xref: /dpdk/lib/vhost/rte_vhost.h (revision 53d3f477)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 
5 #ifndef _RTE_VHOST_H_
6 #define _RTE_VHOST_H_
7 
8 /**
9  * @file
10  * Interface to vhost-user
11  */
12 
13 #include <stdbool.h>
14 #include <stdint.h>
15 #include <sys/eventfd.h>
16 
17 #include <rte_memory.h>
18 #include <rte_mempool.h>
19 
20 #ifdef __cplusplus
21 extern "C" {
22 #endif
23 
24 #ifndef __cplusplus
25 /* These are not C++-aware. */
26 #include <linux/vhost.h>
27 #include <linux/virtio_ring.h>
28 #include <linux/virtio_net.h>
29 #endif
30 
31 #define RTE_VHOST_USER_CLIENT		(1ULL << 0)
32 #define RTE_VHOST_USER_NO_RECONNECT	(1ULL << 1)
33 #define RTE_VHOST_USER_RESERVED_1	(1ULL << 2)
34 #define RTE_VHOST_USER_IOMMU_SUPPORT	(1ULL << 3)
35 #define RTE_VHOST_USER_POSTCOPY_SUPPORT		(1ULL << 4)
36 /* support mbuf with external buffer attached */
37 #define RTE_VHOST_USER_EXTBUF_SUPPORT	(1ULL << 5)
38 /* support only linear buffers (no chained mbufs) */
39 #define RTE_VHOST_USER_LINEARBUF_SUPPORT	(1ULL << 6)
40 #define RTE_VHOST_USER_ASYNC_COPY	(1ULL << 7)
41 #define RTE_VHOST_USER_NET_COMPLIANT_OL_FLAGS	(1ULL << 8)
42 
43 /* Features. */
44 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
45  #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
46 #endif
47 
48 #ifndef VIRTIO_NET_F_MQ
49  #define VIRTIO_NET_F_MQ		22
50 #endif
51 
52 #ifndef VIRTIO_NET_F_MTU
53  #define VIRTIO_NET_F_MTU 3
54 #endif
55 
56 #ifndef VIRTIO_F_ANY_LAYOUT
57  #define VIRTIO_F_ANY_LAYOUT		27
58 #endif
59 
60 /** Protocol features. */
61 #ifndef VHOST_USER_PROTOCOL_F_MQ
62 #define VHOST_USER_PROTOCOL_F_MQ	0
63 #endif
64 
65 #ifndef VHOST_USER_PROTOCOL_F_LOG_SHMFD
66 #define VHOST_USER_PROTOCOL_F_LOG_SHMFD	1
67 #endif
68 
69 #ifndef VHOST_USER_PROTOCOL_F_RARP
70 #define VHOST_USER_PROTOCOL_F_RARP	2
71 #endif
72 
73 #ifndef VHOST_USER_PROTOCOL_F_REPLY_ACK
74 #define VHOST_USER_PROTOCOL_F_REPLY_ACK	3
75 #endif
76 
77 #ifndef VHOST_USER_PROTOCOL_F_NET_MTU
78 #define VHOST_USER_PROTOCOL_F_NET_MTU	4
79 #endif
80 
81 #ifndef VHOST_USER_PROTOCOL_F_SLAVE_REQ
82 #define VHOST_USER_PROTOCOL_F_SLAVE_REQ	5
83 #endif
84 
85 #ifndef VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
86 #define VHOST_USER_PROTOCOL_F_CRYPTO_SESSION 7
87 #endif
88 
89 #ifndef VHOST_USER_PROTOCOL_F_PAGEFAULT
90 #define VHOST_USER_PROTOCOL_F_PAGEFAULT 8
91 #endif
92 
93 #ifndef VHOST_USER_PROTOCOL_F_CONFIG
94 #define VHOST_USER_PROTOCOL_F_CONFIG 9
95 #endif
96 
97 #ifndef VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD
98 #define VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD 10
99 #endif
100 
101 #ifndef VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
102 #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER 11
103 #endif
104 
105 #ifndef VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
106 #define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD 12
107 #endif
108 
109 #ifndef VHOST_USER_PROTOCOL_F_STATUS
110 #define VHOST_USER_PROTOCOL_F_STATUS 16
111 #endif
112 
113 /** Indicate whether protocol features negotiation is supported. */
114 #ifndef VHOST_USER_F_PROTOCOL_FEATURES
115 #define VHOST_USER_F_PROTOCOL_FEATURES	30
116 #endif
117 
118 #define RTE_MAX_VHOST_DEVICE	1024
119 
120 struct rte_vdpa_device;
121 
122 /**
123  * Information relating to memory regions including offsets to
124  * addresses in QEMUs memory file.
