xref: /f-stack/dpdk/lib/librte_vhost/vhost.h (revision 16d80a6d)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2018 Intel Corporation
3  */
4 
5 #ifndef _VHOST_NET_CDEV_H_
6 #define _VHOST_NET_CDEV_H_
7 #include <stdint.h>
8 #include <stdio.h>
9 #include <stdbool.h>
10 #include <sys/types.h>
11 #include <sys/queue.h>
12 #include <unistd.h>
13 #include <linux/vhost.h>
14 #include <linux/virtio_net.h>
15 #include <sys/socket.h>
16 #include <linux/if.h>
17 
18 #include <rte_log.h>
19 #include <rte_ether.h>
20 #include <rte_rwlock.h>
21 
22 #include "rte_vhost.h"
23 #include "rte_vdpa.h"
24 
25 /* Used to indicate that the device is running on a data core */
26 #define VIRTIO_DEV_RUNNING 1
27 /* Used to indicate that the device is ready to operate */
28 #define VIRTIO_DEV_READY 2
29 /* Used to indicate that the built-in vhost net device backend is enabled */
30 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET 4
31 /* Used to indicate that the device has its own data path and configured */
32 #define VIRTIO_DEV_VDPA_CONFIGURED 8
33 
34 /* Backend value set by guest. */
35 #define VIRTIO_DEV_STOPPED -1
36 
37 #define BUF_VECTOR_MAX 256
38 
39 #define VHOST_LOG_CACHE_NR 32
40 
41 /**
42  * Structure contains buffer address, length and descriptor index
43  * from vring to do scatter RX.
44  */
45 struct buf_vector {
46 	uint64_t buf_iova;
47 	uint64_t buf_addr;
48 	uint32_t buf_len;
49 	uint32_t desc_idx;
50 };
51 
52 /*
53  * A structure to hold some fields needed in zero copy code path,
54  * mainly for associating an mbuf with the right desc_idx.
55  */
56 struct zcopy_mbuf {
57 	struct rte_mbuf *mbuf;
58 	uint32_t desc_idx;
59 	uint16_t desc_count;
60 	uint16_t in_use;
61 
62 	TAILQ_ENTRY(zcopy_mbuf) next;
63 };
64 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
65 
66 /*
67  * Structure contains the info for each batched memory copy.
68  */
69 struct batch_copy_elem {
70 	void *dst;
71 	void *src;
72 	uint32_t len;
73 	uint64_t log_addr;
74 };
75 
76 /*
77  * Structure that contains the info for batched dirty logging.
78  */
79 struct log_cache_entry {
80 	uint32_t offset;
81 	unsigned long val;
82 };
83 
84 struct vring_used_elem_packed {
85 	uint16_t id;
86 	uint32_t len;
87 	uint32_t count;
88 };
89 
90 /**
91  * Structure contains variables relevant to RX/TX virtqueues.
92  */
93 struct vhost_virtqueue {
94 	union {
95 		struct vring_desc	*desc;
96 		struct vring_packed_desc   *desc_packed;
97 	};
98 	union {
99 		struct vring_avail	*avail;
100 		struct vring_packed_desc_event *driver_event;
101 	};
102 	union {
103 		struct vring_used	*used;
104 		struct vring_packed_desc_event *device_event;
105 	};
106 	uint32_t		size;
107 
108 	uint16_t		last_avail_idx;
109 	uint16_t		last_used_idx;
110 	/* Last used index we notify to front end. */
111 	uint16_t		signalled_used;
112 	bool			signalled_used_valid;
113 #define VIRTIO_INVALID_EVENTFD		(-1)
114 #define VIRTIO_UNINITIALIZED_EVENTFD	(-2)
115 
116 	/* Backend value to determine if device should started/stopped */
117 	int			backend;
118 	int			enabled;
