1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2017 Intel Corporation
10 * Interface to vhost-user
15 #include <sys/eventfd.h>
17 #include <rte_memory.h>
18 #include <rte_mempool.h>
24 /* These are not C++-aware. */
25 #include <linux/vhost.h>
26 #include <linux/virtio_ring.h>
27 #include <linux/virtio_net.h>
29 #define RTE_VHOST_USER_CLIENT (1ULL << 0)
30 #define RTE_VHOST_USER_NO_RECONNECT (1ULL << 1)
31 #define RTE_VHOST_USER_RESERVED_1 (1ULL << 2)
32 #define RTE_VHOST_USER_IOMMU_SUPPORT (1ULL << 3)
33 #define RTE_VHOST_USER_POSTCOPY_SUPPORT (1ULL << 4)
34 /* support mbuf with external buffer attached */
35 #define RTE_VHOST_USER_EXTBUF_SUPPORT (1ULL << 5)
36 /* support only linear buffers (no chained mbufs) */
37 #define RTE_VHOST_USER_LINEARBUF_SUPPORT (1ULL << 6)
38 #define RTE_VHOST_USER_ASYNC_COPY (1ULL << 7)
39 #define RTE_VHOST_USER_NET_COMPLIANT_OL_FLAGS (1ULL << 8)
42 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
43 #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
46 #ifndef VIRTIO_NET_F_MQ
47 #define VIRTIO_NET_F_MQ 22
50 #ifndef VIRTIO_NET_F_MTU
51 #define VIRTIO_NET_F_MTU 3
54 #ifndef VIRTIO_F_ANY_LAYOUT
55 #define VIRTIO_F_ANY_LAYOUT 27
58 /** Protocol features. */
59 #ifndef VHOST_USER_PROTOCOL_F_MQ
60 #define VHOST_USER_PROTOCOL_F_MQ 0
63 #ifndef VHOST_USER_PROTOCOL_F_LOG_SHMFD
64 #define VHOST_USER_PROTOCOL_F_LOG_SHMFD 1
67 #ifndef VHOST_USER_PROTOCOL_F_RARP
68 #define VHOST_USER_PROTOCOL_F_RARP 2
71 #ifndef VHOST_USER_PROTOCOL_F_REPLY_ACK
72 #define VHOST_USER_PROTOCOL_F_REPLY_ACK 3
75 #ifndef VHOST_USER_PROTOCOL_F_NET_MTU
76 #define VHOST_USER_PROTOCOL_F_NET_MTU 4
79 #ifndef VHOST_USER_PROTOCOL_F_SLAVE_REQ
80 #define VHOST_USER_PROTOCOL_F_SLAVE_REQ 5
83 #ifndef VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
84 #define VHOST_USER_PROTOCOL_F_CRYPTO_SESSION 7
87 #ifndef VHOST_USER_PROTOCOL_F_PAGEFAULT
88 #define VHOST_USER_PROTOCOL_F_PAGEFAULT 8
91 #ifndef VHOST_USER_PROTOCOL_F_CONFIG
92 #define VHOST_USER_PROTOCOL_F_CONFIG 9
95 #ifndef VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD
96 #define VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD 10
99 #ifndef VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
100 #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER 11
103 #ifndef VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
104 #define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD 12
107 #ifndef VHOST_USER_PROTOCOL_F_STATUS
108 #define VHOST_USER_PROTOCOL_F_STATUS 16
111 /** Indicate whether protocol features negotiation is supported. */
112 #ifndef VHOST_USER_F_PROTOCOL_FEATURES
113 #define VHOST_USER_F_PROTOCOL_FEATURES 30
116 struct rte_vdpa_device;
119 * Information relating to memory regions including offsets to
120 * addresses in QEMUs memory file.
122 struct rte_vhost_mem_region {
123 uint64_t guest_phys_addr;
124 uint64_t guest_user_addr;
125 uint64_t host_user_addr;
133 * Memory structure includes region and mapping information.