125  */
126 struct rte_vhost_mem_region {
127 	uint64_t guest_phys_addr;
128 	uint64_t guest_user_addr;
129 	uint64_t host_user_addr;
130 	uint64_t size;
131 	void	 *mmap_addr;
132 	uint64_t mmap_size;
133 	int fd;
134 };
135 
136 /**
137  * Memory structure includes region and mapping information.
138  */
139 struct rte_vhost_memory {
140 	uint32_t nregions;
141 	struct rte_vhost_mem_region regions[];
142 };
143 
144 struct rte_vhost_inflight_desc_split {
145 	uint8_t inflight;
146 	uint8_t padding[5];
147 	uint16_t next;
148 	uint64_t counter;
149 };
150 
151 struct rte_vhost_inflight_info_split {
152 	uint64_t features;
153 	uint16_t version;
154 	uint16_t desc_num;
155 	uint16_t last_inflight_io;
156 	uint16_t used_idx;
157 	struct rte_vhost_inflight_desc_split desc[0];
158 };
159 
160 struct rte_vhost_inflight_desc_packed {
161 	uint8_t inflight;
162 	uint8_t padding;
163 	uint16_t next;
164 	uint16_t last;
165 	uint16_t num;
166 	uint64_t counter;
167 	uint16_t id;
168 	uint16_t flags;
169 	uint32_t len;
170 	uint64_t addr;
171 };
172 
173 struct rte_vhost_inflight_info_packed {
174 	uint64_t features;
175 	uint16_t version;
176 	uint16_t desc_num;
177 	uint16_t free_head;
178 	uint16_t old_free_head;
179 	uint16_t used_idx;
180 	uint16_t old_used_idx;
181 	uint8_t used_wrap_counter;
182 	uint8_t old_used_wrap_counter;
183 	uint8_t padding[7];
184 	struct rte_vhost_inflight_desc_packed desc[0];
185 };
186 
187 struct rte_vhost_resubmit_desc {
188 	uint16_t index;
189 	uint64_t counter;
190 };
191 
192 struct rte_vhost_resubmit_info {
193 	struct rte_vhost_resubmit_desc *resubmit_list;
194 	uint16_t resubmit_num;
195 };
196 
197 struct rte_vhost_ring_inflight {
198 	union {
199 		struct rte_vhost_inflight_info_split *inflight_split;
200 		struct rte_vhost_inflight_info_packed *inflight_packed;
201 	};
202 
203 	struct rte_vhost_resubmit_info *resubmit_inflight;
204 };
205 
206 struct rte_vhost_vring {
207 	union {
208 		struct vring_desc *desc;
209 		struct vring_packed_desc *desc_packed;
210 	};
211 	union {
212 		struct vring_avail *avail;
213 		struct vring_packed_desc_event *driver_event;
214 	};
215 	union {
216 		struct vring_used *used;
217 		struct vring_packed_desc_event *device_event;
218 	};
219 	uint64_t		log_guest_addr;
220 
221 	/** Deprecated, use rte_vhost_vring_call() instead. */
222 	int			callfd;
223 
224 	int			kickfd;
225 	uint16_t		size;
226 };
227 
228 /**
229  * Possible results of the vhost user message handling callbacks
230  */
231 enum rte_vhost_msg_result {
232 	/* Message handling failed */
233 	RTE_VHOST_MSG_RESULT_ERR = -1,
234 	/* Message handling successful */
235 	RTE_VHOST_MSG_RESULT_OK =  0,
236 	/* Message handling successful and reply prepared */
237 	RTE_VHOST_MSG_RESULT_REPLY =  1,
238 	/* Message not handled */
239 	RTE_VHOST_MSG_RESULT_NOT_HANDLED,
240 };
241 
242 /**
243  * Function prototype for the vhost backend to handle specific vhost user
244  * messages.