119 	int			access_ok;
120 	rte_spinlock_t		access_lock;
121 
122 	/* Used to notify the guest (trigger interrupt) */
123 	int			callfd;
124 	/* Currently unused as polling mode is enabled */
125 	int			kickfd;
126 
127 	/* Physical address of used ring, for logging */
128 	uint64_t		log_guest_addr;
129 
130 	uint16_t		nr_zmbuf;
131 	uint16_t		zmbuf_size;
132 	uint16_t		last_zmbuf_idx;
133 	struct zcopy_mbuf	*zmbufs;
134 	struct zcopy_mbuf_list	zmbuf_list;
135 
136 	union {
137 		struct vring_used_elem  *shadow_used_split;
138 		struct vring_used_elem_packed *shadow_used_packed;
139 	};
140 	uint16_t                shadow_used_idx;
141 	struct vhost_vring_addr ring_addrs;
142 
143 	struct batch_copy_elem	*batch_copy_elems;
144 	uint16_t		batch_copy_nb_elems;
145 	bool			used_wrap_counter;
146 	bool			avail_wrap_counter;
147 
148 	struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
149 	uint16_t log_cache_nb_elem;
150 
151 	rte_rwlock_t	iotlb_lock;
152 	rte_rwlock_t	iotlb_pending_lock;
153 	struct rte_mempool *iotlb_pool;
154 	TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
155 	int				iotlb_cache_nr;
156 	TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
157 } __rte_cache_aligned;
158 
159 /* Old kernels have no such macros defined */
160 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
161  #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
162 #endif
163 
164 #ifndef VIRTIO_NET_F_MQ
165  #define VIRTIO_NET_F_MQ		22
166 #endif
167 
168 #define VHOST_MAX_VRING			0x100
169 #define VHOST_MAX_QUEUE_PAIRS		0x80
170 
171 #ifndef VIRTIO_NET_F_MTU
172  #define VIRTIO_NET_F_MTU 3
173 #endif
174 
175 #ifndef VIRTIO_F_ANY_LAYOUT
176  #define VIRTIO_F_ANY_LAYOUT		27
177 #endif
178 
179 /* Declare IOMMU related bits for older kernels */
180 #ifndef VIRTIO_F_IOMMU_PLATFORM
181 
182 #define VIRTIO_F_IOMMU_PLATFORM 33
183 
184 struct vhost_iotlb_msg {
185 	__u64 iova;
186 	__u64 size;
187 	__u64 uaddr;
188 #define VHOST_ACCESS_RO      0x1
189 #define VHOST_ACCESS_WO      0x2
190 #define VHOST_ACCESS_RW      0x3
191 	__u8 perm;
192 #define VHOST_IOTLB_MISS           1
193 #define VHOST_IOTLB_UPDATE         2
194 #define VHOST_IOTLB_INVALIDATE     3
195 #define VHOST_IOTLB_ACCESS_FAIL    4
196 	__u8 type;
197 };
198 
199 #define VHOST_IOTLB_MSG 0x1
200 
201 struct vhost_msg {
202 	int type;
203 	union {
204 		struct vhost_iotlb_msg iotlb;
205 		__u8 padding[64];
206 	};
207 };
208 #endif
209 
210 /*
211  * Define virtio 1.0 for older kernels
212  */
213 #ifndef VIRTIO_F_VERSION_1
214  #define VIRTIO_F_VERSION_1 32
215 #endif
216 
217 /* Declare packed ring related bits for older kernels */
218 #ifndef VIRTIO_F_RING_PACKED
219 
220 #define VIRTIO_F_RING_PACKED 34
221 
222 struct vring_packed_desc {
223 	uint64_t addr;
224 	uint32_t len;
225 	uint16_t id;
226 	uint16_t flags;
227 };
228 
229 struct vring_packed_desc_event {
230 	uint16_t off_wrap;
231 	uint16_t flags;
232 };
233 #endif
234 
235 /*
236  * Declare below packed ring defines unconditionally
237  * as Kernel header might use different names.
