1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2010-2016 Intel Corporation.
7 The vhost library implements a user space virtio net server allowing the user
8 to manipulate the virtio ring directly. In another words, it allows the user
9 to fetch/put packets from/to the VM virtio net device. To achieve this, a
10 vhost library should be able to:
12 * Access the guest memory:
14 For QEMU, this is done by using the ``-object memory-backend-file,share=on,...``
15 option. Which means QEMU will create a file to serve as the guest RAM.
16 The ``share=on`` option allows another process to map that file, which
17 means it can access the guest RAM.
19 * Know all the necessary information about the vring:
21 Information such as where the available ring is stored. Vhost defines some
22 messages (passed through a Unix domain socket file) to tell the backend all
23 the information it needs to know how to manipulate the vring.
29 The following is an overview of some key Vhost API functions:
31 * ``rte_vhost_driver_register(path, flags)``
33 This function registers a vhost driver into the system. ``path`` specifies
34 the Unix domain socket file path.
36 Currently supported flags are:
38 - ``RTE_VHOST_USER_CLIENT``
40 DPDK vhost-user will act as the client when this flag is given. See below
43 - ``RTE_VHOST_USER_NO_RECONNECT``
45 When DPDK vhost-user acts as the client it will keep trying to reconnect
46 to the server (QEMU) until it succeeds. This is useful in two cases:
48 * When QEMU is not started yet.
49 * When QEMU restarts (for example due to a guest OS reboot).
51 This reconnect option is enabled by default. However, it can be turned off
54 - ``RTE_VHOST_USER_DEQUEUE_ZERO_COPY``
56 Dequeue zero copy will be enabled when this flag is set. It is disabled by
59 There are some truths (including limitations) you might want to know while
62 * zero copy is not good for small packets (typically for packet size below
65 * zero copy is really good for VM2VM case. For iperf between two VMs, the
66 boost could be above 70% (when TSO is enabled).
68 * For zero copy in VM2NIC case, guest Tx used vring may be starved if the
69 PMD driver consume the mbuf but not release them timely.
71 For example, i40e driver has an optimization to maximum NIC pipeline which
72 postpones returning transmitted mbuf until only tx_free_threshold free
73 descs left. The virtio TX used ring will be starved if the formula
74 (num_i40e_tx_desc - num_virtio_tx_desc > tx_free_threshold) is true, since
75 i40e will not return back mbuf.
77 A performance tip for tuning zero copy in VM2NIC case is to adjust the
78 frequency of mbuf free (i.e. adjust tx_free_threshold of i40e driver) to
79 balance consumer and producer.
81 * Guest memory should be backended with huge pages to achieve better
82 performance. Using 1G page size is the best.
84 When dequeue zero copy is enabled, the guest phys address and host phys
85 address mapping has to be established. Using non-huge pages means far
86 more page segments. To make it simple, DPDK vhost does a linear search
87 of those segments, thus the fewer the segments, the quicker we will get
88 the mapping. NOTE: we may speed it by using tree searching in future.
90 * zero copy can not work when using vfio-pci with iommu mode currently, this
91 is because we don't setup iommu dma mapping for guest memory. If you have
92 to use vfio-pci driver, please insert vfio-pci kernel module in noiommu
95 * The consumer of zero copy mbufs should consume these mbufs as soon as
96 possible, otherwise it may block the operations in vhost.
98 - ``RTE_VHOST_USER_IOMMU_SUPPORT``
100 IOMMU support will be enabled when this flag is set. It is disabled by
103 Enabling this flag makes possible to use guest vIOMMU to protect vhost
104 from accessing memory the virtio device isn't allowed to, when the feature
105 is negotiated and an IOMMU device is declared.
107 - ``RTE_VHOST_USER_POSTCOPY_SUPPORT``
109 Postcopy live-migration support will be enabled when this flag is set.
110 It is disabled by default.
112 Enabling this flag should only be done when the calling application does
113 not pre-fault the guest shared memory, otherwise migration would fail.
115 - ``RTE_VHOST_USER_LINEARBUF_SUPPORT``
117 Enabling this flag forces vhost dequeue function to only provide linear
118 pktmbuf (no multi-segmented pktmbuf).
120 The vhost library by default provides a single pktmbuf for given a
121 packet, but if for some reason the data doesn't fit into a single
122 pktmbuf (e.g., TSO is enabled), the library will allocate additional
123 pktmbufs from the same mempool and chain them together to create a
124 multi-segmented pktmbuf.
126 However, the vhost application needs to support multi-segmented format.
127 If the vhost application does not support that format and requires large
128 buffers to be dequeue, this flag should be enabled to force only linear
129 buffers (see RTE_VHOST_USER_EXTBUF_SUPPORT) or drop the packet.
131 It is disabled by default.
133 - ``RTE_VHOST_USER_EXTBUF_SUPPORT``
135 Enabling this flag allows vhost dequeue function to allocate and attach
136 an external buffer to a pktmbuf if the pkmbuf doesn't provide enough
137 space to store all data.