245  *
246  * @param vid
247  *  vhost device id
248  * @param msg
249  *  Message pointer.
250  * @return
251  *  RTE_VHOST_MSG_RESULT_OK on success,
252  *  RTE_VHOST_MSG_RESULT_REPLY on success with reply,
253  *  RTE_VHOST_MSG_RESULT_ERR on failure,
254  *  RTE_VHOST_MSG_RESULT_NOT_HANDLED if message was not handled.
255  */
256 typedef enum rte_vhost_msg_result (*rte_vhost_msg_handle)(int vid, void *msg);
257 
258 /**
259  * Optional vhost user message handlers.
260  */
261 struct rte_vhost_user_extern_ops {
262 	/* Called prior to the master message handling. */
263 	rte_vhost_msg_handle pre_msg_handle;
264 	/* Called after the master message handling. */
265 	rte_vhost_msg_handle post_msg_handle;
266 };
267 
268 /**
269  * Device and vring operations.
270  */
271 struct rte_vhost_device_ops {
272 	int (*new_device)(int vid);		/**< Add device. */
273 	void (*destroy_device)(int vid);	/**< Remove device. */
274 
275 	int (*vring_state_changed)(int vid, uint16_t queue_id, int enable);	/**< triggered when a vring is enabled or disabled */
276 
277 	/**
278 	 * Features could be changed after the feature negotiation.
279 	 * For example, VHOST_F_LOG_ALL will be set/cleared at the
280 	 * start/end of live migration, respectively. This callback
281 	 * is used to inform the application on such change.
282 	 */
283 	int (*features_changed)(int vid, uint64_t features);
284 
285 	int (*new_connection)(int vid);
286 	void (*destroy_connection)(int vid);
287 
288 	/**
289 	 * This callback gets called each time a guest gets notified
290 	 * about waiting packets. This is the interrupt handling through
291 	 * the eventfd_write(callfd), which can be used for counting these
292 	 * "slow" syscalls.
293 	 */
294 	void (*guest_notified)(int vid);
295 
296 	void *reserved[1]; /**< Reserved for future extension */
297 };
298 
299 /**
300  * Power monitor condition.
301  */
302 struct rte_vhost_power_monitor_cond {
303 	/**< Address to monitor for changes */
304 	volatile void *addr;
305 	/**< If the `mask` is non-zero, location pointed
306 	 *   to by `addr` will be read and masked, then
307 	 *   compared with this value.
308 	 */
309 	uint64_t val;
310 	/**< 64-bit mask to extract value read from `addr` */
311 	uint64_t mask;
312 	/**< Data size (in bytes) that will be read from the
313 	 *   monitored memory location (`addr`).
314 	 */
315 	uint8_t size;
316 	/**< If 1, and masked value that read from 'addr' equals
317 	 *   'val', the driver should skip core sleep. If 0, and
318 	 *  masked value that read from 'addr' does not equal 'val',
319 	 *  the driver should skip core sleep.
320 	 */
321 	uint8_t match;
322 };
323 
324 /**
325  * Convert guest physical address to host virtual address
326  *
327  * This function is deprecated because unsafe.
328  * New rte_vhost_va_from_guest_pa() should be used instead to ensure
329  * guest physical ranges are fully and contiguously mapped into
330  * process virtual address space.
331  *
332  * @param mem
333  *  the guest memory regions
334  * @param gpa
335  *  the guest physical address for querying
336  * @return
337  *  the host virtual address on success, 0 on failure
338  */
339 __rte_deprecated
340 static __rte_always_inline uint64_t
rte_vhost_gpa_to_vva(struct rte_vhost_memory * mem,uint64_t gpa)341 rte_vhost_gpa_to_vva(struct rte_vhost_memory *mem, uint64_t gpa)
342 {
343 	struct rte_vhost_mem_region *reg;
344 	uint32_t i;
345 
346 	for (i = 0; i < mem->nregions; i++) {
347 		reg = &mem->regions[i];
348 		if (gpa >= reg->guest_phys_addr &&
349 		    gpa <  reg->guest_phys_addr + reg->size) {
350 			return gpa - reg->guest_phys_addr +
351 			       reg->host_user_addr;
352 		}
353 	}
354 
355 	return 0;
356 }
357 
358 /**
359  * Convert guest physical address to host virtual address safely
360  *
361  * This variant of rte_vhost_gpa_to_vva() takes care all the
362  * requested length is mapped and contiguous in process address
363  * space.