135 struct rte_vhost_memory {
137 struct rte_vhost_mem_region regions[];
140 struct rte_vhost_inflight_desc_split {
147 struct rte_vhost_inflight_info_split {
151 uint16_t last_inflight_io;
153 struct rte_vhost_inflight_desc_split desc[0];
156 struct rte_vhost_inflight_desc_packed {
169 struct rte_vhost_inflight_info_packed {
174 uint16_t old_free_head;
176 uint16_t old_used_idx;
177 uint8_t used_wrap_counter;
178 uint8_t old_used_wrap_counter;
180 struct rte_vhost_inflight_desc_packed desc[0];
183 struct rte_vhost_resubmit_desc {
188 struct rte_vhost_resubmit_info {
189 struct rte_vhost_resubmit_desc *resubmit_list;
190 uint16_t resubmit_num;
193 struct rte_vhost_ring_inflight {
195 struct rte_vhost_inflight_info_split *inflight_split;
196 struct rte_vhost_inflight_info_packed *inflight_packed;
199 struct rte_vhost_resubmit_info *resubmit_inflight;
202 struct rte_vhost_vring {
204 struct vring_desc *desc;
205 struct vring_packed_desc *desc_packed;
208 struct vring_avail *avail;
209 struct vring_packed_desc_event *driver_event;
212 struct vring_used *used;
213 struct vring_packed_desc_event *device_event;
215 uint64_t log_guest_addr;
217 /** Deprecated, use rte_vhost_vring_call() instead. */
225 * Possible results of the vhost user message handling callbacks
227 enum rte_vhost_msg_result {
228 /* Message handling failed */
229 RTE_VHOST_MSG_RESULT_ERR = -1,
230 /* Message handling successful */
231 RTE_VHOST_MSG_RESULT_OK = 0,
232 /* Message handling successful and reply prepared */
233 RTE_VHOST_MSG_RESULT_REPLY = 1,
234 /* Message not handled */
235 RTE_VHOST_MSG_RESULT_NOT_HANDLED,
239 * Function prototype for the vhost backend to handle specific vhost user
247 * RTE_VHOST_MSG_RESULT_OK on success,
248 * RTE_VHOST_MSG_RESULT_REPLY on success with reply,
249 * RTE_VHOST_MSG_RESULT_ERR on failure,
250 * RTE_VHOST_MSG_RESULT_NOT_HANDLED if message was not handled.
252 typedef enum rte_vhost_msg_result (*rte_vhost_msg_handle)(int vid, void *msg);
255 * Optional vhost user message handlers.
257 struct rte_vhost_user_extern_ops {
258 /* Called prior to the master message handling. */
259 rte_vhost_msg_handle pre_msg_handle;
260 /* Called after the master message handling. */
261 rte_vhost_msg_handle post_msg_handle;
265 * Device and vring operations.
267 struct vhost_device_ops {
268 int (*new_device)(int vid); /**< Add device. */
269 void (*destroy_device)(int vid); /**< Remove device. */
271 int (*vring_state_changed)(int vid, uint16_t queue_id, int enable); /**< triggered when a vring is enabled or disabled */
274 * Features could be changed after the feature negotiation.
275 * For example, VHOST_F_LOG_ALL will be set/cleared at the
276 * start/end of live migration, respectively. This callback
277 * is used to inform the application on such change.
279 int (*features_changed)(int vid, uint64_t features);
281 int (*new_connection)(int vid);
282 void (*destroy_connection)(int vid);
285 * This callback gets called each time a guest gets notified
286 * about waiting packets. This is the interrupt handling through
287 * the eventfd_write(callfd), which can be used for counting these
290 void (*guest_notified)(int vid);
292 void *reserved[1]; /**< Reserved for future extension */
296 * Convert guest physical address to host virtual address
298 * This function is deprecated because unsafe.
299 * New rte_vhost_va_from_guest_pa() should be used instead to ensure
300 * guest physical ranges are fully and contiguously mapped into
301 * process virtual address space.