238  */
239 #define VRING_DESC_F_AVAIL	(1ULL << 7)
240 #define VRING_DESC_F_USED	(1ULL << 15)
241 
242 #define VRING_EVENT_F_ENABLE 0x0
243 #define VRING_EVENT_F_DISABLE 0x1
244 #define VRING_EVENT_F_DESC 0x2
245 
246 /*
247  * Available and used descs are in same order
248  */
249 #ifndef VIRTIO_F_IN_ORDER
250 #define VIRTIO_F_IN_ORDER      35
251 #endif
252 
253 /* Features supported by this builtin vhost-user net driver. */
254 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
255 				(1ULL << VIRTIO_F_ANY_LAYOUT) | \
256 				(1ULL << VIRTIO_NET_F_CTRL_VQ) | \
257 				(1ULL << VIRTIO_NET_F_CTRL_RX) | \
258 				(1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
259 				(1ULL << VIRTIO_NET_F_MQ)      | \
260 				(1ULL << VIRTIO_F_VERSION_1)   | \
261 				(1ULL << VHOST_F_LOG_ALL)      | \
262 				(1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
263 				(1ULL << VIRTIO_NET_F_GSO) | \
264 				(1ULL << VIRTIO_NET_F_HOST_TSO4) | \
265 				(1ULL << VIRTIO_NET_F_HOST_TSO6) | \
266 				(1ULL << VIRTIO_NET_F_HOST_UFO) | \
267 				(1ULL << VIRTIO_NET_F_HOST_ECN) | \
268 				(1ULL << VIRTIO_NET_F_CSUM)    | \
269 				(1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
270 				(1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
271 				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
272 				(1ULL << VIRTIO_NET_F_GUEST_UFO) | \
273 				(1ULL << VIRTIO_NET_F_GUEST_ECN) | \
274 				(1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
275 				(1ULL << VIRTIO_RING_F_EVENT_IDX) | \
276 				(1ULL << VIRTIO_NET_F_MTU)  | \
277 				(1ULL << VIRTIO_F_IN_ORDER) | \
278 				(1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
279 				(1ULL << VIRTIO_F_RING_PACKED))
280 
281 
282 struct guest_page {
283 	uint64_t guest_phys_addr;
284 	uint64_t host_phys_addr;
285 	uint64_t size;
286 };
287 
288 /* The possible results of a message handling function */
289 enum vh_result {
290 	/* Message handling failed */
291 	VH_RESULT_ERR   = -1,
292 	/* Message handling successful */
293 	VH_RESULT_OK    =  0,
294 	/* Message handling successful and reply prepared */
295 	VH_RESULT_REPLY =  1,
296 };
297 
298 /**
299  * function prototype for the vhost backend to handler specific vhost user
300  * messages prior to the master message handling
301  *
302  * @param vid
303  *  vhost device id
304  * @param msg
305  *  Message pointer.
306  * @param skip_master
307  *  If the handler requires skipping the master message handling, this variable
308  *  shall be written 1, otherwise 0.
309  * @return
310  *  VH_RESULT_OK on success, VH_RESULT_REPLY on success with reply,
311  *  VH_RESULT_ERR on failure
312  */
313 typedef enum vh_result (*vhost_msg_pre_handle)(int vid, void *msg,
314 		uint32_t *skip_master);
315 
316 /**
317  * function prototype for the vhost backend to handler specific vhost user
318  * messages after the master message handling is done
319  *
320  * @param vid
321  *  vhost device id
322  * @param msg
323  *  Message pointer.
324  * @return
325  *  VH_RESULT_OK on success, VH_RESULT_REPLY on success with reply,
326  *  VH_RESULT_ERR on failure
327  */
328 typedef enum vh_result (*vhost_msg_post_handle)(int vid, void *msg);
329 
330 /**
331  * pre and post vhost user message handlers
332  */
333 struct vhost_user_extern_ops {
334 	vhost_msg_pre_handle pre_msg_handle;
335 	vhost_msg_post_handle post_msg_handle;
336 };
337 
338 /**
339  * Device structure contains all configuration information relating
340  * to the device.
341  */
342 struct virtio_net {
343 	/* Frontend (QEMU) memory and memory region information */
344 	struct rte_vhost_memory	*mem;
345 	uint64_t		features;
346 	uint64_t		protocol_features;
347 	int			vid;
348 	uint32_t		flags;
349 	uint16_t		vhost_hlen;
350 	/* to tell if we need broadcast rarp packet */
351 	rte_atomic16_t		broadcast_rarp;
352 	uint32_t		nr_vring;
353 	int			dequeue_zero_copy;
354 	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
355 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
356 	char			ifname[IF_NAME_SZ];
357 	uint64_t		log_size;
358 	uint64_t		log_base;
359 	uint64_t		log_addr;
360 	struct ether_addr	mac;
361 	uint16_t		mtu;
362 
363 	struct vhost_device_ops const *notify_ops;
364 
365 	uint32_t		nr_guest_pages;
366 	uint32_t		max_guest_pages;
367 	struct guest_page       *guest_pages;
368 
369 	int			slave_req_fd;
370 	rte_spinlock_t		slave_req_lock;
371 
372 	int			postcopy_ufd;
373 	int			postcopy_listening;
374 
375 	/*
376 	 * Device id to identify a specific backend device.