364  *
365  * @param mem
366  *  the guest memory regions
367  * @param gpa
368  *  the guest physical address for querying
369  * @param len
370  *  the size of the requested area to map, updated with actual size mapped
371  * @return
372  *  the host virtual address on success, 0 on failure
373  */
374 static __rte_always_inline uint64_t
rte_vhost_va_from_guest_pa(struct rte_vhost_memory * mem,uint64_t gpa,uint64_t * len)375 rte_vhost_va_from_guest_pa(struct rte_vhost_memory *mem,
376 						   uint64_t gpa, uint64_t *len)
377 {
378 	struct rte_vhost_mem_region *r;
379 	uint32_t i;
380 
381 	for (i = 0; i < mem->nregions; i++) {
382 		r = &mem->regions[i];
383 		if (gpa >= r->guest_phys_addr &&
384 		    gpa <  r->guest_phys_addr + r->size) {
385 
386 			if (unlikely(*len > r->guest_phys_addr + r->size - gpa))
387 				*len = r->guest_phys_addr + r->size - gpa;
388 
389 			return gpa - r->guest_phys_addr +
390 			       r->host_user_addr;
391 		}
392 	}
393 	*len = 0;
394 
395 	return 0;
396 }
397 
398 #define RTE_VHOST_NEED_LOG(features)	((features) & (1ULL << VHOST_F_LOG_ALL))
399 
400 /**
401  * Log the memory write start with given address.
402  *
403  * This function only need be invoked when the live migration starts.
404  * Therefore, we won't need call it at all in the most of time. For
405  * making the performance impact be minimum, it's suggested to do a
406  * check before calling it:
407  *
408  *        if (unlikely(RTE_VHOST_NEED_LOG(features)))
409  *                rte_vhost_log_write(vid, addr, len);
410  *
411  * @param vid
412  *  vhost device ID
413  * @param addr
414  *  the starting address for write (in guest physical address space)
415  * @param len
416  *  the length to write
417  */
418 void rte_vhost_log_write(int vid, uint64_t addr, uint64_t len);
419 
420 /**
421  * Log the used ring update start at given offset.
422  *
423  * Same as rte_vhost_log_write, it's suggested to do a check before
424  * calling it:
425  *
426  *        if (unlikely(RTE_VHOST_NEED_LOG(features)))
427  *                rte_vhost_log_used_vring(vid, vring_idx, offset, len);
428  *
429  * @param vid
430  *  vhost device ID
431  * @param vring_idx
432  *  the vring index
433  * @param offset
434  *  the offset inside the used ring
435  * @param len
436  *  the length to write
437  */
438 void rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
439 			      uint64_t offset, uint64_t len);
440 
441 int rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable);
442 
443 /**
444  * Register vhost driver. path could be different for multiple
445  * instance support.
446  */
447 int rte_vhost_driver_register(const char *path, uint64_t flags);
448 
449 /* Unregister vhost driver. This is only meaningful to vhost user. */
450 int rte_vhost_driver_unregister(const char *path);
451 
452 /**
453  * Set the vdpa device id, enforce single connection per socket
454  *
455  * @param path
456  *  The vhost-user socket file path
457  * @param dev
458  *  vDPA device pointer
459  * @return
460  *  0 on success, -1 on failure
461  */
462 int
463 rte_vhost_driver_attach_vdpa_device(const char *path,
464 		struct rte_vdpa_device *dev);
465 
466 /**
467  * Unset the vdpa device id
468  *
469  * @param path
470  *  The vhost-user socket file path
471  * @return
472  *  0 on success, -1 on failure
473  */
474 int
475 rte_vhost_driver_detach_vdpa_device(const char *path);
476 
477 /**
478  * Get the device id
479  *
480  * @param path
481  *  The vhost-user socket file path
482  * @return
483  *  vDPA device pointer, NULL on failure
484  */
485 struct rte_vdpa_device *
486 rte_vhost_driver_get_vdpa_device(const char *path);
487 
488 /**
489  * Set the feature bits the vhost-user driver supports.