304 * the guest memory regions
306 * the guest physical address for querying
308 * the host virtual address on success, 0 on failure
311 static __rte_always_inline uint64_t
312 rte_vhost_gpa_to_vva(struct rte_vhost_memory *mem, uint64_t gpa)
314 struct rte_vhost_mem_region *reg;
317 for (i = 0; i < mem->nregions; i++) {
318 reg = &mem->regions[i];
319 if (gpa >= reg->guest_phys_addr &&
320 gpa < reg->guest_phys_addr + reg->size) {
321 return gpa - reg->guest_phys_addr +
330 * Convert guest physical address to host virtual address safely
332 * This variant of rte_vhost_gpa_to_vva() takes care all the
333 * requested length is mapped and contiguous in process address
337 * the guest memory regions
339 * the guest physical address for querying
341 * the size of the requested area to map, updated with actual size mapped
343 * the host virtual address on success, 0 on failure
345 static __rte_always_inline uint64_t
346 rte_vhost_va_from_guest_pa(struct rte_vhost_memory *mem,
347 uint64_t gpa, uint64_t *len)
349 struct rte_vhost_mem_region *r;
352 for (i = 0; i < mem->nregions; i++) {
353 r = &mem->regions[i];
354 if (gpa >= r->guest_phys_addr &&
355 gpa < r->guest_phys_addr + r->size) {
357 if (unlikely(*len > r->guest_phys_addr + r->size - gpa))
358 *len = r->guest_phys_addr + r->size - gpa;
360 return gpa - r->guest_phys_addr +
369 #define RTE_VHOST_NEED_LOG(features) ((features) & (1ULL << VHOST_F_LOG_ALL))
372 * Log the memory write start with given address.
374 * This function only need be invoked when the live migration starts.
375 * Therefore, we won't need call it at all in the most of time. For
376 * making the performance impact be minimum, it's suggested to do a
377 * check before calling it:
379 * if (unlikely(RTE_VHOST_NEED_LOG(features)))
380 * rte_vhost_log_write(vid, addr, len);
385 * the starting address for write (in guest physical address space)
387 * the length to write
389 void rte_vhost_log_write(int vid, uint64_t addr, uint64_t len);
392 * Log the used ring update start at given offset.
394 * Same as rte_vhost_log_write, it's suggested to do a check before
397 * if (unlikely(RTE_VHOST_NEED_LOG(features)))
398 * rte_vhost_log_used_vring(vid, vring_idx, offset, len);
405 * the offset inside the used ring
407 * the length to write
409 void rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
410 uint64_t offset, uint64_t len);
412 int rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable);
415 * Register vhost driver. path could be different for multiple
418 int rte_vhost_driver_register(const char *path, uint64_t flags);
420 /* Unregister vhost driver. This is only meaningful to vhost user. */
421 int rte_vhost_driver_unregister(const char *path);
424 * Set the vdpa device id, enforce single connection per socket
427 * The vhost-user socket file path
429 * vDPA device pointer
431 * 0 on success, -1 on failure
434 rte_vhost_driver_attach_vdpa_device(const char *path,
435 struct rte_vdpa_device *dev);
438 * Unset the vdpa device id
441 * The vhost-user socket file path
443 * 0 on success, -1 on failure
446 rte_vhost_driver_detach_vdpa_device(const char *path);
452 * The vhost-user socket file path
454 * vDPA device pointer, NULL on failure
456 struct rte_vdpa_device *
457 rte_vhost_driver_get_vdpa_device(const char *path);
460 * Set the feature bits the vhost-user driver supports.
463 * The vhost-user socket file path
467 * 0 on success, -1 on failure
469 int rte_vhost_driver_set_features(const char *path, uint64_t features);
472 * Enable vhost-user driver features.
475 * - the param features should be a subset of the feature bits provided
476 * by rte_vhost_driver_set_features().
477 * - it must be invoked before vhost-user negotiation starts.
480 * The vhost-user socket file path
484 * 0 on success, -1 on failure
486 int rte_vhost_driver_enable_features(const char *path, uint64_t features);
489 * Disable vhost-user driver features.
491 * The two notes at rte_vhost_driver_enable_features() also apply here.
494 * The vhost-user socket file path
496 * Features to disable
498 * 0 on success, -1 on failure
500 int rte_vhost_driver_disable_features(const char *path, uint64_t features);
503 * Get the feature bits before feature negotiation.
506 * The vhost-user socket file path
508 * A pointer to store the queried feature bits
510 * 0 on success, -1 on failure
512 int rte_vhost_driver_get_features(const char *path, uint64_t *features);
515 * Set the protocol feature bits before feature negotiation.
