share=on", make sure the master guest memory is less than 1GB. Current qemu doesn't support large hotplugged device memory. 2) " -chardev socket,id=sock2,path=/opt/vhost-pci-slave1
-netdev type=vhost-user,id=net2,chardev=sock2,vhostforce
-device virtio-net-pci,mac=52:54:00:00:00:02,netdev=net2", make sure that the socket path that associates with the virtio-net device is identical to the slave socket path. -
After guest boots UP the net device, for example:
- in the slave guest: ifconfig eth0 192.168.2.11, where eth0 corresponds to the vhost-pci net device; 2) in the master guest: ifconfig eth0 192.168.2.12, where eth1 corresponds to the virtio net device;
-
Test 4.1 Test with Ping Example of Ping from the slave guest: Ping 192.168.2.12 -I eth0
4.2 Test with Netperf Example of netperf from the slave guest: netperf -H 192.168.2.12 -t TCP_STREAM -l 10