490  *
491  * @param path
492  *  The vhost-user socket file path
493  * @param features
494  *  Supported features
495  * @return
496  *  0 on success, -1 on failure
497  */
498 int rte_vhost_driver_set_features(const char *path, uint64_t features);
499 
500 /**
501  * Enable vhost-user driver features.
502  *
503  * Note that
504  * - the param features should be a subset of the feature bits provided
505  *   by rte_vhost_driver_set_features().
506  * - it must be invoked before vhost-user negotiation starts.
507  *
508  * @param path
509  *  The vhost-user socket file path
510  * @param features
511  *  Features to enable
512  * @return
513  *  0 on success, -1 on failure
514  */
515 int rte_vhost_driver_enable_features(const char *path, uint64_t features);
516 
517 /**
518  * Disable vhost-user driver features.
519  *
520  * The two notes at rte_vhost_driver_enable_features() also apply here.
521  *
522  * @param path
523  *  The vhost-user socket file path
524  * @param features
525  *  Features to disable
526  * @return
527  *  0 on success, -1 on failure
528  */
529 int rte_vhost_driver_disable_features(const char *path, uint64_t features);
530 
531 /**
532  * Get the feature bits before feature negotiation.
533  *
534  * @param path
535  *  The vhost-user socket file path
536  * @param features
537  *  A pointer to store the queried feature bits
538  * @return
539  *  0 on success, -1 on failure
540  */
541 int rte_vhost_driver_get_features(const char *path, uint64_t *features);
542 
543 /**
544  * Set the protocol feature bits before feature negotiation.
545  *
546  * @param path
547  *  The vhost-user socket file path
548  * @param protocol_features
549  *  Supported protocol features
550  * @return
551  *  0 on success, -1 on failure
552  */
553 int
554 rte_vhost_driver_set_protocol_features(const char *path,
555 		uint64_t protocol_features);
556 
557 /**
558  * Get the protocol feature bits before feature negotiation.
559  *
560  * @param path
561  *  The vhost-user socket file path
562  * @param protocol_features
563  *  A pointer to store the queried protocol feature bits
564  * @return
565  *  0 on success, -1 on failure
566  */
567 int
568 rte_vhost_driver_get_protocol_features(const char *path,
569 		uint64_t *protocol_features);
570 
571 /**
572  * Get the queue number bits before feature negotiation.
573  *
574  * @param path
575  *  The vhost-user socket file path
576  * @param queue_num
577  *  A pointer to store the queried queue number bits
578  * @return
579  *  0 on success, -1 on failure
580  */
581 int
582 rte_vhost_driver_get_queue_num(const char *path, uint32_t *queue_num);
583 
584 /**
585  * Get the feature bits after negotiation
586  *
587  * @param vid
588  *  Vhost device ID
589  * @param features
590  *  A pointer to store the queried feature bits
591  * @return
592  *  0 on success, -1 on failure
593  */
594 int rte_vhost_get_negotiated_features(int vid, uint64_t *features);
595 
596 /**
597  * Get the protocol feature bits after negotiation
598  *
599  * @param vid
600  *  Vhost device ID
601  * @param protocol_features
602  *  A pointer to store the queried protocol feature bits
603  * @return
604  *  0 on success, -1 on failure
605  */
606 __rte_experimental
607 int
608 rte_vhost_get_negotiated_protocol_features(int vid,
609 					   uint64_t *protocol_features);
610 
611 /* Register callbacks. */
612 int rte_vhost_driver_callback_register(const char *path,
613 	struct rte_vhost_device_ops const * const ops);
614 
615 /**
616  *
617  * Start the vhost-user driver.
618  *
619  * This function triggers the vhost-user negotiation.