377 	 * It's set to -1 for the default software implementation.
378 	 */
379 	int			vdpa_dev_id;
380 
381 	/* private data for virtio device */
382 	void			*extern_data;
383 	/* pre and post vhost user message handlers for the device */
384 	struct vhost_user_extern_ops extern_ops;
385 } __rte_cache_aligned;
386 
387 static __rte_always_inline bool
388 vq_is_packed(struct virtio_net *dev)
389 {
390 	return dev->features & (1ull << VIRTIO_F_RING_PACKED);
391 }
392 
393 static inline bool
394 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
395 {
396 	uint16_t flags = *((volatile uint16_t *) &desc->flags);
397 
398 	return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
399 		wrap_counter != !!(flags & VRING_DESC_F_USED);
400 }
401 
402 #define VHOST_LOG_PAGE	4096
403 
404 /*
405  * Atomically set a bit in memory.
406  */
407 static __rte_always_inline void
408 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
409 {
410 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
411 	/*
412 	 * __sync_ built-ins are deprecated, but __atomic_ ones
413 	 * are sub-optimized in older GCC versions.
414 	 */
415 	__sync_fetch_and_or_1(addr, (1U << nr));
416 #else
417 	__atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
418 #endif
419 }
420 
421 static __rte_always_inline void
422 vhost_log_page(uint8_t *log_base, uint64_t page)
423 {
424 	vhost_set_bit(page % 8, &log_base[page / 8]);
425 }
426 
427 static __rte_always_inline void
428 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
429 {
430 	uint64_t page;
431 
432 	if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
433 		   !dev->log_base || !len))
434 		return;
435 
436 	if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
437 		return;
438 
439 	/* To make sure guest memory updates are committed before logging */
440 	rte_smp_wmb();
441 
442 	page = addr / VHOST_LOG_PAGE;
443 	while (page * VHOST_LOG_PAGE < addr + len) {
444 		vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
445 		page += 1;
446 	}
447 }
448 
449 static __rte_always_inline void
450 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
451 {
452 	unsigned long *log_base;
453 	int i;
454 
455 	if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
456 		   !dev->log_base))
457 		return;
458 
459 	log_base = (unsigned long *)(uintptr_t)dev->log_base;
460 
461 	/*
462 	 * It is expected a write memory barrier has been issued
463 	 * before this function is called.
464 	 */
465 
466 	for (i = 0; i < vq->log_cache_nb_elem; i++) {
467 		struct log_cache_entry *elem = vq->log_cache + i;
468 
469 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
470 		/*
471 		 * '__sync' builtins are deprecated, but '__atomic' ones
472 		 * are sub-optimized in older GCC versions.
473 		 */
474 		__sync_fetch_and_or(log_base + elem->offset, elem->val);
475 #else
476 		__atomic_fetch_or(log_base + elem->offset, elem->val,
477 				__ATOMIC_RELAXED);
478 #endif
479 	}
480 
481 	rte_smp_wmb();
482 
483 	vq->log_cache_nb_elem = 0;
484 }
485 
486 static __rte_always_inline void
487 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
488 			uint64_t page)
489 {
490 	uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
491 	uint32_t offset = page / (sizeof(unsigned long) << 3);
492 	int i;
493 
494 	for (i = 0; i < vq->log_cache_nb_elem; i++) {
495 		struct log_cache_entry *elem = vq->log_cache + i;
496 
497 		if (elem->offset == offset) {
498 			elem->val |= (1UL << bit_nr);
499 			return;
500 		}
501 	}
502 
503 	if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
504 		/*
505 		 * No more room for a new log cache entry,
506 		 * so write the dirty log map directly.
