2 Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5 Redistribution and use in source and binary forms, with or without
6 modification, are permitted provided that the following conditions
9 * Redistributions of source code must retain the above copyright
10 notice, this list of conditions and the following disclaimer.
11 * Redistributions in binary form must reproduce the above copyright
12 notice, this list of conditions and the following disclaimer in
13 the documentation and/or other materials provided with the
15 * Neither the name of Intel Corporation nor the names of its
16 contributors may be used to endorse or promote products derived
17 from this software without specific prior written permission.
19 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20 "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
23 OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
25 LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26 DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27 THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
29 OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 The vhost library implements a user space virtio net server allowing the user
35 to manipulate the virtio ring directly. In another words, it allows the user
36 to fetch/put packets from/to the VM virtio net device. To achieve this, a
37 vhost library should be able to:
39 * Access the guest memory:
41 For QEMU, this is done by using the ``-object memory-backend-file,share=on,...``
42 option. Which means QEMU will create a file to serve as the guest RAM.
43 The ``share=on`` option allows another process to map that file, which
44 means it can access the guest RAM.
46 * Know all the necessary information about the vring:
48 Information such as where the available ring is stored. Vhost defines some
49 messages to tell the backend all the information it needs to know how to
52 Currently, there are two ways to pass these messages and as a result there are
53 two Vhost implementations in DPDK: *vhost-cuse* (where the character devices
54 are in user space) and *vhost-user*.
56 Vhost-cuse creates a user space character device and hook to a function ioctl,
57 so that all ioctl commands that are sent from the frontend (QEMU) will be
60 Vhost-user creates a Unix domain socket file through which messages are
65 Since DPDK v2.2, the majority of the development effort has gone into
66 enhancing vhost-user, such as multiple queue, live migration, and
67 reconnect. Thus, it is strongly advised to use vhost-user instead of
74 The following is an overview of the Vhost API functions:
76 * ``rte_vhost_driver_register(path, flags)``
78 This function registers a vhost driver into the system. For vhost-cuse, a
79 ``/dev/path`` character device file will be created. For vhost-user server
80 mode, a Unix domain socket file ``path`` will be created.
82 Currently supported flags are (these are valid for vhost-user only):
84 - ``RTE_VHOST_USER_CLIENT``
86 DPDK vhost-user will act as the client when this flag is given. See below
89 - ``RTE_VHOST_USER_NO_RECONNECT``
91 When DPDK vhost-user acts as the client it will keep trying to reconnect
92 to the server (QEMU) until it succeeds. This is useful in two cases:
94 * When QEMU is not started yet.
95 * When QEMU restarts (for example due to a guest OS reboot).
97 This reconnect option is enabled by default. However, it can be turned off
100 - ``RTE_VHOST_USER_DEQUEUE_ZERO_COPY``
102 Dequeue zero copy will be enabled when this flag is set. It is disabled by
105 There are some truths (including limitations) you might want to know while
108 * zero copy is not good for small packets (typically for packet size below
111 * zero copy is really good for VM2VM case. For iperf between two VMs, the
112 boost could be above 70% (when TSO is enableld).
114 * for VM2NIC case, the ``nb_tx_desc`` has to be small enough: <= 64 if virtio
115 indirect feature is not enabled and <= 128 if it is enabled.
117 The is because when dequeue zero copy is enabled, guest Tx used vring will
118 be updated only when corresponding mbuf is freed. Thus, the nb_tx_desc
119 has to be small enough so that the PMD driver will run out of available
120 Tx descriptors and free mbufs timely. Otherwise, guest Tx vring would be
123 * Guest memory should be backended with huge pages to achieve better
124 performance. Using 1G page size is the best.
126 When dequeue zero copy is enabled, the guest phys address and host phys
127 address mapping has to be established. Using non-huge pages means far
128 more page segments. To make it simple, DPDK vhost does a linear search
129 of those segments, thus the fewer the segments, the quicker we will get
130 the mapping. NOTE: we may speed it by using tree searching in future.
132 * ``rte_vhost_driver_session_start()``
134 This function starts the vhost session loop to handle vhost messages. It
135 starts an infinite loop, therefore it should be called in a dedicated
138 * ``rte_vhost_driver_callback_register(virtio_net_device_ops)``
140 This function registers a set of callbacks, to let DPDK applications take
141 the appropriate action when some events happen. The following events are
144 * ``new_device(int vid)``
146 This callback is invoked when a virtio net device becomes ready. ``vid``
147 is the virtio net device ID.