620  *
621  * @param path
622  *  The vhost-user socket file path
623  * @return
624  *  0 on success, -1 on failure
625  */
626 int rte_vhost_driver_start(const char *path);
627 
628 /**
629  * Get the MTU value of the device if set in QEMU.
630  *
631  * @param vid
632  *  virtio-net device ID
633  * @param mtu
634  *  The variable to store the MTU value
635  *
636  * @return
637  *  0: success
638  *  -EAGAIN: device not yet started
639  *  -ENOTSUP: device does not support MTU feature
640  */
641 int rte_vhost_get_mtu(int vid, uint16_t *mtu);
642 
643 /**
644  * Get the numa node from which the virtio net device's memory
645  * is allocated.
646  *
647  * @param vid
648  *  vhost device ID
649  *
650  * @return
651  *  The numa node, -1 on failure
652  */
653 int rte_vhost_get_numa_node(int vid);
654 
655 /**
656  * @deprecated
657  * Get the number of queues the device supports.
658  *
659  * Note this function is deprecated, as it returns a queue pair number,
660  * which is vhost specific. Instead, rte_vhost_get_vring_num should
661  * be used.
662  *
663  * @param vid
664  *  vhost device ID
665  *
666  * @return
667  *  The number of queues, 0 on failure
668  */
669 __rte_deprecated
670 uint32_t rte_vhost_get_queue_num(int vid);
671 
672 /**
673  * Get the number of vrings the device supports.
674  *
675  * @param vid
676  *  vhost device ID
677  *
678  * @return
679  *  The number of vrings, 0 on failure
680  */
681 uint16_t rte_vhost_get_vring_num(int vid);
682 
683 /**
684  * Get the virtio net device's ifname, which is the vhost-user socket
685  * file path.
686  *
687  * @param vid
688  *  vhost device ID
689  * @param buf
690  *  The buffer to stored the queried ifname
691  * @param len
692  *  The length of buf
693  *
694  * @return
695  *  0 on success, -1 on failure
696  */
697 int rte_vhost_get_ifname(int vid, char *buf, size_t len);
698 
699 /**
700  * Get how many avail entries are left in the queue
701  *
702  * @param vid
703  *  vhost device ID
704  * @param queue_id
705  *  virtio queue index
706  *
707  * @return
708  *  num of avail entries left
709  */
710 uint16_t rte_vhost_avail_entries(int vid, uint16_t queue_id);
711 
712 struct rte_mbuf;
713 struct rte_mempool;
714 /**
715  * This function adds buffers to the virtio devices RX virtqueue. Buffers can
716  * be received from the physical port or from another virtual device. A packet
717  * count is returned to indicate the number of packets that were successfully
718  * added to the RX queue.
719  * @param vid
720  *  vhost device ID
721  * @param queue_id
722  *  virtio queue index in mq case
723  * @param pkts
724  *  array to contain packets to be enqueued
725  * @param count
726  *  packets num to be enqueued
727  * @return
728  *  num of packets enqueued
729  */
730 uint16_t rte_vhost_enqueue_burst(int vid, uint16_t queue_id,
731 	struct rte_mbuf **pkts, uint16_t count);
732 
733 /**
734  * This function gets guest buffers from the virtio device TX virtqueue,
735  * construct host mbufs, copies guest buffer content to host mbufs and
736  * store them in pkts to be processed.
737  * @param vid
738  *  vhost device ID
739  * @param queue_id
740  *  virtio queue index in mq case
741  * @param mbuf_pool
742  *  mbuf_pool where host mbuf is allocated.
743  * @param pkts
744  *  array to contain packets to be dequeued
745  * @param count
746  *  packets num to be dequeued
747  * @return
748  *  num of packets dequeued
749  */
750 uint16_t rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
751 	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count);
752 
753 /**
754  * Get guest mem table: a list of memory regions.
755  *
756  * An rte_vhost_vhost_memory object will be allocated internally, to hold the
757  * guest memory regions. Application should free it at destroy_device()
758  * callback.