507 		 */
508 		rte_smp_wmb();
509 		vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
510 
511 		return;
512 	}
513 
514 	vq->log_cache[i].offset = offset;
515 	vq->log_cache[i].val = (1UL << bit_nr);
516 	vq->log_cache_nb_elem++;
517 }
518 
519 static __rte_always_inline void
520 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
521 			uint64_t addr, uint64_t len)
522 {
523 	uint64_t page;
524 
525 	if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
526 		   !dev->log_base || !len))
527 		return;
528 
529 	if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
530 		return;
531 
532 	page = addr / VHOST_LOG_PAGE;
533 	while (page * VHOST_LOG_PAGE < addr + len) {
534 		vhost_log_cache_page(dev, vq, page);
535 		page += 1;
536 	}
537 }
538 
539 static __rte_always_inline void
540 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
541 			uint64_t offset, uint64_t len)
542 {
543 	vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
544 }
545 
546 static __rte_always_inline void
547 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
548 		     uint64_t offset, uint64_t len)
549 {
550 	vhost_log_write(dev, vq->log_guest_addr + offset, len);
551 }
552 
553 /* Macros for printing using RTE_LOG */
554 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
555 #define RTE_LOGTYPE_VHOST_DATA   RTE_LOGTYPE_USER1
556 
557 #ifdef RTE_LIBRTE_VHOST_DEBUG
558 #define VHOST_MAX_PRINT_BUFF 6072
559 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
560 	RTE_LOG(DEBUG, log_type, fmt, ##args)
561 #define PRINT_PACKET(device, addr, size, header) do { \
562 	char *pkt_addr = (char *)(addr); \
563 	unsigned int index; \
564 	char packet[VHOST_MAX_PRINT_BUFF]; \
565 	\
566 	if ((header)) \
567 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
568 	else \
569 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
570 	for (index = 0; index < (size); index++) { \
571 		snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
572 			"%02hhx ", pkt_addr[index]); \
573 	} \
574 	snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
575 	\
576 	VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
577 } while (0)
578 #else
579 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
580 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
581 #endif
582 
583 extern uint64_t VHOST_FEATURES;
584 #define MAX_VHOST_DEVICE	1024
585 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
586 
587 /* Convert guest physical address to host physical address */
588 static __rte_always_inline rte_iova_t
589 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
590 {
591 	uint32_t i;
592 	struct guest_page *page;
593 
594 	for (i = 0; i < dev->nr_guest_pages; i++) {
595 		page = &dev->guest_pages[i];
596 
597 		if (gpa >= page->guest_phys_addr &&
598 		    gpa + size < page->guest_phys_addr + page->size) {
599 			return gpa - page->guest_phys_addr +
600 			       page->host_phys_addr;
601 		}
602 	}
603 
604 	return 0;
605 }
606 
607 static __rte_always_inline struct virtio_net *
608 get_device(int vid)
609 {
610 	struct virtio_net *dev = vhost_devices[vid];
611 
612 	if (unlikely(!dev)) {
613 		RTE_LOG(ERR, VHOST_CONFIG,
614 			"(%d) device not found.\n", vid);
615 	}
616 
617 	return dev;
618 }
619 
620 int vhost_new_device(void);
621 void cleanup_device(struct virtio_net *dev, int destroy);
622 void reset_device(struct virtio_net *dev);
623 void vhost_destroy_device(int);
624 void vhost_destroy_device_notify(struct virtio_net *dev);
625 
626 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
627 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
628 
629 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
630 
631 void vhost_attach_vdpa_device(int vid, int did);
632 void vhost_detach_vdpa_device(int vid);
633 
634 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
635 void vhost_enable_dequeue_zero_copy(int vid);
636 void vhost_set_builtin_virtio_net(int vid, bool enable);
637 
638 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
639 
640 /*
641  * Backend-specific cleanup.