518 * The vhost-user socket file path
519 * @param protocol_features
520 * Supported protocol features
522 * 0 on success, -1 on failure
525 rte_vhost_driver_set_protocol_features(const char *path,
526 uint64_t protocol_features);
529 * Get the protocol feature bits before feature negotiation.
532 * The vhost-user socket file path
533 * @param protocol_features
534 * A pointer to store the queried protocol feature bits
536 * 0 on success, -1 on failure
539 rte_vhost_driver_get_protocol_features(const char *path,
540 uint64_t *protocol_features);
543 * Get the queue number bits before feature negotiation.
546 * The vhost-user socket file path
548 * A pointer to store the queried queue number bits
550 * 0 on success, -1 on failure
553 rte_vhost_driver_get_queue_num(const char *path, uint32_t *queue_num);
556 * Get the feature bits after negotiation
561 * A pointer to store the queried feature bits
563 * 0 on success, -1 on failure
565 int rte_vhost_get_negotiated_features(int vid, uint64_t *features);
568 * Get the protocol feature bits after negotiation
572 * @param protocol_features
573 * A pointer to store the queried protocol feature bits
575 * 0 on success, -1 on failure
579 rte_vhost_get_negotiated_protocol_features(int vid,
580 uint64_t *protocol_features);
582 /* Register callbacks. */
583 int rte_vhost_driver_callback_register(const char *path,
584 struct vhost_device_ops const * const ops);
588 * Start the vhost-user driver.
590 * This function triggers the vhost-user negotiation.
593 * The vhost-user socket file path
595 * 0 on success, -1 on failure
597 int rte_vhost_driver_start(const char *path);
600 * Get the MTU value of the device if set in QEMU.
603 * virtio-net device ID
605 * The variable to store the MTU value
609 * -EAGAIN: device not yet started
610 * -ENOTSUP: device does not support MTU feature
612 int rte_vhost_get_mtu(int vid, uint16_t *mtu);
615 * Get the numa node from which the virtio net device's memory
622 * The numa node, -1 on failure
624 int rte_vhost_get_numa_node(int vid);
628 * Get the number of queues the device supports.
630 * Note this function is deprecated, as it returns a queue pair number,
631 * which is vhost specific. Instead, rte_vhost_get_vring_num should
638 * The number of queues, 0 on failure
641 uint32_t rte_vhost_get_queue_num(int vid);
644 * Get the number of vrings the device supports.
650 * The number of vrings, 0 on failure
652 uint16_t rte_vhost_get_vring_num(int vid);
655 * Get the virtio net device's ifname, which is the vhost-user socket
661 * The buffer to stored the queried ifname
666 * 0 on success, -1 on failure
668 int rte_vhost_get_ifname(int vid, char *buf, size_t len);
671 * Get how many avail entries are left in the queue
679 * num of avail entries left
681 uint16_t rte_vhost_avail_entries(int vid, uint16_t queue_id);
686 * This function adds buffers to the virtio devices RX virtqueue. Buffers can
687 * be received from the physical port or from another virtual device. A packet
688 * count is returned to indicate the number of packets that were successfully
689 * added to the RX queue.
693 * virtio queue index in mq case
695 * array to contain packets to be enqueued
697 * packets num to be enqueued
699 * num of packets enqueued
701 uint16_t rte_vhost_enqueue_burst(int vid, uint16_t queue_id,
702 struct rte_mbuf **pkts, uint16_t count);
705 * This function gets guest buffers from the virtio device TX virtqueue,
706 * construct host mbufs, copies guest buffer content to host mbufs and
707 * store them in pkts to be processed.
711 * virtio queue index in mq case
713 * mbuf_pool where host mbuf is allocated.
715 * array to contain packets to be dequeued
717 * packets num to be dequeued
719 * num of packets dequeued
721 uint16_t rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
722 struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count);
725 * Get guest mem table: a list of memory regions.
727 * An rte_vhost_vhost_memory object will be allocated internally, to hold the
728 * guest memory regions. Application should free it at destroy_device()
734 * To store the returned mem regions
736 * 0 on success, -1 on failure
738 int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
741 * Get guest vring info, including the vring address, vring size, etc.
748 * the structure to hold the requested vring info
750 * 0 on success, -1 on failure
752 int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
753 struct rte_vhost_vring *vring);
756 * Get guest inflight vring info, including inflight ring and resubmit list.