759  *
760  * @param vid
761  *  vhost device ID
762  * @param mem
763  *  To store the returned mem regions
764  * @return
765  *  0 on success, -1 on failure
766  */
767 int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
768 
769 /**
770  * Get guest vring info, including the vring address, vring size, etc.
771  *
772  * @param vid
773  *  vhost device ID
774  * @param vring_idx
775  *  vring index
776  * @param vring
777  *  the structure to hold the requested vring info
778  * @return
779  *  0 on success, -1 on failure
780  */
781 int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
782 			      struct rte_vhost_vring *vring);
783 
784 /**
785  * Get guest inflight vring info, including inflight ring and resubmit list.
786  *
787  * @param vid
788  *  vhost device ID
789  * @param vring_idx
790  *  vring index
791  * @param vring
792  *  the structure to hold the requested inflight vring info
793  * @return
794  *  0 on success, -1 on failure
795  */
796 int
797 rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
798 	struct rte_vhost_ring_inflight *vring);
799 
800 /**
801  * Set split inflight descriptor.
802  *
803  * This function save descriptors that has been consumed in available
804  * ring
805  *
806  * @param vid
807  *  vhost device ID
808  * @param vring_idx
809  *  vring index
810  * @param idx
811  *  inflight entry index
812  * @return
813  *  0 on success, -1 on failure
814  */
815 int
816 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
817 	uint16_t idx);
818 
819 /**
820  * Set packed inflight descriptor and get corresponding inflight entry
821  *
822  * This function save descriptors that has been consumed
823  *
824  * @param vid
825  *  vhost device ID
826  * @param vring_idx
827  *  vring index
828  * @param head
829  *  head of descriptors
830  * @param last
831  *  last of descriptors
832  * @param inflight_entry
833  *  corresponding inflight entry
834  * @return
835  *  0 on success, -1 on failure
836  */
837 int
838 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
839 	uint16_t head, uint16_t last, uint16_t *inflight_entry);
840 
841 /**
842  * Save the head of list that the last batch of used descriptors.
843  *
844  * @param vid
845  *  vhost device ID
846  * @param vring_idx
847  *  vring index
848  * @param idx
849  *  descriptor entry index
850  * @return
851  *  0 on success, -1 on failure
852  */
853 int
854 rte_vhost_set_last_inflight_io_split(int vid,
855 	uint16_t vring_idx, uint16_t idx);
856 
857 /**
858  * Update the inflight free_head, used_idx and used_wrap_counter.
859  *
860  * This function will update status first before updating descriptors
861  * to used
862  *
863  * @param vid
864  *  vhost device ID
865  * @param vring_idx
866  *  vring index
867  * @param head
868  *  head of descriptors
869  * @return
870  *  0 on success, -1 on failure
871  */
872 int
873 rte_vhost_set_last_inflight_io_packed(int vid,
874 	uint16_t vring_idx, uint16_t head);
875 
876 /**
877  * Clear the split inflight status.
878  *
879  * @param vid
880  *  vhost device ID
881  * @param vring_idx
882  *  vring index
883  * @param last_used_idx
884  *  last used idx of used ring
885  * @param idx
886  *  inflight entry index
887  * @return
888  *  0 on success, -1 on failure
889  */
890 int
891 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
892 	uint16_t last_used_idx, uint16_t idx);
893 
894 /**
895  * Clear the packed inflight status.
896  *
897  * @param vid
898  *  vhost device ID
899  * @param vring_idx
900  *  vring index
901  * @param head
902  *  inflight entry index
903  * @return
904  *  0 on success, -1 on failure
905  */
906 int
907 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
908 	uint16_t head);
909 
910 /**
911  * Notify the guest that used descriptors have been added to the vring.  This
912  * function acts as a memory barrier.
913  *
914  * @param vid
915  *  vhost device ID
916  * @param vring_idx
917  *  vring index
918  * @return
919  *  0 on success, -1 on failure
920  */
921 int rte_vhost_vring_call(int vid, uint16_t vring_idx);
922 
923 /**
924  * Get vhost RX queue avail count.