642  *
643  * TODO: fix it; we have one backend now
644  */
645 void vhost_backend_cleanup(struct virtio_net *dev);
646 
647 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
648 			uint64_t iova, uint64_t *len, uint8_t perm);
649 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
650 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
651 
652 static __rte_always_inline uint64_t
653 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
654 			uint64_t iova, uint64_t *len, uint8_t perm)
655 {
656 	if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
657 		return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
658 
659 	return __vhost_iova_to_vva(dev, vq, iova, len, perm);
660 }
661 
662 #define vhost_avail_event(vr) \
663 	(*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
664 #define vhost_used_event(vr) \
665 	(*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
666 
667 /*
668  * The following is used with VIRTIO_RING_F_EVENT_IDX.
669  * Assuming a given event_idx value from the other size, if we have
670  * just incremented index from old to new_idx, should we trigger an
671  * event?
672  */
673 static __rte_always_inline int
674 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
675 {
676 	return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
677 }
678 
679 static __rte_always_inline void
680 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
681 {
682 	/* Flush used->idx update before we read avail->flags. */
683 	rte_smp_mb();
684 
685 	/* Don't kick guest if we don't reach index specified by guest. */
686 	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
687 		uint16_t old = vq->signalled_used;
688 		uint16_t new = vq->last_used_idx;
689 		bool signalled_used_valid = vq->signalled_used_valid;
690 
691 		vq->signalled_used = new;
692 		vq->signalled_used_valid = true;
693 
694 		VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
695 			__func__,
696 			vhost_used_event(vq),
697 			old, new);
698 
699 		if ((vhost_need_event(vhost_used_event(vq), new, old) &&
700 					(vq->callfd >= 0)) ||
701 				unlikely(!signalled_used_valid))
702 			eventfd_write(vq->callfd, (eventfd_t) 1);
703 	} else {
704 		/* Kick the guest if necessary. */
705 		if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
706 				&& (vq->callfd >= 0))
707 			eventfd_write(vq->callfd, (eventfd_t)1);
708 	}
709 }
710 
711 static __rte_always_inline void
712 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
713 {
714 	uint16_t old, new, off, off_wrap;
715 	bool signalled_used_valid, kick = false;
716 
717 	/* Flush used desc update. */
718 	rte_smp_mb();
719 
720 	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
721 		if (vq->driver_event->flags !=
722 				VRING_EVENT_F_DISABLE)
723 			kick = true;
724 		goto kick;
725 	}
726 
727 	old = vq->signalled_used;
728 	new = vq->last_used_idx;
729 	vq->signalled_used = new;
730 	signalled_used_valid = vq->signalled_used_valid;
731 	vq->signalled_used_valid = true;
732 
733 	if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
734 		if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
735 			kick = true;
736 		goto kick;
737 	}
738 
739 	if (unlikely(!signalled_used_valid)) {
740 		kick = true;
741 		goto kick;
742 	}
743 
744 	rte_smp_rmb();
745 
746 	off_wrap = vq->driver_event->off_wrap;
747 	off = off_wrap & ~(1 << 15);
748 
749 	if (new <= old)
750 		old -= vq->size;
751 
752 	if (vq->used_wrap_counter != off_wrap >> 15)
753 		off -= vq->size;
754 
755 	if (vhost_need_event(off, new, old))
756 		kick = true;
757 kick:
758 	if (kick)
759 		eventfd_write(vq->callfd, (eventfd_t)1);
760 }
761 
762 static __rte_always_inline void
763 restore_mbuf(struct rte_mbuf *m)
764 {
765 	uint32_t mbuf_size, priv_size;
766 
767 	while (m) {
768 		priv_size = rte_pktmbuf_priv_size(m->pool);
769 		mbuf_size = sizeof(struct rte_mbuf) + priv_size;
770 		/* start of buffer is after mbuf structure and priv data */
771 
772 		m->buf_addr = (char *)m + mbuf_size;
773 		m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
774 		m = m->next;
775 	}
776 }
777 
778 static __rte_always_inline bool
779 mbuf_is_consumed(struct rte_mbuf *m)
780 {
781 	while (m) {
782 		if (rte_mbuf_refcnt_read(m) > 1)
783 			return false;
784 		m = m->next;
785 	}
786 
787 	return true;
788 }
789 
790 static __rte_always_inline void
791 put_zmbuf(struct zcopy_mbuf *zmbuf)
792 {
793 	zmbuf->in_use = 0;
794 }
795 
796 #endif /* _VHOST_NET_CDEV_H_ */
797