763 * the structure to hold the requested inflight vring info
765 * 0 on success, -1 on failure
768 rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
769 struct rte_vhost_ring_inflight *vring);
772 * Set split inflight descriptor.
774 * This function save descriptors that has been comsumed in available
782 * inflight entry index
784 * 0 on success, -1 on failure
787 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
791 * Set packed inflight descriptor and get corresponding inflight entry
793 * This function save descriptors that has been comsumed
800 * head of descriptors
802 * last of descriptors
803 * @param inflight_entry
804 * corresponding inflight entry
806 * 0 on success, -1 on failure
809 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
810 uint16_t head, uint16_t last, uint16_t *inflight_entry);
813 * Save the head of list that the last batch of used descriptors.
820 * descriptor entry index
822 * 0 on success, -1 on failure
825 rte_vhost_set_last_inflight_io_split(int vid,
826 uint16_t vring_idx, uint16_t idx);
829 * Update the inflight free_head, used_idx and used_wrap_counter.
831 * This function will update status first before updating descriptors
839 * head of descriptors
841 * 0 on success, -1 on failure
844 rte_vhost_set_last_inflight_io_packed(int vid,
845 uint16_t vring_idx, uint16_t head);
848 * Clear the split inflight status.
854 * @param last_used_idx
855 * last used idx of used ring
857 * inflight entry index
859 * 0 on success, -1 on failure
862 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
863 uint16_t last_used_idx, uint16_t idx);
866 * Clear the packed inflight status.
873 * inflight entry index
875 * 0 on success, -1 on failure
878 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
882 * Notify the guest that used descriptors have been added to the vring. This
883 * function acts as a memory barrier.
890 * 0 on success, -1 on failure
892 int rte_vhost_vring_call(int vid, uint16_t vring_idx);
895 * Get vhost RX queue avail count.
900 * virtio queue index in mq case
902 * num of desc available
904 uint32_t rte_vhost_rx_queue_count(int vid, uint16_t qid);
907 * Get log base and log size of the vhost device
916 * 0 on success, -1 on failure
919 rte_vhost_get_log_base(int vid, uint64_t *log_base, uint64_t *log_size);
922 * Get last_avail/used_idx of the vhost virtqueue
928 * @param last_avail_idx
929 * vhost last_avail_idx to get
930 * @param last_used_idx
931 * vhost last_used_idx to get
933 * 0 on success, -1 on failure
936 rte_vhost_get_vring_base(int vid, uint16_t queue_id,
937 uint16_t *last_avail_idx, uint16_t *last_used_idx);
940 * Get last_avail/last_used of the vhost virtqueue
942 * This function is designed for the reconnection and it's specific for
943 * the packed ring as we can get the two parameters from the inflight
950 * @param last_avail_idx
951 * vhost last_avail_idx to get
952 * @param last_used_idx
953 * vhost last_used_idx to get
955 * 0 on success, -1 on failure
958 rte_vhost_get_vring_base_from_inflight(int vid,
959 uint16_t queue_id, uint16_t *last_avail_idx, uint16_t *last_used_idx);
962 * Set last_avail/used_idx of the vhost virtqueue
968 * @param last_avail_idx
969 * last_avail_idx to set
970 * @param last_used_idx
971 * last_used_idx to set
973 * 0 on success, -1 on failure
976 rte_vhost_set_vring_base(int vid, uint16_t queue_id,
977 uint16_t last_avail_idx, uint16_t last_used_idx);
980 * Register external message handling callbacks
985 * virtio external callbacks to register
987 * additional context passed to the callbacks
989 * 0 on success, -1 on failure
992 rte_vhost_extern_callback_register(int vid,
993 struct rte_vhost_user_extern_ops const * const ops, void *ctx);
996 * Get vdpa device id for vhost device.
1001 * vDPA device pointer on success, NULL on failure
1003 struct rte_vdpa_device *
1004 rte_vhost_get_vdpa_device(int vid);
1007 * Notify the guest that should get virtio configuration space from backend.
1012 * wait for the master response the status of this operation
1014 * 0 on success, < 0 on failure
1018 rte_vhost_slave_config_change(int vid, bool need_reply);
1024 #endif /* _RTE_VHOST_H_ */