149 * ``destroy_device(int vid)``
151 This callback is invoked when a virtio net device shuts down (or when the
152 vhost connection is broken).
154 * ``vring_state_changed(int vid, uint16_t queue_id, int enable)``
156 This callback is invoked when a specific queue's state is changed, for
157 example to enabled or disabled.
159 * ``rte_vhost_enqueue_burst(vid, queue_id, pkts, count)``
161 Transmits (enqueues) ``count`` packets from host to guest.
163 * ``rte_vhost_dequeue_burst(vid, queue_id, mbuf_pool, pkts, count)``
165 Receives (dequeues) ``count`` packets from guest, and stored them at ``pkts``.
167 * ``rte_vhost_feature_disable/rte_vhost_feature_enable(feature_mask)``
169 This function disables/enables some features. For example, it can be used to
170 disable mergeable buffers and TSO features, which both are enabled by
174 Vhost Implementations
175 ---------------------
177 Vhost-cuse implementation
178 ~~~~~~~~~~~~~~~~~~~~~~~~~
180 When vSwitch registers the vhost driver, it will register a cuse device driver
181 into the system and creates a character device file. This cuse driver will
182 receive vhost open/release/IOCTL messages from the QEMU simulator.
184 When the open call is received, the vhost driver will create a vhost device
185 for the virtio device in the guest.
187 When the ``VHOST_SET_MEM_TABLE`` ioctl is received, vhost searches the memory
188 region to find the starting user space virtual address that maps the memory of
189 the guest virtual machine. Through this virtual address and the QEMU pid,
190 vhost can find the file QEMU uses to map the guest memory. Vhost maps this
191 file into its address space, in this way vhost can fully access the guest
192 physical memory, which means vhost could access the shared virtio ring and the
193 guest physical address specified in the entry of the ring.
195 The guest virtual machine tells the vhost whether the virtio device is ready
196 for processing or is de-activated through the ``VHOST_NET_SET_BACKEND``
197 message. The registered callback from vSwitch will be called.
199 When the release call is made, vhost will destroy the device.
201 Vhost-user implementation
202 ~~~~~~~~~~~~~~~~~~~~~~~~~
204 Vhost-user uses Unix domain sockets for passing messages. This means the DPDK
205 vhost-user implementation has two options:
207 * DPDK vhost-user acts as the server.
209 DPDK will create a Unix domain socket server file and listen for
210 connections from the frontend.
212 Note, this is the default mode, and the only mode before DPDK v16.07.
215 * DPDK vhost-user acts as the client.
217 Unlike the server mode, this mode doesn't create the socket file;
218 it just tries to connect to the server (which responses to create the
221 When the DPDK vhost-user application restarts, DPDK vhost-user will try to
222 connect to the server again. This is how the "reconnect" feature works.
225 * The "reconnect" feature requires **QEMU v2.7** (or above).
227 * The vhost supported features must be exactly the same before and
228 after the restart. For example, if TSO is disabled and then enabled,
229 nothing will work and issues undefined might happen.
231 No matter which mode is used, once a connection is established, DPDK
232 vhost-user will start receiving and processing vhost messages from QEMU.
234 For messages with a file descriptor, the file descriptor can be used directly
235 in the vhost process as it is already installed by the Unix domain socket.
237 The supported vhost messages are:
239 * ``VHOST_SET_MEM_TABLE``
240 * ``VHOST_SET_VRING_KICK``
241 * ``VHOST_SET_VRING_CALL``
242 * ``VHOST_SET_LOG_FD``
243 * ``VHOST_SET_VRING_ERR``
245 For ``VHOST_SET_MEM_TABLE`` message, QEMU will send information for each
246 memory region and its file descriptor in the ancillary data of the message.
247 The file descriptor is used to map that region.
249 There is no ``VHOST_NET_SET_BACKEND`` message as in vhost-cuse to signal
250 whether the virtio device is ready or stopped. Instead,
251 ``VHOST_SET_VRING_KICK`` is used as the signal to put the vhost device into
252 the data plane, and ``VHOST_GET_VRING_BASE`` is used as the signal to remove
253 the vhost device from the data plane.
255 When the socket connection is closed, vhost will destroy the device.
257 Vhost supported vSwitch reference
258 ---------------------------------
260 For more vhost details and how to support vhost in vSwitch, please refer to
261 the vhost example in the DPDK Sample Applications Guide.