139 This is useful when the vhost application wants to support large packets
140 but doesn't want to increase the default mempool object size nor to
141 support multi-segmented mbufs (non-linear). In this case, a fresh buffer
142 is allocated using rte_malloc() which gets attached to a pktmbuf using
143 rte_pktmbuf_attach_extbuf().
145 See RTE_VHOST_USER_LINEARBUF_SUPPORT as well to disable multi-segmented
146 mbufs for application that doesn't support chained mbufs.
148 It is disabled by default.
150 - ``RTE_VHOST_USER_ASYNC_COPY``
152 Asynchronous data path will be enabled when this flag is set. Async data
153 path allows applications to register async copy devices (typically
154 hardware DMA channels) to the vhost queues. Vhost leverages the copy
155 device registered to free CPU from memory copy operations. A set of
156 async data path APIs are defined for DPDK applications to make use of
157 the async capability. Only packets enqueued/dequeued by async APIs are
158 processed through the async data path.
160 Currently this feature is only implemented on split ring enqueue data
163 It is disabled by default.
165 * ``rte_vhost_driver_set_features(path, features)``
167 This function sets the feature bits the vhost-user driver supports. The
168 vhost-user driver could be vhost-user net, yet it could be something else,
169 say, vhost-user SCSI.
171 * ``rte_vhost_driver_callback_register(path, vhost_device_ops)``
173 This function registers a set of callbacks, to let DPDK applications take
174 the appropriate action when some events happen. The following events are
177 * ``new_device(int vid)``
179 This callback is invoked when a virtio device becomes ready. ``vid``
180 is the vhost device ID.
182 * ``destroy_device(int vid)``
184 This callback is invoked when a virtio device is paused or shut down.
186 * ``vring_state_changed(int vid, uint16_t queue_id, int enable)``
188 This callback is invoked when a specific queue's state is changed, for
189 example to enabled or disabled.
191 * ``features_changed(int vid, uint64_t features)``
193 This callback is invoked when the features is changed. For example,
194 ``VHOST_F_LOG_ALL`` will be set/cleared at the start/end of live
195 migration, respectively.
197 * ``new_connection(int vid)``
199 This callback is invoked on new vhost-user socket connection. If DPDK
200 acts as the server the device should not be deleted before
201 ``destroy_connection`` callback is received.
203 * ``destroy_connection(int vid)``
205 This callback is invoked when vhost-user socket connection is closed.
206 It indicates that device with id ``vid`` is no longer in use and can be
209 * ``rte_vhost_driver_disable/enable_features(path, features))``
211 This function disables/enables some features. For example, it can be used to
212 disable mergeable buffers and TSO features, which both are enabled by
215 * ``rte_vhost_driver_start(path)``
217 This function triggers the vhost-user negotiation. It should be invoked at
218 the end of initializing a vhost-user driver.
220 * ``rte_vhost_enqueue_burst(vid, queue_id, pkts, count)``
222 Transmits (enqueues) ``count`` packets from host to guest.
224 * ``rte_vhost_dequeue_burst(vid, queue_id, mbuf_pool, pkts, count)``
226 Receives (dequeues) ``count`` packets from guest, and stored them at ``pkts``.
228 * ``rte_vhost_crypto_create(vid, cryptodev_id, sess_mempool, socket_id)``
230 As an extension of new_device(), this function adds virtio-crypto workload
231 acceleration capability to the device. All crypto workload is processed by
232 DPDK cryptodev with the device ID of ``cryptodev_id``.
234 * ``rte_vhost_crypto_free(vid)``
236 Frees the memory and vhost-user message handlers created in
237 rte_vhost_crypto_create().
239 * ``rte_vhost_crypto_fetch_requests(vid, queue_id, ops, nb_ops)``
241 Receives (dequeues) ``nb_ops`` virtio-crypto requests from guest, parses
242 them to DPDK Crypto Operations, and fills the ``ops`` with parsing results.
244 * ``rte_vhost_crypto_finalize_requests(queue_id, ops, nb_ops)``
246 After the ``ops`` are dequeued from Cryptodev, finalizes the jobs and
247 notifies the guest(s).
249 * ``rte_vhost_crypto_set_zero_copy(vid, option)``
251 Enable or disable zero copy feature of the vhost crypto backend.
253 * ``rte_vhost_async_channel_register(vid, queue_id, features, ops)``
255 Register a vhost queue with async copy device channel.
256 Following device ``features`` must be specified together with the
261 Async copy device can guarantee the ordering of copy completion
262 sequence. Copies are completed in the same order with that at
265 Currently, only ``async_inorder`` capable device is supported by vhost.
267 * ``async_threshold``
269 The copy length (in bytes) below which CPU copy will be used even if
270 applications call async vhost APIs to enqueue/dequeue data.
272 Typical value is 512~1024 depending on the async device capability.
274 Applications must provide following ``ops`` callbacks for vhost lib to
275 work with the async copy devices:
277 * ``transfer_data(vid, queue_id, descs, opaque_data, count)``
279 vhost invokes this function to submit copy data to the async devices.