925  *
926  * @param vid
927  *  vhost device ID
928  * @param qid
929  *  virtio queue index in mq case
930  * @return
931  *  num of desc available
932  */
933 uint32_t rte_vhost_rx_queue_count(int vid, uint16_t qid);
934 
935 /**
936  * Get power monitor address of the vhost device
937  *
938  * @param vid
939  *  vhost device ID
940  * @param queue_id
941  *  vhost queue ID
942  * @param pmc
943  *  power monitor condition
944  * @return
945  *  0 on success, -1 on failure
946  */
947 __rte_experimental
948 int
949 rte_vhost_get_monitor_addr(int vid, uint16_t queue_id,
950 		struct rte_vhost_power_monitor_cond *pmc);
951 
952 /**
953  * Get log base and log size of the vhost device
954  *
955  * @param vid
956  *  vhost device ID
957  * @param log_base
958  *  vhost log base
959  * @param log_size
960  *  vhost log size
961  * @return
962  *  0 on success, -1 on failure
963  */
964 int
965 rte_vhost_get_log_base(int vid, uint64_t *log_base, uint64_t *log_size);
966 
967 /**
968  * Get last_avail/used_idx of the vhost virtqueue
969  *
970  * @param vid
971  *  vhost device ID
972  * @param queue_id
973  *  vhost queue index
974  * @param last_avail_idx
975  *  vhost last_avail_idx to get
976  * @param last_used_idx
977  *  vhost last_used_idx to get
978  * @return
979  *  0 on success, -1 on failure
980  */
981 int
982 rte_vhost_get_vring_base(int vid, uint16_t queue_id,
983 		uint16_t *last_avail_idx, uint16_t *last_used_idx);
984 
985 /**
986  * Get last_avail/last_used of the vhost virtqueue
987  *
988  * This function is designed for the reconnection and it's specific for
989  * the packed ring as we can get the two parameters from the inflight
990  * queueregion
991  *
992  * @param vid
993  *  vhost device ID
994  * @param queue_id
995  *  vhost queue index
996  * @param last_avail_idx
997  *  vhost last_avail_idx to get
998  * @param last_used_idx
999  *  vhost last_used_idx to get
1000  * @return
1001  *  0 on success, -1 on failure
1002  */
1003 int
1004 rte_vhost_get_vring_base_from_inflight(int vid,
1005 	uint16_t queue_id, uint16_t *last_avail_idx, uint16_t *last_used_idx);
1006 
1007 /**
1008  * Set last_avail/used_idx of the vhost virtqueue
1009  *
1010  * @param vid
1011  *  vhost device ID
1012  * @param queue_id
1013  *  vhost queue index
1014  * @param last_avail_idx
1015  *  last_avail_idx to set
1016  * @param last_used_idx
1017  *  last_used_idx to set
1018  * @return
1019  *  0 on success, -1 on failure
1020  */
1021 int
1022 rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1023 		uint16_t last_avail_idx, uint16_t last_used_idx);
1024 
1025 /**
1026  * Register external message handling callbacks
1027  *
1028  * @param vid
1029  *  vhost device ID
1030  * @param ops
1031  *  virtio external callbacks to register
1032  * @param ctx
1033  *  additional context passed to the callbacks
1034  * @return
1035  *  0 on success, -1 on failure
1036  */
1037 int
1038 rte_vhost_extern_callback_register(int vid,
1039 		struct rte_vhost_user_extern_ops const * const ops, void *ctx);
1040 
1041 /**
1042  * Get vdpa device id for vhost device.
1043  *
1044  * @param vid
1045  *  vhost device id
1046  * @return
1047  *  vDPA device pointer on success, NULL on failure
1048  */
1049 struct rte_vdpa_device *
1050 rte_vhost_get_vdpa_device(int vid);
1051 
1052 /**
1053  * Notify the guest that should get virtio configuration space from backend.
1054  *
1055  * @param vid
1056  *  vhost device ID
1057  * @param need_reply
1058  *  wait for the master response the status of this operation
1059  * @return
1060  *  0 on success, < 0 on failure
1061  */
1062 __rte_experimental
1063 int
1064 rte_vhost_slave_config_change(int vid, bool need_reply);
1065 
1066 #ifdef __cplusplus
1067 }
1068 #endif
1069 
1070 #endif /* _RTE_VHOST_H_ */
1071