280 For non-async_inorder capable devices, ``opaque_data`` could be used
281 for identifying the completed packets.
283 * ``check_completed_copies(vid, queue_id, opaque_data, max_packets)``
285 vhost invokes this function to get the copy data completed by async
288 * ``rte_vhost_async_channel_unregister(vid, queue_id)``
290 Unregister the async copy device channel from a vhost queue.
292 * ``rte_vhost_submit_enqueue_burst(vid, queue_id, pkts, count)``
294 Submit an enqueue request to transmit ``count`` packets from host to guest
295 by async data path. Enqueue is not guaranteed to finish upon the return of
298 Applications must not free the packets submitted for enqueue until the
299 packets are completed.
301 * ``rte_vhost_poll_enqueue_completed(vid, queue_id, pkts, count)``
303 Poll enqueue completion status from async data path. Completed packets
304 are returned to applications through ``pkts``.
306 Vhost-user Implementations
307 --------------------------
309 Vhost-user uses Unix domain sockets for passing messages. This means the DPDK
310 vhost-user implementation has two options:
312 * DPDK vhost-user acts as the server.
314 DPDK will create a Unix domain socket server file and listen for
315 connections from the frontend.
317 Note, this is the default mode, and the only mode before DPDK v16.07.
320 * DPDK vhost-user acts as the client.
322 Unlike the server mode, this mode doesn't create the socket file;
323 it just tries to connect to the server (which responses to create the
326 When the DPDK vhost-user application restarts, DPDK vhost-user will try to
327 connect to the server again. This is how the "reconnect" feature works.
330 * The "reconnect" feature requires **QEMU v2.7** (or above).
332 * The vhost supported features must be exactly the same before and
333 after the restart. For example, if TSO is disabled and then enabled,
334 nothing will work and issues undefined might happen.
336 No matter which mode is used, once a connection is established, DPDK
337 vhost-user will start receiving and processing vhost messages from QEMU.
339 For messages with a file descriptor, the file descriptor can be used directly
340 in the vhost process as it is already installed by the Unix domain socket.
342 The supported vhost messages are:
344 * ``VHOST_SET_MEM_TABLE``
345 * ``VHOST_SET_VRING_KICK``
346 * ``VHOST_SET_VRING_CALL``
347 * ``VHOST_SET_LOG_FD``
348 * ``VHOST_SET_VRING_ERR``
350 For ``VHOST_SET_MEM_TABLE`` message, QEMU will send information for each
351 memory region and its file descriptor in the ancillary data of the message.
352 The file descriptor is used to map that region.
354 ``VHOST_SET_VRING_KICK`` is used as the signal to put the vhost device into
355 the data plane, and ``VHOST_GET_VRING_BASE`` is used as the signal to remove
356 the vhost device from the data plane.
358 When the socket connection is closed, vhost will destroy the device.
360 Guest memory requirement
361 ------------------------
363 * Memory pre-allocation
365 For non-zerocopy non-async data path, guest memory pre-allocation is not a
366 must. This can help save of memory. If users really want the guest memory
367 to be pre-allocated (e.g., for performance reason), we can add option
368 ``-mem-prealloc`` when starting QEMU. Or, we can lock all memory at vhost
369 side which will force memory to be allocated when mmap at vhost side;
370 option --mlockall in ovs-dpdk is an example in hand.
372 For async and zerocopy data path, we force the VM memory to be
373 pre-allocated at vhost lib when mapping the guest memory; and also we need
374 to lock the memory to prevent pages being swapped out to disk.
378 Make sure ``share=on`` QEMU option is given. vhost-user will not work with
379 a QEMU version without shared memory mapping.
381 Vhost supported vSwitch reference
382 ---------------------------------
384 For more vhost details and how to support vhost in vSwitch, please refer to
385 the vhost example in the DPDK Sample Applications Guide.
387 Vhost data path acceleration (vDPA)
388 -----------------------------------
390 vDPA supports selective datapath in vhost-user lib by enabling virtio ring
391 compatible devices to serve virtio driver directly for datapath acceleration.
393 ``rte_vhost_driver_attach_vdpa_device`` is used to configure the vhost device
394 with accelerated backend.
396 Also vhost device capabilities are made configurable to adopt various devices.
397 Such capabilities include supported features, protocol features, queue number.
399 Finally, a set of device ops is defined for device specific operations:
403 Called to get supported queue number of the device.
407 Called to get supported features of the device.
409 * ``get_protocol_features``
411 Called to get supported protocol features of the device.
415 Called to configure the actual device when the virtio device becomes ready.
419 Called to close the actual device when the virtio device is stopped.
421 * ``set_vring_state``
423 Called to change the state of the vring in the actual device when vring state
428 Called to set the negotiated features to device.
432 Called to allow the device to response to RARP sending.
434 * ``get_vfio_group_fd``
436 Called to get the VFIO group fd of the device.
438 * ``get_vfio_device_fd``
440 Called to get the VFIO device fd of the device.
442 * ``get_notify_area``
444 Called to get the notify area info of the queue.